Datasets:

Hendrik-a

/

arxiv

like 0

we apply the quantum optimal control theory based on the krotov method to implement single - qubit and gates and two - qubit cnot gates for inductively coupled superconducting flux qubits with fixed qubit transition frequencies and fixed off - diagonal qubit - qubit coupling . our scheme that shares the same advantage of other directly coupling schemes requires no additional coupler subcircuit and control lines . the control lines needed are only for the manipulation of individual qubits ( e.g. , a time - dependent magnetic flux or field applied on each qubit ) . the qubits are operated at the optimal coherence points and the gate operation times ( single - qubit gates ns ; cnot gates ns ) are much shorter than the corresponding qubit decoherence time . a cnot gate or other general quantum gates can be implemented in a single run of pulse sequence rather than being decomposed into several single - qubit and some entangled two - qubit operations in series by composite pulse sequences . quantum gates constructed via our scheme are all with very high fidelity ( very low error ) as our optimal control scheme takes into account the fixed qubit detuning and fixed two - qubit interaction as well as all other time - dependent magnetic - field - induced single - qubit interactions and two - qubit couplings . the effect of leakage to higher energy - level states and the effect of qubit decoherence on the quantum gate operations are also discussed .

1406.7707

superconducting josephson junction devices and circuits are proving to be promising systems for quantum information processing @xcite . due to the great controllability of the qubits and microwaves in the superconducting circuit systems , single - qubit @xcite , two - qubit @xcite , and three - qubit @xcite operations have been experimentally demonstrated . the next natural step is to develop robust , high - fidelity and scalable gates for larger scale quantum computation . an essential prerequisite for quantum information processing and quantum computation is precise coherent control of quantum systems or quantum bits ( qubits ) . here , we focus our discussion on high - fidelity quantum control for superconducting flux qubit systems . the flux qubits have an advantage over other types of superconducting qubits in the larger anharmonic energy level structure , i.e. the difference between adjacent transition frequencies is larger @xcite and thus less leakage to higher energy - level states . several schemes to implement local qubit operations and controllable couplings using microwaves for multi - flux - qubit systems have been proposed or / and realized . these schemes may be categorized into two groups : ( i ) directly coupling between qubits @xcite and ( ii ) indirectly coupling through an intermediate coupler @xcite . for example , a scheme to control the effective qubit - qubit interaction for directly couping flux qubits with fixed qubit transition frequencies and large detuning between the neighboring qubits using time - dependent magnetic fluxes ( or microwaves ) was proposed in ref . . while this approach is advantageous due to the resonant nature of the coupling , its disadvantage is that at least one of the qubits must be biased away from the coherence optimal point @xcite ; this makes the qubit susceptible to low - frequency flux noise and results in a shorter coherence time . for the case of indirect coupling through an intermediate coupler , the coupling subcircuit or coupling nonlinear element can also be driven with microwaves @xcite . these indirectly coupling schemes have the advantage of enabling the qubits to be operated at their optimal coherence points @xcite . however , in addition to the added circuit complexity , the effective couplings between two qubits in the indirectly coupling schemes are generally smaller so the two - qubit gating time is about @xmath4 ns , at least one or two order(s ) of magnitude longer than that of directly coupling schemes . schemes for microwave controllable coupling , which allow qubits to be operated at optimal bias points and without intermediate coupler subcircuit , exist @xcite . but the effective qubit - qubit couplings in these schemes are usually even smaller than those of the indirectly coupling schemes . smaller two - qubit coupling or equivalently longer two - qubit gating time will usually make the qubits suffer more decoherence effect during the gate operation , thus deteriorating the gate fidelity . the fidelities of single - qubit and two - qubit quantum gate operations in a multi - qubit register for the present directly / indirectly coupling schemes via microwaves in the unitary case ( without considering decoherence effect ) even under the two - level qubit approximation are not perfect @xcite . the microwave pulses for single - qubit gate operations are generally obtained and applied under the approximation that the qubit with large detuning with its neighboring qubits is effectively decoupled with its neighbors even though there exist fixed two - qubit couplings between them . but this decoupling is only valid in the first order of a small parameter that is the ratio of the two - qubit coupling to the detuning . for two - qubit operations , the microwave pulses are commonly constructed and performed under the rotating wave approximation , the adiabatic approximation for the nonlinear coupler , or / and the approximation of neglecting other small residual two - qubit interactions . the qubit - decoupled approximation becomes excellent when the effective qubit - qubit coupling is very small ; however , a large qubit - qubit interaction is favorable for two - qubit operations . therefore , reaching a delicate balance between these two situations results in , for instance , single - qubit gate errors greater than @xmath5 for practical experimental parameters even in the ideal , unitary case of simple two - level approximation for each flux qubit . quantum optimal control theory is a powerful tool that provides a variational framework for finding optimal control field profiles or sequences by maximizing a desired physical objective ( or minimizing a physical cost function ) within certain constraints @xcite . in this paper , we apply the quantum optimal control theory to find the control pulse sequences of externally applied ac magnetic fluxes ( fields ) to implement fast and high - fidelity one - qubit and two - qubit gates on superconducting flux qubits . there have been optimal control studies of quantum gates for superconducting qubits , but all focusing on the gate operations of cooper - pair - box charge qubit systems @xcite . our investigation here is , to our knowledge , the first optimal control study for the flux qubit systems . the flux qubits we consider have fixed direct qubit - qubit couplings and fixed transition frequencies but with large detuning between neighboring qubits , which ensures the qubits are effectively decoupled ( to the first order ) in the absence of time - dependent control fields or signals . our optimal control scheme requires no additional bias or control lines beyond those used for the manipulation of individual qubits . in the absence of the time - dependent magnetic fluxes , the flux qubits in our scheme are biased at the coherence optimal points @xcite to reduce dephasing due to flux noise , i.e.,to have a longer coherence time . furthermore , our optimal control scheme takes into account two - qubit interaction , qubit detuning and other time - dependent magnetic field induced residual single - qubit and two - qubit interactions when performing single - qubit and two - qubit gates . thus the gate fidelities ( errors ) in the unitary ( closed - system ) case can be as high ( low ) as one wishes , limited only by the accuracy of the approximated two - level qubit hamiltonian used and by the machine precision of the computation . besides , the two - qubit cnot gate operations of our scheme is about several folds to two orders of magnitude faster than the directly or / and indirectly microwave controllable coupling schemes @xcite . by considering leakage to higher - energy - level states , the gate errors for single - qubit gates using the optimal control pulse sequences obtained are in the order of @xmath6 and are in the order of @xmath7 for two - qubit cnot gates to take the effect of qubit decoherence into account , we model the qubit dynamics by a quantum master equation with experimentally available relaxation and dephasing rates . the gate errors in the presence of decoheence by considering the master equation are still in the order of @xmath7 for single - qubit gates and in the order of @xmath8 for two - qubit cnot gates . these gate errors are still below the error threshold @xmath9 ( @xmath5 in @xcite ; @xmath10 if surface code error correction is used @xcite ) required for fault - tolerant quantum computation . the paper is organized as follows . we first describe in sec . [ sec : hamiltonian ] the hamiltonian of two inductively coupled flux qubits with each individual qubit controlled by a time - dependent magnetic flux ( field ) . the reduced hamiltonian at the optimal bias point expressed in terms of the two - level qubit basis states is then obtained . in sec . [ sec : qoct ] , a brief description about the quantum optimal control theory for performing state - independent quantum gate operations is presented . the control field pulse sequences obtained by the optimal control theory and their corresponding gate errors and state evolutions are presented in sec . [ sec : ressults ] . it is also shown that our optimal control scheme is notably robust against leakage to states outside the computational basis state space . the effect of qubit decoherence on gate errors is also discussed . a short conclusion with discussions of how to implement the optimal control pulses experimentally is given in sec . [ sec : conclusion ] . ( color online ) schematic illustration of two inductively coupled superconducting flux qubits . the mutual inductance is @xmath11 and there are three junctions in each qubit loop . the external static and time - dependent magnetic fluxes through the @xmath12th qubit are denoted as @xmath13 and @xmath14 , respectively . , scaledwidth=50.0% ] the system we consider here is two flux qubits right next to each other and coupled by their mutual inductance @xmath11 @xcite as schematically illustrated in fig . [ fig : coupled_flux_qubits ] . each qubit loop contains three josephson junctions @xcite , and one of them has an area @xmath15 times smaller than that of the two identical junctions . the larger josephson junctions in the @xmath12th qubit loop have the josephson energy @xmath16 and the corresponding critical current @xmath17 , where @xmath18 is the flux quantum with the planck constant @xmath19 and the elementary charge @xmath20 . the capacitances in the @xmath12th qubit loop satisfy the conditions @xmath21 and @xmath22 , where @xmath23 are the capacitance of the larger josephson junctions . the coupled flux qubit system considered here is exactly the same as that discussed in ref . fluxoid quantization around each qubit loop imposes a constraint on the phase drop across the three junctions @xcite : @xmath24 where @xmath25 is the gauge - invariant phase of each junction , @xmath26 and @xmath27 are , respectively , the static ( dc ) and time - dependent magnetic fluxes applied through the @xmath12th qubit . here the self - inductance is considered negligible as compared to the josephson inductance . the requirement for our coupled - qubit system is to have suitable mutual inductance for qubits placed next to each other as illustrated in fig . [ fig : coupled_flux_qubits ] and independent flux lines to enable control over each individual qubit with external dc bias magnetic fluxes and time - dependent control magnetic fluxes . the qubit - loop size for three - junction flux qubits could range experimentally from submicrometer to micrometer . as long as the design of the qubit loops fulfills the above requirement , the actual size of each qubit loop or the distance between the qubits can have some flexibility . the total hamiltonian of the two coupled three - junction flux qubits reads @xcite : @xmath28\right\ } , \label{eq : htot}\end{aligned}\ ] ] with the redefined phases @xmath29 , @xmath30 , the reduced dc bias magnetic flux @xmath31 , and the reduced time - dependent control magnetic flux @xmath32 . the effective masses are @xmath33 and @xmath34 , which correspond to the effective momenta @xmath35 and @xmath36 . the persistence current in the @xmath12th qubit loop is @xmath37 . \label{eq : i}\end{aligned}\ ] ] here the self - interaction terms of the external time - dependent control magnetic fields @xmath38^{2}$ ] is neglected since they vary only the global phase of the state vector and do not affect the control of the coupled flux qubits . to simulate the dynamics of the coupled flux - qubit system with hamiltonian eq . ( [ eq : htot ] ) , the parameter values of the mutual conductance @xmath11 as well as the josephson energies of the junctions @xmath16 , the capacitance of the larger josephson junctions @xmath39 and the area ratio @xmath15 of the smaller junction to one of the larger junctions in each qubit loop are required . these parameter values can be obtained through the experimental measurement and characterization of the system . the actual values of these parameters we choose for our simulations are given in sec . [ sec : ressults ] . they are all experimentally available or realistic values . we will keep the time - dependent control amplitudes small such that the reduced time - dependent magnetic flux satisfies @xmath40 . small time - dependent control amplitude allows the approximation of @xmath41\sim2\pi f_{c}^{\left(l\right)}(t)$ ] and @xmath42\sim 1 $ ] when expanding the last term of the hamiltonian of eq . ( [ eq : htot ] ) . as a result , the time - dependent hamiltonian becomes linear in @xmath43 or @xmath44 , which can be readily incorporated with the krotov quantum optimal control method that we will employ later . this weak - amplitude approximation of the control fields also keeps the qubit not deviating much from the dc bias point that is set to be the optimal coherence point in our case . moreover , this weak - amplitude approximation helps reduce unwanted possible excitations to the higher - energy - level states outside the computational state space when we make two - level ( qubit ) approximation discussed later . making this weak - amplitude approximation , we obtain from eq . ( [ eq : htot ] ) the hamiltonian similar to that in ref . @xcite as @xmath45 the first term @xmath46 is the single - qubit hamiltonian and reads : @xmath47.\label{eq : hamiltonian03}\end{aligned}\ ] ] the hamiltonian @xmath48 plays the role of a driving hamiltonian representing the interaction between the @xmath12th qubit and its time - dependent magnetic field . it takes the form @xmath49 . \label{driving_ham}\ ] ] the last three terms in eq . ( [ eq : hamiltonian02 ] ) come from the inductive coupling between the two flux qubits . the hamiltonian @xmath50 describes the qubit - qubit interaction controlled by one of the time - dependent magnetic flux [ @xmath51 or @xmath52 and is written as @xmath53 where @xmath54 \label{loop_current}\end{aligned}\ ] ] is the loop current of the @xmath12th qubit when no time - dependent magnetic fluxes are applied [ cf . , eq . ( [ eq : i ] ) ] . the qubit - qubit interaction @xmath55 controlled by two simultaneously applied time - dependent magnetic fluxes through , respectively , the two qubits is @xmath56 we still keep this term , although it is much smaller than the term controlled by a single time - dependent magnetic flux . the hamiltonian @xmath57 describes an always - on interaction between the two flux qubits without time - dependent magnetic fluxes being applied , where @xmath58 is defined in eq . ( [ loop_current ] ) . the hamiltonian , eq . ( [ eq : hamiltonian02 ] ) , in the two - qubit computational state basis @xmath59 , where @xmath60 and @xmath61 are the lowest two energy - level states of @xmath46 of eq . ( [ eq : hamiltonian03 ] ) , becomes @xmath62 \nonumber \\ & & -\sum_{l\neq m=1}^{2}\left[\chi_{1}^{\left(lm\right)}(t)\sigma_{z}^{\left(l\right)}+\chi_{2}^{\left(lm\right)}(t)\sigma_{x}^{\left(l\right)}\right]\nonumber \\ & & -\sum_{l\neq m=1}^{2}\left[\xi_{11}^{\left(lm\right)}(t)\sigma_{z}^{\left(l\right)}\sigma_{z}^{\left(m\right)}+\xi_{22}^{\left(lm\right)}(t)\sigma_{x}^{\left(l\right)}\sigma_{x}^{\left(m\right)}\right . \nonumber \\ & & \qquad \qquad \left . + \xi_{12}^{\left(lm\right)}(t)\sigma_{z}^{\left(l\right)}\sigma_{x}^{\left(m\right)}+\xi_{21}^{\left(lm\right)}(t)\sigma_{x}^{\left(l\right)}\sigma_{z}^{\left(m\right)}\right]\nonumber \\ & & + \theta_{11}(t)\sigma_{z}^{\left(1\right)}\sigma_{z}^{\left(2\right)}+\theta_{22}(t)\sigma_{x}^{\left(1\right)}\sigma_{x}^{\left(2\right)}+\theta_{12}(t)\sigma_{z}^{\left(1\right)}\sigma_{x}^{\left(2\right)}+\theta_{21}(t)\sigma_{x}^{\left(1\right)}\sigma_{z}^{\left(2\right)}\nonumber \\ & & + \lambda_{11}\sigma_{z}^{\left(1\right)}\sigma_{z}^{\left(2\right)}+\lambda_{22}\sigma_{x}^{\left(1\right)}\sigma_{x}^{\left(2\right)}+\lambda_{12}\sigma_{z}^{\left(1\right)}\sigma_{x}^{\left(2\right)}+\lambda_{21}\sigma_{x}^{\left(1\right)}\sigma_{z}^{\left(2\right)},\label{eq : hamiltonian04}\end{aligned}\ ] ] where the single - qubit transition frequencies @xmath63 and the driving hamiltonian amplitudes @xmath64 and @xmath65 are @xmath66 @xmath67 @xmath68 respectively , the other time - dependent controllable interaction strengths , @xmath69 , @xmath70 , @xmath71 and @xmath72 , coming from qubit - qubit inductive interaction are @xmath73 @xmath74 @xmath75 @xmath76 respectively , and the static fixed qubit - qubit interaction strengths @xmath77 are @xmath78 the relevant parameters in eqs . ( [ eq : param01])-([eq : param02 ] ) are @xmath79 which is the mutual inductive energy with respect to the critical current @xmath80 of the larger josephson junctions in each qubit loop , @xmath81 @xmath82 @xmath83 , \label{omega1}\end{aligned}\ ] ] @xmath84 and @xmath85 , \label{delta}\end{aligned}\ ] ] where the operator @xmath86 is defined as @xmath87 the hamiltonian eq . ( [ eq : hamiltonian04 ] ) with parameters defined in eqs . ( [ transition_freq])-([upsilon ] ) is valid for weak time - dependent magnetic fluxes ( fields ) and arbitrary static bias magnetic fluxes . if we take the time dependence of the control magnetic flux to be sinusoidal , i.e. , @xmath88 , and making the relevant rotating wave approximation , we can revert to the hamiltonian of ref . different from ref . @xcite , our control scheme , however , does not require qubits being biased away from the optimal coherence points . in contrast , we set the reduced static or dc bias magnetic fluxes to be @xmath89 so that the qubits are at the optimal coherence points and are thus insensitive to low - frequency flux noise in the first order . at the optimal coherence points of @xmath89 , hamiltonian @xmath46 of eq . ( [ eq : hamiltonian03 ] ) is invariant under the parity transformation of @xmath90 , i.e. , @xmath91 . thus its eigenstates have definite parities and the lowest two energy - level eigenstates have the opposite parities @xcite . since the hamiltonian of eq . ( [ driving_ham ] ) and the loop currents of eq . ( [ loop_current ] ) are odd functions of @xmath90 and the operators @xmath92 from eq . ( [ upsilon ] ) are even functions of @xmath90 at @xmath89 , the parameters @xmath64 , @xmath93 and @xmath94 defined in eqs . ( [ kappa_1 ] ) , ( [ lambda1 ] ) and ( [ omega2 ] ) , respectively , vanish due to the parity symmetry consideration . as a consequence , most parameters in eqs . ( [ eq : param01])-([eq : param02 ] ) vanish , and eq ( [ eq : hamiltonian04 ] ) at the optimal bias points simplifies to @xmath95+\lambda_{22}\sigma_{x}^{\left(1\right)}\sigma_{x}^{\left(2\right)}\nonumber \\ & & -\sum_{l\neq m=1}^{2}\left[\chi_{1}^{\left(lm\right)}(t)\sigma_{z}^{\left(l\right)}+\xi_{12}^{\left(lm\right)}(t)\sigma_{z}^{\left(l\right)}\sigma_{x}^{\left(m\right)}\right]+\theta_{11}(t)\sigma_{z}^{\left(1\right)}\sigma_{z}^{\left(2\right)}.\label{eq : hamiltonian05}\end{aligned}\ ] ] we will use eq . ( [ eq : hamiltonian05 ] ) combining with quantum optimal control theory to find control magnetic field pulses @xmath43 for the implementation of high - fidelity one- and two - qubit gates . quantum optimal control theory enables us to realize accurate state - independent quantum gates by selecting optimal pulse shapes ( arbitrarily shaped pulses and duration ; or continuous dynamical modulation ) for the external control within experimental capabilities @xcite . to perform state - independent optimal control , the equation of motion for the time evolution operator ( or propagator ) @xmath96 is needed : @xmath97u\left(t\right ) , \label{eq : sch1}\ ] ] where @xmath98 is the system hamiltonian and @xmath99 is a time - dependent control field . we choose the trace distance between the desired target gate operation @xmath100 and the actual ( could be nonunitary ) propagator @xmath101 at the final operation time @xmath102 to characterize the gate error : @xmath103^{\dagger}\left[o - u(t)\right]\right\ } \label{error1 } , % \\ % & = & \frac{1}{2n } % \left\langle \left\langle % u\left(t\right)-o\left|u\left(t\right)-o\right\rangle % \right\rangle \right .. % \label{error2}\end{aligned}\ ] ] where @xmath104 is the dimension of the matrix @xmath96 . if the actual time evolution operator @xmath101 is equal to the target gate operation @xmath100 , then @xmath105 with @xmath106 denoting the gate fidelity . sometimes , one is also interested in calculating the errors of the operations ( operators or matrices ) in a subspace of a unitary matrix . when projection is made to the subspace , the projected matrix in the subspace is in general no longer unitary . for instance , when the evolution operator of the multi - level coupled flux - qubit system @xmath101 for the multi - level hamiltonian , eq . ( [ eq : htot ] ) , is projected into the approximated two - level qubit computational state subspace , the projected time evolution operator is no longer unitary . in this case , eq . ( [ error1 ] ) with slight modification [ see eq . ( [ errorp ] ) ] is still an appropriate measure for calculating the gate error . in realistic control problems , it is desirable that the optimal control sequence can provide highest quality ( fidelity ) with minimum energy consumption . therefore , we define the cost function for our optimal control problem as @xmath107^{2}dt , \label{eq : costfun02}\ ] ] where @xmath108 is defined in eq . ( [ error1 ] ) , @xmath109 , a positive shape function , and @xmath110 , a weight , can be adjusted and chosen empirically @xcite . here , the reference field @xmath111 is chosen to be the control sequence in the previous iteration , i.e. , @xmath112 such that the control field energy constraint in the cost function @xmath113 of eq . ( [ eq : costfun02 ] ) has the physical interpretation that the change of the control pulse energy in each iteration is limited @xcite . when the iterative procedure approaches the optimal solution , the change in the control field is minimal or vanishing . therefore , this choice of the reference field @xmath111 ensures that the iterative method is found to reduce the total objective @xmath113 of eq . ( [ eq : costfun02 ] ) by reducing the gate error @xmath108 rather than the total control pulse energy . with eqs . ( [ eq : sch1 ] ) and ( [ eq : costfun02 ] ) and the initial propagator @xmath114 , where @xmath115 is the identity operator , one can then investigate the state - independent quantum gate optimal control problem . we will use the krotov iterative method for the quantum gate optimal control @xcite . the krotov method has several appealing advantages @xcite over the standard gradient optimization methods : ( a ) monotonic increase of the objective with iteration number , ( b ) no requirement for a line search , and ( c ) macrosteps at each iteration . the optimal algorithm following the krotov method @xcite can be found in refs . . we use the reduced hamiltonian , eq . ( [ eq : hamiltonian05 ] ) , with two lowest energy - level states for each qubit to obtain control sequences for single - qubit @xmath0 and @xmath1 gate operations and two - qubit cnot gate operations by optimizing the cost function @xmath113 , eq . ( [ eq : costfun02 ] ) , using the krotov iterative method . simulations on the two - qubit hamiltonian , eq . ( [ eq : htot ] ) , without making the weak - amplitude approximation and with the lowest five energy states for each qubit , using the optimal control field sequences found via eq . ( [ eq : hamiltonian05 ] ) will also be performed for error comparison . this allows us to test how well the optimal control pulses obtained by the weak - amplitude approximation perform , and to determine how severe the leakage to the higher - energy - level states is . the target gate operation @xmath100 is defined in the two - qubit computational state basis . the evolution operator or propagator @xmath101 results from eq . ( [ eq : htot ] ) using the optimal control pulse sequences @xmath116 and @xmath117 obtained from eq . ( [ eq : hamiltonian05 ] ) is in a larger multi - level - state space . we thus define a projection operator @xmath118 to project @xmath101 onto the subspace subtended by the computational basis states , and compare the target operator @xmath100 with the resultant propagator @xmath119 which is in general no longer unitary . the error in this case is defined from eq . ( [ error1 ] ) as @xmath120^{\dagger}[o - pu(t)]\right\ } , % \eta_p=(1/32){\rm tr}\left\{[o - pu(t)]^{\dagger}[o - pu(t)]\right\}. % \left\langle \left\langle % pu\left(t\right)-o\left|\right.pu\left(t\right)-o\right\rangle % \right\rangle . \label{errorp}\ ] ] where @xmath104 is the dimension of the @xmath100 matrix or @xmath121 matrix . we choose the parameters in the two - qubit hamiltonian with experimentally available or realistic values @xcite . the josephson energy of the larger junctions in the qubit loops is chosen to be @xmath122 ghz and the charging energy @xmath123 is chosen to be @xmath124 . the capacitance of the smaller josephson junction is @xmath125 with the ratio @xmath126 . to fit the value of the coupling strength @xmath127 reported in the experiment of ref . , we take the mutual inductance @xmath128 ph and obtain @xmath129 . then other parameters are determined by the calculations through the hamiltonian , eq . ( [ eq : htot ] ) . we then obtain the qubit transition energies @xmath130 and @xmath131 , and the single - qubit driving amplitudes @xmath132 and @xmath133 . the time - dependent coefficients of the last three coupling terms in eq . ( [ eq : hamiltonian05 ] ) are also obtained to be @xmath134 @xmath135 and @xmath136 . we note here that the dynamics and also the optimization results that will be presented later depend on the values of the microscopic parameters . these parameter values can be obtained through the experimental measurement and characterization of the system but always come with some error bars . thus if the actual system is , for example , slightly away from the assumed control point ( a set of values for the system parameters ) chosen for the quantum optimal control theory , then the optimal control pulse sequences sent to the experiment may result in gate fidelities not close to optimal . this issue of imprecise knowledge of the system parameters seems to hinder the practical use and experimental applicability of the quantum optimal control theory . fortunately , there may be ways around , for example , using the adaptive model - free closed - loop feedback control method . we postpone the discussions regarding this until sec . [ sec : conclusion ] . in the following , we first present our optimal control results for single - qubit and two - qubit gates in the unitary case together with brief descriptions of corresponding conventional approaches for comparison . the results considering the effect of leakage to higher - energy - level states and the effect of qubit decoherence are presented subsequently . ( color online ) gate errors @xmath108 as a function of the number of iterations . the gate errors calculated using hamiltonian eq . ( [ eq : hamiltonian06 ] ) decrease as the number of iterations increases . the stopping iteration criteria of error is set to be @xmath137 . the single - qubit gates of the first qubit , the single - qubit gates of the second qubit and the two - qubit cnot gate are achieved at @xmath138 ns , @xmath139 ns and @xmath140 ns , respectively . , scaledwidth=50.0% ] _ conventional approach - _ in most of the studies for the directly coupling schemes , the hamiltonian of two inductively coupled flux qubits at the optimal bias point contains only the first line of eq . ( [ eq : hamiltonian05 ] ) : @xmath141+\lambda_{22}\sigma_{x}^{\left(1\right)}\sigma_{x}^{\left(2\right)}. \label{eq : hamiltonian06}\end{aligned}\ ] ] this is because the coefficients of the last three time - dependent terms of eq . ( [ eq : hamiltonian05 ] ) are several orders of magnitude smaller than the qubit transition frequencies @xmath142 , the single - qubit driving amplitudes @xmath143 and the fixed static qubit - qubit interaction strength @xmath144 . thus neglecting these coefficients does not introduce appreciable error . furthermore , since the strength @xmath144 of the off - diagonal coupling @xmath145 in the presence of large qubit detuning @xmath146 affects the qubit dynamics only in second order in the parameter @xmath147 , the constant coupling @xmath144 term is often neglected at the optimal bias point provided that @xmath147 is rather small . in this case , the qubits are regarded to be effectively decoupled . thus the single - qubit gate , say , @xmath148 gate ( rotation of angle @xmath149 around @xmath150-axis on qubit 1 ) , can be achieved by simply driving the magnetic flux through the first qubit with microwave resonance with the transition frequency of the first qubit , i.e. , @xmath151 , while keeping the time - dependent flux through the second qubit off , i.e. , @xmath152 . when the driving field strength @xmath153 is much smaller than the transition frequencies @xmath63 and qubit detuning @xmath146 in which the rotating - wave approximation can be made , eq . ( [ eq : hamiltonian06 ] ) in the frame rotating with the driving frequency @xmath154 becomes @xmath155 performing single - qubit @xmath156-gate on qubit 1 will also demand qubit 2 to return to its original state up to a global phase at the end of the @xmath156 operation , i.e. , an identity gate on qubit 2 . in this rotating frame , qubit 2 takes @xmath157 to complete an identity gate up to a global phase with @xmath158 being an integer number . taking this time to be the time to complete a @xmath149-pulse on qubit 1 yields the single - qubit driving strength or the rabi frequency to be @xmath159 . by requiring the reduced time - dependent magnetic flux @xmath160 such that the weak - amplitude approximation to obtain eq . ( [ eq : hamiltonian02 ] ) is valid , one obtains the minimum integer number @xmath158 to be @xmath161 . this then leads to @xmath162 and the operation time of @xmath156 gate @xmath163 ns . plugging these numbers into eq . ( [ eq : hamiltonian05 ] ) and eq . ( [ eq : htot ] ) to simulate an @xmath156 gate gives an error of about @xmath164 . this result indicates that the dominant source of error is mainly due to the constant qubit - qubit interaction term of @xmath144 and does not come from the leakage to higher - energy - level states @xcite . indeed , for the experimental parameters used here , the ratio of @xmath165 , and taking the second order correction gives an estimated error also in the order of @xmath5 . this is about the best one can do in a unitary case if neglecting the constant qubit - qubit interaction @xmath144 term . how to include the qubit - qubit interaction term and perform a much more accurate single - qubit gate is not intuitively obvious . _ optimal control approach - _ here , we apply the krotov iterative method to obtain the optimal control sequences for high - fidelity ( low - error ) single - qubit and two - qubit gates taking all the static and time - dependent single - qubit and two - qubit terms or interactions into account . usually , stronger strengths of the control fields result in shorter gate operation times . in our case , the gate operation times are , however , chosen such that the reduced time - dependent magnetic fluxes satisfying the weak - amplitude approximation of @xmath40 . we find that the high - fidelity single - qubit gates on qubit 1 can be achieved at a gating time of @xmath138 ns , and the gating time is @xmath139 ns for qubit 2 . the two - qubit gates will be discussed in the next subsection . as shown in fig . [ fig : error ] , the errors @xmath108 of @xmath166 , @xmath148 , @xmath167 and @xmath168 gates calculated using the reduced hamiltonian , eq . ( [ eq : hamiltonian05 ] ) , decrease with the number of iterations . we note here that for simplicity , we let the positive shape function @xmath169 in eq . ( [ eq : costfun02 ] ) be a constant . in addition , the time - dependent control field @xmath99 in eq . ( [ eq : costfun02 ] ) in our case becomes the reduced time - dependent magnetic fluxes @xmath170 that are dimensionless and required to be smaller than @xmath5 , and thus in our calculations we choose @xmath171 ( ghz)@xmath172 in order to make sure that the iterations converge to the optimal results monotonically . in principle , the gate error for a closed , unitary system can , via the optimal control theory , be as low as one wishes , limited only by the accuracy of the two - level - approximation hamiltonian , eq . ( [ eq : hamiltonian05 ] ) , and by the machine precision of the computation . considering the accuracy of the two - level - approximation hamiltonian used , we set the stopping criteria of error to be @xmath173 , and thus the iterations are terminated when error @xmath137 . a typical set of the control sequences of the reduced time - dependent magnetic fluxes @xmath174 and @xmath175 for the high - fidelity @xmath156 gate is shown in figs . [ fig : x1 ] and [ fig : x1 ] . _ conventional approach - _ many schemes using microwaves have been put forward @xcite to realize an effective controllable qubit - qubit interaction for hamiltonian ( [ eq : hamiltonian06 ] ) or ( [ eq : hamiltonian05 ] ) . however , in most of the schemes , two neighboring qubits are considered effectively decoupled ( as described earlier ) when no time - dependent microwave field is applied . this approximation limits the maximum fidelity ( or minimum error ) these schemes can actually achieve . one of the common goals for gate control is to implement a two - qubit cnot gate , an important universal gate for quantum computation . in all of the schemes , an entangled gate natural to their couplings is implemented , and thus to realize a cnot gate , an additional sequence of single - qubit rotations is needed . for example , if an iswap gate is conveniently implemented in a particular scheme , the cnot gate can be realized by the following sequence cnot@xmath176[r_z(-\pi/2)\otimes r_z(\pi/2)][{\rm iswap}][r_x(\pi/2)\otimes i][{\rm iswap}][i \otimes r_z(\pi/2)]$ ] . typical two - qubit gate operation times in these schemes are in the order of @xmath177 to @xmath178 ns . however , decomposing a cnot gate or a general gate operation into several single - qubit and some entangled two - qubit operations in series makes its operation time normally longer than that of completing the operation in one single run of pulses . moreover , the overall gate error may also become larger as the gate errors of the decomposed gates will add up or accumulate . _ optimal control approach - _ in contrast , the optimal control method has great advantages of enabling the implementation of a cnot gate or other general quantum gates in one single run of pulse or in one single pulse sequence by simply setting the target operation to be the cnot gate or the general quantum gate one wishes to implement . requiring the reduced time - dependent magnetic flux @xmath40 , we then set the cnot gate operation time @xmath179 ns and use the reduced hamiltonian ( [ eq : hamiltonian05 ] ) to find the optimal control sequence . this cnot gating time is about several folds to two orders of magnitude shorter than that of the directly or / and indirectly microwave controllable coupling schemes @xcite . this is because single - qubit and two - qubit operations can be performed in parallel simultaneously ( rather than in series ) in one single run of the optimal control pulses . one can see from fig . [ fig : error ] that the cnot gate error drops quickly with the number of iterations . here , qubit 1 is the control qubit and qubit 2 is the target qubit . the optimal control sequences obtained when the iteration terminates at the point when error @xmath137 are shown in figs . [ fig : cnot ] and [ fig : cnot ] . to investigate the gate error due to the leakage to higher energy - level states outside the computational basis - state space , we apply the control sequences of the single - qubit and two - qubit gates obtained from the optimization of the reduced two - level - approximation hamiltonian ( [ eq : hamiltonian05 ] ) to the multi - level hamiltonian ( [ eq : htot ] ) and calculate the errors @xmath180 of eq . ( [ errorp ] ) in the projected two - qubit computational state subspace . the single - qubit gate errors are found to be @xmath181 , @xmath182 , @xmath183 , and @xmath184 for the @xmath166 , @xmath148 , @xmath167 and @xmath168 gates , respectively . these results indicate that the excitations to higher energy - level states increase slightly the gate errors ; however , the gate errors in the order of @xmath6 are still much smaller than @xmath5 of those conventional approaches that apply directly a microwave @xmath149 pulse by neglecting the qubit - qubit interaction . we also simulate the @xmath156 gate numerically with the multi - level hamiltonian ( [ eq : htot ] ) using the optimal control sequences obtained from the reduced hamiltonian ( [ eq : hamiltonian05 ] ) for initial four different qubit computational basis states @xmath185 , @xmath186 , @xmath187 and @xmath188 . the time evolutions of the state probabilities of the @xmath156 gate are shown in figs . [ fig : x1]-[fig : x1 ] . as expected , after the @xmath156 operation , @xmath189 , @xmath190 , @xmath191 , and @xmath192 . it seems that there is an approximate symmetry under time - reversal between figs . [ fig : x1 ] and [ fig : x1 ] , and between [ fig : x1 ] and [ fig : x1 ] . this is mainly due to the fact that the control pulse sequences for the @xmath156 gate shown in figs . [ fig : x1 ] and [ fig : x1 ] are seemingly symmetric with respect to the midpoint of the operation time . thus the time evolutions of the state probabilities , when the first qubit making the transition between @xmath193 and @xmath194 with the second qubit returning back to the same state , appear to have such a time - reversal symmetry . in our case , the pulse sequences are obtained when the gate error is smaller than the error threshold of @xmath137 for a certain initial guess of the pulse sequence . the truly global optimal pulses should , however , correspond to the minimum value of the cost function obtained by trying all possible initial guesses of the pulse shapes . thus other initial guesses can generate slightly different @xmath156-gate pulse sequences satisfying still the same error threshold requirement but with less symmetry . in fact , the pulse sequence of @xmath175 for the @xmath195 gate we obtain ( not shown here ) does not have an apparently symmetric shape with respect to the midpoint of the operation time as compared to @xmath174 of the @xmath156 gate in figs . [ fig : x1 ] . as a result , the time evolutions of the state probability between @xmath196 ( latexmath:[$|ee\rangle \rightarrow show prominent time reversal symmetry as their corresponding counterparts for the @xmath156 gate . there is , however , indeed an approximate symmetry in time evolutions of the state probabilities in figs . [ fig : x1]-[fig : x1 ] . this approximate symmetry could be understood in terms of hamiltonian eq . ( [ eq : hamiltonian05 ] ) , or more easily in terms of hamiltonian eq . ( [ eq : hamiltonian06 ] ) by neglecting the tiny contributions of the last three time - dependent terms of eq . ( [ eq : hamiltonian05 ] ) . if one puts aside the fixed static qubit - qubit interaction @xmath144 term , then the two qubits in hamiltonian eq . ( [ eq : hamiltonian06 ] ) are decoupled and have the symmetry of exchanging @xmath198 and @xmath199 in the presence of @xmath200 term . as a result , the time evolutions of the state probabilities @xmath201 , @xmath202 , @xmath203 and @xmath204 in figs . [ fig : x1]-[fig : x1 ] , respectively , behave approximately the same . similarly , the approximately same time evolutions for the state probabilities @xmath203 , @xmath204 , @xmath201 and @xmath202 in figs . [ fig : x1]-[fig : x1 ] , respectively , are also observed . applying the optimal control sequences of two - qubit cnot gate shown in figs . [ fig : cnot ] and to multi - level hamiltonian eq . ( [ eq : htot ] ) with initial four different qubit computational basis states @xmath185 , @xmath186 , @xmath187 and @xmath188 , we plot the time evolutions of the state probabilities of the cnot gate in figs . [ fig : cnot]-[fig : cnot ] . when qubit 1 is initially in the state @xmath205 , the two qubits return to their original states ; when qubit 1 is initially in the state @xmath206 , an effective not operation is performed on qubit 2 ( the target qubit ) at the end of the cnot gate operation . this is exactly what we see in figs . [ fig : cnot]-[fig : cnot ] . the error of the cnot gate when including the higher energy - level states is @xmath207 . by comparing this error to the error of the single - qubit gates of @xmath208 , leakage to the higher - energy level states outside the computational basis state space is more appreciable for the cnot gate control pulses . this may be due to the fact that the magnitude of the reduced time - dependent magnetic flux @xmath175 of fig . [ fig : cnot ] for the cnot gate is larger than @xmath174 and @xmath175 for the single - qubit gates [ see , e.g. , @xmath174 and @xmath175 of figs . [ fig : x1 ] and [ fig : x1 ] for the @xmath156 gate ] . nevertheless , the cnot gate error of the two coupled flux qubits is , to our knowledge , still much lower than the currently available schemes when the leakage to higher energy - level states is considered . we have also performed an optimal cnot gate operation with qubit 1 being the target qubit and qubit 2 being the control qubit . this cnot gate can also be accomplished at @xmath140 ns with error @xmath209 . we show below that our optimal control scheme is robust against qubit decoherence . we use the born - markov master equation for the reduced density matrix of the flux qubit system in the lindblad form to model the effect of decoherence on the qubit dynamics @xcite : @xmath210+\sum_{i=1}^{2}\left(\gamma_{1,i}\mathcal{d}\left[\sigma_{i}^{-}\right]\rho(t)+\gamma_{\varphi , i}\mathcal{d}\left[\sigma_{i}^{z}\right]\rho(t)\right ) , \label{eq : mastereq1}\\ & \equiv & \mathcal{l}\rho(t ) , \label{eq : mastereq}\end{aligned}\ ] ] where the hamiltonian @xmath98 is defined in eq . ( [ eq : hamiltonian05 ] ) , the superoperator @xmath211\rho = c\rho c^{\dagger}-\frac{1}{2}c^{\dagger}c\rho-\frac{1}{2}\rho c^{\dagger}c,\ ] ] and @xmath212 and @xmath213 are the relaxation rate and dephasing rate of qubit @xmath214 , respectively . generally speaking , if one starts from a microscopic model in the weak system - environment coupling limit , the argument in the superoperator @xmath215 in the born - markov approximation should rather be operators generating transitions between the energy eigenstates of the coupled - qubit hamiltonian ( [ eq : hamiltonian05 ] ) @xcite . however , for small inter - qubit coupling considered here , these operators in the rotating - wave approximation can be approximately represented by individual qubit transition operators , @xmath216 and @xmath217 . in addition , at finite temperatures one should have a dissipative term describing the influx of energy from the thermal environment into the system . but since typical flux - qubit experiments are conducted at a low temperature around 50 mk @xcite which is much smaller than the transition frequencies of the qubits considered here ( i.e. , @xmath218 ) . as a result , the thermal mean occupation number of the environment modes at the energy of about @xmath219 approaches zero , and the environment may be regarded as an effective zero - temperature bath . these justify the use of the master equation ( [ eq : mastereq1 ] ) to phenomenologically model the decoherence dynamics of the coupled flux - qubit system . the master equation is symbolically expressed in a concise form with a liouville s superoperator @xmath220 in eq . ( [ eq : mastereq ] ) . the equation of motion of the propagator can be found , by substituting @xmath221 into eq . ( [ eq : mastereq ] ) , to be @xmath222 with the relaxation time @xmath223 obtained from the rabi oscillation and the decoherence time @xmath224 from ramsey interference measurements in an recent experiment @xcite that probed the noise spectrum of a superconducting flux qubit , one is able to deduce the realistic values of @xmath212 and @xmath213 using the relations of @xmath225 , @xmath226 and @xmath227 . define the gate error @xmath228 by replacing @xmath229 with @xmath230 in eq . ( [ error1 ] ) for the case considering qubit decoherence . one can perform the optimal control calculation by minimizing the cost function eq . ( [ eq : costfun02 ] ) with @xmath231 using the equation of motion ( [ eq : g ] ) . the single - qubit gate errors considering the effect of decoherence are found to be in the order of @xmath7 , and the errors are in the order of @xmath8 for two - qubit cnot gates . these gate errors are still below the error threshold @xmath9 ( @xmath5 in @xcite ; @xmath10 if surface code error correction is used @xcite ) required for fault - tolerant quantum computation . we have also tested numerically that using a finite - temperature master equation with arguments in the superoperator @xmath215 being the operators generating transitions between the energy eigenstates but with the decay rates and the dephasing rates assuming to be the same as individual single qubits leads to the similar error values with correction only to the digits in @xmath6 or @xmath232 . .[tab : gate_error]summary of the quantum gate errors [ cols="^,^,^,^,^",options="header " , ] we have applied the quantum optimal control method to implement fast and high - fidelity single - qubit and two - qubit gates for two inductively coupled superconducting flux qubits with fixed static off - diagonal qubit - qubit coupling and fixed qubit transition frequencies . table [ tab : gate_error ] summarizes the gate errors calculated with realistic experimental parameters for ideal unitary case , and for the cases considering the effect of leakage state and the effect of qubit decoherence . our optimal control scheme has the following great advantages . ( 1 ) our scheme that shares the same advantage of other directly coupling scheme requires no additional coupler subcircuit and control lines and thus is simple in experimental design . the control lines needed are only for the manipulation of individual qubits ( e.g. , the time - dependent magnetic flux or field through each qubit ) . ( 2 ) quantum gates constructed via our scheme are all with very high fidelity ( very low error ) as our optimal control scheme takes into account the fixed qubit detuning and fixed two - qubit interaction as well as all other time - dependent magnetic - field - induced single - qubit interactions and two - qubit couplings when performing single - qubit and two - qubit gates . ( 3 ) our scheme can cope with noise and decoherence very well as the qubits are biased at the optimal coherence point to reduce the influence of low - frequency flux noise , and the gate operation time ( @xmath3 ns ) is much shorter than the corresponding qubit decoherence time ( a few @xmath233s or larger ) @xcite so that the decoherence effect on such a fast gate is significantly diminished . ( 4 ) a cnot gate or other general quantum gates can be implemented in a single run of pulse sequence rather than being decomposed into several single - qubit and some entangled two - qubit operations in series by composite pulse sequences . a natural question then arises : how to generate these optimal control pulses experimentally ? the pulse sequences shown in fig . [ fig : x1 ] and , and in fig . [ fig : cnot ] and look experimentally challenging , but not impossible . commercial devices ( e.g. , tektronix awg70001a ) for generating arbitrary wave forms with 10 bits of vertical resolution at a sample rate of 50 gsa / s , a bit rate of 12.5 gb / s and a rise / fall time smaller than 27 ps are now available . such a device should enable generation of complex signals in a time scale of sub - nanoseconds to nanoseconds . this high - end arbitrary wave form generator combining with ultrafast josephson electronics should be or about to be able to fulfill the necessary specifications for implementing the optimal control pulse sequences we obtain for the quantum gate operations presented here . another challenge for the experimental implementations and applications of the quantum optimal control theory is one s imprecise knowledge of the quantum system s parameters . typically , the quantum gates constructed by the quantum optimal control theory are computed to very high precision assuming the parameters in the model hamiltonian are exactly known . however , for real systems controlled in the experiments , the parameters and also the hamiltonians usually are not known exactly . this poses a challenge to implement such high - fidelity gates successfully in the laboratory . recently , a hybrid open / closed - loop optimal control method called adaptation by hybrid optimal control ( ad - hoc ) has been proposed @xcite to overcome not only the problem of inaccurate knowledge of the system parameters but also shortcomings of the assumed physical model and errors on the controls themselves . the basic idea is to use the open - loop quantum optimal control theory to find the optimal control pulses with the best available model and parameters of the system . then the pulses are sent to the experiment and their performance are efficiently measured by using , for example , the method of randomized benchmarking @xcite . the closed - loop pulse calibration of ad - hoc , similar to adaptive model - free feedback control ( also referred to as closed - loop laboratory control or learning control ) @xcite , uses the physical system itself as a feedback to calibrate control pulses and optimize their performance . by using , e.g. , the robust and efficient nelder - mead algorithm @xcite , the control pulses are updated experimentally and the procedure is iterated until a target performance is reached or convergence stops . after the calibration is finished , the optimized pulses can then be used . even though the precise experimental parameters are never identified , ad - hoc can reduce the initial gate errors of the numerically obtained optimal control pulses implemented in an experiment by at least an order of magnitude @xcite . thus the quantum optimal control theory is practical and applicable experimentally through , e.g. , the ad - hoc protocol @xcite . a recent experiment demonstrating single - qubit operations ( with fidelity of 0.99 ) and two - qubit entanglement for the electron spins of two proximal nitrogen - vacancy centers in diamond using optimal control has been reported @xcite . it is thus believed that high - fidelity quantum gates with optimal control pulses obtained via our scheme will be realized experimentally in the near future . we are very grateful to prof . yu - xi liu for sending us his note for the derivation of the coupled flux - qubit hamiltonian of ref . @xcite . we acknowledge support from the national science council in taiwan under grant no . 100 - 2112-m-002 - 003-my3 , from the national taiwan university under grants no . 103r891400 , no . 103r891402 and 102r3253 , and from the focus group program of the national center for theoretical sciences , taiwan . 99 j. q. you and f. nori , phys . today * 58 * , 11 , 42 ( 2005 ) ; j. q. you and f. nori , nature * 474 * , 589 ( 2011 ) , and references therein . j. clarke and f. k. wilhelm , nature * 453 * , 1031 ( 2008 ) , and references therein . y. nakamura , yu . a. pashkin , and j. s. tsai , nature * 398 * , 786 ( 1999 ) . d. vion , a. aassime , a. cottet , p. joyez , h. pothier , c. urbina , d. esteve , and m. h. devoret , science * 296 * , 886 ( 2002 ) . i. chiorescu , y. nakamura , c. j. p. m. harmans , and j. e. mooij , science * 299 * , 1869 ( 2003 ) . a. wallraff , d. i. schuster , a. blais , l. frunzio , j. majer , m. h. devoret , s. m. girvin , and r. j. schoelkopf , phys . lett . * 95 * , 060501 ( 2005 ) . j. m. martinis , k. b. cooper , r. mcdermott , m. steffen , m. ansmann , k. d. osborn , k. cicak , s. oh , d. p. pappas , r. w. simmonds , and c. c. yu , phys . rev . lett . * 95 * , 210503 ( 2005 ) . c. rigetti , j. m. gambetta , s. poletto , b. l. t. plourde , j. m. chow , a. d. crcoles , j. a. smolin , s. t. merkel , j. r. rozen , g. a. keefe , m. b. rothwell , m. b. ketchen , and m. steffen , phys . b * 86 * , 100506 ( 2012 ) . j. m. chow , l. dicarlo , j. m. gambetta , f. motzoi , l. frunzio , s. m. girvin , and r. j. schoelkopf , phys . a * 82 * , 040305(r ) ( 2010 ) . h. paik , d. i. schuster , l. s. bishop , g. kirchmair , g. catelani , a. p. sears , b. r. johnson , m. j. reagor , l. frunzio , l. i. glazman , s. m. girvin , m. h. devoret , and r. j. schoelkopf , phys . lett . * 107 * , 240501 ( 2011 ) . j. majer , j. m. chow , j. m. gambetta , j. koch , b. r. johnson , j. a. schreier , l. frunzio , d. i. schuster , a. a. houck , a. wallraff , a. blais , m. h. devoret , s. m. girvin , and r. j. schoelkopf , nature * 449 * , 443 ( 2007 ) . l. dicarlo , j. m. chow , j. m. gambetta , l. s. bishop , b. r. johnson , d. i. schuster , j. majer , a. blais , l. frunzio , s. m. girvin , and r. j. schoelkopf , nature * 460 * , 240 ( 2009 ) . j. m. chow , a. d. crcoles , j. m. gambetta , c. rigetti , b. r. johnson , j. a. smolin , j. r. rozen , g. a. keefe , m. b. rothwell , m. b. ketchen , and m. steffen , phys . lett . * 107 * , 080502 ( 2011 ) . a. dewes , f. r. ong , v. schmitt , r. lauro , n. boulant , p. bertet , d. vion , and d. esteve , phys . lett . * 108 * , 057002 ( 2012 ) . j. m. chow , j. m. gambetta , a. d. crcoles , s. t. merkel , j. a. smolin , c. rigetti , s. poletto , g. a. keefe , m. b. rothwell , j. r. rozen , m. b. ketchen , and m. steffen , phys . rev * 109 * , 060501 ( 2012 ) . l. dicarlo , m. d. reed , l. sun , b. r. johnson , j. m. chow , j. m. gambetta , l. frunzio , s. m. girvin , m. h. devoret , and r. j. schoelkopf , nature * 467 * , 574 ( 2010 ) ; m. neeley , r. c. bialczak , m. lenander , e. lucero , m. mariantoni , a. d. oconnell , d. sank , h. wang , m. weides , j. wenner , y. yin , t. yamamoto , a. n. cleland , and j. m. martinis , nature * 467 * , 570 ( 2010 ) . a. p. peirce , m. a. dahleh , and h. rabitz , phys . a * 37 * , 4950 ( 1988 ) . d. j. tannor , v. a. kazakov and v. orlov , in _ time - dependent quantum molecular dynamics _ , edited by j. broeckhove and l. lathouwers , nato advanced studies institute , series b : physics ( plenum press , new york , 1992 ) , vol . 347 - 360 . j. p. palao and r. kosloff , phys . lett . * 89 * , 188301 ( 2002 ) ; j. p. palao and r. kosloff , phys . a * 68 * , 062308 ( 2003 ) . n. khaneja , t. reiss , c. kehlet , t. schulte - herbrggen , and s. j. glaser , j. magn . reson . * 172 * , 296 ( 2005 ) ; d .- b . tsai , p .- w . chen , and h .- s . goan , phys . a * 79 * , 060306(r ) ( 2009 ) . a. sprl , t. schulte - herbrggen , s. j. glaser , v. bergholm , m. j. storcz , j. ferber , and f. k. wilhelm , phys . a * 75 * , 012302 ( 2007 ) . r. eitan , m. mundt , and d. j. tannor , phys . a * 83 * , 053426 ( 2011 ) . m. wenin and w. ptz , phys . a * 78 * , 012358 ( 2008 ) ; m. wenin and w. ptz , phys . b * 78 * , 165118 ( 2008 ) ; m. wenin , r. roloff , and w. ptz , j. appl . phys . * 105 * , 084504 ( 2009 ) . r. roloff and w. ptz , phys . b * 79 * , 224516 ( 2009 ) ; m. wenin and w. ptz , phys . rev . a * 74 * , 022319 ( 2006 ) . h. jirari , europhys . lett . * 87 * , 40003 ( 2009 ) . p. aliferis and j. preskill , phys . a * 79 * , 012332 ( 2009 ) . d. s. wang , a. g. fowler , and l. c. l. hollenberg , phys . a * 83 * , 020302(r ) ( 2011 ) . a. g. fowler , a. c. whiteside , and l. c. l. hollenberg , phys . lett . * 108 * , 180501 ( 2012 ) , y .- x . liu , j. q. you , l. f. wei , c. p. sun , and f. nori , phys . lett . * 95 * , 087001 ( 2005 ) . f. deppe , m. mariantoni , e. p. menzel , a. marx , s. saito , k. kakuyanagi , h. tanaka , t. meno , k. semba , h. takayanagi , e. solano , and r. gross , nat . phys . * 4 * , 686 ( 2008 ) . h. j. carmichael , _ statistical methods in quantum optics 1 _ ( springer , berline , 1999 ) . c. w. gardiner and p. zoller , _ quantum noise _ , 2nd edition . ( springer - verlag , berlin , 2000 ) . breuer and f. petruccione , _ the theory of open quantum systems _ ( oxford university press , oxford , 2002 ) . d. j. egger , and f. k. wilhelm , arxiv:1402.7193 . e. magesan , j. m. gambetta , b. r. johnson , c. a. ryan , j. m. chow , s. t. merkel , m. p. da silva , g. a. keefe , m. b. rothwell , t. a. ohki , m. b. ketchen , and m. steffen , phys . 109 * , 080505 ( 2012 ) ; e. magesan , j. m. gambetta , and j. emerson , phys . rev . * 106 * , 180504 ( 2011 ) . j. m. chow , j. m. gambetta , l. tornberg , j. koch , l. s. bishop , a. a. houck , b. r. johnson , l. frunzio , s. m. girvin , and r. j. schoelkopf , phys . lett . * 102 * , 090502 ( 2009 ) ;

uv and optical photometry of type ia supernovae ( sne ia ) at low redshift have revealed the existence of two distinct color groups , nuv - red and nuv - blue events . the color curves differ primarily by an offset , with the nuv - blue color curves bluer than the nuv - red curves by 0.4 mag . for a sample of 23 low- sne ia observed with _ swift _ , the nuv - red group dominates by a ratio of 2:1 . we compare rest - frame uv / optical spectrophotometry of intermediate and high- sne ia with uvot photometry and _ hst _ spectrophotometry of low- sne ia , finding that the same two color groups exist at higher- , but with the nuv - blue events as the dominant group . within each red / blue group , we do not detect any offset in color for different redshifts , providing insight into how sn ia uv emission evolves with redshift . through spectral comparisons of sne ia with similar peak widths and phase , we explore the wavelength range that produces the uv / opt color differences . we show that the ejecta velocity of nuv - red sne is larger than that of nuv - blue objects by roughly 12% on average . this velocity difference can explain some of the uv / optical color difference , but differences in the strengths of spectral features seen in mean spectra require additional explanation . because of the different colors for these groups , nuv - red sne will have their extinction underestimated using common techniques . this , in turn , leads to under - estimation of the optical luminosity of the nuv - blue sne ia , in particular , for the high - redshift cosmological sample . not accounting for this effect should thus produce a distance bias that increases with redshift and could significantly bias measurements of cosmological parameters .

1408.1706

type ia supernovae ( sne ia ) are very luminous with a well - understood variation of their peak optical luminosities as a function of their light - curve shapes , the width - luminosity relation ( wlr ; @xmath3m@xmath4(b ) : phillips et al . 1993 ; @xmath3 : riess et al . ( 1996 ) , jha et al . ( 2006a ) ; stretch(s ) : perlmutter et al . ( 1997 ) ; sifto : conley et al . ( 2008 ) ) . the empirically derived width - luminosity relation has allowed sne ia to be utilized to study the expansion of the universe , revealing that the expansion is currently experiencing a period of acceleration ( riess et al . 1998 ; perlmutter et al . 1999 ) . in order to lower the scatter in distance determinations , the larger sample of all sne ia is reduced to a group of so - called `` normal '' sne ia , a group that follow a regular variation of their properties with the width of the peak of their optical light curves ( i.e. , scaling ) . the reduced sample eliminates narrow - peaked sne ia ( see taubenberger et al . 2008 ; krisciunas et al . 2009 ) , sne iax ( li et al . 2003 ; foley et al . 2013 ) , candidates for super - chandrasekhar sne ia ( e.g. , howell et al . 2006 ; scalzo et al . 2010 ) and individually anomalous events ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the reduced sample comprises roughly half of the sne ia discovered , and will be the focus of this work . a key element of the cosmological utilization of sne ia is the assumption that the sn ia phenomenon does not change as the universe has aged , or that any changes continue to follow the empirical laws derived for nearby normal sne ia . for example , the method can work if higher-@xmath1 sne ia tend to have broader optical light curves , as long as the luminosities of those sne still follow the wlr for that peak width . the problem would arise if higher-@xmath1 sne ia had brighter or fainter luminosities for the same light - curve shape . an example of one such problem was reported by sullivan et al . ( 2010 ) , who claim a dependence of sn ia luminosity on the mass and specfic star formation rates of the host galaxy . as these characteristics vary with redshift , a systematic error is introduced to cosmological analyses at the 0.08 magnitude level . spectral modeling makes it clear that iron - peak elements significantly impact the ultraviolet ( uv ) spectral - energy distribution ( sed ) of sne ia ( e.g. , * ? ? ? * ; * ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? theoretical models of the thermonuclear explosion of a sn ia have explored the effects of varying the progenitor metallicity or explosion physics on the resulting @xmath5ni production and distribution , as well as on the expansion velocities ( hflich et al . 1998 ; lentz et al . 2000 ; timmes et al . 2003 ; sauer et al . 2008 ; walker et al . all simulations suggest that the uv emission is much more strongly affected by metallicity variations than the optical emission , and would lead to variations in the uv - optical colors of sne ia with different progenitor metallicities . since the progenitor metallicity is one aspect of a sn ia explosion that would be expected to change with redshift , it is vital to establish the nature of the progenitor metallicity uv sed correlation . however , there is not a consensus about the magnitude of that correlation , underscoring how complicated the challenge is to model the complete time - evolving spectrum of a sn ia . further , as there are multiple classes of progenitor systems suggested ( e.g. , single versus double degenerate ) , as well as multiple explosion scenarios ( e.g. , delayed detonations versus deflagrations versus double detonations ) , it is not clear that progenitor metallicity differences will be the dominant factor in variations of the uv - optical sed @xcite . with all the challenges facing theoretical modeling , observational efforts to characterize the time evolution of the uv - optical sed for a large sample of sne ia might guide the theoretical investigations . the importance of uv emission as a probe of the sn ia event has been recognized for decades , but it has been challenging to perform rest - frame uv observations . efforts to observe higher-@xmath1 sne ia has increased in recent years with many large telescopes being utilized . ellis et al . ( 2008 ) used the lris spectrograph on the keck i telescope to observe supernova legacy survey ( snls ) sne ia , using snls photometry to remove host galaxy light from the spectra ( hereafter e08 ) . e08 found that the dispersion of individual spectra about a mean mid-@xmath1 spectrum increased in the uv . the mean spectrum was not found to have evolved compared to a low-@xmath1 mean spectrum . foley et al . ( 2008 ) performed an independent study that used a collection of large telescopes to observe essence mid-@xmath1 sne ia , also finding that the spectral dispersion increased to the uv . however , uncertain galaxy contamination correction prevented a robust comparison of spectral continua . foley et al . ( 2012 ) employed the lris spectrograph on keck to observe 21 mid-@xmath1 sne ia from the sloan digital sky survey supernova survey ( hereafter f12 ) . f12 was able to remove galaxy contamination from the spectra by modeling the galaxy sed from sdss photometry , and used a low-@xmath1 sample of optical sn ia spectra @xcite combined with a sample of 46 low-@xmath1 uv spectra @xcite . with this high - quality data set , f12 found that while the optical spectra were nearly identical , the uv portion of a mean spectrum of the mid-@xmath1 keck / sdss sample had excess flux relative to a low-@xmath1 mean spectrum . sullivan et al . ( 2009 ) compared a low-@xmath1 mean spectrum , the e08 mid-@xmath1 mean spectrum , and a high-@xmath1 [ @xmath6 sne with _ hst_/acs spectra ; riess et al . ( 2007 ) ] mean spectrum . there was an indication for uv flux excess for the higher redshift samples , but this was discounted because of the relatively small low-@xmath1 sample used in the analysis . cooke et al . ( 2011 ) and maguire et al . ( 2012 ) expanded the low-@xmath1 sample , paying careful consideration to selection effects ; again , there were differences in the uv continuum . a study by balland et al . ( 2009 ) used the eso / vlt telescope to observe snls sne ia , using cfht photometry to remove host galaxy light from the spectra ( hereafter b09 ) . that work concentrated on the variation of spectral features with redshift , finding that higher-@xmath1 spectra have shallower intermediate mass element ( ime ) absorption features , but this was attributed to different color distributions for the samples , as the stretch values were similar given the uncertainties . collectively , these studies show that there is little evolution of the optical emission with redshift , but more in the uv . the direction of that evolution is for the higher-@xmath1 sne ia to be brighter in the uv . at a given redshift , the uv emission exhibits increased dispersion compared to the optical wavelength range . one interpretation of these findings has been improved confidence for the cosmological utility of sne ia in the optical wavelength range , combined with skepticism for the cosmological utility of the uv wavelength range . in section 2 , we compare the @xmath0 colors of low-@xmath1 versus mid- and high-@xmath1 photometry and spectrophotometry . in section 3 , we compare paired spectra from each group . in section 4 , we compare @xmath7 and @xmath2 colors , with section 5 devoted to exploring the cosmological implications of the optical color differences . section 6 summarizes these findings and argues for altering light curve fitting routines to account for the existence of two color groups . the uvot instrument on the _ swift _ satellite is executing a long - term program of observing many nearby sne ia in the uv and optical ( brown et al . 2009 , 2010 , 2012 ; milne et al . 2010 , 2013 ; foley et al . 2012a ; wang et al . 2009a ; bufano et al . 2009 ) . with thousands of epochs of observations of more than 50 sne ia , the uv emission from a large sample of events is being studied to search for universal characteristics as well as for patterns in the variations that are seen . milne et al . ( 2013 ) found that the uv - optical colors of normal sne ia evolve dramatically with epoch and that there exist two major color groups within the subset of normal sne ia . the larger , nuv - red group contained roughly two - thirds of the sample , while a second nuv - blue group was offset from the nuv - red group , evolving bluer by 0.4 mag in @xmath0 ( figure [ u_v_color_curve ] ) . the grouping was found to not correlate with peak - width . although the differences were seen in all uv filters , in this work we will concentrate on the @xmath0 , @xmath7 and @xmath2 colors . the irregular " and muv - blue " minor groups discussed in m13 will be treated as nuv - red in this work . the irregular " group was concluded to be the optically broad - peaked subset of the nuv - red group , and the muv - blue " group had colors similar to nuv - red sne in the @xmath0 and @xmath8 filter pairs . since rest - frame @xmath9 and @xmath10 band emission can be observed in the optical for sne ia at redshifts of @xmath11 , we utilize four studies that have obtained spectra that span rest - wavelength range 30005700 . e08 utilized the low resolution imaging spectrometer ( lris ) on the keck i telescope to obtain 36 host - galaxy subtracted spectra of 36 sne ia at 0.2@xmath12@[email protected] . twenty - three of those spectra span 30005700 and are included in this work . the fully - reduced and subtracted spectra were obtained from the wiezmann interactive supernova data repository , wiserep ( http://www.weizmann.ac.il/astrophysics/wiserep/ ) . b09 utilized the fors1 and fors2 spectrometers to obtain 139 host - galaxy subtracted spectra of 124 snls sne ia . ten of these spectra span 30005700 and are included in this work . the 10 spectra are in the redshift range , 0.38@xmath12@[email protected] . the fully - reduced and subtracted snonly " spectra were obtained from the wiserep site . f12 utilized the lris spectrometer on the keck i telescope to obtain single - epoch spectra of host - galaxy subtracted spectra of 21 sne ia at 0.11@xmath12@[email protected] . nineteen of those spectra span 30005700 and the fully - reduced and subtracted spectra are included in this work . no spectra were rejected based upon reddening selection criteria , meaning that any reddening bias would have to have been employed by the sample selection of the e08 , b09 , f12 and r07 publications . the uvot-@xmath10 filter has some transmission redward of 5700 , but we make no correction for the range of maximum wavelengths ; the maximum wavelength is simply required to be 5700 . the mid-@xmath1 spectra were folded through the uvot @xmath9 and @xmath10 transmission curves to generate spectrophotometry ( tables 1 & 2 ) , utilizing the zero - points from breeveld et al . filters , the breeveld et al . ( 2012 ) zeropoints are the same as poole et al . ( 2008 ) . ] figure [ u_v_color_curve ] shows the spectrophotometry compared to the uvot photometry for the low-@xmath1 sample presented in m13 . the mid-@xmath1 data overlaps the uvot photometry well , and exhibits a separation into two groups as seen in the uvot photometry . interestingly , whereas two - thirds of the uvot sample were nuv - red sne ia , there are more nuv - blue sne ia in the mid-@xmath1 sample . a linear fit to the color evolution of the uvot nuv - red data for epochs @xmath13 @xmath12 5 days allows color evolution with time to be removed , leaving only residual magnitudes relative to the fit . figure [ u_v_histo ] shows a histogram of the residuals , separating the uvot data into nuv - red / blue groups , and leaving the mid-@xmath1 data as a single collection . the nuv - red / blue separation is clear from the histogram , and is reproduced by the mid-@xmath1 sample . the two peaks of the two mid-@xmath1 sample align with the two uvot peaks , suggestive of similar colors within each group . if there is no systematic , erroneous color offset between the derivation of photometry from the low-@xmath1 and mid-@xmath1 samples , the alignment of peaks is suggestive of a lack of color evolution with redshift within the respective red and blue groups . the mean colors of the samples are shown in table 3 , showing that there is a consistent separation between the nuv - red and -blue groups , and that the mean colors within a group exhibit no change in color with redshift . l|c|cccc|cc|c|ccc sn & sample@xmath14 & epoch & @xmath1 & s/@xmath15 & e(b - v)/*c*@xmath16 & @xmath17(min ) & @xmath17(max ) & v(siii)@xmath18 & @xmath0 & @xmath7 & @xmath2 + name & & [ days]@xmath19 & [ km / s ] & & [ mag ] & [ mag ] & [ ang.]@xmath20 & [ ang.]@xmath21 & [ mag]@xmath22 & [ mag]@xmath22 & [ mag]@xmath22 + + 03d1au & e08 & -1.6 & 0.504 & 1.13(02 ) & * 0.02(03 ) * & 2476.2 & 5936.3 & & -0.63 & -0.56 & -0.07 + 03d1dj & e08 & -1.8 & 0.400 & * * & * * & 2675.0 & 6389.3 & & -0.63 & -0.46 & -0.16 + 03d3aw & e08 & -0.8 & 0.449 & 1.066(33 ) & * -0.06(04 ) * & 2584.5 & 6162.9 & & -0.83 & -0.48 & -0.35 + 03d3ay & e08 & -1.0 & 0.371 & 1.054(30 ) & * -0.018(36 ) * & 2728.1 & 6521.3 & & -0.54 & -0.48 & -0.06 + 03d3bh & e08 & -3.5 & 0.249 & * * & * * & 2991.4 & 7152.0 & -11.5697 & -0.73 & -0.65 & -0.08 + 03d3cc & e08 & 9.1 & 0.463 & * * & * * & 2567.2 & 6122.2 & & 0.13 & -0.18 & 0.31 + 03d3cd & e08 & -6.2 & 0.461 & 1.19(09 ) & * -0.00(05 ) * & 2567.3 & 6130.6 & & -0.84 & -0.74 & -0.10 + 03d4ag & e08 & -5.3 & 0.285 & 1.09(02 ) & * -0.03(02 ) * & 2911.2 & 6974.4 & -9.44703 & -0.93 & -0.82 & -0.12 + 03d4cj & e08 & -7.4 & 0.270 & 1.12(01 ) & * -0.05(02 ) * & 2944.9 & 7047.2 & & -0.83 & -0.69 & -0.14 + 04d1hd & e08 & 1.7 & 0.369 & 1.07(01 ) & * -0.06(02 ) * & 2736.0 & 6527.6 & -11.1098 & -0.57 & -0.55 & -0.02 + 04d1pg & b09 & -1.3 & 0.515 & 1.10(02 ) & * 0.12(04 ) * & 2811.0 & 5860.5 & & -0.62 & -0.47 & -0.14 + 04d1rh & e08 & 2.7 & 0.435 & 1.11(03 ) & * -0.02(03 ) * & 2599.5 & 6226.9 & & -0.43 & -0.47 & 0.04 + 04d2fp & b09 & 1.8 & 0.415 & 1.03(02 ) & * -0.00(03 ) * & 3010.5 & 6296.7 & & -0.38 & -0.30 & -0.08 + 04d2gc & e08 & -0.2 & 0.522 & 1.12(03 ) & * 0.05(04 ) * & 2457.9 & 5875.4 & & -0.44 & -0.51 & 0.07 + 04d4 in & e08 & -5.1 & 0.516 & 1.17(02 ) & * -0.04(03 ) * & 2463.7 & 5890.5 & & -0.93 & -0.82 & -0.11 + 04d4jr & b09 & -5.9 & 0.470 & 1.15(02 ) & * -0.02(03 ) * & 2896.9 & 6039.7 & & -0.68 & -0.60 & -0.08 + 04d4jr & e08 & -0.1 & 0.482 & 1.15(02 ) & * -0.02(03 ) * & 2520.2 & 6029.0 & & -0.67 & -0.57 & -0.10 + 2005ik & f12 & 3.0 & 0.320 & 0.872(68 ) & 0.03(10 ) & 2521.9 & 7001.9 & & -0.33 & -0.45 & 0.12 + 2005ix & f12 & 2.9 & 0.260 & 0.971(39 ) & 0.01(01 ) & 2740.5 & 7335.3 & -13.1845 & -0.44 & -0.39 & -0.05 + 2005jc & f12 & 0.8 & 0.215 & 0.995(27 ) & 0.12(04 ) & 2739.9 & 7607.0 & -10.6360 & -0.38 & -0.33 & -0.06 + 2005jd & f12 & 2.7 & 0.320 & 1.090(52 ) & 0.01(01 ) & 2659.9 & 7001.9 & -10.8679 & -0.36 & -0.37 & 0.01 + 2005ji & f12 & 1.0 & 0.214 & * -0.07(09 ) * & 0.01(02 ) & 2775.1 & 7613.3 & -9.82174 & -0.62 & -0.49 & -0.12 + 2005jl & f12 & -0.1 & 0.180 & 1.003(32 ) & 0.10(03 ) & 2783.9 & 7832.7 & -10.2749 & -0.69 & -0.68 & -0.01 + 2005jn & f12 & 3.3 & 0.330 & 1.025(56 ) & 0.01(01 ) & 2557.1 & 6949.3 & -8.84449 & -0.73 & -0.74 & 0.01 + 2005jo & f12 & -1.0 & 0.230 & 1.055(39 ) & 0.14(07 ) & 2932.5 & 7514.3 & -9.54432 & -0.54 & -0.35 & -0.20 + 05d1ix & e08 & -9.1 & 0.490 & 1.06(02 ) & * -0.04(03 ) * & 2503.4 & 6003.4 & & -0.70 & -0.48 & -0.22 + 05d1hk & e08 & -8.6 & 0.263 & 1.15(02 ) & * 0.01(03 ) * & 2957.0 & 7065.9 & & -0.75 & -0.78 & 0.03 + 05d1iy & e08 & -8.1 & 0.248 & * * & * * & 2997.3 & 7164.6 & & -0.61 & -0.75 & 0.14 + 05d2ac & b09 & 2.0 & 0.479 & 1.11(02 ) & * -0.01(03 ) * & 2879.5 & 6003.2 & & -0.20 & -0.31 & 0.11 + 05d2bv & b09 & -0.1 & 0.474 & 0.99(01 ) & * -0.09(03 ) * & 2889.6 & 6023.9 & & -0.63 & -0.46 & -0.18 + 05d2dw & b09 & -5.3 & 0.417 & 1.11(02 ) & * 0.02(03 ) * & 3005.3 & 5736.4 & & -0.78 & -0.53 & -0.25 + 05d4cw & b09 & 7.0 & 0.375 & 0.91(01 ) & * -0.10(03 ) * & 3097.4 & 6457.4 & & -0.15 & -0.15 & -0.00 + 2006pf & f12 & 2.0 & 0.366 & * -0.32(10 ) * & 0.01(01 ) & 2358.6 & 6775.4 & & -0.61 & -0.82 & 0.21 + 2006pq & f12 & -4.4 & 0.193 & * -0.36(13 ) * & 0.04(04 ) & 2651.2 & 7755.7 & -10.8753 & -0.77 & -0.66 & -0.11 + 2007lu & f12 & -0.1 & 0.319 & * -0.47(07 ) * & 0.17(04 ) & 2434.4 & 6975.4 & -11.7068 & -0.41 & -0.51 & 0.10 + 2007lw & f12 & 1.5 & 0.290 & * -0.23(09 ) * & 0.01(02 ) & 2569.7 & 7130.6 & -10.8118 & -0.54 & -0.46 & -0.09 + 2007qu & f12 & -1.2 & 0.310 & * -0.25(14 ) * & 0.01(01 ) & 2466.4 & 7009.6 & -10.4310 & -0.77 & -0.79 & 0.02 + [ cols= " < " , ] [ tab4 ] one alternative explanation to nuv - blue sne ia being the dominant explosion variety at higher redshifts is that the nuv - blue events are more luminous and are over - represented in a magnitude limited sample . there are a number of factors that argue against that explanation . first , none of the samples were rest - frame @xmath9-band dominated when determining candidates , so a luminosity bias would only occur if nuv - blue sne were optically bright . the fact that this separation was not recognized years ago argues against a difference of more than 0.1 - 0.2 mag . further , e08 and f12 studied magnitude selection effects and determined that their samples were not strongly biased . second , all three mid-@xmath1 surveys showed the same nuv - blue dominance , despite the differences between the sample selection for each survey . third , even if there were a luminosity bias of 0.3 mag , the effect would be much too small to shift the nuv - blue : red ratio from 40% to 90% . we conclude that there is a real variation in the nuv - blue to nuv - red ratio with redshift . using the same approach as employed for the low-@xmath1 _ hst _ sn spectra in m13 , we compare spectra of nuv - red / blue pairs . for two sne to be considered `` pairs , '' we require that they have similar optical light - curve shapes ( in the same third of the sample ) , have similar phases ( within 1 day excepting one pair ) , and have _ different _ uv colors . selecting the best matches that meet these requirements allow a direct comparison of sn spectra which largely removes possible differences attributable to light - curve shape ( and thus luminosity ) and phase ; therefore , any differences in the spectra are more likely to be related to the differences in uv colors . we present these spectra pairs in figure [ compspec_mixed ] . as expected , the spectra of sne from the nuv - blue group have a higher uv flux than those of sne from the nuv - red group . interestingly , there is no noticeable difference in the optical continua ( @xmath23 ) of sne from the two groups . this is perhaps not unexpected given the similar optical colors for all sne . but it indicates that this effect is not caused by odd reddening , which would have some subtle difference in the optical continuum , or the confluence of a strong continuum difference offset by differences in spectral features . also notable is that the same uv spectral features are present for both groups . although the strengths of these lines are different for each sn , it is clear that the uv color difference is the result of a difference in a wide band in the uv wavelength range and not the presence / absence of particular lines or dramatically different line strengths ( and unaffected uv continua ) in one particular group . the final clear difference in the two groups is the velocity of multiple spectral features . when the @xmath24 feature is observed with a reasonable signal - to - noise ratio , the nuv - red sn often has a broader , higher - velocity line than the nuv - blue paired sn ( and the nuv - blue spectrum never has a higher - velocity feature for our pairs ) . this is sometimes easier to see in the residual spectra , which shows a positive bump at @xmath25 6100 , indicating excess absorption in the nuv - red spectra . this is perhaps best seen in the comparison of sne 2006pq and 2007ml . since the sne have similar light - curve shapes and the spectra are at similar phases , we conclude that the velocity of the @xmath24 feature is linked to the uv colors . we note that the trend with velocity is not exclusive to ; in particular , h&k shows similar behavior . foley & kasen ( 2011 ) first showed that the intrinsic @xmath26 color of a sn ia was strongly linked to the velocity of the @xmath24 feature near maximum light . the physical explanation for this effect is that higher - velocity spectral features are also broader , creating additional opacity in the blue where line blanketing dominates , while further to the red , electron scattering dominates the opacity , and thus the broader lines do not increase the overall opacity for those wavelengths . this interpretation predicts that other than the obvious velocity differences , spectral differences between low and high - velocity sne should be largely constrained to the bluer wavelengths , that the differences should largely be a difference in the continuum , and that the difference should persist to the uv . figure [ compspec_mixed ] supports this interpretation , and the differences in uv colors may predominantly be caused by different opacities resulting from different ejecta velocities . with a larger uvot sample , an interesting comparison will be between low velocity ( lv ) nuv - blue sne and lv nuv - red sne . if the velocity distributions are similar between those groups , the prediction would be that the @xmath2 colors would be similar . thomas et al . ( 2011 ) reported a correlation between uvot nuv - blue sne ia and the detection of an absorption feature at @xmath17 6580 attributed to . m13 confirmed the correlation for all seven uvot nuv - blue sne in the sample . since the @xmath17 6580 feature is only detected at early epochs ( @xmath27 days ) in normal sne ia , one would only expect to detect this feature in early - epoch spectra that reach to 6650 . six mid-@xmath1 spectra meet those criteria . the comparison spectra of the nuv - blue sn 2006pq shows a notch at the location of the feature , when compared with the nuv - red sn 2007ml , but in general the spectra either cut - off before the feature , were of too low s / n , or were obtained at too late an epoch for identification . there are therefore too few sne from which we can draw strong conclusions . however , several groups have found that high - velocity sne ia , which we associate with nuv - red objects , are far less likely to have detectable ( parrent et al . 2011 ; folatelli et al . 2011 ; silverman & filippenko 2012 ) , consistent with the findings of m13 . f12 generated mean spectra of low-@xmath1 and mid-@xmath1 samples and determined that when normalized to the optical emission , the mid-@xmath1 mean spectrum featured a uv excess , the same result that was found by maguire et al . ( 2012 ) . recognizing the increased fraction of nuv - blue events at mid-@xmath1 , and the similarities of the distributions of low-@xmath1 and mid-@xmath1 colors within the red / blue groups ( shown in figure [ u_v_histo ] ) , the comparison of mean spectra can be revisited with the additional separation between nuv - red and -blue sne . we note that several of the spectra used here were also used by f12 . we used the methods of foley et al . ( 2008 ) and f12 to generate mean spectra and uncertainty spectra corresponding to the boot - strap resampling uncertainty . using only spectra near peak brightness ( @xmath28 t@xmath29 @xmath28 @xmath12 7 days ) and normalizing in the rest - frame wavelength region @xmath30 , mean spectra for the low-@xmath1 sample @xcite and mid-@xmath1 nuv - red and nuv - blue samples are presented in figures [ highz_red_vs_lowz ] to [ highz_red_blue ] . comparing the mean mid-@xmath1 nuv - red and low-@xmath1 mean spectra ( figure [ highz_red_vs_lowz ] ) , we see that the spectra are nearly identical . there are two clear differences : the mid-@xmath1 nuv - red has broader and higher - velocity @xmath176355 and ca h&k features . the difference in the ca h&k feature is exclusively to the blue portion of the complex feature @xcite , which has a significant contribution from @xmath173727 ( but also see childress et al 2013 ) . as the low-@xmath1 sample has both nuv - red and nuv - blue sne , one can extrapolate from the previous comparison of pairs that the relatively small difference in velocity is the result of the nuv - blue sne in the low-@xmath1 sample . another difference in the mid-@xmath1 nuv - red and low-@xmath1 mean spectra is the strength of the 4800 feature , with the mid-@xmath1 nuv - red mean spectrum having a weaker absorption than that of the low-@xmath1 mean spectrum . @xcite reported a similar difference in a comparison of low-@xmath1 and mid-@xmath1 mean spectra . this feature is predominantly caused by absorption with additional absorption from and . because of the blending of this feature , there are multiple plausible explanations for such a difference ranging from excitation / temperature differences to differences in composition . examining figure [ highz_blue_vs_lowz ] , we see more dramatic differences between the mid-@xmath1 nuv - blue and low-@xmath1 mean spectra . the mean mid-@xmath1 nuv - blue spectrum has lower velocity features , a uv excess , and weaker spectral features compared to the mean low-@xmath1 spectrum . specifically , the mid-@xmath1 nuv - blue mean spectrum has excess flux compared to the low-@xmath1 mean spectrum for wavelengths shorter than @xmath25 4000 . this is consistent with what has been seen in other comparisons ( cooke et al . 2011 , maguire et al . 2012 , f12 ) and supports the suggestion that the spectral evolution with redshift is dominated by the change in membership from nuv - red events at low-@xmath1 to nuv - blue events at high-@xmath1 . most spectral features in both the uv and optical , including the ca h&k feature , the `` 4300 '' feature , and the 4800 feature , are weaker in the mid-@xmath1 nuv - blue mean spectrum . in retrospect , a similar behavior can be seen between the intermediate - redshift and low - redshift mean spectra of f12 . considering that many of the same spectra contribute to the f12 mean spectra and the mean spectra presented here , this observation is reassuring . intriguingly , the lower ( @xmath31 ) and higher ( @xmath32 ) redshift subsamples presented by f12 also show this trend with the higher redshift mean spectrum having muted features relative to the lower redshift mean spectrum . we compare the mean mid-@xmath1 nuv - blue and nuv - red spectra in figure [ highz_red_blue ] ; as expected , there are significant spectral differences : the nuv - blue spectrum has lower velocity features , a uv excess , and weaker spectral features . since the mid-@xmath1 nuv - red mean spectrum was similar to the low-@xmath1 mean spectrum , most of the above discussion between the mid-@xmath1 nuv - blue mean spectrum and the low-@xmath1 mean spectrum also applies to this comparison . the wavelength of maximum absorption for the @xmath176355 feature corresponds to a blueshifted velocity , @xmath33 , of @xmath34,@xmath35 @xmath36and @xmath37,@xmath38 @xmath36 , for the nuv - blue and nuv - red mean spectra , respectively . for a subset of the individual mid-@xmath1 spectra , we were able to measure @xmath33 , and its value at @xmath39-band maximum brightness , @xmath40 @xcite . the nuv - blue and nuv - red samples had average @xmath40 of @xmath34,630 and @xmath41,850 @xmath36 , respectively . similarly , the two groups had median values of @xmath34,620 and @xmath41,740 @xmath36 , respectively . the velocity difference in the mean spectra are duplicated in measurements of individual objects . all measurements suggest that the nuv - red sample has an average ejecta velocity that is @xmath2512% larger than that of the nuv - blue sample.6355 line velocities are all blueshifted , and thus negative , but above " and below " will be relative to the absolute magnitude of the velocities . ] of the 15 nuv - blue sne for which we could determine @xmath40 , 14 have velocities below @xmath42 @xmath36 , the nominal separation between `` high - velocity '' and `` low - velocity '' sne ia @xcite . in contrast , 4 of the 8 nuv - red sne for which we could measure @xmath40 had a velocity higher than @xmath42 @xmath36 . m13 found similar results for the low-@xmath1 sample : the nuv - blue objects exclusively had low - velocity ejecta , while half of the nuv - red objects were found in each group . performing a kolmogorov smirnov test , we find that the nuv - blue and nuv - red sne likely have different parent populations with respect to ejecta velocity ( @xmath43 ) . by contrast , performing a kolmogorov smirnov test on the stretch values , we find that the nuv - blue and nuv - red sne likely have similar distributions of stretch values ( @xmath44 ) . we also examined the pseudo - equivalent width ( pew ) of the @xmath45 feature , which has been proposed as an indicator of velocity and color , and is a possible way to further reduce hubble scatter ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? specifically , sne with higher @xmath46 pew tend to have higher maximum - light velocities and redder colors , as one would expect given the velocity - color relation @xcite . for the low-@xmath1 , nuv - blue , and nuv - red composite spectra , we measure pews of 8.0 , 6.4 , and 11.4 , respectively , with an estimated systematic uncertainty of @xmath25 2.0 that greatly dominates over any statistical uncertainty . this measurement is consistent with nuv - blue sne being intrinsically bluer and having lower eject velocities than nuv - red sne . reconcentrating on the general features of the mean spectra , the mid-@xmath1 nuv - blue and nuv - red mean spectra clearly differ in the nuv when normalized in the optical wavelength range . the differences invite the question as to whether the higher dispersion reported in mean uv spectra is wholly due to failing to separate the sne ia into red and blue groups . comparing the dispersion in an optical band ( 45005500 ) with a nuv band ( 30003500 ) for the low-@xmath1 mean spectrum , we find the nuv dispersion to be 4.2 times the optical dispersion . by constrast , that ratio is 2.8 for the mid-@xmath1 nuv - blue mean spectrum and 2.7 for the mid-@xmath1 nuv - red mean spectrum . stated a different way , the nuv - blue and nuv - red mean spectra retain 68% and 63% of the nuv dispersion after separation compared to the low-@xmath1 ( unseparated ) mean spectrum . this suggests that there remains intrinsic dispersion within each group that is independent of nuv - red / blue group membership , but that recognizing the group membership makes the nuv wavelength range closer to being useful relative to the optical wavelength range . utilizing the @xmath0 colors to determine nuv - red versus nuv - blue membership allows the differences in the @xmath7 and @xmath2 colors to be studied . figure [ u_b_color_curve ] shows the @xmath7 color curves of the mid-@xmath1 sample compared to uvot photometry . the figure is similar to figure [ u_v_color_curve ] , but the mid-@xmath1 spectrophotometry has been color - coded red or blue based upon the nuv - red / blue determinations shown in table 1 . a linear fit to the nuv - red uvot photometry is shown as a dashed line , as was performed in m13 . the line was fitted for epochs @xmath28 t @xmath28 @xmath12 6 and offset to the blue by 0.2 magnitudes so that there is a sign change between nuv - red / blue groups . histograms of the residuals to that fit are shown in figure [ u_b_histo ] . there is still a color difference between the two groups , but it is less than for the @xmath0 colors . table 2 shows the lower separation between nuv - red / blue , which lowers from 0.42 and 0.57 mag to 0.34 and 0.38 mag , respectively for the low-@xmath1 and mid-@xmath1 samples , but again shows no evolution of color with redshift within a group . the same method can be employed to study @xmath2 colors . figures [ b_v_color_curve ] and [ b_v_histo ] show the color curves and resulting histograms for the @xmath2 colors . a quadratic fit is utilized for the @xmath2 photometry , using data during the @xmath28 t @xmath28 @xmath12 10 day epoch . the nuv - red and nuv - blue distributions are less separated , with overlap for both samples . table 2 again shows the lower separation between nuv - red / blue , which lowers to 0.10 and 0.17 magnitudes for the low-@xmath1 and mid-@xmath1 samples , but yet again shows no evolution of color with redshift within a group . the color difference between the @xmath2 colors of the nuv - red and -blue groups at both redshifts are potentially problematic for the cosmological utilization of sne ia as distance indicators . the optical colors , particularly the @xmath47 color , is used to estimate the host - galaxy extinction to the sn . current implementations of @xmath47 colors do not differentiate nuv - red from nuv - blue sne ia . as these methods are trained wholly or primarily on nearby , low-@xmath1 sne , they are likely to be dominated by nuv - red sne ia . an algorithm so trained would interpret the bluer @xmath47 color of nuv - blue events to imply very low , even negative extinction . if there really were host - galaxy extinction along the line of sight to that sn , the under - estimation of that extinction would lead to the erroneous impression that the sn is less luminous than it really is . the change in the nuv - red / blue ratio in favor of nuv - blue events , combined with the under - estimation of extinction for nuv - blue events means that many high-@xmath1 sne ia could be systematically more luminous than recognized . it is thus critical to determine if there are differences between the absolute magnitudes of nuv - blue sne ia and nuv - red sne ia . with the recognition of the existence of two groups , it becomes critical to accumulate a sample of nearby nuv - blue and nuv - red sne ia for which the absolute magnitudes can be determined . the lack of color changes with redshift within each group suggests that a new generation of extinction estimation based on low-@xmath1 sne ia would be valid at high-@xmath1 . without absolute magnitudes determinations that separate nuv - blue from nuv - red sne ia , we do not know the exact bias that results from treating both groups as a single sample . however , making the assumption that the absolute optical magnitudes are the same , we can assess the impact of this potential bias . to do this , we perform two similar simulations . for both , we assume that the two groups have a difference in @xmath2 , @xmath48 , that directly impacts the distance modulus such that @xmath49 ( note that the sign of this value is set this way because we defined @xmath50 ) . the average hubble residual bias is then @xmath51 where @xmath52 and @xmath53 are the fraction of nuv - blue and nuv - red sne , and assuming that @xmath54 . we then use the fractions of nuv - blue and nuv - red sne as determined in section 2 to determine the evolution of the hubble residual bias as a function of redshift . the first simulation assumes @xmath55 or @xmath56 , the differences seen for the low-@xmath1 and mid-@xmath1 samples , respectively . we also assume that @xmath57 or @xmath58 , resulting in four distinct possibilities for the hubble residual bias . since there is no specific preference for one possibility over the others , we perform a monte carlo simulation drawing from each of the four possibilities to determine the median bias and the uncertainty on the bias . the second simulation is similar to the first , but uses @xmath59 . this is the measured @xmath60 offset between the low and high - velocity subsamples of the foley & kasen ( 2011 ) sample . as shown above , ejecta velocity is highly correlated with nuv color . although this color difference may not be exactly the difference between the nuv color subclasses , it will be illustrative because of the proportionally small uncertainty for the color difference . for this simulation , we also use both values of @xmath61 and perform a monte carlo simulation to determine the final bias and uncertainty . figure [ hr ] shows the results of the simulations . the hubble residuals shown are relative to a sample that has equal numbers of nuv - blue and nuv - red sne . both predict the hubble residual bias to change with redshift with the bias being negative at low @xmath1 and positive at high @xmath1 . this is expected since nuv - red sne , which should have their extinction overestimated and their distance moduli underestimated , are more prevalent in the local universe . the magnitude of the bias depends on the magnitude of the @xmath2 difference . the first simulation , which uses the measured difference directly from the nuv subclasses indicates a bias of @xmath62 ( @xmath63 ) between @xmath64 and @xmath65 ( @xmath66 ) , while the second simulation finds @xmath67 ( @xmath68 ) for the same redshifts . of course the second simulation has a more statistically significant bias because of the relatively smaller uncertainty for the @xmath2 difference . of particular note is that given the current uncertainties , the first simulation is consistent with zero bias for all redshifts . to investigate the potential effect of two uv - optical color groups on the @xmath69 parameter , we have modified the snana ( v10.34 ) @xcite _ snlc - sim _ routines to allow us to incorporate a filter - dependent bias in the rest - frame _ mlcs2k2 _ model @xcite of the sn . as the swift @xmath9 band is considerably more blue than johnson @xmath70 , we have elected to study the effect for a range of different values from @xmath71 mag ( no bias or nuv - red ) to @xmath72 mag ( strong blue nuv bias ) in increments of @xmath73 mag . to reduce the number of independent parameters , we related the bias in johnson @xmath39 to johnson @xmath70 via @xmath74 . as our analysis of hubble residuals previously indicated a strong trend with redshift , we have elected to simulate all sne ia in the simulation as originating from a single survey modelled after the snls , to avoid conflating redshift - dependent trends arising from different survey properties and selection effects with the signal from the uv bias . we simulate 500 sne ia for each value of the bias , in a flat cosmology with @xmath75 , @xmath76 and three different values of the dark energy equation of state , @xmath77 . additional flux smearing is used to ensure the simulated light curves reproduce the observed scatter in the hubble diagram , and the cosmological fits have reduced @xmath78@xmath25@xmath79 . all sne ia are fit with _ mlcs2k2 _ with priors appropriate for the snls . the effect of the nuv bias on the _ mlcs2k2 _ fit parameters @xmath80 , @xmath81 , and @xmath3 are presented in figure [ fit_param_resid ] . _ mlcs2k2 _ interprets the excess blue color as reduced extinction , and the recovered @xmath81 distribution becomes increasingly asymmetric and skewed to lower values with increase in the bias . however , as @xmath81 is negatively correlated with the distance modulus , @xmath80 , increasing nuv bias leads to an increasing positive bias in the recovered distances . we find two effects with increasing redshift . as rest - frame @xmath82 and @xmath83 redshift out of the observer frame @xmath84 filters , the relative weight of @xmath70 and @xmath39 increases . additionally , at very high-@xmath1 , our simulation samples a very narrow range of low @xmath81 as we are unlikely to find heavily extincted sne ia . however as _ mlcs2k2 _ does not allow negative @xmath81 this leads to a reduced bias in the recovered extinction at high-@xmath1 . consequently , the bias in recovered distance modulus first increases up to @xmath1@xmath85@xmath86 and then decreases . the recovered light curve shape parameter , @xmath3 shows an increasing bias to more negative , or broader , intrinsically brighter models with increase in the strength of the nuv bias . our simulation uses a simple magnitude offset as the bias , and does not alter light curve shape in any band , and we do not find any significant bias in the recovered time of maximum , as we would expect . for each value of the bias , we create a combined sample of unbiased , nuv - red sne ia and nuv - blue sne ia consistent with figure [ blue_red_ratio ] . we use the _ simple - cosfitter _ ( v1.6.11 ) package to estimate the cosmological parameters , @xmath87 and @xmath69 for each combined sample , and compare the results to the cosmological parameters estimated from the full unbiased nuv - red sample . the effect of the nuv bias on the cosmological parameters is shown in figure [ cosmo_fit ] . as increasing the nuv bias leads to an increased positive bias in recovered distance modulus , @xmath80 , we find a bias towards more negative values of the equation of state parameter , @xmath69 from the sne ia data . when combined with the bao prior , the joint constraints are biased towards lower values of @xmath69 and higher values of @xmath88 . however , the tension between our simulated sne ia data and the real bao data also increases with increase in the bias . the @xmath89 sne ia contours do not overlap with the the bao contours in the case of the strong @xmath90 mag bias , and the combined constraints are not meaningful . it is likely that biases over @xmath91 mag are unphysical . the increased tension is not captured by the results from simple parameter marginalization shown in figure [ cosmo_resid ] . there are several limitations in these simulations - given the small sample size presented in this work , there is still considerable uncertainty about the size of the difference between the nuv colors , the physical mechanism that causes it , and the relative ratio of the two groups and it s evolution with redshift . additionally , we have made several simplifying assumptions such as simulating a single survey , and relating the bias in johnson @xmath39 to johnson @xmath70 . more extensive analysis would be premature given the current sample size . the quantitative results shown in figure [ cosmo_resid ] can only be interpreted within the context of these limitations , and the qualitative behavior seen with in light curve fit parameter residuals and the cosmological contours arguably offers more insight . in particular , the tension between our simulated sne ia data , and the real bao constraints is similar to the tension observed between constraints from planck and the union 2.1 supernova set @xcite . there are many sources of systematics or potential new physics that could cause the observed tension between the sne ia data and the results from cmb experiments , and we can not definitely ascribe it to the effect of two groups of uv - optical colors in sne ia . as the size of the _ swift _ sne ia sample increases , and high-@xmath1 searches such as pan - starrs and des probe the rest - frame uv with better s / n , it will become possible to identify and characterize different color groups more thoroughly . absolute magnitudes of the early uvot sne ia were presented in brown et al . ( 2010 ) . that sample was dominated by nuv - red events with a single nuv - blue sn , 2008q and used existing color relations and extinction laws . rather than extend that methodology to the current , larger sample , it is important to wait until an improved treatment of intrinsic colors and extinction is developed , and present the absolute magnitudes of the uvot sample at that time . efforts are also being made to observe sne ia in the nearby hubble flow with uvot to reduce the effect of peculiar and/or thermal velocities on the redshift - derived distances . for the more nearby sne , improved distances using cepheids or other distance indicators would allow the differences seen in the uv photometry as well as spectroscopy @xcite to be compared directly to the absolute magnitudes in the optical . the existence of two uv - optical color groups amongst normal sne ia has been established by uvot photometric observations . three mid-@xmath1 spectroscopic samples have had spectrophotometry performed to produce uvot @xmath9 , @xmath92 and @xmath10 spectrophotometry . the three samples have been combined and compared with the uvot low-@xmath1 sample , finding the same two groups at higher redshift . the colors within each group match the low-@xmath1 colors , but the bluer , nuv - blue group dominates at high-@xmath1 ( 90% ) , the opposite of what is seen at low-@xmath1 ( @xmath2533% ) . this change appears to be the explanation for the difference between high-@xmath1 mean spectra and low-@xmath1 mean spectra that were generated from these samples . a higher redshift sample exhibits an even stronger dominance of nuv - blue events . separating the mid-@xmath1 sample into nuv - red / blue groups , we investigate the @xmath7 and @xmath2 colors . the @xmath7 colors also exhibit a nuv - red / blue separation , with the nuv - blue group again the bluer group , but the two samples begin to overlap . the @xmath2 color feature significant overlap , but the nuv - blue sample is bluer on average by 0.10.2 mag . this presents problems for methods of extinction estimation for sne ia that utilize @xmath47 colors , as existing methods do not differentiate nuv - red / blue events . the use of sne ia as cosmological distance indicators rely upon distant sne ia being similar to nearby sne ia , at least to the extent that there are nearby reference sne ia for all distant sne ia . the finding that nuv - blue events dominate distant sne ia means that the absolute magnitudes , and a separate luminosity - width relation needs to be derived for this group of normal sne ia . we found that the nuv color differences were highly correlated with the measured ejecta velocity . optical color differences have also been reported between normal sne ia that have been divided into two groups based upon the blueshift of a prominent @xmath176355 absorption feature ( @xcite , f12 ) . foley & kasen ( 2011 ) suggested that the physical reason for the correlation between color and ejecta velocity was driven by the connection between the depth of the line - forming region within the ejecta and the opacity . one can interpret this connection as higher ejecta velocities causing overlapping lines and thus increased opacity and redder @xmath47 colors or as increased fe - group abundance increasing the opacity and causing the line - forming region for intermediate - mass elements to be further out in the ejecta and thus at higher velocity . m13 showed that nuv - blue sne ia have low velocities while the nuv - red sne ia are equally mixed between low and high - velocity objects . approximately two - thirds of low-@xmath1 sne ia have low - velocity ejecta ( e.g. , * ? ? * ) . meanwhile , about one - third of low-@xmath1 sne ia are nuv - blue . low-@xmath1 sne ia can be separated into two roughly equal - sized groups : low - velocity / nuv - blue , low - velocity / nuv - red , and high - velocity / nuv - red . the low - velocity / nuv - blue events are bluer in @xmath47 , leading to the observed color differences . this contrasts with the mid-@xmath1 sample , where about three - quarters of the sample are nuv - blue . for the subsample where we could determine their ejecta velocities , the mid-@xmath1 nuv - blue sne were essentially all low - velocity , while half of the mid-@xmath1 nuv - red sne are high - velocity therefore , the three groups listed above , low - velocity / nuv - blue , low - velocity / nuv - red , and high - velocity / nuv - red , have approximate fractions of 76% , 12% , and 12% , respectively . we note that these fractions are seen for a subsample of mid-@xmath1 sne ia , specifically those with relatively high - quality spectra . as detailed by e08 and f12 it is unlikely that these subsamples had selection effects that cause this color difference . however , it is possible that the true fractions for the entire sn ia population at higher redshifts could vary from those reported here . these findings motivate revisiting the sn templates that are used in comparisons with high - redshift sne ia . two sets of templates should be produced , one for nuv - blue sne ia and one for nuv - red sne ia . a key for inclusion into the new template sets will be whether nuv - red / blue membership can be determined . for archival sn ia observations , that would require that at least some observations of a given sn were performed with a detector with acceptable blue sensitivity , to allow clear separation between nuv - red and nuv - blue . for future sn ia observations , these findings suggest placing a high priority on detectors with excellent blue sensitivity , or building a template set from uvot - observed sne ia . an alternative possibility would be for the observations of known nuv - red and nuv - blue sne to reveal another observable characteristic that completely correlates with the nuv - optical along those same lines , it will be important to understand the associations between nuv - red / blue groupings and other characteristics that are independent of peak width , such as : hv / lv grouping , the presence or absence of unburned carbon in early - epoch spectra , whether the late @xmath39 and @xmath93 band photometry follows the lira relation and whether the optical spectra show evidence of strong na d absorption lines . collectively , correlations between these groupings will be a powerful probe of the progenitor metallicity , the binary pair , and the explosion physics of the different varieties of sne ia . p.a.m . acknowledges support from nasa adap grant nnx10ad58 g . is supported by the mitchell postdoctoral fellowship and nsf grant ast-0708873 . all supernova observers thank the mission operations team at penn state for scheduling the thousands of individual uvot target - of - opportunity observations that comprise the uvot dataset . the nasa / ipac extragalactic database ( ned ) was utilized in this work . ned is operated by the jet propulsion laboratory of the california institute of technology , under contract with the national aeronautics and space administration . arsenijevic , m. , fabbro , s. , mourao , a.m. , rica da silva , a.j . 2008 , a&a , 492 , 535 balland , c. et al . 2009 , a&a , 507 , 85 blondin , s. , mandel . k.s . & kirshner , r.p . 2011 , a&a , 526 , a81 + brown , p. j. , et al . 2010 , aj , 721 , 1608 ( b10 ) brown , p. j. , et al . 2012 , apj , 753 , 22 brown , p. j. , et al . 2014 , apj , 787 , 29 bufano , f. , et al . 2009 , apj , 700 , 1456 childress , m. j. , et al . 2013 , mnras , in press , astroph - co 1307.0563 conley , a. , et al . 2008 , apj , 681 , 482 cooke , j. , et al . 2011 , apj , 727 , 35 ellis , r.s . , et al . 2008 , apj , 674 , 51 foley , r. j. , et al . 2008 , apj , 684 , 68 foley , r. j. , et al . 2010 , apj , 708 , 1748 foley , r. j. , sanders , n.e . & kirshner , r.p . 2011 , apj , 742 , 89 foley , r. j. & kasen , d. 2011 , 729 , 55 foley , r. j. , et al . 2012 , apj , 744 , 38 foley , r. j. , et al . 2012 , apj , 753 , l5 foley , r. j. , et al . 2012 , aj , 143 , 113 foley , r. j. , et al . 2013 , apj , 767 , 57 foley , r. j. & kirshner , r.p . 2013 , 769 , l1 folatelli , g. , et al . 2010 , aj , 139 , 120 folatelli , g. , et al . 2012 , apj , 745 , 74 hachinger , s. et al . 2013 , mnras , 429 , 2228 hicken , m. , wood - vasey , w.m . , blondin , s. , challis , p. , jha , s. , kelly , p. , rest , a. , kirshner , r.p 2009 , apj , 700 , 1097 hicken , m. , et al . 2007 , apjl , 669 , l17 hflich , p. , wheeler , j.c . & thielemann , f.k . 1998 , apj , 495 , 617 howell , d.a . , et al . 2006 , nature , 443 , 308 jha , s. , et al . 2006a , aj , 131 , 527 jha , s. , et al . 2006b , aj , 132 , 189 jha , s. , riess , a. & kirshner , r.p . 2007 , apj , 659 , 122 kessler , r. , et al . 2009 , apjs , 185 , 32 kirshner , r.p . , et al . 1993 , apj , 415 , 589 krisciunas , k. et al . 2009 , aj , 138 , 1584 lentz , e.j . , baron , e. , branch , d. , hauschildt , p.h . , nugent , p.e . 2000 , apj , 530 , 966 li , w. , et al . 2001 , pasp , 113 , 1178 lira , p. 1995 , masters thesis , univ . chile maguire , k. et al . 2010 , mnras , 418 , 747 maguire , k. et al . 2012 , mnras , 426 , 2359 mazzali , p.a . 2013 , mnras , astroph 1305.2356 milne , p.a . 2010 , aj , 721 , 1627 ( m10 ) milne , p.a . 2013 , apj , 779 , 23 ( m13 ) nordin , j. , et al . 2011 , apj , 734 , 42 perlmutter , s. , et al . 1997 , apj , 483 , 565 perlmutter , s. , et al . 1999 , apj , 517 , 565 phillips , m. m. 1993 , apj , 413 , l105 phillips , m. m. , lira , p. , suntzeff , n. b. , schommer , r. a. , hamuy , m. , & maza , j. 1999 , aj , 118 , 1766 poole , t. , et al . 2008 , mnras , 383 , 627 rest , a. , et al . 2013 , apj - submitted , astroph 1310.3828 riess , a. g. , press , w. h. , & kirshner , r. p. 1996 , apj , 473 , 88 riess , a. g. , et al . 1998 , aj , 116 , 1009 riess , a. g. , et al . 2007 , apj , 659 , 98 roepke , f.k . 2012 , apj , 750 , l19 roming , p. w. a. , et al . 2005 , space science reviews , 120 , 95 sauer , d. n. , mazzali , p. a. , blondin , s. , filippenko , a. v. , benetti , s. , stehle , m. , challis , p. , kirshner , r. p. , & li , w. 2008 , mnras , 391 , 1605 scalzo , r. et al . 2010 , apj , 713 , 1073 silverman , j. , et al . 2011 , mnras , 410 , 585 silverman , j. m. , ganeshalingam , m. , li , w. , filippenko , a.v . , miller , a.a . , poznanski , d. 2011 , mnras , 410 , 585 silverman , j. , kong , j.j . & filippenko , a.v . 2011 , mnras , 425 , 1819 sullivan , m. , kasliwal , m.m . 2011 , apj , 732 , 118 taubenberger , s. et al . 2008 , mnras , 385 , 75 thomas , r.c . 2011 , apj , 743 , 27 walker , e.s . 2012 , mnras , 427 , 103 wang , x. , et al . 2009a , apj , 697 , 380 wang , x. , et al . 2009b , apj , 699 , 139 wang , x. , et al . 2012 , apj , 749 , 126 yamanaka , m. et al . 2009 , apj , 707 , l118

let us consider a continuum analog of erds - rnyi random graphs , called the quantum random graphs . in these graphs , the vertices are replaced by circles punctured at poisson points of arrivals , and connections are derived through another poisson process on the circle . the quantum random graph is an instance of an inhomogeneous random graph , but it does not belong to the class of rank-1 models . in this paper , we first prove that at criticality the largest connected component of such graph scales like . then , we prove that the scaled and ordered sizes of connected components converge in distribution to the ordered lengths of excursions above zero of a reflected brownian motion with a drift . in the process , we also prove that the surplus edges converge to a point process whose intensity is given by the same reflected diffusion .

1404.5705

the erds - rnyi random graph @xcite is the simplest but probably the most important example of a random graph . in this graph , denoted by @xmath1 each pair of the @xmath2 vertices is connected with probability @xmath3 independently of other pairs . the phase transition phenomenon it exhibits has been greatly studied in the past decades . it has been proved @xcite that for @xmath4 with @xmath5 , the largest component has @xmath6 vertices and the second largest @xmath7 vertices ( as @xmath8 with probability @xmath9 ) ; for @xmath4 with @xmath10 the largest component has @xmath7 vertices ; when @xmath11 , the largest component of @xmath1 has @xmath12 vertices . in 1997 , aldous considered the erds - rnyi random graph inside the scaling window " of the phase transition . he proved in @xcite that the ordered set of the component sizes rescaled by @xmath13 converge to an ordered set of excursion lengths of reflected inhomogeneous brownian motion with a certain drift . this work received a lot of attention since its publication . a vast variety of random graph models studied in the past decades have been shown to exhibit a phase transition phenomenon similar to the erds - rnyi random graph . many of these models have been proved to show same type of behaviour in the near critical regime and some of them have been shown to fall into the different universality class . one of the models above is a random multigraph with @xmath2 vertices constructed using the configuration model with its vertex degrees being independent identically distributed random variables with the distribution @xmath14 . joseph in 2011 @xcite showed that the limit of the ordered set of the critical random multigraph component sizes rescaled appropriately depends on the distribution @xmath14 in the following way . if @xmath14 has finite third moment then the right scaling is @xmath13 and the model falls into the same universality class as the erds - rnyi random graph . however , if @xmath15 as @xmath16 , @xmath17 , @xmath18 then the scaling is @xmath19 and the limit is an ordered set of the excursion lengths of a certain drifted process with independent increments above past minima . riordan @xcite proved the same result as finite third moment but extended it to the scaling window . nachmias and peres in 2007 @xcite among other results proved that random graph obtained by performing percolation on a random @xmath20-regular ( @xmath21 ) graph on @xmath2 vertices with percolation probability @xmath22 , @xmath23 falls into the same universality class as the erds - rnyi random graph . the near critical behavior of the rank-1 model , which is a special case of the general inhomogeneous random graph studied in @xcite , received much attention in the last years . this model can be seen as a random graph with vertices having a weight ( @xmath24 of vertex @xmath25 ) associated to them , and the weight of a vertex moderates its degree in a sense that the probability of connection between two vertices @xmath25 and @xmath26 is @xmath27 , where @xmath28 is a positive constant possibly depending on @xmath2 . we refer the reader to @xcite for further details . in the rank-1 model in the critical window " , the weights @xmath29 are independent identically distributed random variables with finite third moment and the probability of connection is @xmath30 . this model , formulated , however , differently was first proved by aldous in @xcite to belong to the same universality class as the erds - rnyi random graph . later , in @xcite and in @xcite independently , similar result was reproved . the rank-1 model with power - law degrees was considered in @xcite . it has been proved that for exponent @xmath18 , the sizes of the components rescaled by @xmath19 converge to hitting times of certain lvy process . note that for @xmath18 the degrees have finite variance but infinite third moment . the purpose of this paper is to study the near critical behavior of the so - called quantum random graph . as it is explained in @xcite both the motivation and the choice of terminology come from the stochastic geometric ( fortuin kasteleyn type ) representation of the quantum curie - weiss model in transverse magnetic field , which was originally developed in the general ferromagnetic context in @xcite . we refer the reader to @xcite where the stochastic geometric representations of different types are explained in more detail . the quantum random graph , as is explained below , was first introduced in @xcite . it is a quantum version of the classical erds - rnyi random graph . more precisely : [ [ the - model ] ] the model : + + + + + + + + + + there are two parameters @xmath31 $ ] - the inverse temperature , and @xmath32 - the strength of the transverse field . the case @xmath33 corresponds to the ground state . and as it will be explained later the case @xmath34 brings us to the classical erds - rnyi random graphs discussed above . we shall consider only @xmath35 and @xmath36 case in this work . let @xmath37 denote the circle of length @xmath38 . the model is built on the space @xmath39 , that is to each site @xmath40 we attach a copy @xmath41 of @xmath37 , where we use @xmath42 . we shall call the second coordinate of a point in this space - the time coordinate . in each @xmath41 , make finite random number of holes using the poisson point process of holes @xmath43 on @xmath41 with intensity @xmath44 . @xmath43-s are assumed to be independent for different @xmath25-s . the punched circle @xmath45 consists of @xmath46 disjoint connected intervals , @xmath47 of course @xmath48 whenever the cardinality @xmath49 . next , draw finite random number of links between various points of @xmath50 with the same time coordinates , that is between points of the type @xmath51 and @xmath52 where @xmath53 and @xmath54 in the following way . with each ( unordered ) pair of vertices @xmath55 we associate a copy @xmath56 of @xmath37 and a poisson point process of links @xmath57 on @xmath56 with intensity @xmath58 . processes @xmath59 are assumed to be independent for different @xmath60 and also independent of the processes of holes @xmath43 . two intervals @xmath61 and @xmath62 in the decomposition are said to be connected if there exists @xmath63 such that @xmath64 , whereas @xmath65 and @xmath66 . the decomposition @xmath67 of @xmath68 into maximal connected components is , thereby , well defined . we used @xmath69 above to denote the total collection of all the holes . an example for @xmath70 is given in figure [ fig:1 ] . for each fixed @xmath71 the probability @xmath72 . thus , for given @xmath71 the notion @xmath73 of the connected component containing @xmath74 in the above decomposition is well defined . after all the holes are punched and the links are drawn : @xmath75 , where @xmath76 , @xmath77 and @xmath78.,scaledwidth=70.0% ] if @xmath79 then there are no holes and @xmath45 always contains only one connected component , which , of course , equals to @xmath41 itself . in the latter case , the probability that @xmath41 and @xmath80 are connected equals @xmath81 and we are back to a setup similar to that of erds - rnyi . we wish to emphasise here that all the random graph models mentioned in this section mimic the idea of rank-1 random graphs in the sense that the probability of connecting any two vertices depends on the _ weights _ associated with the vertices . it has been noticed in @xcite that our random graph model can be treated as an instance of the general inhomogeneous random graph , if each interval is treated as a vertex . however , in our random graph model the connection between any two vertices depends on the size of overlap of the two intervals , and not only on the individual lengths of the intervals ( which can be considered as weights ) . this property not only separates our model from the class of rank-1 models , but also marks it as the first model which is inherently different from those random graph models which appear to mimic the idea of rank-1 random graphs . the critical curve of the quantum random graph in the @xmath82-coordinate quarter plane is already known ( cf . @xcite ) , and is given by : @xmath83 which follows from comparisons with critical branching process with the offspring distribution given by the _ cut - gamma _ distribution @xmath84 . it has been proved that if @xmath85 such that @xmath86 an emergence of an @xmath6-giant connected component happens in the disjoint decomposition and if @xmath85 such that @xmath87 typical connected component of any point @xmath88 is of order @xmath7 . this is a quantum version of erds - rnyi phase transition phenomenon . the objective of this paper is primarily two fold : first to analyze the component sizes at criticality and then to analyze this model for values of the parameters lying in a small window around the criticality . before stating the results we remark here that our definition of size of a connected component in the quantum random graph model is the number of intervals in the component . we note here that unlike the cases dealt with in @xcite , the exploration process ( or the breadth first walk ) corresponding to our model can visit an already visited vertex ( circle ) . in general , it can do so as many times as the number of intervals sharing the same vertex , which is a _ finite but unbounded random variable_. this difference makes our analysis quite challenging . although , in proving results related to the sizes of component we have managed to surpass the above technical difficulty by proving that these multiple returns take place with very small probability . however , these probabilities are not small enough to be ignored when we consider the scaling limits of ordered set of component sizes in our model . in fact , only the first return plays a crucial role , and the subsequent returns indeed carry such small probabilities that they do not play any major role in our results . the difference between the limiting process in @xcite and in our work is the coefficient @xmath89 . as it becomes clear from the proof , it represents the already explored ( during the first visit ) portion of a vertex ( circle ) . finally , we note here that questions in the spirit of convergence of random graphs as metric spaces are still open for our model primarily because of this precise problem of multiple visits to the same vertex . we see through our proofs that our model belongs to the erds - rnyi universality class primarily due to two reasons : first is the very small probability of multiple returns to the same vertex , and second is the fact that the exploration process in our model seems to forget it s past , and behave almost like a markov process . this feature of our model can be used to check many other inhomogeneous random graph models if they belong to the erds - rnyi universality class . we also note here that the first reason that of small probability of multiple returns can further be relaxed but the second reason that of approximate markovian behaviour is crucial . the first result of this paper complements @xcite , as it answers questions similar to those in @xcite , but at criticality . [ thm : crit : size ] let @xmath90 be the largest component in the critical quantum random graph . then , * there exist constants @xmath91 and @xmath92 , such that for all @xmath93 and @xmath94 , there exists a constant @xmath95 such that @xmath96 * there exist @xmath97 and @xmath98 such that for all @xmath99 and @xmath100 , we have @xmath101 consequently , one can conclude that the size of a typical connected component is @xmath102 , which matches well with the corresponding results for the erds - rnyi random graph . this naturally leads one to believe that the quantum random graph may indeed fall in the same universality class as the erds - rnyi random graph , which is precisely the second result of this paper . however , in order to state the result , we first need to define some related quantities . to begin with , as in @xcite , the surplus edges in a component are , intuitively , those edges , which when removed , make the component become a tree . the precise definition will be given in step ( iv ) of exploration process in section [ subsec : exploration ] , and in that notation @xmath103 next , starting with standard brownian motion @xmath104 , we define a new process @xmath105 where @xmath106 . further , we define the following process @xmath107 and a corresponding marks process @xmath108 , which can be characterized as the counting process for which @xmath109 then , [ thm : joint : cgs ] let @xmath110 denote the ordered sizes of the components in the quantum random graph with parameters \{@xmath111 for some @xmath23 ; and let @xmath112 be the surplus of the @xmath26-th component . then as @xmath113 @xmath114 where @xmath115 denote the ordered lengths of the excursions of @xmath116 above zero , and @xmath117 denote the mark counts corresponding of the process @xmath118 . moreover , the convergence of the rescaled component sizes holds with respect to the @xmath119 topology defined in @xcite . our proofs evolve around an exploration process which is a tool to estimate the size of a connected component in random graphs . rest of this paper is partitioned into two parts . in section [ sec : crit : scale ] , we shall prove theorem [ thm : crit : size ] by adapting the ideas set forth in @xcite to our model . the primary idea is to cleverly construct two counting processes which stochastically dominate ( from above and below ) the original counting process which determines the size of the component . this way we avoid the process of precise counting , and the difficulties associated with it . however , the same can not be done in proving theorem [ thm : joint : cgs ] , since it requires precise estimates , rather than crude upper and lower bounds . in section [ sec : joint : cgs ] , we shall put together the proof of theorem [ thm : joint : cgs ] , which is presented in multiple steps , leading to the main result . two of the main results , which set our analysis apart are propositions [ prop : nu - m ] and [ prop : nu-1 - 0 ] , which are statements about the number of vertices which have already been visited once , twice , or more number of times . these crucial estimates brings our analysis one step closer to that of @xcite , and allows us to follow the program of @xcite . in sections [ subsec : brownian : excursions ] and [ subsec : surplus : edges ] , we prove the convergence of breadth first walk and the process of marking surplus edges to appropriate continuous time analogs , respectively . finally , in section [ subsec : joint : convergence ] , we collate all the results , and prove theorem [ thm : joint : cgs ] . as mentioned in the previous section , the problem of revisiting the same vertex multiple number of times makes it difficult for us to get working estimates for the distribution of the exploration process . we get over this problem by introducing two processes , which stochastically dominate the exploration process from above and below . this enables us to get appropriate upper and lower bounds for the probabilities appearing in theorem [ thm : crit : size ] . in the following section we shall first describe the exploration process . thereafter in sections [ subsec : overcounting ] and [ subsec : free - exploration ] , we shall present the proofs of the two parts of theorem [ thm : crit : size ] . let us start with defining the exploration process corresponding to our model , originally defined in @xcite for one cluster . using this algorithm we shall construct / sample the quantum random graph , interval by interval . first , order the vertices ( circles ) . at time @xmath46 let @xmath120 be the number of active intervals and the number of explored intervals will be @xmath46 . _ * initial stage : * _ take the first vertex ( say @xmath121 ) and sample an interval @xmath122 around @xmath123 in the following way , which is governed by the process of holes . let @xmath124 be two independent exp(@xmath44 ) random variables . if @xmath125 then @xmath126 . otherwise , @xmath127 . in the sequel we shall refer to the distribution of the length of @xmath122 as to @xmath128 . at time @xmath129 the interval @xmath122 is the only active interval . analogous to the exploration process of the erds - rnyi random graph we call all the other parts of the space neutral . since the intervals are not sampled yet it makes more sense to call the whole circles neutral ( or free ) rather than intervals . finally , set @xmath130 . _ * at time @xmath46 : * _ * if @xmath131 , let @xmath132 be the active interval belonging to the vertex with the smallest ordering number . * if @xmath133 and there exists at least one free vertex ( whole neutral circle ) , then sample the interval including @xmath134 with @xmath135 being the first free vertex in the same way as described for @xmath122 above . let @xmath132 be this interval . * if @xmath133 and there is no free vertex , then sample the neutral interval with the smallest beginning point and let @xmath132 be this interval . _ connections of @xmath132 : _ denote by @xmath135 the vertex @xmath132 belongs to . for each vertex @xmath25 run @xmath136 on the @xmath137 . for each link check : * if the resulting link is connected to the already explored interval , simply ignore ( erase ) this link . * if the resulting link is connected to a free vertex , sample an interval @xmath138 around the point of connection @xmath139 in the following way ruled by the process of holes . let @xmath124 be two independent exp(@xmath44 ) random variables . if @xmath125 then @xmath140 . otherwise , @xmath141 . namely , @xmath142 defined above . * if the resulting link is connected to the neutral part @xmath143 of a vertex which already has explored interval(s ) , sample an interval @xmath138 around the point of connection @xmath139 in the following way ruled by the process of holes . let @xmath124 be two independent exp(@xmath44 ) random variables , then @xmath144 . * if the resulting link is connected to an active interval , keep this link . these type of links play the role of surplus edges . namely , if one erases these type of links , the resulting cluster will be a tree . denote by @xmath145 the number of neutral intervals connected to @xmath132 and change their status to active . then set @xmath132 itself explored . note that in order to be able to construct the quantum random graph we need to keep track of all the active and explored intervals ( beginning and ending points ) . by the definition of @xmath120 we have the following recursion : @xmath146 although we do not define the distribution of @xmath145 explicitly , one can get it from _ connections of @xmath132 _ " part of the algorithm . however , it is obvious that the conditional distribution of @xmath145 given @xmath147 is stochastically dominated by bin@xmath148 , with @xmath142 . note that the first time when the process @xmath149 , is when the exploration of the cluster which includes the origin of the process @xmath120 , ends . in other words , let us define @xmath150 , then by @xmath151 the number of explored intervals equals @xmath152 , the size of the cluster which includes the vertex @xmath121 . due to the intricate nature of our model , it is rather difficult to analyze this exploration process without resorting to approximations . thus , we shall define two more counting processes which will dominate the aforementioned exploration process from above and below . these processes will be far more amenable to analyze , hence paving the way for easier computations . let us begin with a collection of independent pairs of random variables @xmath153 such that @xmath154 s are all i.i.d . with @xmath155 , and given @xmath154 , the random variable @xmath156 is distributed as bin@xmath157 . then , define the following process @xmath158 clearly , since the random variables @xmath145s is dominated by @xmath159s , therefore the process @xmath120 is consequently dominated by @xmath160 . however note that this is not a martingale , and define @xmath161 as the martingale extracted out of the process @xmath160 . also , note that @xmath162 which suggests that @xmath163 for all @xmath46 , since @xmath164 , which in turn implies that @xmath165 also dominates @xmath120 . moreover , we can couple @xmath165 to @xmath120 in such a way that @xmath166 for all @xmath46 . note that for a fixed @xmath167 , the set @xmath168 can be rewritten as @xmath169 . next let us define @xmath170 since @xmath171 , @xmath172 implies that either @xmath173 or @xmath174 . therefore , we have @xmath175 thus , in order to get an upper bound for @xmath176 we need to estimate the two probabilities appearing on the right hand side of the above equation . applying optional sampling theorem , together with markov s inequality we get @xmath177 thereafter , let @xmath178 be another copy of @xmath179 , then we have the following which is akin to lemma 5 of @xcite . @xmath180 which is a consequence of the same set of arguments as set forth in lemma 5 of @xcite . thus , we have @xmath181 which implies that the overshoot @xmath182 conditioned on the event @xmath183 is stochastically bounded by the random variable @xmath178 . therefore , using corollary 6 of @xcite with @xmath184 for the process @xmath165 , we get , @xmath185 \le h^2 + 3h , \,\,\,\,\,\,\,\,\,\,\,\,\text{for large enough $ h$,}\ ] ] also note that @xmath186 is a martingale , which can be used to conclude that @xmath187 \ ] ] subsequently , taking @xmath167 so that @xmath188 , and using equations & , we get @xmath189 where @xmath190 . thus , for large enough @xmath167 we can conclude that there exists a @xmath17 such that @xmath191 subsequently , writing @xmath192 , we have @xmath193 let @xmath194 , then @xmath195 implies @xmath196 , which in turn implies @xmath197 now , we shall use this bound on typical component size to obtain a similar bound for the largest component @xmath90 , for which let us introduce @xmath198 as the cumulative size of components , which have at least @xmath199 vertices . clearly , @xmath198 is either zero , or is larger than @xmath199 , therefore , @xmath200 implies that there must be at least one component whose size is larger than @xmath199 , thus resulting in the following set of inequalities , @xmath201 for @xmath202 . next , for any @xmath203 choosing @xmath204 the above inequality simplifies to @xmath205 which proves the first part of theorem [ thm : crit : size ] . now that we have an upper bound , we shall try to get a lower bound , which in turn will shed some light on the order of the size of the largest component . the principal idea to obtain the lower bound is to consider a similar process as above , but one which counts only the free vertices . as it has been mentioned earlier , in our model , each @xmath206 can be revisited many times , and by counting only the free vertices we overrule the possibility of revisiting the same vertex , thus achieving the goal of undercounting . namely , in this process , each component contain at most the same amount of intervals as the original exploration process . this undercounting process is similar to the exploration process in @xcite . however , for the sake of clarity we shall define this process . first order the vertices ( circles ) . at time @xmath207 , let @xmath208 be the number of active intervals and the number of explored intervals will be @xmath46 . * initial stage : * _ take the first vertex ( say @xmath121 ) and sample an interval @xmath122 around @xmath123 in the same way as described in the original exploration process algorithm in the beginning of this section . namely @xmath209 . at time @xmath129 the interval @xmath122 is the only active interval and all other vertices are neutral ( free ) . finally , set @xmath210 . _ * at time @xmath46 : * _ * if @xmath211 , let @xmath212 be the active interval belonging to the vertex with the smallest ordering number . * if @xmath213 and there exists at least one free vertex ( whole neutral circle ) , then sample the interval including @xmath134 with @xmath135 being the first free vertex in the same way as described for @xmath122 above . let @xmath212 be this interval . * if @xmath213 and there is no free vertex , then the process is over . _ connections of @xmath212 : _ denote by @xmath135 the vertex @xmath212 belongs to . for each vertex @xmath25 run @xmath136 on the @xmath214 . for each link check : * if the resulting link is connected to the already explored interval , simply ignore ( erase ) this link . * if the resulting link is connected to a free vertex , sample an interval @xmath215 around the point of connection @xmath139 in the same way as described in the original exploration process algorithm in the beginning of this section . namely , @xmath216 . * if the resulting link is connected to the neutral part of a vertex which already has an explored interval / s , simply ignore ( erase ) this link . * if the resulting link is connected to an active interval , keep this link . denote by @xmath217 the number of neutral intervals connected to @xmath212 and change their status to active . then set @xmath212 itself explored . by the definition of @xmath208 we have the following recursion : @xmath218 define @xmath219 , then the conditional distribution of @xmath217 given @xmath220 and @xmath221 is bin@xmath222 , where @xmath223 . let us begin with @xmath224 , @xmath225 and @xmath226 positive integers , then define @xmath227 with the usual convention that minimum of an empty set is taken as @xmath228 . if @xmath229 , then @xmath230 now equipped with the facts that @xmath231 at criticality and @xmath232 , if we assume @xmath233 , then for large enough @xmath2 we have @xmath234 \ge \frac { ( n - h - k)}{n}\e\left[|i_k^f|\cdot\left(1-\frac{|i_k^f|}{n}\right)\right ] -2\frac hn\left(h+k \right)+ o\left(\frac 1n \right)\ ] ] if we further assume that @xmath235 and @xmath236 , then the above reduces to @xmath237 > \frac 12\ ] ] we can redo these calculations and prove when @xmath238 , in which case @xmath239 . thus implying that @xmath240 is a submartingale . subsequently , using corollary 6 of @xcite , with @xmath184 for the process @xmath241 , we get @xmath242 for @xmath243 . then using similar set of inequalities as in , we get @xmath244 which in turn implies that @xmath245 which is the same as in @xcite . through above estimates what we have learnt is that @xmath246 is of the order of @xmath2 with a very small probability , which means that there exists at least one fairly large cluster . next we need to ensure that this fairly large cluster is not very large . for this , let us define @xmath247 if this set is nonempty , and @xmath248 otherwise . let @xmath249 , and assume @xmath250 , then it implies that @xmath251 . moreover we have @xmath252 also , note that @xmath253 further observing that at criticality @xmath254 , we get @xmath255\\ & \le & 1 + ( 2m_{s-1}-1)\frac{a^f_{\tau_h+s-1}+(\tau_h+s-1)}{n } + \e\left(|i_k^f|-1\right)^2 + 2\frac{a^f_{\tau_h+s-1}+(\tau_h+s-1)}{n}\e\left ( |i_k^f|^2 - |i_k^f|\right)\\ & & + \left ( \frac{a^f_{\tau_h+s-1}+(\tau_h+s-1)}{n}\right)^2\e\left(|i_k^f|^2\right ) + o\left ( \frac{1}{n}\right)\end{aligned}\ ] ] next assuming that @xmath235 and @xmath256 , and noting that we had begun with assuming that @xmath257 and @xmath258 were bounded from below and above by @xmath259 and @xmath224 , respectively , we get @xmath260 \le 2.\ ] ] the above inequality can also be checked even when @xmath261 , thus making @xmath262 a super martingale . from here on , we can follow the arguments set forth in the proof of theorem 2 of @xcite verbatim to conclude that : @xmath263 finally , as in @xcite , choosing @xmath264 and appropriate @xmath224 so as to minimize the right had side of the above inequality , we get @xmath265 , which also satisfies @xmath266 , and makes the rhs of the above inequality less than @xmath267 . writing @xmath268 as the size of the cluster obtained via the exploration process on the free vertices only , we observe that @xmath269 implies @xmath270 , which in turn implies that @xmath271 . however , note that @xmath272 , thus , @xmath273 thus completing the proof of second part of theorem [ thm : crit : size ] . now putting together the two parts of theorem [ thm : crit : size ] , we conclude that the largest component in the critical quantum random graph is indeed @xmath12 , which was the goal of this section . in order to describe our model , one can equivalently consider the length of circle as @xmath274 ( say ) , and run a unit intensity poisson process for creating holes , and another set of i.i.d . poisson processes with intensity @xmath275 for all the pairs of circles for creating links between them . for technical reasons , we shall adopt this interpretation of our model throughout the rest of this work . the distribution of the length of sampled interval @xmath138 , is characterised as @xmath276 where @xmath277 and @xmath278 are two independent , unit intensity exponential random variables . in the notation used in the previous section , we shall write @xmath279 . thus , @xmath280 then , under this notation , using @xcite the critical curve can be expressed as the set of all those @xmath44 and @xmath28 such that @xmath281 subsequently , for any fixed @xmath106 , let us define the critical window as @xmath282 note that we can rewrite @xmath283 ( defined in ) as follows : @xmath284 let @xmath285 be the breadth - first walk associated with the quantum random graph with parameters in the critical window . formally , @xmath286 , @xmath287 , where @xmath145 are the number of neutral intervals connected to the interval which is screened at time @xmath46 , which were first introduced in the algorithm defining the exploration process in . [ thm : scaling : limit ] rescale @xmath288 by defining @xmath289 then @xmath290 in distribution as @xmath113 . _ proof_. from the general theory we know that we can decompose @xmath291 as follows @xmath292 where @xmath293 is a martingale and @xmath294 is a predictable , bounded variation process . then @xmath295 where @xmath296 is a martingale and @xmath297 is a predictable , monotone increasing process . rescaling as in to define @xmath298 , we shall show that for a fixed @xmath299 , as @xmath113 @xmath300 @xmath301 @xmath302 this as a consequence of functional central limit theorem for continuous time martingales , will prove the main result of this theorem . @xmath303 from here onwards , we shall drop the indices @xmath304 from @xmath305 , and all related random variables . [ prop : nu - m ] let @xmath306 denote the number of vertices visited @xmath307 times by time @xmath308 . then , for @xmath309 , and @xmath310 , @xmath311 * proof : * note that by definition , @xmath312 then , applying markov inequality for @xmath313 we get @xmath314 observing that @xmath315 and the probability of connection to the visited site is less or equal than to the free one , therefore , the probability that at step @xmath25 there will appear a connection to the visited site is of order @xmath316 , which implies @xmath317 which proves the lemma for @xmath318 . similar set of arguments can be used to prove the result for @xmath319 . we also note here that for @xmath319 , the order is much smaller . @xmath303 let us define @xmath320 as the sigma algebra containing all the information upto time @xmath46 of the exploration process defined in section [ subsec : exploration ] . then by standard theory we know that @xmath321 which can further be simplified as follows : [ lemma : b - n ] @xmath322,\end{aligned}\ ] ] where @xmath323 , as defined earlier , is the number of vertices visited once until @xmath308 . * proof : * observe that @xmath324 the above sum can be broken down into three categories : intervals that belong to free vertices ; intervals which belong to vertices which have been visited once ; and intervals belonging to vertices which have been visited more than once . using this partition , and proposition [ prop : nu - m ] , we get @xmath325 where we have also used first order approximation of @xmath326 and @xmath327 . subsequently , recalling the arguments used to prove proposition [ prop : nu - m ] , we can show that for @xmath328 , @xmath329 , which in turn implies that the contribution from vertices visited more than twice is @xmath330 , which proves the lemma . we note here that @xmath331 can be an interval which is part of a free vertex , in which case all @xmath154 s have the same distribution . however , in the case when @xmath331 is part of a vertex which has already been visited ( at least once ) , then the distribution of various such @xmath154 s depends on @xmath25 . also , typically for @xmath332 , the interval @xmath331 is part of a free vertex with probability @xmath333 . from here on we consider only those @xmath331 s which are part of free vertices ( which happens with probability@xmath333 ) . [ prop : nu-1 - 0 ] in the notation of lemma [ lemma : b - n ] , for @xmath334 , we have @xmath335 @xmath336 * proof . * writing @xmath337 for the label of the component to which the @xmath25-th interval belongs , we have @xmath338 where @xmath339 is as defined earlier . consequently by proposition [ prop : nu - m ] , we can conclude that @xmath340 therefore , it suffices to prove that @xmath341 however , recall that @xmath342 thus , enough to show that @xmath343 which will , in turn , prove the lemma . corresponding to the breadth - first walk on intervals , it can be shown that : @xmath344 ( refer equations ( 5)-(7 ) of @xcite ) . therefore , we get @xmath345 hence proving is equivalent to proving that @xmath346 for which we shall adapt aldous method to our case . define @xmath347 then , note that for all @xmath348 , equation is satisfied ( by the definition of @xmath349 ) , which in turn implies equation . thus , for @xmath348 , we have proven the main result of this lemma . however , for general @xmath25 , we shall proceed as follows : we shall begin with collecting all bits of results we need to prove . first , we shall begin with recalling @xmath350 from equation , which can further be expressed as @xmath351.\ ] ] then using optional sampling theorem , we note that @xmath352 however , since @xmath350 is a monotone non - decreasing process , and @xmath353 , therefore , @xmath354 next , using the definition of @xmath350 in , and the fact that @xmath355 , we get @xmath356 recall that @xmath357 is stochastically bounded by @xmath358 where @xmath359 , which implies that @xmath360 . however , we also note that @xmath361 for large @xmath2 , therefore we get @xmath362 which in turn implies that @xmath363 next , to analyze @xmath364 , recall its expression from lemma [ lemma : b - n ] , and note that @xmath365 for any collection of intervals @xmath331 and @xmath366 . therefore , we get @xmath367.\end{aligned}\ ] ] however , we also note that @xmath368 , which then implies that @xmath369 similarly , a simple lower bound for @xmath364 can be obtained by dropping the sum over intervals , which have been visited once . @xmath370\end{aligned}\ ] ] now recall the set of equations , and that @xmath371 . therefore , @xmath372 \ ] ] let us analyze the second term in the above expression @xmath373 for the second part of the above expression , recall that @xmath374 for @xmath348 . then using @xmath375 , we can rewrite the above expression as @xmath376 next , by observing that @xmath377 , we conclude that @xmath378 $ ] is at worst @xmath379 . subsequently , combining equations and , we can claim that @xmath380 now that we have all the estimates we need to prove . note that @xmath381 applying equations and , we get @xmath382 thus , @xmath383 which proves that @xmath384 is stochastically bounded as @xmath385 . thus proving the claim in equation , and hence the proposition . @xmath303 equipped with the above proposition , we are now in a position to prove equation . [ lemma : d - n ] @xmath386 which is the assertion of equation . * proof . * recall that @xmath350 can be expressed as @xmath387 = \sum_{i\le k } { \rm var}\left(\eta_i | { { \cal f}}_{i-1 } \right).\ ] ] next , recall the random variables @xmath388 and @xmath156 defined in section [ sec : crit : scale ] , and notice that conditioned on @xmath389 , the random variables @xmath357 , @xmath390 and @xmath156 , are sums of independent bernoulli random variables . therefore , we have @xmath391 note that @xmath156 is independent of @xmath389 , thus we have @xmath392 therefore , @xmath393 next , using proposition [ prop : nu-1 - 0 ] , we get @xmath394 which implies that @xmath395 proving the statement of this lemma . @xmath303 next , we shall prove equation , which will bring us one step closer to proving theorem [ thm : scaling : limit ] . [ lem : m_bar ] equation is true . * proof . * recall that from the definition of @xmath396 , the martingale @xmath397 can be written as @xmath398 in view of this expression , it suffices to prove the following @xmath399 now by the same arguments as used in the proof of lemma [ lemma : d - n ] , we get that @xmath400 . therefore , @xmath401 however recall that , @xmath402 then , using the relationship between expectation of a non - negative random variable , and the sum of its tail probabilities , and using cauchy - schwarz for @xmath403 we get @xmath404.\end{aligned}\ ] ] next , using the standard formula for the fourth moment of binomial distribution , and that @xmath405 , we obtain @xmath406 where @xmath407 $ ] . thus , putting together equations and , and writing @xmath408 we get @xmath409 \ ] ] subsequently , for a fixed @xmath410 , we can break the sum on the right hand side into two parts : @xmath411 and @xmath412 . clearly , the first part can be bounded above by @xmath413 , whereas for the second part , we note that it s a sum of decreasing sequence of numbers , implying that the sum can be bounded above by the number of terms in the sum times the largest of the summands . therefore , we get @xmath414 \ ] ] which after some simple analysis boils down to @xmath415 which , in turn , implies that @xmath416 , thereby proving equation . now we are left with proving equation to finish the proof of theorem [ thm : scaling : limit ] . note that using lemma [ lemma : b - n ] and proposition [ prop : nu-1 - 0 ] , we can write @xmath417 as @xmath418 let us introduce @xmath419 and @xmath420 as two independent random intervals on two distinct free vertices , such that @xmath421 , and @xmath422 . notice that , @xmath423 where @xmath424 is the complement of @xmath420 on @xmath425 . also , observe that at criticality , @xmath426 thus , in view of the above observations , to show that equation is indeed true , it suffices to prove that @xmath427 which , in turn , can be derived as a direct consequence of the following lemma . [ lem : intersection ] @xmath428 in order to prove this lemma , we first need to introduce one more process related to our model , for which we shall recall the exploration process defined in section [ subsec : exploration ] . consider the collection of trees generated by the exploration process by ignoring the surplus edges . we refer to fig.1 of @xcite for an illustration of this tree . let us denote the various levels of the tree as _ generations_. now corresponding to each branch ( a path from the origin to a leaf ) define a process of the points @xmath429 around which intervals are constructed as described in the exploration process . each such process is clearly a markov chain conditioned on the event that each vertex in the cluster is visited only once . [ lem : lim : unif ] consider any process defined as above . let @xmath430 for some fixed @xmath431 be @xmath429 conditioned on interval @xmath366 , which was built around @xmath432 . then , there exists @xmath433 such that writing @xmath434 for normalized lebesgue measure on @xmath425 , and @xmath435 for the measure induced by the random variable @xmath436 on @xmath425 conditioned on the event that all the vertices appearing in tree are visited only once , we have @xmath437 * proof : * consider the process @xmath438 . note that following the same arguments as in the proof of proposition [ prop : nu - m ] , the event that each vertex in the cluster is visited only once has probability @xmath439 . thus @xmath440 is a markov chain on a set of probability @xmath439 . let us call this set @xmath441 . next , writing @xmath442 as the measure on @xmath425 , which is induced by the random variable + @xmath438 , and @xmath434 for the uniform measure on @xmath425 , by proposition [ prop : doeblin ] , we have the desired result.@xmath303 with these results , we shall now embark upon proving lemma [ lem : intersection ] . * proof of lemma [ lem : intersection ] : * in view of the fact that , @xmath443 , we can rewrite as follows @xmath444 next , note that although the conditioning with respect to @xmath389 adds no information to the length of the interval @xmath331 , however the conditioning does tell us the point @xmath429 around which the interval @xmath331 is constructed . therefore , there exists a measurable and bounded @xmath445 , such that @xmath446 . note that the sub - index @xmath25 in this case comes from the exploration process and differs from lemma [ lem : lim : unif ] , where it corresponds to the order on a branch . using this notation , we can introduce @xmath447 , where @xmath277 is a uniform random variable on @xmath425 , then naturally , @xmath448 . with this notation , it suffices to prove the following : @xmath449 by lemma [ lem : lim : unif ] , for any fixed @xmath26 and @xmath450 , we have @xmath451 and writing @xmath452 as the measure induced by @xmath453 conditioned on the event @xmath441 , we have @xmath454 where @xmath455 is a distance on tree . indeed , if @xmath25 and @xmath26 in and @xmath25 , @xmath26 and @xmath450 in belong to the same branch the above equations are direct consequences of the lemma [ lem : lim : unif ] . if @xmath25 and @xmath26 in belong to different branches but @xmath456 then there exists a common ancestor @xmath46 of @xmath25 and @xmath26 such that @xmath457 or @xmath458 implying by lemma [ lem : lim : unif ] the desired result . the same type of arguments imply when @xmath25 , @xmath26 and @xmath450 do not belong to the same branch . let us consider the lhs of equation , since @xmath459 is bounded , therefore @xmath460 as @xmath461 . thus , only the crossproduct terms may have some nontrivial contribution to the lhs of . however , we shall show that the crossproduct terms also vanish when divided by @xmath462 . clearly , @xmath463 -\e\left[(f(u , i))^2\right ] |\\ & \le & |\e\left[(f(x_i , i_j)(f(x_i , i_{j'})\right ] - \e\left[(f(u , i_j)f(u , i_{j'}))\right ] | + | \e\left[(f(u , i_j)f(u , i_{j'}))\right ] -\e\left[(f(u , i))^2\right ] | , \ ] ] where @xmath464 . notice that when restricted to the set @xmath441 , the expected value of @xmath465 converges to the expected value of @xmath466 by equation . however , we also note that the measure of @xmath441 converges to unity , which implies that the first part of the rhs of the above equation converges to @xmath259 . for the second part of the rhs , we notice that as @xmath467 , @xmath468 and @xmath469 are asymptotically independent in total variation , and that the limit does not depend on @xmath26 , we can also interpret this as for a fixed @xmath25 , as we go back far into the past , then the past is independent of the present ( by a simple application of baye s theorem ) . using this kind of argument , it follows that @xmath366 and @xmath470 are asymptotically independent as @xmath471 . ] , therefore , @xmath472 \to \e\left[(f(u , i)\right]^2 .\ ] ] which proves equation , thus proving the result . @xmath303 finally , this proves the claims mentioned in equations , and , which completes the proof of theorem [ thm : scaling : limit ] . another interesting process which appears naturally as an offshoot of the exploration process described in section [ subsec : exploration ] is the counting process corresponding to the surplus edges . recall from step ( iv ) of the regular exploration process from section [ subsec : exploration ] , that surplus edges are generated through connections to active intervals . thus at time @xmath25 , the probability of generating a surplus edge can be expressed as : @xmath473 the primary objective of this section is to prove that , under appropriate scaling , the aforementioned process converges to a counting process whose instantaneous intensity is given by @xmath474 . note that in order to prove the above convergence , it suffices to prove that the respective rates converge to the desired quantity . [ thm : surplus ] @xmath475 as @xmath385 . note that for counting processes , only the time need to be scaled , which explains the multiplicative factor @xmath0 appearing in the statement of the above theorem . * proof . * recall that @xmath476 . next divide collection of active intervals into two classes : those belonging to vertices which have been visited once , and the rest . however , by using an argument similar to the one put forth in proposition [ prop : nu - m ] , we can conclude that the number of intervals in the second group , which corresponds to the number of active intervals belonging to vertices which have been visited more than twice is @xmath477 . therefore , the probability of getting a surplus edge from connections to this class of intervals is @xmath330 . thus the contribution to equation is @xmath478 . simplifying the expression under consideration @xmath479 noting that @xmath480 , we get @xmath481 where recall that @xmath482 . the first part of the limit is primarily a consequence of theorem [ thm : scaling : limit ] , whereas the second part can be obtained by repeating the arguments used in lemma [ lem : intersection ] . @xmath303 as a consequence of the above theorem one can prove the joint convergence of the breadth first walk and the counting process to @xmath483 and a counting process whose intensity in given in the above theorem , respectively . recall the statement of theorem [ thm : joint : cgs ] . in this section , we shall wrap up the proof of this theorem by putting together all the results from previous sections . as pointed out in @xcite , theorem [ thm : joint : cgs ] primarily has two parts : * first is to prove that the excursions of the limit process are matched by the excursions of the breadth first random walk . * second is to arrange these excursions in the decreasing order . this can be achieved if one can ascertain that there exists a random point after which one is sure ( with high probability ) not to see large excursions . in order to settle the first issue , we shall invoke lemmas 7 and 8 of @xcite , which can be applied verbatim to our case , together with theorem [ thm : scaling : limit ] proved in previous subsection . thus , we need only be concerned about the second issue , for which we shall need to prove an appropriate version of lemma 9 of @xcite suited to our case . let us define @xmath484 @xmath485 as a consequence of theorem [ thm : scaling : limit ] @xmath486 [ lem:9:aldous ] let us denote by @xmath487 the probability that the quantum random graph with the parameters @xmath488 , contains a component ( collection of connected intervals ) of size greater than or equal to @xmath489 which does not contain any vertex @xmath25 with @xmath490 . then , @xmath491 * proof : * fix @xmath492 . let @xmath493 be the minimal vertex of the cluster @xmath494 , then for an interval @xmath138 , define @xmath495 then conditioned on component sizes , the labels of the vertices are going to be in random order ( because we are going to order the components in decreasing order ) . for a component @xmath494 of size @xmath496 , writing @xmath493 as the label of the minimal vertex of @xmath494 , define @xmath497 and @xmath498 then note that @xmath499 which implies that the conditional distribution of @xmath500 given @xmath501 , converges to exp@xmath502 as @xmath385 . the conditional exponential distribution implies @xmath503 conditional on cluster sizes being @xmath504 , @xmath505 at this point we note that using the arguments put forth in proposition [ prop : nu - m ] , we can conclude that @xmath506 with probability @xmath507 . then conditioned on cluster sizes being @xmath504 , we get @xmath508\right ) + o(n^{-4/3}).\end{aligned}\ ] ] recall the definition of @xmath509 from , and let us denote by @xmath510 the number of clusters of the size @xmath496 , then @xmath511 clearly , given the size of the component , the minimal vertex of various different components of the same size , are identically distributed , therefore , @xmath512 this , together with implies that @xmath513)+o(n^{-1/3}).\end{aligned}\ ] ] now observe that @xmath514 , then in view of the above set of inequalities , in order to prove the theorem , it suffices to prove that @xmath515 ) < \infty.\ ] ] however , notice that the expected number of components of size larger than @xmath489 , with minimal vertex smaller than @xmath516 is smaller than @xmath517 . moreover , if we define @xmath199 as the size of the over counting tree , then @xmath518 . thus , taking these observations into account , proving is equivalent to proving the following @xmath519 we note here that when @xmath520 , then proving for this case is rather easy . * * note that @xmath521 . also , @xmath522 where @xmath397 is the number of subtrees originating from the root , and @xmath523 is the size of the @xmath25-th subtree . clearly , @xmath524 , implying @xmath525 however note that @xmath526 ( say ) . therefore , @xmath527 which in turn implies that @xmath528 thus proving equation . * * however , for the case when @xmath529 , we shall need to adopt a more stringent approach . let us begin with @xmath530 therefore , it suffices to prove that @xmath531 define @xmath532 @xmath533 and @xmath534 in this notation , need to prove : @xmath535 recall @xmath536 by conditioning on @xmath397 , @xmath537^{m_n}{\rm e}^{-\theta\varepsilon^2})\ ] ] consequently , @xmath538,\ ] ] which can be rewritten as @xmath539,\,\,\ , \forall \theta\ ] ] but recall from equation that we need to prove @xmath540 , which boils down to comparing the above with @xmath541^{m_n-1}\right)$ ] . therefore , if we prove that for all @xmath542 , @xmath543^{m_n-1}\right ) > \theta,\ ] ] this will imply that @xmath544^{m_n-1}\right ) > \frac1{\varepsilon^2 } \log\e\left[\hat{\psi}(\theta\varepsilon^2)^{m_n-1}\right],\ ] ] then by monotonicity of the above function we can conclude that @xmath540 therefore , enough to prove equation which can be restated as @xmath545^{m_n-1}\right)>{\rm e}^{\theta\varepsilon^2}.\ ] ] since @xmath529 the above inequality is equivalent to @xmath546^{m_n-1}-1}{a\varepsilon^2}\right)>\frac{{\rm e}^{\theta\varepsilon^2}-1}{a\varepsilon^2}.\ ] ] let us take @xmath547 such that for all @xmath548 @xmath549 thus , we need to prove that for all @xmath548 @xmath546^{m_n-1}-1}{a\varepsilon^2}\right)>\frac{2\theta}{a}\ ] ] equivalently , @xmath546^{m_n}-[1-c(\theta)\varepsilon]}{a\varepsilon^2}\right)>[1-c(\theta)\varepsilon]\frac{2\theta}{a}.\ ] ] therefore , it is enough to prove @xmath546^{m_n}-[1-c(\theta)\varepsilon]}{a\varepsilon^2}\right)>\frac{2\theta}{a}.\ ] ] however , @xmath550^{m_n}\right ) & = & \e\left(\e\left([1-\varepsilon c(\theta)]^{m_n}| |i| \right)\right)\\ & = & \e\left[\left(1 - \varepsilon c(\theta ) + \varepsilon c(\theta){\rm e}^{-|i|/\lambda n } \right)^n\right]\\ & = & g({{\varepsilon}},\theta,|i|,n)\,\,\,\,\,\,\,\text{(say)}.\end{aligned}\ ] ] then it suffices to prove that @xmath551}{a \varepsilon^2}\right)- c(\theta ) > \frac{2\theta}{a}.\ ] ] note that @xmath552 . therefore , the above expression boils down to @xmath553\right)- c(\theta)>\frac{2\theta}{a}.\ ] ] now note that using the generalized mean value theorem , we have for @xmath554 @xmath555\end{aligned}\ ] ] thus implying that @xmath556\right)- c(\theta)\\ & \ge & \frac{1}{2a}\e\left ( c^2(\theta)\frac{|i|^2}{\lambda^2}\left(1 - { { \varepsilon}}c(\theta)\left(1 - { \rm e}^{-|i|/\lambda n}\right)\right)^{n-2 } + o\left(\frac1n\right ) \right)- c(\theta)\end{aligned}\ ] ] in view of the above and equation , it suffices to prove that @xmath557 equivalently , @xmath558 note that the above looks almost like a quadratic equation in @xmath559 , with the coefficient of the second order term being positive , and for large enough @xmath2 we can have @xmath560 which means that it suffices to prove that @xmath561 however , note that there exists a positive root , say @xmath562 , of this quadratic equation , such that for all @xmath563 , equation is satisfied , which proves the lemma . @xmath303 this lemma also concludes the proof of theorem [ thm : joint : cgs ] . 99 * m. aizenman , a. klein , c.m . newman * ( 1993 ) , percolation methods for disordered quantum ising models , _ mathematics , physics , biology ... , _ r. kotecky , ed . , 124 , world scientic , singapore . * d. aldous . * ( 1997 ) brownian excursions , critical random graphs and the multiplicative coalescent . _ the annals of probabilit_. vol . . 2 , 812 - 854 * s. bhamidi , r. van der hofstad , j. s. h. van leeuwaarden . * ( 2010 ) , scaling limits for critical inhomogeneous random graphs with finite third moments . _ electronic journal of prabability_. 15 , 1682 - 1703 . mr2735378 * s. bhamidi , r. van der hofstad , j. s. h. van leeuwaarden . * ( 2012 ) , novel scaling limits for critical inhomogeneous random graphs . _ the annals of probability_. vol . 40 , no . 6 , 2299 - 2361 . * b. bollobs . * ( 1984 ) , the evolution of random graphs . _ trans . _ 286 , 257 - 274 . * b. bollobs , s. janson , o. riordan . * ( 2007 ) , the phase transition in inhomogeneous random graphs . _ random structures algorithms _ 31 , 3 - 122 . * m. campanino , a. klein , j.f . perez * ( 1991 ) , localization in the ground state of the ising model with a random transverse eld , _ comm . _ 135 , 499515 . * p. erds and a. rnyi . * ( 1960 ) , on the evolution of random graphs._publ . inst . hung . _ 5 , 17 - 61 . * s. janson * ( 2007 ) on a random graph related to quantum theory . _ combinatorics , probability and computing , _ vol . 16 , number 5 , 757 - 766 * a. joseph . * ( 2011 ) the component sizes of a critical random graph with given degree sequence . * d. ioffe . * ( 2009 ) , stochastic geometry of classical and quantum ising models , _ methods of contemporary mathematical statistical physics _ , r. kotecky editor , lnm 1970 , 87 - 26 , springer . * d. ioffe and a. levit . * , ( 2007 ) long range order and giant components of quantum random graphs . _ markov processes and related fields _ , 13 , 469 - 492 . lof . * ( 1986 ) symmetric sampling procedures , general epidemic processes and their threshold limit theorems . _ j. appl . _ 23 , 265 - 282 . * t. uczak . * ( 1990 ) , component behavior near the critical point of the random graph process . _ random structures algorithms _ 1 , 287 - 310 . * a. nachmias and y. peres . * ( 2010 ) the critical random graph , with martingales . _ israel journal of mathematics_. vol . 176 , issue 1 , 29 - 41 . * a. nachmias and y. peres . * ( 2010 ) critical percolation on random regular graphs . _ random structures and algorithms - rsa _ , vol . . 2 , 111 - 148 . * o. riordan . * ( 2012 ) the phase transition in the configuration model . _ combinatorics , probability and computing . 21 , issue 1 - 2 , 265 - 299 * t. turova . * ( 2011 ) survey of scalings for the largest connected component in inhomogeneous random graphs . _ progress in probability_. vol . 64 , 259 - 275 . * t. turova . * ( 2013 ) diffusion approximation for the components in critical inhomogeneous random graphs of rank 1 . _ random structures & algorithms_. vol . 43 , issue 4 , pages 486539 . consider discrete time markov chain @xmath564 on circle of finite length , such that the transition probability density is given by @xmath565 , and that @xmath566 . * proof : * define a new coupled process @xmath568 given by , such that @xmath569 is uniformly distributed on the circle , and that @xmath570 , a fixed point on the circle . the joint process @xmath571 is made to evolve like a mc . define @xmath572 such that @xmath573 which gives @xmath574 next take a sequence of random variables @xmath575 which are i.i.d . then the relationship between the two sequences is given as follows : * if @xmath577 , then let @xmath578 , where @xmath579 is a freshly generated uniform random variable , and then onwards both the chains are equal , and the transition probability density is given by @xmath580 defined earlier . * if @xmath581 , then generate @xmath582 according to @xmath583 , and generate @xmath579 according to @xmath584 .

we have been represented the collective information of estimation procedures of parameters of the open clusters and put them together for showing the importance of clusters to understand their role in stellar evolution phenomenon . moreover , we have been discussed about analytic techniques to determine the structural and dynamical properties of galactic clusters . the members of clusters provide unique opportunity to determine their basic parameters such as , age , metallicity , distance , reddening etc . the membership probability of stars of clusters is assigned through the various approaches and each approach provides different number of probable members of the cluster . here , we have been briefly discussed about various approaches to determine the stellar membership within clusters . open star cluster spatial parameter-dynamical mf

1606.07223

the observational foundations of stellar astrophysics are studies of the sun and stellar clusters @xcite . open star clusters ( ocs ) or galactic open clusters are gravitationally bound system of stars formed together ( i.e. similar identified age ) and their stars are at practically the same distance with similar chemical composition whereas stars in a single cluster are differing only in mass @xcite . hence , they are dynamically associated system of stars which are formed from giant molecular clouds through bursts of star formation @xcite . the study of an individual star provides only limited , frequently inaccurate and uncertain information about it and possibly about the interstellar medium whereas , oscs provide an ideal opportunity to simultaneously study a group of stars located in a relatively small space @xcite . furthermore , it is believed that mostly stars in the galaxy are formed in open clusters @xcite , therefore , ocs are key probes for tracing the star - forming history ( sfh ) @xcite . such objects are excellent laboratories for studying the structure , kinematics and evolution of the galactic disc @xcite with chemical composition @xcite , consquently prescribed as the main building blocks of the stellar populations in our galaxy @xcite . these information of different locations of galaxy are prescribed about galactic evolution processes in the different time span @xcite . overall , ocs are found to be true representatives of the galactic disk population , systematic investigation of their nature , size , number of members , and age @xcite . furthermore , these systems are assigned solar orbits around the milky way centre @xcite . beside these , many questions remain unanswered , such as the location of inner spiral arms or determination of the co - rotation distance @xcite . these ocs are studied to understand the space - age structure , formation and evolution of the galactic disk @xcite and the stellar evolution as well @xcite . they are believed important test beds for star formation , stellar initial mass function ( imf ) and stellar evolution theories @xcite . mostly ocs are poorly studied or even unstudied up to now @xcite . their study is of great interest and attractive work in several astrophysical aspects and verification of their evolution theories . the galactic , radial and vertical , abundance gradient can be studied by ocs @xcite . in addition , these objects are good tools to analyze the large - scale proprieties of the disk of our galaxy and testing tools of the theories of stellar and galactic evolution @xcite . hence , their observational studies have become a traditional benchmark for testing our comprehension of several aspects of the formation and properties of the galactic disk @xcite . these studies are possible due to fact that their stars emerge from the same molecular cloud . the ocs have also provided constraints on the understanding of the step - like abrupt decrease in metallicity at 1 kpc from the sun @xcite . a substantial number of ocs is hidden behind the interstellar material in the galactic plane @xcite . researchers could be identified new clusters due to availability of data of huge infrared ( ir ) surveys and a low interstellar extinction in the ir wavelength range . this low value of extinction gives an opportunity to deeply dig into the spiral arm areas where most of the ocs are concentrated and more than a thousand ocs were discovered by analysing the two micron all sky survey ( 2mass ) @xcite . this survey has proven to be a powerful tool in the analysis of the structure and stellar content of ocs @xcite . this survey is also useful for investigating the star formation process and the stellar evolution @xcite . the structural parameters of ocs hold important information about mass like basic parameters and surrounding galactic tidal field @xcite . the position of a cluster centre and apparent ( angular ) size are needed to identify and describing an oc on the sky @xcite . these common parameters are present in numerous catalogues of available for about 1700 ocs @xcite . on the basis of their ages , they have divided into three groups such as young , intermediate and old - age ocs . the study of young clusters provide information about current star formation processes and are they also becomes key objects for describing questions about galactic structure , whereas observational studies of old and intermediate - age open clusters play an important role to study of theoretical evolution aspects of galaxy @xcite . the young ocs are effectively used to understand the nature of interactions between young stars and interstellar medium due to fact that they are still embedded in the parent nebulous regions @xcite . their members are categorized in various groups such as main sequence ( ms ) , pre - main sequence ( pms ) , etc . as pms stars become older , they contract towards the zero - age main sequence ( zams ) and their core temperatures rise @xcite . the study of dense central region ( the core ) as well as the expanded and sparse region ( the halo or corona ) is necessary to understand cluster evolution @xcite . the core contains relatively bright and massive ( @xmath0 3 @xmath1 ) stars along with low - mass stars @xcite , whereas the corona contains a large number of faint and low - mass ( @xmath2 1 @xmath1 ) stars . these regions has an important bearing on studies related to the mass function ( mf ) and the structure of ocs @xcite . a detailed structural analysis of coronae of ocs is needed to understand the effects of external environments , like the galactic tidal field and impulsive encounters with interstellar clouds , etc . , on dynamical evolution of ocs @xcite . the radial density profile ( rdp ) gives the picture of stellar densities within various concentric radial zones which can be used to identified the sparse ( poorly populated ) and rich clusters . the former ocs do not survive longer than a few hundred myr , whereas later ones may be survive longer @xcite . the fundamental physical parameters of ocs , e.g. distance , reddening , age , and metallicity are play an important role to study about the galactic disk . the depth of photometric analysis of ocs directly depends on the statistical completeness of the sample , especially in the faint - gc tail , and on the reliability of the derived parameters @xcite . the colour colour ( cc ) diagrams of an ocs is valuable tool for studying interstellar extinction in the directions of clusters @xcite . the colour magnitude diagrams ( cmds ) of photometric studies of clusters allow us to estimate their fundamental parameters @xcite . the described diagrams of ocs usually show a well - defined long and broad main sequence ( ms ) @xcite . typically , the determination of basic parameters of ocs via isochrone fitting requires either that the metallicity is estimated , or that an priory value is adopted @xcite . therefore the metallicity becomes a required parameter for the precise determinating of parameters of the ocs . though , many times solar metallicity is often assumed due to the complexity of the required observations and requirement of very indirect methods for its determination @xcite . high resolution spectroscopy is often claimed to be the most accurate method for determining the metallicity value of clusters @xcite , though the value of metellacity of individual member of dense cluster could not be possible through present instruments due to their resolution powers . we also note that individual metallicity studies might well be able to achieve a high precision for a limited number of stars @xcite . the quantitative comparisons between theory and observations of ocs suffer either obscuring material due to their location in the galactic plane or expected many field interlopers of given region @xcite . the small - scale structure , large - scale structure and observational errors produce distortion in redshift which leads to difficulties in determining the real cluster members @xcite . membership selection is therefore of crucial importance in the study of open clusters . the more precision cluster parameters could be computed through the membership analysis . it is well known fact that the proper motions ( pms ) of stars of an oc provide a unique opportunity to obtain membership information for these stars @xcite . these said proper motion values of the stars would be extracted from the online available catalogues @xcite . in recent years , the detailed membership analysis of stars in the cluster field has become a subject of intense investigation , mainly in view to understand the cluster properties @xcite . in nearby clusters ( d @xmath3 200 pc ) , the average large motions of cluster members from background stars makes easy identification of them @xcite . mass spectrum of these objects can be used to study the initial mass function ( imf ) @xcite . the shape of the initial mass function ( imf ) is effective parameter to drawn a theory about the fragmentation of molecular clouds and therefore the formation and development of stellar systems @xcite . the statistical analysis of the spatial distribution of the probable cluster members and their mass function can be useful to drawn some information about their dynamical evolution @xcite . mass segregation and evaporation of low - mass members are the main effects of dynamical interactions @xcite and these interactions are becomes probe to understand the evolving phenomena of the clusters . in the process of evolution , a significant fraction of the cluster stars is gradually lost to the field . dynamical evolution and mass segregation can have a significant effect on the radial structure as well as on the shape of the present - day mass function ( mf ) of open clusters . the mf of oc is usually obtained from the observed luminosity function ( lf ) , which can be obtained from the observed cmds @xcite . the universality of the imf of ocs is still an open question due to elementary considerations about the imf i.e. it depend on star - forming conditions @xcite . present estimations of the observed imf do not constrain the nature of the imf @xcite . identifying systematic variations of the imf with different star - forming conditions would allow us to study early cosmological events @xcite . imf has similar properties in very different environments and should depend only on the process of molecular cloud fragmentation @xcite . the difficulties such as complex corrections for stellar birth rates , life times , etc associated with determining the mass function ( mf ) from field stars are automatically avoided due to similar age and metaliicities of the members of ocs @xcite . the ocs are continuously harassed by external processes aged such as tidal stress and dynamical friction ( from the galactic bulge and disk and from giant molecular clouds ) , and by internal ones , such as mass loss by stellar evolution , mass segregation , and low - mass star evaporation @xcite . the most ocs are disrupting on a timescale of a few hundred myr due to galactic gravitational tidal forces @xcite , therefore most clusters seems to be relatively young . the other hand , the last stage of the evolutionary model of the clusters is believed to be a cluster remnant @xcite . the analysis of observational data becomes indicators to constrain the model of stars formation within the clusters . in a cloud of hydrogen gas , laced with helium and a trace of other elements , something triggers to a physical phenomena collapse , which leads stars formation processes of a range of different masses and begin their lives . the gravitational pull of the parent galaxy will disrupt the clusters and these objects present a snapshot of those stars , which have formed more or less at the same time and at the same place . as a result , these clusters are studied by the astronomers with the great interest . + the theory of stellar structure and evolution within cluster is based on the mass dependent evolution of stellar luminosities and temperatures with age . the dating of young , embedded clusters is considerably more difficult due to the majority of pms stars in the embedded clusters and inaccuracy in the measurement of their luminosities and temperatures . the embedded - cluster phase of evolution lasts somewhere between 3 and 5 myr @xcite . other hand , the ms of older clusters terminates at fainter and cooler levels . the changes of the stars during the end of their lives may be constrains by differences observed among stars away from the ms . star formation processes is occurred in two modes as spontaneous star formation and triggered star formation . the tendency of the interstellar matter of galactic scale is to be condense under gravity into star - forming clouds , which is counteracted by galactic tidal forces . the star formation is take place for that dense cloud , for which , its self gravity to overcome these tidal forces . in the spontaneous star formation ( ssf ) , the huge molecular cloud convert into a protostar by falling inward towards a centre i.e. collapse process of said cloud . this prescribed collapse occurs due to the instability of huge cloud . in the earlier ssf theories , it was believed that it is the result of the gradual cooling of the warm interstellar gas , bringing the thermal jeans mass to a value lower than the mass of the cloud . a cold cloud may be born by turbulence or magnetic tension and pressure or rotation . moreover , collapse of cold cloud occurs due to loses its turbulent energy via any dissipation mechanisms or loosing of magnetic field diffusion or disappearance of cloud s angular momentum due to magnetic tension . the matter accretion of cloud may be occurred due to the some combination of these three energies leads instability and mass change - without any loss of energy @xcite . a significant low mass stars are formed in molecular clouds in the vicinity of massive luminous stars @xcite . the interaction of @xmath4 radiation from @xmath5 stars with their surrounding molecular gas environments is necessary to study about star fromation processes and hydrodynamics of complex dynamic structures @xcite . this formation is take place due to the presence high pressure of gases . these mechanism is divided into three types , which are given as : ( i ) strong , ( ii ) moderate and ( iii ) weak triggering . a diffuse clouds is strike with a gravitational / radiation shock which leads stars formation through the compressed diffuse clouds . this phenomena is known to be * strong * triggering . such type star formation is not possible through the other alternating phenomena . the * moderate * triggering is increased the star formation rate in that molecular cloud region where stars already forms through the spontaneous star formation . thus , the shock increased the compression rate of gas in the * moderate * triggering which leads faster rate of star formation compare to the spontaneous star formation . the * weakly * triggering occurs due to the bulk motion of the gas , which leads varying star formation region with the bulk motion of the gas . furthermore , star formation rate remains constant in this triggering . + triggering is also divided into three groups according to the size of star formation region / rate @xcite . the small scale triggering involves the direct squeezing of pre - existing cloud or globules by high pressure ( nearly surrounds the whole cloud ) . an * intermediate scale triggering * is the compression of a cloud from one side . furthermore , a dense ridge of gas eventually collapses or recollects into denser cores in this triggering leads star formation . the * large scale triggering * is an accumulation of gas into an expanding shell or ring which partially surrounds the pressure source , a star or supernova . the all type of triggering may be result of a gravitational instability due to the density enhancement of a shock moving through a molecular cloud . recurrence of triggered star formation results in sequential star formation @xcite . these triggering processes are occurred due to the various external agents . the brief discussion of these agents are prescribed as below , the supernovae are those massive stars , in which explosions are occurred due to the stellar evolution . these explosions are giving rise to enormous amount of radiation to their surroundings leads to initiate star formation in nearby clouds by their supernovae - driven shock waves . numerical studies @xcite show that shock - waves with velocities in the range of @xmath6 corresponding to supernova explosions at a distance between 10 pc and 100 pc from the molecular cloud can induce such a collapse . massive stars emit very large amount of ultraviolet ( uv ) radiation . the surrounding gas region of these objects is ionized and heated due to the interaction of uv radiation . furthermore , such type regions are defined as the * h ii regions*. since , these regions are contained higher temperature ( @xmath7 k ) and higher pressure compare to its surrounding neutral hydrogen regions , therefore h ii region expands with a typical velocity of the order of @xmath8 . furthermore , its expansion rates decreases with time @xcite . h ii region may undergo with various events during the expansion @xcite . these events are given by various theories / models as follow : ( i ) radiation driven implosion i.e. rdi process @xcite , ( ii ) collect and collapse model @xcite , ( iii ) instability of ionization front , ( iv ) interaction between an h ii region and a nearby filamentary molecular cloud @xcite and ( v ) supersonic turbulent molecular medium @xcite . spiral arms of the galaxy are prominent regions of the ongoing star formation and fulfilled by the presence of young stars , h ii regions , dust and giant molecular clouds @xcite . the said fulfilled matter within arms is the indication of the stars formation processes . the fixed angular speed of rotation of enhanced density regions of spiral arms is defined by the pattern speed and said regions are considered to be waves of compression , i.e. density waves @xcite . the speed of density waves is found to be high for the inner part of the galaxy with reference to the arms , whereas low for the outer part of said arms of the galaxy . a large cloud may sudden compressed due to its interaction with the spiral arm density leads increment of local mass density with the collapse of cloud @xcite . the large scale distribution of molecular clouds of spiral arms can be generated through the interaction between spiral shocks and cold atomic gas @xcite . since , the cluster parameters are highly depends on the observed photometric stellar magnitudes of a particular portion of the sky or field of view ( fov ) of the cluster region . the cluster existent may be probably understood through the crowding behaviour of the stars within the fov of the observed sky region . though , clusters are not having circular shape but it is assumed to be circular for determining the cluster radius in a simple way . a general discussion about determining procedure of the various spatial parameters of the cluster as given below , for determining the cluster radius , it is mandatory to determine the cluster center . it is assumed that the center of cluster is a that point in which the stellar density becomes maximum . the star count method @xcite is a best method for determining the center of any cluster . though it does not guaranteed fact that the point , having maximum most stellar number / density , shows exact cluster characteristics @xcite . in the case of above prescribed fact , we would left maximum stellar point and the second maximum point would considered to be center of the cluster . as a result , the finding procedure of the center becomes iterate procedure until the exact cluster characteristics does not appear . furthermore , the radial stellar densities are used to define the stellar crowding effect within the cluster instead of stellar number . the stellar densities are used to overcome the condition of blank / dark regions of the cluster . such blank / dark regions appear in the small portion of the cluster , however , such effect would be disappears in the case of large scale structure . this fact is well matched with the homogeneity property of the large scale structure of the universe . + the above prescribed radial densities are estimated within various radial zones . if , a total of @xmath9 stars are found within any radial zone ( having area @xmath10 ) then the radial density of that zone estimated by the relation as : @xmath11 . the width of radial zone may be kept in the manner that it does not affected cluster characteristics . the smaller width of the zone of any cluster produces highly varied densities values due to the presence of undefined dark / blank regions within the cluster . the high width of radial zone reduce the precision of estimation value of the cluster radius . furthermore , the said width of the radial zone changes according to the stellar density of the studied cluster . the smooth cluster characteristics may be obtain for low radial width of the dense cluster compare than the width of the sparse cluster . thus the resultant profile of radial stellar densities are played an important role to determine the cluster extant . the radial densities profile ( rdp ) of any cluster contains two axis . the x - axis represents the radial distance , whereas y - axis represents the radial densities . since , it is believed that the field stars are symmetrically distributed in the nearby sky of the cluster , therefore , the cluster region is contained the stellar enhancement with respect to the field region . on this background , it is necessary to consider the field stellar density ( @xmath12 ) within the cluster . this said density must be reduced from the computed stellar densities of the various radial density . there are various statistical formulas to compute the radius of the cluster through these modal radial densities i.e. @xmath13 ( @xmath14 is the observed stellar density of the @xmath15 radial zone ) . the most applied formula for estimating the center density ( @xmath16 ) and core radius ( @xmath17 ) of the cluster is king - empirical formula @xcite as given below , @xmath18 the said core radius of the cluster is defined as the radius where the density , @xmath19 , is half of the central density , @xmath16 @xcite . other hand , the radius of the cluster is defined as the distance from the centre to the point where the stellar density of the cluster merged with the field region . since , it is noticeable fact that the best fitted curve of the king - empirical model and model background density i.e. @xmath12 , are never intercept to each other . since , this fact leads to an infinite extent of the cluster and produce a contradiction to compute the cluster extant , therefore the field stellar density is defined as the sum of 1-@xmath20 error with the @xmath12 . thus , the radius of the cluster is the distance from the center of the cluster to interception point between the best fitted curve and field stellar density . though , the 1-@xmath20 error is the statistical prediction for estimating the cluster radius and the resultant radius would be changed with significance level of the estimation such as 1-@xmath21 or 3-@xmath21 . on this background , the radius of the cluster is also estimated through the estimation error in background stellar density as follow , @xmath22 since , the above prescribed radius depends on the estimation error in the @xmath12 , therefore , it may be varied according to said estimation error . this error depends on the smoothness of the radial density profile , which further depends on the width of the assumed radial zone . the width of radial zone is selected by researcher on the behalf of his own sense . thus , resultant radius may not representing the true radius of the clusters . it is believed that the gravitational attraction among stars of the cluster is vanished beyond the cluster periphery , however it does not satisfied the reasons of sudden disappearing of attraction . though , the cluster members are shown the evaporating nature from the cluster region , so the disappearance of the said weakly gravitational bound may occurred before the end of extent of the cluster . thus , the gravitational bounding among stars of the cluster might not hold together in the adjacent stellar enhancement region of field region . as a result , it is needed to define a radial distance to describe the possible gravitational bounded region of the cluster and such region is defined by the limiting radius . this limiting radius is a radial distance , after which no stars are gravitationally bound to the cluster system @xcite . this said limiting radius is computed by the following relation @xcite , @xmath23 though , the estimation error of the field stellar density is also influenced the value of said radius , however computed value might to be reliable in the sense that the said value with uncertainty lies within the stellar enhancement region . on the basis of limiting radius , the maximum borderline background stellar density i.e. @xmath24 is estimated by the following relation , @xmath25 this limiting radius is also used to compute the concentration parameter ( @xmath26 ) following as @xcite , @xmath27 this relation is directly indication of positive value of @xmath26 for that stellar system , in which the limiting radius is greater than the core radius . furthermore , this parameter may be characterized the density profile of each cluster in such a way that it is independent of its diameter and richness . besides above prescribed radii , we have found tidal radius of the cluster , which is computed through the total cluster mass ( @xmath28 ) as follow @xcite , @xmath29 here we have adopted a galactocentric distance @xmath30 and a milky way - like galaxy with an enclosed mass of @xmath31 @xcite . thus , the tidal radius is estimated using the relation given as @xmath32 @xcite . the various limitation of rdps is described as below @xcite , + ( i ) the rdps is dependent on the limiting magnitudes of the detected stars . it seems that the cluster radius may be increased with the limiting magnitude of detected stars . + ( ii ) in some cases , the weak contrast between cluster and field region in outer region creates inaccuracy for defining the angular size of the cluster . + ( iii ) the exact cluster boundaries has not been defined due to the irregular shape of the cluster . + ( iv ) the identification procedure of the centre seems to be difficult for the sparse and poor cluster due to presence of clump in rdp . + ( v ) the cluster members becomes diminishes with the increasing radial distance due to inability of detection them from the statistically fluctuation of the field stars . the surface number density @xmath33 is defined by the number of stars per magnitude interval in unit area . furthermore , it is a function of magnitude and radius from the cluster center @xcite . the sndps are very effective to estimate the cluster extent and the nature of stellar density within various magnitude bins @xcite . these profiles are highly effected due to the presence of data incompleteness and stellar crowding effect . furthermore , the characteristic curves of this profile are overlap to each other at the periphery of the cluster . as a result , these profiles are least used to determine the cluster extent . beside this , these profiles of clusters are the signature of stellar distribution of various masses leads the revolutionary information about the stellar evolution processes within the clusters . generally , the cluster radius has estimated by considering the spherical shape of the cluster but cluster shape is non - spherical . this non - spherical shape is occurred due to the galactic tidal field @xcite , differential rotation of the galaxy , encounters with giant molecular clouds @xcite . thus , the semi - major ( _ a _ ) and semi - minor ( _ b _ ) axes of the apparent distribution of cluster members gives ellipticity @xmath34 following as , @xmath35 the semi - axes are considered to be the principal axes of an ellipse using characteristic equations for the second - order moments of tangential coordinates of the 1@xmath20 cluster members @xcite . however , this results may be influenced by the membership procedure of the cluster . the reddening of the cluster is a colour - excess but every colour - excess has not representing this value . this prescribed reddening value has computed through the observed stellar magnitudes in @xmath36 and @xmath37 photometric bands . it is represented by the @xmath38 . the solution values of the best fitted isochrone of @xmath39 cmd has provided the its value following as , @xmath40 where @xmath41 and @xmath42 are the observed and intrinsic colour - excess values for detected stars . the above said reddening value may be obtained by any one method from the following described approaches . oscs are very frequently affect by the large colour excesses and this affection may be @xmath43 of the extinction across the whole cluster region @xcite . the solution of linear fit to the main - sequence ( ms ) in the @xmath46 versus @xmath47 plane of the cluster may obtained through the following relation @xcite , @xmath48 those ms stars are selected for the linear fit in above relation , which have @xmath49magnitude less - than or equal to 18 mag . the resulting value of @xmath50 from the above relation would be used to determine the reddening by following munari & carraro @xcite relation , @xmath51 this above prescribed relation is applicable for the @xmath52 furthermore , this relation has used in the certain colour - range as @xmath53 . the reddening , @xmath38 , in the cluster region is estimated using the @xmath54 vs @xmath44 tcd . to determine the reddening , we have to take only those stars which have spectral type earlier than @xmath55 as later type stars may be more effected by the metallicity and background contamination @xcite . in this approach , we would be added some shifted values on the colour - excesses [ @xmath56 and @xmath42 ] of the zero age - main sequence ( zams , @xcite ) , these shifting values are chosen in such a way that they must be satisfied the following relation @xcite , @xmath58 the error in reddening is calculated using the following relation @xcite , @xmath59 where @xmath60 has considered to be 0.01 . the @xmath61 vs @xmath62 diagram of each cluster provides the value of the ratio , @xmath63 through best fitted isochrone and these values are satisfied the relation , @xmath64 @xcite in case of normal reddening law . the reddening , @xmath38 is estimated using the following relations @xcite , @xmath65 where @xmath38 is the reddening . the study of interstellar extinction in the nir is effective to examine the nature of interstellar matter / dust . a @xmath66 vs @xmath67 diagram is also used to determine the interstellar extinction in the near - ir ( nir ) range . we have been added colour - values in such a way that they are approximately satisfied the following given relation , @xmath68 finally , reddening value can be computed through the relation @xmath69 @xcite . these relations are also used to examine the nature of reddening law towards to the clusters region . if , the resultant reddening value through this approach is well matched with the reddening value through @xmath54 vs @xmath44 tcd , then extinction law is said to be normal . generally , normal reddening law is applicable when dust and intermediate stellar gases are absent in the line of sight of the cluster @xcite . in addition , the extinction law is found to be normal for solar neighbourhood , the galactic centre or in moderately young open clusters and associations @xcite . if , the values are not matched to each other , then it becomes indication of the presence of stellar dust / gases . on this background , dust evolution of the star - forming regions becomes one of the more interesting issues in the astronomy due to direct dependency of grin size with the reddening law @xcite . the apparent stellar magnitudes are found to be different for different photometric bands . since , the stars are radiating the stimulated radiation in every bands , therefore , the absolute magnitudes can be computed through the these different apparent magnitudes . it is believed that absolute magnitude of a star is a fixed value in any band , which leads to co - relate the observed colours value with prescribed absolute magnitude . thus , the absorption relations are defined by those transformation relations / equations , which are used for converting the absorbed colour values into absolute magnitudes . the brief notes of these relations are given below , in the view of @xmath70 , the correction in reddening values have been carrying out through absorption relation @xcite as below , @xmath71 @xmath72 a total selective absorption of the stars is estimated through the reddening as follow @xcite , @xmath73 similarly , the @xmath44 and @xmath45 colour excesses may be related as follow @xcite , @xmath74.\ ] ] for simplicity , the mean intrinsic colour i.e. @xmath75 , of the clusters is taken to be 0.0 mag . as the light emitted from the star cluster , it passes through the interstellar dust and gas , hence the scattering and absorption of light takes place since the absorption and scattering of the blue light is more compared to the red light , the stars appear reddened . the reddening law is expected to be different in the presence of dust and gas and its nature is investigated by using @xmath76 tcds @xcite , where @xmath77 is any broad band filter other than @xmath36 and @xmath37 . the @xmath76 tcds have been widely used to separate the influence of the extinction produced by the diffuse interstellar material from that of the intra - cluster medium @xcite . therefore , a best linear fit in the resultant @xmath76 tcd diagram of each cluster gives the value of slope ( @xmath78 ) for the corresponding tcd . assuming the value of @xmath79 for the diffuse interstellar material as 3.1 , a total - to - selective extinction @xmath80 is determined as follow @xcite , @xmath81 the relation @xmath82 @xcite is also used to determine the unknown values of colour - excess through above prescribed value of @xmath80 . theoretical isochrones gives opportunities from the simple age - dating of star clusters , to more complex problems like the derivation of star formation histories of resolved galaxies @xcite . a classical isochrone of a fixed metallicity can be represented by a single line in the hertzsprung - russell diagram @xcite . these prescribed isochrones are providing the distance and age of the cluster by their best fitted solution on the various cmds . the broadness of stellar scattering is mainly caused by the presence of binaries and field stars in the cluster region . it is also noticed that the prescribed broadness of the fainter members are not reducing after the field star decontamination through the statistical cleaning procedures . there are three key issues in the determination of the distance and age of the clusters as follow , ( i ) the accurate metallicity of the cluster , ( ii ) the exact location of the turnoff point on the main sequence and ( iii ) the visual inspected solution of the best fitted isochrone . + the distance and age of the clusters are determining by keeping the fixed reddening and metallicity values . the exact value of the metallicity of the cluster are computing through the spectral analysis of its members spectra . the morphology of the clusters can significantly influence by internal interaction of two - body relaxation , which are result of encounters among member stars and external tidal forces either due to the galactic disc or giant molecular clouds @xcite . in the case of young clusters , the initial morphology should be governed by the initial conditions in the parent molecular cloud @xcite . isodensity contour of a cluster is defined by that region , for which , obtained stellar density is found to be fixed value . since , the stellar density of a cluster is varied from center to its periphery , therefore , the isodensity contour are containing a region , in which , stellar density is found between limits of any given density bins . the contour of higher density seems to be embedded within the contour of lower density . it is also appeared that the contours of higher densities are spread out in the cluster and not connected to each other , whereas outer boundary of other contours are shown similarity with the cluster periphery . the radial existent of these boundary must be least than that of the cluster radius . thus , the prescribed contours seems to be excellent tracer of the initial morphology of the stellar distribution . the mass of cluster ( @xmath83 ) is computed through the value of the galactocentric distance @xmath84 as follows @xcite , @xmath85 where @xmath86 and @xmath36 are oort s constants valid for @xmath87 , @xmath88 and @xmath89 are the tidal radius and gravitational constant @xcite , respectively . these oort s constants of the open clusters subsystem could be easily expressed by their local values as @xcite , @xmath90 furthermore , these local values are provided exact constant as follows , @xmath91 @xmath92 where @xmath93 and @xmath94 . the relative random error of the cluster mass as expressed as @xcite , @xmath95 where @xmath96 and defined as the relative random errors of the tidal radius . half mass radius ( hmr ) of the clusters defined by a radial distance whose periphery contained the half mass of the cluster . the hmr of any cluster is entirely different from the the half physical radius of the clusters . the periphery of hmr is divided the whole cluster regions into the two equal mass parts such as ( i ) the central part and ( ii ) the halo part . other hand , central part within the periphery of half physical is contained the one - third area of the remaining area . thus , the cluster hmr seems to be more than its half physical radius . the prescribed hmr value is useful to estimation of the dynamical relaxation time of the studied cluster . the spatial position of members is changed due to the stellar encounters among them and these encounters lead equi - partition of energies among the cluster members . the data incompleteness fraction is necessary to derive the dynamical properties of the cluster due to the fact that the it is not always possible to detect all the stars present in the ccd frame , particularly towards fainter end @xcite . these incompleteness test are performed through @xmath97 routine in @xmath98 . in this routine , randomly selected artificial stars of known magnitudes and positions are added in the original frames . the cluster characteristics remains unchanged by adding few hundred stars , which probably @xmath99 of the detected stars in the original frame . this said procedure are performed 5 - 10 times for each original image . however , it may also becomes a cause of error for the dynamical properties of the clusters . thus , there are two contributions to the errors of dynamical properties : uncertainties on the correction factor sand uncertainties on the number of stars @xcite . moreover , the correction factor effectively reduce the influence of data incompleteness . on this background , the dynamical behaviour of any cluster is understood as follow , the luminosity function ( lf ) is expressed as total number of cluster members in different magnitude bins @xcite . these members in various magnitude bins are computed through the following relation @xcite , @xmath100 where @xmath9 , @xmath101 and @xmath102 are the cluster members of the cluster , number of observed stars in the cluster and field regions respectively and @xmath103 is the completeness factor for the considered ms brightness in the selected magnitude bins of any photometric band . the k - band luminosity function ( klf ) seems to be a powerful approach to understood the imf of young embedded star clusters on the behalf of several studies @xcite . the klf of these objects may be obtained through following relation @xcite , @xmath104 where @xmath105 is the slope of power law , whereas @xmath106 is total number of stars per unit @xmath107magnitude bin . in the initial process of star formation within ocs , an amount of the stellar mass within in per unit volume known as the initial mass function ( imf ) , which is very effective to determine the subsequent evolution of cluster @xcite . the direct measurement of imf does not possible due to the dynamical evolution of stellar systems though we have been estimated the mass function ( mf ) of cluster . the mf is defined as the relative numbers of stars per unit mass and can be expressed by a power law @xmath108 . the slope , @xmath109 , of the mf can be determined from @xmath110 where @xmath111 is the number of stars per unit logarithmic mass . the masses of mpms can be determined by comparing observed magnitudes with those predicted by a stellar evolutionary model if the age , reddening , distance and metallicity are known . the classical value of mf slope is -1.35 @xcite . this effect may be arisen due to the dynamical evolution or imprint of star formation or both @xcite . through the process of mass segregation , the stellar encounters of the cluster members are gradually increased . the stellar encounters among of members of clusters gradually lead to an increased degree of energy equipartition throughout the lifetime of the cluster @xcite . furthermore , the higher - mass cluster members gradually sink towards the cluster center by transferring their kinetic energy to the more numerous lower - mass stellar members of the cluster @xcite . this phenomena is also statistically verified through kolmogorov - smirnov test . + to understand the mass segregation phenomena , the members of the clusters are subdivided into two / three groups of different stellar magnitude ranges . we have plotted the two dimensional diagram of the normalized commutative frequency with the radial distance of the cluster . the said normalized frequency is defined as the ratio commutative stellar numbers to the total number of members within the given magnitude range . the convex type plot of these defined groups indicates the mass segregation phenomena within the studied cluster . the relaxation time ( @xmath112 ) of a cluster is defined as the time in which the cluster needs from the very beginning to build itself and reach the stability state against the contraction and destruction forces , e.g. gas pressure , turbulence , rotation , and the magnetic field @xcite . in other words , this time ( @xmath113 ) is also defined as the time in which the individual stars exchange energies and their velocity distribution approaches a maxwellian equilibrium and mathematically is expressed as @xcite , @xmath114 where @xmath115 is the number of cluster members , @xmath116 is the radius containing half of the cluster mass and @xmath117 is the average mass of the cluster stars . both theory and simulations show that significant mass segregation among heaviest stars in the cluster core occurs in the local relaxation time , but affecting a large fraction of the mass of the cluster requires a time comparable to the average relaxation time averaged over the inner half of the mass @xcite . if , said value is found to be very low compare to the cluster age , then the studied cluster is said to be dynamically evolved system . similarly , the cluster is defined as the dynamically relax for other cases . whenever , we have to looked towards a small portion of the sky through scientific - eye i.e. telescope , then we are obtained many stellar enhanced region in the somewhere of sky . these stellar enhanced region of the milky - way must be possible stellar clusters . these clusters are mostly influenced by the foreground and background stars . the nature of field star sequences can only be clarified with a difficult membership analysis procedure @xcite . as a result , it is needed to define a factor / procedure for describing the membership reasons / status for detected stars within the precised clusters . since , we have knowledge about position , mean proper motion and observed stellar magnitudes within various photometric bands , therefore , these information may be indicated the possibility of a star belonging to a particular cluster region . this said possibility must be defined by the membership probabilities of star , which are calculated by the various statistical approaches based on the known stellar parameters . in the simple words , the probability associated with the position is identified as the spatial probability , whereas that related to the proper motion is referred as the kinematic probability . similarly , there are some statistical cleaning procedure are based on the observed colour - excess and stellar magnitudes and defined as the statistical probability . the criteria based on these probabilities is expected to yield the most probable members of a cluster . the members of open clusters have the same age , distance , proper motion and spatial velocity , hence they follow a pattern in the colour - magnitude diagrams and proper motion plane . the location of these most probable members in cluster s cmd s then becomes a tracing tool of the best visual fitted isochrones @xcite . as such , the probabilities needed for the estimation of membership of stars within a clusters may be described as follows . these probabilities are statistical values which are computed by using the values of stellar radial distance . the basic assumption of this type probability that the probability of stars at the center of cluster is 1 , whereas 0 for the stars of cluster periphery . furthermore , said probability is considered to be 0 for the outsider stars from the cluster periphery . these probabilities are also subdivided into various categories according to the assumed variation nature of resultant probability as follow , the linear decrements of stellar probabilities with the radial distance is the basic principle of this type probability . for this , the probability of the detected stars has decreased in the relative fraction of radial distance to the cluster radius and it is written in the form of following mathematical expression @xcite , @xmath118 where @xmath119 is the radial distance of @xmath15 star from the center . thus , resultant probabilistic values are dependent on the estimated center of the cluster , whereas independent from their dynamical parameters . the uniform decrements of the probabilistic values with the radial distance is unique property of its . similarly , the values are unaffected by the cluster size . in this probabilistic approach , said values have decreased in the term of square of relative fraction @xmath120 . as a result , the resultant formula have been obtained as , @xmath121 this term is representing the variance of the probability according to the square law . the photometric probability ( @xmath122 ) is computed with reference to the blue and red limits in the @xmath123 vs @xmath37 cmd . the blue sequence may be defined by using empirical zero - age - main - sequence ( zams ) colours @xcite shifted in magnitude and colour to visually match the cluster sequence . the red sequence is defined by a shift of -0.75 mag in @xmath37 and a shift of @xmath124 mag in @xmath123 in order to account for unresolved ms binaries @xcite . stars with b - v colours lying within the binary sequence of the ms are assigned @xmath122 of 1 and are probable members of the cluster , while stars deviating along either direction are assigned a probability as , @xmath125 ^ 2 \}\ ] ] where @xmath126 is the difference between colours from blue or red colour limits and @xmath127 is the photometric error in colour . it is well known that the proper - motion ( pm ) of member stars of a cluster usually differs the values from the mean - proper motion of the cluster . as a result , the deviation of proper - motion of a member star in both @xmath128 and @xmath129 directions with respect to the mean - proper motion of cluster provides a measure of its kinematic probability . the said proper motion values are provided different probabilistic values for a star on the behalf of adopted statistical approaches . we may be seen several membership approach in the literature . here , we are discussed some well known approach following as , in this procedure , the mean - proper motion values of the detected stars ( which are laying within cluster periphery ) has computed by 3@xmath20 clipping with iterative procedure . the mean and @xmath20 values of such proper motion of each cluster have been determined in both the directions and are used to separate the stars not lying within the 3@xmath20 value in both ra and dec directions . using iteration procedure , the stars lying outside 3@xmath20 value from the mean value of proper - motion of cluster / stellar - system in both directions have been completely eliminated and the remaining stars are used to find the mean proper motion of the clusters . the kinematic probability of stars has been derived using these prescribed mean proper motion values through the relation given as @xcite , @xmath130}\ ] ] where @xmath131 and @xmath132 . the @xmath133 and @xmath134 are the proper motion of any star in the direction of @xmath128 and @xmath129 respectively , while @xmath135 and @xmath136 are the uncertainties in measurement of the proper motion of stars along the respective directions . in the past , the kinematic probabilistic algorithm through frequency distribution had been developed for members having proper motions of different observed precisions @xcite . in later work , the correlation coefficient of the field star distribution to the set of parameters describing their distribution on the sky @xcite . the construction of frequency distributions have proposed for the cluster stars ( @xmath137 ) and field stars ( @xmath138 ) @xcite . the frequency function for the @xmath15 star of a cluster can be written as follows , + @xmath139 \right\},$ ] + where @xmath140 and @xmath141are the pms of the @xmath142 star , while @xmath143 and @xmath144 are the cluster s pm centre . @xmath145 is the intrinsic pm dispersion of cluster member stars , and ( @xmath146 ) are the observed errors in the pm components of the @xmath142 star . the frequency distribution for the @xmath142 field star is as follows , + @xmath147 + @xmath148 \right\},$ ] + where @xmath149 and @xmath150 are the field pm centre , while @xmath151 and @xmath152 the observed errors in the pm components . in addition , @xmath153 and @xmath154 are the field intrinsic pm dispersion values and corelation coefficient @xmath155 is calculated by , @xmath156 the spatial distribution may be excluded due to the small observed field of stars @xcite . the distribution of detected stars within observed field can be calculated as , @xmath157 where @xmath158 and @xmath159 are the normalized numbers of stars for the cluster and field ( @xmath160 ) . finally , we have defined the membership probability for the @xmath15 star following as , @xmath161 this procedure is believed to be good indicator of the cluster and filed star separation . since , the contamination by background and foreground objects can not be avoidable during observations , therefore , the effectiveness of the membership determination would be estimated by following relation @xcite , @xmath162 } } { { \sum_{i=1}\limits^{n}}p_{i}{\sum_{i=1}\limits^{n}}(1-p_{i})},\ ] ] where @xmath115 is the total number of stars under consideration for the determination of the membership probability and @xmath163 indicates the probability of @xmath15 star of the cluster . in addition , the resultant larger value of @xmath164 represented larger effectiveness in the determination of membership probability of the stars . this type probability are generally based on the comparison of the cluster and field cmds . according to this method , we removed all cluster stars in a colour - magnitude plane , which fall within a assumed size grid cell . in addition , the field region is taken in such a way that the area of field region must be equal to the area of the cluster region . these equal areas are needed due to symmetric distribution of field stars within observed region of the sky . the field region is also chosen in two ways : ( i ) the loop region around the cluster periphery and ( ii ) one or two degree off from the coordinate of cluster centre . the remaining stars through this approach are called as the statistically cleaned members of the cluster . the statistical probability of above said members is considered to be 1 , whereas 0 for other members . the various statistical probabilistic approach are described as below , the decontaminated cmds have been used to investigated the nature of star cluster candidates and derive their astrophysical properties through various developed astrophysical tools @xcite . it is also believed that the 2mass photometric cmd through field star decontamination has shown to constrain age and distance significantly more than the observed photometry , especially for low - latitude and/or bulge - projected clusters @xcite . in this procedure , the detected stars within the cluster periphery are divided into multi - dimensional ( magnitude and colour axis ) grid of cells . one dimension is referred as the magnitude , whereas others are colour . in the next step , total stellar ( members+filed ) number @xmath165 within cluster periphery is computed for each cell and also estimated the field stellar number @xmath166 in corresponding cells of field region . finally , the similar @xmath166 stars has been subtracted from the cell of cluster region and these remaining stars may be used for further analysis . in this procedure , every field star has its own grid with fixed dimension in such a way that the coordinate of field star becomes centre of its grid . the one dimension is magnitude and others are colours . furthermore , the dimensions of these grids are also intersected to each other . in addition , we have identified those cluster stars , which are laying within assumed dimensions of the grid of field star . after it , the colour - magnitude distances of these stars have been computed from the field star and stars can be arranged in the increasing order of said distances . the star , having shortest distance from the field star , is eliminated from the cluster membership . this described procedure is iterated for every field star . thus , one star from the cluster region has eliminated for a field star . the remaining stars are defined to be statistically cleaned members of the cluster . the stars are distributed overall the cmd and they may be further divided into various sub - types according to their position in the plane of cmd . the brief description about these types as follow , these are o to late - b type stars which can be satisfied the following conditions @xcite , + ( i ) @xmath167 , @xmath168 , @xmath169 , @xmath170 and @xmath171 johnson s @xmath172 ; + ( ii ) an individual distance modulus between @xmath173 and @xmath174 to take into account the effect of binary members and photometric errors @xcite , where @xmath175 and @xmath176 are the mean distance modulus and the width of the gaussian fit of the distance modulus , respectively . the ionization front ( if ) is generated due to the ionization of cluster surrounding by uv radiation of these stars @xcite . the if drives a shock into the pre - existing molecular clumps and compresses it , consequently the clumps become gravitationally unstable and collapse to form next - generation stars @xcite . in the case of anomalous reddening within the cluster , the intrinsic magnitudes of stars , having spectral type earlier than @xmath177 were obtained as follows @xcite , @xmath178,\ ] ] where @xmath179 and @xmath180 is @xmath38 value of individual stars estimated from the @xmath50 method . these stars are used for estimating the reddening value of the cluster . the early type stars have been loosing their masses through the phenomena of ms evolution and this fact is found to be maximum in those young open clusters , which have age to be @xmath181 @xcite . it is believed that about @xmath182 of the early type stars @xmath183 of the young open clusters are showing be - phenomena @xcite . however , this phenomena may be overabundance ( @xmath184 ) due to presence of herbig be stars , which also associated with nebulosity and show large near - ir excess @xcite . the nir excess stars with h@xmath105 emission stars are important targets to study the effect of environment on the disc around the young stellar objects ( ysos ) . before occupying the position in the zams , stars surrounded by optically thick material consisting of an infalling envelope and accretion disc @xcite . disc strength as measured by excesses in intrinsic colours such as @xmath62 or @xmath185 . these colours are depend on the stellar mass and radius , and disc properties such as accretion rate and geometry @xcite . the intrinsic @xmath185-excess stars are those which having the foreground - reddening corrected colour @xmath186 redder than the @xmath187 reddening vector @xcite . it is emphasized that the @xmath185-excess stars are brighter than @xmath188 . the @xmath189-excess stars of the 14 myr open cluster ngc 4755 do not occupy the classical disc locus of unreddened t - tauri stars @xcite , however theoretical models predicts classical full discs @xcite . the disc evolution process around pms stars may be understood through the mass acceleration rate @xcite . the large amount of uv excess may be obtained due to intrinsic properties of late - type pms stars rather than to inappropriate reddening corrections @xcite . the limit of @xmath4 excess from chromospheric activity is seems to be about @xmath190mag @xcite . the accretion luminosity @xmath191 of these stars may be estimated the through the relation @xcite , @xmath192 where @xmath193 . by using a bandwidth ( 700@xmath194)and zero magnitude flux of @xmath195 for the bessell @xmath196 band @xcite , the luminosity ( @xmath197 and @xmath198 ) have been computing from expected magnitude of the photospheric colour of ms stars ( @xmath199 ) and the extinction corrected magnitude of stars with @xmath4 emission ( @xmath200 ) . the values of radius ( @xmath201 ) and mass ( @xmath202 ) and the accretion luminosity are used to estimate the mass accretion rate ( @xmath203 ) through following relation @xcite , @xmath204 where @xmath89 is the gravitational constant . blue stragglers are those stars which are seem to stay on the main sequence longer than expected time through the standard theory of stellar evolution @xcite . furthermore , these high mass objects seem to belong to the thick disc or halo population on the basis of their metallicity or kinematics @xcite . these said members can be identified through the relation between the @xmath44 index of the turnoff and the metallicity of the coeval , metal - poor , thick disc and halo populations @xcite . these objects appear to be above the turnoff leads as a horizontal - branch stars of post - main - sequence stars @xcite . furthermore , these members are also located above the turnoff , blue - ward direction of the cmd and on or near the zams @xcite . the analysis of four lithium - deficient stars indicates that stragglers produced by mass exchange in close binaries @xcite . spectroscopic analysis of radial - velocity studies of metal - poor field blue stragglers suggest that the observed abundances ( including the lack of lithium ) are consistent with stable mass transfer during the asymptotic - branch stage of the primaries @xcite . the giant members are mostly found in the intermediate and old age open star clusters . these members are located in the redder side of the main sequence in the cmd . these stars are those stars which are evaporated from the main sequence due to the dynamical evolution . the possible cause of their evaporation from the ms may be fast evolution rate due to their heavy mass . in addition , the situated above and right side of the turnoff point of best fitted isochrone . moreover , giant members are the brighter members of the cluster . the position of giant members with the ms stars gives the opportunity of better visual fitting of the theoretical isochrone of known metallicity . the wast work is carried out in the field of star cluster properties . the known cluster parameters are assembled in two comprehensive and frequently updated databases , webda and thecatalog of optically visible open clusters and candidates " ( daml02 ) @xcite . these studies are carried out on the basis of various statistical procedures as discussed in the earlier sections of this manuscript . the parameters of cluster @xmath205 is also estimated through basian analysis @xcite . the general discussion about cluster properties of some clusters as given below , the spatial properties of the clusters have been obtained from various photometric catalogues , such as 2mass @xcite , ppmxl @xcite and other available @xmath206 database . basically , the spatial properties of the clusters are varies with the the nature of gathered data . the integrated @xmath207 parameters and the luminosity function of 650 galactic open cluster represented by kharchenko et al 2009 @xcite . the fitting of king s model to open clusters creates difficulties due to the relatively poor stellar population ( compared to globular clusters ) and from the higher degree of contamination by field stars in the galactic disk @xcite . though , the stellar crowding effect may be helped to identify the clusters but some time false estimation may be occurred . recently , dolidze 39 , bh 79 , and ruprecht 103 seems to be field fluctuation instead of the clusters @xcite . the analysis of concentration parameter of the clusters indicates that the angular size of the coronal region is about 6 times the core radius @xcite . the limiting radius of the clusters may vary for individual clusters between about 2@xmath208 and 7@xmath208 @xcite . the cluster radius is found to be larger in infrared bands compare to the optical bands @xcite . the cluster mass could simply be derived from the summation of masses of individual members @xcite . the identification procedure of cluster members is highly influenced by the limiting magnitude of detected stars and uncertainty in the determination of these magnitude . thus , sufficiently accurate and uniform mass estimates for a significant number of ocs in the galaxy will obtained through the deeper surveys and an increasing accuracy of kinematic and photometric data @xcite . since , the broadness of stellar scattering has been reducing through the field star decontamination , therefore we gets a opportunity to obtained more accurate results . for example , the mf slopes @xmath209 and @xmath210 are derived for ngc 637 and 957 by considering the corrections of field star contamination and data incompleteness @xcite . the identification of exact members of any cluster is most crucial and mystical work due to the least information of them . there are several approaches to their identification but each approach has its own limitations . it is believed that the most probable members of cluster may be satisfied the pattern of theoretical isochrones but their are very limiting work for this purpose . the data incompleteness is a major factor to influence the cluster characteristics . this said factor is highly changed the mf values for the fainter members of the clusters leads the deviation of slope of mf from the classical salpeter value . this variation may leading the revolutionary changes within the cluster morphology . however , the addstar routine has been utilized to overcome the effect of said incompleteness but it is highly dependent on the detected stars through the applied photometric techniques . presently , there are no alternative approach to verify the precision of applied completeness factors . + mostly , the dynamical studies are carried out either in visual @xmath37-band or in 2mass - band , but the cross verification of their results is further needed due to not matching results from these studies . the stellar enhancement provides the cluster existence which is highly contaminated by the presence of field stars . in addition , the detection of stars is highly effected by the stellar crowding within the cluster region , which leads low detection rate of field stars of the cluster compare than surrounding . it is also noticed in several work that the cluster periphery has been computed before applying the completeness correction . since , the completeness correction of field and cluster region are found to be different , therefore , it is needed to develop a procedure of estimating the cluster radius , which may be linked with the completeness corrections . the estimation error of stellar magnitude creates major difficultly to determine the characteristics of fainter members of the cluster . the deep observations may reduces this difficulty but there are needed a methodology for increasing the accuracy of these members . + the turn point of ms play a major role in the estimation of distance , age of the cluster through the theoretical isochrones . the identification of this point is highly dependent on the stellar distribution of cmd and appeared width of the ms in said cmd plane . since , the said turn point of the cluster is mostly defined by the visual inspection of the obtained stellar distribution in the cmd , therefore , it is needed to constrain a procedure / model for determining the said turn point , which would be very effectively used to verify the turn point through the visual inspection . the synthetic cmd method and bassian analysis had utilized to constrain the evolutionary isochrone through the observed data points . for these purpose , pre - determined cluster members are used , therefore , the effect of field stars and spread stellar distribution is also demanded some attention . + the determined distance and age of the cluster is also related with the metallicity . generally , the value of metallicity has been selected through the pre - consideration . there are no photometric technique for determining this . the spectroscopic observations of massive stars of the clusters are utilized to determine the metallicity of the cluster . such type spectroscopic observations of the clusters are very limited and also the the spectroscopic study of the very fainter members are not available . on this background , the accurate measurement of the metallicity of the cluster is not available at the present . since , the known cluster parameters depend on the pre - assumed metallicity of the cluster , therefore , the derived fundamental parameters of the cluster may be far away from the reality . on this background , the deep spectroscopic observations would be needed to determine the exact metallicity of the clusters . + it is well known fact that the cluster radius is determined by considering the the spherical shape of the cluster , which is far away from the reality . it is needed to use an equation of other possible shape to determine the cluster extent , which may be change the present understanding towards the clusters morphology . since , the cluster are placed within the three dimensional ( 3d ) space and we are determined their extension by considering them as a two dimensional objects , therefore , the new 3d models are needed to constrain the clear picture of cluster extent . the limiting radii and core radii of the clusters are linked with the density profile , therefore , the identification of exact cluster center becomes most important due to fact that it may be occurred different for the non - spherical clusters . furthermore , the isodensity contour may become signature of the identification of the said center . the fraction of massive stars of a cluster may also signature of its ongoing evolutionary process leads dynamical behaviour and stage of the studied cluster . + the mean proper motion of the cluster is another site to understand the dynamical picture . the analytic study may open an opportunity to understand the future of cluster and varying characteristic of the clusters with time . since , the known proper motion values of the clusters are least reliable due to larger uncertainty in their estimation , therefore , the comprehensive ground based observations needed for this purpose . such type study may further leading to analysis the gravitational field within the cluster . the varying mean proper motion of various stellar mass bins may be useful to verify the mass segregation phenomena within the cluster . it is also needed to study about the possible changes of the proper motion of the cluster members with time . its deep analysis may also helpful to estimate the variation of evaporation rate of cluster members . + the interstellar dust grains / clouds can be identified through the the analysis of tcds . since , said tcds are highly depends on the detected stars in the various bands . the observed magnitudes are varied due to the the quantity of the dust in the direction of cluster , which may reduce the brightness of the stars . since , the interstellar dust / cloud has uniformly distributed within the cluster region , therefore , the reduced fraction of the members are varied according to their mass . thus , it is necessary to applying magnitude corrections on the detected stars . such correction must be changed their distribution on the cmd plane leads changes in the fundamental parameters of the cluster . in addition , these tcds are also effectively used to determine the stellar subcategory and their analysis may helpful to understand the star formation mystery within the cluster . + besides these , there are several oscs , for which , we have been very poor knowledge . the studies of this poor studied cluster and unknown cluster would be increased the database , which would be effectively utilized to constrain / verify the galactic evolution models . the combined results of the clusters may provides information about their evolution process and its dependency on their size . the locations of oscs can be effectively characterized the nature of density waves within the spiral arms . in conclusions , the wide database of the clusters seems to be the building blocks of proposed evolutionary model of the our galaxy and the star formation sites . this research has made use of the vizier catalogue access tool , cds , strasbourg , france . the original description of the vizier service was published in a&as 143 , 23 . gcj is also thankful to ap cyber zone ( nanakmatta ) for providing computer facilities . gcj is also thankful to shree nilamber joshi for providing the friendly environment in the my present working place , which becomes a milestone of my research work . schmidt - kalar th ; 1982 , in : landolt / bornestein , numerical data and functional relationship in science and technology , new series , group vi , vol 26 , scaifers k. , & voigt h. h.(eds . ) springer - verlog , berlin , p.14

we investigate the electron density distribution and the stability of stripe phases in the realistic two - band model with hopping elements between orbitals at ni sites on the square lattice , and compare these results with those obtained for the doubly degenerate hubbard model with two equivalent orbitals and diagonal hopping . for both models we determine the stability regions of filled and half - filled stripe phases for increasing hole doping in the range of , using hartree - fock approximation for large clusters . in the parameter range relevant to the nickelates , we obtain the most stable diagonal stripe structures with filling of nearly one hole per atom , as observed experimentally . in contrast , for the doubly degenerate hubbard model the most stable stripes are somewhat reminiscent of the cuprates , with half - filled atoms at the domain wall sites . this difference elucidates the crucial role of the off - diagonal hopping terms for the stripe formation in lasrnio . the influence of crystal field is discussed as well .

cond-mat0604238

stripe phases are one of the most exciting phenomena of modern condensed matter physics . they have been observed in a variety of systems , including nickelates , @xcite cuprates , @xcite and manganites . @xcite among them , layered la@xmath3sr@xmath4cuo@xmath5 ( lsco ) , la@xmath6nd@xmath7sr@xmath4cuo@xmath5 ( nd - lsco ) , la@xmath3sr@xmath4nio@xmath5 ( lsno ) , and la@xmath8nio@xmath9 ( lno ) compounds play plausibly the most prominent role . however the similarity between them is superficial only , and the stripes in the cuprates differ from the stripes in the nickelates in many respects . for instance , they are dynamical in the former , and static in the latter . in addition , in nd - lsco @xcite and lsco,@xcite one finds the so - called _ half - filled stripes _ , with the density of one doped hole per two atoms along the domain wall ( dw ) . in contrast , it is clear from a variety of experiments that magnetic states within doped nio@xmath10 planes of the nickelates are _ filled stripes _ with density of one doped hole per one atom in a dw . @xcite the question of filling is not the only difference between the nickelate and cuprate stripes , however . neutron diffraction measurements performed on nd - lsco revealed that magnetic peaks are displaced from the antiferromagnetic ( af ) maximum at @xmath11 to the points @xmath12 and @xmath13 and the shift @xmath14 depends linearly on hole doping @xmath15 for @xmath16 , while it is almost constant at higher doping . these values correspond to a superposition of _ vertical _ ( 01 ) and _ horizontal _ ( 10 ) dws . the essentially identical modulation and doping dependence of @xmath14 was observed in superconducting crystals of lsco with @xmath17 . conversely , experiments on lsno established that spin order is characterized by the wave vectors @xmath18 with @xmath19 for @xmath20 , corresponding to a constant charge of one hole / ni ion along a _ diagonal _ dw , in agreement with the predictions made in the pioneering works by zaanen and gunnarsson@xcite and others , @xcite and emphasized recently,@xcite which predicted theoretically the stripe order . more precisely , neutron scattering measurements have revealed static stripe order in the lno samples with @xmath21 , 0.125 , as well as 0.133,@xcite and even over a wider hole doping regime @xmath22 in the case of lsno . @xcite moreover , the incommensurate ( ic ) stripe order persists up to @xmath23 in the nd@xmath3sr@xmath24nio@xmath5 system,@xcite in which the low temperature orthorhombic phase seems to extend to a higher doping region @xmath25 as compared to the la compounds , where the high temperature tetragonal phase is stabilized instead already at @xmath26 . @xcite indications of a charge order ( co ) in doped lsno were also found in electron @xcite and x - ray diffraction studies . @xcite quite recently , the one - dimensional nature of the stripe modulation within nio@xmath10 planes has been directly confirmed in the transmission electron microscopy ( tem ) studies of charge stripes in lsno.@xcite in addition , careful examination of the tem images has shown that at a low temperature stripes are mainly centered on rows of ni atoms . however , a mixture of the so - called site centered ( sc ) and bond centered ( bc ) stripes was also observed in some small regions of the sample . in contrast to lsco , charge and spin order in lsno is characterized by the wave vectors @xmath27 and @xmath28 , see ref . , respectively , with @xmath19 corresponding to a constant charge density of one hole / ni ion along the diagonal stripe . note that the doping dependence of @xmath14 is exactly the same as that found in the seminal hartree - fock ( hf ) calculations within the hubbard model.@xcite experimentally the data give a clear evidence that the stripe order is most stable at hole doping @xmath29 . indeed , @xmath30 and @xmath31 increase linearly with @xmath15 , reach a maximum at @xmath29 with 240 and 180 k , respectively , and then decrease monotonously upon further doping . the particularly robust stability of stripes at @xmath32 stems from the cooperative spin and charge modulation expressed by the coincidence of the spin superlattice peaks at the wave vectors @xmath33 with those of co , given by @xmath34 . one should also note that @xmath14 starts to deviate gradually from the value given by the @xmath19 law above @xmath29 . in fact , co is even more stable than the stripe order at @xmath35 and forms a checkerboard pattern below the transition temperature @xmath36 k. remarkably , with decreasing temperature , a stripe co sets in below @xmath37 k and its incommensurability is twice as large as that of the spin order with the much lower onset temperature @xmath38 k.@xcite the low temperature competition of the checkerboard and stripe order at @xmath35 has also been clearly indicated by the measurements of raman and optical conductivity . @xcite interestingly , above this doping , incommensurability tends to saturate rapidly with the value @xmath39.@xcite apparently , co itself induces commensurate values of @xmath14 and such a commensurability effect seems to be an intrinsic property of the stripe order . at the same time , the increased hole density at the dws suppresses the superexchange energy gain , best optimized when all the holes are accommodated within stripes . therefore , it is the af order that drives @xmath14 to approach the value given by the linear relation @xmath19 below @xmath31 . obviously , when @xmath29 , both effects cooperate which results in the locked - in value of @xmath40 over the entire @xmath41 range . furthermore , the temperature dependence of the specific heat @xmath42 shows a distinct anomaly at the same temperature @xmath43 k suggesting a co transition.@xcite this conjecture is supported by the temperature dependence of the logarithmic resistivity . indeed , as expected for this transition , @xmath44 of the sample with @xmath45 exhibits a steep increase precisely at 240 k.@xcite charge fluctuations around @xmath43 k also lead to conspicuous changes in the optical conductivity spectra @xmath46.@xcite namely , at a high temperature @xmath47 , only a broad peak is observed with a finite low - energy spectral weight . however , when @xmath48 is decreased down to @xmath30 , the low - energy weight below 0.4 ev is gradually transferred to higher energy so that the opening of the charge gap is clearly observed . a special character of @xmath29 as well as @xmath35 doping , is best seen in the magnetic susceptibility @xmath49 and logarithmic resistivity @xmath44 , recorded at 200 k , showing distinct anomalies at these doping levels.@xcite in this context it is important to discuss a peculiar behavior of the hall coefficient @xmath50 and the seebeck coefficient @xmath51 at two representative values of temperature below and above @xmath43 k , i.e. , at 300 k and 210 k. the hall coefficient at 300 k is almost independent of doping and takes a small positive value corresponding to the order of one hole carrier per ni site.@xcite the seebeck coefficient shows a similar nearly constant behavior taking , however , a negative value . in contrast , below @xmath30 , both @xmath50 and @xmath51 change their signs from negative to positive at @xmath29 and their absolute values are larger than those at 300 k. in addition , for samples with @xmath45 and @xmath52 , @xmath50 keeps decreasing with decreasing @xmath48 , so that the number of carriers per ni site gets reduced even down to @xmath53 . these results indicate that the deviation of @xmath15 from @xmath54 can be considered as an electronlike ( @xmath20 ) or holelike ( @xmath55 ) carrier doping into the @xmath56 charge - ordered insulator with three ni sites in the unit cell . hence , for the doping level @xmath29 there is exactly one hole per unit cell and such a state is robust and may be considered as a half - filled one . moreover , it would certainly retain this feature if the incommensurability had followed precisely the relation @xmath57 . however , @xmath14 has a tendency to shift towards @xmath54 , for both sides of the @xmath29 point , which has important implications for the sign of @xmath50 . on the one hand , when @xmath15 is less than @xmath54 , the number of holes is insufficient for filling up the mid - gap states entirely , i.e. , the states inside the charge transfer ( ct ) gap induced by stripe order , and in this case the mid - gap states contain some electrons which become carriers . on the other hand , for @xmath15 larger than @xmath54 , the number of electrons is insufficient for filling up the lower hubbard band entirely , which initially contains holes . consequently , @xmath50 is expected to have the opposite sign to that in the case of @xmath20 ( hole carriers ) . next , the chemical potential shift @xmath58 in lsno for @xmath59 deduced either from x - ray ( xps ) or ultraviolet photoemission spectroscopy @xcite is suppressed . certainly , this phenomenon can not be explained within a simple rigid - band framework in which @xmath60 is expected to shift downwards with increasing hole doping . in fact , an increase of @xmath15 in a system with a spatially uniform hole distribution should enhance the average hole - hole repulsion which , in turn , would result in a higher energy required to add one hole to the system , i.e. , in a larger @xmath61 . therefore , the absence of @xmath58 implies that the average hole - hole repulsion remains nearly unaltered upon doping . such a behavior might be easily explained within a stripe picture in which a constant hole density at the dws implies that the interwall distance decreases and @xmath14 increases linearly with increasing @xmath15 . moreover , a similar suppression of the shift has been found below @xmath62 in lsco , suggesting that an inhomogeneous charge distribution is a common feature of both systems . several methods have been employed to investigate the stripe phases which go beyond the hf approximation , such as : density matrix renormalization group,@xcite slave - boson approximation,@xcite variational local ansatz approximation,@xcite exact diagonalization ( ed ) of finite clusters,@xcite analytical approach based on variational trial wave functions within the string picture,@xcite dynamical mean field theory,@xcite cluster perturbation theory,@xcite and quantum monte carlo simulations.@xcite they all address the crucial role of a proper treatment of local electron correlations in stabilization of the half - filled stripe phases in the cuprates . in spite of this huge effort , it remains unclear whether dws are centered on rows of metal atoms as in sc stripes , or if they are centered on rows of oxygen atoms bridging two metal sites in bc stripes . although no evidence was presented yet , it seems that the degeneracy of @xmath63 orbitals plays an important role in stabilizing filled stripes in the nickelates . in the simplest picture developed for the cuprates , the cu@xmath64 ions forming dws are spinless , while the cu@xmath65 of the af domains carry a spin @xmath66 . in the nickelates , filled dws are formed of the ni@xmath64 ions ( @xmath66 ) , whereas the af domains consist of ni@xmath65 ions ( @xmath67 ) . therefore , a realistic hamiltonian for lsno has to contain , besides the @xmath68 orbital which is occupied by one hole in the parent compounds of the superconducting cuprates ( such as la@xmath10cuo@xmath5 and yba@xmath10cu@xmath69o@xmath70 ) , as well the @xmath71 orbital at each ion , so as to account for the high spin ( @xmath67 ) state of the stoichiometric compound . indeed , when extending their approach to a more realistic four - band peierls - hubbard model for nio@xmath10 planes , zaanen and littlewood have shown@xcite that doped holes prone to form diagonal dws , centered on rows of ni atoms , with a tendency to have a ferromagnetic ( fm ) alignment of the reduced ni magnetic moments at a dw . in fact , subsequent multiband hf calculations emphasized the relevance of the electron - lattice coupling depending on its strength one can obtain either metal- or oxygen - centered structures . @xcite therefore , it appears that the semiclassical theory captures the essence of stripe physics in the nickelates . the stripes were also obtained in the theory using either the hf approximation applied to a multiband ct model , in which both the nickel and oxygen degrees of freedom are explicitely taken into account,@xcite or by the ed of finite clusters within the effective two - band model ( [ eq : h_deg ] ) extended by the coupling of @xmath0 electrons to the lattice.@xcite however , due to a large number of basis states , the latter calculation has solely been done for an eight site cluster allowing only for investigating stripe phases with a small unit cell observed experimentally at high doping levels @xmath29 and @xmath35 . in contrast , the hf approximation used here allows one to describe charge and spin modulation with larger unit cells and hence it should provide an answer to the important question whether the description of nio@xmath10 planes by the simplified effective @xmath0 model described below yields results consistent with the predictions of the ct model.@xcite an important aspect of these studies , qualitatively different from the pioneering work on the cuprates where the relevant models yield dws with _ nonmagnetic _ cu@xmath64 ions,@xcite is that dws in the nickelates are formed of ni@xmath64 ions carrying a _ spin @xmath66 leading to a fm polarization around them . in this work we investigate various stripe phases resulting from an effective hamiltonian for @xmath0 electrons relevant to the nio@xmath10 layers , and compare them with those obtained in the doubly degenerate hubbard ( ddh ) model . we use the hf approximation in order to gain the qualitative insight into possible stripe phases and their stability . this approach is the first step to make in order to identify possible generic instabilities towards stripe phases in nio@xmath10 planes , in analogy to the early work on cuo@xmath10 planes which predicted the existence of stripes in the high @xmath72 cuprates.@xcite the hf method is well suited to compare the energies of different types of magnetically ordered phases , as it approaches the same limit at @xmath73 as better mean - field theories , such as slave - boson approach or the gutzwiller ansatz.@xcite in addition , the hf approach allows one for efficient calculations on large clusters at low temperatures which are necessary to obtain unbiased results concerning the stability of various stripe phases , particularly when the magnetic ( and charge ) unit cells are large at low doping . the paper is organized as follows . the models with orbital degeneracy are introduced in sec . [ sec:2bhm ] . the energy of various stripe phases , filled vs. half - filled , diagonal vs. vertical , together with the influence of the various parameters of the model , are presented in sec . [ sec : gse ] . their structures are further described in sec . [ sec : stripes ] , together with the mechanism which leads to their stability , illustrated by the double occupancy distribution and by the respective densities of states . the paper is summarized in sec . [ sec : sum ] , where we also present our main conclusions . even though doped nickelate lsno is isostructural to its cuprate counterpart lsco , the electronic degrees of freedom in lsno are more involved . in fact , a minimal realistic hamiltonian for lsno must contain , besides the @xmath68 orbital states included in the cuprate oxide models , also the @xmath71 orbital states . such a model of interacting @xmath0 electrons in a 2d ( @xmath74 ) plane may be written as follows , @xmath75 with two orbital flavors , @xmath76 and @xmath77 , forming a basis in the orbital space . the kinetic energy is described by @xmath78 with @xmath79 where @xmath80 stands for an effective @xmath81 hopping matrix element due to the hybridization with oxygen orbitals on ni@xmath82o@xmath82ni bonds , and the off - diagonal hopping @xmath83 along @xmath84 and @xmath85 axis depends on the phase of the @xmath86 orbital along the considered cubic direction . the electron - electron interactions contain only on - site terms , which we write in the following form , @xmath87 where @xmath88 and @xmath89 stand for the intraorbital coulomb and hund s exchange elements . we also used @xmath90 for total electron density operators , given by the sum of densities in orbitals @xmath91 . the last term @xmath92 describes the uniform crystal - field splitting between @xmath86 and @xmath93 orbitals along the @xmath94 axis , @xmath95 the splitting between the @xmath0 orbitals originates from the tetragonal jahn - teller distortion of the nio@xmath70 octahedron . in la@xmath10nio@xmath5 , however , the octahedron where the ni@xmath82o@xmath82ni in - plane ( out - of - plane ) bond lengths are 1.95 ( 2.26 ) ,@xcite respectively , is much less distorted as compared to 1.89 and 2.43 bonds in la@xmath10cuo@xmath5 , @xcite which reflects the difference in electron filling . in what follows we consider only a realistic positive @xmath96 favoring , due to elongated octahedra , the @xmath93 electron occupancy over the @xmath86 occupancy in doped compounds . in order to elucidate the mechanism leading to the stabilization of the stripes , we also investigate the ddh model . it is obtained using the hopping matrix , @xmath97 instead of eq . ( [ eq : h_kin ] ) . note , that the total bandwidth is @xmath98 in the @xmath0 model and @xmath99 in the ddh model ( at @xmath100 ) , but the average diagonal hopping elements are the same this ensures that electrons are approximately similarly correlated for a given @xmath88 in both cases , while the electron - electron interactions have in the ddh model the same form given by eq . ( [ eq : h_int ] ) , provided one labels the orbitals as @xmath15 and @xmath101 . the model given by eq . ( [ eq : h_deg ] ) is very rich . even when restricting oneself to solutions with only two atoms in the unit cell , one obtains an intricate competition between numerous fm and af phases , even on the level of mean field ( mf ) approximation.@xcite in particular , we obtained phase diagrams showing , in the regime away from half filling , a clear tendency towards ferromagnetism for @xmath102 , or at least to the @xmath103-af phase for @xmath104 . in the two - band tight - binding model for degenerate @xmath0 orbitals ( @xmath100 ) . note a different scale for both fermi surfaces . , width=302 ] in the two - band tight - binding model for degenerate @xmath0 orbitals ( @xmath100 ) . note a different scale for both fermi surfaces . , width=302 ] some of the richness of the present model ( [ eq : h_deg ] ) can be traced back to the peculiar features of the electronic dispersion for @xmath0 electrons in two dimensions.@xcite unlike for spins , two bands obtained by diagonalizing the hamiltonian ( [ eq : h_deg ] ) in the noninteracting limit are nondegenerate ( except for some high symmetry directions):@xcite @xmath105 with @xmath106 , @xmath107 , @xmath108 , and @xmath109 for @xmath110 the corresponding fermi surfaces are depicted in fig . [ fig : fs ] . at half filling one piece of the fermi surface is building a `` swiss cross '' , while the second one is building a square . the filling of the upper band is then by 1/4 electron , while that of the lower band is by 3/4 electron per spin . the fermi surface of the upper band shrinks rather fast with increasing doping @xmath15 and below @xmath111 ( @xmath112 ) one finds that this band is empty and all electrons are in the lower band . however , both bands are partly filled in the doping regime @xmath2 relevant for the nickelates , and they will both play a role also when electron interactions are included . at half filling the edges of the pieces of the fermi surface ( fs ) for both bands are connected by the nesting vectors : @xmath113 and @xmath114 . at the same time , the edges of one piece are connected to the edges of the other one by the nesting vector @xmath115 . in addition , at half filling both components of the velocities vanish on their respective fermi surfaces . therefore , several competing instabilities are expected once the interaction is turned on . their relative importance can be estimated by computing the spin - spin correlation function in the noninteracting limit for the corresponding wave - vectors . in the static limit it reads : @xmath116 here the matrices @xmath117 diagonalize the hopping matrix eq . ( [ eq : h_kin ] ) , and @xmath118 are given by eq . ( [ eq : bands ] ) . for @xmath0 electrons for increasing doping , as obtained for : ( a ) @xmath100 , and ( b ) @xmath119 . different lines correspond to : @xmath120 ( solid line ) , @xmath121 ( dotted line ) , and @xmath122 ( dashed line ) . van hove singularity for @xmath120 is obtained at doping higher than that realized experimentally in nickelate lsno oxides : ( a ) @xmath123 , and ( b ) @xmath124 . , width=302 ] since ferromagnetism is an expected feature of the model ( [ eq : h_deg ] ) , we also evaluate the homogeneous spin - spin correlation functions . it shows an intricate interplay of all the three instabilities ( see fig . [ fig : bubble ] ) . they were all reported , regardless of the approach , see for instance refs . and . one finds that the van hove singularity takes place at @xmath125 for the crystal - field splitting @xmath126 , see eq . ( [ eq : ez ] ) . at this point the line where the velocity vanishes has a large component on the fs , and the singularity is logarithmic . note that there is no nesting vector connecting different pieces of the fs , as can be understood by looking at fig . [ fig : fs ] . here one finds a distinct difference between the present @xmath0 model and the ddh model , where instead the fs is nested and takes the shape of a square for both bands at the van hove point ( for @xmath127 ) . therefore , nesting and van hove singularities in the @xmath0 electron model influence the physical properties at different electron densities than in the ddh model . in the following the hamiltonian given by eq . ( [ eq : h_deg ] ) is treated in the hf approximation . we introduce explicity four `` diagonal '' expectation values , @xmath128 , and four `` off - diagonal '' ones , @xmath129 , with @xmath130 and @xmath131 , that we determine self - consistently in order to get a complete characterization of the ground state . due to symmetry the last four fock averages vanish for a 2d square lattice , and one gets the simpler mf hamiltonian : @xmath132 with , @xmath133 and , @xmath134 + \tfrac{1}{2}e_z , \nonumber \\ m^{}_{iz\sigma } & = \tfrac{1}{4}\sum_{i}\big [ ( 3u-5j_h)\langle n_{i}\rangle - \lambda_{\sigma } ( u+ j_h)\langle m_{i}\rangle \nonumber \\ & + ( u-5j_h)\langle o_{i}\rangle + \lambda_{\sigma } ( u- j_h)\langle p_{i}\rangle \big]- \tfrac{1}{2}e_z,\end{aligned}\ ] ] where we introduced @xmath135 for @xmath136 spin and @xmath137 for @xmath138 orbital , and the operators : @xmath139 for total charge and magnetization density , and for orbital anisotropy in the charge and magnetization distribution , respectively . site centered and bond centered stripes represent equally relevant candidates for the ground state . however , it numerically turns out that the energy of sc stripes is substantially larger than the one of the bc stripes , in agreement with the early study of yi et al.@xcite ( see table [ tab : fez0 ] in sec . [ sec : sum ] as well ) . we therefore concentrate on bc stripes . they are made out of pairs of atoms with a fm spin polarization within the model hamiltonian ( [ eq : h_deg ] ) in the wide doping regime @xmath140 . in order to obtain unbiased results , we performed calculations on large clusters implementing the symmetry of stripe phases in reciprocal space . we worked on squared clusters with the linear dimension along the @xmath15 direction chosen as an _ even _ multiplicity of the elementary stripe unit cell dimension . their sizes are listed in table [ tab : egf ] . .1 cm cccccccc & & & & & & & & & & 0.05 & 11 & 88@xmath14188 & 2.8811 & & 11 & 88@xmath14188 & 2.8794 + 0.06 & 11 & 88@xmath14188 & 2.8376 & & 11 & 88@xmath14188 & 2.8326 + 0.07 & 11 & 88@xmath14188 & 2.7946 & & 11 & 88@xmath14188 & 2.7870 + 0.08 & 11 & 88@xmath14188 & 2.7519 & & 11 & 88@xmath14188 & 2.7430 + 0.09 & 10 & 80@xmath14180 & 2.7099 & & 10 & 80@xmath14180 & 2.7001 + 0.10 & 10 & 80@xmath14180 & 2.6677 & & 9 & 72@xmath14172 & 2.6570 + 0.11 & 9 & 72@xmath14172 & 2.6258 & & 8 & 64@xmath14164 & 2.6140 + 0.12 & 8 & 64@xmath14164 & 2.5839 & & 7 & 84@xmath14184 & 2.5712 + 0.14 & 7 & 84@xmath14184 & 2.5001 & & 6 & 72@xmath14172 & 2.4853 + 0.16 & 6 & 72@xmath14172 & 2.4162 & & 6 & 72@xmath14172 & 2.3997 + 0.18 & 5 & 80@xmath14180 & 2.3329 & & 5 & 80@xmath14180 & 2.3130 + 0.20 & 5 & 80@xmath14180 & 2.2485 & & 4 & 80@xmath14180 & 2.2291 + 0.25 & 4 & 80@xmath14180 & 2.0394 & & 4 & 80@xmath14180 & 2.0155 + 0.30 & 3 & 72@xmath14172 & 1.8325 & & 3 & 72@xmath14172 & 1.8029 + 0.40 & 3 & 72@xmath14172 & 1.4619 & & 3 & 72@xmath14172 & 1.4081 + [ tab : egf ] to ensure that the model ( [ eq : h_deg ] ) is indeed relevant for the nickelates , it is necessary to focus on appropriate values of parameters @xmath88 , @xmath89 , and @xmath96 . first of all , x - ray absorption ( xas ) measurements @xcite as well as from xps and bremsstrahlung - isochromat spectroscopy studies @xcite of the electronic structure of lsno suggest that lsno is a ct insulator with nearly the same ct energy @xmath142 as that of nio . therefore , the same parameters could be accepted as used before in the self - consistent born calculations , @xcite which reproduced quite well photoemission spectra of nio . secondly , since our approximation has a tendency to overestimate the effect of the coulomb repulsion in lsno , we used a somewhat smaller value of the coulomb repulsion @xmath88 than that adequate for nio , i.e. , we set @xmath143 ev . next , as the in - plane ni@xmath82o@xmath82ni bond length in la@xmath10nio@xmath5 of 1.95 (see ref . ) is very much the same as the shorter bond of 1.89 in la@xmath10cuo@xmath5,@xcite we set the hopping amplitude @xmath144 between the @xmath145 orbitals and @xmath146 orbitals to be as in lsco , i.e. , @xmath147 ev.@xcite this in turn yields an effective in - plane ni@xmath82ni hopping @xmath148 ev . however , it is more convenient to take the effective @xmath81 hopping element connecting two @xmath149 orbitals along the @xmath94-axis as the energy unit @xmath80 , related to @xmath150 via the slater - koster relation @xmath151 ev , so that @xmath152 . the value of hund s exchange between @xmath153 electrons in nio was estimated as @xmath154 ev.@xcite it is related to @xmath89 for @xmath0 electrons through the simple relation,@xcite @xmath155 where @xmath156 stands for a racah parameter.@xcite taking into account that @xmath157 ev for nio@xmath10,@xcite one finds @xmath158 ev , i.e. , @xmath159 . indeed , it has been shown that @xmath160 ev reproduces the experimental band gap and the magnetic moment of la@xmath10nio@xmath5.@xcite finally , we set @xmath161 ev as a realistic value of the crystal field splitting in the nickelates . on the one hand , band structure calculations in the local density approximation predict the crystal field splitting between @xmath0 orbitals to be 0.5 ev . @xcite on the other hand , xas spectra reveal a somewhat larger splitting of 0.7 ev,@xcite a value also deduced from the optical spectroscopy.@xcite ( [ eq : df ] ) of the vbc ( top ) and dbc ( bottom ) stripe phases with respect to the af phase ( [ eq : df ] ) as functions of doping @xmath15 , obtained at temperature @xmath162 for the @xmath0 model . cluster sizes , distances between dws @xmath163 , and parameters are the same as in table [ tab : egf].,width=302 ] the performed calculations demonstrate a robust tendency towards stripe formation expressed by negative free energy gains of the vertical bond - centered ( vbc ) and diagonal bond - centered ( dbc ) stripe phases with respect to the af phase , @xmath164 for the @xmath0 model ( [ eq : h_deg ] ) . this energy gain @xmath165 increases with doping @xmath15 , as shown in fig . [ fig : f_fafe_gez0 ] . the @xmath166 curves cross each other for decreasing distances between dws @xmath163 which demonstrates the tendency to the gradual formation of stripe phases with smaller unit cells upon increasing doping . thereby , diagonal structures are significantly lower in energy than vertical ones for a given fixed doping , but especially in the large doping regime @xmath167 . the robust stability of the dbc stripe phases with respect to the vbc ones is illustrated more transparently in table [ tab : egf ] , where we compare the ground state free energy @xmath168 for both structures . note that a similar variation of the optimal distance @xmath163 between dws suggests the same optimal stripe filling . as in fig . [ fig : f_fafe_gez0 ] , but with finite crystal field splitting @xmath119 . cluster sizes , distances between dws @xmath163 , and parameters are the same as in table [ tab : egf ] . , width=302 ] the main effect of the crystal field splitting is to reduce the stability of both stripe phases with respect to the uniform af phase . this appears to be rather puzzling at first sight . one finds , however , that realistic positive @xmath96 promotes the @xmath93 orbitals with a narrow band . on the one hand , the electron distribution induced by finite @xmath96 suppresses substantially the af superexchange energy gain which becomes @xmath169 rather than @xmath170 . on the other hand , one would expect that such charge redistribution strongly increases the kinetic energy gain due to a wider band accessible for holes , propagating especially easily along dws where the af order is partly suppressed . nevertheless , this energy gain is easily overcompensated by the kinetic energy loss due to hopping perpendicular to the stripes . in addition , our studies of the bc stripe phases within the single - band hubbard model have shown that the largest kinetic energy gain is released on the bonds connecting pairs of ferromagnetically coupled atoms located in the dws.@xcite note that the fm order of the dw spins is substantially stabilized by the off - diagonal @xmath171 hopping in the @xmath0 model , yielding low - energy charge excitations that lead to the fm superexchange , @xmath172 . altogether , when one orbital is sufficiently favored by finite crystal field over the other one , these low - energy processes are effectively blocked , explaining the enhanced stability of the af order with respect to the bc solutions ( _ cf_. figs . [ fig : f_fafe_gez0 ] and [ fig : f_fafe_gez1 ] ) . a further qualitative point concerns the influence of a finite crystal field splitting @xmath96 between the @xmath86 and @xmath93 orbitals on the stability of dw structures . as depicted in fig . [ fig : f_fafe_gez1 ] , a more realistic value @xmath119 seems not to promote noticeably any stripe phase over another one and one still recovers dbc stripe phases as the ground state . we therefore conclude that it is not the crystal field @xmath96 that is responsible for a different orientation of dws in the nickelates from that observed in the cuprates . later on we shall see that finite @xmath96 has also only a little visible effect on optimal stripe filling . of the vbc ( top ) and dbc ( bottom ) stripe phases with unit cells of length @xmath163 , as a function of doping @xmath15 , obtained for the ddh model . parameters : @xmath173 , @xmath174 , @xmath100 , and @xmath162 . , width=302 ] let us now shortly return to the experimental characterization of the stripe phases in lsno . their periodicity at a given doping @xmath15 is described by both charge @xmath175 and spin @xmath18 wave vectors along the direction diagonal with respect to the ni@xmath82o bond . @xmath14 corresponds to the inverse of the distance @xmath163 between dws,@xcite i.e. , @xmath176 further , @xmath14 first increases continuously with doping @xmath15 and follows the linear relation @xmath57 in the wide doping range @xmath177 , while next it gradually saturates to the value @xmath39.@xcite such a behavior of @xmath14 indicates a fixed hole density of one hole per ni ion in a dw , and is consistent with the hf predictions both in the single and in multiband models.@xcite finally , a fixed hole density along the dws results in the pinning of the chemical potential @xmath60 for @xmath177 , whereas a large ( @xmath178 ev / hole ) downward shift appears in the higher doping regime.@xcite .1 cm cccccccc & & & & & & & & & & 0.050 & 11 & 88@xmath14188 & 3.1095 & & 11 & 88@xmath14188 & 3.1000 + 0.055 & 10 & 80@xmath14180 & 3.0912 & & 11 & 88@xmath14188 & 3.0804 + 0.060 & 9 & 72@xmath14172 & 3.0729 & & 11 & 88@xmath14188 & 3.0610 + 0.065 & 8 & 64@xmath14164 & 3.0547 & & 10 & 80@xmath14180 & 3.0417 + 0.070 & 8 & 64@xmath14164 & 3.0364 & & 9 & 72@xmath14172 & 3.0225 + 0.080 & 8 & 64@xmath14164 & 3.0001 & & 8 & 64@xmath14164 & 2.9839 + 0.090 & 7 & 84@xmath14184 & 2.9636 & & 7 & 84@xmath14184 & 2.9454 + 0.100 & 6 & 72@xmath14172 & 2.9269 & & 7 & 84@xmath14184 & 2.9070 + 0.110 & 5 & 80@xmath14180 & 2.8902 & & 6 & 72@xmath14172 & 2.8683 + 0.120 & 4 & 80@xmath14180 & 2.8538 & & 5 & 80@xmath14180 & 2.8301 + 0.140 & 4 & 80@xmath14180 & 2.7807 & & 5 & 80@xmath14180 & 2.7530 + 0.160 & 3 & 72@xmath14172 & 2.7077 & & 4 & 80@xmath14180 & 2.6760 + 0.180 & 3 & 72@xmath14172 & 2.6347 & & 4 & 80@xmath14180 & 2.5996 + 0.200 & 3 & 72@xmath14172 & 2.5626 & & 3 & 72@xmath14172 & 2.5227 + 0.300 & 3 & 72@xmath14172 & 2.2129 & & 3 & 72@xmath14172 & 2.1531 + 0.400 & 3 & 72@xmath14172 & 1.8809 & & 3 & 72@xmath14172 & 1.8116 + [ tab : ddhf ] it is interesting to establish whether the above results concerning the stripe phases in the ground state and their variation under increasing doping appear solely in the realistic @xmath0 model , or they are generic and remain also a common feature of the ddh model with only diagonal hopping elements given by eq . ( [ eq : h_kinddh ] ) . to allow a meaningful comparison , we set the same values of microscopic parameters as we those chosen for the @xmath0 model , i.e. , @xmath173 , @xmath174 , and we consider only the @xmath100 case imposed by symmetry between the equivalent orbitals . it restricts the solutions of the ddh model to @xmath179 ( see below ) . one finds that the free energy gains @xmath166 of the vbc and dbc stripe phases with respect to the af phase for the ddh model ( fig . [ fig : f_faft2_gez0 ] ) are qualitatively similar to those of the @xmath0 band model . first of all , for a fixed doping @xmath15 diagonal stripe structures are again significantly lower in energy than vertical ones ( _ cf_. also table [ tab : ddhf ] ) . secondly , also in this case , we recover a gradual crossover towards stripe phases with smaller unit cells upon increasing doping . note , however , that in contrast to the predictions made within the @xmath0 model , structures with vertical ( diagonal ) dws separated by a distance @xmath180 are the lowest energy solutions only in a narrow doping regime @xmath181 ( @xmath182 ) , respectively . for a complete characterization of stripe phases we define @xmath183 where @xmath184 is the number of doped holes and @xmath185 is the number of domain walls . our findings concerning the properties of bc stripe phases are summarized in fig . [ fig : enme_g ] showing the doping dependence of the incommensurability @xmath14 , the stripe filling @xmath186 , and the chemical potential @xmath60 for both the vbc ( left ) and dbc ( right ) stripe phases . these quantities for the respective ground states are deduced from figs . [ fig : f_fafe_gez0 ] , [ fig : f_fafe_gez1 ] and [ fig : f_faft2_gez0 ] . in each case the data points correspond to the middle of the stability region of the stripe phase of a given type . the only exception are the @xmath187 phases for them @xmath14 , @xmath186 , and @xmath60 , are plotted for the actual minimum of the free energy @xmath168 . ; ( c ) and ( d ) stripe filling @xmath186 ; ( e ) and ( f ) chemical potential @xmath60 ( in the units of @xmath80 ) , as deduced from the data of figs . [ fig : f_fafe_gez0 ] , [ fig : f_fafe_gez1 ] and [ fig : f_faft2_gez0 ] . filled ( open ) symbols for @xmath0 ( ddh ) model ; circles and squares for @xmath100 and @xmath119 . solid lines in panels ( a ) and ( b ) show the experimental behavior of @xmath14 in lsno given in ref . . , width=302 ] in agreement with the experimental data for lsno , indicated here by a solid line in panels ( a ) and ( b ) of fig . [ fig : enme_g ] , one observes that @xmath14 follows the law @xmath19 up to @xmath188 and then it tends to saturate to the highest possible value for the bc stripe phase , i.e. , @xmath56 , as there are no bc stripe phases in which dws are separated by @xmath189 lattice spacings . further , in the regime where @xmath14 follows linearly the doping level @xmath15 , decreasing stripe periodicity allows the system to maintain nearly fixed filling @xmath186 [ _ cf_. figs . [ fig : enme_g](c ) and [ fig : enme_g](d ) ] , pinning simultaneously the chemical potential @xmath60 , as shown in figs . [ fig : enme_g](e ) and [ fig : enme_g](f ) . remarkably , the ground state of both the vbc and dbc stripe phases is characterized by the optimal filling 0.9 hole / ni , very close indeed to the experimental value one hole / ni ion , and the optimal filling remains almost unaltered in the model with a finite crystal field @xmath119 . regarding the chemical potential shift @xmath58 in the doping regime @xmath190 , one finds that it exceeds the experimental value @xmath191 ev / hole nearly by a factor of 2 . indeed , assuming the effective hopping @xmath192 ev , one obtains @xmath193 ( @xmath194 ) ev / hole for @xmath100 ( @xmath119 ) , respectively . therefore , we conclude that the present effective model can only explain qualitative trends and one needs to carry out calculations within more realistic multiband models with oxygen orbitals included explicitly in order to obtain quantitatively the experimental data . let us verify now whether the established results concerning the doping dependence of @xmath14 and @xmath186 are also obtained within the ddh model . as the structures with vertical ( diagonal ) dws separated by a distance @xmath180 appear in the ddh model only in a narrow doping regime @xmath181 ( @xmath195 ) , respectively , one observes here a fast variation of the optimal distance @xmath163 which should result in a small optimal stripe filling . indeed , as depicted in figs . [ fig : enme_g](a ) and [ fig : enme_g](b ) , the low - doping part of @xmath14 , being linear in @xmath15 , has a larger slope which exceeds the experimental value in lsno roughly by a factor of 2 ( 1.5 ) in the case of the vbc ( dbc ) stripe phase , respectively . consequently , the optimal stripe filling in the former case is substantially reduced down to @xmath196 and in the latter case down to @xmath197 [ see figs . [ fig : enme_g](c ) and [ fig : enme_g](d ) ] . finally , one finds that the chemical potential starts to decrease rapidly at @xmath198 in the vbc stripe phase [ fig . [ fig : enme_g](e ) ] and at slightly larger doping @xmath199 in the dbc phase [ fig . [ fig : enme_g](f ) ] . model ( filled circles ) , or in the ddh model ( open circles ) on a @xmath200 cluster at doping @xmath201 : local hole densities @xmath202 ( top row ) ; modulated magnetization densities @xmath203 ( second row ) ; local hole @xmath204 ( third row ) ; and local modulated magnetic @xmath205 orbital polarization ( bottom row ) . for more clarity , the data points for the ddh model are shifted by one lattice constant from the origin of the coordinate system . parameters : @xmath173 , @xmath174 , @xmath100 , and @xmath162 . , width=302 ] in order to find out the reason of such a vast discrepancy between the predictions made in the @xmath0 and in the ddh model , let us now investigate closer the properties of idealized filled and half - filled dws : half - filled vertical bond - centered ( hvbc ) and half - filled diagonal bond - centered ( hdbc ) . a complete characterization of the charge and magnetization distributions for a system with orbital degeneracy is given by : the local hole density @xmath206 , local modulated magnetization density @xmath203 , local hole orbital polarization @xmath204 , and local modulated magnetic orbital polarization @xmath205 , defined using the operators given in eqs . ( [ eq : opsdef ] ) as follows : @xmath207 here @xmath208 is the local orbital charge density for spin @xmath209 , whereas @xmath210 denotes the local orbital hole density , @xmath211 ( filled circles ) model , or in the ddh model ( open circles ) on a @xmath200 cluster at doping @xmath201 . the meaning of the different panels and shift of the ddh data as in fig . [ fig : fbc ] . parameters : @xmath173 , @xmath174 @xmath100 , and @xmath162 . , width=302 ] to better appreciate the differences between the ddh and the @xmath0 model , we compare in fig . [ fig : fbc ] , the local hole @xmath202 and the modulated magnetization densities @xmath203 of the filled vbc and dbc stripe phases found at temperature @xmath162 in either model on a @xmath200 cluster at doping @xmath201 for the standard parameters , _ i.e. _ , @xmath173 , @xmath174 , and @xmath100 . for completeness we also show the local hole @xmath204 and the local modulated magnetic @xmath205 orbital polarization . these quantities are finite only in the @xmath0 model , while @xmath212 by symmetry in the ddh model . since @xmath213 is positive the holes are located in @xmath86 orbitals , and the polarization of the two sites on the dws is fm ( alternation of signs in @xmath214 at the dws ) . the observed differences , especially pronounced at dw atoms , directly follow from the fact that , a different effect helping to reduce double occupancy at those sites is effective in each model . namely , in the @xmath0 hopping model ( [ eq : h_kin ] ) one finds large positive @xmath204 at dws , which means that it is energetically advantageous to optimize the kinetic energy of holes by putting them into the @xmath86 orbitals where the hopping element @xmath215 is larger and gives a wide band . on the one hand , in the hf the only way to optimize the on - site energy is to develop a strong spin polarization which , in turn , would noticeably reduce the kinetic energy gain . on the other hand , such disadvantageous suppression can be avoided by a strong reduction of the electron density . this explains why the hole density @xmath202 along dws in the @xmath0 model is larger , as compared to the corresponding value found in the ddh case . indeed , in the latter case both bands are equivalent , resulting in @xmath212 . hence this model yields a more localized stripe phase with a larger magnetization @xmath216 at dws than the one obtained in the @xmath0 model ( _ cf_. fig . [ fig : fbc ] ) . for completeness , in fig . [ fig : hfbc ] we compare the hole density , @xmath202 , and the spin density , @xmath203 , profiles of the half - filled stripe phases . the unit cells are smaller by a factor of two and the amplitude of the charge density wave is correspondingly reduced . even though the overall shape of the dws looks very much the same in both models , a larger magnetization @xmath216 at dws is found again within the ddh hamiltonian ( [ eq : h_kinddh ] ) . : intraorbital double occupancy @xmath217 ( top row ) , and two local interorbital @xmath218 ( middle ) and @xmath219 ( bottom ) double occupancies . filled ( open ) circles denote the results found in the @xmath0 ( ddh ) model , respectively . parameters and filled / empty circles as in fig . [ fig : fbc ] . , width=302 ] furthermore , important information about the nature of stripe phases is provided by the averaged intraorbital double occupancy at site @xmath220 , @xmath221 as well as by two local interorbital double occupancies , @xmath222 where @xmath223 . in a system with isotropic charge distribution the double occupancies are site independent , while in stripe phases they vary with a characteristic periodicity following from the size of the stripe unit cell . from the structure of the coulomb interaction eq . ( [ eq : h_int ] ) one expects @xmath224 for a fixed hole density . in a stripe phase these double occupancies are locally suppressed by the magnetic and orbital polarizations in a way that depends both on the shape of the stripe , and on the form of the one - electron hamiltonian . let us first discuss the average double occupancies ( [ eq : d0])-([eq : dt ] ) of the stripe phases depicted in fig . [ fig : fbc ] . remarkably , in the ddh model ( open circles in fig . [ fig : fbcdo ] ) the on - site energy is predominantly optimized by the reduction of the high - energy configurations with opposite spins which lead to intraorbital @xmath217 and interorbital @xmath218 double occupancies , so that the system might create a large number of dws in the unit cell even in the low doping regime , consequently reducing the optimal stripe filling . such a strong reduction of the double occupancies for opposite spins follows from the diagonal hopping which gives a weaker electronic mixing between the sites of opposite spin due to a lower kinetic energy . @xcite : hvbc ( left ) and hdbc ( right ) . different double occupancies as in fig . [ fig : fbcdo ] . parameters and filled / empty circles as in fig . [ fig : hfbc ] . , width=302 ] in contrast , in the @xmath0 model ( filled circles in fig . [ fig : fbcdo ] ) , the coulomb energy is mainly optimized by the reduction of the low - energy interorbital @xmath219 double occupancy resulting in a smaller magnetization @xmath216 at dws as compared to the one found in the ddh model . however , reduced @xmath216 at those sites allows the system to better optimize the kinetic energy gain which then overcompensates a large on - site energy only when the optimal filling is close to one hole per ni site , meaning that for a given doping level , the dws should be separated by a larger distance as compared to predictions made in the ddh model . on the other hand , the robust stability of the dbc stripe phases with respect the vbc ones can be understood as following from a stronger reduction of all double occupancies ( [ eq : d0])-([eq : dt ] ) in the former phase . in fact , the doubly occupied configurations can not be then excited between the nearest neighbor dw sites , leading to a lower total energy . in case of half - filled stripes ( fig . [ fig : hfbcdo ] ) the double occupancies for opposite spins ( @xmath217 and @xmath218 ) are only slightly lower than for filled stripes , but the interorbital double occupancy for equal spins ( @xmath219 ) is significantly higher , reflecting the higher electron filling at dw atoms in these phases . due to the finite hund s exchange considered here , @xmath225 , the high - spin states are promoted . altogether one finds similar generic features seen above for the filled stripes ( fig . [ fig : fbcdo ] ) : ( i ) both double occupancies for the opposite spins are stronger reduced in the ddh model , and ( ii ) the double occupancy by the same spin @xmath226 at the dws is stronger suppressed in the @xmath0 model , resulting from a smaller magnetization at these sites than in the ddh model . this second effects seems to play a role in stabilizing the hdbc stripe phase in the ddh model rather than the dbc phase , found for the realistic @xmath0 model . model at @xmath201 doping : ( a ) partial density of states @xmath227 for @xmath86 orbital , ( b ) partial density of states @xmath228 for @xmath93 orbital , ( c ) total density of states @xmath229 . panel ( d ) shows for comparison the total density of states @xmath229 as found in the dbc stripe phase within in the ddh model . parameters as in fig . [ fig : fbc ] . , width=302 ] important differences between the two models strongly influence the density of states ( dos ) . here we first determine the dos projected on the @xmath230 orbital as : @xmath231 where @xmath232 are the eigenvectors of the mf hamiltonian eq . ( [ eq : h_mf ] ) . it is calculated using a histogram of the corresponding eigenvalues . it is apparent from figs . [ fig : dosf ] and [ fig : doshf ] that the spectra consist of several subbands . however , two hubbard subbands are well visible both for filled and half - filled diagonal bc stripes : lower hubbard band ( lhb ) and upper hubbard band ( uhb ) . the hubbard subbands arise primarily due to the af polarization and the gap between them is proportional to the magnetization . panels ( a ) and ( b ) in figs . [ fig : dosf ] and [ fig : doshf ] show the dos @xmath233 ( [ eq : dosal ] ) , projected on the orbital @xmath86 and @xmath93 , respectively . here , one finds that the mid - gap bands have mainly @xmath86 character close to the gap between them the @xmath86 orbitals optimize the kinetic energy of holes doped into them , whereas the vast majority of the more localized @xmath93 states of dw atoms belongs to the energy regimes within the lhb and close to the uhb . the total dos for the dbc stripes in the @xmath0 model , depicted in fig . [ fig : dosf](c ) , reveals mid - gap states which demonstrate the one - dimensional character of the transport in the stripe phases . note that due to the af spin modulation along the bc dws themselves , one finds as well a distinct gap between two mid - gap bands lying within the mott - hubbard gap . this gap for the filled dbc stipes is lower in the @xmath0 model [ fig . [ fig : dosf](c ) ] than in the ddh model [ fig . [ fig : dosf](d ) ] , as the holes are more delocalized in the former case . this holds as well for the hdbc stripes , see figs . [ fig : doshf](c ) and [ fig : doshf](d ) . we have shown before that the magnetization on the dw sites is larger for the hdbc stripes than for the dbc ones , and therefore the gaps between the mid - gap states are also larger in fig . [ fig : doshf ] than in fig . [ fig : dosf ] . model at @xmath201 doping , panels ( a)-(c ) as in fig . [ fig : dosf ] . panel ( d ) shows for comparison the total density of states @xmath229 as found in the hdbc stripe phase within in the ddh model . parameters as in fig . [ fig : hfbc ] . , width=302 ] remarkably , both for the filled and half - filled diagonal bc stripe phases of figs . [ fig : dosf ] and [ fig : doshf ] the lower mid - gap states are to some extent localized above the lhb in the @xmath0 model , whereas they are clearly almost merged with the lhb in the ddh model . this behavior follows from a smaller on - wall magnetization @xmath203 in the @xmath0 model . therefore , in the ddh case only a few holes occupy the low - lying mid - gap bands , explaining the reason of substantial suppression of the optimal stripe filling in this case . instead , filled stripes are favored in the @xmath0 model they have a larger fraction of holes in the lower mid - gap states . note , however , that also in this case the stripes are metallic , with some spectral weight ( mainly in @xmath86 orbitals ) above @xmath234 . we summarize our key results concerning the stability of stripe phases in the nickelates . we investigated this problem using two hubbard models the realistic model for @xmath0 electrons , and the ddh model which frequently serves as the simplest approach to the orbital degeneracy . @xcite it is quite encouraging that the ddh model gives already diagonal stripe phases as the most stable structures , in contrast to the nondegenerate hubbard model where the vertical ( horizontal ) stripes are found in the hf approximation . however , one finds half - filled bc stripes in the ddh model , and it is only in the realistic @xmath0 model for doped nickelates that one observes a generic tendency to promoting filled stripes over the half - filled ones . as an example of this behavior , we give the free energies of filled and half - filled stripe phases found within both ( @xmath0 and ddh ) models for @xmath201 doping in table [ tab : fez0 ] . first of all , one finds that the energies of bc phases are considerably lower for the realistic model of @xmath0 band due to the higher kinetic energy gains which result from the off - diagonal hopping.@xcite most importantly , for the parameters relevant to lsno used for the data of table [ tab : fez0 ] , one finds that the dbc phase has the lowest energy , _ i.e. _ , indeed experimentally observed _ filled _ diagonal stripes . in contrast , the sc stripe phases with unpolarized dw sites and large double occupancies have much higher energies , and the best of them is not diagonal , but vertical ( horizontal ) one . we also note that even though the ddh model with two equivalent orbitals clearly favors diagonal dws , it stabilizes instead of filled the _ half - filled _ diagonal phases . cccccc & & & model & phase & filled & half - filled & filled & half - filled @xmath0 & bc & * 2.5508 * & 2.5759 & 2.5629 & 2.5756 & sc & 2.7815 & 3.2615 & * 2.7012 * & 3.0729 ddh & bc & 2.8172 & * 2.8135 * & 2.8395 & 2.8354 as in the cuprates , the coexisting charge and magnetic order in diagonal stripe phases in the nickelates is a result of the compromise between the kinetic and magnetic energies the magnetic energy is gained in the af domains , and the kinetic energy is gained mainly along the dws . the bc stripes are favored as then the magnetic energy can be gained ( table [ tab : fez0 ] ) not only in the af domains , but also on the dw magnetic sites . finally , when the dws are filled , more kinetic energy is gained in the @xmath0 model while then the off - diagonal hopping is allowed . altogether , this mechanism shows little sensitivity to the small crystal field splitting . for moderate values of the latter it tends to eliminate stripe phases entirely rather than promoting one stripe phase over another one . in spite of the remarkable success of the present study which gave stable diagonal stripe phases , one has to admit that the predicted electronic properties show some difference to the experimental ones . in fact , our systematic studies of stripe phases completed within the relevant model for @xmath0 orbitals , where several phases separated by different lattice spacing varying from @xmath187 to @xmath235 were considered , have revealed that the optimal stripe filling in the true ground state is slightly less ( @xmath236 hole / ni depending on the crystal field splitting ) than the experimental value of one hole per ni ion . this concerns the entire low doping regime , @xmath237 , where @xmath19 . there may be a few reasons of this discrepancy . first of all , one has to realize that in a multiband model with oxygen orbitals included explicitly the holes would be doped primarily to oxygen orbitals , screening the local moments at ni sites,@xcite in analogy to zhang - rice singlets in the cuprates.@xcite an insulating ground state could then result from oxygen distortions by the peierls mechanism . thus , we argue that the present results could be further improved within a realistic model including not only two @xmath0 orbitals with different hopping elements , but also orbital polarization at oxygen sites . second , it may be expected that the stripe filling will be somewhat changed due to the electron correlation effects beyond the hf approximation used here . we believe that the present work provides a good starting point for future studies of the correlation effects . in summary we have analyzed the stripe formation in the doped nickelates in the realistic model with degeneracy of @xmath0 orbitals , using large clusters . the results obtained with this model were compared with the widely used doubly degenerate hubbard model . even though in both models the distance between the dw is inversely proportional to the doping , the most stable diagonal stripes are found to differ markedly from one model to the other one . in the @xmath0 model , the stripes are filled and nearly insulating , as observed experimentally in a series of layered nickelates . in contrast , in the ddh model they are half - filled and metallic . these latter stripe phases are reminiscent of the stripes observed experimentally in largely doped cuprates . these differences have their roots in the different structure of intersite hopping terms . as the ddh model is closer to @xmath153 than to @xmath0 hopping matrix elements , one might expect that doped insulators with @xmath153 orbital degrees of freedom , as in vanadates or in ruthenates , would promote different stripe phases than those observed in the nickelates . we thank k. rociszewski for insightful discussions . m.r . acknowledges support from the european community under marie curie program number hpmt2000 - 141 . this work was supported by the polish ministry of science and education under project no . 1 p03b 068 26 , and by the ministre franais des affaires etrangres under polonium 09294vh .

the three - body energy - dependent effective interaction given by the bloch - horowitz ( bh ) equation is evaluated for various shell - model oscillator spaces . the results are applied to the test case of the three - body problem (h andhe ) , where it is shown that the interaction reproduces the exact binding energy , regardless of the parameterization ( number of oscillator quanta or value of the oscillator parameter ) of the low - energy included space . we demonstrate a non - perturbative technique for summing the excluded - space three - body ladder diagrams , but also show that accurate results can be obtained perturbatively by iterating the two - body ladders . we examine the evolution of the effective two - body and induced three - body terms as and the size of the included space are varied , including the case of a single included shell , . for typical ranges of , the induced effective three - body interaction , essential for giving the exact three - body binding , is found to contribute% to the binding energy .

nucl-th0404028

techniques in popular use in the nuclear three - body problem include the faddeev@xcite , green s function monte carlo ( gfmc ) , and correlated hyper - spherical harmonics expansion methods ( see , for instance , refs.@xcite ) . these methods have been used with realistic phenomenological potentials ( _ e.g. _ av18@xcite , , cd - bonn@xcite , etc . . . ) with strong short - range interactions , yielding binding energies accurate to within @xmath5%@xcite . all of these methods treat the bare @xmath6 interactions in hilbert spaces that are effectively infinite . because the complexity of the hilbert spaces grows very rapidly with nucleon number a , the extension of such methods to heavier systems becomes increasingly difficult . this motivates another approach : solving a - body problems in more tractable finite `` included spaces , '' while accounting for the missing physics of the `` excluded space '' through an effective interaction . for shell - model - inspired effective theories that define included spaces in terms of harmonic oscillator ( ho ) bases , the missing physics includes both high - momentum and long - wavelength interactions . the former are missing because the included space contains only low - energy oscillator shells , while the latter are connected with the finite extent of the basis states , which are overconfined in an ho potential . shell - model calculations often employ two - body effective interactions , but these are generally determined phenomenologically . in recent years there has been growing success in calculating these effective interactions directly from the underlying bare interaction in a precise and systematic way ( _ e.g. _ refs.@xcite and references within ) . the purpose of this paper is to extend upon this work by calculating the more complicated @xmath7 that could be used in a shell - model - inspired effective theory , through three - body order in the excluded space .. the inclusion of three - body contributions is generally believed essential in building in the density dependence crucial to saturation . the exact effective interaction for the a - body problem is an a - body operator . the calculation of this interaction is clearly as difficult as solving the original a - body problem in an infinite hilbert space . however , there are reasons to hope that such an a - body calculation might be unnecessary . this hope depends on the plausible notion that the effective interaction might converge as an expansion in the number of nucleons interacting in the excluded space . were such an expansion to converge at the three- or four - body cluster level , then standard three- or four - body methods could be used to treat the excluded space . these more complicated but still tractable effective interactions could then be diagonalized in the shell - model - inspired effective theories to produce accurate binding energies and wave functions . there are a couple of arguments that such a scheme might converge . one is the success of the shell model when phenomenological two - body effective interactions are used . this suggests that a good part of the excluded - space physics is two - body . another is the qualitative argument that short - range clustering is increasingly unlikely as the size of the nucleon cluster increases : such spacial clustering is necessary for strong , multi - particle , high - momentum interactions . finally , we know that the pauli principle forbids short - range s - wave interactions above a=4 . additional support for this notion comes from recent large - basis _ ab initio _ no - core shell - model ( ncsm ) calculations@xcite . these calculations use a large but finite ho hilbert space . a lee - suzuki ( ls ) similarity transformation@xcite coupled with a folded - diagram sum@xcite is performed on the bare nn interaction to yield an energy - independent effective interaction . early calculations utilized only the effective two - body interaction as derived from these transformations . hence calculations in large hilbert spaces , or included spaces , were usually needed to ensure convergence which , if judiciously chosen , could improve convergence greatly . ] of the binding energy ( _ i.e. _ the included space had to be large enough such that the contributions to the binding energy from the effective three- and higher - body terms were small ) . recent calculations that now include induced three - body interactions derived from the ls procedure show significant improvement in the convergence of binding energies for the alpha particle and some @xmath8-shell nuclei@xcite . an alternative approach to calculating the induced three - body effective interaction was developed in ref.@xcite using an excitation - energy-_dependent _ three - body interaction based on the self - consistent solution of the bloch - horowitz ( bh ) equation@xcite . this method was successfuly used to calculate the binding energy of @xmath0he . however , both of these works to date ( _ i.e. _ ls and bh procedures ) involve handling the excluded - space physics in a very large but still finite hilbert space . this approximation introduces an extra dependence to the three - body binding on the size of the excluded space . such dependence can only be removed by a true summation over all excluded high - energy modes ( i.e. excluded space of infinte size ) . though this dependence is small due to the fact that results of these previous works are quite accurate ( both groups examined the convergence of their results as a function of the size of the excluded space ) , the rate of convergence clearly depends in detail on how singular the underlying nn interaction is . here we develop a method for calculating the effective three - body interaction that introduces no such high - momentum cutoff , but instead truly integrates over all high - energy modes . the accuracy of our results is only limited by our numerical precision . thus it can be applied with equal confidence to any underlying nn reaction , regardless of that potential s description of short - range physics . this interaction is used in the simplest test cases , @xmath0he/@xmath0h , and the results are compared with those that would result from evaluating the effective interaction only at the two - body level . our treatment is based on solving the bh equation as well , which produces a state - dependent ( and thus excitation - energy - dependent ) hermitian interaction , @xmath9 where @xmath10 here the intrinsic hamiltonian @xmath11 is obtained by summing the relative kinetic energy and potential energy operators @xmath12 and @xmath13 over all nucleon pairs @xmath14 and @xmath15 . the relative kinetic energy is found by subtracting the center - of - mass ( cm ) kinetic energy from the total kinetic energy obtained by summing @xmath16 over all nucleons . @xmath17 is the total mass of the a - body system , @xmath18 is the excluded space projection operator , and @xmath19 is the included - space projection operator that defines the finite , low - energy space in which a direct diagonalization will be done , once @xmath7 is obtained . as @xmath20 is the desired eigenvalue , which is not known @xmath21 @xmath22 , the equations must be solved self - consistently . in the calculations reported here , the av18 potential is used as the bare @xmath6 interaction and no bare three - body interaction is included . because our application is to @xmath0he/@xmath0h , rather than to a heavier nucleus , we will have solved the effective interaction at the a - body level . thus we are not testing the assumption of a cluster expansion here . instead , our focus is on the technique for solving the effective interaction , and its relation to the approximate two - body result . using techniques familiar from faddeev calculations , we first determine an effective two - body - like interaction , @xmath23 , from the scattering @xmath24-matrix . we find that the induced effective three - body interaction can be evaluated perturbatively by iterating on @xmath23 . for a certain range of oscillator parameters @xmath1 , convergence is achieved even for very small included spaces , including the limiting @xmath25 space . we also develop a non - perturbative method . by rearranging the terms of the original expansion for the three - body interaction , the series can be summed exactly through numerical solution of an integral equation . this method is very accurate and efficient for any included space and any oscillator parameter . in section [ preliminaries ] we generalize the two - body formulation presented in ref.@xcite . section [ sect : perturbative ] describes our perturbative treatment of the effective interaction , while section [ sect : nonperturbative ] presents the nonperturbative solution via an integral equation . the associated discussions focus on qualitative issues , with derivations reserved for the appendices . we discuss the results and the need for a simple test of the cluster assumption ( by applying current results to @xmath26he ) in section [ sect : conclusion ] . to guarantee that @xmath7 , like the bare @xmath11 , is translationally invariant one can work in a complete basis of three - nucleon ho slater determinants that includes all configurations with quanta @xmath27 . such a basis , in traditional independent - particle shell - model coordinates , is overcomplete , consisting of subspaces characterized by the eigenvalues of the center - of - mass hamiltonian @xmath28 that do not interact via @xmath7 . for @xmath294 , it is convenient to avoid this overcompleteness by using a jacobi basis , which reduces the @xmath30-body intrinsic - hamiltonian problem to an ( @xmath30 - 1)-body problem . this is the choice we make here . the jacobi included - space ( @xmath31 ) basis corresponding to the ho shell - model basis described above is thus simple , consisting of all relative - coordinate configurations with @xmath27 . ( we note that for @xmath324 , the single - particle basis is the more efficient basis for many - body calculations . however , an @xmath7 developed in a jacobi basis can be easily transformed to single - particle coordinates for use in shell - model - inspired diagonalizations . ) such included spaces for the two - nucleon system are simple to construct . for example , a @xmath2=2 included space consists of spatial wave functions of the form @xmath33 , @xmath34 , and @xmath35 , coupled to spin and isospin wave functions to maintain antisymmetry . this requires @xmath36 to be odd . the corresponding construction of fully anti - symmetrized @xmath30-body wave functions is less trivial @xcite . here we use the codes described in ref.@xcite to generate these for @xmath30=3 . in table [ tab : dimension ] dimensions of various @xmath30=3 included spaces are given as a function of @xmath2 . for the three - body system one can obtain accurate results by diagonalizing the bare @xmath11 in a jacobi basis , provided @xmath2 is made sufficiently large . this was done in ref.@xcite for the av18 potential with @xmath2=60 , resulting in a binding energy accurate to @xmath4 20 kev . the motivation for solving this problem for small @xmath2 using effective interactions is to explore techniques that might be more feasible for larger @xmath30 , where the model - space growth analogous to table [ tab : dimension ] will be much steeper . one goal of the current work is to find techniques that might make the integration over the excluded space more tractable . in our earlier work on the two - body system@xcite we found that this integration is difficult at both long and short distances . we found it convenient in ref.@xcite to remove the long - distance difficulties , the pathologies associated with the overbinding of the ho , at the outset . as the overbinding of the ho becomes arbitrarily large as @xmath37 increases , part of the tail of the wave function remains unresolved in any finite - basis treatment though numerically the contribution of the tail becomes increasingly unimportant in direct diagonalizations as more shells are added . ( this contrasts with the short - range problem , as av18 and other modern potentials are regulated at short distance by some assumed functional form . such potentials can be fully resolved in finite bases , provided @xmath2 is larger than the scale implicit in that functional form . ) in ref.@xcite the long - range behavior was handled by the following rearrangement of eq . ( [ eqn : bh ] ) @xmath38\frac{e}{e - qt}\right\}p,\ ] ] where @xmath39 here @xmath40 and @xmath41 are shorthand for @xmath42 and @xmath43 , respectively . due to space restrictions eq . [ eqn : bhre ] was stated in ref.@xcite without derivation . hence we show its derivation for completeness in appendix [ app : bhfinal ] appearing in ref.@xcite differs from the one shown in eq . [ eqn : teff ] . however , both expressions , from an operator standpoint , are completely equivalent . ] . to calculate included - space matrix elements of eq . [ eqn : bhre ] , it was first convenient to define the states @xmath44 where @xmath45 is some state that resides in the included space ( _ i.e. _ @xmath46 ) . with this definition , matrix elements of eq . [ eqn : bhre ] became @xmath47 our expansion of @xmath7 was formed by expanding the resolvent of @xmath48 ( see eq . [ eqn : veff ] ) in powers of @xmath49 : @xmath50 an iterative procedure was then used to calculate included - space matrix elements of this expansion . at each order we solved for the self - consistent energy @xmath20 . the rearrangement of eq . ( [ eqn : bhre ] ) was applied to the deuteron in ref.@xcite and led to excellent results in perturbation theory for suitably chosen oscillator parameters @xmath1 ( only few orders of @xmath49 were needed to reproduce the deuteron binding energy to within one kev ) . figure [ fig : endshell ] shows that @xmath51 acting on the ho state @xmath45 drastically changes that state s large-@xmath37 behavior : the asymptotic behavior of @xmath52 is @xmath53 , where @xmath54 , @xmath55 being the mass of the particle@xcite . as this is the proper fall - off , the overconfining effects of the ho are thus repaired . the effects are important for the deuteron , which has an extended wave function because of its small binding energy . the operator @xmath51 modifies only those ho states that reside in the last shell of the included space . for all other states within the included space , the operator is identical to unity . the `` endshell - corrected '' states thus control all of the proper asymptotic behavior . the resummation of the kinetic energy operator allows one to adjust basis wave functions to absorb much of the short - range behavior of the @xmath6 interaction into the included space , leaving a weak residual interaction that can be handled perturbatively . by adjusting the oscillator parameter @xmath1 to low values ( @xmath56 fm ) , the residual q - space contributions to the binding energy due to @xmath57 can be evaluated to within @xmath4 one kev in third - order perturbation theory , even for very small included spaces . this is only possible because the kinetic energy has been summed to all orders independent of @xmath1 , thus guaranteeing that the choice of a small @xmath1 will not alter the wave function at large @xmath37 . the optimal @xmath1 presumably provides the best resolution of the hard core without substantially altering the ability of the basis to reconstruct the intermediate - range potential . in the next section we will explore a similar expansion for @xmath0h/@xmath0he . we find modifications are necessary to account for the disparate length scales that come about from the inclusion of a third particle . before starting this discussion , we end this section by presenting compact expressions for the operators @xmath58 and @xmath59 . such operators appear frequently in subsequent formulae . the operator @xmath60 closely resembles the free particle propagator . however , due to the projection operator q in the denominator , calculating matrix elements of @xmath58 requires a little ingenuity . here we show that the excluded - space green s function can be expanded in terms of the much simpler full green s function and operators within the included space that can be inverted easily . following ref.@xcite , we first consider @xmath61 where @xmath62 . collecting terms in this `` schwinger - dyson '' form and noting that @xmath63 gives @xmath64 the last term in eq . [ eqn : step2 ] , @xmath65 , can be rewritten @xmath66 plugging the above equation into eq . [ eqn : step2 ] finally gives @xmath67 similarly , the `` schwinger - dyson '' form of @xmath58 is @xmath68.\ ] ] using eq . ( [ eqn : step3 ] ) then yields @xmath69 this expression is relatively simple to use . the free particle propagator , @xmath70 , is known analytically . indeed , its form is diagonal in momentum space . the operator @xmath71 represents a matrix composed of included - space overlaps of the free particle propagator . since we work within small included spaces , inverting this matrix is not difficult . the included - space matrix elements are easy to calculate , as analytic expressions for @xmath72 for any @xmath30-body system exists . these expressions involve multiple sums over hypergeometric functions and gamma functions . in appendix [ sect:3bodyetme ] we give the relevant expression for the three - body system . to derive an analogous expression for @xmath59 , we first consider @xmath73 where we have substituted eq . [ eqn : finalstep ] in the second line above . with the use of eqs . [ eqn : finalstep ] and [ eqn : eminusqt ] , @xmath59 becomes @xmath74 the physical interpretation of eq . [ eqn : eminusqtq ] is simple : the term on the _ lhs _ represents free propagation in the excluded ( @xmath18 ) space , while the first term on the _ rhs _ ( bottom line ) represents free propagation in the full ( @xmath75 ) space and the second term subtracts off the contribution coming from included - space ( @xmath31 ) propagation . it is convenient to make the following definitions : @xmath76 with these definitions , eqs . [ eqn : finalstep ] and [ eqn : eminusqtq ] become @xmath77 finally , with the expressions above , we can express our perturbative expansion in a concise form , @xmath78 where @xmath79 in the expansion given by eq . [ eqn : bhfinal ] , @xmath80 corresponds to @xmath81 , while @xmath82 corresponds to @xmath83 , and so on . as we stressed in ref.@xcite , matrix elements of the operators shown in eq . [ eqn : gamma_n ] can be simply calculated using a recursive procedure . similarly , one might apply the expansion given by eqs . ( [ eqn : bhfinal ] ) and ( [ eqn : gamma_n ] ) directly to @xmath0he/@xmath0h . as was done in the two - body system , we construct endshell - corrected states to build in the correct aymptotic forms . figure [ fig:3bodyendshell ] shows momentum - space results for one @xmath84 endshell state . as expected , the modified wavefunction acquires a strong peak near zero - momentum , corresponding to large - r corrections . to determine an optimal @xmath1 for subsequent calculations , @xmath11 is minimized within the included space . the results are given in fig . [ fig : he3variational ] for several included spaces . as expected , the endshell states improve the binding of the three - body system and lower the optimal @xmath1 , relative to uncorrected ho results . however , the improved binding energies are still far away from the exact answers . this contrasts with the deuterium results ( see fig.2 of ref . @xcite ) , where 0th - order results using endshell states were accurate ( @xmath4 50 kev ) , and became nearly exact in second- or third - order perturbation theory . the 0th order a=3 results underbind by @xmath85 mev . as fig . [ fig:10hwhe3energy ] shows , these results are not corrected in perturbation theory . successive orders produce wildly oscillating values for the binding energy , in contrast to the rapid convergence found for the deuteron . the difficulty is that a single distance scale @xmath1 does not provide sufficient freedom to describe short - range behavior governed by two relative cordinates . extended states such as those of fig . [ fig : efimov ] are problematic . once one makes a choice of jacobi coordinates , the short - range behavior of some two - body cluster will be difficult to describe . for example , the choice @xmath86 and @xmath87 prevents one from adjusting @xmath1 to account for the short - range interactions of nucleons 1 and 2 . consequently , important hard - core interactions lie outside the included space , leading to nonperturbative behavior . because the deuteron was easily treated , it is possible to sum to all orders the repeated excluded space scatterings of any two - nucleon cluster . such a partial summation should account for the strong repulsive interaction between any two coupled nucleons , leaving only intermediate- and long - range interactions . such interactions can be treated reasonably accurately within the included space using our endshell - corrected states , leaving perturbative corrections . this partial summation corresponds to the faddeev decomposition@xcite . the starting point is again eq . [ eqn : heff_me ] , @xmath88 where we cast @xmath48 in its integral form , @xmath89 here the superscript @xmath0 denotes that this is a three - body effective interaction . now consider the state @xmath90 @xmath91 @xmath92 , which satisfies @xmath93 this can be decomposed into faddeev components @xmath94 , given by @xmath95 where @xmath96 equations [ eqn : psi_sum ] and [ eqn : psi_faddeev ] show that the faddeev components require the solution of three coupled integral equations . for example , @xmath97 satisfies @xmath98 equation [ eqn : psi12_relation ] can be inverted with respect to the @xmath97 faddeev component , giving @xmath99 where @xmath100 analogous equations can be found for the remaining faddeev components . the superscript @xmath101 indicates that the effective interaction given by eq . [ eqn : v12eff ] represents the sum of repeated potential scatterings between the same two nucleons ( in this case , nucleons 1 and 2 ) , while the third nucleon remains a spectator , as illustrated in fig . [ fig : v12eff ] . this expression is exactly the partial sum mentioned above . notice that the middle expression in eq . [ eqn : v12eff ] is very similar to eq . [ eqn : veff ] . furthermore , the integral form of @xmath102 is similar to that of the g - matrix@xcite . yet there are subtleties that differentiate the two . we will return to this point later . the integral equation given by eq . [ eqn : v12eff ] can also be solved exactly in terms of the ( 2 + 1)-body @xmath24-matrices@xcite . in appendix [ sect : exactv12eff ] we show @xmath103^{-1}g_0t_{12},\ ] ] where @xmath104 and @xmath105 is given by eq . [ eqn : invgamma0 ] . the physical interpretation of eq . [ eqn : exactv12eff ] is similar to the one given below eq . [ eqn : eminusqt ] : the first term on the _ rhs _ represents the effective interaction coming from repeated scatterings in the full ( @xmath75 ) space , while the second term subtracts off the contribution from the included ( @xmath31 ) space . the advantage in using the faddeev decompositions comes from exploiting particle exchange symmetries . as the states @xmath106 are fully anti - symmetric , the individual faddeev components can be related to each other by simple permutation operators@xcite . in particular , @xmath107 where the permutation operator @xmath108 is defined as @xmath109 . the last expression in eq . [ eqn : psi12_relation ] then becomes @xmath110 which can be formally inverted with respect to @xmath111 to give @xmath112 where @xmath113 equation [ eqn : v30eff ] defines the induced three - body interaction coming from repeated scatterings of @xmath102 ( see fig . [ fig : resolvent30 ] ) . the operator @xmath108 ensures that every insertion of @xmath102 represents scatterings coming from different pairs of nucleons ( i.e. the rungs of fig . [ fig : resolvent30 ] alternate ) , preventing any double couting . as there are no spectator nucleons in this expression , it is labeled with the superscript @xmath114 . matrix elements @xmath115 are given by @xmath116 where the factor of @xmath117 in the last expression comes from the anti - symmetry of the states . an expansion of eq . [ eqn : v3effme ] can now be found by expanding the resolvent of eq . [ eqn : v30eff ] in powers of @xmath118 , @xmath119 expressions analogous to those of eq . ( [ eqn : gamma_n ] ) can be constructed from @xmath108 and @xmath102 , @xmath120 combining eq . [ eqn : gamma_ns ] with the expansion of eq . [ eqn : v3effme ] gives the final result ( compare with eq . [ eqn : bhfinal ] ) , @xmath121 figure [ fig : threebodyenergies ] shows the convergence of the binding energies for @xmath0h and @xmath0he for several included - space and oscillator parameter choices . the expansion of eq . ( [ eqn : bhfinal3 ] ) converges even for very small included spaces , including @xmath122 ( which corresponds to a single matrix element ) . surprisingly , the most rapid convergence occurs for the smallest included spaces . we do not have a convincing physical explanation for this result . note that the zeroth order results ( i.e. @xmath123 ) all overbind in fig . [ fig : threebodyenergies ] . this is possible since the zeroth order calculation is no longer variational , as @xmath102 now depends on @xmath20 ( see eq . [ eqn : v12eff ] ) . hence there is no prescribed method for finding the optimal @xmath1 , other than by trial and error . ( we stress that fully converged results will be independent of @xmath1 , as we have executed the effective theory faithfully . by an optimal @xmath1 we mean one that will speed the convergence . ) the values of @xmath1 used in fig . [ fig : threebodyenergies ] should be near the optimal . note that these values are much larger than those found for the deuteron calculations of ref.@xcite . since @xmath102 is calculated exactly , all short range two - body correlations are correctly taken into account . hence @xmath1 is no longer forced to small values by the demands of short-@xmath37 physics , but instead can relax to values characteristic of the size of the three - body system . as mentioned earlier , @xmath102 is reminiscent of the @xmath124-matrix found in traditional nuclear many - body theory , which is also an infinite ladder sum in particle - particle propagation . however the differences are substantial . traditional shell - model calculations have always used g - matrix interactions at the two - body level , ignoring any dependence the operator may have on spectator nucleons ( including in general certain violations of the pauli principle ) . the operator @xmath102 , on the other hand , depends explicitly on the spectator nucleon kinetic energy , as well as on the kinetic energies of the interacting nucleons . this dependence is manifest in eq . [ eqn : v12eff ] , as the operator @xmath40 in the resolvent @xmath125 represents the sum of the kinetic energies of all nucleons . this dependence on the spectator nucleon is essential for calculating the correct two - body correlations within the three - body system , as @xmath18 is defined by the quanta carried by the three - body configurations . similarly , for @xmath30-body calculations , the analogous operator , @xmath126 , would carry the dependence of all spectator nucleons . it appears that a perturbative expansion for the binding energy of @xmath0he/@xmath0h results only if one first sums two - body ladder interactions to all orders ( _ i.e. _ @xmath102 ) . this summation softens the two - body interaction , which on iteration can then generate the effective three - body interaction ( _ i.e. _ @xmath127 ) perturbatively . a similar procedure was followed in refs.@xcite , where the triton binding energy was calculated by perturbing in di - baryon fields . the di - baryon fields themselves represent an infinite scattering of two - body interactions , in analogy with @xmath102 . a drawback of the perturbative approach of the previous section is that for each included space defined by @xmath2 , there is only a limited range of @xmath1s for which the expansion converges readily . furthermore , there is no definite procedure for estimating the optimal @xmath1 , as the zeroth - order calculation is no longer variational . thus a nonperturbative procedure would be attractive : any @xmath1 could be chosen , and physical observables would prove to be independent of @xmath1 , as they must for a rigorously executed effective theory . such a non - perturbative summation is indeed possible by reshuffling the terms of eq . ( [ eqn : bhfinal3 ] ) to form a summable geometric series and by numerically solving an integral equation . this approach provides an opportunity to directly compare the relative sizes of @xmath102 and @xmath127 as functions of both @xmath2 and @xmath1 . in this section we only show results for the triton system , as @xmath0he calculations produce similar resultshe results , the reader is referred to ref.@xcite ] . the starting point this time is eq . [ eqn : bhfinal3 ] , @xmath128 where the set @xmath129 is given by eq . [ eqn : gamma_ns ] . this expansion can be rearranged into @xmath130 equation [ eqn:3bhfinalnp ] can be verified by expanding its terms out and directly comparing with eq . [ eqn : bhfinal3repeat ] . the following definitions can be made to simplify eq . [ eqn:3bhfinalnp ] , @xmath131 where the above expressions represent infinite summations . substituting these expressions into eq . [ eqn:3bhfinalnp ] gives @xmath132 the two geometric series in the equation above can be summed to give the final desired result , @xmath133 where we have made the following identifications , @xmath134 the usefulness of eq . ( [ eqn:3bodynpert ] ) depends on having an efficient method for calculating the included - space matrix elements of @xmath135 and @xmath136 . this can be done by first considering the state @xmath137 , which , using eq . [ eqn : gamma_inftys ] , satisfies @xmath138 hence @xmath139 satisfies the fredholm integral equation of the second kind@xcite . there are numerous numerical methods available for solving this particular integral equation . we use a general projection algorithm@xcite . due to the large number of partial waves and the complicated structure of the operator @xmath108 , solving the integral equation by matrix inversion is impractical . once @xmath139 is found , the matrix elements @xmath140 can be evaluated . then the matrix elements @xmath141 are easily calculated since @xmath142 which can be verified using eqs . [ eqn : gamma_ns ] and [ eqn : gamma_inftys ] . hence @xmath143 tables [ tab : triton_results ] and [ tab : tritonnp3half ] show the calculated binding energies for the triton system as a function of the included - space size @xmath2 and oscillator parameter @xmath1 . in table [ tab : triton_results ] we have ignored the small isospin - violating parts of the av18 potential . hence isospin is conserved at @xmath144 . table [ tab : tritonnp3half ] includes isospin - violating contributions . this causes a small admixture of @xmath145 components into the ground state . at about the kev level , the results are independent of the choice of the included space , i.e. , of @xmath1 and @xmath2 . the @xmath4 few kev variations illustrate the level of our numerical precision . note that for the values @xmath1=.83 and 1.17 fm , our calculations of the previous section would give non - converging binding energies . this of course emphasizes the non - perturbative nature of these calculations . note that eq . [ eqn:3bodynpert ] has the correct limiting behavior as @xmath146 , @xmath147 . in this limit the projection operator @xmath148 , giving @xmath149 . thus @xmath150 it is also straightforward to show that @xmath151 where we have used eq . [ eqn : t12def ] to obtain the second line above . finally , it is obvious that @xmath127 vanishes in this limit since @xmath152 hence @xmath153 . tables [ tab:00tritonme]- [ tab:22tritonme ] show the variation of the matrix elements of @xmath102 and @xmath127 with @xmath2 and @xmath1 . isospin - violating effects have also been ignored in these results . as the number of included - space examples is small , it is difficult to extract any limiting behavior . yet it does seem that as @xmath2 increases , matrix elements of @xmath102 approach the bare @xmath154 ( indicated by the symbol @xmath155 in the tables ) . this is especially evident in table [ tab:02tritonme ] , where the effect of renormalization is small ( i.e. the renormalized matrix elements are not too different from their bare matrix elements ) . in the cases involving @xmath127 , limiting behavior is also difficult to extract . indeed , in tables . [ tab:00tritonme ] and [ tab:22tritonme ] matrix elements of @xmath127 tend to grow away from zero with increasing @xmath2 rather than tend toward zero . however , in these cases the renormalization is strong , so that limiting behavior would not be expected for such small @xmath2 . in table [ tab:02tritonme ] the renormalization is much weaker , and the limiting behavior @xmath156 is clearly seen for the @xmath157 and @xmath158 mev examples . it is also clear from tables . [ tab:00tritonme]- [ tab:22tritonme ] that the matrix elements of @xmath102 and @xmath127 have a strong dependence on @xmath2 and @xmath1 ( @xmath159 ) , a dependence often ignored in traditional shell - model calculations . generally the contribution of @xmath160 to the overall binding energy is small compared to @xmath161 ( @xmath162 of @xmath102 ) , as is evident from fig . [ fig : v21_vs_v30 ] . this graph gives the @xmath163 , @xmath164 , and @xmath165 contributions to the binding energy as functions of @xmath2 and @xmath1 . interestingly , for each included - space @xmath2 , there are two values of @xmath1 at which the contribution of @xmath160 completely vanishes . table [ tab : b_zeros ] lists specific values . if this result proves to be a generic property of the three - body @xmath7 in heavier nuclei ( that is , if zeros appear in those calculations at approximately the same value of @xmath1 ) , this could greatly simplify the more complicated included - space diagonalizations required in heavy systems . for example , the three - body contribution could be ignored or could be explored in low - order perturbation theory . this would require establishing that the a=3 zero for @xmath160 does persist at about the same @xmath1 in heavier systems . this seems plausible , as the three - body ladder in a heavy nucleus is quite similar to that in a=3 . we found in section [ sect : perturbative ] that the simple perturbative scheme found in ref.@xcite ( _ i.e. _ eq . ( [ eqn : bhfinal ] ) ) did not directly extend to the three - body system due to the additional length scale introduced by the third nucleon . for any choice of jacobi coordinates , this second length scale excludes a class of hard - core interactions from the included space regardless of the choice of @xmath1 . this problem was circumvented by invoking the faddeev decomposition on @xmath7 , which allowed us to sum to all orders the repeated potential scattering between the same two nucleons , generating @xmath102 . hence the strong repulsive two - nucleon force was treated non - perturbatively . the remaining contribution to @xmath7 , @xmath127 , can then be calculated by summing repeated insertions of @xmath102 on alternate pairs of nucleons . for certain ranges of @xmath1 , this expansion converged after several orders of perturbation even for the smallest allowed included spaces , as shown in fig . [ fig : threebodyenergies ] . in section [ sect : nonperturbative ] we described a non - perturbative method for calculating @xmath7 in the three - body system , summarized in eq . ( [ eqn:3bodynpert ] ) . this allowed a direct comparison of the contributions of @xmath163 , @xmath161 , and @xmath160 . because the calculation was non - perturbative , we were able to explore the dependence of the included - space matrix elements ( and establish the independence of the binding energy ) for a wider range of @xmath1s and @xmath2s . also , for each included space , we found two values of @xmath1 where the total binding was due to @xmath102 , a result that should be explored in heavier systems . the numerical methods included an efficient algorithm for solving integral equations ( _ i.e. _ eq . ( [ eqn:3bodyintegraleq ] ) ) . in the current applications to a=3 systems , the numerical effort was not substantial . work is in progress to apply a similar formalism to the alpha particle . as in the three - body case , one must account for the various length scales inherent to the four - body system . this can be done by invoking the faddeev - yakubovsky decompositions on @xmath7 , in direct analogy to the three - body decompositions presented in this paper . such a procedure will not only yield @xmath163 , @xmath166 and @xmath167 ( direct analogs of @xmath161 and @xmath160 ) , but also @xmath168 . this last expression represents the induced effective four - body interaction . it will be interesting to see if these calculations verify the commonly implicit assumption of the hierarchical sizes of these interactions ( _ i.e. _ @xmath169 ) . the four - body effective interaction is the most complicated lowest - order ( s - wave ) operator that can be constructed . furthermore , alpha - particle clustering is an important phenomenological feature of nuclear structure , apparent even in simple systems like @xmath170be . thus it will be interesting to determine the size of this contribution . the a=4 calculations will also check whether zeros in the three - body effective interaction persist . if they do , we may learn something about their `` trajectories '' in @xmath1 as a is inceased . for completeness , we show the derivation of eq . [ eqn : bhre ] , which was first used in ref.@xcite in deriving a perturbative expansion for the deuteron . we begin by explicitly expressing the bh equation in its various components , @xmath171 next consider the operator @xmath172qt\\ = & t\frac{1}{e - qt}\left[1+qv\frac{1}{e - qh}\right]qt\\ = & t\frac{1}{e - qt}\left[1+qv\left\{\frac{1}{e - qt}+\frac{1}{e - qh}qv\frac{1}{e - qt}\right\}\right]qt\\ = & tq\frac{1}{e - tq}\left[e - qt+qv+qv\frac{1}{e - qh}qv\right]\frac{1}{e - qt}qt , \end{aligned}\ ] ] where the relation @xmath173 was used . a similar calculation gives @xmath174,\\ v\frac{1}{e - qh}qt=&\left[v+v\frac{1}{e - qh}qv\right]\frac{1}{e - qt}qt . \end{aligned}\ ] ] substituting eqs . [ eqn : teffexpand ] and [ eqn : vteffexpand ] into eq . [ eqn : heffexpand ] and using the fact that @xmath175 and @xmath176 gives the desired result : @xmath177 hence @xmath178 where @xmath179 and @xmath180 note that the resolvent in eq . [ eqn : appveff ] is now sandwiched between @xmath41s only , and not @xmath40s . forming a perturbative expansion involves expanding this resolvent as was done in eq . [ eqn : resolvent_expansion ] . the operator @xmath51 acts non - trivially only on endshell states . that is , @xmath181 which can be easily verified by expanding the operator in powers of @xmath182 and noting the fact that the kinetic energy operator @xmath40 can change the principal quantum number @xmath183 by at most @xmath184 . these ` tilde ' states @xmath185 represent the culmination of the infinitely repeated scatterings of the operator @xmath182 on the original states @xmath186 . such scatterings persist at large-@xmath37 even though the potential @xmath41 has already vanished . iterating the last expression of eq . [ eqn : v12eff ] on @xmath23 gives @xmath187 replacing @xmath59 by the relevant expression below eq . [ eqn : invgamma0 ] , eq . [ eqn : v12effexpansion1 ] becomes @xmath188 the various terms of eq . [ eqn : v12effexpansion2 ] can be re - ordered to give @xmath189 where @xmath190 represents an infinite sum of two - body interactions between particles 1 and 2 . the solution to eq . [ eqn : v12sum ] is given by the operator @xmath191 , where @xmath192 as mentioned near the end of sect . [ sect : perturbative ] , the operator @xmath191 differs from the usual two - body @xmath24-matrix since it is imbedded within a three - body space . hence @xmath193 contains a dependence on the jacobi momentum @xmath194 of the third ` spectator ' nucleon . this dependence carries over to the operator @xmath191 . hence the designation @xmath195-body @xmath24-matrix . replacing the infinite sum of eq . [ eqn : v12sum ] by @xmath191 and inserting this into eq . [ eqn : v12effexpansion3 ] gives @xmath196 where @xmath197 is given by eq . [ eqn : gamma21def ] , i.e. @xmath198 equation [ eqn : v12effexpansion4 ] can now be summed geometrically , giving the desired result @xmath199^{-1}g_0t_{12}.\ ] ] the relevant matrix element is @xmath200 replacing the radial integrals by their series expansion and changing to dimensionless variables , eq . [ eqn:3mea ] becomes @xmath201 where @xmath202 and @xmath203 an analytic solution to the integral in eq . [ eqn:3meb ] can be found by first changing variables to cylindrical coordinates : @xmath204 hence the integral becomes @xmath205\\ \times \left[\int_0^{\pi/2}d\phi \ ( cos\phi)^{2(1+l+m+m')}(sin\phi)^{2(1+l+m+m')}\right].\end{gathered}\ ] ] the integrals in square brackets have analytic solutions that can be found in any book of integrals ( e.g. see ref.@xcite ) . they are written here for completeness : @xmath206 @xmath207 this procedure generalizes for matrix elements of @xmath70 for higher - body systems . for an a - body problem within jacobi coordinates , a change of variables to ( a-1)-dimensional spherical coordinates is needed so that the resulting ( a-1 ) integrals ` factorize ' ( e.g. those in square brackets of eq . [ eqn:3mec ] ) . we thank andreas nogga for his insightful discussions and contribution of codes for calculating the permutation operator @xmath108 . this work was supported in part by the division of nuclear physics , office of science , u.s . department of energy , and in part under the auspices of the u. s. department of energy by the university of california , lawrence livermore national laboratory under contract no . w-7405-eng-48 . ( 3217,246)(0,0)[l]@xmath37 ( fm)(2712,1021)(0,0)[l ] @xmath208 fm(2712,1111)(0,0)[l ] @xmath209(2712,1246)(0,0)[l]@xmath210(2743,23)(0,0)(d)(3533,90)(0,0 ) 25(3217,90)(0,0 ) 20(2901,90)(0,0 ) 15(2586,90)(0,0 ) 10(2270,90)(0,0 ) 5(1954,90)(0,0 ) 0(1932,1468)(0,0)[r ] 0.4(1932,1246)(0,0)[r ] 0.3(1932,1024)(0,0)[r ] 0.2(1932,801)(0,0)[r ] 0.1(1932,579)(0,0)[r ] 0(1932,357)(0,0)[r]-0.1(1932,135)(0,0)[r]-0.2(1417,246)(0,0)[l]@xmath37 ( fm)(912,1021)(0,0)[l ] @xmath208 fm(912,1111)(0,0)[l ] @xmath209(912,1246)(0,0)[l]@xmath211(943,23)(0,0)(c)(1733,90)(0,0 ) 25(1417,90)(0,0 ) 20(1101,90)(0,0 ) 15(786,90)(0,0 ) 10(470,90)(0,0 ) 5(154,90)(0,0 ) 0(132,1468)(0,0)[r ] 0.8(132,1246)(0,0)[r ] 0.6(132,1024)(0,0)[r ] 0.4(132,801)(0,0)[r ] 0.2(132,579)(0,0)[r ] 0(132,357)(0,0)[r]-0.2(132,135)(0,0)[r]-0.4(3217,1758)(0,0)[l]@xmath37 ( fm)(2712,2533)(0,0)[l ] @xmath208 fm(2712,2623)(0,0)[l ] @xmath209(2712,2758)(0,0)[l]@xmath212(2743,1535)(0,0)(b)(3533,1602)(0,0 ) 25(3217,1602)(0,0 ) 20(2901,1602)(0,0 ) 15(2586,1602)(0,0 ) 10(2270,1602)(0,0 ) 5(1954,1602)(0,0 ) 0(1932,2980)(0,0)[r ] 1.4(1932,2832)(0,0)[r ] 1.2(1932,2684)(0,0)[r ] 1(1932,2536)(0,0)[r ] 0.8(1932,2388)(0,0)[r ] 0.6(1932,2239)(0,0)[r ] 0.4(1932,2091)(0,0)[r ] 0.2(1932,1943)(0,0)[r ] 0(1932,1795)(0,0)[r]-0.2(1932,1647)(0,0)[r]-0.4(1417,1750)(0,0)[l]@xmath37 ( fm)(912,2550)(0,0)[l ] @xmath208 fm(912,2640)(0,0)[l ] @xmath209(912,2775)(0,0)[l]@xmath213(943,1535)(0,0)(a)(1733,1602)(0,0 ) 25(1417,1602)(0,0 ) 20(1101,1602)(0,0 ) 15(786,1602)(0,0 ) 10(470,1602)(0,0 ) 5(154,1602)(0,0 ) 0(132,2980)(0,0)[r ] 2.5(132,2775)(0,0)[r ] 2(132,2570)(0,0)[r ] 1.5(132,2365)(0,0)[r ] 1(132,2160)(0,0)[r ] 0.5(132,1955)(0,0)[r ] 0(132,1750)(0,0)[r]-0.5 ( 1952,1489)(0,0)[r ] 0.4(1952,1423)(0,0)[r ] 0.2(1952,1357)(0,0)[r ] 0(1952,1292)(0,0)[r]-0.2(1952,1226)(0,0)[r]-0.4(1952,1160)(0,0)[r]-0.6(1952,1094)(0,0)[r]-0.8(1952,1028)(0,0)[r]-1(1952,962)(0,0)[r]-1.2(1952,896)(0,0)[r]-1.4(1952,830)(0,0)[r]-1.6(3291,391)(0,0)@xmath194(3339,696)(0,0)[l ] 8(3282,634)(0,0)[l ] 7(3225,572)(0,0)[l ] 6(3168,511)(0,0)[l ] 5(3112,449)(0,0)[l ] 4(3055,387)(0,0)[l ] 3(2998,325)(0,0)[l ] 2(2941,264)(0,0)[l ] 1(2884,202)(0,0)[l ] 0(2358,115)(0,0)(d)(2358,235)(0,0)@xmath8(2852,186)(0,0 ) 8(2754,221)(0,0 ) 7(2655,257)(0,0 ) 6(2557,293)(0,0 ) 5(2458,328)(0,0 ) 4(2359,364)(0,0 ) 3(2261,400)(0,0 ) 2(2162,435)(0,0 ) 1(2064,471)(0,0 ) 0(152,1489)(0,0)[r ] 0.1(152,1416)(0,0)[r ] 0.08(152,1343)(0,0)[r ] 0.06(152,1270)(0,0)[r ] 0.04(152,1196)(0,0)[r ] 0.02(152,1123)(0,0)[r ] 0(152,1050)(0,0)[r]-0.02(152,977)(0,0)[r]-0.04(152,904)(0,0)[r]-0.06(152,830)(0,0)[r]-0.08(1491,391)(0,0)@xmath194(1539,696)(0,0)[l ] 8(1482,634)(0,0)[l ] 7(1425,572)(0,0)[l ] 6(1368,511)(0,0)[l ] 5(1312,449)(0,0)[l ] 4(1255,387)(0,0)[l ] 3(1198,325)(0,0)[l ] 2(1141,264)(0,0)[l ] 1(1084,202)(0,0)[l ] 0(558,115)(0,0)(c)(558,235)(0,0)@xmath8(1052,186)(0,0 ) 8(954,221)(0,0 ) 7(855,257)(0,0 ) 6(757,293)(0,0 ) 5(658,328)(0,0 ) 4(559,364)(0,0 ) 3(461,400)(0,0 ) 2(362,435)(0,0 ) 1(264,471)(0,0 ) 0(1952,3649)(0,0)[r ] 1.6(1952,3576)(0,0)[r ] 1.4(1952,3503)(0,0)[r ] 1.2(1952,3430)(0,0)[r ] 1(1952,3356)(0,0)[r ] 0.8(1952,3283)(0,0)[r ] 0.6(1952,3210)(0,0)[r ] 0.4(1952,3137)(0,0)[r ] 0.2(1952,3064)(0,0)[r ] 0(1952,2990)(0,0)[r]-0.2(3291,2551)(0,0)@xmath194(3339,2856)(0,0)[l ] 8(3282,2794)(0,0)[l ] 7(3225,2732)(0,0)[l ] 6(3168,2671)(0,0)[l ] 5(3112,2609)(0,0)[l ] 4(3055,2547)(0,0)[l ] 3(2998,2485)(0,0)[l ] 2(2941,2424)(0,0)[l ] 1(2884,2362)(0,0)[l ] 0(2358,2275)(0,0)(b)(2358,2395)(0,0)@xmath8(2852,2346)(0,0 ) 8(2754,2381)(0,0 ) 7(2655,2417)(0,0 ) 6(2557,2453)(0,0 ) 5(2458,2488)(0,0 ) 4(2359,2524)(0,0 ) 3(2261,2560)(0,0 ) 2(2162,2595)(0,0 ) 1(2064,2631)(0,0 ) 0(152,3649)(0,0)[r ] 0.08(152,3539)(0,0)[r ] 0.06(152,3430)(0,0)[r ] 0.04(152,3320)(0,0)[r ] 0.02(152,3210)(0,0)[r ] 0(152,3100)(0,0)[r]-0.02(152,2990)(0,0)[r]-0.04(1491,2551)(0,0)@xmath194(1539,2856)(0,0)[l ] 8(1482,2794)(0,0)[l ] 7(1425,2732)(0,0)[l ] 6(1368,2671)(0,0)[l ] 5(1312,2609)(0,0)[l ] 4(1255,2547)(0,0)[l ] 3(1198,2485)(0,0)[l ] 2(1141,2424)(0,0)[l ] 1(1084,2362)(0,0)[l ] 0(558,2275)(0,0)(a)(558,2395)(0,0)@xmath8(1052,2346)(0,0 ) 8(954,2381)(0,0 ) 7(855,2417)(0,0 ) 6(757,2453)(0,0 ) 5(658,2488)(0,0 ) 4(559,2524)(0,0 ) 3(461,2560)(0,0 ) 2(362,2595)(0,0 ) 1(264,2631)(0,0 ) 0 ( 3196,389)(0,0)[r]triton(3196,506)(0,0)[r]helium3(3196,584)(0,0)[r ] b=1.66 fm(3196,664)(0,0)[r]@xmath220(2730,20)(0,0)(d)(3540,80)(0,0 ) 10(3216,80)(0,0 ) 8(2892,80)(0,0 ) 6(2568,80)(0,0 ) 4(2244,80)(0,0 ) 2(1920,80)(0,0 ) 0(1900,1364)(0,0)[r]-2(1900,1209)(0,0)[r]-4(1900,1053)(0,0)[r]-6(1900,898)(0,0)[r]-8(1900,742)(0,0)[r]-10(1900,587)(0,0)[r]-12(1900,431)(0,0)[r]-14(1900,276)(0,0)[r]-16(1900,120)(0,0)[r]-18(3192,2394)(0,0)[r]triton(3192,2511)(0,0)[r]helium3(3192,2589)(0,0)[r ] b=2.04 fm(3192,2669)(0,0)[r]@xmath221(2720,1532)(0,0)(b)(3540,1592)(0,0 ) 10(3212,1592)(0,0 ) 8(2884,1592)(0,0 ) 6(2556,1592)(0,0 ) 4(2228,1592)(0,0 ) 2(1900,1592)(0,0 ) 0(1880,2876)(0,0)[r]-3(1880,2669)(0,0)[r]-4(1880,2461)(0,0)[r]-5(1880,2254)(0,0)[r]-6(1880,2047)(0,0)[r]-7(1880,1839)(0,0)[r]-8(1880,1632)(0,0)[r]-9(1404,951)(0,0)[r]triton(1404,1068)(0,0)[r]helium3(1404,1146)(0,0)[r ] b=2.04 fm(1404,1226)(0,0)[r]@xmath222(950,20)(0,0)(c)(40,742 ) ( 1740,80)(0,0 ) 10(1424,80)(0,0 ) 8(1108,80)(0,0 ) 6(792,80)(0,0 ) 4(476,80)(0,0 ) 2(160,80)(0,0 ) 0(140,1364)(0,0)[r]-2(140,1226)(0,0)[r]-3(140,1088)(0,0)[r]-4(140,949)(0,0)[r]-5(140,811)(0,0)[r]-6(140,673)(0,0)[r]-7(140,535)(0,0)[r]-8(140,396)(0,0)[r]-9(140,258)(0,0)[r]-10(140,120)(0,0)[r]-11(1408,2375)(0,0)[r]triton(1408,2492)(0,0)[r]helium3(1408,2570)(0,0)[r ] b=1.66 fm(1408,2650)(0,0)[r]@xmath3(960,1532)(0,0)(a)(40,2254 ) ( 1740,1592)(0,0 ) 10(1428,1592)(0,0 ) 8(1116,1592)(0,0 ) 6(804,1592)(0,0 ) 4(492,1592)(0,0 ) 2(180,1592)(0,0 ) 0(160,2876)(0,0)[r]-3(160,2763)(0,0)[r]-3.5(160,2650)(0,0)[r]-4(160,2537)(0,0)[r]-4.5(160,2424)(0,0)[r]-5(160,2311)(0,0)[r]-5.5(160,2197)(0,0)[r]-6(160,2084)(0,0)[r]-6.5(160,1971)(0,0)[r]-7(160,1858)(0,0)[r]-7.5(160,1745)(0,0)[r]-8(160,1632)(0,0)[r]-8.5 ( 0,0 ) , @xmath223 , and @xmath224 for @xmath25 and @xmath225 triton calculations as a function of @xmath1 . the upper pairs of lines refer to @xmath226 , the middle pair @xmath127,and the bottom two @xmath102 . note that the respective sums of all three terms gives the triton binding energy.,title="fig : " ] 0 & -7.6147 ( 218.533 ) & -7.6144 ( 74.113 ) & -7.6148 ( 26.480 ) + 2 & -7.6143 ( 65.260 ) & -7.6144 ( 26.732 ) & -7.6145 ( 12.556 ) + 4 & -7.6144 ( 26.226 ) & -7.6141 ( 15.840 ) & -7.6141 ( 9.068 ) + 6 & -7.6179 ( 13.268 ) & -7.6140 ( 9.899 ) & -7.6141 ( 6.524 ) + 8 & -7.6179 ( 5.985 ) & -7.6136 ( 6.725 ) & -7.6139 ( 5.257 ) + 10 & -7.6182 ( 1.606 ) & -7.6131 ( 4.156 ) & -7.6138 ( 4.093 ) + 8 & 10 & @xmath155 + 60 & 64.51 & 90.00 & 90.00 & 90.00 & 90.00 & 90.00 & 90.00 + & -68.36 & -38.41 & -32.54 & -26.40 & -21.03 & -16.52 & 128.50 + & -3.76 & 8.26 & 22.38 & 37.25 & 51.34 & 64.06 & 0.00 + 30 & 33.59 & 45.00 & 45.00 & 45.00 & 45.00 & 45.00 & 45.00 + & -45.11 & -47.20 & -45.61 & -44.05 & -42.50 & -40.99 & 29.10 + & 3.91 & 3.71 & 6.17 & 8.96 & 11.96 & 15.03 & 0.00 + 15 & 17.67 & 22.50 & 22.50 & 22.50 & 22.50 & 22.50 & 22.50 + & -26.37 & -31.54 & -30.43 & -29.73 & -29.16 & -28.66 & 3.98 + & 1.08 & 2.43 & 4.16 & 5.03 & 5.68 & 6.26 & 0.00 + 10 & @xmath155 + 60 & -33.149 & -47.107 & -46.936 & -46.873 & -46.837 & -45.375 + & 0.146 & -1.771 & -1.692 & -1.016 & -0.330 & 0.000 + 30 & -13.400 & -18.898 & -18.772 & -18.665 & -18.564 & -18.952 + & 0.641 & -0.197 & -0.655 & -0.920 & -1.065 & 0.000 + 15 & -5.053 & -6.964 & -6.911 & -6.884 & -6.857 & -6.769 + & 0.398 & 0.385 & 0.296 & 0.194 & 0.104 & 0.000 + 10 & @xmath155 + 60 & 122.210 & 150.000 & 150.000 & 150.000 & 150.000 & 150.000 + & -11.616 & 11.714 & 15.141 & 18.365 & 20.788 & 89.884 + & 10.381 & 18.639 & 25.302 & 31.312 & 36.794 & 0.000 + 30 & 61.986 & 75.000 & 75.000 & 75.000 & 75.000 & 75.000 + & -16.159 & -12.223 & -11.338 & -10.376 & -9.496 & 21.034 + & 4.571 & 7.211 & 8.547 & 9.790 & 11.033 & 0.000 + 15 & 31.679 & 37.500 & 37.500 & 37.500 & 37.500 & 37.500 + & -10.649 & -10.579 & -10.162 & -9.805 & -9.509 & 4.019 + & 2.144 & 3.365 & 3.782 & 4.071 & 4.335 & 0.000 +

we report optical and infrared spectroscopic observations of the type ia sn 1999ee and the type ib / c sn 1999ex , both of which were hosted by the galaxy ic 5179 . for sn 1999ee we obtained a continuous sequence with an unprecedented wavelength and temporal coverage beginning 9 days before maximum light and extending through day 42 . before maximum light sn 1999ee displayed a normal spectrum with a strong si ii absorption , thus showing that not all slow - declining sne are spectroscopically peculiar at these evolutionary phases . a comparative study of the infrared spectra of sn 1999ee and other type ia supernovae shows that there is a remarkable homogeneity among the branch - normal sne ia during their first 60 days of evolution . sn 1991bg - like objects , on the other hand , display spectroscopic peculiarities at ir wavelengths . sn 1999ex was characterized by the lack of hydrogen lines , weak optical he i lines , and strong he i,20581 , thus providing an example of an intermediate case between pure ib and ic supernovae . we conclude therefore that sn 1999ex provides first clear evidence for a link between the ib and ic classes and that there is a continuous spectroscopic sequence ranging from the he deficient sne ic to the sne ib which are characterized by strong optical he i lines .

astro-ph0203491

the last ten years have witnessed an enormous progress in our knowledge of the optical properties of supernovae ( sne ) of all types . however , comparatively little is still known about these objects in near - infrared ( nir ) wavelengths . given the rapid technological development of nir light detection over recent years , in 1999 we started a program to obtain optical and nir photometry and spectroscopy of nearby sne ( @[email protected] ) . so far , the `` supernova optical and infrared survey '' ( soirs ) program has gathered high - quality observations for @xmath320 sne . so far we have reported results for the bright type ii sn 1999em @xcite . in this paper we report spectroscopic observations of two of the best - observed objects included in our program , the type ia sn 1999ee and the type ib / c sn 1999ex , both of which exploded in the same galaxy within three weeks . sn 1999ee was discovered by m. wischnjewsky on a film taken on 1999 october 7.15 ( jd 2451458.65 ) in the course of the el roble survey @xcite . the sn exploded 10 arcsec east and 10 arcsec south of the nucleus of the spiral galaxy ic 5179 , a very active star - forming galaxy with a heliocentric redshift of 3,498 km s@xmath4 . the optical spectrum taken on 1999 october 9.10 revealed the defining si ii @xmath06355 feature of the type ia class @xcite . the high expansion velocity of 15,700 km s@xmath4 deduced from this line and the faint apparent magnitude at discovery ( @xmath317.5 ) indicated that this object had been found several days before maximum light . at the distance of the host galaxy sn 1999ee offered the promise to reach @xmath5=14.5 about 10 days later , thus proving to be an excellent target for a detailed study of a sn ia early since explosion , both at optical and ir wavelengths . this discovery occurred at the very beginning of one of the soirs follow - up runs , previously scheduled for 1999 october - november , so we decided to give first priority to sn 1999ee . the unique opportunity afforded by sn 1999ee led the eso director general to allocate director s discretionary time to this project in order to secure the best possible data for sn 1999ee . as a result of this effort we obtained a superb dataset of optical / ir photometric and spectroscopic observations of the first 50 days of the evolution of sn 1999ee . three weeks after the discovery of sn 1999ee , a second sn exploded in the same galaxy that hosted sn 1999ee . the discovery of sn 1999ex was made by martin et al . ( 1999 ) on 1999 november 9.51 ut ( jd 2451492.01 ) in the course of the parg automated supernova search at perth observatory . they reported that the sn was not visible in a deep exposure taken on 1999 october 25.58 ( estimated limiting mag 19 ) , and that the object was slowly brightening . a spectrum taken on 1999 november 14.13 @xcite showed that sn 1999ex had close resemblance to that of the type ic sn 1994i taken near maximum light @xcite . this led us initially to classify sn 1999ex as a ic event although we believe that it should be typed as an intermediate ib / c object ( see sec . [ sn99ex.sec ] ) . as soon as this object was discovered we decided to include it in our optical / ir spectroscopic follow - up . the observations gathered for sne 1999ee and 1999ex have an unprecedented temporal and wavelength coverage and afford the possibility to carry out a detailed comparison with atmosphere models . in this paper we report the spectroscopic observations of sne 1999ee and 1999ex and we discuss these results . the spectra presented here are available in electronic form to other researchers ( contact m. h. if interested ) . optical and ir photometry will be published and discussed in detail elsewhere by @xcite and @xcite , respectively . we obtained optical and ir spectra of sne 1999ee and 1999ex with the eso ntt / emmi / sofi , the danish 1.5-m / dfosc equipments at la silla , and the vlt / isaac instrument at cerro paranal between 1999 october 9 and 1999 november 28 . we also obtained three optical spectra with the ctio 4-m telescope on 1999 october 9 , the ctio 1.5-m telescope on 1999 october 27 , and the las campanas dupont 2.5-m telescope on 1999 october 16 . table 1 gives the journal of the observations . the ntt observations comprised three different setups . we used the blue channel of emmi equipped with a tek ccd ( 1024x1024 ) and grating 5 ( 158 lines mm@xmath4 ) which , in first order , delivered spectra with a dispersion of 3.5 pix@xmath4 and a useful wavelength range between 0.330 and 0.525 @xmath6 . with the red channel , ccd tek 2048 , and grating 13 ( 150 lines mm@xmath4 ) the dispersion was 2.7 pix@xmath4 and the spectral coverage included from 0.470 through 1.100 @xmath6 in first order . since this setup had potential second - order contamination beyond @xmath30.6 @xmath6 we decided to take one spectrum with the og530 filter and a second observation without the filter , in order to provide an overlap with the blue spectrum . thus , a single - epoch observation comprised three spectra . the observations with emmi started with calibrations during day time ( bias and dome flat - field exposures ) . the night began with the observation of a spectrophotometric standard ( from the list of @xcite ) through a wide slit of 10 arcsec , after which we observed the sne with a slit of 1 arcsec . before the discovery of sn 1999ex on 1999 november 9 we oriented the slit along the line connecting sn 1999ee and the host galaxy nucleus . on 1999 november 14 and 19 , on the other hand , we oriented the slit along the two sne in order to get simultaneous spectra . since the airmass of our observations was always below 1.5 we do not expect serious systematic errors in the observed fluxes due to atmospheric refraction . we took two exposures of the sne per spectral setup , each of the same length ( typically 300 - 600 sec ) . immediately following this observation we observed a he - ar lamp , at the same position of the sne and before changing the optical setup in order to ensure an accurate wavelength calibration . at the end of the night we observed a second flux standard . we also used the danish 1.5-m telescope and the dfosc instrument at la silla on five nights , between 1999 october 20 and november 28 , in order to improve the temporal coverage of our spectroscopic observations . in all cases we employed a 2k@xmath72k loral ccd and two different grisms to secure a spectral coverage comparable to that obtained with emmi . with grism 3 ( 400 lines mm@xmath4 ) we covered a useful wavelength range between 0.33 and 0.66 @xmath6 with a dispersion of 2.3 pix@xmath4 . with grism 5 ( 300 lines mm@xmath4 ) we sampled the range between 0.53 and 0.98 @xmath6 at 3.1 pix@xmath4 . the red setup produced spectra with significant fringing beyond 0.75 @xmath6 which we did not attempt to remove . we observed the sne with a slit of 2 arcsec . although we did not rotate the slit along the parallactic angle , we obtained the first four spectra with airmass @xmath21.1 . the last spectrum taken on 1999 november 28 , on the other hand , might suffer from atmospheric refraction since we obtained it with an airmass between 1.4 - 2.0 . we took two 1200 sec exposures of the sne per grism , followed by a he - ne arc lamp exposure , and spectra of a flux standard with a 5 arcsec slit . we obtained a spectrum of sn 1999ee on 1999 october 9 ( two days after discovery ) with the r - c spectrograph of the ctio 4-m telescope , a 3k@xmath71k loral ccd , and grating kpgl-2 ( 316 lines mm@xmath4 ) in first order . we took one 1800 sec exposure of the sn ( through thick cirrus ) with a 1.5 arcsec slit oriented along the parallactic angle , a he - ar lamp exposure , and spectra of three flux standards through a 10 arcsec slit . the resulting sn spectrum had a dispersion of 1.9 pix@xmath4 and useful wavelength coverage of 0.33 - 0.87 @xmath6 . second - order contamination was expected beyond 0.66 @xmath6 since we did not include a blocking filter in the optical path . we obtained another spectrum at ctio with the 1.5-m telescope and the r - c spectrograph on 1999 october 27 . in this case we used a 1200@xmath7800 loral ccd , grating 58 ( 400 lines mm@xmath4 ) in second order , a 2 arcsec slit , and a cuso@xmath8 order - blocking filter . the resulting spectrum had a dispersion of 1.1 pix@xmath4 and useful wavelength coverage of 0.37 - 0.50 @xmath6 . we obtained six 1200 sec exposures of sn 1999ee at an airmass @xmath2 1.1 , a he - ar lamp image , and two spectra of flux standards . we also obtained a spectrum of sn 1999ee on 1999 october 16 with the las campanas dupont 2.5-m telescope and the wide field ccd spectrograph . we used a 2048@xmath72048 tek ccd and a blue grism . the resulting spectrum had a dispersion of 3 pix@xmath4 and useful wavelength coverage of 0.36 - 0.92 @xmath6 . second - order contamination was expected beyond 0.66 @xmath6 since we did not use a blocking filter . the reductions consisted in subtracting the overscan and bias from every frame . next , we constructed a normalized flat - field from the quartz - lamp image , duly normalized along the dispersion axis . we proceeded by flat - fielding all of the object frames and extracting 1-d spectra from the 2-d images . we followed the same procedure for the he - ar frames which we used to derive the wavelength calibration for the sne . then we derived a response curve from the two flux stars , which we applied to the sn spectra , in order to get flux calibrated spectra . from the pair of flux - calibrated spectra that we obtained for each spectral setup we removed cosmic rays and obtained a clean spectrum of each sn . the last step consisted in merging the spectra obtained with the different spectroscopic setups . to avoid discontinuities in the combined spectrum we grey - shifted the three spectra relative to each other using the overlap regions . finally , we computed the synthetic @xmath5-band magnitude from the resulting spectra ( following the precepts described in appendix b of @xcite ) and we grey - shifted them so that the flux level matched our observed @xmath5 magnitudes . we checked the spectrophotometric quality of the spectra by computing @xmath9 synthetic magnitudes and comparing them to the observed @xmath9 magnitudes of @xcite . excluding the spectra obtained on 1999 november 14 , this test yielded the following mean differences : @xmath10-@[email protected] , @xmath13-@[email protected] , and @xmath15-@[email protected] , which implies that the relative spectrophotometry at these wavelengths is accurate to 10% or better . the november 14 spectrum , on the other hand , shows @xmath17 fluxes that fall 0.35 mag lower than the photometric @xmath17 magnitudes and @xmath18 fluxes that exceed by 0.2 mag the observed broadband @xmath18 magnitudes , perhaps owing to atmospheric refraction effects . we obtained five ir spectra with the vlt / antu telescope at cerro paranal , between 1999 october 9 and november 28 ( see table 1 ) . we employed the ir spectro - imager isaac @xcite in low resolution mode ( r@xmath3500 ) , with four different gratings that permitted us to obtain spectra in the @xmath19 , @xmath20 , @xmath21 , and @xmath22 bands . we used these gratings in 5@xmath23 , 4@xmath23 , 3@xmath24 and 2@xmath25 order , respectively , which yielded useful data in the spectral ranges 0.984 - 1.136 , 1.109 - 1.355 , 1.415 - 1.818 , and 1.846 - 2.560 @xmath6 . the light detector was a hawaii - rockwell 1024x1024 array . a typical ir observation started during daytime by taking calibrations . we began taking flat - field images using an internal source of continuum light . we obtained multiple on and off image pairs with the same slit used during the night ( 0.6 arcsec ) . we then took xe - ar lamp images ( with the lamp on and off ) with a narrow slit ( 0.3 arcsec ) in order to map geometric distortions . the observations of the sne consisted in an abbaab cycle , where a is an image of the sne offset by 70 arcsec along the slit relative to the b image . this technique of nodding the objects along the slit allowed us to use the a image as an on - source observation and the b image as the off - source sky frame , and viceversa . at each position we exposed for 240 sec , conveniently split into two 120-sec images in order to remove cosmic rays and bad pixels from the final spectra . after completing the abbaab cycle we immediately obtained a pair of on - off arc lamp exposures without moving the telescope or changing optical elements to ensure an accurate wavelength calibration . we then switched to the next grating and repeated the above object - arc procedure until completing the observations with the four setups . for flux calibration we decided to observe a bright solar - analog star , close in the sky to the sne in order to minimize variations in the atmospheric absorptions @xcite . the selected star was hip 109508 , of spectral type g3v , @xmath5=8.0 , @xmath17-@xmath5=0.59 , and located only 3@xmath26 from the sne . in this case we took two ab pairs for each grating . to avoid saturating the detector , we took the shortest possible exposures ( 1.77 sec ) allowed by the electronics that controlled the detector . since the minimum time required before offsetting the telescope was @xmath360 sec , we took ten exposures at each position which provided an exceedingly good signal - to - noise ratio ( s / n ) for the flux standard . our data reduction procedures were explained in detail by @xcite , so we include here only a brief summary . after dividing all of the object images by a normalized flat - field , we performed a first - order sky subtraction by subtracting the a images from the b exposures ( and viceversa ) . we then shifted the a - b image relative to the b - a image in the spatial direction until matching the two spectra , and we added the shifted a - b image to the b - a image so that the resulting frame lacked any sky background , except for the pixel - to - pixel fluctuations expected from photon statistics and readout noise . we then extracted 1-d spectra of the objects from the sky - subtracted frames , making sure to subtract residual dc offset and galaxy light from a window adjacent to the object . for flux calibration we adopted the technique described by @xcite , which consists in dividing the spectrum of interest by a solar - type star to remove the strong telluric ir features , and multiplying the resulting spectrum by the solar spectrum to eliminate the intrinsic features ( pseudonoise ) introduced by the solar - type star . before using the solar spectrum we convolved it with a kernel function in order to reproduce the spectral resolution of the solar - analog standard hip 109508 , we shifted it in wavelength according to the radial velocity ( 68 km s@xmath4 ) of hip 109508 , and we scaled it down to the equivalent of @xmath5=8.0 which corresponds to the observed magnitude of hip 109508 . this technique worked very well to remove telluric lines . on the other hand , it introduced a small systematic error in the flux calibration of the sn due to departures between the solar spectrum and the actual spectral energy distribution of the solar - analog standard . according to atmosphere models the difference in continuum flux for stars with @xmath27=5,500 and 6,000 k ( which correspond to spectral types g8v - f9v , @xcite ) is smaller than 10% in the nir region . a g3v star like hip 109508 has nearly the same spectral type of the sun , so its effective temperature must be close ( within @xmath12100 k ) to that of the adopted solar model . hence , the flux difference between the solar - analog standard and the adopted spectrum should be less than 10% . the @xmath28=0.59 color of hip 109508 suggests little or no reddening so the relative spectrophotometry is probably accurate to 5% or better . the result of these operations were four spectra covering the @xmath19 , @xmath20 , @xmath21 , and @xmath22 bands , which we combined into one final ir spectrum for each sn . given the significant overlap of the @xmath19 and @xmath20 band spectra , we were able to grey - shift the @xmath20 spectrum relative to the @xmath19 spectrum . then we used our broad - band @xmath29 magnitudes to grey - shift the individual spectra . on six occasions we used the ntt / sofi instrument at la silla in service observing mode , in order to complement the ir spectroscopic follow - up . we employed a hawaii - rockwell 1024x1024 light detector , and the blue and red grisms that permitted us to obtain spectra in the ranges 0.95 - 1.64 @xmath6 and 1.53 - 2.52 @xmath6 , with resolutions of 7.0 and 10.2 , respectively . we obtained the spectra with slit widths between 0.6 and 1 arcsec . we nodded the objects along the slit , after exposing for 150 sec at each position . we completed several pairs of ab observations for the sne , we took xe - ne arc images at the position of the targets , and we observed the flux standard hip 109508 . we reduced these data following the same procedure explained above for isaac . the large overlap of 1.53 - 1.64 @xmath6 between the blue and red spectra allowed us to grey - shift the red spectra relative to blue spectra , except on 1999 october 18 when we were only able to obtain the red spectrum . finally , we shifted the resulting spectra using the observed @xmath29 magnitudes @xcite . before analyzing the spectroscopic observations it proves necessary to mention the photometric properties of sn 1999ee . a detailed discussion of the photometric observations can be found in @xcite . in brief , the decline rate of @xmath30@xmath31=0.94 puts sn 1999ee in the group of slowest - declining sne ia . after correcting the observed magnitudes for @xmath22 terms and dust extinction , and assuming a hubble constant of 63 km s@xmath4 mpc@xmath4 @xcite , sn 1999ee had a peak visual magnitude of @xmath32=-19.94 ( jd 2451469.1 ) . these values lie comfortably well with the peak magnitude - decline rate relation for sne ia @xcite . this analysis reveals that sn 1999ee was a luminous type ia event and a very interesting object to check the claim by @xcite that slow - declining events display spectroscopic peculiarities before maximum light . figure [ sn99ee.opt.fig ] displays the optical spectra of sn 1999ee in the rest - frame of the sn , after correcting the observed spectra for the 3,498 km s@xmath4 recession velocity of the host galaxy . we also dereddened the spectra assuming a galactic reddening of @xmath33=0.02 @xcite , and a host galaxy reddening of @xmath33=0.30 derived from the sn optical colors @xcite and optical / ir colors @xcite . the strongest telluric lines are indicated with the @xmath34 symbol . the first spectrum , taken on jd 2451460.56 ( 9 days before maximum ) , exhibited a blue continuum and the characteristic p - cygni profile of the si ii @xmath06355 line of sne ia . other prominent features in this spectrum were the absorption at @xmath33200 attributed to co ii , the ca ii h&k @xmath353934,3968 blend , mg ii @xmath04481 , the blend of lines attributed to fe ii , si ii , s ii around 4550 - 5050 , and the ca ii @xmath358498,8542,8662 triplet . the presence of the na i d @xmath355890,5896 , ca ii @xmath03934 , and ca ii @xmath03968 interstellar lines with equivalent widths of 2.3 , 0.8 , and 0.6 , respectively , revealed a non - negligible amount of absorption in the host galaxy , which agrees with the color analysis of sn 1999ee . by maximum light the `` w '' shaped s ii @xmath355454,5640 feature was well developed and the ca lines were significantly stronger . si ii @xmath04129 was already present although it was quite weak . the si ii @xmath05972 line also was weak relative to si ii @xmath06355 . nugent et al . ( 1995 ) defined the parameter @xmath36(si ii ) as the relative fluxes of this pair of lines and found a tight correlation with the sn luminosity . in this case we obtained @xmath36(si ii)=0.23 which corresponds to the value of a luminous sne ia , in good concordance with the decline rate of @xmath30@xmath31=0.94 and the peak magnitudes derived from the photometric observations . by day 20 the spectrum was much redder . the si ii @xmath06355 line was very weak , the ca triplet had a pronounced p - cygni profile , and the region between 4000 - 5500 was dominated by strong features due to fe ii transitions . the absorption attributed to na i d @xmath355890,5896 was very prominent . overall , these data showed that sn 1999ee was a genuine type ia event with the usual spectral features of other normal sne ia , usually known as branch - normal @xcite . figure [ sn99ee.vel.fig ] presents the expansion velocities derived from the absorption minima of si ii @xmath06355 , after correcting for the recession velocity of the host galaxy . apparently the velocity derived from si ii @xmath06355 decreased rapidly from 16,000 to 10,000 km s@xmath4 between days -7 and -2 , after which the velocity decreased slowly . for comparison we include as solid line the velocities for sn 1990n which was also caught several days before maximum @xcite . both sne showed an inflection in the velocity curve . before maximum , however , the expansion of sn 1999ee was somewhat higher compared to contemporary velocities of sn 1990n . at maximum light and later times , on the other hand , both sne had similar velocities . figure [ sn99ee.ir1.fig ] shows the resulting rest - frame ir spectra of sn 1999ee . the first spectrum , taken nine days before maximum , was quite featureless . a weak absorption line could be seen at 10520 with a weak emission component that seemed to constitute a p - cygni profile . this feature was observed for the first time in an early - time spectrum of sn 1994d and was tentatively identified as due to either he i @xmath010830 or mg ii @xmath010926 @xcite . according to the atmosphere models computed by @xcite , this feature should be due to mg ii since the amount of he in their models is not sufficient to form a line . this implies that the minimum of the mg absorption had an expansion velocity of 11,100 km s@xmath4 . an emission due to fe iii @xcite was also present around 12550 . there was an indication of a p - cygni profile with the absorption minimum at 16160 and peak emission at 16840 . this feature was also present in the sn 1994d spectrum . the theoretical modeling of wheeler et al . suggests that this line is due to si ii , although it might have a fe ii component @xcite . at longer wavelengths the spectrum is even more featureless except for a small p - cygni line with a minimum at 19590 and peak at 20440 . the emission component was also noted in sn 1994d and was attributed to si iii by wheeler et al . by maximum light the mg ii @xmath010926 , fe iii , and si ii feature near 16500 became more prominent and a peak developed around 18100 . one week past maximum the spectrum was dramatically different and became dominated by strong absorption / emission features . the mg ii @xmath010926 line disappeared and a weak feature at 10000 became evident , possibly forming a p - cygni profile . the most remarkable features were the strong and wide ( fwhm @xmath3 12,000 km s@xmath4 ) peaks attributed to blends of co ii , fe ii , and ni ii at 15500 and 17500 @xcite . two weeks after maximum these peaks developed even further and new peaks appeared at 19900 and 21000 . a series of unidentified absorptions also could be observed in the @xmath20 band at 10400 , 10700 , and 11100 . one month after maximum two additional broad emission peaks became evident at 22500 and 23600 which , according to wheeler et al . , are caused by iron - group elements ( co , ni ) as well as intermediate mass elements with appreciable contribution from si . while the broad emissions appeared to redshift significantly as the sn evolved , the features that we identify as absorptions did not show any appreciable change in velocity . figure [ sn99ex.opt.fig ] displays the optical spectra of sn 1999ex in the sn rest - frame . the first spectrum , obtained on jd 2451496.62 ( one day before @xmath17 maximum ) , was characterized by a reddish continuum and several broad absorption / emission features due to ca ii h&k @xmath353934,3968 , fe ii , mg i , na i d @xmath355890,5896 , si ii @xmath06355 , and the ca triplet with a clear p - cygni profile . we also found convincing evidence for he i @xmath356678,7065 with expansion velocities of 6,000 km s@xmath4 ( similar to velocities displayed by the fe , mg , na , si , and o lines ) . it is possible that the feature at @xmath34400 , usually attributed to a blend of mg ii and fe ii , had a contribution due to the he i @xmath04471 . likewise , the feature at @xmath34800 could be a blend of fe ii @xmath04924 and he i @xmath04921 , while the na i d @xmath355890,5896 doublet could have a contribution from he i @xmath05876 . interstellar lines due to na i d @xmath355890,5896 , ca ii @xmath03934 , and ca ii @xmath03968 were observed in absorption with equivalent widths of 2.8 , 1.8 , 0.9 , at the wavelengths corresponding to the rest - frame of the host galaxy . according to the correlation between equivalent widths and reddening for galactic stars @xcite , sn 1999ex was reddened by @[email protected]@xmath37 . however , there appears to be a significant departure of sne from this calibration @xcite , so this estimate is very uncertain . the first spectrum of sn 1999ex bore quite resemblance to that of sn 1994i . @xcite argued that sn 1994i was a type ic event on the grounds that this object 1 ) did not show evidence for hydrogen lines ( ruling out a type ii event ) , 2 ) had a si ii @xmath06355 line more subtle than in classical sne ia , and 3 ) he i @xmath05876 was absent or weak ( and blended with na d ) ( ruling out a type ib event ) . based on the resemblance between these two spectra , we initially classified sn 1999ex as a type ic sn @xcite . the relatively stronger he lines displayed by sn 1999ex suggest that this object could well be an intermediate case between the ib and ic subclasses ( we will return to this point in section [ sn99ex.sec ] ) . during the following two weeks the sn spectrum evolved slowly and most of the lines became stronger . figure [ sn99ex.ir.fig ] displays the ir spectra of sn 1999ex in the sn rest - frame . the most prominent feature in these spectra is the p - cygni profile of he i @xmath010830 which was also observed in sn 1994i . the minimum of the absorption yields an expansion velocity of 8,000 km s@xmath4 , in good agreement with the values derived from the optical spectra . the other evident feature is the p - cygni profile due to he i @xmath020581 , with the same expansion velocity . an unidentified absorption at 9970 - 10045 can be clearly seen , which should have a rest wavelength between 10200 - 10260 . a few weak unidentified lines can be seen also in the @xmath21 band . although type ia sne display a large degree of spectroscopic homogeneity , there are several examples of sne with spectral peculiarities . different subclasses have been identified including sn 1991t - like objects , branch - normal sne , and sn 1991bg - like events @xcite . recently , the lick observatory supernova survey ( loss ) produced a sn sample with well - understood selection effects @xcite which permitted them to assess the intrinsic peculiarity rate of sne ia @xcite . this study showed that the sample of nearby sne ia comprises 20% , 64% , and 16% of 1991t - like , normal , and 1991bg - like objects , respectively . @xcite defined a normal sn as one with prominent features due to si ii @xmath06355 , ca ii h&k @xmath353934,3968 , as well as additional lines of s ii , o i , and mg ii around maximum light . the designation sn 1991 t in such work is not strictly correct since the prototype of the 1991 t class defined by @xcite is sn 1999aa . while sn 1999aa s main difference relative to normal sne is just a relative weakness of the si ii @xmath06355 feature before maximum that becomes almost indistinguishable from normal sne after peak ( see fig . 5 of @xcite ) , sn 1991 t was a much more extreme event in the sense that the pre - maximum spectrum did not display the ca ii h&k @xmath353934,3968 blend shown by branch - normal sne , and also because it remained spectroscopically distinct several days after maximum @xcite . for the sake of clarity , in what follows we refer to the 1991t - like events of li et al . as 1999aa - like objects . the distinguishing feature of sn 1991bg - like objects is the ti ii absorption around 4100 - 4000 . there are many examples of intermediate class objects , which suggests a continuous spectral sequence among sne ia . because 1999aa - like events display slow - declining lightcurves ( @xmath30m@xmath38(b)@xmath391.0 ) and high peak luminosities , and 1991bg - like sne show fast - decline rates ( @xmath30m@xmath38(b)@xmath401.7 ) and low peak luminosities , it is interesting to ask if the photometric behavior can be used to predict spectroscopic peculiarities . as a slow - declining ( @xmath30m@xmath38(b)=0.94 ) and luminous ( @xmath32=-19.94 ) event observed well before maximum light , sn 1999ee provides a good opportunity to investigate whether or not all luminous sne display spectroscopic peculiarities before peak . to examine this point in detail figure [ sn99ee.opt.comp.fig ] ( top ) shows a comparison of the spectrum of sn 1999ee taken 9 days before @xmath17 maximum to that of three other sne including sn 1999aa with a decline rate of @xmath30@xmath31=0.75 @xcite , the caln / tololo sn 1992bc with @xmath30@xmath31=0.87 @xcite , and the normal sn 1990n with @xmath30@xmath31=1.07 @xcite . while sn 1999aa lacked the si ii @xmath06355 feature compared to the branch - normal sn 1990n , both sn 1992bc and sn 1999ee displayed strong and well - defined si features as early as ten days before peak . this shows that _ not all slow - decliners are spectroscopically peculiar before maximum light_. as mentioned above , 1999aa - like events show a normal spectrum at peak brightness , which can be clearly appreciated in the bottom panel of figure [ sn99ee.opt.comp.fig ] from the spectra of sn 1999aa and sn 1999ee . the fact that the spectral differences displayed by the luminous sne are only limited to the very first days of evolution suggests that the 1999aa - like and the branch - normal sne ( represented here by sn 1999ee ) are apparently similar explosions and that spectral diversity at early times probably reflects small differences in mixing or other chaotic behaviors rather than fundamental differences in the character of the explosions . there are a handful of ir spectra published for sne ia @xcite . @xcite recently observed sn 1999by and assembled the first continuous sequence of ir spectra for a sn ia , extending from 3 days before maximum through day + 15 after peak . to our knowledge the ir spectra of sn 1999ee extending from day -9 to + 42 constitute the most complete ir spectroscopic sequence of a sn ia and its branch - normal nature makes it an ideal object in which to study the homogeneity of this class of objects at these wavelengths . we proceed now to address this issue from a sample of five sne ia encompassing a wide range in decline rates and optical spectroscopic properties . the sample comprises the branch - normal sn 1999ee , sn 1998bu @xcite , and sn 1994d @xcite , with @xmath30@xmath31=0.94,1.01,1.32 , respectively @xcite . we also include sn 1999by ( @xmath30@xmath31=1.90 ) which had optical spectra that showed resemblance to sn 1991bg @xcite , and sn 2000cx ( @xmath30@xmath31=0.93 ) which shared some of the spectroscopic peculiarities displayed by sn 1999aa - like events ( e.g. weak si ii @xmath06355 ) before maximum and a `` sui generis '' post - maximum behavior owing to unusually strong fe iii and weak fe ii lines @xcite . figure [ sn99ee.ir.comp3.fig ] ( top ) compares pre - maximum spectra of sn 2000cx , 1999ee , 1994d , and 1999by . the resemblance between the two branch - normal sne ( 1999ee and 1994d ) is remarkable . the largest , yet subtle , difference was the mg ii @xmath010926 feature which was narrower and deeper in sn 1994d . the expansion velocities derived from the absorption minimum was quite similar , @xmath310,500 km s@xmath4 . both the si ii and si iii features near 16500 and 20000 were clearly present on both sne . with the exception of the high mg ii velocity ( @xmath320,000 km s@xmath4 ) , sn 2000cx showed a normal featureless spectrum . since magnesium is destroyed by oxygen burning , it is expected in material which experiences burning at the lowest densities near the surface . the high mg velocity led @xcite to conclude that nuclear burning in sn 2000cx extended farther out than in normal sne , though it is equally likely that in this supernovae a blob containing carbon burning products drifted closer to the surface than in some other sne ia . the greater luminosity and slower decline of sn 2000cx argue for greater @xmath41ni production , and this would manifest itself also in greater excitation in the ejecta . indeed , sn 2000cx showed an emission feature at 12500 attributed to fe iii @xmath3512786,12920,13003 @xcite , which was also present in the normal sn 1999ee but at a lower strength , a weaker than normal si ii feature near 16500 , and a stronger si iii line around 20000 . sn 1999by showed a pre - maximum spectrum noticeably different than that of normal sne . the main difference was due to c i and o i lines , which were clearly absent in the spectra of other sne . @xcite suggested that nuclear burning in sn 1999by did not extend as far out as in normal sne ia . it is possible , however , that the presence of the c i and o i lines was due to a lower excitation evidenced by the smaller @xmath41ni production , the low luminosity and fast decline of sn 1999by . the spectral differences are particularly evident in the @xmath20 band . the mg ii @xmath010926 feature was deeper than normal , although its velocity of @xmath310,000 km s@xmath4 was similar to that observed in normal sne . in the @xmath21 band sn 1999by showed a strong broad emission feature due to a blend of si ii and mg ii lines which was also present in the other sne but with a lower strength , presumably due to a weaker contribution from mg ii . the @xmath22 band spectrum is also bumpier than normal due to mg ii lines . around maximum light the mg ii @xmath010926 feature was present in all three normal sne ( 1999ee , 1998bu , and 1994d ) with similar strengths and profiles ( bottom panel of figure [ sn99ee.ir.comp3.fig ] ) . @xcite noted that the absorption minimum of the mg ii line did not show a shift in wavelength during the pre - maximum evolution of sn 1994d , as opposed to the optical lines which all redshifted with time as the photosphere receded through the ejecta . we confirmed the absence of any significant doppler shift in mg ii from sn 1999ee which , according to @xcite was due to the fact that the photosphere had already receded below the inner edge of the magnesium layer at this early phase . it is interesting to note the absence in these spectra of the ca ii line near 11500 , a feature predicted by the models of @xcite that proves to be a useful diagnostic of the location of the transition layer between complete and incomplete silicon burning . the si ii absorption near 16500 and the spectral features in the @xmath22 band were visible in the normal sne 1999ee and 1998bu with a high degree of similarity . the pre - maximum spectral peculiarities of sn 1999by mentioned above were still evident near maximum light . figure [ sn99ee.ir.comp1.fig ] compares the post - maximum spectra available to us . two weeks after peak the two normal sne 1999ee and 1998bu displayed an impressive spectral homogeneity with strong peaks and valleys at 15000 - 17000 and 22000 - 26000 due to fe ii , co ii , ni ii , and si ii . the presence of fe ii , co ii , ni ii , and si ii is due to the ionization dropping predominantly to these ions , which happen to have strong lines in the near ir . the spectral differences of the fast - decliner sn 1999by persisted at this epoch . while its @xmath20 spectrum had contributions from ca ii lines @xcite that were absent in sne 1999ee and 1998bu , the broad emission around 15000 - 17000 was weaker than normal . at later epochs , the spectra of sne 1999ee and 1998bu became much more complicated , yet their similarity remained remarkable . as a result of this observational campaign we obtained the most complete optical / ir observations of a type ia sn , with an unprecedented wavelength and temporal coverage beginning nine days before maximum light . moreover , since we were able to obtain the ir spectra within one or two days from the optical spectra , it was possible to combine these observations , as shown in figure [ sn99ee.fig ] . this exercise revealed the excellent agreement between the optical and ir fluxes , a result that proved possible by synthesizing broad - band magnitudes and adjusting the flux scales according to the observed magnitudes . this optical / ir sequence shows that , while the pre - maximum optical spectrum was dominated by strong p - cygni profiles of intermediate mass elements like ca ii , si ii , mg ii , s ii , the ir was characterized by a smooth , almost featureless continuum . by maximum light , on the other hand , the ir spectrum was already dominated by broad features , and one week later the ir flux was mostly powered by emission lines of iron group elements like co ii , fe ii , and ni ii ( freshly synthesized in the innermost layers of the ejecta ) that were particularly prominent in the @xmath21 and @xmath22 bands . at the same time a dramatic flux dip around 12000 began to develop . previous spectroscopic observations have already revealed this @xmath20-band deficit in other sne ia @xcite , which is responsible for the red @xmath42 color displayed by sne ia after maximum light @xcite . this spectral sequence lends support to the suggestion by @xcite and @xcite that the `` photosphere '' recedes rapidly to the center of the supernova in the ir while at optical wavelengths the greater opacity arising from the higher spectral density of lines keeps the photosphere at higher velocities . emission at longer wavelengths increases after maximum light due to a shift in ionization to more - neutral species which have greater emissivity in the near ir . in this model , the @xmath20-band deficit is due to the relative absence of lines ( and hence opacity ) in the 12000 region rather than increasing opacity . likewise , the secondary maxima exhibited by the @xmath29 lightcurves @xmath330 days after @xmath17 maximum @xcite are due to the increasing release of energy through lower - optical depth ir transitions . the prominent post - maximum emission features displayed by sn 1999ee in the @xmath21 and @xmath22 bands lend support to this picture . sne are classified according to their spectral features near maximum light @xcite . type ii sne are characterized by prominent hydrogen features and are believed to arise from the core collapse of massive ( @xmath43 8 - 10 m@xmath44 ) stars . the common feature in all type i sne is the lack of hydrogen spectral lines . the strong si ii @xmath06355 is the defining feature of sne ia . their occurrence in all type of stellar environments has led to the general consensus that the progenitors of sne ia are white dwarfs that undergo thermonuclear disruption after a period of mass transfer from binary companions . the proximity of sne ib and ic to massive star formation regions @xcite , on the other hand , provides evidence that these objects result from the core collapse of massive stars . the lack of hydrogen in their spectra is attributed to the loss of their outer hydrogen envelopes by stellar winds or mass transfer to binary companions . there is growing evidence for a close physical connection between sne ii and sne ib from the observations of sne 1987k , 1993j , and 1996cb , which evolved from sne ii at early epochs to sne ib at later times @xcite . the distinguishing difference between type ib and ic sne are the _ optical _ he i lines , which are conspicuous in sne ib and weak or absent in sne ic . it is still a matter of debate whether sne ic constitute an extreme case of sne ib or a completely separate class of objects . the apparent absence of he in the spectra of sne ic motivated the idea that the progenitors of these objects could be nearly bare c+o cores of massive stars @xcite that loose most of their helium by binary transfer or strong stellar winds . however , @xcite showed that helium stars could be the progenitors of both sne ib and sne ic . they showed that the presence or absence of he i lines in the spectrum is determined primarily by the amount of mixing , not the amount of helium present greater mixing of radioactive material increases the excitation of helium leading to stronger lines . significant effort has been put over recent years into determining the presence of helium in the spectra of sne ic in order to better understand the nature of these objects . a detailed inspection of the spectra of the type ic sn 1994i led @xcite to conclude that weak he i lines were probably present in the optical region and that he i @xmath010830 was very prominent , although its doppler shift implied an unusually high expansion velocity @xmath316,600 km s@xmath4 . @xcite confirmed these observations and found evidence that he i @xmath05876 was also present in sn 1994i with a blueshift of @xmath317,800 km s@xmath4 . they also reported high velocity he i @xmath05876 in the spectra of the three best - observed type ic sne ( 1983v , 1987 m , and 1988l ) , which led them to conclude that most , and probably all , sne ic show evidence for optical he i lines . this conclusion , however , was recently questioned by @xcite and @xcite by means of spectral synthesis models which showed that the ir feature near 10250 could be accounted with lines of c i and si i. moreover , @xcite argued that the feature attributed to he i @xmath05876 in the spectrum of sn 1994i could be a blend of other species . evidently , there is no consensus yet about the presence or absence of he in the spectra of sne ic . recently @xcite compiled and analyzed a large collection of spectra of sne ib and ic . this study showed no compelling evidence for he in the spectra of sne ic and no gradual transition from the ib to the ic class , which supported the idea that these objects are produced by different progenitors . our optical spectra of sn 1999ex show evidence for he i absorptions of moderate strength in the optical region , thus suggesting the existence of an intermediate ib / c case . to illustrate this point in figures [ sn99ex.opt.comp1.fig ] and [ sn99ex.opt.comp2.fig ] we show a comparison between sn 1999ex , the best - observed type ic sn 1994i , and the prototypes of the ib ( sn 1984l ) and ic class ( sn 1987 m ) . the near - maximum spectra ( top panel of figure [ sn99ex.opt.comp1.fig ] ) reveal a gradual increase in the strength of all he lines ( indicated with tick marks ) from the ic to the ib class . in the bottom panel of figure [ sn99ex.opt.comp1.fig ] ( one week past maximum ) it is possible to see that , even though the spectra of the type ic sne 1994i and 1987 m are quite similar , sn 1994i shows deeper troughs at the wavelengths of the he i lines , especially at 4471 , 4921 , and 5876 . the spectrum of sn 1994i obtained two weeks past maximum ( figure [ sn99ex.opt.comp2.fig ] ) is particularly interesting as it provides good evidence that the he i @xmath05876 line was indeed present in the spectrum of sn 1994i with a higher expansion velocity than the na i d @xmath05893 doublet . this was also noted by @xcite who quoted an expansion velocity of @xmath316,600 km s@xmath4 for the he line ( see also @xcite for a detailed discussion ) . overall , figures [ sn99ex.opt.comp1.fig]-[sn99ex.opt.comp2.fig ] reveal that there is a smooth spectroscopic sequence ranging from the he deficient type ic sne 1987 m and 1994i , the type ib / c sn 1999ex , and the type ib sn 1984l that is characterized by strong optical he lines . we conclude therefore that sn 1999ex provides first evidence for a link between the ib and ic classes and that there is a continuous sequence of sne ib and ic objects . special attention must be paid to the ir feature near 11,000 which was very prominent in sne 1999ex and 1994i . as mentioned above the definitive identity of this feature in sn 1994i is still uncertain : it could be accounted with lines of c i and si i @xcite or with high velocity he i @xmath010830 @xcite . if the feature in sn 1999ex was he i @xmath010830 it would imply a moderate velocity of 6,000 - 8,000 km s@xmath4 , which matches very well the velocities derived from the fe , na and si lines . the presence of the he i @xmath020581 with the same expansion velocity ( see figure [ sn99ex.ir.fig ] ) suggests that the ir feature in sn 1999ex was indeed due to he i @xmath010830 . altogether our ir spectra of sn 1999ex provide unambiguous proof that he was present in the atmosphere of this intermediate ib / c object . a detailed atmosphere model could be very useful at placing limits on the he mass in the ejecta of sn 1999ex and constraining the nature of its progenitor . as suggested by @xcite the presence of highly blueshifted he i @xmath05876 in sn 1994i suggests that the ir feature was also due to he i @xmath010830 . if so , the difference in he velocities between sne 1994i and 1999ex might prove an interesting clue to the underlying physics of the atmospheres of this class of objects . further optical / ir spectroscopy and photometry will lead us to better understanding the atmospheres of sne ib and ic . finally , given that the ir spectra were obtained within one day from the optical spectra , we were able to combine these observations . figure [ sn99ex.fig ] shows the resulting spectra and the excellent agreement between the optical and ir fluxes . note the prominent he i @xmath0@xmath010830,20581 features . in the course of the soirs program we obtained high - quality spectroscopy of the type ii sn 1999em @xcite . figure [ allsne.fig ] compares maximum - light spectra of the type ii sn 1999em , the type ib / c sn 1999ex , and the type ia sn 1999ee . this figure permits one to compare the different characters of the main classes of sne , all the way from 3,000 to 25,000 . the type ii is distinguished by prominent hydrogen balmer and paschen lines . it also shows the p - cygni profile of the he i @xmath010830 transition . the type ib / c is characterized by the strong he i @xmath3510830,20581 lines in the ir , the he i @xmath05876 line in the optical , and other lines in the optical due to singly ionized ca , fe , mg , si and neutral na and o. it is interesting to note that , while both the type ii and the type ib / c show strong he i @xmath010830 , only the type ib / c shows a significant he i @xmath020581 feature . finally , the type ia at the bottom of this figure is characterized by the strong si ii @xmath06355 and other intermediate - mass elements , and the absence of hydrogen and helium . the observations obtained for sn 1999ee constitute the most complete spectral and temporal coverage ever achieved for a sn ia . its branch - normal character makes it an ideal reference for comparative studies of sne ia . before maximum light sn 1999ee displayed a normal spectrum with a strong si ii @xmath06355 absorption , thus showing that not all slow - declining sne are spectroscopically peculiar at these evolutionary phases . we conclude that the photometric properties of luminous sne ia can not be used to predict spectroscopic peculiarities . from a comparison of the ir spectra of sn 1999ee and other sne ia that encompass a wide range in decline rates we find that there is a remarkable homogeneity among the branch - normal sne ia during their first 60 days of evolution . although the slow - decliner luminous sn 2000cx showed a premaximum featureless ir spectrum similar to that of normal sne , the mg ii @xmath45 line was characterized by a high expansion velocity . the fast - decliner subluminous sn 1999by was noticeably different than the other sne at all epochs . this study reveals that the spectroscopic peculiarities displayed by sn 1991bg - like objects at optical wavelengths are also present in the ir . the fortunate occurrence of sn 1999ex within three weeks and in the same galaxy that hosted sn 1999ee permitted us to obtain optical and ir spectroscopy of a ib / c event . sn 1999ex was characterized for the lack of hydrogen lines , weak optical he i lines , and strong he i @xmath010830,20581 , thus providing an example of an intermediate case between pure ib and ic sne . we conclude therefore that sn 1999ex provides first clear evidence for a link between the ib and ic classes and that there is a continuous spectroscopic sequence ranging from the he deficient sne ic to the sne ib which are characterized by strong optical he i lines . m. h. is very grateful to las campanas and cerro caln observatories for allocating an office and providing generous operational support to the soirs program during 1999 - 2000 . m. h. and j. m. thank the eso , ctio , and las campanas visitor support staffs for their assistance in the course of our observing runs , and the director general of eso for allocating director s discretionary telescope time to this project . we are very grateful to a. filippenko , p. hflich , s. jha , d. leonard , w. li , p. meikle , r. rudy , and j. spyromilio , for making us available their spectra of sne 1984l , 1987 m , 1994d , 1994i , 1999by , 1998bu , and 2000cx . support for this work was provided by nasa through hubble fellowship grant hst - hf-01139.01-a awarded by the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , for nasa , under contract nas 5 - 26555 . j.m . acknowledges support from fondecyt grant 1980172 . this research has made use of the nasa / ipac extragalactic database ( ned ) , which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration . this research has made use of the simbad database , operated at cds , strasbourg , france . axelrod , t. s. 1980 , ph . d. thesis , univ . california at santa cruz baron , e. , branch , d. , hauschildt , p. h. , filippenko , a. v. , & kirshner , r. p. 1999 , , 527 , 739 bowers , e. j. c. , meikle , w. p. s. , geballe , t. r. , walton , n. a. , pinto , p. a. , dhillon , v. s. , howell , s. b. , & harrop - allin , m. k. 1997 , , 290 , 663 branch , d. , fisher , a. , & nugent , p. 1993 , , 106 , 2383 clocchiatti , a. , wheeler , j. c. , brotherton , m. s. , cochran , a. l. , wills , d. , barker , e. s. , & turatto , m. 1996 , , 462 , 462 elias , j. h , frogel , j. a. , hackwell , j. a. , & persson , s. e. 1981 , , 251 , l13 elias , j. h , matthews , k. , neugebauer , g. , & persson , s. e. 1985 , , 296 , 379 filippenko , a. v. 1988 , , 96 , 1941 filippenko , a. v. , matheson , t. , & ho , l. c. 1993 , , 415 , l103 filippenko , a. v. , et al . 1995 , , 450 , l11 filippenko , a. v. 1997 , , 35 , 309 garnavich , p. m. , et al . 2001 , , submitted ( astro - ph/0105490 ) gray , r. o. , & corbally , c. j. 1994 , , 107 , 742 hamuy , m. , suntzeff , n. b. , heathcote , s. r. , walker , a. r. , gigoux , p. , & phillips , m. m. 1994 , , 106 , 566 hamuy , m. , et al . 1996 , , 112 , 2408 hamuy , m. , & phillips , m. m. 1999 , iauc 7310 hamuy , m. , et al . 2001 , , 558 , 615 hernndez , m. , et al . 2000 , , 319 , 223 hflich , p. , khokhlov , a. m. , & wheeler , j. c. 1995 , , 444 , 831 hflich , p. , gerardy , c. l. , fesen , r. a. , & sakai , s. 2002 , , in press ( astro - ph/0112126 ) jha , s. , et al . 1999 , , 125 , 73 krisciunas , k. , hastings , n. c. , loomis , k. , mcmillan , r. , rest , a. , riess , a. g. , & stubbs , c. 2000 , , 539 , 658 krisciunas , k. , et al . 2002 , in preparation leibundgut , b. , kirshner , r. p. , filippenko , a. v. , shields , j. , c. , foltz , c. b. , phillips , m. m. , & sonneborn , g. 1991 , , 371 , l23 li , w. , filippenko , a. v. , & riess , a. g. 2001a , , 546 , 719 li , w. , filippenko , a. v. , treffers , r. r. , riess , a. g. , hu , j. , & qiu , y. l. 2001b , , 546 , 734 li , w. , et al . 2001c , , 113 , 1178 lira , p. , et al . 1998 , , 115 , 234 maiolino , r. , rieke , g. h. , & rieke , m. j. 1996 , , 111 , 537 matheson , t. , filippenko , a. v. , li , w. , leonard , d. c. , & schields , j. c. 2001 , , 121 , 1648 maza , j. , & hamuy , m. , iauc 7272 meikle , w. p. s , et al . 1996 , , 281 , 263 millard , j. , et al . 1999 , , 527 , 746 moorwood , a. f. 1997 , in proc . spie , vol . 2871 , 1146 munari , u. , & zwitter , t. 1997 , , 318 , 269 phillips , m. m , wells , l. a. , suntzeff , n. b. , hamuy , m. , leibundgut , b. , kirshner , r. p. , & foltz , c. b. 1992 , , 103 , 1632 phillips , m. m , lira , p. , suntzeff , n. b. , schommer , r. a. , hamuy , m. , & maza , j. 1999 , , 118 , 1766 pinto , p. a. , & eastman , r. g. 2000 , , 530 , 757 qiu , y. , li . w. , qiao , q. , & hu , j. 1999 , , 117 , 736 rudy , r. j. , lynch , d. k. , mazuk , s. , venturini , c. c. , puetter , r. c. , & hflich , p. 2002 , , 565 , 413 schlegel , d. j. , finkbeiner , d. p. , & davis , m. 1998 , , 500 , 525 spyromilio , j. , pinto , p. a. , & eastman , r. g. 1994 , , 266 , l17 stritzinger , m. , et al . 2002 , in preparation van dyk , s. d. , hamuy , m. , & filippenko , a. v. 1996 , , 111 , 2017 wheeler , j. c. , & harkness , r. p. 1990 53 , 1467 wheeler , j. c. , hflich , p. , harkness , r. p. , & spyromilio , j. 1998 , , 496 , 908 woosley , s. e. , langer , n. , & weaver , t. a. 1995 , , 448 , 315 1999 oct 09 & 460.53 & paranal & vlt / antu & isaac & rockwell1k & & 5,4,3,2 & 0.98 - 2.50 & 2.9,3.6,4.7,7.1 & clear & hamuy , lidman + 1999 oct 09 & 460.56 & tololo & 4-m & r - c spec . & loral3kx1k & kpgl-2 & 1 & 0.33 - 0.87 & 1.9 & cirrus & maza + 1999 oct 11 & 462.52 & la silla & ntt & emmi & tek1024 + 2048 & 5,13 & 1 & 0.33 - 1.01 & 2.7,3.5 & clear & hamuy , doublier + 1999 oct 16 & 467.56 & campanas & 2.5-m & wfccd & tek2048 & blue & & 0.36 - 0.92 & 3.0 & clouds ? & phillips + 1999 oct 18 & 469.49 & la silla & ntt & sofi & rockwell1k & red & & 1.50 - 2.53 & 10.2 & clear & doublier , maza + 1999 oct 18 & 469.60 & la silla & ntt & emmi & tek1024 + 2048 & 5,13 & 1 & 0.33 - 1.00 & 2.7,3.5 & clear & maza + 1999 oct 19 & 470.50 & paranal & vlt / antu & isaac & rockwell1k & & 5,4,3,2 & 0.98 - 2.50 & 2.9,3.6,4.7,7.1 & clear & hamuy , cuby , petr + 1999 oct 20 & 471.56 & la silla & d1.5-m & dfosc & loral2kx2k & 3,5 & 1 & 0.33 - 0.98 & 2.3,3.1 & & pompei + 1999 oct 22 & 473.61 & la silla & ntt & sofi & rockwell1k & blue , red & & 0.95 - 2.47 & 7.0,10.2 & clear & hainaut , lefloch + 1999 oct 25 & 476.51 & la silla & d1.5-m & dfosc & loral2kx2k & 3,5 & 1 & 0.33 - 0.98 & 2.3,3.1 & clear & augusteyn + 1999 oct 26 & 477.48 & la silla & ntt & sofi & rockwell1k & blue , red & & 0.95 - 2.46 & 7.0,10.2 & & hainaut , vanzi + 1999 oct 27 & 478.52 & tololo & 1.5-m & r - c spec . & loral1.2kx0.8k & 58 & 2 & 0.37 - 0.50 & 1.1 & clear & olsen + 1999 oct 29 & 480.51 & la silla & d1.5-m & dfosc & loral2kx2k & 3,5 & 1 & 0.34 - 0.98 & 2.3,3.1 & & augusteyn + 1999 nov 02 & 484.49 & paranal & vlt / antu & isaac & rockwell1k & & 5,4,3,2 & 0.98 - 2.50 & 2.9,3.6,4.7,7.1 & clear & hamuy , lidman , petr + 1999 nov 03 & 485.62 & la silla & ntt & emmi & tek1024 + 2048 & 5,13 & 1 & 0.33 - 1.01 & 2.7,3.5 & & maza + 1999 nov 06 & 488.52 & la silla & ntt & sofi & rockwell1k & blue & & 0.94 - 1.65 & 7.0 & clear & brillant + 1999 nov 06 & 488.53 & la silla & d1.5-m & dfosc & loral2kx2k & 3,5 & 1 & 0.34 - 0.98 & 2.3,3.1 & clear ? & augusteyn + 1999 nov 09 & 491.50 & la silla & ntt & sofi & rockwell1k & blue , red & & 0.95 - 2.51 & 7.0,10.2 & clear & hamuy , brillant + 1999 nov 09 & 491.62 & la silla & ntt & emmi & tek1024 + 2048 & 5,13 & 1 & 0.34 - 1.00 & 2.7,3.5 & clear & hamuy , brillant + 1999 nov 14 & 496.56 & la silla & ntt & sofi & rockwell1k & blue , red & & 0.95 - 2.46 & 7.0,10.2 & clear & hamuy , doublier + 1999 nov 14 & 496.62 & la silla & ntt & emmi & tek1024 + 2048 & 5,13 & 1 & 0.34 - 1.00 & 2.7,3.5 & clear & hamuy , doublier + 1999 nov 18 & 500.53 & paranal & vlt / antu & isaac & rockwell1k & & 5,4,3,2 & 0.98 - 2.50 & 2.9,3.6,4.7,7.1 & clear & serv . + 1999 nov 19 & 501.58 & la silla & ntt & emmi & tek1024 + 2048 & 5,13 & 1 & 0.34 - 1.00 & 2.7,3.5 & clear & hamuy , doublier + 1999 nov 28 & 510.52 & paranal & vlt / antu & isaac & rockwell1k & & 5,4,3,2 & 0.98 - 2.50 & 2.9,3.6,4.7,7.1 & clear & hamuy , lidman , chadid + 1999 nov 28 & 510.63 & la silla & d1.5-m & dfosc & loral2kx2k & 3,5 & 1 & 0.40 - 0.98 & 2.3,3.1 & & pinfield +

a key challenge in systems biology is the elucidation of the underlying principles , or fundamental laws , which determine the cellular phenotype . understanding how these fundamental principles are altered in diseases like cancer is important for translating basic scientific knowledge into clinical advances . while significant progress is being made , with the identification of novel drug targets and treatments by means of systems biological methods , our fundamental systems level understanding of why certain treatments succeed and others fail is still lacking . we here advocate a novel methodological framework for systems analysis and interpretation of molecular omic data , which is based on statistical mechanical principles . specifically , we propose the notion of cellular signalling entropy ( or uncertainty ) , as a novel means of analysing and interpreting omic data , and more fundamentally , as a means of elucidating systems - level principles underlying basic biology and disease . we describe the power of signalling entropy to discriminate cells according to differentiation potential and cancer status . we further argue the case for an empirical cellular entropy - robustness correlation theorem and demonstrate its existence in cancer cell line drug sensitivity data . specifically , we find that high signalling entropy correlates with drug resistance and further describe how entropy could be used to identify the achilles heels of cancer cells . in summary , signalling entropy is a deep and powerful concept , based on rigorous statistical mechanical principles , which , with improved data quality and coverage , will allow a much deeper understanding of the systems biological principles underlying normal and disease physiology . entropy ; network ; signalling ; genomics ; drug resistance ; cancer ; differentiation ; stem cell

1408.1002

recent advances in biotechnology are allowing us to measure cellular properties at an unprecedented level of detail @xcite . for instance , it is now possible to routinely measure various molecular entities ( e.g. dna methylation , mrna and protein expression , snps ) genome - wide in hundreds if not thousands of cellular specimens @xcite . in addition , other molecular data detailing interactions between proteins or between transcription factors and regulatory dna elements are growing at a rapid pace @xcite . all these types of data are now widely referred to collectively as `` omic '' data . the complexity and high - dimensional nature of this omic data presents a daunting challenge to those wishing to analyse and interpret the data @xcite . the difficulty of analysing omic data is further compounded by the inherent complexity of cellular systems . cells are prime examples of organized complex systems , capable of highly stable and predictable behaviour , yet an understanding of how this deterministic behaviour emerges from what is a highly complex and probabilistic pattern of dynamic interactions between numerous intra and extracellular components , still eludes us @xcite . thus , elucidating the systems - biological laws or principles dictating cellular phenotypes is also key for an improved analysis and interpretation of omic data . furthermore , important biological phenomena such as cellular differentiation are fundamentally altered in diseases like cancer @xcite . hence , an attempt to understand how cellular properties emerge at a systems level from the properties seen at the individual gene level is not only a key endeavour for the systems biology community , but also for those wanting to translate basic insights into effective medical advances @xcite . + it is now well accepted that at a fundamental level most biological systems are best modeled in terms of spatial interactions between specific entities ( e.g. neurons in the case of the brain ) , which may or may not be dynamically changing in time @xcite . it therefore seems natural to also use the mathematical and physical framework of networks to help us analyse and interpret omic data at a systems level @xcite . indeed , the cellular phenotype is determined to a large extent by the precise pattern of molecular interactions taking place in the cell , i.e. a molecular interaction network @xcite . although this network has spatial and dynamic dimensions which at present remain largely unexplored due to technological or logistical limitations , there is already a growing number of examples where network - based analysis strategies have been instrumental @xcite . for instance , a deeper understanding of why sustained treatment with egfr inhibitors can lead to dramatic sensitization of cancer cell lines to cytotoxic agents was possible thanks to a systems approach @xcite . another study used reverse engineering network approaches to identify and validate drug targets in glioblastoma multiforme , to be further tested in clinical trials @xcite . what is key to appreciate here is that these successes have been achieved in spite of noisy and incomplete data , suggesting that there are simple yet deep systems biological principles underlying cellular biology that we can already probe and exploit with current technology and data . thus , with future improvements in data quality and coverage , network - based analysis frameworks will play an ever increasing and important role in systems biology , specially at the level of systems analysis and interpretation @xcite . therefore , it is also imperative to develop novel , more powerful , network - theoretical methods for the systems analysis of omic data . + in adopting a network s perspective for the analysis and interpretation of omic data , there are in principle two different ( but not mutually exclusive ) approaches one can take . one possibility is to infer ( i.e. reverse engineer ) the networks from genome - wide data @xcite . most of these applications have done this in the context of gene expression data , with the earliest approaches using clustering or co - expression to postulate gene interdependencies @xcite . partial correlations and graphical gaussian models have proved useful as a means of further refining correlation networks by allowing one to infer the more likely direct interactions while simultaneously also filtering out those which are more likely to be indirect @xcite . these methods remain popular and continue to be studied and improved upon @xcite . other methods have drawn on advanced concepts from information theory , for instance aracne ( `` algorithm for the reconstruction of accurate cellular networks '' ) has been shown to be successful in infering regulatory networks in b - cells @xcite . + in stark contrast to reverse engineering methods , another class of algorithms have used structural biological networks from the outset , using these as scaffolds to integrate with omic data . specifically , by using a structural network one can sparsify the correlation networks inferred from reverse - engineering approaches , thus providing another means of filtering out correlations that are more likely to be indirect @xcite . besides , integration with a structural network automatically provides an improved framework for biological interpretation @xcite . the structural networks themselves are typically derived from large databases , which detail literature curated experimentally verified interactions , including interactions derived from yeast 2 hybrid screens ( y2h ) @xcite . the main example is that of protein protein interaction ( ppi ) maps , which have been generated using a number of different complementary experimental and in - silico approaches , and merging these maps from these different sources together has been shown to be an effective approach in generating more comprehensive high - confidence interaction networks @xcite . ppi networks have been used mainly as a means of integrating and analysing gene expression data ( see e.g. @xcite ) . more recently , this approach has also been successfully applied in the dna methylation context , for instance it has been shown that epigenetic changes associated with age often target specific gene modules and signalling pathways @xcite . + another class of methods that have used structural networks , ppis in particular , have integrated them with gene expression data to define an approximation to the underlying signaling dynamics on the network , thus allowing more in - depth exploration of the interplay between network topology and gene expression . typically , these studies have used the notion of random walks on weighted graphs where the weights are constructed from differential expression statistics , and where the aim is to identify nodes ( genes ) in the network which may be important in dictating the signaling flows within the pathological state . for instance , among these random walk methods is netrank , a modification of the google pagerank algorithm , which was able to identify novel , robust , network based biomarkers for survival time in various cancers @xcite . other random walk based approaches , aimed at identifying causal drivers of specific phenotypes ( e.g. expression or cancer ) , have modeled signal transduction between genes in the causal and phenotypic layers as flows in an electric circuit diagram , an elegant formulation capable of not only identifying the likely causal genes but also of tracing the key pathways of information flow or dysregulation @xcite . random walk theory has also been employed in the development of differential network methodologies . an example is netwalk @xcite , which is similar to netrank but allows differential signaling fluxes to be inferred . this approach was successful in identifying and validating the glucose metabolic pathway as a key determinant of lapatinib resistance in erbb2 positive breast cancer patients @xcite . + another important concept to have emerged recently is that of network rewiring @xcite . this refers to the changes in interaction patterns that accompany changes in the cellular phenotype . network rewiring embodies the concept that it is the changes in the interaction patterns , and not just the changes in absolute gene expression or protein activity , that are the main determinants of the cellular phenotype . that network rewiring may be key to understanding cellular phenotypes was most convincingly demonstrated in a differential epistasis mapping study conducted in yeast cells exposed to a dna damaging agent @xcite . specifically , what this study demonstrated is that responses to perturbations or cellular stresses are best understood in terms of the specific rewiring of protein complexes and functional modules . thus , this conceptual framework of network rewiring may apply equally well to the genetic perturbations and cellular stresses underlying disease pathologies like cancer . + in this article we advocate a network - theoretical framework based on statistical mechanical principles and more specifically on the notion of signalling entropy @xcite . this theoretical framework integrates gene expression ( but in principle also other functional data ) with a ppi network , merging existing concepts such as signaling dynamics ( i.e. random walks ) and network rewiring with that of signalling entropy . in previous work we have shown how signalling entropy ( i ) provides a proxy to the elevation in waddington s epigenetic landscape , correlating with a cell s differentiation potential @xcite , ( ii ) how it can be used to identify signaling pathways and nodes important in differentiation and cancer @xcite , and ( iii ) how it predicts two cancer system - omic hallmarks : ( a ) cancer is characterised by an increase in signalling entropy and ( b ) local signaling entropy changes anti - correlate with differential gene expression @xcite . here , we present and unify the different signaling entropy measures used previously and further explore a novel application of signalling entropy to understanding drug sensitivity profiles in cancer cell lines . specifically , we first use simulated data to justify the existence of an entropy - robustness theorem , and subsequently provide empirical evidence for this theorem by demonstrating that increases in local signalling entropy correlate with drug resistance ( robustness ) . we further show the importance of network topology in dictating the signalling entropy changes underlying drug response . in addition , we provide r - functions implementing the entropy rate calculation and ranking of genes according to differential entropy , all freely available from _ sourceforge.net/projects/signalentropy/files/ _ . loosely defined , entropy of a system , refers to a measure of the disorder , randomness or uncertainty of processes underlying the system s state . in the context of a single cell , signalling entropy will reflect the amount of overall disorder , randomness or uncertainty in how information , i.e. signaling , is passed on in the molecular interaction network . at a fundamental level , all signaling in a cell is probabilistic , determined in part by the relative cellular concentrations of the interacting molecules . hence , in discussing cellular signalling entropy , it is useful to picture a molecular interaction network in which edges represent possible interactions and with the edge weights reflecting the relative probabilities of interaction ( * fig.1a * ) . thus , an interaction network in a high - entropy state is characterised by signaling interaction probabilities that are all fairly similar in value , whereas a low - entropy state will be characterised by specific signalling interactions possessing much higher weights ( * fig.1a * ) . why would this type of signalling entropy , loosely defined as the amount of uncertainty in the signaling interaction patterns , be useful to systems biology ? one way to think of signalling entropy is as representing signaling promiscuity , which has been proposed as a key systems feature underlying the pluripotent or multipotent capacity of cells ( * fig.1b * ) @xcite . indeed , it has been suggested that pluripotency is an intrinsic statistical mechanical property , best defined at the cellular population level @xcite . specifically , it has been demonstrated that pluripotent stem cells exhibit remarkable heterogeneity in gene expression levels , including well known stem cell markers such as _ nanog _ @xcite . it is also well known that a large number of genes , many encoding transcription factors , exhibit low - levels of expression in stem cells , yet simultaneously are being kept in a poised chromatin state , allowing immediate activation if this were required @xcite . thus , in a pluripotent stem cell like state , signal transduction is in a highly egalitarian and , thus , promiscuous state , i.e. a state of high signalling entropy . conversely , differentiation leads , by necessity , to activation of specific transcription factors and pathways and thus to a lowering in the uncertainty of signaling patterns , and thus to a lowering of entropy . we recently demonstrated , using gene expression profiles of over 800 samples , comprising cells of all major stages of differentiation , including human embryonic stem cells ( hescs ) , induced pluripotent stem cells ( ipscs ) , multipotent cell types ( e.g. hematopoietic stem cells ( hscs ) ) , and terminally differentiated cells within these respective lineages , that signalling entropy not only correlates with differentiation potential but that it provides a highly quantitative measure of potency @xcite . indeed , we showed that signalling entropy provides a reasonably good approximation to the energy potential in waddington s epigenetic landscape @xcite . + here we decided to explore the concept of signalling entropy in relation to cellular robustness and specifically to drug resistance in cancer . that signalling entropy may be informative of a cell s robustness is a proposal that stems from a general ( but unproven ) theorem , first proposed by manke and demetrius @xcite : namely , that a system s entropy and robustness are correlated . mathematically , this can be expressed as @xmath0 , which states that a positive change in a system s entropy ( i.e. @xmath1 ) must lead to an increase in robustness ( @xmath2 ) . now , cells are fairly robust entities , having evolved the capacity to buffer the intra - and - extracellular stresses and noise which they are constantly exposed to @xcite . much of this overall stability and robustness likely stems from the topological features of the underlying signaling and regulatory networks , for instance features such as scale - freeness and hierarchical modularity , which are thought to have emerged through natural evolutionary processes such as gene duplication @xcite . however , another key feature which contributes to cellular robustness is cross - talk and signalling pathway redundancy @xcite . pathway redundancy refers to a situation where a cell has the choice of transmitting signals via two or more possible routes . in the language of statistical mechanics , this corresponds to a state of high uncertainty ( or entropy ) in signaling . high signalling entropy could thus underpin a cell s robustness to perturbations , suggesting that a cell s entropy and robustness may indeed be correlated ( * fig.2a * ) . consistent with this , pathway redundancy is also well recognized to be a key feature underlying drug resistance of cancer cells @xcite . ) , then the system s robustness @xmath3 must also decrease , i.e. @xmath4 , so that the product @xmath0 . observe how in the low entropy state , random removal of edges through e.g. inactivating mutations , can lead to deactivation of a key signaling pathway connecting a given start and end nodes ( shown in orange ) . in the high entropy state , the same perturbations do not prevent signal transduction between the two orange nodes . * b ) * depicted are the effects of two major forms of cancer perturbation . in the upper panel , inactivation ( typically of tumour suppressors ) , leads to underexpression and a corresponding loss of correlations / interactions with neighbors in the ppi network . this is tantamount to a state of increased entropy and drug intervention is unlikely to be effective . in the lower panel , we depict the case of an oncogene , which is overexpressed in cancer . this overexpression leads to activation of a specific oncogenic pathway which results in oncogene addiction and increased sensitivity to targeted drug intervention . thus , local signalling entropy and robustness ( as determined by response to a drug ) , may also correlate locally . ] further supporting the notion that entropy and robustness may be correlated , we previously showed that ( i ) cancer is characterised by a global increase in signalling entropy compared to its respective normal tissue @xcite , in line with the observation that cancer cells are specially robust to generic perturbations , and ( ii ) that a gene s differential entropy between normal and cancer tissue is generally speaking anticorrelated with its differential expression @xcite , consistent with the view that cancer cells are specially sensitive to drug interventions that target overexpressed oncogenes , a phenomenon known as oncogene addiction ( * fig.2b * ) @xcite . interestingly , the observations that differential entropy and differential expression are anti - correlated and that cancer is characterised globally by an increase in entropy @xcite , are also consistent with the prevalent view that most driver mutations are indeed inactivating , targeting tumor suppressor genes @xcite . hence , based on all of these observations and insights , we posited that signalling entropy could also prove useful as a means of predicting drug resistance in cancer cells . briefly , we review the definitions and construction of signalling entropy as used in our previous studies @xcite . the construction relies on a comprehensive and high - confidence ppi network which is integrated with gene expression data @xcite ( see * appendix a * ) . briefly , the ppi is used as a scaffold , and edge weights are constructed from the gene expression data to approximate the interaction or signaling probabilities between the corresponding proteins in the ppi . thus , the gene expression data provides the biological context in which to modulate the ppi interaction probabilities . to compute signalling entropy requires the estimation of a stochastic matrix , reflecting these interaction probablities over the network . + the construction of the stochastic matrix can proceed in two different ways . in the earliest studies we used a construction which was done at the level of phenotypes @xcite . under this model , edge weights @xmath5 between proteins @xmath6 and @xmath7 were constructed from the correlation between the expression levels of the corresponding genes @xmath6 and @xmath7 , as assessed over independent samples all representing the same phenotype . estimating the correlations over independent samples , all within the same phenotype , can be viewed as representing a multifactorial perturbation experiment , with e.g. genetic differences between individuals mimicking specific perturbations , and thus allowing influences to be inferred . thus , this approach hinges on the assumption that highly correlated genes , whose coding proteins interact , are more likely to be interacting in the given phenotype than two genes which do not correlate . the use of a ppi as a scaffold is important to filter out significant correlations which only result from indirect influences . the correlations themselves can be defined in many different ways , for instance , using simple pearson correlations or non - linear measures such as mutual information . for example , one way to define the weights is as @xmath8 with @xmath9 describing the pearson correlation coefficient between genes @xmath6 and @xmath7 . this definition guarantees positivity , but also treats positive and negative correlations differently , which makes biological sense because activating and inhibitory interactions normally have completely distinct consequences on downstream signalling . thus , the above weight scheme treats zero or insignificant correlations as intermediate , consistent with the view that an absent interaction is neither inhibitory nor activating . however , other choices of weights are possible : e.g. @xmath10 , which treats negative and positive correlations on an equal footing . once edge weights are defined as above , these are then normalised to define the stochastic matrix @xmath11 over the network , @xmath12 with @xmath13 denoting the ppi neighbors of gene @xmath6 . thus , @xmath11 is the probability of interaction between genes @xmath6 and @xmath7 , and as required , @xmath14 . however , there is no requirement for @xmath11 to be doubly stochastic , i.e. @xmath15 is in general not a symmetric matrix . hence , edges are bi - directional with different weights labeling the probability of signal transduction from @xmath6 to @xmath7 , and that from @xmath7 to @xmath6 ( @xmath16 ) . + an alternative to the above construction of the stochastic matrix is to invoke the mass action principle , i.e. one now assumes that the probability of interaction in a given sample is proportional to the product of expression values of the corresponding genes in that sample @xcite . thus , the ppi is again used as a scaffold to only allow interactions supported by the ppi network , but the weights are defined using the mass action principle , as @xmath17 where @xmath18 denotes the normalised expression intensity value ( or normalised rna - seq read count ) of gene @xmath6 . an important advantage of this construction is that the stochastic matrix is now sample specific , as the expression values are unique to each sample . + given a stochastic matrix , @xmath11 , constructed using one of the two methods above , one can now define a local shannon entropy for each gene @xmath6 as @xmath19 where @xmath20 denotes the degree of gene @xmath6 in the ppi network . the normalisation is optional but ensures that this local shannon entropy is normalised between 0 and 1 . clearly , if only one weight is non - zero , then the entropy attains its minimal value ( 0 ) , representing a state of determinism or lowest uncertainty . conversely , if all edges emanating from @xmath6 carry the same weight , the entropy is maximal ( 1 ) , representing the case of highly promiscuous signaling . in principle , local shannon entropies can thus be compared between phenotypes to identify genes where there are changes in the uncertainty of signaling . in the case where entropies are estimated at the phenotype level , jackknife approaches can be used to derive corresponding differential entropy statistics @xcite . deriving statistics is important because node degree has a dramatic influence on the entropy variance , with high degree nodes exhibiting significantly lower variability in absolute terms , which nevertheless could be highly significant @xcite . in the case where entropies are estimated at the level of individual samples , ordinary statistical tests ( e.g. rank sum tests ) can be used to derive sensible p - values , assuming of course that enough samples exist within the phenotypes being compared . + in addition to the local entropy , it is also of interest to consider statistical properties of the distribution of local entropies , for instance their average . comparing the average of local entropies between two phenotypes would correspond to a comparison of non - equilibrium entropy rates . to see this , consider the formal definition of the entropy rate @xmath21 @xcite , i.e. @xmath22 where @xmath23 is the stationary distribution ( equivalently the left eigenvector with unit eigenvalue ) of @xmath15 ( i.e. @xmath24 ) , and where now @xmath25 note that the entropy rate @xmath21 is an equilibrium entropy since it involves the stationary distribution of the random walker . as such , the entropy rate also depends on the global topology of the network . thus , the entropy rate is a weighted average of the local unnormalized entropies , with the weights specified by the stationary distribution . it follows that comparing the unweighted averages of local entropies reflects a comparison of a non - equilibrium entropy rate since the stationary distribution is never used . the entropy rate constructed using the mass action principle is sample specific . we previously demonstrated that this entropy rate was highly discriminative of the differentiation potential of samples within a developmental lineage , as well as being highly discriminative of normal and cancer tissue @xcite ( * fig.3a * ) . since the entropy rate takes as input a ppi network and a sample s genome - wide expression profile , the significance of the resulting entropy values , as well as that of the difference between entropy rates , also needs to be assessed relative to a null distribution in which the putative information content between network and gene expression values is non - existent . in fact , since the weights in the network determine the entropy rate and these weights are dependent on both the specific network nodes and their respective gene expression profiles , it is natural to assess the significance of the entropy rate by `` destroying '' the mutual information between the network nodes and their gene expression profiles , for instance by randomising ( i.e. permuting ) the gene expression profiles over the network . thus , under this randomisation , the topological properties of the network remain fixed , but the weights are redefined . application of this randomisation procedure to the normal / cancer expression set considered previously @xcite ( * fig.3a * ) shows that the discriminatory potential is significantly reduced upon permuting the gene expression values over the network ( * fig.3b - c * ) . importantly , we observe that the entropy rate is much higher in the normal and cancer states compared to the rates obtained upon randomisation of the gene expression profiles ( * fig.3b * ) , indicating that both normal and cancer states are characterised by a higher level of signaling promiscuity compared to a network with random weights . that the discrimination between normal and cancer is significantly reduced in the randomly weighted network further demonstrates that there is substantial mutual information between the ppi network and the gene expression profiles , thus justifying the signalling entropy approach . a vignette / manual and user - friendly r - scripts that allow computation of the entropy rate is available at the following url : _ sourceforge.net / projects / signalentropy / files/_. here we briefly describe the salient aspects of this package : + _ the input : _ the main r - script provided ( _ compsr _ ) takes as input a user - specified ppi network , and a genome - wide expression vector representing the gene expression profile of a sample . it is assumed that this has been generated using either affy , illumina or rna - sequencing . in principle one ought to use as gene expression value the normalised unlogged intensity ( affy / illu ) or rna - seq count , since this is what should be proportional to the number of rna transcripts in the sample . however , in practice we advise taking the log - transformed normalised value since the log - transformation provides a better compromise between proportionality and regularisation , i.e. some regularisation is advisable since the underlying kinetic reactions are also regular . + _ the output : _ the r - functions provided in the package then allow the user to estimate the global entropy rate for a given sample , as well as the local normalised entropies for each gene in the integrated network . if a phenotype is specified then genes can be ranked according to the correlation strength of their local entropy to the phenotype . thus , the signalling entropy method allows us to assess ( i ) if the overall levels of signalling promiscuity is different between phenotypes , which could be important , for instance , to compare the pluripotent capacity of ipscs generated via different protocols or to assess the stem cell nature of putative cancer stem cells @xcite , and ( ii ) to rank genes according to differential entropy between phenotypes , allowing key signalling genes associated with differentiation , metastasis or cancer to be identified @xcite . our previous observation that signalling entropy is increased in cancer @xcite , and that one of the key characteristics of cancer cells is their robustness to intervention and environmental stresses , suggested to us that high cellular signalling entropy may be a defining feature of a cell that is robust to general perturbations . in addition , cancer cells often exhibit the phenomenon of oncogene addiction , whereby they become overly reliant on the activation of a specific oncogenic pathway , rendering them less robust to targeted intervention @xcite . since oncogenes are normally overexpressed in cancer , it follows by the second cancer system - omic hallmark @xcite , that their lower signalling entropy may underpin their increased sensitivity to targeted drugs ( * fig.2 * ) . based on these insights , we posited that cellular signalling entropy may be correlated to the overall cellular system s robustness to perturbations . + in order to explore the relation between signalling entropy and robustness in a general context , it is useful to consider another network property of the stochastic matrix , namely the global mixing rate . this mixing rate is defined formally as the inverse of the number of timesteps for which one has to evolve a random walk on the graph so that its probability distribution is close to the stationary distribution , independently of the starting node ( * appendix b * ) . this is a notion that is closer to that of robustness or resilience as defined by demetrius and manke @xcite , allowing the mixing rate to be viewed as a crude proxy of a system s overall robustness to generic perturbations . furthermore , the global mixing rate can be easily estimated as @xmath26 where @xmath27 is the second largest ( right ) eigenvalue modulus of the stochastic matrix ( * appendix b * ) . thus , we first asked how this global mixing rate is related to the signalling entropy rate . + for a regular network of degree @xmath28 it can be easily shown that the entropy rate @xmath29 , whilst results on graph theory also show that for sufficiently large regular graphs , @xmath30 @xcite . hence , at least for regular networks a direct correlation exists . it follows that for non - regular graphs with tight degree distributions , e.g. erds - renyi ( er ) graphs , the entropy and mixing rates should also be approximately correlated . indeed , using an empirical computational approach to evaluate the entropy and mixing rates for er graphs with variable entropy rates , we were able to confirm this correlation ( * appendix c , fig.4a * ) . next , we wanted to investigate if this relationship also holds for networks with more realistic topologies than er graphs . hence , we generated connected networks on the order of 500 nodes by random subsampling 1000 nodes from our large protein interaction network ( @xmath31 nodes ) followed by extraction of the maximally connected component ( * appendix d * ) . we verified that these networks possessed approximate scale - free topologies with clustering coefficients which were also significantly higher than for er graphs . as before , for each generated network , stochastic matrices of variable entropy rates were defined . signalling entropy and mixing rates were then estimated for each of these networks , and subsequently averaged over an ensemble of such graphs . as with the random poisson ( er ) graphs , average mixing and entropy rates were highly correlated ( * fig.4b * ) . + having demonstrated a direct correlation between these two measures on simulated data , we next asked if such a correlation could also be present in the full protein interaction networks and with the stochastic matrices derived from real expression data . thus , we computed the global entropy and mixing rates for 488 cancer cell lines from the cancer cell line encylopedia ( ccle ) ( * appendix e * ) @xcite . remarkably , despite the dynamic ranges of both entropy and mixing rates being much smaller ( * fig.4c * ) compared to those of the artificially induced stochastic matrices ( c.f * fig.4a - b * ) , we were still able to detect a statistically significant correlation between entropy and mixing rates , as estimated across the 488 cell lines ( * fig.4c * ) . thus , all these results support the view that global entropy and mixing rates are correlated , albeit only in an average / ensemble sense . in the case of realistic expression and ppi data , the observed correlation between entropy and mixing rates was statistically significant but weak ( * fig.4c * ) . this could be easily attributed to the fact that in real biological networks , the global mixing rate is a very poor measure of cellular robustness . in fact , it is well known that cellular robustness is highly sensitive to which genes undergo the perturbation . for instance , in mice some genes are embryonically lethal , whereas others are not @xcite . robustness itself also admits many different definitions . because of this , we decided to investigate signalling entropy in relation to other more objective measures of cellular robustness . one such measure is drug sensitivity , for instance , ic50 values , which measure the amount of drug dose required to inhibit cell proliferation by 50@xmath32 . according to this measure , a cell that is insensitive to drug treatment is highly robust to that particular treatment . since most drugs target specific genes , we decided to explore the relation , if any , between the local signalling entropy of drug targets and their associated drug sensitivity measures . specifically , we hypothesized that since local entropy provides a proxy for local pathway redundancy , that it would correlate with drug resistance . to test this , we first computed for each of the 8038 genes in the ppi network its local signalling entropy in each of the 488 ccle cancer cell - lines . to identify associations between the 24 drug sensitivity profiles and the 8038 local network entropies , we computed non - linear rank ( spearman ) correlation coefficients across the 488 cancer cell - lines , resulting in @xmath33 correlations and associated p - values . we observed that there were many more significant associations than could be accounted for by random chance ( * fig.5a * ) , with the overall strength of association very similar to that seen between gene expression and drug sensitivity ( * fig.5b * ) . + p - value for each of 134 drug gene target pairs . * d ) * as c ) but for gene expression instead of local entropy . in * c ) & d ) * , we give the distribution of significant positive and negative correlations and the associated binomial test p - value . * e ) * drug target gene pairs ranked according to negative scc ( cyan color ) between signalling entropy and drug sensitivity . only pairs where at least one of entropy or gene expression were significantly associated are shown . upper panels show the scc ( cyan = strong negative scc , white = zero or non - significant scc , magenta = strong positive scc ) , while lower panels show the corresponding p - values with the last row showing the p - value from the likelihood ratio test ( lrt ) assessing the added predictive value of signalling entropy over gene expression . the darker the tones of red the more significant the p - values are , whilst white indicates non - significance . * f ) * a subset of e ) , with pairs now ranked according to the lrt p - value . ] one would expect the targets of specific drugs to be highly informative of the sensitivity of the corresponding drugs . we used the cancerresource @xcite to identify targets of the 24 drugs considered here , and found a total of 154 drug - target pairs . for 134 of these pairs we could compute a p - value of association between local entropy and drug sensitivity with 76 pairs ( i.e. 57@xmath32 ) exhibiting a significant association ( * fig.5c * ) . this was similar to the proportion ( 54@xmath32 ) of observed significant associations between gene expression and drug sensitivity ( * fig.5d * ) . however , interestingly , only 42 of these significant drug - target pairs were in common between the 76 obtained using signalling entropy and the 72 obtained using gene expression . importantly , while the significant associations between gene expression and drug sensitivity involved preferentially positive correlations , in the case of signalling entropy most of the significant correlations were negative ( * fig.5c - d * ) , exactly in line with our hypothesis that high entropy implies drug resistance . thus , as expected , cell - lines with highly expressed drug targets were more sensitive to treatment by the corresponding drug , but equally , drug targets exhibiting a relatively low signalling entropy were also predictive of drug sensitivity . + to formally demonstrate that local signalling entropy adds predictive power over gene expression , we considered bi - variate regression models including both the target s gene expression as well as the target s signalling entropy . using such bivariate models and likelihood ratio tests we found that in the majority of cases where signalling entropy was significantly associated with drug sensitivity that it did so independently of gene expression , adding predictive value ( * fig.5e * ) . top ranked drug - target pairs where signalling entropy added most predictive value over gene expression included topotecan/_tp53 _ and paclitaxel/_myc _ ( * fig.5f * ) . + to further demonstrate that the observed associations between local signalling entropy and drug sensitivity are statistically meaningful , we conducted a control test , in which we replaced in the multivariate model the signalling entropy of the target with a non - local entropy computed by randomly replacing the ppi neighbours of the target with other `` far - away '' genes in the network . for a considerable fraction ( 41@xmath32 ) of drug - target pairs , the original multivariate models including the local entropy constituted better predictive models than those with the non - local entropy ( false discovery rate @xmath34 ) , indicating that the observed associations are driven , at least partly , by the network structure . among the drug - target pairs for which signalling entropy was a significant predictor of drug sensitivity , we observed a striking non - linear association with the topological degree of the targets in the network ( * fig.6a * ) . in fact , for hubs in the network , most of which encode nodes located in the intracellular signaling hierarchy , high signalling entropy was exclusively associated with drug resistance ( negative scc ) . examples included well - known intracellular signalling hubs like _ hdac1 , hdac2 , akt1 , tp53 , stat3 , myc _ ) . some intracellular non - hubs ( e.g. _ casp9 , bcl2l1 , birc3 _ ) also exhibited negative correlations between signalling entropy and drug sensitivity . among targets for which high signalling entropy predicted drug sensitivity , we observed several membrane receptors ( e.g _ erbb2 , erbb3 , egfr , met _ ) and growth factors ( e.g _ hbegf , egf , tgfa _ ) . given that the correlation coefficients were estimated across samples ( cell - lines ) and that the underlying network topology is unchanged between samples , the observed non - linear relation between the directionality of the correlation and node degree is a highly non - trivial finding . we also observed a clear dependence on the main signaling domain of the target , with intracellular hubs showing preferential anti - correlations , in contrast to growth factors and membrane receptors which exhibited positive and negative correlations in equal proportion ( * fig.6b * ) . thus , we can conclude from these analyses that cancer associated changes to the interaction patterns of intra - cellular hubs are key determinants of drug resistance . in particular , in cancers where the local entropy at these hubs is increased , as a result of increased promiscuous signaling , drugs targeting these hubs are less likely to be effective . here we have advocated a fairly novel methodological framework , based on the notion of signalling entropy , to help analyze and interpret functional omic data sets . the method uses a structural network , i.e. a ppi network , from the outset , and integrates this with gene expression data , using local and global signalling entropy measures to estimate the amount of uncertainty in the network signaling patterns . we made the case as to why uncertainty or entropy might be a key concept to consider when analysing biological data . in previous work @xcite , we showed how signalling entropy can be used to estimate the differentiation potential of a cellular sample in the context of normal differentiation processes , as well as demonstrating that signalling entropy also provides a highly accurate discriminator of cancer phenotypes . + in this study we focused on a novel application of signalling entropy to understanding cellular robustness in the context of cancer drug sensitivity screens . our main findings are that ( i ) local signalling entropy measures add predictive value over models that only use gene expression , ( ii ) that the local entropy of drug targets generally correlates positively with drug resistance , and ( iii ) that increased local entropy of intra - cellular hubs in cancer cells is a key hallmark of drug resistance . + these results are consistent and suggestive of an underlying entropy - robustness correlation theorem , as envisaged by previous authors @xcite . here , we provided additional empirical justification for such a theorem , using both simulated as well as real data , and using drug sensitivity measures as proxies for local robustness measures . a more detailed theoretical analysis of local mixing and entropy rates and incorporation of additional information ( e.g. phosphorylation states of kinases , protein expression, .. etc ) when estimating entropies on real data , will undoubtedly lead to further improvements in our systems - level understanding of how entropy / uncertainty dictates cellular phenotypes . from a practical perspective , we have already shown in previous work @xcite how local network entropies could be used to identify key signaling pathways in differentiation . it will therefore be interesting in future to apply the signalling entropy framework in more detail to identify specific signaling nodes / pathways underlying drug sensitivity / resistance . + our results have also confirmed the importance of network topology ( e.g. hubness and therefore scale - freeness ) in dictating drug resistance patterns . thus , it will be interesting to continue to explore the deep relation between topological features such as scale - freeness and hierarchical modularity in relation to the gene expression patterns seen in normal and disease physiology . it is entirely plausible that , although our current data and network models are only mere caricatures of the real biological networks , that underlying fundamental systems biology principles characterising cellular phenotypes can still be gleaned from this data . indeed , our discovery that increased signalling entropy correlates with drug resistance demonstrates that such fundamental principles can already be inferred from existing data resources . + it will also be of interest to consider other potential applications of the signaling entropy method . for instance , one application could be to the identification of functional driver aberrations in cancer . this would first use epigenomic ( e.g. dna methylation and histone modification profiles ) and genomic information ( snps , cnvs ) together with matched gene or protein expression data to identify functional epigenetic / genetic aberrations in cancer . signalling entropy would subsequently be used as a means of identifying those aberrations which also cause a fundamental rewiring of the network . with multi - dimensional matched omic data readily available from the tcga / icgc , this represents another potentially important application of the signalling entropy method . another important future application of the signaling entropy method would be to single - cell data , which is poised to become ever more important @xcite . so far , all signaling entropy analyses have been performed on cell populations , yet single - cell analysis will be necessary to disentangle the entropies at the single - cell and population - cell levels . + in summary , we envisage that signalling entropy will become a key concept in future analyses and interpretation of biological data . aet is supported by the chinese academy of sciences , shanghai institute for biological sciences and the max - planck gesellshaft . rk and ps acknowledge funding under fp7/2007 - 2013/grant agreement nr . we downloaded the complete human protein interaction network from pathway commons ( _ www.pathwaycommons.org _ ) ( jun.2012 ) @xcite , which brings together protein interactions from several distinct sources . we built a protein protein interaction ( ppi ) network from integrating the following sources : the human protein reference database ( hprd ) @xcite , the national cancer institute nature pathway interaction database ( nci - pid ) ( _ pid.nci.nih.gov_ ) , the interactome ( intact ) _ http://www.ebi.ac.uk/intact/ _ and the molecular interaction database ( mint ) _ http://mint.bio.uniroma2.it / mint/_. protein interactions in this network include physical stable interactions such as those defining protein complexes , as well as transient interactions such as post - translational modifications and enzymatic reactions found in signal transduction pathways , including 20 highly curated immune and cancer signaling pathways from netpath ( _ www.netpath.org _ ) @xcite . we focused on non - redundant interactions , only included nodes with an entrez gene i d annotation and focused on the maximally conntected component , resulting in a connected network of 10,720 nodes ( unique entrez ids ) and 152,889 documented interactions . suppose that we have a connected graph with an ergodic markov chain defined on it , given by a stochastic matrix @xmath15 with stationary distribution @xmath35 ( i.e. @xmath24 ) . we further assume that the detailed balance equation , @xmath36 holds . defining the diagonal matrix @xmath37 , the detailed balance equation can be rewritten as @xmath38 . that the matrix @xmath15 is stochastic means that each row of @xmath15 sums to 1 . equivalently , the vector with all unit entries , 1 , is a right eigenvector of @xmath15 , i.e. @xmath39 . note also that the stationary distribution @xmath35 is a left eigenvector of @xmath15 with eigenvalue 1 . the perron - frobenius theorem further implies that all other eigenvalues are less than 1 in magnitude . if detailed balance holds , all eigenvalues are also real . + the global mixing rate can be defined by considering the rate at which the node visitation probabilities of a random walker approaches that of the stationary distribution , independently of the starting position . formally , if we let @xmath40 denote the probability that at time @xmath41 we find the walker at node @xmath6 , then the mixing rate , @xmath42 , is defined by @xcite @xmath43 denoting by @xmath44 the column vector with elements @xmath40 , one can write @xmath45 to determine the elements , @xmath40 , of this vector , it is convenient to first introduce the matrix @xmath46 . this is because @xmath47 , and so @xmath48 can be rewritten in terms of @xmath49 , but also because @xmath49 satisfies the following lemma : @xmath49 has the same eigenvalues as @xmath15 and if @xmath50 is an eigenvector of @xmath49 , then @xmath51 and @xmath52 are right and left eigenvectors of @xmath15 . suppose that @xmath53 . it then follows that @xmath54 in the case of the left - eigenvector , multiply @xmath53 from the left with @xmath55 . then , @xmath56 detailed balance implies that @xmath38 , so @xmath57 . taking the transpose of this implies that @xmath58 is indeed a left - eigenvector of @xmath15 . the significance of the above lemma becomes clear in light of the detailed balance equation , which implies that @xmath59 , and so @xmath49 and @xmath60 can be orthogonally diagonalized . in particular , we can express @xmath40 in terms of the eigenvalue decomposition of @xmath60 , as @xmath61 , where @xmath62 and where @xmath63 is the @xmath64th unit norm eigenvector of @xmath49 . since @xmath65 is the top unit norm eigenvector of @xmath49 ( using previous lemma ) , which has an eigenvalue of 1 , it follows that @xmath66 and hence that @xmath67 it follows that @xmath68 since @xmath69 , we can conclude that as @xmath70 , the rate at which @xmath71 approaches the stationary value @xmath72 is determined by the modulus of the second largest eigenvalue ( the second largest eigenvalue modulus - slem ) . the global mixing rate @xmath42 can thus be estimated as @xmath73 . for large regular graphs of degree @xmath28 , the mixing rate is proportional to @xmath74 @xcite and thus directly proportional to the entropy rate ( @xmath29 for a regular graph of degree @xmath28 ) . by extrapolation , we can thus reasonably assume that for any sufficiently large graph with a tight degree distribution , such as random erdos - renyi ( er ) graphs , that the entropy and mixing rates will also be correlated , albeit perhaps only in an average ensemble sense . the analytical demonstration of this is beyond the scope of this work . hence , we adopt a computational empirical approach to see if the entropy and mixing rates may indeed be correlated . in detail ( and without loss of generality concerning the end result ) we considered er graphs of size 100 nodes and average degree 10 ( other values for these parameters gave similar answers ) . we built an ensemble of 100 distinct er graphs , and for each of these we constructed a family of weighted networks , parameterised by a parameter @xmath75 which controls the level of signalling entropy . specifically , we gradually shifted the weight distribution around each node @xmath6 ( with degree @xmath76 ) , in such a way that @xmath77 for @xmath78 and @xmath79 for @xmath80 , with @xmath81 and with @xmath82 labeling a randomly chosen neighbor of node @xmath6 . thus , @xmath75 is a parameter that directly controls the uncertainty in the information flow , i.e. the entropy rate , since for @xmath83 we have that @xmath84 ( @xmath85 is the symmetric adjacency matrix of the graph ) , whilst for @xmath86 , @xmath87 , i.e. the information flow from node @xmath6 can only proceed along one node ( node @xmath82 ) . thus , one would expect the entropy rate to decrease as @xmath75 is decreased to zero . for each value of @xmath75 we thus have an ensemble of 100 er - graphs , for each of which the entropy and mixing rates can be computed . finally , at each value of @xmath75 the entropy and mixing rates are averaged over the ensemble . the analysis described above was performed also for maximally connected subnetworks generated from the underlying realistic ppi network described earlier . specifically , we randomly subsampled 1000 nodes from the 8038 node ppi network , and extracted the maximally connected subnetwork , which resulted ( on average ) in a subnetwork of approximately 500 nodes . a family of stochastic matrices of variable entropy rates were constructed as explained above and for each resulting weighted network we estimated the entropy and mixing rates . finally , ensemble averages over 100 different realisations were computed . we used the gene expression data and drug sensitivity profiles as provided in the previous publication @xcite . briefly , integration of the gene expression data with our ppi network resulted in a maximally connected component consisting of 8038 genes / proteins . there were 488 cell - lines with gene expression and drug sensitivity profiles for 24 drugs . as a measure of drug response we used the activity area @xcite since this measure gave optimal results when correlating drug response to gene expression levels of well established drug targets . 10 url # 1`#1`urlprefixhref # 1#2#2 # 1#1 l. hood , l. rowen , the human genome project : big science transforms biology and medicine , genome med 5 ( 9 ) ( 2013 ) 79 . n. tcga , integrated genomic analyses of ovarian carcinoma , nature 474 ( 7353 ) ( 2011 ) 609615 . d. hanahan , r. a. weinberg , hallmarks of cancer : the next generation , cell 144 ( 5 ) ( 2011 ) 646674 . a. l. barabasi , n. gulbahce , j. loscalzo , network medicine : a network - based approach to human disease , nat rev genet 12 ( 1 ) ( 2011 ) 5668 . p. creixell , e. m. schoof , j. t. erler , r. linding , navigating cancer network attractors for tumor - specific therapy , nat . 30 ( 9 ) ( 2012 ) 842848 . g. weng , u. s. bhalla , r. iyengar , complexity in biological signaling systems , science 284 ( 5411 ) ( 1999 ) 9296 . a. l. barabasi , network science , philos trans a math phys eng sci 371 ( 1987 ) ( 2013 ) 20120375 . a. e. teschendorff , s. severini , increased entropy of signal transduction in the cancer metastasis phenotype , bmc syst biol 4 ( 2010 ) 104 . j. west , g. bianconi , s. severini , a. e. teschendorff , differential network entropy reveals cancer system hallmarks , sci rep 2 ( 2012 ) 802 . a. l. barabasi , z. n. oltvai , network biology : understanding the cell s functional organization , nat rev genet 5 ( 2 ) ( 2004 ) 101113 . x. robin , p. creixell , o. radetskaya , c. costa santini , j. longden , r. linding , personalized network - based treatments in oncology , clin . j. t. erler , r. linding , network medicine strikes a blow against breast cancer , cell 149 ( 4 ) ( 2012 ) 731733 . m. j. lee , a. s. ye , a. k. gardino , a. m. heijink , p. k. sorger , g. macbeath , m. b. yaffe , sequential application of anticancer drugs enhances cell death by rewiring apoptotic signaling networks , cell 149 ( 4 ) ( 2012 ) 780794 . m. s. carro , w. k. lim , m. j. alvarez , r. j. bollo , x. zhao , e. y. snyder , e. p. sulman , s. l. anne , f. doetsch , h. colman , a. lasorella , k. aldape , a. califano , a. iavarone , the transcriptional network for mesenchymal transformation of brain tumours , nature 463 ( 7279 ) ( 2010 ) 318325 . c. lefebvre , g. rieckhof , a. califano , reverse engineering human regulatory networks , wiley interdiscip rev syst biol med 4 ( 4 ) ( 2012 ) 31125 . e. segal , n. friedman , d. koller , a. regev , a module map showing conditional activity of expression modules in cancer , nat genet 36 ( 10 ) ( 2004 ) 10901098 . r. opgen - rhein , k. strimmer , from correlation to causation networks : a simple approximate learning algorithm , bmc syst biol 1 ( 2007 ) 37 . b. barzel , a. l. barabasi , network link prediction by global silencing of indirect correlations , nat biotechnol 31 ( 8) ( 2013 ) 7205 . s. feizi , d. marbach , m. medard , m. kellis , network deconvolution as a general method to distinguish direct dependencies in , nat biotechnol 31 ( 8) ( 2013 ) 72633 . h. y. chuang , e. lee , y. t. liu , d. lee , t. ideker , network - based classification of breast cancer metastasis , mol syst biol 3 ( 2007 ) 140 . j. dutkowski , t. ideker , protein networks as logic functions in development and cancer , plos comput biol 7 ( 9 ) ( 2011 ) e1002180 . k. mitra , a. r. carvunis , s. k. ramesh , t. ideker , integrative approaches for finding modular structure in biological networks , nat rev genet 14 ( 10 ) ( 2013 ) 71932 . t. s. prasad , k. kandasamy , a. pandey , human protein reference database and human proteinpedia as discovery tools for systems biology , methods mol biol 577 ( 2009 ) 6779 . e. g. cerami , b. e. gross , e. demir , i. rodchenkov , o. babur , n. anwar , n. schultz , g. d. bader , c. sander , pathway commons , a web resource for biological pathway data , nucleic acids res 39 ( database ) ( 2011 ) d685d690 . m. vidal , m. e. cusick , a. l. barabasi , interactome networks and human disease , cell 144 ( 6 ) ( 2011 ) 986998 . i. ulitsky , r. shamir , identification of functional modules using network topology and high - throughput data , bmc syst biol 1 ( 2007 ) 8 . i. w. taylor , r. linding , d. warde - farley , y. liu , c. pesquita , d. faria , s. bull , t. pawson , q. morris , j. l. wrana , dynamic modularity in protein interaction networks predicts breast cancer outcome , nat biotechnol 27 ( 2 ) ( 2009 ) 199204 . j. west , s. beck , x. wang , a. e. teschendorff , an integrative network algorithm identifies age - associated differential methylation interactome hotspots targeting stem - cell differentiation pathways , sci rep 3 ( 2013 ) 1630 . j. roy , c. winter , z. isik , m. schroeder , network information improves cancer outcome prediction , brief bioinform . c. winter , g. kristiansen , s. kersting , j. roy , d. aust , t. knosel , p. rummele , b. jahnke , v. hentrich , f. ruckert , m. niedergethmann , w. weichert , m. bahra , h. j. schlitt , u. settmacher , h. friess , m. buchler , h. d. saeger , m. schroeder , c. pilarsky , r. grutzmann , google goes cancer : improving outcome prediction for cancer patients by , plos comput biol 8 ( 5 ) ( 2012 ) e1002511 . s. suthram , a. beyer , r. m. karp , y. eldar , t. ideker , eqed : an efficient method for interpreting eqtl associations using protein , mol syst biol 4 ( 2008 ) 162 . y. a. kim , s. wuchty , t. m. przytycka , identifying causal genes and dysregulated pathways in complex diseases , plos comput biol 7 ( 3 ) ( 2011 ) e1001095 . k. komurov , m. a. white , p. t. ram , use of data - biased random walks on graphs for the retrieval of context - specific networks from genomic data , plos comput biol 6 ( 8) . k. komurov , j. t. tseng , m. muller , e. g. seviour , t. j. moss , l. yang , d. nagrath , p. t. ram , the glucose - deprivation network counteracts lapatinib - induced toxicity in , mol syst biol 8 ( 2012 ) 596 . s. bandyopadhyay , m. mehta , d. kuo , m. k. sung , r. chuang , e. j. jaehnig , b. bodenmiller , k. licon , w. copeland , m. shales , d. fiedler , j. dutkowski , a. gunol , h. van attikum , k. m. shokat , r. d. kolodner , w. k. huh , r. aebersold , m. c. keogh , n. j. krogan , t. ideker , rewiring of genetic networks in response to dna damage , science 330 ( 6009 ) ( 2010 ) 13851389 . a. califano , rewiring makes the difference , mol syst biol 7 ( 2011 ) 463 . t. ideker , n. j. krogan , differential network biology , mol syst biol 8 ( 2012 ) 565 . c. r. banerji , d. miranda - saavedra , s. severini , m. widschwendter , t. enver , j. x. zhou , a. e. teschendorff , cellular network entropy as the energy potential in waddington s differentiation landscape , sci rep 3 ( 2013 ) 3039 . h. h. chang , m. hemberg , m. barahona , d. e. ingber , s. huang , transcriptome - wide noise controls lineage choice in mammalian progenitor cells , nature 453 ( 7194 ) ( 2008 ) 544547 . b. d. macarthur , i. r. lemischka , statistical mechanics of pluripotency , cell 154 ( 3 ) ( 2013 ) 484489 . c. furusawa , k. kaneko , chaotic expression dynamics implies pluripotency : when theory and experiment meet , biol direct . 15 ( 4 ) ( 2009 ) 17 . c. furusawa , k. kaneko , a dynamical systems view of stem cell biology , science 338 ( 6104 ) ( 2012 ) 2157 . t. i. lee , r. g. jenner , l. a. boyer , m. g. guenther , s. s. levine , r. m. kumar , b. chevalier , s. e. johnstone , m. f. cole , k. isono , h. koseki , t. fuchikami , k. abe , h. l. murray , j. p. zucker , b. yuan , g. w. bell , e. herbolsheimer , n. m. hannett , k. sun , d. t. odom , a. p. otte , t. l. volkert , d. p. bartel , d. a. melton , d. k. gifford , r. jaenisch , r. a. young , control of developmental regulators by polycomb in human embryonic stem cells , cell 125 ( 2 ) ( 2006 ) 301313 . l. demetrius , v. m. grundlach , g. ochs , complexity and demographic stability in population models , theo pop biol 65 ( 2004 ) 211225 . l. demetrius , t. manke , robustness and network evolution - an entropic principle , physica a 346 ( 2005 ) 682696 . t. manke , l. demetrius , m. vingron , an entropic characterization of protein interaction networks and cellular robustness , j r soc interface 3 ( 11 ) ( 2006 ) 843850 . j. stelling , u. sauer , z. szallasi , f. j. doyle , j. doyle , robustness of cellular functions , cell 118 ( 6 ) ( 2004 ) 675685 . j. d. han , n. bertin , t. hao , d. s. goldberg , g. f. berriz , l. v. zhang , d. dupuy , a. j. walhout , m. e. cusick , f. p. roth , m. vidal , evidence for dynamically organized modularity in the yeast protein - protein interaction network , nature 430 ( 6995 ) ( 2004 ) 8893 . j. li , z. yuan , z. zhang , the cellular robustness by gene redundancy in budding yeast , plos genet 6 ( 11 ) ( 2010 ) e1001187 . j. a. engelman , l. chen , x. tan , k. crosby , a. r. guimaraes , r. upadhyay , m. maira , k. mcnamara , s. a. perera , y. song , l. r. chirieac , r. kaur , a. lightbown , j. simendinger , t. li , r. f. padera , c. garcia - echeverria , r. weissleder , u. mahmood , l. c. cantley , k. k. wong , effective use of pi3k and mek inhibitors to treat mutant kras g12d and pik3ca h1047r murine lung cancers , nat . 14 ( 12 ) ( 2008 ) 13511356 . t. sjoblom , s. jones , l. d. wood , d. w. parsons , j. lin , t. d. barber , d. mandelker , r. j. leary , j. ptak , n. silliman , s. szabo , p. buckhaults , c. farrell , p. meeh , s. d. markowitz , j. willis , d. dawson , j. k. willson , a. f. gazdar , j. hartigan , l. wu , c. liu , g. parmigiani , b. h. park , k. e. bachman , n. papadopoulos , b. vogelstein , k. w. kinzler , v. e. velculescu , the consensus coding sequences of human breast and colorectal cancers , science 314 ( 5797 ) ( 2006 ) 268274 . l. d. wood , d. w. parsons , s. jones , j. lin , t. sjblom , r. j. leary , d. shen , s. m. boca , t. barber , j. ptak , n. silliman , s. szabo , z. dezso , v. ustyanksky , t. nikolskaya , y. nikolsky , r. karchin , p. a. wilson , j. s. kaminker , z. zhang , r. croshaw , j. willis , d. dawson , m. shipitsin , j. k. willson , s. sukumar , k. polyak , b. h. park , c. l. pethiyagoda , p. v. pant , d. g. ballinger , a. b. sparks , j. hartigan , d. r. smith , e. suh , n. papadopoulos , p. buckhaults , s. d. markowitz , g. parmigiani , k. w. kinzler , v. e. velculescu , b. vogelstein , the genomic landscapes of human breast and colorectal cancers , science 318 ( 5853 ) ( 2007 ) 11081113 . b. vogelstein , n. papadopoulos , v. e. velculescu , s. zhou , l. a. diaz , k. w. kinzler , cancer genome landscapes , science 339 ( 6127 ) ( 2013 ) 154658 . v. latora , m. baranger , kolmogorov - sinai entropy rate versus physical entropy , phys rev lett 82 ( 3 ) ( 1999 ) 520524 . j. gomez - gardenes , v. latora , entropy rate of diffusion processes on complex networks , phys rev e stat nonlin soft matter phys 78 ( 6 ) ( 2008 ) 065102 . j. friedman , on the second eigenvalue and random walks in random d - regular graphs , combinatorica 11 ( 1991 ) 331262 . j. barretina , g. caponigro , n. stransky , k. venkatesan , a. a. margolin , s. kim , c. j. wilson , j. lehr , g. v. kryukov , d. sonkin , a. reddy , m. liu , l. murray , m. f. berger , j. e. monahan , p. morais , j. meltzer , a. korejwa , j. jan - valbuena , f. a. mapa , j. thibault , e. bric - furlong , p. raman , a. shipway , i. h. engels , j. cheng , g. k. yu , j. yu , p. aspesi jr , m. de silva , k. jagtap , m. d. jones , l. wang , c. hatton , e. palescandolo , s. gupta , s. mahan , c. sougnez , r. c. onofrio , t. liefeld , l. macconaill , w. winckler , m. reich , n. li , j. p. mesirov , s. b. gabriel , g. getz , k. ardlie , v. chan , v. e. myer , b. l. weber , j. porter , m. warmuth , p. finan , j. l. harris , m. meyerson , t. r. golub , m. p. morrissey , w. r. sellers , r. schlegel , l. a. garraway , the cancer cell line encyclopedia enables predictive modelling of anticancer drug sensitivity , nature 483 ( 7391 ) ( 2012 ) 603607 . j. ahmed , t. meinel , m. dunkel , m. s. murgueitio , r. adams , c. blasse , a. eckert , s. preissner , r. preissner , cancerresource : a comprehensive database of cancer - relevant proteins and compound interactions supported by experimental knowledge , nucleic acids res 39 ( database ) ( 2011 ) d960d967 . m. r. speicher , single - cell analysis : toward the clinic , genome med 5 ( 8) ( 2013 ) 74 . k. kandasamy , s. s. mohan , r. raju , s. keerthikumar , g. s. kumar , a. k. venugopal , d. telikicherla , j. d. navarro , s. mathivanan , c. pecquet , s. k. gollapudi , s. g. tattikota , s. mohan , h. padhukasahasram , y. subbannayya , r. goel , h. k. jacob , j. zhong , r. sekhar , v. nanjappa , l. balakrishnan , r. subbaiah , y. l. ramachandra , b. a. rahiman , t. s. prasad , j. x. lin , j. c. houtman , s. desiderio , j. c. renauld , s. n. constantinescu , o. ohara , t. hirano , m. kubo , s. singh , p. khatri , s. draghici , g. d. bader , c. sander , w. j. leonard , a. pandey , netpath : a public resource of curated signal transduction pathways , genome biol 11 ( 1 ) ( 2010 ) r3 . l. lovasz , random walks on graphs : a survey , in : combinatorics , paul erdos is eighty , vol . 2 of mathematical studies , bolyai society , keszthely , hungary , 1993 , pp .

the deterministic network calculus offers an elegant framework for determining delays and backlog in a network with deterministic service guarantees to individual traffic flows . this paper addresses the problem of extending the network calculus to a probabilistic framework with statistical service guarantees . here , the key difficulty relates to expressing , in a statistical setting , an end - to - end ( network ) service curve as a concatenation of per - node service curves . the notion of an _ effective service curve _ is developed as a probabilistic bound on the service received by an individual flow . it is shown that per - node effective service curves can be concatenated to yield a network effective service curve . # 1 _ key words : quality - of - service , service differentiation , statistical service , network calculus .

cs0205001

the deterministic network calculus recently evolved as a fundamental theory for quality of service ( qos ) networks , and has provided powerful tools for reasoning about delay and backlog in a network with service guarantees to individual traffic flows . using the notion of arrival envelopes and service curves @xcite , several recent works have shown that delay and backlog bounds can be concisely expressed in a min - plus algebra @xcite . however , the deterministic view of traffic generally overestimates the actual resource requirements of a flow and results in a low utilization of available network resources . this motivates the search for a statistical network calculus that can exploit statistical multiplexing , while preserving the algebraic aspects of the deterministic calculus . the problem of developing a probabilistic network calculus has been the subject of several studies . kurose @xcite uses the concept of stochastic ordering and obtains bounds on the distribution of delay and buffer occupancy of a flow in a network with fifo scheduling . chang @xcite presents probabilistic bounds on output burstiness , backlog and delays in a network where the moment generating functions of arrivals are exponentially bounded . different bounds for stochastically bounded arrivals are derived by yaron and sidi @xcite and starobinski and sidi @xcite . the above results can be used to determine stochastic end - to - end performance bounds . results on statistical end - to - end delay guarantees in a network have been obtained for specific scheduling algorithms , such as edf @xcite , and gps @xcite , and a class of coordinated scheduling algorithms @xcite . several researchers have considered probabilistic formulations of service curves . cruz defines a probabilistic service curve which violates a given deterministic service curve according to a certain distribution @xcite . chang ( see @xcite , chp . 7 ) presents exercises which hint at a statistical network calculus for the class of ` dynamic f - servers ' . finally , knightly and qiu @xcite derive ` statistical service envelopes ' as time - invariant lower bounds on the service received by an aggregate of flows . with exception of ( @xcite , chp . 7 ) , none of the cited works express statistical end - to - end performance bounds in a min - plus algebra , and it has been an open question whether a statistical network calculus can be developed in this setting . the contribution of this paper is the presentation of a statistical network calculus that uses the min - plus algebra @xcite . the advantage of using the min - plus algebra is that end - to - end guarantees can be expressed as a simple concatenation of single node guarantees , which , in turn , can be exploited to achieve simple probabilistic bounds . we define an _ effective service curve _ , which is , with high certainty , a probabilistic bound on the service received by a single flow . we will show that the main results of the deterministic network calculus carry over to the statistical framework we present . our derivations reveal a difficulty that occurs when calculating probabilistic service guarantees for multiple nodes . we show that the problem can be overcome either by adding assumptions on the traffic at nodes or by modifying the definition of the effective service curve . the results in this paper are set in a continuous time model with fluid left - continuous traffic arrival functions , as is common for network delay analysis in the deterministic network calculus . a node represents a router ( or switch ) in a network . packetization delays and other effects of discrete - sized packets , such as the non - preemption of packet transmission , are ignored . we refer to @xcite for the issues involved in relaxing these assumptions for the analysis of packet networks . when analyzing delays in a network , all processing overhead and propagation delays are ignored . as in the deterministic network calculus , arrivals from a traffic flow to the network satisfy deterministic upper bounds , which are enforced by a deterministic regulator . the remaining sections of this paper are structured as follows . in section [ sec - detcalc ] , we review the notation and key results of the deterministic network calculus . in section [ sec - statcalc ] we introduce effective service curves and present the results for a statistical network calculus in terms of effective service curves . in section [ sec - motivate ] we provide a discussion that motivates our revised definition of an effective service curve . in section [ sec - concl ] , we present brief conclusions . the deterministic network calculus , which was created in @xcite and fully developed in the last decade , provides concise expressions for upper bounds on the backlog and delay experienced by an individual flow at one or more network nodes . an attractive feature of the network calculus is that end - to - end bounds can often be easily obtained from manipulations of the per - node bounds . in this section we review some notation and results from the deterministic network calculus . this section is not a comprehensive summary of the network calculus and we refer to @xcite for a complete discussion . much of the formal framework of the network calculus can be elegantly expressed in a min - plus algebra @xcite , complete with convolution and deconvolution operators for functions . generally , the functions in this paper are non - negative , non - decreasing , and left - continuous , defined over time intervals @xmath0 $ ] . we assume for a given function @xmath1 that @xmath2 if @xmath3 . the _ convolution _ @xmath4 of two functions @xmath1 and @xmath5 , is defined as @xmath6 } \left\ { f(t-\tau ) + g ( \tau)\right\ } \ .\ ] ] the _ deconvolution _ @xmath7 of two functions @xmath1 and @xmath5 is defined as @xmath8 for @xmath9 , the _ impulse function _ @xmath10 is defined as @xmath11 if @xmath1 is nondecreasing , we have the formulas @xmath12 we refer to @xcite for a detailed discussion of the properties of the min - plus algebra and the properties of the convolution and deconvolution operators . let us consider the traffic arrivals to a single network node . the arrivals of a flow in the time interval @xmath13 are given in terms of a function @xmath14 . the departures of a flow from the node in the time interval @xmath13 are denoted by @xmath15 , with @xmath16 . the backlog of a flow at time @xmath17 , denoted by @xmath18 , is given by @xmath19 the delay at time @xmath17 , denoted as @xmath20 , is the delay experienced by an arrival which departs at time @xmath17 , given by @xmath21 we will use @xmath22 and @xmath23 to denote the arrivals and departures in the time interval @xmath24 , with @xmath25 and @xmath26 . we make the following assumptions on the arrival functions . * _ non - negativity . _ the arrivals in any interval of time are non - negative . that is , for any @xmath27 , we have @xmath28 . * _ upper bound . _ the arrivals @xmath29 of a flow are bounded by a subadditive is _ subadditive _ if @xmath30 , for all @xmath31 , or , equivalently , if @xmath32 . ] function @xmath33 , called the _ arrival envelope _ , a function @xmath34 is called an _ envelope _ for a function @xmath1 if @xmath35 for all @xmath36 , or , equivalently , if @xmath37 . ] such that @xmath38 for all @xmath36 . a _ minimum service curve _ for a flow is a function @xmath39 which specifies a lower bound on the service given to the flow such that , for all @xmath40 , @xmath41 a _ maximum service curve _ for a flow is a function @xmath42 which specifies an upper bound on the service given to a flow such that , for all @xmath40 , @xmath43 minimum service curves play a larger role in the network calculus since they provide service guarantees . therefore , we , as the related literature , often refer to a minimum service curve simply as a service curve . if no maximum service curve is explicitly given , one can use @xmath44 , where @xmath45 is the link capacity . the following two theorems summarize some key results of the deterministic network calculus . these results have been derived in @xcite . we follow the notation used in @xcite . * deterministic calculus @xcite . * [ theorem - detcalc ] given a flow with arrival envelope @xmath33 and with minimum service curve @xmath39 , the following hold : 1 . * output envelope . * the function @xmath46 is an envelope for the departures , in the sense that , for all @xmath36 , @xmath47 2 . * backlog bound . * an upper bound for the backlog , denoted by @xmath48 , is given by @xmath49 3 . * delay bound . * an upper bound for the delay , denoted by @xmath50 , is given by @xmath51 the next theorem states that the service curves of a flow at the nodes on its route can be concatenated to define a network service curve , which expresses service guarantees offered to the flow by the network as a whole . * concatenation of deterministic network service curves @xcite . * [ theorem - detnet ] suppose a flow passes through @xmath52 nodes in series , as shown in figure [ fig - net ] , and suppose the flow is offered minimum and maximum service curves @xmath53 and @xmath54 , respectively , at each node @xmath55 . then , the sequence of nodes provides minimum and maximum service curves @xmath56 and @xmath57 , which are given by @xmath58 @xmath56 and @xmath57 will be referred to as _ network service curves _ , and eqs . ( [ eq - def - snetmin])([eq - def - snetmax ] ) will be called the _ concatenation formulas_. with theorems [ theorem - detcalc ] and [ theorem - detnet ] network service curves can be used to determine bounds on delay and backlog for individual flows in a network . there are many additional properties and refinements that have been derived for the deterministic calculus . however , in this paper we will concern ourselves only with the results above . we now approach the network calculus in a probabilistic framework . arrivals and departures from a flow to the network in the time interval @xmath13 are described by random processes @xmath59 and @xmath60 satisfying assumptions ( a1 ) and ( a2 ) . the random processes are defined over an underlying joint probability space that we suppress in our notation . the statistical network calculus makes service guarantees for individual flows , where each flow is allocated a probabilistic service in the form of an ` effective service curve ' . given a flow with arrival process @xmath61 , a _ ( minimum ) effective service curve _ is a nonnegative function @xmath62 that satisfies for all @xmath63 , @xmath64 note that the effective service curve is a non - random function . we omit the corresponding definition of a _ maximum effective service curve_. the following theorem is a probabilistic counterpart to theorem [ theorem - detcalc ] . * statistical calculus . * [ theorem - statcalc ] given a flow with arrival process @xmath61 satisfying assumptions ( a1)(a2 ) , and given an effective service curve @xmath62 , the following hold : 1 . * output envelope . * the function @xmath65 is a probabilistic bound for the departures on @xmath0 $ ] , in the sense that , for all @xmath66 , @xmath67 2 . * backlog bound . * a probabilistic bound for the backlog is given by @xmath68 , in the sense that , for all @xmath63 , @xmath69 3 . * delay bound . * a probabilistic bound for the delay is given by , @xmath70 in the sense that , for all @xmath63 , @xmath71 by setting @xmath72 in theorem [ theorem - statcalc ] , we can recover the bounds of theorem [ theorem - detcalc ] with probability one . * the proof uses on several occasions that the inequality @xmath73 holds for any nonnegative function @xmath5 and for all @xmath74 . to see this inequality , we compute @xmath75 } \bigl\ { \ain ( t - x ) + g ( x ) \bigr\ } \label{eq - detcalc - proof1 - 2}\\ & = & \sup_{x \in [ 0 , t ] } \bigl\ { \ain ( t - x , t + \tau ) - g ( x)\bigr\ } \label{eq - detcalc - proof1 - 3}\\ & \leq & \sup_{x \geq 0 } \bigl\ { a^ * ( \tau + x ) - g ( x)\bigr\ } \label{eq - detcalc - proof1 - 4}\\ & = & a^ * \deconv g ( \tau ) \ . \label{eq - detcalc - proof1 - 5}\end{aligned}\ ] ] eqn . ( [ eq - detcalc - proof1 - 2 ] ) expands the convolution operator . ( [ eq - detcalc - proof1 - 3 ] ) takes @xmath76 inside the infimum and uses that @xmath77 . ( [ eq - detcalc - proof1 - 4 ] ) uses that @xmath78 for all @xmath79 by definition of an arrival envelope , and extends the range of the supremum . finally , eqn . ( [ eq - detcalc - proof1 - 5 ] ) uses the definition of the deconvolution operator . _ proof of the output bound . _ for any fixed @xmath80 , we have @xmath81 eqn . ( [ eq - statcalc - proofi-0 ] ) holds by the definition of the effective service curve @xmath62 . ( [ eq - statcalc - proofi-1 ] ) uses that @xmath82 . eqn . ( [ eq - statcalc - proofi-2 ] ) uses that departures in @xmath13 can not exceed arrivals , that is , @xmath16 for all @xmath40 . finally , eqn . ( [ eq - statcalc - proofi-3 ] ) uses that @xmath83 by eqn . ( [ eq - detcalc - proof1 - 0 ] ) . proof of the backlog bound . _ since @xmath84 and with the definition of the effective service curve , we can write @xmath85 eqn . ( [ eq - statcalc - proofii-1 ] ) holds by definition of the effective service curve . ( [ eq - statcalc - proofii-2 ] ) uses that @xmath86 , and eqn . ( [ eq - statcalc - proofii-8 ] ) uses that @xmath87 by eqn . ( [ eq - detcalc - proof1 - 0 ] ) . proof of the delay bound . _ the delay bound is proven by estimating the probability that the output @xmath60 exceeds the arrivals @xmath88 . @xmath89 eqn . ( [ eq - statcalc - proofiii-1 ] ) uses the definition of the effective service curve @xmath62 . ( [ eq - statcalc - proofiii-2 ] ) uses the definition of the impulse function in eqn . ( [ eq - impulse ] ) and the definition of @xmath50 in eqn . ( [ eq - def - dmax ] ) . ( [ eq - statcalc - proofiii-2a ] ) follows from the associativity of the convolution , and eqn . ( [ eq - statcalc - proofiii-2a ] ) uses the definition of an arrival envelope . @xmath90 a probabilistic counterpart to theorem [ theorem - detnet ] can be formulated as follows . * concatenation of effective service curves . * [ theorem - statnet - intervals - x ] consider a flow that passes through @xmath52 network nodes in series , as shown in figure [ fig - net ] . assume that effective service curves are given by nondecreasing functions @xmath91 at each node ( @xmath92 ) . then , for any @xmath40 , @xmath93 where @xmath94 is an arbitrary parameter . again , we can can recover the deterministic result from theorem [ theorem - detnet ] . by setting @xmath72 , the results in eqn . ( [ eq - statnet - intervals - x ] ) hold with probability one . then by letting @xmath95 , we obtain theorem [ theorem - detnet ] almost surely . * proof . * we proceed in three steps . in the first step , we modify the effective service curve to give lower bounds on the departures simultaneously for all times in the entire interval @xmath0 $ ] . in the second step , we perform a deterministic calculation . the proof concludes with a simple probabilistic estimate . _ step 1 : uniform probabilistic bound on @xmath0 $ ] . _ suppose that @xmath96 is a nondecreasing effective service curve , that is @xmath97:\quad pr \bigl\ { \aout(x)\ge \ain\conv\s^\eps(x ) \bigr\ } \ge 1-\eps\ .\ ] ] we will show that then , for any choice of @xmath94 , @xmath98 : \aout(x)\ge \ain\conv\s^\eps(x - a ) \bigr\ } \ge 1-\eps \frac{t}{a}\ .\ ] ] to see this , fix @xmath94 , set @xmath99 , and consider the events @xmath100 where @xmath101 is the smallest integer no larger than @xmath102 . let @xmath103 $ ] be arbitrary , and let @xmath104 the largest integer with @xmath105 , so that @xmath106 . if @xmath107 occurs , then @xmath108 where we have used the fact that @xmath109 is nondecreasing in the last step . it follows that @xmath110 : \aout(x)\ge \ain\conv\s^\eps(x - a ) \bigr\ } & \ge & pr \bigl\ { \forall j = 1 , \ldots , n:\ \aout(x_j)\ge \ain\conv\s^\eps(x_j ) \bigr\ } \\ & = & pr\bigl\ { \bigcap_{0<j\le n } e_j\bigr\}\\ & \ge & 1-n\eps\ , \label{eq - stronger-4}\end{aligned}\ ] ] which proves eqn . ( [ eq - stronger-2 ] ) . thus , the assumptions of the theorem imply that @xmath111 : & \aout^{h}(x ) \ge \ain^h \conv \s^{h,\eps}\conv \delta_(x ) \ , \quad & h < h \\ & \aout^{h}(t ) \ge \ain^h \conv\s^{h,\eps}(t)\ , \quad & h = h \end{array}\right\ } \ge 1- \eps\left(1 + ( h-1)\frac{t}{a}\right)\ . \label{eq - proof - statnet - x-0}\ ] ] _ step 2 : a deterministic argument . _ suppose that , for a particular sample path , @xmath112 : & \aout^{h}(x ) \ge \ain^h \conv \s^{h,\eps}\conv \delta_(x ) \ , \quad & h < h\ , \\ & \aout^{h}(t ) \ge \ain^h \conv\s^{h,\eps}(t)\ , \quad & h = h\ . \end{array}\right.\ ] ] inserting the first line of eqn . ( [ eq - proof - statnet - x-1 ] ) with @xmath113 into the second line yields @xmath114 } \bigl\ { \inf_{y \in [ 0 , x ] } \bigl\ { \ain^{h-1}(t - x - y ) + \bigl(\s^{h-1,\eps}\conv\delta_a\bigr)(y ) \bigr\ } + \s^{h,\eps}(x ) \bigr\}\\ & = & \ain \conv \bigl(\s^{h-1,\eps } \conv \s^{h,\eps}\conv \delta_{a}\bigr ) ( t)\ .\end{aligned}\ ] ] an induction over the number of nodes shows that eqn . ( [ eq - proof - statnet - x-1 ] ) implies that @xmath115 and @xmath116 satisfy @xmath117 _ step 3 : conclusion . _ we estimate @xmath118 the first inequality follows from the fact that eqn . ( [ eq - proof - statnet - x-1 ] ) implies eqn . ( [ eq - proof - statnet - x-5 ] ) . the second inequality merely uses eqn . ( [ eq - proof - statnet - x-0 ] ) . @xmath90 since the bound in eqn . ( [ eq - statnet - intervals - x ] ) deteriorates as @xmath17 becomes large , theorem [ theorem - statnet - intervals - x ] is of limited practical value . to explain why eqn . ( [ eq - statnet - intervals - x ] ) deteriorates , consider a network as shown in figure [ fig - net ] , with @xmath119 nodes . an effective service curve @xmath120 in the sense of eqn . ( [ eq - eff - servicecurve ] ) at the second node guarantees that , for any given time @xmath17 , the departures from this node are with high probability bounded below by @xmath121 } \bigl\ { \ain^2(t-\tau)+\s^{2,\eps}(\tau ) \bigr\ } \ .\ ] ] suppose that the infimum in eqn . ( [ eq - tech - issue-1 ] ) is assumed at some value @xmath122 . since the departures from the first node are random , even if the arrivals to the first node satisfy the deterministic bound @xmath33 , @xmath123 is a random variable . an effective service curve @xmath124 at the first node guarantees that for any arbitrary but fixed time @xmath125 , the arrivals @xmath126 to the second node are with high probability bounded below by @xmath127 since @xmath123 is a random variable , we can not simply evaluate eqn . ( [ eq - tech - issue-2 ] ) for @xmath128 and use the resulting bound in eqn . ( [ eq - tech - issue-1 ] ) . furthermore , there is , a priori , no time - independent bound on the distribution of @xmath123 . note that the above issue does not arise in the deterministic calculus , since deterministic service curves make service guarantees that hold for all values of @xmath125 . we conclude that , in a probabilistic setting , additional assumptions are required to establish time - independent bounds on the range of the infimum , and , in that way , obtain probabilistic network service curves that do not deteriorate with time . one example of such an assumption is to add the condition that @xmath129 } \bigl\{\ain(t - x ) + \seff(x ) \bigr\ } \right\ } \geq 1 - \eps \ .\ ] ] this condition imposes a limit on the range of the convolution . the condition can be satisfied for a given effective service curve @xmath62 and arrival envelope @xmath33 by choosing @xmath130 such that @xmath131 , which guarantees that @xmath132 } \bigl\ { \ain(t - x)+\seff(x)\bigr\}\ .\ ] ] [ theorem - statnet - intervals - w ] assume that all hypotheses of theorem [ theorem - statnet - intervals - x ] are satisfied , and additionally , that there exists a number @xmath133 such that @xmath134 and @xmath91 satisfy eqn . ( [ eq - eff - servicecurve - t ] ) for @xmath92 . then , for any choice of @xmath94 , @xmath135 is an effective network service curve , with violation probability bounded by @xmath136 more precisely , @xmath137 satisfies @xmath138 } \bigl\{\ain(t - x)+ \s^{net,{\eps'}}(x ) \bigr\ } \right\ } \geq 1 - \eps'\ .\ ] ] the bounds of this network service curve deteriorate with the number of nodes @xmath52 , but , different from theorem [ theorem - statnet - intervals - x ] , the bounds are not dependent on @xmath17 . rather the bounds depend on a time scale @xmath130 as used in eqn . ( [ eq - eff - servicecurve - t ] ) . a key issue , which is not addressed in this paper , relates to establishing @xmath130 for an arbitrary node in the network . * the proof is analogous to the proof of theorem [ theorem - statnet - intervals - x ] , and proceeds in the same three steps . _ step 1 : uniform probabilistic bounds on intervals of length @xmath139 . _ suppose that @xmath62 is a nondecreasing effective service curve satisfying eqn . ( [ eq - eff - servicecurve - t ] ) , and let @xmath140 . fix @xmath94 , set @xmath141 , and consider the events @xmath142 } \bigl\ { \ain(x_j - y)+\s^\eps(y)\bigr\}\right\}\ , \quad j = 0 , \ldots , n-1 , \ ] ] where @xmath101 . if @xmath143 and the event @xmath107 occurs , then a computation analogous to eqn . ( [ eq - stronger-2a ] ) shows that @xmath144 } \bigl\ { \ain(x - y)+ \bigl(\s^\eps\conv\delta_a(y)\bigr)\bigr\}\ .\ ] ] we conclude as in eqs . ( [ eq - stronger-3])-([eq - stronger-4 ] ) that @xmath145 : \aout(x)\ge \ \inf_{y\in [ 0,t+a ] } \bigl\{\ain(x - y ) + \bigl(\seff\conv\delta_a\bigr)(y ) \bigr\}\right\ } & \ge & pr\bigl\ { \bigcap_{0\le j < n } e_j\bigr\}\\ & \ge & 1-\eps\left\lceil\frac{\ell}{a}\right\rceil\ .\end{aligned}\ ] ] _ step 2 : a deterministic argument . _ a computation analogous to eqs . ( [ eq - proof - statnet - x-2])-([eq - proof - statnet - x-5 ] ) shows that @xmath146 : & \aout^{h}(x ) \ge \displaystyle{\inf_{y\in [ 0,t+a ] } } \bigl\{\ain^h ( x - y ) + \bigl(\s^{h,\eps}\conv\delta_a\bigr)(y ) \bigr\ } \quad & h < h,\\ & \aout^{h}(t ) \ge \displaystyle{\inf_{y\in [ 0 , t+a ] } } \bigl\{\ain^h ( t - y ) + \s^{h,\eps}(y ) \bigr\ } & h = h\ , \end{array}\right.\ ] ] implies @xmath147 } \{\ain ( t - x ) + \s^{net,{\eps ' } } ( x ) \ } \ .\ ] ] _ step 3 : conclusion . _ combining steps 1 and 2 , we obtain @xmath148 } \{\ain ( t - x ) + \s^{net,{\eps'}}(x ) \ } \bigr\ } & \ge & pr \bigl\ { \mbox{eqn.~(\ref{eq - proof - concat-3 } ) is satisfied } \bigr\ } \label{eq - proof - concat-8}\\ & \ge & 1- \eps\left(1 + \sum_{h=1}^{h-1 } \left\lceil\frac{(h\!-\!h)t}{a}\right\rceil\right ) \label{eq - proof - concat-9}\\ & \ge & 1- \eps'\ . \end{aligned}\ ] ] here , eqn . ( [ eq - proof - concat-8 ] ) follows from step 2 , and eqn . ( [ eq - proof - concat-9 ] ) follows from the assumptions by choosing @xmath149 for @xmath150 in step 1 . we next define a class of effective service curves where the range of the infimum is bounded independently of time , and then give conditions under which these service curves are also effective service curves in the sense of eqn . ( [ eq - eff - servicecurve ] ) . the resulting effective service curves are valid without adding assumptions on a specific arrival distribution or service discipline . within this context , we obtain an effective network service curve , where the convolution formula has a similarly simple form as in the deterministic network calculus . we define a modified convolution operator by setting , for any @xmath151 , @xmath152 the essential property of this modified operator is that the range over which the infimum is taken is limited to the interval @xmath153 $ ] . note that the function @xmath154 depends on the backlog at time @xmath155 as well as on the arrivals in the interval @xmath156 $ ] . it can be written equivalently as @xmath157 the usual convolution operator is recovered by setting @xmath158 . we now reconsider the definition of a service curve in a deterministic regime . we introduce a revised definition of a ( deterministic ) service curve , which is presented in @xcite , and is referred to as _ adaptive service curve _ in @xcite . a _ ( minimum ) adaptive service curve _ is defined as a function @xmath39 which specifies a lower bound on the service given to a flow such that , for all @xmath159 , with @xmath160 , @xmath161 a maximum adaptive service curve can be defined accordingly .. in our context we set @xmath162 . ] ( [ eq - def - intervalservice ] ) is equivalent to requiring that @xmath39 satisfies eqn . ( [ eq - def - service-1 ] ) for the time - shifted arrivals and departures @xmath163 figure [ fig - delay_backlog ] illustrates the time - shifted arrivals . many service curves with applications in packet networks , such as shapers , schedulers with delay guarantees , and rate - controlled schedulers such as gps , can be expressed in terms of adaptive service curves . by setting @xmath164 , one can see that each adaptive service curve is a service curve . however , the converse does not hold @xcite . we next define a _ ( minimum ) @xmath139-adaptive service curve _ , denoted by @xmath165 , as a function for which eqn . ( [ eq - def - intervalservice ] ) is satisfied whenever @xmath166 . if @xmath167 , we obtain an _ adaptive service curve _ , and drop the superscript in the notation . the difference between a service curve according to eqn . ( [ eq - def - service-1 ] ) and an @xmath139-adaptive service curve is that the former involves arrivals over the entire interval @xmath0 $ ] , while the latter uses information about arrivals and departures in intervals @xmath156 $ ] whose length does not depend on @xmath17 . performing a time shift as in eqn . ( [ eq - timeshift ] ) and applying theorem [ theorem - detnet ] shows that the convolution of @xmath139-adaptive service curves yields an @xmath139-adaptive network service curve . the following lemma shows that for @xmath139 sufficiently large , but finite , an @xmath139-adaptive service curve is a service curve in the sense of eqn . ( [ eq - def - service-1 ] ) . in particular , the conclusions of theorems [ theorem - detcalc ] and [ theorem - detnet ] hold for such service curves . suppose that the arrival function @xmath61 of a flow has arrival envelope @xmath33 . let @xmath165 be an @xmath139-adaptive service curve . if @xmath168 : \ \ a^*(t)\le \smin^\ell(t ) \ , \ ] ] then @xmath169 is an adaptive service curve for intervals of arbitrary length . in particular , @xmath39 satisfies eqn . ( [ eq - def - service-1 ] ) for all @xmath170 . [ lemma - sufficient ] the proof of the lemma is given in the appendix . next we introduce a probabilistic version of the @xmath139-adaptive service curve . we define an _ effective @xmath139-adaptive service curve _ to be a nonnegative function @xmath171 such that @xmath172 for all @xmath173 with @xmath174 . if @xmath167 , we call the resulting function an _ effective adaptive service curve _ , and drop the superscript . note that the infimum in the convolution on the right hand side of eqn . ( [ eq - eff - intervalservice ] ) ranges over an interval of length at most @xmath139 . with this bound on the range of the infimum , we derive the following effective network service curve . technically , @xmath139 plays a similar role as the bound @xmath130 on the range of the convolution in eqn . ( [ eq - eff - servicecurve - t ] ) . * concatenation of effective @xmath139-adaptive service curves . * [ theorem - statnet - intervals - y ] consider a flow passing through nodes numbered @xmath175 , and assume that , at each node , an effective @xmath139-adaptive service curve is given by a nondecreasing function @xmath176 . then the function @xmath177 is an effective @xmath139-adaptive network service curve for any choice of @xmath94 , with violation probability bounded by @xmath178 * proof . * we need to show that , for any @xmath179 with @xmath174 and any choice of the parameter @xmath180 , we have @xmath181 performing a time shift as in eqn . ( [ eq - timeshift ] ) , we may assume without loss of generality that @xmath158 and @xmath182 $ ] . the claim now follows immediately from theorem [ theorem - statnet - intervals - x ] . @xmath90 even though the concatenation formula in theorem [ theorem - statnet - intervals - y ] results in a significant improvement over theorem [ theorem - statnet - intervals - x ] , a drawback of theorem [ theorem - statnet - intervals - y ] is that the construction of the network service curve results in a degradation of the violation probability @xmath183 and introduces a time shift @xmath184 , which grow significantly when @xmath139 and @xmath52 become large . to avoid this successive degradation of the service guarantees , we further strengthen the effective service curve . we define a _ strong effective adaptive service curve _ for intervals of length @xmath139 to be a function @xmath185 which satisfies for any interval @xmath186 of length @xmath139 , @xmath187 \subseteq i_\ell : \ \aout(t_0,t ) \geq \ain\conv_{t_0 } \t^{\ell,\eps}(t ) \bigr\ } \ge 1-\eps\ . \label{eq - eff - strongeffservice}\ ] ] this definition differs from the definition of an effective service curve in eqn . ( [ eq - eff - servicecurve ] ) in two ways : it uses the modified convolution operator , and it provides lower bounds on the departures simultaneously in all subintervals of an interval @xmath186 . with the strong effective adaptive network service curve , we obtain a probabilistic version of a network service curve , with a similar concatenation formula as in the deterministic calculus . this is the content of the following theorem . * concatenation of strong effective adaptive service curves . * [ theorem - statnet - intervals ] consider a flow that passes through @xmath52 network nodes in series . assume that the functions @xmath188 define strong effective adaptive service curves for intervals of length @xmath139 at each node ( @xmath92 ) . then @xmath189 is a strong effective adaptive service curve for intervals of length @xmath139 . note the similarity of the convolution formula in eqn . ( [ eq - statnet - i1 ] ) with the corresponding expression in the deterministic calculus . thus , in the statistical calculus , obtaining a statistical end - to - end service curve via a simple convolution operation comes at the price of significant modifications to the definition of a service curve . * proof . * we need to show that @xmath190 satisfies for any interval @xmath186 of length @xmath139 @xmath191\subseteq i_\ell:\ \aout(t_0,t ) \ge \ain\conv_{t_0 } \t^{net,\ell , h\eps}(t ) \bigr\ } \ge 1-h\eps\ .\ ] ] the argument closely follows steps 2 and 3 from the proof of theorem [ theorem - statnet - intervals - x ] . if , for a particular sample path , @xmath192\subseteq i_\ell,\ \forall h=1,\dots , h:\quad \aout^{h}(t_0,x ) \ge \ain^h \conv_{t_0}\t^{h,\ell,\eps}(x ) \ , \ ] ] then , for any fixed @xmath156\subseteq i_\ell$ ] , the time - shifted arrivals and departures defined by eqn . ( [ eq - timeshift ] ) satisfy @xmath193 by step 2 of proof of theorem [ theorem - statnet - intervals - x ] , this implies @xmath194 reversing the time shift and using that @xmath156 \subseteq i_\ell$ ] was arbitrary , we arrive at @xmath195 \subseteq i_\ell : \ \aout(t ) \ge \ain \conv_{t_0 } \bigl(\t^{1,\ell,\eps } \conv \dots \conv \t^{h,\ell,\eps}\bigr ) ( t)\ .\ ] ] we conclude that @xmath196\subseteq i_\ell:\ \aout(t_0,t ) \ge \ain\conv_{t_0 } \t^{net,\ell , h\eps}(t ) \bigr\ } \\ & & \qquad \ge pr \bigl\{\forall [ t_0,x]\subseteq i_\ell , \,\forall h=1,\dots h:\ \aout^h(t_0,x ) \ge \ain^h\conv_{t_0 } \t^{h,\ell,\eps}(x ) \bigr\ } \\ & & \qquad \ge 1-h\eps\ .\end{aligned}\ ] ] the first inequality follows from the definition of @xmath197 and the fact that eqn . ( [ eq - proof - strong - net-1 ] ) implies eqn . ( [ eq - proof - strong - net-5 ] ) , and the second inequality uses the defining property of strong effective @xmath139-adaptive service curves . @xmath90 a comparison of the definition of the strong effective adaptive service curve in eqn . ( [ eq - eff - strongeffservice ] ) with eqn . ( [ eq - eff - intervalservice ] ) shows that a strong effective adaptive service curve is an effective @xmath139-adaptive service curve which provides service guarantees simultaneously on all subintervals of an interval of length @xmath139 . a comparison of theorem [ theorem - statnet - intervals ] with theorems [ theorem - statnet - intervals - x ] , [ theorem - statnet - intervals - w ] , and [ theorem - statnet - intervals - y ] shows that the more stringent strong effective adaptive service curve expresses the statistical calculus more concisely . therefore , unless additional assumptions are made on the arrival processes and the service curves , the network calculus with strong effective adaptive service curves offers the preferred framework . our next result shows how to construct a strong effective adaptive service curve from an effective adaptive service curve . the lemma indicates that the choice of working with a strong effective adaptive service curve rather than an effective adaptive service curve is purely a matter of technical convenience . [ lemma - construct - strong ] if @xmath171 is a nondecreasing function which defines an effective @xmath139-adaptive service curve for a flow , then , for any choice of @xmath94 , the function @xmath198 is a strong effective service curve for intervals of length @xmath139 , with violation probability given by @xmath199 * proof . * we will show that for any interval @xmath186 of length @xmath139 , @xmath200\subseteq i_\ell:\quad pr\bigl\ { \aout(t_0,t)\ge \ain\conv_{t_0}\s^{\ell,\eps}(t)\bigr\}\ge 1-\eps\ ] ] implies @xmath187\subseteq i_\ell:\ \aout(t_0,t)\ge \ain\conv_{t_0}\t^{\ell,{\eps'}}(t)\bigr\}\ge 1-\eps'\ , \ ] ] where @xmath183 and @xmath201 are as given in the statement of the lemma . by performing a suitable time shift as in eqn . ( [ eq - timeshift ] ) , we may assume without loss of generality that @xmath202 $ ] . the strategy is similar to the construction of strong effective envelopes from effective envelopes in @xcite , and uses the same techniques as the first step in the proof of theorem [ theorem - statnet - intervals - x ] . we first use the fact that the departures satisfy the positivity assumption ( a1 ) to translate service guarantees given on a subinterval into a service guarantee on a longer interval . in the second step , we establish probabilistic bounds for the departures simultaneously in a finite number of subintervals @xmath203 of @xmath186 , and then bound the departures in general subintervals of @xmath186 from below in terms of the departures in the @xmath203 . _ step 1 : a property of the modified convolution . _ let @xmath5 be a nondecreasing function , let @xmath204\subseteq [ t_0,t]$ ] , and @xmath205 . then @xmath206 implies @xmath207 to see this , note that eqn . ( [ eq - proof - convert-1 ] ) implies that either @xmath208 or @xmath209 } \{\ain(t_1,\tau)+g(t_2-\tau)\}\ .\ ] ] if eqn . ( [ eq - proof - convert-3 ] ) holds , then @xmath210 proving eqn . ( [ eq - proof - convert-2 ] ) in this case . we have used that @xmath5 is nondecreasing in the last inequality . if eqn ( [ eq - proof - convert-4 ] ) holds , then @xmath211 } \{\ain(t_1,\tau)+g(t_2-\tau)\ } \label{eq - proof - convert-5}\\ & = & b(t_0 ) + \inf_{\tau\in [ t_1,t_2 ] } \{\ain(t_0,\tau)+g(t_2-\tau)\ } \label{eq - proof - convert-6}\\ & \ge & b(t_0 ) + \inf_{\tau\in [ t_0,t ] } \{\ain(t_0,\tau)+\bigl(g\conv\delta_a)(t_2-\tau)\ } \label{eq - proof - convert-7}\\ & \ge & \ain\conv_{t_0 } \bigl(g\conv\delta_a ) ( t)\ .\end{aligned}\ ] ] which proves eqn . ( [ eq - proof - convert-2 ] ) in the second case . in eqn . ( [ eq - proof - convert-5 ] ) we have used eqn . ( [ eq - proof - convert-4 ] ) . in eqn . ( [ eq - proof - convert-6 ] ) , we have used that @xmath212 and taken @xmath213 under the infimum . ( [ eq - proof - convert-7 ] ) uses the monotonicity of @xmath5 and extends the range of the infimum . _ step 2 : uniform probabilistic bounds on @xmath186 . _ fix @xmath94 , set @xmath214 , and consider the intervals @xmath215\ , \quad 0\le i < j < n\ , \ ] ] where @xmath216 is the smallest integer no less than @xmath217 . consider the events @xmath218 let @xmath156\subseteq [ 0,\ell]$ ] be arbitrary , and choose @xmath219 $ ] be as large as possible . if @xmath220 occurs , we apply step 1 with @xmath221 and @xmath222 , and use that @xmath223 to see that @xmath224 it follows that @xmath225\subseteq [ 0,\ell]:\ \aout(t_0,t)\ge \ain\conv_{t_0 } \bigl(\s^{\ell,\eps}\conv\delta_a\bigr)(t)\bigr\ } \nonumber\\ & & \qquad \ge pr \bigl\ { \bigcap_{0\le i < j < n } e_{ij } \bigr\ } \label{eq - construct - proof-2}\\ & & \qquad \ge 1- n^2\eps/2\ , \label{eq - construct - proof-3}\end{aligned}\ ] ] as claimed . here , eqn . ( [ eq - construct - proof-2 ] ) uses step 2 , and eqn . ( [ eq - construct - proof-3 ] ) uses the definition of @xmath171 . @xmath90 we next show that the adaptive versions of the effective service curve can yield effective service service curves in their original definition . this , however , requires us to add appropriate assumptions on the traffic at a node . the following lemma gives a sufficient conditions for an effective @xmath139-adaptive curve to be an effective service curve in the sense of eqn . ( [ eq - eff - servicecurve ] ) . combining theorem [ theorem - statnet - intervals - y ] with lemma [ lemma - eff - sufficient - y ] yields an effective network service curve , which by theorem [ theorem - statcalc ] guarantees probabilistic bounds on output , backlog , and delay . [ lemma - eff - sufficient - y ] let @xmath171 be a nondecreasing function which defines an effective @xmath139-adaptive service curve for a flow with arrival process @xmath61 , and an 1 . if @xmath226:\ b(t_0)=0 \bigr\}\ge 1-\eps_1\ ] ] for all @xmath227 , then , for any choice of @xmath94 , @xmath228 is an effective adaptive service curve for intervals of arbitrary length , with violation probability @xmath229 . in particular , @xmath230 for all @xmath63 . if the arrival process @xmath61 has arrival envelope @xmath33 and @xmath231 for all @xmath232 , then @xmath233 is an effective adaptive service curve for intervals of arbitrary length , with violation probability @xmath234 . in particular , @xmath235 for all @xmath232 . the lemma should be compared with lemma [ lemma - sufficient ] , as both provide sufficient conditions under which service guarantees on intervals of a given finite length imply service guarantees on intervals of arbitrary length . while the condition on @xmath139 in eqn . ( [ eq - def - t0 ] ) involves only the deterministic arrival envelope and the service curve , eqs . ( [ eq - eff - sufficient - y1 ] ) and ( [ eq - eff - sufficient - y2 ] ) represent additional assumptions on the backlog process . this points out a fundamental difference between the deterministic and the statistical network calculus . * the proof consists of three steps . if eqn . ( [ eq - eff - sufficient - y2 ] ) holds , the first step can be omitted . in the first step , we modify a given effective @xmath139-adaptive service curve to give uniform probabilistic lower bounds on the departure of all intervals of the form @xmath156 $ ] , where @xmath17 is fixed and @xmath236 $ ] . this is analogous to the first step in the proof of theorem [ theorem - statnet - intervals - x ] . the second step contains a deterministic argument . we conclude with a probabilistic estimate . _ step 1 : uniform probabilistic bounds . _ suppose that @xmath237 is a nondecreasing effective @xmath139-adaptive service curve , that is , for any @xmath170 , @xmath238:\quad pr \bigl\ { \aout(t_0,t)\ge \ain\conv_{t_0}\s^{\ell,\eps}(t ) \bigr\ } \ge 1-\eps\ .\ ] ] we will show that then , for any choice of @xmath94 , @xmath239 : \aout(t_0,t)\ge \ain\conv_{t_0}\s^{\ell,\eps}(t - a ) \bigr\ } \ge 1-\eps \ell / a\ .\ ] ] to see this , assume without loss of generality that @xmath240 , and consider the events @xmath241 if @xmath107 occurs , we have for @xmath242 by the first step of the proof of lemma [ lemma - construct - strong ] ( with @xmath243 , @xmath244 , @xmath245 ) , that @xmath246 it follows that @xmath247 : \aout(t)\ge \ain\conv_{t_0}\s^{\ell , \eps}(t - a ) \bigr\ } \nonumber\\ % & & \qquad \ge pr \bigl\ { \forall i=0,\dots , n-1 : \ % \aout(x_j)\ge \ain\conv_{x_j}\s^{\ell , \eps}(x_j ) \bigr\ } \\ & & \qquad = pr\bigl\ { \bigcap_{0 \le i < n } e_j\bigr\}\\ & & \qquad \ge 1-n\eps\ , \end{aligned}\ ] ] which proves eqn . ( [ eq - stronger-6 ] ) in the case @xmath240 . _ step 2 : deterministic argument . _ fix @xmath170 , and suppose that for a particular sample path , we have @xmath248:\ \aout(x , t)\ge \ain\conv_{x } \t^{\ell,\eps}(t)\ , \ ] ] and either @xmath249 : \ b(t_0)=0 \ , \mbox{or } \\ \label{eq - proof - eff - suff-2b } 2 . & & b(t-\ell)\le \s^{\ell,\eps } ( \ell)-a^*(\ell ) \ , \mbox { where $ a^*$ is an arrival envelope . } \end{aligned}\ ] ] in the first case , we can set @xmath250 in eqn . ( [ eq - proof - eff - suff-1a ] ) to obtain @xmath251 in the second case , we note that from eqn . ( [ eq - proof - eff - suff-2b ] ) it follows that @xmath252 which implies that @xmath253 inserting this into eqn . ( [ eq - proof - eff - suff-1a ] ) with @xmath254 yields again eqn . ( [ eq - proof - eff - suff-3 ] ) . _ step 3 : probabilistic estimate . _ if eqn . ( [ eq - eff - sufficient - y1 ] ) holds , we use step 2 to see that @xmath255:\ \aout(x , t)\ge \ain\conv_{x}\t^{\ell,\eps}(t ) , \ \mbox{and}\ \exists t_0 \in [ t-\ell , t ] : \ b(t_0)=0 \end{array } \right\ } \\ & & \qquad \ge 1-(\eps \ell / a+\eps_1)\ , \label{eq - eff - servicecurve - ell-1}\end{aligned}\ ] ] where we have used the result of step 2 in the second line . if in eqn . ( [ eq - eff - sufficient - y2 ] ) holds , we have by step 2 @xmath256 where we have used the definition of @xmath171 in the second line . @xmath90 note that eqs . ( [ eq - eff - servicecurve - ell-1 ] ) and ( [ eq - eff - servicecurve - ell-2 ] ) supply time - independent bounds on the range of the convolution , of the form given in eqn . ( [ eq - eff - servicecurve - t ] ) . a similar , but simpler result holds for strong effective @xmath139-adaptive service curves : [ lemma - eff - sufficient ] given a flow with arrival process @xmath61 , and a strong effective adaptive service curve @xmath185 on intervals of length @xmath139 . assume that for every @xmath170 , either eqn . ( [ eq - eff - sufficient - y1 ] ) or or eqn . ( [ eq - eff - sufficient - y2 ] ) is satisfied . then , for any @xmath257 and any @xmath258 , @xmath259 in particular , for @xmath260 , @xmath261 is an effective service curve in the sense of eqn . ( [ eq - eff - servicecurve ] ) . * proof . * we need to show that under the assumptions of the lemma , we have for any @xmath63 and any @xmath258 @xmath262 by considering time - shifted arrivals and departures as in eqn . ( [ eq - timeshift ] ) , we may assume without loss of generality that @xmath263 . the first step in the proof of lemma [ lemma - eff - sufficient - y ] shows that @xmath264\subset [ t-\ell , t]:\ \aout(x , y)\ge \ain\conv_{x}\t^{\ell,\eps}(x ) , \ \\ \mbox{and}\ \bigl\ { \mbox { either } \ \exists t_0 \in [ t-\ell , t ] : \ b(t_0)=0 \ \\ \qquad \mbox{or } \ b(t-\ell)\le \t^{\ell,\eps } ( \ell)-a^*(\ell ) \bigr\ } \end{array } \right\ } \label{eq - proof - eff - sufficient-1}\\ & & \qquad \ge 1-(\eps+\eps_1)\ , \label{eq - proof - eff - sufficient-2}\end{aligned}\ ] ] as claimed . we have presented a network calculus with probabilistic service guarantees where arrivals to the network satisfy a deterministic arrival bound . we have introduced the notion of _ effective service curves _ as a probabilistic bound on the service received by individual flows in a network . we have shown that some key results from the deterministic network calculus can be carried over to the statistical framework by inserting appropriate probabilistic arguments . we showed that the deterministic bounds on output , delay , and backlog from theorem [ theorem - detcalc ] have corresponding formulations in the statistical calculus ( theorem [ theorem - statcalc ] ) . we have extended the concatenation formula of theorem [ theorem - detnet ] for network service curves to a statistical setting ( theorems [ theorem - statnet - intervals - x ] , [ theorem - statnet - intervals - w ] , [ theorem - statnet - intervals - y ] , and [ theorem - statnet - intervals ] ) . we showed that a modified effective service curve , called _ strong effective adaptive service curve _ yields the simplest concatenation formula . in order to connect the different notions of effective service curves , we have made an additional assumption on the backlog in lemmas [ lemma - eff - sufficient - y ] and [ lemma - eff - sufficient ] . the results in this paper showed that a multi - node version of the statistical network calculus requires us to make assumptions that limit the range of the convolution operation when concatenating effective service curves . such limits on a ` maximum relevant time scale ' , can follow from assumptions on the traffic load ( as in theorems [ theorem - statnet - intervals - w ] , lemma [ lemma - eff - sufficient - y ] and lemma [ lemma - eff - sufficient ] ) , or from appropriately modified service curves . while the question is open whether one can dispense with these additional assumptions , we have made an attempt to justify the need for them . we thank rene cruz for pointing out problems in earlier versions of the effective network service curve . the authors gratefully acknowledge the valuable comments from chengzhi li . 10 r. agrawal , r. l. cruz , c. okino , and r. rajan . performance bounds for flow control protocols . , 7(3):310323 , june 1999 . m. andrews . probabilistic end - to - end delay bounds for earliest deadline first scheduling . in _ proceedings of ieee infocom 2000 _ , pages 603612 , tel aviv , march 2000 . f. l. baccelli , g. cohen , g. j. olsder , and j .- quadrat . . john wiley and sons , 1992 . r. boorstyn , a. burchard , j. liebeherr , and c. oottamakorn . statistical service assurances for traffic scheduling algorithms . , 18(12):26512664 , december 2000 . j. y. le boudec . application of network calculus to guaranteed service networks . , 44(3):10871097 , may 1998 . j. y. le boudec and p. thiran . . springer verlag , lecture notes in computer science , lncs 2050 , 2001 . c. s. chang . stability , queue length , and delay of deterministic and stochastic queueing networks . , 39(5):913931 , may 1994 . c. s. chang . on deterministic traffic regulation and service guarantees : a systematic approach by filtering . , 44(3):10971110 , may 1998 . c. s. chang . springer verlag , 2000 . r. l. cruz . a calculus for network delay , part i : network elements in isolation . , 37(1):114121 , 1991 . r. l. cruz . a calculus for network delay , part ii : network analysis . , 37(1):132141 , january 1991 . r. l. cruz . quality of service guarantees in virtual circuit switched networks . , 13(6):10481056 , august 1995 . r. l. cruz . quality of service management in integrated services networks . in _ proceedings of the 1st semi - annual research review , cwc _ , ucsd , june 1996 . r. l. cruz and c. okino . service gurantees for flow control protocols . in _ proceedings of the 34th allerton conference on communications , control and computating _ , october 1996 . a. elwalid and d. mitra . design of generalized processor sharing schedulers which statistically multiplex heterogeneous qos classes . in _ proceedings of ieee infocom99 _ , pages 12201230 , new york , march 1999 . j. kurose . on computing per - session performance bounds in high - speed multi - hop computer networks . acm sigmetrics92 _ , pages 128139 , 1992 . c. li and e. knightly . coordinated network scheduling : a framework for end - to - end services . in _ proceedings of ieee icnp 2000 _ , osaka , november 2000 . j. qiu and e. knightly . inter - class resource sharing using statistical service envelopes . in _ proceedings of ieee infocom 99 _ , pages 3642 , march 1999 . v. sivaraman and f. m. chiussi . statistical analysis of delay bound violations at an earliest deadline first scheduler . , 36(1):457470 , 1999 . v. sivaraman and f. m. chiussi . providing end - to - end statistical delay guarantees with earliest deadline first scheduling and per - hop traffic shaping . in _ proceedings of ieee infocom 2000 _ , pages 603612 , tel aviv , march 2000 . d. starobinski and m. sidi . stochastically bounded burstiness for communication networks . , 46(1):206212 , 2000 . o. yaron and m. sidi . performance and stability of communication networks via robust exponential bounds . , 1(3):372385 , june 1993 . * appendix * let @xmath165 be an @xmath139-adaptive service curve . we need to show that @xmath265 holds for all @xmath266 with @xmath258 . by considering the time - shifted arrivals and departures as in eqn . ( [ eq - timeshift ] ) , we may assume without loss of generality that @xmath263 . consider intervals @xmath267 $ ] , where @xmath268 is an integer . we will show by induction , that for any integer @xmath268 , @xmath269 applying the definition of an @xmath139-adaptive network service curve with @xmath202 $ ] , we see that eqn . ( [ eq - suff-0 ] ) clearly holds for @xmath270 . for the inductive step , suppose that eqn . ( [ eq - suff-0 ] ) holds for some integer @xmath268 . fix @xmath271 , and let @xmath272 . by the inductive assumption , @xmath273 . ( [ eq - def - intervalservice ] ) says that either @xmath274 or @xmath275 if eqn . ( [ eq - suff-1 ] ) holds , then @xmath276 in eqn . ( [ eq - suff-4 ] ) , we have used eqn . ( [ eq - suff-1 ] ) and the inductive assumption . in eqn . ( [ eq - suff-5 ] ) , we have used that @xmath277 and pulled @xmath278 out of the infimum . in eqn . ( [ eq - suff-6 ] ) , we have inserted @xmath279 , used the assumption that @xmath280 , and extended the range of the infimum . if eqn . ( [ eq - suff-2 ] ) holds , then @xmath281 in eqn . ( [ eq - suff-7 ] ) , we have used eqn . ( [ eq - suff-2 ] ) , and the fact that @xmath282 . in eqn . ( [ eq - suff-8 ] ) , we have taken @xmath283 under the infimum and used that @xmath284 . in eqn . ( [ eq - suff-9 ] ) , we have extended the range of the infimum and used the definition of the convolution .

we present wide - field photometry for about 11,500 stars in the low - metallicity cluster ngc 5466 . we have detected the red giant branch bump for the first time , although it is at least 0.2 mag fainter than expected relative to the turnoff . the number of red giants ( relative to main sequence turnoff stars ) is in excellent agreement with stellar models from the yonsei - yale and teramo groups , and slightly high compared to victoria - regina models . this adds to evidence that an abnormally large ratio of red giant to main - sequence stars is not correlated with cluster metallicity . we discuss theoretical predictions from different research groups and find that the inclusion or exclusion of helium diffusion and strong limit coulomb interactions may be partly responsible . we also examine indicators of dynamical history : the mass function exponent and the blue straggler frequency . ngc 5466 has a very shallow mass function , consistent with large mass loss and recently - discovered tidal tails . the blue straggler sample is significantly more centrally concentrated than the hb or rgb stars . we see no evidence of an upturn in the blue straggler frequency at large distances from the center . dynamical friction timescales indicate that the stragglers should be _ more _ concentrated if the cluster s present density structure has existed for most of its history . ngc 5466 also has an unusually low central density compared to clusters of similar luminosity . in spite of this , the specific frequency of blue stragglers that puts it right on the frequency cluster relation observed for other clusters .

0704.0827

the luminosity function ( lf ) is an observational tool used for analyzing the post - main sequence evolutionary phases of low - mass ( @xmath2 0.5 - 0.8@xmath3 ) metal - poor stars in galactic globular clusters ( ggc ) . because of their age and richness , ggc typically contain hundreds of stars that have evolved off the main sequence . the numbers of stars in evolved phases are directly related to the evolutionary timescales and fuel consumed in each phase @xcite , so that they present us with an opportunity to test this aspect of stellar evolution models . the results of the most stringent tests have been mixed . repeated studies of the metal - poor cluster m30 @xcite have found an excess number of red giant branch ( rgb ) stars relative to main sequence ( ms ) stars . @xcite also uncovered an apparent excess of stars in a combined lf of the metal - poor clusters m68 , ngc 6397 , and m92 . however , the lfs of more metal - rich clusters show no discrepancy ( m5 : sandquist et al . 1996 ; m3 : rood et al . 1999 ; m12 : hargis , sandquist , & bolte 2004 ) . in a survey of 18 clusters , @xcite found good agreement with model predictions with the possible exception of clusters at the high metallicity end . in this paper we present @xmath0 photometry of ngc 5466 , a high galactic latitude globular cluster ( @xmath4 and @xmath5 ) , located in the constellation of botes ( @xmath6 , @xmath7 at a distance of r=15.9 kpc ; harris 1996 ) . ngc 5466 is a loose cluster ( @xmath8 ) with extremely low metallicity ( [ fe / h]@xmath9 ) and subject to little or no reddening , @xmath10 @xcite . in 2 , we describe the process leading to the calibrated photometry , and compare with previous studies of the cluster . in 3 , we compare the observed color - magnitude diagram and observed luminosity function with theoretical models , focusing on the relative number of stars on the lower rgb and around the ms turnoff . finally , in [ bss ] , we present a new examination of the blue straggler population of ngc 5466 . the data used in this study were obtained with the kitt peak national observatory ( kpno ) 0.9 m telescope ( @xmath11 pix@xmath12 ) on the nights of ut dates 1995 may 4 , may 5 , and may 9 . a complete list of the image frames , exposure times , and observing conditions is given in table [ tab31 ] . the images obtained on the three nights were processed using iraf tasks and packages . the reduction involved subtraction and trimming of the overscan region of all images , subtraction of a master bias frame from flats and object frames , and flat fielding of the object frames using images taken at twilight . profile - fitting photometry was done using the daophotii / allstar programs @xcite . we also reduced archival ground - based photometry of the cluster core taken with the high - resolution camera ( hrcam ) on the 3.6 m canada - france - hawaii telescope ( cfht ) . the @xmath13 and @xmath14 images were taken 30 and 31 may 1992 ( observers j. heasley and c. christian ) , and have not previously been described in the literature . the ccd images had 1024 @xmath15 1024 pixels , 013 per pixel , and excellent seeing ( 0.4 - 0.5 arcsec ) . the images were reduced using the archived bias and twilight flat frames , and following a procedure similar to that for the kpno data . this allowed us to get excellent photometry to 2 magnitudes below the turnoff in the cluster core . these images were used entirely for blue straggler identification ( see [ bss ] ) . the conditions at kpno on 1995 may 9 were photometric , and @xcite standard star fields were observed at a range of air masses to determine photometric transformation coefficients . the standard values used for the calibration were chosen from the large compilation of @xcite , which is set to be on the same photometric scale as the earlier @xcite values . we conducted photometry on the standard stars and isolated cluster stars using multiple synthetic apertures . we then used the daogrow @xcite program to construct growth curves to extrapolate measurements to a common aperture size . using the ccdstd program , the standard star transformation equations were found to be : @xmath16 @xmath17 @xmath18 where @xmath19 is the airmass , @xmath20 , @xmath21 and @xmath22 are instrumental magnitudes , and @xmath13 , @xmath23 , and @xmath14 are standard magnitudes . these calibration equations are different than those used for our analysis of m10 @xcite and m12 @xcite , which were observed on the same night , because our @xmath24band exposures did not go as deep as the @xmath23 and @xmath13 exposures . as a result , the @xmath25 color was a better choice for calibrating the @xmath13 photometry down to our faintest observed stars . the calibrated measurements for the standard stars are compared with catalogue values in fig . [ fig311 ] . we note that there is slight evidence of trend in the residuals for the @xmath14 band versus magnitude , which might indicate nonlinearity . this impression is caused by one observation of the pg1323 - 086 field . we did , however , have an additional observation of the same field on the same night having the same exposure time that does not show the same ( small ) trend . because we do not have any reason to eliminate the frame and because its elimination has a minimal effect on the transformation coefficients , we have decided to retain the measurements from the image . aperture photometry for 165 cluster stars was used to calibrate the point - spread function ( psf ) photometry for the cluster . these secondary standard stars were chosen based on relatively low measurement errors and location in relatively uncrowded regions of the cluster . they were chosen from the asymptotic giant branch ( agb ) , upper rgb , and horizontal branch ( hb ) of the cluster in order to cover the entire range of colors covered by cluster stars . the psf - fitting photometry for the three nights of data was combined and averaged after zero - point differences among the frames had been determined and corrected . the zero - point corrections to the standard system were determined after fixing the color - dependent terms at the values measured in the primary standard star calibration . ( this was also done in our studies of m10 and m12 . ) in fig . [ fig312 ] , it can be seen that this procedure does not introduce systematic color- or magnitude - dependent errors . we compared our photometric data set to those of @xcite , @xcite , and @xcite . the magnitude and color comparisons ( @xmath0 in this study versus @xmath26 in @xcite and @xcite , and @xmath27 in @xcite ) as function of magnitude and color are shown in figs . [ fig315 ] [ fig313 ] . though our calibrated magnitudes are slightly brighter than those of @xcite , the differences are small , and there is no color trend . the offsets compared to the @xcite are larger , but again there are no clear color trends . the offsets compared to the @xcite data are also significant , but more notable are slight trends with color . artificial star tests were performed to empirically measure the precision of our photometry and to correct for incompleteness in the detection of stars . we followed the procedure described by @xcite for the calculation of incompleteness corrections as function of position and magnitude . the inputs used for producing the artificial star tests were the reduced @xmath23 and @xmath13 ccd frames , psfs for each object frame , fiducial lines , and an estimate for the initial lf @xcite . artificial stars were randomly placed in cells on a spatial grid and the entire grid was then shifted randomly from run to run in order to ensure the whole imaged field was tested @xcite . each star was placed in a consistent position relative to the cluster center on each image . if a detected star was found to coincide with the input position of an artificial star , it was added to the archive . the new images were reduced using the same procedure applied to the original data set . in this study , a total of 100,000 artificial stars from 50 separate runs were added . the number of artificial stars per trial was chosen so that the effects of crowding on the photometry was qualitatively unchanged . the recovered artificial stars were used to calculate 1 ) median magnitude and color biases ( @xmath28 and @xmath29 , where @xmath30 $ ] ) , 2 ) median external error estimates ( @xmath31/0.6745 $ ] and @xmath32 ) , and 3 ) total recovery probabilities ( @xmath33 , which is the fraction of the stars that were recovered with any output magnitude ) in bins according to projected radius and magnitude . the values for the above quantities are plotted in figs.[fig613 ] [ fig615 ] . finally , an initial estimate of the `` true '' lf and the error distribution , magnitude biases and the total recovery probability ( @xmath34 ) were used to compute the completeness fraction @xmath35 ( the ratio of the predicted number of stars to the actual number of observed stars ) . the completeness fraction results are shown in figure [ fig616 ] . we then interpolated to compute @xmath35 for the radial distance and magnitude of each detected star . for each observed star @xmath36 was added to the appropriate magnitude bin to determined the observed lf . ( note that the completeness fraction was set to 1.0 for star brighter than the turnoff . ) the observed lf along with the upper and lower 1 @xmath37 error bars on log @xmath38 are listed in table [ tab51 ] . because ngc 5466 resides at high galactic latitude , it suffers little if any reddening . though @xcite found a reddening of @xmath39 from the maps of dust ir emission , we adopted @xmath40 @xcite . for our interests in this paper , the small difference is of small importance . most of the comparisons below between observations and theory are _ relative _ , in which reference points ( like the turnoff ) are used to determine magnitude and/or color shifts . this has the benefit of minimizing the influence of uncertainties in reddening and distance modulus ( see below ) . as for abundances , there is only one high - resolution measurement for a cluster star , and it is for the anomalous cepheid v19 . @xcite find [ fe / h]@xmath41 , while @xcite find [ fe / h]@xmath42 using the same data . typically quoted metallicity values include [ fe / h ] @xmath43 @xcite from photoelectric photometry of integrated light in selected filter bands , and [ fe / h ] @xmath44 , which was derived by @xcite ( ultimately from low - resolution spectral scans by @xcite ) . when converted to the widely - used metallicity scale of @xcite , this becomes [ fe / h ] @xmath45 . more work could certainly be done on the composition of ngc 5466 stars , but the evidence so far points to an abundance [ fe / h ] @xmath46 . though the range in the above quoted metallicity values is relatively large for a globular cluster , the exact value is not critical for our purposes since we will primarily be concerned with _ relative _ comparisons . our photometry does not extend faint enough to derive a new distance modulus from subdwarf fitting to the main sequence . @xcite obtained @xmath47 by calibrating the observed luminosity level of the horizontal branch with the relation @xmath48}+0.80 $ ] and adopting a reddening , @xmath49=0.0 and a metallicity , [ fe / h]=@xmath50 . @xcite determined distance moduli @xmath51 from their zero - age hb estimate , assuming no reddening and metallicity on the @xcite scale . we will consider distance moduli in this range . most previous age estimates of ngc 5466 s age have it older than the recent determination of the age of the universe ( @xmath52 gyr ) obtained by the wilkinson microwave anisotropy probe ( wmap ) team @xcite . recent homogeneous studies of ggc indicate that ngc 5466 is coeval with clusters of similar metallicity @xcite . as a result , we will primarily consider ages in the range of 12 to 13 gyr . the color - magnitude diagrams ( cmds ) for ngc 5466 show well - defined rgb , agb , and hb sequences ( see fig . [ fig411 ] ) , and stars extending from the tip of the rgb down to @xmath53 . fiducial sequences for the ms and lower rgb were determined from the mode of the color distribution of stars in magnitude bins . the sgb position was determined using the magnitude distribution of the stars in color bins . the fiducial line for the rest of the rgb was obtained from the mean color of stars in magnitude bins . the fiducial points are listed in table [ tab34 ] . a comparison of the fiducial points derived for ngc 5466 with theoretical isochrones for a range of ages from the teramo @xcite , victoria - regina @xcite , and yonsei - yale @xcite groups is displayed in figures [ fig412 ] and [ rich ] . the isochrones have been shifted in color and magnitude ( aligning the turnoff colors and the magnitudes of the main sequence point 0.05 mag redder than the turnoff ) according to the technique of @xcite . this has the advantage of removing some of systematic uncertainties associated with the color-@xmath54 transformations . [ in our comparisons , we found that the yonsei - yale models could not match the fiducial line for any reasonable set of input parameters when the transformations of @xcite were used . therefore , we only utilize models using the green , demarque , & king ( 1987 ) transformations . even then , we could not find a match with the slope of the upper giant branch for reasonable metallicities ( [ fe / h ] @xmath55 ) . ] on the whole , the shape of the fiducial matches the models well on the main sequence , subgiant branch , and lower giant branch . neither the teramo nor the victoria - regina models include element diffusion processes , while the yonsei - yale isochrones only include he diffusion . however , differences in @xmath54-color transformations are likely to be the cause of some of the differences seen . the number of stars at a given luminosity in post - main - sequence phases is directly proportional to the lifetime spent at that luminosity . it is well known that the lf of the rgb probes the chemical stratification inside a star because the hydrogen abundance being sampled by the thin hydrogen - fusion shell affects the rate of evolution , and hence star counts on the rgb @xcite . setting aside the short pause at the rgb bump , rgb evolution accelerates in a very regular way that is ultimately related to the structure of degenerate core . the relationship between core mass and radius forces the fusion shell to function at strictly controlled density and temperature conditions , which leads to a relationship between core mass and luminosity . this causes the lf to be particularly sensitive to certain physical details , which we discuss in [ rgbms ] below . [ fig617 ] shows the observed luminosity function compared to theoretical lfs for the labeled values of metallicity , exponent of the initial mass function , and a range of age estimates for ngc 5466 , assuming our preferred distance modulus @xmath56 . the theoretical models were normalized to the observed lf at @xmath57 ( sufficiently faint that stellar evolution effects are minimized ) . the models agree well with the observed lf , implying an age of approximately 12 - 13 gyr . @xcite recently discussed theoretical luminosity functions calculated by different groups . one of the primary differences they noted was in the number of giant stars relative to main sequence stars . in order to show these differences in a parameter - independent way , we follow the method of @xcite . in fig . [ fig619 ] , the theoretical lfs were shifted so that a point on the main sequence 0.05 mag redder than the turnoff color point matched the corresponding point on the cluster fiducial line . the reason for using the point ( @xmath58 ) rather than the msto itself is that the ms has a significant slope and curvature at this point , making it possible to accurately measure the point in both observational data and isochrones @xcite . the theoretical models were normalized to the two bins in the observed lf on either side of the turnoff ( @xmath59 and 20.13 ) . as can be seen , age - related differences between the theoretical lfs nearly disappear when this procedure is applied ( see also @xcite ) . however , when different sets of models are compared , there are small differences in the number of rgb stars relative to ms stars , with the victoria - regina models predicting the smallest number of giants and the yonsei - yale models predicting the largest . to quantify the differences , we computed the ratio of the number of stars on the lower giant branch to the number of stars near the main sequence . @xcite introduced this ratio and showed that it is insensitive to age and heavy - element abundance . for the main sequence population , we used star counts in the two bins on either side of the turnoff ( @xmath60 ) , and for the red giant branch we used the counts in the range @xmath61 . we derived model values from the same magnitude ranges _ relative _ to the @xmath58 point on the main sequence . values are compared in table [ tabrat ] . the error in the observed value is dominated by poisson statistical scatter . the yonsei - yale models are in best agreement with the observations , the victoria - regina models are out of agreement by more than 2 @xmath37 , and the teramo models are in between . it is worth examining the possible causes of this difference both because it may help improve the physics inputs for the models and because red giant stars are some of the largest contributors to the integrated light of old stellar populations . @xcite tabulated most of the main physics inputs for the most widely - used model sets . earlier studies @xcite have shown that the lf - shifting method used above eliminates nearly all sensitivity to model input parameters like mass function , age , convective mixing length , and composition inputs with the exception of helium abundance , which we examine first . older models ( e.g. fig . 9 of ratcliff 1987 , fig . 7 of stetson 1991 , fig . 3 of @xcite ) seem to agree that an increase in _ initial _ helium abundance @xmath62 in non - diffusive models by 0.1 results in an increase in the relative number of stars on the rgb ( more precisely , a reduction in the relative number of main sequence stars ) by about 0.07 - 0.08 in @xmath63 . the teramo models ( @xmath64 ) predict about 12% more giant stars relative to main sequence stars compared to the victoria - regina models ( @xmath65)-element enhancement ( [ @xmath66/fe ] = 0.4 ) than the victoria - regina models ( [ @xmath66/fe ] = 0.3 ) , which would tend to _ reduce _ the number of giants relative to main sequence stars . however , because the relative number of rgb and ms stars is not sensitive to small differences in heavy element abundance , this difference is probably unimportant . ] . this difference corresponds to a shift of 0.05 in @xmath63 , which is about an order of magnitude too large for the helium abundance difference . the yonsei - yale models have the lowest assumed helium abundance ( @xmath67 ) , but are the only set of the three that include helium diffusion . the inclusion of helium diffusion reduces the age derived from the turnoff of a globular cluster by about 10 - 15% @xcite , thanks to the inward motion of helium . according to theoretical models ( e.g. fig . 8 of @xcite ) , diffusion has a small effect on the lfs ( @xmath68 a few times @xmath69 in @xmath63 , increasing for increasing age ) , but it does increase the number of giants relative to ms stars . he diffusion reduces the total core hydrogen fuel supply available to an ms star , but in itself this does not strongly modify the lf , just changes the brightness of the turnoff . this magnitude change is eliminated in our lf shifting procedure . the chemical composition profile left in the star after it leaves the main sequence has a greater impact . diffusion reduces the h abundance in the fusion regions , thereby decreasing the evolutionary timescale . according to the models of @xcite , the changes to the he abundance profile are most considerable immediately below the surface convection zone , and just outside the nuclear fusion regions ( where the composition gradient slows the inward settling of helium ) . however , over most of the star , the changes in @xmath62 are limited to 0.01 - 0.02 . because the core portion of the composition profile is consumed on the subgiant branch , the evolution timescales for giant stars are only affected in a minor way , and the appearance of a deep convection zone on the lower giant branch wipes out most of the effects of diffusion for the upper giant branch evolution . in spite of this , the yonsei - yale models have almost 23% more giants than the victoria - regina models , and more than 9% more giants than the teramo models ( relative to ms stars ) . the lower helium abundance in the yonsei - yale models compared to the other models should partially counteract what effects helium diffusion might have had on the rgb / ms ratio as well . thus , it appears that neither he abundance nor he diffusion can completely explain the differences between the yonsei - yale models and the other groups . we can ask whether the lfs show similar disagreements at other metallicities . @xcite made comparisons between the victoria - regina and yonsei - yale theoretical lfs and observational lfs for the clusters m3 @xcite , m5 @xcite , m12 , and m30 . the overall impression from those comparisons was again that the yonsei - yale models ( having he diffusion ) predict more giant stars relative to main - sequence stars than do the victoria - regina models . in fig . [ otherlfs ] , we compare the lfs for these clusters with the teramo models . the degree of agreement or disagreement can be quantified with number ratios of lower rgb and msto stars , similar to the ones we computed earlier for comparisons with ngc 5466 . our calculations are shown in table [ tabrat ] . as can be seen , uncertainties in the metallicity scale have some effect on the comparisons with the observations . the @xcite scale has higher [ fe / h ] values than the @xcite scale , and thus results in lower number ratios . [ rats ] shows the results of comparing the observed ratios with the models for different [ fe / h ] scales . on the @xcite scale ( right panels ) , the observed values seem to be in agreement to within about @xmath70 for most of the models , with the exception of the the lowest metallicity clusters ( m30 and ngc 5466 ) and the lowest helium abundance ( victoria - regina ) models . on the @xcite scale , the yonsei - yale models have the best overall agreement , although the teramo models only deviate noticeably for the lowest metallicity clusters . the victoria - regina models predict too few giants for all of the clusters . the differences from model to model ( as opposed to models versus observation ) point toward deficiencies elsewhere in the physics or computational algorithms used in the stellar evolution codes . the rgb lf is a robust prediction of the models because there is a strong core mass luminosity relationship : the conditions in the hydrogen fusion shell of the giant are strongly dependent on the structure of the degenerate core and are almost independent of the details of the mass or structure of the envelope . as a result of this , we can focus on factors affecting core structure . ( as a non - standard physics example , @xcite describe the way in which core rotation relieves a giant star of some of the need to support itself by gas pressure , which reduces the core temperature and lengthens the evolutionary timescale . ) because model - to - model differences appear even on the faint end of the giant branch , we can set aside factors that only become important to the structure of the star near the tip of the giant branch [ such as neutrino losses and conductive opacities ; see @xcite , for example ] , even though there are significant theoretical uncertainties in these quantities . nuclear reaction rates in the fusion shell can also be neglected , partly because the uncertainties in the reactions appear to be relatively small @xcite , but also because small changes in the reaction rates require only tiny changes in the shell temperature to get the same energy production . this leads us to examine the equation of state ( eos ) in the core . although the behavior of degenerate electrons is thought to be very well understood , their interactions with nuclei can have a measureable effect on the pressure . particles of like charge tend to cluster together , which modifies the free energy of the gas and reduces the gas pressure for given density and temperature . @xcite looked at the effects of the inclusion of coulomb interactions on giant stars , and their results are corroborated by those of @xcite . they found that for a given core mass the fusion shell temperature was higher when the coulomb interactions were included , which leads to faster processing of hydrogen . thus this is another example ( like core rotation ) where modification of the pressure support of the core affects the evolutionary timescales , which results in changes to the luminosity function . the coulomb corrections to the pressure become more important with increasing density for the core , but are small compared to the contribution of the degenerate electrons . all of the model sets we have considered here incorporate coulomb interactions in some form . the teramo group used the most sophisticated `` eos1 '' version of the freeeos , which incorporates coulomb corrections in a form that matches limits in both the weak ( debye - hckel ) and strong ( once - component plasma ) coulomb interaction limits as well as ( less importantly ) electron exchange interactions . the strong interaction limit is most relevant for giant star cores since the strong interaction parameter @xmath71 where @xmath72 is the rms nuclear charge and @xmath73 is the average internuclear distance . the yonsei - yale group used the opal eos tables @xcite , but falls back on the group s older eos [ e.g. @xcite ] for conditions for high densities and temperatures outside the opal tables . while the most recent opal tables probably contain the most complete physical description of the coulomb effect , the opal tables they used ( y .- c . kim , private communication ) were computed prior to recent improvements to account for relativistic electrons @xcite , and as a result cut off at @xmath74 . the yale eos at higher densities _ only _ includes the coulomb effect in the weak debye - hckel limit , which for the highest densities in the core . the victoria - regina models also use a modified version of the eff eos @xcite , with a correction for coulomb corrections in the weak debye - hckel limit @xcite . the differences in the implementation of the coulomb effect may explain the fact that the teramo models generally predict more giants ( relative to the main sequence ) than the victoria - regina models do . however , the smaller coulomb corrections in the yonsei - yale models would tend to result in fewer giants than the teramo models ( although the effects of helium diffusion work in the opposite direction ) . so , we are unable to completely reconcile the differences in the luminosity functions from the three groups . obviously more detailed study is needed by all of the modelling groups to identify the causes of the differences , but such a study is beyond the scope of this paper . still , we believe that helium diffusion and strong interaction coulomb corrections are physical effects that should be considered first . there is , for example , good evidence from helioseismology for helium diffusion in the sun @xcite , despite the surface convection and meridional flow ( e.g. , @xcite ) . it is expected that helium diffusion should also act in globular cluster stars . a detailed study of the effect of equation of state uncertainties has yet to be done [ see , for example , @xcite for a study of uncertainties in other physical inputs ] . use of freeeos would make a study of equation of state effects most stringent since it appears to be capable of modelling the most sophisticated tabular eos ( opal ) , while also having the flexibility to allow individual bits of the physics to be `` turned off '' . as a final warning about the observations , we should remember the lf of the cluster m10 . @xcite found that unusual variations in numbers of rgb stars at different brightness levels in m10 ( a virtual twin to m12 ) . in particular there seemed to a significant excess in the number of stars near the rgb bump in brightness , while the lower rgb appeared normal ( compared to victoria - regina and yonsei - yale models ) . a similar excess may be present in the rgb lf of m13 @xcite . these kinds of variations can not be explained by the `` global '' physics that should apply to all globular cluster stars . these anomalies point toward fluctuations in the stellar initial mass function or composition - dependent effects . a second feature of the lf presented here is a noticeable rgb bump . typically , the rgb bump appears as a peak in the differential lf and as a change of slope in the cumulative lf ( clf ) . the bump provides a measure of the maximum depth reached by the outer convection zone during first dredge - up since it is the result of a pause in the star s evolution when the shell fusion source begins consuming material of constant , lower helium content @xcite . unfortunately , the number of stars occupying the bump gets smaller and the luminosity of the bump increases as the metallicity of the cluster decreases , making the bump harder to detect in metal - poor clusters . a small peak appears in our differential lf at @xmath75 , and a significant ( @xmath76 ) change in slope occurs at the same position as the peak in the differential lf , as shown in fig . [ fig621 ] . the relative brightness of the bump can be measured by comparing to the @xmath13-magnitude of the hb at the level of the rr lyrae instability strip @xmath77 @xcite . this indicator is a function of the total metallicity and the age of the cluster : an increase in metallicity and/or a decrease in age are accompanied by a decrease in luminosity of the bump @xcite . we find @xmath78 mag , and @xmath79 ( from interpolation between the average magnitudes of non - variable hb stars at the blue and red ends of the rr lyrae instability strip ) , giving @xmath80 . in the compilation of @xcite , a zero - age hb reference point was calculated using the relation @xmath81}^2 + 0.236\mbox{[fe / h]}+0.193 $ ] . @xcite found @xmath82 , which is consistent with the value obtained here ( @xmath83 ) . our value of @xmath84 is considerably lower than tabulated values for other clusters with similar metallicities ( m68 : @xmath85 ; m92 : @xmath86 ; m15 : @xmath87 ) . in a separate compilation , @xcite measured a smaller value @xmath88 for m15 , in better agreement with the value for ngc 5466 . ( the difference is primarily because @xcite measured the bump position to be 0.16 mag fainter than @xcite . ) clearly there is still some need for more precise comparisons of bumps in metal - poors clusters with theory . we believe , however , that the brightness of the bump should ultimately be judged using _ hydrogen - fusing _ stars as references because it avoids any effects of the poorly - understood physical processes ( such as the helium flash and/or mass loss ) associated with the creation of an hb star . using the cluster lf ( as seen in figure [ fig619 ] ) , we again find that the observed rgb bump is fainter than model values by at least 0.3 mag when the models are shifted to match the cluster s main sequence . @xcite did similar comparisons between theoretical models and luminosity functions for m3 , m5 , m12 , and m30 taken from the literature . with the exception of m30 ( because the bump could not be identified ) , the position of the bump relative to the turnoff region agreed well with theory . ngc 5466 is thus the most metal - poor cluster this comparison has been done for . so at present we are left with the question of whether this might result from the cluster s low metallicity , or whether we have been the unfortunate victims of a fluctuation in the number of giant stars in this low - mass cluster . we therefore encourage the examination of the luminosity function of more massive metal - poor clusters to settle the question . two recent papers @xcite reported the discovery of tidal streams covering many degrees around ngc 5466 in sloan digital sky survey images . @xcite examined the milky way globular clusters and found ngc 5466 to be a cluster that has probably been strongly affected by disk shocking in the recent past . in our examination of the lf , we found a rather low value for the global main sequence mass function slope . the mass function for a cluster is typically expressed as a power law ( @xmath89 ) , where the slope @xmath90 is the standard salpeter value . generally , the present - day power - law index @xmath91 varies from cluster to cluster . the mass function slopes that best fit the upper lf of ngc 5466 around and above the msto have @xmath92 . ( note that the best fit slope does depend on the models being used : the yonsei - yale models require a flatter slope than the victoria - regina and teramo models . ) such a shallow mass function slope is unusual for a metal - poor cluster . for example , the cluster ngc 5053 has similar metallicity , position relative to the galactic center and plane , and density structure , but still has a steep @xmath93 mass function @xcite . @xcite found that mass function slopes in the range of @xmath94 are influenced primarily by the cluster s position in the galaxy , and to some extent by cluster metallicity . based on both of those factors , ngc 5466 should have a larger mass function slope ( @xmath95 according to the multivariate formula in @xcite ) . like other halo clusters , ngc 5466 s orbit is quite eccentric and will take the cluster more than 30 kpc away from the galactic center @xcite , but it is currently on its way back into the halo after two relatively recent passes through the galactic disk . recent losses of low - mass stars may explain the recent identification of strong tidal tails near the cluster by @xcite . blue stragglers ( bsss ) were first identified by @xcite in the globular cluster m3 . these stars are more massive than the turnoff mass and occupy the space in the cmd just bluer and brighter than the msto . blue stragglers are found in clusters , and relatively more frequently in lower - luminosity clusters @xcite . from the various models proposed for the formation mechanism of bsss , the `` collision '' theory ( involving strong gravitational interactions between previously unassociated single or binary stars ) and the `` mass - transfer '' theory ( in which the more massive star in a binary evolves and during its expansion transfers mass to its companion ) are the strongest possibilities . there is a continuing interest in the study of bsss because they may provide insight into the recent dynamical history of a cluster . in order to identify bsss over the entire observed area of the cluster out to a radius of 116 , we used photometry from three datasets . in the core of the cluster we used the cfht data presented here for the first time . outside of the cfht field , we used the @xmath27 photometry of @xcite , which covered a field 116 on a side centered on the cluster . finally , we used our kpno data for the least crowded outskirts of the cluster . even though ngc 5466 is a very low density cluster , the spatial resolution of the kpno data was such that blends of stars would have resulted in the spurious identification of 10 objects as bss in the intermediate portion of the field . 48 bss candidates were identified in ngc 5466 by @xcite , all located between @xmath96 and @xmath97 from the cluster center . in spite of the low cluster density , we find new bss candidates at all radii and luminosity levels , and find several of their candidates are spurious . according to the cfht photometry , the object with i d 45 from @xcite is a blend of several fainter stars , none of which is a bss . in addition , ids 6 and 24 were identified as blends of stars using the @xcite dataset . our bss list is presented in table [ tab41 ] . the list includes the nine known sx phoenicis stars ( i d 27 , 29 , 35 , 38 , 39 , and 49 , @xcite ; i d 3 ( sx phe 3 ) , 36 ( sx phe 2 ) , and 50 ( sx phe 1 ) , @xcite ) and the three eclipsing binaries ( i d 19 , 30 , and 31 ; @xcite ) . new straggler candidates were given i d numbers that build upon the @xcite list . figure [ fig555 ] shows the cmds used to select the 94 identified bsss in each of the three datasets . in order to use the bsss to constrain cluster dynamics , we compared the normalized cumulative radial distribution of the bsss to the population of the giant branch , as shown in fig . [ fig511 ] . @xcite found a 97.8% probability that their bss sample was more centrally concentrated than red giants in the same magnitude range . because their photometry was taken in conditions of poorer seeing ( compared to our cfht photometry and that of @xcite ) , their samples are likely to be somewhat incomplete near the cluster center . our rgb sample contains 350 stars with magnitude @xmath98 . kolmogorov - smirnov ( k@xmath99s ) probability tests were used to test the hypothesis that both populations were drawn from the same parent population . the k@xmath99s probability that the bsss are drawn from the same radial distribution as the rgb population is @xmath100 , and @xmath101 for the comparison with the hb population . by contrast there is a probability of 0.27 that the rgb and hb samples are drawn from the same population . the concentration of bsss toward the cluster center as compared to the rgb and hb samples is consistent with the idea that they are more massive than individual rgb stars , and as a result have been segregated by mass deeper within the cluster potential well . @xcite recently used samples of stragglers from the cores of 56 globular clusters to show that there was a strong correlation between @xmath102 and integrated cluster @xmath13 magnitude , and a weaker anti - correlation with central density . @xcite examined an additional 13 low - luminosity globular clusters using similar selection criteria . ngc 5466 is an interesting cluster in relation to these samples because it has an integrated luminosity that puts it at the faint end of the piotto et al . sample ( @xmath103 ; @xcite ) , but with a central density that is nearly an order of magnitude lower than any of their clusters [ @xmath104 but comparable to clusters in the sandquist sample . to put ngc 5466 in the context of the piotto et al . and sandquist samples , we selected a subset of our bsss that satisfied the selection criteria in those studies ( brighter than the msto , and bluer than the msto by 0.05 in @xmath105 color ) . from mode fitting to the turnoff region in the @xcite and cfht data , we find @xmath106 and @xmath107 . a color offset of 0.05 in @xmath105 corresponds to an offset of about 0.075 in @xmath108 @xcite for ngc 5466 s metallicity . we find that 75 bsss are brighter than the cluster turnoff ( @xmath109 ) and 0.05 bluer than the cluster turnoff in @xmath105 . we have identified 97 hb stars in our datasets for ngc 5466 , which gives a specific frequency @xmath110 ( with the error estimate from poisson statistics ) . when compared to the @xcite values ( see fig . [ fbss ] ) , ngc 5466 falls within the general trend versus @xmath111 despite the cluster s low central density . on the other hand , ngc 5466 has a lower @xmath112 value than other clusters of similar central density ( but lower total luminosity ) . as discussed by @xcite , this provides additional evidence that the plateau in @xmath112 seen for clusters with @xmath113 is a result of the correlation between cluster integrated magnitude and central density . the lowest luminosity clusters in the sandquist sample ( e3 and palomar 13 ) have central densities comparable to that of ngc 5466 , but bss frequencies that are several times higher . another moderate - luminosity , low - density cluster ( ngc 5053 ; hiner et al . , in preparation ) similar to ngc 5466 shows a comparably low straggler frequency . bsss produced via purely collisional means are not likely to show this kind of behavior . more likely is the scenario proposed by @xcite in which binary stars that would normally produce bsss are destroyed earlier in the cluster s history . more direct observational support for that hypothesis is needed though for example , a detailed study of the variation of the binary star fractions as a function of integrated cluster magnitude . [ fig512 ] shows that the number and frequency @xmath114 of bsss relative to the integrated @xmath13-band flux ( derived from a king model profile ) as a function of radius . both frequencies increase toward the cluster center , and neither shows signs of rising toward larger radii . as recent studies of denser clusters show ( m3 : @xcite ; m55 : @xcite ; 47 tuc : @xcite ; ngc 6752 : @xcite ) , the bss frequency generally decreases at intermediate radii and rises again at larger distances . however , the cluster palomar 13 @xcite , which has a central density similar to ngc 5466 , shows no sign of an increase in straggler frequency at large distance . in more massive clusters , the minimum in the bss frequency is reached approximately where the timescale for dynamical friction equals the age of the cluster @xcite . a similar calculation for the current structure of ngc 5466 indicates that this occurs at about 270(about @xmath115 ) . this appears to be in the outer reaches of the core straggler distribution . because it is likely that ngc 5466 has lost a significant fraction of its mass , we expect that the current density structure of the cluster has not existed throughout its history and that ngc 5466 might have been able to dynamically relax stragglers to its core from larger distances earlier in its history . this may be showing that the global bss population differs significantly between low - density / low - mass clusters and high - density / high - mass clusters . a lack of stragglers at large distance may be signature of large - scale tidal stripping of the cluster , which would remove both stragglers that would normally have formed in primordial binaries in the outer reaches of the cluster and ones that formed in the core but were given velocity kicks into orbits that would take them into the outer reaches . the case against a rise at large radius in palomar 13 is stronger because the cluster has been surveyed out to 19 core radii , while in ngc 5466 , we have only surveyed out to about 10 core radii ( or 7.5 half - mass radii ) . still , ngc 5466 probably should have an even more concentrated distribution of stragglers if its current density structure has existed for most of its history . in more massive clusters , the secondary rise in straggler frequency is observed between 8 and 10 @xmath116 . unfortunately , further study of the stragglers in ngc 5466 will probably be complicated by the strong tidal tails observed in the cluster . examinations of the luminosity functions of globular clusters continue to produce interesting tests of astrophysics . in this study , we found that ngc 5466 has a luminosity function that is in better overall agreement with theoretical models than the anomalous cluster m30 , which has a similar low metallicity . in addition , we found that the relative numbers of red giant and main sequence stars may produce a fairly sensitive test of the physics near the core of red giants specifically , helium diffusion and coulomb interactions . however , we are not yet able to fully explain the differences between sets of theoretical models . recent discoveries of large tidal tails associated with ngc 5466 suggest that this cluster has been strongly disrupted by interactions with the galaxy . our measured flat ( @xmath117 ) main - sequence luminosity function is unusual for a low - metallicity halo cluster . it is , however , consistent with the emerging picture of mass - segregation followed by tidal stripping . we have thoroughly re - examined the blue straggler population in the cluster , and detected a total of 94 . the radial distribution of stragglers is clearly more centrally concentrated than the rgb and hb populations . the frequency of blue stragglers in the cluster is relatively low consistent with the observed anti - correlation between frequency and cluster luminosity , in spite of the cluster s very low central density . we would like to thank the anonymous referee for helpful comments on the manuscript , y. jeon for providing us with an electronic copy of his photometric dataset , y .- c . kim for information on the yonsei - yale isochrones , and s. cassisi for providing us with access to the teramo set of models . this work has been funded through grants ast 00 - 98696 and 05 - 07785 from the national science foundation to e.l.s . and m.b . adelberger , e. g. , et al . 1998 , rev . 70 , 1265 bahcall , j. n. , pinsonneault , m. h. , & wasserburg , g. j. 1995 , rev . phys . , 67 , 781 behr , b. b. 2003 , , 149 , 67 belokurov , v. , evans , n. w. , irwin , m. j. , hewett , p. c. , & wilkinson , m. i. 2006 , , 637 , l29 bergbusch , p. a. 1996 , , 112 , 1061 bjork , s. r. , & chaboyer , b. 2006 , , 641 , 1102 bolte , m. 1994 , , 431 , 223 bono , g. , cassisi , s. , zoccali , m. , & piotto , g. 2001 , , 546 , l109 burles , s. , nollett , k. m. , & turner , m. s. 2001 , , 552 , l1 carretta , e. , & gratton , r. g. 1997 , , 121 , 95 cassisi , s. , salaris , m. , castelli , f. , & pietrinferni , a. 2004 , , 616 , 498 cassisi , s. , salaris , m. , & irwin , a. w. 2003 , , 588 , 862 cho , d .- h . , lee , s .- g . , jeon , y .- b . , & sim , k. j. 2005 , , 129 , 1922 clark , l. l. , sandquist , e. l. , & bolte , m. 2004 , , 128 , 3019 dantona , f. , bellazzini , m. , caloi , v. , pecci , f. f. , galleti , s. , & rood , r. t. 2005 , , 631 , 868 davies , m. b. , piotto , g. , & de angeli , f. 2004 , , 349 , 129 deglinnocenti , s. , weiss , a. , & leone , l. 1997 , , 319 , 487 demarque , p. , woo , j .- h . , kim , y .- c . , & yi , s. k. 2004 , , 155 , 667 dinescu , d. i. , girard , t. m. , & van altena , w. f. 1999 , , 117 , 1792 djorgovski , s. , piotto , g. , & capaccioli , m. 1993 , , 105 , 2148 eggleton , p. p. , faulkner , j. , & flannery , b. p. 1973 , , 23 , 325 fahlman , g. g. , richer , h. b. , & nemec , j. 1991 , , 380 , 124 ferraro , f. r. , beccari , g. , rood , r. t. , bellazzini , m. , sills , a. , & sabbi , e. 2004 , , 603 , 127 ferraro , f. r. , messineo , m. , fusi pecci , f. , de palo , m. a. , straniero , o. , chieffi , a. , & limongi , m. 1999 , , 118 , 1738 ferraro , f. r. & montegriffo , p. 2000 , , 119 , 1282 ferraro , f. r. , paltrinieri , b. , & cacciari , c. 1999 , , 70 , 599 ferraro , f. r. , fusi pecci , f. , cacciari , c. , corsi , c. , buonanno , r. , fahlman , g. g. , & richer , h. b. 1993 , , 106 , 2324 fusi pecci , f. , ferraro , f. r. , crocker , d. a. , rood , r. t. , & buonanno , r. 1990 , , 238 , 95 gallart , c. , zoccali , m. , & aparicio , a. 2005 , , 43 , 387 gnedin , o. y. , & ostriker , j. p. 1997 , , 474 , 223 grillmair , c. j. , & johnson , r. 2006 , , 639 , l17 guenther , d. b. , demarque , p. , kim , y .- c . , & pinsonneault , m. h. 1992 , , 387 , 372 guhathakurta , p. , webster , z. t. , yanny , b. , schneider , d. p. , & bahcall , j. n. 1998 , , 116 , 1757 hargis , j. r. , sandquist , e. l. , & bolte , m. 2004 , , 608 , 243 harpaz , a. , & kovetz , a. 1988 , , 331 , 898 harris , w. e. 1996 , , 112 , 1487 hathaway , d. h. 1996 , , 460 , 1027 jeon , y. , lee , m. g. , kim , s. , & lee , h. 2004 , , 128 , 287 johnson , j. a. , & bolte , m. 1998 , , 115 , 693 kim , y. , demarque , p. , yi , s. k. , & alexander , d. r. 2002 , , 143 , 499 landolt , a. u. 1992 , , 104 , 340 lejeune , t. , cuisinier , f. , & buser , r. 1998 , , 130 , 65 mateo , m. , harris , h. c. , nemec , j. , & olszewski , e. w. 1990 , , 100 , 469 mccarthy , j. k. , & nemec , j. m. 1997 , , 482 , 203 nemec , j. m. & harris , h. c. 1987 , , 316 , 172 nemec , j. , & mateo , m. 1990 , asp conf . ser . 11 : confrontation between stellar pulsation and evolution , 11 , 64 norris , j. e. 2004 , , 612 , l25 olive , k. a. , & skillman , e. d. 2004 , , 617 , 29 piotto , g. , de angeli , f. , king i. r. , djorgovski , s. g. , bono , g. , cassisi , s. , meylan , g. , recio - blanco , a. , rich , r. m. & davies , m. b. 2004 , , 604 , 109 piotto , g. , & zoccali , m. 1999 , , 345 , 485 pollard , d. l. , sandquist , e. l. , hargis , j. r. , & bolte , m. 2005 , , 628 , 729 preston , g. w. , & sneden , c. 2000 , , 120 , 1014 pritzl , b. j. , venn , k. a. , & irwin , m. 2005 , , 130 , 2140 proffitt , c. r. , & michaud , g. 1991 , , 371 , 584 proffitt , c. r. , & vandenberg , d. a. 1991 , , 77 , 473 pryor c. , mcclure , r. d. , fletcher , j. m. , & hesser , j. e. , 1991 , , 102 , 1026 ratcliff , s. j. 1987 , , 318 , 196 renzini , a. & fusi pecci , f. 1988 , , 26 , 199 rogers , f. j. , & nayfonov , a. 2002 , , 576 , 1064 rogers , f. j. , swenson , f. j. , & iglesias , c. a. 1996 , , 456 , 902 rood , r. t. , et al . 1999 , , 523 , 752 rosenberg , a. , aparicio , a. , saviane , i. , & piotto , g. 2000 , , 145 , 451 sabbi , e. , ferraro , f. r. , sills , a. , & rood , r. t. 2004 , , 617 , 1296 salaris , m. , riello , m. , cassisi , s. , & piotto , g. 2004 , , 420 , 911 salaris , m. , & weiss , a.2002 , , 388 , 492 sandage , a. r. 1953 , , 58 , 61 sandquist , e. l. , bolte , m. , langer , g. e. , hesser , j. e. , & mendes de oliveira , c. 1999 , , 158 , 262 sandquist , e. l. , bolte , m. , stetson , p. b. , & hesser , j. e. 1996 , , 470 , 910 sandquist , e. l. 2005 , , 635 , 73 schlegel , d. j. , finkbeiner , d. p. , & davis , m. 1998 , , 500 , 525 searle , l. , & zinn , r. 1978 , , 225 , 357 spergel , d. n. , et al . 2006 , submitted stetson , p. b. 2000 , , 112 , 925 stetson , p. b. 1991 , asp conf . ser . 13 : the formation and evolution of star clusters , 13 , 88 stetson , p. b. 1990 , , 102 , 932 stetson , p. b. 1987 , , 99 , 191 straniero , o. , chieffi , a. , & limongi , m. 1997 , , 490 , 425 vandenberg , d. a. , bergbusch , p. a. , & dowler , p. d. 2006 , , 162 , 375 vandenberg , d. a. , bolte , m. , & stetson , p. b. 1990 , , 100 , 445 vandenberg , d. a. , & clem , j. l. 2003 , , 126 , 778 vandenberg , d. a. , larson , a. m. , & de propris , r. 1998 , , 110 , 98 vandenberg , d. a. , richard , o. , michaud , g. , & richer , j. 2002 , , 571 , 487 vandenberg , d. a. , swenson , f. j. , rogers , f. j. , iglesias , c. a. , & alexander , d. r. 2000 , , 532 , 430 warren , s. r. , sandquist , e. l. , & bolte , m. 2006 , , 648 , 1026 zaggia , s. r. , piotto , g. , & capaccioli , m. 1997 , , 327 , 1004 zoccali , m. , cassisi , s. , piotto , g. , bono , g. , & salaris , m. 1999 , , 518 , l49 zoccali , m. , & piotto , g.2000 , , 358 , 943 zinn , r. & west , m. j. 1984 , , 55 , 45 zinn , r. 1980 , , 241 , 602 rrrrr 1995 may 4 & @xmath23,@xmath13 & 1,1 & 60 & 1.01,1.12 + 1995 may 4 & @xmath23 & 2 & 300 & 1.03 , 1.03 + 1995 may 4 & @xmath23,@xmath13 & 2,1 & 600 & 1.0,1.11,1.0 + 1995 may 5 & @xmath23,@xmath14 & 2,2 & 300 & 1.03,1.01,1.02,1.01 + 1995 may 9 & @xmath23,@xmath13,@xmath14 & 1,1,1 & 60 & 1.12,1.13,1.14 + rrrr 13.532 & 0.5005 & 0.1761 & 0.3010 + 13.832 & @xmath118 & 0.3010 & 1.0000 + 14.132 & @xmath118 & 0.3010 & 1.0000 + 14.432 & 0.5975 & 0.1605 & 0.2575 + 14.732 & 0.8985 & 0.1193 & 0.1651 + 15.032 & 0.6767 & 0.1487 & 0.2279 + 15.332 & 0.6767 & 0.1487 & 0.2279 + 15.632 & 0.5976 & 0.1606 & 0.2575 + 15.932 & 0.8986 & 0.1193 & 0.1651 + 16.232 & 1.2410 & 0.0839 & 0.1041 + 16.532 & 1.0127 & 0.1063 & 0.1411 + 16.832 & 1.0451 & 0.1029 & 0.1351 + 17.132 & 1.3306 & 0.0764 & 0.0928 + 17.432 & 1.4432 & 0.0678 & 0.0804 + 17.732 & 1.4777 & 0.0728 & 0.0876 + 18.032 & 1.6136 & 0.0630 & 0.0738 + 18.332 & 1.7575 & 0.0540 & 0.0617 + 18.632 & 1.7577 & 0.0540 & 0.0617 + 18.932 & 1.9624 & 0.0433 & 0.0481 + 19.232 & 2.2928 & 0.0301 & 0.0324 + 19.532 & 2.5143 & 0.0236 & 0.0249 + 19.832 & 2.6996 & 0.0192 & 0.0201 + 20.132 & 2.7738 & 0.0178 & 0.0185 + 20.432 & 2.8830 & 0.0159 & 0.0165 + 20.732 & 2.9327 & 0.0151 & 0.0157 + 21.032 & 3.0036 & 0.0142 & 0.0146 + 21.332 & 3.0496 & 0.0137 & 0.0142 + 21.632 & 3.1179 & 0.0132 & 0.0136 + rrr 22.198 & 0.578 & 518 + 21.999 & 0.564 & 598 + 21.798 & 0.549 & 724 + 21.596 & 0.532 & 710 + 21.398 & 0.498 & 673 + 21.200 & 0.470 & 670 + 21.004 & 0.459 & 653 + 20.802 & 0.450 & 614 + 20.600 & 0.431 & 568 + 20.400 & 0.418 & 546 + 20.199 & 0.404 & 470 + 20.001 & 0.395 & 453 + 19.805 & 0.396 & 382 + 19.599 & 0.403 & 273 + 19.400 & 0.436 & 235 + 19.206 & 0.495 & 155 + 19.005 & 0.543 & 84 + 18.823 & 0.586 & 70 + 18.588 & 0.607 & 43 + 18.401 & 0.616 & 52 + 18.204 & 0.621 & 40 + 17.989 & 0.636 & 34 + 17.820 & 0.647 & 31 + 17.586 & 0.654 & 23 + 17.403 & 0.672 & 20 + 17.194 & 0.677 & 19 + 16.988 & 0.693 & 11 + 16.784 & 0.722 & 10 + 16.613 & 0.725 & 6 + 16.400 & 0.752 & 12 + 16.199 & 0.771 & 16 + 16.066 & 0.779 & 5 + 15.825 & 0.809 & 7 + 15.602 & 0.827 & 4 + 15.407 & 0.857 & 6 + 15.111 & 0.917 & 1 + 15.006 & 0.930 & 5 + 14.786 & 0.946 & 2 + 14.590 & 0.986 & 11 + 14.438 & 1.044 & 2 + lcclll ngc 5466 & @xmath60 & @xmath119 & & @xmath120 & + vr & & & 0.235 & 0.132 & @xmath50 + vr & & & 0.235 & 0.130 & @xmath121 + t & & & 0.245 & 0.148 & @xmath50 + t & & & 0.245 & 0.146 & @xmath121 + yy & & & 0.230 & 0.162 & @xmath50 + yy & & & 0.230 & 0.160 & @xmath121 + m3 & @xmath122 & @xmath123 & & @xmath124 & + vr & & & 0.235 & 0.170 & @xmath125 + vr & & & 0.235 & 0.158 & @xmath126 + t & & & 0.246 & 0.181 & @xmath125 + t & & & 0.248 & 0.170 & @xmath126 + yy & & & 0.230 & 0.185 & @xmath125 + yy & & & 0.230 & 0.162 & @xmath126 + m5 & @xmath127 & @xmath128 & & @xmath129 & + vr & & & 0.235 & 0.106 & @xmath130 + vr & & & 0.235 & 0.097 & @xmath131 + t & & & 0.248 & 0.114 & @xmath130 + t & & & 0.251 & 0.106 & @xmath131 + yy & & & 0.230 & 0.119 & @xmath130 + yy & & & 0.230 & 0.105 & @xmath131 + m12 & @xmath132 & @xmath133 & & @xmath134 & + vr & & & 0.235 & 0.144 & @xmath130 + vr & & & 0.235 & 0.118 & @xmath135 + t & & & 0.248 & 0.155 & @xmath130 + t & & & 0.251 & 0.144 & @xmath135 + yy & & & 0.230 & 0.150 & @xmath130 + yy & & & 0.230 & 0.137 & @xmath135 + m30 & @xmath136 & @xmath137 & & @xmath138 & + vr & & & 0.235 & 0.174 & @xmath139 + vr & & & 0.235 & 0.153 & @xmath140 + t & & & 0.245 & 0.173 & @xmath139 + t & & & 0.246 & 0.165 & @xmath140 + yy & & & 0.230 & 0.194 & @xmath139 + yy & & & 0.230 & 0.180 & @xmath140 + rrrrrrrrrrcl + 1 & 137.74 & 32.86 & 19.1477 & & 19.0038 & & & & 605 & j & + 1 & 137.74 & 32.86 & 19.1446 & 0.0122 & 19.0182 & 0.0280 & 18.7606 & 0.0400 & 809 & k & + 2 & @xmath141 & @xmath142 & & & 18.7591 & 0.0116 & 18.5972 & 0.0061 & 10313 & c & + 2 & @xmath141 & @xmath142 & 18.8461 & 0.0147 & 18.7209 & 0.0258 & 18.6245 & 0.0504 & 2176 & k & + 3 & @xmath143 & 14.99 & & & 19.1710 & 0.0100 & 18.9595 & 0.0059 & 9339 & c & sx phe ( 3 ) + 3 & @xmath143 & 14.99 & 19.4373 & 0.0155 & 19.2441 & 0.0296 & 19.1073 & 0.0563 & 2129 & k & + 4 & @xmath144 & 83.56 & 19.2057 & & 19.0546 & & & & 646 & j & + 4 & @xmath144 & 83.56 & 19.2266 & 0.0123 & 19.0548 & 0.0256 & 18.8465 & 0.0460 & 2880 & k & + 5 & @xmath145 & 65.06 & 18.5511 & & 18.2697 & & & & 389 & j & + 5 & @xmath145 & 65.06 & 18.5632 & 0.0121 & 18.2740 & 0.0243 & 17.8309 & 0.0255 & 2895 & k & + 7 & @xmath146 & @xmath147 & 18.8010 & & 18.7046 & & & & 504 & j & + 7 & @xmath146 & @xmath147 & 18.8482 & 0.0113 & 18.7379 & 0.0217 & 18.5538 & 0.0386 & 3289 & k & + 8 & @xmath148 & 3.07 & 18.8436 & & 18.6640 & & & & 498 & j & + 8 & @xmath148 & 3.07 & 18.8726 & 0.0120 & 18.6345 & 0.0234 & 18.3367 & 0.0298 & 3358 & k & + 9 & @xmath149 & 83.38 & 19.1757 & & 19.0256 & & & & 626 & j & + 9 & @xmath149 & 83.38 & 19.1385 & 0.0144 & 18.9711 & 0.0257 & 18.7508 & 0.0387 & 2973 & k & + 10 & @xmath150 & 96.04 & 19.3627 & & 19.1882 & & & & 751 & j & + 10 & @xmath150 & 96.04 & 19.4145 & 0.0139 & 19.2190 & 0.0261 & 18.9077 & 0.0494 & 2520 & k & +

we derive thouless - anderson - palmer ( tap ) equations for quantum disordered systems . we apply them to the study of the paramagnetic and glassy phases in the quantum version of the spherical spin - glass model . we generalize several useful quantities ( complexity , threshold level , etc . ) and various ideas ( configurational entropy crisis , etc ) , that have been developed within the classical tap approach , to quantum systems . the analysis of the quantum tap equations allows us to show that the phase diagram ( temperature - quantum parameter ) of the spin - glass model should be generic . in particular , we argue that a crossover from a second order thermodynamic transition close to the classical critical point to a first order thermodynamic transition close to the quantum critical point is to be expected in a large class of systems . lpt - ens/0033 , lpthe/0034 .

cond-mat0011028

glassy systems of extremely diverse types exist in nature . they all share several common features like a very slow , non - equilibrium dynamics . the development of a full theoretical description of the glassy phase is one of the most important challenges in condensed matter physics . a variety of techniques , that range from scaling arguments to mean - field approaches have been , and are still used , with the aim of attempting a satisfactory description of the glassy properties . one of these techniques is due to thouless , anderson and palmer ( tap ) @xcite , who introduced an approach to classical disordered systems based on the study of a free - energy landscape . the key object is the legendre transform of the free - energy @xmath1 with respect to a number of order parameters that are sufficient to describe the transition and the different phases in the system . this function behaves as an effective potential whose minima represent different possible phases . in a classical fully - connected ising model only one order parameter is needed , the global magnetization @xmath2 . the two possible minima of @xmath3 correspond to the two possible states of positive and negative magnetization , @xmath4 . focusing on the sherrington - kirkpatrick ( sk ) mean - field model for spin - glasses , tap showed that _ all _ the local magnetizations @xmath5 , @xmath6 , have to be included in order to derive the relevant free - energy landscape . the extremization condition of the tap free - energy on the @xmath7s leads to the tap equations . it was soon after realized by bray and moore @xcite that the number of solutions to the tap equations for the sk model is exponential in the number of spins in the system , for temperatures below the spin - glass transition @xcite . a very useful alternative derivation of the tap equations was given by plefka @xcite who showed that these equations can also be obtained from a power expansion of the gibbs potential up to second order in the exchange couplings . the advantage of this derivation is twofold : it allows to show convergence of the power expansion for all temperatures and it is easily applicable to other mean - field glassy models . moreover , georges and yedidia @xcite showed that the high temperature series , at fixed order parameter , of the free - energy can be used to derive tap - like equations , and its corrections , for models in finite dimensions or , equivalently , with finite range interactions . the connection between the tap approach and the more standard analysis of the partition function of a disordered model has been exhibited by de dominicis and young @xcite who showed that , for the sk model , one recovers the equilibrium results of the replica or the cavity method @xcite via weighted boltzmann averages over solutions of the tap equations . more recently , the tap approach has been applied to other classical disordered models . in particular , two models that we shall discuss in the following , the spherical and ising @xmath0 spin - glass models @xcite and the ghatak - sherrington ( gs ) model @xcite have been analyzed with this method @xcite . glassy systems , and in particular disordered ones , are characterized by having a very slow dynamics with non equilibrium effects at low temperatures @xcite . mean - field models , like the the spherical @xmath0 spin - glass model @xcite or the sk spin - glass @xcite , capture this phenomenology . the dynamic solution for the evolution starting from random initial conditions , that represent a quench from high temperatures analytically , is intimately connected to the structure and organization of tap solutions . one of the most striking results of the dynamic analysis of @xmath0 spin - glass - like models is that the energy density ( and other one time - quantities ) converges asymptotically to the energy density of high lying solutions of the tap equations . this level has been called _ threshold_. the energy - density in equilibrium is different . this and other related results suggest that an interpretation of the dynamics in terms of a motion in a tap free - energy landscape can be given @xcite . the generalization of the tap approach to dynamics that has been developed in @xcite allows one to make this statement precise : the evolution is determined by a gradient - descent in the tap free - energy landscape with the most important addition of non - markovian terms . usually , glasses can be analyzed with a fully classical approach since their transition temperatures are rather high . nevertheless , in many cases of great interest the critical temperature can be lowered by tuning another external parameter and quantum fluctuations become very important . this is the case for the insulating magnetic compound liho@xmath8y@xmath9f@xmath10 , that is an experimental realization of a quantum spin - glass , and presently receives much attention @xcite . other examples where glassy properties in the presence of quantum fluctuations have been observed are mixed hydrogen bonded ferro - antiferro electric crystals @xcite , interacting electron systems @xcite , cuprates like la@xmath11sr@xmath8cuo@xmath10 @xcite , amorphous insulators @xcite , etc . the quantum fluctuations in liho@xmath8y@xmath9f@xmath10 can be controlled by tuning the strength of an external field that is transverse to the preferred direction of the randomly located magnetic impurities . after a series of experiments presented in @xcite the authors conclusions are : ( 1 ) the samples undergo a paramagnetic to spin - glass transition in the @xmath12 plane , where @xmath13 and @xmath14 is the strength of the transverse field . ( 2 ) the transition is of second order ( in the thermodynamic sense ) close to the classical critical point @xmath15 but crosses over to first order close to the quantum critical point @xmath16 . ( 3 ) the system undergoes out of equilibrium dynamics in the glassy phase as demonstrated by the fact that the dynamics strongly depends on the preparation of the sample for all subsequent times explored experimentally . the theoretical study of quantum spin - glasses started with bray and moore s analysis of the equilibrium properties of the fully connected heisenberg model @xcite . in this article , bray and moore introduced a path - integral representation in imaginary time of the partition function that they analyzed with the replica trick . many articles on the equilibrium of this , and related , mean - field models have been published since @xcite . the static properties of low dimensional models have been studied and it has been shown that , in finite dimensions , griffiths - mccoy singularities are very important close to the quantum critical point @xcite . in all these models , the transition from the paramagnetic to the spin - glass phase has been reported to be of second order throughout . in most classical disordered models studied so far the transition from the disordered to the ordered phase is of second order in the thermodynamic sense . in the exact solution of the sk model , the spin - glass order parameter @xmath17 is continuous at the transition which is of second order in the thermodynamic sense @xcite . in other classical glassy models like the potts glass @xcite or the spherical @xcite and ising @xcite @xmath0 spin - glasses , the order parameter jumps at the transition which , however , is still of second - order in the thermodynamic sense since there is neither a jump in the susceptibility nor a latent heat . a classical model that exhibits a first - order transition is the anisotropic @xmath0 spin - glass , @xmath18 , in which the spins take integer values between @xmath19 and @xmath20 and there is an extra term in the hamiltonian @xmath21 , proportional to a coupling constant @xmath22 , that controls the crystalline tendency . in this case , a crossover from a second - order to a first - order thermodynamic transition in the plane @xmath23 has been exhibited in the exact solution @xcite . the classical ghatak - sherrington ( gs ) model @xcite is another candidate to exhibit a second to first order crossover in the thermodynamic transition . it is the anisotropic extension of the sk model , or the @xmath24 limit of the previous model . in this case , a crossover from a second order to a first order transition in the plane @xmath23 has been exhibited in an _ approximate _ solution ( one step replica symmetry breaking ) @xcite . the exact solution has not been derived yet and it is then not well established if it has a true first - order thermodynamic transition . in quantum problems , first order transitions have been reported in three models . the first one is the so - called `` fermionic ising spin - glass '' analyzed by oppermann and collaborators @xcite . this model , however , is thermodynamically equivalent to the classical gs model discussed above @xcite . the other two models are very similar indeed and they are different ways of extending the classical spherical @xmath0 spin - glass model @xcite to include quantum fluctuations . in one case , the continuous spins are generalized to @xmath25 component vectors and a global spherical constraint as well as commutation relations are imposed @xcite . the other one uses the fact that the spherical @xmath0 spin - glass model can be interpreted as a particle moving in an infinite dimensional hyper - sphere with a random potential . quantization is then done by imposing commutation relations between coordinates and momenta @xcite . the latter can also be interpreted as an extension of the a quantum rotor model @xcite that includes @xmath0 interactions . the relation between the critical properties of the quantum versions of @xmath0 spin - glass models and the experiments in @xcite has been put forward in @xcite . in addition , the connection between the static calculation supplemented by the marginality condition and the analysis of the out of equilibrium dynamics in contact with an environment developed in @xcite was also discussed in @xcite . however , the reason why the transition changes from second to first order close to the quantum critical point was not clear from this analysis . it is one of the aims of this article to clarify this point , and study to what extent one can claim it to be general , with the use of the tap approach . quantum tap equations for the sk model in a transverse field have been presented by ishii and yamamoto @xcite and cesare _ et al _ @xcite . the former use a perturbative expansion of the free - energy in the strength of the transverse field , and then follow closely tap s techniques ; the latter implement a cavity method . the tap equations derived by rehker and oppermann @xcite for the fermionic spin - glass model coincide with the ones presented by yokota @xcite for the classical gs model since these two models are thermodynamically equivalent @xcite . hence , our aim is twofold . on the one hand we want to present a quantum extension of the tap approach to the statistical properties of disordered systems . thus , after a short revision of the classical tap approach in section [ introtap ] , we discuss in section [ sec1:formalism ] the derivation of the quantum tap free - energy and tap equations using a general approach that extends the ones developed by plefka @xcite and georges and yedidia @xcite . the advantage with respect to previous derivations of quantum tap equations is that this method can be applied to any quantum disordered model and it allows to obtain the tap equations as well as the tap free energy . in section [ p - spin ] we present , as an example , the tap free - energy and tap - equations for the quantum extension of the @xmath0 spin spherical spin - glass model studied in @xcite . we show that the tap equations can be easily related to the equations for the order parameter in the matsubara replica approach and also to some of the equations appearing in the real - time dynamic approach . the tap analysis of this model furnishes a benchmark to study the generalization to the quantum case of the methods and interpretations developed for classical systems . section [ transition ] is devoted to the second aim of this article . via the tap approach we show that the same type of phase diagram naturally emerges for all systems having a discontinuous phase transition in their classical limit ( these are models solved by a one - step replica symmetry breaking ansatz within the replica analysis ) . in particular we relate the first - order transition close to the quantum critical point to the structure of metastable states . finally , we present our conclusions in section [ conclu ] . in this section we present a short revision of the classical tap approach to mean - field disordered spin models . the classical tap free - energy @xcite is the legendre transform of the free - energy with respect to local magnetic fields , @xmath26 where @xmath27 are lagrange parameters enforcing the condition @xmath28 . @xmath29 is an effective potential that depends on the local magnetizations . the lagrange conditions @xmath30 , called tap equations , fix the local magnetizations as functions of the local magnetic fields . the solutions @xmath31 of the tap equations are stationary points of @xmath32 . if they are also stable ( all the corresponding eigenvalues of the free - energy hessian are positive ) , they are identified @xcite with pure states , also called tap states . this interpretation was put forward by de dominicis and young @xcite who showed that the partition function in the classical sk model can be written as a weighted sum over the stable solutions of the tap equations : @xmath33 where the index @xmath34 labels different tap states , @xmath35 is an @xmath36-vector encoding the local magnetization in the solution @xmath34 , @xmath37 is the extensive tap free - energy of such solution and the sum runs over all tap solutions . consequently , the static average of any observable can be computed from eq . ( [ eq : weighted ] ) . at low temperatures the tap free - energy has a large number of minima . if one groups different tap states with the same free - energy in sets @xmath38 then the partition function can be written as @xmath39 where the factor @xmath40 is the number of solutions with tap free - energy density @xmath41 . one can now replace the sum by an integral and exponentiate the factor @xmath40 ; this yields @xmath42 where we have taken the continuous limit and introduced the _ complexity _ @xmath43 the configurations that dominate the sum are those having a free - energy density such that it minimizes @xmath44 . the identity between the partition function and the weighted sum over tap solutions has been demonstrated for many others models @xcite and it is generally believed to hold for any mean - field disordered system . in the following we focus on `` discontinuous glassy systems '' @xcite that are characterized by having a discontinuous transition ( the edwards - anderson order parameter , @xmath45 , jumps ) that is still of second order thermodynamically . within the replica analysis of the partition function these models are characterized by a one - step replica symmetry breaking solution below a static transition temperature @xmath46 and a replica symmetric ( rs ) solution , that corresponds to the paramagnetic phase , at @xmath47 . however , for intermediate temperatures @xmath48 there are an exponential in @xmath36 number of non - trivial tap solutions that combine themselves in such a way that the sum ( [ eq : weighted3 ] ) is identical to the rs result . the relationship between metastable states and replicas has been put forward in @xcite . indeed , consider @xmath49 different identical systems ( `` clones '' ) coupled by an attractive , infinitesimal ( but extensive ) interaction . when there exist many pure states all the clones fall into the same state and the free energy for the system of @xmath49 clones reads : @xmath50 on the other hand the computation of the left hand - side of ( [ reltaprep0 ] ) can be performed within the replica formalism : @xmath51 where the overline represents the average over disorder . since the attractive coupling between the @xmath49 clones is infinitesimal , the computation of the right - hand side of ( [ reltaprep1 ] ) reduces simply to the calculation of @xmath52 , where the replica symmetry between the @xmath53 groups of @xmath49-replicas ( @xmath54 ) is _ explicitly _ broken . when the system is in the replica symmetric phase ( @xmath55 ) , this reduces to study one - step solutions non - optimized with respect to @xmath49 : @xmath56 where @xmath57 is the free - energy computed by using replicas , @xmath45 is the edwards - anderson parameter and @xmath49 is the breakpoint ( or the size of the blocks in the replica matrix ) . for simplicity we consider that the inter - state overlap @xmath58 equals zero . the definitions of these parameters are standard in the replica approach @xcite and they will appear in the analysis of the quantum @xmath0 spin - glass model in section 3 . since the integral on the left - hand - side of ( [ reltaprep ] ) is dominated by a saddle point contribution , one finds that , for a given temperature , fixing the value of @xmath49 is equivalent to summing over states with a given energy density @xmath59 . the relationship between @xmath59 and @xmath49 reads @xmath60 note that within this framework one does not have to optimize with respect to @xmath49 . instead , @xmath49 is a free parameter and , by changing the value of @xmath49 , one can consider different groups of metastable states . the analysis of the tap equations reveals three temperature regimes for discontinuous glassy systems : * _ high temperatures _ the system is in the paramagnetic phase , the paramagnetic tap solution @xmath62 , for all @xmath63 , dominates the sum and @xmath64 . @xmath65 is the dynamic critical temperature . above @xmath65 the dynamics starting from a random initial condition converges asymptotically to the paramagnetic solution . * _ intermediate temperatures _ the replica analysis of the partition function indicates that the system is still in the paramagnetic phase . however , the study of the tap equations and the dynamics show that at @xmath65 the paramagnetic solution is fractured in an exponentially large in @xmath36 number of minima of the tap free - energy @xcite . indeed one can recover these results also by the replica method . a careful replica analysis shows that there exist one - step solutions in these temperature regime other than the paramagnetic one . these solutions are in one to one correspondence with groups of states with a given free - energy density ( through the relationship ( [ rela : xf ] ) ) . for instance , one can follow the evolution of the threshold states ( the states with highest free - energy ) by tuning the parameter @xmath49 . for these states , @xmath67 when @xmath68 and decreases at lower temperatures . moreover , the dominant contribution to eq . ( [ eq : weighted3 ] ) is given by the states characterized by @xmath67 , _ i.e. _ those with free - energy density such that @xmath69 these are the threshold states at @xmath65 and other groups when @xmath70 . hence , between @xmath71 saddle - point solutions ( corresponding to @xmath67 ) that are not absolute minima of @xmath37 dominate the integral since their number scales exponentially with @xmath36 . the final result for the free - energy density in this temperature range coincides with the one of the prolongation of the paramagnetic solution ( that actually does not exist ! ) . a naive replica computation fails to signal the difference between a true paramagnetic solution and the ensemble of non trivial tap solutions with @xmath72 . the dynamic approach detects the change in free - energy landscape at @xmath65 since the system can not reach equilibrium for any temperature below @xmath65 @xcite . * _ low temperatures _ @xmath73 . at the static transition temperature the complexity of the tap solutions , which dominates the sum ( [ eq : weighted3 ] ) , vanishes the static transition appears as an _ entropy crisis _ since the part of the total entropy that is related to the large number of states disappears . for @xmath74 the tap states which dominate the integral in eq . ( [ eq : weighted3 ] ) correspond to the equilibrium glassy phase . dynamically , @xmath46 does not play any role . the out of equilibrium dynamics is dominated by the threshold states , which are the highest ones in free - energy and that are characterized by flat directions in the free - energy landscape . note that via the tap approach one can obtain a reasonable justification of the marginality condition @xcite often used to obtain information about the out of equilibrium dynamics starting from a pure equilibrium computation @xcite . indeed the value of @xmath49 fixed by the marginality condition corresponds to the tap states which are marginally stable ( the threshold states ) : the flatness of the free - energy landscape around these states is responsible for aging @xcite . in this section we present a simple procedure to derive tap equations for generic completely connected quantum systems . we also expose the physical meaning of quantum tap equations by the cavity method @xcite . we are aware of two publications where tap equations for quantum systems have been already presented @xcite . with respect to these works our derivation is more systematic , simple and it allows one to obtain the tap equations as well as the tap free - energy for any completely connected quantum disordered system . the formalism that we use to derive tap equations for generic quantum problems is very similar to the one described in @xcite and , it follows even more closely , the one used in @xcite to obtain the dynamical tap equations for classical disordered models . + we focus on systems characterized by the potential energy : @xmath75 where @xmath76 may represent an su(2 ) spin ( @xmath77 ) , a rotor ( @xmath78 ) , a spherical spin ( @xmath79 ) or a space - coordinate ( @xmath79 ) and @xmath80 , the couplings between the different @xmath76 , are independent random variables with zero mean and variance @xmath81 as a consequence the following derivation applies to ( completely connected ) heisenberg models , quantum rotor models and quantum continuous systems . ] . without loss of generality and to simplify the notation we shall suppress the index @xmath82 in the rest of this section . for classical spin - glasses tap showed that all the local magnetizations @xmath7 , @xmath6 , are needed to derive the relevant free - energy density to describe the metastable properties @xcite . if one is interested in the dynamics of classical disordered mean - field systems , one has to legendre transform not only with respect to all time - dependent local magnetizations @xmath83 , but also with respect to the autocorrelation @xmath84 @xmath85 and the linear response @xmath86 @xcite . in order to describe the metastable properties of a quantum disordered model we shall show that it is necessary to legendre transform with respect to the local average coordinates , @xmath87 , and the autocorrelation function in imaginary time , @xmath88 . the quantum tap free - energy reads @xmath89\right|_{\alpha = 1 } \ ; . \label{eq : gamma}\end{aligned}\ ] ] where @xmath90 is the kinetic energy , @xmath91 indicates the functional measure on the configuration space and @xmath82 is a parameter whose role will be clarified in the following . for instance , if @xmath76 are su(2 ) spins @xmath90 is the berry phase and the functional measure is restricted to periodic functions @xmath92 ( with period @xmath93 ) satisfying the constraint @xmath94 . the sources @xmath95 and @xmath96 have the role of lagrange multipliers fixing the average value of the coordinates and the correlation : @xmath97 once the tap free - energy @xmath98 is known , one can derive the tap equations as legendre relations , @xmath99 until now we have not not used the scaling ( [ coupling ] ) and all these definitions can be equally applied to finite dimensional systems . the great simplification due to the mean field character of the interactions in ( [ coupling ] ) is unveiled if one performs a perturbative expansion of eq . ( [ eq : gamma ] ) in @xmath82 and writes @xmath100 as a power series in @xmath34 : @xmath101 in fully - connected models , if one chooses the correct order parameters ( which are @xmath87 , @xmath88 in the quantum case ) , the perturbative expansion ( [ series ] ) around the pure kinetic theory is actually a simple sum over three terms . higher order terms in the series vanish in the thermodynamic limit due to the scaling of @xmath80 with respect to @xmath36 . in more general cases , in finite dimensions , this will not be the case and ( [ series ] ) becomes a @xmath102 expansion around mean field theory , where @xmath103 is the spatial dimension @xcite . let us consider in more detail the terms arising from the expansion ( [ series ] ) . the zeroth - order one is simply @xmath104 , i.e. the free - energy of @xmath36 free spins constrained to have local magnetizations @xmath105 and a global correlation function @xmath106 . this term depends only on the nature of the degrees of freedom , whether they are @xmath107 spins , rotors or space - coordinates . in particular it can be analytically computed only if @xmath108 are spherical spins or space - coordinates . in the other cases one has to resort to approximations or numerical computations . the first - order term is the naive mean field free - energy : @xmath109 note that the decoupling of the spins for @xmath110 is essential to obtain the last identity . the second - order term depends on the correlation function and the overlap function @xmath111 only and equals @xmath112 using the scaling of the couplings with @xmath36 and by the same arguments developed for classical systems @xcite , we have verified that all orders @xmath113 in the series ( [ series ] ) are suppressed in the thermodynamic limit . first of all , let us write the tap equations in a way which allows one to clarify the physical meaning of the different terms : @xmath114 \ , m_i(\tau ' ) \ ; , \nonumber \\ \label{tap2mean } \left . \frac{2}{n}\frac{\delta ( -\beta f)}{\delta c(\tau , \tau ' ) } \right|_{\alpha = 0}&=&g^{cav}(\tau,\tau ' ) = \frac{p}{2 } \left [ q^{p-1}(\tau,\tau ' ) - c^{p-1}(\tau,\tau ' ) \right ] \ ; .\end{aligned}\ ] ] the solutions to these equations are expected to be time - translation invariant since we are developing a description of equilibrium and metastable properties . therefore @xmath115 is indeed independent of the imaginary time and @xmath116 depends only on the difference between @xmath117 and @xmath118 . an understanding of the meaning of the quantum tap equations follows from the analysis of @xmath37 for @xmath110 . indeed , by tracing out all the spins except @xmath108 in the partition function produces a single - site measure ( for @xmath108 ) whose action reads : @xmath119 -\frac{1}{2 \hbar^2 } \int_0^{\beta\hbar } d\tau \int_0^{\beta\hbar } d\tau ' \ ; s_i(\tau)g^{cav } ( \tau,\tau ' ) s_i(\tau ' ) \;,\ ] ] where @xmath120 is the kinetic energy for the spin @xmath108 . as a consequence the tap solutions are the self - consistent relations that relate @xmath121 and @xmath122 ( which are functions of @xmath106 and @xmath105 ) to @xmath106 and @xmath105 obtained from the single - site action ( [ singlesite ] ) . equations ( [ tap1mean ] ) and ( [ tap2mean ] ) show that the action on the _ i_th spin of the @xmath123 remaining ones reduces simply to @xmath115 and @xmath116 . this implies that tracing out all the spins but the _ i_th one produces a gaussian measure for the instantaneous magnetic fields @xmath124 , whose mean and connected two - point correlation function equal respectively @xmath115 and @xmath125 . the expression of @xmath115 and @xmath125 can be justified within the cavity method @xcite . let us focus for simplicity on the @xmath24 case for which @xmath126 where @xmath127 represents the thermal average with respect to the system with the _ i_th site removed . @xmath128 is not simply equal to @xmath129 , which is the mean magnetization for the system of n spins . a correction term , first discovered by onsager , appears : @xmath130j_{i , k } m_{i}\ ] ] plugging eq . ( [ onsager ] ) into eq . ( [ cavp2 ] ) and using the scaling of the couplings with @xmath36 one recovers the expression for @xmath115 given in ( [ tap1mean ] ) in the @xmath24 case . whereas for @xmath116 a similar computation on a completely connected lattice . ] gives back the expression given in ( [ tap2mean ] ) . finally , we remark that the main difference between the classical and the quantum tap approach is that in the latter the cavity interaction consists not only in a cavity field but also in the `` weiss function '' @xmath125 , which is a function of ( imaginary ) time . this already happens in the mean - field theory of quantum non - disordered systems @xcite for which local quantum fluctuations are taken into account exactly , whereas the spatial ones are frozen . for disordered systems , even in the limit of infinite dimensions , one has to take into account not only the local quantum fluctuations but also some spatial fluctuations : all the instantaneous magnetic fields have the same variance but their averaged values fluctuate from site to site . in this section we apply the method of section [ sec1:formalism ] to the study of the quantum spherical @xmath0 spin - glass model . we derive and analyze the tap free - energy density and the tap equations for the local magnetization and correlation function in imaginary time . we relate these equations to the equation for the order parameter in the matsubara replicated approach to equilibrium and in the schwinger - keldysh approach to the non - equilibrium dynamics . a model of a quantum particle with position @xmath131 and momentum @xmath132 that moves on an @xmath36-dimensional random environment is defined as @xmath133= \frac{{\bf p}^2}{2 m } - \sum^{n}_{i_1< ... <i_p } j_{i_1 ... i_p } s_{i_1 } ... s_{i_p } \ ; . \label{eq : action}\ ] ] a lagrange multiplier @xmath134 enforces the averaged spherical constraint @xmath135 the random interaction strengths @xmath136 are taken with zero mean and variance defined in eq . ( [ coupling ] ) . this model is a possible quantum extension of the spherical @xmath0 spin - glass model introduced in @xcite and it is a particular realization of the class defined in ( [ potential ] ) corresponding to space - coordinates @xmath108 constrained to move on a n - dimensional sphere . the zero - th order term of the expansion ( [ series ] ) can be readily computed for this model . by setting @xmath137 , rescaling time according to @xmath138 , and defining the `` quantum parameter '' @xmath139 we obtain the following expression for the quantum tap free - energy ( [ eq : gamma ] ) : @xmath140 the physical parameters @xmath87 and @xmath88 are fixed by the quantum tap equations ( [ legendre ] ) @xmath141 \ , m_i(\tau ' ) \ ; , \nonumber \\ & & z(\tau)\delta ( \tau -\tau ' ) = ( c - q)^{-1}(\tau,\tau ' ) + \delta(\tau-\tau ' ) \frac{1}{\gamma } \frac{\partial^2}{\partial\tau^2 } + \frac{p}{2 } \left [ c^{p-1}(\tau,\tau ' ) - q^{p-1}(\tau,\tau ' ) \right ] \label{eq1 } \ ; .\nonumber\end{aligned}\ ] ] finally , setting @xmath142 and using that at stationarity , @xmath143 , @xmath144 , @xmath145 and @xmath146 , the previous equations are simplified to @xmath147 in the classical case the tap equations admit a large number of solutions at low temperatures . in the following we shall show that this remains the case in a certain regime of @xmath148 and @xmath149 . furthermore we shall classify them by their edwards - anderson parameters . finally , we shall exhibit several properties of the tap solutions valid at low temperatures . let us analyze in detail the equations for the local magnetizations , @xmath150 . first of all we note that a simple equation that relates @xmath45 to the potential energy density derives from eq . ( [ eq2 ] ) . in fact , by multiplying eq . ( [ eq2 ] ) by @xmath151 and summing over @xmath152 one obtains ( for @xmath153 ) @xmath154 where we introduced the discrete fourier transform of the correlation @xmath155 following kurchan _ et al _ @xcite , we introduce the _ angular _ variables @xmath156 and define the _ angular potential energy density _ @xmath157 for any fixed energy level @xmath158 , eq . ( [ eqq1 ] ) becomes a second - order polynomial equation for @xmath45 ; the solution is determined by @xmath159 and @xmath160 , called the threshold value , is given by @xmath161 the right - hand - side of eq . ( [ eqq1final ] ) has to be real . this imposes the condition @xmath162 since @xmath158 is a negative quantity . for each sign in eq . ( [ eqq1final ] ) , its left - hand - side has a bell - shape as a function of @xmath45 . it vanishes at @xmath163 and @xmath164 and attains its maximum at @xmath165 . hence , at fixed values of @xmath158 and @xmath148 , eq . ( [ eqq1final ] ) has none or _ two _ solutions , @xmath166 , with @xmath167 ( we assume , as expected , that @xmath168 is positive for all @xmath117 ) . in the classical case , the minus sign in eq ( [ eqq1final ] ) leads to a value of @xmath45 that is a minimum of the tap free - energy for all @xmath169 . the edwards - anderson parameter determined in this way has the expected physical behavior @xcite . in appendix a we show that in the quantum case one has to choose the minus sign in eq . ( [ eqq1final ] ) , too . thus , @xmath45 is determined by @xmath170 this equation still has two solutions . it can be proven that the solution with the larger absolute value of @xmath45 has the correct physical properties . in particular , it is connected to the classical solution , and it is then the solution to be kept . thus there is a one to one correspondence between @xmath45 and @xmath158 . it is of particular interest , as we shall show below , the _ threshold solution _ @xmath171 . in this case the equation for @xmath45 becomes @xmath172 note that this equation coincides with the one found with the matsubara formalism using the marginality condition to fix the block size @xmath49 in the replica matrix @xcite . furthermore , it coincides with the equation for the dynamic value of the edwards - anderson parameter @xmath173 obtained from the study of the real - time dynamics of the quantum model evolving in contact with an ohmic quantum environment @xcite , when one takes first the thermodynamic limit , next the long - time limit of the system s dynamics in contact with the environment and , finally , the strength of the coupling to the environment to zero . the relationship between tap , matsubara and dynamical approach will be discussed in section [ reltapmatdyn ] . the equations ( [ eqq1final ] ) reveal an interesting structure of the tap equations ( [ eqc - qstat ] ) and ( [ eqmstat ] ) . for a given value of the angular potential energy @xmath158 the tap equations decouple in two different sets : eqs . ( [ eqc - qstat ] ) and ( [ eqq1final ] ) , with the spherical condition on @xmath174 , determine the correlation function , the spherical parameter and the edwards - anderson parameter ; whereas eqs . ( [ eqmstat ] ) determine the angular variables only . they read @xmath175 where we have defined @xmath176 for a given value of the angular potential energy @xmath158 eqs . ( [ angular ] ) allow one to determine the angular part of the tap solutions . in general for a given value of @xmath158 , eqs . ( [ eqc - qstat ] ) and ( [ eqq1final ] ) determine the correlation function , the spherical parameter and the edwards - anderson parameter _ in a unique way_. ( an exception to this rule are the paramagnetic solutions which however do not correspond to any @xmath177 . ) as a consequence , the multiplicity of tap solutions is entirely due to eq . ( [ angular ] ) which , for certain values of @xmath177 , can admit an exponential ( in @xmath36 ) number of solutions @xmath178 . the complexity as a function of @xmath158 is then defined as @xmath179 equation ( [ angular ] ) already appears at the classical level . the complexity has been computed by crisanti and sommers @xcite and cavagna _ et al . _ @xcite with the following result . there are typically no solutions for @xmath180 , whereas for @xmath181 the complexity reads @xmath182 where @xmath183 is the value at which @xmath184 vanishes . a plot of this function is traced in fig . [ comp.fig ] for @xmath185 . in the classical case the tap equations separate in two sets : @xmath36 equations for the angular variables and one for the edwards - anderson parameter . the former admit an exponential number of solution and are studied from a statistical point of view ( one computes the number of solution , and the typical properties of solution corresponding to a given @xmath186 ) , whereas the latter can be easily solved . in the quantum case the analysis of the equations for the angular variables is identical to the one used for classical systems . the analog of the equation that determines @xmath45 becomes now a differential equation for @xmath187 that has to be studied numerically . this differential equation can be mapped exactly onto the ones analysed with the replica method , as we shall show in section ( [ reltapmatdyn ] ) , and its numerical solution can be found in @xcite . in this section we perform a low - temperature and low @xmath149 approximation , also discussed in @xcite , that allows one to obtain some qualitative results that remain valid for the exact solution . at low temperature and low @xmath149 , the extension of the imaginary time - interval diverges @xmath188 $ ] and the periodic correlation @xmath187 is expected to have a rapid decay , over a short time - interval , from @xmath189 to its `` asymptotic '' value , say , at @xmath190 . moreover the `` regular '' part of the correlation , that we define as @xcite @xmath191 can be assumed to be small . therefore we can expand the tap free - energy in powers of @xmath192 . up to terms of the order of @xmath193 we obtain @xmath194 where we have focused on the space of time translation invariant ( tti ) functions ( since the tap solutions are tti this does not imply a loss of generality ) . within this approximation the tap equations become quadratic in fourier space , @xmath195 and yield @xmath196 by taking @xmath197 and comparing to eq . ( [ eqq1finalnew ] ) one obtains @xmath198 the spherical constraint reads @xmath199 where @xmath200 the integral in eq . ( [ eq : constaint ] ) has to be taken on the interval @xmath201 $ ] such that the square root is real . in the low temperature limit , we approximate @xmath202 and , by changing variables in the integral , we obtain @xmath203 with @xmath204 equation ( [ eq : gamma - q1 ] ) yields a relation between @xmath205 , @xmath45 and @xmath158 of the form @xmath206 with @xmath207 a numerical constant . for each @xmath158 , there is a solution with a physically meaningful value of @xmath45 that is close to @xmath189 , until reaching a critical @xmath208 . this value tells us when the tap solutions associated to @xmath158 disappear . it can be easily proven that @xmath209 is a growing function of @xmath158 ; hence , @xmath208 is a decreasing function of @xmath158 . this implies that the tap solutions that are at the threshold level disappear first than those that are at lower values of @xmath158 . this is again similar to the dependence of the classical tap solutions with temperature @xcite : the solutions corresponding to the threshold level disappear at a lower temperature than the ones corresponding to the equilibrium level @xmath210 and , more generally , @xmath211 if @xmath212 . the low frequency behavior of the spectral density @xmath213 of the threshold states is gapless , @xmath214 whereas all the other states ( @xmath215 ) have a gap @xmath216 in their excitation spectrum , @xmath217 furthermore we have studied the dependence of the free - energy on @xmath218 and @xmath149 in the low temperature limit . plugging the solution ( [ qapprox ] ) into ( [ tapapprox ] ) we find , after a tedious computation , @xmath219 where @xmath45 satisfies eq . ( [ eqq1finalnew ] ) . this expression allows one to study the evolution of the free - energy of the tap states as a function of @xmath149 . we have found that if one knows a tap solution at zero temperature and zero @xmath149 one can follow it continuously in @xmath205 . as in the classical case tap solutions do not cross , merge nor divide in this model . figure [ fgamma.fig ] summarizes these results in a schematic way . finally , note the special role of the threshold states , which are gapless ( contrary to the others ) , the first ones to disappears and the ones with highest free - energy density . in the classical case one can check the stability of tap states . in the quantum case this is a difficult task that has to be performed numerically . in the following we shall limit ourself to prove that the tap solutions characterized by @xmath220 ( threshold states ) are characterized by zero modes and are hence marginally stable . we expect that a complete stability analysis will confirm that the tap states characterized by @xmath215 are stable . let us focus on the reduced free - energy hessian @xmath221 evaluated in a tap solution @xmath31 . this matrix depends on @xmath222 only through their difference . therefore it is diagonal in fourier space . focusing on zero frequency , the original problem reduces to the diagonalization of the following matrix : @xmath223 where @xmath224 . the density of eigenvalues of @xmath225 has been computed in @xcite and , except for the isolated eigenvalue corresponding to the eigenvector @xmath226 , it is a semicircular law centered in @xmath227 with width @xmath228 . consequently , threshold states are characterized by a vanishing fraction of zero modes . the classical limit of eqs . ( [ eqc - qstat ] ) and ( [ eqmstat ] ) yields the classical tap equations computed by kurchan _ in fact , in the classical limit , @xmath229 and the parameter @xmath134 is fixed by integrating eq . ( [ eqc - qstat ] ) between @xmath230 and @xmath231 . this yields @xmath232 by inserting this value of @xmath134 in eq . ( [ eqmstat ] ) we obtain @xmath233 that coincide with the classical tap equations for the local magnetizations . the equation that fixes @xmath45 as a function of @xmath158 in the classical limit is simply obtained from eq . ( [ eqq1final ] ) by setting @xmath234 . in section [ introtap ] we have recalled the relationship between tap , replica and dynamical approaches in the classical case . in this subsection we show how these connections are generalized to quantum systems . via the replica analysis in the matsubara imaginary - time framework and within a 1step rsb ansatz , the order parameter is the @xmath235 matrix @xmath236 which is fully described by : @xmath53 identical diagonal elements @xmath237 that depend on the the imaginary time @xmath238 , @xmath239 constant elements @xmath45 that occupy the @xmath240 blocks around the diagonal , the remaining @xmath241 elements @xmath242 that in the absence of an external field are identically zero . in the @xmath243 limit , the three parameters @xmath237 , @xmath45 and @xmath49 , together with the value of the lagrange multiplier that enforces the averaged spherical constraint determine the full solution of the problem @xcite . the connection between tap and the matsubara approaches is obtained by identifying the edwards - anderson parameters @xmath45 in the two approaches , @xmath187 with the @xmath238-dependent diagonal parameter @xmath237 in @xmath236 , and the lagrange multipliers . in particular we have shown that subtracting the equation obtained for @xmath244 from the one corresponding to @xmath245 one obtains eq . ( [ eqc - qstat ] ) . in the matsubara approach one has another equation for @xmath45 in which @xmath49 acts as an external parameter . therefore by fixing the value of the breakpoint one fixes the value of @xmath45 . as in the classical case two different recipes to fix the breaking point parameter @xmath49 , namely optimization and the marginality condition , lead to the static and dynamic transitions , respectively . in the tap approach the role of @xmath49 is played by @xmath177 that enters the equation for @xmath45 as a parameter . we have found that the relationship between @xmath49 and @xmath246 is encoded in @xmath247 where @xmath248 is the complexity defined in ( [ complexity0 ] ) , see appendix b. this suggests that in the a quantum problem the relationship ( [ reltaprep ] ) is generalized to @xmath249 using eq . ( [ me ] ) we have found that the matsubara equations for @xmath45 and @xmath250 and the tap equations for @xmath45 and @xmath168 coincide . for instance , the tap equations for the highest tap states ( threshold states ) and the lowest tap states coincide with the ones obtained in the matsubara approach by using the marginality condition and the extremization with respect to @xmath49 , respectively . moreover , as another confirmation of eq . ( [ reltaprepquantum ] ) we have verified that the free - energy obtained from the matsubara computation @xcite equals the one obtained in the tap approach for all values of @xmath93 and @xmath149 . in other words , we have checked that @xmath251 as a consequence the phase diagram that follows from the tap approach coincides with the one obtained in @xcite . . the study of the real - time dynamics of the p - spin quantum model evolving in contact with an ohmic quantum environment has been performed in @xcite . the dynamical behavior is characterized by two regimes . at high temperature and high @xmath149 the system equilibrates in the paramagnetic state via an equilibrium dynamics . whereas at low temperature and low @xmath149 the systems ages and remains out of equilibrium also at infinite times . as in the classical case we have found that the long - time out of equilibrium dynamics is dominated by the threshold states . this is proven by the fact that the equations for @xmath45 and @xmath187 within the tap approach coincide with the ones obtained from the study of the real - time dynamics , when the following limits are taken in its precise order : @xmath252 . in order words , when one takes first the thermodynamic limit , next the long - time limit of the system s dynamics in contact with the environment and , finally , the strength of the coupling to the environment @xmath253 to zero . notice that this equivalence holds for the paramagnetic states also . finally , we have shown that the relationship between the effective temperature @xcite arising in the asymptotic out of equilibrium regime @xcite and the complexity is , as classically , @xmath254 note that the connection between tap and real - time dynamics is done by identification of several equations . a more precise analysis , along the lines of @xcite , should prove the full equivalence of the two methods . in this section we present some general arguments that allow one to predict the phase diagram of discontinuous glassy systems . since the free - energy landscape plays a key role , we expect these results to have a certain degree of universality and to apply to this entire class of disordered systems . let us focus on two limiting regions of the phase diagram : around the classical phase transition ( @xmath255 ) and around the quantum phase transition ( @xmath256 ) . in the former case the physics is well known and it is reviewed in section [ introtap ] . the effect of switching on weak quantum fluctuations consists only in a weak variation of the complexity @xmath248 . for this reason the effect of quantum fluctuations reduces simply to a variation of the thermodynamic ( @xmath257 ) and the dynamic ( @xmath258 ) transition temperatures ( respectively lines @xmath259 and @xmath260 in fig . [ diag.fig ] ) . at zero temperature and low @xmath149 the system is in the glassy phase ( gp ) , whereas at very high @xmath149 quantum fluctuations destroy the glassy phase and the system is a quantum paramagnet ( qpm ) . as a consequence one expects that a quantum phase transition should divide these two regimes at a certain value @xmath261 . at zero temperature the complexity is expected to remain a smooth function of the free - energy density since it is bounded by the logarithm of the number of energy minima divided by @xmath36 , which is a finite quantity independent of temperature . ] . consequently , equation ( [ eq : weighted3 ] ) implies that the sum over the exponential number of glassy states is always dominated by the lowest ones in free - energy since the @xmath262 term in the exponential largely dominates in this limit ( this is different from the classical problem in which other states dominate between @xmath46 and @xmath65 ) . at zero temperature these are the states with lower angular potential energy , _ i.e. _ with @xmath263 . now , from fig . [ comp.fig ] we conclude that @xmath264 at zero temperature and for all @xmath205 . for this reason the mechanism behind the transition must be totally different from the classical one . the transition can not be related a configurational entropy that vanishes when approaching @xmath265 from above ( `` entropy crisis '' ) since this quantity is always zero at zero temperature . indeed , according to eq . ( [ me ] ) , if we assume that @xmath266 when @xmath267 , then @xmath268 for all @xmath205 in the glassy phase . in the paramagnetic phase instead @xmath67 . thus , @xmath49 must jump at the transition . if the edwards - anderson parameter also jumps at @xmath265 , the susceptibility is discontinuous , and the transition is of first order thermodynamically . as in the previous case the effect of switching on thermal fluctuations reduces simply , for low @xmath148 , to a variation of @xmath269 ( line ( 2 ) in fig . [ diag.fig ] ) . another hint on the difference between classical and quantum phase transition can be gained by a technical remark . it is well known that the paramagnetic solution of the classical problem remains stable in the low temperature phase . this is a spurious solution of the mean field equations which has to be discarded in the analysis of the low temperature regime . in the quantum case , one also expects to find a spurious paramagnetic solution , which is the continuation of the classical paramagnet to low temperatures . this solution exists to the left of line @xmath259 in fig . [ diag.fig ] , consequently one expects coexistence of two paramagnetic solutions : a physical one which is the continuation of the quantum paramagnet valid at low temperatures and high @xmath205 and a spurious one which is the continuation of the classical paramagnet . in the classical case the transition is of second order even if the order parameter jumps discontinuously . this peculiar behavior is due to the fact that near the transition the paramagnetic state is fractured into an exponential number of states which continuously become the ones responsible for the glassy phase at low temperature . this is not possible at zero temperature ( the quantum paramagnet is not formed by a collection of glassy states ) and therefore it is reasonable to expect a quantum first order phase transition between the glass phase and the quantum paramagnet . increases and vanish exactly at @xmath261 . in this case the glassy states could be grown up continuously from the quantum paramagnet . ] finally , note that @xmath270 on the point b. we then expect that a first order transition line separating the qpm from the cpm starts at this point . this line should end on a point @xmath174 given that for very large value of @xmath149 and @xmath148 the quantum and thermal fluctuations are so strong that the system becomes non interacting and in this case only one paramagnetic phase exists . in the analysis of the @xmath185 spherical spin - glass model @xcite the line @xmath271 has not been found . we conjecture that in this case the line @xmath271 is so short that it is very difficult to find numerically . furthermore , within the accuracy of the algorithm , the dynamic and static critical lines collapse at the point b. in the quantum model studied in @xcite instead this line has been detected and it was demonstrated in this paper that its length increases with @xmath0 . now that the equilibrium phase diagram is completely predicted from a qualitatively point of view , we can focus on the non - equilibrium regime . as noted previously , low quantum fluctuations simply change the values of @xmath258 but do not change qualitatively the dynamic transition which remains second order in the sense that the asymptotic energy is continuous across the transition , but its derivative is not . this remains true until the line @xmath272 reaches the line @xmath271 . after this point the dynamic transition between the the quantum paramagnet and the threshold states becomes first - order , i.e. the asymptotic energy is not continuous across the transition . this , of course , is very difficult to see numerically since the discontinuity has a very small value . in summary , through some general arguments based on the tap approach we have predicted a phase diagram that should have a certain degree of universality since its form is determined by the qualitative form of the free - energy landscape . indeed , not only the quantum @xmath0 spin spherical model exhibits the phase diagram displayed in fig . [ diag.fig ] but some other classical and quantum models share exactly the organization of phases and transitions @xcite . in this paper we have derived tap equations for a large class of mean field disordered quantum system . moreover we have applied the tap approach to the quantum version of the spherical @xmath0 spin model . the study of this system , whose real - time dynamics and statics have been analyzed in @xcite , has furnished an ideal benchmark to generalize to the quantum case several concepts developed for classical disordered systems . armed with this knowledge , founded on the study of the free - energy landscape , we have shown that the same phase diagram , presented in fig . [ diag.fig ] , naturally emerges in a large class of quantum disordered systems , those having a classical discontinuous transition . whether other models like the sk model in a transverse field , or its soft spin version studied in @xcite , also have such crossover in the transition from the disordered to the ordered phase is an issue that deserves revision . for the moment , no study of models with classical continuous transition has shown this feature . however , it might have been masked by the methods used in previous studies . the soft sk model might be the easiest example where to answer this equation via , _ e.g. _ , a careful application of the replicated matsubara approach . we would like to stress that the tap approach furnishes an alternative and more transparent route to replicas which has also the advantage of showing explicitly the weakness of the mean field description . let us cite one example . the marginality prescription in the replica approach becomes the more transparent statement that the non - equilibrium dynamics is dominated by the tap states which are marginally stable , _ i.e. _ the flatness of the free - energy landscape around these states is responsible for aging . concerning one of the weaknesses of the mean field description we would like to underline that the enormous number of pure states ( with different free - energy densities ) found for mean field models can not persist in finite dimensions and the majority of them should become _ metastable _ states . how this changes the mean field scenario is an active domain of research for classical systems @xcite . we remark that interesting continuations of our work concern , on the one hand , the application of the tap approach to different quantum mean - field models @xcite and , on the other hand , the generalization of the static quantum tap approach to real - time dynamics ( for classical systems this has been done in @xcite ) . this would allow one to show the relationship between long - time dynamics and free - energy landscape for quantum systems directly . finally , the precise definition of a `` quantum state '' is a delicate matter and merits further analysis . in this paper we have simply called `` state '' a minimum of the tap free - energy density . one possible way to verify the existence and stability of these states is by studying the dynamics of this system starting from particular initial conditions as done in @xcite for the classical model . this study is underway @xcite . * acknowledgments * we wish to thank d. r. grempel , j. kurchan , g. lozano and c. a. da silva santos for very useful discussions . g. b. is supported by the center of material theory , rutgers university , nj , usa . l. f. c. thanks ecos - sud for a travel grant and financial support from the grant `` algorithmes doptimisation et systmes quantiques dsordonns '' , aci - jeunes chercheurs , 2000 . in this appendix we show that the equation for @xmath45 that leads to a solution with the correct physical properties is eq . ( [ eqq1final ] ) with the minus sign . indeed , we search a solution that corresponds to a minimum of the tap free - energy . the full stability analysis , that involves the evaluation of the complete hessian of the tap free - energy , is rather hard and has to be done numerically ( for instance , the form of @xmath273 can only be obtained numerically ) . + however , we can still perform a partial analysis that suffices to justify the choice of the minus sign . let us concentrate on the following diagonal elements of the hessian : @xmath274 \left[\frac{1}{(\tilde{c}(0)-\beta q_{\sc ea})^{2}q_{\sc ea}}-\frac{p(p-1)}{2}q_{\sc ea}^{p-3 } \right ] \ ; , \\ \label{c(0)-stab } \frac{\partial ( -\beta f ) } { \partial \tilde{c}(0 ) \partial \tilde{c}(0 ) } & = & \frac{-1}{(\tilde{c}(0)-q\beta ) ^{2 } } + \frac{p(p-1)}{2\beta } \int_{0}^{\beta } d\tau c^{p-2}(\tau ) \ ; .\end{aligned}\ ] ] from @xmath275 s definition we find @xmath276 since @xmath45 is fixed by eq . ( [ eqq1final ] ) , the second factor on the right - hand - side of eq . ( [ q - stab ] ) is positive ( negative ) for @xmath277 ( @xmath278 ) . a stable solution corresponds to a negative value of ( [ q - stab ] ) and ( [ c(0)-stab ] ) , therefore one has to take the solution @xmath279 for @xmath277 and @xmath280 for @xmath278 . moreover , since for @xmath281 the right hand side of ( [ eqq1final ] ) is positive then we obtain that @xmath282 and @xmath283 are positive and using that @xmath284 and imposing that ( [ c(0)-stab ] ) has to be negative , we obtain : @xmath285 but this is impossible for @xmath278 , i.e. it is not possible to have a consistent stable @xmath286 solution . in this appendix we give an argument in favor of the relation @xmath287 in eq . ( [ me ] ) . let us start by assuming that it does hold and see that it leads to the equation linking @xmath49 , @xmath45 and @xmath187 in the matsubara approach . first of all we write the derivative of @xmath248 with respect to @xmath59 as a derivative with respect to @xmath246 . this can be easily done by noticing that differentiating @xmath59 in eq . ( [ eq : gamma ] ) with respect to @xmath246 at @xmath288 and @xmath149 fixed is equivalent to differentiating @xmath59 with respect to @xmath246 at @xmath289 and @xmath290 fixed because @xmath59 is stationary in @xmath291 and @xmath290 . as a consequence we find @xmath292 the derivative @xmath293 can be easily computed from eq . ( [ complexity ] ) . one arrives at @xmath294 in summary , starting from the tap approach at fixed @xmath158 we obtain @xmath45 and @xmath187 . assuming then that eq . ( [ me ] ) holds we obtain the equation linking @xmath45 and @xmath49 in the matsubara approach . the equations for @xmath187 in the tap and matsubara approaches , once @xmath45 is fixed , are identical . hence we have proven that eq . ( [ me ] ) leads to the matsubara results in @xcite . in j. kurchan , j. phys . * i * it is shown that there is a problem with the calculation in @xcite since , for topological reasons , the number of tap solutions computed should actually be one . two solutions to this puzzle have been proposed , one in this article and another one in @xcite . j - p bouchaud , l. f. cugliandolo , j. kurchan and m. mzard , _ out of equilibrium dynamics in spin glasses and other glassy systems _ , in _ spin - glasses and random fields _ , a. p. young ed . ( world scientific , 1998 ) . e. vincent , j. hammann , m. ocio , j - p bouchaud and l. f. cugliandolo , _ slow dynamics and aging in glassy systems _ , in sitges 1996 , ed . m. rub ( springer - verlag , 1997 ) . p. nordblad and p. svendlidh , _ experiments on spin - glasses _ in _ spin - glasses and random fields _ , a. p. young ed . ( world scientific , 1998 ) . w. wu , b. ellmann , t. f. rosenbaum , g. aeppli and d. h. reich , phys . lett . * 67 * , 2076 ( 1991 ) . w. wu , d. bitko , t. f. rosenbaum and g. aeppli , phys . lett . * 71 * , 1919 ( 1993 ) . t. f. rosenbaum , j. phys . * c8 * , 9759 ( 1996 ) . j. brooke _ al _ , science * 284 * , 779 ( 1999 ) . e. courtens , j. phys . ( paris ) * 43 * l199 ( 1982 ) , phys . lett . * 52 * 69 ( 1984 ) . e. matsushita and t. matsubara , prog . phys . * 71 * 235 ( 1984 ) . r. pirc , b. tadic and r. blinc , z. phys . * b61 * 69 ( 1985 ) , phys . rev . * b36 * 8607 ( 1987 ) . s. rogge , d. natelson and d. d. osheroff , phys . lett . * 76 * , 3136 ( 1996 ) . s. rogge , d. natelson , b. tigner and d. d. osheroff , phys . tev . * b55 * , 11256 ( 1997 ) . d. natelson , d. rosenberg and d. d. osheroff , phys . * 80 * , 4689 ( 1998 ) . p. shukla and s. singh , phys . lett . * a81 * , 477 ( 1981 ) . t. yamamoto and h. ishii , j. phys . * c20 * , 6053 ( 1987 ) . v. dobrosavleljevic and d. thirumalai , j. phys . * a23 * l767 , ( 1990 ) . j. miller and d. a. huse , phys . lett . * 70 * , 3147 ( 1993 ) . y. y. goldschmidt , phys . * b41 * , 4858 ( 1990 ) . t. kope , phys . rev . * b52 * , 9590 ( 1995 ) . t. giamarchi and p. le doussal , phys . * b53 * 15206 ( 1996 ) . d. r. grempel and m. j. rozenberg , phys . 80 * , 389 ( 1998 ) ; _ ibid _ * 81 * , 2550 ( 1998 ) . a. georges , o. parcollet and s. sachdev , phys . . lett . * 85 * , 840 ( 2000 ) . d. s. fisher , phys . lett . * 69 * , 534 ( 1992 ) . f. igloi and h. rieger , * b57 * , 11404 ( 1998 ) . m. y. guo , r. n. bhatt and d. a. huse , phys . lett . * 72 * , 4137 ( 1994 ) . h. rieger and a. p. young , lett . * 72 * , 4141 ( 1994 ) . a. h. castro neto and b. a. jones , cond - mat/0003085 . o. motrunich , k. damle , d. a. huse , phys . lett . * 84 * , 3434 ( 2000 ) and cond - mat/0005543 .

intracluster planetary nebulae are a useful tracer of the evolution of galaxies and galaxy clusters . we analyze our catalog of 318 intracluster planetary nebulae candidates found in 0.89 square degrees of the virgo cluster . we give additional evidence for the great depth of the virgo cluster s intracluster stellar population , which implies that the bulk of the intracluster stars come from late - type galaxies and dwarfs . we also provide evidence that the intracluster stars are clustered on the sky on arcminute scales , in agreement with tidal - stripping scenarios of intracluster star production . although significant systematic uncertainties exist , we find that the average fraction of intracluster starlight in the virgo is , and may be higher if the intracluster stars have a large spatial line - of - sight depth . we find that the intracluster star density changes little with radius or projected density over the range surveyed . these results , along with other intracluster star observations , imply that intracluster star production in virgo is ongoing and consistent with the cluster s known dynamical youth .

astro-ph0407274

the concept of intracluster starlight was first proposed by @xcite , who claimed to detect excess light between the galaxies of the coma cluster . follow - up photographic searches for intracluster luminosity in coma and other rich clusters ( e.g. , * ? ? ? * ; * ? ? ? * see vchez - gmez 1999 and feldmeier 2000 for reviews ) produced mixed results , and it was not until the advent of ccds that more precise estimates of the amount of intracluster starlight were made ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? however , even these modern measurements carry a substantial uncertainty , due to the extremely low surface brightness of the phenomenon . typically the brightest intracluster regions are less than @xmath1 of the brightness of the night sky in the @xmath2 band ( e.g. , * ? ? ? * ) , and measurements of this luminosity must also contend with the problems presented by scattered light from nearby bright objects and the contribution of faint discrete sources . recently , a great deal of progress has been made in the study of intracluster light ( icl ) . our ability to measure the icl in abell clusters has grown to the point that reliable surface photometry is now possible at levels several magnitudes ( 0.1% to .01% in the @xmath2 and @xmath3 bands ) lower than that of the sky ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? these studies are making it possible to study the evolution of the icl as a function of redshift , cluster substructure , galactic density , and galactic morphology . meanwhile , high - resolution n - body models now have the ability to follow hundreds of interacting cluster galaxies within a cosmological context @xcite . these high - quality simulations are making testable predictions concerning the production of intracluster light . when combined with earlier and recent semi - analytic studies @xcite there is now sufficient theoretical insight to begin to interpret the observations . another major advance in the study of the icl is our growing ability to study individual intracluster objects such as globular clusters @xcite and in particular , individual intracluster stars , specifically planetary nebulae ( ipne ) , red giants ( irgs ; @xcite ) and supernovae ( isne ; @xcite ) . perhaps the most useful of these objects are the planetaries . ipne have several unique features that make them ideal for probing intracluster starlight ( for a complete review , see * ? ? ? in the light of [ ] @xmath45007 , planetary nebulae ( pne ) are extremely luminous , and can be observed out to distances of @xmath5 mpc with present day telescopes . planetary nebulae are also excellent tracers of stellar luminosity ( ciardullo , jacoby , & ford 1989 ; ciardullo et al . 1989 ) , and their number counts can be scaled to produce estimates of the total amount of unseen light . in addition , through the [ ] @xmath45007 planetary nebula luminosity function ( pnlf ) , pne are precise distance indicators ( jacoby et al . 1992 ; ciardullo 2002b ) ; therefore , the observed shape of the pnlf provides information on the line - of - sight distribution of the intracluster stars . finally , since the [ ] @xmath45007 emission is in a narrow line , ipn velocities can be determined via moderate ( @xmath6 ) resolution spectroscopy , making kinematical studies of the icl possible . in this paper , we utilize the 318 ipn candidates detected in our photometric survey of virgo ( feldmeier , ciardullo , & jacoby 1998 ; hereafter paper i ; feldmeier et al . 2003a ; hereafter paper ii ) to place limits on the amount and distribution of intracluster starlight in that cluster . in 2 , we review the observations and reduction procedures which produced the ipn catalog . in 3 , we use the ipn data to estimate the three - dimensional structure of virgo s diffuse stellar component . we compare the distribution of icl to that of the galaxies and the hot intracluster medium , and present evidence that virgo s ipne are clumped on small scales . in 4 , we compute the total amount of intracluster starlight that must be present in virgo , discuss the systematic uncertainties present in this number , and compare the large scale distribution of ipne with that of two common parameters of cluster dynamical evolution : cluster radius and projected galaxy density . we discuss the implications of our results in 5 and combine our data with the observation of intracluster red giants , to place constraints on the population of the intracluster stars and on their origins . we stress that not all of the 318 candidates are genuine ipne : we estimate ( ciardullo et al . 2002a ) that @xmath7 20% of the ipn candidates are likely background objects that are themselves of considerable scientific interest ( e.g. , kudritzki et al . 2000 ; rhoads et al . 2000 ; stern et al . 2000 ) . for this paper , we adopt @xmath8 mpc as the distance to the virgo cluster core . this number is consistent with the pnlf distances to the elliptical galaxies of the core @xcite , and enables us to compare our ipn measurements directly with the pnlf observations of ellipticals . we note that such a distance is @xmath9 mpc smaller than the mean distance estimated from cepheid variables by the _ hst _ distance scale key project and others ( pierce et al . 1994 ; saha et al . 1997 ; freedman et al . such a discrepancy is not serious , especially since there is growing evidence that the ellipticals of virgo are not necessarily at the same distance as the spirals @xcite . additionally , the median distances of the two methods are consistent within the errors , and the difference in the mean is solely due to the most distant galaxy ngc 4639 ( d = @xmath10 ; freedman et al . 2001 ) . in any case , all quantities in this paper that are dependent on this distance will be expressed as a function of @xmath11 , where @xmath12 mpc . the detection , photometry , and astrometry of the 318 ipn candidates were reported in full in paper ii . here , we briefly summarize the results . the images for this survey were taken over the course of three observing runs from march 1997 to march 1999 at the kitt peak national observatory ( kpno ) using the 4 m telescope and the prime focus ccd camera for runs 1 and 2 ( pfccd ) and the noao mosaic i imager ( mosaic ; muller et al . 1998 ) for run 3 . the data are composed of images taken through an [ ] @xmath45007 narrow - band filter , and a medium - band continuum filter . ipn candidates appear as point sources in the former filter , but disappear in the latter filter . the ipn candidates were detected semi - automatically using two separate methods , and objects in , or adjacent to , virgo cluster galaxies have been culled from our sample . in addition , any source known to be present in the continuum , or shown to be a background galaxy through follow - up spectroscopy has been removed . we stress , however , that only 10% of the sources have follow - up spectra , and those spectra that do exist have low signal - to - noise . we will return to this point later . after identifying the ipn candidates , we determined [ ] @xmath45007 magnitudes by comparing their fluxes to spectrophotometric standards defined by stone ( 1977 ) and massey et al . we define our [ ] @xmath45007 magnitudes in the same way as jacoby ( 1989 ) : @xmath13 where @xmath14 is in ergs @xmath15 s@xmath16 . equatorial coordinates were obtained for each intracluster planetary nebula candidate by comparing its position to those of usno - a 2.0 astrometric stars ( monet 1998 ; monet et al . 1996 ) on the same frame . the magnitude limit of each frame was determined by adding artificial stars to the data and estimating the recovery fraction as a function of magnitude ( e.g. , stetson 1987 ; bolte 1989 ; harris 1990 ) . to be conservative , the 90% completeness level has been adopted as the limiting magnitude for each field . the field locations are shown graphically in paper ii , figure 1 . the resulting luminosity functions for each field , binned into 0.2 magnitude intervals , are shown in figure [ fig : pnlf ] . the original data for field 1 has been superseded by the field 7 results ( paper ii ) . the first step towards understanding the mechanisms which produce intracluster stars is to compare the stars spatial distribution with that of the better - studied components of galaxy clusters : the cluster galaxies , the hot intracluster gas , and the invisible dark matter , which dominates the total mass . for example , if most intracluster stars are removed early in a cluster s lifetime ( merritt 1984 ) , the distribution of this diffuse component will be smooth and follow the cluster potential . however , if a significant portion of the stars are removed at late epochs via galaxy encounters and tidal stripping ( e.g. , richstone & malumuth 1983 ; moore et al . 1996 ) , then the icl should be clumpy and have a non - relaxed appearance . in particular , the `` harassment '' models of moore et al . ( 1996 ) and moore , lake , & katz ( 1998 ) first predicted that many intracluster stars will exist in long ( @xmath17 mpc ) tidal tails , which may maintain their structure for gigayears . since virgo is thought to be dynamically young ( tully & shaya 1984 ; binggelli , popescu & tammann 1993 and references within ) , it is possible that such structures are still present and observable . since planetary nebulae closely follow the distribution of starlight in galaxies ( e.g. , ciardullo , jacoby , & ford 1989 ; ciardullo et al . 1989 ) , they should be an ideal tracer of virgo s intracluster population . here , we detail our efforts to determine the spatial distribution of the intracluster stars through the analysis of the ipn data . from theory ( e.g. , dopita et al . 1992 ; stanghellini 1995 ; mndez & soffner 1997 ) , and observations ( e.g. , ciardullo et al . 2002b , and references therein ) , it has been shown that bright planetary nebulae follow a well - defined luminosity function in the light of [ ] @xmath45007 . in particular , the [ o iii ] planetary nebula luminosity function ( pnlf ) at the brightest magnitudes is well fit by an exponential function with a sharp bright - end cutoff @xmath18 , i.e. , @xmath19 since there is no reason to believe that ipne would not also follow the pnlf , it is possible to use this cutoff to probe distances in the intracluster environment . however , an implicit assumption of the pnlf method is that all the pne that make up the observed luminosity function are effectively at the same distance ( e.g. , in a normal galaxy ) . in a cluster such as virgo , the finite depth of the cluster can distort the pnlf , as ipne at many different distances can contribute to the observed luminosity function . in this case , the pnlf of the ipne will be the convolution of the intrinsic luminosity function and the density function of the ipne along the line - of - sight . thus , the shape of the observed luminosity function can be exploited to learn about the three - dimensional distribution of the icl . unfortunately , deriving the complete line - of - sight distribution is not trivial . mathematically , the function is : @xmath20 \ast e(m)\ ] ] where @xmath21 is the final observed luminosity function , @xmath22 is the empirical pnlf , @xmath23 is the density of intracluster pne as a function of distance modulus , @xmath24 , @xmath25 is the apparent luminosity function of any contaminating objects , and @xmath26 is the photometric error function of the observations . although the intrinsic luminosity function is known and the error function is , in principle , derivable from our photometry ( this function is not necessarily a gaussian ; * ? ? ? * ) , the luminosity function of the contaminants is not currently not well - known . this places a fundamental limit on our ability to measure @xmath23 . with an improved estimate of @xmath25 , we will be able to do a much better job at constraining this function . for the present , we focus on a much simpler case : the magnitude of the brightest ipn candidate in each field . given the empirical pnlf , there is a maximum absolute magnitude @xmath18 , that a pne may have in the light of [ ] @xmath45007 . if we assume that the brightest observed ipn candidate in each field has this absolute magnitude , we can derive its upper limit distance , and therefore estimate the distance to the front edge of the virgo intracluster population . for this paper , we adopt a value of @xmath18 of -4.51 ( ciardullo et al . 2002b ) . figure [ fig : depth ] displays our upper limit distances , along with similar limits found by other authors using similar techniques @xcite . field 8 has been excluded from the analysis , since the ipn density of the field is consistent with zero ( paper ii ) . the extraordinary ipn candidate i-33 of @xcite , which is over 0.8 magnitudes brighter than all other objects in our survey , has also been omitted from the diagram . for comparison , figure [ fig : depth ] also shows the cepheid - based distances to six virgo spirals @xcite and the pnlf distances of six virgo early - type galaxies @xcite . note that virgo is known to have a complex structure with at least two main subclusters : cluster a , which is dominated by the giant elliptical m87 ( and contains the background m84/m86 subgroup ; bhringer et al . 1994 ; schindler , binggeli , & b " ohringer 1999 ) , and cluster b , which contains the giant elliptical m49 @xcite . the respective subclusters for each field are labeled in the figure . several results are apparent from the data . first , there is a general offset in the upper limit distances between ipn observations in subclusters a and b. according to our data , the front of subcluster a is at a distance of 12.7 @xmath27 0.4 mpc , while subcluster b begins at 14.1 @xmath27 0.8 mpc . these distance estimates use all the datasets and include all possible systematic errors common to the pnlf method ( see ciardullo et al . 2002b for a discussion of these systematic effects ) . if we take an average of the brightest @xmath28 magnitudes of our fields ( 27 , and the m87 field of ciardullo et al . 1998 ) weighted solely by the photometric errors , we find that the average magnitude of the brightest ipne candidates in subcluster a is @xmath29 , while that of the brightest ipne in subcluster b is @xmath30 , corresponding to a difference at the 4.9@xmath31 level . if , instead of using just the brightest ipne , we fit the data to the empirical pnlf using the method of maximum likelihood ( ciardullo et al . 1989 ) , then the upper limit distance to each field decreases by @xmath9 mpc ( field 4 excepted , since that region does not contain enough planetaries for a reliable pnlf analysis ) . this systematic shortening of the distances is expected , since it is more probable that the brightest ipne have magnitudes slightly fainter than @xmath18 , rather than precisely at @xmath18 . nevertheless , the analysis confirms the fact that intracluster stars of subclump b is substantially more distant than those of subclump a. the systematic difference between the upper limit distances is most likely due to the internal geometry of virgo : observations by yasuda , fukugita , & okamura ( 1997 ) and federspiel , tammann , & sandage ( 1998 ) demonstrate that the galaxies of subcluster b have a mean distance modulus @xmath32 mag more distant than those of subcluster a. moreover , the two discrepant points in this relation , field 4 and subaru 1 , can be easily explained through other means . field 4 only has two objects in its photometrically complete sample , so small number statistics affect its distance determination . ( the probability of one of these pne being within @xmath33 mag of @xmath18 is only @xmath34% . ) subaru 1 , on the other hand , is centered on the m84/m86 subclump of cluster a , and this system is known to lie @xmath33 mag behind the virgo core ( e.g. , binggeli , popescu , & tammann 1993 ; bhringer et al . 1994 ; schindler , binggeli , & b " ohringer 1999 ; jerjen , binggeli & barazzi 2003 ) . therefore , the upper limit distance of subaru 1 probably reflects the distance to this background subgroup , and not the virgo cluster as a whole . a second property which is obvious from figure [ fig : depth ] is that the intracluster distance to subcluster a is clearly much closer than the 15 mpc assumed for the cluster core . as has been pointed out previously ( ciardullo et al . 1998 ; paper i ; arnaboldi et al . 2002 ) , this is due to a selection effect : it is much easier to detect an ipne located on the front edge of the cluster than one located on the back edge . thus , our upper limit ipn distances represent measurements to the front edge of virgo s icl , not the system s mean distance . nevertheless , the depth implied from our measurements is remarkable . if we take the data at face value , then the ipn distribution has a line - of - sight radius of 4.2 + ( 15 * ( d@xmath35 - 1 ) ) mpc . if we compare this radius to the classical radius of virgo on the sky ( @xmath36 , or @xmath37 mpc ; * ? ? * ; * ? ? ? * ; * ? ? ? * ) we find that the virgo cluster is more than 2.6 times as deep as it is wide . virgo is nowhere near a spherical cluster : it contains considerable substructure , and is elongated significantly along our line of sight . our inferred depth for the virgo cluster is much greater than the @xmath17 mpc value associated with the early - type galaxies of the cluster s core ( i.e. , * ? ? ? * ; * ? ? ? * ; * ? ? ? on the other hand , this depth agrees extremely well with that derived from tully - fisher observations of the cluster s spirals . according to @xcite and @xcite , virgo is highly elongated along the line - of - sight , with a width - to - depth ratio of @xmath38:4 , and an overall depth of @xmath39 mpc . much of this depth lies background to the core ellipticals , but the line of spirals does extend into the foreground . this conclusion is strengthened by the discovery of hydrogen deficient spirals on the outskirts ( @xmath40 mpc ) of virgo , which have likely moved through the cluster core @xcite . moreover , it is also in agreement with the results of @xcite , who find a virgo cluster depth of 6 mpc from sbf measurements of its dwarf ellipticals . since simulations @xcite show that the majority of intracluster stars are ejected into orbits similar to that of their parent galaxies , our measurements imply that the bulk of virgo s intracluster stars originate in late - type galaxies and dwarfs . if ipne were derived from elliptical galaxies , the depth of the icl would be much less . a complication from this analysis is that we might be inadvertently combining two different components of icl together in our analysis : a `` core '' component associated with the elliptical galaxies , and a component of stars removed from galaxy - galaxy interactions in the infalling regions of virgo . we believe this to be unlikely for a number of reasons . first , modern models of intracluster star production imply that the bulk of intracluster stars are released when their parent galaxies and groups pass through pericenter ( see mihos 2004a ; willman et al . 2004 for some graphical examples ) . the relative amount of intracluster star production is much less on the outskirts of a galaxy cluster . second , if such a component existed , the lower relative velocities of such encounters would imply that there would still be luminous `` parent '' galaxies nearby our ipn fields ( 100 - 200 kpc ) in the foreground , assuming that the interactions happened 1 - 2 gyr ago . this does not appear to be the case ( see 4.2 ) . finally , there is independent evidence that galaxies that have traveled through the cluster core can return to radial distances of @xmath41 mpc . as mentioned previously , observations of h i deficient spiral galaxies show depths that are even larger that those derived from the ipne @xcite . since the accepted explanation for h i deficiency is ram pressure stripping by the hot gas in the cluster core @xcite , the existence of these galaxies imply that at least some virgo objects have extended radial orbits . moreover , although the apocenter distances of these galaxies are somewhat uncertain , due to errors in their tully - fisher distances and other effects @xcite , there is good reason to believe that a depth of @xmath41 mpc is not unreasonable . using standard cosmological simulations , @xcite found that the maximum rebound radius reached by infalling galaxies is roughly between 1 and 2.5 times the cluster virial radii . when they scaled this result to the virgo cluster , @xcite found a corresponding rebound depth of 4.1 mpc , remarkably similar to that found for the ipn distribution . although the precise agreement between the infall models and the ipn distances is probably fortuitous , it does suggest that our measured depth for the intracluster stars is plausible . models by @xcite , which predict a maximum orbital radius of @xmath42 virial radii , further support our result . icl comparisons with the hot intracluster medium are less straightforward . multiple studies with _ rosat _ and _ asca _ , demonstrate that the x - ray emission from virgo s intracluster medium extends @xmath41 to 5 degrees ( @xmath43 mpc ) from m87 , and generally follows the galaxy distribution . however , the line - of - sight depth of this component is unclear . although intracluster gas is dissipative , and the majority of clusters are only mildly aspherical in their x - ray distributions ( mohr et al . 1995 ) , there are cases where the x - ray contours are considerably elongated ( up to a factor of three in depth ; mohr et al . 1995 ) . therefore , although the data suggest that the ipne are more widely distributed in depth than the gaseous intracluster medium , this can not be proven . our conclusions above have two caveats . first , from the spectroscopic observations of kudritzki et al . ( 2000 ) and freeman et al . ( 2000 ) , it is clear that there are contaminating objects as bright as @xmath44 present within our virgo survey data . therefore , the upper limits displayed in figure [ fig : depth ] could be affected by these interlopers . however , there are a number of reasons to believe that our estimates are accurate . the most powerful argument in support of our distance estimates is the magnitude offset between the results from subclusters a and b. it is difficult to see how this difference could be caused by background contaminating objects , since to first order , these objects are scattered uniformly across the sky . second , given a contamination probability of @xmath45 ( freeman et al . 2000 ; ciardullo et al . 2002a ) , it is unlikely that the majority of the bright ipn candidates are background galaxies . finally , the spectroscopic observations of ciardullo et al . ( 2002a ) definitively show that the brightest ipn candidates in front of m87 have [ ] @xmath45007 emission , and are not background objects . from these data , the icl extends at least @xmath17 mpc in front of virgo s central galaxy . similar results have been found by narrow - band h@xmath46 imaging and limited spectroscopic follow - up by @xcite and @xcite . a more troubling issue concerns the possibility of intracluster h ii regions . gerhard et al . ( 2002 ) recently discovered an isolated h ii region in the outer halo of ngc 4388 or possibly in intracluster space . such objects would emit [ ] @xmath45007 , and , if smaller than @xmath47 pc in size , would be indistinguishable from ipne without deep continuum images or balmer - line spectroscopy . fortunately , such objects are rare . even if these objects are distributed throughout virgo s intracluster space , then the density reported by gerhard et al . ( 2002 ) implies a contamination fraction of only @xmath48 . additionally , @xcite claim that the isolated h ii region discovered by gerhard et al . ( 2002 ) was formed from gas that was ram - pressure stripped from ngc 4388 , and now is falling back into the galaxy . if this is correct , it is less likely that large numbers of isolated h ii regions will be found far away from their parent galaxies ( however , see ryan - weber et al . 2003 for some examples ) . spectroscopic observations of the brightest ipne would remove any uncertainty . another useful way to probe the distribution of virgo s intracluster light is through the locations of the ipne on the sky . these are shown in figure [ fig : spatial ] . for reference , at the distance of virgo , @xmath49 corresponds to @xmath50 kpc . studying the ipne in this way allows us to use more of the ipn candidates in the analysis and investigate the intracluster light distribution on smaller spatial scales . it is obvious from the figure that at least some of our fields contain significant evidence of substructure . for example , of the 16 ipne of field 2 , none are present in the bottom third of the ccd frame . this segregation is not instrumental none of the other fields exhibit that property . from simple binomial statistics , the probability of the bottom third of the field being empty is less than @xmath51 . of course , the eye can be fooled , and _ a posteriori _ statistics are of questionable validity . what is needed is an unbiased statistical test of the spatial distribution of ipn candidates . the traditional approach for such an analysis is the two - point angular correlation function ( peebles 1980 ) , which is defined as the joint probability of finding two objects in two particular elements of solid angle on the sky . since the accuracy of this method depends on the number of object pairs in the sample , the limited of ipn candidates in fields 26 make this statistic too noisy for diagnostic purposes . a simpler procedure that requires less data is quadrat analysis ( see diggle 1983 and references therein ) . this method is a two - dimensional analog to the more common three dimensional `` counts - in - cells '' test used in cosmological surveys ( peebles 1980 ) , and was first used in studies of galaxy clusters by @xcite . like the two point correlation function , quadrat analysis compares the observed ipn distribution against a random distribution . unlike correlation functions , however , quadrat analysis has the ability , at least in theory , to test for the presence of sheets and filaments . the principle of quadrat analysis is simple . if one divides a field into @xmath52 regions , and the objects are distributed randomly , then the number of objects present within each region should be distributed as the @xmath53 statistic , @xmath54 where @xmath55 is the observed number of objects in region @xmath56 and @xmath57 is the mean surface density of objects multiplied by the region s area . note that there are no constraints on the size of each region , nor its shape . however , given the limited number of ipn candidates present in our data , we chose to simply divide our ccd fields into rectangles and test for large - scale non - uniformity . to implement our test , we began by dividing each ccd field into @xmath58 , @xmath59 , and @xmath60 rectangular regions , and calculating the effective area of each region . note that this second step was not straightforward : bright stars , galaxies , bad pixels , and gaps between the ccds in our mosaic fields ( fields 7 and 8) all act to limit our ipn survey area . to account for these effects , we very carefully masked out all those regions where the ipn detections would be abnormally difficult ( or impossible ) , and restricted our analyses to the areas outside these regions . our next step was to choose samples of ipne for analysis . our ipn detections on a frame are 90% complete down to a limiting magnitude of @xmath61 ; if we restrict our study to objects brighter than this , then we can be assured that our quadrat analysis will not be biased . however , it is possible ( and even likely ) that the probability of detecting an ipn below the completeness limit is independent of frame position , and only a function of shot noise . if this is the case , then we can use our entire sample of ipne ( which is typically a factor of @xmath17 larger than the statistically complete sample ) , and still obtain a result that is bias - free . to test for this possibility , we performed our quadrat analysis on both the statistically complete sample , and on our entire sample of ipn candidates . for fields 4 and 5 however , there are simply too few objects to measure the spatial distribution for the statistically complete sample . tables [ table : quad - all ] and [ table : quad - comp ] contain the results of our analysis . from the table , it is clear that the ipne in several of our fields are non - randomly distributed . the best evidence for this effect occurs in field 7 , where the statistically complete sample of ipne are non - randomly distributed at the 95% confidence level , and the entire sample of ipne are non - random with 99.9% confidence . fields 2 and 4 also show some evidence for clustering , although the signal is weaker . when their entire ipn samples in these fields are used , random distributions are excluded with 95% confidence . an examination of figure [ fig : spatial ] reveals that the features found by the quadrat analysis are easily seen by eye . the signal for field 2 comes from the absence of objects in the lower portion of the frame , while field 4 s non - random result is generated by the cluster of objects in the upper left - hand corner . finally , the large @xmath53 of field 7 comes from the east - west division of ipne and the two large areas where ipn candidates are seen . these non - uniformities range in scale from @xmath62 to @xmath63 , and are in agreement with the conclusions of okamura et al . ( 2002 ; see their figure 1 ) who find an inhomogeneous distribution of ipne in their subaru 1 field . how do these results compare with the `` harassment '' scenarios of moore et al . ( 1996 ) ? by both visual inspection of the moore et al . ( 1996 ) simulations , and order - of - magnitude arguments , the width of tidal debris arcs should be @xmath7 50 kpc , which corresponds to @xmath7 12@xmath64 in the virgo cluster . observations of such arcs in the coma and centaurus clusters ( trentham & mobasher 1998 ; gregg & west 1998 ; calcaeo - roldn , et al . 2000 ) , confirm that tidal tails are often @xmath65 to @xmath66 kpc in width ( @xmath67 ) . this implies that a tidal debris arc would fill a significant portion of a pfccd frame , and up to one one - third of a mosaic field . therefore , the non - uniformities detected in our analysis are more likely to indicate small scale inhomogeneities , either within a longer tidal arc , or possibly unrelated to arcs . more wide - field data will be needed to unambiguously search for long thin tidal debris arcs in the virgo cluster . we note that since the density of contaminating sources is approximately 20% , it is unlikely that they could have produced the signal found by our quadrat analysis . the evidence from both lyman - break galaxy surveys @xcite , and higher redshift lyman-@xmath46 galaxy surveys ( ouchi et al . 2003 ) , suggests that high redshift galaxies cluster on somewhat smaller angular scales ( @xmath68 ) than seen in our data . still , the best way to confirm our clustering results is to obtain velocities for the ipn candidates . if these objects are part of a tidal debris arc , or other related structure , they will be strongly correlated in velocity , as well as position . adding this third dimension is crucial to the interpretation of these data . in principle , determining the amount of intracluster luminosity from the observed numbers of ipne is straightforward . theories of simple stellar populations ( e.g. , renzini & buzzoni 1986 ) have shown that the bolometric luminosity - specific stellar evolutionary flux of non - star - forming stellar populations should be @xmath69 stars - yr@xmath16-@xmath70 , ( nearly ) independent of population age or initial mass function . if the lifetime of the planetary nebula stage is @xmath71 yr ( kwok 2000 ) , and if the pnlf of equation ( 2 ) is valid to @xmath72 mag below the pnlf cutoff , then every stellar system should have @xmath73 pn-@xmath70 . according to equation ( 2 ) , approximately one out of ten of these pne will be within 2.5 mag of @xmath18 . thus , under the above assumptions , most stellar populations should have @xmath74 pn-@xmath70 . the observed number of ipne , coupled with equation ( 2 ) , can therefore be used to deduce the total luminosity of the underlying stellar population . unfortunately , there are a number of systematic uncertainties associated with this conversion that must be addressed before we can use ipn surface densities to estimate the intracluster luminosity . we discuss each of these uncertainties below and determine a conservative correction for each effect . although stellar evolution predicts a constant @xmath75 value for all non star - forming populations , observations present a more complicated picture . ciardullo ( 1995 ) found that in a sample of 23 elliptical galaxies , lenticular galaxies , and spiral bulges , the observed value of @xmath75 never exceeded @xmath76 pn-@xmath70 but was often significantly less , with higher luminosity galaxies having systematically smaller values of @xmath75 , down to values of @xmath77 for m87 and @xmath78 for m49 . a lower value for @xmath75 has also been inferred for the populations of galactic globular clusters . the reason for this discrepancy is unclear . it is possible that in old or metal - rich populations , not all stars go through the planetary nebula phase @xcite ; such an hypothesis is supported by the anti - correlation between @xmath75 and the presence of extreme horizontal branch stars in elliptical galaxies @xcite . alternatively , the pnlf may not always follow the simple exponential law of equation ( 2 ) . ciardullo et al . ( 2004 ) have recently shown that in star - forming populations , the pnlf has a distinctive ` dip ' @xmath17 mag below the pnlf cutoff ; if similar features occur in older populations , then the extrapolation to the faint end of the luminosity function may be inaccurate . in either case , since the total amount of intracluster luminosity depends on @xmath75 , the variation of @xmath75 with stellar population can be a significant source of error , especially considering that our constraints on the age and metallicity of the stellar population are still weak ( paper i ; durrell et al . 2002 ) . to correct for this effect , we adopted the results of durrell et al . ( 2002 ) , who used the _ hubble space telescope _ to measure the luminosity function of red giants in virgo s intracluster space . models demonstrate that these data can be easily scaled to give the total luminosity of the intracluster population in this field ( e.g. , * ? ? ? * ) . by comparing the numbers of ipne in field 7 , which surrounds the _ hst _ field , with rgb and agb star counts , durrell et al ( 2002 ) found a value of @xmath79 pn-@xmath70 for virgo s intracluster population . this value is similar to that derived for the bulges of spiral galaxies , and we adopt it here for our luminosity calculations . however , even this determination has a substantial systematic uncertainty . as we have noted in 3.2 , the intracluster stars are not uniformly distributed on the sky , and the _ hst _ survey field is only @xmath80 the size of field 7 . thus there is the possibility of a density mismatch between the two measurements due to the spatial structure of the intracluster light . at this time , we can not determine the amount of this effect : additional intracluster red giant fields will be needed for a more accurate comparison . a second systematic error associated with our measurements comes from the fact that samples of ipn candidates are not pristine . approximately 20% of our ipn candidates are background objects of various kinds , of which the most common are lyman-@xmath46 galaxies at z @xmath81 3.12 ( e.g. , kudritzki et al . 2000 ; rhoads et al . 2000 ; stern et al . the only absolutely secure way to eliminate these contaminants is through spectroscopy : true ipne have narrow ( @xmath34 km s@xmath16 ) emission lines , show additional flux at [ o iii ] @xmath82 , and have an [ o iii ] @xmath83 to h@xmath46 emission line ratio of @xmath84 @xcite . this distinguishes ipne from background galaxies ( which have broader lines and no companion emission ) or intergalactic h ii regions ( which usually have an h@xmath46 line that is brighter than [ o iii ] @xmath85 ) . unfortunately , since we do not yet have follow - up spectroscopy for our ipn candidates , we must statistically remove the background contaminants using the methods detailed in paper ii and in ciardullo et al . ( 2002a ) . using a set of control fields , and identical search procedures as the virgo ipn survey , we determined the surface density of ipn - like objects , and adopted that density as our mean background ( 69@xmath27 23 per square degree brighter than @xmath86 ) . this density is corrected in each field for any known background objects ( paper ii ) . there are a number of complications to this background subtraction . first , as noted in paper ii , the ciardullo et al . ( 2002a ) survey only covered 0.13 deg@xmath87 of sky , and thus detected only nine unresolved ipne - like objects . consequently , the estimate of background density has a large poissonian error . of greater concern is the effect that cosmic large - scale structure has on the background source counts . recent observations by ouchi et al . ( 2003 ) find that lyman-@xmath46 galaxies are already clustered at z=4.86 , which implies even greater clustering at z @xmath81 3 . combined with the z @xmath88 lyman - break galaxy results of @xcite and @xcite , it is almost certain that our background estimation is affected by the large - scale structure exhibited by emission - line galaxies . we do note that our contamination fraction is similar to the @xmath89 value found by freeman et al . ( 2000 ) from limited follow - up spectroscopy and the @xmath90 estimate of @xcite based on a narrow - band survey of a field in the leo i galaxy group . ultimately additional blank field measurements and spectroscopic follow - ups will be needed to further reduce the uncertainty . finally , as noted in 3.2 , the ipn candidates of virgo have a significant line - of - sight depth . because our observations do not reach all the way through the cluster , our density determination is biased : we are more likely to detect ipne on the near side of the cluster than the far side . in paper i , we created two extreme models for the ipn distribution : a minimal icl model , wherein all the ipne are at a single distance , and a maximum icl model , in which the intracluster stars are distributed isotropically throughout a sphere of radius 3 mpc . the difference in densities between these two models is a factor of @xmath42 . therefore , we are faced with a quandary : do we assume a spherically symmetric model , and multiply the measured densities by a factor of three , or do we adopt the single distance model that we know to be incorrect ? in order to be conservative , we have chosen the latter option , and adopt a single distance model for all of our intracluster fields . in the future , when our studies of the line - of - sight densities are completed , we will be able to obtain a more accurate answer . in conclusion , although there are significant systematic uncertainties relating the number of ipne to the total amount of starlight , we believe we have reasonable corrections to these effects which are conservative , and supported by the data . to determine the amount of luminosity in each of our survey fields , we began by adopting the pnlf of equation ( 2 ) , and convolving it with the photometric error versus magnitude relation derived from our photometry . we then computed the likelihood of this model luminosity function fitting the observed statistical sample of pne as a function of two variables , system distance and total pn population @xcite . we then integrated over the former variable , reduced the latter variable by the fraction of contamination expected for each field , and then translated the pn populations into intracluster luminosity via our best - fit value for @xmath75 . these results appears in table [ table : luminosity ] , column 2 . the quoted errors are the formal uncertainties associated with the maximum likelihood solutions ; these are usually asymmetric due to the shape of the [ ] @xmath45007 planetary nebulae luminosity function . in the case of field 3 , we have shown previously that the pne in this field are a combination of ipne , and pne in the halo of m 87 proper , and that at least 10 of these objects can legitimately called ipne ( paper i ) . therefore , to place limits on this field s intracluster light , we assumed that 1 ) all of the pne candidates in this field are intracluster , or 2 ) that only the brightest 10 candidates are intracluster . these two models are referred to as `` 3h '' and `` 3l '' respectively . for fields 4 and 5 , due to the very small number of objects above the completeness limit , the uncertainties are poorly constrained . in these cases , we replaced the maximum - likelihood error bars with the errors expected from poissonian statistics . for field 8 , the ipn density is consistent with zero , so no analysis has been performed in this region . as expected , the amount of intracluster light is significant : the intracluster luminosity contained within our average pfccd survey field is comparable to that of the large magellanic cloud . the implied bolometric luminosity surface densities , denoted as @xmath91 , are given in table [ table : luminosity ] , column 3 . for reference , these densities are roughly @xmath66 times less than that of the milky way s disk at the solar neighborhood @xcite . if we adopt a bolometric correction consistent with a late - type stellar population , ( @xmath92 ) , these bolometric luminosity surface densities can be translated into @xmath2-band surface brightnesses . these are given in table [ table : luminosity ] , column 4 . although the surface brightnesses are quite faint ( 56 magnitudes below that of the night sky ) , they are reachable with the newer generation of ccd icl surveys now being performed , provided that precise and accurate sky subtraction can be achieved ( gonzalez et al . 2000 ; feldmeier et al . 2002 ; krick & bernstein 2003 ; mihos et al . 2004 , in prep . ) . with the luminosity of virgo s intracluster stars now determined , we can next compare this number to the amount of light contained within the cluster s galaxies . this is non - trivial , because as we mentioned previously , the structure of virgo is unrelaxed , and difficult to model accurately . to estimate this quantity , we used the galaxy catalog of @xcite to find the total @xmath93-band magnitudes of all member galaxies brighter than the catalog s limit of @xmath94 . ( this limit is adequate for our purpose , since the contribution of fainter objects to virgo s total luminosity is negligible . ) we then converted these @xmath93-magnitudes to @xmath3 , using the observed @xmath95 colors of early - type galaxies @xcite and a mean @xmath96 color for late - type galaxies @xcite . finally , with the @xmath3 magnitudes of all the galaxies in hand , we computed the radial profile of virgo in bolometric light , using an @xmath3-band bolometric correction of @xmath97 . note that to do this , we must determine the center of the cluster . @xcite have shown that the luminosity density of virgo peaks @xmath98 nw of the central elliptical , m87 . however , based on the cluster s kinematics @xcite , distances @xcite , and x - ray properties @xcite , it is likely that this offset is due to the contribution of the m84/m86 group , which is falling into virgo from behind . we therefore used m87 as the central point , and computed the @xmath3-band luminosity density of virgo s galaxies in a series of circular apertures centered on the galaxy . the luminosity surface densities we derive are given in table [ table : luminosity ] , column 5 , which we denote as @xmath99 . for fields 2 and 6 , which are in subcluster b of virgo , the circular aperture model derived above is not applicable , since that subcluster has an irregular galaxy distribution . therefore , to set a robust lower limit on the fraction derived for fields 2 and 6 , we assumed a value for @xmath99 that is representative of the center of subcluster a. finally , we computed the fraction of intracluster light in each field by dividing @xmath91 by the sum of @xmath99 and @xmath91 . at this point , to make error analysis more tractable , we also adopted the average uncertainty of our asymmetric density error bars . based on all of these assumptions , we derive intracluster star fractions between 4% and 41% , depending on the field . when averaged over all our fields , the fraction of intracluster light in virgo is 18.5% @xmath27 3.0% ( excluding field 8) and 15.8% @xmath27 3.0% ( with field 8 included ) . the quoted uncertainties include only the random component ; the systematic uncertainties are dominated by the uncertain line - of - sight depth of the intracluster stars , and the error associated with our estimate of @xmath75 . we estimate that these systematic errors could be as large as 5% . however , by assuming a single - limit distance for the intracluster starlight , we have underestimated the total amount of intracluster light , and therefore the fractions quoted above are conservative . to ensure that our values for @xmath99 are reasonable , we have also calculated local luminosity surface densities for each field to compare against our ipn luminosity densities . to do this , we defined two densities , @xmath100 and @xmath101 , to be the amount of bolometric luminosity density associated with the 25 and 50 @xcite galaxies closest to the center of each of our survey fields , omitting clear background galaxies . we then used these densities to normalize our ipn measurements in exactly the same way as described above . the results are given in table [ table : localluminosity ] . for any individual field , there can be a large difference between the locally and globally derived luminosity fractions . this is expected , given the clumpy , unrelaxed state of the cluster , and the limited number of luminous galaxies near each field . however , the weighted mean of these two local determinations for the entire cluster is 15.7% @xmath27 1.4% ( including field 8) , essentially identical to our previous result . these fractions agree with single field studies of virgo s intracluster population . arnaboldi et al . ( 2002 ) found a fraction of at least 17% from the observations of field rcn1 , and okamura et al . ( 2002 ) , and arnaboldi et al . ( 2003 ) estimated a fraction of 10% from ipn data of the subaru 1 field . similarly , ferguson , tanvir & von hippel ( 1998 ) found fractions of between 4% and 12% using _ hst _ observations of irg s , and durrell et al . ( 2002 ) obtained a fraction of 15@xmath102% using the red giant measurements of two _ hst _ fields . however , it is important to note that each of these studies makes different assumptions concerning the calculation of intracluster luminosity , the amount of light bound to galaxies , and the contamination of background sources . so none of these numbers is unassailable . how much stellar mass does this amount of intracluster light imply ? if we take our value for the median luminosity surface density of intracluster light ( @xmath103 kpc@xmath87 ) , and assume a bolometric correction of @xmath92 ( a value typical of older stellar populations ) and a mass - to - light ratio of @xmath104 @xcite , then the derived surface mass density of virgo s intracluster stars is @xmath105 mpc@xmath87 d@xmath106 . although this mass density seems large , when compared to a detailed mass model of the three virgo subclusters @xcite , and after correcting for the different assumed distances to the virgo core in the two calculations ( 15 mpc versus 20 mpc ) , we find that the intracluster stars add on average of @xmath81 17% to the stellar mass of the cluster . in any case , as in most galaxy clusters , the majority of the baryonic mass of virgo is in the intracluster gas @xcite , and the baryonic mass itself is only a fraction of the total mass . intracluster stars do not solve the missing - mass problem in galaxy clusters , as was originally determined by @xcite . the radial distribution of the ipne compared to that of the cluster light is shown in figure [ fig : radial ] . as can be clearly seen , the ipn density falls slowly , if it all , and certainly more slowly than that of the galaxies . if we fit the ipn densities with a simple power law in radius we find that the least - squares slope is @xmath107 . given the uncertainties associated with the measurements , this result is marginally consistent with the simulations of dubinski ( 1998 ) , and napolitano ( 2003 ) , which predict like profiles for the intracluster light . since galaxy morphology correlates strongly with galactic density @xcite , it is conceivable that ipne density also correlates with this quantity . to test this hypothesis , we obtained the local projected galaxy density for each field by taking the catalog of @xcite , and finding all virgo galaxies within 100@xmath11 kpc in radius ( @xmath108 ) of each field center . by using this catalog , we are sensitive to objects at least as faint as @xmath109 , and in some cases , two magnitudes fainter , which puts our density measurement into the domain of intermediate dwarf galaxies . this number density , denoted as n@xmath110 , is given in table 3 , column 7 , and plotted against the ipn density in figure [ fig : dens ] . as can be clearly seen , there appears to be little or no correlation between intracluster light density and projected galaxy density . after fitting the data to a simple power - law , we find a slope of -0.1 @xmath111 in units of log @xmath112 . this null result is somewhat surprising , since if intracluster light is produced from galaxy interactions , a correlation with density might be expected . we repeated this test over radii of 200 , 500 , and 1000@xmath11 kpc respectively , and found similar null results ( no positive slope at greater than the 2 @xmath31 level ) . these null results may be due to the significant scatter in the projected density , as well as the known uncertainties in the intracluster stellar densities . alternatively , the dynamic range of density surveyed in this analysis may be too small ( a factor of @xmath17 ) compared to that needed to see the morphology - density relation . finally , our measurements may suffer from an unavoidable selection effect . in order to avoid confusion with pne which are bound in the halos of galaxies , we must conduct our intracluster light survey in regions where there are no bright objects . in the past , these locations may have been occupied by bright galaxies , and the intracluster stars may have been released from these objects . however , given the large amount of time that could have elapsed between the release of these stars and the present epoch , it is possible that the ipne have drifted more than the adopted 1001000@xmath11 kpc radii . more intracluster observations over a larger range of cluster densities will be needed to address this issue further . when we combine our own ipn data , the results from intracluster red giant observations @xcite , and other ipn surveys @xcite , we come to a reasonably consistent picture of virgo s stellar intracluster population . there is a moderate ( 1020% ) amount of intracluster light in the virgo cluster , significantly more than found in the galaxy groups of m 81 ( @xmath113 3% ; feldmeier et al . 2003b ) and leo i ( @xmath114 ; castro - rodrguez et al . 2003 ) , but much less than is present in the rich clusters of coma ( @xmath7 50% ; bernstein et al . 1995 ) , abell 1689 ( @xmath81 30% ; tyson & fischer 1995 ) , and the compact group hcg 90 ( 38%48% ; * ? ? ? this is consistent with the view that virgo is a dynamically young cluster still in the process of formation @xcite . as virgo dynamically evolves , the amount of intracluster light should increase . from the large depth derived from the ipne , it seems clear that a significant portion of virgo s intracluster stars originate from late - type galaxies whose highly radial orbits take them in and out of the cluster core . the intermediate value of @xmath75 also strongly implies that the bulk of the ipne come from lower - luminosity galaxies , and not giant ellipticals ( m@xmath115 ; ciardullo 1995 ) . this conclusion is supported by the intracluster red giant star observations of @xcite . their analysis of the intracluster red giant luminosity function shows that virgo s intracluster stars have a mean metallicity in the range @xmath116 } < -0.2 $ ] . such a value is more metal - rich than a typical dwarf galaxy @xcite , yet more metal - poor than the giant ellipticals that inhabit the virgo cluster core ( i.e. , * ? ? ? * ; * ? ? ? moreover , the value of @xmath75 derived for the ipn population , and the ipn properties described in @xcite and paper i suggest that the intracluster stars of virgo are not extremely old or young . if the ipn were extremely old , then the depth of the virgo cluster would be even higher than its large derived value ( paper i ) , and measurements of @xmath75 in young stellar populations such as the magellanic clouds give values near the theoretical maximum ( ciardullo et al . these facts , along with the lack of correlation between ipn density and galactic environment , imply that the production of intracluster stars is an ongoing process . the above properties appear consistent with tidal - stripping from galaxy interaction ( richstone & malumuth 1983 ) scenarios such as `` galaxy harassment '' ( moore et al . 1996 ; moore , lake , & katz 1998 ) `` tidal stirring '' ( mayer et al . 2001 ) , and `` galaxy threshing '' ( bekki et al . 2003 ) and are inconsistent with tidal stripping by the mean cluster field very early in the cluster s lifetime ( merritt 1984 ) . however , a few caveats are in order . first , since we derive properties through the ipne , any stellar population that is extremely old and metal - poor ( jacoby et al . 1997 ) , or very metal - rich ( ciardullo 1995 ) , will not produce large numbers of ipne , and so will be missed in our survey . if there was a large amount of intracluster star production at early times , it would be extremely old , and we would be insensitive to this population . the close agreement between the irg and the ipn observations makes the existence of a large pop ii / iii component unlikely , but some extremely old intracluster stars may still exist . also , since ipne are weighted by stellar luminosity , the properties of rarer populations , such as those produced by tidally - stripped dwarf galaxies , will be poorly represented in our survey . it is almost certain that some of virgo s intracluster light originated from fainter dwarfs that have undergone severe tidal disruption and now exist as a population of ultra - compact dwarf galaxies ( drinkwater et al . 2003 ; karick et al . 2003 ) . there are a number of further studies that can improve the results given in this paper . additional ipn fields , especially ones at larger cluster radii and at lower galaxy densities , would be invaluable in confirming our null results with respect to galaxy density and radial distribution . deep imaging of a single field would also assist in better defining the line - of - sight distribution of ipne . but the most important observations needed are follow - up spectroscopy for a large number of candidate ipne . such observations would remove the uncertainty in our results due to background sources , produce better ipn magnitudes , and most critically , allow comparison of virgo s intracluster population with dynamical models . we thank m. arnaboldi , k. freeman , r. kudritzki , r. mndez , and chris mihos for useful discussions , and their infinite patience . we also thank an anonymous referee for their comments which improved this paper . j. feldmeier acknowledges salary support from j.c . mihos under grant ast-9876143 . this work was also partially supported by nsf grant ast 0071238 , ast 0302030 , and nasa grant nag 5 - 9377 . feldmeier , j.j . , durrell , p.r . , ciardullo , r. , & jacoby , g.h . 2003b , in iau symposium no . 209 planetary nebulae : their evolution and role in the universe , ed . s. kwok , m. dopita , & r. sutherland ( san francisco : asp ) , 605 krick , j. e. , & bernstein , r. 2003 , in carnegie observatories astrophysics series , vol . 3 : clusters of galaxies : probes of cosmological structure and galaxy evolution , ed . j. s. mulchaey , a. dressler , and a. oemler ( pasadena : carnegie observatories ) ccccc 2 & 2 & 7.48 & 3 & 0.06 + 2 & 3 & 10.9 & 8 & 0.21 + 2 & 4 & 26.3 & 15 & 0.04 + + 3 & 2 & 5.03 & 3 & 0.17 + 3 & 3 & 10.4 & 8 & 0.24 + 3 & 4 & 11.2 & 15 & 0.74 + + 4 & 2 & 2.73 & 3 & 0.43 + 4 & 3 & 14.2 & 7 & 0.05 + 4 & 4 & 20.2 & 15 & 0.17 + + 5 & 2 & 3.91 & 3 & 0.27 + 5 & 3 & 1.98 & 8 & 0.98 + 5 & 4 & 9.19 & 15 & 0.87 + + 6 & 2 & 4.08 & 3 & 0.25 + 6 & 3 & 6.51 & 8 & 0.59 + 6 & 4 & 12.1 & 15 & 0.67 + + 7 & 2 & 2.91 & 3 & 0.41 + 7 & 3 & 14.9 & 8 & 0.06 + 7 & 4 & 44.3 & 15 & @xmath117 + + 8 & 2 & 4.07 & 3 & 0.25 + 8 & 3 & 5.69 & 8 & 0.68 + 8 & 4 & 15.8 & 15 & 0.39 + ccccc 2 & 2 & 3.88 & 3 & 0.28 + 2 & 3 & 5.96 & 8 & 0.65 + 2 & 4 & 21.3 & 15 & 0.13 + + 3 & 2 & 6.34 & 3 & 0.10 + 3 & 3 & 12.3 & 8 & 0.14 + 3 & 4 & 14.2 & 15 & 0.51 + + 6 & 2 & 5.71 & 3 & 0.13 + 6 & 3 & 6.02 & 8 & 0.65 + 6 & 4 & 14.1 & 15 & 0.52 + + 7 & 2 & 1.29 & 3 & 0.73 + 7 & 3 & 7.24 & 8 & 0.51 + 7 & 4 & 25.2 & 15 & 0.05 + + 8 & 2 & 6.16 & 3 & 0.10 + 8 & 3 & 10.4 & 8 & 0.24 + 8 & 4 & 17.3 & 15 & 0.30 + ccccccc 2 & 3.9@xmath118 & 8.4 & 27.4 & 15.2 & 0.35 @xmath27 0.22 & 12 + 3h & 9.0@xmath119 & 19.9 & 26.5 & 107 & 0.16 @xmath27 0.02 & 26 + 3l & 2.0@xmath120 & 4.5 & 28.1 & 107 & 0.04 @xmath27 0.01 & 26 + 4 & 2.4@xmath121 & 8.3 & 27.4 & 13.5 & 0.37 @xmath27 0.26 & 34 + 5 & 3.5@xmath122 & 9.5 & 27.3 & 13.5 & 0.41 @xmath27 0.23 & 11 + 6 & 1.7@xmath123 & 4.5 & 28.1 & 15.2 & 0.23 @xmath27 0.08 & 20 + 7 & 6.5@xmath124 & 3.2 & 28.4 & 13.5 & 0.19 @xmath27 0.04 & 31 + 8 & 0 & & & & & 21 + weighted avg . & & & & & 0.158 @xmath27 0.03 & + cccccc 2 & 8.4 & 4.2 & 0.67 @xmath27 0.42 & 18.9 & 0.31 @xmath27 0.20 + 3h & 19.9 & 48.8 & 0.29 @xmath27 0.04 & 23.4 & 0.46 @xmath27 0.06 + 3l & 4.5 & 48.8 & 0.08 @xmath27 0.02 & 23.4 & 0.16 @xmath27 0.05 + 4 & 8.3 & 7.7 & 0.52 @xmath27 0.36 & 6.0 & 0.58 @xmath27 0.41 + 5 & 9.5 & 11.1 & 0.46 @xmath27 0.26 & 7.8 & 0.55 @xmath27 0.31 + 6 & 4.5 & 41.4 & 0.10 @xmath27 0.04 & 13.9 & 0.24 @xmath27 0.08 + 7 & 3.2 & 15.9 & 0.17 @xmath27 0.03 & 11.3 & 0.22 @xmath27 0.04 + 8 & 0 & & & & + weighted avg . & & & 0.133 @xmath27 0.02 & & 0.218 @xmath27 0.03 +

polarization observables in neutron - deuteron scattering are calculated to next - to - next - to - next - to - leading order ( ) in pionless effective field theory ( ) . at the two - body-wave contact interactions are found to be important contributions to the neutron vector analyzing power , , and the deuteron vector analyzing power , . extracting the two - body-wavecoefficients from two - body scattering data and varying them within the expectedtheoretical errors provides results that are consistent ( at the level ) with experimental data at low energies . cutoff dependence of the correction of the doublet-wave scattering amplitude suggests the need for a new three - body force at , which is likely one that mixes wigner - symmetric and wigner - antisymmetric three - body channels .

1512.03774

the maximum of the nucleon vector analyzing power , @xmath3 , and the deuteron vector analyzing power , @xmath4 , in neutron - deuteron ( @xmath7 ) and proton - deuteron ( @xmath8 ) scattering are significantly underpredicted by existing three - body calculations at low energies ( nucleon laboratory energies below @xmath9 mev ) . this is known as the three - nucleon analyzing power problem or as the @xmath10 puzzle ( see , e.g. , refs . @xcite ) . both phenomenological potential model calculations ( pmc ) @xcite and more modern potentials derived from chiral effective field theory @xcite have not resolved this discrepancy . pmc have shown that @xmath10 is very sensitive to the values of the two - body @xmath11 phase shifts @xcite and that the experimental neutron - proton ( @xmath12 ) scattering data does not give enough latitude to simultaneously fit the two - body @xmath11 wave phase shifts and @xmath10 @xcite . for low energies ( @xmath13 ) nuclear systems can be described by a theory containing only contact interactions between nucleons and possible external currents . this theory , known as pionless effective field theory ( @xmath1 ) ( see , e.g. ref . @xcite for a review ) , has been used to calculate nucleon - nucleon ( @xmath14 ) scattering @xcite , deuteron electromagnetic form factors @xcite , @xmath12 capture @xcite , and neutrino - deuteron scattering in the two - body sector @xcite . progress in the three - body sector includes the calculation of @xmath7 scattering @xcite and @xmath8 scattering @xcite . the first calculations of @xmath7 scattering relied on the partial resummation technique @xcite , which resummed certain higher order contributions and therefore were not strictly perturbative in the @xmath1power counting . references @xcite introduced a technique to calculate @xmath7 scattering amplitudes strictly perturbatively that is no more numerically expensive than the partial resummation technique . with the strictly perturbative approach @xmath7 scattering was calculated to next - to - next - to - leading order ( @xmath15 ) including the two - body @xmath16-mixing term @xcite . the two - body @xmath16-mixing term provides the first non - zero contribution to the polarization observables in @xmath7 scattering . however , that work did not investigate polarization observables since they are not expected to be reproduced well at @xmath15 . the @xmath15 calculation is lo in the polarization observables since it gives the first non - zero contribution to them . building on this @xmath15 calculation , we present in this paper results for polarization observables in @xmath7 scattering to next - to - next - to - next - to - leading order ( @xmath0 ) in @xmath1 . at @xmath0 there are new contributions from shape parameter corrections as well as the two - body @xmath2-wave contact interactions that are found to give the dominant contribution to @xmath10 ( as already identified in pmc ) . at @xmath15 a term that causes splitting between the neutron - neutron ( @xmath17 ) and @xmath12 @xmath18 scattering lengths should be included . experimentally the difference in scattering lengths is roughly 3% and therefore numerically is considered a @xmath15 correction in the @xmath1power counting . this term has been included in previous two - body calculations @xcite , but was not treated strictly perturbatively as it is in this work . this paper is organized as follows . in sec . [ sec:2b ] all necessary two - body physics for the three - body calculation is presented . section [ sec:3b ] discusses how the @xmath7 scattering amplitudes are calculated and introduces the concept of @xmath2-wave auxiliary fields to calculate the three - body contribution from the two - body @xmath2-wave contact interactions . in sec . [ sec : observables ] expressions for the polarization observables in @xmath7 scattering are introduced . polarization observable results in @xmath1are shown in sec . [ sec : results ] , and finally we conclude in sec . [ sec : conclusion ] . the two - body @xmath6-wave lagrangian up to and including @xmath0 in the @xmath19-parametrization @xcite is given by @xmath20+y_{s}\left[\hat{s}_{a}^{\dagger}\hat{n}^{t}\bar{p}_{a}\hat{n}+\mathrm{h.c.}\right],\end{aligned}\ ] ] where @xmath21 ( @xmath22 ) is the spin - triplet iso - singlet ( spin - singlet iso - triplet ) auxiliary field , and @xmath23 ( @xmath24 ) projects out the spin - triplet iso - singlet ( spin - singlet iso - triplet ) combination of nucleons . the subscript `` 2 '' indicates that eq . ( [ eq : lags ] ) includes only two - body terms . at lo the parameters are fit to reproduce the bound state and virtual bound state poles in the @xmath25 and @xmath18 channels , respectively . the majority of nlo , @xmath15 , and @xmath0 parameters in the @xmath19-parametrization are then fit to ensure the poles remain unchanged and have the correct residues . the @xmath0 parameters @xmath26 and @xmath27 are fit to reproduce the effective range expansion ( ere ) shape corrections about the poles of the @xmath25 and @xmath18 channels , respectively . carrying out this procedure yields @xcite @xmath28 where @xmath29 is a scale introduced via dimensional regularization with the power divergence subtraction scheme @xcite . all physical observables must be @xmath29-independent . the value @xmath30 mev ( @xmath31 mev ) is the deuteron bound state momentum ( @xmath18 virtual bound state momentum ) , and @xmath32 ( @xmath33 ) the residue of the @xmath25 ( @xmath18 ) bound state ( virtual bound state ) pole . for the shape parameter correction about the @xmath25 ( @xmath18 ) pole we use @xmath34 ( @xmath35 ) . the @xmath15 parameter @xmath36 mev , and is fit to the splitting between the virtual bound state momentum in the @xmath12 and @xmath17 spin - singlet channels . there is also a separate parameter for the splitting between the virtual bound state momentum in the @xmath12 and proton - proton ( @xmath37 ) spin - singlet channels in the absence of coulomb , but this is not relevant for the @xmath7 system and is not considered here . at @xmath15 there is a contribution from two - body @xmath16-mixing given by the lagrangian @xmath38+\mathrm{h.c.}.\ ] ] the parameter @xmath39 is fit to the asymptotic @xmath40 mixing ratio of the deuteron wavefunction , yielding @xcite @xmath41 where @xmath42 is the asymptotic @xmath40 mixing ratio of the deuteron wavefunction @xcite . two - body @xmath2-wave contact interactions first occur at @xmath0 . the @xmath11 terms are given by the @xmath11 lagrangian @xcite , @xmath43+c_{2}^{({}^3\!p_{2})}\left[2\delta_{xw}\delta_{yz}+2\delta_{xz}\delta_{yw}-\frac{4}{3}\delta_{xy}\delta_{wz}\right]\right)\\\nonumber & \hspace{3cm}\times\frac{1}{4}(\hat{n}^{t}\mathcal{o}_{xya}^{(1,p)}\hat{n})^{\dagger}(\hat{n}^{t}\mathcal{o}_{wza}^{(1,p)}\hat{n}),\end{aligned}\ ] ] where @xmath44 and the projector is defined as @xmath45 . note that the projector @xmath46 differs from the projector in ref . @xcite because we consider @xmath14 scattering , not just @xmath12 scattering . at @xmath0 the two - body @xmath47 contact interaction also appears : @xmath48 but it does not contribute to the polarization observables in our calculation at this order . the operator @xmath49 is defined by @xmath50 where the projector is @xmath51 . fitting the coefficients to the @xmath12 nijmegen phase shifts @xcite yields the values @xmath52 these @xmath53 values are in good agreement with those found in ref . @xcite . we take them as the central values of an experimental fit , but there is a substantial theoretical @xmath1error associated with that fit ; these coefficients are @xmath0 for @xmath7 scattering but are lo in two - body @xmath2-wave scattering . at lo the power counting mandates that an infinite number of diagrams be summed , yielding the lo dibaryon propagators @xcite @xmath54 the lo deuteron wavefunction renormalization is the residue about the @xmath25 bound state pole , which gives @xmath55 the form of higher order dibaryon propagators and wavefunction renormalization constants can be found in refs . @xcite . for this work the form of these higher order corrections will not be explicitly needed . rather , the higher order corrections will naturally be included in the integral equations , and diagrams with corrections attached to external dibaryon propagators will give higher order deuteron wavefunction renormalization contributions in the on - shell limit . the @xmath56 correction to the @xmath7 scattering amplitude using the methods introduced in ref . @xcite is given by the integral equation represented in fig . [ fig : pertcorrectiondiagrams ] . because this integral equation is not yet projected in spin or partial waves , it includes both doublet and quartet channel contributions . the single line represents a nucleon , the double line a spin - triplet dibaryon , and the double dashed line a spin - singlet dibaryon . a thick solid line denotes a sum over both spin - triplet and spin - singlet dibaryons . the oval with a 0 " inside is the lo @xmath7 scattering amplitude , the oval with the @xmath57 inside is the @xmath57th order correction to the @xmath7 scattering amplitude , the circle with the @xmath57 inside is a @xmath56 correction to the dibaryon propagator , and the rectangle with the @xmath57 inside is a @xmath56 three - body " correction . single lines represent nucleons , double lines spin - triplet dibaryons , and double dashed lines spin - singlet dibaryons . thick solid lines denote a sum over both spin - triplet and spin - singlet dibaryons . the lo @xmath7 scattering amplitude is the oval with a 0 " inside , and the oval with the @xmath57 inside is the @xmath56 correction to the @xmath7 scattering amplitude . the circle with the @xmath57 inside is the @xmath56 correction to the dibaryon propagators ( see fig . [ fig : twobodycontributions ] ) , and the rectangle with the @xmath57 inside is the @xmath57th order three - body " correction ( see fig . [ fig : n3londscatteringdiagrams]).,title="fig:",width=377 ] single lines represent nucleons , double lines spin - triplet dibaryons , and double dashed lines spin - singlet dibaryons . thick solid lines denote a sum over both spin - triplet and spin - singlet dibaryons . the lo @xmath7 scattering amplitude is the oval with a 0 " inside , and the oval with the @xmath57 inside is the @xmath56 correction to the @xmath7 scattering amplitude . the circle with the @xmath57 inside is the @xmath56 correction to the dibaryon propagators ( see fig . [ fig : twobodycontributions ] ) , and the rectangle with the @xmath57 inside is the @xmath57th order three - body " correction ( see fig . [ fig : n3londscatteringdiagrams]).,title="fig:",width=377 ] perturbative corrections to the dibaryon propagators in the @xmath19-parametrization @xcite are given in fig . [ fig : twobodycontributions ] . higher order corrections to dibaryon propagators ( used in the diagrams of fig . [ fig : pertcorrectiondiagrams ] ) . the nlo ( n=1 ) corrections are range corrections from @xmath58 and @xmath59 . at @xmath15 ( n=2 ) the dibaryons receive further range corrections @xmath60 and @xmath61 in the @xmath19- + parametrization , as well as the @xmath62 correction from splitting between the @xmath17 and @xmath12 spin - singlet scattering lengths . the @xmath0 ( n=3 ) corrections are from higher order range corrections @xmath63 and @xmath64 in the @xmath19-parametrization , and shape parameter corrections @xmath26 and @xmath27 . , width=415 ] the nlo contributions @xmath58 and @xmath59 are from range corrections , and the @xmath15 terms @xmath60 and @xmath61 are from higher order corrections to @xmath58 and @xmath59 , respectively . the @xmath15 correction , @xmath65 , arises from the splitting between the @xmath66 scattering length for @xmath17 and @xmath12 scattering . correction to the dibaryon propagator from @xmath60 and @xmath61 in this formalism . however , there is still a correction from the splitting term @xmath65 in the ere . ] at @xmath67 there are corrections @xmath63 and @xmath64 to the effective range corrections and also shape parameter corrections @xmath26 and @xmath27 . the three - body " contributions to the integral equation are given by the diagrams in fig . [ fig : n3londscatteringdiagrams ] . three - body " contributions to integral equations ( used in the diagrams of fig . [ fig : pertcorrectiondiagrams ] ) . the lo terms are nucleon exchange plus in the doublet @xmath6-wave channel the lo three - body force ( dark square ) . the nlo term is a nlo correction to the lo three - body force . at @xmath15 there are contributions from the two - body @xmath16-mixing term ( coupling indicated by pale square ) , the @xmath15 correction to the lo three - body force , @xmath68 , and a new energy dependent three - body force , @xmath69 . the @xmath0 contributions are from the two - body @xmath2-wave contact interactions ( green circle ) , the @xmath0 correction to the lo three - body force , and the @xmath0 correction to the @xmath15 energy dependent three - body force.,width=396 ] lo diagrams are given by nucleon exchange and the lo three - body force in the doublet @xmath6-wave channel . the nlo contribution comes from a nlo correction to the lo three - body force in the doublet @xmath6-wave channel . at @xmath15 there is a contribution from the two - body @xmath16-mixing term , a @xmath15 correction to the lo three - body force , and a new energy dependent three - body force in the the doublet @xmath6-wave channel . finally , at @xmath0 there are contributions from the @xmath53 and @xmath70 two - body @xmath2-wave contact interactions as well as @xmath0 corrections to the lo three - body force and the energy dependent @xmath15 three - body force . up to @xmath0 all three - body forces occur in the doublet @xmath6-wave channel . projecting the @xmath57th order amplitude for @xmath7 scattering in fig . [ fig : pertcorrectiondiagrams ] into a partial wave basis yields the set of integral equations in cluster configuration ( c.c . ) @xcite space @xmath71 where @xmath72 is a matrix of lo dibaryon propagators in c.c . space , and the first subscript in all terms appearing in the integral equations refers to the order of a term ( @xmath73 is lo , @xmath74 is nlo , etc . ) . @xmath75 and @xmath76 are vectors in c.c . space defined by @xmath77 t^{j;nt\to ns}_{n;l's',ls}(k , p , e ) \end{array}\right ) , \ \ \mathbf{v}_{p}=\left(\begin{array}{c } 1 \\ 0 \end{array}\right),\ ] ] where @xmath78 is the amplitude for @xmath7 scattering , @xmath79 the amplitude for a neutron and deuteron going to a nucleon and spin - singlet dibaryon , and @xmath76 projects out diagrams in c.c . space corresponding to a spin - triplet dibaryon in the initial state . the value @xmath80 ( @xmath6 ) refers to the initial orbital ( total spin ) angular momentum , @xmath81 ( @xmath82 ) to the final orbital ( total spin ) angular momentum , and @xmath83 to the total angular momentum ( orbital plus total spin angular momentum ) . all integral equations are calculated half off - shell in the center of mass ( c.m . ) frame , with @xmath84 being the outgoing off - shell momentum and @xmath85 the incoming on - shell momentum such that the total energy of the @xmath7 system is given by @xmath86 . for this calculation all partial waves up to @xmath87 are included . all three - body + integrals are regulated using a sharp cutoff , which in the @xmath88 " notation is defined by @xmath89 the `` @xmath90 '' notation defines the schur product ( element wise matrix multiplication ) of c.c . space vectors . for this set of integral equations the lo kernel in c.c . space is given by @xmath91 -\frac{4\pi}{qp}q_{l}\left(\frac{q^{2}+p^{2}-m_{n}e - i\epsilon}{qp}\right)\left ( \begin{array}{cc } 1 & 0\\ 0 & 0 \end{array}\right ) & , \ \ s=\nicefrac{3}{2}\\ \end{array}\right.,\end{aligned}\ ] ] where at this order there is no mixing between different partial waves or splitting of different @xmath83-values . the lo three - body force , @xmath92 , is fit to the doublet @xmath6-wave @xmath7 scattering length , @xmath93 fm . for details of how the fit is performed see ref . the functions @xmath94 are legendre functions of the second kind and are defined as used here differs from the conventional definition by a phase factor of @xmath95 . ] @xmath96 with @xmath97 being the standard legendre polynomials . note that @xmath98 does not exist . at nlo the only contribution to the kernel comes from the nlo correction to the three - body force , yielding @xmath99 where @xmath100 is the nlo correction to the lo three - body force , which is again fit to the doublet @xmath6-wave @xmath7 scattering length @xcite . the dibaryons also receive a correction at nlo , which is given by the c.c . space vector @xmath101 \frac{(z_{s}-1)}{2\gamma_{s}}\left(\gamma_{s}+\sqrt{\frac{3}{4}p^{2}-m_{n}e - i\epsilon}\right ) \end{array}\right).\ ] ] the @xmath15 kernel receives contributions from the two - body @xmath16-mixing term , the @xmath15 correction to the lo three - body force , @xmath68 , and a new energy dependent three - body force , @xmath69 . in c.c . space these contributions give @xmath102_{zx}=\frac{y \ , y_{sd } \ , m_{n}}{2}\left(z_{sd}^{(1)}(j , l',s',l , s , x , z)\frac{1}{kp}\left[4p^{2}q_{l}(a)+k^{2}q_{l'}(a)\right]\right.\\\nonumber & + z_{sd}^{(1)}(j , l , s , l',s',z , x)\frac{1}{kp}\left[p^{2}q_{l}(a)+4k^{2}q_{l'}(a)\right]\\\nonumber & \left.+\sum_{l''}\left[z_{sd}^{(2)}(j , l',s',l , s , x , z , l'')+z_{sd}^{(2)}(j , l , s , l',s',z , x , l'')\right]q_{l''}(a)\right)\\\nonumber & -\pi \left(h_{{\mathrm{n}^{2}\mathrm{lo}}}+\frac{4}{3}(m_{n}e+\gamma_{t}^{2})h_{2}^{({\mathrm{n}^{2}\mathrm{lo}})}\right)\delta_{l0}\delta_{ll'}\delta_{ss'}\delta_{s\nicefrac{1}{2}}(-1)^{x+z},\end{aligned}\ ] ] where the subscripts @xmath103 " and @xmath104 " refer to the matrix element in c.c . space and @xmath105 . the value @xmath106 ( @xmath107 ) corresponds to an initial ( final ) spin - triplet dibaryon state and @xmath108 ( @xmath109 ) to an initial ( final ) spin - singlet dibaryon state . functions @xmath110 and @xmath111 are defined with 3@xmath112-symbols , yielding @xmath113 1 & s ' & \frac{1}{2 } \end{array}\right\}}\\\nonumber & \hspace{5cm}\times{\left\{\begin{array}{ccc } 2 & 1 & x\\[1 mm ] \frac{1}{2 } & s & s ' \end{array}\right\}}{\left\{\begin{array}{ccc } s ' & 2 & s\\[1 mm ] l & j & l ' \end{array}\right\}}{c_{l,2,l'}^{0,0,0}},\end{aligned}\ ] ] and @xmath114 z & s ' & \frac{1}{2 } \end{array}\right\}}\\[2mm]\nonumber & \times\left({\left\{\begin{array}{ccc } \frac{1}{2 } & x & s\\[1 mm ] 1 & l '' & l \\[1 mm ] s ' & l ' & j \end{array}\right\}}+{\left\{\begin{array}{ccc } \frac{1}{2 } & 1 & s'\\[1 mm ] l ' & j & l '' \end{array}\right\}}{\left\{\begin{array}{ccc } l & x & l''\\[1 mm ] \frac{1}{2 } & j & s \end{array}\right\}}+\frac{1}{3}(-1)^{1+l''+l}\frac{1}{\widehat{s}\widehat{l}}\delta_{ll'}\delta_{ss'}\right){c_{l,1,l''}^{0,0,0}}{c_{l'',1,l'}^{0,0,0}},\end{aligned}\ ] ] where the hat is defined as @xmath115 . at @xmath15 there is also a correction to the dibaryon propagators from @xmath60 , @xmath61 , and @xmath65 . in c.c . space this is @xmath116 \frac{(z_{s}-1)^{2}}{2\gamma_{s}}\left(\gamma_{s}+\sqrt{\frac{3}{4}p^{2}-m_{n}e - i\epsilon}\right ) & -\,\ , \frac{2}{3}\delta_{s}^{({\mathrm{n}^{2}\mathrm{lo}})}d_{s } \ ! \left(e-\frac{p^{2}}{2m_{n}},p\right ) \end{array}\right ) .\ ] ] the factor of 2/3 in front of @xmath65 comes from the isospin projection . @xmath65 is only associated with the @xmath17 spin - singlet dibaryon propagator , which contributes 2/3 to the total isospin invariant nucleon spin - singlet dibaryon amplitude ( @xmath79 ) . the @xmath0 kernel contains a correction to the lo three - body force , @xmath117 , and a correction to the @xmath15 energy dependent three - body force , @xmath118 , which gives @xmath119 griesshammer @xcite argues that at @xmath0 there is a new divergence that mixes the wigner - symmetric and wigner - antisymmetric channels of the doublet @xmath6-wave , but that the need for a new counter - term is suppressed due to the pauli principle and should occur two orders higher . however , birse @xcite suggests that the pauli principle is automatically included in the asymptotic analysis and therefore the need for a new counter - term will be at @xmath0 . we find that fitting @xmath117 to the doublet @xmath6-wave @xmath7 scattering length and @xmath120 to the triton binding energy yields a @xmath0 correction to the doublet @xmath6-wave @xmath7 scattering amplitude that is not properly renormalized . this supports the claim made by birse that a new three - body force that mixes wigner - antisymmetric and wigner - symmetric channels in the doublet @xmath6-wave will be necessary at @xmath0 . this will be addressed in future work . an important additional contribution at @xmath0 comes from the two - body @xmath2-wave contact interactions . these will be dealt with using a slightly different method . there are also corrections to the dibaryon propagators at this order coming from @xmath63 and @xmath64 , and the shape parameter corrections @xmath26 and @xmath27 , which yield @xmath121\\[3 mm ] \left(\gamma_{s}+\sqrt{\frac{3}{4}p^{2}-m_{n}e - i\epsilon}\right)\left[\frac{(z_{s}-1)^{3}}{2\gamma_{s}}+\rho_{s1}\left(\frac{3}{4}p^{2}-m_{n}e-\gamma_{s}^{2}\right)\right ] \end{array}\right).\ ] ] the numerical solution of the integral equations is carried out by means of the hetherington - schick method @xcite , which solves the equations along a contour in the complex plane . using the solution along the contour in the integral equations , the scattering amplitude along the real axis can be solved . by rotating into the complex plane the fixed singularity of the deuteron pole is avoided as well as the branch cut singularities that occur above the deuteron breakup energy . the methods in refs . @xcite allow the hetherington - schick method to be used to calculate diagrams with the full off - shell lo scattering amplitude without calculating the full off - shell scattering amplitude . in order to calculate to large cutoffs and obtain sufficient numerical accuracy most mesh points are clustered for momenta with magnitude less than @xmath122 . for momenta with magnitude greater than @xmath122 far fewer mesh points are used since the amplitude is decaying as a power law @xcite . in this way , calculations to large cutoffs can be obtained using a reasonable number of mesh points , with a numerical accuracy of less than 1% . it is important to show convergence to large cutoffs because observables in this eft should be independent of the cutoff . any deviations to the result as the cutoff becomes large indicates missing physics ( such as a neglected counterterm ) . .5 cm the contribution to the @xmath0 kernel from the two - body @xmath2-wave contact interaction is shown in the last box in fig . [ fig : n3londscatteringdiagrams ] . this diagram is one - loop and can be solved analytically and projected out in an angular momentum basis . however , the resulting forms for all partial waves are cumbersome in numerical calculations . in order to circumvent this it is convenient to introduce a @xmath2-wave auxiliary field via the lagrangian @xmath123\\\nonumber & \hspace{2cm}+\frac{1}{2}y^{{{\ensuremath{{}^{1}\mathrm{p}_{1}}\xspace}}}\left[\left({\hat{\mathcal{p}}}_{i}^{{{\ensuremath{{}^{1}\mathrm{p}_{1}}\xspace}}}\right)^{\dagger}\hat{n}^{t}i\mathcal{o}_{i}^{(0,p)}\hat{n}+\mathrm{h.c.}\right].\end{aligned}\ ] ] this approach is equivalent to using the two - body @xmath2-wave contact interactions in eqs . ( [ eq : pwavelagrupak ] ) and ( [ eq : pwave1p1 ] ) . the @xmath0 contribution from the two - body @xmath2-wave contact terms is given by the coupled integral equations in fig . [ fig : twobodypwave ] . the p " amplitude is defined in the boxed region of fig . [ fig : twobodypwave ] , where the double line with a zig - zag represents a @xmath2-wave dibaryon propagator . the @xmath2-wave dibaryon propagator is simply given by a constant , since the scattering volumes in the two - body @xmath2-waves are of natural size and are therefore perturbative and do not require resumming . the constant term can be factored out of all numerical expressions and then reintroduced at the end to obtain the final expression . unboxed diagrams are the integral equations for the @xmath0 contribution to the @xmath7 scattering amplitude from the two - body @xmath2-wave contact interactions . the double lines with a zig - zag in the middle are the @xmath2-wave dibaryon propagator , given by @xmath124 . the boxed diagrams represent the equation for the `` p '' amplitude used in the unboxed integral equations above . the notation 3@xmath125 " in the oval indicates that this is a @xmath0 correction but one that only involves the two - body @xmath2-wave contributions.,width=377 ] the p " amplitude is given by @xmath126_{1}+\mathbf{k}^{j({}^{2r+1}\!p_{z})}_{l's',ls}(q , p , e)\otimes \mathbf{t}_{0;ls , ls}^{j}(k , q , e),\ ] ] where @xmath127=0 or 1 , the quantity @xmath128 ( @xmath107 ) for the @xmath11 ( @xmath47 ) two - body @xmath2-wave contact interaction channels ( channel ) , and the kernel function @xmath129 is a vector in c.c . space . for the first term on the right hand side only one element of the kernel in c.c . space is taken since only states with an initial spin - triplet dibaryon are needed . the kernel function for eq . ( [ eq : pamp ] ) is the tree level diagram in the box in fig . [ fig : twobodypwave ] . projected onto a partial wave basis this yields @xmath130_{x}=-\frac{m_{n } \ , y \ , y^{{\ensuremath{{}^{3}\mathrm{p}_{z}}\xspace}}}{4kp}\mathcal{z}^{{\ensuremath{{}^{3}\mathrm{p}_{z}}\xspace}}\left(j , l',s',l , s , x , z\right)\left(2k \ , q_{l'}(a)+p \ , q_{l}(a)\right)\ ] ] for the @xmath11 coefficients , and @xmath131_{x}=-\frac{m_{n } \ , y \ , y^{{{\ensuremath{{}^{1}\mathrm{p}_{1}}\xspace}}}}{4kp}\mathcal{z}^{{{\ensuremath{{}^{1}\mathrm{p}_{1}}\xspace}}}\left(j , l',s',l , s , x\right)\left(2k \ , q_{l'}(a)+p \ , q_{l}(a)\right)\ ] ] for the @xmath47 coefficient . functions @xmath132 and @xmath133 are defined by @xmath134 1 & s & \frac{1}{2 } \end{array}\right\}}\\\nonumber & \hspace{1cm}\times{\left\{\begin{array}{ccc } \frac{1}{2 } & 1 & s\\[1 mm ] 1 & s ' & z \end{array}\right\}}{\left\{\begin{array}{ccc } s & 1 & s'\\[1 mm ] l ' & j & l \end{array}\right\}}{\left\{\begin{array}{ccc } 1-x & \frac{1}{2 } & \frac{1}{2}\\[1 mm ] 1 & \frac{1}{2 } & \frac{1}{2 } \end{array}\right\}}{c_{l,1,l'}^{0,0,0}}\end{aligned}\ ] ] and @xmath135 l ' & j & l \end{array}\right\}}\delta_{s,\frac{1}{2}}{c_{l,1,l'}^{0,0,0}}.\end{aligned}\ ] ] as before , @xmath106 corresponds to an initial state spin - triplet dibaryon propagator and @xmath108 corresponds to an initial state spin - singlet dibaryon propagator . the advantage of these kernel functions in the @xmath2-wave auxiliary field method is that they only contain legendre functions of the second kind , which are already calculated to solve the lo @xmath7 scattering amplitude . therefore , no additional work has to be done to calculate the values of these functions along the mesh points of our integral equations . in the more conventional approach new functions involving @xmath136 and @xmath137 have to be calculated for each partial wave , and the higher the partial wave the more complicated the functional forms become . the @xmath2-wave auxiliary field method is a far more transparent and numerically efficient means by which to calculate corrections from two - body @xmath2-wave contact terms to @xmath7 scattering . using the p " amplitude the @xmath0 correction to the @xmath7 scattering amplitude from the two - body @xmath2-wave contact interactions in fig . [ fig : twobodypwave ] is given by the coupled integral equations @xmath138^{t}\otimes t_{l''s'',ls}^{j({}^{2r+1}\!p_{z})}(k , q , e)\\\nonumber & \hspace{5cm}+\mathbf{k}^{j}_{0;l's',l's'}(q , p , e)\otimes \mathbf{t}_{3_{\mathrm{p}};l's',ls}^{j}(k , q , e).\end{aligned}\ ] ] the term @xmath139 is the @xmath2-wave dibaryon propagator , and can be removed from the integral equations by an appropriate renormalization . in the inhomogeneous term we have used time reversal symmetry to write the kernel for a @xmath2-wave dibaryon going to an @xmath6-wave dibaryon in terms of the kernel for an @xmath6-wave dibaryon going to a @xmath2-wave dibaryon . the factor of @xmath140 is due to time reversal symmetry and comes from the fact that under time reversal @xmath141 finally , the coefficients @xmath142 and @xmath143 must be fit to @xmath12 scattering data . this can be done by performing gaussian integration on the @xmath2-wave auxiliary fields in eq . ( [ eq : pwavelag ] ) and matching to the coefficients in eqs . ( [ eq : pwavelagrupak ] ) and ( [ eq : pwave1p1 ] ) , or by performing a matching calculation using the lagrangians of eqs . ( [ eq : pwavelag ] ) , ( [ eq : pwavelagrupak ] ) , and ( [ eq : pwave1p1 ] ) . in order to match these coefficients the identities @xmath144 @xmath145 and @xmath146 are used , where the indices on the left are spherical ( @xmath147 ) and those on the right cartesian ( @xmath148 ) . matching coefficients yields @xmath149 then using eq . ( [ eq : cpjnpfit ] ) the ratio of coefficients @xmath142 and @xmath143 can be fit to @xmath12 scattering data . a factor of @xmath150 can be removed from the @xmath0 correction to the @xmath7 scattering amplitude by appropriate renormalization of the amplitude . the calculated amplitude is then renormalized by this factor so that we can consider different values of @xmath150 without needing to recalculate the scattering amplitude . using the @xmath2-wave auxiliary field method is equivalent to starting with the one loop diagram for the two - body @xmath2-wave contact interaction diagram in fig . [ fig : n3londscatteringdiagrams ] and projecting it in partial waves before carrying out the loop integration . the resulting angular loop integration is trivial , leading to an integral over the magnitude of the loop momentum of an integrand of products of legendre functions of the second kind . we find that using the @xmath2-wave auxiliary field allows us to treat this @xmath0 part of the correction in direct parallel with earlier corrections ; note that the boxed diagrams in fig . [ fig : twobodypwave ] mimic those found in the @xmath6-wave case . the @xmath2-wave auxiliary field method effectively trades performing a one loop diagram analytically for a tree level diagram that is then used to numerically reproduce the contributions from the one loop diagram . in the on - shell limit the scattering amplitudes become the transition matrix @xmath151 defined in the partial wave basis by @xmath152 in order to calculate polarization observables the transition matrix in the spin - basis is needed . this is related to the transition matrix in the partial wave basis via @xmath153 where @xmath154 ( @xmath155 ) is the initial ( final ) deuteron spin component in the @xmath104-direction and @xmath156 ( @xmath157 ) the initial ( final ) neutron spin component in the @xmath104-direction . the unpolarized cross section is given by summing over all final spins and averaging over all initial spins of the square of the transition matrix , yielding @xmath158 where the bar over @xmath159 denotes that it is unpolarized . polarizing the initial neutron leads to three different polarization observables denoted @xmath10 , @xmath160 , and @xmath161 . these correspond to polarizing the neutron along each of the respective axes . in the madison conventions @xcite the @xmath104-direction is defined by the momentum of the incoming beam , @xmath162 , and the @xmath163-direction by @xmath164 , where @xmath165 ( @xmath166 ) is the incoming ( outgoing ) momentum of the neutron . the polarization observables @xmath160 and @xmath161 are parity - violating observables and @xmath161 has been considered elsewhere in @xmath1 @xcite . the cross section due to a transversely polarized neutron beam is given by @xmath167},\ ] ] where @xmath168 is the azimuthal angle , @xmath162 defines the @xmath104-direction , and the direction of polarization defines the @xmath103-axis . an analyzing power @xmath169 can be derived from the transition matrix in the spin - basis using density matrix techniques @xcite , which yield @xmath170}{\displaystyle\sum_{m_{1},m_{2}}\sum_{m_{1}',m_{2}'}\left|m_{m_{1}',m_{2}';m_{1},m_{2}}\right|^{2}}.\ ] ] the resulting form will contain a @xmath171 that can be factored out to give the expression for @xmath3 in eq . ( [ sigay ] ) . polarizing the initial deuteron gives four polarization observables @xmath172 , @xmath173 , @xmath174 , and @xmath175 . other polarization observables exist but are related by rotational symmetry or are parity - violating or violate time - reversal symmetry . the differential scattering cross section in terms of these polarization observables is given by @xcite @xmath176,\end{aligned}\ ] ] where @xmath177 , @xmath178 , @xmath179 , and @xmath180 are numbers giving the amount of respective polarization . using density matrix techniques the vector polarization @xmath181 is given by @xmath182}{\displaystyle\sum_{m_{1},m_{2}}\sum_{m_{1}',m_{2}'}\left|m_{m_{1}',m_{2}';m_{1},m_{2}}\right|^{2}},\ ] ] where @xmath171 can be factored out to give @xmath4 . the tensor polarizations using density matrix techniques are given by @xmath183 @xmath184}{\displaystyle\sum_{m_{1},m_{2}}\sum_{m_{1}',m_{2}'}\left|m_{m_{1}',m_{2}';m_{1},m_{2}}\right|^{2}},\ ] ] and @xmath185}{\displaystyle\sum_{m_{1},m_{2}}\sum_{m_{1}',m_{2}'}\left|m_{m_{1}',m_{2}';m_{1},m_{2}}\right|^{2}},\ ] ] where again the @xmath168 dependence can be factored out leaving only @xmath186 dependence . the polarization observables can also be derived using the techniques in ref . @xcite , which has the advantage of writing the angular dependence in terms of legendre polynomials , where their coefficients are given by the scattering amplitudes . this gives analytical insight into how the shape of polarization observables are related to the scattering amplitudes . the @xmath1differential scattering cross section up to @xmath15 at a neutron lab energy of @xmath187 mev is shown in fig . [ fig : sigma ] and compared with data from schwarz et al . the solid green line is the lo prediction , the dashed blue line the nlo prediction , and the red band the @xmath15 prediction with a 6% error estimate from the @xmath1power counting . relatively good agreement at @xmath15 is observed with respect to the experimental data , and the minimum at @xmath15 coincides with that of the available experimental data . theoretical errors are not shown on the lo and nlo results , but we see that the @xmath1treatment converges on the data . the @xmath1power counting predicts a naive error estimate of @xmath188 for the n@xmath189lo scattering amplitudes , where @xmath190 and @xmath191 . [ cols="^,^ " , ] the dashed lines are @xmath1results for @xmath10 for several sets of @xmath53 coefficients varied by 15% around their central values . top left : @xmath192 mev , experimental data from neidel et al . top right : @xmath193 mev , experimental data from neidel et al . bottom : @xmath187 mev , the solid line is a pmc calculation from kievsky et al . using av-18+ur @xcite , with experimental data from mcaninch et al . @xcite . in the following , @xmath194 " stands for 15 percent above central values given in eq . ( [ eq : cpjnpfit ] ) ; @xmath195 " is at central value ; and @xmath196 " is 15 percent below central value . the coefficient values ( @xmath197,@xmath198,@xmath199 ) used to produce the curves shown are ( from lowest @xmath1curve to highest @xmath1curve on the plots ) : big dots ( green)=@xmath200 ; small dots ( blue)=@xmath201 ; long dash ( red)=@xmath202 ; long - dash - dot ( purple ) = @xmath203 ; short - dash - dot ( orange ) = @xmath204 ; double - dot ( black ) = @xmath205.,width=340 ] while the position of the peak of @xmath10 depends primarily on the minimum of the cross section , its magnitude depends primarily on the two - body @xmath2-wave contact interaction @xmath53 terms . the contribution to @xmath10 from the @xmath16-mixing term is negligible at about three orders of magnitude smaller than the contribution from the two - body @xmath2-wave contact interactions . in fig . [ fig : aybands ] the different dashed curves correspond to different choices for the @xmath53 coefficients . these coefficients are fit to the nijmegen @xmath2-wave phase shifts for @xmath12 scattering , yielding the central values in eq . ( [ eq : cpjnpfit ] ) . but we expect a substantial @xmath1theoretical error associated with this fit because these terms are the leading @xmath1terms contributing to @xmath14 @xmath2-wave scattering . therefore we vary each of the three @xmath53 coefficients by 15 percent from their central values . representative curves are shown in fig . [ fig : aybands ] . the most obvious conclusion is that when appropriate theoretical errors are applied to the @xmath53 coefficients , a large range of @xmath5 s can be accommodated . this motivates us to go to higher order . cc the solid red line is the @xmath1prediction ( without theoretical error bars ) for deuteron polarization observables in @xmath7 scattering , the dashed green line pmc calculations using av18+ur for @xmath7 scattering @xcite , and the dotted blue line pmc calculations using av18+ur for @xmath8 scattering @xcite . all experimental data is for @xmath8 scattering from shimizu et al . @xcite at a laboratory deuteron energy of @xmath206 mev.,title="fig:",width=340 ] & the solid red line is the @xmath1prediction ( without theoretical error bars ) for deuteron polarization observables in @xmath7 scattering , the dashed green line pmc calculations using av18+ur for @xmath7 scattering @xcite , and the dotted blue line pmc calculations using av18+ur for @xmath8 scattering @xcite . all experimental data is for @xmath8 scattering from shimizu et al . @xcite at a laboratory deuteron energy of @xmath206 mev.,title="fig:",width=340 ] + the solid red line is the @xmath1prediction ( without theoretical error bars ) for deuteron polarization observables in @xmath7 scattering , the dashed green line pmc calculations using av18+ur for @xmath7 scattering @xcite , and the dotted blue line pmc calculations using av18+ur for @xmath8 scattering @xcite . all experimental data is for @xmath8 scattering from shimizu et al . @xcite at a laboratory deuteron energy of @xmath206 mev.,title="fig:",width=340 ] & the solid red line is the @xmath1prediction ( without theoretical error bars ) for deuteron polarization observables in @xmath7 scattering , the dashed green line pmc calculations using av18+ur for @xmath7 scattering @xcite , and the dotted blue line pmc calculations using av18+ur for @xmath8 scattering @xcite . all experimental data is for @xmath8 scattering from shimizu et al . @xcite at a laboratory deuteron energy of @xmath206 mev.,title="fig:",width=340 ] while the curves shown in fig . [ fig : aybands ] are just a representative sample , we note the following scaling : as @xmath197 increases from the central value in eq . ( [ eq : cpjnpfit ] ) , with the other coefficients held constant , @xmath5 decreases substantially at all energies . @xmath5 is most sensitive to variations of this coefficient about its central value when the other two coefficients are held constant within their allowed variation . next in sensitivity are changes in @xmath198 , and then changes in @xmath199 ( again with other coefficients held constant within their allowed variation ) . but for the @xmath207 values , @xmath5 increases as these coefficients increase . the contributions of the @xmath53 coefficients to @xmath10 are not independent since if the @xmath53 coefficients are all equal their total contribution to @xmath10 is zero . all of the @xmath10 observables are calculated at a cutoff of @xmath208 mev . it is necessary to use a large cutoff , because the three - body @xmath209 , @xmath210 , and @xmath211 channels go asymptotically like @xmath212 @xcite and as a function of @xmath213 converge slowly . channel the leading asymptotic behavior comes from the wigner - antisymmetric piece , which is equivalent to the asymptotic behavior of the @xmath209 channel @xcite . ] a new three - body counter - term for the @xmath209 , @xmath210 , and @xmath211 channels will be needed at order @xmath214 @xcite and is necessary for a @xmath215 calculation . in addition , the channel @xmath216 goes asymptotically like @xmath217 . despite the slow rate of convergence , varying this channel within its cutoff variation was found to have almost no effect on the observables studied here at the order to which we are working . however , given that a three - body force occurs in this channel at order @xmath218 @xcite , a new three body force must be included in a @xmath215 calculation . all other channels converge much faster and are well converged at cutoffs of a few thousand mev . the deuteron polarization observables are given in fig . [ fig : tnn ] , where the solid red line is the @xmath1prediction ( not including theoretical errors ) using the central values for @xmath53 coefficients given in eq . ( [ eq : cpjnpfit ] ) , the long - dashed green line a pmc for @xmath7 using av-18+ur @xcite , and the dotted blue line a pmc for @xmath8 using av-18+ur @xcite . all of the data shown is for @xmath8 scattering from shimizu et al . @xcite at a laboratory deuteron energy of @xmath206 mev . the results for @xmath7 scattering should roughly agree with those of @xmath8 scattering at higher energies and backward angles where coulomb effects are less important . rough qualitative agreement is observed for all polarization observables . the contribution from @xmath16-mixing is significant for all polarization observables except @xmath172 , where the @xmath16-mixing contribution is three orders of magnitude smaller than the two - body @xmath2-wave contact interaction contributions . in fact , @xmath172 changes in the same manner that @xmath5 changes when the two - body @xmath2-wave coefficients are varied . the contribution from the two - body @xmath2-wave contact interactions for @xmath173 is roughly the same size as @xmath16-mixing contributions , negligible for @xmath174 , and about two orders of magnitude smaller than @xmath16-mixing for @xmath175 . using the techniques in refs . @xcite , polarization observables in @xmath7 scattering have been calculated to @xmath0 in @xmath1 . the polarization observables in @xmath1receive non - zero contributions from the two - body @xmath16-mixing term and the two - body @xmath2-wave contact interactions , @xmath53 . contributions from the two - body @xmath2-wave contact interactions were calculated by the introduction of a @xmath2-wave auxiliary field . this approach leads to great analytical and numerical simplifications in the calculation of the two - body @xmath2-wave contact interaction contributions to @xmath7 scattering amplitudes at @xmath0 , and will be useful for higher order @xmath1calculations and halo eft calculations as well . we find that the @xmath10 and @xmath172 polarization observables are dominated by contributions from the two - body @xmath2-wave contact interactions while @xmath16-mixing contributions are three orders of magnitude smaller . the @xmath173 observable receives roughly equal contributions from @xmath16-mixing and two - body @xmath2-wave contact interactions , @xmath174 only has contributions from @xmath16-mixing , and @xmath175 primarily receives contributions from @xmath16-mixing with contributions from the two - body @xmath2-wave contact interactions being two orders of magnitude smaller . deuteron polarization observables for @xmath7 scattering were compared with available @xmath8 data and pmc . rough qualitative agreement is found at larger angles where effects from the coulomb interaction are less important . the nucleon vector analyzing power @xmath10 was calculated at @xmath192 , 1.9 , and @xmath219 mev and was found to vary widely given the theoretical error associated with the two - body @xmath2-wave coefficients . however , the position of the @xmath10 peak is predicted well in @xmath1 . significantly , we find that we can account for all the low energy @xmath5 data considered here so long as we allow the extracted coefficients @xmath53 to vary within their expected @xmath1theoretical errors . hence there is no disagreement at the moment between @xmath1and experiment within theoretical error . but this motivates us to pursue a higher order calculation . in calculating the @xmath7 scattering amplitudes the cutoff was taken to be @xmath208 mev . a large cutoff was necessary to achieve convergence in the three - body @xmath0 @xmath209 , @xmath210 , and @xmath211 channels as they behave asymptotically like @xmath212 @xcite , and therefore have slow convergence . the asymptotic form of the @xmath209 , @xmath210 , and @xmath211 channels will lead to the need for a three - body force at @xmath214 @xcite . some authors would advocate that order - by - order in the @xmath1expansion the cutoff variation with respect to @xmath213 should grow smaller ( for a discussion of this see ref . in such a power counting scheme the three - body force occurring at @xmath214 would be promoted to @xmath0 . although this is a desirable property for a power counting it is not rigorously motivated in the three - body sector . in future work we will calculate @xmath220 contributions , which include the @xmath214 three - body force terms in the @xmath209 , @xmath210 , and @xmath211 channels . unfortunately , these three - body forces must be fit to three - body data in the @xmath209 , @xmath210 , and @xmath211 channels , which are expected to be important for obtaining an accurate description of @xmath5 because @xmath2-waves are the first angular momenta that cause splitting among @xmath83-values . the three - body forces can be fit to @xmath10 at one energy and then predictions can be made for other energies . at @xmath220 there will also be a new two - body @xmath16-mixing term , a new energy dependent three - body force in the doublet @xmath6-wave channel , and a three - body @xmath16-mixing term that can be fit to the asymptotic @xmath40-mixing ratio of the triton wavefunction . in this @xmath1calculation the doublet @xmath6-wave channel was only calculated to @xmath15 . the doublet @xmath6-wave channel has only one @xmath83-value and therefore no splitting for different @xmath83-values occurs . as a result the doublet @xmath6-wave only contributes up to nlo in the numerator for all polarization observables . the denominator of all polarization observables is given by the unpolarized @xmath7 scattering cross section . in a strictly perturbative approach we would expand the denominator . since the first non - zero contribution to the numerator of polarization observables occurs at @xmath15 the doublet @xmath6-wave contribution from the expanded denominator would again only be needed to nlo for a @xmath0 calculation . however , we find that the peak of @xmath10 depends on the minimum of the cross section that is only reproduced well at @xmath15 . therefore we resum certain higher order contributions into the denominator and keep the cross section expanded perturbatively to @xmath15 in the denominator . future calculations will include the @xmath0 doublet @xmath6-wave channel and calculate the @xmath0 cross section . this should not significantly change the results since good agreement is already observed with experimental data at @xmath15 for the cross section . calculation of the doublet @xmath6-wave channel to @xmath0 is likely complicated by the requirement of an additional wigner - antisymmetric to wigner - symmetric three - body force . griesshammer @xcite claims that a wigner - antisymmetric to wigner - symmetric three - body force should occur at @xmath221 due to suppression from the pauli principle , while birse @xcite claims that it should occur at @xmath0 as the pauli principle is already included in a naive asymptotic analysis . having fit @xmath117 to the doublet @xmath6-wave scattering length and @xmath120 to the triton binding energy we find the doublet @xmath6-wave @xmath7 scattering amplitude is not properly renormalized , suggesting the need for a new three - body force as claimed by birse . future work will address this new three - body force in order to have a complete @xmath0 calculation . also at @xmath0 and higher orders the divergences that must be renormalized become worse , leading to potential numerical issues , especially at higher cutoffs due to numerical fine tuning . finally , calculating polarization observables in @xmath8 scattering is of interest due to the larger data set available for such interactions . such calculations are complicated due to coulomb interactions , but have been performed in @xmath1 in the quartet and doublet @xmath6-wave channels in which coulomb effects are treated perturbatively " @xcite . higher partial waves will be needed but are in principle straightforward to include . the main stumbling block to calculations in @xmath8 scattering are three - body forces in the doublet @xmath6-wave channel . it was shown in ref . @xcite that at nlo the same three - body forces could not be used to renormalize both @xmath7 and @xmath8 scattering . this pattern likely persists to @xmath15 where a new energy dependent three - body force occurs . in the @xmath7 system the three - body forces are fit to the doublet @xmath6-wave @xmath7 scattering length and the triton binding energy . for @xmath8 scattering fitting to the @xmath222 binding energy is straightforward , but fitting the doublet @xmath6-wave @xmath8 scattering length is complicated due to coulomb effects . therefore an appropriate renormalization condition will need to be found for @xmath8 scattering at @xmath15 in order to investigate the large @xmath8 data set for polarization observables . all of the diagrams used in these calculations need to be projected into the partial wave basis . one approach is to construct all necessary projectors and use them to project all diagrams onto respective partial waves @xcite . the advantage of this approach is that it makes projecting out diagrams very easy . however , the downside is that a projector must be constructed for every channel of interest and only @xmath6- and @xmath2-wave projectors have been published to date . instead of the projector method we adopt a racah algebra approach that , while computationally more intensive , gives all partial waves at once . the contribution from a generic diagram is given by @xmath223_{yx},\ ] ] where @xmath224 ( @xmath225 ) is the initial ( final ) dibaryon spin polarization , @xmath226 ( @xmath227 ) the initial ( final ) dibaryon isospin polarization , @xmath228 ( @xmath229 ) the initial ( final ) nucleon spin , and @xmath230 ( @xmath231 ) the initial ( final ) nucleon isospin . the subscripts @xmath163 and @xmath103 pick out a component of the c.c . space matrix , with @xmath106 ( @xmath108 ) corresponding to an initial spin - triplet ( spin - singlet ) dibaryon and @xmath232 ( @xmath233 ) corresponding to a final spin - triplet ( spin - singlet ) dibaryon . a generic contribution is projected onto a partial wave basis by @xmath234_{yx}=\frac{1}{4\pi}\sum_{\chi}{c_{l , s , j}^{m_{l},m_{s},m}}{c_{l',s',j}^{m_{l'},m_{s'},m}}{c_{x,\frac{1}{2},s}^{i,\alpha , m_{s}}}{c_{y,\frac{1}{2},s'}^{j,\beta , m_{s'}}}{c_{1-x,\frac{1}{2},\frac{1}{2}}^{a , a,-\frac{1}{2}}}{c_{1-y,\frac{1}{2},\frac{1}{2}}^{b , b,-\frac{1}{2}}}\\\nonumber & \hspace{2cm}\times\int\!\!d\omega_{q}\int\!\!d\omega_{\ell}\left[\left(\boldsymbol{\mathcal{k}}^{jb}_{ia}({\vec{\boldsymbol{q}}},{\vec{\boldsymbol{\boldsymbol{\ell}}}})\right)^{\beta b}_{\alpha a}\right]_{yx}y_{l}^{m_{l}}(\hat{\mathbf{q}})\left(y_{l'}^{m_{l'}}(\hat{\boldsymbol{\ell}})\right)^{*},\end{aligned}\ ] ] where the first two clebsch - gordan coefficients couple orbital and spin angular momentum , the next two couple dibaryon and nucleon spin , and the final two couple nucleon and dibaryon isospin . @xmath235 sums over all magnetic quantum numbers . in order to treat all elements of a c.c . space matrix simultaneously we define the operators @xmath236 using the wigner - eckart theorem these operators when projected out give @xmath237 and @xmath238 as an example , a contribution from the @xmath16-mixing term is given by @xmath223^{(sd)}_{yx}=\frac{1}{a+\hat{\mathbf{q}}\cdot\hat{\boldsymbol{\ell}}}\left[\sigma_{m}{s_{y}^{j}}^{\dagger}{t_{1-y}^{b}}^{\dagger}\right]^{\beta b}_{\alpha a}{\vec{\boldsymbol{q}}}_{i}{\vec{\boldsymbol{q}}}_{-m}(-1)^{m}\delta_{1x}.\ ] ] using eq . ( [ eq : projection ] ) and the identity @xmath239 the projection of this contribution in the partial wave basis becomes @xmath240 the sum over magnetic quantum numbers can then be simplified via racah algebra yielding @xmath241^{(sd)}_{yx}=2\sqrt{3\widehat{y}\widehat{(1-y)}\widehat{s}\widehat{s'}\widehat{l}\widehat{l''}}(-1)^{\frac{1}{2}+x+y+l''+l+s+s'-j}\\\nonumber & \hspace{3cm}\times{\left\{\begin{array}{ccc } y & \frac{1}{2 } & \frac{1}{2}\\[1 mm ] 1 & s ' & \frac{1}{2 } \end{array}\right\}}{\left\{\begin{array}{ccc } l '' & 1 & x\\[1 mm ] \frac{1}{2 } & s & s ' \end{array}\right\}}{\left\{\begin{array}{ccc } s ' & l '' & s\\[1 mm ] l & j & l ' \end{array}\right\}}{c_{1,1,l''}^{0,0,0}}{c_{l , l'',l'}^{0,0,0}}\ell^{2}q_{l}(a)\delta_{1x}.\end{aligned}\ ] ] we would like to thank d.r . phillips , m.r . schindler , and w. tornow for discussions important for the completion of this work . this material is based upon work supported by the u.s . department of energy , office of science , office of nuclear physics , under award number de - fg02 - 05er41368 and award number de - fg02 - 93er40756 ( jv ) .

we develop a model for the distribution of the ism and star formation in galaxies based on recent studies that indicate that galactic disks stabilise to a constant stability parameter , which we combine with prescriptions of how the phases of the ism are determined and for the star formation law ( sfl ) . the model predicts the gas surface mass density and star formation intensity of a galaxy given its rotation curve , stellar surface mass density and the gas velocity dispersion . this model is tested on radial profiles of neutral and molecular ism surface mass density and star formation intensity of 12 galaxies selected from the things sample . our tests focus on intermediate radii ( 0.3 to 1 times the optical radius ) because there are insufficient data to test the outer disks and the fits are less accurate in detail in the centre . nevertheless , the model produces reasonable agreement with ism mass and star formation rate integrated over the central region in all but one case . to optimise the model , we evaluate four recipes for the stability parameter , three recipes for apportioning the ism into molecular and neutral components , and eight versions of the sfl . we find no clear - cut best prescription for the two - fluid ( gas and stars ) stability parameter and therefore for simplicity , we use the approximation ( ) . we found that an empirical scaling between the molecular to neutral ism ratio ( ) and the stellar surface mass density proposed by works marginally better than the other two prescriptions for this ratio in predicting the ism profiles , and noticeably better in predicting star formation intensity from the ism profiles produced by our model with the sfls we tested . thus in the context of our modeled ism profiles , the linear molecular sfl and the two - component sfl work better than the other prescriptions we tested . we incorporate these relations into our ` constant disk ' ( cq - disk ) model . * keywords : * galaxies : ism - kinematics and dynamics - star formation

1307.2896

the processes in the interstellar medium ( ism , or gas ) that determine its distribution , structure , and the formation of stars are multiple , very complex , and operate on atomic to galactic size scales ( see e.g. * ? ? ? this makes it difficult to develop a comprehensive model for the distribution of gas and star formation in galaxies from first principles . fortunately , normal galaxies are usually in a marginally stable equilibrium state . here we use this fact as the basis of a model for the distribution of gas in galaxies , and with some additional assumptions we extend this into a model of the star formation distribution . our aim is to construct an easy to implement model for the distribution of gas and star formation in galaxies that can be used to compare to observations and to easily create realistic simulated galaxies . our approach also allows difficult to observe properties of galaxies ( e.g. the molecular ism distribution ) to be inferred from those that are relatively easy to determine or infer ( the stellar mass profile and rotation curve ) . star formation has long been believed to be related to disk ( in)stability . major studies ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) have been carried out on gravitational disk stability since the 1960s and the theory of this subject is well developed . however , as early as 1972 , @xcite claimed that the whole galactic disk should be marginally stable due to negative feedback mechanism . he used this assumption to predict the gas surface density profile ; his model roughly matches the observations in the outer regions of his sample galaxies but overestimates the densities in the inner half of the galaxies . since then , observational evidence for quirk s statement that disk galaxies are usually in a marginally stable state has mounted and been noticed in various studies ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? the gravitational stability parameter @xmath3 ( see section [ threshold ] for details ) for a single component thin disk was first derived by @xcite . @xcite used a sample of tens of galaxies to show that there is very little star formation where @xmath3 is above some critical threshold . @xcite confirmed the result using more recent data . however a single fluid @xmath3 does not accurately indicate the stability of a real disk which contains both stars and gas . @xcite derived the two - fluid ( stars and gas ) stability parameter @xmath0 for a thin disk . @xcite improved the derivation by considering the stellar component as collisionless , yielding a rigorous but elaborate form of the stability parameter , @xmath4 . @xcite proposed an approximation for the two - fluid stability parameter , @xmath1 , which is widely used because of its simple form . more recently , @xcite reexamined the @xcite approximation and gave a simple but more accurate effective two - fluid stability parameters for both an infinitesimally thin ( @xmath5 ) and finite thickness ( @xmath6 ) galactic disks . inspired by the @xcite and recent studies on disk stability theory and star formation laws , we develop and test the ` constant @xmath3 disk ' ( cq - disk ) model , which we define as : a two - fluid ( gas+stars ) axisymmetric thin disk will evolve into a marginally stable state with a constant stability parameter @xmath0 through out the galactic disk . the basic idea for this model is to use the constancy of the two - fluid stability parameter to predict the distribution of gas and then use empirical molecular - to - neutral gas ratio ( @xmath2 ) relations and the best star formation law ( sfl ) to predict the distribution of neutral and molecular gas as well as star formation rate . here we compare the observed distribution of the ism , and star formation in a small sample of galaxies to our model in order to test our assumptions and to optimise the model by trialling different recipes for its key ingredients ( @xmath0 , @xmath2 , and the sfl ) . the outline of our paper is as follows : section [ bkintro ] briefly introduces background information , e.g. the gravitational stability parameter , molecular - to - neutral gas ratio model , and star formation laws ; section [ model ] presents the detailed model we construct ; section [ data ] gives a brief description of our sample galaxies ; section [ results ] shows the results ; in section [ discussion ] we discuss the assumption and implications of our model ; and in section [ application ] we discuss further tests and possible uses of our model . through out this paper , we are only concerned with the large scale star formation in the galactic disk , hence we suppose the disk is axisymmetric and focus on the radial profiles of various physical quantities . as our research progressed it became useful to divide the galactic disk in to three regions : central ( within @xmath7 ) , intermediate ( @xmath8 ) and outer ( @xmath9 ) , where @xmath10 is the optical radius , defined by where the @xmath11 band surface brightness reaches 25 mag arcsec@xmath12 . in general , our model works best in the intermediate radii , and much of the testing of the model is limited to these radii . the profiles we create often do not match the observations in detail in the central region , although , as we will show , the match to quantities integrated over the central region is reasonable . we can not test our star formation models in the outer disk because the star formation intensity was not measured beyond @xmath10 in our data sources . in @xcite we discuss the expected ism structure in the outer disk . there we showed that a gas dominated constant @xmath3 outer disk should have a surface mass density fall - off with the same profile as the dark matter for systems with a flat rotation curve , thus explaining the close relationship between dark matter and hi @xcite . the other reason that we make these divisions is that the rcs are generally more reliable in the intermediate disk : there are less or even no data points in the central regions , while often warps and asymmetries are present in the outer regions . @xcite showed that galactic disks can be unstable to axis - symmetric perturbations , and that the stability of a thin disk to such a perturbation can be represented quantitatively in a single parameter @xmath3 . following @xcite , @xcite , and @xcite , we express the stability parameter for a gaseous disk : @xmath13 where @xmath14 is the gas velocity dispersion . typically @xmath14 is assumed to be constant ( e.g. * ? ? ? * hereafter l08 ) , although observations show that it typically slowly varies with radius . @xmath15 is the gravitational constant , @xmath16 is the gas surface mass density and @xmath17 is the epicyclic frequency which can be calculated from the rotation curve using @xmath18 where @xmath19 is the circular velocity of the gas at a distance @xmath20 from the galactic centre . @xmath3 parameter can be an indicator for widespread star formation : a large stability parameter ( @xmath21 ) means that pressure and centrifugal forces are sufficient to support the disk and thus it is stable , while a small @xmath22 means the gravity exceeds the internal support and the disk will collapse , resulting in widespread star formation @xcite . for a single component stellar disk , we can use a similar parameter , @xmath23 , to indicate gravitational stability of the stars @xcite , but redefined as the equation shown above by later authors e.g. @xcite in order to use a more consistent expression with regards to @xmath24 . ] : @xmath25 where @xmath26 is the star surface mass density and @xmath27 is the radial component of the stellar velocity dispersion @xcite . we follow the estimation of @xmath27 given by l08 : assuming a typical fixed shape to the velocity dispersion ellipsoid , we have @xmath28 where @xmath29 is the z - direction component of the stellar velocity dispersion and using the standard relationship between this , disk scale height @xmath30 , and mass density for an isothermal disk ( e.g. * ? ? ? * ) , we have @xmath31 here @xmath30 is assumed to be a constant through out the galaxy disk ( e.g. as done by l08 ) . since , @xmath30 is difficult to measure unless the galaxy is viewed edge - on , we follow l08 and use the exponential stellar disk s scale length , @xmath32 , to estimate it . using the average flattening ratio @xmath33 , measured by @xcite , we then obtain @xmath34 galactic disks are composed of both gas and stars . thus , we need to consider both of them in a more sophisticated disk stability theory . there have been numerous studies on the two - fluid disk stability theory and therefore different forms of two - fluid stability parameters , @xmath35 . here we introduce four of them : @xcite explored the gravitational stability of a two - fluid thin disk . later - on , @xcite extended this work by treating the stellar part of the disk as collisionless and gave an explicit expression for the total stability parameter @xmath4 : @xmath36 where @xmath37 and @xmath38 , with @xmath39 the wave number of the instability being considered . we take @xmath39 to be the wavenumber of the most unstable mode in our calculations ( l08 ) . in this paper , we make a coarse searching through the parameter space of @xmath39 and find the value which gives the lowest gas surface mass density . we find that the galaxies studied here have @xmath40 , consistent with the result of l08 . although the @xcite derivation of the two - fluid stability parameter is quite rigorous , it is complicated and wavenumber dependent , making it hard to use . @xcite give a much simpler form , @xmath1 : @xmath41 it is widely used by astronomers ( e.g. * ? ? ? * ) although the analysis deriving it has been criticised @xcite . @xcite improved the @xcite approximation and proposed an effective @xmath3 parameter for a two - fluid infinitesimally thin disk , @xmath5 : @xmath42 where @xmath43 and @xmath44 is defined above in eq . [ eq_y ] . in reality , disks are not infinitesimally thin , and the thickness of the disk has a stabilising effect . therefore , @xcite also give an expression for a disk of finite thickness , @xmath6 : @xmath45 where @xmath46 is a factor , by which the stability parameter of each component increases , reflecting the effect of the thickness of the disk . the factor @xmath46 depends on the ratio of vertical to radial velocity dispersion : @xmath47 . so that we have @xmath48 for gas , and @xmath49 for stars . @xcite showed that @xmath24 values for 15 disk galaxies are almost a constant . similarly @xcite used more recent measurements to show that @xmath24 is constant over a large fraction of the outer disks of galaxies where hi dominates the mass . @xcite tested the @xcite @xmath1 on more disk galaxies and got a similar result . l08 used the @xcite @xmath4 in testing the star formation threshold for 23 nearby galaxies and get a remarkably flat @xmath4 thorough the optically bright part of galactic disks with @xmath50 . one plausible explanation for this phenomena is self - regulation by star formation @xcite : an unstable disk ( low @xmath3 ) will collapse and commence widespread star formation . the star formation will consume most of the gas in a short time @xcite , also the resulting winds from high mass stars and supernovae explosions will expel part of the gas out of the galactic disk @xcite and heat up the remaining gas @xcite . this will increase @xmath3 by lowering @xmath16 and increasing @xmath14 and thus make the disk more stable . on the other hand , a stable disk ( high @xmath3 ) will have suppressed star formation because the gas can not collapse . this means less heating of the ism through winds and supernovae , and hence a relative cooling and lowering of @xmath14 , thus decreasing @xmath3 . thus a negative feedback system is set up and it will finally come to a stable equilibrium state . here we suppose that galaxies equilibrate to the same @xmath3 at all radii through this local feedback cycle : regions of low @xmath3 heat up locally due to feedback , while regions of high @xmath3 cool due to relative lack of star formation . we test whether @xmath3 is constant within the optically bright portion of galaxies , and consider whether @xmath3 may be constant within a galaxy , but vary from galaxy to galaxy . this might happen if the balance in the local support between epicyclic frequency @xmath17 and velocity dispersion @xmath51 varies due to changes in the relative importance of angular momentum and feedback , perhaps due to initial mass function variations @xcite or metallicity dependent cooling . it is useful to have an accurate prescription for determining the molecular to neutral ratio ( @xmath52 ) because these phases of the ism are measured with separate observations and because it is generally believed that star formation is more related to the the molecular rather than the atomic gas . in practice the molecular data has been more expensive to obtain than data and requires use of tracers of h@xmath53 like co emission . the advent of new facilities like sma , carma and alma may change these economics . various simple prescriptions for determining @xmath2 have been raised in previous studies ( cf . l08 and references therein ) . l08 used high quality hi and co data to assess four of the most popular @xmath2 prescriptions and found two that worked well . the first is purely empirical and involves just the stellar surface mass density ( we call this the sr relation or @xmath54 ) @xmath55 the other uses hydrostatic pressure @xmath56 as a parameter ( the pr relation or @xmath57 ) and is @xmath58 where @xmath56 is given by @xcite as @xmath59 where @xmath29 can be calculated using eq . [ sigma_sz ] . although l08 found that @xmath56 is not as good of a predictor of @xmath2 as the stellar surface mass density , it has a more solid physical basis @xcite . another physically intuitive model is described by ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * hereafter kr relation or @xmath60 ) and tested by @xcite using numerical simulations . in this model the molecular gas can only exist in a region well shielded from the uv radiation field . the approximate solution is @xmath61 @xcite , where @xmath62 and where @xmath63 is a dimensionless number representing the scaled radiation field , @xmath64 where @xmath65 is the dust cross section per hydrogen atom and @xmath66 is the mean mass per nucleus . following @xcite , we take the approximation that @xmath67 and @xmath68 with @xmath69 being the metallicity normalised to the solar value . from eq . [ eq_molk ] - [ eq_zprime ] we expect some troublesome asymptotic behaviour in low density and low metallicity regions because a small @xmath70 ( @xmath71 ) leads to a negative @xmath60 . we therefore set negative @xmath60 values to zero in our code , which means there is no molecular gas in low density and low metallicity regions . we test all three relations to estimate @xmath2 in our calculation ( using the total gas surface mass density derived using our model , cf model description in section [ model ] ) and see which prescription best models the @xmath72 and @xmath73 profiles . the sfl is usually defined as the relationship between star formation rate and ism properties , especially density . the sfl allows one to calculate the star formation rate given gas density and some other galactic parameters or vice versa . this is usually expressed as the surface density of star formation , @xmath74 , defined as the sfr per unit area , or in other words the star formation intensity . there have been numerous sfls and here we summarise and test the most popular ones in the literature ( l08 ; * ? ? ? * ; * ? ? ? * and references therein ) . [ sfls ] probably the most well known and widely used sfl is the schmidt - kennicutt law @xcite . it is a simple empirical correlation between star formation intensity and total gas : @xmath75 where @xmath76 is a coefficient with unit of [ @xmath77 , @xmath16 is the surface mass density of the gas ( both neutral and molecular ) , and @xmath78 @xcite . here we fix @xmath79 in our tests of @xmath80 . the power law index @xmath81 of schimidt - kennicutt law is very close to 1.5 , which can be explained by arguing that stars form in a free - fall time scale in a gas disk with fixed scale height : since the free fall time scale @xmath82 is proportional to the inverse square root of the local gas density @xmath83 , then for a fixed scale height @xmath84 , the star formation intensity is @xcite : @xmath85 therefore the star formation intensity in a free - fall timescale with fixed scale height can be written as @xmath86 where @xmath87 has the same unit as @xmath76 . so this version of the sfl is nearly the same as the schmidt - kennicutt sfl with a slightly higher @xmath88 . if the disk scale height is not fixed but set by hydrostatic equilibrium , then the star formation law ( still assuming stars form in disk free - fall time scale ) is @xcite @xmath89 where @xmath90 has the same unit as @xmath76 . if we instead assume that a constant fraction of the ism is consumed in a dynamical timescale , we would have @xmath91 where @xmath92 is a dimensionless coefficient and @xmath93 is the orbital angular frequency @xcite . following the suggestion of this functional form by @xcite and @xcite , @xcite showed that this form of the sfl was able to account for the star formation intensity in his sample of normal galaxies and circum - nuclear star burst , and worked equally well as @xmath80 . @xcite presented a star formation model which assumes that star formation is triggered by gmc collisions in a shearing disk . this predicts that @xmath94 where @xmath95 is a dimensionless coefficient , @xmath96 and @xmath97 is the circular velocity at a particular galactocentric radius @xmath20 . note @xmath98 for a flat rotation curve . the gmc collision sfl is only meant to be valid in regions where a significant fraction of the gas is in gravitationally bound clouds , which typically means @xmath99 . however , here we test it over a wider range of densities . @xcite and @xcite studied 12 nearby spiral galaxies at sub - kpc resolution and concluded that the sfr is proportional to the molecular content : @xmath100 where the coefficient @xmath101 has the same unit as @xmath76 . @xcite provided a turbulence - regulated star formation model to predict the sfr by assuming stars primarily form in molecular clouds that are virialised and supersonically turbulent and that the probability distribution of densities is lognormal . the intensity of star formation by their model is given by @xmath102 where the coefficient @xmath103 has the same unit as @xmath76 , @xmath104 is the mass fraction of gas in giant molecular clouds ( gmcs ) and can be approximated as @xmath105 , and @xmath106 @xcite . @xcite presented a two - component star formation law , @xmath107 where the coefficient @xmath108 has the same unit as @xmath76 , and @xmath109 is a ` critical ' gas surface mass density . in regions with @xmath110 , gmcs have an internal pressure that far exceeds the ambient gas pressure and the star formation time scale is independent of the environment ; whilst in regions with @xmath111 , the star formation time scale depends on the metallicity and the clumping . our model requires at least three inputs : the rotation curve , gas velocity dispersion and the stellar mass distribution . the metallicity is also needed if using the kr relation ( cf . [ rmol ] ) . the basic assumption for our model is that the galactic disk is in a gravitationally marginal stable state and the @xmath2 relations and sfls are valid for all the galaxies . our algorithm is as follows : 1 ) . determine the radial total gas distribution ( @xmath112 ) assuming a marginally stable disk . we make the assumption that the two - fluid stability parameter , @xmath113 , is constant through out the whole galactic disk . we test all four two - fluid stability parameters ( @xmath4 , @xmath1 , @xmath5 , @xmath6 , cf section [ threshold ] ) on our sample galaxies and pick out the one most suitable for our model and then adopt it for all further calculations . we define ` most suitable ' in the first place as having the flattest q profile with the smallest rms deviation over all or most of the sample galaxies , and as a secondary consideration the simplest form . based on eq . ( [ qg])-([qrw2 ] ) , we calculate the gas surface mass density @xmath16 from the rotation curve @xmath114 , gas velocity dispersion and stellar surface mass density @xmath26 by fixing the stability parameter @xmath0 to the average value at intermediate radii . 2 ) . determine the radial molecular ( @xmath115 ) and neutral ( @xmath116 ) gas distribution we test three different ways to determine @xmath2 : the sr , pr and kr relations ( cf . section [ rmol ] ) . once we have @xmath2 we determine the molecular gas surface mass density based on our model predicted @xmath117 from step 1 @xmath118 and neutral gas surface mass density @xmath119 we then compare the derived molecular and neutral gas surface mass density to the observations to determine which @xmath2 relation fits the data best . 3 ) . determine the star formation intensity ( @xmath120 ) after obtaining the @xmath72 and @xmath73 , we use the sfl(s ) introduced in section [ sfls ] to calculate the star formation intensity @xmath74 . following @xcite , we fit each galaxy using every sfl described in section [ sfls ] . the results are evaluated to determine which sfl(s ) fit the observed @xmath74 best so as to implement in future application of our model . in order to construct our model , we need the following quantities as inputs : the rotation curve @xmath114 , the stellar surface mass density @xmath26 , the gas velocity dispersion @xmath14 , and metallicity . also , for comparison with the observations , we need data for the neutral gas surface mass density @xmath72 , the molecular gas surface mass density @xmath73 and the star formation intensity @xmath74 . the best sample , which has all the data listed above , is a subset of the things sample ( l08 ; de blok et al . it is composed of five dwarf galaxies ( ddo 154 , ic 2574 , ngc 7793 , ngc 2403 , and ngc 925 ) and seven spiral galaxies ( ngc 3198 , ngc 4736 , ngc 6946 , ngc 3521 , ngc 5055 , ngc 2841 , and ngc 7331 ) . the @xmath26 , @xmath72 and @xmath73 are tabulated in the paper of l08 , although the dwarf galaxies do not have co data and thus do not have @xmath73 . the @xmath114 is provided by @xcite . l08 used a fixed value of @xmath121 . however , in general @xmath14 is observed to vary in galaxies @xcite . therefore , we tried both the measured @xmath14 and a fixed @xmath122 when calculating the @xmath0 . the measured @xmath14 are from a. c. primo & f. walter ( private communication ) and plotted in fig . [ fig_vsig ] the metallicity data is taken from @xcite . we use a linear fit for each galaxy based on its metallicity zero point and radial gradient @xcite provided by moustakas ( private communication ) . the solar metallicity is taken as @xmath123(o / h)=@xmath124 @xcite . for the rotation curve @xmath114 , we fit the data using the universal rotation curve ( urc ) @xcite . the urc can be parameterised as @xmath125 where @xmath126 is the radial variable . nominally , @xmath127 is the radius encircling @xmath128 of the light , @xmath129 is the velocity at @xmath130 , while @xmath131 and @xmath132 are constants that depend on the luminosity . here we keep the @xmath129 , @xmath127 , @xmath131 , and @xmath132 as free parameters . the rotation curve data and their urc fits are shown in fig . [ vrot ] . the main advantage of using such a smooth rotation curve fit is to have a well defined and realistic derivative of @xmath114 . this is important when calculating @xmath17 ( eq . [ e : kappa ] ) and thus essential for any version of @xmath0 . we note that an overly tight fit to all the kinks and wiggles in the rotation curve amplifies these features in the derivative , producing unrealistic results , especially when the fitted @xmath114 profile declines faster than keplerian . it is likely that these small amplitude variations in @xmath114 result from noise in the data and small scale non - circular motions due to bars , spiral arms , and asymmetries . the urc parameters are fitted using the mpfit software package @xcite to @xmath114 . we also tried two other rotation curve parameterisations : the exponential form ( l08 ) and the arctan form @xcite . generally , the urc model fits the data best , therefore we use the urc fit for all the rotation curves in this paper . the urc model fitting sometimes results in unphysical parameters ( e.g. the fitted @xmath133 of ddo 154 is 12.1 kpc , far greater than the observation , 1.2 kpc ; l08 ) , but we use them anyway because we are only concerned with getting a smooth form which can best represent the data . the four two - fluid q models are fitted to a constant ( in forms of @xmath134 ) using the mpfit software package . in order to assess the quality of the fit we use the `` deviation value '' @xmath135 which we define as @xmath136 where @xmath137 is the number of valid annuli in the galaxy . @xmath135 has units of dex ; values of @xmath135 closest to zero indicate the model that most closely represents the observations . the fitted values and their corresponding deviation values are shown in table [ qerr ] . because the observed @xmath0 are less constant in the central and outer regions , we only use the data within the intermediate disk during the fitting . in terms of flatness , the average @xmath135 listed at the bottom of the table [ qerr ] indicate that @xmath4 is marginally the best recipe , followed by @xmath6 , @xmath1 , and @xmath5 . it is satisfying that the form of @xmath0 that is most rigorously defined yields the lowest average @xmath135 , because it suggests that the feedback processes are truly working to maintain disks at a constant stability , at least over the intermediate radii . but since all these @xmath0 recipes result in typically a factor of @xmath138 deviations about the mean , we choose to use @xmath1 in our calculations as the easiest to implement , and sufficient for our purposes . l c c c c c c c c c c c c c c [ qerr ] ddo154 & 4 & 3.50 & 0.10 & 2.80 & 0.10 & 3.26 & 0.16 & 4.48 & 0.19 & 50.00 & 7.10 & 8.70 & 0.20 + ic2574 & 27 & 1.46 & 0.21 & 1.26 & 0.20 & 1.30 & 0.21 & 1.84 & 0.21 & 134.00 & 8.70 & 9.30 & 0.13 + ngc7793 & 22 & 1.77 & 0.19 & 1.70 & 0.21 & 1.79 & 0.18 & 2.34 & 0.19 & 115.00 & 9.50 & 9.10 & 0.55 + ngc2403 & 33 & 1.96 & 0.18 & 1.89 & 0.20 & 1.98 & 0.17 & 2.61 & 0.17 & 134.00 & 9.70 & 9.50 & 0.49 + ngc0925 & 22 & 1.82 & 0.22 & 1.36 & 0.23 & 1.44 & 0.27 & 1.98 & 0.25 & 136.00 & 9.90 & 9.82 & 0.43 + ngc3198 & 14 & 1.82 & 0.16 & 1.42 & 0.15 & 1.58 & 0.23 & 2.09 & 0.21 & 150.00 & 10.10 & 10.12 & 0.37 + ngc4736 & 16 & 1.85 & 0.27 & 1.73 & 0.29 & 1.78 & 0.28 & 2.20 & 0.29 & 156.00 & 10.30 & 8.95 & 3.45 + ngc6946 & 24 & 1.55 & 0.26 & 1.21 & 0.31 & 1.53 & 0.23 & 2.05 & 0.20 & 186.00 & 10.50 & 10.01 & 0.31 + ngc3521 & 18 & 1.57 & 0.17 & 1.19 & 0.21 & 1.50 & 0.20 & 1.99 & 0.19 & 227.00 & 10.70 & 10.15 & 0.66 + ngc5055 & 25 & 1.67 & 0.18 & 1.31 & 0.28 & 1.69 & 0.20 & 2.29 & 0.16 & 192.00 & 10.80 & 10.25 & 0.83 + ngc2841 & 15 & 3.99 & 0.17 & 2.52 & 0.19 & 2.94 & 0.21 & 3.66 & 0.20 & 302.00 & 10.80 & 10.11 & 0.14 + ngc7331 & 20 & 1.86 & 0.13 & 1.33 & 0.16 & 1.67 & 0.22 & 2.20 & 0.20 & 244.00 & 10.90 & 10.25 & 1.04 + mean & & 2.07 & 0.187 & 1.64 & 0.211 & 1.87 & 0.213 & 2.48 & 0.205 & 168.83 & 10.45 & 9.93 & 0.72 + figure [ q_vsig ] shows @xmath1 for each galaxy using two cases for the gas velocity dispersion : a fixed @xmath14(=11km / s ; l08 ) and the measured @xmath14 . apparently , there is little difference in these two situations . furthermore , we calculate the @xmath14 needed to maintain a constant @xmath1 and plot it on top of the measured @xmath14 in figure [ fig_vsig ] . in most cases or at nearly all radii , the needed @xmath14 is within 1@xmath51 uncertainty of the data and very close to a constant @xmath139 as well . therefore , in order to keep the simplicity of our model , we use the fixed @xmath140 in the following calculations . radial profiles . black open circles are @xmath1 calculated using @xmath139 and black solid lines are their best constant fit ; green open diamonds are @xmath1 calculated using measured @xmath14 , and green dash lines are their best constant fit . error bars show 1@xmath51 uncertainty . ] needed in order to keep a constant @xmath1 with observed rotation curve and @xmath16 and @xmath26 . red solid and dashed lines are the needed gas velocity dispersion calculated using the average @xmath0 value for the particular galaxy from table [ qerr ] and a fixed @xmath141 respectively . measured gas velocity dispersion with 1@xmath51 uncertainty are also over - plotted as black dots and error bars . the @xmath14 data are from primo & walter ( private communication ) . ] we next calculate the neutral and molecular gas surface densities using the three @xmath2 relations : sr , pr and kr ( cf . section [ rmol ] ) . we use the fitted @xmath1 values tabulated for each galaxy in table [ qerr ] to derive the gas distribution . the calculated @xmath72 and @xmath73 radial profiles are shown in fig . [ sigma_hi ] and fig . [ sigma_h2 ] respectively . we calculate deviation values @xmath142 and @xmath143 for the three @xmath2 relations ( @xmath144 , @xmath145 , etc . ) using equations analogous to eq . [ eq_epsilon_q ] to compare how well the @xmath72 , and the @xmath73 observations and model agree . a cursory comparison of fig . [ sigma_hi ] with fig . [ sigma_h2 ] suggests that the hi profiles do not fit the data as well as the h@xmath53 profiles . however , this is largely illusory - the @xmath72 profiles are much flatter than the @xmath73 profiles for which a larger display range is required . comparison of the @xmath142 and @xmath146 values in table 2 demonstrate that in general the hi profiles are fit by the model about as well or better than the h@xmath53 profiles . all three @xmath2 model predictions agree with the data ( cf . table [ rmol_tab ] ) reasonably well with the average @xmath147 and the average @xmath148 , equivalent to fractional errors better than 0.6 and 0.9 respectively . the kr relation is the best in terms of the @xmath142 values and the sr relation is the best in terms of @xmath146 values . however , in some low metallicity regions and/or galaxies ( e.g. ddo 154 and ic 2547 ) the kr relation results in no molecular gas . we do sometimes have a very bad fit ( e.g. @xmath72 of ngc 2841 ) or even failure ( e.g. @xmath72 of ngc 4736 ) . the failure is caused by the big dip in the measured @xmath1 curve ( fig . [ q_vsig ] ) which means the disk is already nearly unstable from just the stars . the predicted hi surface mass density is usually flat over the radii considered here , but the model sometimes has a big divergence from the data in the outer regions where the observed hi has a steep decline , e.g. in ngc 925 and ngc 6946 . note that the @xmath149 relations are tested using our cq - disk model predicted @xmath117 , not the observed total gas surface mass density . the best @xmath149 relation we find here need not necessarily be the best one would find when applying to observed ism profiles . however , our results are consistent with those of l08 , who found the sr relation best in determining the molecular contents using observed values . radial profiles . black dots and error bars are measured data with 1@xmath51 uncertainty ; and red solid , green dash and blue dot lines are model derived @xmath72 using sr , pr and kr @xmath2 relations ( cf . section [ rmol ] ) respectively . ] radial profiles . conventions follow fig.[sigma_hi ] . there are only 7 galaxies shown here because we do not have co data for the other 5 galaxies . ] l c c c c c c c c c c c [ rmol_tab ] ddo154 & 4 & 0.02 & 0.13 & 0.02 + ic2574 & 27 & 0.06 & 0.14 & 0.06 + ngc7793 & 22 & 0.24 & 0.27 & 0.13 + ngc2403 & 33 & 0.18 & 0.21 & 0.10 + ngc0925 & 22 & 0.08 & 0.13 & 0.08 + ngc3198 & 14 & 0.06 & 0.11 & 0.05 + ngc4736 & 9 & 0.64 & 0.51 & 0.53 + ngc6946 & 24 & 0.13 & 0.13 & 0.11 + ngc3521 & 18 & 0.10 & 0.16 & 0.09 + ngc5055 & 25 & 0.14 & 0.11 & 0.17 + ngc2841 & 15 & 0.25 & 0.23 & 0.20 + ngc7331 & 20 & 0.10 & 0.13 & 0.06 + median & & 0.13 & 0.14 & 0.10 + mean & & 0.17 & 0.19 & 0.14 + & & & & + ngc3198 & 4 & 0.20 & 0.24 & 0.31 + ngc6946 & 17 & 0.23 & 0.26 & 0.24 + ngc3521 & 12 & 0.24 & 0.35 & 0.21 + ngc5055 & 12 & 0.20 & 0.25 & 0.20 + ngc2841 & 8 & 0.13 & 0.38 & 0.47 + ngc7331 & 8 & 0.21 & 0.25 & 0.17 + median & & 0.21 & 0.26 & 0.24 + mean & & 0.20 & 0.29 & 0.27 following @xcite , we test all sfls introduced in section [ sfls ] using the derived @xmath72 and @xmath73 ( cf . section [ result_gas ] ) as the input gas surface mass density values . for each galaxy , we derive the best - fit coefficients ( e.g. @xmath76 , @xmath92 , etc . ) for these sfls and measure the resultant uncertainty , @xmath150 , which is defined in a similar way to eq . [ eq_epsilon_q ] . since we have three different @xmath2 models , we also have three different sets of @xmath72 and @xmath73 ( cf . section [ result_gas ] ) and therefore three different fitted sfl coefficients.the coefficients and the corresponding @xmath151 values are listed in table [ sfl_parameter1 ] and [ sfl_parameter2 ] . the resulting @xmath74 profiles from the sfls are plotted on top of the observed data in figs . [ sfr_s ] , [ sfr_p ] , and [ sfr_k ] . comparing the curves in fig . [ sfr_s ] , [ sfr_p ] , and [ sfr_k ] the sfls that depend on separating the molecular and neutral phases generally fit better than the sfls that just depend on the total cool and cold ism content . the predictions based on the sr relation performs better than pr and kr relations in terms of sfr predictions . the kr relation generally performs the worse in terms of sfr predictions . in addition , we see from table [ sfl_parameter2 ] that the coefficient of the fit strongly correlates with stellar mass ( as seen in the ordering of the table rows ) when using the kr prescription for @xmath2 . this means that sfl with this prescription is not universal . the @xmath152 values in tables [ sfl_parameter1 ] and [ sfl_parameter2 ] also support the view that the the molecular sfls using the sr @xmath2 relation generally work the best . the exception to this rule is the dynamical time sfl ( sfr@xmath153 ) which outperforms the turbulence regulated molecular sfl ( sfr@xmath154 ) using the sr relation and all molecular sfls using the pr and kr relation . however , the values in tables [ sfl_parameter1 ] and [ sfl_parameter2 ] are the fit results using the intermediate radii data only . in terms of the simultaneous fit to all data points it is the third best sfl ( @xmath155 ) following the linear molecular sfl ( best , @xmath156 ) and the two component sfl ( second best , @xmath157 ) . this is slightly different from the results of @xcite who found the two component ( kmt ) sfl to be the best sfl in terms of rms dispersion . the best fit to all data for the linear molecular sfl yields a coefficient @xmath158 using the sr relation . this is slightly higher than the l08 value , @xmath159 , but well within its 2@xmath51 uncertainty . the higher coefficient implies 80% more efficient star formation and shorter molecular gas cycling times than those derived by l08 . note that the gmc collision sfl is meant to be applied to molecular gas rich regions only ( cf . section [ sfl_cc ] ) . it is therefore not surprising that our application of this sfl in regions that are not dominated by molecular gas , such as central region of ngc 2841 , results in a poor match to the observed @xmath160 . once more , one should bear in mind that the best sfl is selected based on our model predicted gas contents ( @xmath161 or @xmath162 ) . this is different from the standard approach of using the observed @xmath161 or @xmath162 to predict @xmath160 . most of the best - fit sfl coefficients ( a or b coefficients in table [ sfl_parameter1 ] and [ sfl_parameter2 ] ) , except the coefficient of the gmc collision sfl ( @xmath95 ) and the coefficient of the kmt sfl ( @xmath108 ) , are on the order of but a little bit larger than those of @xcite , who apply the sfls onto observed gas surface mass densities . since we have slightly different sample galaxies with @xcite , it is possible to have very different fitting values , however in fact , we do have similar @xmath95 and @xmath108 values to those of @xcite for galaxies in the overlapping part of our and tan s sample . the model - observation deviation ( @xmath151 values in table [ sfl_parameter1 ] & [ sfl_parameter2 ] ) have similar situation : they are also generally larger than but on the order of those in ( * ? ? ? * the @xmath63 values ) . for example , for the galaxy ngc 7331 , we find @xmath163 for the linear molecular sfl while @xcite determines @xmath164 for this sfl . however , the combination of the sr prescription for @xmath2 and the linear molecular sfl is claimed to provide the best results by l08 , which is consistent with our results . l c c c c c c c c c c c [ sfl_parameter1 ] ddo154 & 4 . & 2.51 & 0.155 & 3.28 & 0.156 & 12.49 & 0.159 & 4.01 & 0.062 & 41.62 & 0.128 + ic2574 & 27 . & 1.66 & 0.274 & 2.14 & 0.274 & 7.49 & 0.272 & 9.49 & 0.273 & 33.78 & 0.343 + ngc7793 & 22 . & 6.84 & 0.491 & 8.86 & 0.489 & 32.29 & 0.484 & 11.61 & 0.458 & 42.52 & 0.577 + ngc2403 & 33 . & 6.71 & 0.394 & 8.72 & 0.393 & 32.04 & 0.387 & 10.98 & 0.308 & 57.46 & 0.407 + ngc0925 & 22 . & 3.30 & 0.200 & 4.27 & 0.194 & 15.42 & 0.167 & 20.39 & 0.204 & 64.03 & 0.273 + ngc3198 & 14 . & 4.96 & 0.200 & 6.41 & 0.192 & 22.42 & 0.145 & 10.51 & 0.152 & 40.55 & 0.246 + ngc4736 & 9 . & 12.38 & 0.838 & 12.65 & 0.856 & 14.13 & 1.394 & 0.63 & 1.077 & 14.77 & 1.293 + ngc6946 & 24 . & 16.23 & 0.355 & 19.27 & 0.344 & 43.65 & 0.294 & 23.42 & 0.270 & 82.37 & 0.360 + ngc3521 & 18 . & 5.43 & 0.401 & 6.43 & 0.379 & 14.31 & 0.257 & 6.91 & 0.296 & 20.21 & 0.399 + ngc5055 & 24 . & 6.83 & 0.520 & 8.58 & 0.505 & 25.73 & 0.418 & 11.27 & 0.418 & 30.35 & 0.468 + ngc2841 & 15 . & 5.79 & 0.341 & 7.18 & 0.390 & 20.22 & 0.676 & 3.50 & 0.375 & 26.96 & 0.605 + ngc7331 & 20 . & 4.31 & 0.232 & 5.32 & 0.212 & 14.44 & 0.103 & 5.92 & 0.101 & 14.86 & 0.111 + all & 232 & 9.04 & 0.557 & 11.10 & 0.543 & 28.06 & 0.523 & 12.10 & 0.385 & 39.18 & 0.501 + median & & 5.79 & & 7.18 & & 20.22 & & 10.51 & & 40.55 + mean & & 6.41 & & 7.76 & & 21.22 & & 9.89 & & 39.12 + l c c c c c c [ sfl_parameter2 ] sr relation + ddo154 & 30.94 & 0.060 & 77.70 & 0.142 & 71.09 & 0.058 + ic2574 & 17.70 & 0.255 & 83.59 & 0.204 & 38.85 & 0.251 + ngc7793 & 13.22 & 0.169 & 40.33 & 0.291 & 29.35 & 0.168 + ngc2403 & 11.33 & 0.108 & 27.99 & 0.165 & 24.20 & 0.125 + ngc0925 & 9.83 & 0.195 & 61.09 & 0.165 & 20.74 & 0.232 + ngc3198 & 6.90 & 0.181 & 16.08 & 0.193 & 14.19 & 0.229 + ngc4736 & 11.49 & 0.459 & 1.19 & 1.654 & 11.71 & 0.651 + ngc6946 & 14.80 & 0.159 & 15.28 & 0.197 & 23.46 & 0.146 + ngc3521 & 5.02 & 0.167 & 4.08 & 0.090 & 7.83 & 0.095 + ngc5055 & 6.72 & 0.269 & 7.81 & 0.135 & 12.51 & 0.196 + ngc2841 & 3.75 & 0.296 & 4.96 & 0.827 & 7.25 & 0.485 + ngc7331 & 4.37 & 0.056 & 3.84 & 0.261 & 7.65 & 0.138 + all & 9.51 & 0.300 & 8.37 & 0.642 & 15.61 & 0.334 + median & 11.33 & & 16.08 & & 20.74 + mean & 11.34 & & 28.66 & & 22.40 + pr relation + ddo154 & 3.86 & 0.124 & 11.38 & 0.066 & 8.84 & 0.126 + ic2574 & 2.71 & 0.254 & 10.78 & 0.284 & 5.88 & 0.253 + ngc7793 & 8.03 & 0.432 & 19.49 & 0.494 & 17.80 & 0.428 + ngc2403 & 8.04 & 0.327 & 20.86 & 0.330 & 17.62 & 0.320 + ngc0925 & 4.83 & 0.140 & 30.42 & 0.163 & 10.45 & 0.133 + ngc3198 & 5.73 & 0.101 & 13.57 & 0.098 & 12.06 & 0.083 + ngc4736 & 11.67 & 0.832 & 1.18 & 1.926 & 11.90 & 1.040 + ngc6946 & 13.47 & 0.300 & 14.65 & 0.268 & 21.79 & 0.266 + ngc3521 & 4.56 & 0.309 & 3.90 & 0.194 & 7.28 & 0.228 + ngc5055 & 7.08 & 0.391 & 8.29 & 0.249 & 13.42 & 0.324 + ngc2841 & 4.82 & 0.565 & 5.64 & 1.116 & 9.01 & 0.770 + ngc7331 & 4.13 & 0.103 & 3.75 & 0.145 & 7.36 & 0.058 + all & 8.66 & 0.440 & 8.13 & 0.503 & 15.08 & 0.407 + median & 5.73 & & 11.38 & & 12.06 + mean & 6.58 & & 11.99 & & 11.95 + kr relation + ngc7793 & 76.85 & 0.396 & 223.23 & 0.426 & 167.85 & 0.406 + ngc2403 & 23.20 & 0.422 & 45.28 & 0.504 & 46.87 & 0.445 + ngc0925 & 15.66 & 0.254 & 98.27 & 0.230 & 32.49 & 0.259 + ngc3198 & 9.09 & 0.898 & 19.59 & 0.915 & 17.84 & 0.931 + ngc4736 & 12.46 & 0.454 & 1.18 & 1.499 & 12.71 & 0.647 + ngc6946 & 14.11 & 0.268 & 14.88 & 0.255 & 22.54 & 0.238 + ngc3521 & 5.07 & 0.193 & 4.08 & 0.206 & 7.83 & 0.204 + ngc5055 & 6.44 & 0.190 & 7.54 & 0.232 & 11.90 & 0.209 + ngc2841 & 3.60 & 0.637 & 4.62 & 0.869 & 6.80 & 0.732 + ngc7331 & 4.22 & 0.295 & 3.67 & 0.409 & 7.24 & 0.360 + all & 9.38 & 0.441 & 8.18 & 0.639 & 15.10 & 0.475 + median & 12.46 & & 14.88 & & 17.84 & + mean & 17.07 & & 42.23 & & 33.49 & + radial profiles using the sr prescription for @xmath2 . black dots and error bars are measured @xmath74 from l08 . thick solid lines are the three best fit sfls : red , linear molecular sfl ; blue , turbulence regulated sfl ; green , two - component sfl . dash lines are other sfl predictions : dark green , sk law ; orange , free - fall with fixed scale height ; sandy brown , free - fall with variable scale height ; yellow , orbital time scale ; purple , gmc collision . ] radial profiles using the pr prescription for @xmath2 . conventions follow fig.[sfr_s ] . ] radial profiles using the kr prescription for @xmath2 . conventions follow fig.[sfr_s ] . ] the basic assumption of our model is that the two - fluid stability parameter @xmath0 is a constant . as we can see from fig . [ q_vsig ] , @xmath1 is roughly constant , especially at intermediate radii . the variation can be large , up to a factor of 3 over the whole radial range sampled but the variations within the intermediate radii regions are smaller , less than a factor @xmath165 except for three cases ngc4736 , ngc 6946 , and ngc5055 , where the variations are a factor of @xmath166 ( see table [ qerr ] ) . there is also systematic variations with galactocentric radius : in 7/12 cases @xmath1 rises towards the centre , sometimes quite sharply ( ngc7793 , ngc2403 , ngc5055 , ngc2841 ) , while in 3/4 of the sample @xmath1 increases at large radii , mostly beyond @xmath10 . this systematic variation may cause over prediction of gas and sfr surface densities in the central and outer regions . a rapid central rise of @xmath1 can have several explanations , which are discussed in section [ centraldisk ] . the slow rise of @xmath1 beyond @xmath10 might be caused by undetected molecular gas in the outer disk , or may be due to the limited supply of ism as pointed out by @xcite . the gas velocity dispersion is hard to measure @xcite and has large uncertainties ( cf . [ fig_vsig ] ) . however , as shown in fig . [ q_vsig ] the shape of the @xmath1 profiles is not greatly affected by choosing a constant @xmath14 or adopting the observed @xmath14 profiles . in four cases ( ngc3198 , ngc3521 , ngc2841 , and ngc7331 ) the profiles are noticeably shifted vertically between the two options . this is because the measured @xmath14 is significantly higher than the assumed value of 11 km s@xmath167 . we can also see from fig . [ fig_vsig ] that the @xmath14 needed to keep a constant @xmath1 ( using the observed @xmath26 and @xmath16 in eq . [ qg ] , [ qs ] and [ qws ] ) is pretty constant and close to @xmath168 and well within @xmath169 uncertainty of the measured gas velocity dispersion . it might be surprising that @xmath1 does not vary much no matter whether we use a constant @xmath170 or the measured @xmath14 . this is because the definition of @xmath1 ( eq . [ qws ] ) is similar to the equivalent resistance of two resistors connected in parallel and @xmath23 is usually smaller than @xmath24 in the inner and intermediate disk regions . thus , @xmath1 is usually dominated by the @xmath23 value . this also illustrates that it is the total stability parameter , @xmath0 , instead of @xmath24 that matters most and the interaction between gas and stars plays an important role in balancing the stability of galactic disk . figures [ sfr_p ] - [ sfr_k ] , suggest that our model overestimates @xmath16 and/or @xmath74 in the central disk for many of the sample galaxies . in e.g. ngc 2841 , ngc 2403 and ngc 7331 , the model overestimates @xmath16 and @xmath74 by more than an order of magnitude . this result is similar to that of @xcite although he used a different stability criterion @xcite . @xcite argued that this is because the density - wave - induced shocks , which are very strong in the inner parts of the galaxy , make the gas in the post shock regions dense enough to be jeans unstable . thus their stability criteria , which does not consider shocks , may not be a good representation for the inner disk stability . however , our model does work well in the central part of other galaxies , e.g. ngc 7793 , and furthermore , our model sometimes underestimates the central disk sfr of galaxies like ngc 6946 and ngc 925 . possible reasons for the model - observation discrepancy in the central disk could be the fitted functional form of the rotation curves are not good enough to represent the data ( cf . section [ rcs ] ) , or the central disk is not a suitable place to apply our model because it is dominated by a bulge instead of being pure disk . despite these complications and the poor fit to the central region in detail , fig . [ sfr_ratio ] shows that the sfr integrated over the central region from our model agrees with the data to a similar level of accuracy as it does over the intermediate radii . the specific model we are using here employs the @xmath1 recipe for @xmath0 , the sr formulation of @xmath2 , and the linear molecular sfl @xmath171 . the relatively good agreement in the central region is because usually the model only fails badly in the very central part of the disk , typically covering an area less than 50% of the central disk . therefore , the integrated sfr does not deviate from the observation very much . the exception to this is ngc 2841 , where the modelled sfr profiles shown in figs . [ sfr_s ] - [ sfr_k ] deviate strongly from the observed over the entire central disk . . solid symbols are for intermediate disk and open symbols are for central disk . the specific model we are using here is the @xmath1 recipe for @xmath0 , the sr formulation of @xmath2 , and the linear molecular sfl @xmath171 . ] since @xmath3 is dependent on the derivative of the rotation curve ( through @xmath17 ) , the precise shape of the smooth fit must have a quite strong effect on the result , especially in the inner region , where the rotation curve rises rapidly . in order to test this effect , we use the exponential parameterisation to fit the rc of ngc 2403 and use these new fitting results to recalculate the gas and star formation distribution . the results are shown in fig . [ ngc2403_exp ] . the exponential form results in a shallower slope in the rising part of the rc and therefore a lowered gas and star formation surface density in the central region compared to the result of the urc fitting ; whilst the urc form fitting results in a steeper slope and thus a higher gas and star formation surface density in the central region . as expected , different forms of rc fittings make little difference in the middle and outer disk for galaxies with a flat rc . . red dots are calculated using the fitted urc form and green triangles are calculated using fitted exponential rotation curve . lower - left panel is hi surface mass density and lower - right panel is sfr surface density . in these panels the black dots with error bars are measured data with uncertainties . the red lines are calculated using the fitted urc rotation curve and green lines are calculated using the fitted exponential rotation curve , in both cases the appropriate constant @xmath1 is adopted . solid lines are results using the sr prescription for @xmath2 and dash lines use the pr prescription . the sfl used here is the linear molecular sfl . ] examination of figs . [ sigma_hi ] [ sfr_k ] and tables [ rmol_tab ] - [ sfl_parameter2 ] demonstrate that the cq - disk model does a reasonable job ( factor of @xmath166 agreement ) of matching the overall shape of the , co , and sfr radial profiles especially at intermediate radii . figure [ q_vsig ] indicates that these radii are where @xmath0 is not only fairly flat but at its minimum , i.e. the most unstable but uniformly so . this may be because at these radii star formation feedback is effective and the timescales for ism flows are sufficient for profiles to evolve to nearly constant stability . as emphasised in sec 6.2 , the model has the most difficulties fitting the detailed profiles in the central regions of galaxies , but still is fairly reasonable in predicting quantities integrated out to @xmath172 . even at intermediate radii , the model curves do not match the profiles in close detail , with the observed profiles having small scale bumps , dips , and kinks . often these local enhancements can also be seen in the profiles of @xmath0 , hi , h@xmath53 , and sfr ( e.g. the kinks in the profiles in ic 2574 and ngc 3521 at @xmath173 kpc ) . these indicate that local enhancements in the gas density are reflected in the sfr profiles as expected by any reasonable sfl , but result in the @xmath3 profiles not being flat in detail . while disks may be evolving towards a uniform stability , local deviations may build up due to processes beyond the scope of this model ( e.g. internal resonances , external perturbations , minor mergers ) . hence the cq - disk model is not as well suited to modelling the details of individual galaxies as the overall structure of galaxies in general . our sample is relatively small . further high quality observations would be useful to further test and optimise our model . new and developing generations of instrumentation including the jvla , askap , meerkat , westerbork+apertif for observations , alma , carma , and sma for co ( and thus h@xmath53 ) , wise , panstarrs , lsst for modelling stellar content , combined with star formation surveys such as hags @xcite , singg @xcite , and 11hugs @xcite as well as the wide availability of images of star formation from galex and wise data mean that it will be possible to do studies similar to ours for well selected samples of hundreds or thousands of galaxies in the near future . examination of figures [ sigma_hi ] [ sfr_k ] demonstrates that the various @xmath2 and sfl prescriptions diverge from each other the most in the outer and central regions of galaxies . for the reasons discussed in sec . 6.2 the centre is more difficult to model . so we concentrate on how further study of lsb galaxies and outer disks may improve our model . the sr and pr recipes diverge from each other there because the former only depends on the stellar surface mass density , while the hydrostatic pressure used in the pr prescription becomes more dominated by gas . the metallicity dependence of the kr relation combined with the metallicity gradients typically seen in the bright parts of galaxies @xcite will also drive a radial gradient compared to the sr and pr prescriptions . lsb regions are also important to test because of indications of low level star formation in extended uv disk galaxies @xcite and dominated regions @xcite . this low intensity star formation likely signifies imf variations @xcite which should be incorporated in future versions of a more comprehensive cq - disk model . comparison of the integrated mass and star formation rate may provide a sensitive test of our model since figures 4 , and 6 - 8 show that and star formation are largely segregated in galaxies , and the divergent behaviour of the various model profiles suggests that the /sfr ratio is likely to be dependent on the @xmath2 and sfl prescriptions . we will test this using data from the singg and sungg star formation surveys ( zheng et al . 2013 , in prep ) . the cq - disk model may prove useful in galaxy simulations . for example , currently when generating the initial distribution of particles for n body simulations of interacting galaxies one often creates disks with the same scale length for stars and gas ( e.g. * ? ? ? * ; * ? ? ? * ) or giving the stellar disks exponential profiles with differing scale lengths ( e.g. * ? ? ? * ) . assuming a constant @xmath0 disk provides alternative easy to implement initial conditions for detailed simulations of galaxies . one could also use the cq - disk model to enhance semi - analytic cosmological simulations , i.e. in a manner similar to that done by @xcite who modeled the detectability of galaxies in future surveys . a similar application of the cq - disk model will allow better line profile models , as well as models of h@xmath53 ( co ) , and star formation in the same volume . we have developed a simple ` constant @xmath3 disk ' ( cq - disk ) model for predicting the distribution of ism and star formation in galaxies based on the assumption that the two - fluid instability parameter ( @xmath0 ) of the galactic disk is a constant . the model predicts the gas surface mass density and star formation intensity given the rotation curve , stellar surface mass density and the gas velocity dispersion . in this paper we compared radial profiles of hi , and h@xmath53 surface mass density and star formation intensity from a sample of 12 galaxies from l08 . in order to optimise our model , we tried various prescriptions for calculating @xmath0 , the ratio @xmath174 , and the star formation law ( sfl ) . we find that * the @xmath0 profiles are fairly flat over the intermediate radii of the disk , with variations of a factor of @xmath138 about the mean , no matter which recipe of @xmath0 is employed . the @xcite formulation of @xmath0 has the strongest physical basis of the recipes we tried , and also marginally the flattest @xmath0 profiles . however it is the most difficult to implement since it requires the wavelength of the most unstable mode to be derived . the @xcite approximation , is the most practical recipe for @xmath0 in terms of its very simple form , and the one we have adopted in our model . * we tested three prescriptions of @xmath2 by comparing the observed surface mass densities of neutral and molecular gas with our model predictions . all three prescriptions produce @xmath72 and @xmath73 profiles that match the observed profiles with typical variations better than a factor of 2 about the mean . overall , the empirical scaling of @xmath2 with stellar surface mass density ( sr ) proposed by @xcite produces the best fits to both the @xmath72 and @xmath73 profiles , and therefore we favor this model . while the @xcite self - shielding prescription kr produces typically better matches to the @xmath72 profiles than the sr relation , the matches to the @xmath73 profiles are worse . although the best @xmath149 relation selected here is based on the predicted @xmath117 from our cq - disk model , it is consistent with the l08 results based on observed neutral and molecular ism profiles . * we tested eight versions of the sfl , five that depend on just the total ism content ( defined as combined neutral and molecular component ) and three that require the ism be separated in to neutral and molecular phases . the latter three were tested with all three prescriptions of @xmath2 that we trialled . the linear molecular sfl , sfr@xmath175 produces the best matches to the @xmath74 profiles when used with the sr @xmath2 relation , with the models typically agreeing with the observed profiles to within a factor of 2 . the two - component sfl sfr@xmath176 , combined with the sr relation does second best with agreement to typically a factor of 2.2 , while the orbital time sfl sfr@xmath153 , at third best in terms of @xmath177 value , is the best of the single component ism sfls with agreement to within a factor of @xmath178 over the intermediate radii disk . again , the best sfls here are selected based on our model predicted @xmath179 as well as @xmath73 , not the observed gas surface mass densities . however , the results are consistent with those from l08 and @xcite who do use observed ism profiles for their tests . this consistency also supports our cq - disk model in an indirect way . * the modelled star formation intensity profiles ( figs [ sfr_s ] , [ sfr_p ] , [ sfr_k ] ) match the observations best at intermediate radii , and show the largest deviations in the central region ( where @xmath180 ) . this suggest that a more elaborate model is needed to explain the ism and gas distribution in the centres of galaxies , especially those with bulges or a steeply rising rotation curve . however , integrating the sfr over the entire central region area we find that the models agree with the observations within a factor of 3 in all but one case . this indicates the systematic discrepancy integrated over galaxy centres generally is not severe . since we are testing a model for the distribution of star formation and gas in galaxies that we base on inferences from things team results , there is no surprise that our model works well for large portions of galaxies selected from the things sample . our future papers will test our model on galaxies selected independently . one advantage of our models is that they are easy to calculate given the rotation curve and the distribution of stellar mass . this may prove to be a useful advantage in terms of implementation compared to more detailed models like that of @xcite . however to handle the central regions , some reasonable modifications may be required , such as requiring gas and sfr to only be at @xmath181 for galaxies with @xmath182 above some fiducial value . we thank a. leroy and e. de blok for kindly providing the things data . we thank the referee very much for thoughtful comments and suggestions , which largely improved the content and clarity of our paper . we also thank anahi caldu primo and fabian walter for providing the gas velocity dispersion data . zz also thanks icrar for hospitality during his visit to perth , western australia . this work is funded in part by galex gi grant nnx09af85 g .

we introduce a model of binary data corruption induced by a brownian agent ( active random walker ) on a-dimensional lattice . a continuum formulation allows the exact calculation of several quantities related to the density of corrupted bits ; for example the mean of , and the density - density correlation function . excellent agreement is found with the results from numerical simulations . we also calculate the probability distribution of in , which is found to be log - normal , indicating that the system is governed by extreme fluctuations . = 10000 0.9truecm

cond-mat9811237

brownian motion is one of the fundamental processes in nature . originally observed in the irregular motion of pollen grains by the botanist brown@xcite , and cast into the language of the diffusion equation by einstein@xcite , it has now been applied , in the mathematical framework of random walks@xcite , to an enormous variety of processes in the physical sciences and beyond . a very rich field of research has been built up around the behavior of a random walk coupled to a disordered environment@xcite , a good example being the anomalous diffusion of electrons in a disordered medium@xcite . also , one can consider a random walker _ being _ the disordering agent in its environment . applications of the latter include the tagged diffusion of atoms in a crystal@xcite , or magnetic disordering due to a wandering vacancy@xcite . in this paper we will introduce a particularly simple example of an active random walker ( or brownian agent ) disordering its environment . although the model is interesting in its own right , we believe it will have a useful application to the study of data corruption in ultra - small storage devices . before pursuing this connection , we shall briefly describe the model ( which will be more carefully defined in the next section ) . the two main features of the model are first , the brownian agent ( ba ) performs a pure random walk it is not affected by the environment in any way . second , the environment is bistable . that is to say , it is composed of elements which may only exist in one of two possible states ( see ref.@xcite for a loosely related random walk process ) . thus we can consider the environment to be composed of binary data ( our favored realization ) , magnetic spins , chemical species a and b , and so on . as the ba wanders through the environment it has a certain probability to switch the value of an element in its immediate vicinity . thus , if we start with a system in which all elements exist in the same state ( ` up ' say ) , and introduce the ba at the origin , then after some time , there will be a region around the origin in which the elements will be found in a mixture of ` up ' and ` down ' states . naturally , the linear size of the region will grow on average as @xmath4 . the more subtle question concerns the degree of disordering which exists for elements within this region , and also their spatial correlations .. as we shall see , the statistics of the disordered elements are very rich . this is most convincingly demonstrated by the dominance of extreme fluctuations ; for instance , the distribution of disordered elements is log - normal . thus , typical and average events are quite distinct , and become ever more so as time proceeds . before giving an outline of the paper we shall say a few words about the potential relevance of this system to data corruption . with the advent of semiconductor memories ( for dynamic random access memories ( dram s ) and various types of read only memories ( rom s ) ) , there has been a tremendous drive within the semiconductor industry to produce ever - smaller memory devices@xcite . there are many properties ( _ e.g. _ stability , power consumption , volatility , and cost ) which must be balanced in the design of such devices . these factors determine the type of material used , and the geometry , dimension and architecture of the device . ( for instance , three dimensional arrays have a very efficient address structure , and are stable against interference from bombarding alpha particles , but are very expensive to produce@xcite . ) one of the main issues is the stability , or reliability , of the device . in semiconductor memories , there are many physical effects which can create hard errors ( destruction or corruption of the device itself ) or soft errors ( corruption of the data stored in the device ) . in the latter category , the most common problems originate from electron clouds caused by alpha particles , but soft errors may also arise from electromigration and charge diffusion@xcite . the key point is that different corruption mechanisms operate on different time scales ( leading to the famous bath - tub curve of device reliability@xcite ) . it is therefore important to know on what time - scales one should expect significant corruption from a given process . the model we propose here ( namely data corruption via a ba ) is probably not relevant for today s semiconductor devices , since there are so many ` mesoscopic ' processes occurring on the level of a flip - flop , that subtle correlations due to a ba will be washed out . however , we can look ahead to the new generation of ( quantum ) storage devices , in which a single electron ( controlled in a gate via coulomb blockade ) can store one bit of data . in this case , a microscopic ba may indeed play an important role in data corruption , and it will be necessary to understand its time - scales and efficacy of operation , so that we can minimize its influence . this paper constitutes a first step towards gaining such an understanding . the outline of this paper is as follows . in the next section we shall carefully define the model ( using discrete space and time ) within the master equation formulation of stochastic processes . we shall derive some general statistical properties of the process , but we shall not enter into any explicit calculations . this is deferred to sections iii - v in which we introduce a very simple continuum theory for the process , which is motivated by viewing the process as a stochastic cellular automaton . in section iii we derive this continuum theory and , using the complementary descriptions of quantum mechanics ( _ i.e. _ the schrdinger equation and the feynman path integral ) , we shall demonstrate its equivalence to the master equation formulation of section ii . we then examine the case of one spatial dimension in section iv . the model is tractable , and a great deal of information may be derived concerning the mean density of disordered elements , their spatial correlations , and finally their entire distribution function . in section v , we briefly study higher dimensions , and derive some general statistical properties for the process , for an arbitrary spatial coupling between the ba and its environment . we also derive an expression for the mean density of disordered elements in two dimensions . in section vi we present results from extensive numerical simulations of the discrete process . in all cases , we find good agreement between the simulation results and the predictions of the continuum theory . we end the paper with a summary of the work , and some ideas for future study . we consider binary data bits on a @xmath0-dimensional hypercubic lattice . for convenience we shall represent each bit by an ising spin @xmath5 , where the index @xmath6 represents a discrete lattice vector . the spin takes the value @xmath7 for a data bit which is uncorrupted ( corrupted ) . thus , the initial configuration is a lattice of spins , all of which take the value @xmath8 . [ we prefer to describe the system almost exclusively in terms of the spin variables . thus we shall use phrases such as ` magnetization density ' , or ` global magnetization ' . the translation of these quantities to the corresponding properties for corrupted data bits is immediate as one only need replace @xmath5 by @xmath9 , where @xmath10 denotes the presence ( with value unity ) or absence ( with value zero ) of a corrupted bit . similarly , we shall often refer to the average magnetization density @xmath11 , which is related to the average density of corrupted bits @xmath12 via @xmath13 . ] we denote the position of the ba by the lattice vector @xmath14 . each time step , the ba has a probability @xmath15 to make a jump to one of its ( @xmath16 ) nearest neighbors . for the sake of generality , we will not insist that the ba always flips a spin ( _ i.e. _ changes a data bit ) as it moves . thus , on a given jump , we allow the ba to flip the spin at the site it is leaving , with a probability @xmath17 . in this section we shall describe the process via a master equation@xcite . namely , we shall define the dynamics through the evolution of the distribution @xmath18 , which is the probability that at time @xmath19 the ba is at position @xmath14 , and the spins have configuration @xmath20 . given the above rules , the master equation takes the form @xmath21 where @xmath22 represent the @xmath16 orthogonal lattice vectors ( which have magnitude @xmath23 ) . the natural quantities to extract from the distribution @xmath24 are conditional averages . the simplest is the mean value of the spin at position @xmath25 at time @xmath19 , given the ba is at position @xmath14 . this is defined via @xmath26 higher order conditional averages may be defined accordingly . performing the spin trace over the master equation with a weight of @xmath5 yields @xmath27\\ & - & { pq\over d } \ \theta ( { \bf r},t \ | \ { \bf r } ) \sum \limits _ { \bf l } \delta _ { { \bf r } , { \bf r}+{\bf l } } \ .\end{aligned}\ ] ] the above equation has a physically appealing form . the rate of change of @xmath28 has two contributions . the first is lattice diffusion , as given by the first sum on the r.h.s . the second contribution vanishes unless the spin in question is in the immediate vicinity of the ba , in which case it acts as a sink . at this point in the discussion it is worthwhile to consider the continuum limit . namely we take the time scale @xmath29 and the lattice scale @xmath23 to zero , and define a diffusion constant @xmath30 . we also introduce a coupling @xmath31 . then replacing the kronecker @xmath32-function in eq.([lomageq ] ) by a dirac @xmath32-function we find @xmath33 it is important to note that this continuum equation is not strictly derived from eq.([lomageq ] ) , as we have not proved that the continuum limit exists . in fact , we shall find that for @xmath34 , the lattice scale @xmath23 is crucial , and consequently we must soften the dirac @xmath32-function to a function @xmath35 which is sharply peaked ( over a linear scale @xmath23 ) with unit integral over @xmath36 . one actually expects this to be the case , as eq.([lomco ] ) is the imaginary - time schrdinger equation for a particle under the influence of a repulsive @xmath37-function potential . [ note the independent spatial variable in this quantum system is @xmath14 , with the variable @xmath6 simply labelling the position of the potential . ] it is well known@xcite that the repulsive @xmath32-function potential is ` invisible ' to the particle for @xmath34 , and one usually cures this by smearing the potential just as described above . the quantum mechanics analogy will prove useful in the next section when we construct an alternative continuum model . before leaving this section we shall indicate the derivation of a non - trivial statistical relation hidden inside eq.([lomageq ] ) . first , we must define the initial condition . as mentioned before , given that up - spins denote uncorrupted data , the initial value of each spin is @xmath8 . there is a slight subtlety of definition regarding the value of the spin at the site where the ba is initially planted . ( this position shall be taken to be the origin , without any loss of generality . ) we shall take this spin to be initially @xmath38 so that immediately after the ba has moved away the spin at the origin has value @xmath8 . thus we have @xmath39 and consequently , @xmath40 we may obtain the following two averages from @xmath41 . the first is the average value of the spin at the origin . this is simply given by @xmath42 . the second average is the quantity @xmath43 , which corresponds to averaging the value of the spin at the site where the ba happens to be at time @xmath19 . one can prove that @xmath44 for all @xmath19 . we arrive at the above result by essentially solving the partial difference equation ( [ lomageq ] ) using discrete fourier and laplace transforms . the details can be found in appendix a. this result is useful for proving a more physically relevant relation . let us denote the average global magnetization by @xmath45 \ , \ ] ] where we have defined it relative to the ( infinite ) initial magnetization . this quantity essentially measures the average of the total number of corrupted bits ( up to a factor of 2 ) . summing eq.([lomageq ] ) over @xmath6 and @xmath14 gives @xmath46 then using ( [ iden ] ) we can rewrite the above relation in the form @xmath47 in other words , the rate of change of the mean global magnetization is proportional to the mean magnetization density at the origin . this is a non - trivial relation between a global and a local quantity . in principle , one can obtain exact results for many interesting quantities ( like the mean magnetization density , or correlation functions ) by directly solving for the conditional averages , as illustrated in appendix a. however , we prefer to obtain results from a continuum theory ; partly because the calculations are a little easier , but more importantly because we can access more sophisticated properties of the system , such as the probability distribution of the coarse - grained magnetization density . in this section we shall motivate a particularly simple continuum description of the data corruption process , and show its equivalence to the discrete theory of the previous section . there is an alternative method of characterizing the evolution of the system , other than using the evolution of the probability distribution @xmath18 via the master equation . this method consists of writing the local rules for the process in the spirit of a stochastic cellular automaton ( sca)@xcite . let us focus on the case that at each time step the ba makes a random jump to one of its nearest neighbors , and that the spin at the site which it leaves behind , definitely flips . this corresponds to setting @xmath48 . the local rules for such a process are easily written down . let us denote the time - dependent position of the ba by @xmath49 , a randomly chosen unit lattice vector by @xmath50 , and the time - dependent value of the spin at site @xmath6 by @xmath51 . then we have @xmath52 we are interested in a continuum limit of these two rules . the first is nothing more than a random walk . we take the lattice vector @xmath49 to be a continuum vector ( _ i.e. _ each of the @xmath0 components is a real number ) , and we replace the random unit lattice vector @xmath50 by a continuum vector @xmath53 , each component of which is a uncorrelated gaussian random variable with zero mean ( _ i.e. _ @xmath54 is a white noise process ) . the correlator of @xmath55 is given by @xmath56 where here and henceforth , angled brackets indicate an average over the noise ( or equivalently the paths of the ba ) . then , on taking @xmath57 , eq.([scadis1 ] ) assumes the form @xmath58 which is the familiar equation for a continuum random walker where @xmath59 is the diffusion constant@xcite . the second sca rule is more complicated to generalize to the continuum . as a first step let us define a coarse - grained magnetization density @xmath60 in the following way . we imagine defining a large region around the lattice site @xmath6 and summing all the spins in that region . their sum ( suitably normalized ) constitutes @xmath60 , with the label @xmath6 denoting a point in the @xmath36 continuum . an entirely analogous procedure is used in motivating the landau - wilson free energy functional from the ising model of ferromagnetism@xcite . the difficulty in our case is that we can not derive a closed equation for @xmath61 from the discrete rule ( [ scadis2 ] ) . we therefore make the following approximation . splitting the r.h.s . of ( [ scadis2 ] ) into two pieces , we see that the first may be taken over to the l.h.s . which may then be taken to be a time derivative in the limit of @xmath57 . the second piece resembles a decay term centered at @xmath62 . so , we postulate that the coarse - grained magnetization density satisfies @xmath63 where @xmath64 is a phenomenological parameter which describes how strongly the magnetization density is coupled to the ba . we stress that the field @xmath60 is a function of the continuous space and time variables @xmath6 and @xmath19 , and a _ functional _ of the path @xmath49 of the ba . now , the above heuristic derivation of the continuum theory was based on a sca for the case in which the ba always moves ( @xmath65 ) , and for which the spin located at the previous ba position is always flipped ( @xmath66 ) . in general @xmath15 and @xmath17 are both less than unity . intuitively we expect a very simple renormalization of our phenomenological parameters as @xmath15 and @xmath17 are changed . the diffusion constant @xmath59 should be proportional to @xmath15 , and the strength of the spin - ba coupling @xmath64 should be proportional to both @xmath15 and , more importantly , @xmath17 . thus we see a very close correspondence between @xmath59 and @xmath64 in the current continuum theory , and the parameters @xmath67 and @xmath68 which were introduced in the continuum limit ( [ lomco ] ) of the discrete equation ( [ lomageq ] ) . in fact they are identical , as will emerge in the following discussion . one of the positive features of the continuum theory as described by ( [ scacont2 ] ) is that one may immediately integrate the equation to find the magnetization density as an explicit functional of the path of the ba . as an initial condition we take @xmath69 for all @xmath6 . the subtlety encountered in the discrete theory concerning the initial value of the spin at the origin disappears here since the coarse grained function @xmath61 is not sensitive to the value of one inverted spin . straightforward integration of ( [ scacont2 ] ) yields @xmath70 \ .\ ] ] it is important to note at this stage that the magnetization density @xmath61 is clearly positive for all @xmath6 and @xmath19 . therefore within our continuum formulation , we have ignored paths of the ba which create large patches containing a majority of down spins ( _ i.e. _ corrupted bits ) . such patches will occur , but their frequency of occurrence is certainly very small since the system starts in a completely uncorrupted state . for instance , the probability for the ba to create a purely negative domain of @xmath71 spins is of the order @xmath72 . therefore , so long as we coarse - grain the original spin model over a sufficiently large scale , we can be confident that the most important configurations have been retained in the continuum theory . ultimately , one must justify such an approximation _ a posteriori _ by comparison with either results from the discrete theory , or from numerical simulations . as we shall see , both of these support the current continuum model and the approximations contained therein . we shall now connect the continuum theory as given by ( [ solcon ] ) , and the continuum limit ( [ lomco ] ) of the discrete theory , as given by ( [ lomageq ] ) . the mean magnetization density in the discrete theory is given by @xmath73 where in the continuum limit , we have replaced the sum over ba positions by an integral , and the field @xmath28 satisfies the imaginary - time schrdinger equation as given in ( [ lomco ] ) . in the alternative continuum theory , we can find the mean magnetization density by averaging the coarse - grained density @xmath74 over all paths @xmath49 . each path is weighted by a gaussian factor @xmath75 where @xmath76 is a normalization factor . therefore we can write the average of @xmath61 as a functional integral @xmath77 \right \rbrace \ , \\ & = & { \cal n } \int d^{d}r_{f } \int \limits ^{{\bf r}_{f } } { \cal d}{\bf r}(t ' ) \ { \rm exp } \left \lbrace - \int \limits _ { 0}^{t } dt ' \ \left [ { 1\over 2d ' } \left ( { d{\bf r}\over dt ' } \right ) ^{2 } + \lambda ' \delta ^{d } ( { \bf r}-{\bf r}(t ' ) ) \right ] \right \rbrace \ , \end{aligned}\ ] ] where we have used ( [ solcon ] ) in going from the first line to the second , and we have introduced the final position of the ba ( _ i.e. _ * r*(t ) ) as a free integration variable @xmath78 in rewriting the second line as the third . the reason for this cosmetic change is to make explicit the fact that @xmath79 can be expressed as a spatial integral over the final ba position , where the integrand is itself a path integral over ba trajectories . this path integral is nothing more than a re - expression of the solution of an imaginary time schrdinger equation ( using the well - known feynman path integral formulation of quantum mechanics@xcite ) for a particle in a repulsive @xmath32-function potential . we can now see the connection : eq . ( [ mmd2 ] ) is an exact restatement of eqs . ( [ lomco ] ) and ( [ mmd1 ] ) , with the identification @xmath80 and @xmath81 . [ so , henceforth we shall drop the primes in the material parameters . ] to summarize , by utilizing the complementary formulations of quantum mechanics via the schrdinger equation and the feynman path integral , we have shown that the continuum limit of the master equation is identical to the continuum theory constructed at the beginning of this section . two final points are in order . first , as noted in the previous section , the @xmath32-function potential must be replaced by a smeared function @xmath35 for @xmath34 ; thus in our continuum theory encapsulated in eq.([scacont2 ] ) , we shall make a similar replacement when studying two or higher dimensions . second , we have established a connection between the master equation and eq.([scacont2 ] ) only at the level of the first moment . it is straightforward to extend each formulation to higher order correlation functions , and indeed one finds an exact correspondence . for instance , we can define the conditional spin - spin correlation function within the discrete theory @xmath82 using the master equation one can show that in the continuum limit this function satisfies the schrdinger equation for a particle under the influence of two repulsive @xmath37-function potentials located at @xmath25 and @xmath83 . similarly , we can construct the coarse - grained two - point correlation function from eq.([solcon ] ) by evaluating @xmath84 . it is easy to see that this quantity is given by an integral over the analogous path - integral for two repulsive @xmath32-function potentials . having completed our formulation of a simple continuum theory , and shown its equivalence to the continuum limit of the master equation , we shall proceed to the next section in which we present a comprehensive solution of the model in one dimension . in this section we restrict ourselves to one dimension . this does not necessarily mean a single chain of sites . rather , we shall exclusively study the continuum theory of the last section , and in this case , for large enough times , @xmath3 refers to any system which has an infinite longitudinal dimension , and finite transverse dimensions ( for instance an infinitely long strip ) . this is the case , since as time proceeds , the correlation length will eventually become greater than the transverse size of the system , thereby only allowing the longitudinal fluctuations to continue growing , as is the case in a strictly one - dimensional system . the continuum model described in the previous section can be viewed as a ` non - conserved ' cousin of the continuum theory of vacancy mediated diffusion ( a process which in the spin language conserves magnetization ) introduced recently@xcite . an exact analysis of the latter theory was possible using infinite order perturbation theory in the spin - ba coupling @xmath68 . we shall use the same technique here , as it leads rather directly to a full solution . alternatively , one may solve the schrdinger equation for the conditional averages . however , there are some important quantities ( like the distribution of the magnetization density ) which can not be easily recovered from the latter approach . our starting point is the integrated solution of the continuum formulation as given in eq.([solcon ] ) . first , we shall derive an expression for the magnetization density @xmath85 . performing a direct average of eq.([solcon ] ) and expanding in powers of @xmath86 , we have @xmath87 where @xmath88 , and for @xmath89 @xmath90^{n } \right \rangle \ .\ ] ] we refer the reader to appendix b in which the above average is explicitly calculated . the result is @xmath91 where @xmath92 is the probability density of the ba . the structure of eq.([solchi ] ) is that of an @xmath93-fold convolution , so we may utilize a laplace transform to good effect . we have ( for @xmath89 ) @xmath94 performing the sum over these functions as dictated by eq.([avden ] ) we find @xmath95 \ .\ ] ] we note in passing that a similar result is easily derived for any @xmath96 . the case of @xmath34 is more complicated as the function @xmath97 is no longer integrable . this expression for the laplace transform of @xmath11 is exact . this will prove to be important when we come to evaluate the distribution function of @xmath98 . the inverse of the laplace transform is given by @xmath99 + \exp \left ( { \lambda |x| \over d } + { \lambda ^{2}t \over 2d } \right ) \ { \rm erfc } \left [ \lambda \left ( { t \over 2d } \right ) ^{1/2 } + { |x|\over ( 2dt)^{1/2 } } \right ] \ , \ ] ] where @xmath100 and @xmath101 are error functions@xcite . considering the long time behavior of the above expression , we have for @xmath102 @xmath103 \ .\ ] ] one can also retrieve the spatial behavior with little effort . for small @xmath104 we have @xmath105 the large @xmath104 behavior has two regimes : @xmath106 it is interesting to note that apart from the natural diffusion scale @xmath107 , there is a larger ` ballistic ' scale @xmath108 in the system , beyond which the disordering efficacy of the ba is much reduced , since it makes so few visits to these distant sites . there is no simple ( i.e. single length ) scaling form for @xmath85 . next we consider the continuum analogues of eqs.([iden])-([glomag3 ] ) . we define the average global magnetization ( relative to its initial value ) as @xmath109 \ , \ ] ] which may be compared to the discrete version in eq.([glomag ] ) . integrating and averaging the continuum model ( [ scacont2 ] ) yields @xmath110 which is to be compared with eq.([glomag2 ] ) . this last equation indicates that we may explicitly find an expression for @xmath111 by calculating the time derivative of the spatial integral of @xmath112 . this may be done at the level of the perturbation series ( [ avden ] ) , from which one may show that the following relation holds exactly , for all times : @xmath113 which is the continuum analogue of eq.([iden ] ) . finally , combining eqs.([glomagc2 ] ) and ( [ idenc ] ) we have @xmath114 thus , the non - trivial relation between the rate of change of the global magnetization , and the mean of the magnetization density at the origin , is seen to be exact within the continuum model ( which complements the exact relation @xmath115 found in the discrete framework ) . similar results are easily obtained for all @xmath96 . in section v we shall derive a slightly more complicated form of @xmath116 ( involving the smearing function @xmath117 ) which is appropriate for higher dimensional systems . directly from eqs.([resden0 ] ) and ( [ glomagc3 ] ) we note that @xmath118 ; thus , the average number of corrupted bits in @xmath3 increases as the square root of time . we now turn to spatial correlations in the system . these are most easily probed via the two - point correlation function @xmath119 \right ) \right \rangle \ , \end{aligned}\ ] ] where we have used the solution ( [ solcon ] ) in the second line . this average can be calculated using infinite order perturbation theory in @xmath68 , just as was used to evaluate @xmath85 . we write @xmath120 with @xmath121 . for @xmath89 , a given term @xmath122 can be explicitly evaluated by making integral representations of the @xmath93 @xmath32-functions , and performing the average over the paths @xmath123 ( as described in appendix b ) . thus one has @xmath124 \ , \end{aligned}\ ] ] with the notation @xmath125 . this @xmath126-fold integral can be reduced using laplace transform in time , such that the integrals over @xmath127 may be performed , as described in appendix c. the result is @xmath128 \right ) ^{n } \ .\ ] ] summing over these functions with a weight of @xmath129 and inverting the laplace transform using a bromwich integral@xcite , we have @xmath130 } \ , \ ] ] where , as usual , the contour @xmath131 is parallel to the imaginary axis , and to the right of any singularities . we have rescaled space and time as @xmath132 and @xmath133 . this integral may be evaluated for large @xmath134 in the following way . we re - express the integral as an expansion in powers of @xmath135 ( which is _ not _ the same as our original expansion in powers of @xmath68 ) . so we have @xmath136 with @xmath137 \ , \end{aligned}\ ] ] where the second line is the explicit form of the integral after integrating around the only singularity a branch point located at @xmath138 . the integral over @xmath139 may be simplified for @xmath140 to give @xmath141 in these rescaled units , we have from eq.([resden0 ] ) @xmath142 ( for large @xmath134 ) . thus , we may resum the functions @xmath143 to find @xmath144 \ , \end{aligned}\ ] ] where @xmath145 is a jacobi theta function ( with norm @xmath17)@xcite . note , we have written the last line in unscaled variables , and we see that the ratio of the correlation function to @xmath146 does not depend on @xmath68 for large times . the behavior of @xmath147 in the limits of large and small @xmath104 are as follows . for large @xmath104 , the fields at the origin and at @xmath104 will be uncorrelated , so that @xmath148 , the latter result following since @xmath149 for @xmath150 . at the other extreme , as @xmath151 , @xmath152 . referring to the exact solution of the continuum model , eq.([solcon ] ) , one can see that the second moment of the magnetization density is actually given exactly by @xmath153 ( where the last ( optional ) argument indicates the parametric dependence on the spin - ba coupling ) . so for long times we take the expression for @xmath154 given in eq.([resden0 ] ) and replace @xmath155 by @xmath156 . therefore @xmath157 for @xmath140 . thus , the limits of the function @xmath158 are 1/2 ( for small @xmath104 ) and unity ( for large @xmath104 ) , which is naturally consistent with the analytic form given above in terms of the jacobi theta function . in section vi we shall compare this expression with results from a numerical simulation of the discrete model described earlier . to complete our study of the properties of this system in one dimension , we shall consider the complete probability distribution @xmath159 of the magnetization density . we shall be able to calculate this exactly , since i ) we can see from ( [ solcon ] ) that the @xmath160 moment of the density is related to the mean density with a replacement @xmath161 ; and ii ) we have an _ exact _ expression for the mean density ( albeit in the laplace transform variable @xmath162 ) . the first point is a fortuitous property of our continuum model which we should certainly exploit . the second property is less obvious . one might imagine that , given we know all density moments via the first , even the asymptotic form for the mean density would be sufficient to calculate the probability distribution ( for large times ) . this is not the case as we shall see the complete analytic structure of @xmath163 is required in order to reconstruct the distribution @xmath164 . we define @xmath164 via @xmath165 where @xmath166 is the stochastic field solution given in eq . ( [ solcon ] ) . we can re - express the @xmath37-function using a frequency integral , and expand in powers of the field as follows : @xmath167 \rangle \\ \nonumber & = & \int \limits _ { -\infty}^{\infty } { d\omega \over 2\pi } \ e^{-i\omega \phi } \ \sum \limits _ { n=0}^{\infty } { ( i\omega ) ^{n}\over n ! } \langle \phi _ { r}(x , t)^{n } \rangle \\ & = & \int \limits _ { -\infty}^{\infty } { d\omega \over 2\pi } \ e^{-i\omega \phi } \ \sum \limits _ { n=0}^{\infty } { ( i\omega ) ^{n}\over n ! } m(x , t ; n\lambda ) \ , \end{aligned}\ ] ] the last line following from the property i ) alluded to above . so the laplace transform ( over time ) of @xmath164 is given in terms of the laplace transform of @xmath168 . from eq.([ltden ] ) we have @xmath169 \\ \nonumber \\ & = & { { \hat g}(0,s)-{\hat g}(x , s ) \over s{\hat g}(0,s ) } \ + \ { { \hat g}(x , s ) \over s{\hat g}(0,s)(1+n\lambda { \hat g}(0,s ) ) } \ , \end{aligned}\ ] ] where the second line follows from some algebraic manipulations . the first term is easily handled as it is independent of @xmath93 . thus the sum over @xmath93 for this term ( as is required in eq.([deltaex ] ) ) yields a factor of @xmath170 which finally yields a factor of @xmath171 when integrated over @xmath172 . the second term is more interesting . details of how to perform the sum over @xmath93 and the frequency integral may be found in appendix d. the final result for @xmath173 reads @xmath174 \ .\ ] ] this form for the laplace transform of the probability distribution may be easily generalized for any @xmath96 . finally we must invert the laplace transform . to this end we note that @xmath175 \ .\ ] ] inserting ( [ ltgreen ] ) into ( [ ltprob ] ) and inverting the transform we have our final result @xmath176 \ + \ { 1 \over ( \pi t)^{1/2 } } \ { 1 \over { \tilde \lambda } \phi } \exp \left \lbrace - \left [ { |x| \over ( 2dt)^{1/2 } } - { \log ( \phi ) \over 2{\tilde \lambda } t^{1/2 } } \right ] ^{2 } \right \rbrace \ , \ ] ] where @xmath100 is the error function@xcite , and we have defined @xmath177 . this is illustrated in fig.2 for @xmath178 ( and @xmath179 ) , and three different times corresponding to @xmath180 , @xmath181 and @xmath182 . in particular , the probability distribution for the magnetization density at the origin takes the form @xmath183 \ , \ ] ] which is a pure log - normal distribution . this result is very revealing , as it shows that the fluctuations in this system are extreme . for instance , we have already seen that the _ mean _ value of the density ( at the origin ) decays as @xmath184 . however , if one asks how the _ typical _ ( or most likely ) value decays , one can see from ( [ lognor ] ) that @xmath185 . thus , as time proceeds , the typical value of @xmath61 decays to zero exponentially fast , whilst the mean decays slowly as @xmath186 . this is possible because the log - normal distribution has a long tail , extending out to the extreme value of @xmath187 . in fact the end point of the distribution ( _ i.e. _ @xmath188 ) also decays as @xmath186 which is consistent with the known persistence properties of a random walker in @xmath3 ( namely , the probability of a walker _ never _ having returned to the origin after time @xmath19 decays as @xmath189 ) . in fig.3 we illustrate @xmath190 for three different times . as a final remark , we note that if we erroneously use the asymptotic form ( [ resden0 ] ) for @xmath191 to build the distribution function , we find that @xmath192 is equal to @xmath193 ; thus emphasizing the fact that we need the entire analytic form of @xmath191 to successfully construct the distribution @xmath164 . this ends a rather long section on the analytic properties of the continuum model in @xmath3 . in the next section we shall briefly study the case of higher dimensions , and then in section vi we shall compare our results with some numerical simulations performed on the original discrete model as we mentioned several times in the previous two sections , the continuum theory requires some regularization for @xmath34 . this can be most easily ( and physically ) accomplished by smearing the @xmath32-function interaction between the ba and the spins . thus , our continuum model takes the form @xmath194 where @xmath35 is a normalized function which is sharply peaked over a region of linear dimension @xmath23 around the point @xmath6 . [ a good choice would be @xmath195 . ] in this section we shall analyze some basic properties of eq.([scacontd ] ) for general @xmath117 . then we will use a somewhat more crude approach to estimate the mean magnetization density ( at the origin ) as a function of time in @xmath196 . although we have generalized our continuum theory somewhat , we can still make substantial headway by first integrating the equation of motion , and then using infinite order perturbation theory . the first step yields @xmath197 \ , \ ] ] whilst the second consists of expanding this equation in powers of @xmath68 and averaging term by term : @xmath198 where @xmath199 , and for @xmath89 @xmath200^{n } \right \rangle \ .\ ] ] by making a fourier representation of the interaction function @xmath117 , performing the average over paths ( see appendix b ) , and finally laplace transforming in time , we arrive at @xmath201 \ , \ ] ] with the convention @xmath202 . we shall be concerned with two quantities . the first is a smeared mean magnetization density near the origin , and the second is the global magnetization . these are given by @xmath203 and @xmath204\ ] ] respectively . we shall prove that @xmath205 which is the smeared analogue of the global versus ( strictly ) local relation ( [ glomagc3 ] ) we proved for @xmath206 . comparing eqs.([avdend ] ) and ( [ defd1 ] ) we see that @xmath207 where in laplace transform space @xmath208 similarly we have @xmath209 with @xmath210 for @xmath89 , @xmath211 , so that @xmath212 using eq.([chid ] ) we may evaluate the above integral to give @xmath213 performing the integral over @xmath214 immediately yields @xmath215 thus comparing eqs.([magsmex ] ) , ( [ glomagsmex ] ) and ( [ finalc ] ) we see the validity of relation eq.([idend ] ) . as a corollary , by integrating the averaged equation of motion ( [ scacontd ] ) over space , we have @xmath216 when compared with eq.([idend ] ) , the above relation gives us @xmath217 which is the smeared version of the local relation ( [ idenc ] ) proven in section iii . we note that , although we have been concerned with a sharply peaked interaction function , the relations ( [ idend ] ) and ( [ corol2 ] ) hold for _ any _ function @xmath218 . this ends the more rigorous part of the present section . in the remainder we shall just mention some explicit results for the mean local magnetization density ( at the origin ) , which are obtained with a cruder regularization . the difficulty with making headway using the smoothing function , is that the @xmath93-fold integrals over the @xmath117 s are intractable ( unless one can find a particularly ` friendly ' form for @xmath117 . ) as an alternative approach , we return to the sharp dirac @xmath32-function as used in section iv . we remarked that the @xmath93-fold convolution integrals were divergent due to the non - integrability of @xmath219 for @xmath34 . to evade this difficulty we can simply impose a cut - off into the integration limits . this is closely connected to introducing a microscopic time scale into the temporal correlations of the ba . such an regularization procedure was used in ref . @xcite , and the results so obtained were shown to be equivalent to previously known exact results@xcite . so we shall use the same procedure here , but with due caution . first , we consider @xmath196 . in a precisely analogous way to the calculation in section iv , we expand the field solution ( [ solcon ] ) in powers of @xmath68 and average term by term . thus , we have ( cf . eqs . ( [ avden ] ) - ( [ solchi ] ) ) @xmath220 where @xmath221 , and for @xmath89 @xmath222 in this case , the probability distribution of the ba at the origin has the form @xmath223 . note we have inserted the microscopic time regulator @xmath224 in the limits of the time integrals . our strategy is to evaluate the time integrals one by one , keeping only the most singular term at each step . we shall use the general result ( for @xmath225 ) @xmath226^{m } \over ( t-\tau)\tau } \sim { ( m+2 ) \over ( m+1 ) } \ { [ \log ( t / t_{0})]^{m+1 } \over t } \ .\ ] ] therefore we have the dominant contribution @xmath227^{n}\over ( 2\pi d)^{n } } \ .\ ] ] inserting this result into eq . ( [ avden2 ] ) and summing over @xmath93 we have the asymptotic form @xmath228^{-1 } \ .\ ] ] thus the magnetization at the origin does decay to zero for large times , but logarithmically slowly . from the relation ( [ idend ] ) we see that the mean global magnetization increases as @xmath229 . the same kind of analysis can be repeated for @xmath230 , and one finds that @xmath231 saturates to a constant for large times , which implies that @xmath232 for large times . these results are easily understood from the recurrent properties of the ba ( _ i.e. _ a random walker returns to its starting point with probability one , only for @xmath233 ) . it would be more interesting to derive the distribution of the magnetization density for @xmath234 , but this requires the more careful regularization method involving the smoothing function @xmath35 and thus lies beyond the scope of the present work . [ in @xmath196 one can capitalize on the slightly crude result ( [ avdenasy ] ) obtained using the cut - off @xmath224 , combined with the exact property @xmath235 to derive a form for the distribution function @xmath236 . such an approach yields @xmath237 , with @xmath238 . however , this result is not to be taken seriously , since we need the whole analytic structure of @xmath239 in order to derive @xmath24 , as evidenced in the previous section . ] we have performed extensive numerical simulations of the discrete model , as defined in section ii , in order to test the results obtained in the last two sections from the continuum theory . in all of the simulations for which we present results , we have set the hopping rate @xmath15 of the ba , along with the flipping probability @xmath17 , to unity . we have experimented with decreasing the flipping probability , and have found that its only effect is to renormalize the effective spin - ba coupling @xmath68 , such that @xmath240 , as expected . most of our results are obtained from a one - dimensional chain of sites . the chain length is unimportant , so long as one ensures that the ba has never touched the edges in any of its realizations up to the latest time at which data is extracted . generally we average over between @xmath241 and @xmath242 realizations ( or runs ) depending on the desired quality of the data . such simulations required a few days on a dec alpha 233 mhz workstation . in a given run , at each time step the ba is moved left or right with equal probability and the spin it leaves behind is flipped . each run starts with the same initial configuration ; namely all spins up , except the spin at the origin ( which is the starting site of the ba ) which is pointed down . ( this means that all spins are up after one time step , since the ba has moved away from the origin and flipped the down spin . ) in fig.4 we show the measured mean magnetization density at the origin , along with the quantity @xmath243 ( cf . eq.([lomag ] ) ) . they are seen to be identical thus confirming relations eq.([iden ] ) and its continuum counterpart ( [ idenc ] ) . the solid line is the asymptotic prediction ( [ resden0 ] ) from the continuum theory . it is seen to be in good agreement with the data , as the line has a slope of ( -1/2 ) . from the fit of this log - log plot we can read off the effective value of @xmath68 , since from ( [ resden0 ] ) the prefactor of @xmath186 is given by @xmath244 . ( the diffusion constant for the lattice random walk is unity . ) we have fitted the data to @xmath245 with @xmath246 , which yields @xmath247 . in fig.5 we plot the small @xmath248 dependence of @xmath85 on a log - log scale . the data is well fitted by the prediction given in eq.([resdensx1 ] ) . in fig.6 we plot @xmath249 versus @xmath248 . note that good data collapse is found for intermediate values of @xmath248 . we have been unable to numerically probe the ballistic scale @xmath108 . ( note , the theoretical curves shown in the last two figures are plotted with no free parameters . ) in fig.7 we plot the discrete time derivative of the total number of down spins @xmath250 ( which is @xmath251 ) , along with @xmath146 . the two curves are indistinguishable within the numerical noise , thus confirming the global / local relation ( [ glomag3 ] ) . this also provides secondary confirmation of the continuum form of this relation eq.([glomagc3 ] ) with @xmath252 . in fig.8 we plot the ratio of the measured two point correlation function ( divided by @xmath146 ) versus @xmath248 . note that it varies from 1/2 ( at small @xmath104 ) to unity ( at large @xmath104 ) as expected . the data from two different times is shown , and one sees excellent agreement with the theoretical prediction ( [ corrfinal ] ) , which is plotted with _ no _ free parameters . this agreement provides very strong evidence for the validity of our whole continuum approach . briefly , we mention simulations in @xmath196 . higher dimensional simulations are not too difficult as one is only ever moving the single ba at each time step . in fig.9 , we show @xmath253 and @xmath254 . the data for the two functions are identical verifying the discrete relation ( [ iden ] ) in two dimensions , as well as confirming the continuum result ( [ corol2 ] ) . we have plotted the inverse of these functions against @xmath255 in order to compare with the theoretical prediction ( [ avdenasy ] ) . again , good agreement is found , thereby confirming the less rigorous method by which the two dimensional result was obtained . finally we mention our attempt to measure the probability distribution of the coarse grained magnetization density at the origin ( in @xmath3 ) , which was found from the continuum theory to be a log - normal distribution . clearly it does not make sense to measure moments of the spin at the origin , since the odd ( even ) moments are equal to @xmath256 ( unity ) . therefore , we define a coarse grained magnetization over a patch of spins . if the patch is taken too small , the coarse - graining will be ineffective , whilst if the patch is taken too large , the ba will take a long time to leave the patch , and the asymptotic behavior will be numerically inaccessible . so we have compromised and have used a patch containing 21 spins . we have binned the patch magnetization from @xmath242 independent runs and generated the histograms shown in fig.10 . note , that because the patch size is modest , the histograms have non - zero weight in the negative @xmath61 region , in contrast to the strict continuum limit . however , we do see that for @xmath61 near unity , the histograms have a robust tail , which is the signature that extreme fluctuations are important . in this paper we have introduced and analyzed a simple model of data corruption due to a brownian agent ( ba ) . in section ii we introduced a discrete version of the model , which consists of a ba flipping bits ( or spins ) on a lattice . the model is non - trivial since the value of a given spin depends very sensitively on the path of the ba ( _ i.e. _ whether the spin has been visited an odd or even number of times ) . we presented a master equation formulation of the model and derived an equation of motion for the conditional average of the magnetization density . in the continuum limit , this quantity was seen to satisfy an imaginary - time schrdinger equation ( itse ) for a particle in a repulsive @xmath37-function potential . higher - order conditional averages also satisfy itse s with an additional repulsive @xmath32-function potential for each spin being averaged . we also proved that there is an exact proportionality between the rate of change of the mean global magnetization @xmath257 ( as defined in eq.([glomag ] ) ) and the mean magnetization density at the origin , thus revealing a non - trivial statistical relation between a global and a local quantity . in section iii we recast the discrete model as a stochastic cellular automaton ( sca ) . from the local rules of the sca , we postulated a particularly simple continuum theory written in terms of a coarse - grained magnetization density @xmath60 ( which is a functional of the walk @xmath49 of the ba ) . we showed that averages of this stochastic field may be recast as integrals over the final ba position , the integrands of which are imaginary - time path integrals ( itpi s ) . these itpi s were shown to describe quantum systems of repulsive @xmath37-function potentials identical to those revealed through the itse formulation of the master equation . in this way we have confirmed that the continuum theory is a good representation of the underlying discrete model . [ it is worth mentioning that averages of @xmath258 correspond to the quantum mechanics of an _ attractive _ @xmath32-function potential . in this case there is the possibility of a sharp transition in behavior for @xmath230 as one varies @xmath155 , since bound states only exist for a sufficiently attractive well . ] in section iv we thoroughly examined the properties of the continuum theory for @xmath3 . first , we derived an exact expression for the evolution of the magnetization density @xmath85 , and found its spatial variation for small and large @xmath104 . in particular we found @xmath259 for large times ; and also the existence of both a diffusive and a ballistic scale in @xmath85 . using the exact result for @xmath85 enabled us to prove that the rate of change of @xmath257 is proportional to @xmath146 for all times ( with proportionality constant @xmath68 ) which is the continuum analog of the global / local relation that was proved in section ii . we then studied the two - point correlation function @xmath147 . an exact expression was found for the laplace transform of this quantity , from which we were able to extract its long - time behavior , which is expressed in terms of a jacobi theta function as shown in eq.([corrfinal ] ) . the correlation function has an asymptotic scaling form ; namely the ratio @xmath158 depends only on @xmath248 . finally in section iv , we examined the probability density @xmath164 of the magnetization density . using a fortuitous property of the original continuum theory which enables us to represent the @xmath160 moment of the magnetization density in terms of @xmath85 , along with our exact expression for this latter quantity , we were able to derive an exact expression for @xmath159 as shown in eq.([probxt ] ) . in particular , setting @xmath102 reveals that @xmath260 is a pure log - normal distribution . this last result emphasizes the importance of extreme fluctuations in this system . for instance , @xmath146 decays as @xmath186 , whilst the typical value of the magnetization density at the origin ( i.e. the mode of @xmath164 ) decays as @xmath261 . in section v we briefly examined higher dimensions . for @xmath34 it is necessary to regularize the continuum theory , most appropriately with a smeared interaction between the ba and the spins . using an arbitrary interaction function @xmath35 ( which has a linear scale @xmath23 ) we were able to prove a more general form of the global / local relation , as shown in eq.([idend ] ) . we then concentrated on @xmath196 , and using a cruder regularization ( namely , introducing a microscopic correlation time @xmath224 ) we were able to derive an expression for the asymptotic decay of @xmath231 . this decay ( cf . eq.([avdenasy ] ) ) is seen to be logarithmically slow . in section vi we presented our results from extensive numerical simulations of the original lattice model . most of our numerical work examines the case of @xmath3 . we measured the spatial and temporal variation of the mean magnetization density , the mean global magnetization , and the two - point correlation function @xmath262 . in all cases we found excellent agreement between our data and the theoretical predictions arising from the continuum model . in particular , the agreement between the theoretical form for @xmath147 and the numerical data is very satisfying , as there are no free parameters to adjust . we also measured @xmath253 in two dimensional simulations , and found good agreement with the predicted logarithmic decay . finally we attempted to measure the probability distribution of the magnetization density , by binning the magnetization of a spin patch ( containing 21 spins ) from @xmath242 realizations . the results are not of high enough quality to directly compare with the derived log - normal distribution ; however , we do see clear evidence of a long robust tail of the histogram for values of @xmath61 near unity , which is a clear signature of the importance of extreme fluctuations . in conclusion , we have introduced and solved a model in which a ba interacts with a bimodal environment ( _ i.e. _ a medium containing two types of particles , spins , bits , etc . ) . our primary application has been an environment composed of bits of data , which the ba steadily corrupts . we have been interested in the statistical correlations and fluctuations of the disordered medium , and our exact results ( arising from a simple continuum model ) reveal the correlations to be non - trivial and the fluctuations to be extreme in nature . these statements are made quantitative by the form of the two - point correlation function , and the probability distribution of the density of corrupted bits . there are many directions for future work , foremost amongst which are : i ) calculating the distribution @xmath164 in @xmath196 and determining its sensitivity to the smearing function @xmath117 , ii ) investigating autocorrelation effects in one dimension , iii ) refining the model to take into account relevant factors ( like a back - coupling between the environment and the ba , or an asymmetry in the flipping probability ) which will arise if one tries to make a stronger connection to real processes . the model may also be seen to be a very simplified version of other systems . for instance the bistable medium can be taken to be composed of two chemical species a and b ( with vanishingly low mobility ) and the ba to be a high mobility catalyst , inducing a reversible reaction between a and b ( and _ vice versa _ ) . alternatively we can think of the ba as a wandering impurity in an ionic crystal ( such as an anion or cation vacancy in _ nacl _ ) or a semiconductor compound ( such as _ zn _ in _ gaas _ ) , which has a small probability of reordering the local bi - atomic structure as it passes through a given unit cell@xcite . the data corruption process appears to us the most interesting application , as well has being the most potentially relevant . this is especially true given the enormous efforts dedicated to creating memory storage devices of ever - decreasing size . such miniaturization will lead to new causes of soft error production@xcite , amongst which will inevitably be found the brownian agent . the authors would like to thank r. burghaus , m. howard , w. loinaz , a. mckane , b. schmittmann and r. zia for useful discussions . the authors also gratefully acknowledge financial support from the division of materials research of the national science foundation . in this appendix we outline the solution to the discrete equation ( [ lomageq ] ) for the conditional average @xmath41 . this is achieved by use of discrete fourier and laplace transforms defined via @xmath263 = \sum \limits _ { n=0}^{\infty } z^{n } \ \sum \limits _ { \bf r } \theta ( { \bf r},n \delta t\ | \ { \bf r } ) \exp ( i{\bf k } \cdot { \bf r } ) \ , \ ] ] ( with @xmath264 ) . by self - consistently determining the function @xmath265 , one can explicitly solve for the double transform of @xmath266 . inverting the discrete fourier transform and summing over @xmath14 one finds ( with the particular choice of initial condition ( [ ict ] ) ) @xmath267 = { 1 \over ( 1-z ) } \left \lbrace { 1 - 2(1-z ) \int \limits _ { \rm bz } d^{d}k \ e({\bf k},z ) \over 1 + { zq \over d } \int \limits _ { \rm bz } d^{d}k \ e({\bf k},z ) \ { \bf l } e^{-i{\bf k}\cdot { \bf l } } } \right \rbrace \ , \ ] ] where @xmath268 indicates integration over the first briulloin zone , and @xmath269 \right \rbrace ^{-1 } \ .\ ] ] similarly one can find an explicit expression for @xmath270 $ ] . the expressions can be shown to be equal thus proving the relation ( [ iden ] ) as given in the main text . in this appendix we outine the procedure for averaging the expression in eq.([defchi ] ) . first , we time order the integrals , thereby absorbing the factor of @xmath271 . then , making an integral representation of each dirac @xmath37-function , we have @xmath272 \bigr \rangle \ .\end{aligned}\ ] ] using the solution of eq.([scacont1 ] ) , we may rewrite the average in ( [ app2_1 ] ) as @xmath273 \right \rangle \ .\ ] ] the gaussian average over the noise may now be performed , and the above expression reduces to @xmath274 \right \rbrace \ .\ ] ] we now insert this expression back into eq.([app2_1 ] ) , and make the change of variables @xmath275 . the integrals over @xmath276 are easily performed and we arrive at eq.([solchi ] ) in the main text . in this appendix we outline the evaluation of the @xmath126-fold integral for @xmath277 given in eq.([corrcn ] ) . we note first that the time integrals have the form of a @xmath93-fold convolution . thus , we can laplace transform the expression to find @xmath278 } \ .\end{aligned}\ ] ] we make the change of variable @xmath275 , and rearrange the integrals to give @xmath279^{-1 } \cdots \int { dk_{n}'\over 2\pi } \ [ s+(d/2){k_{n}'}^{2}]^{-1 } \\ & & \ \ \ \ \ \ \ \ \ \ \ \ \ \times \prod \limits _ { m=1}^{n } \bigl [ 1 + \exp [ -i ( k_{m}'-k_{m-1}')x ] \bigr ] \ , \end{aligned}\ ] ] with the understanding that @xmath280 . we now multiply out the product which gives us @xmath281 terms . these are grouped into @xmath282 sets , the @xmath283 set containing @xmath284 terms which are equal after integration . thus each term in the @xmath283 set ( within the integrals ) may be taken to contain @xmath11 factors of @xmath285 ( where @xmath286 ) and @xmath287 factors of unity . using the integral @xmath288 } = { 1 \over ( 2sd)^{1/2 } } \exp \left [ - \left ( { 2s\over d } \right ) ^{1/2}|x| \right ] \ , \ ] ] we can reduce eq.([app3_2 ] ) to the form @xmath289 ^{n } \exp \left [ - m\left ( { 2s\over d } \right ) ^{1/2}|x| \right ] \ .\ ] ] performing the binomial sum gives eq.([ltcorrn ] ) in the main text . in this appendix we outline the derivation of eq.([ltprob ] ) from eqs.([deltaex ] ) and ( [ ltdengen ] ) . the only non - trivial aspect of the derivation is the frequency integral and sum ( over @xmath93 ) of the second term in eq.([ltdengen ] ) . ignoring the prefactor of that term ( namely @xmath290 ) we must evaluate a quantity @xmath291 , which has the form @xmath292 in order to perform the sum , we introduce the integral representation @xmath293 \ .\ ] ] the sum over @xmath93 now reconstitutes an exponential function , and we have @xmath294 the integral over @xmath172 is easily done to give @xmath295 finally , changing variables to @xmath296 we have @xmath297 \ .\ ] ] using this result for @xmath298 in conjunction with eqs.([deltaex ] ) and ( [ ltdengen ] ) we have ( [ ltprob ] ) in the main text . for more diverse examples see i. s. herschberg and r. paans , informatie , * 28 * , 10 ( 1986 ) ; f. b. cohen , in _ information security - the next decade _ , eds . j. h. p. eloff and s. h. von solms ( chapman and hall , london , 1995 ) .

from extensive molecular dynamics simulations on immiscible two - phase flows , we find the relative slipping between the fluids and the solid wall everywhere to follow the generalized navier boundary condition , in which the amount of slipping is proportional to the sum of tangential viscous stress and the uncompensated young stress . the latter arises from the deviation of the fluid - fluid interface from its static configuration . we give a continuum formulation of the immiscible flow hydrodynamics , comprising the generalized navier boundary condition , the navier - stokes equation , and the cahn - hilliard interfacial free energy . our hydrodynamic model yields interfacial and velocity profiles matching those from the molecular dynamics simulations at the molecular - scale vicinity of the contact line . in particular , the behavior at high capillary numbers , leading to the breakup of the fluid - fluid interface , is accurately predicted . * molecular scale contact line hydrodynamics + of immiscible flows *

cond-mat0210534

immiscible two - phase flow in the vicinity of the contact line ( cl ) , where the fluid - fluid interface intersects the solid wall , is a classical problem that falls beyond the framework of conventional hydrodynamics @xcite . in particular , molecular dynamics ( md ) studies have shown relative slipping between the fluids and the wall , in violation of the no - slip boundary condition @xcite . there have been numerous ad - hoc models @xcite to address this phenomenon , but none was able to give a quantitative account of the md slip velocity profile in the molecular - scale vicinity of the cl . while away from the moving cl the small amount of relative slipping was found to follow the navier boundary condition ( nbc ) @xcite , i.e. , relative slipping proportional to the tangential viscous stress , in the molecular - scale vicinity of the cl the nbc failed totally to account for the near - complete slip . this failure casts doubts on the general applicability of the nbc to immiscible flows and hinders a continuum formulation of the hydrodynamics in the cl region . in particular , a ( possible ) breakdown in the hydrodynamic description for the molecular - scale cl region has been suggested @xcite . in another approach @xcite , it was shown the md results can be reproduced by continuum finite element simulations , provided the slip profile extracted from md is used as input . this work demonstrated the feasibility of the hybrid algorithm , but left unresolved the problem concerning the boundary condition governing the cl motion . without a continuum hydrodynamic formulation , it becomes difficult or impossible to have realistic simulations of micro- or nanofluidics , or of immiscible flows in porous media where the relative wetting characteristics , the moving cl dissipation , and behavior over undulating solid surfaces may have macroscopic implications . from md simulations on immiscible two - phase flows , we report the finding that the generalized navier boundary condition ( gnbc ) applies for all boundary regions , whereby the relative slipping is proportional to the sum of tangential viscous stress and the uncompensated young stress . the latter arises from the deviation of the fluid - fluid interface from its static configuration @xcite . by combining gnbc with the cahn - hilliard ( ch ) hydrodynamic formulation of two - phase flow @xcite , we obtained a consistent , continuum description of immiscible flow with material parameters ( such as viscosity , interfacial tension , etc ) directly obtainable from md simulations . the convective - diffusive dynamics in the vicinity of the interface and the moving cl also means the introduction of two phenomenological dynamic parameters whose values can be fixed by comparison with one md flow profile . once the parameter values are determined from md simulations , our continuum hydrodynamics can yield predictions matching those from md simulations ( for different couette and poiseuille flows ) . our findings suggest the no - slip boundary condition to be an approximation to the gnbc , accurate for most macroscopic flows but failing in immiscible flows . these results open the door to efficient simulations of nano- or microfluidics involving immiscible components , as well as to macroscopic immiscible flow calculations , e.g. , in porous media , that are physically meaningful at the molecular level @xcite . the latter is possible , for example , by employing the adaptive method based on the iterative grid redistribution introduced in ref . this method has demonstrated the capability of resolving , at the same time , both the global behavior of a partial differential equation solution with coarse mesh , and a strong singularity in a localized region with a refined local mesh of over @xmath0 ratio to the coarse mesh . the md simulations were performed for both the static and dynamic configurations in couette and poiseuille flows . the two immiscible fluids were confined between two parallel walls separated along the @xmath1 direction , with the fluid - solid boundaries defined by @xmath2 ( see fig . [ fig1 ] for couette geometry ) . interaction between the fluid molecules was modeled by a modified lennard - jones ( lj ) potential @xmath3 $ ] , where @xmath4 is the distance between the molecules , @xmath5 and @xmath6 are the energy scale and range of interaction , respectively , and @xmath7 for like molecules and @xmath8 for molecules of different species . each of the two walls was constructed by two ( or more ) [ 001 ] planes of an fcc lattice ( see appendix [ app - wall - fluid ] ) , with each wall molecule attached to the lattice site by a harmonic spring . the mean - squared displacement of wall molecules was controlled to obey the lindemann criterion . the wall - fluid interaction was also modeled by a lj potential @xmath9 , with energy and range parameters @xmath10 and @xmath11 , and a @xmath12 for specifying the wetting property of the fluid . both @xmath13 and @xmath9 were cut off at @xmath14 . the mass of the wall molecule was set equal to that of the fluid molecule @xmath15 , and the average number densities for the fluids and wall were set at @xmath16 and @xmath17 , respectively . the temperature was controlled at @xmath18 , where @xmath19 is boltzmann s constant . moving the top and bottom walls at a constant speed @xmath20 in the @xmath21 directions , respectively , induced the couette flow @xcite . applying a body force @xmath22 to each fluid molecule in the @xmath23 direction induced the poiseuille flow @xcite . periodic boundary conditions were imposed on the @xmath23 and @xmath24 boundaries of the sample . most of our md simulations were carried out on samples consisting 6144 atoms for each fluid and 2880 atoms for each wall . the sample is @xmath25 by @xmath26 along the @xmath23 and @xmath24 , respectively , and @xmath27 . our md results represent time averages over 20 to 40 million time steps . for technical details of our md simulations , we followed those described in ref . @xcite . two different cases were considered in our simulations . the symmetric case refers to identical wall - fluid interactions for the two fluids ( both @xmath28 ) , which leads to a flat static interface in the @xmath29 plane with a @xmath30 contact angle . the asymmetric case refers to different wall - fluid interactions , with @xmath28 for one and @xmath31 for the other . the resulting static interface is a circular arc with a @xmath32 contact angle . we measured six quantities in the couette - flow steady states of @xmath33 , @xmath27 for the symmetric case and @xmath34 , @xmath27 for the asymmetric case : @xmath35 , the slip velocity relative to the moving wall ; @xmath36 , the tangential force per unit area exerted by the wall ; the @xmath37 , @xmath38 components of the fluid stress tensor ( @xmath39 denotes the outward surface normal ) , and @xmath40 , @xmath41 . we denote the region within @xmath42 of the wall the boundary layer ( bl ) . it must be thin enough to render sufficient precision for measuring @xmath35 , while thick enough to fully account for the tangential wall - fluid interaction force , due to the finite range of the lj interaction . thus it is not possible to do md measurements strictly at the fluid - solid boundary , not only because of poor statistics , but also because of this intrinsic limitation . the wall force can be identified by separating the force on each fluid molecule into wall - fluid and fluid - fluid components . for @xmath43 the fluid molecules can detect the atomic structure of the wall . when coupled with kinetic collisions with the wall molecules , there arises a nonzero tangential wall force that varies along the @xmath1 direction and saturates at @xmath44 . @xmath36 is the saturated total tangential wall force per unit wall area ( fig . [ fig2 ] ) . in appendix [ app - wall - fluid ] we give account of our md results on both the tangential and normal components of the wall force , plus the effect(s ) of increasing the wall thickness in our simulations from @xmath45 layers of wall molecules to @xmath46 layers and to infinite layers ( by using the continuum approximation beyond the @xmath46 layers ) . spatial resolution along the @xmath23 and @xmath1 directions was achieved by evenly dividing the sampling region into bins , each @xmath47 by @xmath48 in size . @xmath35 was obtained as the time average of fluid molecules velocities inside the bl , measured with respect to the moving wall ; @xmath36 was obtained from the time average of the total tangential wall force experienced by the fluid molecules in the bl , divided by the bin area in the @xmath49 plane ; @xmath50 was obtained from the time averages of the kinetic momentum transfer plus the fluid - fluid interaction forces across the constant-@xmath51 bin surfaces , and @xmath52 was measured as the time - averaged velocity component(s ) within each bin . for the contribution of intermolecular forces to the stress , we have directly measured the fluid - fluid interaction forces across bin surfaces instead of using the irving - kirkwood expression @xcite , whose validity was noted to be not justified at the fluid - fluid or the wall - fluid interface ( see the paragraph following equation ( 5.15 ) in the above reference ) . in appendix [ app - md - stress ] we give some details on our md stress measurements . as reference quantities , we also measured @xmath53 , @xmath54 , @xmath55 in the static ( @xmath56 ) configuration . in addition , we measured in both the static and dynamic configurations the average molecular densities @xmath57 and @xmath58 for the two fluid species in each bin to determine the interface profile . the shear viscosity @xmath59 and the interfacial tension @xmath60 were also determined . we have also measured the interface and velocity profiles for the poiseuille flow in the asymmetric case , as well as for couette flows with different @xmath20 and @xmath61 in the symmetric case . in the presence of a fluid - fluid interface , the static fluid stress tensor @xmath62 contains static young stress ( surface tension ) as well as those stresses arising from wall - fluid interaction . for the consideration of moving cl , we will be concerned with the part of the fluid stress tensor which is purely dynamic in origin , i.e. , arising purely from the hydrodynamic motion of the fluid ( and the cl ) . in the notations below , the over tilde will denote the quantity to be the difference between that quantity and its static part . thus if @xmath63 is the total stress , we will be concerned only with the hydrodynamic part , denoted by @xmath64 . we note that in the absence of body forces , force equilibrium in bulk fluid is governed by the relation @xmath65 . as shown in fig . [ fig2 ] , this relation is altered in the bl , where the fluid - wall interaction means the existence of a dynamic , tangential force density @xmath66 such that @xmath67 inside the bl . the total tangential force exerted by the wall on the fluid is given by @xmath68 per unit wall area . in steady state , this wall force is necessarily balanced by the tangential fluid force @xmath69 . here @xmath70 means taking partial derivative with respect to @xmath23 , @xmath1 , or surface normal . we now present evidences to show that everywhere on the boundaries , relative slipping is proportional to @xmath71 ( the gnbc , see also equation ( [ gnbc3 ] ) below ) : @xmath72 where @xmath73 is the slip coefficient and @xmath71 can be written as @xmath74 where we have used the fact that @xmath75 . ( more strictly , @xmath76 because there is no fluid below @xmath77 and hence no momentum transport across @xmath77 ) here the @xmath1 coordinate is for the lower fluid - solid boundary ( same below ) , with the understanding that the same physics holds at the upper boundary , and @xmath78 for the lower boundary . we have verified the steady state force balance @xmath79 on the two boundaries ( inset to fig . [ fig3 ] ) @xcite . this relation reflects the fact that at steady state , the frictional force exerted by the solid wall on the moving / slipping fluid is fully accounted for in @xmath71 . thus the gnbc ( or nbc ) can be expressed in either @xmath71 or @xmath80 , but not both . in fig . [ fig3 ] we show the measured md data for the symmetric and asymmetric cases in the couette geometry . the symbols represent the values of @xmath71 measured in the bl . the solid lines represent the values of @xmath71 calculated from @xmath81 by using @xmath82 for the symmetric case and @xmath83 , @xmath84 for the asymmetric case away from the cl region ( straight line segments in fig . [ fig3 ] ) , and @xmath85 in the cl region @xcite , with @xmath35 and @xmath86 obtained from md simulations . it is seen that for the lower boundary ( upper right panel ) , the md data agree well with the predictions of eq . ( [ gnbc1 ] ) . for the upper boundary ( lower left panel ) the straight line segments also agree well with eq . ( [ gnbc1 ] ) . however , there is some discrepancy in the interfacial region of the upper boundary that seems to arise from a `` shear thinning '' effect of decreasing @xmath73 at very large tangential stresses @xcite . the fact that the wall force density is distributed inside a thin bl and vanishes beyond the bl necessitates the form of @xmath71 as defined by eq . ( [ gnbc2 ] ) . however , it is intuitively obvious that the fluids would experience almost the identical physical effect(s ) from a wall force density @xmath87 , concentrated strictly at the fluid - solid boundary with the same total wall force per unit area . in the inset to fig . [ fig2 ] , it is shown that the md - measured wall force density is a sharply peaked function . the sharp boundary limit involves the approximation of replacing this peaked function by @xmath88 . the replacement of a diffuse boundary by a sharp boundary can considerably simplify the form of the gnbc , because local force balance along @xmath23 then requires @xmath89 away from the boundary @xmath77 . integration of this relation from @xmath90 to @xmath91 yields @xmath92 and as a consequence ( by comparing with eq . ( [ gnbc2 ] ) ) @xmath93 . therefore , @xmath94 changes from @xmath76 to @xmath95 at @xmath77 , leading to @xmath96 . comparing with the diffuse boundary , we see that the form of the equation remains the same , but the bl is now from @xmath97 to @xmath90 . thus the gnbc ( [ gnbc1 ] ) becomes @xmath98 in the sharp boundary limit . the tangential stress @xmath99 can be decomposed into a viscous component and a non - viscous component : @xmath100 . in fig . [ fig4 ] we show that away from the interfacial region the tangential viscous stress @xmath101 is the only nonzero component , but in the interfacial region @xmath102 is dominant , thereby accounting for the failure of nbc to describe the cl motion . therefore away from the cl region the nbc is valid , but in the interfacial region the nbc clearly fails to describe the cl motion . we wish to clarify the origin of @xmath103 and @xmath55 as the dynamic and static young stresses , respectively , so that @xmath104 is the uncompensated young stress . as shown in the inset to fig . [ fig4 ] , the integrals ( across the interface ) of @xmath103 ( @xmath105 , calculated by subtracting the viscous component @xmath106 from the total tangential stress @xmath38 ) and @xmath55 are equal to @xmath107 and @xmath108 , respectively , at different values of @xmath1 , i.e. , @xmath109 and @xmath110 , where @xmath111 and @xmath112 are respectively the dynamic and static interfacial angles at @xmath1 @xcite . here @xmath113 denotes the integration across the fluid - fluid interface along @xmath23 . these results clearly show the origin of the extra tangential stress in the interfacial region to be the interfacial ( uncompensated ) young stress . thus the gnbc is given by @xmath114(0 ) -\tilde{\sigma}_{nx}^y(0).\ ] ] here only one component of the viscous stress is nonzero , due to @xmath115 at the boundary ; and @xmath116 is denoted the uncompensated young stress , satisfying @xmath117 , with @xmath118 being a microscopic dynamic ( static ) contact angle at the fluid - solid boundary . the fact that @xmath119 away from the cl shows that the gnbc implies nbc for single phase flows . due to the diffuse nature of the bl in md simulations , the contact angle @xmath118 can not be directly measured . nevertheless , they are obtainable through extrapolation by using the integrated interfacial curvature within the bl . that is , in the sharp boundary limit the force balance in the fluids is expressed by @xmath120 . integration in @xmath1 across the bl gives @xmath121 integration ( of eq . ( [ angle1 ] ) along @xmath23 ) across the fluid - fluid interface then yields @xmath122 -\int_{\rm int}dx{\sigma}_{nx}^v(z_0)+\int_{\rm int}dx{\sigma}_{nx}^v(0 ) + \gamma{\cal k}_d-\gamma{\cal k}_s=0.\ ] ] where @xmath123 $ ] is the change of the @xmath1-integrated @xmath124 across the interface , @xmath125 and @xmath126 denote the dynamic and static @xmath1-integrated interfacial curvatures : @xmath127 and @xmath128 here @xmath123 $ ] , @xmath129 , @xmath130 , and @xmath131 are obtainable from md simulations , @xmath132 for the circular static interfaces , while @xmath133(0)$ ] may be obtained by extrapolating from the values of tangential viscous stress at @xmath134 , @xmath135 , and @xmath136 . therefore the microscopic dynamic contact angle @xmath137 can be obtained from eq . ( [ angle2 ] ) . in appendix [ md - continuum - correspondence ] we give a more detailed account of the relationship between the md measured stresses and the stress components in the continuum hydrodynamics . the above extrapolation is based on this correspondence . we have measured the @xmath1-integrated @xmath138 in the bl . the dominant behavior is a sharp drop across the interface , as shown in fig . [ fig5 ] for both the symmetric and asymmetric cases . the value of @xmath137 obtained is @xmath139 for the symmetric case and @xmath140 for the asymmetric case at the lower boundary , and @xmath141 at the upper boundary . these values are noted to be very close to @xmath142 . yet the small difference between the dynamic and static ( microscopic ) contact angles is essential in accounting for the near complete slip in the cl region . in essence , our results show that in the vicinity of the cl , the tangential viscous stress @xmath143 as postulated by the nbc can not give rise to the near - complete cl slip without taking into account the tangential young stress @xmath144 in combination with the gradient of the ( bl - integrated ) normal stress @xmath37 . for the static configuration , the normal stress gradient is balanced by the young stress , leading to the young s equation . it is only for a moving cl that there is a component of the young stress which is no longer balanced by the normal stress gradient , and this uncompensated young stress is precisely the additional component captured by the gnbc but missed by the nbc . ( [ gnbc3 ] ) to serve as a boundary condition in hydrodynamic calculations , we need to derive the local value of the uncompensated young stress @xmath145 from a continuum formulation of the immiscible flow hydrodynamics . such a formulation is important for studying the macroscopic implications of moving cl s under scenarios beyond the capability of md simulations . as a first - order approximation , we formulate a hydrodynamic model based on the gnbc and the ch free energy functional @xcite that has been successful in the calculations of fluid - fluid interfacial phenomena : @xmath146=\int d{\bf r } \left[\displaystyle\frac{1}{2}k\left(\nabla\phi\right)^2+f(\phi)\right],\ ] ] where @xmath147 , @xmath148 , and @xmath149 , @xmath4 , @xmath150 are parameters which can be directly obtained from md simulations through the interface profile thickness @xmath151 @xcite , the interfacial tension @xmath152 , and the two homogeneous equilibrium phases given by the condition of @xmath153 , yielding @xmath154 ( @xmath155 in our case ) . to derive the effects of the ch free energy @xmath156 on immiscible flow hydrodynamics , let us consider a composition field @xmath157 . a displacement of the molecules from @xmath158 to @xmath159 induces a local change of @xmath160 , @xmath161 , to the first order in @xmath162 . the associated change in @xmath156 is given by the sum of a body term and a surface term : @xmath163 + \int ds\left[\sigma_{ni}^yu_i\right],\ ] ] where @xmath164 is the capillary force density , with @xmath165 being the chemical potential , and @xmath166 is the tangential young stress due to the spatial variation of @xmath160 at the fluid - solid boundary ( @xmath167 ) . hence the two coupled equations of motion are the navier - stokes equation ( with the addition of the capillary force density ) and the convection - diffusion equation for @xmath157 : @xmath168= -\nabla p + \nabla\cdot{\mbox{\boldmath$\sigma$}}^v + \mu\nabla\phi+\rho_m{\bf g}_{ext},\ ] ] @xmath169 together with the incompressibility condition @xmath170 . here @xmath171 is the fluid mass density , @xmath172 is the pressure , @xmath173 denotes the viscous part of the stress tensor , @xmath174 is the external body force density ( for poiseuille flows ) , and @xmath175 is the phenomenological mobility coefficient . four boundary conditions are required to solve eqs . ( [ he1 ] ) and ( [ he2 ] ) . two are given by the impermeability condition , i.e. , the normal components of the fluid velocity and diffusive flux are zero : @xmath115 and @xmath176 . the form of the other two differential boundary conditions may be obtained from the total free energy @xmath177=f[\phi]+\int ds\gamma_{wf}(\phi),\ ] ] plus our knowledge of gnbc . here @xmath178 is the interfacial free energy per unit area at the fluid - solid boundary . we use @xmath179 to denote a smooth interpolation between @xmath180 , with @xmath181 given by @xmath182 ( young s equation ) . it should be noted that the form of the smooth interpolation has very little effect on the final results . hence we have chosen a simple interpolation function . similar to eq . ( [ ch - force ] ) , the change in @xmath183 due to the displacement of the molecules from @xmath158 to @xmath159 is given by @xmath184 + \int ds\left[\tilde{\sigma}_{ni}^yu_i\right],\ ] ] where @xmath185\partial_i\phi,\ ] ] is the uncompensated young stress @xcite ( see below ) . the continuum ( differential ) form of the gnbc ( [ gnbc3 ] ) is therefore given by @xmath186(0 ) + \left [ l(\phi)\partial_x\phi \right](0),\ ] ] where @xmath187(0)$ ] , with @xmath188 , is the differential expression for @xmath189 in equation ( [ gnbc3 ] ) . here @xmath190 is @xmath191 as seen in eq . ( [ young - stress ] ) , and @xmath192\partial_x\phi = \partial_x\gamma_{wf}(\phi)$ ] @xcite equals to @xmath193 , in accordance with the static force balance relation @xmath194 . from @xmath195(0 ) = \gamma\cos\theta_d^{surf}$ ] @xcite and @xmath196 , we see that @xmath197(0 ) = \gamma(\cos\theta_d^{surf}-\cos\theta_s^{surf}),\ ] ] in agreement with @xmath187(0)$ ] being the uncompensated young stress . another boundary condition may be inferred from the fact that @xmath198 is the euler - lagrange equation at the fluid - solid boundary for minimizing the total free energy @xmath199 $ ] . that is , @xmath198 corresponds with the equilibrium ( static ) condition where @xmath200 . the boundary relaxation dynamics of @xmath160 is plausibly assumed to be the first - order extension of that correspondence for a nonzero @xmath201 : @xmath202,\ ] ] where @xmath203 is a ( positive ) phenomenological parameter . motivated by the methods presented in @xcite , a second order scheme is designed to solve the ch hydrodynamic model , comprising the dynamic equations and the four boundary conditions . details of the numerical algorithm are presented in appendix [ app - numerical ] . besides those parameters which can be directly obtained from md simulations , @xmath175 and @xmath203 are treated as fitting parameters , determined by comparison with md results ( values given in the caption to fig . [ fig6 ] ) . in figs . [ fig1 ] and [ fig6 ] we show that the continuum model can indeed quantitatively reproduce the interface and velocity profiles from md simulations , including the near - complete slip ( @xmath204 in fig . [ fig6 ] ) of the cl , the fine features in the molecular - scale vicinity of the cl , and the fast pressure variation in the bl ( inset to fig . [ fig6 ] ) , with its implied large interfacial curvature . we wish to emphasize that for the comparison with the symmetric case , the parameters in the continuum model , including those in the gnbc , are directly obtained from the md simulations , whose velocity profiles are then fitted by those from the hydrodynamic calculations with optimized @xmath175 and @xmath203 values . thus the comparison with the asymmetric ( couette ) case , with @xmath205 directly evaluated from md simulation data , is _ without adjustable parameters_. we have also obtained @xmath206 and @xmath207 for the symmetric and asymmetric ( the lower boundary ) cases shown in figs . [ fig1]a and [ fig1]b , respectively . both are in excellent agreement with their extrapolated values in md simulations . for the upper boundary in the asymmetric case , our calculated @xmath208 , which differs somewhat from the md extrapolation value of @xmath141 . this difference is a reflection of the discrepancy seen in fig . however , it is noteworthy that the difference in the dynamic contact angles does not show up in the velocity profiles , which agree well . to further verify that the boundary conditions and the parameter values are local properties and hence applicable to flows with different macroscopic conditions , we have varied the wall speed @xmath20 , the system size @xmath61 , and the flow geometry to check that the same set of parameters plus the gnbc are valid for reproducing ( a ) the velocity profiles from a different set of couette - flow simulations in the symmetric configurations , shown in fig . [ fig7 ] , as well as ( b ) the velocity profiles of the poiseuille flow simulations in the asymmetric case , shown in fig . the remarkable overall agreement in all cases affirms the validity of gnbc and the hydrodynamic model @xcite , as well as justifies the replacement of the diffuse fluid - solid boundary ( force density ) by a sharp boundary . another comparison is the dissipation incurred by the moving cl in the couette flow geometry . rate of heat generation per unit wall area is given by @xmath209 . from this we have to subtract a small but constant relative slipping away from the interface , @xmath210 , where @xmath211 is a slip length for fluid 1 if @xmath212 and for fluid 2 if @xmath213 . the resulting heat generation rate due to the cl is @xmath214 ( for one wall ) , where @xmath215 is the length of the cl and @xmath216 defines the width of the cl region : @xmath217 thus cl dissipation is equivalent to a segment , @xmath218 , of dissipation by single phase flow . figure [ fig9 ] shows the variation of @xmath216 as a function of capillary number @xmath219 for the symmetric case of couette flow . close to @xmath220 the value of @xmath216 increases rapidly , in good agreement with the md results , and beyond which the continuum model failed to converge . this corresponds to the breakup of the interface observed in md simulations @xcite . in summary , we have found for the first time the boundary condition that yields near - complete slipping of the cl , in good agreement with md results on the molecular scale . it should also be noted , however , that the present continuum formulation can not calculate fluctuation effects that are important in md simulations . long range interactions , e.g. , that due to van der waals interaction , have also been ignored . the latter is potentially important in the calculations involving wetting layers . partial support from hkust s ehia funding and hong kong rgc grants hkust 6176/99p and 6143/01p is hereby acknowledged . we have measured both the tangential and normal components of the wall force exerted on the fluids . both components vary along the @xmath1 direction and saturate somewhere away from the fluid - solid boundary . the tangential component saturates ( by @xmath221 ) at @xmath134 , which is well inside the wall - fluid interaction range ( @xmath222 , smaller than the cut - off distance @xmath14 for the wall - fluid interaction potential @xmath9 ) . on the other hand , the normal component is @xmath223 of the saturation value at @xmath134 and @xmath221 at @xmath224 . the different saturation ranges of the tangential and normal components may be understood as follows . for a fluid molecule close to the solid wall , the interaction with one particular ( the closest ) wall molecule can be much stronger than all the others . as this fluid molecule moves laterally but still remaining its close proximity to the wall , it would thus experience a strong periodic modulation in its interaction with the wall . this lateral inhomogeneity is an important source for the tangential component of the wall force . away from the fluid - solid boundary , each fluid molecule can interact with many wall molecules on a nearly equal basis . thus the modulation amplitude of the wall potential would clearly decrease with increasing distance from the wall . hence the tangential wall force tends to saturate at the relatively short range of @xmath44 . on the contrary , the normal wall force directly arises from the wall - fluid interaction , independent of whether the wall potential is `` rough '' or not . consequently , the normal wall force saturates much slower than the tangential component . the md results presented in this paper were obtained from simulations using solid walls constructed by two [ 001 ] planes of an fcc lattice . we have also carried out md simulations using thicker confining walls . first we changed the number of molecular layers ( [ 001 ] planes of fcc lattice ) from two to four in constructing each of the two walls . the wall - fluid interaction potential @xmath9 were still cut off at @xmath14 . it turned out that neither component of the wall force shows any noticeable change . the reason is that for the tangential component , the two outer planes are too distant to contribute to the roughness of the wall potential , while for the normal component , the fluid molecules closest to the wall are separated from the two outer planes by a distance @xmath225 . consequently , both the interface and velocity profiles do not show any noticeable change . additional wall layers do not contribute to the perceived modulation of the wall potential by the fluid molecules . nevertheless , they can still affect the tangential wall force by modifying the organization of the fluid molecules near the wall . such organization is governed by the wall - fluid interaction and can be greatly influenced by the normal wall force . to see the effects of normal wall force due to additional wall layers , we used four [ 001 ] planes of an fcc lattice plus a half - space continuum in constructing a wall . the first four solid layers show the atomic structure detectable by the fluid molecules while the half - space continuum models the deeper solid layers . the wall - fluid interaction was modelled as follows . for an in - range pair of fluid and wall molecules separated by a distance @xmath226 , the interaction potential is still @xmath9 . here the wall molecule must be from one of the four solid layers . in addition to this short - range interaction , the fluid molecules can also experience the long - range interaction potential due to ( 1 ) the distant wall molecules in the four solid layers and ( 2 ) the continuum . for ( 1 ) we integrated the @xmath227 term in @xmath9 over the out - of - range ( @xmath228 ) area of the solid layers while for ( 2 ) we integrated the same term over the half - space continuum . according to this model , only the in - range ( @xmath226 ) part of the solid wall shows atomic structure to a fluid molecule while the out - of - range ( @xmath228 ) part is effectively a half - space continuum . we found that the effect of the long - range normal wall force ( for @xmath229 ) is to attract the fluid molecules to the wall . in fact the average number density in the bl can increase by @xmath230 once the long - range force is included . as a result , the slip coefficient @xmath231 increases by @xmath232 . this results in small but visible changes in the interface and velocity profiles . these tests have convinced us that by using two [ 001 ] planes of an fcc lattice to model the solid wall , we have captured the dominant wall - fluid interaction . in fact , using two molecular layers to model the solid wall has been extensively practiced in the past md simulations @xcite , although in some instances more molecular layers have also been used @xcite , where the accurate modeling of the normal component of the wall - fluid interaction force is important . irving and kirkwood @xcite have shown that in the hydrodynamic equation of motion ( momentum transport ) , the stress tensor ( flux of momentum ) may be expressed in terms of molecular variables as @xmath233 where @xmath234 is the kinetic contribution to the stress tensor , given by @xmath235 \left[\displaystyle\frac{{\bf p}_i}{m_i}-{\bf v}({\bf r},t)\right ] \delta({\bf x}_i-{\bf r}),f\right\rangle,\ ] ] and @xmath236 is the contribution of intermolecular forces to the stress tensor , given by @xmath237 here @xmath238 , @xmath239 , and @xmath240 are respectively the mass , momentum , and position of molecule @xmath241 , @xmath242 is the local average velocity , @xmath243 is the force on molecule @xmath241 due to molecule @xmath244 , @xmath245 is the probability distribution function @xmath246 which satisfies the normalization condition @xmath247 and the liouville equation @xmath248,\ ] ] with @xmath249 being the potential energy of the system , and @xmath250 means taking the average for a probability distribution function @xmath245 . although widely employed in the stress measurements in md simulations , the above expression for @xmath236 ( eq . ( [ u - stress ] ) ) represents only the leading term in an asymptotic expansion , accurate when the interaction range is small compared to the range of hydrodynamic variation @xcite . this can be seen as follows . consider that all the molecules interact via a pair potential @xmath251 such that the intermolecular force @xmath252 for @xmath253 . accordingly , eq . ( [ u - stress ] ) can be rewritten as @xmath254 where @xmath255 is the pair density defined by @xmath256 it has been shown ( see the appendix in ref . @xcite ) that according to the definition that @xmath257 is the force acting across @xmath258 , the full expression for @xmath236 is given by @xmath259.\ ] ] it is readily seen that eq . ( [ u - stress-1 ] ) may be obtained from eq . ( [ u - stress-2 ] ) by keeping only the lowest order term in a taylor s series in @xmath260 , i.e. , @xmath261 that means @xmath262 must be negligible compared with @xmath263 . here @xmath264 is on the order of the range of intermolecular force . this approximation , however , can not be justified at the fluid - fluid or the wall - fluid interface , where @xmath262 can be comparable in magnitude to @xmath255 . in the study of moving cl , it is of great importance to obtain the correct information about stress distributions at both the fluid - fluid and the wall - fluid interfaces . therefore , we have directly measured the @xmath23 component of fluid - fluid interaction forces acting across the @xmath51 bin surfaces , in order to obtain the @xmath265 component of @xmath236 . for example , in measuring @xmath266 at a given @xmath1-direction bin surface , we recorded all the pairs of molecules interacting across that surface . here `` interacting across '' means that the line connecting a pair of molecules intersects the bin surface . for those pairs , we then computed @xmath266 at the given bin surface using @xmath267 where @xmath268 is the area of @xmath1-direction bin surface , @xmath269 indicate all possible pairs of molecules interacting across the bin surface , with molecule @xmath241 being `` inside of @xmath270 '' and molecule @xmath244 being `` outside of @xmath270 '' ( molecule @xmath241 is below molecule @xmath244 ) , and @xmath271 is the @xmath23 component of the force on molecule @xmath241 due to molecule @xmath244 . a schematic illustration is shown in fig . [ fig - stress ] . for comparison , we have measured the @xmath272 and @xmath273 components of @xmath236 using the discrete version of irving - kirkwood expression ( [ u - stress ] ) : @xmath274 where @xmath275 is the volume of sampling bin , @xmath241 runs over fluid molecules in the sampling bin , @xmath244 runs over fluid molecules in interaction with molecule @xmath241 , and @xmath276 means taking the time average . we found that far from the the fluid - fluid and the wall - fluid interfaces , the results based on the irving - kirkwood expression agree well with those from direct force measurement , whereas near the fluid - fluid or the wall - fluid interface , the two results show appreciable differences ( up to @xmath277 ) , especially for the @xmath273 component at the fluid - fluid interface . we want to note the correspondence between the md - measured stresses and the continuum hydrodynamic stress components . this correspondence is essential to obtaining the microscopic contact angle @xmath137 , defined in the continuum hydrodynamic model but not directly measurable in md simulations . the gnbc for the diffuse bl is given by @xmath278 + \left[\sigma_{zx}(z_0)-\sigma_{zx}^0(z_0)\right],\ ] ] which involves only md measurable quantities . to obtain the contact angle @xmath137 from md results , we need to interpret the md - measured quantities in terms of the various continuum variables in the hydrodynamic model . in so doing it is essential to note the following . \(1 ) @xmath37 can be decomposed into a molecular component and a hydrodynamic component : @xmath279 . meanwhile , @xmath54 can be composed into the same molecular component and a hydrostatic component : @xmath280 . the molecular component @xmath281 exists even if there is no hydrodynamic fluid motion or fluid - fluid interfacial curvature . in particular , @xmath281 in the bl depends on the wall - fluid interactions . the change of the bl - integrated @xmath281 across the fluid - fluid interface equals the change in the wall - fluid interfacial free energy , i.e. , @xmath282=\delta\gamma_{wf } = \gamma_{wf}(\phi_+)-\gamma_{wf}(\phi_-)$ ] . on the other hand , the hydrodynamic component @xmath283 in @xmath37 results from the hydrodynamic fluid motion and fluid - fluid interfacial curvature . in the static ( @xmath56 or @xmath284 ) configuration , @xmath283 becomes the hydrostatic component @xmath285 in @xmath54 . \(2 ) @xmath286 can be decomposed into a viscous component plus a young component : @xmath287 with @xmath288 and @xmath289 . \(3 ) @xmath290 is the static young stress : i.e. , @xmath291 . with the help of the above relations , integration of eq . ( [ extrap - gnbc ] ) across the fluid - fluid interface yields @xmath292 + \int_{\rm int}dx\sigma_{zx}^v(z_0)+\gamma\cos\theta_d(z_0 ) -\delta\left[\int_0^{z_0}dz\sigma_{xx}^{hs}\right ] -\gamma\cos\theta_s(z_0),\ ] ] where @xmath293 $ ] is the change of the @xmath1-integrated @xmath294 across the interface . according to the laplace s equation , the change of the hydrostatic @xmath1-integrated normal stress is directly related to the static @xmath1-integrated curvature @xmath126 : @xmath295.\ ] ] note that @xmath126 vanishes in the symmetric case . substituting eq . ( [ extrap - static ] ) into eq . ( [ extrap - gnbc - int1 ] ) then yields @xmath296 if interpreted in the continuum hydrodynamic formulation with a sharp fluid - solid boundary , the last term in the right - hand side of eq . ( [ extrap - gnbc - int2 ] ) , @xmath182 , is the net wall force along @xmath23 arising from the wall - fluid interfacial free energy jump across the fluid - fluid interface , in accordance with the young s equation @xmath297 . on the other hand , the sum of the first three terms on the right - hand side of eq . ( [ extrap - gnbc - int2 ] ) is the net fluid force along @xmath23 exerted on the three fluid sides of a bl fluid element in the interfacial region , due to the hydrodynamic motion of the fluids . to obtain an extrapolated value for the contact angle @xmath137 from eq . ( [ extrap - gnbc - int2 ] ) , we turn to the stokes equation in the bl : @xmath298 obtained from the @xmath23-component of eq . ( [ he1 ] ) by dropping the inertial and external forces . integration in @xmath1 of eq . ( [ extrap - stokes ] ) across the bl , together with the integration along @xmath23 across the fluid - fluid interface yields @xmath299 + \int_{\rm int}dx\sigma_{zx}^v(z_0)+\gamma\cos\theta_d(z_0 ) -\int_{\rm int}dx\sigma_{zx}^v(0)-\gamma\cos\theta_d^{surf}=0.\ ] ] here we have made use of two relations : ( 1 ) @xmath300 in the sharp interface limit @xcite , with @xmath301 being the interfacial curvature and @xmath302 the location of the interface along @xmath23 . ( 2 ) @xmath303 is the dynamic @xmath1-integrated curvature @xmath304 . the local force balance along @xmath23 is expressed by eq . ( [ extrap - stokes ] ) . accordingly , the force balance along @xmath23 for the bl fluids in the integration region is expressed by eq . ( [ extrap - stokes - int ] ) , where @xmath305 $ ] is the net force on the left and right ( constant-@xmath23 ) surfaces , @xmath306 is the tangential force on the @xmath134 surface , and @xmath307 is the tangential force on the @xmath77 surface . substituting eq . ( [ extrap - stokes - int ] ) into eq . ( [ extrap - gnbc - int2 ] ) and identifying the normal stress @xmath308 with @xmath283 , we obtain @xmath309 which is identical to the integration of the continuum gnbc ( eq . ( [ he3 ] ) ) along @xmath23 across the fluid - fluid interface . in summary , to obtain eq . ( [ extrap - gnbc - int3 ] ) from eq . ( [ extrap - gnbc - int1 ] ) , we have used both @xmath310 and @xmath311 , whose integrated expressions are given by eq . ( [ extrap - static ] ) and eq . ( [ extrap - stokes - int ] ) , respectively . we note that @xmath312 is equivalent to the relation @xmath120 ( integrated expressions given by eqs . ( [ angle1 ] ) and ( [ angle2 ] ) ) , which has been used to obtain @xmath137 through extrapolation in sec . [ md - evidence ] . we present our numerical algorithm for solving the continuum hydrodynamic model , comprising the dynamic equations ( [ he1 ] ) and ( [ he2 ] ) and the four boundary conditions @xmath115 , @xmath313 , plus eqs . ( [ he3 ] ) and ( [ he4 ] ) . we pay special attention to the application of boundary conditions , and restrict our analysis to the couette flow because the generalization to poiseuille flow is straightforward . to obtain a set of dimensionless equations suitable for numerical computations , we scale @xmath160 by @xmath314 , length by @xmath315 , velocity by the wall speed @xmath20 , time by @xmath316 , and pressure / stress by @xmath317 . in dimensionless forms , the convection - diffusion equation reads @xmath318 the navier - stokes equation reads @xmath319=-\nabla p+ \nabla^2{\bf v}+{\cal b}(-\nabla^2\phi-\phi+\phi^3 ) \nabla\phi,\ ] ] the relaxation of @xmath160 at the fluid - solid boundary is governed by @xmath320,\ ] ] and the gnbc becomes @xmath321^{-1 } v^{slip}_x={\cal b } \left[\partial_n\phi-\displaystyle\frac{\sqrt{2}}{3}\cos\theta_s^0 s_\gamma(\phi)\right]\partial_x\phi-\partial_nv_x.\ ] ] here @xmath322 and @xmath323 . five dimensionless parameters appear in the above equations . they are ( 1 ) @xmath324 , which is the ratio of a diffusion length @xmath325 to @xmath326 , ( 2 ) @xmath327 , ( 3 ) @xmath328 , which is inversely proportional to the capillary number @xmath219 , ( 4 ) @xmath329 , which is the ratio of @xmath330 ( of velocity dimension ) to @xmath20 , and ( 5 ) @xmath331 , which is the ratio of the slip length @xmath332 to @xmath326 . for immiscible couette flows , there are four variables @xmath160 , @xmath40 , @xmath41 , and @xmath172 to be solved in a two - dimensional ( 2d ) system ( in the @xmath333 plane ) . we want to solve the convection - diffusion equation and the navier - stokes equation in a 2d system of length @xmath334 ( along @xmath23 ) and height @xmath335 ( along @xmath1 ) . here @xmath334 must be large enough to allow the single phase flows ( far from the fluid - fluid interface ) to approach uniform shear flows . a finite - difference scheme is employed as follows . ( 1 ) @xmath336 and @xmath337 equally spaced levels are introduced in the @xmath23 and @xmath1 directions , respectively . grid size is given by @xmath338 and @xmath339 along @xmath23 and @xmath1 , respectively . ( 2 ) each variable ( @xmath340 ) is defined at @xmath341 sites distributed from @xmath342 to @xmath343 and from @xmath344 to @xmath345 , represented by the array @xmath346 , with @xmath347 and @xmath348 . here @xmath349 , with @xmath350 and @xmath351 . ( 3 ) in applying the various boundary conditions , `` ghost '' sites outside the system , i.e. , @xmath352 , @xmath353 , @xmath354 , or @xmath355 , may appear in the discretization scheme . the values of the variables at the ghost sites are determined separately from the various boundary conditions , detailed below . ( 4 ) first and second spatial derivatives along @xmath356 ( = @xmath23 or @xmath1 ) are represented by @xmath357/{2\delta_\zeta}$ ] and @xmath358 /{\delta_\zeta^2}$ ] . with the chemical potential @xmath359 given by @xmath360-\phi_{i , j}+\phi_{i , j}^3,\ ] ] the discretized convection - diffusion equation is @xmath361_{i , j}={\cal l}_d \left[\displaystyle\frac{\mu_{i+1,j}-2\mu_{i , j}+\mu_{i-1,j}}{\delta_x^2}+ \displaystyle\frac{\mu_{i , j+1}-2\mu_{i , j}+\mu_{i , j-1}}{\delta_z^2}\right],\ ] ] with @xmath362_{i , j}= v_{xi , j}\displaystyle\frac{\phi_{i+1,j}-\phi_{i-1,j}}{2\delta_x}+ v_{zi , j}\displaystyle\frac{\phi_{i , j+1}-\phi_{i , j-1}}{2\delta_z}.\ ] ] the boundary conditions at @xmath363 can be easily applied using @xmath364 and @xmath365 for single - phase uniform shear flows . here we focus on the boundary conditions at @xmath366 : @xmath313 and eq . ( [ dmlrelax ] ) . we spell out the numerics for the lower boundary @xmath367 , with the understanding that the same can be applied to the upper boundary . to solve the discretized convection - diffusion equation ( [ disccde ] ) at the lower boundary @xmath367 , we need the values of @xmath359 at @xmath367 and @xmath354 . we also need the values of @xmath359 at @xmath367 to solve the same equation at @xmath368 . according to eq . ( [ discmu ] ) , @xmath359 at @xmath367 and @xmath354 can not be directly evaluated from @xmath369 with @xmath347 and @xmath348 . but they can still be determined from the boundary conditions at @xmath370 . @xmath359 at @xmath354 is obtained from @xmath313 at @xmath367 as @xmath371 to obtain @xmath359 at @xmath367 , we need to determine @xmath369 at @xmath354 . this can be done by requiring that eqs . ( [ dmlcde ] ) and ( [ dmlrelax ] ) yield the same @xmath372 at @xmath370 . the discretized convection - diffusion equation is given by eq . ( [ disccde ] ) while the discretized relaxation equation for @xmath160 at the boundary @xmath367 is given by @xmath373_{i , j } = -{\cal v}_s\left [ \displaystyle\frac{\phi_{i , j-1}-\phi_{i , j+1}}{2\delta_z}- \displaystyle\frac{\sqrt{2}}{3}\cos\theta_s^0 s_\gamma(\phi_{i , j } ) \right].\ ] ] equating the right - hand side of eq . ( [ disccde ] ) at @xmath367 ( with @xmath374 fixed by eq . ( [ cdebc1 ] ) and other @xmath375 s given by eq . ( [ discmu ] ) ) with that of eq . ( [ cdebc2 ] ) leads to a tridiagonal system of linear equations for @xmath369 ( @xmath369 coupled with @xmath376 and @xmath377 ) at @xmath354 . solving this tridiagonal system determines @xmath369 at @xmath354 , from which we obtain @xmath359 at @xmath367 by using eq . ( [ discmu ] ) . we now turn to the navier - stokes equation ( [ dmlnse ] ) with the incompressibility condition @xmath378 . the difficulty in solving the navier - stokes equation is the lack of a time evolution equation for the pressure @xmath172 . in the following , we will introduce a numerical method based on the pressure poisson equation @xcite . taking the divergence of momentum equation ( [ dmlnse ] ) and applying the incompressibility condition , we obtain the pressure poisson equation @xmath379 + { \cal b}\nabla\cdot[(-\nabla^2\phi-\phi+\phi^3)\nabla\phi].\ ] ] dotting the momentum equation ( [ dmlnse ] ) with the surface normal at the fluid - solid boundary and using @xmath115 , we obtain for eq . ( [ ppe ] ) the boundary condition @xmath380 at @xmath366 . in addition , we use @xmath381 and @xmath382 for the values of @xmath383 and @xmath384 at the boundaries @xmath363 . this reflects the single phase flow given by eq . ( [ single - couette ] ) . from the momentum equation ( [ dmlnse ] ) and the pressure poisson equation ( [ ppe ] ) , we derive a diffusion equation @xmath385 for @xmath386 . with @xmath378 given at time @xmath387 , and in order to ensure that @xmath388 remains divergence - free at @xmath389 , we must impose the additional boundary condition @xmath378 at all times @xmath390 . we will show that this boundary condition is needed in solving for @xmath172 in a finite - difference scheme . in order to solve the pressure poisson equation , we need to evaluate @xmath391_{i , j}$ ] for @xmath347 and @xmath348 , @xmath392_{i , j}$ ] for @xmath393 and @xmath348 , and @xmath394_{i , j}$ ] for @xmath347 and @xmath395 . for @xmath396 , we have @xmath397_{i , j}=0,\ ] ] for @xmath393 and @xmath348 ; @xmath398_{i , j}= & 2{\cal r}\left [ \displaystyle\frac{v_{xi+1,j}-v_{xi-1,j}}{2\delta_x } \displaystyle\frac{v_{zi , j+1}-v_{zi , j-1}}{2\delta_z } -\displaystyle\frac{v_{zi+1,j}-v_{zi-1,j}}{2\delta_x } \displaystyle\frac{v_{xi , j+1}-v_{xi , j-1}}{2\delta_z}\right ] \\ & + { \cal b}\mu_{i , j}\left ( \displaystyle\frac{\phi_{i+1,j}-2\phi_{i , j}+\phi_{i-1,j}}{\delta_x^2}+ \displaystyle\frac{\phi_{i , j+1}-2\phi_{i , j}+\phi_{i , j-1}}{\delta_z^2}\right)\\ & + { \cal b}\left(\displaystyle\frac{\mu_{i+1,j}-\mu_{i-1,j}}{2\delta_x } \displaystyle\frac{\phi_{i+1,j}-\phi_{i-1,j}}{2\delta_x } + \displaystyle\frac{\mu_{i , j+1}-\mu_{i , j-1}}{2\delta_z } \displaystyle\frac{\phi_{i , j+1}-\phi_{i , j-1}}{2\delta_z}\right ) , \end{array}\ ] ] for @xmath399 and @xmath400 ; and @xmath398_{i , j}= & 2{\cal r } \displaystyle\frac{v_{xi+1,j}-v_{xi-1,j}}{2\delta_x } \displaystyle\frac{v_{zi , j+1}-v_{zi , j-1}}{2\delta_z } \\ & + { \cal b}\mu_{i , j}\left ( \displaystyle\frac{\phi_{i+1,j}-2\phi_{i , j}+\phi_{i-1,j}}{\delta_x^2}+ \displaystyle\frac{\phi_{i , j+1}-2\phi_{i , j}+\phi_{i , j-1}}{\delta_z^2}\right)\\ & + { \cal b}\displaystyle\frac{\mu_{i+1,j}-\mu_{i-1,j}}{2\delta_x } \displaystyle\frac{\phi_{i+1,j}-\phi_{i-1,j}}{2\delta_x } , \end{array}\ ] ] for @xmath399 and @xmath395 ( where @xmath401 and @xmath402 ) . we see that @xmath160 and @xmath41 at ghost sites of @xmath403 appear in the last expression . the ghost @xmath160 s have already been determined in solving the convection - diffusion equation , while the ghost @xmath41 s are determined through the additional boundary condition @xmath378 : @xmath404 for @xmath399 , and @xmath395 . for @xmath405 , we have @xmath406_{i , j}=0\ ] ] for @xmath393 and @xmath348 ; @xmath407_{i , j}=0\ ] ] for @xmath393 and @xmath395 ; and @xmath407_{i , j}= \displaystyle\frac{v_{zi , j+1}+v_{zi , j-1}}{\delta_z^2}+{\cal b } \mu_{i , j}\displaystyle\frac{\phi_{i , j+1}-\phi_{i , j-1}}{2\delta_z}\ ] ] for @xmath399 and @xmath395 ( where @xmath401 ) . the last expression involves the ghost @xmath160 s and @xmath41 s at @xmath403 . given the above values of @xmath391_{i , j}$ ] and @xmath408_{i , j}$ ] , we apply a 2d fast fourier transformation to solve @xmath409 ( up to a constant ) for @xmath347 and @xmath348 . the discretized navier - stokes equation is given by @xmath410 for @xmath399 and @xmath348 , and @xmath411 for @xmath399 and @xmath400 , together with the boundary conditions that @xmath412 at @xmath395 and @xmath413 given by eq . ( [ single - couette ] ) at @xmath393 . equation ( [ discnsex ] ) at @xmath395 involves @xmath160 and @xmath40 at ghost sites of @xmath403 . the ghost @xmath160 s come from @xmath359 at @xmath395 , and have already been determined . the ghost @xmath40 s are determined from the discretized gnbc @xmath414^{-1}v_{xi , j}^{slip}= { \cal b}\left[\displaystyle\frac{\phi_{i , j-1}-\phi_{i , j+1 } } { 2\delta_z}- \displaystyle\frac{\sqrt{2}}{3}\cos\theta_s^0 s_\gamma(\phi_{i , j } ) \right]\displaystyle\frac{\phi_{i+1,j}-\phi_{i-1,j}}{2\delta_x } -\displaystyle\frac{v_{xi , j-1}-v_{xi , j+1}}{2\delta_z},\ ] ] at the lower boundary @xmath367 with @xmath415 , and @xmath416^{-1}v_{xi , j}^{slip}= { \cal b}\left[\displaystyle\frac{\phi_{i , j+1}-\phi_{i , j-1 } } { 2\delta_z}- \displaystyle\frac{\sqrt{2}}{3}\cos\theta_s^0 s_\gamma(\phi_{i , j } ) \right]\displaystyle\frac{\phi_{i+1,j}-\phi_{i-1,j}}{2\delta_x } -\displaystyle\frac{v_{xi , j+1}-v_{xi , j-1}}{2\delta_z},\ ] ] at the upper boundary @xmath417 with @xmath418 . in summary , to solve the dynamic equations ( [ he1 ] ) and ( [ he2 ] ) , we need to use @xmath364 and eq . ( [ single - couette ] ) at @xmath363 , with @xmath115 , @xmath313 , plus eqs . ( [ he3 ] ) and ( [ he4 ] ) at @xmath366 . in particular , in applying the boundary conditions at @xmath366 , values of @xmath160 , @xmath40 , and @xmath41 at ghost sites have to be introduced and solved for . we outline the scheme for time discretization and integration . for simplicity we only describe the forward euler time stepping . in the following a superscript @xmath39 denotes consecutive time instants and @xmath419 is the time interval . * step 3 * : compute @xmath424 and @xmath425 at all the interior sites ( except those fixed by the boundary conditions at all times ) using @xmath426 and @xmath427 according to eqs . ( [ disccde ] ) , ( [ discnsex ] ) , and ( [ discnsez ] ) in discretized time . here the ghost @xmath422 , @xmath420 , and @xmath421 determined in step 1 and @xmath423 solved in step 2 are needed . blake first proposed that the uncompensated young stress contributes to the relative slipping , in the context of an adsorption / desorption model . see t. d. blake , _ dynamic contact angles and wetting kinetics _ , in wettability , edited by j. c. berg ( marcel dekker , inc . , 1993 ) p.251 . for the contribution of intermolecular forces to the stress , we have directly measured the fluid - fluid interaction forces across bin surfaces instead of using the irving - kirkwood expression ( j. h. irving and j. g. kirkwood , j. chem . * 18 * , 817 ( 1950 ) ) , whose validity was noted to be not justified at the fluid - fluid or the wall - fluid interface ( see the paragraph following equation ( 5.15 ) in the above reference ) . this relation is obtained by assuming the two fluids interacting independently with the wall , so that @xmath71 may be expressed as the weighted average of @xmath428 and @xmath429 . the desired expression is obtained by noting @xmath430 to within @xmath431 . according to the mechanical definition of interfacial tension , @xmath432 equals to the integral ( across the interface ) of the difference between the normal and parallel components of the pressure , @xmath433 $ ] , where @xmath434 is along the interface normal , and @xmath435 and @xmath436 are the pressure - tensor components normal and parallel to the interface , respectively ( far from the interface @xmath437 ) . see j. g. kirkwood and f. p. buff , j. chem . * 17 * , 338 ( 1949 ) . the laplace s equation , plus the fact that @xmath109 and @xmath110 for curved ( fluid - fluid ) interfaces , can be derived from the above expression for @xmath432 , together with force balance considerations . our md simulation results thus confirm this definition of the surface tension . from md simulations , the interface @xmath438 , with a profile very accurately described by the @xmath439 form predicted by the ch free energy . the small value of @xmath326 implies negligible diffusion across the interface . for gently curved interfaces , @xmath195= \int_{\rm int}d\phi(k\partial_l\phi)\cos\theta_d^{surf}$ ] , where @xmath440 means taking partial derivative with respect to ( fluid - fluid ) interface normal @xmath434 , with @xmath441 approximated by @xmath442 . as @xmath443 , we have @xmath195=\gamma\cos\theta_d^{surf}$ ] . agreement between continuum hydrodynamics and md simulation has been demonstrated by treating the slip profile ( @xmath35 along the wall , derived from md simulations ) and the dynamic contact angle as inputs to the continuum model , see ref . the significance of our results lies in the fact that no such information have been put in by hand .

using the rhessi satellite as a compton polarimeter , a recent study claimed that the prompt emission of grb021206 was almost fully linearly polarized . this was challenged by a subsequent reanalysis . we present an novel approach , applying our method to the same data . we identify compton scattering candidates by carefully filtering events in energy , time , and scattering geometry . our polarization search is based on time dependent scattering rates in perpendicular directions , thus optimally excluding systematic errors . we perform simulations to obtain the instrument s polarimetric sensitivity , and these simulations include photon polarization . for grb021206 , we formally find a linear polarization degree of% , concluding that the data quality is insufficient to constrain the polarization degree in this case . we further applied our analysis to grb030519b and found again a null result .

astro-ph0405525

one of the most outstanding problems in present - day models of the energy release in @xmath1-ray bursts ( grb ) relates to the conversion of the liberated energy into observed electromagnetic radiation . in relativistic fireball models , the pressure of the photon - lepton plasma itself leads to relativistic expansion ( e.g. , @xcite ) . baryons are then accelerated through their coupling to the electrons . at a more basic level , the origin of the expanding fireball requires a definitive source of energy release that may itself be a source for mass acceleration . @xcite and @xcite discuss the formation of relativistically expanding jets from hyperaccreting , stellar - mass black holes that are formed as a consequence of iron core collapse of a rotating , massive ( @xmath2 ) star . alternative models for the extraction of rotational ( disk and black hole ) energy and its conversion to expanding shells or jets involve magnetohydrodynamic processes , including reconnection and dynamo operation in the accretion disks ( e.g , @xcite ) . the most direct evidence for the presence of magnetic fields in grb comes from their emission spectrum which is now widely interpreted in terms of synchrotron emission from relativistic electrons . whether there is a direct connection between magnetic fields produced in the immediate environment of the grb or the precursor star and the fields that are , at much larger distances , responsible for the observed synchrotron emission is not clear . @xcite argue that the required field strengths _ for grb afterglow shocks _ exceed those to be expected from dragging a pre - existing progenitor field along the expanding shell , and that they also exceed field strengths that could be produced by compression of interstellar magnetic fields . a field - generating mechanism intrinsic to the shocks would thus be required . the situation is , however , less clear for the prompt grb emission . the recent report @xcite of strong linear polarization in the _ prompt _ @xmath1-ray emission from grb021206 @xcite therefore stirred some excitement . the observations were performed with the rhessi satellite @xcite . from extensive modeling , the authors derived a polarization degree of @xmath3% , compatible with the assumption of maximum polarization . the emitting electrons are commonly thought to be accelerated in collisionless shocks by the fermi mechanism to a energy distribution , @xmath4 , with typically @xmath5 . in that case , standard synchrotron theory predicts that the intrinsic polarization degree @xmath6 is a function of @xmath7 , namely @xmath8 @xcite . this value therefore constitutes a maximum for a homogeneous magnetic field . for @xmath9 , one thus finds @xmath10 . tangled magnetic fields with different field vectors relative to the line of sight will in general reduce @xmath6 . the implications of these observations are far - reaching : not only does this observation further support the synchrotron radiation model , it also seems to require nearly homogeneous magnetic fields over the source visible to the observer ( a solid angle of @xmath11 ) . @xcite argue that magnetic fields dragged by the expanding shell from the surface of the exploding object are too weak to produce the prompt grb , requiring additional turbulent shock - generated fields that , however , would produce unpolarized emission . therefore , they suggest that the magnetic fields produced in the central engine are responsible for driving the fireball themselves . alternatively , post - shock dynamos could generate the highly ordered magnetic fields ; but then it will have to be shown that the instabilities occur on large spatial scales @xcite . @xcite calculated the pulse - averaged polarization degree for a relativistically expanding shell that contains a global toroidal field . the synchrotron emission is again assumed to be from power - law distributed electrons . their calculations predict a maximum of 60% for the polarization fraction , depending on the spectral parameters of the grb . this view is not unequivocal . @xcite points out that a slight off - axis orientation of the relativistically expanding jet produces a strong linear polarization signal even in the presence of random magnetic fields . for further views on this , we refer to the summary in lyutikov et al . ( 2003 ) and references cited therein . in any case , the prospect of diagnosing magnetic field structures in the very emission region of the prompt grb deserves upmost attention . @xcite have reanalyzed the rhessi data of grb021206 and cast serious doubts on the polarization measurements reported earlier by @xcite . @xcite conclude that the signal reported by @xcite is either spurious or is not related to polarization , and they claim that no statistically meaningful statement can be made for the degree of linear polarization for this particular grb . we have reanalyzed grb021206 and find problematic issues in both previous analyses . we find , in line with @xcite , clear evidence that much of the signal claimed by @xcite is induced not by source polarization but by accidental coincidences of two unrelated photons arriving at the same time in two different detectors . we essentially confirm the null results of @xcite , but we present a polarization analysis that compares simultaneous scattering rates of orthogonal detector pairs and thus does not need a complicated normalization . additionally , after implementing photon polarization in the geant3 software , we made extensive simulations of rhessi s response to a fully polarized grb . we also add one further grb to this analysis , again finding no statistically significant evidence for non - zero polarization . the structure of the paper is as follows : we start with a description of the rhessi satellite and its relevant features for polarization analysis in [ sec : rhessi ] . next , we present the rhessi data of grb021206 in [ sec : data ] . in our data analysis ( [ sec : method ] ) , we first select coincidence events ( [ sec : c_types ] ) , before we search for a polarization signal in [ sec : polanal ] . then , in [ sec : simul ] , we present our simulations of a fully polarized grb . the results of the polarization analysis of grb021206 are presented in [ sec : results ] , together with the results from our simulation of a fully polarized grb . after a quick look at grb030519b - another candidate for polarization analysis - we compare our analysis with previous works ( [ sec : comp ] ) and discuss in general the suitability of rhessi as a grb - polarimeter in [ sec : req ] . the reuven ramaty high energy solar spectroscopic imager ( rhessi , @xcite ) is a nasa small explorer ( smex ) mission that was launched on 2002 february 5 into a low earth orbit ( 600 km altitude , @xmath12 inclination ) . rhessi s primary science goal is to study solar flares in x - rays and gamma rays by using high resolution imaging spectroscopy . the mission is equipped with a single instrument made up of two main systems : an imager and a spectrometer . the imager @xcite consists of a 1.5@xmath13 m long tube with a set of 9 grids mounted on the front end of the tube and an identical set mounted on the rear end . the system makes use of the rotational modulation principle to obtain images of solar flares @xcite . rhessi rotates with a period of @xmath14s and points always towards the sun . the spectrometer consists of nine large germanium detectors cooled to 75@xmath13k . only weak lateral shielding was applied around the ge - detectors . this causes rhessi to be an all sky monitor for photons with energies above roughly 50@xmath13kev . thus , in addition to solar observations , rhessi is also suited for grb observations . the arrangement of rhessi s nine coaxial germanium detectors is shown in fig . [ fig : dets ] . they have a diameter of 7.1@xmath13 cm and a height of 8.5@xmath13 cm each . the cylinders are electrically segmented in a thin front segment ( about 1.5@xmath13 cm thick ) and a thick rear segment ( about 7@xmath13 cm thick ) , for details see @xcite . in reality the values of the front versus rear thickness may vary significantly from detector to detector . the front segments are able to detect photons from 3@xmath13kev to 2.7@xmath13mev with a resolution ( fwhm ) of @xmath15 1@xmath13kev ( at 100@xmath13kev ) , while the rear segments are sensitive from about 20@xmath13kev to 17@xmath13mev with a resolution of @xmath15 3@xmath13kev ( at 1@xmath13mev ) . a shielding plate with holes for the detectors is placed on top of the cryostat which protects the rear segments from solar photons scattered in the cryostat materials . similar graded z cylinders surround the front segments and protect them from low energy background photons ( @xmath16kev ) coming from the side . with respect to grb observations , rhessi s detection threshold depends on the effective shielding encountered by the incoming gamma rays . in the particular case of grb021206 , which was located at an angle of @xmath17 with respect to the sun s direction , the lower threshold of the rear segments was about 100@xmath13kev . this is mainly caused by absorption in the front segments and in the grid structures of the imager , absorption in the shielding plate , and absorption in the support modules located on the spacecraft equipment deck ( see fig . [ fig : hessi ] ) . these support modules are placed with a symmetry of @xmath18 in order to balance the spacecraft . the uneven mass distribution of the support modules together with the rotation of rhessi leads to a periodic variability of the detection threshold , especially for grbs coming in under such a small polar angle . rhessi was originally designed to measure linear polarization of hard x - rays from the sun with the help of a be scatterer mounted in the plane of the spectrometer @xcite . solar photons pass through predefined holes in the imager tube and , after compton scattering in the be block , are detected in the neighboring rear segments . for compton polarimetry of gamma rays coming from non - solar sources ( and therefore missing the holes for the be scatterer ) , one rather uses coincidence measurements between two detectors , one of which works as a scatterer . the azimuthal distribution of the compton scattered photons depends on the photon polarization : polarized photons are preferentially scattered in the direction perpendicular to the initial polarization direction . this azimuthal modulation is maximal for scattering angles of about @xmath18 . therefore , rhessi s polarization sensitivity is highest for photons approaching the satellite from directions perpendicular to the detector plane , i.e. close to its rotation axis ( as was the case for grb021206 ) . the coincidence rate of two detectors is a function of the angle between the photon polarization direction and the line through the detector centers . as this angle changes with the rotation of the satellite , the coincidence rate may exhibit a characteristic modulation pattern . its amplitude is proportional to the polarization degree , and its period is @xmath19 . the maximum amplitude that a 100% polarized grb would produce is an intrinsic function of the instrument and is obtained from monte carlo simulations . the detector signals are amplified by charge - sensitive preamplifiers before being sent to the instrument data processing unit ( idpu ) for further analog and digital processing : amplification , digitalization , time stamping etc .. there are eighteen independent readout channels , one for each detector segment . each detected event is defined by its detector number and segment , its arrival time and its energy channel . the event time is measured in units of binary microseconds ( @xmath20b@xmath21s@xmath22s ) . the absolute time is known with an accuracy of a few milliseconds . the data are stored event by event in the onboard memory . at the time of the grbs studied here , front and rear segments of detector 2 were coupled . its energy resolution is therefore much worse ( @xmath23kev at @xmath24kev ) . the readout of detector 2 was provided only via the front channel . in order to save onboard memory , an automatic `` decimation '' procedure is applied to events in the rear segments : only one out of @xmath25 rear events is stored for energies below a predefined threshold . this rear decimation is switched on mainly when the spacecraft is at high magnetic latitudes , where electron precipitation and cosmic ray background are high . this was the case during grb021206 with @xmath26 and energy channels below 1024 ( @xmath27kev ) . we developed our own routines for the polarization search . the basic data structure we use is the event list that contains for each detected photon its arrival time , its energy and the corresponding detector segment number . the basic analysis steps are described below . the light curve of grb021206 , extracted from the event list for all entries with energies @xmath28 is shown in fig . [ fig : ltc ] . for the polarization analysis , we selected a time interval that corresponds to exactly one rotation of rhessi ( dashed lines in fig . [ fig : ltc ] ) : @xmath29 where @xmath30 is the photon arrival time since 2002 december 6 22:49:00 ut . we find a total of about @xmath31 photons in this time interval and energy band . the corresponding energy spectrum is shown in fig . [ fig : rspec ] ( summed over all detectors and segments ) . the step around 380 kev is due to decimation in the rear segments . the 511 kev line originates from positron creation and annihilation in the spacecraft and the atmosphere . the total numbers of counts in each detector segment are listed in table [ tab : counts ] , using the time and energy interval of eq . ( [ eq : e0 ] ) and eq . ( [ t0_cut ] ) . detector 2 has no rear counts and a high total count rate since it is treated as an undecimated front segment . the number of counts in the other detectors varies significantly , especially in the front segments , whose thicknesses are different from detector to detector . a photon that is compton scattered from one detector into another one makes signals in two detectors , and thus makes two coincident entries ( indices @xmath32 and @xmath33 ) in the event list . naively , one would expect the real coincidences to happen within the same binary microsecond . it turns out , however , that real coincidences in the rhessi data can have a time difference @xmath34 up to @xmath35b@xmath21s ( see below ) . compton scattering is the most important interaction of photons with germanium at energies from 150@xmath13kev up to a few mev . the energy @xmath36 of a compton scattered photon is @xmath37 where @xmath38 is the initial photon energy , @xmath39 the electron mass and @xmath40 the scattering angle between in- and out - going photon . the kinetic energy @xmath41 of the recoil electron is @xmath42 for a 150@xmath13 kev photon that makes a compton scattering with @xmath43 the energy of the recoil electron is @xmath44kev . in our observation , low energy noise dominates below about 15 kev ( see fig . [ fig : rspec ] ) . we therefore require for the energies @xmath45 and @xmath46 of a coincidence pair ( `` energy cut '' ) : @xmath47 if we assume that the total energy of the scattered photon is measured in one of the ge detectors , and the total energy @xmath41 of the recoil electron in the other one , we can calculate the scattering angle @xmath40 from the above equation ( [ eq : compton ] ) , namely @xmath48 by taking into account the geometry of the detectors and the space in between we allow @xmath40 within @xmath49 . this rather generous angle range corresponds to @xmath50 . for each event pair ( energies @xmath45 and @xmath46 ) we can therefore apply an additional energy cut ( `` kinematical cut '' ) , namely @xmath51 because we do not know which detector observed the scattered photon and which one the recoil electron , we have to test both combinations . concerning the accepted detector pairs , we distinguish between spatially _ close _ and _ distant _ detector pairs , see fig . [ fig : dets ] . among the 36 possible pairs we have 19 close and 17 distant pairs . the chance of a real compton scattered event between distant detectors is considerably smaller than between close detectors . for pairs like e.g. detectors 2 and 6 , it is essentially zero because detector 7 is located in between . since the mean free path of a compton scattered photon is about 1 to 2@xmath13 cm , it would be stopped or scattered in detector 7 . indeed , when the analysis was performed with the distant detector pairs only , an insignificant number of compton scattering candidates was found ( see section [ sec : distant ] ) . the acceptance of only close detector pairs is called `` close pairs cut '' . in order to find the time difference @xmath52 of _ real _ coincidences , we study the number of pairs as a function of @xmath52 for _ any _ two events ( not only two consecutive entries in the event list ) fulfilling ( [ t0_cut ] ) and ( [ e0_cut ] ) , and occurring in a pair of close detectors . fig . [ fig : dtplot ] shows six plots which represent the possible combinations between the involved electronics ( f : front segments , r : rear segments , and 2 : the unsegmented detector 2 ) . rf , for instance , stands for pairs for which one signal occurred in a rear segment and the other in a front segment , and @xmath52 is defined as the time difference between front and rear . negative @xmath52-values indicate cases where the front signal occurred earlier than the rear signal . in order to obtain positive and negative time differences in the cases of rr or ff pairs , we arbitrarily choose to subtract the time of the detector with the lower number from the time of the partner with the higher number . this does not affect our further analysis . rr,2xdec stands for coincidences between two rear segments , where both energies @xmath45 and @xmath46 are within the decimated range . the structure in the @xmath52-plot of the rr,2xdec - pairs shows a period of @xmath53b@xmath21s , which is due to the decimation procedure with @xmath54 . decimation in the rear segment works like a clocked veto @xcite : during @xmath20b@xmath21s , the events in all detectors are accepted , during the next @xmath55b@xmath21s they are all rejected . the real coincidences are clearly visible in fig . [ fig : dtplot ] . we find that they appear at different time differences @xmath52 , mainly at : @xmath56 these time bins are indicated in fig . [ fig : dtplot ] by the vertical gray lines , and they are used as `` @xmath52-cut '' . coincidences outside these time bins are accidental coincidences . we think that the different @xmath52 are caused by different delays in the readout channels : the rear segment electronics is fastest , the time stamp in the front segments is on average @xmath57b@xmath21s later , and detector 2 is about @xmath58b@xmath21s later . the high photon rate during grb021206 introduces numerous accidental coincidences . their rate can be determined from fig . [ fig : dtplot ] . we use the accidental coincidences with @xmath59b@xmath21s @xmath60 @xmath61b@xmath21s ( so - called `` time shifted coincidences '' ) and interpolate them around @xmath62 ( see fig . [ fig : dtplot ] ) . the time differences @xmath63 used to estimate the number of accidentals should on the one hand be higher than the longest readout time of the events , i.e. , longer than a few b@xmath21s . on the other hand , it should be smaller than the shortest time scale of variability in the grb light curve . the total number of coincidences ( @xmath64 ) as well as the number of accidental coincidences ( @xmath65 ) can be determined for any time interval , leading to time dependent functions @xmath66 and @xmath67 . the number @xmath67 is proportional to the square of the total count rate , and therefore shows a strong time variation . some of the real coincidences observed during the grb are induced by various background components . we call this number @xmath68 . it could show a rotation angle dependent time variation and can be estimated by interpolation of the data before and after the grb : @xmath69 where @xmath70 denotes a time bin before and @xmath71 a time bin after the grb , with equal rotation angle of rhessi as at time @xmath72 . the remaining number of coincidences , namely @xmath73 are called `` compton scattering candidates '' and includes the real compton scattering events . a closer look at all the close detector pairs that fulfill the energy cut ( [ e0_cut ] ) , the kinematical cut ( [ e_special ] ) , and and the @xmath52-cut ( [ dt_cut ] ) shows that many of them are part of a multiple coincidence ( 3 or more detector segments firing at the same time ) . it often happens that a photon undergoes a forward compton scattering in a front segment before being compton scattered in the rear segment into a neighboring detector . in other cases , several detectors observe a signal . this is relatively more often the case before and after the grb . multiple coincidences can be caused by nuclear reactions , bremsstrahlung , or pair / photon cascades originating from the passage of high energy cosmic rays through the spacecraft . none of these multiple coincidences contribute to a polarization signal . we therefore do not accept them . we identify multiple coincidences by looking at all accepted coincidences . let @xmath74 be the time of a coincidence , more exactly , the earlier time of the two events involved . two coincidences are members of a multiple coincidence if @xmath75b@xmath21s . all coincidences that fulfill this condition are rejected ( so - called `` no - multiples cut '' ) . however , members of multiple coincidences are also members of time shifted coincidences used to determine the number of accidental coincidences . in order to obtain the correct number of accidental coincidences , we clean the initial event list from members of multiple coincidences . we then search again for real and time shifted coincidences . in order to search for a polarization signal we first note that all close detector pairs can naturally be grouped into four scattering directions ( see fig . [ fig : dets ] ) , each separated by @xmath76 . according to a rhessi - fixed coordinate system , we call these directions the @xmath77 direction , the @xmath76 direction , the @xmath18 direction , and the @xmath78 direction . the true pair angles , measured from detector center to detector center , vary slightly . the mean deviation from the nominal value is @xmath79 with a maximum of @xmath80 . this is much less then the width of scattering angles accepted by a detector pair , which is of the order @xmath81 . assuming that polarized photons at a time @xmath82 were scattered preferentially in the @xmath78 direction , then , one eighth of a satellite rotation later , they would be scattered preferentially in the @xmath18-direction . after another rotation by @xmath76 , they would be scattered in the @xmath76-direction , and so on . we determine the number of compton scattering candidates for the 4 scattering directions and call them @xmath83 , with @xmath84 \ { 0 , 45 , 90 , 135 } . for a grb with a strongly varying light curve , such as grb021206 , the @xmath85 are strongly correlated with the light curve . let @xmath86 be the normalized coincidence light curves , @xmath87 where the sums are over the entire time interval used for polarization analysis ( typically one or more full rotations , e.g. eq . ( [ t0_cut ] ) ) . we can then define the following asymmetries : @xmath88 for an _ unpolarized _ grb , we make the plausible assumption that the @xmath85 are approximately proportional to each other , i.e. we assume that there exists a time dependent function @xmath89 and factors @xmath90 such that we can write @xmath91 . the factors @xmath90 take into account the efficiencies of the four scattering directions , while @xmath89 is closely related to the light curve of the grb . in that case , the asymmetries ( [ eq : asym ] ) are always zero ( within statistical errors ) . in the case of a _ polarized _ grb , an additional sinusoidal time dependence is introduced : @xmath92 the observation of a non - zero modulation amplitude @xmath93 would be an indication of a grb polarization . simulations show that @xmath93 is of order @xmath94 for a @xmath24% polarized grb . the expected periodicity of the modulation is @xmath95 . the absolute phase @xmath96 is unknown , and the relative phase shifts are fixed at @xmath97 , @xmath98 , @xmath99 , and @xmath100 . from eqs . ( [ eq : norm_cs_ltc])-([eq : ltc_mod ] ) , we obtain @xmath101 the definition of the asymmetries @xmath102 and @xmath103 uses only data . by making a fit , the best values for the observable modulation amplitude @xmath93 and the phase @xmath96 can be determined . in order to obtain the real polarization degree of a grb , @xmath93 should be compared with @xmath104 obtained from the simulation of a 100% polarized grb , analyzed the same way . the standard rhessi software provides energy deconvolution routines , using simulated response functions , but only for solar photons . therefore we have rerun the simulation code , but for off - axis photons . all essential parts of the satellite are built into a mass model using the geant3.21 particle tracking code @xcite . the model provides an exact description of the ge spectrometer and its adjacent elements , e.g. graded - z shielding , detector electronics and cryo - cooler . other components and structures that are positioned at larger distances from the ge detectors , maintain their mass and composition while some approximations were made for their geometry ( e.g. simplified shapes ) and internal arrangement ( e.g. using material of mean density and z - number ) . the total mass of the spacecraft is 291@xmath13 kg , including 131@xmath13 kg of scientific payload . especially for photons that arrive at an angle of less than @xmath105 with respect to the sun s direction ( as was the case for grb021206 ) , the mass distribution on the equipment deck affects the light curves and spectra . the most relevant components responsible for such effects are the solid state recorder at @xmath78 above ( i.e. sun side of ) the equipment deck , the cryo - cooler power converter at @xmath106 ( above the equipment deck ) , the idpu at @xmath76 ( below the equipment deck ) , and the battery cells at @xmath107 ( below the equipment deck ) , see fig . [ fig : hessi ] ( or @xcite , fig.10 ) . they all weigh of the order 10@xmath13 kg . using the mass model of the spacecraft and the spectrometer , the rhessi response function can be generated . for this purpose the spacecraft model in the geant code is illuminated by a parallel flux of gamma rays with a predefined energy spectrum and direction . all primary particles , like scattered photons , as well as secondary particles , like recoil electrons , are traced . finally , the energy deposition in each germanium detector segment is stored . photon polarization is not included in geant3 . we wrote specific routines to fully include photon polarization , using the same formalism as in the geant4 extension packet for low energy compton scattering @xcite . the momentum vector of the scattered photon and its polarization direction were generated in accordance with the klein - nishina formula taking into account its initial polarization . in order to analyze the simulated data with the same program as the real data , the output of the geant - code was prepared to look like real data . the effects of the readout electronics were simulated by delaying the front segment events by 1.6@xmath108s and the events in detector 2 by 2.6@xmath108s ( see end of section [ sec : coinc ] ) . next , the time was converted into integers ( in units of b@xmath21s ) and the energy was converted into channels . the rear segment decimation was simulated by accepting rear segment events with energy channel @xmath109 only if @xmath110 . we also added events as cosmic induced background . in order to obtain good statistics , we simulated the grb lasting over many rotations ( typically 10 to 40 ) . a very high photon rate per rotation would introduce a very high accidental coincidence rate . we start the data analysis of grb021206 by identifying possible coincidences . possible coincidences are event pairs that have a time difference @xmath111b@xmath21s , belong to the time interval given in eq . ( [ t0_cut ] ) , and fulfill the energy cut ( [ e0_cut ] ) . we find 16542 such pairs , see table [ tbl : cut_eff ] . using the method described in section [ sec : c_types ] to determine @xmath65 and @xmath68 , we find 1641 compton scattering candidates . the effects of successively applying the @xmath52-cut , the close pairs cut , the kinematical cut , and the no - multiples cut are listed in table [ tbl : cut_eff ] . the @xmath52-cut and the close pairs cut affect the number of compton scattering candidates only slightly , but are both very effective in reducing the number of accidental coincidences , as can be seen from table [ tbl : cut_eff ] . the kinematical cut seems less effective . however , simulations show that the modulation amplitude is about 15% higher when using the kinematical cut . the sequence of the energy cut , the @xmath52-cut , the close pairs cut and the kinematical cut is interchangeable . the last line of table [ tbl : cut_eff ] is obtained after applying the no - multiples cut . the no - multiples cuts is not interchangeable with the other cuts . the quite big effect of the no - multiples cut can be explained by the fact that multiple coincidences can lead to comparably many double coincidences . a triple coincidence of , e.g. , detector 2 , rear segment 3 and rear segment 5 can lead to three accepted coincidences . multiple coincidences that include 4 or more detector segments will cause even more double coincidences . the time dependent total coincidence rate @xmath66 of grb021206 is plotted in fig . [ fig : t1hist ] ( crosses ) , together with the number @xmath67 of accidentals ( gray filled histogram ) . both @xmath64 and @xmath65 were determined as described in section [ sec : c_types ] . it can be seen that a large fraction of all coincidences during the grb are accidentals . before and after the grb , we observe background coincidences . the number of accidental coincidences is very small before and after the grb . we note at this point that the energies of the background coincidences are on average higher than those of the grb coincidences . we also observed that the percentage of multiple coincidences is higher among the background coincidences than among the grb induced coincidences . we obtain a total of @xmath112 coincidences in the time interval between the two dashed long vertical lines of fig . [ fig : t1hist ] , and @xmath113 accidental coincidences ( the error is @xmath114 because the interpolation of the accidentals relies on 10 times better statistics ) . from the time interval before and after the grb ( indicated by the short dashed lines in fig . [ fig : t1hist ] ) we determine the number of background coincidences as @xmath115 . thus we find @xmath116 compton scattering candidates . the cited errors are purely statistical . the time distributions of the compton scattering candidates during the grb ( @xmath85 ) are plotted in fig . [ fig : diffhist ] for the four possible scattering directions ( see section [ sec : polanal ] ) . the two directions in each plot make an angle of @xmath18 . if , in case of a polarized burst , one curve showed a relative maximum compared to the other curve , 4 bins later ( @xmath15 1.0@xmath13s ) the other curve would show a relative maximum . additionally , the two plots are related by the fact that a relative maximum in the light curve of the @xmath78 direction ( upper plot ) would appear 2 bins ( @xmath15 0.5@xmath13s ) later in the @xmath18 direction ( lower plot ) . no such effect can be seen by visual inspection . the asymmetries @xmath117 and @xmath118 , defined in eq . ( [ eq : asym ] ) , are plotted in fig . [ fig : asym ] , together with the best fit of the function ( [ eq : fitfun ] ) . since the two curves are connected by a fix phase , we fit them simultaneously in order to obtain the two free parameters , namely the asymmetry amplitude @xmath119 and the phase @xmath96 . the fit result for the modulation amplitude is @xmath120 with a @xmath121 of 14.0 ( 28 dof ) . we have also studied grb030519b @xcite as another good candidate for polarization analysis . grb030519b is also very strong and occurred at an angle of @xmath122 with respect to the direction toward the sun , i.e. it came from the antisolar side at an anlge of only @xmath123 . there was no decimation during this observation . with the method described in section [ sec : method ] , but including two full rotations from 14:04:53.80 to 14:05:02.062 ut , we obtain a total of @xmath124 coincidences , @xmath125 accidental coincidences , and the number of background coincidences is @xmath126 , resulting in @xmath127 compton scattering candidates . more than half of the coincidences originate from background . the more than three times higher value for the background coincidences ( compared with grb021206 ) can be explained by the two times longer grb time interval , and by the undecimated data . the fitted amplitude is @xmath128 with a @xmath121 of 11.7 ( 28 dof ) . the initial energy spectrum is simulated between 25 kev and 5 mev with a spectral shape that follows a simple power law : @xmath129 , where @xmath130 is the number of photons per energy interval . we made simulations of fully polarized grbs with power law indices @xmath131 , @xmath132 , and @xmath133 . these simulations were made with uniform time distribution , corresponding to a constant flux during the burst . between 150 kev and 2 mev , a power law index of @xmath134 gives a detected spectrum that agrees well with the one of grb021206 , see fig . [ fig : comp_spec ] . we also made a high - statistics simulation of a fully polarized grb with @xmath134 taking into account the highly variable light curve of grb021206 . after applying decimation , we found @xmath135 compton scattering candidates . the asymmetry plot for this simulation is shown in fig . [ fig : cdiff_sim ] . the fitted modulation amplitude is @xmath136 the fit result for the phase , @xmath137 , is in good agreement with the simulated value of @xmath138 . the @xmath121 in the fit is @xmath139 ( 28 dof ) . the error bars have different length because of the statistical errors of the underlying light curve . a uniform time distribution gives error bars of similar length . the fitted asymmetry modulations @xmath104 are plotted as a function of @xmath140 in fig . [ fig : simresult ] . at @xmath141 we have two simulations with uniform time distribution ( called a and b ) and one taking into account the light curve variation ( called ltc ) . the values obtained by including decimation are marked with diamonds , undecimated results are marked by triangles . the mean value of the uniform simulations , with decimation , is @xmath142% , in good agreement with eq . ( [ eq : my100 ] ) . we conclude from fig . [ fig : simresult ] : ( i ) within statistical errors , decimation does not affect the modulation amplitude @xmath143 , ( ii ) @xmath104 does not depend significantly on the power law index @xmath140 in the range 2.4 2.8 , ( iii ) @xmath104 does not depend on the light curve variation . this can be understood because each time bin can be thought of as an independent experiment . whether they are simulated with more or less statistics should not bias the result . the simulations also show that less than half of the so - called compton scattering candidates are due to the process we want to analyze : an incoming photon undergoes its first interaction within a rhessi detector and is then compton scattered into a second detector , without additional intervening interaction . some photons undergo a first interaction before they reach a germanium detector ( of the order 20% ) . other photons are scattered twice in the first detector before being detected in a second detector ( of the order 40% ) . we also made simulations with unpolarized photons and a power law index @xmath134 . we obtain ( decimation is applied ) @xmath144% for uniform time distribution and @xmath145% for a simulation taking into account the light curve of grb021206 . both fits agree with @xmath146 within statistics . simulations of grb030519b gave comparable values for the modulation amplitude @xmath104 as the simulations of grb021206 . the fraction of accidental coincidences was chosen to be small in these simulations . a high fraction of accidental coincidences would increase the statistical errors . as a test , we determined the number of compton scattering candidates as described in section [ sec : c_types ] , for the 17 distant detector pairs instead of the close detector pairs . from simulations _ we obtain a ratio @xmath147 with the data of _ grb021206 _ , we obtain for the distant detector pairs : @xmath148 , @xmath149 , @xmath150 , and therefore @xmath151 . combining eq . ( [ eq : n - c - grb021206 ] ) with eq . ( [ eq : ratio - dc ] ) , we would expect @xmath152 . this is slightly higher than the observed number . with the data of _ grb030519b _ , we obtain for the distant detector pairs : @xmath153 , @xmath154 , @xmath155 , and therefore @xmath156 . combining eq . ( [ eq : n - c - grb030519b ] ) with eq . ( [ eq : ratio - dc ] ) , we would expect @xmath157 . this is in excellent agreement with the observed number . the results show that the distant pairs do not contribute significantly to the number of compton scattering candidates . we select and identify coincidences according to the parameters given in sect . [ sec : coinc ] and [ sec : c_types ] . the efficiencies of the cuts used are listed in table [ tbl : cut_eff ] . these narrow cuts lead to a comparably small error for the number of compton scattering candidates ( see section [ sec : comp ] ) . by comparing scattered events in two directions that make a relative angle of 90@xmath158 , we eliminate most systematic effects induced by the strongly varying light curve . the asymmetry plot we present in fig . [ fig : asym ] is entirely based on observational data ; simulations are only necessary to find the possible degree of polarization . when processing simulated data of a fully polarized grb with our method , we find a modulation @xmath104 of the order 20 to 25% ( see fig . [ fig : simresult ] ) . the observed value @xmath93 has to be compared with the simulated value @xmath104 to obtain the polarization degree @xmath159 of a grb : @xmath160 from theoretical considerations ( see sect . [ sec : intro ] ) , a maximum polarization of @xmath161 is expected . thus , the maximum observable modulation amplitude @xmath162 is of the order 15% . the data of both grbs presented , namely grb021206 and grb030529b , agree with zero polarization ( see eq . ( [ eq : mp - grb021206 ] ) and eq . ( [ eq : mp - grb030519b ] ) ) . but given the large error bars ( of the order 9% for both grbs ) , the maximum modulation of about 15% can not be excluded . the formula ( [ eq : pgrb ] ) can be written in the form @xmath163 . in the case of grb021206 we obtain a mean value of @xmath164% for the polarization degree ( using eq . ( [ eq : mp - grb021206 ] ) and eq . ( [ eq : my100 ] ) ) . however , proper error propagation is complicated in this case . a simple estimate can be obtained by using the 1@xmath165 limits of @xmath93 and @xmath104 : @xmath166% and @xmath167% . therefore , the result of the polarization analysis can be summarized as @xmath168 the way we treat multiple coincidences and how we estimate the number of accidental coincidences is not fully consistent . if the photon flux is so high that the chance for triple accidental coincidences is not negligible , our method underestimates the number of compton scattering candidates systematically . when we determine @xmath64 we do not accept any triple coincidences including triple _ accidental _ coincidences , but afterwards we again subtract the contribution from triple accidental coincidences , thus obtaining a number @xmath169 that is too low . this underestimate is smaller , the narrower the cuts for accepting coincidences . that is the reason why we apply the no - multiples cut as last cut . in the case of grb021206 , most of the triple coincidences are physical . but some triple coincidences are purely accidental , and a comparable number of triple coincidences are mixed , i.e. a real coincidence is accidentally accompanied by an event in a third detector . we estimate that the real number of compton scattering candidates would be of the order 50 coincidences higher than the number given in eq . ( [ eq : n - c - grb021206 ] ) . an effect of the same order is expected for the distant detector pairs . the numbers cited in sect . [ sec : distant ] for grb021206 show that the observed number ( @xmath170 ) of compton scattering candidates for the distant pairs is @xmath1570 coincidences less than the expected number ( @xmath171 ) . this discrepancy is indeed of the order 50 . to determine this number exactly is beyond the scope of this work , especially since it would only slightly affect the result of the polarization analysis and does not affect the conclusion . the estimate given above is fully sufficient for our purposes . in the case of grb030519b , the chance for triple accidental coincidences is very small . the number given in eq . ( [ eq : n - c - grb030519b ] ) is therefore correct . this is further confirmed by the numbers cited in sect . [ sec : distant ] , where @xmath170 and @xmath171 agree very well for grb030519b . another point to discuss is the normalization we use in eq . ( [ eq : norm_cs_ltc ] ) , and the assumption that the efficiencies of the four scattering directions are proportional to each other . if statistics were so good , that this approximations could not be used , the @xmath121 of the fit of the asymmetry plot would be too high . but even with the high - statistics monte carlo simulations we have performed , our fits of the asymmetry plot gave acceptable results . another advantage of our eqs . ( [ eq : norm_cs_ltc ] ) and ( [ eq : asym ] ) is that the so defined asymmetry does only depend on data and thus can not suffer from systematic errors of the simulations . we conclude that our approximations are good enough within the accuracy necessary for this work . we note that the light curve of grb021206 has two maxima around 22:49:15 ut and around 22:49:17 ut . they are separated by roughly half a rotation , i.e. by the period that a possible polarization signal would have . effects that are related to the count rate can therefore possibly be mistaken for a polarization signal . the accidental coincidences on the one hand , are proportional to the square of the total grb light curve : @xmath172 . if accidental coincidences are mistaken for grb induced real coincidences and normalized with the light curve , an effect @xmath173 would be observed . background coincidences on the other hand , are constant with time ( at least approximately ) . if background coincidences are mistaken for grb induced real coincidences and normalized with the light curve , this would result in an effect @xmath174 . we observed for example that when we did not subtract the background coincidences , the coincidences from distant detectors did not cancel out but showed a significant modulation . since we identify the compton scattering candidates carefully , our analysis should not suffer from these effects . by comparing simultaneous scattering rates of perpendicular directions , we further minimize systematic effects related to the light curve . in table [ tbl : comp ] we compare our number of accepted coincidences with the results presented by @xcite ( hereafter cb ) and @xcite ( hereafter rf ) . we would like to emphasize that the total number of coincidences ( @xmath64 ) depends _ strongly _ on the cuts used . the number of compton scattering candidates ( @xmath169 ) is less dependent on the cuts used . its error @xmath175 , however , depends again on @xmath64 , namely @xmath176 , see eq . ( [ eq : nc ] ) . we can not confirm the high number of compton scattering candidates found by cb , our value being more than a factor 10 lower . on the other hand , we confirm the result presented by rf , although our error is 3 times smaller . we can obtain a high number of coincidences ( @xmath64 ) by widening the acceptance cuts , see first line of table [ tbl : cut_eff ] . but by widening the cuts we accept much more accidentals , but only a few more compton scattering candidates . however , we can not reproduce cb s high number @xmath64 and , at the same time , their relatively small @xmath177-ratio . the cuts used in cb are not described by the authors . a few details became very recently available in @xcite . in table [ tbl : comp ] we use the following cuts in order to reproduce cb s result : ( i ) time interval @xmath178 , ( ii ) as energy cut : @xmath179 and @xmath180 , ( iii ) as @xmath52-cut : @xmath181b@xmath21s , ( iv ) no - multiples cut ; the kinematical cut is not applied and all detector pairs are accepted . our slightly higher @xmath64 is probably due to a different treatment of multiple coincidences . in order to extract the polarization signal , cb rely strongly on the simulation of a non - polarized grb . as a simple test of how well simulations and data agree , we can use the ratio of compton scattering candidates to the total 0.15 - 2.0@xmath13mev light curve events : @xmath182 . since @xmath169 depends on the cuts used , @xmath183 does as well . we use simulations with a low fraction of accidental coincidences in order to minimize systematic effects . such simulations give values around @xmath184% , depending slightly on the power law index @xmath140 , namely @xmath185% . our data analysis gives @xmath186% , a value slightly smaller than from simulations . the likely reason is described in sect . [ sec : our - disc ] . if the chance of triple accidental coincidences is not negligible , as in the case of grb021206 , @xmath169 is underestimated , and therefore @xmath187 as well . we conclude that @xmath187 and @xmath188 agree reasonably well using our method and simulations . if we use similar cuts to cb ( see last paragraph ) , we obtain @xmath189% . from fig . 1 of cb we can estimate the total number of grb photons in the 150 to 2000 kev band to be @xmath190 ( in agreement with the number obtained from the event list ) . cb obtain @xmath191 compton scattering candidates , resulting in @xmath192% , i.e. in a value 7 times higher than expected from simulations . this could be another indication that cb s number of compton scattering candidates is too high . if cb s high number of compton scattering candidates was partly due to accidental coincidences , the effect discussed in sect . [ sec : systemtic ] were possibly responsible for the polarization signal found by cb . cb s value for the modulation factor , @xmath193% , which they obtained from simulations and analytical estimates , agrees very well with our value , if we omit the kinematical cut ( [ e_special ] ) . in that case , we obtain @xmath194% with a statistical error of @xmath195% . the main differences between our and rf s analysis in determining the number of compton scattering candidates are the following : 1 . instead of the @xmath52-cut ( [ dt_cut ] ) , rf mention to use @xmath196 b@xmath21s . the quite high rate at @xmath197b@xmath21s in fig . 2 of rf is interpreted by them as probably due to real coincidences , whereas it is caused by the decimation procedure ( see fig . [ fig : dtplot ] ) . + our very narrow @xmath52-cut is one of the main reason for our much smaller @xmath64 and @xmath65 . 2 . instead of accepting only close detector pairs rf accept also distant pairs . accepting distant detector pairs introduces mainly background noise ( see section [ sec : distant ] ) . we determine the number of accidental coincidences from a time shifted coincidence window . this has the advantage of being independent of the high variability of the grb light curve . different efficiencies of the detectors and effects due to the decimation procedure are also taken into account . by making the time window for the time shifted coincidences wide , we obtain a relatively small statistical error in @xmath65 . 4 . instead of our energy cut ( [ e0_cut ] ) , rf use 150 kev @xmath198 2000 kev . this narrower energy cut eliminates real coincidences . rf replace front / rear coincidences within the same detector by a single event at the earlier time stamp . photons that made a first interaction in the front segment before being compton scattered in the rear segment into another detector mostly lost their polarization information in the first interaction . such events do not contribute to the polarization signal . rf use the same @xmath199s long time interval as cb , whereas we use @xmath200s (= one full rotation ) . we also tried to reproduce rf ( see table [ tbl : comp ] ) by using the following cuts in our analysis : ( i ) time interval @xmath201 , ( ii ) as energy cut : @xmath202 , ( iii ) as @xmath52-cut : @xmath203b@xmath21s , ( iv ) no - multiples cut ; all detector pairs are accepted and no kinematical cut is applied . since rf used @xmath204b@xmath21s as @xmath52-cut ( presumably , the meaning of rf s @xmath205 is not entirely clear ) a cut not easily feasible with our program , that takes into account the repeating 4b@xmath21s - structure caused by decimation we added @xmath206 to @xmath64 and @xmath65 in order to estimate the effect of the wider @xmath52-cut . rf s slightly higher value for @xmath64 than our reproduction can be explained by point 5 above . rf s polarization analysis is not clear to us . their basic eq . ( 4 ) is valid for compton scattering coincidences . however , it is used for all coincidences , including about 80% accidentals , for which an expression like @xmath207 would be more correct . we further do not understand how to get from rf s equation ( 7 ) to equation ( 9 ) . concerning rf s section 7 , we agree with rf that the uncertainty in cb s simulation of an unpolarized grb is not discussed in cb . but , in our view , the uncertainty in the simulation of an unpolarized grb and the uncertainty in the simulations of @xmath104 are not as closely related as rf claim . a further evidence that the number of accidental coincidences is much higher than presented in cb comes from the analysis of rf . rf attempted to reproduce cb although they had to estimate cb s cuts . in fig . 7b of rf , they reproduced the `` angular light curve '' of cb by using a narrower energy cut than cb , but accepting @xmath208b@xmath21s . this clearly increases only the number of accidental coincidences . /bin / sh : this : command not found the fact that rf were able to reproduce cb s angular light curve by using very different cuts can be understood , if both sets of cuts , cb s and rf s , mostly accept accidental coincidences ( r.e . rutledge , private communication ) . a search for grb polarization using rhessi data is feasible only for very strong grbs . a fully polarized grb would yield a modulation amplitude @xmath209 . in order to distinguish between _ no _ polarization and _ full _ polarization , the error @xmath210 of the fitted modulation amplitude should be around 0.05 or smaller . for both grbs presented in this article we find @xmath211 ( see eqs . ( [ eq : mp - grb021206 ] ) and ( [ eq : mp - grb030519b ] ) ) . at first sight , it may seem surprising that both errors are similar , even though the 25 - 100 kev fluence of grb021206 as measured by ulysses was more than 5 times higher than for grb030519b ( see @xcite and @xcite ) . the main reason is the decimation of the rhessi data . if decimation had not been active during grb021206 , rhessi would have observed about 2.8 times more compton scattering candidates , about 2.8 times more accidental coincidences , and about 2.2 times more background coincidences . thus , the error @xmath210 would be about a factor @xmath212 smaller , i.e. @xmath213 . this would be just above the limit to distinguish between _ no _ polarization and _ full _ polarization . another reason for the similar @xmath210 is the better @xmath214 ratio in the case of grb030519b , namely 0.72 compared to 0.56 for grb021206 . the error @xmath175 , and therefore the error @xmath210 , depends on the total number of coincidences : @xmath176 , see eq . ( [ eq : nc ] ) . in the case of grb021206 , @xmath64 is dominated by accidental coincidences , while for grb030519b @xmath64 is dominated by background coincidences . the optimal case , however , would be if @xmath64 were dominated by compton scattering candidates . a grb suitable for polarization analysis with rhessi should have a fluence comparable to grb021206 and the data should not be decimated . furthermore , it should last about @xmath215s in order to be neither dominated by accidental coincidences nor by background coincidences . for such a grb , a polarization degree of 100% would be distinguishable from zero polarization at a confidence level of several standard deviations . since grb021206 was one of the strongest grbs ever observed , the probability for such a grb to occur is small , given that it should also come from a direction close to rhessi s rotation axis . the possibility that prompt grb emission is strongly linearly polarized at @xmath1-ray energies deserves attention with the advent of detectors that can potentially measure such polarization . polarization carries fundamental information on the orientation of magnetic fields in the source and eventually helps confine the magnetic field generation mechanism . its detection is difficult due to various interactions in presently available detector systems . the rhessi satellite can in principle be used as a polarimeter by using the direction dependence of compton scattering in conjunction with the rotation of the satellite . the effects are subtle , however , and require accurate knowledge of the mass distribution of the satellite and the detector geometry . the methods essentially analyze compton scattered grb photons that induce coincidences in detector pairs . a principal difficulty is the separation of such events from accidental and background coincidences in the same detector pairs . two previous publications @xcite have addressed this problem , arriving at essentially opposite conclusions for the same observed grb . whereas @xcite claim to detect maximum polarization ( @xmath216% ) , @xcite challenge this result and find no significant constraint for the polarization degree . we have revisited the problem of grb polarization measurements with the rhessi satellite . our basic test case is grb021206 used by @xcite and @xcite for their respective analyses . by applying well justified selection criteria for coincidences in energy , time , and scattering angle , we found @xmath217 compton scattering candidates . by plotting these 770 events as `` coincidence light curves '' for the four different scattering directions , we could define an asymmetry and search for a possible polarization signal . we compare the result of our polarization analysis with simulations of a 100% polarized grb . we can not reject the null hypothesis that the burst is unpolarized , but neither can we significantly detect any non - zero polarization . the maximum possible synchrotron polarization degree would , for the measured spectral index of the burst , be of order 70% , fully compatible with the data . we conclude that _ no statement on @xmath1-ray polarization can be made for grb021206_. our result contradicts the statements of @xcite , who find more than 9000 compton scattering candidates ( @xmath169 ) and claim to see a polarization signal at the 5.7@xmath165 level . the main problem in their analysis is the number of accidental coincidences ( @xmath65 ) that is , in our view , not determined correctly . @xcite obtain their signal by comparing a measured `` angular light curve '' with simulations of an unpolarised grb . if they do not use a correct @xmath218-ratio in their simulation , then the presented simulated data points can not be trusted . our result agrees in many aspects with the reanalysis presented by @xcite . by making a more sophisticated coincidence selection , we obtain much smaller errors , however . a few points of caution in the polarization analysis of @xcite have been mentioned in our presentation . we therefore conclude that rhessi has not yet measured the degree of polarization of any observed grb . implications on magnetic field orientation based on rhessi results are therefore premature , and the physics of magnetic field generation and structure in the @xmath1-ray source of grb must rely on alternative information for the time being . on the other hand , our analysis suggests that at least in principle , rhessi does offer the capability of measuring polarization . a grb with comparable fluence as grb021206 , but lasting @xmath219s , would be required to significantly distinguish maximum polarization of a grb from zero polarization . in addition , a grb suitable for polarization analysis should come from a direction close to rhessi s rotation axis . we thank o. wigger , a. mchedlishvili and e. kirk for many helpful discussions and encouragement . we also thank our referee , d.m . smith ( scipp , university of california , santa cruz ) , for careful reading of our manuscript and his constructive comments . blandford , r. d. , & znajek , r. l. 1977 , mnras , 179 , 433 boggs , s.e . , & coburn , w. , 2004 , x - ray polarimetry workshop , held at slac , stanford , california , 9 - 11 february . electronic versions of papers available under : http://heasarc.gsfc.nasa.gov/docs/heasarc/polar/polar.html cern program library office , 1993 , geant detector description and simulation tool ( cern geneva , switzerland ) coburn , w. , & boggs , s.e . , 2003 , nature 423 : 415 curtis , d. w. 1999 , ftp://apollo.ssl.berkeley.edu/pub/hessi/instrument/documents/hsi_idpu_020a.pdf galeev , a. a. , rosner , r. , & vaiana , g. s. 1979 , apj , 229 , 318 geant4 , http://wwwasd.web.cern.ch/wwwasd/geant4/geant4.html heitler , w. 1954 , the quantum theory of radiation ( oxford clarendon press ) hurford , g.j . 2002 , solar physics * 210 : * 61 hurley , k. et al . 2002 , grb circular network 1727 , 1728 , and 2281 hurley , k. et al . 2003 , grb circular network 2237 katz , j. 1994 , apj , 422 , 248 katz , j. 1997 , apj , 490 , 633 lamb , d. et al . 2003 , grb circular network 2235 and 2238 lin , r.p . et al . 2002 , solar physics * 210 : * 3 lyutikov , m. , pariev , v.i . , & blandford , r.d . 2003 , apj , 597 , 998 macfadyen , a. i. , & woosley , s. e. 1999 , apj , 524 , 262 mcconnell , m.l . 2002 , solar physics * 210 : * 125 medvedev , m. k. , & loeb , a. 1999 , apj , 526 , 697 mszros , p. , & rees , m. j. 1997 , apj , 482 , l29 piran , t. 1999 , phys . rep . , 314 , 575 rutledge , r.e . , & fox , d.b . , 2004 , mnras , vol . 350 , p. 1288 rybicki , g. b. , & lightman , a. p. 1979 , radiative processes in astrophysics ( new york : jws ) smith , d.m . , et al . 2002 , solar physics * 210 : * 33 thompson d. j. 1996 , in gamma - ray bursts : 3rd huntsville symposium , aip conf . proc 384 , eds . c. kouveliotou , m. briggs , g. fishman ( new york : aip ) waxman , e. 2003 , nature , 423 , 388 woosley , s. e. , & macfadyen , a. i. 1999 , a&as , 138 , 499 zehnder , a. , et al . 2002 , spie proc . 4853 , waikoloa , hawaii crrr 1 & 6971 & 8147 & 15118 + 2 & 24788 & 0 & 24788 + 3 & 6274 & 7387 & 13661 + 4 & 5782 & 8280 & 14062 + 5 & 6676 & 6523 & 13199 + 6 & 4306 & 7482 & 11788 + 7 & 5293 & 7500 & 12793 + 8 & 6106 & 7850 & 13956 + 9 & 3651 & 7797 & 11448 + total & 69847 & 60966 & 130813 + lcrrrr time interval & eq . ( [ t0_cut ] ) & & & & + @xmath220 and energy cut & eq . ( [ e0_cut ] ) & & & & + @xmath220 and @xmath181b@xmath21s & & 16542 & 13468.8 & 1432.0 & 1641 @xmath221 141 + @xmath52-cut & eq . ( [ dt_cut ] ) & 6030 & 3322.6 & 1172.9 & 1535 @xmath222 83 + close pairs cut & [ sec : coinc ] & 4135 & 1868.6 & 790.5 & 1476 @xmath222 69 + kinematical cut & eq . ( [ e_special ] ) & 2647 & 1093.7 & 512.6 & 1041 @xmath222 55 + no - multiples & [ sec : c_types ] & 2141 & 1081.0 & 290.3 & 770 @xmath222 49 + crrlr cb & @xmath223 & @xmath224 & @xmath225 & @xmath226 + rf & @xmath227 & @xmath228 & @xmath229 & @xmath230 + present work & @xmath231 & @xmath232 & @xmath233 & @xmath234 + `` our cb '' & @xmath235 & @xmath236 & @xmath237 & @xmath238 + `` our rf '' & @xmath239 & @xmath240 & @xmath241 & @xmath242 +

we propose and analyze several schemes for cooling bosonic and fermionic atoms in an optical lattice potential close to the ground state of the no tunnelling regime . some of the protocols rely on the concept of algorithmic cooling , which combines occupation number filtering with ideas from ensemble quantum computation . we also design algorithms that create an ensemble of defect - free quantum registers . we study the efficiency of our protocols for realistic temperatures and in the presence of a harmonic confinement . we also propose an incoherent physical implementation of filtering which can be operated in a continuous way . k = 10000

cond-mat0605198

ultracold atoms stored in optical lattices can be controlled and manipulated with a very high degree of precision and flexibility . this places them among the most promising candidates for implementing quantum computations @xcite and quantum simulations of certain classes of quantum many body systems @xcite . quantum simulation would allow us to understand physical properties of certain materials at low temperatures that so far have eluded a theoretical description or numerical simulation . however , both quantum simulation and quantum computation with this system face a crucial problem : the temperature in current experiments is too high . in this paper we propose and analyze several methods to decrease the temperature of atoms in optical lattices and thus to reach the interesting regimes in quantum simulations , as well as to prepare defect free registers for quantum computation . so far , several experimental groups have been able to load bosonic or fermionic atoms in optical lattices and reach the strong interaction regime @xcite . the analysis of experiments in the tonks gas regime indicates a temperature of the order of the width of the lowest bloch band @xcite , and for a mott insulator ( mi ) a temperature of the order of the on - site interaction energy has been reported @xcite . for fermions one observes temperatures of the order of the fermi energy @xcite . those temperatures severely restrict the physical phenomena that can be observed and the quantum information tasks that can be carried out with these lattices . one may think of several ways of cooling atoms in optical lattices . since the process of loading atoms into the lattice may lead to additional heating @xcite we focus here on methods that operate once the lattice potential has been raised . for example , one may sympathetically cool the atoms in the lattice using a second bose einstein condensate @xcite . a completely different approach is the _ filtering _ scheme of rabl _ et al . it operates in the no - tunnelling regime , transferring atoms between optical lattices so as to create a configuration with one atom per site . such a loading scheme can , however , originate holes due to imperfections in the original cloud . our analysis of filtering in the presence of a harmonic trap has shown that these holes , which are preferably located at the borders of the cloud , result in a considerable amount of entropy @xcite . here we propose and investigate several cooling schemes which overcome the limitations of filtering . the first set of schemes uses discrete operations to make atoms in different sites interact , thus concentrating the entropy on some atoms which are then expelled from the lattice . due to the similarity with quantum information processing , we term this kind of methods _ algorithmic cooling_. the second set of cooling methods combine filtering with either particle hopping or evaporative cooling techniques . all our cooling protocols are based on translation invariant operations ( i.e. do not require single site addressing ) and consider the residual harmonic confinement present in current experiments . although we will be mostly analyzing their effects on bosonic atoms , they can also be trivially generalized to fermions . this work has several spin - offs . first of all , we have designed algorithms that very efficiently remove all residual defects from a mott insulator , thus producing an ensemble of perfect registers for quantum computing @xcite . second , we propose how to create pointer atoms at the endpoints of a perfect quantum registers and show how these pointers can be used to tailor the register to a specific length . finally , since filtering is an important ingredient of all our algorithms , we have also designed a new incoherent version of filtering procedure which is better adapted to our protocols than the adiabatic scheme from ref . @xcite . the paper is organized as follows . we start in sect . [ init ] describing our system in terms of the bose - hubbard model and discussing the properties of current experimental states , such as densities , temperature and entropy . we also summarize the basic tools and figures of merit which underlie our cooling schemes . [ sect_filtering ] is devoted to a particular tool , filtering . we describe the effect of filtering on the entropy and temperature for realistic initial states and study the optimal choice of experimental parameters . in the last part we propose an implementation of filtering which is similar to a radio - frequency knife . once we have described all tools , in sect . [ sect_gscool ] we introduce several protocols for cooling to the ground state . this includes the algorithmic cooling in subsect . [ algo - cool ] and two other protocols in the following subsections . based on similar ideas , sect . [ sect_qrcool ] presents a quantum protocol that produces an ensemble of defect - free quantum registers and shows how to create pointer atoms on these registers . we conclude this work with some remarks concerning possible variants and combinations of our cooling schemes . in appendix a we present a detailed description of the numerical method that we use to simulate classically correlated states . we consider a gas of ultra - cold bosonic atoms which have been loaded into a three dimensional ( 3d ) optical lattice . this lattice is created by six laser beams of wave vector @xmath0 propagating along three orthogonal directions . if the laser light is off - resonant with any atomic transition , the ac stark effect induces a periodic potential on the atoms of the form : @xmath1 with a strength or `` lattice depth '' @xmath2 proportional to the dynamic atomic polarizability and the laser intensity . the gaussian profile of the laser beams creates an additional harmonic confinement which can be compensated by additional magnetic or optical confining elements @xcite . in the following we will mostly be concerned with one - dimensional ( 1d ) lattices . in other words we will assume that the lattice potential is so strong along two directions , @xmath3 that tunnelling is only allowed along the third one . we will also assume that the confinement along all directions is still much stronger than the atomic interaction strength . under these conditions , the atoms can be described using a single band bose - hubbard model ( bhm ) @xcite , which for a lattice of length @xmath4 reads @xmath5 .\ ] ] the parameter @xmath6 denotes the hopping matrix element between two adjacent sites , @xmath7 is the on - site interaction energy between two atoms and the energy @xmath8 accounts for the strength of the harmonic confinement . second quantization operators @xmath9 and @xmath10 create and annihilate , respectively , a particle on site @xmath11 , and @xmath12 is the occupation number operator . since the tunnelling rate decreases exponentially with the trapping strength , @xmath2 , while the on - site interaction @xmath7 remains almost constant @xcite , we have adopted this last value as the unit of energy for our work . even more important , this exponential dependence allows one to reach the strong interaction regime @xmath13 . at @xmath14 the ground state changes from a superfluid ( sf ) to a mott insulator ( mi ) @xcite . in the sf regime particles are delocalized over all lattice sites . in 1d without harmonic trap the sf ground state reads : @xmath15 where @xmath16 is the particle number . in the mi regime particles are localized at individual lattice sites . for integer filling factors @xmath17 the ground state is given by : @xmath18 the presence of a harmonic trap typically leads to a wedding cake structure of mi states with different @xmath19 . for very shallow traps one obtains only a single mi phase , which has filling @xmath20 and which is centered around the bottom of the trap . in the fock basis with respect to lattice sites this mi state can be written as : @xmath21 this state will be the target ground state for our cooling schemes . throughout this paper we will work with 1d thermal states in the grand canonical ensemble , which are characterized by two parameters : the temperature @xmath22 and the chemical potential @xmath23 . we are particularly interested in the no - tunnelling limit , which ensures that single particle eigenfunctions even at the bottom of the trap are localized on individual lattice wells @xcite . ] , @xmath24 . in this limit the hamiltonian ( [ bhm ] ) becomes diagonal in the fock basis of independent lattice sites : @xmath25 and the density matrix becomes a product of thermal states for each lattice site , @xmath26 this simplifies calculations considerably and for instance the von - neumann entropy can be written as the sum over single site entropies @xmath27 let us now study thermal states of the form ( [ rho_ind ] ) in more detail . in fig . [ fig_parameters ] we have depicted graphically all relevant energy scales in the no - tunnelling limit . the chemical potential determines the size of the cloud via the relation @xmath28 , where @xmath29 denotes the site at which @xmath30 . for @xmath31 singly occupied sites at the border of the cloud become energetically degenerate with doubly occupied sites in the center . , chemical potential @xmath23 , and harmonic confinement @xmath32 . the characteristic size of the atomic cloud is given by @xmath33 . ] the analysis of recent experiments in the mott regime @xcite reveals a substantial temperature of the order of the on - site interaction energy @xmath7 . this result is consistent with our own numerical calculations and translates into an entropy per particle @xmath34 . the particle number in a 1d tube of a 3d lattice as in @xcite typically ranges between @xmath35 and @xmath36 particles . a representative density distribution corresponding to such initial conditions ( with @xmath37 ) is plotted in fig . [ fig_initial_1d]a . ( @xmath38 in mi phase ) and equal particle number @xmath37 . at hopping rate @xmath39 ( solid ) the temperature is given by @xmath40 and at @xmath41 ( @xmath42 ) ( dashed ) one obtains @xmath43 . the harmonic confinement is fixed at @xmath44 ( @xmath45 in transverse direction ) . ( b ) the separation into two fermionic phases becomes clearly visible in the density profile of a thermal state at low temperatures . numerical parameters : @xmath46 , @xmath47 , @xmath48 , @xmath49 , ( @xmath37 , @xmath50).,title="fig : " ] ( @xmath38 in mi phase ) and equal particle number @xmath37 . at hopping rate @xmath39 ( solid ) the temperature is given by @xmath40 and at @xmath41 ( @xmath42 ) ( dashed ) one obtains @xmath43 . the harmonic confinement is fixed at @xmath44 ( @xmath45 in transverse direction ) . ( b ) the separation into two fermionic phases becomes clearly visible in the density profile of a thermal state at low temperatures . numerical parameters : @xmath46 , @xmath47 , @xmath48 , @xmath49 , ( @xmath37 , @xmath50).,title="fig : " ] in this example the inverse temperature is given by @xmath51 . since our cooling protocols lead to even lower temperatures , we will from now on focus on the _ low temperature regime _ , @xmath52 . moreover , we will only consider states with at most two particles per site , which puts the constraint @xmath53 on the chemical potential . such a situation can either be achieved by choosing the harmonic trap shallow enough or by applying an appropriate filtering operation @xcite . under the assumptions @xmath54 and @xmath55 we have shown in @xcite that the density distribution of the initial state ( [ rho_ind ] ) can be separated into regions that are completely characterized by fermionic distribution @xmath56 to be more precise , at the borders of atomic cloud , @xmath57 , the mean occupation number is given by @xmath58 with @xmath59 , while in the center of the trap , @xmath60 , one has : @xmath61 with @xmath62 and effective chemical potential @xmath63 . note also that the underlying mi phase in the center of the trap is well reproduced by the function @xmath64 , which originally has been derived for the border region . as a consequence , the density distribution for the whole lattice can be put in the simple form : @xmath65 , which corresponds to two fermionic phases i and ii , sitting on top of each other [ fig . [ fig_initial_1d]b ] . in other words , the initial state of our system can effectively be described in terms of non - interacting fermions , which can occupy two different energy bands i and ii , with dispersion relations @xmath66 and @xmath67 , respectively [ fig . [ fig_bands_2 ] ] . so far this fermionic interpretation is just based on the properties of the density profile of thermal states in the mi regime . in order to put this fermionic picture on more general grounds one can fermionize the bhm ( [ bhm ] ) directly . in @xcite we have shown that the dynamics at finite @xmath6 can effectively be described by the following fermionic hamiltonian : @xmath68 . \label{h_fermi}\end{aligned}\ ] ] here , the fermionic operators @xmath69 and @xmath70 refer to energy band i and ii , respectively . this effective description is self - consistent as long as the probability of finding a particle hole pair is negligible , i.e. @xmath71 . we have shown above that this fulfilled for thermal states at low temperatures and at negligible tunnelling . later we will see to what extent the validity can be extended to finite tunnelling rates . and @xmath72 , where @xmath11 denotes the lattice site and @xmath7 is the interaction energy . increasing the harmonic trap strength from @xmath8 to @xmath73 increases the chemical potential to @xmath74 so that the population of the upper band becomes energetically favorable . in the bosonic picture this process corresponds to the formation of doubly occupied sites . ] in general , the state of the system will however exhibit both classical and quantum correlations . in this case we have to represent the state in terms of a matrix - product state ( mps ) @xcite . this concept can also be used to compute 1d thermal states numerically @xcite . based on this method one can , for instance , estimate how the temperature of a 1d tube changes when passing from the mi to the sf regime . assuming that the process is thermodynamically adiabatic one obtains the new temperature with the following procedure . we tune the parameters of a thermal state in the sf regime until the expectation values for the entropy and particle number match the corresponding values of the initial thermal state in the mi regime . this can be illustrated with the following example . starting from a representative state in the no - tunnelling regime with @xmath75 and @xmath37 we have to tune the temperature to @xmath76 at @xmath41 ( @xmath42 ) in order to leave @xmath77 and @xmath16 unchanged [ fig . [ fig_initial_1d]a ] . hence , in the sf regime one faces a substantial temperature of the order of the width of the lowest bloch band . note , however , that this is only a lower bound to the true temperature , because our approach does not include any sort of heating processes induced by the adiabatic evolution . we will show below that algorithmic protocols are suited both for ground state cooling and the initialization of quantum registers . the goal in both cases is to create a pure state under the constraint of keeping a large number of particles . given the fact that our protocols converge to the desired family of states , we can measure the performance of the protocol by computing the mixedness of the state . this mixedness can in turn be quantified using the von neumann entropy @xmath78 ( [ vonneumann ] ) . in some cases , such as for finite hopping @xmath6 we will not be able to compute the von - neumann entropy efficiently . we will then refer to the rnyi entropy @xmath79 which is a lower bound @xmath80 and can be evaluated using mps [ appendix a ] . in order to assess the efficiency of a protocol in achieving our objectives we define two figures of merit . the ratio of the entropies _ per _ particle after and before invoking the protocol , @xmath81 , quantifies the amount of cooling . the ratio of the final and initial number of particles , @xmath82 , quantifies the particle loss induced by the protocol . note that these figures of merit can sometimes be misleading and should therefore be applied with care . in the case of ground state cooling the entropy is only a good figure of merit if the state of the system after the cooling protocol , @xmath83 , is close to thermal equilibrium . if this is not fulfilled , we use an effective thermal state , @xmath84 , with the same number of particles , @xmath16 , and energy , @xmath85 , to compute the figures of merit . for instance , the final entropy is given by @xmath86 . given that our system can somehow thermalize , these will be indeed the properties of our final state . finally , it is important to point out , that other variables , like energy or temperature , are not very well suited as figures of merit , because they depend crucially on external system parameters like the trap strength . note also that in the case of quantum registers it can be erroneous to assume that a finite value of the final entropy implies the existence of defects . for example , we will propose a protocol below that generates a state which his an incoherent superposition of perfect quantum registers with varying length and position . this state has some residual entropy but it is ideal for quantum information processing . all our cooling protocols rely on a set of translationally invariant quantum operations that can be realized in current experiments with optical lattices . these operations are : _ ( i ) particle transfer : _ depending on the occupation numbers , an integer number of particles is transferred between internal states @xmath87 and @xmath88 . this process can be described by the unitary operation @xmath89 are fock states with @xmath90 and @xmath91 atoms in internal states @xmath87 and @xmath88 , respectively . note that for this unitary operator it holds @xmath92 . certain operations , like @xmath93 or @xmath94 , have already been demonstrated experimentally in an entanglement interferometer @xcite . _ ( ii ) lattice shifts : _ we denote by @xmath95 the operations which shift the @xmath88 lattice @xmath96 steps to the right . for example , @xmath97 transforms the state @xmath98 into @xmath99 are removed from the system . we denote this operation by @xmath100 . it transforms the state @xmath101 or by coupling this state resonantly to an untrapped state . _ ( v ) filtering : _ this means particle transfer operations of the form @xmath102 , followed by @xmath100 , for all @xmath103 . after tracing out the subsystem @xmath88 , the filter operation is described by a completely positive map acting solely on atoms in state @xmath87 : @xmath104 : & & \sum_{n , m } \rho_{n , m}|n\rangle\langle m| \to\\ & & \sum_{n , m\leq m}\rho_{n , m}|n\rangle\langle m| + \sum_{n > m}\rho_{n , n } |m\rangle\langle m| . \nonumber\end{aligned}\ ] ] the first proposal for a coherent implementation of the operation @xmath105 [ fig . [ fig_p1_lattice ] ] appeared in ref . more recently , a scheme based on resonant control of interaction driven spin oscillations has been put forward @xcite . in @xcite we have proposed an ultra - fast coherent implementation of @xmath105 relying on optimal laser control . in this paper we will propose an incoherent realization of filtering , which has the special virtue that it can be applied continuously . which aims at producing a uniform filling of one atom per site . defects arising from holes can not be corrected . ] we now study the filtering operation @xmath105 in eq . ( [ fn ] ) in more detail . this operation is especially relevant for cooling because it produces a state close to the ground state of the mi regime . thus , it can serve as a benchmark which has to be beaten by alternative cooling schemes . in this section we begin summarizing the analytical results about filtering from ref . @xcite and discuss the conditions to reach optimal cooling efficiency . we close this section with a possible experimental implementation of filtering which is based on an incoherent coupling of atoms to a continuum of untrapped states . and entropy @xmath78 before ( left ) and after ( right ) the application of the filter operation @xmath105 . the final particle distribution can be well described by eq . ( [ nf ] ) ( dashed line ) . numerical parameters for initial state : @xmath106 , @xmath107=1 , @xmath108 , @xmath109 and @xmath48 . figures of merit : @xmath110 and @xmath111 . ] in fig . [ fig_p1 ] we have depicted the particle and entropy distributions before and after one filtering step , @xmath105 . one observes that a nearly perfect mi phase with filling factor @xmath20 is created in the center of the trap . defects in this phase are due to the presence of holes and concentrate at the borders of the trap . this behavior is reminiscent of fermions , for which excitations can only be created within an energy range of order @xmath112 around the fermi level . this numerical observation can easily be understood with our previous analysis of the initial state . filtering removes phase ii , which is due to doubly occupied sites , and leaves the fermionic phase i unaffected [ fig . [ fig_p1]a ] . the fermionic picture allows us to to find a simple estimate for the final entropy . from the expression for the density of states , @xmath113 , one immediately obtains that the number of states ( lattice sites ) within an energy range of @xmath114 around the chemical potential @xmath23 is given by : @xmath115 this parameter is characteristic for the tail width of the density distribution [ fig . [ fig_p1 ] ] and will in the following be central for the analysis of our protocols . since the final entropy must be localized at these sites we expect that @xmath116 . indeed , the rigorous derivation in @xcite yields @xmath117 , with @xmath118 . here , we have introduced the final number of particles @xmath119 , which can be well approximated by the parameter @xmath120 : @xmath121 hence , the final entropy per particle can be written as : @xmath122 which reflects the size of the region of defects @xmath123 in units of the system size @xmath120 . for the special choice @xmath124 ( or equivalently ) one finds the following expressions for our figures of merit @xcite : @xmath125 with numerical parameters @xmath126 and @xmath127 . this result shows that filtering becomes more efficient with decreasing temperature , since @xmath128 and @xmath129 for @xmath130 . it is important to note that the state after filtering is not an equilibrium state , because it is energetically favorable that particles tunnel from the borders to the center of the trap , thereby forming doubly occupied sites . however , the system can easily be brought to equilibrium by adapting the trap strength . while tunnelling is still suppressed , one has to decrease the strength of the harmonic confinement to a new value @xmath73 , with @xmath131 . the system is then in the equilibrium configuration @xmath132 ( [ nf ] ) with rescaled parameters @xmath133 and @xmath134 . this observation shows that it is misleading to infer the cooling efficiency solely from the ratio @xmath135 , because it depends crucially on the choice of @xmath73 . let us now study how the cooling efficiency of filtering depends on the initial state variables @xmath8 , @xmath136 and @xmath23 . since the initial temperature is dictated by the experimental setup , we consider only the trap strength @xmath8 and the chemical potential @xmath23 ( which can be varied via the particle number @xmath16 ) as free parameters . since our figures of merit are computed per particle we expect a weak dependence on @xmath16 and therefore focus on the @xmath8 dependence . this can be studied in terms of the mean central occupation number @xmath137 . for instance , in the special case @xmath138 , we have obtained the analytical expression ( [ cool_eff ] ) for the ratio @xmath81 , which exhibits a strong temperature dependence . in contrast , for @xmath139 one finds that the cooling efficiency becomes independent of the temperature : @xmath140 for @xmath52 . this can be understood from the presence of a @xmath141 mi phase in the center of the trap , which does not contain entropy . in the opposite regime @xmath142 the protocol has no cooling effect at all . for general @xmath143 , we have computed the figures of merit numerically exact [ fig . [ fig_p1_opt ] ] . for fixed initial entropy @xmath107 and particle number @xmath144 as a function of the harmonic confinement . the change of the trap is indicated by either the mean central occupation number @xmath143 or the inverse trap strength @xmath145 . we plot the figures of merit @xmath81 ( solid ) , @xmath82 ( dashed ) and the weighted quantity @xmath146 ( dotted ) . the initial parameters are @xmath147 and ( a ) @xmath148 , ( b ) @xmath149 , ( c ) @xmath150 and ( d ) @xmath151 . ] one observes that for initial entropies @xmath152 a central filling @xmath153 is always close to optimal . a special situation arises when we approach zero temperature . as shown in fig . [ fig_p1_opt]d , the quantity @xmath81 changes periodically when decreasing the trap strength @xmath154 . in this regime an additional mi phase with @xmath141 is present in the center of the trap and the splitting of the entropy between the lower and upper fermionic band depends dramatically on the value of the harmonic confinement . in particular , one can find trap strengths ( e.g. @xmath155 in fig . [ fig_p1_opt]d ) at which the final entropy approaches zero and only a comparatively small number of particles is lost ( @xmath156 ) . for shallower traps the @xmath141 mi phase starts to collapse . in this regime it becomes very difficult to find proper thermal states which match the initial conditions in terms of entropy and particle number . finally , the central occupation approaches one and the cooling protocol leaves the initial state unchanged . it is also interesting to study the efficiency of filtering when acting on the full 3d lattice . in this case only a small subset of 1d tubes will satisfy the optimal initial conditions . using the parameters of sect . [ init ] ( @xmath149 and @xmath157 ) we obtain that @xmath158 and @xmath159 . the relatively large particle loss results from the high densities in the center of the trap . this is also the main reason why the protocol performs worse compared to a 1d tube with @xmath149 as shown in fig . [ fig_p1_opt]b . filtering constitutes a fundamental tool in all our cooling protocols . as shown before @xcite filter operations can be realized by coherently transferring particles between two internal states and then removing these particles . what we will show below is that both processes can be combined producing an incoherent evolution that gives rise to the completely positive map @xmath105 ( [ fn ] ) . are resonantly coupled to a continuum of untrapped states and thereby removed from the system . ] the experimental procedure for achieving the map @xmath160 is very simple . we will use two atomic states : one atomic state shall be confined by an optical lattice , deep in a mi phase , while the other one will be in a continuum of untrapped states which are free to escape the lattice . we will couple the trapped and untrapped states with two raman lasers which have a relative detuning of the order of the interaction energy , @xmath161 [ fig . [ fig_levels ] ] . as long as the coupling is active , the lasers will depopulate lattice sites which have too many atoms , @xmath162 , while leaving other sites untouched . if the untrapped states are such that they have few atoms and they are quickly expelled from the trap ( for instance by a magnetic field gradient ) , these states will behave for practical purposes like a thermal bath in a vacuum state . if the coherence time of this bath , physically determined by the time free atoms spend close to the trapped ones , is very short compared to the rabi frequency @xmath163 of the raman transition , we will be allowed to write a master equation for the trapped atoms . the solution of this equation converges at large times to the desired filtered state , e.g. @xmath164 for @xmath105 . conceptually , this mechanism is equivalent to the frequency knife from evaporative cooling , where atoms containing too much kinetic energy are expelled from the trap in order to lower the temperature . in our case , however , it is the interaction energy we get rid off and , as a side effect , we make the filling of the lattice more uniform . compared to the optimal control scheme in @xcite , the operation of this much simpler method is not very fast . from the solution of the master equation it follows that states with occupation @xmath165 decay after a time which is of the order of the inverse rabi frequency , @xmath166 . the main source of errors arises from the non - resonant coupling of the @xmath167 state with the reservoir . the probability of a defect ( empty site ) will approximately be given by : @xmath168 . hence , for @xmath169 we get an operation time @xmath170 which is comparable to the time scale of the adiabatic scheme @xcite . however , our incoherent scheme has two big advantages . first , it can be applied continuously . second , it does not put any constraints on the interaction energies of the two species . this holds under the assumption that untrapped atoms are expelled so quickly from the trap , that they do not interact significantly with the trapped atoms . we have seen that the residual entropy after filtering is concentrated at the borders of the density distribution . particles on these sites are also the only source of energy excitations because all doubly occupied sites have been removed . in the following we will propose several protocols which selectively remove particles at the borders , thereby bringing the system closer to its ground state . . ( a ) operation @xmath171 : particles are transferred to state @xmath88 ( red ) . ( b ) operation @xmath172 : lattice @xmath88 has been shifted @xmath173 sites to the right . ( c ) iteration of the sequence : @xmath174 , @xmath175 , @xmath97 . if a site is occupied by two particles of different species , then both particles are removed . afterwards lattice @xmath88 is shifted one site to the left . after @xmath176 iterations lattice @xmath88 is shifted @xmath177 sites to the left . ( d ) the same sequence as in ( c ) is applied , with the difference that lattice @xmath88 is shifted to the right . the final distribution of atoms in state @xmath87 is sharper compared to the initial distribution ( dotted ) . numerical parameters : @xmath106 , @xmath178 ( @xmath149 ) , @xmath179 , @xmath173=42 , @xmath180 , @xmath181 , @xmath182 . after equilibration : @xmath183 ( @xmath184 ) . ] this is a cooling method based on a set of discrete steps which resemble a quantum information processing protocol . the steps are the following : ( i ) we begin with a cloud in thermal equilibrium in the no - tunneling regime , having two or less atoms per site , all in internal state @xmath87 . this can be ensured with a filtering operation @xmath185 . ( ii ) we next split the particle distribution into two , with an operation @xmath171 [ fig . [ fig_gs_cool]a ] . ( iii ) the @xmath87 and @xmath88 atoms are shifted away from each other until both ensembles barely overlap . we then begin moving the clouds one against each other , removing in the process atoms from doubly occupied sites . experimentally , this can be achieved by introducing a third internal level @xmath186 and applying the unitary operation @xmath174 in generalization of ( [ u_nm ] ) . this sequence sharpens the density distribution of both clouds and it is iterated for a small number of steps , of order @xmath123 [ fig . [ fig_gs_cool]c ] . ( iv ) the atoms of type @xmath88 are moved again to the other side of the lattice and a process similar to ( iii ) is repeated [ fig . [ fig_gs_cool]d ] . ( v ) remaining atoms in state @xmath88 can now be removed . notice that the final particle distribution will never be perfectly sharp [ fig . [ fig_gs_cool]d ] because this protocol is intrinsically limited by thermal and quantum fluctuations in the initial state . qualitatively , in order to remove a particle of type @xmath87 from the tail , it has to coincide with a particle of type @xmath88 from the other cloud . errors arise when particles do not coincide and are thus proportional to the fluctuations of the density . if we want to clean about @xmath123 sites or remove @xmath123 particles , the errors will be @xmath187 . in the limit @xmath52 , we obtain the following scaling behavior for the final entropy per particle @xmath188 where @xmath120 ( [ n_bar ] ) is the number of particles in @xmath87 after step ( ii ) and @xmath144 is the initial particle number . this simple estimate is already in good agreement with the exact result in @xcite , which takes also particle losses into account . our findings show that the algorithmic protocol becomes more efficient with increasing initial particle number . this is in contrast to filtering , for which @xmath189 is independent of @xmath144 . , @xmath190 ( @xmath191 ) , @xmath192 , @xmath193 , @xmath194 , @xmath195 , @xmath196 ; after equilibration : @xmath197 ( @xmath198 ) ; dashed line : @xmath199 , @xmath200 , @xmath201 . ( b ) final entropy per particle @xmath189 as a function of the initial particle number @xmath144 for fixed @xmath150 ( circles ) . for large @xmath144 one obtains a @xmath202 dependence ( solid ) . for comparison we plot @xmath189 after filtering ( diamonds ) which is expected to be independent of @xmath144 ( dotted ) . small variations in @xmath189 can be attributed to the specific choice of the harmonic confinement.,title="fig : " ] , @xmath190 ( @xmath191 ) , @xmath192 , @xmath193 , @xmath194 , @xmath195 , @xmath196 ; after equilibration : @xmath197 ( @xmath198 ) ; dashed line : @xmath199 , @xmath200 , @xmath201 . ( b ) final entropy per particle @xmath189 as a function of the initial particle number @xmath144 for fixed @xmath150 ( circles ) . for large @xmath144 one obtains a @xmath202 dependence ( solid ) . for comparison we plot @xmath189 after filtering ( diamonds ) which is expected to be independent of @xmath144 ( dotted ) . small variations in @xmath189 can be attributed to the specific choice of the harmonic confinement.,title="fig : " ] we have verified numerically the scaling ( [ sf_gscool ] ) [ fig . [ fig_gs_cool_sqrtn]b ] . the numerical simulation is by no means trivial because the protocol establishes classical correlations both among lattice sites and also among internal states . to reproduce these correlations we have resorted to representing the classical density matrices using mps [ appendix a ] . these numerical simulations show that the final density distribution is very close to a thermal distribution [ fig . [ fig_gs_cool_sqrtn]a ] . indeed if we apply one or few iterations of the protocol , the final state will still contain some holes on the tails and be close to thermal equilibrium , as the computation of the rnyi entropy entropy @xmath203 ( [ renyi ] ) before and after equilibration shows [ fig . [ fig_gs_cool_sqrtn]a ] . on the other hand , for a large number of iterations , the final density matrix will be an incoherent superpositions of perfect uniform mi states which differ on their length and position [ similar to fig . [ fig_p1a_mps ] ] . these states are suitable for quantum computation but are far from equilibrium . it is clear that we could cool the atoms to the ground state with @xmath204 efficiency if the density distributions of @xmath87 and @xmath88 atoms were perfectly correlated . it seems also that we can improve the performance in realistic situations by removing the classical correlations which are established during the process [ fig . [ fig_gs_cool_sqrtn]a ] . we also increase the performance if we break the correlations between the atomic species at the end of the protocol while leaving the inter site correlations untouched . each iteration of the protocol would then reduce the total entropy by a factor of @xmath205 @xcite . from an experimental point of view , this means that the algorithm will work better when using multiple independent clouds to clean each other . these clouds may come from loading the lattice with atoms in different internal states , from splitting the lattice into multiple condensates , or simply by using the clouds from different 1d tubes to clean each other@xmath206 many other possibilities can be conceived . the protocol can be further improved by selecting specific particle number subspaces from the density matrix before restoring thermal equilibrium ( see sect . [ sect_qrcool ] for details ) . note also that we need not get rid of the atomic cloud in state @xmath88 . for appropriate initial trap parameters , it is preferably to move this cloud back to the center and to pump all atoms back into internal state @xmath87 . the resulting state will exhibit a mi shell structure with two plateaus at densities @xmath167 and @xmath207 . this is a simple way to engineer ground states of tighter traps . finally , let us remark that this protocol will only correct inhomogeneities along one direction . in a real experiment the protocol should be repeated along the other three directions . we have seen that algorithmic schemes perform much better than simple filtering . here , we propose two non - algorithmic ground state cooling schemes which combine filtering with other quantum operations available in optical lattices . the basic idea of this protocol is to first transfer particles from the tails of the cloud to the center of the trap , to form doubly occupied sites , and then to remove these particles with some filtering operation . to be more precise , one has to iterate the following sequence of operations : ( i ) allow for some tunnelling while the trap is adjusted in order to reach a central occupation of @xmath208 . ( ii ) suppress tunnelling and perform the filtering . ( iii ) slightly open the trap so that final distribution resembles a thermal one ( see sect . [ sect_filtering ] ) . these steps are repeated sequentially until the entropy per particle converges . this protocol has been analyzed in detail in @xcite and the results can be summarized as follows . if one allows for equilibration then a state asymptotically close to the ground state can be reached after only a few filtering cycles . without equilibration the final entropy is limited by the initial probability of finding a defect in the center of the trap . , in terms of the rnyi entropy , of filtering @xmath105 as a function of the hopping rate @xmath209 . for the numerical computation in terms of mps we use a lattice of length @xmath210 , which typically contains @xmath211 particles . the trap strength is chosen to be @xmath212 and @xmath124 , yielding @xmath213 . ] alternatively , one can think of an implementation of sequential filtering that operates at fixed but non - zero hopping rate . such a protocol would clearly profit from the fact that adiabatic changes of the lattice potential , which lead to additional heating , are not required . a key ingredient is the generation of doubly occupied sites in the center . one has to fix the hopping rate at a value , at which there exists a coupling between doubly occupied sites at the center and singly occupied sites at the borders . our analysis of the bhm in terms of an effective two - band fermionic model @xcite shows , however , that this occurs typically only deep in the sf regime for @xmath214 . we have studied the sequential filtering at fixed @xmath6 numerically , computing the cooling efficiency as a function of the hopping rate . the results are shown in fig . [ fig_dynamic_3 ] . for high temperatures the ratio @xmath81 is rather independent of @xmath6 and we achieve some cooling . for very small temperatures , @xmath215 , it changes dramatically with @xmath6 . this is a clear signature of the quantum phase transition , which is expected to occur at @xmath216 in the thermodynamic limit . while for @xmath217 the state is well described in terms of independent wells and filtering works very efficiently , for @xmath218 particles are delocalized over the lattice , and filtering causes heating rather than cooling . summing up , since we require a significant value of the hopping , @xmath219 , this second variant of sequential filtering can only be used as an initial step , when temperatures are still comparatively high . particles located at the tails of the density distribution can also be removed with a method similar to evaporative cooling or a frequency knife . the idea is to make use of inhomogeneous on - site energies and to choose the detuning @xmath220 of a radiation field in such a way that only atoms located at specific lattice sites are resonantly coupled to another internal state . using a magnetic field gradient it has been demonstrated that individual lattice sites can be resolved within an uncertainty of about five sites @xcite . in our case the spatial dependence of the on - site energies is naturally provided by the harmonic trapping potential . however , in order to make use of this inhomogeneity one has to ensure that the atoms are coupled to an internal state which responds differently to the ac - stark shift induced by the lattice laser beams . experimentally , this can , for instance , be achieved with a setup similar to the one for the creation of spin - dependent lattices @xcite . there , one atomic species is trapped exclusively by @xmath221 polarized laser light , whereas the other species is trapped predominantly by @xmath222 polarized light . hence , an optically untrapped internal state can simply be realized by using only @xmath222 polarized laser light for creating the optical lattice . let us now estimate the requirements on the the rabi frequency @xmath163 of the transition depending on the trap strength @xmath8 . for simplicity , we consider two internal states , and a configuration for which the excited state exhibits zero on - site energy at each lattice site . we are interested on removing particles which are typically located at a distance @xmath29 from the center of the trap . hence , for resonant coupling of these atoms we must choose the detuning to be @xmath223 . particles at sites @xmath224 feel an effective detuning @xmath225 , where @xmath226 denotes the particle number ( [ n_bar ] ) . the probability that a particle at site @xmath224 is transferred to the excited state is then given by : @xmath227 in order to locally address the site @xmath29 reasonably well one has to demand : @xmath228 . for typical harmonic trapping frequencies in the mi regime , @xmath229 hz @xcite , and @xmath230 this translates into @xmath231 khz . experimentally it should therefore be feasible to resolve individual sites with an uncertainty of a few lattice sites and thus to sharpen the density profile within this uncertainty . note that this scheme clearly profits from a large number of particles per tube . another advantage is that it preserves the spherical symmetry of the density distribution in a 3d lattice . finally let us remark that this method can easily be incorporated in the continuous filtering scheme [ fig . [ fig_levels ] ] proposed in sect . [ sect_filtering ] , since there the atoms are also coupled to untrapped states . a perfect one - dimensional quantum register is a connected array of commensurately filled lattice sites . for most purposes the filling factor is @xmath20 . this state appears naturally as the unique ground state of the bhm in the no - tunnelling limit and in the presence of harmonic confinement . hence , any efficient ground state cooling protocol will produce a good quantum register . moreover , we have seen in the previous section , that algorithmic protocols can be used to create an _ ensemble _ of quantum registers rather than a unique one ( see also @xcite ) . in this section we will propose and analyze two alternative algorithms for the creation of a quantum register ensemble . as compared to previous proposals @xcite these schemes require only a small number of operations , which makes them more appealing for experimental implementation . the first protocol produces registers with filling @xmath20 , whereas the second protocol is optimized for filling @xmath141 . in addition , we propose how to transform atoms at the endpoints of each register into `` pointer '' atoms , which then enables addressing individual lattice sites . this can be used , for instance , to create registers of equal length . even more important it offers the opportunity to perform ensemble quantum computation in this system . we start from a rather cold cloud , @xmath52 , which has been subject to fundamental filtering operations @xmath160 ( [ fn ] ) . the result is an almost perfect mi in the center of the trap with some residual defects ( or holes ) which are predominantly localized at the borders [ see fig . [ fig_p1 ] ] . since we operate solely in the no - tunnelling regime these defects can not redistribute nor evaporate . our goal is to remove all these defects and we will achieve it by applying nearest - neighbor quantum gates which simulate inelastic collisions between particles and holes . simply put , whenever a particle sits next to a hole , the particle will be annihilated . this process is analogous to spin flips in ferromagnets and the formation of domains of equal magnetization . thus , these algorithmic schemes can also be understood as controlled equilibration and cooling of defects . for all protocols that will be proposed below , the resulting state is a mixture of perfect ( up to defects in the central mi phase ) quantum registers , which differ only in their length and lateral position . the entropy of this state will be of the order @xmath232 , where the parameter @xmath123 ( [ delta ] ) quantifies the translational uncertainty in the initial density distribution . note , however , even though the final entropy is very small , these protocols are typically not suited for ground state cooling . the reason is that the final state is far from thermal equilibrium . numerically we find that the value of the entropy after equilibration is comparable to the value before invoking the protocol . however , if one makes use of the pointer atoms to select only registers with a specific length then these protocols indeed lead to cooling . the operation sequence is illustrated in fig . [ fig_p1a_lattice ] . we first merge oddly aligned pairs of sites using a superlattice @xcite and empty sites with only one atom . the sites are again split and the original lattice is restored . this operation is then repeated several times , alternating between even and oddly aligned sites , until the _ total _ entropy reaches a minimum . we have analyzed the performance of this protocol under realistic conditions using the mps description to compute the at most classically correlated density matrix [ appendix a ] . in fig . [ fig_p1a_mps ] we plot the typical density distribution after different steps of the protocol . a single step changes very little the density but decreases dramatically the entropy ( from @xmath233 to @xmath234 ) . indeed , the first steps account for the elimination of most defects . after a few iterations the value of the total entropy starts to saturate because the density matrix has collapsed to a classical ensemble of ( almost ) defect - free quantum registers [ fig . [ fig_p1a_mps ] ] . as a consequence of the protocol , the number of atoms per register must be even , which leads to steps in the final density profile [ fig . [ fig_p1a_mps ] ] . ( dotted ) , after one iteration ( dashed ) , after eight iterations ( solid ) . parameters of the initial state : @xmath235 , @xmath236 ( @xmath237 ) , @xmath145=1000 , @xmath238 . parameters after filtering : @xmath239 , @xmath240 ( @xmath233 , @xmath241 ) . parameters after iteration 1 : @xmath242 , @xmath243 , @xmath244 ; iteration 2 : @xmath245 , @xmath246 , @xmath247 ; iteration 7 : @xmath248 , @xmath249 , @xmath250 ; iteration 8 : @xmath251 , @xmath252 , @xmath253 . stopping the algorithm at iteration 3 gives a minimum in the entropy _ per _ particle : @xmath254 . upon reaching thermal equilibrium this value has increased to @xmath255 ( @xmath256 ) which is comparable to the value after filtering . however , when selecting the subspace containing only registers of length @xmath257 then the protocol leads to cooling . the entropy per particle in thermal equilibrium will then be given by @xmath258 . right : the step like structure of the density distribution allows one to deduce the states which contribute significantly to the final density matrix . the states can be classified according to their particle number and their lateral position . , title="fig : " ] ( dotted ) , after one iteration ( dashed ) , after eight iterations ( solid ) . parameters of the initial state : @xmath235 , @xmath236 ( @xmath237 ) , @xmath145=1000 , @xmath238 . parameters after filtering : @xmath239 , @xmath240 ( @xmath233 , @xmath241 ) . parameters after iteration 1 : @xmath242 , @xmath243 , @xmath244 ; iteration 2 : @xmath245 , @xmath246 , @xmath247 ; iteration 7 : @xmath248 , @xmath249 , @xmath250 ; iteration 8 : @xmath251 , @xmath252 , @xmath253 . stopping the algorithm at iteration 3 gives a minimum in the entropy _ per _ particle : @xmath254 . upon reaching thermal equilibrium this value has increased to @xmath255 ( @xmath256 ) which is comparable to the value after filtering . however , when selecting the subspace containing only registers of length @xmath257 then the protocol leads to cooling . the entropy per particle in thermal equilibrium will then be given by @xmath258 . right : the step like structure of the density distribution allows one to deduce the states which contribute significantly to the final density matrix . the states can be classified according to their particle number and their lateral position . , title="fig : " ] clearly the fix point of this protocol would be a state with no particles at all . let us now estimate how many iterations @xmath259 of the protocol are required to reach a state with reasonable particle number and tolerable defect probability . first , we have to point out that there are two sources of defects : ( i ) holes in the central mi phase and ( ii ) particles at the borders which are disconnected from the central mi phase and which have not yet been erased by the protocol . the probability for defects of the first kind is negligible in the limit of low temperatures @xmath260 [ see sect . [ sect_initstate ] ] . defects of kind ( ii ) can be assessed by the following observation : in order to erase a connected array of @xmath259 particles , which is separated by at least one empty site from a central mi state , one requires exactly @xmath261 iterations of the protocol . the probability of finding defects after @xmath259 iterations is then given by the probability of having states with array size larger than @xmath259 in the initial density matrix . this probability can easily be computed numerically exact from the distribution ( [ nf ] ) . the derivation of closed expressions is however difficult . nevertheless , one can get a good estimate for the optimal number of iterations based on the following argument . the characteristic tail width of the initial particle distribution ( [ nf ] ) is @xmath123 ( [ delta ] ) . hence , the occurrence of arrays of size @xmath262 is already very unlikely . this implies that after roughly @xmath123 iterations of protocol 1 we expect to have registers with negligible defect probability . this can be illustrated with an example . for the initial state in fig . [ fig_p1a_mps ] we have @xmath263 and the relative difference in the total entropy after the seventh and eighth iteration has reduced to @xmath264 , which implies a defect probability of the same order of magnitude . the typical size of the registers after @xmath123 iterations is roughly @xmath265 , where @xmath120 ( [ n_bar ] ) is the characteristic size of the initial mi state . this implies @xmath266 , otherwise no particles remain in the system . since @xmath267 , one has to require the low temperature regime @xmath52 . as mentioned above , the algorithmic ground state cooling protocol of the previous section can also be used for the creation of quantum registers . however , the ground state cooling algorithm involves @xmath268 operations as compared to @xmath269 operations of the current protocol . note also that the example presented in fig . [ fig_p1a_mps ] indicates that this protocol does not lead to cooling . after restoring thermal equilibrium , the entropy has reached a value comparable to the one after filtering . we have repeated this analysis for various initial conditions and our results confirm this observation . however , if one selects a particle number subspace with @xmath16 being larger than the mean value than our protocol can indeed be used for cooling . according to the data in fig . [ fig_p1a_mps ] the entropy per particle can be reduced by roughly @xmath270 as compared to filtering . we start from a state which contains only empty or doubly occupied sites . this can be achieved by applying filtering operation @xmath185 , followed by @xmath94 and @xmath100 . the protocol is depicted in fig . [ fig_p2a_lattice ] . we begin with transferring one particle per site to state @xmath88 . then the @xmath88lattice is shifted one site to the left and the same operation as before is performed . afterwards one empties single occupied sites . this procedure allows to remove defects in a correlated way . occupied sites which have an unoccupied site to the right are emptied . since the probability of finding particles in the center is close to one , the central mi is preserved except for losses at the right border . this procedure is repeated until all particle disconnected from the central mi phase become annihilated and only perfect mi phases in the center remain . one step of the protocol can be summarized in the following sequence of operations : @xmath171 , @xmath271 , @xmath171 , @xmath100 , @xmath94 , @xmath100 . following the discussion of protocol 1 , this sequence has to be applied approximately @xmath272 times , where @xmath123 denotes again the characteristic tail width of the initial particle distribution . the factor two stems from the fact that at each step of the protocol the size of `` particle islands '' in the tails , as well as the central mi phase , is reduced only by one as compared to two in protocol 1 . the final density matrix looks very similar to the one after protocol 1 , with the difference that also registers with odd number of atoms appear . given an ensemble of quantum registers we will now show how to create pointer atoms . to be more precise , our goal is to selectively transfer the two end atoms of each register to a different internal level . these pointer atoms enable single site addressing which can used to engineer registers of specific length . our scheme relies on the same set of operations that is used in current experiments for entangling atoms located at different lattice sites @xcite . we consider quantum registers with one atom per site . initially all atoms are in internal state @xmath87 . a hadamard transformation puts the atoms in the coherent superposition state latexmath:[$(|a{\rangle}+ @xmath88lattice one site to the right and waits an appropriate time until the on site interaction between species @xmath87 and @xmath88 yields a collisional @xmath274phase . this means that on each site the state @xmath275 is transformed into @xmath276 . after two lattice shifts to the left one waits again until a @xmath274phase has built up . then the lattice is shifted back to its original position and a second hadamard operation is performed . the resulting state is a again a product state but the end atoms have been promoted to level @xmath88 . these atoms can be considered as pointer atoms because they mark the beginning ( and the end ) of each register and can thus be used to access every site within the register in a deterministic way . in practice , only one pointer atom is needed . the second pointer atom , e.g. the one on the left , can easily be removed by applying the following operation sequence : @xmath97 , @xmath277 , @xmath278 , @xmath171 . the pointer atoms can now be used to create an ensemble of registers of fixed length , a feature which is desired for quantum computation and which can even lead to cooling . we first present a protocol that requires only two internal states of the atoms . then we show that the algorithm can be simplified considerably when a third internal level is included . in both cases we start from a situation where all registers have one pointer atom in state @xmath88 which is located at the right most occupied site of the register . * protocol based on two internal states : * the central idea is to remove atoms selectively from the system by transferring them to the pointer level @xmath279 . we first show how to discard all registers which are shorter than a desired length @xmath259 . we start with the sequence : @xmath280 , @xmath281 , @xmath282 , @xmath278 . this ensures that registers of length @xmath283 are protected against further modifications , because their pointer atoms have been removed . next we promote atoms in doubly occupied sites to the pointer level and shift this ( two atom ) pointer one site to the right . if the pointer hits an occupied site then one atom in @xmath87 will be removed . iteration of this process removes all atoms on the right of the current pointer position . to be more precise , one has to iterate @xmath284 times the sequence : @xmath93 , @xmath271 , @xmath282 , @xmath285 , @xmath100 . in a last step the remaining doubly occupied sites are emptied via @xmath93 , @xmath100 . after creating new pointer atoms the minimal register length is given by @xmath286 . again we keep only the pointer atom on the right side . let us now show how to shorten registers of length @xmath287 to length @xmath288 . we first apply the sequence : @xmath97 , @xmath280 , @xmath289 , @xmath290 , @xmath97 . for the target register of length @xmath288 this merely transfers one atom from the right end of the register to the left end . for larger registers one obtains a two atom pointer that can now be used to select and discard all atoms located left of the pointer . this can be accomplished by iterating the sequence : @xmath285 , @xmath100 , @xmath93 , @xmath97 until all registers with appreciable weight in the density matrix have been shortened to the desired size . * protocol based on three internal states : * this protocol is based on the operation @xmath291 , which transfers an atom from @xmath87 to @xmath186 , given that a pointer atom is present . this way one can use the pointer as a `` marker '' , which allows one to produce registers of desired length @xmath259 in a very simple manner one first marks an array of @xmath259 atoms and then discards all atoms which have remained in level @xmath87 . the algorithm can be summarized as follows : apply the sequence @xmath292 ; iterate @xmath284 times the sequence @xmath293 ; finally apply the operation @xmath294 . we have proposed various ground state cooling schemes that allow to reduce the temperature in current optical lattice setups considerably . in particular , we have presented algorithmic cooling schemes which combine filtering with concepts inspired by ensemble quantum computation . we have shown that these algorithmic protocols can be designed both for ground state cooling and the creation of defect - free quantum registers . we have tested and analyzed our protocols under realistic experimental conditions , with special emphasis on the presence of a harmonic trap . we found that the residual entropy of all our protocols depends crucially on one parameter which accounts for the translational uncertainty of the initial particle distribution . a special virtue of our schemes is , that they rely on general concepts which can easily be adapted to different experimental situations . for instance , little modifications ensure that our protocols can be applied both to bosonic and fermionic systems . a second advantage of our protocols is , that they are designed for the no tunnelling regime and hence do not necessarily require equilibration processes induced by particle hopping and elastic collisions . hence , they allow to approach the ground state of the no - tunnelling regime in very short times . in this sense , the collection of cooling schemes presented in this article can be considered as a toolbox which is tailored for cooling atoms in optical lattice setups . the tool ( or combination of tools ) which is best suited for a given purpose , can be chosen according to the characteristic features of a specific experimental setup . for instance , in the case of large systems at high temperatures , one can think of combining filtering with the frequency knife method which is then followed by the algorithmic ground state cooling protocol . or ground state cooling can be combined with adiabatic transformation of the hamiltonian so as to produce ground states of models different from the simple bose - hubbard considered here . and one should also keep in mind that a 3d lattice structure offers a large variety of possibilities , which have not been fully explored yet . the methods introduced here greatly enhance the potential of optical lattice setups for future applications and might pave the way to the experimental realization of quantum simulation and quantum computation in this system . we also hope that the concepts introduced in this work might trigger further research in the direction of ground state cooling in optical lattices . this work was supported in part by the eu ist projects quprodis and covaqial , the dfg ( sfb 631 ) and the `` kompetenznetzwerk quanteninformationsverarbeitung der bayerischen staatsregierung '' . our algorithmic protocols establish classical correlations between different lattice sites , when applied to thermal states in the no - tunnelling regime . hence , a description in terms of independent wells ( as in ( [ rho_ind ] ) ) is no longer adequate . therefore we refer to a representation in terms of matrix product states ( mps ) @xcite . to be more precise , for a 1d lattice of length @xmath4 we want to map a classical density matrix of the form @xmath295 onto a pure state in mps form : @xmath296 here , @xmath297 denote matrices of dimension @xmath298 and is the occupation number of site @xmath11 . the matrices at the endpoints , @xmath299 and @xmath300 , are @xmath301 and @xmath302 vectors , respectively . the mapping from @xmath303 to @xmath304 can easily be accomplished by setting : @xmath305 . local operations on @xmath303 , like unitary operations @xmath308 ( [ u_nm ] ) or filter operations ( [ fn ] ) , amount to transformations of the local matrices @xmath309 in the mps picture . for illustration , let us consider a completely positive map @xmath310 which acts on a local state @xmath311 at site @xmath11 . in kraus representation one can write @xmath312 with the kraus operators @xmath313 satisfying the completeness relation @xmath314 . the mps matrices transform according to @xmath315 non - local operations that involve more than one lattice site are more complicated to implement . as an example , let us consider the most complicated case , which occurs in our protocols : a spin - dependent lattice shift @xmath97 which shifts the lattice @xmath88 one site to the left . for this we have to consider two species of atoms . in generalization of ( [ mps ] ) , the mps matrices @xmath316 are now labelled with two physical indices , @xmath317 and @xmath318 , referring to states @xmath319 and @xmath88 , respectively . it will turn out to be convenient to rewrite these matrices in tensor form : @xmath320 , with @xmath321 . to begin with let us consider only the first two lattice sites . we start out with tracing over subsystem @xmath88 at the first site : @xmath322 , because we want to omit sites which pass the system boundaries after shifting . the induced error is negligible , given that sites at the boundaries are not populated . we then multiply the matrix of the second site and perform the lattice shift : @xmath323 , where we use einstein summation convention . after relabelling @xmath324 we perform a singular value decomposition : @xmath325 with @xmath326 being a diagonal square matrix and @xmath327 . the new mps matrix for site @xmath328 can then be defined as : @xmath329 accordingly , we define @xmath330 and iterate this scheme until the end of the lattice . note , that the dimension @xmath331 of the mps matrices can , in principle , increase exponentially with the number of lattice shifts . if @xmath331 becomes larger than a desired value @xmath332 one has to resort to a truncation method similar to the one proposed for mixed quantum states @xcite . however , in practice , we find that @xmath331 increases only linearly with the number of lattice shifts , given that each shift is followed by a filter operation . thus , with our method we can fairly easy simulate protocols exactly , i.e. without truncation , which involve up to @xmath333 lattice shifts on lattices with up to @xmath334 sites .

we study the transport properties of a four - terminal corner junction made by etching a two - dimensional topological insulator to form a quantum point contact ( qpc ) . the qpc geometry enables inter - boundary tunneling processes allowing for the coupling among states with different helicity , while the tight confinement in the qpc region activates charging effects leading to the coulomb blockade physics . peculiar signatures of these effects are theoretically investigated using a scattering field theory modified to take into account the electron - electron interaction within a self - consistent mean - field approach . the current - voltage characteristics and the current fluctuations ( noise ) are derived beyond the linear response regime . universal aspects of the thermal noise of the corner junction made of helical matter are also discussed .

1406.4819

the recent discovery of topological insulators ( tis ) , both in three and in two dimensions , represents a promising advancement towards quantum technology of matter@xcite . in fact , microscopic laws of physics are invariant under time reversal , but the transport of energy and information in real devices is an irreversible process . this irreversibility originates from the device - environment coupling and limits , for instance , the possibility of quantum computation . the tis , bulk gapped materials exhibiting conducting channels at the boundaries@xcite , represent a possible technological platform to limit the decoherence induced by dissipative phenomena . indeed , the peculiar electronic behavior predicted in devices based on topological materials , which can be designed as quantum spin - hall systems@xcite , could provide efficient spin injection protocols avoiding the complexity of working with nanoscale ferromagnetism@xcite . principles found in tis could inspire the next - generation spintronic devices operating with low power consumption and/or performing topological protected quantum computation . a specific example of two - dimensional ti is the hgte / cdte quantum well which has been demonstrated , both theoretically and experimentally , to have a single pair of helical edge states leading to a quantized conductance plateau when the fermi energy lies in the bulk gap @xcite . quantized transport along the hgte boundaries can be conveniently explained by an edge channel picture ( fig . [ fig : fig1 ] ( b ) ) : two states with opposite spin orientation propagate along opposite device edges in the same direction thus leading to a quantization@xcite of the conductance in unit of @xmath0 . + at the aim of exploiting topological protection and ballistic transport , several proposals of ti - based spin - transistor have been presented , being the working principles based on the sensitivity of the surface currents to the magnetization direction of a thin ferromagnetic barrier@xcite , the aharonov - bohm and fabry - prot interferometry @xcite , the field effect of a single gated hgte nanoconstriction @xcite . + the simplest non - trivial ti - based device to probe a variety of transport properties exhibited by edge states can be obtained by creating a narrow constriction , i.e. a qpc , etched on a _ quantum spin hall insulator bar _ as depicted in fig . [ fig : fig1 ] ( a ) . near - equilibrium spin and charge transport properties of similar two- and four - terminal geometries have been studied both in the non - interacting@xcite and in the helical luttinger liquid regime@xcite . the emphasis of the previous studies was mainly on the conductance and the zero - frequency noise . recently , finite frequency noise has been studied@xcite putting in evidence that tunneling processes of single- or two - particle contribute differently to the current fluctuations , the latter observation being important in discriminating multiple - quasiparticle ( agglomerate ) tunneling events in correlated systems . + however , in order to compare the theoretical models with the experiments performed on real devices one needs to take simultaneously into account non - equilibrium effects , non - perturbative inter - edge coupling , charging effects and coulomb blockade physics , non - linear transport regimes . while most of these requirements are commonly met in free fermion models described by scattering approaches , the charging effects are completely missed . on the other hand , when interaction effects are taken into account inter - edge coupling is considered at perturbative level , while the effect of the interaction is modeled as uniformly distributed along the edge modes . the latter condition seems to be unlikely in real systems where charging effects are relevant only in the qpc region , i.e. where the tight confinement of the electronic densities enables unscreened coulomb interaction effects . furthermore , the perturbative nature ( with respect to the inter - edge coupling ) of the interacting theory limits its use to the linear response regime . + the aim of the present work is to give a comprehensive description of the interplay of all the above mentioned factors in the transport properties of a two - dimensional topological insulator by studying a minimal model of a four - terminal topological corner - junction . we give distinctive features of correlations and interference processes among helical edge states , useful for future experiments . in particular , our analysis , based on a modified scattering field theory , is focussed on a qpc geometry obtained by etching a two - dimensional ti giving rise to a 4-terminal corner junction [ fig . [ fig : fig1 ] ( a ) ] . in this geometry inter - edge tunneling processes take place , while the tight confinement in the qpc region activates charging effects leading to a coulomb - blockade - like physics . we describe all the peculiar signatures of these effects in the charge current and its fluctuation even beyond the linear response regime and suggest how to control edge - state transport by means of all electrical gating ( side - gate control ) . + the organization of the paper is the following . in sec . [ sec : ham ] we introduce an effective one - dimensional model to describe edge states at the surface of a two - dimensional topological insulator and their coupling in the qpc geometry . here we present the model of four - terminal setup and its operational configuration . a discussion on the symmetry - preserving mean - field approximation of the interacting part of the hamiltonian is also provided in subsec . [ ssec : spmfapp ] . in sec . [ sec : scattering - field ] we introduce the scattering field approach and the basic formalism used to calculate the current - voltage curves and the current - current correlations in terms of the scattering matrix elements , with emphasis on the role of the charging energy effect on the qpc . the modified boundary conditions and the self - consistency constraints are presented in subsec . [ ssec : bcs - self ] . results are presented in sec . [ sec : results ] . in particular , in subsec . [ ssec : iv - diff - cond ] we show the results concerning current - voltage characteristics and the differential conductance , with particular care in defining the properties of the coulomb blockade regime . in subsec . [ ssec : cur - cur - noise ] we introduce the general expression of the finite - frequency current - current correlation which also includes finite temperature effects , non - linear regime and non - equilibrium , non - local information and interaction effects . thermal noise ( [ ssec : thermal - noise ] ) , local ( [ ssec : transport - fluct ] ) and non - local ( [ ssec : non - local - transport - fluct ] ) transport - induced fluctuations are carefully discussed and compared to the non - interacting case . finally , we discuss our conclusions in sec . [ sec : concl ] . in the following we formulate a minimal model describing the electronic transport in the corner junction depicted in fig.[fig : fig1 ] ( a ) . in this system the confinement effects at the qpc locally activate the spin - orbit interaction and cause unscreened charging effects ; while the former is responsible for spin - flipping tunneling effects , the latter gives origin to the coulomb repulsion and possibly coulomb blockade physics . we thus consider an effective one - dimensional hamiltonian @xmath1 describing edge states localized along the top and bottom boundaries of a quantum well defining a two - dimensional topological insulator and pinched to form a qpc : @xmath2 @xmath3 point - like constriction ( i.e. linear dimension @xmath4 , @xmath5 being the fermi wave length ) obtained by etching a two - dimensional topological insulator . a side - gate ( @xmath6 ) can be used to tune the electronic density in the constriction , while a four - terminal geometry is considered . @xmath7 sketch of the bands structure of a two - dimensional topological insulator . helical states , having linear energy - momentum dispersion , describe the system as long as the particle energy ranges inside the bulk insulating gap @xmath8 . coulomb interaction effects are assumed to be relevant only in the constriction region ( qpc).,title="fig : " ] @xmath7 point - like constriction ( i.e. linear dimension @xmath4 , @xmath5 being the fermi wave length ) obtained by etching a two - dimensional topological insulator . a side - gate ( @xmath6 ) can be used to tune the electronic density in the constriction , while a four - terminal geometry is considered . @xmath7 sketch of the bands structure of a two - dimensional topological insulator . helical states , having linear energy - momentum dispersion , describe the system as long as the particle energy ranges inside the bulk insulating gap @xmath8 . coulomb interaction effects are assumed to be relevant only in the constriction region ( qpc).,title="fig : " ] where@xcite : @xmath9 @xmath10 represents the right ( left ) mover electron annihilation operator with spin @xmath11 , while @xmath12 stands for the normal ordering of the operator @xmath13 with respect to the equilibrium state defined by occupied energy levels below the fermi sea . in our description , without loss of generality , we assume that spin-@xmath14 right movers @xmath15 and spin-@xmath16 left movers @xmath17 flow along the top boundary while the spin-@xmath16 right movers @xmath18 and spin-@xmath14 left movers @xmath19 flow along the bottom boundary . at the nanoconstriction ( @xmath20 ) the edge modes are coupled by transverse confinement effects and the overlap between edge states belonging to different boundaries leads to the twofold effect , one of producing a local reactivation of the spin - orbit coupling@xcite and the other one of enabling coulomb repulsion effects . indeed , at the qpc ( @xmath20 ) inter - boundary tunneling events may take place and the only terms which preserve time - reversal symmetry@xcite can be distinguished in spin - preserving and spin - flipping tunneling described by the following contributions to the hamiltonian : @xmath21 @xmath22 where @xmath23 , @xmath24 is the chirality , while @xmath25 are the space - dependent tunneling amplitudes : @xmath26 . here the local modification of the spin - orbit coupling@xcite at @xmath20 governed by @xmath27 produces an effective spin - flipping mechanism . concerning the electron - electron interaction , the only terms that preserve the time - reversal symmetry of the system are the dispersive @xmath28 and the forward scattering @xmath29 where @xmath30 denotes the electron density with @xmath31 and spin @xmath32 . thus , in eq . ( [ eq : hamiltonian ] ) , the coulomb repulsion originated by unscreened charges at the qpc is accounted for @xmath33 . the local nature of the interaction suggests that the coupling functions @xmath34 , @xmath35 are different from zero only at the qpc , while the case of pure coulomb repulsion requires @xmath36 . accordingly , we assume @xmath37 and @xmath38 , with @xmath39 , where @xmath40 is the area of the qpc . in the presence of side gates @xmath41 acting at top and bottom sides of the qpc , an additional term appears in the hamiltonian : @xmath42 where the local action of the gate is modeled by @xmath43 , being @xmath44 the length of the qpc . + as a final comment , we notice that the delta - like nature of the potentials introduced in @xmath45 does not limit the generality of our model as long as the low energy regime is considered . indeed , an incoming electron of momentum @xmath46 is not able to resolve the internal structure of the qpc region if the relation @xmath47 is fulfilled . thus within the low energy regime ( @xmath48 ) our results are not affected by the simplified assumptions we made , while beyond this regime model - dependent features are expected . however , the energy scale @xmath49 , estimated by assuming @xmath50 m / s and @xmath51 nm , takes a value which is several times bigger than the topological insulator bulk gap @xmath8 ( @xmath52 ev , @xmath53 being the qpc length expressed in nm ) . thus the low energy condition which ensures the generality of the model is always satisfied as long as we consider the energy range in which the topological edge states represent the relevant modes involved in the charge transport . the model of corner junction described so far contains the term @xmath45 which is difficult to be treated within the scattering matrix formalism . since we are interested in describing the coulomb blockade physics , which is a fundamentally classical effect , it is sufficient a mean field treatment of the interaction . in performing this , however , a symmetry preserving approximation is required . indeed , the original hamiltonian describes a time reversal invariant system , which is a property to be preserved after the approximation has been made . in general the mean field treatment substitutes the interaction @xmath45 with a single particle operator of the form @xmath54 . the time reversal operator @xmath55 , being @xmath56 the conjugation operator , produces the field transformation @xmath57 , with @xmath58 , and the notation @xmath59 ( @xmath60 ) denoting the chirality ( spin projection ) opposite to @xmath61 ( @xmath32 ) . as a consequence the approximated interaction hamiltonian preserves the time reversal invariance , i.e. @xmath62 , provided that @xmath63 and @xmath64 . the standard mean field approximation of @xmath45 leads to the single particle operator @xmath65 , where @xmath66 are electrostatic potentials describing the coulomb repulsion originated by the electron densities @xmath67 at the qpc . these potentials , to be determined self - consistently , have the appropriate symmetry properties ( i.e. @xmath68 and @xmath69 ) to make @xmath70 a time reversal preserving approximation of @xmath45 . thus in the following we make the substitution @xmath71 . we now formulate a scattering field theory _ la _ bttiker@xcite able to describe coherent spin and charge transport in the system shown in fig . [ fig : fig1 ] ( a ) . + the charge or spin current operators @xmath72 in first quantization are written as follows : @xmath73 where @xmath74 stands for the pauli matrix , @xmath75 is the identity matrix acting on the spin sub - space of the hilbert space given by the tensor product @xmath76 ( @xmath77 , @xmath78 ) . to build a scattering field theory , one first defines the scattering field corresponding to each terminal @xmath79 in terms of the incoming ( @xmath80 ) and outgoing ( @xmath81 ) electron operators , according to : @xmath82\\\nonumber \hat{\psi}_2(x , t ) & = & \int de \frac{e^{-iet/\hbar}}{\sqrt{h v_f}}\bigl[\hat{a}_{r\uparrow}(e;x)+\hat{b}_{l\downarrow}(e;x)\bigl]\\\nonumber \hat{\psi}_3(x , t ) & = & \int de \frac{e^{-iet/\hbar}}{\sqrt{h v_f}}\bigl[\hat{a}_{l\downarrow}(e;x)+\hat{b}_{r\uparrow}(e;x)\bigl]\\\nonumber \hat{\psi}_4(x , t ) & = & \int de \frac{e^{-iet/\hbar}}{\sqrt{h v_f}}\bigl[\hat{a}_{l\uparrow}(e;x)+\hat{b}_{r\downarrow}(e;x)\bigl],\end{aligned}\ ] ] where @xmath83 with @xmath84 and the wavevector @xmath46 ( and similarly for @xmath85 ) . + then , the second - quantized current operators @xmath86 in the terminal @xmath87 are explicitly given by ( @xmath88 ) : @xmath89,\ ] ] where @xmath90 , @xmath91 , @xmath92 and @xmath93 . in writing eq . ( [ eq : current_op ] ) we made use of the fourier transform @xmath94 $ ] , while the following correspondence has been made : @xmath95^t=[b_{l\uparrow } , b_{l\downarrow } , b_{r\uparrow } , b_{r\downarrow}]^t$ ] , @xmath96^t=[a_{r\downarrow } , a_{r\uparrow } , a_{l\downarrow } , a_{l\uparrow}]^t$ ] . + the expectation value @xmath97 can be computed making use of the scattering relation @xmath98 and from quantum statical average @xmath99 , being @xmath100 the fermi - dirac distribution with electrochemical potential @xmath101 . after direct computation we get : @xmath102f_{j}(e).\ ] ] the charge current fluctuations are described by the operators @xmath103 and can be computed within the scattering formalism . in this context @xmath104 , while the current - current correlation function@xcite @xmath105 contains non - local information on the spectral density of the current fluctuations , @xmath106 , being the fourier transform defined as @xmath107 $ ] . the non - local spectral density @xmath108\times\\\nonumber & & \sum_{m,\mu}\mathcal{a}^{(i)}_{m \mu}(e , e')\mathcal{a}^{(j)}_{\mu m}(e , e')f_{m}(e)(1-f_{\mu}(e'))+\\\nonumber & & + ( i \rightarrow j , j \rightarrow i ; \tau \rightarrow -\tau),\end{aligned}\ ] ] is explicitly written in terms of scattering matrix elements according to the relation : @xmath109.\ ] ] eq . ( [ eq : spectral_density ] ) contains information on the fluctuation properties of the charge currents flowing through the junction and contains the equilibrium noise contribution ( thermal noise ) and the non - equilibrium ( shot - noise ) term . thus effects of finite bias , temperature and frequency @xmath110 are simultaneously taken into account in our formulation . here it is worth to mention that in our notation the kirchhoff s law is written as follows : @xmath111 due to our choice of the area vectors encircling the lead regions . ( [ eq : kl ] ) takes the form @xmath112 when the substitution @xmath113 is made . as the effect of the mentioned substitution the spectral density is redefined accordingly : @xmath114 . although the physical equivalence of the area vector conventions , in the following we will work within the new conventions ( @xmath115_{new}= \epsilon_{i}(-)^{i}[\hat{\mathbf{j}}^{(i)}_{c}]_{old}$ ] , @xmath116_{new}=\epsilon_{i}\epsilon_{j}(-)^{i+j}[\mathcal{s}_{ij}(\tau)]_{old}$ ] ) to better compare our results with the existing literature . + in order to capture non - equilibrium charging effects at the qpc a self - consistent evaluation of @xmath117 is required . in our model , however , interaction effects are approximated by ultralocal potentials and thus the mean field terms @xmath117 are determined by the electron densities @xmath118 at the qpc . within the scattering formalism the particle density at the qpc can be evaluated according to the formula : @xmath119 where the non - equilibrium distribution @xmath120f_{i}(e)$ ] accounts for the perturbation of the occupation number of a scattering state due to the presence of the qpc , while @xmath121 represents the one dimensional density of states . as will be clear in the following discussion , the scattering matrix of the problem is energy - independent and all the integrals present in the theory can be performed analytically . in the following discussion , however , simplifying approximations are made in order to put in evidence the physics behind the formalism . in particular , in the experimental relevant temperature range ( low temperature regime ) eq . ( [ eq : nedistr ] ) is well approximated by its zero temperature value@xcite and thus the charge density imbalance at the qpc takes the form : @xmath122,\ ] ] where @xmath123 is the quantum capacitance of the junction , while @xmath124 represents the charge density imbalance induced by a shift of the electron density with respect to the equilibrium value @xmath125 , where @xmath126 is the chemical potential and @xmath8 is the bulk gap . ( [ eq : charge_imb ] ) shows that changing the electrochemical potential of the four terminals of the same amount @xmath127 produces the charge density imbalance @xmath128 . this is equivalent to the field effect produced by a back - gate acting below the whole hetero - structure . the scattering matrix @xmath129 is a four by four unitary matrix whose diagonal entries vanish by helicity and time - reversal symmetry while all the other entries can be explicitly determined as a function of the tunneling amplitudes @xmath130 by imposing the proper boundary conditions ( bcs ) on the wave functions@xcite . since the wave functions are continuous in the regions @xmath131 and @xmath132 , we only have to impose the matching conditions where dirac delta potentials are present , i.e. at @xmath20 . by using the equation of motion of the quantum field @xmath133 , i.e. @xmath134\psi_{\alpha\sigma}(x)+\\\nonumber & & + 2\hbar v_f\bigl[\gamma_{sp}\psi_{\bar{\alpha}\sigma}(x)+\gamma_{sf}\xi_{\alpha}\psi_{\alpha\bar{\sigma}}(x)\bigl]\delta(x)=0,\end{aligned}\ ] ] with @xmath135+u_{g},\ ] ] being @xmath136 the interaction parameter [ notice that @xmath137 and @xmath138 , and explicitly taking into account the properties of the dirac delta potential under integration , one obtains the following matching conditions ( @xmath139 , @xmath140 ) : @xmath141+\gamma_{sp}[\psi^{(+)}_{\bar{\alpha}\sigma}+\psi^{(-)}_{\bar{\alpha}\sigma}]+\\\nonumber & + & \xi_{\alpha}\gamma_{sf}[\psi^{(+)}_{\alpha\bar{\sigma}}+\psi^{(-)}_{\alpha\bar{\sigma}}]+u^{(+)}_{\alpha\sigma}\psi^{(+)}_{\alpha\sigma}+u^{(-)}_{\alpha\sigma}\psi^{(-)}_{\alpha\sigma}=0.\end{aligned}\ ] ] in eq . ( [ eq : bcs ] ) we make the substitution @xmath142/2 $ ] which is equivalent to consider the spatial average @xmath143 of the self - generated coulomb potential computed by using the values taken just after and before the constriction . due to the time reversal symmetry we have @xmath144 and @xmath145 , while the mean field self - consistency is provided by the equations @xmath146+u_g\\\nonumber & & u_{2}=\frac{u}{2}\sum_{\mu=\pm}[\langle\rho_{r\downarrow}\rangle^{(\mu)}+\langle\rho_{l\uparrow}\rangle^{(\mu)}]+u_g.\end{aligned}\ ] ] the explicit form of eqs . ( [ eq : self ] ) in terms of the scattering matrix elements is given by the nonlinear equations in the unknown parameters @xmath147 and @xmath148 : @xmath149 being the @xmath150 matrix expression reported in appendix [ app : a ] . ( [ eq : self2 ] ) must be solved self - consistently since the matrix elements of @xmath150 depends on @xmath151 and @xmath148 themselves through the scattering matrix elements ( see eq . ( [ eq : m ] ) of appendix [ app : a ] ) . the dimensionless parameters @xmath152 represent the electrochemical potential shift imposed to the i - th lead as the effect of the voltage bias and are defined by @xmath153 , while the dimensionless quantity @xmath154 is related to the field effect induced by the side - gate @xmath41 . the scattering matrix elements , which completely define the self - consistency relations , can be determined making the substitutions @xmath155 within the eqs . ( [ eq : bcs ] ) also taking into account the scattering relation @xmath156 . the scattering matrix @xmath157 $ ] determined according to the above procedure is given in eq . ( [ eq : sm ] ) of appendix [ app : b ] and parametrically depends on the self - consistency parameters @xmath147 and @xmath148 to be determined by solving the nonlinear problem given by eqs . ( [ eq : self2 ] ) . the scattering matrix @xmath157 $ ] depends explicitly on the details of the qpc encoded by the tunnel probabilities @xmath158 and presents an hidden dependence on the voltage bias . importantly , it does not depend on the energy of the scattering processes . it is also worth to mention that eq . ( [ eq : sm ] ) provides a system description beyond the linear response theory and correctly reproduces the limit of negligible interaction ( i.e. @xmath159)@xcite and of uncoupled boundaries ( i.e. @xmath160 , @xmath161 ) . in the latter case the scattering matrix describes the reflectionless transport of helical states moving along separate boundaries and its non - vanishing elements are simply the phase factors @xmath162 and @xmath163 . in particular the renormalization of the tunneling amplitudes promoted by the interaction is well captured by the present formalism . for instance , supposing @xmath164 and assuming @xmath165 as a small perturbation , we get @xmath166 and @xmath167 where the renormalized tunneling parameter is related to its bare value by the relation @xmath168 . the renormalized tunneling describes a lowering of inter - edges scattering probability induced by the interaction ( i.e. @xmath169 ) . + this behavior is in qualitative agreement with a renormalization group argument on the spin preserving and spin - flipping scattering within a helical luttinger liquid model@xcite . when these terms become relevant , the tunneling amplitudes depend as a power - law on the luttinger parameter , which is linear in the coulomb interaction @xmath170 , for small @xmath170 . + once the scattering matrix has been determined , the charge current and its fluctuation can be computed . in the following we present the current - voltage characteristics and the current fluctuations assuming that the chemical potential is located in the middle of the topological insulator bulk gap , i.e. @xmath171 . as a consequence , the energy shift imposed by the voltage bias to the local electrochemical potentials takes the form @xmath172 , while the dimensionless energy shifts @xmath173 are chosen in the interval @xmath174 - 1,1[$ ] , the latter condition ensuring that only topological edge states are involved in the charge transport . within a real topological heterostructure going beyond this limit implies that the charge current is partially supported by states ( not included in the theory ) having energy above the insulating gap . + for a qpc of nanometric size ( @xmath175 nm ) , the dimensionless parameter @xmath176 can be estimated to be of order @xmath177 . the latter estimate comes from identifying @xmath178 with the charging energy @xmath179 , being @xmath180 the capacitance of the nanometric constriction varying in the af range . we study the current flowing through the system assuming the @xmath181 limit which is appropriate to describe experiments performed in thermal bath with temperature ranging from tens of mk to few kelvin . under this condition , since the scattering matrix does not depend on the energy , the charge current @xmath182 flowing through the i - th lead can be written as @xmath183v_{j},\ ] ] where @xmath184 is used as current unit . in particular , if the value of the bulk gap is @xmath185 , @xmath186 takes values in the na range , i.e. @xmath187 . despite the linear response form of eq . ( [ eq : charge_curr ] ) , the hidden dependence on the applied voltages of the scattering matrix determines non - linear current - voltage characteristics . the deviation from the ohmic behavior and the current suppression at low bias originates from the coulomb repulsion at the qpc and represents a clear signature of the coulomb blockade regime . since the electron density within the constriction can be tuned by using the side - gate @xmath188 , the coulomb blockade regime can be switched on and off by using all - electrical means@xcite . in order to put in evidence the renormalization effects induced by the coulomb interaction , we focus on a crossed voltage - bias configuration for which the leads @xmath189 and @xmath190 are grounded ( i.e. @xmath191 ) , while a finite voltage drop @xmath192 is applied to the leads @xmath193 and @xmath194 , i.e. @xmath195 and @xmath196 . accordingly , the explicit expression of the charge currents @xmath197 and @xmath198 in terms of the scattering matrix elements can be written in the form : @xmath199\frac{v}{2}\\ \label{eq : curr - scatt2 } & & i_{ch}^{(2)}=\frac{e^2}{h}\bigl[1+|s_{24}|^2\bigl]\frac{v}{2},\end{aligned}\ ] ] while the charge conservation implies that @xmath200 and @xmath201 . the usefulness of the crossed bias configuration in recognizing the renormalized junction parameters is evidenced in eqs . ( [ eq : curr - scatt1])-([eq : curr - scatt2 ] ) . indeed , one expects that @xmath198 is essentially linear with respect to the applied bias since @xmath202 is only a weak perturbation , while important renormalization effects and nonlinear behavior are expected for @xmath197 . ( full line ) and @xmath198 ( dashed line ) as a function of the voltage drop @xmath192 . the remaining model parameters have been fixed as follows : @xmath203 , @xmath204 , @xmath205 and @xmath206 . the charge conservation implies @xmath200 and @xmath201 , so that @xmath207 is respected . here and in figs . 3 , 4 , 7 , 8 we set @xmath208 to shorten the axis label . ] in fact , in the limit of small inter - edge coupling , i.e. @xmath209 , and @xmath164 , one obtains @xmath210\frac{v}{2}\\ & & i_{ch}^{(2)}\approx\frac{e^2}{h}\frac{v}{2}.\end{aligned}\ ] ] the functional form of @xmath197 is useful in identifying the renormalized tunneling amplitude @xmath211 as a function of the applied bias when experimental data are analyzed . this is confirmed by fig . [ fig : fig2 ] , where the charge currents @xmath197 ( full line ) and @xmath198 ( dashed line ) are represented as a function of @xmath192 , while fixing the remaining parameters as shown in the figure caption . the @xmath197 _ vs _ @xmath192 curve shows a clear coulomb blockade region at low bias , while an ohmic behavior is detected for @xmath198 . as a function of the applied voltage @xmath212 and of the side - gate voltage @xmath188 . the junction parameters are : @xmath203 , @xmath204 , @xmath213 . darker areas represent lower conductance values , while each contour line is labeled by the corresponding conductance value . at low bias , the coulomb blockade region is evident . ] to further characterize the coulomb blockade regime , in fig . [ fig : fig3 ] we show the normalized differential conductance @xmath214 as a function of the applied voltage @xmath212 and of the side - gate voltage @xmath188 , while the remaining parameters are fixed as done in fig . [ fig : fig2 ] . the minimum conductance value is obtained for @xmath215 and @xmath206 , while the conductance progressively increases and the linearity of the current - voltage curve is gradually recovered as the voltage of the side gate @xmath188 is moved from the value -3 . we have also verified that for different values of @xmath216 the coulomb blockade region is always centered at a side - gate value @xmath217 . this feature is evident in fig . [ fig : fig4 ] where the differential conductance @xmath218 is shown as a function of the bias voltages for @xmath219 ( upper left panel ) , @xmath220 ( upper right panel ) , @xmath221 ( lower left panel ) and @xmath222 ( lower right panel ) , while the junction parameters have been fixed as @xmath203 , @xmath204 . at low bias ( @xmath223 ) , the coulomb blockade regime is induced by side - gate voltages @xmath224 $ ] , and the extension of the coulomb blockade region is of the order of @xmath225 along the @xmath188 axis . the coulomb blockade regime vanishes for side - gate values fulfilling the condition @xmath226 . apart from specific aspects related to the multi - terminal nature of the device , here we notice that @xmath227 plays the same role of the critical voltage @xmath228 defining the coulomb diamond extension in the conventional coulomb blockade theory . furthermore , despite the zero - temperature limit considered here , the transition from coulomb blockade and conducting regime as a function of @xmath192 is not sharp ( see for instance the behavior of @xmath197 close to @xmath229 in fig . [ fig : fig2 ] ) . this behavior originates from the multi - terminal nature of the system which allows a non - thermal smearing of the current - voltage curves due to the current leakage towards a different electrode . thus , the current leakage plays the same role of the thermal fluctuations ( at finite temperature ) in removing the coulomb blockade . as a function of the applied voltage @xmath212 and of the side - gate voltage @xmath188 . the junction parameters are : @xmath203 , @xmath204 , while @xmath219 for the upper left panel , @xmath220 for the upper right panel , @xmath221 for the lower left panel and @xmath222 for the lower right panel . the coulomb blockade regime is removed when @xmath230 verifies the condition @xmath226 . ] exploiting the energy independence of the scattering matrix of the problem , we can write the finite frequency spectral density of the current fluctuation @xmath231 in the following form : @xmath232,\end{aligned}\ ] ] where we introduced the integral function @xmath233,\end{aligned}\ ] ] which can be analytically determined . the frequency resolved quantity @xmath231 [ see eq . ( [ eq : spectral ] ) ] takes simultaneously into account finite temperature and finite bias effects and presents a non - linear dependence with respect to the applied voltage . thus , within this framework , non - equilibrium effects under non - linear response of the system can be fully accounted for . in the following we focus on the zero - frequency noise component @xmath234 . experimentally , @xmath234 is determined from the flat region of the spectrum ( frequencies above few tens of khz ) where one can neglect the contribution of @xmath235 noise , which is instead dominant at low frequency . since the flicker noise completely masks any frequency dependence of @xmath231 , the only experimentally accessible information is encoded in the zero frequency fluctuation . in the absence of applied bias ( @xmath236 ) thermally activated particles can gain a sufficient amount of energy to escape from a given electrode . as a consequence , within a given time interval , a particles number fluctuation is possible even in the absence of a net dc current . these equilibrium processes , which can be of local and non - local nature , are mathematically characterized by taking the zero - frequency limit of eq . ( [ eq : spectral ] ) assuming a grounded configuration of the electrodes@xcite ( @xmath236 ) . using the time reversal symmetry ( @xmath237 ) and the conservation of the helicity ( @xmath238 ) , the local component of thermal noise detected in the ith lead can be written as @xmath239 while the non - local component ( @xmath240 ) of the thermal fluctuation takes the form @xmath241 both the local and non - local components of the thermal fluctuation have a multiplicative universal factor @xmath242 and a factor depending on the insulating gap @xmath8 . at low temperature , however , @xmath243 approaches @xmath189 and thus the local component of the thermal noise becomes a universal quantity which does not depend on the conductance of the corner junction . the latter property is specific of the helical nature , while in conventional materials ( not - helical ) the thermal noise is affected by the not universal value of the junction conductance . the temperature @xmath244 below which @xmath245 becomes universal depends on the value of the insulating bulk gap of the topological insulator . in fig . [ fig : fig4 ] we report the temperature behavior of the local component of the thermal noise normalized to @xmath246 for @xmath247 mev , @xmath248 mev , @xmath249 mev . the considered values of @xmath8 are , respectively , appropriate for the two - dimensional topological insulators alsb / inas / gasb / alsb , cdte / hgte / cdte and @xmath250/graphene/@xmath250 @xcite . . the universal behavior of the thermal noise is reached below a certain temperature @xmath244 which is related to the value of the bulk insulating gap of the topological insulator assumed as energy cut - off of the one - dimensional theory . different values of @xmath8 are considered : @xmath251 mev , appropriate for alsb / inas / gasb / alsb ; @xmath252 mev , appropriate for cdte / hgte / cdte ; @xmath253 mev , appropriate for @xmath250/graphene/@xmath250@xcite . ] as shown in fig . [ fig : fig5 ] , the universal character of @xmath245 becomes evident as the ratio @xmath254 increases , i.e. when the bath temperature decreases below a certain gap - dependent temperature @xmath244 [ ] . in the case of a cdte / hgte / cdte nanostructure ( @xmath255 mev ) , the helical nature of the system can be probed at a bath temperature @xmath256 k , which is much higher than the mk range usually required to verify the emergence of a topological phase in the quantum well . accordingly , the measure of the thermal noise @xmath245 of the junction contains information on the helical nature of the system , the latter being already evident at bath temperature of few kelvin . + information on the junction parameters @xmath158 can instead be deduced from the non - local component of the thermal fluctuations described by @xmath257 , @xmath240 . for instance , a corner junction with negligible reactivation of the spin - orbit coupling within the qpc region is described by @xmath164 and presents negligible non - local fluctuations between a couple of electrodes only connected by spin - flipping tunneling processes , i.e. @xmath258 . on the other hand , for @xmath259 , the ratio between the correlation functions @xmath260 is a temperature independent quantity depending on the relative frequency of spin - preserving and spin - flipping tunneling events . at sufficiently low temperature @xmath261 and thus the non - local thermal fluctuation can be studied as a function of the side - gate voltage @xmath230 . the latter analysis is performed in fig . [ fig : fig6 ] where @xmath261 _ vs _ @xmath188 curves are shown . , @xmath262 , @xmath263 as a function of the side - gate voltage @xmath188 . the junction parameters are : @xmath203 , @xmath204 , @xmath213 . lower panel : @xmath262 as a function of the side - gate voltage @xmath188 . the different curves are obtained by fixing @xmath203 and @xmath213 , while setting the spin - flipping strength to @xmath164 ( top curve ) , 0.05 , 0.1 , 0.15 ( bottom curve).,title="fig : " ] + , @xmath262 , @xmath263 as a function of the side - gate voltage @xmath188 . the junction parameters are : @xmath203 , @xmath204 , @xmath213 . lower panel : @xmath262 as a function of the side - gate voltage @xmath188 . the different curves are obtained by fixing @xmath203 and @xmath213 , while setting the spin - flipping strength to @xmath164 ( top curve ) , 0.05 , 0.1 , 0.15 ( bottom curve).,title="fig : " ] as a general remark , we observe that all the non - local correlations shown in fig . [ fig : fig6 ] are negative . this is consistent with the fact that in thermal equilibrium and at vanishing frequency @xmath264 , not only the particles flux is conserved but also its fluctuation . since an increasing flux at one terminal must be compensated by a decreasing of the current at another terminal , non - local correlations must be negative ( or at best zero ) . this argument fails under special conditions only in case of a bosonic system . + in the upper panel of fig . [ fig : fig6 ] we show @xmath265 , @xmath262 , @xmath263 as a function of the side - gate voltage @xmath188 , while fixing the remaining parameters as @xmath203 , @xmath204 , @xmath213 . thus we expect that , under coulomb blockade regime ( @xmath206 ) , the non - local correlations @xmath266 present a maximum or a minimum depending on the couple of terminals considered . this behavior is directly related to the scattering amplitudes dependence on the side - gate voltage . indeed , the amplitude @xmath267 is of order of one far from the coulomb blockade region , while it is reduced when the coulomb blockade is reached . under the latter condition , the amplitudes @xmath268 and @xmath269 are fed in order to preserve the charge conservation law . the dependence of @xmath270 on the spin - flipping tunneling rate @xmath27 is shown in the lower panel of fig . [ fig : fig6 ] . in that figure , the @xmath271 _ vs _ @xmath188 curves present a minimum for @xmath206 ( coulomb blockade regime ) which becomes deeper when the value of @xmath27 is increased . since @xmath272 for @xmath164 , non - local correlations provide a direct measure of the spin - orbit interaction reactivation in the qpc region . + thus an accurate analysis of the thermal fluctuations allows a complete characterization of the junction parameters and put in evidence the helical nature of the system . we now consider current fluctuations in the presence of a steady state current and assume the zero - temperature limit . this regime is suitable for characterizing the non - equilibrium topological phase at the mk temperature range . under these assumptions and considering the low - frequency regime @xmath264 , the diagonal terms @xmath273 of the integral function in eq . ( [ eq : spectral ] ) are negligible and thus the transport fluctuations can be written as @xmath274 where @xmath275 is well approximated by the expression @xmath276.\ ] ] once again we recall that eq . ( [ eq : spectral_tr ] ) is non - linear with respect to the applied bias due to the hidden bias dependence of the scattering matrix , while the low - frequency correlations sum rule @xmath277 has been numerically verified with accuracy within the numerical error ( @xmath278 in dimensionless units ) . the statistical properties of current fluctuations in the ith lead can be described by introducing the fano factor @xmath279 . fano factor of 1 indicates poissonian fluctuation processes , while sub - poissonian ( super - poissonian ) fluctuations are characterized by @xmath280 ( @xmath281 ) . within the crossed bias configuration ( i.e. @xmath191 , @xmath195 and @xmath196 ) here considered , introducing the notation @xmath282 , one can explicitly derive the fano factors which have the following expressions : @xmath283 they depend on the applied bias @xmath192 and on the side - gate voltage @xmath188 via the scattering amplitudes @xmath284 . in fig . [ fig : fig7 ] we plot the fano factor characterizing the fluctuation processes of the lead 1 ( upper panel ) and 2 ( lower panel ) as a function of the side - gate voltage @xmath230 and by fixing the junction parameters as done in fig . [ fig : fig3 ] . the different curves both in the upper and lower panel are computed taking bias voltages @xmath285 , @xmath286 , @xmath287 , respectively . _ vs _ @xmath230 curves computed assuming junction parameters : @xmath203 , @xmath204 , @xmath213 ( as fixed in fig . 3 ) . the horizontal line @xmath288 is a guide for the eyes indicating the fano factor of a poissonian process for comparison . lower panel : @xmath289 _ vs _ @xmath230 computed by setting the junction parameters as in the upper panel . the different curves in the upper and lower panel refer to different values of applied voltage : @xmath285 , @xmath286 , @xmath287.,title="fig : " ] + _ vs _ @xmath230 curves computed assuming junction parameters : @xmath203 , @xmath204 , @xmath213 ( as fixed in fig . 3 ) . the horizontal line @xmath288 is a guide for the eyes indicating the fano factor of a poissonian process for comparison . lower panel : @xmath289 _ vs _ @xmath230 computed by setting the junction parameters as in the upper panel . the different curves in the upper and lower panel refer to different values of applied voltage : @xmath285 , @xmath286 , @xmath287.,title="fig : " ] the upper panel of fig . [ fig : fig7 ] shows that by tuning the side - gate voltage within the coulomb blockade region a super - poissonian noise is observed with a strong enhancement of the fano factor up to a value of 5 . sub - poissonian fluctuations are instead detected outside the blockade region . the coulomb blockade regime can be overcome by using the side - gate voltage or by increasing the applied voltage bias . the fano factor @xmath289 associated to the fluctuation in the lead 2 ( see lower panel ) is instead quite insensitive to the effect of @xmath230 which only produces a moderate increase of a sub - poissonian fano factor when the coulomb blockade region is reached . the occurrence of super - poissonian noise can be ascribed to different mechanisms even though , in all the cases , the coulomb blockade physics seems to be fundamental for the shot noise enhancement@xcite . in the lead 1 two conditions favorable to the shot - noise enhancement are verified : ( i ) two channels , with different transparencies , are available for transport , i.e. the paths linking the couple of electrodes @xmath290 and @xmath291 ; ( ii ) the coulomb blockade regime makes the tunneling events correlated and thus the electrons are transferred in bunches when transport takes place through the more transparent channel . + on the other hand , the charge current in the lead 2 presents an ohmic behavior coming from the interaction of an open channel ( carrying a current @xmath292 ) with a low - transparency channel ( carrying a current @xmath293 ) . the open channel does not contribute to the current fluctuations , while the opaque one is responsible for the sub - poissonian value of @xmath289 since @xmath294 . + from the experimental side , it has been reported in ref . [ ] that the presence of coexisting current paths , namely a hopping ( diffusive ) path in parallel with a resonant tunneling process , hampers the manifestation of an enhanced fano factor . in that case , the measured fano factor @xmath295 results from a weighted average of the diffusive fano factor @xmath296 , associated to the hopping background , and of the fano factor @xmath297 describing the resonant tunneling process ( @xmath298 ) . when the diffusive and the resonant channels contribute equally to the current , the measured fano factor @xmath295 is near equal to @xmath299 , while a further reduction of @xmath295 is expected if the diffusive path dominates the transport . the latter situation , however , appears favored in devices with large transverse dimension ( @xmath300 in the case of ref . [ ] ) where the presence of diffusive channels alternative to the direct tunneling transmission can not be excluded . on the other hand , the possibility to observe effective fano factors of order 10 is supported by ref . [ ] where edge - channel transport with quantized conductance @xmath301 has been induced by using high magnetic field in the range 9 - 11 t. thus , in our opinion , the fano factor enhancement here reported can in principle be detected in a topological corner junction where the presence of diffusive effects should be limited by the insulating bulk of the topological insulator . within the qpc region , however , the presence of impurity states coupled to the edge - channels can provide diffusive links which lower the effective fano factor . for the above reasons , experimental investigations are required in order to clarify the role of the topological protection in promoting super - poissonian fluctuations with giant fano factor in coulomb blockade regime . here we report the analysis of the non - local current correlations assuming the zero temperature limit . starting from eq . ( [ eq : spectral_tr ] ) , we derive the expressions of @xmath302 , @xmath303 and @xmath304 as a function of the scattering matrix elements : @xmath305 these correlation functions present non - linear dependence on the applied bias @xmath192 ( encoded within the scattering matrix elements ) and contain information on the tunneling amplitude renormalization in coulomb blockade regime . in the absence of charging effects and for @xmath306 ( @xmath307 ) eqs . ( [ eq : non - loc - corr ] ) can be written in closed form in terms of the bare junction parameters @xmath165 and @xmath27 according to the following expressions : @xmath308 where the notation @xmath309 has been introduced . ( [ eq : non - loc - corr - free ] ) show that in the absence of spin - flipping tunneling , i.e. for @xmath164 , the only non - vanishing correlation is @xmath303 , being @xmath310 . since the above observation remains true also in the case of non - vanishing coulomb interaction , studying the non - local current correlations provides the opportunity to further characterize the junction and , in particular , allows the detection of the spin - orbit interaction reactivation in the constriction region . for corner junctions characterized by a moderate reactivation of the spin - orbit effect ( @xmath311 ) , since @xmath312 , @xmath313 and @xmath314 , the above equations predict the following relation among the strengths of the correlation functions : @xmath315 . + moreover the non - local correlations contain information on the coulomb blockade regime and on the tunneling amplitudes renormalization due to the interaction . this analysis is performed in fig . [ fig : fig8 ] where the correlation functions ( in unit of @xmath316 ) @xmath302 , @xmath303 , and @xmath304 are shown as a function of the applied bias @xmath212 and of the side - gate voltage @xmath188 . in particular , all the correlation functions show negative values and present a double - minimum structure within the coulomb blockade region . the two minima are located at @xmath317 and @xmath206 , the latter bias configuration defining the edge of the coulomb blockade region . thus , the absolute value of the non - local correlations is maximized when the coulomb blockade regime is going to be overcome ( see fig . [ fig : fig2 ] ) . interestingly , the presence of the two minima in the correlation functions is directly related to the interaction and its renormalization effect on the tunneling amplitudes . this is supported by the fact that , in the absence of charging effects , the correlation functions present a bias dependence proportional to @xmath318 [ see eqs . ( [ eq : non - loc - corr - free ] ) ] which does not allow the presence of minima . moreover , figs . [ fig : fig8 ] confirm the relation @xmath315 which maintains its validity also in the presence of charging effects . we theoretically studied the transport properties of a corner - junction realized by using a two - dimensional topological insulator shaped in the form of a four - terminal qpc . the coulomb interaction effects , at mean field level , and the reactivation of the spin - orbit coupling due to the tight confinement of the electron densities within the qpc region are taken into account by implementing a self - consistent scattering field theory able to capture the coulomb blockade physics . in this framework , the charging effects are described by a low - energy scattering matrix parametrized by two self - consistent parameters , namely @xmath147 and @xmath148 , and the resulting theoretical approach takes simultaneously into account non - equilibrium effects and coulomb blockade physics beyond the linear response regime . assuming a crossed - bias configuration ( @xmath191 and @xmath319 ) , we studied the current - voltage characteristics of the device and the current - current local and non - local correlations ( noise ) . the dc transport of the device shows a coulomb blockade regime which can be removed by using the side - gate and the bias voltage , while the differential conductance presents a structure reminiscent of a single coulomb diamond whose extension along the side - gate axis is proportional to the interaction value @xmath216 . concerning the fluctuation properties of the current flowing through the device , we studied the thermal noise and the transport noise . it has been proven that the ( local ) thermal noise takes the universal value @xmath242 below a gap - dependent characteristic temperature @xmath244 , the latter property being relevant in discriminating devices with helical properties . the non - local thermal fluctuations contain instead important information on the tunneling amplitudes and thus can help in characterizing the junction properties . the coulomb blockade region has been further characterized by evaluating the fano factor which provides a measure of the transport fluctuations in a given electrode . it has been found a sub - poissonian character of the fluctuation outside the coulomb blockade region and a super - poissonian value of the fano factor as the effect of the blockade imposed by the charging energy within the qpc . the non - local transport fluctuations exhibit a double - minimum structure as a function of the applied bias . the double - minimum is peculiar of a non - vanishing coulomb interaction within the qpc and thus is not present when the charging energy of the constriction is neglected . the theory presented in this work contains all the relevant ingredients needed to describe the transport properties of a corner - junction realized using helical matter ( i.e. a topological insulator ) and allows a direct comparison with forthcoming experimental works . the experimental implementation of our proposal presents similar difficulties of the ones described in ref . [ ] even though the topological corner - junction setup does not require the use of high magnetic fields to induce edge states , the latter being an important simplification of the experimental situation . finally , it s worth mentioning that the universal nature of the thermal fluctuation of a corner - junction realized using helical matter can be an important signature to probe the emergence of a topological phase at cryogenic temperatures ( few kelvins ) . the latter point appears to be even more important in testing topological properties of the next - generation room temperature topological insulators@xcite . the explicit expression of the @xmath150 matrix in terms of the scattering matrix elements takes the following form : @xmath320 the explicit expression of the scattering matrix @xmath157 $ ] in terms of the bare tunneling amplitudes @xmath321 and of the self - consistency potentials @xmath322 takes the following form : @xmath323 the authors gratefully acknowledge d. ferraro and a. braggio for helpful comments . discussions with c. barone and s. pagano on the 1/f noise in nanodevices are also acknowledged . one of the authors ( r. c. ) acknowledges support from the project firb-2012-hybridnanodev . 99 m. z. hasan and c. l. kane , rev . phys . * 82 * , 3045 ( 2010 ) . qi and s. -c . zhang , rev . mod . phys . * 83 * , 1057 ( 2011 ) . c. day , phys . today * 61 * , 19 ( 2008 ) . c. l. kane and e. j. mele , phys lett . * 95 * , 146802 ( 2005 ) ; c. l. kane and e. j. mele , phys . rev . lett . * 95 * , 226801 ( 2005 ) . a. inhofer , d. bercioux , phys . b * 88 * , 235412 ( 2013 ) . b. a. bernevig , t. l. hughes , and s .- c . zhang , science * 314 * , 1757 ( 2006 ) . m. konig , s. weidmann , c. brune , a. roth , h. buhmann , l. w. molenkampf , x. -l . qi , and s. -c . zhang , science * 318 * , 766 ( 2007 ) . a. roth , c. brne , h. buhmann , l. w. molenkamp , j. maciejko , x .- qi , s .- c . zhang , science * 325 * , 294 ( 2009 ) . m. bttiker , science * 325 * , 278 ( 2009 ) . b. d. kong , y. g. semenov , c. m. krowne , and k. w. kim , appl . . lett . * 98 * , 243112 ( 2011 ) . j. maciejko , e .- a . kim , and x .- qi , phys . b * 82 * , 195409 ( 2010 ) . f. dolcini , phys . b * 83 * , 165304 ( 2011 ) . r. citro , f. romeo and n. andrei , phys . b * 84 * , 161301(r ) ( 2011 ) . p. virtanen and p. recher , phys . b * 83 * , 115332 ( 2011 ) . v. krueckl and k. richter , phys . lett . * 107 * , 086803 ( 2011 ) . l. b. zhang , f. cheng , f. zhai , and k. chang , phys . b * 83 * , 081402(r ) ( 2011 ) . p. sternativo , f. dolcini , phys . b * 89 * , 035415 ( 2014 ) . f. romeo , r. citro , d. ferraro , and m. sassetti , phys . b * 86 * , 165418 ( 2012 ) . hou , e .- a . kim , and c. chamon , phys . lett . * 102 * , 076602 , ( 2009 ) . a. strm and h. johannesson , phys . 102 * , 096806 ( 2009 ) . t. l. schmidt , phys . lett . * 107 * , 096602 ( 2011 ) . g. dolcetto , s. barbarino , d. ferraro , n. magnoli and m. sassetti , phys . b * 85 * , 195138 ( 2012 ) . d. ferraro , g. dolcetto , r. citro , f. romeo , m. sassetti , phys . b * 87 * , 245419 ( 2013 ) . j .- souquet , p. simon , phys . b * 86 * , 161410(r ) ( 2012 ) . lee , y .- l . lee , and c .- h . chung , phys . b * 86 * , 235121 ( 2012 ) . where not specified the integration domain is given by @xmath324 $ ] , i.e. @xmath325 . j. c. y. teo and c. l. kane , phys . b * 79 * , 235321 ( 2009 ) ; c .- x . liu , j. c. budich , p. recher , and b. trauzettel , phys . b * 83 * , 035407 ( 2011 ) ; j. i. vayrynen and t. ojanen , phys . rev * 106 * , 076803 ( 2011 ) . b. a. bernevig , and s .- c . zhang , phys . lett . * 96 * , 106802 ( 2006 ) ; x .- qi , t. l. hughes , and s .- c . zhang , phys . b * 78 * , 195424 ( 2008 ) . m. bttiker , phys . b * 46 * , 12485 ( 1992 ) . v. a. khlus , sov jetp * 66 * , 1243 ( 1987 ) . at low temperature we have : @xmath326 where the insulating bulk gap energy @xmath8 represents the energy cut - off of the edge - states theory . t. giamarchi , `` quantum physics in one dimension '' , ( oxford university press , oxford , 2004 ) . coulomb blockade physics has been detected in @xmath327-based quantum dots in s. cho , d. kim , p. syers , n. p. butch , j. paglione , m. s. fuhrer , nano lett . * 12 * , 469 ( 2012 ) . for the grounded configuration ( @xmath328 , @xmath329 ) the self - consistent equations given in eqs . ( [ eq : self2 ] ) are directly solved by @xmath330 . thus the scattering matrix of the problem is given in closed form by eq . ( [ eq : sm ] ) with the substitution @xmath330 . moreover , in this case , coulomb interaction just redefines the zero of the side - gate voltage @xmath230 . y. ando , j. phys . . jpn . * 82 * , 102001 ( 2013 ) . l. kou , b. yan , f. hu , s .- c . wu , t. o. wehling , c. felser , c. chen , t. frauenheim , nano lett . * 13 * , 6251 ( 2013 ) . the universality of the equilibrium noise of a multi - terminal quantum spin hall insulator has been recently pointed out in j. m. edge , j. li , p. delplace , and m. bttiker , phys . lett . * 110 * , 246601 ( 2013 ) . y. zhang , l. dicarlo , d. t. mcclure , m. yamamoto , s. tarucha , c. m. marcus , m. p. hanson , and a. c. gossard , phys . rev . lett . * 99 * , 036603 ( 2007 ) . o. zarchin , y. c. chung , m. heiblum , d. rohrlich , and v. umansky , phys . * 98 * , 066801 ( 2007 ) . n. ubbelohde , c. fricke , f. hohls , and r. j. haug , phys . b * 88 * , 041304(r ) ( 2013 ) . e. onac , f. balestro , b. trauzettel , c. f. j. lodewijk , and l. p. kouwenhoven , phys . lett . * 96 * , 026803 ( 2006 ) . s. s. safonov , a. k. savchenko , d. a. bagrets , o. n. jouravlev , y.v . nazarov , e. h. linfield , and d. a. ritchie , phys . lett . * 91 * , 136801 ( 2003 ) .

we study the density estimation problem with observations generated by certain dynamical systems that admit a unique underlying invariant lebesgue density . observations drawn from dynamical systems are not independent and moreover , usual mixing concepts may not be appropriate for measuring the dependence among these observations . by employing the-mixing concept to measure the dependence , we conduct statistical analysis on the consistency and convergence of the kernel density estimator . our main results are as follows : first , we show that with properly chosen bandwidth , the kernel density estimator is universally consistent under-norm ; second , we establish convergence rates for the estimator with respect to several classes of dynamical systems under-norm . in the analysis , the density function is only assumed to be hlder continuous which is a weak assumption in the literature of nonparametric density estimation and also more realistic in the dynamical system context . last but not least , we prove that the same convergence rates of the estimator under-norm and-norm can be achieved when the density function is hlder continuous , compactly supported and bounded . the bandwidth selection problem of the kernel density estimator for dynamical system is also discussed in our study via numerical simulations . kernel density estimation , dynamical system , dependent observations ,-mixing process , universal consistency , convergence rates , covering number , learning theory

1607.03792

dynamical systems are now ubiquitous and are vital in modeling complex systems , especially when they admit recurrence relations . statistical inference for dynamical systems has drawn continuous attention across various fields , the topics of which include parameter estimation , invariant measure estimation , forecasting , noise detection , among others . for instance , in the statistics and machine learning community , the statistical inference for certain dynamical systems have been recently studied in @xcite , just to name a few . we refer the reader to a recent survey in @xcite for a general depiction of this topic . the purpose of this study is to investigate the density estimation problem for dynamical systems via a classical nonparametric approach , i.e. , kernel density estimation . the commonly considered density estimation problem can be stated as follows . let @xmath4 , @xmath5 , @xmath6 be observations drawn independently from an unknown distribution @xmath7 on @xmath8 with the density @xmath2 . density estimation is concerned with the estimation of the underlying density @xmath2 . accurate estimation of the density is important for many machine learning tasks such as regression , classification , and clustering problems and also plays an important role in many real - world applications . nonparametric density estimators are popular since weaker assumptions are applied to the underlying probability distribution . typical nonparametric density estimators include the histogram and kernel density estimator . in this study , we are interested in the latter one , namely , _ kernel density estimator _ , which is also termed as _ parzen - rosenblatt estimator _ @xcite and takes the following form @xmath9 here , @xmath10 is a bandwidth parameter and @xmath11 is a smoothing kernel . in the literature , point - wise and uniform consistency and convergence rates of the estimator @xmath12 to the unknown truth density @xmath2 under various distance measurements , e.g. , @xmath13 , have been established by resorting to the regularity assumptions on the smoothing kernel @xmath14 , the density @xmath2 , as well as the decay of the bandwidth sequence @xmath15 . besides the theoretical concerns on the consistency and convergence rates , another practical issue one usually needs to address is the choice of the bandwidth parameter @xmath16 , which is also called the _ smoothing parameter_. it plays a crucial role in the bias - variance trade - off in kernel density estimation . in the literature , approaches to choosing the smoothing parameter include least - squares cross - validation @xcite , biased cross - validation @xcite , plug - in method @xcite , the double kernel method @xcite , as well as the method based on a discrepancy principle @xcite . we refer the reader to @xcite for a general overview and to @xcite for more detailed reviews . note that studies on the kernel density estimator mentioned above heavily rely on the assumption that the observations are drawn in an i.i.d fashion . in the literature of statistics and machine learning , it is commonly accepted that the i.i.d assumption on the given data can be very much restrictive in real - world applications . having realized this , researchers turn to weaken this i.i.d assumption by assuming that the observations are weakly dependent under various notions of weakly dependence which include @xmath17-mixing , @xmath18-mixing , and @xmath19-mixing @xcite . there has been a flurry of work to attack this problem with theoretical and practical concerns , see e.g. , @xcite and @xcite , under the above notions of dependence . as a matter of fact , the assumed correlation among the observations complicates the kernel density estimation problem from a technical as well as practical view and also brings inherent barriers . this is because , more frequently , the analysis on the consistency and convergence rates of the kernel density estimator is proceeded by decomposing the error term into bias and variance terms , which correspond to data - free and data - dependent error terms , respectively . the data - free error term can be tackled by using techniques from the approximation theory while the data - dependent error term is usually dealt with by exploiting arguments from the empirical process theory such as concentration inequalities . as a result , due to the existence of dependence among observations and various notions of the dependence measurement , the techniques , and results concerning the data - dependent error term are in general not universally applicable . on the other hand , it has been also pointed out that the bandwidth selection in kernel density estimation under dependence also departures from the independent case , see e.g. , @xcite . in fact , when the observations @xmath20 are generated by certain ergodic measure - preserving dynamical systems , the problem of kernel density estimation can be even more involved . to explain , let us consider a discrete - time ergodic measure - preserving dynamical system described by the sequence @xmath21 of iterates of an unknown map @xmath22 with @xmath23 and a unique invariant measure @xmath7 which possesses a density @xmath2 with respect to the lebesgue measure ( rigorous definitions will be given in the sequel ) . that is , we have @xmath24 where @xmath25 is the initial state . it is noticed that in this case the usual mixing concepts are not general enough to characterize the dependence among observations generated by @xcite . on the other hand , existing theoretical studies on the consistency and convergence rates of the kernel density estimator for i.i.d . observations frequently assume that the density function @xmath2 is sufficiently smooth , e.g. , first - order or even second - order smoothness . however , more often than not , this requirement can be stringent in the dynamical system context . for instance , the lasota - yorke map @xcite admits a density @xmath2 which only belongs to the space @xmath26 , i.e. , functions of bounded variation . this is also the case for the @xmath18-map in example [ beta_map ] ( see subsection [ sec::subsec::dynamicalsys ] ) . therefore , studies on kernel density estimation mentioned above with dependent observations , in general , may not be applicable . for more detailed comparison we refer to section [ candd ] . in this study , the kernel density estimation problem with observations generated by dynamical systems is approached by making use of a more general concept for measuring the dependence of observations , namely , the so - called @xmath0-mixing process ( refer to section [ sec::preliminary ] for the definition ) . proposed in @xcite and recently investigated in @xcite and @xcite , the @xmath0-mixing concept is shown to be more general and powerful in measuring dependence among observations generated by dynamical systems and can accommodate a large class of dynamical systems . recently , a bernstein - type exponential inequality for @xmath0-mixing processes was established in @xcite and its applications to some learning schemes were explored in @xcite and @xcite . our main purpose in this paper is to conduct some theoretical analysis and practical implementations on the kernel density estimator for dynamical systems . the primary concern is the consistency and convergence rates of the kernel density estimator with observations generated by dynamical systems . the consistency and convergence analysis is conducted under @xmath1-norm , and @xmath3-norm , respectively . we show that under mild assumptions on the smoothing kernel , with properly chosen bandwidth , the estimator is universally consistent under @xmath1-norm . when the probability distribution @xmath7 possesses a polynomial or exponential decay outside of a radius-@xmath27 ball in its support , under the hlder continuity assumptions on the kernel function and the density , we obtain almost optimal convergence rates under @xmath1-norm . moreover , when the probability distribution @xmath7 is compactly supported , which is a frequently encountered setting in the dynamical system context , we prove that stronger convergence results of the estimator can be developed , i.e. , convergence results under @xmath3-norm which are shown to be of the same order with its @xmath1-norm convergence rates . finally , with regard to the practical implementation of the estimator , we also discuss the bandwidth selection problem by performing numerical comparisons among several typical existing selectors that include least squares cross - validation and its variants for dependent observations as well as the double kernel method . we show that the double kernel bandwidth selector proposed in @xcite can in general work well . moreover , according to our numerical experiments , we find that bandwidth selection for kernel density estimator of dynamical systems is usually ad - hoc in the sense that its performance may depend on the considered dynamical system . the rest of this paper is organized as follows . section [ sec::preliminary ] is a warm - up section for the introduction of some notations , definitions and assumptions that are related to the kernel density estimation problem and dynamical systems . section [ sec::consistency_convergence ] is concerned with the consistency and convergence of the kernel density estimator and presents the main theoretical results of this study . we discuss the bandwidth selection problem in section [ sec::bandwidth_selection ] . all the proofs of section [ sec::consistency_convergence ] can be found in section [ proofs ] . we end this paper in section [ sec::conclusion ] . throughout this paper , @xmath28 is denoted as the lebesgue measure on @xmath8 and @xmath29 is an arbitrary norm on @xmath8 . we denote @xmath30 as the centered ball of @xmath8 with radius @xmath27 , that is , @xmath31 and its complement @xmath32 as @xmath33 recall that for @xmath34 , the @xmath35-norm is defined as @xmath36 , and the @xmath37-norm is defined as @xmath38 . let @xmath39 be a probability space . we denote @xmath40 as the space of ( equivalence classes of ) measurable functions @xmath41 with finite @xmath42-norm @xmath43 . then @xmath40 together with @xmath44 forms a banach space . moreover , if @xmath45 is a sub-@xmath46-algebra , then @xmath47 denotes the space of all @xmath48-measurable functions @xmath49 . finally , for a banach space @xmath50 , we write @xmath51 for its closed unit ball . in what follows , the notation @xmath52 means that there exists a positive constant @xmath53 such that @xmath54 , for all @xmath55 . with a slight abuse of notation , in this paper , @xmath56 and @xmath57 are used interchangeably for positive constants while their values may vary across different lemmas , theorems , and corollaries . in this subsection , we first introduce the dynamical systems of interest , namely , ergodic measure - preserving dynamical systems . mathematically , an ergodic measure - preserving _ dynamical system _ is a system @xmath58 with a mapping @xmath59 that is measure - preserving , i.e. , @xmath60 for all @xmath61 , and ergodic , i.e. , @xmath62 implies @xmath63 or @xmath64 . in this study , we are confined to the dynamical systems in which @xmath65 is a subset of @xmath8 , @xmath66 is a probability measure that is absolutely continuous with respect to the lebesgue measure @xmath67 and admits a unique invariant lebesgue density @xmath2 . in our study , it is assumed that the observations @xmath68 are generated by the discrete - time dynamical system . below we list several typical examples of discrete - time dynamical systems that satisfy the above assumptions @xcite : [ logistic_map ] the logistic map is defined by @xmath69,\ ] ] with a unique invariant lebesgue density @xmath70 [ gauss_map ] the gauss map is defined by @xmath71 with a unique invariant lebesgue density @xmath72 [ beta_map ] for @xmath73 , the @xmath18-map is defined as @xmath74 with a unique invariant lebesgue density given by @xmath75}(x),\ ] ] where @xmath76 is a constant chosen such that @xmath2 has integral @xmath64 . we now introduce the notion for measuring the dependence among observations from dynamical systems , namely , @xmath0-mixing process @xcite . to this end , let us assume that @xmath77 is a measurable space with @xmath78 . let @xmath79 be an @xmath80-valued stochastic process on @xmath81 , and for @xmath82 , denote by @xmath83 the @xmath46-algebra generated by @xmath84 . let @xmath85 be a measurable map . @xmath86 is denoted as the @xmath87-image measure of @xmath66 , which is defined as @xmath88 , @xmath89 measurable . the process @xmath90 is called _ stationary _ if @xmath91 for all @xmath92 . denote @xmath93 . moreover , for @xmath94 satisfying @xmath95 , we denote the correlation of @xmath96 and @xmath97 by @xmath98 it is shown that several dependency coefficients for @xmath90 can be expressed in terms of such correlations for restricted sets of functions @xmath96 and @xmath97 @xcite . in order to introduce the notion , we also need to define a new norm , which is taken from @xcite and introduces restrictions on @xmath96 and @xmath97 considered here . let us assume that @xmath99 is a subspace of bounded measurable functions @xmath100 and that we have a semi - norm @xmath101 on @xmath99 . for @xmath102 , we define the @xmath0-norm @xmath103 by @xmath104 additionally , we need to introduce the following restrictions on the semi - norm @xmath101 . [ seminorm_assump ] we assume that the following two restrictions on the semi - norm @xmath101 hold : * @xmath105 for all constant functions @xmath102 ; * @xmath106 . note that the first constraint on the semi - norm in assumption [ seminorm_assump ] implies its shift invariance on @xmath107 while the inequality constraint can be viewed as an abstract _ chain rule _ if one views the semi - norm as a norm describing aspects of the smoothness of @xmath108 , as discussed in @xcite . in fact , it is easy to show that the following function classes , which are probably also the most frequently considered in the dynamical system context , satisfy condition @xmath109 in assumption [ seminorm_assump ] . moreover , they also satisfy condition @xmath110 in assumption [ seminorm_assump ] , as shown in @xcite : * @xmath111 : the class of bounded functions on @xmath80 ; * @xmath112 : the class of bounded variation functions on @xmath80 ; * @xmath113 : the class of bounded and @xmath17-hlder continuous functions on @xmath80 ; * @xmath114 : the class of lipschitz continuous functions on @xmath80 ; * @xmath115 : the class of continuously differentiable functions on @xmath80 . let @xmath81 be a probability space , @xmath77 be a measurable space , @xmath116 be an @xmath80-valued , stationary process on @xmath65 , and @xmath117 be defined by for some semi - norm @xmath101 . then , for @xmath118 , we define the @xmath0-mixing coefficients by @xmath119 and the time - reversed @xmath0-mixing coefficients by @xmath120 let @xmath121 be a strictly positive sequence converging to @xmath122 . we say that * @xmath90 is _ ( time - reversed ) @xmath0-mixing _ * with rate @xmath121 , if we have @xmath123 for all @xmath118 . moreover , if @xmath121 is of the form @xmath124 for some constants @xmath125 , @xmath126 , and @xmath127 , then @xmath128 is called * _ geometrically _ ( time - reversed ) @xmath0-mixing*. from the above definition , we see that a @xmath0-mixing process is defined in association with an underlying function space . for the above listed function spaces , i.e. , @xmath111 , @xmath112 , @xmath113 , @xmath114 and @xmath115 , the increase of the smoothness enlarges the class of the associated stochastic processes , as illustrated in @xcite . note that the classical @xmath19-mixing process is essentially a @xmath0-mixing process associated with the function space @xmath111 . note also that not all @xmath17-mixing processes are @xmath0-mixing , and vice versa . we refer the reader to @xcite for the relations among @xmath17- , @xmath19- and @xmath0-mixing processes . on the other hand , under the above notations and definitions , from theorem @xmath129 in @xcite , we know that logistic map in example [ logistic_map ] is geometrically time - reversed @xmath0-mixing with @xmath130 while theorem @xmath131 in @xcite ( see also chapter @xmath132 in @xcite ) indicates that gauss map in example [ gauss_map ] is geometrically time - reversed @xmath0-mixing with @xmath133 . example [ beta_map ] is also geometrically time - reversed @xmath0-mixing with @xmath133 according to @xcite . for more examples of geometrically time - reversed @xmath0-mixing dynamical systems , the reader is referred to section @xmath134 in @xcite . for the smoothing kernel @xmath11 in the kernel density estimator , in this paper we consider its following general form , namely , @xmath135-dimensional smoothing kernel : [ kernel ] a bounded , monotonically decreasing function @xmath136 is a * @xmath135-dimensional smoothing kernel * if @xmath137 the choice of the norm in definition [ kernel ] does not matter since all norms on @xmath8 are equivalent . to see this , let @xmath138 be another norm on @xmath8 satisfying @xmath139 . from the equivalence of the two norms on @xmath8 , one can find a positive constant @xmath53 such that @xmath140 holds for all @xmath141 . therefore , easily we have @xmath142 in what follows , without loss of generality , we assume that the constant @xmath143 in definition [ kernel ] equals to @xmath64 . [ lemma_kernel ] a bounded , monotonically decreasing function @xmath136 is a @xmath135-dimensional smoothing kernel if and only if @xmath144 from the above discussions , it suffices to consider the integration constraint for the kernel function @xmath11 with respect to the euclidean norm @xmath145 . we thus have @xmath146 where @xmath147 is the volume of the unit ball @xmath148 of the euclidean space @xmath149 . this completes the proof of lemma [ lemma_kernel ] . let @xmath150 and denote @xmath151 as the indicator function . several common examples of @xmath135-dimensional smoothing kernels @xmath152 include the naive kernel @xmath153}(r)$ ] , the triangle kernel @xmath154}(r)$ ] , the epanechnikov kernel @xmath155}(r)$ ] , and the gaussian kernel @xmath156 . in this paper , we are interested in the kernels that satisfy the following restrictions on their shape and regularity : [ assump_kernel ] for a fixed function space @xmath99 , we make the following assumptions on the @xmath135-dimensional smoothing kernel @xmath11 : * @xmath11 is hlder continuous with exponent @xmath18 with @xmath157 $ ] ; * @xmath158 ; * for all @xmath159 , we have @xmath160 and there exists a function @xmath161 such that @xmath162 it is easy to verify that for @xmath163 , assumption [ assump_kernel ] is met for the triangle kernel , the epanechnikov kernel , and the gaussian kernel . particularly , condition @xmath164 holds for all these kernels with @xmath101 being the lipschitz norm and @xmath165 . moreover , as we shall see below , not all the conditions in assumption [ assump_kernel ] are required for the analysis conducted in this study and conditions assumed on the kernel will be specified explicitly . we now show that given a @xmath135-dimensional smoothing kernel @xmath11 as in definition [ kernel ] , one can easily construct a probability density on @xmath8 . let @xmath11 be a @xmath135-dimensional smoothing kernel and @xmath166 be a probability measure on @xmath8 . then , for @xmath167 , @xmath168 is called a * @xmath11-smoothing of @xmath166*. it is not difficult to see that @xmath169 defines a probability density on @xmath8 , since fubini s theorem yields that @xmath170 let us denote @xmath171 as @xmath172 note that @xmath173 also induces a density function on @xmath8 since there holds @xmath174 . for the sake of notational simplification , in what follows , we introduce the convolution operator @xmath175 . under this notation , we then see that @xmath169 is the density of the measure that is the convolution of the measure @xmath166 and @xmath176 . recalling that @xmath7 is a probability measure on @xmath8 with the corresponding density function @xmath2 , by taking @xmath177 with @xmath178 , we have @xmath179 since @xmath180 and @xmath181 , from proposition ( 8.8 ) in @xcite we know that @xmath182 is uniformly continuous and bounded . specifically , when @xmath166 is the empirical measure @xmath183 , the _ kernel density estimator for dynamical systems _ in this study can be expressed as @xmath184 from now on , for notational simplicity , we will suppress the subscript @xmath185 of @xmath186 and denote @xmath187 , e.g. , @xmath188 . in this section , we study the consistency and convergence rates of @xmath189 to the true density @xmath2 under @xmath1-norm and also @xmath3-norm for some special cases . recall that @xmath189 is a nonparametric density estimator and so the criterion that measures its goodness - of - fit matters , which , for instance , includes @xmath1-distance , @xmath190-distance and @xmath3-distance . in the literature of kernel density estimation , probably the most frequently employed criterion is the @xmath190-distance of the difference between @xmath189 and @xmath2 , since it entails an exact bias - variance decomposition and can be analyzed relatively easily by using taylor expansion involved arguments . however , it is argued in @xcite ( see also @xcite ) that @xmath1-distance could be a more reasonable choice since : it is invariant under monotone transformations ; it is always well - defined as a metric on the space of density functions ; it is also proportional to the total variation metric and so leads to better visualization of the closeness to the true density function than @xmath190-distance . the downside of using @xmath1-distance is that it does not admit an exact bias - variance decomposition and the usual taylor expansion involved techniques for error estimation may not apply directly . nonetheless , if we introduce the intermediate estimator @xmath182 in , obviously the following inequality holds @xmath191 the consistency and convergence analysis in our study will be mainly conducted in the @xmath1 sense with the help of inequality . for some specific case , i.e. , when the density @xmath2 is compactly supported , we are also concerned with the consistency and convergence of @xmath189 to @xmath2 under @xmath3-norm . in this case , there also holds the following inequality @xmath192 it is easy to see that the first error term on the right - hand side of or is stochastic due to the empirical measure @xmath193 while the second one is deterministic because of its sampling - free nature . loosely speaking , the first error term corresponds to the variance of the estimator @xmath189 , while the second one can be treated as its bias although or is not an exact error decomposition . in our study , we proceed with the consistency and convergence analysis on @xmath189 by bounding the two error terms , respectively . our first theoretical result on bounding the deterministic error term shows that , given a @xmath135-dimensional kernel @xmath11 , the @xmath1-distance between its @xmath11-smooth of the measure @xmath7 , i.e. , @xmath182 , and @xmath2 can be arbitrarily small by choosing the bandwidth appropriately . moreover , under mild assumptions on the regularity of @xmath2 and @xmath11 , the @xmath3-distance between the two quantities possesses a polynomial decay with respect to the bandwidth @xmath194 . [ approximationerror ] let @xmath11 be a @xmath135-dimensional smoothing kernel . * for any @xmath195 , there exists @xmath196 such that for any @xmath197 $ ] we have @xmath198 * if @xmath11 satisfies condition @xmath110 in assumption [ assump_kernel ] and @xmath2 is @xmath17-hlder continuous with @xmath199 , then there holds @xmath200 we now show that the @xmath1-distance between @xmath182 and @xmath2 can be upper bounded by their difference ( in the sense of @xmath3-distance ) on a compact domain of @xmath8 together with their difference ( in the sense of @xmath1-distance ) outside this domain . as we shall see later , this observation will entail us to consider different classes of the true densities @xmath2 . the following result is crucial in our subsequent analysis on the consistency and convergence rates of @xmath189 . [ comparisionnorms ] assume that @xmath11 is a @xmath135-dimensional smoothing kernel that satisfies conditions @xmath109 and @xmath110 in assumption [ assump_kernel ] . for @xmath201 and @xmath202 , we have @xmath203 we now proceed with the estimation of the stochastic error term @xmath204 by establishing probabilistic oracle inequalities . for the sake of readability , let us start with an overview of the analysis conducted in this subsection for bounding the stochastic error term . in this study , the stochastic error term is tackled by using capacity - involved arguments and the bernstein - type inequality established in @xcite . in the sequel , for any fixed @xmath205 , we write @xmath206 and we further denote the centered random variable @xmath207 on @xmath65 as @xmath208 it thus follows that @xmath209 and consequently we have @xmath210 and @xmath211 as a result , in order to bound @xmath204 , it suffices to bound the supremum of the empirical process @xmath212 indexed by @xmath213 . for any @xmath214 , there holds @xmath215 the second term of the right - hand side of the above equality can be similarly dealt with as in the proof of theorem [ comparisionnorms ] . in order to bound the first term , we define @xmath216 as the function set of @xmath207 that corresponds to @xmath217 which lies on a radius-@xmath27 ball of @xmath8 : @xmath218 the idea here is to apply capacity - involved arguments and the bernstein - type exponential inequality in @xcite to the function set @xmath216 and the associated empirical process @xmath219 . the difference between @xmath189 and @xmath182 under the @xmath3-norm can be bounded analogously . therefore , to further our analysis , we first need to bound the capacity of @xmath216 in terms of covering numbers . let @xmath220 be a metric space and @xmath221 . for @xmath195 , the * @xmath222-covering number * of @xmath223 is denoted as @xmath224 where @xmath225 . for a fixed @xmath202 , we consider the function set @xmath226 the following proposition provides an estimate of the covering number of @xmath227 . [ kernelspacecn ] let @xmath11 be a @xmath135-dimensional smoothing kernel that satisfies conditions @xmath109 in assumption [ assump_kernel ] and @xmath228 $ ] . then there exists a positive constant @xmath229 such that for all @xmath230 $ ] , we have @xmath231 we now establish oracle inequalities for the kernel density estimator under @xmath1-norm , and @xmath3-norm , respectively . these oracle inequalities will be crucial in establishing the consistency and convergence results of the estimator . recall that the considered kernel density estimation problem is based on samples from an @xmath80-valued @xmath0-mixing process which is associated with an underlying function class @xmath99 . as shown below , the established oracle inequality holds without further restrictions on the support of the density function @xmath2 . [ oracleinequalityl1 ] suppose that assumption [ assump_kernel ] holds . let @xmath232 be an @xmath80-valued stationary geometrically ( time - reversed ) @xmath0-mixing process on @xmath81 with @xmath117 being defined for some semi - norm @xmath101 that satisfies assumption [ seminorm_assump ] . then for all @xmath233 , @xmath202 and @xmath234 , there exists an @xmath235 such that for all @xmath236 , with probability @xmath66 at least @xmath237 , there holds @xmath238 here @xmath239 will be given explicitly in the proof . our next result shows that when the density function @xmath2 is compactly supported and bounded , an oracle inequality under @xmath3-norm can be also derived . [ oracleinequalityinftynorm ] let @xmath11 be a @xmath135-dimensional kernel function that satisfies conditions @xmath109 and @xmath164 in assumption [ assump_kernel ] . let @xmath79 be an @xmath80-valued stationary geometrically ( time - reversed ) @xmath0-mixing process on @xmath81 with @xmath117 being defined for some semi - norm @xmath101 that satisfies assumption [ seminorm_assump ] . assume that there exists a constant @xmath240 such that @xmath241 and the density function @xmath2 satisfies @xmath242 . then for all @xmath233 and @xmath243 , there exists an @xmath244 such that for all @xmath245 , with probability @xmath66 at least @xmath246 , there holds @xmath247 here @xmath248 will be given explicitly in the proof . in theorem [ oracleinequalityinftynorm ] , the kernel @xmath11 is only required to satisfy conditions @xmath109 and @xmath164 in assumption [ assump_kernel ] whereas the condition that @xmath158 for some @xmath249 is not needed . this is again due to the compact support assumption of the density function @xmath2 as stated in theorem [ oracleinequalityinftynorm ] . we now present results on the universal consistency property of the kernel density estimator @xmath189 in the sense of @xmath1-norm . a kernel density estimator @xmath189 is said to be _ universally consistent _ in the sense of @xmath1-norm if @xmath189 converges to @xmath2 almost surely under @xmath1-norm without any further restrictions on the probability distribution @xmath7 . [ consistencyl1 ] let @xmath11 be a @xmath135-dimensional smoothing kernel that satisfies conditions @xmath109 and @xmath164 in assumption [ assump_kernel ] . let @xmath232 be an @xmath80-valued stationary geometrically ( time - reversed ) @xmath0-mixing process on @xmath81 with @xmath117 being defined for some semi - norm @xmath101 that satisfies assumption [ seminorm_assump ] . if @xmath250 then the kernel density estimator @xmath251 is universally consistent in the sense of @xmath1-norm . the consistency result in theorem [ consistencyl1 ] is independent of the probability distribution @xmath7 and is therefore said to be universal . in this subsection , we will show that if certain tail assumptions on @xmath7 are applied , convergence rates can be obtained under @xmath1-norm . here , we consider three different situations , namely , the tail of the probability distribution @xmath7 has a polynomial decay , exponential decay and disappears , respectively . [ convergenceratesl1 ] let @xmath11 be a @xmath135-dimensional smoothing kernel that satisfies assumption [ assump_kernel ] . assume that the density @xmath2 is @xmath17-hlder continuous with @xmath252 . let @xmath232 be an @xmath80-valued stationary geometrically ( time - reversed ) @xmath0-mixing process on @xmath81 with @xmath117 being defined for some semi - norm @xmath101 that satisfies assumption [ seminorm_assump ] . we consider the following cases : 1 . @xmath253 for some @xmath254 and for all @xmath255 ; 2 . @xmath256 for some @xmath257 , @xmath254 and for all @xmath255 ; 3 . @xmath258 for some @xmath259 . for the above cases , if @xmath260 with @xmath239 the same as in theorem [ oracleinequalityl1 ] , and the sequences @xmath16 are of the following forms : 1 . @xmath261 ; 2 . @xmath262 ; 3 . @xmath263 ; then with probability @xmath66 at least @xmath264 , there holds @xmath265 where the convergence rates 1 . @xmath266 ; 2 . @xmath267 ; 3 . @xmath268 . in subsection [ oracle_inequalites ] , when the density function @xmath2 is bounded and compactly supported , we establish oracle inequality of @xmath189 under @xmath3-norm . combining this with the estimate of the deterministic error term in theorem [ approximationerror ] _ ( ii ) _ under @xmath3-norm , we arrive at the following result that characterizes the convergence of @xmath189 to @xmath2 under @xmath3-norm . [ convergencerateslinfty ] let @xmath11 be a @xmath135-dimensional smoothing kernel that satisfies conditions @xmath109 and @xmath164 in assumption [ assump_kernel ] . let @xmath232 be an @xmath80-valued stationary geometrically ( time - reversed ) @xmath0-mixing process on @xmath81 with @xmath117 being defined for some semi - norm @xmath101 that satisfies assumption [ seminorm_assump ] . assume that there exists a constant @xmath240 such that @xmath241 and the density function @xmath2 is @xmath17-hlder continuous with @xmath199 and @xmath242 . then for all @xmath245 with @xmath248 as in theorem [ oracleinequalityinftynorm ] , by choosing @xmath269 with probability @xmath66 at least @xmath264 , there holds @xmath270 in theorems [ convergenceratesl1 ] and [ convergencerateslinfty ] , one needs to ensure that @xmath260 with @xmath239 as in theorem [ oracleinequalityl1 ] and @xmath271 with @xmath248 as in theorem [ oracleinequalityinftynorm ] , respectively . one may also note that due to the involvement of the term @xmath272 , the numbers @xmath239 and @xmath248 depend on the @xmath16 . however , recalling that for the triangle kernel , the epanechnikov kernel , and the gaussian kernel , we have @xmath273 , which , together with the choices of @xmath16 in theorems [ convergenceratesl1 ] and [ convergencerateslinfty ] , implies that @xmath239 and @xmath248 are well - defined . it should be also remarked that in the scenario where the density function @xmath2 is compactly supported and bounded , the convergence rate of @xmath189 to @xmath2 is not only obtainable , but also the same with that derived under @xmath1-norm . this is indeed an interesting observation since convergence under @xmath3-norm implies convergence under @xmath1-norm . this section presents some comments on the obtained theoretical results on the consistency and convergence rates of @xmath189 and compares them with related findings in the literature . we highlight that in our analysis the density function @xmath2 is only assumed to be hlder continuous . as pointed out in the introduction , in the context of dynamical systems , this seems to be more than a reasonable assumption . on the other hand , the consistency , as well as the convergence results obtained in our study , are of type with high probability " due to the use of the bernstein - type exponential inequality that takes into account the variance information of the random variables . from our analysis and the obtained theoretical results , one can also easily observe the influence of the dependence among observations . for instance , from theorem [ consistencyl1 ] we see that with increasing dependence among observations ( corresponding to smaller @xmath274 ) , in order to ensure the universal consistency of @xmath251 , the decay of @xmath16 ( with respect to @xmath275 ) is required to be faster . this is in fact also the case if we look at results on the convergence rates in theorems [ convergenceratesl1 ] and [ convergencerateslinfty ] . moreover , the influence of the dependence among observations is also indicated there . that is , an increase of the dependence among observations may slow down the convergence of @xmath189 in the sense of both @xmath1-norm and @xmath3-norm . it is also interesting to note that when @xmath274 tends to infinity , which corresponds to the case where observations can be roughly treated as independent ones , meaningful convergence rates can be also deduced . it turns out that , up to a logarithmic factor , the established convergence rates under @xmath3-norm , namely , @xmath276 , match the optimal rates in the i.i.d . case , see , e.g. , @xcite and @xcite . as mentioned in the introduction , there exist several studies in the literature that address the kernel density estimation problem for dynamical systems . for example , @xcite conducted some first studies and showed the point - wise consistency as well as convergence ( in expectation ) of the kernel density estimator . the convergence rates obtained in their study are of the type @xmath277 , which are conducted in terms of the variance of @xmath189 . the notion they used for measuring the dependence among observations is @xmath17-mixing coefficient ( see @xmath278 in @xcite ) . considering the density estimation problem for one - dimensional dynamical systems , @xcite presented some studies on the kernel density estimator @xmath189 by developing a central limit theorem and apply it to bound the variance of the estimator . further some studies on the kernel density estimation of the invariant lebesgue density for dynamical systems were conducted in @xcite . by considering both dynamical noise and observational noise , point - wise convergence of the estimator @xmath189 in expectation was established , i.e. , the convergence of @xmath279 for any @xmath280 . note further that these results rely on the second - order smoothness and boundedness of @xmath2 . therefore , the second - order smoothness assumption on the density function together with the point - wise convergence in expectation makes it different from our work . in particular , under the additional assumption on the tail of the noise distribution , the convergence of @xmath281 for any fixed @xmath280 is of the order @xmath282 with @xmath283 . concerning the convergence of @xmath189 in a dynamical system setup , @xcite also presented some interesting studies which in some sense also motivated our work here . by using also the @xmath0-mixing concept as adopted in our study to measure the dependence among observations from dynamical systems , she presented the point - wise convergence of @xmath189 with the help of hoeffding - type exponential inequality ( see proposition 3.1 in @xcite ) . the assumption applied on @xmath2 is that it is bounded from below and also @xmath17-hlder continuous ( more precisely , @xmath2 is assumed to be _ @xmath17-regular _ , see assumption 2.3 in @xcite ) . hence , from the above discussions , we suggest that the work we present in this study is essentially different from that in @xcite . this section discusses the model selection problem of the kernel density estimator by performing numerical simulation studies . in the context of kernel density estimation , model selection is mainly referred to the choice of the smoothing kernel @xmath11 and the selection of the kernel bandwidth @xmath194 , which are of crucial importance for the practical implementation of the data - driven density estimator . according to our experimental experience and also the empirical observations reported in @xcite , it seems that the choice of the kernel or the noise does not have a significant influence on the performance of the estimator . therefore , our emphasis will be placed on the bandwidth selection problem in our simulation studies . in the literature of kernel density estimation , various bandwidth selectors have been proposed , several typical examples of which have been alluded to in the introduction . when turning to the case with dependent observations , the bandwidth selection problem has been also drawing much attention , see e.g. , @xcite . among existing bandwidth selectors , probably the most frequently employed ones are based on the cross - validation ideas . for cross - validation bandwidth selectors , one tries to minimize the integrated squared error ( ise ) of the empirical estimator @xmath189 where @xmath284 note that on the right - hand side of the above equality , the last term @xmath285 is independent of @xmath194 and so the minimization of @xmath286 is equivalent to minimize @xmath287 it is shown that with i.i.d observations , an unbiased estimator of the above quantity , which is termed as least squares cross - validation ( lscv ) , is given as follows : @xmath288 where the leave - one - out density estimator @xmath289 is defined as @xmath290 when the observations are dependent , it is shown that cross - validation can produce much under - smoothed estimates , see e.g. , @xcite . observing this , @xcite proposed the modified least squares cross - validation ( mlscv ) , which is defined as follows @xmath291 where @xmath292 is set to @xmath64 or @xmath134 as suggested in @xcite and @xmath293 the underlying intuition of proposing mlscv is that when estimating the density of a fixed point , ignoring observations in the vicinity of this point may be help in reducing the influence of dependence among observations . however , when turning to the @xmath1 point of view , the above bandwidth selectors may not work well due to the use of the least squares criterion . alternatively , @xcite proposed the double kernel bandwidth selector that minimizes the following quantity @xmath294 where @xmath295 and @xmath296 are kernel density estimators based on the kernels @xmath11 and @xmath297 , respectively . some rigorous theoretical treatments on the effectiveness of the above bandwidth selector were made in @xcite . our purpose in simulation studies is to conduct empirical comparisons among the above bandwidth selectors in the dynamical system context instead of proposing new approaches . in our experiments , observations @xmath298 are generated from the following model @xmath299 where @xmath300 , @xmath46 is set to @xmath301 and the initial state @xmath25 is randomly generated based on the density @xmath2 . for the map @xmath302 in , we choose logistic map in example [ logistic_map ] and gauss map in example [ gauss_map ] . we vary the sample size among @xmath303 , implement bandwidth selection procedures over @xmath304 replications and select the bandwidth from a grid of values in the interval @xmath305 $ ] with @xmath306 equispaced points . here , @xmath307 is set as the minimum distance between consecutive points @xmath308 @xcite , while @xmath309 is chosen according to the _ maximal smoothing principle _ proposed in @xcite . throughout our experiments , we use the gaussian kernel for the kernel density estimators . in our experiments , we conduct comparisons among the above - mentioned bandwidth selectors which are , respectively , denoted as follows : * lscv : the least squares cross - validation given in ; * mlscv-1 : the modified least squares cross - validation in with @xmath310 ; * mlscv-2 : the modified least squares cross - validation in with @xmath311 ; * dkm : the double kernel method defined in where the two kernels used here are the epanechnikov kernel and the triangle kernel , respectively . in the experiments , due to the known density functions for logistic map and gauss map , and in accordance with our previous analysis from the @xmath1 point of view , the criterion of comparing different selected bandwidths is the following absolute mean error ( ame ) : @xmath312 where @xmath313 are @xmath314 equispaced points in the interval @xmath315 $ ] and @xmath314 is set to @xmath316 . we also compare the selected bandwidth with the one that has the minimum absolute mean error which serves as a * _ baseline _ * method in our experiments . the ames of the above bandwidth selectors for logistic map in example [ logistic_map ] and gauss map in example [ gauss_map ] over @xmath317 replications are averaged and recorded in tables [ logistic_map_table ] and [ gauss_map_table ] below . in figs.[logistic_map_figure ] and [ gauss_map_figure ] , we also plot the kernel density estimators for logistic map in example [ logistic_map ] and gauss map in example [ gauss_map ] with different bandwidths and their true density functions with different sample sizes . the sample size of each panel , in figs.[logistic_map_figure ] and [ gauss_map_figure ] , from up to bottom , is @xmath318 , @xmath319 and @xmath320 , respectively . in each panel , the densely dashed black curve represents the true density , the dotted blue curve is the estimated density function with the bandwidth selected by the baseline method while the solid red curve stands for the estimated density with the bandwidth selected by the double kernel method . all density functions in figs.[logistic_map_figure ] and [ gauss_map_figure ] are plotted with @xmath306 equispaced points in the interval @xmath321 . .the ames of different bandwidth selectors for logistic map in example [ logistic_map ] [ cols="^,^,^,^,^,^ " , ] + + + + from tables [ logistic_map_table ] and [ gauss_map_table ] , and figs.[logistic_map_figure ] and [ gauss_map_figure ] , we see that the true density functions of logistic map and gauss map can be approximated well with enough observations and the double kernel method works slightly better than the other three methods for the two dynamical systems . in fact , according to our experimental experience , we find that the bandwidth selector of the kernel density estimator for a dynamical system is usually ad - hoc . that is , for existing bandwidth selectors , there seems no a universal optimal one that can be applicable to all dynamical systems and outperforms the others . therefore , further exploration and insights on the bandwidth selection problem in the dynamical system context certainly deserve future study . on the other hand , we also notice that due to the presence of dependence among observations generated by dynamical systems , the sample size usually needs to be large enough to approximate the density function well . this can be also seen from the plotted density functions in figs.[logistic_map_figure ] and [ gauss_map_figure ] with varying sample sizes . aside from the above observations , not surprisingly , from figs.[logistic_map_figure ] and [ gauss_map_figure ] , we also observe the _ boundary effect _ @xcite from the kernel density estimators for dynamical systems , which seems to be even more significant than the i.i.d case . from a practical implementation view , some special studies are arguably called for addressing this problem . _ ( i ) _ since the space of continuous and compactly supported functions @xmath322 is dense in @xmath323 , we can find @xmath324 such that @xmath325 therefore , for any @xmath326 , we have @xmath327 where @xmath173 is defined in and the last inequality follows from the fact that @xmath328 the above inequality is due to young s inequality ( 8.7 ) in @xcite . moreover , there exist a constant @xmath329 such that @xmath330 and a constant @xmath331 such that @xmath332 now we define @xmath333 by @xmath334}(\|x\| ) k(\|x\|)\end{aligned}\ ] ] and @xmath335 by @xmath336 then we have @xmath337 moreover , we have @xmath338 } \bigg ( \frac{\|x\|}{h } \bigg ) k \bigg ( \frac{\|x\|}{h } \bigg ) - k \bigg ( \frac{\|x\|}{h } \bigg ) \biggr| \ , \mathrm{d}x \\ & = \int_{\mathbb{r}^d } \bigl| { \boldsymbol{1}}_{[-r , r]}(\|x\| ) k(\|x\| ) - k(\|x\| ) \bigr| \ , \mathrm{d}x \\ & = \int_{h_r } k(\|x\| ) \ , \mathrm{d}x \leq \frac{\varepsilon}{9 \|\bar{f}\|_1}. \end{split}\end{aligned}\ ] ] finally , for @xmath201 , we have @xmath339 since @xmath340 is uniformly continuous , there exists a constant @xmath341 such that for all @xmath342 and @xmath343 , we have @xmath344 therefore , we obtain @xmath345 and consequently the assertion can be proved by combining estimates in and . _ ( ii ) _ the @xmath17-hlder continuity of @xmath2 tells us that for any @xmath280 , there holds @xmath346 we thus have completed the proof of theorem [ approximationerror ] . the following lemma , which will be used several times in the sequel , supplies the key to the proof of theorem [ comparisionnorms ] . [ generalrandestimate ] let the assumptions of theorem [ comparisionnorms ] hold and @xmath347 be defined in . then , for an arbitrary probability measure @xmath166 on @xmath8 , we have @xmath348 for a positive constant @xmath349 , we have @xmath350 on the other hand , it is easy to see that @xmath351 if and only if @xmath352 . now we set @xmath353 . in this case , if we additionally have @xmath354 , then @xmath355 . therefore , we come to the following estimate @xmath356 we thus have shown the assertion of lemma [ generalrandestimate ] . we decompose @xmath357 as follows @xmath358 combining the two estimates in and , we obtain the desired conclusion . to prove proposition [ kernelspacecn ] , we need the following lemmas . [ transnumber ] let @xmath220 and @xmath359 be metric spaces and @xmath360 be an @xmath17-hlder continuous function with constant @xmath53 . then , for @xmath221 and all @xmath326 we have @xmath361 let @xmath362 be an @xmath222-net of @xmath223 , that is , @xmath363 . for @xmath364 , we set @xmath365 . now , it only suffices to show that this gives a @xmath366-net of @xmath367 . in fact , supposing that @xmath368 , then there exists @xmath369 such that @xmath370 . this implies @xmath371 therefore , we have @xmath372 . that is , @xmath373 is a @xmath366-net of @xmath367 . this completes the proof of lemma [ transnumber ] . [ remark_covering number ] we remark that when @xmath80 is a banach space with the norm @xmath29 , then for any @xmath374 there holds @xmath375 [ differentnormcn ] let @xmath138 be another norm on @xmath8 . then for all @xmath230 $ ] we have @xmath376 it is a straightforward conclusion of proposition 1.3.1 in @xcite and lemma 6.21 in @xcite . [ holderconstant ] let @xmath11 be a @xmath135-dimensional smoothing kernel that satisfies conditions @xmath109 in assumption [ assump_kernel ] . let @xmath167 be the bandwidth parameter , and @xmath347 be defined in for any @xmath280 . then we have @xmath377 where @xmath53 is a positive constant . from the definition of @xmath347 and the fact that @xmath11 is a @xmath135-dimensional @xmath18-hlder continuous kernel , we have @xmath378 where @xmath53 is a positive constant . the desired conclusion is thus obtained . lemma [ holderconstant ] reveals that @xmath227 is the image of hlder continuous map @xmath379 with the constant @xmath380 . by lemmas [ transnumber ] and [ differentnormcn ] we obtain @xmath381 where @xmath229 is a constant independent of @xmath222 . this completes the proof of proposition [ kernelspacecn ] . the following bernstein - type exponential inequality , which was developed recently in @xcite , will serve as one of the main ingredients in the consistency and convergence analysis of the kernel density estimator . it can be stated in the following general form : [ bernsteininequality ] assume that @xmath90 @xmath382 is an @xmath80-valued stationary geometrically ( time - reversed ) @xmath0-mixing process on @xmath81 with @xmath117 be defined by for some semi - norm @xmath101 satisfying condition @xmath110 in assumption [ seminorm_assump ] , and @xmath93 . moreover , let @xmath383 be a function such that @xmath102 with @xmath384 and assume that there exist some @xmath385 , @xmath386 , and @xmath387 such that @xmath388 , @xmath389 , and @xmath390 . then , for all @xmath243 , @xmath391 , and @xmath392 with probability @xmath66 at least @xmath393 , there holds @xmath394 let the notations @xmath347 and @xmath207 be defined in and , respectively , that is , @xmath395 , and @xmath396 . we first assume that @xmath159 is fixed and then estimate @xmath397 by using bernstein s inequality in theorem [ bernsteininequality ] . for this purpose , we shall verify the following conditions : obviously , we have @xmath398 . moreover , simple estimates yield @xmath399 and @xmath400 finally , the first condition in assumption [ seminorm_assump ] and condition @xmath164 in assumption [ assump_kernel ] imply @xmath401 now we can apply the bernstein - type inequality in theorem [ bernsteininequality ] and obtain that for @xmath402 , for any fixed @xmath159 , with probability @xmath66 at most @xmath403 , there holds @xmath404 where @xmath405 consider the function set @xmath406 . we choose @xmath407 such that @xmath408 is a minimal @xmath409-net of @xmath410 with respect to @xmath411 . noticing the following relation @xmath412 we know that @xmath413 is an @xmath222-net of @xmath216 with respect to @xmath411 . note that here we have @xmath414 , since the net is minimal . from proposition [ kernelspacecn ] , we know that there exists a positive constant @xmath53 independent of @xmath222 such that @xmath415 . from the estimate in and a union bound argument , with probability @xmath66 at least @xmath416 , the following estimate holds @xmath417 by a simple variable transformation , we see that with probability @xmath66 at least @xmath246 , there holds @xmath418 recalling that @xmath408 is an @xmath409-net of @xmath227 , this implies that , for any @xmath419 , there exists @xmath420 such that @xmath421 . then we have @xmath422 and consequently @xmath423 , we have @xmath424 this together with inequality implies that for any @xmath419 , there holds @xmath425 consequently we have @xmath426 now recall that for @xmath427 and @xmath428 , hlder s inequality implies @xmath429 this tells us that @xmath430 moreover , there holds @xmath431 we now set @xmath432 and obtain @xmath433 . thus we have @xmath434 now we need to estimate the corresponding integral over @xmath32 . by definition we have @xmath435 from lemma [ generalrandestimate ] we obtain @xmath436 and @xmath437 since @xmath202 , we can construct a function @xmath108 with @xmath438 and there exists a function @xmath439 such that @xmath440 . applying bernstein inequality in theorem [ bernsteininequality ] with respect to this function @xmath108 , it is easy to see that when @xmath441 , with probability @xmath66 at least @xmath442 , there holds @xmath443 where @xmath444 this implies that with probability @xmath66 at least @xmath445 , there holds @xmath446 and consequently we obtain @xmath447 by combining estimates in and , and taking @xmath448 , we have accomplished the proof of theorem [ oracleinequalityl1 ] . [ remark_on_cx ] let us briefly discuss the choice of the function @xmath439 in the proof of theorem [ oracleinequalityl1 ] . for example , in the case @xmath449 , we can choose @xmath450 then we have @xmath451 and therefore , @xmath452 is well - defined . moreover , it is easily seen that even for smoother underlying functions classes like @xmath453 we can construct a function @xmath108 such that @xmath454 . recalling the definitions of @xmath347 and @xmath207 given in and , we have @xmath211 to prove the assertion , we first estimate @xmath397 for fixed @xmath159 using the bernstein inequality in theorem [ bernsteininequality ] . for this purpose , we first verify the following conditions : obviously , we have @xmath398 . then , simple estimates imply @xmath455 and @xmath456 finally , the first condition in assumption [ seminorm_assump ] and condition @xmath164 in assumption [ assump_kernel ] yield @xmath401 therefore , we can apply the bernstein inequality in theorem [ bernsteininequality ] and obtain that for @xmath245 , for any fixed @xmath159 , with probability @xmath66 at least @xmath457 , there holds @xmath458 where @xmath459 let us consider the following function set @xmath460 and choose @xmath461 such that @xmath462 is a minimal @xmath409-net of @xmath463 with respect to @xmath411 and @xmath464 . as in the proof of theorem [ oracleinequalityl1 ] , one can show that @xmath465 is an @xmath222-net of @xmath466 . again from proposition [ kernelspacecn ] we know that there holds @xmath467 . this in connection with implies that the following union bound @xmath468 holds with probability @xmath66 at least @xmath246 . for any @xmath469 , there exists a @xmath420 such that @xmath470 . then we have @xmath471 and consequently with probability @xmath66 at least @xmath246 , there holds @xmath472 for any @xmath469 . by setting @xmath432 , we obtain @xmath473 . thus , with probability @xmath66 at least @xmath246 , we have @xmath474 by taking the supremum of the left hand side of the above inequality over @xmath217 , we complete the proof of theorem [ oracleinequalityinftynorm ] . without loss of generality , we assume that @xmath475 . since @xmath476 , theorem [ approximationerror ] implies that @xmath477 . we set @xmath478 and we can also assume w.l.o.g that @xmath479 . moreover , there exists a constant @xmath480 such that @xmath481 for any @xmath482 , we select @xmath483 . then there exists a constant @xmath484 such that @xmath485 for all @xmath486 . on the other hand , with the above choice of @xmath487 , we have @xmath488 thus , for all @xmath489 , we have @xmath490 thus , following from theorem [ oracleinequalityl1 ] , when @xmath185 is sufficient large , for any @xmath326 , with probability @xmath66 at least @xmath491 , there holds @xmath492 therefore , with properly chosen @xmath493 , one can show that @xmath251 converges to @xmath2 under @xmath1-norm almost surely . we have completed the proof of theorem [ consistencyl1 ] . _ ( i ) _ combining the estimates in theorem [ oracleinequalityl1 ] and theorem [ comparisionnorms ] , we know that with probability @xmath66 at least @xmath494 , there holds @xmath495 let @xmath496 and later we will see from the choices of @xmath16 and @xmath487 that there exists some constant @xmath53 such that @xmath497 can be bounded by @xmath498 . therefore , with probability @xmath66 at least @xmath499 there holds @xmath500 by choosing @xmath501 _ ( ii ) _ similar to case _ ( i ) _ , one can show that with probability @xmath66 at least @xmath499 there holds @xmath502 by choosing @xmath503 _ ( iii ) _ from theorem [ comparisionnorms ] we see that with confidence @xmath499 , there holds @xmath504 where @xmath16 is chosen as @xmath505 the proof of theorem [ convergenceratesl1 ] is completed . the desired estimate is an easy consequence if we combine the estimates in theorem [ oracleinequalityinftynorm ] and theorem [ approximationerror ] _ ( ii ) _ and choose @xmath506 we omit the details of the proof here . in the present paper , we studied the kernel density estimation problem for dynamical systems admitting a unique invariant lebesgue density by using the @xmath0-mixing coefficient to measure the dependence among observations . the main results presented in this paper are the consistency and convergence rates of the kernel density estimator in the sense of @xmath1-norm and @xmath3-norm . with properly chosen bandwidth , we showed that the kernel density estimator is universally consistent . under mild assumptions on the kernel function and the density function , we established convergence rates for the estimator . for instance , when the density function is bounded and compactly supported , both @xmath1-norm and @xmath3-norm convergence rates with the same order can be achieved for general geometrically time - reversed @xmath0-mixing dynamical systems . the convergence mentioned here is of type with high probability " due to the use of a bernstein - type exponential inequality and this makes the present study different from the existing related studies . we also discussed the model selection problem of the kernel density estimation in the dynamical system context by carrying out numerical experiments .

we prove that the general symmetric tensor in of rank is identifiable , provided that is smaller than the generic rank . that is , its waring decomposition as a sum of powers of linear forms is unique . only three exceptional cases arise , all of which were known in the literature . our original contribution regards the case of cubics ( ) , while for we rely on known results on weak defectivity by ballico , ciliberto , chiantini , and mella .

1504.00547

we denote by @xmath0 the space of symmetric tensors on @xmath4 ; such tensors can be identified with homogeneous polynomials of degree @xmath5 in @xmath6 variables , which are also referred to as _ forms_. in this symmetric setting , the most natural tensor rank decomposition is the classical waring decomposition , which expresses a symmetric tensor as a sum of powers of linear forms . precisely , every form @xmath7 has a minimal expression @xmath8 where @xmath9 are linear forms @xcite ; the minimal number of summands @xmath1 is called the symmetric rank of @xmath10 , since in the correspondence between forms and symmetric tensors , powers of linear forms correspond to tensors of rank @xmath11 . a natural question concerns the number of summands required for representing a general form in @xmath12 . this problem is elementary for @xmath13 , which corresponds to the case of symmetric matrices . for @xmath14 , the question was answered by alexander and hirschowitz in @xcite . letting @xmath15 they proved that the general @xmath7 with @xmath16 has rank @xmath17 , which is called the _ generic rank _ , unless the space @xmath12 is one of the so - called defective cases @xmath18 for @xmath19 and @xmath20 , where the generic rank is @xmath21 . when the rank of a waring decomposition is strictly smaller than @xmath22 , we say that this decomposition is of _ subgeneric rank_. it is worth noting that , in our notation , being of _ subgeneric rank _ is not always equivalent to being of _ rank smaller than the one of a general tensor _ , because in the defective cases above a general tensor has rank strictly bigger than @xmath22 . the alexander hirschowitz theorem implies that the generic tensor of subgeneric rank admits only a finite number of alternative waring decompositions @xcite . in this paper , we shall be concerned with proving that the is an element of the @xmath1-secant variety of the veronese variety @xmath23 , which is an irreducible variety @xcite , this terminology is warranted . ] generic tensor of fixed subgeneric rank admits precisely one waring decomposition , modulo permutations of the summands and scaling by @xmath5-roots of unity . more precisely , the main result of this paper is the following theorem . [ thm_main ] let @xmath16 . the general tensor in @xmath0 of subgeneric rank @xmath24 with @xmath22 as in ( [ eqn_exp_rank ] ) has a _ unique _ waring decomposition , i.e. , it is _ identifiable _ , unless it is one of the following cases : 1 . @xmath25 , @xmath26 , and @xmath27 ; 2 . @xmath28 , @xmath29 , and @xmath30 ; 3 . @xmath2 , @xmath31 , and @xmath27 . in all of these exceptional cases , there are exactly two waring decompositions . the three exceptional cases were already known in the literature . the first two cases are classical ; see remark 4.4 in @xcite and remark 6.5 in @xcite . the third case was recently found by ranestad and voisin ; see the proof of lemma 4.3 in @xcite . a uniform treatment of these cases is presented in proposition [ prop : n235 ] . our original contribution establishes that there are no more exceptions to identifiability for cubics . the proof of theorem [ thm_main ] is based on the study of the geometric concepts of _ weak defectivity _ , developed in @xcite , and _ tangentially weak defectivity _ , developed in @xcite . indeed , from this point of view , the theorem can be reformulated in the following way , which is a result of independent interest . [ thm_main2 ] let @xmath16 , @xmath22 as in ( [ eqn_exp_rank ] ) , and @xmath24 . then , the common singular locus of the space of hypersurfaces of degree @xmath5 in @xmath32 that are singular at @xmath1 general points , consists of exactly these @xmath1 points , except in the following cases : 1 . @xmath25 , @xmath26 , and @xmath27 . the unique sextic plane curve singular at @xmath33 general points is a double cubic , so that its singular locus is an elliptic cubic curve ; 2 . @xmath28 , @xmath29 , and @xmath30 . the net of quartic surfaces singular at @xmath34 points consists of reducible quadrics , so that the common singular locus is the base locus of the pencil of quadrics through @xmath34 general points , which is an elliptic normal curve of degree @xmath35 ; 3 . @xmath2 , @xmath31 , and @xmath27 . the common singular locus of the pencil of cubic 4-folds singular at @xmath33 general points is the unique elliptic normal curve of degree @xmath36 through these @xmath33 points . furthermore , the above exceptional cases are the only instances where there exists a unique elliptic normal curve of degree @xmath6 in @xmath32 through @xmath1 general points . in this formulation , the theorem was already partially proved : the case @xmath37 was proved by chiantini and ciliberto @xcite ; for @xmath3 it was proved by ballico ( * ? ? ? * theorem 1.1 ) ; and for @xmath38 with @xmath39 it was proved by mella ( * ? ? ? * theorem 4.1 ) . consequently , the original contribution of this paper concerns the case of cubics , i.e. , @xmath2 , which we solve completely in the subgeneric case . this answers the question posed in remark 4.4 in @xcite . we notice that ballico @xcite proved an even stronger result for @xmath3 . namely , he showed that a general hypersurface of degree @xmath3 in @xmath32 that is singular in @xmath1 general points , is singular only at these @xmath1 points ( except for the exceptional cases ( 1 ) and ( 2 ) of theorem [ thm_main2 ] ) . this is equivalent to showing that the veronese variety @xmath23 is not @xmath1-weakly defective , while our result only says that it is not _ @xmath1-tangentially _ weakly defective . we wonder whether the above list also gives the classification of all @xmath1-weakly defective veronese varieties @xmath23 , even for @xmath2 . symmetric tensors of general rank are not expected to admit only a finite number of waring decompositions , because the expected dimension @xmath40 of the @xmath17-secant variety of the veronese variety @xmath23 may exceed the dimension @xmath41 of the embedding space @xmath12 . therefore , at least a curve s worth of alternative waring decompositions of a general symmetric tensor is anticipated in these cases . however , if @xmath42 is integer , then a general symmetric tensor is still expected to admit only a finite number of waring decompositions . the approach pursued in this paper , i.e. , proving not tangential weak defectivity , can not handle tensors of the generic rank . other approaches need to be considered in this setting . in fact , mella @xcite formulated a conjecture about the cases where the expression in ( [ eq : waring ] ) is still expected to be unique even for general symmetric tensors . in @xcite , further evidence for this conjecture was given ; in addition , the analogous problem for nonsymmetric tensors was also considered . even though the general symmetric tensor is not of subgeneric rank , the setting considered in this paper is nevertheless important in applications where one is mostly interested in the identifiability of symmetric tensors of subgeneric rank . for instance , anandkumar , ge , hsu , kakade , and telgarsky @xcite consider statistical parameter inference algorithms based on decomposing symmetric tensors for a wide class of latent variable models . the identifiability of the waring decomposition then ensures that the recovered parameters , which correspond with the individual symmetric rank-@xmath11 terms in waring s decomposition , are unique , and , thus , admit an interpretation in the application domain . the rank of the waring decomposition , in these applications , is invariably much smaller than the generic rank . as general sources on tensor decomposition , we refer to @xcite . in analogy to theorem [ thm_main ] , we mention that the results in @xcite give broad evidence to the analogous problem in the setting of nonsymmetric tensors , i.e. , that a general nonsymmetric tensor of subgeneric rank admits a unique tensor rank decomposition , unless it is one of the exceptional cases that have already been proved in @xcite . in the proofs by induction of several theorems , we rely on the software macaulay2 @xcite for proving the base cases . the two scripts we used are available as ancillary files in the arxiv submission of this paper . the content of the paper is the following . in section [ sec : newexample ] , we present a uniform treatment of the exceptional cases appearing in theorems [ thm_main ] and [ thm_main2 ] . remark [ rem : genesis ] also discusses our initial motivation for studying the topic of this paper . section [ sec : cubics ] contains the proof of the main theorem . thereafter , the connection between weakly defective varieties and the dual varieties to secant varieties , including a description of the dual varieties of all weakly defective examples appearing in theorem [ thm_main2 ] , is explored in section [ sec : dual ] . in particular , theorem [ thm : dualveronese ] contains the description of cubic hypersurfaces in @xmath43 that can be written as the determinant of a @xmath44 matrix with linear entries . in section [ sec : algorithm ] , we give an explicit criterion allowing to check if a given waring decomposition is unique . this algorithm is an extension to the symmetric case of the one provided in @xcite for general tensors . [ [ acknowledgements ] ] * acknowledgements * + + + + + + + + + + + + + + + + + + we want to thank c. ciliberto , i. domanov , j.m . landsberg , m. mella , l. oeding , k. ranestad and f. russo for useful discussions . in particular , k. ranestad pointed out the use of gale transforms to prove proposition [ prop : n235 ] and informed us about lemma 4.3 of @xcite which contains a proof of the third exceptional case in theorems [ thm_main ] and [ thm_main2 ] . we thank i. domanov for pointing out that improved specific identifiability results can be obtained by considering reshapings of the tensor . the first and second author wish to thank the simons institute for the theory of computing in berkeley , ca for generous support . the following classical result shows that the values @xmath45 , which appear in theorems [ thm_main ] and [ thm_main2 ] , have a special role for elliptic normal curves . [ prop : n235 ] assume that there are only finitely many elliptic normal curves passing through @xmath46 general points in @xmath32 . then , @xmath26 , @xmath47 , or @xmath48 and , correspondingly , @xmath49 . in these three cases , there is a unique elliptic normal curve in @xmath32 passing through @xmath50 general points . elliptic normal curves of degree @xmath51 in @xmath52 depend on @xmath53 parameters , which is the dimension of the space of sections of the normal bundle . the passage of the curve through a point in @xmath52 imposes @xmath54 conditions , which is the codimension of the curve . therefore , we may expect finitely many elliptic normal curves through @xmath46 _ general _ points in @xmath52 only if @xmath55 . this implies that @xmath56 divides @xmath57 , hence @xmath56 divides @xmath35 , which gives the values @xmath45 . moreover , @xmath58 . in case @xmath26 and @xmath59 , the elliptic curve is a plane cubic , and it is unique . in case @xmath29 and @xmath60 , an elliptic normal curve is a complete intersection of two quadrics . thus , if @xmath61 is the pencil of quadrics through @xmath34 general points @xmath62 , then @xmath63 is the unique elliptic normal curve through the @xmath64 s . in case @xmath31 and @xmath59 , the existence and the uniqueness of the curve was found by coble ( * ? ? ? * theorem 19 ) by applying a gale transform see @xcite for a nice review and reducing to the case @xmath26 and @xmath59 ; a modern treatment was given by dolgachev ( * ? ? ? * theorem 5.2 ) . next , we analyze the case of cubic hypersurfaces in @xmath43 that are singular at @xmath33 general points . [ prop : veneroni ] we have the following two results . a. the @xmath65-minors of a @xmath44 matrix with linear entries on @xmath43 define a ( sextic ) elliptic normal curve in @xmath43 . b. if @xmath66 is a ( sextic ) elliptic normal curve in @xmath43 , then the variety of secant lines @xmath67 is a complete intersection of two cubic hypersurfaces on @xmath43 , each one being the determinant of a @xmath44 matrix with linear entries on @xmath43 . part ( i ) is well known : the curve is obtained by cutting the segre variety @xmath68 , i.e. , the variety of @xmath44 matrices of rank @xmath11 , with a linear space @xmath43 . claim ( ii ) follows by ( * ? ? ? * lemma 2.9 ) . [ thm : elliptic9 ] let @xmath69 be general points in @xmath43 . let @xmath66 be the elliptic normal sextic curve through these points . a cubic that is singular at @xmath69 contains @xmath67 and is singular on @xmath66 . by proposition [ prop : veneroni ] , in the pencil of cubics containing @xmath67 , the general element is singular along @xmath66 . this pencil fills the space of cubics that are singular at @xmath69 , which is @xmath65-dimensional by the alexander hirschowitz theorem @xcite . these observations lead to a different proof of the third exceptional case in theorem [ thm_main ] . the general tensor in @xmath70 of rank @xmath33 has exactly two waring decompositions as sum of @xmath33 powers of linear forms . in the language of @xcite , we have to prove that the secant order of @xmath71 is @xmath65 . by , this is equal to the secant order of the @xmath33-contact locus @xmath66 , which corresponds to the third veronese embedding of an elliptic normal sextic curve in @xmath43 , by theorem [ thm : elliptic9 ] . thus , @xmath66 is an elliptic curve of degree @xmath72 in @xmath73 , whose secant order is @xmath65 by ( * ? ? ? * proposition 5.2 ) . [ coro : veneroni ] let @xmath26 , @xmath47 , or @xmath48 . the unique elliptic normal curve that is mentioned in proposition [ prop : n235 ] , which passes through @xmath74 general points in @xmath32 , can be constructed as the singular locus of a general hypersurface of degree @xmath75 that is singular in the @xmath74 points . the cases @xmath76 have already been considered in the proof of proposition [ prop : n235 ] . the case @xmath31 follows from theorem [ thm : elliptic9 ] . [ rem : genesis ] it was only at the completion of this manuscript that we became aware of ranestad and voisin s proof of the third exceptional case that appears in theorems [ thm_main ] and [ thm_main2 ] . our initial motivation for studying this problem arose because the third case was unexpectedly in our minds suggested by a computational analysis performed by the third author , who ran the algorithm that we present in section [ sec : algorithm ] , for hypersurfaces of degree @xmath5 in @xmath32 singular at the maximal number of random points , i.e. , @xmath77 with @xmath22 as in ( [ eqn_exp_rank ] ) , for all reasonably small values of @xmath5 , @xmath78 . it took a while to realize what happened , because this third case was missing in ( * ? ? ? * theorem 6.1.2 ) . actually , theorem 6.1.2 of @xcite only intended to collect previous results by ballico @xcite , ciliberto and chiantini @xcite , and mella @xcite , which are individually correct . the second author takes the responsibility to have first overlooked the assumption @xmath3 in summarizing and reporting the results of @xcite . for @xmath38 , theorem [ thm_main ] was known with the additional assumption @xmath79 ; see ( * ? ? ? * theorem 4.1 ) . from the theoretical proof that we present in section [ sec : cubics ] , we can conclude that the third case was the last exception . therefore , theorem 6.1.2 in @xcite remains true if the third case @xmath80 is added to the list of exceptions . exactly the same remark applies to the formulation of theorem 2.3 in @xcite and theorem 12.3.4.3 in @xcite . we informed the coauthors of @xcite of the problem , and they accepted the above conclusion . we turn our attention to the proof of theorem [ thm_main ] in the case of cubics , i.e. , @xmath2 . given @xmath78 , we define @xmath81 it is the generic rank for cubic polynomials for @xmath82 . in other words , a cubic polynomial on @xmath32 singular at @xmath83 general points vanishes identically for @xmath82 @xcite . some elementary algebra shows that @xmath84 if @xmath85 , while @xmath86 if @xmath87 . for the sake of future reference , let us state explicitly the following consequence of the alexander hirschowitz theorem @xcite . [ ahcubics ] the space of cubic hypersurfaces on @xmath32 that are singular at @xmath88 general points has dimension a. @xmath6 if @xmath85 , or b. @xmath89 if @xmath90 . in addition , the space of cubic hypersurfaces on @xmath32 , @xmath91 , that are singular at @xmath83 general points is empty . to complete the proof of the theorems [ thm_main ] and [ thm_main2 ] , it remains to show the following result , which refines theorem [ ahcubics ] . [ cubiche ] the space of cubic hypersurfaces in @xmath32 that are singular at @xmath88 general points has dimension a. @xmath6 if @xmath85 , or b. @xmath89 if @xmath87 , and , in addition , its common singular locus consists only of these @xmath88 points , provided that @xmath92 . in order to prove theorem [ cubiche ] , we may assume @xmath93 , since the cases with @xmath94 ( as well as the case of cubics in @xmath43 singular at @xmath34 points ) can be checked separately using the approach described in section [ sec : algorithm ] . the outline of our proof of theorem [ cubiche ] is as follows . in section [ sec : codim3 ] , we will prove case ( i ) by induction on subspaces of codimension @xmath47 , adopting an approach that is mainly inspired by ( * ? ? ? * section 5 ) , where an alternative proof of theorem [ ahcubics ] was given . to prove case ( ii ) , the aforementioned technique needs a modification . we will construct an inductive proof on subspaces of codimension @xmath47 and @xmath35 ; in the inductive step , we will rely , additionally , on the argument of case ( i ) . this strategy will be presented in section [ sec : codim4 ] . in the rest of this section , if @xmath95 is a set of simple points in @xmath32 and @xmath96 is a linear subspace , we denote with @xmath97 the space of degree @xmath5 polynomials in @xmath98 vanishing at all of the points in @xmath95 . moreover , if @xmath99 is a a set of double ( singular ) points , we denote by @xmath100 the space of degree @xmath5 polynomials in @xmath98 vanishing on all of the points in @xmath101 and whose derivatives vanish on all of the points in @xmath99 . the notation @xmath102 denotes the subspace of codimension @xmath103 whose ideal is @xmath104 . we start by proving three auxiliary results . [ proprep ] let @xmath93 , and let @xmath105 be general subspaces of codimension @xmath47 . let @xmath106 , respectively @xmath107 , with @xmath108 be three general points on @xmath109 , respectively @xmath110 . let @xmath111 with @xmath112 be two general points on @xmath113 . then , the space of cubic hypersurfaces in @xmath32 that contain @xmath114 and that are singular at the eight points @xmath115 has dimension @xmath47 . furthermore , the common singular locus is contained in @xmath116 . the base cases @xmath117 , @xmath118 and @xmath34 can be proved with the macaulay2 script ` generic - identifiability.m2 ` that is provided as an ancillary file to the arxiv version of this paper . using this software , we may compute the following dimensions : @xmath119 so that the claim on the codimension follows . the code also proves the statement about the singular locus . for @xmath120 , the statement follows by induction on @xmath78 . indeed , we may choose coordinates such that @xmath121 , @xmath122 , @xmath123 . in this setting it is clear that there are no quadrics that contain @xmath116 , and moreover every cubic containing @xmath116 is a cone with vertex in @xmath124 . thus , for a general hyperplane @xmath125 , the castelnuovo sequence ( see ( * ? ? ? * equation ( 1 ) ) ) induces an inclusion @xmath126 hence , if we specialize the eight points to the hyperplane @xmath127 , we get an inclusion @xmath128 then , our statement follows by induction . the singular locus is a cone with vertex @xmath129 over the singular locus of the base case @xmath130 . following the output of the software for the case @xmath130 , we can guess the common singular locus of the cubic hypersurfaces in @xmath131 , for @xmath132 . it turns out that in some examples but we believe in general it is given by the union of the three linear subspaces @xmath133 , @xmath134 , @xmath135 and by @xmath34 linear subspaces of codimension @xmath118 , each containing one of the @xmath34 points , and three of them contained in @xmath109 , three of them contained in @xmath110 , and two of them contained in @xmath113 . [ proprep2 ] let @xmath136 , and let @xmath137 be subspaces of codimension three . let @xmath106 , respectively @xmath107 , with @xmath138 be general points on @xmath139 , respectively @xmath140 . let @xmath141 be general points . then , the space of cubic hypersurfaces in @xmath32 containing @xmath142 and singular along the set of @xmath143 points @xmath144 has dimension @xmath6 . its common singular locus contains the linear space @xmath133 and is @xmath145-dimensional at the points @xmath146 and @xmath147 . the base cases @xmath31 , @xmath36 , and @xmath118 can be proved with the macaulay2 script ` generic - identifiability.m2 ` . running the software , we find the following dimensions @xmath148 these values indeed correspond to the claimed dimensions . for @xmath149 , the statement follows by induction from @xmath150 to @xmath78 . indeed , given a third general subspace @xmath151 of codimension @xmath47 , we get the exact sequence @xmath152 where the dimensions of the three spaces in the sequence are respectively @xmath153 , @xmath154 , and @xmath155 . let us specialize @xmath156 of the points @xmath157 to @xmath134 , @xmath156 of the points @xmath158 to @xmath135 , and the two points @xmath159 to @xmath113 . then , we obtain a sequence @xmath160 where the trace @xmath161 satisfies the assumptions on @xmath162 , so that we can apply the induction . notice that the residual ( left ) space satisfies the hypotheses of proposition [ proprep ] and has dimension @xmath47 . since the common singular locus of the cubics containing @xmath142 and singular at @xmath163 must be contained in the common singular locus of the leftmost @xmath47-dimensional space , it follows by proposition [ proprep ] that its components through @xmath146 and @xmath147 must be contained in @xmath113 . after the degeneration , the space of cubics @xmath164 still has dimension at most @xmath165 by induction . hence , by semicontinuity it follows that its dimension is indeed equal to @xmath6 . the common singular locus can not be positive dimensional at points @xmath146 and @xmath147 , because otherwise it should be of positive dimension in the trace ( right space ) , where by induction we know that it is @xmath145-dimensional . [ proprep3 ] let @xmath93 , and let @xmath166 be a subspace of codimension @xmath47 . if @xmath85 , then the space of cubic hypersurfaces in @xmath32 that contain @xmath109 and that are singular at @xmath167 general points @xmath157 and at @xmath78 general points @xmath168 has dimension @xmath6 . moreover , its common singular locus is @xmath145-dimensional at the @xmath78 points @xmath64 . the statement can be checked for @xmath117 , @xmath118 with the macaulay2 script ` generic - identifiability.m2 . ` let @xmath120 and @xmath85 . consider the sequence @xmath169 where @xmath110 is a general subspace of codimension @xmath47 . denoting by @xmath99 the union of the double points supported at the points @xmath106 and @xmath64 , we get @xmath170 we specialize @xmath171 of the points @xmath106 to @xmath133 and @xmath172 of the points @xmath64 to @xmath110 . we can assume that at least one point @xmath64 that is a contained in a positive dimensional component of the singular locus is not specialized . thus , we left @xmath172 general points on @xmath109 and @xmath65 general points in @xmath32 . let us note that we can not apply induction from @xmath150 to @xmath78 to determine the dimension of the right space contrary to the strategy that was employed in the proof of the foregoing propositions in this section because then we would have to specialize @xmath150 of the points @xmath64 to @xmath110 , hereby losing control over the singular locus . instead , we note that we can immediately use proposition 5.4 of @xcite ( in @xmath173 ) on the trace ( right space ) ; it turns out to be empty . on the residual ( left space ) , proposition [ proprep2 ] can be invoked , proving that it has dimension @xmath6 . now , if the singular locus would have a positive dimensional component , then , since the dimension of the space of cubics is constant along the specialization ( it equals @xmath6 ) , we would get a deformation of the singular locus , which should be of positive dimension at every point . this , however , contradicts proposition [ proprep2 ] , hereby concluding the proof . we are now ready to prove the first part of theorem [ cubiche ] . we fix a linear subspace @xmath174 of codimension @xmath47 and consider the exact sequence @xmath175 where @xmath176 is the space of cubic polynomials on @xmath32 and the quotient space @xmath177 is isomorphic to the space of cubic polynomials on @xmath109 . then , we specialize to @xmath109 as many points as possible in such a way that the trace with respect to @xmath109 imposes independent conditions on the cubics of @xmath109 . to be precise , we have @xmath178 double points and we specialize @xmath179 of them to @xmath109 , leaving @xmath78 points outside . then , the result follows from theorem 5.1 of @xcite on the trace ( right space ) , which turns out to be empty , and by proposition [ proprep3 ] on the residual ( left space ) , which has dimension @xmath6 . if the contact locus has positive dimension , then , since the dimension of the space of cubics is constant and equal to @xmath51 in the degeneration , we would get a deformation of the singular locus with a positive dimension at every point , contradicting proposition [ proprep3 ] and concluding the proof . for proving the second case in theorem [ cubiche ] , we need to introduce several other auxiliary results on configurations that involve subspaces of codimension three and four . these configurations are covered in propositions [ proprep : codim433 ] through [ proprep3:codim4 ] . [ proprep : codim433 ] let @xmath93 , and let @xmath105 be general subspaces of codimension @xmath35 , @xmath47 , and @xmath47 , respectively . let @xmath180 be general points on @xmath109 . let @xmath107 , respectively , @xmath111 with @xmath181 be four general points on @xmath110 , respectively @xmath113 . then , the space of cubic hypersurfaces in @xmath32 that contain @xmath116 and are singular at the @xmath182 points @xmath183 is empty . the proof is similar to the proof of proposition [ proprep ] . the macaulay2 code proves the base cases @xmath117 , @xmath118 , @xmath34 and @xmath33 . for @xmath120 , we may choose coordinates such that @xmath184 , @xmath185 , and @xmath186 . then , the statement follows by induction on @xmath78 . indeed , as in the proof of proposition [ proprep ] , the space @xmath187 is empty , thus for a general hyperplane @xmath125 , the castelnuovo sequence induces an embedding @xmath188 and , moreover , every cubic in the left space is a cone with vertex at @xmath124 . hence , by specializing the @xmath182 points to the hyperplane @xmath127 , we get : @xmath189 then , the statement follows by induction . [ proprep2:codim433 ] let @xmath190 , let @xmath191 , and let @xmath192 be subspaces of codimension @xmath35 and @xmath47 , respectively . let @xmath106 with @xmath193 be general points on @xmath139 . let @xmath107 with @xmath194 be general points on @xmath140 . then , the space of cubic hypersurfaces in @xmath32 that contain @xmath195 and are singular at all the @xmath106 s and @xmath107 s and at four general points @xmath196 , is empty . the macaulay2 script proves the base case @xmath197 . for @xmath198 with @xmath199 , the statement follows by induction from @xmath150 to @xmath78 . indeed , given a third general subspace @xmath151 of codimension @xmath47 , we get the exact sequence @xmath200 where the dimensions of the three spaces in the sequence are respectively @xmath201 , @xmath202 and @xmath203 . let @xmath99 denote the union of the double points supported at the @xmath64 s , @xmath106 s and @xmath107 s . assume that we specialize @xmath204 of the points @xmath157 to @xmath134 , @xmath205 of the points @xmath158 to @xmath135 , and the four points @xmath206 to @xmath113 . then , we obtain a sequence @xmath207 where the trace @xmath208 satisfies the assumptions on @xmath162 , so that we can apply induction . then , we may conclude , as the residual ( left space ) satisfies the hypotheses of proposition [ proprep : codim433 ] , and , consequently , it is empty . [ proprep3:codim433 ] let @xmath190 , @xmath191 , and @xmath166 be a subspace of codimension four . then , the space of cubic hypersurfaces in @xmath32 that are singular at @xmath209 general points @xmath106 on @xmath139 ( and , thus , contain @xmath139 , by theorem [ ahcubics ] ) and at @xmath210 general points @xmath211 is empty . the macaulay2 script ` generic - identifiability.m2 ` proves the case @xmath197 . for @xmath212 with @xmath199 , the statement follows by the sequence @xmath213 where @xmath110 is a general subspace of codimension @xmath47 . if we denote by @xmath99 the union of the double points supported at the points @xmath106 and @xmath64 , then we get the sequence @xmath214 then , we specialize @xmath215 of the points @xmath106 to @xmath133 and @xmath216 of the points @xmath64 to @xmath110 . the trace ( right space ) contains exactly @xmath217 double points and turns out to be empty by induction . thus , there remain @xmath150 general points on @xmath109 and @xmath35 general points on @xmath32 ; we can then use proposition [ proprep2:codim433 ] on the residual ( left space ) to conclude . [ proprep : codim4 ] let @xmath149 , and let @xmath105 be general subspaces of codimension respectively @xmath35 , @xmath35 , and @xmath47 . let @xmath106 , respectively @xmath107 , with @xmath218 be general points on @xmath109 , respectively @xmath110 . finally , let @xmath111 with @xmath219 be general points on @xmath113 . then , the space of cubic hypersurfaces in @xmath32 that contain @xmath220 and that are singular at the @xmath221 points @xmath222 has dimension @xmath11 . in other words , there is a unique cubic hypersurface @xmath223 through @xmath114 and singular at @xmath99 . furthermore , the singular locus of @xmath223 is contained in @xmath116 . the proof is similar to the proof of proposition [ proprep ] . the macaulay2 code ` generic - identifiability.m2 ` proves the base cases @xmath130 , @xmath33 , and @xmath224 . for @xmath225 , we may choose coordinates such that @xmath184 , @xmath226 , and @xmath227 ; then , the statement follows by induction on @xmath78 . indeed , as in the proofs of proposition [ proprep ] and proposition [ proprep : codim433 ] , we let @xmath125 be a general hyperplane , so that the castelnuovo sequence induces the inclusion @xmath228 because the space @xmath229 is empty . hence , by specializing the @xmath221 points on the hyperplane @xmath127 , we get an exact sequence : @xmath230 now the statement follows by induction . [ proprep2:codim4 ] let @xmath149 , @xmath90 , and @xmath231 be subspaces of codimension @xmath35 . let @xmath106 and @xmath107 , where @xmath232 , be general points on @xmath139 and @xmath140 , respectively . then , the space of cubic hypersurfaces in @xmath32 that contain @xmath195 and are singular at the @xmath233 points @xmath234 , @xmath235 , and at an additional set of five general points @xmath168 , @xmath236 , has dimension @xmath237 . furthermore , its common singular locus , which contains the linear space @xmath133 , is @xmath145-dimensional at each of the points @xmath238 . the case @xmath130 is handled in the ` generic - identifiability.m2 ` macaulay2 script . for @xmath239 with @xmath240 , the statement follows by induction on @xmath46 . given a third general subspace @xmath113 of codimension @xmath47 , we get the exact sequence @xmath241 where the dimensions of the three spaces in the sequence are respectively @xmath242 , @xmath243 and @xmath244 . let @xmath99 denote the union of the double points supported at @xmath245 , @xmath106 and @xmath107 with @xmath246 . then , we specialize @xmath247 of the points @xmath157 to @xmath134 , @xmath247 of the points @xmath248 to @xmath135 , and the points @xmath245 to @xmath113 . we thus obtain a sequence @xmath249 where the trace @xmath161 satisfies the assumptions on @xmath162 , so that we can apply induction . then , the residual ( left space ) satisfies the hypotheses of proposition [ proprep : codim4 ] and has dimension one . moreover , the common singular locus has to be contained in the common singular locus of the left @xmath11-dimensional space . after the degeneration , the space @xmath250 still has dimension less than or equal to @xmath251 , by induction , and , therefore , its dimension equals @xmath237 , by semicontinuity . the common singular locus can not be positive dimensional at the points @xmath245 , because otherwise it should be positive dimensional in the trace ( right space ) , while we know that it is @xmath145-dimensional there by induction . [ proprep3:codim4 ] let @xmath149 , @xmath90 , and @xmath166 be a subspace of codimension @xmath35 . then , the space of cubic hypersurfaces in @xmath32 that contain @xmath109 and are singular at @xmath209 general points @xmath157 and at @xmath252 general points @xmath168 has dimension @xmath237 . furthermore , its singular locus is of dimension @xmath145 at all of the points @xmath64 . the statement follows by the sequence @xmath253 where @xmath110 is a general subspace of codimension @xmath35 . denoting by @xmath99 the union of the double points supported at the points @xmath106 s and @xmath64 s , we get @xmath254 suppose that the singular locus would be of positive dimension at one of the @xmath64 s , say , at @xmath255 . then , we specialize @xmath256 of the points @xmath106 to @xmath133 and @xmath257 of the points @xmath258 to @xmath110 . the trace ( right space ) contains exactly @xmath259 double points and is empty because of proposition [ proprep3:codim433 ] . there remain @xmath257 general points on @xmath109 and @xmath48 general points one of which is @xmath260on @xmath32 . we can use proposition [ proprep2:codim4 ] on the residual ( left space ) , which has dimension @xmath237 . by assumption , the singular locus has positive dimension at @xmath260 ; however , the dimension of the space of cubics is constant and equal to @xmath237 through the degeneration , so that we have a deformation of the singular locus , which is of positive dimension at every point . in particular , the singular locus will be of positive dimension at all the points that were not specialized to @xmath110 , hereby contradicting proposition [ proprep2:codim4 ] . we conclude that our initial assumption must have been false , so that no general points @xmath261 can exist where the singular locus is of positive dimension . we fix a codimension four linear subspace @xmath262 and we use the exact sequence @xmath263 where , as above , @xmath176 is the space of cubic polynomials on @xmath32 and the quotient space @xmath177 is isomorphic to the space of cubic polynomials on @xmath109 . we specialize @xmath259 points on @xmath109 , leaving @xmath252 points outside . then , the result follows from theorem 5.1 of @xcite on the trace ( right space ) , which turns out to be empty and by proposition [ proprep3:codim4 ] on the residual ( left space ) . if the contact locus would have a positive dimension , then , since the dimension of the space of cubics is constant and equal to @xmath237 in the degeneration , we would get a deformation of the singular locus , which should be of positive dimension at every point ; however , this contradicts proposition [ proprep3:codim4 ] , hereby concluding the proof . denote by @xmath264 the tangent space to the projective variety @xmath265 at the point @xmath266 . following the notation of @xcite we say that @xmath267 is _ not @xmath46-weakly defective _ if the general hyperplane @xmath127 containing the linear span of the tangent spaces at @xmath46 general points @xmath268 , i.e. , @xmath269 , is tangent to @xmath267 only at finitely many points . this is equivalent with saying that the @xmath46-contact locus with respect to @xmath270 and @xmath127 is zero - dimensional . for any projective variety @xmath267 , we will denote by @xmath271 the dual variety to @xmath267 . note that the dual of the secant variety @xmath272 contains the points corresponding to hypersurfaces of degree @xmath5 in @xmath32 with @xmath46 general singular points , and it has codimension @xmath273 , where @xmath46 is the expected value for the codimension . let @xmath274 and let @xmath275 be the @xmath46-secant variety of @xmath267 . then , the following are equivalent : a. the general hyperplane @xmath127 containing @xmath276 for general @xmath270 is tangent to @xmath267 only at @xmath270 , i.e. , the @xmath46-contact locus with respect to @xmath270 and @xmath127 consists exactly of the points @xmath270 , b. @xmath267 is not @xmath46-weakly defective , and c. @xmath277^{\vee}=n - k$ ] , that is a general hyperplane tangent to @xmath275 is tangent along a linear space of projective dimension @xmath278 . \(i ) @xmath279 ( ii ) follows from . ( ii ) @xmath279 ( iii ) follows from terracini s lemma . it is interesting to describe the dual varieties of @xmath280 in the exceptional cases of theorems [ thm_main ] and [ thm_main2 ] . they have dimension smaller than expected . [ thm : dualveronese ] the following dual varieties correspond to the exceptional cases appearing in theorems [ thm_main ] and [ thm_main2 ] . a. @xmath281 contains the plane sextics which are double cubics . it has codimension @xmath72 . b. @xmath282 contains the quartic surfaces which are reducible in a pair of quadrics . it has codimension @xmath283 . c. @xmath284 contains the cubic @xmath35-folds which can be written as the determinant of a @xmath44 matrix with linear entries . it has codimension @xmath72 . to compute the dimension in third case , note that the hilbert scheme of elliptic normal sextic curves in @xmath43 has dimension @xmath201 . so the cubic hypersurfaces coming from this construction have dimension @xmath285 , and @xmath286 . we remark that the defective veronese varieties according to the classification of alexander and hirschowitz @xcite ( see @xcite for the equations of the defective secant varieties ) yield the following dual varieties a. @xmath287 , for @xmath19 , contains quartic hypersurfaces which are double quadrics . it has codimension @xmath288 . b. @xmath289 contains cubic @xmath47-folds which can be written as the determinant of a @xmath44 symmetric matrix with linear entries . it has codimension @xmath221 . indeed , it is birational to the hilbert scheme of quartic rational normal curves which has dimension @xmath290 . while the generic symmetric tensor of subgeneric rank is expected to admit a unique waring decomposition , _ specific _ tensors , whose waring decomposition is assumed to be known , may admit multiple decompositions . we proceed by presenting an approach for certifying specific identifiability of symmetric tensors of small rank by checking not tangential weak defectivity of the @xmath1-secant variety of a veronese variety in the given point . the strategy is an adaption of the algorithm from @xcite to the setting of identifiability with respect to the veronese variety @xmath291 . as such , the presented condition will only be a sufficient condition ; that is , if the criterion does not apply , then the outcome of the test is inconclusive . on the other hand , if the criterion applies , then the given input tensor is @xmath1-identifiable and of symmetric rank @xmath1 . throughout this section , it is assumed that we are handed a waring decomposition @xmath292 wherein the point @xmath293 is the degree @xmath5 veronese embedding of the vector @xmath294 . in other words , we know the points @xmath64 appearing in the decomposition . the goal only consists of certifying that @xmath295 is @xmath1-identifiable the decomposition(s ) are not sought . to this end , the strategy in @xcite suggests a two - step procedure : prove that @xmath295 is a smooth point , and verify the hessian criterion . in principle , the method can be applied for all tensors of subgeneric rank , however , in practice the range of applicability of the algorithm in section [ sec_algorithm ] is restricted by the lack of good techniques for certifying smoothness . for this reason , the results of @xcite may apply in a wider range than the results we present , by combining reshapings of higher - order tensors into tensors of order three with proposition 1.32 in @xcite , as we were kindly informed by its authors . we will nevertheless present an example of an identifiable tensor in @xmath296 of rank @xmath224 whose identifiability can not be proved by the state - of - the - art specific identifiability criteria from the literature . it is important to stress that we discuss the general setting of degree @xmath297 veronese embeddings . we will restrict our attention to nondefective @xmath1-secants of @xmath298 , because identifiability will not hold for general tensors on a defective @xmath1-secant variety . this is the interesting setting , because the alexander hirschowitz theorem @xcite stipulates that most @xmath299 are nondefective . [ infinitepoints ] let @xmath291 be a nondefective veronese variety , and let @xmath300 with @xmath22 as in ( [ eqn_exp_rank ] ) . assume that we are given a _ point @xmath301 if the linear span of the tangent spaces to @xmath298 at the @xmath64 s , i.e. , @xmath302 has the expected dimension , i.e. , @xmath303 , and if , in addition , the _ @xmath1-tangential contact locus _ @xmath304 is zero - dimensional at every @xmath305 , then @xmath295 is @xmath1-identifiable , @xmath1 is its symmetric rank , and @xmath306 is its unique decomposition . the proof is obtained by repeating the proofs of ( * ? ? ? * lemma 4.3 , lemma 4.4 , and theorem 4.5 ) , therein substituting the segre variety with the veronese variety @xmath298 . we present a simplification of the proof of ( * ? ? ? * theorem 4.5 ) . there is an open neighborhood of @xmath307 consisting of points for which smoothness , terracini s lemma @xcite , and the absence of a contact locus will hold . the variety must thus be generally identifiable , hence the projection @xmath308 onto the first factor of the usual abstract secant variety @xmath309 is a birational morphism . after we find another decomposition @xmath310 , we get that the fiber @xmath311 contains the two points @xmath312 and @xmath313 . terracini s lemma implies that the connected component of the fiber passing through @xmath312 can not be positive dimensional , hence , it contains just this unique point . since @xmath308 is birational and @xmath295 is a smooth point of @xmath314 , we have a contradiction with zariski s main theorem . we note a minor omission in the formulation of theorem 4.5 in @xcite , where we forgot to include the condition that @xmath315 should be of the expected dimension . it is clear from the proof of aforementioned theorem that this condition must hold , as can be understood from the invocation of ( * ? ? ? * lemma 4.3 ) . if one chooses @xmath1 _ random _ points @xmath316 , then lemma [ infinitepoints ] may be invoked to prove generic @xmath1-identifiability . in this way , one can handle the cases @xmath317 , @xmath318 and @xmath319 , which were not covered by the proof in section [ sec : cubics ] . the case @xmath320 is trivial , because there is only one point , which is naturally identifiable . for practically verifying lemma [ infinitepoints ] , we need a sufficiently explicit description of the @xmath1-contact locus . this is obtained as follows . interpreting a point @xmath321 as a power of a linear form , say @xmath322 where @xmath323 is a basis of @xmath32 , it follows immediately that the tangent space is given by @xmath324 if we choose the standard monomial basis @xmath325 for @xmath326 , then this tangent space can be represented in a straightforward manner as a @xmath327 matrix of constants , say @xmath328 . the cartesian equations of @xmath315 may then be constructed by computing the kernel of the matrix @xmath329^t$ ] . the number of such equations should be precisely @xmath330 ; otherwise , the first condition in lemma [ infinitepoints ] concerning the dimension of the tangent space would be violated . let us denote the cartesian equations as @xmath331 where the vector @xmath332_{0\le i_1 \le i_2 \le \cdots \le i_d\le n}$ ] is the @xmath333th basis vector of the kernel of the matrix of constants @xmath334 . for imposing that a point @xmath335 is contained in @xmath315 , it should obey the cartesian equations , i.e. , @xmath336 for all @xmath337 . that is , the equations ( [ eqn_eqns_in_deal ] ) define the ideal - theoretic equations for @xmath338 . it similarly follows that deriving the equations ( [ eqn_eqns_in_deal ] ) with respect to @xmath339 and substituting @xmath339 for , respectively , @xmath340 results in the ideal - theoretic equations of the intersection @xmath341 ; naturally , the @xmath340 should be treated as new variables . the number of equations thus constructed equals @xmath342 . to determine that @xmath343 is zero - dimensional at each @xmath64 , it suffices to verify that the codimension of the tangent space , i.e. , the derivative of the equations of the ideal , is @xmath78 at each of the @xmath64 s . this tangent space can be represented by a matrix @xmath127 of size @xmath344 , which contains only constants when it is evaluated at one of the @xmath64 s . the rank of @xmath127 coincides with the dimension of the contact locus and can be computed using simple linear algebra . as was remarked in @xcite , @xmath127 can be interpreted as a `` stacked hessian '' matrix @xmath345 $ ] , wherein @xmath346 is the hessian matrix of partial derivatives @xmath347_{i , j=0}^n = \left [ \frac{\partial^2}{\partial x_j \partial x_i } q_k(x_0,x_1,\ldots , x_n ) \right]_{i , j=0}^n;\end{aligned}\ ] ] this is the reason why we call the above approach of verifying lemma [ infinitepoints ] the _ hessian criterion_. the hessian criterion in lemma [ infinitepoints ] may only be applied to smooth points of @xmath299 . one approach for proving smoothness consists of verifying that the local equations of @xmath299 are of the expected degree . such equations are known in the case when the number of terms @xmath1 in the symmetric decomposition is sufficiently small . a standard nontrivial set of local equations is generated by the @xmath348-minors of the usual symmetric flattenings ; see ( * ? ? ? * theorem 7.3.3.3 ) and ( * ? ? ? * theorems 4.5a and 4.10a ) . for veronese embeddings of odd degree , the young flattenings from ( * ? ? ? * section 4 ) apply in a wider range than the standard symmetric flattenings ; however , they are more involved to explain and implement . our discussion will focus on the simple symmetric flattenings , which can still handle a respectable number of cases for veronese embeddings of degree at least four . for degree three veronese embeddings , the young flattenings that were described in @xcite should be employed . the strategy that we present for proving that @xmath295 corresponds to a smooth point is , essentially , based on ( * ? ? ? * theorems 4.10a and 4.5a ) and ( * ? ? ? * theorem 7.3.3.3 ) and consists of obtaining local equations of the @xmath1-secant variety @xmath299 in @xmath295 . crucial to this approach are the _ symmetric flattenings _ , which , we recall , may be defined as follows . let @xmath349 , then we can define the map @xmath350 we have the following . [ lem_smoothness ] let @xmath291 be the veronese variety , let @xmath351 , and let @xmath352 . assume that we are given a point @xmath353 let @xmath354 be the following linear space : @xmath355 i.e. , the symmetric product of the kernel and the complement of the image of @xmath356 . if @xmath357 then @xmath295 is a smooth point of @xmath358 . the subspace @xmath354 is the normal space at @xmath295 of the locus of @xmath348-minors of the catalecticant matrix @xmath359 ; see , for example , ( * ? ? ? 5.3.3.1 ) . if @xmath354 has the expected dimension @xmath360 , then the locus of @xmath348-minors of the catalecticant matrix @xmath359 is smooth at @xmath295 and of the expected dimension @xmath361 . the @xmath1-secant variety @xmath358 is contained in that locus , being of the expected dimension @xmath361 by the alexander - hirschowitz theorem , so it too has to be smooth at @xmath295 . a natural question concerning the foregoing lemma concerns the maximum value of @xmath1 for which it can be applied . that is , if we pick a sufficiently general smooth point @xmath362 , what is the maximum value of @xmath1 for which lemma [ lem_smoothness ] can prove that @xmath295 is , indeed , smooth ? a lower bound follows immediately from the work of iarrobino and kanev ( * ? ? ? * theorem 4.10a ) : [ prop_ik_lower_bound ] let @xmath291 be the veronese variety , let @xmath351 , and let @xmath363 then , lemma [ lem_smoothness ] can be applied to all points of an irreducible component of @xmath299 minus some zariski - closed set . in table [ tab_bounds ] , some values of the lower bound in proposition [ prop_ik_lower_bound ] are tabulated along with a sharp maximum value of @xmath1 for which the equations generated by lemma [ lem_smoothness ] generate an irreducible component of @xmath299 . the values of this alleged sharp upper bound were computed by taking random points on this variety and verifying lemma [ lem_smoothness ] ; as such , they are only true with high probability . it is clear from the table that the lower bound in proposition [ prop_ik_lower_bound ] is not sharp . .the maximum value @xmath1 for which lemma [ lem_smoothness ] applies to all points in an irreducible component of @xmath299 minus some zariski - closed set is displayed as the middle set of columns ( @xmath364 ) for each degree @xmath365 of the veronese embedding @xmath291 . the left set of columns ( @xmath366 ) shows the lower bound from proposition [ prop_ik_lower_bound ] , for every @xmath5 . the right set of columns ( @xmath367 ) shows the maximum value of @xmath1 for which kruskal s criterion is applicable , for every @xmath5 . a @xmath368 indicates that the value could not be computed within a reasonable time . values displayed in boldface indicate the widest range for @xmath1 for a particular combination of the degree @xmath5 and size @xmath78 . [ cols="^,^,^,^,^,^,^,^,^,^,^,^,^,^,^,^,^,^,^,^,^,^,^ " , ] an implementation in macaulay2 of the above sufficient condition for smoothness based on a symmetric flattening is included in the ` specific - identifiability.m2 ` file that is provided with the arxiv version of this paper . in addition , this file contains an implementation of a smoothness test based on young flattenings for veronese embeddings of degree three . for the sake of completeness , we present an algorithm that attempts to prove the identifiability of a given waring decomposition @xmath369 , where @xmath291 , by checking the sufficient conditions in lemma [ infinitepoints ] and lemma [ lem_smoothness ] . it operates as follows . 1 . construct a matrix representation of the span @xmath370 . if @xmath371 , then the algorithm terminates , claiming that it can not prove the identifiability of @xmath295 . 2 . construct the symmetric flattening @xmath372 for @xmath351 . if @xmath373 , then the algorithm terminates , claiming that it can not prove identifiability of @xmath295 . 3 . compute the matrix @xmath374 . if @xmath375 , then the algorithm terminates , claiming that it can not prove identifiability of @xmath295 . 4 . compute a basis of the kernel of @xmath315 . denote the number of equations by @xmath376 . 5 . for every point @xmath64 , @xmath377 , perform the following : 1 . construct the hessians @xmath346 for @xmath378 evaluated at the point @xmath64 and stack them into the matrix @xmath127 . 2 . if @xmath379 , then the contact locus is of positive dimension at @xmath64 . the algorithm halts , claiming that it can not prove identifiability of @xmath295 . 6 . the algorithm proclaims that the waring decomposition @xmath307 is unique and that @xmath295 is a smooth point of @xmath299 . it is instructive to investigate the largest value of @xmath1 for which the above algorithm may be expected to prove identifiability of a sufficiently general point @xmath362 on a generically identifiable veronese variety @xmath298 . as the hessian criterion applies for all tensors of subgeneric rank , it follows that the range of applicability is bounded only by the smoothness test . that is , the highlighted columns in table [ tab_bounds ] contain the relevant values . a popular criterion for testing identifiability that is applicable for veronese embeddings of any degree @xmath297 is the so - called kruskal condition @xcite . let @xmath293 be some specific points . in the symmetric setting , kruskal s condition states that if @xmath380 where @xmath46 is the largest number such that every subset of @xmath381 consisting of @xmath46 vectors is linearly independent . for points in general configuration , the maximum value for @xmath46 is thus @xmath6 . a comparison between the proposed criterion for specific identifiability and kruskal s criterion is also featured in table [ tab_bounds ] . consider the following @xmath224-term waring decomposition in @xmath382 : @xmath383 the identifiability of this example can not be handled with domanov and de lathauwer s criterion , because it only applies for waring decompositions with @xmath384 . the sufficient condition presented in this paper , on the other hand , is applicable up to @xmath182 terms . therefore , we can run the algorithm presented above . this example may be verified with the ` specific - identifiability.m2 ` script that is provided with the arxiv version of this paper . in the first step , the @xmath385 matrix representing the span of @xmath315 is constructed . its rank is @xmath386 , as expected . the algorithm proceeds with the construction of young flattening , checking they generate an irreducible component . then , a basis of the kernel of @xmath315 is computed , containing @xmath387 equations . note that the dimension of @xmath354 and the codimension of @xmath315 should always be equal in the approach for certifying identifiability that was proposed in this section . then , for each of the points , the @xmath388 hessian matrices are computed and stacked into a @xmath389 matrix @xmath127 . the rank of @xmath127 equals @xmath36 for each of the points . in addition , the kernel of @xmath390 consists of a single vector that must be a multiple of the coefficient vector of @xmath64 ; for example , the vector in the kernel of the stacked hessian @xmath127 corresponding to the last term in the waring decomposition of @xmath295 is a multiple of its coefficient vector @xmath391 $ ] . for each of the points , this is indeed the case . finally , the algorithm positively concludes that @xmath295 admits a unique waring decomposition , i.e. , @xmath295 is identifiable . c. ciliberto , _ geometric aspects of polynomial interpolation in more variables and of waring s problem _ , progress in mathematics , vol . european congress of mathematics , barcelona , july 1014 , 2000 , volume i , birkhuser basel , 2001 . i. domanov and l. de lathauwer , _ on the uniqueness of the canonical polyadic decomposition of third - order tensors part ii : uniqueness of the overall decomposition _ , siam j. matrix anal . appl . * 34 * ( 2013 ) , 876903 .

we report the discovery of a strong over - density of galaxies in the field of a radio galaxy at ( 4c65.22 ) based on our broad - band and narrow - band ( h ) photometry with subaru telescope . we find that h emitters are located in the outskirts of the density peak ( cluster core ) dominated by passive red - sequence galaxies . this resembles the situation in lower - redshift clusters , suggesting that the newly discovered structure is a well - evolved rich galaxy cluster at . our data suggest that the color density and stellar mass density relations are already in place at , mostly driven by the passive red massive galaxies residing within kpc from the cluster core . these environmental trends almost disappear when we consider only star - forming ( sf ) galaxies . we do not find sfr density or ssfr density relations amongst sf galaxies , and the location of the sf main sequence does not significantly change with environment . nevertheless , we find a tentative hint that star - bursting galaxies ( up - scattered objects from the main sequence ) are preferentially located in a small group at-mpc away from the main body of the cluster . we also argue that the scatter of the sf main sequence could be dependent on the distance to the nearest neighboring galaxy .

1405.4165

recent observations have established that star - forming ( sf ) galaxies show a tight correlation between stellar mass ( @xmath7 ) and star - formation rate ( sfr ) . this sfr@xmath7 correlation ( so - called sf main sequence ) is investigated in the local universe ( e.g. @xcite ; @xcite ) , as well as in the distant universe out to @xmath8 ( e.g. @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . the sfr at a fixed mass evolves with time , reflecting the cosmic star formation history . the presence of this tight sfr@xmath7 correlation suggests that stellar mass has always been an important parameter that regulates sf activity in galaxies across cosmic time . the scatter around the sf main sequence is reported to be very small ( @xmath50.3 dex level at any redshifts ) , but the deviation around the sf main sequence may be an important parameter as it is likely to reflect the variation of gas accretion / exaustion history of galaxies ( e.g. @xcite ; @xcite ; @xcite ) . some recent studies attempt to identify the origin of the scatter and the key parameter that makes the strongest impact on galaxy evolution . for example , @xcite studied how the galaxy structure and the mode of sf activity depend on the position on the sfr@xmath7 diagram . they find that the upper envelope of the sf main sequence tends to be dominated by dusty galaxies ( characterized by high sfr@xmath9/sfr@xmath10 ratio ) with high sersic index ( _ n _ ) , suggesting a rapid build - up of mass in the nuclear regions of these systems , due to e.g. galaxy - galaxy interactions / mergers . [ cols="^,^,^,^,^,^,^ " , ] we use photometric redshifts ( @xmath11 ) to select non - h@xmath1-emitting passive members ( with strong 4000 break ) which can not be identified with the above h@xmath1 emitter selection . we here restrict the sample to those detected at @xmath122@xmath13 level with at least four bands out of the six broad - band images ( @xmath14 ) . the best - fit photometric redshifts are computed with a simple template fitting following the standard @xmath15 minimizing statistics using eazy code ( @xcite ) . the distribution of the derived @xmath11 is shown in fig . the photometric redshifts derived with only six broad - band filters may not be very accurate , but it is worth mentioning that there is a clear redshift spike at @[email protected] near the radio galaxy ( top panel of fig.[fig : zphot_histogram ] ) . in the following analysis , we apply a photo-@xmath16 cut of @xmath17 ( dotted lines in fig.[fig : zphot_histogram ] ) to select member galaxies associated to the radio galaxy . with this criteria , we select 61 galaxies in total , among which 20 galaxies are within the central 250 kpc region members and h@xmath1 emitters , and exploit this point as the cluster center ( see also fig . [ fig : map ] ) . ] . as shown in fig . [ fig : color_select ] ( black circles ) , most of the photo-@xmath16 selected galaxies satisfy the _ bzk _ and _ rjk _ criteria . unfortunately , we can not fully quantify the photo-@xmath16 accuracy because no spec-@xmath16 information is available in the observed field . however , we emphasize that our results do not change even if we apply simple color selections without relying on the photometric redshift . we here crudely test the contamination level by using the ukidss / uds data , where @xmath18 data are publicly available ( @xcite ) . after applying the same magnitude cuts , we derive photo-@xmath16 of the uds galaxies in the same way as described above . we find that @xmath62250 galaxies ( out of @xmath625000 sources ) satisfy the @xmath17 criteria over the entire @xmath60.8 deg@xmath19 uds field , yielding the expected contamination rate of @xmath60.8 arcmin@xmath20 . based on this estimate , we expect @xmath60.6 galaxies in the central 250 kpc region . therefore , the central 250 kpc region of 4c65.22 field is indeed over - dense by a factor of @xmath630@xmath21 compared to the control field ( see also 4 ) . we also expect @xmath622 galaxies within 4@xmath227@xmath23 moircs fov . the number density of the photo-@xmath16 members in the 4c65.22 field is still significantly higher than general field ( by a factor of @xmath63@xmath21 ) , but this analysis suggests that the over - density is diluted when averaged over the fov . for star - forming galaxies , the photometric redshifts may be much less reliable because of their flat featureless broad - band seds . we find that only 20 h@xmath1 emitters ( 45% ) satisfy the photo-@xmath16 criteria of 1.3@xmath24@[email protected] ( see blue - hatched histograms in fig.[fig : zphot_histogram ] ) . in particular , we find that the photometric redshift does not work well for h@xmath1 emitters with red optical colors , and that the photo-@xmath16 errors of such red emitters tend to be slightly larger than those of blue emitters . this probably reflects the well - known age / dust degeneracy , making it extremely difficult to construct a complete sample of sf galaxies using a limited set of broad - band filters . we remind the readers that this point is an important advantage of the narrow - band survey : it allows us to construct a clean star - forming galaxy sample within a narrow redshift slice at a high completeness level ( e.g. @xcite ; @xcite ) . in the case of our current dataset , @xmath655% of sf galaxies might have been missed out with the photo-@xmath16 selection alone , but those galaxies are now _ rescued _ with our nb - based approach . in summary , we selected the final member galaxies with the following two criteria . first , we select 61 galaxies with @xmath25 . second , we include all 44 h@xmath1 emitters ( selected in 3.1 ) regardless of their photometric redshifts . note that 20 out of the 44 h@xmath1 emitters satisfy the above photo-@xmath16 criteria as well . in these ways , our final member catalog contains 85 galaxies in total , and this sample will be used in the remaining part of this paper . we note that a fraction of the selected h@xmath1 emitters might be agns . in general , the fraction of agns in h@xmath1-selected galaxies at @xmath26 is expected to be @xmath61015% ( @xcite ; @xcite ) . however , unfortunatately , it is not possible to quantify the agn contribution in our own sample , without multi - wavelength imaging / spectroscopic follow - up observations . in particular , the environmental dependence of the agn activity in the distant universe is still an open question and beyond the scope of this paper . -band data . the size of this image is 100@xmath27100@xmath28 , same as the right - hand panel of fig . north is up , and east is to the left . + [ fig : image ] ] based on the sample selection described in 3 , we here report the discovery of a strong over - density of @xmath3 galaxies around the radio galaxy 4c65.22 . in fig . [ fig : map ] , we show the spatial distribution of the selected member galaxies . we here plot photo-@xmath16 selected galaxies ( with 1.3@xmath24@[email protected] ; black circles ) and the h@xmath1 emitters ( blue squares ) . the member galaxies on the red sequence ( @xmath29 ; see 4.2 ) are marked with red symbols . it is evident that there is a clear density peak of @xmath3 galaxies near the radio galaxy . the radio galaxy ( shown with yellow star in fig . [ fig : map ] ) is located at @xmath30140 kpc away from the density peak , not in the very center of the core . we already showed in fig . [ fig : zphot_histogram ] a clear redshift spike at @xmath31 within @xmath4@xmath24250 kpc from the density peak . this result does not change even if we consider `` all '' member galaxies including h@xmath1 emitters . the number density of all the member galaxies ( i.e. photo-@xmath16 member and h@xmath1 emitters ) within the 250 kpc circle is 28.9 arcmin@xmath32 , which is a factor of @xmath1210@xmath21 higher than the average in the outer field ( 2.3 arcmin@xmath20 ) . we note that a small @xmath11 peak at @xmath3 can also be seen in the bottom panel of fig . [ fig : zphot_histogram ] . this probably reflects the presence of large - scale structures extending beyond the 250 kpc circle . although we need spectroscopic follow - up observation for confirming such relatively poor structures , it is highly possible that some of the small groups as well as the filamentary structures traced by h@xmath1 emitters seen in fig . [ fig : map ] are really associated to the central cluster . another important result drawn from fig . [ fig : map ] is that the cluster central region ( @xmath4@xmath5250 kpc ) is clearly dominated by red - sequence galaxies without h@xmath1 emission ( see also fig . [ fig : image ] for a three - color representation of our @xmath33 images in the central cluster region ) . interestingly , this `` suppression radius '' of @xmath6200 kpc is consistent with that reported in another x - ray selected rich cluster at a similar redshift ( e.g. @xcite ) , suggesting this newly discovered system is a well - evolved rich galaxy cluster at @xmath2 . the presence of a number of h@xmath1 emitters ( at @xmath34 ) surrounding the core further supports our discovery of the rich cluster associated to the radio galaxy . we note that no extended x - ray emission is reported in this field so far ; e.g. by the rosat all - sky cluster survey ( @xcite ) . this is not surprising given the shallowness of the all - sky survey data . however , it would be highly possible that we can detect an extended x - ray emission associated with this cluster if we can perform a deep x - ray follow - up observation in the future . in fig . [ fig : colmag ] , we show color magnitude diagrams for galaxies located at @xmath4@xmath24250 kpc and those outside this annulus . an interesting feature recognized in this plot is the clear `` deficit '' of blue galaxies in the central cluster region ( @xmath4@xmath24250 kpc ) , in stark contrast to the outer field . there still exist a few h@xmath1 emitters within the @xmath35250 kpc circle , but these h@xmath1 emitting galaxies in the cluster core region show redder colors than typical h@xmath1 emitters in the field environment . this clear lack of low - mass blue galaxies may represent an accelerated galaxy evolution in high-@xmath16 cluster environments ; it is most likely explained that blue sf galaxies falling into the cluster are quickly transformed into red / massive quiescent population . it is also likely that the quenching process accompanies a quick mass growth via e.g. mergers or galaxy galaxy interactions , otherwise we can not fully explain the deficit of low - mass blue galaxies _ and _ the excess of red massive galaxies in the cluster core region at the same time . in fig . [ fig : colmag ] , the red sequence is only visible in its bright end , consistent with some recent studies of @xmath36 clusters ( e.g. @xcite ; @xcite ) . the lack of the faint end of the red sequence is unlikely due to the selection effect : we verified that the number of faint red galaxies does not increase even if we use @xmath37-band detected catalog instead of nb - selected catalog . we also show in fig . [ fig : fraction ] the cumulative ( and differential ) fraction of each population in our total sample as a function of the distance from the density peak out to @xmath351 mpc . this plot further supports our claim that red galaxies are strongly clustered within @xmath38250 kpc . it is evident that the fraction of red galaxies sharply drops at @xmath39kpc , which coincides with the place where we see the dramatical increase in h@xmath1 emitters and blue galaxies . the kolmogorov - smirnov test ( ks - test ) suggests with @xmath1299.9% confidence level that the distributions of red galaxies and h@xmath1 emitters ( or blue galaxies ) are drawn from a different parent population . we therefore conclude that most of blue star - forming galaxies falling into the highest - density cluster core are forced to migrate toward the red sequence , resulting in the clear deficit of low - mass blue galaxies in the central part of this cluster . using the newly discovered cluster field as a laboratory , we here take a more detailed look at the environmental dependence of galaxy properties at @xmath3 . for each member galaxy , we calculate the local surface number density ( @xmath41 ) , where @xmath42 denotes the distance to the fifth - nearest neighbor of each member galaxy . in the following analysis of this paper , we use this local density @xmath43 as an environmental index , but our results do not change even if we use @xmath44 or @xmath45 . we first focus on galaxy colors . in fig.[fig : color_density ] ( left ) , we show @xmath46 ( rest - frame @xmath47 ) colors of member galaxies as a function of @xmath43 . this plot clearly demonstrates that the color density relation is already in place at @xmath3 , consistent with the visual impression from fig . [ fig : map ] . we here compute the spearman s rank correlation coefficient ( @xmath48 ) . with a sample size of @xmath49@xmath5084 , we can conclude that there is a significant correlation between the two variables : a null hypothesis that there is no correlation between @xmath46 and @xmath43 is ruled out at @xmath1299% confidence level . the correlation becomes much less significant when we consider only h@xmath1 emitters . in this case , the spearman s rank correlation coefficient is @xmath51 . with a sample size of @xmath52 , we can not rule out a null hypothesis that there is no significant correlation between @xmath46 and @xmath43 for h@xmath1 emitters . therefore we suggest that the color density relation is primarily driven by passive red galaxies in the highest - density cluster core , equivalently by the lack of blue galaxies in the highest - density region . we note , however , that there is a weak correlation between the ew(h@xmath1+[nii ] ) and galaxy colors in the sense that redder h@xmath1 emitters tend to have lower ew ( see blue squares in fig . 8 , where we use different symbol size for high - ew / low - ew emitters ) . in other words , our definition of the h@xmath1 emitters ( @[email protected] ) may potentially fail to select low - ew ( red ) h@xmath1 emitters . therefore we need to keep in mind that the results on the color density relation amongst h@xmath1 emitters could be highly sensitive to the ew cut used for the sample selection . it is possible that the color density relation is partly produced by the correlation between galaxy stellar mass and environment . we here investigate the environmental dependence of stellar masses ( @xmath7 ) of the member galaxies . the @xmath7 of galaxies are derived using @xmath54-band photometry with @xmath55 correction based on the @xmath56 color as the following . @xmath57 where @xmath58 indicates the @xmath59 correction value . these conversion equations are derived using the model galaxies developed by @xcite assuming @xcite imf . it is demonstrated that this `` one - color method '' can provide reasonable @xmath7 values out to @xmath60 ( see @xcite ) . in fig . [ fig : color_density ] ( right ) , we plot the derived @xmath7 as a function of @xmath43 . our data suggest that there exists a weak @xmath7density relation in the observed field , again driven by the massive galaxies in the highest - density cluster core . by computing the spearman s rank correlation coefficient ( @xmath61 ) , and with a sample size of @xmath6284 , we can rule out the null hypothesis that there is no @xmath7@xmath43 relationship with @xmath3097% confidence level . however , if we focus only on the h@xmath1 emitters , the spearman s rank correlation coefficient is @xmath51 , suggesting no significant correlation between the two variables with the sample size of @xmath52 . our data thus suggest that the color density and @xmath7density relation are already in place at @xmath3 , mostly driven by passive red / massive galaxies in the highest - density region . however , the correlations almost disappear when we consider only sf galaxies . an important advantage of the nb h@xmath1 survey is that we can measure h@xmath1 flux of all emitters within the observed fov . we here describe how we derive the h@xmath1-based sfrs of h@xmath1 emitters , and then describe how the derived sfrs depend on environment . first of all , we calculate the h@xmath1 + line flux ( @xmath63}}$ ] ) , continuum flux density ( @xmath64 ) and rest - frame equivalent width ( ew@xmath65 ) of each h@xmath1 emitter based on its broad - band and narrow - band photometry with the following equations : @xmath66}}=\delta_{\rm{nb}}\frac { f_{\rm{nb } } - f_{h } } { 1-\delta_{\rm{nb}}/\delta_{h}}\ ] ] @xmath67 @xmath68 } } ) = ( 1+z)^{-1 } \frac{f_{\rm{h\alpha } + [ nii]}}{f_c}.\ ] ] the @xmath69 ( @xmath70@xmath71 m ) and @xmath72 ( @xmath73@xmath71 m ) are the fwhms of the broad - band ( @xmath74 ) and nb1657 filters , @xmath75 and @xmath76 are the flux densities at @xmath74-band and at nb1657 , respectively . we then multiply 4@xmath77@xmath78 by @xmath79}}$ ] to derive the luminosity @xmath80}}$ ] , where @xmath81 is the luminosity distance of 1.109@xmath2110@xmath82 mpc at @xmath2 . we correct for the line contribution into the total nb fluxes , using the empirical calibration between /(h@xmath1 + ) flux ratio and ew@xmath65(h@xmath1 + ) for local sdss galaxies presented by @xcite . the median of the derived /(h@xmath1 + ) ratio for our h@xmath1 emitter sample is 0.19 , consistent with the h@xmath1 emitter sample at @xmath83 ( /(h@xmath1+)@xmath500.22 ) shown by @xcite . although there remains a large scatter around this relation ( @xmath60.4 dex ; e.g. @xcite ) , it is impossible to measure [ nii ] line flux for all galaxies because it requires deep spectroscopic observation for all of them . therefore our procedure is the best effort at this moment , and at least it would be more realistic than the convensional constant ( e.g. 30% ) correction . we then compute the dust - free h@xmath1-based star formation rates , sfr@xmath84 , using the @xcite relation for @xcite imf ; sfr@xmath84[@xmath85/yr]@xmath86[ergs@xmath32 ] . we note that our h@xmath1 emitter selection criteria shown in 3.1 approximately correspond to ew@xmath88(h@xmath1+)@xmath8920 and sfr@xmath84@xmath893@xmath85/yr . finally , we correct for the dust extinction effect . we here consider two independent methods to estimate dust extinction at h@xmath1 ( @xmath90 ) . the first method is to use @xmath90@xmath7 relation established for local galaxies by @xcite . the @xmath90@xmath7 correlation is likely to be unchanged up to @xmath3 ( @xcite ; @xcite ) , and this method is often used in studies of distant galaxies as a convenient method to predict dust extinction effect for individual galaxies . the second approach is to use sfr@xmath84/sfr@xmath10 ratio for predicting @xmath90 . because h@xmath1 is less sensitive to dust extinction effect than rest - frame uv light , the sfr@xmath84/sfr@xmath10 ratio can be used as an indicator of dust attenuation ( @xcite ; @xcite ) . in our case , the @xmath91-band flux density ( @xmath92@xmath301800 ) can be converted to the rest - frame uv luminosity density , and then we derive sfr@xmath10 using @xcite calibration ; sfr@[email protected]@xmath2110@xmath93[ergs@xmath32hz@xmath32 ] . the observed sfr@xmath84/sfr@xmath10 ratio is translated to the dust extinction value using the following equation : @xmath94 @xmath95 we here adopt @[email protected] and @[email protected] from @xcite law with @[email protected] . one thing we need to treat carefully is an extra extinction toward the nebular regions ( @xmath99 value in the above equations ) . in the @xcite prescription , the @xmath99 value was shown to be 0.44 ( i.e. @xmath100 ) . however , recent studies claim that this @xmath99 value is likely to be different for high-@xmath16 sf galaxies ( e.g. @xcite ) . @xcite empirically linked the sfr@xmath84/sfr@xmath10 ratio to @xmath101 for @xmath102 sf galaxies . they derived the best - fitted value of @xmath103 , and we adopt this value to our analysis . we note that the method may not be perfect enough ( e.g. sfr@xmath84/sfr@xmath10 flux ratio could depend on galactic age ; see @xcite for a detailed discussion ) , but we expect this method is more realistic than the simple @xmath7-dependent correction shown above . in fig . [ fig : extinction ] ( left ) , we plot the derived @xmath90 values against their stellar mass for all h@xmath1 emitters . the extinction effect ( derived by the sfr@xmath84/sfr@xmath10 ratio ) increases with stellar mass , showing a reasonable agreement with the local @xmath90@xmath7 relation . also , the derived @xmath90 values are higher for red h@xmath1 emitters ( see middle panel of fig . [ fig : extinction ] ) , suggesting the redder h@xmath1 emitters tend to be dustier . it may be interesting to note that there is a large scatter around the @xmath90@xmath7 relation , and that there exist some exceptionally dusty galaxies with @xmath90@xmath893 mag , which can not be identified with the simple @xmath7-dependent correction emitters showing @xmath90@xmath104 ( i.e. sfr@xmath84/sfr@xmath10@xmath105 ) . one of these sources ( id#930 ) is consistent with @xmath90@xmath106 within its error bar ( probably a dust - free galaxy ) , whilst the other source ( id#13 ) shows an unrealistic negative @xmath90 value . this is because the @xmath91-band flux of id#13 is contaminated by its close companion galaxy ( due to the poorer psf size of our @xmath91-band data ) . although we do not remove this source from our analyses , it does not affect our conclusion at all . ] . also , there may be a weak correlation between @xmath90 and environment for the same h@xmath1 emitters sample ( see fig . [ fig : extinction ] right ) . although the statistical significance is low ( only @xmath61@xmath13 level ) and there is a large uncertainty in deriving @xmath90 of individual galaxies , this result may suggest that sf galaxies in high - density environments tend to be more highly obscured by dust ( by @xmath60.5 mag level ) than field counterparts . this is qualitatively consistent with our recent study ( @xcite ) , where we showed a higher dust extinction of cluster galaxies with a 24@xmath71 m stacking analysis for @xmath107 galaxies . finally , we show in fig . [ fig : sfr_density ] the sfrs and specific sfrs ( ssfr@xmath50sfr/@xmath7 ) of all h@xmath1 emitters as a function of local galaxy density ( @xmath108 measured with all member galaxies ) . the left- and right - hand panels show the results based on the two independent dust extinction correction described above . we do not find any significant sfr density or ssfr density correlation in any of the four panels in fig . [ fig : sfr_density ] , suggesting that sf activity of sf galaxies does not significantly change with environment . by computing the spearman s rank correlation coefficients with a sample size of @xmath52 , we can not reject a null hypothesis that there is no significant relationship between the two variables in any of these four panels . we caution that our results are based on the purely h@xmath1-selected galaxies ( i.e. for sf galaxy population ) . the clear excess of non - h@xmath1-emitting red - sequence galaxies in the very high - density region ( as shown in figs . [ fig : map][fig : colmag ] ) implies an anti - correlation between ( s)sfr and environment , although we can not measure sfrs of galaxies without h@xmath1 emission . the important goal of this study is to test the environmental dependence of the sf main sequence at @xmath3 using the h@xmath1-selected galaxies in the newly discovered cluster field . in fig . [ fig : ms ] , we show the sfr@xmath7 diagrams for h@xmath1 emitters in the 4c65.22 field . the left- and right - hand panels show the result from different dust extinction correction : derived from @xmath7 ( left panel ) and from sfr@xmath84/sfr@xmath10 ratio ( right panel ) . in this plot , the h@xmath1 emitters in high - density environment ( @xmath109 ; top @xmath61/3 ) are shown with the red circles , while the other h@xmath1 emitters are shown with the blue circles . regardless of the dust extinction correction procedure , we find that the location of the sf main sequence is independent of environment at @xmath3 . in the bottom panels of fig . [ fig : ms ] , we also show the distribution of sfr offsets from the main sequence ( @xmath110ms@xmath111 ) for galaxies in high - density ( red filled histogram ) and low - density environment ( blue hatched histogram ) . again , we do not see any strong difference between the two environment samples : the ks test suggests that it is unlikely that the two distributions are drawn from the same parent population in either case of the dust extinction procedure , further supporting our conclusion that the location of the sf main sequence does not significantly change with environment . one potentially interesting finding from fig . [ fig : ms ] is that the h@xmath1 emitters in the higher - density region ( red symbols ) tend to populate the higher envelope of the sf main sequence in the massive end ( with @xmath112 ) . we attempt to quantify this trend by calculating the fraction of galaxies with @xmath113ms@xmath114 ( i.e. galaxies with boosted activity ) . the fraction turns out to be 56@xmath11531% and 33@xmath11519% for high- and low - density sample , respectively . the error bars are clearly large due to the limited sample size , preventing us from drawing any firm conclusion . however , at least to say , those massive galaxies with boosted activity in high-@xmath16 cluster environments should be an interesting population under the influence of environmental effects . finally , we investigate the origin of the scatter of the @xmath110ms@xmath111 value within our observed field . in the top panel of fig . [ fig : map_ms ] , we plot the spatial distribution of the h@xmath1 emitters by dividing the sample into three categories based on the @xmath110ms@xmath111 value : `` bursty '' galaxies ( @xmath116ms@xmath117 ) , `` normal '' galaxies ( @xmath118ms@xmath119 ) , and `` semi - passive '' galaxies ( @xmath116ms@xmath119 ) . the sample is of course small , but it may be interesting to note that the north - west group located @xmath61 mpc away from the main body of the cluster tend to have a large number of `` bursty '' galaxies , implying that star - burst activities are most frequently triggered in poor group environments . in contrast , we can see a mix of bursty / normal / semi - passive population in the periphery of the cluster core . the local density ( @xmath43 ) of galaxies in the cluster periphery is similar to those in the north - west group , making it hard to identify the site of starburst activity with the local density approach alone ( see bottom - left panel of fig . [ fig : map_ms ] ) . to further investigate the environmental origin of the scatter of the sf main sequence , we show in the bottom - right panel of fig . [ fig : map_ms ] the @xmath110ms values as a function of the distance to the nearest neighboring member galaxy ( @xmath120 ) . in this plot , we find a tentative hint that galaxies having a close companion ( with @xmath5200 kpc ) tend to show a larger scatter around the main sequence , whilst isolated galaxies tend to be located on the main sequence . the sample is small , but this plot may suggest a potential link between the galaxy galaxy encounter and the deviation around the sf main sequence i.e . a galaxy tends to show a boosted / truncated sf activity once they become a satellite of another galaxy ( c.f . @xcite ; @xcite ) . however , because of the mixture of bursty / normal / passive population , we can not detect an environmental variation in the location ( or zero point ) of the sf main sequence . it has been a long observational challenge to find a galaxy cluster at @xmath122 until recently . however , an increasing number of studies have identified clusters at such frontier redshifts with various techniques ( e.g. @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) , and the nature of such high-@xmath16 clusters and their member galaxies are now being investigated . some studies point out that sf activity in high-@xmath16 cluster cores are comparable to ( or even higher than ) general field galaxies ( e.g. @xcite ; @xcite ; @xcite ; @xcite ) ; hence galaxy clusters at these redshifts are interpreted as being in the `` transitional phase '' or final quenching period ( @xcite ; @xcite ; @xcite ; @xcite ) . interestingly , our newly discovered cluster reported in this paper resembles more like @xmath123 clusters , characterized by a strong concentration of passive red galaxies in the cluster center and star - forming population surrounding the core ( e.g. @xcite ; 2011 ) . our finding of the suppressed sf activity in the cluster core is also similar to the situation in the well - studied rich cluster xmmuj2235.3@xmath1242557 at @xmath125 ( @xcite , @xcite , @xcite ; @xcite ) . recently , @xcite studied an x - ray detected @xmath126 group in chandra deep field south . based on the sed and morphological analyses using the candels data , these authors suggest that most galaxies in this @xmath127 group are quiescent early - type systems ( with a few of them specroscopically confirmed ) . the size of the spatial extension of quiescent galaxies in the system reported by @xcite is also qualitatively consistent with that of our newly discovered system at @xmath2 ( @xmath6200300 kpc scale ) . in our current work , by utilizing the narrow - band h@xmath1 imaging technique , we find a clear `` deficit '' of blue galaxies in the central cluster region ( @xmath4@xmath24250 kpc ) . we do find a few h@xmath1 emitters within the @xmath35250 kpc circle , but those h@xmath1 emitting galaxies in the cluster core show redder colors than typical blue h@xmath1 emitters seen in general field environment ( 4.2 ) . we anticipate that the clear lack of low - mass blue galaxies in the cluster central region is due to the rapid sf quenching and rapid mass growth triggered by environmental effects . if a low - mass blue galaxy enters a rich cluster environment , its sf activity will soon be quenched . this quenching mechanism should accompany an intense short - lived starburst or merging event , because the red - sequence galaxies dominating the cluster core tend to be much more massive than typical blue sf galaxies in the surrounding field . the color density relation has been investigated out to @xmath36 or even beyond ( e.g. @xcite ; @xcite ; @xcite ) . our data suggest that the color density relation is already in place around the radio galaxy at @xmath3 . however , the correlation becomes much weaker once we remove non - h@xmath1-emitting galaxies , suggesting that the color density relation is a product of passive red - sequence galaxies dominating the highest - density cluster core . a similar result was drawn from morphological analysis of cluster / field galaxies at @xmath127 by @xcite . these authors showed that the morphologies of sf galaxies in cluster and field environments at @xmath127 show no significant difference . we note , however , that our data covers only a limited fov ( [email protected] mpc@xmath19 ) , and it is possible that we do not sample galaxies in very low - density environments . we also note that the definition of the h@xmath1 emitters in this study is ew@xmath65(h@xmath1@xmath129[nii])@xmath8920 , which might be too large if we consider local sf galaxies ( e.g. @xcite ) . at @xmath3 , however , it is expected that the ew@xmath65(h@xmath1@xmath129[nii])@xmath8920 criterion allows us to select a fairly complete sample of normal main - sequence galaxies . therefore we do not expect that the ew cut does not make any strong impacts on our results , but we need to keep in mind that our survey can potentially miss low - ew ( but still star - forming ) galaxies . this is an inevitable problem associated with any nb - based study , and we need future deep imaging / spectroscopic observations to construct a more complete sample of galaxies down to such less active population . it is worth mentioning that recent studies on @xmath130 proto - clusters suggest that sf galaxies in higher - density environments tend to have redder colors , implying a color density correlation _ amongst star - forming galaxies _ ( see @xcite ; @xcite ) . @xcite showed that most of h@xmath1 emitters with red @xmath131 colors in the core of the pks1138262 proto - cluster at @xmath132 are very massive system ( with @xmath133 ) . such an accelerated galaxy stellar mass growth is also suggested by some recent studies on @xmath8 proto - clusters ( @xcite ; @xcite ; @xcite ; @xcite ) . in our current analysis , we find a number of red massive galaxies _ without star formation _ in the core of the @xmath3 cluster . the stellar masses of these passive population are comparable to those of massive sf galaxies in the @xmath130 proto - cluster enviromnents . if we assume that those massive sf galaxies reported in @xmath130 proto - cluster environments are the direct progenitors of passive red galaxies seen in the @xmath3 cluster core , they are expected to have stopped their sf activity within a relatively short time interval between @xmath60 and @xmath3 . there have been a lot of debates on the `` reversal '' of the sfr density relation in the distant universe . @xcite first noted this possibility with the multi - wavelength dataset in the goods fields ( see also @xcite ; @xcite ) . this reversal of the sf density relation is still under debate , and it is likely that the results could be uncertain depending on the sample definitions and/or the definitions of environment ( @xcite ; @xcite ; @xcite ; @xcite ) . at least for star - forming galaxies , recent studies have obtained a similar conclusion that there is no significant sfr density or ssfr density correlation amongst sf galaxies ( @xcite ; @xcite ; 2012 ; @xcite ; @xcite ) . the results drawn from our current work is consistent with those recent studies . we caution that the `` flat '' sfr density relation may not hold at @xmath8 . as we demonstrated in @xcite , sf galaxies in proto - cluster environments tend to be more massive than those in underdense regions . although most of the h@xmath1 emitters in proto - cluster environments tend to be located on the sf main sequence , they are skewed to the massive end of the main sequence ; hence their sfrs tend to be higher ( on average ) than those in general field environments . a vital approach to understanding the environmental effects on sf galaxies is to examine the environmental impacts on the sf main sequence . recent studies have investigated the sf main sequence mainly for distant _ field _ galaxies . those studies have confirmed the existence of the sf main sequence out to @xmath134 or above , but the environmental dependence of sf main sequence is much poorly understood due to the lack of well - defined sf galaxy samples in clusters / groups in the distant unverse . in @xcite , we studied the environmental dependence of the sf main sequence using our purely h@xmath1-selected galaxy samples at @xmath135 in distant cluster environments and general field environments . we found that there is no significant environmental variation in the sf main sequence out to @xmath60 ( with a difference of @xmath60.2 dex at maximum ) . in that work , we used @xmath7-dependent extinction correction for all h@xmath1 emitters , which is reported to be unchanged ( at least on average ) throughout the cosmic time ( @xcite ) . however , we need to test the validity of this procedure by comparing the results derived with different methods of dust extinction correction . in our current analysis , we test this concern by estimating the h@xmath1 dust extinction in a more realistic way using the sfr@xmath84/sfr@xmath10 ratio . however , we again find that the location of the sf main sequence does not change with environment ( with a difference of @xmath50.1 dex level ) , confirming the conclusion drawn by @xcite . we caution , however , that the dust extinction issue is still an important key question to fully assess the real enviromental impacts on the sf main sequence . indeed , we do find a marginal trend that the dust extinction effect ( @xmath90 ) is weakly correlated with local density in the sense that sf galaxies in higher - density environment tend to be more highly obscured by dust . a similar implication was obtained for @xmath107 galaxies by @xcite . in that work , we used sfr@xmath9/sfr@xmath84 ratio ( with 24@xmath71 m stacking analysis ) to show that sf galaxies in high - density environments tend to be dustier ( by @xmath60.5 mag level ) than those in underdense environments , in qualitative agreement with the current work . our studies thus suggest a possibility that sf galaxies surviving in high-@xmath16 cluster environments tend to be dustier , and perhaps the nature / mode of sf activity in cluster environments may be different from those in general field environments . we recall that the environmental impacts on the dust properties of galaxies are still very pooly understood at this moment , and there are some contradicting results on this issue ( @xcite ; @xcite ) . therefore , a special care must be taken on the dust extinction correction when we interprete the environmental dependence of the sf main sequence , and this key issue needs to be studied more in detail with larger galaxy samples in the future . finally , we comment on the possibility of the environmental impacts on the `` scatter '' of the sf main sequence . as we showed in fig . [ fig : map_ms ] , we do not find a correlation between @xmath110ms and @xmath43 , consistent with the environmental independence of the location of the main sequence . on the other hand , we find a tentative hint that the @xmath110ms value more strongly deviates for those having a close companion within @xmath5200 kpc , while more isolated galaxies tend to be more steadily located on the sf main sequence . the sample is clearly too small to draw any firm conclusion , but this result may provide us with a new insight on the origin of the scatter of the sf main sequence . if a galaxy approaches another galaxy , their sf activity would be disturbed ( some are boosted and others are truncated ) due to the galaxy - galaxy interaction or halo gas stripping ( satellite quenching ) . this would explain the larger scatter of the @xmath110ms for those with small @xmath120 , and the apparent environmental independence of the location of the sf main sequence . we performed a broad - band and narrow - band ( h@xmath1 ) imaging survey of a radio galaxy field ( 4c65.22 ) at @xmath0 over the 7@xmath224@xmath23 fov ( corresponding to [email protected] mpc@xmath19 ) with moircs / suprime - cam on the subaru telescope . based on this new dataset , we find a rich cluster candidate around the radio galaxy . we also studied an environmental dependence of galaxy properties around this newly discovered structure at @xmath3 . our findings are summarized as the following . \(1 ) with the photo-@xmath16 and h@xmath1 emitter selections , we discovered a strong over - density of galaxies around 4c65.22 . we find that the cluster central region ( @xmath136@xmath5250 kpc from the density peak ) is clearly dominated by passive red - sequence galaxies without h@xmath1 emission , while the h@xmath1-emitting galaxies are preferentially located in the cluster outskirts , showing a sharp decline in the h@xmath1 emitter fraction toward the cluster center . this spatial segregation is similar to that seen in lower redshift clusters , suggesting that the newly discovered structure is a well - matured system . \(2 ) the color density and @xmath7density relations are already in place at @xmath3 . these environmental trends are mostly driven by passive red / massive galaxies residing in the cluster central region , whilst such environmental trends become much weaker when we consider only h@xmath1 emitters . this is also the case for sfr density and ssfr density relation . an excess of non - h@xmath1-emitting red - sequence galaxies in the cluster core strongly suggests that the sf activity is suppressed in the very rich environment at @xmath2 , but the sf activity amongst sf galaxies are almost independent of environment . \(3 ) there may exist a weak correlation between dust attenuation ( derived with sfr@xmath84/sfr@xmath10 ratio ) and local environment for sf galaxies at @xmath2 : galaxies in high - density environments tend to be dustier by @xmath60.51.0 mag level . however , even if we take this point into account , we can not see a detectable environmental variation in the location of the sf main sequence , consistent with recent studies . we do not find any correlation between the sfr offset from the sf main sequence ( @xmath110ms ) and @xmath43 , but we find a tentative hint that galaxies having a close companion ( at a moderate distance of @xmath5200 kpc ) tend to be more largely scattered from the sf main sequence than more isolated galaxies . the broad - band and narrow - band imaging data used in this paper are collected at the subaru telescope , which is operated by the national astronomical observatory of japan ( naoj ) . we thank the referee for reviewing our paper and providing us with useful comments which improved the paper . y.k . , k.t . , and m.h . acknowledge support from the japan society for the promotion of science ( jsps ) through jsps research fellowships for young scientists . this work was financially supported in part by a grant - in - aid for the scientific research ( nos.21340045 ; 24244015 ) by the japanese ministry of education , culture , sports and science . + _ facilities : _ . alberts , s. , pope , a. , brodwin , m. , et al . 2014 , , 437 , 437 balogh , m. , eke , v. , miller , c. , et al . 2004 , , 348 , 1355 bassett , r. , papovich , c. , lotz , j. m. , et al . 2013 , , 770 , 58 bauer , a. e. , conselice , c. j. , prez - gonzlez , p. g. , et al . 2011 , , 417 , 289 bauer , a. e. , grtzbauch , r. , jrgensen , i. , varela , j. , & bergmann , m. 2011 , , 411 , 2009 bertin , e. , & arnouts , s. 1996 , , 117 , 393 best , p. n. , lehnert , m. d. , miley , g. k. , rttgering , h. j. a. 2003 , , 343 , 1 bhringer , h. , voges , w. , huchra , j. p. , et al . 2000 , , 129 , 435 brammer , g. b. , van dokkum , p. g. , & coppi , p. 2008 , , 686 , 1503 brinchmann , j. , charlot , s. , white , s. d. m. , et al . 2004 , , 351 , 1151 brinkmann , w. , chester , m. , kollgaard , r. , et al . 1999 , , 134 , 221 brodwin , m. , stanford , s. a. , gonzalez , a. h. , et al . 2013 , , 779 , 138 buat , v. , boselli , a. , gavazzi , g. , & bonfanti , c. 2002 , , 383 , 801 bunker , a. j. , warren , s. j. , hewett , p. c. , & clements , d. l. 1995 , , 273 , 513 calzetti , d. , armus , l. , bohlin , r. c. , kinney , a. l. , koornneef , j. , & storchi - bergmann , t. 2000 , , 533 , 682 cardelli , j. a. , clayton , g. c. , & mathis , j. s. 1989 , , 345 , 245 cooke , e. a. , hatch , n. a. , muldrew , s. i. , rigby , e. e. , & kurk , j. d. 2014 , , 440 , 3262 cooper , m. c. , coil , a. l. , gerke , b. f. , et al . 2010 , , 409 , 337 cooper , m. c. , newman , j. a. , weiner , b. j. , et al . 2008 , , 383 , 1058 cucciati , o. , iovino , a. , marinoni , c. , et al . 2006 , , 458 , 39 daddi , e. , dickinson , m. , morrison , g. , et al . 2007 , , 670 , 156 dressler , a. 1980 , , 236 , 351 elbaz , d. , dickinson , m. , hwang , h. s. , et al . 2011 , , 533 , a119 elbaz , d. , daddi , e. , le borgne , d. , et al . 2007 , , 468 , 33 fassbender , r. , nastasi , a. , bhringer , h. , et al . 2011 , , 527 , l10 galametz , a. , stern , d. , pentericci , l. , et al . 2013 , , 559 , a2 galametz , a. , stern , d. , stanford , s. a. , et al . 2010 , , 516 , a101 garn , t. , et al . 2010 , , 402 , 2017 garn , t. , & best , p. n. 2010 , , 409 , 421 gobat , r. , daddi , e. , onodera , m. , et al . 2011 , , 526 , a133 gmez , p. l. , et al . 2003 , , 584 , 210 goto , t. , et al . 2003 , , 55 , 757 greene , c. r. , gilbank , d. g. , balogh , m. l. , et al . 2012 , , 425 , 1738 grtzbauch , r. , bauer , a. e. , jrgensen , i. , & varela , j. 2012 , , 423 , 3652 grtzbauch , r. , conselice , c. j. , bauer , a. e. , et al . 2011 , , 418 , 938 gunn , j. e. , & stryker , l. l. 1983 , , 52 , 121 hayashi , m. , kodama , t. , koyama , y. , et al . 2014 , , 439 , 2571 hayashi , m. , kodama , t. , tadaki , k .- , koyama , y. , & tanaka , i. 2012 , , 757 , 15 hayashi , m. , kodama , t. , koyama , y. , tadaki , k .- i . , & tanaka , i. 2011 , , 415 , 2670 hayashi , m. , kodama , t. , koyama , y. , tanaka , i. , shimasaku , k. , & okamura , s. 2010 , , 402 , 1980 hatch , n. a. , kurk , j. d. , pentericci , l. , et al . 2011 , , 415 , 2993 hilton , m. , lloyd - davies , e. , stanford , s. a. , et al . 2010 , , 718 , 133 hwang , h. s. , elbaz , d. , dickinson , m. , et al . 2011 , , 535 , a60 ibar , e. , sobral , d. , best , p. n. , et al . 2013 , , 434 , 3218 ichikawa , t. , et al . 2006 , , 6269 , 38 ideue , y. , nagao , t. , taniguchi , y. , et al . 2009 , , 700 , 971 iye , m. , et al . 2004 , , 56 , 381 kajisawa , m. , ichikawa , t. , yamada , t. , et al . 2010 , , 723 , 129 kajisawa , m. , kodama , t. , tanaka , i. , yamada , t. , & bower , r. 2006 , , 371 , 577 kashino , d. , silverman , j. d. , rodighiero , g. , et al . 2013 , , 777 , l8 kennicutt , r. c. , jr . 1998 , , 36 , 189 kennicutt , r. c. , jr . , & kent , s. m. 1983 , , 88 , 1094 kinney , a. l. , calzetti , d. , bohlin , r. c. , et al . 1996 , , 467 , 38 kodama , t. , hayashi , m. , koyama , y. , et al . 2013 , iau symposium , 295 , 74 kodama , t. , tanaka , i. , kajisawa , m. , et al . 2007 , , 377 , 1717 kodama , t. , balogh , m. l. , smail , i. , bower , r. g. , & nakata , f. 2004 , , 354 , 1103 kodama , t. , smail , i. , nakata , f. , okamura , s. , & bower , r. g. 2001 , , 562 , l9 kodama , t. , arimoto , n. , barger , a. j. , & aragon - salamanca , a. 1998 , , 334 , 99 kollgaard , r. i. , feigelson , e. d. , laurent - muehleisen , s. a. , et al . 1995 , , 449 , 61 kollgaard , r. i. , brinkmann , w. , chester , m. m. , et al . 1994 , , 93 , 145 koyama , y. , smail , i. , kurk , j. , et al . 2013 , , 434 , 423 koyama , y. , kodama , t. , tadaki , k .- i . , et al . 2013 , , 428 , 1551 koyama , y. , kodama , t. , nakata , f. , shimasaku , k. , & okamura , s. 2011 , , 734 , 66 koyama , y. , kodama , t. , shimasaku , k. , hayashi , m. , okamura , s. , tanaka , i. , & tokoku , c. 2010 , , 403 , 1611 koyama , y. , kodama , t. , tanaka , m. , shimasaku , k. , & okamura , s. 2007 , , 382 , 1719 kurk , j. , cimatti , a. , zamorani , g. , et al . 2009 , , 504 , 331 kurk , j. d. , pentericci , l. , rttgering , h. j. a. , & miley , g. k. 2004 , , 428 , 793 kurk , j. d. , rttgering , h. j. a. , pentericci , l. , et al . 2000 , , 358 , l1 lawrence , a. , warren , s. j. , almaini , o. , et al . 2007 , , 379 , 1599 lin , l. , jian , h .- y . , foucaud , s. , et al . 2014 , , 782 , 33 matsuda , y. , smail , i. , geach , j. e. , et al . 2011 , , 416 , 2041 muzzin , a. , wilson , g. , demarco , r. , et al . 2013 , , 767 , 39 muzzin , a. , wilson , g. , yee , h. k. c. , et al . 2012 , , 746 , 188 lacy , m. , hill , g. j. , kaiser , m .- e . , & rawlings , s. 1993 , , 263 , 707 lewis , i. , et al . 2002 , , 334 , 673 li , i. h. , glazebrook , k. , gilbank , d. , et al . 2011 , , 411 , 1869 lidman , c. , rosati , p. , tanaka , m. , et al . 2008 , , 489 , 981 madau , p. , ftac13erguson , h. c. , dickinson , m. e. , et al . 1996 , , 283 , 1388 mcgee , s. l. , balogh , m. l. , wilman , d. j. , et al . 2011 , , 413 , 996 miyazaki , s. , et al . 2002 , , 54 , 833 nastasi , a. , fassbender , r. , bhringer , h. , et al . 2011 , , 532 , l6 newman , a. b. , ellis , r. s. , andreon , s. , et al . 2013 , arxiv:1310.6754 noeske , k. g. , weiner , b. j. , faber , s. m. , et al . 2007 , , 660 , l43 ouchi , m. , et al . 2004 , , 611 , 660 patel , s. g. , kelson , d. d. , holden , b. p. , franx , m. , & illingworth , g. d. 2011 , , 735 , 53 papovich , c. , momcheva , i. , willmer , c. n. a. , et al . 2010 , , 716 , 1503 park , c. , & hwang , h. s. 2009 , , 699 , 1595 peng , y .- j . , lilly , s. j. , renzini , a. , & carollo , m. 2012 , , 757 , 4 peng , y .- j . , lilly , s. j. , kova , k. , et al . 2010 , , 721 , 193 price , s. h. , kriek , m. , brammer , g. b. , et al . 2013 , arxiv:1310.4177 rodighiero , g. , cimatti , a. , gruppioni , c. , et al . 2010 , , 518 , l25 saintonge , a. , tacconi , l. j. , fabello , s. , et al . 2012 , , 758 , 73 salpeter , e. e. 1955 , , 121 , 161 santini , p. , fontana , a. , grazian , a. , et al . 2009 , , 504 , 751 santos , j. s. , altieri , b. , popesso , p. , et al . 2013 , , 433 , 1287 santos , j. s. , fassbender , r. , nastasi , a. , et al . 2011 , , 531 , l15 schlegel , d. j. , finkbeiner , d. p. , & davis , m. 1998 , , 500 , 525 scoville , n. , arnouts , s. , aussel , h. , et al . 2013 , , 206 , 3 shimakawa , r. , et al . , 2014 , mnras , submitted smail , i. , geach , j. e. , swinbank , a. m. , et al . 2014 , , 782 , 19 sobral , d. , smail , i. , best , p. n. , et al . 2013 , , 428 , 1128 sobral , d. , best , p. n. , matsuda , y. , et al . 2012 , , 420 , 1926 sobral , d. , best , p. n. , smail , i. , geach , j. e. , cirasuolo , m. , garn , t. , & dalton , g. b. 2011 , , 411 , 675 stanford , s. a. , brodwin , m. , gonzalez , a. h. , et al . 2012 , , 753 , 164 steidel , c. c. , adelberger , k. l. , shapley , a. e. , et al . 2005 , , 626 , 44 stott , j. p. , smail , i. , edge , a. c. , et al . 2007 , , 661 , 95 suzuki , r. , et al . 2008 , , 60 , 1347 tacconi , l. j. , neri , r. , genzel , r. , et al . 2013 , , 768 , 74 tadaki , k. i. , kodama , t. , tanaka , i. , et al . 2013 , , 778 , 114 tadaki , k. i. , kodama , t. , ota , k. , et al . 2012 , , 423 , 2617 tadaki , k. i. , kodama , t. , koyama , y. , et al . 2011 , , 63 , 437 tanaka , i. , breuck , c. d. , kurk , j. d. , et al . 2011 , , 63 , 415 tanaka , m. , finoguenov , a. , mirkazemi , m. , et al . 2013 , , 65 , 17 tanaka , m. , finoguenov , a. , & ueda , y. 2010 , , 716 , l152 tanaka , m. , goto , t. , okamura , s. , shimasaku , k. , & brinkmann , j. 2004 , , 128 , 2677 tran , k .- v . h. , papovich , c. , saintonge , a. , et al . 2010 , , 719 , l126 venemans , b. p. , rttgering , h. j. a. , miley , g. k. , et al . 2007 , , 461 , 823 venemans , b. p. , kurk , j. d. , miley , g. k. , et al . 2002 , , 569 , l11 villar , v. , gallego , j. , prez - gonzlez , p. g. , pascual , s. , noeske , k. , koo , d. c. , barro , g. , & zamorano , j. 2008 , , 677 , 169 vulcani , b. , poggianti , b. m. , finn , r. a. , et al . 2010 , , 710 , l1 whitaker , k. e. , van dokkum , p. g. , brammer , g. , & franx , m. 2012 , , 754 , l29 wijesinghe , d. b. , hopkins , a. m. , brough , s. , et al . 2012 , , 423 , 3679 wuyts , s. , frster schreiber , n. m. , nelson , e. j. , et al . 2013 , , 779 , 135 wuyts , s. , frster schreiber , n. m. , van der wel , a. , et al . 2011 , , 742 , 96 yagi , m. , kashikawa , n. , sekiguchi , m. , doi , m. , yasuda , n. , shimasaku , k. , & okamura , s. 2002 , , 123 , 66 zeimann , g. r. , stanford , s. a. , brodwin , m. , et al . 2013 , , 779 , 137 ziparo , f. , popesso , p. , finoguenov , a. , et al . 2013 , , 2575

the detrended cross - correlation coefficient has recently been proposed to quantify the strength of cross - correlations on different temporal scales in bivariate , non - stationary time series . it is based on the detrended cross - correlation and detrended fluctuation analyses ( dcca and dfa , respectively ) and can be viewed as an analogue of the pearson coefficient in the case of the fluctuation analysis . the coefficient works well in many practical situations but by construction its applicability is limited to detection of whether two signals are generally cross - correlated , without possibility to obtain information on the amplitude of fluctuations that are responsible for those cross - correlations . in order to introduce some related flexibility , here we propose an extension of that exploits the multifractal versions of dfa and dcca : mfdfa and mfcca , respectively . the resulting new coefficient not only is able to quantify the strength of correlations , but also it allows one to identify the range of detrended fluctuation amplitudes that are correlated in two signals under study . we show how the coefficient works in practical situations by applying it to stochastic time series representing processes with long memory : autoregressive and multiplicative ones . such processes are often used to model signals recorded from complex systems and complex physical phenomena like turbulence , so we are convinced that this new measure can successfully be applied in time series analysis . in particular , we present an example of such application to highly complex empirical data from financial markets . the present formulation can straightforwardly be extended to multivariate data in terms of the-dependent counterpart of the correlation matrices and then to the network representation .

1506.08692

standard correlation measures like the pearson correlation coefficient and the cross - correlation function require stationary data in order to provide reliable results , which is a requirement that is hard to fulfill in many real - world situations ( the financial and physiological data are the negative examples here @xcite ) . ( by stationarity we mean stability of the probability distribution functions of the data over time ; from this perspective nonstationarity can be produced both by the long - range autocorrelations and by the pdf s heavy tails that make any signal length effectively insufficient . ) this problem can to some degree be resolved by replacing original signals with the corresponding detrended fluctuations , i.e. by considering the power - law correlations between the polynomially detrended walks , which are much more stationary than the orginal data . this approach was introduced in a context of autocorrelations in @xcite as the detrended fluctuation analysis ( dfa ) and immediately gained popularity among the researchers working with empirical data across many disciplines . the modification of dfa oriented towards detection of the power - law cross - correlations is known as the detrended cross - correlation analysis ( dcca ) @xcite and it can be used in analysis of bivariate and multivariate empirical data @xcite . among the reasons behind the popularity of dfa was its ability of detecting fractal character of signals , which was subsequently extended to the multifractal case ( the mfdfa method @xcite ) , which also proved very useful if applied to empirical data , especially owing to its superior reliability if compared to other methods @xcite . dcca was also generalized in order to be applicable to signals with multifractal cross - correlations and the resulting mfdcca / mfdxa algorithm @xcite also attracted some attention @xcite . however , this generalization raises controversy as at some stage it requires to neglect the detrended covariance sign to avoid obtaining complex values , which leads to inevitable loss of information about analyzed signals and , consequently , to incorrect results @xcite . this drawback has recently been removed with a new sign - sensitive method of the multifractal detrended cross - correlation analysis with an acronym mfcca @xcite that is more consistent than mfdcca as a multifractal generalization of dcca ( for more details see section ii ) . although the dfa and dcca methods were designed to deal with nonstationary signals , they are related to each other exactly in the same way as the variance and the covariance analyses are related to each other in the case of stationary data . therefore , by exploiting both these methods , an analogue of the pearson coefficient was introduced . it is known as the detrended cross - correlation coefficient @xmath0 @xcite and serves as a tool for quantifying strength of correlations in the fluctuations of detrended signals at a given time scale @xcite . the main advantage of @xmath0 over the pearson coefficient is its ability to quantify correlations in nonstationary signals @xcite . what is important is that the signals under study may either be fractal or non - fractal since @xmath0 is defined for a single scale . it is also worth mentioning that there is a counterpart of @xmath0 constructed for the detrended moving average cross - correlation analysis ( dmca ) @xcite but considering it here exceeds the objectives of this work . by definition , the detrended cross - correlation coefficient is insensitive to higher - order statistics of fluctuations , i.e. , other than the simple covariance . this means that values of @xmath0 can not indicate whether the detected cross - correlations between two signals originate from the fluctuations of all amplitudes equally likely or rather some specific range of amplitudes gives a dominant contribution , while the remaining fluctuations can be much less correlated or even completely uncorrelated . one might easily imagine situations , in which such insensitivity can be considered a serious disadvantage of the method . for instance , let each nonstationary signal in a pair be a mixture of two processes , one of which is correlated in both signals and has a relatively small amplitude , while the other is unique to each signal and has a relatively high amplitude . the coefficient @xmath0 would probably indicate that the signals are somehow cross - correlated but it would not bring any information allowing one to identify the amplitude of the cross - correlated components . in order to avoid such insensitivity , here we propose a generalization of the detrended cross - correlation coefficient in such a way that it becomes sensitive to correlations in the fluctuations of a selected amplitude range . the easiest way to do this is by introducing a @xmath2-dependent detrended cross - correlation coefficient @xmath1 ( @xmath3 ) based on the so - called @xmath2-dependent fluctuation functions @xmath4 from mfdfa and mfcca @xcite . the idea is based on the fact that from a number of different values contibuting to a sum , one can select specific values ( e.g. , the large , the medium , or the small ones ) by rising all the values in the sum to some power ( high positive , small positive , or negative one , respectively ) . like @xmath0 , the coefficient @xmath1 is not related to fractal properties of signals so it can be used to quantify cross - correlations between any signals . as by its definition @xmath1 is intended to be a tool for analyzing nonstationary signals , we expect that it can find broad applications in the studies of empirical data from natural complex systems : physical , biological , social , financial , etc . in the remaining part of the paper , we present the formal definition of @xmath1 ( section ii ) , show examples of how it works if applied to the computer - generated signals representing different stochastic processes ( section iii ) and to empirical data from the financial markets ( section iv ) , and finally present the main conclusions ( section v ) . a fundamental quantity for the detrended fluctuation analysis as well as for any type of its derivative methods is the variance ( covariance ) @xmath5 ( @xmath6 ) of the detrended signals @xmath7 ( @xmath8 stands for either @xmath9 or @xmath10 ) . let us consider a pair of time series @xmath11 and @xmath12 divided into @xmath13 separate boxes of length @xmath14 ( i.e. , @xmath15 boxes starting from the opposite ends ) . a detrending procedure consists of calculating in each box @xmath16 ( @xmath17 ) the residual signals @xmath7 equal to the difference between the integrated signals and the @xmath18th - order polynomials @xmath19 fitted to these signals : @xmath20 in the present work we use @xmath21 . the covariance and the variances of @xmath9 and @xmath10 in a box @xmath16 are defined as : @xmath22 where @xmath8 again means either @xmath9 or @xmath10 . these quantities can be used to define a family of the so - called fluctuation functions of order @xmath2 @xcite : @xmath23 |f_{xy}^2(s,\nu)|^{q/2},\\ f_{zz}^q(s ) = { 1 \over 2 m_s } \sum_{\nu=0}^{2 m_s - 1 } \left[f_{zz}^2(s,\nu)\right]^{q/2}. \label{eq::variance.q}\end{aligned}\ ] ] the above definition of @xmath24 guarantees that : ( i ) no imaginary part occurs in @xmath24 ( only the absolute values are raised to a real power @xmath25 ) and ( ii ) by preserving signs of the covariances @xmath26 , no information is lost while taking the absolute values . for @xmath27 the above definitions are reduced to a simpler form : @xmath28 that may be interpreted as an average covariance and average variances for the boxes of size @xmath14 . in the standard application of the fluctuation functions in mfdfa and mfcca , one observes the dependence of @xmath29 and @xmath30 on the scale @xmath14 and looks for a convincing scaling behavior : @xmath31^{1/q } \sim s^{\gamma(q)}$ ] and/or @xmath32^{1/q } \sim s^{\delta(q)}$ ] , which indicates a fractal structure of the signals ( monofractal for the constant functions : @xmath33 and @xmath34 , multifractal otherwise ) . the structure of eqs . ( [ eq::covariance ] ) and ( [ eq::variance ] ) , which resembles the ordinary covariance and variance , respectively , suggested one to introduce the detrended cross - correlation coefficient @xcite : @xmath35 owing to its normalized range of values @xcite : @xmath36 with @xmath37 in the case of uncorrelated signals , @xmath38 in the case of a perfect cross - correlation , and @xmath39 in the case of a perfect anticorrelation , the coefficient @xmath40 can be used to quantify the strength of cross - correlations between the detrended signals @xmath7 on different scales @xmath14 and to compare this strength among different signal pairs @xcite . the structure of eqs . ( [ eq::covariance.average ] ) and ( [ eq::variance.average ] ) indicates that all the boxes contribute to the fluctuation functions and the correlation coefficient with the same weight , irrespective of how large ( or how small ) is @xmath26 and @xmath41 in a particular box @xmath16 . this means that , by using solely the coefficient @xmath0 , it is impossible to observe how the boxes that are characterized by the fluctuations of a specific amplitude range contribute to the overall cross - correlations . the following definition of a new @xmath2-dependent detrended cross - correlation ( @xmath2dcca ) coefficient @xmath42 allows one to overcome this constraint : @xmath43 as the real exponent @xmath2 plays the role of a respective filter here . for @xmath27 we restore the definition ( [ eq::rho.dcca ] ) of @xmath0 , for @xmath44 the boxes with high values of @xmath26 and @xmath41 contribute to @xmath42 the most , while for @xmath45 the boxes with relatively small values do it the most significantly . the more deviated from the value @xmath27 the exponent @xmath2 is , the more extreme fluctuations in the corresponding boxes contribute to the coefficient @xmath42 . note that no form of the @xmath14-dependence in eqs . ( [ eq::covariance.q ] ) and ( [ eq::variance.q ] ) has been assumed , so the signals under study do not need to be fractal at all . for @xmath46 , values of the @xmath1 coefficient are bound within the same range as @xmath0 , i.e. , @xmath47 in order to show this , we fix the scale @xmath14 and prove the following cauchy - schwarz - like inequality : @xmath48 ^ 2 \le f_{xx}^q(s ) f_{yy}^q(s ) , \quad q \ge 0 . \label{eq::cauchy - schwarz - like}\ ] ] first , observe that from the relation : @xmath49 ( @xmath50 , @xmath51 ) it follows for any two boxes @xmath52 that : @xmath53^{\alpha } \le \\ \nonumber \le \left [ f_{xx}^2(s,\nu ) f_{yy}^2(s,\mu ) \right]^{2 \alpha } + \left [ f_{xx}^2(s,\mu ) f_{yy}^2(s,\nu ) \right]^{2 \alpha } \\ \quad \label{eq::lemma}\end{aligned}\ ] ] since @xmath54 . now , to simplify the notation , we temporarily neglect the signum function in eq . ( [ eq::covariance.q ] ) and assume that all the covariances @xmath26 are positive . we then start from the l.h.s . of eq . ( [ eq::cauchy - schwarz - like ] ) : @xmath55 ^ 2 = { 1 \over 4 m_s^2 } \lbrace \sum_{\nu=0}^{2 m_s-1 } \left [ f_{xy}^2(s,\nu ) \right]^{q/2 } \rbrace^2 = \\ \nonumber = { 1 \over 4 m_s^2 } \sum_{\nu=0}^{2 m_s-1 } \left [ f_{xy}^2(s,\nu ) \right]^q + \\ \nonumber + { 1 \over 4 m_s^2 } \sum_{\nu=0}^{2 m_s-1 } \sum_{\mu=\nu+1}^{2 m_s-1 } 2 \left [ f_{xy}^2(s,\nu ) f_{xy}^2(s,\mu ) \right]^{q/2 } \le \\ \nonumber \le { 1 \over 4 m_s^2 } \sum_{\nu=0}^{2 m_s-1 } \left [ f_{xx}^2(s,\nu ) f_{yy}^2(s,\nu ) \right]^{q/2 } + \\ \nonumber + { 1 \over 4 m_s^2 } \sum_{\nu=0}^{2 m_s-1 } \sum_{\mu=\nu+1}^{2 m_s-1 } \lbrace \left [ f_{xx}^2(s,\nu ) f_{yy}^2(s,\mu ) \right]^{q/2 } + \\ \nonumber + \left [ f_{xx}^2(s,\mu ) f_{yy}^2(s,\nu ) \right]^{q/2 } \rbrace = \\ \nonumber = { 1 \over 4 m_s^2 } \sum_{\nu=0}^{2 m_s-1 } \sum_{\mu=0}^{2 m_s-1 } \left [ f_{xx}^2(s,\nu ) f_{yy}^2(s,\mu ) \right]^{q/2 } = \\ \nonumber = f_{xx}^q(s ) f_{yy}^q(s ) , \end{aligned}\ ] ] where we exploited the cauchy - schwarz inequality : @xmath56 ^ 2 \le f_{xx}^2(s,\nu ) f_{yy}^2(s,\nu)$ ] , the implication : @xmath57 for @xmath46 , and eq . ( [ eq::lemma ] ) . in general , the covariances @xmath26 can be negative and we must take this fact into consideration . fortunately , the following relation always holds : @xmath58 |f_{xy}^2(s,\nu)|^{q/2 } \le \\ \nonumber \le { 1 \over 2 m_s } \sum_{\nu=0}^{2 m_s - 1 } |f_{xy}^2(s,\nu)|^{q/2 } \le \sqrt { f_{xx}^q(s ) f_{yy}^q(s ) } , \end{aligned}\ ] ] which ends the proof of the inequalities ( [ eq::cauchy - schwarz - like ] ) and ( [ eq::normalized ] ) . for @xmath59 the situation looks different , because the implication @xmath57 is false in this case . this means that the denominator in eq . ( [ eq::rho.q ] ) may be arbitrarily small as compared to the numerator modulus and @xmath42 can then assume either large positive or large negative values : @xmath60 for some scales @xmath14 . this can be evident for uncorrelated or partially correlated signals , while it is unlike for perfectly correlated ones . therefore , the interpretation of the values of @xmath42 for @xmath59 is a delicate issue that will be discussed in more detail in section iii . in this section we present examples of the application of the new coefficient @xmath1 to the computer - generated time series representing different stochastic processes , in which the cross - correlations can fully be controlled , and then we discuss the performance of the coefficient in those cases . since @xmath1 is oriented towards analysis of nonstationary data , we prefer to employ stochastic models that produce signals with long memory and/or heavy - tailed probability distribution functions . long memory , which produces trends , and heavy tails of the fluctuations p.d.f.s , which can produce apparent trends , are the principal sources of nonstationarity in data from natural complex systems , therefore their presence in the model signals is highly desired . we focus on two models : the autoregressive fractionally integrated moving average ( arfima ) and the markov - switching multifractal ( msm ) . the former produces fractal signals with long memory and , despite its origin in financial economics @xcite , it is broadly used to model anomalous diffusion in various fields of science , like atmosphere physics and geophysics @xcite , astrophysics @xcite , biology and physiology @xcite , and many other . the latter can be viewed as a version of a random walk in random time @xcite . it is based on the stochastic multiplicative cascades and creates signals with long memory , heavy - tailed p.d.f . , and with arbitrary length ( unlike more typical cascading processes whose length is limited to the consecutive powers of the number of branches ) @xcite . msm can be exploited to model various cascade - like phenomena , e.g. , turbulence @xcite and financial volatility @xcite and to predict future evolution of the corresponding observables . first , we study behavior of @xmath42 for a pair of maximally correlated signals . in order to prepare such signals , we use the arfima process @xcite defined by the following formulas : @xmath61 where the parameters @xmath62 ( @xmath63 ) , fulfilling the condition : @xmath64 , characterize the temporal range of the linear autocorrelations in @xmath65 and are strictly related to the hurst exponents : @xmath66 . the quantity @xmath67 is called weight and defined by : @xmath68 the time series @xmath69 and @xmath70 are correlated due to a common noise term @xmath71 being an i.i.d . gaussian random variable . the arfima signals are ( mono)fractal , so their fluctuation functions @xmath32^{1/q}$ ] form the families of power - laws presented in the upper panels of fig . [ fig::arfima.correlated ] . the fractally cross - correlated nature of @xmath69 and @xmath70 can be seen in fig . [ fig::arfima.correlated](c ) as a family of parallel power - law functions @xmath31^{1/q}$ ] . careful inspection of the three panels : fig . [ fig::arfima.correlated](a)-(c ) suggests that the fluctuation amplitudes of the signals under study are maximally cross - correlated , so we may expect that a properly defined @xmath2dcca coefficient should reflect this fact with its value being close to unity . indeed , in fig . [ fig::arfima.correlated](d ) , @xmath72 irrespective of @xmath14 and @xmath2 . this invariance can also be a consequence of the stationarity of the detrended arfima processes . for signals with heavy tails of the pdfs ( which preserve their non - stationary character after detrending ) , we would expect more sizable deviations of @xmath42 from unity even if such signals were on average strongly cross - correlated . this is because large fluctuations that may occur in different signals at different moments can have strong impact on covariance of the signals in the related boxes ( up to the considerably large widths @xmath14 ) , which can suppress it in these particular boxes and thus influence @xmath24 . the higher @xmath73 is , the stronger effect of this type may happen . -dependent fluctuation functions @xmath74^{1/q}$ ] and @xmath75^{1/q}$ ] for two arfima time series sharing the same gaussian noise term . a family of the fluctuation functions for different values of @xmath2 , @xmath76 are shown for each time series ( the topmost one represents @xmath77 ) . the linear dependence of @xmath74^{1/q}$ ] and @xmath75^{1/q}$ ] on the double logarithmic plots indicates a fractal character of the analyzed signals . ( c ) the @xmath2-dependent cross - fluctuation function @xmath31^{1/q}$ ] for the same pair of the time series indicating fractal cross - correlations between them . ( d ) the @xmath2-dependent detrended cross - correlation coefficient @xmath42 for different values of @xmath2 . note the almost perfect cross - correlation for any @xmath14 and @xmath2 . ] in order to test the @xmath2dcca coefficient @xmath42 in the absence of cross - correlations , we take the same signals as before : @xmath69 and @xmath70 and destroy their temporal structure by random shuffling . however , before we show the complete results , we have to discuss the problem of @xmath78 for @xmath59 . [ fig::arfima.unbound](a)(b ) shows @xmath79 for ( a ) @xmath80 and ( b ) @xmath81 . as one can see , its values can span a number of orders of magnitude with the more negative @xmath2 is , the larger this range can be . the unbound character of @xmath42 makes any inference on the strength of the cross - correlations virtually impossible . as it has already been mentioned in section ii , the high values of @xmath79 are produced if the denominator in eq . ( [ eq::rho.q ] ) is much smaller than the numerator . this situation can predominantly happen if the signals under study are not cross - correlated or are cross - correlated weakly . this suggests that a value of @xmath79 that deviates much from 1 in any direction may be viewed as an indicator of a lack of cross - correlations . taking this into consideration , one can redefine the @xmath2dcca coefficient in the following way : @xmath82^{-1 } & \textrm{if } \quad |\rho_q(s)| > 1 . \end{cases } \label{eq::rho.star}\ ] ] now the @xmath83 coefficient remains always within the interval @xmath84 even if @xmath59 . two typical cases can be distinguished in this situation : ( i ) fig . [ fig::arfima.unbound](c ) exhibits that for @xmath80 the coefficient @xmath83 indicates the expected almost - zero level of cross - correlations ; ( ii ) for @xmath81 ( fig . [ fig::arfima.unbound](d ) ) @xmath83 is clearly non - zero , but it is highly unstable with respect to @xmath14 , bouncing between positive and negative values . both these cases , if found in an analysis , may thus be considered the indicators of uncorrelated signals . for simplicity , from now on we will omit the star in the notation of the @xmath2dcca coefficient : @xmath85 -dependent detrended cross - correlation coefficient @xmath1 as a function of the temporal scale @xmath14 for sample values of @xmath59 : @xmath80 ( left ) and @xmath81 ( right ) . ( top ) scale dependence of the originally defined @xmath42 for which values outside the standard interval @xmath84 are allowed ( the modulus @xmath79 is taken in order to show both the positive ( black circles ) and the negative ( red / grey squares ) values on the logarithmic axis ) . dashed lines indicate the @xmath86 level . ( bottom ) the modified coefficient @xmath83 defined by eq . ( [ eq::rho.star ] ) with its values being in the interval @xmath84 . those values of @xmath42 that were inverted to obtain @xmath83 are denoted by circles . ] complete results of @xmath42 for the randomized arfima signals are presented in fig . [ fig::arfima.shuffled ] . in the light of the above discussion , we are justified to conclude that the analyzed signals are not cross - correlated at any scale @xmath14 and at any @xmath87 . since for @xmath88 and @xmath89 we observe some deflections of @xmath42 from zero , we have to perform a test that can resolve whether these deflections are statistically significant . in order to do this , we calculate @xmath42 for @xmath90 pairs of the randomized arfima signals and estimate the dispersion of @xmath42 . the related standard deviations @xmath91 are shown in fig . [ fig::arfima.deviation ] . a direct comparison between fig . [ fig::arfima.shuffled ] and fig . [ fig::arfima.deviation ] leads us to a conclusion that the deflections from zero observed for @xmath88 in fig . [ fig::arfima.shuffled ] can not be considered statistically significant . although we show here the results for the arfima processes only , the qualitatively similar results can be obtained for time series representing other examples of the uncorrelated statistical processes , which supports our claim that @xmath42 correctly estimates the level of cross - correlations between uncorrelated signals . calculated for a pair of the randomized arfima time series . in each panel , @xmath42 for different @xmath2 is displayed . statistical significance of the results can be assessed by comparing these results with fig . [ fig::arfima.deviation ] . the inverted values of @xmath42 for @xmath59 are denoted by circles . ] ( black solid lines ) and standard deviation @xmath91 ( blue / grey lines with error bars ) of the coefficient @xmath42 calculated for 10,000 pairs of the shuffled arfima time series . in each panel , the result for different @xmath2 is displayed . ] there is a good moment here to stress that a necessary condition for obtaining the correct results ( i.e. , the lack of cross - correlations ) shown in fig . [ fig::arfima.shuffled ] is a proper definition of the @xmath2-dependent fluctuation function @xmath24 . such a definition requires preserving the signs of the covariances @xmath26 exactly as it was done in eq . ( [ eq::covariance.q ] ) . unfortunately , it often happens in literature that those signs are neglected and only the moduli of the covariances are considered ( e.g. , @xcite ) . what consequences can this approach have is illustrated in fig . [ fig::arfima.shuffled.nosign ] . we take the same data as in fig . [ fig::arfima.shuffled ] and calculate a modified coefficient @xmath92 that uses the no - sign definition of @xmath24 . as one can see , for @xmath93 this coefficient falsely indicates the presence of statistically significant positive cross - correlations across all the scales , even though we know that such correlations can not exist between the independently randomized signals . ( using the no - sign definition of @xmath24 in eq . ( [ eq::covariance.q ] ) ) calculated for the same pair of the randomized arfima time series that was already used in fig . [ fig::arfima.shuffled ] . observe that for @xmath93 the coefficient gives a false indication of positive cross - correlations . apparent statistical significance of these results can be seen by comparing them with fig . [ fig::arfima.deviation ] . ] the above examples of the almost perfectly correlated and the almost perfectly uncorrelated signals form the two extremes that are the least interesting trivial cases between which one can find a spectrum of much more interesting cases where the signals may reveal partial cross - correlations . for example , the correlations that are restricted to the signals specific components or that are transient in time . we shall discuss here both these situations . first , we test sensitivity of the @xmath2dcca approach if the cross - correlations are transient . in order to do this , we take two cross - correlated arfima time series ( with @xmath72 as in section iiia ) and randomize a part of the data points in one of these time series . after this operation , the modified time series consists of a fraction @xmath94 of the original data points in its center and a fraction @xmath95 of the shuffled points . sample results of these calculations are shown in fig . [ fig::sensitivity ] for different choices of @xmath94 . it is clear that in this particular case the method gives statistically significant indication of the cross - correlated character of fluctuations for as low as @xmath96 correlated data points in both signals . this indication is restricted to the small and medium temporal scales ( @xmath97 ) only , however . in order to obtain similar significance for the larger scales up to @xmath98 , one has to increase the correlated fraction to @xmath99 , so that @xmath100 , where @xmath101 is the length of the time series . coefficient calculated based on the arfima time series ( @xmath102 ) . an original time series 1 is compared with a modified time series 2 , in which only a fraction @xmath94 of the data points is preserved , while the rest is randomized ( black lines ) . the results for the null - hypothesis of the uncorrelated time series are also shown for reference as the @xmath103 standard deviation of 100 independent realizations of the shuffled surrogates ( magenta / grey lines symmetric with respect to @xmath104 ) . ] second , we assume that each time series under study consists of different components that have different amplitudes . only some of these components are cross - correlated between the time series , while the other ones remain uncorrelated . for instance , a pair of time series may be cross - correlated only via large or via medium fluctuations and uncorrelated in small ones . we therefore expect that the @xmath2dcca analysis will allow us : ( i ) to identify the existence of the cross - correlations and ( ii ) to show , what fluctuation amplitudes carry these cross - correlations . the arfima processes are not ideal for being a subject of the present analysis since their fluctuations are essentially gaussian distributed and the range of their amplitudes is rather small . the method works fine also in this case , but here a much more instructive example can be the non - stationary processes with heavy - tailed fluctuations , where the amplitudes of the large and the small fluctuations differ considerably . the heavy - tailed fluctuations often appear in empirical data recorded from both natural and social systems , so our choice is realistic and it does not limit the applicability domain of the method . let us consider time series generated according to the markov - switching multifractal model developed in @xcite . this is a multiplicative , hierarchical model that is able to reproduce the multifractal characteristics of some types of empirical data ( for example , the financial volatility ) . from the point of view of the present analysis , it is important that it produces the ( unsigned ) signals with heavy - tailed probability distribution functions , whose fluctuations can exhibit long - range , multiscale auto - correlations . according to the msm model , an observable @xmath69 is given by : @xmath105 where @xmath106 is a gaussian random variable and @xmath107 plays a role of discrete time . @xmath108 is called the instantaneous volatility and defined as a product of a constant factor @xmath109 and @xmath110 multipliers @xmath111 drawn from the binary or the lognormal distribution : @xmath112 in the binomial case , @xmath113 $ ] , @xmath114 , while in the lognormal case : @xmath115 @xcite . each multiplier @xmath111 changes its value at time @xmath107 with probability : @xmath116 where @xmath117 and @xmath118 . here we choose the cascades with @xmath119 levels and @xmath120 branches . we also omit the gaussian - distributed factor @xmath106 in eq . ( [ eq::msm ] ) as being unnecessary from the point of view of our analysis , because the complexity of @xmath69 is related to the instantaneous volatility @xmath108 . as a consequence of the stochastic character of the multiplier values , two independently produced signals would not be cross - correlated . a pair of the multiscale cross - correlated time series can nevertheless be obtained from an msm time series by copying the set of its multipliers @xmath121 and adding a small amount of noise to each one : @xmath122 , where @xmath71 is an i.i.d . gaussian random variable and @xmath123 . dcca coefficient @xmath42 calculated for a pair of time series from the lognormal markov - switching multifractal ( msm ) model with @xmath119 , @xmath120 , @xmath124 , and @xmath125 ( see text for the description of these parameters ) . four selected values of @xmath2 are shown denoted by different lines . @xmath27 corresponds to the standard coefficient @xmath0 . ( a ) the original time series . ( b ) both time series are filtered ( see text ) in order to remove the correlations among the small fluctuations ( @xmath126 ) . ( c ) the randomized data ( single realization of the shuffled surrogates ) without any temporal correlations . ( d ) both time series are filtered in order to remove the correlations among the medium and large fluctuations ( @xmath127 ) . for clarity of the pictures , the inverted values of @xmath42 are not distinguished . ] our objective is to show that the @xmath2dcca coefficient @xmath42 can provide us with more information on the cross - correlation structure of these time series that the ordinary coefficient @xmath40 can . we generate two lognormal msm time series of length @xmath128 with @xmath124 . the parameter @xmath125 guarantees that their detrended fluctuations are cross - correlated . indeed , fig . [ fig::msm.good.examples](a ) displays that for @xmath76 the coefficients @xmath42 detect strong cross - correlations ( for @xmath102 even the maximum ones over all the scales ) . what then happens if we apply a filter that randomizes all the data points with small amplitude @xmath126 ? action of this filter is equivalent to removing the cross - correlations among the small fluctuations and preserving the ones among the medium and large fluctuations . [ fig::msm.good.examples](b ) presents the results obtained for such filtered signals . one can see that while the cross - correlations at the longer scales ( roughly , for @xmath129 ) are still detected by @xmath42 for any @xmath2 , for @xmath130 the @xmath2dcca coefficients indicate a significant decrease of the cross - correlations at the shorter scales ( @xmath131 ) . this decrease is more explicit for @xmath59 and for @xmath80 the corresponding coefficient indicates that the signals are completely uncorrelated . as such small values of @xmath2 are related to very small fluctuations , this result is exactly what one might expect for the filtered signals . on the other hand , for @xmath132 the filtering does not introduce any change in @xmath42 and this is also perfectly compliant with the _ a priori _ expectation that for the large fluctuations the cross - correlations should survive . now let us look at the results for @xmath27 in fig . [ fig::msm.good.examples](a)(b ) : a change is hardly seen . thus , in this case the @xmath0 coefficient proves totally insensitive to a significant change in the correlation structure of the data . another example of such an evident lack of sensitivity of @xmath0 can be seen in bottom panels of fig . [ fig::msm.good.examples ] . if the same msm time series as above are completely shuffled , all their temporal correlations are destroyed . [ fig::msm.good.examples](c ) shows that the coefficients @xmath42 do not detect any cross - correlations accordingly . the results for the large scales are within statistical error even if they are far from zero ; among them the coefficient for @xmath27 is almost flat for the majority of scales except the largest ones . these results can be compared with the results obtained for the filtered signals ( fig . [ fig::msm.good.examples](d ) ) , in which only the correlations among the small fluctuations ( @xmath133 ) are preserved and all the other correlations are removed via data shuffling . now the coefficients @xmath42 indicate the presence of cross - correlations for the medium scales ( for @xmath134 , @xmath135 , and @xmath80 ) and for the small scales ( for @xmath134 and @xmath135 ) . such values of @xmath2 correspond to the small fluctuations . in contrast , for @xmath27 the related coefficient does not show anything significant and behaves roughly similar to the no - correlation case of fig . [ fig::msm.good.examples](c ) . the two further examples illustrating usefulness of the @xmath2dcca coefficient are obtained from the binary msm model producing a pair of the @xmath110=20-level binary multiplicative cascades with the @xmath136-level multipliers @xmath111 drawn from the set @xmath137 $ ] and the set @xmath138 $ ] , respectively . the resulting time series are of length @xmath139 and their values belong to the interval ( 0.01,100 ) . in order to prepare the top panels of fig . [ fig::msm.subtle.examples](a)(b ) , we modify both time series in such a way that we leave the original values if they do not exceed a threshold of @xmath140 and randomize the values above this threshold . then we create copies of the so - modified time series and once again randomize their values if they are below @xmath141 . by doing this , we get the two pairs of signals : ( i ) a pair of signals without the largest fluctuations , ( ii ) a pair of signals without the largest and the smallest fluctuations . now we compare the coefficients @xmath42 calculated for these pairs . first , we look at @xmath142 ( black solid lines ) . in both top panels , we see that @xmath142 deviates significantly from unity for the scales @xmath131 , which suggests that the cross - correlations are not uniform across the fluctuation amplitudes . however , based solely on @xmath27 we can not state decisively , which fluctuations correspond to this decreased correlations . as regards a comparison between the panels ( a ) and ( b ) , there is some visible difference in @xmath142 for the middle scales ( @xmath143 ) , but again one can not decide what is the origin of this difference . dcca coefficient @xmath42 calculated for a pair of time series from the binary markov - switching multifractal ( msm ) model with @xmath144 and @xmath145 $ ] ( cascade no . 1 ) or @xmath146 $ ] ( cascade no . different values of @xmath2 are denoted by different lines . the case of @xmath27 corresponds to the standard dcca coefficient @xmath0 . ( top ) @xmath42 for ( a ) a pair of time series with randomized fluctuations of large amplitudes @xmath147 and for ( b ) a pair of time series with randomized fluctuations of large amplitudes @xmath147 and small amplitudes @xmath148 . ( bottom ) @xmath42 for ( c ) a pair of time series with randomized fluctuations of medium amplitudes @xmath149 and for ( d ) a pair of time series with randomized fluctuations of small and medium amplitudes @xmath150 . for clarity of the pictures , the inverted values of @xmath42 are not distinguished . ] let us now include @xmath151 . for @xmath77 the values of @xmath42 in both panels are even smaller than their counterparts for @xmath27 and this is especially strong for the smallest scales @xmath152 . the opposite relation is seen for @xmath134 , where @xmath153 is much larger than @xmath142 . this means that by increasing @xmath2 we obtain a decrease of @xmath42 , which can correctly be interpreted as a manifestation of uncorrelated behavior of the largest fluctuations in both pairs of the signals . if we then compare the coefficients for @xmath134 and @xmath135 , we realize that they are systematically larger in ( a ) than in ( b ) . even for @xmath81 the instability of @xmath154 suggest a lack of cross - correlations . all this means that the small fluctuations are weaker cross - correlated in ( b ) than in ( a ) , exactly as it is expected from the signal construction in both cases . next we once more use the time series generated by the binary msm model . we randomize the fluctuations in the middle range of the amplitudes : @xmath149 , which comprises the most frequent fluctuations in both time series , so that only the small and the large fluctuations remain cross - correlated . the results for this pair of signals are shown in fig . [ fig::msm.subtle.examples](c ) . a copy of each time series is created and now the small fluctuations are a subject of additional randomization . thus , we obtain the second pair of time series with only the large fluctuations that are still cross - correlated ( fig . [ fig::msm.subtle.examples](d ) ) . the difference between these pairs is the existence ( c ) and the lack ( d ) of the cross - correlations among the fluctuations with small amplitude . on the one hand , for @xmath27 and @xmath77 we do not observe any difference in @xmath42 between ( c ) and ( d ) as they are close to unity in both panels . on the other hand , the values of @xmath42 for @xmath135 and @xmath80 are substantially smaller for small @xmath14 in ( d ) than in ( c ) . taking these results altogether , one can correctly conclude that both signal pairs are cross - correlated in the large fluctuations and uncorrelated in the small ones . it has to be stressed that drawing this conclusion would be impossible if we restricted our analysis to the @xmath27 case of @xmath0 . ( the uncorrelated character of the fluctuations of the medium amplitudes can be revealed in other comparative analyses , but here in this example such a range of amplitudes is beyond our interest . ) financial data is known to be cross - correlated across different temporal scales @xcite , so it can serve as a suitable subject to show practical application of the @xmath2dcca coefficient . we choose data from two large financial markets : the american stock market represented by the stocks traded at new york stock exchange ( nyse ) or nasdaq and the foreign currency exchange market ( forex ) , which is a global market . in the former case , our data set comprises high - frequency recordings of the stock prices corresponding to the 100 largest american companies over the years 1998 - 99 . for each considered stock @xmath107 , the corresponding time series represents the logarithmic price increments ( returns ) @xmath155 sampled at a fixed time interval @xmath156 min . the length of such time series is @xmath157 data points . the sample results for two pairs of stocks : ( a ) microsoft@xmath158intel and ( b ) microsoft@xmath1583 m are displayed in fig . [ fig::stocks ] and tested against the null hypothesis that the cross - correlations are random ( standard deviation of 100 independent realizations of the shuffled surrogates ) . dcca coefficient @xmath42 calculated for two pairs of time series representing stock price logarithmic returns sampled with @xmath159 min frequency . ( top ) two stocks from the same industrial sector : microsoft vs. intel . ( bottom ) two stocks from different industrial sectors : microsoft vs. 3 m . each panel for both pairs shows @xmath42 calculated for different value of @xmath2 ( heavy black line ) together with the mean @xmath160 and the standard deviation @xmath91 obtained from 100 independent realizations of the shuffled surrogate data ( thin blue / grey lines with error bars ) . the inverted values of @xmath42 for @xmath59 are denoted by circles . ] dcca coefficient @xmath42 calculated for two pairs of time series representing stock price logarithmic returns sampled with @xmath159 min frequency . ( top ) two stocks from the same industrial sector : microsoft vs. intel . ( bottom ) two stocks from different industrial sectors : microsoft vs. 3 m . each panel for both pairs shows @xmath42 calculated for different value of @xmath2 ( heavy black line ) together with the mean @xmath160 and the standard deviation @xmath91 obtained from 100 independent realizations of the shuffled surrogate data ( thin blue / grey lines with error bars ) . the inverted values of @xmath42 for @xmath59 are denoted by circles . ] dcca coefficients @xmath160 calculated for the stock pairs divided into three groups : the stocks representing different industrial sectors ( solid black ) , the stocks representing the same sector but different subsectors ( dashed red / grey ) , and the stocks representing the same subsector ( dash - dotted green / grey ) . dotted lines denote the zero - correlation level . ] for @xmath161 corresponding to small fluctuations , there is no statistically significant indication of cross - correlations between any pair of stocks . genuine cross - correlations are however visible starting from @xmath135 up to high positive @xmath2s . one can see here that even if a vast majority of the @xmath42 values are originally larger than 1 ( see the red dots representing the inverted values of the coefficient for @xmath135 ) , the cross - correlations may be considered genuine as long as the function @xmath42 is approximately stable ( i.e. , without strong fluctuations ) over some range of the scales . it is interesting to notice the weakening of the cross - correlations if one goes from small positive @xmath2s towards larger ones . it may be interpreted as a manifestation of the fact that the most cross - correlated fluctuations are those of the medium amplitudes , while the largest fluctuations are less correlated among the stocks . one can also infer that in both examples , the coefficients @xmath42 are larger for the small scales and smaller for the large scales . however , it can be the opposite for different choice of the stocks ( not shown ) so there is no regularity here . by looking at both parts of fig . [ fig::stocks ] , one can see that there is an important difference between them : the correlations are much stronger ( larger values of @xmath42 ) for the stocks representing mutually related industrial sectors ( microsoft@xmath158intel , fig . [ fig::stocks](a ) ) , while they are weaker for the stocks from unrelated sectors ( microsoft@xmath1583 m fig . [ fig::stocks](b ) ) . that this is a systematic effect one can learn from fig . [ fig::sectors ] , where three groups of the stock pairs are considered , representing different levels of the industrial similarity between companies : ( i ) the stocks corresponding to different industrial sectors , ( ii ) the stocks representing the same sector but different subsectors , and ( iii ) the stocks corresponding to the same subsector ( according to @xcite ) . obviously , the similarity level increases here gradually from ( i ) to ( iii ) . we calculate @xmath42 for each pair of stocks ( 4950 pairs total : 4218 pairs in ( i ) , 548 pairs in ( ii ) , and 184 pairs in ( iii ) ) and then average the obtained functions over the pairs within each group . [ fig::sectors ] confirms that the values of @xmath160 systematically increase if the pairs consist of the stocks with increasing industrial similarity . it is worth to mention that the correlations that lead to the non - zero values of @xmath42 are nonlinear . we infer this from a failed test of the null hypothesis that the surrogate signals with the same pearson autocorrelation function can reproduce the @xmath42 functions . [ fig::fourier.surrogates ] documents the results of this test , showing that , on average , the fourier - phase shuffling leads to complete destruction of the correlations that @xmath1 is sensitive to ( compare with fig . [ fig::stocks](a ) ) . this is especially evident for the positive @xmath2s , for which the standard deviation @xmath91 obtained from 100 independent realizations of the surrogates is relatively small , especially for the small and medium scales @xmath14 . for the large scales the mfdfa and mfcca procedures are less efficient in detrending ( the related polynomial degree @xmath21 is too low in this case ) and this produces the non - zero deviations of @xmath160 observed in fig . [ fig::fourier.surrogates ] . dcca coefficient @xmath160 and its standard deviation @xmath91 calculated for the microsoft@xmath158intel pair ( the same as in fig . [ fig::stocks](top ) ) after randomizing the fourier phases of the original signals , which destroyed all the statistical dependences except for the pearson autocorrelation function . 100 independent realizations of such surrogate data are used ( thin blue / grey lines with error bars ) . ] the second empirical example comes from the forex . we consider high - frequency ( @xmath162=1 min ) logarithmic returns of the exchange rates among a set of major currencies recorded over the years 2004 - 2008 . [ fig::currencies ] presents @xmath42 calculated for two such time series corresponding to the usd / eur and gbp / usd rates . since euro and the british pound are positively cross - correlated ( both are currencies used in the european union countries ) and since the us dollar appears in the numerator of the first rate and in the denominator of the second rate , these time series reveal negative cross - correlations of their detrended fluctuations . they are the strongest for @xmath135 and @xmath163 , while their strength decreases if we move away from these values in both directions . one can also see in the topmost panels of fig . [ fig::currencies ] that even for @xmath164 there is statistically significant indication of genuine cross - correlations for large @xmath14 ( if compared with the standard deviation of the shuffled surrogates ) . we infer from these results that the most strongly cross - correlated fluctuations of the forex data are those of the medium amplitudes . the large - amplitude fluctuations are relatively weaker but still substantially cross - correlated and even the fluctuations with the small amplitudes reveal weak cross - correlations ( unlike their uncorrelated counterparts from the stock market in fig . [ fig::stocks ] ) . dcca coefficient @xmath42 calculated for a pair of time series representing returns of the usd / eur and gbp / usd exchange rates sampled with @xmath165 min frequency . each panel shows @xmath42 calculated for different value of ( heavy black line ) together with the mean @xmath160 and the standard deviation @xmath91 obtained from 100 independent realizations of the shuffled surrogate data ( thin blue / grey lines with error bars ) . the inverted values of @xmath42 for @xmath59 are denoted by circles . ] we proposed a new measure of the multiscale detrended cross - correlations between a pair of time series , called the @xmath2dcca coefficient . this coefficient forms actually a family of functions @xmath42 that depend on the exponent @xmath2 and the temporal scale @xmath14 . like in the case of the multifractal , @xmath2-dependent fluctuation functions , by varying a value of @xmath2 , we can amplify those fluctuations that are of specific amplitudes : e.g. , the small ones for @xmath166 or the large ones for @xmath167 , and study the structure of the cross - correlations among such fluctuations . a specific case of @xmath1 for @xmath27 corresponds to the dcca coefficient @xmath0 already known from literature @xcite , so from this perspective the @xmath1 coefficient may be viewed as its generalization . for @xmath46 , the interpretation of @xmath1 is straightforward as its value is confined in the range @xmath168 : the maximum value of @xmath169 means that the corresponding detrended fluctuations of both signals under study show exactly the same correlation structure , @xmath170 means that such fluctuations are completely anticorrelated , and @xmath104 indicates that they are independent of each other . for @xmath59 , on the other hand , the coefficient @xmath171 may assume arbitrarily large values , which can lead to interpretation problems , but this may happen especially if the detrended fluctuations of the two signals are uncorrelated . knowing that , we may be concerned about the exact values of @xmath1 in this case . for practical reasons , we overcome this problem by replacing the original values @xmath172 by their inversion and thus forcing them to fit into the normal interval @xmath84 . in the case of no correlations , a course of the function @xmath42 becomes strongly unstable and its values wildly fluctuate . this effect can thus serve as an optical indication of the signal independence . in contrast , if the signal fluctuations are cross - correlated , we obtain largely stable behavior of @xmath42 even for @xmath87 and even if the original values of @xmath79 slightly exceed 1 . therefore , in this case a lack of wild fluctuations of @xmath42 and its non - zero value can signalize the existence of cross - correlations . all this means that the family of the coefficients @xmath1 is a well - defined tool of the multiscale detrended cross - correlation analysis . obviously , because of the more demanding interpretation of results for @xmath59 , one may restrict an analysis to the @xmath46 case , but this is equivalent to ignoring the cross - correlation structure of small fluctuations , so we do not recommend it . in order to illustrate the performance of this new measure , we applied it to a few selected data sets representing long - memory processes : arfima and the markov - switching multifractal . we showed that @xmath1 is able to identify correctly specific cross - correlations that remain undetected by the standard dcca cross - correlation coefficient @xmath0 as well as it is able to distinguish between the processes whose detrended fluctuations are cross - correlated in different amplitude ranges @xmath158 a task that is infeasible with @xmath0 . we also carried out the @xmath1 analysis based on sample empirical data from the financial markets and found that the detrended cross - correlations among such data also depend on the detrended - fluctuation amplitude . straightforward and likely fruitful for a more global correlation analysis would be to form the @xmath2-dependent counterparts of the conventional correlation matrices for larger sets of multivariate data and to study their spectral properties as it is commonplace in the standard correlation matrix analysis . by proceeding further in this direction , one may then also consider constructing the @xmath2-dependent graphs . on the other hand , a parallel approach would be defining the @xmath2-generalized version of the dmca coefficient @xmath173 @xcite in a full analogy to our approach . t. mikosch , c. stric , rev . 86 * , 378 ( 2004 ) . bassler , j.l . mccauley , g.h . gunaratne , proc . usa * 104 * , 17287 ( 2007 ) . pagan , j. econometrics * 45 * , 267 ( 2002 ) . schmitt , d. chetalova , r.schfer , t. guhr , epl * 103 * , 58003 ( 2013 ) . g. mayer - kress , integ . sci . * 29 * , 205 - 210 ( 1994 ) . peng , s. havlin , h.e . stanley , a.l . goldberger , chaos * 5 * , 82 ( 1995 ) . j. kohlmorgen , k .- mller , j. rittweger , k. pawelzik , biol . cybern . * 83 * 73 ( 2000 ) . bernaola - galvn , p.ch . ivanov , l.a.n . amaral , h.e . stanley , phys . lett . * 87 * , 168105 ( 2001 ) . j. perell , j. masoliver , a. kasprzak , r. kutner , phys . e * 78 * 036108 ( 2008 ) . j. kwapie , s. drod , phys . rep . * 515 * , 115 ( 2012 ) . k . peng , s.v . buldyrev , s. havlin , m. simons , h.e . stanley , a.l . goldberger , phys . e * 49 * , 1685 ( 1994 ) . b. podobnik , h.e . stanley , phys . lett . * 100 * , 084102 ( 2008 ) . b. podobnik , d. horvatic , a.m. petersen , h.e . stanley , proc . usa * 106 * , 22079 ( 2009 ) . n. xu , p. shang , s. kamae , nonlin . dyn . * 61 * , 207 ( 2010 ) . g.f . zebende , p.a . da silva , a. machado filhoa , physica a * 390 * , 1677 ( 2011 ) . qian , y .- m . liu , z .- q . jiang , b. podobnik , w .- x . zhou , h.e . stanley , arxiv:1504.02435 ( 2015 ) . kantelhardt , s.a . zschiegner , a. bunde , s. havlin , e. koscielny - bunde , h.e . stanley , phys . a * 316 * , 87 ( 2002 ) . p. owicimka , j. kwapie , s. drod , physica a * 347 * , 626 ( 2005 ) . muzy , e. bacry , r. baile , p. poggi , europhys . lett . * 82 * , 60007 ( 2008 ) . ivanov , l.a.n . amaral , a.l . goldberger , s. havlin , m.g . rosenblum , z.r . struzik , h.e . stanley , nature * 399 * , 461 ( 1999 ) . udovichenko , p.e . strizhak , theor . exp . chem . * 38 * , 259 ( 2002 ) . a. witt and b. d. malamud , surv . geophys . 34 , 541 ( 2013 ) . l. calvet , a. fisher , rev . 84 * , 381 ( 2002 ) . s. drod , j. kwapie , p. owicimka , r. rak , epl * 88 * , 60003 ( 2009 ) . s. drod , j. kwapie , p.owicimka , r. rak , new j. phys . * 12 * , 105003 ( 2010 ) . e. koscielny - bunde , j.w . kantelhardt , p. braund , a. bunde , s. havlin , j. hydrology * 322 * , 120 ( 2006 ) . kantelhardt , e. koscielny - bunde , d. rybski , p. braun , a. bunde , s. havlin , j. geophys . atmos . * 111 * , d01106 ( 2006 ) . m. ausloos , phys . e * 86 * , 031108 ( 2012 ) . jafari , p. pedram , l. hedayatifar , j. stat . p04012 ( 2007 ) . p. owicimka , j. kwapie , i. celiska , s. drod , r. rak , arxiv:1106.2902 ( 2011 ) . d. grech , g. pamua , physica a * 387 * , 4299 ( 2008 ) . e. maiorino , l. livi , a. giuliani , a. sadeghian , a. rizzi , physica a * 428 * , 302 ( 2015 ) . d. makowiec , a. dudkowska , r. gaaska , a. rynkiewicz , physica a * 388 * , 3486 ( 2009 ) . l. kristoufek , phys . e * 91 * , 022802 ( 2015 ) . s. drod , p. owicimka , phys . e * 91 * , 030902(r ) ( 2015 ) . g. rotundo , m. ausloos , c. herteliu , b. ileanu , physica a * 429 * , 109 ( 2015 ) . p. owicimka , j. kwapie , s. drod , phys . e * 74 * , 016103 ( 2006 ) . zhou , phys . e * 77 * , 066211 ( 2008 ) . z .- q . jiang , w .- x . zhou , phys . e * 84 * , 016106 ( 2011 ) . chen , physica a * 390 * , 297 ( 2011 ) . z. li , x. lu , physica a * 391 * , 3930 ( 2012 ) . wang , c. xie , acta phys . pol . b * 43 * , 2021 ( 2012 ) . p. owicimka , s. drod , m. forczek , s. jadach , j. kwapie , phys . e * 89 * , 023305 ( 2014 ) . s. chen - hua , l. chao - ling , s. ya - li , physica a * 419 * , 417 ( 2015 ) . zebende , phys . a * 390 * , 614 ( 2011 ) . vassoler , g.f . zebende , phys . a * 391 * , 2438 ( 2012 ) . zebende , m.f . da silva , a. machado filho , phys . a * 392 * , 1756 ( 2013 ) . reboredo , m.a . rivera - castro , g.f . zebende , energy econ . * 42 * , 132 ( 2014 ) . l. kristoufek , physica a * 402 * , 291 - 298 ( 2014 ) . l. kristoufek , physica a * 406 * , 169 - 175 ( 2014 ) . b. podobnik , z .- q . jiang , w .- x . zhou , h.e . stanley , phys . e * 84 * , 066118 ( 2011 ) . j. hosking , biometrika * 68 * , 165 ( 1981 ) . gil - alana , j. appl . climatol . * 51 * , 1904 ( 2012 ) . f. yusof , i.l . kane , z. yusop , aip conf . proc . * 1643 * , 446 ( 2015 ) . k. burnecki , a. weron , j. stat . mech . * 2014 * , p10036 ( 2014 ) . k. torre , d. delignires , l. lemoine , brit . . psychol . * 60 * , 85 ( 2007 ) . a. leite , a.p . rocha , m.e . silva , chaos * 23 * , 023103 ( 2013 ) . saakian , phys . rev . e * 85 * , 031142 ( 2012 ) . t. lux , _ the markov - switching multifractal model of asset returns : gmm estimation and linear forecasting of volatility _ , economic working paper no . 2006/17 ( christian - albrechts - universitt kiel , 2006 ) . b. rypdal , k. rypdal , j. goephys . res . * 116 * , a02202 ( 2011 ) . l. calvet , a. fisher , _ multifractal volatility : theory , forecasting , and pricing _ , academic press ( 2008 ) . r. liu , t. di matteo , t. lux , physica a * 383 * , 35 ( 2007 ) . yahoo ! finance , _ http://finance.yahoo.com _ . l. kristoufek , physica a * 428 * , 194 ( 2015 ) .

spectra for 2d stars in the 1.5d approximation are created from synthetic spectra of 1d non - local thermodynamic equilibrium ( nlte ) spherical model atmospheres produced by the phoenix code . the 1.5d stars have the spatially averaged rayleigh - jeans flux of a k3 - 4 iii star , while varying the temperature difference between the two 1d component models ( ) , and the relative surface area covered . synthetic observable quantities from the 1.5d stars are fitted with quantities from nlte and local thermodynamic equilibrium ( lte ) 1d models to assess the errors in inferred values from assuming horizontal homogeneity and lte . five different quantities are fit to determine the of the 1.5d stars : ubvri photometric colors , absolute surface flux seds , relative seds , continuum normalized spectra , and tio band profiles . in all cases except the tio band profiles , the inferred value increases with increasing . in all cases , the inferred value from fitting 1d lte quantities is higher than from fitting 1d nlte quantities and is approximately constant as a function of within each case . the difference between lte and nlte for the tio bands is caused indirectly by the nlte temperature structure of the upper atmosphere , as the bands are computed in lte . we conclude that the difference between values derived from nlte and lte modelling is relatively insensitive to the degree of the horizontal inhomogeneity of the star being modeled , and largely depends on the observable quantity being fit .

1404.7028

red giant stars rank among the brightest stars in the galaxy , being generally much brighter in the visible band than main sequence stars of the same spectral type . but this only partially accounts for their brightness as large portion of their flux is emitted in the near - ir . this , combined with their enormous surface areas , gives them such large luminosities that they are easily observable even in very remote stellar populations . red giants grant us a tool for probing nearly all regions of the galaxy using a common indicator , a feat unparalleled by most other types of stars . [ [ section ] ] because many red giants are low to intermediate mass stars that have evolved beyond the main sequence , generally found in older stellar populations , their abundances can be indicators of early galactic chemical evolution . for example , by comparing observations of galactic bulge giants with those of giants located in the thin and thick disks and halo , it has been shown that the bulge likely experienced similar formation timescales , chemical evolution histories , star formation rates and initial mass functions as the local thick disk population @xcite . bulge and disk giants show some differences in their chemical abundances , with the bulge giants showing a higher relative abundance of select elements than the disk giants , suggesting more rapid chemical enrichment , possibly by ejecta from supernovae of types ia and ii @xcite . observations of red clump giants in the bulge have also produced additional evidence of a central bar @xcite , with their apparent visual magnitudes being brighter at some galactic latitudes than others . [ [ section-1 ] ] the tip of the red giant branch ( trgb ) can also be used as a standard candle to determine distances to nearby galaxies . the distance moduli obtained from the @xmath2 pass - band magnitude of the trgb are comparable with those from primary distance indicators like cepheids and rr lyraes @xcite , and in some cases even suggest a reevaluation of the metallicity dependence and zero point calibration of the cepheid distance scale @xcite . the trgb method even has advantages over other distance determinations like those of cepheids and rr lyraes : @xmath3 the trgb method requires much less telescope time than variable stars ; @xmath4 the @xmath2 magnitude of the trgb is insensitive to the variation of metallicity for @xmath5}<-$ ] 0.7 ; and @xmath6 the trgb suffers less from extinction problems than cepheids , which are in general located in star - forming regions @xcite . much of what is now known about all types of stars comes from fitting the predicted quantities from atmospheric models to observations . estimates of the solar chemical abundances come from fitting synthetic spectral line profiles and equivalent widths to those observed in the sun @xcite . calibrations of stellar parameters , such as @xmath7 , @xmath8 , @xmath1 , @xmath9 , and @xmath10 , for different spectral types are found from fitting models @xcite . beyond studying single stars , model atmospheres can be used to determine qualities of larger structures as well . the age - metallicity and color - metallicity relations of globular clusters can be determined from the abundances of individual red giants within the clusters @xcite . [ [ section-2 ] ] in this work , we will explore the limitations of two of the simplifying assumptions of atmospheric modeling : horizontal homogeneity and local thermodynamic equilibrium ( lte ) . both of these assumptions have been generally adopted because they are more computationally practical than the alternatives , requiring less time and fewer resources to arrive at a result . by comparison , horizontal inhomogeneity requires model atmospheres to be calculated in two or three geometric dimensions ( 2d or 3d models ) instead of just one , and non - lte ( nlte ) requires that each level population be computed in statistical equilibrium ( se ) using iterative processes . [ [ section-3 ] ] however , horizontal homogeneity and lte both limit how realistic a model can be . simulations of red giant atmospheres performed in 3d have confirmed that turbulent surface convection causes horizontal inhomogeneities to form @xcite , such as visually observable surface features like solar granulation @xcite . these features are known to lead to detectable effects , such as altering predicted line strengths and shapes and , thus , inferred elemental abundances @xcite . for 3d models of red giant atmospheres , whose modeling parameters span the ranges of 3600 k @xmath11 5200 k , 1.0 @xmath12 3.0 , and -3.0 @xmath13}\leq$ ] 0.0 , granules have been shown to span a range of sizes from as small as on the order of 10@xmath14 cm to as large as 2 @xmath15 10@xmath16 cm , with the majority on the order of 10@xmath17 cm @xcite . the cooler stars and stars with lower values of @xmath18 generally display larger features . for the same set of 3d models , the root mean square ( rms ) temperature variation among these features at optical depth unity is usually in the range of @xmath19 2 to 5 @xmath20 , or @xmath19 200 to 300 k for the parameters listed above , although variations can reach 2000 k between the hottest and coolest areas @xcite . most of the 3d models reported in the literature have @xmath1 @xmath21 4500 k , with @xmath22 varying by @xmath19 200 k. [ [ section-4 ] ] for stars exhibiting horizontal inhomogeneities , where the temperature varies across the features , @xmath1 is no longer a well defined quantity . by definition , @xmath1 is derived from the stefan - boltzmann law , where the bolometric luminosity ( @xmath23 ) of a star is proportional to the fourth power of its @xmath1 ( @xmath24 ) . the @xmath24 of a horizontally inhomogeneous star may be similarly found by summing the @xmath23 of each of the inhomogeneous components , weighted by their relative surface coverage , and taking the fourth root of the result . this quantity is defined empirically by intrinsic properties of real or model stars and is independent of fitting models to observable quantities . alternatively , another model fitting independent @xmath1 can be defined from the long wavelength tails of stellar spectra . because flux has a linear dependence on @xmath1 in the rayleigh - jeans limit , an estimate of the @xmath1 can be made from measuring the absolute surface flux of the r - j tail . this dependence is used by the infrared flux method @xcite to determine @xmath1 from the r - j tails of spectra ( @xmath25 ) . [ [ section-5 ] ] the key idea for a horizontally inhomogeneous star having the same @xmath24 as a horizontally homogeneous star , is that the seds will differ based on the variation of the modeling parameters , such as @xmath1 , across the inhomogeneous surface . for example , using the planck function , @xmath26 to describe the shape of a stellar continuum illustrates the issue directly @xcite . a star with a range of differing temperatures across the surface does not have a directly obvious value of @xmath1 that should be used , and because of the equation s non - linear dependence on @xmath1 , this is an important question . the higher temperature material will contribute disproportionately more flux than the lower temperature material , with a dependence on wavelength , and will alter the spectrum from that of a horizontally homogeneous star of the same @xmath24 accordingly . [ [ section-6 ] ] the most noticeable departure of lte models from observed stars comes from comparing computed and observed spectral features and seds . for 1d models of red giants , nlte models have been shown to be more accurate than lte models in predicting the overall monochromatic flux ( @xmath27 ) levels of seds and strengths of individual spectral lines , with the notable exceptions of molecular absorption bands and the near - uv band flux @xcite . calculating the molecular level populations in nlte is computationally demanding , and is not handled in many atmospheric modeling codes . both lte and nlte models over - predict the near - uv @xmath27 levels of cool red giants . in the case of the near - uv @xmath27 levels , nlte models are worse in the over - prediction than their lte counterparts . the nlte effects of fe group elements on the model structure and @xmath27 distribution have been shown to be much more important for predicting a sed than the nlte effects of all the light metals combined , and serve to substantially increase the near - uv @xmath27 levels as a result of nlte fe i overionization @xcite . the magnitude of this effect has been shown to be inversely proportional to the completeness of the fe i atomic model used in the atmospheric modeling @xcite , discussed in detail in section [ phoenix ] . these failures of 1d nlte models to predict observable quantities may be , in part , related to the exclusion of horizontal inhomogeneities in the models . our primary goal is investigating the relation between errors introduced in determining a star s @xmath1 from assuming horizontal homogeneity and those introduced by assuming lte . we look to determine how distinguishable 3d hydro and nlte effects are when looking at low resolution diagnostics ( photometric colors and overall seds ) . the error inherent in assuming lte is expected to remain approximately constant for different levels of horizontal inhomogeneity , as any changes in a lte spectrum caused by the inhomogeneities should also be represented to a similar degree in the corresponding nlte spectrum , as the relative difference in the flux distribution between lte and nlte should remain roughly constant within the @xmath1 range studied in this work . [ [ section-7 ] ] section [ methods ] outlines the parameters used for the 1d nlte and lte model grids , and the methods used in creating and processing the 1.5d nlte spectra . section [ results ] presents the results of fitting 1d seds and spectra to the 1.5d seds and spectra . section [ summary ] gives a brief summary and discussion of the results . in general , the calculation of a 2d model atmosphere requires a very substantial increase in computational effort over that needed for 1d models . however , there are two regimes of validity for which the situation simplifies ; these are often referred to as 1.5d models . the valid regimes are either 1 ) when the geometric scale of the inhomogeneities is so small that they are very optically thin , and every emergent ray may be assumed to have fully sampled both warm and cool components of the atmosphere many times , or 2 ) that the inhomogeneities are large enough that they are optically thick to their own radiation so that the flux is similar to that of two spatially unresolved stars . regardless , both regimes require that the inhomogeneities be small enough and randomly distributed so that there is no correlation between component @xmath1 values and limb darkening . by assuming the average size of surface features to be that of 3d hydro modeled red giant granules , we satisfy both conditions of the larger inhomogeneity regime . [ [ section-8 ] ] we construct our 1.5d seds by taking the linear average of two 1d nlte seds . two - dimensional information , such as the difference in temperature among features or the relative portion of the stellar surface covered by the different features , can be approximated by choosing the parameters of the 1d components being averaged , and the averaging weights . these 1.5d nlte seds and derived colors , approximating 2d effects , are treated as artificial `` observations '' of real stars that are fitted with a library of nlte and lte 1d trial spectra for inferring the @xmath1 of the 1.5d stars . a grid of 50 1d stellar model atmospheres and corresponding synthetic spectra was produced for this purpose using the phoenix code . the phoenix code can be used to model nlte atmospheres and spectra of stellar objects throughout the h - r diagram @xcite . phoenix version 15 is utilized for all modeling calculations in this work . it is important to include nlte effects in model atmospheres and synthetic spectra . together , they can lead to a difference in the blue and near - uv bands flux of up to 50% over corresponding phoenix lte models of red giant stars as a result of nlte fe i overionization @xcite , displayed in fig . [ fig : exampleseds ] . it is also understood that nlte radiative equilibrium in these stars is model - dependent and is known to depend on the completeness of the atomic model of fe i and the completeness and accuracy of the collisional cross - sections for fe i ionization . while the free electron collisional cross - sections for fe i used by phoenix are robust , it does not take into account h collisions , and it has been cautioned that this may cause phoenix to overestimate the effects of fe i overionization @xcite . additionally , it was shown by @xcite that using a more complete fe i atomic model than that used by phoenix will reduce overionization effects by providing more fe i high energy excited states to facilitate recombinations from fe ii . they also found that the greater the number of energy levels within @xmath28 = k@xmath1 of the ground state ionization energy ( @xmath29 ) , the more accurate the nlte ionization equilibrium solution . the number of energy levels in our model fe i atom within @xmath28 = k@xmath1 of @xmath29 ranges from one to four as @xmath1 increases from 4250 ( @xmath25 of our 1.5d spectra ) to 4750 k ( our hottest 1.5d component ) . this has been found to be too few to accurately compute the nlte recombination rate , and we expect to over - estimate the nlte fe i overionization . however , we note that the 1.5d target spectra and the 1d library of fitted nlte spectra were computed with the _ same _ fe i atomic model . therefore , we expect the difference between fitted 1d nlte @xmath1 values and @xmath24 to be dominated by horizontal inhomogeneity effects . we also note that the magnitude of any difference in inferred @xmath1 value between fitting with 1d nlte spectra and 1d lte spectra will be over estimated , especially for any diagnostics involving wavebands bluer than the v filter , where the overionization effects are dominant . therefore we concern ourselves primarily with how this difference depends on the degree of inhomogeneity in the 1.5d spectra being fit , and ignore the magnitude . [ [ section-9 ] ] the stellar parameters adopted for our @xmath25 equivalent 1d star are @xmath1 = 4250 k , log @xmath7 = 2.0 , and [ m / h ] = -0.5 with the alpha elements , from o to ti , enhanced by [ a / h ] = + 0.3 dex . the nlte models were constructed by treating the 20 atomic species listed in table [ tab : nlte species list ] as nlte species when calculating energy level populations , all of them in the neutral state and most in the singly ionized state as well @xcite . phoenix does not calculate level populations for molecular species in nlte ; molecules are treated in lte for nlte models and spectra . [ [ section-10 ] ] nlte radiative equilibrium is complex in that any given transition may either heat or cool the atmosphere with respect to lte , depending on how rapidly @xmath30 increases inward at the transition wavelength , whether the transition is located in the wien or rayleigh - jeans regime , and whether the transition is a net heater or cooler in lte with respect to the gray atmosphere @xcite . for our @xmath1 range , all of the nlte models exhibit a surface cooling effect in the outermost layers of the atmosphere . this is displayed in fig . [ fig : examplettau ] as a lower temperature in the nlte models than in the lte ones for a given @xmath31 . for the majority of the upper atmosphere , between @xmath32 10@xmath33 and 10@xmath34 , the opposite is seen , with the nlte models having a higher temperature than the lte ones . [ [ section-11 ] ] to prepare the grid of models , we varied the @xmath1 value from 3550 k to 4750 k , with @xmath35 = 50 k , and produced both lte and nlte models . upon convergence of a model structure , two different spectra were synthesized for each model , the fully line blanketed sed and the pure continuum sed for use in normalizing the flux in the line blanketed seds . this calculation is always performed in lte even when synthesizing the continuum of a nlte model because phoenix can not omit the nlte b - b transitions in the calculations when synthesizing a spectrum . in all spectral synthesis cases , the seds are sampled from @xmath36 = 3000 to 13000 @xmath37 as shown in table [ tab : wavelength - grid - spacing ] , with variable spacings , @xmath38 , to approximately preserve the spectral resolution , @xmath39 , across the full range at @xmath40 300000 to 400000 . in addition to this , the nlte seds are sampled at supplementary points that ensure each nlte spectral line is critically sampled ; these points are automatically distributed over each line by phoenix . [ [ section-12 ] ] finally , to increase the apparent formal numerical resolution of the grid s @xmath1 range sampling , additional seds were linearly interpolated between neighboring seds in the grid to reach a final apparent temperature resolution of 25 k. this was done to smooth the final fitted @xmath1 value versus degree of inhomogeneity relation . however , we note that this does not decrease the formal uncertainty of the @xmath1 determination . linear interpolation of the flux values was used instead of interpolating the log flux values because at the small ratio of @xmath41 , the relative difference of the monochromatic flux ( @xmath42 ) between the two methods was less than 1.0 % at all wavelength sampling points . seventeen unique 1.5d seds were produced to serve as artificial 2d targets by linearly averaging two synthetic nlte 1d seds . these were individually distinguished by the differences in the @xmath1 values of their warm and cool 1d components , @xmath0 . this @xmath0 corresponds to the contrast of inhomogeneities . it was enforced that , for each 1.5d sed , the linear average of their component @xmath1 values be 4250 k , effectively giving each of the 1.5d seds a @xmath25 = 4250 k. the production of 1.5d seds by the linear averaging of the component fluxes in this way results in continuum levels in the rayleigh - jeans tails being identical within a few percent , and was chosen to maintain a consistent observational property among all of the 1.5d seds . [ [ section-13 ] ] two different weighting schemes were used in creating the 1.5d seds : 1 ) evenly weighting both components as a simple reference case ; and 2 ) weighting the hot component at a ratio of 2:1 to the cool component . this ratio approximately represents the relative solar surface area covered by granules and intergranular lanes respectively @xcite . these two methods simulate a variation in surface coverage , or filling factor ( ff ) , of the hot and cool features . hereafter , the evenly weighted method is referred to as having a 1:1 ff , and the method with the hot component being weighted at 2:1 as referred to as having a 2:1 ff . to fully explore the effects of horizontal inhomogeneity , we take the upper limit of 300 k for the rms temperature difference across the features found from 3d modeling , and double it to 600 k , for a maximum realistic value of @xmath0 . we then extend this value up to 1000 and 1050 k , for the 1:1 and 2:1 ff respectively , to test how large the rms temperature variations would have to be before 1d estimates of @xmath1 become very misleading , with significant departures from @xmath24 . table [ tab:1.5d - sed ] displays the components used for the two schemes , at each @xmath0 that exactly preserves both the @xmath25 value as 4250 k and the respective ff , as well as displaying the resultant 1.5d @xmath24 as found from the average of the components bolometric luminosities . [ [ section-14 ] ] we attempted to recover the @xmath1 values of the 1.5d stars using three different methods of fitting 1d models : ubvri photometry , spectrophotometry , and spectroscopy . common to each method of post - processing , the 1.5d seds were interpolated from their initial wavelength distribution to a new distribution with a constant @xmath38 of 0.006 @xmath37 . as a preliminary test to see which diagnostics would be expected to return @xmath1 values close to @xmath24 , and which would return values differing from @xmath24 , we examined the relative difference between two of our 1:1 ff spectra , @xmath0 = 200 and 1000 k , and 1d models with their equivalent @xmath24 . these spectra were convolved with a gaussian kernel of fwhm = 50 @xmath37 , representative of the nominal resolution element of the observed spectrophotometry in the burnashev catalogue @xcite , prior to taking the differences . while there is little difference between the 1d and 1.5d in the near - ir and parts of the visible wavebands , they begin to diverge rapidly in the uv as seen in fig . [ fig : safe - zone ] . this suggests that diagnostic tools focused on redder wavelengths are more likely to return @xmath1 values close to @xmath24 . a more detailed high - resolutions analysis at the level of individual spectral lines could reveal specific diagnostics that are robust against 2d effects , but was beyond the scope of this study . for each of the nlte and lte 1d and nlte 1.5d seds , photometric colors were produced using bessel s updated johnson - cousins ubvri photometry @xcite . the transmission data for each filter were interpolated to the wavelength distribution using quadratic splines . five different color indices , u@xmath43-b@xmath43 , b - v , v - r , v - i , and r - i , were calculated to milli - magnitude precision for each of the seds . it should be noted that the u band is of limited accuracy as the short wavelength cutoff in the observational u band is not defined by the instrumental configuration , but rather ozone absorption in earth s atmosphere . because ozone column density varies by @xmath1920 % with geographic location and season , it is difficult to accurately reproduce u band photometry synthetically from models . these values were calibrated with a phoenix nlte synthesized sed for the standard star vega ( @xmath44 lyr , hr7001 , hd172167 ) , using a single - point photometric calibration independent of color . [ [ section-15 ] ] while three of the five color indices for the 1d seds behaved as expected from the behavior of @xmath45 , with their values monotonically increasing for decreasing @xmath1 values over the range of the 1d seds , both the u@xmath43-b@xmath43 and b - v indices stopped increasing and started decreasing at @xmath1 = 3900 k for the lte seds , and at @xmath1 = 3800 k for the nlte seds . this turnover is observed in stars , but between @xmath1 = 3540 k and 3380 k for the u - b index @xcite . the phenomenon is caused by spectral features in the b filter growing in strength more rapidly with decreasing @xmath1 values than those in the u filter , and reverses the trend in the color index expected from @xmath45 . likewise , the spectral features in the v filter grow more rapidly than those in the b filter . in this case , the incorrect prediction of the @xmath1 value of the turnover is caused by the excess blue and uv flux common to phoenix models of cool red giant stars @xcite . [ fig : turnover ] shows the b - v index value trend , including turnover , for both the lte and nlte 1d seds . [ [ section-16 ] ] to quantify the errors from fitting 1d nlte and lte photometry to nlte 2d stars , the library of 1d color index values was fitted to each of the 1.5d color index values . the closest matching @xmath1 value was found by means of inspection . whichever 1d sed had the smallest difference between its value and the 1.5d value for each index was chosen as the match . for example , fig . [ fig : turnover ] shows the value of the b - v index for the 1.5d sed of 1:1 ff with @xmath46 = 1000 k including the region of closest match to 1d b - v values . because of the degeneracy caused by the turnover in the u@xmath43-b@xmath43 and b - v indices and the limiting @xmath1 numerical resolution of the 1d grid , it was possible that the best matching @xmath1 value would fall outside of the range of temperatures enclosed by the cool and hot 1d components , and would not be the correct match . automatically limiting the best match to within the components @xmath1 range was determined to be too restrictive , so the closest three matches for both of the indices were found and ranked in order , whereupon the correct best match was chosen by inspection to logically fit the @xmath1 range . [ [ spectrophotometry ] ] spectrophotometry + + + + + + + + + + + + + + + + + absolute surface flux seds , @xmath47 , of the 1.5d models were fitted with 1d seds to find the closest match to the the total energy radiated by the 1.5d stars ; the relative flux seds were fitted with 1d relative seds to find the closest match to the overall shape of the 1.5d seds . for the relative flux seds , the surface flux values were normalized to the average flux within a 10 @xmath37 window ( @xmath48 ) located between @xmath36 = 12750 and 12760 @xmath37 in the rayleigh - jeans tail of each sed . a 10 @xmath37 window was chosen over a single data point to avoid having selected a point that unreliably represents the continuum level in different seds . this window was chosen over others in the rayleigh - jeans tail because , it regularly had the fewest and weakest spectral features contained within its bounds . [ fig : pinned ] displays the relative version of the nlte seds with @xmath1 values of 4750 k , 4250 k , and 3550 k. [ [ spectroscopy ] ] spectroscopy + + + + + + + + + + + + continuum normalized spectra were fitted with 1d spectra to determine the closest match based upon the relative strength of the spectral features . continuum normalization was done for the 1d spectra by dividing the blanketed spectra by the corresponding continuum spectra . for the 1.5d seds , the same continuum normalization process was used , where the 1.5d synthetic continua were generated the same way that the 1.5d seds were , by linearly averaging the two corresponding 1d synthetic continua together . initial fitting of the full wavelength range showed that the 1.5d spectra could be fit well by 1d spectra in the blue / uv and the near ir wavebands , but showed a large discrepancy over most of the optical waveband between @xmath36 = 4800 and 8000 @xmath37 . this discrepancy , caused primarily by the distinctive tio molecular absorption bands present in cool star spectra @xcite , prompted a second fitting diagnostic for the continuum normalized spectra , by restricting the fit to the wavelength range between @xmath36 = 5500 and 8000 @xmath37 . an mgh band centered at 5167 @xmath37 was found to overlap with the tio band near this wavelength range , and depended differently on @xmath1 than the tio bands , affecting the quality of the fits . the overlapping mgh and tio bands and additional nearby tio bands are displayed in fig . [ fig : lineid ] , illustrating the different temperature dependence of the features . [ [ section-17 ] ] each of the high resolution 1d and 1.5d seds were convolved with a gaussian kernel of fwhm = 50 @xmath37 . the smoothed seds were then re - sampled to a much coarser @xmath38 spacing to more accurately reflect the number of degrees of freedom available when comparing with an observed sed , but still fine enough to critically sample every feature remaining after the smoothing convolution . this amounted to a sample of 4000 wavelength points spaced between @xmath36 = 3100 and 12900 @xmath37 , with 100 @xmath37 at each end of the original wavelength range having been lost to convolution edge effects . for both @xmath47 and relative surface fluxes , the new points were evenly distributed in logarithmic space instead of linear space , amounting to taking the log of the upper and lower bounds of the wavelength region , distributing the new sample points evenly between these values , and then exponentiating everything to return to linear space . this resulted in a distribution of wavelength points with smaller @xmath38 at shorter wavelengths and larger @xmath38 at longer wavelengths . the motivation behind this was to attribute additional weight to spectral features located in the blue end of the sed when determining a best fit , without arbitrarily weighting select wavelength regions more heavily or determining a best fit for these regions separate from the rest of the sed altogether . extra weighting was attributed to the blue band of the wavelength range because there are more spectral features per @xmath38 interval there , and it is more sensitive to changes in temperature than the red band , making it a more sensitive diagnostic tool for determining @xmath1 . the continuum normalized spectra were re - sampled in linear wavelength space to not grant any of the spectral features additional weight in the final fitting process . unlike the spectrophotometric seds , the relative strengths of individual spectral features are of interest here , not the overall shape of the spectrum . [ [ section-18 ] ] in all cases , the best fit 1d spectrum was determined by minimizing a modified pearson @xmath49 test statistic , of the form @xmath50 where @xmath51 is the number of degrees of freedom , @xmath52 is the observed frequency of a phenomenon to be fitted , and @xmath53 is the modeled or expected frequency of the phenomenon . the 1d and 1.5d fluxes were treated as the modeled and observed frequencies , respectively , because a spectral flux is in some sense the frequency at which photons of given energies emerge from stars . for each 1.5d sed , every 1d sed was individually treated as the model value to create a test statistic value . the 1d sed with the minimum @xmath49 value was chosen as the best fitting sed from which the 1.5d @xmath1 value was inferred . in each fitting case , the result was determined to be significant at a confidence level of @xmath54 = 0.05 . as a check for self consistency , the 1d nlte fitted results are expected to be bounded by the hot and cold components @xmath1 values for a given 1.5d star , but this restriction is not expected of the 1d lte fitted results . for all cases except one , the best fitting @xmath1 values for the 1.5d stars showed an increasing trend with increasing @xmath0 . this is expected from the non - linear dependence of @xmath55 on temperature , where a hotter @xmath1 value produces disproportionately more flux than a cooler one , and the averaged flux in 1.5d stars will be more than a @xmath1 value of 4250 k would suggest for a 1d star . this effect is even more pronounced at bluer wavelengths . taking the derivative of @xmath55 with respect to temperature , @xmath56 inspection reveals that for changing temperatures , @xmath55 exhibits larger relative changes at shorter wavelengths , causing the flux at the blue end of the spectrum to increase disproportionately faster than the rest of the wavelength range with increasing @xmath1 . [ [ section-19 ] ] for the case of the tio bands fitted in the continuum normalized spectra , the reverse trend was instead seen ; a decreasing best fit @xmath1 value with increasing @xmath0 . this is expected from the non - linear dependence of molecule formation on gas temperature @xcite . all fitting results are considered as having a formal uncertainty of @xmath5725 k , one half of the original numerical temperature resolution of the 1d grid . the fit results are summarized in tables [ tab : master - results - table-1 ] and [ tab : master - results - table-2 ] , while tables [ tab : master - difference - table-1 ] and [ tab : master - difference - table-2 ] display the differences between the fit results and the 1.5d @xmath24 . results for comparing with 1d lte spectra are limited to diagnostics only involving v - band photometry and redder , or equivalent wavelength ranges , as discussed in section [ phoenix ] . figs . [ fig : nlte - photometry - results ] and [ fig : nlte - photometry - results-2:1 ] presents each of the five photometric color indices nlte best fitting @xmath1 values as functions of @xmath0 for all of the 1.5d stars . each of the color indices shows increasing best fit @xmath1 values for the 1.5d stars with increasing @xmath0 . the slope of the @xmath58 relation is different for the five indices , creating , for a given 1.5d star , a spread of best fit @xmath1 values among the indices that grows with increasing @xmath0 . the slopes of the individual color indices fits are steeper for the bluer indices , and nearly flat for the r - i index over the @xmath0 range . such a spread in the fits shows that horizontal inhomogeneity effects are non - constant across the spectrum . this spread can not be resolved at a 25 k resolution for @xmath59 150 k , and grows as large as 375 k for 1.5d stars with 1:1 ff , and as large as 250 k for 2:1 ff . [ [ section-20 ] ] for the 2:1 ff fits , the spread is not as large as the 1:1 ff , because , while the 2:1 1.5d star is created with more hot material than cool material , the @xmath1 values of both components are lower than the respective components of a 1:1 1.5d star having the same @xmath0 , outweighing the contribution of more hot material by having less flux to contribute from the hot component . this resulted in either the same or lower temperature fits for a given @xmath0 . additionally , because the cool component of the 2:1 1.5d stars has a lower @xmath1 value than that of a 1:1 1.5d star , the blue and uv regions are even more severely line - blanketed , contributing even less flux than the difference in 1d @xmath1 values alone would suggest , and resulting in lower temperature fits for the color indices at shorter wavelengths , noticeably u@xmath43-b@xmath43 and b - v . [ [ nlte ] ] nlte + + + + to quantify the magnitude of nlte effects , we defined the quantity @xmath60 as the difference between the fitted nlte @xmath1 value and the fitted lte @xmath1 value for a 1.5d star . examining each color index individually for the effects of nlte , it is seen that both choices of ff return lte best fit @xmath1 values that are always hotter than the nlte best fits for a given @xmath0 , and that @xmath60 is approximately constant within one numerical temperature resolution unit for all choices of @xmath0 for each color index . this constant value depends on color index , suggesting that nlte effects are non - constant across the spectrum , but they are temperature independent over the range of @xmath1 values in the 1d grid . for the u@xmath43-b@xmath43 index , the magnitude of @xmath60 is at least 100 k larger than the other indices because of the greater flux in the blue and uv from nlte fe i overionization , forcing the lte seds to have a higher @xmath1 value to match the flux . [ fig : u - b - results ] displays the results for the u@xmath43-b@xmath43 index , showing a nearly constant value for @xmath60 of -150 k to -175 k. the plateau in the lte 1:1 ff @xmath58 relation above @xmath0 = 800 k is not a breaking of this constant @xmath60 , but rather the best fit @xmath1 values were restricted by the upper limit of the 1d grid of models . both choices of ff show an increasing trend in nlte best fitting @xmath1 values for increasing @xmath0 . for values of @xmath59 300 k , both choices of ff produce the same best fitting @xmath1 value for a given @xmath0 value , whereas for values of @xmath61 400 k , the 1:1 ff 1.5d seds produce hotter @xmath1 values than the 2:1 ff seds . the 1:1 ff seds reached a best fitting @xmath1 value of 4450 k and the 2:1 ff seds reached 4425 k at maximum @xmath0 . [ [ nlte-1 ] ] nlte + + + + comparing the lte results with the nlte results for both choices of ff , and for every value of @xmath0 , the lte best fitting @xmath62 values are hotter than the nlte values . because nlte stars produce more blue and uv flux for a given @xmath1 value than lte stars , the lte 1d seds returned higher best fitting @xmath1 values to match the @xmath47 value of the 1.5d sed . for the 1:1 ff 1.5d seds , @xmath60 varied between 25 k and 50 k , whereas for the 2:1 ff seds , @xmath60 was nearly constant at 50 k , only dropping to 25 k for @xmath0 = 900 k. the results of fitting 1d nlte relative flux seds to the 1.5d seds are presented in fig . [ fig : pin - nlte - results ] . both choices of ff show an increasing trend in best fitting @xmath1 values for increasing @xmath0 with the 1:1 ff best fitting @xmath1 values being hotter than the 2:1 best fits at every value of @xmath63 100 k. the 1:1 ff seds reached a best fitting @xmath1 value of 4550 k and the 2:1 ff seds reached 4475 k at maximum @xmath0 . because the logarithmically spaced @xmath36 grid was utilized to place more weight on bluer wavelengths for the fits , the 1.5d sed matches well with the best fitting 1d sed for @xmath64 4500 @xmath37 , but the 1d sed fails to predict the shape of the 1.5d sed around the peak of the spectrum ; here , the 1.5d sed would be fit by a sed with less flux at these wavelengths relative to @xmath48 , suggesting a cooler @xmath1 value for the best fitting 1d sed . [ [ nlte-2 ] ] nlte + + + + the lte best fitting @xmath1 values are hotter than the nlte values for every value of @xmath0 . for both choices of ff , @xmath60 varied between 50 k and 75k . the best fit @xmath1 values below @xmath0 = 500 k are approximately equal at each @xmath0 for both choices of ff , and only begin to differ above @xmath0 = 600 k , with the 1:1 ff 1.5d stars having hotter fitted values . the inequality reaches a maximum @xmath1 value difference of 100 k at the maximum @xmath0 , with the 1:1 ff 1.5d stars reaching a maximum @xmath1 value of 4650 k. [ [ section-21 ] ] fig . [ fig : rectified - comp ] shows the 1.5d 1:1 ff continuum normalized spectrum with @xmath0 = 1000 k plotted with the best fitting 1d nlte spectrum and two bracketing 1d nlte spectra . the best fitting 1d spectrum is seen to be a good fit to the 1.5d spectrum for the blue and red ends of the @xmath36 range , but is a poor match between @xmath65 4500 and 8000 @xmath37 . in this region , the 1.5d spectrum exhibits stronger spectral features than the best fitting 1d spectrum would imply , suggesting that there is a cooler best fitting @xmath1 value for this region . additional fits omitting the poorly matched region were performed and the resultant @xmath1 values were found to differ by less than one numerical temperature resolution unit from the fits performed over the entire range , at maximum @xmath0 . [ [ nlte-3 ] ] nlte + + + + comparing the nlte results with lte results , it is seen that for both ff values , the lte best fit @xmath1 values are hotter than the nlte fits , but @xmath60 increases with increasing @xmath0 . for the 2:1 ff 1.5d stars , @xmath66 reaches a maximum of 175 k , and for the 1:1 ff stars , it is potentially even larger but the lte results plateau above @xmath0 = 800 k , similar to the u@xmath43-b@xmath43 photometric color index in section [ sec : ubvri - photometry ] , and for the same reasons . fig . [ fig : tio - nlte - results ] shows nlte best fitting @xmath1 values inferred from fitting to continuum normalized spectra restricted to the tio bands located between @xmath36 = 5500 and 8000 @xmath37 for all 1.5d stars . a decreasing trend of fitted @xmath1 value with increasing @xmath0 is observed , dropping as low as @xmath1 = 4000 k at maximum @xmath0 . this agrees with the observation from section [ sec : continuum - normalized - spectra ] that the best fitting @xmath1 values for this @xmath36 should be cooler than those found for the same star when fitting the entire visible band . fig [ fig : tio - comp ] shows the tio fitting region of the 1.5d 1:1 @xmath0 = 1000 k with the best fitting 1d spectrum found for this region , as well as the 1d spectrum found for fitting the entire @xmath36 range as a comparison . both of the ff values produce similar results when fit with nlte 1d stars , the fits being separated by no more than one numerical temperature resolution unit at any @xmath0 . the 1:1 ff 1.5d stars have the hotter fitted @xmath1 values at @xmath0 values greater than 600 k. the tio molecular bands grow in strength so rapidly with decreasing @xmath1 that the hot component does not dominate the fit in the tio band region as it does for the overall continuum normalized spectra . the cool component now dominates the shape , with the hot component only mitigating the effect , pulling the fits to lower temperatures with increasing @xmath0 . [ [ nlte-4 ] ] nlte + + + + the lte results are hotter than the respective nlte results at each @xmath0 for both choices of ff , as seen in fig . [ fig : tio - results ] . this is not caused directly by an nlte effect , such as fe i overionization as in the previous four methods , as phoenix does not compute molecules in nlte . instead it is an indirect effect of the difference of lte and nlte @xmath67 structures in the upper atmosphere , and only 1.5d modeling of this style has revealed its role . for a given @xmath1 , nlte models are generally hotter than lte models in the upper atmosphere above @xmath31 = 1 , as seen in fig . [ fig : examplettau ] . because of this , more tio molecules will collisionally dissociate in the upper atmosphere of a nlte star than an lte star , and nlte stars will form weaker absorption features . therefore , nlte best fits will be required to have lower @xmath1 values to match the strength of the tio absorption features . in this case @xmath60 is non constant as a function of @xmath0 , increasing from 50 k at @xmath0 = 0 k , to a local maximum of 100 k , then decreasing back to 75 k at maximum @xmath0 , for both ff . the goal of this work has been to analyze the effects of massively nlte atmospheric modeling combined with 2d horizontal inhomogeneities on @xmath1 values inferred from seds and line profiles . the stellar atmosphere and spectrum synthesis code phoenix was used to generate a grid of spherical stellar atmosphere models in both lte and nlte , and to synthesize spectra for the models . spectra of target 2d `` observed '' stars were produced in the 1.5d approximation by linearly averaging two nlte 1d spectra together under two different weighting schemes ( ff ) , such that the @xmath25 was 4250 k and the temperature difference between the 1d was as large as @xmath0 = 1050 k. the grid of lte and nlte 1d seds and spectra were fit to the observations to infer @xmath1 values for the 1.5d stars using five different approaches . all inferred @xmath1 values and differences between inferred @xmath1 and 1.5d @xmath24 are considered to have a formal uncertainty of 25 k , half of one temperature resolution unit in our pre - interpolated 1d grid . photometric colors of 1d stars computed from synthetic ubvri photometry were compared to 1.5d colors to assess the errors in photometrically derived @xmath1 values . for the five color indices and both values of ff , the inferred value of @xmath1 was seen to increase with @xmath0 , and increased at a greater rate for indices that involved bluer wavebands . when the lte and nlte results were compared , the @xmath1 values inferred from fitting lte colors were systematically higher than their nlte counterparts . the magnitude of @xmath60 was approximately constant as a function of @xmath0 in all cases . the value was largest when comparing u@xmath43-b@xmath43 inferred @xmath1 values , and decreased for redder indices . absolute surface flux 1d seds were fit to the 1.5d seds to assess how changes to the predicted bolometric flux introduced by the modeling assumptions affect the inferred value of @xmath1 . for both values of ff , the inferred value of @xmath1 was seen to increase with @xmath0 . this approach showed the lowest overall error of any of the full @xmath36 distribution fitting approaches in the inferred value of @xmath1 ; only 110 k difference between the fit @xmath1 value and @xmath24 at maximum @xmath0 . again , the inferred @xmath1 values from fitting lte seds were systematically higher than those from fitting nlte seds , and the magnitude of @xmath60 was approximately constant as a function of @xmath0 . relative 1d seds normalized to the average continuum flux in a 10 @xmath37 window in the r - j tail were fit to the 1.5d seds to assess how the modeling assumptions affect overall shape of the sed and the temperature sensitive spectral features located in the blue and near - uv bands . for both values of ff , the inferred value of @xmath1 was seen to increase with @xmath0 . the inferred @xmath1 values from fitting lte seds were systematically higher than those from fitting nlte seds , and the magnitude of @xmath60 was approximately constant as a function of @xmath0 . these three complimentary methods ( photometric colors , absolute surface flux seds , and relative seds ) show results consistent with each other . of the three spectrophotometric methods , the photometric colors give both the highest and the lowest errors on the estimates of the @xmath1 . at maximum @xmath0 , the u@xmath43-b@xmath43 index fitting returned up to 340 k higher than @xmath24 , while the r - i index returned as low as 60 k less than @xmath24 . the v - i fitting resulted in error values similar to that of the absolute seds , with fitted @xmath1 values as high as 90 k above @xmath24 . the relative sed fitting returned error values similar to the v - r index , as high as 210 and 240 k above @xmath24 respectively . together , these results reinforce that red / near - ir photometry is more reliable for diagnosing @xmath1 in stars with significant horizontal inhomogeneity . continuum normalized 1d spectra spanning a wavelength distribution between @xmath36 = 3000 and 13000 @xmath37 were fit to the 1.5d spectra to assess how the modeling assumptions change the predicted strength of spectral features . for both values of ff , the inferred value of @xmath1 was seen to increase with @xmath0 . it is important to note that this result is consistent with the spectrophotometric results , even though it is arrived at through a complimentary method . this approach showed the highest overall error of any of the full spectrum fitting approaches in the inferred value of @xmath1 ; 310 k above @xmath24 at maximum @xmath0 . the inferred @xmath1 values from fitting lte spectra were systematically higher than those from fitting nlte spectra , and the magnitude of @xmath60 was approximately constant as a function of @xmath0 . predicted line profiles for the important 1d tio bands spanning a wavelength range between @xmath36 = 5500 and 8000 @xmath37 were fit to the 1.5d tio bands to assess how the strength of molecular features found primarily in the cold component of the 1.5d stars affect the inferred value of @xmath1 . this was the only approach to show the inferred value of @xmath1 decreasing with increasing @xmath0 . the rapid nonlinear growth of molecular features with decreasing temperature became the dominant aspect in determining the @xmath1 value , over the nonlinear contribution to the average flux from the higher temperature 1.5d component . the inferred @xmath1 values from fitting lte spectra were still systematically higher than those from fitting nlte spectra , and the magnitude of @xmath60 was approximately constant as a function of @xmath0 . in this work we have shown that the approximations of both horizontal homogeneity and lte introduce errors in the value of @xmath1 inferred from fitting quantities derived from models to observed quantities . by assuming both horizontal homogeneity and lte , the inferred @xmath1 values may differ from the @xmath24 of a star by 340 k or more , depending on the quantity used to infer the @xmath1 . of the two values of ff , 1:1 produced hotter values of inferred @xmath1 in general . in simulating horizontal inhomogeneities it was seen that the bolometric flux of a 1.5d star increased with @xmath0 , and a percentage of the flux was redistributed at bluer wavelengths . spectral features in general appeared to have been produced by a star hotter than the @xmath25 , except that strong molecular features found in cooler stars were also present in the spectra . furthermore , for all five approaches , @xmath60 is approximately independent of @xmath0 , and we conclude that the magnitude of the effect of nlte on @xmath1 derived from fitting 1d models is approximately independent of the thermal contrast characterizing the degree of the horizontal inhomogeneity of the star being modeled , and only dependent on the observable quantity being fit . while @xmath0 was the only parameter varied in the scope of this study , other modeling parameters such as log @xmath7 and [ fe / h ] are expected to have an impact on @xmath1 @xmath24 and @xmath1 @xmath25 derived in various ways for horizontally inhomogeneous stars . both @xcite and @xcite have shown the rms temperature variations to increase with decreasing log @xmath7 , and @xcite has also shown them to increase with decreasing [ fe / h ] . likewise , the values of the fitted @xmath1 @xmath24 at a given @xmath0 are expected to differ with the @xmath25 of the 1.5d stars . investigating these parameters requires extensive additions to our 1d grid of models , as well as an updated fe i atomic model , and will be explored in a future study . we would like to thank the nserc discovery grant program and saint mary s university s faculty of graduate studies and research for funding this work . we would also like to acknowledge comupte canada member acenet for providing us with all computational resources and cpu time . alves - brito , a. , melendez , j. asplund , m. , _ et al . _ , 2010 , , 513 , a35 asplund , m. , 2005 , , 43 , 481 asplund , m. , grevesse , n. , sauval , a. j. , _ et al . _ , 2009 , , 47 , 481 baron , e. & hauschildt , p. h. , 1998 , , 495 , 370 bergemann , m. , kudritzki , r. p. , davies , b. , _ et al . _ , 2013 , _ eas publications series _ , 60 , 103 bessell , m. s. , 1990 , , 102 , 1181 brott , i. & hauschildt , p. h. , 2005 , _ esa special publication _ , 576 , 565 burnashev v.i . , 1985 , abastumanskaya astrofiz . obs . bull . 59 , 83 caffau , e. , ludwig , h .- , steffen , m. , _ et al . _ , 2011 , , 268 , 255 carretta , e. , bragaglia , a. , gratton , r. g. , _ et al . _ , 2010 , , 516 , a55 carretta , e. & gratton , r. g. , 1997 , , 121 , 95 chiavassa , a. , collet , r. , casagrande , l. , _ et al . _ , 2010 , , 524 , a93 collet , r. , asplund , m. , & trampedach , r. , 2007 , , 469 , 687 collet , r. , asplund , m. , & trampedach , r. , 2008 , , 79 , 649 collet , r. , nordlund , . , asplund , m. , _ et al . _ , 2009 , , 80 , 719 cox , a. n. , 2000 , allen s astrophysical quantities ( 4th ed . ; new york , ny : springer ) cunha , k. & smith , v. v. , 2006 , , 651 , 491 davies , b. , kudritzki , r .- p . , plez , b. , _ et al . _ , 2013 , , 767 , 3 dobrovolskas , v. , kuinskas , a. , steffen , m. , _ et al . _ , 2013 , , 559 , a102 hauschildt , p. h. , baron , e. , & allard , f. , 1997 , , 483 , 390 hayek , w. , asplund , m. , collet , r. , _ et al . _ , 2011 , , 529 , a158 kuinskas , a. , ludwig , h .- g . , steffen , m. , _ et al . _ , 2013a , memorie della societa astronomica italiana supplementi , 24 , 68 kuinskas , a. , steffen , m. , ludwig , h .- et al . _ , 2013b , , 549 , a14 lee , m. g. , freedman , w. l. , & madore , b. f. , 1993 , , 417 , 553 ludwig , h .- & kuinskas , a. , 2012 , , 547 , a118 magic , z. , collet , r. , asplund , m. , _ et al . _ , 2013a , , 557 , a26 magic , z. , collet , r. , hayek , w. , _ et al . _ , 2013b , , 560 , a8 makarov , d. , makarova , l. , rizzi , l. , _ et al . _ , 2006 , , 132 , 2729 martins , f. , schaerer , d. , & hillier , d. j. , 2005 , , 436 , 1049 mashonkina , l. , gehren , t. , shi , j .- r . , _ et al . _ , 2011 , , 528 , a87 mashonkina , l. , ludwig , h .- g . , korn , a. , _ et al . _ , 2013 , memorie della societa astronomica italiana supplementi , 24 , 120 mathur , s. , hekker , s. , trampedach , r. , _ et al . _ , 2011 , , 741 , 119 melndez , j. , asplund , m. , alves - brito , a. , _ et al . _ , 2008 , , 484 , l21 pilachowski , c. a. , sneden , c. , & wallerstein , g. , 1983 , , 52 , 241 ramrez , i. & melndez , j. , 2005 , , 626 , 446 rizzi , l. , tully , r. b. , makarov , d. , _ et al . _ , 2007 , , 661 , 815 ross , j. e. & aller , l. h. , 1976 , science , 191 , 1223 salaris , m. & cassisi , s. , 1998 , , 298 , 166 samadi , r. , belkacem , k. , ludwig , h .- _ , 2013 , , 559 , a40 sheminova , v. a. , 2012 , , 280 , 83 short , c. i. , campbell , e. a. , pickup , h. , _ et al . _ , 2012 , , 747 , 143 short , c. i. & hauschildt , p. h. , 2003 , , 596 , 501 short , c. i. & hauschildt , p. h. , 2005 , , 618 , 926 short , c. i. & hauschildt , p. h. , 2006 , , 641 , 494 short , c. i. & hauschildt , p. h. , 2009 , , 691 , 1634 stanek , k. z. , udalski , a. , szymanski , m. , _ et al . _ , 1997 , , 477 , 163 tremblay , p .- e . , ludwig , h .- g . , freytag , b. , _ et al . _ , 2013 , , 557 , a7 uitenbroek , h. & criscuoli , s. , 2011 , , 736 , 69 h & 80/3160 & + he & 19/37 & + li & 57/333 & 55/124 + c & 228/1387 & + n & 252/2313 & + o & 36/66 & + ne & 26/37 & + na & 53/142 & 35/171 + mg & 273/835 & 72/340 + al & 111/250 & 188/1674 + si & 329/1871 & 93/436 + p & 229/903 & 89/760 + s & 146/349 & 84/444 + k & 73/210 & 22/66 + ca & 194/1029 & 87/455 + ti & 395/5279 & 204/2399 + mn & 316/3096 & 546/7767 + fe & 494/6903 & 617/13675 + co & 316/4428 & 255/2725 + ni & 153/1690 & 429/7445 + [ tab : nlte species list ] 3000 - 4000 & 0.010 & 350000 + 4000 - 5000 & 0.013 & 346000 + 5000 - 6000 & 0.016 & 344000 + 7000 - 8000 & 0.023 & 326000 + 8000 - 11000 & 0.027 & 352000 + 11000 - 13000 & 0.037 & 324000 + [ tab : wavelength - grid - spacing ] 1050 & & & & 4600 & 3550 & 4330 + 1000 & 4750 & 3750 & 4336 & & & + 900 & 4700 & 3800 & 4320 & 4550 & 3650 & 4309 + 800 & 4650 & 3850 & 4305 & & & + 750 & & & & 4500 & 3750 & 4292 + 700 & 4600 & 3900 & 4293 & & & + 600 & 4550 & 3950 & 4281 & 4450 & 3850 & 4277 + 500 & 4500 & 4000 & 4272 & & & + 450 & & & & 4400 & 3950 & 4265 + 400 & 4450 & 4050 & 4264 & & & + 300 & 4400 & 4100 & 4258 & 4350 & 4050 & 4257 + 200 & 4350 & 4150 & 4254 & & & + 150 & & & & 4300 & 4150 & 4252 + 100 & 4300 & 4200 & 4251 & & & + 0 & 4250 & 4250 & 4250 & 4250 & 4250 & 4250 + 1000 & 4675 & 4650 & 4575 & 4425 & 4300 & 4450 & 4550 & 4650 & 4025 + 900 & 4625 & 4575 & 4500 & 4400 & 4275 & 4425 & 4500 & 4575 & 4025 + 800 & 4550 & 4525 & 4450 & 4375 & 4275 & 4400 & 4450 & 4525 & 4050 + 700 & 4500 & 4450 & 4400 & 4350 & 4275 & 4375 & 4425 & 4475 & 4075 + 600 & 4450 & 4400 & 4375 & 4325 & 4275 & 4350 & 4375 & 4425 & 4100 + 500 & 4375 & 4350 & 4325 & 4300 & 4275 & 4325 & 4350 & 4375 & 4125 + 400 & 4350 & 4325 & 4300 & 4275 & 4250 & 4300 & 4325 & 4350 & 4175 + 300 & 4300 & 4300 & 4275 & 4275 & 4250 & 4275 & 4300 & 4300 & 4200 + 200 & 4275 & 4275 & 4275 & 4250 & 4250 & 4275 & 4275 & 4275 & 4225 + 100 & 4250 & 4250 & 4250 & 4250 & 4250 & 4250 & 4250 & 4250 & 4250 + 0 & 4250 & 4250 & 4250 & 4250 & 4250 & 4250 & 4250 & 4250 & 4250 + 1050 & 4575 & 4575 & 4550 & 4425 & 4300 & 4425 & 4475 & 4550 & 4000 + 900 & 4525 & 4525 & 4475 & 4375 & 4250 & 4400 & 4450 & 4500 & 4025 + 750 & 4475 & 4450 & 4425 & 4325 & 4250 & 4350 & 4400 & 4450 & 4050 + 600 & 4400 & 4375 & 4350 & 4300 & 4250 & 4325 & 4350 & 4400 & 4075 + 450 & 4350 & 4325 & 4300 & 4275 & 4250 & 4300 & 4325 & 4350 & 4125 + 300 & 4300 & 4275 & 4275 & 4275 & 4250 & 4275 & 4275 & 4300 & 4200 + 150 & 4250 & 4250 & 4250 & 4250 & 4250 & 4250 & 4250 & 4275 & 4250 + 0 & 4250 & 4250 & 4250 & 4250 & 4250 & 4250 & 4250 & 4250 & 4250 + 1000 & 4600 & 4475 & 4325 & 4100 + 900 & 4525 & 4425 & 4325 & 4100 + 800 & 4475 & 4400 & 4300 & 4125 + 700 & 4425 & 4375 & 4300 & 4175 + 600 & 4400 & 4350 & 4300 & 4200 + 500 & 4350 & 4325 & 4300 & 4225 + 400 & 4325 & 4300 & 4300 & 4250 + 300 & 4300 & 4300 & 4300 & 4275 + 200 & 4275 & 4300 & 4300 & 4300 + 100 & 4275 & 4275 & 4300 & 4300 + 0 & 4275 & 4275 & 4275 & 4300 + 1050 & 4575 & 4450 & 4325 & 4075 + 900 & 4500 & 4400 & 4300 & 4100 + 750 & 4450 & 4375 & 4275 & 4125 + 600 & 4375 & 4325 & 4275 & 4175 + 450 & 4325 & 4300 & 4300 & 4225 + 300 & 4300 & 4300 & 4300 & 4275 + 150 & 4275 & 4275 & 4300 & 4300 + 0 & 4275 & 4275 & 4275 & 4300 + 1000 & 339 & 314 & 239 & 89 & -36 & 114 & 214 & 314 & -311 + 900 & 305 & 255 & 180 & 80 & -45 & 105 & 180 & 255 & -295 + 800 & 245 & 220 & 145 & 70 & -30 & 95 & 145 & 220 & -255 + 700 & 207 & 157 & 107 & 57 & -18 & 82 & 132 & 182 & -218 + 600 & 169 & 119 & 94 & 44 & -6 & 69 & 94 & 144 & -181 + 500 & 103 & 78 & 53 & 28 & 3 & 53 & 78 & 103 & -147 + 400 & 86 & 61 & 36 & 11 & -14 & 36 & 61 & 86 & -89 + 300 & 42 & 42 & 17 & 17 & -8 & 17 & 42 & 42 & -58 + 200 & 21 & 21 & 21 & -4 & -4 & 21 & 21 & 21 & -29 + 100 & -1 & -1 & -1 & -1 & -1 & -1 & -1 & -1 & -1 + 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 + 1050 & 245 & 245 & 220 & 95 & -30 & 95 & 145 & 220 & -330 + 900 & 216 & 216 & 166 & 66 & -59 & 91 & 141 & 191 & -284 + 750 & 183 & 158 & 133 & 33 & -42 & 58 & 108 & 158 & -242 + 600 & 123 & 98 & 73 & 23 & -27 & 48 & 73 & 123 & -202 + 450 & 85 & 60 & 35 & 10 & -15 & 35 & 60 & 85 & -140 + 300 & 43 & 18 & 18 & 18 & -7 & 18 & 18 & 43 & -57 + 150 & -2 & -2 & -2 & -2 & -2 & -2 & -2 & 23 & -2 + 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 + 1000 & 264 & 139 & -11 & -236 + 900 & 205 & 105 & 5 & -220 + 800 & 170 & 95 & -5 & -180 + 700 & 132 & 82 & 7 & -118 + 600 & 119 & 69 & 19 & -81 + 500 & 78 & 53 & 28 & -47 + 400 & 61 & 36 & 36 & -14 + 300 & 42 & 42 & 42 & 17 + 200 & 21 & 46 & 46 & 46 + 100 & 24 & 24 & 49 & 49 + 0 & 25 & 25 & 25 & 50 + 1050 & 245 & 120 & -5 & -255 + 900 & 191 & 91 & -9 & -209 + 750 & 158 & 83 & -17 & -167 + 600 & 98 & 48 & -2 & -102 + 450 & 60 & 35 & 35 & -40 + 300 & 43 & 43 & 43 & 18 + 150 & 23 & 23 & 48 & 48 + 0 & 25 & 25 & 25 & 50 +

the seyfert 1 galaxy zw 229 - 015 is among the brightest active galaxies being monitored by the _ kepler _ mission . in order to determine the black hole mass in zw 229 - 015 from h reverberation mapping , we have carried out nightly observations with the kast spectrograph at the lick 3 m telescope during the dark runs from june through december 2010 , obtaining 54 spectroscopic observations in total . we have also obtained nightly-band imaging with the katzman automatic imaging telescope at lick observatory and with the 0.9 m telescope at the brigham young university west mountain observatory over the same period . we detect strong variability in the source , which exhibited more than a factor of 2 change in broad h flux . from cross - correlation measurements , we find that the h light curve has a rest - frame lag of days with respect to the-band continuum variations . we also measure reverberation lags for h and h and find an upper limit to the h lag . combining the h lag measurement with a broad h width of km s measured from the root - mean - square variability spectrum , we obtain a virial estimate of for the black hole in zw 229 - 015 . as a _ kepler _ target , zw 229 - 015 will eventually have one of the highest - quality optical light curves ever measured for any active galaxy , and the black hole mass determined from reverberation mapping will serve as a benchmark for testing relationships between black hole mass and continuum variability characteristics in active galactic nuclei .

1103.2153

the nasa _ kepler _ mission @xcite is designed to search for exoplanets by detecting transit events , and early results have demonstrated the spectacular quality of _ kepler _ s ultra - precise photometry ( e.g. , * ? ? ? in addition to its primary goal of obtaining light curves of galactic stars , _ kepler _ is monitoring a small number of active galactic nuclei ( agns ) of various types , including seyfert galaxies , quasars , and blazars . _ kepler_will provide the best optical light curves ever obtained of agns , vastly exceeding the quality of any previous observations in terms of both sampling cadence and photometric accuracy . this will enable unprecedented measurements of the characteristic properties of agn optical variability . one of the key motivations for measurement of high - precision agn light curves is to determine the power spectrum of continuum fluctuations and to search for any characteristic timescales of variability ( in the form of breaks or slope changes in the power spectrum of variations ) , which might be related to light - crossing , dynamical , or thermal timescales in the accretion disk . at x - ray energies , long - duration light curves have been used to search for breaks in the power spectral density ( e.g. , * ? ? ? * ; * ? ? ? * ) , and the break timescales are observed to be dependent on both black hole mass and luminosity across a range extending from x - ray binaries to quasars @xcite . there is accumulating evidence that the optical fluctuations of agns also contain characteristic timescales that may depend on black hole mass and luminosity @xcite . ground - based optical light curves of agns are limited in quality by nightly and seasonal sampling , weather , and seeing - dependent errors , but _ kepler _ can provide exquisite data that will for the first time enable detailed measurements of the optical variability properties of individual objects over timescales ranging from hours up to ( potentially ) a few years . unobscured ( type 1 ) agns monitored by _ kepler _ can provide powerful tests of proposed relationships between variability properties , black hole mass , and luminosity that have been derived from larger samples of objects with lower - quality ground - based light curves . in order to fully exploit the results of these _ kepler _ agn observations , independent measurements of black hole masses in _ kepler_-field agns are needed . here , we present the results of a six - month , ground - based , optical reverberation - mapping program for the _ kepler _ target zw 229 - 015 ( at @xmath10 ; * ? ? ? * ) , with the goal of determining the black hole mass in this galaxy from its h@xmath0 reverberation lag . this galaxy ( also known as cgcg 229 - 015 or 2masx j19052592 + 4227398 ) was first identified as a type 1 seyfert by @xcite . since then , no follow - up observations of this agn have been reported in the literature at all , aside from its inclusion in _ rosat _ all - sky survey source catalogs @xcite . nevertheless , zw 229 - 015 is one of the nearest and brightest agns in the _ kepler _ field , and it was selected as a monitoring target for the first two cycles of _ kepler _ observations . to assess its suitability for ground - based monitoring , we obtained a test spectroscopic exposure at the lick observatory 3 m shane telescope in april 2010 . this observation reconfirmed the seyfert 1 classification , and we then selected it for the reverberation mapping program reported here . in this paper , we describe our imaging and spectroscopic observations of zw 229 - 015 , the measurements of the continuum and emission - line light curves and the broad emission - line reverberation lag , and an estimate of the black hole mass based on the h@xmath0 lag and line width . reverberation measurements require a light curve of the agn continuum , and we obtained @xmath1-band photometry of zw 229 - 015 following the same strategy used in recent reverberation programs ( e.g. , * ? ? ? * ; * ? ? ? although ground - based measurements can not approach the accuracy or sampling cadence of the _ kepler _ data , they allowed us to monitor the target s variability continuously during the course of our observing program . furthermore , the @xmath1-band photometry gives a nearly line - free measurement of the galaxy s continuum flux , while the broad _ kepler _ spectral response includes the h@xmath3 and h@xmath0 lines from the agn . we obtained regularly scheduled @xmath1-band imaging at the 0.76 m katzman automatic imaging telescope ( kait ) at lick observatory , and at the 0.9 m telescope at the brigham young university west mountain observatory ( wmo ) . at both telescopes , zw 229 - 015 was observed each night during the monitoring period , except when precluded by weather or instrument problems . we also obtained a small number of images at other facilities , as described below . figure [ wmoimage ] illustrates the field surrounding zw 229 - 015 . the galaxy has a barred spiral morphology , with a hubble type of roughly sba . the kait 0.76 m robotic telescope @xcite uses a @xmath11 pixel site ccd camera with a scale of 08 pixel@xmath7 . a single 300 s exposure of zw 229 - 015 was obtained on each night , weather permitting , from june 3 through december 13 ( ut dates are used throughout this paper ) . during the monitoring period , kait had two month - long shutdowns for repairs , first in august and again from mid - october to mid - november . the median seeing for all kait exposures was 215 . the automatic kait data - processing pipeline includes bias subtraction and flatfielding . the 0.9 m telescope at the west mountain observatory is one of five small research telescopes operated by brigham young university . it was installed and tested in late 2009 and began working in a partially queue - scheduled mode during the 2010 observing season . the observations secured for this investigation were scheduled as part of an external guest investigator program . from june 5 until september 1 , observations were made using a @xmath12 pixel kaf 09000 ccd camera with a scale of 05 pixel@xmath7 . after september 2 , observations were made using a @xmath13 pixel fairchild 3041-uv ccd camera with a scale of 06 pixel@xmath7 . the typical observing procedure was to secure two exposures of zw 229 - 015 while it was close to meridian transit on each night that conditions permitted telescope operations . exposures were 300 s from the start of monitoring until september 7 , and were decreased to 200 s thereafter because of the higher sensitivity of the fairchild ccd . observations continued until december 13 , but only a few exposures were obtained after november 5 due to poor weather . standard - star fields from @xcite were observed on a few photometric nights . the median seeing for the wmo exposures was 189 . the standard processing pipeline for wmo data includes bias subtraction and flat - fielding . the lick 1 m nickel reflector was used to observe zw 229 - 015 during a few nights while kait was shut down for repairs . the nickel direct imaging camera contains a thinned , loral @xmath13 pixel ccd with a scale of 0184 pixel@xmath7 , and the observations were obtained using @xmath14 binning for the ccd readout . on one photometric night , @xcite standard - star fields were also observed . overscan subtraction and flat - fielding using twilight sky flats were done following standard procedures in iraf . zw 229 - 015 was observed on five nights in august and september using the seoul national university 4096 @xmath15 4096 pixel camera ( snucam ; * ? ? ? * ) and @xmath1-band filter at the maidanak observatory 1.5 m telescope in uzbekistan ( fairchild ccd ; 0266 pixel@xmath7 ) . the processing pipeline for snucam data included overscan subtraction and flat - fielding . lccc 1 & 19:05:19.49 & 42:28:57.7 & @xmath16 + 2 & 19:05:35.95 & 42:27:16.2 & @xmath17 + 3 & 19:05:33.71 & 42:28:50.5 & @xmath18 + 4 & 19:05:19.98 & 42:27:37.9 & @xmath19 + 5 & 19:05:27.28 & 42:26:48.8 & @xmath20 + 6 & 19:05:30.05 & 42:26:21.4 & @xmath21 [ compstarstable ] our light - curve measurements followed the same procedures used by @xcite . first , each image was cleaned of cosmic - ray hits using the la - cosmic procedure @xcite implemented in iraf . aperture photometry on zw 229 - 015 and several comparison stars was performed using the iraf task ` phot ` . ultimately , six comparison stars were selected ; these are illustrated in figure [ wmoimage ] . the photometric measurements were carried out with a range of aperture sizes . we found that the 4 aperture radius gave the cleanest light curves , and our final photometry is based on this aperture size . the background sky annulus had inner and outer radii of 8 and 144 . when more than one exposure was taken per night at a given telescope , the measurements were averaged together to give a single data point for that night . photometric calibration was accomplished by carrying out aperture photometry on the @xcite standard stars observed at wmo and the nickel telescope on nights that were judged to be photometric by the observers . we used only @xmath1-band exposures and did not attempt to derive any color - dependent terms for the transformation between instrumental and calibrated @xmath1 magnitudes . we found excellent agreement between the photometric calibrations for the night of september 26 at wmo and the night of august 12 at the nickel telescope , and took the average of the two calibrations to obtain our final photometric scale . table [ compstarstable ] lists the positions and magnitudes of the six comparison stars . the light curves from each telescope were separately normalized to @xmath1-band ( vega - system ) magnitudes based on the comparison - star photometry . to check for small systematic differences between the photometric scales of the four telescopes , which could be present because color terms were not measured , we determined the average offset between the kait and wmo light curves for the 53 nights when the object was observed at both telescopes . the weighted average offset was @xmath22 mag , and we applied this shift of 0.01 mag to the kait light curve as an empirical correction to bring it into agreement with the wmo photometric scale . similar offsets of 0.010 and 0.020 mag were applied to the nickel and maidanak data points , respectively , in order to bring them into agreement with the combined kait and wmo light curve , based on the weighted average shift determined from nights when the galaxy was observed at more than one telescope . lcc 5351.9785 & k & @xmath23 + 5352.8036 & w & @xmath24 + 5352.9688 & k & @xmath25 + 5353.9834 & k & @xmath26 + 5354.9619 & k & @xmath27 + [ phottable ] the final light - curve data for zw 229 - 015 are listed in table [ phottable ] . some observations taken in cloudy conditions had very low signal - to - noise ratio ( s / n ) , and we excluded any measurement having a formal uncertainty greater than 0.03 mag . we also discarded measurements from a few images that suffered from very poor flat - fielding or other problems . the resulting light curve is illustrated in figure [ vlightcurve ] . the photometric uncertainties listed in table [ phottable ] are the values returned by the photometry routine , which depend on photon - counting statistics for the object and background as well as on readout noise . these are likely to be underestimates of the true uncertainties , since additional errors can result from flat - fielding irregularities , point - spread function variations , and other issues . one way to estimate the actual photometric uncertainties is to examine the magnitude fluctuations in the comparison - star light curves , which are presumed to be intrinsically nonvariable . comparison stars 3 , 4 , and 5 have @xmath1 magnitudes very similar to that of zw 229 - 015 . for each of these stars we examined the scatter of measurements about the mean magnitude , and found standard deviations of 0.0110.016 mag ( see table [ compstarstable ] ) , while the median value of the formal photometric error on a single measurement for these stars is 0.006 mag . this discrepancy shows that the photon - counting statistics are usually not the dominant contributor to the error budget , except when the images have very low s / n . we address this issue by adding an additional error term in quadrature to the photometric uncertainties prior to carrying out the cross - correlation measurements , as described in [ reverbsection ] . our photometric aperture includes a substantial portion of the host galaxy , including essentially the entire bulge and much of the bar and disk . we do not attempt to subtract off this host - galaxy starlight contribution to the light - curve points , because our ground - based imaging does not have sufficiently high resolution to decompose the agn point source and bulge into separate components . removing this constant contribution of host - galaxy starlight would not affect the measurement of the reverberation lag . a condensed version of the @xmath1-band light curve was prepared for the cross - correlation measurements by taking a weighted average of any photometric data taken within 5 hours of each other . this produces a light curve with at most one data point per night . the condensed light curve was then converted from magnitudes to linear flux units . the spectroscopic monitoring program at the lick 3 m shane telescope was carried out by several independent observing teams . unlike other recent reverberation datasets that were obtained via long - duration , dedicated campaigns of classically scheduled observing nights , we chose to obtain synoptic monitoring of this single target by enlisting the entire community of observers using the kast spectrograph at lick during summer and fall 2010 . the kast spectrograph is mounted at the cassegrain focus of the 3 m telescope for about half of each lunation , and the spectrograph is used for a variety of programs including supernova and transient follow - up observations , stellar spectroscopy , and agn surveys . during each kast observing night , the regularly scheduled observers took a 1200 s exposure of zw 229 - 015 to contribute to this monitoring program . this strategy enabled us to collect a spectroscopic dataset with almost nightly sampling during each dark run , except for weather - related losses . the spectroscopic monitoring began with one observation in may , and then continued with nightly observations during each dark run ( weather permitting ) from june through mid - december . from june through september , nearly every night was clear enough to observe , while october through december had more substantial weather losses , as is typical for mt . hamilton ( the site of lick observatory ) . table [ spectable ] gives a log of the kast observations for the 54 nights when useful data were obtained . we have omitted data from two nights when zw 229 - 015 was observed through very thick clouds , resulting in a s / n that was too low to be useful for any measurements . each observing team used a different instrumental setup for the spectrograph , designed for their primary science programs . on the blue side of the spectrograph , all observers used the 600-line grism , which gives a dispersion close to 1.0 pixel@xmath7 . the spectral range falling on the blue - side detector is controlled by adjusting the ccd position via a manually movable stage , resulting in a fixed wavelength coverage for the entire night s observations . most observing teams set the ccd to a position that covered @xmath285520 , placing the h@xmath0 and [ ] @xmath294959 , 5007 lines on the blue ccd . however , on several nights a bluer setup covering @xmath305150 or @xmath315135 was used , and with this setup the [ ] @xmath325007 line fell off the edge of the blue ccd , while the @xmath324959 line was still present . for the red side of the spectrograph , several gratings are available and different gratings were installed on different nights . since we could not ensure the same degree of consistency in the red - side observations as was obtained on the blue side , the red - side observations were a lower priority for this project . on some nights zw 299 - 015 was not observed at all on the red side . when it was observed , gratings of 300 , 600 , and 1200 lines mm@xmath7 were used for the red - side observations , with dispersions ranging from 1.2 to 4.6 pixel@xmath7 . in most cases , the observers used a d55 dichroic to separate the blue and red beams of the spectrograph , while on a few nights a mirror was used that directed all of the light to the blue side . when the 300 or 600 lines mm@xmath7 gratings were used , there was usually a small region of wavelength overlap between the blue and red spectra . as a result of the heterogeneous nature of the red - side data , this paper focuses primarily on the analysis of the blue - side observations , which were all taken with the same grism and homogeneously observed except for the variations in spectral range described above . with a single exception ( the night of 2010 august 16 ) , all observations of zw 229 - 015 were obtained with the spectrograph slit at a position angle of pa = 565 . this angle was chosen so as to place a bright foreground star ( marked with a square in figure [ wmoimage ] ) in the slit , in order to aid in centering the primary target . a 4 slit width was used for all exposures of zw 229 - 015 and for the standard star bd+284211 . a 1200 s exposure of zw 229 - 015 and a 60 s exposure of bd+284211 were taken on most nights . on a few nights , the observers adjusted the exposure time due to weather constraints or took more than one exposure of the agn . the airmass was smaller than 1.2 for 72% of the spectroscopic observations of zw 229 - 015 , and smaller than 1.5 for all but the final three exposures , when the object was observed at airmasses between 1.5 and 2.0 . at the end of the observing season in december , when the target was observed at relatively high airmass , the fixed slit pa of 565 was within 1020 of the optimal parallactic angle @xcite for the exposures , resulting in minimal differential light losses . table [ spectable ] also lists the `` image quality , '' measured as the full width at half - maximum intensity ( fwhm ) of the spatial profile of the galactic foreground star in the spectroscopic exposure . this image quality is primarily a measure of atmospheric seeing , but on some nights poor focus or occasional guiding errors added significantly to the spatial profile width . on clear nights , the blue - side observations typically resulted in a s / n per pixel of @xmath3345 in the extracted spectra at 4700 . dome - flat exposures and observations of arc - line lamps were taken during each afternoon . dome flats were usually observed through a 2 slit , since that slit width was used for the observers primary programs . the difference in slit width between the agn exposures and dome - flat exposures did not have any significant impact on the quality of the blue - side flattening , but on the red side it left strong fringing residuals at wavelengths longer than @xmath34 . reduction of the kast data followed standard procedures , including overscan subtraction , flat - fielding , unweighted ( nonoptimal ) extractions , and wavelength calibration using the line - lamp exposures . an extraction width of 65 was used . flux calibration and a linear shift to the wavelength scale based on the wavelengths of night - sky emission lines were then applied to the spectra . the final blue - side spectra were binned to a uniform linear scale of 1 pixel@xmath7 . cosmic - ray residuals were removed from the reduced spectra by interpolation over the affected pixels . for nights when more than one exposure was taken , the multiple exposures were combined into a single reduced spectrum . error spectra were also extracted and propagated through the full sequence of reductions . figure [ zw229spectrum ] illustrates the reduced spectrum from a single observation . we measured the centroid wavelength of the [ ] @xmath325007 line in each spectrum where the line was available . the median result is @xmath35 , corresponding to an emission - line redshift of @xmath36 , consistent with the value from @xcite . the fairly large dispersion in the [ ] wavelength measurements is most likely a result of variations in the centering of the agn in the wide 4 slit in each observation . in order to intercalibrate the flux scales of the spectra and align them to a consistent wavelength scale , we employed the scaling procedure described by @xcite , which is based on the assumption that the narrow [ ] emission - line flux is intrinsically constant over the timescale of the monitoring program . this method applies small wavelength shifts and a multiplicative flux scaling to each spectrum , as well as a convolution with a gaussian broadening kernel , and finds the values of the shift , kernel width , and flux scaling factor that minimize the residual differences between the individual spectrum and a reference spectrum over a small wavelength region containing a narrow emission line and some surrounding continuum . according to @xcite , their scaling method can yield accuracies of 5% or better , but higher accuracy ( @xmath372% ) can be obtained with the best - quality datasets ( e.g. , * ? ? ? * ; * ? ? ? the flux scale of the reference spectrum was normalized by scaling it to match the mean [ ] @xmath325007 flux measured from eight nights that were considered to be photometric by the observers . for those eight nights , we find a mean observed [ ] @xmath325007 flux of @xmath38 erg @xmath39 s@xmath7 , with a standard deviation of 4% , as measured from the flux - calibrated spectra prior to applying the scaling procedure . when applying the @xcite procedure to the individual spectra , we used the [ ] @xmath324959 emission line as the flux scaling reference line , instead of the @xmath325007 line . the reason for this modification was that 10 of the spectroscopic observations used the blue - side ccd position that cut off the @xmath325007 emission line . the wavelength range in common to all of the blue - side exposures extended to 5130 , including the @xmath324959 emission line and about 20 redward of the line . by using the @xmath324959 line as the flux reference , all of the spectra could be treated identically in the flux scaling procedure . to examine the accuracy of the spectral scaling , we measured the flux in the @xmath324959 emission line from each spectrum after the scaling was applied , along with its uncertainty from the propagated error spectrum . we then determined the normalized excess variance @xmath40 of the @xmath324959 light curve , following the definition used by @xcite : @xmath41,\ ] ] where @xmath42 is the total number of observations , @xmath43 is the mean flux , and @xmath44 and @xmath45 are the individual flux values and their uncertainties . we find @xmath46 , indicating that the residual errors in flux scaling contribute a spurious scatter to the light curves at the @xmath47% level . these residuals are primarily due to seeing variations ( particularly from the worst - seeing nights ) , miscentering of the target in the slit , and night - to - night variations in the telescope and spectrograph focus . since zw 229 - 015 exhibited dramatic , high - amplitude variability during the monitoring period , this residual scatter only has a small effect on the h@xmath0 light curve . figure [ figmean ] illustrates the mean and root - mean - square ( rms ) spectra based on the full dataset , constructed after applying the spectral scaling procedure to the data . the rms spectrum essentially gives the standard deviation of flux values at a given wavelength , and illustrates the relative degree of variability of different wavelength regions of the spectrum . the broad h@xmath0 , h@xmath4 , and h@xmath5lines clearly stand out in the rms spectrum , and the variable continuum rises strongly toward the blue end of the observed wavelength range . in addition , a broad @xmath324686 line is evident in the rms spectrum , although it is too weak to be visible in the mean spectrum , where only the narrow component is clearly detected . the flux scaling of the red - side data is more difficult to establish accurately . on the red side , there are no strong and isolated narrow emission lines , so the @xcite scaling routine can not be applied directly . additionally , on some nights the standard star was not observed simultaneously on the blue and red sides , resulting in flux scaling offsets when conditions were not photometric or the seeing was poor . in order to place the red - side data on a flux scale that was approximately consistent with the blue - side data , we first scaled each red spectrum by the scale factor used for the corresponding blue spectrum . this worked well for most nights , but left obvious offsets between the blue - side and red - side flux scales for a few nights . for those nights , an additional scaling factor was applied to the red - side data in order to bring the spectrum into agreement with the corresponding blue - side spectrum over the region of wavelength overlap or near - overlap . ( we note that in a dedicated monitoring campaign with homogeneous observations , these flux scaling problems can be mitigated by using a spectrograph setup with substantial wavelength overlap between the red and blue sides , and by observing standard stars simultaneously on both arms of the spectrograph . ) the h@xmath0 light curve was measured by first subtracting a local , linear continuum underlying the line . the continuum fitting windows were set to 48604910 and 51105130 . the h@xmath0 flux was then measured by direct integration of the spectrum over 49305050 . we similarly measured the light curves of h@xmath4and h@xmath5 . the emission - line summation windows for h@xmath4 and h@xmath5 were 44204500 and 41754245 , respectively , while the continuum windows were 43004400 and 45504600 for h@xmath4 , and 41004150 and 42754325 for h@xmath5 . on the red side , the emission - line window for h@xmath3 was 66006825 , and the continuum windows were 65006560 and 69407020 . all of these emission - line measurements include contributions from the narrow components of the balmer lines , which are presumed to be nonvariable over the timescale of the monitoring project . in the mean spectrum , the narrow component of h@xmath0 contributes @xmath48% of the total h@xmath0 flux . in addition , the h@xmath4 integration window includes the [ ] @xmath324363 line , and the h@xmath3 window includes [ ] @xmath296548 , 6583 . we do not attempt to decompose these emission blends into separate components . llcccccc 2010 - 05 - 19 & 5335.995 & 2.2 & 43 & & @xmath49 & @xmath50 & @xmath51 + 2010 - 06 - 08 & 5355.978 & 2.1 & 45 & @xmath52 & @xmath53 & @xmath54 & @xmath55 + 2010 - 06 - 11 & 5358.954 & 2.9 & 37 & @xmath56 & @xmath57 & @xmath58 & @xmath59 + 2010 - 06 - 13 & 5360.985 & 2.5 & 40 & @xmath60 & @xmath61 & @xmath62 & @xmath63 + 2010 - 06 - 15 & 5362.928 & 2.7 & 40 & @xmath64 & @xmath65 & @xmath66 & @xmath67 + [ spectable ] figure [ alllightcurves ] displays the balmer - line light curves . the h@xmath0 light curve is the cleanest of the four , because of the proximity of the h@xmath0 line to the [ ] line used as the flux scaling reference . the h@xmath3 light curve is substantially noisier than that of h@xmath0 , despite the higher flux of h@xmath3 , due to the issues in matching the flux scales of the red - side and blue - side spectra . nevertheless , all of the lines clearly show the same pattern of flux variations throughout the monitoring period . the variability amplitude of the light curves can be quantified using the statistics @xmath68 and @xmath69(e.g . * ; * ? ? ? * ; * ? ? ? * ) , where @xmath68 is the ratio of maximum to minimum flux observed , and @xmath69 is calculated according to @xmath70 where @xmath71 and @xmath72 are the variance and the mean - square measurement uncertainty in the fluxes , and @xmath73 is the average flux in the light curve . for the h@xmath0light curve of zw 229 - 015 , we find @xmath74 and @xmath75 . by both of these measures , zw 229 - 015 displayed strong variability , exceeding the values of @xmath68 and @xmath69 observed for the entire sample of 13 agns observed in the 2008 lick agn monitoring project during a shorter 2-month campaign @xcite . figure [ alllightcurves ] also displays the light curve of the continuum region 35503800 . ( we refer to this spectroscopic @xmath76-band continuum measurement as the @xmath77 light curve . ) this is the region most dominated by the agn continuum in the blue - side spectra . the spectroscopic continuum light curve is considerably noisier than the @xmath1-band light curve , as expected due to the combination of seeing - dependent slit losses , miscentering of the agn in the slit , and the nonparallactic slit orientation . nevertheless , the measurements illustrate that the @xmath76-band continuum generally follows the same pattern of variations as the @xmath1-band light curve , with a factor of @xmath47 variability amplitude during the monitoring period . although it is difficult to see in figure [ alllightcurves ] , the continuum and emission - line light curves do exhibit a small temporal offset . to demonstrate this more clearly , figure [ lczoom ] illustrates the central portion of the @xmath1-band continuum light curve ( in flux units ) and the h@xmath0 light curve , after rescaling to approximately the same flux level . aligning the two light curves by eye , it is evident that the h@xmath0 variations lag the continuum by @xmath78 days . to determine the reverberation lag quantitatively , we employed the interpolation cross - correlation function ( icf ) method @xcite , following the prescriptions described by @xcite , @xcite , and @xcite . the icf method deals with unevenly sampled time series by interpolating linearly between data points to obtain temporally matched time series for the continuum and emission - line light curves . for a given trial value of the lag time @xmath79 , the continuum light curve is shifted forward in time by @xmath79 days , and the cross - correlation is computed for the region of temporal overlap between the two light curves . the value of the cross - correlation function ( ccf ) for a given lag is computed twice : first by interpolating the shifted continuum light curve to the time steps of the emission - line light curve , and then by interpolating the emission - line light curve to the time steps of the shifted continuum light curve . the two interpolated versions of the ccf are then averaged together to give the final ccf . we computed the ccf for lag values ranging from @xmath80 to 30 days , in increments of 0.25 days . once the ccf is computed , the emission - line lag is determined in two ways : the peak of the ccf defines @xmath81 , while the centroid of the ccf for points above 80% of the peak value of the ccf gives @xmath82 . the value of @xmath81 and its uncertainty are quantized in units of the sampling interval used for the cross - correlation , in this case 0.25 days . black hole masses from reverberation data are most often determined using @xmath82 @xcite . the final cross - correlation results and their uncertainties were determined by following the monte carlo bootstrapping procedure described by @xcite , @xcite , and @xcite . in this method , a large number of modified realizations of the light curves are created and the cross - correlation function is computed for each realization of a pair of continuum and emission - line light curves . each simulated light curve is generated by selecting @xmath83 points randomly from the original light curve , where @xmath83 is the number of points in the original light curve and the random selection allows a particular point to be chosen more than once . if a particular point is selected @xmath84 times , then its uncertainty is reduced by a factor of @xmath85 . then , the simulated light curve is deviated by adding random gaussian noise based on the uncertainty in each data point . the emission - line and continuum light curves are each resampled @xmath86 times by this procedure , and for each pair of simulated light curves the cross - correlation function is determined . from these simulations , we build up distributions of values of @xmath82and @xmath81 . the final adopted values of @xmath82 and @xmath81 are defined to be the median values from each distribution . the @xmath87 uncertainty ranges on @xmath82 and @xmath81 are given by the range of lag values around the median that exclude the highest 15.87% and lowest 15.87% of lag values in the distribution . when the continuum and emission - line light curves have very similar observational cadences , the two interpolated versions of the ccf are typically very similar in shape . however , if the two light curves have very different temporal sampling , then the two ccfs can differ substantially . as described by @xcite , the two interpolated versions of the ccf are usually averaged together to obtain the final result , unless there is a strong reason to prefer one interpolation over the other . we tested whether this might affect the lag measurements by measuring @xmath81 and @xmath82 separately for the two interpolated versions of the ccf , and comparing them with the results obtained for the averaged ccf . although the two interpolated ccfs differed somewhat in shape , the measured lags were very similar in both cases , and the derived values of @xmath82 differed by only 0.18 days for the two separate ccfs . given this close agreement , we chose to use the averaged ccf for our final results . as discussed by @xcite , the reliability of cross - correlation lag measurements can be improved by subtracting off any long - term secular variations from the light curves that occur on timescales much longer than the reverberation timescale . in practice , this is often accomplished by subtracting a simple linear fit from the light curves . we found that the measurements of @xmath81 and @xmath82 were virtually unchanged regardless of whether the light curves were linearly detrended , and detrending with a quadratic fit ( in an attempt to remove the long - timescale curvature of the light curves ) only modified the measured lags by less than their @xmath87 uncertainties . for our final measurements , we chose to use the linearly detrended light curves . we also tested to see whether the cross - correlation measurements would be affected by removal of the last few spectroscopic data points , during the period of poor weather when the sampling was very sporadic , and found that the results for @xmath81 and @xmath82 again were consistent with our final adopted values within the @xmath87 uncertainties . as described above in [ photsection ] , the photometric errors listed in table [ phottable ] are likely to underestimate the true uncertainties . to account approximately for additional sources of error , we added 0.01 mag in quadrature to all of the @xmath1-band photometric errors prior to computing the cross - correlations . we found that this change had only a small effect on the cross - correlation results : the derived value of @xmath82 was either @xmath88 or @xmath89 days using the original photometric errors or the expanded error bars , respectively . our final measurement of the ccf gives observed - frame h@xmath0 lags of @xmath90 days and @xmath91 days , consistent with the general expectation of a @xmath78 day lag from inspection of the light curves ( as in figure [ lczoom ] ) . converting to the rest frame of zw 229 - 015 , the lags are @xmath92 and @xmath93 days . all of the lag measurements are listed in table [ lagtable ] . figure [ ccfs ] displays the autocorrelation function of the @xmath1-band light curve and the h@xmath0 vs. @xmath1-band ccf . the peak value of the ccf is 0.86 , indicating a fairly strong correlation between the continuum and emission - line variations . we also measured the h@xmath0 lag relative to the @xmath77-band light curve and found observed - frame lags of @xmath94 and @xmath95 days . within the substantial uncertainties , these results are consistent with the lag measurements carried out against the @xmath1 band . cross - correlating the @xmath1-band light curve against the @xmath77 continuum , we find values of @xmath81 and @xmath82 that are consistent with zero within their @xmath87 uncertainties , so there is no significant detection of a lag between continuum bands . higher - cadence observations would be needed in order to search for any cross - band continuum lags ( e.g. , * ? ? ? in addition to the featureless continuum of the agn , the @xmath77 spectral region also includes emission from the `` small blue bump , '' which contains both balmer continuum and emission . for ngc 5548 , @xcite showed that the small blue bump had a lag relative to the uv continuum that was similar to the lag of ly @xmath3 , consistent with the expectation that the small blue bump emission originates from the broad - line region ( blr ) . the nondetection of a lag between the @xmath77 and @xmath1 bands in zw 229 - 015 suggests that either the small blue bump makes only a small contribution to the @xmath77 band , or that the balmer continuum has an extremely short lag time relative to the primary continuum . the latter possibility would be consistent with the decreasing lags measured for higher - order balmer lines ( see [ balmerlags ] below ) . there is a small contribution to the @xmath1-band flux from the broad h@xmath0 line , and also an extremely small contribution from h@xmath3 at the red end of the johnson @xmath1 filter passband . this can potentially bias the lag measurement , by adding a component to the continuum light curve that has zero temporal shift with respect to the emission - line light curve . consequently , the lag measured from the cross - correlation peak will tend to be lower than the true lag . in order to assess the magnitude of this effect , we carried out synthetic photometry on zw 229 - 015 using a combined blue - side and red - side spectrum constructed from a night with good overlap between the blue and red sides . using the iraf synphot package , we calculated the @xmath1-band magnitude of the spectrum twice : first on the original spectrum , and then on a modified version of the spectrum in which the h@xmath0 and h@xmath3 emission lines were completely removed by manually interpolating over them . the two synthetic @xmath1-band measurements differed by only 0.02 mag , indicating that the broad balmer lines make only a very small contribution ( @xmath96 ) to the broad - band @xmath1 magnitude . to obtain a quantitative estimate of the magnitude of this bias on the lag measurement , we carried out a set of monte carlo simulations . we simulated a large number ( @xmath86 ) of agn light curves using the method of @xcite , for a power - density spectrum of the form @xmath97 normalized to a continuum rms variability of 20% over a 180-day period . a simple delta - function transfer function with a lag of 4 days was used to model the emission - line response to the continuum fluctuations . the `` contaminated '' @xmath1-band light curve was simulated by adding the emission - line light curve to the continuum light curve , weighted so that 2% of the total flux was from the emission - line contribution . with each set of simulated light curves we measured the emission - line lag relative to the uncontaminated continuum light curve , and relative to the continuum light curve with the added emission - line contribution . the simulations confirm that the bias in the lag measurements is very small for zw 229 - 015 . when the emission - line contribution to the continuum light curve is at the observed 2% level as in zw 229 - 015 , the simulations show that the measurement of @xmath82 is biased on average by only @xmath98 days ( with an rms scatter of 0.05 days in the bias for the full ensemble of simulations ) , relative to the measurement obtained by cross - correlating against the pure continuum light curve . for zw 229 - 015 , the bias is much smaller than the overall uncertainty in the lag measurement , but it can become significant for higher levels of emission - line contamination . we carried out additional simulations , and found that if the emission - line contribution to the broad - band light curve increases to 10% , for example , the average bias in the @xmath82 measurement is @xmath99 days ( for the same model parameters described above ) . for future reverberation campaigns that use broad - band photometry to measure the continuum light curves , it will be important to carry out simulations like these to assess the likely magnitude of this effect on each individual object . the size of the bias would depend on both the emission - line equivalent width and on the target s redshift , since higher redshifts would move the broad h@xmath0 line closer to the peak transmission of the @xmath1 filter . alternatively , another approach would be to subtract off the emission - line flux directly from the broad - band light curve before measuring the cross - correlation lag . lcc h@xmath3 vs. @xmath1 & @xmath100 & @xmath101 + h@xmath0 vs. @xmath1 & @xmath89 & @xmath102 + h@xmath4 vs. @xmath1 & @xmath103 & @xmath104 + h@xmath5 vs. @xmath1 & @xmath105 & @xmath106 + h@xmath0 vs. @xmath77 & @xmath107 & @xmath108 + @xmath1 vs. @xmath77 & @xmath109 & @xmath110 + [ lagtable ] the behavior of the reverberation lag as a function of velocity across broad emission lines is dependent on the kinematics of the blr , and recent ground - based programs have successfully obtained velocity - resolved measurements of reverberation lags for h@xmath0 and other lines in several agns @xcite . our zw 229 - 015 dataset is not ideally suited for velocity - resolved reverberation measurements , because the h@xmath0lag is fairly short and the monthly gaps in the spectroscopic light curve limit the measurement accuracy for lags of individual velocity segments of the h@xmath0 line . nevertheless , we are still able to obtain some rudimentary information on the velocity dependence of the reverberation lag across the h@xmath0 line profile . light curves were measured for several independent velocity segments across the h@xmath0 line , and the individual segment light curves were cross - correlated against the @xmath1-band continuum . the results are illustrated in figure [ vdelay ] . six velocity segments were used for the final measurements . in the line wings , where the variability amplitude is relatively low , the measurements have large uncertainties ; the lags measured for the blue - wing and red - wing segments are consistent with zero within their @xmath87 uncertainties . ( we used a broader velocity segment for the blue wing because subdividing the blue wing into even smaller segments led to very poor cross - correlation results , with some subsegments having negative lags . ) with these caveats in mind , the velocity - resolved behavior seen in figure [ vdelay ] shows marginal evidence for a slightly longer lag in the line core relative to the mean lag , while the lag in the line wings is shorter . a symmetric blue - red response in the velocity - delay map , with shorter lags in the line wings , is qualitatively consistent with keplerian motion of the blr clouds ( e.g. , * ? ? ? * ) , but due to the large uncertainties in the velocity - resolved measurements we refrain from drawing any specific conclusions about the blr kinematics . the results do illustrate that zw 229 - 015 is a promising candidate for further velocity - resolved reverberation work . given the strong and rapid variability in this object , a high - cadence monitoring campaign could obtain much better results , and a reconstruction of the two - dimensional transfer function ( i.e. , the distribution of lag time as a function of velocity ) might be obtainable ( e.g. , * ? ? ? we measured cross - correlation lags for the other balmer lines following the same methods that were used for h@xmath0 , including the linear detrending . the balmer - line reverberation lags follow the same pattern found previously for other agns , where the h@xmath3 line has the longest lag and higher - order balmer lines have progressively shorter lags ( e.g. , * ? ? ? * ; * ? ? ? the lag of h@xmath5 is too small to be resolved significantly , and our measurement of @xmath82indicates a @xmath87 upper limit of @xmath111 days . as discussed by @xcite , this trend in balmer - line lags is the result of the differences in optical depths for the balmer lines in the blr , and the same behavior is seen in blr photoionization models @xcite . for example , the h@xmath3 line has the highest total optical depth through the blr , and the observed h@xmath3 emission originates primarily from larger radii in the blr where the optical depth is smaller . using the @xmath82 measurements , we find a ratio of balmer - line reverberation lags of @xmath79(h@xmath3):@xmath79(h@xmath0):@xmath79(h@xmath4 ) = 1.31:1.00:0.87 , which is within the range observed by @xcite for agns having similar emission - line lags of a few days . virial estimates of black hole masses based on reverberation - mapping data employ the width of broad emission lines to give the typical velocity for blr gas . following procedures often employed for reverberation - mapping data ( e.g. , * ? ? ? * ) , we measured the fwhm and the line dispersion ( or second moment ) of the h@xmath0line from both the mean and the rms spectra . in both cases , a linear continuum was fitted to regions on either side of the h@xmath0 line , and then subtracted off . the fwhm was measured directly from the continuum - subtracted line profile , and the line dispersion @xmath112(in velocity units ) was calculated as @xmath113 / \left[\sum s_i\right ] - \lambda_0 ^ 2 \right),\ ] ] where @xmath114 is the flux density at wavelength bin @xmath115 , and @xmath116 is the flux - weighted centroid wavelength of the line profile . while the narrow component of h@xmath0 essentially vanishes in the rms spectrum , it must be removed from the mean spectrum before the broad - line width can be measured . the narrow [ ] @xmath324959 emission line has a nearly gaussian profile with fwhm = 6.6 , and we used this as a model to subtract off the narrow component of h@xmath0 . to determine the final values and uncertainties for the line widths , we followed the monte carlo method described by @xcite and @xcite , creating randomly resampled realizations of the mean and rms spectra . to construct each realization , we randomly selected 54 spectra from our dataset , without regard to whether an individual spectrum was already selected , and then determined the mean and rms spectra for that randomly sampled set . for a total of 1000 realizations , we measured fwhm and @xmath112 for h@xmath0 , from both the mean and rms spectra . in each realization , the borders of the continuum - fitting windows were shifted randomly over a range of 10 in order to account for the line - width uncertainty resulting from the specific choice of continuum regions . the final value and uncertainty for each line - width measurement ( fwhm and @xmath112 ) are taken to be the median value and standard deviation determined from the full ensemble of simulations . from the rms spectrum , we measure fwhm = @xmath117 km s@xmath7 and @xmath118 km s@xmath7 for h@xmath0 . the mean spectrum gives fwhm = @xmath119 and @xmath120 km s@xmath7 for the broad h@xmath0 line . after removal of the narrow component from the mean spectrum , the broad h@xmath0 line centroid in the observed frame is @xmath121 . the redshift inferred from the broad h@xmath0 line is therefore 0.0271 , and there is no significant velocity offset between the narrow [ ] line and the broad h@xmath0emission . the observed line widths are affected by instrumental broadening in the 4 wide slit , but only by a very small amount . the instrumental line width adds in quadrature to the observed line width ( as in @xmath122 , or similarly for the fwhm values ) . we can obtain an upper limit to the instrumental broadening by measuring the widths of arc - lamp lines observed through the 4 slit , since the arc lamps uniformly illuminate the entire slit width while the agn does not ( at least under typical seeing conditions ) . from the line at 5086 , we find an upper limit to the instrumental broadening of fwhm@xmath123 , corresponding to an instrumental velocity width of fwhm@xmath124 km s@xmath7 or an instrumental dispersion of @xmath125 km s@xmath7 . this instrumental broadening is small enough that the maximum possible contribution of instrumental broadening to the observed @xmath112 of the broad h@xmath0 line is only 0.9% of the observed line dispersion . to estimate the likely value of the instrumental dispersion @xmath126 , we followed the method described by @xcite , using an observation of [ ] through a narrower slit to determine the intrinsic width of the line . the initial test spectrum of zw 229 - 015 , taken on 2010 april 18 , was observed through a 2 slit . in that exposure , the [ ] @xmath325007 line has a raw fwhm of 5.44 , and the 5086 arc line has fwhm = 4.18 . assuming that the agn nearly fills the slit in the 2 observation ( roughly the size of the seeing disk ) , this implies an intrinsic line width of fwhm = 203 km s@xmath7 for [ ] . the instrumental broadening for the 4 slit can then be estimated as the difference in quadrature between the observed width of [ ] in the 4 slit ( fwhm = 373 km s@xmath7 ) and the intrinsic [ ] width measured from the 2 observation . this implies an instrumental contribution of fwhm@xmath127 km s@xmath7 or @xmath128 km s@xmath7 for the 4 slit , consistent with the upper limits derived previously . the corresponding corrections to the measured broad h@xmath0 line width for this instrumental broadening are very small . from the rms spectrum , the corrected h@xmath0 widths are fwhm = @xmath129 km s@xmath7and @xmath6 km s@xmath7 , and in the mean spectrum the corrected h@xmath0 widths are fwhm = @xmath130 and @xmath131 km s@xmath7 . measurements of reverberation lag and broad - line width can be combined to give an estimate of black hole mass , under the assumption that the line width is primarily due to virial motion of broad - line gas ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . from the virial theorem , we have @xmath132 where @xmath79 is the reverberation lag and @xmath133 gives the mean radius of the blr , @xmath134 is some measure of the broad - line width in velocity units ( typically either fwhm or the line dispersion @xmath135 ) , and @xmath136 is a scaling factor which is dependent on the geometry and kinematics of the blr as well as our viewing angle . since these parameters of the blr are unknown , it has become customary to use a single value for @xmath136 that should ideally represent the mean value of the virial normalization factor for an appropriate ensemble of agns . the adopted value of @xmath136 can be based either on some set of assumptions about blr kinematics ( e.g. , * ? ? ? * ) , or by finding the normalization that brings reverberation - mapped agns into best agreement with the local @xmath137 relation of quiescent galaxies @xcite . for consistency with the majority of recent reverberation work , we use @xmath112(h@xmath0 ) measured from the rms line profile as the measure of @xmath134 , and @xmath138(h@xmath0 ) as the measure of blr size ; these parameters have been shown to yield the most robust mass estimates @xcite . for the virial normalization factor , we use @xmath139 , derived from the full available sample of reverberation - mapped agns having measured stellar velocity dispersions @xcite . combining the rest - frame value @xmath140 with @xmath6 km s@xmath7 , we obtain a `` virial product '' of @xmath141 @xmath9 . applying the virial normalization factor @xmath142 , this gives an estimated black hole mass of @xmath143 @xmath9 . we note that the quoted uncertainties on @xmath144 above only include the propagated errors on the lag and line - width measurements , but not the ( unknown ) systematic uncertainty on the applied value of @xmath136 . from examination of the scatter of reverberation - mapped agns about the best - fitting @xmath137 relation , the typical uncertainty in the value of @xmath136 applied to any individual agn is probably at the level of a factor of @xmath145 ( e.g. , * ? ? ? * ; * ? ? ? * ) , so the actual error in the estimated value of @xmath144 is dominated by the uncertainty in @xmath136 and not by the measurement errors in the h@xmath0 lag or width . a long - term goal of reverberation mapping work is to use velocity - resolved measurements to directly constrain the blr geometry and kinematics so that accurate black hole masses can be derived for individual agns , but such analysis is beyond the scope of this work . we can estimate the eddington ratio @xmath146 for zw 229 - 015 using the @xmath1-band continuum flux and a bolometric correction . without high - resolution optical images or an accurate model of the stellar population , we can not directly determine the amount of starlight contamination to the @xmath1-band magnitudes or spectroscopic continuum fluxes , but a rough approximation will suffice to estimate @xmath146 . according to the extinction map of @xcite , the galactic foreground extinction toward zw 229 - 015 is @xmath147 mag . we make the simplifying assumption that the extinction correction and starlight contamination correction would roughly compensate for each other , and take the rest - frame 5100 flux density from the mean spectrum as an estimate of the intrinsic ( extinction - free ) agn flux density . we find @xmath148(5100 ) @xmath149 erg s@xmath7 @xmath39 in the mean spectrum , corresponding to @xmath150(5100 ) @xmath151 erg s@xmath7 for a luminosity distance of 119 mpc ( assuming wmap-7 cosmological parameters with @xmath152 km s@xmath7 mpc@xmath7 ; larson et al . 2011 ) . applying the same bolometric correction used by @xcite of @xmath153(5100 ) , we obtain an estimated @xmath154 erg s@xmath7 , and @xmath155 . this value is within the typical range found for nearby reverberation - mapped seyferts @xcite . the possible influence of radiation pressure on black hole masses derived from reverberation mapping has been the subject of much recent discussion @xcite . the impact of radiation pressure on the motion of blr clouds is likely to be strongest for agns radiating at high luminosity , but zw 229 - 015 has a rather modest eddington ratio and any corrections for radiation pressure should be relatively minor . @xcite proposed a modification to the virial equation , adding a term proportional to the agn continuum luminosity , as in @xmath156 where @xmath157 at 5100 , and where the numerical value of @xmath136 differs from the value used to normalize the `` standard '' virial equation ( equation [ virialequation ] ) . using the sample of reverberation - mapped agns having measured stellar velocity dispersions , they estimated best - fit values of @xmath158 and @xmath159 . applying this formalism to zw 229 - 015 , we obtain an estimate of @xmath160 @xmath9 for zw 229 - 015 . this value is lower than the pure virial estimate for @xmath144 , due to the combination of the smaller @xmath136 value derived by @xcite and the relatively low luminosity of zw 229 - 015 , but it differs from the pure virial estimate by only 12% , or less than @xmath87 . from examination of the orbits of blr clouds under the influence of radiation pressure , @xcite proposed an alternative variant of the virial equation as a new mass estimator , dependent on the agn continuum luminosity and broad - line fwhm . applying their mass estimator ( their equation 18 ) and using fwhm(h@xmath0 ) measured from the mean spectrum , we obtain @xmath161 @xmath9 for zw 229 - 015 . in either case , the corrections to @xmath144 for radiation - pressure effects are not very large , and they are far smaller than the overall uncertainty in the value of the virial normalization factor @xmath136 for an individual agn . this dataset was obtained using a somewhat unconventional strategy , in which the spectroscopic observations were done on a nightly basis , but only during the dark runs . we find that this method can deliver high - quality reverberation results , provided that a well - sampled photometric light curve ( not having large gaps ) is also obtained . since there are many facilities that can deliver nightly , queue - scheduled imaging observations without monthly gaps , this is now a straightforward approach to measurement of long - term light curves and reverberation lags for additional agns . in terms of the measurement accuracy for the h@xmath0 lag , our results for zw 229 - 015 are similar in quality to what was obtained for several objects with similar lag times of a few days during the 2008 lick agn monitoring project @xcite , a program comprising 64 mostly consecutive nights of spectroscopic observations along with nightly photometry . our results demonstrate that zw 229 - 015 will be an excellent target for a variety of future investigations . its strong and rapid variability is noteworthy in comparison to many other objects having similar black hole masses and luminosities ( e.g. , * ? ? ? * ) , and it is very fortunate that such a nearby and highly variable agn happens to be in the _ kepler _ field of view . while the _ kepler _ light curve for zw 229 - 015 is not yet available , we anticipate that it will enable fundamentally new investigations into the properties of agn optical continuum variability . there are only a few other low - redshift seyfert galaxies being monitored by _ kepler _ , and although their variability properties are currently unknown , it seems likely that zw 229 - 015 will be the most important object of this small sample . as such , it is crucial that zw 229 - 015 continue to be monitored by _ kepler _ for the longest possible duration during the mission . additional observations of zw 229 - 015 can be used to measure the galaxy s bulge properties , to place it on the correlations between black hole mass and bulge luminosity or mass , and to add it to the sample used to calibrate the relationship between blr radius and continuum luminosity following the methods described by @xcite . similarly , a measurement of the bulge velocity dispersion for zw 229 - 015 can be used to add this galaxy to the @xmath137 relation for reverberation - mapped agns @xcite . our wide - slit observations are not well suited to carrying out measurements of stellar kinematics , but stellar absorption features including the triplet absorption lines are easily visible in the lick spectra , so it should be straightforward to measure the bulge velocity dispersion with suitable new observations . other observing programs for zw 229 - 015 are currently in progress , including near - infrared monitoring with _ spitzer _ to measure the dust reverberation timescale . we will also revisit our h@xmath0 reverberation measurement when the _ kepler_light curve becomes available , since the sampling and precision of the _ kepler _ data may lead to an improved determination of the emission - line lag . to summarize our conclusions : we have measured a rest - frame reverberation lag of @xmath162 days for the h@xmath0 line in zw 229 - 015 . combining this result with the broad h@xmath0 line width , we obtain a virial estimate of @xmath163 @xmath9for the black hole in this galaxy , under the assumption of a virial normalization factor of @xmath142 . the mass estimate does not change significantly if we adopt the method of @xcite to obtain a mass estimate corrected for the dynamical influence of radiation pressure in the blr . the _ kepler _ target zw 229 - 015 is destined to be one of the most important agns for investigations of the origin and characteristics of optical continuum variability . as recent observations have begun to reveal connections between optical continuum variability characteristics and black hole masses in agns , this estimate of @xmath144 will be a valuable asset for future interpretation of the _ kepler _ light curve for zw 229 - 015 . research by a.j.b . , m.l.n . , and j.l.w . at uc irvine has been supported by nsf career grant ast0548198 . a.v.f.s group is grateful for the financial support of nsf grant ast0908886 , the tabasgo foundation , gary and cynthia bengier , and the richard and rhoda goldman fund ; also , j.m.s . thanks marc j. staley for a graduate fellowship . kait and its ongoing operation were made possible by donations from sun microsystems , inc . , the hewlett - packard company , autoscope corporation , lick observatory , the nsf , the university of california , the sylvia & jim katzman foundation , and the tabasgo foundation . the west mountain observatory 0.9 m telescope has been supported by nsf grant ast0618209 . , e.g.h . , and j.w.m . would like thank the department of physics and astronomy at brigham young university for continued support of research efforts at the west mountain observatory . tommaso treu s research group at ucsb has been supported by nsf career grant nsf0642621 and a packard fellowship . research by j.h.w . has been supported by basic science research program through the national research foundation of korea funded by the ministry of education , science and technology ( 2010 - 0021558 ) . we thank rick edelson for discussions that motivated this project . we are grateful to the lick observatory staff for their assistance in obtaining the spectroscopic observations , and to the staff of the maidanak observatory for obtaining data for this project . we also thank tabitha buehler , carl melis , dovi poznanski , x. prochaska , and ben zuckerman for contributing observations . this research has made use of the nasa / ipac extragalactic database ( ned ) which is operated by the jet propulsion laboratory , california institute of technology , under contract with nasa .

we investigate the correlation between the mass of a central supermassive black hole and the total gravitational mass of the host galaxy ( ) . the results are based on 43 galaxy - scale strong gravitational lenses from the sloan lens acs ( slacs ) survey whose black hole masses were estimated through two scaling relations : the relation between black hole mass and srsic index ( ) and the relation between black hole mass and stellar velocity dispersion ( ) . we use the enclosed mass within , the radius within which the density profile of the early type galaxy exceeds the critical density of the universe by a factor of 200 , determined by gravitational lens models fitted to hst imaging data , as a tracer of the total gravitational mass . the best fit correlation , where is determined from relation , is over 2 orders of magnitude in . from a variety of tests , we find that we can not reliably infer a connection between and from the relation . the relation provides some of the first , direct observational evidence to test the prediction that supermassive black hole properties are determined by the halo properties of the host galaxy .

0909.0269

supermassive black holes ( smbh ) are believed to reside in nearly all galaxies @xcite and the masses of these smbh ( @xmath5 ) show correlations with host galaxy properties , implying that smbh and galaxy formation processes are closely linked @xcite . previous studies have shown correlations between @xmath5 and galaxy s effective stellar velocity dispersion ( @xmath9 ) , bulge luminosity ( @xmath10 ) , s@xmath1rsic index ( _ n _ ) and stellar mass of the bulge component ( @xmath11 ) @xcite . some of the challenges faced by current models of smbh formation and evolution include reproducing and maintaining these scaling relations regardless of the events that take place during galaxy evolution driven by the process of hierarchical mass assembly @xcite . these scaling relations are not only important tests of the models of smbh formation and evolution but also powerful predictive tools from which we can infer smbh masses of galaxies that are located at higher redshifts . in this paper , we examine the evidence for a scaling relation between @xmath5 and the _ total _ mass of the host galaxy ( denoted as @xmath0 ) . most self regulating theoretical models of smbh formation predict a fundamental connection between @xmath5 and @xmath0 of the host galaxy @xcite . one of the most important predictions of the galaxy models , which study the interaction between the dark matter haloes of galaxies and baryonic matter settling into the gravitational potential to form the bulge and smbh , is that halo properties determine those of the bulge component and smbh @xcite . however , observational evidence for such a scaling relation has been sparse since measurement of the total mass is non - trivial . results from @xcite examined some of the first _ indirect _ observational evidence for the existence of a @xmath8 relation . in @xcite , the correlation between bulge velocity dispersion ( @xmath12 ) and the observed circular velocity ( @xmath13 ) , for a sample of 20 elliptical galaxies and 16 spiral galaxies , is translated into an equivalent @xmath8 correlation . although @xmath12 can be translated into @xmath5 through @xmath14 relation in a straightforwards manner , an estimate of @xmath0 is dependent on the conversion between @xmath13 and virial velocity , the velocity of the galactic halo at the virial radius . @xcite uses results from @xmath15cdm cosmological simulations to derive @xmath0 . following @xcite , several studies have examined the @xmath16 relation and its implications for galaxy formation and evolution @xcite . the primary goal of this study is to extend the observational evidence to support the @xmath17 relation using an attractive , alternative , and _ direct _ method to measure the total mass of the host galaxy . in recent years , strong gravitational lensing has emerged as a powerful tool to probe the mass profiles , ranging from individual galaxies to clusters of galaxies @xcite . due to the fact deflection of a photon passing an intervening massive object is independent of the deflector s dynamical state , gravitational lensing does not suffer from difficulties associated with dynamical mass measurements of galaxies , where assumptions regarding orbital motions of tracers can lead to complications . thus , gravitational lensing is a more robust method to estimate the total mass , including dark matter , around early - type galaxies and infer the existence of an isothermal mass profile ( @xmath18 ) in various systems @xcite the mass enclosed within the einstein radius ( @xmath19 ) which is measured from the lens model that produces the best fit to the observed multiple images , is a direct probe of both luminous and dark mass in a galaxy . in conjunction with velocity dispersion and surface brightness profiles of the lens galaxy , measurement of @xmath19 from a lens model can be used effectively to constrain the luminous and dark matter profiles . in this study , we utilize the @xmath19 as a tracer of the total mass and use the @xmath20 mass profile of the lensing galaxy to determine the mass contained within a redshift - independent circular aperture . we also attempt to derive the @xmath8 relation using purely photometric tracers of @xmath5 and @xmath0 . therefore , we examine the possibility of using @xmath2 relation @xcite as the primary method to estimate black hole masses . surface brightness profiles of the bulge component of most galaxies , in particular e / s0 , can be described by the s@xmath1rsic law @xcite as follows : @xmath21 } ) } \label{eq : sersic}\end{aligned}\ ] ] where @xmath22(r ) is the surface brightness at radius @xmath23 and s@xmath1rsic index ( @xmath24 ) is the degree of light concentration . previous studies in the literature indicate that the quantity _ n _ varies monotonically with galaxy magnitude . therefore , the existence of a @xmath25 relation can be inferred given the dependence of @xmath5 on the galaxy magnitude @xcite and the connection between galaxy magnitude and s@xmath1rsic index @xcite . @xcite discuss the most recent version of the @xmath25 relation and indicate that the dependence between the quantities @xmath26 and @xmath27 can be best represented by a log - quadratic relation . the main motivation for using this relation is that measurement of _ n _ requires only imaging data , which is easier to acquire than spectroscopic data at a given redshift ; therefore , @xmath8 could be easily extended to large samples of higher redshift gravitational lenses , which may not have spectroscopic measurements . furthermore , _ n _ is a distant - independent quantity , an added advantage for estimating black hole masses of high redshift galaxies that may not have secure redshift measurements . this paper is structured as follows . in 2 we present a brief description of the target selection . in 3 we describe the following analysis procedures : 3.1 measurements of @xmath0 ; 3.2 : deriving the connection between @xmath5 and @xmath0 using @xmath3 relation ; 3.3 : deriving the connection between @xmath5 and @xmath0 using @xmath2 relation . in 4 we present the results of our analysis . in 5 and 6 we discuss the implications of our results and give some concluding remarks regarding this project . we assume the following cosmological terms for all computations in the paper : @xmath28 , @xmath29 , @xmath30 and @xmath31 . unless otherwise noted all scaling relations in this paper are defined as linear relationships in log - log space and all logarithms assume a base of 10 . our sample is a subset of the galaxy - scale strong gravitational lenses from the sloan lens acs ( slacs ) survey @xcite , a hubble space telescope ( hst ) imaging survey that is optimized to detect bright early - type lens galaxies with faint lensed sources . these lenses were initially selected from spectra of galaxies of the sdss luminous red galaxy ( lrg ) and main galaxy samples @xcite . after a best - fit template of the continuum was subtracted from the spectrum of each galaxy , the residual spectra were scanned for nebular emission lines at a redshift higher than that of the target galaxy . galaxies whose residual spectra exhibited at least three atomic transitions at a common background redshift were identified as lens candidates @xcite and imaged using the wide - field channel ( wfc ) of the advanced camera for surveys ( acs ) through f435w and f814w filters ( programs 10174 , 10587 & 10886 ) . a detailed description of the gravitational lenses discovered by the slacs survey and subsequent analysis can be found in the series , the sloan lens acs survey 1 - 8 @xcite . to date , slacs survey has discovered 131 galaxy - scale strong gravitational lenses with snapshot or deep hst acs imaging , secure spectroscopic redshifts for both lens and source galaxies in each system and stellar velocity dispersion , @xmath32 , measured within the 3 " sdss spectroscopic fiber @xcite . from the 131 slacs lenses , 70 systems are classified as `` grade - a '' systems by @xcite , where the grade - a classification implies a definite lens as determined by hst imaging . we use 43 `` grade - a '' lenses , from programs 10174 , 10587 and 10886 , as the sample for our analysis . we do not use lens systems that do not have a simple lens model due to nearby companion galaxies ( 7 lens systems ) or systems where @xmath33 value is not available ( 14 lens systems ) because the median signal - to - noise ( snr ) was too low within the 3 " sdss spectroscopic fiber @xcite . the most important advantage of using the slacs galaxy sample to determine the @xmath8 relation is that we have an accurate measurement of the total mass profile of each galaxy . however , slacs lenses are located at distances ( @xmath34 ) where the smbh mass can not be measured directly through its influence on the surrounding gas or stars . therefore , we use two scaling relations to determine the smbh masses of the galaxy sample : @xmath3 relation @xcite and @xmath2 relation @xcite . we use parameters from the strong lens modeling performed by @xcite to obtain @xmath0 . @xcite use a singular isothermal ellipsoid ( sie ) lens model to describe foreground lens plane which generates the multiple images of the source . the sie lens model is parameterized by the angular einstein radius , @xmath35 , which relates to the physical parameters of the mass model as follows : @xmath36 where @xmath37 and @xmath38 are angular diameter distances between the lens and source plane and observer and source plane respectively . the surface brightness profile of the background galaxy is described by a single or multiple s@xmath1rsic ellipsoid profile and the model lensed image , for a given sie lens model , is produced by the forward ray tracing method . the model parameters are adjusted until the images match the observations and the solutions are optimized using a merit function , such as @xmath39 @xcite . we use the lens model parameters given by @xcite , i.e. @xmath35 , and integrate the best - fit mass profile to obtain the total mass ( both luminous and dark ) within a projected radii of choice . during the preliminary analysis , we use the inverted lensing equation to compute mass inside the einstein radius ( @xmath19 ) , as a tracer of @xmath0 , assuming that the lens and source galaxies are aligned along the optical axis . however , @xmath19 is not a measurement of mass within a @xmath40 aperture since the projected radius characterized by the einstein radius for a given mass distribution is redshift dependent . we therefore use @xmath4 , the radius within which the mean density of the lensing galaxy exceeds the critical density of the universe ( @xmath41 ) by a factor of 200 , as a `` standard aperture '' . the use of @xmath4 as a radial aperture is also consistent with theoretical studies of the @xmath8 relation @xcite . we integrate the best - fit mass profile to obtain the mass contained within a projected radius of @xmath4 ( @xmath42 ) as a tracer of the total mass as characterized below : @xmath43 adopting an aperture - corrected " mass measurement ( @xmath42 ) , instead of @xmath19 , does not affect the overall form of the @xmath8 relation ( the slope and zero - point ) discussed in the following sections . results of @xcite show a lack of correlation between the quantities @xmath44 and mass or @xmath45 , where @xmath46 is the stellar velocity dispersion corrected to an aperture of @xmath47 . these results are consistent with the near isothermal nature of the radial profile ; thus , using a radial aperture other than @xmath48 , to derive the total mass , does not introduce an additional dependence on the assumed mass model . we also observe a relationship of @xmath49 , consistent with the fact that an aperture corrected mass measurement does not affect the observed @xmath8 relation . to obtain the associated 1-@xmath50 error of @xmath42 , we use gaussian error propagation adopting an empirical error of 2% on the measured einstein radii following @xcite ( @xmath51 ) . table 1 lists the unique sdss identifiers for slacs lenses , redshift of the lens galaxy of a system and the derived values of @xmath42 ( hereafter denoted as @xmath0 ) for each lensing system . we first use the @xmath3 relation to estimate @xmath5 of the early type lens sample . the velocity dispersion measurements are derived from spectra from the sdss 3 " diameter fiber . the sdss velocity dispersion measurements are corrected to a standard aperture , of radius equal to @xmath52 , using the power - law relation by @xcite , similar to smbh and fundamental plane studies @xcite . the mean difference between sdss velocity dispersion measurements and the aperture corrected measurements ( hereafter denoted as @xmath9 ) are @xmath53 ; therefore , the aperture correction does not significantly influence the overall form of the @xmath8 relation discussed in the following sections . we use @xmath3 relation by @xcite , obtained from measurements of @xmath5 from dynamically detected central black holes : @xmath54 with an intrinsic scatter of @xmath55 dex . we assume that errors on @xmath56 and best - fit coefficients of the @xmath57 relation are uncorrelated and use gaussian error propagation to determine the 1-@xmath50 error on the quantity @xmath58 . the values of @xmath56 and its associated error for each lensing system and smbh mass estimates obtained from @xmath3 relation are listed in table 1 . we also attempt to use the @xmath25 relation to estimate @xmath5 of the slacs lens sample @xcite . we perform a careful and detailed decomposition of hst acs images to obtain the s@xmath1rsic indices of the slacs lens galaxies . although our sample of lens galaxies were observed in at least two acs filters , mainly f814w and f435w , we exclusively use f814w ( broad i - band ) data due to high snr of the images . furthermore , redder magnitudes are a better tracer of mass in comparison to b - band magnitudes . to keep our decomposition method consistent with that of @xcite , we use a two - component model to describe the surface brightness profiles of slacs lens galaxies . in addition to the s@xmath1rsic profile , which describes the surface brightness of the bulge component in each galaxy , we include a second disk component characterized by a simple exponential profile : @xmath59 where @xmath60 is the surface brightness at radius @xmath23 . to obtain the best possible accuracy of the data analysis process , we take particular care during all intermediate steps leading to the bulge - disk composition . therefore , we obtain the raw images and perform careful cosmic ray removal , distortion correction , manual mask production and determination of the best possible acs point - spread function ( psf ) . we obtain the acs images from the hst archive at the canadian astronomy data centre ( cadc ) , where the images are processed by calacs calibration software . imaging data from programs 10174 and 10587 are 420 s single - exposure snapshot data ; therefore , we perform an additional cosmic ray identification step using the l.a.cosmic software ( lacos ) @xcite , which distinguishes between undersampled sources and cosmic rays . we then use the multidrizzle reduction package to obtain distortion - free acs images , where the distortion solution is applied to correct all pixels to equal areas . we use the photometry package sextractor @xcite to produce a catalog of galaxies in each acs image . furthermore , we use the segmentation images generated by sextractor , which deblends each source in the field , to produce manual masks of the lensed features around the early type galaxy before lens modeling . we perform the two - dimensional decomposition of each early - type lens galaxy using galaxy image 2d ( gim2d ) , which is publicly available to users @xcite . gim2d uses the metropolis algorithm to derive the best - fit parameter values and confidence intervals , for a surface brightness model of choice , through monte - carlo sampling of the likelihood function . since our surface brightness model consist of a bulge and a disk component , we explore the full range of bulge fraction ( b / t = 0.0 - 1.0 ) such that the metropolis algorithm converges to an accurate quantitative morphology of galaxies classified as e - types . on average , we find that most slacs lenses have a significant bulge component ( @xmath61 ) and a bulge plus disk light profile produces a better fit to the lens galaxy , significantly minimizing the residuals when the galaxy model is subtracted from the input galaxy image . for gim2d lens models , we take particular care to define the acs point spread function ( psf ) , which is extremely position- and time- dependent . we investigate a variety of psf models to determine the most suitable psf for convolution with gim2d galaxy models . the psf models are as follows . 1 . a star extracted from the field . a psf produced at the location of the galaxy , in the non - drizzled image , using tiny tim software @xcite . we insert the tiny tim generated psf into an empty acs - wfc field , at the location of the galaxy , and multidrizzle the resulting image to obtain a distortion - free psf model . 3 . a psf produced at the location of the galaxy using the principle component analysis ( pca ) models in the acs psf library @xcite . we examined the residual images produced by subtracting the lens surface brightness model , each convolved with the psf models described above , to determine the best - fit psf for gim2d lens modeling . the psf model produced from the acs library gives the best results , minimizing the core residuals for a fixed surface brightness model . therefore , we extract the s@xmath1rsic indices from the best - fit bulge plus disk decomposition , using the acs psf library model for convolution . an example of the difference between residual images produced by various psf models are shown in figure [ fig : psfcompare ] . we also examine the reliability of the bulge - disk decomposition by comparing our results for the slacs lens sample relative to the scaling relations for spheroidal components of the sdss galaxies . since slacs lenses were initially derived from sdss luminous red galaxy ( lrg ) and main samples , structural parameters measured from acs imaging should be in agreement with those of sdss galaxies . figure [ fig : kormendy ] shows the comparison of effective bulge radius ( @xmath62 ) , i - band magnitude of the bulge component ( @xmath63 ) and aperture - corrected bulge velocity dispersion ( @xmath64 ) of the slacs lenses to the sdss bulge parameters . we construct the scaling relations for sdss galaxies from the bulge - disk decomposition of 77523 galaxies and velocity dispersions from the sdss database . the structural parameters derived from acs imaging are in good agreement with those of sdss galaxies , which confirms the accuracy of our bulge - disk decomposition ( in addition to the minimal residuals seen in the galaxy images after subtracting the galaxy models ) . the outlier in figure [ fig : kormendy ] is sdss j0959 + 0410 , a disk galaxy with a small bulge fraction ( @xmath65 ) . previous studies , which examine the positions of bulges with disk - like features ( also referred to as @xmath66 ) in structural parameter space find that pseudobulges with low @xmath67 deviate from the scaling relations of classical bulges @xcite . using the best - fit parameter for _ n _ from gim2d models , we estimate the smbh masses using the log - quadratic @xmath68 relation by @xcite : @xmath69 ^ 2 \label{eq : grahammn}\end{aligned}\ ] ] with an intrinsic scatter of @xmath70 dex . similar to @xcite , we assume a measurement error of 20% on the values of _ n _ to obtain the 1-@xmath50 error on @xmath58 . table 1 lists the s@xmath1rsic indices and bulge fractions of the slacs lenses from gim2d modeling , estimates of @xmath5 for each lens system . in the following section , we combine the correlation between the primary observable quantities ( @xmath72 ) with @xmath3 relation to derive the @xmath8 relation . a tight correlation between the quantities @xmath73 and @xmath74 is apparent from figure [ fig : sigmamtot ] . from spearman s rank test , we obtain a correlation coefficient ( @xmath75 ) of 0.84 ( degrees of freedom = 41 ) , which indicates a strong positive correlation between the quantities @xmath73 and @xmath74 , with a @xmath76 confidence level that the correlation has not occurred by chance . to quantify this correlation , we use the @xmath39-fitting routine by @xcite , which implements a generalized form of the least - squares fitting routine by @xcite . @xcite routine accounts for intrinsic scatter ( @xmath77 ) beyond the observational errors , for a relation of interest , by adding @xmath77 in quadrature to the error in the dependent variable . initial fits to the observed relation between @xmath78 and @xmath74 , incorporating observational errors in both variables , gives a large reduced @xmath39 value ( @xmath79 ) indicative of intrinsic scatter in the relation . therefore , we perform the fits by incorporating observational errors in both variables and intrinsic scatter in the @xmath80-variable . the value of @xmath77 is determined by requiring that @xmath81 is unity . the results of the fitting routine gives the following best fit correlation : @xmath82 } \label{eq : sigmamtoteq}\end{aligned}\ ] ] with an intrinsic scatter of 0.037 dex in @xmath78 . combining equations [ eq : msigma ] and [ eq : sigmamtoteq ] , we derive the following @xmath8 relation : @xmath83 \label{eq : observational}\end{aligned}\ ] ] figure [ fig : frommsigma ] shows @xmath9 transformed into smbh masses using equation [ eq : msigma ] . we assume that measurement errors of the velocity dispersions and best - fit coefficients of the @xmath3 relation are uncorrelated and use gaussian error propagation to determine the 1-@xmath50 error on the quantity @xmath84 . we also perform a direct fitting step to @xmath0 and the secondary observable quantity , @xmath5 , to confirm the validity of the derived @xmath8 relation . we find @xmath85 with @xmath86 and @xmath87 , which is in agreement with the result shown in equation [ eq : observational ] . in this section , we discuss the @xmath89 relation of the slacs lenses , shown in figure [ fig : nmtot ] , and figure [ fig : frommn ] which shows @xmath90 converted to smbh masses using equation [ eq : grahammn ] @xcite . similar to 4.1 , we use gaussian error propagation ( assuming a measurement error of 20% on @xmath90 ) with intrinsic scatter of the @xmath8 relation ( @xmath91 ) added in quadrature , to determine the 1-@xmath50 error of the quantity @xmath92 . results of @xcite indicate that @xmath93 relation is comparable to the @xmath3 relation ; therefore , we expect both scaling relations to yield @xmath8 relations with similar level of scatter . if the expected equivalence between @xmath94 and @xmath57 relations hold , we can then extend the @xmath8 relation to gravitational lens samples which span a higher dynamical range in the total mass at various redshifts . unfortunately , however , the trends observed in figures [ fig : nmtot ] and [ fig : frommn ] are very different from the strong positive correlations shown in figures [ fig : sigmamtot ] and [ fig : frommsigma ] . from a spearman s rank test , we obtain a weak negative @xmath89 correlation ( @xmath95 for degrees of freedom = 41 ) for the trend observed in figure [ fig : nmtot ] with only a @xmath96 confidence level that the correlation has not occurred by chance . a spearman s rank test performed on the @xmath8 relation shows similar statistical properties to that of the @xmath89 relation . before concluding if the correlation observed in figure [ fig : frommsigma ] is a better representation of a @xmath8 relation , we investigate various possibilities that may cause the inconsistency observed between figures [ fig : frommsigma ] and [ fig : frommn ] . first , we attempted to reproduce the @xmath2 relation , obtained by @xcite , for the slacs lens sample . in figure [ fig : graham ] , we plot @xmath97 values for the slacs lenses , versus the corresponding s@xmath1rsic indices from the bulge - disk decomposition . the log - quadratic @xmath2 relation by @xcite is overlaid on the slacs data for comparison . figure [ fig : graham ] clearly indicates that the log - quadratic @xmath98 relation can not be reproduced with the @xmath97 and @xmath90 values for the slacs lens sample . a weak but different correlation between @xmath97 and @xmath24 is apparent from figure [ fig : graham ] . we perform a spearman s rank test and derive a weak , negative correlation , @xmath99 at the @xmath100 significance level , between these quantities . the correlation observed in figure [ fig : graham ] is in contrast to the log - linear or log - quadratic @xmath25 relation discussed in literature @xcite . previous studies of morphology and scaling relations of bulge dominated galaxies claim that @xmath24 scales monotonically with the galaxy bulge magnitude ( equivalently bulge luminosity ) @xcite . if this correlation holds , we expect @xmath24 to correlate positively with @xmath5 due the dependence between @xmath5 and bulge luminosity ( @xmath10 ) @xcite . since figure [ fig : graham ] contradicts the existence of a log - quadratic @xmath25 relation , we also test the correlation between @xmath24 and i - band bulge magnitude ( @xmath101 ) for the slacs lens sample . the results of gim2d bulge - disk decomposition of slacs galaxies are shown in figure [ fig : magvsn ] . a spearman s rank test to determine degree of correlation between @xmath101 and @xmath24 shows that there is only a very weak correlation between the two quantities ( @xmath102 ) ; furthermore , the rank correlation we observe between @xmath101 and @xmath24 is significant at a level larger than 5% . although we do not observe a direct correlation between galaxy bulge magnitude and the s@xmath1rsic index , our results are limited by the narrow range of magnitudes ( @xmath103 ) in the slacs lens sample . furthermore , our results are consistent with those of @xcite , who also find that @xmath24 for all slacs lenses is uncorrelated with the measurements of lensing mass , dynamical mass , luminosity and velocity dispersion of the sample . given the trend observed in figure [ fig : magvsn ] , we can not confidently expect @xmath5 to correlate with @xmath24 as claimed by previous studies @xcite . the possible origin of @xmath25 relation has not been examined theoretically and is beyond the scope of this paper . for the remainder of this study , we use the @xmath3 relation to determine the black hole masses . the correlation observed between @xmath5 and @xmath0 , in figure [ fig : frommsigma ] , is in remarkable agreement with theoretical predictions @xcite and local observations @xcite of the @xmath8 relation . in the following section , we compare our results to several black hole formation scenarios posed by various theoretical studies and examine the implications of our findings . equation [ eq : observational ] , which quantifies the trend observed in figure [ fig : frommsigma ] , shows that @xmath5 scales non - linearly with @xmath0 and that efficiency of black hole formation increases with total mass . theoretical models that reproduce the observed luminosity function of high - redshift quasars @xcite predict that @xmath5 scales as a power law of the circular velocity of the galactic halo ( denoted as @xmath105 and also referred to as virial velocity ) in which the black hole resides : @xmath106 the @xmath107 relation shown above can be converted into an equivalent @xmath8 relation by considering the dependence between @xmath105 and halo mass ( equivalent to @xmath0 in this study ) , @xmath108 . the resulting correlation between @xmath5 and @xmath109 is as follows : @xmath110 the circular velocity of a given halo mass is redshift - dependent ; therefore , an important aspect of this analytical prediction is the evolution of the @xmath111 relation with time @xcite . the slope of the @xmath111 relation , @xmath112 , is a valuable indicator of various formation scenarios which result in observed black hole populations . in the process of hierarchical mass assembly , formation of smbh is driven by mergers of galaxy haloes . a linear relation between @xmath5 and @xmath109 , where @xmath113 , results from a formation scenario where the black holes residing in the merging haloes coalesce without additional gas accretion . a slope of @xmath114 is characteristic of a merger where the growth of the resulting black hole is dominated by an accretion process where a significant gas fraction from the merger product is driven in to the central accreting region @xcite . within this formation scenario , the fraction of baryons accreted on to the central regions that feed the black hole is significantly larger for more massive haloes , due to a deeper potential well ; therefore , massive haloes host larger smbh . in figure [ fig : combinedevolution ] , we plot two analytical predictions for the @xmath8 relation @xcite , which correspond to two formation scenarios , for three different epochs ( @xmath115 ) . both types of evolutionary tracks indicate that @xmath5 increases with the total mass of the host galaxy at any given redshift and that a galaxy of given total mass hosts a more massive black hole at a higher redshift relative to a lower redshift . within the context of hierarchical mass assembly in @xmath116 cosmology , the decrease in the growth of @xmath5 relative to @xmath0 at lower redshifts can be caused by processes such as a decrease in merger rates and the gas fraction that is available to fuel the central smbh . another important physical process involved in the evolution of @xmath8 relation is the feed - back regulated growth of smbh @xcite . a black hole shines at a fraction , @xmath117 , of its eddington luminosity ( @xmath118 ) following a merger and returns a fraction of this energy into the surrounding galactic gas . a black hole shining at its limiting @xmath118 can unbind the surrounding galactic gas if the energy liberated from the black hole is sufficient to overcome the gravitational binding energy of the gas . as the mass of a black hole increases through merger driven processes discussed above , the energy output can approach the limit where it is sufficient to unbind the entire galactic gas . the unbound galactic gas that escapes into the halo is heated beyond the virial temperature and can not cool during the dynamical time of the quasar ; therefore , this mechanism can eventually terminate the accretion process that feeds the central black hole . the growth of a smbh via a merger - driven , feed - back regulated mechanism implies a relation of @xmath119 ( the required rate of energy deposition to unbind a self - gravitating system is proportional to @xmath120 ) @xcite , leading to a slope of @xmath121 in the @xmath8 relation . evolutionary tracks shown in solid blue lines in figure [ fig : combinedevolution ] @xcite are examples of the formation scenario described above . the evolutionary tracks with a shallower slope , similar to the dashed lines shown in figure [ fig : combinedevolution ] @xcite , may be indicative of a modified feed - back regulated growth mechanism . if the galactic gas heated by the energy output of the black hole cools before the black hole reaches its critical eddington limit , additional energy ( up to a factor of @xmath122 ) would be required to unbind the cool gas component . this formation scenario leads to dependence of @xmath123 , resulting in a slope of @xmath124 in the @xmath8 relation . it is worth noting that @xcite assumes the @xmath3 relation by @xcite , @xmath125 , which yields a relation of @xmath126 . therefore , we recompute the results of @xcite using the @xmath3 relation used throughout this study @xcite , which gives a dependence of @xmath127 . within the context of formation and evolution of smbh , it is extremely important to accurately determine the parameters that quantify the slope and the evolution of @xmath8 relation . these parameters provide significant insight into the dominant formation scenarios that lead to the observed black hole populations at various redshifts . due to the narrow range of redshifts in the slacs lens sample , the @xmath8 relation we derive can be considered as an evolutionary track for the mean redshift of the slacs sample ( @xmath128 ) , as shown in figure [ fig : combinedevolution ] ( solid black line ) . the @xmath8 relation we derive from the slacs lens sample provides a unique opportunity to compare the theoretical tracks with observational evidence of the @xmath8 relation . this is shown in figure [ fig : comparisonevolution ] , where we compare the observational @xmath8 relation to the theoretical predictions of the @xmath8 relation at @xmath129 . the dashed region indicates the upper and lower 1-@xmath130 bounds of the observational form of the @xmath8 relation found in this study . inspection of figure [ fig : comparisonevolution ] shows that our results are in excellent agreement with analytical predictions of feed - back regulated growth . however , from our results , it is difficult to distinguish the importance of gas cooling in the black hole formation process . the @xmath8 relation found in this study is strongly suggestive that halo properties determine those of the galaxy and its black hole . this link is also observed in the fundamental plane ( fp ) , the two - dimensional projection of the three - dimensional space defined by the quantities of surface brightness ( @xmath131 ) , effective radius ( @xmath62 ) and central velocity dispersion ( @xmath132 ) , of early - type galaxies @xcite . results of @xcite show that a sample of 53 slacs lenses define a fp that is consistent with the general population of early - type galaxies from sdss : @xmath133 the ratio @xmath134 for the slacs lenses is @xmath135 @xcite , indicative of a universal isothermal mass profile ( also known as the bulge - halo conspiracy " ) . therefore , we can replace the quantities @xmath62 and @xmath132 in the fp with @xmath4 ( given the trend observed between the quantities @xmath136 and @xmath137 ) and @xmath138 . combining this with equations [ eq : totalmass ] and [ eq : msigma ] , we derive : @xmath139 ignoring the dependence of equation [ eq : slacsfp2 ] on @xmath131 , since it only varies weakly relative to the other variables , we can extract the connection between @xmath5 and @xmath0 from the fp of slacs lenses . the slope of 1.30 of the @xmath8 relation derived from the slacs fp @xcite is in agreement with our results within the 1-@xmath130 bounds . the manifestation of a @xmath8 relation within the fp of the slacs lenses further strengthens the existence of the observed @xmath8 relation . we require @xmath140 measurements of both @xmath5 and @xmath0 to fully compare the manifestation of @xmath8 relation within the fp and its implications for regularity of early - type galaxy formation scenarios . an additional advantage of the results from this study lies within the method used to derive @xmath0 . the traditional method of estimating total gravitational mass of a galaxy , used in black hole studies , is to convert the @xmath141 circular velocity ( from galaxy rotation curves ) into the velocity of the galactic halo ( also referred to as the @xmath142 velocity ) . the inferred total mass of a galaxy differs depending on the method used to relate observed circular velocity to the virial velocity . the effect on the @xmath8 relation from varying assumptions of the connection between observed circular velocity ( @xmath13 ) and virial velocity ( @xmath143 ) , discussed in @xcite , is also shown in figure [ fig : frommsigma ] . the dashed line shows the best - fit @xmath8 relation obtained by @xcite using cosmological prescriptions of @xcite to relate @xmath13 and @xmath143 ( @xmath144 ) and the dot - dashed line shows the resulting relation where @xmath145 ( @xmath146 ) . from the relations obtained by @xcite , it is evident that the slope of the @xmath8 is affected by the method assumed to relate @xmath13 and @xmath143 . gravitational lens modeling is independent of such dynamical assumptions and provides an elegant alternative method to determine the total mass . therefore , the @xmath8 relation characterized by this independent method is not only a complementary comparison to existing observational evidence but also provides valuable insight to determine the dominant physical processes of smbh growth . we use a sample of 43 early - type galaxies , which exhibit galaxy - scale strong gravitational lensing , to derive the scaling relation between black hole mass , @xmath5 , and the total mass , @xmath0 , of the host galaxy . in this study , we use gravitational lens modeling to directly measure the total mass and the mass profile of the galaxy rather than converting the observed circular velocity into a total mass , the traditional method that is used in black hole studies . we use two alternative scaling relations , @xmath3 and @xmath25 to estimate the black hole masses of the lens sample . we obtain a tight correlation between @xmath9 and @xmath0 in the log - log space . in conjunction with the @xmath3 relation , we derive the observational form of the @xmath8 relation that is consistent with no intrinsic scatter . we do not find a significant correlation between @xmath90 and @xmath0 . from a variety of tests , we find that we can not confidently estimate black hole masses with the @xmath98 relation . the scaling relation we observe between @xmath5 and @xmath0 is non - linear and is in agreement with theoretical predictions of the growth of black holes and observational studies of the local @xmath8 relation . the observed @xmath8 relation is also consistent with the studies of the fundamental plane of slacs lenses @xcite , which is suggestive of a unified scenario where the properties of the host halo determine those of the resulting galaxy and black hole formed through hierarchical merging . the observed non - linear correlation between @xmath5 and @xmath0 indicates that massive halos are more efficient in forming black holes and the slope of the @xmath8 relation is suggestive of a merger - driven , feed - back regulated process for the growth of black holes . we wish to thank laura ferrarese , chien peng , christopher pritchet , jon willis , and the members of the slacs team for valuable discussions and useful comments . we also wish to thank the anonymous referee whose helpful comments improved this paper . this research used the facilities of the canadian astronomy data centre operated by the national research council of canada with the support of the canadian space agency . the authors gratefully acknowledge the support from the national research council of canada and nserc through discovery grants . adams , f.c . , graff , d.s . , mbonye , m. , & richstone , d.o . 2003 , , 591 , 125 adams , f.c . , graff , d.s . , & richstone , d.o . 2001 , , 551 , l31 baes , m. , buyle , p. , hau , g.k.t . , & dejonghe , h. 2003 , , 341 , l44 bertin , e. , & arnouts , s. 1996 , , 117 , 393 bolton , a. , burles , s. , koopmans , l.v.e . , treu , t. , gavazzi , r. , moustaka , l.a . , wayth , r. , & + schlegel , d.j . 2008 , , 682 , 946 bolton , a. , treu , t. , koopmans , l.v.e . , gavazzi , r. , moustakas , l.a . , burles , s. , schlegel , d.j . , & wayth , r. 2008 , , 684 , 248 bolton , a. , burles , s. , koopmans , l.v.e . , treu , t. , & moustakas , l.a . 2006 , , 638 , 703 bullock , j.s . , kolatt , t.s . , sigad , y. , somerville , r.s . , kravtsov , a.v . , klypin , a.a . , primack , j.r . , & dekel , a. 2001 , , 321 , 559 buyle , p. , ferrarese , l. , gentile , g. , dejonghe , h. , baes , m. , & klein , u. 2006 , , 373 , 700 cattaneo , a. 2001 , , 324 , 128 cattaneo , a. , haehnelt , m.g . , rees , m.j . 1999 , , 308 , 77 courteau , s. , mcdonald , m. , widrow , l.m . , & holtzman , j. 2007 , , 655 , l21 croton , d.j . 2009 , , 394 , 1109 croton , d.j . , 2006 , , 365 , 11 di matteo , t. , springel , v. , & hernquist , l. 2005 , , 433 , 604 di matteo , t. , croft , r.a.c . , springel , v. , & hernquist , l. 2003 , , 593 , 56 djorgovski , s. , & davis , m. 1987 , , 313 , 59 dye , s. & warren , s. 2007 , arxiv:0708.0787 eisenstein , d.j . , et al . 2001 , , 122 , 2267 el - zant , a.a . , shlosman , i. , begelman , m.c . , & frank , j. 2003 , , 590 , 641 ferrarese , l. , et al . 2006 , , 164 , 334f ferrarese , l. , & ford , h. 2005 , , 116 , 523 ferrarese , l. 2002 , , 578 , 90 ferrarese , l. , & merritt , d. 2000 , , 539 , l9 fisher , d.b . , & drory , n. 2008 , , 136 , 773 gavazzi , r. , treu , t. , koopmans , l.v.e . , bolton , a.s . , moustakas , l.a . , burles , s. , & + marshall , p.j . 2008 , , 677 , 1046 gavazzi , r. , treu , t. , rhodes , j.d . , koopmans , l.v.e . , bolton , a.s . , burles , s. , massey , r.j . , & + moustakas , l.a . 2007 , , 667 , 176 gebhardt , k. , et al . 2000 , , 539 , l13 graham , a.w . , & driver , s.p . 2007 , , 655 , 77 graham , a.w . , & guzm@xmath147n , r. 2003 , , 125 , 2936 graham , a.w . , erwin , p. , caon , n. , & trujillo , i. 2001 , , 563 , l11 g@xmath148ltekin , k. , et al . 2009 , , 698 , 198 haehnelt , m.g . , & kauffmann , g. 2000 , , 318 , l35 haehnelt , m.g . , natarajan , p. , & rees , m.j . 1998 , , 300 , 817 halkola , a. , hildebrandt , h. , schrabback , t. , lombardi , m. , brada , m. , erben , t. , schneider , p. , & wuttke , d. 2008 , , 481 , 65 hopkins , p.f . , hernquist , l. , cox , t.j . , di matteo , t. , robertson , b. , & springel , v. 2005 , , 630 , 716 hopkins , p.f . , hernquist , l. , cox , t.j . , di matteo , t. , robertson , b. , & springel , v. 2005 , , 630 , 705 jee , m.j . , blakeslee , j.p . , sirianni , m. , martel , a.r . , white , r.l . , & ford , h.c . 2007 , , 119 , 1403j jerjen , h. , binggeli , b. , & freeman , k.c . 2000 , , 119 , 593 j@xmath149rgensen , i. , franx , m. , & kj@xmath150rgaard , p. 1995 , , 276 , 1341 koopmans , l.v.e . , treu , t. , bolton , a.s . , burles , s. , & moustakas , l.a . 2006 , , 649 , 599 kormendy , j. , & richstone , d. 1995 , , 33 , 581 krist , j. 1993 , , 52 , 536 lauer , t.r . , et al . 2007 , , 664 , 226 marconi , a. , & hunt , l.k . 2003 , , 589 , l21 mclure , r.j . , jarvis , m.j . , targett , t.a . , dunlop , j.s . , & best , p.n . 2006 , , 368 , 1395 moustakas , l.a . , 2007 , , 660 , l31 monaco , p. , salucci , p. , & danese , l. 2000 , , 311 , 279 pizzella , a. , corsini , e.m . , bont , sarzi , m. , coccato , l. , & bertola , f. 2005 , , 631 , 785 press , w.h . , flannery , b.p . , teukolsky , s.a . , & vetterling , w.t . 1992 , numerical recipes in fortran ( cambridge : cambridge univ . press ) robertson , b. , hernquist , l. , cox , t.j . , di matteo , t. , hopkins , p.f . , martini , p. , & springel , v. 2006 , , 641 , 90 s@xmath1rsic , j.l . 1968 , atlas de galaxies australes ( cordoba : observatorio astronomica ) silk , j. , & rees , m.j . 1998 , , 331 , l1 simard , l. , et al . 2002 , , 142 , 1 springel , v. , di matteo , t. , & hernquist , l. 2005 , , 620 , l79 springel , v. di matteo , t. , & hernquist , l. 2005 , , 361 , 776 strauss , m.a . , et al . 2002 , , 124 , 1810 tremaine , s. , et al . 2002 , , 574 , 740 treu , t. , gavazzi , r. , gorecki , a. , marshall , p.j . , koopmans , l.v.e . , bolton , a.s . , moustakas , l.a . , & burles , s. 2009 , , 690 , 670 treu , t. , koopmans , l.v . , bolton , a.s . , burles , s. , & moustakas , l.a . 2006 , , 640 , 662 van dokkum , p. 2001 , , 113 , 1420 volonteri , m. , haardt , f. , & madau , p. 2003 , . 582 , 559 weiner , b.j . 2006 , , 653 , 1049 wyithe , j.s.b . , & loeb , a. 2003 , , 595 , 614 wyithe , j.s.b . , & loeb , a. 2002 , , 581 , 886 wyithe , j.s.b . , & padmanabhan , t. 2006 , , 372 , 1681 rsic bulge + exponential disk ) convolved with a multidrizzled tiny - tim psf ( panel 2 ) , natural psf ( panel 3 ) and acs - library generated psf ( panel 4 ) is subtracted from the observed image . residual image where the galaxy model is convolved with a tiny - tim psf shows artifacts such as boxy - core feature , due to the finite size of the tiny - tim psf , evident in panel 2 . statistics of the pixels in panel 2 show that the mean value of the core residuals is comparable to the background level ; therefore , the artifact introduced by the finite tiny - tim psf does not significantly affect the quality of the overall fit . in general , we find that both natural and acs - library generated psfs produce minimal core residuals . however , most slacs fields are relatively devoid of stars , suitable for convolution ; therefore , we use acs - library generated psf for gim2d galaxy modeling . ] ) , i - band magnitude of the bulge ( @xmath63 ) and aperture corrected bulge velocity dispersion ( @xmath130 ) . the scaling relations between bulge structural parameters for sdss early type galaxies are constructed from a sample 77523 galaxies and are shown in gray - scale . the gray - scale is scaled to 20% of the peak value of the central distribution to show that the structural parameters of the slacs lenses lie well within the distribution that encompasses the sdss galaxies . the outlier in this figure is sdss j0959 + 0410 . ] and @xmath0 , for the sample of early type slacs lenses . velocity dispersion values from sdss pipeline are corrected to an aperture of radius equal to @xmath151 using the empirical relation @xcite . the error bars correspond to the 1-@xmath130 error of the aperture corrected velocity dispersion , from gaussian error propagation , taking the measurement errors of sdss velocity dispersions into account . total masses of the slacs lenses are derived from strong lens modeling parameters of @xcite . the 1-@xmath130 errors of @xmath0 , which are smaller than the data points , are incorporated into the fitting routines but not shown in this plot . the solid line correspond to the best - fit @xmath72 relation for all slacs lenses in the logarithmic space . ] with @xmath9 converted into @xmath5 using @xmath8 relation @xcite . the error bars correspond to 1-@xmath50 error on the quantity @xmath152 . the solid line corresponds to the @xmath8 relation for all slacs lenses derived using the @xmath72 and @xmath8 relations . the dashed line represents the resulting @xmath153 relation obtained by @xcite , where @xmath0 is computed from cosmological simulations @xcite relating the the observed circular velocity ( @xmath154 ) to the virial velocity of the host halo ( @xmath155 ) . the dotted line shows the resulting @xmath153 relation if @xmath0 is computed using @xmath156 @xcite . ] and @xmath0 , for the sample of early type slacs lenses . the s@xmath1rsic indices are derived from the best - fit bulge+disk decomposition of the slacs lenses . the error bars correspond to the measurement errors on the quantity @xmath90 . total masses of the slacs lenses are derived from strong lens modeling parameters of @xcite . the 1-@xmath50 errors of @xmath0 , which are smaller than the data points , are incorporated into the fitting routines but not shown in this plot . ] with @xmath90 converted into @xmath5 using @xmath25 relation @xcite . the error bars correspond to 1-@xmath50 error on the quantity @xmath152.the 1-@xmath50 error of @xmath0 is smaller than the data points , and is not shown in the plot . ] relation to slacs data . @xmath5 values are derived from the @xmath3 relation and s@xmath1rsic indices are best - fit results from the gim2d bulge - disk decomposition . error bar on the legend represent the typical 1-@xmath50 uncertainty of @xmath157 , derived from gaussian error propagation using measurement errors from the sdss velocity dispersions ( @xmath158 ) , and a measurement error of 20% on the quantity @xmath24 . the log - quadratic relation by @xcite is shown by the solid black line . ] versus @xmath159 for slacs lens sample . the values of @xmath90 are the best fit @xmath160 s@xmath1rsic indices from the bulge+disk decomposition using gim2d . the values of @xmath101 are the best - fit , extinction - corrected , i - band @xmath160 magnitudes from the bulge+disk decomposition . ] relation from various analytical predictions for @xmath161 . the pale blue , solid lines are predictions from @xcite , where @xmath162 . the dashed lines are predictions from @xcite , where @xmath127 , for the same epochs as @xcite ( epochs increasing in the same order as shown in the labels for @xcite ) . the thick , solid black line is a comparison of the observed @xmath8 relation from the slacs lens sample ( equation [ eq : observational ] ) , at @xmath128 . ] relation to theoretical predictions of the @xmath8 relation at @xmath129 , the mean redshift of the slacs lens sample . the solid black line correspond to the best - fit result from this study and the shaded region shows the 1-@xmath50 upper and lower limits of the @xmath8 relation . the solid red and blue lines correspond to the theoretical predictions of @xcite and @xcite of the @xmath8 relation respectively . ] lccccccc sdss j0029 - 0055 & 0.23 & 0.76 & @xmath163 & 7.29@xmath164 & @xmath165 & 2.65@xmath164 & 1.04@xmath166 + sdss j0037 - 0942 & 0.20 & 0.94 & @xmath167 & 3.38@xmath164 & @xmath168 & 5.69@xmath164 & 2.40@xmath166 + sdss j0216 - 0812 & 0.33 & 0.43 & @xmath169 & 8.50@xmath170 & @xmath171 & 1.52@xmath172 & 4.03@xmath166 + sdss j0252 + 0039 & 0.28 & 0.58 & @xmath173 & 1.19@xmath172 & @xmath174 & 6.48@xmath170 & 1.27@xmath166 + sdss j0330 - 0020 & 0.35 & 0.88 & @xmath175 & 9.65@xmath164 & @xmath176 & 1.88@xmath164 & 1.51@xmath166 + sdss j0728 + 3835 & 0.21 & 0.95 & @xmath177 & 7.21@xmath164 & @xmath178 & 1.90@xmath164 & 1.73@xmath166 + sdss j0737 + 3216 & 0.32 & 0.85 & @xmath179 & 6.70@xmath164 & @xmath180 & 1.28@xmath172 & 2.41@xmath166 + sdss j0822 + 2652 & 0.24 & 0.81 & @xmath181 & 4.41@xmath164 & @xmath182 & 4.67@xmath164 & 2.00@xmath166 + sdss j0912 + 0029 & 0.16 & 1.00 & @xmath183 & 1.35@xmath164 & @xmath184 & 1.09@xmath172 & 4.36@xmath166 + sdss j0935 - 0003 & 0.35 & 0.27 & @xmath185 & 6.89@xmath170 & @xmath186 & 3.33@xmath172 & 4.46@xmath166 + sdss j0936 + 0913 & 0.19 & 0.79 & @xmath187 & 6.27@xmath164 & @xmath188 & 3.42@xmath164 & 1.48@xmath166 + sdss j0946 + 1006 & 0.22 & 0.40 & @xmath189 & 2.61@xmath170 & @xmath190 & 5.91@xmath164 & 2.32@xmath166 + sdss j0955 + 0101 & 0.11 & 0.77 & @xmath191 & 1.05@xmath172 & @xmath192 & 1.14@xmath164 & 1.21@xmath166 + sdss j0956 + 5100 & 0.24 & 0.90 & @xmath193 & 9.48@xmath164 & @xmath194 & 1.11@xmath172 & 3.24@xmath166 + sdss j0959 + 0410 & 0.13 & 0.14 & @xmath195 & 2.10@xmath170 & @xmath196 & 2.25@xmath164 & 1.08@xmath166 + sdss j0959 + 4416 & 0.24 & 0.64 & @xmath197 & 1.87@xmath164 & @xmath198 & 3.99@xmath164 & 1.65@xmath166 + sdss j1020 + 1122 & 0.28 & 0.58 & @xmath199 & 3.20@xmath164 & @xmath200 & 7.16@xmath164 & 2.76@xmath166 + sdss j1029 + 0420 & 0.10 & 0.84 & @xmath201 & 6.24@xmath164 & @xmath202 & 1.81@xmath164 & 9.82@xmath203 + sdss j1106 + 5228 & 0.10 & 0.79 & @xmath204 & 5.90@xmath164 & @xmath205 & 4.94@xmath164 & 1.49@xmath166 + sdss j1112 + 0826 & 0.27 & 0.67 & @xmath206 & 5.63@xmath170 & @xmath207 & 1.32@xmath172 & 3.08@xmath166 + sdss j1134 + 6027 & 0.15 & 0.53 & @xmath208 & 1.55@xmath164 & @xmath209 & 3.80@xmath164 & 1.51@xmath166 + sdss j1142 + 1001 & 0.22 & 0.48 & @xmath210 & 1.98@xmath164 & @xmath211 & 2.81@xmath164 & 1.68@xmath166 + sdss j1143 - 0144 & 0.11 & 0.90 & @xmath212 & 1.46@xmath164 & @xmath213 & 4.70@xmath164 & 2.52@xmath166 + sdss j1204 + 0358 & 0.16 & 0.72 & @xmath214 & 4.37@xmath164 & @xmath215 & 5.78@xmath164 & 1.72@xmath166 + sdss j1205 + 4910 & 0.22 & 0.78 & @xmath216 & 5.66@xmath164 & @xmath217 & 6.17@xmath164 & 2.37@xmath166 + sdss j1213 + 6708 & 0.12 & 0.89 & @xmath218 & 1.19@xmath172 & @xmath219 & 5.76@xmath164 & 1.70@xmath166 + sdss j1250 + 0523 & 0.23 & 0.70 & @xmath191 & 1.05@xmath172 & @xmath220 & 3.89@xmath164 & 1.46@xmath166 + sdss j1402 + 6321 & 0.20 & 1.00 & @xmath221 & 6.38@xmath164 & @xmath222 & 4.40@xmath164 & 2.61@xmath166 + sdss j1403 + 0006 & 0.19 & 0.66 & @xmath223 & 1.18@xmath172 & @xmath224 & 1.74@xmath164 & 1.18@xmath166 + sdss j1416 + 5136 & 0.30 & 0.55 & @xmath225 & 6.07@xmath170 & @xmath226 & 4.04@xmath164 & 2.31@xmath166 + sdss j1420 + 6019 & 0.06 & 0.75 & @xmath227 & 1.29@xmath164 & @xmath228 & 1.74@xmath164 & 9.39@xmath203 + sdss j1430 + 4105 & 0.29 & 0.74 & @xmath229 & 3.00@xmath164 & @xmath230 & 1.18@xmath172 & 3.76@xmath166 + sdss j1443 + 0304 & 0.13 & 0.89 & @xmath231 & 1.17@xmath172 & @xmath232 & 1.74@xmath164 & 9.46@xmath203 + sdss j1451 - 0239 & 0.13 & 0.93 & @xmath233 & 5.46@xmath164 & @xmath234 & 2.21@xmath164 & 1.17@xmath166 + sdss j1525 + 3327 & 0.36 & 0.58 & @xmath235 & 3.23@xmath164 & @xmath236 & 5.06@xmath164 & 3.03@xmath166 + sdss j1531 - 0105 & 0.16 & 1.00 & @xmath237 & 5.37@xmath164 & @xmath238 & 5.55@xmath164 & 2.35@xmath166 + sdss j1538 + 5817 & 0.14 & 0.86 & @xmath239 & 1.12@xmath172 & @xmath240 & 1.06@xmath164 & 1.17@xmath166 + sdss j1627 - 0053 & 0.21 & 0.67 & @xmath241 & 3.27@xmath164 & @xmath242 & 7.91@xmath164 & 2.11@xmath166 + sdss j1630 + 4520 & 0.25 & 0.86 & @xmath243 & 2.68@xmath164 & @xmath244 & 6.15@xmath164 & 3.02@xmath166 + sdss j1636 + 4707 & 0.23 & 0.64 & @xmath245 & 1.33@xmath164 & @xmath246 & 3.23@xmath164 & 1.53@xmath166 + sdss j2238 - 0754 & 0.14 & 0.80 & @xmath247 & 3.41@xmath164 & @xmath248 & 1.47@xmath164 & 1.44@xmath166 + sdss j2303 + 1422 & 0.16 & 0.97 & @xmath249 & 2.71@xmath164 & @xmath250 & 3.83@xmath164 & 2.57@xmath166 + sdss j2341 + 0000 & 0.19 & 0.52 & @xmath251 & 9.02@xmath164 & @xmath252 & 1.44@xmath164 & 1.87@xmath166

we present initial results from the first systematic survey of luminous quasars . quasars at 5.5 , the post - reionization epoch , are crucial tools to explore the evolution of intergalactic medium , quasar evolution and the early super - massive black hole growth . however , it has been very challenging to select quasars at redshifts 5.3 5.7 using conventional color selections , due to their similar optical colors to late - type stars , especially m dwarfs , resulting in a glaring redshift gap in quasar redshift distributions . we develop a new selection technique for 5.5 quasars based on optical , near - ir and mid - ir photometric data from sloan digital sky survey ( sdss ) , ukirt infrared deep sky surveys - large area survey ( ulas ) , vista hemisphere survey ( vhs ) and wide field infrared survey explorer ( wise ) . from our pilot observations in sdss - ulas / vhs area , we have discovered 15 new quasars at 5.3 5.7 and 6 new lower redshift quasars , with sdss z band magnitude brighter than 20.5 . including other two 5.5 quasars already published in our previous work , we now construct an uniform quasar sample at 5.3 5.7 with 17 quasars in a 4800 square degree survey area . for further application in a larger survey area , we apply our selection pipeline to do a test selection by using the new wide field j band photometric data from a preliminary version of the ukirt hemisphere survey ( uhs ) . we successfully discover the first uhs selected 5.5 quasar .

1703.03526

high - redshift ( z @xmath4 5 ) quasars are important tracers to study the early universe . however , they are difficult to be found due to both low spatial density and high contaminants from cool dwarfs when using color selection . although more than 300,000 quasars are now known ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , only @xmath3 290 quasars are at @xmath5 5 . in the distribution of quasar redshift , there is an obvious gap of known quasars at 5.3 @xmath6 5.7 , due to their similar optical colors to that of late - type stars ( see the redshift distribution in sec . 4 ) . only @xmath3 30 known quasars have been found in this redshift gap over a wide magnitude range ( 17.5 @xmath6 26 mag ) ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? compared to the studies at lower redshift and higher redshift , this gap posts significant limit on the study of quasar evolution from @xmath1 5 to 6 , over the post - reionization epoch . observations of the gunn - peterson effect using absorption spectra of high redshift quasars suggest that reionization is just completing at @xmath7 6 , possibly with a tail to @xmath7 5.5 @xcite . therefore , the physical conditions of the post reionization igm , at @xmath1 5 - 6 , provides the basic boundary conditions of models of reionization , such as the evolution of igm temperature , photon mean free path , metallicity and the impact of helium reionization ( e.g. * ? ? ? they place strong constraints on reionization topology as well as on the sources of reionization and chemical feedback by early galaxy population . ly@xmath8 opacity measurement directly probes the evolution of igm . following @xcite , several new measurements about the ly@xmath8 opacity at 5 @xmath6 6 are given @xcite , but igm statistics are still poorly constrained at @xmath9 @xcite . moreover , a quasar sample in this redshift range is also a key to study the evolution of quasar luminosity function ( qlf ) and black hole growth . at high redshift , the qlf and black hole evolution have been measured at @xmath1 5 and 6 @xcite . however , at @xmath1 5.5 they are still poorly measured due to the lack of a complete quasar sample . @xcite derived the quasar spatial density at @xmath1 4 , 4.9 and 6 , and fitted a luminosity - dependent density evolution model to the combined dataset . they concluded that the quasar number density evolution steepens at high redshift , such that luminous quasars decline as a population more steeply at @xmath5 5 than from @xmath10 4 to @xmath10 5 ( also * ? ? ? however , the exact evolution of quasar density from @xmath11 = 5 to 6 is unclear because of the small size and high incompleteness of existing @xmath1 5.5 quasar sample . the quasar number density at @xmath12 is also needed to estimate the contribution of quasars to the ionizing background just after the reionization epoch . @xcite suggested that there was a rapid black hole mass growth phase after @xmath1 6 . study of black hole growth at @xmath1 4.8 supports that the notion of fast smbh growth at this epoch , corresponding to probably the first such phase for most smbh @xcite . studying bh growth properties at @xmath0 will fill in the missing link between @xmath13 and 6 . to answer the questions posted above , a large , uniformly selected sample of quasars at @xmath14 is needed . however , so far there has not ever been a complete quasar survey at @xmath0 . as shown in 2 , broad - band colors of @xmath0 quasars are very similar to those of much more numerous m dwarfs , when a small number of passbands are used . therefore , to avoid the lager number of star contaminations , previous quasar selections have always excluded the region of m dwarf locus in @xmath15 color - color diagram . as a result , most surveys of high - redshift quasars have avoided the color space occupied by @xmath12 quasars and are highly incomplete at this redshift . to construct a large uniform @xmath1 5.5 quasar sample , a more effective selection to separate quasars from m dwarf in this most contaminated region is required . in this paper , we report initial results from a new search that focuses on the selection of @xmath0 quasars . our new color selection criteria based on optical , near- and mid - ir colors have yielded 17 quasars in the redshift range of 5.3 @xmath2 5.7 during the pilot observation described here . our optical / ir color selection technique and candidate selection using a combination of existing and new imaging surveys are described in section 2 . the details of our spectroscopy observations and new discoveries are presented in section 3 and section 4 . in section 5 , we discuss the completeness of our new selection and also report a test selection and first discovery using the preliminary version of the ukirt hemisphere survey ( uhs ) photometric data . a summary is given in section 6 . in this paper , we adopt a @xmath16cdm cosmology with parameters @xmath17 = 0.728 , @xmath18 = 0.272 , @xmath19 = 0.0456 , and h@xmath20 = 70 @xmath21 @xcite . photometric data from the sloan digital sky survey ( sdss ) are in the sdss photometric system @xcite , which is almost identical to the ab system at bright magnitudes ; photometric data from ir surveys are in the vega system . all sdss data shown in this paper are corrected for galactic extinction . at @xmath1 5 , most quasars are undetectable in u - band and g - band because of the presence of lyman limit systems ( llss ) , which are optically thick to the continuum radiation from the quasar @xcite . meanwhile , lyman series absorption systems begin to dominate in the r - band and ly @xmath8 emission moves to the @xmath22band . the @xmath23/@xmath24 color - color diagram is often used to select @xmath11 @xmath3 5 quasar candidates in previous studies @xcite . at higher redshift , the @xmath24 color becomes redder and most @xmath5 5.1 quasars begin to enter the m dwarf locus in the @xmath15 color - color diagram , which makes it very difficult to select @xmath25 5.2 quasars only with optical colors , especially at @xmath1 5.5 , where quasars have essentially the same optical colors as m dwarfs ( see figure 1 ) . previous selections focused on the region in the right - bottom of the @xmath15 diagram . : the color track of quasar at @xmath11 = 5 to 6 ( red dots and line ) with a step of @xmath26 , generated by calculating the mean colors of simulated quasars at each redshift bin . the simulated quasar sample used here is the same sample described in sec . the contours show the locus of m dwarfs , from early type to late type . the cyan contours denote m1-m3 dwarfs , the orange contours denote m4-m6 dwarfs and the purple contours denote m7-m9 dwarfs . clearly , @xmath27 quasars are serious contaminated by late type m dwarfs . @xmath28 : the spectrum of an average of simulated @xmath1 5.5 quasars compared with the spectrum of a typical m5 dwarf . the dashed lines represent normalised sdss @xmath29 , @xmath30 and @xmath11 bandpasses from left to right . for the comparison , we scaled both the quasar spectrum and m dwarf spectrum to @xmath31 . in this case , the synthetic sdss @xmath29 and @xmath11 bands magnitudes of quasar are 22.1 and 18.8 . for m5 dwarf , the magnitudes are 21.8 in @xmath29 band and 18.9 in @xmath11 band . it is obvious that their @xmath23 and @xmath24 colors are too similar to be distinguished using optical colors alone.,title="fig:",scaledwidth=40.0% ] : the color track of quasar at @xmath11 = 5 to 6 ( red dots and line ) with a step of @xmath26 , generated by calculating the mean colors of simulated quasars at each redshift bin . the simulated quasar sample used here is the same sample described in sec . the contours show the locus of m dwarfs , from early type to late type . the cyan contours denote m1-m3 dwarfs , the orange contours denote m4-m6 dwarfs and the purple contours denote m7-m9 dwarfs . clearly , @xmath27 quasars are serious contaminated by late type m dwarfs . @xmath28 : the spectrum of an average of simulated @xmath1 5.5 quasars compared with the spectrum of a typical m5 dwarf . the dashed lines represent normalised sdss @xmath29 , @xmath30 and @xmath11 bandpasses from left to right . for the comparison , we scaled both the quasar spectrum and m dwarf spectrum to @xmath31 . in this case , the synthetic sdss @xmath29 and @xmath11 bands magnitudes of quasar are 22.1 and 18.8 . for m5 dwarf , the magnitudes are 21.8 in @xmath29 band and 18.9 in @xmath11 band . it is obvious that their @xmath23 and @xmath24 colors are too similar to be distinguished using optical colors alone.,title="fig:",scaledwidth=40.0% ] in figure 1 , we plot the quasar color track from @xmath11 = 5 to @xmath11 = 6 in the @xmath15 color - color diagram comparing with locus of m dwarfs and also show the comparison between spectra of @xmath1 5.5 quasar and a typical m5 dwarf . as shown , from earlier type to later type , m dwarfs show a redder @xmath23 color . m dwarfs earlier than m3 have a bluer @xmath23 color than quasars and can be more easily rejected . m dwarfs later than m4 have their continua peak at 8000 - 12000 @xmath32 @xcite and can seriously contaminate quasar selection . quasars at 5.3 @xmath2 5.7 have strong ly@xmath8 emission line at 7600 @xmath33 8150 @xmath32 and a power law continuum redward of ly@xmath8 emission with a average index of @xmath34 = @xmath330.5 . the locus of late type m dwarfs ( m4 @xmath33 m8 ) almost overlap the whole region of @xmath1 5.5 quasars . @xcite proposed a new color selection criteria for @xmath35 quasars , by adding photometric data from wide field infrared survey explorer ( wise ) and at the same time relaxing the @xmath15 color cuts . allwise w1-w2 color of allwise can separate @xmath25 5.1 quasars from m dwarfs due to the redder w1-w2 color of high redshift quasars than that of late - type stars . for quasars at @xmath1 5.5 , if we include the whole region overlapped by m dwarf locus on the @xmath15 color - color diagram , we will include a huge number of m dwarfs . in this case , although using w1-w2 color can help to reject a part of m dwarfs , those remained m dwarfs will still result in a high contamination rate . more colors are needed to further reject m dwarfs . nir photometry covering the wavelength range from 9000 @xmath32 to 2@xmath36 m ( @xmath37 bands ) will effectively distinguish @xmath7 5.5 quasars from late - type m dwarfs . the spectral energy distributions of @xmath38 quasars are mainly dominated by a power - law spectrum with a slope @xmath39 at the wavelength range from ly@xmath8 to h@xmath40 , which is flatter than that of m dwarfs , a grey body spectrum . so quasars have redder @xmath41 , @xmath42w1 and @xmath43w2 colors than that of m dwarfs . especially , at @xmath0 , the h@xmath8 and h@xmath40 emission lines in quasar spectra shift to w2 and w1 bands respectively . the @xmath42w1 , @xmath43w2 and even w1-w2 colors of quasars become more redder and distinguishable form m dwarfs . we thus add @xmath44-w1 , @xmath41 and @xmath45 w2 colors , together with the @xmath46 color - color diagram and w1-w2 color cut , to construct our new color selection criteria of @xmath0 quasars . figure 2 shows the loci of quasars and m dwarfs in various color - color diagrams used in our new selection . for quasars , we used synthetical @xmath47,@xmath48 , w1 and w2 bands photometric data from a sample of simulated quasar spectra at 5.3 @xmath25.7 . the details of this simulated quasar sample can be found in sec . we used the same photometric systems as sdss , the ukirt infrared deep sky surveys ( ukidss ) - large area survey ( ulas ) and wise . for m dwarfs , we used the spectroscopically identified m dwarfs from querying the sdss dr10 catalog , and obtained their nir photometric data from ulas and allwise . as shown , in the @xmath46 color - color diagram , m dwarfs locate in a similar region as @xmath1 5.5 quasars . on the other hand , on the @xmath42w1/w1@xmath33w2 and @xmath49/@xmath43w2 color diagrams , most of simulated quasars could be separated from stars due to their redder @xmath42w1 , w1@xmath33w2 , @xmath49 and @xmath43w2 colors . to improve the efficiency of selection and reduce the size of candidate sample , we restrict our selection to the cleanest regions ( purple dashed lines in fig . 2 ) , which are still able to include most quasars . we used the following photometric datasets to select candidates of @xmath0 quasars . in optical range , we used sdss dr10 photometry catalog which covered @xmath50 deg@xmath51 with @xmath52 , @xmath53 , @xmath29 , @xmath30 and @xmath11 bands . in near - infrared wavelengths , the published large area covered deep nir photometric database with @xmath54 , @xmath44 , @xmath55 and @xmath56(@xmath57 ) bands are ulas @xcite covering @xmath3 4000 @xmath58 in the northern sky and the vista hemisphere survey ( vhs ; * ? ? ? * ) covering the whole southern sky . to expand the available survey area , although vhs focus on the southern sky , we also did the selection in the overlap area between vhs dr3 and sdss , a @xmath3 800 square degree field . for w1 and w2 bands , we used allwise dataset . the allwise program combined photometric data from the wise cryogenic @xcite and neowise @xcite post - cryogenic survey phases and mapped the entire sky with w1 , w2 , w3 and w4 ( 3.4 , 4.6 , 12 , and 22 @xmath36 m ) bands . allwise have high detection completeness of know quasars over almost all redshifts @xcite . due to the shallower detections in w3 and w4 bands , we only used w1 and w2 data . therefore , we carried out a quasar survey for @xmath1 5.5 quasars based on sdss , ulas / vhs and allwise photometric data within a @xmath3 4800 @xmath58 field . the allwise detection completeness of high redshift quasars ( @xmath59 ) is 50% at z band magnitude @xmath3 20.5 @xcite and decreases rapidly towards fainter end . allwise w1 and w2 , especially w2 , are not deep enough to detect @xmath1 5.5 quasars with sdss @xmath11 band magnitude fainter than 20.5 . we thus limited our selection and main candidate sample with @xmath60 20.5 mag . we started the candidate selection from a catalog of candidates which met our optical cuts , and selected only pointed sources . we limited our selection area to galactic latitude b @xmath4 20@xmath61 or b @xmath62 due to the quickly increasing star contaminations at lower galactic latitude . the optical color selection criteria we used for candidate selection are summarised as following , also shown in figure 2 . here we did not limit the signal - to noise ratio ( @xmath63 ) in @xmath29 band since @xmath29 band may be the drop - out band . @xmath64 @xmath65 @xmath66 @xmath67 @xmath68 we limited @xmath15 colors and required drop - outs in @xmath52 and @xmath53 bands to ensure the redshift range of quasars . we then cross - matched optical selected objects with ulas dr10 & vhs dr3 and allwise data using a 2@xmath69 cross radius . we required objects to be detected in @xmath29 , @xmath30 , @xmath11 , @xmath44 , @xmath55 , @xmath56 , w1 and w2 bands . we selected sources which meet our sdss - ulas / vhs - allwise @xmath42w1/w1@xmath33w2 and @xmath49/@xmath43w2 color cuts . a @xmath70 color cut was added here for further rejection of stars . to ensure the w1 and w2 photometry quality , we limited the signal - to - noise ratio to be higher than 5 in w1 and 3 in w2 . we did not further limit the @xmath63 in @xmath37 bands as the photometry in these bands are deep enough . actually , all sources that met our selection had @xmath71 in @xmath72 bands . the nir color selections we used are listed as following ( also in figure 2 ) . @xmath73 @xmath74 @xmath75 @xmath76 @xmath77 @xmath78 we visually checked images and removed targets with suspicious detections , such as multiple peaked objects or being affected by bright stars . the effect on color selection from the difference between ulas and vhs photometry systems , especially in @xmath56(@xmath57 ) band , is much smaller than photometric uncertainties . thus we used the same color cuts for ulas and vhs detected objects . using this selection method , we selected @xmath3 1000 quasar candidates in total as our main sample . optical spectroscopy for the identification of @xmath7 5.5 quasar candidates were carried out using several facilities : the 6.5 m mmt telescope and the 2.3 m bok telescope in the u.s . , the 2.3 m anu telescope in australia , the lijiang 2.4 m telescope ( ljt ) in china . our observations started in the fall of 2014 . up to date , we have observed 93 candidates from our main sample based on their brightness , colors and positions . we have discovered 21 new quasars . three quasars , j0155 + 0415 , j2207@xmath330416 and j2225 + 0330 were also in our candidate list but had been observed and published earlier as @xmath1 5 candidates @xcite . all information of spectroscopic identifications of these 24 quasars is listed in table 1 . we observed about 50 candidates using the red channel spectrograph @xcite on the mmt 6.5 m telescope . we used the 270 @xmath79 grating centred at 7500 @xmath32 ( 8500 @xmath32 ) , providing coverage from @xmath3 5700 to 9300 @xmath32 ( @xmath3 6700 to 10300 @xmath32 ) . we used the @xmath80 or @xmath81 slit based on seeing condition , providing resolutions of @xmath82 and @xmath83 , respectively . we also used the wide field spectrograph ( wifes ; * ? ? ? * ; * ? ? ? * ) , an integral - field double - beam image - slicing spectrograph on the anu 2.3 m telescope at siding spring observatory , to observe seven of our quasar candidates . they were observed using grating r3000 on wifes which gives a resolution of @xmath84 at wavelengths between 5300@xmath32 and 9800 @xmath32 . the lijiang 2.4 m telescope is located at lijiang observatory , yunnan observatories , chinese academy of sciences ( cas ) . it is equipped with the yunnan faint object spectrograph and camera ( yfosc ) which can take spectra followed by photometric images with a very short switching time . we used grism 5 ( g5 ) , with dispersion of 185 @xmath32/mm and wavelength coverage from 5000 to 9800 @xmath32 . we used a @xmath85 slit which yields a resolution of @xmath86 . some candidates were observed using the boller and chivens spectrograph ( b&c ) on steward observatory s 2.3 m bok telescope at kitt peak with the g400 grating and 2@xmath875 slit which gives a resolution of @xmath88 and @xmath89 @xmath90 wavelength coverage . all spectra taken on the 2.4 m telescope , 2.3 m bok telescope , and mmt telescope were reduced using standard iraf routines . the wifes data were reduced with a python based pipeline pywifes @xcite . the flux of all spectra were calibrated using standard stars observed on the same night and then scaled to sdss @xmath30-bands magnitudes for absolute flux calibration . c c c l c c c c c j010806.60 + 071120.6 & 5.53 & 19.57 & sso2.3m / wifes & 1800.0 & r3000 & 1.0 & 2014 - 10 - 15 + j011353.75 + 055951.1 & 5.00 & 20.45 & sso2.3m / wifes & 1800.0 & r3000 & 1.0 & 2014 - 10 - 15 + j082933.10 + 250645.6 & 5.35 & 19.67 & mmt / red & 300.0 & g270 & 1.0 & 2015 - 03 - 14 + [email protected] & 5.32 & 20.14 & mmt / red & 250.0 & g270 & 1.5 & 2015 - 11 - 06 + j095712.20 + 101618.5 & 5.14 & 19.61 & ljt / yfosc & 3000.0 & g5 & 1.8 & 2015 - 02 - 18 + [email protected] & 5.55 & 20.00 & sso2.3m / wifes & 3600.0 & r3000 & 1.0 & 2015 - 05 - 15 + j102201.91 + 080122.2 & 5.30 & 19.12 & mmt / red & 300.0 & g270 & 1.5 & 2015 - 11 - 06 + j113308.78 + 160355.7 & 5.61 & 19.71 & mmt / red & 250.0 & g270 & 1.5 & 2015 - 11 - 06 + j113414.23 + 082853.3 & 5.69 & 20.31 & mmt / red & 600.0 & g270 & 1.5 & 2015 - 03 - 16 + [email protected] & 5.31 & 19.23 & mmt / red & 300.0 & g270 & 1.0 & 2015 - 05 - 10 + j114946.45 + 074850.6 & 5.66 & 20.30 & mmt / red & 900.0 & g270 & 1.0 & 2015 - 03 - 13 + [email protected] & 5.25 & 20.08 & sso2.3m / wifes & 3600.0 & r3000 & 1.0 & 2015 - 05 - 12 + j131929.23 + 151305.0 & 4.50 & 20.05 & mmt / red & 900.0 & g270 & 1.0 & 2015 - 03 - 13 + [email protected] & 5.67 & 18.89 & ljt / yfosc & 2500.0 & g5 & 1.8 & 2015 - 02 - 14 + j151339.64 + 085406.5 & 5.47 & 19.89 & mmt / red & 600.0 & g270 & 1.0 & 2015 - 05 - 09 + j152712.86 + 064121.9 & 5.57 & 19.95 & mmt / red & 300.0 & g270 & 1.0 & 2015 - 05 - 10 + [email protected] & 5.61 & 20.31 & mmt / red & 600.0 & g270 & 1.5 & 2015 - 11 - 06 + [email protected] & 5.22 & 19.14 & sso2.3m / wifes & 1200.0 & r3000 & 1.0 & 2015 - 07 - 20 + j233008.71 + 095743.7 & 5.30 & 19.78 & sso2.3m / wifes & 1800.0 & r3000 & 1.0 & 2015 - 08 - 15 + [email protected] & 5.25 & 19.61 & sso2.3m / wifes & 1500.0 & r3000 & 1.0 & 2015 - 07 - 20 + j235824.04 + 063437.4 & 5.32 & 19.54 & sso2.3m / wifes & 1800.0 & r3000 & 1.0 & 2014 - 10 - 16 + j015533.28 + 041506.7 & 5.37 & 19.26 & bok / b&c & 2400.0 & r400 & 2.5 & 2014 - 10 - 28 + [email protected] & 5.53 & 18.95 & ljt / yfosc & 2400.0 & g5 & 1.8 & 2014 - 10 - 22 + j222514.38 + 033012.5 & 5.24 & 19.47 & bok / b&c & 2400.0 & r400 & 2.5 & 2014 - 10 - 19 from our sdss - ulas / vhs - allwise selected candidate sample , we have observed 93 candidates and discovered 21 quasars . there are 15 new quasars in the redshift range of 5.3 @xmath2 5.7 . others are lower redshift quasars but all at redshift @xmath5 5 , except one broad absorption line quasar with @xmath91 4.50 . there are also 3 quasars in our target list which were already observed and published as @xmath1 5 candidates @xcite . two of them are z @xmath3 5.5 quasars , the other one is at @xmath10 5.24 . therefore , in the pilot observed sample , we get a @xmath3 25% selection success rate for quasars and @xmath3 18% for the redshift range of 5.3 @xmath6 5.7 . these quasars form an uniformly selected sample of @xmath1 5.5 quasar with 17 quasars in the magnitude limit of sdss @xmath10 20.5 . most of the other 72 observed candidates are m dwarfs . few of them can only be ruled out as quasars since there are no emission features but can not be identified further . we measure the redshifts by visually matching observed spectrum to quasar template using an eye - recognition assistant for quasar spectra software ( asera ; * ? ? ? the matching is based on @xmath92 , @xmath93 , nv , oi / siii , and siiv emission lines . the typical uncertainty of our redshift measurement is around 0.03 and will be smaller for higher s / n spectra . we do not include the systematic offset of ly@xmath8 emission line ( e.g. , shen et al.2007 ) , which is typical @xmath3 500 km / s and much smaller than the uncertainty of matching . we calculate the absolute magnitude at rest - frame 1450 , @xmath94 , by fitting a power - law continuum of each observed spectrum . we assume an average quasar uv continuum slope of @xmath34= @xmath95 @xcite , due to the fact that our spectra do not cover wide enough wavelength range for direct slope measurement . we normalise the power - law continuum to match the visually identified continuum windows that contain minimal contribution from quasar emission lines and sky oh lines . those quasar spectra used for fitting are all scaled by using their sdss @xmath30-band magnitude , the uncertainties of power - law continuum fitting are much smaller than the photometric errors , therefore the uncertainties of @xmath94 are comparable to sdss @xmath30-band photometric errors . the redshifts , @xmath94 and photometric information of our new quasars are listed in table 2 . in figure 3 , we plot the redshift distribution of our new quasars , compared with all previously known quasars and sdss - ulas / vhs - allwise detected known quasars in the magnitude limit @xmath60 20.5 mag . our discoveries , including the two which have been published as @xmath1 5 quasars , almost double the number of known quasars at @xmath1 5.5 with @xmath11 band magnitude brighter than 20.5 . all spectra of new quasars are presented in figure 4 . lllp1.25cmp1.25cmp1.25cmp1.25cmp1.25cmp1.25cmp1.25cmp1.25 cm j010806.60 + 071120.6 & 5.53 & @xmath3327.19 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + j011353.75 + 055951.1 & 5.00 & @xmath3325.83 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + j082933.10 + 250645.6 & 5.35 & @xmath3326.98 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + [email protected] & 5.32 & @xmath3326.25 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + j095712.20 + 101618.5 & 5.14 & @xmath3326.75 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + [email protected] & 5.55 & @xmath3326.96 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + j102201.91 + 080122.2 & 5.30 & @xmath3327.63 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + j113308.78 + 160355.7 & 5.61 & @xmath3327.49 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + j113414.23 + 082853.3 & 5.69 & @xmath3326.41 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + [email protected] & 5.31 & @xmath3327.44 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + j114946.45 + 074850.6 & 5.66 & @xmath3326.40 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + [email protected] & 5.25 & @xmath3326.27 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + j131929.23 + 151305.0 & 4.50 & @xmath3325.62 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + [email protected] & 5.67 & @xmath3327.76 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + j151339.64 + 085406.5 & 5.47 & @xmath3326.81 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + j152712.86 + 064121.9 & 5.57 & @xmath3326.92 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + [email protected] & 5.61 & @xmath3326.24 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + [email protected] & 5.22 & @xmath3327.13 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + j233008.71 + 095743.7 & 5.30 & @xmath3326.75 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + [email protected] & 5.25 & @xmath3326.34 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + j235824.04 + 063437.4 & 5.32 & @xmath3327.26 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + j015533.28 + 041506.7 & 5.37 & @xmath3327.10 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + [email protected] & 5.53 & @xmath3327.77 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + j222514.38 + 033012.5 & 5.24 & @xmath3327.17 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] * [email protected] * , @xmath11 = 5.67 . it is one of the most luminous new quasar with @xmath94 = @xmath97 . * j152712.86 + 064121.9 * , @xmath11 = 5.57 . it is a weak line quasar with a very weak ly@xmath8 emission line and no other obvious emission features . we measure the redshift by matching the continuum to template and the redshift uncertainty is a little larger than others . * j131929.23 + 151305.0 * , @xmath11 = 4.50 . the quasar has the lowest redshift in our new discoveries . it is a broad absorption line quasar with redder colors than normal lower redshift quasars , that is why it could be selected by our selection . * j113414.23 + 082853.3 * , @xmath11 = 5.69 . this one is the highest redshift quasar in this sample with strong ly@xmath8 emission and strong igm absorption blueward of ly@xmath8 . the sample obtained now is not yet a complete survey sample since we have only spectroscopically observed a small part of candidates , in which case it is hard to estimate the incompleteness of spectroscopy . but we want to first calculate completeness to demonstrate the effectiveness of the selection here . it will be used in later works . to first calculate the completeness of our color selection criteria , we generate a sample of simulated quasars using the quasar model from @xcite . we extend this model toward redder wavelengths to cover the allwise w1 , w2 bands for quasars at @xmath11 = 5 to 6 ( mcgreer et al . in prep . based on this model , a total of @xmath3 200,000 simulated quasars have been generated and evenly distributed in the ( @xmath94 , @xmath11 ) space of 5@xmath66 and @xmath98 . we assign optical photometric errors using the magnitude - error relations from the sdss main survey . for @xmath48 bands , we use the ulas photometric errors . the magnitude - error relations between ulas and vhs are similar in the flux limit of our selection . the little differences , including the difference between ulas k and vhs ks will not affect the selection function too much . here to show the redshift and magnitude dependent selection completeness , we only use the relations from ulas photometry . we will consider a separate vhs photometry based selection function in the further works for complete sample construction and luminosity function measurement . allwise detection depth is highly dependent on sky position , which will affect the detection incompleteness and photometric uncertainties . we model the coverage - dependent detection incompleteness and photometric uncertainties of allwise using the allwise coverage map within sdss - ulas / vhs area . we follow the procedure outlined in @xcite . figure 5 represents our selection function for @xmath0 quasars . the selection function shows a high completeness in the redshift range of 5.3 @xmath6 5.7 with a @xmath3 91% mean completeness at @xmath9926.5 . in the range of @xmath100 , the mean selection probability is around 40% . towards fainter end , the completeness decreases quickly due to the increasing detection incompleteness and photometric uncertainties of allwise w1 & w2 bands . as shown , the selected region with high completeness is extended to lower redshift to @xmath1 5.2 . that is caused by the slow evolution of quasars in @xmath23 and @xmath24 colors from @xmath10 5.1 to 5.3 ( see figure 1 ) . to include most of @xmath1 5.5 quasars , we use a relative relax @xmath46 cut and thus can select some lower redshift quasars , which is consistent with our result shown in figure 3 . at high redshift end , our @xmath101 2.2 cuts will restrict selected sample into @xmath1025.7 due to the quick increasing of quasar @xmath24 color from @xmath10 5.7 to 5.8 . so there is a sharp edge at @xmath10 5.7 . our selection using ulas / vhs photometric data is limited in a small area . to study the igm and quasar number density , a larger sample is required . therefore , the coming larger area optical / nir surveys will provide good opportunity for @xmath1 5.5 quasar selection . we used data from a preliminary version of the uhs . this is a j - band survey of the northern sky ( 0@xmath103 @xmath104 decl . @xmath104 60@xmath103 ) to a depth of j = 19.6 , supplementing the area already covered by ukidss . the survey was begun by the ukidss consortium , but is being completed by the new operators of ukirt , university of arizona and university of hawaii . the data is initially proprietary but is intended to be public in due course , through the same interface as ukidss , i.e. the wfcam science archive . the survey is briefly described by @xcite and will be fully reported in dye et al . ( 2016 in preparation ) . the combination of ulas and uhs will provides a complete j band map in the northern sky with the depth matching sdss . to test our selection with uhs j band , we selected a test candidate sample in the spring sky . we cross matched ( 2@xmath69 ) @xmath46 selected candidates with uhs and allwise catalog . we then used the same @xmath44-w1/w1-w2 color - cut as discussed above but without cuts related to @xmath55 & @xmath56 bands . due to the lack of @xmath55 and @xmath56 photometry , the fraction of m dwarf contaminations will increase , here we just focused on the bright candidates with sdss @xmath60 19.5 to reduce the number of candidates . we observed 9 selected candidates by using palomar p200/dbsp spectrograph and mmt / red channel spectrograph in april and may , 2016 . we discovered the first uhs j band selected @xmath1 5.5 quasar j101637.71 + 254131.9 = [email protected] , @xmath30 = [email protected] , @xmath11 = [email protected] , @xmath44 = [email protected] , w1 = [email protected] and w2 = [email protected] . ] at @xmath10 5.64 . the spectrum was obtained using p200/dbsp with grating g316 ( r @xmath3 960 at 7500 ) , 1@xmath875 slit and 500s@xmath1053 exposure time on april 27 & 30 in 2016 ( see spectrum in figure 4 ) . data was also reduced by using standard iraf routines . this quasar is a luminous quasar with @xmath94 = -27.81 . we measure the redshift and @xmath94 using the same method discussed in sec . 4.1 . uhs j band photometric data is helpful for the large area @xmath1 5.5 quasar survey . if more nir data will be provided , we can significantly reduce the number of candidates , such as using pan - starrs 1 ( ps1 ) data @xcite . ps1 covers the entire sky above declination @xmath106 in the @xmath107 filters . the area coverage and depth , especially in the reddest narrower bands @xmath11 & @xmath108 , make ps1 to be a good choice for @xmath1 5.5 quasar selection . therefore , combined ps1 and uhs data , a large uniform and completeness @xmath1 5.5 quasar sample can be expected . besides , the overlap area between ps1 and vhs will add a new optical / nir covered area for quasar selection in the southern sky . the obvious redshift gap of known quasars at 5.3 @xmath1095.7 becomes a limitation of the study of igm evolution , quasar number density and bh evolution from higher redshift to lower redshift over the post - reionaiztion epoch . this gap is caused by the same colors of @xmath1 5.5 quasar and late - type stars in broad optical bands . to explore quasars at this redshift range , we develop a new selection method for @xmath1 5.5 quasars and build the first sample of quasars at 5.3 @xmath2 5.7 . main results of our works are listed as following . * in addition to the traditional @xmath15 color - color diagram , we add new color - color selection criteria based on ulas / vhs @xmath44,@xmath55,@xmath56 , and allwise w1&w2 bands . we have done a pilot survey for @xmath1 5.5 quasars with sdss z band magnitude brighter than 20.5 using our new selection pipeline . * we have discovered 21 new quasars from our sdss - ulas / vhs - allwise selected main candidate sample . there are 15 new quasars in the redshift range of 5.3 @xmath2 5.7 and 5 quasars at redshift 5 @xmath6 5.3 . the other one is a broad absorption line quasar with @xmath91 4.50 . there are also 3 quasars in our target list but already being observed as @xmath1 5 candidates @xcite . 2 of them are z @xmath3 5.5 quasars . therefore , we construct the first uniform @xmath1 5.5 quasar sample with 17 quasars in the magnitude limit of sdss @xmath60 20.5 . * the selection function shows a high completeness at @xmath11026 , which can be expected to provide a sample of new @xmath1 5.5 quasars for measurement of quasar luminosity function at this redshift gap . * for the further application of a wide field quasar survey and to construct a larger sample , we have tried to construct the selection pipeline using uhs j band data . from our sdss - uhs - allwise selected test sample , we discovered the first uhs selected @xmath1 5.5 luminous quasar . in a subsequent paper , we will present the final complete sample of our @xmath1 5.5 quasar survey and the first measurement of quasar lf at this epoch . the evolution of quasar density from @xmath10 5 to 6 will also be constrained . our selection focus on luminous quasars , so the new quasar sample at this redshift range provide a valuable dataset to study the igm evolution in the tail of reionization . we will expand our selection to large survey area using new dataset such as the uhs , ps1 and the vlt survey telescope ( vst ) atlas @xcite . in this work , we only use color box to select @xmath0 quasar candidates . recently , several modern techniques , probabilistic selections based on bayesian model or extreme deconvolution method , have been explored to search quasars at different redshift range ( e.g. * ? ? ? * ; * ? ? ? the extreme deconvolution method in @xcite required a sample of quasars used as training set . while at @xmath12 , there were few known quasars and most of known quasars located in the right - bottom region in @xmath15 diagram due to selection criteria . there was no good training sample representing typical colors of @xmath12 quasars before . our new quasar sample and selection method could provide a new training sample for future probabilistic selection . besides , an extension of the bayesian model from @xcite to include more nir colors will also be expected to be useful for @xmath1 5.5 quasar selection . additional , in future , the variability ( e.g. lsst ) will also play an important role on high redshift quasar selection . we thank the referee for providing helpful comments and suggestions . j. yang , x .- b . wu and f. wang thank the supports by the nsfc grant no.11373008 and 11533001 , the strategic priority research program the emergence of cosmological structures of the chinese academy of sciences , grant no . xdb09000000 , the national key basic research program of china 2014cb845700 , and from the ministry of science and technology of china under grant 2016yfa0400703 . j. yang , x. fan and i. d. mcgreer acknowledge the support from the us nsf grant ast 11 - 07682 and ast 15 - 15115 .. funding for the lijiang 2.4 m telescope is provided by chinese academy of sciences and the people s government of yunnan province . this research uses data obtained through the telescope access program ( tap ) , which has been funded by the strategic priority research program `` the emergence of cosmological structures '' ( grant no . xdb09000000 ) , national astronomical observatories , chinese academy of sciences , and the special fund for astronomy from the ministry of finance in china . we acknowledge the use of the lijiang 2.4 m telescope , the mmt 6.5 m telescope , the bok telescope , anu 2.3 m telescope and palomar hale 5 m telescope . observations obtained with the hale telescope at palomar observatory were obtained as part of an agreement between the national astronomical observatories , chinese academy of sciences , and the california institute of technology . this work was partially supported by the open project program of the key laboratory of optical astronomy , national astronomical observatories , chinese academy of sciences . we acknowledge the use of sdss photometric data . funding for sdss - iii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , and the u.s . department of energy office of science . the sdss - iii web site is http://www.sdss3.org/. sdss - iii is managed by the astrophysical research consortium for the participating institutions of the sdss - iii collaboration including the university of arizona , the brazilian participation group , brookhaven national laboratory , university of cambridge , carnegie mellon university , university of florida , the french participation group , the german participation group , harvard university , the instituto de astrofisica de canarias , the michigan state / notre dame / jina participation group , johns hopkins university , lawrence berkeley national laboratory , max planck institute for astrophysics , max planck institute for extraterrestrial physics , new mexico state university , new york university , ohio state university , pennsylvania state university , university of portsmouth , princeton university , the spanish participation group , university of tokyo , university of utah , vanderbilt university , university of virginia , university of washington , and yale university . this publication makes use of data products from the wide - field infrared survey explorer , which is a joint project of the university of california , los angeles , and the jet propulsion laboratory / california institute of technology , and neowise , which is a project of the jet propulsion laboratory / california institute of technology . wise and neowise are funded by the national aeronautics and space administration . we acknowledge the use of the ukidss data , and the vista data . baados , e. , venemans , b. p. , decarli , r. , et al . 2016 , arxiv:1608.03279 becker , g. d. , bolton , j. s. , madau , p. , et al . 2015 , , 447 , 3402 bolton , j. s. , becker , g. d. , raskutti , s. , et al . 2012 , , 419 , 2880 cardelli , j. a. , clayton , g. c. , & mathis , j. s. 1989 , , 345 , 245 childress , m. j. , vogt , f. p. a. , nielsen , j. , & sharp , r. g. 2014 , , 349 , 617 cool , r. j. , kochanek , c. s. , eisenstein , d. j. , et al . 2006 , , 132 , 823 chambers , k. c. , magnier , e. a. , metcalfe , n. , et al . 2016 , arxiv:1612.05560 dipompeo , m. a. , bovy , j. , myers , a. d. , & lang , d. 2015 , , 452 , 3124 dopita , m. , hart , j. , mcgregor , p. , et al . 2007 , , 310 , 255 dopita , m. , rhee , j. , farage , c. , et al . 2010 , , 327 , 245 douglas , l. s. , bremer , m. n. , stanway , e. r. , & lehnert , m. d. 2007 , , 376 , 1393 fan , x. , carilli , c. l. , & keating , b. 2006 , , 44 , 415 fan , x. , strauss , m. a. , schneider , d. p. , et al . 1999 , , 118 , 1 jiang , l. , fan , x. , annis , j. , et al . 2008 , , 135 , 1057 jiang , l. , mcgreer , i. , fan , x. , et al . 2016 , , submitted kaiser , n. , aussel , h. , burke , b. e. , et al . 2002 , , 4836 , 154 kaiser , n. , burgett , w. , chambers , k. , et al . 2010 , , 7733 , 77330e kashikawa , n. , ishizaki , y. , willott , c. j. , et al . 2015 , , 798 , 28 kirkpatrick , j. d. , kelly , d. m. , rieke , g. h. , et al . 1993 , , 402 , 643 komatsu , e. , dunkley , j. , nolta , m. r. , et al . 2009 , , 180 , 330 lawrence , a. 2013 , astrophysics and space science proceedings , 37 , 271 lawrence , a. , warren , s. j. , almaini , o. , et al . 2007 , , 379 , 1599 lupton , r. h. , gunn , j. e. , & szalay , a. s. 1999 , , 118 , 1406 mainzer , a. , bauer , j. , grav , t. , et al . 2011 , , 731 , 53 matute , i. , masegosa , j. , mrquez , i. , et al . 2013 , , 557 , a78 mcgreer , i. d. , jiang , l. , fan , x. , et al . 2013 , , 768 , 105 mcgreer , i. d. , mesinger , a. , & dodorico , v. 2015 , , 447 , 499 mclean , i. s. , mcgovern , m. r. , burgasser , a. j. , et al . 2003 , , 596 , 561 mcmahon , r. g. , banerji , m. , gonzalez , e. , et al . 2013 , the messenger , 154 , 35 mortlock , d. j. , patel , m. , warren , s. j. , et al . 2012 , , 419 , 390 pris , i. , petitjean , p. , aubourg , . , et al . 2014 , , 563 , a54 pris , i. , petitjean , p. , ross , n. p. , et al . 2016 , arxiv:1608.06483 richards , g. t. , fan , x. , newberg , h. j. , et al . 2002 , , 123 , 2945 romani , r. w. , sowards - emmerd , d. , greenhill , l. , & michelson , p. 2004 , , 610 , l9 schmidt , g. d. , weymann , r. j. , & foltz , c. b. 1989 , , 101 , 713 schneider , d. p. , richards , g. t. , hall , p. b. , et al . 2010 , , 139 , 2360 schlegel , d. j. , finkbeiner , d. p. , & davis , m. 1998 , , 500 , 525 shanks , t. , metcalfe , n. , chehade , b. , et al . 2015 , , 451 , 4238 simpson , c. , mortlock , d. , warren , s. , et al . 2014 , , 442 , 3454 stern , d. , djorgovski , s. g. , perley , r. a. , de carvalho , r. r. , & wall , j. v. 2000 , , 119 , 1526 trakhtenbrot , b. , netzer , h. , lira , p. , & shemmer , o. 2011 , , 730 , 7 vanden berk , d. e. , richards , g. t. , bauer , a. , et al . 2001 , , 122 , 549 wang , f. , wu , x .- b . , fan , x. , et al . 2016 , , 819 , 24 willott , c. j. , albert , l. , arzoumanian , d. , et al . 2010a , , 140 , 546 willott , c. j. , delorme , p. , reyl , c. , et al . 2010b , , 139 , 906 wright , e. l. , eisenhardt , p. r. m. , mainzer , a. k. , et al . 2010 , , 140 , 1868 yang , j. , wang , f. , wu , x .- b . 2016 , accepted by apj yuan , h. , zhang , h. , zhang , y. , et al . 2013 , astronomy and computing , 3 , 65

minimum divergence methods are popular tools in a variety of statistical applications . we consider tubular model adequacy tests , and demonstrate that the new divergences that are generated in the process are very useful in robust statistical inference . in particular we show that family of-divergences can be alternatively developed using the tubular model adequacy tests ; a further application of the paradigm generates a larger superfamily of divergences . we describe the properties of this larger class and its potential applications in robust inference . along the way , the failure of the first order influence function analysis in capturing the robustness of these procedures is also established .

1611.05224

tests of goodness - of - fit constitute an important component of statistical methodology . they assess the agreement between the observed data and the model family through appropriate test statistics indicating the extent of possible error in our model based inference . the standard approach of assessing the goodness - of - fit is generally based on suitable chi - square type divergences , such as the pearson s chi - square ( pcs ) and the likelihood divergence ( ld ) . the latter divergence generates the likelihood ratio chi - square . many other chi - square type divergences are available in the literature . for example the power divergence ( pd ) family of @xcite provides several other choices beyond these two . standard approaches to goodness - of - fit testing are also described in many other sources such as @xcite , @xcite , @xcite etc . however , as noted in @xcite , all these tests generally fail to pick out any specific model when the sample size is sufficiently large as the specified model would be eventually edged out by a more complicated model and hence these tests fail to assess if the specified model is actually a good approximation or not . it may be that the specified model ( simple and easily workable ) is already very close to the true data generating distribution ( i.e , the discrepancy between them is very small , within practically acceptable range ) , yet these chi - square type goodness - of - fit tests might get rejected for large sample sizes in favor of more complicated models which have marginally lower discrepancy but are otherwise hard to work with . in such situations any practitioner will prefer to accept the simpler model as the error is already in acceptable range , but these chi - square type tests will not help them in this context . to get around this problem , @xcite developed a measure of model adequacy " based on a mixture method for contingency tables . let @xmath1 denote the true data generating probabilities for the cells in the contingency table and consider a parametric model family of cell probabilities denoted by @xmath2 . then , one can express @xmath1 as a two - point mixture of a model element @xmath3 and an unspecified modeling error probability @xmath4 ; this mixture is given by @xmath5 , where @xmath6 is the weight of error in modeling through @xmath7 . then , the model adequacy index or the mixture index of fit @xmath8 can be defined as the smallest possible value of this @xmath6 indicating the minimum error proportion that can not be explained through the model family @xmath7 . so , whenever @xmath8 is large the assumed model is far from the true distributions . on the other hand , if @xmath8 is small and less than a pre - specified error limit ( say , @xmath9 ) one can deduce any inference based on the assumed model within the allowed error margin . thus , the quantity @xmath10 indeed measures the distance " of the assumed model from the true distribution allowing the evaluation of the model downplaying the role of the sample size . note that this @xmath8 can be re - written as @xcite @xmath11,\ ] ] where @xmath12 is the cell index and the quantity within the third bracket is in fact a statistical distance ( or , divergence ) between the model element @xmath13 and the true probability @xmath1 . then the model is said to be adequate at level @xmath14 " if @xmath15 and the test for this hypothesis is known as the model adequacy test " . more generally , let us use the term density " for both the discrete or continuous probabilities and consider the class @xmath16 of all density functions which are absolute continuous with respect to some common dominating measure @xmath17 . in case of continuous variables this dominating measure is the lebeague measure while in case of discrete variables it is a suitable counting measure . define a general statistical divergence @xmath18 between two density functions @xmath1 and @xmath13 as a non - negative function from @xmath19 to @xmath20 which equals zero if and only if @xmath21 identically . for such a general statistical divergence , we can say that the assumed model @xmath22 is adequate at level @xmath14 with respect to the true probability @xmath1 if the distance of @xmath1 from the model family @xmath7 defined as @xmath23 is less than or equal to @xmath14 . so our general model adequacy test considers the null hypothesis @xmath24 note that , this null hypothesis indeed says that the true probability @xmath1 is within a tolerance region @xmath25 about @xmath7 ( also called the model tube " ) defined as @xmath26 further the model tube @xmath25 is an union of open balls of radius @xmath14 , the pre - specified tolerance limit , around each model elements and hence the null hypothesis in ( [ eq : mat ] ) , alternatively expressed as @xmath27 , indicates that the true probability @xmath1 belongs to the radius-@xmath14 open ball around at least one model element . note that , in the cases where the null hypothesis is true , then @xmath1 may belong to the radius-@xmath14 open ball around more than one model element all of which can be adopted given the tolerance limit @xmath14 . given @xmath14 , one can construct suitable test statistics for solving the problem of model adequacy . @xcite , with their particular choice of @xmath28 , considered the likelihood ratio test for the corresponding model adequacy hypothesis . however , the distance function involved in @xmath8 is not everywhere differentiable which leads to non - standard asymptotics @xcite as well as severe computational difficulties @xcite . other works in the related context of tolerance region based goodness - of - fit were done by @xcite , @xcite and @xcite ; these works were mostly based on euclidean distances and not quite general for all parametric models . further , a challenging task in model adequacy test is the selection of the tolerance limit @xmath14 , which was also not quite easy for these approaches . to resolve these issues , @xcite proposed to construct the tolerance region in the null hypothesis ( [ eq : mat ] ) by the simpler , easily interpretable and everywhere differentiable kullback - leibler divergence defined as @xmath29 note that the kld is adjoint to the popular likelihood divergence ( ld ) given by @xmath30 @xcite went on to construct the likelihood ratio test statistic ( based on the ld ) for the model adequacy test with @xmath31kld and illustrated several nice properties of this approach . specifically , the particular geometric structure of the kld in connection with the likelihood ratio yields a nice algorithm to choose the tolerance limit @xmath14 with interesting applications . the above discussion invokes a natural question : can we obtain some larger superfamily of statistical divergences , that could be further helpful in robust statistical inference like the pd or the gkl families , through more general model adequacy tests ? in this paper , we try to answer this interesting question by exploring some model adequacy tests based on more general divergences ( rather than kld and ld only ) . we first consider a recent generalization of the pd family , namely the two - parameter @xmath0-divergence family of @xcite , and illustrate that it can also be obtained from a suitable model adequacy test based on appropriate generalizations of the ld and kld in section [ sec : inerpret_sd ] . these generalizations of the ld and the kld are indeed seen to be the special cases of the @xmath0-divergence family . so , we further generalize our model adequacy test by replacing kld and ld by appropriate members of the general @xmath0-divergence family and explore its properties in section [ sec : mat_sd ] . these require some non - trivial generalization of the work of @xcite in order to provide a general class of divergence based model adequacy tests which increases the novelty of this work . further , as expected , this general @xmath0-divergence based model adequacy tests again generate another _ new _ and larger super - family of divergences which , interestingly , contains both the @xmath0-divergence family and the gkl family as its subclasses . this is another novel discovery within the literature of the density based divergences and the related robust statistical inference . we refer to this larger three parameter superfamily of divergences as the generalized @xmath0-divergence ( gsd ) family and explore some of its basic properties in section [ sec : gsd ] . the potential applications of this gsd superfamily in robust statistical inference have been investigated in section [ sec : gsd_inf ] along with the asymptotic and robustness properties of the resulting estimators . section [ sec : illustratios ] presents some numerical illustrations of the performance of gsd based inference and the paper ends with some concluding remarks in section [ sec : conclusion ] . before concluding this section , we summarize , in an itemized manner , the major contributions of this paper . * we explore the process of generating new divergence families from suitable model adequacy tests . this technique is not entirely unknown in the literature ; however it certainly has not received adequate attention and its application has been limited to the one parameter gkl and pd families . * we demonstrate that the two parameter family of @xmath0-divergences , which has been recently proposed and studied in the literature , can be generated through appropriate model adequacy tests in the same spirit . * we demonstrate that a further application of this technique leads to the generation of an entirely new family of divergences which is a super - class containing the @xmath0-divergence family as a special case . * an exploration of the properties of the divergences within this super - class demonstrates that the divergences within this super - class which appear to provide the best combinations of robustness and efficiency do not belong to any of the previously studied subclasses of this super - class . this indicates that studying the properties of the individual subclasses may not be sufficient to identify the best " divergence in respect of robust parametric estimation , so that this super - class does make some positive value addition to the literature . * a study of the minimum divergence estimators generated by the above superclass reveals the limitation of the first order influence function analysis in assessing their robustness . the @xmath0-divergence family of density - based divergences , recently developed by @xcite , is a two - parameter @xmath47 generalization of the pd family that connects each member of the pd family ( having parameter @xmath48 ) at @xmath49 to the @xmath50 divergence at @xmath51 . this family is defined as @xmath52d\mu , \label{eq : s_div_gen}\ ] ] where @xmath53 and @xmath54 . clearly , @xmath55 . also the above form is defined only when @xmath56 and @xmath57 ; for @xmath58 or @xmath59 the corresponding @xmath0-divergence measure is defined by the continuous limit of ( [ eq : s_div_gen ] ) as @xmath60 or @xmath61 respectively . further , at the choice @xmath62 , this @xmath0-divergence family contains another popular divergence family , namely the density power divergence ( dpd ) family of @xcite having parameter @xmath63 , defined as @xmath64 note that the @xmath0-divergences measures are not symmetric in general . but it becomes symmetric if and only if either @xmath65 ( which generates the @xmath50 divergence ) , or @xmath66 . the latter case represents an interesting subclass referred to as the @xmath0-hellinger distances ( shd ) in @xcite and is defined by @xmath67 it connects the hellinger distance at @xmath68 to @xmath50-distance at @xmath51 . note that just as the hellinger distance represents the self adjoint member of the pd family , any other cross section of the class of @xmath0-divergences for a fixed value @xmath69 has a self adjoint member in @xmath70 . the applications of @xmath0-divergences in robust parametric inferences have been described in @xcite , @xcite and @xcite . it has been illustrated that some members of the @xmath0-divergence family generate more robust inference compared to its existing members within the pd and dpd subfamilies . in this paper , we demonstrate that one can obtain the @xmath0-divergence family from the suitable model adequacy test following the arguments of the previous section . since the @xmath0-divergence family is an extension of the pd family , we should be able to obtain it as a solution of suitable generalization of the testing problem ( [ eq : mat_pd ] ) . so , it is intuitive to construct the required model adequacy test based on similar extensions of the kld and ld divergences through the parameter @xmath69 . we consider the particular subfamily within the @xmath0-divergence family having this property ( corresponding to @xmath58 and @xmath59 respectively ) which are given by @xmath71 and @xmath72 for all @xmath73 . in particular , at the choice @xmath68 , the above families simplify , respectively , to the kld and the ld measures . further , @xcite have noted a special geometric structure between the kld and the ld which cancels the curvatures of each other to generate the new family of divergence measures . it is the property that any one of these two divergences can be obtained from the other just by interchanging @xmath1 and @xmath13 , i.e. , in the language of @xcite , the ld and kld are adjoint to each other ; they are also symmetrically opposite to each other within the pd family , with the hellinger distance ( hd ) measure being the point of symmetry . this special geometry also holds for our sld and skl families at any given @xmath69 , i.e. , @xmath74 further , they are also symmetrically opposite to the only self - adjoint subfamily , namely the shd family , within the @xmath0-divergence family of divergences ; this fact can be easily verified from the plots of the lines @xmath58 and @xmath59 in the @xmath75 plane , as shown in figure [ fig : ab=0 ] , which are clearly symmetric around the line @xmath76 for each @xmath77 . the line @xmath51 is the asymptote of both these lines as @xmath78 @xmath79 respectively which justifies the fact that @xmath0-divergences with @xmath51 are independent of the choice of @xmath48 and are always self - adjoint too . these geometrical properties of the skl and sld families intuitively indicate that the use of the members of the skl and sld families at any particular @xmath77 in constructing the model adequacy test would possibly generate the corresponding cross - section of the @xmath0-divergence family with the same @xmath69 . we prove this intuition rigorously in the rest of this section . ) and the sld family ( @xmath59 ) on the @xmath75-plane of the @xmath0-divergence family , relative to the self - adjoint members ( dashed line , at @xmath80-@xmath81),scaledwidth=50.0% ] take any fixed @xmath77 and a tolerance limit @xmath14 . we want to test whether our model is adequate in terms of the @xmath82 divergence at level @xmath14 . this is equivalent to testing for the null hypothesis @xmath83 let us construct a divergence based test ( rather than the lrt ) for this null hypothesis using the @xmath84 with the same tuning parameter @xmath69 . these divergence based tests are quite robust compared to the lrt as explored in @xcite and @xcite . then , based on a sample of size @xmath85 , the resulting test statistic would be @xmath86 , where @xmath87 is the estimated density under the null hypothesis obtained by minimizing the skl divergence between the true density @xmath1 and the model @xmath13 over all possible model densities satisfying the null hypothesis . thus , @xmath87 is nothing but the solution of the constrained optimization problem @xmath88 as in @xcite , we transfer this constrained optimization problem to an equivalent simpler unconstrained optimization problem . for simplicity , let us first assume that the model family contains only one element , i.e. , @xmath89 . then , the constrained problem ( [ eq : constrained1_sd ] ) simplifies to @xmath90 the constrained optimization problem ( [ eq : constrained2_sd ] ) has a unique solution on the boundary of @xmath25 having the form @xmath91 for some @xmath42 $ ] , a function of given @xmath14 satisfying @xmath92 . [ thm : solution_sd_mat ] * proof : * the proof follows easily by applying the lagrange multiplier method . @xmath93 . further , starting from a fixed @xmath42 $ ] we can see that the much simpler unconstrained problem @xmath94 , ~~\bar{\tau } = 1- \tau , \label{eq : unconstrained_sd}\ ] ] has the solution @xmath95 of the same form as ( [ eq : sol_form_sd ] ) . as argued in @xcite , there is a direct one to one correspondence between the parameters @xmath96 and @xmath14 in the two optimization problems in ( [ eq : unconstrained_sd ] ) and ( [ eq : constrained2_sd ] ) respectively through the relation @xmath97 . thus our targeted quantity @xmath87 in our model adequacy test statistic @xmath86 can be easily obtained as the solution to the simpler unconstrained optimization problem in ( [ eq : unconstrained_sd ] ) with a suitable @xmath96-value satisfying @xmath97 . now , suppose the model family @xmath7 contains more than one element indexed by the parameter @xmath32 as @xmath98 . then , we should extend the unconstrained problem in ( [ eq : unconstrained_sd ] ) to the optimization problem @xmath99 \equiv \min_\theta\min_{p } \left[\tau { \rm skl}_\alpha(g , p ) + \bar{\tau } { \rm sld}_\alpha(p , f_\theta)\right ] . \label{eq : unconstrained2_sd}\ ] ] again we can show that the required solution to the constrained optimization problem in ( [ eq : constrained1_sd ] ) is nothing but the solution to this unconstrained problem in ( [ eq : unconstrained2_sd ] ) with some appropriate value of @xmath96 satisfying @xmath100 . further , the inner minimization in ( [ eq : unconstrained2_sd ] ) is of the form ( [ eq : unconstrained_sd ] ) and has the explicit solution @xmath101 . substituting this solution , the final optimization problem in ( [ eq : unconstrained2_sd ] ) becomes @xmath102 . \label{eq : unconstrained3_sd}\ ] ] however , a routine algebra shows that @xmath103\nonumber\\ & = & \left[\tau { \rm skl}_\alpha(g , g^\tau f_\theta^{\bar{\tau } } ) + \bar{\tau } { \rm sld}_\alpha(g^\tau f_\theta^{\bar{\tau } } , f_\theta)\right]\nonumber\\ & = & \tau \bar{\tau}s_{(\alpha , \lambda_\tau)}(g , f_\theta),\end{aligned}\ ] ] where @xmath104 and @xmath105 is a function of @xmath96 . therefore , our required solution @xmath95 is nothing but the model element @xmath106 with @xmath107 being the minimum @xmath0-divergence estimator defined as @xmath108 thus , our model adequacy test based on skl@xmath109 independently yields the form of the @xmath0-divergence measure with appropriate parameter and also directly depends on the corresponding minimum @xmath0-divergence estimator which has nice robustness properties with respect to the outliers in the sample data @xcite . further , noting that , given @xmath110 , @xmath111 is a one - to - one function of @xmath96 , and any @xmath0-divergence measure with parameter @xmath112 and @xmath113 $ ] can be obtained from the model adequacy based on the skl with the same @xmath69 and the appropriate value of @xmath14 obtained from @xmath114 . therefore the @xmath0-divergence family has a direct one to one correspondence with the model adequacy test based on skl and sld . the nice and interesting interpretation of the @xmath0-divergence measures as discussed in the previous section invokes a natural question . what if we start with a more general divergence family instead of the sld and skl ? will we get an even larger superfamily of divergences in that case ? in order to answer these questions , we now develop a general model adequacy test based on suitable members of the @xmath0-divergence family . noting the special geometric curvatures of the skl and sld families and the fact that the whole @xmath0-divergence family is symmetric about @xmath115 , we may try to replace the skl and sld by suitable members of the general @xmath0-divergence family . more precisely , for any @xmath112 , we want to consider two @xmath0-divergence measures with parameters @xmath116 and @xmath117 in place of sld@xmath109 and skl@xmath109 in such a way that their geometric curvature cancel each other just like skl and sld . that is , these two members should be symmetrically opposite with respect to the line @xmath76 and satisfy @xmath118 from the derivations of the previous section , we may take @xmath119 for some @xmath120 and then the above requirements give us the choice @xmath121 . so , we consider the null hypothesis @xmath122 and construct the model adequacy test based on @xmath123 . note that the choice @xmath124 yields the model adequacy test of the previous section based on skl and sld . based on a sample of size @xmath85 , we take an estimate @xmath1 of the true density and define the test statistics for the model adequacy test of the hypothesis ( [ eq : mat_gen ] ) as given by @xmath125 where @xmath87 is the best model element ( with respect to the @xmath126 measure ) in the model tube @xmath25 and is defined as @xmath127 thus , the @xmath87 can be obtained as the solution of the constrained optimization problem @xmath128 proceeding as in the previous section , we can again show that the unique solution to the above constrained optimization problem with general model family @xmath129 is nothing but the unique solution @xmath95 of the unconstrained optimization problem @xmath130 \equiv \min_\theta\min_{p}\left[\tau s_{(\alpha , \frac{\gamma}{1-\alpha})}(g , p ) + \bar{\tau}s_{(\alpha , \frac{\alpha-1-\gamma}{1-\alpha})}(p , f_\theta)\right ] , \label{eq : unconstrained_gen}\ ] ] for some @xmath131 $ ] that depends on the tolerance limit @xmath14 through the relation @xmath132 to see this , we again consider the one element model family as @xmath89 . then , the following theorem presents the solution of the simplified version of the constrained problem ( [ eq : constrained1_gen ] ) given by @xmath133 the proof of the theorem again follows from the method of lagrange multiplier and is hence omitted . however , we can easily check that the solution @xmath95 as given in the above theorem is also the unique solution to the much simpler unconstrained optimization problem @xmath136 , \label{eq : unconstrained1_gen}\ ] ] for the @xmath96 value satisfying @xmath135 . thus , our general model adequacy test based on the @xmath0-divergences yields a divergence function through the quantity @xmath140 . this function @xmath140 is can be easily extended to the cases @xmath144 and @xmath145 through continuous limits of the above form and define a three - parameter family of proper statistical divergence measures . further the choice @xmath124 reduces the general divergence measure @xmath140 to the suitable @xmath0-divergence measure with parameters @xmath69 and @xmath111 ( defined in ( [ eq : lambda_tau ] ) ) . so , we refer to this general three parameter family as the generalized super - divergence ( gsd ) family " and discuss its properties in detail in the next section . note that given the generated sample data , we can easily compute the model adequacy test statistics by using the minimum gsd estimators ( mgsdes ) of the model parameter @xmath32 . and the performance and robustness of the proposed model adequacy test directly depend on that of the minimum gsd estimators . in the rest of this paper , we primarily illustrate the properties of the gsd and the corresponding estimators in detail and leave the detailed analysis of this general model adequacy test resulting from the gsd and the mgsde for our future research work . the generalized @xmath0-divergence ( gsd ) family , as derived in the previous section , is a three parameter family of statistical divergences @xmath140 defined in equation ( [ eq : def0_gsd ] ) for @xmath112 , @xmath146 and @xmath147 with @xmath148 . we can easily extend this family to be defined over the parameter values @xmath149 $ ] , @xmath42 $ ] and @xmath150 through their continuous limits in the form given by equation ( [ eq : def0_gsd ] ) . in table [ tab : spcl ] , we have listed several divergences or families of divergences which belong to the gsd family . many ( although not all ) of the divergences ( or , classes of divergences ) listed in table [ tab : spcl ] are obtained as limiting members of the gsd family . .existing divergences as special cases of the gsd family [ cols= " < , < " , ] [ tab : spcl ] thus , the generalized @xmath0-divergence family contains the existing divergence families like the @xmath0-divergence family ( and hence the pd and dpd families ) and the gkl family of @xcite as its limiting special cases and yields many more interesting new divergence measures . however , one interesting limiting subfamily of this gsd family arises at the choice @xmath145 which gives @xmath151 this family is an interesting generalization of the ordinary gkl family over the parameter @xmath149 $ ] defined in equation ( [ eq : gkl ] ) and simplifies to the gkl family at @xmath68 . first , we show that all the members of the gsd family represent proper statistical divergences . to see this , we can rewrite the gsd family ( [ eq : def0_gsd ] ) in terms of the pearson residual @xmath152 as @xmath153 where @xmath154 , ~~\delta\geq -1.\ ] ] this function @xmath155 is strictly convex in on @xmath156 and satisfies the relations @xmath157 here @xmath158 and @xmath159 denote the first and second order derivatives respectively with respect to the argument . using these properties , we get the following theorem . all the functions @xmath140 in the generalized @xmath0-divergence family for @xmath160 $ ] and @xmath150 define proper statistical divergence in the sense that , for any two densities @xmath161 , @xmath162 [ thm : gsd_pro1 ] examining the forms of the generalized @xmath0-divergence measures , we can obtain several interesting properties of this superfamily , which are listed in the following theorem . for any two densities @xmath163 , the generalized @xmath0-divergence measure @xmath140 satisfies the following properties : 1 . for any given @xmath69 and @xmath164 , the gsd is adjoint with respect to @xmath165 , i.e. , @xmath166 2 . at @xmath165 , the gsd measure is self - adjoint and symmetric in its arguments yielding an interesting subfamily of gsd given by @xmath167.\ ] ] 3 . the gsd measure also become symmetric and self - adjoint at @xmath168 , where it becomes independent of the parameter @xmath96 and coincides with the shd family . [ thm : gsd_pro2 ] from the above properties and the limiting forms of the gsd measures , we can clearly observe that all the members of the gsd family are not distinct ; they become identically equal at two or more choice of the parameter combination @xmath169 . for example , @xmath170 therefore , it would be an interesting future challenge for a mathematician to find the underlying geometry and functional topological properties of the gsd family and characterize all the distinct members within this large superfamily . for the present paper , we try to illustrate this fact through a suitable numerical example in section [ sec : illustratios ] . let us consider the parametric model family of densities @xmath171 . let @xmath172 denote the distribution function corresponding to @xmath35 . we are interested in the estimation of the parameter @xmath32 . let @xmath173 denote the distribution function corresponding to the true density @xmath1 . the minimum gsd functional @xmath174 at @xmath173 is then defined by the relation @xmath175 provided the minimum exists . from its definition , the gsd functional @xmath174 is fisher consistent under the assumption that the model is identifiable . when @xmath173 is outside the model , @xmath176 represents the best fitting parameter , and @xmath177 is the model element closest to @xmath1 in the gsd sense . for simplicity , we suppress the subscript @xmath178 for @xmath179 . given the observed sample data , we estimate the parameter @xmath32 by minimizing the divergence @xmath180 over @xmath181 , where @xmath34 is some non - parametric estimate of the true density @xmath1 based on the observed sample . when the model is discrete , a simple choice for @xmath34 is given by the relative frequencies ; for continuous models there is no such simple choice and we need some nonparametric density estimator . thus , the estimating equation for the minimum gsd estimator ( mgsde ) is given by @xmath182 where @xmath183 and @xmath184 . \label{eq : k_delta}\ ] ] note that , the function @xmath185 satisfies the relations @xmath186 and @xmath187 . therefore , this estimating equation of the mgsde is an _ unbiased estimating equation _ at the model @xmath188 . further , at any fixed @xmath69 , the estimating equation of the mgsdes differ only in terms of the function @xmath189 , just like the case of minimum @xmath0-divergence estimators ( msdes ) . hence , the robustness properties of the resulting estimators are expected to depend at least partially on the form of this @xmath189 function . however , all the divergences within the gsd family produce affine invariant estimators as described in the following proposition . consider the affine transformation @xmath190 for a non - singular matrix @xmath191 and vector @xmath192 of the dimension as of @xmath193 . then we have @xmath194 where @xmath195 . therefore the divergence measure @xmath140 is not itself affine invariant , but the corresponding minimum divergence estimator is affine equivariant . * proof : * note that , under the above transformation , the jacobian formula gives us the relation between the densities of @xmath193 and @xmath196 as @xmath197 then the proposition follows from this relation and the form of the @xmath0-divergence . @xmath198 the influence function is a popular and useful indicator of the first - order robustness and efficiency of any estimator . influence function in fact indicates the asymptotic effect of the infinitesimal contamination on the properties of the estimator in the neighborhood of the true distribution @xcite . more precisely , the influence function of any statistical functional @xmath199 at the true distribution @xmath173 is defined as @xmath200 where @xmath201 is the contaminated distribution @xmath202 , @xmath203 is the contamination proportion and @xmath204 denote the degenerate distribution at the contamination point @xmath205 . the influence function of a robust estimator should be bounded ; its non - boundedness indicates that the first order asymptotic bias of the estimator may diverge to infinity under contamination . we will now examine the first order robustness of the proposed minimum gsd estimator by deriving its influence function . let us consider the minimum gsd functional @xmath206 as defined in ( [ eq : msde_functional ] ) , which satisfies the estimating equation ( [ eq : s - divergence_est_equation ] ) with @xmath34 replaced by @xmath1 . then , a straightforward differentiation of the estimating equation yields the required influence function as presented in the following theorem . [ thm : msde_if ] consider the general set - up of the minimum gsd estimator as mentioned in the previous subsection . then , the influence function of the minimum gsd functional @xmath207 is given by @xmath208\ ] ] where @xmath209 , @xmath210 and @xmath211 with @xmath212 and @xmath213 denoting the gradient with respect to @xmath32 . [ cor : msde_if0 ] under the set - up of theorem [ thm : msde_if ] , suppose the true density @xmath1 belongs to the model family @xmath7 with @xmath214 . then the influence function of the minimum gsd functional has the simpler form @xmath215 . \label{eq : s_div_if_model}\end{aligned}\ ] ] the most interesting and remarkable observation here is that the influence function of the mgsde at the model is independent of the parameters @xmath164 and @xmath96 and depends only on the parameter @xmath69 . hence the influence function analysis predicts similar behavior ( in terms of first order robustness and efficiency ) for all minimum gsd estimators with the same value of the parameter @xmath69 irrespective of the other two parameters @xmath164 and @xmath96 . also note that , this influence function is the same as that of the minimum dpd estimators and also the minimum @xmath0-divergence estimators with the same values of @xmath69 . thus the influence functions of the mgsdes have bounded re - descending natures except in the case @xmath49 where it is unbounded . figure [ fig : if_s_div ] shows the nature of the influence functions for the poisson - mean ( discrete case ) and normal - mean ( continuous case ) . therefore , as per the first order influence function analysis , all the mgsdes are robust for all @xmath216 and non - robust at @xmath68 . under the @xmath217 model at the @xmath218 ( first panel ) and the normal @xmath219 model at the @xmath220 ( second panel),title="fig:",scaledwidth=45.0%,scaledwidth=35.0% ] under the @xmath217 model at the @xmath218 ( first panel ) and the normal @xmath219 model at the @xmath220 ( second panel),title="fig:",scaledwidth=45.0%,scaledwidth=35.0% ] in actual practice , the picture given by the first order influence function analysis often leads to an inaccurate prediction of the actual performance of the mgsde . our simulation studies in section [ sec : illustratios ] will demonstrate that some members of the gsd family having bounded influence functions generate highly non - robust estimators ; these choices include small positive @xmath69 , @xmath221 and @xmath222 . on the other hand some members of the gsd family with @xmath68 , @xmath223 and @xmath224 ( large ) , in spite of having unbounded influence functions , generate highly robust estimators . hence , the classical first order influence function analysis can not portray the true robustness picture for the minimum gsd estimators . as explained above , it may fail in both counts ; it may label strongly robust estimators as unstable , and may declare highly unstable estimators as being strongly robust . an appropriate second order influence function analysis may provide a more accurate description of the distortion of the estimators due to contamination . for brevity we do not present the second order analysis in this paper ; however this analysis reaffirms and strengthens all the illustrations and conclusions of @xcite . let us now describe the consistency and asymptotic distribution of the proposed minimum gsd estimators . for simplicity , we will consider only the case of discrete models in this paper . suppose @xmath225 are @xmath85 independent and identically distributed observations from a discrete probability mass function ( pmf ) @xmath1 modeled by the parametric family @xmath226 and let the distributions be supported , without loss of generality , on @xmath227 . we assume that both @xmath1 and @xmath7 belong to the class @xmath16 , where the dominating measure @xmath17 is now the counting measure over the support @xmath228 . under this set - up , we can easily get an estimate @xmath34 of the true pmf @xmath1 through the relative frequencies defined as @xmath229 , where @xmath230 denote the indicator function of the event @xmath231 . so , we can get the mgsde of @xmath32 by minimizing the gsd measure between two probability vectors @xmath232 and @xmath233 and hence the corresponding estimating equation is given by ( [ eq : s - divergence_est_equation ] ) with @xmath234 replaced by @xmath235 and the integral replaced by summation over @xmath228 : @xmath236 where now we have @xmath237 . now , in order to prove the asymptotic properties of the mgsde , we consider the matrix @xmath238 as defined in ( [ eq : j_g ] ) and define @xmath239 , \label{eq : v_g}\ ] ] where @xmath240 represents the variance under the true density @xmath1 . further , we also make the following assumptions : 1 . the model family @xmath241 is identifiable . 2 . the probability mass functions @xmath35 of the model have common support so that the set @xmath242 is independent of @xmath32 . the true pmf @xmath1 also supported on @xmath228 . 3 . there exists open subset @xmath243 for which the best fitting parameter @xmath244 is an interior point and for almost all @xmath245 , the pmf @xmath246 admits all third derivatives of the type @xmath247 for all @xmath248 . the matrix @xmath238 , defined in equation ( [ eq : j_g ] ) , is positive definite . the quantities @xmath249 and + @xmath250 are bounded for all @xmath251 and for all @xmath248 . 6 . for almost all @xmath245 , there exists functions @xmath252 , @xmath253 , @xmath254 that dominate , in absolute value , @xmath255 for all @xmath256 and which are uniformly bounded in expectation with respect to @xmath1 and @xmath257 for all @xmath248 . the functions @xmath258 and @xmath259 with @xmath260 are uniformly bounded for all @xmath248 . then we have the following theorem stating the consistency and asymptotic normality of the mgsde under the discrete models . for brevity in presentation , its proof has been moved to the appendix . [ thm : gs - divergence_asymptotic ] under the set - up of discrete models as mentioned above and assumptions ( a1)(a7 ) , the following results hold : 1 . there exists a consistent sequence @xmath261 of roots to the minimum gsd estimating equation ( [ eq : discrete_est_equation ] ) . the asymptotic distribution of @xmath262 is @xmath263-dimensional normal with vector mean @xmath264 and covariance matrix @xmath265 . under the assumption of theorem [ thm : gs - divergence_asymptotic ] , let us further assume that the true pmf @xmath1 belongs to the model family @xmath7 with @xmath188 . then , @xmath266 has a simpler asymptotic distribution given by @xmath267 , where @xmath268 , @xmath269 and @xmath270 . interestingly , the asymptotic distribution of the mgsde at the model is independent of the parameters @xmath164 and @xmath96 . this is also expected from the fact that first order influence functions of these estimators do not depend on @xmath164 and @xmath96 . hence , all the mgsde with the same @xmath69 ( but with different @xmath164 and @xmath96 ) have the same asymptotic efficiency at the assumed model and this efficiency is also quite high for all most common parametric models for small positive values of @xmath69 . this can be seen by noting the fact that the asymptotic variance and hence the efficiency at the model is exactly same as that of the msde or the minimum dpd estimators with the same value of @xmath69 and their high efficiencies have already been illustrated by ( * ? ? ? * table 1 ) and ( * ? ? ? * table 9.1 ) respectively . in this section , we will present a simulation study to describe the finite sample performances of the proposed mgsdes for different choice of the parameters @xmath69 , @xmath164 and @xmath96 . since the minimum gsd estimation under continuous model requires complicated kernel smoothing techniques , in this paper we will consider the discrete model only so that the minimization of the gsd measure can be done simply by using the relative frequencies . let us consider the poisson distribution with the true parameter value @xmath271 . we generate samples of size @xmath272 from the @xmath218 distribution and compute the minimum gsd estimators of @xmath32 for different values of the tuning parameters @xmath69 , @xmath164 and @xmath96 . repeating the process 1000 times , we compute the empirical bias and mse of the mgsdes of @xmath32 as given by @xmath273 where @xmath274 denotes the mgsde obtained at the @xmath275-th replication . these empirical biases and the mses are reported in table [ tab : sim0 ] for some particular values of tuning parameters . these show the performance of the proposed mgsdes under pure data where there are no outlier issues . with any sample of finite size , the poisson model inevitably leads to zero frequencies for infinitely many cells . some of the divergences within our gsd family have positive powers of the density @xmath1 in the denominator and consequently are not defined when there are one or more empty cells . this includes , for example , all cressie - read divergences with @xmath276 . the empty cell problem is a specific and extreme case of the general inlier problem " , where the observed frequencies are substantially lower than the expected frequencies . the gsd with @xmath68 , @xmath124 and @xmath222 is a case in question , where the divergence is not defined for empty cells . in order to keep a clear focus in the present context , we defer the consideration of the inlier issue in relation to the gsd to be taken up on a future occasion . next we examine the robustness of our proposal with respect to contamination/ outliers in the observed data . so , we repeat the above simulation study but after randomly contaminating 10% of each sample by observations from a @xmath277 distribution ; the latter distribution is well separated from the true data generating distribution , which is poisson(5 ) . the corresponding empirical bias and mse values obtained are reported in table [ tab : sim10 ] . if these bias and mse under the contaminated scenario turn out to be similar to that obtained in case of pure data that will indicate the stability / robustness of the corresponding estimators against the insertion of outlying observations . the major findings from these simulation results given in tables [ tab : sim0][tab : sim10 ] about the robustness of our proposed mgsdes along with their pure data performances have been listed below . 1 . under pure data , the absolute bias is minimum at the maximum likelihoods estimator ( mle , corresponding to @xmath278 ) as expected . however most of the other members of the gsd family also generate competitive values of the empirical biases the mse under pure data is also minimum at mle and it increases as @xmath69 increases for any fixed @xmath48 and @xmath96 . this fact is also expected from the ( theoretical ) asymptotic variance of the mgsdes which increases with @xmath69 at the model . however , the mses of most of the other mgsdes are also quite competitive and hence the loss in efficiency under pure data is not a very serious concern while using those mgsdes . 3 . under the contaminated scenario , the bias and mse become quite high for the mle with respect to the pure data case . but several other members of the gsd family generate stable estimators that do not depart significantly away from their values in pure data . the stable mgsdes generally correspond to the larger choice of @xmath69 and @xmath96 . in particular , both the bias and mse generally show a decreasing pattern as we increase the value of the tuning parameters @xmath69 or @xmath96 . thus , in terms of stability our proposed estimators appear to do better as @xmath69 ( or @xmath96 ) increases when the other two parameters are held constant . the effect of the third tuning parameter @xmath164 on the robustness of the mgsdes is also quite interesting . for larger values of @xmath96 and @xmath69 , there does not have any significant effect of @xmath164 on the robustness of the estimator . but for small values of @xmath69 or @xmath96 , @xmath164 affects the robustness of the corresponding mgsdes significantly the mgsdes with negative values of @xmath164 still generate stable bias and mses under contamination indicating a strong degree of robustness , but those corresponding to positive @xmath164 become even more unstable than the mle . [ tab : sim0 ] therefore , combining the above findings , we can see that the proposed minimum gsd estimators are highly robust under data contaminations and yield only a small loss in efficiency under pure data . this makes our proposal very useful in real practice with contaminated data , generating highly robust inference . further note that , under the contamination scenario considered here , the best estimators within the mgsde family in terms of both bias and mse correspond to the generalized @xmath0-divergence measure with @xmath279 , @xmath280 and @xmath281 . roughly , estimators within these choices of tuning parameters appear to provide the best compromise between efficiency at the model and robustness under data contamination . interestingly , these gsd measures do not appear to belong to any of the existing divergence measures like pd , gkl and @xmath0-divergences ; in fact they are far separated from the existing ones . hence the development of this larger gsd family does not limited only to a theoretical generalization in an academic interest . rather , they produce new minimum gsd estimators which generates more robust inference compared to the other existing minimum divergence estimators . another interesting findings of this simulation exercise is that the robustness of the proposed mgsdes depends directly on all the three tuning parameters @xmath69 , @xmath164 and @xmath96 . however , the classical ( first order ) influence functions of the mgsdes predict robustness behavior to be independent of the tuning parameters @xmath96 and @xmath164 . this clearly indicates the limitation of the ( first order ) influence function analysis in assessing the robustness of the minimum gsd estimators . [ tab : sim10 ] in this paper , we have discussed the divergence based model adequacy tests and its link with some new divergence families . we have shown that the recent general family of the @xmath0-divergence can also be obtained from suitable model adequacy tests . we have also considered a more general model adequacy test based on the @xmath0-divergence measures which further generates a much larger superfamily of divergences that contains the @xmath0-divergence family and also the well - know generalized kullback - leibler divergence family as its special case , among others . we have also discussed several interesting properties of this new divergence family that we call the generalized @xmath0-divergence family ( gsd ) . we have indicated also the potential application of the gsd family in robust parametric inference ; derived the asymptotic properties of the minimum gsd estimators and illustrated their robustness through influence function analysis and suitable numerical illustrations . in this pursuit , we have observed and demonstrated the limitations of the first order influence function in assessing the robustness of these minimum divergence estimators . the research of this paper in fact opens up a lots of interesting questions for future researchers , both mathematicians and statisticians . the new generalized @xmath0-divergence family is seen to have several interesting properties with some identical members within this family . so , it would be an interesting future problem to investigate the topological properties of the gsd measures to characterize the distinct members of this divergence family . further , the application of the minimum gsd estimators in several statistical inference problems other than parametric estimation will also provide a great value addition to the literature of the robust inference . however , the most interesting future work in this case would be to generalize the connection between the model adequacy test and resulting divergence family . we have seen that starting from a model adequacy test based on one particular divergence one gets a one - parameter family of divergences ( pd family or the gkl family ) ; the model adequacy test based on a one parameter divergence family ( skl or sld ) generates a two - parameter divergence family ( @xmath0-divergence family ) and staring from the model adequacy test based on the two parameter @xmath0-divergence family we obtained a three - parameter divergence family . up to this point it seems meaningful since each time we generate a divergence which is better than the existing ones in terms of combining good efficiency and robustness properties . so , a natural question in this regard is how long we can extend this fact to generate even larger and larger superfamily of divergences . noting that all the members of the three parameter gsd family are not distinct , this process may eventually end after finite steps where we will get the largest such divergence family including all possible distinct divergences ; however this needs many more research to arrive at any such conclusions . we hope to pursue some of these extensions in our future endeavors . we will assume that the assumptions ( a1)(a7 ) hold and consider some useful lemmas first from @xcite . 1 . @xmath285 \le n^{\frac{k}{2}}e_g[|\delta_n(x ) - \delta_g(x)|]^k \le \left [ \frac{g(x)(1-g(x))}{f_{\theta}^2(x)}\right]^{\frac{k}{2}}$ ] . @xmath286 \le \frac{2g(x)(1-g(x))}{f_{\theta}(x)}$ ] . now , let us define the quantities @xmath289 and + @xmath290 , where @xmath291 and @xmath292 note that these two quantities @xmath293 and @xmath294 are different from those considered in @xcite in the context of msdes since the function @xmath189 involved here is completely different from that in case of @xmath0-divergences . so , the next two lemmas , which yield the asymptotic distributions of these quantities in analogue to lemma 3 and lemma 4 of @xcite , needs suitably adjusted proof as described below . * proof :* by lemma 2.15 of @xcite ( or , * ? ? ? * lemma 25 ) , there exists some positive constant @xmath299 such that @xmath300 also , by lemma [ lem : lemma_3.1 ] , @xmath301 \le \beta \frac{g^{1/2}(x)}{f_{\theta}(x)}$ ] . + and by lemma [ lem : lemma_3.2 ] , @xmath302 = \beta e_g[\eta_n(x ) ] \rightarrow 0 $ ] as @xmath287 . thus we get , @xmath303 f_{\theta}^{1+\alpha}(x)|u_{\theta}(x)| \nonumber \\ & \le & \beta \sum_x g^{1/2}(x ) f_{\theta}^{\alpha}(x)|u_{\theta}(x)| % \nonumber \\ & < & < \infty ~~\mbox { ( by assumption ( a5))}. \nonumber \end{aligned}\ ] ] so , by dominated convergence theorem ( dct ) , @xmath304 as @xmath287 . + hence , by markov inequality , @xmath305 as @xmath287 . * proof :* note that , by the previous lemma [ lem : lemma_3.3 ] , the asymptotic distribution of @xmath293 and @xmath294 are the same . now , we have @xmath308\right ) \nonumber \\ & ~\displaystyle\mathop{\rightarrow}^\mathcal{d } & z \sim n ( 0 , v_g ) ~~~ \mbox{[by central limit theorem ( clt)]}. \nonumber ~~~~~~~~~~~~~~~~{\square } \end{aligned}\ ] ] now , we proof our theorem [ thm : gs - divergence_asymptotic ] using these lemmas . + * proof of consistency ( part(a ) ) : * consider the behavior of @xmath309 on a sphere @xmath310 which has radius @xmath311 and center at @xmath244 . we will show , for sufficiently small @xmath311 , the probability tends to one that @xmath312 so that the gsd measure has a local minimum with respect to @xmath32 in the interior of @xmath310 . at this local minimum , the estimating equation must be satisfied . hence , for any @xmath313 sufficiently small , the minimum gsd estimating equation has a solution @xmath261 within @xmath310 with probability tending to one as @xmath287 . now taking taylor series expansion of @xmath309 about @xmath314 , we get @xmath315 where @xmath316 lies between @xmath244 and @xmath32 . now , let us consider each terms one - by - one . for the linear term @xmath317 , we consider @xmath318 where @xmath319 is @xmath320 evaluated at @xmath314 and @xmath189 is defined in ( [ eq : k_delta ] ) . here , we will prove that @xmath321 as @xmath287 and note that the right hand side of above is zero by definition of the minimum gsd estimator . clearly by assumption ( a7 ) and the fact that @xmath322 almost surely ( a.s . ) by strong law of large number ( slln ) , it follows that @xmath323 for any @xmath324 in between @xmath319 and @xmath325 , uniformly in @xmath245 . so , by using the one - term taylor series expansion , @xmath326 however , by lemma [ lem : lemma_3.1](1 ) , we have @xmath327 \le \frac{\left[g(x)(1-g(x))\right]^{1/2}}{f_{\theta^g}(x ) \sqrt n } \rightarrow 0 , ~~~ ~~~ as ~ ~ n \rightarrow \infty.\nonumber \end{aligned}\ ] ] and , by lemma [ lem : lemma_3.1](2 ) along with assumption ( a5 ) , we get @xmath328 % \nonumber \\ & & ~~~~~ \leq 2c_1 \sum_x g^{1/2}(x ) f_{\theta^g}^{\alpha}(x ) |u_{j\theta^g}(x ) | ~ < ~ \infty . \nonumber % \\ & & ~~~~~~~~~~~~~~~~~~~~~~\mbox{[by assumption ( a5 ) ] } \nonumber \end{aligned}\ ] ] hence , by dominated convergence theorem ( dct ) , we get , @xmath329 \rightarrow 0 , \nonumber \end{aligned}\ ] ] as @xmath287 . then the markov inequality yields the desired result . thus , we get @xmath330 therefore , with probability tending to one , @xmath331 , where @xmath311 is the radius of @xmath310 and @xmath263 is the dimension of @xmath32 . next let us consider the quadratic term @xmath332 . clearly @xmath333 . \nonumber \end{aligned}\ ] ] first we will prove that @xmath334 to see this we get , from assumption ( a7 ) as in ( [ eq:39 ] ) , that @xmath335 for every @xmath324 lying in between @xmath319 and @xmath325 , uniformly in @xmath245 . so , by using the one - term taylor series expansion , @xmath336 thus , we get @xmath337 since by assumption ( a5 ) , @xmath338 , the desired claim in ( [ eq:45 ] ) follows by the similar argument that were used to prove the claim in ( [ eq:38 ] ) . in a similar fashion , we also get @xmath339 so , combining ( [ eq:45 ] ) , ( [ eq:47 ] ) and ( [ eq:48 ] ) , we have @xmath340 but we know that @xmath341 here the absolute value of the first term in the above expression ( [ eq:50 ] ) is @xmath342 with probability tending to one . and , the second term in ( [ eq:50 ] ) is a negative definite quadratic form in the variables @xmath343 . letting @xmath344 be the largest eigenvalue of @xmath345 , the quadratic form is @xmath346 . combining them , we get appropriate constants @xmath347 and @xmath348 such that for @xmath349 , we have @xmath350 with probability tending to one . finally , considering the cubic term @xmath351 , we note that @xmath352 or , @xmath353 where @xmath354 . here we will prove that all the terms in the rhs of the above expression are bounded . for the first term ( [ i ] ) , we use ( [ eq:46 ] ) to get @xmath355 < \infty . ~~~~ \mbox{[by clt and assumption ( a6)]}\nonumber \end{aligned}\ ] ] thus term ( [ i ] ) is bounded . now for the second term ( [ ii ] ) , we again use ( [ eq:39 ] ) to get @xmath356 < \infty , ~~~~ \mbox{[by clt and assumption ( a6)]}\nonumber \end{aligned}\ ] ] so that term ( [ ii ] ) is also bounded . similarly the terms ( [ iii ] ) , ( [ iv ] ) , ( [ v ] ) and ( [ viii ] ) are bounded as in case of term ( [ ii ] ) and using ( [ eq:39 ] ) and assumption ( a6 ) . next for the term ( [ vi ] ) , we will consider the following : @xmath357 so that @xmath358 also , @xmath359 so , @xmath360 < \infty , ~ \mbox{[by assumption ( a6)]}\nonumber % \\ & & ~~~~~~~ \mbox{[by assumption ( a6 ) ] } , \end{aligned}\ ] ] where @xmath361 . thus the term ( [ vi ] ) is bounded and also similarly the rest of the terms ( [ vii ] ) , ( [ ix ] ) , ( [ x ] ) and ( [ xi ] ) are bounded . hence , we get @xmath362 on the sphere @xmath310 with probability tending to one . combining these three inequalities we have @xmath363 which is strictly negative for @xmath364 . thus , for any sufficiently small @xmath311 , there exists a sequence of roots @xmath365 to the minimum gsd estimating equation such that @xmath366 converges to one , where @xmath367 denotes the @xmath50-norm . it remains to show that we can determine such a sequence independent of @xmath311 . to see this , we let @xmath368 to be the root closest to @xmath244 . this exists since the limit of a sequence of roots is again a root by the continuity of the gsd measure . this completes our proof of the consistency part . + * proof of the asymptotic normality ( part ( b ) ) : * to prove the asymptotic normality , we have to expand @xmath369 in taylor series about @xmath314 to get @xmath370 where @xmath371 lies in between @xmath32 and @xmath244 . suppose @xmath261 is the solution of the minimum gsd estimating equation that is consistent by part ( a ) of the theorem . replace @xmath32 by @xmath261 in the above equation ( [ eq:59 ] ) so that the lhs of the equation becomes zero and we get that here , the first term within the bracketed quantity in the rhs of ( [ eq:60 ] ) converges to @xmath238 with probability tending to one and the second bracketed term is an @xmath373 term ( as shown in the proof of the consistency part ) . further lemma [ lem : lemma_3.4 ] gives us that @xmath374 f_{\theta^g}^{1+\alpha}(x)u_{\theta^g}(x ) \nonumber \\ & = & s_{1n}|_{\theta=\theta^g } \mathop{\rightarrow}^\mathcal{d } n_p(0 , v_g ) . \nonumber \end{aligned}\ ] ] bregman , l. m. ( 1967 ) . the relaxation method of finding the common point of convex sets and its application to the solution of problems in convex programming . _ 7 _ , 200217 . original article is in _ zh . vychisl . mat . mat . _ , * 7 * , pp . 620631 , 1997 . ghosh , a. , harris , i. r. , maji , a. , basu , a. and pardo , l. ( 2016 ) . a generalized divergence for statistical inference . _ bernoulli _ , to appear . pre - print technical report ( 2013 ) at _ biru/2013/3 _ , bayesian and interdisciplinary research unit , indian statistical institute , kolkata , india .

six examples of the compact , isolated high velocity clouds ( chvcs ) identified by braun and burton ( ) , but only marginally resolved in single dish data , have been imaged with the westerbork synthesis radio telescope . the 65 confirmed objects in this class define a dynamically cold system , with a global minimum for the velocity dispersion of only 70 km s , found in the local group standard of rest . the population is in - falling at 100 km s toward the local group barycenter . these objects have a characteristic morphology , in which one or more compact cores is embedded in a diffuse halo . the compact cores typically account for 40% of the line flux while covering some 15% of the source area . the narrow line width of all core components allows unambiguous identification of these with the cool condensed phase of , the cnm , with kinetic temperature near 100 k , while the halos appear to represent a shielding column of warm diffuse , the wnm , with temperature near 8000 k. we detect a core with one of the narrowest emission lines ever observed , with intrinsic fwhm of no more than 2 km s and 75 k brightness . from a comparison of column and volume densities for this feature we derive a distance in the range 0.5 to 1 mpc . we determine a metallicity for this same object of 0.04 to 0.07 solar . comparably high distances are implied by demanding the stability of objects with multiple cores , which show relative velocities as large as 70 km s on 30 arcmin scales . many of the compact cores show systematic velocity gradients along the major axis of their elliptical extent which are well - fit by circular rotation in a flattened disk system . two out of three of the derived rotation curves are well - fit by navarro , frenk , and white ( 1997 ) cold dark matter profiles . these kinematic signatures imply a high dark - to - visible mass ratio of 1050 , for mpc , which scales as . the implied dark matter halos dominate the mass volume density within the central 2 kpc ( 10 arcmin ) of each source , providing a sufficent hydrostatic pressure to allow cnm condensation . the chvc properties are similar in many respects to those of the local group dwarf irregular galaxies , excepting the presence of a high surface brightness stellar population .

astro-ph9912417

the possibility of an extragalactic deployment of high velocity clouds has been considered in various contexts by ( among others ) oort ( @xcite ) , verschuur ( @xcite ) , eichler ( @xcite ) , einasto et al . ( @xcite ) , giovanelli ( @xcite ) , bajaja et al . ( @xcite ) , burton ( @xcite ) , wakker and van woerden ( @xcite ) , blitz et al . ( @xcite ) , and braun and burton ( @xcite ) . blitz et al . ( @xcite ) interpret several general hvc properties in terms of the hierarchical structure formation and evolution of galaxies . in this context , the extended hvc complexes would be the nearby objects currently undergoing accretion onto the galaxy , while the more compact , isolated ones would be their distant counterparts in the local group environment . it is striking that the population of anomalous velocity splits naturally into two rather distinct components when observed with the spatial sampling and sensitivity of modern surveys , like the leiden / dwingeloo survey ( lds : hartmann & burton @xcite ) in the north and the parkes multibeam survey ( see putman & gibson @xcite ) in the south . with this new perspective , it has become clear that the large majority of the 561 objects previously cataloged as distinct hvcs by wakker and van woerden ( @xcite ) are the local , low contrast maxima of extended diffuse complexes with angular sizes of tens of degrees . in addition to this diffuse , yet lumpy , component there appear to be several hundred intrinsically compact objects with angular sizes of about 1 degree . objects of intermediate angular size appear to be rather rare . braun and burton ( @xcite ) identified and confirmed 65 examples of compact , isolated high velocity clouds ( the chvcs ) which plausibly represent a homogeneous subsample of the high velocity clouds , in a single physical state , and arguably before their physical properties have been strongly influenced by the radiation field of the milky way or of m31 , or by a gravitational encounter with one of these major systems . braun and burton showed , in particular , that the velocity dispersion of the chvc sample is minimized in a reference frame consistent within the observational errors to the local group standard of rest . this minimization provides a quantitative demonstration of local group deployment . within this frame , the chvc ensemble is dynamically quite cold , with a dispersion of only 70 km s@xmath0 , although strongly infalling into the local group barycenter at a velocity of about 100 km s@xmath0 . most hvcs have been identified simply by their anomalous velocity emission in total power surveys at an angular resolution of 05 or coarser . information on characteristic intrinsic linear scales , on resolved spectral properties which might reveal , for example , opacity information , and on such kinematic properties as intrinsic widths or possible rotation , have been largely unavailable . of the sample of 65 compact , isolated hvcs catalogued by braun and burton ( @xcite ) , only two had been subject to interferometric imaging . wakker and schwarz ( @xcite ) used the westerbork array to show that both chvc114@xmath410@xmath4430 and chvc111@xmath406@xmath4466 are characterized by a core / halo morphology , with only about 40% of the single dish flux recovered on angular scales of tens of arcmin , and , furthermore , that the linewidths of the single dish spectra of these two sources were resolved into components of some 5 width or less . both of the imaged systems display systematic velocity gradients along the major axis of an elliptical distribution , which wakker and schwarz judged to be suggestive of rotation in self gravitating systems at local group distances . if the chvc objects are in fact a population of unevolved sub - dwarf galaxies scattered throughout the local group , then they might reveal some morphological characteristics which would not be consistent with the expectations of other suggested scenarios , in particular for objects ejected by a galactic fountain ( e.g. shapiro and field @xcite , bregman @xcite ) or located within the galactic halo ( moore et al . @xcite ) . the westerbork imaging discussed in this paper reveals a characteristic core / halo morphology for the chvcs , very narrow linewidths in the cores , and , in many cases , a signature of rotation . these properties are strongly suggestive of self - gravitating systems at local group distances , and specifically resemble the gaseous components of some dwarf galaxies . on the contrary , if the chvc objects were produced relatively locally by an energetic mechanism responsible for a galactic fountain , then their properties might have been expected to include large linewidths , motions not ordered by rotation , and a characteristic morphology other than that of a core / halo . high resolution imaging also makes it possible to provide specific targets for optical observations . deep optical imaging would help clarify the distinction between the chvcs and ( sub)dwarf galaxies , and any indication of a stellar population would allow a direct distance determination . if the chvcs are at local group distances , then the diffuse h@xmath5 emission surrounding them is expected to be weaker than that associated with high velocity gas in complexes lying within the halo of the milky way or extending from the magellanic clouds ( bland - hawthorn & maloney @xcite ) . furthermore , high resolution imaging is necessary in order to interpret existing observations aimed at detecting an hvc in absorption against a star in the halo of the milky way or against an extragalactic background source of continuum radiation . in either case , the absorption experiment targets an extremely small area . it is not clear if the failure to detect absorption in at least some of the published accounts should be attributed to the hvc in question not in fact covering the continuum source , rather than to the metallicity characterisics of the hvc ( for the cases of a negative result toward an extragalactic background sources ) or to the hvc being at a larger distance ( for the cases of a negative result toward halo stars ) . the results shown below indicate that only a small portion of each object is characterized by the high column density emission from the chvc cores , and thus that selection of suitable probes for absorption line measurements must take the actual column density distribution into account . if the compact hvcs discussed here are distributed throughout the local group as primordial objects , either surviving as remnants of the formation of the local group galaxies , or still raining in to these systems in a continuing evolution , then such objects would also be expected near other groups of galaxies . several searches for such objects beyond the local group are currently in progress . the results presented here on the small angular sizes of the chvcs generally and the even smaller surface covering factor of the high column density cores must be considered when designing an optimum search strategy or when calculating the expected detection statistics for a given experiment directed at other galaxies or groups of galaxies . a conclusive detection experiment of this type has not yet been carried out . our discussion is organized as follows . we begin by describing the method of sample selection in [ sec : samp ] , proceed with a description of the newly acquired observations in [ sec : data ] and continue with a presentation and discussion of our results in [ sec : results ] and [ sec : disc ] . we close by briefly summarizing our results in [ sec : summ ] . because high resolution imaging had previously only been obtained for two objects of the chvc class , we chose a sample from the catalog of braun and burton ( @xcite ) which spanned a wide range of the source properties , paying particular attention to source position both on the sky and in radial velocity , as well as to the physical attributes of linewidth and total flux density . although only six chvc sources were imaged in our program , the sources are distributed widely in galactic coordinates , span radial velocities of @xmath6 , vary in linewidth from 6 to 95 , and in line flux from 25 to 300 jy . the source properties are summarized in table [ tab : sample ] . cccccc name & ra(2000 ) & dec(2000 ) & lds flux & lds fwhm & lds structure + chvc@xmath7 & ( h m ) & ( ) & ( jy ) & ( ) & ( a @xmath8 b @ pa ) + 069 + 04@xmath4223 & 19 50.1 & 33 41 & 86 & 34.0 & 10 @xmath8 09 @ 330@xmath9 + chvc115 + 13@xmath4275 & 22 56.9 & 74 33 & 105 & 95.4 & 11 @xmath8 06 @ 114@xmath9 + chvc125 + 41@xmath4207 & 12 24.0 & 75 36 & 235 & 5.9 & 13 @xmath8 07 @ 292@xmath9 + chvc191 + 60 + 093 & 10 36.9 & 34 10 & 66 : & 29.6 & 10 @xmath8 09 @ 170@xmath9 + chvc204 + 30 + 075 & 08 27.5 & 20 09 & 305 : & 33.9 & 08 @xmath8 06 @ 215@xmath9 + chvc230 + 61 + 165 & 10 55.2 & 15 28 & 26 : & 29.3 & 14 @xmath8 07 @ 182@xmath9 + observations of the six chvc fields were obtained with the wsrt array between 12/11/98 and 17/12/98 . one twelve hour integration was obtained for each field in the standard wsrt array configuration having a shortest baseline of 36 meters . on source integration was bracketed by half hour observations of the calibration sources 3c286 and 3c48 . at the time of the observations , 11 of the 14 telescopes of the array were equipped with the upgraded receivers having a system temperature of about 27 k in the 1150 to 1850 mhz band . ( as of 11/3/99 the entire array is equipped with a receiver package covering the range from 250 mhz to 8 ghz with eight feed systems mounted in a prime focus turret . ) the interim wsrt correlator system available at the time was used to provide 256 uniformly weighted spectral channels in two linear polarizations across 2.5 mhz centered on the @xmath10 velocity of each source . the effective velocity resolution was 1.2 times the channel spacing of 2.06 km s@xmath0 . standard gain and bandpass calibrations were carried out after editing the data and deleting all baselines effected by shadowing . a component model for the continuum emission from each field was derived from an image made from the average of the line free spectral channels . this model was subtracted directly from the visibility data . the block of spectral channels containing line emission was imaged with a visibility based clean deconvolution proceeding down to a flux level of twice the rms noise level at a variety of spatial resolutions . uniform weighting of the visibility data was employed , together with a series of gaussian tapers decreasing to 30% amplitude at projected baselines of 1.25 , 2.5 , 5 , and 10 k@xmath11 , resulting in spatial resolutions of about 120 , 60 , 30 , and 20 arcsec , respectively . in a few cases , some residual continuum emission was still present in the data cubes ; in those cases , several spectral channels from both edges of the cube were averaged together and subtracted from the entire cube . the typical rms noise levels in the deconvolved cubes were 2.0 , 1.7 , 1.4 , and 1.3 mjy per beam per spectral channel at the four spatial resolutions noted above . the corresponding brightness sensitivities were , respectively , 0.085 , 0.29 , 0.95 , and 2.0 k in a single channel of 2.06 km s@xmath0 width . ( flux per beam and brightness temperature are related as usual by @xmath12 , or @xmath13 for @xmath14 , the beam area in arcsec@xmath15 . ) expressed as an optically thin column density , the sensitivities correspond to 0.32 , 1.1 , 3.6 , and 7.4 @xmath16 @xmath17 , respectively , for emission filling the beam and extending over a single velocity channel . while some emission profiles originating in the cool atomic phase are indeed as narrow as the single channel velocity width ( as we will see below ) , emission profiles originating in the warm neutral phase will have substantially broader linewidths . for example , the thermal linewidth of a component with kinetic temperature , @xmath18 k , will be 21 km s@xmath0 fwhm . at this velocity resolution , our column density sensitivity is degraded to 1.0 , 3.5 , 11 , and 24 @xmath19 @xmath17 . this somewhat counter intuitive correspondence of column density sensitivity with the expected linewidth must be borne in mind . narrow lines are significantly easier to detect than broad ones . moment images ( zero , first , and second ) were generated from each cube after employing a blanking criterion for inclusion of each pixel in the weighted sum . this involved demanding a brightness in excess of about 2@xmath20 after smoothing the cube by a factor of three both spatially and in velocity . images of integrated emission were corrected for the primary beam response of the instrument , which is well approximated ( at 1420 mhz ) by a circular gaussian with 2110 arcsec fwhm . while structures extending over as much as 10 arcmin in a single spectral channel were adequately recovered in the resulting images , there were also indications of the presence of more diffuse features which could not be adequately imaged . these were apparent due to the artifacts they induced ; specifically , the so called short spacing bowl surrounding regions of bright extended emission . even with careful windowing during the deconvolution it was not possible to completely eliminate such artifacts . consequently , a significant fraction of the total flux detected in the single dish observations could not be recovered . the integrated flux detected in the reconstructed images after primary beam correction varied from less than 1% to as much as 55% of that detected in the lds . the percentages of total fluxes contributed by the dense cores are indicated for each observed field in table [ tab : properties ] . ccccc name & wsrt flux & core flux & core area & distance + chvc@xmath7 & ( jy ) & % & % & ( kpc ) + 069 + 04@xmath4223 & 21.9 & 26 & 3.4 & @xmath21(4050 ) + chvc115 + 13@xmath4275 & 20.7 & 20 & 10 . & @xmath21(40 - 100 ) + chvc125 + 41@xmath4207 & 130 . & 55 & 25 & 5001000 + chvc191 + 60 + 093 & 0.22 & 0.3 & 0.2 & @xmath217 : + chvc204 + 30 + 075 & 116 . & 38 & 48 & @xmath21(40130 ) + chvc230 + 61 + 165 & 0.65 & 2.5 & 0.3 & @xmath214 : + a crude attempt was made to correct the interferometric images of integrated for the non detection of diffuse emission features . an elliptical gaussian with dimensions and orientation as measured in the lds and an integral flux sufficient to recover the lds total was added to the interferometric images . these parameters are summarized for the six fields in table [ tab : properties ] . the resulting composite images are presented in turn below . the lowest column density contours in these composite images follow the elliptical outlines of the lds source model , while the higher levels are dominated by the compact structures detected interferometrically . although not appropriate for detailed analysis on spatial scales between about 10 and 30 arcmin , these composite images are consistent with all of the data currently in hand and are indicative of the column densities likely to be present in diffuse structures as well as their overall spatial extent . in at least one case ( discussed below ) we were able to verify with independent data that the resulting composite provided a reasonable representation of the source structure and extent . in this section , we briefly summarize the structural and kinematic data obtained for each of the six observed fields using the wsrt . the moment images and a few representative spectra for this field are shown at 1 arcmin resolution in fig . [ fig : h069o ] . the integrated distribution is dominated by a bright elongated concentration ( clump a ) of some 15 arcmin extent . several fainter knots lie about 10 arcmin further to the south , and are connected to each other by a low brightness bridge of emission ( more clearly seen in fig . [ fig : h069 m ] ) . spectra toward some of the brightest emission regions all have ( @xmath22 ) peak brightnesses of about 2 k , ( @xmath23 ) relatively broad line profiles of about 15 km s@xmath0 fwhm , and ( @xmath24 ) a centroid velocity near @xmath4235 km s@xmath0 . at several local @xmath25 maxima within clump a the profiles become doubly peaked . these will be referred to again below . the velocity field displays at least one systematic trend : a moderate velocity gradient , oriented along pa 30@xmath9 ( east of north ) along the major axis of clump a from about @xmath4230 to @xmath4240 km s@xmath0 and extending over some 10 arcmin . the distribution of profile linewidth shows that clump a has a moderately high linewidth throughout , but also has several distinct regions with apparent dispersions as high as 8 km s@xmath0 . these correspond to regions of line splitting rather than to simple line broadening , while the overall velocity centroid is unchanged . two such regions are observed , centered near @xmath26 and @xmath27 . the two peaks in the line profile are separated by about 10 km s@xmath0 in both cases , while the angular extent is only some 90 arcsec . the integrated distribution in chvc069 + 04@xmath4223 , after inclusion of a simple representation of the diffuse emisison derived from the lds data , is shown in fig . [ fig : h069 m ] . the brightest knots have column densities just in excess of 10@xmath28 @xmath17 , while the diffuse underlying envelope peaks at about @xmath29 @xmath17 . the basic data for this field are shown at 2 arcmin resolution in fig . [ fig : h115o ] . this object is resolved in the wsrt data into a collection of approximately 10 sub structures . each clump is between about 1 and 10 arcmin in size , while the ensemble of clumps is distributed over a region of about 30 arcmin diameter . spectra towards several of the clumps show that each feature has a peak brightness of about 1 k and an intrinsic velocity width of between 10 and 15 km s@xmath0 fwhm . on the other hand , each feature has a distinct centroid velocity so that the collection spans the interval of @xmath4300 to @xmath4220 km s@xmath0 . the global distribution of the line of sight velocities shows no obvious pattern . high , low , and intermediate velocity clumps are distributed across the field . some of the most extreme velocities do occur in close spatial proximity , however , near @xmath30 ; but comparable velocities are also seen elsewhere . more careful examination of the individual clumps reveals that many of them have significant velocity gradients oriented preferentially along their long axes . this is true for the clumps centered near @xmath31 , @xmath32 , @xmath33 , and @xmath34 . the typical gradient observed is some 10 km s@xmath0 over an angular extent of 5 arcmin . the image of velocity dispersion shown in the lower left panel of fig . 3 serves primarily to illustrate the regions where multiple velocity components overlap along the line of sight . the morphology of the regions having multiple velocity components does not seem peculiar , but instead suggests simply line of sight overlap of distinct features rather than a physical interaction . the composite image of integrated in chvc115 + 13@xmath4275 is shown in fig . [ fig : h115 m ] . column densities in the knots do not exceed about @xmath35 @xmath17 , while the broad underlying distribution peaks between 5 and @xmath36 @xmath17 the compact high velocity cloud chvc125 + 41@xmath4207 is particularly interesting . because of the high brightness of this source , higher angular resolution could be employed for the data presentation . basic data at 28 arcsec resolution are shown in fig . [ fig : h125o ] . the source has a complex filamentary morphology extending over some 45 arcmin . several compact cores of only 1 to 2 arcmin extent are seen within the eastern segment ( clump a ) of the complex . the spectrum toward the brightest compact core is quite remarkable , in having a peak brightness of 75 k as well as a linewidth which is completely unresolved with our effective velocity resolution of 2.47 km s@xmath0 . this width is one of the narrowest ever measured for the emission line . velocity channels adjacent to the line peak have intensities down to about 20% of maximum while being spaced by only 2.06 km s@xmath0 . a spectrum toward the more diffuse western knot ( clump b ) has a more modest peak brightness of some 30 k and a linewidth of about 6 km s@xmath0 fwhm . the most prominent systematic trend in the velocity field is a north south gradient over clump b of about 8 km s@xmath0 over 10 arcmin . clump a has a shallower north south velocity gradient of about half this magnitude . the fainter filaments have velocities which do not follow this simple pattern . the image of velocity dispersion delineates those regions with extremely narrow linewidth from those which are only moderately narrow . both of the compact cores in the eastern complex as well as several other more diffuse structures are completely unresolved in velocity . clump b is less extreme in this regard . one small region of clump b , centered near @xmath37 , deserves special mention . this minor local maximum in the @xmath25 distribution displays clear velocity splitting in the emission profile , amounting to some 5 km s@xmath0 over a region of about an arcmin in diameter , as shown in fig . [ fig : h125s ] . a second region displaying a similar phenomenon is centered near @xmath38 . both regions are about 90 arcsec in diameter . the composite distribution of _ apparent _ column density is shown in fig . [ fig : h125 m ] . the emphasis on the term `` apparent '' is particularly apt : as we will see below , the compact cores are likely to have opacities of about 2 , so that actual column densities are about twice the apparent ones in these directions . even so , the apparent column density exceeds @xmath39 @xmath17 over much of the source , while the underlying halo reaches values of more than @xmath29 @xmath17 near the knots . in this particular case , we can obtain some quality assessment of our simple composite image of @xmath25 , in which we have attempted to compensate for the limited sensitivity to diffuse structures inherent in our interferometric data . this source has recently been observed with the lovell telescope at jodrell bank by de vries et al . ( @xcite ) . comparison of the @xmath29 @xmath17 contour in the jodrell bank image with our composite image shows good agreement in both source size and orientation . the data for this compact hvc are shown at 1 arcmin resolution in fig . [ fig : h204o ] . several moderately bright clumps are seen over a region of some 35 arcmin in size . the large angular extent of this object in a given velocity channel challenges reconstruction of the brightness distribution with the single 12hour wsrt coverage which we obtained . some questions remain regarding the reconstruction fidelity , particularly in regard to the most northerly feature which extends down to the 10% level of the primary beam , and as such , will have had any imperfections magnified by almost a factor of 10 in the displayed result . the spectra toward the more compact local maxima have peak brightnesses between 2 k and 8 k and linewidths which are generally some 15 km s@xmath0 fwhm . doubly peaked spectra are seen in two regions ; in one case where two components seem to overlap along the line of sight and in another where localized line splitting is observed ( as noted below ) . the large elliptical feature in the south central part of the field ( clump a ) shows a well defined north south velocity gradient running from about 55 to 80 km s@xmath0 over some 12 arcmin . the large elliptical feature in the north east ( clump b ) has a similar gradient along pa @xmath435@xmath9 running from 45 to 70 km s@xmath0 over 20 arcmin . four other major clumps with more nearly discrete line of sight velocities fill out this system . velocity dispersions are typically substantially less than 10 , except near @xmath40 where clump a appears to overlap with the smaller south easterly clump , and near @xmath41 where two velocity peaks separated by about 20 km s@xmath0 are seen over the 60 arcsec extent of this local maxima in @xmath25 . column densities of a few times @xmath42 @xmath17 are seen toward the brighter knots in the composite reconstruction shown in fig . [ fig : h204 m ] . a very substantial diffuse component is inferred for this field , reaching column densities of a few times @xmath43 @xmath17 over about 30 arcmin near the source centroid . the two final sources in our sample are both quite faint and primarily diffuse , albeit small in angular size . the wsrt data at 2 arcmin resolution are shown in fig . [ fig : h191o ] . low brightness clumps of 5 to 10 arcmin extent are detected in both these cases . peak brightnesses of less than 0.5 k are seen , coupled with moderately narrow linewidths of some 6 km s@xmath0 fwhm . the northern clump of chvc191 + 60 + 093 has a moderately strong velocity gradient along pa @xmath44 extending from 87 to 95 km s@xmath0 over 5 arcmin . the velocity field of chvc230 + 61 + 165 shows no clear systematic trend . the composite reconstructions shown in fig . [ fig : h191 m ] and fig . [ fig : h230 m ] indicate peak column densities of only about @xmath43 @xmath17 , embedded in a diffuse component of only 25 @xmath45 @xmath17 . the six compact , isolated high velocity clouds which we have imaged with moderately high spatial resolution span a wide range of source morphologies , but nevertheless share several noteworthy properties . in each case we detect a number of relatively compact clumps which range in size from a few arcmin to perhaps 20 arcmin in extent . these are local enhancements in the column density distribution with peak column densities in the range @xmath43 to @xmath46 @xmath17 . each clump is characterized by its own line of sight velocity and relatively narrow line widths at 60 arcsec resolution in the range of 2 to 15 km s@xmath0 fwhm . at the narrow linewidth extreme , very stringent limits on the kinetic temperature are implied which we discuss below . but even linewidths as large as 15 km s@xmath0 fwhm are significantly less than the 21 km s@xmath0 fwhm which corresponds to the 8000 k equilibrium temperature of the warm neutral medium ( wnm ) . based on these modest linewidths we can already conclude that the in the chvc clumps must be in the form of the cool neutral medium ( cnm ) with typical equilibrium temperatures in the range of 50 to 200 k ( e.g. wolfire et al . @xcite , @xcite ) . each clump has a smoothly varying line of sight velocity across its extent , often in the form of a systematic gradient along the major axis of an elliptical distribution . we will address the best resolved cases of velocity gradients below . most of the imaged chvc systems are composed of multiple clumps . since each clump has a distinct velocity , the total chvc linewidth is determined primarily by the velocity spread between the clumps , and to a lesser degree by systematic velocity gradients , rather than by the intrinsic linewidth of a single entity . there are a few cases where there appears to be partial line of sight overlap of two distinct clumps , leading to a doubly peaked emission profile within a single 60 arcsec beam . in the most extreme case studied so far , chvc115 + 13@xmath4275 , the inter clump velocity spread amounts to some 70 km s@xmath0 . this case will also be addressed in more detail below . in addition to the clumps , we see evidence , in all cases studied , for a substantial diffuse halo component of emission . the clumps account for as little as 1% to as much as 55% of the total emission which was detected in the single dish observations , although a typical flux fraction is about 40% , as was also found previously by wakker and schwarz ( @xcite ) for the two chvcs which they imaged . the halos have typical spatial fwhm of about 1 degree and column densities within the inner 30 arcmin between 2@xmath45 and 2@xmath47 @xmath17 . in table [ tab : properties ] we also include an indication of the surface area of the high colum density cores relative to the diffuse halos . this was defined as the ratio of the area with column density exceeding 5@xmath45@xmath17 in the wsrt - only images to the source area determined in the lds . these values vary over a wide range , but are typically lower by a factor of about two from the fractional flux , with a mean of 15% . the data currently available suggest a characteristic two phase structure . the diffuse morphology of the halo is consistent with a component of wnm shielding the embedded clumps of cnm . thermodynamic modeling of the ( which we address in the next subsection ) actually requires such a nested geometry of the warm and cool phases when the entire structure is immersed in an isotropic radiation field . comparison of the emission spectra observed interferometrically with those seen in total power supports this conjecture , since the diffuse components generally have linewidths exceeding the 21 km s@xmath0 fwhm thermal linewidth of an 8000 k gas . this remains a difficult issue to address in full with the limited data currently in hand . resolution , total power imaging with the up graded arecibo telescope would allow direct assessment of the degree of sub structure as well as the local lineshape of the halo component . this could enable unambiguous determination of the nature of the halo gas . a tight correlation of emission brightness temperature with opacity has been established for both the galaxy and for m31 ( braun and walterbos @xcite ) ; a comparable correlation pertains for the resolved high brightness structures seen in many nearby galaxy disks ( braun @xcite , @xcite ) which have a typical size of about 150 pc and a velocity fwhm of less than about 5 km s@xmath0 . the high brightness temperatures seen in chvc125 + 41@xmath4207 , amounting to about 75 k , are therefore already an indication that relatively high opacities are likely to be present . in this case a lower limit to the opacity in the bright cores can be inferred by comparing the kinetic temperature which follows from the linewidth with the emission brightness temperature . as illustrated in fig . [ fig : h125o ] , the linewidth in the cores of clump a is completely unresolved with the available 2.47 km s@xmath0 resolution . an upper limit to the intrinsic linewidth of about 2 km s@xmath0 fwhm corresponds to an upper limit on the kinetic temperature of 85 k. for an isothermal gas the brightness temperature , @xmath48 , is related to the spin temperature , @xmath49 , by @xmath50 ) ; assuming equality of the kinetic and spin temperatures allows a lower limit on the opacity to be derived of @xmath51 . this must be regarded as a firm lower limit since any line broadening by turbulence , which the linewidth observed already constrains to be of order 1 km s@xmath0 or less , would only serve to bring the implied kinetic temperature even closer to the observed brightness temperature , yielding a higher @xmath52 . similarly , excitation effects which might suppress the spin temperature with respect to the kinetic temperature would only serve to exasperate this problem . ( the interesting possibilities for constraining ism physics with this remarkable source have prompted us to request new wsrt observations with substantially higher velocity resolution . ) plausible lower limits to the actual column density toward the two compact cores in clump a of chvc125 + 41@xmath4207 are therefore about @xmath46 @xmath17 ( where the apparent column is some @xmath53 @xmath17 ) . the angular size of these features can be determined approximately from the apparent column density distribution shown in fig . [ fig : h125o ] . mean spatial fwhm extents of about 90 arcsec are observed , although their flat topped appearance suggests that opacity effects may already be partially responsible for this measured size . the fwhm of the actual column density distribution is likely to be somewhat smaller . these compact cores are seen in projection against more extended regions with apparent column densities of 1 to 2@xmath54 @xmath17 . the thermodynamics of in a variety of astrophysical settings has been studied by wolfire et al . ( @xcite ) . a stable cool phase of is expected when both @xmath55 a sufficient column of shielding gas is present , and @xmath56 the thermal pressure is sufficiently high . equilibrium kinetic temperatures of 85 k are predicted to occur over a range of volume densities , depending on the dust to gas ratio , on the gas phase metallicity , and on the intensity of the interstellar radiation field . these dependancies are illustrated in figs . 57 of wolfire et al . ( @xcite ) . since the thermal pressure in the mid plane of our galaxy in the solar neighborhood is fairly well constrained to be about @xmath57 @xmath58 k and this pressure is believed to decline substantially with height above the plane , an upper limit to the volume density of some 24 @xmath58 can be assigned to with 85 k kinetic temperature . assuming that the opaque cores in chvc125 + 41@xmath4207 have a roughly spherical geometry , which is indeed suggested by their projected spatial appearance and that the volume filling factor of cool is unity , then the column density , angular size , and volume density can be used to estimate the distance to the source , according to @xmath59 . this provides a lower limit to the distance to chvc125 + 41@xmath4207 of 31 kpc . of course at a distance of 30 kpc , the ambient thermal pressure is likely to have declined by more than an order of magnitude from the galaxy mid plane value ( e.g. wolfire et al . @xcite ) , leading to a corresponding linear increase in the distance estimate . at a potential distance of few 100 kpc , it becomes important to reconsider the ionization and heating conditions that would apply . specific calculations of equilibrium conditions within the local group environment were made available to us by wolfire , sternberg , hollenbach , and mckee ( private communication ) for two bracketing values of the neutral shielding column density , namely 1 and 10@xmath47 @xmath17 , a metallicity of 0.1 solar ( which we discuss further below ) , and a dust to gas mass ratio of 0.1 times the solar neighborhood value . equilibrium volume densities at @xmath60 k of 3.5 and 0.65 @xmath58 are found for shielding columns of 1 and 10@xmath47 @xmath17 , respectively , as can be seen in fig . [ fig : phasep ] . the corresponding distances are in the range of 210 to 1100 kpc . comparison with fig . [ fig : h125o ] suggests that shielding columns in the higher part of this range are most relevant for the two opaque cores under consideration , suggesting a distance in the range 0.5 to 1 mpc . we stress that the resulting distance estimate is not dependent on details of the thermodynamic calculation , but only on the assumed thermal pressure in the source environment . placing chvc125 + 41@xmath4207 at a distance of , say , 3 kpc would require assigning it a volume density of some 250 @xmath58 . the well constrained temperature of @xmath60 k , would then imply a thermal pressure , @xmath61 @xmath58 k , a value vastly exceeding those observed in the local ism . one might argue that a high pressure may be related to the high speed re entry of the object into the galaxy given its line of sight velocity in the local standard of rest frame of @xmath4200 km s@xmath0 . it is difficult to reconcile such a scenario with one of the narrowest emission lines ever detected , showing no measurable component of turbulent line broadening . a moderately low thermal pressure , on the other hand , amounting to @xmath62 @xmath58 k for this high latitude , non turbulent source , seems plausible on rather general grounds . a source distance in the range 0.5 to 1 mpc follows directly . in this case a consistent solution for the heating and ionization of the source is also in hand , as outlined above . unfortunately , this method of distance determination can not be applied more generally in our current sub sample . none of the cores in the other chvcs we have imaged has a comparably high opacity . the typical peak brightness seen in other cores amounts to only about 2 k , suggesting opacities of about 2% , for @xmath63 k. with such low opacity , the likelihood of encountering unit filling factor of the gas along the line - of - sight becomes vanishingly small , so that the comparison of column and volume densities gives only a very weak lower limit to the source distance . applying the same method blindly to the other cores in our current sample and assuming a volume density of 1 @xmath58 throughout , returns lower limits to the distance which vary substantially within an individual chvc as shown in table [ tab : properties ] . the method is particularly suspect for the low brightness cores ( 0.5 k peak ) of chvc191 + 60 + 093 and chvc230 + 61 + 165 , where lower limits of 7 and 4 kpc are found . elsewhere , lower limits to the distance in the range 40 to 130 kpc are implied . the source chvc125 + 41@xmath4207 is one of the few compact hvcs for which it has been possible to measure absorption against a background uv source . bowen and blades ( @xcite ) detected unsaturated absorption at a corrected velocity of @xmath64 ( bowen et al . @xcite ) toward the seyfert galaxy mrk 205 at @xmath65 . they derive a column density of @xmath66 @xmath17 for this component . this line of sight passes just to the south east of the high @xmath25 portion of clump b shown in fig . [ fig : h125 m ] . the column density we infer in this direction from our composite image is @xmath67 @xmath17 . wakker and van woerden ( @xcite ) quote a value of @xmath68 @xmath17 in a nearby direction . analysis of the recent jodrell bank data ( de vries et al . @xcite ) would provide an improved estimate . jenkins et al . ( @xcite ) provide extensive data for the gas phase abundance of relative to in the solar neighborhood . they derive relative gas phase abundances of log(@xmath69 and @xmath45.0 , respectively , for low and high volume density lines of sight . the relative abundance we infer , log(@xmath70 , is about a factor of 0.07 down from the local ism value appropriate for low volume densities . passing as it does through only the outer reaches of the atomic halo of chvc125 + 41@xmath4207 , this seems the most appropriate for comparison . we have taken care to refer to comparable measurements of gas phase metal abundance in comparing the chvc125 + 41@xmath4207 data with that of the solar neighborhood . the local ism value is believed to be substantially depleted by deposition on dust grains from its total value , log(@xmath71 . if the dust abundance of chvc125 + 41@xmath4207 is less than that in the solar neighborhood , which is quite likely in most formation scenarios , then dust depletion would presumably be less effective and the metal abundance might then lie somewhere in the range 0.04 to 0.07 relative to the solar value . an interesting kinematic phenomenon to emerge from these high resolution observations of the compact high velocity clouds is the occurrence of several spatially localized regions which display splitting of the emission profiles without a related change in the velocity of the centroid of the profile . an example of the phenomenon is given by the series of spectra shown in fig . [ fig : h125s ] . five such regions were detected , two in each of chvc069 + 04@xmath4223 and chvc125 + 41@xmath4207 , and one in chvc204 + 30 + 075 . each of these cases is associated with a relatively high column density , in excess of about @xmath42 @xmath17 . these regions are quite distinct from those instances of enhanced linewidth which are the result of localized line of sight overlap of extended features having different velocities . that phenomenon is also observed on several occasions ( as noted above ) but it can typically be recognized as such by the presence of two distinct velocity systems which are each spatially extended and have uncorrelated morphology . the angular extent of these split line regions is confined to 1 or 2 arcmin in all cases . the degree of velocity splitting varies from only about 5 km s@xmath0 in chvc125 + 41@xmath4207 , to 10 km s@xmath0 in chvc069 + 04@xmath4223 , and to as high as 20 km s@xmath0 in chvc204 + 30 + 075 . small depressions in the line core of galactic emission features have often been interpreted as evidence for so called self absorption by the cool opaque core of a warmer semi opaque atomic structure . such an interpretation does not seem appropriate for the phenomena under discussion , since the line strength between the peaks declines to a rather low brightness temperature . the spatially localized and symmetric distribution of line shapes seen in fig . [ fig : h125s ] seems more suggestive of an organized inflow or outflow with a substantial degree of spherical symmetry . at a distance of 700 kpc , these regions would have linear sizes of 300 pc , making them comparable to some of the low luminosity super shells detected in nearby galaxies which are powered by the supernovae and stellar winds of a stellar association ( e.g. mashchenko et al . @xcite ) . a common kinematic pattern observed within the chvc clumps which we have imaged is a velocity gradient oriented along the major axis of a roughly elliptical distribution . specific examples of such gradients were noted above for several of the objects imaged . the observed magnitudes of the velocity gradients vary from 0.5 to 2 km s@xmath0arcmin@xmath0 . the cases which are best resolved are clump a of chvc069 + 04@xmath4223 and clumps a and b of chvc204 + 30 + 075 , each of which is about 20 arcmin in angular size . this same pattern was also seen in the two chvcs imaged by wakker and schwarz ( @xcite ) , namely chvc114@xmath410@xmath4430 and chvc111@xmath406@xmath4466 . since these well resolved examples of systematic gradients in the velocity fields are reminiscent in form and amplitude of the `` spider '' diagrams seen for distributions in some dwarf galaxies , we carried out standard tilted ring fits to assess the extent to which they could also be modeled by rotation in a flattened disk system . in the usual way , we began by allowing all kinematic parameters to vary freely from ring to ring , and then solved sequentially for the best fitting kinematic center , for the systemic velocity , for the position angle of the receding line of nodes , and , finally , for the kinematic inclination . holding all of these best fitting parameters fixed , we then fit only for the rotation velocity of each ring . robust solutions for circular rotation were found in all three cases and these are shown in fig . [ fig : kinenfw ] . the fits display slowly rising rotation velocity with radius ; the rise is continuous to some 15 km s@xmath0 in the case of chvc069 + 04@xmath4223a , and flattens out at 500 to 600 arcsec to values of 15 and 20 km s@xmath0 for chvc204 + 30 + 075a and b. an estimate of the contained dynamical mass is given by @xmath72 , while the mass of gas , @xmath73 , where @xmath74 is the integrated flux in units of jykm s@xmath0 and there is an assumed contribution by helium of 40% by mass . at an assumed distance of 0.7 mpc , these three clumps have @xmath75 , @xmath76 , and @xmath77 m@xmath78 , @xmath79 , @xmath80 , and @xmath81 m@xmath78 , and dark to visible mass ratios of @xmath82 , 36 , and 29 , respectively . the derived value of @xmath83 scales with the assumed distance as @xmath3 , given the dependencies noted above . in addition to this crude estimate of dynamical mass , we can compare our derived rotation curves with those expected for a cold dark matter halo as parameterized by navarro , frenk , and white ( nfw , @xcite ) . these authors find that such halos tend to have a `` universal '' shape in their extensive numerical simulations , so that the density profile and resulting rotation velocity are determined simply from the halo mass . we have calculated halo properties using the ( web retrieved 19/09/99 version ) code of navarro , frenk , and white ( @xcite ) assuming @xmath84=0.3 , @xmath85=0.7 , h@xmath86= 60 km s@xmath0 mpc@xmath0 and a cluster abundance normalized fluctuation spectrum . the early plateau and subsequent rise in the derived rotation velocity of chvc069 + 04@xmath4223a is not well - fit by an nfw halo . for comparison , the expected rotation curve of a @xmath87 m@xmath78 ( within the virial radius of 3.4 kpc ) nfw halo is overlaid on the data . while this curve crudely matches the last few measured points , the shape at smaller radii deviates substantially . on the other hand , rather good agreement with the data is found for both chvc204 + 30 + 075a and b with halo masses of @xmath88 , and @xmath89 m@xmath78 ( within 9.3 and 12.6 kpc ) at an assumed distance of 0.7 mpc . these rotation curves are overlaid on the data in fig . [ fig : kinenfw ] . note how well the shape of these rotation curves is reproduced by the standard nfw halo profile . the very high total linewidth seen for chvc115 + 13@xmath4275 was found to arise from the broad range of distinct line of sight velocities of the ten individual clumps which make up this source . the angular separation of the clumps is some 30 arcmin , while their velocity centroids are separated by as much as 70 km s@xmath0 . since this source is very isolated in position and velocity ( see the appropriate panel of fig . 1 of braun and burton @xcite ) a chance superposition of ten unrelated components is unlikely . it is interesting to consider the stability of this collection of clumps under several scenarios . if , for instance , this collection were located at a distance of 5 kpc , the distribution of clumps would have a linear diameter of 44 pc and would double in size on a timescale of only @xmath90 yr . such a short dynamical timescale would imply that we were witnessing a rather special moment in the evolution of this source . if , on the other hand , this source were located at a distance of 0.7 mpc and were self gravitating , the angular radius and half velocity width of 15 arcmin and 35 km s@xmath0 could be used to calculate a dynamical mass from @xmath91 m@xmath78 . the corresponding mass of gas at this distance , @xmath73 is @xmath92 m@xmath78 . the dark to visible ratio at this assumed distance is @xmath83 = 51 , scaling , as in the previous subsection , with @xmath3 . a comparable analysis can be applied to chvc043@xmath93 , for which arecibo imaging data was obtained by giovanelli ( @xcite ) . this object has only two primary core components which extend over about 40 arcmin , and are separated by 60 km s@xmath0 in velocity , while the total line flux is 260 jy - km s@xmath0 . using the half - separations and line flux yields @xmath94 m@xmath78 , @xmath95 m@xmath78 and @xmath83 = 20 for @xmath96 = 0.7 mpc . since only two core components are involved , geometric effects of orientation are more likely to play a role in this calculation than in the case of chvc115 + 13@xmath4275 . the imaging obtained here for our chvc sub sample reveals striking similarities between the gas properties of the chvcs and those of low mass galaxies . the comparison with the local group dwarf irregulars , leo a and sagdig , studied at high resolution by young and lo ( @xcite , @xcite ) is particularly apt . both of these objects display the same general morphology and kinematics as the chvcs : high column density clumps characterized by narrow linewidths ( about 8 km s@xmath0 fwhm ) and angular sizes of a few arcmin embedded in a diffuse halo of lower column density gas with broader linewidths ( about 20 km s@xmath0 fwhm ) extending over 10 s of arcmin . young and lo reach the same conclusion presented here regarding the phase content of the dwarf irregulars , namely that cool condensations of cnm ( with @xmath97100 k ) are found within wnm envelopes ( with @xmath978000 k ) . the peak brightness temperatures detected from the cnm in leo a , 73 k , are almost identical to those we see in the compact clumps of chvc125 + 41@xmath4207 . the velocity fields of both leo a and sagdig are also very similar to those of the chvcs . modest velocity gradients of comparable magnitude are observed ( about 10 km s@xmath0 over 10 arcmin ) along one position angle in addition to a substantial unstructured component . unlike the chvcs , these galaxies contain comparable gas and stellar masses , each corresponding to about 10@xmath98 m@xmath78 in the case of leo a and about 10@xmath99 m@xmath78 in sagdig . young and lo also comment on a small region of about 2.5 arcmin diameter within leo a which they suggest might be the expansion or contraction signature of an shell , comparable to the five cases of line splitting within chvcs which we describe above . the dwarf elliptical companions of m31 , ngc 185 and ngc 205 ( young and lo @xcite ) are rather different in their gas properties . while both the cnm and wnm components of are seen in these two cases , all of the neutral gas is concentrated to the central five arcmin ( 1 kpc ) of each galaxy and the global kinematics are less systematic . the strong interaction of these objects with m31 ( ngc 205 in particular is on an orbit that is likely to have recently passed through the disk of m31 ) and the high mass density of stars ( stellar masses about 100 times greater than the 10@xmath100m@xmath78 in gas ) makes a direct comparison with the chvcs less obvious . an intermediate and less perturbed example is given by the more isolated local group dwarf spheroidal lgs 3 ( young and lo @xcite ) . in this case , the stellar mass is likely only a few times that of the 10@xmath101 m@xmath78 in gas . here the is also confined to only the inner 5 arcmin ( corresponding to 1 kpc ) , shows little systematic kinematics , and the cool condensed phase is almost entirely absent . although peak column densities of somewhat more than 10@xmath28 @xmath17 are observed over a region of a few arcmin extent , there is no sign of ongoing star formation . extending the comparison to even higher mass systems provides some further insight into the conditions necessary for allowing the condensation of cnm clumps in a galactic environment . several low mass spirals were observed in the high resolution sample of braun ( @xcite , @xcite ) , including the nearby sbm galaxy ngc 2366 and two of the nearest low surface brightness spirals , ngc 247 ( of type sc ) and ngc 4236 ( sbd ) . a filamentary network of cnm gas is seen in all these systems , which accounts for about 80% of the line flux from only about 15% of the surface area within @xmath102 ( the radius where the average face - on surface brightness of light in the b band has declined to 25 mag arsec@xmath103 ) . just beyond @xmath102 the flux fraction of cnm gas disappears entirely , even though the wnm continues out to a radius of almost twice @xmath102 . the result is that the _ total _ flux fraction due to cnm in these galaxies is about 70% , substantially more than the 40% cnm fraction in the chvcs . comparison with the thermodynamics of wolfire et al . ( @xcite ) suggest that the edge of the cnm disk should occur where the thermal pressure has declined to a value of @xmath104 , below which cnm condensation is no longer possible . the numerical value of @xmath104 depends on the dust and gas phase metal abundance , the radiation field intensity and the shielding column of neutral gas . values for @xmath104/k of about 200 @xmath58 k seem indicated under conditions of moderate metallicity and radiation field - intensity expected for the low mass spirals above . comparison with fig . [ fig : phasep ] suggests that slightly lower values for @xmath104/k of about 100 @xmath58 k should be expected under the low metallicity , low field intensity conditions that might apply to the chvcs . what pressures might be expected in the mid - plane of a disk like self gravitating distribution of @xmath1 ? the concept of hydrostatic pressure equilibrium demands that the weight of the overlying medium is balanced by the sum of the contributions to the gas pressure . this can be expressed as @xmath105 where @xmath106 is the gas volume density at height @xmath107 from the mid - plane . in this form , @xmath108 refers to the mass surface density between @xmath109 to @xmath110 , not between @xmath111 to @xmath110 . the total pressure is composed of magnetic , cosmic ray , and turbulent components in addition to the thermal component which is of relevance to the calculation of the hi kinetic temperature . in a high mass galactic disk , the thermal pressure might amount to only about 1/4 of the total pressure , while in a low mass chvc with few if any sources of internal turbulence and energetic particles , the thermal pressure might be the dominant component . if for simplicity we assume a constant gas density , @xmath112 , for @xmath113 for some full disk thickness , @xmath114 , then @xmath115 for @xmath107 less than @xmath116 . for the moment we will neglect contributions to the mass surface density by stars or dark matter , which if present would lead to proportionally higher pressures . equation [ eqn : hydro ] then becomes simply @xmath117 inserting numerical values into eqn . [ eqn : press ] yields @xmath118 thus , whenever the face on column density exceeds about 10@xmath28 @xmath17 in a disk like system we can expect to see the cool condensations of the cnm . given the quadratic dependence on column density in eqns . [ eqn : hydro][eqn : n20 ] , we might expect the transition from purely wnm to a cnm / wnm mix to be quite abrupt . this expectation is borne out quite dramatically at the edge of the cnm disks in spiral galaxies which indeed occurs very near a face on column density of 12@xmath810@xmath28 @xmath17 . if a significant contribution to the mass surface density within one gas scale height were due to stars or dark matter then the minimum column density needed for cnm condensation would decrease . for the chvcs , our current data suggest a transition column density nearer 10@xmath119 @xmath17 , rather than the 10@xmath28 @xmath17 seen in nearby spirals . this may be an indication that the mass surface density of dark matter is already a significant contribution . we can assess this possibility by considering the mass volume density of the implied dark halos . the density distribution of the @xmath89 m@xmath78 nfw halo which gave a good fit to the derived rotation velocity of the core in chvc204 + 30 + 075b ( fig . [ fig : kinenfw ] ) is shown in fig . [ fig : rhoplot ] . at a thermal pressure of @xmath120 @xmath58 k , the 8000 k wnm halos of the cnm cores will have a volume density of about 0.01 @xmath58 . comparison with fig . [ fig : rhoplot ] shows that the dark halo will provide a dominant contribution to the mass density out to several kpc . within the central region , of perhaps 2 kpc ( or 10 arcmin at @xmath2 mpc ) in diameter , the mass density is enhanced by more than a factor of 10 over that of the gas alone . in consequence , we can expect to achieve values of @xmath104/k of about 100 @xmath58 k in the chvcs whenever the column densities exceed a few times 10@xmath119 @xmath17 . we present some of the first high resolution imaging of a sub sample of the compact high velocity clouds ( chvcs ) . previously we had demonstrated that the spatial and kinematic properties of the chvcs were strongly suggestive of a local group origin ( braun & burton @xcite ) , although the typical distance could not be well constrained from the low resolution ( 05 ) data then in hand . our imaging has revealed that these objects have a characteristic morphology , consisting of one or more quiescent , low dispersion compact cores embedded in a diffuse , warmer halo . this is consistent with what was seen previously by wakker & schwarz ( @xcite ) for the two chvcs which they imaged . the compact cores can be unambiguously identified with the cool neutral medium ( cnm ) of condensed atomic hydrogen , since their linewidths are in all cases significantly narrower than even the thermal linewidth of the warm neutral medium ( wnm ) . in one object we detect some of the narrowest emission lines yet seen , with an intrinsic fwhm of 2 km s@xmath0 . the halo linewidths , while difficult to measure directly , appear broad enough to be consistent with a wnm origin . such a nested geometry of cnm and wnm is in agreement with the expectation from thermodynamics since a significant neutral shielding column is required for the stable existence of cnm when exposed to an ionizing radiation field ( e.g. wolfire et al . @xcite ) . the flux fraction in compact cores varies between about 1% and 50% in our chvc sub sample , but is typically 40% , as also seen in the two cases studied by wakker and schwarz ( @xcite ) . the surface covering factor of the moderate column density core emission relative to the diffuse halo is usually a factor two lower , namely about 15% . high resolution imaging of the column density distribution of chvc125 + 41@xmath121 has permitted an improved determination of the metal abundance in this object when combined with a previous measurement of unsaturated absorption ( bowen & blades @xcite , bowen et al . @xcite ) of between 0.04 and 0.07 solar . the current program has also allowed meaningful distance estimates for chvcs to be made for the first time . distances in the range 0.5 to 1 mpc follow from two independent approaches . the first method utilizes the fact that we have resolved cool , opaque knots of within one of the six imaged systems ( chvc125 + 41@xmath121 ) . by equating the observed angular size of these knots with the line of sight depth , we derive a distance from a comparison of the column and volume densities . estimates of the column density follow directly from our data , while specific thermodynamic calculations for volume densities under the relevant physical conditions were made available to us by wolfire , sternberg , hollenbach , and mckee ( private communication ) . the second approach allows only a combination of source distance and dark matter fraction to be derived , since it relies on a comparison of dynamical with visible mass under the assumption that the system is gravitationally bound . for this approach , the chvc115 + 13@xmath122 system proved particularly illuminating , since a collection of some 10 compact cores is distributed over a 30 arcmin diameter region within a common diffuse envelope , while having discrete line of sight velocities distributed over a 70 km s@xmath0 range . at an assumed distance , @xmath2 mpc , the dark to visible mass ratio is @xmath123 , with @xmath83 scaling as @xmath3 . a common feature observed in the chvcs is that the compact cores have an elliptical shape with a clear velocity gradient aligned with the major axis . these kinematic data can be well modeled by rotation in flattened disk systems . rotation curve fits to the best resolved cores within chvc069 + 04@xmath124 and chvc204 + 30 + 075 , yield rather high dark - to - visible mass ratios of @xmath83 = 1040 at d = 0.7 mpc , with @xmath83 scaling , as above , with 1/d . moreover , the shape of the rotation curves in two of these three cases is very well described by that due to a cold dark matter halo as parameterized by navarro et al . ( @xcite ) . well resolved imaging has also revealed another intriguing aspect of the chvcs . five instances of localized line splitting are detected . these are areas of 60 to 90 arcsec extent in regions of moderately high column density in which the emission profile becomes doubly peaked in velocity , without displaying a shift in the velocity centroid . this appears to be a signature of localized outflow or inflow . at a distance of 0.7 mpc they would correspond to feaures of a few hundred pc in extent , comparable in size to the stellar wind driven bubbles surrounding young stellar associations . comparison of the chvc properties with those of nearby low mass galaxies shows striking similarities , particularly with the local group dwarf irregulars . the required thermal pressure for condensation of the cool cnm cores , @xmath104/k @xmath125 100 @xmath58 k , can be achieved in hydrostatic equilibrium of a self gravitating disk like distribution at the observed shielding column of wnm of about 10@xmath119 @xmath17 if the significant mass density of the dark matter halos is taken into account within the inner 2 kpc . our study of the compact high velocity clouds has helped to clarify the nature of these objects . all of the evidence points to these being strongly dark matter dominated , self gravitating objects of modest mass in the local group potential at distances between perhaps 0.3 and 1 mpc with physical dimensions in of about 10 kpc . typical masses of a few times 10@xmath99 m@xmath78 are indicated , with associated dark masses of some 10@xmath126 m@xmath78 . the total number of such objects associated with the local group , while still uncertain , may be about 200 . it seems likely that these are the objects which recent numerical simulations suggest should still be associated with massive galaxies in a poor environment at the current epoch ( klypin et al . @xcite , moore et al . @xcite ) . the star formation history within these objects is still very much a mystery . high stellar densities , like those seen in dwarf spheroidal systems in the local group , are clearly not associated with the centroids of the gas distribution . a large spatial separation of the gas and associated stellar distribution , due for example to tidal effects , is not expected to be a common circumstance since the population is dynamically so cold . the peak column densities currently observed within the chvc cores only rarely exceed 10@xmath28 @xmath17 . it is plausible that these values are insufficient to allow widespread star formation at this time . on the other hand , finite yet low metallicities have been measured in at least one source , indicating that some enrichment has clearly taken place in the past . it is quite possible that associated old stellar populations with a low surface density will be found in deep searches which can distinguish such stars from the galactic foreground . we are grateful to m.g . wolfire , a. sternberg , d. hollenbach , and c.f . mckee for providing the equilibrium temperature curves for in an intergalactic radiation field shown in fig . [ fig : phasep ] as well as stimulating correspondence on the ism in general , and to u. schwarz and h. van woerden for showing us , in advance of publication , the jodrell bank data of de vries et al . ( @xcite ) on the source chvc125 + 41@xmath4207 . the westerbork synthesis radio telescope is operated by the netherlands foundation for research in astronomy , under contract with the netherlands organization for scientific research . bajaja e. , morras r. , pppel w.g.l . , 1987 , pub . 69 , 237 bland - hawthorn j. , maloney p.r . , 1999 , in stromlo workshop on high - velocity clouds , eds . gibson , b.k . & putman , m.e . , ( asp conf 166 ) , 212 blitz l. , spergel d.n . , teuben p.j . , hartmann d. , burton w.b . , 1999 , apj 514 , 818 bowen d.v . , blades j.c . , 1993 , apj 403 , l55 bowen d.v . , blades j.c . , pettini m. , 1995 , apj 448 , 662 braun r. , 1995 , a&as 114 , 409 braun r. , 1997 , apj 484 , 637 braun r. , 1998 , in interstellar turbulence , eds . franco j. , carraminana a. ( cambridge : cambridge university press ) , 52 braun r. , burton w.b . , 1999 , a&a 341 , 437 braun r. , walterbos r.a.m . , 1992 , apj 386 , 120 bregman j.n . , 1980 , apj 236 , 577 burton w.b . , 1997 , in the physics of the galactic halo , eds . lesch h. , dettmann r .- j . , mebold u. , schlickeiser r. ( berlin : akademie verlag ) , 15 de vries g. , schwarz u.j . , smoker j. , davies r.d . , van woerden h. , 1999 , private communication eichler d. , 1976 , apj 208 , 694 einasto j. , haud u. , jeveer m. , kaasik a. , 1976 , mnras 177 , 357 giovanelli r. , 1981 , aj 86 , 1468 hartmann d. , burton w.b . , 1997 , `` atlas of galactic neutral hydrogen '' ( cambridge : cambridge university press ) jenkins e.b . , savage b.d . , spitzer l. , 1986 , apj 301 , 35 klypin a. , kravtsov a.v . , valenzuela o. , prada f. , 1999 , apj 522 , 82 mashchenko s.y . , thilker d.a . , braun r. , 1999 , a&a 343 , 352 moore b. , ghigna s. , governato g. , lake g. , quinn t. , stadel j. , tozzi p. , 1999 , apj 524 , 19 navarro j.f . , frenk c.s . , white s.d.m . , 1997 , apj 490 , 493 oort j.h . , 1966 , bull . netherlands 18 , 421 oort j.h . , 1970 , a&a 7 , 381 oort j.h . , 1981 , a&a 94 , 359 putman , m.e . , gibson , b.k . , 1999 , pasa 16 , 70 shapiro p.r . , field g.b . , 1976 , apj 205 , 762 verschuur g.l . , 1975 , ara&a 13 , 257 wakker b.p . , schwarz u. , 1991 , a&a 250 , 484 wakker b.p . , van woerden h. , 1991 , a&a 250 , 509 wakker b.p . , van woerden h. , 1997 , ara&a 35 , 217 wolfire m.g . , hollenbach d. , mckee c.f . , tielens a.g.g.m . , bakes e.l.o . , 1995 , apj 443 , 152 wolfire m.g . , mckee c.f . , hollenbach d. , tielens a.g.g.m . , 1995 , apj 453 , 673 young l.m . , lo k.y . , 1996 , apj 462 , 203 young l.m , lo k.y . , 1997a , apj 476 , 127 young l.m . , lo k.y . , 1997b , apj 490 , 710

we consider the dirichlet boundary value problem for divergence form elliptic operators with bounded measurable coefficients . we prove that for uniform domains with ahlfors regular boundary , the bmo solvability of such problems is equivalent to a quantitative absolute continuity of the elliptic measure with respect to the surface measure , i.e. . this generalizes a previous result on lipschitz domains by dindos , kenig and pipher ( see ) .

1602.00717

consider an open bounded domain @xmath1 ( @xmath2 ) . let @xmath3 be an operator defined as @xmath4 , where @xmath5 be a real @xmath6 matrix on @xmath7 that is bounded measurable and uniformly elliptic : that is , there exists a constant @xmath8 such that @xmath9 for all @xmath10 and @xmath11 . the matrix @xmath12 is not assumed to be symmetric . we say @xmath7 is regular for @xmath3 , if for any continuous function @xmath13 , the corresponding solution @xmath14 to the dirichlet problem @xmath15 is continuous in @xmath16 . in particular @xmath17 on the boundary in the classical sense . it was proved in @xcite that @xmath7 is regular for @xmath3 if and only if it is regular for the laplacian . assume @xmath7 is regular , then by the riesz representation theorem , for any @xmath11 there is a probability measure @xmath18 on @xmath19 such that @xmath20 for an arbitrary boundary value @xmath13 and its corresponding solution @xmath14 . the measure @xmath21 is called the elliptic measure ( harmonic measure if @xmath22 ) of @xmath7 at @xmath23 . from , we see that the information about the boundary behavior of solutions to dirichlet problem is encoded in the elliptic measure @xmath21 . if @xmath7 is connected , the elliptic measures @xmath21 and @xmath24 at different points @xmath25 are mutually absolute continuous , moreover they satisfy @xmath26 . thus for the problem we are concerned with , we just need to study the elliptic measure at a fixed point @xmath27 . denote @xmath28 . for lipschitz domains , by the work of @xcite , @xcite and @xcite , we know the elliptic measure @xmath29 and the solutions @xmath14 to enjoy numerous properties , most notably - ( see section [ sect : prelim ] ) . these pointwise estimates are important on their own , but they also serve as a toolkit to build global pde estimates for the solutions to the dirichlet problem . on the other hand , when the surface measure @xmath30 is well defined on @xmath19 , a natural question is : what is the relationship between the measures @xmath29 and @xmath30 ? in fact , the answer to this measure theoretic question is closely related to the boundary behavior of solutions to this pde problem . let @xmath7 be a regular domain such that the surface measure @xmath31 on the boundary is locally finite . here @xmath32 denotes the @xmath33-dimensional hausdorff measure restricted to @xmath19 . for @xmath34 , we say the problem is solvable in @xmath35 if there exists a universal constant @xmath36 such that for any boundary value @xmath13 and its corresponding solution @xmath14 , @xmath37 where @xmath38 is the maximal function of @xmath14 in the non - tangential cone @xmath39 ( the definition is specified in section [ sect : prelim ] ) . from we know @xmath40 is comparable to the hardy - littlewood maximal function @xmath41 with respect to @xmath29 . provided that @xmath30 is doubling , the theory of weights tells us that @xmath42 for the laplacian on lipschitz domains , dahlberg @xcite proved the harmonic measure @xmath43 ; therefore is solvable in @xmath35 for @xmath44 . when @xmath45 , follows trivially from the maximal principle . however , if @xmath22 and @xmath46 is the upper half plane , the problem is also solvable in the bmo space , that is , if @xmath47 , its harmonic extension @xmath14 has the property that @xmath48 is a carleson measure on @xmath7 ( see @xcite , and also section 4.4 theorem 3 of @xcite ) . in addition , the carleson measure norm of @xmath49 is equivalent to the bmo norm of @xmath50 . ( for the definition of carleson measure , see definition [ def : carl ] . ) this bmo solvability also holds for lipschitz domains , when @xmath49 is replaced by @xmath51 and @xmath52 ( see @xcite ) . recall @xmath53 is a quantitative version of absolute continuity with respect to @xmath30 , and that @xmath54 . ( see definition [ def : ainfty ] and its following paragraph . ) by the work of dahlberg , we know the elliptic measure @xmath55 if @xmath3 is a small perturbationof the laplacian ( see @xcite ) . later in @xcite , the smallness assumption was removed using harmonic analysis methods ( see also @xcite ) . more precisely , in @xcite the authors showed that if @xmath56 and @xmath57 is a perturbation of @xmath58 , then @xmath59 . assume the operator @xmath3 satisfies @xmath0 , then @xmath60 note that there is some ambiguity with @xmath61 : the fact that is not @xmath62 solvable does not necessarily imply @xmath63 . a natural question arises : is there a solvability criterion that directly characterizes @xmath0 ? in 2009 , dindos , kenig and pipher showed that for lipschitz domains , the elliptic measure @xmath64 if and only if the problem is bmo - solvable , i.e. for any continuous function @xmath13 , the carleson measure of @xmath65 is controlled by the bmo norm of @xmath50 ( see @xcite ) . the dirichlet problem has been studied on domains less regular than lipschitz ones , in particular @xmath66 domains , non - tangential accessible ( nta ) domains and uniform domains . these non - smooth domains arise naturally in free boundary problems and geometric analysis . roughly speaking , uniform domains are domains of which each ball centered at the boundary contains an interior ball comparable in size ( interior corkscrew condition , see definition [ def : icc ] ) and that satisfy a notion of quantified connectivity ( harnack chain condition , see definition [ def : hcc ] ) . nta domains are uniform domains whose exterior also satisfies the corkscrew condition . it is easy to see @xmath67 the notion of nta domain was introduced by jerison and kenig in the pioneer work @xcite , where they proved the set of properties - hold for the laplacian on nta domains . in @xcite the authors addressed the @xmath35 solvability of on chord - arc domains , i.e. nta domains with ahlfors regular boundary ( see definition [ def : adr ] ) . on the other hand , by the independent work of @xcite and @xcite , certain properties , notably the boundary comparison principle , are shown to hold for the laplacian on uniform domains ( the domains studied in the latter paper also require ahlfors regular boundary ) . in section [ sect : br ] we show that all the properties - hold for uniform domains with ahlfors regular boundary . moreover , we prove an equivalent characterization of the absolute continuity of elliptic measures by pde solvability condition : for uniform domains with ahlfors regular boundary , the elliptic measure @xmath0 if and only if the dirichlet problem is bmo solvable . in @xcite , the authors studied the geometric implication of the harmonic measure @xmath68 . for a uniform domain with ahlfors regular boundary , they showed @xmath69 which characterizes the absolute continuity of the harmonic measure from a geometric point of view . we should point out that some of the implications in had been proved earlier in @xcite , @xcite and @xcite . the complement of the four - corner cantor set inside a large ball @xmath70 is an example of a uniform domain with ahlfors regular boundary , whose boundary is purely unrectifiable . thus its harmonic measure @xmath71 by . our study of general elliptic operators raises the following question : if @xmath72 , is there a uniformly elliptic operator @xmath73 such that @xmath0 ? more generally , for a uniform domain with purely unrectifiable boundary , is there an @xmath3 such that @xmath0 ? if so , what geometric information does this carry , and can we characterize the corresponding matrix @xmath12 ? in this paper we assume quantified connectivity of the domain , in order to get a quantified characterization of the absolute continuity of @xmath29 with respect to @xmath30 , i.e. @xmath0 . recent work by several authors has addressed the question of the relation between @xmath29 and @xmath30 with no connectivity assumption on the domain . we refer interested readers to @xcite , @xcite and @xcite . the plan for this paper is as follows . we state the definitions and main theorem in section [ sect : prelim ] . in section [ sect : br ] we establish some preliminary lemmas for the elliptic measure and solutions to problem on uniform domains with ahlfors regular boundary . in section [ sect : showbmos ] we assume the elliptic measure is an @xmath74 weight with respect to the surface measure , and show the bmo solvability , i.e. the carleson measure of @xmath75 is bounded by the bmo norm of the boundary data . section [ sect : showainfty ] is devoted to the proof of @xmath0 under the assumption of the bmo solvability for . in section [ sect : converse ] we show that if @xmath0 , the converse to the carleson measure estimate holds . throughout this paper , we always assume @xmath1 is open and bounded , and @xmath2 . [ def : icc ] the domain @xmath7 is said to satisfy the * interior corkscrew condition * ( resp . exterior corkscrew condition ) if there exists @xmath76 and @xmath77 such that for any @xmath78 and any @xmath79 , there exists a corkscrew point ( or non - tangential point ) @xmath80 ( resp . @xmath81 ) such that @xmath82 define the non - tangential cone @xmath83 at @xmath78 with aperture @xmath84 as follows @xmath85 and define the truncated cone @xmath86 . we will omit the super - index @xmath87 when there is no confusion . the interior corkscrew condition in particular implies @xmath88 is nonempty as long as the aperture @xmath89 . the non - tangential maximal function is defined as @xmath90 and the square function is defined as @xmath91 we also consider truncated square function @xmath92 , and the only difference is that the non - tangential cone @xmath39 is replaced by the truncated cone @xmath93 . [ def : hcc ] the domain @xmath7 is said to satisfy the * harnack chain condition * if there are universal constants @xmath94 and @xmath95 , such that for every pair of points @xmath12 and @xmath96 in @xmath7 satisfying @xmath97 there is a chain of open harnack balls @xmath98 in @xmath7 that connects @xmath12 to @xmath96 . namely , @xmath99 , @xmath100 , @xmath101 and @xmath102 for all @xmath103 . moreover , the number of balls @xmath104 . 1 . if two points @xmath105 do not satisfy , we can simply take the balls @xmath106 and @xmath107 to connect them . 2 . we often want the balls to satisfy @xmath108 to be able to enlarge them . this is possible . in fact , if @xmath7 satisfy the harnack chain condition , then for any @xmath109 , there is a constant @xmath110 such that @xmath7 satisfies the above harnack chain condition with the comparable size condition replaced by @xmath111 the number of balls @xmath112 may increase , but is still of the order @xmath113 . moreover , the ratio between @xmath110 and @xmath114 is fixed : @xmath115 balls satisfying the condition are called * harnack balls * with constants @xmath116 . if @xmath7 satisfies ( 1 ) the interior corkscrew condition and ( 2 ) the harnack chain condition , then we say @xmath7 is a * uniform domain*. if in addition , @xmath7 satisfies the exterior corkscrew condition , we say it is an * nta ( non - tangential accessible ) domain*. 1 . there is an alternative and equivalent definition of the uniform domain . see @xcite definition 2.14 and the paragraph following the definition for details . the uniform domain is also known as one - sided nta domain in the literature : by comparison , the nta domain introduced by jerison and kenig @xcite is a uniform domain whose complement also satisfies the corkscrew condition . for any @xmath78 and @xmath117 , let @xmath118 denote the surface ball @xmath119 , and @xmath120 denote the carleson region above @xmath121 . we always assume @xmath122 , so that @xmath123 . [ def : adr ] we say that the boundary of @xmath7 is * ahlfors regular * if there exist constants @xmath124 , such that for any @xmath78 and any radius @xmath117 , @xmath125 where @xmath126 is the surface measure . [ def : carl ] a measure @xmath49 in @xmath7 is a * carleson measure * if there exists a constant @xmath36 such that for all surface ball @xmath127 , @xmath128 we refer to the smallest such constant as the carleson measure norm of @xmath49 . [ def : ainfty ] let @xmath49 and @xmath129 be two finite borel measures on @xmath19 . we say @xmath130 if for any @xmath131 , there exists @xmath132 such that for any surface ball @xmath121 and any @xmath133 , @xmath134 the @xmath74 property is in fact symmetric : @xmath130 if and only if @xmath135 . if this holds , there is a density function @xmath136 ; moreover there exists @xmath137 such that @xmath138 and satisfies a reverse hlder inequality @xmath139 for all surface balls @xmath121 . in other words , @xmath140 . the purpose of this paper is to prove the following theorem : [ thm : main ] let @xmath1 be a uniform domain with ahlfors regular boundary and the operator @xmath73 , where @xmath141 is a real @xmath6 matrix that is bounded and uniformly elliptic on @xmath7 . then the elliptic measure @xmath0 if and only if the problem is bmo solvable , namely the following statement holds : there exists a constant @xmath142 such that for any continuous function @xmath13 , if @xmath143 in @xmath7 with @xmath17 on @xmath19 , then @xmath65 is a carleson measure , and @xmath144 moreover , the reverse inequality to also holds , that is , there exists a constant @xmath95 such that @xmath145 before we start the proof , we make the following observation : the carleson measure norm of @xmath65 is in some sense equivalent to the integral of the truncated square function . suppose @xmath146 is a surface ball . for any @xmath147 , we define @xmath148 . let @xmath149 be the point such that @xmath150 . then @xmath151 since @xmath19 is ahlfors regular , this implies @xmath152 . thus @xmath153 on one hand , @xmath154 on the other hand , @xmath155 where @xmath156 . therefore @xmath157 for any @xmath158 . the boundary hlder regularity of the solution to the dirichlet problem has been proved for the nta domain in @xcite . by tools from potential theory , we show the boundary hlder regularity holds for uniform domains with ahlfors regular boundary . the author becomes aware of theorem [ arcdc ] and [ thm : osc ] through discussions with prof . tatiana toro , for which the author is grateful . a domain @xmath7 is * wiener regular * , if for any lipschitz function @xmath159 , there exists a solution @xmath160 to @xmath143 with dirichlet boundary data @xmath50 . a classic result of littman , stampacchia and weinberger @xcite shows that @xmath7 is regular for the elliptic operator @xmath73 if and only if it is regular for the laplacian . therefore wiener regular is indeed a property of the domain , so we do not need to specify which operator we refer to in this definition . @xmath7 is wiener regular if and only if for any @xmath78 , @xmath161 for any compact set @xmath162 , the capacity is defined as follows : @xmath163 the following condition has been explored extensively by aikawa ( the condition has been mentioned without name in the following work @xcite ) . see @xcite @xcite @xcite for example . a domain @xmath7 is said to satisfy the * capacity density condition ( cdc ) * if there exist constants @xmath164 such that for any @xmath165 and any @xmath79 , we have @xmath166 clearly if the domain @xmath7 satisfies the cdc , it satisfies , hence @xmath7 is wiener regular . what is relevant in our case is the following : [ arcdc ] if the domain @xmath7 has ahlfors regular boundary , it satisfies the cdc . in particular , @xmath7 is wiener regular . for any set @xmath167 contained in a ball of radius @xmath168 , its hausdorff content and its capacity satisfy @xmath169 where the hausdorff content @xmath170 is defined as @xmath171 for the proof see @xcite page 193 - 194 . we claim @xmath172 for ahlfors regular boundary @xmath19 . let @xmath173 . it is clear from the definition that @xmath174 . let @xmath175 be balls in @xmath176 such that @xmath177 . without loss of generality we may assume @xmath178 . since @xmath19 is ahlfors regular , for each @xmath179 @xmath180 thus by the countable sub - additivity of the hausdorff measure , @xmath181 this is true for any arbitrary covering @xmath182 of @xmath167 , hence is true for the infimum . by the definition @xmath183 therefore @xmath184 by and , @xmath185 therefore @xmath186 the domain @xmath7 satisfies cdc . [ thm : osc ] let @xmath187 be an open , bounded domain . there exists a constant @xmath142 such that the following holds . suppose @xmath188 are functions satisfying @xmath189 @xmath190 then for any @xmath191 , any @xmath192 such that @xmath193 and any @xmath194 , we have @xmath195 where @xmath196 is the oscillation of a function @xmath197 on a set @xmath167 : @xmath198 this theorem , together with the cdc condition , implies the boundary hlder regularity of solutions with vanishing boundary data . [ prop : bhr ] let @xmath7 be a domain satisfying cdc with constants @xmath199 . there exist constants @xmath200 and @xmath142 such that the following holds . for any @xmath201 and any @xmath192 with @xmath202 , let @xmath203 be the solution of @xmath204 then for any @xmath205 , @xmath206 let @xmath207 be an open bounded domain , then we can apply theorem [ thm : osc ] to the function @xmath14 on @xmath187 . for any @xmath205 , let @xmath168 be the constant such that @xmath208 . then @xmath209 since @xmath210 , by the cdc condition @xmath211.\ ] ] hence @xmath212 where @xmath213 . therefore @xmath214 after we established the boundary regularity of non - negative solutions ( proposition [ prop : bhr ] ) , one can prove many other properties of the solutions and the elliptic measure . we quote them as lemmas here . if the matrix @xmath12 is symmetric , the proof is exactly the same as in the nta case ( see @xcite or @xcite ) ; if @xmath12 is non - symmetric , see @xcite theorem 1.11 , ( 1.12 ) , ( 1.13 ) and ( 1.14 ) . [ lm : lbomega ] let @xmath7 be a uniform domain with ahlfors regular boundary.then there exists @xmath215 such that for any @xmath201 and any @xmath216 , we have @xmath217 here @xmath112 and @xmath218 are the constants mentioned in the definitions of the interior corkscrew condition and cdc , respectively . [ lm : bhp ] let @xmath14 be a non - negative solution in @xmath219 with vanishing boundary data on @xmath220 . then @xmath221 suppose @xmath222 , then @xmath223 there is a constant @xmath94 such that , for any surface ball @xmath146 , let @xmath224 be its doubling surface ball , we have @xmath225 in @xcite , the authors proved the boundary comparison principle on nta domains using the maximal principle , the doubling of harmonic measure and p. jones geometric localization theorem ( @xcite ) . the geometric localization theorem says the following : if @xmath7 is an nta domain , there are constants @xmath94 and @xmath77 such that for any @xmath158 and @xmath226 , there exists a domain @xmath227 such that @xmath228 and moreover @xmath227 is an nta domain , whose nta constants only depend on the nta constants of @xmath7 . to prove the boundary comparison principle in our setting , and later to prove theorem [ thm : bmogreen ] , we need a similar geometric localization theorem for uniform domains with ahlfors regular boundary . jones construction works for uniform domains , but does not have good property on the boundary . to take full advantage of ahlfors regularity in our setting , we use the carleson box construction in @xcite . [ thm : glt ] let @xmath7 be a uniform domain with ahlfors regular boundary . there are constants @xmath94 and @xmath77 such that for any @xmath201 and @xmath226 , there exists a domain @xmath227 such that @xmath229 and moreover , @xmath227 is a uniform domain with ahlfors regular boundary , where the constants only depend on the corresponding constants for @xmath7 . since @xmath230 is ahlfors regular , it has a nested grid @xmath231 such that @xmath232 for any @xmath233 , and each @xmath234 has diameter less than @xmath235 and contains a ball of radius comparable to @xmath235 . ( for their precise properties , see lemma 1.15 in @xcite ) . for any surface ball @xmath146 , there exists @xmath236 such that @xmath237 and @xmath238 . let @xmath239 be the carleson box defined as ( 3.52 ) ( see also ( 3.47 ) , ( 3,43 ) ) in @xcite . the authors of @xcite proved in lemma 3.61 that @xmath239 is a 1-sided nta domain ( i.e. uniform domain ) with ahlfors regular boundary ; and in lemma 3.55 they proved there is some @xmath240 such that @xmath241 from the definition of @xmath239 and the fact that @xmath238 , it is clear that we also have @xmath242 let @xmath14 and @xmath243 be non - negative solutions in @xmath244 with vanishing boundary data on @xmath220 . then @xmath245 once we have the geometric localization theorem , the proof proceeds the same as lemma ( 4.10 ) in @xcite . assume @xmath55 . for any continuous function @xmath13 , let @xmath14 be the solution to the elliptic problem @xmath246 with boundary data @xmath50 , we want to show that @xmath247 is a carleson measure , and in particular , @xmath248 let @xmath146 be any surface ball on the boundary . denote the constant @xmath249 , and let @xmath250 be its concentric surface ball . let @xmath251 @xmath252 @xmath253 and let @xmath254 be the solutions to @xmath143 with boundary data @xmath255 . clearly @xmath256 is a constant , so its carleson measure is trivial . ( note that if @xmath168 is big enough such that @xmath257 , then @xmath258 , and @xmath259 is trivial . in this case we only need to consider @xmath260 and @xmath261 . ) we have shown in section [ sect : prelim ] ( see and ) that @xmath262 since @xmath263 , it follows that @xmath264 by hlder inequality , for @xmath265 we have @xmath266 under the assumption @xmath55 , the following theorems are at our disposal : [ thm : nu ] assume @xmath267 for some @xmath268 , then the elliptic problem @xmath143 is @xmath269solvable : if @xmath14 is a solution with boundary data @xmath270 , then @xmath271 where @xmath272 is the non - tangential maximal function , and @xmath273 . [ thm : nusu ] assume @xmath274 , then if @xmath275 with boundary data @xmath50 , and @xmath276 , we have @xmath277 for any @xmath278 . note that the continuous function @xmath260 satisfies @xmath279 . if @xmath280 , @xmath260 must assume both positive and negative values on the boundary . thus there must exist some @xmath27 such that @xmath281 . therefore we can apply theorem [ thm : nu ] and theorem [ thm : nusu ] to @xmath261 , and get @xmath282 combining , and , we obtain @xmath283 to show similar estimate for @xmath284 , let @xmath285 be a whitney decomposition of @xmath286 , then for all @xmath287 , @xmath288 . on each whitney cube @xmath289 , we have the following cacciopoli type estimate , @xmath290 summing up , we get @xmath291 recall @xmath292 denotes @xmath293 , and @xmath294 denotes @xmath295 . let @xmath296 be the solutions to @xmath143 with non - negative boundary data @xmath297 , then @xmath298 and @xmath299 . let @xmath300 we have the following lemma : [ lm : vbmobd ] the function @xmath243 defined in satisfies * @xmath301 for all @xmath302 . * @xmath303 for all @xmath304 . here @xmath200 is the degree of boundary hlder regularity for non - negative solutions and it only depends on @xmath305 and the ellipticity of @xmath12 ( see ) . by the definition , we know the function @xmath243 vanishes on @xmath306 . note that @xmath307 by the choice of @xmath306 , @xmath243 is a non - negative solution in @xmath308 and vanishes on @xmath309 . let @xmath12 be a corkscrew point in @xmath310 , by lemma [ lm : bhp ] we have @xmath311 let @xmath312 be the radon - nikodym derivative of @xmath313 with respect to @xmath30 . because @xmath68 , there exists some @xmath137 such that @xmath314 for any surface ball @xmath315 . let @xmath316 be the radon - nikodym derivative of @xmath21 with respect to @xmath313 , i.e. @xmath317 then @xmath318 let @xmath315 be any surface ball contained in @xmath319 , let @xmath320 be a corkscrew point in @xmath321 . then by corollary 1.3.8 @xcite ( it follows easily from the boundary caparison principle ) @xmath322 on the other hand , by the boundary regularity of @xmath323 in @xmath324 , we have @xmath325 combining and we have @xmath326 for any @xmath315 contained in @xmath327 . therefore @xmath328 combining , and , @xmath329 therefore @xmath330 for any @xmath304 , let @xmath331 be a boundary point such that @xmath150 . note that @xmath332 so @xmath333 . we consider the dirichlet problem in @xmath334 . note that @xmath335 hence the triangle inequality implies @xmath336 note that @xmath337 , @xmath243 is a non - negative solution in @xmath338 which vanishes on @xmath339 . by the boundary hlder regularity ( proposition [ prop : bhr ] ) , set inclusion and the first part of this lemma , we conclude @xmath340 using lemma [ lm : vbmobd ] and , we get @xmath341 note that @xmath342 , we need the following lemma . [ lm : delta ] for any @xmath343 , we have @xmath344 if @xmath345 , we can prove directly : @xmath346 if not , the proof is more delicate . for @xmath347 , let @xmath348 and @xmath349 then @xmath350 for any @xmath351 , there exists @xmath149 such that @xmath150 . then @xmath352 in other words @xmath353 . consider a covering of @xmath354 by balls of radius @xmath355 @xmath356 by vitali covering lemma it has a countable subcover @xmath357 where @xmath358 and the balls @xmath359 are pairwise disjoint . the fact that @xmath360 implies @xmath361 for each @xmath362 . and the pairwise disjointness of @xmath363 s implies there are only finitely many of them . in fact , @xmath364 note that @xmath365 independent of @xmath362 , and @xmath366 let @xmath367 be the number of @xmath363 s . combining , and we get @xmath368 thus @xmath369 . by the definition of @xmath370 , @xmath371 thus @xmath372 combined with we get @xmath373 the last inequality is because @xmath374 . combining and , we get @xmath375 and together give the carleson measure estimate . let @xmath121 be an arbitrary surface ball . let @xmath50 be a continuous non - negative function supported in @xmath121 and @xmath14 is the corresponding solution with boundary value @xmath50 . in particular @xmath14 is non - negative . consider another surface ball @xmath315 that is @xmath376distance away from @xmath121 and also has radius @xmath168 , and denote @xmath377 . then by assumption , @xmath378 we have shown in that @xmath379 in order to lower bound the square function @xmath380 , we need to decompose the non - tangential cone @xmath381 as follows . for any @xmath382 and any @xmath383 , denote @xmath384 as a stripe in the cone @xmath381 at height @xmath385 , and denote @xmath386 a union of @xmath387 stripes . the above figure illustrates these notations , even though it over - simplifies the shape of the cone @xmath39 and the relations between different radii . recall that @xmath50 is supported in @xmath121 , @xmath14 is the corresponding solution with boundary value @xmath50 , and @xmath315 is disjoint from @xmath121 . hence @xmath14 vanishes on @xmath315 , and we have the following poincar type estimate : [ lm : poincare ] there exist an aperture @xmath388 , and integers @xmath389 , such that the following poincar inequality holds for all @xmath382 , @xmath390 the constants @xmath389 and @xmath36 only depend on @xmath305 and @xmath87 . we are going to sketch the proof first , and then fill in the details using harnack chains . we need to make a brief remark on the notation before we start . so far we have been using @xmath391 and @xmath392 to distinguish solid integrals and integrals on the surface , but we only need solid integral in the proof of this lemma . to simplify the notation , we simply use @xmath392 for solid integral in the scope of this proof . the following lemma is the standard poincar type inequality ( see ( 7.45 ) and lemma 7.16 in @xcite ) . let @xmath393 be a ball in @xmath176 . if the function @xmath394 , then @xmath395 here @xmath396 , @xmath397 is the @xmath398dimensional lebesgue measure of @xmath393 and @xmath399 is the volume of the unit ball in @xmath176 . let @xmath400 denote the radius of @xmath393 , then it follows from the above lemma that @xmath401 assume there is a harnack chain @xmath402 from @xmath403 to @xmath404 , by the triangle inequality we have @xmath405 assume @xmath406 and @xmath407 are two intersecting balls with comparable sizes @xmath408 where @xmath409 are two constants . we want to estimate @xmath410 by the integral of @xmath411 . by @xmath412 and , we know @xmath413 therefore @xmath414 hence by , @xmath415 similarly @xmath416 therefore @xmath417 assume the radii of each pair of consecutive balls in the harnack chain have the same ratio bounds as in , then it follows from and that @xmath418 on the other hand , by assumption the last ball @xmath419 , we have @xmath420 where @xmath320 is the corkscrew point in @xmath421 . the second inequality is by , and the last one by . since @xmath422 is a non - tangential ball and @xmath14 is non - negative , by the harnack principle @xmath423 for a constant @xmath424 . hence @xmath425 combine and we get @xmath426 plugging back in , we obtain @xmath427 choose @xmath428 big enough such that @xmath429 then we can absorb the first term on the right hand side of to the left , and obtain @xmath430 when @xmath428 is fixed , the number of balls @xmath431 is also fixed , thus the above inequality does not depend on @xmath112 . this is the above poincar inequality for non - tangential balls . after we establish , in order to prove the poincar inequality for non - tangential cones , we need to cover @xmath432 by balls ; we also need to choose the harnack chain carefully so that the integration region @xmath433 in the right hand side of is contained in a non - tangential cone , possibly of a bigger aperture and wider stripe . first we make the following simply observation : [ obs : one ] let @xmath393 be a harnack ball with constants @xmath116 ( see the definition of harnack balls in ) . assume @xmath393 contains a non - tangential point @xmath434 with @xmath435 , then @xmath436 if in adition @xmath437 , then @xmath438 , and we can get more precise estimate of @xmath393 : [ obs : two ] let @xmath393 be a harnack ball with constants @xmath116 . assume @xmath393 contains a non - tangential point @xmath439 , then @xmath440 moreover , the following observations can be made by induction : [ obs : three ] let @xmath393 and @xmath441 be two harnack balls with constants @xmath442 . assume @xmath443 and @xmath393 contains some point @xmath437 , then @xmath444 where @xmath445 is an integer depending only on @xmath114 ( more precisely , @xmath445 is such that @xmath446 ) . [ obs : four ] inductively , if @xmath447 are harnack balls with constants @xmath442 such that @xmath448 , and @xmath449 contains some point @xmath450 , then @xmath451 let @xmath12 and @xmath96 be arbitrary points in @xmath432 and @xmath452 respectively . then @xmath453 and @xmath454 by the harnack chain condition , there is a chain of open harnack balls @xmath98 satisfying @xmath455 that connects @xmath12 and @xmath96 , and the number of balls @xmath431 . by observation [ obs : two ] , the balls @xmath449 and @xmath456 are non - tangential balls of aperture @xmath457 : @xmath458 simply computation shows that the sizes of two consecutive harnack balls are comparable : @xmath459 recall we showed in section [ sect : poincare ] ( see ) that the comparable size condition implies @xmath460 for all @xmath103 , with a constant @xmath461 depending on @xmath462 . choose an appropriate @xmath114 so that the enlarged ball @xmath463 still lies in @xmath7 , and moreover , its distance to the boundary is still comparable to its diameter . more precisely , it is an easy exercise to show that if we choose @xmath464 , the enlarged balls @xmath465 s are still harnack balls with modified constants : @xmath466 denote @xmath467 ( it stands for integration tube ) . by and observation [ obs : four ] , @xmath468 where @xmath469 depend on the constants of the harnack balls @xmath470 , the number of balls @xmath471 and the @xmath87 we start with . thus by @xmath472 to summarize , for any @xmath473 , we can find a harnack ball @xmath393 containing @xmath12 , and satisfies @xmath474 , and @xmath475 where the aperture @xmath476 , and @xmath477 are constants depending on @xmath87 . we cover @xmath432 by such harnack balls : @xmath478 and each ball @xmath479 has radius comparable to @xmath385 . from the covering we can extract a vitali sub - covering : @xmath480 such that @xmath481 are pairwise disjoint . by the definition and , the set @xmath482 is contained in an annulus with small radius @xmath483 and big radius @xmath484 , so its lebesgue measure is bounded by @xmath485 up to a constant . by the disjointedness of @xmath481 s and the fact that each @xmath486 has radius comparable to @xmath385 , we can show that the number of balls in the vitali covering is bounded by a constant @xmath487 . the proof is exactly the same as and in lemma [ lm : delta ] , so we omit it here . finally , by and , we have @xmath488 this finishes the proof of lemma [ lm : poincare ] . recall in the beginning of section [ sect : showainfty ] , we have and for a continuous non - negative function @xmath50 supported in @xmath121 and for a surface ball @xmath315 of radius @xmath168 and @xmath376distance away from @xmath121 . combining and , we get @xmath489 where @xmath388 is the aperture in lemma [ lm : poincare ] , and the square function is defined as @xmath490 here we use the square function defined in a bigger cone @xmath491 , so that we can lower bound the square function using the poincar type inequality : @xmath492 where @xmath493 is a corkscrew point at the scale @xmath385 . in the last inequality , we used the interior corkscrew condition , thus each stripe of cone @xmath494 contains a ball of radius comparable to @xmath495 ( as long as @xmath87 is chosen to be big , say @xmath496 , where @xmath112 is the constant in the definition of the corkscrew condition ) . by the harnack principle @xmath497 , where @xmath240 is a constant independent of @xmath14 and @xmath103 . thus @xmath498 recall for any @xmath499 , the point @xmath500 is a corkscrew point in @xmath501 . let @xmath96 be the corkscrew point in @xmath502 , again by the harnack principle we have @xmath503 . therefore @xmath504 plugging this back in we get @xmath505 and thus @xmath506 let @xmath12 be a corkscrew point in @xmath121 . since @xmath121 and @xmath315 have the same radius @xmath168 , and they are @xmath376distance apart , by the harnack principle @xmath507 . by the assumption @xmath50 is supported on @xmath121 , @xmath508 the last equality is due to the estimate of @xmath509 when @xmath12 is a corkscrew point in @xmath286 and @xmath510 ( see @xcite corollary 1.3.8 ) . as a result , we proved the following estimate : let @xmath50 be a non - negative continuous function supported on @xmath121 , then @xmath511 where the constant @xmath36 is independent of @xmath50 . let @xmath121 be a surface ball with radius @xmath168 . for @xmath131 fixed , we shall find an @xmath512 , such that for any @xmath513 , @xmath514 in fact , since @xmath30 and @xmath313 are borel measures , we may assume @xmath167 is an open subset of @xmath121 . let @xmath515 be a small constant to be determined later , we define the function @xmath516 where @xmath517 is the hardy - littlewood maximal function with respect to @xmath30 , and @xmath518 since @xmath519 is a space of homogeneous type , we can adapt the arguments in @xcite very easily and show that @xmath520 for a constant @xmath12 independent of @xmath167 . hence @xmath50 is a bmo function , and @xmath521 moreover , it is clear to see from the definitions and that @xmath522 and @xmath523 on the open set @xmath167 . let @xmath524 be a point outside of @xmath525 , then @xmath526 . since @xmath133 and @xmath306 contains @xmath527 , in order for @xmath528 to be nonempty , the diameter of @xmath529 is at least @xmath168 . thus by ahlfors regularity , @xmath530 therefore @xmath531 for all @xmath532 . this means , as long as @xmath133 is such that @xmath533 by we have @xmath534 hence @xmath535 outside of @xmath525 . in other words , @xmath536 next we want to use a mollification argument to approximate @xmath50 by continuous functions , such that their bmo norms are uniformly bounded by the bmo norm of @xmath50 . let @xmath537 be a radial - symmetric smooth function on @xmath176 such that @xmath538 on @xmath539 , @xmath540 and @xmath541 . let @xmath542 define for all @xmath524 @xmath543 in fact @xmath544 is a ( renormalized ) convolution of @xmath50 by @xmath545 restricted to the boundary @xmath19 . the following lemma summarizes the properties of these @xmath544 s , the proof of which is postponed to appendix [ sect : feprop ] . [ lm : fe ] the following properties hold for @xmath544 s : 1 . each @xmath544 is continuous , and is supported in @xmath546 ; 2 . there is a constant @xmath36 ( independent of @xmath547 ) such that @xmath548 ; 3 . @xmath549 for all @xmath527 in their support @xmath546 . the last property and fatou s lemma imply @xmath550 since each @xmath544 is continuous , we can apply , @xmath551 combining , and , we get @xmath552 on the other hand , since @xmath553 and @xmath313 is a doubling measure , @xmath554 therefore @xmath555 as long as the condition , i.e. @xmath556 is satisfied . in other words , @xmath557 . in this section we are going to prove the following proposition . [ prop : reversecarl ] assume @xmath0 . if @xmath143 in @xmath7 with boundary data @xmath13 , then @xmath558 as long as the right hand side is finite . since @xmath0 , a classical result in harmonic analysis says @xmath559 . we may as well prove for @xmath560 . in the light of previous work @xcite and @xcite , jerison and kenig studied the dirichlet problem with bmo boundary data and proved the following theorem for the laplacian on nta domains ( see @xcite theorem 9.6 ) . the main ingredients of their proof are the properties , and a geometric localization theorem . in section [ sect : br ] we show these properties hold for general elliptic operators on uniform domains with ahlfors regular boundary , therefore a similar proof is applicable in our case . we include the proof here for the sake of completeness . [ thm : bmogreen ] there exists a constant @xmath142 such that @xmath561 on condition that the right hand side is bounded . let @xmath118 be an arbitrary surface ball . by theorem [ thm : glt ] there is a domain @xmath562 satisfying @xmath563 and @xmath562 is a uniform domain with ahlfors regular boundary . assume the radius @xmath168 is such that @xmath564 . by the interior corkscrew condition , we can find a point @xmath565 and @xmath566 . for the elliptic operator @xmath3 on @xmath562 , let @xmath129 be the elliptic measure , and @xmath567 be the green s function with pole at @xmath568 . we claim @xmath569 for any surface ball @xmath570 . in fact , let @xmath571 and @xmath572 be corkscrew points of @xmath121 and @xmath315 respectively , and @xmath573 be the radius of @xmath315 . apply to the domains @xmath7 and @xmath562 , we get @xmath574 thus @xmath575 similarly @xmath576 note that @xmath577 , by the boundary comparison principle @xmath578 combining , and we get @xmath579 since @xmath580 and @xmath129 is a doubling measure , we have @xmath581 . therefore @xmath582 the above estimate in particular implies @xmath583 here we choose constant @xmath584 , so that @xmath585 . hence we have the global estimate on @xmath562 ( see lemma 1.5.1 in @xcite ) @xmath586 recall and @xmath587 , in other words , @xmath588 note that @xmath589 , by considering harnack chains in @xmath590 or using the boundary comparison principle , implies @xmath591 thus @xmath592 on the other hand , by the harnack principle , on @xmath593 @xmath594 by the choice of @xmath568 we know @xmath595 and @xmath596 . let @xmath597 such that @xmath598 , then by properties and , @xmath599 for any @xmath600 . plugging into , we get @xmath601 by the maximal principle and the bound on green s function , @xmath602 note that the last inequality is independent of @xmath562 and @xmath568 . combining and , @xmath603 summing up and , we get @xmath604 together with and , we deduce @xmath605 recall for @xmath147 , the set @xmath606 is defined as @xmath607 . by and the properties , , we have @xmath608 thus by changing the order of integration , @xmath609 where @xmath610 . since @xmath68 , there exists @xmath137 such that the radon - nikodym derivative @xmath611 . let @xmath612 be the conjugate of @xmath613 , i.e. @xmath614 , we have @xmath615 the second inequality is because @xmath313 is a doubling measure . we claim the following theorem about the truncated square function : [ tm : squarecarleson ] for any @xmath616 , @xmath617 on condition that the right hand side is finite . here @xmath618 denotes any surface ball of radius @xmath168 . assume this theorem holds , then follows from combining , , and , which concludes the proof of proposition [ prop : reversecarl ] . now we are going to prove the theorem by combining several lemmas about the truncated square function . in @xcite , the authors have proved similar lemmas for the square function ( see proposition 4.5 , lemma 4.6 and lemma 6.2 ) , and we are adapting their arguments to truncated square function . the proof of the following two lemmas is similar to the case of non - truncated square function , so we postpone it to appendix [ sect : trsq ] . [ lemma : open ] for any @xmath619 , the set @xmath620 is open in @xmath19 . [ lemma : diffapper ] let @xmath616 . assume @xmath621 is an aperture bigger than @xmath87 , and @xmath622 is a surface ball of radius @xmath168 , then @xmath623 moreover , we have the following good-@xmath461 " inequality between the truncated square function and the carleson type function @xmath624 [ lm : goodlambda ] there exist an aperture @xmath625 and a constant @xmath142 such that for any surface ball @xmath626 and any @xmath627 , the following estimate holds : @xmath628 let @xmath629 . consider the open set @xmath630 . similar to lemma 3.7 in @xcite , let @xmath631 be a whitney decomposition of @xmath632 , such that for each @xmath362 , @xmath633 where @xmath634 , and @xmath635 . we claim that for all @xmath636 such that @xmath637 , we have @xmath638 this is clearly true if the left hand side is empty , so we may assume without loss of generality that it is nonempty , and @xmath639 if @xmath640 for some @xmath641 , we define @xmath642 then @xmath643 note that @xmath644 we need to consider two cases . _ case 1 . _ assume @xmath645 is such that @xmath646 let @xmath647 be arbitrary , recall that @xmath648 , thus @xmath649 if @xmath640 and @xmath650 , we have @xmath651 and @xmath652 in other words @xmath653 . hence @xmath654 therefore @xmath655 if @xmath640 and @xmath656 , recall @xmath657 and , we have @xmath658 hence @xmath659 where the carleson type function is defined in . by , we know that @xmath660 . in addition , @xmath30 is an ahlfors regular measure and @xmath661 . therefore it follows from that @xmath662 where @xmath36 is a constant depending on @xmath87 and the constants @xmath462 , the lower and upper bounds of the ahlfors regularity of @xmath30 . on the other hand , @xmath663 thus @xmath664 recall @xmath665 for all @xmath641 ( see ) , therefore @xmath666 _ case 2 . _ assume @xmath645 is such that @xmath667 . ( note that if @xmath668 , the proof is done . ) we only consider the @xmath636 s such that @xmath669 , so assume @xmath670 is a point in the intersection . in particular @xmath671 , so @xmath672 . so @xmath673 on the other hand , if @xmath674 for some @xmath675 , then @xmath676 thus it follows that @xmath677 . in particular , for all the @xmath640 for some @xmath641 , @xmath678 in other words , @xmath679 . similar to , one can show @xmath680 therefore @xmath681 this finishes the proof of . summing up for @xmath362 s such that @xmath682 , we obtain @xmath683 the last inequality is because @xmath684 is a decomposition of the set @xmath685 . this also implies @xmath686 therefore @xmath687 the last equality is because @xmath388 and @xmath688 combining and , we get @xmath689 by lemma [ lm : goodlambda ] , @xmath690 here @xmath691 stands for the carleson measure defined by the function @xmath14 , and thus by definition , @xmath692 apply lemma [ lemma : diffapper ] to the right hand side of , the inequality becomes @xmath693 where @xmath694 is a constant and @xmath695 . if the radius @xmath168 is small enough such that @xmath696 , we can rewrite the above inequality in the average integral form : @xmath697 pick @xmath698 small enough ( depending on @xmath87 ) such that @xmath699 , and denote @xmath700 , then @xmath701 theorem 6.1 in @xcite states the following global estimate @xmath702 thus @xmath703 we claim the contractionestimate , together with the global estimate implies @xmath704 for @xmath168 arbitrary , let @xmath362 be a positive integer such that @xmath705 , then @xmath706 . ( the constants of equivalence only depends on @xmath12 and the ahlfors regularity of @xmath30 . in particular they do not depend on @xmath168 or @xmath707 . ) by and , we have @xmath708 hence @xmath709 to simplify the notations , we denote @xmath710 where @xmath114 and @xmath110 are the constants in the inequalities and respectively , and @xmath711 . thus we can rewrite and as @xmath712 and @xmath713 induction on , we obtain @xmath714 let @xmath715 , then becomes @xmath716 note the constant @xmath36 is independent of @xmath707 and @xmath168 , we conclude that @xmath717 this finishes the proof of the theorem [ tm : squarecarleson ] , hence the conclusion follows . we are going to prove lemma [ lm : fe ] in this section . note that @xmath544 as in is well defined since @xmath718 on the other hand we also have @xmath719 the constants @xmath114 and @xmath110 are independent of @xmath547 . _ proof of ( 1 ) . _ for any surface ball @xmath720 , we denote @xmath721 as the surface ball @xmath722 . since @xmath50 is supported in @xmath525 , it is clear that each @xmath544 is supported in @xmath723 . thus all @xmath544 s are supported in @xmath546 if we only consider @xmath547 smaller than @xmath168 , the radius of @xmath121 . note that @xmath724 since @xmath725 , for any @xmath726 we have @xmath727 thus @xmath728 so for any @xmath131 fixed , @xmath729 is a continuous function with respect to @xmath524 . since @xmath730 , we can prove similarly that @xmath731 is also continuous . thus @xmath732 is continuous with respect to @xmath524 . _ proof of ( 2 ) . _ next we want to show @xmath733 . fix @xmath131 . let @xmath461 be a real number to be determined later . let @xmath734 be an arbitrary surface ball with center @xmath735 and radius @xmath736 . we consider two cases . _ if @xmath737 , by the definition and the estimates , , @xmath738 the last inequality is true if we let @xmath739 be the constant satisfying @xmath740 thus @xmath741 since @xmath742 . _ if @xmath743 , by the definition and the estimate , @xmath744 note @xmath745 it follows from that @xmath746 since @xmath747 . we have proved the following : there exists a constant @xmath36 such that for any @xmath547 and any surface ball @xmath306 , one can find a constant @xmath748 such that @xmath749 the constant @xmath36 does not depend on either @xmath547 or @xmath306 , therefore we conclude @xmath733 for all @xmath547 . _ proof of ( 3 ) . _ the fact @xmath750 is obviously true when @xmath751 . for an arbitrary @xmath8 , consider @xmath752 such that @xmath753 . for fixed @xmath527 , there exists some @xmath754 such that @xmath755 . it means that @xmath756 in particular , there exists some surface ball @xmath315 containing @xmath527 such that @xmath757 then for any point @xmath758 , we also have @xmath759 , and hence @xmath760 . consider all @xmath544 with @xmath761 , we have @xmath762 , hence @xmath763 therefore @xmath764 for all @xmath8 and all @xmath527 satisfying @xmath765 . thus @xmath766 for all @xmath527 . assume @xmath78 satisfying @xmath767 then there exists @xmath768 such that @xmath769 for @xmath770 fixed , we claim there exists @xmath131 such that @xmath771 for any @xmath772 . in fact , @xmath773 where @xmath187 is the set difference between @xmath774 and @xmath775 . therefore @xmath788 , where @xmath789 is defined as the union of the following three sets ( corresponding to the above three cases ) @xmath790 @xmath791 @xmath792 similarly @xmath793 , where @xmath794 is defined as above with @xmath795 replaced by @xmath780 . using the triangle inequality and @xmath781 again , we get @xmath796 , where @xmath797 @xmath798 @xmath799 recall @xmath187 is the set difference between @xmath800 and @xmath801 , so @xmath802 in particular , @xmath803 in @xmath187 . hence @xmath804 for an arbitrary @xmath515 , we want to find @xmath805 small enough , so that the right hand side of is bounded by @xmath806 as long as @xmath781 . note that @xmath807 , so @xmath808 therefore the integral @xmath809 is small as long as the lebesgue measures of @xmath810 , @xmath811 and @xmath812 are all small enough . both @xmath810 and @xmath811 are contained in annuli of radius @xmath813 , so their lebesgue measures are small if we choose @xmath547 small enough ( depending on @xmath770 ) . for the set @xmath812 , rewrite it as @xmath814 the function @xmath815 is a lipschitz function in @xmath816 , and @xmath817 . if we choose @xmath818 , then for any @xmath819 , @xmath820 namely , @xmath821 and @xmath822 is lipschitz on @xmath812 . by the co - area formula , @xmath823 on the other hand , @xmath824 therefore by , we may choose @xmath547 small enough ( depending on @xmath87 ) such that @xmath825 is small , which in turn implies @xmath826 would be small . more precisely , we have shown that we can choose @xmath547 small enough such that @xmath827 thus we conclude from and that @xmath828 therefore @xmath829 for all @xmath830 . this finishes the proof that @xmath831 is open in @xmath19 . we prove the estimate by duality : let @xmath168 be the conjugate of @xmath832 , namely @xmath833 , then @xmath834 extending @xmath835 to all of @xmath19 by setting it to be zero outside of @xmath121 , and still denote the extension @xmath835 . for any @xmath23 , let @xmath836 be a boundary point such that @xmath837 . recall that @xmath146 . by fubini theorem , @xmath838 where @xmath839 is defined to be @xmath840 since @xmath30 is ahlfors regular , there exists a constant @xmath36 such that @xmath841 , where @xmath842 is the hardy - littlewood maximal function of @xmath835 with respect to @xmath30 . let @xmath843 , then @xmath844 let @xmath845 , @xmath846 , then @xmath847 thus it follows from that @xmath848 we want to switch the order of integration in the above integral . assume @xmath11 and @xmath78 satisfy @xmath849 in particular @xmath850 . @xmath852 thus we can bound the right hand side of by the following : @xmath853 because @xmath854 , @xmath855 for any function @xmath835 supported in @xmath121 with @xmath856 norm one . by the definition , @xmath857 this finishes the proof of lemma [ lemma : diffapper ] . c. kenig , h. koch , j. pipher , t. toro , _ a new approach to absolute continuity of elliptic measure , with applications to non - symmetric equations_. adv . math . * 153 * ( 2000 ) , no . 2 , 231 - 298 , doi:10.1006/aima.1999.1899 . mr1770930 ( 2002f:35071 ) w. littman , g. stampacchia , h.f . weinberger , _ regular points for elliptic equations with discontinuous coefficients_. annali della scuola normale superiore di pisa - classe di scienze 17.1 - 2 ( 1963 ) : 43 - 77 .

we consider the effect of energy dependent propagation of relativistic electrons in a spatially inhomogeneous medium in order to interpret the broad - band nonthermal radiation of the young shell - type supernova remnant ( snr ) cassiopeia a. a two - zone model is proposed that distinguishes between compact , bright steep - spectrum radio knots and the bright fragmented radio ring on the one hand , and the remainder of the shell - the diffuse ` plateau ' - on the other hand . in the framework of this model it is possible to explain the basic features of the spectral and temporal evolution of the synchrotron radiation of cas a if one assumes that these compact structures correspond to sites of efficient electron acceleration producing hard spectra of accelerated particles with power - law indices . the resulting energy distribution of radio electrons in these compact structures becomes significantly steeper than the electron production spectrum on timescales of the energy dependent escape of these electrons into the surrounding diffuse plateau region . we argue that the steepness , rather than the hardness , of the radio spectra of compact bright structures in clumpy sources can in general be considered as a typical signature of sites where strong electron acceleration has built up high gradients in the spatial distribution of radio electrons . subsequent diffusive escape then modifies their energy distribution , leading to potentially observable spatial variations of spectral indices within the radio source . qualitative and quantitative interpretations of a number of observational data of cas a are given . predictions following from the model are discussed .

astro-ph0001187

cassiopeia a is the youngest of the known galactic supernova remnants ( snrs ) whose birth probably dates back to 1680 ( ashworth @xcite ) . it is one of the most prominent and well studied radio sources on the sky ( e.g. bell et al . @xcite ; tuffs @xcite ; braun et al . @xcite ; anderson et al . @xcite hereafter arlpb ; kassim et al . @xcite , hereafter kpde ; etc . ) , whose synchrotron radiation probably extends into the hard x - ray region ( allen et al . @xcite ; favata et al . most of the radiation of both nonthermal and thermal origin comes from a shell region enclosed between two spheres , with angular radii @xmath1 and @xmath2 , corresponding to spatial radii @xmath3 and @xmath4 , respectively , for a distance of @xmath5 ( reed et al . the former corresponds to the mean radius of the assumed blast wave , while the latter is supposedly the mean radius of the reverse shock in the freely expanding ( upstream ) ejecta which are heating the gas to temperatures @xmath6 , thus creating the hot thermal x - ray component ( becker et al . @xcite ; fabian et al . @xcite ; jansen et al . @xcite ) . cas a is considered to be a powerful particle accelerator , with an energy content in relativistic electrons , estimated from simple equipartition arguments , in the range @xmath7 ( e.g. chevalier et al . @xcite ; arlpb ) . this value should be increased by more than an order of magnitude for the total energetics in relativistic particles if the high ratio between relativistic protons and electrons observed in cosmic rays ( cr ) holds in cas a. however , the mechanisms and sites of particle acceleration in cas a have not yet been identified . observations of cas a at optical wavelengths show strong line emission from 2 main types of compact structures . the quasi stationary flocculi are thought to trace the circumstellar medium ( e.g. fesen et al . @xcite ) , while numerous fast moving knots ( fmk ) represent dense clumps of supernova ejecta moving ballistically with average velocities of @xmath8 from the time of explosion ( see e.g. reed et al . @xcite , and references therein ) . the fmks have been invoked as sites of particle acceleration by scott & chevalier ( @xcite ) who proposed second order fermi acceleration of particles in the turbulence created in the wake of fmks overtaking the shell . another possibility is a first order fermi acceleration process at the bow shocks driven ahead of fmks as proposed by jones et al . ( @xcite ) . these authors showed that these dense gas ` bullets ' can effectively accelerate electrons at the stage of deceleration and subsequent fast destruction of the fmks by kelvin - helmholtz and rayleigh - taylor instabilities . perhaps the most straightforward suggestion , however , is that efficient electron acceleration occurs directly in regions of high radio brightness in other words , that these regions are bright not only because of enhanced magnetic field , but also due to local enhancement of relativistic electrons . such regions have been suggested to be sites of electron acceleration by a number of authors ( e.g. bell @xcite , dickel & greisen @xcite , cowsik & sarkar @xcite ) . in cas a they are represented by rather compact structures which include @xmath9 radio knots ( tuffs @xcite , arlpb ) located in the shell , several paraboloidal radio features suggested to be bow shocks associated with decelerated ejecta ( braun et al . @xcite ) , and the bright fragmented radio ring at the projected radius of the ( supposedly ) reverse shock / contact discontinuity . examples of all these morphological structures are depicted in fig.1 . any theory of particle acceleration in cas a has to address a wide range of spatial and spectral characteristics of the observed nonthermal radiation from the radio to infrared ( ir ) , and possibly to x - ray regimes , which can be summarized as follows : * for @xmath10 the total radio flux is well approximated by a power - law @xmath11 with an index @xmath12 ( e.g. baars et al . @xcite ) ; * for @xmath13 the spectrum @xmath14 essentially flattens , and turns over below 15mhz ( baars et al . @xcite ) ; * the secular decline of the fluxes seems to be frequency - dependent , with a decline rate at the level of @xmath15 in the frequency range @xmath16 ( rees @xcite ; hook et al . 1992 ) , and @xmath17 at @xmath18 ( dent et al . @xcite ; osullivan & green @xcite ) . * the contribution of the diffuse emission of the shell ( _ plateau _ ) to the total flux of cas a at 5ghz amounts to @xmath19 , and the second half is due to the fragments of bright radio ring ( @xmath20 ) and radio knots + bow shocks ( @xmath21 ) ( tuffs @xcite ) . the total flux at the epoch 1987 was about 750jy ( see arlpb ) . * individual radio knots show a wide spread of spectral indices in the range from @xmath22 to @xmath23 ( rosenberg @xcite ; tuffs @xcite ; arlpb ; ar96 ) . * for @xmath24 the radio spectrum flattens to @xmath25 ( mezger et al . @xcite ) which possibly extends to the ir region if the flux of continuum emission measured at 6@xmath26 by tuffs et al . ( @xcite ) has a synchrotron origin . * recent observations reveal hard x - ray emission that extends with a power - law photon index @xmath27 up to 120kev ( allen et al . @xcite ; favata et al . @xcite ) ; a synchrotron origin of this radiation implies acceleration of electrons to multi - tev energies . detailed studies based on high resolution mapping of cas a at ghz frequencies have suggested that the identification of bright compact radio features with electron acceleration sites can be problematic ( arlpb ; anderson & rudnick @xcite , hereafter ar96 ) . a particularly interesting finding of arlpb and ar96 is a significant and rather unexpected correlation between the spectral index of the knots and their projected position in the shell , as well as their radio brightness : the steeper radio knots reside mostly in the outer regions of the shell , at @xmath28 , and tend to be brighter . the analysis in ar96 has shown that a self - consistent explanation of these correlations would be problematic if one assumes that the radio knots are the sites of efficient acceleration of the electrons . for an snr as young as cas a the radiative energy losses can not modify the spectrum of radio electrons . therefore in the framework of a ` standard ' spatially homogeneous source model approach , one has to attribute the radio spectral indices observed to the source spectra of the electrons . the observed brightness / steepness trend of the radio knots would then apparently rule out effective electron acceleration in the knots , because such a process implies a hardening ( see e.g. berezhko & vlk @xcite ) , rather than a steepening of the particle spectra , typically to the power - law index @xmath29 . there does exist however a natural way to modify the spectra of radio electrons , if we abandon the standard approach of a spatially uniform source for treatment of these electrons . spectral modifications become unavoidable if we take into account _ energy - dependent _ propagation and escape of relativistic particles from the regions of higher concentration in an inhomogeneous medium . the efficiency of this process depends on the spatial gradients in the energy distribution @xmath30 of particles , and timescales of the spectral modifications can be as short as the escape time @xmath31 . this effect is widely used for the interpretation of the galactic cr spectra in the framework of diffusive or leaky box models , but it has not yet been given proper attention in studies of radio emission in the cr sources themselves . in this paper we consider the consequensies of energy dependent propagation of relativistic electrons for the interpretation of the observed spectral , spatial and temporal characteristics of the broad - band nonthermal emission of cas a. in sect . 2 we introduce the ` two - zone ' model for a spatially non - uniform radio source , separating compact regions with a high density of relativistic electrons ( zone 1 ) from the rest of the shell where the electron density is significantly lower ( zone 2 ) . in sect . 2 we consider in a qualitative way possible consequensies of such an approach for the interpretation of the observed radio data . in sect . 3 we derive the system of kinetic equations for the electron energy distributions in the two zones . in sect . 4 we assume that zone 1 components correspond to the sites of efficient electron acceleration , and show that this scenario is able to explain the broad band non - thermal radiation data of cas a. in sect . 5 we study the opposite scenario , which assumes that the enhancement of the electron density in zone 1 is caused not by active acceleration of particles there , but rather only by compression of the background population of relativistic electrons in the shell . we show that interpretation of the data within this latter scenario is problematic . in sect . 6 we summarize the observational features which the model can explain , and in sect . 7 we discuss implications and predictions of the model . to lowest approximation cas a might be represented as a homogeneous shell containing magnetic field , relativistic electrons and gas . although not reflecting the real pattern of the source , this ` single - zone ' approximation allows estimates of basic parameters such as the mean magnetic field @xmath32 and energetics in radio electrons . in this approximation cowsik & sarkar ( @xcite ) have derived a lower limit to @xmath33 , comparing the expected bremsstrahlung flux of gev electrons with the upper limit @xmath34 of sas-2 ( fichtel et al . @xcite ) and cos b detectors . with the recent upper limits of the egret telescope , @xmath35 ( esposito et al . @xcite ) , this constraint on @xmath32 is significantly strengthened . in fig.2 we present the fluxes of the bremsstrahlung and synchrotron radiations calculated for 3 values of the mean magnetic field : @xmath36 , @xmath37 and @xmath38 . for the mean gas density ( in terms of ` h - atoms ' ) we take @xmath39 , corresponding to @xmath40 of matter in the shell ( fabian et al . @xcite , jansen et al . @xcite , reed et al . @xcite ) , and use a mean atomic @xmath41 derived by cowsik & sarkar ( @xcite ) from the elemental abundance estimates of chevalier & kirshner ( @xcite ) in cas a. continuous injection of electrons into the shell , with a spectrum @xmath42 where @xmath43 and an exponential cutoff energy @xmath44 , is assumed . the chosen @xmath45 explains the fluxes of hard x - rays above 10kev as synchrotron emission for the case of @xmath46 . however , this value of @xmath32 is unacceptable because the bremsstrahlung flux produced by gev radio electrons is too high . the egret upper limit for the @xmath47-ray flux of cas a requires at least @xmath48 . simultaneously we have to assume a significant flattening in the source spectrum of electrons below 100mev in order to avoid contradictions with the observed x - ray fluxes at 100kev . for a single power - law distribution of electrons extending with @xmath49 down to the mev region , the lower limit to the magnetic field is @xmath50 . note however , that for @xmath51 the radiative energy losses of the electrons establish a spectral break at optical or lower frequencies , in which case the spatially homogeneous single - zone model fails to explain the observed hard x - ray fluxes by synchrotron radiation . a more realistic model for cas a has to account for the inhomogeneities in the radio brightness distribution and for the observed spread in the power - law exponents of the radio spectra of individual structures within @xmath52 . the simplest approximation for a non - uniform radio source corresponds to a ` two - zone ' model , where the source with total volume @xmath53 consists of 2 different regions , or zones , with volumes @xmath54 and @xmath55 , with 2 different spatial densities ( concentrations ) of radio electrons @xmath56 and @xmath57 . propagation effects may result in a significant difference between the two - zone and single - zone models only if these densities are essentially different . therefore the principal definition for these two zones is that the concentration of electrons in zone 1 is much higher than in zone 2 , @xmath58 . the energy distributions of the electrons in these zones can be different . therefore in the power - law approximation @xmath59 the exponents @xmath60 and @xmath61 can be different . because the spectral ratio @xmath62 may thus depend on energy , it is convenient to specify that the condition @xmath58 relates first of all to electrons with energies @xmath63 which are typically responsible for radio emission in the ` 1 ghz ' band . in order to apply the two - zone model to cas a , we have to foresee some observational manifestations of the radio emission of zones 1 and 2 . it is reasonable to expect that not only the electron densities , but also the mean magnetic fields @xmath64 and @xmath65 in the two zones differ , with the magnetic field presumably being higher in the regions of higher electron densities . therefore , the radio emissivities in these zones will be strongly contrasting , with regions belonging to zone 1 characterized by a much larger emissivity than regions in zone 2 . it is also plausible that zone 1 components should be rather compact , with a small volume filling factor in the source , or otherwise the overall emission would be strongly dominated by this single zone . application of this criterion to cas a immediately places all bright fragments of the radio ring in zone 1 . other constituents of zone 1 are the radio knots ( and bow shocks ) most of which are characterized by a strongly enhanced emissivity . we note however that in reality the knots embrace a wide range of emissivities , and because the simplified two - zone approach allows only 2 different electron densities ( and emissivities ) , some compact structures with relatively low brightness should be more correctly classified as belonging to zone 2 . this may have a physical basis enhanced emissivity due to mild compression of magnetic fields , but without strong excess in the electon densities . zone 2 then consists of the rest of the ( spherical ) shell between the radio ring and the outer edge of cas a , i.e. the main body of the radio plateau , but also includes some low - brightness radio knots . for purposes of practical classification of low - intensity knots it may prove useful to resort also to the second potentially observable distinction between zones 1 and 2 different mean spectral indices @xmath66 and @xmath67 of their intrinsic emission . due to a faster escape of higher energy electrons from the regions of high concentration , one could generally expect that the radiation spectra produced in zone 1 should be _ steeper _ , @xmath68 ( see below sect . because the spectral index of the total emission @xmath69 is to be maintained , an immediate conclusion is that the radiation spectrum of the diffuse zone 2 should be noticeably _ flatter _ , than the mean , i.e. @xmath70 . because of the complicated morphology , a quantitative treatment of the spectral indices of different types of structures is problematic in practice . in order to assess the characteristic values for @xmath71 , we note that the line of sight integrated radio brightness seen towards individual radio knots shows a broad range of spectral indices from @xmath72 to @xmath73 , as measured by ar96 between @xmath74 and @xmath75 . these spectral indices however do not always correspond directly to the _ intrinsic _ spectral indices of the knots , due to background contamination from the plateau . although the detailed least square fitting procedure for spectral index measurements used in ar96 ( their sect . 2.4 ) largely removes this problem for knots with high contrast to the plateau , for faint knots at the limit of confusion with the plateau , the spectral index will be biased towards the ( presumably flatter ) index of that emission . in this regard , the result of ar96 presented in their fig.6d appears as very informative . it shows that most of the radio knots found at large angular distances @xmath76 ( the _ projected _ @xmath77 ) belong to the steep - spectrum population : from 62 such knots in total , 45 have indices @xmath78 , and 22 ( i.e. @xmath79 ! ) belong to the ` extreme ' end of the observed spectral indices @xmath80 . because at these projected distances the brightness of the plateau is generally reduced with projected radius @xmath81 . it is plausible ( though not quantitatively investigated ) that this trend is due to the population of steep spectra radio knots detected by ar96 , which at lower angular resolution would contribute more flux than the plateau at angular distances @xmath82 ; the intrinsic spectrum of the plateau emission might be then flat . ] , this may indicate that the intrinsic spectral index of the radio knots on average could be much steeper than the mean @xmath83 , and that the spectral index @xmath84 of the ` extremely steep ' population of the observed knots might be not so extreme , but rather a representative value for the intrinsic index of the knots . an ( indirect ) indication of this suggestion could be also another result shown in fig . 6d of ar96 : 83 from 123 knots found at angular distances close to the radio ring , @xmath85 , where the brightness of the plateau is relatively high , belong to the population of knots with an ` intermediate ' steepness @xmath86 . as discussed below in sect . 6 , the model does not exclude a physical reason for such geometrical correlation between the knot index and its proximity to the ring . however , this effect can be explained also as a result of a line - of - sight mixture of _ comparable _ fluxes of the steep - spectrum knots with intrinsic @xmath87 and of the flat - spectrum diffuse plateau emission with @xmath88 ( see this sect . below ) . moreover , if we believe that the real geometry of the shell is quasi - spherical , it is difficult to suggest any other reasonable explanation , except for invoking the flux contamination effect , why only 2 knots from those 123 show the index @xmath89 , whereas the expected number of ` extremely steep ' knots to be located at deprojected radii @xmath90 but at the angular distances close to the `` ring '' should have been comparable with ) , we would still expect significantly more than 2 knots to be seen within projected distances @xmath85 ] the number @xmath92 . we can conventionally define that all knots which have a moderate brightness and show spectral index @xmath93 belong to zone 2 . then , given the tendency for brighter knots to be steeper ( ar96 ) , this classification would retain most of the radio knots ( at least in terms of their overal flux ) in zone 1 . at the same time , excluding thus the flat - spectrum knots from zone 1 , for the average intrinsic spectral index of the remaining population of knots one can reasonably suppose @xmath94 or so . it is possible that the intrinsic spectral index of the zone 1 components identified with the bright fragments of the radio ring is also close to this value . the low frequency measurements of woan & duffet - smith ( @xcite ; hereafter wds90 ) at angular resolution @xmath95 show that the average spectral index of the bright large radio components correlated with the ring is @xmath96 , i.e. again significantly larger than the mean @xmath83 of the total emission . if we accept that the contribution of the diffuse emission , zone 2 , may be flat , the intrinsic spectral index of the fragmented radio ring should be even larger . then @xmath97 may be a reasonable value for the average spectral index of the entire zone 1 . for qualitative discussions in this section we will use @xmath98 . we note however that our model does not pretend to predict characteristic spectral indices with the accuracy of the second digit after the comma . the first zone will thus include a number @xmath99 individual components , i.e. basically the bright fragmented radio ring , the ` radio ' bow shocks , and the main part of all radio knots . besides , zone 1 could also contain some regions of high concentration of electrons and high emissivities which are not immediately distinguished on the intensity maps either because of their very small sizes or because of strong contamination by , and confusion with the diffuse plateau emission . the latter may be the case if we believe that the fragmented `` radio ring '' actually represents a chain of large but geometrically flat ( thin ) structures of the `` radio sphere '' which are observed mostly ` edge - on ' . then at angular distances interior to the ring , @xmath100 or so , such fragments observed ` face - on ' may easily merge with the plateau emission and show up as diffuse ` clouds ' of only moderately enhanced brightness ( see fig . 1 ) . the flux produced in zone 1 can be estimated if we remember that at 5 ghz the ( _ background subtracted _ ) flux of the radio ring fragments makes up @xmath101 , and that radio knots produce @xmath102 of the total flux ( tuffs @xcite ) . the latter value should be somewhat reduced for the flux of knots to be removed from zone 1 and classified in zone 2 . on the other hand , it seems quite possible that up to @xmath103 of the total flux should be assigned to the ` unseen ' population of zone 1 ( especially , to face - on structures of the `` radio sphere '' ) . therefore , we can estimate the overall flux @xmath104 of zone 1 as @xmath105 of the total , concluding that the fluxes of zone 1 and zone 2 at 5 ghz are about the same , @xmath106 . the spectral index @xmath67 in the diffuse zone 2 can be then predicted if we note that the total flux has the form @xmath107 in order for the mean spectral index between @xmath108 and @xmath109 to be equal to @xmath110 , one needs a flux ratio @xmath111 then for @xmath112 , @xmath113 , and @xmath114 , we find @xmath115 . it may somewhat vary assuming slightly different @xmath116 or @xmath66 ( or @xmath83 , which can be also slightly less than 0.77 , see fig.2 ) . in particular , @xmath117 may be rather close to 0.6 if we allow for the idea that the ` unseen ' face - on structures of the `` radio sphere '' could contribute into zone 1 a third of what is contributed by the edge - on `` radio ring '' . remarkably , these values of @xmath117 correlate well with the lower range of spectral indices of the flat - spectrum radio knots . this suggests that flat - spectrum radio knots , with spectral indices close to 0.6 ( or flatter than 0.7 as classified above ) may indeed represent the regions of moderate local compression of the magnetic field and/or relativistic electrons of the diffuse shell , and therefore may give information about the spectrum of the background population of electrons there . thus , the model suggests a rather flat intrinsic spectrum of the radio plateau in the range @xmath118 , which corresponds to the spectral index of electrons @xmath119 . for zone 1 , @xmath120 implies a rather steep electron spectrum with @xmath121 . taking into account the enhanced brightness of the compact zone 1 , we will refer to zone 1 structures as the compact bright steep - spectrum radio ( cbsr ) components . even though still highly simplified , this two - zone approach suggests a qualitative interpretation of several observed radio features of cas a. in particular , it allows us to solve the problem of the low - frequency turnover below 20mhz . although the measurements of kpde at @xmath122 show evidence for thermally absorbing low - temperature gas in the central @xmath123 region of cas a , the thermal absorption can not result in any noticeable effect in the shell where @xmath124 . on the other hand , in the case of a spatially homogeneous source ( the shell ) with a mean field @xmath125 , the process of synchrotron self - absorption results in a turnover of the radio spectrum only at frequencies below 5mhz ( see fig.2 ) . to explain the observed turnover below @xmath126 , in a single - zone approach one needs @xmath127 . but this assumption boosts the magnetic field energy up to @xmath128 , and therefore is hardly acceptable . the situation essentially changes if one takes spatial inhomogeneities in the distribution of radio electrons and magnetic fields into account . indeed , in the framework of the proposed two - zone approach the radio flux at frequencies well below 1 ghz is dominated by the steep - spectrum component , i.e. one has to require synchrotron self - absorption mainly of the zone 1 flux at @xmath129 . because of the very high density of the radio electrons and of the magnetic field in compact zone 1 components , synchrotron self - absorption there should be expected at frequencies significantly higher than in the single - zone model . in order to estimate the model parameters needed for synchrotron self - absorption to occur below @xmath126 , we note that for a power - law distribution of electrons @xmath130 the synchrotron luminosity can be expressed as @xmath131 where @xmath132 with @xmath133 . @xmath134 is a weak function of @xmath135 that changes within @xmath136 for @xmath137 , and we will assume @xmath138 . with the same accuracy , the synchrotron absorption coefficient is given by @xmath139 where @xmath140 is the volume in units of @xmath141 . the volume of zone 1 can be represented in the form @xmath142 , where @xmath143 is the mean thickness of individual cbsr components along the line of sight and @xmath144 is their overall surface in the plane of the sky . requiring the synchrotron opacity to be significant at @xmath129 ( which implies @xmath145 at @xmath146 ) we obtain @xmath147 for the external radius of the plateau region @xmath148 this leads to an area filling factor @xmath149 where @xmath150 . for @xmath151 in zone 1 ( see sect . 4 ) , the ratio @xmath152 . from fig.1 it is apparent that the area filling factor of all bright radio features is indeed of the order of 10% . from eq.(5 ) it follows that the characteristic length scale for @xmath153 cbsr components of the zone 1 should be of order @xmath154 . if we separate the radio knots from the larger fragments of the radio ring , and take into account that the flux @xmath155 of the knots constitutes about a third of @xmath104 , the mean radius of the knots can be estimated as @xmath156 . at the distance 3.4kpc this corresponds to the angular sizes of about @xmath157 , very similar to what is really observed . the superposition of flat and steep spectral components , with @xmath106 around @xmath75 , results in a flattening of the total radio spectrum at higher frequencies . observations of mezger et al . ( @xcite ) do indicate a noticeable flattening of the cas a spectrum in the wavelength range from @xmath158 to 1 mm , with power law index @xmath159 that possibly extends up to the near - ir region if the flux of continuum emission measured at @xmath160 by tuffs et al . ( @xcite ) is synchrotron in origin ( see fig.2 ) . another important effect related to such composite fluxes consists in the possibility to give a new interpretation of the claimed secular flattening of the radio spectra ( dent et al . @xcite ; osullivan & green @xcite ) . namely , the flattening of the radio spectra in time could be easily explained as being due to different rates of decline of the fluxes of the two zones , with the plateau emission dropping more slowly . observations at 5ghz have shown that the cbsr components are fading more rapidly than the total emission , which then requires that the plateau must be fading less rapidly ( see table 6.1 of tuffs @xcite ) . if the plateau indeed were to have a flatter spectral index than the cbsr components , this would result in a flattening of the spectral index of cas a with time . note that previously , in the framework of the single - zone approach , the secular flattening could be explained only by the assumption of a gradual flattening of the spectrum of electrons either at higher energies or in time ( scott & chevalier @xcite , chevalier et al . @xcite , cowsik & sarkar @xcite , ellison & reynolds @xcite ) . another advantage of a spatially inhomogeneous model is its ability to explain the basic characteristics of the observed nonthermal radiation in terms of very efficient acceleration processes operating in cas a , which generally result in a source function of accelerated particles with hard spectral indices @xmath161 ( e.g. efficient fermi acceleration at strong shock fronts ) . even though the spectral index formally corresponding to the mean @xmath162 is @xmath163 , the assumption of @xmath164 much harder than @xmath165 becomes possible if one takes into account the energy dependent propagation and escape of electrons from regions of high concentration on timescales @xmath166 . for example , assuming that the sites of efficient particle acceleration are located in zone 1 , one could expect that the electron energy distribution @xmath167 formed there on timescales @xmath168 would have the power - law index @xmath169 . on the other hand the leakage from zone 1 corresponds to injection of relativistic electrons into surrounding zone 2 with the rate @xmath170 where @xmath171 . with a reasonable source function index @xmath172 and energy - dependent escape with @xmath173 , the values of @xmath174 and @xmath175 needed for interpretation of the cas a radio data could be reproduced . thus , the energy distribution of electrons @xmath176 in the compact regions of efficient acceleration , which results in the enhanced density of relativistic electrons , may be actually much steeper than @xmath177 in the surrounding volume . interestingly , although the electron distribution @xmath177 in the plateau region would show the hard spectral index of acceleration @xmath178 , in principle no acceleration of electrons needs occur there at all . it is important to note , however , that on a qualitative level the same power - law indices @xmath61 and @xmath60 for both zones could be expected also in the case when relativistic electrons were accelerated in zone 2 , while the electron density in cbsr structures would be enhanced due to local compression of the electrons , but not efficient acceleration in zone 1 . to distinguish between these 2 basic scenarios , quantitative modelling is needed . approximating the momentum distribution function of relativistic electrons as an isotropic function @xmath179 of the momentum @xmath180 , the kinetic equation for the energy distribution @xmath181 of relativistic particles ( @xmath182 ) can be written as : @xmath183 \ ; .\end{aligned}\ ] ] here @xmath184 is the fluid velocity , @xmath185 is the energy loss rate of electrons . assuming isotropic diffusion , @xmath186 is the scalar spatial diffusion coefficient , whereas @xmath187 $ ] is a functional standing for various stochastic acceleration terms of electrons . in the two - zone model the source with volume @xmath53 is subdivided into zone 1 composed of @xmath188 compact structures , with volume @xmath189 , spatially immersed in the zone 2 with @xmath190 . the set of equations describing the evolution of the total energy distribution function of particles in these two zones , @xmath191 , can be found by integration of eq.(7 ) over @xmath192 ( @xmath193 ) . integration of the term @xmath194 over @xmath54 results in @xmath195 . the volume integral of the two first terms in the right hand side of eq.(7 ) is reduced to the sum of @xmath188 integrals on the surface @xmath196 separating the zone 1 components from zone 2 : @xmath197\,{\rm d}^3 r & = & \\ \sum_{i=1}^{k}\ ; [ \ , \oint_{s_i } d\,({\bf e}\ , { \rm grad}_{\bf r } f)\ , { \rm d } s & - & \oint_{s_i } ( { \bf e \ , u } ) f \ , { \rm d } s \ , ] \ ; . \nonumber\end{aligned}\ ] ] here * e * is the unit vector perpendicular to the surface element @xmath198 , directed outward from zone 1 to zone 2 . these terms describe the rate of diffusive and convective exchange of particles between the two zones . to proceed further , let us separate those parts of the interface @xmath199 where the scalar product ( @xmath200 ) has positive and negative signs : @xmath201 . the surface @xmath202 corresponds to the regions of plasma outflow from zone 1 into zone 2 , so the integral over @xmath202 of the second term in eq.(8 ) describes the convective escape of electrons , @xmath203 , where @xmath204 @xmath205 is characteristic thickness of , and @xmath206 is the plasma velocity in cbsr components . the first integral in eq.(8 ) over the surface @xmath202 can be simplified if we approximate @xmath207 , where @xmath208 and @xmath209 are the volume averaged distribution functions of relativistic particles in zones 1 and 2 , respectively , and @xmath210 is the characteristic thickness of the transition layer between these zones . taking into account that the volume of a body can be expressed as @xmath211 , where @xmath212 is the thickness and @xmath213 is a factor depending on the shape of the body , we find @xmath214 where @xmath215 has the meaning of characteristic diffusive escape time of relativistic electrons from the regions of higher electron densities : @xmath216 here @xmath217 corresponds to the mean of the spatial diffusion coefficient @xmath218 on the surface @xmath202 . taking further into account that @xmath219 ( @xmath193 ) , the diffusive escape ( _ exchange _ ) term is reduced to the form @xmath220 the integrals over the surface @xmath221 in eq.(8 ) describe the inflow of electrons from zone 2 , with some rate @xmath222 , through the frontal surfaces of zone 1 components . for a mainly adiabatic compression of relativistic particles ( see sect . 5 ) , @xmath222 could be directly connected with the energy distribution @xmath223 of electrons in zone 2 . one could however expect the formation of strong shocks due to fast motion of cbsr components with respect to the surrounding plasma , so that the upstream electrons could be accelerated before entering zone 1 . in that case @xmath222 would correspond to the source function of electrons accelerated on the shock fronts and injected into zone 1 . an essential part of the diffusive shock acceleration process is given by the adiabatic compression term in eq.(7 ) ( third term ) in the shock region near the surface @xmath224 . therefore the volume integral of that term in a thin region adjacent to @xmath224 is to be formally incorporated into the source function @xmath222 . the integral in the main part of the volume @xmath54 describes adiabatic energy losses ( or gain ) of electrons , and can be combined with the volume integral of the forth term in eq.(7 ) as : @xmath225 where @xmath226 is the mean rate of energy losses of electrons , which now includes also the adiabatic energy loss term @xmath227 . finally , the volume integral of the last term in eq.(7 ) describes internal sources of accelerated electrons @xmath228 inside the volume @xmath54 . the final equation then reads : @xmath229 here @xmath230 corresponds to the total injection rate of electrons in the first zone , and @xmath231 is the overall `` diffusive + convective '' escape time of particles @xmath232^{-1}\ ; .\ ] ] because @xmath233 generally increases for decreasing @xmath234 , @xmath231 becomes energy _ independent _ at sufficiently small energies when @xmath235 . the range of actual energy dependence of @xmath231 is limited also at high energies since the diffusive escape time can not be less than the light travel time across the cbsr components @xmath236 . this obvious requirement formally follows from the condition that the characteristic lengthscale @xmath237 for spatial gradients in the distribution function @xmath238 can not be less than the mean electron scattering path @xmath239 , since otherwise the diffusion approximation implied in eq.(7 ) fails . taking into account that @xmath240 , from eq.(11 ) we find that indeed @xmath241 . assuming @xmath242 , a reasonable approximation for the diffusive escape time is then @xmath243 from eq.(11 ) follows that generally the power - law index @xmath244 for diffusive escape time should be smaller than the index @xmath245 of the diffusion coefficient : since faster diffusion of more energetic particles tends to smooth out the gradients of the distribution function @xmath238 more effectively , the characteristic lengthscale @xmath246 should increase with energy . then for a power - law approximation @xmath247 the index @xmath248 , so @xmath249 . this conclusion is important since it allows us to assume a power - law index @xmath250 needed for modification of the electron spectral index , not excluding at the same time efficient shock acceleration of particles in the bohm diffusion limit which corresponds to @xmath251 . integration of eq.(7 ) over the volume of zone 2 results in an equation for @xmath252 : @xmath253 where @xmath254 is the mean energy loss rate , and @xmath255 is the source function of electrons in zone 2 . the second and third terms here result from the same surface integrals on the interface between zones 1 and 2 as in eq.(8 ) , but they have opposite signs since now the direction of the unit vector * e * is reversed . note that formally the volume integrals in eq.(8 ) result in two additional surface integrals describing particle exchange between zone 2 and the exterior on its outer boundary . however , this actually extends the two - zone model to a three - zone one , therefore we neglect here these exchange terms in order to remain in the framework of the two - zone approach . the set of kinetic equations ( 14 ) and ( 17 ) describes the evolution of relativistic electrons in the general case of an expanding inhomogeneous source when the electron energy losses and particle exchange rates between two zones are time - dependent . given these parameters and the source functions @xmath256 , the electron energy distributions can be found numerically by iterations , using analytic solutions to the kinetic equation of electrons in an expanding homogeneous medium ( atoyan & aharonian @xcite ) . to reduce the number of free parameters , we shall consider basically the case of a stationary source because the current energy distributions depend mostly on the injection of the electrons during recent ( largest ) timescales , @xmath257 . only for calculations of the secular decrease of the fluxes we shall use the general results for an expanding source , taking into account adiabatic losses and decreasing average magnetic fields . let us first consider the hypothesis that the compact bright radio components are the main sites of electron acceleration , and neglect possible acceleration in zone 2 , i.e. @xmath258 . the electron injection function in zone 1 is approximated as @xmath259 where we have omitted the suffix ` 1 ' in @xmath260 . for calculations we assume equal radii @xmath261 and magnetic fields @xmath64 for zone 1 components . since the thickness @xmath205 of a spherical object corresponds to its diameter , eq.(9 ) gives @xmath262 . for @xmath263 , and @xmath206 less than the sound speed @xmath264 in the gas with @xmath265 and mean atomic @xmath266 ( corresponding to @xmath267 , cowsik & sarkar @xcite ) , the time @xmath268 is larger than the source age @xmath269 . therefore the precise value of @xmath206 will not significantly affect the results . the minimum escape time of electrons from the zone 1 is taken as @xmath270 . the time @xmath271 in eq.(1 ) defines the ratio of the magnetic fields in two zones needed for an explanation of the flux ratio @xmath272 . indeed an effective modification of the injection spectrum of radio electrons in the gev region implies that at these energies the overall escape time @xmath273 , and that it is significantly smaller than @xmath269 . the leakage of the electrons from @xmath54 will result in @xmath274 electrons accumulated in the surrounding zone 2 . using also eq.(3 ) , we then find @xmath275 thus , for @xmath276 the ratio @xmath277 . in fig.3 we show the energy distributions of electrons @xmath278 calculated the injection spectrum with @xmath279 and @xmath280 , and the escape time parameters @xmath281 and @xmath282 . at gev energies the power law index of electrons in zone 1 is indeed strongly modified , almost reaching the maximum theoretical value @xmath283 . however both at lower and higher energies there is some flattening of @xmath284 . at energies @xmath285 the flattening is connected with the increase of the escape time to @xmath286 , which becomes comparable to the age of cas a , reducing thus the efficiency of the ecape . at high energies @xmath287 is limited by the energy - independent minimum escape time @xmath288 in eq.(16 ) . this effect can be seen in fig.3 for the dot - dashed curve at @xmath289 . note , however , that for the self - consistently calculated spectrum ( solid curve ) of electrons in zone 1 the flattening is noticeable already at energies @xmath290 where @xmath291 becomes comparable to @xmath292 . at these energies the gradients in the spatial distribution of relativistic electrons in the source become small , reducing the net exchange of particles between the zones , which again results in a less efficient modification of @xmath293 . the synchrotron and bremsstrahlung fluxes are presented in fig.4 , and fig . 5 shows the spectral indices at different frequencies . because the synchrotron radiation in zone 1 is steep , it becomes dominant at low frequencies . the compactness of zone 1 components leads to synchrotron self absorption at @xmath13 . at frequencies above 5ghz the plateau emission dominates resulting in a noticeable flattening of the overall spectrum in the far infrared ( fir ) . all these results are in agreement with expectations derived qualitatively in sect . 2 . for the model parameters used in fig.4 the energy content in relativistic electrons in zones 1 and 2 is @xmath294 and @xmath295 , respectively , and the energy in the magnetic fields @xmath296 and @xmath297 . the total energy in relativistic electrons , @xmath298 , can be estimated analytically taking into account that : ( a ) the shape of the overall energy distribution of electrons @xmath299 repeats the power law injection @xmath300 until the break energy @xmath301 due to synchrotron losses ( in zone 2 ) , and ( b ) the total number of gev electrons @xmath302 ( with @xmath303 ) is defined mainly by electrons in zone 2 , @xmath304 , which can be estimated using eq.(3 ) . for @xmath305 we find @xmath306 for @xmath279 and @xmath307 this equation results in @xmath308 , in agreement with the numerical results . in fig.6 we show the fluxes calculated for @xmath309 and different values of the magnetic fields in zones 1 and 2 . for all 3 cases the spectra of synchrotron radiation coincide up to the ir , but at higher frequencies they differ because of radiative break induced by synchrotron losses . note that for interpretation of the x - ray fluxes above 10kev one needs higher values of the exponential cutoff energy for higher magnetic field @xmath64 : @xmath310 for @xmath311 and @xmath312 for @xmath313 . for @xmath314 the calculated synchrotron fluxes remain significantly below the observational data even assuming @xmath315 . however , the synchrotron losses in high magnetic fields effectively limit the maximum energy of electrons to values well below @xmath316 . for example , in the case of diffusive shock acceleration , the acceleration rate can be estimated as @xmath317 , where @xmath318 is the diffusion coefficient and @xmath319 is the speed of cbsr components with respect to the surrounding plasma . equating this rate in the bohm diffusion limit to the synchrotron loss rate , the characteristic maximum energy of electrons can be estimated as : @xmath320 comparison with the values of @xmath45 in fig . 6 shows that diffusive shock acceleration could be consistent with synchrotron origin of the hard x - rays from cas a for @xmath321 . higher values of @xmath64 would require a more efficient acceleration mechanism . but anyway , the value of @xmath322 is a very conservative upper limit of our model for the mean magnetic field in the bright radio structures if the measured hard x - ray flux has synchrotron origin . at the same time , the lower limit for @xmath64 is about 1 mg , which is needed for consistency of the bremsstrahlung with the flux upper limits of egret and osse detectors ( see fig.6 ) . it is interesting that magnetic fields corresponding to the energy equipartition with relativistic electrons in the cbsr components are in the range @xmath323 . another feature to be addressed in this section is the wavelength - dependent rate @xmath324 of the secular decline of radio fluxes . for calculations of this rate we need to specify the rates of the magnetic field decline and the adiabatic losses of electrons . for a spherical source ( shell ) homogeneously expanding with speed @xmath325 , the adiabatic losses @xmath326 , where @xmath327 corresponds to the current expansion time ( ` age ' ) of the outer shell . in the ` standard ' power - law approximation for the magnetic field declining with the shell radius as @xmath328 , the characteristic decline time @xmath329 . for a magnetic field ` frozen ' into the spherically expanding source @xmath330 , but @xmath331 if the fields are effectively created in the shell . the assumption @xmath332 results in @xmath333 . concerning zone 1 , we note that along with the compact radio structures with declining brightness observations show also a large number of brightening components , and the timescales of the flux variations in the individual knots can be as small as few tens of years ( tuffs @xcite , ar96 ) . an estimate of the average time @xmath334 of the magnetic field decrease in cbsr components is thus problematic . therefore we consider @xmath334 as a free model parameter , and neglect adiabatic losses of electrons there ( if any especially as they are supposed to be the sites of effective acceleration ) . calculations of the secular decline rate of the fluxes at different frequencies are presented in fig . the straight heavy dashed line shown corresponds to the decline rate deduced by dent et al . ( @xcite ) and baars et al . ( @xcite ) from comparison of the decline rate @xmath335 measured then at 81mhz by scott et al . ( @xcite ) with the data in the 1 - 10ghz region . later measurements have shown that the decline rate around @xmath336 is at the same level @xmath337 as at @xmath338 , which has placed doubt on the frequency dependence of the decline rate in general ( rees @xcite ; hook et al . however , the recent result of osullivan & green ( @xcite ) that the flux decline rate at 15ghz is about @xmath339 , seems to confirm the effect of slower decline of radio fluxes at high frequencies . figure 7 shows that practically all the observational data can be well explained assuming the average magnetic field decline time in zone 1 of order @xmath340 . note that a significant drop of the secular decline rate predicted at @xmath341 is connected with the gradual decrease of synchrotron self - absorption in zone 1 . consider now the hypothesis that acceleration of electrons takes place only in zone 2 and is negligible in zone 1 . in this case the source function @xmath342 for zone 2 is in the form of eq.(18 ) , while @xmath343 for zone 1 is defined only by the rate @xmath344 of advection of relativistic electrons from the plateau region . the required for zone 1 relativistic electron density enhancement could be then supposed as being due to adiabatic compression of the zone 2 electrons on the front surface @xmath221 of fastly moving cbsr components , and their subsequent accumulation in zone 1 over timescales @xmath345 . neglecting in eq.(7 ) the diffusion term on the surface @xmath221 , in compliance with the assumption of negligible diffusive acceleration there , and retaining only the convection and the adiabatic compression terms , the equation describing the transport of the electrons through @xmath221 reads : @xmath346 where @xmath347 is the momentum distribution function , @xmath348 is the spatial coordinate and @xmath349 is the plasma speed along the @xmath348-axis perpendicular to the compression layer . general solution to this equation reads : @xmath350 $ ] , where @xmath351 is an arbitrary initial function . connecting the electron distribution functions across the layer , the injection rate of relativistic electrons from zone 2 into zone 1 is found : @xmath352 where @xmath353 is the mean compression ratio , and @xmath354 and @xmath206 correspond to the plasma speeds of the plasma in the upstream and downstream regions in the rest frame of the cbsr components . considering possible values of @xmath354 , note that for the mean radial expansion age of radio knots @xmath355 within @xmath356 ( tuffs @xcite ; anderson & rudnick @xcite ) the characteristic speed of the knots may be estimated as @xmath357 . taking into account that surrounding plasma in the shell is moving in the same direction , and as evidenced by a shorter x - ray expansion time scales @xmath358 ( koralesky et al . @xcite ; vink et al . @xcite ) , appears to be overtaking most of the compact radio knots , the value @xmath359 seems a reasonable estimate for the relative speed of radio knots with respect to the surrounding medium . for the bright radio ring @xmath360 ( tuffs @xcite ) , corresponding to @xmath361 , while the speed of freely expanding ejecta upstream of the reverse shock , i.e. at @xmath362 , is about @xmath363 . thus , for the radio ring @xmath364 , and we can take @xmath365 as a reasonable maximum value of @xmath354 for the zone 1 . the compression ratio @xmath366 can be approximated as the ratio of magnetic fields @xmath367 . the energy distributions of the radio electrons can be connected as @xmath368 using the relation @xmath369 . ( 3 ) and ( 24 ) result in @xmath370 where @xmath371 . since the volume filling factor of the compact zone 1 structures is @xmath372 or less , and since for effective modification of the electron spectra by escape one needs @xmath373 , the ratio of the magnetic fields in the two zones should be rather high , @xmath374 . the results of numerical calculations are shown in fig.8 . the total flux seems to fit the data practically equally well as in fig.4 earlier . however , a closer look to these figures reveals significant differences . first of all , in fig.8 the agreement with the radio data is reached due to a much higher contribution of zone 1 to the overall flux . in particular , at 5ghz the partition of the fluxes corresponds to @xmath375 , instead of @xmath376 . secondly , although here we have assumed high @xmath377 , the calculations show that the spectral index of the electrons swept up from zone 2 , @xmath378 , was steepened in zone 1 only to values @xmath379 , well below the ` expected ' value 2.95 . obviously , this is far insufficient for explanation of the observed steep spectral indices of the radio knots . the reason for difficulties arising in the scenario that assumes only passive compression , but not active acceleration of electrons in zone 1 , can be understood from fig.9 where we show the spatial densities @xmath380 of the electrons in zones 1 and 2 . the dot - dashed line shows the spectrum of the electrons which would be formed in the zone 1 if one neglects the term @xmath381 in eq.(14 ) . this spectrum is steep . but when the effect of spatial density gradients between the two zones is taken into account correctly , the situation dramatically changes . the energy dependent propagation can result in a significant steepening of the source spectra only in the regions of higher spatial densities of relativistic particles . otherwise diffusive exchange of particles tends to equalize @xmath291 with @xmath292 resulting in @xmath382 independently of @xmath244 , until the radiative losses become important . thus , for an effective modifications of the initial source spectrum of the radio electrons the condition @xmath383 should be satisfied . the ratio of electron densities expected at 1gev can be estimated from eqs . ( 25 ) and ( 26 ) ( assuming @xmath384 ) as : @xmath385 for the parameters used in fig.8 this equation predicts @xmath386 , in agreement with numerical calculations , which is smaller by a factor of @xmath387 than similar ratio in fig.3 . calculations show that it is not easy to increase this ratio significantly , remaining within the ` adiabatic compression ' scenario . thus , a self - consistent explanation of the observed radio fluxes of cas a is very problematic , unless we assume an effective acceleration of electrons in the cbsr components in order to build up sizeable gradients in the spatial distribution of the radio electrons in the source . the spatially non - uniform source model , even in its simplest two - zone form , allows us to unify into a single picture many observational data on the broad - band nonthermal radiation of cas a. the following is the summary of observational features of cas a which could be explained in the framework of this model . the brighter radio structures tend to be steeper ( ar96 ) to the extent that the energy density of relativistic electrons there would be higher . in principle , even assuming the same hard power law index for the acceleration , e.g @xmath388 , and @xmath389 for the escape of the electrons , it is possible to explain the observed variations of the spectral indices of individual compact radio structures from @xmath390 to @xmath391 . such variations can be connected with two effects . ( a ) the intrinsic index of some knots will be flatter than the maximum possible @xmath392 if the local gradient @xmath393 is not sufficiently high , and if the characteristic escape time @xmath394 of gev electrons from these knots is larger than @xmath395 . the impact of the latter effect can be seen in figs . 10 and 11 , where we show the fluxes and spectral indices calculated in the modified two - zone model approach , subdividing the cbsr structures in two groups and assuming 2 different spatial / temporal scales for larger ( ` ring ' ) and smaller ( ` knot ' ) components in zone 1 . ( b ) the intrinsic spectrum of individual knots could be steep , up to @xmath396 , but the observed spectrum may be flatter , depending on the degree of flux contamination by the flat - spectrum radiation from the diffuse plateau along the line of sight to the knot . this effect may be relevant mainly to those cases when the background subtraction is problematic ( e.g. , for the structures with insufficiently high contrast to the plateau ) . the strong correlation between the spectral index and the projected position of radio knots observed by ar96 could be connected with both effects ( a ) and ( b ) . indeed , the brightness of the diffuse plateau seems to decrease from the radio ring towards the outer edge of the shell more rapidly than one would expect merely on the base of the projection effect in a spherical geometry ( see fig.7 from arlpb ) . this implies that the density @xmath397 of radio electrons could be higher at distances closer to the ring . then the steepest knots would be found predominantly closer to the edge of the plateau , because ( i ) the local ratio @xmath393 might be higher far from the reverse shock , and ( ii ) contribution of the flat spectrum plateau emission to the steep spectrum flux of the knots is smaller in directions to the projected periphery of the shell . the cutoff of the radio spectrum below 20mhz is explained by the synchrotron self - absorption of the flux of zone 1 . this interpretation is possible basically because the model predicts that the intrinsic fluxes of the compact bright structures are much steeper than the spectra in the diffuse plateau . then at low frequencies the flux of cbsr components should dominate the total radio emission . therefore synchrotron self - absorption of only this flux , which becomes possible due to high density of particles ( and fields ) in those compact structures , is sufficient to comply with the data . it is worth noticing that due to variations of the physical parameters in the individual cbsr components contributing to zone 1 flux , one could generally expect that the synchrotron absorption of that flux would be in reality smoother than it is shown in fig . 4 where we have assumed the same parameters for all counterparts of zone 1 . the position of the characteristic cutoff frequency for an individual component @xmath398 is found from the condition @xmath399 where @xmath400 is its opacity . from eqs . ( 3 ) and ( 4 ) it follows that @xmath401 , with @xmath402 where @xmath403 is the knot flux ( luminosity ) at some fixed frequency ( say 1ghz ) . because of the very strong dependence of @xmath404 on @xmath405 , a significant broadening of the synchrotron turnover frequency for the ensemble of cbsr components ( say by a factor 2 ) would be expected only if the _ luminosity - weighted _ dispersion @xmath406 of the parameter @xmath407 is very large ( @xmath408 ) . the fact that the observed turnover is sharp may then imply a quite reasonable possibility that the distribution of the parameters @xmath409 is not far from ` gaussian noise ' . in that case one would normally expect @xmath410 ( or perhaps even less , given the expected correlation of @xmath403 and @xmath411 ) , so the synchrotron turnover position would be effectively dispersed , or broadened , only by a factor roughly @xmath412 , i.e. about @xmath413 around the mean position . note that the overall flux in fig.10 , where 2 significantly different sizes for the large and small components are assumed , show practically the same sharp cutoff below 20mhz as in fig.4 . observations at low frequencies by kpde , at a resolution of about @xmath414 , show that the spectral index of cas a between 333mhz and 1.38ghz is approximately constant , @xmath415 , at angular radii @xmath416 , but that it is quickly increasing to @xmath417 as @xmath418 . in principle , this effect might be connected with the steep spectrum radio knots which have been detected by ar96 but can not be resolved in the low frequency maps of kpde . because of the rapid decline of the plateau brightness at large angular distances from the ring , the cbsr components ( the radio knots and bow shocks ) increasingly contribute to the ` diffuse ' overal flux closer to the periphery . we do not exclude that perhaps in the framework of a more spatially structured model ( that would allow large scale inhomogeneities in the plateau region itself ) one could expect also some steepening of the intrinsic diffuse flux . however , given much slower and less efficient spectral modifications for the large scale inhomogeneities , the propagation effects alone would be able to explain only rather moderate steepening of the intrinsic plateau emission . if the acceleration is efficient so that the primary spectrum is really hard , the steepening to values @xmath419 seems to require a dominating contribution of the compact structures in the overall flux detected by kpde . the two - zone model suggests a possible interpretation for the ` discrepancy ' between the spectral index @xmath420 found by kpde at the radius of the bright radio ring , and observations of wds90 who found @xmath421 if _ only _ the bright components at approximately the same radial distances are considered . both results might be understood if we take into account that the spectral index found by kpde corresponds to the flux of two zones integrated along the geometrical circles of different fixed radii , and hence would be more affected by the flat plateau flux than the result found in wds90 ( compare the heavy and thin solid lines in fig.11 ) . the flattening of the radiation spectrum observed at millimeter wavelengths ( mezger et al . @xcite ) is a natural consequence of the ` flat + steep ' representation of the total flux in the two - component approach , with approximately equal contributions from these components around 5ghz . the two - zone model also predicts that the flux measured by tuffs et al . ( @xcite ) around @xmath422 should be predominantly synchrotron in origin . the unusual dependence of the secular decline of radio fluxes on frequency , which is at a constant level at @xmath423 but seems dropping at higher frequencies , is explained by varying contributions of the flat and steep spectrum components at different frequencies . this interpretation suggests that on average the net flux of the bright compact components in cas a drops faster than the diffuse plateau emission , in agreement with observations of tuffs ( @xcite ) . the synchrotron origin of the x - rays above 10kev can be explained if electrons in the compact cbsr components are accelerated to energies of few tens of tev . these energies can be reached , e.g. , by diffusive shock acceleration in the bohm limit , which seems seems a plausible mechanism for production of relativistic electrons at the reverse shock presumably connected with the bright radio ` ring ' . in principle , for the radio knots as well the diffusive acceleration at the bow shocks could result in the high electron energies needed for x - ray emission . perhaps , another possibility for acceleration of electrons in the radio knots could be connected with reconnection of the magnetic field lines strongly amplified , as shown by jones et al . ( @xcite ) , in thin turbulent layers behind the bow shocks of dense gas ` bullets ' at the stage of their fast disruption by rayleigh - taylor and kelvin - helmholtz instabilities . hard spectral indices for accelerated particles weaken the constraints on the energetics of relativistic electrons in cas a. for example , in the case of @xmath424 , and @xmath425 and @xmath426 used in fig.4 , the total energy @xmath427 . the energy @xmath428 needed in the case of @xmath429 for a uniform shell with the same magnetic field @xmath430 is by a factor of 10 larger . the radio fluxes of the prototype young snr cas a show a large variety of spectral and temporal features which are very difficult to incorporate into one self - consistent picture , if one remains in the framework of a single - zone ( i.e. uniform ) model approach and thus has to attribute the spectral indices deduced from radio observations to the source spectra of accelerated electrons , since radiation losses can not modify the energy distribution of radio electrons during the lifetime of cas a. the interpretation of the radio data essentially changes if we take into account that energy dependent propagation of relativistic particles is able to modify their energy distribution in a spatially inhomogeneous source on timescales much shorter than the radiative loss time . the efficiency and extent of these modifications depend on the strength of the gradients in the spatial distribution of the particles in the source . the simplest spatially inhomogeneous model is the one where the radio source is subdivided into two zones with different energy distributions of relativistic electrons and different magnetic fields . the basic assumption of the model is that the number density of radio electrons in the compact bright radio components ( predominantly included in zone 1 ) is much higher than in the surrounding diffuse shell between the reverse shock and the blast wave ( zone 2 ) . the fulfillment of this assumption is contingent on an efficient acceleration of electrons in these radio bright components . in principle , acceleration of electrons in the diffuse shell is not needed at all . note , however , that the model does not exclude a contribution from such acceleration provided that it is not so powerful as to noticeably reduce the gradients in the spatial density of electrons established by acceleration in the cbsr components , which would otherwise diminish the efficiency of spectral modifications in zone 1 . the possibility for exchange of particles between these zones on an energy dependent timescale @xmath431 , with @xmath432 , allows us to suggest a single hard power law injection spectrum of electrons with an index @xmath433 , implying an efficient acceleration process . then , depending on the relativistic electron density contrast and physical parameters of individual component @xmath398 , the leakage of electrons from those cbsr components can result in the steepening of the injection spectrum up to @xmath434 , in agreement with the maximal spectral indices of the knots @xmath435 . the energy distribution of radio electrons @xmath177 in the extended plateau will show the hard power - law index of injection , @xmath436 . as a general remark we note that energy - dependent propagation in any non - uniform medium would always tend to flatten the energy distribution of particles in the regions with _ lower _ density , i.e. which are typically more extended . the two - component decomposition of the overall flux , with @xmath437 at 5ghz , predicts that the sites of current or very recent acceleration will become more pronounced in the high resolution ( few arcsec ) maps at lower frequencies , where the contribution of the diffuse plateau to the overall flux will decrease . this should result in a significant increase of the brightness contrast for the remaining compact structures ( which belong to zone 1 ) , reaching its maximum around 40mhz ( i.e. at the minimum frequencies not affected by synchrotron self - absorption ) . at frequencies @xmath438 the bright compact structures will quickly disappear , but the plateau emission , in the form of a weak diffuse shell , may become dominant again . this shell would probably have an apparent radius less than @xmath439 if the emissivity of the shell is indeed strongly decreasing towards the blast wave , implying very low brightness at large radii . correlations between the spectral index and the geometrical thickness of the radio knots of similar brightness could be expected on the high resolution maps of cas a taken at 1.4 - 5ghz . the radio knots with a smaller thickness , but the same brightness , would tend to be steeper since the escape times in those knots could be smaller . note however that the escape time is not the only parameter affecting the efficiency of spectral modifications . in particular , the ` steepness - compactness ' correlation could be different for knots in the fading stage ( which are ` older ' , with declining injection of new particles ) and in the brightening stage . besides , such a correlation could be significantly affected by different magnetic fields in the knots . nevertheless , the search for correlations between spectral index and compactness of the knots seems worthwhile . at frequencies above 10ghz the brightness contrast will decrease , and the cbsr structures may appear less pronounced . the total emission will be dominated by the diffuse flat - spectrum plateau , and the spectral index may decrease to @xmath440 ( see fig.11 ) . the structure of zone 2 , in particular the outer edge of the shell , will be better discernible . an important prediction concerns the character of the long term evolution of the radio pattern of cas a. one can expect that during the next @xmath441 , when the total emission will become dominated by the diffuse plateau and the compact radio structures may become less common , the remnant will have a much more uniform radio shell with a spectral index @xmath442 , and cas a may become similar to tycho s snr ( e.g. see klein et al . interestingly , the overall efficiency of electron acceleration at that time might be even lower than at present , while the radio spectrum will be significantly harder reflecting the spectrum of the acceleration which is taking place presently . more generally , the energy dependent propagation of radio electrons in a spatially inhomogeneous medium can explain the trend ( see e.g. green @xcite ; jones et al . @xcite ) that young clumpy shell - type snrs often exhibit radio spectra that are significantly steeper than those of older ones showing a typical spectral index @xmath443 . at x - ray frequencies the appearance of cas a may be very different in the soft x - ray and hard x - ray domains . at photon energies below @xmath444 the radiation is dominated by thermal emission of the gas in the extended region from the reverse shock up to the blast wave . apart from the thermal diffuse emission we could expect a noticeable contribution of the nonthermal x - rays from compact radio structures . this radiation could be more easily distinguished from the thermal emission in the case of compact bright radio knots at the periphery of cas a , if the x - ray spectra would appear relatively flat in the 1 - 5kev region ( see figs . 4 and 11 ) and the line emission would be deficient . these observations will be possible with the high spectral and angular resolutions of xmm and astro - e . in hard x - rays above 10kev the appearance of cas a may significantly change . in the case of a nonthermal origin of this radiation , all acceleration sites of the highest energy electrons may become clearly visible . note , however , that the angular resolution of the detectors needed to reveal significant spatial changes at these energies must be better than @xmath445 . the observation of such hard nonthermal x - ray fluxes from the compact radio knots and radio ring would impose an upper limit on the magnetic field in these structures @xmath446 , with the probable value expected in the range @xmath447 . the model predictions for magnetic fields in the diffuse shell are @xmath448 . in summary , the steepness of radio spectra of bright and compact structures in clumpy radio sources is not an indication of inefficient acceleration , but rather a natural consequence of very efficient acceleration which builds up high spatial gradients of relativistic electrons in those sources increasing the efficiency of spectral modifications due to their energy dependent escape . the authors thank larry rudnick for very useful discussions , and the anonymous referee for very helpful comments . rjt thanks nrao for hospitality during his visits to the vla in 1984 and 1985 and to rick perley , steve gull and martin brown for support in the observations and data reduction which led to fig.1 . the work of ama was supported through the verbundforschung astronomie / astrophysik of the german bmbf under the grant no . 05 - 2hd66a(7 ) .

we have carried out a deep narrow - band imaging survey of six fields with heavy - element quasar absorption lines , using the goddard fabry - perot ( fp ) system at the apache point observatory ( apo ) 3.5-meter telescope . the aim of these observations was to search for redshifted ly- emission from the galaxies underlying the absorbers at and their companion galaxies . the 3 sensitivity levels ranged between and erg s in observed - frame ly- flux . no significant ly- emitters were detected at level . the absence of significant ly- emission implies limits on the star formation rate ( sfr ) of 0.9 - 2.7 yr per 2-pixel x 2-pixel region , if no dust attenuation is assumed . we compare our results with those from other emission - line studies of absorber fields and with predictions for global average sfr based on the models of cosmic chemical evolution . our limits are among the tightest existing constraints on ly- emission from galaxies in absorber fields , but are consistent with many other studies . in the absence of dust attenuation , these studies suggest that sfrs in a large fraction of objects in the absorber fields may lie below the global mean sfr . however , it is possible that dust attenuation is responsible for the low emission line fluxes in some objects . it is also possible that the star - forming regions are compact and at smaller angular separations from the quasar than the width of our point spread function and , get lost in the quasar emission . we outline future observations that could help to distinguish between the various possibilities .

astro-ph0509182

a great deal of progress has been made in the last decade in the studies of distant galaxies . the average star formation history of the universe has also been estimated from emission properties of galaxies detected in deep imaging and redshift surveys such as the canada - france redshift survey and the hubble deep field ( e.g. , lilly et al . 1996 ; madau et al . 1996 , 1998 ) . a large population of bright galaxies with high star formation rates ( sfrs ) has been uncovered by means of the lyman - break technique ( e.g. , steidel et al . . however , such flux - limited studies may not be adequate for investigating the evolution of normal galaxies , since such studies are biased toward the brighter or the more actively star forming galaxies . in principle , a less biased sampling of distant galaxies may be obtained by means of the absorption lines they superpose in the radiation from background quasars . the presence of heavy - element absorption lines in quasar spectra preselects regions that have undergone some degree of star formation . there is nothing special about the lines of sight to the absorbers except for the convenient placement of background quasars , assuming that most of the absorbers are intergalactic material and not ejected material near the qso at high , special relativistic velocities . thus , it is very likely that these absorbers are surrounded by other galactic or protogalactic objects at the same redshifts . thus in principle , deep images of quasar absorber fields may reveal star - forming regions at high redshift . since a well - known signature of star forming galaxies is strong ly-@xmath0 or h-@xmath0 emission , a natural strategy to look for the absorber galaxies or their companions is to search for the redshifted emission lines . several previous studies have attempted to detect continuum and line emission from galaxies underlying quasar absorbers . at low redshifts , [ o ii ] or [ o iii ] emission has been detected in mg ii systems with narrow - band imaging ( e.g. , yanny 1990 ; yanny , york , & williams 1990 ; bergeron & boisse 1991 ; yanny & york 1992 ) . searches for low-@xmath9 damped lyman - alpha absorbers ( dlas ) have often imaged , and sometimes spectroscopically confirmed , galaxies with various morphologies ( spirals , irregulars , low surface brightness galaxies etc.e.g . , steidel et al . 1994 , 1995 ; lebrun et al . 1997 ; bowen et al . 2001 ; cohen 2001 ; turnshek et al . 2001 ) . at high redshifts , however , it has been much more difficult to identify and study in detail the galaxies responsible for quasar absorption systems . there have been a few detections of ly-@xmath0 emission in quasar absorber fields ( e.g. lowenthal et al . 1991 ; francis et al . 1996 ; roche , lowenthal , & woodgate 2000 ) . however , most other attempts to detect ly-@xmath0 emission from high-@xmath9 intervening ( @xmath10 ) dlas have produced either non - detections or weak detections ( e.g. smith et al . 1989 ; hunstead et al . 1990 ; lowenthal et al . 1995 ; djorgovski et al . most of the few confirmed ly-@xmath0 detections for high - redshift dlas have been for absorbers with @xmath11 , which may differ from the cosmologically more interesting general population of dlas with @xmath10 ( e.g. warren & moller 1996 ; moller & warren 1998 ; fynbo et al . 1999 ) . most attempts to detect dlas in h-@xmath0 have either yielded non - detections or detected objects separated by large angular distances from the quasars , rather than objects close to quasar sightlines ( see , e.g. , teplitz , malkan , & mclean 1998 ; bechtold et al . 1998 ; mannucci et al . 1998 , bunker et al . these objects , while unlikely to be the absorbing galaxies themselves , are still interesting because they are likely to be companion galaxies in the same cluster as the absorber . a summary of previous attempts to detect high-@xmath9 dlas in emission is given by kulkarni et al . ( 2000 ) . with the goal of obtaining a large sample of sfr estimates in the fields of high - redshift absorbers , we have started a lyman-@xmath0 imaging survey for absorbers at @xmath12 using a fabry - perot imager as a tunable narrow - band filter . here we report our results for six fields with known heavy - element absorption systems . sample selection , observations , and data reduction are described in section 2 . the reduced images and photometry are presented in section 3 . section 4 presents a comparison of our results with previous studies of the same fields and with results of other ly-@xmath0 emitter studies . section 4 also compares the sfr constraints from our study with the estimates from other emission - line searches for quasar absorber galaxies . a comparison with predictions based on global sfr models is also presented . the blue and vis - broad etalons in the goddard space flight center ( gsfc ) fabry perot ( fp ) imaging system have optimum sensitivity and resolution in the wavelength range @xmath13 . we therefore restricted our search to the redshift range @xmath14 . we searched the york et al . ( 1991 ) catalog of heavy - element quasar absorbers for absorbers with ( i ) @xmath15 ( ii ) @xmath16 to avoid absorbers possibly associated with the quasars , and ( iii ) with well - detected mixed ionization species ( si ii , al ii or o i in addition to c iv and/or si iv ) . six such objects were finally observed . two of these six systems are dlas , and one is a sub - dla . in addition to these six systems , we also observed the well - studied field of the radio galaxy 53w002 containing ly-@xmath0 emitters at @xmath17 ( pascarelle et al . 1996a , 1996b ; keel et al . 1999 ) , as a calibration object . table 1 lists the general properties of the objects observed . the fp instrument used for our observations has a choice of six queensgate 50 mm diameter piezo - electrically driven , capacitance - stabilized etalons which can be tuned to any wavelength in the range 4000 - 10000 , and with resolution from 2.5 to 28 fwhm depending on the etalon and the wavelength . the system throughput including the ccd is 20% in the red and 5% in the blue . the instrument is used at the nasmyth f/10.3 focus of the apo 3.5 m telescope . behind the telescope focus a field lens and collimator lens collimate the light through the etalon and an order - sorting blocking filter , and a camera lens refocuses onto the ccd . the camera has a stis technology site 2048x2048 ccd with 21 @xmath18 m pixels , 5 e@xmath19 rms readout noise , mpp ( inverted operation ) , minichannels , and a very high efficiency down to 3400 . the quantum efficiency is 82% at 4000 , 90% at 7000 and 53% at 9000 . this provides a large field of view , low noise , smooth bias levels , high charge transfer efficiency , and high sensitivity with uv , visible and ir etalons . the instrument can be controlled from the observatory control room via the apache point fiber optics communication system , or remotely over the internet , using command line control and a vnc gui monitoring system . the wavelength transmitted by an fp system is a function of the field angle . @xmath20 $ ] , where @xmath21 is the camera lens focal length , 200 mm , and @xmath22 is the distance on the detector from the optical axis . at the apo f/10.3 nasmyth focus , @xmath23 \biggr\}$ ] , with @xmath24 in arcseconds on the sky . for example , for @xmath25 = 4100 , @xmath26 , @xmath27 = 11.1 . the effective monochromatic field of view depends on the tolerable fraction of the spectral width . table 2 lists the details of our observations . the observations were carried out primarily with the blue etalon during 9 runs between october 2000 and may 2004 . the seeing varied between 1.0@xmath28 and 1.8@xmath28 fwhm . conditions were not photometric for many of the nights . the narrow - band exposures were obtained setting the fp to the air wavelengths of the expected ly-@xmath0 emission lines at the absorber redshifts . in some cases , additional narrow - band exposures were also obtained with the vis - broad etalon at wavelength settings offset by 400 km s@xmath5 redward of the absorber redshifts . the wavelength calibration was checked throughout using scans of lines from an ar lamp . a temperature control program was run throughout to automatically keep the fp wavelength setting fixed by correcting for effects of temperature variations every five minutes . the fwhms for the different fields were 5.6 - 14.7 , as measured from the width of the calibration lamp lines . a @xmath29 binning was used , resulting in a plate scale of 0.366 @xmath28 per binned pixel . the full circular field of view corresponds to @xmath30 in diameter . of course , given the variation in the central wavelength as a function of angle from the axis , the central @xmath31 diameter region is sampled within @xmath32 km s@xmath5 of the central wavelength expected at the absorber redshift at 4100 . total narrow - band integration times were 320 - 600 minutes per field , split into 8 - 15 exposures of 40 minutes each . in addition , b - band exposures were obtained for all fields to sample the uv continuum near the ly-@xmath0 line . a pentagonal dither pattern was used to aid with flat - fielding and cosmic ray removal . twilight sky flats were obtained for each wavelength setting . standard star exposures were obtained for hz44 . additionally , images of the 53w002 field were also used for photometric calibration . data reduction was carried out using standard iraf tasks . the biases on each night were combined to get an average bias frame . the darks , flats and the object frames were corrected for the overscan region using the ccdproc task . all the images were trimmed to a size of 561 x 561 pixels or @xmath33 , which contains the full field of view of the fabry perot . the darks , the flats and the object frames on each night were bias - corrected with the average bias corresponding to that night . the flat frames and the object frames were then dark - corrected with the combined dark for each night . the flat frames were combined using the flatcombine task to produce average flats corresponding to different wavelengths and filter settings . in some cases , twilight or dome flats were not available for a particular wavelength and filter setting . in such cases , dark - corrected object frames were combined ( with all objects removed ) to create an average sky flat that was later used for flatfielding . object frames for a particular wavelength and filter settings were flat fielded using the average flats in the same setting . bias , dark and flat frames from other nights in the same season were used in a very few cases where they were not available on the same night . the iraf task cosmicrays was used to remove cosmic rays from each object frame . a bad pixel mask was created for each image and the bad pixels were removed using the iraf tasks ccdproc or fixpix . to account for the varying degrees of extinction encountered over different nights , we carried out an empirical correction . the flux counts for unsaturated bright stars were measured in each image using the phot task in iraf . the counts were corrected for images with different exposure times . for each star , the maximum value of counts @xmath34 from different images was taken as an indicator of the unextinguished flux level of the @xmath35th star . the flux ratio for the @xmath35th star in the @xmath36th image @xmath37 was then calculated for the best four or five stars in all images to estimate the extinction factors . the extinction factors thus derived for these four or five stars in each given image were then averaged to get an indication of the extinction for that image . finally , each image was multiplied by this average empirical extinction factor to get an extinction corrected image . all extinction - corrected images for each field and filter / wavelength setting were registered and shifted so as to match the coordinates of a reference star in all frames . this was done using the imexam , lintran and imshift tasks . some images were rotated using the geomap and gregister tasks to match the coordinates of all stars in different images , to correct for a small rotational offset between some images taken at the same pointing in different observing seasons . finally , the individual exposures for a given object and given wavelength and filter setting were combined using the iraf task imcombine to get the final narrow - band and broad - band images . figures 1(a)-6(a ) show @xmath38 sections of the reduced broad - band ( b ) images of the quasar absorber fields of q0216 + 080 , q0636 + 680 , q0956 + 123 , q1209 + 093 , q1442 + 101 , and q2233 + 131 , respectively . the corresponding panels ( b ) show the narrow - band images of the same fields obtained with the blue etalon . the stripes at the borders of some images are an artifact of the coadding of the dithered images . the quasar is absent or considerably dim in the narrow - band images for q0216 + 080 , q1209 + 093 , and q2233 + 131 since the absorbers being studied in these systems are dlas or sub - dlas . figure 7 shows the calibration field 53w002 . the expected b - band continuum must be subtracted from the observed narrow - band fluxes in order to determine if a statistically significant redshifted ly-@xmath0 excess exists for any object seen in both bands . we estimated the continuum in the narrow - band images by scaling the b images using the relative photometric calibrations of the two images . to do this , we subtracted a scaled b band image from the narrow - band image , aligning the images spatially , and adjusting the scaling factor so as to minimize variance in the central portion of the subtracted image . furthermore , to minimize the effect of different seeing in the broad - band and narrrow - band images , the broad and narrow - band images were convolved with a gaussian of the same fwhm before subtraction in idp3 . the subtraction was carried out interactively , using the interactive data language ( idl , version 6.1 ) program image display paradigm-3 ( idp-3 ) version 2.7 , written by d. lytle and e. stobie ( see , e.g. , lytle et al . 1999 ) figs . 1(c)-7(c ) show the resultant continuum - subtracted images for the six quasar absorber fields and the calibration field of 53w002 . finally , figs . 1(d)-7(d ) show @xmath39 close - up views of the continuum - subtracted images . all of the objects in the narrow - band images disappeared almost completely after subtraction of the continuum . the slight residuals left at the positions of some objects arise because of the difficulty in matching the point spread functions ( psfs ) perfectly in the broad and narrow - band images . no significant ly-@xmath0 emission at @xmath7 level was detected from any object in any of the quasar absorber fields . the few vis - b images obtained for three of the fields ( q0216 + 080 , q0636 + 680 , q2233 + 131 ) were also analyzed in a similar manner , and showed no ly-@xmath0 detections either . no significant objects were found even if the images were smoothed using gaussian filters of about 2.5 or 5 pixels fwhm . finally , we also compared our images with broad - band images from the sloan digital sky survey ( sdss ) where available . a few very faint features are seen in our images for q0216 + 080 , but these features are indistinguishable from the noise . in q0636 + 680 and q0956 + 123 , some residual flux is seen close to the quasar after continuum subtraction . however , these regions close to the quasars are not well sampled at our modest angular resolution . thus , it is not clear whether the residuals seen near q0636 + 680 and q0956 + 123 are significant features or artifacts , without obtaining higher resolution images of these fields . in the case of q0956 + 123 , the feature apparent near the left upper corner of fig . 3(d ) is an artifact arising from a residual multipixel cosmic ray event in a single narrow - band image . in the field of q2233 + 131 , three bright galaxies are seen : 17.4@xmath28 east and 1.7@xmath28 south of the quasar ; 28.2@xmath28 east and 51.7@xmath28 south of the quasar ; and 35.2@xmath28 east , 30.2@xmath28 north of the quasar ( fig . all of these galaxies are seen in the sdss . however , all of these galaxies are far too bright to be at @xmath40 ( @xmath41 respectively from sdss ) . it is more likely that the excess emission seen in these objects is [ o ii ] @xmath42 emission from interloper galaxies at @xmath43 . we plan to obtain spectra of these galaxies in the near future to determine their redshifts . to estimate the limits on the ly-@xmath0 fluxes for the absorbers in our fields , we used our calibration observations of the field of the radio galaxy 53w002 . this field is known to have ly-@xmath0 emitters at @xmath44 ( pascarelle et al . 1996a , 1996b ; keel et al . 1999 ) . for calibration purposes , we used the brightest objects seen in our continuum - subtracted narrow - band images of this field , i.e. , objects 3 and 1 of keel et al . ( 1999 ; table 2 ) , labeled o1 and o2 in the bottom panel of figure 7 . aperture photometry of these two objects was carried out using the iraf task phot with aperture sizes @xmath45 times the seeing fwhm . using the ly-@xmath0 fluxes for these from table 2 of keel et al . ( 1999 ) and the counts measured for those objects in our images , we estimated the photometric calibration of our images . to correct for sensitivity differences at the different wavelengths used for our quasar absorbers and the calibration field 53w002 , we used our observations of standard star hz44 . using the photometry of this star in images obtained at the different wavelength settings , and adopting the absolute spectral distribution of this star from oke ( 1990 ) , we estimated the relative sensitivity differences of the fp at the different wavelengths . we next estimated the noise level in the continuum - subtracted blue - etalon images for each field . to do this , we considered a @xmath46 region centered on the quasar and determined the noise profile over this region using idp-3 . while doing this , we disregarded a circular region centered on the quasar with a radius of roughly twice the seeing fwhm to avoid effects of imperfect seeing matches between the broad and narrow band images . we then measured the mean noise level per pixel in concentric annuli of radii ranging from 2 times the seeing to 5 @xmath28 , and finally corrected for the sensitivity differences at the various wavelengths . the noise level per pixel was found to be nearly constant everywhere ( within a few @xmath47 ) in this region . the mean value of the noise per pixel was used to estimate the noise level per 2 pixel x 2 pixel region , assuming poisson statistics . based on this , we estimated the 3 @xmath2 observed - frame ly-@xmath0 point source flux sensitivity reached in our images . if dust attenuation is assumed to be small , these ly-@xmath0 flux limits can be converted to limits on the star formation rates ( sfrs ) . these point - source sfr limits implied by the non - detections of ly-@xmath0 emission in our quasar absorber fields are listed in table 3 and are in the range of 0.9 - 2.7 m@xmath48 yr@xmath5 . here , we have used the prescription of kennicutt ( 1998 ) for converting h-@xmath0 luminosity @xmath49 to the sfr , i.e. , @xmath50 and assumed a ratio @xmath51 for case - b recombination . we also used our continuum - subtracted images to estimate the flux limits for a diffuse foreground absorbing galaxy . to do this , we used the mean value of the noise per pixel to estimate the flux limit over the entire 10 @xmath52 region . this flux limit was used to calculate the limiting ly-@xmath0 luminosity and hence the limiting sfr in the @xmath53 region . at @xmath54 , this region corresponds to a size of @xmath55 kpc@xmath56 centered on the quasar , for the comsology adopted here . the diffuse - source flux limits are in the range of 2.6 - 7.4 @xmath57 erg s@xmath5 @xmath6 summed over the @xmath53 region . the corresponding summed sfr limits are in the range of 12.3 - 36.4 m@xmath48 yr@xmath5 over the entire @xmath53 region . for most of our targets , no other imaging information exists in any waveband on a scale comparable to our fields of view . for q0216 + 080 and q2233 + 131 , hst nicmos h - band images are available ( warren et al . the nicmos images of q0216 + 080 show two objects with @xmath58 and 24.34 , located 1.4@xmath28 and 3.8@xmath28 away from the quasar , at position angles of -104.8 and 97.8 degrees east of north , respectively . however , the closer object is likely to be an artifact since it occurs on a diffraction spike . for q2233 + 131 , the nicmos h - band image revealed two objects with @xmath59 and 25.12 , located 2.8@xmath28 and 3.3@xmath28 away from the quasar , at position angles of 158.6 and 68.3 degrees east of north , respectively . of these , the former object has been identified with a lyman limit system at @xmath60 ( djorgovski et al . none of these objects are seen in our b - band or nb images . this could be because of the difference in angular resolutions of our study and the nicmos study . in either case , no information on star - forming emission - line objects is available from these broad - band images . the only additional emission - line constraints available are for q0216 + 080 and q2233 + 131 . the narrow - band imaging study of deharveng et al . ( 1990 ) found no ly-@xmath0 emitters in the q0216 + 080 field , placing a 3 @xmath2 upper limit of @xmath61 erg s@xmath5 @xmath6 on the ly-@xmath0 emission flux . our observations of this field have provided a ly-@xmath0 flux limit @xmath62 times tighter than that of deharveng et al . ( 1990 ) . for q2233 + 131 , our limit agrees closely ( within @xmath63 ) with that of lowenthal et al . ( 1995 ) . [ [ the - space - density - of - ly - alpha - emitters - comparison - with - other - studies ] ] the space density of ly-@xmath0 emitters : comparison with other studies ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ to understand whether the non - detections of ly-@xmath0 emitters ( laes ) in our fields are consistent with other surveys for ly-@xmath0 emitters , we now examine results from some recent lae searches . stiavelli et al . ( 2001 ) detected 58 lae candidates over a field of 1200 arcmin@xmath56 at @xmath64 . based on this , they deduced a completeness - corrected space density of 0.07 laes per arcmin@xmath56 with ly-@xmath0 fluxes above @xmath65 erg @xmath6 s@xmath5 . palunas et al . ( 2004 ) detected 37 absorbers in a 46 @xmath66 46 field of view . based on this , they derived the space density of laes to be 0.019 arcmin@xmath67 in a redshift interval @xmath68 at @xmath69 . these ( stiavelli et al . 2001 , palunas et al . 2004 ) studies imply 0.49 and 0.42 laes per arcmin@xmath56 per unit redshift , respectively . at @xmath70 , lae searches by rhoads et al . ( 2000 ) and malhotra & rhoads ( 2002 ) have found space densities of 4000 deg@xmath67 per unit redshift , i.e. 1.1 lae per arcmin@xmath56 per unit redshift . these other lae searches cover much wider fields of view , and larger redshift ranges @xmath71 ( wider filter bandwidths ) than our study . adopting a mean space density of 0.45 laes per arcmin@xmath56 per unit redshift at @xmath72 from stiavelli et al . ( 2001 ) and palunas et al . ( 2004 ) , and using @xmath73 as a typical redshift range covered by our study , one would expect @xmath74 absorber in our fields with effective monochromatic coverage of @xmath75 , if our fields were similar to the other lae fields . on the other hand , the lae searches of rhoads et al . ( 2000 ) , stiavelli et al . ( 2001 ) , and malhotra & rhoads ( 2002 ) are not selected by the presence of a quasar absorption system . the presence of a well - established heavy - element quasar absorber with mixed ionization implies the existence of a region that has had some star formation . at @xmath69 , our complete fields of view of about 3.4 diameter cover about 2.2 mpc@xmath56 around the absorbing sightline ( assuming @xmath76 , @xmath77 , and @xmath78 km s@xmath5 mpc@xmath5 ) . given the gradient in the central wavelength , the roughly monochromatic ( within @xmath32 km s@xmath5 of the absorber ) fields of view of about 2.1 diameter cover about 0.8 mpc@xmath56 around the quasar sightline . since the absorbing galaxy certainly must lie within this region , one might expect a higher lae density in our fields than that in a blind field . francis et al . ( 1996 , 2004 ) and palunas et al . ( 2004 ) did indeed find a number of laes in wide - field surveys surrounding a field with 3 sub - dlas . however , this region appears to be a filament with higher density than a typical field region . furthermore , the laes in this filament are separated from 2 of the 3 absorber sightlines by more than 2 in radius ( see fig . 4 of palunas et al . such objects would thus not be detectable in a field of view such as ours . thus , overall , the lack of laes in our observations are not inconsistent with the space density of laes found by other studies . we note , however , that detailed comparisons would require a full consideration of the luminosity function of laes since our observations reach fainter flux limits than many of the other lae searches . [ [ the - star - formation - rates - of - quasar - absorbers - and - their - companions ] ] the star formation rates of quasar absorbers and their companions ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ fig . 8 plots our sfr limits in the fields of quasar absorbers together with the results of other searches for ly-@xmath0 , h-@xmath0 , h-@xmath79 , [ o ii ] , and [ o iii ] emission in quasar absorber fields . the filled red triangles show our apo ly-@xmath0 limits , while the unfilled black triangles at @xmath80 show our limits from previous h-@xmath0 imaging with hst / nicmos ( kulkarni et al . 2000 , 2001 ) . other data are from yanny et al . ( 1987 ) ; yanny , york , & williams ( 1990 ) ; yanny ( 1990 ) ; hunstead et al . ( 1990 ) ; deharveng et al . ( 1990 , 1995 ) ; giavalisco et al . ( 1994 ) ; lowenthal et al . ( 1995 ) ; francis et al . ( 1996 ) ; warren & mller ( 1996 ) ; bergvall et al . ( 1997 ) ; mannucci et al . ( 1998 ) ; teplitz et al . ( 1998 ) ; bunker et al . ( 1999 ) ; fynbo et al . ( 1999 , 2000 ) ; bouche et al . ( 2001 ) ; van der werf et al . ( 2000 ) ; mller et al . ( 2002 ) ; meyer , thompson , & mannucci ( 2003 ) ; mller , fynbo , & fall ( 2004 ) ; christensen et al . ( 2004 , 2005 ) ; schulte - ladbeck et al . ( 2004 ) ; chen , kennicutt , & rauch ( 2005 ) ; and weatherley et al . ( 2005 ) . in total , we have plotted 71 detections and 30 upper limits in fig . we note , however , that many of the emission line detections are for candidates that have not yet been confirmed with spectroscopy or multiple narrow - band imaging . for the few objects where multiple emission lines have been detected , we have plotted the values from h-@xmath0 , h-@xmath79 , or [ o ii ] , in that order of preference . all the shown candidate detections and 3 @xmath2 upper limits have been normalized to @xmath76 , @xmath77 , and @xmath78 km s@xmath5 mpc@xmath5 . once again , the prescription of kennicutt ( 1998 ) has been used for the conversion from the h-@xmath0 luminosity @xmath49 to the sfr , and a ratio @xmath51 for case - b recombination has been assumed . no correction has been made for dust attenuation , but we examine this issue in more detail below . our apo fp constraints are among the lowest , but are clearly consistent with many other measurements . the open inverted triangles at @xmath81 are based on spectroscopic ( slit ) observations of h-@xmath0 . the unfilled circles are candidate objects with h-@xmath0 emission detected at the absorber redshift , but are often more than @xmath82 away from the quasar , and have a lower precision in redshift . the curves in fig . 8 are useful for comparing the data with global predictions based on the luminosity density of galaxies from the deep galaxy imaging surveys . since most of the data in fig . 8 are for objects in dla fields , we show the calculations of bunker et al . ( 1999 ) for the predicted cross - section - weighted sfr in dlas . these models use two alternative sets of assumptions about the number of absorbers per unit redshift and the distribution of global sfr among individual absorbers at a given redshift . the thick and thin solid curves show the ld5 and ld0 `` large disk '' calculations ( for @xmath83 and @xmath84 respectively ) , of bunker et al . based on the closed - box pei & fall ( 1995 ) models . the ld models assume that dlas are the progenitors of spiral galaxies , with space density equal to that of local spirals , but with size and sfr of each dla larger in the past compared to that of a local spiral . the dashed curve shows the h5 and h0 predictions of bunker et al . ( 1999 ) ( for @xmath83 and @xmath84 respectively ) for the `` hierarchical '' hypothesis , which assumes that there were multiple dlas at high redshift corresponding to every present - day spiral ( i.e. that dlas were sub - galactic fragments that later merged to form present - day spirals ) . thus in the h5 and h0 models , dlas have the same distribution of gas cross - section sizes as in local spirals , but with a higher space density in the past , and have smaller sfrs than in the corresponding `` large - disk '' models . the ld curve , if computed for @xmath76 , @xmath77 , and @xmath78 km s@xmath5 mpc@xmath5 , would lie between the ld5 and ld0 curves , since the distance and time scales for the former cosmology are intermediate between those for @xmath85 and @xmath86 . this can also be seen by comparing the comoving global sfr predicted by the pei & fall ( 1995 ) closed - box models for @xmath83 and @xmath84 , with that predicted by the @xmath87-cold dark matter ( @xmath87-cdm ) hydrodynamical simulations of nagamine , cen , & ostriker ( 2000 ) . thus , a large fraction of the observed sfrs would appear to fall significantly below the prediction of the large - disk scenario . the h curve for this @xmath87-cdm cosmology would be identical to the h5 and h0 curves ( since the latter curves coincide with each other ) . a large fraction of the data points in fig . 8 thus appear to be more consistent with the hierarchical scenario , suggesting perhaps that many absorbers or their companion galaxies may be arising in star - forming dwarf galaxies or sub - galactic fragments that merged later to form present - day galaxies . however , several detections and upper limits , including our fp limits , appear to be considerably below even the hierarchical prediction . indeed , about 63@xmath88 of the detections and about 73@xmath88 of the limits plotted in fig . 8 are less than 5 @xmath8 yr@xmath5 . such large local deviations for the absorber fields from the global mean would be surprising . taken at face value , fig . 8 suggests that the absorption - based view of the cosmic star formation history could be quite different from the emission - based view , i.e. , from the star formation history inferred on the basis of the direct galaxy imaging surveys such as the hubble deep field . low sfrs for galaxies in absorber fields would also be consistent with the low global metallicities found in dlas ( e.g. , kulkarni et al . 2005 and references therein ) . similar suggestions have also been made by wolfe et al . ( 2003 ) , who inferred relatively high sfr values for some dlas based on c ii * absorption , but found the high sfrs to be inconsistent with the low metallicities observed in the dlas . we note , however , that while the c ii * method may potentially offer an interesting way to estimate sfrs in dlas , it seems less direct and more model - dependent than the emission - line based constraints . there are also other possible interpretations of fig 8 . it is possible that the true sfrs in galaxies in the absorber fields are higher , but their emission lines , especially ly-@xmath0 , are attenuated by dust . it is well known that resonance scattering of ly-@xmath0 , in the presence of dust , can lead to quenching of the lyman-@xmath0 emission in high-@xmath89 systems ( e.g. , charlot & fall 1991 ) . presence of dust in some ly-@xmath0 emitting regions is also suggested from the recent detection of 24 @xmath18 m emission in spitzer space telescope images of laes at @xmath72 in an absorber field ( colbert et al . on the other hand , there are several reasons to expect that ly-@xmath0 emission may still be seen from absorbers and their companion galaxies , in at least some cases . first , the very knowledge of a dla means that the quasar in which it appears is not greatly affected by dust attenuation . ly-@xmath0 is easily seen from regions of such low extinction . second , the dust - to - gas ratios in dlas inferred from relative abundances such as [ cr / zn ] or [ fe / zn ] are much smaller than in the milky way ( e.g. , pettini et al . 1997 ; khare et al . 2004 ; and references therein ) . in a recent study of a large number of sdss quasar spectra , we have recently found evidence for a statistically significant but small amount of dust in quasar absorbers ( york et al . finally , there are several examples of high-@xmath9 ly-@xmath0 emitters with very little dust attenuation ( e.g. , giavalisco et al . 1994 ; francis et al . 1996 ; hu , cowie , & mcmahon 1998 ; kudritzki et al . 2000 ) . the effect of dust attenuation is expected to be much less severe in the h-@xmath0 line . indeed , the sfr constraints based on h-@xmath0 detections in most high-@xmath9 dla candidates or companions shown in fig 8 are considerably higher than the constraints from the ly-@xmath0 emission searches . considering detections alone , the median sfrs from h-@xmath0 , ly-@xmath0 , and [ oii ] are 35.9 , 3.2 , and 0.3 @xmath8 yr@xmath5 , respectively . if both detections and limits are considered , the median sfrs from h-@xmath0 , ly-@xmath0 , and [ oii ] are 28.2 , 2.4 , and 0.3 @xmath8 yr@xmath5 treating the limits as detections , and 18.8 , 0.1 , and 0.3 @xmath8 yr@xmath5 treating the limits as zeros . it is tempting to think that the difference between the median h-@xmath0 and ly-@xmath0 values could be partly caused by dust extinction . however , it is not clear whether this is a significant effect , since ly-@xmath0 measurements ( candidates or limits ) are not available for most of the h-@xmath0 candidates . in the few objects that show detections of multiple emission lines , the sfr estimates from the available lines are usually consistent within a factor of @xmath90 . also , some of the h-@xmath0 candidates are at large angular separations from the quasars and have less accurate redshifts . thus , there is no clear indication that ly-@xmath0 in the h-@xmath0 candidates is attenuated by dust . in any case , dust alone may not explain low ly-@xmath0 emission in every individual case , because the extent of ly-@xmath0 attenuation also depends on several other factors , such as the orientation , the topology of the h ii region , and the distribution of stars , gas and dust within the absorber . finally , dust attenuation can not explain why nearly all of the sfrs inferred from [ o ii ] and @xmath91 of the sfrs inferred from h-@xmath0 are low . another possibility is that the star - forming regions in the absorber galaxies are compact , lie directly in front of the quasars , and hence get lost in the quasar point spread function ( psf ) in our study and other ground - based studies so far . for example , if the absorbing galaxies had star - forming cores comparable in size to those in the lyman - break galaxies [ typically @xmath92 kpc ] aligned perfectly in front of the quasars , it would not be possible to resolve them with studies such as our own ( and most other ground - based studies used in fig . in other words , the sfr constraints derived from such studies may be more appropriate for the companion galaxies of the absorbers , rather than the absorbers themselves . it is difficult to rule out this possibility without high - resolution imaging data . however , the chance of having the star forming region exactly in front of the quasar in _ each _ case is small . indeed , the absence of emission within the ly-@xmath0 absorption profiles in the published spectra for most dlas ( including those from our study ) suggests that they do not have star - forming regions perfectly aligned with the background quasars . even if the absorber galaxies themselves were small , they are unlikely to be isolated objects , since such small galaxies are expected to occur as companions to larger galaxies within a few hundred kpc . in any case , it would still be surprising why the few existing high - resolution imaging studies of absorber galaxies have often failed to detect h-@xmath0 emission or found relatively weak ly-@xmath0 emission ( e.g. , bouche et al . 2001 ; kulkarni et al . 2000 , 2001 ; mller et al . 2002 ) . future narrow - band imaging studies of more high - redshift quasar absorber fields would help to understand whether the low ly-@xmath0 fluxes we find are representative of the absorber galaxies . furthermore , more narrow - band imaging or spectroscopic studies in the near - infrared will give access to the h-@xmath0 emission line from the @xmath93 absorbers . a systematic comparison of the h-@xmath0 and ly-@xmath0 data for the same absorbers can help to understand whether the emission lines in the galaxies in absorber fields are simply attenuated by dust , or whether these galaxies truly have low sfrs . it will thus be interesting to look at ly-@xmath0 emission from the objects reported to show strong h-@xmath0 emission , and to look at h-@xmath0 near the quasars for which we have limits on ly-@xmath0 . finally , more high - resolution narrow - band imaging and/or spectroscopic studies would be especially important to understand whether the star formation in the absorber galaxies is restricted to compact regions . this paper is based on observations obtained with the apache point observatory 3.5-meter telescope , which is owned and operated by the astrophysical research consortium . we thank jonathan brinkmann for assistance with the fabry - perot setup and calibration at the apo . we also thank andrew bunker for helpful discussion and betty stobie for providing and assisting with the idp-3 package . finally , we are grateful to an anonymous referee whose comments helped to improve this paper . vpk , dgt , and jm acknowledge support from the national science foundation grant ast-0206197 and from the university of south carolina research foundation . bechtold , j. , elston , r. , yee , h. k. c. , ellingson , e. , & cutri , r. m. 1998 , in ` the young universe : galaxy formation and evolution at intermediate and high redshift ' , ed . s. dodorico , a. fontana , and e. giallongo , asp conf . series , ( pasp ) , 146 , 241 _ * fig . 1 * apo images of the field of q0216 + 080 . ( a ) @xmath95 b - band image ( top left ) ; ( b ) @xmath95 narrow - band image before continuum subtraction ( top right ) ; ( c ) @xmath95 narrow - band image after continuum subtraction ( bottom left ) ; and ( d ) @xmath39 narrow - band image after continuum subtraction ( bottom right ) . see attached gif images . _ _ measurements of star formation rates ( in @xmath8 yr@xmath5 ) for candidate objects in quasar absorber fields , from narrow - band imaging and spectroscopic searches for ly-@xmath0 , h-@xmath0 , h-@xmath79 , [ o ii ] and [ o iii ] emission lines . data points are from our apo ly-@xmath0 survey ( this work ) , our previous nicmos h-@xmath0 imaging ( kulkarni et al . 2000 , 2001 ) , and other literature ( see text ) . most of the higher values are based on candidate h-@xmath0 emitters , often located far from the quasar lines of sight . the thick solid ( upper ) and thin solid ( middle ) curves show , respectively , the calculations of bunker et al . ( 1999 ) for the predicted cross - section - weighted sfr in the large - disk scenario for @xmath83 and @xmath84 . the dashed curve shows the calculations of bunker et al . ( 1999 ) for the hierarchical scenario for @xmath83 and @xmath84 . all of these curves are based on the closed - box global sfr models of pei & fall ( 1995 ) . _ llllllr qso & r.a.(j2000 ) & dec.(j2000 ) & @xmath96 & @xmath97 & known ions & @xmath98 + & & & & & & ( ) + q0216 + 080&02:18:57.3 & + 08:17:28.0&2.9957&2.2931&civ , aliii , alii,& ... @xmath99 + & & & & & siii , feii ; dla + q0636 + 680&06:42:04.3 & + 67:58:35.6 & 3.1775&2.3112&civ , cii , alii & ... + q0956 + 123&09:58:52.3 & + 12:02:43.2 & 3.306&2.3104&civ , siiv , alii&1.42 + q1209 + 093&12:11:34.9 & + 09:02:22.8 & 3.297&2.5822&civ , alii , feii;&21.2 + & & & & & dla + q1442 + 101&14:45:16.5 & + 09:58:36.2 & 3.535&2.5632&ovi , civ , siiv , & 1.40 + & & & & & ciii , siiii , siii , oi + q2233 + 131&22:36:19.2&+13:26:20.0&3.2978&2.5537&civ , cii , siiv , siiii,&5.35 + & & & & & siii , oi ; sub - dla + 53w002&17:14:14.7&+50:15:29.7 & ... & ... & ... & ... + ( calibration & ... & ... & ... & ... & ... & ... + field ) & ... & ... & ... & ... & ... & ... + q0216 + 080 & 2001/11/10&05:55:28.87&1.111&blue&4050/160&4002.2&2400&14.69 + ... & 2001/11/10&06:36:18.15&1.111&blue&4050/160&4002.2&1200 & + ... & 2001/11/11&05:42:48.66&1.102&blue&4050/160&4002.2&2400 & + ... & 2001/11/11&06:52:30.75&1.125&blue&4050/160&4002.2&1660 & + ... & 2001/11/12&05:20:37.43&1.111&blue&4050/160&4001.3&2400 & + ... & 2001/11/12&06:05:54.94&1.100&blue&4050/160&4001.3&2400 & + ... & 2001/12/15&03:17:36.77&1.107&blue&4050/160&4002.2&2400 & + ... & 2001/12/15&04:00:38.97&1.101&blue&4050/160&4002.2&2400 & + ... & 2001/12/15&04:50:23.34&1.139&blue&4050/160&4002.2&2400 & + ... & 2001/12/15&05:32:31.49&1.216&blue&4050/160&4002.2&2400 & + ... & 2002/11/03&06:59:28.92 & 1.106 & vis - b & 4050/160&4008.0 & 2400 & + ... & 2002/11/03&07:52:40.87&1.163&vis - b&4050/160&4008.0 & 2400 & + ... & 2002/11/08&07:42:27.53&1.180&vis - b&4050/160&4008.0 & 1800 & + & & & & & & & + ... & 2000/10/26&05:23:00.92&1.201& ... &b&4400&480 & + ... & 2000/10/26&05:32:58.88&1.180& ... &b&4400&480 & + ... & 2000/10/26&05:42:02.58&1.164& ... &b&4400&480 & + ... & 2000/10/26&05:51:06.28&1.150& ... &b&4400&480 & + ... & 2000/10/26&06:00:28.58&1.137& ... &b&4400&480 & + ... & 2001/12/15&01:21:29.11&1.338& ... &b&4400&300 & + ... & 2001/12/15&01:27:45.25&1.314& ... &b&4400&300 & + ... & 2001/12/15&01:34:03.80&1.292& ... &b&4400&300 & + ... & 2001/12/15&01:40:22.75&1.272& ... &b&4400&300 & + ... & 2001/12/15&01:46:35.10&1.254& ... &b&4400&300 & + & & & & & & & + & & & & & & & + q0636 + 680 & 2000/10/26&10:51:55.81&1.229&blue&4050/160&4024&2400&14.66 + ... & 2000/10/26&11:33:12.48&1.224&blue&4050/160&4024&2400 & + ... & 2001/02/22&03:20:22.82&1.225&blue&4050/160&4024.3&2400 & + ... & 2001/02/22&04:18:32.24&1.232&blue&4050/160&4024.3&2400 & + ... & 2001/02/22&05:06:11.28&1.260&blue&4050/160&4024.3&2400 & + ... & 2001/12/15&07:15:20.01&1.236&blue&4050/160&4024.2&2400 & + ... & 2001/12/15&07:56:38.23&1.224&blue&4050/160&4024.2&2400 & + ... & 2001/12/15&08:39:13.21&1.228&blue&4050/160&4024.2&2400 & + ... & 2001/12/15&09:27:39.32&1.252&blue&4050/160&4024.2&2400 & + ... & 2002/03/12&02:56:26.66&1.228&blue&4050/160&4024.2&2400 & + ... & 2002/03/12&03:38:47.01&1.247&blue&4050/160&4024.2&2400 & + ... & 2002/11/03&10:45:35.05&1.224&vis - b&4050/160&4030.0&2400 & + ... & 2002/11/03&11:29:28.69&1.229&vis - b&4050/160&4030.0&2400 & + ... & 2002/11/05&09:44:12.32&1.242&vis - b&4050/160&4030.0&2400 & + & & & & & & & + ... & 2000/10/26&10:07:53.96&1.251& ... &b&4400&480 & + ... & 2000/10/26&10:17:24.49&1.244& ... &b&4400&480 & + ... & 2000/10/26&10:26:39.20&1.239& ... &b&4400&480 & + ... & 2000/10/26&10:35:43.11&1.235& ... &b&4400&480 & + ... & 2001/02/25&05:16:15.67&1.281& ... &b&4400&300 & + ... & 2001/02/25&05:22:41.63&1.288& ... &b&4400&300 & + ... & 2001/02/25&05:29:27.60&1.296& ... &b&4400&300 & + & & & & & & & + & & & & & & & + q0956 + 123&2001/02/22&06:23:39.98&1.077 & blue&4050/160 & 4023.3&2400&14.79 + ... & 2001/02/22&07:24:08.73&1.078&blue&4050/160 & 4023.3&2400 & + ... & 2001/02/22&08:16:52.25&1.135&blue&4050/160 & 4023.3&2400 & + ... & 2001/02/22&09:24:17.19&1.312&blue&4050/160 & 4023.3&2400 & + ... & 2001/02/25&08:38:17.40&1.204&blue&4050/160&4023.4&2400 & + ... & 2001/05/22&03:18:47.08&1.258&blue&4050/160&4023.3&2400 & + ... & 2002/03/12&05:13:05.87&1.078&blue&4050/160&4023.2&2400 & + ... & 2002/03/12&05:55:28.03&1.071&blue&4050/160&4023.2&2400 & + ... & 2002/03/12&06:44:56.70&1.104&blue&4050/160&4023.2&2400 & + & & & & & & & + ... & 2001/02/25&06:10:45.09&1.078& ... &b&4400&300 & + ... & 2001/02/25&06:17:40.16&1.074& ... &b&4400&300 & + ... & 2001/02/25&06:24:01.83&1.072& ... &b&4400&300 & + ... & 2001/02/25&06:30:58.02&1.070& ... &b&4400&300 & + ... & 2001/02/25&06:37:32.38&1.069& ... &b&4400&300 & + ... & 2001/02/25&06:49:41.66&1.069& ... &b&4400&300 & + ... & 2001/02/25&07:06:17.69&1.074& ... &b&4400&300 & + ... & 2001/02/25&07:12:54.06&1.078& ... &b&4400&300 & + & & & & & & & + & & & & & & & + q1209 + 093 & 2001/05/20&04:42:28.87&1.154&blue&4300/135&4353.6&2400&7.01 + ... & 2001/05/20&05:28:37.23&1.260&blue & 4300/135 & 4353.6&2400 & + ... & 2001/05/22&05:11:26.45&1.233&blue & 4300/135&4353.5&2400 & + ... & 2001/05/23&03:41:53.18&1.101&blue & 4300/135&4353.7&2400 & + ... & 2001/05/23&04:24:29.14&1.145&blue & 4300/135&4353.6&2400 & + ... & 2002/03/11&08:52:16.27&1.119&blue & 4300/135&4353.5&2400 & + ... & 2002/03/12&08:25:38.11&1.101&blue&4300/135&4353.5&2400 & + ... & 2002/03/12&09:17:40.82&1.159&blue&4300/135&4353.5&2400 & + & & & & & & & + ... & 2001/05/23&05:22:34.56&1.277& ... &b&4400&300 & + ... & 2001/05/23&05:37:57.81&1.332& ... &b&4400&300 & + ... & 2001/05/23&05:44:12.36&1.357& ... &b&4400&300 & + ... & 2001/05/23&05:50:36.43&1.384& ... &b&4400&300 & + ... & 2001/05/23&05:56:55.87&1.413& ... &b&4400&300 & + ... & 2001/05/23&06:03:07.01&1.444& ... &b&4400&300 & + ... & 2001/05/23&06:09:17.35&1.477& ... &b&4400&300 & + ... & 2001/05/23&06:15:38.11&1.514& ... &b&4400&300 & + & & & & & & & + q1442 + 101&2001/05/20&07:29:32.74&1.170&blue&4300/135&4330.5&2400&6.10 + ... & 2001/05/20&08:17:28.39&1.299&blue&4300/135&4330.4&2400 & + ... & 2001/05/21&06:37:53.56&1.104&blue&4300/135&4330.6&2400 & + ... & 2001/05/21&07:27:17.70&1.174&blue&4300/135&4330.5&2400 & + ... & 2001/05/21&08:09:09.42&1.284&blue&4300/135&4330.4&2400 & + ... & 2001/05/22&06:15:29.60&1.091&blue&4300/135&4330.5&2400 & + ... & 2001/05/22&06:57:37.14&1.130&blue&4300/135&4330.5&2400 & + ... & 2001/05/23&06:51:26.89&1.127&blue&4300/135&4330.5&2400 & + ... & 2001/05/23&07:33:07.00&1.204&blue&4300/135&4330.5&2400 & + & & & & & & & + ... & 2001/02/25&10:03:28.53&1.153& ... &b&4400&300 & + ... & 2001/02/25&10:10:44.33&1.141& ... &b&4400 & 300 & + ... & 2001/02/25&10:18:19.75&1.130& ... &b&4400&300 & + ... & 2001/02/25&10:26:38.82&1.120& ... &b&4400&300 & + ... & 2001/02/25&11:21:21.65&1.085& ... &b&4400&300 & + ... & 2001/02/25&11:32:29.47&1.085& ... &b&4400&300 & + ... & 2001/02/25&11:41:09.15&1.087& ... &b&4400&300 & + ... & 2001/02/25&11:48:41.57&1.089& ... &b&4400&300 & + ... & 2001/02/25&11:55:50.50&1.0938& ... &b&4400&300 & + ... & 2001/05/23&08:21:14.63&1.359& ... &b&4400&300 & + ... & 2001/05/23&08:27:35.43&1.387& ... &b&4400&300 & + ... & 2001/05/23&08:33:59.98&1.418& ... &b&4400&300 & + & & & & & & & + & & & & & & & + q2233 + 131&2000/10/26&03:45:44.77&1.066&blue&4300/135&4319&2400&5.59 + ... & 2000/10/27&01:38:14.59&1.150&blue&4300/135&4314.3&2400 & + ... & 2000/10/27&02:41:43.61&1.070&blue&4300/135&4317.0&2400 & + ... & 2000/10/27&03:32:30.68&1.062&blue&4300/135&4318&2400 & + ... & 2000/10/27&04:20:34.25&1.097&blue&4300/135&4318&2400 & + ... & 2000/10/27&05:12:21.30&1.190&blue&4300/135&4318.5&2400 & + ... & 2000/10/27&06:11:24.96&1.398&blue&4300/135&4319&2400 & + ... & 2001/11/10&02:29:02.38&1.060&blue&4313/116&4318.9&2400 & + ... & 2001/11/10&03:10:40.90&1.081&blue&4313/116&4318.9&2400 & + ... & 2001/11/10&03:52:49.45&1.136&blue&4313/116&4318.9&2400 & + ... & 2001/11/10&04:34:24.56&1.233&blue&4313/116&4318.9&2400 & + ... & 2001/11/11&02:01:10.21&1.062&blue&4313/116&4318.9&2400 & + ... & 2001/11/11&02:45:46.09&1.066&blue&4313/116&4318.9&2400 & + ... & 2001/11/12&02:18:11.07&1.060&blue&4313/116&4318.9&2400 & + ... & 2001/11/12&03:01:43.69&1.080&blue&4313/116&4318.9&2400 & + ... & 2002/11/04&03:08:18.76 & 1.064&vis - b&4300/135&4325.0&1740 & + ... & 2002/11/06&01:29:15.00&1.104&vis - b&4300/135&4325.0&2400 & + ... & 2002/11/06&02:31:17.24&1.060&vis - b&4300/135&4325.0&800 & + ... & 2002/11/06&05:02:00.19&1.268&vis - b&4300/135&4325.0&2400 & + & & & & & & & + ... & 2000/10/26&02:42:34.68&1.072& ... &b&4400&480 & + ... & 2000/10/26&02:53:33.87&1.066& ... &b&4400&480 & + ... & 2000/10/26&03:06:41.39&1.061& ... &b&4400&480 & + ... & 2000/10/26&03:17:01.37&1.060& ... &b&4400&480 & + ... & 2000/10/26&03:28:07.59&1.060& ... &b&4400&480 & + ... & 2001/11/11&03:54:43.13&1.147& ... &b&4400&2400 & + ... & 2000/10/25&05:26:24.96&1.205& ... &b&4400&1200 & + & & & & & & & + & & & & & & & + 53w002&2001/05/20&09:45:49.27&1.086&blue&4155/160&4128.4&2400&8.31 + ... & 2001/05/21&09:34:32.16&1.079&blue&4155/160&4128.5&2400 & + ... & 2001/05/21&10:15:54.17&1.126&blue&4155/160&4128.6&2400 & + ... & 2001/05/22&09:08:42.61&1.063&blue&4155/160&4128.3&2400 & + ... & 2001/05/23&09:31:39.77&1.084&blue&4155/160&4128.3&2400 & + & & & & & & & + ... & 2001/05/22&09:57:53.15&1.108 & ... & b&4400&300 & + ... & 2001/05/22&10:04:47.73&1.117 & ... & b&4400&300 & + ... & 2001/05/22&10:10:54.66&1.125 & ... & b&4400&300 & + ... & 2001/05/22&10:17:17.82&1.134 & ... & b&4400&300 & + ... & 2001/05/22&10:23:55.19&1.145 & ... & b&4400&300 & + ... & 2001/05/22&10:34:30.57&1.163 & ... & b&4400&300 & + ... & 2001/05/22&10:41:24.96&1.176 & ... & b&4400&300 & + ... & 2001/05/22&10:52:14.56&1.197 & ... & b&4400&300 & + ... & 2001/05/23&08:44:33.56&1.054 & ... & b&4400&300 & + ... & 2001/05/23&08:51:20.34&1.057 & ... & b&4400&300 & + ... & 2001/05/23&08:57:47.50&1.060 & ... & b&4400&300 & + ... & 2001/05/23&09:05:17.92&1.064 & ... & b&4400&300 & + & & & & & & & + llcc qso & @xmath97 & @xmath102(erg s@xmath5 @xmath6 ) ) & sfr ( @xmath8 yr@xmath5)@xmath103 + q0216 + 080&2.2931&@xmath104&@xmath105 + q0636 + 680&2.3112&@xmath106&@xmath107 + q0956 + 123&2.3104&@xmath108&@xmath105 + q1209 + 093&2.5822&@xmath109&@xmath110 + q1442 + 101&2.5632&@xmath111&@xmath112 + q2233 + 131&2.5537&@xmath113&@xmath114 + @xmath100 : 3 @xmath2 upper limits on lyman-@xmath0 flux for a 2 pixel x 2 pixel region , corresponding to a physical size of @xmath115 kpc x 6.0 kpc at @xmath116 ; @xmath117 : 3 @xmath2 upper limits on sfr .

an interesting observation was reported by corrigan - sasaki that all the frequencies of small oscillations around equilibrium are _ quantised _ " for calogero and sutherland ( c - s ) systems , typical integrable multi - particle dynamics . we present an analytic proof by applying recent results of loris - sasaki . explicit forms of ` classical ' and quantum eigenfunctions are presented for c - s systems based on any root systems . yukawa institute kyoto + yitp-03 - 51 + hep - th/0308052 + august 2003 * quantum & classical eigenfunctions in + calogero & sutherland systems + + * i. loris and r. sasaki yukawa institute for theoretical physics , + kyoto university , kyoto 606 - 8502 , japan + dienst theoretische natuurkunde , vrije universiteit brussel , + pleinlaan 2 , b-1050 brussels , belgium

hep-th0308052

in a recent paper @xcite simple theorems pertaining to the correspondence between quantum and classical dynamics are proved for the general multi - particle quantum mechanical systems with discrete eigenvalues . the theorems relate quantum mechanical eigenvalues and eigenfunctions to the properties of the classical mechanical system at equilibrium . corresponding to each quantum eigenfunction , a ` _ classical eigenfunction _ ' is defined whose eigenvalue is given by the ` main part ' , that is the order @xmath3 part , of the quantum eigenvalue . for the ` _ elementary excitations _ ' these classical and quantum eigenvalues are nothing but the eigenfrequencies of the normal modes of the small oscillations at equilibrium . we apply these theorems to the calogero and sutherland @xcite systems , typical integrable multi - particle dynamics with long range interactions based on root systems @xcite . the theorems provide an analytic proof for the interesting observations made by corrigan - sasaki @xcite concerning the classical and quantum integrability in calogero and sutherland systems . explicit forms of the classical and quantum eigenfunctions for the elementary excitations are presented for the calogero and sutherland systems based on any root systems . these exemplify another aspect of the close relationship between the classical and quantum integrability in calogero and sutherland systems . this paper is organised as follows . in section two , basic formulation of multi - particle quantum mechanics in terms of the _ prepotential _ @xcite is briefly reviewed . after the reformulation of the quantum mechanical wavefunctions at equilibrium , the main theorem of the loris - sasaki paper @xcite is recapitulated . in section three the basic concepts of the calogero and sutherland systems are summarised . section four and five are the main part of this paper , presenting the classical and quantum eigenfunctions of the calogero systems ( section four ) and sutherland systems ( section five ) . the final section is for summary and comments . let us start with a basic formulation of multi - particle quantum mechanical system in terms of a _ prepotential _ @xcite and later we will discuss its relationship with the corresponding classical ( @xmath4 ) dynamics . the dynamical variables are the coordinates @xmath5 and their canonically conjugate momenta @xmath6 , subject to the heisenberg commutation relations or the poisson bracket relations : @xmath7=i\hbar\delta_{j\,k},\qquad [ q_{j},q_{k}]= [ p_{j},p_{k}]=0,\\ \{q_{j},p_{k}\}=\delta_{j\,k},\qquad \{q_{j},q_{k}\}= \{p_{j},p_{k}\}=0.\end{aligned}\ ] ] we will adopt the standard vector notation in @xmath8 : @xmath9 in which @xmath10 is the number of particles . in quantum theory , the momentum operator @xmath11 acts as a differential operator : @xmath12 throughout this paper we discuss the standard hamiltonian system @xmath13 in which we have assumed for simplicity that all the particles have the same mass , which is rescaled to unity . let us start with mild assumptions that the system has a unique and _ square integrable _ ground state @xmath14 : @xmath15 and that it has a finite ( or an infinite ) number of _ discrete _ eigenvalues : @xmath16 here we adopt the convention that the ground state energy is vanishing , by adjusting the constant part of the potential @xmath17 , see below . since the above time - independent schrdinger equation is real for a self - adjoint hamiltonian and that the ground state has no _ node _ we express the ground state eigenfunction as @xmath18 in which a real function @xmath19 is called a _ prepotential _ @xcite . by simple differentiation of ( [ grstate ] ) , we obtain @xmath20\psi_0,\ ] ] which results in @xmath21\right\}\psi_0=0.\ ] ] in other words , we can express the potential ( plus the ground state energy ) in terms of the prepotential @xcite @xmath22 . \label{potform}\ ] ] by removing the obvious @xmath3-dependent terms , let us define a _ potential @xmath23 : @xmath24 conversely , ( [ potform ] ) is a riccati equation determining the prepotential @xmath25 for a given potential @xmath17 ( or @xmath26 ) . needless to say , it does not matter if the prepotential can be expressed in terms of elementary functions or not . now let us consider the equilibrium point of the classical potential @xmath26 ( [ clpot ] ) . the classical hamiltonian @xmath27 has a stationary solution at the classical equilibrium point , @xmath28 , @xmath29 . there could be , in general , many stationary points of the classical potential @xmath26 , among which we will focus on the ` _ maximum _ ' point @xmath30 of the ground state wavefunction @xmath14 @xcite : @xmath31 by expanding the classical potential @xmath26 around @xmath30 ( [ vmin ] ) , we obtain @xmath32 since @xmath33 , ( [ clpot ] ) . thus the eigen ( angular ) frequencies ( ( frequency)@xmath34 ) of small oscillations near the classical equilibrium are given as the eigenvalues of the hessian matrix @xmath35 ( @xmath36 ) : @xmath37,\quad \widetilde{v}_c=\mbox{matrix}\left[\left.\ { \partial^2 v_c\over{\partial q_j\partial q_k}}\right|_{\bar{q}}\right]=\widetilde{w}^2 . \label{vwmat}\ ] ] let us express the discrete eigenfunctions in product forms @xmath38 in which @xmath39 obeys a simplified equation with the similarity transformed hamiltonian @xmath40 @xcite : @xmath41 in which @xmath42 is the laplacian and a linear differential operator @xmath43 is defined for any smooth function @xmath44 as : @xmath45 here we adjust the normalisation of the eigenfunctions @xmath46 so that the corresponding _ classical _ " eigenfunctions @xmath47 are finite ( non - vanishing ) in the limit @xmath4 : @xmath48 by taking the classical limit ( @xmath4 ) of ( [ sthameq ] ) and considering ( [ eigexp ] ) , ( [ htilform ] ) , we arrive at an ` _ eigenvalue equation _ ' for the _ classical _ " wavefunctions @xmath49 in which the operator @xmath43 is defined above ( [ cleigen ] ) . conversely one could start with the above eigenvalue equation . one defines the classical eigenfunctions as its solutions satisfying certain regularity conditions . then the _ quantum _ eigenfunction @xmath39 could be considered as an @xmath3-_deformation _ of the _ classical _ eigenfunction @xmath50 . for the calogero and sutherland systems to be discussed below , there is a one - to - one correspondence between the classical and quantum eigenfunctions . for generic multi - particle quantum mechanical systems , the situation is less clear . the _ classical _ eigenfunctions have the following remarkable properties : [ section ] [ prop21 ] the product of two _ classical _ eigenfunctions @xmath51 and @xmath52 is again a _ classical _ eigenfunction with the eigenvalue @xmath53 , @xmath54 [ prop22 ] the classical eigenfunctions vanish at the equilibrium @xmath30 @xmath55 [ prop23 ] the derivatives of a classical eigenfunction at the equilibrium @xmath30 form an eigenvector of the hessian matrix @xmath35 , iff @xmath56 @xmath57 obviously the hessian matrix @xmath35 has at most @xmath10 different eigenvalues and eigenvectors . the classical eigenfunctions @xmath58 , @xmath59 for which @xmath60 will be called _ elementary excitations _ " . at equilibrium , each corresponds to the _ normal coordinate _ of the small oscillations with the eigen ( angular ) frequency @xmath61 . the elementary excitations are the generators of all the classical eigenfunctions . in other words , any classical eigenfunction can be expressed as @xmath62 or a linear combination thereof with the same eigenvalue @xmath63 . the above type of classical eigenfunctions are obviously non - elementary and they have zero gradient at equilibrium , for example , @xmath64 . because of this property , the representation of the elementary excitations is not unique except for some lower members . these results provide a basis of the analytical proof of the observations made in corrigan - sasaki paper @xcite on the correspondence / contrast between the _ classical _ and _ quantum _ integrability in calogero - moser systems . it should be mentioned that perelomov s recent work @xcite asserts essentially our proposition 2.3 for the special cases of the quantum - classical eigenvalue correspondence of the sutherland systems . throughout this section we have assumed that the prepotential @xmath25 is independent of the planck s constant @xmath3 , for simplicity of the presentation . the main content of this section is valid even if @xmath25 depends on @xmath3 , so long as @xmath65 is well - defined . a celebrated example that @xmath66 diverges is the hydrogen atom , for which the classical equilibrium does not exist . in this case the quantum - classical correspondence does not make sense and the present formulation does not apply . in the subsequent sections we will show many explicit examples of the _ classical _ and _ quantum _ eigenfunctions and their relationship . a calogero - moser system is a multi - particle hamiltonian dynamics associated with a root system @xmath67 of rank @xmath10 . this is a set of vectors in @xmath8 invariant under reflections in the hyperplane perpendicular to each vector in @xmath67 : @xmath68 the set of reflections @xmath69 generates a finite reflection group @xmath70 , known as a coxeter ( or weyl ) group . a calogero - moser system is integrable both at the classical and quantum levels for various choices of the long range interaction potentials ; rational ( @xmath71 ) , rational with a harmonic confining potential , trigonometric ( @xmath72 ) , hyperbolic ( @xmath73 ) and elliptic with the weierstrass function ( @xmath74 ) potential . in the rest of this paper we will discuss the rational case ( with a harmonic confining potential ) under the name of calogero system @xcite and the trigonometric potential case to be called sutherland system @xcite . both quantum hamiltonians have an infinite number of discrete eigenvalues . the prepotentials are @xmath75 in which @xmath76 is the prepotential of the theory without the harmonic confining potential and @xmath77 in these formulae , @xmath78 is the set of positive roots and @xmath79 are real _ positive _ coupling constants which are defined on orbits of the corresponding coxeter group , _ i.e. _ they are identical for roots in the same orbit . for crystallographic root systems there is one coupling constant @xmath80 for all roots in simply - laced models , and there are two _ independent _ coupling constants , @xmath81 for long roots and @xmath82 for short roots in non - simply laced models . we will give the explicit forms of @xmath25 in later sections . throughout this paper we put the scale factor in the trigonometric functions to unity for simplicity ; instead of the general form @xmath83 , we use @xmath84 . we also adopt the convention that long roots have squared length two , @xmath85 , unless otherwise stated . these prepotentials determine the potentials : @xmath86 \begin{displaystyle } { 1\over2}\sum_{\rho\in\delta_+ } { g_{\rho}(g_{\rho}-\hbar ) \rho^{2}\over{\sin^2(\rho\cdot q)}}-e_0 , \qquad\qquad\quad\ \mbox{sutherland}. \end{displaystyle } \end{array } \right . \label{fullpot}\end{aligned}\ ] ] the sutherland systems are integrable , both at the classical and quantum levels , for the crystallographic root systems , that is those associated with simple lie algebras : \{@xmath87 models , it is customary to introduce one more degree of freedom , @xmath88 and @xmath89 and embed all of the roots in @xmath90 . [ embedding ] ] , @xmath91 , @xmath92 , @xmath93 , @xmath94 , @xmath95 , @xmath96 , @xmath97 and @xmath98 and the so - called @xmath99 . on the other hand , the calogero systems are integrable for any root systems , crystallographic and non - crystallographic . the latter are @xmath100 , @xmath101 , and @xmath102 , the dihedral group of order @xmath103 . the prepotential @xmath25 ( [ calw ] ) , ( [ sutw ] ) , and hence the ( classical ) potential @xmath17 ( [ fullpot ] ) and the hamiltonian are coxeter ( weyl ) invariant : @xmath104 which is the symmetry of the entire calogero - moser systems . this results in the fact that the ground state @xmath14 and all the other eigenfunctions are are coxeter ( weyl ) invariant @xcite : @xmath105 the quantum calogero and sutherland systems are not only integrable but also _ exactly solvable _ @xcite , that is , the similarity transformed hamiltonians ( [ htilform ] ) are _ lower triangular _ in certain basis of the hilbert space . the eigenvalues can be read off easily from the diagonal matrix elements of @xmath40 . the exact eigenvalues of the excited states in the calogero system are an integer multiple of the oscillator quantum @xmath106 : @xmath107 here @xmath108 are non - negative _ quantum numbers _ and @xmath109 and the integers @xmath110 , @xmath59 are called the _ exponents _ of the root system @xmath67 : [ cols="^,<,^,<",options="header " , ] + table i : the degrees @xmath111 at an elementary excitation exists . the coupling constant(s ) @xmath112 of the rational @xmath71 potentials shifts _ only _ the ground state energy @xmath113 in ( [ fullpot ] ) : @xmath114 in which the first term is the zero - point energy of the oscillators . for a given non - negative integer @xmath115 , let @xmath116 be the number of different solutions of @xmath117 then the energy eigenvalue @xmath118 has the degeneracy @xmath116 . the exact eigenvalues of the excited states in the sutherland @xcite system are specified by the _ dominant highest weight _ @xmath119 : @xmath120 here @xmath108 are non - negative _ quantum numbers _ , @xmath121 , @xmath59 are the _ fundamental weights _ and @xmath122 and @xmath123 are called the _ weyl vector _ and a _ deformed weyl vector_. the ground state energy @xmath113 in ( [ fullpot ] ) is solely determined by @xmath123 : @xmath124 for the general discussion of quantum calogero and sutherland systems for any root system along the present line of arguments , the quantum integrability , lax pairs , quantum eigenfunctions , creation - annihilation operators etc , we refer to @xcite . a rather different approach by heckman and opdam @xcite to calogero - moser models with degenerate potentials based on any root system should also be mentioned in this connection . the eigenfunctions of the sutherland systems are sometimes called heckman - opdam s jacobi polynomials . those for the @xmath125 series are known as the jack polynomials @xcite . in the following two sections , we will show the classical and quantum eigenfunctions of the elementary excitations in calogero systems ( section [ rationals ] ) and in sutherland systems ( section [ trigs ] ) . for brevity and clarity of the presentation , we present the eigenfunctions of the ` _ reduced theory _ ' in which most of the coupling constants are put to unity . to be more precise , for simply - laced theories ( @xmath125 , @xmath126 , @xmath127 , @xmath128 and @xmath129odd ) ) we put the coupling constant unity , @xmath130 . for non simply - laced theories ( @xmath131 , @xmath132 , @xmath133 , @xmath134 and @xmath129even ) ) we put the coupling constant for long roots unity @xmath135 and keep the coupling constant for short roots intact , @xmath136 . the angular frequency of the harmonic confining potential is also put to unity , @xmath137 . let us introduce elementary symmetric polynomials as useful ingredients for expressing the eigenfunctions . the degree @xmath138 elementary symmetric polynomial in @xmath10 variables , @xmath139 , @xmath140 is defined by the expansion of a generating function @xmath141 the basis of the quantum eigenfunctions @xmath46 of the calogero system is the coxeter ( weyl ) invariant polynomials in the coordinates @xmath142 . in order to express the eigenfunctions in a closed form , let us introduce the similarity transformed hamiltonian @xmath143 _ without the harmonic confining potential _ : @xmath144 in which @xmath145 is the euler derivative measuring the degree of a monomial . the hamiltonian @xmath143 maps a coxeter invariant polynomial to another _ with degree two less _ @xmath146=-2\hat{h}_r,\ ] ] which implies for an arbitrary parameter @xmath147 @xmath148=\exp[\kappa\hat{h}_r]\mathcal{d}.\ ] ] the _ lower triangularity _ of the hamiltonian @xmath143 means that the exponential operator contains only finite powers of @xmath143 , up to @xmath149 $ ] , when applied to a coxeter invariant polynomial of degree @xmath115 . by multiplying @xmath106 on both sides and choosing @xmath150 , we obtain @xmath151= \exp[\hat{h}_r/2\omega\hbar]\omega\hbar\mathcal{d}.\ ] ] thus we arrive at a formula of an eigenfunction of @xmath40 with the eigenvalue @xmath152 ( @xmath115 being a non - negative integer ) , starting from an arbitrary _ homogeneous _ coxeter invariant polynomial @xmath153 of degree @xmath115 : @xmath154\phi_n(q),\quad \mathcal{d}\phi_n(q)=n\phi_n(q ) . \label{qeigform2}\end{aligned}\ ] ] a similar formula was derived in @xcite for the theories based on the @xmath125-series of root systems . there are @xmath116 ( [ partitions ] ) linearly independent coxeter invariant homogeneous polynomials , which is equal to the degeneracy of the eigenspace of @xmath155 . among them there are special eigenfunctions which are linear combinations of the coxeter invariant _ homogeneous _ polynomials such that they are annihilated by @xmath143 : @xmath156 the number of homogeneous eigenfunctions is @xmath157 , which is much less than the total dimensionality of the eigenspace , @xmath116 . the simplest class of quantum eigenfunctions depends only on @xmath158 , ( [ qpdefs ] ) : @xmath159 if we define @xmath160 and @xmath161 , ( with @xmath113 defined in ( [ e0def ] ) ) , we obtain the associated laguerre polynomial in @xmath162 as the quantum eigenfunction @xmath163(-1)^n{x^n\over{n!}}= \sum_{j=0}^n(-1)^j{\alpha+n\choose n - j}{x^j\over{j!}}=l^{(\alpha)}_n(x ) . \label{laguerrepoly}\ ] ] this class of universal eigenfunctions is known from the early days of calogero systems @xcite . it is easy to verify proposition [ prop22 ] and [ prop23 ] for this eigenfunction , since the classical limit is @xmath164 in which @xmath165 @xcite . the classical counterpart of the above general result ( [ qeigform1 ] ) , ( [ qeigform2 ] ) is simply obtained as the @xmath4 limit : @xmath166\phi_n(q),\quad \mathcal{d}\phi_n(q)=n\phi_n(q).\end{aligned}\ ] ] since the operator @xmath167 ( @xmath43 ) satisfies the leibnitz rule @xmath168 , we obtain corresponding to proposition 2.1 ( [ cleigen2 ] ) @xmath169\phi_n(q)\phi_m(q)= \{\exp[\hat{a}_r/2\omega]\phi_n(q)\}\ , \{\exp[\hat{a}_r/2\omega]\phi_m(q)\}.\label{mulkform}\ ] ] the classical eigenfunctions of the elementary excitations are the generators of all the classical eigenfunctions . the quantum eigenfunctions of the elementary excitations play a less prominent role . the product of two quantum eigenfunctions is no longer a quantum eigenfunction , since the laplacian @xmath42 and thus the hamiltonian @xmath40 do not enjoy the leibnitz rule . here we will show explicitly the classical and quantum eigenfunctions of the elementary excitations for the calogero systems . the knowledge of the classical equilibrium and the eigenvectors of the hessian matrix @xmath35 ( [ vwmat ] ) helps to determine the classical eigenfunctions . calogero and collaborators discussed the classical equilibrium problem of the @xmath170 calogero system about quarter of a century ago @xcite . a modern version in terms of the prepotential was developed by corrigan - sasaki @xcite . following the usual convention we embed the root vectors in @xmath90 as : @xmath171 the prepotential for the full and _ reduced _ theory read @xmath172 we discuss the reduced theory for simplicity and brevity . the equations ( [ vmin ] ) determining the maximum of the ground state wavefunction @xmath14 read @xmath173 these determine @xmath174 , @xmath175 to be the zeros of the hermite polynomial @xmath176 @xcite , @xmath177 the hessian @xmath178 has eigenvalues @xmath179 , which are exactly the quantum eigenvalues ( divided by @xmath106 ) of the elementary excitations listed in table i. ( the lowest eigenvalue 1 belongs to the center of mass degree of freedom which is completely decoupled from the other modes . ) here are our new results on the classical and quantum eigenfunctions . the @xmath138-th eigenvector of @xmath35 has a simple form @xcite @xmath180 in which @xmath181 is a polynomial of degree @xmath138 of a single variable . they obey the following three term recursion relation : @xmath182 the orthogonality relations of the eigenvectors @xmath183 read simply as @xmath184 these are ` orthogonal polynomials of a discrete variable ' @xcite . in the present case , the discrete variable is obviously the zeros of the hermite polynomial . a simple representation of the elementary excitations is provided by the elementary symmetric polynomials in @xmath142 , @xmath185 , @xmath186 ( [ genfun ] ) which are obviously weyl invariant . by applying the operator @xmath167 ( [ ardef ] ) on the generating function ( [ genfun ] ) @xmath187 and noting @xmath188 and @xmath189 , we obtain @xmath190 through partial fraction decomposition . this leads to @xmath191 we obtain the classical eigenfunctions for the elementary excitations : @xmath192s_k(\{q_j\})=\sum_{l=0}^{[k/2]}{(r+1-k+2l ) ! \over{(r+1-k)!4^ll!}}s_{k-2l}(\{q_j\}).\ ] ] since @xmath193 are harmonic polynomials @xmath194 they are at the same time quantum eigenfunctions : @xmath195 with some calculation one can verify propositions 2.22.3 , that is @xmath196 and its derivative gives the above function @xmath197 , @xmath198 . indeed , as @xmath199 , one finds that @xmath200\nonumber\\ & \equiv & ( -)^{k-1}p_{k-1}(\bar q_j ) . \label{recdetails}\end{aligned}\ ] ] the polynomial @xmath181 of degree @xmath138 ( with @xmath201 ) can not vanish in all the @xmath202 points @xmath203 ( with @xmath204 ) . hence @xmath205 and it follows that @xmath206 is indeed an elementary excitation . as we now know that the expressions ( [ vkpform ] ) are eigenvectors corresponding to different eigenvalues of the matrix @xmath207 , it follows that the @xmath181 are orthogonal polynomials of a discrete variable . hence they obey a three term recurrence relation of type @xmath208 . the coefficients of this recurrence are obtained from the definition ( [ recdetails ] ) for different @xmath138 , _ i.e. _ @xmath209 , @xmath210 , @xmath211 . let us note that the rational @xmath212 and @xmath213 systems are identical with the @xmath214 system . the root vectors of @xmath214 are expressed neatly in terms of an orthonormal basis of @xmath215 as : @xmath216 the prepotential for the full and _ reduced _ theory read @xmath217 we discuss the reduced theory . assuming @xmath218 , the equations ( [ vmin ] ) determining the maximum of the ground state wavefunction @xmath14 read @xmath219 and determine @xmath220 , @xmath59 , as the zeros of the associated laguerre polynomial @xmath221 , @xcite , @xmath222 . the hessian @xmath178 has eigenvalues @xmath223 , which are exactly the quantum eigenvalues ( divided by @xmath106 ) of the elementary excitations listed in table i. the new results on the classical and quantum eigenfunctions are as follows . the @xmath138-th eigenvector of @xmath35 has a simple form @xmath224 in which the polynomials @xmath225 obey the following three term recursion relation : @xmath226 the orthogonality relations of the eigenvectors @xmath183 again correspond to those of orthogonal polynomials of a discrete variable : @xmath227 a simple representation of the elementary excitations is provided by the elementary symmetric polynomials in @xmath228 , @xmath229 , @xmath230 , which are obviously weyl invariant . by applying the operator @xmath167 ( [ ardef ] ) and the laplacian @xmath42 on the generating function ( [ genfun ] ) @xmath231 , we obtain @xmath232 for the former formula , as in the @xmath125-series case , the partial fraction decomposition is used . these mean @xmath233 from which we obtain the classical eigenfunctions for the elementary excitations : @xmath234s_k(\{q_j^2\})=\sum_{l=0}^{k}(-)^l { r+l - k\choose l}{\gamma(r+l - k+\gamma ) \over{\gamma(r - k+\gamma)}}s_{k - l}(\{q_j^2\ } ) . \nonumber\end{aligned}\ ] ] the corresponding quantum eigenfunctions have very similar forms , since the action of @xmath235 on @xmath236 ( [ braonsk ] ) , ( [ brtrionsk ] ) is the same as that of @xmath43 with @xmath237 replaced by @xmath238 : @xmath239s_k(\{q_j^2\})=\sum_{l=0}^{k}(-)^l { r+l - k\choose l}{\gamma(r+l - k+\gamma+\hbar/2 ) \over{\gamma(r - k+\gamma+\hbar/2)}}s_{k - l}(\{q_j^2\ } ) . \nonumber\end{aligned}\ ] ] the correspondence between the ( classical ) eigenfunctions ( [ brclas ] ) , ( [ brquant ] ) and the classical eigenvectors ( [ brosc ] ) ( and ( [ bprec1 ] ) , ( [ bprec ] ) ) is established in an identical manner as in the @xmath125-case . the root vectors of @xmath240 are : @xmath241 the prepotential for the full and _ reduced _ theory read @xmath242 we discuss the reduced theory . the equilibrium position @xmath243 are the zeros of the laguerre polynomial @xmath244 , that is one of @xmath174 s is vanishing @xcite let us choose @xmath245 . the hessian @xmath178 has eigenvalues @xmath246 , which are exactly the quantum eigenvalues ( divided by @xmath106 ) of the elementary excitations listed in table i. the @xmath138-th eigenvector of @xmath35 has a simple form @xmath247 in which the polynomials @xmath225 obey the following three term recursion relation : @xmath248 which is a special case of that of @xmath214 with @xmath249 . the classical and quantum eigenfunctions @xmath250 , , @xmath251 and @xmath252 , , @xmath253 have the same form as in @xmath214 case ( [ brclas ] ) , ( [ brquant ] ) with @xmath249 . the special elementary excitation of the @xmath240 theory belongs to the @xmath10-th eigenvalue . the classical eigenfunction is the same as the quantum one @xmath254 for the @xmath127-series of root systems we consider only the reduced theory , that is @xmath255 . the roots are normalised @xmath256 , for all root systems @xmath257 and @xmath258 . the generating function is defined in terms of the weights belonging to the * 27 * dimensional representation : @xmath260 these are _ minimal weights _ having the properties @xmath261 for @xmath262 and @xmath263 by applying the operator @xmath167 ( [ ardef ] ) and using the above properties of the weight vectors , we obtain @xmath264 or in terms of the @xmath236 , @xmath265 some lower members of @xmath236 , which depend on @xmath266 are @xmath267 the elementary excitations are for @xmath2682 , 5 , 6 , 8 , 9 , 12 , ( see table i ) for which @xmath269 s are functionally independent . the corresponding quantum eigenfunctions are : @xmath270,\\ \phi_8&=&s_8+(57 + 13\hbar)s_6-{15/2}(1-\hbar)(q^2)^3\\ & & \hspace*{2mm}+{5(6+\hbar)(2217 + 313\hbar)/24}\left [ 3(q^2)^3 - 8(9+\hbar)q^2 + 6(9+\hbar)(12+\hbar)\right],\\ \phi_9&=&s_9-{35/6}(15 + 4\hbar ) s_5\times[q^2 - 2(9 + 2\hbar)],\\[6pt ] \phi_{12}&= & s_{12 } + \frac{\left(1725534 + 1146267\hbar + 188608\hbar^2\right ) } { 6144\left(3 + \hbar \right ) \left ( 36 + 11\hbar \right)}(q^2)^6 + \frac{23}{576 } s_5 ^ 2q^2 + \frac{101\,}{192}s_6 ^ 2\nonumber\\ & & + \frac{\left ( 222939 + 146112\hbar + 23798\hbar^2 \right ) \ , } { 960\,\left ( 3 + \hbar \right ) \ , \left ( 36 + 11\hbar \right)}(q^2)^3s_6 - \frac{\left(3216 + 937\hbar \right ) \ , } { 1920\,\left ( 3 + \hbar \right)}(q^2)^2s_8\end{aligned}\ ] ] it should be noted that @xmath271 is a classical and quantum eigenfunction . the twelfth eigenfunction @xmath272 is chosen to be a homogeneous one . the generating function is defined in terms of the weights belonging to the * 56 * dimensional representation : @xmath274 these are _ minimal weights _ having the properties @xmath261 for @xmath275 and @xmath276 moreover @xmath277 is _ even _ , _ i.e. _ if @xmath278 then @xmath279 . this is why the odd order terms in @xmath162 vanish in ( [ e7calgen ] ) . by applying the operator @xmath167 ( [ ardef ] ) and using the above properties of the weight vectors , we obtain @xmath280 or in terms of the @xmath281 , @xmath282 some lower members of @xmath283 are @xmath284 the elementary excitations are for @xmath2682 , 6 , 8 , 10 , 12 , 14 , 18 , ( see table i ) for which @xmath269 s are again functionally independent . the lower members of the corresponding quantum eigenfunctions are : @xmath285,\\ \phi_8&=&s_8+(555/2 + 165\hbar/4)s_6 - 315(1-\hbar)(q^2)^3\\ & & \quad + ( 536948 + 102416\hbar+4851\hbar^2)45/32 \left [ ( q^2)^2 - 3(14+\hbar)(q^2 - 7(18+\hbar)/8)\right].\nonumber\end{aligned}\ ] ] we have not succeeded in deriving an equation for a generating function similar to ( [ acalgeneq ] ) , ( [ bcalgeneq ] ) , ( [ e6calgeneq ] ) , ( [ e7calgeneq ] ) . we start from a weyl invariant power sum basis in terms of roots @xmath286 for those eight @xmath138 s listed in the @xmath258 row of table i. they are functionally independent . for example , @xmath287 . although the derivation of the classical and quantum eigenfunctions for elementary excitations is straightforward , some results are too lengthy to present . for want of proper and convenient notation , we show only some lower members of the classical eigenfunctions : @xmath288 verification of the propositions [ prop21][prop23 ] in section two is easy but tedious calculation . the long roots of @xmath133 are the roots of @xmath289 @xmath290 and the short roots are the union of vector , spinor and anti - spinor weights of @xmath289 : @xmath291 we will consider a reduced theory in which @xmath292 and the short root coupling is denoted by @xmath136 : @xmath293 let us introduce the elementary symmetric polynomials in @xmath294 , as in the @xmath131 ( @xmath126 ) series ( [ genfun ] ) : @xmath295 which are not weyl invariant , except for @xmath296 . a weyl invariant basis for the elementary excitations are for degree 2 , 6 , 8 and 12 ( see table i ) polynomials @xcite : @xmath297 the quantum eigenfunctions for the elementary excitations are : @xmath298 , \\ \phi_8&=&\phi_8+(3+\hbar)\,\phi_6+(3+\hbar ) \left(2(2+\gamma)+\hbar\right)/8\,\phi_2 ^ 2\nonumber\\ & & - ( 3+\hbar)\left(2(2+\gamma)+\hbar\right)\left(4(1+\gamma)+\hbar\right)/8\left[{\vphantom{\mbox{\oneeight i}}}2\phi_2- \left(6(1+\gamma)+\hbar\right)\right],\\ \phi_{12}&=&\phi_{12}+(3 + 2\hbar)\left({\vphantom{\mbox{\onefour i}}}2\phi_8+(3+\hbar)\phi_6\right)\!/2\,\phi_2 -(3 + 2\hbar)\left(6(1+\gamma)+5\hbar\right)\phi_8\nonumber\\ & & + ( 3+\hbar)(3 + 2\hbar)\left(2(2+\gamma)+\hbar\right)/24\phi_2 ^ 3- ( 3+\hbar)(3 + 2\hbar)\left(4(1+\gamma)+3\hbar\right)\phi_6\nonumber\\ & & -(3+\hbar)(3 + 2\hbar)\left(2(1+\gamma)+\hbar\right)\left(2(2+\gamma)+\hbar\right)/16\ ! \nonumber\\ & & \hspace*{15mm}\times \left[{\vphantom{\mbox{\oneeight i}}}5\phi_2 ^ 2 - 6\left(4(1+\gamma)+\hbar\right)\phi_2 + 2\left(4(1+\gamma)+\hbar\right ) \left(6(1+\gamma)+\hbar\right)\right].\end{aligned}\ ] ] the dihedral group of order @xmath103 , @xmath299 , is the group of orthogonal transformations that preserve a regular @xmath300-sided polygon in two dimensions . if all the roots are chosen to have the same length @xmath301 , they can be parametrised as : @xmath302 for odd @xmath300 all of the roots are in the same orbit of the reflection group but for even @xmath300 there are two orbits , one consisting of the @xmath303 with odd @xmath304 and the other with even @xmath304 . thus the @xmath299 calogero system has one coupling constant @xmath305 for odd @xmath300 and two couplings @xmath306 and @xmath307 for even @xmath300 on top of the frequency @xmath308 of the harmonic confining potential . the complete set of quantum eigenfunctions ( @xmath309 ) are given for all rank two calogero systems in @xcite , with @xmath310 , @xmath311 and @xmath312 . so we concentrate on the elementary excitations with explicit @xmath3 dependence . the coxeter invariant polynomials exist at degree 2 and @xmath300 @xcite : @xmath313 where @xmath314 s are @xmath315 if we introduce the two - dimensional polar coordinates system and the rank of the root system @xmath10 , which in this case is 2 of @xmath316 . ] for @xmath317 , @xmath318 the two coxeter invariant polynomial variables read @xmath319 the essential part of the quantum eigenfunctions are a laguerre polynomial in @xmath320 , ( [ laguerrepoly ] ) times a jacobi polynomial in @xmath321 , thus the separation of variables is achieved . as above , let us consider the reduced theory , @xmath255 for odd @xmath300 and @xmath322 and @xmath323 for even @xmath300 . we have @xmath324 and @xmath325 thus @xmath326 is a classical and quantum eigenfunction for odd @xmath300 . for even @xmath300 we have @xmath327 which can be expressed as the jacobi polynomial of degree one @xcite . the classical equilibrium point is @xmath328 for odd and even @xmath300 , respectively . verification of proposition [ prop22 ] is straightforward . the non - crystallographic coxeter groups of @xmath100 and @xmath101 are the symmetry groups of the icosahedron and four - dimensional 600-cell , respectively . the former consists of 30 roots and the latter 120 . let us start from coxeter invariant power sum bases in terms of roots @xmath329 the quantum eigenfunctions of the elementary excitations in @xmath330 are : @xmath331,\\ \phi_{10}&=&\phi_{10}-{(215 + 126\hbar)\over12}\left[4q^2 - 15(2+\hbar)\right]\phi_6 + { 5\over6}(58 + 63\hbar)\,(q^2)^4+{75\over4}(109 + 45\hbar)\,(q^2)^3\nonumber\\ & & -{25\over32}(10 + 3\hbar)(770 + 454\hbar+63\hbar^2 ) \left[{\vphantom{\mbox{\twozero i}}}8(q^2)^2-(6+\hbar)\left({\vphantom{\mbox{\onefour i}}}10q^2 - 3(10+\hbar)\right)\right].\end{aligned}\ ] ] some of the quantum eigenfunctions of the elementary excitations in @xmath332 are : @xmath333.\end{aligned}\ ] ] the other two eigenfunctions are too lengthy to be reported . as shown in [ roots ] the eigenstates of the sutherland system are specified by _ dominant highest weight _ @xmath119 . the basis of the classical and quantum eigenfunctions are thus the sum of the exponentials of @xmath334 taken for the entire weyl orbit of @xmath119 , @xmath335 , which we denote as @xmath336 as usual , if the multiplicity @xmath337 is vanishing @xmath338 , it is not written . for example , the basis corresponding to the fundamental weights @xmath339 , , @xmath340 are @xmath341 , @xmath342 , , @xmath343 : @xmath344 the operator @xmath43 is lower triangular and the laplacian @xcite is diagonal in this basis : @xmath345 as shown below there is a marked difference in the forms of the eigenfunctions between the classical ( @xmath125 , @xmath131 , @xmath132 and @xmath126 ) and the exceptional ( @xmath127 , @xmath133 and @xmath134 ) root systems . the elementary excitations of the @xmath125-series sutherland system can have the same classical and quantum eigenfunctions as in the calogero case . the classical equilibrium of the @xmath170 sutherland system is rather trivial @xcite . the prepotential for the full and _ reduced _ theory read @xmath346 the equations ( [ vmin ] ) determining the maximum of the ground state wavefunction @xmath14 read @xmath347=0 , \quad j=1,\ldots , r+1,\ ] ] which are satisfied by the _ equally - spaced _ configuration @xmath348 . the hessian @xmath178 has eigenvalues @xmath349 , which can be expressed as @xmath350 with @xmath122 defined in ( [ weylvec ] ) . in these formulas the trivial eigenvalue 0 , coming from the translational invariance , is removed . the @xmath138-th eigenvector of @xmath35 is simply @xmath351 . the orthogonality condition of the eigenvectors @xmath183 read simply as @xmath352 . let us introduce a generating function @xmath353 it is easy to see that the symmetric polynomial @xmath236 is equal to the basis @xmath354 ( [ psidefs ] ) up to a term proportional to the center of mass " @xmath355 which is orthogonal to all the @xmath170 roots . the generating function @xmath356 satisfies @xmath357 which translate into @xmath358 therefore @xmath359 is a classical and quantum eigenfunction of the @xmath138-th elementary excitation with eigenvalues @xmath360 and @xmath361 of @xmath43 ( [ cleigen ] ) and @xmath40 ( [ htilform ] ) , respectively . the absence of quantum corrections is a general property shared by eigenfunctions belonging to _ minimal weights _ all the fundamental representations of the a - series root systems are minimal . since the @xmath131 and @xmath132 root systems are closely related , @xmath362 for @xmath363 , many formulas for the eigenfunctions etc take similar forms . it is advantageous to write these expressions in parallel so that the similarity and differences can be well appreciated . the prepotentials for the _ reduced _ theory read @xmath364 the equations ( [ vmin ] ) determining the maximum of the ground state wavefunction @xmath14 read @xmath365 & & \hspace*{24mm}j=1,\ldots , r,\nonumber\\[-8pt ] \sum_{k\ne j}^r{1\over{\bar{x}_j-\bar{x}_k}}+ { \gamma\over{2}}{1\over{\bar{x}_j-1}}+{\gamma\over{2}}{1\over{\bar{x}_j+1}}&=&0 , \hspace*{34 mm } \mbox{(c)},\end{aligned}\ ] ] for @xmath366 . they determine @xmath367 as the zeros of jacobi polynomials @xcite : @xmath368 because of the identity @xmath369 the ( b ) case always has one zero at @xmath370 . let us choose @xmath371 . the hessian @xmath178 has eigenvalues @xmath372 in which the last one of the b - series belongs to the spinor representation ( @xmath340 ) . the @xmath138-th eigenvector of @xmath35 has a form @xmath373 in which the polynomials @xmath374 of a single variable @xmath162 obey the three term recursion relations : @xmath375 for the b - series and @xmath376 for the c - series . the orthogonality conditions for these discrete variable polynomials are @xmath377 with @xmath367 being the zeros of a jacobi polynomial . let us introduce a generating function @xmath378 it is easy to see that the symmetric polynomial @xmath236 is proportional to the basis @xmath354 ( [ psidefs ] ) : @xmath379 the generating function satisfies @xmath380 these mean in turn @xmath381 the quantum eigenfunctions for the elementary excitations are : @xmath382 } \frac{(r+2l - k)!\gamma(r - k+\gamma+\hbar/2 + 1/2)}{4^l l ! ( r - k ) ! \gamma(r - k+\gamma+\hbar/2+l+1/2)}s_{k-2l } , \quad k=1,\ldots , r,\quad \mbox{(c)}.\end{aligned}\ ] ] the corresponding eigenvalues are : @xmath383 the eigenfunction for the spinor weight in b - series @xmath384 ( [ sutbspin ] ) has no quantum corrections . the representations ( [ sutbv1 ] ) , ( [ sutbv2 ] ) and the recursions ( [ sutbprec1])([sutcprec1 ] ) are obtained from these eigenfunctions . the reduced prepotential of the d - series sutherland system is obtained by removing the short ( long ) root coupling @xmath237 terms from those of the b- and c- series ( [ rwbsut ] ) , ( [ rwcsut ] ) . this results in the change of the eigenvectors @xmath183 ( [ sutbv1 ] ) , ( [ sutbv2 ] ) @xmath385 ( [ sutdv1 ] ) , ( [ sutdv2 ] ) and emergence of another eigenfunction associated with anti - spinor weights @xmath253 ( [ sutdspin ] ) which receives no quantum corrections . the equations ( [ vmin ] ) determining the maximum of the ground state wavefunction @xmath14 has a solution @xmath245 , @xmath386 and with @xmath366 , @xmath387 being the zeros of the jacobi polynomial @xmath388 or equivalently of the gegenbauer polynomial @xmath389 : @xmath390 the hessian @xmath178 has eigenvalues @xmath391,\end{aligned}\ ] ] in which the exceptional one is doubly degenerate corresponding to the ` fish tail ' of the @xmath126-series dynkin diagram . the corresponding eigenvectors of @xmath35 are @xmath392 the polynomials @xmath374 of a single variable @xmath162 obey simple three term recursion relations : @xmath393 the generating function has the same form as in the b , c cases : @xmath378 it is easy to see that the symmetric polynomial @xmath236 is proportional to the basis @xmath354 ( [ psidefs ] ) and that the two additional bases are : @xmath394 they satisfy @xmath395 these imply for @xmath236 : @xmath396 thus we arrive at the quantum eigenfunctions corresponding to the elementary excitations : @xmath397 } \frac{(r+2l - k)!\gamma(r - k+\hbar/2 + 1/2)}{4^l l ! ( r - k ) ! \gamma(r - k+\hbar/2+l+1/2)}s_{k-2l } , \quad k=1,\ldots , r-2,\\ \phi_{r-1}&=&\prod_{j=1}^r\sin q_j,\qquad \phi_r=\prod_{j=1}^r\cos q_j , \label{sutdspin}\end{aligned}\ ] ] with the eigenvalues of @xmath40 @xmath398.\ ] ] again , the eigenfunctions corresponding to the spinor and anti - spinor weights receive no quantum corrections . these are minimal weights . for the sutherland systems based on exceptional root systems , e , f and g , the method of the generating functions seems not so useful as in the classical root systems cases , because of the ` exceptional ' character . the equilibrium points of the potentials are not related to known classical polynomials in contrast to the cases discussed above . new polynomials describing the equilibria were introduced by odake and sasaki @xcite . here we will construct the eigenfunctions corresponding to the fundamental weights ( elementary excitations ) starting from the basis @xmath341 , , @xmath343 ( [ psidefs ] ) . there is no universally accepted way of naming the simple roots and fundamental weights of the exceptional root systems . we show our conventions in terms of the dynkin diagrams . the symmetry of the dynkin diagram is reflected in the structure of the eigenvalues and eigenfunctions , too . dynkin diagram with the numbers of the simple roots attached . ] the spectra of the hessian @xmath178and the corresponding @xmath40 in the order of @xmath399 are : @xmath400 the quantum eigenfunctions are listed in the order of increasing energy eigenvalues and the values of the @xmath401 : @xmath402 the orbits in sub - leading terms of an eigenfunction are contained in the _ lie algebra representation _ specified by the dominant weight of the leading term . constant terms correspond to zero weights . for example , the highest weight representation specified by @xmath403 ( [ e64 ] ) consists of the weyl orbits of @xmath403 , @xmath404 , @xmath405 and some zero weights . the lowest two eigenfunctions ( [ e6mini ] ) consist of single orbits belonging to the * 27 * and @xmath406 representations , which are minimal . thus they do not receive quantum corrections . they corresponds to the left and right ends of the diagram , fig.[fig : e6 ] . the fundamental weight @xmath405 ( [ e6roots ] ) corresponds to the adjoint representation , containing all the roots and the rank number of zero weights . the constant term in ( [ e6roots ] ) , @xmath407 reflects the number of roots 72 and the highest exponent 11 which is the ` _ height _ ' of the _ highest root _ , that is @xmath405 in the present case @xcite . the longer the dominant weight @xmath408 becomes , the more complicated structure has the corresponding eigenfunction . these are common features of all the eigenfunctions of the sutherland systems . the spectra of the hessian @xmath178 and the corresponding @xmath40 are : @xmath409 which have no degeneracy . the dynkin diagram has no symmetry . dynkin diagram with the numbers of the simple roots attached . ] the corresponding quantum eigenfunctions and the values of the @xmath401 are : @xmath410 the first corresponds to the * 56 * dimensional representation which is minimal . it has no @xmath3 dependence . the second corresponds to the set of roots ( adjoint representation ) with 126 roots and the highest exponent being 17 . the last expression ( [ e74 ] ) is much longer than its classical counterpart @xmath411 for the @xmath258 eigenfunctions , we will present the classical ones simply because of the lack of space . the spectra of the hessian @xmath178 and the corresponding @xmath40 in the order of @xmath412 are : @xmath413 which has no degeneracy . the dynkin diagram has no symmetry . dynkin diagram with the numbers of the simple roots attached . ] the classical eigenfunctions for six lower elementary excitations are : @xmath414 most of the denominators contain the exponents of @xmath258 , @xmath415 . this is a common feature shared by all the root systems but seen most clearly in the exceptional root systems cases . the two unlisted eigenfunctions @xmath416 and @xmath417 have simply too many terms to be presented here . for @xmath416 , @xmath418 , the number of elements in the weyl orbit of @xmath419 is 241920 and the highest weight representation of @xmath419 is 146325270 dimensional . the eigenfunction @xmath416 contains 14 terms corresponding to the dominant characters in the lie algebra representation of the highest weight @xmath419 , that is @xmath419 , @xmath420 , @xmath421 , @xmath422 , @xmath423 , @xmath424 , @xmath425 , @xmath426 , @xmath427 , @xmath405 , @xmath428 , @xmath339 , @xmath429 and zero weights . for @xmath417 , @xmath430 , the number of elements in the weyl orbit of @xmath403 is 483840 and the highest weight representation of @xmath403 is 6899079264 dimensional . the eigenfunction @xmath417 contains 24 terms and some of their coefficients are ratios of enormously large integers . the sutherland systems based on @xmath133 and @xmath134 are interesting because of the interplay of the long and short root couplings . while we show the dynkin diagram of @xmath133 to indicate our convention of the simple roots naming , we simply agree that @xmath431 is the short simple root of @xmath134 , thus @xmath432 is the long simple root . dynkin diagram with the numbers of the simple roots attached . ] the spectra of the hessian @xmath178 and the corresponding @xmath40 of @xmath133 in the order of @xmath433 are : @xmath434 the quantum eigenfunctions are listed in the order of increasing energy eigenvalues and the values of the @xmath401 : @xmath435 \left[{\vphantom{\mbox{\onefour i}}}(2+\gamma)(3+\gamma)(4 + 3\gamma)\right]^{-1}\psi_1 \nonumber\\ & & \ \qquad\qquad \ + 24\gamma\left[{\vphantom{\mbox{\onefour i}}}28 + 37\gamma+27\gamma^2\right ] \left[{\vphantom{\mbox{\onefour i}}}(2+\gamma)(3+\gamma)(4 + 3\gamma)\right]^{-1}\psi_4 \nonumber\\ & & \ + 96\left[{\vphantom{\mbox{\onefour i}}}24 + 30\gamma+85\gamma^2 + 67\gamma^3 + 30\gamma^4\right]\!\ ! \left[{\vphantom{\mbox{\onefour i}}}(2+\gamma)(3+\gamma)(4 + 3\gamma)(3 + 2\gamma)\right]^{-1}. \label{f42}\end{aligned}\ ] ] here we listed the classical eigenfunction @xmath436 ( [ f42 ] ) for the highest elementary excitation , simply for display reasons . the spectra of the hessian @xmath178 and the corresponding @xmath40 of @xmath134 in the order of @xmath339 , @xmath405 are : @xmath437 the quantum eigenfunctions are listed in the order of increasing energy eigenvalues and the values of the @xmath401 : @xmath438 the general theorem relating classical and quantum mechanics ( section [ qmechanics ] ) is applied to the calogero and sutherland systems , typical integrable multi - particle dynamics associated with root systems and having long range interactions . the classical and quantum eigenfunctions for the elementary excitations are constructed explicitly ( section [ rationals ] , [ trigs ] ) , and their relation to the eigenmodes of small oscillations ( of the corresponding classical system ) is worked out in full . in particular , we obtain new representations for the eigenvectors ( of small oscillations ) in terms of orthogonal polynomials of a discrete variable ( the discrete variable being the zeros of well - known classical polynomials ) . it turns out that the quantum eigenfunctions are very closely related to the classical counterparts . as a special case , the quantum eigenfunction of the sutherland system belonging to a minimal representation consists of a single weyl orbit and it has exactly the same form as the classical one , that is the quantum corrections are absent . the next simplest case , those belonging to the adjoint representations is fully described by the number of roots and the highest exponents . as shown in many explicit examples , the classical and quantum eigenfunctions are fully described in terms of the roots , weights , exponents and characters , etc . we do believe that this is the case for any eigenfunctions of the calogero and sutherland systems . to demonstrate this assertion for any particular theory and universally for all the calogero and sutherland systems is a good challenge . i.l . is a post - doctoral fellow with the f.w.o .- vlaanderen ( belgium ) . f. calogero , solution of the one - dimensional @xmath115-body problem with quadratic and/or inversely quadratic pair potentials " , j. math . * 12 * ( 1971 ) 419 - 436 ; b. sutherland , `` exact results for a quantum many - body problem in one - dimension . ii '' , phys * a5 * ( 1972 ) 1372 - 1376 ; j. moser , `` three integrable hamiltonian systems connected with isospectral deformations '' , adv . math . * 16 * ( 1975 ) 197 - 220 . olshanetsky and a.m. perelomov , completely integrable hamiltonian systems connected with semisimple lie algebras " , inventions math . * 37 * ( 1976 ) , 93 - 108 ; `` classical integrable finite - dimensional systems related to lie algebras '' , phys . rep . * c71 * ( 1981 ) , 314 - 400 . e.corrigan and r.sasaki , quantum vs classical integrability in calogero - moser systems " , j. phys . a * 35 * ( 2002 ) 7017 - 7062 , hep - th/0204039 ; s. odake and r. sasaki , `` polynomials associated with equilibrium positions in calogero - moser systems , '' j. phys . a * 35 * ( 2002 ) 8283 - 8314 , hep - th/0206172 ; o. ragnisco and r. sasaki , quantum vs classical integrability in ruijsenaars - schneider systems " , preprint yitp-03 - 09 , hepth/0305120 . g. j. heckman and e. m. opdam , `` root systems and hypergeometric functions i '' , comp . math . * 64 * ( 1987 ) , 329352 ; g. j. heckman , `` root systems and hypergeometric functions ii '' , comp . math . * 64 * ( 1987 ) , 353373 ; e. m. opdam , `` root systems and hypergeometric functions iii '' , comp . math . * 67 * ( 1988 ) , 2149 ; `` root systems and hypergeometric functions iv '' , comp . math . * 67 * ( 1988 ) , 191209 . r. stanley , some combinatorial properties of jack symmetric function `` , adv . math . * 77 * ( 1989 ) 76 - 115 ; i.g . macdonald , symmetric functions and hall polynomials '' , second edition , oxford university press , 1995 . k. sogo , a simple derivation of multivariable hermite and legendre polynomials " , j. phys . jpn * 65 * ( 1996 ) 3097 - 3099 ; n. gurappa and p. k. panigrahi , equivalence of the calogero - sutherland model to free harmonic oscillators " , phys . rev . * b59 * ( 1999 ) r2490-r2493 , cond - mat/9710035 ; n. gurappa and p. k. panigrahi , free harmonic oscillators , jack polynomials , and calogero - sutherland systems " , phys . rev . * b62 * ( 2000 ) 1943 - 1949 , hep - th/9910123 . gambardella , exact results in quantum many - body systems of interacting particles in many dimensions with @xmath439 as the dynamical group " , j. math * 16 * 1172 - 1187 ( 1975 ) . f. calogero , `` on the zeros of the classical polynomials '' , lett . nuovo cim . * 19 * ( 1977 ) 505 - 507 ; equilibrium configuration of one - dimensional many - body problems with quadratic and inverse quadratic pair potentials " , lett . nuovo cim . * 22 * ( 1977 ) 251 - 253 ; eigenvectors of a matrix related to the zeros of hermite polynomials " , lett . nuovo cim . * 24 * ( 1979 ) 601 - 604 ; `` matrices , differential operators and polynomials '' , j. math . phys . * 22 * ( 1981 ) 919 - 934 . f. calogero and a.m. perelomov , `` properties of certain matrices related to the equilibrium configuration of one - dimensional many - body problems with pair potentials @xmath440 and @xmath441 '' , commun . * 59 * ( 1978 ) 109 - 116 . a.erdlyi et al . higher transcendental functions " ii , mcgraw - hill , new york , 1953 ; a.f.nikiforov , s.k.suslov and v.b . uvarov , classical orthogonal polynomials of a discrete variable " , springer , berlin 1991 .

fusion data forc+c ,c+c andc+c are analyzed by coupled - channels calculations that are based on the m3y+repulsion , double - folding potential . the fusion is determined by ingoing - wave - boundary conditions ( iwbc ) that are imposed at the minimum of the pocket in the entrance channel potential . quadrupole and octupole transitions to low - lying states in projectile and target are included in the calculations , as well as mutual excitations of these states . the effect of one - neutron transfer is also considered but the effect is small in the measured energy regime . it is shown that mutual excitations to high - lying states play a very important role in developing a comprehensive and consistent description of the measurements . thus the shapes of the calculated cross sections forc+c andc+c are in good agreement with the data . the fusion cross sections forc+c determined by the iwbc are generally larger than the measured cross sections but they are consistent with the maxima of some of the observed peak cross sections . they are therefore expected to provide an upper limit for the extrapolation into the low - energy regime of interest to astrophysics .

1112.0496

the fusion of carbon nuclei is an important reaction in the description of type ia supernovae and other astronomical events in the cosmos like the superburst of an accreting neutron star . there is , unfortunately , a very large uncertainty in the predicted fusion cross sections for @xmath1c+@xmath1c that are needed in a stellar environment @xcite , partly because the cross sections are difficult to measure down to the low energies that are of interest , and partly because the data contain strong resonance structures that make it difficult to extrapolate the measured cross sections to low energy . in order to get some constraints on the extrapolation it is instructive to analyze the existing fusion data for @xmath0c+@xmath0c by trentalange et al . @xcite and charvet et al . @xcite , and also of the @xmath1c+@xmath0c data by dayras et al . @xcite and notani et al . @xcite , because these data do not exhibit strong resonance features . the @xmath0c+@xmath0c fusion data @xcite are analyzed by coupled - channels calculations that include couplings to the one - phonon quadrupole and octupole excitations in projectile and target , as well as mutual excitations of these states . the effect of one - neutron transfer , which can proceed to the @xmath1c+@xmath2c mass partition with positive @xmath3 value is also studied , and so is the effect of the neutron exchange with zero q value in @xmath1c+@xmath0c reactions . the fusion cross sections are determined by ingoing - wave - boundary conditions ( iwbc ) that are imposed at the minimum of the pocket in the entrance channel potential . the calculated cross sections defined this way are fairly smooth as functions of the center - of - mass energy and they are well suited to analyze the fusion data for @xmath0c+@xmath0c @xcite and @xmath1c+@xmath0c @xcite . the @xmath1c+@xmath1c fusion data , on the other hand , contain a lot of structures or resonances , and it is beyond the scope of this investigation to try to reproduce these data in detail . it is , however , of great interest to see how the calculated cross sections , obtained from the iwbc , compare to the data and how they possibly can be used to put constraints on the extrapolation to very low energies . the coupled equations are solved using either a standard woods - saxon @xcite or the m3y+repulsion , double - folding potentials @xcite . the coupled - channels effects on the calculated fusion cross sections are relatively modest compared to the large enhancement of several orders of magnitude that are commonly seen in calculations of heavy - ion fusion reactions . for the carbon systems , the enhancement of fusion compared to the no - coupling limit is typically a factor of 2 at energies far below the coulomb barrier . however , it is a challenge to explain the data in detail because the coupled - channels calculations are sensitive to channels that have rather high excitation energies . one problem is that these channels are closed at low beam energies but that problem can be solved by imposing the correct , decaying state boundary conditions at large distances between the reacting nuclei @xcite . another problem is that the nuclear structure of high - lying states is sometimes very uncertain . the m3y+repulsion potential was introduced in ref . @xcite to explain the hindrance that has been observed in the fusion of many heavy - ion systems at very low energies @xcite . the hindrance phenomenon was first observed as a suppression at very low energies of @xmath4ni+@xmath5y fusion data compared to calculations that were based on a standard woods - saxon potential @xcite . a simple explanation of the phenomenon is the existence of a shallow pocket in the entrance channel potential which forces the fusion cross section to vanish as the center - of - mass energy approaches the minimum energy of the pocket @xcite . important issues have been whether the fusion hindrance also occurs in light - ion systems , and how it will affect the extrapolation of measured cross section to the low energies that are of interest to astrophysics @xcite . the evidence for quasi - molecular resonances , observed in the elastic scattering @xcite and fusion reactions @xcite of @xmath1c+@xmath1c are usually explained as resonances in a shallow two - body potential between the reacting nuclei . we will show that the analysis of the @xmath0c+@xmath0c and @xmath1c+@xmath0c fusion data supports the idea of a fusion hindrance and the existence of a shallow pocket in the entrance channel potential . the fusion cross sections reported in refs . @xcite are based on measurements of the characteristic @xmath6-rays emitted from some of the evaporation residues that are produced , mostly those associated with the proton , neutron , and @xmath7 decay . the total fusion cross sections were obtained with the aid of statistical model calculations which is a major source of the systematic uncertainty . the systematic error is difficult to estimate but it can be quite large . for example , the systematic error of the absolute cross sections quoted in ref . @xcite is @xmath815% . since some of the systematic error concerns the overall normalization of the measured cross sections , and not so much the shape ( i. e. , the energy dependence of the cross section ) , it is of interest to adopt an adjustable overall normalization when analysing the data . the ingredients of the coupled - channels calculations are presented in the next section . the analysis of the @xmath0c+@xmath0c fusion data is discussed in section iii , where the optimum parameters of the m3y+repulsion potential , such as the radius parameter of @xmath0c and the diffuseness associated with the repulsive term , are determined . these parameters are used together with a parameterization of the experimental density of @xmath1c as input to the coupled - channels calculations of the fusion cross sections for @xmath1c+@xmath0c and @xmath1c+@xmath1c which are presented in sections iv and v , respectively . the systematics of the low - energy fusion cross sections for the 3 systems of carbon isotopes is discussed in section vi . finally , the conclusions of this work are presented in section vii . the ingredients in the coupled - channels calculations include the ion - ion potential , the nuclear structure input , the strength and q - value of the transfer reactions , and the definition of the fusion cross section . the coupled - channels technique that is used has been applied previously and is described , for example , in refs . all of the details will therefore not be repeated here . it is emphasized that the calculations are performed in the rotating frame approximation @xcite which makes it feasible to include the most important reaction channels in the calculations . in this approximation , the magnetic quantum number @xmath9 of the entrance channel is conserved . in reactions between nuclei with @xmath10 ground states , this implies that the @xmath9 quantum number remains zero . in reactions with odd nuclei one would have to repeat the calculations for each initial value of @xmath9 and the average fusion cross section should be compared to the measurements . in the fusion of @xmath1c+@xmath0c , for example , the ground state of @xmath1c is a @xmath10 state and @xmath0c has a @xmath11 ground state . however , the cross sections for @xmath9 = -1/2 and + 1/2 are identical so it is sufficient to calculate the cross section once . besides the commonly used woods - saxon potential , the ion - ion potential that will be used in this paper is the so - called m3y+repulsion potential @xcite . it consists of the conventional m3y potential and a repulsive term which is discussed below . the m3y double - folding potential is defined as @xmath12 where @xmath13 is the m3y effective nucleon - nucleon interaction derived from the reid potential @xcite . the densities @xmath14 of the @xmath1c and @xmath0c carbon isotopes are parametrized in terms of the symmetrized fermi functions defined in the appendix of ref . one advantage of this parametrization is that the mean square radius is given by the simple expression , @xmath15 in terms of the radius @xmath16 and the diffuseness @xmath17 . the diffuseness of the densities for both nuclei is set equal to 0.44 fm so the only parameters that need to be specified are the matter radii of the reacting nuclei . this choice of the diffuseness is in good agreement with the tail of the measured point - proton density of @xmath1c @xcite which is illustrated in fig . the calculated point - proton distribution has the radius @xmath16 = 2.155 fm consistent with the measured rms charge radius . the radii that reproduce the point - proton ( pp ) rms radii of @xmath1c and @xmath0c , extracted from the measured root - mean - square ( rms ) charge radii @xcite , are shown in the last few lines of table i. the quoted radius for @xmath1c is expected to be a realistic estimate of the matter radius of @xmath1c and it will therefore be used in calculations of the m3y double - folding potentials . the matter radius of @xmath0c is possibly larger than the radius of @xmath1c because of the valence neutron . that is what one would expect by adding a neutron to an inert @xmath1c core nucleus . a simple estimate based on a single - particle , woods - saxon potential gives a mean square radius of @xmath18 = 8.655 @xmath19 for the @xmath20 valence neutron . combined with the point - proton rms radius of @xmath1c , @xmath21 = 5.462(9 ) @xmath19 , one can now estimate the mean square matter radius @xmath22 of @xmath0c by @xmath23 the result is a rms matter radius of 2.378 fm . the associated radius of a fermi - function distribution with diffuseness @xmath17 = 0.44 fm is 2.228 fm according to the last line of table i. this is a useful reference value for the discussion below . a critical part of the m3y+repulsion , double - folding potential is the repulsive part @xcite , which is generated by a contact interaction , @xmath24 . the densities that are used in calculating the repulsive , double - folding potential have the same radius as those that are used in the calculation of the m3y double - folding potential but the diffuseness @xmath25 is chosen differently ( usually smaller . ) the strength @xmath26 of the repulsion , on the other hand , is adjusted for a given value of @xmath25 so that the nuclear incompressibility @xmath27 = 234 mev is produced ( see ref . @xcite for details . ) with this constraint there are three parameters that must be specified before the m3y+repulsion potential can be calculated , namely , the matter radii of the two reacting nuclei , and the diffuseness parameter @xmath25 associated with the repulsion . the parameters that give the best fit to trentalange s @xmath0c+@xmath0c fusion data are the @xmath0c radius @xmath16 = 2.28 fm and the diffuseness @xmath25 = 0.31 fm . how these parameters were determined is described in detail in section iii.b . the radius @xmath16 = 2.28 fm is slightly larger than the 2.228 fm that was estimated from eq . ( [ rms ] ) . that may not be unreasonable as discussed in ref . @xcite because using a larger radius in a coupled - channels calculation is a way to compensate for the dynamic polarization of states that are not included explicitly in the calculation . the discrepancy between the estimated rms radius , eq . ( [ rms ] ) , and the value extracted from the fit to the fusion data is only 1.2 % . the entrance channel potential determined by the best fit parameters is shown by the solid curve in fig . [ pot1313]a . also shown in fig . [ pot1313 ] are the potentials that will be applied in calculating the fusion of @xmath1c+@xmath0c and @xmath1c+@xmath1c . they were obtained by setting the radius of @xmath1c equal to the radius @xmath16 = 2.155 fm of the point - proton distribution of @xmath1c ( see table i ) , and the diffuseness associated with the repulsion was set to @xmath25 = 0.31 fm . the pure m3y double - folding potentials shown in fig . [ pot1313 ] are unphysical because they are much deeper for overlapping nuclei than the ground state energy of the compound nucleus . the ( blue ) dashed curve in fig . [ pot1313]a is the entrance channel potential for a standard woods - saxon ( ws ) potential with the depth @xmath28 = -39.31 mev , radius @xmath29 = 5.268 fm , and diffuseness @xmath17 = 0.63 fm . here the radius was adjusted to optimize the fit to trentalange s fusion data as discussed in section iii.a . this potential is deeper than the m3y+repulsion potential and has a minimum that is close to the energy of the compound nucleus @xmath30 mg . the three entrance channel potentials shown in fig . [ pot1313]a , which are based on the m3y , the m3y+repulsion and the woods - saxon potential , produce essentially the same coulomb barrier height . @xmath31 6 mev . the most important features of the m3y+repulsion potential are that it produces a shallow pocket and a relatively thick barrier . these features were utilized in ref . @xcite to explain the hindrance of fusion that has been observed in the fusion of many medium mass systems at extreme subbarrier energies . the analysis of the @xmath0c+@xmath0c fusion data discussed in the next section confirms that a shallow pocket in the entrance channel is a general feature that helps explain the energy dependence of fusion data . the nuclear couplings that excite the low - lying states of projectile and target are derived from the macroscopic description of surface excitations @xcite . in this description , the surface of nucleus @xmath32 is parametrized as @xmath33 where @xmath34 are the ( static or dynamic ) deformation amplitudes of the nucleus , and @xmath35 specifies a spatial direction . the distortion of the nuclear surface , @xmath36 is an operator that can excite and de - excite the nucleus ( see chapter ii.4 of ref . the excitation can be vibrational , in which case the intrinsic hamiltonian is a harmonic oscillator , or it can be rotational , in which case @xmath37 = @xmath38 @xmath39 , where @xmath38 is the static deformation parameter and @xmath40 is the direction of the symmetry axis of the deformed nucleus . in the macroscopic description of heavy - ion reactions @xcite , the nuclear potential between two colliding nuclei is parametrized in terms of the surface distortions @xmath41 as follows @xmath42 where @xmath35 is the direction of the center of mass distance between projectile and target . in the rotating frame approximation , which is a simplifications that is commonly used and which will be used in coupled - channels calculations , the direction of @xmath35 defines the z - axis , so that @xmath43 in the model of heavy - ion fusion reactions described in ref . @xcite , the nuclear potential is expanded up to second order in the surface distortions , @xmath44 @xmath45 , \label{uexp}\ ] ] where the argument of @xmath46 and its derivatives is @xmath47 . the second - order term in this expression has been renormalized so that the ground state expectation value of that term is zero . the ground state expectation value of the first - order term ( proportional to @xmath48+@xmath49 ) will also vanish if the ground states of the two reacting nuclei are @xmath10 states . the ground state expectation of the entire expression , eq . ( [ uexp ] ) , is therefore identical to the ` bare ' interaction , i. e. , @xmath50 this implies that the m3y+repulsion potential which is calculated using the ground state densities of the reacting nuclei can be used as the bare interaction , @xmath46 . the diagonal matrix element of the first - order term in eq . ( [ uexp ] ) can be non - zero , for example , in the @xmath51 state of a deformed nucleus . the calculation of diagonal and off - diagonal matrix elements is outlined in the appendix . details of how to calculate the matrix elements of the second - order couplings in eq . ( [ uexp ] ) , both for vibrational and rotational excitations , are given in ref . @xcite . the nuclear structure input to the coupled - channels calculations is listed in table ii . the nuclei @xmath1c and @xmath0c are both considered deformed with oblate quadrupole shapes . the connection between the deformation parameter @xmath38 and the measured @xmath52 values is summarized in the appendix . for @xmath1c one can extract the quadrupole deformation parameter @xmath53 = 0.57(2 ) from the measured strength of the @xmath54 transition . the value of this deformation parameter produces an intrinsic quadrupole moment of @xmath55 = -19.5 @xmath19 ( assuming an oblate shape ) . the quadrupole moment of the @xmath51 state is therefore @xmath56 = -2/7@xmath55 = 5.57 @xmath19 , which is consistent with the measured value of @xmath57 @xmath19 @xcite . also shown in the table is the structure information about the @xmath58 transition which is included in the calculations as part of the two - phonon quadrupole excitation of @xmath1c . values of quadrupole deformation parameter @xmath53 of @xmath0c , extracted from eq . ( [ belwu ] ) of the appendix and the known strengths of the quadrupole transition from the 1/2@xmath59 ground to the 3/2@xmath59 and the 5/2@xmath59 excited states , are shown in the last column of table ii . they are seen to be almost identical , which indicates that @xmath0c is a fairly good rotor . assuming the deformation parameter @xmath53 = 0.495 and an oblate shape , one obtains the intrinsic quadrupole moment @xmath55 = -17.9 @xmath19 . the strength of the octupole transition in @xmath0c quoted in table ii appears to be very large and almost as large as the octupole strength in @xmath1c . this is misleading because what matters is the off - diagonal matrix element of the octupole amplitude , which is given by eq . ( [ matel ] ) of the appendix . in @xmath1c the matrix element is @xmath60 whereas in @xmath0c it is ( with @xmath9 = @xmath27= 1/2 ) @xmath61 the matrix element of the octupole amplitude is therefore reduced in @xmath0c by the factor @xmath62 0.655 relative to the expression eq . ( [ octc12 ] ) for @xmath1c . the coupled - channels calculations that are performed are similar to those presented in ref . @xcite for the fusion of @xmath63al with different germanium isotopes . in addition to the nuclear couplings derived from the nuclear interaction , eq . ( [ uexp ] ) , the coulomb interaction ( eq . ( 2 ) of ref . @xcite ) is included in the calculations to first order in the deformation amplitudes . the one - neutron transfer reactions that will be considered all involve the @xmath20 orbit , both in the initial and final states . one example is the ground state to ground state transfer reaction , @xmath1c(@xmath0c,@xmath1c)@xmath0c , which has zero q value . the spectroscopic factors for the initial and final state @xmath11 state in @xmath0c were set equal to 0.75(10 ) , which is the value recommended in table iii of ref . @xcite . the other example is the one - neutron transfer from @xmath0c+@xmath0c to the ground states of @xmath1c+@xmath2c which has a q - value of + 3.2 mev . the spectroscopic factor for the initial state is the same as above . it is set equal to 2 * 0.75 in the calculations because the transfer can take place from either of the two @xmath0c nuclei in the entrance channel . the spectroscopic factor for the @xmath20 orbit in the @xmath10 ground state of @xmath2c is 1.63(33 ) according to table iii of ref . @xcite . the transfer form factors that are used are the so - called quesada single - particle form factors @xcite which in the past turned out to be fairly realistic @xcite . they have here been calibrated against calculations performed with the computer code ptolemy @xcite of the one - neutron transfer cross in @xmath1c+@xmath0c collisions at a 5.2 mev center - of - mass energy . the fusion cross section is calculated from the ingoing flux which is determined by the ingoing - wave - boundary conditions ( iwbc ) that are imposed at the minimum of the pocket in the entrance channel potential . this approximation ignores the internal structure of the combined di - nuclear or compound system and the calculated cross sections are usually fairly smooth functions of energies , at least at energies near and below the coulomb barrier . at energies far above the coulomb barrier , there are sometimes numerical problems in the coupled - channels calculations which can cause an erratic behavior of the calculated cross section as a function of energy . one can overcome these problems by applying a weak imaginary potential @xcite @xmath64 which acts near the position @xmath65 of the pocket in the entrance channel potential . the fusion cross section is then defined as the sum of the absorption cross section and the cross section for the ingoing flux . a weak imaginary potential will be applied in the following when the calculated fusion cross section is shown in a linear plot at energies far above the coulomb barrier . the parameters that are used are @xmath66 = -5 mev and @xmath67 = 0.5 fm . no imaginary potential will be used in the calculations at energies below the coulomb barrier . the 1/2@xmath59 ground state spin of @xmath0c causes some concern when calculating the scattering and fusion of @xmath0c+@xmath0c . the fermionic nature of @xmath0c requires that the total wave function for @xmath0c+@xmath0c be anti - symmetric . by coupling the intrinsic @xmath11 ground state spins of the reacting nuclei one obtains a total spin of @xmath68=0 ( anti - symmetric ) and @xmath68=1 ( symmetric ) . the wave function for the relative motion of the two nuclei must therefore be symmetric for @xmath68=0 , i. e. , it consists of even partial waves , and it must be antisymmetric for @xmath68=1 , consisting of odd partial waves . the fusion cross section is therefore calculated as the weighted sum @xmath69 it turns out that the contributions to the fusion cross section from odd and even @xmath70-values are almost identical . the biased weighting of the two contributions in eq . ( [ fusa ] ) does therefore not have much effect . in calculations of the fusion and elastic scattering of two identical spin zero @xmath1c nuclei , one should only consider the contribution from even partial waves which is doubled because of the symmetry of two identical particles . in reactions of @xmath1c+@xmath0c , on the other hand , the contribution from even and odd partial waves have the same weight . in this section the @xmath0c+@xmath0c fusion data by trentalange et al . @xcite and charvet et al . @xcite are analyzed in terms of coupled - channels calculations . the channels that are included are enumerated in fig . [ spc1313 ] . the ( red ) solid lines represent the elastic channel and the channels associated with the excitation of the states shown in table ii , in both projectile or target . the calculation based on these channels is referred to as the one - phonon ( ph1 ) calculation and has a total of 7 channels . the ( blue ) dashed lines in fig . [ spc1313 ] are the 6 mutual excitations of the lowest quadrupole and octupole excitations in projectile and target . these include the four mutual states @xmath71 , where @xmath72 = @xmath73 or @xmath74 , @xmath32=1,2 , and @xmath75 and @xmath76 belong to different nuclei , and the two @xmath77 mutual excitations that belong to the same nucleus . together with the seven channels of the one - phonon calculation that gives a total of 13 channels . this calculation is referred to as the mutual excitation calculation or ch13 . the calculation ch13 will be compared to the one - phonon calculation ( ph1 ) described above and the no - coupling limit which has only one channel . there are other excitations , for example , the two - phonon excitations which are poorly known and mutual excitations that involve the @xmath78 state but they are all ignored in the following . it can be very difficult to see small discrepancies between measured and calculated cross sections , in particular , when they are plotted on a conventional logarithmic scale . it is therefore useful to use other representations that amplify certain features of the comparison , such as the @xmath68 factor for fusion , which emphasizes the behavior at low energies . another representation , which is very useful when coupled - channels effects are modest , is the so - called enhancement factor , which was used in the early days when the enhancement of subbarrier fusion was first discovered . it is defined as the ratio of a cross section relative to the cross section calculated in the no - coupling limit , and it will be used both for measurements as well as coupled - channels calculations , as long as it is indicated which no - coupling limit is used as a reference . finally , the comparison of data and coupled - channels calculations will also be made in terms of the ratio of the measured and calculated cross sections . if the measured and calculated cross sections agree , the ratio should be close to one . the one - neutron transfer discussed earlier is included as an independent degree of freedom . that implies that the transfer can take place from any excited state and it is calculated with the same form factor which describes the ground state to ground state transition . the full calculation ( with 13 excitation channels mentioned above and the one - neutron transfer ) will therefore contain 26 channels and is referred to as the ch26 calculation . the analysis of the data is performed in terms of the @xmath79 per data point , @xmath80 , which is minimized by adjusting certain parameters of the ion - ion potential . the overall normalization of the calculated cross sections is also adjusted by a multiplicative factor @xmath81 , in order to improve the fit to the data . the motivation for this adjustment is the large uncertainty in the absolute normalization of the measured cross sections that was mentioned in the introduction . the shape ( i.e. , the energy - dependence ) of the measured cross section is assumed to be more accurately determined . the uncertainty in the data analysis will therefore include the statistical uncertainty and an adopted systematic error of only 3% , which is much smaller than the 15% experimental error @xcite . the first set of calculations are based on a standard woods - saxon potential with typical parameters from ref . the diffuseness of the potential is set to @xmath17 = 0.63 fm and the depth is -39.31 mev . the radius of the potential , @xmath29 = 5.268 fm , was adjusted to optimize the fit to the data of ref . @xcite . a comparison to the data is made in fig . [ frat1313ws ] in terms of the enhancement of the cross sections with respect to the no - coupling limit . the enhancement of the measured cross sections is seen to deviate from the enhancement of the coupled - channels calculation ( ch26 ) by up to 20% . moreover , the effects of mutual excitations and one - neutron transfer are seen to be modest by comparing to the no - transfer calculations ( ch13 ) and the one - phonon calculations ( ph1 ) . it is very unlikely that the agreement with the data can be improved much by adjusting the strengths of the couplings to the excitation or transfer channels . it is shown in the next subsection that a much better agreement with the data is achieved by applying the m3y+repulsion instead of the woods - saxon potential . the agreement with the data in fig . [ frat1313ws ] can not be improved much by scaling the calculated cross sections with an adjustable scaling factor @xmath81 in the data analysis . the best fit for @xmath81 = 1 has a @xmath80 = 8.7 . it can be reduced to @xmath80 = 6.8 by adjusting both the scaling factor @xmath81 and the radius @xmath29 of the woods - saxon well . the values that give the best fit are @xmath81 = 0.885 and @xmath29 = 5.345 fm . another observation in fig . [ frat1313ws ] is that the data sets of trentalange et al . @xcite and charvet et al . @xcite differ by up to 10% in the overlapping energy regime . however , this is not a serious problem because the systematic error of both experiments is about 15% . one advantage of the m3y+repulsion potential is that it has an additional adjustable parameter besides the radius @xmath16 of the density , namely , the diffuseness parameter @xmath25 of the density that is used in calculating the repulsive part of the potential . this parameter controls the depth of the pocket and the thickness of the barrier in the entrance channel potential . the determination of the best fit parameters is illustrated in fig . [ xki2 ] . for a fixed value of @xmath25 , the fusion cross sections were calculated for different values of the radius @xmath16 of @xmath0c . in each case the scaling factor @xmath81 was adjusted to give the best @xmath80 . the results are shown by the dashed curves . the minimum @xmath80 for each dashed curve defines the solid curve , and the minimum of this curve defines the absolute best fit ( abf ) to the data . the parameters of the minimum are @xmath25 = 0.31 fm , the matter radius @xmath16 = 2.28 fm , and the scaling factor @xmath81 = 0.843 . they are compared in table 1 to other parameters that were obtained for fixed choices of the radius of @xmath0c . the @xmath80 of the best fit to trentalange s data is small according to table 1 , @xmath80 = 1.0 , although the adopted systematic error of the analysis was set to only 3% . the scaling factor of the best fit , @xmath81 = 0.843 , is not unreasonable but it is slightly extreme because the systematic uncertainty in the absolute cross section of the measurement was estimated to be 15% @xcite . it is therefore of interest to discuss whether the other parameters of the best fit are realistic . the diffuseness @xmath25 associated with the repulsion , @xmath25 = 0.31 fm , falls below the range of values @xmath25 = 0.4 0.43 fm that have been obtained in the analysis of fusion data for other symmetric heavy - ion systems @xcite . the @xmath0c radius of the best fit , @xmath16 = 2.28 fm , produces an rms matter radius of 2.407 fm according to table 1 which is slightly larger than the rms matter radius of 2.378 fm that was estimated in eq . ( [ rms ] ) . this result is not unreasonable as discussed in connection with an analysis of the fusion data for @xmath82ca+@xmath82ca @xcite . there it was pointed out that a larger radius may simulate the effect of the dynamic polarization of excited states which are not included explicitly in the coupled - channels calculations . if the scaling parameter @xmath81 is kept fixed at @xmath81=1 in the data analysis , the best fit is achieved with a smaller @xmath0c radius , namely , @xmath16 = 2.17 fm . this value is almost as small as the radius for the point - proton density distribution of @xmath0c . the fit is poor , with a @xmath80 = 2.75 as shown in the first line of table i. the value of the diffuseness parameter @xmath25 which produces the best fit in this case is @xmath25 = 0.33 . the ratios of the measured and calculated cross sections discussed above are illustrated in fig . [ frf1313 ] . the ratio of trentalange s data @xcite to the best fit ( bf ) solution with radius @xmath16 = 2.17 and @xmath81 = 1 is shown by the open symbols . the ratio of the same data set to the absolute best solution ( abf ) with radius @xmath16 = 2.28 fm is shown by the solid ( red ) points . also shown at higher energies is the ratio of charvet s data @xcite to the same solution ( with radius @xmath16 = 2.28 fm . ) it is seen that the ratios for the absolute best solution are essentially constants ( @xmath83 0.84 for trentalange s data , and @xmath83 0.92 for chatvet s data up to 10 mev ) . the two data sets differ by up to 10% in the overlapping energy regime . however , this is not a serious problem as mentioned earlier because the systematic error of both experiments is about 15% . the ratio for best fit ( bf ) solution for @xmath81=1 exhibits some energy dependence in fig . [ frf1313 ] . it is of course possible that the data do contain some structures ( similar to those observed in the @xmath1c+@xmath1c fusion data ) that should not be accounted for by the type of coupled - channels calculations performed here . the absolute best fit ( abf ) solution , obtained with the @xmath0c radius @xmath16 = 2.28 fm and @xmath81 = 0.843 , could therefore be misguided or misleading . on the other hand , the absolute best fit does have some attractive features as argued above : the quality of the fit is excellent , and the extracted rms matter radius is only 1.2% larger than expected . the absolute best fit requires a scaling of the calculation , @xmath81 = 0.843 , which is not unreasonable considering the large 15% systematic error of the experiment . the parameters of the abf are therefore adopted in the following . in order to illustrate what makes it possible to reproduce the energy dependence of trentalange s data so well when using the m3y+repulsion potential , we show in fig . [ frat1313 ] the enhancement of the calculations and the data relative to the no - coupling calculation . in contrast to the results for the woods - saxon potential shown in fig . [ frat1313ws ] , there is now a strong sensitivity to mutual excitations . this is caused by a larger second derivative of the ion - ion potential ( stemming from the onset of the strong repulsion ) , and consequently , a stronger quadratic coupling of the entrance channel to mutual excitations . an interesting observation in fig . [ frat1313 ] is that the neutron transfer does not have much influence on the calculated cross section in the energy regime of the experiment . this is seen by comparing the calculations labeled ch13 and ch26 . the influence shows up at energies below 3 mev , so it will have an impact on the extrapolation of the cross sections to lower energies . another interesting feature in fig . [ frat1313 ] is the structure of the data at low energies . the small peaks at 3.6 and 4.2 mev , for example , could be remnants of similar structures observed in the fusion data for @xmath1c+@xmath1c . the calculated @xmath68 factors for the fusion of @xmath0c+@xmath0c are compared in fig . [ fsf1313 ] to the data of ref . the data have here been divided by the scaling factor @xmath81 = 0.843 which optimizes the agreement with the full calculation labeled ch26 . the calculation that is based on one - phonon excitations only ( ph1 ) does no reproduce the energy dependence of the data between 3 and 6 mev . the calculation that includes mutual excitations but no transfer ( the ch13 calculation ) is seen to reproduce the corrected data very well . finally , the additional coupling to the one - neutron transfer produces some enhancement in the full ch26 calculation but that occurs at energies below the range of the experiment . the calculated cross sections obtained with all 26 coupled channels and the parameters determined in the previous subsection are compared in fig . [ flf1313 ] to the data of ref . the calculation exceeds the data on average by only 4% , which is small compared to the 15% systematic error of the experiment . the coupled - channels effects are fairly modest at high energies ; this can be seen by comparing to the no - coupling calculation which is shown by the thick dashed curve . the results of a maximum angular momentum cutoff in the coupled - channels calculations are shown for @xmath84 = 2 - 12 . having determined the radius of @xmath0c and the diffuseness @xmath25 associated with the repulsive part of the double - folding potential in the previous section , one can now predict the ion - ion potential for @xmath1c+@xmath0c , provided the density of @xmath1c is known . here the parameters of the point - proton density of @xmath1c given in table 1 are adopted . the predicted m3y+repulsion potential is shown in fig . [ pot1313]b . it is similar to @xmath0c+@xmath0c potential shown in fig . [ pot1313]a . the main difference between the two figures is that the ground state of the compound nucleus @xmath85 mg is located at a higher energy than @xmath30 mg because of the unpaired neutron in @xmath85 mg . the excitation energies that are considered in the calculations of the fusion cross sections for @xmath1c+@xmath0c are shown in fig . [ spc1213 ] . the solid ( red ) lines represent the excitations shown in table i which together with the elastic channel give a total of 7 channels . one of these states is the @xmath86 hoyle state of @xmath1c which is included because its coupling to the @xmath51 is known . the ( blue ) dashed lines in fig . [ spc1213 ] are the 6 mutual excitations that are considered . they include the @xmath87 mutual excitation in @xmath1c , the @xmath77 mutual excitation in @xmath0c , and the four @xmath71 mutual excitations in projectile and target with @xmath75 = @xmath51 or @xmath88 and @xmath76 = @xmath73 or @xmath74 . two - phonon excitations ( except the @xmath86 hoyle state which has already been mentioned ) and mutual excitations that involve the @xmath78 state in @xmath0c are ignored . the total number of excitation channels is therefore 7 + 6 = 13 . when the one - neutron transfer with zero q value is included in the calculations as described in sect . ii.d , the number of channels is doubled . the full calculation , which includes the one - neutron transfer and all of the 13 excitation channels shown in fig . [ spc1213 ] will therefore have 26 channels and it is referred to as ch26 . a major problem , which causes some uncertainty in the predictions made by the coupled - channels calculations , is the high excitation energy and the large strength of the octupole transition in @xmath1c ( see table 1 . ) a large excitation energy implies decaying boundary conditions at large separations of projectile and target instead of the scattering boundary conditions , and this can cause numerical problems . a large octupole strength implies a strong sensitivity to mutual excitations that involve the @xmath88 state and some of these excitations are poorly known as explained below . the octupole excitation of @xmath0c , on the other hand , is not a problem because the excitation energy of the @xmath74 state is much lower than in @xmath1c , and the effective octupole strength is reduced as shown in eq . ( [ octupole ] ) . the enhancement of the @xmath1c+@xmath0c fusion relative to the no - coupling limit is shown in fig . [ frat1213 ] for different coupled - channels calculations . the full calculation ch26 has a very strong peak near 4.7 mev . the calculation ch24 , which excludes the mutual @xmath87 excitation in @xmath1c , has a much weaker peak . there are two reasons why the mutual @xmath87 excitation in @xmath1c has such a large influence on subbarrier fusion . one reason is the large octupole strength of @xmath1c . the other reason is that the direct coupling of the ground state to mutual excitations is governed by the second derivative of the ion - ion potential , according to eq . ( [ uexp ] ) , and this quantity is particularly large for the m3y+repulsion potential as pointed out in section iii.c . it is very interesting that the enhancement of the data @xcite with respect to the no - coupling limit also exhibits a peak structure in fig . [ frat1213 ] . the peak is located near 4.3 mev , somewhat below the positions of the two calculated peaks . the influence of transfer can be seen by comparing the thin curves c12 and ch13 , which do not include the influence of transfer , to the corresponding thick curves ch24 and ch26 , which include the effect . the influence is relatively modest but it does shift the calculated peaks towards the peak position of the data . the @xmath68 factor for the fusion of @xmath1c+@xmath0c is illustrated in fig . [ fsf1213 ] . the calculation ` ws ch26 ' is based on the standard woods - saxon potential with radius @xmath29 = 5.352 fm , depth @xmath28 = -38.6 mev , and diffuseness @xmath17 = 0.63 fm , which was illustrated in fig . [ pot1313]b . it includes all 26 channels considered in this sections and makes a very good fit to the data above 4 mev . however , a strong hindrance of the data sets in at energies below 4 mev . the comparison shows that the fusion of @xmath1c+@xmath0c is an example on the fusion hindrance phenomenon which has been observed in many medium heavy systems @xcite . the two calculations ch24 and ch26 that are based on the m3y+repulsion potential reproduce the data in fig . [ fsf1213 ] at the lowest and also at high energies . the calculation ch26 exceeds the data between 4.5 and 5.5 mev by up to 50% . the calculation ch24 is in better agreement with the data but there are still some deviations . the measured @xmath68 factor is rather flat between 3 and 4 mev and has a modest peak near 4.3 mev . the calculated @xmath68 factors do not exhibit any sharp peak structures but rise slowly with decreasing energy . the two calculations ch24 and ch26 represent extreme views of the influence of the mutual excitation of the @xmath51 and @xmath88 states in @xmath1c , i. e. , of a quadrupole excitation built on the @xmath88 state or an octupole excitation built on the @xmath51 state . the calculation ch24 completely ignores it , whereas the calculation ch26 exaggerates it by assuming that the @xmath51 and @xmath88 excitations are independent modes of excitation . in a realistic , microscopic description , the mutual excitation of the @xmath51 and @xmath88 states will not be one discrete excitation but will be fragmented and spread out over a range of excitation energies . this is indeed the result of shell model calculations that are discussed in section vi . similar predictions are made for the fusion of @xmath1c+@xmath1c . the entrance channel potential that is used is shown in fig . [ pot1313]c . it was obtained with the parameters of the point - proton density of @xmath1c shown in table i , and the diffuseness associated with the repulsive part of the m3y+repulsion interaction was again set equal to the value @xmath25 = 0.31 fm which was determined in the analysis of the @xmath0c+@xmath0c fusion data . the excitation energies that are considered in the coupled - channels calculations are shown in fig . [ spc1212 ] . the solid ( red ) lines represent the excitations shown in table i which together with the elastic channel give a total of 7 channels . the second @xmath10 state is hoyle state . the ( blue ) dashed lines in fig . [ spc1212 ] are the mutual excitations that are considered in the calculations . they include the three mutual excitations : @xmath89 , @xmath87 and @xmath90 , with a one - phonon excitation in each @xmath1c nucleus , and the two mutual excitations : @xmath87 , with both one - phonon excitations belonging to the same nucleus . there is also a @xmath91 mutual excitation indicated in the figure ( with one octupole excitation in each nucleus ) but it is ignored because the excitation energy is very high . the total number of channels consisting of the 7 channels mentioned above ( represented by the solid ( red ) lines in fig . [ spc1212 ] ) and the 5 mutual excitation channels ( the ( blue ) dashed lines in fig . [ spc1212 ] ) is therefore 12 . the associated coupled - channels calculation is referred to as ch12 . the calculation ch12 includes the mutual excitation of the @xmath51 and @xmath88 states , not only when the two states belong to different nuclei but also when they belong to the same nucleus . since the description of the @xmath87 mutual excitation in terms of two independent modes of excitation is questionable when the two excitations belong to the same nucleus , it is of interest to perform a calculation that does not include that kind of mutual excitation . such a calculation contains 10 channels and is referred to as ch10 . the ch10 calculation does contain mutual excitations of the @xmath51 and @xmath88 states but only when the two one - phonon excitations belong to different nuclei . the two calculations are compared in fig . [ ff1212 ] to the measured cross sections of ref . the ch12 calculation exceeds the data substantially between 3 and 5.5 mev , whereas the ch10 calculation is lower and intersects with some of the measured peak cross sections . the @xmath68 factors for the fusion of @xmath1c+@xmath1c predicted by the ch10 and ch12 calculations are compared in fig . [ fsf1212 ] to the data of five experiments @xcite . the symbols for the four measurements @xcite are the same as in fig . [ ff1212 ] , whereas the symbol for spillane s measurements @xcite is indicated in the figure . the calculation ch12 exceeds all of the data points shown in this figure . it is considered an extreme upper limit because it exaggerates , as discussed above and in the previous section , the influence of the mutual @xmath87 excitation within the same nucleus . the calculation ch10 , which is much lower , is considered a lower limit of the prediction . the large difference between the ch10 and ch12 calculations shown in fig . [ fsf1212 ] is very unfortunate . it illustrates the difficulty and uncertainty in predicting the fusion cross section due to the poor empirical knowledge of excitations that are built on the @xmath51 and @xmath88 states . we shall eliminate that problem in the next section by applying the results of shell model calculations . another uncertainty , which is similar in magnitude , is due to the empirical value @xmath92 = 0.90(7 ) @xcite . this is illustrated in the figure by the thin lines and represents an error band of up to @xmath825% . the @xmath68 factor one obtains in the no - coupling limit is shown by the lowest dashed curve in fig . [ fsf1212 ] . it is a factor of 2 to 3 below the measured peak cross sections and a factor of 2 to 5 above the minima of the measured cross sections . in other words , it does not provide a background nor does it provide an upper limit of the measured cross sections . it is closer to the smooth experimental cross section constructed by yakovlev et al . the discussion in the previous sections shows that it is possible to develop a fairly comprehensive description of the fusion data for the three carbon systems within the coupled - channels approach . a disturbing trend in the analysis of the @xmath0c+@xmath0c fusion data @xcite is the need for dividing the data by the factor @xmath81 = 0.843 , i. e. , for increasing the measured cross sections by 19% . however , that is not a very serious problem because the systematic error of the experiment @xcite is 15% . the need for a renormalization of the @xmath0c+@xmath0c data is possibly an experimental problem because it is difficult to explain why the measured fusion cross sections for @xmath1c+@xmath0c @xcite are larger than the measured cross sections for @xmath0c+@xmath0c @xcite . naively , one would expect the cross sections for the smaller system to be smaller . this is indeed the ordering that is observed in the fusion of carbon isotopes with a thorium target @xcite . of course , the problem could be in the absolute normalization of @xmath1c+@xmath0c data @xcite . whatever the explanation is , the basic problem is illustrated in fig . [ frfcall ] , where the measured cross sections for the two systems have both been normalized to the ch26 calculation for the @xmath0c+@xmath0c system discussed in sections iii.b . it is seen that the normalized cross sections are fairly constant but the cross sections for the smaller @xmath1c+@xmath0c system are larger , on average by the factor 1.02/0.84 = 1.21 ( c. f. the caption to fig . [ frfcall ] ) . the 21% deviation between the two measurements is within the large uncertainties of the two experiments , each being of the order of 15 to 30% but it does show an unexpected trend . clearly , it is desirable to have the experimental uncertainties reduced in future measurements . another disturbing feature discussed in the previous section is the large uncertainty in the cross section predicted for the fusion of @xmath1c+@xmath1c . half of the uncertainty concerns the influence of the @xmath87 mutual excitation which is poorly known experimentally @xcite . one way to overcome this problem is to rely on shell model calculations @xcite which predict that e2 excitations built on the @xmath88 state and e3 excitations built on the @xmath51 state primarily populate a @xmath93 state which is about 4 mev above the @xmath88 state . that is similar to the model we have used for the mutual @xmath87 excitation . however , the b - values of these transitions are only about half of the b - values obtained for the corresponding e2 and e3 transitions from the ground state to the first @xmath51 and @xmath88 states in @xmath1c . the cross sections one obtains for the three fusion reactions discussed in this work , using the best fit parameters determined in the previous sections , and the shell model results described above , are compared to the data in fig . [ fsfcallsm ] in terms of the @xmath68 factor . the data for @xmath0c+@xmath0c have been divided by the scaling factor @xmath94 = 0.843 which results in an excellent agreement with the data . the calculations of the fusion of @xmath1c+@xmath1c and @xmath1c+@xmath0c are ch12 and ch26 calculations , respectively , which are similar to those discussed in section iv and v. the only difference is that the parameters for the mutual @xmath87 excitation in @xmath1c have been replaced by shell model predictions @xcite . the calculated @xmath68 factors for the fusion of @xmath1c+@xmath1c and @xmath1c+@xmath0c are essentially predictions because the parameters of ion - ion potential and the density of @xmath0c were determined in the analysis of the @xmath0c+@xmath0c fusion data , whereas the density of @xmath1c was determined by a fit to the experimental point - proton distribution . the @xmath1c+@xmath0c data @xcite shown in fig . [ fsfcallsm ] form a plateau between 2.5 and 4 mev and has a peak near 4.3 mev . these features are not quite reproduced by the calculation but the fit to the data is not poor ; the fit to the data of ref . @xcite is stable with respect to the scaling factor @xmath81 and to an overall energy shift @xmath95 in the center of mass energy , i. e. , the @xmath79 per data point has a minimum for @xmath81 = 1 and @xmath95 = 0 . although there is still some uncertainty in the predicted @xmath68 factor for the fusion of @xmath1c+@xmath1c , it appears that the calculation shown in fig . [ fsfcallsm ] is consistent with the maxima of the measured peak cross sections . the peak at 4.924 mev is within the error band of the calculation . the 3 low - energy ( triangular shaped ) data points are from ref . they exceed the calculation tremendously but that is not a problem either because those 3 data points are questionable as discussed in ref . @xcite . the calculation ( with its error band ) is therefore expected to provide a useful guidance and an upper limit for the extrapolation to the low energies that are of interest to astrophysics . we have explored the fusion of different combinations of carbon isotopes within the coupled - channels approach . the fusion cross section is determined by ingoing - wave - boundary conditions that are imposed at the minimum of the pocket in the entrance channel potential . it turned out that the measured cross sections for the fusion of @xmath0c+@xmath0c can not be explained accurately at low energies when a standard woods - saxon potential is applied in the calculations . however , the data can be explained fairly well by applying the shallow potential that is produced by the m3y+repulsion potential . this is a characteristic feature of the fusion hindrance phenomenon which has been observed in many medium - heavy systems . one reason the shallow potential makes it possible to explain the carbon data by coupled - channels calculations is that the high - lying excitations become closed channels at low center - of - mass energies , not only with respect to reactions but also to fusion . the coupled - channels calculations are surprisingly sensitive to mutual excitations and this seems to be justified by the comparison to the data . thus the calculations that include one - phonon excitations alone can not explain the data so well , neither when the standard woods - saxon nor the m3y+repulsion potential is applied . the data are reproduced much better when mutual excitations are included and the shallow m3y+repulsion potential is applied . the reason is that the direct coupling to these high - lying states is proportional to the second derivative of the ion - ion potential and this quantity is much larger for the m3y+repulsion than for the standard woods - saxon potential because of the strong repulsion . the large sensitivity to the excitation of high - lying states is particularly critical when the mutual excitations involve the strong octupole excitation of @xmath1c . the uncertainty in the strength of the octupole excitation , and also in the strength of transitions to high - lying states that are built on the @xmath51 and @xmath88 states , makes it difficult to predict the fusion cross sections very accurately when @xmath1c is involved . however , by adopting the nuclear structure properties of the high - lying states predicted by the shell model , one obtains a fairly good description of the @xmath1c+@xmath0c fusion data . the prediction of the cross section for the fusion of @xmath1c+@xmath1c , based on the ingoing - wave - boundary conditions and the shell model prediction of the coupling to high - lying states , exceeds most of the measured cross sections and is consistent with the maxima of the observed peak cross sections . thus it appears that the calculation provides an upper limit of the measured cross sections . the calculation is therefore expected to provide an upper limit for the extrapolation of cross sections into the unexplored territory of very low energies . on the experimental side it is very important to get a better absolute normalization of the measured fusion cross sections . the current systematic errors are fairly large . this was exploited in the analysis of the @xmath0c+@xmath0c fusion data by focusing on the shape of the measured cross section , whereas the absolute normalization was treated as an adjustable parameter . in this way it was possible to calibrate the m3y+repulsion potential and achieve an excellent fit to the data . it is of great interest to know whether the renormalization that was applied to the data can be justified by more accurate measurements . * acknowledgments*. the authors are grateful to b. a. brown for providing the shell model calculations . this work was supported by the u.s . department of energy , office of nuclear physics , contract no . de - ac02 - 06ch11357 . one of us ( x.t . ) is supported by the nsf under grant no . phy-0758100 and phy-0822648 , and the national natural science foundation of china under grant no . 11021504 , and the university of notre dame . the matrix elements of the couplings between two heavy ions are generated by matrix elements of the static or dynamic deformation amplitudes @xmath37 ( see for example ref . the matrix elements are expressed in terms of the reduced matrix elements @xmath96 , @xmath97 the off - diagonal reduced matrix elements can be extracted from the measured @xmath98 values according to the formula @xmath99 if the @xmath100-value is expressed in weisskopf units the relation is @xmath101 ^ 2}{4\pi } \frac{|\langle i_2||\alpha_\lambda || i_1\rangle|^2}{2i_2 + 1}.\ ] ] the diagonal matrix elements , on the other hand , can be obtained from the measured quadrupole moments , @xmath102 they are related to the intrinsic quadrupole moments @xmath55 and the @xmath27 quantum number by @xmath103 for a deformed nucleus with a static deformation @xmath38 the deformation amplitude is @xmath37 = @xmath104 , where @xmath40 is the orientation of the symmetry axis . in this case the expressions for the matrix elements between states @xmath105 are @xmath106 this implies ( by comparing to eq . ( [ matrixelm ] ) that the reduced matrix elements are @xmath107 @xmath101 ^ 2}{4\pi } \ \beta_\lambda^2 \ \frac{2i_1 + 1}{2i_2 + 1 } \ \langle i_1k \ \lambda 0| i_2k\rangle^2 . \label{belwu}\ ] ] 99 l. r. gasques _ et al . _ , c * 72 * , 025806 ( 2005 ) . s. trentalange , s. c. wu , j. l. osborne , and c. a. barnes , nucl . phys . * a483 * , 406 ( 1988 ) . j. l. charvet , r. dayras , j. m. fieni , s. joly , and j. l. uzureau , nucl . a*376 * , 292 ( 1982 ) . r. a. dayras , r. g. stokstad , z. e. switkowski , and r. m. wieland , nucl . a*265 * , 153 ( 1976 ) . m. notani _ _ , nucl . phys . * a824 * , 192c ( 2010 ) . r. a. broglia and a. winther , _ frontiers in physics lecture notes series : heavy - ion reactions _ ( addison - wesley , redwood city , ca , 1991 ) . s. misicu and h. esbensen . c * 75 * , 034606 ( 2007 ) . h. esbensen , phys . c * 77 * , 054608 ( 2008 ) . c. l. jiang , b. b. back , h. esbensen , r. v. f. janssens , and k.e . rehm , phys . c * 73 * , 014613 ( 2006 ) . c. l. jiang _ et al . * 89 * , 052701 ( 2002 ) . c. l. jiang , k. e. rehm , b. b. back , and r. v. f. janssens , phys . c * 75 * , 015803 ( 2007 ) ; phys . c * 79 * , 044601 ( 2009 ) . d. a. bromley , j. a. kuehner , and e. almquist , phys . lett . * 4 * , 365 ( 1960 ) . e. almquist , d. a. bromley , and j. a. kuehner , phys . lett . * 4 * , 515 ( 1960 ) . g. michaud , phys . c * 8 * , 525 ( 1973 ) . b. dasmahapatra , b. cujec , and f. lahlou nucl . phys . * a384 * , 257 ( 1982 ) . h. esbensen , phys . c * 68 * , 034604 ( 2003 ) . h. esbensen and s. misicu , phys . c * 76 * , 054609 ( 2007 ) . g. bertsch , j. borosowicz , h. mcmanus , and w. g. love , nucl . phys . * a284 * , 399 ( 1977 ) . h. de vries , c.w . de jager and c. de vries , atomic data nuclear data tables , * 36 * , 495 - 536 ( 1987 ) . i. angeli , atomic data and nuclear data tables , * 87 * , 185 - 206 ( 2004 ) . h. esbensen , c. l. jiang , and a. m. stefanini , phys . c * 82*. 054621 ( 2010 ) . evaluated nuclear data structure files , nuclear data center , brookhaven nat . [ http://www.ndc.bnl.gov ] . w. j. vermeer , m. t. esat , j. a. kuehner , r. h. spear , a. m. baxter , and s. hinds , phys . lett . * 122*b , 23 ( 1983 ) . j. lee , m. b. tsang and w. g. lynch , phys . c * 75 * , 064320 ( 2007 ) . j. m. quesada , g. pollarolo , r. a. broglia , and a. winther , nucl . a442 * , 381 ( 1985 ) . h. esbensen and s. landowne , nucl . phys . * a492 * , 473 ( 1989 ) . m. h. macfarlane and s. c. pieper , code ptolemy : argonne laboratory report no . anl-76 - 11 ( 1978 ) . m. g. mazarakis and w. e. stephens , phys . c * 7*,1280 ( 1973 ) . m. d. high and b. cujec , nucl . phys . * 282 * , 181 ( 1977 ) . e. f. aguilera _ et al . _ , c * 73 * , 064601 ( 2006 ) . h. w. becker , k. u. kettner , c. rolfs , h. p. trautvetter , z. phys . * 303 * , 305 ( 1981 ) . t. spillane _ et al . lett . * 98 * , 122501 ( 2007 ) . d. g. yakovlev , m. beard , l. r. gasques , and m. wiescher , phys . c * 82 * , 044609 ( 2010 ) . m. alcorta _ et al . _ , rev . lett . * 106 * , 172701 ( 2011 ) . b. a. brown ( private communications . ) j. zickefoose , u. conn . thesis ( 2010 ) . m. notani , x. tang _ et al . _ , university of notre dame ( unpublished information , 2011 ) . @xmath81 & @xmath80 + @xmath0c & 2.17 & 2.345 & 0.33 & 1 & 2.75 + @xmath0c & 2.28 & 2.407 & 0.31 & 0.843 & 1.00 + @xmath1c ( pp ) & 2.155 & 2.337(2 ) & & & + @xmath0c ( pp ) & 2.146 & 2.332(3 ) & & & + @xmath0c , eq . ( [ rms ] ) & 2.228 & 2.378 & & & + b(e@xmath108 ) ( w.u . ) & @xmath109 + @xmath1c & @xmath110 & 0 & & @xmath55 = -19.5 @xmath19 & 0.570 + & @xmath51 & 4.439 & e2 : @xmath111 & 4.65(26 ) & 0.570 + & @xmath86 & 7.654 & e2 : @xmath58 & 8.0(11 ) & 0.236 + & @xmath88 & 9.641 & e3 : @xmath112 & 12(2 ) @xcite & 0.90(7 ) + @xmath0c & @xmath11 & 0 & & @xmath55 = -17.9 @xmath19 & 0.495 + & @xmath73 & 3.6845 & e2 : @xmath113 & 3.5(8 ) & 0.495 + & @xmath78 & 7.547 & e2 : @xmath114 & 3.1(2 ) & 0.465 + & @xmath74 & 3.8538 & e3 : @xmath115 & 10(4 ) & 0.82 +

we summarize recent results presented in the astrophysics session during a conference on `` frontiers of contemporary physics '' . we will discuss three main fields ( high - energy astrophysics , relativistic astrophysics , and cosmology ) , where astrophysicists are pushing the limits of our knowledge of the physics of the universe to new frontiers . since the highlights of early 1997 were the first detection of a redshift and the optical and x - ray afterglows of gamma - ray bursts , as well as the first well - documented flares of tev - blazars across a large fraction of the electromagnetic spectrum , we will concentrate on these topics . other topics covered are black holes and relativistic jets , high - energy cosmic rays ,-astronomy , extragalactic magnetic fields , and cosmological models . # 1&#2(#3)=@urnal , _ # 1 * # 2 ( @ur 255=#3 ) , * _ @ur255<100 255 by 1900 255 @urnal @=reference foul up : nested macros = you forgot & or ( ) after the last # 1&#2,#3(#4 ) , # 1 * # 2 * , # 3 ( 19#4 ) ps . ps

astro-ph9711066

astrophysics has become one of the most rapidly evolving disciplines in physics over the last two decades . an aggressive expansion , mainly driven by new technology , has pushed the limits for the observer further in photon energy , sensitivity , and spatial resolution and new sub - disciplines are being added to astronomy and astrophysics at a breathtaking speed . young fields like x- and @xmath1-ray astronomy already play an integral role in the scientific community , while tev- and neutrino - astronomy are knocking at the door . though even the established astronomical sub - disciplines have legions of their own frontiers , we will here concentrate on the very forefront of astronomy where astronomy is pushing the very frontiers of physics itself . three of those fields were selected for this workshop : \a ) high - energy astrophysics , where we can study photons and particles from cosmic accelerators with energies way above what can be produced in laboratory on earth unfortunately , we even do not understand those cosmic accelerators yet . \b ) relativistic astrophysics , where we can study general relativity in the strong limit never reached in our solar system . black holes are one example , which most astronomers consider already well established while many physicists wait for the final proof and there is hope . \c ) cosmology , the most fundamental , yet sometimes also most fantastic part of physics , where astronomy provides the basic data and sets the framework for any cosmological model . despite the importance of all those questions , a small workshop like this is of course always dominated by the personal interests of the scientists present and the and the most recent discoveries . the spring of 1997 was especially marked by two major new observational developments : first , the afterglow of the enigmatic gamma ray bursts ( grbs ) was observed in x - rays , the optical and also in the radio . one also obtained the first lower limit for the redshift of a grb , demonstrating fairly convincingly , that grbs are cosmological . second , blazars , this fascinating subclass of quasars , were observed to tens of tev photon energy ( marking the advent of tev astronomy ) , and that with variability of hours . one also observed correlated variability episodes with the optical and x - rays . this gives hope that the basic physical mechanisms underlying these phenomena may be closer to an understanding ; they may be even related after all , and also relate to many other high energy phenomena such as zev cosmic rays and cosmic magnetic fields . therefore the authors of this summary decided not to stand in the way of hot , new trends at a conference like this and gave rather unequal weight to the different topics , putting here more emphasis on these new developments . gamma ray bursts ( grbs ) , short flashes of gamma - photons on the sky , have captured the imagination of theorists for decades after they were initially discovered by satellites to monitor nuclear explosions in space . until early 1997 all we had were speculations . this changed dramatically when the italian - dutch satellite bepposax started to provide sufficiently good positions for new grbs , that for the first time made radio , optical , and x - ray detections of their afterglow possible . the high point of these discoveries was the first redshift of a grb @xcite , demonstrating that , in at least one case , the grb came from a cosmological distance . pre-1997 properties of grbs have been reviewed in hartmann @xcite , and fishman & meegan @xcite . they are isotropically distributed over the sky to all tests made ( using the batse data from the compton gamma ray observatory ) , with the current catalogue of over 1100 bursts , have no repeaters as far as can be stated with certainty , and have a count - rate relation already suggestive of cosmological distances ( a downward deviation from a 3/2 powerlaw in cumulative numbers at low flux levels ) . their _ typical _ fluence is @xmath2 , which translates at cosmological distances to an energy required reminiscent of a supernova @xcite . the modern theoretical attempts to interpret grbs center on ideas by mszros and rees @xcite , as well as paczyski @xcite . in these models a relativistic shock is caused by a fireball expanding into a surrounding medium , such as the interstellar medium or a stellar wind ( or jet ? ) , accelerating electrons / positrons to a very high energy , which then produce the electromagnetic radiation observed in grbs and their afterglow . the low level of associated radiation at other wavelengths limits the baryonic load ( _ i.e. _ , the content in thermal protons , which would be expected to cause thermal radiation ) of the emitting regions to very low amounts , and also constrains the scale of the emitting region to lengths much larger than a neutron star . it appears that all the emission seen is actually non - thermal . such models readily lend themselves to the modification to hadronic particle populations @xcite , so as to accelerate also baryons to very high energy , and use the ensuing hadronic interactions to produce gammas as as well as high energy protons . it then becomes of interest to ask whether a combination of hypothetical models can be built , which produce both very high energy cosmic ray particles and the gamma rays which are observed . we note that there appears to be no correlation between the arrival directions of gamma ray bursts @xcite and the super - galactic plane ( _ i.e. _ a ` local ' galaxy over - density ) , in contrast to the finding for the very high energy cosmic rays @xcite ( see below ) . also , there is no correlation between the gamma - ray bursts of the third batse catalogue as well as the pre - gro bursts and the arrival directions of high energy cosmic rays from the haverah park set , covering approximately the same period of time @xcite . eli waxman showed in his contribution that now quantitative models for the emission observed from grbs are available which can be tested . he was able to provide a lengthy discussion of the recently detected grb 970508 , including the determination of a lower limit to its redshift ( see also @xcite ) . other recent work is described in @xcite @xcite @xcite @xcite and in iau circulars . this list is only exemplary . we briefly outline the essential features of the fireball model @xcite , considering for simplicity only the external shockwave propagating into the surrounding medium : a large amount of energy is released and starts an expansion with a highly relativistic velocity , of order @xmath3 . rather akin to the normal sedov solution for supernova remnants , the heated interstellar medium dominates the expansion after the initial burst . there is a shell of relativistic matter with radius @xmath4 , and thickness @xmath5 in the observers frame ; the density in the shell frame is given by @xmath6 , and the internal energy @xmath7 , disregarding any heavy elements . the heated energy of the ism shell is @xmath8 where @xmath9 is the energy of the explosion , and the factor @xmath10 comes from sharing the energy between ejecta and the ism . the factor @xmath11 in front of the internal energy in the shock frame @xmath12 derives from the transformation to the observers frame . this then gives the evolution of @xmath1 with time @xmath13 , which again is rather reminiscent of the sedov relation for the expansion velocity of a shock in the interstellar medium ( _ e.g. _ @xcite ) . light is emitted over @xmath14 , and the electron population starts at the relativistic thermal energy @xmath15 with @xmath16 being some factor of order unity . the electrons continue to higher energies with a powerlaw determined by the shock . the synchrotron emission has its minimum frequency at @xmath17 , where @xmath18 is the charge of the electron . this translates to @xmath19 and so the temporal behavior of the spectrum is expected to be @xmath20 where @xmath21 is the spectral index of the synchrotron radiation . this is a testable case : for grb 970228 the spectrum was determined between optical and x - ray wavelengths and gives @xmath22 and consequently the temporal behavior has a scaling as @xmath23 to be compared with an observed @xmath24 behavior . this is a magnificent and positive test and was one of the main points of eli waxman s lecture . at the time of writing the optical afterglow has been observed to fade steadily still after six months for grb970228 . particles can be accelerated at internal shocks in the grb outflow , which may be responsible for the rapid variability in some grbs @xcite . eli waxman proposed that in the shocks produced by the explosion both electrons and protons get accelerated ; the absolute limit to acceleration for energetic protons is the spatial constraint , assuming that acceleration is sufficiently rapid : @xmath25 where @xmath1 is the lorentz factor of the shock and the expression is in the shock frame . scaling the magnetic field with the assumption of equipartition ( introducing the factor @xmath26 ) in this model we get @xmath27 this then leads to a maximum proton energy in the shock frame of @xmath28 using the empirical scaling for the numerical relationship between lorentz factor and radius from radio observations @xcite @xmath29 where @xmath30 . the maximum energy written in the observers frame is then given by @xmath31 demonstrating that very high energies are possible . going to the smaller radius , where the grb event first becomes visible , near @xmath32 , where the lorentz factor of the shock is near 300 , obviously we have there @xmath33 to within the uncertainties sufficient to account for the highest energies observed in cosmic rays , if we accept @xmath34 . of course , everything depends then on which density of the interstellar medium is relevant ; if we take the density recently determined from rosat observations for the hot interstellar medium @xcite , of @xmath35 , then the maximum energy is too small , but if we take the average density as determined from neutral hydrogen measurements of order 1 particle per ccm , then the proposal becomes viable . so , eli waxman argued that such high energy protons may account for the observed ultra high energy cosmic rays . one serious question , however , is whether the overall energetics of the fireball are reasonable @xcite or actually exceed the level given by any conceivable model of neutron star mergers or other stellar collapses . eli waxman also went through the exercise to estimate the contribution to a universal high energy neutrino background , which we all hope to see soon with amanda at the south pole . of course , the latest news from the may 8 burst which had come out the night before his lecture dominated everything . the observation that this burst was at cosmological distances , gives one of longest sought answers in astrophysics of the last decade and is a quantum leap in our understanding of the enigmatic grbs . with those observations at hand we can now for the first time test the quantitative predictions of models , like the one presented here . besides grbs , blazars presumably black hole powered relativistic jets , pointing at the observer are the other main source of high - energy photons . they have now been observed at tev energies , where they sometimes have their strongest electromagnetic output and also demonstrate extreme variability down to small fractions of an hour . the key point of these tev observations of blazars is to decide which emission mechanism is the dominant one : are hadronic or leptonic interactions the basis for the high - energy emission ? in the hadronic interaction model protons are accelerated to high energies . here , we assume a shock which accelerates protons and electrons and wish to consider what the maximum photon energy can be before klein - nishina effects or other losses cut off the spectrum for an observer . the shock velocity is taken to be @xmath36 , the magnetic field strength @xmath37 , the fraction of the homogeneous magnetic field energy in turbulence @xmath38 , the natural logarithm of the wavenumber @xmath39 range ratio of turbulence as @xmath40 , then the maximum lorentz factor for protons can be written as @xcite : @xmath41 here the maximum energy is limited by synchrotron losses and we have adopted a saturated spectrum of turbulence with @xmath42 , different from biermann & strittmatter @xcite . we note , that using the limit of a relativistic shock velocity and setting all other parameters to their limits corresponds to setting the acceleration time scale equal to the larmor cycle time , which would seem to be a rather extreme limit ; however , in order to derive a strong upper limit , we will go to this limit below ( see also the arguments on electron maximal energies regarding the crab nebula by de jager and harding @xcite ) . of course , a strongly relativistic shock would modify this argument as seen from an outside observer @xcite . therefore , going to this absolute limit we will set @xmath43 the expression for the maximum lorentz factor of electrons is @xmath44 . the klein nishina cutoff becomes important when the photon energy in the frame of the collision approaches the rest mass of the electron , and so when @xmath45 including the lorentz - factor of the bulk motion @xmath46 we obtain @xmath47 now , the magnetic field strength is expected to approximately scale as @xmath48 therefore , going to this rather extreme limit , and using the region close to the black hole , the maximum photon energy comes out to be @xmath49 this is in fact quite close to the maximum derived in various models in the literature , _ e.g. _ , @xcite . therefore , should the observations conclusively demonstrate that the spectrum continues straight to above 10 tev photon energy ( for which there may be some indication ) , then the leptonic models would begin to have serious difficulties . going through the same argument with hadronic processes , we obtain @xmath50 and so for the same parameters as above , this maximum is @xmath51 as a result , hadronic processes are much more readily able to account for photon energies in the high tev range , _ e.g. _ , @xcite . however , the present observations do not yet allow a final judgment on this question . protheroe & biermann @xcite argued recently that the infrared radiation field from the molecular cloud torus , expected to exist in all agn in the unified scheme " , will terminate all tev photons unless they are emitted above the torus , that is a distance from the central engine of order 0.1 to 1 pc . this argument is safe by a factor of 1000 , _ i.e. _ even if the luminosity of the torus is 1000 times weaker , tev photons still have a hard time to escape . this means that many models in the literature would fail and especially the leptonic models would have problems . however , the feature used in many of these models , a small distance to the central engine , is not always essential , and so we can expect this class of models to get rejuvenated . the final judgment will come from the high energy neutrino observations , which are a firm prediction of the hadronic models . in this vein , wolfgang rhode reviewed the current situation : within the last few years blazars have been investigated with different techniques ( satellites , cherenkov telescopes , air shower arrays ) above photon energies of more than 1 gev . the high energy part ( @xmath5210 tev ) of the spectrum is of special interest , as noted above . first , the detection of photons far above 10 tev would favor accelerated protons as the primary high energy particles . secondly , the distance up to which a @xmath5210 tev photon can be observed depends on the density of infrared photons at @xmath53 ev in the universe . an observation of a distance dependence of the high energetic photon flux from blazars at various distances could thus provide the possibility to measure directly the cosmological density of the infrared ( ir ) background radiation @xcite , and therefore provide a crucial check on the early evolution of galaxies and their activities , both in their starburst mode as well as in their central activity @xcite . it appears that modeling the observed gamma - ray background is consistent with a flat photon spectrum of index -2 @xcite for the agn responsible for the background , consistent with a typical spectrum for blazars . mannheim _ et al . _ @xcite investigated systematically a sample of 13 blazars within the field of view of the hegra ( high energy gamma ray astronomy ) experiment , which were all close enough ( z @xmath54 0.1 ) to be not absorbed at photon energies of several ten tev . this sample was later enlarged to 30 sources . the whipple group discovered two blazars , mrk 421 and mrk 501 , at tev energies @xcite . these observations were confirmed by the two prototype telescopes ( threshold about 1.5 tev ) of the hegra telescope array during the setup phase of the array @xcite . both groups also looked at other sources and established low upper limits @xcite . in spring 1997 mrk 501 was observed to show an eightfold flux enhancement over the last two months before the meeting ( may 1997 ) which was still continuing@xcite . the hegra telescope system observed mrk 421 up to more than 5 tev @xcite , and mrk 501 up to about 20 tev @xcite . as reported elsewhere by a team led by stefan wagner @xcite a campaign to simultaneously monitor the bl lac object mrk 421 at optical , x - ray and gamma - ray wavelengths was done together with the asca - satellite and the whipple observatory . the variations are correlated on time scales of one to several days . in one case the optical brightness has been observed to vary within 60 seconds , putting extreme constraints on any theoretical model . besides the two now famous sources , the sample of 30 blazar sources was further investigated by the hegra team with several independent data sets and analysis techniques @xcite . in autumn 1996 a possible detection of a cumulative signal and a marginal detection of the source 0116 + 319 in three of five data sets were reported by the hegra collaboration @xcite . a test of the z dependence of the significance of the detection of all sources of these data sets suggests , that the detected tev photons for individual blazars with [email protected] was consistent with zero significance , while the distribution of detection significances for tev emission from [email protected] blazars was above zero even though they individually did not reached the 5@xmath55 level . as already pointed out in @xcite two other hegra data sets do not show this behavior . such different results , nevertheless , are still consistent with the present knowledge of the highly variable time structure of the sources . including all systematic uncertainties , the mean maximum photon energy of the blazars as a combined sample is expected to be between 30 tev and 70 tev . at higher energies a detection of signals from this object class would require a large surface array @xcite . the detection of photons of mrk 501 up to about 20 tev now shows that the calculations of the infrared background given in @xcite led to a severe overestimate of the ir photon density . as pointed out in @xcite an independent upper limit can be tentatively calculated by using the fact , that only up to [email protected] significant photon detection excesses were found in the blazar sample . one obtains a ir - photon density close to the lower estimates of macminn and primack @xcite . this conclusion has recently been confirmed by a calculation of berezinsky _ et al . _ @xcite . the high energy tev photons observed with both detector types ( cherenkov telescopes and surface array ) seem presently to suggest that hadronic processes are required in blazars ; we note once again , that high - energy protons are in fact directly observed here on earth . an ultimate test for many of these models is the detection of high energy neutrinos . wolfgang rhode briefly discussed some ramifications of the recent discovery @xcite that the canonical model to explain the gamma - ray emission spectrum of our galaxy assumed to be produced by cosmic rays impinging on thermal gas and dust in the galaxy provides an excellent fit to the spatial variations , but a contradiction with the spectrum . it appears as if the cosmic ray spectrum responsible for the interaction ( p - p collisions leading to pions which decay ) is quite a bit flatter than the spectrum which is both indicated by direct observation and by radio observations of other galaxies . in an extensive collaboration involving groups at nasa , at bartol , in wuppertal and in bonn , it has been shown , that using this inference one can predict anew the neutrino spectrum of the galaxy giving a flux of neutrinos about ten times higher than expected so far . hence through gamma - ray observations we can learn directly something about the in - situ properties of high - energy particles far away from earth . 0.5 cm as discussed by francis halzen in this volume , intense efforts are under way ( amanda ) to directly detect extragalactic neutrinos from hadronic processes in the universe . in this workshop alice l. bean reported on an alternative method to detect these by radio emission from air - showers in ice the rice experiment @xcite . consider a high energy ( @xmath52100 tev ) neutrino coming from underneath the south pole ice sheet , and causing an electromagnetic shower . this shower causes a radio emission pulse to travel in a cone - shaped surface through the ice . in the rice experiment radio antennas have been lowered into the ice , and can pick up any such radio emission . rice will consist of an array of compact radio ( 100 to 1000 mhz ) receivers buried in the ice at the south pole . during the 1995 - 96 and 1996 - 97 austral summers , several receivers and transmitters were deployed in bore holes drilled for the amanda project , at depths of 141 to 260 m. at present alice bean _ _ are testing the setup with a receiver and try to develop algorithms to pick up any signal from the background . only in coincidence with the amanda array can such a nice and old idea be fully developed and properly tested . if it works , it will be a great boost , because radio technology is well developed and mature . the absolute record holder in energy , however , are high - energy cosmic rays , for which the origin and transport through the intergalactic magnetic field were discussed by peter biermann : the recent detection of several cosmic ray events with energies beyond @xmath57 ev is challenging astrophysical theories @xcite . theoretically , the microwave background does not allow particles beyond @xmath58 ev to reach us from cosmological distances @xcite . however , we now have a significant number of clear events of particles with energies beyond this limit even though these highest energy cosmic rays clearly can not be contained in our galactic disk and therefore must originate further outside . biermann briefly sketched the various proposals to explain these high energy events , such as monopoles @xcite , the decay of exotic particles @xcite , shocks in the large scale structure of the universe @xcite , compact objects @xcite , gamma - ray bursts ( see above ) , large scale shocks in our galactic halo @xcite , galaxy collisions @xcite , clusters of galaxies @xcite , active galactic nuclei @xcite , and , specifically , radio galaxies @xcite . in any model in which the cosmic rays arrive from nearby cosmological distances , say , from sources related to galaxies , we can make some strong predictions : the clustering of arrival directions on the sky ought to correspond to the source clustering for energies at which intergalactic scattering by magnetic fields is no longer important , and for which the cosmologically local structure of the universe is still inhomogeneous . above @xmath59 ev the arrival directions of cosmic rays , as seen by the haverah park array @xcite , the akeno array @xcite , and also by a combination of all experiments @xcite , are no longer isotropic , but appear to partially cluster towards the super - galactic plane , the locus of cosmologically nearby normal galaxies , and radio galaxies . some local enhancements of the very high - energy cosmic rays may be due to several identifiable radio galaxies ; one such candidate is the radio galaxy 3c134 @xcite . on the other hand , using the known distribution of candidate sources such as radio galaxies @xcite , and not just the simplified notion of the super - galactic plane @xcite , one can simulate the clustering of arrival directions @xcite : one result is that the clustering to the super - galactic plane as derived from the source locations should be weak , and might not be seriously detectable with the statistics available . clearly the transport of cosmic rays @xcite through the inter - galactic medium and knowledge of cosmological magnetic field strengths is important . simulations @xcite of the formation of cosmological structure , for example allow to determine the spatial inhomogeneity of cosmic magnetic fields . such simulations , however , do not give an absolute number for the strength of the magnetic field . combining these simulations with observations of the `` rotation measure '' ( of polarized light ) to distant radio sources allows then to deduce upper limits for the strength of the magnetic field . these upper limits are of order @xmath60 along the filaments and sheets of the galaxy distribution . in one case , the sheet outside the coma cluster , there is a definitive estimate of the strength of the magnetic field consistent with this range @xcite . such estimates are almost three orders of magnitude higher than hitherto assumed , usually based on derivation using the same data , but assuming a _ homogeneous _ universe ( which we know is wrong ) . high energy cosmic ray particles can be either captured or strongly scattered in such magnetic filaments and sheets , depending on the initial transverse momentum . the cosmological background in radio and x - ray wavelengths will have contributions from these intergalactic filaments and sheets , should the magnetic fields really be as high as @xmath61 . we conclude that the magnetic field structure in the universe is likely to be just as extremely inhomogeneous as the galaxy distribution . hence , some fraction of highly energetic particles should be trapped inside the sheets of the baryonic matter distribution , and so produce the weak correlation with the super - galactic plane detected . : astrophysical jets can be the largest and most impressive signs of the energetic phenomena one commonly associates with black holes and active galactic nuclei ( agn ) and heino falcke reviewed our continuously expanding picture of those gigantic cosmic plasma accelerators . in recent years it was found that , rather than only a minority of sources , almost all types of black holes seem to produce those outflows . besides the well known radio - loud quasars and radio galaxies this includes radio - quiet quasars , seyferts , liners , and x - ray binaries ( stellar mass black holes ) as well . those jets can substantially influence their environment and are often the site for intense energetic phenomena , _ e.g. _ the production of gamma - rays and cosmic - ray particles of the highest energies known today . the energy - budget of those large , extended jets is probably mainly controlled by the accretion rate onto the central object and a still mysterious effect that causes a dichotomy in the jet emissivity relative to the accretion power . heino falcke discussed the evidence for astrophysical jets in various classes of agn and their basic parameters which are crucial for the modeling of all energetic phenomena that have been linked to agn . it is often argued that the escape speed from the central object is an important factor that determines the terminal jet speed . if that is true and since we believe that most of the agn are powered by a black hole one may expect that if an agn produces a jet it should _ always _ be relativistic . consequently the crucial question then becomes : which classes of agn have jets ? in @xcite and @xcite the authors wrote down a hypothesis , simply stating that since black holes do not have many free parameters , agn should be similar in their basic properties ( `` the universal engine '' , @xcite and hence one should _ ab initio _ assume that all agn have relativistic jets rather than only a few sub - classes . as it turned out , this hypothesis , in its simplicity , was surprisingly successful . however , there are interesting difficulties : one finds a clear dichotomy between radio - loud and radio - quiet quasars and it is often assumed that the radio - quiet quasars do not have a relativistic jet at all , while vla observations of the steep - spectrum radio - loud pg quasars @xcite and @xcite have clearly established , that those sources have large scale radio jets . heino falcke therefore asked : what would be the consequences , if radio - quiet quasars too would have relativistic jets ? as for radio - loud quasars , the most prominent sources would be those which are pointing towards us and are relativistically boosted . in an optically selected sample , we would expect that , if radio - quiet quasars have relativistic jets , some of the quasars are accidentally pointing towards us , thus producing a population of ` weak blazars ' with a number of predictable properties . and indeed , miller _ _ @xcite and falcke et al . @xcite were able to identify a small sample of radio - intermediate quasars ( riq ) which met all the requirements for being relativistically boosted , intrinsically _ radio - quiet _ quasars @xcite , thus strongly suggesting that in fact _ all _ rather than just 10% of quasars have relativistic jets . besides radio - quiet quasars , there is another important regime where one should find relativistic jets . quasars are powered by black holes with high accretion rates , but what happens if the accretion rate decreases ? will those jets die completely ? ho _ et al . _ @xcite found that roughly one half of all nearby galaxies show signs of optical activity in the nucleus , in the form of liner or seyfert spectra , and quite a number also show nuclear radio emission . this could indicate the presence of a black hole engine @xcite . falcke and collaborators conclude @xcite that indeed the radio cores in liners are part of the central engine since optical and radio fluxes are correlated . moreover , we can compare the radio and emission - line luminosities with the jet / disk model by falcke & biermann @xcite . the model predicted a specific radio / nuclear luminosity correlation for low - power agn and is based on the assumption that accretion disk luminosity and jet power in agn are coupled by a universal constant . the liners fall exactly into the range predicted for low - luminosity , _ radio - loud _ jets . this result not only strongly suggests that liners do have powerful nuclear radio jets ( _ e.g. _ m87 ; ngc4258 , @xcite ; m81 , @xcite , etc . ) but is also consistent with mildly relativistic lorentz factors around @xmath62 as used in the model . that should be compared with lorentz factors of @xmath63 derived with the same method for radio - loud quasars @xcite . 0.5 cm paul wiita reported on work done with gopal - krishna and vasant kulkarni ( both at the ncra of the tata institute at pune , india ) on the status of the so called unified scheme " which is used to explain the differences of the most powerful radio jets just in terms of different orientation and obscuration : a key argument in favor of orientation based unification schemes is the finding that among the most powerful 3crr radio sources the ( apparent ) median linear size of quasars is smaller than that of radio galaxies , which supports the idea that quasars are a subset of radio galaxies , distinguished by being viewed at smaller angles to the line of sight . recent measurements of radio sizes for a few other low frequency samples are , however , not in accord with this trend , leading to the claim that orientation may not be the main difference between radio galaxies and quasars . wiita pointed out that this `` inconsistency '' can be removed by making allowance for the temporal evolution of sources in both size and luminosity , as inferred from independent observations . this approach can also readily explain the other claimed `` major discrepancy '' with the unified scheme , namely , the difference between the radio luminosity size correlations for quasars and radio galaxies . some of this work is reported in @xcite . 0.5 cm * black holes : * the scales of relativistic jets are still large compared to the event horizon of a black hole , but even black holes are slowly coming into sight ! _ falcke pointed out that the mass of the black hole in the center of our galaxy is now determined with a very a high precision ( @xmath64 ) by the measurement of proper motion of stars @xcite , pushing the central dark mass density to @xmath65pc@xmath66 ( _ i.e. _ the mass of a whole galaxy concentrated in less then the volume between the solar system and the next stars ) . any alternatives to a black hole , _ e.g. _ an ultra compact cluster of stellar remnants seems to be ruled out now @xcite thus making the galactic center source sgr a * the best supermassive black hole candidate today . interestingly , the mass is so large that the photon horizon of this black hole has a diameter corresponding to an angular resolution of 27 @xmath67arcsecond . such a resolution will in fact be reached by planned vlbi ( very long baseline interferometry ) experiments operating at 220 ghz . since this black hole also has an ultra compact emission region estimated to be of a similar size , radiating at just this frequency , this leaves the tantalizing possibility to directly image the horizon of a black hole at least in one case . 0.5 cm paul wiita also reported on work done with gang bao and ying xiong , ( also at georgia state university ) and with petr hadrava ( at the astronomical institute of the czech academy ) on polarization variability as a signature of black holes : in regions where electron scattering dominates the opacity above accretion disks , x - ray radiation originating there should be partially linearly polarized . both observations of rapid x - ray variability and theoretical studies suggest that this inner disk region is unstable and could appear clumpy . he showed how variations in the orbital parameters of the bright spots and the angle between the line of sight and the disk axis affect the observed polarization . the amplitudes of both the changes in the degree of polarization and the angle of the plane of polarization are energy - dependent . they are relatively independent of the physical mechanism producing the polarization . this feature is directly created by the gravitational bending of light rays by the central black hole and it is apparently unique to a system including a black hole and an accretion disk . this work is reported in @xcite . 0.5 cm dallas kennedy reported on work done with k. s. gopinath ( u. florida ) and j. m. gelb ( u. texas ) on relativistic landau states of electrons in intense astrophysical magnetic fields , encountered especially in neutron stars at @xmath68 @xmath69 gauss ( lanl astro - ph/9702014 and astro - ph/ 9703108 ) . the classical and semi - classical orbits of relativistic charged particles were outlined for motion on a spherical surface , in an intense magnetic dipole background . the dipole and rotational axes in general should not be aligned , if the star s magnetic dynamo is to be self - sustaining . kinematic regimes differ depending on the relative sizes of energy , canonical azimuthal angular momentum , and magnetic field strength in rescaled units . magnetic flux enclosed by the orbits is quantized very close to the poles . open questions relating to the state of electronic matter near neutron star surfaces were sketched . subsequent work has extended this calculation to the full 3-d problem at finite density , with electrons in local field - transverse planes , but including the gravitational and hadronic structure only as given backgrounds ( lanl astro - ph/9707196 ) . further questions include the magnetically - induced structural changes below the star surface and changed state of matter , as well as observational signals of such exotic `` quantum hall - like '' surface physics . a larger and perhaps more entertaining review was given at this conference by rocky kolb . here we summarize some of the interesting twists to cosmological models presented during the astrophysics workshop . david hochberg reported on course - graining , structure formation and the transition to large - scale homogeneity in the universe @xcite : newtonian hydrodynamics plus frw cosmology combine to yield a good description of the matter dominated universe at large scales . under assumptions of vorticity - free flow and validity of the zeldovich approximation ( that the gravitational acceleration is parallel to the velocity ) the ensuing dynamics can be re - cast in terms of a cosmological version of the ( massive ) kardar - parisi - zhang ( kpz ) equation , which has enjoyed extensive application in the study of surface growth phenomena . here , he applied it to the problem of the growth and distribution of large - scale structure in the universe . using the well established techniques of the dynamic re - normalization group to study the scaling properties of the solutions of the kpz equation , he calculated the power law behavior and attendant exponent of the galaxy to galaxy correlation function and showed that the transition to large scale homogeneity is an necessary consequence of the course - graining . 0.5 cm arjun berera reported on an attempt to determine the largest scale of primordial density perturbations beyond the hubble radius with cobe - dmr and the implications for early universe cosmology : causality imposes rigorous constraints which suppress super - horizon scale coherence . the power spectrum of scalar primordial density perturbations is modified by inclusion of a super - hubble suppression scale in order to respect this constraint mandated by causality . a recent analysis of cobe - dmr data was presented , in which measurements were made of the super - hubble suppression scale , the spectral index and the amplitude . theoretical implications of this analysis focus on the warm inflation scenario , which in part motivated this cobe analysis . a summary of the scenario was presented which included discussion about self - consistency , avoidance of a re - heating period , relevance to open universe and primordial seeds of density perturbations and magnetic fields . although the details of the scenario are model dependent , this so identified regime is an outcome only of friedmann cosmology . he asked if this modification would have any significant effects on the mechanisms to produce large magnetic fields . 0.5 cm as mention in the previous cosmic ray section , the origin of magnetic fields is still an elusive goal in our understanding of the universe ; many models have been proposed , from dynamos working in stars , in accretion disks , in entire galaxies , and in large scale accretion flows ; primordial magnetic fields have many attractive features , since they may obviate some of the difficulties faced by the dynamo models . recently improved instrumental capabilities over the past decade or so have greatly improved our knowledge of the extent and strength of cosmic magnetic fields . some surprising discoveries have emerged and phil kronberg summarized our latest knowledge of the strength and extent of magnetic fields in galaxies , galaxy clusters , and what little inkling has been gained about widespread intergalactic magnetic fields . the status of our knowledge is well described by kronberg @xcite in two excellent reviews . phil kronberg included some basic theory of magnetic field regeneration , and some constraints imposed by recent experimental data . kronberg also explained the observational methods of probing magnetic fields in outer space , and mentioned future prospects linked to experiments with gamma- and cosmic ray detectors , and at low frequency radio and sub - millimeter bands . the information came from measurements of the faraday rotation of the plane of linear polarization turned by the transfer through an ionized and magnetized medium . this rate of angle rotation with increasing wavelength is defined as rotation measure ( rm ) ( radians / m@xmath70 ) , and is proportional to the integral of the electron density along the line of sight , folded with the parallel magnetic field components . on the sky most of the effects are dominated by our own galaxy . looking further one finds the following : \a ) in normal galaxies such as our own the magnetic fields are of order 3 to 10 @xmath67gauss @xcite , with a lower limit of about 1 @xmath67gauss . \b ) in the halo of our galaxy the magnetic field is @xmath71 0.2 @xmath67gauss , over a scale of a few kpc @xcite . \c ) in starburst galaxies such as m82 the magnetic field can be very much stronger , with values up to @xmath71 30 @xmath67gauss possible @xcite . \d ) in clusters of galaxies the magnetic field strength is subject to some uncertainty in the number of magnetic field reversals along lines of sight through the cluster @xcite , but the overall rotation of the polarization plane of background radio sources suggests values of a few @xmath67gauss , while some cooling - flow clusters with strong radio galaxies are known to have values about ten times higher . \e ) outside one cluster , the coma cluster , kim _ et al . _ @xcite have estimated the strength of the magnetic field near 0.1 @xmath67gauss , assuming equipartition between magnetic fields and relativistic particles . \f ) across cosmological distances there is only an _ upper limit _ of @xmath71 1 nanogauss , derived from rotation measure data measured along lines of sight to cosmologically distant radio - quasars @xcite , assuming a reversal scale of the magnetic field structure of 1 mpc , and otherwise cosmologically homogeneous properties of the intergalactic medium . if we were to assume that the reversal scale is the same as the bubble structure of the galaxy distribution , then this limit would be @xmath71 200 picogauss . \g ) we do not ( yet ) have a successful theory to account for the origin of cosmological magnetic fields , neither in galaxies , nor in clusters . stars such as the sun show evidence for a fast dynamo acting to reverse the magnetic field every 11 years , and so it is often assumed , but has not finally been demonstrated , that galaxies do the same on the much larger scale @xcite . in clusters of galaxies , the magnetic field can energetically be provided by the radio galaxies @xcite . the observation of normal magnetic fields in a galaxy at fairly large redshift suggests that magnetic fields can build up over cosmologically fairly short time scales @xcite . on the other hand , the strength and topology of magnetic fields in galaxies appears to exclude an origin as simply primordial @xcite . one recent attempt to simulate the growth of cosmological magnetic fields has been made by kulsrud and collaborators @xcite , using the structure formation itself as the source in a battery process @xcite . another argument has been that massive stars and compact accretion disks in agn may be all that is needed , also starting with a battery process @xcite . finally , ralston reported on his much debated claim of cosmological birefringence , published with nodlund . ralston asked the question , whether there is bi - refringence of the universe , based upon an analysis of radio galaxy data @xcite . if so , this would have been a major change for our understanding of the universe @xcite . he finds a big effect for redshifts larger than 0.3 from a large database of previously published measurements of polarization vectors of galaxies . while the proposal has spawned an intense debate with a number of counter examples on the lanl electronic preprint server @xcite and in the popular press , ralston himself concluded barring hidden systematic effects , the analysis indicates a new cosmological effect " . clearly the debate demonstrates , that most researchers still would put more weight on the first part of the sentence and favor a systematic effect in the analysis of the data to interpret nodland s & ralston s findings , before making drastic changes to existing cosmological models . compact objects from the centers of the gamma ray bursts to the the presumed black holes at the focus of the agn span about ten powers of ten in mass . their associated outflows , whether explosive , non - steady , steady or quiescent , have been the focus of much work over the past several decades . those are the laboratories where new physics can and will be learned . it appears that basically the same concepts may help us to understand much of what we observe : everywhere we look at a compact object such as a black hole , there seems to be an accretion flow ( disk ) and a jet , with gigantic and relativistic shock waves running through these jets , and accelerating particles . another common analogy is simply the formation of a compact object with a concurrent explosion , be it as a supernova , or as a gamma ray burst . in the case of a gamma ray burst again a relativistic flow seems indicated , and we come back to the basic agn language , and maybe the same physics . many questions still remain : are the zev energy particles , presumably protons , really derived from shock waves in such jets from the most powerful radio galaxies we know ? or do they come from the gamma ray bursts ? or , most excitingly perhaps , do they come from the decay of particles of gut - scale energies ? are the emissions that we observe in gamma ray bursts or in tev blazars , all derivable from leptonic processes , or do they require hadronic processes to get started ? what will neutrino astronomy bring us ? can we connect all these observations to the structure formation of the universe , whether it is the cosmic magnetic fields or the first seeds of black holes ? these and many more tasks are waiting for us and thanks to impressive developments in recent years , _ we can be certain that the `` golden years of astrophysics '' are not over yet_. pb and hf wish to thank all participants for the discussions at the meeting as well as before and afterwards . for this article we have in part used some of the material the participants have generously provided to us . pb also wishes to thank karl mannheim , giovanna pugliese and jrg rachen for comments , as well as abhas mitra , tsvi piran , ray protheroe , gnther sigl , and todor stanev for intense discussions on various topics also covered in this workshop . beck , r. , _ et al . _ : 1996 _ ann . astroph . _ * 34 * , 155 - 206 berera , a. , fang , l.z . : 1994 _ phys . letters _ * 72 * , 458 - 461 berezinsky v.s . , grigorueva s.i . : 1988 , _ astron . & _ * 199 * , 1 - 12 . biermann , p.l : : 1997 _ j. of physics g _ * 23 * , 1 - 27 biermann , p.l , kang , h. , ryu , d. : 1997 in proc . `` extremely high energy cosmic rays '' , tokyo , ed . m. nagano , p. 79 - 88 , astro - ph/9709250 biermann , p.l . , _ et al . _ : 1997 in proc . moriond meeting ( january 1997 ) in press , astro - ph/9709252 bietenholz , m.f . , bartel , n. , rupen , m.p . , _ et al . _ : 1996 _ astroph . j. _ * 457 * , 604 - 609 bttcher , m. , schlickeiser , r. : 1996 _ astron . & _ * 306 * , 86 - 96 carroll , s.m . , field , g.b . : 1997 astro - ph/9704263 ( in press ) cesarsky , c. , ptuskin , v. : 1993 23rd icrc , university of calgary , vol . 2 , og 9.3 , p. 341 - 344 . costa , e. _ et al . _ : 1997 _ nature _ * -387 * , 783 - 785 , astro - ph/9706065 cox , d.p . : 1972 _ astrophys . j. _ * 178 * , 159 - 168 dar , a. : 1997 astro - ph/9709231 daum , k. _ et al . _ : 1995 _ zeitschrift f. phys . c _ * 66 * , 417 . dermer , c.d . , schlickeiser , r. : 1992 _ science _ * 257 * , 1642 - 1647 . falcke , h. , biermann , p.l . : 1995 _ astron . & _ * 293 * , 665 - 682 . falcke , h. , malkan , m.a . , biermann , p.l . : 1995 _ astron . & _ * 298 * , 375 - 394 . falcke , h. 1996a , in : `` jets from stars and galactic nuclei '' , lecture notes in physics 471 , w. kundt ( ed . ) , springer , p. 1934 fishman , g.j . , meegan , c.a . : 1995 _ ann . _ * 33 * , 415 - 458 frail , d.a . , _ et al . _ : 1997 _ nature _ * 389 * , 261 - 263 gaisser , t.k . : 1990 _ cosmic rays and particle physics _ , cambridge univ . press galama , t. _ et al . _ : 1997 _ nature _ * 387 * , 479 - 481 geddes , j. , quinn , t.c . , wald , r.m . : 1996 _ astrophys . j. _ * 459 * , 384 - 389 . goldman , t. , hochberg , d. , laflamme , r. and perez - mercader , j. : 1996 _ physics letters _ * a 222 * , 177 guarnieri , a. _ et al . _ : 1997 ( in press ) astro - ph/9707164 gurbatov , s.n . , saichev , a.i . , shandarin , s.f . : 1985 _ sov . dokl . _ * 30 * , 921 - 923 gurbatov , s.n . , saichev , a.i . , shandarin , s.f . : 1985 _ monthly not . soc . _ * 236 * , 385 - 402 han , j.l , qiao , g.j . : 1994 _ astron . & _ * 288 * , 759 - 772 hartmann , d.h . : 1994 in _ high energy astrophysics _ , ed . j.m . matthews , world scientific , p. 69 - 106 hartmann , d.h . , briggs , m.s . , mannheim , k. : 1995 _ astrophys . j. _ * * ( in press ) hayashida , n. , _ et al . _ : 1996 _ phys . letters _ * 77 * , 1000 - 1003 heckman , t.m . : 1980 _ astron . & astroph . _ * 87 * , 152 hermann , g. et al . , xxv icrc97 , durban , og 10.3.19 ( ? ? ? ? ) herrnstein , j.r . et al . _ : 1997 _ astrophys . j. letters _ * 475 * , l17 - l20 hillas , a.m. : 1984 _ ann . astrophys . _ * 22 * , 425 - 444 . ho , l.c . , filippenko , a.v . , sargent , w.l.w . : 1995 _ astroph . j. suppl . _ * 98 * , 477 ho , l.c . , _ et al . _ : 1997 astrophys j. , in press hochberg , d. and perez - mercader , j.:1996 _ general relativity and gravitation _ * 28 * , 1427 hunter , s.d . _ et al . _ : 1997 _ astrophys . j. _ * 481 * , 205 - 240 hurley , k. _ et al . _ : 1997 _ astrophys . j. letters _ * 485 * , l1 - l4 , astro - ph/9705128 iau circular no . * 6592 * , 19th march 1997 . jager , o.c . de , harding , a.k . : 1992 _ astrophys . j. _ * 396 * , 161 - 172 jokipii j.r . , morfill g. : 1987 , _ astrophys . j. _ * 312 * , 170 - 177 kang , h. , ryu , d. , jones , t.w . : 1996 _ astroph . j. _ * 456 * , 422 - 427 . kang , h. , rachen , j.p . , biermann , p.l . : 1997 _ monthly not . * 286 * 257 - 267 karle , a. _ et al . _ : 1995 _ physics letters b _ * 347 * , 161 - 170 kellermann , k.i . , sramek , r. , schmidt , m. , _ et al . _ : 1994 _ astron . j. _ * 108 * , 1163 kephardt , t.w . , weiler , t.j . : 1996 _ astroparticle phys . _ * 4 * , 271 - 279 kim , k .- et al . _ : 1989 _ nature _ * 341 * , 720 - 723 . et al . _ : 1990 _ astrophys . j. _ * 355 * , 29 - 37 . et al . _ : 1991 _ astrophys . j. _ * 379 * , 80 - 88 . krawczynski , h. _ et al . _ : 1997 astro - ph/9611044 krennrich , f. _ et al._:1997 _ astroph . j. _ * 481 * , 758 - 763 gopal krishna , wiita , p.j . , hooda , j.s . : 1996 _ astron . & _ * 316 * , l13 - l16 gopal krishna , kulkarni , v.k . , wiita , p.j . : 1996 _ astrophys . j. letters _ * 463 * , l1 - l4 gopal krishna , wiita , p.j . : 1996 _ astrophys . j. _ * 467 * , 191 - 196 kronberg , p.p . _ et al . _ : 1985 _ astrophys . j. _ * 291 * , 693 - 707 kronberg , p.p . , _ et al . _ : 1992 _ astrophys . j. _ * 387 * , 528 - 535 kronberg , p.p . : 1994 _ rep . phys . _ * 57 * , 325 - 382 . kronberg , p.p . : 1997 in extremely high energy cosmic rays " , university of tokyo , ed . m. nagano , p. 89 - 103 kuehne , r. : 1997 astro - ph/9708109 kulsrud , r.m . , cen , r. , ostriker , j.p . , ryu , d. : 1996 _ astrophys . j. _ * 480 * , 481 - 491 macdonald , j. , stanev , t. , biermann , p.l . : 1991 _ astrophys . j. _ * 378 * , 30 - 38 . macminn , d. and primack , j. : 1996 _ space sci . _ * 75 * , 413 . magnussen , n. _ et al . _ : 1997 , xxv icrc97 , durban , og 4.3.15 mallozzi , r. _ et al . _ : 1995 _ astrophys . j. _ * 454 * , 597 - 603 mannheim , k. biermann , p.l . : 1989 _ astron . & _ * 221 * , 211 - 220 . mannheim , k. , krlls , w. m. , biermann , p.l . : 1991 _ astron . & _ * 251 * , 723 - 731 . mannheim , k. , stanev , t. , biermann , p.l . : 1992 _ astron . & astroph . _ * 260 * , l1 - l3 . mannheim , k. : 1993 _ astron . & astroph . _ * 269 * , 67 - 76 . mannheim , k. : 1993 _ phys . * d 48 * , 2408 . mszros , p. & rees , m. : 1994 _ mon . not . r. astron . letters _ * 269 * , l41 - l43 . mszros , p. , rees , m.j . : 1997 _ astrophys . j. _ * 476 * , 232 - 237 mszros , p. , rees , m.j . : 1997 _ astrophys . j. letters _ * 482 * , l29 - l32 metzger , m.r . _ et al . _ : 1997 _ nature _ * 387 * , 878 - 880 meyer , h. _ et al . _ : 1996 , proc . of the heidelberg workshop on gamma - ray emitting agn ( 1996 ) , 37 . milgrom , m. , usov , v. : 1995 _ astroph . j. letters _ * 449 * , l37 - l40 . milgrom , m. , usov , v. : 1995 _ astroparticle physics _ * 4 * , 365 - 369 . miller , p. , rawlings , s. , & saunders , r. : 1993 _ monthly not . soc . _ * 263 * , 425 miralda - escud , j. , waxman , e. : 1996 _ astrophys . j. _ * 462 * , l59 - l62 . mitra , a. : _ astroph . j. _ * * ( in press , astro - ph/9710063 ) nagano , m. : 1997 editor , extremely high energy cosmic rays " tokyo , university of tokyo nodlund , b. , ralston . j.p . : 1997 _ phys . letters _ * 78 * , 3043 - 3046 nodlund , b. , ralston , j.p . : 1997 _ phys . letters _ * 79 * , 1958 norman , c.a . , melrose , d.b . , achterberg , a. : 1995 _ astrophys . j. _ * 454 * , 60 - 68 paczyski , b. : 1986 _ astrophys . j. letters _ * 308 * , l43 - l46 paczyski , b. : 1997 _ astrophys . j. letters _ ( submitted , astro - ph/9710086 ) panaitescu , a. , mszros , p. : 1997 astro - ph/9703187 panaitescu , a. _ et al . _ : 1997 _ astrophys . j. _ * 482 * , astro - ph/9610064 paradijs , j. van _ et al . _ : 1997 _ nature _ * 386 * , 686 - 689 parker , e.n . : 1958 _ astrophys . j. _ * 128 * , 664 - 676 . parker , e.n . : 1992 _ astrophys . j. _ * 401 * , 137 - 145 peacock , j.a . : 1981 _ monthly not . soc . _ * 196 * , 135 - 162 petry , d. _ et al . _ : 1996 _ astron . & astroph . _ * 311 * , l13 . petry , d. _ et al . _ : 1997 , xxv icrc97 , durban , og 4.3.1 piran , t. : 1992 _ astrophys . j. letters _ * 389 * , l45 - l48 piro , l. _ et al . _ : 1997 ( a&a , in press ) astro - ph/9707215 protheroe , r.j . , szabo , a.p . : 1992 _ phys . letters _ * 69 * , 2885 - 2888 . protheroe , r.j . , johnson , p.a . : 1996a _ astroparticle physics _ * 4 * , 253 - 269 ( err . * 5 * , 215 ) protheroe , r.j . , johnson , p.a . : 1996b in `` proc . 4th international workshop on theoretical and phenomenological aspects of underground physics ( taup95 ) '' , ed . m. fratas , _ nucl b. , proc . _ * * , ( in press ) protheroe , r.j . , stanev , t. : 1996 _ phys . rev . letters _ ( submitted , astro - ph9605036 ) protheroe , r.j . , biermann , p.l . : 1996 _ astroparticle physics _ * * ( in press , astro - ph/9608052 ) . m. punch , m. _ et al . _ : 1992 _ nature _ * 358 * , 477 . rhode , w . _ et al . _ : 1996 _ astroparticle phys . _ * 4 * , 217 - 225 rhode , w. : 1997 to appear in the proc . of the xxxiind rencontres de moriond . rhode , w. , meyer , h. : 1997 in proc . `` relativistic jets in agn '' , cracow 1997 , ostrowski et al . agn " , cracow 1997 , ostrowski et al . ( eds . ) sahu , k.c . _ et al . _ : 1997 _ nature _ * 387 * , 476 - 478 , astro - ph/9705184 sahu , k.c . , livio , m. , petro , l. : 1997 _ astrophys . j. _ * * ( in press ) astro - ph/9706225 sari , r. : 1997 astro - ph/9706078 shaver , p.a . , pierre , m. : 1989 _ astron . & _ * 220 * , 35 - 41 . shaver , p.a . : 1991 _ austral . _ * 44 * , 759 - 769 shemi , a. : 1995 _ monthly not . soc . _ * 275 * , 115 - 120 . sigl , g. , _ et al . _ : 1995 _ science _ * 270 * , 1977 - 1980 . sikora , m. _ et al . _ : 1997 _ astroph . j. _ * 484 * , 108 - 117 snowden , s.l . _ et al . _ : 1997 _ astrophys . j. _ * 485 * , ( in press ) mpe preprint no . 390 spinrad , h. , _ et al . _ : 1985 _ publ . pacific _ * 97 * , 932 - 961 . sreekumar , p. _ et al . _ : 1997 _ astrophys . j. _ * * ( in press ) , astro - ph/9709257 stanev , t. , biermann , p.l . , & gaisser , t.k . : 1993 , _ astron . & . _ * 274 * , 902 - 908 . stanev , t. , _ et al . _ : 1995 _ phys . letters _ * 75 * , 3056 - 3059 . stanev , t. , schaefer , r. , watson , a. : 1996 _ astroparticle physics _ * 5 * , 75 - 78 stanev , t. : 1997 _ astrophys . j. _ * 479 * , 290 - 295 , stanev , t. , franceschini , a. : 1997 preprint stecker , f. : 1968 _ phys . rev . letters _ * 21 * , 1016 - 1018 . stern , b. , poutanen , j. : 1997 ( in press ) astro - ph/9703167 stern , b. , svensson , r. , poutanen , j. : 1997 astro - ph/9701171 uchihori , y. _ et al . _ : 1997 , in proc . extremely high energy cosmic rays " , ed . m. nagano , university of tokyo meeting , p. 50 - 60 vietri , m. : 1995 _ astrophys . j. _ * 453 * , 883 - 889 vietri , m. : 1996 _ montly not . soc . _ * 278 * , l1 - l4 vietri , m. : 1997 ( submitted ) astro - ph/9706060 vietri , m. : 1997 _ astrophys . j. letters _ * 478 * , l5 - l8 , astro - ph/9701024 wagner , s.j . : 1997 in proc . `` relativistic jets in agn '' , cracow 1997 , ostrowski et al . blazars wagner , s.j . _ et al . _ : 1997 _ astrophys . j. letters _ * * ( in press ) wardle , j.f.c . , perley , r.a . , cohen , m.h . : 1997 _ phys . letters _ * 79 * , 1801 - 1804 waxman , e. : 1995a _ phys . letters _ * 75 * , 386 - 389 . waxman , e. : 1995b _ astrophys . j. letters _ * 452 * , l1 - l4 waxman , e. , fisher , k.b . , piran , t. : 1996 _ astrophys . j. _ * 483 * , 1 - 7 ( astro - ph/9604005 ) waxman , e. : 1997 astro - ph/9701231 waxman , e. : 1997 , _ astrophys . j. letters _ * 485 * , l5 - l8 , astro - ph/9704116 waxman , e. : 1997 astrophys . j. letters * 489 * , ( in press ) , astro - ph/9705229 waxman , e. : 1997 ( to appear in proc . icrr symposium on extremely high energy cosmic rays ) astro - ph/9612061 waxman , e. : 1997 _ astrophys . j. letters _ ( submitted ) , astro - ph/9709190 waxman , e. , kulkarni , s.r . , frail , d.a . : 1997 _ astrophys . j. _ * * ( submitted ) astro - ph/9709199 westerhoff , s. _ et al . _ : 1995 _ ap . * 4 * , 119 zatsepin g.t . , kuzmin v.a . : 1966 , _ jetph lett . _ * 4 * , 78 .

we present extensive monte - carlo spin dynamics simulations of the classical xy model in three dimensions on a simple cubic lattice with periodic boundary conditions . a recently developed efficient integration algorithm for the equations of motion is used , which allows a substantial improvement of statistics and large integration times . we find spin wave peaks in a wide range around the critical point and spin diffusion for all temperatures . at the critical point we find evidence for a violation of dynamic scaling in the sense that independent components of the dynamic structure factor require different dynamic exponents in order to obtain scaling . below the critical point we investigate the dispersion relation of the spin waves and the linewidths of and find agreement with mode coupling theory . apart from strong spin wave peaks we observe additional peaks in which can be attributed to two - spin wave interactions . the overall lineshapes are also discussed and compared to mode coupling predictions . finally , we present first results for the transport coefficient of the out - of - plane magnetization component at the critical point , which is related to the thermal conductivity ofhe near the superfluid - normal transition .

cond-mat9905295

the theoretical investigation of classical spin systems has played a key role in the understanding of phase transitions , critical behavior , scaling , and universality @xcite . in particular , the claasical ising , the xy , and the heisenberg model are the most relevant spin models in three dimensions . each of these simple models represents a universality class which , apart from the spatial dimensionality and the range of the interactions , is characterized by the number of components @xmath3 of the order parameter , e.g , the magnetization in the case of ferromagnetic models . despite their simplicity these spin systems continue to be of high relevance within the framework of _ dynamic _ behavior near critical points @xcite ( see also ref.@xcite for a recent review ) . the ising @xmath4 , the xy @xmath5 , and the heisenberg @xmath6 universality class can be extended towards _ dynamic _ universality classes , which in addition to their static properties are characterized by the set of conservation laws @xcite . special attention must be paid to the presence of energy conserving driving terms in the equations of motion which lead to propagating modes ( spin waves ) below the critical temperature and thus modify the dynamics @xcite . the discrete nature of ising spins does not allow such terms so that its dynamics is always of relaxational type ( see ref.@xcite for a complete classification ) . the simplest spin model which allows propagating modes is a particular version of the ferromagnetic xy model @xmath5 . the dynamics here is characterized by a nonconserved order parameter which is dynamically coupled to a _ conserved _ quantity ( see sec.ii ) . the presence of spin waves reduces the value of the dynamic critical exponent @xmath7 as compared to pure critical relaxation @xcite . the same is true for the isotropic heisenberg model for which the @xmath8 component magnetization vector is always conserved in the presence of energy conserving driving terms . if the model is ferromagnetic , the magnetization is the order parameter . however , for an _ antiferromagnet _ , the magnetization plays the role of a conserved vector which is dynamically coupled to the _ nonconserved _ order parameter ( staggered magnetization ) . this difference in the conservation laws causes the classical heisenberg ferro- and antiferromagnet to be in different dynamic universality classes although they belong to the same static universality class . due to their fundamental role in the understanding of the critical dynamics in magnets heisenberg ferro- and antiferromagnets have been thoroughly studied analytically by mode coupling theories ( see ref.@xcite for a general overview ) especially in presence of dipolar interactions @xcite and numerically by spin dynamics in @xmath9 @xcite and in @xmath10 @xcite and by methods closely related to molecular dynamics @xcite . the xy model may be viewed as a heisenberg ferromagnet with an easy - plane @xmath11 anisotropy such that the order parameter has only two components . planar ferromagnets are realized by layered compounds such as k@xmath12cuf@xmath13 @xcite and rb@xmath12crcl@xmath13 @xcite which almost act as two - dimensional systems . the best results available today have been obtained on cocl@xmath12 intercalated in graphite @xcite , where a crossover from two - dimensional to three - dimensional behavior in the correlations has been observed below the kosterlitz - thouless temperature . apart from the evident interpretation as a planar ferromagnet the xy model captures a larger variety of phenomena than the ising or the heisenberg model . despite its continuous @xmath14 symmetry the xy model undergoes a continuous phase transition at a finite temperature in two dimensions , known as the kosterlitz - thouless transition @xcite . rather than by the onset of long - ranged order the transition is solely characterized by a diverging correlation length , when the critical temperature is approached from above . due to a peculiar conspiracy between the spatial dimensionality @xmath9 and the number of spin components @xmath15 configurations of bound and free vortices dominate the critical behavior of the xy model , where the unbinding of vortex pairs marks the point of the phase transition @xcite . naturally , many attempts to describe the critical dynamics of the xy model in @xmath9 theoretically are based on the dynamics of vortices and vortex pairs @xcite . according to analytical and numerical investigations for the ferromagnetic case @xcite the in - plane component @xmath16 and the out - of - plane component @xmath17 of the dynamic structure factor are expected to have central peaks above the transition . the line shapes of these peaks are predicted to be squared lorentzian and gaussian , respectively . below the transition only spin - wave peaks are expected @xcite . to test these specific predictions much numerical effort has been spent on spin dynamics simulations of the xy model in @xmath9 @xcite . although dynamic finite - size scaling and the value of the dynamical exponent @xmath18 has been confirmed to a high degree of confidence @xcite , the measured lineshapes of @xmath0 @xcite are apparently not well captured by analytical theories ( see ref.@xcite for details ) . it is therefore instructive to measure @xmath0 for the xy model in @xmath10 for which configurations of bound or free vortices do not play any special role for the critical behavior and should therefore not provide particularly noticable contributions to the structure factor . in @xmath10 the dynamics of the planar ferromagnet has been investigated by mode coupling theory @xcite and specific predictions concerning line shapes and -widths have been made which can be compared with our data ( see sec.iv ) . it is well known that the @xmath19-transition of @xmath2he is the in the xy universality class , but the applications of the xy model for the physics of @xmath2he reach far beyond that . the spin dynamics for the xy model is the lattice analogue of the dynamical model e ( symmetric planar ferromagnet @xcite ) which _ asymptotically _ also describes the _ critical dynamics _ of @xmath2he near the @xmath19-line @xcite . if one therefore studies the transport properties of the xy model near the critical point @xmath20 , one should obtain lattice analogues of the corresponding transport coefficients of @xmath2he near the @xmath19-transition . in this respect the aforementioned conserved quantity plays a particularly interesting part , because it is related to the entropy density in @xmath2he and its associated transport coefficient corresponds to the thermal conductivity of @xmath2he @xcite which is an experimentally accessible quantity @xcite . below the critical temperature spin waves in the xy model then correspond to travelling waves of second sound in @xmath2he . these propagating modes cause the thermal conductivity to diverge at the lambda transition of bulk @xmath2he @xcite . in a finite system like our simulation sample , one therefore expects critical finite - size rounding of the thermal conductivity , which can be studied in the framework of the spin dynamics simulation and which should also be observable in experiments . to what extent the spin dynamics simulation actually captures the critical dynamics of @xmath2he is a rather delicate question . although the _ asymptotic _ behavior is described by model e , the actual _ crossover _ to the asymptotic behavior , i.e. , the decay of _ nonasymptotic _ corrections is governed by the specific heat exponent @xmath21 @xcite , which is so small that the true asymptotic behavior will never be seen in a simulation . from the point of view of analytic theory this means that in order to capture the nonansymptotic effects present in experiments one has to replace model e by the more complicated model f @xcite . from the point of view of spin dynamics simulations this means that one has to look for sources of such nonasymptotic behavior artificially generated by the simulation method and other nonasymptotic corrections not captured by the model or the method ( see sec.ii ) . apart from these problems , it should also be mentioned that the dynamical model e has two renormalization group fixed points . one of these fixed points yields dynamic scaling with a single dynamic exponent @xmath22 in @xmath23 dimensions , whereas the other gives rise to a weak _ violation _ of dynamic scaling . theoretical arguments @xciteand experimental evidence indicate that the latter fixed point is the stable one for @xmath2he in @xmath10 , i.e. , the critical dynamics is characterized by two different dynamic exponents @xmath24 ( order parameter ) and @xmath25 ( conserved quantity ) which fulfill the scaling relation @xmath26 . their difference @xmath27 has the nature of a dynamic wegner exponent and is known as the _ transient _ exponent @xcite . the remainder of the paper is organized as follows . in sec.ii we present the model and the simulation methods used to generate equilibrium configurations and to obtain the critical point of the model and its static critical exponents . furthermore , the equations of motion and the method used to integrate them numerically are presented . in sec.iii we briefly discuss the static critical behavior of our model and present an accurate estimate of the critical temperature . sec.iv is devoted to the discussion of the dynamic structure factor and the comparison with predictions of analytic theory . in sec.v we present first results for the lattice analogue of the thermal conductivity and discuss its scaling properties . a summary and prospects for future work are given in sec.vi . unless otherwise stated statistical errors quoted in this work correspond to one standard deviation . the system under investigation is given by a ferromagnetic heisenberg model with the strongest possible easy - plane anisotropy . the model hamiltonian reads @xmath28 where @xmath29 denotes a nearest neighbor pair of spins on a simple cubic lattice in three dimensions . the lattice contains @xmath30 lattice sites in each direction and in order to avoid surface effects periodic boundary conditions are applied . each spin @xmath31 is a classical spin @xmath32 with the normalization @xmath33 . the easy - plane anisotropy in eq.([hxy ] ) is the strongest possible in the sense that the @xmath7-components of the spins do not couple , so that eq.([hxy ] ) looks like the standard hamiltonian for the usual ( plane rotator ) xy model . as a starting point for the spin dynamics a sequence of equilibrium configurations is needed to provide initial conditions for the equations of motion . these configurations are obtained from a monte - carlo simulation of the model hamiltonian given by eq.([hxy ] ) . the monte - carlo algorithm chosen is a hybrid scheme , where each hybrid monte - carlo step ( mcs ) consists of 10 updates each of which can be one of : one metropolis sweep of the whole lattice , one single cluster wolff update @xcite , or one overrelaxation update of the whole lattice @xcite . the metropolis algorithm updates the lattice sequentially in the standard way . according to the detailed balance condition we choose the acceptance probability @xmath34 for a single spin flip , where @xmath35 is the change in configurational energy according to eq.([hxy ] ) and @xmath36 . the wolff algorithm also works the standard way @xcite , except that _ only _ the @xmath37 and @xmath38 components of the spins are used for the cluster growth . this means that a cluster update never changes the @xmath7-component of any spin so that the wolff algorithm is nonergodic in this case . our cluster update is still a valid monte - carlo step in the sense that it fulfills detailed balance , however , in order to provide a valid monte - carlo algorithm it has to be used together with the metropolis algorithm described above in a hybrid fashion . the use of wolff updates allows us to take advantage of improved estimators @xcite for magnetic quantities . the overrelaxation part of the algorithm performs a microcanonical update of the configuration in the following way . the local configurational energy has the functional form of a scalar product of the spins , where according to eq.([hxy ] ) only the @xmath37 and @xmath38 components are involved . with respect to the sum of its nearest neighbor spins each spin has a transverse component in the @xmath39 plane which does not enter the scalar product . the overrelaxation algorithm simply scans the lattice sequentially , determines this transverse component for each lattice site and flips its sign . this does not change the local configurational energy @xmath40 and by virtue of the usual metropolis acceptance function @xmath41 the update is always accepted . along with this simple operation the sign of @xmath42 is flipped with probability @xmath43 at each lattice site which according to eq.([hxy ] ) also does not change the energy of the configuration . this overrelaxation algorithm is similar to the one used in ref.@xcite and it quite efficiently decorrelates subsequent configurations over a wider range of temperatures around the critical point than does the wolff algorithm . typically , we use three metropolis _ ( m ) _ , five single cluster wolff _ ( c ) _ , and two overrelaxation updates _ ( o ) _ in a hybrid monte - carlo step in the critical region of our xy model . the inividual updates are mixed automatically in the program so that the update sequence _ ( m c c m o c m c c o ) _ is generated as one hybrid monte - carlo step in this case . the random number generator we use is the shift register generator r1279 given by the recursion relation @xmath44 for @xmath45 . generators like this are known to cause systematic errors in combination with the wolff algorithm @xcite ; however , for lags @xmath46 as large as the ones used here these errors will be far smaller than typical statistical errors . they are further reduced by the hybrid nature of our algorithm @xcite . the spin dynamics of the xy model is defined by the equations of motion @xmath47 where @xmath48 ist the hamiltonian defined by eq.([hxy ] ) . one may interpret eq.([xyeqmot ] ) as the direct classical analogue of the heisenberg equations of motion for spin operators , where @xmath49 so that energies and frequencies are measured in the same units . from the symmetry of eq.([hxy ] ) it is evident that the components @xmath50 and @xmath51 of the magnetization @xmath52 are not conserved under the dynamics given by eq.([xyeqmot ] ) . note that the two - component vector @xmath53 is the _ order parameter _ of the xy model . the @xmath7- or _ out - of - plane _ component @xmath54 of the magnetization @xmath55 is just the conserved quantity within the framework of model e dynamics we have already referred to in sec.i @xcite . note that eq.([hxy ] ) is invariant with respect to the transformation @xmath56 which is a symmetry required by model e. for the comparison of the critical spin dynamics with the critical dynamics according to model e it is important to realize , that the configurational _ energy _ is an _ additional _ constant of motion , because eq.([xyeqmot ] ) , in contrast to the coarse grained model e , does not contain relaxation . whether energy conservation is a reasonable assumption for the dynamics of the xy model or any other classical spin model is a question of the time scales to be resolved . the most important time scale for our investigation is set by the propagating modes ( spin waves ) in the system and for these the configurational energy is indeed constant . within the time scale of the spin waves thermal averages can therefore be replaced by averages over the initial configurations from which the time integration of eq.([xyeqmot ] ) is started . for much longer times relaxation processes ( equilibration with the heat bath ) come into play which violate energy conservation and render our spin dynamics approach invalid . in the vicinity of the critical point energy conservation becomes particularly important , because the dynamic universality class may change under the influence of an additional conservation law . if model e is augmented by energy conservation ( model e , see ref.@xcite ) it turns out that the energy asymptotically decouples from the order parameter @xmath53 and the conserved _ out - of - plane _ magnetization @xmath54 and model e critical behavior is restored . however , energy conservation may introduce corrections to the asymptotic finite - size scaling behavior @xcite which decay very slowly and may cause ambiguities in the scaling analysis of the spin dynamics data . note that these corrections are generated by the spin dynamics method . the equations of motion given by eq.([xyeqmot ] ) are integrated numerically for each initial spin configuration by a recently developed decomposition method @xcite . this method guarantees exact energy conservation and conservation of spin length @xmath57 and conserves @xmath54 within its numerical truncation errors . for the present study a second order integrator is used with the time step @xmath58 . this time step guarantees sufficient accuracy with respect to the conservation of @xmath54 and is much faster than well - known predictor corrector methods @xcite . for some accuracy and stability tests the time step has been increased to @xmath59 which still yields sufficient accuracy for the dynamic structure factor . fourth order integrators are much more accurate as far as @xmath54 conservation is concerned , but their internal complexity makes them much slower than a second order method for the same time step @xcite . moreover , statistical errors are not decrased significantly by fourth order methods and we therefore only report results obtained by the second order method . the equations of motion are integrated to a final time of @xmath60 and thermal averages are taken over 1000 initial configurations . all error bars of dynamic quantities correspond to one standard deviation . the simulations have been performed on various dec alpha axp , ibm rs6000 , and hp risc8000 workstations both at the rwth aachen and the bugh wuppertal . the basic ingredient for the spin dynamics simulation is provided by the sequence of initial spin configurations , which has to be generated according to the canonical ensemble in order to provide well defined thermal averages . therefore , the static behavior of the xy model and especially the location of @xmath20 have to be determined first . for this purpose we employ the hybrid monte - carlo scheme described above for lattice sizes @xmath30 between @xmath61 and @xmath62 . for each system size and temperature we perform 10 blocks of @xmath63 hybrid steps for equilibration followed by @xmath64 hybrid steps for measurements . each measurement block yields an estimate for all static quantities of interest and from these we obtain our final estimates and estimates of their statistical error following standard procedures . the integrated autocorrelation time of our hybrid algorithm is determined by the autocorrelation function of the energy or , equivalently , the _ modulus _ @xmath65 of the order parameter , which yield the slowest modes for the wolff algorithm . the autocorrelation times are generally rather short , at @xmath20 ( see below ) they range from about 5 hybrid mcs for @xmath61 to about 10 hybrid mcs for @xmath62 . for comparison , the autocorrelation function for the order parameter itself yields an autocorrelation time of less than one hybrid mcs . the values for the equilibration and measurement periods given above thus translate to roughly 100 and 1000 autocorrelation times , respectively , which is sufficient for all practical purposes . in order to obtain the best statistics for magnetic quantities a measurement is made after every hybrid mcs . the critical temperature @xmath20 is determined by monitoring the temperature and size dependence of the binder cumulant ratio . specifically , we measure the cumulants @xmath66 it turns out that @xmath67 is more sensitive to changes in temperature and system size so that we only use @xmath67 for the final fine - tuning of the temperature @xmath68 . for convenience the temperature is expressed as the dimensionless reduced coupling @xmath69 , where @xmath70 denotes the critical point . from standard procedures @xcite we obtain @xmath71 . for comparison we mention that @xmath72 for the standard plane rotator xy model on a simple cubic lattice in @xmath10 @xcite . by ignoring corrections to scaling and averaging over the measurements for @xmath73 and @xmath74 we find the estimates @xmath75 for the values @xmath76 and @xmath77 of the cumulants defined by eq.([u1u2 ] ) at the critical point . note that for @xmath78 the value of @xmath67 remains within two standard deviations of @xmath76 for all @xmath30 . our estimate for @xmath77 is within two standard deviations of the corresponding estimate @xmath79 found in ref.@xcite which already gives some evidence that the planar heisenberg variant of the xy model studied here is indeed a member of the static xy universality class . the critical exponents are estimated from the critical finite - size scaling behavior of the average modulus @xmath80 of the order parameter , the average square @xmath81 of the order parameter , and the temperature derivative of the latter . at @xmath82 , i.e. , @xmath83 one finds the leading scaling behavior @xmath84 with the system size @xmath30 , where @xmath85 , @xmath86 , and @xmath87 are the critical exponents of the order parameter , the susceptibility , and the correlation length , respectively . during the data analysis it turns out that corrections to scaling can be ignored within the statistical error of the quantities in eq.([fss ] ) . from our estimate @xmath88 and @xmath89 20 , 24 , 30 , 36 , 40 , 50 , 60 , and 80 we find the following values for the critical exponents : @xmath90 these values satisfy the scaling relation @xmath91 . furthermore , we directly obtain from eq.([expon ] ) @xmath92 , @xmath93 , @xmath94 , and @xmath95 in very good agreement with previous monte - carlo estimates @xcite and renormalization group theory for the @xmath96 ginzburg - landau model @xcite . the critical exponent @xmath97 of the specific heat can be obtained from , e.g. the hyperscaling relation , but the statistical error of @xmath87 is too large to exclude logarithmic behavior @xmath98 . apart from this deficiency our simulations confirm xy - like critical behavior for our version of the xy model ( see eq.([hxy ] ) ) quite accurately . in the following @xmath99 is measured in units of @xmath100 chosen such that @xmath101 . we continue this section with a short discussion of the static spin - spin correlation function ( structure factor ) @xmath102 at the critical temperature . the static structure factor is the spatial fourier transform of the spin - spin correlation function @xmath103 where @xmath104 refer to spin the components , @xmath105 and @xmath106 denote lattice vectors , and the thermal average is indicated by @xmath107 . note that @xmath108 due to translational invariance and reflection symmetry of the system . the spatial fourier transform is therefore given by a cosine transform which we use in the form @xmath109 with respect to the spontaneous magnetization of the xy model @xmath110 has a transverse component @xmath111 and a longitudinal component @xmath112 . in our monte - carlo simulation we measure @xmath113 and @xmath114 by _ rotating _ the coordinate system of the spins around the @xmath7-axis such that the random but finite magnetization vector @xmath53 is aligned with the @xmath38-direction . the @xmath37- and @xmath38-components of the spins in the rotated frame then correspond to the transverse and the longitudinal spin components , respectively , and their correlation functions yield @xmath115 and @xmath116 . it should also be mentioned that the out - of - plane component @xmath117 of the static structure factor is independent of @xmath118 and does not show any critical behavior . according to eq.([gq ] ) the normalization of @xmath110 is such that @xmath119 , where @xmath120 is the static susceptibility . note that @xmath121 by definition and that @xmath122 , where @xmath123 is the magnetic susceptibility with respect to the _ modulus _ of the magnetization . we furthermore limit the discussion to the @xmath124 direction , because other lattice directions do not provide new information on a simple cubic lattice . the components of @xmath118 are always measured in units of the inverse lattice contant . at @xmath20 the longitudinal component @xmath112 of the static structure factor can be described by the model function @xmath125 where @xmath126 is taken from eq.([expon ] ) , @xmath127 ^ 2 $ ] captures lattice effects @xcite , and the finite - size scaling function @xmath128 is chosen as a simple generalization of a lorentzian : @xmath129^{-(2-\eta)/2 } .\ ] ] the parameters @xmath130 and @xmath131 are determined from a fit to the data , where @xmath130 and @xmath131 are independent of @xmath30 . we have chosen @xmath132 and @xmath133 as obtained from a fit to the data for @xmath134 . for @xmath135 and @xmath62 @xmath130 and @xmath131 are found within less than 1% of these values , for @xmath61 they are about 3% smaller . despite its simplicity the model function given by eq.([glq ] ) captures the shape of the monte - carlo estimate for the longitudinal component of the structure factor remarkably well . the finite - size scaling analysis of our data for @xmath112 for small @xmath136 , is shown in fig.[glscal ] . deviations from finite - size scaling set in at @xmath137 for @xmath61 , the data for @xmath135 , 60 , and 80 collapse within the error bars up to @xmath138 . within the symbol sizes data collapse is obtained up to @xmath139 , where lattice effects set in . for comparison the model scaling function @xmath140 ( @xmath132 and @xmath133 , see eq.([glx ] ) ) is shown by the dashed line in fig.[glscal ] . the true scaling form of @xmath112 for small @xmath136 is captured rather well by @xmath128 . however , the choice of the model function is , of course , not unique . the dynamic structure factor @xmath0 is the space - time fourier transform of the position and time displaced spin - spin correlation function @xmath141 which we define by ( see eq.([gij ] ) ) @xmath142 the indices @xmath143 refer to spin components , @xmath105 and @xmath106 are lattice vectors , and @xmath144 and @xmath145 are moments in time to which the initial spin configuration has evolved according to the equations of motion ( see eq.([xyeqmot ] ) ) . the average @xmath146 is taken over the set of initial configurations as described in sec.ii . note that @xmath147 is also symmetric with respect to @xmath105 and @xmath106 ( see eq.([gij ] ) ) . as the second argument only the time displacement enters , because the equations of motion are invariant under the transformation @xmath148 , @xmath149 ( see eqs.([hxy ] ) and ( [ xyeqmot ] ) ) . the space - time fourier transform is therefore also given by the cosine transform ( see eq.([gq ] ) ) @xmath150 note that the normalization in eq.([sqw ] ) is such that @xmath151 , where @xmath110 is the static structure factor defined by eq.([gq ] ) . the out - of - plane component @xmath17 of the dynamic structure factor is associated with the conserved out - of - plane magnetization @xmath54 , i.e. , @xmath152 ( see also eq.([sqw ] ) ) is a constant in time . the in - plane magnetization ( order parameter ) @xmath53 is not conserved under the spin dynamics . although the time scale set by the motion of @xmath50 and @xmath51 is considerably larger than typical time scales set by spin waves , all initial differences between longitudinal and transverse components of the spin correlations completely disappear during the integration of the equations of motion . we therefore only discuss the average of the @xmath153 and the @xmath154 component of @xmath0 and refer to it as the `` in - plane component '' @xmath16 . we now turn to the discussion of @xmath0 above , below , and at the critical point . the correlation functions are measured up to a time displacement of @xmath155 away from criticality , where only smaller systems are considered . at the critical temperature we measure correlations up to @xmath156 for systems with up to @xmath134 lattice sites in each direction . in order to avoid effects of criticality we choose the temperature @xmath157 in the following . due to the absence of critical finite - size scaling at this temperature we limit the spin dynamics simulations to smaller systems with @xmath61 , 24 , and 30 . for better momentum resolution we only present results obtained for @xmath158 , smaller @xmath30 yield identical results with a lower resolution in @xmath118 . we also limit the presentation to @xmath0 in the @xmath124 direction , other lattice directions provide essentially the same information . in fig.[sxzfit ] we show @xmath159 and @xmath160 for @xmath161 as functions of the dimensionless frequency @xmath162 . both components apparently display a central peak without any additional features . for the values of @xmath68 and @xmath136 used in fig.[sxzfit ] one expects a central peak of lorentzian shape @xcite . as displayed by the solid line in fig.[sxzfit ] a simple lorentzian of the form @xmath163 ^ 2}\ ] ] characterized by an amplitude @xmath164 and a width @xmath165 captures the shape of @xmath166 very well up to @xmath167 , where the intensity has already dropped by an order of magnitude . for larger @xmath168 @xmath166 decays faster than a lorentzian . if one tries to fit the line shape of @xmath169 also with a lorentzian for small @xmath168 , it turns out that a better fit is obtained from a superposition of symmetrically placed lorentzians . we use @xmath170 ^ 2 } + { a_{zz}(q ) \over 1 + \left[(\omega + \omega(q ) ) / \gamma_{zz}(q ) \right]^2 } , \ ] ] where @xmath171 denotes the spin wave frequency which is used as an additional fit parameter . the dashed line in fig.[sxzfit ] displays the lorentzian fit for @xmath169 according to eq.([lzz ] ) , where @xmath172 is indeed finite within its statistical error of roughly @xmath173 . however , eq.([lzz ] ) only captures the shape of @xmath169 up to @xmath174 . if eq.([lxx ] ) ( @xmath175 ) is used instead , this frequency range becomes even smaller . this discrepancy in frequency range between the in - plane and the out - of - plane components of @xmath0 may be due to the difference in time scales between in - plane and out - of - plane modes as predicted by mode coupling theory @xcite . spin wave signatures are still visible in @xmath169 at @xmath157 as shown in fig.[szzatc ] , where @xmath169 is plotted for the first four momenta @xmath176 , @xmath177 . for @xmath178 a very broad maximum becomes visible which moves to the right as @xmath136 increases . the appearance of spin wave signatures in @xmath169 above @xmath20 is expected from renormalization group theory for model e dynamics @xcite . for the @xmath136 values used in fig.[szzatc ] the shape of @xmath169 is captured quite well by the lorentzian defined by eq.([lzz ] ) . due to the small enhancement of the line intensity over the signal level at @xmath179 the peak position @xmath171 and linewidth @xmath180 are not very well defined . we therefore refrain from a more detailed analysis at this point . the frequency dependence of the in - plane component @xmath166 is dominated by a central peak as in fig.[sxzfit ] for all @xmath136 . only for large @xmath136 near the brillouin zone boundary a shoulder appears in @xmath166 near the position of the spin - wave peak in @xmath169 . we illustrate this in fig.[sxzpi ] , where @xmath166 and @xmath169 are shown for @xmath181 . note that @xmath169 , which is much smaller than @xmath166 for small @xmath136 ( see fig.[sxzfit ] ) , becomes comparable to @xmath166 in magnitude for large @xmath136 . in a qualitative sense we have recovered the same behavior as observed by spin dynamics simulations of the xy model in @xmath9 above the kosterlitz - thouless transition @xcite . as before we want to avoid critical finite - size effects also below @xmath20 and we therefore choose @xmath182 in the following . again relatively small systems are sufficient for this investigation . our main results have been obtained for @xmath158 , smaller systems yield the same results at a lower @xmath136 resolution . the dynamics of the xy model below @xmath20 is dominated by spin waves which are visible in @xmath0 as pronounced peaks at the spin wave frequency @xmath171 as shown in fig.[sxzsw ] for @xmath161 in the @xmath124 direction . the peak intensity of @xmath166 is more than one order of magnitude larger than the corresponding peak intensity of @xmath169 . @xmath166 also displays a pronounced central peak , where the intensity exceeds the intensity of @xmath169 by over three orders of magnitude . in a qualitative sense this is again the same behavior as observed in @xmath9 @xcite , where the central peak in @xmath166 was not expected by analytical theory @xcite . we also find some fine structure at low intensities in @xmath166 for which no theoretical predictions exist . in fig.[sxzsw ] we have marked these additional resonances by arrows . as demontrated already in @xmath9 @xcite these signals can be interpreted as two - spin wave peaks in the following way . in order to produce a contribution to @xmath166 at @xmath161 in the @xmath124 direction one can combine _ two _ spin waves at @xmath161 , one in the @xmath183 direction , which is eqivalent to the @xmath124 direction on a simple cubic lattice , and one in the @xmath184 direction ( not shown ) , where the latter has a higher energy ( frequency ) . the _ difference _ between the corresponding frequencies is marked as @xmath185 in fig.[sxzsw ] , their _ sum _ is marked as @xmath186 . a second way to get a contribution to @xmath166 at @xmath161 in the @xmath124 direction is to combine two spin waves in the @xmath124 direction , one at @xmath161 the other at @xmath187 , where again the latter has a higher energy . sum _ of the corresponding frequencies is marked as @xmath188 in fig.[sxzsw ] their _ difference _ almost coincides with the position of the spin wave peak and can therefore not be resolved . in order to monitor the @xmath136 dependence of the spin wave frequency @xmath171 and the line widths @xmath189 and @xmath180 of the spin wave peaks , we again employ fits to simple lorentzians . for the in - plane component we generalize eq.([lxx ] ) to include the spin wave peaks : @xmath190 ^ 2 } + { a_{xx}(q ) \over 1 + \left[(\omega - \omega(q ) ) / \gamma_{xx}(q ) \right]^2 } + { a_{xx}(q ) \over 1 + \left[(\omega + \omega(q ) ) / \gamma_{xx}(q ) \right]^2}.\ ] ] for the out - of - plane component we use eq.([lzz ] ) which captures the shape of @xmath169 for sufficiently small frequencies in a satisfactory way . the dispersion relation @xmath171 can be obtained from eq.([lxxsw ] ) or eq.([lzz ] ) , where the latter yields slightly smaller error bars , because @xmath169 displays a sharper spin wave maximum . the estimates for @xmath171 , @xmath189 , and @xmath180 depend on the frequency range over which eqs.([lxxsw ] ) and ( [ lzz ] ) are fitted to @xmath166 and @xmath169 , respectively . we have chosen a frequency window around the spin wave peak , where for @xmath166 the central peak has been subtracted first . the error in the dispersion relation and the line widths is estimated by varying the size of the frequency window from about 1.2 times the half width to about twice the half width of the peak . the dispersion relation @xmath171 obtained from this procedure for the @xmath124 direction is shown in fig.[omq ] , the correponding zero temperature dispersion relation shown by the solid line . according to linear spin wave theory one obtains @xmath191 for @xmath192 where @xmath10 in our case . as expected , the spin wave frequencies are renormalized to fall below the @xmath192 dispersion curve . the functional form of @xmath171 can be captured by a fourier series . a convenient choice is @xmath193 where the fourier coefficients @xmath194 are rapidly decreasing . if eq.([omqfour ] ) is truncated after the second term one obtains @xmath195 and @xmath196 from a least square fit which connects all data points within their error bars as shown by the dashed line in fig.[omq ] . the line width @xmath189 of the spin wave peak in the in - plane component of the structure factor is shown in fig.[gxxq](a ) . as before the data can be analyzed by a fourier series with rapidly decreasing coefficients . a convenient choice here is @xmath197 where @xmath97 refers to @xmath37 ( in - plane ) and @xmath7 ( out - of - plane ) . a fit to @xmath198 with only two coefficients yields @xmath199 and @xmath200 which is shown by the dashed line in fig.[gxxq](a ) . the statistical error of @xmath201 and @xmath202 is about @xmath203 . the data for small @xmath136 are represented quite well , whereas near the brillouin zone boundary significant deviations occur . these can be reduced very quickly by including higher fourier modes in the fit ( not shown ) , where the higher fourier coefficients decrease rapidly in magnitude . here , we are primarily interested in the @xmath136 dependence of the line width near the center of the brillouin zone ( see below ) . a corresponding analysis has been performed for the line width @xmath180 of the spin wave peak in the out - of - plane component of the structure factor . the result is shown in fig.[gxxq](b ) , where the fourier coefficients in the fit ( dashed line ) are given by @xmath204 and @xmath205 . again the small @xmath136 behavior is captured very well by the fit , but near the brillouin zone boundary deviations occur which can also be reduced very quickly by including higher fourier modes . the limit @xmath206 is of particular interest for the line widths , because it reflects the influence of conservation laws on the dynamics . the line width @xmath207 can be interpreted as the relaxation rate of the ( non - conserved ) order parameter and therefore @xmath207 should be positive . the line width @xmath208 is the relaxation rate of the out - of - plane magnetization @xmath54 which is _ conserved _ , i.e. , @xmath208 should _ vanish_. from the fourier fits shown in fig.[gxxq ] one finds @xmath209 and @xmath210 with statistical errors of about @xmath211 in both cases . the extrapolation of @xmath189 and @xmath180 to @xmath212 is shown in fig.[gamq2 ] . the @xmath136 dependence of the line widths for small @xmath136 is quadratic as anticipated by the fourier fits shown in fig.[gxxq ] . from a fit to a straight line we find @xmath213 ( solid line ) , whereas @xmath214 ( dashed line ) within its statistical error in agreement with the conservation laws . finally , we note that the @xmath215 dependence of the linewidth @xmath216 is in agreement with the prediction of mode coupling theory @xcite . the exploration of critical dynamics and dynamic scaling naturally requires most of the numerical effort . in order to reach the scaling regime large systems are required and we have therefore performed simulations for @xmath89 20 , 24 , 30 , 40 , and 60 . our prime objective here is the test of dynamic finite - size scaling which we assume to be valid in the form @xcite @xmath217 at the critical point , where @xmath218 is the static structure factor discussed in sec.iii and @xmath219 denotes the dynamic critical exponent . note , that the @xmath118 dependence is reduced to a dependence only on @xmath220 , i.e. , isotropic scaling in space has been assumed . the index @xmath97 again refers to the spin component @xmath37 ( in - plane ) or @xmath7 ( out - of - plane ) , where @xmath221 and @xmath222 , respectively . in order to estimate @xmath219 we scale our data according to eq.([sqwscal ] ) and test the result for data collapse in the limit of small frequencies . the static structure factor needed for normalization in eq.([sqwscal ] ) is given by the equal - time correlations which we also extract from the spin dynamics data for consistency . in order to provide numerical estimates of the dynamic exponents @xmath25 and @xmath24 first we analyze the dispersion relations @xmath223 and @xmath224 at the critical point as obtained from a lorentz fit to the spin wave peak of @xmath225 ( see eq.([lzz ] ) ) and a lorentz fit to @xmath226 according to eq.([lxx ] ) . the line shape of @xmath166 is dominated by a strong central peak ( see below ) so that we restrict the analysis of the full dispersion relation to @xmath223 . the result in the @xmath124 direction is shown in fig.[omtc ] for @xmath227 and @xmath135 . the data apparently collapse onto a single curve and show a linear behavior for @xmath228 down to @xmath229 , where @xmath171 becomes nonlinear . from finite - size scaling one expects the scaling form @xmath230 at @xmath20 for sufficiently small q. for @xmath231 and @xmath89 20 , 24 , 30 , 40 and 60 we obtain an approximation of the small @xmath136 behavior of @xmath232 which is shown in fig.[omzztc](a ) . according to eq.([omscal ] ) @xmath232 should vary as @xmath233 for fixed @xmath234 . a least square fit to the data shown in fig.[omzztc](a ) yields the dynamic exponent @xmath235 which differs substantially from the mode - coupling prediction @xmath236 @xcite . from the scaling relation @xmath237 @xcite we obtain @xmath238 which can be tested against the small @xmath136 behavior of @xmath224 . the result is shown in fig.[omzztc](b ) . for the larger systems @xmath158 , 40 , and 60 the data agree with the power law , but for smaller lattice sizes @xmath61 and @xmath227 systematic deviations occur . if we exclude these smaller systems from the power - law fit shown in fig.[omzztc](a ) we find @xmath239 which is at the upper error boundary of the previous estimate given by eq.([zm ] ) . an alternative estimate of the dynamic exponents is provided by the _ median frequency _ @xmath240 which is defined by the relation @xmath241 note that we have normalized the dynamic structure factor such that its integral over all frequencies yields the static structure factor . we evaluate eq.([median ] ) with the trapezoidal rule . systematic errors induced by this simple numerical integration method are smaller than statistical uncertainties coming from the statistical errors of @xmath226 and @xmath225 . if the integration is performed with simpsons rule the same results are obtained within statistical errors . according to eq.([sqwscal ] ) we expect the scaling behavior @xmath242 of the median frequency at the critical point . due to the presence of a strong central peak in @xmath226 the median frequency @xmath243 for @xmath134 turns out to be of the same size as the frequency resolution of our data . we therefore limit the analysis of @xmath244 to the system sizes @xmath89 20 , 24 , 30 , and 40 . the other median frequency @xmath245 can be determined accurately for all system sizes . the result for @xmath246 is displayed in fig.[ommed ] . from a least square fit of @xmath243 to a power law for @xmath61 , 24 , 30 , and 40 we obtain @xmath247 . if only the data for @xmath158 and 40 are used we obtain @xmath248 . both estimates are consistent with eq.([zm ] ) and the scaling relation @xmath249 . the latter estimate is displayed by the solid line in fig.[ommed ] . the corresponding fit of @xmath250 for @xmath251 yields @xmath252 which is also within the error bar of our previous estimate ( see eq.([zm ] ) , dashed line in fig.[ommed ] ) . the spread in the values for the dynamic exponents obtained from the dispersion relations for small @xmath136 and the median frequencies is considerable . in order to reconcile all estimates with eq.([zm ] ) almost the full width of the error interval ( one standard deviation ) is needed . however , eq.([zm ] ) represents a reasonable mean value of all estimates discussed so far and we therefore adopt it as our final estimate for @xmath25 and use @xmath253 to determine @xmath254 . as a final test of eq.([zm ] ) we use eq.([sqwscal ] ) in order to obtain a scaling plot of the dynamic structure factor . for @xmath231 , @xmath158 , 40 and 60 and @xmath255 the resulting scaling plot for @xmath225 is shown in fig.[szzscal ] . the corresponding plot for @xmath226 with @xmath256 is displayed in fig.[sxxscal ] . data collapse can only be expected for sufficiently small @xmath168 . scaling works quite well for all frequencies up to the spin wave peak , where the intensity decreases slightly with increasing @xmath30 , but the error bars are still overlapping . the quality of the data collapse in not so evident from @xmath226 ( see fig.[sxxscal ] ) . at @xmath179 the scaled intensity for @xmath134 deviates from the ones for @xmath158 and 40 by about one standard deviation . this could be an effect of the finite statistical sample . the line shape of the central peak in @xmath166 still fits rather well to a lorentzian . only in the vicinity of @xmath179 are the data also compatible with the gaussian lineshape expected from mode coupling theory @xcite . the above scaling analysis provides quite strong evidence for a violation of dynamic scaling in the sense that two different dynamic exponents are required in order to obtain scaling in the dispersion relations , the median frequencies , and the two components of the dynamic structure factor . however , we can not prove from our data whether the estimate @xmath257 , which indicates the violation of dynamic scaling , corresponds to the transient exponent @xmath258 @xcite or if the measured difference consitutes an _ effective _ exponent for system sizes which are too small to ignore additional corrections . as pointed out in sec.i the presence of energy conservation in our spin dynamics simulation gives rise to such corrections governed by the exponent @xmath259 @xcite . on the basis of our data we can not exclude the possibility , that these corrections yield the dominant contribution to the measured effective exponent @xmath258 . for a clarification of this point more information is needed from analytical theories of dynamic finite - size scaling @xcite and from simulations @xcite . we have already mentioned in the introduction that the transport properties of the xy model near the critical point provide lattice analogues of the corresponding transport coefficients of @xmath2he near the @xmath19-transition . the conserved out - of - plane component @xmath54 of the magnetization is the lattice analogue of the entropy density in @xmath2he and its associated transport coefficient therefore corresponds to the thermal conductivity of @xmath2he which is of experimental interest . in principle the thermal conductivity can be extracted from an extrapolation of the characteristic frequency @xmath260 of @xmath169 given by @xmath261 @xcite to @xmath212 . however , in order to obtain a reliable extrapolation at the critical point a very high momentum resolution for @xmath206 is required and this can only be realized with very large systems ( see fig.[omtc ] and ref.@xcite ) . we therefore resort to an alternative approach already considered in ref.@xcite , we express the thermal conductivity by a current - current correlation function for a suitably chosen current density @xmath262 at lattice site @xmath263 @xcite . in order to identify the current density @xmath264 we reexamine the @xmath7-component of the equation of motion ( see eq.([xyeqmot ] ) ) which reads @xmath265 where the sum is over all nearest neighbors of lattice site @xmath263 . we define the @xmath266th component @xmath267 @xmath268 of the current density @xmath264 associated with the lattice point @xmath263 by @xcite @xmath269 where the notation @xmath270 denotes the nearest neighbor of the lattice site @xmath263 in the @xmath266th lattice direction . for the case of the simple cubic lattice studied here @xmath271 , @xmath272 , and @xmath273 can be visualized as the unit vectors of a cartesian coordinate system aligned with the lattice . the lattice divergence of the current density according to eq.([jik ] ) at lattice site @xmath263 is then given by @xmath274 which is just the negative r.h.s . of eq.([xyeqmotz ] ) . note that the lattice spacing has been set to unity . in the spirit of hydrodynamics eq.([xyeqmotz ] ) can be interpreted as second order discretization of an equation of continuity of the form @xmath275 . the density @xmath276 and the current @xmath264 defined by eq.([jik ] ) are located on staggered meshes , i.e. , the current density component @xmath267 associated with lattice site @xmath263 is located half way between @xmath263 and @xmath270 , so that each of the finite differences appearing in eq.([divj ] ) is located at lattice site @xmath263 just as the l.h.s . of eq.([xyeqmotz ] ) . this displacement of spins and currents on the lattice has no further consequences in our case , because we have applied periodic boundary conditions in all lattice directions . according to eq.(44 ) of ref.@xcite we find the expression @xmath277 for the transport coefficient @xmath1 , where @xmath278 and the same normalization as in eq.([sqw ] ) has been used . note that the out - of - plane static susceptibility @xmath279 needed for normalization in eq.([dqw ] ) is essentially constant as a function of system size @xmath30 at fixed temperature . in the following we will analyze @xmath1 only for @xmath82 , where @xmath280 corresponds to the thermal conductivity measured in @xmath2he experiments . according to eq.([xyeqmotz ] ) the current density @xmath264 has the scaling dimension @xmath281 , because @xmath282 has the scaling dimension zero . from eq.([dqw ] ) we then find @xmath283 as the scaling dimension of @xmath1 so that naive finite - size scaling yields @xmath284 at the critical point @xcite . we therefore expect the scaling form @xmath285 at @xmath82 , where @xmath286 is just a normalization factor . a corresponding scaling plot of @xmath287 versus @xmath288 for @xmath158 , 40 , and 60 and @xmath212 is shown in fig.[d0wscal ] , where we have used our estimate @xmath255 . the statistical noise in the data for @xmath1 is considerably larger than the noise in the data for the structure factor ( see also ref.@xcite ) . the spread of the data points in fig.[d0wscal ] is of the same magnitude as the scatter of the data in each individual data set displayed . in view of these statistical uncertainties our data scale reasonably well for small @xmath168 and confirm the scaling exponent @xmath289 ( see eq.([zm ] ) ) of the transport coefficient @xmath1 . for @xmath290 @xmath1 scales accordingly as shown in fig.[dqwscal ] for @xmath231 . the position of the spin wave maximum appears to be shifted to the right as compared to the spin wave peak in @xmath225 . according to eq.([dqwxzz ] ) scaling is obtained up to a value of about 0.6 of the scaling argument for both @xmath212 and @xmath231 . according to fig.[dqwscal ] the spin wave maximum itself appears to be outside the scaling regime for the system sizes used here . for @xmath290 @xmath1 tends to zero as @xmath291 within the statistical errors . the thermal conductivity @xmath292 is shown in fig.[d00l ] as a function of the system size . the overall behavior of @xmath19 with @xmath30 is captured quite well by the expected power law ( full line ) for @xmath251 and even the data point for @xmath227 is only one standard deviation away . systems with @xmath61 or less may be too small to be in the scaling regime . in view of the considerable statistical error in the data slowly varying corrections to scaling as discussed in the previous section can not be identified . in any case more theoretical information in needed in order to provide a reliable background for the interpretation of spin dynamics data of transport coefficients in the critical regime @xcite . the easy - plane heisenberg ferromagnet belongs to the xy universality class which has been demonstrated by the evaluation of the critical exponents and the static structure factor . unlike the standard plane rotator xy model the planar ferromagnet is endowed with a reversible spin dynamics which can be efficiently simulated by recently developed decomposition methods . due to spatial and temporal symmetries the spin dynamics of the planar ferromagnet is expected to be in the same dynamic universality class as @xmath2he near the superfluid normal transition , but may have different corrections to scaling . the data have been compared to field - theoretic and mode - coupling predictions with the following main results . 1 . above the critical regime the spin dynamics data show a strong lorentzian central peak in the in - plane component @xmath226 in agreement with mode coupling theory . the out - of - plane component @xmath225 displays a pseudo spin - wave peak which is also of lorentzian shape and becomes increasingly prominent as @xmath136 is increased . the presence of this peak is in accordance with field - theoretic predictions for @xmath293 . 2 . below the critical regime strong spin wave peaks occur in both components of the dynamic structure factor . the shape of these peaks is captured very well by lorentians as expected from mode - coupling theory . in addition to the spin wave peak @xmath226 displays a central peak of lorentzian shape and additional multi spin - wave peaks which do not appear in @xmath225 . for small @xmath136 the dispersion relation @xmath171 is linear in @xmath136 and the line widths @xmath189 and @xmath180 are quadratic in @xmath136 as expected from mode - coupling theory . the qualitative agreement with the spin dynamics data for @xmath9 suggests that the dynamics of the two - dimensional xy model may not be captured by vortex dynamics theories . 3 . at @xmath20 , in contrast to mode - coupling theory and in agreement with field - theoretic predictions , @xmath294 and @xmath225 require different dynamic exponents in order to obtain scaling : @xmath295 and @xmath296 , whereas mode coupling theory yields @xmath297 . the out - of - plane component @xmath226 is dominated by a strong central peak . a shoulder at a finite frequency indicates the presence of a spin wave signal . in @xmath225 a strong spin wave peak remains the dominating feature . the line shapes are still compatible with lorentzians , the central peak in @xmath166 is only compatible with a gaussian shape very close to @xmath179 . the transport coefficient @xmath1 , which provides access to the lattice analogue of the thermal conductivity of @xmath2he within the xy model , has been investigated at @xmath20 . the statistical fluctuations of the data are much stronger than those in the data for the structure factor which makes the scaling analysis less unique . however , scaling in agreement with the previously obtained dynamic exponents is obtained . the transport coefficient also shows a strong spin wave resonance for finite @xmath136 and vanishes for @xmath291 in this case . the thermal conductivity @xmath292 scales with the system size in the expected way within the error bars for sufficiently large systems . m. krech gratefully acknowledges many helpful discussions with v. dohm and financial support of this work through the heisenberg program of the deutsche forschungsgemeinschaft . this research was supported in part by nsf grant # dmr - 9727714 and by the computer center of the rwth aachen . 99 d. j. amit , _ field theory , the renormalization group , and critical phenomena _ ( mcgraw - hill , new york , 1978 ) . g. parisi , _ statistical field theory _ ( addison - wesley , wokingham , 1988 ) . p. c. hohenberg and b. i. halperin , rev . phys . * 49 * , 435 ( 1977 ) . d. p. landau and m. krech , j. phys . cond . mat . * 11 * , r1 ( 1999 ) . k. kawasaki in _ phase transitions and critical phenomena _ vol . 5a , edited by c. domb and m. s. green ( academic , new york , 1976 ) . e. frey and f. schwabl , adv . phys . * 43 * , 577 ( 1994 ) . a. keren , phys . lett . * 72 * , 3254 ( 1994 ) . k. chen and d.p . landau , phys . b * 49 * , 3266 ( 1994 ) . a. bunker , k. chen , and d.p . landau , phys . b * 54 * , 9259 ( 1996 ) . d. c. rapaport and d. p. landau , phys . e * 53 * , 4696 ( 1996 ) . k. hirakawa and h. yoshizawa , j. phys . . jpn . * 47 * , 368 ( 1979 ) . m. t. hutchins , p. day , e. janke , and r. pynn , j. mag . mag . mat . * 54 * , 673 ( 1986 ) . d. g. wiesler , h. zabel , and s. m. shapiro , z. phys . b * 93 * , 277 ( 1994 ) . j. m. kosterlitz and d. j. thouless , j. phys . c * 5 * , l124 ( 1972 ) ; j. phys . c * 6 * , 1181 ( 1973 ) . d. l. huber , phys . b * 26 * , 3758 ( 1982 ) . f. g. mertens , a. r. bishop , g. m. wysin , and c. kawabata , phys . lett . * 59 * , 117 ( 1987 ) ; f. g. mertens , a. r. bishop , m. e. gouvea , and g. m. wysin , j. de physique c * 8 * , 1385 ( 1988 ) ; f. g. mertens , a. r. bishop , g. m. wysin , and c. kawabata , phys . b * 39 * , 591 ( 1989 ) . m. e. gouvea , g. m. wysin , a. r. bishop , and f. g. mertens , phys . b * 39 * , 11840 ( 1989 ) . a. r. vlkel , f. g. mertens , a. r. bishop , and g. m. wysin , phys . b * 43 * , 5992 ( 1991 ) . d. p. landau and r. w. gerling , j. mag . magn . mater . 104 - 107 , 246 ( 1992 ) . h. g. evertz and d. p. landau , phys . rev . b * 54 * , 12302 ( 1996 ) . j. e. r. costa and b. v. costa , phys . b * 54 * , 994 ( 1996 ) ; b. v. costa , j. e. r. costa , and d.p . landau , j. appl . phys . * 81 * , 5746 ( 1997 ) . s. thoma , e. frey , and f. schwabl , phys . b * 43 * , 5831 ( 1991 ) . b. i. halperin , p. c. hohenberg , and e. d. siggia , phys . b * 13 * , 1299 ( 1976 ) . c. de dominicis and l. peliti , phys . b * 18 * , 353 ( 1978 ) . v. dohm , z. phys . b * 33 * , 79 ( 1978 ) ; v. dohm and r. a. ferrel , phys 67a * , 387 ( 1978 ) . g. ahlers , phys . * 21 * , 1159 ( 1968 ) ; w. y. tam and g. ahlers , phys . b * 32 * , 5932 ( 1985 ) . j. a. lipa , d. r. swanson , j. a. nissen , t. c. p. chui , and u. e. israelsson , phys , rev . * 76 * , 944 ( 1996 ) . v. dohm , phys . rev . b * 44 * , 2697 ( 1991 ) . u. wolff , phys . * 62 * , 361 ( 1989 ) . m. hasenbusch , nucl . phys . * b333 * , 581 ( 1990 ) . a. m. ferrenberg , d. p. landau , and y. j. wong , phys . rev . lett . * 69 * , 3382 ( 1992 ) ; l. n. shchur and h. w. j. blte , phys e * 55 * , r4905 ( 1997 ) . a. m. ferrenberg and d. p. landau , unpublished . m. krech , a. bunker , and d. p. landau , comp . * 111 * , 1 ( 1998 ) ; j. frank , w. huang , and b. leimkuhler , j. comp . phys . * 133 * , 160 ( 1997 ) . k. chen , a. m. ferrenberg , and d. p. landau , phys . rev . b * 48 * , 3249 ( 1993 ) . a. p. gottlob and m. hasenbusch , physica a * 201 * , 593 ( 1993 ) . j. c. le guillou and j. zinn - justin , j. phys . ( france ) lett . * 46 * , l137 ( 1985 ) ; r. guida and j. zinn - justin , j. phys . a * 31 * , 8103 ( 1998 ) . d. s. ritchie and m. e. fisher , phys . b * 5 * , 2668 ( 1972 ) . m. tpler , diplomarbeit rwth aachen , 1998 . m. krech , to appear in _ computer simulation studies in condensed matter physics xii _ , edited by d. p. landau , s. p. lewis , and h. b. schttler ( springer verlag , heidelberg , berlin , 1999 ) . n. a. lurie , d. l. huber , and m. blume , phys . b * 9 * , 2171 ( 1974 ) . l. p. kadanoff and p. c. martin , ann . phys . * 24 * , 419 ( 1963 ) .

this paper presents the first results of an h imaging survey of galaxies in the central regions of the a2151 cluster . a total of 50 sources were detected in h , from which 41 were classified as secure members of the cluster and 2 as likely members based on spectroscopic and photometric redshift considerations . the remaining 7 galaxies were classified as background contaminants and thus excluded from our study on the h properties of the cluster . the morphologies of the 43 h selected galaxies range from grand design spirals and interacting galaxies to blue compacts and tidal dwarfs or isolated extragalactic hii regions , spanning a range of magnitudes of mag . from these 43 galaxies , 7 have been classified as agn candidates . these agn candidates follow the(h ) vs. relationship of the normal galaxies implying that the emission associated with the nuclear engine has a rather secondary impact on the total h emission of these galaxies . a comparison with the clusters coma and a1367 and a sample of field galaxies has shown the presence of cluster galaxies with(h ) lower than expected for their , a consecuence of the cluster environment . this fact results in differences in the(h ) vs.(h ) and(h ) distributions of the clusters with respect to the field , and in cluster to cluster variations of these quantities , which we propose are driven by a global cluster property as the total mass . in addition , the cluster h emitting galaxies tend to avoid the central regions of the clusters , again with different intensity depending on the cluster total mass . for the particular case of a2151 , we find that most h emitting galaxies are located close to the regions with the higher galaxy density , offset from the main x - ray peak . overall , we conclude that both , the global cluster environment as well as the cluster merging history play a non negligible role in the integral star formation properties of clusters of galaxies .

0907.1370

the hercules supercluster is probably among the most massive structures in the local universe ( chincarini et al . 1981 ; barmby & huchra 1998 ) . this supercluster is composed of three smaller clusters : a2151 , a2147 and a2152 , that are probably bound ( barmby & huchra 1998 ) . optical and x - ray studies suggest that a2151 could still be in the process of collapsing , given the lack of hi deficiency in the spiral population ( giovanelli & haynes 1985 ; dickey 1997 ) and the irregular distribution of the hot icm and its low x - ray flux ( magri et al . 1988 ; huang & sarazin 1996 ) . on top of that a2151 has at least three distinct subclusters with varying dwarf to giant galaxy ratios ( bird et al . 1995 ; snchez - janssen et al . 2005 ) . the main aim of this paper is to study the population of galaxies with recent star formation in a2151 , as traced by their h@xmath0 luminosities . it has been largely discussed in the literature the effect of the environment on the h@xmath0 emission of galaxies , the commonly accepted conclusion being that the total h@xmath0 luminosity of galaxies is lowered in high density environments ( e.g. gavazzi et al . 2002 ; domnguez et al . 2002 ; lewis et al . 2002 ; gmez et al . 2003 ; balogh et al . 2004 ; tanaka et al . 2004 ; rines et al . 2005 ) . however , most of these studies suffer from two main drawbacks : ( a ) some of them use information on the h@xmath0 emission of galaxies from fiber spectroscopy , as it is the case of those works based on sdss and 2df data . it is known that there is a bias on the total h@xmath0 flux measured in fiber surveys for low redshift galaxies , imposed by the size of the fiber on the sky . ( b ) in most cases the samples of galaxies were selected on the basis of their optical fluxes and miss star forming dwarf galaxies . the inclusion of these dwarf galaxies in the large surveys devoted to the study of the star formation in cluster galaxies imposes new challenges to the models of evolution of galaxies in dense environments ( reverte et al . 2007 ) . for these reasons , we stress the relevance of deep h@xmath0 imaging surveys that allow to capture the totality of the h@xmath0 emission of galaxies and to include all the h@xmath0 emitting sources down to a given h@xmath0 flux , disregarding their continuum emission . our previous work in this direction ( iglesias - pramo et al . 2002 ) highlights the necessity of expanding the available databases of h@xmath0 selected samples of clusters of galaxies in order to allow a proper comparison with field samples and to improve our understanding of the role played by the environment on the activity of star formation of galaxies . in this paper we present the first results of an h@xmath0 survey of the a2151 cluster and a comparison of the h@xmath0 properties of similar galaxy samples in other clusters and in the local field . a total of 50 galaxies with h@xmath0 emission were detected . in section 2 we describe the observations . in section 3 we explain the data reduction procedure . section 4 describes the selection of the sample . section 5 shows the main results derived from the observations related to the h@xmath0 properties of the sample galaxies and compared with other samples from the literature . finally , in section 6 we discuss the implications of these results and present the main conclusions . throughout the paper we assume a cosmology with @xmath5 km s@xmath6 mpc@xmath6 , @xmath7 and @xmath8 . the data were obtained during two observational runs . the first one was from 6 to 7 of may 2003 at the 256 cm nordic optical telescope ( not ) located at the roque de los muchachos observatory . the instrumentation used was the alfosc spectrograph in direct imaging mode with a redshifted h@xmath0 filter centered at @xmath9 and with fwhm of @xmath10 for the emission line ( on - band frames ) , and a gunn-@xmath11 centered at @xmath12 and with fwhm of @xmath13 broadband filter for the continuum ( off - band frames ) . the field of view was a circle of approximately 6 arcmin in diameter with a plate scale of 0.188 arcsec pix@xmath6 . in this run , five pointings were obtained , covering an area of about 141 arcmin@xmath14 . the second run was during 4 and 5 of june 2003 at the 420 cm william herschel telescope ( wht ) , located also at the roque de los muchachos observatory . the instrumentation used was the prime focus ( pf ) camera with two redshifted h@xmath0 filters centered at @xmath15 and @xmath16 , and with fwhm of @xmath17 and @xmath18 respectively for the emission line ( on - band frames ) , and a harris-@xmath19 centered at @xmath20 with a fwhm of @xmath21 broadband filter for the continuum ( off - band frames ) . the nominal field of view was [email protected] arcmin@xmath23 with a plate scale of 0.24 arcsec pix@xmath6 . unfortunately , for the narrowband filters , the convergence of the pf beam reduced the useful field of view to approximately 33 arcmin@xmath14 . during this run 11 pointings were obtained , covering a total area of about 363 arcmin@xmath14 . in table [ fil ] we present a log of the pointings , the filters employed at each telescope and the corresponding exposure times of the frames . the average seeing measured for all the frames varied between 11.3 . the total coverage of our sample is 0.15 deg@xmath23 ( equivalent to about 7% of the total abell area assuming an abell radius for a2151 of 0.82 deg from solanes , 2001 ) including the areas of higher density of galaxies . the data were reduced employing the iraf package . all images were bias subtracted and flat - field corrected employing twilight sky flats . for the majority of the galaxies detected , three or more exposures were obtained . then , these images were combined employing the iraf task imcombine with the median operation . this was done in order to increase the signal to noise ratio and to wipe out the cosmic rays . for few galaxies with less than three exposures , the cosmic rays were removed employing the iraf task credit . net h@xmath0 frames were produced by subtracting the off - band from the on - band frames . for this , all the off - band frames were scaled to the on - band frames under the assumption that the resulting average net flux of field stars is null . given that some stellar spectra may show balmer absorption lines or other features that could affect the scaling procedure , a minimum of ten field stars per frame were used . on - band exposures of several spectrophotometric standard stars from oke ( 1990 ) were obtained each night in both observational runs in order to flux calibrate our data . the flux calibration was carried out following barth et al . the fluxes were corrected for galactic extinction using the reddening maps of schlegel et al . ( 1998 ) and the extinction law of cardelli et al . ( 1989 ) . in order to verify the quality of our photometric calibration , we make use of the sdss database and of the fact that the effective wavelengths of the broad band filters used in this work are close enough to each other and to the one corresponding to the sdss-@xmath24 filter , 6390 for the harris-@xmath19 and 6250 for the sdss-@xmath24 filters respectively . ] . thus , under the assumption of a nearly flat shape of the red continuum of the galaxies the magnitudes obtained with the three filters in a given photometric system are directly comparable . as a first steep we derive ab magnitudes , where @xmath25 is the effective frequency of each of the broad band filters and @xmath26 is the flux density at this frequency in erg s@xmath6 @xmath27 hz@xmath6 . ] for a sample of non - saturated stars in our off - band frames and compare them to their sdss-@xmath24 magnitudes . we show in figure [ cal ] the comparison between our ab magnitudes and the sdss-@xmath24 ones for a sample of stars present in our frames . the median of the difference between the two sets of data is about 0.03 mag , with a standard deviation of 0.1 mag . this result illustrates that the zero point of our calibration fairly agrees with that of sdss . the fluxes for each emitting galaxy were extracted employing the focas package , that runs under iraf . we applied this code to the net h@xmath0 frames and detected objects with flux larger than 3@xmath28 over the sky background and within an area larger than 30 pix ( approximately equivalent to the seeing disk ) . given the typical values of the @xmath28 of the sky background in our frames , the limiting surface brightness for detection in this work is approximately @xmath29 erg @xmath27 s@xmath6 arcsec@xmath30 . this work is part of a more ambitious project devoted to the complete coverage in h@xmath0 of the central parts of a sample of clusters representative of the local universe . with this aim , we begun an h@xmath0 imaging follow - up of galaxies in a2151 for which the radial velocities , extracted from ned , were consistent with the known velocity distribution of this cluster km s@xmath6 , which corresponds to 3@xmath28 around the systemic velocity of the cluster . ] . during the realization of this work the database of radial velocities of galaxies in this cluster was enlarged , mainly due to the contribution of the sdss . thus , the h@xmath0 flux could be measured for a larger number of galaxies present in our frames . this fact turned our pointed observations into an h@xmath0 volume limited survey . the resulting total area covered by our observations corresponds to approximately 0.2 square degrees and is mostly concentrated in an area within the inner 15 arcmin ( in radius ) of the cluster , as it is shown in figure [ espacial ] . the h@xmath0 flux can only be measured here for galaxies whose h@xmath0 line is redshifted within the transmittance window of the narrow band filter . for this reason in each frame we initially restrict our sample to galaxies with available spectroscopic redshift and for which the h@xmath0 line is shifted into the wavelength range where the transmittance of the filter is higher than 50% of the peak transmitance , thus those galaxies detected in h@xmath0 as described in the previous section are included in our sample . this restriction results in a degree of incompleteness of the sample of galaxies for which an h@xmath0 flux was measured . in order to estimate this incompleteness effect for the central regions of a2151 , we present in figure [ histo_velo ] the velocity distribution for all the galaxies with measured radial velocity from spectroscopic redshift in sdss within different regions around the center of a2151 . the sample is split into four subsamples according to the distance to the center of the cluster . as it can be seen in figure [ histo_velo ] , the distribution of galaxies in a2151 is broad and peaked around 10000 km s@xmath6 for galaxies within the inner 5 arcmin of the center . in the annulus between 5 and 10 arcmin , the distribution gets narrower and still centered at 10000 km s@xmath6 . when we move to galaxies in the annulus between 10 and 20 arcmin , the velocity distribution becomes again broader and peaked at 12000 km s@xmath6 . finally , in the annulus between 20 and 30 arcmin from the center of the cluster the velocity distribution of galaxies gets again narrower and centered at 11000 km s@xmath6 . this implies that for the fields observed at the not the h@xmath0 flux of some cluster galaxies is lost because of our transmittance filter restriction . these galaxies ( approximately 22% of the total sdss sample ) are the ones populating the low velocity tail of the distribution corresponding to the inner 10 arcmin ( in radius ) . on the other hand , this problem is not present for the fields observed with the wht because most of the cluster galaxies fulfill the transmittance filter criterion imposed above . since most of the outer fields were observed with the wht ( see figure [ espacial ] ) , we conservatively estimate the incompleteness of our h@xmath0 sample to be less than 20% . in addition to these galaxies for which the radial velocity is known from spectroscopic observations , our a2151 sample also includes those sources detected in emission as explained in section [ data_red ] but whose radial velocity is not known . the fact that these objects show up in our net h@xmath0 frames as positive detections suggest that they are very likely galaxies belonging to a2151 . however , we can not exclude the possibility that they are background objects for which we are detecting redshifted [ oiii ] , h@xmath31 , [ oii ] or ly@xmath0 emission . one way to disentangle whether these are cluster or background ( contaminant ) objects is using the sdss photometric redshift ( photo - z ) , which is available for all of them . at the mean redshift of a2151 ( @xmath32 ) , the emission line contaminants must be at @xmath33 ( at this value of @xmath34 , the [ oiii]5007 line shifts to 6800 , which corresponds to the wavelength of h@xmath0 at the redshift of a2151 ) . it is well known that the accuracy of the photo - z is much lower than that of the spec - z . in particular , the typical @xmath28 declared for the sdss photo - z is of the order of 0.05 , which is larger than the fwhm of our narrow band filters . however , given that the expected contamination comes from galaxies at redshifts @xmath33 , we can use the sdss photo - z information to identify the contaminants included in our sample . for this , we make use of the three sdss photo - z estimations available for most galaxies of our sample . given the already mentioned accuracy of the sdss photo - z and the expected redshift of the contaminants , we assume that all galaxies of our sample with photo - z@xmath35 must be background contaminants . as for all except two of our galaxies ( one with and the other without available spec - z ) we have three different photo - z estimations , we will split our sample into four categories : ( a ) those galaxies with spectroscopic @xmath34 within 3@xmath28 of the systemic velocity of the cluster , are flagged `` 0 '' and , by default , considered secure members of the cluster ; ( b ) those with no spectroscopic @xmath34 and at least two photo - z estimations consistent with photo - z @xmath36 , are flagged `` 1 '' and considered secure members of the cluster ; ( c ) those with no spectroscopic @xmath34 and just one photo - z estimation consistent with photo - z @xmath36 , are flagged `` 2 '' and considered likely members of the cluster ; and ( d ) those with no spectroscopic @xmath34 and three photo - z estimations consistent with photo - z @xmath35 are flagged `` 3 '' and considered contaminants . the only galaxy with no spectroscopic @xmath34 and with just one available photo - z estimation , was classified as background contaminant . in order to assess the robustness of our photo - z based criterion to identify the contaminants we tested it with the subsample of galaxies with available spec - z . the result was that all these galaxies were classified as secure members of the cluster ( as it was found from their spectroscopic radial velocities ) . the sdss spectra allow us to look for the presence of agns in our sample . we show in figure [ agn ] the classical diagnostic diagram [ nii]6584/h@xmath0 vs. [ oiii]5007/h@xmath31 ( initially proposed by baldwin , phillips & terlevich 1981 ) for our galaxies with the line separating starbursts and agns proposed by kauffmann et al . only those galaxies of the sample for which these four emission lines were detected in sdss are shown in the plot . a total of five galaxies lie clearly above the sequence of kauffmann et al . kug1602 + 175 ( # 9 ) , ngc6050 ( # 21 ) , ic1182 ( # 39 ) , pgc057064 ( # 29 ) and kug1603 + 179a ( # 33 ) and two of them only marginally ugc10190 ( # 27 ) and sdssj160531.84 + 174826.1 ( # 34 ) . also , broad balmer emission lines are apparent in ic1182 ( # 39 ) and pgc057064 ( # 29 ) . in addition ic1182 ( # 39 ) and sdssj160531.84 + 174826.1 ( # 34 ) are reported to be x - ray sources in ned , which could be indicative of the presence of an agn . however , as shown by rosa - gonzlez et al . ( 2007 ) , the evidence of x - ray emission does not imply unequivocally the presence of an agn since it may come from a recent stellar population . for this reason , we prefer not to include the detection in x - ray as one of our criteria to flag the agn candidates in our sample . ic1182 ( # 39 ) was studied in detail by moles et al . ( 2004 ) and they concluded that although the line ratio diagnostics are not conclusive about the nature of ic1182 ( # 39 ) , they discard the active nature of the nuclear emission of this galaxy based on other criteria . sdssj160531.84 + 174826.1 ( # 34 ) has been reported to host an intermediate - mass black hole , besides being a dwarf disk galaxy ( dong et al . 2007 ) . after these considerations , galaxies in our sample fulfilling any of the two criteria mentioned above are flagged as agn candidates in table [ lista ] . finally , our sample contains a total of 50 galaxies detected in h@xmath0 whose main properties are shown in table [ lista ] : ( 1 ) i d . number ; ( 2 ) galaxy name from ned database or from sdss ; ( 3 ) right ascension ( j2000.0 ) ; ( 4 ) declination ( j2000.0 ) ; ( 5 ) h@xmath0 @xmath37 [ nii ] flux in units of 10@xmath38 erg @xmath27 s@xmath6 uncorrected for galactic extinction ; ( 6 ) h@xmath0 @xmath37 [ nii ] equivalent width in ; ( 7 ) radial ( spectroscopic ) velocity in km s@xmath6 from ned or sdss , when available ; ( 8) @xmath3 in mag , from the sdss @xmath39 magnitudes , assuming a distance to a2151 of 158.3 mpc and an average correction of @xmath40 ( fukugita et al . 1995 ) ; ( 9 ) flag accounting for the cluster membership : `` 0 '' for galaxies with spectroscopic radial velocity , secure members of the cluster ; `` 1 '' for secure members of the cluster from the photo - z criterion ; `` 2 '' for likely members of the cluster from the photo - z criterion ; `` 3 '' for contaminants from the photo - z criterion ; ( 10 ) flag for agn candidates ; ( 11 ) hubble type from leda ( those galaxies with no available hubble type in leda have been classified as `` comp '' from the sdss database , is almost 0.5 times compared to that of the total sample . ] because of their compact appearance in the optical frames ) . figure [ mosaic ] shows the optical frames of the 50 h@xmath0 selected galaxies in the same order as they appear listed in table [ lista ] . we describe in some lines the most remarkable features regarding the h@xmath0 emission properties of the sample : * the reported velocities for the galaxies ngc6050 ( # 21 ) and ugc10190 ( # 27 ) are near the detection limits for the fwhm of our filters , so it is possible that the asymmetric distribution of the h@xmath0 emission of these galaxies could be an effect of their rotation velocity fields , that make the h@xmath0 line lay outside the transmission curves of the filters . * pgc057077 ( # 38 ) is classified in leda as e. however , its sdss spectrum is typical of a star forming galaxy , with strong emission line features , the h@xmath0 emission extending well over the whole galaxy . * two of the galaxies are associated to tidal features [ d97]ce-060 ( # 43 ) and ic1182[s72]d ( # 42 ) and because of their low luminosities they can be considered tidal dwarf candidates . * some galaxies can be associated to galaxy interactions because of their disturbed morphologies and their close projected distance to other cluster members : ngc6045 ( # 13 ) , sdssj160508.81 + 174545.3 ( # 14 ) , ngc6050 ( # 21 ) , sdssj160305.24 + 171136.1 ( # 4 ) , sdssj160304.20 + 171126.7 ( # 5 ) , pgc057064 ( # 29 ) , leda084724 ( # 44 ) , + sdssj160547.17 + 173501.6 ( # 45 ) . * two objects could be isolated hii regions associated to brighter galaxies : + sdssj160528.84 + 174906.2b ( # 31 ) , sdssj160523.67 + 174828.8 ( # 22 ) . * finally , there is a candidate of star formation induced by the ram pressure exerted by the igm : leda 1543586 ( # 15 ) . the h@xmath0 emission of this galaxy is clearly assymetric ( see figure 5 ) with most knots in the side facing the cluster center and almost no emission in the opposite side . this pattern of h@xmath0 emission has been already observed in cluster galaxies ( gavazzi et al . 2001 ) and is associated with the ram pressure exerted by the intergalactic medium on the interstellar medium of disk galaxies . a detailed investigation of the structural properties of the a2151 emission line galaxies will be presented in a forthcoming paper . in this section we discuss on the main statistical properties of the h@xmath0 emission for our a2151 h@xmath0 emitting galaxies . as it has been shown in the previous section , some of the selected emission line galaxies are very likely background contaminants . given that we are interested in the properties of the cluster members , those galaxies considered contaminants ( i.e. flagged `` 3 '' in table [ lista ] ) will be discarded in the subsequent analysis . thus , hereafter we will refer to the subsample of secure ( flags `` 0 '' and `` 1 '' ) and likely ( flag `` 2 '' ) members of the cluster as the cluster galaxies . for clarity , hereafter we will refer to h@xmath41[nii ] as h@xmath0 , although we warn that no correction for [ nii ] contamination was performed for our sample galaxies ( neither for the other samples used throughout this work ) . however , if required for other purposes an efficient correction for the [ nii ] contamination can be succesfully performed ( e.g. following reverte ( 2008 ) ) . figure [ lha_ewha ] shows the h@xmath0 luminosities vs. h@xmath0 equivalent widths for our sample galaxies . the points appear spread in the plot occupying the locus limited by @xmath42h@xmath43 erg s@xmath6 and @xmath44h@xmath45 . the minimum value of @xmath4(h@xmath0 ) measured for our sample is 2 ( see also table 2 ) . we stress the importance of this point since this value has been used in the literature as the limit between active and passive star forming galaxies ( rines et al . 2005 ) , thus meaning that our sample contains all the active star forming galaxies in the surveyed regions of a2151 . the candidate agn galaxies are on average more h@xmath0 luminous than the total sample , reflecting the low fraction of low luminosity agns with respect to high luminosity ones . figure [ mb_lha ] shows a linear relation between @xmath3 and @xmath2(h@xmath0 ) , similar to what has been found for virgo cluster ( see figure 8 of iglesias - pramo et al . 2002 , with data taken from goldmine , gavazzi et al . 2003 ) . again the candidate agn galaxies follow the same trend as the rest . the similar behavior of the candidate cluster agns and the rest of the galaxies suggests that their nuclear h@xmath0 emission , resulting from the non thermal uv radiation , would be small compared to the emission of the gas ionized by the stellar uv radiation . this plot also gives an idea about the sensitivity of our survey . as it can be seen , we detect galaxies down to a value of @xmath46(h@xmath0)/(erg s@xmath6 ) @xmath47 39 and @xmath48 mag , implying that our h@xmath0 selected sample contains not only the bright spirals of a2151 but also a non negligible fraction of star forming dwarf galaxies . for the subsequent discussion we make use of h@xmath0 fluxes and equivalent widths for other volume limited samples of galaxies , namely the samples of coma and a1367 emitters of a1367 in iglesias - pramo et al . ( 2002 ) contains several galaxies for which no information of the redshift was available by that time . given that photo - z information is currently available from the sdss database , we have applied to those galaxies the same criterion of membership to the cluster as applied to our a2151 galaxies in this paper . for this reason , the sample of h@xmath0 emitters of a1367 used in this paper is smaller than the one in iglesias - pramo et al . further details on this point will be given in a forthcoming paper . ] ( from iglesias - pramo et al . 2002 ) and the h@xmath0 survey of the local 11 mpc volume ( from kennicutt et al . 2008 , hereafter k08 ) . the cluster samples will serve as a representation of other high density environments in the nearby universe , and the k08 sample will be taken as a control field sample and it does not contain massive clusters since the two most massive structures of the very local universe ( i.e. the virgo and ursa major clusters ) have been excluded . figure [ lha_mb_all ] shows the @xmath2(h@xmath0 ) vs. @xmath3 for the four galaxy samples under consideration . the k08 galaxies , delineate an ( almost linear ) relatively tight relation with the scatter increasing at low luminosities . the cluster samples follow the same relation although showing a larger scatter . this scatter is mainly due to the presence of cluster galaxies showing , for a given @xmath3 , a lower @xmath2(h@xmath0 ) than the one expected from the k08 sample . the fraction of such galaxies appears maximum for coma and minimum for a2151 . they are mostly bright galaxies , although at least one of them shows magnitude as faint as @xmath49 mag . the k08 sample reaches luminosity values as low as @xmath50h@xmath51/(erg s@xmath6)@xmath52 , as shown in figure [ lha_mb_all ] . the three cluster samples reach luminosities as low as @xmath50h@xmath51/(erg s@xmath6)@xmath53 , though few cluster dwarfs have been meassured with @xmath2(h@xmath0 ) lower than this value . therefore in order to make a robust comparison with k08 we limit our analysis to @xmath50h@xmath51/(erg s@xmath6)@xmath54 39 . figure [ contour ] shows the contours of the @xmath2(h@xmath0 ) vs. @xmath4(h@xmath0 ) plot for the three clusters , overimposed to those of the k08 sample . as it can be seen , the k08 galaxies occupy a large region in the plot with a maximum between @xmath55h@xmath56 erg s@xmath6 and @xmath57h@xmath58 . in contrast , the cluster samples show different topologies in this figure . the distribution of coma presents a well defined maximum , peaking at @xmath4(h@xmath0)@xmath47 1.3 and @xmath46(h@xmath0)/(erg s@xmath6 ) @xmath47 40.6 . this peak slightly shifted to higher @xmath2(h@xmath0 ) can be also identified in the distributions of a1367 and a2151 as a relative maximum . in addition , these two clusters show central maxima near that of the k08 sample , a feature that is missing for coma . besides the fact that k08 sample extends towards lower values of @xmath2(h@xmath0 ) , we remark the different topology of the distributions presented by the clusters with respect to the field sample within the range of @xmath2(h@xmath0 ) and @xmath4(h@xmath0 ) common to the four samples . in the discussion , we argue that these different topologies in fact can reflect a direct imprint of the environment . figure [ histo_lha ] shows the histogram of the @xmath2(h@xmath0 ) distributions of the four samples . the total histogram of the three clusters , made out of a combination of the three original histograms weighted each one by the inverse of the surveyed volume , is shown in the bottom panel ( black dashed line ) together with the histogram of the k08 sample ( continuous line ) . the histogram of coma shows a broad maximum between @xmath59h@xmath60erg s@xmath61 whereas those of a1367 and a2151 show a plateau in this range and a peak at @xmath62h@xmath60erg s@xmath63 . the histogram of the k08 sample shows a monotonic increase until it reaches a maximum at @xmath50h@xmath60erg s@xmath64 and then it shows a decrease for lower luminosities . in order to compare the combined cluster distribution to that of the k08 sample , the cluster distribution has been normalized to the k08 in the range @xmath65h@xmath60erg s@xmath66 . it can be seen that the cluster distribution shows a different shape from k08 in the range @xmath55h@xmath60erg s@xmath67 , reflecting a relative deficit of cluster star - forming galaxies with respect to the field sample . taking a look to figure [ lha_mb_all ] , this range of @xmath2(h@xmath0 ) is mainly populated by star - forming dwarfs and outliers from the k08 relation , these last being a likely analog of the spirals with weak h@xmath0 emission detected in virgo cluster by koopmann & kenney ( 2004 ) . figure [ dist_cum ] shows the cumulative distribution of the h@xmath0 detections versus the distance to the center of the cluster for coma , a1367 and a2151 . the center of each cluster has been taken from ned and is coincident with the corresponding maximum in the galaxy distribution . galaxies are separated in three categories : ( a ) @xmath4(h@xmath0 ) @xmath68 ; ( b ) @xmath69 @xmath70 @xmath4(h@xmath0 ) @xmath71 ; and ( c ) @xmath4(h@xmath0 ) @xmath70 20 ( hereafter ew40 , ew20 and ew0 galaxies respectively ) . we have restricted the samples to those galaxies inside the inner 1 mpc of the center of each cluster since the observations cover reasonably well these regions for the three clusters . as can be seen from the figure , the three clusters show very different behaviors : for a2151 , the three categories follow each other from 0.15 mpc to 0.5 mpc from the center . beyond this distance there are no ew40 galaxies whereas ew20 and ew0 galaxies still can be found until distances of 1 mpc . for coma , the inner 0.3 mpc are almost completely devoid of star forming galaxies . most of the ew40 and ew20 galaxies can be found at distances larger than 0.7 mpc from the center . on the contrary , the ew0 galaxies are homogeneously distributed between 0.3 mpc and 1 mpc . for a1367 the distribution of galaxies is almost homogeneous from 0.15 mpc to 1 mpc for the three categories and it shows an intermediate behavior between coma and a2151 . summarizing , our results show that the ew40 galaxies ( those that most actively form stars ) are centrally concentrated in a2151 whereas avoid the inner regions in coma and a1367 . galaxies with intermediate star formation activity ( the ew20 ones ) tend to follow the ew40 ones in the three clusters , at least for radii shorter than @xmath72 mpc . the ew0 galaxies are the most homogeneously distributed among the three categories for the three clusters . in the previous section we have seen how the environment affects the h@xmath0 emission of cluster galaxies with respect to those in less dense environments . the existence of cluster galaxies with low h@xmath0 emission has already been reported in the literature ( e.g. kennicutt 1983 ; koopmann & kenney 2004 ) , and has been atrributed to mechanisms like galaxy interactions or ram pressure stripping among others ( boselli & gavazzi 2006 , and references therein ) . in fact , it has been found a dependence of star formation rate on local density that holds even at distances will outside the virialized region of the cluster ( e.g. lewis et al . 2002 ; gmez et al . 2003 ; balogh et al . 2004 ; rines et al . 2005 ) in the sense that the higher the local density of galaxies the lower fraction of star forming galaxies . this effect , known as the star formation - density relation seems to be independent of the global environment considered ( cluster , loose group , compact group , etc ) . in addition , poggianti et al . ( 2006 ) have found a relation between the fraction of emission line galaxies and the cluster mass within the virialized region . from a general point of view , it is not unexpected the existence of an imprint of the global cluster environment on properties related to the h@xmath0 luminosity of cluster galaxies : in the previous section a relative deficit of cluster star - forming galaxies with respect to the field sample has been found . in addition , some cluster to cluster differences have been reported : we have shown in figure [ contour ] that the well defined topology of k08 distribution , showing a well defined single maximum , is not reproduced by the cluster samples which show different topologies . these results can be interpreted assuming a sequence parameterized by the total mass of the clusters : as the mass of the cluster increases and the intergalactic medium becomes denser and hotter , the ram pressure stripping becomes more efficient at removing the gas component from galaxies . since dwarf galaxies present smaller escape velocities , the ram pressure stripping is expected to be more efficient for them than for giant galaxies ( e.g. mori & burkert 2000 ) . the measured x - ray luminosities are 40.5@xmath2210@xmath73/4.5@xmath2210@xmath73/0.9@xmath2210@xmath73 erg s@xmath6 for coma / a1367/a2151 ( magri et al . 1988 ) = 50 km s@xmath6 mpc@xmath6 was assumed by these authors to estimate the x - ray luminosities . ] , and the velocity dispersions are 1008/879/691 km s@xmath6 for coma / a1367/a2151 ( struble & rood 1999 ) . according to this sequence , as cluster mass increases , an increasing fraction of low luminosity star - forming galaxies should be devoided of gas ( mori & burkert 2000 ; boselli et al . 2008 ) and the ram pressure stripping can be observed in more luminous galaxies . thus , for very massive clusters few ( if any ) low luminosity star - forming galaxies would be observed in h@xmath0 , and the h@xmath0 emission of high luminosity galaxies would suffer from severe gas stripping . this scenario is qualitatively illustrated by figure [ lha_mb_all ] , where in the case of coma the outliers from the k08 relation are mainly bright ( non dwarf ) galaxies . for a1367 and a2151 these outliers shift towards lower luminosity . this rearrangement of the h@xmath0 emission in cluster galaxies is reflected also in the change of topolgy and distribution shown in figures [ contour ] and [ histo_lha ] , respectively . the spatial distribution of star forming galaxies in the clusters shows a clear effect of the cluster environment on their recent star formation activity ( figure [ dist_cum ] ) : for coma and a1367 the ew40 and ew20 galaxies avoid the central regions of the clusters and most of them are located beyond 0.5 mpc from the cluster center , illustrating the existence of mechanisms quenching the star formation . however , this does not hold for a2151 : in this cluster all the star forming galaxies are distributed in a similar way with clustercentric distance and most of them can be located at distances lower than 0.5 mpc from the cluster center . interestingly , for a2151 we can find most of the strong h@xmath0 emitting galaxies ( ew40 ) at very short distances from the cluster center , contrary to what is seen in the other two clusters . taking a look to the x - ray map ( magri et al . 1988 , their plate 6 ) and to the 2mass and sdss galaxy distributions of this cluster , we can see that the maximum of the cluster galaxy distribution is not coincident with the primary x - ray maximum , but with a secondary one located @xmath74 arcmin to the east of the primary maximum . this implies that most a2151 galaxies are not embedded in a hot x - ray intergalactic medium . in the case of coma and a1367 , despite they also show some degree of x - ray structure , the maxima of their galaxy distributions are located well inside the x - ray emitting regions . the optical and x - ray structure of a2151 suggests that an evolved structure , coincident with the primary x - ray maximum , is incorporating a less evolved structure , coincident with the maximum of the galaxy distribution and the secondary x - ray maximum , thus giving rise to a cluster merger event . most h@xmath0 emitters in a2151 are located around the secondary x - ray maximum , in the regions of higher density of a2151 galaxies . this last point gives support to the idea that cluster mergers could trigger episodes of star formation ( owen et al . 1999 ; poggianti et al . 2004 ; johnston - hollitt et al . 2008 ) . putting all these results together , we conclude that both , the global cluster environment as well as the cluster merging history play a non negligible role in the integral star formation properties of clusters of galaxies . further observations are required to confirm these points . in a near future a larger and homogeneous h@xmath0 survey covering a larger and homogeneous sample of nearby clusters will help to shed light on these and other points regarding the recent star formation history of cluster galaxies . we want to thank the anonymous referee for her / his interesting comments and suggestions that have contributed to the improvement of this manuscript . thanks are given to p. papaderos for his help and a. boselli for many useful comments and suggestions in the initial stages of this work . we acknowledge financial support by the spanish ministerio de ciencia e innovacin grants aya2007 - 67645-c03 - 02 . this research has made use of the nasa / ipac extragalactic database ( ned ) which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration . 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 . iraf is distributed by the national optical astronomy observatories , which are operated by the association of universities for research in astronomy , inc . , under cooperative agreement with the national science foundation . the wht is operated on the island of la palma by the isaac newton group in the spanish observatorio del roque de los muchachos of the instituto de astrofsica de canarias . the data presented here have been taken using alfosc , which is owned by the instituto de astrofsica de andaluca ( iaa ) and operated at the nordic optical telescope under agreement between iaa and the nbifafg of the astronomical observatory of copenhagen . baldwin , j. a. , phillips , m. m. , & terlevich , r. 1981 , , 93 , 5 balogh , m. l. , baldry , i. k. , nichol , r. , miller , c. , bower , r. , & glazebrook , k. 2004 , , 615 , l101 barmby , p. , & huchra , j.p . 1998 , aj , 115 , 6 barth , c.s . , cepa , j. , vlchez , j.m . , & dottori , h.a . 1994 , aj , 108 , 2069 bird , c.m . , davis , d.m . , & beers , t.c . 1995 , aj , 109 , 920 boselli , a. , & gavazzi , g. 2006 , , 118 , 517 boselli , a. , boissier , s. , cortese , l. , & gavazzi , g. 2008 , , 674 , 742 cardelli , j. a. , clayton , g. c. , & mathis , j. s. 1989 , , 345 , 245 chincarini , g. , thompson , l.a . , & hood , h.j . 1981 , apj , 249,47 dickey , j.m . 1997 , aj , 113 , 1939 domnguez , m. j. , zandivarez , a. a. , martnez , h. j. , merchn , m. e. , muriel , h. , & lambas , d. g. 2002 , , 335 , 825 dong , x. , et al . 2007 , , 657 , 700 fukugita , m. , shimasaku , k. , & ichikawa , t. 1995 , , 107 , 945 gavazzi , g. , boselli , a. , mayer , l. , iglesias - paramo , j. , vlchez , j. m. , & carrasco , l. 2001 , , 563 , l23 gavazzi , g. , boselli , a. , pedotti , p. , gallazzi , a. , & carrasco , l. 2002 , , 396 , 449 gavazzi , g. , boselli , a. , donati , a. , franzetti , p. , & scodeggio , m. 2003 , , 400 , 451 giovanelli , r. & haynes , m.p . 1985 , apj , 292 , 404 gmez , p.l . , nichol , r.c . , miller , c.j . , balogh , m.l . , goto , t. , zabludoff , a.i . , romer , a.k . , bernardi , m. , sneth , r. , hopkins , a.m. , and 7 coauthors 2003 , apj , 584 , 210 huang , z. , & sarazin , c.l . 1996 , apj , 461 , 622 iglesias - pramo , j. , boselli , a. , cortese , l. , vlchez , j.m . , & gavazzi , g. 2002 , a&a , 384,383 johnston - hollitt , m. , sato , m. , gill , j. a. , fleenor , m. c. , & brick , a .- m . 2008 , , 390 , 289 kauffmann , g. , et al . 2003 , , 346 , 1055 kennicutt , r. c. , jr . 1983 , , 88 , 483 kennicutt , r. c. , jr . , lee , j. c. , funes , s. j. , jos g. , sakai , s. , & akiyama , s. 2008 , , 178 , 247 koopmann , r. a. , & kenney , j. d. p. 2004 , , 613 , 866 lewis , i. , balogh , m. , de propris , r. , couchm w. , bower , r. , offer , a. , bland - hawthorn , j. , baldry , i.k . , baugh , c. , bridges , t. , and 23 coauthors 2002 , mnras , 334 , 673 magri , c. , haynes , m.p . , forman , w. , jones , c. , & giovanelli , r. 1988 , apj , 333 , 136 moles , m. , bettoni , d. , fasano , g. , kjrgaard , p. , varela , j. , & milvang - jensen , b. 2004 , , 418 , 495 mori , m. , & burkert , a. 2000 , , 538 , 559 oke j.b . 1990 , aj 99 , 1621 owen , f. n. , ledlow , m. j. , keel , w. c. , & morrison , g. e. 1999 , , 118 , 633 poggianti , b. m. , bridges , t. j. , komiyama , y. , yagi , m. , carter , d. , mobasher , b. , okamura , s. , & kashikawa , n. 2004 , , 601 , 197 poggianti , b. m. , et al . 2006 , , 642 , 188 reverte , d. , vlchez , j. m. , hernndez - fernndez , j. d. , & iglesias - pramo , j. 2007 , , 133 , 705 reverte , d. 2008 , ph.d . thesis , universidad de granada rines , k. , geller , m. j. , kurtz , m. j. , & diaferio , a. 2005 , , 130 , 1482 rosa - gonzlez , d. , burgarella , d. , nandra , k. , kunth , d. , terlevich , e. , & terlevich , r. 2007 , , 379 , 357 snchez - janssen , r. , iglesias - pramo , j. , muoz - tun , c. , aguerri , j.a.l . , & vlchez , j.m . 2005 , a&a 434 , 521 schlegel , d.j . , finkbeiner , d.p . , & davis , m. 1998 , apj , 500 , 525 struble , m. f. , & rood , h. j. 1999 , , 125 , 35 solanes , j.m . 2001 ` constructing the universe with clusters of galaxies ' , ed . f. durret & d. gerbal . ( electronic book ) tanaka , m. , goto , t. , okamura , s. , shimasaku , k. , & brinkmann , j. 2004 , , 128 , 2677 .pointings of the a2151 survey : ( 1 ) and ( 2 ) r.a . and dec . of the approximate central position of the pointings ; ( 3 ) telescope used to obtain the data : not ( nordic optical telescope ) , wht ( william herschell telescope ) ; ( 4 ) central wavelength ( ) of the filter used to measure the h@xmath0 emission ; ( 5 ) broadband filter employed to measure the continuum emission ; ( 7 ) total exposure time of the h@xmath0 frames . [ cols="^,^,^,^,^,^ " , ] ccccccccccc i d . # & galaxy & ra & dec & @xmath75(h@xmath0+[nii ] ) & ew@xmath76 & @xmath77 & @xmath3 & & hubble + number & name & j2000.0 & j2000.0 & ( @xmath78erg @xmath27 s@xmath6 ) & ( ) & ( km s@xmath6 ) & ( mag ) & member & agn & type + & & ( hh mm ss ) & ( dd mm ss ) & & & & & & & + # 1 & sdssj160258.11 + 171227.8 & 16 02 58.1 & 17 12 27.9 & [email protected] & 49@xmath795 & & @xmath80 & 3 & - & comp + # 2 & sdssj160258.70 + 171138.1 & 16 02 58.7 & 17 11 38.2 & [email protected] & 30@xmath794 & & @xmath81 & 2 & - & comp + # 3 & sdssj160302.54 + 171007.0 & 16 03 02.5 & 17 10 07.1 & [email protected] & 5@xmath792 & & @xmath82 & 3 & - & comp + # 4 & sdssj160305.24 + 171136.1 & 16 03 05.2 & 17 11 36.2 & [email protected] & 4@xmath791 & 9930 & @xmath83 & 0 & - & comp + # 5 & sdssj160304.20 + 171126.7 & 16 03 04.2 & 17 11 26.7 & [email protected] & 31@xmath792 & 10830 & @xmath84 & 0 & - & comp + # 6 & leda084703 & 16 03 05.7 & 17 10 20.4 & 27@xmath791 & 53@xmath792 & 10093 & @xmath85 & 0 & - & sm + # 7 & leda084710 & 16 04 20.4 & 17 26 11.2 & 37@xmath792 & 23@xmath796 & 10723 & @xmath86 & 0 & - & sb + # 8 & kug1602 + 174a & 16 04 39.0 & 17 20 59.9 & 18@xmath793 & 11@xmath792 & 10626 & @xmath87 & 0 & - & sb + # 9 & kug1602 + 175 & 16 04 45.4 & 17 26 54.3 & 134@xmath798 & 32@xmath792 & 10638 & @xmath88 & 0 & y & sc + # 10 & kug1602 + 174b & 16 04 47.6 & 17 20 52.0 & 56@xmath795 & 22@xmath792 & 10581 & @xmath89 & 0 & - & sb + # 11 & @xmath90d97@xmath91$]ce-200 & 16 05 06.8 & 17 47 02.0 & [email protected] & 29@xmath792 & 9927 & @xmath92 & 0 & - & s + # 12 & leda084719 & 16 05 07.1 & 17 38 57.8 & 15@xmath793 & 8@xmath791 & 10162 & @xmath93 & 0 & - & sbc + # 13 & ngc6045 & 16 05 07.9 & 17 45 27.6 & 192@xmath7913 & 22@xmath792 & 9986 & @xmath94 & 0 & - & sbc + # 14 & sdssj160508.81 + 174545.3 & 16 05 08.8 & 17 45 45.4 & [email protected] & 36@xmath793 & & @xmath95 & 1 & - & comp + # 15 & leda1543586 & 16 05 10.5 & 17 51 16.1 & [email protected] & 30@xmath792 & 9905 & @xmath96 & 0 & - & sc + # 16 & ic1173 & 16 05 12.6 & 17 25 22.3 & 98@xmath7910 & 14@xmath792 & 10431 & @xmath97 & 0 & - & sabc + # 17 & sdssj160520.64 + 175201.5 & 16 05 20.6 & 17 52 01.5 & [email protected] & 32@xmath792 & & @xmath98 & 2 & - & comp + # 18 & sdssj160520.58 + 175210.6 & 16 05 20.6 & 17 52 10.7 & [email protected] & 20@xmath792 & 10350 & @xmath99 & 0 & - & comp + # 19 & sdssj160521.50 + 175337.8 & 16 05 21.5 & 17 53 37.8 & [email protected] & 12@xmath792 & & @xmath100 & 1 & - & comp + # 20 & @xmath90d97@xmath91$]ce-143 & 16 05 21.7 & 17 51 56.3 & [email protected] & 11@xmath792 & 11587 & @xmath101 & 0 & - & comp + # 21 & ngc6050 & 16 05 23.4 & 17 45 25.8 & 38.@xmath797 & 6@xmath791 & 9572 & @xmath102 & 0 & y & sabc + # 22 & sdssj160523.67 + 174828.8 & 16 05 23.7 & 17 48 28.9 & [email protected] & 29@xmath799 & & @xmath103 & 1 & - & comp + # 23 & sdssj160523.66 + 174832.3 & 16 05 23.7 & 17 48 32.3 & [email protected] & 27@xmath796 & & @xmath104 & 1 & - & comp + # 24 & sdssj160524.27 + 175329.3 & 16 05 24.3 & 17 53 29.4 & [email protected] & 46@xmath793 & & @xmath105 & 1 & - & comp + # 25 & sdssj160524.65 + 175213.0 & 16 05 24.7 & 17 52 13.0 & [email protected] & 81@xmath7916 & & @xmath106 & 3 & - & comp + # 26 & @xmath90dkp87@xmath91$]160310.21 + 175956.7 & 16 05 25.0 & 17 51 50.6 & 3@xmath791 & 22@xmath796 & & @xmath107 & 1 & - & comp + # 27 & ugc10190 & 16 05 26.3 & 17 41 48.6 & 14@xmath793 & 8@xmath791 & 11077 & @xmath108 & 0 & y & sc + # 28 & sdssj160526.83 + 175218.6 & 16 05 26.8 & 17 52 18.7 & [email protected] & 36@xmath798 & & @xmath109 & 3 & - & comp + # 29 & pgc057064 & 16 05 27.2 & 17 49 51.6 & 50@xmath7912 & 21@xmath797 & 10254 & @xmath110 & 0 & y & sab + # 30 & sdssj160528.22 + 175212.3 & 16 05 28.2 & 17 52 12.3 & [email protected] & 15@xmath797 & & @xmath111 & 1 & - & comp + # 31 & sdssj160528.84 + 174906.2b & 16 05 28.6 & 17 49 13.7 & [email protected] & 190@xmath7930 & & @xmath112 & 1 & - & comp + # 32 & sdssj160528.84 + 174906.2 & 16 05 28.8 & 17 49 06.2 & [email protected] & 11@xmath796 & & @xmath113 & 1 & - & comp + # 33 & kug1603 + 179a & 16 05 30.6 & 17 46 07.2 & 98@xmath794 & 54@xmath792 & 11189 & @xmath114 & 0 & y & sb + # 34 & sdssj160531.84 + 174826.1 & 16 05 31.8 & 17 48 26.2 & [email protected] & 26@xmath792 & 9402 & @xmath115 & 0 & y & comp + # 35 & sdssj160532.06 + 174617.8 & 16 05 32.1 & 17 46 17.9 & [email protected] & 12@xmath792 & & @xmath116 & 3 & - & comp + # 36 & sdssj160533.36 + 174548.3 & 16 05 33.4 & 17 45 48.3 & [email protected] & 41@xmath793 & & @xmath117 & 1 & - & comp + # 37 & sdssj160533.98 + 174554.1 & 16 05 34.0 & 17 45 54.1 & [email protected] & 15@xmath792 & & @xmath118 & 1 & - & comp + # 38 & pgc057077 & 16 05 34.2 & 17 46 11.8 & 87@xmath792 & 148@xmath795 & 10233 & @xmath119 & 0 & - & e + # 39 & ic1182 & 16 05 36.8 & 17 48 07.5 & 375@xmath7911 & 111@xmath793 & 10240 & @xmath120 & 0 & y & s0/a + # 40 & sdssj160539.93 + 175243.9 & 16 05 39.9 & 17 52 43.9 & [email protected] & 16@xmath797 & & @xmath109 & 3 & - & comp + # 41 & sdssj160540.47 + 175102.3 & 16 05 40.5 & 17 51 02.4 & [email protected] & 18@xmath798 & & @xmath121 & 3 & - & comp + # 42 & ic1182:[s72]d & 16 05 41.9 & 17 47 58.4 & [email protected] & 195@xmath797 & 9983 & @xmath122 & 0 & - & comp + # 43 & @xmath90d97@xmath91$]ce-060 & 16 05 44.8 & 17 42 20.2 & [email protected] & 2@xmath792 & 11100 & @xmath123 & 0 & - & comp + # 44 & leda084724 & 16 05 45.5 & 17 34 56.7 & 30@xmath791 & 30@xmath794 & 12206 & @xmath124 & 0 & - & sbc + # 45 & sdssj160547.17 + 173501.6 & 16 05 47.2 & 17 35 01.7 & [email protected] & 6@xmath792 & & @xmath125 & 1 & - & comp + # 46 & leda3085054 & 16 05 56.0 & 17 42 33.9 & [email protected] & 69@xmath794 & 11145 & @xmath126 & 0 & - & s + # 47 & sdssj160556.98 + 174304.1 & 16 05 57.0 & 17 43 04.1 & [email protected] & 60@xmath793 & & @xmath113 & 1 & - & comp + # 48 & leda140568 & 16 06 00.1 & 17 45 52.0 & [email protected] & 15@xmath792 & 11959 & @xmath84 & 0 & - & s + # 59 & cgcg108 - 149 & 16 06 35.3 & 17 53 33.2 & 35@xmath793 & 10@xmath794 & 11049 & @xmath127 & 0 & - & saba + # 50 & pgc057185 & 16 06 48.2 & 17 38 51.6 & 11@xmath792 & 6@xmath792 & 11246 & @xmath128 & 0 & - & sbc + ) sdss sample . the dot - dashed lines correspond to the transmittance profiles of the narrow band filters used in this work . the horizontal lines represent the interval where the transmittance is larger than 50% of the peak transmitance for each filter . the upper x - axis indicates the wavelength ( in ) of the redshifted h@xmath0 line corresponding to the radial velocity indicated in the lower x - axis.,width=604 ] 6584/h@xmath0 vs. [ oiii]5007/h@xmath31 for the h@xmath0 emitting galaxies with sdss spectra . only galaxies for which the four lines are detected in emission are plotted . the solid line corresponds to the separation between starbursts and agns of kauffmann et al . the names of the seven galaxies located to the right of the line are indicated.,width=604 ] frames of the a2151 emitting galaxies with the r contours overimposed . the h@xmath0 surface brightness is indicated by the color scale to the right of each image in 10@xmath130erg / cm@xmath14/s / arcsec@xmath14 . the superposed contours are in ab magnitudes arcsec@xmath30 . the lowest contour is at 24 mag arcsec@xmath30 and the increment is 0.2 mag arcsec@xmath30,title="fig:",width=226 ] frames of the a2151 emitting galaxies with the r contours overimposed . the h@xmath0 surface brightness is indicated by the color scale to the right of each image in 10@xmath130erg / cm@xmath14/s / arcsec@xmath14 . the superposed contours are in ab magnitudes arcsec@xmath30 . the lowest contour is at 24 mag arcsec@xmath30 and the increment is 0.2 mag arcsec@xmath30,title="fig:",width=377 ] frames of the a2151 emitting galaxies with the r contours overimposed . the h@xmath0 surface brightness is indicated by the color scale to the right of each image in 10@xmath130erg / cm@xmath14/s / arcsec@xmath14 . the superposed contours are in ab magnitudes arcsec@xmath30 . the lowest contour is at 24 mag arcsec@xmath30 and the increment is 0.2 mag arcsec@xmath30,title="fig:",width=302 ] frames of the a2151 emitting galaxies with the r contours overimposed . the h@xmath0 surface brightness is indicated by the color scale to the right of each image in 10@xmath130erg / cm@xmath14/s / arcsec@xmath14 . the superposed contours are in ab magnitudes arcsec@xmath30 . the lowest contour is at 24 mag arcsec@xmath30 and the increment is 0.2 mag arcsec@xmath30,title="fig:",width=302 ] ( h@xmath0 ) vs. @xmath4(h@xmath0 ) plot for the clusters . those of k08 are also plotted in the three panels for comparison . the black cross indicates the position of the maximum of the k08 distribution.,width=604 ] + [ nii ] luminosity for the detected galaxies . the black line in the bottom panel corresponds to the combined distribution of the three clusters weighted each one to the corresponding surveyed volume and arbitrarily normalized.,width=604 ] emitting galaxies versus the distance to the center of the cluster for coma , a1367 , a2151 . blue , brown and orange lines correspond to galaxies with @xmath4(h@xmath0 ) @xmath132 40 , 20 @xmath70 @xmath4(h@xmath0 ) @xmath70 40 and @xmath4(h@xmath0 ) @xmath70 20 respectively.,title="fig:",width=302 ] emitting galaxies versus the distance to the center of the cluster for coma , a1367 , a2151 . blue , brown and orange lines correspond to galaxies with @xmath4(h@xmath0 ) @xmath132 40 , 20 @xmath70 @xmath4(h@xmath0 ) @xmath70 40 and @xmath4(h@xmath0 ) @xmath70 20 respectively.,title="fig:",width=302 ] emitting galaxies versus the distance to the center of the cluster for coma , a1367 , a2151 . blue , brown and orange lines correspond to galaxies with @xmath4(h@xmath0 ) @xmath132 40 , 20 @xmath70 @xmath4(h@xmath0 ) @xmath70 40 and @xmath4(h@xmath0 ) @xmath70 20 respectively.,title="fig:",width=302 ]

in this contribution we describe computational tools that permit the evaluation of multi - loop scattering amplitudes in supergravity , in terms of amplitudes in super - yang - mills theory . we also discuss the remarkable ultraviolet behavior of supergravity , which follows from these amplitudes , and is as good as that of super - yang - mills theory through at least four loops . ucla/11/tep/104 su - itp-11/08 slac pub14380 cern ph th/2011 - 030 + saclay ipht

1103.1848

it is well known that quantum gravity is non - renormalizable by power counting , due to the dimensionful nature of newton s constant , @xmath3 . string theory cures these divergences by introducing a new length scale , related to the string tension , at which particles are no longer point - like . the question we wish to address in this contribution is whether a non - point - like theory is actually necessary for perturbative finiteness . perhaps with enough symmetry a point - like theory of quantum gravity could have an ultraviolet - finite perturbative expansion . in particular , we shall consider the theory of gravity with the maximal supersymmetry compatible with having particles of at most spin two the ungauged version of @xmath0 supergravity @xcite . the on - shell ultraviolet divergences of @xmath0 supergravity , _ i.e. _ those which can not be removed by field redefinitions , can be probed by studying the ultraviolet behavior of multi - loop on - shell amplitudes for graviton scattering . such scattering amplitudes would be very difficult to compute in a conventional framework using feynman diagrams . however , tree amplitudes in gravity can be expressed in terms of tree amplitudes in gauge theory , by making use of the kawai - lewellen - tye ( klt ) relations @xcite , or more recent relations found by three of the present authors @xcite . loop amplitudes can be constructed efficiently from tree amplitudes via generalized unitarity @xcite , particularly in theories with maximal supersymmetry . using these methods , the four - graviton amplitude in @xmath0 supergravity has been computed at two @xcite , three @xcite and ( most recently ) four loops @xcite . aspects of this program have been reviewed previously in refs . @xcite . there are many other proposals for making sense of quantum gravity with point - like particles . for example , the asymptotic safety program @xcite proposes that the einstein action for gravity flows in the ultraviolet to a nontrivial , lorentz - invariant fixed point . it has also been suggested that the ultraviolet theory could break lorentz invariance @xcite . in contrast to these two particular approaches , here we will do conventional perturbation theory around a ( possible ) gaussian fixed point . the remainder of this article is organized as follows . in we review what is known about the potential counterterms for @xmath0 supergravity , based on constraints coming from both @xmath0 supersymmetry and @xmath4 invariance . in we briefly mention the connection between amplitude divergences in various dimensions and the associated counterterms , for both @xmath0 supergravity and the ( closely related ) @xmath1 super - yang - mills theory . in we review the klt relations between gravity and gauge tree amplitudes . in we review how generalized unitarity permits the efficient reconstruction of multi - loop amplitudes from tree amplitudes . in we show how the combination of unitarity and the klt relations simplifies the computation of @xmath0 supergravity loop amplitudes , by relating them to ( planar and non - planar ) loop amplitudes in @xmath1 super - yang - mills theory . finally , in we describe the four - graviton amplitudes that have been determined at two , three and four loops using these methods , and we discuss their ultraviolet properties . we present our conclusions in . in field theory , ultraviolet divergences are associated with local counterterms . the divergences that survive in on - shell scattering amplitudes should respect the symmetries of the theory ; in theories of gravity the counterterms should be generally covariant . thus they are expressible as products of the riemann tensor @xmath5 , along with covariant derivatives @xmath6 acting on it . ( if matter is present , then the energy momentum tensor @xmath7 can also appear . ) the loop - counting parameter @xmath8 has mass dimension @xmath9 , while the riemann tensor has mass dimension 2 : @xmath10 . therefore , by dimensional analysis , an @xmath11-loop counterterm has the generic form ( suppressing all lorentz indices ) @xmath12 for some power @xmath13 . nonlinear field redefinitions of the einstein action allow the removal of the ricci tensor @xmath14 and scalar @xmath15 from potential counterterms . after making such redefinitions , the only available one - loop counterterm ( in a theory of pure gravity ) , @xmath16 , is equivalent to the gauss - bonnet term . the latter is a total derivative , and can not be generated in perturbation theory . this fact explains why pure gravity is finite at one loop , although there are divergences if matter is present @xcite . in any pure supergravity theory , _ i.e. _ one in which all states are related by supersymmetry to the graviton , there are also no divergences at two loops . the reason is that the unique potential counterterm , @xmath17 , is incompatible with @xmath18 supersymmetry . when sandwiched between four graviton plane - wave states , @xmath19 produces a nonzero matrix element @xcite for helicity configurations @xmath20 that are forbidden by supersymmetry ward identities @xcite . again , if matter super - multiplets are present , other counterterms are available , and lower - loop divergences are possible , even at one loop @xcite . in pure supergravity , the first potential counterterm appears at three loops @xcite , and is often abbreviated as @xmath21 . it has long been known to be compatible with not just @xmath18 supersymmetry , but the full @xmath0 , because it appears as the first subleading term ( after the einstein action ) in the low - energy limit of the four - graviton scattering amplitude in type ii closed superstring theory @xcite , r^4 invariants are @xmath22 , @xmath23 , @xmath24 , and @xmath25 stands for any of the @xmath26 four - point amplitudes in @xmath27 supergravity ( after removing the gravitational coupling constant ) . the @xmath21 operator was ruled out as a counterterm for @xmath0 supergravity by analyzing the ultraviolet behavior of the three - loop four - graviton amplitude @xcite ( see ) . recently , elvang , freedman and kiermaier @xcite studied the constraints of @xmath0 supersymmetry on counterterms of higher operator dimension , and also with more than four powers of the riemann tensor . the latter only affect amplitudes with more than four external legs . the first non - vanishing @xmath28-graviton tree amplitudes are the maximally - helicity - violating ( mhv ) ones , which contain two negative graviton helicities , and @xmath29 positive helicities . for mhv amplitudes , the supersymmetry ward identities @xcite imply that the amplitudes , divided by a simple prefactor , are bose symmetric @xcite . all non - vanishing four - point amplitudes are mhv ( for gravitons only , the only non - vanishing case is @xmath30 ) . therefore @xmath0-supersymmetric on - shell counterterms of the form @xmath31 can be classified in terms of bose - symmetric polynomials @xmath32 of degree @xmath33 , where @xmath34 . this analysis leads to one independent operator each of the form @xmath21 and @xmath31 for @xmath35 , with multiple operators appearing first at order @xmath36 . by dimensional analysis , @xmath31 counterterms are associated with divergences in @xmath37 at loop order @xmath38 . all five - point amplitudes are mhv as well ( for gravitons only , either @xmath39 or its parity conjugate ) , so the bose - symmetry constraints ( in more variables ) are still valid . for @xmath40 , a much more sophisticated analysis of the @xmath0 supersymmetry ward identities on next - to - mhv amplitudes is required @xcite . the upshot is that @xmath0 supersymmetry alone is sufficient to rule out all counterterms through seven loops except for @xmath21 , @xmath41 and @xmath42 . ( earlier work ruled out the four - loop counterterms @xmath43 and @xmath44 @xcite . ) however , there is another constraint on counterterms in @xmath0 supergravity in @xmath37 , and that is invariance under the continuous symmetry @xmath4 , a non - compact form of the exceptional lie group @xmath45 @xcite . the theory contains 70 massless scalars , which parametrize the coset space @xmath46 . the non-@xmath47 part of @xmath4 is realized nonlinearly , through motions on the coset manifold parametrized by the scalar fields . therefore it imposes another set of amplitude ward identities , which are associated with soft limits as the momentum of one or more scalar particles approaches zero @xcite . for example , in the limit that a single scalar becomes soft , all the matrix elements of a potential counterterm should vanish . if @xmath4 is also a symmetry at the quantum level , then these properties can be used to constrain potential counterterms . the @xmath47 subgroup of @xmath4 was shown to be non - anomalous at one - loop long ago @xcite . more recently , bossard , hillmann and nicolai @xcite , using a formulation for the vector fields that has manifest electric - magnetic duality , but is not lorentz covariant , have extended this result to the full @xmath4 symmetry , and to all orders in perturbation theory . in ref . @xcite it was found that the single - soft - scalar limit was non - vanishing for the operator @xmath48 generated by string theory , where @xmath49 is the dilaton ( plus terms generated by @xmath0 supersymmetry ) . this result suggested that the @xmath21 operator might be ruled out as a counterterm . a more refined analysis @xcite isolated the matrix elements generated solely by the @xmath47-singlet operator @xmath21 ( _ i.e. _ removing effects of the dilaton ) , and still found a nonvanishing single - soft limit , thereby demonstrating at the amplitude level that the @xmath21 counterterm is not allowed by linearized @xmath4 . later this analysis was extended @xcite to a large set of higher - dimension operators , and has served to rule out , via @xmath4 , the @xmath41 and @xmath42 potential counterterms mentioned above , as well as to constrain the potential seven - loop counterterm to have a unique form , corresponding to @xmath50 ( plus terms generated by @xmath0 supersymmetry ; see ref . @xcite ) . in other words , the seven - loop finiteness of @xmath0 supergravity in @xmath37 can be assessed purely by computing the four - graviton scattering amplitude . similar conclusions about the finiteness of @xmath0 supergravity through seven loops ( as well as results concerning first divergences in higher - dimensional versions of the theory ) , were arrived at in ref . @xcite , based also on the @xmath4 invariance of counterterms , but using three different lines of analysis , including the dimensional reduction of higher - dimensional counterterms . a seven - loop , @xmath0 supersymmetric counterterm was constructed long ago @xcite , but that construction was not manifestly @xmath4 invariant . more recently it was found @xcite that this candidate counterterm can be identified with the volume of the on - shell @xmath0 superspace , and that it is @xmath4 invariant , although it is still possible that it vanishes after using the classical field equations . however , it seems more likely that the volume coincides with the @xmath50 potential counterterm that passes the @xmath4 constraints also studied in ref . @xcite . on the other hand , ref . @xcite has discussed the constraints from @xmath4 in the context of a light - cone superspace approach , and argues that the theory is perturbatively finite to all loop orders . even if a counterterm is allowed by all known symmetries , that does not necessarily mean that its coefficient is nonzero . only an explicit computation can determine this property for certain . seven - loop four - graviton scattering amplitudes are still a bit beyond present technology . however , the four - loop amplitude can , and has been , computed @xcite , and furthermore it also allows access to the @xmath50 potential counterterm , albeit in a different spacetime dimension . in general , we can test the ultraviolet behavior of the four - graviton scattering amplitude in @xmath0 supergravity at any loop order @xmath11 by increasing the spacetime dimension @xmath51 associated with the loop - momentum integration , until the amplitude starts to diverge . it is instructive to compare this behavior with the corresponding behavior of the maximally supersymmetric gauge theory , @xmath1 super - yang - mills theory ( @xmath1 sym ) . the latter theory is known to be finite to all loop orders in @xmath37 @xcite . however , it diverges in @xmath52 . the critical dimension @xmath53 , in which the theory first diverges as @xmath51 increases , depends on the number of loops , and is given by the formula @xcite , d_c(l)|_sym = 4 + ( l>1 ) . [ neq4dcl ] the surprising result from the four - graviton computations to be described below , is that , through four loops , @xmath0 supergravity is just as well behaved , d_c(l)|_sugra = 4 + ( l=2,3,4 ) . [ neq8dcl ] in both theories , the one - loop case is special , and the first divergence is in eight dimensions ( @xmath54 ) . clearly , the equality between ( [ neq4dcl ] ) and ( [ neq8dcl ] ) must break at some point , if @xmath0 supergravity is to diverge in four dimensions . for @xmath1 sym , the divergences in the critical dimension are all associated with a single type of counterterm , for @xmath55 , of the general form @xmath56 , where @xmath57 is the gluon field strength and the color structure is generic . given this fact , and recalling that the loop - counting parameter for gauge theory is dimensionless , while that for gravity is dimensionful , @xmath3 , the only way that the two formulas for @xmath53 can coincide , is if each successive @xmath0 supergravity divergence in the critical dimension for @xmath58 is associated with a counterterm with two more derivatives ( additional powers of the curvature beyond @xmath21 would not produce a divergence in the four - graviton amplitude ) . indeed , the associated higher - dimensional counterterms have the form @xmath59 , @xmath58 . thus the @xmath50 potential counterterm would correspond to the divergence of the four - loop four - graviton scattering amplitude in @xmath60 . ( we do not yet know for sure whether the amplitude diverges in this dimension ; we do know that it does not diverge in lower dimensions . ) furthermore , when the divergence in the five - loop amplitude is computed , one of two things must happen : either ( 1 ) the equality of eqs . ( [ neq4dcl ] ) and ( [ neq8dcl ] ) must break , or else ( 2 ) the appropriate operator for describing the five - loop divergence in the critical dimension must be @xmath61 . because this operator has two _ more _ derivatives on it than the potential seven - loop counterterm in @xmath37 , @xmath62 , possibility ( 2 ) would be a strong indicator that this counterterm is not present . on the other hand , there have been predictions , based on the general structure of contributions in a world - line formalism using the non - minimal pure spinor formalism @xcite , that the equally good ultraviolet behavior of @xmath0 supergravity and @xmath1 sym will break at five loops . clearly the ultraviolet behavior of the five - loop four - graviton amplitude is an important outstanding question , which will shed strong light on the potential seven - loop divergence of @xmath0 supergravity in four dimensions . in the remainder of this contribution , we will outline the technical tools that have made possible the computation of the complete four - graviton scattering amplitude in @xmath0 supergravity through four loops , as well as the extraction of its ultraviolet divergence in the appropriate higher spacetime dimensions . as mentioned in the introduction , tree amplitudes in gravity can be expressed in terms of tree amplitudes in gauge theory , specifically as bilinear combinations of gauge amplitudes . the reason this will prove so useful to us is that , by using generalized unitarity , we will be able to chop the gravity loop amplitudes up into products of gravity trees . then we can use the gravity - gauge relations to write everything in terms of products of gauge - theory trees , products which actually appear in cuts of gauge loop amplitudes . in this way , multi - loop gauge amplitudes provide the information needed to construct multi - loop gravity amplitudes . the original gravity - gauge tree amplitude relations were found by kawai , lewellen and tye @xcite , who recognized that the world - sheet integrands needed to compute tree - level amplitudes in the closed type ii superstring theory were essentially the square of the integrands appearing in the open - superstring tree amplitudes . klt represented the closed - string world - sheet integrals over the complex plane as products of contour integrals , and then deformed the contours until they could be identified as integrals for open - string amplitudes , thus deriving relations between closed- and open - string tree amplitudes . because the low - energy limit of the perturbative sector of the closed type ii superstring in @xmath37 is @xmath0 supergravity , and that of the open superstring is @xmath1 sym @xcite , as the string tension goes to infinity the klt relations express any @xmath0 supergravity tree amplitude in terms of amplitudes in @xmath1 sym . more recently , there have been a variety of studies of `` klt - type '' relations from various perspectives @xcite . one set of relations , found by three of the present authors @xcite , follows from @xcite a color - kinematic duality satisfied by gauge theory amplitudes . these relations promise to greatly simplify future computations of @xmath0 supergravity loop amplitudes @xcite . however , in this article we will only describe the use of the klt relations , because those were employed in the two- , three- and four - loop supergravity computations reviewed here . the klt relations for @xmath0 supergravity amplitudes are bilinear in the @xmath1 sym amplitudes , for two complementary reasons : ( 1 ) integrals over the complex plane naturally break up into pairs of contour integrals , and ( 2 ) the @xmath0 supergravity fock space naturally factors into a product of `` left '' and `` right '' @xmath1 sym fock spaces , = [ ] _ l [ ] _ r . [ neq8from4 ] the @xmath63 massless states of @xmath0 supergravity are tabulated in the upper half of . each state can be associated with a unique pair of states from @xmath1 sym , which has 16 massless states ( excluding color degrees of freedom ) , tabulated in the lower half of the table . for example , the eight helicity @xmath64 gravitino states are products of helicity @xmath65 gluon and helicity @xmath66 gluino states in two possible ways : @xmath67 , @xmath68 , @xmath69 . @xmath70 & @xmath9 & @xmath71 & @xmath72 & @xmath73 & @xmath74 & @xmath75 & @xmath76 & @xmath77 & @xmath78 + [ 2pt ] # of states & @xmath76 & @xmath79 & @xmath80 & @xmath81 & @xmath82 & @xmath81 & @xmath80 & @xmath79 & @xmath76 + field & @xmath83 & @xmath84 & @xmath85 & @xmath86 & @xmath87 & @xmath88 & @xmath89 & @xmath90 & @xmath91 + [ 2pt ] + @xmath70 & & & @xmath72 & @xmath73 & @xmath74 & @xmath75 & @xmath76 & & + [ 2pt ] # of states & & & @xmath76 & @xmath92 & @xmath93 & @xmath92 & @xmath76 & & + field & & & @xmath94 & @xmath95 & @xmath96 & @xmath97 & @xmath98 & & + [ 2pt ] in the open string theory , color degrees of freedom for gluons appear as chan - paton factors , but these factors are not present in the closed string . hence the gauge theory amplitudes appearing in the klt relations are those from which the chan - paton factors have been stripped off , which are known in the qcd community as color - ordered subamplitudes ( see _ e.g. _ ref . @xcite for a review ) . the full color - dressed gauge - theory tree amplitude @xmath99 is given as a sum over permutations of the color - ordered subamplitudes @xmath100 , _ n^(\{k_i , a_i } ) = g^n-2 _ s_n / z_n ( t^a_(1 ) t^a_(2 ) t^a_(n ) ) a_n^((1 ) , ( 2 ) , , ( n ) ) , [ antreedef ] where @xmath101 is the gauge coupling , @xmath102 is an adjoint index , @xmath103 is a generator matrix in the fundamental representation of @xmath104 , the sum is over all @xmath105 inequivalent ( non - cyclic ) permutations @xmath106 of @xmath28 objects , and the argument @xmath107 of @xmath100 labels both the momentum @xmath108 and state information ( helicity @xmath109 , _ etc . _ ) . in the case of supergravity tree amplitudes , @xmath110 , only powers of the gravitational coupling @xmath111 have to be stripped off , where @xmath111 is related to newton s constant by @xmath112 . we define @xmath113 by _ n^(\{k_i } ) = ( ) ^n-2m_n^(1,2, ,n ) . [ mntreedef ] then the first few klt relations have the form , m_3^(1,2,3 ) & = & i a_3^(1,2,3 ) _ 3^(1,2,3 ) , [ kltthree ] + m_4^(1,2,3,4 ) & = & - i s_12 a_4^(1,2,3,4 ) _ 4^(1,2,4,3 ) , [ kltfour ] + m_5^(1,2,3,4,5 ) & = & i s_12 s_34 a_5^(1,2,3,4,5 ) _ 5^(2,1,4,3,5 ) + ( 2,3 ) , [ kltfive ] + m_6^(1,2,3,4,5,6 ) & = & - i s_12s_45 a_6^(1,2,3,4,5,6 ) + & & + & & + ( 2,3,4 ) , [ kltsix ] where @xmath114 , and `` @xmath115 '' indicates a sum over the @xmath116 permutations of the @xmath117 arguments of @xmath118 . here @xmath100 indicates a tree amplitude for which the external states are drawn from the left - moving fock space @xmath119_l$ ] in the tensor product ( [ neq8from4 ] ) , while @xmath120 denotes an amplitude from the right - moving copy @xmath119_r$ ] . [ unitarityfigure ] 5.2 truein the scattering matrix is a unitary operator between in and out states : @xmath121 , or in terms of the more standard `` off - forward '' scattering matrix , @xmath122 , 2 t = t^t , [ tunit ] where disc@xmath123 . this simple relation generates the well - known unitarity relations , or cutting rules @xcite , for the discontinuities ( or absorptive parts ) of perturbative amplitudes . if one inserts a perturbative expansion for @xmath124 into , say t_4 & = & g^2 t_4^+ g^4 t_4^+ g^6 t_4^+ , [ t4 ] + t_5 & = & g^3 t_5^+ g^5 t_5^+ g^7 t_5^+ , [ t5 ] for the four- and five - point amplitudes , then one obtains the unitarity relations shown in . at order @xmath125 , the discontinuity in the one - loop four - point amplitude is given by the product of two order @xmath126 four - point tree amplitudes . the product must be summed over all possible intermediate states crossing the cut ( indicated by the dashed line in ) , and integrated over all possible intermediate momenta . at two loops , or order @xmath127 , there are two possible types of cuts : the product of a tree - level and a one - loop four - point amplitude ( @xmath128 ) , and the product of two tree - level five - point amplitudes ( @xmath129 ) . to get the complete scattering amplitude , not just the absorptive part , one could try to reconstruct the real part via a dispersion relation . however , in the context of perturbation theory , an easier method is available , because one knows that the amplitude could have been calculated in terms of feynman diagrams . therefore it can be expressed as a linear combination of appropriate feynman integrals , with coefficients that are rational functions of the kinematic variables . the unitarity method @xcite matches the information coming from the cuts against the set of available loop integrals in order to determine these rational coefficients . using unitarity in @xmath130 dimensions @xcite , one can also determine the so - called `` rational terms '' , which have no cuts in @xmath37 . _ generalized unitarity _ @xcite consists of imposing more than the minimal number of cut lines . it often simplifies enormously the information required to compute many terms in the amplitude @xcite , especially in highly supersymmetric theories @xcite . provides an example of generalized unitarity at the multi - loop level . one starts with an ordinary three - particle cut for a three - loop four - point amplitude . the information in this cut can be extracted more easily by cutting the one - loop five - point amplitude on the right - hand side of the cut , decomposing it into the product of a four - point tree and a five - point tree , in three inequivalent ways . 5.2 truein illustrates a particular class of generalized unitarity cuts , in which all cut momenta are allowed to be real . it is possible , however , to impose more and more on - shell constraints on intermediate legs , dissolving the amplitude into products of more tree amplitudes , each with fewer legs ( and hence simpler ) . for four - point amplitudes , the maximal cuts are the limiting cases in which all tree amplitudes are three - point ones , which can be dissolved no further . shows how one of the real - momentum configurations in generates several maximal cuts ( which contain complex momenta ) . the _ method of maximal cuts _ @xcite for constructing a multi - loop amplitude begins with the evaluation of the maximal cuts , and the construction of a candidate ansatz for the loop - momentum integrand that is consistent with them . for simplicity we will discuss here the evaluation of four - dimensional cuts , that is , cuts in which the cut loop momenta are taken to be in four dimensions . for complete generality the cut loop momenta should be in @xmath51 dimensions . however , for the four - point amplitudes in maximally supersymmetric gauge theory or gravity , the @xmath51-dimensional cuts have yet to reveal any new terms , beyond those found using the four - dimensional cuts @xcite . for real momenta , the kinematics of the three - point process with all massless legs is singular all three momenta must be parallel . however , for complex momenta it is perfectly nonsingular @xcite . the maximal cuts for four - point amplitudes are enumerated simply by drawing all cubic graphs . their evaluation is also very simple , for four - dimensional cuts , because three - point tree amplitudes are always given by a simple expression in the usual spinor products , in either @xmath131 or @xmath132 , where @xmath133 ( @xmath134 ) is the two - component positive - chirality ( negative - chirality ) spinor associated with the massless momentum @xmath108 . for example , for three gluons there are only two non - vanishing amplitudes , a_3^(1 ^ -,2 ^ -,3^+ ) = i 1.2 ^ 41.22.33.1 , a_3^(1^+,2^+,3 ^ - ) = -i 1.2 ^ 41.22.33.1 . [ mmpppm ] there are two types of three - point complex kinematics ; for each type , one of the two amplitudes in is non - vanishing and the other one vanishes @xcite . three - point amplitudes for gravity can be obtained directly as products of two gauge amplitudes , using . 5.2 truein even though the maximal cuts are very simple to evaluate analytically , they provide a great deal of information , and an ansatz that satisfies the maximal cuts is an excellent starting point for constructing the full answer . for example , for the contributions to four - gluon scattering in @xmath1 sym that are planar ( the dominant terms in the large @xmath135 limit ) , the maximal cuts find all terms present in the amplitude at one , two and three loops . they only start to miss planar terms at four loops ( and non - planar terms at three loops ) . the remaining terms , whether planar or non - planar , can be found systematically by collapsing one propagator in each maximal cut to generate the next - to - maximal cuts ; one more propagator to generate the next - to - next - to - maximal cuts ; and so on . at each stage the ansatz is improved by adding more terms in order to fit the new information . each additional term should contain at least one power of an inverse ( collapsed ) propagator @xmath136 , corresponding to the fact that it was invisible on the maximal cut ( @xmath137 ) , and only became visible on the next - to - maximal cut ( @xmath138 ) . the process of amplitude construction terminates when no more terms need to be added . then the amplitude can be checked , by a comparison ( usually numerical ) against a complete , or `` spanning '' @xcite , set of unitarity cuts . the general strategy @xcite we have adopted for computing multi - loop @xmath0 supergravity amplitudes is to first compute the loop - momentum integrands for the corresponding amplitudes in @xmath1 sym . the integrands are described by a sum of feynman integrals for cubic graphs , with standard scalar propagator factors and additional numerator polynomials . in the four - point case , the @xmath139 such integral has the form , @xmath140 where @xmath141 , @xmath142 , are the three independent external momenta , @xmath143 are the @xmath11 independent loop momenta , and @xmath144 are the momenta of the @xmath145 propagators ( internal lines of the graph @xmath13 ) , which are linear combinations of the @xmath143 and the @xmath141 . as usual , @xmath146 is the @xmath51-dimensional measure for the @xmath147 loop momentum . the numerator polynomial @xmath148 is a polynomial in both internal and external momenta . the color factor @xmath149 can be written as a product of structure constants @xmath150 for the gauge group . it can also be written diagrammatically , using three - vertices for @xmath150 factors , and lines ( propagators ) for @xmath151 contractions . in this form , it is given just by the associated cubic graph . these integrands can then be cut in any desired fashion . through the klt relations , they provide the data needed to evaluate very efficiently the generalized cuts for @xmath0 supergravity . in particular , the @xmath0 supergravity cuts require a sum over the 256 states in the @xmath0 supergravity multiplet , for every cut line . however , the corresponding cut @xmath1 sym loop integrands already contain a sum over the 16 states in the @xmath1 sym multiplet . the klt relations express the @xmath0 supergravity cuts as sums of products of two copies of @xmath1 sym cuts . the @xmath0 sum factorizes as , _ = _ [ ] _ l _ [ ] _ r , and the @xmath1 sums have already been carried out in the course of constructing the @xmath1 sym integrand . because gravity has no notion of color , planar and non - planar contributions can not be separated in graviton amplitudes . the klt relations therefore must relate the gravity cuts to both planar and non - planar gauge theory cuts . in other words , the complete @xmath1 sym amplitude , both planar ( large @xmath135 ) and non - planar terms , is required in this method . 5.2 truein sketches how the method works for a particular generalized cut at three loops . the @xmath0 supergravity cut contains one four - point tree amplitude and two five - point ones . we use the klt relations ( [ kltfour ] ) and ( [ kltfive ] ) . we relabel them , and use the fact that @xmath152 is totally symmetric in legs 1,2,3,4 to rewrite them as , m_4^(_1,_2,_3,_4 ) & = & - i s__1_2s__2_3s__1_3 a_4^(_1,_2,_3,_4 ) _ 4^(_1,_2,_3,_4 ) , [ kltnew ] + m_5^(1,2,_2,_1,_5 ) & = & -i s__5 1 s_2_2 a_5^(1,2,_2,_1,_5 ) _ 5^(1,_1,2,_2,_5 ) + ( 1,2 ) , + m_5^(4,3,_3,_4,_5 ) & = & -i s__5 4 s_3_3 a_5^(4,3,_3,_4,_5 ) _ 5^(4,_4,3,_3,_5 ) + ( 3,4 ) . in this way , both occurrences of the four - point @xmath1 sym amplitude carry the same cyclic ordering as the @xmath0 supergravity one , as shown in the figure . one of the two five - point amplitudes carries the same ordering , as shown in the left copy . this copy can be evaluated using the planar @xmath1 sym amplitude . the other five - point amplitude is twisted , leading to the right copy , which is non - planar , so it requires non - planar terms in the @xmath1 sym amplitude . a reflection symmetry under the permutation @xmath153 is preserved by this representation . the two - fold permutation sum in @xmath154 in leads to a four - fold permutation sum in the figure ; one must add the permutations @xmath155 , @xmath156 , and @xmath157 . note that for terms that are detected in the maximal cuts , because of the simple relation between gravity and gauge three - point amplitudes ( ) , the numerator factors are always simply squared in passing from gauge theory to gravity . 4.7 truein the full two - loop four - point amplitude in @xmath1 sym is given by @xcite a_4^&= & - s_12 s_23 a_4^ , [ twoloopym ] where @xmath158 are the scalar planar and non - planar double box integrals shown in , and @xmath159 are color factors constructed from structure constant vertices , with the same graphical structure as the corresponding integral . the quantity @xmath160 is totally symmetric under gluon interchange , and its square is the @xmath21 matrix element in , up to a factor of @xmath107 . because all terms in are detected by the maximal cuts , the complete two - loop four - point amplitude in @xmath0 supergravity is found simply by squaring the prefactors in ( and removing the color factors , as appropriate for gravity ) : m_4^&= & -i ( s_12 s_23 a_4^)^2 + & = & s_12 s_23 s_13 m_4^. + [ twoloopgr ] because the loop integrals appearing in the two amplitudes , , are precisely the same , the critical dimension @xmath161 is automatically the same for both theories at two loops . this value is @xmath162 , the dimension in which the two - loop , seven - propagator integrals , @xmath163 , are log divergent , in agreement with . the two - loop @xmath0 supergravity divergence is associated with a counterterm of the form @xmath41 in @xmath164 . this type of counterterm is permitted by the field - theoretic duality constraints of ref . @xcite . at three loops , the integrand of the @xmath1 sym four - point amplitude begins to have dependence on the loop - momentum in its numerator , as well as ( non - planar ) terms that can not be detected in the maximal cuts . for this reason , the three - loop @xmath0 supergravity amplitude , in its initial two forms @xcite , was not given by simply squaring the @xmath1 sym results except for a subset of the graphs that could be inferred using only two - particle cuts . more recently , three of the present authors rearranged the three - loop @xmath1 sym amplitude so as to make manifest its color - kinematic duality @xcite . in this form the @xmath0 supergravity amplitude can once again be found by a simple squaring procedure . here we will give the amplitudes in the form found in ref . @xcite , which requires only the nine cubic graphs shown in . ( three more cubic graphs , containing three - point subdiagrams , enter the solution in ref . @xcite . ) 5.5 truein both the @xmath1 sym and @xmath0 supergravity amplitudes are described by giving the loop - momentum numerator polynomials @xmath165 for these graphs . in addition , the @xmath1 sym graphs are multiplied by the corresponding color structure , as in . . the numerator factors @xmath165 for the integrals @xmath166 in for @xmath1 super - yang - mills theory . the first column labels the integral , the second column the relative numerator factor . an overall factor of @xmath160 has been removed . the invariants @xmath167 and @xmath168 are defined in . [ cols="^,^",options="header " , ] .4 cm gives the values of @xmath165 for @xmath0 supergravity , in a form @xcite which is also manifestly quadratic in the loop momenta . ( in the first version of the amplitude @xcite , the quadratic nature was not yet manifest . ) comparing the two sets of numerators , we see that the @xmath0 supergravity ones are the squares of the @xmath1 sym ones , up to contact terms , as expected from the klt relations . for example , in graphs ( e)(g ) , @xmath169 , so @xmath170 ^ 2 $ ] ( modulo @xmath136 terms ) . because the numerator factors for both @xmath0 supergravity and @xmath1 sym are manifestly quadratic in the loop momenta , the critical dimensions @xmath53 at three loops remain equal , @xmath171 for @xmath172 . indeed , when the ultraviolet poles in the integrals for @xmath0 supergravity are evaluated , no further cancellation is found , and the resulting pole is m_4^ , d=6 - 2 |_pole = 5 _ 3(4)^9 ( s_12 s_23 s_13)^2 m_4^ , .3 cm [ threeloopd6div ] corresponding to a counterterm of the form @xmath42 in @xmath173 . again , the existence of this counterterm is consistent with the field - theoretic duality constraints of ref . @xcite . the form of the divergence ( [ threeloopd6div ] ) was reproduced from string - theoretic duality arguments in ref . @xcite ; however , the rational number predicted there does not agree with . whether or not this indicates an issue in decoupling massive states from string theory to obtain @xmath0 supergravity @xcite remains unclear . 5.5 truein at four loops , the same general strategy still works , but the bookkeeping issues are greater @xcite . one can start by classifying the cubic vacuum graphs . at three loops there were only two ; at four loops there are five , shown in . the next step is to decorate the five vacuum graphs with four external legs to get the cubic four - point graphs . as at lower loops , graphs containing triangles ( three propagators or fewer on a loop ) or other three point subgraphs can be dropped . ( this statement would not be true for representations obeying the color - kinematic duality , as at three loops @xcite . ) ( a ) only gives rise to triangle - containing graphs , so it can be dropped . altogether there are 50 cubic four - point graphs with nonvanishing numerators . graphs ( b ) and ( c ) do generate four - point graphs without triangles , but the numerators for all such graphs can be determined , up to possible contact terms , by iterated two - particle cuts . because of the structure of these cuts @xcite , the associated numerator polynomials turn out to be very simple . graphs ( d ) , and particularly ( e ) , give rise to the most complex numerators . 5.6 truein the method of maximal cuts was used to determine the numerator polynomials for @xmath1 sym . at four loops , the maximal cuts have 13 cut conditions @xmath137 . then near - maximal cuts with only 12 cut conditions are considered , followed by ones with 11 cut conditions . at this point the @xmath1 sym ansatz is complete ; no more terms need to be added . the result was verified by comparison against a spanning set of generalized cuts . in we show three of the 50 numerator polynomials . these three are associated with the one non - planar cubic vacuum graph ( e ) , and they have the most complex numerators . integral ( 50 ) is required for the ansatz for the integrand to match various cuts . however , it integrates to zero and has vanishing color factor , so it does not contribute to the @xmath1 sym amplitude . in constructing the amplitude , it proved very useful to have simple pictorial rules that allow one to generate numerator polynomials for many graphs from those for other graphs , either at the same loop order or at lower loop order . an old rule @xcite , called the _ rung rule _ , applies whenever a graph has a two - particle cut . a newer rule is the _ box cut rule _ it can be applied to any graph that contains a four - point subdiagram , and it generates that graph s numerator polynomial ( modulo certain contact terms ) from the polynomials associated with particular lower - loop graphs . together , these rules are quite powerful ; of the 50 graphs , only four have neither two - particle cuts nor box cuts . ( three of the four appear in . ) after the @xmath1 sym amplitude was computed , the 50 numerator polynomials for the @xmath0 supergravity amplitude were then constructed , using information provided by the klt relations . the results are quite lengthy , but are provided as mathematica readable files in ref . @xcite , along with some tools for manipulating them . from the numerator polynomials for the @xmath0 supergravity amplitude , the amplitude s ultraviolet behavior could be extracted , by expanding the integrals in the limit of small external momenta , relative to the loop momenta @xcite . unlike the three - loop representations in refs . @xcite , the ultraviolet behavior for the form in ref . @xcite is not manifest . that means that each integral is more divergent than the sum , and hence subleading terms in the expansion are required . it is necessary to expand to third order , in order to show that @xmath0 supergravity is as well behaved as @xmath1 sym at four loops , in this representation of the amplitude . more concretely , the numerator polynomials , omitting an overall factor of @xmath174 , have a mass dimension of 12 , _ i.e. _ each term is of the form @xmath175 , where @xmath33 and @xmath176 stand respectively for external and loop momenta . the maximum value of @xmath117 turns out to be 8 for every integral . the integrals all have 13 propagators , so they have the form @xmath177 . the amplitude is manifestly finite in @xmath37 , because @xmath178 . ( this result is not unexpected , given the absence of a @xmath43 counterterm @xcite . ) the amplitude is not manifestly finite in @xmath179 ; to see that requires cancellation of the @xmath180 , @xmath181 and @xmath182 terms , after expansion around small @xmath33 . the cancellation of the @xmath180 terms is relatively simple , because one can simply set the external momenta @xmath108 to zero inside the integrals that appear . at this point , the potentially divergent integrals all reduce to one of two types of scalar vacuum integrals there are no loop - momentum tensors appearing in the numerator , and no doubled propagator factors in the denominator . in fact , only two of the five vacuum graphs in appear , ( d ) and ( e ) . collecting all terms , one finds that the coefficients of ( d ) and ( e ) both vanish . the cancellation of the @xmath181 terms ( and the @xmath183 terms ) is trivial : using dimensional regularization , with no dimensionful parameter , lorentz invariance does not allow an odd - power divergence . the most intricate cancellation is that of the @xmath182 terms , corresponding to the vanishing of the coefficient of the potential counterterm @xmath42 in @xmath179 . in the expansion of the integrals to the second subleading order as @xmath184 , thirty different four - loop vacuum integrals are generated . these integrals often have doubled ( and sometimes tripled ) propagators , arising from the taylor expansion of the loop - momentum integrand in the external momentum . some integrals also contain tensors in the loop - momentum in their numerators . however , there are consistency relations between the integrals , corresponding to the ability to shift the loop momenta by external momenta before expanding around @xmath185 . these consistency relations are powerful enough to imply the cancellation of the ultraviolet pole in @xmath186 . as a check , we evaluated all 30 ultraviolet poles directly , with the same conclusion . we did not yet evaluate the ultraviolet pole near @xmath187 ( the critical dimension for @xmath1 sym at this loop order ) , so in principle it could cancel , although that seems unlikely to be the case . in summary , the four - loop four - point amplitude of @xmath0 supergravity is ultraviolet finite for @xmath188 @xcite , the same bound found for @xmath1 super - yang - mills theory . finiteness in @xmath189 is a consequence of nontrivial cancellations , beyond those already found at three loops @xcite . these results provide the strongest direct support to date for the possibility that @xmath0 supergravity might be a perturbatively finite quantum theory of gravity . in every explicit computation to date , through four loops , the ultraviolet behavior of @xmath0 supergravity has proven to be no worse than that of @xmath1 super - yang - mills theory . on the other hand , there are several recent arguments @xcite in favor of the existence of a seven - loop counterterm @xcite of the form @xmath50 . as argued in , the five - loop four - graviton scattering amplitude , when evaluated in higher dimensions for the loop momentum , should provide a fairly decisive test for what will happen at seven loops . although this computation is difficult , it may well prove feasible using new ideas related to the color - kinematic duality @xcite . suppose that @xmath0 supergravity turns out to be finite to all orders in perturbation theory . this result still would not prove that it is a consistent theory of quantum gravity at the non - perturbative level . there are at least two reasons to think that it might need a non - perturbative ultraviolet completion : 1 . the ( likely ) @xmath190 or worse growth of the coefficients of the order @xmath11 terms in the perturbative expansion , which for fixed - angle scattering , would imply a non - convergent behavior @xmath191 . 2 . the fact that the perturbative series seems to be @xmath4 invariant , while the mass spectrum of black holes is non - invariant ( see _ e.g. _ ref . @xcite for recent discussions ) . qed is an example of a perturbatively well - defined theory that needs an ultraviolet completion ; it also has factorial growth in its perturbative coefficients , @xmath192 , due to ultraviolet renormalons associated with the landau pole . yet for small values of @xmath193 qed works extremely well : it predicts the anomalous magnetic moment of the electron to 10 digits of accuracy . also , there are many pointlike non - perturbative ultraviolet completions for qed , namely asymptotically free grand unified theories . are there any imaginable pointlike completions for @xmath0 supergravity ? maybe the only completion is string theory ; or maybe this can not happen because of the impossibility of decoupling non - perturbative string states not present in @xmath0 supergravity @xcite . another question is whether @xmath0 supergravity might point the way to other , more realistic finite ( or well behaved ) theories of quantum gravity , having less supersymmetry and ( perhaps ) chiral fermions . one step in this direction could be to examine the multi - loop behavior of theories that can be thought of as spontaneously broken gauged @xmath0 supergravity @xcite , which are known to have improved ultraviolet behavior at one loop @xcite . in any event , the excellent perturbative ultraviolet behavior of @xmath0 supergravity has already provided many surprises . although the theory may not itself be of direct phenomenological interest , perhaps it will some day lead to more realistic theories also having excellent ultraviolet behavior . as a `` toy model '' for a pointlike theory of quantum gravity , it has been extremely instructive , and further exploration will no doubt be fruitful as well . -.3 cm l.d . thanks the organizers of the xvi@xmath194 european workshop on string theory in madrid for the opportunity to present this work , and g. dallagata , s. ferrara , and f. zwirner for useful conversations . this research was supported by the us department of energy under contracts de ac0276sf00515 , de fg0391er40662 and de - fg02 - 90er40577 ( oji ) , by the us national science foundation under grant phy-0855356 , and by the alfred p. sloan foundation . j.j.m.c . gratefully acknowledges the stanford institute for theoretical physics for financial support . h.j.s research is supported by the european research council under advanced investigator grant erc - adg-228301 . the figures were generated using jaxodraw @xcite , based on axodraw @xcite . b. de wit and d. z. freedman , nucl . b * 130 * , 105 ( 1977 ) . e. cremmer , b. julia and j. scherk , phys . b * 76 * , 409 ( 1978 ) . e. cremmer and b. julia , phys . b * 80 * , 48 ( 1978 ) ; nucl . b * 159 * , 141 ( 1979 ) . h. kawai , d. c. lewellen and s. h. h. tye , nucl . b * 269 * , 1 ( 1986 ) . z. bern , j. j. m. carrasco and h. johansson , phys . d * 78 * , 085011 ( 2008 ) [ 0805.3993 [ hep - ph ] ] . r. j. eden , p. v. landshoff , d. i. olive and j. c. polkinghorne , _ the analytic s matrix _ ( cambridge university press , 1966 ) . z. bern , l. j. dixon and d. a. kosower , nucl . b * 513 * , 3 ( 1998 ) [ hep - ph/9708239 ] . z. bern , l. j. dixon and d. a. kosower , jhep * 0001 * , 027 ( 2000 ) [ hep - ph/0001001 ] ; + jhep * 0408 * , 012 ( 2004 ) [ hep - ph/0404293 ] . z. bern , v. del duca , l. j. dixon and d. a. kosower , phys . d * 71 * , 045006 ( 2005 ) [ hep - th/0410224 ] . r. britto , f. cachazo and b. feng , nucl . b * 725 * , 275 ( 2005 ) [ hep - th/0412103 ] . z. bern , l. j. dixon , d. c. dunbar , m. perelstein and j. s. rozowsky , nucl . b * 530 * , 401 ( 1998 ) [ hep - th/9802162 ] . z. bern , j. j. carrasco , l. j. dixon , h. johansson , d. a. kosower and r. roiban , phys . * 98 * , 161303 ( 2007 ) [ hep - th/0702112 ] . z. bern , j. j. m. carrasco , l. j. dixon , h. johansson and r. roiban , phys . rev . d * 78 * , 105019 ( 2008 ) [ 0808.4112 [ hep - th ] ] . z. bern , j. j. carrasco , l. j. dixon , h. johansson and r. roiban , phys . lett . * 103 * , 081301 ( 2009 ) [ 0905.2326 [ hep - th ] ] . z. bern , j. j. carrasco , l. j. dixon , h. johansson and r. roiban , phys . * d82 * , 125040 ( 2010 ) [ 1008.3327 [ hep - th ] ] . s. weinberg , in _ understanding the fundamental constituents of matter _ a. zichichi ( plenum press , new york , 1977 ) ; s. weinberg , in _ general relativity _ , eds . s. w. hawking and w. israel ( cambridge university press , 1979 ) , p. 700 ; + m. niedermaier and m. reuter , living rev . rel . * 9 * , 5 ( 2006 ) . p. hoava , phys . rev . d * 79 * , 084008 ( 2009 ) [ 0901.3775 [ hep - th ] ] . g. t hooft and m. j. g. veltman , annales poincare phys . a * 20 * , 69 ( 1974 ) . m. t. grisaru , phys . b * 66 * , 75 ( 1977 ) . s. deser , j. h. kay and k. s. stelle , phys . lett . * 38 * , 527 ( 1977 ) . e. tomboulis , phys . b * 67 * , 417 ( 1977 ) . m. t. grisaru , h. n. pendleton and p. van nieuwenhuizen , phys . d * 15 * , 996 ( 1977 ) ; + m. t. grisaru and h. n. pendleton , nucl . b * 124 * , 81 ( 1977 ) . p. van nieuwenhuizen and j. a. m. vermaseren , phys . b * 65 * , 263 ( 1976 ) . s. ferrara and b. zumino , nucl . b * 134 * , 301 ( 1978 ) . s. deser and j. h. kay , phys . b * 76 * , 400 ( 1978 ) . p. s. howe and u. lindstrm , nucl . b * 181 * , 487 ( 1981 ) . r. e. kallosh , phys . b * 99 * , 122 ( 1981 ) . d. j. gross and e. witten , nucl . b * 277 * , 1 ( 1986 ) . h. elvang , d. z. freedman and m. kiermaier , jhep * 1011 * , 016 ( 2010 ) [ 1003.5018 [ hep - th ] ] . j. m. drummond , p. j. heslop , p. s. howe and s. f. kerstan , jhep * 0308 * , 016 ( 2003 ) [ hep - th/0305202 ] . r. kallosh , jhep * 0909 * , 116 ( 2009 ) [ 0906.3495 [ hep - th ] ] . m. bianchi , h. elvang and d. z. freedman , jhep * 0809 * , 063 ( 2008 ) [ 0805.0757 [ hep - th ] ] . n. arkani - hamed , f. cachazo and j. kaplan , jhep * 1009 * , 016 ( 2010 ) [ 0808.1446 [ hep - th ] ] . r. kallosh and t. kugo , jhep * 0901 * , 072 ( 2009 ) [ 0811.3414 [ hep - th ] ] . n. marcus , phys . b * 157 * , 383 ( 1985 ) . d. z. freedman and e. tonni , 1101.1672 [ hep - th ] . g. bossard , p. s. howe and k. s. stelle , jhep * 1101 * , 020 ( 2011 ) [ 1009.0743 [ hep - th ] ] . g. bossard , p. s. howe , k. s. stelle , phys . b * 682 * , 137 ( 2009 ) [ 0908.3883 [ hep - th ] ] . r. kallosh , jhep * 1012 * , 009 ( 2010 ) [ 1009.1135 [ hep - th ] ] . s. mandelstam , nucl . b * 213 * , 149 ( 1983 ) ; + p. s. howe , k. s. stelle and p. k. townsend , nucl . b * 214 * , 519 ( 1983 ) ; + l. brink , o. lindgren and b. e. w. nilsson , phys . b * 123 * , 323 ( 1983 ) . j. bjrnsson and m. b. green , jhep * 1008 * , 132 ( 2010 ) [ 1004.2692 [ hep - th ] ] . j. bjrnsson , jhep * 1101 * , 002 ( 2011 ) [ 1009.5906 [ hep - th ] ] . m. b. green , j. h. schwarz and l. brink , nucl . b * 198 * , 474 ( 1982 ) . j. m. drummond , m. spradlin , a. volovich and c. wen , phys . d * 79 * , 105018 ( 2009 ) [ 0901.2363 [ hep - th ] ] ; + n. e. j. bjerrum - bohr , p. h. damgaard , b. feng and t. sondergaard , phys . rev . d * 82 * , 107702 ( 2010 ) [ 1005.4367 [ hep - th ] ; jhep * 1009 * , 067 ( 2010 ) [ 1007.3111 [ hep - th ] ] ; + b. feng and s. he , jhep * 1009 * , 043 ( 2010 ) [ 1007.0055 [ hep - th ] ] ; + b. feng , s. he , r. huang and y. jia , jhep * 1010 * , 109 ( 2010 ) [ 1008.1626 [ hep - th ] ] ; + n. e. j. bjerrum - bohr , p. h. damgaard , t. sondergaard and p. vanhove , jhep * 1101 * , 001 ( 2011 ) [ 1010.3933 [ hep - th ] ] . z. bern , t. dennen , y. t. huang and m. kiermaier , phys . d * 82 * , 065003 ( 2010 ) [ 1004.0693 [ hep - th ] ] z. bern , j. j. m. carrasco and h. johansson , phys . lett . * 105 * , 061602 ( 2010 ) [ 1004.0476 [ hep - th ] ] . j. j. m. carrasco and h. johansson , to appear . l. j. dixon , in _ qcd & beyond : proceedings of tasi 95 _ , ed . d. e. soper ( world scientific , 1996 ) [ hep - ph/9601359 ] . l. d. landau , nucl . * 13 * , 181 ( 1959 ) ; + s. mandelstam , phys . rev . * 115 * , 1741 ( 1959 ) ; + r. e. cutkosky , j. math . * 1 * , 429 ( 1960 ) . z. bern , l. j. dixon , d. c. dunbar and d. a. kosower , nucl . b * 425 * , 217 ( 1994 ) [ hep - ph/9403226 ] ; nucl . b * 435 * , 59 ( 1995 ) [ hep - ph/9409265 ] . z. bern and a. g. morgan , nucl . b * 467 * , 479 ( 1996 ) [ hep - ph/9511336 ] ; + z. bern , l. j. dixon , d. c. dunbar and d. a. kosower , phys . b * 394 * , 105 ( 1997 ) [ hep - th/9611127 ] . z. bern , m. czakon , l. j. dixon , d. a. kosower and v. a. smirnov , phys . d * 75 * , 085010 ( 2007 ) [ hep - th/0610248 ] . z. bern , j. j. m. carrasco , h. johansson and d. a. kosower , phys . d * 76 * , 125020 ( 2007 ) [ 0705.1864 [ hep - th ] ] . e. i. buchbinder and f. cachazo , jhep * 0511 * , 036 ( 2005 ) [ hep - th/0506126 ] ; + f. cachazo and d. skinner , 0801.4574 [ hep - th ] ; + f. cachazo , 0803.1988 [ hep - th ] ; + f. cachazo , m. spradlin and a. volovich , phys . d * 78 * , 105022 ( 2008 ) [ 0805.4832 [ hep - th ] ] ; + m. spradlin , a. volovich and c. wen , phys . 085025 ( 2008 ) [ 0808.1054 [ hep - th ] ] . m. h. goroff and a. sagnotti , phys . b * 160 * , 81 ( 1985 ) ; nucl . b * 266 * , 709 ( 1986 ) . e. witten , commun . * 252 * , 189 ( 2004 ) [ hep - th/0312171 ] . z. bern , j. s. rozowsky and b. yan , phys . b * 401 * , 273 ( 1997 ) [ hep - ph/9702424 ] . m. b. green , j. g. russo and p. vanhove , jhep * 1006 * , 075 ( 2010 ) [ 1002.3805 [ hep - th ] ] . m. b. green , h. ooguri and j. h. schwarz , phys . * 99 * , 041601 ( 2007 ) [ 0704.0777 [ hep - th ] ] . a. a. vladimirov , theor . phys . * 43 * , 417 ( 1980 ) [ teor . mat . fiz . * 43 * , 210 ( 1980 ) ] ; + n. marcus and a. sagnotti , nuovo cim . a * 87 * , 1 ( 1985 ) . m. bianchi , s. ferrara and r. kallosh , phys . b * 690 * , 328 ( 2010 ) [ 0910.3674 [ hep - th ] ] ; jhep * 1003 * , 081 ( 2010 ) [ 0912.0057 [ hep - th ] ] . s. ferrara and b. zumino , phys . b * 86 * , 279 ( 1979 ) . d. binosi and l. theussl , comput . commun . * 161 * , 76 ( 2004 ) [ hep - ph/0309015 ] ; + d. binosi , j. collins , c. kaufhold and l. theussl , comput . commun . * 180 * , 1709 ( 2009 ) [ 0811.4113 [ hep - ph ] ] . j. a. m. vermaseren , comput . commun . * 83 * , 45 ( 1994 ) .

we use strong - coupling perturbation theory , the variational cluster approach ( vca ) , and the dynamical density - matrix renormalization group ( ddmrg ) method to investigate static and dynamical properties of the one - dimensional bose hubbard model in both the mott - insulating and superfluid phases . from the von neumann entanglement entropy we determine the central charge and the transition points for the first two mott lobes . our dmrg results for the ground - state energy , momentum distribution function , boson correlation function decay , mott gap , and single particle - spectral function are reproduced very well by the strong - coupling expansion to fifth order , and by vca with clusters up to 12 sites as long as the ratio between the hopping amplitude and on - site repulsion , , is smaller than 0.15 and 0.25 , respectively . in addition , in the superfluid phase vca captures well the ground - state energy and the sound velocity of the linear phonon modes . this comparison provides an authoritative estimate for the range of applicability of these methods . in strong - coupling theory for the mott phase , the dynamical structure factor is obtained from the solution of an effective single - particle problem with an attractive potential . the resulting resonances show up as double - peak structure close to the brillouin zone boundary . these high - energy features also appear in the superfluid phase which is characterized by a pronounced phonon mode at small momenta and energies , as predicted by bogoliubov and field theory . in one dimension , there are no traces of an amplitude mode in the dynamical single - particle and two - particle correlation functions .

1203.1120

the ability to place ultracold bosonic atoms in optical lattices offered new prospects in the study of quantum many - particle systems @xcite , mainly because , in contrast to solid - state realizations , the properties of the system can be manipulated in a very controlled way by tuning the particle density , the lattice depth , the trapping potential and the interactions between the particles @xcite . likewise , the spatial dimension and coordination number of the optical lattice , the degree of disorder , or the coupling strength to external fields might be changed @xcite . hence , in these experiments , specific lattice hamiltonians can be engineered and analyzed , including quantum phase transitions between gapped and itinerant phases . a prominent example is the transition between mott insulating ( mi ) and superfluid ( sf ) phases which results from the competition between the particles kinetic energy and their mutual on - site repulsion . in this way , subtle quantum correlation effects become observable on a macroscopic scale . hubbard hamiltonian captures the essential physics of interacting bosons in optical lattices @xcite . the ground - state phase diagram of this model in two and three dimensions has been determined by analytical , perturbative methods @xcite and numerical , quantum monte - carlo techniques @xcite . the one - dimensional ( 1d ) case , which can be realized experimentally @xcite , is also accessible by qmc @xcite , and particularly rewarding to study because the physics in 1d normally is rather peculiar @xcite . on a linear chain with @xmath1 sites and periodic boundary conditions ( pbc ) the bose hubbard hamiltonian reads @xmath2 here , @xmath3 , @xmath4 , and @xmath5 are the boson creation , annihilation and particle number operators on site @xmath6 . the grand - canonical hamiltonian is given by @xmath7 where @xmath8 is the thermodynamic chemical potential and @xmath9 counts the total number of particles . for @xmath10 particles ( atoms ) in the system , the ( global ) filling factor is @xmath11 . in eq . , the hopping of the bosons between neighboring sites is characterized by the tunneling amplitude @xmath12 , while @xmath13 is the on - site interaction which we choose to be repulsive , @xmath14 ; recently , ngerl et al . investigated an unstable crystal of bosons with @xmath15 @xcite . accordingly , the physics of the bose hubbard model is governed by the ratio between kinetic energy and interaction energy , @xmath16 . if , for given chemical potential @xmath8 , @xmath17 is larger than a critical value the bosons are `` superfluid '' . below @xmath18 , the system becomes mott insulating , characterized by an integer filling factor @xmath19 . in experiments , @xmath17 can be varied over several orders of magnitude , by modifying the depth of the lattice through quantum optical techniques whereby sf and mi phases can be realized . from a theoretical point of view , the calculation of the boundaries between the sf and mi phases in the @xmath20 ground - state phase diagram is particularly demanding because quantum phase transitions in one dimension often are of kosterlitz thouless type @xcite with exponentially small mott gaps in the vicinity of the transition . and the electron transfer amplitude @xmath0 . the boundaries delimiting the first two mott lobes were determined by dmrg , using system sizes up to @xmath21 and obc @xcite , see text . [ fig : pd ] ] the numerical density - matrix renormalization group ( dmrg ) method @xcite is well suited to address one - dimensional interacting particle systems @xcite . in fact , the @xmath20 ground - state phase diagram of the 1d bose hubbard model has been obtained fairly accurately using this technique @xcite , see fig . [ fig : pd ] . since multiple occupancies pose serious technical problems , the maximal boson number per site in dmrg is constrained to be five . note that dmrg naturally works at fixed @xmath22 , i.e. , in the canonical ensemble . this leads to the definition of two chemical potentials for finite systems , @xmath23 @xcite where @xmath24 denotes the ground - state energy . in the mi state we have a finite gap , @xmath25 , whereas the chemical potential is continuous in the sf phase , @xmath26 . in one dimension , the delocalized sf state is not macroscopically occupied but rather characterized by an algebraic divergence of the momentum distribution @xcite . the localized mi state is incompressible , as usual , and characterized by an integer particle density and a gap in the single - particle spectrum @xcite . the regions in the @xmath20 phase diagram where the density @xmath19 is pinned to integer values are termed mott lobes . their special shape is conditioned by the strong phase fluctuations existing in a 1d system . close to the boundaries of the mott lobes , the mott gap is exponentially small . the precise position of the mott dips can be obtained from the tomonaga luttinger parameter @xcite . a detailed theoretical understanding of the bose hubbard model requires the calculations of ( dynamical ) correlation functions which poses a hard problem for which no exact solution exists . recall that the 1d bose hubbard model at @xmath27 ( soft - core bosons ) is not integrable . consequently , a large variety of approximative approaches were suggested and elaborated for the bose hubbard model and its variants during the last two decades ; for a recent review see ref . [ ] . in the sf phase , ( weakly ) interacting bosons at low energies are well described as a tomonaga luttinger liquid @xcite . however , close to the sf to mi transition , the precise character of the spectrum is still under debate . this particularly concerns the question whether or not a second , gapped mode besides the standard sound mode , as obtained from mean - field theory @xcite , can be seen in the single - particle spectral function or in the dynamical structure factor . in the mi phase , strong - coupling expansions in @xmath16 give reliable analytical results . the ground - state energy of all mott lobes was determined to second order by freericks and monien @xcite , and was improved up to order @xmath28 for the lowest mott lobe , @xmath29 , by damski and zakrzewski @xcite . they also provided the series expansion for the local particle - density fluctuations to order @xmath30 , a high - order series expansion for the single - particle density matrix @xmath31 for @xmath32 , and gave the corresponding expressions for the ground - state density - density correlation function @xmath33 for @xmath32 ; for results for @xmath34 and @xmath35 , respectively , see ref . the fourier transformation of @xmath36 provides the momentum distribution @xmath37 . the result for @xmath37 to third order in @xmath17 was re - derived by freericks et al . @xcite using a different method . in contrast to higher dimensions , @xmath38 , the convergence of the strong - coupling expansion series in 1d is rather questionable . these problems become apparent , e.g. , in the calculation of the critical value @xmath18 for the transition between the mott insulator and the superfluid phase . for example , the series expansion for the superfluid susceptibility constructed by eckardt et al . @xcite determines @xmath18 very accurately in @xmath38 but fails for @xmath39 where a reentrant superfluid phase is predicted @xcite . high - order expansions are also possible for the single - particle gap @xcite . the mott transition in one dimension is of kosterlitz thouless ( kt ) type so that the gap becomes exponentially small close to the transition , which can not be reproduced easily within a third - order strong - coupling expansion @xcite . in order to obtain a good approximation of the critical value for the transition , elstner and monien @xcite proposed a scaling analysis for the gap . based on this idea , freericks et al . @xcite used a ( 6,7)pad approximant for the square of the logarithm of the single - particle gap to find @xmath40 for @xmath29 , in good agreement with the dmrg value ; for another scheme , see heil and von der linden @xcite . in the present paper , we first refine and extend the perturbative strong - coupling approach in order to analyze the single - particle spectral function and the dynamical structure factor . for the latter quantity , we obtain higher - order corrections from the corresponding green s function . secondly , in order to relax the strong - coupling condition , we employ the variational cluster approach ( vca ) that is applicable in both the mott insulating and the superfluid phase @xcite . for the calculation of spectral properties in the sf phase , the vca can be reformulated in terms of a pseudo - particle approach , whereby single - particle excitations within a cluster are approximately mapped onto particle - like excitations @xcite , or in terms of the self - energy functional approach @xcite . thirdly , we perform large - scale dmrg calculations : ( i ) to access the whole parameter space of the bose hubbard model and ( ii ) to benchmark the reliability of the used analytical strong - coupling and numerical vca techniques . while in the past dmrg has been successfully applied to investigate the ground - state properties of the bose hubbard model @xcite , dmrg results for dynamical properties at zero temperatures are rare ( in contrast to fermionic systems ) , but highly desirable because superfluids in optical lattices can be studied by momentum - resolved bragg spectroscopy @xcite . the outline of this paper is as follows . in sect . [ sec : pa ] we describe perturbative approaches to the bose hubbard model , and present a detailed derivation of the strong - coupling results for static and dynamical quantities . [ sec : vca ] and [ sec : dmrg ] sketch the specifics of the vca and dmrg , respectively , when applied to the bose hubbard model . [ sec : results ] contains our main results . in particular , we discuss how the von neumann entanglement entropy can be calculated from dmrg and how it can be used to determine the kt transition point in the mott lobes . next , we determine the ground - state energy , the boson correlation function , and the momentum distribution function . lastly , we analyze the photoemission spectra and dynamical structure factors . in all cases , we compare analytical and numerical results . finally , sect . [ sec : s ] summarizes our findings . for weak interactions , we use the perturbative results obtained by bogoliubov @xcite ( see fetter and walecka , chap . 35 @xcite ) , for a weakly interacting bose gas with contact interaction and density @xmath11 . from the text - book formulae we find for the 1d bose hubbard hamiltonian at @xmath29 @xmath41 where @xmath42 is the bare dispersion of ( [ hamil ] ) shifted by @xmath43 , @xmath44 is the dispersion of the bogoliubov quasiparticles , and @xmath45 is the number of particles in the condensate . here , @xmath46 are the crystal momenta for pbc . the bogoliubov ground - state energy reads @xmath47 two problems with the bogoliubov theory become apparent when we consider some limits . first , we address the limit @xmath48 for @xmath44 , @xmath49 . therefore , in the thermodynamic limit , the integral in eq . ( [ ooops ] ) is logarithmically divergent , and @xmath50 in 1d results , in agreement with field theory @xcite . this , however , invalidates the starting point of the bogoliubov approximation . second , we can not apply the theory for large @xmath51 because @xmath52 so that @xmath53 for large @xmath51 , in contrast with the exact limit , @xmath54 . the analytical result for the ground - state energy in bogoliubov theory is found , e.g. using mathematica @xcite , as @xmath55 the small-@xmath13 expansion is @xmath56 corrections are of the order @xmath57 which is formally beyond the validity of the bogoliubov expansion which ignores terms of the order @xmath58 . for the bosonic hubbard model , an @xmath16 strong - coupling expansion easily permits the calculation of the ground state for @xmath59 , @xmath60 because it is non - degenerate for the mott lobe with integer filling @xmath11 . likewise , the energy levels of a single - hole excitation , @xmath61 , and of a single - particle excitation , @xmath62 , can be determined to high order in @xmath17 because the perturbation theory for these energy levels also starts from non - degenerate states , e.g. , for @xmath29 , @xmath63 when we employ the unitary harris lange transformation @xcite , the strong - coupling hamiltonian of the bose hubbard model can be derived in a systematic way , @xmath64 in practice , a finite order in the expansion of @xmath65 is kept . when we retain @xmath66 for @xmath67 , we denote this the ` @xmath68th - order approximation ' . in @xmath68th order we thus keep @xmath69 terms in the expansion for @xmath70 whose terms obey @xmath71_{-}=0 $ ] for @xmath72 . to order @xmath69 , the number of double occupancies is conserved by @xmath70 . this defines the construction principle for the operators @xmath73 . the leading order terms for @xmath66 and @xmath74 are given by @xmath75 \hat{s}_2&=&\sum_{d_1,d_2 } \frac{-\hat{p}_{d_1 } \hat{t } \hat{p}_{d_1 } \hat{t } \hat{p}_{d_2 } + \hat{p}_{d_1 } \hat{t } \hat{p}_{d_2 } \hat{t } \hat{p}_{d_2}}{(d_1-d_2)^2 } \\[3pt ] & & + \!\ ! \frac{\hat{p}_{d_1 } \hat{t } \hat{p}_{d_3 } \hat{t } \hat{p}_{d_2}}{2(d_1-d_2 ) } \frac{[d_1-d_3+d_2-d_3]}{(d_1-d_3)(d_2-d_3 ) } \nonumber \ ; , \\[6pt ] \hat{h}_0&= & \sum_{d}\hat{p}_d \hat{t } \hat{p}_{d } \ ; , \label{hath0 } \\[6pt ] \hat{h}_1&= & \sum_{d_1,d_2 } \frac{\hat{p}_{d_1 } \hat{t } \hat{p}_{d_2 } \hat{t } \hat{p}_{d_1}}{d_1-d_2}\ ; , \end{aligned}\ ] ] where @xmath76 is the projection operator onto the subspace of eigenstate with @xmath77 interactions , @xmath78 . in the above sums it is implicitly understood that all indices @xmath79 are mutually different . a compact formula for the recursive generation of higher orders can be found in ref . @xcite . in our analysis , we use a computer program to generate orders @xmath80 in @xmath66 and @xmath74 @xcite . for the exact ground state of the mott insulator we have @xmath81 where @xmath82 is the exact ground - state energy . within the strong - coupling expansion we then find @xmath83 where @xmath84 is the ground state of @xmath85 , see eq . ( [ defphi0 ] ) . since the harris lange transformation is unitary , operators and ground - state expectation values translate according to @xmath86 the series expansion for @xmath65 to @xmath68th order contains @xmath68 powers of the kinetic energy operator @xmath87 . therefore , local operators @xmath88 translate into cluster operators which involve the sites @xmath89 with @xmath90 . the range of @xmath70 scales accordingly : the strong - coupling theory generates a cluster expansion . for fixed momentum @xmath91 , the exact eigenstates of @xmath70 with one extra particle or one hole in @xmath84 , eq . ( [ defphi0 ] ) , are given by the hole and particle states defined in eqs . ( [ defphi - hole ] ) , ( [ defphi - particle ] ) . in this sector , we thus obtain the ground - state energy and the single - particle excitation energies from @xmath92 e_{\rm p}(k ) & = & \langle \phi_{\rm p}(k ) \ ; , \\[3pt ] e_{\rm h}(k ) & = & \langle \phi_{\rm h}(k ) |\hat{h}|\phi_{\rm -e_0 \ ; .\end{aligned}\ ] ] up to and including 6th order in @xmath17 , we obtain for the ground - state energy per site @xmath93}}{4ul}=-x^2+x^4+\frac{68}{9}x^6+{\cal o}(x^8)\;,\ ] ] in agreement with ref . @xcite . the single - hole and single - particle excitations energies are @xmath94 and @xmath95 the single - particle gap is calculated from @xmath96 which results in @xmath97 in agreement with ref . @xcite . the single - particle spectral functions are obtained from @xmath98 where @xmath99 is the excitation energy of the exact eigenstates @xmath100 of @xmath70 with @xmath101 bosons , measured from the ground - state energy , and @xmath102 for pbc . obviously , the single - particle gap @xmath103 is obtained from @xmath104 . for the calculation of the spectral function , we need the weight factors @xmath105 up to and including third order in @xmath17 we find @xmath106^{2 } \ ; , \label{whole } \\ w_{\rm p}(k ) & = & 2 \bigl [ 1-\frac{7}{4 } x^2 + \left(2x-\frac{15}{4 } x^{3}\right)\cos(k ) \nonumber \\ & & + 8x^{2}\cos(2k ) + 18x^{3 } \cos(3k ) \bigr]^{2 } \ ; . \label{wparticle}\end{aligned}\ ] ] the weights @xmath107 are those of the lower and upper hubbard bands which are energetically closest to the single - particle gap and separated by @xmath13 in the atomic limit . in higher orders of the strong - coupling expansion , secondary hubbard bands appear in the single - particle spectral function @xcite . this can most easily be seen from the weights which express the overlap of the exact excited eigenstates of @xmath70 with the states @xmath108 see eqs . ( [ k+ ] ) , ( [ k- ] ) . with an amplitude of the order @xmath109 , the state @xmath110 contains a component with two neighboring holes and one doubly occupied site in a row . this component is _ not _ in the original subspace with @xmath111 and contributes to the upper hubbard band with weight @xmath112 . therefore , for the weight of the lower hubbard band we have @xmath113 the state @xmath114 contains configurations with a triple occupancy and a neighboring hole to the left or right . their amplitude up to order @xmath109 is @xmath115 . they contribute to the secondary hubbard band centered around @xmath116 to order @xmath109 and @xmath117 . components with two double occupancies and a quadruple occupancies have an amplitude proportional to order @xmath109 and thus contribute to the bands centered around @xmath118 and @xmath119 , respectively , with weights of the order of @xmath112 . up to and including order @xmath117 , the secondary hubbard band around @xmath116 has the weight @xmath120 therefore , the total weight for the upper hubbard bands is given by @xmath121 , in agreement with the sum rule @xmath122 = w_{\rm uhb}(k)-w_{\rm lhb}(k ) = 1 \ ; , \label{sr1}\ ] ] which follows directly from the definition of the spectral function . another check results from the momentum distribution sum rule , @xmath123 up to and including third order in @xmath17 , our results for @xmath37 agree with those found in refs . @xcite . for the density - density correlation function we focus on @xmath124 so that we do not have to consider terms of the form @xmath125 . we define the states @xmath126 where the density operator in momentum space is given by ( @xmath127 ) @xmath128 then , we can express the dynamical structure factor in the form @xmath129 where @xmath130 are the exact eigenstates of @xmath70 in the sector with @xmath131 bosons . _ leading order contribution . _ the dynamical structure factor was calculated analytically within mean - field theory @xcite , bosonization @xcite , and lowest - order strong - coupling theory @xcite . the strong - coupling result to leading order is readily obtained from the exact eigenstates of @xmath132 , eq . ( [ hath0 ] ) , in the sector with one hole and one double occupancy . the subspace @xmath133 is spanned by the @xmath134 orthonormal states ( @xmath135 ) @xmath136 the states @xmath137 obey the effective single - particle schrdinger equation @xmath138 as expected for a translational invariant system , the center - of - mass momentum @xmath139 with @xmath140 is conserved . the leading - order contribution to the states @xmath141}\rangle$ ] from ( [ qstatesdef ] ) is given by @xmath142}\rangle = \sqrt{2}\left [ ( 1-e^{{\rm i } q})|q,1\rangle + ( 1-e^{-{\rm i } q})|q , l-1\rangle \right ] \ ; , \label{qstatesleadingorder}\ ] ] i.e. , double occupancy and hole are nearest neighbors . ( [ pairmotion ] ) describes a single particle on an open chain with @xmath143 sites which reflects the fact that hole and double occupancy can not be on the same site , @xmath135 . in contrast to the fermionic case , the hard - core constraint is not sufficient to determine the phase shift between hole and double occupancy because their tunnel amplitudes differ by a factor of two . therefore , the scattering phase shift between double occupancy and hole is not trivial @xcite . this is in contrast to the mean - field approach @xcite where the bare dispersions for hole and double occupancy enter eq . ( [ skomeagfromexacteigenstates ] ) . the normalized double - occupancy hole eigenstates are given by @xmath144 with @xmath145 ( @xmath146 ) where the two - particle phase shift @xmath147 follows from @xmath148= \frac{2\sin(q)}{1 + 2\cos(q ) } \ ; .\ ] ] the energies of the eigenstates @xmath149 of @xmath150 are given by @xmath151 the overlap with the states in eq . ( [ qstatesleadingorder ] ) defines the oscillator strengths in eq . ( [ qstatesdef ] ) , @xmath152}\rangle & = & \sqrt{2 } \sqrt{\frac{2}{l}}\sin(k ) \bigl [ ( 1-e^{{\rm i}q})e^{-{\rm i}\phi(q ) } \nonumber \\ & & -(1-e^{-{\rm i}q})(-1)^{m_k}e^{-{\rm i}\phi(q)(l-1 ) } \bigr]\ ; , \end{aligned}\ ] ] so that , in the thermodynamic limit ( @xmath153 ) , we obtain for the weights @xmath154 where we dropped the cross terms because their contribution to the structure factor vanishes due to the fast oscillations of @xmath155 . the dynamical structure factor becomes for @xmath124 @xmath156}(q,\omega)&= & 2 \left(\frac{4t\sin(q/2)}{u}\right)^2 \label{finals1 } \\ & & \times \int_{0}^{\pi } \frac{{\rm d}k}{\pi } \sin^2(k ) \delta\left(\omega - u - e(q , k)\right)\ ; .\nonumber \end{aligned}\ ] ] finally , for @xmath157 we obtain ( @xmath158 ) @xmath159}(q,\omega)= \biggl(\frac{4\sin(q/2)}{u}\biggr)^2\ ! \frac{\sqrt{20 + 16\cos(q)-(\omega - u)^2}}{2\pi(5 + 4\cos(q))}\ ] ] for the dynamical structure factor to leading order @xcite . _ second - order and higher - order contributions . _ for the next order in the @xmath160-expansion we must calculate the action of @xmath161}$ ] on the states @xmath137 , where we use @xmath162}$ ] . the correction to eq . ( [ pairmotion ] ) reads @xmath163 \nonumber \\ & & -2 \left[(1-\delta_{l , l-1})(1-\delta_{l , l-2})(1+e^{2{\rm i}q})|q , l+2\rangle\right ] \nonumber \\ & & + 2\delta_{l,1 } \bigl [ ( -\frac{1}{4}+e^{-{\rm i}q}+e^{{\rm i}q } ) |q,1\rangle \nonumber \\ & & \hphantom{+ 2\delta_{l,1 } \bigl [ } + ( \frac{1}{4}+e^{-{\rm i}q}+e^{-2{\rm i}q } ) |q , l-1\rangle \bigr ] \nonumber \\ & & + 2\delta_{l , l-1 } \bigl [ ( \frac{1}{4}+e^{{\rm i}q}+e^{2{\rm i}q } ) |q,1\rangle \nonumber \\ & & \hphantom{+ 2\delta_{l , l-1 } \bigl [ } + ( -\frac{1}{4}+e^{-{\rm i}q}+e^{{\rm i}q } ) |q , l-1\rangle \bigr ] \ ; . \label{nextorderschroedinger}\end{aligned}\ ] ] the effective single - particle problem contains an overall energy shift @xmath164 , a nearest - neighbor transfer @xmath165 with amplitude @xmath166 as before , an additional next - nearest neighbor transfer from @xmath167 with amplitude @xmath168 , and a potential at the chain ends , @xmath169 and @xmath170 . now that the potential links the two chain ends , it is computational advantageous to treat the problem on a ring instead of a chain . the potential is readily generalized according to eq . ( [ nextorderschroedinger ] ) . for a ring , the potential also contains the terms @xmath171 and @xmath172 and their complex conjugates so that the potential links four neighboring sites . moreover , the extension of @xmath132 from a chain to a ring generates corrections to @xmath173 and @xmath174 . the @xmath175-corrections to the states @xmath176 ( [ qstatesdef ] ) read @xmath177}\rangle = 3\sqrt{2 } \left [ ( 1-e^{2{\rm i } q})|q,2\rangle + ( 1-e^{-2{\rm i } q})|q , l-2\rangle \right ] \ ; . \label{qstatessecondorder}\ ] ] as the potential @xmath178 , the dynamical structure factor to second order involves the four neighboring sites @xmath179 . the calculation of all eigenstates of @xmath180 is not feasible in the thermodynamic limit . to calculate the dynamical structure factor we address the corresponding green s function @xmath181 for the structure factor in leading order , eq . ( [ qstatesleadingorder ] ) requires the four green s functions @xmath182 for @xmath183 . the second order requires the green s function for @xmath184 . this cluster principle generalizes to higher orders , i.e. , in @xmath68th order we have to calculate a @xmath185 matrix of green s functions for a potential which links @xmath186 neighboring sites . the green s function of a particle in a potential of finite range is readily calculated @xcite . we start from @xmath187 , where @xmath188 describes the free particle motion over the ring with dispersion relation @xmath189 ; up to second order , we have @xmath190 + 13(t^2/u)-4(t^2/u ) [ \cos(2k)+\cos(2k-2q)]$ ] with @xmath191 . in the thermodynamic limit , the free green s function is readily calculated , @xmath192 where we use the fact that the free states are plane waves . we calculate the free green s functions for @xmath193 with the help of the residue theorem . therefore , their real and imaginary parts are available with high accuracy for all real frequencies @xmath124 . with the help of the operator identity @xmath194 we derive the green s function ( [ greenfunction ] ) from the equation @xmath195 this matrix equation has the formal solution @xmath196 in @xmath68th - order perturbation theory , the potential and the required green s functions have the same range @xmath186 on the lattice . therefore , the matrix problem in ( [ matrixproblem ] ) reduces to the inversion of a @xmath197-matrix for fixed @xmath198 . the green s function calculation provides higher - order corrections to @xmath199}(q,\omega)$ ] . in sect . [ sec : results ] we show results for the dynamical structure factor to fifth order in the @xmath160-expansion in the region around @xmath200 . the green s function calculation does _ not _ cover the higher hubbard sub - bands . the first contribution to the structure factor @xmath201 beyond @xmath202 occurs around @xmath116 with intensity @xmath203 . in the regions where strong - coupling perturbation theory is reliable , @xmath204 , the secondary bands contribute only a few percent of the total weight . when we include the term to order @xmath203 in the frequency - integrated structure factor , the sum - rule for the structure factor is fulfilled , i.e. , we reproduce the terms @xmath205 of the ground - state density - density correlation function for @xmath32 @xcite up to and including @xmath112 . the basic idea of the vca is to approximate the self - energy @xmath206 of a strongly correlated , physical system @xmath207 by the self - energy of an exactly solvable reference system @xmath208 @xcite . both the physical and the reference system share the same interaction but differ in their single - particle terms . the optimal self - energy is determined self - consistently from a stationary condition on the grand - canonical potential @xmath209 , @xmath210 to evaluate this expression , the self - energy is parameterized by the single - particle parameters of the reference system . in fact , this idea is quite general and allows to unify ( cluster)-dynamical mean field theory and vca within the same theoretical framework depending on the choice of the reference system @xcite . in the case of the vca , the reference system is chosen to be a cluster decomposition of the physical system with modified single - particle parameters . furthermore , the reference system is selected such that it can be solved exactly . in principle , any many - body cluster solver at hand can be used which provides the dynamic single particle green s function . here , we use lanczos exact diagonalization @xcite . originally , vca was introduced for fermionic systems @xcite . for correlated lattice bosons , it first has been in use to investigate the normal , mott insulating phase @xcite . in refs . @xcite vca has been extended to the superfluid phase . this extension adopts the nambu notation and is applicable to a large class of lattice models that exhibit a condensed phase . since vca is in the end a form of a cluster mean - field approach , it can obviously not comprise fluctuations at length scales larger than the cluster size . this means that in the case of power - law decaying correlations as present here , these are spuriously replaced by long - range order in the vca . this is a common issue of all mean - field like approaches . despite this drawback , vca still provides reliable results for many observables such as the ground - state energy , the sound velocity of the phonon excitations , and the single - particle spectral function . explicitly , the grand potential for bosonic systems with normal and the superfluid components is given by @xcite @xmath211^{-1 } \langle \hat a \rangle - \langle \hat { a'}\rangle^{\dagger } [ g{_{(0)}}^{\prime}]^{-1 } \langle\hat a'\rangle \ ; , \label{eq : omvca}\end{aligned}\ ] ] where @xmath212 is the interacting green s function , @xmath213 is the single - particle hamiltonian matrix , @xmath214 denotes expectation values of the nambu boson operators consisting of both creation and annihilation operators , and the subscript ` ( 0 ) ' indicates that the green s function is evaluated at zero wavevector and zero frequency . in , the prime marks again reference system quantities . the first line of is identical to the expression in the normal phase @xcite apart from the fact that the green s functions are considered to be in nambu space and thus contain anomalous parts which also account for the factor @xmath215 . the second line takes care of the condensation of bosons , which in one dimension is an artifact of the dynamical and self - consistent mean - field treatment , as discussed above . to obtain the results presented below , we always use the chemical potential of the reference system and a field which breaks the @xmath216 symmetry on the level of the reference system as variational parameters . in the mott phase we also determine the intercluster hopping and the boundary energies of the reference system self - consistently @xcite . having found the stationary point of the grand potential @xmath209 with respect to the variational parameters , we evaluate the dynamical single particle green s function @xmath217 of the physical system @xcite . from that we calculate the single - particle spectral function @xmath218 . the static density - density correlation functions can be obtained from the fourier transform of the momentum distribution function , as specified in ref . the dmrg allows us to calculate static , dynamic and spectral properties of the 1d bose hubbard model with high precision for fairly large system sizes . the main obstacle is related to the fact that , in principle , each lattice site can be occupied by infinitely many bosonic particles . therefore , one has to introduce a cutoff @xmath219 , the maximum number of bosons per site taken into account . the dmrg results are nonetheless unbiased and numerically exact , if the dependence on @xmath219 can be proven to be negligible and a careful finite - size extrapolation to the thermodynamic limit ( @xmath153 ) has been performed . within the ground - state dmrg technique @xcite the energy functional @xmath220 is minimized in a variational subspace in order to find the ground - state wave function @xmath221 and energy @xmath222 whereby the dmrg energy error is proportional to the weight of the density - matrix eigenstates discarded in the renormalization process . increasing the number @xmath223 of density - matrix eigenstates kept , the discarded weight can be reduced systematically . practically , the ground - state dmrg procedure mostly consists of two steps . during the infinite - system algorithm the system size is enlarged by two sites at each step and this operation has to be continued until the whole system reaches the desired system size @xmath1 . subsequently , a finite - system algorithm is used , where several sweeps through a lattice of fixed size @xmath1 are performed . thereby , the lattice is divided in two blocks with @xmath224 respectively @xmath225 sites where @xmath226 . this sweeping improves the quality of the results obtained in the infinite - system algorithm . we note that this procedure is perfectly suited to compute the von neumann entanglement entropy on - the - fly in the finite - system algorithm . from this quantity , the kt transition point of the bose hubbard model can be determined accurately . the dmrg procedure can also be used to minimize the following functional @xcite : @xmath227 here , @xmath228 is the ( time - independent ) hamilton operator and @xmath229 denotes the quantum operator of the physical quantity to be analyzed ; @xmath230 is its hermitian conjugate and @xmath231 . once this minimization has been carried out , the dynamical correlation function @xmath232 can be evaluated . here , the small real number @xmath233 shifts the poles of the correlation function in the complex plane , i.e. , @xmath233 leads to a lorentzian broadening of the peaks of the corresponding spectral function given in lehmann representation as @xmath234 within this so - called dynamical dmrg ( ddmrg ) technique , the sweeps in the finite - system algorithm are repeated until both functionals , @xmath235 and @xmath236 , take their minimal values . investigating the bose hubbard model by dmrg , we keep up to @xmath237 bosonic particles per site . furthermore , we use @xmath238 density - matrix eigenstates in the dmrg runs for the ground - state expectation values . then , the discarded weight is typically smaller than @xmath239 . in the ddmrg calculations we keep @xmath240 states to determine the ground state during the first five dmrg sweeps , and afterwards use @xmath241 states for the calculation of the dynamical properties . previously , the kt transition point between the superfluid and insulating phases has been determined from the luttinger parameter @xmath242 @xcite , which can be extracted from the density - density correlation function by dmrg , yielding @xmath243 ( @xmath244 ) for @xmath29 ( @xmath245 ) @xcite . although @xmath246 is not defined in the mi , @xmath247 is finite and continuous over the kt transition because the mott gap is exponentially small . therefore , it can be used within a dmrg finite - size extrapolation procedure . the quantum phase transition should become manifest in the system s entanglement properties as well @xcite . an important measure to quantify the entanglement of two subsets of an interacting quantum system is the von neumann entanglement entropy , which shows a logarithmic scaling for critical systems @xcite . to determine the critical point between a tomonaga luttinger liquid and gapped ( or ordered ) phases for more subtle situations , e.g. for frustrated spin models , spinless fermion models with nearest - neighbor interaction or fermion - boson transport models , the use of the entanglement entropy difference has been demonstrated to be advantageous @xcite . from eq . , for the 1d bose hubbard model with @xmath29 ( upper panel ) and @xmath245 ( lower panel ) . data obtained by dmrg for lattices up to @xmath248 with pbc . the closed symbols indicate the maximum value for each system size . an extrapolation of the @xmath0 values at these maxima to the thermodynamic limit provides the kosterlitz thouless transition point , see insets ; here , the lines correspond to a polynomial fit . the vertical dashed lines in the main panels mark the kosterlitz thouless transition point . [ kttransition],title="fig : " ] + from eq . , for the 1d bose hubbard model with @xmath29 ( upper panel ) and @xmath245 ( lower panel ) . data obtained by dmrg for lattices up to @xmath248 with pbc . the closed symbols indicate the maximum value for each system size . an extrapolation of the @xmath0 values at these maxima to the thermodynamic limit provides the kosterlitz thouless transition point , see insets ; here , the lines correspond to a polynomial fit . the vertical dashed lines in the main panels mark the kosterlitz thouless transition point . [ kttransition],title="fig : " ] for a block of length @xmath224 in a periodic system of the system size @xmath1 , the von neumann entropy , @xmath249 , is given by @xmath250 , with the reduced density matrix @xmath251 . one finds for pbc @xcite , @xmath252 + s_1\ ; , \label{von_neumann_entropy}\end{aligned}\ ] ] where @xmath253 is the central charge . when one evaluates the entropy difference @xmath254 using dmrg with open boundary conditions ( obc ) @xcite , it includes the effect of the non - universal constant @xmath255 . therefore , the values for @xmath256 can not be extrapolated systematically . here , we follow the alternative scheme proposed by nishimoto @xcite . we subtract @xmath257 from @xmath258 to obtain @xmath259}{\ln\left[\cos(\pi / l)\right]}\;. \label{eq - cstar}\end{aligned}\ ] ] as @xmath153 , in the sf regime , the quantity @xmath260 scales to the central charge @xmath261 @xcite . [ kttransition ] displays @xmath260 for the 1d bose hubbard model . advantageously , we can use periodic boundary conditions for the calculation of this quantity . as shown in the insets , the position of the maximum in @xmath262 can be reliably extrapolated to the thermodynamic limit . in this way we get the cone point of the mott lobes @xmath263 for @xmath29 and @xmath264 for @xmath245 ( in units of @xmath13 ) , in excellent agreement with the previous estimates from the obc finite - size scaling of @xmath242 @xcite . the ground - state energy @xmath82 of the 1d bose hubbard model has been determined analytically in the weak- and strong - coupling cases . for weak interactions , the bogoliubov result was given in eq . , with the small-@xmath13 expansion given by eq . . for strong interactions , an expansion up to 14th order in @xmath16 was obtained by damski and zakrzewski @xcite : @xmath265}}{4ul}&= & -x^2+x^4+\frac{68}{9}x^6 -\frac{1267}{81}x^8+\frac{44171}{1458}x^{10 } \nonumber\\[2pt ] & & -\frac{4902596}{6561}x^{12}-\frac{8020902135607}{2645395200}x^{14 } \\[2pt ] & & + { \cal o}(x^{16 } ) \nonumber \;. \label{expansion - gs}\end{aligned}\ ] ] , as a function of interaction strength @xmath0 for @xmath29 . weak - coupling and strong - coupling results are compared with the @xmath153 extrapolated dmrg data obtained from chains up to @xmath21 with obc . for the vca calculations ( crosses ) a cluster with @xmath266 ( @xmath267 ) sites is used in the mi ( sf ) phase.[fig : gs ] ] fig . [ fig : gs ] compares these perturbative results with our vca and dmrg data . the vca reproduces the dmrg results almost perfectly for all interaction strengths @xmath268 . the strong - coupling series expansion is also in accordance with the numerical exact data , surprisingly even beyond the kt transition point , i.e. , for @xmath269 . note that in ref . [ ] the ground - state energy was compared with numerical data obtained for a system with @xmath270 sites only . hence , in their figure , the deviation starts at about @xmath271 . clearly , the quality of the strong - coupling approximation improves as higher - order corrections are taken into account , cf . the 4th- , 10th- , and 14th - order results . [ fig : gs ] also shows the range of validity of the corresponding weak - coupling approaches . surprisingly , the bogoliubov result is applicable up to the mott transition point . in order to characterize the correlations in the ground - state of the interacting bose gas described by the bose hubbard model , it is instructive to look at the distance dependence of the expectation values @xmath272 , which , with appropriate normalization , constitute the matrix elements of the one - particle density matrix @xcite . + ) for decreasing interaction strengths @xmath273 ( a ) , 0.1 ( b ) , 0.15 ( c ) , and 0.2 ( d ) . dmrg results are obtained for a chain with @xmath21 sites and obc . to minimize the boundary effects we place @xmath6 and @xmath224 symmetrically around the center of the system . the vca data were calculated using clusters with @xmath267 ( green triangles ) , 8 ( blue squares ) , and 12 ( red circles ) . [ corr_func],title="fig : " ] + in the gapless sf state , the boson single - particle correlation function @xmath274 shows a power - law decay with an exponent determined by the tomonaga luttinger parameter @xmath242 @xcite . in the insulating ( gapped ) mi , the bosonic correlations decay exponentially ( at large distances ) , which is demonstrated by the semi - logarithmic representation in fig . [ corr_func ] . at very strong couplings , the excitation gap is large and therefore can be obtained very accurately within vca . as @xmath17 becomes larger , i.e. , @xmath13 becoming smaller at fixed @xmath12 , the correlations are significant over many lattice sites . in this regime , we find noticeable deviations of the vca results if @xmath275 is too small , see panel ( d ) . the fourier - transformed single - particle density matrix gives the momentum distribution function @xmath276 to third order in @xmath16 , strong - coupling theory predicts for the first mott lobe @xcite : @xmath277}(k)&=&1+(8x-16x^3)\cos(k ) \nonumber \\ & & + 36 x^2 \cos(2k)+176 x^3\cos(3k)\;.\label{3rd - nk}\end{aligned}\ ] ] in fig . [ nk_mi ] we compare the strong - coupling expansion ( [ 3rd - nk ] ) with the dmrg and vca numerics . within the first mott lobe from dmrg with @xmath248 and pbc ( symbols ) , vca ( dashed lines ) , and third - order strong - coupling expansion ( solid lines ) . the inset in panel ( d ) shows the dependence of the vca results on the cluster size @xmath275 for @xmath278 in comparison with the dmrg data.[nk_mi],scaledwidth=47.0% ] while for @xmath279 , where the momentum distribution is rather flat indicating weak site - to - site correlations , all methods essentially agree [ see panel ( a ) ] , small deviations between analytical and numerical approaches appear for @xmath280 [ panel ( b ) ] . the oscillations emerging for @xmath281 in the third - order strong - coupling theory are clearly an artifact . the vca reproduces the density distribution @xmath37 very well . however , it fails quantitatively for @xmath48 and @xmath282 if the cluster used is not large enough , see the inset in panel ( d ) . when we approach the mi - sf kt transition point by raising @xmath17 , @xmath283 will rapidly increase with system size . in 1d , of course , @xmath283 will not attain a macroscopic value in the thermodynamic limit because no true condensate develops @xcite . instead , we have from eq . @xmath284 thus far , it is difficult to reproduce this algebraic divergence in the sf phase ; see , however , ref . [ ] , where @xmath285 was determined by dmrg . the single - particle excitations associated with the injection and emission of a boson with wave vector @xmath91 and frequency @xmath286 , are described by the spectral functions @xmath287 and @xmath288 , see eqs . and , respectively . these quantities can be evaluated by vca @xcite and ddmrg @xcite . for the bose hubbard model the following sum - rules hold [ cf . ] : @xmath289&=&1\;,\\ \int_{-\infty}^{\infty } { \rm d}\omega a^-(k,\omega)&=&n(k)\;.\end{aligned}\ ] ] summing over momenta @xmath91 , the density of states @xmath290 follows as @xmath291 where @xmath292 . within the ddmrg framework , however , it is much more appropriate to calculate @xmath290 directly , instead of performing the @xmath91-summation of @xmath293 . first , we discuss the spectral function , @xmath294 , and the density of states , @xmath290 , in the mi regime . the ddmrg spectra for fixed @xmath91 consist of two lorentzians of width @xmath295 , which is the broadening introduced in the ddmrg procedure , cf . [ sec : dmrg ] . [ spec - strong ] shows the quasiparticle dispersions ( squares ) extracted from lorentz fits to the maxima in @xmath293 . the quality of the fits suggests that the quasiparticle life - time is very large . because of the large mott gap separating single - particle and single - hole quasiparticle bands , the vca can work with small cluster sizes and the quasiparticle spectra are in perfect agreement with the ddmrg data . the same holds for the strong - coupling results . in fact , for large interactions , each site is singly occupied in the ground state . as a consequence , a hole or doubly occupied site can move almost freely through the system . from this consideration , the leading - order expression for the quasiparticle dispersions results @xcite , see eqs . ( [ ehole ] ) and ( [ eparticle ] ) . we note in passing that the simple mean - field approach by van oosten et al . @xcite fails to reproduce the quasiparticle dispersion already for @xmath296 , see ref . as the on - site interaction further weakens , the mott gap gradually closes ; the corresponding results are depicted in fig . [ spec - intermediate ] . obviously , strong - coupling theory becomes imprecise at @xmath297 , and completely fails at @xmath298 . there also vca shows some artificial gap features near the brillouin zone boundary , which do not show up in ddmrg . in the @xmath91-@xmath286 plane ( left panels ) and density of states @xmath290 ( right panels ) for the 1d bose hubbard model with @xmath29 and @xmath279 ( upper panels ) , @xmath299 ( lower panels ) we compare ddmrg data for a system with @xmath248 and obc ( squares ) with the results of vca for @xmath266 and @xmath300 ( density plots ) and strong - coupling expansions ( [ ehole ] ) and ( [ eparticle ] ) ( lines).[spec - strong],title="fig:",scaledwidth=47.0% ] + in the @xmath91-@xmath286 plane ( left panels ) and density of states @xmath290 ( right panels ) for the 1d bose hubbard model with @xmath29 and @xmath279 ( upper panels ) , @xmath299 ( lower panels ) . we compare ddmrg data for a system with @xmath248 and obc ( squares ) with the results of vca for @xmath266 and @xmath300 ( density plots ) and strong - coupling expansions ( [ ehole ] ) and ( [ eparticle ] ) ( lines).[spec - strong],title="fig:",scaledwidth=47.0% ] in the superfluid phase , the elementary excitations concentrate around the region @xmath301 , see fig . 7 in ref . [ ] , which indicates the formation of a `` condensate '' . in accordance with bogoliubov theory and field theory @xcite , the low - energy , low - momentum excitations dominate the single - particle spectrum . as can be seen from fig . [ spec - sf ] , our spectral function indeed exhibits a phonon mode whose excitation energy for a system in the thermodynamic limit is linear in @xmath91 and gapless at @xmath302 . yet , for finite - size systems a gap is present whose magnitude is inversely proportional to the system size . our ddmrg data demonstrate that the gap almost vanishes already for a obc system with 64 sites . a similar behavior has been observed in qmc calculations which employ the directed - loop method @xcite . ( upper panels ) , @xmath303 ( middle panels ) and @xmath304 ( lower panels ) . notations are the same as in fig . [ spec - strong ] . [ spec - intermediate],title="fig:",scaledwidth=48.0% ] + ( upper panels ) , @xmath303 ( middle panels ) and @xmath304 ( lower panels ) . notations are the same as in fig . [ spec - strong ] . [ spec - intermediate],title="fig:",scaledwidth=48.0% ] + ( upper panels ) , @xmath303 ( middle panels ) and @xmath304 ( lower panels ) . notations are the same as in fig . [ spec - strong ] . [ spec - intermediate],title="fig:",scaledwidth=48.0% ] within the vca , we find a larger gap as compared to ddmrg , due to the fact that we solve only four - site clusters exactly which are subsequently coupled perturbatively . in ref . [ ] we showed that the gap at @xmath302 decreases with increasing cluster size , suggesting the correct behavior in the infinitely large cluster limit . along the linear goldstone modes , the spectral weight obtained by means of vca exhibits fringes and a series of minigaps . this behavior is most likely a result of the cluster decomposition and subsequent periodization of the green s function @xcite . however , it should be emphasized that the slopes of the phonon mode obtained by the two methods agree very well . a universal feature of systems with broken @xmath216 symmetry is that , in addition to a gapless goldstone mode , a gapped amplitude mode should be present . whereas the goldstone modes correspond to phase fluctuations , the amplitude modes arise from fluctuations in the magnitude of the order parameter . this behavior can be sketched by a mexican hat potential for the order parameter @xcite . it has been argued in ref . [ ] that the amplitude modes are sharp excitations in the quasiparticle sense only for dimensions @xmath305 for which they were detected experimentally @xcite . for @xmath306 the decay of the amplitude modes into goldstone modes is very efficient and , thus , the weight observed in the susceptibilities can be redistributed over a large frequency range . this renders an observation of the amplitude modes difficult . in 2d it was demonstrated theoretically that the coupling to the amplitude modes can be improved by evaluating susceptibilities for the kinetic energies @xcite or for operators that resemble the rotationally invariant structure of the mexican hat potential @xcite . this should result in clearer signals for the amplitude modes in the respective response functions . indeed , in setups with ultracold atoms , recent lattice modulation experiments , which couple directly to the amplitude modes , provide evidence for their existence in 2d @xcite . ( left panel ) , @xmath307 ( right panel ) . compared are ddmrg dispersions with vca density plots for @xmath267 and the dispersion of the condensate excitations @xmath44 from bogoliubov theory ( black lines , other symbols are the same as in fig . [ spec - strong ] ) ; see eq . with .[spec - sf],scaledwidth=48.0% ] in 1d where no true condensate exists , the spectral smearing of the amplitude modes is believed to be even more pronounced . our numerical dmrg and vca results for the spectral function and the dynamical structure factor , which both couple to the gapless goldstone mode and the amplitude mode , give no indication for the latter . therefore , we do not expect that an amplitude mode will be observed in bragg spectroscopy experiments for bosons in one dimension . it is quite remarkable that vca reproduces the overall character of the single - particle spectrum consisting of goldstone modes only , despite of the fact that , technically , a spurious condensate has to be introduced to treat the superfluid phase , see sec . [ sec : vca ] for a detailed discussion . in the mi phase of the 1d bose hubbard model for different @xmath0 where @xmath29 . ddmrg data were obtained for an @xmath308 site system with pbc , using @xmath309 . red crosses mark the positions of the maximum in each @xmath139-sector . [ sqwmi_density ] ] we now turn to the dynamical density - density response function . we carry out large - scale ddmrg calculations of the dynamic structure factor , @xmath201 , and compare the results with the predictions of strong - coupling theory where appropriate . in strong coupling , we show results in fourth - order approximation . the agreement with the ddmrg data for @xmath310 improves noticeably when we calculate expectation values with @xmath311}\rangle$ ] from ( [ qstatesdef ] ) , i.e. , we keep the states to fifth order in @xmath17 . since ddmrg provides @xmath201 with a finite broadening @xmath233 , it turns out to be useful to convolve the strong - coupling result with the lorentz distribution @xcite , @xmath312}\;.\end{aligned}\ ] ] _ mott phase . _ [ sqwmi_density ] illustrates the change of the intensity distribution of @xmath201 in the @xmath313-@xmath286 plane as @xmath16 increases in the mi regime . for small @xmath17 , deep in the mi , the spectral weight is concentrated around @xmath314 in the region @xmath315 ( cf . the upper left panel ) . this meets the predictions of the strong - coupling theory @xcite . in this regime the structure factor is dominated by the primary band . when @xmath17 increases , the maximum of @xmath201 acquires an appreciable dispersion ; simultaneously the overall intensity of the density - density response strengthens , see the middle panels of fig . [ sqwmi_density ] and also fig . [ sqwmi ] . as the system approaches the mi - sf transition point , the excitation gap closes . concomitantly , we find a significant redistribution of the spectral weight to smaller @xmath313 values , see the lower panels of fig . [ sqwmi_density ] . in fig . [ sqwmi ] we show constant - moment scans of @xmath201 at @xmath316 and @xmath317 . for @xmath273 and @xmath318 , the agreement between the broadened strong - coupling results and the ddmrg data for @xmath201 is excellent . as @xmath17 becomes larger than @xmath319 , the strong - coupling theory yields a double - peak structure in @xmath320 . when we increase the lattice size and reduce @xmath233 , this feature also appears in our ddmrg data for @xmath321 . therefore , this feature is not an artifact of the strong - coupling approach even though the strong - coupling expansion overestimates the double - peak structure for @xmath310 . the strong - coupling expansion solves an effective single - particle problem in a ( finite - range , attractive ) potential . such a potential gives rise to a non - trivial spectrum ( resonances and possibly bound states ) . the energy levels of the effective single - particle problem lead to non - trivial spectral signatures in the dynamical structure factor . ( left panels ) , @xmath317 ( right panels ) . ddmrg data ( circles ) were obtained for @xmath248 , pbc , and @xmath309 ; at @xmath321 ( @xmath317 ) also for @xmath21 , pbc , and @xmath322 ( green crosses ) . blue solid ( red dashed ) lines give the corresponding results of the strong - coupling theory with @xmath323 ( @xmath309 ) . please note the different scales of the ordinates.[sqwmi ] ] _ superfluid phase . [ skwsf ] and [ sqwsft0.40qpi ] present the corresponding results for the dynamical structure factor in the sf phase . at small momenta , bogoliubov theory gives the correct slope of the dispersion which we derive from the maximum of the ddmrg data for @xmath201 . note that the dispersion @xmath324 in ( [ be ] ) is identical to the predictions from field theory @xcite . bogoliubov s dispersion overestimates the ddmrg maxima for larger momenta and higher energies , as observed experimentally for a 3d setup @xcite . as compared to the mi phase , the density - density response has higher intensity in the sf state . interestingly , we also find a shoulder in @xmath320 , which may form a double peak as @xmath153 , @xmath325 , see the right - hand panel of fig . [ sqwsft0.40qpi ] . this high - energy double peak in the sf phase resembles the structure seen in the mi phase . in our opinion , this rules out an interpretation of the second peak as signature of a massive higgs mode @xcite . in the superfluid phase of the bose hubbard model with @xmath29 . again we use @xmath308 , pbc , and @xmath309 . the yellow line gives the bogoliubov result . [ skwsf ] ] at @xmath316 and @xmath317 . only ddmrg data are shown . circles ( crosses ) mark the results for @xmath248 , pbc , and @xmath309 ( @xmath21 , pbc , and @xmath322 ) . [ sqwsft0.40qpi ] ] the aim of this paper was twofold : ( i ) to provide extensive numerical ( d)dmrg data for static and dynamical quantities of the one - dimensional bose hubbard model at integer filling , mostly for @xmath326 ; ( ii ) to compare the ( d)dmrg results with the analytical strong coupling perturbation theory and the numerically inexpensive vca and thereby explore their merits and limitations in the most demanding case of one dimension . the ground - state energy from dmrg compares favorably with results from vca and from perturbation theory . for static correlation functions such as the single - particle density matrix and the momentum distribution , the comparison between dmrg data and strong - coupling perturbation theory ( vca ) shows that the latter are reliable for @xmath327 ( @xmath328 ) , for doable implementations . we calculated dynamical quantities such as the single - particle spectral function and the dynamical structure factor . in the superfluid phase , the response at low energies is dominated by the quasi - condensate , in agreement with predictions from field theory and bogoliubov theory . the latter provides the correct result for the phonon mode despite the fact that it is based on the incorrect assumption of a true condensate . for finite interactions and at higher energies , the dynamical structure factor is broad and reflects the physics of the mott insulator . the overall character of the single - particle spectrum and the sound velocity of the phonon modes are reproduced by vca for larger values of @xmath17 . the strong - coupling results for the dynamical structure factor helped us to interpret our numerical ddmrg data because the latter are spectrally broadened for finite system sizes . the two - particle correlation function in the mott phase reflects the ( scattering ) states of a doubly occupied site and a hole with a hard - core repulsion and a ( weak ) longer - ranged attraction giving rise to a double - peak structure in the dynamical structure factor near the boundary of the brillouin zone . our numerical work can be compared with experiments only after the parabolic confinement potentials will have been taken into account . important as it is to confine the atoms to the optical lattice , the confinement potential often is so strong that the density profile contains several mott regions with different integer fillings and transition regions between them . in this case , the structure factor at low energies describes the dynamical response of the ` wedding - cake ' density profile @xcite . there are two other directions to extend our work . the mott gap in the bose hubbard model resembles a band gap . therefore , it is interesting to see how bound states ( ` excitons ' ) form in this gap in the presence of a nearest - neighbor attraction . a second route to extend our work is the inclusion of a disorder potential @xcite so that the smearing and closing of the mott gap as a function of the disorder can be studied . work in this direction is in progress . the authors thank i. danshita , g. hager , e. jeckelmann , h. monien , s. nishimoto , and w. zwerger for valuable discussions . financial support by the deutsche forschungsgemeinschaft through the sfb 652 and by the austrian science fund ( fwf ) p24081-n16 are gratefully acknowledged .

we introduce nsabc/ nice - structured algebraic block cipher using-bit word arithmetic , a-bit analogous of skipjack with-bit key . the skipjack s internal 4-round feistel structure is replaced with a-bit , 2-round cascade of a binary operation that permutes a text word under control of a key word . the operation , similarly to the multiplication in idea , bases on an algebraic group over-bit words , so it is also capable of decrypting by means of the inverse element of in the group . the cipher utilizes a secret-bit tweak an easily changeable parameter with unique value for each block encrypted under the same key that is derived from the block index and an additional-bit key . a software implementation for takes circa 9 clock cycles per byte on x86 - 64 processors .

1105.3388

in the today s world full of crypto algorithms , one may wonder what makes a block cipher attractive . in the authors opinion , the answer to the question is one word : elegance . if something looks nice , then there is a big chance that it is also good . an elegant specification makes it easier to memorize . memorability makes it easier to realize and to analyze , that allows for fruitful cryptanalytic results , leading to deeper understanding which , in turn , makes greater confidence in the algorithm . the elegance comprises the following features : * few algebraic operations . using of many operations results in hardly - tractable and possibly undesirable interactions between them . * simple and regular key schedule . a complex key schedule , which effectively adds another , unrelated , function to the cipher , results in hardly - tractable and possibly undesirable interactions between the functions . idea , a secure block cipher designed by xuejia lai and james l. massey @xcite is an example of elegance . besides being elegant with an efficient choice and arrangement of algebraic operations , it is elegant for some more features : * the use of incompatible group operations , where _ incompatible _ means there are no simple relations ( such as distributivity ) between them . the incompatibility eliminates any exploitable algebraic property thus makes it infeasible to solve the cipher algebraically . * the use of modular multiplication . multiplication produces huge mathematical complexity while consuming few clock cycles on modern processors . it thus greatly contributes to security and efficiency of the cipher . however , idea uses multiplication modulo the fermat prime @xmath8 which does not exist for @xmath9 or @xmath7 , making it not extendable to machine word lengths nowadays . furthermore , its key schedule is rather irregular due to the rotation of the primary key . skipjack , a secure block cipher designed by the u.s . national security agency @xcite , is another example of elegant design . besides being elegant with an efficient , simple and regular key schedule , it is elegant for one more feature : the use of two ciphers an outer cipher , or _ wrapper _ , consisting of first and last rounds , and an inner cipher , or _ core _ , consisting of middle rounds . the terms `` core '' and `` wrapper '' were introduced in the design rationale of a structural analogous of skipjack : the block cipher mars @xcite . mars s designers justify this two - layer structure by writing that it breaks any repetitious property , it makes any iterative characteristic impossible , and it disallows any propagation of eventual vulnerabilities in either layer to the other one , thus making attacks more difficult . the wrapper is primarily aimed at fast diffusion and the core primarily at strong confusion . as claude e. shannon termed in his pioneer work @xcite , _ diffusion _ here refers to the process of letting each input bit affect many output bits ( or , equivalently , each output bit be affected by many input bits ) , and _ confusion _ here refers to the process of letting that affection very involved , possibly by doing it multiple times in very different ways . if a cipher is seen as a polynomial map in the plaintext and the key to the ciphertext , then the methods of diffusion and confusion can be described as the effort of making the polynomials as complete as possible , i.e. such that they contain virtually all terms at all degrees . approach is very evident in the structure of skipjack ( see figure [ fig : the - full - cipher-2 ] ) . skipjack ( as opposed to mars ) was moreover sought elegant as the wrapper there is , in essence , the inverse function of the core . however , skipjack uses an s - box that renders it rather slow , hard to program in a secure and efficient manner , and not extendable to large machine word lengths , as such . this article describes an attempt to combine the elegant idea of using incompatible and complex machine - oriented algebraic operations in idea with the elegant structure of skipjack into a scalable and tweakable block cipher called nsabc nice - structured algebraic block cipher . nsabc is scalable . it is defined for every even word length _ w_. it encrypts a 4__w__-bit text block under a 5__w__-bit key , thus allows scaling up with 8-bit increment in block length and 10-bit increment in key length . nsabc is tweakable . it can use an easily changeable 4__w__-bit parameter , called _ tweak _ @xcite , to make a unique version of the cipher for every block encrypted under the same key . included in the specification is a formula for changing the tweak . nsabc makes use of entirely the overall structure of skipjack , including the key schedule , and only replaces the internal 4-round feistel structure of skipjack with another structure . the new structure consists of two rounds of the binary operation @xmath10 , that encrypts a text word _ x _ using a key word _ z _ and a key - dependent word _ e_. the operation @xmath11 is derived from an algebraic group over _ w_-bit words taking @xmath12 as the unit element , so it is also capable of decrypting by means of the inverse element of _ z _ in the group . the two rounds are separated by an exclusive - or ( xor ) operation that modifies the current text word by a tweak word . nsabc is put in public domain . as it bases on skipjack , eventual users should be aware of patent(s ) that may be possibly held by the u.s . government and take steps to make sure the use is free of legal issues . we ( the designers of nsabc ) are not aware of any patent related to other parts of the design . the rest of the article is organized as follows . section 2 defines operations and notations . section 3 specifies the cipher . section 5 gives numerical examples . section 4 suggests some implementation techniques . section 6 concludes the article . source code of software implementations are given in the appendices . throughout this article , _ w _ denotes the machine word length . we use the symbols @xmath13 , @xmath14 , @xmath15 and @xmath16 to denote addition , subtraction ( and arithmetic negation ) , multiplication and multiplicative inversion , respectively , modulo @xmath17 ( unless otherwise said ) . we use the symbols @xmath18 and @xmath19 to denote bit - wise complement and exclusive - or ( xor ) _ _ on _ w_-bit operands ( unless otherwise said ) . we write @xmath20 to denote leftward rotation ( i. e. cyclic shift toward the most significant bit ) of _ x _ , that is always a _ w_-bit word , by _ n _ bits . for even _ w _ , the symbol @xmath21 denotes swapping the high and low order halves , i.e. @xmath22 let s define binary operation @xmath23 by @xmath24 and binary operation @xmath6 by @xmath25 the bivariate polynomials on the right hand side are permutation polynomials in either variable for every fixed value of the other variable [ riv99 ] . in other words , @xmath23 and @xmath6 are quasi - group operations . furthermore , @xmath23 is a group operation over the set of _ w_-bit numbers . this fact becomes obvious by considering an alternative definition for the @xmath23 operation @xcite : it can be done by dropping the rightmost bit , which is always `` 1 '' , of the product modulo @xmath26 of the operands each appended with an `` 1 '' bit . symbolically , @xmath27/2\;(\mathrm{\textrm{mod}\:}2^{w})\ ] ] the group defined by @xmath23 is thus isomorphic to the multiplicative group of odd integers modulo @xmath26 , via the isomorphism @xmath28 the unit ( i.e. , identity ) element of the group is 0 . the inverse element of _ x _ , denoted @xmath29 , is @xmath30 the following relations are obvious . @xmath31\ ] ] @xmath32\ ] ] since the unary operator @xmath14 is an involution , the following relations hold . @xmath33 @xmath34 notice that 0 is the right unit element w.r.t . the operation @xmath6 . hence @xmath35 which means that @xmath36 is also the right inverse element of _ y _ w. r. t. the @xmath6 operation . since @xmath37 holds for every _ x _ , the following relations hold . @xmath38 @xmath39 let _ e _ be a fixed _ w_-bit number . let s define binary operations @xmath40 and @xmath11 by @xmath41 @xmath42 then @xmath40 and @xmath11 are quasi - group operations over the set of _ w_-bit numbers . this follows from a more general fact that the right - hand side trivariate polynomials are permutations in either variable while keeping the other two fixed [ riv99 ] . actually , the symbols @xmath40 and @xmath11 each defines an entire family of binary operations , of which each is uniquely determined by _ furthermore , from the definition it immediately follows that @xmath40 is a group operation , namely , the group is isomorphic to one defined by @xmath23 via the isomorphism @xmath43 the unit element of the group is _ e_. the inverse element of _ x _ in the group , denoted @xmath44 , is @xmath45\,\boxtimes\,\left[2(x - e)\boxplus1\right]^{-1}\ ] ] simple calculation proves the following relations . @xmath46\ ] ] @xmath47\ ] ] @xmath48 @xmath49 @xmath50 notice that _ e _ is also the right unit element w. r. t. @xmath11 , and @xmath51 also the right inverse element of _ y _ w. r. t. @xmath11 . the operation @xmath11 , which is non - commutative and non - associative , will be used for encryption and , due to the existence of right inversion , also for decryption . we write multi - part data values in _ string _ ( or _ number _ ) notation or _ tuple _ ( or _ vector _ ) notation . in string notation , the value is written as a sequence of symbols , possibly separated by space(s ) that are insignificant . in tuple notation , the value is written as a sequence , in parentheses , of comma - separated symbols . for examples , _ z y x _ and 43 210 are in string notation , @xmath52 and @xmath53 are in tuple notation . the string notation indicates _ high - first _ order : the first ( i.e. leftmost ) symbol denotes the most significant part of the value when it is interpreted as a number . conversely , the tuple notation indicates _ low - first _ order : the first symbol denotes the least significant part of the value when it is interpreted as a number . for examples , to interpret a 3-word number , @xmath54 denotes the most significant word of @xmath55 and @xmath56 denotes the least significant word of @xmath57 . the same value may appear in either notation . thus , for example , for every _ a _ , _ b _ , _ c _ and _ d _ , @xmath58 the term _ part _ introduced above usually refers to `` word '' , but it may also refer to `` digit '' [ of a number ] , `` component '' [ of a tuple or vector ] , as well as group thereof . if , for example _ x _ , _ y _ , _ z _ , _ t _ are 1-digit , 2-digit , 3-digit and 4-digit values respectively , then @xmath59 means @xmath60 , @xmath61 , @xmath62 and @xmath63 . note that the `` string notation '' and `` number notation '' being used as synonyms does not mean that big - endian data ordering is mandated . in order to avoid security irrelevant details , we do not specify endianess . we nevertheless provide a `` reference '' implementations in c++ , where every octet string is considered as a [ generally multi - word ] number with the first octet taken as the least significant one . the implementation thus interprets octet strings as numbers in little - endian order . let @xmath64 denote the permutation that reverses the word order of a non - empty word string . for example , for @xmath65 , ` \mathtt{0x0123abcd}\mathrm{^{r}}=\mathtt{0xcdab2301 } ` let @xmath66 denote the permutation that swaps the high order and low order halves of every word of a non - empty word string . for example , for @xmath65 , ` \mathtt{0x0123abcd}\mathrm{^{s}}=\mathtt{0x1032badc } ` the operator @xmath19 on word strings denote word - wise application of @xmath19 . for example , @xmath67 unless otherwise said , operators @xmath13 and @xmath14 on word strings denote word - wise modular addition and subtraction , respectively . for example , @xmath68 @xmath69 let @xmath70 denote the word - wise application of the inversion operator @xmath70 on a word string . for example , @xmath71 given word strings _ e _ and _ x _ of the same length , let @xmath72 denote the word - wise application of the inversion operator @xmath44 on every word _ x _ of _ x _ with the index - matching word of _ e _ taken as the [ right ] unit element _ e_. for example , @xmath73 operations on word strings are used in this article only to express the decryption function explicitly . this section provides details of nsabc/_w_. from now on _ w _ , the word length , must be even . throughout this article , _ x _ denotes a 4__w__-bit plaintext block , _ y _ a 4__w__-bit ciphertext block , _ z _ a 5__w__-bit key , _ t _ a 4__w__-bit secret _ tweak _ , i.e. , a value that is used to encrypt only one block under the key , _ u _ a _ w_-bit _ unit key _ , i.e. an additional key that generates right unit elements for the underlying quasi - groups . tweaking is optional . it may be disabled by keeping constant ( like _ z _ and _ u _ ) while encrypting many blocks . when tweaking is disabled , nsabc becomes a conventional , non - tweakable , block cipher . mathematically , the cipher is given by two functions , 00.00.0000 which encrypts _ x _ under control of _ z _ , _ t _ and _ u _ , which decrypts _ y _ under control of _ z _ , _ t _ and _ u _ , satisfying the apparent relation @xmath74 the function encrypt is defined in terms of four functions : 00.00.0000 a _ text encryption _ function that encrypts a plaintext block using a _ key schedule _ , a _ unit schedule _ and a _ tweak schedule _ ; a _ key expansion _ function , that generates the key schedule from _ z _ ; a _ unit element _ function , that generates the unit schedule from _ u _ ; and a _ tweak expansion _ function , that generates the tweak schedule from _ t_. 00.00.0000 4__w__-bit plaintext block 5__w__-bit key 4__w__-bit tweak _ w_-bit unit key 00.00.0000 4__w__-bit ciphertext block @xmath75 an explicit relation for encrypt is given in algorithm [ alg : function - encrypt ] . an explicit relation for decrypt is given in algorithm [ alg : function - decrypt ] . mechanically , encryption is performed on a conceptual processor with a 4-word _ text register _ @xmath76 , a 5-word _ key register _ @xmath77 , a 4-word _ tweak register _ @xmath78 and a word _ unit register _ _ u_. the key register is initially loaded with the key _ z_. the tweak register is initially loaded with the tweak _ t_. the unit register is initially loaded with the unit key _ u_. the text register is initially loaded with the plaintext block _ x _ and finally it contains the ciphertext block _ y_. the concrete , vector , notation here specifies the order of words so , for example , @xmath56 is initially loaded with the least significant word of _ x _ and finally it contains the least significant word of _ y_. the text register @xmath76 is initially loaded with the plaintext block _ x _ and finally it contains the ciphertext block _ y_. text encryption proceeds in 32 rounds of operations . a round is of either _ type a _ or _ type b_. the rounds are arranged in four _ passes _ : firstly eight rounds of type a , then eight rounds of type b , then eight rounds of type a again , finally eight rounds of type b again . for _ k_-th round , @xmath79 , the text word @xmath56 is permuted , i.e. it is updated by an execution unit called _ g - box _ that implements a permutation _ g _ on the set of word values , and the contents of the text word @xmath56 are mixed , by exclusive - or ( xor ) , into an other text word that is either @xmath80 or @xmath81 . the order of operations and the target of mixing depend on the round type : * for an a - typed round ( see figure [ fig : rounds ] part a ) , _ g _ applies first , then the mixing takes place and targets @xmath80 . that is , the contents of @xmath56 enters the _ g_-box , the output value of the _ g_-box is stored back to @xmath56 , then the contents of @xmath56 and @xmath80 are xored and the result is stored to @xmath80 . the words @xmath54 and @xmath81 are left unchanged . * for a b - typed round ( see figure [ fig : rounds ] part b ) , the mixing takes place and targets @xmath81 first , then _ g _ applies . that is , the contents of @xmath56 and @xmath81 are xored and the result is stored to @xmath81 , then the contents of @xmath56 enters the _ g_-box , the output value of the _ g_-box is stored back to @xmath56 . the words @xmath80 and @xmath54 are left unchanged . representative rounds . ] besides the text input , the _ g_-box also takes as its inputs an ordered pair of _ w_-bit _ key words _ @xmath82 ( depicted by @xmath83 in fig . [ fig : rounds ] and [ fig : permutation - g ] ) , an ordered pair of _ w_-bit _ unit words _ @xmath84 ( depicted by @xmath85 in fig . [ fig : rounds ] and [ fig : permutation - g ] ) , and a _ w_-bit _ tweak word _ @xmath86 ( depicted by in fig . [ fig : rounds ] and [ fig : permutation - g ] ) . the details on how key words , unit words and tweak words are generated and used will be given in the subsequent subsections . the encryption round is completed with a rotation by one word toward the least significant word on the text register , i.e. the text register is modified by simultaneous loading the word @xmath56 with the contents of the word @xmath80 , @xmath80 with the contents of @xmath54 , @xmath54 with the contents of @xmath81 , and @xmath81 with the contents of @xmath56 . 00.00.0000 4__w__-bit plaintext block 64__w__-bit key schedule 64__w__-bit unit schedule 32__w__-bit tweak schedule 00.00.0000 4__w__-bit ciphertext block for @xmath87 loop if @xmath88 then @xmath89 @xmath90 elsif @xmath91 then @xmath92 @xmath89 end if end loop relations : : @xmath93 for @xmath88 : @xmath94 @xmath95 @xmath96 @xmath97 for @xmath91 : @xmath98 @xmath95 @xmath99 @xmath97 for @xmath100 : @xmath101 @xmath102 @xmath103 @xmath104 @xmath105 the tweak register @xmath78 is initially loaded with the tweak _ t_. the tweak words are generated in 32 rounds of operations . for _ k_-th round , @xmath79 , the value of the word @xmath106 of the tweak register is taken as the tweak word @xmath86 [ which enters the _ g_-box in the _ k_-th encryption round ] . then , similarly to the text register , the tweak register is rotated by one word toward the least significant word ( see figure [ fig : rounds ] ) . 00.00.0000 4__w__-bit tweak 00.00.0000 32__w__-bit tweak schedule @xmath107 for @xmath87 loop @xmath108 @xmath109 end loop relations : : @xmath110 for @xmath100 : @xmath111 @xmath112 @xmath113 @xmath114 @xmath115 @xmath116 note . for @xmath117 , the tweak schedule is @xmath118 the key register @xmath77 is initially loaded with the key _ z_. the key words are generated in 64 rounds of operations . for _ k_-th round , @xmath119 , the value of the word @xmath120 of the key register is taken as the key word @xmath121 [ which enters the _ g_-box of the _ k_/2-th encryption round as the first key word if _ k _ is even , or as the second key word if _ k _ is odd ] . the key register is then rotated by one word toward the least significant word . the rotation is similar to that on the text register and the tweak register . ( see figure [ fig : rounds ] . ) 00.00.0000 5__w__-bit key 00.00.0000 64__w__-bit key schedule @xmath122 for @xmath123 loop @xmath124 @xmath125 end loop relations : : @xmath126 for @xmath127 : @xmath128 @xmath129 @xmath130 @xmath131 @xmath132 @xmath133 @xmath134 note . for @xmath135 , the key schedule is @xmath136 the unit register _ u _ is initially loaded with the unit key _ u_. unit words are generated in 64 rounds of operations . for _ k_-th round , @xmath119 , the value of the unit register _ u _ is taken as the unit word @xmath137 [ which , similarly to the key word @xmath121 , enters the _ g_-box of _ k_/2-th encryption round as the first unit word if _ k _ is even , or as the second unit word if _ k _ is odd ] . the register is then added modulo @xmath17 by the [ key - dependent ] constant @xmath138 to become ready for the next round . ( see figure [ fig : rounds ] . ) 00.00.0000 _ w_-bit unit key 00.00.0000 64__w__-bit unit schedule @xmath139 for @xmath123 loop @xmath140 @xmath141 end loop relations : : @xmath142 for @xmath143 : @xmath144 @xmath145 @xmath146 or , equivalently , for every @xmath147 : @xmath148 note . given unit key _ u _ , the unit schedule is @xmath149 the _ g_-box implements a permutation _ ( see figure [ fig : permutation - g ] ) that takes as argument a text word and is parametrized by an ordered pair of key words @xmath150 , an ordered pair of unit words @xmath151 and a tweak word @xmath152 to return a text word as the result . the _ g_-box operates on a word register that initially contains the argument and finally contains the result . the _ g_-box proceeds in two rounds , each consisting of an operation @xmath11 followed by a half - word swap s. the two rounds are separated by an exclusive - or ( xor ) operation . for the first round , the operation @xmath11 takes the contents of the register as the left operand , @xmath153 as the right operand , and @xmath154 as its right unit element . the result is stored back to the register . the register is then modified by operation s , i.e. swapping the contents of its high and low order halves . for the inter - round xor operation , the register is modified by xoring its contents with the tweak word @xmath152 and storing the result back to it . for the second round , the register is processed similarly to the first round with @xmath155 and @xmath156 being used instead of @xmath153 and @xmath154 , respectively . permutation g.,title="fig : " ] 00.00.0000 _ w_-bit text word pair of _ w_-bit key words pair of _ w_-bit unit words _ w_-bit tweak word 00.00.0000 _ w_-bit text word @xmath157 @xmath158 @xmath159 @xmath160 @xmath158 @xmath161 relation : : @xmath162 [ [ notes ] ] notes + + + + + 1 . the cipher uses 64 distinct instances from the family of operations @xmath11 . alternatively , it may be seen as using 64 identical instances of the single operation @xmath6 or @xmath23 , but operands and result of each instance are `` biased '' by adding or subtracting the constant @xmath154 ( or @xmath156 ) that is specific to the instance , and furthermore , being seen as @xmath23 , the left operand enters and the result leaves it in altered sign . like skipjack , the _ g_-box permutes @xmath153 ( or @xmath155 ) while keeping @xmath4 and other parameters fixed . unlike skipjack , the _ g_-box does nt permute the word @xmath163 where @xmath164 and @xmath165 stand for the high and the low order half respectively . 4 . unlike skipjack , diffusion in the _ g_-box is incomplete , i.e. not every input bit affects all output bits . indeed , the _ v_-th bit of the argument , with @xmath166 , affects only all bits of the low order half and bits _ v _ through @xmath167 of the result ; bits @xmath168 through @xmath169 remain unaffected . 5 . if @xmath170 then _ g _ becomes the identity . decryption can be easily derived from encryption . namely , if @xmath171 then it immediately follows that 00.00.0000 4__w__-bit ciphertext block 5__w__-bit key 4__w__-bit tweak _ w_-bit unit key 00.00.0000 4__w__-bit plaintext block @xmath172 @xmath173 in other words , encrypting the cipher block in reverse half - word order ( @xmath174 ) using the tweak in reverse half - word order ( @xmath175 ) , the unit schedule in reverse word order ( @xmath176 ) , and the key schedule consisting of inverse words of one expanded from the key in reverse word order ( @xmath177 ) , where the inversions are of the quasi - groups defined by the operation @xmath11 and each quasi - group is uniquely given by its right unit that is the index - matching word of the encryption unit schedule in reverse word order ( @xmath176 ) , recovers the plain block in reverse half - word order ( @xmath178 ) . [ [ notes-1 ] ] notes + + + + + 1 . the full cipher is illustrated in figure [ fig : the - full - cipher-1 ] , where @xmath179 , @xmath180 , @xmath135 and @xmath117 ( the unit schedule is omitted ) . the figure is obtained by `` unrolling '' ( i.e. eliminating all rotations of ) the dataflow graph of the full cipher that would be obtained by cascading the individual rounds as in figure [ fig : rounds ] . 2 . the overall structure is up to word indexing identical to that of skipjack . the word re - indexing , which is cryptographically insignificant , was introduced to ease description and illustration . 3 . like skipjack , decryption is similar to encryption . to decrypt with skipjack , one swaps adjacent words in the cipher and the plain block and swaps adjacent word pairs in the key . to decrypt with nsabc , one reverses the word order , i.e. , swaps the first and the last words as well as the second first and the second last ones in the text block , the tweak and the key . for skipjack , one also swaps high and low order halves of every word . for nsabc , one swaps high and low order halves of every word but that of the key . 4 . unlike skipjack , just swapping the words and half - words does nt turn encryption into decryption one needs to invert key words too . thus although encrypt and decrypt can be expressed explicitly in terms of crypt , decrypt can not be expressed explicitly in terms of encrypt . the full cipher , by `` unrolling''.,title="fig : " ] the 4__w__-bit secret tweak _ t _ is used to encrypt only one block [ under a given key _ z _ and unit key _ _ u _ _ ] . in order to encrypt multiple blocks the tweak is derived from the block index and a 4__w__-bit [ additional ] key , called _ tweak key _ , as follows . let @xmath181 denote the tweak used to encrypt _ j_-th block . for the first block ( @xmath182 ) , the tweak key is used as the tweak directly : @xmath183 the subsequent tweak is computed from the current tweak by the recurrent relation : @xmath184 or , equivalently , @xmath185 where all operands are regarded as 4__w__-bit numbers and all operators are defined on 4__w__-bit arithmetic , i.e. mod @xmath186 . [ [ notes-2 ] ] notes + + + + + 1 . the third relation , where @xmath187 conveniently designates the [ unnamed ] tweak key , is meant for random access . the family of functions @xmath188 , parametrized by the tweak key @xmath187 , is not @xmath189-almost 2-xor universal according to definition in @xcite . eventual application of this family in the liskov - rivest - wagner construction , i.e. encryption by @xmath190 , is therefore impossible . 2 . for efficient random access , applications may opt to use non - flat spaces of the block index _ j_. for example , an application that encrypts relational databases may define the index in the format @xmath191 , where @xmath192 is database number , @xmath193 is table number within the database , @xmath194 is row number within the table , @xmath195 is field number within the row and @xmath196 is block number within the field . 3 . tweaking must be disabled when the cipher is used as a permutation , i. e. to generate a sequence of unique numbers . 4 . tweaking should be enabled in all other modes of operation . for example , a non - tweakable block cipher can generate a sequence of independent numbers by encrypting a counter block in cipher block chaining ( cbc ) mode ; nsabc can generate a similar sequence with virtually the same cycle length by encrypting a constant block in a `` tweaked cbc '' mode . an encipherment in nsabc/16 with @xmath197 results in @xmath198 table [ tab : contents - of - registers ] lists states of the [ conceptual ] processor during the encipherment , i.e. the contents of all registers at the start of round _ k _ for _ k _ = 0 , 1 , 2 , ... , 64 for key schedule and unit schedule , and _ k _ = 0 , 1 , 2 , ... , 32 for tweak schedule and text encryption . the start of round 64 ( 32 ) conveniently means the end of round 63 ( 31 ) , which is that of the entire algorithm . = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = unit key register tweak register text register k u z4 z3 z2 z1 z0 k t3 t2 t1 t0 x3 x2 x1 x0 = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = 0 1998 88880777006600050000 0 0001002203334444 0123456789abcdef 1 4cc9 00008888077700660005 2 7ffa 00050000888807770066 1 4444000100220333 388401234567b12f 3 b32b 00660005000088880777 4 e65c 07770066000500008888 2 0333444400010022 1e90388401235bf7 5 198d 88880777006600050000 6 4cbe 00008888077700660005 3 0022033344440001 60ac1e903884618f 7 7fef 00050000888807770066 8 b320 00660005000088880777 4 0001002203334444 499160ac1e907115 9 e651 07770066000500008888 10 1982 88880777006600050000 5 4444000100220333 c2d7499160acdc47 11 4cb3 00008888077700660005 12 7fe4 00050000888807770066 6 0333444400010022 f1efc2d749919143 13 b315 00660005000088880777 14 e646 07770066000500008888 7 0022033344440001 03d2f1efc2d74a43 15 1977 88880777006600050000 16 4ca8 00008888077700660005 8 0001002203334444 273d03d2f1efe5ea 17 7fd9 00050000888807770066 18 b30a 00660005000088880777 9 4444000100220333 1615c2d703d2f1ef 19 e63b 07770066000500008888 20 196c 88880777006600050000 10 0333444400010022 a9b6e7fac2d703d2 21 4c9d 00008888077700660005 22 7fce 00050000888807770066 11 0022033344440001 18c0aa64e7fac2d7 23 b2ff 00660005000088880777 24 e630 07770066000500008888 12 0001002203334444 b049da17aa64e7fa 25 1961 88880777006600050000 26 4c92 00008888077700660005 13 4444000100220333 851857b3da17aa64 27 7fc3 00050000888807770066 28 b2f4 00660005000088880777 14 0333444400010022 71f82f7c57b3da17 29 e625 07770066000500008888 30 1956 88880777006600050000 15 0022033344440001 d5f0abef2f7c57b3 31 4c87 00008888077700660005 32 7fb8 00050000888807770066 16 0001002203334444 e5118243abef2f7c 33 b2e9 00660005000088880777 34 e61a 07770066000500008888 17 4444000100220333 94fae51182433f15 35 194b 88880777006600050000 36 4c7c 00008888077700660005 18 0333444400010022 7bb394fae511f9f0 37 7fad 00050000888807770066 38 b2de 00660005000088880777 19 0022033344440001 11747bb394faf465 39 e60f 07770066000500008888 40 1940 88880777006600050000 20 0001002203334444 d14f11747bb345b5 41 4c71 00008888077700660005 42 7fa2 00050000888807770066 21 4444000100220333 0385d14f11747836 43 b2d3 00660005000088880777 44 e604 07770066000500008888 22 0333444400010022 873b0385d14f964f 45 1935 88880777006600050000 46 4c66 00008888077700660005 23 0022033344440001 cb9b873b03851ad4 47 7f97 00050000888807770066 48 b2c8 00660005000088880777 24 0001002203334444 d6fccb9b873bd579 49 e5f9 07770066000500008888 50 192a 88880777006600050000 25 4444000100220333 d4cf0385cb9b873b 51 4c5b 00008888077700660005 52 7f8c 00050000888807770066 26 0333444400010022 779f53f40385cb9b 53 b2bd 00660005000088880777 54 e5ee 07770066000500008888 27 0022033344440001 8cecbc0453f40385 55 191f 88880777006600050000 56 4c50 00008888077700660005 28 0001002203334444 c93a8f69bc0453f4 57 7f81 00050000888807770066 58 b2b2 00660005000088880777 29 4444000100220333 2e1a9ace8f69bc04 59 e5e3 07770066000500008888 60 1914 88880777006600050000 30 0333444400010022 8038921e9ace8f69 61 4c45 00008888077700660005 62 7f76 00050000888807770066 31 0022033344440001 d4be0f51921e9ace 63 b2a7 00660005000088880777 64 e5d8 07770066000500008888 32 0001002203334444 88b14e700f51921e this section provides methods for efficient software implementation for two types of environment : memory - constrained , such as embedded computers , and memory - abundant , such as servers and personal computers . the function encrypt can be implemented without using any writeable memory on a processor with at least 16 word registers : * 4 for @xmath76 text register * 5 for @xmath77 key register * 4 for @xmath78 tweak register * 1 for _ u _ unit register , * 1 for the constant value @xmath138 , and * 1 for _ k _ round index . indeed , the schedules _ k _ , _ l _ and _ c _ can vanish because every word of them , once produced , can be consumed immediately , provided that the functions ke , te , ue and crypt are programmed to run in parallel and synchronized with each increment of _ k_. source code of this implementation is given in appendix a. unlike encrypt , decrypt needs memory for the key schedule because on - the - fly modular multiplicative inversion is too slow to be practical . in this environment , modes of operation that avoid decrypt ( i.e. ones using encrypt to decrypt ) are thus preferable . the quasi - group operation @xmath11 can be evaluated by only one multiplication and one addition . indeed , @xmath199 where _ x _ is a text word , _ z _ key word , and @xmath200 @xmath201 so , instead of using the @xmath202 pairs , one may pre - compute the @xmath203 pairs once and use them many times . the cipher is parallelizable . the following procedure executes all 32 rounds in 20 steps , of which half performing two or three parallel evaluations of _ g_. recall that @xmath204 is the result of _ g _ in round _ k_. 1 . compute @xmath205 2 . compute @xmath206 3 . compute @xmath207 4 . compute @xmath208 5 . compute @xmath209 6 . compute @xmath210 , @xmath211 in parallel 7 . compute @xmath212 , @xmath213 in parallel 8 . compute @xmath214 , @xmath215 , @xmath216 in parallel 9 . compute @xmath217 , @xmath218 in parallel 10 . compute @xmath219 , @xmath220 in parallel 11 . compute @xmath221 12 . compute @xmath222 13 . compute @xmath223 14 . compute @xmath224 15 . compute @xmath225 16 . compute @xmath226 , @xmath227 in parallel 17 . compute @xmath228 , @xmath229 in parallel 18 . compute @xmath230 , @xmath231 , @xmath232 in parallel 19 . compute @xmath233 , @xmath234 in parallel 20 . compute @xmath235 , @xmath236 in parallel the procedure becomes evident by examining the dataflow graph of the cipher , shown in figure [ fig : the - full - cipher-2 ] , which is obtained by `` unrolling '' the one in figure [ fig : the - full - cipher-1 ] . here `` unrolling '' means introducing a rotation so that the g - boxes with congruent round indices ( mod 3 ) lay on a straight line . the full cipher , by another `` unrolling '' . ] on a x86 - 64 processor in 32-bit mode ( @xmath9 ) , the procedure takes about 256 clock cycles , i.e. 256/16 = 16 clock cycles per byte encrypted . ( the source code of this implementation is given in appendix b. ) in 64-bit mode ( @xmath7 ) , it takes about 384 clock cycles , i.e. 384/32 = 12 clock cycles per byte . the procedure may be also coded twice , i.e. it may be run in two instances in parallel on a single core of the processor , with the second instance delayed by a few steps after the first , to encrypt two blocks possibly under different tweaks and/or keys . this method has shown to be effective for x86 - 64 processors in 64-bit mode , resulting in about 9 clock cycles per byte . we defined nsabc , a block cipher utilizing a group operation that is essentially modular multiplication of machine words , a powerful operation available on many processors . nsabc was meant to be elegant . it uses no s - boxes or `` magic '' constants . it uses only machine word - oriented algebraic operations . it makes use of the simple and regular structure of skipjack which has become publicly known for over a decade sufficient time to be truly understood . it is elegant to be easily memorizable , realizable and analyzable . nsabc bases on some valuable design of a well - reputed agency in the branch . we therefore believe that it is worth analysis and it can withstand rigorous analysis . if this happens to be true , then we may have a practical cipher with 256-bit blocks , allowing to encrypt enormous amount of data under the same key , and with 320-bit keys , allowing to protect data over every imaginable time . in cipher design there is always a trade - off between security and efficiency , and designers always have to ask : `` what do we want , a very strong and fairly fast cipher , or fairly strong but very fast ? '' nsabc reflects the authors view on the dilemma . if skipjack is regarded as very strong and just fairly fast , then nsabc may be regarded as a design emphasizing the second aspect make it very fast , abeit just fairly strong . for _ w_-bit word length , nsabc key length is 5__w _ _ bits , optionally plus 5__w _ _ bits more , whilst the true level of security is yet to be determined . on the other hand , on a modern 64-bit processor it takes only 9 clock cycles to encrypt a byte . .... typedef uint32_t word ; //---------------------------------------------------------------------- static word o(word x , word y , word e ) { return 2*x*y + ( 1 - 2*e)*(x - y + e ) ; } //---------------------------------------------------------------------- static word g(word x , word z0 , word z1 , word u0 , word u1 , word t ) { x = o(x , z0,u0 ) ; x = _ rotl(x,16 ) ; x ^= t ; x = o(x , z1,u1 ) ; x = _ rotl(x,16 ) ; return x ; } //---------------------------------------------------------------------- void encrypt ( word y[4 ] , word const x[4 ] , word const z[5 ] , word const t[4 ] , word u ) { word x0 = x[0 ] , x1 = x[1 ] , x2 = x[2 ] , x3 = x[3 ] , z0 = z[0 ] , z1 = z[1 ] , z2 = z[2 ] , z3 = z[3 ] , z4 = z[4 ] , t0 = t[0 ] , t1 = t[1 ] , t2 = t[2 ] , t3 = t[3 ] , u0 = u , u1 = u0 + 2*u + 1 ; for(int k = 0 ; k < 32 ; k++ ) { if(k & 8) // b - round { x3 ^= x0 ; x0 = g(x0 , z3 , z4 , u0 , u1 , t0 ) ; } else // a - round { x0 = g(x0 , z3 , z4 , u0 , u1 , t0 ) ; x1 ^= x0 ; } word x = x0 ; x0 = x1 ; x1 = x2 ; x2 = x3 ; x3 = x ; word z = z0 ; z0 = z2 ; z2 = z4 ; z4 = z1 ; z1 = z3 ; z3 = z ; word t = t0 ; t0 = t1 ; t1 = t2 ; t2 = t3 ; t3 = t ; u0 = u1 + 2*u + 1 ; u1 = u0 + 2*u + 1 ; } y[0 ] = x0 ; y[1 ] = x1 ; y[2 ] = x2 ; y[3 ] = x3 ; } .... .... void expandkey ( word m[64 ] , word n[64 ] , word const z[5 ] , word u ) { word z0=z[0 ] , z1=z[1 ] , z2=z[2 ] , z3=z[3 ] , z4=z[4 ] ; word u = u ; for ( int k=0 ; k<64 ; k++ ) { m[k ] = 2*(z3 - u ) + 1 ; n[k ] = ( 2*u - 1)*(z3 - u ) ; u + = 2*u + 1 ; word z = z0 ; z0=z1 ; z1=z2 ; z2=z3 ; z3=z4 ; z4=z ; } } //---------------------------------------------------------------------- static inline word g ( word x , word t , word m0 , word m1 , word n0 , word n1 ) { x * = m0 ; x + = n0 ; x = _ rotl(x,16 ) ; x ^= t ; x * = m1 ; x + = n1 ; x = _ rotl(x,16 ) ; return x ; } //---------------------------------------------------------------------- void crypt ( word y[4 ] , word const x[4 ] , word const t[4 ] , word const m[64 ] , word const n[64 ] ) { // step 1 word const g0 = g(x[0 ] , t[0 ] , m[0 ] , m[1 ] , n[0 ] , n[1 ] ) ; // step 2 word const g1 = g(x[1]^g0 , t[1 ] , m[2 ] , m[3 ] , n[2 ] , n[3 ] ) ; // step 3 word const g2 = g(x[2]^g1 , t[2 ] , m[4 ] , m[5 ] , n[4 ] , n[5 ] ) ; // step 4 word const g3 = g(x[3]^g2 , t[3 ] , m[6 ] , m[7 ] , n[6 ] , n[7 ] ) ; // step 5 word const g4 = g(g0^g3 , t[0 ] , m[8 ] , m[9 ] , n[8 ] , n[9 ] ) ; // step 6 word const g5 = g(g1^g4 , t[1 ] , m[10],m[11],n[10],n[11 ] ) ; word const g11= g(g4 , t[3 ] , m[22],m[23],n[22],n[23 ] ) ; // step 7 word const g6 = g(g2^g5 , t[2 ] , m[12],m[13],n[12],n[13 ] ) ; word const g9 = g(g5 , t[1 ] , m[18],m[19],n[18],n[19 ] ) ; // step 8 word const g7 = g(g3^g6 , t[3 ] , m[14],m[15],n[14],n[15 ] ) ; word const g10= g(g6 , t[2 ] , m[20],m[21],n[20],n[21 ] ) ; word const g13= g(g6^g9 , t[1 ] , m[26],m[27],n[26],n[27 ] ) ; // step 9 word const g8 = g(g4^g7 , t[0 ] , m[16],m[17],n[16],n[17 ] ) ; word const g14= g(g4^g10 , t[2 ] , m[28],m[29],n[28],n[29 ] ) ; // step 10 word const g12= g(g5^g8 , t[0 ] , m[24],m[25],n[24],n[25 ] ) ; word const g15= g(g5^g8^g11 , t[3 ] , m[30],m[31],n[30],n[31 ] ) ; // step 11 word const g16= g(g6^g9^g12 , t[0 ] , m[32],m[33],n[32],n[33 ] ) ; // step 12 word const g17= g(g4^g10^g13^g16 , t[1 ] , m[34],m[35],n[34],n[35 ] ) ; // step 13 word const g18= g(g5^g8^g11^g14^g17 , t[2 ] , m[36],m[37],n[36],n[37 ] ) ; // step 14 word const g19= g(g15^g18 , t[3 ] , m[38],m[39],n[38],n[39 ] ) ; // step 15 word const g20= g(g16^g19 , t[0 ] , m[40],m[41],n[40],n[41 ] ) ; // step 16 word const g21= g(g17^g20 , t[1 ] , m[42],m[43],n[42],n[43 ] ) ; word const g27= g(g20 , t[3 ] , m[54],m[55],n[54],n[55 ] ) ; // step 17 word const g22= g(g18^g21 , t[2 ] , m[44],m[45],n[44],n[45 ] ) ; word const g25= g(g21 , t[1 ] , m[50],m[51],n[50],n[51 ] ) ; // step 18 word const g23= g(g19^g22 , t[3 ] , m[46],m[47],n[46],n[47 ] ) ; word const g26= g(g22 , t[2 ] , m[52],m[53],n[52],n[53 ] ) ; word const g29= g(g22^g25 , t[1 ] , m[58],m[59],n[58],n[59 ] ) ; // step 19 word const g24= g(g20^g23 , t[0 ] , m[48],m[49],n[48],n[49 ] ) ; word const g30= g(g20^g26 , t[2 ] , m[60],m[61],n[60],n[61 ] ) ; y[1 ] = g20^g26^g29 ; // step 20 word const g28= g(g21^g24 , t[0 ] , m[56],m[57],n[56],n[57 ] ) ; word const g31= g(g21^g24^g27 , t[3 ] , m[62],m[63],n[62],n[63 ] ) ; y[2 ] = g21^g24^g27^g30 ; // step 21 y[0 ] = g22^g25^g28 ; y[3 ] = g31 ; } //---------------------------------------------------------------------- // multiplicative inverse of x ( mod 2**32 ) , x odd . // source code by thomas pornin , usenet 2009 . word inverse(word x ) { word y = 2 - x ; // xy = = 1 mod 4 y * = 2 - x*y ; // xy = = 1 mod 16 y * = 2 - x*y ; // xy = = 1 mod 256 y * = 2 - x*y ; // xy = = 1 mod 65536 y * = 2 - x*y ; // xy = = 1 mod 4294967296 return y ; } //---------------------------------------------------------------------- void invertkey ( word im[64 ] , word in[64 ] , word const m[64 ] , word const n[64 ] ) { // m , n , i m , in must not overlap ! for(int k=0 ; k < 64 ; k++ ) { im[k ] = inverse ( m[63-k ] ) ; in[k ] = - n[63-k ] * im[k ] ; } } //---------------------------------------------------------------------- void icrypt ( word x[4 ] , word const y[4 ] , word const t[4 ] , word const im[64 ] , word const in[64 ] ) { word xrs[4 ] , trs[4 ] ; trs[0 ] = _ rotl(t[3],16 ) ; trs[1 ] = _ rotl(t[2],16 ) ; trs[2 ] = _ rotl(t[1],16 ) ; trs[3 ] = _ rotl(t[0],16 ) ; xrs[0 ] = _ rotl(y[3],16 ) ; xrs[1 ] = _ rotl(y[2],16 ) ; xrs[2 ] = _ rotl(y[1],16 ) ; xrs[3 ] = _ rotl(y[0],16 ) ; crypt ( xrs , xrs , trs , i m , in ) ; x[0 ] = _ rotl(xrs[3],16 ) ; x[1 ] = _ rotl(xrs[2],16 ) ; x[2 ] = _ rotl(xrs[1],16 ) ; x[3 ] = _ rotl(xrs[0],16 ) ; } //---------------------------------------------------------------------- // testing against the reference implementation void test ( ) { int const ntimes = 10000 ; int const nrep = 100 ; word x[4 ] , y[4 ] , t[4 ] , z[5 ] , m[64 ] , n[64 ] , im[64 ] , in[64 ] ; for(int i=0 ; i<5 ; i++ ) z[i ] = random_word ( ) ; for(int i=0 ; i<4 ; i++ ) t[i ] = random_word ( ) ; word u = random_word ( ) ; // correctness of the optimized implementation expandkey ( m , n , z , u ) ; for(int n = ntimes ; n ; n-- ) { for(int i=0 ; i<4 ; i++ ) x[i ] = random_word ( ) ; memcpy ( y , x , sizeof(x ) ) ; for(int m = nrep ; m ; m-- ) { encrypt ( x , x , z , t , u ) ; crypt ( y , y , t , m , n ) ; } if ( memcmp(y , x , sizeof(x ) ) ! = 0 ) cout < < " crypt : incorrect encryption ! " < < endl ; } // invertibility of the optimized implementation invertkey ( i m , in , m , n ) ; for(int n = ntimes ; n ; n-- ) { for(int i=0 ; i<4 ; i++ ) x[i ] = random_word ( ) ; memcpy ( y , x , sizeof(x ) ) ; for(int m = nrep ; m ; m-- ) crypt ( y , y , t , m , n ) ; for(int m = nrep ; m ; m-- ) icrypt ( y , y , t , im , in ) ; if ( memcmp(y , x , sizeof(x ) ) ! = 0 ) cout < < " icrypt : incorrect decryption ! " < < endl ; } } //////////////////////////////////////////////////////////////////////// ....

the properties of the bright and energetic type ic sn 1997ef are investigated using a monte carlo spectrum synthesis code . analysis of the earliest spectra is used to determine the time of outburst . the changing features of the spectrum and the light curve are used to probe the ejecta and to determine their composition , verifying the results of explosion calculations . since synthetic spectra computed using our best explosion model co100 are only moderately good reproductions of the observations , the inverse approach is adopted , and a density structure is derived by demanding that it gives the best possible fit to the observed spectrum at every epoch analysed . it is found that the density structure of model co100 is adequate at intermediate velocities ( 500025000 km s ) , but that a slower density decline ( ) is required to obtain the extensive line blending at high velocities ( 2500050000 km s ) which is the characterising feature of this and other energetic type ic supernovae . also , the inner ` hole ' in the density predicted by the model is found not to be compatible with the observed evolution of the spectrum , which reaches very low photospheric velocities at epochs of about 2 months . the ` best fit ' density distribution results in somewhat different parameters for the sn , namely an ejecta mass of 9.6m ( v. 7.6m in co100 ) and an explosion kinetic energy oferg ( v.erg in co100 ) . this revised value of the kinetic energy brings sn 1997ef closer to the value for the ` prototypical ' type ic ` hypernova ' sn 1998bw . the abundance distribution of model co100 is found to hold well . the modified density structure is used to compute a synthetic light curve , which is found to agree very well with the observed bolometric light curve around maximum . the amount of radioactiveni produced by the sn is confirmed at 0.13m . in the context of an axisymmetric explosion , a somewhat smaller kinetic energy than that of sn 1998bw may have resulted from the non alignment of the symmetry axis of the sn and the line of sight . this might also explain the lack of evidence for a gamma ray burst correlated with sn 1997ef .

astro-ph0007222

sn 1997ef in ugc4107 was recognised as a peculiar and interesting object as soon as its first spectra were taken by the harvard - cfa team ( garnavich et al . 1997a , b , c ) . the spectra displayed very broad features , quite unlike those of any other sn known to date , so that it was not even clear whether what was observed was an absorption or an emission spectrum . continued observation revealed a light curve typical of a sn deriving from a compact object ( a sn ia or a sn ic ) , but much broader than the templates for both of these classes . later spectra showed more resolved spectral lines , clarifying that the broad features were actually very extended absorption blends , and that the sn was therefore in the photospheric epoch . lines of ca ii , o i and fe ii were strong , and the si ii 6347 , 6371 line appeared to be comparatively weak , so the sn was classified as type ic , as also supported by the overall similarity with the spectra of other sne ic such as sn 1994i ( filippenko 1997 ; millard et al . 1999 ) . typical models for a low mass , low kinetic energy sn ic ( nomoto et al . 1994 ; iwamoto et al . 1994 ) gave good fits to the light curve , but the synthetic spectra computed using such models , while yielding the correct types of spectral lines , completely failed to reproduce the observed large line width , which we estimated as at least 20000 km s@xmath0for some of the strongest lines ( iwamoto et al . 2000 , hereafter paper i ) . soon after sn 1997ef , however , the extraordinarily bright and powerful type ic sn 1998bw was discovered as the optical counterpart of grb980425 ( galama et al . a great deal of study , both observational and theoretical , was and still is being devoted to this sn , which was soon realised to be a massive type ic event of exceptionally large kinetic energy ( @xmath6 ergs , iwamoto et al . 1998 ; woosley , eastman & schmidt 1999 ; branch 2000 ) , where the link with the grb may lie in a rather asymmetric explosion ( macfadyen & woosley 1999 , hflich et al . 1999 , khokhlov et al . 1999 , maeda et al . it was only natural then also to analyse sn 1997ef as a powerful ` hypernova ' . paper i presents a detailed study of the light curve and the spectrum using two models , one for a ` normal ' sn ic ( model co60 : m@xmath7= 4.4m@xmath2 , ke = 10@xmath8 erg ) and another for a ` hypernova ' ( model co100 : m@xmath7= 7.6m@xmath2 , @xmath9 erg ) . both models have a @xmath5ni mass of 0.15m@xmath2 . in paper i we showed that both models reproduce the light curve reasonably well , but only the more massive and energetic model co100 can also produce reasonable synthetic spectra , and therefore they opted for the ` hypernova ' model and suggested that sn 1997ef may be a somewhat less extreme case than sn 1998bw . a possible connection with a grb has been searched for in archives for sn 1997ef . the result was that grb971115 could be compatible with sn 1997ef in position and time of occurrence ( wang & wheeler 1998 ) , but the statistical significance of the correlation is much weaker than for sn 1998bw and grb980425 . in this paper , we concentrate on the spectroscopy of sn 1997ef . this is interesting in many respcts . first , it is important that we not only fit the spectra and their evolution using one explosion model , but that we also understand which lines are present in the spectrum . since we have available a sufficiently large number of observed spectra , we can follow the spectral evolution and thus probe different depths in the ejecta and verify the composition as a function of velocity , and also describe how the spectrum changes as a function of luminosity as the sn moves along the light curve . we use for this purpose the monte carlo ( mc ) code originally described in mazzali & lucy ( 1993 ) and further improved and discussed by lucy ( 1999 ) and mazzali ( 2000 ) . the code includes the effect of photon branching , which must be very important in a situation where the high expansion velocities lead to extensive line blanketing . also , since the code uses the bolometric luminosity as input , we can compare the required @xmath10 values with those of a light curve calculation relaxing the assumption @xmath11 made in paper i and thus improve the comparison between model @xmath12 from light curve calculations and observed @xmath13 mags . furthermore , we use spectrum synthesis to date the earliest spectra , and thus to determine with some accuracy the time of explosion , which is important for both the grb connection and the light curve calculation . finally , having noticed that even synthetic spectra obtained from co100 do not give wonderfully good fits to the observed spectra , we adopt the ` inverse ' approach and try to determine the density and abundance distribution with velocity by looking for a best fit to the spectra . this method was pioneered by branch ( 2000 ) , who suggested for sn 1997ef a much flatter density dependence ( @xmath14 ) than what is predicted by model co100 ( where the density index is @xmath15 . consequently , for sn 1997ef branch ( 2000 ) obtains a much larger ejecta mass ( 6m@xmath2 above @xmath16 km s@xmath0 ) and kinetic energy ( @xmath17erg above the same velocity ) . since we study several spectra we can control @xmath18 as a function of @xmath19 , and hence build our own @xmath20 . we do confirm that the spectra require a flatter density dependence , but we show that this is only necessary at high velocity ( @xmath21 km s@xmath0 ) . examining spectra at advanced epochs , we also show that the predicted density ` hole ' at low velocities is not compatible with observations . we therefore derive new values for both m@xmath7and @xmath22 . we take this process one step further by computing a synthetic light curve from our derived @xmath20 , and show that this gives a very good fit to the observed light curve of sn 1997ef . in the rest of this paper we first describe how we proceeded to date the spectra , then discuss the analysis of 6 spectra , reviewing the results and their implications . then we review the properties of the model for sn 1997ef as we derived it , and discuss how it differs from model co100 , which was computed to fit the light curve , not the spectra . we also show the synthetic light curve obtained from our ad hoc model for @xmath20 . finally , we review and discuss our findings , including possible reasons why sn 1997ef was a hypernova but no grb was apparently linked to it . we have selected for modelling 6 spectra of sn 1997ef obtained by the harvard - cfa group ( garnavich et al . in prep . ; see paper i , figs . 8 and 9 ) . we have chosen well exposed , high s / n spectra , with a large wavelength coverage , and tried to sample the light curve as uniformly as possible . the first three spectra are from epochs around maximum ( which was not observed either spectroscopically or photometrically but was probably reached around dec 10 , 1997 at @xmath23 . the dates of the spectra are nov 29 , which is very soon after discovery , dec 5 and dec 17 . all the spectra show very broad lines . they are all early enough that they are sensitive to changes in the kinetic energy and to the assumed epoch of outburst . the next two spectra have dates dec 24 and jan 1 , which is around the time when the sn enters the tail of the light curve . the lines in these spectra becoming progressively narrower , indicating that the high velocity part of the ejecta is becoming optically thin . the final spectrum has date jan 26 , 1998 , which is well on the light curve tail . line velocities are quite small , suggesting that the spectrum is formed deep in the ejecta . one of the challenges for a good light curve calculation is to follow the evolution of the photospheric velocity with time . even model co100 was not perfect in this respect ( cf . paper i , figure 7 ) . since the epoch of maximum was not observed , light curve models are not very tightly constrained . since the rise time to maximum depends on the structure of the model and the abundance distribution , the range of allowed parameters could be effectively costrained if at least the time of explosion could be determined , albeit with some uncertainty . this could be attempted using spectrum synthesis and the measured photospheric velocity as determined in paper i from the velocity of the si ii line . it could be expected that this line is formed somewhat above the photosphere at very early epochs , like in sne ia ( e.g. sn 1994d , patat et al . 1996 ) . since our synthetic spectra are computed using the luminosity @xmath10 , the epoch @xmath24 and the photospheric velocity @xmath25 as input , an appropriate value of @xmath24 must be found that gives a photospheric radius @xmath26 which , when combined with @xmath10 , produces temperatures in agreement with the observed spectrum . in particular , although the continuum is not well determined in the very complex spectra of sn 1997ef , spectral lines serve a double role : they depend on the temperature ( and hence on @xmath27 and @xmath10 ) through their ratios , and on @xmath25 through their displacement . not many lines are visible in the earliest spectra , but at least si ii 6346 , 6371 and the broad complex which can most likely be identified as o i 7774 , o i8447 and the ca ii ir triplet could be used to guide the calculation of the synthetic spectra . models for nov 29 show that acceptable fits are found for @xmath28 days and @xmath29 km s@xmath0 . greater epochs require smaller velocities , so that @xmath27 always has a value of about @xmath30 cm . this radius , combined with the value of @xmath10 , which depends almost entirely on the assumed distance , gives a temperature appropriate to get a good overall fit to the spectrum . if we assume e.g. @xmath31 days , then we find a good overall fit to the spectrum for @xmath32 km s@xmath0 , which is too small and is reflected on the line shift . the opposite happens if @xmath33 days . in order to strenghten our result , we applied the same technique to the dec 5 spectrum , and we found acceptable fits for @xmath34 days and @xmath35 km s@xmath0 . therefore , the two epochs give consistent results , and we can estimate the time of outburst to have been nov 2022 . since the fits obtained for @xmath36 days on nov 29 and @xmath37 days on dec 5 were particularly good , we selected nov 20 as our reference epoch , also for the comparison with the light curve and the photospheric velocities . we will show the fits in the next sections , when discussing the various epochs . in figure 1 we show as an example the nov 29 spectrum and two synthetic spectra computed as above for @xmath36 days ( dashed line ) and @xmath31 days ( dotted line ) , respectively , using model co100 . the difference in velocity is clearly visible . these results depend on the selection made for two other quantities , distance and overall ejecta mass , as discussed by mazzali & schmidt ( 2000 ) in the context of type ia sne . for all models , we adopted a distance modulus of 33.63 , i.e. a distance of 53.1 mpc , as estimated from a recession velocity of 3450 km s@xmath0(p . garnavich , 2000 , private communication ) and @xmath38 km s@xmath0mpc@xmath0 . we assume no extinction ( @xmath39 , which is justified by the absence of a narrow interstellar na i d line from the spectrum of sn 1997ef at all phases . as for the ejected mass , we already showed in paper i that this must be large in order to fit both the light curve and the spectra . therefore in this paper we use the solution found in paper i , model co100 , as a reference . this is to our knowledge the earliest spectrum available of sn 1997ef . although it is somewhat noisy and it does not extend very far to the red , it nevertheless includes most of the main features , which made sn 1997ef such a peculiar and interesting object . the spectrum ( figure 2 ) has essentially no continuum , and it is characterised by four broad minima , at roughly 3900 , 4700 , 6000 and 7500 . the minima are separated by three rather sharp peaks , at 4300 , 5200 and 6300 , respectively . the minima are so broad and the peaks so strong that the spectrum could be confused for an emission spectrum . both branch ( 2000 ) and paper i showed that the real nature of the spectrum is a superposition of absorption lines . we already pointed out that the ` emission peaks ' are the result of photon travel in a medium of large line opacity . photons redshift their way through the envelope until they find a portion of the spectrum where line opacity is low because of the intrinsic distribution in wavelength of the atomic transitions . we call such a region a ` line - free window ' . a large number of photons escape through these regions - all the photons coming from the optically thick region immediately to the blue , and this leads to a large flux and to the observed ` peaks ' . while the width of the peak depends solely on the width of the line - free window , its strength depends both on the flux at the lower boundary ( the photosphere ) and on the width of the blanketing region which feeds photons to it : the broader the blanketing region , the more photons must reach the window to escape , resulting in a higher peak flux . this can be easily seen by inspection of the spectrum . in paper i we showed that a synthetic spectrum based on model co100 gives a reasonably good reproduction of the observations , and allows us to identify the four absorptions as follows : 3900 : ca ii h&k + ni ii , co ii and si iilines ; 4700 : mostly fe ii lines ; 6000 : si ii 6347 , 6371 plus weaker si ii lines and na i d further to the blue ; 7500 : o i 7774 . in spectra computed with model co100 the o i 7774 and the ca ii ir triplet are well separated , but unfortunately the observed spectrum does not extend far enough to the red to show the ca ii region . with reference to that synthetic spectrum , which is shown again here as the dashed line in figure 2 , we notice that none of the absorptions extend quite far enough to the blue . for the two bluer troughs , which are blends of very many lines , this may possibly be due to the lack of lines resulting from neglecting some element or not using a sufficiently complete line list , but this seems unlikely for the two redder troughs , where a few very strong lines dominate . the synthetic spectrum has @xmath36 days , @xmath40 ( erg s@xmath0 ) , @xmath41 km s@xmath0 . the observed @xmath13 magnitude ( 16.7 ) and the synthetic one ( 16.75 ) agree quite well , but the spectrum has a large bolometric correction ( 0.28 ) , which indicates that the assumption @xmath42 made in paper i when fitting the light curve is at least weak . the observed si ii line velocity is 19000 km s@xmath0 , and the core of the si ii absorption is well reproduced even using model co100 . therefore , in order to improve the synthetic spectrum , we modified the density only above 20000 km s@xmath0 . while the original co100 model has @xmath43 in the outer part , we made the outer density radial dependence flatter . as a consequence , high velocity absorption becomes stronger . of course , this happens first in the strongest lines , the ca ii h&k doublet and ir triplet first and foremost . the strength of these lines increases dramatically , because they come from low excitation levels which are still highly populated even at the very low densities and temperatures of the outer envelope ( @xmath44k ) , and so these lines increase greatly in width . next come the si ii and o i lines . for these lines , the drop in the level population with radius is much steeper , since they come from levels on average about 10ev above the ground state , and so they do become broader , but not so noticeably , and their cores do not shift much towards the blue . all other lines , strong and weak are similarly affected , to a higher or lesser degree depending on the ionisation and excitation of the levels . since we wanted to introduce as little change from model co100 as possible , we choose the steepest density profile which gave a good fit to the spectrum . this was obtained with a density power law index @xmath45 at @xmath21 km s@xmath0 . when this change is introduced the ca ii h&k doublet , the si ii line and the o i7774 line broaden to the point that the observed absorption troughs are very well reproduced . further to the red , in the unobserved part of the spectrum , the ca ii ir triplet is predicted to broaden and blend with o i 7774 . the contribution of o i 8446 , which is also a strong line , is nevertheless small compared to that of ca ii . the blending of o i 7774 and the ca ii ir triplet is an important discriminant for the density structure at high velocities . fortunately , the dec . 5 spectrum covers that region . the blue sides of the peaks at 4300 and 5000 are also fitted much better , as the increased density causes the emission to set in at higher velocities . the only region where no improvement is seen is near 4500 . the fe ii and co ii lines causing the absorption near 4750 are not sufficiently strong to extend to the blue , and there are not enough strong lines that absorb near 44004500 . it is possible that we are either missing lines or ignoring some relevant element . interestingly , branch ( 2000 ) , who used a code that is very efficient for identifying lines , seemed to have the same problem . the modification to the outer density profile has only a small influence on the total mass , since it adds only about 0.25m@xmath2 in the outer ejecta , but it does lead to a significant increase in the overall kinetic energy , since all the mass is added to the high velocity shells , where it represents a dramatic increase , of 12 orders of magnitude . the kinetic energy above 20000 km s@xmath0 increases from @xmath46 to about @xmath47erg , and the total kinetic energy increases from @xmath48 to @xmath3erg . this is quite a large change , but it is still smaller than the value adopted for sn 1998bw . also , both the outer slope and the revised @xmath22 are , respectively , steeper and smaller than the values suggested by branch ( 2000 ) . we will discuss the implications of these revised values later . an increase in @xmath22 only makes sn 1997ef more like a hypernova . the best fitting ad hoc model has the same input parameters as the spectrum computed using co100 . the increased outer density makes the temperature only slightly higher . the synthetic spectrum has @xmath49 , @xmath50 and @xmath51 , so that @xmath52 . therefore comparing @xmath53 with @xmath13 is misleading , since @xmath13 rises more rapidly than @xmath53 . finally , a word of comment about the presence of he in the spectrum . in sn 1998bw , he i 10830 was almost certainly observed in absorption at early epochs ( patat et al . 2000 ) . its presence could be understood if the he shell was not completely lost before the explosion . unfortunately , ir spectra are not available for sn 1997ef , so the best place to look for he is he i 5868 . this line is very close to na i d , which is rather strong and contributes to the broad 6000 trough with an absorption near 5450 . a small he mass does not produce a strong line unless very large departure coefficients are assumed , and in any case a strong he i line is not necessary to fit the spectrum . therefore , although we can not be final , we do not think he i 5868 is present in the nov 29 spectrum . the second spectrum in our analysis has a good s / n and a very good wavelength coverage , extending from 3100 to 8700 . it is actually quite similar to the 29 nov . one , but more detail is visible ( figure 3 ) , e.g. the structure of the complex 6000 absorption and , very importantly , the ca ii ir triplet is visible , and clearly blended with o i 7774 , which is a strong argument in favour of a ` flat ' outer density distribution , as discussed above . that the feature measured at 7500 is a blend is clear , given the presence of two distinct absorptions , whose wavelengths match those of the two component multiplets , as shown also by the spectrum computed with model co100 . the synthetic spectrum based on model co100 was shown in paper i and again here as the dashed line in figure 3 . it has parameters @xmath37 days , @xmath54 ( erg s@xmath0 ) , @xmath55 km s@xmath0 . both @xmath10 and @xmath27 have values similar to those on nov . 29 , therefore the temperatures are similar and so are the two spectra , although the photosphere is quite a bit deeper on dec . this is shown also by the si ii line velocity , which is reduced to 13000 km s@xmath0 . the observed @xmath13 mag on dec 5 was 16.5 , and the co100 spectrum reproduces it well ( @xmath56 ) . the problem with that model is essentially the same as that discussed for the nov 29 model : all troughs do not extend far enough to the blue . furthermore , o i 7774 and the ca ii ir triplet do not blend anywhere near as much as they should when the original model is used . in that spectrum the ca ii ir triplet extends only to about 7850 , i.e. a ca iivelocity of @xmath57 km if the ca ii ir triplet is to blend with the o i line , it must reach at least 7650 in the blue wing , i.e. @xmath58 km s@xmath0 . this is consistent with the modified density structure adopted for the nov 29 spectrum . since @xmath25 is much lower than the values at which the density has been modified ( @xmath59 km s@xmath0 ) , we can use the same density structure as for nov 29 . the spectrum we computed for the modified structure and the same input parameters as above is shown in figure 3 . the improvements are apparent , and they concern mostly the ca ii ir triplet + o i 7774 feature in the red and the blue region near 4000 , which is impoved by the strengthening and broadening of ca ii h&k . improvements in the trough at 4700 are again only marginal , as are those to the si ii line region . surprisingly perhaps , the si ii line does not become broader , but this can be understood since the high velocity regions where the density has been enhanced are far above the photosphere and are of too low density now to cause much si ii absorption . the situation is made more complicated by the apparent development of an absorption shoulder near 5900 . we can not offer a clear identification for this shoulder , except for several lines of o i at 6157 , but these are not very strong . the shoulder was not present on nov 29 , nor is it on dec 17 . another , stronger shoulder is visible near 5750 . this is where si ii 5958 , 5979falls , and the synthetic spectra show that this doublet , combined with na i d further to the red , is strong enough to produce a shoulder stronger than the observed one without resorting to he i. therefore we can confirm the non - detection of he i 5876 . note that the si abundance in this model was lower than on nov 29 . this was necessary to fit the core of the si ii 6347 , 6371 line . even the modified density does not give a perfect synthetic spectrum , but the excellent behaviour in the red confirms that we are adopting the correct measures when using an extended outer envelope . if a density law flatter than @xmath60 is adopted in an attempt to improve the si ii line two problems emerge : 1 ) o i 7774 extends too far to the blue and 2 ) the increasing strength of ca ii h&k causes too much blocking of the uv spectrum and , consequently , a synthetic peak at 4500 much stronger than what is observed . these problems would be even worse in the nov 29 spectrum , and so we chose to retain the @xmath60 solution as a compromise and suggest that some alternative line may be the reason for at least the shoulder at 5900 . our best - fitting spectrum has @xmath61 , @xmath62 and again a large @xmath63 , so that @xmath64 . the increased @xmath65 is apparent if one looks at the ratio of the two main peaks , at 4300 and 5200 . therefore , @xmath53 is significantly smaller than @xmath13 at this epoch , which is near maximum in the @xmath13 light curve but appears to be still very much on the rising branch of the bolometric curve . this third spectrum has again a limited wavelength coverage , but it is the first one available after dec 5 , and it is therefore useful to study the evolution of the sn immediately following @xmath13 maximum . the spectrum , which is displayed in figure 4 has @xmath66 . the overall aspect of the spectrum is similar to the previous two epochs , which indicates that the temperature conditions have probably not changed greatly . however , closer inspection , and in particular a wavelength comparison of this and the two earlier spectra , reveals that all the main absorption and emission peaks have shifted to the red by 100 - 200 , depending on the wavelength . a similar , but smaller shift also occurred between nov 29 and dec 5 . the immediate interpretation is that we are observing the effect of the inward motion of the photosphere through the ejecta as they expand and thin out . note that this recession takes place essentially in the lagrangian mass / velocity frame , and not in the observer s ( radius ) frame , much like in the plateau phase of some sne ii . the si ii velocity measured from this spectrum is in fact only 8000 km s@xmath0 . the change in line absorption velocity is larger between dec 5 and dec 17 than it was between nov 29 and dec 5 because the time interval between the spectra is larger . all absorptions move by about the same amount , and the emissions move as well . this is obvious given our explanation for what the emissions really are : if the blanketed regions move , the opacity windows must move along with them , just like in p - cygni profiles caused by a single line , a situation of which the case of sn 1997ef represents after all just an extreme extension . looking in more detail , the shoulder which affected the si ii line near 5900 on dec 5 seems to have disappeared on dec 17 , leaving serious questions about its identity . the absorption near 5000 now shows more structure , with minima at 4800 and 4950 and two shoulders at 5100 and 5300 , which resembles quite closely the typical profile observed in sne ia near maximum and soon thereafter . in sne ia the entire feature is attributed mostly to fe ii lines , and this was confirmed in sn 1997ef on both nov 29 and dec 5 . the synthetic spectrum computed with co100 is shown as the dashed line in figure 4 . the structure in the fe ii absorption is reproduced reasonably well , and the width problem now does not affect either the feature at 4300 or that near 7500 . this is probably because the outer high velocity regions are so far above the photosphere now that they do not really cause any significant absorption , and so the original model co100 gives an acceptable description of the density structure in the intermediate velocity regions . unfortunately this spectrum does not cover the ca ii ir triplet , which is predicted not to blend with o i 7774 in the co100 model . the synthetic spectrum was computed with the following parameters : @xmath67 days , @xmath68 ( erg s@xmath0 ) , @xmath69 km s@xmath0 . the temperatures are lower than on dec 5 but comparable to those on nov 29 , and @xmath27 is actually larger than on both previous epochs , although the mass above the photosphere is now as large as 5m@xmath2 . compared to the dec 5 model , the abundance of o is reduced and that of s increased , which is the behaviour expected as deeper and deeper regions of the ejecta are probed . the reduced velocity is the reason for the de - blending of many features . there are two major shortcomings in the model : 1 ) the position of the peak at 5500 is not correctly reproduced . this is clearly the consequence of the absorption at 5400 not being correctly reproduced , so that the ` optical depth window ' starts too far to the blue . this feature is not well reproduced in the spectra of sne ia either , so it is quite likely that some fe - group lines are missing or their strength is not correct . 2 ) the si iiline is too narrow , or rather the very triangular shape of the blue side of the absorption can not be reproduced . only another si ii line is active in that region ( si ii 5979 , observed near 5800 ) , but it is too weak . in any case it would be difficult to explain the observed profile with just one line , so we really have no explanation for what might affect the spectrum . we have also modelled the spectrum using our modified density distribution ( figure 4 , continuous line ) . as we could expect , since the near - photospheric region is not affected by the modification , at this epoch the two synthetic spectra are essentially identical . the spectrum computed with the modified outer density profile has @xmath70 and @xmath71 . the bolometric correction is still large , @xmath72 , and @xmath73 . it is very interesting that the model luminosity , as obtained from fitting the spectrum , is actually larger on dec 17 than on dec 5 . the bolometric light curve appears then quite different from both the @xmath13 and the @xmath74 curves . the large bolometric correction is now caused mostly by the red part of the spectrum , with the synthetic @xmath75 band still rising so that the bolometric light curve rises slowly to a rather delayed maximum . near - ir photometry would have been useful to confirm this prediction . the 24 dec spectrum is the first of a set of three spectra modelled in this paper which extend the time coverage beyond the near - maximum epochs considered also in paper i. the 24 dec spectrum is shown in figure 5 . it does not extend far to the red , but in the blue it has a good s / n down to 3700 , thus covering the ca ii h&k region . the epoch of the spectrum is 35 days after explosion , for which we chose nov 20 . the spectrum has @xmath76 , a drop of 0.2 mag in only one week from dec 17 , and lies therefore well in the declining part of the @xmath13 light curve . the si ii line is now quite narrow , and its core has a velocity of 5400 km s@xmath0 . this and all other absorptions continue to move to the red . following the trend already noticeable on dec 17 , many small features which earlier blended into the broad troughs are beginning to be resolved , marking the progressive inward motion of the photospere . the broad absorption near 6000 is the region that changed the most with respect to dec 17 . in particular , a separate absorption is now visible at 5700 . we modelled the spectrum using model co100 with the outer modification introduced as discussed above . as usual , @xmath10 and @xmath25 were selected so as to get as good a fit as possible to the observations . the input parameters for the best model , which is shown as the continuous line in figure 5 , were @xmath77 days , @xmath40 ( erg s@xmath0 ) and @xmath78 km s@xmath0 . note that @xmath10 has the same value as on nov 29 , indicating that the bolometric light curve is now declining . the photospheric velocity is now very low , as is also indicated by the si ii line velocity , and is actually slightly below the inner edge of the density distribution of model co100 ( cf . paper i , figure 3 ) . therefore , in order to produce a reasonable spectrum we had to introduce a second change to the co100 density structure , extending it down to @xmath79 . we used @xmath80 for this inner extension . this is only a first approximation , but it is obvious that if there is a photosphere there must be opacity right down to it , and this is not compatible with the ` density hole ' predicted by model co100 below 5300 km s@xmath0 . since the photosphere is so deep , the inclusion of the flat outer density profile derived from the earlier spectra has no effect on the synthetic spectrum . even when the flat outer density is included , o i 7774 and the ca ii ir triplet are predicted to be separated in wavelength . this region is not covered on dec 24 , but our prediction will be confirmed in the jan 1 spectrum ( see next section ) . the small inner extension to the density does not affect the total mass much , so that at this epoch the total mass above the photosphere is 8.4m@xmath2 . a major change in this model is in the abundances . as shown in figure 7 of paper i , the oxygen - dominated envelope extends down to 6500 km s@xmath0 . on dec 24 the photosphere is well below this point , and si , s and the fe - group have larger abundances . we fitted the abundances to obtain a best fit , and found that s is the element whose abundance must increase most , at the expense of o. the behaviour of the abundances as determined from spectrum synthesis is reviewed in the discussion . most line identifications are the same as on earlier epochs . the absorption at 5700 is due to na i d . overall , the synthetic spectrum is a good match to the observed one . the major shortcoming is that the emission peak near 6500 is not reproduced . net emission may be present there , although the wavelength does not match that of the expected line of [ o i ] 6300 . there is a wavelength coincidence with h@xmath81 , but this is not expected to be a strong feature in a sn ic ejecta . the synthetic spectrum has @xmath82 and @xmath83 . the bolometric correction is still large , @xmath84 , so that @xmath52 . therefore , it appears that bolometric maximum was reached near dec 24 , much later than @xmath13 or @xmath74 maximum . a lot of the flux is still released in the near - ir , in particular in the ca ii ir triplet peak . the next spectrum in our series is rather close in epoch to the previous one but , with an epoch of 42 days , it captures a moment when the light curve begins to settle on the tail . although the s / n is only moderately high , the ample wavelength coverage ( 3400 to 8900 ) makes it an interesting spectrum to model . the spectrum is shown in figure 6 . comparing it to the dec 24 spectrum , several signs of development can easily be noticed : 1 . the ratio of the three peaks in the blue , at 4000 , 4500 and 5500 is much steeper , indicating a shift of the flux towards the red which clearly must be the consequence of a reduced temperature . the absorption complex near 6000 has now split into two components . one is centred near 5700 and is due to na i d , which is now sufficiently strong and isolated because of the reduction in strength of neighbouring lines coming from more highly excited levels that it gives rise to its own p - cygni emission , near 5900 . to the red of na i d the si ii line is still strong , near 6250 , and is accompanied by a weaker absorption near 6100 , which was also present on dec 24 . two absorptions are developing at about 6800 and 7100 . the o i 7774 line and the ca ii ir triplet are now well separated , as was already predicted by our synthetic spectrum on dec 24 . the o i line is rather narrow , while the ca ii absorption is broader . this is a consequence of the different run of excitation with radius ( excitation falls more steeply for o than for ca , and so does the line optical depth ) . 5 . emission in the ca ii ir triplet is now very strong , much stronger than the corresponding absorption : this indicates that net emission is beginning to be an important contribution to spectrum formation . the line blueshift is further reduced : the si ii line now has a measured @xmath85 km s@xmath0 . this is now significantly smaller than than the inner boundary of the original co100 density distribution . since the essentially photospheric nature of the spectrum is proved by the progressive evolution of all features , a viable model must include a low velocity extension to the density profile . therefore , we modelled the spectrum using as a starting point the co100 density distribution modified at high velocity as described earlier . this modification has no effect on the emerging spectrum at these later epochs : the maximum observable matter velocity , measured at the blue edge of the ca ii ir triplet is only 23000 km s@xmath0 . we extended the density distribution inwards to @xmath86 km s@xmath0 , and tried various density laws below 5000 km s@xmath0 . our first conclusion is that the density must be increasing inwards : if we use a decreasing density a shell forms near 5000 km s@xmath0which affects the lines ; if the density is constant below 5000 km s@xmath0the value near the photosphere is too low and the lines are too weak . on the other hand , if the density increases too steeply the synthetic lines tend to be too sharp . as a best approximate solution we adopted a power law with @xmath80 . our final model has parameters @xmath87 days , @xmath88 ( erg s@xmath0 ) , @xmath89 km s@xmath0 , which is close to the observed si ii velocity . it is interesting that the simultaneous reduction in both @xmath10 and @xmath25 means that the electron temperature @xmath90 ranges between 4500 and 6500k , which is close to the values it had on previous epochs . this explains why the naure of most line features is unchanged throughout the evolution of the sn . the synthetic spectrum is shown as the thin line in figure 6 . overall , the quality of the fit is rather good , confirming the correctness of our assumption . major shortcomings are : 1 ) the strength of the peak at 4500 . this is probably due to having somewhat too much opacity in the uv . 2 ) the strength and position of the peak near 5500 . this was a problem also on dec 24 and we explained it as the consequence of missing line strength near 5400 . 3 ) flux is missing in the ca ii ir triplet emission . this is because net emission - following collisional excitation - is not included in our code . the difference between the observed and synthetic peak is to be ascribed to net emission , which is responsible for about half of the total line flux . other parts of the spectrum are very well reproduced , e.g. the o i and ca ii ir triplet absorption , the structure of the si ii line region , where the weak absorption at 6000 is attributed to fe ii 6148 and the emission near 6500 is now correctly reproduced , confirming that the peak on 24 dec was probably not [ o i ] 6300 . the weak absorptions at 6800 and 7100correspond to lines of o i 7001 and to several fe ii lines , respectively , but the synthetic lines are too weak . fe ii lines are becoming stronger as the abundance of fe - group elements is higher , while that of o is further decreased . the inward extension of the density affects the ejecta mass , adding about 1.5m@xmath2 between 3000 and 5000 km s@xmath0 , but it only has a small effect on the kinetic energy , adding only @xmath91erg in that velocity shell . the synthetic spectrum has @xmath92 , which compares well with the observed @xmath93 . synthetic @xmath94 , showing that the spectrum is rather red . the @xmath75 magnitude drops less than both @xmath74 and @xmath13 , and the bolometric correction is still large , @xmath95 , so that @xmath96 , but this neglects the net emission in the ca ii ir triplet , so the actual value is probably smaller . this is the last spectrum available , corresponding to an epoch well on the light curve tail , with @xmath97 . the spectrum , shown in figure 7 , shows a clear evolution from jan 1 . in particular , the flux peak has moved to the red quite noticeably , indicating a temperature drop . looking at the spectrum in detail , one can notice that the features in the blue have changed little , and so has the region furthest to the red , although the ca ii ir triplet emission is now stronger . on the other hand , the region around 6000 has changed quite significantly . the flux peak has moved from 5500 to 6000 , the absorptions at 5700 ( na i d ) and at 6200 are much stronger . the si iiline has essentially disappeared , all that is left being a weak feature near 6400 , which is too red to be identified with the si ii doublet . in fact , a si ii velocity can not be measured at this epoch . still , many features have persisted from earlier epochs , and these have shifted further to the red , indicating the presence of significant amounts of material at very low velocity in the ejecta . the overall nature of the spectrum is still photospheric , but the incidence of net emission is higher than on jan 1 , which will limit our ability to fit the spectrum using the mc code . we used the density distribution discussed in the previous section , including both the outer @xmath1 part above 25000 km s@xmath0 , which has no effect on the synthetic spectrum , and the inner @xmath98 extension between 3000 and 5000 km s@xmath0 , but we had to extend the density distribution further inwards , so that we could obtain an appropriate temperature and spectrum . in keeping with the hydrodynamical prediction of a ` density hole ' in the centre of the ejecta , we tried to use the largest possible value of the density power law index allowed by the quality of the synthetic spectrum . we selected a model based on the overall fit , but we could not reproduce the observations as well as on previous epochs . we could accomodate an innermost density law @xmath99 below 3000 km s@xmath0 , which is an attempt to include an inner ` hole ' in density . this extension adds a further 0.5m@xmath2 to the ejecta mass , but only less than @xmath100erg to the explosion kinetic energy . our best fit spectrum , shown in figure 7 , has parameters @xmath101 days , @xmath102 ( erg s@xmath0 ) and @xmath103 km s@xmath0 . the temperature is significantly lower than on jan 1 , with @xmath104 ranging from 6000 to 4000k . because of the rather flat density structure at low velocity , the temperature is almost constant ( @xmath105k ) at @xmath106 km s@xmath0 , which is where the spectrum is formed . this helps to keep the lines sufficiently broad . the abundances are similar to those of jan 1 , but oxygen is further reduced and s is increased , reflecting the expected trend as smaller and smaller velocities are sampled . admittedly , the synthetic spectrum is not a very good fit , but almost all the observed features are at least reproduced . the two emission peaks at 4600 and 5500 are much too strong in the model , which affects the synthetic @xmath13 magnitude . the observed value is @xmath107 , and the synthetic spectrum has @xmath108 . the @xmath74 flux is also somewhat overestimated at @xmath109 . possible reasons for the excessive strength of the synthetic peaks at 4600 and 5400were already given in the previous section . in the @xmath13 region , an absorption near 5600 is also not reproduced . a similar feature was also observed on jan 1 , near 5650 , but it was weaker then , while it was absent on dec 24 . the absorption might be attributed to he i 5876 at a velocity of about 13000 km s@xmath0 . he may be expected to be found in the ejecta as a leftover of the star s he - shell . if he is distributed in a shell , he atoms may only be excited at a rather advanced epoch , when the @xmath110-rays and positrons from the decay of @xmath5co can penetrate the ejecta and reach the shell . however , the velocity implied by the position of the absorption is low for a hypotetical he shell , and the apparent shift of the feature between jan 1 and jan 26 does not support this scenario either . alternatively , he can be produced with 56ni in the deepest layer as the result of alpha - rich freezeout ( see paper i , figures 5 and 6 ) . in this case he would be found together with co / fe , so the he velocity could take any value depending on the 56ni mixing . however , in this case non - thermal excitation of he would take place immediately , and we should see the he i line as soon as the photosphere reaches the layer where he is present . if he is located at about 14000 km s@xmath0 , we should have seen the he iline as early as dec 5 . the above problems notwithstanding , as long as no other candidate identification is available he must be regarded as at least a possibility . observatons of he i 10830 would certainly help settle this issue . in any case , given the moderate strength of the line , the he mass involved may not have to be large . the model predicts a moderately strong absorption at 6300 , attributing it to a blend of the si ii doublet and , mostly , of fe ii 6417 , 6456 . this feature is present also in the observed spectrum , but it is much weaker . the reason for the discrepancy is clearly the emission at 6300 , which must signal the onset of the corresponding [ o i ] line . another net emission , which is not reproduced by the model , is visible near 7300 . this is clearly ca ii ] 7292 , 7324 . together with the ca ii ir triplet , these are the only three net emissions in the optical spectrum . this is a typical situation for sne ic at these epochs ( cf . sn 1987 m at 60 days after maximum , swartz et al . 1993 ) . comparing the spectra of sn 1997ef and sn 1987 m , apart from the obviously broader lines in sn 1997ef , it is possible to note that in sn 1987 m the fe iilines are weaker and that the candidate he i line is not seen . sn 1987 m is thought to have produced roughly as much @xmath5ni as sn 1997ef ( 0.15m@xmath2 ; nomoto , filippenko , & shigeyama 1990 , paper i ) , but this could be an overestimation . classical , low energy sne ic are not expected to eject any he , but this might not be the case for sn 1997ef , and it certainly is not for the other hypernova , sn 1998bw , where he i 10830 is observed ( patat et al . 2000 ) . on the other hand , at a comparable epoch the [ o i ] 6300 emission is much stronger in sn 1998bw than in sn 1997ef , supporting the idea that sn 1998bw came from a more massive progenitor . further detailed study of the he ( and o ) content of hypernovae would certainly be worthwhile . the synthetic spectrum has a small @xmath65 ( 0.01 ) , and @xmath111 . the actual value of @xmath65 may be even smaller , and possibly negative , since the model @xmath13 flux is certainly overestimated by about 0.15 mag , and the net emission in the ca ii ir triplet is not included in our estimate . therefore , the value of @xmath53 has an uncertainty of at least 0.2 mag . as we have discussed above at great length , modelling the time evolution of the spectrum of sn 1997ef has revealed two main inconsistencies with the explosion model co100 , which we used in paper i to reproduce both the light curve and the near - maximum spectra . the basic parameters of the observed and synthetic spectra are recapped in table 1 . firstly , the high line velocities observed near maximum in several strong lines , notably the ca ii ir triplet and the si ii doublet , could not be reproduced with the rather steep density profile of co100 , but require a flatter outer density law , @xmath1 , at @xmath21 km s@xmath0 . secondly , the fact that the january spectra are still predominantly photospheric in nature , showing low line velocities ( @xmath112 km s@xmath0 ) is not compatible with the presence in co100 of an inner ` density hole ' . this ` hole ' was the result of depositing the kinetic energy at a radius in the progenitor which results in the ejection of exactly the amount of @xmath5ni necessary to reproduce the sn tail luminosity . matter below that radius , known as the ` mass cut ' , is assumed to fall back onto the compact remnant . clearly , material must exist at low velocities , and we found that an inner extension below 5000 km s@xmath0with a @xmath80 density law down to 3000 km s@xmath0 and a @xmath113 law below that , to reproduce the inner ` density hole ' , is the best overall solution , although we still can not get perfect fits of those very late epochs , even ignoring the net emission which is clearly present in some lines . in figure 8 we show the density profile of the original co100 model and our modified one . the outer density extension adds significanlty to the kinetic energy ( @xmath114erg ) , but only marginally to the ejecta mass ( 0.25m@xmath2 ) , while the inner extension adds about 1.65m@xmath2 to @xmath115 but has a negligible effect ( @xmath116erg ) on the explosion kinetic energy . our modified model has a total @xmath117m@xmath2 and @xmath118erg , compared to @xmath119m@xmath2 and @xmath120erg for co100 . these new values reinforce the classification of sn 1997ef as a hypernova , which is based on the @xmath22 being larger than @xmath47erg . since we have a @xmath1 density law on the outside , and an essentially flat law deep inside , and since our density distribution was derived starting from that of co100 , whose relic is the steep density gradient in the intermediate part , one might argue that a @xmath1 density law might work well throughout the ejecta . we have tested such a possibility , but although it works reasonably well for the outer part , at @xmath121 km s@xmath0 , it gives rise to very sharp - lined spectra at the later epochs , when the photosphere is in the flat part of the ejecta , at @xmath122 km s@xmath0 . also , joining directly the inner flat part and the outer @xmath1 part beyond 10000 km s@xmath0 would give a sn with extremely large @xmath115 and @xmath22 , and the synthetic spectra for the epochs when the photosphere is near 10000 km s@xmath0 would have very deep lines . we have to ask ourselves the question what could give rise to the observed deviation from the hydrodynamical model . the flat outer part may be due to mass loss at high rate during the presupernova stages , or to a transition between the oxygen shell and an outer helium shell , some evidence for which may be present as a weak he i 5876 line in the two january spectra , although at a velocity ( @xmath123 km s@xmath0 ) which is smaller than what would be expected . branch ( 2000 ) offered a similar solution , and his ad hoc density law was even flatter @xmath14 than ours . as a result he offers an estimate of the kinetic energy above 7000 km s@xmath0 as @xmath17erg . our value above the same velocity is somewhat smaller but comparable , @xmath124erg . also , he suggests that the ejecta mass above 7000 km s@xmath0 is @xmath125m@xmath2 , and our value is again smaller but comparable , @xmath126m@xmath2 . the location of the inner density cutoff , on the other hand , was determined essentially by demanding that the correct mass of @xmath5ni be ejected from the progenitor . clearly , the estimate of the position of the cut - off was incorrect . although it is possible that this may be due to an incorrect calculation of the progenitor s structure , it is more tempting to attribute the problem to some asymmetry in the explosion . if the explosion was asymmetric , similar to what has been suggested for the hypernova sn 1998bw in order to explain its connection with a grb , it is quite possible that most @xmath5ni was produced near the beam axis , while away from that axis burning would be less efficient , and would terminate at intermediate elements such as si or s ( maeda et al . 2000 ) . clearly one then has to place the mass cut deeper in order to achieve the require ejected mass of @xmath5ni . this is also supported by our derived abundance distribution , which suggests that the s abundance is still increasing at the lowest velocities sampled , about 2000 km s@xmath0 . the abundance distribution as derived from our models is displayed in figure 9 , but we must keep in mind that , since we used homogenised compositions above the photosphere , all sharp composition boundaries are smoothed out . on the other hand , no clear spectral evidence - such as sudden changes in the properties of the lines - is visible for strict abundance stratification . the presence of an inner density core was also suggested in paper i to explain the observed deviation of the synthetic light curve computed with model co100 from the observations at advanced phases : the synthetic light curve has a steeper slope than the observed one , indicating that @xmath110-ray trapping is more efficient than in the model . this could be achieved if @xmath5ni was distributed deeper than it is in co100 . in order to verify our findings , we used our derived density and abundance distribution and computed a synthetic light curve in spherical symmetry . we used two different codes , one based on monte carlo methods and the other , a radiation hydrodynamics code , which was used in the light - curve computation in paper i. the monte carlo code is based on the simple code developed for sne ia described in cappellaro et al . @xmath110-ray deposition in the expanding ejecta is computed following the random walk of gamma - ray photons adopting a constant @xmath110-ray opacity @xmath127@xmath128g@xmath0 . once they deposit their energy , the @xmath110-rays are assumed to generate optical photons on the spot . the random walk of the optical photons through the ejecta is then also followed in monte carlo , assuming a constant optical opacity @xmath129 . the monte carlo scheme is able to model efficiently the random walk of photons through the ejecta - and thus to take into account the delay between the emission of a photon and its escaping from the sn nebula , which determines the initial shape of the light curve . the value of @xmath129 depends on the temperature and composition , and it is possible that it changes with time . thus we had to find a convenient approximate value for this particular model of sn 1997ef . since @xmath129 affects the light curve essentially near maximum , we found that a value @xmath130 @xmath128g@xmath0 gives a good fit to the light curve around maximum . this value is smaller than for sne ia , as might be expected since line opacity is much stronger in the fe - dominated ejecta of a sn ia . the synthetic light curve in monte carlo is compared to the one obtained with the radiation hydrodynamics code . as described in paper i , the code solves a multi - frequency radiative transfer equation in the fluid s comoving frame , using the feautrier method iteratively with an approximate lambda operator . the energy equation of gas plus radiation and the radiation momentum equation are coupled to the transfer equation in order to follow the evolution of gas temperatures . the code also uses average opacities , and the values used were the same as in the monte carlo code . the synthetic bolometric light curves are shown in figure 10 . except for a slight difference around the maximum , which is probably to be ascribed to monte carlo noise , the codes produce very similar light curves . therefore , it is safe to use the simple monte carlo code in calculating bolometric luminosities if a reasonable opacity is chosen . the synthetic light curves reproduce our derived bolometric curve reasonably well around maximum . the @xmath13 light curve can consequently be reproduced if the bolometric correction as resulting from our synthetic spectra is taken into account . in both codes a @xmath5ni mass of 0.13m@xmath2 was used . in order to achieve a rapid rise of the light curve , consistent with our dating , we had to distribute @xmath5ni homogeneously throughout the ejecta . both light curves reproduce the maximum quite well , but their decline on the tail is much too steep . the observed @xmath13 light curve after about day 60 appears to follow the @xmath5co decline rate , implying constant and complete deposition of about 0.07m@xmath2 of @xmath5ni . however , our codes predict for those advanced epochs a deposition fraction of less than 0.5 and steadily decreasing with time . in figure 11 we compare the observed velocity of the si ii doublet with the run of @xmath131 as derived from the two light curve codes and with the values used as input for the calculation of the synthetic spectra . the two codes are in good agreement with one another , although the mc values are consistently larger than those of the radiation hydro model . this is probably due to different zoning in the two codes . however , the observed points are lower than both models , and the difference increases with the epoch . the velocities used for the spectral calculations , on the other hand , start lower than the observed values but slowly approach them . this is understandable , as it can be expected that at early times the si ii line forms above the photosphere since it is very strong . later , as the line becomes weaker , it forms closer and closer to the photosphere . the fact that the value of @xmath131 tends to flatten out at advanced epochs , and does not reproduce the observations , is not new ( see , e.g. , iwamoto et al . 1998 on sn 1998bw ) , and it probably indicates that the assumption of a gray photosphere is not good at advanced epochs . however , the difference at early phases can not be due to that effect . one possibility is that we have somewhat overestimated the density at the highest velocities in our spectral calculations , where we were guided more by the ca ii ir triplet than by the si ii doublet . note that the velocity derived from model co100 in paper i was much smaller than the observed values . another possibility is that our assumption of a gray opacity again fails in those outermost layers , as shown perhaps by the fact that only the ca ii ir triplet is active there . note that all of these remarks apply to a spherically symmetric situation . another inconsistency between the light curve and spectral calculations is that if we integrate the ni mass as the sum of the abundances of fe , co and ni in the various synthetic spectra we obtain an fe - group mass of only @xmath132m@xmath2above 2000 km s@xmath0 , which is much less that what is necessary to power the light curve . it is unlikely that our spectrum synthesis may have given such a large error , especially since the fe lines are already very strong in most spectra . so we have a siuation where at early times we need more surface @xmath5ni to power the light curve than we see in the spectra , while at late times the @xmath5ni we have used to reproduce the light curve at peak does not deposit sufficiently to reproduce the light curve slope and brightness . the late light curve appears to require the complete deposition of @xmath133m@xmath2 of @xmath5ni . the two problems may be independent , but they may also be related . one possibility to explain the late light curve behaviour is that , in a spherically symmetric scenario , additional @xmath5ni ( @xmath134m@xmath2 ) could be located deep in the ejecta , below 2000 km s@xmath0 . if the density is sufficiently high there the @xmath110-rays produced by the decay of this @xmath5ni may deposit completely . the presence of very low velocity ejecta would further reduce the mass of the expected compact object , and may be in contradiction with the distribution of @xmath5ni in models of the progenitor evolution and explosion . this would not affect the situation at early times . another possibility is that the ejecta may not be spherically symmetric , in either mass or abundance distribution , or both . if some @xmath5ni is ejected at low velocities together with other elements in some direction , the @xmath110-rays produced by its decay chain could be efficiently trapped even at advanced epochs . the optical photons produced by the thermalisazion of @xmath110-rays and positrons in this hypothetical high density region would only be able to escape at advanced phases , and so they would have little effect on the early time spectra but they could power the late light curve . at the same time , some material has been ejected at high velocity , possibly in a jet - like form , to reproduce the sharp rise of the light curve by high - velocity @xmath5niand to explain the broad lines of si ii and ca ii ( @xmath135 km s@xmath0 ) at early epochs . in this scenario , the uniform mixing of @xmath5ni which was used to reproduce the sharp rise of the light curve might refer only to the @xmath5niejected in the jet , while the low velocity @xmath5ni would enter play only later . this scenario might help explaining the discrepancy between the light curve and spectroscopic @xmath5ni masses . if most of the @xmath5ni is ejected in a preferred direction ( e.g. in the jet that might be observed as a grb if the viewing angle is favourable ) , this @xmath5ni may power the early light curve by depositing its @xmath110-rays in the neighbouring ejecta , which have lower velocity , higher density and a smaller fe - group abundance . however , if we view the event from an angle sufficiently far from the jet direction , the spectrum we see would be produced in a region where the abundances are different from those found in the jet hence the smaller spectroscopic mass of @xmath5ni . this is a possibility for sn 1997ef , since there is no positive identificatin of a grb counterpart to the sn event . the possibility that the explosion was asymmetric is in line with existing models of energetic supernova explosions linked to ghrb s ( macfadyen & woosley 1999 , hflich et al . 1999 , khokhlov et al . 1999 , maeda et al . 2000 ) , and it deserved further study with multi - dimensional radiation hydrodynamics codes . we have shown in this paper how spectrum synthesis can be used not only to verify explosion models of sne , but also to improve on them . given our findings , it would now be very interesting to see hydrodynamic calculations of the explosion , in one or more dimensions , which reproduce closely the observed bolometric light curve . the derived density distribution , ejecta mass and kinetic energy may or may not be close to what we have obtained spectroscopically , but in any case such a model would have to be tested also for its ability to reproduce the sn spectral evolution . our study also shows that fitting an observed @xmath13 curve with a synthetic bolometric one may be somewhat misleading , since the bolometric correction can be significant . we have discussed in 9 the reason for the difference between the density distribution derived from spectral modelling and that obtained from 1d hydrodynamic calculations of the explosion of the stripped core of a massive star , and remarked that the presence of significant amounts of ejected matter at low velocity may be due to some asymmetry in the explosion . a similar conclusion was reached for the other type ic hypernova , sn 1998bw , but in that case it was based on somewhat different evidence , most importantly the connection with a grb ( e.g. wheeler 2000 ) . the early - time spectra of sn 1998bw have even broader and more blended lines than those of sn 1997ef , but a clear evolution towards lower line velocities and less line blending is not seen as clearly as in sn 1997ef , while the relatively early development of nebular emission is seen in both objects . this may be understandable since although the ejecta mass of sn 1998bw was larger than that of sn 1997ef , even if we use our upwards revised value for the latter , the kinetic energy of sn 1998bw appeared to be much larger ( nomoto et al . branch ( 2000 ) in a spectroscopic study similar to the one he performed for sn 1997ef , suggested an extremely large @xmath22 for sn 1998bw ( @xmath136erg above 7000 km s@xmath0 , i.e. a factor 2 larger than his value for sn 1997ef above the same velocity ) . we have performed a preliminary analysis of sn 1998bw using the technique described in this paper , and find a similarly large value ( nakamura et al . 1999 ) . if sn 1997ef was also a highly asymmetric explosion , although weaker in energy and ejecting a smaller mass than sn 1998bw , why was a grb not positively detected ? the most likely possibilities are that either the jet - axis of sn 1997ef was not oriented exacly along the line - of - sight , or that a weaker explosion energy reduced the beaming so that a grb was not formed . both scenarios might be able to explain many of the observed features of sn 1997ef , in particular the lower expansion velocities and the smaller measured @xmath22 . when we apply a spherically symmetric model , we estimate @xmath22 by integrating a 1d model around a sphere . therefore , if the sn is observed on or very close to the jet - axis , like sn 1998bw , @xmath22 could be grossly overestimated hflich et al . 1999 ) . on the other hand , the difference between the velocity of the spectral lines in the two sne is real , and so is the difference in luminosity at late times . that sn 1998bw produced nore @xmath5ni is also confirmed by the estimate for the @xmath5ni mass ( @xmath137m@xmath2 ) obtained from the nebular lines , which does not depend much on the asymmetry ( see danziger et al . 1999 ; nomoto et al . 2000 ) . therefore , even though inclination may be a factor , that there is some intrinsic difference between the two objects appears unavoidable , at least as long as the relative distance estimate is reliable . finally , we note that another sn 1997ef - like type ic hypernova was recently observed , sn 1998ey , whose spectra appear to be identical to those of sn 1997ef , and for which there is also no observed grb counterpart ( garnavich et al . this coincidence is striking , and calls for the accumulation of more data on type ic hypernova candidates . * acknowledgements * this work has been supported in part by the grant - in - aid for scientific research ( 12640233 , 12740122 ) and coe research ( 07ce2002 ) of the ministry of education , science , culture and sports in japan . it is a pleasure to thank p.garnavich and j.danziger for useful communications and the referee , d.branch , for constructive remarks . nomoto , k. , mazzali , p. , nakamura , t. , iwamoto , k. , maeda , k. , suzuki , t. , turatto , m. , danziger , i.j . , & patat , f. 2000 , in ` supernovae and gamma - ray bursts ' , ed . m. livio et al . , cambridge university press , cambridge , in press ( astro - ph/0003077 ) rrccccccccccccc 29 nov & 9 & 42.13 & 15500 & 19072 & -12.65 & 1.05 & 6123 & 7725 & 17.45 & 16.74 & 16.7 & 0.29 & -16.600 5 dec & 15 & 42.15 & 9500 & 13962 & -12.30 & 3.38 & 6128 & 9447 & 17.35 & 16.58 & 16.5 & 0.40 & -16.650 17 dec & 27 & 42.20 & 7500 & 8011 & -12.67 & 5.16 & 5291 & 6648 & 17.72 & 16.65 & 16.6 & 0.20 & -16.775 24 dec & 34 & 42.13 & 4900 & 5378 & -12.71 & 7.56 & 5602 & 7602 & 17.83 & 16.71 & 16.8 & 0.32 & -16.500 1 jan & 42 & 42.00 & 3600 & 3775 & -12.78 & 8.48 & 5426 & 6830 & 18.41 & 17.16 & 17.3 & 0.22 & -16.250 26 jan & 67 & 41.80 & 1950 & & -13.51 & 8.97 & 5232 & 5701 & 19.06 & 17.85 & 18.0 & 0.01 & -15.775

we develop a fast variational approximation scheme for gaussian process ( gp ) regression , where the spectrum of the covariance function is subjected to a sparse approximation . our approach enables uncertainty in covariance function hyperparameters to be treated without using monte carlo methods and is robust to overfitting . our article makes three contributions . first , we present a variational bayes algorithm for fitting sparse spectrum gp regression models that uses nonconjugate variational message passing to derive fast and efficient updates . second , we propose a novel adaptive neighbourhood technique for obtaining predictive inference that is effective in dealing with nonstationarity . regression is performed locally at each point to be predicted and the neighbourhood is determined using a measure defined based on lengthscales estimated from an initial fit . weighting dimensions according to lengthscales , this downweights variables of little relevance , leading to automatic variable selection and improved prediction . third , we introduce a technique for accelerating convergence in nonconjugate variational message passing by adapting step sizes in the direction of the natural gradient of the lower bound . our adaptive strategy can be easily implemented and empirical results indicate significant speedups . example.eps gsave newpath 20 20 moveto 20 220 lineto 220 220 lineto 220 20 lineto closepath 2 setlinewidth gsave .4 setgray fill grestore stroke grestore

1306.1999

gaussian process ( gp ) models provide a flexible , probabilistic approach to regression and are widely used . however , application of gp models to large data sets is challenging as the memory and computational requirements scale as @xmath0 and @xmath1 respectively , where @xmath2 is the number of training data points . various sparse gp approximations have been proposed to overcome this limitation . a unifying framework of existing sparse methods is given in @xcite . we consider the stationary sparse spectrum gp regression model introduced by @xcite , where the spectrum of the covariance function is sparsified instead of the usual spatial domain . the ssgp algorithm developed by @xcite for fitting this model uses conjugate gradients to optimize the marginal likelihoood with respect to the hyperparameters and spectral points . comparisons with other state - of - the - art sparse gp approximations such as the fully independent training conditional model ( first introduced as sparse pseudo - input gp in * ? ? ? * ) and the sparse multiscale gp @xcite , showed that ssgp yielded significant improvements . however , optimization with respect to spectral frequencies increases the tendency to underestimate predictive uncertainty and poses a risk of overfitting in the ssgp algorithm . in this paper , we develop a fast variational approximation scheme for the sparse spectrum gp regression model , which enables uncertainty in covariance function hyperparameters to be treated . in addition , we propose an adaptive local neighbourhood approach for dealing with nonstationary data . although accounting for hyperparameter uncertainty may be of little importance when fitting globally to a large data set , local fitting within neighbourhoods results in fitting to small data sets even if the full data set is large , and here it is important to account for hyperparameter uncertainty to avoid overfitting . our examples show that our methodology is particularly beneficial when combined with the local fitting approach for this reason . our approach also allows hierarchical models involving covariance function parameters to be constructed . this idea is implemented in the context of functional longitudinal models by mensah _ ( 2014 ) so that smoothness properties of trajectories can be related to individual specific covariates . gps have diverse applications and various methods have been developed to overcome their computational limitations for handling large data sets . a good summary of approximations used in modelling large spatial data sets is given in @xcite . computational costs can also be reduced through local gp regression as a much smaller number of training data is utilized in each partition . this approach has been considered in machine learning ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and in spatial statistics ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? @xcite propose fitting gp models in local neighbourhoods which are defined online for each test point . however , covariance hyperparameters are estimated only for a subset of all possible local neighbourhoods . different local experts are then combined using a mixture model capable of handling multimodality . our idea of using adaptive nearest neighbours in gp regression is inspired by techniques in classification designed to mitigate the curse of dimensionality @xcite . for each test point , we fit two models . in the first instance , the neighbourhood is determined using the euclidean metric . lengthscales estimated from the first fitting are then used to redefine the distance measure determining the neighbourhood for fitting the second model . experiments suggest that this approach improves prediction significantly in data with nonstationarities , as hyperparameters are allowed to vary across neighbourhoods adapted to each query point . weighting dimensions according to lengthscales downweights variables of little relevance and also leads to automatic variable selection . our approach differs from methods where local neighbourhoods are built sequentially to optimize the choice of the neighbourhood . examples include @xcite and @xcite , where the gaussian likelihood is approximated by the use of an ordering and conditioning on a subset of past observations . in @xcite , an empirical bayes mean - square prediction error criterion is optimized . while greedy searches usually rely on fast updating formulae available only in the gaussian case , our approach works in non - gaussian settings as well . @xcite suggest making neighbourhoods non - local to improve learning of covariance parameters , but local neighbourhoods may work better when the motivation is to handle nonstationarity . @xcite make a connection between discrete spatial markov random fields and continuous gaussian random fields with covariance functions in the matrn class . for fitting the sparse spectrum gp regression model , we derive a variational bayes ( vb , * ? ? ? * ) algorithm that uses nonconjugate variational message passing @xcite to derive fast and efficient updates . vb methods approximate the intractable posterior in bayesian inference by a factorized distribution . this product density assumption is often unrealistic and can lead to underestimation of posterior variance @xcite . however , optimization of a factorized variational posterior can be decomposed into local computations that only involve neighbouring nodes in the factor graph and this often gives rise to fast computational algorithms . vb has also been shown to be able to give reasonably good estimates of the marginal posterior distributions and excellent predictive inferences ( e.g. * ? ? ? * ; * ? ? ? variational message passing @xcite is a general - purpose algorithm that allows vb to be applied to conjugate - exponential models @xcite . nonconjugate variational message passing extends variational message passing to nonconjugate models by assuming that the factors in vb are members of the exponential family . we use nonconjugate variational message passing to derive efficient updates for the variational posteriors of the lengthscales , which are assumed to be gaussian . @xcite use vb for spatial modelling via gp , where they also treat uncertainty in the covariance function hyperparameters . however , they propose using importance sampling within each vb iteration to handle the intractable expectations associated with the covariance function hyperparameters . variational inference has also been considered in machine learning for sparse gps that select the inducing inputs and hyperparameters by maximizing a lower bound to the exact marginal likelihood @xcite , and heteroscedastic gp regression models where the noise is input dependent @xcite . vb is known to suffer from slow convergence when there is strong dependence between variables in the factors . to speed up convergence , @xcite propose parameter expanded vb to reduce coupling in updates , while @xcite considered partially noncentered parametrizations . here , we introduce an adaptive strategy to accelerate convergence in nonconjugate variational message passing , which is inspired by adaptive overrelaxed bound optimization methods @xcite . previously , @xcite showed that nonconjugate variational message passing is a natural gradient ascent algorithm with step size one and step sizes smaller than one correspond to damping . here , we propose using step sizes larger than one which can help to accelerate convergence in fixed point iterations algorithms ( see * ? ? ? * ) . instead of searching for the optimal step size , we use an adaptive strategy which ensures that the lower bound increases after each cycle of updates . empirical results indicate significant speedups . @xcite considered combining parameter - wise updates to form a diagonal direction for a line search . a general iterative algorithm for computing vb estimators ( defined as means of variational posteriors ) has also been proposed by @xcite and its convergence properties investigated for normal mixture models . section [ ssgpmodel ] describes the sparse spectrum gp regression model and section [ variational inference ] develops the nonconjugate variational message passing algorithm for fitting it . section [ adaptive strategy ] presents an adaptive strategy for accelerating convergence in nonconjugate variational message passing . section [ pred distn ] discusses how the predictive distribution can be estimated and the measures used for performance evaluation . section [ neigh ] describes the adaptive neighbourhood approach for local regression . section [ eg ] considers examples including real and simulated data and section [ conclusion ] concludes . given a data set @xmath3 , we assume each output @xmath4 is generated by an unknown latent function @xmath5 evaluated at the input , @xmath6 , and independently corrupted by additive gaussian noise such that @xmath7 a gp prior is assumed over @xmath8 for @xmath9 . for any set of inputs @xmath10 , @xmath11^t$ ] has a joint gaussian distribution , @xmath12 , where @xmath13 is a covariance matrix . we assume that the mean of the process is zero . it is straightforward to allow for a nonzero mean , but a zero mean is sufficient for the examples in this paper . the entries of @xmath13 are given by @xmath14 , where @xmath15 and @xmath16 is some stationary covariance function . for example , we consider the stationary squared exponential covariance function , @xmath17 where @xmath18 , @xmath19^t)$ ] and @xmath20 for @xmath21 . @xcite introduced a novel perspective on gp approximation by sparsifying the spectrum of the covariance function . they considered the linear regression model , @xmath22 where @xmath23 , @xmath24 are independent and identically distributed as @xmath25 and @xmath26 is a @xmath27-dimensional vector of spectral frequencies . the power spectral density of a stationary covariance function @xmath16 is @xmath28 and @xmath29 is proportional to a probability density @xmath30 such that @xmath31 . when @xmath32 are drawn randomly from @xmath30 , @xcite showed that can be viewed as a sparse gp that approximates the full stationary gp by replacing the spectrum with a discrete set of spectral points . from , the probability density @xmath30 associated with the squared exponential covariance function in is @xmath33 . if @xmath32 is generated randomly from @xmath34 , then @xmath35 is a random sample from @xmath30 . from , a sparse gp approximation to @xmath8 is @xmath36 , \label{spec2}\end{aligned}\ ] ] where @xmath37 , @xmath38^t$ ] is a vector of lengthscales and @xmath39 denotes element by element multiplication of two vectors . within the sparse gp approximation , we can allow the components of @xmath40 to be negative . let @xmath41^t$ ] and @xmath42^t$ ] . note that in ( [ sqexpcov ] ) @xmath43 appears as its square in @xmath44 so that @xmath45 remains positive semidefinite . ignoring the non - negativity constraint allows us to use a gaussian variational posterior for @xmath40 . the associated expectations in the variational lower bound can then be derived in closed form ( see section [ variational inference ] ) . this is a highly novel aspect of our algorithm allowing a fast method that still handles covariance function hyperparameter uncertainty . this is especially important when fitting locally as described in section 6 where training datasets may be small . the squared exponential covariance function also implements automatic relevance determination since the magnitude of @xmath43 is a measure of how relevant the @xmath46th variable is . when @xmath43 goes to zero , the covariance function becomes almost independent of the @xmath46th variable , essentially removing it from inference . see @xcite for more discussion . using the stationary sparse gp approximation in ( [ spec2 ] ) , we consider variational inference for @xmath47 \\ + \epsilon_i , \;\;\text{where}\;\ ; \epsilon_i \sim n(0,\gamma^2).\end{gathered}\ ] ] let @xmath48^t$ ] , @xmath49^t$ ] , @xmath50^t$ ] and @xmath51^t$ ] , where @xmath52^t.\end{gathered}\ ] ] then this model can be written as @xmath53 where @xmath54 . for bayesian inference , we assume the priors : @xmath55 , @xmath56 and @xmath57 , where the hyperparameters @xmath58 , @xmath59 , @xmath60 and @xmath61 are assumed to be known . the density function of a random variable @xmath62 distributed as @xmath63 is @xmath64 , where @xmath65 and @xmath66 . while inverse - gamma priors are more commonly used for variance parameters in hierarchical models due to the conditional conjugacy relationship with gaussian families , @xcite recommends use of the half - cauchy family as priors because resulting inferences can be sensitive to inverse - gamma hyperparameters when variance estimates are close to zero . we made the same observation in our experiments with inverse - gamma priors for @xmath67 and @xmath68 . in particular , predictive inferences are sensitive to inverse - gamma priors in local regressions ( see section [ neigh ] ) , where only a small neighbourhood is used for fitting at each test point . we consider variational inference for the sparse spectrum gp regression model in ( [ model2 ] ) . let @xmath69 be the set of unknown parameters and @xmath70 be the true posterior of @xmath71 . in variational approximation , @xmath70 is approximated by a @xmath72 for which inference is more tractable , and the kullback - leibler divergence between @xmath72 and @xmath70 is minimized . this is equivalent to maximizing a lower bound @xmath73 on the log marginal likelihood @xmath74 , where @xmath75 , @xmath76 and @xmath77 denotes expectation with respect to @xmath72 . next , we review some important results in vb and nonconjugate variational message passing , which will be used to construct the variational algorithm . in vb , @xmath72 is assumed to factorize into @xmath78 for some partition @xmath79 of @xmath71 . the optimal densities may be obtained from @xmath80 where @xmath81 denotes expectation with respect to @xmath82 ( see , e.g. * ? ? ? * ) . for conjugate - exponential models , the optimal densities have the same form as the priors and it suffices to update the parameters of @xmath83 , such as in variational message passing @xcite . however , for nonconjugate models , the optimal densities will not belong to recognizable density families . apart from the product assumption , nonconjugate variational message passing @xcite further assumes each @xmath84 is a member of some exponential family , that is , @xmath85 where @xmath86 is the vector of natural parameters and @xmath87 are the sufficient statistics . hence , we only have to find each @xmath86 that maximizes the lower bound @xmath73 . nonconjugate variational message passing can be interpreted as a fixed point iterations algorithm where updates are obtained from the condition that the gradient or natural gradient ( see * ? ? ? * ; * ? ? ? * ) of @xmath73 with respect to each @xmath86 is zero when @xmath73 is maximized . suppose @xmath88 , @xmath89 and let @xmath90 denote the variance - covariance matrix of @xmath91 . provided @xmath92 is invertible , @xcite showed that the natural gradient of @xmath73 with respect to @xmath86 is @xmath93 therefore , the update for each @xmath86 is @xmath94 where the summation is over all factors in @xmath95 , the neighbourhood of @xmath96 in the factor graph of @xmath97 . updates in nonconjugate variational message passing reduce to those in variational message passing when the factors @xmath98 are conjugate ( see * ? ? ? * ; * ? ? ? however , unlike variational message passing , the lower bound @xmath73 is not guaranteed to increase at each step and convergence problems may be encountered sometimes . @xcite suggest using damping to fix convergence problems . when @xmath99 , @xcite showed that the update in can be simplified to @xmath100^{-1 } , \\ \mu_{\theta_i}^q & \leftarrow \mu_{\theta_i}^q + \sigma_{\theta_i}^q\sum_{a \in n(\theta_i)}\negthickspace \frac{\partial s_a}{\partial \mu_{\theta_i}^q}. \end{aligned}\ ] ] here @xmath101 denotes the vector obtained by stacking the columns of a matrix @xmath102 under each other , from left to right in order . we consider a variational approximation of the form @xmath103 from , the optimal densities @xmath104 and @xmath105 are @xmath106 and @xmath107 , where @xmath108 and @xmath109 , @xmath110 , @xmath111 . the variational parameter updates of @xmath112 , @xmath113 , @xmath114 and @xmath115 can also be derived from . as @xmath116 can be arbitrarily large or small , @xcite suggest evaluating @xmath117 efficiently using quadrature . a discussion can be found in appendix b of @xcite and we follow their methods . for @xmath118 , @xmath119 is not a conjugate factor and we use nonconjugate variational message passing . assuming @xmath120 , updates for @xmath121 and @xmath122 can be derived using and matrix differential calculus ( see * ? ? ? the expectations with respect to @xmath123 in are given in appendices a and b. let @xmath124 denote the set of variational parameters . an iterative scheme for finding @xmath125 is given in algorithm [ alg1 ] . a unique aspect of our variational scheme is the way covariance function uncertainty is handled , with the expectations involving @xmath40 in the lower bound computable in closed form . in particular , @xmath126 and @xmath127 can be evaluated in closed form ( see appendix a ) . let @xmath128 be partitioned as @xmath129 $ ] where @xmath102 , @xmath130 and @xmath131 are all @xmath132 matrices . in algorithm 1 , we define @xmath133 , @xmath134 for @xmath135 , @xmath136 , @xmath137 . the lower bound @xmath73 defined in is commonly used for monitoring convergence . it can be evaluated in closed form ( see appendix b ) and is given by @xmath138 the above expression applies only after the updates in steps 5 and 6 of algorithm [ alg1 ] have been made . in the sparse spectrum gp regression model , @xmath139 and @xmath140 are intimately linked . each time the lengthscales @xmath141 are changed by a small amount , the amplitudes ( @xmath140 ) will have to respond to this change in order to match the observed @xmath142 . in , we have assumed that the variational posteriors of @xmath40 and @xmath140 are independent so that expectations with respect to @xmath123 are tractable and closed form updates can be derived for a fast algorithm . however , strong dependence between @xmath40 and @xmath140 implies that only small steps can be taken in each cycle of updates and a large number of iterations will likely be required for algorithm [ alg1 ] to converge . to accelerate convergence , we propose modifying the updates in steps 1 and 2 . let @xmath143 be the natural parameter of @xmath118 and @xmath144 be the update of @xmath143 in nonconjugate variational message passing . @xcite showed that nonconjugate variational message passing is a natural gradient ascent method with step size one . at iteration @xmath145 , we consider @xmath146 where @xmath147 . when @xmath148 , reduces to the update in nonconjugate variational message passing . taking @xmath149 may be helpful when updates in nonconjugate variational message passing fail to increase @xmath73 . from our experiments , instability in algorithm 1 usually occur within the first few iterations . beyond that , the algorithm is usually quite stable and taking larger steps with @xmath150 can result in significant speed - ups . indicates conventional path in fixed point iterations while the dot dash line indicates path to convergence with a step size greater than 1.,scaledwidth=49.0% ] recall that nonconjugate variational message passing is a fixed point iterations algorithm . figure [ fixedpointplot ] illustrates in a single variable case ( where we are solving @xmath151 ) how taking steps larger than one can accelerate convergence . instead of taking @xmath152 , consider @xmath153 , where @xmath154 and @xmath150 . the solid line starting from @xmath155 indicates the conventional path in fixed point iterations while the dot dash line indicates the path with a step size greater than 1 . the dot dash line moves towards the point of convergence faster than the solid line . however , it may overshoot if @xmath156 is too large . in algorithm [ alg2 ] , we borrow ideas from @xcite to construct an adaptive algorithm where @xmath156 is allowed to increase by a factor @xmath157 after each cycle of updates whilst @xmath73 is on an increasing trend and we revert to @xmath148 when @xmath73 decreases . the adaptive nonconjugate variational message passing algorithm is given in algorithm [ alg2 ] . in appendix c , we show that reduces to the updates : @xmath158^{-1 } \\ \text{and } \;\ ; \mu_\lambda^q \leftarrow \mu_\lambda^q + a_t\ , \sigma_\lambda^q \sum_{a \in n(\lambda ) } \frac{\partial s_a}{\partial \mu_\lambda^q}. \end{gathered}\ ] ] step 3(b ) has been added as a safeguard as the updated @xmath122 may not be symmetric positive definite due to rounding errors or when @xmath156 is large . in this case , we propose reducing the step size by a factor @xmath159 until all eigenvalues of @xmath122 are positive . it is useful to insert step 3(b ) in algorithm [ alg1 ] after @xmath122 has been updated as well as it can serve as damping . for both algorithms [ alg1 ] and [ alg2 ] , we initialize @xmath121 as @xmath160^t$ ] ( which is one half of the amplitudes of the inputs after any rescaling ) , @xmath122 as @xmath161^t$ ] , @xmath115 as @xmath162 , @xmath114 as @xmath163 , and @xmath164 and @xmath165 are initialized using the updates in steps 34 of algorithm [ alg1 ] . we set the maximum number of iterations as 500 and the algorithms are deemed to have converged if the relative increase in @xmath73 is less than @xmath166 . @xcite recommend taking the factor @xmath159 to be close to but more than 1 . using this as a guide , we have experimented with @xmath159 taking values 1.1 , 1.5 and 2 . while all these values lead to improvement in efficiency , we find @xmath167 to be more favourable , as the step sizes increase rather slowly when @xmath168 and too fast when @xmath169 , leading to many failed attempts to improve @xmath73 . while algorithm [ alg2 ] does not necessarily converge to the same local mode as algorithm [ alg1 ] , results from the two algorithms are usually very close . algorithm [ alg2 ] sometimes demonstrates the ability to avoid local modes with the larger steps that it takes . we compare and quantify the performance of the two algorithms in section [ pendulum eg ] . note that in algorithm 2 , each failed attempt to improve @xmath73 is also counted as an additional iteration in step 5(b ) even though step 1 does not have to be reevaluated . we note that algorithms 1 and 2 are not guaranteed to converge due to the fixed point updates in nonconjugate variational message passing . however , convergence issues can usually be mitigated by rescaling variables and varying the initialization values . as the fixed point updates may not result in an increase in @xmath73 , it is possible to compute @xmath73 after performing the updates and reduce @xmath156 if necessary . however , this requires computing a lower bound of a more complex form than at each iteration . our experiments indicate that a decline in @xmath73 is often due to @xmath122 not being symmetric positive definite , and hence installing step 3(b ) suffices in most cases . we also find that checking the simplified form of @xmath73 in at the end of each cycle and simply reverting @xmath156 to 1 if necessary is more economical . if premature stopping occurs in algorithms 1 or 2 due to a decrease in the lower bound at some iteration , this can be detected by examination of the lower bound values and remedied if needed by damping where values @xmath149 are considered . let @xmath170 and @xmath171 be the training and testing data sets respectively . let @xmath172 be the set of spectral frequencies randomly generated from @xmath34 . bayesian predictive inference is based on the predictive distribution , @xmath173 assuming @xmath174 is conditionally independent of @xmath131 given @xmath140 , @xmath40 and @xmath175 . we replace @xmath176 with our variational approximation @xmath177 so that @xmath178 from ( [ postpred ] ) , the posterior predictive mean of @xmath174 is @xmath179 where @xmath180^t\end{gathered}\ ] ] and @xmath181 can be computed using results in appendix a. the posterior predictive variance is @xmath182 in the examples , we follow @xcite and evaluate performance using two quantitative measures : normalized mean square error ( nmse ) and mean negative log probability ( mnlp ) . these are defined as @xmath183 the mnlp is implicitly based on a normal predictive distribution for @xmath174 with mean @xmath184 and variance @xmath185 , @xmath186 . we propose a new technique of obtaining predictive inference by fitting models locally using adaptive neighbourhoods . our proposed approach consists of two stages : for each test point @xmath187 , @xmath186 , 1 . we first find the @xmath16 nearest neighbours of @xmath187 in @xmath131 ( that are closest to @xmath187 in terms of euclidean distance ) and denote the index set of these @xmath16 neighbours by @xmath188 . we use algorithm 2 to fit a sparse spectrum gp regression model , @xmath189 , to @xmath190 . 2 . next , we use the variational posterior mean of the lengthscales , @xmath121 , from @xmath189 to define a new distance measure : @xmath191 where the dimensions are weighted according to @xmath192 . this will effectively downweight or remove variables of little or no relevance . using this new distance measure , we find the @xmath16 nearest neighbours of @xmath187 in @xmath131 and denote the index set of these @xmath16 neighbours by @xmath193 . we use algorithm 2 to fit a sparse spectrum gp regression model , @xmath194 , to @xmath195 and use the variational posterior from @xmath194 for predictive inference . in summary , the first fitting @xmath196 is used to find out which variables are more relevant in determining the output . from , a large value of @xmath197 indicates that the covariance drops rapidly along the dimension of @xmath198 and hence the neighbourhood should be shrunk along the @xmath198th dimension . using @xmath121 from the first fit as an estimate of the lengthscales , the neighbourhood is then adapted before performing a second fitting @xmath199 to improve prediction . we do not recommend iterating the fitting process further since this may result in cyclical behaviour with the neighbourhood successively expanding and contracting along a certain dimension as the iterations proceed . in the examples , when the ssgp algorithm is implemented using this adaptive neighbourhood approach , we replace the variational posterior mean value @xmath121 ( which does not exist for the ssgp method since it does not estimate a variational posterior distribution for @xmath40 ) by the point estimates of the lengthscales @xmath200 obtained by the ssgp approach . the adaptive neighbourhood approach is well - placed to handle data with nonstationarities as stationarity is only assumed locally and local fitting can adapt the noise and the degree of smoothing to the nonstationarities . adapting the neighbourhood can also be very helpful in improving prediction when there are many irrelevant variables due to automatic relevance determination implemented via the lengthscales . a major advantage of the variational approach is that it allows uncertainty in the covariance hyperparameters to be modelled within a fast computational scheme . this is especially important when fitting using local neighbourhoods as plug - in approaches to estimating hyperparameters will tend to underestimate predictive uncertainty when the data set is small . this approach is advantageous for dealing with large data sets as well . as we only consider fitting models to a small subset @xmath16 of data points at each test point , a smaller number of basis functions @xmath201 might suffice . while the computational requirements grow linearly with the number of prediction locations , this approach is trivially parallelizable to get a linear speed - up with the number of processors we compare the performance of the variational approach with the ssgp algorithm using three real data sets : the pendulum data set , the rainfall - runoff data set and the auto - mpg data set . the implementation of ssgp in matlab is obtained from http://www.tsc.uc3m.es/~miguel/simpletutorialssgp.php . there are two versions of the ssgp algorithm : ssgp ( fixed ) uses fixed spectral points while ssgp ( optimized ) optimizes the marginal likelihood with respect to the spectral points . we will only consider ssgp ( fixed ) . we observe some sensitivity in predictive performance to the basis functions and adopt the following strategy for better results : for each implementation of algorithm [ alg1 ] ( or [ alg2 ] ) , we randomly generate ten sets of spectral points from @xmath34 , perform 2 iterations of the algorithm , and select the set with the highest attained lower bound to continue to full convergence . a similar strategy was used by @xcite to initialize the ssgp algorithm . due to the zero mean assumption , we center all target vectors , @xmath142 by subtracting the mean @xmath202 from @xmath142 . in the examples , `` va '' refers to the variational approximation approach implemented via algorithm [ alg2 ] , `` global '' refers to using the entire training set for fitting while `` local '' refers to the adaptive neighbourhood approach described in section [ neigh ] . the pendulum data set ( available at http://www.tsc.uc3m.es/~miguel/simpletutorialssgp.php ) has @xmath203 covariates and contains 315 training points and 315 test points . the target variable is the change in angular velocity of a simulated mechanical pendulum over 50 ms and the covariates consist of different parameters of the system . @xcite used this example to show that ssgp ( optimized ) can sometimes fail due to overfitting . we rescale the input variables in the training set to lie in @xmath204 $ ] and consider the number of basis functions , @xmath205 . we compare the performance of algorithm [ alg2 ] with ssgp ( fixed ) using nmse and mnlp values averaged over ten repetitions . we set @xmath167 , @xmath206 for the half - cauchy priors , following @xcite and @xcite and @xmath207 , @xmath208 for the lengthscales in algorithm [ alg2 ] . for this data set which is quite small , we note that the adaptive neighbourhood approach did not yield significant improvements as all inputs are relevant and there is no strong nonstationarity . hence we report only results for global fits , which are shown in figure [ pendulumnmsemnlp ] . the nmse and mnlp values produced by algorithm [ alg2 ] are comparable with that of ssgp ( fixed ) for small @xmath209 and are better for large @xmath209 . on the whole , algorithm [ alg2 ] produces reasonably good nmse performance and is less prone to overfitting than the ssgp algorithm . the ability of the variational approach to treat uncertainty in the covariance function hyperparameters reduces underestimation of predictive uncertainty , resulting in better mnlp performance . ( global va ) and global ssgp ( fixed ) and averaged over ten repetitions plotted against number of basis functions @xmath201.,scaledwidth=49.5% ] with @xmath167 while empty circles correspond to algorithm [ alg1].,scaledwidth=49.5% ] next , we compare the performance of algorithm [ alg1 ] with algorithm [ alg2 ] both in terms of efficiency and the lower bound attained at convergence . we use algorithm [ alg1 ] to re - perform the runs for @xmath210 , using the same sets of spectral points that were used in algorithm 2 . these runs are indexed from 1 to 40 ( there are ten repetitions for each @xmath209 ) . figure [ pendulumlbcomparisons ] shows a plot of the lower bound attained at convergence on the left and a plot of the number of iterations required for convergence on the right for each of the 40 runs . figure [ pendulumlbcomparisons ] indicates that , except for runs 3 and 40 , the lower bound attained by algorithms [ alg1 ] and [ alg2 ] are almost indistinguishable . however , algorithm [ alg2 ] required a much smaller number of iterations to converge than algorithm [ alg1 ] . excluding runs 3 and 40 where the lower bound attained by algorithms [ alg1 ] and [ alg2 ] differs significantly , using algorithm [ alg2 ] instead of algorithm [ alg1 ] leads on average to a reduction of 49% in the number of iterations required for convergence . the highest reduction observed is 84% at run 9 . at run 3 , algorithm [ alg2 ] was able to escape a local mode and attained a higher lower bound at convergence . however , at run 40 , it was caught in a local mode . we re - perform run 40 using @xmath168 and it turns out that algorithm [ alg2 ] was then able to attain the same lower bound as algorithm [ alg1 ] but in around half the number of iterations . with @xmath167 while dashed line corresponds to algorithm [ alg1 ] ) . right : plot of the adaptive step size @xmath211 used in algorithm [ alg2 ] against iteration number @xmath212.,scaledwidth=49.5% ] the typical behaviour of algorithm [ alg2 ] is illustrated in figure [ pendulumrun33 ] . on the left is a plot of the lower bound against iteration number and on the right is a plot of the adaptive step size @xmath211 used in algorithm [ alg2 ] against iteration number @xmath212 for run 33 . the step size typically increases by a factor of 1.5 at each iteration but falls back to 1 when the lower bound fails to increase . the step size may also be reduced by factors of 1.5 due to the requirement that the covariance matrix be symmetric positive definite in step 2(b ) of algorithm [ alg2 ] . the reduction in the number of iterations that algorithm [ alg2 ] takes to converge as compared to algorithm [ alg1 ] is 74% for run 33 . for the next four subsections , our discussion concerns performance of the adaptive neighbourhood approach . we compare the performance using two real datasets : the rainfall - runoff data set and the auto - mpg data set . we fit these data globally using ssgp ( fixed ) , va and mcmc , and compare results with the adaptive neighbourhood approach , implemented using both ssgp ( fixed ) and algorithm [ alg2 ] with factor @xmath167 . for the priors , we set @xmath206 , @xmath207 . for @xmath59 , we set @xmath213 for the rainfall - runoff data where a less smooth mean function is expected and @xmath214 for the auto - mpg data . the prior variance for the lengthscales can be chosen empirically by predictive performance on a test set or using prior knowledge . prior knowledge about the hyperparameters in the covariance function can be elicited by thinking about the prior degree of expected correlation of the mean function for covariates separated by lag one in each dimension when the covariates are standardized . for both the global ssgp ( fixed ) and va approach , we consider the number of basis functions @xmath215 . in addition , we generate ten artificial covariates on top of the existing covariates in both the rainfall - runoff and auto - mpg data set to test the capability of algorithm [ alg2 ] in automatic relevance determination . in this example , we consider data from a deterministic rainfall - runoff model , which is a simplification of the australian water balance model ( awbm , * ? ? ? the awbm estimates catchment streamflow using time series of rainfall and evapotranspiration data and is widely used in australia for estimating catchment water yield or design flood estimation . the model has three parameters - the maximum storage capacity @xmath216 , the base flow index _ bfi _ and the baseflow recession factor @xmath13 . we have model simulations for around eleven years of average monthly potential evapotranspiration and daily rainfall data for the barrington river catchment , located in new south wales , australia . the model was run for 500 different values of the parameters @xmath217 generated using a maximin latin hypercube design . this data contains 500 data points for each of 3700 days , with a total of 1.85 million data points . for each day , the total rainfall is also recorded . a subset of this data has been studied in @xcite . [ htb ! ] even though the size of the data is large , the computational demands of the adaptive neighbourhood approach will depend mostly on the number of query points and the neighbourhood size . this makes our approach highly suitable for this data set . this is especially true since emulation of the model will be most interesting near values of peak rainfall input and generally for events of hydrological significance , where there might be a flood risk for example . so the proportion of interesting query points in this example is a small fraction of the total data set size and furthermore we expect the model output to vary rapidly in some parts of the parameter space but very little in other parts so the ability of the local method to smooth adaptively is very attractive for this problem . we will consider prediction for the two days with the highest rainfall inputs . we take awbm streamflow response as the target @xmath142 , and @xmath216 and @xmath13 as covariates , omitting _ bfi_. a small amount of independent normal random noise with standard deviation 0.01 was added to @xmath142 to avoid degeneracies in regions of the space where the response tends to be identically zero . for each day , we randomly selected 100 data points as the test set and use the remaining 400 data points as the training set . these data are highly nonstationary with large flat regions , a few rapidly varying regions and the noise level changes a lot over the space . figure [ rainfall2covmnlpnmse ] shows the nmse and mnlp values averaged over ten repetitions for the rainfall - runoff data with peak rainfall . for global ssgp ( fixed ) , we observe a slight improvement in nmse values as @xmath209 increases , while mnlp values remain largely constant at around 3.75 even for large @xmath209 . due to the nonstationary nature of this data , a global stationary fit does very poorly in mnlp . for the adaptive neighbourhood approach , we consider neighbourhoods of size @xmath218 , fixing the number of basis functions , @xmath219 . for the local methods , the dotted lines correspond to results from the initial fitting where the @xmath16 nearest neighbours are determined based on euclidean distance . the solid lines correspond to results from the final fit where the @xmath16 nearest neighbours are determined using the new distance measure with dimensions weighted according to the lengthscales . the improvement brought about by adapting the neighbourhood is more apparent in va than in ssgp ( fixed ) . ! ] figure [ rainfall2covmnlpnmse_2 ] shows the nmse and mnlp values averaged over ten repetitions for the rainfall - runoff data with the second highest rainfall . in this example we also observe that a global stationary fit does very poorly in mnlp , again due to the nonstationary nature of the data . similarly , when adapting the neighbourhood approach , there are greater improvements in va than in ssgp ( fixed ) . it is clear that the adaptive neighbourhood approach is critical for this data set where the mean function varies rapidly over some parts of the space but very little over other parts . the variational approach performs very well when using just a small neighbourhood about each test point both in terms of nmse and mnlp . figure [ s=68 ] illustrates how the neighbourhood of a test point changes from the initial to the final fit for the case @xmath220 , when algorithm [ alg2 ] was being used . the plot on the left shows the neighbours ( denoted by circles ) of a test point ( denoted by solid circle ) determined using euclidean distance . the plot on the right shows the neighbours of the same test point determined using the new distance measure . in this case , the component of @xmath121 corresponding to the covariate @xmath216 is much larger than that corresponding to the covariate @xmath13 , resulting in the neighbourhood being shrunk along the @xmath216 axis . the adapted neighbourhood leads to an improvement in the estimation of the predictive mean and especially the predictive variance of the test point . + m & va & ssgp & mcmc ( 2000 iterations ) + + 20 & 2.626 & 0.177 & + 40 & 8.315 & 0.268 & + 60 & 17.545 & 0.491 & 2068.596 + 80 & 32.920 & 0.638 & + 100 & 55.223 & 0.836 & + + + k & va & ssgp & mcmc ( 2000 iterations ) + + 20 & 257.773 & 16.361 & 653.690 + 40 & 247.744 & 17.964 & 2253.453 + 60 & 265.194 & 22.695 & 5182.298 + 80 & 320.416 & 23.091 & 9229.747 + 100 & 279.356 & 24.774 & 14441.828 + table [ comtimes ] shows the computation times of the va , ssgp ( fixed ) and mcmc algorithms on the rainfall - runoff data with peak rainfall input . we ran the mcmc using rstan @xcite on a dual processor windows pc 3.30 ghz workstation and both ssgp ( fixed ) and va in matlab using a 3.2 ghz intel core i5 quad core imac . for the global approach , computation times for ssgp ( fixed ) and va are averaged over 10 repetitions , while mcmc is based on a single run with 2000 iterations . for the adaptive neighbourhood approach , table [ comtimes ] shows the total time it takes to run all 100 test points . note that the timing for the local approaches can be significantly reduced by parallelizing . in terms of computation speed , ssgp ( fixed ) is the fastest followed by va . we observe that mcmc is substantially slower than the other methods and the computation time increases significantly when the size of neighbourhood increases . we do not observe such significant increase in computation times for va and ssgp ( fixed ) . we consider rainfall - runoff data on the day with peak rainfall and generate ten additional covariates artificially . as both covariates @xmath216 and @xmath13 lie in the interval @xmath221 $ ] , we simulate each of the ten additional covariates randomly from the uniform distribution on the interval @xmath221 $ ] . we compare the performance of ssgp ( fixed ) and algorithm [ alg2 ] using a global fit with the adaptive neighbourhood approach . we set @xmath167 in algorithm [ alg2 ] and use the same priors as in section [ rainfall ] . for the global approach , we consider the number of basis functions , @xmath215 while for the adaptive local neighbourhood approach , we consider neighbourhoods of size @xmath218 , fixing the number of basis functions , @xmath219 . the results are shown in figure [ rainfall12covmnlpnmse ] . for the global approach , the results of ssgp ( fixed ) are quite similar to those in the 2 covariates case . for the local approach , a small neighbourhood with @xmath222 does not work well for both ssgp ( fixed ) and va , indicating that a larger neighbourhood is likely required for high dimensional problems . there is a clear improvement in the mnlp values from adapting the neighbourhood according to the lengthscales and algorithm 2 achieved the lowest mnlp values among the methods that were studied , using a smaller neighbourhood . the mnlp values achieved by algorithm [ alg2 ] are close to those attained in section [ rainfall ] , indicating that the adaptive neighbourhood approach is effective in eliminating covariates of little relevance . the variational approach is able to provide significant improvement in this aspect and is much more robust to overfitting for small neighbourhoods . however , the nmse values obtained in the adaptive neighbourhood approach are higher than those obtained in the global approach . finally , we note that a good neighbourhood size is dependent on the number of covariance function parameters to be estimated and on the degree of nonstationarity , which is very much problem specific . some experimentation with different neighbourhood sizes is probably necessary . in this example , we consider the automobile city - cycle fuel consumption in miles per gallon ( auto - mpg ) data taken from the cmu statistics library . this dataset was used in the 1983 american statistical association exposition and is available at http://archive.ics.uci.edu/ml/datasets.html . the dataset contains 398 instances and nine attributes . @xcite used this data to predict the attribute mpg " , which is the city - cycle fuel consumption in miles per gallon . the other eight attributes include two multi - valued discrete , four continuous attributes and two categorical variables . we drop the two categorical variables , car name and origin , and keep the four continuous attributes and two multi - valued discrete variables . six of the data points are removed as they have missing entries in some of the input variables . we randomly select 80 data points as the test set , and use the remaining 312 data points as the training set . figure [ auto ] shows the nmse and mnlp values averaged over ten repetitions . for the global ssgp ( fixed ) and va methods , we observe slight improvements in both the nmse and mnlp values as @xmath209 increases . the mnlp and nmse values for mcmc and global va are also better than for global ssgp ( fixed ) . for the adaptive neighbourhood approach , we consider neighbourhoods of size @xmath223 , while fixing the number of basis function @xmath219 . for local va , the final fit is slightly better than the initial fit . there is an improvement brought about by adapting the neighbourhood as figure [ auto ] shows that the mnlp values of both the initial and final fits are lower than the mcmc method for neighbourhood size of 40 , 60 , 80 and 100 . for local ssgp ( fixed ) , we observe that their performance is worse than mcmc . moreover , it seems that the initial fit is better than the final fit , for neighbourhood size of 80 and 100 . this may be because the lengthscales are not accurate enough to be used for the final fit . we also examined the local ssgp approach with larger neighbourhood sizes of 150 , 200 and 250 . we found that , at a neighbourhood size of 150 , the performance of the final fit of local ssgp ( fixed ) ( mnlp and nmse of 2.38 and 0.129 respectively ) is slightly better than mcmc . adapting the neighbourhood approach is still more apparent in the variational approach as it is able to achieve mnlp and nmse value of 2.26 and 0.117 respectively at neighbourhood size of 60 . we now consider the auto - mpg data and look at the influence of irrelevant covariates on the model . this is again done by generating ten additional covariates artificially and randomly from the uniform distribution on the interval [ 0,1 ] . once again , like the rainfall - runoff data , it seems that a larger neighbourhood is required to attain the best performance for the variational approach when irrelevant covariates are added . in this example , for the variational approach , we found that neighbourhood size of 100 produces the best peformance with mnlp and nmse values of 2.30 and 0.127 respectively . again , after examining the local ssgp ( fixed ) approach with larger neighbourhood sizes , we found that it attains the best performance ( mnlp and nmse of 2.43 and 0.141 respectively ) at a neighbourhood size of 150 . in order to explain why there is a difference in the stability of the adaptive neighbourhood approach between va and ssgp ( fixed ) , we examined the estimated predictive mean and variance for one test point from the auto - mpg test set with 10 simulated irrelevant covariates . we implement the adaptive neighbourhoods approach based on just the initial fitting , which uses the shortest euclidean distance . figure [ posterior_mean ] shows 100 posterior predictive means and variances from ssgp ( fixed ) and va with the adaptive neighbourhood approach . in the 100 replications , only the spectral points change . since va accounts for hyperparameter uncertainty , it is more robust towards the choice of spectral points . we observe that the posterior predictive means and variances are concentrated around a smaller range of values even when the size of neighbourhood is small . on the other hand , for local ssgp ( fixed ) , the posterior predictive means vary more for different choices of the spectral points with the values ranging from 0 to 20 and with many of the posterior predictive variances small when the size of the neighbourhood is small . in this paper , we have presented a nonconjugate variational message passing algorithm for fitting sparse spectrum gp regression models where closed form updates are possible for all variational parameters , except for the evaluation of @xmath116 . we note that @xmath116 can be evaluated very efficiently using quadrature and there is almost no computational overhead when compared to updates based on conditionally conjugate inverse - gamma priors for the variance parameters . however , half - cauchy priors lead to much better predictive inference especially in the adaptive neighbourhood approach where the amount of training data is small . a bayesian approach has been adopted for parameter estimation which allows covariance function hyperparameter uncertainty to be treated and empirical results suggest that this improves prediction ( especially in the mnlp values ) and prevents overfitting . we also propose a novel adaptive neighbourhood technique for obtaining predictive inference which is adept at handling data with nonstationarities and this approach can be extended to large data sets as well . the simulated data sets showed that weighting the dimensions according to the lengthscales estimated from an initial fit is very effective at downweighting variables of little relevance , leading to automatic variable selection and improved prediction . in addition , we introduce a technique for accelerating convergence in nonconjugate variational message passing by taking step sizes larger than one in the direction of the natural gradient of the lower bound . we do not attempt to search for the optimal step size but adopt an adaptive strategy that can be easily implemented , and empirical results indicate significant speed - ups . algorithm [ alg2 ] is thus an attractive alternative for fitting sparse spectrum gp regression models , which is stable , robust to overfitting for small data sets and capable of dealing with highly nonstationary data as well when used in combination with the adaptive neighbourhood approach . we thank lucy marshall for supplying the rainfall - runoff data set . linda tan was partially supported as part of the singapore delft water alliance s tropical reservoir research programme . david nott , ajay jasra and victor ong s research was supported by a singapore ministry of education academic research fund tier 2 grant ( r-155 - 000 - 143 - 112 ) . we also thank the referees and associate editor for their comments which have helped improved the manuscript . references amari , s. : natural gradient works efficiently in learning . neural computation 10 , 251276 ( 1998 ) attias , h. : inferring parameters and structure of latent variable models by variational bayes . in laskey , k. and prade , h. ( eds . ) proceedings of the 15th conference on uncertainty in artificial intelligence , pp . morgan kaufmann , san francisco , ca ( 1999 ) attias , h. : a variational bayesian framework for graphical models . in solla , s. a. , leen , t. k. and mller , k .- advances in neural information processing systems 12 , pp . mit press , cambridge , ma ( 2000 ) blei , d. m. and jordan , m. i. : variational inference for dirichlet process mixtures . bayesian analysis 1 , 121144 ( 2006 ) boughton , w. : the australian water balance model . environmental modelling and software 19 , 943956 ( 2004 ) braun , m. and mcauliffe , j. : variational inference for large - scale models of discrete choice . journal of the american statistical association 105 , 324335 ( 2010 ) gelman , a. : prior distributions for variance parameters in hierarchical models . bayesian analysis 1 , 515533 ( 2006 ) gramacy , r. b. and apley , d. w. ( 2014 ) . local gaussian process approximation for large computer experiments . journal of computational and graphical statistics . to appear . haas , t. c. : local prediction of a spatio - temporal process with an application to wet sulfate deposition . journal of the american statistical association 90 , 11891199 ( 1995 ) hastie , t. and tibshirani , r. : discriminant adaptive nearest neighbor classification . ieee transactions on pattern analysis and machine intelligence 18 , 607616 ( 1996 ) hoffman , m. d. , blei , d. m. , wang , c. and paisley , j. : stochastic variational inference . journal of machine learning research 14 , 13031347 ( 2013 ) honkela , a. , valpola , h. and karhunen , j. : accelerating cyclic update algorithms for parameter estimation by pattern searches . neural processing letters 17 , 191203 ( 2003 ) huang , h. , yang , b. and hsu , c. ( 2005 ) . triple jump acceleration for the em algorithm . in han , j. , wah , b. w. , raghavan , v. , wu , x. and rastogi , r. ( eds . ) proceedings of the 5th ieee international conference on data mining , pp . ieee computer society , washington , dc , usa . , mallicka , b. k. and holmesa , c. c. : analyzing nonstationary spatial data using piecewise gaussian processes . journal of the american statistical association 100 , 653668 ( 2005 ) knowles , d. a. , minka , t. p. : non - conjugate variational message passing for multinomial and binary regression . in shawe - taylor , j. , zemel , r. s. , bartlett , p. , pereira , f. and weinberger , k. q. ( eds . ) advances in neural information processing systems 24 , pp . red hook , ny : curran associates , inc . ( 2011 ) lzaro - gredilla , m. , quionero - candela , j. , rasmussen , c. e. and figueiras - vidal , a. r. : sparse spectrum gaussian process regression . journal of machine learning research 11 , 18651881 ( 2010 ) lzaro - gredilla , m. and titsias , m. k. : variational heteroscedastic gaussian process regression . in getoor , l. and scheffer , t. ( eds . ) proceedings of the 28th international conference on machine learning , pp . omnipress , madison , mi , usa ( 2011 ) lindgren , f. , rue , h. and lindstrm , j. : 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 ( 2011 ) magnus , j. r. and neudecker , h. : matrix differential calculus with applications in statistics and econometrics . wiley , chichester , uk ( 1988 ) nguyen - tuong , d. , seeger , m. and peters , j. : model learning with local gaussian process regression . advanced robotics 23 , 20152034 ( 2009 ) nott , d. j. , tan , s. l. , villani , m. and kohn , r. : regression density estimation with variational methods and stochastic approximation . journal of computational and graphical statistics 21 , 797820 ( 2012 ) ormerod , j. t. and wand , m. p. : explaining variational approximations . the american statistician 64 , 140153 ( 2010 ) park , s. and choi , s. : hierarchical gaussian process regression . in sugiyama , m. and yang , q. ( eds . ) proceedings of 2nd asian conference on machine learning , pp . 95110 ( 2010 ) qi , y. and jaakkola , t. s. : parameter expanded variational bayesian methods . in schlkopf , b. , platt , j. and hofmann , t. ( eds . ) advances in neural information processing systems 19 , pp . 10971104 . mit press , cambridge ( 2006 ) quinlan , r. : combining instance - based and model - based learning . in proceedings on the tenth international conference of machine learning , 236 - 243 . university of massachusetts , amherst . morgan kaufmann ( 1993 ) quionero - candela , t. and rasmussen , c. e. : a unifying view of sparse approximate gaussian process regression . journal of machine learning research 6 , 19391959 ( 2005 ) rasmussen , c. e. and williams , c. k. i. : gaussian processes for machine learning . mit press , cambridge , ma ( 2006 ) ren , q. , banerjee , s. , finley , a. o. and hodges , j. s. : variational bayesian methods for spatial data analysis . computational statistics and data analysis 55 , 31973217 ( 2011 ) salakhutdinov , r. and roweis , s. : adaptive overrelaxed bound optimization methods . in fawcett , t. and mishra , n. ( eds . ) proceedings of the 20th international conference on machine learning , pp . 664671 . aaai press , menlo park , california ( 2003 ) snelson , e. and ghahramani , z. : sparse gaussian processes using pseudo - inputs . in weiss , y. , schlkopf , b. and platt , j. ( eds . ) advances in neural information processing systems 18 , pp . 12571264 . mit press , cambridge , ma . ( 2006 ) snelson , e. and ghahramani , z. : local and global sparse gaussian process approximations . in meila , m. and shen , x. ( eds ) jmlr workshop and conference proceedings volume 2 : aistats 2007 , pp . 524531 ( 2007 ) stan development team : rstan : the r interface to stan , version 2.5.0 . http://mc-stan.org/rstan.html ( 2014 ) stein , m. l. , chi , z. and welty , l. j. : approximating likelihoods for large spatial data sets . journal of the royal statistical society : series b 66 , 275296 ( 2004 ) tan , l. s. l. and nott , d. j. : variational inference for generalized linear mixed models using partially non - centered parametrizations . statistical science 28 , 168188 ( 2013 tan , l. s. l. and nott , d. j. : a stochastic variational framework for fitting and diagnosing generalized linear mixed models . bayesian analysis ( 2014 ) . doi : 10.1214/14-ba885 titsias , m. k. : . variational learning of inducing variables in sparse gaussian processes . in van dyk , d. and welling , m. ( eds . ) proceedings of the 12th international conference on artificial intelligence and statistics , pp . 567574 ( 2009 ) urtasun , r. and darrell , t. : sparse probabilistic regression for activity - independent human pose inference . in ieee conference on computer vision and pattern recognition 2008 , pp . 18 ( 2008 ) vecchia , a. v. : estimation and model identication for continuous spatial processes . journal of the royal statistical society : series b 50 , 297312 ( 1988 ) walder , c. , kim , k. i. and schlkopf , b. : sparse multiscale gaussian process regression . in mccallum , a. and roweis , s. ( eds . ) proceedings of the 25th international conference on machine learning pp . acm press , new york ( 2008 ) wand , m. p. , ormerod , j. t. , padoan , s. a. and frhrwirth , r. : mean field variational bayes for elaborate distributions . bayesian analysis 6 , 847900 ( 2011 ) wand , m. p. : fully simplified multivariate normal updates in non - conjugate variational message passing . journal of machine learning research 15 , 13511369 ( 2014 ) wang , b. and titterington , d. m. : inadequacy of interval estimates corresponding to variational bayesian approximations . in cowell , r. g. and ghahramani , z. ( eds . ) , _ proceedings of the 10th international workshop on artificial intelligence and statistics _ , pp . 373380 . society for artificial intelligence and statistics ( 2005 ) wang , b. and titterington , d. m. : convergence properties of a general algorithm for calculating variational bayesian estimates for a normal mixture model . bayesian analysis 3 , 625650 ( 2006 ) winn , j. and bishop , c.m . : variational message passing . journal of machine learning research 6 , 661694 ( 2005 ) suppose @xmath224 and @xmath225 , @xmath226 are fixed vectors the same length as @xmath40 . let @xmath227 and @xmath228 , then @xmath229 \end{gathered}\ ] ] @xmath230 \end{gathered}\ ] ] @xmath231\end{gathered}\ ] ] by setting @xmath232 in the first and third expressions , we get @xmath233 @xmath234=\exp\{i \mu^t ( t_1-t_2)- \tfrac{1}{2}(t_1-t_2)^t \sigma ( t_1-t_2)\}$ ] implies @xmath235 & = e\{\cos(t_1^t\lambda)\cos(t_2^t\lambda ) \\ & \quad + \sin(t_1^t\lambda)\sin(t_2^t\lambda)\ } \\ & = \exp\{-\tfrac{1}{2}(t_1-t_2)^t \sigma ( t_1-t_2)\ } \\ & \quad \cdot \cos\{\mu^t ( t_1-t_2)\ } \end{aligned}\ ] ] and @xmath236 & = e\{\sin(t_1^t\lambda)\cos(t_2^t\lambda ) \\ & \quad -\cos(t_1^t\lambda)\sin(t_2^t\lambda)\ } \\ & = \exp\{-\tfrac{1}{2}(t_1-t_2)^t \sigma ( t_1-t_2)\ } \\ & \quad \cdot \sin\{\mu^t ( t_1-t_2)\}. \end{aligned}\ ] ] replacing @xmath226 by @xmath237 , we get @xmath238 & = e\{\cos(t_1^t\lambda)\cos(t_2^t\lambda ) \\ & \quad -\sin(t_1^t\lambda)\sin(t_2^t\lambda)\ } \\ & = \exp\{-\tfrac{1}{2}(t_1+t_2)^t \sigma ( t_1+t_2)\ } \\ & \quad \cdot \cos\{\mu^t ( t_1+t_2)\ } \end{aligned}\ ] ] and @xmath239 & = e\{\sin(t_1^t\lambda)\cos(t_2^t\lambda ) \\ & \quad + \cos(t_1^t\lambda)\sin(t_2^t\lambda)\ } \\ & = \exp\{-\tfrac{1}{2}(t_1+t_2)^t \sigma ( t_1+t_2)\ } \\ & \quad \cdot \sin\{\mu^t ( t_1+t_2)\}. \end{aligned}\ ] ] ( [ e1])+([e3 ] ) gives the first equation of the lemma , ( [ e1])-([e3 ] ) gives the second and ( [ e2])+([e4 ] ) gives the third . @xmath240 using lemma 1 , we have @xmath241^t,\ ] ] where @xmath242 \end{aligned}\ ] ] and @xmath133 for @xmath135 , @xmath136 . we also have @xmath243 where @xmath244 $ ] , where @xmath245 , @xmath246 , @xmath247 are all @xmath248 matrices and @xmath249 @xmath250 , @xmath251 for @xmath136 , @xmath137 . from ( [ lb ] ) , the lower bound is given by @xmath252 where @xmath253 the terms in the lower bound can be evaluated as follows : @xmath254 \\ \cdot { \mathcal{h}(n , c_\gamma^q , a_\gamma^2)}/{\mathcal{h}(n-2,c_\gamma^q , a_\gamma^2)}\end{gathered}\ ] ] @xmath255 @xmath256 @xmath257 @xmath258 @xmath259 @xmath260 @xmath261 @xmath262 putting these terms together and making use of the updates in steps 5 and 6 of algorithm [ alg1 ] gives the lower bound in . it can be shown ( see * ? ? ? * ; * ? ? ? * ) that the natural parameter of @xmath120 is @xmath263 where @xmath264 is a unique @xmath265 matrix that transforms @xmath266 into @xmath101 for any @xmath267 symmetric square matrix @xmath102 , that is , @xmath268 . we use @xmath266 to denote the @xmath269 vector obtained from @xmath101 by eliminating all supradiagonal elements of @xmath102 . @xcite is a good reference for the matrix differential calculus involved in the derivation below . from and 7 ) , we have @xmath270 where @xmath271 and @xmath272 are evaluated at @xmath273 let @xmath274 the first line of simplifies to @xmath275 the second line of gives @xmath276

we report new _ spitzer space telescope _ observations from the irac and mips instruments of the young ( 3 myr ) orionis cluster . the populous nature of this cluster makes it a good target for statistically - significant studies of disk emission as a function of mass . we identify 336 stars as members of the cluster using optical and near - infrared color magnitude diagrams . using the spectral energy distribution ( sed ) slopes in the irac spectral range , we place objects in several classes : non - excess stars , stars with optically thick disks ( like classical t tauri stars ) , class i ( protostellar ) candidates , and stars with `` evolved disks '' ; the last exhibit smaller irac excesses than optically thick disk systems . in general , this classification agrees with the location expected in irac - mips color - color diagrams for these objects . we find that the evolved disk systems are mostly a combination of objects with optically thick but non - flared disks , suggesting grain growth and/or settling , and transition disks , systems in which the inner disk is partially or fully cleared of small dust . in all , we identify 7 transition disk candidates and 3 possible debris disk systems . there appears to be a spatial extension of infrared excess sources to the north - east , which may be associated with the young ( myr ) embedded cluster ngc 2024 . as in other young stellar populations , the fraction of disks depends on the stellar mass , ranging from% for stars in the herbig ae / be mass range ( ) to% in the t tauri mass range ( 1 - 0.1 ) . we find that the disk fraction does not decrease significantly toward the brown dwarf candidates (.1 ) . the irac infrared excesses found in stellar clusters and associations with and without central high mass stars are similar , suggesting that external photoevaporation is not very important in many clusters . finally , we find no correlation between the x - ray luminosity and the disk infrared excess , suggesting that the x - rays are not strongly affected by disk accretion .

astro-ph0701476

observations have shown and theory predicts that disks around young stars are an inescapable consequence of star formation , given angular momentum conservation during protostellar cloud core collapse @xcite . as the resulting protoplanetary disks age , their excesses at near - infrared ( nir ) and mid - infrared wavelengths decrease . previous studies have indicated that the timescale for this evolution in disk emission is strongly dependent on the stellar mass @xcite ranging from 5 - 7 myr for objects in the mass range of the t tauri stars ( tts , with spectral type k5 or later ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) to @xmath53 myr for objects in the mass range of the herbig ae / be ( haebe ) stars ( spectral type f5 or earlier ; * ? ? ? the observed decreases in infrared emission can result from : grain growth to sizes much larger than the wavelength of observation ; dust settling to the disk midplane , which reduces the `` flaring '' of the disk and thus the amount of energy radiated @xcite ; clearing of small dust particles by large bodies in the disk ; or some combination of these processes . since grain growth and settling to the mid - plane occur fastest in the ( warmer ) inner disk @xcite , the disk emission is expected to decline faster at short wavelengths than at long wavelengths , and there is observational evidence for this @xcite . of particular interest are `` transition disk '' objects , which have an inner , optically thin disk region , possibly produced by planet clearing , combined with an outer , optically thick disk @xcite . prior to the complete or nearly - complete clearing seen in transition disks , one would expect to observe optically - thick but reduced infrared emission from the inner disk , for which there is some evidence ( e.g. , * ? ? ? * ; * ? ? ? * ) . however , additional large samples of stars are desirable to make the result more statistically significant , as stars of the same mass and age show a wide range of disk emission properties @xcite . the @xmath1 orionis cluster is of particular interest because it is reasonably near and relatively populous , making possible statistically - significant studies of disk properties as a function of stellar mass . moreover , @xmath1 orionis has an age @xmath6 myr at which one might expect the beginnings of disk evolution to become evident . it is part of the orion ob1b sub - association which has an age of 1.7 - 7 myr@xcite . low - mass members of the @xmath1 orionis cluster were first reported by @xcite , who found over 80 x - ray sources and spectroscopically identified more than 100 low mass , pre - main sequence ( pms ) members lying within 1 degree from the star @xmath1 ori . the distance calculated by hipparcos for this star ( 352@xmath7 ) agrees , within the uncertainties of hipparcos , with the hipparcos distance calculated for the overall population of the stellar subassociation ob1b ( 439@xmath833 ; * ? ? ? * ) , which is statistically more reliable . the estimated age of the cluster is 2 - 4 myr @xcite . since the @xmath1 orionis cluster is relatively near and the reddening toward the center of the cluster is low ( e(b - v)=0.05 mag ; * ? ? ? * ; * ? ? ? * ) , this stellar cluster is an excellent laboratory to study young stars in a entire range of mass , from the massive and multiple o9.5 type star @xmath1 ori to the lowest mass objects , like brown dwarfs and free - floating planets ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the unprecedented sensitivity and spatial resolution provided by the _ spitzer space telescope _ in the near- and mid - infrared windows are powerful tools to expand significantly our understanding of star and planet formation processes . in this contribution , we analyze the near- and mid - infrared properties of stars in the @xmath1 orionis cluster , ranging in mass from haebe stars to the substellar limit . this paper is organized as follows . in [ sec : obs ] we present the observational data . in [ sec : selmem ] we describe the selection of cluster members . we analyze the observations and describe the results on [ sec : res ] , and give our conclusions in [ sec : conc ] . we have observed a field of @xmath01730 arcmin@xmath9 on the @xmath1 orionis cluster using the four channels ( 3.6 , 4.5 , 5.8 , and 8.0 ) of the infrared array camera ( irac , * ? ? ? * ) ; 90% of this field was also observed using the 24 band of the multiband imaging spectrometer for _ spitzer _ ( mips , * ? ? ? the irac observations presented here were taken on october 9 , 2004 . the field of view was covered by a @xmath10 position mosaic ( 20 overlap ) , with three dithered exposures at each position . images were obtained in high dynamic range ( hdr ) mode , whereby a short integration ( 1 second ) is immediately followed by a long integration ( 26.8 seconds ) . standard basic calibrated data ( bcd ) products from version s13.2 of the _ spitzer _ science center s irac pipeline were used to make the final mosaics . post - bcd data treatment was performed using custom idl software ( gutermuth et al . 2004 , gutermuth et al . 2006 ) that includes modules for detection and correction of bright source artifacts , detection and removal of cosmic ray hits , construction of the long and short exposure hdr mosaics , and the merger of those mosaics to yield the final science images . mips observations were made using the medium scan mode with half - array cross scan overlap , resulting in a total effective exposure time of 80 seconds . the 24 images were processed using the mips instrument team data analysis tool ( dat ) , which calibrates the data and applies a distortion correction to each individual exposure before combining into a final mosaic @xcite . figures [ fig : field ] and [ fig : field2 ] show the irac color and the mips 24 images of the @xmath1 orionis cluster . the lower - right panel in figure [ fig : field2 ] shows a map of dust infrared emission @xcite illustrating the location of the cluster in the ob1b sub - association , which can be traced by the ringlike structure centering at ra@xmath084 and dec@xmath0 - 2 @xcite . in this panel , the boxes show the positions of the irac ( small box ) and mips ( large box ) scan regions . the isocontours are estimators of galactic extinction ( for = 0.2 , 0.4 , 0.6 & 0.8 mag ) from the 100 map of dust emission ( see , * ? ? ? the galactic extinction and the background at 8 and 24 is quite low compared with other regions with similar ages , which improves the precision and sensitivity of the photometry . point source detections were carried out individually on each irac band using photvis ( version 1.09 ) , an idl gui - based photometry visualization tool developed by r. gutermuth using the daophot modules ported to idl as part of the idl astronomy user library @xcite . more than 10,000 sources were detected in at least one _ spitzer _ band . we extracted the photometry of these objects using the _ apphot _ package in iraf , with an aperture radius of 3.7 and a background annulus from 3.7 to 8.6 . we adopted zero - point magnitudes for the standard aperture radius ( 12 ) and background annulus ( 12 - 22.4 ) of 19.665 , 18.928 , 16.847 and 17.391 in the [ 3.6 ] , [ 4.5 ] , [ 5.8 ] and [ 8.0 ] bands , respectively . aperture corrections were made using the values described in irac data handbook @xcite . a preliminary list of 1682 objects ( hereafter _ sample 1 _ ) was created by selecting those objects having photometric measurements in all irac bands with errors less than 0.15 magnitudes . we obtained point source photometry at 24 with iraf/_daophot _ point spread function fitting , using an aperture size of about 5.7 and an aperture correction factor of 1.73 derived from the stinytim psf model . the absolute flux calibration uncertainty should be 4% @xcite . because of the contamination by the bright star @xmath1 ori in the irac and mips images , stars located within 30 of this o star were not included in the preliminary list . optical magnitudes were obtained from the cida equatorial variability survey which is being carried out using the quest i camera @xcite installed at the jurgen stock telescope ( a 1-m schmidt telescope ) at the venezuelan national observatory of llano del hato . the camera , an array of 4x4 ccds , is designed to work in driftscan mode , which is a very efficient way to survey large areas of the sky . each scan of the sky was reduced and calibrated with the standard quest software and the method described in @xcite for scans centered at @xmath11=-1 . we used data from multiple scans centered at @xmath12 . the scans have a width of 2.2@xmath13 in @xmath11 . the range in right ascension shown in figure 1 is covered completely by these optical observations . however , there is no data in a small region from -2.52@xmath13 @xmath14 -2.44@xmath13 , corresponding to the gap between ccds in the quest i camera . most objects with an optical counterpart have more than 10 measurements , so we can identify the variables using differential photometry @xcite . we do not have spectroscopic confirmation for most of the probable members in the @xmath1 orionis cluster . we must use photometric criteria to make a robust selection of members , as described below . figure [ fig : agn1 ] shows the [ 3.6 ] versus [ 5.8]-[8.0 ] color - magnitude diagram for _ sample 1 _ ( [ sec2:ir ] ) . the bulk of the sample has [ 5.8]-[8.0]@xmath00 and [ 3.6]@xmath514.5 ; below this limit the number of objects with photospheric colors ( [ 5.8]-[8.0]@xmath00 ) decreases drastically , and the contamination from extragalactic sources is expected to be more than 50% @xcite . we identify extended sources ( crosses ) by computing the radial profile for each object using the _ radprof _ task in iraf , and designate as extended sources those with fwhm values more than 3 standard deviations from the median in all irac bands ; typically this means fwhm@xmath32 . most objects located below the [ 3.6 ] band limiting magnitude ( solid line ) in figure [ fig : agn1 ] are identified in this way as extended sources and are likely to be extragalactic objects . we therefore initially restrict our sample to objects above the solid line in figure [ fig : agn1 ] . some galaxies have very red colors because of strong pah ( polycyclic aromatic hydrocarbon ) emission ; with features at 7.7 , 8.6 , 11.2 , 12.7 , 16.4 , and 17.1 correlated with star formation rates @xcite . @xcite showed that the [ 4.5 - 5.8 ] versus [ 5.8 - 8.0 ] and [ 3.6 - 5.8 ] versus [ 4.5 - 8.0 ] color - color diagrams can be used to eliminate most of these objects . as shown in figure [ fig : pah ] , objects below and to the right of the solid lines are likely to be pah - rich galaxies . we find that most of the pah - rich galaxy candidates have extended radial profiles as inferred from point - source fitting . visual inspection of the irac images also shows that many of these candidates are diffuse objects . in addition to the galaxies , other objects with similar colors turn out to be faint companions of brighter stars , thus calling into question the accuracy of the photometry . only one object ( open square ) with a stellar profile is located in the pah - rich galaxies regions ; this star ( hd294268=so411 ) has been identified as a young star with spectral type f5 @xcite . we therefore eliminate the rest of the objects below the solid lines in figure [ fig : pah ] , resulting in a sample of 1280 stellar sources ( hereafter _ sample 2 _ ) . about 40% of the objects in _ sample 2 _ do not have optical counterparts in the cida equatorial variability survey ; most of these objects are fainter than the limiting magnitude ( [email protected] ) or brighter than the saturation limit ( [email protected] ) of this survey . some of the objects in _ sample 2 _ also are located on the gap in declination of the survey . we augmented the v optical data set using photometry from @xcite and for the brightest objects from the @xcite catalog . virtually all the objects in _ sample 2 _ ( 99% ) have 2mass counterparts @xcite . figure [ fig : comp ] shows the distributions of j magnitudes for all 2mass sources in the irac field ( open histogram ) and for the 2mass sources included in _ sample 2 _ ( striped histogram ) . comparison of these two histograms indicates that _ sample 2 _ is essentially complete to about j=14.0 . assuming a reddening of e(b - v)=0.05 @xcite , a distance of d=440 pc @xcite and the 3 myr isochrones from @xcite , the equivalent completeness limit in mass is @xmath0 0.15 @xmath4 ( spectral type @xmath0 m5 ) . for comparison , the distribution of j magnitudes for our mips detections ( solid histogram ) is shown in figure [ fig : comp ] . most of brightest objects ( j@xmath511 ; @xmath01.5 @xmath4 ) have 24 detections . _ sample 2 _ includes 214 stars previously found to be members by other investigators ( hereafter previously known members ; see the appendix [ sec : members ] ) . of these objects , the membership of 104 stars were confirmed spectroscopically , while the other 110 stars were assigned membership based only on optical photometry . in addition , 105 objects in our sample have been identified as x - ray sources by @xcite , of which 49 objects have not been identified previously as spectroscopic or photometric members . figure [ fig : cmd_sel ] shows the v versus v - j color - magnitude diagram for _ sample 2_. filled squares ( spectroscopic members ) and filled triangles ( photometric members ) represent previously known members ; both distributions share the same region in figure [ fig : cmd_sel ] suggesting that optical photometric selection can be effective in identifying low mass members ( see also ; * ? ? ? * ; * ? ? ? the stars with @xmath15 roughly follow the 3 myr isochrone of ( * ? ? ? * dashed curve ) , while the redder stars depart significantly from the theoretical isochrone ; this may be due to incompleteness in the opacity tables at low temperatures , resulting in theoretical colors that are too blue for v - j @xmath3 4 @xcite . filled circles represent the higher mass stars selected by @xcite with high membership probabilities ( @xmath350% ) ; the plus symbols represent bright stars with very low probabilities of membership . the mean of the distribution of previously known cluster members can be characterized roughly as straight line in the v versus v - j diagram , with a standard deviation @xmath1(v - j ) @xmath16 mag . as a first approximation , we define the photometric region of probable members as the region above the dotted line in figure [ fig : cmd_sel ] . this line represents the lower 3 @xmath1 deviation from the mean distribution line . the x - ray detected stars @xcite , labeled by open squares , are mostly located in the region of the previously known cluster members ( most of these objects were identified previously as cluster members ) . therefore , x - ray emission can be used as an additional criterion supporting membership in young stellar groups ( e.g. ; * ? ? ? * ; * ? ? ? * ; * ? ? ? figure [ fig : cmd_sel2 ] shows the j versus j - k color - magnitude diagram for _ sample 2_. stars with j @xmath3 11.5 have similar j - k colors , in agreement with the trend expected for the 3 myr isochrone from @xcite . for these stars , the distribution of j - k color can be represented as a gaussian function centered at j - k=0.94 with @xmath1(j - k ) = 0.06 . we adopt the border between members and non - members to be the 3 @xmath1 deviation line ( dotted line ) . for stars with [email protected] , we define the region of probable members by using the x - ray sources in addition to the known members of the cluster . the crosses represent non - members defined using figure [ fig : cmd_sel ] ( below the dotted line ) . some of these non - members , with j@xmath313 , appear in the region of probable members in figure [ fig : cmd_sel2 ] , so some contamination would result from selecting stars using 2mass colors alone . our preliminary member sample consists of stars located redward from dotted lines in figures [ fig : cmd_sel ] and [ fig : cmd_sel2 ] , x - ray sources @xcite , and previously known members ( see [ sec : members ] ) located redward from the dotted line in figure [ fig : cmd_sel2 ] . our preliminary uncertain member sample of the cluster includes those objects selected using only the criterion defined in figure [ fig : cmd_sel2 ] , where the contamination of non - members could be important and additional confirmation is required . the most important source of contamination are field m dwarfs . given the galactic latitude of the @xmath1 orionis cluster ( [email protected] deg ) , m giant stars are not expected to be a significant source of contamination ( @xmath55% ) in the cluster @xcite . we identify 3 possible m giants stars using the j - h versus h - k color - color diagram @xcite , including them in the uncertain members sample . @xcite showed that young stars located in the general region of the cluster consist of two populations with different mean ages and distances superimposed along the line of sight : one with an age of 3 myr located at 439 pc ( consistent with the @xmath1 orionis cluster ) and the other with 10 myr located at 326 pc , which could be associated with the orion ob1a association @xcite . however , most of the objects belonging to the older stellar group are located north - ward from the main population of @xmath1 orionis cluster , so that the contamination by the older stars in our survey should be small , @xmath1710% ( jeffries 2006 , private communication ) . in particular , 11 out 75 stars included in our preliminary member sample have radial velocities from @xcite consistent for stars in the older population but inconsistent with those of the @xmath1 orionis cluster these stars were included , with a note , in the uncertain member sample . only two stars in our uncertain members sample have radial velocity expected of stars in the @xmath1 orionis cluster ; we included these in our final membership list . the final catalogs include 336 stars in the member sample and 133 stars in the uncertain member sample . in the member sample , 132 stars have optical photometry from the cida survey , of which 57 stars ( 43% ) were classified as variable stars using the method described in @xcite ; since young stars are expected to exhibit photometric variability ( e.g. ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , this is an additional criterion supporting membership . table [ tab : members ] shows irac , mips and v photometry data for stars in the _ member sample _ according to the criteria discussed above . column 1 shows the internal running identification number in our sample ; columns 2 and 3 are the stellar coordinates ; columns 4 , 5 , 6 and 7 give irac magnitudes in the bands [ 3.6 ] , [ 4.5 ] , [ 5.8 ] and [ 8.0 ] , respectively ; column 8 gives mips ( 24 ) magnitudes ; columns 9 and 10 show the v magnitudes and its sources ; column 11 indicates if the star has been identified as variable in the cida survey ; column 12 indicates if the star is a x - ray source @xcite ; column 13 shows the disk classification based on on the irac and mips analysis ( see [ sec : res ] ) ; the last column gives the references for the _ previously known members_. table [ tab : uncert ] shows irac and mips photometry for stars included in the _ uncertain member sample_. the information shown in this table in columns from 1 to 8 is the same as in table [ tab : members ] . the last column shows the disk classification based on the irac and mips analysis discussed in the following section . figure [ fig : irac_cc ] shows two irac color - color diagrams , [ 5.8]-[8.0 ] versus [ 3.6]-[4.5 ] ( left panels ) and [ 4.5]-[5.8 ] versus [ 3.6]-[4.5 ] ( right panels ) , for the members ( upper panels ) and uncertain members ( lower panels ) . different symbols show the disk classification based on the irac sed slope ( [ res : slope ] ) . particularly , triangles , squares and diamonds represent stars with different levels of infrared emission , while circles are non - excess stars . symbols surrounded by open squares show objects with small or no excesses in the irac bands but large excesses at 24 ( see below ) . the dashed box in the [ 3.6 ] - [ 4.5 ] versus [ 5.8 ] - [ 8.0 ] color - color diagram indicates the the region predicted for stars with optically - thick disks ( [ res : slope ] ) by the models of @xcite the upper - left panel of figure [ fig : irac_cc ] shows that most of stars with significant irac excess ( squares ; hereafter irac class ii stars ) have colors in agreement with model colors for optically - thick disks . the stars without irac excesses ( hereafter irac class iii stars ) are located around the photospheric region ( [ 5.8]-[8.0]@xmath00 , [ 3.5]-[4.5]@xmath00 ) . a few systems are found in the region between the irac class iii and class ii stars ; these objects with weak irac excess ( triangles ) are discussed in more detail in [ res : slope ] . two of these objects ( the stars so638 and s0299 ) have [ 3.6]-[4.5]@xmath170 , while exhibiting significant emission at longer wavelengths . three other objects ( diamonds ) have very large excesses which may suggest the presence of a protostellar envelope ( hereafter irac class i candidates ) ; so457 ( iras05358 - 0238 ) , so927 ( an emission line object * ? ? ? * ) , and so411 ( hd294268 ) . the last two objects have [ 3.6]-[4.5 ] colors expected for class ii systems , so these objects are more likely to be objects with optically - thick disks and high 8 fluxes due to strong silicate emission or rising seds at wavelength longer than 5.8 ( [ res : sed ] ) . in the upper - right panel of figure [ fig : irac_cc ] , the [ 4.5]-[5.8 ] colors of irac class ii stars shows less scatter in comparison with the [ 5.8]-[8.0 ] colors . there may be some pah background contamination present in the [ 8.0 ] band ; however , this also could indicate that intrinsic differences in disk flaring or other properties that may be more easily detected at @xmath18 m @xcite . for example , the irac class ii stars located redward from the dashed box ( so462 , so774 and so1266 ) and the star s0411 ( with [ 5.8]-[8.0]@xmath31.5 ) in the upper - left panel , have a flat or rising seds at wavelengths longer than 5.8 ( see [ res : sed ] ) . in the lower panels , nine irac class ii stars are located in the region expected for stars with optically thick disks , and thus are likely to be cluster members . one irac class ii star ( so1356 ) , located below the model region , has an excess at [ 5.8]-[8.0 ] , but not at [ 3.6]-4.5 ] or [ 4.5]-[5.8 ] . this star was not covered in the mips field , so it is not clear if the excess observed at 8 originates from pah background contamination or from emission from an outer disk . the object so1293 ( represented by the triangle ) has a larger excess at [ 4.5]-[5.8 ] than at [ 5.8]-[8.0 ] ; it is faint ( j=15.64 ) and so could have a longer wavelength excess below the mips detection limit . five uncertain members with very large infrared excesses ( irac class i candidates ) , are plotted in the irac color - color diagrams : so950 , so336 , so361 , s0916 and so668 . because agn are located in the same region as the class i objects in these diagrams @xcite and are too faint to easily detect spatial extension , additional data are needed to confirm these objects as protostellar members . figure [ fig : mips_cc ] shows k-[24 ] versus j - h color - color diagrams for members ( upper panel ) and uncertain members ( lower panel ) . we sort the objects into three regions , depending upon the magnitude of their k-[24 ] excesses ( see also * ? ? ? one region consists of k-[24 ] @xmath19 , corresponding to systems with undetectable or very uncertain long - wavelength excesses , given photometric errors . a second region encompasses objects with k-[24 ] @xmath33.3 , corresponding to optically - thick disk emission and/or protostellar envelope emission . finally , the intermediate - excess region 0.5@xmath5 k-[24 ] @xmath5 3.3 is consistent with colors typical of optically - thin debris disks [ the famous @xmath20 pic debris disk system , which is one of the debris disk system with largest 24 excess known , has k-[24]@xmath03.3 ; ( e.g. , * ? ? ? * ; * ? ? ? there are seven members with very large [ 24 ] excesses but low irac excesses ( symbols surrounded by open squares ) ; so908 , so818 , so1267 , so897 , so299 , so587 and so1268 . the last object has no detectable irac excess . these stars are possible `` transition disk '' systems , in which the inner disk is completely or nearly - completely cleared of small dust but the outer disk is optically thick ( * ? ? ? * see [ res : cc ] ) . in the lower panel , the uncertain member so120 is also a transition disk candidate . there are two early type members which are located in the intermediate - excess region : so913 ( j - h=-0.04 ) has no irac excess and so956 ( j - h=0.10 ) has a weak irac excess ( [ res : slope ] ) . these stars are `` debris disk candidates '' . the star so981 ( j - [email protected] ) shows no irac excess but has a modest 24 excess , and thus is a likely debris disk system around a k early type star . however , a more detailed study of the debris disk candidates in this young stellar cluster is necessary to clarify the actual nature of their 24 excesses , since contribution from primordial disk material could be present @xcite . there are 5 low mass star members ( j - [email protected] ) with weak irac excess , in which the excesses observed at the irac bands and at 24 are consistent with a overall decrease of infrared emission from the inner and outer disks . @xcite showed the utility of characterizing excess infrared emission using the irac sed slope @xmath21 determined from the [ 3.6]-[8.0 ] color , where @xmath21=@xmath22/dlog[\lambda]$ ] ) . in this system , irac class i candidates , defined as systems with @xmath23 , are usually protostars ; disk - bearing systems , having @xmath24 , include irac class ii stars and objects with weak irac excess ( the lower limit depends upon spectral type and photometric errors , see below ) ; and irac class iii stars have essentially no inner disk emission , with @xmath25 ( again , depending upon spectral type and errors ) . figure [ fig : alphadef ] shows @xmath21 versus the [ 8.0 ] magnitude for members ( filled symbols ) and uncertain members ( open symbols ) of the cluster . dashed lines represent the regions of differing levels of infrared excess as discussed above . we subdivide the disk - bearing stars into two categories : irac class ii stars ( squares , @xmath21@xmath3 - 1.8 ) ; and systems with less irac infrared excess , which @xcite termed `` anemic '' disks and which we call here evolved disks " ( -2.56 @xmath26 ) . because the photometric uncertainties in @xmath21 increase with [ 8.0 ] , a single straight line in figure [ fig : alphadef ] to separate irac class iii stars ( in the irac wavelength range ) and evolved disks is not useful . @xcite avoided this problem by applying their criteria to separate class iii and evolved disks only to the brighter stars with spectral types m4.5 or earlier ( the number of evolved disks would be overestimated in the lowest mass bins due to large photometric errors ; c. lada , private communication ) . we distinguish between irac class iii stars ( circles ) and evolved disk systems ( triangles ) using the mean error of @xmath21 ( @xmath27 ) calculated from the photometric errors for stars within 0.5 magnitude bins . irac class iii stars are those with excesses , if any , smaller than the 3 @xmath27 deviation ( solid lines ) from the median @xmath21 . we wish to examine the mass dependence of disk emission , but we do not have spectral types for most of these stars . as a proxy for mass , we use the j magnitude , which is relatively unaffected by disk excess emission , and convert to approximate mass ranges using the 3 myr isochrone from @xcite for low mass ( @xmath281 m@xmath29 ) stars and @xcite high mass ( @xmath31 m@xmath29 ) stars , assuming a distance of 440 pc and e(b - v)=0.05 . figure [ fig : alphaj ] shows @xmath21 versus j , with the regions expected for objects in different ranges of mass : haebe range ( @xmath32 m@xmath29 ) , intermediate mass tts ( imtts ) range ( 1.0 - 2.0 m@xmath29 ) , tts range ( 0.1 - 1.0 m@xmath29 ) and brown dwarfs range ( @xmath50.1 m@xmath29 ) . most of the objects with optically thick and evolved disks are located in the mass range of t tauri stars . using the results from figure [ fig : alphaj ] , we derive the fractions of stars with optically - thick disks ( filled circles + dotted line ) and all disk - bearing stars ( open circles + dashed line ) versus j magnitude , with corresponding mass ranges indicated ( see also table [ tab : frac ] ) . the lowest fraction of disks is observed in the highest mass stars ( haebe range ) , in agreement with the fraction calculated for the overall population of ob1b sub - association @xcite . there is a marginal evidence that the disk fraction declines toward lower masses ( into the brown dwarf range ) , in agreement with results from @xcite for the young ( 2 - 3 myr ) stellar group ic 348 . nevertheless , including the error bars , the fraction is also consistent with no mass dependence toward lower masses . the fraction of thick disks in the tts mass range agrees with the result from ( * ? ? ? * % ; ) for low mass stars . only 15@xmath85% of disk - bearing stars in the tts mass range are disk evolved objects , approximately in agreement with results from the 4 myr - old cluster tr 37 ( 10% ; * ? ? ? * ) , although with better statistics and higher photometric accuracy . if we assume a steady - state evolution of disks , and also assume that the evolved disks are going through a phase of clearing of small dust particles in the inner disk , then our observed frequency suggests a timescale of inner disk clearing @xmath30 myr @xmath31 myr . figure [ fig : sed ] shows seds ( data from irac , mips , 2mass and optical b , v , r and i magnitudes when are available ) for selected members . the first 5 rows show the seds of stars in the mass range of tts ( 0.1 - 1.0 m@xmath29 ) , while the last row shows the seds of stars in the mass range of haebe stars ( @xmath32 m@xmath29 ) . seds ( solid lines ) are normalized to the j band and are sorted by different levels of ir excesses indicating differences in the disks . each panel is labeled with the internal running identification number and the classification from the irac sed slope analysis ( [ res : slope ] ) . the dotted line represents the median sed ( table [ tab : sed_ctts ] ) of optically thick disk stars in the mass range of tts defined in figure [ fig : alphaj ] ( hereafter , optically thick disk median sed ) ; error bars denote the quartiles of the distribution . this typical disk emission in the @xmath1 orionis cluster is lower than the median in taurus ( * ? ? ? * dot - dashed line in the first upper - left panel ) , which has an age of @xmath01 - 2 myr , indicating differences due to evolutionary effects such as grain growth and/or settling to the disk midplane ( see also * ? ? ? * ; * ? ? ? * ) , or lower accretion - rate @xcite . the dashed lines represent the median photospheric sed ( table [ tab : sed_wtts ] ) in the t tauri mass range . the first row of figure [ fig : sed ] shows the seds of evolved disk objects ( so905 , so1057 , so759 and so587 ) which exhibit modest infrared disk emission . the lower level of infrared excess emission could be explained by decreasing the height of the irradiation surface in the disks due to a higher degree of settling , which diminishes the degree of flaring , thus producing flatter disk structures @xcite . the second row of seds ( so120 , so1268 , so299 and so897 ) shows some transition disk candidates , which exhibit little or no excess in irac bands but 24 emission similar to optically thick disks . this indicates a removal of small dust particles in the inner disks but not depletion or removal in the outer disk . in the next rows , the stars so818 , so1267 , so908 , so1156 , so540 and so1266 show somewhat weaker irac excesses . the stars so73 , so927 , so1154 , so1153 , so457 , and so668 show larger infrared excesses than the optically thick median , which might indicate more highly flared disk structure @xcite , edge - on disks , or a contribution from an envelope . so927 has strong emission in the h@xmath21 @xcite . so1153 ( v510 ori ; hh 444 ) and so1154 ( hh 445 ) are herbig - haro objects producing collimated outflows @xcite . finally , so457 has been identified by @xcite as a class i object . in the last row of figure [ fig : sed ] , seds of earlier - type members are compared with photospheres of stars with similar spectral types ( long dashed line * ? ? ? * ) . the star so724 ( b2 ; * ? * ) shows small excess at 8 but has no excess at 24 , this star was classified in [ res : slope ] as an evolved disk object possibly due to a background pah contamination at 8 . the stars so913 ( b9.5 ; * ? ? ? * ) and so956 ( a8 ; * ? ? ? * ) show small or no excesses in the irac bands but have significant excesses at 24 , consistent with being debris disk stars . finally , the star so411 , classified as class i candidate in [ res : slope ] , shows small excesses at wavelengths shorter than 5.8 and a rising sed at longer wavelengths . this emission line star , with spectral type f5 @xcite , could be an accreting transition disk object in the haebe mass range . this object alternatively could be considered to be an imtts , as f5 is the usual transition spectral type between haebe and imtts ( see * ? ? ? most of the disk classifications using the irac sed slope ( [ res : slope ] ) agree with the results from analyzing seds . however , a few objects were reclassified taking into account the overall seds ; these objects are labeled in tables [ tab : members ] and [ tab : uncert ] . figure [ fig : disk_evol ] shows the fraction of stars in the tts mass range with near - infrared disk emission in different stellar groups , as a function of age @xcite . recent _ spitzer _ results for the young stellar clusters ngc7129 @xcite , taurus @xcite , @xmath32 chameleontis @xcite , tr37 and ngc7160 @xcite , ic348 @xcite , and upper scorpius @xcite have been included in this plot . the disk fractions shown in figure [ fig : disk_evol ] include stars with optically thick and evolved disks ( disk - bearing stars ) detected mostly using irac photometry . the @xmath1 orionis cluster ( filled square ) follows ( within the scatter of the distribution ) the disk fraction trend outlined by in other stellar groups @xcite . aside from the decrease in the disk fraction with age , the amount of infrared disk emission also decreases with age . figure [ fig : slopehist ] shows the distribution of @xmath21 for low mass stars in the stellar groups taurus @xcite , ic348 @xcite , @xmath1 orionis ( this work ) and tr 37 @xcite . as in figure [ fig : sed ] , the youngest group , taurus , has a larger population of stars with robust optically thick disks ( @xmath21 @xmath3 -1 ) . in contrast , in the more evolved stellar groups most disk - bearing stars have an @xmath21 @xmath5 - 1 , indicating a reduction in the height of the disk photosphere ( less disk flaring ) in the inner regions . the taurus sample suggests a relatively clear break between stars without irac excesses and optically - thick disk systems @xcite , although the sample size is relatively small . the @xmath1 orionis sample also suggests a small fraction of objects with intermediate or evolved disk excesses , as discussed previously . the evolved disk fraction is less certain in tr 37 , due to the larger photometric errors for that more distant cluster . the ic 348 sample suggests a larger fraction of evolved disks than seen in @xmath1 ori ; the reality of this possible difference requires further study . in contrast to ic 348 , the @xmath1 orionis and tr 37 clusters include high mass stars located in their center ( o9 in @xmath1 orionis cluster and o7 in tr37 ) ; however , the medians of @xmath21 are similar in these stellar groups ( figure [ fig : slopehist ] ) . this suggests that external photoevaporation in @xmath1 orionis and tr 37 may be not very important for disk dissipation . figure [ fig : frac1 ] shows the spatial distribution of members with and without infrared excesses . as shown in the upper right panel , the space density has a peak with a fwhm of 0.25@xmath33 ( 1.9 pc ) . the upper - left panel shows the ratio of systems with disks to the total members in radial distance bins defined by the semi - circles east - west versus the radial distance from the center of the cluster . solid circles ( + dotted line ) represent the fraction of thick disks , while open circles ( + dashed line ) represent the total fraction of disk - bearing stars , defined in [ res : sed ] . the error bars represents the statistical @xmath34 error in our disk fractions , where n is the number of stars with disks . similarly , the lower - right panel shows the fraction of disks in radial distance bins defined by the semi - circles north - south . in contrast to the result of @xcite , using k and l band observations , there is evidence for a higher disk fraction near the cluster center . however , we can not rule out a small , relatively uniform contamination of non - members which could also produce this effect . there is a slight increase in the disk fraction in the north - east direction ; in the direction toward ngc2024 ( age @xmath5 1 myr , d@xmath0440 pc ; * ? ? ? * ; * ? ? ? * ) , located at @xmath01.0 deg from the @xmath1 orionis center ; this fraction could be affected by disk - bearing members of this younger stellar group . since ngc2024 is centrally concentred , the fraction could be also affected by stars extends toward the horsehead nebula ( located at @xmath00.6 deg from the @xmath1 orionis center ) , which also shows several class ii objects in separate _ maps @xcite . spectroscopic studies are necessary to test whether the disk fractions in the outer regions are affected by non - members . a sub - sample of 81 stars have x - ray luminosities from @xcite , including 8 evolved disks , 21 optically thick disks and 52 class iii stars . figure [ fig : alphax ] shows the cumulative distribution of class iii stars and stars with optically thick disks as a function of x - ray luminosity . the significance level of 64% in a kolmogorov - smirnov ( ks ) test indicates that class iii stars have similar x - ray luminosities as stars with disks . this suggests that the presence of optically thick disks have small or no influence on x - ray emission , such as might be caused by an accretion shock , in agreement with the results from @xcite . stars selected as uncertain members are relatively faint ( [email protected] ) . from 133 stars in this sample , only 11 objects have irac sed slopes consistent with the presence of disks ( 10 thick disk + 1 evolved disks ) . so , the fraction of disk - bearing stars for the uncertain members ( [email protected] % ) is very low compared with the fraction of disk - bearing stars expected for members with j@xmath313 in the @xmath1 orionis cluster ( [email protected] ) . assuming that the source of non - members are class iii stars and the disk - bearing stars are actual members of the cluster , we can expect that at least 84 class iii stars ( @xmath070 % of the uncertain member sample ) do not belong to the @xmath1 orionis cluster and are likely to be field m dwarfs or other sources of contaminations . additional spectroscopic studies are essential to determine membership for this sample . we have used the irac and mips instruments on board the _ spitzer _ space telescope to study the frequencies and properties of disks around 336 members of the 3 myr old @xmath1 orionis cluster , with results for another 133 objects of uncertain membership . we confirm and strongly improve upon previous results showing that disk fractions are much lower around a - type and intermediate - mass t tauri stars , implying that disk evolution is much more rapid for stars with masses @xmath35 . in particular , we find that only a few percent of all a - f stars have optically thick disks , with perhaps another @xmath36 % having detectable optically - thin , probable debris disks . this fraction of probable debris disk is lower that those found in the older orion ob1b ( 38% ) and ob1a ( 46% ) sub - associations , suggesting that @xmath1 orionis cluster could be in a phase between the clearing of the primordial disk and the formation of large icy objects , which produce the second - generation dust observed in debris disk systems @xcite . the disk frequency @xmath37% we find for t tauri stars is reasonably consistent with studies of other populations , perhaps a bit lower . we also find no evidence that the disk fraction decreases significantly among the highest - mass brown dwarfs . overall , disk frequencies follows the overall trend of decreasing disk frequency with increasing age observed in other young stellar populations . about 15% of the t tauri disk systems exhibit weak irac excesses inconsistent with simple optically - thick disk models . if these systems are in the process of clearing small dust particles from their inner disks , the observed fraction suggests that the timescale for this evolution is @xmath38 myr . we also find that the infrared excesses among optically - thick disks in @xmath1 ori are systematically lower than the excesses observed from younger taurus molecular cloud stars . this result confirms and strongly extends previous results indicating an overall decrease in the level of emission of optically - thick disks with age , which may indicate an overall settling of dust particles to the disk midplanes , or grain growth , or both . the populous nature and relatively near distance of the @xmath1 orionis cluster allows us to improve the statistics of these correlations , providing an important step toward a full characterization of protoplanetary disk evolution in clustered environments , in which most stars form . finally , we do not find correlation between x ray luminosity and the disk infrared excess , suggesting that the presence of optically thick disks have small or not influence on the x ray emission . thanks to r.d . jeffries , d. stern and c. lada for insightful communications . this publication makes use of data products from the cida equatorial variability survey , obtained with he j. stock telescope at the venezuela national astronomical observatory , which is operated by cida for the ministerio de ciencia y tecnologa of venezuela , and 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 . this work is based on observations made with the _ spitzer space telescope _ ( go-1 0037 and go-1 0058 ) , which is operated by the jet propulsion laboratory , california institute of technology under a contract with nasa . support for this work was provided by university of michigan and nasa grant nag5 - 13210 . the @xmath1 orionis cluster has been the subject of several photometric and spectroscopic studies to identify members of its population .. analyzing the photometric and kinematic properties of the bright population in the cluster , @xcite determined an angular size of 0.40 degrees and identified 7 stars with photometric and kinematic membership probabilities larger than 50% . since the cluster is reasonably close , relatively young and has low reddening , many of the previous studies have focused on identifying low mass stars and substellar objects . adopting an age interval of 2 - 7 myr and the hipparcos distance of the star @xmath1 ori ( 352@xmath7 pc ) , @xcite presented several lists of low mass stars , brown dwarfs and free - floating planets belonging to the @xmath1 orionis cluster , selected from optical and nir color - magnitude diagrams . adopting an age of 2.5 myr and the distance calculated for ob1b subassociation ( 440 pc ) , @xcite reported photometric membership for a sample of low mass stars ( 193 objects ) based on the v versus v - i color - magnitude diagram . @xcite found that the cluster has @xmath0160 members in the mass range from 0.2 @xmath4 to 1.0 @xmath4 . adopting an age of 3 myr and a distance of 350 pc , @xcite reported photometric membership for 135 candidates of very low mass stars ( @xmath50.4 @xmath4 ) using the color magnitude diagrams , i versus i - j , and photometric variability . @xcite presented intermediate and low resolution spectra including the h@xmath21 and @xmath396708 lines and confirmed 25 low mass stars and 2 brown dwarfs as members of the cluster . @xcite confirmed 51 low mass stars and brown dwarfs as members using low resolution spectroscopic analysis of forbidden lines , as well as @xmath396708 and h@xmath21 lines . @xcite presented an intermediate - resolution spectroscopic study of 76 photometrically selected stars ranging in mass from 0.04 @xmath4 to 0.3 @xmath4 ; 57 stars were confirmed as members on the basis that they exhibit @xmath396708 in absorption , a weak doublet ( 8183,8195 ) indicating low surface gravity consistent with pms status , and radial velocities expected for the kinematic properties of the cluster . @xcite confirmed as members 38 low - mass stars using high - resolution spectroscopic analysis of doublet . @xcite and @xcite showed that optical photometric selection alone is very effective to identify low mass members belonging to the cluster , with small contamination by non - members for stars with i@xmath517 magnitudes . recently , @xcite compiled 59 spectroscopic confirmed members of the cluster and studied the circumstellar disks of the sample using nir data . @xcite detect 175 x - ray sources down to the substellar limit . half of this sample could be new members of the cluster based on the x - ray emission as indicator of youth . megeath , s. t. , et al . 2005b , massive star birth : a crossroads of astrophysics , iau symposium proceedings of the international astronomical union 227 , held 16 - 20 may , italy , edited by cesaroni , r. ; felli , m. ; churchwell , e. ; walmsley , m. cambridge : cambridge university press , pp.383 - 388 . llllllllllllll 9 & 84.32781 & -2.67275 & 10.99 @xmath8 0.03 & 10.97 @xmath8 0.03 & 10.80 @xmath8 0.03 & 10.92 @xmath8 0.03 & & 15.40 @xmath8 0.05 & q & 0 & & iii & 9 + 27 & 84.34599 & -2.54623 & 13.00 @xmath8 0.03 & 12.94 @xmath8 0.03 & 12.86 @xmath8 0.04 & 12.99 @xmath8 0.05 & & 18.42 @xmath8 0.02 & 9 & & & iii & + 41 & 84.35401 & -2.86730 & 10.30 @xmath8 0.03 & 10.31 @xmath8 0.03 & 10.28 @xmath8 0.03 & 10.25 @xmath8 0.03 & & & & & & iii & + 52 & 84.36209 & -2.36497 & 10.54 @xmath8 0.03 & 10.56 @xmath8 0.03 & 10.50 @xmath8 0.03 & 10.49 @xmath8 0.03 & & 13.77 @xmath8 0.08 & q & 0 & & iii & + 59 & 84.36683 & -2.60175 & 12.65 @xmath8 0.03 & 12.62 @xmath8 0.03 & 12.51 @xmath8 0.04 & 12.52 @xmath8 0.04 & & 17.64 @xmath8 0.05 & q & 0 & & iii & + 60 & 84.36792 & -2.40508 & 12.87 @xmath8 0.03 & 12.82 @xmath8 0.03 & 12.82 @xmath8 0.04 & 12.80 @xmath8 0.05 & & 18.08 @xmath8 0.11 & q & 0 & & iii & 9 + 73 & 84.37888 & -2.39521 & 10.90 @xmath8 0.03 & 10.44 @xmath8 0.03 & 10.04 @xmath8 0.03 & 9.11 @xmath8 0.03 & 5.64 @xmath8 0.03 & 16.28 @xmath8 0.10 & q & 1 & & ii & 9 + 74 & 84.37941 & -2.52922 & 9.46 @xmath8 0.03 & 9.39 @xmath8 0.03 & 9.30 @xmath8 0.03 & 9.32 @xmath8 0.03 & 9.19 @xmath8 0.04 & & & & & iii & + 77 & 84.38132 & -2.40748 & 10.99 @xmath8 0.03 & 10.98 @xmath8 0.03 & 10.93 @xmath8 0.03 & 10.93 @xmath8 0.03 & & 15.83 @xmath8 0.11 & q & 1 & & iii & 9 + 82 & 84.38269 & -2.75514 & 9.24 @xmath8 0.03 & 9.21 @xmath8 0.03 & 9.16 @xmath8 0.03 & 9.20 @xmath8 0.03 & 8.95 @xmath8 0.04 & 9.42 @xmath8 0.02 & & & & iii & + 92 & 84.38811 & -2.89332 & 10.59 @xmath8 0.03 & 10.74 @xmath8 0.03 & 10.64 @xmath8 0.03 & 10.55 @xmath8 0.03 & & 14.99 @xmath8 0.04 & q & 0 & & iii & 9 + 115 & 84.40181 & -2.24377 & 10.74 @xmath8 0.03 & 10.72 @xmath8 0.03 & 10.67 @xmath8 0.03 & 10.70 @xmath8 0.03 & & 15.49 @xmath8 0.10 & q & 0 & & iii & 9 + 116 & 84.40192 & -2.69906 & 14.19 @xmath8 0.03 & 14.02 @xmath8 0.03 & 13.98 @xmath8 0.06 & 13.97 @xmath8 0.09 & & & & & & iii & 1,2,3,6 + 117 & 84.40270 & -2.56669 & 11.82 @xmath8 0.03 & 11.83 @xmath8 0.03 & 11.81 @xmath8 0.03 & 11.76 @xmath8 0.04 & & 17.12 @xmath8 0.05 & q & 0 & & iii & 9 + 123 & 84.40760 & -2.76226 & 11.51 @xmath8 0.03 & 11.47 @xmath8 0.03 & 11.40 @xmath8 0.03 & 11.44 @xmath8 0.03 & & 16.37 @xmath8 0.05 & q & 1 & & iii & 9 + 124 & 84.40766 & -2.83598 & 9.49 @xmath8 0.03 & 9.54 @xmath8 0.03 & 9.45 @xmath8 0.03 & 9.44 @xmath8 0.03 & 9.65 @xmath8 0.05 & & & & & iii & + 147 & 84.42406 & -2.48557 & 9.51 @xmath8 0.03 & 9.50 @xmath8 0.03 & 9.47 @xmath8 0.03 & 9.53 @xmath8 0.03 & 9.59 @xmath8 0.04 & 10.90 @xmath8 0.07 & 7 & & & iii & + 164 & 84.43706 & -2.49920 & 9.91 @xmath8 0.03 & 9.90 @xmath8 0.03 & 9.91 @xmath8 0.03 & 9.89 @xmath8 0.03 & 9.80 @xmath8 0.05 & 10.95 @xmath8 0.09 & 7 & & & iii & + 165 & 84.43856 & -2.48111 & 12.65 @xmath8 0.03 & 12.58 @xmath8 0.03 & 12.55 @xmath8 0.04 & 12.49 @xmath8 0.04 & & 18.76 @xmath8 0.13 & 9 & & & iii & 9 + 168 & 84.43889 & -2.73678 & 8.96 @xmath8 0.03 & 9.02 @xmath8 0.03 & 8.95 @xmath8 0.03 & 8.95 @xmath8 0.03 & 8.94 @xmath8 0.04 & 10.66 @xmath8 0.07 & 7 & & & iii & + 194 & 84.45671 & -2.60506 & 10.12 @xmath8 0.03 & 10.17 @xmath8 0.03 & 10.06 @xmath8 0.03 & 10.10 @xmath8 0.03 & 10.03 @xmath8 0.06 & & & & & iii & + 209 & 84.46288 & -2.43542 & 13.64 @xmath8 0.03 & 13.53 @xmath8 0.03 & 13.68 @xmath8 0.05 & 13.38 @xmath8 0.07 & & & & & & iii & 2,3 + 214 & 84.46497 & -2.59045 & 10.81 @xmath8 0.03 & 10.80 @xmath8 0.03 & 10.79 @xmath8 0.03 & 10.72 @xmath8 0.03 & & 15.64 @xmath8 0.01 & 9 & & 1 & iii & 8,9 + 219 & 84.46685 & -2.60126 & 13.92 @xmath8 0.03 & 13.81 @xmath8 0.03 & 13.73 @xmath8 0.05 & 13.70 @xmath8 0.07 & & & & & 0 & iii & 6 + 220 & 84.46747 & -2.56052 & 11.85 @xmath8 0.03 & 11.85 @xmath8 0.03 & 11.78 @xmath8 0.03 & 11.77 @xmath8 0.04 & & 16.89 @xmath8 0.03 & 9 & & 0 & iii & 9,12 + 229 & 84.47091 & -2.55951 & 9.45 @xmath8 0.03 & 9.46 @xmath8 0.03 & 9.43 @xmath8 0.03 & 9.42 @xmath8 0.03 & 9.40 @xmath8 0.04 & & & & 1 & iii & + 240 & 84.47515 & -2.74462 & 11.92 @xmath8 0.03 & 11.88 @xmath8 0.03 & 11.86 @xmath8 0.03 & 11.82 @xmath8 0.03 & & 17.02 @xmath8 0.05 & q & 0 & & iii & 8,9 + 243 & 84.47665 & -2.65823 & 8.51 @xmath8 0.03 & 8.68 @xmath8 0.03 & 8.55 @xmath8 0.03 & 8.51 @xmath8 0.03 & 8.47 @xmath8 0.03 & 10.91 @xmath8 0.07 & 7 & & 1 & iii & 7 + 244 & 84.47683 & -2.72711 & 8.99 @xmath8 0.03 & 9.06 @xmath8 0.03 & 8.98 @xmath8 0.03 & 8.96 @xmath8 0.03 & 8.98 @xmath8 0.04 & & & & & iii & + 247 & 84.47852 & -2.68587 & 12.04 @xmath8 0.03 & 11.80 @xmath8 0.03 & 11.45 @xmath8 0.03 & 10.82 @xmath8 0.03 & 8.41 @xmath8 0.03 & 18.55 @xmath8 0.06 & q & 0 & 0 & ii & 5,9 + 251 & 84.47960 & -2.46006 & 12.45 @xmath8 0.03 & 12.36 @xmath8 0.03 & 12.30 @xmath8 0.04 & 12.46 @xmath8 0.04 & & 18.49 @xmath8 0.10 & 9 & & & iii & 9 + 254 & 84.48156 & -2.55146 & 13.66 @xmath8 0.03 & 13.35 @xmath8 0.03 & 12.94 @xmath8 0.04 & 12.32 @xmath8 0.04 & 9.70 @xmath8 0.05 & & & & 0 & ii & 2,3,12 + 256 & 84.48218 & -2.40933 & 13.79 @xmath8 0.03 & 13.74 @xmath8 0.03 & 13.56 @xmath8 0.06 & 13.77 @xmath8 0.10 & & & & & & iii & 2,3 + 260 & 84.48530 & -2.73263 & 11.91 @xmath8 0.03 & 11.96 @xmath8 0.03 & 11.85 @xmath8 0.03 & 11.86 @xmath8 0.03 & & 16.84 @xmath8 0.05 & q & 0 & & iii & 9 + 271 & 84.48935 & -2.64562 & 12.92 @xmath8 0.03 & 12.76 @xmath8 0.03 & 12.58 @xmath8 0.04 & 12.03 @xmath8 0.04 & 9.40 @xmath8 0.04 & & & & 0 & ii & 1,2,3,6,8,12 + 275 & 84.49076 & -2.91668 & 11.62 @xmath8 0.03 & 11.62 @xmath8 0.03 & 11.60 @xmath8 0.03 & 11.55 @xmath8 0.03 & & 14.86 @xmath8 0.04 & q & 1 & & iii & + 283 & 84.49328 & -2.69059 & 12.06 @xmath8 0.03 & 11.96 @xmath8 0.03 & 11.90 @xmath8 0.03 & 11.96 @xmath8 0.04 & & 18.92 @xmath8 0.07 & q & 0 & 0 & iii & 6,9,12 + 293 & 84.49872 & -2.85090 & 9.61 @xmath8 0.03 & 9.70 @xmath8 0.03 & 9.62 @xmath8 0.03 & 9.59 @xmath8 0.03 & 9.67 @xmath8 0.04 & & & & & iii & + 297 & 84.50228 & -2.75267 & 11.56 @xmath8 0.03 & 11.47 @xmath8 0.03 & 11.47 @xmath8 0.03 & 11.43 @xmath8 0.03 & & 17.75 @xmath8 0.06 & q & 0 & 1 & iii & 9,12 + 299 & 84.50394 & -2.43548 & 11.68 @xmath8 0.03 & 11.76 @xmath8 0.03 & 11.66 @xmath8 0.03 & 11.33 @xmath8 0.03 & 6.76 @xmath8 0.03 & 16.70 @xmath8 0.11 & q & 1 & & td@xmath9 & 9 + 300 & 84.50434 & -2.76042 & 10.12 @xmath8 0.03 & 9.70 @xmath8 0.03 & 9.29 @xmath8 0.03 & 8.64 @xmath8 0.03 & 5.87 @xmath8 0.03 & 17.59 @xmath8 0.07 & q & 1 & & ii & 9 + lllllllll 33 & 84.34808 & -2.74109 & 14.49 @xmath8 0.03 & 14.46 @xmath8 0.04 & 14.42 @xmath8 0.07 & 14.44 @xmath8 0.15 & & iii + 47 & 84.35836 & -2.34652 & 13.53 @xmath8 0.03 & 13.55 @xmath8 0.03 & 13.47 @xmath8 0.05 & 13.80 @xmath8 0.11 & & iii + 57 & 84.36604 & -2.51190 & 13.45 @xmath8 0.03 & 13.45 @xmath8 0.03 & 13.33 @xmath8 0.04 & 13.52 @xmath8 0.08 & & iii + 71 & 84.37812 & -2.40533 & 14.45 @xmath8 0.03 & 14.48 @xmath8 0.04 & 14.51 @xmath8 0.11 & 14.21 @xmath8 0.13 & & iii + 81 & 84.38234 & -2.40128 & 14.36 @xmath8 0.03 & 14.27 @xmath8 0.04 & 14.19 @xmath8 0.07 & 14.13 @xmath8 0.14 & & iii + 87 & 84.38534 & -2.75422 & 13.67 @xmath8 0.03 & 13.70 @xmath8 0.04 & 13.86 @xmath8 0.08 & 13.96 @xmath8 0.15 & & iii + 107 & 84.39636 & -2.44933 & 12.10 @xmath8 0.03 & 12.01 @xmath8 0.03 & 11.91 @xmath8 0.03 & 11.94 @xmath8 0.03 & & iii + 120 & 84.40674 & -2.42888 & 13.48 @xmath8 0.03 & 13.49 @xmath8 0.03 & 13.35 @xmath8 0.05 & 13.30 @xmath8 0.09 & 9.95 @xmath8 0.05 & td@xmath40 + 219 & 84.46685 & -2.60126 & 13.92 @xmath8 0.03 & 13.81 @xmath8 0.03 & 13.73 @xmath8 0.05 & 13.70 @xmath8 0.07 & & iii@xmath41 + 127 & 84.40806 & -2.80576 & 12.78 @xmath8 0.03 & 12.82 @xmath8 0.03 & 12.80 @xmath8 0.04 & 12.73 @xmath8 0.04 & & iii + 134 & 84.41515 & -2.30742 & 14.14 @xmath8 0.03 & 13.93 @xmath8 0.03 & 13.87 @xmath8 0.06 & 13.74 @xmath8 0.08 & & iii + 154 & 84.43077 & -2.44282 & 14.48 @xmath8 0.03 & 14.45 @xmath8 0.04 & 14.35 @xmath8 0.08 & 14.15 @xmath8 0.09 & & iii + 174 & 84.44174 & -2.45832 & 13.71 @xmath8 0.03 & 13.63 @xmath8 0.03 & 13.61 @xmath8 0.05 & 13.53 @xmath8 0.08 & & iii + 188 & 84.45503 & -2.36186 & 14.14 @xmath8 0.03 & 14.17 @xmath8 0.04 & 13.94 @xmath8 0.06 & 14.06 @xmath8 0.13 & & iii + 207 & 84.46257 & -2.32070 & 14.13 @xmath8 0.03 & 14.18 @xmath8 0.03 & 14.14 @xmath8 0.07 & 14.20 @xmath8 0.11 & & iii + 221 & 84.46809 & -2.44730 & 14.24 @xmath8 0.03 & 14.22 @xmath8 0.03 & 14.29 @xmath8 0.07 & 14.10 @xmath8 0.12 & & iii + 236 & 84.47451 & -2.48824 & 14.48 @xmath8 0.03 & 14.46 @xmath8 0.04 & 14.07 @xmath8 0.07 & 14.13 @xmath8 0.13 & & iii + ccccc 0.44 & -1.506 & -1.254 & -1.654 & 22 + 0.55 & -1.002 & -0.842 & -1.274 & 51 + 0.64 & -0.738 & -0.588 & -0.928 & 60 + 0.79 & -0.308 & -0.220 & -0.376 & 52 + 1.24 & 0.0 & 0.0 & 0.0 & 64 + 1.66 & -0.037 & -0.005 & -0.065 & 64 + 2.16 & -0.209 & -0.125 & -0.245 & 64 + 3.6 & -0.563 & -0.400 & -0.645 & 64 + 4.5 & -0.733 & -0.580 & -0.836 & 64 + 5.8 & -0.910 & -0.748 & -0.978 & 64 + 8.0 & -0.962 & -0.804 & -1.076 & 64 + 24 & -1.238 & -1.029 & -1.430 & 58 + ccccc 0.44 & -1.360 & -1.206 & -1.550 & 60 + 0.55 & -0.863 & -0.610 & -1.166 & 116 + 0.64 & -0.597 & -0.400 & -0.870 & 128 + 0.79 & -0.204 & -0.144 & -0.324 & 97 + 1.235 & 0 & 0 & 0 & 137 + 1.662 & -0.057 & -0.037 & -0.073 & 137 + 2.159 & -0.257 & -0.241 & -0.273 & 137 + 3.6 & -0.748 & -0.733 & -0.777 & 137 + 4.5 & -1.038 & -1.007 & -1.083 & 137 + 5.8 & -1.306 & -1.278 & -1.348 & 137 + 8.0 & -1.693 & -1.660 & -1.723 & 137 + 24 & -3.259 & -3.198 & -3.431 & 42@xmath40 + ccc haebe & [email protected] & [email protected] + imtts & [email protected] & [email protected] + tts & [email protected] & [email protected] + bdwarfs & [email protected] & [email protected] + total & [email protected] & [email protected] +

+ the random - phase approximation with second - order screened exchange ( rpa+sosex ) is a model of electron correlation energy with two caveats : its accuracy depends on an arbitrary choice of mean field , and it scales as operations and memory for electrons . we derive a new algorithm that reduces its scaling to operations and memory using controlled approximations and a new self - consistent field that approximates brueckner coupled - cluster doubles ( bccd ) theory with rpa+sosex , referred to as brueckner rpa ( brpa ) theory . the algorithm comparably reduces the scaling of second - order mller - plesset ( mp2 ) perturbation theory with smaller cost prefactors than rpa+sosex . within a semiempirical model , we study h dissociation to test accuracy and h rings to verify scaling .

1303.3847

density functional theory ( dft ) and coupled - cluster ( cc ) theory are the two dominant paradigms for the computation of many - electron ground states with complementary capabilities . both theories are built upon the independent - orbital and self - consistent field ( scf ) structure of hartree - fock ( hf ) theory @xcite . dft proves the existence of an exact density functional @xcite for which there are approximations @xcite enabling routine simulation of thousands of electrons @xcite that scale as @xmath1 operations and @xmath3 memory for @xmath2 electrons . cc theory is a systematically improvable hierarchy of methods @xcite indexed by @xmath6 that scale as @xmath7 operations and @xmath8 memory . in practice , dft is limited by accuracy and cc theory is limited by cost . the random phase approximation ( rpa ) is a natural point of convergence in the ongoing development of cc theory and dft . rpa emerges from truncated versions of cc methods as reduced - complexity methods that retain a significant fraction of the cc correlation energy @xcite . rpa also occurs when dft is used to approximate the polarization function in the adiabatic - connection fluctuation - dissipation formula for the correlation energy @xcite . this formula might become a part of more accurate density functionals , but it has a higher cost and complexity @xcite than conventional density functionals . in both cases , a balance of cost and accuracy is being sought with rpa . the electron correlation energy model that we consider in this paper is the rpa correlation energy plus the second - order screened exchange ( sosex ) energy @xcite . rpa+sosex is exact for one electron and to second order in perturbation theory . it agrees with quantum monte carlo benchmarks of the uniform electron gas @xcite to within 0.002 ha / electron @xcite . benchmarks on inhomogeneous systems show mean absolute errors of 0.002 ha / atom for cohesive energies of solids in a 5-solid test set @xcite , 0.008 ha / molecule for atomization energies of molecules in the g2 - 1 test set @xcite , and 0.010 ha / atom for correlation energies of first and second row atoms @xcite . these atomic and molecular rpa+sosex results approach but fail to surpass the accuracy of second - order mller - plesset ( mp2 ) perturbation theory @xcite and the b3lyp density functional @xcite . also , any rpa+sosex energy depends on a mean field choice and is not unique @xcite . the goal of this paper is to further develop rpa+sosex as a compromise between the cost of dft and the accuracy of cc theory . known algorithms for rpa+sosex @xcite and similar models @xcite require @xmath0 operations and either @xmath1 or @xmath9 memory . by utilizing an auxiliary basis set @xcite , fast interaction kernel summation @xcite , and a low - rank approximation of energy denominators @xcite , we design a new algorithm to reduce the cost to @xmath1 operations and @xmath3 memory . we define a precise rpa+sosex total energy with a unique choice of mean field based on brueckner orbitals @xcite that is designed to approximate brueckner coupled - cluster doubles ( bccd ) theory @xcite . this is referred to as brueckner rpa ( brpa ) theory . it is important to be pragmatic about the short - term value of brpa theory and a fast rpa+sosex algorithm . despite its construction , brpa is not a good approximation of bccd . the approximations used to retain the rpa+sosex form are too crude . for a given basis , a fast rpa+sosex calculation will have a large cost relative to conventional dft . with the continued reduction of average errors in density functionals @xcite and a correlation of dft and rpa error outliers @xcite , it is unclear whether the cost is warranted without additional research . the paper proceeds as follows . brpa theory is derived in sec . [ brpatheory ] as a truncation of bccd theory . the scf structure of brpa theory is emphasized . the three main components of a fast rpa+sosex algorithm are introduced in sec . [ fast_components ] . this includes a nonstandard choice of primary and auxiliary basis , which simplifies the cost accounting and algorithm design . in sec . [ rpastructure ] , the tensor structure in existing rpa algorithms @xcite is reviewed and extended to rpa+sosex . the novel structure of rpa+sosex theory enables a significant reduction in the number of its variables . fast and conventional rpa+sosex algorithms are designed in sec . [ fast_design ] as pseudocode . a detailed leading - order cost analysis is given instead of a crude scaling analysis . we study applications to a semiempirical h@xmath5 model in sec . [ applications ] . the accuracy of brpa theory is tested on h@xmath4 , and the scaling of rpa+sosex algorithms is confirmed with h@xmath5 calculations for large @xmath2 . the implications for future work are discussed in sec . [ discussion ] . this includes implementation in established electronic structure codes , basis set convergence problems , and further development of correlation models . we conclude in sec . [ conclusions ] with a brief consideration of rpa as a distinct electronic structure paradigm . the common origin of all post - hf methods is the many - electron hamiltonian in second quantization notation @xcite , @xmath10 where @xmath11 are the fermion lowering operators of spin - orbitals . spin - orbital indices are labeled by @xmath12 . any tensor or product of tensors with repeated indices has an implicit sum over those indices unless they appear unrepeated in a tensor or product of tensors or in an explicit sum in the same equation . this notation is related to the standard bracket notation @xcite as @xmath13 where @xmath14 and @xmath15 are spin - space coordinates , @xmath16 is the single - electron hamiltonian , @xmath17 is the inter - electron interaction , @xmath18 are the spin - orbital wavefunctions , and @xmath19 is a tensor antisymmetrization operation . the operators have symmetries @xmath20 and @xmath21 . we define a reference slater determinant , @xmath22 , by splitting the set of orbitals into virtual orbitals labeled by @xmath23 and @xmath2 occupied orbitals labeled by @xmath24 . @xmath22 is defined by @xmath25 uniquely up to a global phase . when combined with fermion anticommutation , @xmath26 and @xmath27 , ( [ phi_define ] ) is sufficient to calculate all expectation values of @xmath22 . hf theory posits the minimization of @xmath28 and diagonalization of particle and hole subspaces , @xmath29 and @xmath30 . for local minima in @xmath31 , this is equivalent to diagonalization of a fock matrix , @xmath32 or equivalently a fock operator in spin - space , @xmath33 with the canonical scf structure between @xmath18 and @xmath34 . the total energy consistent with the fock matrix is @xmath35 the orbital energies are approximate excitation energies , @xmath36 which is known as koopmans theorem @xcite . the basic structure of cc theory is to solve @xmath37 approximately as a projected eigenvalue problem @xcite . it defines a cluster operator , @xmath38 , from the linear span of a set of operators , @xmath39 , omitting identity , @xmath40 , which only alters the normalization , @xmath41 at the singles - and - doubles level of theory ( ccsd ) , @xmath42 is the hf ground state and @xmath43 . in its brueckner variant ( bccd ) , @xmath42 is varied and @xmath44 . we use nonstandard notation to simplify the tensor form of the bccd equations . the first is @xmath45 with partial symmetry constraints : @xmath46 but not @xmath47 . the second is a non - hermitian brueckner matrix @xcite , @xmath48 this reduces the total energy to a form similar to eq . ( [ hf_energy ] ) , @xmath49 also , the single - excitation equations simplify to @xmath50 . the most complicated part of bccd theory is the double - excitation equations . we separate out a `` ring '' tensor , @xmath51 which simplifies the remaining equations to @xmath52 only @xmath53 appears in the bccd tensor equations . changes in @xmath54 that do not alter @xmath53 are a redundancy in the theory . we construct brpa theory by truncating the bccd tensor equations to extract the rpa+sosex correlation model . we crudely attempt to minimize errors by minimizing the number of truncations . only eq . ( [ bccd_double ] ) is truncated by ( 1 ) removing the `` ladder '' terms on the second line of the equation , ( 2 ) reducing the ring tensor to a `` direct - ring '' tensor , @xmath55 , for @xmath56 and ( 3 ) reducing @xmath57 to its hermitian part , @xmath58 , to guarantee orthogonal orbitals with real energies . the resulting equations depend on @xmath54 but only define @xmath53 . to define @xmath54 , we expand the set of equations by removing all ` @xmath59 ' , @xmath60 these are known as the rpa riccati equations @xcite . they have a unique , positive - definite `` stabilizing '' solution @xcite that evolves over @xmath61 from the unique @xmath62 solution to @xmath63 . in conjunction with eqs . ( [ direct_ring ] ) and ( [ brpa_double ] ) , it is convenient to describe brpa theory with scf structure . we expand eq . ( [ fock ] ) to the diagonalization of a generalized fock matrix , @xmath64 with a static and hermitian self - energy matrix , @xmath65 , in analogy to many - body green s function theory @xcite . with the choice @xmath66 eq . ( [ general_fock ] ) contains @xmath50 , eqs . ( [ hf_energy ] ) and ( [ cc_energy ] ) are analogous , and @xmath67 diagonalization enables a rearrangement of eq . ( [ brpa_double ] ) , @xmath68 it is convenient to write @xmath65 using a non - hermitian matrix , @xmath69 whereas @xmath70 in hf theory . a generalized scf cycle is used for brpa calculations . for a given @xmath71 , we solve eq . ( [ diag_riccati ] ) iteratively for @xmath54 as an inner scf cycle . we then calculate @xmath72 from @xmath54 , which is used to recalculate @xmath71 in the outer scf cycle . this is appropriate for reducing the number of @xmath72 calculations when there is a large relative cost to calculate @xmath72 over @xmath54 . the scf cycle is summarized as @xmath73 . the inner scf cycle can be avoided by using a first - order approximation for @xmath54 instead of solving eq . ( [ diag_riccati ] ) , @xmath74 @xmath75 in eq . ( [ sigma ] ) defines @xmath76 , which is similar to a canonical transformation mp2 ( ct - mp2 ) theory @xcite . other theories use @xmath77 to calculate @xmath78 but calculate @xmath71 using an inconsistent @xmath65 . brueckner cc2 ( bcc2 ) theory @xcite uses @xmath79 and conventional mp2 theory uses @xmath70 . a generalized koopmans theorem @xcite applies when @xmath78 and @xmath71 are calculated from the same @xmath54 and @xmath65 , @xmath80 which generalizes eq . ( [ koopmans ] ) . we assume @xmath57 to be real for @xmath81 . this partially corrects for the absence of electron correlation in koopmans theorem , but it still has an `` orbital relaxation '' error . because solutions to the brpa equations depend on the choice of orbital occupations , @xmath82 is not exactly the difference between a pair of converged brpa total energies . in contrast , many - body green s function theory is formally exact @xcite . none of the three components of the fast brpa algorithm are fundamentally new . however , some details of their use are nonstandard . we review these details and how they compare with modern standards in the gaussian - orbital and planewave - pseudopotential electronic structure methodologies . in this paper , we use a grid in @xmath14 with @xmath83 points as both the primary and auxiliary basis , labeled by @xmath84 . @xmath85 is a basis set efficiency factor . the relation between primary and orbital fermion operators is @xmath86 with orthonormal structure , @xmath87 . single - electron operators in the primary basis are @xmath88 . this notation transparently interchanges with basis - free notation ( e.g. @xmath89 ) . in terms of @xmath90 , @xmath91 with number operators , @xmath92 , and a `` kernel '' , @xmath93 , from @xmath94 this is similar to a tensor hypercontraction ( thc ) form @xcite for @xmath95 however , here we use it as a theoretical basis to simplify accounting and not necessarily as a computational basis . a computational basis must enable decomposition of @xmath95 similar to eq . ( [ v_thc ] ) to be suitable for a fast brpa calculation . its kernel , @xmath93 , must be amenable to fast summation methods discussed in sec . [ fastvsum ] . a general form , @xmath96 , is acceptable if the `` vertex '' , @xmath97 , is sparse . in @xmath97 , @xmath98 are primary basis indices , and @xmath99 is an auxiliary basis index . these primary and auxiliary bases need not be orthogonal or equal , but we assume both to simplify method development . in gaussian - orbital electronic structure , both primary and auxiliary basis functions are atom - centered polynomials times gaussians . the resolution - of - identity ( ri ) form of @xmath95 is @xcite @xmath100 for primary , @xmath101 , and auxiliary , @xmath102 , basis functions and a matrix inverse , ` @xmath103 ' . the ri vertex is not sparse , which is also the case in cholesky @xcite and pseudospectral @xcite decompositions . the thc vertex is sparse but requires @xmath104 operations to be calculated at present @xcite . in a standard ri - rpa calculation @xcite , @xmath105 for the cc - pvtz basis and @xmath106 for its auxiliary basis @xcite of a frozen - core c atom ( @xmath107 ) . in planewave - pseudopotential electronic structure , all the basis functions are planewaves , but pseudopotentials augment the primary basis inside atomic spheres @xcite . no augmentation of the auxiliary basis is a source of errors when core - valence polarization is important @xcite . planewaves have a sparse vertex because the fast fourier transform ( fft ) enables efficient grid operations . also , the coulomb kernel , @xmath93 , is diagonal in the planewave basis . in a standard planewave rpa calculation @xcite , @xmath108 for the primary basis and @xmath109 for the auxiliary basis of a frozen - core c atom in diamond at equilibrium . in this paper , we build low - rank approximations to energy denominators using numerical quadratures of three integrals . the first integral is over the imaginary frequency axis , @xmath110 with @xmath111 and a quadrature of @xmath112 points , @xmath113 , and weights , @xmath114 , indexed by @xmath115 . we use a permuted index , @xmath116 , to write a required quadrature symmetry : @xmath117 and @xmath118 . the remaining two integrals are @xmath119 with closed counterclockwise contours , @xmath120 separating @xmath121 from @xmath122 , and @xmath123 separating @xmath124 from @xmath125 . their quadratures are @xmath126 points , @xmath127 , indexed by @xmath128 and @xmath129 points , @xmath130 , indexed by @xmath131 with different weights , @xmath132 , for each @xmath113 . we combine @xmath133 and @xmath134 into one index , @xmath135 , for notational convenience . with a 4-parameter model of the orbital energy spectrum , @xmath136 $ ] and @xmath137 $ ] , and a target error , @xmath138 , we analytically construct numerical quadratures in appendix [ quadrature_section ] . comparable to other results @xcite , the quadrature sizes are @xmath139 without an explicit @xmath2-dependence . when @xmath140 is zero or very small , a low - energy cutoff must be introduced . in conjunction with eq . ( [ quadrature1 ] ) , the fast rpa algorithm also requires an approximate product closure relation , @xmath141 for @xmath142 , this equation is closed by partial fractions , @xmath143 for @xmath144 , this requires coefficients , @xmath145 , of a finite difference approximation to @xmath146 for @xmath147 . it is a standard linear approximation problem that we solve by minimizing a root - mean - square ( rms ) error metric , @xmath148 in practice , we find that @xmath149 is proportional to @xmath138 . numerical quadratures of the laplace transform @xcite are an alternative low - rank energy denominator approximation , @xmath150 numerical methods for calculating quadratures are known @xcite . the main advantage of eq . ( [ laplace_quadrature ] ) over eqs . ( [ quadrature1 ] ) and ( [ quadrature2 ] ) is the single quadrature summation rather than nested summations . however , eqs . ( [ quadrature1 ] ) and ( [ quadrature2 ] ) have better numerical behavior in the case of complex orbital energies and lead to equations with connections to many - body green s function theory @xcite . kernel interpolation ( i.e. skeleton decomposition @xcite ) is yet another low - rank approximation of energy denominators , @xmath151_{ef } \frac{1}{\omega_{bj } + \omega_f } , \notag \\ [ \mathbf{k}]_{ef } & = \frac{1}{\omega_e + \omega_f } .\end{aligned}\ ] ] it is not based on reduction of an integral to quadrature , thus it does not have a simple exact limiting expression . however , it is exact if @xmath152 or @xmath153 is in @xmath154 and may be efficient for treating isolated spectral features and large interior energy gaps . in this paper , fast @xmath93 summation methods are accounted for by assigning @xmath155 operations to @xmath156 . @xmath157 is an efficiency factor with either weak or no @xmath2-dependence for fast methods . further details are not important for cost analysis . fast @xmath93 summation methods are ubiquitous in electronic structure . ffts are used in planewave - pseudopotential codes to solve the poisson equation and apply the hamiltonian to an orbital . the fast multiple method ( fmm ) is used in gaussian - orbital codes to efficiently calculate the matrix elements of the hartree potential @xcite . except for planewave calculations of the fock exchange @xcite , these fast methods are not a bottleneck and their @xmath157 values do not contribute to the leading - order cost . the leading - order cost of the fast brpa algorithm has a dependence on @xmath157 . this is commonly the case for classical @xmath2-body simulations of electrostatics in molecular dynamics and gravity in astrophysics . as a result , the method development in those fields has been driven to develop new fast summation techniques . such modern innovations include the accelerated cartesian expansion @xcite ( ace ) , multilevel summation method @xcite ( msm ) , and hierarchical matrix decompositions @xcite . however , electronic structure applications have special requirements for summation methods such as the ability to treat point nuclei , smooth valence electron charge , and the multiple length scales of core electron charge in a unified framework . applications such as brpa might motivate this type of development . fast summation methods exist for other @xmath93 besides the coulomb kernel . the fast brpa algorithm applies to any @xmath93 for which @xmath157 is small for any reason . such generality might be appealing , but further improvements to accuracy or efficiency could result from better use of structure within the coulomb kernel and elementary operations beyond just @xmath156 . an example is the low off - diagonal rank of the coulomb kernel shared by a general class of structured matrices @xcite . there are three distinct approaches to calculating the rpa correlation energy . the first is calculating @xmath54 by solving the rpa riccati equation in eq . ( [ brpa_double ] ) and evaluating @xmath158 the rpa riccati equation is equivalent to the rpa symplectic eigenvalue problem @xcite that determines electron - hole excitation energies . @xmath159 is also half the sum of the difference between these excitation energies and the corresponding energies in the tamm - dancoff approximation @xcite . these energies are the poles of frequency - dependent polarization functions and this sum is extracted from them by the residue theorem when used on the adiabatic - connection fluctuation - dissipation ( acfd ) formula for @xmath159 . this formula can be rearranged as @xmath160 where @xmath161 is the correlated part of the rpa two - body density matrix averaged over interaction strength @xcite . we use the well - studied structure of @xmath161 as a reference point for the discussion of structure in @xmath54 , which has yet to be elucidated . in an auxiliary basis , the acfd formula for @xmath159 is @xcite @xmath162 , \ ] ] where @xmath163 is the rpa polarization function for interaction strength @xmath164 . @xmath163 is defined in relation to its @xmath165 value as @xmath166 the screened coulomb interaction is related to @xmath163 as @xmath167 using eq . ( [ w_def2 ] ) , we encapsulate @xmath164 in eq . ( [ acfd_rpa ] ) with @xmath168_{xy } = p_{xz}^0(\omega ) v_{zy } , \notag \\ \overline{w}_{xy}(\omega ) & = - 2 v_{xz } [ \mathbf{e}(\omega)^{-1 } + \mathbf{e}(\omega)^{-2 } \ln(\mathbf{i } - \mathbf{e}(\omega))]_{zy } , \end{aligned}\ ] ] and extract @xmath161 by splitting @xmath169 over its internal indices , @xmath170 other forms @xcite of @xmath161 combine the ` @xmath171 ' and ` @xmath172 ' sectors and use different but equivalent energy denominators . from observing eqs . ( [ diag_riccati ] ) , ( [ quadrature1 ] ) , and ( [ c_def ] ) , it is reasonable to expect similar tensor structure in @xmath161 and @xmath54 . to that end , we expand @xmath95 in eq . ( [ diag_riccati ] ) using eq . ( [ v_thc ] ) and regroup terms , @xmath173 this can be rewritten to define @xmath174 without reference to @xmath54 , [ almost_dyson ] @xmath175 the iteration of eq . ( [ s_def ] ) starting from @xmath176 elucidates a factorization ansatz that further simplifies @xmath174 , @xmath177 it enables a reduction of eq . ( [ almost_dyson ] ) to dyson - like equations , [ new_dyson]@xmath178 . \label{b_def}\end{aligned}\ ] ] this uses a divided difference , @xmath179 = [ f(x ) - f(y ) ] / ( x - y)$ ] , and an analytic reconstruction of @xmath180 with the form @xmath181 for a closed counterclockwise contour @xmath182 separating @xmath183 from the poles of @xmath184 , which are disjoint if @xmath180 is finite . we maximize the superficial similarity of @xmath161 and @xmath54 by again regrouping terms in @xmath54 and applying eq . ( [ quadrature1 ] ) , @xmath185 while both @xmath161 and @xmath54 produce the same value of @xmath159 , their corresponding sosex - like values , @xmath186 and @xmath187 , are not equal and begin to differ at third order in perturbation theory @xcite . however , @xmath188 and @xmath189 are numerically similar for small molecules @xcite . in a brpa calculation based on sec . [ scf_brpa ] , @xmath54 will appear in its own calculation and also in the calculation of @xmath72 . there are @xmath190 variables in @xmath54 , which is the memory bottleneck . direct calculation of @xmath174 in eq . ( [ almost_dyson ] ) and of @xmath72 from @xmath174 will reduce the number of variables to @xmath191 if the full storage of @xmath192 is avoided . @xmath180 is not directly useful because it contains @xmath193 variables . nevertheless , the interpolation of @xmath180 from an effective quadrature of eq . ( [ reconstruct ] ) reduces it to @xmath194 variables and enables fast algorithms . here we design fast and conventional algorithms for mp2 and rpa+sosex calculations . in each of these four designs , we begin by identifying the tensor equations to be evaluated . many of these equations are sufficiently complicated that an algorithm for efficient evaluation is not immediately obvious . we use simple pseudocode to decompose these equations into intermediate variables and elementary operations . algorithm costs are then easy to account for in the pseudocode . without the simple primary and auxiliary basis sets in sec . [ basis_sets ] , this design process becomes significantly more difficult . even so , the results here demonstrate what is possible with a practical basis set if sufficient effort is given to algorithm design . the pseudocode is grouped into functions . memory usage is delineated at the beginning of the function declaration with inputs then outputs separated by a semicolon in the function argument and a workspace statement that lists all temporary variables . the body of each function contains ` for ' loops and elementary ` @xmath195 ' operations that denote the calculation of expression @xmath196 and its storage in variable @xmath197 . each bottleneck is commented with its leading - order operation count in the cost polynomial of @xmath198 . we count as one operation each addition , multiplication , and division . real and complex are not distinguished in operation and variable counts . the basic design strategies for minimizing cost are reuse of calculations and avoidance of concurrent data storage . all computational costs that are leading order in @xmath2 are arranged as either matrix - matrix multiplications with the minimum matrix dimension maximized or as fast @xmath93 summations . this choice hides the cost of data movement in efficient implementations of linear algebra and @xmath93 summation . these strategies suffice for the design of serial algorithms , but parallel algorithms will need to explicitly account for communication costs . a common element of the four algorithms is calculation of @xmath77 or @xmath72 directly in the primary basis . the scf equations from sec . [ scf_brpa ] in the primary basis are analogous to eq . ( [ scf_space ] ) , [ main ] @xmath199 with @xmath78 and @xmath65 also calculated in the primary basis . algorithm [ mp2c ] calculates @xmath200 using eqs . ( [ t0 ] ) and ( [ sigma_xy ] ) by substituting @xmath201 and @xmath75 . it is comparable to an ri - mp2 algorithm @xcite when eq . ( [ main ] ) is used to calculate @xmath78 from @xmath200 . both require @xmath202 operations , but the @xmath191 memory cost of ri - mp2 reduces to an @xmath203 memory cost by avoiding storage of the dense ri vertex in eq . ( [ ri_vertex ] ) . the intermediate calculation of @xmath200 instead of a direct calculation of @xmath78 enables the self - consistent mp2 methods in sec . [ scf_brpa ] . @xmath204 @xmath205 @xmath206 @xmath207 @xmath208 @xmath209 @xmath210 @xmath211 @xmath212 @xmath213 @xmath214 @xmath215 because of the @xmath54 structure found in sec . [ t_struct ] , the scf inner loop proposed in sec . [ scf_brpa ] only needs to determine @xmath174 rather than @xmath54 . we rearrange eq . ( [ almost_dyson ] ) into a residual tensor , @xmath216 and recast the scf inner loop as solving @xmath217 . algorithm [ inbrpac ] calculates @xmath218 with @xmath202 operations and @xmath191 memory . as in the mp2 case , the operation count is the same as other rpa+sosex algorithms @xcite , but avoiding the storage of @xmath54 reduces the @xmath190 memory cost . algorithm [ outbrpac ] calculates @xmath219 using eqs . ( [ t_factor ] ) and ( [ sigma_xy ] ) . it is comparable in cost with the @xmath218 inner loop calculations , which violates the assumption in sec . [ scf_brpa ] of an inexpensive inner loop . as a result , it is more efficient to calculate @xmath218 and @xmath219 concurrently and solve the orbital self - consistency and @xmath217 problems simultaneously . there will be one iterative cycle as opposed to inner iterations nested within outer iterations . this is an example of modifying the overall design of an algorithm based on a change in the relative cost of its components . @xmath220 @xmath221 @xmath222 @xmath223 @xmath224 @xmath212 @xmath225 @xmath226 @xmath227 @xmath228 @xmath222 @xmath229 @xmath230 @xmath231 @xmath232 @xmath233 @xmath234 @xmath235 @xmath236 @xmath237 mp2 corresponds to @xmath238 in terms of the structure discussed in sec . [ t_struct ] . however , @xmath239 is still an element of fast mp2 calculations . to efficiently calculate @xmath239 , we introduce a mean - field green s function , @xmath240 with eqs . ( [ quadrature1 ] ) and ( [ quadrature2 ] ) , we reduce @xmath239 in eq . ( [ a0_def ] ) to [ a0_simple ] @xmath241 given @xmath242 , we need @xmath243 operations to form @xmath244 . repeated application of eqs . ( [ quadrature1 ] ) and ( [ quadrature2 ] ) to @xmath200 enables its decomposition into @xmath242 and rearrangement into @xmath245 @xmath200 is calculated by algorithm [ mp2f ] in @xmath246 operations and @xmath247 memory . this improves on the @xmath193 operations used by thc - mp2 @xcite and @xmath104 operations used by atomic - orbital laplace mp2 with integral screening @xcite . neither method uses fast @xmath93 summation , which enables the extra factor of @xmath2 speedup in the new algorithm . @xmath248 @xmath249 @xmath250 @xmath251 @xmath252 @xmath253 @xmath254 @xmath255 @xmath256 @xmath257 @xmath258 @xmath259 @xmath260 @xmath261 @xmath262 @xmath263 @xmath264 @xmath265 @xmath266 @xmath267 @xmath268 @xmath269 @xmath270 @xmath271 as in eq . ( [ a0_split ] ) , we use eq . ( [ quadrature1 ] ) to approximate @xmath192 with @xmath272 and postulate a similar form for @xmath180 , @xmath273 with @xmath274 . using eq . ( [ closure ] ) , this reduces eq . ( [ new_dyson ] ) to @xmath275 for @xmath276 . these equations are equivalent to @xmath277 for @xmath278 @xmath279 is calculated by algorithm [ inbrpaf ] in @xmath280 operations and @xmath194 memory . compared to the conventional algorithm in sec . [ brpac_section ] , this is more efficient in operations when @xmath281 and in memory when @xmath282 . the calculation of @xmath283 at quadrature points in eq . ( [ c_def ] ) is noniterative and has the same cost , but we lack efficient formulas relating @xmath283 to @xmath219 . @xmath284 @xmath285 @xmath286 @xmath287 @xmath288 @xmath289 @xmath290 @xmath291 @xmath292 @xmath293 @xmath294 @xmath295 @xmath290 @xmath296 @xmath297 in terms of half - transformed green s functions , @xmath298 eqs . ( [ quadrature1 ] ) , ( [ quadrature2 ] ) , ( [ closure ] ) , ( [ t_factor ] ) , and ( [ a_ansatz ] ) reduce @xmath54 to @xmath299 this bears some resemblance to the approximate thc - ccsd form @xcite of @xmath54 . @xmath300 appears in the brpa analog of eq . ( [ fast_sigma0 ] ) , @xmath301 @xmath219 is calculated by algorithm [ outbrpaf ] using @xmath302 operations and @xmath247 memory . a fast brpa calculation of @xmath219 is a factor of @xmath303 more expensive than a fast mp2 calculation of @xmath200 in the @xmath304 regime . the calculations in eq . ( [ fast_sigma ] ) are especially difficult when memory limitations prohibit concurrent storage of @xmath300 , @xmath305 , and @xmath306 . full evaluation of @xmath306 requires @xmath307 operations , and without concurrent storage it must be evaluated twice overall . without the simple primary and auxiliary basis structure from sec . [ basis_sets ] , pseudocode that is equivalent to algorithm [ outbrpaf ] will become significantly longer and more difficult to design . @xmath308 \mathbf{workspace } : \ { & u^e_f , w_f^e , x_f^e , y^e_f , z_f^e , x_{\underline{a}\underline{i}}^e , v_x^e , x_x^{ef } , \\ & x_x^{\underline{a}\underline{i } } , x_x^{e\underline{p } } , y_x^{e\underline{p } } , w_{xy}^e , x_{xy}^e , x_{xy}^{\underline{i } } , f_{xy}^{\underline{p } } \ } \end{aligned}$ ] @xmath309 @xmath310 @xmath311 @xmath312 @xmath313 @xmath314 @xmath315 @xmath316 @xmath317 @xmath318 @xmath319 @xmath320 @xmath321 @xmath259 @xmath322 @xmath323 @xmath324 @xmath325 @xmath326 @xmath327 @xmath328 @xmath329 @xmath330 @xmath331 @xmath332 @xmath333 @xmath334 @xmath335 @xmath336 @xmath337 @xmath334 @xmath338 @xmath339 @xmath340 @xmath341 before committing further to its development , it is prudent to test the accuracy of brpa theory against other popular total energy methods and performance of the fast mp2 and brpa algorithms against their conventional counterparts . direct use of the algorithms in sec . [ fast_design ] is possible for hamiltonians that are limited to a zero - differential - overlap ( zdo ) form @xcite . there have been proposals to calculate electron correlation within a semiempirical framework @xcite , but they remains undeveloped . we consider h@xmath5 because it gives us fine control over the number of electrons and also because h@xmath4 is the simplest two - electron molecule with an internal coordinate . the zdo form of h@xmath4 is an extended hubbard model with @xmath342 and @xmath343 \normalsize , \ \ \ \mathbf{v } = \scriptsize \left [ \begin{array}{cccc } u & v & u & v \\ v & u & v & u \\ u & v & u & v \\ v & u & v & u \end{array } \right ] \normalsize , \ ] ] in the notation of eq . ( [ aux_h ] ) . it has a natural pairwise extension to the @xmath344 case . separate h@xmath4 and h@xmath5 hamiltonians are fit in appendix [ hn_fit ] . the h@xmath4 version is fit to reproduce exact , hf , and mp2 total energies . the h@xmath5 version is fit to a simple distance - dependent form that gives proper asymptotic behavior . these models have limited transferability , which is a standard caveat of semiempirical modeling with simple hamiltonian forms . symmetric dissociation of h@xmath4 is used as a critical test of new electron correlation methods @xcite . the model in eq . ( [ h_h2 ] ) has simple formulas for hf , mp2 , brpa , and exact energies , @xmath345 dissociation curves are shown in fig . [ h2energy]a . brpa is a uniform improvement over hf but has three times the error of mp2 or b3lyp at equilibrium . these methods fail at dissociation , but brpa is stable at twice the b3lyp error as mp2 diverges . rpa+sosex total energies are strongly modulated by the choice of mean field . as shown in fig . [ h2energy]c , errors in h@xmath4 near equilibrium can be corrected by significantly reducing the gap . this is typical behavior , and the smaller orbital gaps in dft reference mean fields systematically improve total energies @xcite over hf - based calculations @xcite . reducing gaps to reduce errors conflicts with the generalized koopmans theorem in eq . ( [ general_koopmans ] ) that requires the brpa gap to approximate the physical gap . on h@xmath4 , brpa increases the hf gap towards the exact gap , but some methods for calculating excitations such as @xmath346 theory decrease the hf gap @xcite . one possible resolution is to derive an rpa+sosex model where orbital energies are not required to approximate electron addition and removal energies , such as by modifying @xmath163 in eq . ( [ acfd_rpa ] ) with exchange terms @xcite . dissociation failures can be fixed by enabling symmetry breaking in the reference state . as shown in fig . [ h2energy]d , the error now peaks at around twice the equilibrium bond length . mp2 now has uniformly less error than brpa , but brpa fixes the derivative discontinuity in the mp2 energy surface . symmetry breaking is needed for size consistency at this level of theory when considering the dissociation of a closed - shell molecule into open - shell fragments . it also serves as a simple model of static correlation . broken symmetries can be restored with a projection @xcite , but the resulting total energy corrections are not size consistent . an accurate , size - consistent model of electron correlation should be able to repair symmetries broken by the reference state accurately but not necessarily exactly . a single example such as h@xmath4 is not enough to determine error statistics and compare the average errors of total energy methods . however , while errors on a system as important and simple as h@xmath4 remain large , it suffices to discount methods . 99 a. szabo and n. s. ostlund , _ modern quantum chemistry : introduction to advanced electronic structure theory _ ( dover , new york , 1996 ) . p. hohenberg and w. kohn , http://dx.doi.org/10.1103/physrev.136.b864[phys . rev . ] * 136 * , b864 ( 1964 ) . w. kohn and l. j. sham , http://dx.doi.org/10.1103/physrev.140.a1133[phys . * 140 * , a1133 ( 1965 ) . j. p. perdew and a. zunger , http://dx.doi.org/10.1103/physrevb.23.5048[phys . rev . b ] * 23 * , 5048 ( 1981 ) . r. briddon and m. j. rayson , http://dx.doi.org/10.1002/pssb.201046147[phys . status solidi b ] * 248 * , 1309 ( 2011 ) . r. j. bartlett and m. musia , http://dx.doi.org/10.1103/revmodphys.79.291[rev . phys . ] * 79 * , 291 ( 2007 ) . g. e. scuseria , t. m. henderson , and d. c. sorensen , http://dx.doi.org/10.1063/1.3043729[j . chem . phys . ] * 129 * , 231101 ( 2008 ) . f. furche , http://dx.doi.org/10.1063/1.2977789[j . chem . phys . ] * 129 * , 114105 ( 2008 ) . h. eshuis , j. yarkony , and f. furche , http://dx.doi.org/10.1063/1.3442749[j . chem . phys . ] * 132 * , 234114 ( 2010 ) . a. grneis , m. marsman , j. harl , l. schimka , and g. kresse , http://dx.doi.org/10.1063/1.3250347[j . chem . ] http://dx.doi.org/10.1063/1.3250347[phys . ] * 131 * , 154115 ( 2009 ) . d. m. ceperley and b. j. alder , http://dx.doi.org/10.1103/physrevlett.45.566[phys . lett . ] * 45 * , 566 ( 1980 ) . d. l. freeman , http://dx.doi.org/10.1103/physrevb.15.5512[phys . b ] * 15 * , 5512 ( 1977 ) . j. paier , b. g. janesko , t. m. henderson , g. e. scuseria , a. grneis , and g. kresse , http://dx.doi.org/10.1063/1.3317437[j . chem . phys . ] * 132 * , 094103 ( 2010 ) ; * 133 * , 179902 ( 2010 ) . a. heelmann , http://dx.doi.org/10.1063/1.3590916[j . chem . phys . ] * 134 * , 204107 ( 2011 ) . x. ren , p. rinke , g. e. scuseria , and m. scheffler , http://dx.doi.org/10.1103/physrevb.88.035120[phys . b ] * 88 * , 035120 ( 2013 ) . j. paier , x. ren , p. rinke , g. e. scuseria , a. grneis , g. kresse , and m. scheffler , http://dx.doi.org/10.1088/1367-2630/14/4/043002[new j. phys . ] * 14 * , 043002 ( 2012 ) . c. van alsenoy , http://dx.doi.org/10.1002/jcc.540090607[j . chem . ] * 9 * , 620 ( 1988 ) . l. greengard and v. rokhlin , http://dx.doi.org/10.1016/0021-9991(87)90140-9[j . comput . phys . ] * 73 * , 325 ( 1987 ) . j. almlf , http://dx.doi.org/10.1016/0009-2614(91)80078-c[chem . phys . lett . ] * 181 * , 319 ( 1991 ) . r. k. nesbet , http://dx.doi.org/10.1103/physrev.109.1632[phys . rev . ] * 109 * , 1632 ( 1958 ) . l. z. stolarczyk and h. j. monkhorst , http://dx.doi.org/10.1002/qua.560260827[int . j. quantum chem . , symp . ] * 26 * , 267 ( 1984 ) . l. goerigkab and s. grimme , http://dx.doi.org/10.1039/c0cp02984j[phys . chem . phys . ] * 13 * , 6670 ( 2011 ) . a. j. cohen , p. mori - snchez , and w. yang , http://dx.doi.org/10.1021/cr200107z[chem . rev . ] * 112 * , 289 ( 2012 ) . t. koopmans , http://dx.doi.org/10.1016/s0031-8914(34)90011-2[physica ] * 1 * , 104 ( 1934 ) . ( in german ) b. p. molinari , http://dx.doi.org/10.1137/0311021[siam j. control ] * 11 * , 262 ( 1973 ) . t. n. lan and t. yanai , http://dx.doi.org/10.1063/1.4809983[j . chem . phys . ] * 138 * , 224108 ( 2013 ) . y. akinaga , y. kawashima , and s. ten - no , http://dx.doi.org/10.1016/j.cplett.2011.02.044[chem . phys . lett . ] * 506 * , 276 ( 2011 ) . i. lindgren and s. salomonson , http://dx.doi.org/10.1002/qua.944[int . j. quantum chem . ] * 90 * , 294 ( 2002 ) . l. hedin , http://dx.doi.org/10.1103/physrev.139.a796[phys . rev . ] * 139 * , a796 ( 1965 ) . r. m. parrish , e. g. hohenstein , t. j. martnez , and c. d. sherrill , http://dx.doi.org/10.1063/1.4768233[j . chem . ] http://dx.doi.org/10.1063/1.4768233[phys . ] * 137 * , 224106 ( 2012 ) . h. koch , a. snchez de mers , and t. b. pedersen , http://dx.doi.org/10.1063/1.1578621[j . phys . ] * 118 * , 9481 ( 2003 ) . t. j. martnez and e. a. carter , http://dx.doi.org/10.1063/1.466350[j . chem . phys . ] * 100 * , 3631 ( 1994 ) . f. weigend , a. khn , and c. httig , http://dx.doi.org/10.1063/1.1445115[j . ] * 116 * , 3175 ( 2002 ) . e. blchl , http://dx.doi.org/10.1103/physrevb.50.17953[phys . b ] * 50 * , 17953 ( 1994 ) . j. harl , l. schimka , and g. kresse , http://dx.doi.org/10.1103/physrevb.81.115126[phys . b ] * 81 * , 115126 ( 2010 ) . the @xmath85 values are generated by vasp using the reported geometries and planewave cutoffs . for an energy cutoff @xmath78 in ha and a volume per electron @xmath347 in bohr , @xmath348 using basic accounting of planewaves inside a cubic unit cell in real space and a spherical fermi sea in momentum space . l. lin , j. lu , l. ying , and w. e , http://dx.doi.org/10.1007/s11401-009-0201-7[chinese ann . b ] * 30 * , 729 ( 2009 ) . a. takatsuka , s. ten - no , and w. hackbusch , http://dx.doi.org/10.1063/1.2958921[j . phys . ] * 129 * , 044112 ( 2008 ) . s. a. goreinov , e. e. tyrtyshnikov , and n. l. zamarashkin , http://dx.doi.org/10.1016/s0024-3795(96)00301-1[linear algebra ] http://dx.doi.org/10.1016/s0024-3795(96)00301-1[appl . ] * 261 * , 1 ( 1997 ) . c. a. white , b. g. johnson , p. m.w . gill , and m. head - gordon , http://dx.doi.org/10.1016/0009-2614(94)01128-1[chem . ] http://dx.doi.org/10.1016/0009-2614(94)01128-1[phys . lett . ] * 230 * , 8 ( 1994 ) . j. paier , r. hirschl , m. marsman , and g. kresse , http://dx.doi.org/10.1063/1.1926272[j . chem . phys . ] * 122 * , 234102 ( 2005 ) . b. shanker and h. huang , http://dx.doi.org/10.1016/j.jcp.2007.04.033[j . comput . phys . ] * 226 * , 732 ( 2007 ) . r. d. skeel , i. tezcan , and d. j. hardy , http://dx.doi.org/10.1002/jcc.10072[j . chem . ] * 23 * , 673 ( 2002 ) . s. brm , l. grasedyck , and w. hackbusch , http://dx.doi.org/10.1016/s0955-7997(02)00152-2[eng . * 27 * , 405 ( 2003 ) . g. jansen , r .- f . liu , and j. g. ngyn , http://dx.doi.org/10.1063/1.3481575[j . ] * 133 * , 154106 ( 2010 ) . m. feyereisen , g. fitzgerald , and a. komornicki , http://dx.doi.org/10.1016/0009-2614(93)87156-w[chem . phys . lett . ] * 208 * , 359 ( 1993 ) . e. g. hohenstein , r. m. parrish , and t. j. martnez , http://dx.doi.org/10.1063/1.4732310[j . chem . phys . ] * 137 * , 044103 ( 2012 ) . m. hser , http://dx.doi.org/10.1007/bf01113535[theor . acta ] * 87 * , 147 ( 1993 ) . e. g. hohenstein , r. m. parrish , c. d. sherrill , and t. j. martnez , http://dx.doi.org/10.1063/1.4768241[j . chem . ] http://dx.doi.org/10.1063/1.4768241[phys . ] * 137 * , 221101 ( 2012 ) . i. fischer - hjalmars , http://dx.doi.org/10.1063/1.1696232[j . phys . ] * 42 * , 1962 ( 1965 ) . w. thiel , http://dx.doi.org/10.1021/ja00396a021[j . chem . soc . ] * 103 * , 1413 ( 1981 ) . t. m. henderson and g. e. scuseria , http://dx.doi.org/10.1080/00268976.2010.507227[mol . phys . ] * 108 * , 2511 ( 2010 ) . p. mori - snchez , a. j. cohen , and w. yang , http://dx.doi.org/10.1103/physreva.85.042507[phys . a ] * 85 * , 042507 ( 2012 ) . f. caruso , d. r. rohr , m. hellgren , x. ren , p. rinke , a. rubio , and m. scheffler , http://dx.doi.org/10.1103/physrevlett.110.146403[phys . lett . ] * 110 * , 146403 ( 2013 ) . a. heelmann and a. grling , http://dx.doi.org/10.1080/00268976.2011.614282[mol . phys . ] * 109 * , 2473 ( 2011 ) . j. hubbard , http://dx.doi.org/10.1098/rspa.1958.0003[proc . a ] * 243 * , 336 ( 1957 ) . c. a. jimnez - hoyos , t. m. henderson , t. tsuchimochi , and g. e. scuseria , http://dx.doi.org/10.1063/1.4705280[j . chem . phys . ] * 136 * , 164109 ( 2012 ) . see the supplementary material at http://arxiv.org/e-print/1303.3847 for a c implementation of the algorithms in sec . [ fast_design ] . e. anderson , z. bai , c. bischof , s. blackford , j. demmel , j. dongarra , j. du croz , a. greenbaum , s. hammarling , a. mckenney , and d. sorensen , http://www.netlib.org/lapack/lug/[_lapack users guide , third edition _ ] ( siam , philadelphia , 1999 ) . j. j. dongarra , j. du croz , i. s. duff , and s. hammarling , http://dx.doi.org/10.1145/77626.79170[acm trans . ] http://dx.doi.org/10.1145/77626.79170[math . * 16 * , 1 ( 1990 ) . in inbrpaf , tensors of the form @xmath349 and @xmath350 have symmetries @xmath351 and @xmath352 . mp2f contains these symmetries and adds @xmath353 , @xmath354 , and @xmath355 . outbrpaf contains these symmetries and adds @xmath356 . only half the entries of these symmetric tensors need to be stored and calculated , which halves the memory and operation counts . complex arithmetic doubles memory usage and fft runtimes and triples matrix multiplication runtimes over real arithmetic . the additional cost of complex arithmetic is mitigated for all the operations except matrix multiplication ( zgemm ) , which is reduced to 1.5 times the cost model . m. frigo and s. g. johnson , http://dx.doi.org/10.1109/jproc.2004.840301[proc . ieee ] * 93 * , 216 ( 2005 ) . the leading - order cost model assumes that all parameters are much larger than one , which is false for @xmath342 . we correct the conventional algorithm costs with @xmath357 for the operation counts , @xmath358 for the mp2 memory cost , and @xmath359 for both brpa memory costs . d. g. truhlar , http://dx.doi.org/10.1016/s0009-2614(98)00866-5[chem . phys . lett . ] * 294 * , 45 ( 1998 ) . t. kato , http://dx.doi.org/10.1002/cpa.3160100201[commun . pure appl . ] * 10 * , 151 ( 1957 ) . l. kong , f. a. bischoff , and e. f. valeev , http://dx.doi.org/10.1021/cr200204r[chem . rev . ] * 112 * , 75 ( 2012 ) . n. d. drummond , m. d. towler , and r. j. needs , http://dx.doi.org/10.1103/physrevb.70.235119[phys . b ] * 70 * , 235119 ( 2004 ) . s. goedecker , http://dx.doi.org/10.1103/revmodphys.71.1085[rev . ] * 71 * , 1085 ( 1999 ) . m. gell - mann and k. a. brueckner , http://dx.doi.org/10.1103/physrev.106.364[phys . rev . ] * 106 * , 364 ( 1957 ) . k. sawada , http://dx.doi.org/10.1103/physrev.106.372[phys . ] * 106 * , 372 ( 1957 ) . m. baranger , http://dx.doi.org/10.1103/physrev.120.957[phys . rev . ] * 120 * , 957 ( 1960 ) . a. d. mclachlan and m. a. ball , http://dx.doi.org/10.1103/revmodphys.36.844[rev . phys . ] * 36 * , 844 ( 1964 ) . b. jancovici and d. h. schiff , http://dx.doi.org/10.1016/0029-5582(64)90578-4[nucl . phys . ] * 58 * , 678 ( 1964 ) . a. j. glick , h. j. lipkin , and n. meshkov , http://dx.doi.org/10.1016/0029-5582(65)90864-3[nucl . phys . ] * 62 * , 211 ( 1965 ) . j. harris and r. o. jones , http://dx.doi.org/10.1088/0305-4608/4/8/013[j . f ] * 4 * , 1170 ( 1974 ) . h. eshuis , j. e. bates , and f. furche , http://dx.doi.org/10.1007/s00214-011-1084-8[theor . * 131 * , 1084 ( 2012 ) . x. ren , p. rinke , c. joas , and m. scheffler , link:10.1007/s10853 - 012 - 6570 - 4[j . sci . ] * 47 * , 7447 ( 2012 ) . p. y. ayala and g. e. scuseria , http://dx.doi.org/10.1063/1.478256[j . chem . phys . ] * 110 * , 3660 ( 1999 ) . s. y. willow , k. s. kim , and s. hirata , http://dx.doi.org/10.1063/1.4768697[j . chem . phys . ] * 137 * , 204122 ( 2012 ) . d. neuhauser , e. rabani , and r. baer , http://dx.doi.org/10.1021/ct300946j[j . theory comput . ] * 9 * , 24 ( 2013 ) . j. p. perdew and k. schmidt , http://dx.doi.org/10.1063/1.1390175[aip conf . proc . ] * 577 * , 1 ( 2001 ) . e. i. zolotarev , zap . nauk , st . petersburg * 30 * , 5 ( 1877 ) ; reprinted in _ collected works _ ( izdat . akad . nauk sssr , leningrad , 1932 ) , vol . 2 , pp . 1 - 59 . ( in russian ) a. a. gonar , http://dx.doi.org/10.1070/sm1969v007n04abeh001107[math . ussr sb . ] * 7 * , 623 ( 1969 ) . m. j. frisch , g. w. trucks , h. b. schlegel , g. e. scuseria , m. a. robb , j. r. cheeseman , g. scalmani , v. barone , b. mennucci , g. a. petersson , h. nakatsuji , m. caricato , x. li , h. p. hratchian , a. f. izmaylov , j. bloino , g. zheng , j. l. sonnenberg , m. hada , m. ehara , k. toyota , r. fukuda , j. hasegawa , m. ishida , t. nakajima , y. honda , o. kitao , h. nakai , t. vreven , j. a. montgomery , jr . , j. e. peralta , f. ogliaro , m. bearpark , j. j. heyd , e. brothers , k. n. kudin , v. n. staroverov , r. kobayashi , j. normand , k. raghavachari , a. rendell , j. c. burant , s. s. iyengar , j. tomasi , m. cossi , n. rega , j. m. millam , m. klene , j. e. knox , j. b. cross , v. bakken , c. adamo , j. jaramillo , r. gomperts , r. e. stratmann , o. yazyev , a. j. austin , r. cammi , c. pomelli , j. w. ochterski , r. l. martin , k. morokuma , v. g. zakrzewski , g. a. voth , p. salvador , j. j. dannenberg , s. dapprich , a. d. daniels , . farkas , j. b. foresman , j. v. ortiz , j. cioslowski , and d. j. fox , http://www.gaussian.com/[gaussian 09 , revision c.01 ] , gaussian , inc . , wallingford ct , 2009 .

we investigate the accretion of angular momentum onto a protoplanet system using three - dimensional hydrodynamical simulations . we consider a local region around a protoplanet in a protoplanetary disk with sufficient spatial resolution . we describe the structure of the gas flow onto and around the protoplanet in detail . we find that the gas flows onto the protoplanet system in the vertical direction crossing the shock front near the hill radius of the protoplanet , which is qualitatively different from the picture established by two - dimensional simulations . the specific angular momentum of the gas accreted by the protoplanet system increases with the protoplanet mass . at jovian orbit , when the protoplanet mass is , where is jovian mass , the specific angular momentum increases as . on the other hand , it increases as when the protoplanet mass is . the stronger dependence of the specific angular momentum on the protoplanet mass for is due to thermal pressure of the gas . the estimated total angular momentum of a system of a gas giant planet and a circumplanetary disk is two - orders of magnitude larger than those of the present gas giant planets in the solar system . a large fraction of the total angular momentum contributes to the formation of the circumplanetary disk . we also discuss the satellite formation from the circumplanetary disk .

0801.3305

until now , more than 200 extrasolar planets ( or exoplanets ) have been detected mainly by measuring the radial motion of their parent star along the line of sight . almost all exoplanets observed by this method are giant planets , like jupiter and saturn in our solar system , because massive planets are preferentially observed . although these planets are supposed to be formed in the disk surrounding the central star ( i.e. , circumstellar disk or protoplanetary disk ) , their formation process has not been fully understood yet . in the core accretion scenario @xcite , a solid core or protoplanet with @xmath6 , where @xmath7 is the earth mass , captures a massive gas envelope from the protoplanetary disk by self - gravity to become a gas giant planet . the evolution of the gaseous protoplanet has been studied with the approximation of spherical symmetry including radiative transfer ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? @xcite showed that rapid gas accretion is triggered when the solid core mass exceeds @xmath8 , and the protoplanet quickly increases their mass by gas accretion . however , the angular momentum of the accreting gas was ignored in these studies , because they assumed spherical symmetry . since the gas accretes onto the solid core with a certain amount of the angular momentum , a circumplanetary disk forms around the protoplanet , analogously to the formation of a protoplanetary disk around a protostar . the difference in the disk formation between the protostar and protoplanet is the region from which the central object acquires the angular momentum . the protostar acquires the angular momentum from a parent cloud , while the protoplanet acquires it from the shearing motion in the protoplanetary ( circumstellar ) disk . in addition , the gravitational sphere of the protostar spreads almost infinitely , while the gravitational sphere ( i.e. , the hill sphere ) of the protoplanet is limited in the region around the protoplanet because the gravity of the central star exceeds that of the protoplanet outside the hill sphere . the numerical simulations are useful to investigate the gas accretion onto a protoplanet and its circumplanetary disk ( hereafter we just call them as a protoplanet system ) . @xcite studied giant planet formation using one - dimensional quasi - spherical approximation with angular momentum transfer . they showed that as the protoplanet contracts , outer layers of the envelope containing sufficient specific angular momentum remain in bound orbit , and form a circumplanetary disk . however , owing to the spherical symmetry , the accretion flow from the protoplanetary disk to the protoplanet could not be investigated in their study . to study the accretion flow onto a protoplanet and the acquisition process of the angular momentum in detail , a multidimensional simulation is necessary . @xcite investigated the gas flow around a protoplanet with relatively low resolution , and found that the spin rotation vector of the protoplanet becomes parallel to the orbital rotation vector . recently , the flow pattern was carefully investigated by many authors @xcite . however , since the main purpose of these studies was to clarify the planet migration process in a large scale ( i.e. , outside the hill radius ) , they did not investigate the region near the protoplanet ( i.e. , inside the hill radius ) with sufficient resolution . thus , in their simulations , we can not study the gas stream inside the hill radius . to investigate the accretion flow onto the protoplanet system , we need to cover a large spatial scale from the region far from the hill sphere to that in the proximity to the protoplanet . for jupiter , since the hill radius is @xmath9 , where @xmath10 is jovian radius , we have to resolve at least @xmath111000 times different scales . to cover a large dynamical range in scales , a few authors used the nested - grid method . @xcite investigated the relation between the spiral patterns within the hill radius and migration rate using three - dimensional nested - grid code . although they resolved the region inside the hill radius , they did not investigate the structure in the proximity to the protoplanet because they adopted a sink cell at @xmath12 that corresponds to @xmath13 at jovian orbit ( 5.2 au ) . thus , we can not observe a circumplanetary disk at @xmath14 in their calculation . @xcite also investigated the gas flow around a protoplanet using two - dimensional nested - grid code . they resolved the region from @xmath15 to @xmath16 . they found that a circumplanetary disk with @xmath17 is formed around the protoplanet . @xcite and @xcite investigated the evolution of the protoplanet system using three - dimensional nested - grid code . they found that the gas flow pattern in three dimensions is qualitatively different from that in two dimensions : the gas is flowing into the protoplanet system only in the vertical direction in three - dimensional simulations . in the present study , we calculated the evolution of the protoplanet system using three - dimensional nested - grid code . we found that after the flow around the protoplanet reaches a steady state , the angular momentum accreting onto the protoplanet system is well converged regardless of both the cell width and the size of the sink cell region , while the mass accretion rate is not well converged . although we calculated the evolution of the protoplanet system with spatial resolution much higher than previous studies , we still need further higher spatial resolution to determine the mass accretion rate onto the protoplanet . thus , in this paper , we focus on the gas flow onto and around the protoplanet system and the accretion process of the angular momentum , and do not deal with the mass accretion rate ( we plan to investigate the mass accretion rate with higher spatial resolution using a higher - performance computer in a subsequent paper ) . note that the specific angular momentum accreting onto the protoplanet system with fixed protoplanet mass does not strongly depend on the mass accretion rate as described in the following sections . the structure of the paper is as follows . the frameworks of our models are given in 2 and the numerical method is described in 3 . the numerical results are presented in 4 . 5 is devoted for discussions . we summarize our conclusions in 6 . we consider a local region around a protoplanet using shearing sheet model ( e.g. , * ? ? ? we assume that the temperature is constant and the self - gravity of the disk is negligible . the orbit of the protoplanet is assumed to be circular on the equatorial plane of the circumstellar disk . we set up local rotating cartesian coordinates with the origin at the protoplanet and the @xmath18- , @xmath19- , and @xmath20-axis are radial , azimuthal , and vertical direction of the disk , respectively . we solve the equations of hydrodynamics without self - gravity : @xmath21 @xmath22 where @xmath23 , @xmath24 , @xmath25 , @xmath26 , and @xmath27 are the gas density , velocity , gas pressure , effective potential , and keplerian angular velocity of the protoplanet , respectively . in the above equations , the curvature terms are neglected . we adopt an isothermal equation of state , @xmath28 where @xmath29 is the sound speed . the keplerian angular velocity of the protoplanet is given by @xmath30 where @xmath31 , @xmath32 , and @xmath33 are the gravitational constant , mass of the central star , and orbital radius of the protoplanet , respectively . the effective potential @xmath26 is given by @xmath34 where @xmath0 and @xmath35 are the mass of the protoplanet , and the distance from the center of the protoplanet ( e.g. , * ? ? ? the first term is composed of the gravitational potential of the central star and the centrifugal potential , and higher orders in @xmath18 , @xmath19 and @xmath20 are neglected . the second term is the gravitational potential of the protoplanet . using the hill radius @xmath36 equation ( [ eq : phi ] ) can be rewritten as @xmath37 our initial settings are similar to @xcite and @xcite . the gas flow has a constant shear in the @xmath18-direction as @xmath38 for hydrostatic equilibrium , the density is given by @xmath39 where @xmath40 ( @xmath41 ) is the surface density of the unperturbed disk . the scale height @xmath42 is related to the sound speed @xmath29 as @xmath43 . in the standard solar nebular model @xcite , the temperature @xmath44 , sound speed @xmath29 , and gas density @xmath45 are given by @xmath46 where @xmath47 and @xmath48 are the protostellar and solar luminosities , @xmath49 where @xmath50 is the mean molecular weight of the gas composed mainly of h@xmath51 and he , and @xmath52 respectively . when @xmath53 and @xmath54 are adopted , using equations ( [ eq : omegap ] ) , ( [ eq : nebular - temp ] ) , and ( [ eq : nebular - cs ] ) , the scale height @xmath42 can be described as @xmath55 our basic equations can be normalized by unit time , @xmath56 and unit length , @xmath57 . the density is also scalable in equations ( 1 ) and ( 2 ) since we neglect the self - gravity of the disk in equation ( 2 ) . we normalize the density by @xmath58 . hereafter the normalized quantities are expressed with tilde on top , e.g. , @xmath59 , @xmath60 , @xmath61 , etc . the non - dimensional unperturbed velocity and density are given by @xmath62 @xmath63 thus , non - dimensional equations corresponding to equations ( [ eq : basic-1 ] ) , ( [ eq : basic-2 ] ) , ( [ eq : basic-3 ] ) , and ( [ eq : phi - eff ] ) are @xmath64 @xmath65 @xmath66 @xmath67 where @xmath68 is a unit vector directed to the @xmath20-axis . thus , the gas flow is characterized by only one parameter , the non - dimensional hill radius @xmath69 . in this paper , we adopt @xmath70 ( see , table 1 ) . as functions of the orbital radius and the mass of the central star , the parameter @xmath71 are related to the actual mass of protoplanet in the unit of jovian mass @xmath72 as @xmath73 for example , in the model with @xmath74 , @xmath75au and @xmath76 , the protoplanet mass is @xmath77 ( model m04 in table [ table : table1 ] ) . hereafter , we call model m04 ` the fiducial model . ' for each model , the protoplanet mass for @xmath75au and @xmath76 is presented in table [ table : table1 ] . in our parameter range , at jovian orbit ( @xmath75au ) , protoplanets have masses of @xmath78 . we will show our results assuming @xmath79 au and @xmath80 in the following . we will discuss the dependence on the orbital radius @xmath33 in [ sec : jupiter - saturn ] . in subsequent sections , we use non - dimensional quantities ( e.g. , @xmath81 , @xmath82 , @xmath83 , and @xmath84 ) when we show the structure of the protoplanet system ( figs . 1 , 4 , 5 , 7 , 8 , 12 , and 13 ) . on the other hand , to compare physical quantities derived from numerical results with those of the present jupiter , we use dimensional quantities at jovian orbit when we show the time evolution or radial distribution ( figs . 2 , 3 , 6 , 9 , 10 , and 11 ) of the mass @xmath85 and ( specific ) angular momentum @xmath86 ( @xmath87 ) of the protoplanet system , in which the conversion coefficient from physical quantities at jovian orbit ( 5.2au ) into those at any orbit ( @xmath33 ) is described in each axis . in addition , for convenience , we redefine the time unit as @xmath88 that corresponds to the orbital period of the protoplanet . when @xmath75au is assumed , @xmath89 corresponds to @xmath90yr . to estimate the angular momentum acquired by a protoplanet system from the protoplanetary disk , we need to cover a large dynamic range of spatial scale . using the nested - grid method ( for details , see * ? ? ? * ; * ? ? ? * ) , we cover the region near the protoplanet by the grids with high spatial resolution , and region remote from the protoplanet by the grids with coarse spatial resolution . each level of rectangular grid has the same number of cells ( @xmath91 ) , but cell width @xmath92 depends on the grid level @xmath93 . the cell width is reduced 1/2 with increasing the grid level ( @xmath94 ) . in a fiducial model , we use 8 grid levels ( @xmath95 ) . the box size of the coarsest grid @xmath96 is chosen to @xmath97 , and that of the finest grid @xmath98 is @xmath99 . the cell width in the coarsest grid @xmath96 is @xmath100 , and it decreases with @xmath101 as the grid level @xmath93 increases . thus , the finest grid has @xmath102 . we assume the fixed boundary condition in the @xmath103- and @xmath104-direction and the periodic boundary condition in the @xmath105-direction . in real units , using the standard solar nebular model , the scale height at jovian orbit ( @xmath1065.2au ) is @xmath107au . the computational domain in azimuthal direction corresponds to @xmath108au = @xmath109au , which is equivalent to the circumference of jovian orbit around the sun @xmath110 ( @xmath111au ) . although we imposed a periodic boundary condition in azimuthal direction , this domain size is valid except for ignoring the curvature . note that the computational domain is not necessarily the same as a real circumference of a planet as long as it is sufficiently large . to verify our results , in some models , we calculate the evolution of the protoplanet system adopting different levels of the finest grid ( or different maximum grid levels ) , @xmath112 = 5 , 6 , 7 , 9 , and 10 . in these models , the cell width of @xmath113 is @xmath114 ( @xmath115 at @xmath75au ) that corresponds to 33 times jovian radius , while that of @xmath116 is @xmath117 ( @xmath118 at @xmath75au ) that corresponds to 1.02 time jovian radius . the maximum grid level @xmath112 , and the cell width of the finest grid for each model are summarized in table [ table : table1 ] . at first , we show the evolution of the protoplanet system calculated with @xmath95 . the cell width of the maximum grid level ( @xmath98 ) that covers the region in the proximity of the protoplanet is @xmath119 ( @xmath120 cm at @xmath75au ) . figure [ fig:1 ] shows the evolution for model mn04 ( fiducial model ) , in which the protoplanet with @xmath121 is adopted for @xmath75au . the upper panels in figure [ fig:1 ] ( [ fig:1]_a _ _ e _ ) show the time sequence of the region far from the protoplanet ( @xmath122 ) with @xmath123 1 , 2 , and 3 grid levels , in which three different grid levels are superimposed , while the lower panels ( fig . [ fig:1]_f__j _ ) show the region near the protoplanet ( @xmath124 ) with @xmath123 3 , 4 , and 5 . each lower panel is 4 times magnification of each upper panel . in these panels , the protoplanet is located at the origin ( @xmath82 , @xmath83 , @xmath84 ) = ( 0 , 0 , 0 ) . the elapsed time @xmath125 is denoted in each upper panel . the central density @xmath126 is also denoted in each upper panel . figures [ fig:1]_a__c _ and [ fig:1]_f__h _ show that the density is enhanced in the narrow band with the spiral pattern that is distributed from the upper - left to the lower - right region . the density gaps that appears on the right ( left ) side of the spiral pattern in the region of @xmath127 ( @xmath128 ) are also seen in these panels . in figure [ fig:1]_c _ , the density of the spiral pattern around the protoplanet is @xmath129 , while that of the gap is @xmath130 . thus , there are a density contrast of @xmath131 between the spiral patterns and gaps . in addition , the central density increases up to @xmath132 for @xmath133 . figures [ fig:1]_h__j _ show a round shape near the protoplanet . since the hill radius is @xmath134 in this model , the gravity of the protoplanet is predominant in the region of @xmath135 . thus , the central region of @xmath136 has the round structure . figure [ fig:1 ] shows that the density distributions in panels _ c__e _ ( or _ h__j _ ) are similar . except for the calculation with @xmath116 , in all models , the structure around the protoplanet hardly changes for @xmath133 , which seems that the steady state is already achieved for @xmath133 . @xcite investigated the evolution of the protoplanet system as the same condition as ours but in two dimensions , and showed that the gas stream around the protoplanet is in a steady state after a short timescale of @xmath137 . in [ sec : test ] , we showed the evolution of the protoplanet system with the maximum grid level @xmath95 . in this subsection , to check the convergence of our calculation , we compare the evolutions of the protoplanet system with different maximum grid levels ( or different cell width of the finest grid ) . since our purpose is to investigate the angular momentum acquired by the protoplanet system , we use the average specific angular momentum as a measure of the convergence . as a function of the distance from the protoplanet @xmath138 , we define the average specific angular momentum @xmath139 as @xmath140 where the mass @xmath141 and angular momentum @xmath142 are integrated from the center @xmath143 to distance @xmath138 . here , we adopt @xmath144 ( i.e. , @xmath145 ) . the dependence of @xmath139 on @xmath138 is discussed in [ sec : dis - evo ] . we often show the mass and ( specific ) angular momentum of the protoplanet system in real units at @xmath75au to compare numerical results with present values of gaseous planets . the dimensional values @xmath146 , @xmath147 , and @xmath148 at @xmath75au can be converted into @xmath149 , @xmath150 , and @xmath151 at any orbit as @xmath152 @xmath153 and @xmath154 in the following , we show the values at 5.2au . when we refer to dimensional physical quantities without any mention of orbit , they are the values at @xmath79 au . the evolution of @xmath155 for models m04l5m04l9 ( see , table [ table : table1 ] ) are shown in figure [ fig:2 ] . in these models , the protoplanet mass is fixed , and only the maximum grid level ( or cell width of the finest grid ) is changed . figure [ fig:2 ] shows that the average specific angular momentum rapidly increases initially , then it saturates at a certain value in each model . although we adopted the same mass of the protoplanet , the saturation levels of the average specific angular momenta are different . the average specific angular momentum saturates at @xmath156 for model m04l5 , while it is saturated at @xmath157 for models m04l7 , m04 , m04l9 , and m04l10 . thus , there are a little differences for models with @xmath158 . for example , model m04l7 has @xmath159 , while model m04l9 has @xmath160 at @xmath161 . therefore , the average specific angular momenta are sufficiently converged in @xmath162 or @xmath163 within a relative error of about 15% . in the following , we safely calculate the evolution of the protoplanet system with @xmath95 of the maximum grid level . when we adopt the maximum grid level @xmath95 , the cell width is @xmath164 . in real units , when the protoplanet is located at 5.2au , the cell width corresponds to @xmath165 cm whose size is 4.1 times the jovian radius . interior to the gas giant planet , there exists a solid core with @xmath6 in mass and @xmath166 cm in size @xcite . the size of the solid core is much smaller than the cell width adopted in our calculation . to investigate the evolution of a protoplanet through gas accretion , in principle , we need to resolve a solid core with sufficiently small cell size ( @xmath167 cm ) however , since our purpose is to investigate the angular momentum flowing into the protoplanet system , we do not always need to resolve a central solid core and a protoplanet . as discussed in @xcite , the gas around the central region makes an artificial pressure gradient force , which may affect the gas accretion onto the protoplanet system . to check this , we also calculated the evolution of the protoplanet system adopting the sink cell in some models . we parameterized the size of the sink : @xmath168 and @xmath169 ( models m04s01 and m04s03 ) . these models are also summarized in table [ table : table1 ] . in real units , the radius of the sink in model m04s01 is @xmath170 ( @xmath171 jovian radius ) , while that in model m04s03 is @xmath172 ( @xmath173 jovian radius ) . during the calculation , we remove the gas from the region inside the sink radius in each time step , and integrate the removed mass and angular momentum that are assumed as the mass and angular momentum of the protoplanet system . after the steady states are achieved at @xmath174 , we estimate the average specific angular momenta @xmath139 as a function of @xmath138 . figure [ fig:3 ] shows the distribution of the average specific angular momentum @xmath175 against the distance from the protoplanet @xmath35 in real units for each model . the solid line represents @xmath175 without a sink cell , while the other lines represent those with sink cells . inside the sink radius , the average specific angular momentum is assumed as a constant value . in figure [ fig:3 ] , the vertical dotted line indicates the hill radius . in figure [ fig:3 ] the solid line shows that @xmath175 increases from the center to a peak around the hill radius , and drops sharply just outside the hill radius . the drop indicates that the angular momentum becomes negative at @xmath176 . thus , the rotational direction turns around between the region inside and outside the hill radius . as shown in @xcite , @xcite , and @xcite , the protoplanet formed by the gas accretion in the protoplanetary disk has a prograde spin , and thus it has a positive ( specific ) angular momentum . on the other hand , gas far outside of hill sphere seems to rotate retrogradely against the protoplanet because it rotates with nearly keplerian velocity with respect to the central star [ @xmath177 , as shown in eq . ( [ eq : shear ] ) ] . as a result , gas inside the hill radius has a positive angular momentum , while that outside the hill radius has a negative angular momentum . figure [ fig:3 ] shows that @xmath175 in models with the sink cell depends on the radius of the sink in the region @xmath178 . however , they do not depend sensitively on the size of the sink in the region of @xmath179 . for example , at @xmath180 , model without the sink ( model m04 ) has @xmath181 , while model m04s03 has @xmath182 . thus , difference of the average specific angular momentum among these models at the hill radius is @xmath183% . this difference decreases with the sink radius . in this study , we focus on the angular momentum of the protoplanet system , not the planet itself . as shown in figure [ fig:3 ] , the angular momentum acquired by the protoplanet system ( planet + circumplanetary disk ) extends up to the hill radius , in which almost all the angular momentum distributed in the region of @xmath184 or @xmath185 . in the following , we calculate the evolution of the protoplanet system without the sink cell . figure [ fig:4 ] shows the gas structure around the hill sphere for model m04 at @xmath186 . figure [ fig:4 ] left panel shows the structure on the cross section in the @xmath187 plane , in which red lines indicate the streamlines . in this panel , the gas enters @xmath188 grid from upper @xmath83 boundary for @xmath189 ( from lower @xmath83 boundary for @xmath190 ) and goes downward ( upward for @xmath190 ) according to the keplerian shear motion . the shocks ( crowded contours near the hill radius ) are seen in the upper right and lower left region from the protoplanet . in this model , the shock front almost corresponds to the hill radius ( fig . [ fig:4 ] left panel ) . when the gas approaches the protoplanet , the streamlines are bent by the gravity of the protoplanet . according to @xcite , the gas flow is divided into three region : the pass - by region ( @xmath191 ) , the horseshoe region ( @xmath192 , and @xmath193 ) , and the planet atmosphere region ( @xmath194 ) . note that although @xcite classified the flow pattern in their two dimensional calculation , their classification is useful for global flow pattern in three dimensions . in the pass - by region , the flow is first attracted toward the protoplanet , and then causes a shock after passing by the protoplanet . at the shock front , the density reaches a local peak and the streamlines bend suddenly . on the other hand , the gas entering the horseshoe region turns round by the coriolis force and goes back . the outermost streamlines in the horseshoe region ( i.e. , the streamlines passing very close to the protoplanet ) pass through the shock front , while the gas on the streamlines far from the protoplanet does not experience the shock . in the atmospheric region , the gas is bound by the protoplanet and forms a circumplanetary disk that revolves circularly around the protoplanet in the prograde ( counterclockwise ) direction . although the streamlines on @xmath187 plane ( fig . [ fig:4 ] left panel ) are similar to those in recent two dimensional calculations ( e.g. , * ? ? ? * ; * ? ? ? * ) , there are important differences . in two - dimensional calculations , a part of the gas near the hill sphere can accrete onto the protoplanet . @xcite showed that gas only in a narrow band that distributed from the lower left to the upper right region against the protoplanet for @xmath128 spirals inward toward the protoplanet passing through the shocked region and finally accretes onto the protoplanet ( for details , see figs . 4 , and 8 of * ? ? ? * ) . on the other hand , in our three - dimensional calculation , gas only flows out from the hill sphere and thus does not accrete onto the protoplanet on the midplane . figure [ fig:4 ] left panel shows that although gas flows into the hill sphere , a part of the gas flows out from the central region . figure [ fig:4 ] right panel is three - dimensional view at the same epoch as the left panel . in this panel , only the streamlines flowing into the high - density region of @xmath195 are drawn for @xmath196 which are inversely integrated from the high - density region . this panel clearly shows the gas flowing into the protoplanet in the vertical direction . to investigate gas flowing into the protoplanet system in detail , in figure [ fig:5 ] , we plot the streamlines at the same epoch as figure [ fig:4 ] with different grid levels ( @xmath93=3 , 5 , and 7 ) . in this figure , each upper panel shows three - dimensional view , while each lower panel shows the structure on the cross section in the @xmath197 plane . note that , in lower panels , the streamlines are projected onto the @xmath197 plane . figure [ fig:5]_a _ shows only the streamlines in a narrow bundle flowing into the protoplanet system . this feature is similar to that shown in two - dimensional calculations @xcite . however , the streamlines in figure [ fig:5]_a _ indicate that gas rises upward near the shock front and then falls into the central region in the vertical direction . gas flowing in the vertical direction spirals into the inner region ( fig . [ fig:5]_c _ ) . in this process , vortices appear as shown in figure [ fig:5]_d_. as shown in figure [ fig:4 ] left panel , also in figure [ fig:5]_d _ , gas is flowing out from the central region on @xmath187 plane . gas in the proximity of the protoplanet rotates circularly in the prograde direction as shown in figure [ fig:5]_e_. figure [ fig:5]_f _ shows that a part of the gas flowing into the upper boundary of @xmath198 grid level contributes to the disk formation around the protoplanet , while a remainder is bent and flows out from the central region . when we look down the protoplanetary disk from the above along the @xmath84-axis , streamlines may seem to be almost the same as those in two - dimensional calculations . however , gas moves also in the vertical direction : streamlines go upward near the shock front ( @xmath199 ) , and vertically falls into the central region at @xmath195 . this feature of streamlines is also seen in models m04s01 and m04s03 , in which the sink cell is adopted . thus , different features of streamlines in two- and three - dimensional calculations are not caused by the effect of the pressure gradient force , but caused by the dimensions ( because the same feature appears in both models with and without the sink ) . this flow pattern is also seen in other three - dimensional calculation @xcite . in two - dimensional calculation , since the vertical motion is restricted , the flow pattern is different from that in three dimensions . this difference affects the accretion rate onto the protoplanet and migration rate . @xcite showed that the migration rate is different between two- and three - dimensional calculations . in the present study , however , since we focus on the angular momentum of a protoplanet system , we do not discuss them any more . we will discuss the accretion and migration rate in the subsequent papers . finally , we comment on the circumplanetary disk . in figure [ fig:5 ] lower panels , the green surface ( i.e. , iso - density surface ) indicates the high - density structure around the protoplanet . these panels show the disk - like structure in the proximity of the protoplanet , and the disk becomes thinner as it approaches the protoplanet ( i.e. , the origin ) . we discuss the circumplanetary disk in [ sec : dis - disk ] . we have shown the evolution of the protoplanet system for model with 0.4@xmath2 in [ sec : typical ] . in this subsection , we investigate the evolution of the protoplanet system with different protoplanet masses . figure [ fig:6 ] upper panels show the accumulated masses ( eq . [ eq : mr ] ) within @xmath200 ( @xmath201 , fig . [ fig:6]_a _ ) and @xmath202 ( @xmath203 , fig . [ fig:6]_b _ ) against the elapsed time for different models , while figure [ fig:6 ] middle panels show the corresponding angular momenta ( eq . [ eq : jr ] ) in the same regions ( fig . [ fig:6]_c _ for @xmath204 , and fig . [ fig:6]_d _ for @xmath205 ) . in figure [ fig:6 ] , both masses and angular momenta for all models increase rapidly for @xmath206 . this is because the protoplanet with mass of @xmath207 suddenly appears in the protoplanetary disk at @xmath208 . however , this rapid growth phase ( @xmath206 ) is not real , because the gas is considered to begin to accrete onto the protoplanet when the mass of the solid core exceeds @xmath6 @xcite . the growth rates of the mass and angular momentum begin to decrease at @xmath209 in all models , then both masses and angular momenta increase with an almost constant rate until the end of the calculation ( @xmath210 ) . @xcite calculated the mass accretion rate onto the protoplanet as @xmath211 thus , the growth time @xmath212 is @xmath213yr ( i.e. , @xmath214 ) . in our calculation , we continue to calculate the evolution of the protoplanet system for @xmath215yr ( @xmath216 ) by fixing the planet mass . we think that this treatment is not problematic , since the growth timescale is longer than our calculation time . note that since @xcite calculated the evolution of the protoplanet system in two dimensions , the growth rate might be different from that in three - dimensional calculations . figure [ fig:6 ] lower panels show the evolution of the average specific angular momentum in the region of @xmath200 ( fig . [ fig:6]_e _ for @xmath217 ) and @xmath202 ( fig . [ fig:6]_f _ for @xmath218 ) . the masses and angular momenta flowing into the protoplanet system increase with a constant rate for @xmath206 , while the average specific angular momenta are saturated at certain values for @xmath219 . this saturation means that flow around the protoplanet is in the steady state . figures [ fig:6]_e _ and _ f _ also indicate that the average specific angular momentum brought into the protoplanet system increases with the mass of the protoplanet . in figures [ fig:6]_e _ and _ f _ , the average specific angular momenta in the region of @xmath200 is larger than those in the region of @xmath202 indicating that the protoplanet system has a larger average specific angular momentum in the more distant place from the protoplanet . we will investigate the angular momentum acquired in the protoplanet system in [ sec : dis - evo ] . figures [ fig:7 ] and [ fig:8 ] show the density distributions ( upper panels ) and jacobi energy contours ( lower panels ) on the cross section in the @xmath187 ( fig . [ fig:7 ] ) and @xmath197 plane ( fig . [ fig:8 ] ) around the hill sphere after the steady state is achieved ( @xmath220 ) for models m02 ( left panels ) , m04 ( middle panels ) , and m06 ( right panels ) . the white - dotted line in each panel indicates the hill radius @xmath71 . in each upper panel , the shock appears from the upper left to lower right near the hill radius . these shocks are frequently seen in similar calculations @xcite . in figure [ fig:7 ] upper panels , the round structures are seen in the proximity to the protoplanet ( @xmath195 ) , while the ellipsoidal structures are seen in the regions of @xmath221 . this is because gas distributed near the protoplanet is more strongly bound by the protoplanet . figure [ fig:8 ] upper panels show that contours of the central region sags in the center of a concave structure , and thin disks are formed around the protoplanet ( @xmath195 ) . in addition , the butterfly - like structure is also seen inside the hill radius in figure [ fig:8 ] upper panels . these structures are considered to be formed by the rapid rotation of the central circumplanetary disk : similar structure is seen in a rapidly rotating protostar ( e.g. , fig.1 of * ? ? ? we will discuss the disk structure in [ sec : dis - disk ] . in contrast to celestial mechanics , it is difficult to find fluid elements bound by the protoplanet because thermal energy is important in addition to the gravitational and kinetic energies . to discern gas bound by the gravity of the protoplanet , we use the jacobi energy as an indicator @xcite . in our unit , the jacobi energy is given by @xmath222 in equation ( [ eq : jacobi ] ) , the first term is the kinetic energy , the second and third terms are the tidal energy of the central star , and the forth term is the gravitational energy of the protoplanet . the jacobi energy is the conserved quantity in a rotating system . in the case of fluid , the jacobi energy is not strictly appropriate because the thermal energy is ignored . however , we can use this for rough estimation of gas bound by the protoplanet . we determine fluid elements bound by the protoplanet from the contour of the jacobi energy in the lower panels in figures [ fig:7 ] and [ fig:8 ] . fluid elements with lower jacobi energy are strongly bound by the protoplanet as described in equation ( [ eq : jacobi ] ) . lower panels of figures [ fig:7 ] and [ fig:8 ] show that inside the region of @xmath223 , each contour has a closed ellipse . since fluid elements move on this closed orbit , it is considered that gas distributed in the region of @xmath223 is bound by the protoplanet when the thermal effect can be ignored . we discuss the thermal effect in [ sec : dis - evo ] . on the other hand , although fluid elements exist inside the hill radius , outside @xmath224 , they are not bound by the protoplanet , because the contours of the jacobi energy straddle the hill radius . for example , in lower left panel of figure [ fig:7 ] , the contour of @xmath225 straddle the hill radius , and thus fluid elements with this jacobi energy freely move on this contour . thus , although these elements transiently stay inside the hill radius , they flow out from the hill sphere . in summary , figures [ fig:7 ] and [ fig:8 ] suggest that fluid elements inside @xmath226 are bound by the protoplanet . figure [ fig:9 ] upper panel shows the accumulated mass @xmath227 for different models after the steady state is achieved ( @xmath174 ) . in this panel , a thin solid line indicates the initial value ( or the value of the protoplanetary disk ) , and the circle is the hill radius @xmath228 for each model . the accumulated mass in any model is larger than the initial value , because gas flows into the hill sphere . this panel indicates that the massive protoplanet has a massive envelope . since these mass distributions are in a steady state at the fixed mass of the protoplanet , it can be considered that different curves correspond to snapshots at different evolution phases . namely , a gas envelope increases its mass with time ( or the protoplanet mass ) . outside the hill radius , the accumulated mass in each model converges to the initial value indicating that the mass distribution for @xmath229 does not change from the initial state because the influence of the protoplanet is small . figure [ fig:9 ] lower panel shows the absolute value of the angular momentum @xmath230 . there are spikes in all models , at which the sign of the angular momentum is reversed . as shown in [ sec : sink ] , the angular momentum has a positive sign around the protoplanet ( @xmath231 ) , while it becomes negative outside the hill radius ( @xmath229 ) . the sign of the angular momentum is reversed outside the hill radius for models m01 , m02 , m04 , and m06 , while it is reversed inside the hill radius for model m005 . this is because the protoplanet system for model m005 does not acquire a sufficient mass and angular momentum owing to the shallow gravitational potential and relatively large thermal pressure ( for details , see [ sec : dis - evo ] ) . figure [ fig:9 ] upper panel shows that the protoplanet system has the envelope mass of only @xmath232 for model m005 , in which the protoplanet mass is @xmath233 . except for model m005 , the angular momenta gradually decrease after they reach their peak around @xmath234 , then it becomes negative at @xmath235 . thus , the angular momenta bound by the protoplanet system are limited in the region of @xmath236 at the maximum . figure [ fig:9 ] lower panel shows that the angular momentum keeps an almost constant value around the hill radius ( @xmath237 ) , until they are reversed . this means that the angular momentum with plus sign and that with minus sign are mixed in this region , as shown in figure [ fig:4 ] left panel . as a result , wherever we estimate the angular momentum in the range of @xmath238 , we can obtain almost the same values of the angular momentum . the distributions of the average specific angular momentum @xmath175 for different models are shown in figure [ fig:10 ] . the circles in this figure mean the hill radii @xmath228 . the crosses indicate the jacobi radii @xmath239 inside which gas is considered to be bound by the gravity of the protoplanet . we determine the jacobi radii from the contour of the jacobi energies as in figures [ fig:7 ] and [ fig:8 ] lower panels . the jacobi radii in all models are distributed in the range of @xmath240 . figure [ fig:10 ] indicates that a more massive protoplanet has an envelope with larger amount of the specific angular momentum . thus , the specific angular momentum accreting onto the protoplanet system increases as the protoplanet mass increases . the rapid drops at large radii indicate the reverse of the rotation axis as shown in figures [ fig:3 ] and [ fig:9 ] . in figure [ fig:10 ] , in model m01 ( @xmath241 ) , @xmath175 at @xmath180 is twice of that at @xmath242 , while , in models m02 , m04 , m06 , m1 , and m3 ( @xmath243 ) , there are little differences between the average specific angular momentum derived from the hill radius @xmath244 and that derived from jacobi radius @xmath245 . thus , we can safely estimate the average specific angular momentum using either the hill radius or jacobi radius for models with @xmath246 . to properly calculate the angular momentum of the protoplanet system , we have to calculate the planetary growth from the solid core with @xmath6 to the present mass . however , it takes huge computation time to calculate all evolution phases . thus , we estimate the angular momentum of the protoplanet system according to the following procedure : ( i ) we calculate the average specific angular momenta of the gas flowing into the protoplanet system at the fixed masses of the protoplanet ( i.e. , under the same parameter @xmath71 ) in models changing the mass of the protoplanet , then ( ii ) derive the relation between the average specific angular momentum and mass of the protoplanet , and describe it as a function of the protoplanet mass , and ( iii ) estimate the angular momentum of the protoplanet system integrating the average specific angular momentum by mass up to the present value of gas giant planets . at first , we analytically estimate the specific angular momentum of the protoplanet system , then compare it with numerical results . we assume that the gas that overcomes the hill potential flows into the hill sphere ( or a protoplanet system ) with the kepler velocity of the protoplanet at the hill radius @xmath228 . when the protoplanet mass is @xmath0 , the kepler velocity at @xmath228 is given by @xmath247 note that @xmath248 is constant when @xmath33 is fixed . the specific angular momentum can be written as @xmath249 thus , the specific angular momentum is proportional to @xmath250 . in addition , when the mass of the central star is fixed , the specific angular momentum is proportional to @xmath251 thus , the specific angular momentum increases with 2/3 power of the protoplanet mass . figure [ fig:11 ] shows the average specific angular momentum derived from all models against the protoplanet mass . the average specific angular momenta are estimated in the region of @xmath252 ( @xmath253 ) , @xmath254 ( @xmath255 ) , @xmath256 ( @xmath257 ) , and @xmath258 ( @xmath259 ) . although we fixed the protoplanet mass in each model , we can consider the horizontal axis in figure [ fig:11 ] as time sequence of the protoplanet system . figure [ fig:11 ] clearly shows that more massive protoplanet can acquire the envelope with larger average specific angular momentum . although the average specific angular momenta for @xmath260 differ from those for @xmath261 for models with @xmath262 , there are little difference for models with @xmath263 . in addition , when we adopt the average specific angular momenta in the region of @xmath264 , we underestimate them for any model as shown in figure [ fig:11 ] . therefore , we can properly estimate the average specific angular momentum of the protoplanet system in any region of @xmath265 for models with @xmath266 . in figure [ fig:11 ] , the red and blue lines are fitting formulae of the evolution of the average specific angular momentum as a function of the protoplanet mass for @xmath267 ( low mass , @xmath268 ; red line ) , and @xmath269 ( high mass , @xmath270 : blue line ) . they are given by @xmath271 and @xmath272 figure [ fig:11 ] shows that the evolution of the average specific angular momentum for @xmath273 is well described by @xmath274 which corresponds to the equation ( [ eq : ana - j ] ) . on the other hand , the growth rate of the average specific angular momentum for @xmath275 is larger than @xmath274 , and can be fitted by @xmath3 . we ignored the thermal effect when we derive equation ( [ eq : ana - j ] ) . when @xmath0 is small , the gas flowing into the protoplanet system is affected relatively strongly by the thermal pressure . we can estimate the mass at which the gravity dominates the thermal pressure force from the balance between the thermal pressure gradient and gravitational forces . near the hill radius , the thermal pressure gradient force is more dominant than the gravity when the kepler speed is slower than the sound speed ( i.e. , @xmath276 ) . on the other hand , when @xmath277 , the gas flow is controlled mainly by the gravity of the protoplanet even near the hill radius . using equation ( [ eq : kepler ] ) , @xmath278 is realized when @xmath279 which corresponds to @xmath280 of the protoplanet mass at @xmath79au . thus , the gas flow is largely affected by the thermal pressure for @xmath281 , while it is not so affected by the thermal pressure for @xmath282 . in figure [ fig:11 ] , the evolution of the angular momentum for @xmath283 well corresponds to the analytical solution ( @xmath4 ) indicating that the thermal effect is negligible for @xmath284 . however , for @xmath285 , the average specific angular momentum is smaller than @xmath270 ( blue line ) . thus , when the protoplanet is @xmath275 , the thermal effect is not negligible for the acquisition process of the angular momentum because the gravitational potential is relatively shallow . thus , the thermal pressure seems to remain important for @xmath286 , although @xmath278 realized at @xmath287 . to verify the relation of the thermal and gravitational effects , in figure [ fig:12 ] , we plot the ratio of the azimuthal to kepler velocity ( @xmath288 ) around the protoplanet on the midplane for models m02 , m04 , and m06 . in this figure , closed contours inside the hill radius indicate that gas revolves around the protoplanet . for example , the closed contour of @xmath289 in figure [ fig:12 ] means that gas rotates along the contour with 50% of the kepler velocity . the black circles represent contours of @xmath290 , inside which gas revolves around the protoplanet with supersonic velocity . thus , the gravitational force of the protoplanet is dominant inside the black circles , while the thermal pressure gradient force is dominant outside the black circles . figure [ fig:12 ] shows that the radius of the @xmath290 contour increases with the protoplanet mass indicating that the region that dominated by gravity of the protoplanet extends with the protoplanet mass . as the protoplanet mass increases , the flow speed inside the hill radius approaches to the kepler velocity , and the region dominated by the gravity of the protoplanet spreads outward . in this way , since the thermal effect decreases as the protoplanet mass increases , the growth rate of the average specific angular momentum approaches to the analytical solution ( eq . [ eq : ana - j ] ) . in this paper , we investigated the evolution of the protoplanet system under the isothermal approximation , which might not be problematic in the circumplanetary disk , but may not be valid in the very proximity to the protoplanet . to investigate the thermal effect around the protoplanet , we adopted the sink cell in some models . as shown in [ sec : sink ] , the cell width in the fiducial model having @xmath95 is about 4 jovian radius . thus , gas inside @xmath291 has the same thermal energy . if the protoplanet has almost the same size as the present gas giant planet , the thermal energy around the protoplanet may be overestimated in model without the sink cell , because an actual protoplanet is embedded in the small part of the innermost cell . on the other hand , when we adopt the sink cell , the thermal energy around the protoplanet is underestimated , because the thermal energy is artificially removed from the sink cell . however , as shown in [ sec : sink ] , when the sink radius is much smaller than the hill radius ( @xmath292 ) , there are little differences in the angular momentum acquired by the protoplanet system between models with and without the sink cell . this is because a large part of the angular momentum of the protoplanet system is distributed around the hill radius ( @xmath293 ) as shown in [ sec : res - ang ] . the jovian radius is much smaller than the hill radius ( @xmath294 ) . in addition , the angular momentum flowing into the protoplanet system is determined by the shearing motion in the region of @xmath234 ( see [ sec : dis - evo ] ) . therefore , under the assumption that the protoplanet is smaller than the innermost cell width , it is expected that the thermal effect from the protoplanet is sufficiently small for the acquisition process of the angular momentum when we are using sufficient smaller cells than the hill radius . @xcite , @xcite , and @xcite suggested that gaseous protoplanet has a large envelope with high temperature . for example , @xcite showed that , for jovian case , gas distributed in the range of @xmath295 cm behaves isothermally , while that distributed in the range of @xmath296 cm behaves adiabatically in their spherically symmetric calculations ( see fig.4 of @xcite ) . thus , gas in the range of @xmath296 cm ( @xmath297 ) has higher temperature than the ambient medium . however , it is expected that the gaseous envelope hardly affect the angular momentum of the protoplanet system , because the size of the envelope ( @xmath297 ) is much smaller than the hill radius ( @xmath294 ) . a large part of the angular momentum is distributed in the range of @xmath298 , as shown in [ sec : res - ang ] . on the other hand , the size of the circumplanetary disk may be affected by the thermal envelope . thus , when we investigate the formation of the circumplanetary disk , we have to include the realistic thermal evolution around the protoplanet . however , it is difficult to study the thermal evolution around the protoplanet , because we need to solve the radiation hydrodynamics in three dimensions . in a subsequent paper , we will investigate the effect of the thermal envelope under simple assumptions . in [ sec : dis - evo ] , we fixed the physical quantities as those at @xmath75au . however , in our calculation , since we use the dimensionless quantities , we can rescale those at any orbits @xmath33 . under the standard model @xcite , we can generalize equation ( [ eq : jlm ] ) as @xmath299 we assume that the protoplanet system acquires the gas with the average specific angular momentum of equation ( [ eq : fit3 ] ) when the protoplanet has a mass of @xmath275 . thus , to estimate the angular momentum of the protoplanet system , we need to integrate equation ( [ eq : fit3 ] ) by the mass until the present values of gas giant planets . for example , the angular momentum in our model for jupiter is @xmath300 which is about 30 times larger than that of present jupiter ( @xmath301g@xmath302 ) . on the other hand , the angular momentum in our model for saturn is @xmath303 which is 50 times larger than that of present saturn ( @xmath304g@xmath302 ) , where @xmath305 is the saturnian mass . note that the orbital angular momenta of the _ present _ jovian and saturnian satellites can be ignored in the protoplanet system because they are considerably smaller than the jovian and saturnian spin angular momenta . the above estimation corresponds to the angular momentum of proto planet - disk system , which includes the central planet and protoplanetary disk . we showed that the angular momentum of the protoplanet system is @xmath306 times larger than present spin of the gaseous planet . the angular momenta flowing into the hill sphere are distributed into the spin of the protoplanet and orbital motion of the circumplanetary disk . it is expected that a large fraction of the total angular momentum contributes to the formation of the circumplanetary disk and the residual contributes to the spin of the planet . although the fraction of the angular momentum of the circumplanetary disk to the spin of protoplanet is not correctly estimated in our calculation , a part of the angular momentum is certainly distributed into the disk . thus , the angular momentum transfer and dissipation mechanism for the circumplanetary disk are necessary to follow further evolution of the protoplanet system . @xcite proposed the despin mechanism of the protoplanet by planetary dipole magnetic field , in which an initially rapidly rotating protoplanet can be spun down to the present value by the magnetic interaction between the protoplanet and circumplanetary disk . when the circumplanetary disk is formed around the protoplanet , it is possible to form satellites in the disk . while there are many scenarios for the satellite formation ( e.g. , * ) , regular satellites around gas giant planets are supposed to be formed in the gaseous disk as for the planet formation in the protoplanetary disk ( e.g. , * ? ? ? * ; * ? ? ? observations showed that the regular satellites around jupiter and saturn are distributed only in the close vicinity of the planet ( @xmath307 ) , and are on prograde orbits near the equatorial plane . thus , it is expected that these regular satellites formed in the circumplanetary disk . we discussed the angular momentum of the protoplanet system in [ sec : dis - evo ] . as the angular momentum flowing into the hill sphere is brought into both the protoplanet and circumplanetary disk , we can not estimate the fraction of angular momentum for the circumplanetary disk . when we assume that the centrifugal force is balanced with the gravity of the protoplanet , we can derive the centrifugal radius @xmath308 as @xmath309 to quantify the disk size , we adopt the specific angular momentum in equation ( [ eq : cent ] ) as those in equation ( [ eq : fit3 ] ) . the centrifugal radius for the proto - jovian disk with @xmath310 and @xmath311au is @xmath312 which is 22 times as large as jovian radius ( @xmath313 ) . the galilean satellites , which are regular satellites , are distributed in the range of @xmath314 , which is consistent with the centrifugal radius derived from our calculation . note that the disk size is expected to be larger than equation ( [ eq : r - est ] ) owing to the thermal effect around the protoplanet . in the same way , we estimate the centrifugal radius of proto - saturnian disk as @xmath315 which is 68 times as large as saturnian radius ( @xmath316 cm ) . the saturnian representative regular satellites ( mimas , enceladus , tethys , dione , rhea , titan , hyperion , and iapetus ) are distributed in the range of @xmath317 , which is also corresponding to the centrifugal radius of our result . figure [ fig:13 ] shows the structure around the protoplanet at the end of the calculation ( @xmath174 ) for model m1 , in which the protoplanet has mass of @xmath318 . each right panel is 4 times magnification of each left panel . figures [ fig:13]_a _ , _ b _ , and _ c _ show the shock fronts outside the hill radius , and a thick disk between the shock front and the hill sphere . in the proximity to the protoplanet , a concave structure is seen in figures [ fig:13]_e _ and _ f_. the contours in these figures rapidly drop around the rotation axis ( i.e. , @xmath20-axis ) . thus , even inside the hill sphere , except for the central region , the disk has a thick torus - like structure . figure [ fig:13]_g _ , _ h _ , and _ i _ show the structure in the very proximity to the protoplanet . on the midplane , the elliptical structure is seen in a large scale ( fig . [ fig:13]_d _ ) , while an almost round shape is seen in the proximity to the protoplanet ( fig . [ fig:13]_g _ ) because the force acting on gas in this region is dominated by the gravity of planet . the contours in figures [ fig:13]_h _ and _ i _ show a very thin disk in the region of @xmath319 ( or @xmath320 ) . thus , the regular satellites might be formed in these circumplanetary disks . however , to study the satellite formation in more detail , we need more realistic calculations . to study the gas flow pattern and angular momentum accretion onto the protoplanet system , we have calculated the evolution of the protoplanet system in the circumstellar disk . firstly , we have investigated the dependence of the angular momentum accreting onto the protoplanet system on spatial resolution with different cell widths , and confirmed its convergence for the cell width much smaller than the hill radius . next , adopting the sink cell whose size is comparable to or slightly larger than the radius of the present gaseous planets in the solar system , we have checked that , in both models with and without the sink , the thermal effect around the protoplanet barely affects the gas flow pattern and angular momentum of the protoplanet system . thirdly , with sufficiently high spatial resolution , we have calculated the evolution of the protoplanet system with the protoplanet mass in the range of @xmath321 , where @xmath322 and @xmath2 are the protoplanet and jovian masses , respectively . the following results are obtained : * the gas flow pattern in three dimensions is qualitatively different from that in two dimensions : the gas is flowing onto the protoplanet system mainly in the vertical direction in three - dimensional simulations . * the specific angular momentum increases as @xmath3 when the protoplanet mass is @xmath1 , while it increases as @xmath4 when @xmath323 . * the angular momentum of the protoplanet system is @xmath306 times larger than the present spin of the gaseous planets in the solar system . * a thin disk is formed only in the region of @xmath324 , where @xmath10 is the radius of the planets . this location agree with the orbital radii of regular satellites around the present gaseous planets in the solar system . these conclusions imply that the protoplanet system can acquire sufficient angular momentum from the circumstellar disk , and despin mechanisms of the protoplanet system are necessary to realize the present spin of the gaseous planets in the solar system . we are planning higher - resolution simulations to further investigate the evolution of the angular momentum of the gas giant planets in near future . we have greatly benefited from discussion with h. tanaka , and t. muto . we also thank t. hanawa for contribution to the nested - grid code . numerical calculations were carried out with fujitsu vpp5000 at the center for computational astrophysics , the national astronomical observatory of japan . si is grateful for the hospitality of kitp and interactions with the participants of the program `` star formation through cosmic time '' . this work is supported by the grants - in - aid from mext ( 15740118 , 16077202 , 16740115 , 18740104 ) . plane are plotted . each lower panel ( @xmath325 ) is four times spatial magnification of each upper panel ( @xmath96 ) . three levels of grids are shown in each upper ( @xmath1231 , 2 , 3 ) and lower ( @xmath1233 , 4 , 5 ) panel . level of outermost grid is denoted in the upper left corner . the elapsed time @xmath125 and central density @xmath326 on the midplane are denoted in the top of each panel . the velocity scale in the unit of the sound speed is denoted in the bottom of each panel . , width=491 ] ( _ left _ ) , 5 ( _ middle _ ) , and 7 ( _ right _ ) . each upper panel is the structure in three dimensions shown in birds - eye view ( _ right _ ) , and each lower panel is the structure on the cross section in the @xmath197 plane . the gas streamlines ( _ red lines _ ) , density structure ( _ color _ ) , and velocity vectors ( _ arrows _ ) are plotted in each panel . the size of the domain is shown in each panel . , width=566 ] plane in each upper panel for models m02 ( _ left _ ) , m04 ( _ middle _ ) , and m06 ( _ right _ ) . the jacobi energy ( _ false color , and contours _ ) are plotted in each lower panel for the same models as in each upper panel . three levels of grid ( @xmath1234 , 5 , and 6 ) are superimposed in each panel . the dotted - circle represents the hill radius . , width=566 ] ( @xmath253 ) , @xmath328 ( @xmath259 ) , @xmath329 ( @xmath255 ) , and @xmath330 ( @xmath257 ) against the protoplanet mass . blue and red lines are the fitting formulae for @xmath331 ( blue ) and @xmath332 ( red ) , respectively . , width=566 ] ( _ color _ and _ contours _ ) around the protoplanet for models m02 ( left ) , m04 ( middle ) , and m06 ( right ) . three levels of grids are superimposed in each panel . the outer and inner white - dashed - circles represent the hill radius @xmath333 ( outer ) and the half of the hill radius @xmath334 ( inner ) , respectively . the black circle represents the contour of @xmath335 , inside which gas rotates with supersonic velocity @xmath336 . , width=566 ] ( upper panels ) , @xmath197 ( middle panels ) , and @xmath337 ( lower panels ) plane with different outermost grid levels @xmath325 ( left ) , 5 ( right ) and 7 ( right ) for model m1 . three grid levels are superimposed in each panel . right panels _ h _ , and _ i _ are eight times enlargement of left panels _ a _ , _ b _ , and _ c_. the dotted circles represent the hill radii @xmath333 . , width=566 ]

in the horndeski s most general scalar - tensor theories with second - order field equations , we derive the conditions for the avoidance of ghosts and laplacian instabilities associated with scalar , tensor , and vector perturbations in the presence of two perfect fluids on the flat friedmann - lematre - robertson - walker ( flrw ) background . our general results are useful for the construction of theoretically consistent models of dark energy . we apply our formulas to extended galileon models in which a tracker solution with an equation of state smaller than is present . we clarify the allowed parameter space in which the ghosts and laplacian instabilities are absent and we numerically confirm that such models are indeed cosmologically viable .

1110.3878

the problem of dark energy is one of the most serious problems faced by cosmologists and particle physicists . the cosmological constant is a simplest candidate , but it is plagued by the huge difference of energy scales between the theoretical and observed values @xcite . many alternative models to the cosmological constant have been proposed to approach the origin of dark energy @xcite . dark energy models can be broadly classified into two classes : ( i ) modified matter models , and ( ii ) modified gravity models . in the class ( i ) the modified matter source is introduced to drive the late - time cosmic acceleration . the representative model in this class is quintessence @xcite , in which the potential energy @xmath1 of a slow - rolling scalar field @xmath2 is the source for dark energy . another model is k - essence @xcite , in which a non - linear term with respect to the field kinetic energy @xmath3 leads to the cosmic acceleration . the representative models that belong to the class ( ii ) are those based on @xmath4 theories @xcite , @xmath5 theories @xcite , brans - dicke theories @xcite , dvali - gabadazde - porrati ( dgp ) braneworld @xcite , and galileon gravity @xcite ( see ref . @xcite for reviews ) . the lagrangian in @xmath4 theories is an arbitrary function of the ricci scalar @xmath6 , which corresponds to the particular class of brans - dicke theories ( brans - dicke parameter @xmath7 @xcite ) . the @xmath5 theories consist of a function of @xmath6 and the gauss - bonnet term @xmath8 , in which case there are two independent scalar degrees of freedom in a general background @xcite . in galileon gravity , the lagrangian is constructed such that the field equations are invariant under the galilean symmetry @xmath9 in the limit of minkowski spacetime @xcite . one of those terms is given by @xmath10 , which appears in the dgp model as a result of the mixing of a brane - bending mode with a transverse graviton @xcite . in the galileon model there exists a stable late - time de sitter solution where @xmath11constant @xcite . moreover the model can be consistent with local gravity constraints through the vainshtein mechanism @xcite . recently deffayet _ et al . _ @xcite derived the action for the most general scalar - tensor theories having second - order field equations . this action is equivalent to that derived by horndeski @xcite in 1974 in the context of lovelock gravity @xcite ( see also ref . in fact the horndeski s action is sufficiently general to accommodate most of the dark energy models proposed in literature . moreover , the gauss - bonnet couplings @xmath12 @xcite , the generalized galileon term @xmath13 @xcite , the derivative couplings @xmath14 with the einstein tensor @xmath15 @xcite also belong to the class of the horndeski s action @xcite . for such general theories the full background and perturbation equations were recently derived in ref . @xcite in the presence of non - relativistic matter . in the context of inflation several authors computed the power spectra of scalar and tensor perturbations @xcite as well as the non - linear parameter of primordial non - gaussianities @xcite . if we modify gravity from general relativity we need to worry about the presence of ghosts as well as laplacian instabilities . in the presence of two perfect fluids ( radiation and non - relativistic matter ) we shall derive conditions for the avoidance of ghosts and laplacian instabilities associated with scalar , tensor , and vector perturbations in the horndeski s theories . this is of particular importance for the construction of theoretically consistent modified gravitational models of dark energy . we apply our results to extended galileon models in which a tracker solution is present . this tracker corresponds to the generalization of that found in ref . @xcite for the covariant galileon . unlike the covariant galileon the equation of state of dark energy @xmath16 can take the values close to @xmath0 during the evolution from the matter era to the accelerated epoch . this property should allow for the tracker to be compatible with observations . this paper is organized as follows . in sec . [ backsec ] the background equations of motion are derived on the flat flrw spacetime in the presence of two perfect fluids . in sec . [ avoidancesec ] we find conditions for the absence of ghosts and laplacian instabilities by deriving the second - order action for scalar , tensor , and vector perturbations . in sec . [ applicationsec ] we apply our general formulas to kinetically driven dark energy models , which cover the covariant galileon as a specific case . not only we identify a theoretically consistent parameter space but also we numerically integrate the field equations to confirm the analytic estimation . [ consec ] is devoted to conclusions . in order to avoid the ostrogradski instability @xcite it is desirable to keep the equations of motion at second order in derivatives . the most general 4-dimensional scalar - tensor theories having second - order field equations are described by the lagrangian @xcite @xmath17 where @xmath18\,,\\ { \cal l}_{5 } & = & g_{5}(\phi , x)\ , g_{\mu\nu}\,(\nabla^{\mu}\nabla^{\nu}\phi)-\frac{1}{6}\ , g_{5,x}\,[(\box\phi)^{3}-3(\box\phi)\,(\nabla_{\mu}\nabla_{\nu}\phi)\,(\nabla^{\mu}\nabla^{\nu}\phi)+2(\nabla^{\mu}\nabla_{\alpha}\phi)\,(\nabla^{\alpha}\nabla_{\beta}\phi)\,(\nabla^{\beta}\nabla_{\mu}\phi)]\,.\label{eachlag5}\end{aligned}\ ] ] here @xmath19 and @xmath20 ( @xmath21 ) are functions with respect to a scalar field @xmath2 and its kinetic energy @xmath22 , @xmath6 is the ricci scalar , and @xmath23 is the einstein tensor . @xmath24 and @xmath25 ( @xmath21 ) correspond to the partial derivatives of @xmath20 with respect to @xmath26 and @xmath2 respectively , i.e. @xmath27 and @xmath28 . the above lagrangian was first discovered by horndeski in a different form @xcite , which is equivalent to the lagrangian ( [ lagsum ] ) @xcite . in addition to the scalar field @xmath2 we take into account two barotropic perfect fluids , whose energy densities are @xmath29 and @xmath30 . then the total action is given by @xmath31 where @xmath32 is the determinant of the metric @xmath33 , @xmath34 and @xmath35 are the lagrangians of two perfect fluids respectively . let us consider a flat flrw background with the metric @xmath36 , where @xmath37 is the cosmic time , @xmath38 is the lapse function , and @xmath39 is the scale factor . varying the action ( [ action ] ) with respect to the @xmath38 and @xmath39 respectively , we obtain @xmath40g_{5,\phi}+4hx\dot{\phi}\ , g_{5,\phi\phi}=-p_{a}-p_{b}\,,\label{be2}\end{aligned}\ ] ] where a dot represents a derivative with respect to @xmath37 , @xmath41 is the hubble parameter , and @xmath42 are the pressures of the two perfect fluids . varying the action ( [ action ] ) with respect to @xmath43 , it follows that @xmath44 where @xmath45 the two perfect fluids obey the following continuity equations @xmath46 equations ( [ be1 ] ) , ( [ be2 ] ) , ( [ fieldeq ] ) , ( [ rhoreq ] ) , and ( [ rhomeq ] ) are not independent because of the bianchi identities . the field equation ( [ fieldeq ] ) can be derived by using the other equations . in this section we study the stability of the flat flrw background for the action ( [ action ] ) in the presence of two perfect fluids . let us first consider scalar metric perturbations @xmath47 , @xmath48 , and @xmath49 with the line element @xcite @xmath50 here we have gauged away a scalar perturbation @xmath51 that appears in the form @xmath52 in front of the term @xmath53 . this fixes the spatial part of the gauge - transformation vector @xmath54 . in the following we choose the uniform - field gauge for which the field @xmath2 has no perturbations , that is , @xmath55 . this fixes the temporal part of the vector @xmath54 . the two perfect fluids can be described in terms of the following lagrangian @xmath56\,,\ ] ] where @xmath57 and @xmath58 ( @xmath59 ) correspond to the chemical potential and the entropy per particle respectively . we will employ the method introduced in ref . @xcite to study the perfect fluid from a lagrangian point of view in order to extract the conditions for the absence of ghost and laplace instabilities . in the following we will summarize and simplify the method given in ref . @xcite ( see also ref . @xcite ) . since we are interested in those fluids with equations of state of the kind @xmath60 ( @xmath59 ) , we consider @xmath61 as functions of @xmath57 alone . here the chemical potential of each fluid is defined as @xmath62 , where @xmath63 is the number density of the fluid @xmath64 . in fact , it is sufficient to give the equations of state @xmath65 . we define the fluid 4-velocity @xmath66 associated with the chemical potential @xmath67 , as @xmath68 , where @xmath69 is a scalar field . the normalization condition for the 4-velocity allows us to write the 4-velocity of the fluid @xmath70 , as @xmath71 . after we perturb the field @xmath72 as @xmath73 , we can expand the matter lagrangian at second order and then perform the field redefinition @xmath74 , where @xmath75 is chosen to represent the independent scalar degree of freedom of the fluid @xmath70 . along the same lines , the independent scalar degree of freedom for the fluid @xmath76 corresponds to @xmath77 . since , at linear order , the scalar fluids do not contribute to the vector perturbations , it is sufficient to study their scalar contributions alone in order to derive the conditions for the avoidance of ghosts and laplacian instabilities . we will discuss this issue in appendix a. we perturb the action @xmath78 up to the second order . after integrations by parts , the second - order action is given by @xmath79\,,\label{saction}\end{aligned}\ ] ] where @xmath80 , and @xmath81 the equation of motion for @xmath48 gives rise to the following momentum constraint @xmath82 which can be used to integrate out the field @xmath47 . after replacing @xmath47 in eq . ( [ saction ] ) , by using eq . ( [ eq : hamconstr ] ) , we need to integrate the terms @xmath83 and @xmath84 by parts , so that the action ( [ saction ] ) reduces to @xmath85,\ ] ] where @xmath86 , and @xmath87 , @xmath88 , @xmath89 , @xmath90 are the components of the @xmath91 matrices @xmath92 , @xmath93 , @xmath94 , @xmath95 . imposing the matrix @xmath92 to be positive definite leads to the no - ghost conditions @xmath96 and @xmath97 , @xmath98 . the latter two conditions are automatically satisfied for radiation ( @xmath99 ) and non - relativistic matter ( @xmath100 ) . the speed of propagation @xmath101 for the fields can be found by solving the following discriminant equation @xmath102 this has two trivial solutions @xmath103 and @xmath104 , which are not negative for radiation and non - relativistic matter . in order to avoid the laplacian instability associated with the remaining solution of eq . ( [ cs2 ] ) we require that @xmath105}{w_{1}(4w_{1}w_{3}+9w_{2}^{2})}\geq0\,.\label{cscon}\ ] ] we also consider tensor perturbations @xmath106 characterized by @xmath107 , where @xmath106 is traceless ( @xmath108 ) and divergence - free ( @xmath109 ) . we decompose @xmath106 into the two polarization modes , @xmath110 , where the matrices @xmath111 and @xmath112 are normalized to be @xmath113 , ( where @xmath114 ) and @xmath115 in fourier space . the second - order action for tensor perturbations can be written as @xmath116\,,\ ] ] where the conditions for the avoidance of ghosts and laplacian instabilities are given by @xmath117 for the consistency of the theories given by the action ( [ action ] ) the conditions ( [ qscon ] ) , ( [ cscon ] ) , ( [ qtcon ] ) , and ( [ ctcon ] ) need to be satisfied . finally , as we will show in appendix a , the vector modes do not add any conditions to those derived above . in this section we shall study the cosmology based on the extended galileon models by taking into account the conditions for the avoidance of ghosts and laplacian instabilities derived in sec . [ avoidancesec ] . the covariant galileon without the field potential @xcite corresponds to @xmath118 where @xmath119 ( @xmath120 ) are dimensionless constants , @xmath121 is the reduced planck mass , and @xmath122 is a constant having the dimension of mass . in this case it is known that there exists a cosmological tracker solution characterized by the condition @xmath123constant @xcite . along the tracker the dark energy equation of state evolves as @xmath124 ( radiation era ) @xmath125 @xmath126 ( matter era ) @xmath125 @xmath127 ( de sitter era ) . however , the tracker solution for the covariant galileon is not favored by the joint data analysis of supernovae ia ( snia ) , cosmic microwave background ( cmb ) , and baryon acoustic oscillations ( bao ) @xcite . this comes from the unusual evolution of @xmath16 during the matter era away from @xmath0 . only the solutions that approach the tracker at late times can be allowed observationally . let us consider the generalization of the covariant galileon to find tracker solutions with different @xmath16 . we take the following functions @xmath128 where @xmath119 and @xmath61 ( @xmath120 ) are dimensionless constants , and @xmath129 ( @xmath120 ) are constants having dimensions of mass , @xmath130 , which recovers the dvali - turner model @xcite . ] . for two perfect fluids we consider radiation ( energy density @xmath131 , equation of state @xmath99 ) and non - relativistic matter ( energy density @xmath132 , equation of state @xmath133 ) . for the choice ( [ gegali ] ) the field equations ( [ be1 ] ) and ( [ be2 ] ) can be written as @xmath134 where @xmath135 and @xmath136 . for the covariant galileon ( @xmath137 , @xmath138 ) there is a tracker solution with @xmath139constant , in which case all the terms in eq . ( [ rhode ] ) are proportional to @xmath140 . we search for tracker solutions characterized by the condition @xmath141 where the power @xmath142 is a real constant . if we choose the following powers , all the terms in eq . ( [ rhode ] ) are proportional to @xmath143 : @xmath144 note that the covariant galileon corresponds to @xmath145 and @xmath146 . let us study whether the tracker really exists or not for the powers given by eq . ( [ power ] ) . before doing so we first discuss conditions for the existence of a de sitter solution characterized by @xmath11constant and @xmath147constant . we introduce the following variable @xmath148 for the simplification of the background equations it is convenient to define @xmath149 where @xmath150 is the hubble parameter at the de sitter solution . for the covariant galileon one has @xmath151 . from eqs . ( [ back1 ] ) and ( [ back2 ] ) we find that the de sitter fixed point where @xmath152 and @xmath153 is present under the following conditions @xmath154\,,\label{c3}\end{aligned}\ ] ] where @xmath155 is the value of @xmath156 at the de sitter solution , and @xmath157 in order to discuss the cosmological dynamics it is convenient to introduce the following variables @xmath158^{\frac{p+2q}{1 + 2q}}\,,\qquad\omega_{r}\equiv\frac{\rho_{r}}{3h^{2}m_{{\rm pl}}^{2}}\,.\label{eq : defr1r2}\ ] ] the solution ( [ eq : deftrack ] ) , in terms of the variable @xmath159 , is given by @xmath160 . the de sitter fixed point corresponds to @xmath161 . we will only consider cosmological dynamics for which @xmath159 and @xmath162 are both positive at all times : otherwise the inverse relations of ( [ eq : defr1r2 ] ) may be ill - defined . on the other hand , we will see later on that a viable cosmology can not allow a change of sign for the variable @xmath162 . defining the dark energy density parameter @xmath163 , it follows that @xmath164 - 3\alpha(2p+4q+1)\right\ } + 4\beta(p+3q+1)\bigr].\quad\;\,\label{eq : omde}\end{aligned}\ ] ] the friedmann equation gives the relation @xmath165 . for the initial conditions where @xmath159 is positive but small ( @xmath166 ) , we require that @xmath167 in order to have @xmath168 . in the following we will replace @xmath169 with @xmath170 . we can obtain the differential equations for @xmath159 , @xmath162 , and @xmath171 after deriving @xmath172 and @xmath173 from eqs . ( [ be2 ] ) and ( [ fieldeq ] ) , e.g. , @xmath174 , where a prime represents a derivative with respect to @xmath175 . the r.h.s . of these differential equations can be expressed in terms of @xmath159 , @xmath162 , and @xmath171 together with the coefficients @xmath176 , @xmath177 , @xmath178 , and @xmath142 . the equation for @xmath159 is given by @xmath179/\delta\,,\label{r1eq}\end{aligned}\ ] ] where @xmath180\,.\end{aligned}\ ] ] in addition to eq . ( [ r1eq ] ) we write the remaining equations of motion for the variables @xmath162 and @xmath171 . we find that the second einstein equation is equivalent to @xmath181\}\bigr\}\nonumber \\ & & \qquad{}+\frac{2r'_{2}}{r_{2}}(2q+1)\bigl\ { r_{1}^{\frac{p-1}{2q+1}}r_{2}\{3\beta(p+q)+r_{1}^{2}[3\alpha p-3\beta(p+q)+p+3\alpha q]-3\alpha r_{1}(p+q)\}+2q\bigr\}\nonumber \\ & & \qquad{}-3(2q+1)(p+2q)\{r_{1}^{\frac{p-1}{2q+1}}r_{2}[r_{1}^{3}(4\beta-3\alpha-2)+3\alpha r_{1}-4\beta]+2\}-2(2q+1)(p+2q)\,\omega_{r}=0\,,\end{aligned}\ ] ] whereas the equation of continuity for the radiation fluid leads to @xmath182 these two equations , combined with eq . ( [ r1eq ] ) , completely determine the dynamics of the system . from eq . ( [ r1eq ] ) we find that there is a fixed point characterized by @xmath183 which corresponds to the tracker solution with @xmath184 . considering the homogenous perturbations @xmath185 along the solution @xmath160 , it follows that @xmath186 for @xmath187 the tracker is stable along the @xmath159 direction provided that @xmath188 along the tracker the dark energy density parameter is simply given by @xmath189 for @xmath190 the r.h.s . of eq . ( [ con ] ) is within the range @xmath191 . as long as the condition ( [ con ] ) is satisfied , which includes the case of the covariant galileon , the solutions stay at the tracker . with the increase of @xmath162 the solutions finally reach the de sitter fixed point characterized by @xmath160 and @xmath192 . along the tracker the equations for @xmath162 and @xmath171 are given by @xmath193 combining these equations , it follows that @xmath194 integration of this equation leads to @xmath195 where @xmath196 is a constant . since @xmath197 , the evolution of the variable @xmath162 is @xmath198 . since we want @xmath162 to be subdominant at early times and to grow on the tracker solution , we further impose @xmath199 substituting eq . ( [ r2ana ] ) into ( [ omertra ] ) , we obtain the integrated solution @xmath200 where @xmath201 is a constant . therefore , the dynamics on the tracker depends only on the free parameter @xmath202 and the two initial conditions @xmath203 . for a particular choice of @xmath204 such as @xmath205 ( which corresponds to the covariant galileon ) , it is possible to derive the explicit solution for @xmath171 in terms of @xmath206 @xcite . for general @xmath204 , however , we can not necessarily find an explicit expression for @xmath171 . the density parameter of dark energy along the tracker is given by @xmath207 . writing the density parameters of dark energy and radiation today ( @xmath208 ) as @xmath209 and @xmath210 respectively and using eqs . ( [ omede ] ) , ( [ r2ana ] ) , and ( [ omerre ] ) , the coefficients @xmath196 and @xmath201 are found to be @xmath211 where @xmath212 is the density parameter of non - relativistic matter today . the dark energy equation of state @xmath213 and the effective total equation of state @xmath214 along the tracker are @xmath215 in the early cosmological epoch in which the condition @xmath216 is satisfied , one has @xmath217 and @xmath218 , respectively . hence the evolution of @xmath16 during the radiation and matter eras is given by @xmath219 and @xmath220 , respectively . at the de sitter fixed point ( @xmath192 and @xmath221 ) it follows that @xmath222 . the tracker solution for the covariant galileon ( @xmath205 ) is incompatible with observations because @xmath16 is away from @xmath0 during the matter and radiation eras @xcite . for the compatibility with observations we require that @xmath223 in the regime @xmath224 the conditions for the avoidance of ghosts and laplacian instabilities are given by @xmath225r_{2}>0\,,\label{qs1}\\ c_{s}^{2 } & \simeq & \bigl\{4p^{3}(\omega_{r}+3)-2p^{2}\{(\omega_{r}+3)(6\beta-3\alpha+2)-2q[3\omega_{r}+11 - 3(\alpha-2\beta)(\omega_{r}+3)]\}\nonumber \\ & & -3\{\beta(\omega_{r}+3)+8q^{3}(\omega_{r}+5)(\alpha-2\beta)-2q^{2}(7\omega_{r}+27)(\alpha-2\beta)+q[3\alpha(\omega_{r}+3)-2\beta(5\omega_{r}+17)]\}\nonumber \\ & & -p\{(\omega_{r}+3)(3\alpha-12\beta-1)+4q^{2}[(\alpha-2\beta)(9\omega_{r}+33)-2(\omega_{r}+5)]\nonumber \\ & & + q[12(2\beta-\alpha)(3\omega_{r}+10)+6\omega_{r}+22]\}\bigl\}/[24q^{2}(2p+4q-1)\{p-3(\alpha-2\beta)q\}]\geq0\,,\label{cs1}\\ \frac{q_{t}}{m_{{\rm pl}}^{2 } } & \simeq & \frac{1}{8}\left[2 + 3(\alpha-2\beta)r_{2}\right]>0\,,\label{qt1}\\ c_{t}^{2 } & \simeq & 1-\frac{3\{2[2(\alpha-2\beta)q+3\beta]p+8(\alpha-2\beta)q^{2}+\beta(16q-3)+\beta(2p+4q-1)\omega_{r}\}}{4q(2p+4q-1)}r_{2}\geq0\,.\label{ct1}\end{aligned}\ ] ] these results reproduce those derived in refs . @xcite for the covariant galileon ( @xmath226 ) . let us consider the case in which the parameters @xmath176 , @xmath177 , @xmath178 , @xmath142 are not very different from the order of unity . since @xmath224 the conditions ( [ qt1 ] ) and ( [ ct1 ] ) are automatically satisfied . we see here that @xmath162 can not change its sign , as this implies the violation of the condition @xmath227 . since we only consider the case @xmath228 at the initial stage , the scalar ghost can be avoided for @xmath229>0\,.\label{con1}\ ] ] for @xmath178 and @xmath142 satisfying @xmath230 , the laplacian instability of the scalar perturbation is absent as long as the numerator in eq . ( [ cs1 ] ) is positive . for the covariant galileon this corresponds to the condition @xmath231 @xcite . we study the stability of the late - time de sitter solution characterized by @xmath232 . at the de sitter fixed point the system is effectively described by one scalar degree of freedom @xmath49 , with the second - order action @xmath233 $ ] . for homogenous perturbations ( comoving wavenumber @xmath234 ) the scalar perturbation obeys the equation of motion @xmath235 , whose solution is @xmath236 where @xmath196 and @xmath201 are constants . now we are considering the de sitter solution where @xmath237 is constant , in which case @xmath238 does not vary in time . since @xmath239 , the second term in eq . ( [ phit ] ) decays in proportion to @xmath240 . hence the de sitter solution in our theory is classically stable against homogeneous perturbations . the conditions for the avoidance of ghosts and laplacian instabilities ( against inhomogeneous perturbations ) at @xmath232 are given by @xmath241}{[2p-6(\alpha-2\beta)q-3\alpha+6\beta-2]^{2}}>0\,,\label{qs2}\\ c_{s}^{2 } & = & \{6\beta+4p^{2}+p\,[9(\alpha-2\beta)^{2}+3\alpha-12\beta+4q(6\beta-3\alpha+2)-2]+3(\alpha-2\beta)[3\beta+q(9\alpha-12\beta-8q+6)]\}\nonumber \\ & & { } \times\frac{3(2\beta-\alpha)(2q+1)+2p-2}{6(6\beta-3\alpha-2)(p+2q)(2p+4q-1)(p-3\alpha q+6\beta q)}\geq0\,,\label{cs2}\\ \frac{q_{t}}{m_{{\rm pl}}^{2 } } & = & \frac{1}{8}\left(3\alpha-6\beta+2\right)>0\,,\label{qt2}\\ c_{t}^{2 } & = & \frac{2(2p+4q-1)-3\alpha}{(2p+4q-1)(3\alpha-6\beta+2)}\geq0\,.\label{ct2}\end{aligned}\ ] ] equation ( [ r1eq ] ) shows that there is another fixed point characterized by @xmath243 let us study the behavior of the solutions in the regime @xmath242 and @xmath224 . in doing so we consider the parameter space with @xmath244 and @xmath245 . then the equations for @xmath159 , @xmath162 , and @xmath171 are approximately given by @xmath246}{2(2q+1)(2p+6q-1)}\ , r_{2}\,,\label{r2ap}\\ \omega_{r } ' & \simeq & \omega_{r}(\omega_{r}-1)\,.\label{omer}\end{aligned}\ ] ] from eq . ( [ omer ] ) there are two fixed points : @xmath247 and @xmath221 . during the radiation era ( @xmath247 ) integration of eqs . ( [ r1ap ] ) and ( [ r2ap ] ) gives @xmath248 whereas during the matter era ( @xmath221 ) one has @xmath249 the variable @xmath159 increases for @xmath244 and @xmath245 , which is followed by the approach to the tracker solution . whether the variable @xmath162 grows or not depends on the values of @xmath178 and @xmath142 . if @xmath250 and @xmath245 , for example , @xmath162 decreases in the matter era . however the dynamics will in general change as the solutions approach the tracker , @xmath251 . the dark energy equation of state @xmath16 and the total effective equation of state @xmath252 in the regime @xmath242 and @xmath224 are approximately given by @xmath253 if @xmath145 and @xmath146 , we have @xmath254 @xcite . for @xmath255 the ghosts and laplacian instabilities can be avoided for @xmath256>0\,,\label{qt3}\\ c_{t}^{2 } & \simeq & 1+\frac{3(4p+12q-5 - 3\omega_{r})}{4p+12q-2}\beta r_{1}^{(p-1)/(2q+1)}r_{2}\geq0\,.\label{ct3}\end{aligned}\ ] ] in the regime @xmath242 and @xmath224 the conditions ( [ qt3 ] ) and ( [ ct3 ] ) are automatically satisfied . from eq . ( [ r1eq ] ) we see that in general there are other two more complicated fixed points , @xmath257 , those which satisfy the equation @xmath258r_{j}+2\beta(p+3q)=0\,,\label{eq : spur}\ ] ] with @xmath259 . whether or not these fixed points are viable or not depends on their stability and the chosen parameters of the model . then we will consider only those models for which either there are no real solutions to eq . ( [ eq : spur ] ) , or , if they exist , they are placed outside the range of interest corresponding to the interval @xmath260 . therefore we need to fulfill the following condition @xmath261\,,\label{eq : otherfp}\ ] ] where @xmath262^{2}-8\beta p(3\alpha-4\beta+2)(p+3q)\,.\end{aligned}\ ] ] let us discuss the relevant constraints we have found so far in order to restrict the allowed parameter space . therefore we have @xmath263 & > 0\ , , & & \textrm{in order to have } \left.q_{s}\right|_{{\rm de\ sitter}}>0\,,\\ q\left[p-3(\alpha-2\beta)q\right ] & > 0\ , , & & \textrm{in order to have } \left.q_{s}\right|_{r_{1}=1,r_{2}\ll1}>0\,,\label{ds2}\\ \left.c_{s}^{2}\right|_{{\rm de\ sitter } } & \geq0\ , , & & \!\!\left.c_{t}^{2}\right|_{{\rm de\ sitter}}\geq0\,,\\ \left.c_{s}^{2}\right|_{r_{1}=1,r_{2}\ll1,\omega_{r}=0 } & \geq0\ , , & & \!\!\left.c_{s}^{2}\right|_{r_{1}=1,r_{2}\ll1,\omega_{r}=1}\geq0\,.\label{eq : condm}\end{aligned}\ ] ] the conditions ( [ eq : conda])-([eq : condd ] ) impose @xmath264 this implies that negative values of @xmath178 and @xmath142 are excluded . furthermore , the condition ( [ eq : conde ] ) implies @xmath265 altogether we find the following allowed parameter space @xmath266,\label{eq : conds}\end{aligned}\ ] ] where @xmath267 although the above conditions have been derived for different initial conditions , we impose them to be true at the same time . this is because we do not know a priori initial conditions in the early universe , as we have not specified how inflation works in these models . to be more concrete , let us consider the theory with @xmath145 and @xmath268 . in this case the dark energy equation of state along the tracker is given by @xmath269 during the matter era . the condition ( [ eq : conds ] ) reduces to @xmath270\,.\label{eq : allcondspec}\ ] ] adding the condition ( [ eq : otherfp ] ) , the parameter space reduces to @xmath271.\label{eq : allandoth}\end{aligned}\ ] ] during the transition from the regime @xmath160 and @xmath224 to the de sitter attractor ( @xmath160 and @xmath192 ) the tensor propagation speed squared @xmath272 can be negative , depending on the model parameters . for @xmath145 and @xmath268 we have @xmath273(5+r_{2})}\,,\label{ctex}\ ] ] along the tracker . since the transition occurs at late times , the contribution of the radiation density parameter can be neglected in eq . ( [ ctex ] ) . if @xmath274 and @xmath275 , then @xmath272 is close to 1 . for positive @xmath176 and @xmath177 of the order of unity it happens that @xmath272 becomes negative . taking the derivative of eq . ( [ ctex ] ) with respect to @xmath162 , @xmath272 has a minimum ( the second derivative @xmath276 , at @xmath277 , is always positive in the region @xmath278 and when the conditions ( [ eq : allcondspec ] ) are satisfied ) at @xmath279 where @xmath280\,.\ ] ] in eq . ( [ r2so ] ) we have chosen the solution with @xmath228 , because another solution is always negative if the conditions ( [ eq : allcondspec ] ) are satisfied . ) plane ( the area in black ) for @xmath145 and @xmath268 . we have used all the available conditions considered so far , i.e. ( [ eq : allandoth ] ) and ( [ eq : r2mincond]).,width=453 ] plugging the solution ( [ r2so ] ) into eq . ( [ ctex ] ) , we find that @xmath281 therefore the allowed parameter space is characterized by @xmath282\lor[\delta_{{\rm min}}\geq0\land(0<r_{2,{\rm min}}<1)\land\left.c_{t}^{2}\right|_{r_{2,{\rm min}}}\geq0]\,.\label{eq : r2mincond}\ ] ] the first inequality comes from setting @xmath283 to be an imaginary number such that there is no minimum for @xmath272 . the second inequality allows the minimum of @xmath272 to be outside the interested range for @xmath162 . in fig . [ fig : allow ] we plot the allowed parameter space for @xmath145 and @xmath268 in the @xmath284 plane constrained by the conditions ( [ eq : allandoth ] ) and ( [ eq : r2mincond ] ) . , @xmath169 , @xmath171 , @xmath252 , and @xmath16 versus the redshift @xmath285 for @xmath145 , @xmath268 , @xmath286 , @xmath287 , and @xmath288 . the initial conditions are chosen to be @xmath160 , @xmath289 , and @xmath290 at @xmath291 . in this case the solution is on the tracker from the beginning.,width=307,height=316 ] [ omega ] versus @xmath285 for @xmath145 , @xmath268 , @xmath286 , @xmath287 , and @xmath288 with several different initial conditions . the solid line corresponds to the tracker with the initial conditions same as those given in fig . [ omega ] . the initial conditions for the cases ( a)-(c ) are ( a ) @xmath292 , @xmath293 , @xmath290 at @xmath294 , ( b ) @xmath295 , @xmath296 , @xmath290 at @xmath291 , and ( c ) @xmath297 , @xmath298 , @xmath299 at @xmath300 , respectively.,width=307,height=307 ] [ wde ] versus @xmath285 for the same model parameters and initial conditions as those given in fig . [ wde ] . the solid line represents the tracker solution , whereas the cases ( a ) , ( b ) , and ( c ) correspond to the evolution for the initial conditions as those given in fig . [ wde].,width=307,height=288 ] [ sghost ] versus @xmath285 for the same model parameters and initial conditions as those given in fig . the solid line represents the tracker solution , whereas the cases ( a ) , ( b ) , and ( c ) correspond to the evolution for the initial conditions as those given in fig . [ wde].,width=307,height=307 ] [ cs ] in order to confirm the analytic estimation given above , we solve the differential equations of @xmath159 , @xmath162 , and @xmath171 numerically for given values of @xmath176 , @xmath177 , @xmath178 , and @xmath142 . in particular we study the cosmological evolution for @xmath145 and @xmath268 , in which case the parameters @xmath176 and @xmath177 need to satisfy the constraints ( [ eq : allandoth ] ) and ( [ eq : r2mincond ] ) . in fig . [ omega ] we plot the evolution of @xmath301 , @xmath169 , @xmath171 , @xmath252 , and @xmath16 versus the redshift @xmath302 for @xmath145 , @xmath268 , @xmath286 , @xmath287 , and @xmath288 with the initial conditions corresponding to the tracker from the beginning ( @xmath160 ) . for this choice of @xmath176 and @xmath177 the conditions ( [ eq : allandoth ] ) and ( [ eq : r2mincond ] ) are satisfied . note that the coefficients @xmath119 ( @xmath303 ) are known to be @xmath304 , @xmath305 , @xmath306 , and @xmath307 from eqs . ( [ c2])-([ab ] ) . furthermore both the fixed points given in eq . ( [ eq : spur ] ) exist in the region @xmath308 . figure [ omega ] shows that the sequence of radiation ( @xmath309 , @xmath247 ) , matter ( @xmath310 , @xmath311 ) , and de sitter ( @xmath312 , @xmath313 ) eras is in fact realized . the dark energy density parameter grows as @xmath314 toward the de sitter attractor ( @xmath232 ) . as estimated by eq . ( [ wdetra ] ) , the evolution of @xmath16 during the radiation and matter eras are given by @xmath315 and @xmath269 , respectively . after the end of the matter - dominated epoch the solution approaches the de sitter attractor with @xmath127 . in fig . [ wde ] we show the variation of @xmath16 for the same model parameters as those used in fig . [ omega ] with a number of different initial conditions . the approach to the tracker occurs later for smaller initial values of @xmath159 . this can be clearly seen in the numerical simulations of the cases ( a)-(c ) in fig . the case ( a ) corresponds to the early tracking , whereas the case ( c ) to the late - time tracking with smaller initial values of @xmath159 . in the regime characterized by @xmath242 and @xmath224 , the analytic estimation ( [ wdeap ] ) gives @xmath316 and @xmath317 during the radiation and the matter eras , respectively . these analytic values of @xmath16 are in good agreement with their numerical values for the late - time tracking solution ( such as the case ( c ) in fig . [ wde ] ) . in fig . [ sghost ] the evolution of the quantity @xmath238 is shown for the same model parameters and the initial conditions as those given in fig . [ wde ] . for the tracker our numerical simulations show that @xmath238 grows according to eq . ( [ qs1 ] ) in the regime @xmath224 ( i.e. @xmath318 ) , which finally approaches the value ( [ qs2 ] ) at the de sitter solution . for the initial conditions with @xmath242 and @xmath224 we find that the early evolution of @xmath238 is well described by eq . ( [ qs3 ] ) , i.e. @xmath319 during the radiation era . as we can see in fig . [ sghost ] , the evolution of @xmath238 shifts to that of the tracker after the solutions reach the regime around @xmath160 . provided that @xmath228 initially , @xmath238 always remains to be positive . as we see in eqs . ( [ qt1 ] ) and ( [ qt3 ] ) the quantity @xmath320 is close to the value @xmath321 in the regime @xmath224 , independently of the values of @xmath159 . numerically we confirmed that , for both the initial conditions @xmath160 and @xmath242 , @xmath320 stays the value around @xmath321 until recently and then it finally approaches the value ( [ qt2 ] ) at the de sitter solution . provided that @xmath322 the no - ghost condition for the tensor perturbation is always satisfied . in fig . [ cs ] we illustrate the evolution of @xmath323 for the same model parameters and the initial conditions as those given in fig . [ wde ] . the analytic estimation ( [ cs1 ] ) for the tracker gives @xmath324 and @xmath325 during the radiation and matter dominated epochs respectively , which show good agreement with the numerical result in fig . finally @xmath323 approaches the value @xmath326 with a smooth transition from the matter era to the de sitter epoch . in the regime @xmath242 and @xmath224 the analytic estimation ( [ cs3 ] ) gives @xmath327 . in fact the numerical simulations for the cases ( a ) , ( b ) , and ( c ) reproduce this value before the solutions reach the tracker . since @xmath328 from the radiation era to the de sitter epoch , the laplacian instability of the scalar perturbation is absent . versus @xmath285 along the tracker for ( i ) @xmath145 , @xmath268 , @xmath286 , @xmath287 , @xmath288 , and ( ii ) @xmath145 , @xmath268 , @xmath329 , @xmath330 , @xmath288 . in both cases the initial conditions are chosen to be @xmath160 , @xmath289 , and @xmath290 . in the case ( ii ) the condition ( [ eq : r2mincond ] ) is violated , so that @xmath272 temporally becomes negative during the transition from the regime @xmath224 to the regime @xmath192.,width=307,height=307 ] [ ct ] figure [ ct ] depicts the evolution of @xmath272 for the tracker with two different combinations of @xmath176 and @xmath177 . the case ( i ) corresponds to the model parameters and the initial conditions as those given in fig . [ omega ] . as estimated by eq . ( [ ct1 ] ) , the tensor propagation speed squared along the tracker is very close to 1 but slightly less than 1 in the early cosmological epoch . in the case ( i ) the model parameters satisfy the condition ( [ eq : r2mincond ] ) , so that @xmath272 takes a positive minimum value 0.745 at @xmath331 . then the solution finally approaches the de sitter attractor with @xmath332 . in the case ( ii ) , in which the condition ( [ eq : r2mincond ] ) is violated , the minimum value of @xmath272 is negative . this shows that the condition ( [ eq : r2mincond ] ) is in fact required to avoid the laplacian instability of the tensor perturbation . we have also run our numerical code for the initial conditions with @xmath333 and found that the evolution of @xmath272 is similar to that for the tracker . the only difference is that the tensor propagation speed is slightly superluminal in the regime @xmath333 . in fact , eq . ( [ ct3 ] ) shows that @xmath272 is slightly larger than 1 under the conditions ( [ pqcon ] ) and ( [ betacon ] ) . we have also carried out the numerical simulations for other values of @xmath334 and confirmed the accuracy of the analytic estimation . if we choose larger values of @xmath142 for given @xmath178 , the dark energy equation of state @xmath16 along the tracker approaches @xmath0 . in the limit where @xmath335 the scalar propagation speed squared ( [ cs2 ] ) at the de sitter solution reduces to @xmath336 and hence we require the condition @xmath337 . on the other hand the condition ( [ ds2 ] ) reduces to @xmath338 for @xmath335 . hence , if @xmath16 along the tracker is close to @xmath0 , the allowed parameter space tends to be smaller . in the horndeski s most general scalar - tensor theories we derived conditions for the avoidance of ghosts and laplacian instabilities associated with scalar , tensor , and vector perturbations . the four conditions ( [ qscon ] ) , ( [ cscon ] ) , ( [ qtcon ] ) , and ( [ ctcon ] ) need to be satisfied for the theoretical consistency . vector perturbations do not give rise to any additional conditions to those derived for scalar and tensor perturbations . the horndeski s action covers most of the dark energy models proposed in literature and hence our formulas are general enough to apply them to concrete models with second - order field equations . we proposed new kinetically driven dark energy models described by the functions ( [ gegali ] ) , which cover the covariant galileon as a specific case . for the choice of the powers @xmath61 ( @xmath120 ) given in eq . ( [ power ] ) we showed the existence of the tracker solution along which @xmath339constant . finally the solutions approach a stable de sitter attractor at which @xmath11constant . along the tracker the dark energy equation of state during the matter dominance is found to be @xmath340 . the covariant galileon ( @xmath145 and @xmath146 ) corresponds to @xmath126 during the matter era , which is not favored from the combined data analysis of snia , cmb , and bao . the extended galileon model we proposed can alleviate this problem because @xmath16 can be close to @xmath0 for @xmath178 smaller than @xmath142 . we clarified the theoretically allowed parameter space in which the ghosts and laplacian instabilities are absent . for @xmath145 and @xmath268 we carried out numerical simulations to check the evolution of the background quantities ( like @xmath16 and @xmath301 ) as well as the quantities such as @xmath323 , @xmath272 , @xmath238 , and @xmath341 . as we estimated analytically , the dark energy equation of state for the tracker evolves as @xmath315 ( radiation era ) , @xmath269 ( matter era ) , and @xmath127 ( de sitter era ) , see fig . [ wde ] . for the initial conditions with @xmath242 , @xmath16 starts to evolve from the value estimated by eq . ( [ wdeap ] ) . the approach to the tracker occurs later for smaller initial values of @xmath159 . for the values of @xmath176 and @xmath177 which are inside the allowed parameter space , our numerical simulations show that @xmath323 , @xmath272 , @xmath238 , and @xmath341 remain to be positive in the cosmic expansion history . note that the condition ( [ eq : r2mincond ] ) is important to avoid that @xmath272 becomes negative during the transition from the matter era to the de sitter epoch . while we showed the cosmological evolution for one choice of @xmath178 and @xmath142 , we also confirmed that the analytic estimation is trustable for other values of @xmath178 and @xmath142 . in the limit that @xmath342 the dark energy equation of state for the tracker mimics that in the @xmath343cdm model . it will be of interest to see how the combined data analysis of snia , cmb , and bao places constraints on the tracker solution in the extended galileon models . in order to confront this model with the observations of large scale structure and weak lensing , we also need to study the evolution of matter density perturbations as well as gravitational potentials . we leave these issues for future work . a.d.f . and s.t . were supported by the grant - in - aid for scientific research fund of the jsps nos . 10271 and 30318802 . is also supported by the thep center . also thanks financial support for the grant - in - aid for scientific research on innovative areas ( no . 21111006 ) . we thank alexander vikman for his initial collaboration on this work . is grateful to the organizers of sixth aegean summer school and the dark universe conference for their kind hospitalities during which a part of this work was done . the no - ghost conditions for the vector modes in the presence of two perfect fluids can be found by using the method in ref . let us consider the perturbed metric @xmath344 where @xmath345 . in order to describe the vector perturbations at linear order , for the perfect fluid we write @xmath346 , where @xmath347 and @xmath348 are the velocity potentials for the fluid @xmath70 . we can choose the background values for @xmath347 and @xmath348 as follows : @xmath349 , and @xmath350 . then , for a plane wave in fourier space , we have @xmath351 . by splitting @xmath352 as @xmath353 , and choosing the gauge for which @xmath354 , ( where @xmath355 is the component of @xmath356 perpendicular to @xmath357 ) , and the arbitrary background quantities such that @xmath358 , we find that @xmath357 is parallel to @xmath356 and both are perpendicular to @xmath359 . after expanding the action at second order in the fields and integrating out the auxiliary fields , we obtain @xmath360\,,\ ] ] where @xmath361 , @xmath362 , and @xmath363 are time - dependent coefficients . the two no - ghost conditions can be written as @xmath364}{2\,[2(1+w_{{a}})\rho_{{a}}+2\,\rho_{{b}}(1+w_{b})+w_{{1}}{(k / a)}^{2}]}}>0\,,\label{eq : vecghost}\\ q_{11}^{v}q_{22}^{v}-(q_{12}^{v})^{2 } & = & a^{10}\,{\frac{w_{{1}}{(k / a)}^{2}(1+w_{{a}})\rho_{{a}}(1+w_{{b}})\rho_{{b}}}{4[2(1+w_{a})\,\rho_{{a}}+2\,(1+w_{b})\rho_{{b}}+w_{{1}}{(k / a)}^{2}]}}>0\,.\end{aligned}\ ] ] these conditions are satisfied for @xmath365 ( which corresponds to the condition ( [ qtcon ] ) ) and @xmath366 , and @xmath367 . in general relativity with one single perfect fluid , only the condition ( [ eq : vecghost ] ) holds , and it agrees with the result in ref . @xcite , when @xmath368 . hence the vector perturbations do not provide additional constraints to those derived for scalar and tensor perturbations . v. sahni and a. a. starobinsky , int . j. mod . d * 9 * , 373 ( 2000 ) ; s. m. carroll , living rev . rel . * 4 * , 1 ( 2001 ) ; t. padmanabhan , phys . rept . * 380 * , 235 ( 2003 ) ; p. j. e. peebles and b. ratra , rev . mod . * 75 * , 559 ( 2003 ) ; e. j. copeland , m. sami and s. tsujikawa , int . j. mod . d * 15 * , 1753 ( 2006 ) ; s. tsujikawa , arxiv:1004.1493 [ astro-ph.co ] . y. fujii , phys . rev . d * 26 * , 2580 ( 1982 ) ; l. h. ford , phys . d * 35 * , 2339 ( 1987 ) ; c. wetterich , nucl . phys b. * 302 * , 668 ( 1988 ) ; b. ratra , p. j. e. peebles , phys . rev . * d37 * , 3406 ( 1988 ) ; t. chiba , n. sugiyama and t. nakamura , mon . not . * 289 * , l5-l9 ( 1997 ) ; r. r. caldwell , r. dave and p. j. steinhardt , phys . 80 * , 1582 ( 1998 ) . c. armendariz - picon , t. damour and v. f. mukhanov , phys . * b458 * , 209 - 218 ( 1999 ) ; t. chiba , t. okabe and m. yamaguchi , phys . * d62 * , 023511 ( 2000 ) ; c. armendariz - picon , v. f. mukhanov and p. j. steinhardt , phys . rev . * 85 * , 4438 - 4441 ( 2000 ) . s. capozziello , int . j. mod . d * 11 * , 483 ( 2002 ) ; s. capozziello , s. carloni and a. troisi , recent res . . astron . astrophys . * 1 * , 625 ( 2003 ) ; s. m. carroll , v. duvvuri , m. trodden and m. s. turner , phys . rev . d * 70 * , 043528 ( 2004 ) . w. hu and i. sawicki , phys . d * 76 * , 064004 ( 2007 ) ; a. a. starobinsky , jetp lett . * 86 * , 157 ( 2007 ) ; s. a. appleby and r. a. battye , phys . b * 654 * , 7 ( 2007 ) ; l. amendola and s. tsujikawa , phys . b * 660 * , 125 ( 2008 ) ; s. tsujikawa , phys . d * 77 * , 023507 ( 2008 ) . s. m. carroll , a. de felice , v. duvvuri , d. a. easson , m. trodden and m. s. turner , phys . * d71 * , 063513 ( 2005 ) ; s. nojiri , s. d. odintsov , phys . b631 * , 1 - 6 ( 2005 ) ; i. navarro and k. van acoleyen , jcap * 0603 * , 008 ( 2006 ) ; a. de felice and s. tsujikawa , phys . * b675 * , 1 - 8 ( 2009 ) ; a. de felice and t. suyama , jcap * 0906 * , 034 ( 2009 ) ; a. de felice and t. suyama , phys . rev . * d80 * , 083523 ( 2009 ) ; a. de felice , d. f. mota and s. tsujikawa , phys . rev . * d81 * , 023532 ( 2010 ) . t. p. sotiriou and v. faraoni , rev . * 82 * , 451 ( 2010 ) ; a. de felice and s. tsujikawa , living rev . * 13 * , 3 ( 2010 ) ; s. tsujikawa , lect . notes phys . * 800 * , 99 - 145 ( 2010 ) ; t. clifton , p. g. ferreira , a. padilla and c. skordis , arxiv:1106.2476 [ astro-ph.co ] . c. deffayet , g. r. dvali , g. gabadadze and a. i. vainshtein , phys . d * 65 * , 044026 ( 2002 ) ; m. porrati , phys . b * 534 * , 209 ( 2002 ) ; m. a. luty , m. porrati and r. rattazzi , jhep * 0309 * , 029 ( 2003 ) . s. m. carroll _ et al . , _ phys . d * 71 * , 063513 ( 2005 ) ; s. nojiri , s. d. odintsov and m. sasaki , phys . d * 71 * , 123509 ( 2005 ) ; g. calcagni , s. tsujikawa and m. sami , class . * 22 * , 3977 ( 2005 ) ; a. de felice , m. hindmarsh and m. trodden , jcap * 0608 * , 005 ( 2006 ) . n. chow and j. khoury , phys . rev . * d80 * , 024037 ( 2009 ) ; f. p. silva and k. koyama , phys . rev . d * 80 * , 121301 ( 2009 ) ; c. de rham and a. j. tolley , jcap * 1005 * , 015 ( 2010 ) ; t. kobayashi , h. tashiro and d. suzuki , phys . rev . d * 81 * , 063513 ( 2010 ) ; t. kobayashi , phys . rev . * d81 * , 103533 ( 2010 ) ; a. de felice and s. tsujikawa , jcap * 1007 * , 024 ( 2010 ) ; a. de felice , s. mukohyama and s. tsujikawa , phys . * d82 * , 023524 ( 2010 ) ; k. hinterbichler , m. trodden and d. wesley , phys . * d82 * , 124018 ( 2010 ) ; t. kobayashi , m. yamaguchi and j. yokoyama , phys . 105 * , 231302 ( 2010 ) ; d. f. mota , m. sandstad and t. zlosnik , jhep * 1012 * , 051 ( 2010 ) ; c. burrage , c. de rham , d. seery and a. j. tolley , jcap * 1101 * , 014 ( 2011 ) ; k. hirano and z. komiya , arxiv:1012.5451 [ astro-ph.co ] ; g. goon , k. hinterbichler and m. trodden , arxiv:1109.3450 [ hep - th ] ; z. -g . liu , j. zhang and y. s. piao , arxiv:1105.5713 [ astro-ph.co ] ; h. wang , t. qiu and y. s. piao , arxiv:1110.1795 [ hep - ph ] . l. amendola , phys . * b 301 * , 175 ( 1993 ) ; c. germani and a. kehagias , phys . lett . * 105 * , 011302 ( 2010 ) ; phys . lett . * 106 * , 161302 ( 2011 ) ; c. germani and y. watanabe , jcap * 1107 * , 031 ( 2011 ) ; g. gubitosi , e. v. linder , phys . lett . * b703 * , 113 - 118 ( 2011 ) .

we study the physical electron in quantum electrodynamics expanded on the light - cone fock space in order to address two problems : ( 1 ) the physics of the electron s anomalous magnetic moment in nonperturbative qed , and ( 2 ) the practical problems of ultraviolet regularization and renormalization in truncated nonperturbative light - cone hamiltonian theory . we present results for computed in a light - cone gauge fock space truncated to include one bare electron and at most two photons ; i.e. , up to two photons in flight . the calculational scheme uses an invariant mass cutoff , discretized light - cone quantization ( dlcq ) , a tamm dancoff truncation of the fock space , and a photon mass regulator . we introduce new weighting methods which greatly improve convergence to the continuum within dlcq . nonperturbative renormalization of the coupling and electron mass are carried out , and a limit on the magnitude of the effective physical coupling strength is computed . a large renormalized coupling strength is then used to make the nonperturbative effects in the electron anomalous moment from the one - electron , two - photon fock state sector numerically detectable .

hep-ph9806541

many years ago feynman issued the following challenge@xcite : `` it seems that very little physical intuition has yet been developed in this subject [ of quantum electrodynamics ] . in nearly every case we are reduced to computing exactly the coefficient of some specific term . we have no way to get a general idea of the result to be expected . as a specific challenge , is there any method of computing the anomalous moment of the electron which , on first rough approximation , gives a rough approximation to the @xmath2 term and a crude one to @xmath3 ; and when improved , increases the accuracy of the @xmath3 term , yielding a rough estimate to @xmath4 and beyond ? '' this challenge was taken up by drell and pagels@xcite , who used a sideways dispersion relation and low - energy theorems for compton scattering@xcite to construct consistency conditions for the anomalous moment . their approach did meet with some success , particularly in understanding the sign of the @xmath2 term ; however , the dispersion relation requires an ultraviolet cutoff , and low - energy approximations of the integrand are not completely adequate . we propose to meet feynman s challenge by using discretized light - cone@xcite quantization@xcite ( dlcq)@xcite . by constructing the dressed electron state in fock space we can in principle compute physical properties of the electron exactly@xcite . in practice , various truncations are required , but the approach remains nonperturbative . instead of producing an expansion in @xmath2 , we produce an expansion in fock particles , _ i.e_. the number of photons in flight . the computation is equivalent to a selective summation of graphs to all orders , but is actually done by diagonalizing a matrix approximation of the light - cone mass - squared operator . in this form the calculation becomes a testing ground for techniques of nonperturbative renormalization . when two - photon intermediate states are allowed , graphs such as the multiloop graph in fig . [ fig : samplegraphs](a ) enter the calculation and are summed to all orders . even at the one - photon level , the calculation is nonperturbative because there are infinite - order contributions from graphs of the sort in fig . [ fig : samplegraphs](b ) and ( c ) . however , in the case of fig . [ fig : samplegraphs](c ) , crossed - photon graphs and z graphs are needed to cancel a divergence at zero longitudinal momentum for any instantaneous electron . because of these cancellations , we place the nonperturbative part of the one - photon contribution into a two - photon calculation . = the selection of the graphs to be summed is driven by the truncations made and is designed to make the calculation tractable . the truncation in particle number is physically reasonable ; the work of drell and pagels@xcite shows that states with few photons in flight are dominant . as the actual number of particles is varied and as different truncations of the interactions are explored , one can gain a better understanding of the physics of the anomalous moment . thus our calculation can be viewed as the beginning of a possible program , with systematic improvements available . it might even become a model for how to proceed with nonperturbative calculations in quantum chromodynamics ( qcd ) . in the work presented here we regulate the theory by an invariant - mass cutoff , which limits the light - cone energy of the fock states included in the basis , and by a tamm - dancoff truncation@xcite of the number of constituents . these restrictions keep the numerical calculation to a very reasonable size but they do complicate the renormalization . the truncations turn the bare parameters of the theory into fock - sector dependent functions of momentum@xcite and require careful construction of appropriate counterterms@xcite to correctly approximate the solution to the original theory@xcite . we fix these functions by applying conditions on mass eigenvalues and vertices in the presence of spectator constituents . two alternative renormalization procedures now exist . one is the similarity transformation developed by wilson and gazek@xcite and wegner@xcite , where counterterms are generated perturbatively as the hamiltonian matrix is narrowed in the range of allowed light - cone energies ; the final hamiltonian matrix is used nonperturbatively . this approach has been applied by perry and co - workers@xcite . the other is based on introduction of pauli villars regulators@xcite before the quantization and numerical schemes are selected , so that counterterms can be supplied simply by adjusting the bare parameters . this approach has just recently been successfully tested in a simple model@xcite , and it should soon be considered for the anomalous - moment problem studied here . we compute the anomalous moment@xcite from a spin - flip matrix element of the plus component of the electromagnetic current@xcite . we approximate the fock - state expansion of the dressed electron with a truncation to no more than two photons and one electron . the eigenvalue problem for the wave functions becomes a coupled set of three integral equations . to construct these equations we use the light - cone hamiltonian derived by tang _ et al_.@xcite , regulated by the invariant - mass cutoff . the photon mass is taken to be one tenth of the electron mass , to help control infrared divergences . the coupling strength is set at @xmath5 , because limitations on numerical accuracy make nonperturbative effects discernible only at large coupling . the calculation is not an attempt to compete with the accuracy of the perturbative calculations by kinoshita and co - workers@xcite . the bare electron mass in the one - photon sector is computed from the one - loop correction allowed by the two - photon states , where one photon is a spectator . we then require that the bare mass in the no - photon sector be such that physical mass is an eigenvalue of the light - cone hamiltonian . the three - point bare coupling is also sector dependent . there are no vacuum polarization effects , because pair production is removed by the tamm dancoff truncation . however , the truncation violates the ward identity so that vertex and wave function renormalization do not cancel@xcite . a consequence of this is that the physical coupling is limited to a cutoff - dependent finite range of allowed values . we compute the critical coupling that defines the upper limit . the vertex renormalization is fixed by considering the proper part of the transition amplitude for photon absorption by an electron at zero photon momentum . a means by which this transition amplitude can be computed from the lowest eigenstate is constructed . full diagonalization of the hamiltonian is not required ; however , the renormalization condition and the eigenvalue problem must be solved as a coupled system . the wave function renormalization is directly available from the bare amplitude in the fock state expansion . to within the accuracy of the calculation , the values for the anomalous moment become constant for an ultraviolet cutoff sufficiently large . however , most four - point graphs that arise in the bound - state problem are log divergent . to any order the divergences cancel if all graphs are included , but the tamm dancoff truncation spoils this . the resulting logarithmic effects are not detectable in the numerical results . the calculations presented here build on the significant amount of work that has been done recently on the use of light - cone quantization@xcite in the construction of solvable bound - state problems for strongly interacting theories . the coordinates used are based on the choice of @xmath6 as the time coordinate , where @xmath7 is the ordinary time and @xmath8 any cartesian spatial coordinate . a variety of field theories have been analyzed in this way . most theories considered have been simple model theories in one space dimension ; however , there have been studies of three dimensional theories , including the wick cutkosky model@xcite , the yukawa model@xcite , quantum electrodynamics ( qed)@xcite , and quantum chromodynamics ( qcd)@xcite . there has also been some work on nonperturbative scattering calculations@xcite and on the stationary phase approximation to the soliton in @xmath9@xcite . much of this work has involved numerical studies . brodsky , pauli , and co - workers have analyzed various one - dimensional theories ( yukawa@xcite , qed@xcite , and qcd@xcite ) and have devoted a considerable amount of effort to three - dimensional qed@xcite . some work on the three - dimensional wick cutkosky model@xcite has been done by sawicki and co - workers , ji and furnstahl@xcite , and wivoda and hiller@xcite . one - dimensional scalar theories , @xmath10 and @xmath11 , have been studied by harindranath and vary@xcite . work on the one - dimensional yukawa model has been done by harindranath , shigemitsu and perry@xcite ; this was based on a tamm dancoff truncation@xcite and used basis - function methods as well as a discretization technique . the basis - function methods have been extended to the three - dimensional case by gazek _ et al_.@xcite . a preliminary treatment of qcd in three dimensions has been attempted by hollenberg and co - workers@xcite . dimensional reduction of qcd to an effective theory in @xmath12 dimensions has also been considered@xcite . other nonperturbative approaches applicable to qcd include lattice gauge theory@xcite , sum rules@xcite , and schwinger . a particularly successful form of lattice theory has been developed by lepage and collaborators@xcite who use tadpole - improved actions to reduce discretization errors . the hamiltonian form of lattice theory@xcite is actually similar to the approach usually taken in light - cone quantization , in that a hamiltonian operator is constructed and partially diagonalized@xcite . there has also been work on combinations of the lattice with light - cone methods in the transverse lattice method@xcite and in direct use of a light - cone lattice@xcite . two important aspects of qcd that all these methods address are vacuum structure and symmetry breaking . in light - cone quantization , the vacuum appears to be the trivial perturbative vacuum . this has the advantage that one can compute massive states immediately without first computing the vacuum state . however , in equal - time quantization , the nontrivial structure of the qcd vacuum is known to be important . this paradox of the trivial vacuum has received much attention . nonperturbative analyses of various light - cone models indicate that interactions induced by zero modes@xcite and other considerations@xcite play important roles in generating effects such as symmetry breaking@xcite that are usually associated with the vacuum . the progress made recently in the application of light - cone quantization owes much to earlier work@xcite the development then was aimed at perturbation theory , and in particular its application to deep inelastic scattering , but much has been carried over to bound - state problems . new work on perturbation theory has also been done@xcite . an outline of the remainder of the paper is as follows . the discretized light - cone formulation of the anomalous moment problem is given in sec . [ sec : dlcq ] . the nonperturbative mass and coupling renormalization that we use are described in sec . [ sec : renormalization ] . finite corrections associated with photon zero modes and with ambiguities in infinite renormalizations are discussed in sec . [ sec : finitecorrections ] . our results are presented in sec . [ sec : results ] , and a brief summary is given in sec . [ sec : summary ] . we define light - cone coordinates@xcite by @xmath13 momentum variables are similarly constructed as @xmath14 the time variable is taken to be @xmath15 , and time evolution of a system is then determined by @xmath16 , the operator associated with the momentum component conjugate to @xmath15 . usually one seeks stationary states obtained as eigenstates of @xmath16 . frequently the eigenvalue problem is expressed in terms of a light - cone hamiltonian@xcite ( mass - squared operator ) @xmath17 as @xmath18 where @xmath19 is the mass of the state , and @xmath20 and @xmath21 are momentum operators conjugate to @xmath22 and @xmath23 . it is convenient to work in a fock basis @xmath24 where @xmath20 and @xmath21 are diagonal , with @xmath25 the number of particles and @xmath26 ranging between 1 and @xmath25 . to simplify the notation only one particle type is included explicitly . the state @xmath27 is given by an expansion @xmath28_n\,[d^2k_\perp]_n\ , \psi_n(x,\bbox{k}_\perp ) |n : p^+=xp^+,\bbox{p}_\perp = x\bbox{p}_\perp+\bbox{k}_\perp\rangle\,,\ ] ] with @xmath29_n=4\pi\delta(1-\sum_{i=1}^nx_i ) \prod_{i=1}^n\frac{dx_i}{4\pi\sqrt{x_i}}\,,\;\;\ ; [ d^2k_\perp]_n=4\pi^2\delta(\sum_{i=1}^n\bbox{k}_{\perp i } ) \prod_{i=1}^n\frac{d^2k_{\perp i}}{4\pi^2}\,,\ ] ] @xmath30 the total light - cone momentum , and @xmath31 interpreted as the wave function of the contribution from states with @xmath25 particles . the solution of ( [ eq : eigenprob ] ) in principle yields these wave functions . the anomalous moment @xmath0 is computed from the standard form factor @xmath32 at zero momentum transfer : @xmath33 in the standard light - cone frame@xcite where @xmath34 the form factor can be computed from the spin - flip matrix element of the plus component of the current : @xmath35 brodsky and drell@xcite have given a reduction of this matrix element to a convenient form that depends directly on the wave functions . from this we have @xmath36_n\,[d^2k_\perp]_n\ , \psi_{n\uparrow}^*(x,\bbox{k}_\perp ) \sum_{i\neq j}x_i\frac{\partial}{\partial k_{1i } } \psi_{n\downarrow}(x,\bbox{k}_\perp)\,,\ ] ] where @xmath37 is the fractional charge of the struck constituent . up to this point , we have used formally exact expressions . a key approximation to be made is the truncation of all sums to a finite number of particles . the result is the light - cone equivalent of the tamm - dancoff approximation@xcite . the eigenvalue problem becomes a finite set of equations that are in principle solvable . however , the truncation has many consequences for the renormalization of the theory@xcite and for comparisons to feynman perturbation theory@xcite . some of these consequences are discussed in sec . [ sec : renormalization ] . in addition , qed requires regularization and renormalization . to regularize it , we use a cutoff on the invariant mass of the allowed fock states@xcite @xmath38 this limits the relative transverse momentum @xmath39 of each constituent and keeps the longitudinal momentum away from zero . the latter aspect is important for control of spurious infrared singularities , which are discussed in [ sec : infrared ] . an additional cutoff , that limits the change in invariant mass across any matrix element of the hamiltonian@xcite , could be considered . when only states with at most one photon and no pairs are retained , and instantaneous interactions are neglected , brodsky and drell@xcite have shown that eq . ( [ eq : aelc ] ) reduces to @xmath40 ^ 2}\,,\ ] ] which in the limit of @xmath41 becomes @xmath42 because the instantaneous interactions are higher order in @xmath2 , this is the leading perturbative result . the integrals involved can all be done analytically even for finite cutoff , although the final form is not instructive . for @xmath43 , the resulting formula yields the standard schwinger@xcite contribution of @xmath44 at infinite cutoff . in general this provides a point of comparison for numerical calculations with one or more photons . the inclusion of the dependence on the photon mass in the analytic result is crucial for comparison with numerical results calculated with nonzero @xmath45 because the mass dependence is quite strong , as can be seen in fig . [ fig : aevsmgamma ] . = the most systematic approach to discretization of the eigenvalue problem is the method originally suggested by pauli and brodsky@xcite , discretized light - cone quantization ( dlcq ) . in essence it is the replacement of integrals by trapezoidal approximations , with equally - spaced intervals in the longitudinal and transverse momenta @xmath46 the length scales @xmath47 and @xmath48 determine the resolution of the calculation . because the plus component of momentum is always positive , the limit @xmath49 can be exchanged for a limit in terms of the integer _ resolution_@xcite @xmath50 the combination of momentum components that defines @xmath51 is then independent of @xmath47 . the longitudinal momentum fractions @xmath52 become ratios of integers @xmath53 . because the @xmath54 are all positive , dlcq automatically limits the number of particles to no more than @xmath55 . the integers @xmath56 and @xmath57 range between limits associated with some maximum integer @xmath58 fixed by the invariant - mass cutoff . a finite matrix problem is then obtained without an explicit tamm - dancoff truncation ; however , this number of particles is much too large in practice for numerical treatments of three - dimensional theories . we use antiperiodic boundary conditions for the fermions and periodic boundary conditions for the photons . these restrict the integers @xmath25 associated with longitudinal momenta to being odd for fermions and even for photons . the description of the dressed electron state must then use odd values of @xmath55 . in most applications , dlcq is introduced at the level of second quantization . this can yield a compact expression of the eigenvalue problem . recently , a transformation of the dlcq hamiltonian to a gaussian basis has been suggested@xcite ; however , the steps for renormalization in that basis have not been worked out . the application of dlcq to qed is summarized in @xcite , which we use as a starting point . this includes use of light - cone gauge @xcite , with @xmath59 . modifications of this gauge choice due to zero modes@xcite are discussed in sec . [ sec : finitecorrections ] . = the fundamental interactions of light - cone qed are illustrated in fig . [ fig : qedgraphs ] . for the calculation reported here we do not include any pair production processes . the instantaneous photon interactions are then completely excluded because each fock state has only one fermion . we also exclude the fourth diagram ( and its conjugate ) to decouple two - photon states from the bare electron state ; this simplifies the calculation and limits the role of two - photon states to that of providing the basis for inclusion of crossed - photon graphs . there is also a technical modification of the interaction associated with the third diagram of fig . [ fig : qedgraphs ] which is discussed in sec . [ sec : infrared ] . after the exclusions have been made , the light - cone hamiltonian becomes @xmath60 with @xmath61 . the discrete form of the spin-@xmath62 eigenstate @xmath63 is @xmath64 where @xmath65 . according to the eigenvalue equation @xmath66 the amplitudes @xmath67 must satisfy the following ( discretized ) integral equations : @xmath68 @xmath69\right ) \psi_{1s}(\underline{n}',\underline{m}';s_1,\lambda_1 ) } \hspace{0.5 in } \nonumber \\ & & = \frac{k}{4\pi l_\perp^2 } \sum_{\underline{n},\underline{m } } \delta_{\underline{n}+\underline{m},\underline{k } } \frac{1}{\sqrt{m'm } } \left\{e_0 ^ 2\delta_{\lambda_1,-2s_1}\frac{1}{k } \psi_{1s}(\underline{n},\underline{m};s_1,-2s_1 ) \right . \nonumber \\ & & \hspace{1 in } \left .- e_1(\underline{n}')e_1(\underline{n})\delta_{\lambda_1,2s_1 } \frac{\theta(n - m')}{n - m ' } \psi_{1s}(\underline{n},\underline{m};s_1,2s_1)\right\ } \nonumber \\ & & + e_0\frac{km_e}{2\sqrt{\pi}l_\perp}\frac{1}{\sqrt{m ' } } \left(\frac{1}{n'}-\frac{1}{k}\right ) \delta_{\lambda_1,2s}\delta_{s_1,-s}\psi_{0s } \label{eq : psi1s } \\ & & + e_0\frac{k\sqrt{\pi}}{\sqrt{2}l_\perp^2}\frac{1}{\sqrt{m ' } } \left\{\bbox{\epsilon}_{\perp\,,\lambda_1}^*\cdot \left(\frac{\bbox{m}'_\perp}{m'}-\frac{\bbox{n}'_\perp}{n'}\right ) \delta_{\lambda_1,2s}\delta_{s_1,s } + \bbox{\epsilon}_{\perp\,,\lambda_1}^*\cdot \frac{\bbox{m}'_\perp}{m'}\delta_{\lambda_1,-2s}\delta_{s_1,s}\right\ } \psi_{0s } \nonumber \\ & & + e_1(\underline{n } ' ) \frac{km_e}{2\sqrt{\pi}l_\perp}\sum_{\underline{n},\underline{m } } \delta_{\underline{n}+\underline{m}+\underline{m'},\underline{k } } \frac{1}{\sqrt{2m}}\left(\frac{1}{n}-\frac{1}{n+m}\right ) \nonumber \\ & & \hspace{1 in } \times \left\{\psi_{2s}(\underline{n},\underline{m},\underline{m}';-s_1,2s_1,\lambda_1 ) + \psi_{2s}(\underline{n},\underline{m}',\underline{m } ; -s_1,\lambda_1,2s_1)\right\ } \nonumber \\ & & + e_1(\underline{n } ' ) \frac{k\sqrt{\pi}}{\sqrt{2}l_\perp^2}\sum_{\underline{n},\underline{m } } \delta_{\underline{n}+\underline{m}+\underline{m'},\underline{k } } \frac{1}{\sqrt{2m}}\left\ { \bbox{\epsilon}_{\perp\ , , 2s_1}\cdot \left(\frac{\bbox{m}_\perp}{m}-\frac{\bbox{n}_\perp}{n}\right ) \psi_{2s}(\underline{n},\underline{m},\underline{m}';s_1,2s_1,\lambda_1 ) \right . \nonumber \\ & & \hspace{1 in } + \bbox{\epsilon}_{\perp\ , , 2s_1}\cdot \left(\frac{\bbox{m}_\perp}{m}-\frac{\bbox{n}_\perp}{n}\right ) \psi_{2s}(\underline{n},\underline{m}',\underline{m};s_1,\lambda_1,2s_1 ) \nonumber \\ & & \hspace{1 in } + \bbox{\epsilon}_{\perp\ , , -2s_1}\cdot \left(\frac{\bbox{m}_\perp}{m}-\frac{\bbox{n}'_\perp}{n'}\right ) \psi_{2s}(\underline{n},\underline{m},\underline{m}';s_1,-2s_1,\lambda_1 ) \nonumber \\ & & \hspace{1 in } \left . + \bbox{\epsilon}_{\perp\ , , -2s_1}\cdot \left(\frac{\bbox{m}_\perp}{m}-\frac{\bbox{n}'_\perp}{n'}\right ) \psi_{2s}(\underline{n},\underline{m}',\underline{m};s_1,\lambda_1,-2s_1 ) \right\}\ , , \nonumber\end{aligned}\ ] ] and @xmath70\right ) \psi_{2s}(\underline{n}',\underline{m}'_1,\underline{m}'_2 ; s_1,\lambda_1,\lambda_2 ) \nonumber \\ & & = \frac{km_e}{2\sqrt{2\pi}l_\perp } \left\{\delta_{\lambda_2,-2s_1 } \frac{e_1(\underline{n}'+\underline{m}'_2)}{\sqrt{m'_2 } } \left(\frac{1}{n'}-\frac{1}{n'+m'_2}\right ) \psi_{1s}(\underline{n}'+\underline{m}'_2,\underline{m}'_1;-s_1,\lambda_1 ) \right . \nonumber \\ & & \hspace{0.5 in } \left . + \delta_{\lambda_1,-2s_1 } \frac{e_1(\underline{n}'+\underline{m}'_1)}{\sqrt{m'_1 } } \left(\frac{1}{n'}-\frac{1}{n'+m'_1}\right ) \psi_{1s}(\underline{n}'+\underline{m}'_1,\underline{m}'_2;-s_1,\lambda_2 ) \right\ } \nonumber \\ & & \frac{k\sqrt{\pi}}{2l_\perp^2}\left\ { \delta_{\lambda_2,2s_1 } \frac{e_1(\underline{n}'+\underline{m}'_2)}{\sqrt{m_2 } } \bbox{\epsilon}_{\perp\ , , \lambda_2}^*\cdot \left(\frac{\bbox{m}'_{\perp 2}}{m'_2}-\frac{\bbox{n}'_\perp}{n'}\right ) \psi_{1s}(\underline{n}'+\underline{m}'_2,\underline{m}'_1;s_1,\lambda_1 ) \right . \label{eq : psi2s } \\ & & \hspace{0.5 in } + \delta_{\lambda_1,2s_1 } \frac{e_1(\underline{n}'+\underline{m}'_1)}{\sqrt{m_1 } } \bbox{\epsilon}_{\perp\ , , \lambda_1}^*\cdot \left(\frac{\bbox{m}'_{\perp 1}}{m'_1}-\frac{\bbox{n}'_\perp}{n'}\right ) \psi_{1s}(\underline{n}'+\underline{m}'_1,\underline{m}'_2;s_1,\lambda_2 ) \nonumber \\ & & \hspace{0.5 in } + \delta_{\lambda_2,-2s_1 } \frac{e_1(\underline{n}'+\underline{m}'_2)}{\sqrt{m_2 } } \bbox{\epsilon}_{\perp\ , , \lambda_2}^*\cdot \left(\frac{\bbox{m}'_{\perp 2}}{m'_2 } -\frac{\bbox{n}'_\perp+\bbox{m}'_{\perp 2}}{n'+m'_2}\right ) \psi_{1s}(\underline{n}'+\underline{m}'_2,\underline{m}'_1;s_1,\lambda_1 ) \nonumber \\ & & \hspace{0.5 in } \left . + \delta_{\lambda_1,-2s_1 } \frac{e_1(\underline{n}'+\underline{m}'_1)}{\sqrt{m_1 } } \bbox{\epsilon}_{\perp\ , , \lambda_1}^*\cdot \left(\frac{\bbox{m}'_{\perp 1}}{m'_1 } -\frac{\bbox{n}'_\perp+\bbox{m}'_{\perp 1}}{n'+m'_1}\right ) \psi_{1s}(\underline{n}'+\underline{m}'_1,\underline{m}'_2;s_1,\lambda_2 ) \right\}\ , . \nonumber\end{aligned}\ ] ] in anticipation of the discussion of renormalization in sec . [ sec : renormalization ] , bare masses @xmath71 and @xmath72 and bare couplings @xmath73 and @xmath74 have been introduced . these equations are solved numerically , with the first step being the use of ( [ eq : psi2s ] ) to eliminate @xmath75 from ( [ eq : psi1s ] ) . once a solution is obtained for one value of total spin @xmath62 , the solution for the opposite spin can be computed directly from @xmath76 and @xmath77 these follow from the symmetries of the integral equations . other symmetries lead to relationships between amplitude components , which can be summarized as follows : @xmath78 and @xmath79 where the functions @xmath80 , @xmath81 , @xmath82 , and @xmath83 are real . the problem can then be reduced to a smaller matrix problem for these real functions . for @xmath84 we store @xmath85 , @xmath86 , @xmath87 , and @xmath88 . for @xmath89 we store @xmath90 , @xmath91 , @xmath92 , @xmath93 , and @xmath94 . the use of symmetry reduces the matrix storage requirement by a factor of 8 . the hermitian matrix of the original eigenvalue equation ( [ eq : psi1s ] ) can be expressed as a real symmetric matrix in the reduced equation by using a two - component representation of complex arithmetic : @xmath95 and @xmath96 the leading perturbative result@xcite is recovered by keeping only @xmath97 terms on the right - hand side of ( [ eq : psi1s ] ) . this equation can then be immediately solved for @xmath98 , which can be used to form a discrete approximation to ( [ eq : aeintegrals ] ) . the approximation includes a finite difference approximation to the derivatives that appear in ( [ eq : aelc ] ) and therefore is not simply a trapezoidal approximation to ( [ eq : aeintegrals ] ) . results obtained with ordinary dlcq show an irregular dependence on the numerical parameters @xmath55 and @xmath58 , which interferes with extrapolation to infinite resolution . the causes of the irregularities have been determined to be the numerical approximation of the derivative in the formula ( [ eq : aelc ] ) for @xmath0 and boundary effects in the numerical integrations . the error in the derivative can be controlled by choosing @xmath99 and @xmath100 . the bound on @xmath55 is consistent with the resolution needed to resolve the one - photon peak in the @xmath0 integrand when @xmath101 , which is the photon mass we use . smaller values of @xmath45 shift the peak to smaller photon momenta and would increase the lower bound on @xmath55 . the shape of the integrand for various values of @xmath45 is illustrated in fig . [ fig : onephoton ] . the mass sensitivity of the numerical convergence rate is shown in table [ tab : onephoton ] . = rddddd & + + @xmath55 & 0 & 0.001 & 0.01 & 0.1 & 0.3162 + 11 & 0.8182 & 0.8182 & 0.8173 & 0.7455 & 0.5041 + 21 & 0.9048 & 0.9047 & 0.9025 & 0.7708 & 0.5079 + 41 & 0.9512 & 0.9512 & 0.9461 & 0.7728 & 0.5090 + 81 & 0.9753 & 0.9752 & 0.9648 & 0.7731 & 0.5093 + 161 & 0.9876 & 0.9873 & 0.9699 & 0.7732 & 0.5094 + 321 & 0.9938 & 0.9933 & 0.9703 & 0.7732 & 0.5094 + 641 & 0.9969 & 0.9959 & 0.9703 & 0.7732 & 0.5094 + 1281 & 0.9984 & 0.9968 & 0.9703 & 0.7732 & 0.5094 + 2561 & 0.9992 & 0.9969 & 0.9703 & 0.7732 & 0.5094 + @xmath102 & 1.0000 & 0.9969 & 0.9703 & 0.7732 & 0.5094 + the integration boundary effects are more difficult to control . these effects arise from use of the dlcq grid which is incommensurate with the integration domain . at the boundaries , the trapezoidal rule misses contributions beyond the last grid point ; this error is not a smooth function of the grid spacing . to overcome this error , one can replace the trapezoidal rule by open - closed newton cotes formulas tailored specifically to the boundary @xcite . grid points near the boundary are then associated with unequal integration weights . the unequal weights must be taken into account in normalization sums and symmetrization of the hamiltonian matrix , but this is easily done . one can even consider use of simpson s rule , although this does not appear useful in the anomalous moment calculation . the improvement brought by these weighting methods can be dramatic , as shown in ref . here we are interested in ultraviolet divergences associated with large @xmath39 . electron self - energy contributions , which are divergent , shift the mass and , through wave function renormalization , change the coupling . these induce @xmath103 and @xmath104 dependencies in the eigenvalues . in the discrete truncated problem these effects depend on the fock sector considered@xcite . for example , an electron in a fock state for which a transition to a state with more photons is not allowed , perhaps due to truncation in photon number , will not experience any self - energy corrections . if one additional photon is allowed , but not instantaneous interactions , only single loops can occur . if two or more additional photons can appear , then an infinite number of overlapping loops can contribute to self - energy corrections , a truly nonperturbative situation . in each case , the leading divergence is removed by introduction of counterterms associated with bare masses that are sector dependent and momentum dependent@xcite . to be specific , consider the case where there are at most two photons and only one electron . the fock - state expansion can be written schematically as @xmath105 here @xmath106 and @xmath107 are column vectors that contain the amplitudes for individual fock states with one and two photons , respectively . the eigenvalue problem ( [ eq : eigenprob ] ) becomes a coupled set of three integral equations ( [ eq : psi0s ] ) , ( [ eq : psi1s ] ) , and ( [ eq : psi2s ] ) , which we write more compactly as @xmath108 where @xmath71 is the bare electron mass and the vectors @xmath109 and the tensors @xmath110 are integral operators obtained from @xmath51 . we now require that @xmath71 be such that @xmath111 is an eigenvalue . the second and third equations can be solved for @xmath112 and @xmath113 . then the first equation yields @xmath71 . normalization of @xmath27 fixes the value of @xmath114 . suppose now that this two - photon problem is embedded in some larger problem where one needs to know the bare mass of an electron in a fock state that can couple to fock states obtained by adding at most two photons . the same set of equations can be applied , with all constituents in the lowest fock state , other than the electron , acting as spectators . one need only replace @xmath115 in ( [ eq : integraleqns ] ) by @xmath116 and @xmath111 by @xmath117 , with @xmath118 and @xmath119 the longitudinal momentum fraction and transverse momentum of the initial electron . notice that @xmath71 is now a function of @xmath118 and @xmath119 . this can be generalized to cases with more photons , and reduced to the case of only one contributing photon . thus one obtains a mechanism for a sector - dependent , momentum - dependent mass renormalization that is used from the top @xmath25-photon sector down to the bare electron state @xmath120 . the last step automatically includes the solution of the full eigenvalue problem for the dressed electron state . for the one - photon case embedded in the two - photon problem we have @xmath121 the second photon is a spectator . the coupling to @xmath107 then induces the one - loop self - energy correction with this spectator present . the explicit form is obtained from ( [ eq : psi1s ] ) and ( [ eq : psi2s ] ) with @xmath122 and with any interaction involving the spectator dropped . . [ eq : psi2s ] can then be solved for @xmath107 and the result substituted into the modified ( [ eq : psi1s ] ) to obtain . @xmath123 -\frac{k}{n}\left[m_e^2+(\bbox{n}_\perp \pi / l_\perp)^2\right ] -\frac{k}{m}\left[m_\gamma^2+(\bbox{m}_\perp \pi / l_\perp)^2\right ] } \nonumber \\ & & \hspace{0.25 in } + e_1 ^ 2(\underline{n}')\frac{k^2 \pi}{4 l_\perp^4 } \sum_{\underline{n},\underline{m } } \frac{\delta_{\underline{n}+\underline{m},\underline{n}'}/m \left\{\left(\bbox{m}_\perp / m-\bbox{n}_\perp / n\right)^2 + \left(\bbox{m}_\perp / m-\bbox{n}'_\perp / n'\right)^2 \right\}\psi_{1s}(\underline{n}',\underline{m}';s_1,\lambda_1 ) } { \frac{k}{n'}\left[m_e^2+(\bbox{n}'_\perp \pi / l_\perp)^2\right ] -\frac{k}{n}\left[m_e^2+(\bbox{n}_\perp \pi / l_\perp)^2\right ] -\frac{k}{m}\left[m_\gamma^2+(\bbox{m}_\perp \pi / l_\perp)^2\right]}\ , . \nonumber\end{aligned}\ ] ] the kronecker deltas from helicity conservation have been used to simplify the result , and only terms in which the second photon is a spectator have been kept . rearrangement of the coefficient of @xmath98 yields @xmath124 -\frac{k}{n}\left[m_e^2+(\bbox{n}_\perp \pi / l_\perp)^2\right ] -\frac{k}{m}\left[m_\gamma^2+(\bbox{m}_\perp \pi / l_\perp)^2\right ] } \nonumber\end{aligned}\ ] ] as the one - loop mass . if electron - positron pairs are included , the photon mass is renormalized and must be treated in the analogous fashion . in general , the two mass renormalizations are coupled , and must be carried out simultaneously . all of the steps in mass renormalization depend on knowing all couplings . this information is actually not immediately available because the couplings are to be renormalized . the bare coupling for the electron - photon three - point vertex depends on the initial and final momenta of the electron and on the sectors between which the coupling acts@xcite . the momentum dependence is present because the amount of momentum available constrains the extent to which higher order corrections can contribute . similarly , the sector dependence makes itself felt when the number of additional particles in higher - order corrections is restricted . we fix these bare coupling functions by matching photon absorption amplitudes to the fundamental three - point vertex . the amplitudes are computed from the numerical eigenfunction of the light - cone hamiltonian . therefore , the coupling renormalization conditions and the mass eigenvalue problem form a coupled set of equations that are solved iteratively . when vacuum polarization is absent , the bare coupling @xmath73 is related to the physical coupling @xmath125 by @xmath126 where @xmath127 is the initial electron momentum and @xmath128 the final momentum . the renormalization functions @xmath129 and @xmath130 are generalizations of the usual constants@xcite . the wave function renormalization function @xmath131 is easily computed since it is the probability of the bare electron fock state in the dressed electron state . in the earlier notation of eq . ( [ eq : schematicexpansion ] ) , we have @xmath132 where @xmath133 is the light - cone momentum of the dressed electron . the amplitude @xmath114 must be computed in a basis where only allowed particles appear . for example , if the vertex is the photon absorption process , @xmath134 must be computed with one less photon in the basis than in the basis used for @xmath135 . from this example one can see that the tamm - dancoff approximation has destroyed the usual ward identity . the function @xmath136 can be fixed by considering the transition amplitude @xmath137 for photon absorption by an electron at zero photon momentum . the proper part of this amplitude , meaning that without self - energy corrections to the legs , is required to be proportional to the elementary three - point no - flip vertex @xmath138 when @xmath139 : @xmath140 in the limit , only @xmath141 dependence can remain .. ] numerically the limit can be taken by using a photon with momentum @xmath142 ; in the dlcq limit of @xmath143 and @xmath144 , this momentum becomes zero . of course , if this particular state does not satisfy the cutoff , a state with slightly larger longitudinal momentum must be used instead . the full transition amplitude can be computed from solutions to the eigenvalue problem ( [ eq : eigenprob ] ) . let @xmath145 be the free light - cone hamiltonian with physical masses . the eigenstates of @xmath145 are then the asymptotic states of the electron and photon . the transition is driven by the interaction @xmath146 . define resolvents for the free and full hamiltonians as @xmath147 with @xmath62 the square of the center - of - mass energy . the @xmath148 matrix can be formally expressed in terms of these as @xmath149 when sandwiched between the initial and final states , this yields @xmath150 where the @xmath151 are eigenstates of @xmath51 with eigenvalues @xmath152 and bare - electron amplitudes @xmath153 . in the limit that @xmath62 becomes @xmath154 , we obtain @xmath155 in which @xmath156 is the dressed electron state and @xmath157 . the connection between @xmath137 and @xmath158 is made by considering the matrix element of @xmath159 . we have @xmath160 = the factors on the right are illustrated in fig . [ fig : propervertex ] . the second factor contains no intermediate @xmath161 states and the initial photon is absorbed before @xmath162 appears as an intermediate state . in the third factor , the initial photon remains a spectator throughout . the sum runs over all possible combinations of these forms and yields @xmath163 where @xmath164 is the propagator for the electron in the presence of the initial photon as a spectator . in the limit where @xmath62 and @xmath165 approach @xmath154 , we obtain @xmath166 this then reduces to an expression for @xmath158 @xmath167 thus the solution of the eigenvalue problem for only one state can be used to compute @xmath136 . full diagonalization of @xmath51 is not needed . because @xmath136 is needed in the construction of @xmath51 , the eigenvalue problem and the renormalization conditions must be solved simultaneously . this leads to an iterative procedure that begins with an initial guess for the bare coupling functions . one then computes bare masses and new bare couplings . the process is repeated until convergence is attained . this must be done from the top sector down ; the bare masses in any one sector and the bare couplings between any two depend only on the sectors above , the ones with more photons . the structure of the hamiltonian matrix can then be determined once and for all at these levels and then used in the determination of the structure at the levels further below . when the fock basis is limited to no more than one photon , and instantaneous interactions are neglected , the renormalization conditions are quite simple . we have from ( [ eq : tfiformula ] ) @xmath168 and from this , with ( [ eq : couplingrenormcond ] ) and ( [ eq : tproperformula ] ) , @xmath169 where the last equality follows from the unavailability of any state that can correct the initial electron line when a photon spectator is present . the bare charge is then given by @xmath170 the subscript of 1 corresponds to use in couplings between one and two - photon states . we now consider the solution of the problem in the case of a basis with no more than two photons . ( [ eq : e1 ] ) provides the solution for the bare coupling between one and two - photon states and , through spectator dependence of @xmath114 , makes @xmath74 a function of the final electron momentum . we then need to consider the bare coupling between the bare electron and the one - photon states . on substitution of eqs . ( [ eq : couplingrenormcond ] ) and ( [ eq : tproperformula ] ) , eq . ( [ eq : e0 ] ) becomes @xmath171 this is a nonlinear equation for @xmath73 because @xmath137 has a complicated dependence on this bare charge . to make this dependence explicit , we first use the fact that @xmath156 is an eigenstate of @xmath172 to reduce ( [ eq : tfiformula ] ) to the form @xmath173 \psi_0\psi_1^*(1 - 2/k,\hat{x}\pi / l_\perp)\,.\ ] ] the amplitude @xmath174 satisfies the middle equation of ( [ eq : integraleqns ] ) , which can be written as @xmath175\cdot\bbox{\psi}_1 = m_e^2\bbox{\psi}_1\,,\ ] ] where @xmath176 is an effective interaction obtained by integrating out the @xmath75 amplitude and the @xmath177 term is the finite instantaneous fermion interaction . the coupling to @xmath114 contains @xmath73 as a simple factor . we extract this in the following definitions : @xmath178 the scaled amplitude @xmath179 can then be obtained as a formal solution to eq . ( [ eq : psi1eq ] ) that shows all @xmath73 dependence is independent of @xmath73 , as can be seen from ( [ eq : psi1s ] ) and ( [ eq : psi2s ] ) . ] @xmath180 with @xmath181 the amplitude for two - photon states is then given by @xmath182 where @xmath183 is a rectangular matrix independent of @xmath73 . the discrete normalization condition is @xmath184\,.\ ] ] this yields @xmath185 which can be used with ( [ eq : tfiexplicit ] ) and ( [ eq : e0new ] ) to obtain @xmath186}\,,\ ] ] where @xmath187 is independent of @xmath73 . the phases of @xmath188 and @xmath189 are such that the right - hand side is real , as it must be . the remaining implicit dependence on @xmath73 is in @xmath190 , which is given by ( [ eq : psi1prime ] ) , and in @xmath191 , with @xmath183 independent of @xmath73 . notice that @xmath192 and @xmath193 are independent of @xmath73 and need to be computed only once . the equation for @xmath73 is best solved iteratively after it is squared to eliminate the square root on the right hand side . a real solution exists only for a finite range of the physical coupling @xmath125 . this is an artifact of the tamm dancoff truncation and the consequent failure of the ward identity . the value of the critical coupling @xmath194 , the upper limit of the allowed range , can be found by studying the @xmath195 limit . in this limit we find @xmath196 ^ 2 } { 4\pi z_{2i}v_{fi}^{\prime\,2 } ( |\bbox{\psi}_1^\prime|^2+|\bbox{\psi}_2^\prime|^2 ) } \left[m_e^2-\frac{m_e^2+\pi^2/l_\perp^2}{1 - 2/k } -\frac{m_\gamma^2+\pi^2/l_\perp^2}{2/k}\right]^2\,,\ ] ] with @xmath197 calculable as @xmath198 . values of the critical coupling are plotted in fig . [ fig : critcoupling ] . the change from a basis with no more than one photon to a basis with no more than two is quite small . to stay within the limit imposed by this result we will use a physical coupling of @xmath199 . = a nonzero photon mass @xmath45 is used to eliminate the usual infrared singularities . because the calculation deals with a charged system , there would otherwise be considerable difficulty with soft photons@xcite . on the light cone , there are other singularities not removed by the photon mass . they are associated with contributions that involve zero longitudinal momentum . the fundamental four - point vertices can be infrared singular , in the limit of zero longitudinal momentum for the instantaneous fermion . they must be allowed to cancel against iterations of the three - point vertices which are also singular . this constrains the bare couplings in the four - point vertices to forms derived from the three - point coupling @xmath73 . the pairs of diagrams are shown in fig . [ fig : infraredcancel ] . the first pair does not actually involve a singularity ; however , we do match the four - point coupling to the iterated three - point coupling . the second requires basis states with two photons , which are available in the calculation . the third pair requires the presence of electron - positron pairs in the basis , or an effective interaction in the hamiltonian . neither is included at present , and therefore this piece of the instantaneous interaction must also be excluded from the hamiltonian . detection of @xmath200 in the instantaneous interaction can be easily done to exclude this graph . = other infrared singularities are associated with the emission and absorption of real photons with longitudinal momentum near zero@xcite . in perturbative calculations these are regulated by the mandelstam leibbrandt prescription@xcite . viewed in @xmath15-ordered perturbation theory , each intermediate state contributes a denominator in which the light - cone energy @xmath201 of the photon becomes large , and each vertex can contain a factor of @xmath202 . if @xmath203 is separately regulated , with some lower cutoff @xmath204 , graphs with multiple photons will contribute powers of @xmath205 or even @xmath206 . for a tamm - dancoff approximation to a charged system , these can not be expected to cancel . the choice of the invariant mass cutoff ( [ eq : cutoff ] ) instead couples the regulation of small @xmath203 to that of large @xmath39 . the combination prevents the small @xmath203 region of integration from making large contributions except in cases where there are already ultraviolet transverse divergences . these spurious infrared infinities are then handled by the mass and coupling renormalization discussed in this section . there remains a logarithmic divergence associated with four - point graphs of the sort illustrated in fig . [ fig : fourpointgraphs](a ) and ( b ) . if all graphs of this order are included in a perturbative calculation , the logarithms cancel . however , the tamm - dancoff truncation of the present calculation excludes some graphs , such as the one shown in fig . [ fig : fourpointgraphs](c ) , and the cancellation can no longer take place . in a nonperturbative calculation one must include the equivalent of diagrams with an arbitrary number of interlocked loops , such as fig . [ fig : fourpointgraphs](b ) , which are also logarithmically divergent . the needed counterterm is of the form @xmath207 but can not be found analytically without summing all the infinite chains of interlocked loops . = one way to approach the construction of @xmath208 is to fit compton amplitudes to data@xcite . this will require development of techniques to describe scattering processes within dlcq . a generalization of earlier work@xcite on the inversion of the full greens function may be useful as a means for computing the @xmath148 matrix and thus scattering amplitudes . we do not consider this further here . the results presented in sec . [ sec : results ] do still contain the @xmath104 divergence . as applied to qed , dlcq requires the use of periodic boundary conditions for the photon field . this is because photons couple to fermion bilinears , which are automatically periodic , even if the preferred antiperiodic boundary conditions are used for the fermions . for fields periodic in the longitudinal direction @xmath22 , there are contributions from zero modes@xcite , modes independent of @xmath22 that correspond to zero longitudinal momentum . as shown by pauli and kalloniatis@xcite , these modes prevent the choice of ordinary light - cone gauge because the zero - mode piece of @xmath209 can not be gauged away . instead this piece must satisfy a constraint equation . in fact , careful application of dlcq to most bosonic theories will result in constraint equations that relate the zero - mode contribution to the normal - mode operators in a nonlinear , nontrivial way . for qed there are zero modes in all the components of the photon field , for which constraint equations must be solved . what is more , the constraint equation for the dependent piece of the fermion field , which is easily solved in light - cone gauge in the continuum , becomes coupled to the zero - mode constraint equations . a formulation of the coupled system of constraints has been given by kalloniatis and robertson@xcite . extension of these constraints to include a nonzero photon mass is straightforward . the constraint equations are difficult to solve , even in simpler theories@xcite . this is partly because they couple states with different @xmath210 and require study of convergence as a function of some @xmath210 cutoff . the difficulty is also due to the need for an ultraviolet cutoff and renormalization of masses and couplings . because the renormalization is formulated in terms of solutions to the mass eigenvalue problem and because the hamiltonian can not be formed until the zero - mode contribution is known , the problem expands to a very large nonlinear system of simultaneous equations . as a result of these difficulties , the calculations discussed here do not include zero modes . however , some progress has been made recently by kalloniatis@xcite in the solution of constraint equations for su(2 ) yang - mills in @xmath12 dimensions coupled to massive adjoint scalars . for theories such as qed where symmetry breaking effects are not expected , solution of constraint equations may not be necessary . one can instead treat the end - point behavior of photon amplitudes in a manner similar to that of the `` ladder relations '' studied by antonuccio and dalley@xcite . behavior of amplitudes at small longitudinal momentum , as extracted from the integral equations , can be used to construct effective interactions that include zero modes to leading order in @xmath211 . this is equivalent to the approach used in@xcite where the behavior of the exchange kernel was studied in a scalar theory to determine the effective interaction@xcite . to keep zero - mode terms to higher order in @xmath211 would actually be inconsistent with dlcq s neglect of higher order non - zero - mode terms . in the work of ref . @xcite inclusion of the zero - mode contributions @xmath212 did improve convergence . the whole issue may actually be moot when the invariant mass cutoff ( [ eq : cutoff ] ) is used . this cutoff explicitly excludes contributions from states with zero longitudinal momentum . the meaning of this exclusion for nondynamical fields is unclear . the calculations that showed zero modes to be useful for convergence@xcite did not employ the invariant mass cutoff . new calculations need to be carried out specifically to study the effect of cutoff choice on the importance of zero modes . the use of light - cone coordinates , combined with the tamm - dancoff truncation in particle number and the invariant mass cutoff , explicitly break symmetries of the theory@xcite . in particular , rotational symmetry about the transverse axes is broken because the associated operators involve the interaction and therefore change particle number . the change in particle number can not be accommodated in field theory without allowing an infinite number of particles . restoration of such symmetries can be accomplished by the addition of finite counter - terms to the hamiltonian@xcite including adjustment of the `` vertex mass , '' which appears in the spin - flip vertex , relative to the `` kinetic mass''@xcite . the ambiguities associated with the infinite counterterms allow such finite terms to exist@xcite . restoration of symmetries is then viewed as a source of conditions by which these finite parts can be determined . in practice , this might involve study of processes@xcite such as compton scattering@xcite or electron - electron scattering . given the tamm - dancoff truncation , an alternative is to view the eigenvalue problem as a few - body problem@xcite for which the correct effective hamiltonian and the generators of translations , rotations , and boosts must satisfy the usual poincar algebra@xcite . the effective operators might be constructed by adding minimal finite corrections to their field - theoretic forms . the finite corrections are determined by the requirement that the poincar algebra be satisfied . for the results presented here , no attempt has been made to include these finite corrections . an accurate dlcq calculation for a basis with at most one photon can be easily done when instantaneous interactions are neglected . the accuracy can be verified directly because the integrals that yield @xmath0 can be performed analytically@xcite . in the limit of infinite cutoff and zero photon mass this reproduces the schwinger@xcite result of @xmath213 . the only coupling renormalization is a trivial wave function renormalization . the dlcq result at various cutoff values is shown in fig . [ fig : results ] for a photon mass of @xmath214 . weighting methods@xcite are a critical part of the calculation . = calculations with a basis that includes at most two photons have been done at five different values of the cutoff for a coupling of @xmath199 . they are also shown in fig . [ fig : results ] . the two - photon contribution adds approximately 40% . this is much larger than the order of magnitude ( @xmath215 or 3% ) that one would expect . it is also opposite in sign to the sommerfield petermann contribution@xcite of @xmath216 to the anomalous moment . we attribute this large difference to the absence of z graphs . the basis sizes involved are on the order of 1 to 4 million , which translates to solution of linear systems with 4000 to 10,000 variables once the two - photon states are integrated out and symmetries of the one - photon states are used . the values obtained from dlcq were extrapolated to @xmath217 and @xmath218 by fits to @xmath219 . exclusion of the last term provided an estimate of the error in the fit , which is reflected in the error bars in fig . [ fig : results ] . the values of @xmath55 ranged from 21 to 31 and those of @xmath58 from 7 to 9 , 10 or 11 , depending on matrix size limitations . the time required for an extrapolated value at a fixed cutoff is roughly 10 hours on a cray x - mp , using less than 32 million words of memory . this seems quite competitive with older lattice methods , where a quenched qcd calculation of heavy - light meson wave functions@xcite required 300 hours on a cm-200@xcite , but does not yet match the effort required with the latest methods@xcite , for which calculation of the b meson magnetic form factor might require 50 hours on a good pc@xcite . the nonperturbative calculation of the anomalous moment of the electron @xmath0 , besides being of intrinsic interest itself , exposes many important issues for nonperturbative calculations within gauge theories which occur in the context of a truncated fock space . these include nonperturbative mass and coupling renormalization , control of spurious infrared singularities , determination of zero - mode contributions , and the construction of finite counterterms which restore symmetries . each of these has been addressed in the preceding sections , and the first two have been incorporated into a dlcq calculation where as many as two photons are included in the basis . we have presented results for @xmath0 computed in a light - cone - gauge fock space truncated to include one bare electron and at most two photons ; _ i.e. _ , up to two photons in flight . the calculational scheme uses an invariant mass cutoff , discretized light - cone quantization ( dlcq ) , a tamm dancoff truncation of the fock space , and a photon mass regulator . we have utilized new weighting methods which greatly improve convergence to the continuum within dlcq . a large renormalized coupling strength @xmath1 is then used to make the nonperturbative effects in the electron anomalous moment from the one - electron , two - photon fock state sector numerically detectable . results are given in fig . [ fig : results ] . the disagreement between these results and what one would expect from perturbation theory at order @xmath3 indicates that the effect of z graphs needs to be included in a systematic way . this can be done as an effective interaction , to avoid expansion of the fock basis to include pair states . the corresponding piece of the instantaneous fermion interaction , as depicted in fig . [ fig : infraredcancel](c ) , must then also be included to maintain an infrared cancellation . new counterterms : one piece of the infinite renormalization is missing in the calculation . as discussed in section [ sec : fourptgraphs ] , it requires a new nonperturbative counterterm for the logarithmic divergences present in diagrams of the type shown in fig . [ fig : fourpointgraphs](a ) and ( b ) . the divergence arises because the tamm dancoff truncation prevents certain cancellations . construction of the counterterm will likely require analysis of scattering processes . zero modes in dlcq : before full consideration of photon zero modes is undertaken , we recommend renewed study of zero modes in a scalar theory where the constraint equation can be solved exactly@xcite . this may show that , when the invariant mass cutoff ( [ eq : cutoff ] ) is used , zero modes do not make a significant numerical contribution . if instead there is an important contribution , it should be computed only to leading order in the numerical resolution , to be consistent with the level of approximation used in the basic dlcq approach . use of symmetries in dlcq renormalization : the restoration of symmetries should then complete construction of the light - cone hamiltonian . one can normalize to specific physical processes@xcite or take an abstract approach based on the algebra of the poincar generators@xcite . higher fock states : once the two - photon calculation is fully under control , the addition of @xmath220 states can be considered . this will require analysis of photon mass and wave function renormalization . many of the complications of the light - cone fock state analysis presented here can be traced to the complexity of sector - dependent renormalization . given such complications the newly developed alternative of pauli villars regularization@xcite may be the preferred approach . within such a scheme , it is also likely that the limitation to a small number of photons can be relaxed . the analysis presented here is the first step in a systematic program to compute physical quantities in gauge theory systematically utilizing a light - cone fock expansion . it will also be interesting to use these methods and the present knowledge of the dressed - electron state in qed in order to systematically construct the neutral positronium state as a composite of a dressed electron and positron . such an analysis can serve as the prototype for systemic nonperturbative construction of colorless bound state hadrons in qcd . this work was supported in part by the minnesota supercomputer institute through grants of computing time and by the department of energy . we thank g. mccartor , r. perry , st . gazek , d. robertson , and a. kalloniatis for helpful discussions .

we present strictly efficient schemes for scalable measurement - based quantum computing using continuous - variable systems : these schemes are based on suitable non - gaussian resource states , ones that can be prepared using interactions of light with matter systems or even purely optically . merely gaussian measurements such as optical homodyning as well as photon counting measurements are required , on individual sites . these schemes overcome limitations posed by gaussian cluster states , which are known not to be universal for quantum computations of unbounded length , unless one is willing to scale the degree of squeezing with the total system size . we establish a framework derived from tensor networks and matrix product states with infinite physical dimension and finite auxiliary dimension general enough to provide a framework for such schemes . since in the discussed schemes the logical encoding is finite - dimensional , tools of error correction are applicable . we also identify some further limitations for any continuous - variable computing scheme from which one can argue that no substantially easier ways of continuous - variable measurement - based computing than the presented one can exist .

1112.2641

to realize a quantum computer in the circuit model , it is crucial to have precise control over each of the carriers of quantum information . in addition to keeping the quantum state of the computer protected from unavoidable noise induced by the environment , it is necessary to implement suitable quantum gates , usually in the form of one and two qubits gates . the latter is of particular difficulty , especially when optical quantum systems are used because their interaction is either weak or merely induced by measurements . the paradigm of measurement - based quantum computing ( mbqc ) as pioneered by raussendorf and briegel @xcite and substantially generalized by gross and eisert @xcite allows to get rid of the necessity of performing unitary operations to implement a quantum circuit . instead , the actual computation is performed by preparing a multipartite entangled state , the resource , in a first step followed by adaptively chosen local measurements on this resource . the important improvement stems from the fact that the resource is universal , i.e. , it can be prepared independently of the algorithm one wants to perform . this means that the presumably difficult step , the one which involves entangling operations , can be performed off - line . this resource - preparation may also be probabilistic as it is possible to wait with the implementation of the algorithm until the resource is available . what is more , individual addressing in the final read out step is anyway also required in the circuit model , so this step does not add any further difficulty to the scheme . quantum computing based on continuous - variable ( cv ) quantum optical systems differs from the more conventional notion based on their discrete analogues in not making use of single photons as the carriers of quantum information @xcite . instead , it relies to a large extent on gaussian states and their manipulation . the notable advantage is that gaussian states are easier to prepare in the laboratory , the corresponding interactions are often stronger and easier to accomplish , and some measurements , e.g. , homodyne detection , can be performed with an efficiency substantially surpassing the one of single - photon measurements . however , quantum information protocols using gaussian states only , gaussian operations , and gaussian measurements suffer from serious limitations as in this setting neither entanglement distillation @xcite nor error correction against gaussian errors @xcite is possible . also , since one can easily efficiently keep track of first and second moments , any gaussian evolution of gaussian states can be efficiently simulated on a classical computer @xcite , clearly ruling out as the possibility of universal quantum computing . mbqc based on gaussian resource states has been extensively discussed in the literature @xcite due to several features : they are a direct generalization of the well known qubit cluster - state to the continuous - variable regime , they allow for universal quantum computing with gaussian measurements and a single non - gaussian one , and they can be prepared with present day experimental techniques , as already demonstrated . however , if they are not formed out of idealized infinitely squeezed states ( ones that are not contained in hilbert space and would require infinite energy in preparation ) but rather from physical states possessing only finite squeezing , they suffer from exponentially decaying localizable entanglement . this limitation , which applies to any gaussian resource state irrespectively of the permitted class of measurements @xcite , implies that full error correction and the machinery of fault - tolerant quantum computing @xcite , in parts yet to be developed for such continuous - variable systems , appears to be necessary even when both state preparation and measurements are perfect . notably , those restrictions only apply when trying to perform a quantum computation of an _ unbounded _ length . for any _ finite _ length , there exists , for any required accuracy , a physical cluster state such that any quantum operation up to this length is possible with this accuracy . because the widely assumed superior power of quantum computers compared to classical ones manifests itself most prominently in the scaling of the runtime with the problem size , the situation of a quantum computation with unbounded length is in the focus of attention in this work . in addition to schemes based on gaussian cluster states , two more classes of schemes relying on cv - quantum optics have been proposed . the first one is based on superpositions of coherent states which are called schrdinger cat states or , when they have very low amplitude , kitten states @xcite . they can be , in approximation , created by subtracting photons from squeezed states and allow for universal quantum computing by passive operations such as beam - splitters and photon - counting detection only . however , they also suffer from quite severe problems : the probabilistic nature of the quantum operations stemming from the use of non - overlapping basis - elements seems to be the most fundamental one , giving rise to significant overheads . the second approach combines the advantageous features of both discrete and continuous - variable optics @xcite . in these schemes , quantum information is carried by qubits and single qubit operations are performed directly on them . two - qubit operations , in contrast , are performed by letting both qubits interact with a strong cv - mode , the qubus . with this article , we pursue two different , but complementary goals : on the one hand , we aim at clarifying the boundary between settings where cv - mbqc is possible and such situations where this is not the case . for this reason , we will develop a general framework capable of describing quantum computation in the measurement - based paradigm , regardless of the dimension of the carriers of quantum information . within this picture , we can identify some serious limitations giving rise to challenges that have to be overcome . on the other hand , we introduce a first strictly efficient scheme relying only on a simple controlled rotation , which can be realized by an atom - light interaction of the jaynes - cummings type or purely optically by the kerr effect , for the creation of the resource . on the measurement side , we require beamsplitters , phase - space displacers , and photon counting measurements . while this scheme is not fleshed out in all detail of its concrete physical implementation , it should be clear that quantum optical implementations of such ideas are conceivable . the present article is organized as follows : first , we discuss what operations are possible with continuous - variable quantum systems and introduce a framework based on matrix product states ( mps ) to describe mbqc in a general setting . after a discussion of the properties which a cv - mbqc scheme needs to posses in order to be called theoretically efficient , we show the problems of achieving those requirements with gaussian measurements on non - gaussian resource states . last , we provide an example for a feasible scheme and discuss in detail how efficient mbqc can be performed in this situation . when performing tasks of quantum information with continuous - variable quantum optics different classes of operations are considered to be of different difficulty . this is also true for measurements where the achievable efficiency greatly differs between the various methods . before continuing with the discussion , we remind the reader of some basic properties of gaussian states and operations while also taking the opportunity to set the notation . we consider a single light mode . the energy eigenvectors of the unit oscillator are denoted by @xmath0 with @xmath1 . the annihilation operator @xmath2 acts on them according to @xmath3 . the commutator relation with its adjoint , the creation operator , is @xmath4=1 $ ] , setting @xmath5 . the eigenvectors of the photon - number operator @xmath6 with @xmath7 are called the number states or fock states . one can now define the canonical operators or quadratures as @xmath8 and for @xmath9 $ ] the family of rotated quadrature operators @xmath10 the latter family of observables is the one that captures homodyne detection @xcite . gaussian unitaries are the ones which can be written as @xmath11 where @xmath12 contains no terms in higher than quadratic order in @xmath13 and @xmath14 ( or , equivalently , in @xmath2 and @xmath15 ) . there are three classes of gaussian single - mode unitary operators into which all gaussian unitary gates can be decomposed . the first ones are corresponding to the application of the displacement operator @xmath16 for @xmath17 . such a transformation is reflected in the heisenberg picture by a map of the form @xmath18 optically , such a transformation can be realized by mixing the mode with a second mode , which is in a strong coherent state , on a beam splitter in the limit of vanishing reflectivity . in the second one are the transformations generated by the clockwise rotation operator @xmath19 , which can be realized by a phase shifter , acting on the canonical operators as @xmath20 , @xmath21 , while in the last one are those generated by the squeezing operator @xmath22,\ ] ] acting as @xmath23 , @xmath24 . the single - mode gaussian unitary operations form a ( non - compact ) group which we denote by @xmath25 . to complete the set of operators , which are necessary to implement arbitrary _ multi - mode _ gaussian operations , we introduce the ( absorption - free ) beam splitter acting on two modes @xmath26 and @xmath27 by @xmath28\ ] ] where @xmath29 and @xmath30 are the transmission and reflection coefficient , respectively . the gaussian operations can be divided into two classes : phase shifters and beam splitters do not change the total number of photons and are called passive operations . to implement a squeezer or a displacer , on the other hand , one does need additional photon sources which makes them more difficult to realize . there are also non - gaussian operations which are within the realm of present experimental techniques : the cross kerr effect gives rise to a non - linear coupling between two modes ; its hamiltonian reads @xmath31 it can be realized by transmitting two modes together through an optically non - linear medium . as non - linear optical effects are weak and absorption needs to be kept low , the achievable effective coupling strength is quite small , i.e. , @xmath32 , where @xmath33 is the interaction time . for the most relevant situation , where one of the modes carries a state with many photons while the other one is in the single photon regime , the hamiltonian ( [ eq : crosskerr ] ) has already been experimentally implemented to perform quantum non - demolition measurement ( qnd ) @xcite . the related hamiltonian @xmath34 for a positive @xmath35 describes the coupling of a two - level atom to a light mode in the dispersive limit of the jaynes - cummings model . in this situation , the effective coupling strength can be increased by placing the mode and the atom inside an optical cavity . a particularly strong interaction can be achieved with a technique known as electromagnetically induced transparency ( eit ) which is routinely used to exchange quantum information between between light modes and atomic vapors and for which also experiments with single atoms exist @xcite . we now turn to the kind of measurements that we will consider feasible for the purposes of this work . the most important gaussian measurement scheme is homodyne dectection which correponds to the observable @xmath36 in eq . ( [ eq : rotq ] ) . it is realized by combining the mode with a strong laser , called the local oscillator , in an interferometer , measuring the intensities on both output ports , and subtracting the results . another important type of gaussian measurement is eight - port homodyning corresponding to a direct measurement of the q - function , i.e.,@xmath37 for @xmath17 , where @xmath38 are the state vectors of the non - orthogonal and overcomplete coherent states @xcite . non - gaussian measurements are in many instances more difficult to realize than gaussian ones , and usually with significantly lower detection efficiencies . the most feasible instance of a non - gaussian measurement reflects a single - photon detector which can only distinguish between the absence of photons and the presence of one or more photons . the corresponding povm - elements are @xmath39 and @xmath40 . a photon - number resolving detector is more challenging to implement but with time - multiplexing @xcite or superconducting nano - wires @xcite , it is possible to perform photon counting for the first few number states with reasonable efficiency . in this section , we introduce a general framework to describe measurement - based quantum computing . a mbqc scheme consists two ingredients : first , a resource state and second a set of possible , or allowed local measurements . we start with a description of quantum wires which are used for single - qudit processing @xcite and discuss their coupling to fully universal resources afterwards . the formalism of matrix product states , which was originally introduced to describe a certain class of entangled one - dimensional many - body states @xcite is extremely useful to capture the essentials of quantum computing in the measurement based model @xcite . we start by describing one - dimensional continuous - variable quantum wires which can be used to carry a single qubit of quantum information . the techniques we develop are independent of the actual physical implementation but we will often describe them with terms of quantum optics and have such a system in mind . we consider a chain of @xmath41 quantum systems with dimension @xmath42 which are called lattice sites . we also allow for the situation @xmath43 which describes cv light modes . we say that a state is physical if it has finite mean energy , i.e. , for a single mode : @xmath44 . we choose a basis ( or a countably infinite hilbert - space basis ) @xmath45 of @xmath46 , which we call the computational basis , associate to any basis element a @xmath47-dimensional matrix @xmath48\in\mathcal{m}_d({{\mathbbm{c}}})$ ] , and write a translationally invariant matrix product state ( mps ) as @xmath49\ldots a[i_2]a[i_1]|r\rangle|i_l,\ldots , i_2,i_1\rangle \label{eq : mpsdef}\ ] ] where @xmath50 . this vector space is the one where the quantum computation will take place and is called correlation space . the matrices must fulfill the completeness relation @xmath51^\dag a[i]={\mathbbm{1}},\ ] ] for rendering deterministic computation feasible . an important fact is that all physical states can be approximated arbitrary well by a finite - dimensional mps . as any finite - dimensional state can be written as a mps @xcite , it only remains to show that one can truncate states with finite energy . we write the mean total photon number of a state @xmath52 with @xmath53 modes as @xmath54 where the matrix elements of @xmath52 are the ones of the fock basis . let @xmath55 be the non - normalized state obtained from @xmath52 by setting @xmath56 if one of the indices is larger than @xmath57 . this state fulfills @xmath58 denoting the trace norm , which is the relevant norm for the distinguishability of quantum states , by @xmath59 , we get from ref . @xcite @xmath60 if we assume the mean photon number ( the energy ) _ per mode _ to be constant , the truncation error is independent of the system size . this means , for a given required accuracy , we can choose a cut - off number @xmath57 independently of the length of the intended computation . we now show how single - qudit mbqc can be performed with such a matrix product state , following and extending the line of reasoning used in ref . let us assume , the first mode is measured in the computational basis with result @xmath53 . then , the remaining system is described by the state vector @xmath61\ldots a[i_2]a[k]|r\rangle|i_l,\ldots , i_2\rangle . \label{eq : mpsdef2}\ ] ] this can be interpreted as the action of the matrix @xmath62 $ ] on the right boundary vector @xmath63 . to interpret this as the action of a quantum gate it is necessary that @xmath62 $ ] is proportional to a unitary matrix , i.e. , @xmath62^\dag a[k]=p(k){\mathbbm{1}}$ ] where @xmath64 is the probability with which the measurement result @xmath53 is obtained . when the measurement basis is continuous , we denote by @xmath65 the probability density while @xmath66 denote the corresponding probability measure . because we will also discuss measurements where the corresponding matrices are not proportional to unitary ones , we note that the probability in this more general situation is @xmath67|\psi\rangle\|^2 $ ] , where @xmath68 is the state vector of the correlation system . if the measurement is not performed in the computational basis but in another one and the result corresponds to a projection to the state vector @xmath69 , the matrix applied on @xmath63 reads @xmath70=\sum_{i=1}^{d_p}\langle x|i\rangle a[i]\ , . \label{eq : skewmeasurement}\ ] ] note that the basis @xmath71 may be continuous and/or overcomplete and that measurements in such bases naturally occur in cv - quantum optics . we also allow this basis to consist of improper eigenstates reflecting an idealized homodyning detection . we can now already note an almost trivial but very important necessary requirement for a single - qudit mbqc scheme : if there exists no allowed measurement basis such that @xmath72 $ ] is proportional to a unitary matrix for almost all @xmath73 , it is not even possible to transport a single @xmath47-dimensional qudit . an equivalent formulation is that @xmath74 must not depend on the state vector @xmath68 of the correlation space . if this was the case , a measurement would yield information about @xmath68 , which would clearly destroy coherence . a possible way of preparing a matrix product state with bond dimension @xmath47 is a sequential interaction with a @xmath47-dimensional auxiliary system @xcite . this picture proves to be extraordinarily useful when discussing quantum wires . let the interaction between this auxiliary system and a local physical system be described by @xmath75 and assume the physical system to be initialized in the state vector @xmath76 . then , the matrix elements of the mps matrices read @xmath77_{j , k}=\left(\langle k|\otimes\langle i|\right)\hat{u}\left(|j\rangle\otimes|\psi\rangle\right).\ ] ] in this picture , which is sketched in figure [ fig : seqprep ] , the correlation system is identified with the auxiliary system , and a single qudit gate is performed by first letting the auxiliary system interact with a mode in a known state and then measuring the mode . it is important to note that this picture works regardless of the actual way the resource state is prepared . for this reason , we will use the picture of sequential preparation to be more intuitive while maintaining full generality . for example , resource states can also arise as ground states of local hamiltonians or be prepared by the action of translationally invariant quantum cellular automata @xcite . if the goal is not to process a @xmath47-dimensional qudit but only a @xmath78-dimensional one with @xmath79 , the above requirement is too strong . the @xmath78-dimensional space is called the computational subspace . we introduce the encoding @xmath80 with @xmath81 and the encoded matrices @xmath82=v^\dag a[i]v\,,\quad b\in\mathcal{m}_d({{\mathbbm{c}}})\ , . \label{eq : enca}\ ] ] if they are proportional to unitary matrices for all @xmath83 , it is possible to process a single @xmath78 dimensional qudit . one can observe that @xmath84\ldots a_{b_1}[i_1 ] v|\psi\rangle\\ = \langle\phi| b_{b_l}[i_l]\ldots b_{b_1}[i_1]|\psi\rangle\end{gathered}\ ] ] where @xmath85 . the subscripts @xmath86 denote the chosen measurement basis . this means , for all measurements , the state with the matrices @xmath87 behaves exactly like the one with the matrices @xmath88 . thus , we can just proceed as if we had a state with a @xmath78-dimensional correlation space from the very beginning . this situation is not the most general one . the isometry ( [ eq : encoding ] ) can depend on the measurement basis and the outcome of the previous steps . in this case , the computational subspace is not fixed but changes during the computation . it is reasonable to demand that at least the initial and the final encoding coincide so that it is possible to work with a fixed in- and out - coupling , provided by @xmath89 . in this case ( [ eq : enca ] ) becomes @xmath90v_{b_{l-1}}[i_{l-1}]\ldots v_{b_1}[i_1]v^\dag_{b_1}[i_1]a_{b_1}[i_1]v|\psi\rangle\\ = \langle\phi| b_{b_l}[i_l]\ldots b_{b_1}[i_1]|\psi\rangle\end{gathered}\ ] ] where @xmath91=v^\dag_{b_k}[i_k]a[i_k]v_{b_{k-1}}[i_{k-1}]$ ] . when several quantum wires are used to perform a quantum computation , it is also necessary to demand the changing encoding to return to the initial encoding given by @xmath89 , whenever a coupling occurs . if this was not the case , the couplings would have to depend on the history of measurements and their results in both of the wires . this would clearly be against the spirit of a measurement - based protocol , where all the adaptation lies in the choice of measurement bases only . the infinite - dimensional matrix product states used in this work bear some resemblance to the continuous matrix product states which where introduced in refs . @xcite to describe one - dimensional bosonic quantum fields . in contrast to the discrete structure of bosonic modes described here , they are formulated without an underlying lattice . however , such cmps can be obtained by a suitable continuum limit from infinite - dimensional matrix product states . in this limit , only two independent mps matrices remain while the rest can be calculated from them which is an important difference to the present situation . interpretation of single - qudit mbqc as sequential interaction with an ancilla . the physical sites are initially in the state vector @xmath76 and interact with the correlation system through @xmath92 . depending on the measurement results in @xmath93 and @xmath94 , the unitary gate @xmath95 is applied on @xmath96 . ] to go beyond single - qudit processing and achieve quantum computational universality , wires have to be coupled . physically , this can be performed in different ways . one possibility is to let the auxiliary systems of the sequential preparation scheme interact . however , this might be very challenging to do in reality , both when working with atoms , due to the difficulty of controlling them , and with light , due to weak non - linear interaction . for this reason , joint measurements are preferable to couple two wires which is also shown in figure [ fig : coupling ] . in an optical situation , this could mean to combine two modes belonging to two different wires on a beam splitter and measuring both output ports . this is a broadening of the usual definition of mbqc where only local measurements are performed while we now also allow for two - local ones . on the other hand , there also exists a closely related third way which is also based on the beam - splitter interaction but strictly stays within the measurement - based paradigm . the idea is to perform the coupling independently of the executed algorithm according to some fixed scheme . this step then belongs to the preparation of the resource while the subsequently performed measurements are purely local . let @xmath97 be the coupling between the two modes , @xmath98 and @xmath99 the measurement bases after interaction . then , the applied operation reads @xmath100=\sum_{i , j}\langle x_1,x_2|\hat{w}|i , j\rangle a[i]\otimes a[j].\ ] ] coupling of two quantum wires . the interaction of two physical systems through @xmath97 and the subsequent measurements ( @xmath93 and @xmath94 ) induces a entangling gate on the two correlation systems . ] in this section , we give three condition a mbqc - scheme has to fulfill to be called efficient and comment on their importance . to facilitate the derivation , we restrict ourselves to the , arguably most important , case of @xmath101 while stressing that the requirements for universal computing on qudits with larger dimension are completely analogous . when discussing requirements , one has to distinguish between two different notions : one the one hand , there are _ theoretical _ requirements , of the kind that , if they are met , efficient scalable quantum computing is possible in the absence of noise . going away from this idealized situation , on the other hand , two new features arise . firstly , in practice , any quantum operation is affected by noise , which error - correction may allow to counter to some extent . secondly , any actual quantum computation is , needless to say , finite . thus , the practical requirements that might be sufficient to sustain a computation may differ from the theoretical ones . however , as we are primarily interested in scalable quantum computing and for reasons of clarity we will focus on the theoretical requirements and comment on other practical implications whenever it is in order . we start with the theoretical requirements for quantum wires ( which process single qubits ) and state the first necessary condition . [ req : requirement ] let @xmath102 be the state vector of the correlation system after @xmath103 steps in the quantum wire when the initial state vector was @xmath68 . we say that transport of a single qubit is possible if there exists some @xmath104 such that for all @xmath103 @xmath105 here , @xmath95 is the best attempt at undoing the action of the quantum wire on @xmath68 . this requirement ensures that computations are not limited in length . if encodings of single logical qubits in multiple wires are used , this requirement can be weakened to @xmath106 . for universal single - qubit quantum computing to be possible , it is not enough to merely transport quantum information . it is also necessary that any one - qubit unitary @xmath107 can be efficiently approximated . because it depends on the measurement outcome which operation is applied , we can not hope for this to be possible deterministically . instead , the measurements induce a random walk on @xmath108 which can be controlled by choosing the measurement basis in the next step depending on the gate implemented so far . for efficient quantum computing , the average number of steps must not increase too fast with the desired accuracy . we set @xmath109:= { p(k)}^{-1/2}a_\mathcal{b}[k]\ ] ] where @xmath110 denotes the measurement basis and define @xmath111=\prod_{i=1}^n\bar{a}_{\mathcal{b}_i}[k_i].\ ] ] [ req : approximation ] let @xmath107 and @xmath104 . there exits a sequence ( possibly adaptive ) of bases @xmath112 such that the expectation value for an approximation @xmath113 $ ] fulfilling @xmath114 after at most @xmath103 measurements in the wire satisfies @xmath115 here , @xmath116 denotes the operator norm , which is the meaningful figure of merit in this case . of course , single qubit operations are not sufficient for universality but an entangling operation is also needed : [ req : coupling ] denote with @xmath117 the controlled-@xmath118 gate . the expectation value for an approximation @xmath119 fulfilling @xmath120 after at most @xmath103 measurements on both of the quantum wires involved , where both single - site and two - site measurements are allowed , must satisfy @xmath121 requirements [ req : approximation ] and [ req : coupling ] guarantee that every quantum circuit can be simulated in correlation space with a polylogarithmic overhead . this is the commonly used requirement in quantum computing . when one is content with a polynomial overhead , ( [ eq : approximation ] ) and ( [ eq : coupling ] ) can both be relaxed to @xmath122 for some constant @xmath123 . a computational model fulfilling only this weaker condition is still of interest when the ultimate aim is to perform a quantum algorithm which provides exponential speed - up over the corresponding classical one ( like shor s algorithm ) . however , whenever the speed - up is only polynomial ( like in grover s search algorithm ) , the stronger polylog scaling is clearly necessary for the quantum algorithm to have any advantage . when one aims at the implementation of constant size quantum circuits only , requirements 2 and 3 are not necessary as one does not need to implement the quantum gates to arbitrary accuracy but only to some @xmath104 which is determined by the total length of the computation . the last feature to demand from a mbqc model is the ability to initialize the correlation system into some known state and to perform a measurement of the correlation space in the computational basis . [ req : iniandreadout ] for efficient initialization of the correlation system to be possible , there must exist some measurement sequence such that , independently from its initial state vector @xmath68 , @xmath124 here @xmath102 is the state vector after @xmath103 steps in the wire and it must be true that @xmath125 . for readout , it is necessary that after @xmath103 steps with @xmath126 the combined normalized action @xmath111 = \prod_{j = n}^1\bar{a}_{\mathcal{b}_j}[\mathbf{k_j}]\ ] ] fulfills either @xmath127-|0\rangle\langle 0|\|\le{\varepsilon}$ ] or @xmath127-|1\rangle\langle 1|\|\le{\varepsilon}$ ] . the first requirement just means that one can approach one of the computational basis states fast enough , while the second one ensures an efficient implementation of an approximative measurement in the computational basis . while measurement - based computational schemes that do not respect these requirements may be conceivable in principle , within the framework presented here they are very natural indeed and necessary for universal quantum computing . in the previous section we have discussed the requirements a mbqc scheme has to fulfill in order to be called efficient . we now show some consequences of these requirements which will help us both to find classes of schemes which can not be efficient and such that are . this seems an important enterprise in order to arrive at `` no - go results '' , so to identify serious limitations that have to be circumvented in one way or the other . for many important situations , @xmath128 in requirement [ req : requirement ] vanishes exponentially in @xmath103 as the following observation shows , which by virtue of the above statement implies that efficient quantum computing is not possible . [ obs : transport ] given a basis @xmath110 , let @xmath129 , with @xmath130 . if for all @xmath131 , @xmath72\not\propto u$ ] with @xmath107 , transport is impossible when measuring in this basis . proof : we calculate the fidelity @xmath132\in\mathbb{su}_2}\min_{|\psi\rangle}\frac{|\langle\psi|u[x]a[x]|\psi\rangle|^2}{p(x)}.\ ] ] because @xmath130 and @xmath133 for all @xmath134 , there exists a non - empty compact interval @xmath135 such that there is some @xmath136 with @xmath137 for all @xmath138 . using an argument from ref . @xcite , one can show that this implies the exponential decay of @xmath128 , showing that requirement [ req : requirement ] is not satisfied . thus , for transport to be possible , there must exist a basis such that almost all measurement outcomes correspond to some @xmath72 $ ] with @xmath72^\dag a[x]\propto{\mathbbm{1}}$ ] . for transport over a fixed length , observation [ obs : transport ] does not apply because even if the fidelity decays exponentially fast , it might be still large enough for the task under question . we show a simple sufficient condition , which will be used later when discussing examples for which requirement [ req : approximation ] is true : [ obs : randwalkcond ] assume that for any target gate @xmath107 there exists some basis @xmath110 and some @xmath129 for any @xmath139 with @xmath140 where @xmath141 is the desired accuracy . when @xmath142-u\|\le{\varepsilon}\ ] ] for all @xmath143 , requirement [ req : approximation ] is fulfilled . proof : the expected number of tries can be bounded by @xmath144 the most important important situation covered by observation [ obs : randwalkcond ] is the one where the measurements have discrete results and for every @xmath107 a basis exists which contains at least a single element which fulfills the requirements of observation [ obs : randwalkcond ] . sufficient conditions for initialization and read - out of the correlation system to be possible are provided by the following observation : [ obs : iniandreadout ] let there exist some @xmath145 and a basis @xmath110 containing some subset @xmath129 of `` non - unitary measurements '' . assume that @xmath146\propto u[x]\left[\begin{array}{cc } s_1(x ) & 0 \\ 0 & s_2(x ) \end{array}\right]v^\dag\ ] ] for all @xmath134 , where @xmath147,v\in \mathbb{su}_2 $ ] and @xmath148 , @xmath149 are the ( not necessarily ordered ) singular values of @xmath72 $ ] which fulfill either @xmath150 or @xmath151 . in this case , efficient read - out and initialization of the correlation system is possible . proof : the situation is most transparent when @xmath152 . in this case , a measurement result @xmath73 corresponds to a projection of the correlation space state to the state vector @xmath153 . in the more general case the measurement is not projective but merely weak . by performing the unitary operation @xmath154 $ ] , which is efficiently possible due to requirement [ req : approximation ] , and repeating the measurement , a projective measurement in the basis @xmath155 can be approximated . this approximation is efficient , due to the existence of a finite gap between @xmath156 and @xmath157 occurring with finite probability . the independence of @xmath89 of @xmath73 is crucial : if this is not the case , the basis in which the correlation space measurement will occur can not be fixed . this will lead to a destruction of quantum information resulting in a failure of the mbqc scheme . initialization of the correlation system in the state vector @xmath158 can be performed in the very same manner . one just performs the measurement procedure outlined above after which the correlation system is in @xmath147|0\rangle$ ] or @xmath147|1\rangle$ ] , depending on the outcome . applying now the gates @xmath147^\dag$ ] or @xmath159^\dag$ ] , respectively , achieves the required initialization . when devising a mbqc scheme , there are several fundamental limitations concerning the use of gaussian states and gaussian measurements . the first important fact is that all protocols involving only gaussian measurements on gaussian states can be simulated efficiently on a classical computer , ruling out the possibility for any quantum speed - up @xcite . however , for an _ ideal _ gaussian cluster state , universality can be achieved by adding a single non - gaussian measurement to the toolbox @xcite . on the other hand , any _ physical _ gaussian quantum wire , including an one - dimensional gaussian cluster state can not fulfill requirement [ req : requirement ] even when allowing for non - gaussian measurements @xcite . we now discuss further limitations and consider the situation where the interaction ( [ eq : interactionunitary ] ) can be written as @xmath160 up to local unitaries on the correlation system , where @xmath161 . this class contains states which allow for transport of a qubit in a wire by gaussian measurements only . an example is given by @xmath162 where @xmath163 is a parameter characterizing the interaction strength . taking a coherent state vector @xmath164 with @xmath165 , @xmath166 as input , @xmath167 , @xmath168 , a @xmath169-quadrature measurement with result @xmath73 applies in correlation space @xmath170={\rm diag}\,(e^{ip_0(x-\frac{q_0}{2})},e^{-ip_0(x-\frac{q_0}{2})})\ ] ] with probability @xmath171 . thus , requirement [ req : requirement ] is fulfilled . however , efficient single - qubit gates are impossible due to the lack of control . the only possible measurement ( up to squeezing ) for which @xmath172 $ ] is unitary are phase - space displacements followed by @xmath169-quadrature homodyne detection as for all others , the projectors do not fulfill @xmath173 which implies that the probabilities of the measurements depend on the state of the correlation system and , therefore , the applied matrix is not unitary . writing the displacement as @xmath174 , we get @xmath175={\rm diag}\,(e^{i(p_0+\delta_p)(x-\frac{q_0+\delta_q}{2})},e^{-i(p_0-\delta_p)(x-\frac{q_0+\delta_q}{2})})\ ] ] with probability @xmath176 . up to a redefinition @xmath177 and some global phase , ( [ eq : ttt3 ] ) is constant in @xmath178 , i.e. , there is no way of controlling the random walk , making it impossible to meet requirement [ req : approximation ] . the same argument holds for any controlled gaussian operation and for any gaussian input state . when allowing for non - gaussian input states , some amount of control is indeed possible , as the following example shows : consider the interaction of eq . ( [ eq : controlledrot ] ) together with a superposition of two different photon numbers as an input , @xmath179 . choosing @xmath180 , the post - interaction state reads @xmath181 and the normalized matrices @xmath182 and @xmath183 for the @xmath169- and @xmath65-quadrature measurements are @xmath184=\,&{\rm diag}\,(e^{-i\phi(x)},e^{i\phi(x)}),\\ \bar{a}_p[x]=\,&{\rm diag}\,(e^{i\phi(x)},e^{-i\phi(x)}),\end{aligned}\ ] ] with @xmath185 and @xmath186 thus , there exist two gaussian measurements which allow to control the random walk but even this control is not enough to meet requirement [ req : approximation ] . because this is a general feature of any mbqc - scheme relying solely on gaussian measurements we will discuss it in detail . we consider an mps with @xmath101 with the continuous family of matrices @xmath187 $ ] with @xmath188 . as the projectors describing eight - port homodyning are not orthogonal , they will turn the state of the remaining system into a mixed one when measuring a mode , and we do not need to consider this measurement . because all relevant gaussian measurements can be expressed as the application of squeezing , rotating , and displacing , followed by a @xmath169-quadrature measurement , we can write @xmath189=\cos(\theta)a_q[\lambda x+\delta_q]+\sin(\theta)a_p[x/\lambda+\delta_p]\ ] ] where @xmath190=\frac{1}{\sqrt{2\pi } } \int{\rm d}x\,a_q[x]e^{iqx},\ ] ] and where we have omitted global phase factors . if single - qubit transport is possible in this wire , we can , without loss of generality , assume that @xmath187:=a_{0,0,0,0}[x]$ ] is proportional to a unitary matrix for all relevant @xmath73 . as both squeezing by a finite @xmath191 and shifting in phase space does merely result in a redefinition of @xmath73 in eq . ( [ eq : allgauss ] ) , we can restrict ourselves to the discussion of @xmath192= & \cos(\theta ) a_{q}[x]+ \sin(\theta)a_{p}[x]\nonumber\\ = & \sum_{n=0}^{d_p-1 } e^{i\theta n}\psi_n(x)a[n]\end{aligned}\ ] ] where @xmath193 are the energy eigenfunctions of the harmonic oscillator . we assume that the physical dimension is finite and discuss the required changes due to an infinite physical dimension afterwards . we now argue that requirement [ req : approximation ] can not be fulfilled in the present situation : [ obs : nocontrol ] let @xmath194 $ ] be as in ( [ eq : allgauss2 ] ) with finite @xmath195 . then , there exist some constants @xmath196 , @xmath197 , and @xmath198 such that for all @xmath107 , @xmath199 $ ] , and all @xmath200 @xmath201-u\|\le{\varepsilon})<c{\varepsilon}^\lambda,\ ] ] where @xmath202 denotes the probability for the random variable @xmath73 which corresponds to a quadrature measurement with angle @xmath203 . if this is true , the expected number of tries to implement any unitary is bounded as @xmath204 for @xmath200 , and requirement [ req : approximation ] is not fulfilled . the proof is given in the appendix this result , which can be summarized as `` no quantum computation with gaussian measurements only '' is complementary to the one reported in ref . @xcite which can , in turn , be summarized as `` no quantum computing with gaussian states only '' . these results apply only to the model where all operations are noiseless and error - correction is not performed . together , this means , both non - gaussian states and non - gaussian measurements are simultaneously necessary for continuous variable mbqc . however , the two limitations are of very different nature . the reason for the former is that there exists not enough localizable entanglement which means that the quantum information gets destroyed along the wire and transport or teleportation is not possible . the root of the latter limitation is the insufficient possibility to control what quantum gate is performed . up to now , we have only considered resource - states where the physical dimension is finite . we demonstrate the difference between states with and without support on a finite number of fock states with the help of an example where the resource state is described by the matrices @xmath146=\frac{1}{\sqrt{\mathcal{n}}}\left(f(x){\mathbbm{1}}+ig(x)z\right)\ ] ] with @xmath205 to ensure the normalization condition @xmath206^\dag a[x]={\mathbbm{1}}.\ ] ] both @xmath207 and @xmath208 are chosen to be real , even functions . thus , also their fourier transforms @xmath209 , @xmath210 are real and the matrices corresponding to @xmath36 are @xmath211=\frac{1}{\sqrt{\mathcal{n}}}\bigl((f(x)\cos(\theta)\\ + \tilde{f}(x)\sin(\theta)){\mathbbm{1}}+i(g(x)\cos(\theta)+\tilde{g}(x)\sin(\theta))z\bigr),\end{gathered}\ ] ] which are all unitary . if the resource state is finite - dimensional , @xmath207 , @xmath208 , @xmath209 , and @xmath210 are of the form @xmath212 and the probability for @xmath213 $ ] to differ too much from the desired unitary is too large . if we allow the resource state to be infinite dimensional , the situation changes . in particular , it is possible that @xmath214 for @xmath215 for some constant @xmath123 . this allows to find a resource state where a @xmath169-quadrature measurement implements some gate with a finite probability . this is a striking example for a qualitative difference between a proper continuous - variable state and all of its finite dimensional truncations . one might now be tempted to use this new insight to devise a mbqc scheme which allows for theoretically efficient single - qubit processing with gaussian measurements only , but this is impossible . this can be immediately seen by truncating the state according to ( [ eq : meanphoton ] ) . then , the above results imply that single qubit mbqc is impossible when the energy per mode stays constant when scaling the system . if , on the other hand , the energy per mode is increased with the length of the computation , this no - go result does not hold any longer . the reasons for the existence of such a severe limitation lies in the strong notion of efficiency . if eq . ( [ eq : coupling ] ) in requirement [ req : approximation ] is weakened to ( [ eq : coupling2 ] ) , this restriction is not present anymore , and it is possible to devise a mbqc - scheme based on gaussian measurements on a non - gaussian resource . to do so , one can take the resource defined by the matrices ( [ eq : aq ] ) . after changing the variable and neglecting global phases , one gets @xmath216=s(-p_0q_0)s(-2p_0x)\ ] ] with @xmath217 where @xmath218 is the phase gate . assume that the target gate is @xmath219 . chose @xmath220 and @xmath221 such that @xmath222 and , therefore , @xmath223=s(\theta)$ ] . using @xmath224 , one can bound the probability that the distance of the actually implemented gate to the desired one is larger than @xmath141 as @xmath225-s(\theta)\|>{\varepsilon})\le{\mathbbm{p}}(|2p_0x|>{\varepsilon})\le cp_0{\varepsilon}\ ] ] where @xmath226 is some constant . this means , it is possible to approximate an arbitrary phase gate in @xmath227 steps . changing the interaction unitary to contain an additional hadamard gate on the correlation space and using the decomposition detailed below together with the composition law for errors , it is easy to show that any @xmath107 can be approximated in @xmath228 steps . in this section , we present a scheme which allows for efficient cv - mbqc . the limitations discussed above force us to use both non - gaussian resource states and non - gaussian measurements . even though the experimental requirements for its realization are indeed challenging , the scheme does only use primitives which all have already been demonstrated to be feasible in proof - of - principle experiments . we choose the interaction unitary in eq . ( [ eq : interactionunitary ] ) to be @xmath229 where @xmath163 is a parameter . we initialize the modes in a coherent state vector @xmath164 with @xmath166 . ( [ eq : exunitary ] ) describes a controlled rotation in phase space . the class of measurements corresponding to unitary evolution is given by the displaced photon - counting measurements , i.e. , projections onto the state vector @xmath230 where @xmath188 . with the help of ( [ eq : defalpha ] ) , the applied gates read @xmath231=h\left[\begin{array}{cc } e^{-in\phi(x ) } & 0 \\ 0 & e^{in\phi(x)}\end{array}\right]\ ] ] where @xmath232 the corresponding probabilities are @xmath233 and fulfill , for every @xmath73 , the normalization condition @xmath234 . any single - qubit unitary operation can be efficiently approximated with those operations : every @xmath107 can be written as @xmath235 @xmath236 is the hadamard gate , and @xmath237 $ ] . rewriting this as @xmath238 we have decomposed every gate into four applications of ( [ eq : axn ] ) . inverting ( [ eq : phix ] ) yields @xmath239 to implement @xmath219 , one only has to choose @xmath73 such that @xmath240 for some small @xmath103 where @xmath241 and @xmath242 suffice and the former is only needed for @xmath243 . for any fixed @xmath244 and @xmath245 , the maximally necessary shift @xmath246 is bounded and with ( [ eq : probx ] ) there exists a lower bound @xmath247 for all relevant @xmath73 and @xmath103 . using ( [ eq : su2decomp2 ] ) , one can see that the probability to obtain any chosen @xmath107 in four steps is lower bounded by @xmath248 . thus , the conditions for observation [ obs : randwalkcond ] are fulfilled and efficient single - qubit processing is possible . we now turn to the coupling of two wires . let the two correlation space qubits be prepared in @xmath249 for @xmath250 . then , the state vectors after the interaction with the light mode reads @xmath251 displacing now both modes by @xmath252 , rotating the second mode by an angle of @xmath253 , and setting @xmath254 , the joint state vector becomes @xmath255 a balanced beam - splitter transforms two coherent states as @xmath256 ) yields @xmath257 the matrices corresponding to a photon - counting measurement in both modes with results @xmath258 and @xmath259 respectively read @xmath260=&{\rm diag}\,(e^{-\frac{i\pi n_1}{4}},e^{-\frac{i7\pi n_1}{4}},e^{-\frac{i3\pi n_1}{4}},e^{-\frac{i5\pi n_1}{4}})\nonumber\\ & \times{\rm diag}\,(e^{-\frac{i7\pi n_2}{4}},e^{-\frac{i\pi n_2}{4}},e^{-\frac{i5\pi n_2}{4}},e^{-\frac{i3\pi n_2}{4}})\end{aligned}\ ] ] which is , up to a local unitary , a controlled phase gate , i.e. , @xmath261={\rm diag}\,(e^{-\frac{i\pi(n_1 + 7n_2)}{4}},e^{-\frac{i\pi(3n_1 + 5n_2)}{4}})\\ \otimes{\rm diag}\,(1,e^{-\frac{i\pi(6n_1 - 6n_2)}{4}})\\\times{\rm diag}\,(1,1,1,e^{-i\pi(-n_1+n_2)}).\end{gathered}\ ] ] if @xmath262 is even , @xmath263 $ ] is not entangling , and the local operations can be efficiently undone as described above . if @xmath262 is odd , a @xmath264-gate is implemented up to local correction . as this situation occurs with finite probability for any choice of @xmath265 and @xmath203 , this means that requirement [ req : coupling ] is fulfilled . it remains to show that initialization and read - out are possible . to measure in the computational basis , we shift the mode after the interaction by @xmath266 which turns the joint state of qubit and mode to @xmath267 counting the photons leads to a matrix fulfilling the conditions of requirement [ req : iniandreadout ] . note that there is some asymmetry : if the counted number of photons is larger than one , the qubit is in @xmath268 . if the result is zero , one has to undo the gate @xmath269 and repeat the procedure . due to the gap between @xmath26 and @xmath270 , this is efficiently possible . this concludes the proof that the proposed scheme is efficient . as already mentioned , the coupling scheme discussed so far is not completely in the mbqc - paradigm because it involves the adaptive coupling of sites . however , this is not necessary : consider two wires which are coupled every @xmath271 sites by the beam - splitter described by eq . ( [ eq : beamsplittera ] ) . because @xmath272 , two consecutive couplings can be undone by appropriately choosing local operations between them . from eq . ( [ eq : probx ] ) it follows that for every block of four sites , there exists a probability @xmath273 that the desired gate is realized and probability of failure in at least one of the wires is upper bounded by @xmath274 . let @xmath103 be total number of gates , the total probability of failure is bounded by @xmath275 for suitable constants @xmath276 and @xmath277 . thus , for a fixed probability of success , @xmath53 has only to grow as @xmath278 , which means that the need for two - site measurements can be removed for the price of a polylog overhead . for a quantum circuit consisting of more than two wires , as depicted in figure [ fig : realmbqc ] , every wire should be alternately coupled to its left and right neighbor . when the circuits consists of @xmath279 quantum wire with all have a length @xmath103 , the respective overhead behaves as @xmath280 . non - adaptive coupling of four quantum wires where the wiggly lines denote the probabilistic measurement - based @xmath264 gate . by scaling the distance between the single couplings polylogarithmically in the length of the computation , the total probability of failure can be made arbitrarily small . ] even if a model fulfills the theoretical requirements it will still be affected by noise . thus , error correction will be necessary to build a scalable quantum computer . in the scheme presented above , one plausible dominant source of errors is reflecting the finite efficiency of the photon - counting measurement needed to perform both single and two qubit gates . if photon loss happens , a `` wrong '' operation is applied in the two - dimensional correlation space . when the rate of these qubit errors is low enough , techniques of error correction and fault - tolerant quantum computing may be applied to make a computation with unbounded length possible at the expense of a polylogarithmic overhead @xcite . this contrasts with the errors stemming from the use of finitely squeezed states in the scheme based on gaussian cluster states for which no method of error correction exists yet to achieve fault tolerance . beside of the obvious errors stemming from decoherence and imperfections in the measurement procedure , the presented continuous variable schemes have an important additional source of errors . for example , when performing a shifted photon counting , uncertainty in the parameter of the shift operator @xmath281 results in uncertainty of the applied operator . we now investigate how those errors scale with the length of the computation . we restrict ourselves to the single - qubit case while noting that the error analysis for multiple qubits is completely analogous . consider the situation where the target gate reads @xmath282 while the gate actually applied is @xmath283 . we assume that the error induced by a single step is bounded , i.e. , for all @xmath83 , @xmath284 if all @xmath285 are also unitary , the total error grows linearly in @xmath103 and can be bounded as @xmath286 . if this is not the case , an tight upper bound to the error is @xmath287 i.e. , it grows , in the worst case , exponentially in @xmath103 . in the mbqc - scheme based on gaussian cluster states as discussed , e.g. , in refs . @xcite , the implemented gates are not unitary for physical resource states which is the reason for the exponential decay of the transport fidelity @xcite . in this work , we have investigated the possibilities provided by measurement - based quantum computing with continuous - variable resources . after introducing a framework allowing for the description of a huge class of possible resource states , we have clarified which conditions are necessary for a cv - mbqc scheme to be viable . those conditions lead to new limitations . especially , we have shown that , without the use of not yet developed continuous - variable error correction , scalable quantum computing with gaussian measurements alone is impossible , even if the resource state is non - gaussian , complementing the prior result which prohibits mbqc with gaussian resources , even if the measurements are non - gaussian . as the second main result , we gave an explicit example of an efficient mbqc scheme where the resource could be created by a simple interaction between light modes in coherent states and some qubit degrees of freedom . processing is then performed by shifted photon counting while entangling two - qubit gates rely on simple beam splitters . an analysis of errors highlighted the qualitative differences of the two major kinds of errors . two major conclusions can be drawn from our findings : first , continuous - variable measurement based quantum computing is an extraordinarily difficult enterprise . second , when non - gaussian resource states are combined with non - gaussian measurements , efficient schemes do exist , and in this work , we introduce such first strictly efficient schemes . while this may be disappointing when aiming at a scalable optical quantum computer , one does need to keep in mind , that our statements about efficiency only address the asymptotic behavior . thus , the limitation do not at all rule out the usefulness of much simpler protocols in situations where only a few modes are used . such situations include quantum repeaters and hybrid entanglement - distribution protocols for applications particularly in quantum cryptography . hence , for such purposes , both the purely gaussian setting as well as those non - gaussian schemes that asymptotically do not give rise to universal quantum computing can well be feasible as elementary tools for entanglement distillation schemes . it is the hope that the framework established here fosters further such investigations . we would like to thank the eu ( qessence , compas , minos ) , the bmbf ( quorep ) , and the euryi for support . 10 r. raussendorf and h. j. briegel , phys . * 86 * * , 5188 ( 2001 ) . r. raussendorf and h. j. briegel , quant . inf . comp . * 6 * , 433 ( 2002 ) . d. gross and j. eisert , phys . lett . * 98 * , 220503 ( 2007 ) . d. gross , j. eisert , n. schuch , and d. perez - garcia , phys . a * 76 * , 052315 ( 2007 ) . j. eisert and m. b. plenio , int . j. quant . inf . * 1 * , 479 ( 2003 ) . s. l. braunstein and p. van loock , rev . ph / ys . * 77 * , 513 ( 2005 ) . j. eisert , s. scheel , and m. b. plenio , phys . rev . lett . * 89 * , 137903 ( 2002 ) . j. fiurek , phys . lett . * 89 * , 137904 ( 2002 ) . g. giedke and j. i. cirac , phys . a * 66 * , 032316 ( 2002 ) . j. niset , j. fiurek , and n. j. cerf , phys . lett . * 102 * , 120501 ( 2009 ) . s. d. bartlett , b. c. sanders , s. l. braunstein , and k. nemoto , phys . lett . * 88 * , 097904 ( 2002 ) . n. c. menicucci , p. van loock , m. gu , c. weedbrook , t. c. ralph , and m. a. nielsen , phys . * 97 * * , 110501 ( 2006 ) . n. c. menicucci , s. t. flammia , and o. pfister , phys . * 101 * , 130501 ( 2008 ) . s. t. flammia , n. c. menicucci , and o. pfister , j. phys . b * 42 * , 114009 ( 2009 ) . r. ukai , j .- yoshikawa , n. iwata , p. van loock , and a. furusawa , phys . a * 81 * , 032315 ( 2010 ) . m. gu , c. weedbrook , n. c. menicucci , t. c. ralph , and p. van loock , phys . a * 79 * , 062318 ( 2009 ) . m. ohliger , k. kieling , and j. eisert , phys . a * 82 * , 042336 ( 2010 ) . m. a. nielsen and c. m. dawson , phys . a * 71 * , 042323 ( 2005 ) . r. raussendorf and j. harrington , phys . lett . * 98 * , 190504 ( 2007 ) . t. c. ralph , a. gilchrist , g. j. milburn , w. j. munro , and s. glancy , phys . a * 68 * , 042319 ( 2003 ) . t. p. spiller , k. nemoto , s. l. braunstein , w. j. munro , p. van loock , and g. j. milburn , new . j. phys * 8 * , 30 ( 2006 ) . van loock , w. j. munro , k. nemoto , t. p. spiller , t. d. ladd , s. l. braunstein , and g. j. milburn , phys . a * 78 * , 022303 ( 2008 ) . w. p. schleich , _ quantum optics in phase space _ , wiley - vch ( 2001 ) . p. grangier , j. a. levenson , and j .- poizat , nature * 396 * , 537 ( 1998 ) . d. achilles , c. silberhorn , c. sliwa , k. banaszek , i. a. walmsley , m. j. fitch , b. c. jacobs , t. b. pittman , and j. d. franson , j. mod . opt . * 51 * , 1499 ( 2004 ) . a. divochiy , f. marsili , d. bitauld , a. gaggero , r. leoni , f. mattioli , a. korneev , v. seleznev , n. kaurova , o. minaeva , g. goltsman , k.g . lagoudakis , m. benkhaoul , f. levy , and a. fiore , nature photonics * 2 * , 302 ( 2008 ) . s. rebic , s. m. tan , a. s. parkins , and d. f. walls , j. opt . b. * 1 * , 490 ( 1999 ) . s. parkins , nature * 465 * , 699 ( 2010 ) . d. gross and j. eisert , phys . a * 82 * , 040303(r ) ( 2010 ) . m. fannes , b. nachtergaele , and r. f. werner , comm . * 144 * * , 443 ( 1992 ) . d. perez - garcia , f. verstraete , m. m. wolf , and j. i. cirac , quant . comp * * 7 * * , 401 ( 2007 ) . m. ohliger , v. nesme , d. gross , and j. eisert , arxiv:1111.0853 . c. schn , k. hammerer , m. m. wolf , j. i. cirac , and e. solano , phys . a * 75 * , 032311 ( 2007 ) . g. k. brennen and a. miyake , phys . * 101 * , 010502 ( 2008 ) . x. chen , b. zeng , z. gu , b. yoshida , and i. l. chuang , phys . lett . * 102 * , 220501 ( 2009 ) . f. verstraete and j. i. cirac , phys . lett . * 104 * , 190405 ( 2010 ) . t. j. osborne , j. eisert , and f. verstraete , phys . 105 * , 260401 ( 2010 ) . to find an upper bound to @xmath288-u\|\le{\varepsilon})$ ] , we first lower bound the probability that @xmath289 for some suitably chosen @xmath290 . then , for @xmath291 , we upper bound @xmath292-u\|$ ] by a polynomial in @xmath73 and use that the probability for a polynomial to be larger than @xmath141 can not grow too fast for small @xmath141 . we calculate @xmath293-u\|\ge\|\frac{a_\theta[x]}{\sqrt{p_\theta(x)}}-u\|_2/\sqrt{2}\ ] ] where @xmath294 denotes the frobenius norm . we now use ( [ eq : allgauss2 ] ) and the fact that @xmath295 where @xmath296 denotes the @xmath103th hermite polynomial . we also use @xmath297 and that @xmath298 is upper - bounded by a constant independent of @xmath203 . this allows us to find non - negative polynomials @xmath299 and @xmath300 , where the coefficients are continuous functions of @xmath95 and @xmath203 with @xmath292-u\|\ge f_{u,\theta}(x)/g_{u,\theta}(x)$ ] . as the probability density @xmath298 is of the form @xmath301 we get @xmath302 for some suitable @xmath303 , @xmath304 . now , we assume that @xmath305 . in this case , @xmath300 can be upper bounded by @xmath306 for some @xmath307 and some even @xmath308 . note that the maximal degree of all polynomials only depends on @xmath195 . assume that @xmath299 has @xmath309 zeros @xmath310 . around any of these zeros , @xmath299 can be lower bounded by @xmath311 for some @xmath312 and some even @xmath313 . using this and ( [ eq : boundlargedev ] ) , we chose some @xmath314 such that @xmath315-u\|\le { \varepsilon}\right)\le { \varepsilon}\\ & + \sum_{i=1}^m{\mathbbm{p}}_\theta\biggl(c_4(x - x_i)^s\le c_3\left(c_1\log\left(\frac{c_2}{{\varepsilon}}\right)\right)^r{\varepsilon}\biggr).\label{eq : app0}\end{aligned}\ ] ] for all @xmath316 . because @xmath298 is upper bounded , there exist a constant @xmath317 such that @xmath318 for all @xmath316 . inserting ( [ eq : app1 ] ) into ( [ eq : app0 ] ) , one obtains ( [ eq : impcontrol ] ) . as all constants depend on @xmath95 and @xmath203 in a continuous way , and @xmath95 and @xmath203 are both taken from compact domains , there exist a set of constants such that ( [ eq : impcontrol ] ) is simultaneously true for all @xmath95 and @xmath203 . thus , observation [ obs : nocontrol ] holds .

we review a geometric approach to classification and examination of quantum correlations in composite systems . since quantum information tasks are usually achieved by manipulating spin and alike systems or , in general , systems with a finite number of energy levels , classification problems are usually treated in frames of linear algebra . we proposed to shift the attention to a geometric description . treating consistently quantum states as points of a projective space rather than as vectors in a hilbert space we were able to apply powerful methods of differential , symplectic and algebraic geometry to attack the problem of equivalence of states with respect to the strength of correlations , or , in other words , to classify them from this point of view . such classifications are interpreted as identification of states with ` the same correlations properties ' i.e. ones that can be used for the same information purposes , or , from yet another point of view , states that can be mutually transformed one to another by specific , experimentally accessible operations . it is clear that the latter characterization answers the fundamental question ` what can be transformed into what _ via _ available means ? ' . exactly such an interpretations , i.e , in terms of mutual transformability can be clearly formulated in terms of actions of specific groups on the space of states and is the starting point for the proposed methods . center for theoretical physics , polish academy of sciences , al . lotnikw 32/46 , 02 - 668 warszawa , poland icfo - institut de ciencies fotoniques , the barcelona institute of science and technology , 08860 castelldefels ( barcelona ) , spain

1701.03536

quantum entanglement - a direct consequence of linearity of quantum mechanics and the superposition principle - is one of the most intriguing phenomena distinguishing quantum and classical description of physical systems . quantum correlated ( e.g. , entangled ) states of composite systems possess features unknown in the classical world , like the seemingly paradoxical nonlocal properties exhibited by the famous einstein - podolsky - rosen analysis of completeness of the quantum theory . recently , with the development of quantum information theory they came to prominence as the main resource for several applications aiming at speeding up and making more secure information transfers ( see , e.g. , @xcite ) . a novel kind of quantum correlations , called _ quantum discord _ , different from entanglement , but also absent in the macroscopic world , was discovered @xcite , @xcite adding one more element to `` mysteries of quantum mechanics '' as seen from the classical point of view . although typically a quantum system , as , e.g. , a harmonic oscillator or a hydrogen atom , is described in terms of an infinitely dimensional hilbert space , for most quantum - information applications the restriction to finite dimensions suffices , since usually the active role in information processing play only spin degrees of freedom or only few energy levels are excited during the evolution . from the mathematical point of view such finite - dimensional quantum mechanics seems to mount a smaller challenge than in the infinite - dimensional case - the tool of choice is here linear algebra rather than functional analysis . nevertheless , understanding of correlations in multipartite finite dimensional quantum systems is still incomplete , especially for systems with many components or consisting of particles governed by quantum statistics ( bosons , fermions ) @xcite . the statistical interpretation of quantum mechanics disturbs a bit the simple linear - algebraic approach to quantum mechanics - vectors of state ( elements of a finite - dimensional hilbert space @xmath2 ) should be of the unit norm . obviously , physicist are accustomed to cope with this problem in a natural way by `` normalizing the vector and neglecting the global phase '' . nevertheless , it is often expedient to implement this prescription by adopting a suitable mathematical structure , the projective space @xmath3 , already from the start . the projective space is obtained from the original hilbert space @xmath2 by identifying vectors , etc . , and the short one @xmath4 etc . for elements ( vectors ) of @xmath2 . ] differing by a scalar , complex , non - zero factor , @xmath5 . we will denote elements ( points ) of @xmath3 by @xmath6 , etc . and , if we want to identify a particular equivalence class of e vector @xmath7 , by @xmath8 $ ] , etc . obviously , both approaches , the linear - algebraic ( plus normalization and neglecting the global case ) picture and the projective one are equivalent . following the former we loose linearity , so which advantages we could expect instead ? we answered this question in our paper @xcite , where we propose , by working in the projective space , to apply completely new ( in this context ) techniques to analyze the phenomenon of entanglement . the approach has given a deeper insight into the unexpectedly rich geometric structure of the space of states and enabled the use of recently developed advanced methods of complex differential , algebraic and symplectic geometry . the most efficient characterization of quantum correlations is achieved by identifying states that are ` equally correlated ' or , in other words , states that can be mutually transformed _ via _ methods allowed by rules of quantum mechanics without destroying quantum correlations . this is the case of pure quantum operations that are _ local _ , i.e. restricted to act independently on each subsystem . as required by quantum mechanics , such operations are unitary in the relevant hilbert space @xmath2 , but form only a subgroup of the whole group @xmath9 as limited by the restriction to subsystems . for these reasons they are called _ local unitary _ ( lu ) operations . a complementary insight is provided by adding more operations , like measurements performed locally , i.e. , each restricted to a single subsystem , and possible communication among subsystems _ via _ classical means of communication . such operations are dubbed locc - _ local operations with classical communication_. they usually destroy purely quantum correlations , but in any case do not increase them , so if one state can be transformed to another one _ via _ locc transformation , the latter is not more ( and usually less ) correlated than the former . this allows quantifying correlations by constructing _ entanglement measures _ , quantities characterizing states that remain unchanged under lu operations and do not increase under locc ones . technically it is convenient to group together states related to each other by invertible stochastic local operations and classical communication ( slocc ) two states , @xmath10 and @xmath7 belong to the same slocc class if both transformations , @xmath11 and @xmath12 can be achieved by means of locc with non - zero probability . hence , in general , if two states belong to two distinct slocc classes , they have different correlation properties and might be not exchangeable for achieving the same quantum informational tasks . as mentioned above the lu operations form a subgroup @xmath13 of the whole unitary group @xmath9 of the relevant hilbert space @xmath2 ( since the global phase of a state is irrelevant , we can use the special unitary group @xmath14 instead ) . the slocc also form a group @xmath15 that happens to be the complexification of @xmath13 , i.e. , @xmath16 . in contrast to @xmath13 , which is compact as a subgroup of the unitary group , @xmath15 is not compact . the exact forms of @xmath13 and , consequently , @xmath15 depend on the problem in question ( number and dimensionality of subsystems , distinguishability of particles constituting them ) . the actions of @xmath13 and @xmath15 on @xmath2 are naturally and easily transferred to the projective space @xmath3 _ via _ @xmath17=[u\psi]$ ] . the projective space has a reach geometrical structure . in particular , in a natural way , it is a khler manifold , i.e. , a complex symplectic riemannian manifold on which all three structures are compatible . symplecticity means that it can be treated as a kind of a classical phase space for a hamiltonian system . the action of a group on such a manifold reflects symmetries of the dynamical system , and the methods of analysis of such systems can be borrowed from classical mechanics . in modern formulation of classical mechanics ( see , e.g. , @xcite , @xcite ) one employs the full power of symplectic geometry . so far the above outlined geometric treatment of correlations was restricted to pure states . there exists , however , its natural extension to mixed states . instead of the projective space @xmath3 one considers another manifold - an orbit of the whole @xmath14 group through a chosen mixed state ( i.e. , the set of all mixed density matrices obtained for the chosen one @xmath18 _ via _ by unitary conjugations , @xmath19 , @xmath20 ) , in other words all density matrices with the same spectrum . such an orbit posses a natural khlerian structure given by the so called kirillov - kostant - souriau form . this observation allows to apply the same geometric techniques as in the pure - state case . as outlined in previous sections , one of the basic problems in the theory of quantum correlations is the classification of states with respect to local operations performed independently on subsystems of a given system by two classes of such operations lu and slocc , both being groups acting on the space of states , a unitary one @xmath13 and its complexification @xmath16 , respectively . the space of pure states is the projective space @xmath3 , and for mixed states , the space of spectral ( i.e. , possessing the same eigenvalues ) density matrices . the underlying hilbert space for a composite system has a form the tensor product of the hilbert spaces of the subsystems or , in the case of indistinguishable components , an appropriate ( anti)-symmetrization of the product . in both cases of pure and mixed states the above described spaces of states spaces have a natural geometric structure of a symplectic manifolds ( see appendix for a brief description of relevant mathematical concepts used in the paper ) . for applications in the theory of quantum correlations we will consider the following particular instances . 1 . @xmath21 ( the complex projective space ) , and 1 . for pure states of @xmath22 distinguishable particles the hilbert space is the tensor product , @xmath23 , where @xmath24 is the @xmath25-dimensional hilbert space of @xmath26-th particle , the action of @xmath27 on @xmath28 is defined in the natural way in terms of the tensor product , i.e. @xmath29 , the complexified group is thus @xmath30 , and the lie algebras read @xmath31 , @xmath32 ; 2 . for pure states of @xmath22 @xmath33-state bosons @xmath34 ( the @xmath22-th symmetrized tensor power of @xmath35 ) , where @xmath35 is the hilbert space of a single boson , the group @xmath36 acts diagonally , @xmath37 , and , consequently , @xmath38 , @xmath39 , @xmath40 ; 3 . for pure states of @xmath22 @xmath33-state fermions , @xmath41 ( the @xmath22-th anti - symmetrized tensor power of @xmath35 ) , where @xmath35 is the hilbert space of a single fermion , the group @xmath36 again acts diagonally ( i.e , by the same element of the group on each factor in the anti - symmetrized product ) , @xmath38 , algebras @xmath39 , @xmath40 . isospectral density matrices @xmath42 for systems of distinguishable particles with the hilbert space and the groups @xmath13 and @xmath15 as in 1(a ) above but now acting _ via _ conjugations ( the adjoint action ) on the lie algebra @xmath43 . in fact density matrices are not exactly elements of respective lie algebras since the latter are traceless . nevertheless a shift by a constant multiple of the identity yields the desired normalization to @xmath44 , and the requirement of positivity restricts them to a subset of so ` shifted ' lie algebra ( see below ) . for @xmath3 the kirillov - kostant - souriau form providing a symplectic structure , calculated at a point @xmath45\in \mathbb{p}(\mathcal{h})$ ] reads @xmath46v \rangle}}}{2{\ensuremath{\langle v | v \rangle}}},\end{gathered}\ ] ] whereas for @xmath42 at a density matrix @xmath47 , @xmath48),\end{gathered}\ ] ] the momentum map @xmath49 for pure states of @xmath22 distinguishable particles is given by @xmath50)=\frac{i}{2}\{\rho_1([v])-\frac{1}{n_1}i_{n_1},\rho_2([v])-\frac{1}{n_2}i_{n_2},\ldots,\rho_l([v])-\frac{1}{n_l}i_{n_l}\},\end{gathered}\ ] ] where @xmath51)$ ] is the @xmath26-th reduced one - particle density matrix of a state @xmath45\in \mathbb{p}(\mathcal{h})$ ] and @xmath52 is the identity operator on the @xmath25-dimensional hilbert space @xmath24 . for pure states of @xmath22 @xmath33-state indistinguishable particles ( bosons or fermions ) it reads @xmath53)=\frac{i}{2}(\rho_1([v])-\frac{1}{d}i)$ ] , @xmath54)$ ] , and for the case of mixed states on a single orbit @xmath55 by the formula similar to ( [ dist ] ) but with @xmath56 being now the reduced one - particle density matrices of a state @xmath57 . in all above considered cases the momentum map relates a state to its one - particle reduced density matrices . this observation paves a way of relating the local unitary equivalence to the properties of the reduced states or , in other words , to properties of the momentum map . since @xmath13 is a compact group its orbits are closed . to check if two states belong to the same orbit it suffices to find the ( finitely generated @xcite ) ring of @xmath13-invariant polynomials , i.e.polynomials that are constant on orbits of @xmath13 , since each orbit is uniquely determined by the values of all generating polynomials in an an arbitrary point of it . the equivariant momentum map , @xmath58 , maps @xmath13-orbits in the space of states @xmath59 onto orbits of the adjoint action of @xmath13 in @xmath43 . consequently , each @xmath60-invariant polynomial @xmath61 on @xmath43 , when composed with @xmath62 , gives a @xmath13-invariant polynomial @xmath63 on @xmath59 . the invariant polynomials for the adjoint action of @xmath64 are generated by traces of powers not larger than @xmath65 of @xmath66 . combining this with the fact that the momentum map is given by reduced one - particle density matrices we conclude that traces of their powers are @xmath13-invariant polynomials on @xmath59 . if the pre - image of every adjoint orbit from @xmath67 consists of exactly one @xmath13-orbit in @xmath59 , a @xmath13-orbit on @xmath59 ( i.e. , a set of equally correlated states ) can be identified by the traces of powers of the reduced one - particle density matrices . since the traces of powers of a matrix determine its spectrum ( and _ vice versa _ ) we can decide upon local unitary equivalence of states by examining the spectra of their reduced one - particle density matrices , hence , in principle by measurements restricted performed independently in each laboratory only . alas , the above described situation is quite exceptional . typically many @xmath13-orbits are mapped onto one adjoint orbit . hence , even if one - particle density matrices of two states have the same spectra we need additional @xmath13-invariant polynomials to decide their @xmath13-equivalence @xcite , @xcite . the number of the additional polynomials characterizes , in a certain sense , the amount of additional information that can not be obtained from local measurements , but needs to be inferred form non - local ones , i.e. , involving effectively more than one subsystem . the whole problem can be looked upon as a quantum version of the classical ` marginal problem ' , where we try to recover a probability distribution from its marginals ( here , recover a quantum state from its reduced density matrices ) @xcite . let us , however , concentrate for a moment on cases ( exceptional , as it follows from the above remarks ) , when we infer local unitary equivalence upon examining the spectra of the reduced states . as pointed above , the situation when the set of @xmath13-invariant polynomials on @xmath59 is given by the composition of @xmath68-invariant polynomials on @xmath43 with the momentum map @xmath58 occurs only when the pre - image of every adjoint orbit from @xmath69 is exactly one @xmath13-orbit in @xmath59 . let us thus consider the fibre of the momentum map @xmath62 over @xmath70 , @xmath71 i.e. the set of all points in @xmath59 that are mapped to the point @xmath72 by the momentum map . the position of @xmath73 with respect to the orbit @xmath74 is of the key importance for the above situation . orbits of @xmath13 in @xmath59 are in 1 - 1 correspondence with the adjoint orbits in @xmath75 if and only if the fibres @xmath76 are contained in orbits @xmath74 , since in this case all points mapped to @xmath72 are on the same orbit ( that can be thus identified , when we know @xmath72 ) . it is easy to see that the tangent space @xmath77 at @xmath78 is @xmath79-orthogonal to the tangent space @xmath80 at the same point @xmath78 , i.e. , for each @xmath81 and @xmath82 we have @xmath83 . indeed , let @xmath84 , @xmath85 , be a basis in the lie algebra @xmath86 . the corresponding vector fields @xmath87 at @xmath88 ( c.f . eq . ( [ eq : fundvfield ] ) in the appendix ) span the tangent space @xmath89 to the orbit through @xmath88 at @xmath88 . on the other hand , the fiber @xmath73 is a common level set of the functions @xmath90 , @xmath91 which means that the derivatives of @xmath90 vanish on the tangent to @xmath73 , @xmath92 for @xmath93 . define now the kernel of @xmath94 , @xmath95 from the definition of the momentum map ( see appendix ) we have @xmath96 hence @xmath97 if and only if @xmath98 is @xmath79-orthogonal to all @xmath87 . since @xmath87 span @xmath89 we have @xmath99 where by @xmath100 we denote the space of @xmath79-orthogonal vectors to @xmath101 . combining this with the observation above that @xmath102 vanish on @xmath103 , we find that @xmath104 and , consequently , @xmath105 . the above reasoning clearly does not depend on the choice of a particular point in @xmath73 , i.e. , as announced , @xmath106 a submanifold @xmath107 of a symplectic manifold @xmath59 is called coisotropic if for arbitrary @xmath108 we have @xmath109 . we conclude thus that if @xmath110 is coisotropic then @xmath111 . indeed from ( [ fact1 ] ) and the coisotropy of @xmath110 at each @xmath112 we have @xmath113 . hence , in this case examining whether some @xmath78 belongs to @xmath110 ( and , consequently whether @xmath78 and @xmath88 are lu - equivalent ) reduces to checking whether their corresponding one - particle reduced states have the same spectra @xcite . in order to characterize all systems for which lu - equivalence can be decided using the spectra of one - particle density matrices we need to identify those whose ( at least ) generic @xmath13-orbit is coisotropic . as it turns out , such systems need to satisfy some group theoretic conditions @xcite . to reach the final solution we have to study not only the action of @xmath13 but also of its complexification @xmath114 on @xmath59 . note that the group @xmath15 is much bigger than @xmath13 and therefore the number of @xmath15-orbits is smaller than number of @xmath13-orbits in @xmath59 . if @xmath15 has an open dense orbit on @xmath59 then we call @xmath59 _ almost homogenous manifold _ @xcite . such @xmath59 has a finite number of @xmath15-orbits . the almost homogenity of @xmath59 with respect to @xmath15-action is a necessary condition for deciding @xmath13-equivalence on @xmath59 using the momentum map @xcite . it is , however , not a sufficient condition , what can be seen from the example of the three - qubit system , where there are exactly @xmath115 orbits of @xmath116 action , but states @xmath117 and @xmath118 , where @xmath119 is an orthonormal basis in @xmath120 , satisfy @xmath121 but are not @xmath13-equivalent as they belong to different @xmath15-orbits @xcite . an important role in the formulation of the sufficient condition is played by the borel subgroup of the group @xmath15 . by definition a borel subgroup @xmath122 is a maximal connected solvable subgroup of the group @xmath15 . for example , for @xmath123 , the group of upper - triangular matrices , that is a stabilizer of the standard full flag in @xmath124 , ( i.e. , a collection of subspaces with the dimension growing by one , such that the given one includes all subspaces with smaller dimensions ) , @xmath125 where @xmath126 is a basis in @xmath2 , is an example of a borel subgroup . generally , any two borel subgroups are conjugated by an element of @xmath15 . therefore , in the considered example , @xmath122 is a borel subgroup of @xmath15 if and only if when it stabilizes some standard full flag . the crucial notion for the @xmath13-equivalence problem is the notion of a _ spherical space_. if @xmath15 is a group and @xmath127 its subgroup we call @xmath128 , i.e. , the space obtained from @xmath15 by identifying points connected by elements of @xmath127 , a _ g - homogenous space_. in this space the group @xmath15 acts in a natural way , simply by the group product followed by identification of elements connected by elements of @xmath127 , hence we can consider actions of @xmath15 or its subgroups on @xmath129 , in particular of a borel subgroup . this leads to the following definition . a @xmath15-homogenous space @xmath128 is a _ spherical homogenous space _ if and only if some ( and therefore every ) borel subgroup @xmath130 has an open dense orbit in @xmath129 . now observe that an orbit of @xmath15 in the initially considered manifold @xmath59 through a point @xmath88 can be identified with some @xmath15-homogenous space . indeed we can take as @xmath127 the subgroup that stabilizes @xmath88 , than all other points on the orbit are obtained by actions of @xmath15 that effectively move @xmath88 . all these considerations allow the following definition . if @xmath15 has an open dense orbit @xmath128 in @xmath59 and @xmath129 is a spherical homogenous space then @xmath59 is called a _ spherical embedding _ of @xmath128 . such @xmath59 is also called _ almost homogenous spherical space _ ( with respect to the action of @xmath15 ) . its relevance stems from the following brion s theorem @xcite , [ brion result](brion ) let @xmath13 be a connected compact lie group acting on connected compact khler manifold @xmath131 by a hamiltonian action and let @xmath132 . the following are equivalent 1 . @xmath59 is a spherical embedding of the open @xmath15-orbit . 2 . for every @xmath133 the fiber @xmath73 is contained in @xmath74 . in other words , the second point above assures that the momentum map separates all @xmath13-orbits , i.e. , from @xmath134 it follows @xmath135 . in our case the group @xmath13 is the local unitary group , i.e. , the group of the product of the local unitaries . the conclusion is that if the relevant @xmath136 ( the space of states of the whole system ) is a spherical embedding of an open orbit of the corresponding group of the product of the local special linear groups ( i.e. the complexification of the local unitary group ) then lu - equivalence can be decided upon spectra of the reduced one - particle states . as it should be clear from the above presented reasoning we thus need to find open dense orbit of @xmath15 in @xmath3 and show that @xmath3 is a spherical embedding of it ( in short , that the action of @xmath15 on @xmath3 is _ spherical _ ) . in @xcite we use brion s theorem and show that open dense orbits of the borel subgroup exist for systems of two fermions , two bosons and two distinguishable particles in arbitrary dimensions . therefore , for such system the lu - equivalence can be decided using reduced one - particle density matrices . from the previous section we conclude that typically we are confronted with undecidability of local unitary equivalence upon examining the spectra of reduced one - particle density matrices . the problem can be be looked upon from another point of view , using tools borrowed from algebraic geometry , by exhibiting causes for such a situation . in particular , one can show that an obstacle is the existence of so - called exceptional states . to describe them we define the rank of a state @xcite . to this end we consider the images of three maps corresponding to the cases of distinguishable particles , bosons and fermions , respectively , 1 . the segre map , @xmath137 , @xmath138 , ... , [ v_l ] ) \lo [ v_1\otimes ... \otimes v_l ] $ ] . 2 . the veronese map , @xmath139 , @xmath140 \lo [ v^{\otimes l}]$ ] . 3 . the plcker map , @xmath141 , where @xmath142 is the grassmanian , i.e. , the space of all @xmath22-dimensional subspaces of @xmath35 , @xmath143 $ ] . the rank of a state with respect to @xmath144 is defined as @xmath145 = { \rm rk}_\xx[\psi ] = \min \{r\in\nn : \psi={x}_1+\dots + { x}_r , \ [ x_j]\in\xx\}\,,\ ] ] where @xmath146 is the image of the corresponding map ( 1.-3 . above ) . in words , it means that the rank is the minimal number of states needed to obtain the state in question from states in @xmath144 . the sets of states of rank @xmath147 will be denoted by @xmath148\in\pp(\mc h ) : { \rm rk}[\psi ] = r \}$ ] . they are not closed and it turns out that there are states in @xmath149 of a certain rank @xmath147 that can be approximated with an arbitrary precision by states of a lower rank . an example of an exceptional state is a 3-qubit @xmath150-state of rank 3 , @xmath151 . it can be obtained as the limit of a sequence of rank-2 states obtained from the ghz state , @xmath152 . indeed in @xmath153 we have , @xmath154 \stackrel{a\rw 0}{\lw}[w ] , \quad a(a ) = \frac{1}{\sqrt{2 } } \begin{pmatrix } a & a \\ -a^{-1 } & a^{-1 } \end{pmatrix}\ ] ] it was proved @xcite that their existence is an obstacle preventing from inferring the local unitary equivalence of states from the image of the momentum map , i.e. , from the spectra of the reduced one - particle density matrices . the main theorem of @xcite reads : [ theo spher<->no except ] suppose that we have one of the following three configurations of a state space @xmath155 , a complex reductive lie group @xmath15 acting irreducibly on @xmath155 , and a variety of coherent states @xmath156 , which is the unique closed @xmath15-orbit in the projective space @xmath153 . \(i ) @xmath157 , @xmath158 , @xmath159 . \(ii ) @xmath160 , @xmath161 , @xmath162 . \(iii ) @xmath163 , @xmath161 , @xmath164 . then the action of @xmath15 on @xmath165 ( resp . @xmath166 on @xmath167 ) ) is spherical if and only if there are no exceptional states in @xmath168 ( resp . @xmath169 ) with respect to @xmath146 . in other words , sphericity of the representation is equivalent to the property that states of a given rank can not be approximated by states of lower rank . combining this theorem with the results described in the previous section we infer that the existence of exceptional states is an obstacle for deciding lu - equivalence using the momentum map i.e. , upon reduced spectra . as already mentioned above , information about the spectra of the reduced one - particle density matrices does not allow to decide whether two states of the whole system belong to the same orbit of the local unitary transformation , i.e. whether they have the same correlation properties . the condition @xmath170)=\mu(k.[\phi])$ ] , where @xmath171 $ ] denotes the orbit in @xmath3 of @xmath13 through @xmath8 $ ] , etc . , is still a necessary one for the @xmath13-equivalence of states @xmath172 $ ] and @xmath8 $ ] . the image of the momentum map , @xmath75 consists of adjoint orbits in @xmath173 . each adjoint orbit intersects the cartan subalgebra @xmath174 ( maximal commutative subalgebra of @xmath173 at a finite number of points . this statement expresses the fact that each matrix from @xmath173 can be diagonalized by the adjoint action ( conjugation ) of an element of the group @xmath13 ( just like a ( anti-)hermitian matrix can be diagonalized by some unitary transformation ) . the cartan subalgebra , in this case , is represented by diagonal matrices , but the eigenvalues ( diagonal elements at the point of intersection ) are ordered in some specific way . by permutations we can order them differently , reaching in this way some other point of intersection . thus different intersection points are connected by the action of some subgroup of the permutation group - the weyl group . to make the situation unambiguous we impose a concrete order , for example nonincreasing , which means that we choose the intersection point that belongs to a subset of diagonal matrices ( i.e. , elements of @xmath175 with nonincreasing diagonal entries ) , called the positive weyl chamber and denoted by @xmath176 . now we can define @xmath177 to be the map satisfying @xmath178)=\mu(k.[\phi])\cap\mathfrak{t}_{+}$ ] . it assigns to a state @xmath172 $ ] the ordered spectra of the ( shifted ) one - particle reduced density matrices . by taking the intersection of the whole image of the momentum map , @xmath75 , with the positive weyl chamber , @xmath176 , one obtains the set @xmath179 parametrizng the orbits by elements of @xmath180 @xcite . the convexity theorem of the momentum map @xcite states that @xmath181 is a convex polytope , referred to as the kirwan polytope . the necessary condition for states @xmath182 $ ] and @xmath183 $ ] to be @xmath13-equivalent , can be therefore formulated as @xmath184)=\psi([\phi_2])$ ] . to decide whether they are really equivalent we need to inspect additional invariants . eg . , additional invariant polynomials . it is shown in @xcite that for @xmath22-qubit states satisfying the necessary condition , the number of additional invariant polynomials strongly depends of the spectra of the one - qubit reduced density matrices , i.e. , on the point in the polytope @xmath181 . for @xmath185 , the number of additional polynomials is given by by the dimension of the reduced space @xmath186 . in @xcite @xmath187 was analyzed for arbitrary @xmath185 . for multi - qubit systems the inequalities describing the polytope @xmath181 are known @xcite . denote by @xmath188 an increasingly ordered spectrum of the @xmath26-th reduced density matrix and by @xmath189 the shifted spectrum , @xmath190 . then , @xmath181 is given by @xmath191 and @xmath192 . methods used in @xcite to compute the dimensions of spaces @xmath193 , are different for points @xmath194 belonging to the interior of @xmath181 and for points from the boundary of the polytope . for more than two qubits , the polytope is of full dimension , hence a generic @xmath13-orbit in the space of states @xmath59 has the dimension of @xmath13 @xcite . using the regularity of @xmath62 @xcite we get that for points @xmath194 from the interior of the polytope the dimension of the reduced space reads : @xmath195 points belonging to the boundary of @xmath181 can be grouped into three classes : ( i ) @xmath196 of @xmath197 are equal to @xmath198 , ( ii ) at least one of inequalities @xmath192 is an equality , ( iii ) @xmath196 of @xmath197 are equal to @xmath199 . in case @xmath200 , inequalities that yield @xmath181 reduce to an analogical set of inequalities for the @xmath201-qubit polytope . therefore , @xmath202 . states that are mapped to points that fall into case ( ii ) belong to the @xmath203-orbit through the @xmath22-qubit @xmath150-state , @xmath204={\mbox{$| 01\ldots1 \rangle$}}+{\mbox{$| 101\ldots1 \rangle$}}+\ldots+{\mbox{$| 1\ldots10 \rangle$}}$ ] @xcite . as it is shown in @xcite , the closure of such an orbit is an almost homogeneous spherical space . therefore , fibers of the momentum map are contained in @xmath13-orbits ( see section [ sec : spherical ] ) , i.e.@xmath205 . case ( iii ) , where @xmath196 of @xmath197 are equal to @xmath199 , is the most difficult one , as it requires the use of some more advanced tools from the geometric invariant theory ( git ) @xcite . here a key role is played by _ stable states _ @xcite , i.e.states for which @xmath206)=0 $ ] and @xmath207 @xcite . for a symplectic action of a compact group @xmath208 the existence of stable states implies that @xmath209 , where @xmath210 is the momentum map for the @xmath208-action . the strategy taken in @xcite for case ( iii ) is the following . the group @xmath13 can be divided into @xmath211 , where @xmath212 , @xmath213 and @xmath214 acts on the first @xmath196 qubits . the action of @xmath214 yields the momentum map , which assigns to a state its first @xmath196 one - qubit reduced density matrices . therefore , @xmath215 consists of states whose first @xmath196 reduced density matrices are maximally mixed , while the remaining @xmath201 reduced matrices are arbitrary . further one constructs a state that is git stable with respect to the action of @xmath216 on @xmath3 . hence , @xmath217 . furthermore , the quotient @xmath218 is a symplectic variety itself . because the actions of @xmath214 and @xmath219 commute , we can consider the action of @xmath219 on @xmath218 . the momentum map for @xmath220 acting on @xmath218 gives the remaining @xmath221 one - qubit reduced density matrices . the polytope of @xmath222 is of full dimension , i.e.of dimension @xmath221 . by the formula for the dimension of the reduced space for points from the interior of the polytope , we get : @xmath223 which is the desired result for the case ( iii ) . the three parts of the boundary of @xmath224 for four qubits . the numbers denote @xmath225 . if the number is missing , then @xmath226 . taken from @xcite.,width=529 ] a classification of states with respect to slocc @xcite can be , as already mentioned , treated as a complementary one to the lu - equivalence characterization of quantum correlations . for reasons briefly explained below such a classification is not an easy task and its intricacies are still not fully understood . to a large extend the arising problems were fairly exhaustively explained in @xcite . for the considered multipartite systems reversible slocc operations correspond to elements of the complexification @xmath227 of the local unitary group @xmath13 , and two states are slocc equivalent if and only if they belong to the same @xmath15-orbit . recall that the problem of @xmath13-equivalence is solvable by means of @xmath13-invariant polynomials . as the group @xmath15 is reductive , ( which means that it is a complexification of its maximal compact group - in this case the compact group @xmath13 ) , the hilbert - nagata theorem @xcite ensures that the ring of @xmath15-invariant polynomials is finitely generated , just like in the compact case of the local unitary group @xmath13 . nevertheless , the problem of @xmath15-equivalence turns out to be significantly different from the problem of @xmath13-equivalence.the essence of this difference is the fact that the group @xmath15 is not compact and thus @xmath15-orbits do not have to be closed . for two vectors @xmath228 and @xmath229 on two non - intersecting orbits , the closures of the orbits can intersect . since @xmath15-invariant polynomials are continuous functions , they are not able to distinguish between orbits @xmath230 and @xmath231 in such a case . it is only possible to distinguish between orbits whose closures have non - empty intersection , in particular between closed @xmath15-orbits . in purely topological language this ` pathology ' can be linked to the fact that the orbit space @xmath232 , i.e. , the quotient space identifying states on the same orbit , is not longer a hausdorff space - not every pair of points is separated by open sets . the @xmath15-equivalence of states is thus intimately linked to the structure of the orbit space resulting from the action of a non - compact reductive group on a vector space @xmath28 ( equivalently on the projective space @xmath233 ) . two orbits of @xmath234 and @xmath235 in @xmath2 are called @xmath236-equivalent if and only if there exists a sequence of orbits @xmath237 such that the closures of each two consecutive ones intersect , @xmath238 . the relation of @xmath236-equivalence divides @xmath15-orbits into equivalence classes ( @xmath236-classes ) . it turns out that every @xmath236-class contains exactly one closed @xmath15-orbit contained in the closure of every @xmath15-orbit belonging to the considered @xmath236-class . the equivalence classes are thus parametrized by closed @xmath15-orbits and @xmath15-invariant polynomials distinguish between @xmath15-orbits belonging to different @xmath236-classes @xcite . among all @xmath236-classes we distinguish those corresponding to the zero vector - they form the so - called null cone @xcite . this class must be removed if we want to consider the quotient space at the projective level . after removing from the projective space @xmath233 points corresponding to vectors from the null cone we are left with so called _ semistable _ points , @xmath239 . two points @xmath240 are @xmath236-equivalent if there are vectors of @xmath241 , such that @xmath242 $ ] and @xmath243 $ ] and on the level of the hilbert space @xmath244 . the quotient space obtained from the semistable points by @xmath236-equivalence relation is denoted by @xmath245 and is a projective algebraic variety . it is known in the literature under the name _ git quotient _ @xcite and we will call it _ git space_. points of the git space correspond to @xmath236-classes of semistable points and are in one - to - one correspondence with closed @xmath15-orbits . it turns out that every closed @xmath15-orbit in @xmath246 contains exactly one @xmath13-orbit from @xmath247 @xcite . therefore we get the following equivalence @xmath248 , giving thus a description the git space in terms of the momentum map ( see figure 2 ) [ fig : git ] . taken from @xcite.,title="fig : " ] the set of closed @xmath15-orbits is given by the action of @xmath15 on @xmath247 , i.e. @xmath249 . among the semistable points we distinguish the so - called stable points @xmath250 @xcite . the existence of a single stable point makes @xmath251 an open dense subset of @xmath252 , i.e. almost every semistable point is stable @xcite . for the stable point @xmath253 the @xmath236-equivalence class consists of exactly one closed @xmath15-orbit . for semistable but not stable points this class always consists of an infinite number of @xmath15-orbits . vectors belonging to the null cone , i.e. @xmath236-class whose closed @xmath15-orbit is the zero vector , may represent important states from the point of view of quantum correlations . for example , the w - state @xmath254 and separable states belong to the null cone but their quantum properties are significantly different . therefore we need a finer procedure dividing @xmath15-orbits , one that includes the git construction and also provides mathematically and physically well - defined stratification of the null cone . a key role is played here by the function @xmath255 , i.e. , the norm of the momentum map . it has a clear mathematical and physical interpretation . according to the definition given by klyachko @xcite , the total variance of state @xmath45\in\mathbb{p}(\mathcal{h})$ ] with respect to the symmetry group @xmath256 is given by @xmath257)=\frac{1}{{\ensuremath{\langle v | v \rangle}}}\left(\sum_{i=1}^{\dim k}{\mbox{$\langle v |$}}\xi_{i}{}^{2}{\mbox{$| v \rangle$ } } -\frac{1}{{\ensuremath{\langle v | v \rangle}}}\sum_{i=1}^{\dim k}{\mbox{$\langle v |$}}\xi_{i}{\mbox{$| v \rangle$}}^{2}\right)=c-4\cdot\left\vert \mu\right\vert ^{2}([v ] ) , \label{eq : variance}\end{gathered}\ ] ] where @xmath258 form an orthonormal basis of algebra @xmath43 and @xmath236 is a @xmath45$]-independent constant . the function @xmath259)$ ] can be also expressed as the expectation value of the casimir operator @xcite , @xmath260 for the irreducible representation of @xmath13 on the symmetrized tensor product @xmath261 @xcite . in this case we have @xmath262 and @xmath263\right)$ ] . finally , @xmath264 is directly related to the linear entropy , which is a linear function of the total variance . a point @xmath45\in \mathbb{p}([v])$ ] is a critical point of @xmath264 if it is a solution of @xmath53).v=\lambda v$ ] @xcite . critical points of @xmath264 can be therefore divided into two categories . the first includes all @xmath13-orbits belonging to @xmath247 . these are called minimal critical points and for them @xmath264 reaches a global minimum . the minimal critical points correspond to states with maximum total variance and maximum linear entropy . the other critical points are given by some @xmath13-orbits in the null cone . for these points @xmath53)\neq 0 $ ] and @xmath53)v=\lambda v$ ] . therefore , in the null cone we distinguish @xmath15-orbits passing through the critical @xmath13-orbits . the relationship between critical points of @xmath264 , @xmath236-equivalence and git construction becomes clear if we consider the gradient flow of @xmath265 @xcite . the gradient of @xmath265 is well defined as the projective space @xmath266 is a khler manifold , and therefore is equipped with a well defined metric determined by its riemannian structure ( see appendix ) . the gradient flow is tangent to @xmath15-orbits and carries points towards critical @xmath13-orbits . two points @xmath267 are equivalent from the point of view of the gradient flow if they are taken by it to the same critical @xmath13-orbit . this definition is consistent with the @xmath236-equivalence definition . however , it is at the same time more general because it allows an extension of the concept of equivalence to the null cone . the situation in the null cone is more complex as the critical @xmath13-orbits do not need to be in one fiber of @xmath268 ( recall that @xmath269 is given by @xmath270 ) . nevertheless , the polytope @xmath271 has a finite number of points @xmath272 for which @xmath273 contains critical @xmath13-orbits . let @xmath274 denote the set of critical @xmath13 orbits mapped by @xmath268 on @xmath275 and @xmath276 be the set of all the points that are taken by gradient flow of @xmath265 to @xmath274 . the quotient space @xmath277 , obtained from @xmath276 by dividing @xmath276 by the equivalence relation induced from the gradient flow and the space @xmath278 are isomorphic algebraic varieties ( see figure 3 ) . and @xmath274 , with two exemplary critical @xmath13-orbits , @xmath279 and @xmath280 . the arrows represent the gradient flow of @xmath281.taken from @xcite.,title="fig : " ] the above described construction is thus analogous to the git one . for @xmath282 we get that @xmath283 and @xmath284 . using the so defined equivalence relation we can think of a quotient space @xmath3 by @xmath15 , i.e , the space identifying points on the same @xmath15-orbit , as of the space consisting of a finite number of projective algebraic varieties : @xmath285 in the above formula we abused slightly the notation writing @xmath286 , since , as explained above this is not a ` good quotient ' from the point of view of the @xmath15-action . the map @xmath268 has another important property , namely not only @xmath271 is a convex polytope but also the image of every @xmath15-orbit @xmath287 has this property @xcite . a finite number of varieties @xmath288 is the result of the fact that @xmath276 can be equivalently defined as those @xmath289 for which polytopes @xmath287 share the nearest point to the origin . but the momentum map convexity theorem for @xmath15-orbits ensures that the number of such polytopes is finite @xcite , so the number of manifolds @xmath274 is also finite . summarizing , we obtained the correspondence given in table 1 . .a dictionary [ cols="<,<",options="header " , ] the critical points of @xmath290 , or of the linear entropy , play a key role in understanding the generalized slocc classes . finding the critical states by direct application of the definition is a computationally difficult task , as it requires solving the equation @xmath53).v=\lambda v$ ] , i.e. finding eigenvectors and eigenvalues of a matrix depending nonlinearly on a vector it acts on , rendering thus seemingly straightforward eigenvectors - eigenvalues problem effectively nonlinear . in @xcite a slightly more tractable method on an interplay between momentum maps for abelian and non - abelian lie groups was proposed . for a compact group @xmath13 , we denote by @xmath291 its maximal torus , which is a maximal compact and connected abelian subgroup . for example , when @xmath292 , a maximal torus consists of unitary diagonal matrices with the determinant one . a momentum map @xmath293 for the action of @xmath291 on @xmath59 is given by the composition of @xmath58 with the projection on the cartan subalgebra @xmath294 . therefore , we have @xmath53)=\mu_t([v])+\alpha$ ] , where @xmath295 . by the convexity theorem , @xmath296 is a convex polytope . for abelian groups , the convexity theorem specifies the vertices of the polytope @xcite . the vertices are given by the set of weights @xmath297 , where @xmath298 are the fixed points for the action of @xmath291 on @xmath59 is fixed by the action of @xmath291 iff @xmath299 . the critical points of @xmath300 , must satisfy a similar condition as the critical points of @xmath264 , i.e. @xmath301).v=\lambda v$ ] . therefore , for @xmath302 a point @xmath45 $ ] is a critical point iff @xmath303 , i.e. @xmath45 $ ] is a fixed point for @xmath304 and @xmath301)=\beta$ ] . the fixed - point set @xmath305 for the action of @xmath306 is not a symplectic variety . however , weakening the definition of @xmath305 by demanding points @xmath45 $ ] to satisfy @xmath307),\beta\rangle=\langle\beta,\beta\rangle$ ] instead of being mapped to @xmath308 , we get a symplectic variety @xmath309 . points from @xmath309 are sent by @xmath310 to the hyperplane that is perpendicular to @xmath308 and contains @xmath308 @xcite . clearly , @xmath311 . the set @xmath309 is a @xmath291-invariant symplectic variety , hence , by the convexity theorem , we have that @xmath312 is a convex polytope , which is spanned by a subset of weights from @xmath313 . the definition of @xmath309 implies that @xmath308 is the closest to zero point of this polytope . in other words , @xmath45\in m$ ] is a critical point of @xmath300 iff it is mapped to a minimal convex combination of weights , @xmath308 , and @xmath45\in z_\beta$ ] @xcite . is the image of state @xmath314 , points @xmath315 correspond to the biseparable states and @xmath316 . taken from @xcite . ] function @xmath264 is @xmath13-invariant , therefore we can restrict our consideration to critical points satisfying @xmath53)\in \mathfrak{t}_+$ ] . for such states we have @xmath53)=\mu_t([v])$ ] and @xmath45 $ ] is a critical point of @xmath264 iff it is a critical point of @xmath300 . let us denote by @xmath317 the set of all minimal combinations of weights from @xmath313 that belong to @xmath176 . then , a state @xmath45 $ ] is a critical one if and only if @xmath53)\in \mathcal{b}$ ] and @xmath45\in z_\beta$ ] . critical sets are therefore of the form @xmath318 , where @xmath319 . in @xcite the above reasoning was applied to compute the critical points of the linear entropy for pure states of @xmath22 qubits . set @xmath313 is the image under @xmath62 of the basis states @xmath320 , where @xmath321 , hence @xmath322 . we discuss the algorithm of finding the minimal combinations of weights and list the results up to @xmath323 ( the construction of the set of minimal combinations of weights is shown on figure 4 ) . we also show that for @xmath324 , the set @xmath325 , where @xmath326 is spanned by the basis states whose weights span @xmath308 . moreover , we show when sets @xmath327 are nonempty and for each @xmath328 we describe a construction of a state that is mapped to @xmath308 . we conclude that the number of critical values of the linear entropy grows super - exponentially with @xmath22 . in previous sections we discussed applications of the momentum map in two significant problems of the theory of quantum correlations , lu and slocc equivalence of pure states . in @xcite it is shown how methods of symplectic and algebraic geometry can be applied to some concrete problems involving mixed states , namely to exhibit geometric and topological aspects of of quantum correlations for separable ( non - entangled ) states . the existence of quantum correlations for multipartite separable mixed states can be regarded as one of the most interesting quantum information discoveries of the last decade . in 2001 ollivier and urek @xcite and independently henderson and vedral @xcite introduced the notion of quantum discord as a measure of the quantumness of correlations . quantum discord is always non - negative @xcite . the states with vanishing quantum discord are called _ states_. they form the boundary between classical and quantum correlations @xcite . bipartite pointer states can be identified with the so - called classical - quantum , @xmath329 states @xcite . an important subclass of @xmath329 states are classical - classical , @xmath330 states @xcite . for @xmath331 , where @xmath332 and @xmath333 a state is @xmath330 if it can be written as @xmath334 , where @xmath335 is an orthonormal basis in @xmath336 and @xmath337 in @xmath338 . a state @xmath18 is a @xmath329 state if it can be written as @xmath339 , where @xmath340 are the density matrices on @xmath338 . both classes are of measure zero in @xmath341 @xcite . for pure states , separable states are exactly zero - discord states . it was shown in @xcite that pure separable states are geometrically distinguished in the state space and belong to the unique symplectic @xmath13-orbit in @xmath233 . for mixed states , already for two particles it is easy to see that there are infinitely many symplectic @xmath13-orbits and there are separable states through which @xmath13-orbits are not symplectic . thus a simple extension of the results of @xcite , even for two - particle mixed states is not possible . in @xcite four facts concerning geometric and topological characterizations of @xmath330 and @xmath329 states are shown . they extend results of @xcite to mixed states : ( 1 ) the set of @xmath329 states is the closure of all symplectic orbits of @xmath342 action , ( 2 ) the set of @xmath330 states is the closure of all symplectic orbits of @xmath343 action , ( 3 ) the set of @xmath329 states is exactly the set of @xmath342 orbits whose euler - poincar characteristics @xmath344 do not vanish , ( 4 ) the set of @xmath330 states is exactly the set of @xmath343 orbits whose euler - poincar characteristics @xmath344 do not vanish . the space of all density matrices is not a symplectic space ( the symplectic form is degenerate ) . nevertheless , as already mentioned in the introduction ( see also appendix ) , the set of density matrices with the fixed spectrum @xmath42 , which is the adjoint orbit of @xmath345 through @xmath18 is symplectic . therefore , the action of the above given groups @xmath13 on @xmath42 leads to existence of the momentum map @xmath346 @xcite . in order to check if a given orbit @xmath347 ( the action of @xmath13 on @xmath348 is the adjoint action ) is or is not symplectic it is enough to consider the restriction of the momentum map @xmath62 to @xmath347 . then @xmath347 is symplectic if this restriction is bijective . the computational conditions for @xmath62 to be bijective are given in the kostant - sternberg theorem @xcite . let us note that since @xmath349 is mapped by @xmath62 onto adjoint orbit in @xmath43 , non - symplecticity of @xmath349 ( the degeneration of symplectic form on @xmath349 ) can be measured as @xmath350 . for two qubits the @xmath330 states , in a fixed basis , form a 3-dimensional simplex and therefore it is possible to see how the closure of the symplectic @xmath351 orbits forms the set of @xmath330 states ( figure . 5 , 6 and 7 ) . in @xcite we also discuss existence of khler structure and show that it is present on all considered symplectic @xmath13-orbits . for finding euler - poincar characteristics @xmath344 we use the hopf - samelson theorem @xcite . this theorem says that for action of a compact group @xmath13 on a manifold @xmath59 the euler - poincar characteristics @xmath344 of the orbit @xmath352 passing through @xmath133 is given by : 1 . if the maximal torus @xmath291 of @xmath13 is contained in @xmath353 then @xmath354 , where @xmath355 and @xmath356 are weyl groups of @xmath13 and @xmath353 respectively . 2 . otherwise , @xmath357 . in @xcite it is shown that orbits of the discussed groups through @xmath330 and @xmath329 states are the only orbits with stabilizer subgroups containing maximal torus . the ranks of the weyl groups are calculated and a formula for @xmath344 is given . states of two qubits . the large dot : @xmath358 , the dotted lines : @xmath359 , elsewhere : @xmath360 . taken from @xcite.,width=302 ] for orbits through @xmath330 states of two qbits . the thick dashed line : @xmath361 , the lined surfaces : @xmath362 , elsewhere : @xmath363.taken from @xcite.,width=302 ] for orbits through @xmath330 states of two qbits . the thick dashed line : @xmath364 , the lined surfaces : @xmath365 , the dotted lines and elsewhere : @xmath366.taken from @xcite.,width=302 ] one of the basic problems in the theory of quantum correlations is the classification of states with respect to local operations performed independently on subsystems of a given system . two such classes are particularly interesting , ( 1 ) local unitary ( lu ) operations , and ( 2 ) slocc - stochastic local operations with classical communication . mathematically , these operations correspond to the action of some compact group @xmath367 in case ( 1 ) and its complexification @xmath16 in case ( 2 ) . the space of pure states ( after neglecting the global phase ) is the projective space @xmath3 , and for mixed states , the space of isospectral density matrices is an adjoint orbit of the unitary group @xmath14 action . in both cases , these spaces have a natural geometric ( khler ) structure , and therefore in particular they are symplectic manifolds . since the action of compact group on @xmath59 preserves the symplectic structure there exists the momentum map . in the cases considered here , it assigns to a state of @xmath22 particles its reduced one - particle density matrices , and therefore is directly related to partial trace over @xmath368 particles . this observation opened new possibilities for the analysis employing tools and methods of symplectic and algebraic geometry . we showed how such an approach can be applied to analyze / solve the following problems . * when is information contained in one - particle reduced density matrices sufficient to solve the problem of lu - equivalence ? * what are obstacles in the case when such an information is insufficient , i e. , when the lu - equivalence can not be established upon examining spectra of reduced matrices ? * how many additional invariant polynomials ( except those directly derived from one - qubit density matrices ) are needed to solve the lu - equivalence problem ? how his number depends on the spectra of reduced matrices ? * how to classify states under slocc operations ? such a classification should be effective , use straightforwardly calculated quantities and yield a finite number of generalized slocc classes . * how to characterize geometrically mixed states with zero discord , more specifically of @xmath330 and @xmath329 states ? the proposed methods are , in principle applicable , to distinguishable as well as indistinguishable particles , but concrete results for the latter are scarce what shows a direction for further studies . another direction of future research may concentrate on looking for other ways to identify critical orbits important for slocc equivalence , for example by employing other approach to stratification described in section [ sec : gita ] @xcite . it should allow moving a state to the appropriate critical orbit case a one parameter slocc subgroup and thus have a clear operational meaning since one could decide to which class a state belongs by acting on it by a one parameter family of slocc ( hence physical ) operations . a symplectic manifold is an abstract generalization of a phase - space in classical mechanics . in the following we will invoke notions known in c classical mechanics to illustrate some features stemming from this generalizations . mathematically , a symplectic manifold is a smooth manifold @xmath59 equipped with a closed , @xmath369 , and non - degenerate differential two - form @xmath79 . the nondenegeracy means that if @xmath101 is a tangent field to @xmath59 such that at each @xmath370 and each tangent vector @xmath371 at @xmath372 we have @xmath373 then @xmath101 must vanish everywhere on @xmath59 . it implies that a symplectic manifold is always even - dimensional . a _ khler manifold _ is a complex symplectic manifold equipped with additional riemannian structure @xmath374 ( a metric , i.e. positive - definite symmetric two - form ) , such that all three structures ( complex , symplectic and riemannian ) are compatible _ vis . _ , @xmath375 , or , equivalently , @xmath376 . it also means that @xmath377 is a hermitian metric i.e. , @xmath378 a symplectic manifold @xmath379 is a natural geometric structure for classical hamiltonian mechanics . a symplectic manifold is a classical mechanical phase - space equipped with the structure of poisson brackets . they are defined in terms of @xmath79 in the following way . for an ( appropriately smooth ) function @xmath380 on the phase space we define a tangent vector field @xmath381 _ via _ @xmath382 i.e. action of the two form @xmath79 on the vector field @xmath381 gives a one form ( as it should ) equal to the differential of @xmath380 . then we set for the poisson bracket of two functions @xmath380 and @xmath127 , @xmath383 the dynamics of a system is determined by a hamilton function @xmath127 _ via _ canonical ( hamilton ) equations of motion @xmath384 for an arbitrary phase - space function @xmath380 . using now the definitions of the fields @xmath385 , @xmath381 and the poisson brackets we have . @xmath386 hence the vector field @xmath385 determines at each point @xmath370 the direction in which this point moves under the dynamics generated bu the hamilton function @xmath127 . the momentum map appears always when a lie group @xmath13 acts on a symplectic manifold preserving the symplectic structure . in classical mechanics this corresponds to a situations when we have a group of canonical ( i.e. preserving the poison brackets ) symmetries . in the paper @xmath13 is always a connected compact semi - simple matrix lie group , in fact a subgroup of a unitary group . let us denote the action of @xmath13 on @xmath59 by @xmath387 , @xmath133 , @xmath388 . for each element of the lie algebra @xmath389 of the group @xmath13 we define the _ fundamental vector field _ @xmath391 where @xmath392 - smooth vector fields on @xmath59 . invoking again the classical mechanical origin of the presented concepts we may call @xmath393 a generator of the one dimensional subgroup of symmetries . the vector field @xmath394 at @xmath370 points to the direction in which the phase - space point @xmath372 moves under the action of this one - parameter subgroup . under additional conditions is semi - simple and @xmath59 has the trivial first de rham cohomology group ] fulfilled in all cases considered here , there is a well defined function @xmath395 , such that @xmath396 . upon referring to the definitions above , we see thus that @xmath397 plays a role of a hamilton function for the ` motion ' of phase - space points under the one - parameter subgroup in question , where the parameter plays a role of the time . in classical mechanics @xmath398 would be called a generating function for the one parameter group . the functions @xmath395 can be chosen to be linear in @xmath399 and thus define the unique map @xmath400 , where @xmath401 is the dual vector space of @xmath43 , i.e. the space of linear forms on the linear space @xmath43 . hence , @xmath72 is defined by @xmath402 , where @xmath403 denotes the action of a form @xmath194 on a vector @xmath404 . the function @xmath400 is is called the _ momentum map_. the group @xmath13 acts also on its lie algebra @xmath43 by the adjoint action @xmath405 . it has its natural dual on the dual space @xmath401 , the coadjoint action @xmath406 the set of points obtained by the action of a group on a manifold on a particular point @xmath372 we call the orbit of the group ( thought the point @xmath372 ) . in particular the set @xmath407 is the orbit through a point @xmath372 of the action of @xmath13 on @xmath59 , and @xmath408 - a coadjoint orbit of @xmath13 through @xmath409 . to shorten the notation we use occasionally @xmath74 to denote the orbit of @xmath13 through @xmath88 , whatever the action of @xmath13 and the manifold on which it acts are . each coadjoint orbit is equipped with a natural symplectic structure given by the so - called kirillov - kostant - souriau form . to define it let us first construct the fundamental vector field for the coadjoint action of the one - parameter subgroup of @xmath13 generated by a lie - algebra element @xmath393 at some point @xmath410 , @xmath411 now we define the kirillov - kostant - souriau form at each point @xmath194 on a coadjoint orbit by @xmath412\right\rangle,\ ] ] where@xmath413 $ ] is the lie bracket in the lie algebra @xmath43 . for a semi - simple @xmath13 the momentum map is equivariant , i.e. @xmath414 for any @xmath133 and @xmath388 . orbits of @xmath13-action on @xmath59 are therefore mapped by @xmath62 onto orbits of the coadjoint action in @xmath401 . coadjoint and adjoint orbits ( i.e. the orbits of the adjoint action of @xmath13 on its lie algebra @xmath43 ) can be identified in all cases we consider in the paper ( i.e. for @xmath13 compact as a subgroup of a unitary group ) using a non - degenerate scalar product on @xmath43 defined by @xmath415 , i.e , identifying a linear form @xmath410 with the vector @xmath399 such that @xmath416 for all @xmath417 . we use this identification throughout the paper and therefore we treat the moment map @xmath62 as a map from @xmath59 to @xmath43 rather than @xmath418 . the authors gratefully acknowledge the support of the erc grant qolaps . b. kostant and s. sternberg , symplectic projective orbits . new directions in applied mathematics , papers presented april 25/26 , 1980 , on the occasion of the case centennial celebration ( hilton , p. j. and young , g. s. , eds . ) new york : springer , 81 - 84 , 1982

we use automated surface photometry and pattern classification techniques to morphologically classify galaxies . the two - dimensional light distribution of a galaxy is reconstructed using fourier series fits to azimuthal profiles computed in concentric elliptical annuli centered on the galaxy . both the phase and amplitude of each fourier component have been studied as a function of radial bin number for a large collection of galaxy images using principal component analysis . we find that up to 90% of the variance in many of these fourier profiles may be characterized in as few as 3 principal components and their use substantially reduces the dimensionality of the classification problem . we use supervised learning methods in the form of artificial neural networks to train galaxy classifiers that detect morphological bars at the 85 - 90% confidence level and can identify the hubble type with a scatter of 1.5 steps on the 16-step stage axis of the revised hubble system . finally , we systematically characterize the adverse effects of decreasing resolution and on the quality of morphological information predicted by these classifiers . * date of this version : * _ subject headings : _ galaxies : automated morphological classification

astro-ph0110275

one of the earliest galaxy classification schemes , discussed by hubble ( 1926 ) , was based on the visual appearance of two - dimensional images . this and other early schemes , based largely on photographic images in roughly the bandpass , used global image properties such as the visually perceived bulge - to - disk ratio and the degree of azimuthal surface brightness symmetry as major classification criteria . even these early systems discussed the presence of spiral structure and the overall characteristics of the spiral arms : grand design of high or low pitch angle verses patchy or multiple arms ( later termed flocculent spirals ) . the early hubble system delineated spirals into a parallel sequence of barred and unbarred disk galaxies in a two - dimensional system . a refinement of this approach by de vaucouleurs ( 1959 ) , established a three - dimensional classification volume with the major axis being the hubble stage ( e , s , i ) and the two other axes representing family ( barred or unbarred ) and variety ( ringed or non - ringed ) . the revised hubble system ( hereafter referred to as the rhs ) was designed to describe a continuum of morphological properties . early workers in the field strove to systematize the degree of visual structures we see in galaxies ( bulge / disk ratio , spiral arms , bars , rings ) in an effort to understand the physical processes that formed these different galaxies . a useful classification system must relate members of different classes to well understood general properties of the objects being systematically classified . it has long been known ( de vaucouleurs 1977 , buta 1994 , and roberts and haynes 1994 ) that the stage axis of the hubble sequence produces smooth , strong correlations with well known global properties : color , surface brightness , maximum rotational velocity , and gas content . these measured quantities are linked directly to very important physical properties : stellar population fraction , the surface density of stars , total mass of the system , and the rate of conversion of gas to stars . as discussed by odewahn and aldering ( 1995 ) , such correlations may be smooth and significant in a mean sense , however the degree of accidental scatter in these relations is quite large . in most cases this scatter is much larger than that introduced by measurement error and is clearly cosmic in origin . the family and variety estimates of the rhs produce much less significant correlations with such properties . buta and combes ( 1995 ) , in an excellent review of these morphological features , make clear that family and variety convey information about the specific dynamical properties of disk systems : the degree of differential rotation , the presence of certain families of resonant orbits among the stellar component of the disk , and the response of the viscous disk components ( the gas ) to the global and local properties of the galaxy potential well . galaxy classification systems are generally built on the visual appearance in an image , and hence data sets based on these systems can be gathered from imaging surveys alone . the mean correlations described above are then used to draw inferences about important physical processes describing galaxies via these catalogs . in other words , morphological classification provided a `` cheap '' and direct means of acquiring large statistical samples describing galaxies . in some respects , the qualitative nature of this classification approach , and the dearth of recognized experts who classified in some established system gave the approach a less than deserved poor reputation . however , even recent comparisons among the estimates of experienced human classifiers ( naim 1995 ) show a reliable level of repeatability is achieved . with an expansive increase in the quantity and quality of galaxy images in the last decade data via hst and a host of ground - base imaging surveys , interest has returned to the field of morphological classification ( driver 1995a , b ; glazebrook 1995 ; abraham 1996 ; odewahn 1996 ) . many workers have developed automated image analysis systems that provide quantitative estimates of galaxy types ( okamura 1984 , whitmore 1984 , spiekermann 1992 , storrie - lombardi 1992 , abraham 1994 , odewahn 1995 , han 1995 , odewahn 1997 ) . in one the most physically meaningful approaches , rix and zaritsky ( 1995 ) showed in a small sample of spiral galaxies that kinematic distortions in the global velocity field of a galaxy can be linked to global asymmetry measurements in the light distribution . a discussion of novel feature spaces for identifying peculiar galaxies is presented by naim ( 1997 ) . all of these studies have produced a wealth of information about the systematic properties of galaxies , particularly in the area of comparing the low and high redshift universe ( abraham 1996 , odewahn 1996 ) . however , it must be conceded that none of these classification systems are truly morphological in nature . in general , each method uses one or more two - dimensional parameter spaces based on global image properties and produces a type estimate via mean correlations like those discussed above using some multivariate pattern classification approach ( linear parameter space divisions , principal component analysis , decision tree , artificial neural network ) . in this paper we describe a truly morphological approach to galaxy classification based on the fourier reconstruction of galaxy images and the subsequent pattern analysis based on the amplitude and phase angle of the fourier components used . this method can quantitatively detect the presence of spiral arms and bars . in addition , it provides a systematic means of describing the degree of large - scale global asymmetry in a galaxy , a property known to correlate loosely with hubble stage , but one which is also strongly linked to important physical events such as merging and tidal interaction . in section 2 we describe our machine - automated classification method . in section 3 we compare independent sets of visual estimates of morphological types in the rhs that establish a well understood set of training / testing cases . these date are comprised of a local sample from ground - based imaging and a distant sample collected from the hst archive . in section 4 we demonstrate a practical application using these data to train and assess the performance of several fourier - based galaxy classifiers . we discuss the role of systematic and accidental errors in its use caused by varying image resolution and . in section 5 we summarize the results and lay the groundwork for future morphological studies of distant galaxy populations . the two - dimensional luminosity distribution observed in most galaxies most often displays a high degree of azimuthal symmetry . this justifies the use of one - dimensional radial surface brightness profiles in describing the radial stellar surface density distribution . such profiles are then decomposed into constituent photometric components ( de vaucouleurs 1958 , de vaucouleurs and simien 1986 , kent 1986 , de jong and van der kruit 1994 ) . this profile is extracted from the galaxy image in many ways , but most often by computing mean flux density in elliptical annuli having the shape and orientation of the faint surface brightness , outer isophotes . parameterizing this profile , with a concentration index , will produce a measurement that correlates well with bulge - to - disk ratio and hence with hubble type . local departures from this overall radial symmetry in the form of high spatial frequency components ( arms , bars , rings ) form the basis for additional criteria to visual galaxy classification systems like the rhs . a method which quantitatively describes departures from azimuthal symmetry in galaxy images would seem to be a rich starting place for developing a machine automated classification system , either of the supervised learning variety ( odewahn 1992 , weir 1995 ) or the unsupervised variety ( mahonen 1995 ) . towards this end , we have adapted a moments - based image analysis technique ( odewahn 1989 ) to fit fourier components to fixed - grid azimuthal profiles in elliptical annuli to reconstruct galaxy images . one may think of this step as an optimized data compression technique for reducing the dimensionality of our ultimate pattern classification problem : the recognition and characterization of morphological features in galaxies . the technique of fourier decomposition presented here is used to quantify the two dimensional luminosity distributions of galaxies . similar approaches have been used to study the shapes of isophotes in elliptical galaxies by bender and mollenhoff ( 1987 ) , and applications to the luminosity distribution are described by lauer ( 1985 ) , buta ( 1987 ) , and ohta ( 1990 ) . with this methodology , we can quantify the amplitude and phase of the bar and spiral arm components in a galaxy image . this technique is very useful for modeling the bar luminosity distribution for systems in which a simple elliptical bar model was not sufficient , as was first demonstrated by elmegreen and elmegreen ( 1985 ) in describing the bar luminosity distributions exhibited by galaxies over a range of hubble types . artificial neural networks ( ann s ) are systems of weight vectors , whose component values are established through various machine learning algorithms . these systems receive information in the form of a vector ( the input pattern ) and produce as output a numerical pattern encoding a classification . they were designed to simulate groups of biological neurons and their interconnections and to mimic the ability of such systems to learn and generalize . discussion of the development and practical application of neural networks is well presented in the literature ( see mcclellan and rumelhardt 1988 ) . in astronomical applications , this technique has been applied with considerable success to the problem of star - galaxy separation by odewahn ( 1992 ) . a neural network classifier was developed storrie - lombardi ( 1992 ) to assign galaxy types on the basis of the photometric parameters supplied by the eso - lv catalog of lauberts and valentijn ( 1989 ) . applications of this method to large surveys of galaxies using photographic schmidt plate material are discussed in odewahn ( 1995 ) and naim ( 1995 ) . more recent discussions of star - galaxy separation using ann s are presented by andreon ( 2000 ) , mahonen & frantti ( 2000 ) , and and philip ( 2000 ) . an extremely thorough discussion of galaxy classification with ann methods is given by bazell ( 2000 ) . the neural network literature quoted above contains a wealth of information on the theoretical development of neural networks and the various methods used to establish the network weight values . we summarize here only the practical aspects of the ann operation and the basic equations for applying a feed - forward network . the input information , in our case the principal components formed from the fourier profiles discussed in section 3.2 , is presented to the network through a set of input nodes , referred to as the input layer . each input node is `` connected '' via an information pathway to nodes in a second layer referred to as the first hidden layer . in a repeating fashion , one may construct a network using any number of such hidden layers . for the backpropagation networks discussed here , we have used ann s with two hidden layers . the final layer of node positions , referred to as the output layer , will contain the numerical output of the network encoding the classification . this system of nodes and interconnections is analogous to the system of neurons and synapses of the brain . information is conveyed along each connection , processed at each node site , and passed along to nodes further `` upstream '' in the network . this is referred to as a feed - forward network . the information processing at each node site is performed by combining all input numerical information from upstream nodes , @xmath1 , in a weighted average of the form : @xmath2 the @xmath3 subscript refers to a summation of all nodes in the previous layer of nodes and the @xmath4 subscript refers to the node position in the present layer , i.e. the node for which we are computing an output . each time the ann produces an output pattern , it has done so by processing information from the input layer . this input information is referred to as the input pattern , and we use the @xmath5 subscript to indicate input pattern . hence , if there are 5 nodes in the previous layer ( @xmath3 ) , each node ( @xmath4 ) in the current layer will contain 5 weight values ( @xmath6 ) and a constant term , @xmath7 , which is referred to as the bias . the final nodal output is computed via the activation function , which in the case of this work is a sigmoidal function of the form : @xmath8 hence , the information passed from node @xmath4 in the current layer of consideration , when the network was presented with input pattern @xmath5 , is denoted as @xmath9 . this numerical value is subsequently passed to the forward layer along the connection lines for further network processing . the activation values computed in the output layer form the numerical output of the ann and serve to encode the classification for input pattern , @xmath5 . in order to solve for the weight and bias values of equation 1 for all nodes , one requires a set of input patterns for which the correct classification is known . this set of examples is used in an iterative fashion to establish weight values using a gradient descent algorithm known as backpropagation . in brief , backpropagation training is performed by initially assigning random values to the @xmath6 and @xmath7 terms in all nodes . each time a training pattern is presented to the ann , the activation for each node , @xmath9 , is computed . after the output layer is computed , we go backwards through the network and compute an error term which measures of the change in the network output produced by an incremental change in the node weight values . based on these error terms , all network weights are updated and the process continues for each set of training patterns . as discussed in the next section , precautions must be taken in this process to avoid problems from over - training ( simply memorizing the input pattern set as opposed to generalizing the problem ) . the goal of the present work is to develop a system that automatically detects the presence of specific types of morphological structures in galaxy images . this is clearly a more complex problem compared to the use of a few global photometric parameters to predict types , and so we desired to use an additional method of neural network classification to confirm the results of our classical backpropagation approach . philip ( 2000 ) develop a neural network based on bayes principal that assumes the clustering of attribute values ( the input parameters ) . the method considers the error produced by each training pattern presented to the network , and weights are updated on the basis of the bayesian probability associated with each attribute for a given training pattern . in this approach , the probability density of identical attribute values flattens out while attributes showing large differences from the mean get boosted in importance . this is referred to as a difference boosting neural network ( dbnn ) . we have applied this new approach as a check on our backpropagation network results , however this method has the added advantage that it trains to convergence in a much faster time than is required by classical backpropagation methods . in this paper we develop a technique that is a pre - processing step for producing input to a supervised classifier in the form of an artificial neural network ( odewahn 1997 ) . in this approach , the radial surface brightness profile of a galaxy is computed in elliptical annuli centered on the galaxy center . the rational here , valid in most cases , is that we are isolating the disk component of the galaxy and hence establishing a way of defining the equatorial plane of the system . the position , shape and orientation of these annuli are determined using classical isophotal ellipse fitting techniques . within each annulus , we compute the run of flux density with position angle , @xmath10 , in the equatorial plane of the galaxy , to form the azimuthal surface brightness profile . to describe each azimuthal profile , and reduce the dimensionality of our classification problem , each azimuthal profile is modeled by the following fourier series : @xmath11 for computational efficiency , the fourier terms are computed for each azimuthal profile using the following moment relations : @xmath12 @xmath13 @xmath14 where @xmath10 is the angle ( in the equatorial plane of the galaxy ) with respect to the photometric major axis , and m is an integer . hence , if we assume galaxy disks to be thin and suffer no internal extinction , this approach crudely removes inclination effects and allows us to compare galaxy images independent of orientation . the fourier series in this paper were computed up to m=5 . fourier amplitudes , describing the relative amplitude of each component are computed using the following expression : @xmath15 finally , a phase angle describing the angular position of the peak signal contributed by each component can be computed . we compute the phase angle of the @xmath16 component as : @xmath17 a practical application of this method to a barred spiral galaxy imaged with hst is shown in figure [ model_sample_1 ] . another example for an elliptical galaxy image which contains no high spatial frequency image structure is shown in figure [ model_sample_2 ] . through experimentation with both hst- and ground - based images , we have determined that using 17 elliptical annuli and up to @xmath18 in equation [ eq : fourier0 ] consistently reproduces the basic morphological features of most galaxy images . a considerable amount of experimentation went into determining how one should establish the radial bin intervals in this procedure . a series of tests were made using unequal linear bins that overlapped and logarithmic binning interval . we experimented with normalizing the bin radius by the effective radius , @xmath19 , so that all galaxies could be compared on more uniform spatial scale . this approach proved problematic since it produced profile scales that were highly compressed for early - type systems , a highly expanded in late - type systems . in such cases , the inner regions if the late - type galaxies , where much of the important morphological structure is present , was under - resolved . additionally , we found that measurement of the effective radius of galaxies in low resolution , low hst images can be systematically low , and using @xmath19 in this manner will introduce a substantial bias when comparing local and distant samples . in the end , we found that using equal size radial bins extending to an outer , low surface brightness optical isophote produce the most robust results for comparing morphological properties among galaxy images . it should be noted that one can justify the use of higher order fourier components for cases of high resolution galaxy images . in experiments with some of the deep palomar 60@xmath20 images described in section 3 , we found that the bars of many galaxies have a strong contrast relative to the disk , and high spatial frequency components ( up to @xmath21 ) are required to reconstruct such a sharp feature . incorporating so many spatial frequencies in a general image classifier is impractical , but one might hope to use some type of information about high frequency structure : if the inclusion of such components in the model significantly improves the model fit , then this is important information . perhaps a good number of barred galaxies can be identified if we simply search for images that require high order fourier components in the inner annuli to build a good fit model . the fourier - reconstructed images described above are comprised of radially - dependent sets of fourier amplitudes and phase angles . using 17 annuli , 6 fourier amplitudes ( including the @xmath22 term ) and the corresponding phase angles to parameterize each image , we thus describe each galaxy with a 221 element classification vector . some human classifiers have insisted that at least 10,000 resolution elements are required in a galaxy image that is to be morphologically classified ( de vaucouleurs , private communication ) . more liberally , we might consider that one should have at least a @xmath23 image of a galaxy , or 2500 pixels , to convey morphological information . hence , using the fourier description of a galaxy reduces the number of parameters describing the image information content by a factor of more than ten compared to that used by a human classifier . an input classification vector with over 100 elements is rather large for most pattern classifiers trained in the presence of noise . to distill our image parameterization further , we have chosen to characterize the radial properties of each fourier component . in other words , just as we often describe the radial surface brightness profile of a galaxy using several model parameters , so too can we describe the radial trends in each set of amplitude and phase estimates computed with equations [ eq : fourier4 ] and [ eq : fourier5 ] . the mean radial trends in both the amplitude and phase of the coefficients in equations [ eq : fourier4 ] and [ eq : fourier5 ] are shown in figures [ mfprof1 ] and [ mfprof2 ] for three types of galaxies . in these figures the mean phase or amplitude of a given fourier component is plotted on the y - axis as a function of elliptical annulus number on the x - axis . we divided a sample of 196 galaxies from section 3 into three family categories : a ( unbarred system ) , ab ( transitional systems ) , and b ( barred systems ) and computed the mean radial profiles for the amplitude and phase coefficients of the @xmath24 and @xmath25 coefficients . as noted by elmegreen and elmegreen ( 1985 ) and ohta ( 1990 ) , these are the dominant terms needed to reproduce the light distributions of most barred galaxies . in figure [ mfprof1 ] we see a clear trend among the @xmath24 and @xmath25 amplitudes : barred systems have significant power in the inner rings , ab systems have systematically lower power , and a galaxies have the lowest amount of @xmath24 and @xmath25 power in even the inner elliptical annuli . these mean profiles were computed with 71 a , 73 ab and 61 b galaxies taken from the analysis of section 4 . the 1-@xmath26 error of each point in the a - system profiles are shown in figures [ mfprof1 ] and [ mfprof2 ] , and it is clear that a single point from the profile of one galaxy will have little discriminatory power in distinguishing bar class . however , if we combine the information of the points from annuli with ring numbers between 3 and 12 , then we can expect to derive a robust estimate of whether a bar is present in the inner regions of a galaxy image . in this region of the reconstructed images , the mean fourier profiles clearly delineate the presence of a bar . in figure [ mfprof2 ] we plot the corresponding mean profiles for the @xmath24 and @xmath25 phase angles . although these profiles are less sensitive to bar presence , the phase terms will be important in distinguishing barred galaxies from purely spiral systems . in the case of a barred galaxy , the @xmath24 and @xmath25 phase angles remain fixed with ring number ( radius ) , whereas with a two - armed spiral pattern we expect to see a systematic variation of @xmath24 and @xmath25 phase angles with radius since the flux density peaks are changing their position angle smoothly as we progress outward in the radial annuli . as can be seen in figure [ mfprof2 ] we still observe a gradual progression in the trends of these phase angle profiles for the a , ab , and b galaxy sets . to summarize , the radial @xmath24 and @xmath25 amplitude and phase angle profiles of a galaxy carry important information about the nature of the 2-d structure present in a galaxy image , and specifically allow us to robustly quantify the presence of a linear bar structure . following the innovative work of han ( 1995 ) , we performed a principal component analysis ( pca ) of each set of fourier profiles for our collections of galaxies that had reliable stage and family classifications . for most fourier coefficients it was determined that 85% to 90% of the profile variance could be described with only 2 or 3 principal components . hence , using the eigenvectors determined from this analysis , we can characterize the fourier profile of a random galaxy using only 2 or 3 parameters . should we use the amplitude and phase of only the @xmath24 and @xmath25 components to characterize a galaxy , then we can parameterize the light distribution using 4 coefficients @xmath27 2 principal components @xmath28 8 parameters . such an image vector constitutes a very reasonable size for training a supervised classifier with the 100 - 200 patterns ( galaxies ) available in this work . in figures [ mfprof1 ] and [ mfprof2 ] we have divided the galaxies by family class , but these fourier - based principal components may be used to identify stage - related image traits also . in figure [ pc_demo ] we demonstrate our principal component method by showing pca - based image parameter spaces that are symbol coded by stage and family class . the development of a pattern classifier based on supervised learning , a classifier that works in a predefined system , requires a data sample consisting of the input classification information ( the image parameters ) and the corresponding target patterns ( the galaxy types ) . this sample is referred to as the training set , and a portion of it is held back from the actual supervised learning process ( in our case backpropagation training ) to be used as a test sample to make an unbiased judgement of classifier performance during and after training . as we discuss below , nearly all of the past morphology - based studies of high redshift galaxies have focused on the stage axis only . since our aim is to develop morphological classifiers that estimate the family and variety of galaxies based on fourier amplitude strengths and phases , we have visually reclassified a large number of high , well - resolved hst galaxy images in the rhs to establish a statistically significant testing sample . although the present sample is based on 35 medium - deep and deep wfpc2 fields , the number of galaxy images from this sample that are suitable for such classification is only 146 . hence , for the present study we incorporated the local galaxy samples from ground - based ccd observations by de jong and van der kruit ( 1995 ) , hereafter j95 , and frei ( 1996 ) , hereafter f96 , adding 124 more galaxies to the sample . for both the local and distant hst samples we inter - compare our classification sets in order to objectively define a weighted mean training set whose systematic and accidental errors are well understood . from such a sample we may establish a fair `` ground - truth '' sample of objects for use in training and testing pattern classifiers . a large collection of visual type estimates in several bandpasses from wfpc2 deep fields were discussed in odewahn ( 1996 ) and driver ( 1995b ) . similar visual classification sets have been assembled since that time for the hubble flanking fields ( odewahn 1997 ) and the bbps fields ( cohen 2001 ) . to begin our assessment of classifier errors , we have analyzed a large number of past visual stage estimates made in the f814w bandpass images . we inter - compared all of the stage estimates common to the 5 classifiers of odewahn ( 1996 ) and the dwg sample and derived linear transformations to convert all classifiers to a common system . the vast majority of classifications were from one classifier ( 1000 from odewahn , hereafter referred to as sco ) . it was decided that the odewahn classes , which match well the rc3 mean type system , would define the hst classification system . following odewahn and de vaucouleurs ( 1993 ) we then use a residual analysis to determine the accidental scatter associated with each classifier , and hence a relative weight associated with each classification of an hst - observed galaxy . finally , a catalog of 1306 galaxy types was assembled : 1000 with single odewahn estimates with mean error of @xmath29 type steps and 306 weighted mean types ( 109 having 5 independent estimates ) with mean error of @xmath30 type steps . the galaxies in this catalog have stage estimates from the the f814w images and objective errors estimated for each type . these data will hereafter be referred to as the hst1 set . the purpose of our current work is to develop automated techniques that identify morphological structures , and with the exception of the bright bbp galaxy types estimates of sco , the hst1 types are comprised only of estimates of the stage axis in the rhs ( e , s0 , sa , ... such data sets are fine for the type classifiers discussed in odewahn ( 1996 ) , however for more advanced classification experiments we desired a set of faint hst - observed galaxies with classifications in the rhs : stage , family and variety . to collect this set we extracted 4534 galaxy images in f814w from 39 different wfpc2 fields . the majority of these images were contributed by moderate depth wfpc2 parallel fields , but the deep 53w002 and hdf - n fields were also used . from this sample we established , purely through visual inspection , the group of 146 galaxy images having sufficient and resolution to allow good morphological classification . as a second sample , we binned 125 ground - based ccd images of f96 ( in gunn g ) and j95 ( in johnson b ) to a resolution similar to that of the hst images . the manipulation of all galaxy images used in this work , whether for the purpose of human visual classification or for automated image analysis was performed with a variant of the morpho package discussed in odewahn ( 1997 ) . a new version of this galaxy morphological classification software system called lmorpho , which is optimized for use under the linux operating system , provides an image database manager that allows users to select galaxies for analysis based on morphological traits , photometric properties or many other meta - data properties ( bandpass , image size , or pixel scale ) . beyond the image database manager of lmorpho , the system provides tools performing automated galaxy surface photometry , basic astronomical image processing , and a variety of multivariate statistical analysis and pattern classification tasks . our galaxy image libraries were independently inspected by three classifiers using an lmorpho graphical galaxy classification tool . the three classifiers were sco , windhorst ( hereafter raw ) , and cohen ( hereafter shc ) , and each visually estimated the stage , family , and variety for each of the 272 galaxy images in the adopted sample . for a variety of reasons ( insufficient or resolution , high inclination , etc ... ) certain aspects of the classification ( usually the family and variety ) were uncertain and these were designated as unknown . following the method used for the hst1 sample , we next inter - compared these type catalogs in order to establish the systematic and accidental errors associated with each set . the results of this are summarized in table [ tab1 ] and figure [ tcompare ] . for the local galaxy samples , we compared each classifier to the weighted mean types in the rhs contained in the rc3 and discussed in buta ( 1994 ) . for the distant sample we compared each classifier s types with the weighted mean types of hst1 . although many of the hst1 types are indeed weighted means of several independent estimates , the majority are single estimates from sco and hence we should regard the hst1 set as an independent ( sco - dominated ) unit - weight catalog . for both the local and distant galaxy samples , a preliminary linear regression analysis was performed to determine transformation relations for converting to the rc3 and hst1 systems respectively . an impartial regression analysis , using each catalog as the dependent variable , was used to determine the mean relationship between all possible catalog pairs . for each sub - sample ( local and distant ) , small scale and zeropoint shifts were applied to the sco , raw and shc stage estimates to bring all classifications onto a uniform system . it is informative to note that the size and sense of the scale corrections were uniform by classifier irrespective of local or distant sample : sco and raw systematically classified galaxies later at the 8% level , and shc systematically classified galaxies earlier at the 6% level . these scale adjustments were statistically significant at the @xmath31 level . the zeropoint shifts , statistically significant at the @xmath32 level , were general less than one step on the 16-step stage axis of the rhs ( with the exception of a @xmath33 step correction for the distant raw sample ) . in each case ( local and distant ) mean transformation equations were derived for the sco , raw , and shc classification sets . following these small , but statistically justified , transformations a second correlation analysis was performed to verify system uniformity and to compute the mean variance associated with each catalog comparison . this analysis is illustrated in the top panel of figure [ tcompare ] for the local sample and in the bottom panel for the distant ( hst ) sample . with all stage estimates transformed to a common mean system , we followed the methodology of odewahn and de vaucouleurs ( 1993 ) to compute the mean standard deviation of the @xmath34 residuals in each comparison ( these values are plotted in each panel of figure [ tcompare ] ) . these estimates were combined via a system of linear equations to calculate the standard deviation associated with each individual catalog . the results , summarized in table [ tab2 ] , show first that each classifier ( sco , raw , shc ) is able to produce stage estimates for the local galaxies in the rc3-defined system with scatter of between 1 and 2 steps on the rhs . this is quite consistent with a similar study by naim ( 1995 ) using 831 galaxies typed from schmidt images by 6 expert classifiers . as expected , the r.m.s . scatter estimated for the distant galaxy samples are somewhat larger , but still in the 1.5 to 2 step range . using the systematic corrections and sample weights derived for the hst classification sets we have compiled a catalog of weighted mean revised hubble types for our distant galaxy samples . in this way we produce not only a final catalog of higher quality types compared to that from any single classifier , but we are able to derive error estimates for each galaxy . this last point is important from the point of view of assessing the quality of any automated morphological classifier . galaxies whose morphological properties are uncertain to a number of human classifiers are most likely peculiar in some aspect and hence should not be included in the training or testing of a generic machine classifier . weighted mean types and errors for our local galaxy sample were adopted from the rc3 . representative galaxies from each of these samples , broken into stage and family groups , are shown in figure [ sample_gals ] . although agreement between stage classifications was found to be quite satisfactory for developing adequate test / train samples for automatic stage classifiers , the uniformity of the family ( barred vs. unbarred ) and the variety ( ringed vs. non - ringed ) classes was less than satisfactory . from the hst image samples , all 3 classifiers agreed on the family assignment only 30% of the time and on the variety assignment only 25% of the time . this is clearly due to a lack of the and resolution required for such morphological classification in galaxy images . in reality , some of the ground - based images also lacked sufficient to allow for the unambiguous assignment of family and variety assignment . hence , for the purposes of developing a robust morphological classifier capable of dealing with these image properties we chose to develop a third , higher weight set of galaxy images . during the course of a large program to acquire photometric calibration fields for the digital palomar sky survey ( gal 2000 ) one of us ( sco ) initiated a program for non - photometric conditions of imaging large , bright well - classified galaxies from the list of `` best - classified '' galaxies in buta ( 1994 ) these galaxies had classifications agreed upon by several human experts and in fact represent proto - types of the various hubble stages . each galaxy was imaged in gunn gri or cousins b for 10 to 15 minutes with the palomar 60-inch using ccd13 . conditions were generally not photometric , but thin cirrus and seeing no worse than 2.5 arcsec fwhm was tolerated for this program . when possible , multiple images of the same galaxy were obtained so that we might characterize how accidental errors in the surface photometry will affect output from the automated morphological classifier . in table [ p60table ] we summarize 30 images in gunn g and cousins b of 20 different galaxies selected for experimenting with with automated family and variety classification . here we describe the application of the method to a set of hst archival images . using the results of section 2 , we assembled samples of local and distant galaxies having weighted mean stage estimates and family classifications that at least two human classifiers agreed upon . we show a representative sample of these galaxies in figure [ sample_gals ] . as we shall discuss below , the criteria for building a family classifier training set had to be somewhat liberal in order to compile a usable experimental sample . for stage classification , a series of backpropagation networks using different input parameter vectors and network layer architectures were experimented with . in the case of the family classifier , we used two different types ann classifiers : a backpropagation - trained feed forward network , and a difference boosting neural network ( dbnn ) developed by philip ( 2000 ) . a set of 262 images were collected using the sample of galaxies having @xmath35 and weighted mean stage estimates from the analysis of section 3 . most galaxy classification systems provide criteria for estimating the morphological type of an edge - on system , however because we are concerned here with the analysis of two - dimensional morphological structures , we chose to exclude such high - inclination systems . an lmorpho task was used to perform automated surface photometry of each image , and then an interactive mosaic viewer was used to inspect the ellipse fit to the low surface brightness isophote in each galaxy postage stamp image . this was done to verify that no improper image processing errors occurred due to confusion from nearby images or other image defects . ten galaxy stamps were rejected in this step , the majority of these being relatively faint hst - observed galaxies that extended too close to the edge of the wfpc2 field . the human classifiers dealt well with `` masking '' such edge effects when the visual classification of section 2 was performed , however the sky fitting and surface brightness contouring routines of lmorpho were unable to properly handle such situations . future automated morphological surveys will of course have to deal more effectively with this type of error , but for the purpose of this work we chose to simply delete these galaxies from any machine classification experiments . for each of the 252 postage stamp images of well - classified galaxies remaining in our experimental sample the full set of morphologically - dependent fourier image parameters discussed in section 3 were computed with lmorpho . a series of backpropagation neural networks were trained following the procedures outlined in odewahn ( 1997 ) using 3 different types of input vector sets and 6 different network architectures . in every training case , two hidden layers of the same size were used to map n - dimension input vectors to an 8-node output layer , one node for each 2-step interval of the stage axis of the rhs . in practice , a final type was assigned for each input vector using the weighted mean value of the output node value . this allows one to assign not only the highest weight classification , but the rms scatter in this node - weighted mean value can be used to assign a classification confidence . hence , while any output pattern will result in the assignment of an estimated type , patterns where most of the output signal is carried in one major node will have a high confidence assignment , and patterns where the output pattern is distributed over many nodes will be assigned low confidence . the image parameter catalogs were collected into a single binary catalog which can be used in lmorpho to interactively generate symbol - coded parameter space plots like those shown in figures [ pars_stage ] and [ pars_family ] . with this package the user can quickly view many different parameter spaces and judge which one shows the largest degree of type separation ( whether stage or family ) by determining how well the different type - coding symbols are separated in the plot . an important feature in this graphical tool is that any point may be marked and used for a variety of uses . in one mode the user can choose to view on a real - time basis the galaxy image of any selected data point . in this way one is able to determine if a particular outlier is due to improper image processing or some truly unique morphological circumstance . in the latter case , such data are retained for use in classifier training . in the former , these patterns would be rejected from use in gathering training and testing samples . as the images used in this particular series of classification experiments were generally of high quality , fewer than 5% of the original 252 postage stamps were rejected in this manner . for only a few hundred galaxy images ( or even less than a few thousand ) this process of parameter space inspection is trivial to complete in a few minutes . finally , we made a preliminary set of ann training runs using different input parameter sets as well as subdividing the training / testing data in different subsets . through this process we uncovered an additional 28 objects that consistantly gave highly discrepant training results , the image parameter combinations for these sources were highly abnormal compared to the bulk of the data , and these sources were rejected from further experiments . this represented about 9% of our original training set , but determining the source of peculiarity for these galaxies will require a larger , more diverse collection of images for future experiments . we determined a number of useful input parameter combinations that showed good segregation by morphological type and a series of different image parameter set patterns were formulated . each type of pattern would form the input layer to an ann classifier . for clarity , we assigned a running integer value to each type of input , and we refer to this as the feature set number . we indicate the feature set number for each training exercise in column 8 of table 6 . in table [ parlist ] we summarize the lmorpho image parameters that were selected for the stage and family classifiers developed in this paper . since different image parameters are used in each classifier , we list in column 2 the feature set number for each input pattern using each image parameter . the selection of training and testing samples was carried out in the same manner for all input vector types . as summarized in table [ tbreakdown ] all galaxies were binned into the 8 type bins corresponding to the 8 nodes of the ann output layer . each such bin was divided into two sets : one for training and one for testing . one is always tempted to use more galaxies in the training sample so that the network has a larger number of patterns for generalizing the problem . one danger in backpropagation training is over - training : beyond some number of training iterations the network weights could be adjusted so as to simply memorize the training pattern sets as opposed to generalizing the mapping of input vectors to output classes . to prevent this , one should use an independent test data set to judge classifier performance . the backpropagation code in lmorpho computes classifier statistics for both a training sample and a testing sample at each user - specified iteration in which the updated ann weights are stored . in a post - training phase , each set of statistics is inspected by the code to determine the optimal training cycle . the algorithm used to isolate this iteration uses the slope and rms scatter of the correlation between the target stage prediction and the ann stage prediction . both the training and testing data sets are used . the case of over - training is detected by a progressive drop in the rms scatter for the training data with a flat or even increasing rms scatter for the test sample . an optimal training cycle is selected that minimizes the rms scatter for training data before over - training occurs . additionally , the slope of the linear regression must be close to unity within some user - specified tolerance , usually set at 20% . we summarize the results of our backpropagation training to develop fourier - based ann classifiers in table [ stageclassstats ] . the slope ( @xmath36 ) and scatter ( @xmath37 ) about a linear regression between the ann - predicted and the targeted mean hubble stage are tabulated for both the training and testing data sets for the selected optimal training epoch . additionally , as with the type comparison analysis in section 3 , we compute a scatter ( @xmath38 ) based on direct differences between the ann and target values following odewahn and de vaucouleurs ( 1993 ) . this statistic reflects more honestly the scatter to be expected for the user of such an ann - derived catalog . as expected , scatter among the training data is systematically lower than for the test data , since the error function being minimized in the backpropagation training is formed with the training data . however , the scatter derived for the test data is usually about 2-steps on the rhs stage axis , and hence very reasonable for most scientific pursuits with morphological data . in figure [ net_results ] we compare mean neural network classifier types from 3 different anns ( marked with asterisks in table [ stageclassstats ] ) to weighted mean visual types for two sets of data : the training data and the testing data . to summarize , the test data , comprised of sets of input classification parameters and their corresponding target types , were never presented to the ann classifiers during backpropagation training to establish the network weight values . as such , this data sample represents a fair comparison by which we can judge true network performance . we include in figure [ net_results ] the same correlation for the training sample . in summary , the fourier - based stage classifiers developed in this experiment clearly perform as well as a human classifier . these results are certainly encouraging if our goal is simply to estimate revised hubble types for a large number of digital images , but a more important point must be stressed . we have shown that our method of parameterizing a galaxy image preserves the information content needed to emulate the process used by a human classifier . as with the bar classification work described below , our ultimate goal will be to use such a digital image analysis to search for more direct correlations among galaxy properties and ultimately develop a more physically meaningful classification system beyond the rhs . as discussed in section 3 , the visual classifiers divided the family estimates into the three bins of the rhs : a , ab , b. the difference between each class is somewhat vague and subject to personal bias . it is not surprising that we found very few cases where all three classifier agree that a galaxy was of the ab family class . as was discussed for figure [ mfprof1 ] , the fourier profiles for ab galaxies are , in the mean , intermediate between the a and b galaxies . however , the ab galaxies show a markedly large variance which is probably due to the wide variance in ab classification criteria used by the human classifiers . the presence of such uncertainty made it impossible to train an effective 3-division stage classifier . for the present work , we approached the problem using an unambiguous set of a and b galaxies and attempted to develop a robust classifier for identifying them using the fourier - based eigenvector approach discussed in section 3 . when larger samples of morphologically classifiable galaxy images are collected , it is anticipated that a more thorough study of bar strength parameters will allow the development of a more sophisticated automated family classifier . for the present set of classification experiments we gathered images of 71 a galaxies and 61 b galaxies drawn from the analysis of section 3 . in this case , any galaxy which had family classifications that were agreed upon by at least 2 of the 3 human classifiers were adopted for use . to clarify the problem , we used only a and b classes for classifier experiments . after a series of parameter space inspections , such as those described in the stage classification section , we chose to use the first principal components of the @xmath24 and @xmath25 amplitude profiles . we found that additional discrimination was added in many cases using the first principal component of the @xmath25 phase profile and adopted this for use in the classifier input vector . it was found that including parameters that are highly correlated with bulge to disk ratio ( the principal components of the normalized flux profile ) did not add significantly to detecting the presence of a bar . this was surprising in that it is well known that bar shapes and properties are correlated with hubble stage , and hence bulge to disk ratio ( elmegreen and elmegreen 1986 ) . in future work with much larger samples of galaxies , the use of terms related to the shape of the surface brightness profile may prove useful for this reason : pattern classifiers can better interpret the meaning of morphological shape parameters like the first principal components of the @xmath24 and @xmath25 amplitude profiles if some information related to hubble stage is provided . for the present small sample of potentially useful training samples , we decided to exclude the use of surface brightness profile shape parameters for the training of a family classifier . we added two simple global image parameters that contribute information about bar presence that were independent of the fourier model approach . in determining the shape and orientation of the optimal elliptical aperture for measuring each galaxy , lmorpho computes a series of ellipse fits to progressively fainter surface brightness levels . the initial surface brightness levels are high and generally sample the bright inner region of a galaxy where the bars are found . it was found that simple parameters using these elliptical isophote parameters could be formed that gave useful information about the presence of a bar that would be independent of the image models derived with the fourier method . we define the minimum axis ratio of the series of ellipse fits to be @xmath39 , and the axis ratio of the largest ellipse ( i.e. the fit to the lowest surface brightness level ) to be @xmath40 . in general , a barred galaxy will have @xmath41 , but with the condition that the major axis length of the isophote measured for @xmath39 is significantly smaller than that measured at the isophote for @xmath40 . to discriminate this condition we formed the parameter @xmath42 , where @xmath43 is the semi - major axis length measured at @xmath39 and @xmath44 is the semi - major axis length measured at @xmath40 . additionally , we formed @xmath45 , to measure the contrast of the ellipticity variation in the galaxy image . it was found that for many galaxies , a plot with @xmath46 on the y - axis and @xmath47 on the x - axis places b galaxies in the upper left corner and a galaxies in the lower right corner . of course , there is a rather substantial overlapping zone in the low @xmath46 and low @xmath47 region of this plot . hence , while such a simple scheme can not discriminate all a , b galaxies , it can provide high weight information for some systems . simple initial ann classification experiments provided evidence that these parameters , when added to the fourier - based parameters , produced better discrimination for a , b samples and hence we included them in the training of our final family classifier . two types of family classifiers were trained with the input vector parameters shown in table [ parlist ] . first , the backpropagation network code used the previously discussed stage classifiers was used to train a network consisting of 2 hidden layers having 12 nodes each . this network mapped an 8-element input vector to a 2-node output layer . patterns with more power produced in output node 1 were classified as a , and patterns with more power produced in output node 2 were classified as b. many of the dimensions in this 8-dimensional classification space produce a good split between the a and b populations and hence the backpropagation training converged relatively quickly . as with the stage classifiers , a sample of pure test patterns was retained and used to assess network performance at each stage in the backpropagation weight update process . this procedure gives us a truly independent check on the classifier performance and an effective means of guarding against over - training . a total of 30 training epochs were used in the backpropagation training , and epoch 15 was selected as providing optimal performance with a success rate performance of 92% among training patterns and 85% among test patterns . in a second experiment the dbnn method of philip ( 2000 ) was applied to the identical training and testing patterns as for the backpropagation ann . after boosting , the training set was found to produce a 94.1% successful classification rate and the test set was found to produce a success rate of 87.5% . the performance of the dbnn classifier was found to be marginally better than the backpropagation - trained network , however a large sample of training data will be needed to assess if this improvement is significant . nevertheless , both methods produced automated family classifiers that are able to discriminate bar presence with a roughly 90% probability of success . image quality plays a crucial role in determining the extent to which morphological classification can be performed on a galaxy . a gradual decrease in image resolution and will systematically degrade the quality of morphological classifications , both by human and machine methods . with stage estimation , one might expect that we lose the ability to differentiate the bulge and disk components , making it extremely difficult to work at the early stage of the hubble sequence ( differentiate e , s0 and early spiral systems ) . at the late hubble stages , we lose many of the low surface brightness features in the out disk regions that are crucial for differentiating among the sd , sm and i m systems . with family classification , one can expect that gradual image degradation will make it impossible to detect the presence of a bar , much less characterize the properties of the bar ( see van den bergh 2001 ) . in the past , human classifiers , in particular those dealing with low resolution and low schmidt plate images ( nilson 1973 , corwin 1985 ) , have generally taken such effects into account by dropping the amount of detail in the literal type assigned to a galaxy . in other words , a system might be classified simply as `` s '' rather that `` sbc '' , if the image quality is sufficient only to differentiate the difference between early- , mid- , or late - type galaxy . such effects must be accounted for in any automated system geared towards the recognition of morphological features . one must know if there is a strong systematic bias towards , for instance , missing bar or spiral features as resolution decreases . perhaps if such effects are well understood we may hope to correct large statistical samples in some systematic way . at the very least , we must quantify the levels of low or resolution that can be tolerated before the quality of automated classifications falls below some minimum tolerance required by a given science goal . we chose to define resolution as the number of pixels in a galaxy image contained within the isophotal ellipse used to integrate the isophotal magnitude in lmorpho . this is a reasonable approach for hst images where , for most filters , each pixel approaches the resolution of the optical image . for ground based images we should use the size of the seeing disk , which is generally several times larger than the pixel size . in our present study all of the ground based images have been block averaged to produce pixel sizes that are comparable to the seeing disk and hence we can use this uniform resolution definition for all of the galaxies analyzed . as for defining the in a galaxy image we used two approaches . in the first , we simply compute the ratio of the mean signal ( above sky ) per pixel to the standard deviation of the local sky measure . in the second approach , we define the mean signal , @xmath48 , to be the zeroeth order term in the fourier series fitted to the azimuthal distribution in each model annulus , and the noise , @xmath49 , is the r.m.s . scatter about that fit . this latter approach incorporates the poisson noise associated with the detected galaxy signal and accounts for large scale structural changes across the azimuthal profile . in practice , the two methods had very different ranges , but were found to be well correlated . the method 1 approach yielded estimates in the range 100 to 1000 , and the fourier - based method produced values in the range of 3 to 20 . it was found that the early - type galaxies , having little large scale structure in their images , produced the tightest correlation between these two types of estimators . for practical reasons , we used the more easily computed method 1 values in the image classification experiments described in this section . to characterize how decreased and resolution will affect the quality of our fourier - based ann classifiers , we selected sets of galaxies having extremely high and resolution . these systems were found to be classified correctly by the stage and family classifiers discussed in the previous sections . most of these images were taken from the ground - based datasets obtained with the palomar 60@xmath20 discussed in section 3 , but a few of the highest resolution hst images were also included . for samples having well classified stage and family types , we selected galaxies with images that had at least 2000 pixels contained within the optimal elliptical aperture . these sets of images were then systematically re - sampled using an lmorpho package designed for this experiment to produce postage stamp images of galaxies having progressively lower and resolution . as this exercise was designed to study the general systematic effects of image degradation on morphological classification , no attempt was made to model the noise properties of a given detector , like wfpc2 or stis ( as in odewahn 1997 ) . for each re - sampled image , an extra component of gaussian noise was added to the original image . we modeled decreasing resolution in two ways . first , we consider the case of high sample rate with image blurring , the case typically encountered with ground - based observations of nearby galaxies . in the second , and more relevant case , we considered a low sample rate with high optical resolution , the image suffers from a high degree of pixelation . to simulate this effect , we simply block averaged the original galaxy images to a lower image sampling rate . this latter effect is well known to anyone who has classified large numbers of distant galaxies on wfpc2 images and is the one , as we will show , which dominates our ability to determine distant galaxy morphological classification . the results for the backpropagation ann stage classifiers are summarized in figure [ systematics_stage ] . for this experiment a simple success rate calculation was not appropriate . we desire to know not only whether a classifier fails to predict the correct stage estimate within some number of type bins , but in which direction a misclassification occurs . if the classifier consistently pushed stage estimates in an early ( towards e , s0 ) or a late ( toward i m ) direction then a substantial bias will be introduced in the scientific interpretation of the ann - generated morphological catalog . hence , we chose to compute the trend in type stage residuals , computed in the sense of @xmath50 , as a function of resolution and , where @xmath51 is the true galaxy type . these trends were computed using overlapping bins in resolution or , with the mean stage residual and the rms about that mean computed in each bin . a set of 1901 re - sampled images were processed with the ann classifiers discussed in section 4.2 to estimate hubble stage . in figure [ systematics_stage ] we show trends in stage residuals for three sets of target stage intervals : e - s0 , sa - sbc , and sc - im . in each case we have used overlapping x - axis bins of width 0.2 dex , and we expect to observe the degradation in classifier performance as image quality is lowered . spiral and irregular stages show a moderate trend in positive mean offset ( the ann classifies these galaxies with an earlier value ) and increased rms scatter about the mean with decreasing resolution and . of more significance , e+s0 galaxies exhibit a mean negative offset ( the ann classifies these systems to be later than the target value ) of 2 - 3 hubble steps . this occurs for even the highest resolution and highest galaxy images in our present sample . additionally , this mean offset and increasing rms scatter in t - types for the e+s0 galaxies is clearly steeper than for spiral and irregular galaxies . a variety of network architectures and input parameter vectors were experimented with in order to correct this problem . little improvement was obtained due in large part to the lack of a sufficiently large sample of images for such galaxies . as discussed in cohen ( 2001 ) , e+s0 in the field are relatively rare for 22 , and hence a very large number of hst parallel observations are needed to build up a significant image library . work is currently underway to collect such images from the many archival hst observations of moderate - redshift clusters . clusters contain not only a large number of galaxies in a single wfpc2 field , but the morphological fractions of these galaxies are skewed to the early - types . the analysis of such large samples will potentially eliminate the mean negative stage offsets observed for the e+s0 samples of figure [ systematics_stage ] , however it is doubtful that we can hope for a comparable decrease in the rms of such stage classifications in the low and resolution regimes . it is clear that images of sufficiently high quality are needed to differentiate many of the subtle structural properties of early - type galaxies , as is clear from the work of i m ( 2000 ) . the larger field size , higher quantum efficiency , and especially increased resolution of acs , compared to wfpc2 , will make this the instrument of choice on hst for identifying large samples of moderate - redshift e+s0 systems in the near future . the results for the backpropagation ann family classifier are summarized in figure [ systematics_bar ] . unlike the stage classifiers , our family classifiers predict only two states : a or b. hence , we are able to characterize the output results as success or failure cases . we therefore chose to plot the distribution of and resolution for ann - predicted classes for a sample of barred galaxies that agree with the target value and for those that fail to agree . as expected , in figure [ systematics_bar ] the success cases cluster in the high , high resolution area of the plot , and the failure cases cluster most heavily in the low and resolution region . for clarity , we also plot the success rate trends binned by and resolution . for identifying the presence of a bar , it is quite clear from these mean relations that some critical image resolution is required . for our present samples and fourier - based ann classifiers , we are unable to effectively identify barred systems at better than a 70% success rate until we obtain galaxy images with at least 1000 resolution elements ( @xmath52 3 , where @xmath52 is the number of image pixels above the isophotal threshold ) . the trend with is less steep , but a mean ( per pixel ) of around 1000 is also required to reliably identify bars . the same experiment was carried out for a sample of unbarred ( a ) galaxies with different , but expected , results . in this case , galaxies that are successfully classified as a at high and resolution generally retain that classification at low and resolution . in other words , the ann classifiers rarely turn a galaxies into b galaxies , as is true for human classifiers . in general , as we lose image resolution we begin to loose the ability to identify morphological bars , but we do not tend to contaminate b galaxy samples with misclassified a galaxies , an important point when considering the the frequency of barred systems at high redshift . in this work we have developed a methodology for classifying galaxies based on the strength of fourier components in the azimuthal light distribution . we have used pattern recognition techniques to develop machine classifiers which map the fourier mode information to the revised hubble system , a well known galaxy classification system . we must stress that the basic motivation here is not simply to create a machine - based replication of the revised hubble system . the rhs was developed to describe a variety of properties , some of which are directly related with two - dimensional morphological structures in galaxies . we therefore have chosen to use this classification system to guide our development of new quantitative classification systems . we have demonstrated here a method of extracting galaxy image information in a way which permits us to automate the recognition of bars and hubble type , as defined by the family and stage axes of the rhs . in a sense , this work represents a proof - of - concept study . the current method is able to recognize many of the same structural properties of galaxies that have guided the study of galaxy morphology for the last fifty years . having established how well the method works , under a variety of image and resolution regimes , we can entertain the prospect of searching for refinements in the classification system itself . in other words , the next step is to apply various multivariate statistical techniques and non - supervised learning methods to determine if there are relationships among the fourier - based model image parameters that convey more directly relationships among galaxies . for instance , strength of the @xmath24 and @xmath25 fourier components in the inner regions of disk galaxies might be used to form a continuous measure of bar strength , and we might dispense with the 3-cell family axis of the rhs all together . continuity along each axis of rhs was recognized long ago ( de vaucouleurs 1959 ) , but human classifiers lacked the image measurement ability to classify galaxies in anything other than a set of bins . in addition , as discussed in section 4.3 , we may now incorporate information on image quality , as measured by and resolution , in determining what measurable morphological information is available in a galaxy image . finally , disk systems in the late stages of the rhs show increasing amounts of global image asymmetry . such asymmetries are sometimes linked to tidal encounters with nearby galaxies . how are these asymmetries related ? is the increased degree of disk asymmetry caused exclusively by encounters , and is the high frequency of asymmetric systems observed at @xmath53 0.5 the result of an increased interaction rate ? strength of the @xmath54 fourier component , as measured by the first principal component of the @xmath54 amplitude profile , may be used to measure this asymmetry independent of the presence of other morphological structures like bars and arms , something that is not possible with less sophisticated image - rotation asymmetry measurements . such asymmetry measurements may be correlated closely with the internal kinematics of the galaxy ( rix & zaritsky 1996 ) or the present level of active star formation in a disk and hence will convey more direct insight into the physical processes driving galaxy formation and evolution . we acknowledge support from nasa grants ar.6385.01.95a , go-6609.01 - 95a , and ar-7534.01 - 96a . abraham , r. , tanvir , n.r . , santiago , b. , ellis , r. s. , glazebrook , k. g. & van den bergh , s. , 1996 , , 279 , l47 c c c c c c c c sco & rc3 & l & [email protected] & [email protected] & 0.898 & 119 & 1.60 + shc & rc3 & l & [email protected] & [email protected] & 0.810 & 123 & 2.13 + raw & rc3 & l & [email protected] & [email protected] & 0.889 & 121 & 1.61 + shc & sco & l & [email protected] & [email protected] & 0.739 & 123 & 2.53 + raw & sco & l & [email protected] & [email protected] & 0.929 & 120 & 1.38 + raw & shc & l & [email protected] & [email protected] & 0.792 & 121 & 2.21 + & & & & & & & + sco & hst1 & d & [email protected] & [email protected] & 0.911 & 117 & 1.92 + shc & hst1 & d & [email protected] & [email protected] & 0.786 & 125 & 2.86 + raw & hst1 & d & [email protected] & [email protected] & 0.810 & 137 & 2.73 + sco & shc & d & [email protected] & [email protected] & 0.795 & 116 & 2.84 + raw & sco & d & [email protected] & [email protected] & 0.860 & 113 & 2.41 + raw & shc & d & [email protected] & [email protected] & 0.729 & 123 & 3.27 + [ tab1 ] the first two columns indicate the stage catalogs used on each axis to fit the relation @xmath56 . the column labeled s indicates whether a sample of local ( l ) or distant ( d ) galaxies were used . the correlation coefficient , r , the number of galaxies in the fit , n , and the standard deviation of the stage residuals , @xmath26 are also tabulated . l l c ngc 1744 & sb(s)d & 1 g + ngc 4501 & sa(rs)b & 2b + ngc 4519 & sb(rs)d & 1 g + ngc 4618 & sbm & 2 g + ngc 5850 & sb(r)b & 1g,1b + ngc 5985 & sab(r)b & 1 g + ngc 5377 & sb(s)a & 1 g + ngc 6070 & sa(s)c & 3g,1b + ngc 6340 & sa(s)0 & 1 g + ngc 6384 & sab(r)bc & 1g,2b + ngc 6643 & sa(rs)c & 1 g + ngc 6764 & sb(s)bc & 1 g + ngc 6824 & sa(s)b & 1 g + ngc 6951 & sab(rs)bc & 2 g + ngc 7217 & sa(r)ab & 2 g + ngc 7457 & sa(rs)0 & 1 g + ngc 7479 & sb(s)c & 1 g + ngc 7741 & sb(rs)cd & 1 g + pgc 61512 & sb(s)dm & 1 g + [ p60table ] l c l pc1_fnorm & 1,2,3 & first pc of normalized flux profile + pc2_fnorm & 1,2,3 & second pc of normalized flux profile + pc1_1ta & 2,3 & first pc of @xmath54 amplitude profile + pc1_2ta & 1,2,3,4 & first pc of @xmath24 amplitude profile + pc2_2ta & 1 & second pc of @xmath24 amplitude profile + pc1_4ta & 1,2 & first pc of @xmath25 amplitude profile + pc2_4ta & 1,2,4 & second pc of @xmath25 amplitude profile + pc1_4tp & 1,2 & first pc of @xmath25 phase profile + pc2_4tp & 1,2,4 & second pc of @xmath25 phase profile + s / n & 3 & mean measured in elliptical aperture + @xmath65 & 3 & number of image pixels + @xmath47 & 4 & bar parameter 3 + @xmath46 & 4 & bar parameter 4 + [ parlist ] c l c c 1 & 19 & -8.0 -4.5 & e to e0 - 1 + 2 & 6 & -4.5 -2.5 & e+ to l0- + 3 & 13 & -2.0 -0.5 & l0 to l0 + + 4 & 18 & -0.5 1.5 & s0a to sa + 5 & 34 & 1.5 3.5 & sab to sb + 6 & 91 & 3.5 5.5 & sbc to sc + 7 & 30 & 5.5 7.5 & scd to sd + 8 & 12 & 7.5 10.5 & sdm to i m + [ tbreakdown ] l l l l l l c c c 0.920 & 0.93 & 2.09 & 0.908 & 2.25 & 2.09 & 5800 & 1 & 9 + 0.953 & 1.15 & 1.71 & 0.968 & 2.29 & 1.71 & 2900 & 1 & 11 + 1.045 & 1.16 & 1.68 & 0.949 & 2.46 & 1.68 & 2600 & 1 & 12 @xmath66 + 0.935 & 1.20 & 1.84 & 0.959 & 2.37 & 1.84 & 2000 & 1 & 13 + 0.705 & 1.71 & 2.57 & 0.722 & 1.92 & 2.57 & 1000 & 1 & 14 + 0.931 & 1.78 & 2.28 & 0.906 & 1.92 & 2.28 & 4200 & 2 & 9 + 0.974 & 1.35 & 1.82 & 0.909 & 2.06 & 1.82 & 4300 & 2 & 11 + 0.936 & 1.56 & 2.25 & 0.914 & 2.06 & 2.25 & 2300 & 2 & 12 + 0.938 & 1.13 & 1.87 & 0.986 & 2.51 & 1.87 & 3900 & 2 & 13 @xmath66 + 1.081 & 1.32 & 1.74 & 1.005 & 1.88 & 1.74 & 3600 & 2 & 14 + 1.009 & 1.44 & 1.71 & 0.978 & 1.71 & 1.71 & 1800 & 3 & 9 + 0.878 & 1.16 & 2.59 & 0.803 & 1.77 & 2.59 & 1000 & 3 & 11 + 1.016 & 0.93 & 1.54 & 0.946 & 1.83 & 1.54 & 900 & 3 & 12 @xmath66 + 0.917 & 1.36 & 2.07 & 0.965 & 1.62 & 2.07 & 500 & 3 & 13 + 1.025 & 0.79 & 1.38 & 1.045 & 1.79 & 1.38 & 900 & 3 & 14 + [ stageclassstats ] asterisks indicate networks that gave the best performance for a given node architecture and these were used to compute mean types in figure [ net_results ] . the tr superscript refers to training results and te refers to testing results . the l column indicates the number of nodes in each of the two hidden layers used in every ann . as explained in the text , the feature set number indicates the set of image parameters fed to each network .

accretion onto central massive black holes in galaxies is often modelled with the bondi solution . in this paper we study a generalization of the classical bondi accretion theory , considering the additional effects of the gravitational potential of the host galaxy , and of electron scattering in the optically thin limit . we provide a general analysis of the bias in the estimates of the bondi radius and mass accretion rate , when adopting as fiducial values for the density and temperature at infinity the values of these quantities measured at finite distance from the central black hole . we also give general formulae to compute the correction terms of the critical accretion parameter in relevant asymptotic regimes . a full analytical discussion is presented in the case of an hernquist galaxy , when the problem reduces to the discussion of a cubic equation , therefore allowing for more than one critical point in the accretion structure . the results are useful for observational works ( especially in the case of low - luminosity systems ) , as well as for numerical simulations , where accretion rates are usually defined in terms of the gas properties near the black hole . [ firstpage ] galaxies : elliptical and lenticular , cd accretion : spherical accretion x - rays : galaxies x - rays : ism

1602.04598

the bondi solution of accretion on a point mass ( bondi 1952 ) , due to its inherent simplicity , is a standard tool for the interpretation of observations of the accretion phenomenon , and the starting point for the development of recipes for the mass accretion rate , to be adopted for example in semi - analytical models and numerical simulations that lack the resolution to study gas transport down to parsec scale . in cosmological simulations and semi - analytical models of the early growth of massive black holes ( hereafter mbhs ) , and of the co - evolution of mbhs and their host galaxies , the bondi accretion rate is used to link the mass supply to the accretion disks surrounding mbhs with the density and temperature of their environment ( e.g. , fabian & rees 1995 , volonteri & rees 2005 , di matteo et al . 2005 , hopkins et al . 2006 , booth & schaye 2009 , park & ricotti 2011 , wyithe & loeb 2012 , hirschmann et al . 2014 , inayoshi et al . 2015 , degraf et al . 2015 , curtis & sijacki 2015 ) . another important application of the classical bondi model is to estimate the mass accretion rate on mbhs at the center of galaxies , by using observed values of the gas density and temperature in the vicinity of the mbh ( e.g. , loewenstein et al . 2001 ; baganoff et al . 2003 , pellegrini 2005 , 2010 ; allen et al . 2006 ; rafferty et al . 2006 ; mcnamara et al . 2011 ; wong et al . 2014 ; russell et al . in this view , one assumes that these values represent the true boundary conditions ( i.e. , at infinity ) for the bondi problem ( see quataert & narayan 2000 ) . however , it is recognized that even the knowledge of the true boundary conditions would not be enough for a proper treatment of mass accretion on mbhs at the center of galaxies : first , because the mbh is not isolated , being at the bottom of the host galaxy potential well ; second , because the radiation emitted by the inflowing material interacts with the material itself , with the consequent establishment of unsteady accretion ( for luminosities of the order of @xmath0 or greater , where @xmath1 is the eddington luminosity ; e.g. , cowie et al . 1978 ) ; and finally , because the flow also gets mass and energy from the inputs due to stellar evolution ( e.g. , ciotti et al . 1991 ) . when the last two of the above circumstances are important , bondi accretion can not be applied ; during phases of moderate accretion , instaed , the problem can be considered almost steady , so that bondi accretion can be considered a first , reliable approximation of the real situation . in this paper we present a quantification of the bias on the estimates of the bondi radius and mass accretion rate , that is introduced when adopting as boundary values for the density and temperature those at arbitrary but finite distances from the mbhs . first we derive the exact formulae for this bias , in case of radiation pressure due to electron scattering , and of the additional gravitational potential of a galaxy ; we also derive the asymptotic expansion of these formulae close to the mbh . these formulae contain a critical accretion parameter , that in general can be determined only numerically . then , we present a technique to obtain the analytical expressions for the critical accretion parameter , in some special cases . a full analytical discussion of the critical points is presented for a hernquist galaxy model . we finally solve numerically the bondi problem for a mbh at the center of a galaxy , and including the effect of radiation pressure due to electron scattering , and compare the numerical results with the analytical ones . the paper is organized as follows . in section 2 we recall the main properties of the classical bondi solution , and we present a preliminary analysis of the mass accretion bias introduced by considering as boundary values for the density and temperature those at points along the solution at finite distance from the mbh . in section 3 we add to the bondi solution , in a self - consistent way , the effect of electron scattering , and in section 4 we consider the full case of the bondi solution in presence of radiation feedback and a galaxy potential . in section 5 the particular case of a hernquist galaxy is presented , building numerically the accretion solution , and also providing a full analytical discussion of the problem . for all these cases we derive the formulae that allow to recover the true accretion rate from fiducial estimates of the bondi radius and accretion rate obtained by assuming classical bondi accretion . the main conclusions are summarized in section 6 . finally , three appendixes contain technical details and relevant formulae useful in analytical and numerical studies . as the present investigation builds on the classical bondi ( 1952 ) accretion model , it is useful to recall its main properties . the classical bondi theory describes spherically - symmetric , steady accretion of a spatially infinite gas distribution onto an isolated central mass , in our case a mbh , of mass @xmath2 . the self - gravity , angular momentum and viscosity of the accreting gas , as well as magnetic fields and feedback phenomena , are not considered . the gas is taken to be perfect , and subject to polytropic transformations ; thus its pressure ( @xmath3 ) and density ( @xmath4 ) are related by : @xmath5 where @xmath6 is the polytropic index , @xmath7 is the proton mass , @xmath8 is the mean molecular weight , @xmath9 is the boltzmann constant , and @xmath10 and @xmath11 are respectively the gas pressure and the density at infinity . the polytropic gas sound speed is @xmath12 note that @xmath13 is not necessarily the adiabatic index , so that in the bondi theory the entropy of the gas can change along the radial streamlines ( in fact , polytropic transformations have a constant specific heat , e.g. , chandrasekhar 1939 ) , and in principle , once the solution is known , one could compute the heat balance of each fluid element as it moves toward the mbh , adiabatic index @xmath14 , and specific heat at constant volume @xmath15 , the molar specific heat is @xmath16 . therefore , when @xmath17 , a fluid element loses energy as it moves inward and heats . ] . in spherical symmetry the time - independent continuity equation is : @xmath18 where @xmath19 is the gas radial velocity , and @xmath20 is the time - independent accretion rate on the mbh . the bernoulli equation , with the appropriate boundary conditions at infinity , becomes : @xmath21 where , from eq . ( 1 ) and @xmath22 , @xmath23 is given by : @xmath24,\ ] ] where @xmath25 is the sound speed of the gas at infinity . in the isothermal case , @xmath26 and @xmath27 . in the case of an adiabatic transformation , @xmath28 is the enthalpy per unit mass , while it is just proportional to it for a generic polytropic transformation . a scale length of fundamental importance for the problem , the so - called bondi radius , is naturally defined as appears in the numerator of the definition of the bondi radius . ] @xmath29 and eqs . ( 3)-(4 ) are then recast in dimensionless form by introducing the normalized quantities : @xmath30 and the mach number @xmath31 . for @xmath22 , eqs . ( 3)-(4 ) then become @xmath32 where @xmath33 is the dimensionless accretion parameter : once known , it fixes the accretion rate for assigned @xmath2 and boundary conditions for the accreting gas . note how from eqs . ( 7)-(8 ) it follows that all physical quantities can be expressed in terms of the radial profile of the mach number . by elimination of @xmath34 in eq . ( 8) , the bondi problem reduces to the solution of the equation @xmath35 where @xmath36 as well known , @xmath37 can not be chosen arbitrarily : for @xmath38 , both @xmath39 and @xmath40 have a minimum ( that we indicate with @xmath41 and @xmath42 , respectively ) , thus to satisfy eq . ( 10 ) @xmath43 requires that @xmath44 , i.e. , that @xmath45 . equation ( 10 ) then implies @xmath46 it is easy to show that , for @xmath47 : @xmath48 so that for the classical bondi problem one has : @xmath49 note that @xmath50 for @xmath51 , and @xmath52 for @xmath53 . in the isothermal case , the analogous of eq . ( 10 ) is : @xmath54 where now @xmath55 solutions exist only for @xmath56 , i.e. , for @xmath57 . simple algebra shows that @xmath58 so that @xmath59 in the isothermal case coincides with the limit of eq . ( 14 ) for @xmath60 . in practice , to solve the bondi problem means to obtain the radial profile of @xmath61 , for given @xmath62 . unfortunately , eqs . ( 10 ) and ( 15 ) do not have an explicit solution , and must be solved numerically . we do not enter in the details of the solutions ( see , e.g. , bondi 1952 ; frank , king & raine 1992 ; krolik 1998 ) and , as common in similar studies , we restrict to the critical case @xmath63 ; in this case , @xmath64 is also the sonic radius . among the two critical solutions , we consider that with increasing mach number approaching the center . in the following , the function @xmath40 in eqs . ( 11 ) and ( 16 ) is generalized by considering the effect of radiation pressure due to electron scattering , and the additional gravitational field of the host galaxy . in those more general cases , we provide the formulae for the true mass accretion rate , together with its estimates obtained using values of density and temperature at any radius @xmath65 along the ( new ) solution , while assuming classical bondi accretion . the critical solutions for each case are constructed with the aid of a numerical code built on purpose : we first determined numerically the position and the value of the absolute minimum of @xmath66 ( sects . 4 and 5 ) , and , after having determined @xmath59 , we evaluated the solution over the whole radial range . here we introduce the general procedure that will be considered in the next sections to estimate the bondi radius and mass accretion rate , for the basic case of the classical bondi solution . to keep the notation simple , in the following we use the symbol @xmath67 to indicate the critical value @xmath59 . later we will use the same procedure after having included in the problem the effects of radiation pressure due to electron scattering ( sect . 3 ) , and of the gravitational potential of the host galaxy ( sects . 4 and 5 ) . for assigned values of @xmath11 , @xmath68 , @xmath13 and @xmath2 , the classical bondi accretion rate is given by eq . ( 9 ) : @xmath69 in practice , when dealing with observations or numerical simulations , one inserts in eq . ( 18 ) the values of @xmath70 and @xmath71 at a finite distance @xmath65 from the mbh , and considers them as `` proxies '' for @xmath11 and @xmath68 . this procedure gives an _ estimated _ value of the bondi radius ( that we call @xmath72 ) and mass accretion rate ( that we call @xmath73 ) . here we investigate how much these @xmath72 and @xmath73 depart from the true values @xmath74 and @xmath20 , as a function of @xmath65 , under the assumption that the bondi solution holds at all radii . the fiducial bondi radius and mass accretion rate are then defined as : @xmath75 in particular , @xmath72 can be conveniently normalized to @xmath74 as : @xmath76 and , from eqs . ( 18)-(19 ) , the ratio @xmath77 can be expressed in terms of @xmath78 , independently of the boundary conditions @xmath11 and @xmath68 , as : @xmath79^{5 - 3\gamma\over 2(\gamma-1)}.\ ] ] obviously , for @xmath80 , one has @xmath81 , by definition of @xmath74 and @xmath72 [ eqs . ( 6 ) and ( 19 ) ] ; or , equivalently , because @xmath82 ; in turn , one also has that @xmath83 . note also how , for @xmath47 , @xmath72 is always smaller than @xmath74 , since the gas sound speed increases inward ; for @xmath26 , the sound speed is constant , and then @xmath84 , independently of the distance from the center . for the mass accretion rate , in the monoatomic adiabatic case @xmath85 , @xmath86 , independently of the distance from the center , i.e. , there is no bias in the estimated mass accretion rate . for @xmath26 , instead , using eq . ( 20 ) , the bias is just given by @xmath87 at the radius @xmath65 where the measure is taken , i.e. , @xmath88 . a more quantitative insight in the behavior of eqs . ( 20)-(21 ) is derived from the asymptotic expansion of @xmath61 near the center , which is obtained after computing the expansion of @xmath40 for @xmath89 , and of @xmath90 for @xmath91 , in eqs . ( 11 ) and ( 16 ) . for @xmath92 , near the center @xmath93 , and the leading terms of eqs . ( 20)-(21 ) read : @xmath94 @xmath95 for @xmath47 , eq . ( 22 ) confirms that @xmath78 decreases as @xmath65 approaches the center . for @xmath96 ( @xmath97 ) , there is an interesting result : @xmath98 , i.e. , independently of the position @xmath65 at which the temperature to derive @xmath72 is taken , it is always concluded that the fiducial bondi radius is placed at a larger radius ( @xmath99 ) , and by the same factor . for what is concerning the mass accretion rate @xmath73 , from eq . ( 23 ) one has that @xmath100 , if @xmath101 ; thus , @xmath73 significantly overestimates @xmath20 for @xmath89 , more than for any other larger @xmath102 . note that in eq . ( 22 ) it is possible to express @xmath78 in terms of @xmath103 , and then in eq . ( 21 ) to obtain @xmath20 in terms of @xmath73 and @xmath103 . in this way , the true @xmath20 can be recovered from densities and temperatures taken at some ( small ) distance @xmath65 from the center , as @xmath104 . this represents a useful result for observational and numerical studies , affected by instrumental or grid resolution . the properties above , together with the trend of @xmath105 with @xmath106 for different @xmath13 , are illustrated by fig . [ fig : bias ] , that uses the numerical results of our code . the figure shows how @xmath72 is smaller than @xmath74 , and how this underestimate increases with @xmath65 decreasing , and for increasing @xmath13 . the lower panel of fig . [ fig : bias ] shows the trend of @xmath107 with @xmath106 , for different @xmath13 ; one sees that the use in eq . ( 19 ) of @xmath108 instead of @xmath11 , of @xmath109 instead of @xmath25 , and of @xmath72 instead of @xmath74 , leads to an overestimate of the true accretion rate @xmath20 ( except for @xmath96 ) . for @xmath110 , the overestimate of @xmath20 is significant . the numerical results in fig . [ fig : bias ] , for @xmath111 , are in excellent agreement with those provided by the analytical asymptotic expansions near the center in eqs . ( 22)-(23 ) ( not shown in this figure for clarity ; but see the following fig . as well known , the bondi solution is a purely hydrodynamical flow , where heat exchanges are implicitly described by the polytropic index . therefore , for given polytropic index , and in absence of shock waves ( as for example in the transition from subsonic to supersonic regime for the subcritical @xmath112 case ) , one could follow the entropy evolution of each fluid element along the radial streamline , and determine the reversible heat exchanges . however , in real accretion the energetics can be dominated by the irreversible emission of energy near the mbh , that , for the radiative component , is usually expressed as @xmath113 where @xmath114 is the efficiency of the release of the accretion energy , and @xmath115 is the mass accretion rate ; in the classical bondi accretion , @xmath116 . in principle @xmath114 can depend on luminosity @xmath117 or on @xmath115 ( especially at low luminosities , as in the adaf family [ narayan & yi 1995 ] , and its variants ) . at high accretion rates , the efficiency is of the order of @xmath118 , and the accretion is likely unsteady , so that bondi accretion can not be used ( e.g. , ciotti & ostriker 2012 for a review ) . the emitted radiation interacts with the surrounding medium and modifies the accretion process : the radiation effects can be sufficiently strong to stop accretion ( the so - called negative feedback ) , and shut - off the central agn . stationarity is almost impossible in these circumstances ( binney & tabor 1995 ; ciotti & ostriker 1997 , 2001 ; park et al . 2014 ) . however , when restricting to low accretion rates and considering only electron scattering , in the optically thin regime it is possible to generalize the classical bondi accretion solution by taking into account the radiation pressure effect ( taam , fu & fryxell 1991 ; fukue 2001 ; lusso & ciotti 2011 ) . in fact , for electron scattering in the optically thin regime and in spherical symmetry , the effective force experienced by a gas element can be written as : @xmath119 where @xmath120 is the eddington luminosity , and @xmath121@xmath122 is the thomson cross section . in the optically thin regime , being @xmath123 independent of radius , the radiation feedback can be implemented in eq . ( 11 ) as a correction that reduces the gravitational force of the mbh by the factor @xmath124 ; thus , for @xmath22 , the function @xmath66 in eq . ( 11 ) becomes : @xmath125 we do not give the analogous formula for the isothermal case ( eq . 16 ) , because , as in the classical case , the relevant quantities can be obtained by taking the limit for @xmath126 of the formulae in the general polytropic case . one can repeat step by step the analysis of sect . 2 , and show that the ( unique ) minimum of @xmath66 in eq . ( 26 ) is reached for @xmath127 being @xmath128 , the position of the minimum moves inward with respect to the classical case in eq . ( 13 ) , and the value of @xmath42 is just that of the classical bondi accretion ( eq . 13 ) reduced by the factor @xmath129 . since the minimum of @xmath39 is independent of electron scattering , the critical value of the new accretion parameter @xmath130 , at a given @xmath13 , is : @xmath131 where @xmath67 is the critical parameter in the corresponding classical case . being @xmath128 , @xmath130 is lower than in the classical model . the true accretion rate , that we now call @xmath132 , is also reduced with respect to the classical value @xmath20 , for given @xmath2 , @xmath13 , and boundary conditions at infinity : @xmath133 note that @xmath132 enters the value of @xmath124 through the accretion luminosity @xmath117 , so that eq . ( 29 ) can be seen as an implicit equation for @xmath132 . remarkably , this equation can be explicitly solved by introducing the eddington mass accretion rate @xmath134 in general @xmath114 depends on the accretion rate , but in the following for illustrative purposes we assume @xmath135 . from eqs . ( 29)-(30 ) , one obtains explicitly @xmath136 in terms of @xmath137 , by solving the quadratic equation : @xmath138 ( see lusso & ciotti 2011 ) . for example , at low accretion rates @xmath139 , while when @xmath20 would diverge to infinity , one reaches the asymptotic accretion rate @xmath140 we now apply to the bondi solution with electron scattering the same procedure of sect . 2.1 , to quantify the differences , as a function of radius , between the true ( @xmath74 ) and estimated ( @xmath72 ) bondi radius , and the true ( @xmath132 ) and estimated ( @xmath73 ) accretion rate , where @xmath72 and @xmath73 are defined as in eq . it is easy to show that : @xmath141 @xmath142^{5 - 3\gamma\over 2(\gamma-1)}= { 1\over\chi^2 } \left[{{r_{\rm b}}\over{r_{\rm e}}(x)}\right]^{5 - 3\gamma\over 2(\gamma-1)},\ ] ] where the density and temperature profiles are now those appropriate for accretion with electron scattering . for @xmath143 , by definition , @xmath144 . again , as for the classical bondi problem in sect . 2.1 , @xmath72 is always smaller than @xmath74 for @xmath47 , since the gas sound speed increases inward is easily understood when considering that the problem with electron scattering is just the classical bondi problem on a mbh of reduced mass equal to @xmath145 . ] ; for @xmath146 , @xmath84 independently of the distance @xmath65 from the center . for the mass accretion rate , again in analogy with what found for the classical bondi problem , we have that @xmath147 for @xmath148 ; that , for @xmath96 , @xmath149 independent of @xmath65 ; and that , for @xmath146 , @xmath150 . note that now the resulting bias depends not only on the distance @xmath65 where the density and temperature are taken , but also on the value of @xmath136 , through the parameter @xmath124 . the asymptotic analysis shows that near the center : @xmath151 @xmath152 in particular , the r.h.s . in eqs . ( 35)-(36 ) are just the r.h.s . ( 22)-(23 ) for the classical bondi problem , multiplied by @xmath153 and @xmath154 , respectively . therefore , @xmath78 again decreases for @xmath47 and @xmath155 . for @xmath156 , @xmath72 scales linearly with @xmath65 , as @xmath157 , and again @xmath99 , but by a larger factor than in classical bondi accretion . for @xmath158 , @xmath73 overestimates the true accretion rate @xmath132 ; for @xmath146 , @xmath159 . again , as in the classical bondi problem , it is possible to recover the true accretion rate from @xmath72 and @xmath73 near the center , as @xmath160 . the resulting quadratic equation for @xmath132 can be easily solved after writing the correction coefficient @xmath124 in terms of @xmath132 , following the procedure described below eq . ( 30 ) . the properties above are illustrated by fig . 2 , that shows the numerical and asymptotic solutions for @xmath78 [ eqs . ( 33 ) and ( 35 ) ] , and for @xmath161 [ eqs . ( 34 ) and ( 36 ) ] , as a function of the distance @xmath65 , for different values of @xmath123 , and @xmath13 fixed to the representative value 1.4 . the figure gives a quantification of the bias on the estimates of the bondi radius and mass accretion rate : @xmath72 is always an underestimate of @xmath74 , as expected , while @xmath162 is always an overestimate of the true accretion rate , even by a large factor if @xmath163 ( and , of course , increasing for larger @xmath123 ) . we can now discuss the full problem , i.e. , we investigate how standard bondi accretion is modified by the additional potential of the host galaxy , and by electron scattering . we then generalize the previous conclusions about the fiducial bondi radius @xmath72 and mass accretion rate @xmath73 obtained using quantities at finite distance from the mbh . we assume spherical symmetry for the host galaxy , so that its gravitational potential can be written in full generality as @xmath164 where @xmath165 , @xmath166 and @xmath167 are respectively the total galaxy mass , a characteristic scale - length , and the dimensionless galaxy potential . by introducing the parameters @xmath168 the effective total gravitational potential to be inserted in the expression for the function @xmath66 can be written as : @xmath169.\ ] ] the addition of the galaxy potential @xmath170 changes the function @xmath66 in eqs . ( 11 ) and ( 16 ) , and consequently also the values of @xmath171 and of the critical @xmath67 ( that now we call @xmath172 ) ; the function @xmath39 is unaffected by the addition of the galaxy . thus , @xmath172 is now given in the polytropic case by eq . ( 12 ) with the new value of @xmath42 , and again the isothermal case can be obtained as a limit for @xmath60 of the polytropic problem . for a generic galaxy model , it is no longer possible to obtain an analytical expression for @xmath173 , @xmath42 , and @xmath172 , and they must be determined numerically . as we will see in sect . 5 , the galaxy potential can produce more than one minimum for the function @xmath66 ; in this case it is easy to conclude that the general considerations after eqs . ( 11 ) and ( 16 ) refer to the _ absolute _ minimum of @xmath66 , and so @xmath173 gives - along the critical solution - the location of the _ sonic point _ obeys the identity @xmath174 where @xmath175 is the total gravitational potential . ] . of course , when @xmath176 ( or @xmath177 ) , the galaxy contribution vanishes , @xmath178 , and the formulae of sect . 3 hold . instead , by setting @xmath179 , one can determine the sole gravitational effects of the host galaxy . in appendix a we describe how to compute the correction terms for @xmath67 due to the presence of a generic galaxy model , in some special cases , while in appendix b an important property of monoatomic adiabatic accretion in generic galaxy potentials is derived , i.e. , that @xmath180 , where the last identity comes from eq . ( 28 ) with @xmath96 . following the procedure described in sects . 2 and 3 , we evaluate the deviation of @xmath72 and @xmath73 from the true values of @xmath74 and of the mass accretion rate , that we now indicate as @xmath181 . the presence of a galaxy changes the accretion rate on the central mbh ; eq . ( 9 ) still holds , but now : @xmath182 where @xmath20 is the bondi accretion rate , for the same chosen boundary conditions @xmath11 and @xmath25 , in absence of the galaxy and of radiation pressure . the expressions for @xmath72 and @xmath73 are given by the analogous of eqs . ( 20)-(21 ) and ( 33)-(34 ) , where now the physical variables are taken along the solution in presence of the galaxy and of electron scattering . ( 19 ) then one has : @xmath183 and @xmath184^{5 - 3\gamma\over 2(\gamma-1)}.\ ] ] for @xmath143 , by definition , @xmath144 , and , @xmath185 . as in the previous simpler accretion cases , for @xmath146 one has that @xmath84 , independently of the distance from the center , and @xmath186 . for @xmath96 , and from the result of appendix b , @xmath187 , independent of @xmath65 . additional properties of @xmath172 will be discussed in the next section . repeating the asymptotic analysis of sects . 2 and 3 , near the center we now derive : @xmath188 @xmath189 as in the other cases , @xmath78 decreases for @xmath22 and @xmath190 . for @xmath96 , @xmath72 scales linearly with @xmath65 , as @xmath191 ( because , for @xmath96 , @xmath180 , see appendix b ) , and again @xmath99 for @xmath192 , but by a larger factor than in classical bondi accretion . remarkably , near the center it is again possible to obtain the true @xmath181 by using the fiducial @xmath73 and @xmath72 . in fact , for @xmath155 , @xmath172 cancels out from eqs . ( 42)-(43 ) , and @xmath193 . this quadratic equation can be easily solved for @xmath181 as in sect . 3 , after expressing the radiative correction coefficient @xmath124 in term of @xmath181 , following the procedure described below eq . ( 30 ) . before proceeding with the solution of this accretion problem , we need to examine whether @xmath1 changes in presence of a galaxy . if this is the case , then the dimensionless accretion luminosity @xmath194 and the associated @xmath124 would be a function of the specific model investigated . however , it can be shown that for any given galaxy model , characterized by the cumulative mass distribution @xmath195 , the eddington luminosity depends on radius as @xmath196{l_{\rm edd}},\ ] ] where @xmath1 is the classical value whose expression is given below eq . therefore , @xmath197 is minimum at the center , with @xmath198 , and steady accretion - also in presence of a galaxy - can not exist for @xmath199 . this implies that @xmath123 and @xmath124 , also in presence of a galaxy , are still defined as in sect . in order to provide quantitative estimates for the trends of @xmath105 and @xmath200 , we need to adopt a specific galaxy model , to determine @xmath172 . as already noticed , the value of @xmath172 is known once the absolute minimum @xmath201 is known ; this , in turn , requires the determination of @xmath173 . as a realistic galactic potential we consider that corresponding to the hernquist ( 1990 ) density profile , that describes well the mass distribution of early - type galaxies ; however , we stress that several results below remain true also for other galaxy models , such as the so - called @xmath13-models ( dehnen 1993 , tremaine et al . the gravitational potential of the hernquist model is : @xmath202 where the scale - length @xmath166 is related to the galaxy effective radius @xmath203 as @xmath204 . thus , from eq . ( 39 ) for @xmath22 one has : @xmath205 and the formula for the isothermal case is obtained from eq . ( 16 ) . before presenting the numerical results for the hernquist galaxy model , a preliminary discussion is useful in order to obtain hints on the expected behavior of the solution . for the hernquist model the analysis is sufficiently simple to prove some interesting results , yet the case already illustrates the difficulties encountered in accretion problems in galaxy potentials . basically , the following analysis focuses on the behavior ( number and location ) of the critical points of @xmath66 , on the determination of the absolute minimum @xmath42 , and finally on the dependence of @xmath172 on the model parameters . the full analysis of the number and location of minima of eq . ( 47 ) is given in appendix c , where eq . ( c1 ) gives the expression for @xmath206 when @xmath47 ; for @xmath146 , @xmath206 is obtained as the limit for @xmath60 of the expression for @xmath206 in eq . ( c1 ) . from the expression for @xmath40 one sees that in three cases the determination of @xmath173 and @xmath42 is trivial . the first is when @xmath207 ( or @xmath176 ) , then the galaxy contribution vanishes , and eqs . ( 27)-(28 ) are recovered . the second case corresponds to @xmath208 , when the problem reduces to bondi accretion onto a mbh of mass @xmath209 . for these two cases the position of the only minimum of the function @xmath66 ( i.e. , of the sonic radius ) , and the critical value @xmath172 , for @xmath210 , are given by : @xmath211 where @xmath67 is the critical value for the corresponding classical bondi problem ( sect . a third simple case is when @xmath212 but @xmath96 : @xmath206 is definite positive [ eq . ( c3 ) ] , so that the minimum of @xmath66 is reached at the center ( as for the adiabatic bondi problem without the galaxy ) . simple algebra shows that the hernquist model with @xmath213 obeys the ( very weak ) hypoteses of appendix b , and so the general identity @xmath214 holds . note that this value differs from that in eq . ( 48 ) pertinent to the @xmath208 case : in fact , in appendix b we also show that the @xmath96 case for the hernquist model is singular for @xmath215 , as also manifested by the fact that @xmath216 while @xmath217 . as a strictly related result , again from appendix b , it follows that accretion in the @xmath96 case is impossibile in a generic galaxy models without a central mbh , because @xmath218 and so @xmath219 ( as confirmed by setting @xmath220 in the identity above ) . as shown in appendix c , for accretion in the general hernquist galaxy model with central mbh and electron scattering , @xmath173 can be obtained by solving an algebraic ( cubic ) equation , thus providing in principle the explicit expression for @xmath42 and @xmath172 , at variance with the common situation encountered for other galaxy models . the solution presents no difficulites , however it is quite cumbersome , and here we just recall that , depending on the specific values of @xmath221 , @xmath222 and @xmath13 , there can be a single minimum for @xmath66 , or two minima and one maximum . in appendix c we provide analytical formulae that can be easily implemented in numerical studies , allowing to determine , for any given choice of @xmath221 , @xmath222 and @xmath13 , the number and the location of the critical points of @xmath66 . altough the explicit expression for @xmath172 can be easily constructed , here we just give some hints on its behavior by considering cases of perturbative analysis ( namely , for @xmath177 or @xmath176 , and for @xmath215 ) , along the lines described in appendix a ( eq . a6 and following for details ) . for example , for @xmath26 , the results in appendix c3 show that there is only one critical point for @xmath66 in the considered cases , and expanding around the unperturbed minimum ( whose position is given in eq . 27 for @xmath177 or @xmath176 , and in eq . 48 for @xmath215 ) , one finds that the leading term of @xmath172 is : @xmath223 where @xmath224 . a similar analysis can be performed for @xmath225 . a careful expansion around the absolute minimum of @xmath66 , as determined from appendix c , shows that for @xmath177 ( or @xmath176 ) , and for @xmath215 we have respectively : @xmath226}\right]^{\gamma+1\over 2(\gamma-1 ) } , } \\ \\ \displaystyle{(\chi + { { \cal r}})^2 \left[1-{16\ { { \cal r}}\ \xi ( \gamma-1)\over ( 1+{{\cal r}})^2 ( \gamma+1 ) ( 5 - 3\gamma ) } \right]^{\gamma+1\over 2(\gamma-1)}. } \end{cases}\ ] ] note how eq . ( 50 ) for @xmath227 recovers both cases of eq . ( 49 ) . as expected , however , the second of eq . ( 50 ) is singular for @xmath228 , for the reasons described in appendix b. note that , at variance with the bondi accretion with electron scattering only , now the electron scattering coefficient @xmath124 can not be factorized in the expression for @xmath172 , i.e. , @xmath172 in general can not be factorized as eq . ( 28 ) as the product of the electron scattering coefficients times a function relative to the galaxy accretion without electron scattering effects . it follows that in presence of a galaxy the procedure described in sect . 4 [ eqs . ( 31)-(32 ) ] can not be applied analytically . we stress that all the formulae above have been also derived by direct expansion of the exact solution of eq . ( c2 ) , thus providing an independent test of the expansion procedure in appendix a. for a hernquist galaxy potential it is not possible to obtain the analytical expression for the critical accretion parameter @xmath172 ; thus , we determined numerically the density and temperature profiles of the bondi model with electron scattering and a hernquist galaxy , and calculated @xmath172 , @xmath72 and @xmath229 . the results for @xmath172 are plotted in figure [ fig : ltot ] , where we show the trend of @xmath172 with respect to @xmath230 , for fixed @xmath231 and @xmath232 , with respect to @xmath233 for @xmath234 and @xmath232 , and finally with respect to @xmath123 for @xmath234 and @xmath231 . in the left panel , for a fixed @xmath222 , for which we assumed a realistic value is of the order of few tens of pc , as expected for the mbhs at the center of early type galaxies , the value @xmath231 corresponds to @xmath166 of few kpc , i.e. , a reasonable value for @xmath235 . ] of 100 , @xmath172 increases as @xmath221 increases , i.e. , as the galaxy total mass increases . in the middle panel , by setting @xmath234 , as dictated by the magorrian relation ( magorrian et al . 1998 ) , we see that @xmath172 decreases as @xmath222 increases , since the effect fo the galaxy is vanishing ( and in fact @xmath172 reaches down to @xmath130 ) . finally , in the right panel @xmath172 goes to zero for increasing values of @xmath123 . note that for the adopted , illustrative values of @xmath221 and @xmath222 , the analysis in appendix c3 ( see also fig . c1 ) shows that there are three critical points for @xmath66 ( two minima and one maximum ) , so that @xmath172 is obtained from the absolute minimum . these results for @xmath172 are also compared to those obtained for the previous simpler models , i.e. , the @xmath67 of the classical bondi model , the lower @xmath130 of the bondi model with electron scattering , and that of the bondi model with only the hernquist galaxy . the latter keeps always larger than @xmath172 , since the effect of the electron scattering is to lower @xmath67 ( eq . 28 ) ; of course , it tends to @xmath67 of the classical bondi model for @xmath176 ( i.e. , for @xmath236 ) , and for @xmath207 ( i.e. , very large @xmath237 , and then negligible effect of the galaxy ) . also , @xmath238 for @xmath176 and for @xmath207 ( both cases in which the effect of the galaxy potential is vanishing ) . in fig . 4 we represent @xmath72 and @xmath229 , as a function of @xmath239 , for different values of @xmath13 and @xmath123 . the figure presents a quantification of the bias on the estimates of the bondi radius and of the true mass accretion rate . even in this more complex case , @xmath72 provides an underestimate of @xmath240 . for @xmath96 , @xmath241 if @xmath242 , otherwise @xmath162 is always an overestimate of the true accretion rate ( increasing for larger @xmath123 ) . for the other @xmath13 s , @xmath162 provides an overestimate for @xmath243 ( more severe for larger @xmath123 ) , and an underestimate for @xmath244 ( less severe for larger @xmath123 ) . due to its simplicity , the classical bondi accretion theory remains the standard paradigm against which more realistic descriptions are confronted , and observations are interpreted . also , bondi accretion is adopted to get a simple approximation for the mass accretion rate in semi - analytical models and numerical simulations , particularly when numerical resolution is not high enough to probe in a self - consistent way the regions near the central mbh . given the wide use of the classical bondi theory , and considered that accretion on the mbhs at the center of galaxies is certainly more complicated than the description provided by this theory , the motivation for this work was to generalize the theory , including the radiation pressure feedback due to electron scattering in the optically thin approximation , and the effect of a general gravitational potential due to a host galaxy . all the hypotheses of classical bondi accretion ( stationarity , absence of rotation , spherical symmetry ) were maintained . in addition , the present work provides a quantititative answer to a major question , namely what is the bias induced on the estimates of the bondi radius and mass accretion rate when adopting as bona - fide values of the hydrodynamical variables their values at some finite distance from the center . this issue is relevant for observational ( e.g. , pellegrini 2005 , 2010 ; russell et al . 2015 ) and numerical works ( e.g. , di matteo et al . 2005 , park et al . 2014 , degraf et al . the main results can be summarized as follows . \1 ) for the three models of classical bondi accretion , of accretion with electron scattering , and of accretion on a mbh at the center of a galaxy with electron scattering , we provide the exact formulae for @xmath78 ( where @xmath74 is fixed for a chosen value of @xmath245 ) , as a function of the distance @xmath65 , the mach number @xmath246 , and the accretion parameter ; these can be used when the temperature is taken at any @xmath65 along the `` modified '' bondi solution . we then give the exact formulae for the mass accretion bias ( respectively @xmath77 , @xmath161 , and @xmath247 ) in terms of the bias on the bondi radius ( @xmath78 ) . for a quantitative estimate of the bias , the knowledge of the numerical solution of the associated accretion problem is required . \2 ) the formulae above allow to get some general results without previous knowledge of the exact numerical solution , for particular @xmath13 values . in the monoatomic adiabatic case ( @xmath96 ) , @xmath73 is a constant fraction of the true mass accretion rate ( @xmath20 , or @xmath132 , or @xmath248 ) , independently of the distance @xmath65 from the mbh . the fraction is exactly unity in case of classical accretion ( @xmath249 ) , and it is given by the ratio of the critical accretion parameters in the other cases : with radiation pressure , @xmath250 ; with electron scattering and a galaxy potential , @xmath251 ( see sect . 5.1 for more details and the particular case of a hernquist galaxy ) . in the isothermal ( @xmath26 ) case , there is no bias in the estimated value of the bondi radius ( @xmath84 ) , independently of the distance from the center ; the bias provided by @xmath73 is proportional to the density of the accreting material . \3 ) the trends of @xmath73 and @xmath72 near the center come from the asymptotic expansion of the solutions for @xmath252 and @xmath253 : the bias can be written in terms of the distance @xmath65 and of the value of the critical accretion parameter . in particular , for @xmath47 , @xmath72 provides a larger and larger underestimate of @xmath74 as @xmath65 is approaching the center [ in all cases @xmath254 . @xmath73 correspondingly provides an overestimate of @xmath20 , @xmath132 , and @xmath181 . \4 ) from the asymptotic expansion of the solutions near the center , for all three accretion models considered , it is also shown how to recover the true value of the mass accretion rate from @xmath72 and @xmath73 measured at some distance @xmath65 from the center . these formulae are useful for observational and numerical works . \5 ) when the analytical solution is known for an accretion case , a general asymptotic technique is given that allows to obtain the correction terms for the value of the critical accretion parameter for slightly different models , avoiding the need of a numerical solution of the accretion problem ( appendix a ) . \6 ) in the illustrative case of a hernquist galaxy model , the determination of @xmath173 and @xmath172 reduces to the study of a cubic equation . the case is fully discussed analytically ( appendix c ) , and it is shown that more than one critical point for @xmath66 can exist . we provide a simple analytical framework to determine whether one or two minima occur , and the correction terms for @xmath172 , determined with the method of point 5 ) . \7 ) we finally solved numerically the accretion problem for the three models considered , to provide quantitative estimates of the bias . the analytical formulae turned out to be in perfect agreement with the numerical results . this work reveals how using values of density and temperature at various radii when deducing accretion properties of mbhs at the center of galaxies produces values of the mass accretion rate that should be taken with care . we here obtained , however , results sufficiently general to provide correction factors to be used in various observational and numerical works . > from the discussion in sect . [ sec : class ] it follows that the critical value of the accretion parameter @xmath67 in the bondi theory is given by eq . therefore , @xmath67 can be computed explicitly when @xmath42 is . for example , this is possible in the classical bondi problem ( also in presence of electron scattering , as shown in sect . 3 ) , but in general this is impossible in presence of the galaxy potential . for this reason in order to determine the value of @xmath67 we must resort to numerical evaluation of @xmath42 . it is then useful to be able to determine analytically the first correction terms to the value of @xmath67 due to the presence of the galaxy . in the following we present the general procedure for such determination . the first step is to cast the function @xmath40 of the problem under scrutiny as the sum of two terms : one ( @xmath255 ) so that the position of the minimum , @xmath256 , can be explicitly determined , @xmath257 . the second term ( @xmath258 ) is the perturbation term , depending on some small ordering parameter @xmath259 . in practice , we suppose we can write @xmath260 with @xmath261 . if all the required regularity conditions are satisfied , the position of the new minimum , @xmath262 , is given by @xmath263=0,\ ] ] with @xmath264 by expansion of eq . ( a2 ) and order balance one can in principle determine the perturbation coefficients in eq . ( a3 ) at the desidered order . for example , provided that @xmath265 , it can be proved that @xmath266_{(\epsilon , x)=(0,{x_0})},\ ] ] and in such cases when @xmath267 @xmath268_{(\epsilon , x)=(0,{x_0})}.\ ] ] once the coefficients are determined , @xmath42 is obtained by expansion of eq . ( a1 ) : for example , in case of a regular perturbation @xmath258 , it is easy to show that at the first order @xmath269_{\epsilon=0}.\ ] ] in sect . 5 simple perturbation cases are described for the case of an hernquist galaxy . one is obtained for @xmath177 , so that @xmath270 , the unperturbed problem is the bondi accretion ( with or without electron scattering ) , with @xmath255 given in eq . ( 26 ) , and @xmath256 given by eq . ( 27 ) , while @xmath271 . a second , strictly related case , is that of @xmath272 so that the unperturbed problem is the same as in the previous case , but now @xmath273 . finally , we consider @xmath274 . in this case the galaxy core radius vanishes , so that the unperturbed problem is obtained by considering the bondi accretion on a mbh of mass @xmath209 , with unpertburbed solution @xmath256 given by eq . ( 48 ) , while @xmath275 . here we show that for @xmath96 the function @xmath206 relative to bondi accretion with electron scattering and in presence of a galaxy , is definite positive for @xmath276 , so that @xmath66 is monotonically increasing and the ( only ) minimum over the physical domain is attained for @xmath277 . in fact , for a spherical galaxy model of density profile @xmath278 , finite total mass @xmath165 , and with a central mbh of mass @xmath2 , the total potential in presence of electron scattering can be written as ( see eq . 39 and binney & tremaine 1987 ) @xmath279 with the introduction of the dimensionless total potential @xmath280 , the general form of eq . ( 47 ) reads @xmath281 so that @xmath282 where @xmath283 , and this concludes the proof . it follows that for model galaxies with @xmath284 , for @xmath96 one has @xmath285 so that @xmath180 . the hernquist model with @xmath213 satisfies this hypothesis , but not when @xmath208 . this is the reason behind the fact that the @xmath96 case for the hernquist model is singular , in the sense that the properties of the @xmath208 case can not be obtained as the limit for @xmath215 , as mentioned in sect . . note that for @xmath286 eq . ( 47 ) shows that the exponent of @xmath106 in eq . ( b2 ) is @xmath287 , and the procedure above now confirms that @xmath206 can change sign , depending on the specific shape of @xmath288 . in the general case of bondi accretion with electron scattering in a hernquist galaxy with a central mbh and @xmath289 ( the case @xmath208 is trivial , as discussed in sect . [ sec : hernquistgal ] ) , the critical points of @xmath66 are placed at the zeroes of @xmath290 where @xmath291\over 2 } x -{\xi^2(5 - 3\gamma)\chi\over 4 } , \end{split}\ ] ] as obtained from eq . ( 47 ) , and @xmath292 . in the isothermal and in the adiabatic monoatomic cases this reduces to @xmath293 , } \end{cases}\ ] ] respectively , and in the last case the positivity for @xmath294 is apparent , in accordance with the general result of appendix [ app : adia ] . note that in the limiting case of @xmath220 , i.e. when formally @xmath295 , the problem reduces to bondi accretion ( without electron scattering ) in the galaxy potential well only : therefore the obtained formulae can used also to discuss this special class of solutions . for @xmath296 , @xmath213 and @xmath297 , the constant term in eq . ( c2 ) is negative , while the cubic term is positive . this means that @xmath66 presents always at least one minimum at @xmath294 . however , it may happens that for specific values of the parameters there are three positive zeros of @xmath66 , corresponding , for increasing @xmath106 , to a minimum , a maximum , and a minimum of @xmath66 , respectively . this is relevant not only for the theoretical implications but also for numerical investigations . in fact , the position of the absolute minimum of @xmath66 is usually obtained by numerical analysis , and it is of great help to know whether there just one minimum or it is necessary to investigate what of the minima is the absolute one , as required in order to compute the critical accretion parameter @xmath67 . for assigned values of the parameters the existence and the position of the zeros of @xmath206 can be determined by using the sturm method or from the theory of cubic equations . in fact , after standard reduction of eq . ( c2 ) to canonical form @xmath298 , one consider the sign of the quantity @xmath299 . when @xmath300 the original equation has one real ( and so positive ) root and two complex conjugate roots . for @xmath301 , there are three real roots , and for @xmath302 there are a real root and a double ( and so real ) root . in the case of three real roots , their positivity can be determined from the descartes sign rule . the function @xmath303 associated with eq . ( c2 ) is the product of the factor @xmath304 times a cubic polynomial that can be ordered as a function of @xmath222 or @xmath221 as : @xmath305\xi - \\ & \ ( 5 - 3\gamma)^2(\chi+{{\cal r}})^2 \left [ 4(\gamma-1)^2{{\cal r}}-p_6 \right]=\\ & \ - 4(5 - 3\gamma)^2(\gamma-1){{\cal r}}^3 - ( 4p_1\xi+p_2 ) { { \cal r}}^2 + \\ & \ 2(p_3\xi^2 - 2p_4\xi -p_5){{\cal r}}+ ( \gamma+1)\left[4\xi+(5 - 3\gamma ) \chi \right]^3 , \end{split}\ ] ] where the coefficients are @xmath306 for assigned values of @xmath307 , and @xmath124 , is then straightforward to determine numerically the number of real zeros of @xmath28 by using eqs ( c4)-(c5 ) , and successively their positivity checking the number of sign variations in the coefficients of eq . ( c2 ) . in the next sections we will use eq . ( c4 ) to deduce some analytical property of the critical points in some limit case . in terms of @xmath222 , the leading term of eq . ( c4 ) is positive , so that @xmath300 for fixed @xmath221 and sufficiently large values of @xmath222 , and there is only one minimum for @xmath66 , as expected from physical considerations . the displacement of the minimum position with respect to eq . ( 27 ) can be determined by using the method in appendix a. for very small values of @xmath222 and @xmath308 instead the problem presents three real solutions for @xmath309 : a careful application of descartes rule shows that all the three zeros are positive . in fact , reducing @xmath222 at fixed @xmath221 , the function @xmath66 flattens in the outer regions , until @xmath302 and an horizontal flex appears , while the only minimum is placed near the position @xmath173 in eq . ( 48 ) : again , the correttive terms can be obtained from appendix a. reducing further @xmath222 , the double solution splits , and the minimum - maximum - minimum structure appears . the outer minimum deepens until it becomes the absolute minimum , approaching @xmath173 given in eq . ( 48 ) , while the other two zeroes merge at @xmath310 . with some work it is possible to obtain the first terms of the asymptotic expansion for @xmath215 of the positions of the three critical points of eq . ( c2 ) , ordered for increasing distance from the origin : @xmath311 where @xmath312 $ ] . it is apparent how the reality condition for @xmath313 and @xmath314 the first two zeroes matches the reality condition obtained from eq . ( c4 ) . moreover , again with some work from eq . ( c4 ) , it can be proved that the reality condition of the three zeroes for @xmath315 and small @xmath222 requires that @xmath316 , so that the three critical points in eq . ( c6 ) collapse into the origin for @xmath317 , consistently with the general result of appendix b for adiabatic accretion . a different situation is obtained when considering eq . ( c4 ) as a function of @xmath221 . it is apparent how @xmath300 for @xmath176 , and there is only one real solution , as expected , because the problem reduces to the classical bondi problem . the perturbed position of the minimum can be determined again from appendix a , with @xmath318 given in eq . instead , @xmath301 for very large values of @xmath221 and @xmath22 , and so eq . ( c2 ) admits three real solutions . descartes sign rule shows that the three zeros are positive . by using the order - balance technique it can be shown that their asymptotic leading term at fixed @xmath13 and @xmath222 ( in order of increasing distance from the center ) is given by @xmath319 one could be worried by the possibility that for @xmath315 , the solution @xmath314 becomes larger than @xmath320 : however it can be shown that once the reality condition is verified this can not happens , because in order to have @xmath301 , the mass ratio @xmath221 must diverge faster than @xmath321 for @xmath315 . the @xmath26 case is discussed in the following section . a particularly simple case is the isothermal one . in fact , for @xmath26 eq . ( c4 ) leads to the study of the sign of the quadratic polynomial @xmath322 . therefore , at variance with the @xmath22 cases , in isothermal accretion @xmath300 for large values of @xmath221 and fixed @xmath222 , and only one ( positive ) minimum exists , with the sonic point given by the last identity in eq . ( c7 ) with @xmath26 . in addition , @xmath300 for large values of @xmath222 and fixed @xmath221 , with the sonic point placed at @xmath173 in eq . ( 27 ) with @xmath26 . finally @xmath300 independently of @xmath221 when @xmath323 , and so again there is only one ( positive ) minimum . for @xmath324 there are two positive values of the galaxy - to - mbh mass ratio @xmath325 so that @xmath301 for @xmath326 : descartes sign rule shows that the three zeros are all placed at @xmath294 . in particular , for @xmath207 , @xmath327 and @xmath328 . summarizing , in the isothermal case , for @xmath329 there is only one minimum @xmath330 , for @xmath331 there is only one minimum @xmath332 , and there is only one minimum for large values of @xmath221 ( @xmath222 ) and fixed @xmath222 ( @xmath221 ) . the situation is clearly illustrated in fig . l.c . is grateful to g. bertin and j. ostriker for useful discussions . l.c . and s.p . were supported by the miur grant prin 2010 - 2011 , project `` the chemical and dynamical evolution of the milky way and local group galaxies '' , prot . 2010ly5n2 t . ciotti , l. and ostriker , j. p. , 2012 , agn feedback in elliptical galaxies : numerical simulations , in _ hot interstellar matter in elliptical galaxies _ , astrophysics and space science library , vol . 378 , p.83 , eds . kim , s. pellegrini cowie , l. l. , ostriker , j. p. , stark , a. a. , 1978 , apj , * 226 * , 1041 curtis , m. , sijacki , d. , 2015 , mnras , * 454 * , 3445 datta , s. , 2016 , arxiv:1601.01307 degraf , c. , di matteo , t. , treu , t. , feng , y. , woo , j .- h . , park , d. 2015 , mnras , * 454 * , 913 dehnen , w. , 1993 , mnras , * 265 * , 250 frank , j. , king , a. and raine , d. , 1992 , _ accretion power in astrophysics . _ , camb . astrophys . 21 , cambridge university press hernquist , l. , 1990 , apj , * 356 * , 359 hirschmann , m. , dolag , k. , saro , a. , bachmann , l. , borgani , s. , burkert , a. , 2014 , mnras , * 442 * , 2304 hopkins , p. , hernquist , l. , cox , t.j . , 2006 , apjs , * 163 * , 1 jaffe , w. , 1983 , mnras , * 202 * , 995 krolik , j. h. , 1998 , _ active galactic nuclei : from the central black hole to the galactic environment _ , princeton university press loewenstein , m. , mushotzky , r. f. , angelini , l. , arnaud , k. a. , quataert , eliot 2001 , apj , * 555 * , 21 lusso , e. , and ciotti , l. , 2011 , a&a , 525 , 115 magorrian , j. , tremaine , s. , richstone , d. , et al . 1998 , aj , * 115 * , 2285 mcnamara , b. r. , rohanizadegan , m. , nulsen , p. e. j. 2011 , apj , * 727 * , 39 narayan , r. and yi , i. , 1995 , apj , * 452 * , 710 pacucci , f. , volonteri , m. , ferrara , a. 2015 , mnras , * 452 * , 1922 park , k. , ricotti , m. 2011 , apj , * 739 * , 2 park , k. , ricotti , m. , di matteo , t. , et al . 2014 , mnras , * 445 * , 2325 pellegrini , s. 2005 , apj , * 624 * , 155 pellegrini , s. 2010 , apj , * 717 * , 640 quataert , e. and narayan , r. , 2000 , apj , * 528 * , 236 rafferty , d. a. , mcnamara , b. r. , nulsen , p. e. j. , & wise , m. w. 2006 , apj , * 652 * , 216 russell , h. r. , fabian , a. c. , mcnamara , b. , broderick , a. e. , 2015 , mnras , * 451 * , 588 taam , r. e. , fu , a. , fryxell , b. a. , 1991 , apj , * 371 * , 696 tremaine , s. , richstone , d. o. , byun , y .- i . , et al . 1994 , aj , * 107 * , 634 volonteri , m. , rees , m.j . 2005 , apj , * 633 * , 624 wong , ka - wah , irwin , j.a . , shcherbakov , r. v. , yukita , m. , million , e. t. , bregman , j. n. 2014 , apj , * 780 * , 9 wyithe , j. s. b. , loeb , a. 2012 , mnras , * 425 * , 2892

we present a comprehensive re - analysis of stellar photometric variability in the field of the open cluster m37 following the application of a new photometry and de - trending method to mmt / megacam image archive . this new analysis allows a rare opportunity to explore photometric variability over a broad range of time - scales , from minutes to a month . the intent of this work is to examine the entire sample of over 30,000 objects for periodic , aperiodic , and sporadic behaviors in their light curves . we show a modified version of the fast periodogram algorithm ( f ) and change - point analysis ( cpa ) as tools for detecting and assessing the significance of periodic and non - periodic variations . the benefits of our new photometry and analysis methods are evident . a total of 2306 stars exhibit convincing variations that are induced by flares , pulsations , eclipses , starspots , and unknown causes in some cases . this represents a 60% increase in the number of variables known in this field . moreover , 30 of the previously identified variables are found to be false positives resulting from time - dependent systematic effects . new catalog includes 61 eclipsing binary systems , 92 multiperiodic variable stars , 132 aperiodic variables , and 436 flare stars , as well as several hundreds of rotating variables . based on extended and improved catalog of variables , we investigate the basic properties ( e.g. , period , amplitude , type ) of all variables . the catalog can be accessed through the web interface ( http://stardb.yonsei.ac.kr/ ) .

1505.04495

the rich open cluster m37 ( ngc 2099 ) in the constellation auriga has been photometrically monitored to search for new variables with different motivations @xcite . the combination of short and deep exposures with similar sky coverage allowed time - series photometry of hundreds to thousands of stars from the brightest to the dimmest within the cluster field ( @xmath1 ) . their studies indicate that the fraction of variable sources @xmath2 increases as photometric precision of the survey gets better . @xcite performed the first variability survey for the central ( @xmath3 ) field of the m37 with a @xmath4-filter photometry and found 7 variable stars among 2300 stars ( @xmath5 ) . due to poor photometric quality , their variables were with large variability and included three w uma - type eclipsing binaries ( ebs ) , two high - amplitude pulsating stars , and two long - period detached eb candidates . @xcite and @xcite used the same data set to search for short - period pulsating variables ( e.g. , @xmath6 scuti- and @xmath7 doradus - type stars ) and rotating variables of later spectral types ( f k ) , respectively . using @xmath8-band time - series observations , these two studies found the total of 153 variable stars among 12,000 stars ( @xmath9 ) , including seven previously known variables , in the field of the cluster ( @xmath10 ) . lastly , the most extensive variability studies in this field were carried out by @xcite as the by - products of a deep survey for transiting planets in the m37 field ( @xmath11 ) with the 6.5 m mmt telescope . the number of known variable stars increased by a factor of @xmath1210 . by applying several statistical algorithms , they identified 1445 variables among 23,000 stars ( @xmath13 ) , of which 99% were new discoveries @xcite . this catalog of variable stars provides the largest homogeneous set of stellar rotation periods for a cluster older than 500 myr , which showed the crucial role of the rotation in the stellar angular momentum evolution @xcite . in this paper , we search new variable phenomena from the same data set of @xcite with vastly improved photometric method ( @xcite , hereafter paper i ) . in section 2 , we give a brief description of new light curves and its temporal properties . we present a comprehensive re - analysis of photometric variability in section 3 . in section 4 , we present a new catalog of variable stars and various variability characteristics . we compare our findings with previous works and discuss the nature of each variability class separately in section 5 . between two consecutive frames ( top left ) and the exposure times @xmath14 ( top right ) . we use logarithmic scale for the horizontal axis in these two panels . the bottom panel is the distribution of exposure times as a function of the modified julian date ( mjd).,scaledwidth=50.0% ] we refer the reader to paper i for a detailed description of our photometric reduction and light curve production applied to the archival imaging data of mmt transit survey . here we provide only a brief summary of the key features . in paper i , we introduced new methods for robust high - precision photometry from well - sampled images of a non - crowded field with strongly varying psf . the archival light curves from the original image subtraction procedure exhibit many unusual outliers , and more than 20% of data get rejected by the simple filtering algorithm adopted by early analysis . we note that some of these outliers are results of intrinsic variability of short time scales . in order to achieve better photometric precisions and also to utilize all available data , the entire imaging database were re - analyzed with our multi - aperture indexing photometry and a set of sophisticated calibration procedures . this algorithm finds an optimal aperture for each star with a maximum signal - to - noise ratio , and also treat peculiar situations where photometry returns misleading information with more optimal photometric index . we also applied photometric de - trending algorithm based on a hierarchical clustering method . our method removes systematic variations that are shared by light curves of nearby stars while true variabilities are preserved . consequently , our method utilizes nearly 100% of available data and reduce the rms scatter several times smaller than original analysis for brighter stars . our new data set includes 30,294 objects , including both point and extended sources . for blended targets , the original image subtraction light curve is still of higher precision than the present aperture photometry light curves , and thus for these objects the archival light curve is used . in order to pick out such cases , we utilized a simple method for quantifying the level of blending ( see @xcite ) . @xmath15 where @xmath16 for light curve points @xmath17 with uncertainties @xmath18 , and @xmath19 is the same statistic measured with respect to a 4th order polynomial fit . we adopt the value @xmath20 for selecting significant blended light curves , and then count the number of blended ones ( 3746 objects ) . figure [ fig1 ] shows that the cumulative fraction function of temporal intervals @xmath21 between two consecutive frames and the exposure times @xmath22 . this observing cadence is sensitive to variability on timescales ranging from tens of seconds ( 30 seconds ) to one month ( @xmath1231.7 days ) . in particular , the time resolution allows an analysis of the short - term variability , which is less than @xmath1280 seconds over 50% of the data and @xmath12150 seconds over 90% of the data . variability features can be divided into two broad categories : periodic and non - periodic feature . the former is easier to detect and model using statistical and model - specific fitting procedures as a form of fourier series ( e.g. , @xcite ) . but the latter is rather ambiguous , since it is not always possible to characterize flux variation and to define a reliable range for some parameters . to overcome this difficulty , @xcite calculated many basic statistics that can describe the distribution of fluxes even in the limit of few data points . one interesting approach is to select stochastically varying features in quasar - like sources . the correlated variability of quasars is well parameterized by a damped random walk model , providing a simple statistical description of their characteristic timescales and amplitudes ( e.g. , @xcite ) . in this work we use a modified version of the fast @xmath0 periodogram algorithm ( f@xmath0 ) and the change - point analysis ( cpa ) method for detecting and assessing the significance of periodic and non - periodic signals , respectively . we first searched a periodic feature by f@xmath0 technique that calculates the minimized @xmath0 as a function of frequency at the desired number of harmonics @xcite . with the best - fit frequency , we compared the ratio between the variance of the data after subtracting the best - fit model and full - amplitude of model function . this algorithm use a fourier series truncated at harmonic @xmath23 as a model function : @xmath24 the fitting coefficients ( @xmath25 , @xmath26 , and @xmath27 ) obtained in each harmonic model provide a reasonable description of a generic periodic feature , irrespective of the feature shape , as discussed by @xcite . figure [ fig2 ] shows example light curves of both periodic ( new variable star ; v1983 ) and long - term ( v1092 ) variable stars in our sample , respectively . because the optimal choice of @xmath23 depends on the true shape of the signal , the fit quality is evaluated for each model with standard @xmath0 statistics , and then checked the residual ( phase - folded ) light curves after the subtraction of the best - fit model . to test the significance of a signal , we use the ratio between the full - amplitude of the model function ( @xmath28 = @xmath29 ) and the rms scatter ( @xmath30 ) of the data after model subtraction : @xmath31 the lower limit for the ratio is empirically set to @xmath32 = 0.95 based on light curve inspections . @xmath32-values less than 0.95 indicate that we do not expect any significant ( semi- ) periodic signals for a given light curve . even when the source is poorly described by the harmonic fit , this indicator provides a hint of a generic periodic signal ( see right panel of figure [ fig2 ] ) . most variability detection methods require conventional models that are mainly focused on the periodic signals , and are not suitable for the study of arbitrary - shaped , non - periodic , and sporadic variations . also , in many cases , signal estimation is equated with smoothing of data for de - noising . this sometimes discards vital information in time series data . we introduce a non - parametric method to extract all significant features based on the cpa using local statistics . change - point analysis is a method to identify abrupt variations in sequential data . it is widely used in the statistics and data mining communities as well as in the field of time domain astronomy ( see e.g. , @xcite and reference therein ) . using a combination of cumulative sum scheme ( cusum ) and bootstrap rank statistics @xcite , we produce a series of estimated change - points which correspond to the moments of apparent systematic changes . for a given dataset \{@xmath33 , @xmath34 , @xmath35 , @xmath36 } of size @xmath37 , the cusum values are given by @xmath38 for @xmath39 = @xmath40 , where @xmath41 = 0 and the mean value @xmath42 for the sample of @xmath37 values of @xmath43 . an inflection point at which the sign of the cusum slope changes is used to determine whether a given interval of data should be kept as one ( @xmath44=@xmath45=@xmath35=@xmath46=@xmath42 ) or subdivided into two subintervals ( @xmath44=@xmath35=@xmath47@xmath48@xmath49=@xmath46 ) . based on @xmath50 bootstrap samples that are randomly re - ordered original values , we estimate the confidence level ( c.l . ) to reduce the false positives by random noise . here are the main procedure of each bootstrap process : * we generate a bootstrap sample of @xmath37 dataset , denoted \{@xmath51 , @xmath52 , @xmath35 , @xmath53 } , by randomly re - ordering the original @xmath37 values without replacement . * based on the bootstrap dataset , we calculate the bootstrap cusum @xmath54 by using equation ( 4 ) , and then obtain the maximum , minimum , and difference of the bootstrap cusum . @xmath55 where @xmath56 @xmath57 * we check whether the bootstrap difference @xmath58 is less than the original difference @xmath59 . it is straightforward to derive estimates of confidence level as follows : @xmath60 where @xmath61 typically more than 90% confidence level is suitable for most cases . when a sudden change is detected , the location of change - point is initially determined as follow : @xmath62 } \left|s_{t}\right|,\ ] ] where @xmath63 denotes the last point before the change occurred . the time - series data is split into two segments on each side of the change point , and the analysis is repeated for each segment until no more significant change point is detected . once this set is generated , all the change - point locations and their confidence levels are re - estimated in a backward elimination manner . using only the sample in the segment bounded by adjacent change - points , we re - calculate these parameters detected by the forward procedure to obtain more accurate estimations . when any initial change - point fails the criteria in the backward procedure , we eliminate this change - point estimate . this process repeats until no further improvement is possible . the final change points thus define the segments ( i.e. , piecewise constant level sets ) , characterized by the start and end time of a given interval with its mean and variance . after this procedure , signal detection process is performed for each segment separately . to detect significant outlying features occurring at specific levels in light curve , we adopt a simple criteria similar to micro - lensing alert system @xcite in the presence of hetero - scedastic measurement errors ( @xmath64 ) : @xmath65 @xmath66 @xmath67 @xmath68 where the mean @xmath69 and deviation @xmath70 are the local statistics for a given segment , @xmath64 is the photometric error at epoch @xmath71 , and @xmath72 is the number of consecutive points which satisfy the equations ( 8a8c ) . the different combination of @xmath73 values are used for maximize the detection efficiency of significant outlying features . the values of @xmath73 were taken to be at least larger than 3 , 1 , and 3 , respectively . as a result , statistically significant events were flagged as candidate flare sources if at least three contiguous points satisfy the above criteria . example of flare - like feature detection is shown in the left panel of figure [ fig3 ] . by applying the cpa method to the entire light curves , we efficiently identified several hundred instances of abrupt brightness changes without any smoothing or interpolation of the raw data . we use the same limits of cpa thresholds to find eclipsing - like event detection , except the first term of equation ( 8) , i.e. , @xmath74 . the cpa method is useful to detect the moments of eclipse ingress , center , or egress in cases where the eclipsing pattern is not repeated or the coverage is not sufficient for the detection through conventional period analysis ( e.g. , box - fitting least - squares method ) . example of eclipsing - like feature detection is shown in the right panel of figure [ fig3 ] . we will discuss this issue further in section 5.1 . for periodicity , we first used the vartools light curve analysis program to determine likely periods of variable stars . this code computes three different periodograms for the light curves;(i ) lomb - scargle ( ls : @xcite ) , ( ii ) analysis of variance ( aov : @xcite ) , and ( iii ) box - fitting least - squares ( bls : @xcite ) algorithms . because each method is sensitive to different types of periodic variables , it is often useful to combine results from many different analysis of periodicity . basically we adopt the similar treatment as in @xcite . we limited the search range of these algorithms to periods between 0.01 and 30 days with relatively fine frequency resolution of less than 0.01/@xmath75 ( @xmath75 is the time - span of the light curve ) . we also obtained period estimates from the statistical technique sigspec @xcite . it calculates the spectral significance ( sig ) levels based on the analytical solution for discrete fourier transform amplitude spectrum , including dependencies on frequency and phase . in recent years , this package is mainly used in the study of stellar pulsation ( i.e. , asteroseismology ) , which shows up as small , periodic variations in the brightness of the stars ( e.g. , @xcite ) . in order to further crosscheck results of periodicity analysis , we also performed a multi - step period search algorithm msperiod @xcite . figure [ fig4 ] shows examples of three selected light curves ; power spectrum and phased diagram corresponding to the highest peak . the differences in the various algorithms depend strongly on the shape of variation . one case ( v2256 ) shows that all periods are essentially identical ( @xmath76 0.0784 days ) , while in the other cases the derived period differ a lot . visual inspections are necessary for v2173 and v1482 to choose the most plausible periods ; @xmath77 and 2.595 respectively . we provide a catalog of the total 2306 stars that exhibits convincing variations that are induced by flares , pulsations , eclipses , starspots , or in some cases , unknown causes . this catalog is made available in electronic form at http://stardb.yonsei.ac.kr/. table [ taba1 ] gives the basic information about previously known and new variable stars that we detected in the field of m37 . the columns are as follows : : the variable star ids for the first 1483 stars are from the numbering system of @xcite . new variable i d numbers are given in ascending order by their right ascension from v1484 to v2306 . : each object is designated by the combination of right ascension and declination for j2000 epoch ( e.g. , 055536.78 + 323345.97 ) . : we basically used the dominant periods which were highly scored by several methods . for ebs , we checked the existence of secondary eclipse when phased at twice the period . in cases where our periodicity analysis is affected by a large scatter of light curves due to photometric bias or source blending , we used periods taken from the previous catalog @xcite , and then denoted by @xmath78 in that column . we also found evidences of multiperiodicity in some pulsating stars ( see section 5.2 ) . : the range of variability for periodic variables was derived from phase diagram in which the shape of the light curve was modeled by equation ( 2 ) . this amplitude is defined as the full - amplitude of model function ( @xmath79 = @xmath29 ) . when this was not possible , i.e. , cases of ebs and irregulars , we used the original light curve to estimate the amplitude . : @xmath80 magnitudes for many of our stars are available from the sloan digital sky survey data release 7 ( sdss dr7 ) and two micro all sky survey ( 2mass ) . each catalog is positionally matched to mmt sources with an average distance of 0.6@xmath81 . in the case of sdss , the photometric quality at low galactic latitudes is not guaranteed to be accurate to the sdss quoted limits of 2% in color , and photometric depths are about two magnitude shallower than our data . similarly , 2mass is not deep enough to detect low - mass members in the cluster . : we assigned five variability types similar to the variability classification scheme proposed by the gcvs @xcite . variable stars are grouped according to major astrophysical reasons for variability , such as flare ( f ) , pulsating ( p ) , rotating ( r ) , eclipsing binary ( eb ) , and aperiodic variables ( a ) , including variable candidates ( :) . further details are given in section 5 . if a variable belongs to more than one type of variability , it is indicated by combination of individual types , e.g. , f+eb , p+eb : . a total of 30 stars listed as variables in @xcite turned out to be not real variable in our study . their new amplitude spectrum often does not show anything significant ( figure [ fig5 ] ) and no apparent periodicity in the light curve . these false positives are listed in table [ tab1 ] . rcrcrr 1 & v27 & 055118.34 + 323201.71 & 32 & 20.938 & 16.364 + 2 & v85 & 055125.32 + 324341.29 & 28 & 20.862 & 7.668 + 3 & v88 & 055125.46 + 324315.32 & 28 & 21.820 & 1.646 + 4 & v120 & 055128.52 + 322339.10 & 35 & 17.196 & 6.859 + 5 & v155 & 055131.10 + 323756.98 & 30 & 16.847 & 7.424 + 6 & v166 & 055131.96 + 323600.24 & 31 & 20.895 & 10.451 + 7 & v224 & 055137.29 + 324011.78 & 29 & 18.165 & 4.571 + 8 & v230 & 055138.02 + 324011.65 & 29 & 16.690 & 4.595 + 9 & v250 & 055139.59 + 324013.27 & 29 & 17.282 & 4.551 + 10 & v330 & 055151.25 + 323527.70 & 22 & 16.831 & 5.428 + 11 & v337 & 055151.72 + 323510.02 & 22 & 17.920 & 5.442 + 12 & v342 & 055152.18 + 323522.44 & 22 & 19.061 & 5.438 + 13 & v347 & 055152.55 + 323536.03 & 22 & 15.574 & 5.500 + 14 & v367 & 055153.88 + 323459.77 & 22 & 17.463 & 5.515 + 15 & v390 & 055155.42 + 323523.56 & 22 & 18.268 & 5.481 + 16 & v408 & 055156.77 + 323534.61 & 22 & 15.733 & 5.481 + 17 & v433 & 055157.99 + 323528.59 & 22 & 16.617 & 5.481 + 18 & v492 & 055202.16 + 324517.30 & 19 & 20.849 & 16.586 + 19 & v973 & 055235.48 + 322104.58 & 18 & 16.847 & 5.089 + 20 & v1076 & 055241.54 + 322107.71 & 18 & 20.262 & 10.927 + 21 & v1237 & 055258.14 + 322152.09 & 9 & 16.775 & 5.692 + 22 & v1260 & 055300.53 + 323445.47 & 4 & 19.851 & 15.895 + 23 & v1282 & 055303.05 + 323621.95 & 4 & 21.530 & 1.666 + 24 & v1305 & 055305.05 + 323447.83 & 4 & 18.166 & 7.966 + 25 & v1330 & 055306.74 + 323442.52 & 4 & 18.061 & 8.143 + 26 & v1371 & 055310.61 + 323748.13 & 3 & 20.902 & 9.934 + 27 & v1372 & 055310.65 + 323740.82 & 3 & 20.985 & 10.633 + 28 & v1400 & 055312.89 + 323005.46 & 6 & 16.770 & + 29 & v1421 & 055314.64 + 322656.12 & 7 & 20.735 & 15.013 + 30 & v1423 & 055315.29 + 323051.00 & 6 & 20.728 & 12.388 + in this work , we found 823 new variables ( v1484v2306 ) and identified types of their variability . the discovery rate of new variables is increased by 60% in comparison with the existing catalog . the new variables are primarily 436 flaring stars , 30 sharp ebs , 48 low - amplitude pulsating variables , and 65 aperiodic variables , as well as many other rotating variables . we briefly explored how variability fractions depends on their brightness / period and variable types . as shown in figure [ fig6 ] , short - period , low - amplitude pulsating stars and episodic flare events are dominantly found at the bright- and faint - end of magnitude , respectively . in figure [ fig7 ] , we plot variability fraction as a function of period . the period range of new variables is very similar to that reported by @xcite . of these , most of new pulsating candidates show @xmath6 scuti - like characteristics ( @xmath82 0.3 days ) . also the majority of periodic variability can be explained by rotational modulation of surface features ( i.e. , starspots ) . magnitude . previously known variables ( v1v1483 ) , newly detected variables ( v1484v2306 ) , and total number of observed stars are marked with dashed lines , gray solid lines , and black solid lines , respectively.,scaledwidth=45.0% ] .,scaledwidth=50.0% ] time - scales and amplitudes of variability are intrinsic properties of the star and are independent of the distance @xcite . these two quantities can easily be measured from the light curves . we extracted the basic properties ( e.g. , period , amplitude ) of all periodic variables using the fourier decomposition technique . for comparison purposes , we utilized the flare parameters : ( i ) flare duration @xmath83 that is defined as the time - scales of flare events at 80% of its peak level , and ( ii ) flare amplitude @xmath84m@xmath85 that is magnitude at the moment of flare maximum ( s. -w . chang et al . , in preparation , hereafter paper iii ) . we separate the variable objects into photometrically selected candidate cluster and field members provided by @xcite , respectively . since the @xmath86 color covers a relatively wide range from @xmath87 to 4.5 , we used @xmath86 as a separator between a sequence of cluster and that of field population . as can be seen in figure [ fig8 ] , short period ( @xmath88 1 day ) , small amplitude pulsating stars are found at the blue - end of the distribution of both cluster and field sources , while episodic flare events are dominantly found from the sources in the red color range . our new data strengthen the relationship between the color of stars and their variability parameters derived for cluster members previously ( figure [ fig8 ] ) . the cluster members show a tight and nearly linear relation between color index and rotation period in the range of 1 @xmath88 @xmath89 @xmath88 20 days , although the change in slope occurs after the knee point at @xmath90 @xmath91 7 days . this sequence forms a curved , diagonal band from bright blue stars to faint red ones , whose periods are increasing with increasing color index ( or magnitude ) , while for non - cluster members there is no clear correlation . the color - amplitude relation for the cluster members shows that , statistically , cooler stars have greater amplitudes in the case of both periodic variables and flares . as in the previous study on stellar rotation ( see section 6 and figures 14 , 15 and 17 in @xcite ) , there is a strong correlation between rossby number and amplitude , which can be interpreted as stars rotating faster having more spots , larger spots , and/or greater spot - photosphere contrast than slower rotating stars . cooler stars have on average lower rossby numbers at a particular age , so they end up having larger amplitudes . figure [ fig9 ] illustrates the color - magnitude diagram ( cmd ) for all objects and all variable stars in the m37 field . in the new catalog , we found that about half of variable objects are distributed along the cluster main - sequence , and the other half are located in the field or in the outskirts of the cluster sequence . in order to identify the latter stars , we compared our data with the trilegal model to reproduce the cmd at this location of the galaxy . most of the input parameters are described in the @xcite . for this analysis , we set the pointing parameters for the field center ( @xmath92 ; @xmath93 ) and total field area ( @xmath94 = 0.16 deg@xmath95 ) , a limiting magnitude as @xmath96 = 23.5 and a salpeter initial mass function with cutoff at 0.01 m@xmath97 . the remaining parameter values are fixed to default values that correspond to their model calibration . the model output consists mostly of thin disk stars ( 96% ) with small contribution of old stars in the thick disk and halo ( 4% ) . non - member , variable stars may be related in some way to the thin disk population , but their distributions are not fully explained by the model . vs. @xmath71 diagram of both observed ( left ) and simulated ( right ) data . photometrically selected cluster region is plotted as red line in both panels . ] in the following subsections , we investigate the nature of each variability class : ( i ) 61 eb systems , ( ii ) 92 multiperiodic variable stars , ( iii ) 132 aperiodic variables , and ( iv ) 436 flare stars , respectively . in order to understand variability properties , we utilized both color - magnitude and color - color diagrams to separate variable objects into broad magnitude ranges or spectral classes , to the extent possible . our photometric time - series data provide the light curves of 61 ebs . figure [ fig10 ] shows phased light curves of 8 newly discovered ebs . we categorize ebs into three groups according to the shape of phased light curves , and discuss each group below . ebs in this category are either algol - type systems or @xmath98 lyrae - type systems . to derive their geometric parameters , we used the detached eclipsing binary light curve ( debil ) fitter by @xcite , which employs a simple model of a perfectly detached binary system in a classical 2-body orbit . table [ tab2 ] lists parameters derived for all detached / semi - detached binaries in the m37 field . ebs with well determined periods are included in debil analysis . special notes are given below for interesting features . [ [ v2 ] ] v2 + + this bright eb system was missed by previous study due to magnitude cut , but it clearly shows evidences of eclipse in our new light curve . [ [ v587 ] ] v587 + + + + the light curve of this eb system contains two distinct periodic signals , the 3.6173 days modulation with a small amplitude ( @xmath99 @xmath91 0.056 mag ) and an additional period at 3.2857 days ( orbital period ) . since an ellipsoidal variation should have the same period as the orbit , the first signal is more likely a consequence of starspots and rotational modulation .. after 3.6173-d modulation is removed , the light curve clearly shows the signature of a typical algol - type binary with both primary and secondary eclipses . [ [ v1482 ] ] v1482 + + + + + this eb system is a newly recognized algol - type binary with a @xmath6 sct - type pulsating component . we have clearly detected an orbital period of 2.5972 days including multiple primary and secondary eclipses . the short period variability caused by stellar pulsation is shown in the out - of - eclipse light curves . in order to determine the oscillation modes of the pulsation , a frequency analysis was performed in the out - of - eclipse data using the program period04 @xcite . our analysis not only shows peaks as a consequence of binary effects ( @xmath100 = 2.5638 and @xmath101 = 1.3006 ) but also reveals the presence of the dominant pulsation frequency @xmath102 = 0.0465 with a small amplitude ( @xmath99 = 0.0035 mag ) . rcrrrrrrrrrrl 1 & & 13.677 & & & & & & & & & & not enough data + 2 & & 14.840 & & & & & & & & & & not enough data + 3 & & 22.549 & 0.7780 & @xmath1030.08 & 0.17 & 0.15 & 22.81 & 24.12 & 1.00 & 0.65 & 283.4 & d + 4 & & 20.463 & 0.6477 & & & & & & & & & sd ; no secondary minimum ? + 5 & & 22.121 & 0.9123 & @xmath1030.01 & 0.10 & 0.04 & 22.23 & 24.59 & 1.00 & 0.53 & 129.5 & d + 6 & & 17.392 & 3.4460 & 0.01 & 0.10 & 0.08 & 17.96 & 18.35 & 1.00 & 0.33 & 232.1 & d + 7 & & 19.474 & 2.1995 & @xmath1030.08 & 0.08 & 0.07 & 20.15 & 20.30 & 1.00 & 0.75 & 239.7 & d + 8 & & 17.080 & 1.6676 & & & & & & & & & d ; no secondary minimum ? + 9 & & 16.572 & 3.2857 & 0.52 & 0.06 & 0.02 & 16.58 & 21.24 & 1.00 & 0.31 & 209.4 & d + rm ( p=3.6173 days ) ( new ) + 10 & & 16.669 & 1.8324 & 0.06 & 0.28 & 0.06 & 16.73 & 19.80 & 0.98 & 0.01 & 77.2 & sd + 11 & & 21.635 & 2.0272 & & & & & & & & & d ; no secondary minimum ? + 12 & & 17.383 & 0.7310 & @xmath1030.04 & 0.43 & 0.34 & 17.61 & 18.91 & 0.99 & 0.77 & 73.3 & d + 13 & & 16.736 & 1.9732 & 0.00 & 0.16 & 0.05 & 16.75 & 21.05 & 1.00 & 0.27 & 357.0 & sd ( new ) + 14 & & 18.730 & 5.4956 & 0.03 & 0.04 & 0.04 & 19.21 & 19.84 & 1.00 & 0.55 & 267.5 & d + rm + 15 & & 16.977 & 3.6948 & @xmath1030.66 & 0.06 & 0.01 & 16.98 & 22.60 & 1.00 & 0.29 & 268.3 & d + 16 & & 21.869 & 2.2834 & 0.05 & 0.07 & 0.06 & 22.40 & 22.87 & 1.00 & 0.16 & 268.3 & d ( new ) + 17 & & 19.742 & 1.1919 & @xmath1030.41 & 0.12 & 0.02 & 19.77 & 23.56 & 1.00 & 0.86 & 271.6 & d + 18 & & 14.874 & 2.1898 & @xmath1030.05 & 0.16 & 0.06 & 15.01 & 17.19 & 0.99 & 0.84 & 280.1 & d + @xmath7 doradus component ( p=0.941/1.174 days ) + 19 & & 17.006 & 5.9680 & 0.12 & 0.07 & 0.07 & 17.47 & 18.15 & 1.00 & 0.54 & 89.3 & d + 20 & v1482 & 15.597 & 2.5972 & 0.02 & 0.14 & 0.08 & 15.71 & 18.14 & 0.99 & 0.17 & 76.1 & d + @xmath6 scuti component ( p=0.0465 days ) ( new ) + 21 & v2138 & 14.181 & 3.4051 & 0.04 & 0.09 & 0.04 & 14.28 & 16.85 & 1.00 & 0.63 & 191.7 & d ( new ) + 22 & v2297 & 15.863 & 3.5109 & @xmath1030.01 & 0.12 & 0.02 & 15.88 & 20.42 & 1.00 & 0.32 & 229.7 & d ( new ) + the light curves of contact binary stars ( also known as w uma - type variable stars ; ew ) are very much different from those for detached binaries . our new catalog lists 22 contact binaries of which four are new discoveries ( v840 , v1031 , v1250 , v1463 ) . [ [ v37-v706-v1031 ] ] v37 , v706 , v1031 + + + + + + + + + + + + + + + + these systems are one of the fastest orbiting contact binary systems ( @xmath90 @xmath91 0.22 days ) . it is known that observed distribution of orbital periods has a cut - off period limit at 0.22 days ( e.g. , @xcite ) . we have not found anything faster in our study . rccrcrrr 1 & & 055223.89 + 322826.47 & 15.604 & 1 & 53732.28145 & 1.0 & 0.011 + & & & & 2 & 53734.26454 & 2.6 & 0.142 + & & & & 3 & 53756.09692 & 3.1 & 0.134 + 2 & & 055226.24 + 324044.62 & 15.774 & 1 & 53731.14064 & 1.1 & 0.074 + & & & & 2 & 53745.26403 & 4.7 & 0.171 + 3 & & 055232.65 + 322208.16 & 16.348 & 1 & 53755.08741 & 2.2 & 0.046 + 4 & & 055236.13 + 323001.13 & 16.012 & 1 & 53734.15888 & 2.2 & 0.054 + & & & & 2 & 53756.24427 & 1.1 & 0.051 + 5 & & 055237.96 + 323415.64 & 16.569 & 1 & 53732.14053 & 1.5 & 0.019 + & & & & 2 & 53756.17928 & 1.8 & @xmath88 0.030 + 6 & & 055311.98 + 322353.58 & 17.717 & 1 & 53726.20141 & 0.4 & 0.208 + & & & & 2 & 53730.29695 & 0.7 & 0.146 + & & & & 3 & 53733.22359 & 1.6 & 0.184 + & & & & 4 & 53734.41413 & 0.4 & 0.159 + & & & & 5 & 53737.34213 & 1.2 & 0.161 + & & & & 6 & 53740.33536 & 0.9 & 0.124 + & & & & 7 & 53753.25941 & 0.7 & 0.203 + & & & & 8 & 53756.17828 & 1.2 & 0.157 + 7 & v1731 & 055157.49 + 324534.90 & 15.823 & 1 & 53728.25931 & 3.3 & 0.233 + & & & & 2 & 53732.11914 & 2.5 & 0.282 + 8 & v1748 & 055159.96 + 322433.64 & 14.846 & 1 & 53740.46821 & 0.5 & 0.012 + & & & & 2 & 53742.46911 & 0.5 & 0.015 + 9 & v1757 & 055201.06 + 322408.79 & 17.189 & 1 & 53731.14064 & 0.4 & 0.022 + 10 & v1889 & 055216.11 + 322822.95 & 14.971 & 1 & 53742.42521 & 1.6 & 0.100 + & & & & 2 & 53756.13038 & 3.8 & 0.082 + 11 & v1892 & 055216.39 + 322609.00 & 19.328 & 1 & 53734.15888 & 2.0 & 0.178 + 12 & v1909 & 055218.62 + 324100.16 & 13.281 & 1 & 53731.34461 & 0.0 & 0.019 + & & & & 2 & 53753.15826 & 2.0 & 0.291 + 13 & v1915 & 055218.84 + 324434.08 & 21.189 & 1 & 53755.21194 & 1.9 & 0.273 + 14 & v2213 & 055301.96 + 323744.91 & 16.433 & 1 & 53725.17392 & 2.0 & 0.230 + & & & & 2 & 53733.15900 & 0.5 & 0.025 + 15 & & 055302.02 + 322327.62 & 15.094 & 1 & 53732.11914 & 2.6 & 0.091 + & & & & 2 & 53735.16000 & 1.9 & 0.087 + & & & & 3 & 53754.14737 & 0.8 & 0.051 + with our cpa analysis , we discovered 15 new eclipsing binary candidates that have eclipsing - like or transit - like features undergoing eclipse , eclipse - ingress , or eclipse - egress phases . table [ tab3 ] summarize observed eclipsing features of the light curves , which are missed by the standard analysis software . their orbital periods are not well constrained . the light curve of each event can be described by the first time of dimming points @xmath104 , the time duration between the start time of eclipse and the end time of eclipse @xmath105 , and the depth of observed dimming points @xmath106 . figure [ fig11 ] illustrates examples of light curves of this category . [ [ v791 ] ] v791 + + + + this detached eb system has deep primary and secondary eclipses ( 1213% and 1% , respectively ) , and has out - of - eclipse variability at the 2% level with period @xmath124.0631 days . after pre - whitening with the rotational modulation , our analysis indicates possible orbital period @xmath107 = 2.4277 days . [ [ v926 ] ] v926 + + + + this object was known to be a long - period ( 16.5777 days ) variable star . our ls analysis supports the conclusion of previous study , but it has sinusoidal - shaped periodic signal in our light curve . the ls periodogram shows two power peaks with periods similar to previous one ( rank 1 = 15.7291 days ) and another shorter than one day ( rank 2 = 0.9376 days ) . one sudden eclipse event was caught during its return to quiescence brightness . [ [ v1018 ] ] v1018 + + + + + we caught two eclipses , but the star seems to vary all the time in a strange manner . by using the plavchan algorithm @xcite provided in nasa exoplanet archive periodogram service , we identified two types of systems which can be confused with a transiting planetary system . one type is connected with the grazing ebs , and it shows v- or u - shape eclipses . the other type is the eclipsing binary systems with a large primary star , in which case the eclipse produces a flat bottomed light curve with small deeming . figure [ fig12 ] shows the phased light curves of actual cases of these contaminants . the observed shapes of the eclipsing events resemble transit - like signals , but depths of mid - eclipse are more than a few % of the total light ( v985 : 5% ; v1483 : 14% ; v1731 : 24% ; v1889 : 10% ) . multi - periodicity is a common feature found in nearly all type of pulsating stars such as @xmath6 scuti stars , rr lyraes , and cepheids . our catalog contains many examples of multiperiodic variable stars which deserve in - depth analysis . we performed multiple - frequency analysis of 92 short - periodic variables with the program period04 @xcite , looking for the presence of multi - periodicity . with an iterative pre - whitening procedure , discrete fourier transformations are calculated until there is no meaningful change in the fit residuals . each light curve was fitted using an equation of the form : @xmath108 where @xmath25 is a mean level and @xmath109 , @xmath110 , and @xmath111 are the frequency , amplitude , and phase for each successive peak found in the amplitude spectrum . we adopt a conservative approach in selecting the statistical significant peaks from the amplitude spectrum . in general , a signal / noise ( s / n ) amplitude ratio of 4.0 is a good criterion for independent frequencies , equivalent to 99.9% certainty of variability @xcite . for combination modes with known values , the s / n amplitude ratio of @xmath123.5 is used which is equivalent to 90% certainty . after removal of all significant peaks in the amplitude spectrum , the noise level was estimated on the residual amplitude spectrum over 5 c / d boxes around each frequency . the resulting light curve of one example multiperiodic variable is shown in figure [ fig13 ] . the multi - period model is in excellent agreement with the observed light curve . is shown as a solid curve . the full amplitude of this variation is about 0.04 mag in @xmath96-band light curve . ] table [ taba2 ] summarizes all candidate frequencies identified in 92 multiperiodic light curves with their fitted fourier parameters . all the detected frequencies are classified as independent modes ( @xmath100 , @xmath101 , @xmath102 , @xmath35 ) , combination modes ( @xmath112 ) , and harmonic mode ( @xmath113 ) . by using the method described in section 3.1 , we detected long - term aperiodic variability of 132 variable candidates with time - scales similar to the total observing time span ( @xmath1231 days ) . we visually inspected each light curve to confirm its variable nature , and this result is illustrated in terms of time sequence ( figure [ fig14 ] ) . due to the lack of periodicity , the light curves of aperiodic variables are simply characterized by the level of photometric variability ( @xmath99 ) , the dispersion about the mean level ( @xmath114 ) , and the reduced chi - square assuming a constant median value for the light curve ( @xmath115 ) . @xmath116 where @xmath117 is the photometric uncertainty of each data point @xmath43 , and @xmath118 is the total number of measurements . the full parameters are described in table [ taba3 ] . these variables show relatively low levels of variability with @xmath99 @xmath88 0.03 mag ( 60% ) and @xmath99 @xmath88 0.05 mag ( 80% ) , respectively . 31 days ) . ] values for sdss photometry system , we used the extinction coefficients compiled in @xcite . ] we used the combined sdss and 2mass photometry to identify basic properties of these sources . since color ranges are much smaller in the near - infrared color - color diagram , the stellar locus is not as well defined as in optical color - color diagram ( figure [ fig15 ] ) . we infer spectral types from @xcite , in which stars in this sample cover spectral types from approximately a5 to m5 and can be either main - sequence and giant stars . the cmd indicates that many of these aperiodic stars are field stars not associated with m37 . the origin of aperiodic nature is difficult to explain . possible speculations include that ( i ) some of these may be cases where long - period field giants are detected as low - amplitude photometric variables on timescales of weeks or longer ( e.g. , @xcite ) , ( ii ) some of them may be semi - regular variables exhibiting both fast and slow variabilities ( e.g. , @xcite ) , and ( iii ) as mentioned in @xcite , the rapid amplitude variation may be flickering , nova - like outbursts seen in many symbiotic variables ( e.g. , @xcite ) . further observations of much longer duration will be necessary to properly characterize these aperiodic variables . earlier analysis of m37 data ( hartman et al . 2008a ) could not address flare events properly because many data points were removed as outliers . our new pipeline , with nearly 100% data utilization , offers opportunity to investigate flare stars more thoroughly . by cpa analysis , we found that 436 stars ( 625 flare events ) show flaring activity characterized by a rapid flux increase and a slower decay . as expected , the location of these flaring stars on color - magnitude and color - color diagrams suggest that most of them are late - type main - sequence stars ( k4m6 ) in m37 open cluster ( figure [ fig16 ] ) . flares are also observed from earlier spectral types ( a0k3 ) , but these are only a few and probably belong to field stars . the underlying mechanism of flares in stars cooler than about f5 is essentially the same as in the sun . it is believed that these cool stars can generate a magnetic field through a dynamos at the interface between the radiative core and the convective envelope @xcite . thus , flaring is an inevitable result of strong magnetic field . but flares in hot stars , especially in a - type stars , are not easy to explain in the context of dynamo process because these stars should not generate strong magnetic fields in their thin convection layer and are incapable of storing larger energy in that region @xcite . the observed flare signature of one a - type star ( v72 ) might be the result of binarity . if an a - type main sequence star has an active m dwarf secondary star , such binary star would look like a flare - generating a - type star . the number of flare events is increased by a factor of two in the new analysis because our new calibration procedure provides more accurate @xmath96-band light curves ( figure [ fig17 ] ) . in table [ taba4 ] and figure [ fig18 ] , we list the parameters of flare events discovered by our cpa analysis . they have a broad range of peak amplitude ( @xmath119 = @xmath120 mag ) and typical durations of a few minutes to several hours , irrespective of their spectral type . more detailed analysis of flare statistics and its relation to magnetic activity and stellar rotation will be presented in subsequent papers . -band data in terms of both photometric accuracy and data recovery rate . ] we cataloged and investigated a variety of photometric variables in the field of m37 using our new light curves , which were obtained from archival imaging data of one - month long exo - planet transit survey with the 6.5-m mmt telescope . this new dataset allows a rare opportunity to explore different types of variability of short ( @xmath12minutes ) and long ( @xmath12one - month ) time - scales simultaneously . we find 61 eclipsing binary systems ( eb ) , 92 multiperiodic variable stars ( mp ) , 132 aperiodic variables ( a ) , and 436 flare stars ( f ) , as well as several hundreds of rotating variables ( r ) . our work has increased the number of known variable stars in this field by 60% . the new variable catalog has been made available on the web ( http://stardb.yonsei.ac.kr/ ) . we would like to thank the anonymous referee for helpful suggestion . this research was supported by basic science research program of the national research foundation of korea ( 2011 - 0030875 ) . i.b . acknowledges the support from kasi - yonsei drc program of korea research council of fundamental science and technology ( drc-12 - 2-kasi ) . benk , j. m. , kolenberg , k. , szab , r. , et al . 2010 , , 409 , 1585 benzm , a. o. & gdel , m. 2010 , , 48 , 241 breger , m. , stich , j. , garrido , r. , et al . 1993 , , 271 , 482 butler , n. r. & bloom , j. s. 2011 , , 141 , 93 chang , s .- w . , byun , y .- i . , & hartman , j. d. 2015 , , 149 , 135 ( paper i ) ciardi , d. r. , braun , k. v. , bryden , g. , et al . 2011 , , 141 , 108 covey , k. r. , ivezi , . , schlegel , d. , et al . 2007 , , 134 , 2398 cutri , r. m. , skrutskie , m. f. , van dyk , s. , et al . 2003 , the irsa 2mass all - sky point source catalog , nasa / ipac infrared science archive , http://irsa.ipac.caltech.edu/applications/gator/ debosscher , j. , sarro , l. m. , aert , c. , et al . 2007 , , 475 , 1159 debosscher , j. , sarro , l. m. , lpez , m. , et al . 2009 , , 506 , 519 devor , j. 2005 , , 628 , 411 eyer , l. & mowlavi , n. 2008 , j. phys . ser . , 118 , 012010 girardi , l. , grebel , e. k. , odenkirchen , m. , & chiosi , c. 2004 , , 422 , 205 girardi , l. , groenewegen , m. a. t. , hatziminaoglou , e. , & da costa , l. 2005 , , 436 , 895 glicenstein , j. -f . 2001 , in asp conf . 239 , microlensing 2000 : a new era of microlensing astrophysics , ed . j. w. menzies & penny d. sackett . ( san francisco , ca : asp ) , 28 gromadzki , m. , mikoajewska , j. , whitelock , p. , & marang , f. 2009 , acta astron . , 59 , 169 gromadzki , m. , mikolajewska , j. , & soszynski , i. 2013 , arxiv:1312.6063 grubrbauer , m. , kolenberg , k. , rowe , j. f. , et al . 2007 , , 379 , 1498 guggenberger , e. , kolenberg , k. , nemec , j. m. , et al . 2012 , , 424 , 649 hartman , j. d. , gaudi , b. s. , holman , m. j. , et al . 2008a , , 675 , 1233 hartman , j. d. , gaudi , b. s. , holman , m. j. , et al . 2008b , , 675 , 1254 hartman , j. d. , gaudi , b. s. , pinsonneault , m. h. , et al . 2009a , , 691 , 342 hartman , j. d. , gaudi , b. s. , holman , m. j. , et al . 2009b , , 695 , 336 henry , gregory w. , fekel , francis c. , henry , stephen m. , & hall , douglas s. 2000 , , 130 , 201 irwin , j. , irwin , m. , aigrain , s. , et al . , , 375 , 1449 kalirai , j. s. , ventura , p. , richer , h. b. , et al . 2001 , , 122 , 3239 kang , y. b. , kim , s .- l . , rey , s .- c . , et al . 2007 , , 119 , 239 kelly , b. c. , bechtold , j. , & siemiginowska , a. 2009 , , 698 , 895 kiss , l. l. , szab , gy . m. , szildi , k. , frsz , g. , srneczky , k. , & csk , b. 2001 , , 376 , 561 kovcs , g. , zucker , s. , & mazeh , t. 2002 , , 391 , 369 kuschnig , r. , weiss , w. w. , gruber , r. , bely , p. y. , jenkner , h. 1997 , , 328 , 544 lenz p. , breger m. 2005 , coast , 146 , 53 lomb , n. r. 1976 , , 39 , 447 messina , s. , distefano , e. , parihar , p. , et al . 2008 , , 483 , 253 molnar , lawrence a. , van noord , daniel m. , & steenwyk , steven d. 2013 , arxiv:1310.0539 richards , j. w. , starr , d. l. , butler , n. r. , et al . 2011 , , 733 , 10 rucinski , s. m. 1992 , , 103 , 960 rucinski , s. m. 2002 , , 114 , 1124 palmer , d. m. 2009 , , 695 , 496 plavchan , p. , jura , m. , kirkpatrick , d. , cutri , roc m. , & gallagher , s. c. 2008 , , 175 , 191 press , w.h . & rybicki , g.b . 1989 , , 338 , 277 press , w. h. , teukolsky , s. a. , vetterling , w. t. , & flannery , b. p. 1992 , numerical recipes in c. 2nd ed . ( new york : cambridge university press ) reegen , p. 2007 , , 467 , 1353 samus , n. n. , durlevich , o. v. , kazarovets , e. v. , et al . 2009 , vizier online data catalog : b / gcvs , 1 , 2025 scargle , j. d. 1982 , , 263 , 835 schtz , n. & holschneider , m. 2011 , , 84 , 021120 schwarzenberg - czerny , a. 1989 , , 241 , 153 shin , m .- s . & byun , y .- 2004 , jkas , 37 , 79 taylor , w. 2000 , in change - point analyzer 2.0 shareware program , taylor enterprises , libertyville , illinois . , http://www.variation.com/cpa ccrrrrrrrrrrrrc & 055220.39 + 323319.65 & & & & 13.677 & 0.285 & 0.191 & & 13.406 & 0.609 & 12.447 & 0.270 & 0.063 & eb + & 055216.58 + 322815.32 & & & 16.792 & 14.840 & 0.564 & 0.290 & 14.471 & 14.653 & 0.688 & 13.539 & 0.287 & 0.134 & eb + & 055233.01 + 323241.91 & 0.4225 & 0.320 & 17.982 & 15.789 & 0.560 & 0.329 & 15.284 & 16.051 & 0.728 & 14.267 & 0.355 & 0.012 & eb + & 055253.26 + 323301.47 & 0.5582 & 0.235 & 18.079 & 15.634 & 0.762 & 0.477 & 15.093 & 15.948 & 0.938 & 13.951 & 0.497 & 0.101 & eb + & 055300.63 + 322450.81 & 0.2788 & 0.310 & 18.036 & 15.915 & 0.459 & 0.481 & 15.235 & 16.184 & 0.585 & 14.421 & 0.359 & 0.084 & mp + & 055150.53 + 323234.70 & 0.1098 & 0.343 & 18.218 & 15.973 & 0.591 & 0.173 & 15.622 & & & 14.739 & 0.342 & -0.009 & mp + & 055239.09 + 323631.27 & 0.3577 & 0.433 & 19.858 & 17.252 & 1.039 & 0.056 & 17.065 & 17.890 & 0.815 & 15.779 & 0.446 & 0.196 & eb + & 055234.32 + 323218.70 & 0.1195 & & & & & & & & & & & & u + & 055214.91 + 322441.35 & 0.0909 & & 14.873 & 13.233 & 0.137 & 0.128 & 13.114 & 13.298 & 0.370 & 12.348 & 0.171 & 0.042 & p + & 055200.49 + 323648.19 & 0.9432 & 0.075 & 16.844 & 14.151 & 0.928 & 0.436 & 13.493 & 14.762 & 0.531 & 12.356 & 0.567 & 0.131 & p + rccrrcrrrl 1 & & 055616.79 + 322514.20 & 15.915 & 3.5872 & f@xmath121 & 310.1 & 0.3331 & 68.9 & + & & & & 7.1768 & f@xmath122(=2f@xmath121 ) & 35.4 & 0.2906 & 11.3 & + & & & & 10.7619 & f@xmath123(=3f@xmath121 ) & 22.8 & 0.0418 & 10.3 & + & & & & 5.6768 & f@xmath124 & 19.5 & 0.2369 & 5.7 & + & & & & 14.3473 & f@xmath125 & 13.8 & 0.9990 & 7.1 & + 2 & & 055506.89 + 323303.20 & 15.973 & 9.1040 & f@xmath121(=f@xmath123-f@xmath122 ) & 343.4 & 0.5433 & 172.7 & + & & & & 18.2081 & f@xmath122(=f@xmath123-f@xmath121 ) & 89.9 & 0.2201 & 53.6 & + & & & & 27.3122 & f@xmath123(=f@xmath121+f@xmath122 ) & 27.6 & 0.9237 & 19.7 & + & & & & 8.7545 & f@xmath124 & 15.4 & 0.5801 & 8.1 & + & & & & 11.9319 & f@xmath125 & 14.5 & 0.5557 & 6.3 & + & & & & 36.4178 & f@xmath126(=2f@xmath122 ) & 10.1 & 0.0527 & 10.4 & + & & & & 14.9868 & f@xmath127 & 8.8 & 0.9887 & 4.5 & + & & & & 16.8695 & f@xmath128 & 7.9 & 0.7280 & 4.5 & + 3 & & 055519.81 + 323541.23 & 15.245 & 8.4261 & f@xmath121 & 79.0 & 0.9052 & 84.8 & + & & & & 16.8561 & f@xmath122(=2f@xmath121 ) & 7.6 & 0.6488 & 8.3 & + & & & & 17.7044 & f@xmath123 & 8.0 & 0.5895 & 9.1 & + & & & & 16.4751 & f@xmath124 & 6.9 & 0.5617 & 7.3 & + & & & & 8.8787 & f@xmath125 & 4.9 & 0.4440 & 5.2 & + & & & & 12.0801 & f@xmath126 & 4.8 & 0.8357 & 5.0 & + & & & & 9.2640 & f@xmath127 & 5.9 & 0.6453 & 6.4 & + & & & & 10.6455 & f@xmath128 & 5.4 & 0.0597 & 6.0 & + & & & & 13.5604 & f@xmath129 & 4.8 & 0.8005 & 4.9 & + & & & & 16.2578 & f@xmath130 & 5.1 & 0.6379 & 5.3 & + & & & & 8.0470 & f@xmath131 & 4.5 & 0.8268 & 4.7 & + & & & & 2.3825 & f@xmath132 & 5.2 & 0.7274 & 3.6 & + & & & & 4.3812 & f@xmath133 & 4.3 & 0.3942 & 4.2 & + & & & & 14.9483 & f@xmath134 & 3.9 & 0.8404 & 4.0 & + 4 & & 055600.99 + 323041.25 & 15.873 & 12.7176 & f@xmath121 & 14.4 & 0.0994 & 8.1 & + & & & & 7.8896 & f@xmath122 & 13.8 & 0.4901 & 6.0 & + & & & & 12.7698 & f@xmath123 & 10.5 & 0.0175 & 6.0 & close to f@xmath121 + & & & & 14.4564 & f@xmath124 & 9.6 & 0.0884 & 6.5 & + 5 & & 055527.32 + 322543.01 & 16.511 & 10.0341 & f@xmath121(=f@xmath132-f@xmath122 ) & 23.0 & 0.4259 & 137.9 & + & & & & 10.4706 & f@xmath122(=f@xmath132-f@xmath121 ) & 13.0 & 0.8171 & 77.9 & + & & & & 17.1389 & f@xmath123 & 8.7 & 0.9050 & 51.9 & + & & & & 14.0376 & f@xmath124 & 6.1 & 0.6801 & 33.6 & + & & & & 13.0968 & f@xmath125 & 3.7 & 0.1837 & 20.9 & + & & & & 12.8019 & f@xmath126 & 3.1 & 0.2056 & 17.4 & + & & & & 15.0841 & f@xmath127 & 2.1 & 0.7887 & 11.6 & + & & & & 13.6263 & f@xmath128 & 1.3 & 0.1260 & 7.1 & + & & & & 11.1091 & f@xmath129 & 1.4 & 0.2454 & 8.7 & + & & & & 27.1718 & f@xmath130 & 1.0 & 0.7296 & 6.7 & + rccrrrrrrc 1 & & 055122.14 + 322942.40 & 16.742 & 0.009 & 0.096 & 443.5 & 4730 & + 2 & & 055123.93 + 324357.22 & 15.837 & 0.003 & 0.017 & 14.0 & 4706 & + 3 & & 055124.73 + 324312.36 & 18.360 & 0.004 & 0.028 & 1.8 & 4730 & 13.260 + 4 & & 055125.42 + 323600.45 & 17.564 & 0.002 & 0.017 & 1.6 & 4730 & 21.199 + 5 & & 055127.04 + 323225.57 & 15.458 & 0.002 & 0.010 & 7.7 & 4592 & + 6 & & 055128.19 + 322931.14 & 17.854 & 0.004 & 0.029 & 3.4 & 4730 & + 7 & & 055129.26 + 323235.44 & 18.515 & 0.005 & 0.029 & 1.4 & 4730 & + 8 & & 055130.48 + 324053.05 & 18.348 & 0.005 & 0.033 & 1.9 & 4730 & 7.709 + 9 & & 055131.24 + 323059.61 & 17.688 & 0.008 & 0.046 & 12.9 & 4730 & + 10 & & 055131.87 + 324233.83 & 17.482 & 0.004 & 0.026 & 5.1 & 4730 & + rccrcrrrc 1 & & 055120.47 + 322200.75 & 20.690 & 1 & 53742.40307 & 0.093 & 0.17 & + & & & & 2 & 53742.46247 & 0.135 & 0.08 & + 2 & & 055121.46 + 322856.90 & 19.990 & 1 & 53730.34740 & 0.221 & 0.55 & + & & & & 2 & 53733.19093 & 0.176 & 0.21 & + & & & & 3 & 53735.19767 & 0.039 & 0.49 & + & & & & 4 & 53737.33376 & 0.188 & 0.30 & + 3 & & 055123.45 + 322948.83 & 20.809 & 1 & 53732.22482 & 0.061 & 0.32 & + & & & & 2 & 53732.35513 & 0.133 & 1.11 & + & & & & 3 & 53742.35427 & 0.281 & 0.23 & + 4 & & 055124.44 + 324407.71 & 15.143 & 1 & 53730.32306 & 0.010 & 1.47 & + 5 & & 055124.78 + 323440.56 & 18.368 & 1 & 53728.24480 & 0.026 & 0.17 & + & & & & 2 & 53742.44376 & 0.117 & 0.24 & + & & & & 3 & 53753.25302 & 0.139 & 0.16 & + 6 & & 055124.82 + 322937.32 & 20.017 & 1 & 53726.39533 & 0.046 & 0.10 & var + 7 & & 055125.41 + 323800.03 & 18.086 & 1 & 53726.27563 & 0.041 & 0.58 & var + & & & & 2 & 53727.30272 & 0.025 & 0.46 & + 8 & & 055125.71 + 323106.38 & 20.553 & 1 & 53734.23990 & 0.072 & 0.20 & var + & & & & 2 & 53756.08268 & 0.079 & 0.60 & + 9 & & 055126.78 + 324024.73 & 17.971 & 1 & 53726.43908 & 0.030 & 0.29 & + 10 & & 055127.86 + 322854.79 & 22.181 & 1 & 53727.19210 & 0.360 & 0.58 & var +

various problems of geometry , topology and dynamical systems on surfaces as well as some questions concerning one - dimensional dynamical systems lead to the study of closed surfaces endowed with a flat metric with several cone - type singularities . in an important particular case , when the flat metric has trivial holonomy , the corresponding flat surfaces are naturally organized into families which appear to be isomorphic to moduli spaces of holomorphic one - forms . one can obtain much information about the geometry and dynamics of an individual flat surface by studying both its orbit under the teichmller geodesic flow and under the linear group action on the corresponding moduli space . we apply this general principle to the study of generic geodesics and to counting of closed geodesics on a flat surface . ( 0,0)(0,0 ) ( 170,-490)proceedings of the international congress ( 170,-500)of mathematics , madrid , spain , 2006 ( 170,-510)2006 , ems publishing house primary 57m50 , 32g15 ; secondary 37d40 , 37d50 , 30f30 . flat surface , teichmller geodesic flow , moduli space , asymptotic cycle , lyapunov exponent , interval exchange transformation , renormalization .

math0609399

consider a collection of vectors @xmath0 in @xmath1 and construct from these vectors a broken line in a natural way : a @xmath2-th edge of the broken line is represented by the vector @xmath3 . construct another broken line starting at the same point as the initial one by taking the same vectors in the order @xmath4 , where @xmath5 is some permutation of @xmath6 elements . by construction the two broken lines share the same endpoints ; suppose that they bound a polygon like in fig . [ zorich : fig : suspension ] . identifying the pairs of sides corresponding to the same vectors @xmath3 , @xmath7 , by parallel translations we obtain a surface endowed with a flat metric . ( this construction follows the one in @xcite . ) the flat metric is nonsingular outside of a finite number of cone - type singularities corresponding to the vertices of the polygon . by construction the flat metric has trivial holonomy : a parallel transport of a vector along a closed path does not change the direction ( and length ) of the vector . this implies , in particular , that all cone angles are integer multiples of @xmath8 . ( 20,2)(20,2 ) ( 142,-48)@xmath9 ( 168,-24)@xmath10 ( 222,-7)@xmath11 ( 270,-42)@xmath12 ( 115,-93)@xmath12 ( 163,-135)@xmath11 ( 216,-114)@xmath10 ( 245,-90)@xmath9 the polygon in our construction depends continuously on the vectors @xmath3 . this means that the combinatorial geometry of the resulting flat surface ( its genus @xmath13 , the number @xmath14 and types of the resulting conical singularities ) does not change under small deformations of the vectors @xmath3 . this allows to consider a flat surface as an element of a family of flat surfaces sharing common combinatorial geometry ; here we do not distinguish isometric flat surfaces . as an example of such family one can consider a family of flat tori of area one , which can be identified with the space of lattices of area one : @xmath15 { \operatorname{so}(2,\mathbb r)}\ \phantom{\backslash\ , { \operatorname{sl}(2,\mathbb r)}\ ! / } { \operatorname{sl}(2,\mathbb z)}\end{array } \quad = \quad \begin{array}{l } { { \mathbb h}^2}/\\ [ -\halfbls]\phantom{{{\mathbb h}^2}/ } { \operatorname{sl}(2,\mathbb z)}\end{array}\ ] ] the corresponding `` modular surface '' is not compact , see fig . [ zorich : fig : space : of : flat : tori ] . flat tori representing points , which are close to the cusp , are almost degenerate : they have a very short closed geodesic . similarly , families of flat surfaces of higher genera also form noncompact finite - dimensional orbifolds . the origin of their noncompactness is the same as for the tori : flat surfaces having short closed geodesics represent points which are close to the multidimensional `` cusps '' . ( -11,0)(-11,0 ) ( -70,-40)neighborhood of a ( -70,-50)cusp = subset of ( -70,-60)tori having short ( -70,-70)closed geodesic we shall consider only those flat surfaces , which have trivial holonomy . choosing a direction at some point of such flat surface we can transport it to any other point . it would be convenient to include the choice of direction in the definition of a flat structure . in particular , we want to distinguish the flat structure represented by the polygon in fig . [ zorich : fig : suspension ] and the one represented by the same polygon rotated by some angle different from @xmath8 . consider the natural coordinate @xmath16 in the complex plane . in this coordinate the parallel translations which we use to identify the sides of the polygon in fig . [ zorich : fig : suspension ] are represented as @xmath17 . since this correspondence is holomorphic , it means that our flat surface @xmath18 with punctured conical points inherits the complex structure . it is easy to check that the complex structure extends to the punctured points . consider now a holomorphic 1-form @xmath19 in the complex plane . when we pass to the surface @xmath18 the coordinate @xmath16 is not globally defined anymore . however , since the changes of local coordinates are defined as @xmath17 , we see that @xmath20 . thus , the holomorphic 1-form @xmath19 on @xmath21 defines a holomorphic 1-form @xmath22 on @xmath18 which in local coordinates has the form @xmath23 . it is easy to check that the form @xmath22 has zeroes exactly at those points of @xmath18 where the flat structure has conical singularities . reciprocally , one can show that a pair ( riemann surface , holomorphic 1-form ) uniquely defines a flat structure of the type described above . in an appropriate local coordinate @xmath24 a holomorphic 1-form can be represented in a neighborhood of zero as @xmath25 , where @xmath26 is called the degree of zero . the form @xmath22 has a zero of degree @xmath26 at a conical point with cone angle @xmath27 . the sum of degrees @xmath28 of zeroes of a holomorphic 1-form on a riemann surface of genus @xmath13 equals @xmath29 . the moduli space @xmath30 of pairs ( complex structure , holomorphic 1-form ) is a @xmath31-vector bundle over the moduli space @xmath32 of complex structures . the space @xmath30 is naturally stratified by the strata @xmath33 enumerated by unordered partitions of the number @xmath29 in a collection of positive integers @xmath34 . any holomorphic 1-forms corresponding to a fixed stratum @xmath33 has exactly @xmath14 zeroes , and @xmath35 are the degrees of zeroes . note , that an individual stratum @xmath33 in general does not form a fiber bundle over @xmath32 . it is possible to show that if the permutation @xmath5 which was used to construct a polygon in fig . [ zorich : fig : suspension ] satisfy some explicit conditions , vectors @xmath36 representing the sides of the polygon serve as coordinates in the corresponding family @xmath33 . consider vectors @xmath37 as complex numbers . let @xmath37 join vertices @xmath38 and @xmath39 of the polygon . denote by @xmath40 the resulting path on @xmath18 joining the points @xmath41 . our interpretation of @xmath37 as of a complex number implies that @xmath42 the path @xmath40 represents a relative cycle : an element of the relative homology group @xmath43 of the surface @xmath18 relative to the finite collection of conical points @xmath44 . relation above means that @xmath45 represents a period of @xmath22 : an integral of @xmath22 over the relative cycle @xmath40 . in other words , a small domain in @xmath46 containing @xmath47 $ ] can be considered as a local coordinate chart in our family @xmath33 of flat surfaces . we summarize the correspondence between geometric language of flat surfaces and the complex - analytic language of holomorphic 1-forms on a riemann surface in the dictionary below . [ cols="^,^ " , ] note that the vector space @xmath48 contains a natural integer lattice @xmath49 . consider a linear volume element @xmath50 in the vector space @xmath51 normalized in such a way that the volume of the fundamental domain in the `` cubic '' lattice @xmath52 equals one . consider now the real hypersurface @xmath53 defined by the equation @xmath54 . the volume element @xmath50 can be naturally restricted to the hypersurface defining the volume element @xmath55 on @xmath56 . the total volume @xmath57 of every stratum is finite . the values of these volumes were computed by a. eskin and a. okounkov @xcite . consider a flat surface @xmath18 and consider a polygonal pattern obtained by unwrapping @xmath18 along some geodesic cuts . for example , one can assume that our flat surface @xmath18 is glued from a polygon @xmath58 as on fig . [ zorich : fig : suspension ] . consider a linear transformation @xmath59 of the plane @xmath1 . the sides of the new polygon @xmath60 are again arranged into pairs , where the sides in each pair are parallel and have equal length . identifying the sides in each pair by a parallel translation we obtain a new flat surface @xmath61 which , actually , does not depend on the way in which @xmath18 was unwrapped to a polygonal pattern @xmath62 . thus , we get a continuous action of the group @xmath63 on each stratum @xmath33 . considering the subgroup @xmath64 of area preserving linear transformations we get the action of @xmath64 on the `` unit hyperboloid '' @xmath56 . considering the diagonal subgroup @xmath65 we get a continuous action of this one - parameter subgroup on each stratum @xmath33 . this action induces a natural flow on the stratum which is called the _ teichmller geodesic flow_. the actions of the groups @xmath64 and @xmath66 preserve the measure @xmath55 . both actions are ergodic with respect to this measure on each connected component of every stratum @xmath56 . the following basic principle ( which was was first used in the pioneering works of h. masur @xcite and of w. veech @xcite to prove unique ergodicity of almost all interval exchange transformations ) appeared to be surprisingly powerful in the study of flat surfaces . suppose that we need some information about geometry or dynamics of an individual flat surface @xmath18 . consider the `` point '' @xmath18 in the corresponding family of flat surfaces @xmath67 . denote by @xmath68 the closure of the @xmath63-orbit of @xmath18 in @xmath69 . in numerous cases knowledge about the structure of @xmath70 gives a comprehensive information about geometry and dynamics of the initial flat surface @xmath18 . moreover , some delicate numerical characteristics of @xmath18 can be expressed as averages of simpler characteristics over @xmath70 . we apply this general philosophy to the study of geodesics on flat surfaces . actually , there is a hope that this philosophy extends much further . a closure of an orbit of an abstract dynamical system might have extremely complicated structure . according to the optimistic hopes , the closure @xmath70 of a @xmath63-orbit of any flat surface @xmath18 is a nice complex - analytic variety , and all such varieties might be classified . for genus two the latter statements were recently proved by c. mcmullen ( see @xcite and @xcite ) and partly by k. calta @xcite . the following theorem supports the hope for some nice and simple description of orbit closures . suppose that the closure in the stratum @xmath67 of a @xmath63-orbit of some flat surface @xmath18 is a complex - analytic subvariety . then in cohomological coordinates @xmath46 this subvariety is represented by an affine subspace . in this section we study geodesics on a flat surface @xmath18 going in generic directions . according to the theorem of s. kerckhoff , h. masur and j. smillie @xcite , for any flat surface @xmath18 the directional flow in almost any direction is uniquely ergodic . this implies , in particular , that for such directions the geodesics wind around @xmath18 in a relatively regular manner . namely , it is possible to find a cycle @xmath71 such that a long piece of geodesic pretends to wind around @xmath18 repeatedly following this asymptotic cycle @xmath72 . rigorously it can be described as follows . having a geodesic segment @xmath73 and some point @xmath74 we emit from @xmath75 a geodesic transversal to @xmath76 . from time to time the geodesic would intersect @xmath76 . denote the corresponding points as @xmath77 . closing up the corresponding pieces of the geodesic by joining the starting point @xmath78 and the point @xmath79 of @xmath2-th return to @xmath76 with a path going along @xmath76 we get a sequence of closed paths defining the cycles @xmath80 . these cycles represent longer and longer pieces of the geodesic . when the direction of the geodesic is uniquely ergodic , the limit @xmath81 exists and the corresponding asymptotic cycle @xmath82 does not depend on the starting point @xmath83 . changing the transverse interval @xmath76 we get a collinear asymptotic cycle . when @xmath18 is a flat torus glued from a unit square , the asymptotic cycle @xmath72 is a vector in @xmath84 and its slope is exactly the slope of our flat geodesic in standard coordinates . when @xmath18 is a surface of higher genus the asymptotic cycle belongs to a @xmath85-dimensional space @xmath86 . let us study how the cycles @xmath87 deviate from the direction of the asymptotic cycle @xmath72 . choose a hyperplane @xmath88 in @xmath89 orthogonal ( transversal ) to the asymptotic cycle @xmath72 and consider a parallel projection to this screen along @xmath72 . projections of the cycles @xmath90 would not be necessarily bounded : directions of the cycles @xmath90 tend to direction of the asymptotic cycle @xmath72 provided the norms of the projections grow sublinearly with respect to @xmath91 . let us observe how the projections are distributed in the screen @xmath88 . a heuristic answer is given by fig . [ zorich : fig : asymptotic : cycle : deviation ] . we see that the distribution of projections of the cycles @xmath90 in the screen @xmath88 is anisotropic : the projections accumulate along some line . this means that in the original space @xmath92 the vectors @xmath90 deviate from the asymptotic direction @xmath93 spanned by @xmath72 not arbitrarily but along some two - dimensional subspace @xmath94 containing @xmath93 , see fig . [ zorich : fig : asymptotic : cycle : deviation ] . moreover , measuring the norms @xmath95 of the projections we get @xmath96 thus , the vector @xmath90 is located approximately in the two - dimensional plane @xmath94 , and the distance from its endpoint to the line @xmath93 in @xmath94 is at most of the order @xmath97 , see fig . [ zorich : fig : asymptotic : cycle : deviation ] . ( 0,0)(0,0 ) ( 0,0 ) ( 0,0)(0,0 ) ( 197,-85)@xmath90 ( 224,-60)@xmath97 ( 269,-105)@xmath98 ( 0,0 ) ( 0,0)(0,0 ) ( 105,-190)@xmath99 ( 102,-210)@xmath100 ( 65,-220)@xmath101 ( 35,-208)@xmath102 ( 20,-181)@xmath103 ( 35,-142)@xmath104 ( 75,-120)@xmath105 ( 0,0 ) ( 0,0)(0,0 ) ( 265,-180)asymptotic plane @xmath106 ( 0,0 ) ( 0,0)(0,0 ) ( 90,-70)direction of the ( 90,-80)asymptotic cycle ( 0,0 ) ( 0,0)(0,0 ) ( 90,-20)@xmath107 consider now a new screen @xmath108 orthogonal to the plane @xmath94 . now the screen @xmath109 has codimension two in @xmath110 . taking the projections of @xmath90 to @xmath109 along @xmath94 we eliminate the asymptotic directions @xmath93 and @xmath94 and we see how the vectors @xmath90 deviate from @xmath94 . on the screen @xmath109 we observe the same picture as in fig . [ zorich : fig : asymptotic : cycle : deviation ] : the projections are again located along a one - dimensional subspace . coming back to the ambient space @xmath110 , this means that in the first term of approximation all vectors @xmath90 are aligned along the one - dimensional subspace @xmath93 spanned by the asymptotic cycle . in the second term of approximation , they can deviate from @xmath93 , but the deviation occurs mostly in the two - dimensional subspace @xmath94 , and has order @xmath97 where @xmath111 . in the third term of approximation we see that the vectors @xmath90 may deviate from the plane @xmath94 , but the deviation occurs mostly in a three - dimensional space @xmath112 and has order @xmath98 where @xmath113 . going on we get further terms of approximation . however , getting to a subspace @xmath114 which has half of the dimension of the ambient space we see that , in there is no more deviation from @xmath114 : the distance from any @xmath90 to @xmath114 is uniformly bounded . note that the intersection form endows the space @xmath110 with a natural symplectic structure . it can be checked that the resulting @xmath13-dimensional subspace @xmath114 is a lagrangian subspace for this symplectic form . a rigorous formulation of phenomena described heuristically in fig . [ zorich : fig : asymptotic : cycle : deviation ] is given by the theorem below . by convention we always consider a flat surface together with a choice of direction which is called the vertical direction , or , sometimes , `` direction to the north '' . using an appropriate homotethy we normalize the area of @xmath18 to one , so that @xmath115 . we chose a point @xmath116 and a horizontal segment @xmath76 passing through @xmath78 ; by @xmath117 we denote the length of @xmath76 . the interval @xmath76 is chosen in such way , that the interval exchange transformation induced by the vertical flow has the minimal possible number @xmath118 of subintervals under exchange . ( actually , almost any other choice of @xmath76 would also work . ) we consider a geodesic ray @xmath119 emitted from @xmath78 in the vertical direction . ( if @xmath78 is a saddle point , there are several outgoing vertical geodesic rays ; choose any of them . ) each time when @xmath119 intersects @xmath76 we join the point @xmath120 of intersection and the starting point @xmath78 along @xmath76 producing a closed path . we denote the homology class of the corresponding loop by @xmath90 . let @xmath22 be the holomorphic 1-form representing @xmath18 ; let @xmath13 be genus of @xmath18 . choose some euclidean metric in @xmath121 which would allow to measure a distance from a vector to a subspace . let by convention @xmath122 . [ zorich : th : mf ] for almost any flat surface @xmath18 in any stratum @xmath56 there exists a flag of subspaces @xmath123 in the first homology group of the surface with the following properties . choose any starting point @xmath83 in the horizontal segment @xmath76 . consider the corresponding sequence @xmath124 of cycles . the following limit exists @xmath125 where the nonzero asymptotic cycle @xmath126 is poincar dual to the cohomology class of @xmath127 $ ] , and the one - dimensional subspace @xmath128 is spanned by @xmath72 . for any @xmath129 one has @xmath130 and @xmath131 where the constant depends only on @xmath18 and on the choice of the euclidean structure in the homology space . the numbers @xmath132 are the top @xmath13 lyapunov exponents of the teichmller geodesic flow on the corresponding connected component of the stratum @xmath133 ; in particular , they do not depend on the individual generic flat surface @xmath18 in the connected component . it should be stressed , that the theorem above was formulated in @xcite as a conditional statement : under the conjecture that @xmath134 there exist a lagrangian subspace @xmath114 such that the cycles are in a bounded distance from @xmath114 ; under the further conjecture that all the exponents @xmath135 , for @xmath136 , are distinct , there is a complete lagrangian flag ( i.e. the dimensions of the subspaces @xmath137 , where @xmath138 , rise each time by one ) . these two conjectures were later proved by g. forni @xcite and by a. avila and m. viana @xcite correspondingly . currently there are no methods of calculation of individual lyapunov exponents @xmath135 ( though there is some experimental knowledge of their approximate values ) . nevertheless , for any connected component of any stratum ( and , more generally , for any @xmath139-invariant suborbifold ) it is possible to evaluate the _ sum _ of the lyapunov exponents @xmath140 , where @xmath13 is the genus . the formula for this sum was discovered by m. kontsevich ; morally , it is given in terms of characteristic numbers of some natural vector bundles over the strata @xmath33 , see @xcite . another interpretation of this formula was found by g. forni @xcite ; see also a very nice formalization of these results in the survey of r. krikorian @xcite . for some special @xmath139-invariant suborbifolds the corresponding vector bundles might have equivariant subbundles , which provides additional information on corresponding subcollections of the lyapunov exponents , or even gives their explicit values in some cases , like in the case of teichmller curves considered in the paper of i. bouw and m. mller @xcite . theorem [ zorich : th : mf ] illustrates a phenomenon of _ deviation spectrum_. it was proved by g. forni in @xcite that ergodic sums of smooth functions on an interval along trajectories of interval exchange transformations , and ergodic integrals of smooth functions on flat surfaces along trajectories of directional flows have deviation spectrum analogous to the one described in theorem [ zorich : th : mf ] . l. flaminio and g. forni showed that the same phenomenon can be observed for other parabolic dynamical systems , for example , for the horocycle flow on compact surfaces of constant negative curvature @xcite . [ [ idea - of - the - proof - renormalization . ] ] idea of the proof : renormalization . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the reason why the deviation of the cycles @xmath87 from the asymptotic direction is governed by the teichmller geodesic flow is illustrated in fig . [ zorich : fig : renormalization ] . in a sense , we follow the initial ideas of h. masur @xcite and of w. veech @xcite . fix a horizontal segment @xmath76 and emit a vertical trajectory from some point @xmath75 in @xmath76 . when the trajectory intersects @xmath76 for the first time join the corresponding point @xmath141 to the original point @xmath75 along @xmath76 to obtain a closed loop . here @xmath142 denotes the first return map to the transversal @xmath76 induced by the vertical flow . denote by @xmath143 the corresponding cycle in @xmath144 . let the interval exchange transformation @xmath145 decompose @xmath76 into @xmath6 subintervals @xmath146 . it is easy to see that the `` first return cycle '' @xmath143 is piecewise constant : we have @xmath147 whenever @xmath75 and @xmath148 belong to the same subinterval @xmath149 , see fig . [ zorich : fig : renormalization ] . it is easy to see that @xmath150 the average of this sum with respect to the `` time '' @xmath91 tends to the asymptotic cycle @xmath72 . we need to study the deviation of this sum from the value @xmath151 . to do this consider a shorter subinterval @xmath152 as in fig . [ zorich : fig : renormalization ] . its length is chosen in such way , that the first return map of the vertical flow again induces an interval exchange transformation @xmath153 of @xmath6 subintervals . new first return cycles @xmath154 to the interval @xmath152 are expressed in terms of the initial first return cycles @xmath155 by the linear relations below ; the lengths @xmath156 of subintervals of the new partition @xmath157 are expressed in terms of the lengths @xmath158 of subintervals of the initial partition by dual linear relations : @xmath159 is completely determined by the initial interval exchange transformation @xmath160 and by the choice of @xmath161 . ( 100,50)(100,50 ) ( 230,-240)@xmath162 ( 300,-70)@xmath163 ( 360,-140)a ) ( 160,-400)b ) ( 345,-400)c ) ( 111,-48)@xmath9 ( 144,-9)@xmath10 ( 196,-5)@xmath11 ( 232,-37)@xmath12 ( 116,-95)@xmath12 ( 157,-126)@xmath11 ( 204,-119)@xmath10 ( 236,-84)@xmath9 to construct the cycle @xmath90 representing a long piece of leaf of the vertical foliation we followed the trajectory @xmath164 of the initial interval exchange transformation @xmath142 and computed the corresponding ergodic sum . passing to a shorter horizontal interval @xmath161 we can follow the trajectory @xmath165 of the new interval exchange transformation @xmath166 ( provided @xmath167 ) . since the subinterval @xmath152 is shorter than @xmath76 we cover the initial piece of trajectory of the vertical flow in a smaller number @xmath168 of steps . in other words , passing from @xmath169 to @xmath170 we accelerate the time : it is easy to see that the trajectory @xmath171 follows the trajectory @xmath172 but jumps over several iterations of @xmath169 at a time . this approach would not be efficient if the new first return map @xmath166 would be more complicated than the initial one . but we know that passing from @xmath169 to @xmath170 we stay within a family of interval exchange transformations of some fixed number @xmath6 of subintervals , and , moreover , that the new `` first return cycles '' @xmath154 and the lengths @xmath156 of the new subintervals are expressed in terms of the initial ones by means of the @xmath173-matrix @xmath174 , which depends only on the choice of @xmath161 and which can be easily computed . our strategy can be now formulated as follows . one can define an explicit algorithm ( generalizing euclidean algorithm ) which canonically associates to an interval exchange transformation @xmath160 some specific subinterval @xmath161 and , hence , a new interval exchange transformation @xmath166 . similarly to the euclidean algorithm our algorithm is invariant under proportional rescaling of @xmath76 and @xmath152 , so , when we find it convenient , we can always rescale the length of the interval to one . this algorithm can be considered as a map @xmath175 from the space of all interval exchange transformations of a given number @xmath6 of subintervals to itself . applying recursively this algorithm we construct a sequence of subintervals @xmath176 and a sequence of matrices @xmath177 describing transitions form interval exchange transformation @xmath178 to interval exchange transformation @xmath179 . taking a product @xmath180 we can immediately express the `` first return cycles '' to a microscopic subinterval @xmath181 in terms of the initial `` first return cycles '' to @xmath76 . considering now the matrices @xmath174 as the values of a matrix - valued function on the space of interval exchange transformations , we realize that we study the products of matrices @xmath174 along the orbits @xmath182 of the map on the space of interval exchange transformations . when the map is ergodic with respect to a finite measure , the properties of these products are described by the oseledets theorem , and the cycles @xmath90 have a deviation spectrum governed by the lyapunov exponents of the cocycle @xmath174 on the space of interval exchange transformations . note that the first return cycle to the subinterval @xmath181 ( which is very short ) represents the cycle @xmath90 corresponding to a very long trajectory @xmath183 of the initial interval exchange transformation . in other words , our renormalization procedure @xmath175 plays a role of a time acceleration machine : morally , instead of getting the cycle @xmath90 by following a trajectory @xmath183 of the initial interval exchange transformation for the exponential time @xmath184 we obtain the cycle @xmath90 applying only @xmath185 steps of the renormalization map @xmath175 on the space of interval exchange transformations . it remains to establish the relation between the the cocycle @xmath174 over the map @xmath175 and the teichmller geodesic flow . conceptually , this relation was elaborated in the original paper of w. veech @xcite . first let us discuss how can one `` almost canonically '' ( that is up to a finite ambiguity ) choose a zippered rectangles representation of a flat surface . note that fig . [ zorich : fig : renormalization ] suggests the way which allows to obtain infinitely many zippered rectangles representations of the same flat surface : we chop an appropriate rectangle on the right , put it atop the corresponding rectangle and then repeat the procedure recursively . this resembles the situation with a representation of a flat torus by a parallelogram : a point of the fundamental domain in fig . [ zorich : fig : space : of : flat : tori ] provides a canonical representative though any point of the corresponding @xmath186-orbit represents the same flat torus . a `` canonical '' zippered rectangles decomposition of a flat surface also belongs to some fundamental domain . following w. veech one can define the fundamental domain in terms of some specific choice of a `` canonical '' horizontal interval @xmath76 . namely , let us position the left endpoint of @xmath76 at a conical singularity . let us choose the length of @xmath76 in such way that the interval exchange transformation @xmath142 induced by the first return of the vertical flow to @xmath76 has minimal possible number @xmath118 of subintervals under exchange . among all such horizontal segments @xmath76 choose the shortest one , which length is greater than or equal to one . this construction is applicable to almost all flat surfaces ; the finite ambiguity corresponds to the finite freedom in the choice of the conical singularity and in the choice of the horizontal ray adjacent to it . since the interval @xmath76 defines a decomposition of ( almost any ) flat surface into `` zippered rectangles '' ( see fig . [ zorich : fig : renormalization ] ) we can pass from the space of flat surfaces to the space of zippered rectangles ( which can be considered as a finite ramified covering over the space of flat surfaces ) . teichmller geodesic flow lifts naturally to the space of zippered rectangles . it acts on zippered rectangles by expansion in horizontal direction and contraction in vertical direction ; i.e. the zippered rectangles are modified by linear transformations @xmath187 . however , as soon as the teichmller geodesic flow brings us out of the fundamental domain , we have to modify the zippered rectangles decomposition to the `` canonical one '' corresponding to the fundamental domain . ( compare to fig . [ zorich : fig : space : of : flat : tori ] where the teichmller geodesic flow corresponds to the standard geodesic flow in the hyperbolic metric on the upper half - plane . ) the corresponding modification of zippered rectangles ( chop an appropriate rectangle on the right , put it atop the corresponding rectangle ; repeat the procedure several times , if necessary ) is illustrated in fig . [ zorich : fig : renormalization ] . now everything is ready to establish the relation between the teichmller geodesic flow and the map @xmath175 on the space of interval exchange transformations . consider some codimension one subspace @xmath188 in the space of zippered rectangles transversal to the teichmller geodesic flow . say , @xmath188 might be defined by the requirement that the base @xmath76 of the zippered rectangles decomposition has length one , @xmath189 . this is the choice in the original paper of w. veech @xcite ; under this choice @xmath188 represents part of the boundary of the fundamental domain in the space of zippered rectangles . teichmller geodesic flow defines the first return map @xmath190 to the section @xmath188 . the map @xmath191 can be described as follows . take a flat surface of unit area decomposed into zippered rectangles @xmath192 with the base @xmath76 of length one . apply expansion in horizontal direction and contraction in vertical direction . for some @xmath193 the deformed zippered rectangles can be rearranged as in fig . [ zorich : fig : renormalization ] to get back to the base of length one ; the result is the image of the map @xmath191 . actually , we can first apply the rearrangement as in fig . [ zorich : fig : renormalization ] to the initial zippered rectangles @xmath192 and then apply the transformation @xmath194 the two operations commute . this gives , in particular , an explicit formula for @xmath193 . namely let @xmath195 be the width of the rightmost rectangle and let @xmath196 be the width of the rectangle , which top horizontal side is glued to the rightmost position at the base @xmath76 . ( for the upper zippered rectangle decomposition in fig . [ zorich : fig : renormalization ] we have @xmath197 and @xmath198 . ) then @xmath199 recall that a decomposition of a flat surface into zippered rectangles naturally defines an interval exchange transformation the first return map of the vertical flow to the base @xmath76 of zippered rectangles . hence , the map @xmath191 of the subspace @xmath188 of zippered rectangles defines an induced map on the space of interval exchange transformations . it remains to note that this induced map is exactly the map @xmath175 . in other words , the map @xmath190 induced by the first return of the teichmller geodesic flow to the subspace @xmath188 of zippered rectangles is the suspension of the map @xmath175 on the space of interval exchange transformations . we complete with a remark concerning the choice of a section . the natural section @xmath188 chosen in the original paper of w. veech @xcite is in a sense too large : the corresponding invariant measure ( induced from the measure on the space of flat surfaces ) is infinite . choosing an appropriate subset @xmath200 one can get finite invariant measure . moreover , the subset @xmath201 can be chosen in such way that the corresponding first return map @xmath202 of the teichmller geodesic flow is a suspension of some natural map @xmath203 on the space of interval exchange transformations , see @xcite . according to the results of h. masur @xcite and w. veech @xcite the teichmller geodesic flow is ergodic which implies ergodicity of the maps @xmath204 and @xmath203 . to apply oseledets theorem one should , actually , consider the induced cocycle @xmath205 over this new map @xmath203 instead of the cocycle @xmath174 over the map @xmath175 described above . consider a flat surface @xmath18 ; we always assume that the flat metric on @xmath18 has trivial holonomy , and that the surface @xmath18 has finite number of cone - type singularities . by convention a flat surface is endowed with a choice of direction , refereed to as a `` vertical direction '' , or as a `` direction to the north '' . since the flat metric has trivial holonomy , this direction can be transported in a unique way to any point of the surface . a geodesic segment joining two conical singularities and having no conical points in its interior is called _ saddle connection_. the case when boundaries of a saddle connection coincide is not excluded : a saddle connection might join a conical point to itself . in this section we study saddle connections and closed regular geodesics on a generic flat surface @xmath18 of genus @xmath206 . in particular , we count them and we explain the following curious phenomenon : saddle connections and closed regular geodesics often appear in pairs , triples , etc of parallel saddle connections ( correspondingly closed regular geodesics ) of the same direction and length . when all saddle connections ( closed regular geodesics ) in such configuration are short the corresponding flat surface is almost degenerate ; it is located close to the boundary of the moduli space . a description of possible configurations of parallel saddle connections ( closed geodesics ) gives us a description of the multidimensional `` cusps '' of the strata . the results of this section are based on the joint work with a. eskin and h. masur @xcite and on their work @xcite . a series of beautiful results developing the counting problems considered here were recently obtained by ya . vorobets @xcite . [ [ counting - closed - geodesics - and - saddle - connections . ] ] counting closed geodesics and saddle connections . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + closed geodesics on flat surfaces of higher genera have some similarities with ones on the torus . suppose that we have a regular closed geodesic passing through a point @xmath207 . emitting a geodesic from a nearby point @xmath75 in the same direction we obtain a parallel closed geodesic of the same length as the initial one . thus , closed geodesics appear in families of parallel closed geodesics . however , in the torus case every such family fills the entire torus while a family of parallel regular closed geodesics on a flat surfaces of higher genus fills only part of the surface . namely , it fills a flat cylinder having a conical singularity on each of its boundaries . typically , a maximal cylinder of closed regular geodesics is bounded by a pair of closed saddle connections . reciprocally , any saddle connection joining a conical point @xmath208 to itself and coming back to @xmath208 at the angle @xmath5 bounds a cylinder filled with closed regular geodesics . a geodesic representative of a homotopy class of a curve on a flat surface is realized in general by a broken line of geodesic segments with vertices at conical points . by convention we consider only closed _ regular _ geodesics ( which by definition do not pass through conical points ) or saddle connections ( which by definition do not have conical points in its interior ) . everywhere in this section we normalize the area of a flat surface to one . let @xmath209 be the number of saddle connections of length at most @xmath210 on a flat surface @xmath18 . let @xmath211 be the number of maximal cylinders filled with closed regular geodesics of length at most @xmath210 on @xmath18 . it was proved by h. masur that for any flat surface @xmath18 both counting functions @xmath212 grow quadratically in @xmath210 . namely , there exist constants @xmath213 such that @xmath214 for @xmath210 sufficiently large . recently ya . vorobets has obtained uniform estimates for the constants @xmath215 and @xmath216 which depend only on the genus of @xmath18 , see @xcite . passing from _ all _ flat surfaces to _ almost all _ surfaces in a given connected component of a given stratum one gets a much more precise result , see @xcite : for almost all flat surfaces @xmath18 in any stratum @xmath33 the counting functions @xmath209 and @xmath211 have exact quadratic asymptotics @xmath217 moreover , the siegel veech constants @xmath218 ( correspondingly @xmath219 ) coincide for almost all flat surfaces @xmath18 in each connected component @xmath220 of the stratum . [ [ phenomenon - of - higher - multiplicities . ] ] phenomenon of higher multiplicities . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + note that the direction to the north is well - defined even at a conical point of a flat surface , moreover , at a conical point @xmath221 with a cone angle @xmath222 we have @xmath223 different directions to the north ! consider some saddle connection @xmath224 $ ] with an endpoint at @xmath221 . memorize its direction , say , let it be the north - west direction . let us launch a geodesic from the same starting point @xmath221 in one of the remaining remaining @xmath225 north - west directions . let us study how big is the chance to hit @xmath226 ones again , and how big is the chance to hit it after passing the same distance as before . we do not exclude the case @xmath227 . intuitively it is clear that the answer to the first question is : `` the chances are low '' and to the second one is `` the chances are even lower '' . this makes the following theorem ( see @xcite ) somehow counterintuitive : for almost any flat surface @xmath18 in any stratum and for any pair @xmath228 of conical singularities on @xmath18 the function @xmath229 counting the number of pairs of parallel saddle connections of the same length joining @xmath221 to @xmath226 has exact quadratic asymptotics @xmath230 where the siegel veech constant @xmath231 depends only on the connected component of the stratum and on the cone angles at @xmath221 and @xmath226 . for almost all flat surfaces @xmath18 in any stratum one can not find neither a single pair of parallel saddle connections on @xmath18 of different length , nor a single pair of parallel saddle connections joining different pairs of singularities . analogous statements ( with some reservations for specific connected components of certain strata ) can be formulated for arrangements of @xmath232 parallel saddle connections . the situation with closed regular geodesics is similar : they might appear ( also with some exceptions for specific connected components of certain strata ) in families of @xmath233 distinct maximal cylinders filled with parallel closed regular geodesics of equal length . a general formula for the siegel veech constant in the corresponding quadratic asymptotics is presented at the end of this section , while here we want to discuss the numerical values of siegel veech constants in a simple concrete example . we consider the principal strata @xmath234 in small genera . let @xmath235 be the corresponding counting function , where @xmath223 is the number of distinct maximal cylinders filled with parallel closed regular geodesics of equal length bounded by @xmath210 . let @xmath236 the table below ( extracted from @xcite ) presents the values of @xmath237 for @xmath238 . note that for a generic flat surface @xmath18 of genus @xmath13 a configuration of @xmath239 cylinders is not realizable , so we do not fill the corresponding entry . @xmath240k & g=1 & g=2 & g=3 & g=4 \\ \hline & & & & \\ [ -\halfbls]&&&&\\ [ -2\halfbls]\ 1\ & \cfrac{1}{2}\cdot\cfrac{1}{\zeta(2)}\approx 0.304 & \cfrac{5}{2}\cdot\cfrac{1}{\zeta(2)}\approx 1.52 & \cfrac{36}{7}\cdot\cfrac{1}{\zeta(2)}\approx 3.13 & \cfrac{3150}{377}\cdot\cfrac{1}{\zeta(2)}\approx 5.08 \\ [ -1\halfbls]&&&&\\ \hline & & & & \\ [ -\halfbls]&&&&\\ [ -2\halfbls]\ 2\ & - & - & \cfrac{3}{14}\cdot\cfrac{1}{\zeta(2)}\approx 0.13 & \cfrac{90}{377}\cdot\cfrac{1}{\zeta(2)}\approx 0.145 \\ [ -\halfbls]&&&&\\ \hline & & & & \\ [ -\halfbls]&&&&\\ [ -2\halfbls]\ 3\ & - & - & - & \cfrac{5}{754}\cdot\cfrac{1}{\zeta(2)}\approx 0.00403 \\ [ -\halfbls]&&&&\\ \hline \end{array}\ ] ] comparing these values we see , that our intuition was not quite misleading . morally , in genus @xmath241 a closed regular geodesic belongs to a one - cylinder family with `` probability '' @xmath242 , to a two - cylinder family with `` probability '' @xmath243 and to a three - cylinder family with `` probability '' only @xmath244 ( where `` probabilities '' are calculated proportionally to the siegel veech constants @xmath245 ) . [ [ rigid - configurations - of - saddle - connections - and - cusps - of - the - strata . ] ] rigid configurations of saddle connections and `` cusps '' of the strata . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + a saddle connection or a regular closed geodesic on a flat surface @xmath18 persists under small deformations of @xmath18 inside the corresponding stratum . it might happen that any deformation of a given flat surface which shortens some specific saddle connection necessarily shortens some other saddle connections . we say that a collection @xmath246 of saddle connections is _ rigid _ if any sufficiently small deformation of the flat surface inside the stratum preserves the proportions @xmath247 of the lengths of all saddle connections in the collection . it was shown in @xcite that all saddle connections in any rigid collection are _ their directions and lengths can be expressed in terms of integrals of the holomorphic 1-form @xmath22 along corresponding paths , this implies that homologous saddle connections @xmath248 are parallel and have equal length and either all of them join the same pair of distinct singular points , or all @xmath249 are closed loops . this implies that when saddle connections in a rigid collection are contracted by a continuous deformation , the limiting flat surface generically decomposes into several connected components represented by nondegenerate flat surfaces @xmath250 , see fig . [ zorich : fig : dist : sad : mult ] , where @xmath223 might vary from one to the genus of the initial surface . let the initial surface @xmath18 belong to a stratum @xmath33 . denote the set with multiplicities @xmath251 by @xmath252 . let @xmath253 be the stratum ambient for @xmath254 . the stratum @xmath255 of disconnected flat surfaces @xmath256 is referred to as a _ principal boundary _ stratum of the stratum @xmath257 . for any connected component of any stratum @xmath257 the paper @xcite describes all principal boundary strata ; their union is called the principal boundary of the corresponding connected component of @xmath257 . ( -160,0)(-160,0 ) ( 0,0 ) ( 0,0)(0,0 ) ( -153,-36)@xmath258 ( -6,-46)@xmath259 ( 5,0 ) ( 0,0)(0,0 ) ( -94,-127)@xmath260 ( 0,0 ) ( 0,0)(0,0 ) ( 20,-12)@xmath261 ( 155,-22)@xmath262 ( 0,0 ) ( 0,0)(0,0 ) ( 80,-180)@xmath263 ( 20,-117)@xmath261 ( 145,-126)@xmath262 ( 84,-140)@xmath264 ( 70,-140)@xmath265 ( 79,-125)@xmath266 the paper @xcite also presents the inverse construction . consider any flat surface @xmath267 in the principal boundary of @xmath257 ; consider a sufficiently small value of a complex parameter @xmath268 . one can reconstruct the flat surface @xmath269 endowed with a collection of homologous saddle connections @xmath270 such that @xmath271 , and such that degeneration of @xmath18 contracting the saddle connections @xmath249 in the collection gives the surface @xmath272 . this inverse construction involves several surgeries of the flat structure . having a disconnected flat surface @xmath273 one applies an appropriate surgery to each @xmath254 producing a surface @xmath274 with boundary . the surgery depends on the parameter @xmath275 : the boundary of each @xmath274 is composed from two geodesic segments of lengths @xmath276 ; moreover , the boundary components of @xmath274 and @xmath277 are compatible , which allows to glue the compound surface @xmath18 from the collection of surfaces with boundary , see fig . [ zorich : fig : dist : sad : mult ] as an example . a collection @xmath278 of homologous saddle connections determines the following data on combinatorial geometry of the decomposition @xmath279 : the number of components , their boundary structure , the singularity data for each component , the cyclic order in which the components are glued to each other . these data are referred to as a _ configuration _ of homologous saddle connections . a configuration @xmath280 uniquely determines the corresponding boundary stratum @xmath281 ; it does not depend on the collection @xmath119 of homologous saddle connections representing the configuration @xmath280 . the constructions above explain how configurations @xmath280 of homologous saddle connections on flat surfaces @xmath269 determine the `` cusps '' of the stratum @xmath257 . consider a subset @xmath282 of surfaces of area one having a saddle connection shorter than @xmath275 . up to a subset @xmath283 of negligibly small measure the set @xmath284 might be represented as a disjoint union @xmath285 of neighborhoods @xmath286 of the corresponding `` cusps '' @xmath280 . here @xmath280 runs over a finite set of configurations admissible for the given stratum @xmath287 ; this set is explicitly described in @xcite . when a configuration @xmath280 is composed from homologous saddle connections joining _ distinct _ zeroes , the neighborhood @xmath286 of the cusp @xmath280 has the structure of a fiber bundle over the corresponding boundary stratum @xmath281 ( up to a difference in a set of a negligibly small measure ) . a fiber of this bundle is represented by a finite cover over the euclidean disc of radius @xmath275 ramified at the center of the disc . moreover , the canonical measure in @xmath286 decomposes into a product measure of the canonical measure in the boundary stratum @xmath281 and the euclidean measure in the fiber ( see @xcite ) , so @xmath288 we warn the reader that the correspondence between compactification of the moduli space of abelian differentials and the deligne mamford compactification of the underlying moduli space of curves is not straightforward . in particular , the desingularized stable curve corresponding to the limiting flat surface generically _ is not _ represented as a union of riemann surfaces corresponding to @xmath250 the stable curve might contain more components . [ [ evaluation - of - the - siegelveech - constants . ] ] evaluation of the siegel veech constants . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + consider a flat surface @xmath18 . to every closed regular geodesic @xmath119 on @xmath18 we can associate a vector @xmath289 in @xmath1 having the length and the direction of @xmath119 . in other words , @xmath290 , where we consider a complex number as a vector in @xmath291 . applying this construction to all closed regular geodesic on @xmath18 we construct a discrete set @xmath292 . consider the following operator @xmath293 from functions with compact support on @xmath1 to functions on a connected component @xmath294 of the stratum @xmath295 : @xmath296 function @xmath297 generalizes the counting function @xmath211 introduced in the beginning of this section . namely , when @xmath298 is the characteristic function @xmath299 of the disc of radius @xmath210 with the center at the origin of @xmath1 , the function @xmath300 counts the number of regular closed geodesics of length at most @xmath210 on a flat surface @xmath18 . for any function @xmath301 with compact support the following equality is valid : @xmath302 where the constant @xmath303 does not depend on the function @xmath304 . note that this is an exact equality . in particular , choosing the characteristic function @xmath299 of a disc of radius @xmath210 as a function @xmath304 we see that for any positive @xmath210 the _ average _ number of closed regular geodesics not longer than @xmath210 on flat surfaces @xmath305 is exactly @xmath306 , where the siegel veech constant @xmath303 does not depend on @xmath210 , but only on the connected component @xmath307 . the theorem of eskin and masur cited above tells that for large values of @xmath210 one gets approximate equality @xmath308 `` pointwisely '' for almost all individual flat surfaces @xmath309 . it is proved in @xcite that the corresponding siegel veech constant @xmath310 coincides with the constant @xmath303 in equation above . actually , the same technique can be applied to count separately pairs , triples , or any other specific configurations @xmath280 of homologous saddle connections . every time when we find a collection of homologous saddle connections @xmath248 representing the chosen configuration @xmath280 we construct a vector @xmath311 . since all @xmath312 are homologous , we can take any of them as @xmath249 . taking all possible collections of homologous saddle connections on @xmath18 representing the fixed configuration @xmath280 we construct new discrete set @xmath313 and new functional @xmath314 . theorem of eskin and masur and theorem of veech @xcite presented above are valid for @xmath315 . the corresponding siegel veech constant @xmath316 responsible for the quadratic growth rate @xmath317 of the number of collections of homologous saddle connections of the type @xmath280 on an individual generic flat surface @xmath18 coincides with the constant @xmath318 in the expression analogous to . formula can be applied to @xmath319 for any value of @xmath210 . in particular , instead of taking large @xmath210 we can choose a very small @xmath320 . the corresponding function @xmath321 counts how many collections of parallel @xmath275-short saddle connections ( closed geodesics ) of the type @xmath280 we can find on a flat surface @xmath305 . for the flat surfaces @xmath18 outside of the subset @xmath322 there are no such saddle connections ( closed geodesics ) , so @xmath323 . for surfaces @xmath18 from the subset @xmath324 there is exactly one collection like this , @xmath325 . finally , for the surfaces from the remaining ( very small ) subset @xmath326 one has @xmath327 . eskin and masur have proved in @xcite that though @xmath321 might be large on @xmath328 the measure of this subset is so small ( see @xcite ) that @xmath329 and hence @xmath330 this latter volume is almost the same as the volume @xmath331 of the neighborhood of the cusp @xmath280 evaluated in equation above , namely , @xmath332 ( see @xcite ) . taking into consideration that @xmath333 and applying siegel veech formula to @xmath334 we finally get @xmath335 which implies the following formula for the siegel veech constant @xmath336 : @xmath337 sums of the lyapunov exponents @xmath140 discussed in section [ zorich : s : generic : geodesics ] are closely related to the siegel veech constants . according to the theorems of h. masur @xcite and of w. veech @xcite teichmller geodesic flow is ergodic on every connected component of every stratum of flat surfaces . thus , the lyapunov exponents @xmath338 of the teichmller geodesic flow responsible for the deviation spectrum of generic geodesics on a flat surface ( see section [ zorich : s : generic : geodesics ] ) , or siegel veech constants responsible for counting of closed geodesics on a flat surface ( see section [ zorich : s : closed : geodesics : on : flat : surfaces ] ) are specific for each connected component of each stratum . the fact that the strata @xmath56 are not necessarily connected was observed by w. veech . in order to formulate the classification theorem for connected components of the strata @xmath33 we need to describe the classifying invariants . there are two of them : spin structure and hyperellipticity . both notions are applicable only to part of the strata : flat surfaces from the strata @xmath339 have even or odd spin structure . the strata @xmath340 and @xmath341 have a special hyperelliptic connected component . consider a flat surface @xmath18 from a stratum @xmath339 . let @xmath342 be a smooth closed path on @xmath18 ; here @xmath343 is a standard circle . note that at any point of the surfaces @xmath18 we know where is the `` direction to the north '' . hence , at any point @xmath344 we can apply a compass and measure the direction of the tangent vector @xmath345 . moving along our path @xmath346 we make the tangent vector turn in the compass . thus we get a map @xmath347 from the parameter circle to the circumference of the compass . this map is called the _ gauss map_. we define the _ index _ @xmath348 of the path @xmath349 as a degree of the corresponding gauss map ( or , in other words as the algebraic number of turns of the tangent vector around the compass ) taken modulo @xmath350 . @xmath351 it is easy to see that @xmath348 does not depend on parameterization . moreover , it does not change under small deformations of the path . deforming the path more drastically we may change its position with respect to conical singularities of the flat metric . say , the initial path might go on the left of @xmath352 and its deformation might pass on the right of @xmath352 . this deformation changes the @xmath353 . however , if the cone angle at @xmath352 is of the type @xmath354 , then @xmath355 does not change ! this observation explains why @xmath348 is well - defined for a free homotopy class @xmath356 $ ] when @xmath357 ( and hence , when all cone angles are odd multiples of @xmath8 ) . the lemma above shows that the parity of the spin structure is an invariant of connected components of the strata of those abelian differentials ( equivalently , flat surfaces ) , which have zeroes of even degrees ( equivalently , conical points with cone angles which are odd multiples of @xmath8 ) . a flat surface @xmath18 may have a symmetry ; one specific family of such flat surfaces , which are `` more symmetric than others '' is of a special interest for us . recall that there is a one - to - one correspondence between flat surfaces and pairs ( riemann surface @xmath363 , holomorphic 1-form @xmath22 ) . when the corresponding riemann surface is hyperelliptic the hyperelliptic involution @xmath364 acts on any holomorphic 1-form @xmath22 as @xmath365 . we say that a flat surface @xmath18 is a hyperelliptic flat surface if there is an isometry @xmath366 such that @xmath367 is an involution , @xmath368 , and the quotient surface @xmath369 is a topological sphere . in flat coordinates differential of such involution obviously satisfies @xmath370 . in a general stratum @xmath33 hyperelliptic flat surfaces form a small subspace of nontrivial codimension . however , there are two special strata , namely , @xmath340 and @xmath341 , for which hyperelliptic flat surfaces form entire hyperelliptic connected components @xmath371 and @xmath372 correspondingly . note that in the stratum @xmath341 there are hyperelliptic flat surfaces of two different types . a hyperelliptic involution @xmath373 may fix the conical points or might interchange them . it is not difficult to show that for flat surfaces from the connected component @xmath372 the hyperelliptic involution interchanges the conical singularities . the remaining family of those hyperelliptic flat surfaces in @xmath341 , for which the hyperelliptic involution keeps the saddle points fixed , forms a subspace of nontrivial codimension in the complement @xmath374 . thus , the hyperelliptic connected component @xmath372 does not coincide with the space of all hyperelliptic flat surfaces . each of the strata @xmath396 , @xmath397 of the moduli space of flat surfaces of genus @xmath398 has two connected components : the hyperelliptic one , and one having odd spin structure . the other strata are connected for genus @xmath398 . as a much more serious accessible introduction to teichmller dynamics i can recommend a collection of surveys of a. eskin @xcite , g. forni @xcite , p. hubert and t. schmidt @xcite and h. masur @xcite , organized as a chapter of the handbook of dynamical systems . i also recommend recent surveys of h. masur and s. tabachnikov @xcite and of j. smillie @xcite especially in the aspects related to billiards in polygons . the part concerning renormalization and interval exchange transformations is presented in the survey of j .- c . yoccoz @xcite . the ideas presented in the current paper are illustrated in more detailed way in the survey @xcite . considerable part of results presented in this survey is obtained in collaboration . i use this opportunity to thank a. eskin , m. kontsevich and h. masur for the pleasure to work with them . i am grateful to m .- c . vergne for her help in preparation of pictures . i highly appreciate careful and responsible editing by i. zimmermann . r. krikorian , dviation de moyennes ergodiques , flots de teichmller et cocycle de kontsevich zorich ( daprs forni , kontsevich , zorich ) . sminaire bourbaki 927 ( 2003 ) . _ astrisque _ * 299 * ( 2005 ) , 5993 . m. kontsevich , lyapunov exponents and hodge theory . in _ `` the mathematical beauty of physics '' ( saclay , 1996 ) , ( in honor of c. itzykson)_. adv . phys . 24 . world sci . publishing , river edge , nj 1997 , 318332 . j. smillie , the dynamics of billiard flows in rational polygons . in : _ dynamical systems , ergodic theory and applications _ ( ed . by ya . g. sinai ) , encyclopaedia of mathematical sciences 100 , math . physics 1 . springer verlag , berlin 2000 , 360382 . vorobets , periodic geodesics on generic translation surfaces . in : _ algebraic and topological dynamics _ ( ed . by s. kolyada , yu . i. manin and t. ward ) . contemporary mathematics 385 , ams , providence , ri , 2005 , 205258 . yoccoz , continuous fraction algorithms for interval exchange maps : an introduction . in _ `` frontiers in number theory , physics and geometry i. on random matrices , zeta functions , and dynamical systems '' _ , p. cartier , b. julia , p. moussa , p. vanhove ( editors ) , springer verlag , berlin 2006 , 403437 . a. zorich , how do the leaves of a closed 1-form wind around a surface . in _ `` pseudoperiodic topology''_. ( ed . by v. i. arnold , m. kontsevich , a. zorich ) . ams translations , ser . 197 , ams , providence , ri , 1999 , 135178 . a. zorich , flat surfaces . in _ `` frontiers in number theory , physics and geometry . volume i. on random matrices , zeta functions , and dynamical systems '' _ , p. cartier , b. julia , p. moussa , p. vanhove ( editors ) , springer verlag , berlin 2006 , 439585 .

the final - state interaction in multichannel decay processes is systematically studied in the hadronic picture with application to decay in mind . since the final - state interaction is intrinsically interwoven with the decay interaction in this case , no simple phase theorem like `` watson s theorem '' holds for experimentally observed final states . we first solve exactly the two - channel problem as a toy model in order to clarify the issues . the constraints of the two - channel approximation turns out to be too stringent for most decay modes , but realistic multichannel problems are too complex for useful quantitative analysis at present . to alleviate the stringent constraints of the two - body problem and to cope with complexity beyond it , we introduce a method of approximation that is applicable to the case where one prominent inelastic channel dominates over all others . we illustrate this approximation method with the amplitude of the decay fed by the intermediate states of a charmed - meson pair . even with our approximation we need more accurate information of strong interactions than we have now . nonetheless we are able to obtain some insight in the issue and draw useful conclusions on general features on the strong phases .

0710.5534

the well - known phase theorem @xcite holds for the final - state interaction ( fsi ) of decay processes when the final state consists of a single eigenstate of scattering . while no simple nontrivial extension is known in the case of multichannel final states , some calculations were made in the past with unjustified extension of the single - channel phase theorem@xcite . a two - channel problem was studied with a certain class of @xmath2-matrix and the correct observation was made that inelastic channels are the main source of strong phases in many @xmath0 decay modes@xcite . however , it is not easy to obtain quantitatively reliable results from the two - channel model . taking the large limit of open channels , the statistical model @xcite was proposed as an alternative approach . quantitatively , however , it is short of predictive power since it does not ask for detailed knowledge of strong interaction . when one approaches the problem in the quark - gluon picture , one faces inability or large uncertainty in computing contributions of the soft collinear constituents numerically . it is fair to say that at present we are far from successful computation of a fsi phase in multichannel decay processes . in this paper we first study the two - channel problem in detail . the problem is solvable in a reasonably compact form without approximation or assumption if relevant information is available about strong interaction physics and decay branching fractions . the general solution to the two - channel toy - model shows how the elastic scattering phases and the channel coupling contribute to the total fsi phase . it concludes in agreement with donoghue _ et al_@xcite that if a large strong phase emerges in the @xmath0 decay into two light - mesons , its major source is coupling to decay channels that have large branching fractions . although it points to the source of problems in strong phases , the simple two - channel model is inapplicable to @xmath0 decay . we proceed to the case of more than two decay channels . even the three - channel problem is mathematically too complicated for solving in a compact form . on the physics side our knowledge of strong interaction at total energy 5 gev ( @xmath3 ) is not good enough to carry through the analysis with precision . to cope with formidable complexity of the problem , we introduce an approximation method that works in the case that , aside from the channel of our interest , one inelastic channel dominates over all others . this is different from an approximate two - channel problem . it can happen , for instance , to the two - body light - hadron decays of @xmath0 meson when they couple to the charmed - meson pair states . we apply our approximation method to the decay @xmath1 and make semiquantitative analysis with knowledge of hadron physics currently available to us . two basic ingredients in discussion of fsi are unitarity and time - reversal . in the standard model the decay interaction is sum of effective local operators @xmath4 , each of which has a cp - violating ( and therefore t - violating ) phase factored out as @xmath5 the t - violating `` weak phase '' @xmath6 arises from the ckm elements . in computing decay amplitudes we work separately on the @xmath7-invariant part @xmath8 of the decay interaction . in the case that @xmath7-violation is more general , we can break up the weak interactions @xmath9 into @xmath7-even and @xmath7-odd parts as @xmath10 so that @xmath11 then both @xmath12 and @xmath13 are @xmath7-even since @xmath14 under time reversal . and @xmath15 for the first term @xmath16 of eq . ( [ sm ] ) . ] we should compute the decay amplitudes for @xmath12 and @xmath13 separately and take a suitable sum of them at the end . therefore it is sufficient to consider in general only @xmath7-even weak interactions . no matter what method one may use , computation of fsi phases must always be made separately for different decay operators . because two decay operators generate two different fsi phases for the same decay process even if the net quantum numbers of operators are identical . we consider in this paper only final - state interaction of strong interaction though it is in principle easy to include electromagnetic fsi . we shall refer to fsi phases also as strong phases in this paper . with @xmath7-invariance , the strong @xmath2-matrix operator obeys @xmath17 we can always choose phases of states such that the @xmath7-invariant @xmath2-matrix elements ( @xmath18 ) are not only unitary but symmetric ; @xmath19 since @xmath20 and @xmath21 under time reversal . it is emphasized here that the requirement of eq . ( [ symmetric ] ) fixes the phases of states except for the overall sign of @xmath22 . and @xmath23 , @xmath24 and @xmath25 would acquire phases of opposite signs @xmath26 so that the equality @xmath27 would break down . ] specifically for the eigenchannels of the @xmath2-matrix @xmath28 and @xmath29 , it holds by definition that @xmath30 where @xmath31 is the eigenphase shift . in the case of decay matrix elements , the initial state is a one - particle state that is stable with respect to strong interaction . since the initial decaying state is an asymptotic state with respect to strong interaction , there is no distinction between `` in '' and `` out '' states . for @xmath0 decay @xmath32 where we choose the @xmath0 meson at rest . ( [ b ] ) removes an arbitrary unphysical phase from the state @xmath33 too . a simple relation results from time reversal of the decay matrix element @xmath34 when the final state @xmath35 is an eigenstate of @xmath2 . t - invariance of @xmath8 leads to @xmath36 inserting the completeness relation @xmath37 next to @xmath8 in the right - hand side , we obtain from eq . ( [ t ] ) with eq . ( [ eigen ] ) @xmath38 with the decay amplitude @xmath34 into the eigenchannel @xmath39 denoted by @xmath40 , this relation reads @xmath41 namely , @xmath42 . this is the well - known phase theorem usually referred to as watson s theorem@xcite . it is a powerful theorem when the final state is an eigenstate of @xmath2 . it has been an important tool of analysis in hyperon decay and @xmath43 decay . but this relation is of little use when rescattering has inelasticity . we mean by _ inelasticity _ that observable final states are linear combinations of eigenstates whose weights are _ not _ simply determined by the clebsch - gordan coefficients of isospin symmetry or su(3 ) coefficients . in such cases the phase of the elastic scattering amplitude has little to nothing to do with the fsi phase of the corresponding decay amplitude , as we shall see it in a moment . usefulness of the phase theorem is thus limited to the decay of low - mass particles where rescattering is purely elastic up to isospin structure . if the @xmath44 meson mass were sufficiently above 1 gev , for instance , @xmath45 of definite isospin would no longer be an eigenstate of @xmath2-matrix even approximately . the state @xmath46 and @xmath47 would enter an @xmath2-matrix eigenchannel with @xmath45 and composition of such an eigenstate depends on low - energy dynamics of the transition among @xmath45 , @xmath46 , and @xmath47 . in the case of @xmath0 decay an experimentally observed final state is a linear combination of many different @xmath2-matrix eigenstates so that the net fsi phase results from the eigenphases weighted with the decay amplitudes of @xmath48 . for instance , the @xmath45 final state in @xmath49 of @xmath50 decay . the state @xmath51 ( in @xmath52-wave ) is far from being an @xmath2-matrix eigenstate at energy @xmath53 . if we want to use the phase theorem eq . ( [ ph ] ) , we must expand @xmath54 in the strong @xmath2-matrix eigenstates at @xmath53 . however , we have little knowledge of these eigenstates since their composition depends sensitively on strong interaction at long and intermediate distances . experimentally the two - body channels do not account for all final states in @xmath0 decay . three and four particle final states of the same @xmath55 may be significant . unless there is a good reason to believe that channel coupling is negligible among these final states , the strong @xmath2-matrix eigenstates at @xmath53 are made of many different particle states ( @xmath45 , @xmath46 , @xmath56 , @xmath57 , @xmath58 ) . therefore , if we expand the @xmath45 state at total energy @xmath53 , it is a linear combination of many different eigenstates of strong @xmath2-matrix . we would have to know these expansion coefficients in order to determine the fsi phase of @xmath45 final state with the phase theorem . let us formulate what we have described above . when an observable final state @xmath59 is not @xmath2-matrix eigenstate , we expand it in the eigenstates of @xmath2-matrix @xmath60 as - matrix eignestates by @xmath61 , @xmath62 and the observed particle states by @xmath63 , @xmath64 , @xmath65 , @xmath58 . ] @xmath66 we choose that the @xmath2-matrix is symmetric ( @xmath67 ) in the basis of the observable states . ( cf eq./ ( [ symmetric ] ) ) then the expansion coefficients @xmath68 are real , that is , the transformation matrix @xmath69 in eq . ( [ expansion ] ) is an orthogonal matrix . ( see appendix if proof is needed . ) with this expansion and eq . ( [ ph ] ) , @xmath70 we are able to write the right - hand side of eq . ( [ multichannel ] ) in terms of observable decay amplitudes @xmath71 s as @xmath72 where @xmath73 has been used . this is the fundamental relation in discussion of fsi . the physical picture is simple and clear . ( fig.1 ) the input is unitarity and time - reversal of the @xmath2-matrix aside from choice of unphysical phases of states . dynamical information of strong interaction is fed through the eigenphases @xmath31 and the orthogonal mixing matrix @xmath69 . in addition , we must provide relative magnitude of the amplitudes @xmath74 as independent pieces of input from weak interaction . consequently the phase of @xmath74 depends not only on strong interaction but also on weak interaction . it is clear here that the fsi phase of a decay amplitude has virtually nothing to do with the phase of the elastic scattering amplitude @xmath75 of @xmath76 at 5 gev . although it looks almost futile to go any further , the purpose of this paper is to extract something useful for b decay out of eq . ( [ fsi ] ) . we focus on the fsi phases of the two - body decay modes of the @xmath0 meson . the phases of three - body decay amplitudes depend on the sub - energies of three particles . it is only the phases integrated over the sub - energies with the total energy fixed to @xmath53 that enter eq . ( [ fsi ] ) . we need the @xmath2-matrix elements @xmath77 from experiment . to be concrete , let us consider the elastic @xmath78 scattering amplitude as an example . the argument below is identical for other two - body channels of light mesons . in @xmath0 decay the relevant partial - wave channel is @xmath78 in @xmath52-wave ( @xmath79 ) with isospin @xmath49 and @xmath80 . although the high - energy @xmath45 scattering can not be directly measured in experiment , we can make a reasonable estimate about elastic @xmath45 scattering since the center - of - momentum energy 5 gev is in the high - energy asymptotic region well above the @xmath45 resonances . to the open charm channels@xcite is one of interesting subjects of our study later in this paper . ] the two - body light - hadron scattering in the high energy asymptotic region was studied theoretically and experimentally in the 1960 s . the regge theory describes this physics well . the properties of the regge trajectories can be deduced from meson - nucleon and nucleon - nucleon scattering even though we have no meson - meson scattering experiment . we give a very brief review@xcite of our knowledge in this area of the past time since it is an important input in our study of fsi . first of all , the pomeron exchange dominates in high - energy elastic scattering . ( fig 2 ) the pomeron may include a cut and can be more a complicated singularity than a simple pole in @xmath81-plane unlike the non - pomeron trajectories such as @xmath82 and @xmath83 . at the level of numerical accuracy of our discussion , however , we treat the pomeron as a simple pole at @xmath84 with the intercept @xmath85 and a vanishingly small slope . this entails factorization of the @xmath81-plane residue into product of two vertices . since isospin is zero for the pomeron , it contributes equally to the @xmath49 and @xmath80 states of the crossed channels @xmath45 . with these properties of the pomeron we obtain necessary pieces of information on @xmath45 scattering from @xmath86 and @xmath87 scattering at high energies . the invariant amplitude for the asymptotic elastic scattering is parametrized with the regge parameters of the pomeron in the form @xmath88 where @xmath89 and @xmath90 are the factorized residue and the trajectory of the pomeron , respectively , in terms of the invariant momentum transfer @xmath91 . rise of @xmath92 does not set in at @xmath935 gev . we may safely ignore it in our numerical work . we have included in @xmath94 the @xmath91-dependence of some other kinematical factors such as @xmath95 . the value of @xmath96 is at our choice , normally chosen to be 1 gev@xmath97 . ] we can fix @xmath98 by the optical theorem @xmath99 and the factorization relation @xmath100 at @xmath101 @xcite . the value @xmath102 is in line with the empirical quark counting rule , @xmath103@xcite . the width of the forward peak is more or less universal to all elastic hadron scatterings@xcite . we approximate the pomeron slope to zero ( @xmath104 ) and fit the forward elastic peak with the standard exponential form @xmath105 . the diffraction peak parameter is fixed by experiment to @xmath106 by the elasticity @xmath107 at @xmath101@xcite . this value of @xmath108 reproduces the @xmath91 dependence of @xmath109 that falls by roughly three orders of magnitude from @xmath110 to @xmath111 for @xmath87 and @xmath86 scattering@xcite . it is justified by factorization to choose the @xmath45 forward peak parameter equal to that of @xmath87 and @xmath86 . the @xmath45 invariant amplitude at high energies is thus set to @xmath112 where we choose @xmath113 . the uncertainty in @xmath108 is primarily due to whether one estimates it with @xmath107 of @xmath114 or @xmath87 and @xmath115 . the partial - wave amplitudes @xmath116 can be projected out of @xmath117 . the result for the @xmath52-wave is : @xmath118 which leads to the @xmath52-wave @xmath2-matrix with @xmath119 , @xmath120 hereafter we shall often parametrize strength of elastic scattering by @xmath121 as @xmath122 with eq . ( [ partial ] ) , the pomeron contribution to the @xmath2-matrix of @xmath45 scattering at @xmath53 is @xmath123 the partial - wave amplitudes @xmath116 extracted from the flat pomeron amplitude are purely imaginary for all @xmath124 . the amplitude @xmath125 approaches asymptotically the imaginary axis below the center of the argand diagram ( shown schematically for @xmath49 in fig . if one described the high inelasticity of high - energy @xmath45 scattering by an absorptive black sphere potential , one would have @xmath126 ( _ i.e. , _ @xmath127 ) . in this limit @xmath128 for all @xmath124 s by shadow scattering effect , which is in disagreement with experiment . although the numerical value in the right - hand side of eq . ( [ spi ] ) has been extracted for the @xmath45 channel , it is much the same for other two - meson channels . with a help of the @xmath129 cross section @xmath130@xcite we obtain @xmath131 where uncertainties are comparable to @xmath132 quoted for @xmath45 in eq . ( [ spi ] ) or a little larger . the values in eqs . ( [ pi ] ) and ( [ k ] ) are the pomeron contribution alone . the nonleading regge exchanges generate a small imaginary part for @xmath133 . the relevant trajectories are the @xmath82 and @xmath83 in the case of @xmath45 . their contributions can be estimated with a few additional theoretical inputs and the exchange degeneracy of @xmath82 and @xmath83 . ] from the cross section difference @xmath134 at @xmath101@xcite . within the uncertainty due to @xmath91-dependence of the residue @xmath135 , their contributions to @xmath136 are at the level of @xmath137 for @xmath45 . some may wonder about validity of extracting the @xmath52-wave amplitude from the forward peak region alone . the @xmath52-wave amplitude has a flat angular dependence so that the contribution of @xmath125 extends equally to all directions ( @xmath138 ) . on the other hand experiment shows that the forward peak falls off by more than three orders of magnitude and there is no sign of the @xmath52-wave contribution at large angles . but this is no surprise . the @xmath52-wave amplitude at large angles is canceled by the partial - wave amplitudes of up to @xmath139 which are rapidly oscillatory in angular dependence as @xmath140/\sqrt{l\sin\theta}$ ] ( for @xmath141 ) . now our task is to extract useful pieces of information from eq . ( [ fsi ] ) with the high - energy elastic @xmath2-matrix of eq . ( [ pi ] ) or eq . ( [ k ] ) . let us first count how many dynamical quantities are involved in the most general @xmath142 fsi relation , eq . ( [ fsi ] ) . the unitary and symmetric @xmath2 matrix contains @xmath143 independent parameters ; @xmath144 eigenphase shifts and @xmath145 rotation angles of @xmath69 . to solve for @xmath74 s of observable channels in eq . ( [ fsi ] ) , therefore , we must feed the @xmath143 dynamical parameters of strong interaction . this is not sufficient to determine @xmath74 uniquely . although the fsi relation eq . ( [ fsi ] ) may look as if it introduced @xmath146 constraints through the real and imaginary parts , only a half of them , namely @xmath144 of them are actually independent . gives a constraint only on the phase of @xmath147 , not its magnitude . in the case of @xmath144 channels , something similar happens : the phases of @xmath40 for _ eigenchannels _ are determined when @xmath77 are completely specified , but their magnitudes @xmath148 are not . ] we must provide the relative magnitudes of @xmath40 or @xmath74 as an additional input in order to determine the fsi phases uniquely . the magnitude of a decay amplitude is determined primarily by weak interaction , _ i.e. , _ the property of decay operators . knowledge of strong interaction alone can never determine multichannel fsi phases . we need to know interplay of strong and weak interactions . nobody is capable of tackling this problem for a general value @xmath144 . we will therefore be content with studying the fsi relation first in the simple manageable case of @xmath149 and then searching a sensible approximation in more complicated and realistic cases . although the two - channel problem is the next to the simplest , there has been no serious attempt to study this case in the past , probably with a good reason as we see below . although it may not look much relevant to the @xmath0 decay of the real world , we have a chance to see through general characteristics of coupled channel effects . for instance , how is the fsi phase of @xmath45 channel affected by the @xmath46 channel ? if one of the charm - anticharm channels such as @xmath150 strongly couples to the @xmath45 channel , how does this channel affect the fsi of the @xmath45 channel ? is an order of magnitude larger than that to @xmath45 . the @xmath151 channel has a huge branching fraction because of the robust @xmath152 transition . ] while simple - minded perturbative calculations have been undertaken in the past , we would like to study these questions systematically with the two - channel toy model that incorporates unitarity . we can write the general @xmath7-invariant @xmath2-matrix of @xmath153 with three parameter ( @xmath154 for @xmath149 ) in the form of @xmath155 where @xmath156 and @xmath157 are abbreviated as @xmath158 and @xmath159 in the second line . substituting this @xmath2-matrix in the fsi relation eq . ( [ fsi ] ) , we obtain the constraints on the real and imaginary parts or the magnitudes and phases of the decay amplitudes defined by sign by the symmetry condition @xmath67 on the @xmath2-matrix . ] @xmath160 where the phases @xmath161 are the fsi phases ( the strong phases ) of channel 1 and 2 . we have in mind @xmath162 for @xmath45 and @xmath163 for either @xmath46 or @xmath150 of @xmath49 . the constraining equation of eq . ( [ fsi ] ) can be written out for the real and imaginary parts as @xmath164 where @xmath165 and @xmath166 are the @xmath153 matrices . to be explicit , @xmath167 as we have pointed out above , eq . ( [ relation2 ] ) contains only two independent constraints , not four . indeed , one can show that two eigenvalues of the @xmath168 matrix in the right - hand side are unity and generate no constraint . the two remaining eigenvalues are @xmath169 and generate constraints . we now feed our dynamical input eq . ( [ epsilon ] ) in the slightly different notation ; @xmath170 where @xmath121 is real but not necessarily very small in magnitude . for our discussion later we choose @xmath121 to be @xmath171 , as suggested by the pomeron dominance in elastic scattering . it is easy to include the nonleading regge contributions and relax the condition @xmath172 . fixing @xmath121 amounts to setting two parameters in @xmath2-matrix of eq . ( [ two ] ) so that we are left with one parameter out of three . when we fix the @xmath173 component as in eq . ( [ s11 ] ) , it is more convenient to parametrize the @xmath2-matrix in the form @xmath174 where the angle @xmath175 is the remaining single parameter of the @xmath2-matrix . it is related to two eigenphase shifts @xmath176 defined in eq . ( [ two ] ) by @xmath177 note that once @xmath173 is given , magnitude of the channel coupling @xmath178 in eq . ( [ two ] is fixed by unitarity and no longer a free parameter . let us first study the case where the partial - wave amplitude @xmath180 of rescattering is stronger in the second channel than in the first channel , namely , _ i.e. , _ @xmath181 . this may serve as a toy - model of @xmath45 and @xmath151 ; branching is larger than that of @xmath182 . the @xmath183 and @xmath184 channels can not make @xmath79 . ] the @xmath151 scattering of @xmath49 is presumably strong because of the near - threshold enhancement and/or broad excited charmonium resonances . for illustration we consider the extreme case that the rescattering in the second channel is maximally strong relative to that in the first channel . this is realized by choosing @xmath185 in eq . ( [ two ] ) . the symmetric unitary @xmath2-matrix takes the form of @xmath186 in terms of partial - wave amplitudes , this gives @xmath187 and @xmath188 for the first and second channels , respectively , with @xmath189 . by substituting this @xmath2-matrix in the fsi relations , we obtain ( see eq . ( [ delta ] ) ) @xmath190 the real and the imaginary parts of @xmath147 are still independent of each other . the phases @xmath161 can not be determined uniquely even after the @xmath2-matrix is fully specified . but the phase @xmath191 is related to @xmath192 by @xmath193 even when @xmath175 of @xmath194 is equal to @xmath195 ( normally called as `` resonant '' ) , @xmath192 is not necessarily equal to @xmath195 . although this may look puzzling at the first sight , it is not . to determine @xmath161 uniquely , we need to feed one more piece of information . for instance , if the value of the ratio @xmath196 is supplied , we can determine @xmath191 and @xmath192 individually . by eliminating @xmath192 from eq . ( [ a ] ) we obtain the relation that determines @xmath191 in terms of @xmath196 and @xmath121 : @xmath197 . \label{delta1}\ ] ] the ratio @xmath196 contains information of weak interactions . in multichannel decay , weak interaction plays a very important role in determining the fsi phases . there is one shortcoming of this two - channel toy model : as one sees in eq . ( [ delta1 ] ) , the ratio @xmath196 must lie in the range of @xmath198 for this model to be applicable . at the lower boundary of @xmath196 , it happens that @xmath199 , while @xmath200 at the upper boundary of @xmath196 . when @xmath196 is in between , both @xmath191 and @xmath192 take nonzero values even though all elements of the @xmath2-matrix are real . this is an important point to be emphasized . the phase of @xmath147 can arise from the process @xmath201 through the intermediate state @xmath202 . some may wonder why the ratio @xmath196 is constrained in the two - channel toy model . with channel coupling present , one channel feeds the other by fsi to the direction to equalize magnitudes of @xmath203 and @xmath204 . the fsi not only generates phases for @xmath147 and @xmath205 but also alters their magnitudes . highly asymmetric @xmath203 and @xmath204 are incompatible with the fsi connecting the two channels unless @xmath206 , _ i.e. , _ @xmath207 . when channel 2 is @xmath46 , we expect that the elastic @xmath46 scattering is asymptotic at @xmath209gev and very similar to the elastic @xmath45 scattering : @xmath210 in this case unitarity and symmetry require that the off - diagonal element @xmath211 should be purely imaginary ( @xmath212 in eq . ( [ two ] ) ) : @xmath213 this is the case that was studied by donoghue _ et al_@xcite . the fsi relation eq . ( [ fsi ] ) leads us to @xmath214 here again the fsi phases are uniquely determined only after the ratio @xmath215 is given . the phase @xmath191 is expressed in terms of @xmath196 and @xmath121 by the same relation as eq . ( [ delta1 ] ) . the ratio @xmath196 lies in the same range as eq . ( [ range ] ) . in contrast to the case of the maximum @xmath216 , the amplitude @xmath205 and @xmath147 are @xmath217 out of phase , one real and the other purely imaginary , at the lower and upper boundaries of the range for @xmath196 . once we have explored the two extreme cases above , it is not difficult to find the general solution for an arbitrary value of @xmath175 in eq . ( [ two ] ) . by rewriting the fsi relation in @xmath218 , we can reduce it to the case of @xmath219 above . we obtain the solution for general @xmath175 as @xmath220 the first and second lines of eq . ( [ general ] ) reduce to those of the previous cases ; @xmath221 and @xmath222 as @xmath223 and @xmath224 , respectively . the third line relates the fsi phases @xmath191 and @xmath192 to each other with parameter @xmath121 and @xmath175 . the ratio @xmath196 is related to @xmath191 and @xmath121 exactly in the same way as in eq . ( [ delta1 ] ) . consequently @xmath196 is also restricted to the same range , eq . ( [ range ] ) . we can actually find the exact solution even when we include the very small imaginary part of @xmath225 in @xmath173 due to the low - ranking regge trajectories in elastic scattering . in this most general case it is convenient to write the @xmath2-matrix in the form @xmath226 unitarity and symmetry fixes the phase of @xmath211 as shown above once those of @xmath173 and @xmath227 are given . for this reason one earlier literature@xcite assigned the angles @xmath228 and @xmath229 to the `` initial - state '' and `` final - state '' interactions of scattering , hinting that the phases of decay amplitudes @xmath147 and @xmath205 acquire @xmath228 and @xmath229 , respectively , by fsi . unfortunately this interpretation was wrong . as we see below , the phases of @xmath230 are @xmath231 _ plus _ additional contributions that depend on mixing of the channels and weak interaction . it should also be pointed out that the simple phase relation like @xmath232 holds only for the @xmath2-matrix of @xmath153 , not of more than two channels . to solve the fsi relation with eq . ( [ sg ] ) we factor out the `` elastic phases '' @xmath228 and @xmath229 of the diagonal @xmath2-matrix elements @xmath233 ( _ not _ of the partial - wave amplitudes @xmath116 ) from the decay amplitudes by introducing @xmath234 by @xmath235 ( @xmath236 ) . then the fsi relation @xmath237 reduces to the form identical to eq . ( [ s2 ] ) . therefore the solution for @xmath230 can be immediately written as eq . ( [ solution2 ] ) : @xmath238 where @xmath239 and @xmath240 are real the first and the second line of eq . ( [ solutiong ] ) requires @xmath196 to remain in the same range as in the previous two special cases . ( [ range ] ) ) . the relation of eq . ( [ delta1 ] ) is trivially modified as @xmath241 . \label{deltag}\ ] ] this relation is the bottom line of the general two - channel toy model : the total fsi phase @xmath191 of @xmath147 is sum of the rescattering phase @xmath228 of @xmath242 , not of the elastic partial - wave amplitude ( @xmath243 ) , plus the the phase due to rescattering through the second channel . it can not be over - emphasized that in the presence of inelasticity the phase of @xmath242 is very different from the phase of the partial - wave amplitude . for pomeron - dominated scattering , for instance , @xmath244 since @xmath116 is purely imaginary for all @xmath124 , but @xmath245 or @xmath246 since @xmath173 is real and positive for @xmath247 . the phases of @xmath248 and @xmath116 would be equal only in the elastic limit where @xmath249 and @xmath250 . it makes no sense whatsoever even as an approximation to equate the fsi phase with the phase of the elastic partial - wave amplitude @xmath251 in @xmath0 decay . it is worth mentioning here that the solutions of the two - channel problem , eqs . ( [ solutiong ] ) and ( [ deltag ] ) , apply to the @xmath252 decay into @xmath253 and @xmath254 . the @xmath253 and @xmath255 yields add up to over 99% of the observed nonleptonic final states . in this case the lower ranking regge trajectories contribute more to the diagonal @xmath2-matrix elements than in the light hadrons channels of @xmath0 decay . while the two - channel toy model casts light on many important issues , it has one undesirable feature that the ratio @xmath196 is restricted within the rather narrow range set by eq . ( [ range ] ) . numerically , @xmath256 this constraint limits applicability of the two - channel toy - model to @xmath0 decay modes . we must extend to more than two channels to study @xmath0 decay . however , as the coupled channels increases , the number of dynamical unknowns quickly increases in the fsi relation . in order to keep our problem manageable , we must introduce some approximation that keeps mathematical complexity under control . going back to our fundamental equation eq . ( [ fsi ] ) , we consider the situation where one inelastic channel makes a dominant feed back to the elastic channel ( channel 1 ) and all other inelastic channels are either unimportant individually or largely cancel among them . we are specifically interested in the case , @xmath257 such that @xmath258 in this case we can truncate the sum over the inelastic channels at @xmath163 ; @xmath259 the fsi relation for the channel 2 reads @xmath260 since unlike the off - diagonal @xmath261 the diagonal s - matrix element @xmath262 contains the term unity , we expect that @xmath227 is @xmath263 or a substantial fraction of it unless the term @xmath264 is cancelled accidentally by @xmath265 with high accuracy . in comparison , @xmath211 represents a small leakage into a dominant inelastic channel in the present case . therefore @xmath266 where the both factors in the right - hand side are small . in other words , the effect of the single elastic channel back on a robust inelastic channel @xmath267 is negligible when the coupling @xmath268 between them is feeble . therefore , magnitude of the first term @xmath269 in the right - hand side of eq . ( [ t2 ] ) is much smaller than that of @xmath270 by the assumptions made in eqs . ( [ assumption ] ) and ( [ assumption2 ] ) . therefore , we may drop the first term in eq . ( [ t2 ] ) . then no information of the channel 1 is needed to solve eq . ( [ t2 ] ) for @xmath205 . therefore we solve only eq . ( [ t1 ] ) and obtain a relation between @xmath147 and @xmath205 . solving eq . ( [ t2 ] ) for @xmath205 may be hard . in the numerical exercise later we do not attempt to compute for @xmath205 theoretically in terms of other inelastic channels , but resort to experiment for information of @xmath205 . the @xmath2-matrix now need not satisfy unitarity in the subsector of channel 1 and 2 . it can be written in general as @xmath271 where two real parameters @xmath272 and @xmath273 have been introduced to describe inelasticity of scattering . when the channel 2 is also a two - body light - hadron channel , the value of @xmath273 is @xmath274 and @xmath275 . in @xmath0 decay the branching fractions to two - body light - mesons are much smaller than those to charmed meson pairs by the property of weak interaction . the decay @xmath1 through @xmath276 is a typical example since @xmath277 is much smaller than @xmath225 but @xmath204 is much larger than @xmath203 . it has been speculated that the presence of the @xmath276 channel may generate a large fsi phase for @xmath278@xcite . we will examine this possibility later . the fsi relation can be solved for @xmath230 even in the presence of the additional parameter @xmath272 by the rephasing technique that we have used earlier . when we express @xmath147 and @xmath205 in terms of @xmath279 and @xmath280 instead of @xmath239 and @xmath240 , the solution of eq . ( [ fsi ] ) with eq . ( [ s ] ) is : @xmath281 + \frac{i}{2-\epsilon } [ a_2\cos(\chi_2+\chi_{\kappa})+b_2\sin(\chi_2+\chi_{\kappa})]\biggr ) \nonumber \\ a_2 & = & a_2 + ib_2 , \label{solutions } \end{aligned}\ ] ] as we have pointed out , the parameter @xmath273 is not needed to express the relation between @xmath147 and @xmath205 in the truncated approximation . although @xmath282 enters @xmath147 , we can express the phase @xmath191 without @xmath282 by using experimental knowledge of @xmath196 . rewriting eq . ( [ solutions ] ) with the magnitudes and phases , we obtain a simple generalization of the previous relation , @xmath283 the right - hand side of eq . ( [ solutiont ] ) gives the contribution of the channel coupling that is to be added to the elastic contribution @xmath228 . the ratio @xmath196 is now bounded as @xmath284 if @xmath272 is small , that is , if the leakage into the channel 2 is small , large values can be accommodated for @xmath196 . therefore the truncated model is applicable to more general situations than the two - channel toy model that we have discussed . the ratio @xmath196 contains information of weak interaction . it is amusing to see in eq . ( [ solutiont ] ) that the strong phase @xmath191 coincides with the small `` elastic phase '' @xmath228 of @xmath242 when @xmath205 takes the smallest value , @xmath285 , in the allowed range of eq . ( [ rangeg ] ) . the channel coupling effect @xmath286 is the strongest when @xmath196 takes the upper limit @xmath287 in eq . ( [ rangeg ] ) . if we proceed further and include two prominent inelastic channels , the fsi relation for @xmath147 is : @xmath288 if we continue along this line and incorporate more inelastic channels , the fsi equation for channel 1 turns into @xmath289 the first term in the right - hand side can be viewed as counteraction of elastic rescattering . it affects not only on the phase of @xmath147 but also causes long - distance enhancement or suppression on magnitude of @xmath147 depending on the force in the elastic channel . with @xmath290 , eq . ( [ fsig ] ) reveals an interesting general feature of the multichannel fsi . separating the real and imaginary parts of eq . ( [ fsig ] ) , we have in the case of real @xmath173 @xmath291 eq . ( [ fsiab ] ) shows that elastic rescattering @xmath173 enhances the inelastic rescattering effect by @xmath292 @xmath293 for the real part @xmath239 of @xmath147 and and suppresses it for the imaginary part @xmath240 by @xmath294 ( @xmath295 ) . this characteristic depends only on @xmath173 , which we belive we know fairly accurately . despite its simplicity ( [ fsiab ] ) contains useful information . for instance , _ if _ the transition to and from channel 1 can be described by the born terms of @xmath91- and @xmath296-channel exchanges , the off - diagonal partial - wave amplitudes are real so that the off - diagonal @xmath2-matrix @xmath297 @xmath298 are purely imaginary . therefore the real decay amplitudes @xmath71 ( @xmath299 ) of the inelastic channels contribute to the imaginary part @xmath240 of channel 1 in this case . even if there is no resonance in the process , the phase @xmath191 can be very large in this way . if furthermore the `` inelastic '' decay amplitudes @xmath71 ( @xmath299 ) happen to be all real , the phase @xmath191 would be @xmath300 . this is not surprising : in the language of dispersion theory the on - shell inelastic intermediate states generate an absorptive part for @xmath147 that turns out to be purely imaginary in this situation . ( [ fsiab ] ) involves one dynamical input : the elastic scattering amplitudes of light mesons are almost purely imaginary ( pomeron - dominated ) . some numerical exercise is called for to show relevance of our endeavor to the @xmath0 decay of the real world . contrary to the initial optimism that had prevailed before @xmath0 physics experiment started , analysis of experiment seems to suggest that the fsi phases of some two - body decay amplitudes appear to be much larger than what we expected in the original short - distance picture@xcite . the word @xmath278 puzzle has been coined for the unexpectedly large tree - contribution and/or fsi phases in @xmath278 modes . as the perturbative technique has become more sophisticated , people have come to agree that emission and absorption of soft and collinear quarks and gluons plays an important role in many decay modes@xcite . such soft constituents contribute to the fsi phases involving long - distance physics . while one can parametrizes such contributions in the soft - collinear theory , one can not evaluate them numerically in perturbative argument . our @xmath2-matrix approach also has its own drawback : while elastic scattering of two light hadrons at energy @xmath53 has been well understood , we know less about their inelastic scattering . nonetheless we would like to show here that our method may be useful in some of @xmath0 decay modes . it is believed that the decay @xmath1 occurs primarily with the penguin interaction @xmath301 $ ] . in the penguin process the coupling to channels such as @xmath302 need to be studied as a source of the strong phase of the @xmath278 amplitude . however , it has been argued that @xmath278 can be produced indirectly with the tree interaction @xmath303 $ ] as well through the charmed - meson - pair states such as @xmath304 and @xmath276@xcite . the @xmath278 amplitude of this process acquires a strong phase different from the direct penguin amplitude . what can our approach say about this problem ? the branching fractions have been measured for the following two - body channels that couple to @xmath278 in @xmath305 decay@xcite : @xmath306 all of them can make @xmath307 , the spin - parity of @xmath278 . the states @xmath278 and @xmath302 can be either in @xmath308 or @xmath309 while all other modes are only in @xmath308 . from the regge phenomenology on elastic @xmath278 scattering , we have already estimated @xmath121 for the pomeron contribution in section iii . the @xmath82 and @xmath83 trajectories with exchange degeneracy allow us to estimate the small imaginary part of @xmath173 . by adding the @xmath82 and @xmath83 regge contributions , we obtain numerically @xmath310 the @xmath311 contributions to the scattering amplitudes @xmath7 are dominantly imaginary for @xmath308 and real for @xmath309 , as we expect from @xmath52 to @xmath91-channel duality . therefore they generate a larger imaginary part for @xmath312 in the @xmath309 channel . this is very different from our intuitive picture in the single - channel case . though we do not attach errors to our estimate , errors as large as factor two are possible for the phases in eq . ( [ nonleading ] ) . estimate of @xmath211 is less reliable than @xmath173 owing to very indirect experimental information and to larger theoretical uncertainties . we eliminate the @xmath313 channel from our consideration since the regge residue or the coupling of @xmath314 to @xmath315 is highly suppressed ( `` ozi - forbidden '' ) . in contrast @xmath302 can couple to @xmath278 without such suppression . while a loose upper bound has been set on @xmath302 experimentally ( see eq . ( [ br ] ) ) , we believe that the @xmath302 channel is more important than the @xmath313 channel . we may use the regge phenomenology to estimate @xmath211 for the @xmath302 state of longitudinal polarizations in the final state . for the scattering @xmath316 , the leading regge poles are @xmath317 and @xmath279 which are exchange degenerate . trajectory generates a backward peak is generated , but it is less important than the forward peak . we neglect the backward peak contribution to @xmath211 . ] the couplings of @xmath318 and @xmath319 are known on the mass shells of @xmath317 and @xmath279 , respectively , from low - energy spectroscopy . the corresponding @xmath320 couplings of @xmath317 and @xmath279 are obtained by an su(3 ) rotation . but we need to extrapolate them to the off - shell @xmath82 and @xmath279 to relate them to the regge residues . this extrapolation is a major source of uncertainty . if we ignore the extrapolation , they are at the same level in magnitude as the nonleading regge contributions in @xmath173 ; @xmath321 using the ratio @xmath196 computed with the measured branching fraction and the upper bound listed in eq . ( [ br ] ) , we reach the crude estimate , @xmath322 despite large uncertainty of these numbers we may conclude with eq . ( [ rhokstar ] ) that the inelastic term @xmath323 of the @xmath302 channel in the @xmath278 mode is not significant relative to the elastic term @xmath324 . it is certainly not a major source of the strong phase for the @xmath278 amplitude in the standard penguin decay ( not through @xmath325 ) . the same line of estimate suggests that the @xmath326 state is neither important to the fsi phase of @xmath278 . some may wonder about one - pion exchange in @xmath327 . the reggeized pion exchange amplitude for @xmath328 is down by another power of @xmath329 relative to those of @xmath317 and @xmath279 exchanges . the small denominator of the pion propagator does not enhance the amplitude near forward direction because the lorentz structure requires the amplitude near the pion pole to be proportional to @xmath330 the both factors in the numerator vanish up to @xmath331 for longitudinally polarized @xmath332 and @xmath82 in the forward direction ( @xmath333 ) . they eliminate a forward peak from the pion pole . the regge theory predicts that the rest of the scattering amplitude falls sharply ( @xmath334 ) off the forward direction . therefore the pion exchange can be dismissed . then we feel safe to conclude with eq . ( [ estimate ] ) that the coupling of @xmath278 to the @xmath302 channel is not important in determining the fsi phases for the @xmath278 modes . the @xmath335 and @xmath336 channels can couple to the @xmath278 channel of @xmath308 through the @xmath279 regge exchange . although this contribution has the same @xmath53 dependence as the @xmath82 and @xmath83 regge exchanges , the @xmath279 regge residues with @xmath337 are most likely smaller than the residue with @xmath315 : we can estimate from the @xmath279 decay branching that , after the @xmath338-wave phase space factor is separated , the on - shell @xmath279 couplings to @xmath339 and @xmath340 are about factor 20 smaller than those to @xmath315 . therefore neither @xmath335 nor @xmath336 compete with @xmath302 in the final state . therefore we may leave out @xmath335 and @xmath336 from our consideration . the contributions from nonresonant three - body final states are harder to estimate since computation of @xmath211 is next to impossible . there are many nonresonant multiparticle channels with relatively minor branching fractions . if rescattering of @xmath278 to multibody channels is quasi - diffractive with no quantum number exchange , the pomeron can contribute . in such scattering the final states consist of two lumps of relatively small invariant masses that carry the same flavors as @xmath44 and @xmath341 . they are likely to end up in two - body states of highly excited meson states , for instance , @xmath342 . while this is a possibility , none of such modes have been positively identified so far in measurement . genuine nonresonant three - body channels are probably not a major source of the fsi phases , unless their contributions add up by constructive interference to a large value . in fact , it is conceivable that they sum up in random phases into relatively a small number@xcite . that is one motivation when we have introduced the truncated approximation . our tentative conclusion on the @xmath278 amplitudes ( @xmath343 ) of the light - quark penguin decay operators is that the fsi phase produced by coupling to the inelastic channels is insignificant . analysis of the @xmath278 amplitudes in search of the weak phases was started more than ten years ago . with little knowledge of the strong phases , however , the analysis could be carried out only by assuming that the long - distance strong phases be negligible@xcite . the charmed meson - pair channels are very different . since they are the ckm dominant tree - decay final states , their branching fractions are a few orders of magnitude larger than that of the penguin - dominated decay . they can annihilate into @xmath278 . in the quark picture this process can be viewed as the on - shell contribution of the @xmath325 penguin to @xmath278 . some call this process as `` charming penguin''@xcite . among the charmed meson pairs , the most prominent decay channel of @xmath79 is @xmath344 . its amplitude can be estimated with eq . ( [ br ] ) as @xmath345 one important question here is how much of the observed total @xmath346 is the `` charming penguin '' contribution . since we expect that long - distance physics enters the on - shell process of charmed - meson pairs at energy @xmath53 , it is not easy to evaluate its magnitude . some argue that it can be very large@xcite . however , a counter argument was made to advocate the short - distance argument@xcite . theorists have not come to consensus on magnitude of this contribution . therefore we insert one fudge factor @xmath347 here for this contribution to @xmath147 as @xmath348 where @xmath347 , the fraction of the @xmath325 contribution , may be as large as a half or even more@xcite . it must be settled by theory rather than by experiment . with this fudge factor ( [ solutiont ] ) turns into @xmath349 where @xmath350 . even if the spill - over of @xmath344 into @xmath278 is as tiny as one tenth of percent ( @xmath351 ) , the @xmath344 channel may control the fsi phase of the @xmath278 amplitude that comes through @xmath325 . to proceed further , we must look into the transition @xmath352 . ( fig.4 ) application of the regge theory is questionable to the charm - pair channels since the charmed meson masses are around 2 gev and the total energy is a little above 5 gev . departing from the regge theory , let us make the born approximation in @xmath91-channel exchange . the exchanged mesons are the charmed mesons @xmath353 , @xmath354 . for @xmath353-exchange , we know the @xmath355 coupling from the decay @xmath356 and can compute the @xmath357 coupling by an su(3 ) rotation of the @xmath355 coupling . the differential cross section rises toward the forward direction when the @xmath353 meson is exchanged . for @xmath358 exchange , we can deduce the @xmath359 coupling from the @xmath355 coupling with the heavy quark spin symmetry and rotate it into the @xmath360 coupling by su(3 ) . in contrast to the @xmath353 exchange , the lorentz structure of the vertex @xmath361 of @xmath358 exchange cancels a forward peak that would be otherwise generated by the @xmath358 propagator . for this reason the @xmath358-exchange is less important . when we compute the @xmath353 contribution in the feynman diagram with the on - shell couplings , we obtain @xmath362 . but this is obviously a nonsense . the reason is that we have ignored the form - factor damping effect of the exchanged off - shell @xmath353 . for an order - of - magnitude estimate we may multiply a factor of @xmath363 as a form - factor effect where @xmath364 is the binding scale of the charmed mesons then our estimate goes down by more than an order of magnitude from @xmath362 to @xmath365 . this latter value is probably closer to reality . it is roughly in line with the rule of thumb ; in the quark picture a pair creation probability of @xmath325 is suppressed by about @xmath366 relative to light - quark pair creation of @xmath367 , where the quark masses are the constituent masses . this rule works roughly for @xmath368 and @xmath325 production in high - energy collision . if we use this rule , we obtain @xmath369 from eq . ( [ rhokstar ] ) with @xmath370 . therefore we choose as our best guess @xmath371 as we have noted earlier , @xmath211 is purely imaginary for real @xmath125 . the number of eq . ( [ dds ] ) is obviously an order - of - magnitude estimate at best . with @xmath372 and tentatively @xmath373 , a crude central value of our estimate for @xmath374 is @xmath375= 0.04 . \label{dds2}\ ] ] we now substitute all these numbers in eq . ( [ solutiont ] ) of the truncated approximation . we take the number of eq . ( [ dds2 ] ) as a ballpark figure and sweep the value of @xmath376 by a factor two across this value to see what fsi phase can be generated for @xmath278 of @xmath308 by the channel coupling to the @xmath344 channel . the result is plotted in fig . the value of @xmath377 is constrained between 0.5 and 1.5 by eq . ( [ solutiont ] ) and sweeps in the region between two vertical broken lines . within the uncertainty of @xmath374 , the fsi phase @xmath191 of the @xmath278 amplitude through @xmath325 can be any value between @xmath378 and @xmath379 . ( [ solutiont ] ) does not determine the sign of @xmath191 since it does not contain full information of @xmath205 . because of the large uncertainty of @xmath374 , we can not constrain @xmath191 meaningfully at present . even @xmath380 is not reliably excluded . keeping the uncertainty of our estimate in mind , we should state here only that the channel coupling to @xmath276 is capable of producing a very large strong phase for the charming penguin @xmath278 amplitude , in particular in the case that the `` charming penguin amplitude '' is a sizable fraction of the total amplitude . quantitatively reliable computation of the fsi phase will be possible after we have obtained a better theoretical estimate of @xmath211 for @xmath352 as well as magnitude of the decay amplitude through @xmath325 . until then we must not set this strong phase to zero but leave it as an unknown parameter to be determined by fit to experimental data . if such experimental fit clearly requires a large fsi angle for the @xmath278 mode of @xmath308 but not of @xmath309 , we shall be able to assert that the on - shell @xmath325 intermediate state plays an important role in the decay @xmath1 . the statement above can be made for many other two - body light - hadron modes . the @xmath45 mode couples to @xmath46 whose branching fraction is nearly an order of magnitude larger than that of @xmath45 . but perturbative calculation of the @xmath381 transition@xcite by the born diagrams without off - shell damping can easily overestimate it . with an estimate of @xmath211 along the same line as for @xmath382 , the right - hand side of eq . ( [ solutiont ] ) comes out to be negative in the case of @xmath383 , it implies that the branching fraction of @xmath46 is not large enough to affect the fsi phase for the @xmath45 amplitudes of the tree decay @xmath384 ( @xmath385 = light quarks ) . the @xmath45 of @xmath49 can be fed also by the charmed hadron channel @xmath151 whose branching fraction is two orders of magnitude larger than that of @xmath45 . for the @xmath45 amplitude through @xmath386 , coupling to @xmath151 is the most important source of the strong phase . to obtain the strong phases and to compare with experiment , we again need to know about relative importance of the two classes of decay ; @xmath387 and @xmath388 for @xmath389 . in contrast , the tree - dominated ckm - favored decay modes such as @xmath390 have no wide open inelastic channels . transition to the @xmath391 channel is as insignificant as the transition to @xmath328 in the @xmath278 mode . since the @xmath391 decay branching fraction is comparable to that of @xmath392 and @xmath211 is much smaller than @xmath173 ( cf eqs . ( [ rhokstar ] ) and ( [ estimate ] ) ) , of its elastic scattering , we expect that @xmath393 and therefore the channel coupling contribution is unimportant . it means that the fsi phases are small in these modes and that simple short - distance calculation of the phases can produce an answer not far from reality . one of our purposes is to solve the most general two - channel model exactly and to clarify the mechanism of generating strong phases in this toy model . we have seen above that existence of competing channels completely changes the strong phase from that of the `` elastic rescattering phase '' in all cases of two - channel @xmath2-matrix . unitarity plays an important role here . the other purpose is to introduce a feasible approximation scheme which may be applicable to cases of special interest in @xmath0 decay . we have applied this method to the decay @xmath1 and have made semiquantitative analysis . but its outcome is not numerically satisfying because of limitation in available knowledge about off - diagonal scattering . we have extended our analysis to the general multichannel case and come to one simple interesting observation : though it may sound odd , inelastic scattering tend to enhance the strong phase of the elastic channel most when inelastic amplitudes are real ( _ i.e. , _ @xmath394 = imaginary ) rather than imaginary , if the strong phases of the inelastic decay amplitudes are small . ( [ fsiab ] ) . ) it is a big challenge to go beyond the two - channel approximation . as the number of channels @xmath144 increases , the number of parameters in the fsi relation increases as @xmath143 for strong interaction . in addition we need piece of information from weak interaction . in the truncated approximation we have kept only a single dominant one among the inelastic channels . our assumption is that all other inelastic channels are less important or else sum up in random phases to become numerically insignificant . if the higher inelastic channels do not sum randomly , there must be some underlying dynamical reason for it . in such a case some approach orthogonal to ours may have advantage . for instance , it is an approach based on quarks and gluons instead of hadrons . the major source of the strong ( fsi ) phase in the multichannel case is the transition to inelastic channels . consequently accurate computation of the fsi phases depends on knowledge of the transition @xmath2-matrix at energy @xmath53 between a channel of our interest and dominant inelastic channels . that is , we need to know dynamics of long and intermediate distances at this energy . the present author is of the opinion that quantitatively we have a little better handle on hadron physics numerically than on the soft and collinear quarks and gluons in this territory . but opinions probably divide among physicists of different generations . the author is grateful to c. w. bauer for updating him in the theoretical status of the charming penguin . this work was supported by the director , office of science , office of high energy and nuclear physics , division of high energy physics , of the u.s . department of energy under contract de ac0205ch11231 . expansion of the observable state @xmath395 in the @xmath2-matrix eigenstates @xmath60 is defined by @xmath396 at this stage the matrix @xmath69 is assumed to be only unitary , not necessarily orthogonal . make time - reversal on the scattering amplitude from eigenstate @xmath61 to observable state @xmath64 in our phase convention of states under time reversal : @xmath397 operating @xmath398 on eq . ( [ expansion1 ] ) , we obtain @xmath399 substitution of eqs . ( [ expansion1 ] ) and ( [ expansion2 ] ) in eq . ( [ hybridt ] ) gives us @xmath400 which proves that @xmath69 is orthogonal : @xmath401 k. watson , _ phys . rev . _ * 95 * , 228 ( 1954 ) ) . + k. aizu , proc . theor . phys . 1953 ( kyoto - tokyo science council , 1954 ) , p.200 . e. fermi , nuovo cim . * 2 * , 17 ( 1955 ) . see also _ elementary particles _ , ( yale university press , new haven , ct , 1951 ) , pp . 58 - 64 . for a critical review of such earlier literature on the fsi phases , see m. suzuki , _ proceedings of the third international conference on b physics and cp violation _ ( taipei , taiwan , december 1999 ) , edited by h. -y . cheng and w. -s . hou , world scientific , pp . 259 - 265 , and literature therein . donoghue , e. golowich , a.a . petrov , and j.m . soares , phys . lett . * 77 * , 2178 ( 1996 ) . r. n. cahn and m. suzuki , hep - ph/9708208 . l. l. wolfenstein and m. suzuki , phye . d * 60 * , 074019 ( 1999 ) . l. wolfenstein , private communications . m. chiuchini , e. franco , g. martinelli , and l. silvestrini , nucl . phys . * b501 * , 271 ( 1997 ) . d. b. collins and e. j. squires , _ regge poles in particle physics _ ( springer - verlag , berlin , 1968 ) . w .- yao et al . , _ review of particle physics _ , j. phys . g * 33 * , 1 ( 2006 ) . e. m. levin and l. l. frankfurt , jetp lett . * 2 * , 65 ( 1965 ) . d. b. collins and e. j. squires , ibid , pp . 193 - 250 . r. j. n. phillips and w. rarita , phys . rev . * 139 * , b1336 ( 1965 ) . et al . , _ cern / hera 70 - 2 ; e. flaminio _ et al . , _ , cern / hera 70 - 3 . e. bracci _ et al . , _ high - energy reactions analysis group , cern / hera-72 - 1 . j. gillespie , final state interactions ( holden - day , san francisco , ca , 1964 ) . a. b. kaidalov and m. i. vysotsky , hep - ph/0704.0404 c. w. bauer , d. pirjol , i. z. rothstein , and i. w. stewart , phys . d * 70 * , 054015 ( 2004 ) ; c. w. bauer , i. z. rothstein , and i. w. stewart , phys . d * 74 * , 034010 ( 2006 ) . m. beneke , g. buchalla , m. neubert , and c. t. sachrajda , phys . d * 72 * , 098501 ( 2005 ) . m. beneke , g. buchalla , m. neubert , and c. t. sachrajda , phys . * 83 * , 1914 ( 1999 ) . for a short review , see for instance s. fleming , hep - ph/0601257 and th . feldmann , hep - ph/0610192 and references therein . j. chay , h - n . li , and s. mishima , arxiv:0711.2953 and references therein . w .- yao et al . , _ review of particle physics _ , j. phys . g * 33 * , 1 ( 2006 ) and 2007 partial update for 2008 . falk , a.l . kagan , y. nir , and a.a . petrov , phys . d * 57 * , 4290 ( 1998 ) ; a.a . petrov , hep - ph/9903366 .

we discuss the problem of two particles interacting via short - range interactions within a harmonic - oscillator trap . the interactions are organized according to their number of derivatives and defined in truncated model spaces made from a bound - state basis . leading - order ( lo ) interactions are iterated to all orders , while corrections are treated in perturbation theory . we show explicitly that next - to - lo and next - to - next - to - lo interactions improve convergence as the model space increases . in the large - model - space limit we regain results from a pseudopotential . arbitrary scattering lengths are considered , as well as a generalization to include the non - vanishing range of the interaction .

1001.5071

much experimental progress has recently been achieved in the area of ultra - cold atoms . it is now possible @xcite to make systems of atoms confined in optical lattices formed by laser beams and at the same time control the strength of the two - body interaction with magnetic fields . the atoms are cooled down to extremely low temperatures and , in the limit of low tunneling , each lattice site may be regarded as a harmonic - oscillator ( ho ) well , which is independent from the others and contains only a few atoms . using a feshbach resonance , the two - body interaction can be fine - tuned so that the @xmath0-wave scattering length @xmath1 is made much larger than the range @xmath2 of the interaction . by doing so , the theoretical problem of a system of trapped particles interacting via short - range forces can be realized experimentally . such atomic systems present remarkable similarities to some nuclear physics systems . indeed , the two - nucleon scattering length is much larger than the range of the nuclear force , set by the pion mass . in such a situation , the physical properties of few - body systems are universal , that is , they depend mainly on @xmath1 , not on the details of the two - body interaction . at sufficiently low energies , an effective field theory ( eft ) has been formulated which turns this separation of scales into an expansion in powers of @xmath3 @xcite . except for isospin , which does play a role in the relative relevance of few - body forces , the version of this eft used in nuclear physics @xcite is formally indistinguishable from the theory describing atomic systems @xcite . ( nevertheless , the underlying theories for the two cases are very different . ) as a consequence , atomic systems characterized by large scattering lengths can be studied with techniques developed in nuclear physics and , conversely , provide an excellent testing ground for few- and many - body methods that can be further applied , with little or no change , to the description of nuclear systems at low energies . the no - core shell model ( ncsm ) is one of the most flexible _ ab initio _ methods used to obtain the solution to the non - relativistic schrdinger equation for many - nucleon systems @xcite . currently , it is the only such method able to reach medium - mass nuclei with no restrictions to closed or nearly - closed shells and at the same time to handle local and non - local interactions on the same footing . it uses a discrete single - particle basis typically a ho basis and , being a numerical method , relies on a suitable truncation of the model space accessible to nucleons in the form of a maximum number @xmath4 of accessible shells above the minimum configuration . this requires the use of effective interactions ( not only among nucleons but also among nucleons and external probes such as photons , when they are present ) to account for effects left out by the truncation to a finite model space . the method of choice has been the construction of effective operators via unitary transformations . this involves the use of an approximation , the so - called cluster approximation " , which is not _ a priori _ controlled , and which poses challenges for the description of low - momentum observables @xcite . an alternative is to use eft to construct effective interactions that are consistent with the underlying theory of qcd directly in the model spaces where many - nucleon calculations are carried out @xcite . in this paper , we consider the problem of two trapped atoms with large scattering length , @xmath5 , from the perspective of eft interactions solved with the ncsm . the main simplification with respect to the nuclear case is that the trapping potential provides a natural single - particle basis ; its ho length @xmath6 does not need to be removed at the end of the calculation because it represents the long - distance physics of the trap . as long as @xmath7 , the trapped system should still exhibit universal behavior , although for @xmath8 it is significantly different from that of the untrapped system . we use eft techniques to expand the interaction between particles as a series of contact interactions with an increasing number of derivatives , and calculate energy levels inside the trap by explicitly solving the schrdinger equation . we extend here the work initiated in ref . @xcite , in which only the leading - order ( lo ) interaction was investigated . we include explicitly corrections up to the next - to - next - to - leading - order ( n@xmath9lo ) , and show how they provide improved convergence as @xmath4 increases , at least as long as they are treated in perturbation theory . finite and infinite values of the scattering length @xmath1 are considered , and we show that in the @xmath10 limit our lo results converge to the levels of a pseudopotential @xcite with the same scattering length , obtained by busch _ starting at nlo a finite value of the effective range @xmath11 is allowed as well , and at higher orders other effective - range expansion parameters could be included similarly . the @xmath10 limit of our subleading - order eft provides a derivation of the generalized busch _ et al . _ relation @xcite . with the two - body system thus understood , we can use our method to calculate the energies of larger trapped systems @xcite . other approaches to the same problem of course exist in the literature . they are frequently based on a specific form for the interparticle potential ( see , for example , ref . closest in spirit to ours is probably the approach @xcite where an effective short - range interaction is fitted to several levels of the pseudopotential at unitarity , but diagonalized exactly . what distinguish our framework are its _ i ) _ systematic character , in the form of a controlled expansion ; and _ ii ) _ generality , since no assumptions about the form of the short - range interactions are needed . these features allow the same method to apply across fields . of course , the low - energy observables themselves should be in agreement among different approaches , as long as they are calculated properly . the paper is organized as follows . we first show in sec . [ sect2 ] the general formalism of our approach , including a detailed description of the treatment beyond leading order . we illustrate the method in different situations ( finite and infinite scattering length , negligible and non - vanishing effective range , _ etc . _ ) in sec . [ sect3 ] . we conclude and discuss future applications in sec . [ sect4 ] . appendices [ appd ] , [ appa ] , and [ wf_nlo ] provide some of the details omitted in the main text . we consider a non - relativistic system of two particles of reduced mass @xmath12 that interact with each other in the @xmath0 wave . for definiteness , we think of two - component fermions , which support a single @xmath0 channel and interact also in @xmath13 and higher waves , but the framework can be straightforwardly applied to bosons and other fermions . in free space , the properties of this system can be characterized by scattering phase shifts for each partial wave of angular momentum @xmath14 at relative on - shell momentum @xmath15 , @xmath16 . when the relative momentum is much smaller than the inverse of the range @xmath2 of the interparticle interaction , @xmath17 ( we use units in which @xmath18 and @xmath19 ) , the phase shifts are given by the effective range expansion ( ere ) ; for example , in the @xmath0 wave , @xmath20 where @xmath1 , @xmath11 , @xmath21 , @xmath22 are , respectively , the scattering length , effective range , shape parameter , and higher ere parameters not shown explicitly . a similar expansion exists for @xmath13 and higher partial waves . generically , the sizes of ere parameters are set by @xmath2 , for example @xmath23 . the ere ( [ ere ] ) is an expansion in powers of @xmath24 . given a desired precision , the ere can be truncated and the system described by a finite number of parameters . potentials that generate the same values for this finite number of ere parameters can not be distinguished at this precision level : they all generate the same wavefunction for distances beyond the range of the force , @xmath25 . such potentials are said to belong to the same universality class . most interesting is the case where the depth of the interaction potential is fine - tuned so that an @xmath0-wave bound state is near threshold and the associated scattering length is large , @xmath5 . then the physics of the bound state is largely independent of the potential ; for example , for @xmath26 the wavefunction of a state of energy @xmath27 is @xmath28\frac{r}{a_2 } + { \cal o}\left(\frac{r^2}{a_2 ^ 2}\right)\right\}. \label{aswf}\ ] ] physics in the unitarity limit , @xmath29 , is controlled by the zero - energy bound state and exhibits a higher degree of universality . the ere description is useful because it is model independent . however , it does not directly provide a basis for the study of many - body systems . this can be accomplished using the ideas of eft @xcite . since at low momenta the details of the interaction can not be resolved , we can expand the interparticle interaction @xmath30 as a taylor series in momentum space , as done in app . [ appd ] ; in coordinate space , @xmath31 \delta(\vec{r } ) + \delta(\vec{r}\ , ' ) \left[\nabla^2\delta(\vec{r})\right]\right\ } \nonumber\\ & & + c_4 \left\{\left[\nabla\,'^4\delta(\vec{r}\,')\right ] \delta(\vec{r } ) + \delta(\vec{r}\ , ' ) \left[\nabla^4\delta(\vec{r})\right ] + 2 \left[\nabla\,'^2\delta(\vec{r}\,')\right ] \left[\nabla^2\delta(\vec{r})\right ] \right\ } + \ldots \label{taylorcoord}\end{aligned}\ ] ] where @xmath32 , @xmath33 , and @xmath34 are parameters , and `` '' denote interactions that contribute at higher orders . because the interactions are singular , an ultraviolet ( uv ) cutoff @xmath35 has to be introduced to solve the schrdinger equation . in order for observables to be independent of @xmath35 ( renormalization - group invariance ) , the parameters @xmath36 have to depend on @xmath35 this is merely a consequence of the fact that the uv cutoff is an arbitrary separation between the short - range dynamics included explicitly in the dynamics ( through high virtual momenta ) versus that included implicitly in the potential ( through its parameters ) . contributions to observables obtained from @xmath30 ( [ taylorcoord ] ) can also be organized in powers of @xmath24 . it has , in fact , been shown @xcite that in the two - body sector the eft expansion reproduces the ere ( [ ere ] ) at each power of @xmath24 . in the generic situation , one can simply treat the whole potential in perturbation theory . when @xmath5 , however , the @xmath32 term in eq . ( [ taylorcoord ] ) needs to be solved exactly , while the remaining terms can still be accounted for in perturbation theory @xcite . these higher - order terms represent range and subtler effects in the @xmath0 wave , and @xmath13 and higher waves . here we are interested in the large - scattering - length scenario , when the two particles are trapped in an ho potential of frequency @xmath37 . the ho introduces a third scale , the length @xmath38 . in the relative frame , the hamiltonian for this system reads @xmath39 + v. \label{hamil}\ ] ] as long as @xmath40 , details of the interparticle potential remain irrelevant and universality is not destroyed . in the following we consider various values for the ratio @xmath41 , which are , as discussed in sec . [ sec : intro ] , of interest in both atomic and nuclear physics . we want to set up a perturbative approach ; we thus write the hamiltonian ( [ hamil ] ) for the relative motion as @xmath42 with the energy and wavefunction decomposed accordingly , @xmath43 and @xmath44 the superscript @xmath45 corresponds to the order in perturbation theory of the different terms . here we order interactions according to the power counting of ref . @xcite and for convenience we split the parameters @xmath36 in eq . ( [ taylorcoord ] ) among different orders . the leading - order ( lo ) hamiltonian is @xmath46 + c_0^{(0 ) } \delta(\vec{r } ) , \label{lopot}\ ] ] and the corresponding wavefunction @xmath47 is the solution of the schrdinger equation @xmath48 the next - to - leading - order ( nlo ) correction to the potential is @xmath49 + 2\left[\vec{\nabla } \delta(\vec{r})\right]\cdot \vec{\nabla } + 2\delta(\vec{r})\nabla^2\right\ } , \label{nlopot}\ ] ] and the first - order corrections to the energy , @xmath50 , and to the wavefunction , @xmath51 , are obtained in first - order perturbation theory . that is , they are such that @xmath52 the next - to - next - to - leading - order ( n@xmath9lo ) correction to the potential @xmath53 is given by @xmath54 + 2\left[\vec{\nabla } \delta(\vec{r})\right]\cdot \vec{\nabla } + 2\delta(\vec{r})\nabla^2\right\ } \nonumber\\ & & + c_4^{(2 ) } \left\{\left[\nabla^4\delta(\vec{r})\right ] + 4\left[\vec{\nabla } \nabla^2\delta(\vec{r})\right]\cdot \vec{\nabla } + 4\left[\vec{\nabla } \vec{\nabla } \delta(\vec{r})\right ] \cdot \cdot \vec{\nabla}\vec{\nabla } \right.\nonumber\\ & & \left . \qquad\quad + 4\left[\vec{\nabla } \delta(\vec{r})\right]\cdot \vec{\nabla}\nabla^2 + 2\delta(\vec{r})\nabla^4\right\}. \label{nnlopot}\end{aligned}\ ] ] the corrections @xmath55 and @xmath56 are obtained from @xmath57 which just means perturbation theory to first order in @xmath53 and to second order in @xmath58 . extension to higher orders is straightforward . we work on a basis of ho wavefunctions @xmath59 with energies @xmath60 . useful properties of these wavefunctions are summarized in app . [ appa ] . in the ho basis the singularity of the potential ( [ taylorcoord ] ) can be tamed by imposing a maximum number of shells @xmath4 @xcite , which corresponds to a uv momentum cutoff @xmath61 we therefore expand the wavefunction ( [ wf ] ) in the ho basis in the finite space , @xmath62 where @xmath63 are coefficients to be determined . since the two - body potentials up to n@xmath9lo only support @xmath0 waves , to this order the eigenfunctions with @xmath64 are simply the ho wavefunctions @xmath59 with eigenvalues @xmath65 up to the energy @xmath66 . for @xmath67 , on the other hand , the levels are affected by the interparticle potential . denoting by @xmath68 the maximum value of the radial quantum number ( that is , @xmath68 is the largest integer smaller than , or equal to , @xmath69 ) and omitting the labels @xmath67 and @xmath70 , the @xmath0 wavefunction can be written as @xmath71 in terms of the ho @xmath0 wavefunctions @xmath72 , @xmath73^{-1/2 } e^{-r^2/2b^2 } l_n^{(1/2)}\left(r^2/b^2\right ) , \label{hoswf}\end{aligned}\ ] ] where @xmath74 is the generalized laguerre polynomial . the resulting energies ( [ e ] ) will depend on @xmath4 as well as @xmath37 , @xmath75 . since @xmath4 is arbitrary , we want the energies not to depend sensitively on @xmath4 . this can not be achieved in general , but it can for the shallow levels of interest that is , those with @xmath76 , which are dominated by physics at distances @xmath25 . as we show in the following , this is accomplished by allowing the @xmath77 to depend on both @xmath4 and @xmath37 , @xmath78 . nevertheless , at any order a residual @xmath4 dependence introduces an error in the calculation of shallow levels , which should be proportional to powers of @xmath79 . at the end of the calculation we want to take @xmath4 sufficiently large , @xmath80 , so that this error is not larger than the error proportional to powers of @xmath2 stemming from the truncation of eq . ( [ taylorcoord ] ) . the physics at lo is obtained by diagonalizing the two - body hamiltonian ( [ lopot ] ) . the approach is similar to the treatment in a free - particle basis @xcite , with the difference that we work here only with bound states , which naturally are closely related to ho states because of the presence of the trap . renormalization of the interaction at lo has already been discussed in ref . @xcite , but for completeness we repeat the derivation here . we start with the schrdinger equation ( [ sch0 ] ) for the wavefunction of the two - fermion system , @xmath47 . one can show , by inserting eq . ( [ expansions ] ) into eq . ( [ sch0 ] ) and projecting on a ho basis state @xmath81 , that the expansion coefficients are given by @xmath82 where @xmath83 is itself a combination of the unknown coefficients . direct substitution of the expansion coefficients back into eq . ( [ expansions ] ) yields @xmath84 from the consistency condition at @xmath85 we obtain the relation @xcite that an energy @xmath86 has to satisfy : @xmath87 -\frac{\pi}{2 } \frac{\gamma\left(\frac{3}{4}-\frac{e^{(0)}(\omega)}{2\omega}\right ) } { \gamma\left(\frac{1}{4}-\frac{e^{(0)}(\omega)}{2\omega}\right ) } \right\ } , \label{c_0_bis}\end{aligned}\ ] ] where @xmath88 in terms of the generalized hypergeometric function @xmath89 . ( [ c_0_bis ] ) we used eqs . ( [ suml01 ] ) and ( [ suml02 ] ) . ) finally , the constant @xmath90 is fixed by the chosen normalization of @xmath91 . ( it can be calculated using eq . ( [ suml04 ] ) . ) note that this constant is energy dependent ; when necessary we denote it by @xmath92 . the rhs of eq . ( [ c_0_bis ] ) clearly depends on @xmath68 and @xmath37 , and so does @xmath93 . in a given model space , the coupling constant @xmath93 can be fixed to reproduce one observable , in this case one energy of the two - body system . if , to be definite , we take that energy as the ground - state energy in the trap , @xmath94 then @xmath95 is determined from @xmath96 -\frac{\pi}{2 } \frac{\gamma\left(\frac{3}{4}-\frac{e_0(\omega)}{2\omega}\right ) } { \gamma\left(\frac{1}{4}-\frac{e_0(\omega)}{2\omega}\right ) } \right\}. \label{c_0_bisbis}\end{aligned}\ ] ] while one energy in the spectrum is fixed in all model spaces , the rest of the energy spectrum runs with the model space : the remaining energies @xmath97 satisfy eq . ( [ c_0_bis ] ) and in general depend not only on @xmath37 but also on @xmath68 . these energies should converge as @xmath98 to finite values @xmath99 . however , once @xmath35 in eq . ( [ lambda ] ) exceeds @xmath100 , the theoretical errors are dominated by the physics of the effective range @xmath11 , which was left out of lo . in the two - nucleon system , the effective range is much smaller than the scattering length , but is finite . in an atomic system near a feshbach resonance , the range is usually neglected , although it should become relatively more important as one moves away from the resonance . in either case , the lo , which ignores the range , contains energy - dependent errors of @xmath101 . in addition , in both cases , the truncation to a model space excludes physics of momenta beyond @xmath35 , which introduces further energy - dependent errors of @xmath102 . in other words , the truncation induces contributions to the effective range , the shape parameter , and so on , governed by @xmath35 rather than by @xmath100 . the role of contributions beyond lo is to correct for these two types of energy - dependent errors : nlo for @xmath103 errors and higher orders for higher powers of @xmath15 . in the untrapped system , the power counting for a system with a large @xmath0-wave scattering length @xmath1 is such that corrections beyond lo have to be treated as perturbations @xcite . it is one of our goals in this paper to show that treating higher orders in perturbation theory in the presence of the trap allows for a systematic improvement of the two - body energies . at nlo , we include corrections as first - order perturbations on top of the lo wavefunction @xmath104 . as discussed above , in lo we fix @xmath32 so that one of the states has the observed " energy . the nlo correction in eq . ( [ taylorcoord ] ) introduces a new parameter @xmath33 that can be chosen so that a second energy level is fixed . however , the nlo term induces , in general , a non - vanishing correction to the energy used to fix @xmath32 . one should , therefore , readjust @xmath32 so that in nlo we reproduce the two observables ( energy levels ) at the same time . in order to keep track of this change , it is convenient to split @xmath32 into an lo piece @xmath93 , which remains unchanged , and an nlo piece @xmath105 , as done in eq . ( [ nlopot ] ) , and treat the latter in perturbation theory as well . thus , for energies we have @xmath106 \right\ } \nonumber\\ & = & \frac{\kappa^{(0)2}}{c_0^{(0)2 } } \left\{c_0^{(1)}+ 4\mu c_2^{(1)}\left[e^{(0 ) } - \frac{4c_0^{(0)}}{3\pi^{2 } b^3 } \frac{\gamma\left(n_{max}+\frac{5}{2}\right ) } { \gamma\left(n_{max}+1\right)}\right ] \right\ } , \label{eq : corrnlo}\end{aligned}\ ] ] where we used eq . ( [ suml03 ] ) . the requirement that two energy levels have the correct positions in the spectrum fixes the amount of change from lo , thus providing two equations that determine the unknown coupling constants @xmath105 and @xmath107 in each model space . in the case when the lowest level @xmath108 is already fixed to a given ( experimental or theoretical ) value , @xmath109 . however , in the case with finite non - negligible range , one can alternatively choose that in lo @xmath93 be fixed to a level of the two - body spectrum without a range , while in nlo that level is shifted to the correct position with range , thus requiring that @xmath110 . these two alternatives are the ho - basis equivalent to fixing @xmath93 in a free - particle basis to , respectively , a known binding energy ( such as the deuteron binding energy ) or the scattering length . in either case , if we take , say , the first excited level @xmath111 to be reproduced at nlo in addition to the ground state , the two equations for the determination of @xmath112 and @xmath113 can be written as @xmath114 from eqs . ( [ eq : corrnlo ] ) and ( [ 2eqs ] ) , we can easily solve for @xmath112 and @xmath113 : @xmath115 and @xmath116 with these coupling constants fixed , eq . ( [ eq : corrnlo ] ) provides values for the other energy levels . the form of the nlo wavefunction is a bit more complicated ; in app . [ wf_nlo ] we show that , up to higher - order terms , @xmath117 where @xmath118 . \label{a1}\end{aligned}\ ] ] of course , just as in lo , the levels not used as input at nlo will have errors @xmath119 . as we show explicitly later , the magnitude of these errors is smaller than at lo , since more physics has been accounted for . if further precision is desired , we can continue the procedure to higher orders . in this paper we consider one order more , n@xmath9lo , so as to show the systematic trend of improvement but we omit details that can be obtained straightforwardly , if painfully . the correction @xmath55 to the energy is obtained using perturbation theory up to the second order . in addition to the second - order correction from @xmath58 ( [ nlopot ] ) , one has the first - order correction from @xmath53 ( [ nnlopot ] ) : @xmath120 the potential @xmath53 affects the energy levels that were fixed already , so again it is convenient to compensate for this by adding in eq . ( [ nnlopot ] ) perturbative shifts @xmath121 and @xmath122 with respect to the lower - order parameters . these two parameters , together with the four - derivative parameter @xmath123 , are determined so that three energy levels are fixed to the correct values . taking the lowest three levels @xmath124 , @xmath125 , to be fixed , the three equations for the determination of @xmath126 , @xmath127 , and @xmath128 are @xmath129 obviously , higher corrections can be added in a similar fashion . in the absence of a trap , the lo eft is formally equivalent @xcite in the infinite - cutoff limit to the pseudopotential @xcite . as we show here , the situation is the same in the presence of the ho potential , where the pseudopotential was solved in ref . we consider explicitly the eft to nlo , in order to derive also the first corrections to the pseudopotential in the trap . the wavefunction to nlo for a finite value of @xmath68 is given in eq . ( [ wf2_nlo ] ) . by having @xmath130 and using eq . ( [ suml ] ) one obtains : @xmath131 where @xmath132 is the confluent hypergeometric function , and we have introduced the energy @xmath133 which is the limit of the energy ( in units of @xmath37 ) of the two - body system in the trap . the second term in eq . ( [ wf_nlo_inf ] ) was obtained by using the completeness of the ho basis , @xmath134 the singularity of this term is mitigated by the pre - factor @xmath135 , which vanishes as @xmath79 for large cutoff . it stems from the enhancement of high virtual momenta due to the second derivatives in @xmath58 . as shown in app . [ appd ] , for an equivalent energy - dependent potential this term is absent . for small , non - vanishing values of @xmath136 , _ i.e. _ @xmath137 , use of eq . ( [ smallx ] ) gives @xmath138 by identification of this wavefunction with the wavefunction in the untrapped case , eq . ( [ aswf ] ) , we obtain the relation between the energy of the trapped two - body system and the ere parameters : @xmath139 in our framework , eq . ( [ transcend ] ) is to be interpreted in perturbation theory . at lo only the scattering length is taken into account , and the energies @xmath140 are given by @xmath141 a relation first found in ref . @xcite using the pseudopotential @xcite . the latter can be viewed as a renormalization of the delta - function interaction @xcite . indeed , the strength of the lo delta - function interaction has to decrease with the increase of the model - space size in order for it to give sensible results . one can see from eq . ( [ asygr ] ) that for large @xmath68 the first term in eq . ( [ c_0_bisbis ] ) grows as @xmath142 . taking the limit of eq . ( [ c_0_bisbis ] ) and using eq . ( [ transcendlo ] ) , one finds @xmath143 . \label{c_0run}\end{aligned}\ ] ] in the @xmath144 limit this running is exactly the one found in free space @xcite : the non - trivial rate of change with the cutoff @xmath35 is controlled by @xmath145 . this is not surprising since the large-@xmath35 behavior should be independent of the long - range physics of the trap . at nlo , the effective range appears : eq . ( [ transcend ] ) becomes @xmath146 which can be solved to this order as @xmath147 in terms of the digamma function @xmath148 . in the presence of range , all levels change from lo to nlo . the range in fact controls the asymptotic behavior of the nlo coupling constants . in the large-@xmath68 limit we find from eqs . ( [ c01 ] ) and ( [ c21 ] ) that @xmath149 \label{c01asym}\ ] ] and @xmath150 . \label{c21asym}\ ] ] again , this is the same running as in free space @xcite , as it should be . note that it qualitatively changes for @xmath151 . in this case , @xmath152 , which means , given our choice of levels to fix the coupling constants at nlo , that @xmath109 and @xmath153 . as a consequence the ratio of energies in eq . ( [ c21 ] ) goes to @xmath154 as @xmath155 , and @xmath156 and @xmath157 approach constants . at n@xmath9lo the power counting of ref . @xcite indicates that no new ere term should be included . this is a reflection of the fact that a fine - tuning in @xmath1 does not in general lead to an enhancement in the shape parameter @xmath21 : in order to go from @xmath158 to @xmath159 two orders are needed . therefore at n@xmath9lo eq . ( [ transcend ] ) is unmodified . it is only at n@xmath160lo that we need to account for a non - vanishing @xmath0-wave shape parameter , also with an @xmath0-wave potential of the form of eq . ( [ nnlopot ] ) ( but with different parameters ) . for fermions , n@xmath160lo also contains a @xmath13-wave interaction to account for the @xmath13-wave scattering volume . it is clear that the procedure can be continued to higher orders , and we expect for @xmath67 levels @xmath161 where @xmath162 and @xmath163 is given by the ere ( [ ere ] ) . this extension to subleading orders agrees with refs . @xcite . waves with @xmath164 can also be examined with the same method we developed here , but we leave details for future work . the importance of eq . ( [ transcend ] ) lies in the link between the energies inside the ho well , @xmath165 , and the scattering parameters , @xmath1 , @xmath11 , _ etc . _ it is the analog of lscher s formula @xcite , which links the levels inside a cubic box and the same scattering parameters . as such , eq . ( [ transcend ] ) provides the energy levels necessary to fix the coupling constants in finite model spaces when the ere parameters are known . in the nuclear case , for example , the scattering parameters have been determined from data , so eq . ( [ transcend ] ) can be used to fix the parameters of the pionless eft without relying on a fit to nuclear levels , as done in ref . an extension of ref . @xcite to this case is in progress . in the atomic case , the lowest energy @xmath166 of two trapped particles has been measured @xcite . one can use @xmath166 directly as input in eq . ( [ c_0_bisbis ] ) . alternatively , this energy was found @xcite to be in good agreement with the lowest state of the theoretical spectrum obtained under a pseudopotential assumption @xcite , in which the eigenvalues are determined by the scattering length @xmath1 . in ref . @xcite we used this theoretical lo energy as input ; we confirmed that the two - fermion spectrum of the underlying pseudopotential is reached asymptotically as @xmath130 , and we calculated the properties of three- and four - fermion systems . in subleading orders , we can now include effective range and higher effects , which might account for the small discrepancies @xcite between the theory and experiment . we should stress that , because the bare coupling constants are not observables , their sizes have no direct physical meaning . what matters is the total contribution of a given order to an observable . for example , for @xmath167 , @xmath168 , and yet , the nlo energy shift ( [ energyexpansion_nlo ] ) is small as long as @xmath11 is sufficiently small . we show explicit results for energies in the next section . in this section , we illustrate the approach described above in a few cases of interest for atomic and nuclear systems . we initially consider situations where the range of the interaction can be neglected ; we first look at the unitarity limit , @xmath169 , and then finite @xmath41 . finally , we consider the case where the range of the interaction is non - negligible _ albeit _ still small with respect to the scattering length . in all cases , we use eq . ( [ transcend ] ) to provide the asymptotic levels in the trap . we start by considering a system of two trapped particles at unitarity , characterized by @xmath169 and @xmath170 . in the untrapped case , this situation can be realized by considering an attractive potential given by , for instance , a square well with a fine - tuned depth . asymptotically , the wavefunction at zero energy behaves as @xmath171 and the scattering length @xmath1 is then infinite . in the presence of the harmonic trap , the only scale is set by @xmath6 , or , equivalently , @xmath37 , so the solutions of eq . ( [ transcend ] ) have to be constants ; indeed , they are given by the poles of the gamma functions in the denominators , _ @xmath172 where @xmath173 is an integer . at each order we use a finite number of these energies to determine the interaction parameters in each model space : @xmath174 at lo , @xmath174 and @xmath175 at nlo , and @xmath174 , @xmath175 , and @xmath176 at n@xmath9lo . using the ground - state energy to determine @xmath93 , eq . ( [ c_0_bisbis ] ) simplifies to @xmath177 which at large cutoff becomes @xmath178 . \label{c_0rununit}\ ] ] analogous expressions can be derived for the other coupling constants . the resulting running of the coupling constants is shown in figs . [ c0_lambda_unitarity ] , [ c2_lambda3_unitarity ] , and [ c4_lambda5_unitarity ] . one can see that , in agreement with sec . [ bush_formula_limit ] , a coupling constant @xmath36 behaves asymptotically as @xmath179 . with the interaction parameters so determined , we calculate the remaining energies , which depend on @xmath68 . results for some excited states , from the second ( @xmath180 ) to the fourth ( @xmath181 ) , are shown in fig . [ states_unitarity ] for different values of the cutoff @xmath68 in the dimensionless combination @xmath182 . at lo and nlo , all states change with @xmath182 ; at n@xmath9lo , the second excited - state energy is used as input . as we showed in sec . [ bush_formula_limit ] , these calculated energies converge as @xmath68 increases to the values given in eq . ( [ trans_unit ] ) . the plot explicitly shows , in addition , that the convergence with respect to @xmath68 to the exact value is increased as higher - order corrections are added in perturbation theory . of the second ( @xmath176 ) , third ( @xmath183 ) , and fourth ( @xmath184 ) excited - state energies , in units of @xmath37 , for two particles in a trap with the interaction characterized by @xmath169 and @xmath170 . results at lo ( green diamonds ) , nlo ( red squares ) , and n@xmath9lo ( black circles ) are compared with the exact values ( blue solid lines ) given by eq . ( [ transcend ] ) . the coupling constants are fixed so that the ground - state , additionally the first excited - state , and additionally the second excited - state energies are reproduced in lo , nlo , and n@xmath9lo , respectively . ( note that , indeed , at n@xmath9lo the second excited state in the leftmost panel is constant on top of the exact value for all values of @xmath182 . ) ] in order to further study the effects of a finite @xmath68 , we consider eq . ( [ transcend ] ) from a different angle . hitherto , we have used it simply to fix observables ( energy levels ) . now we employ it also to extract phase shifts . in a given model space , characterized by a given @xmath68 , we calculate the energy levels @xmath185 . the lowest levels used in the fitting of coupling constants are , of course , exact , while all others deviate from the exact values . the momentum @xmath186 associated with an energy level @xmath185 is @xmath187 so that the phase shifts for discrete values of @xmath15 , @xmath188 , can be determined by simply inverting eq . ( [ transcendallo ] ) : @xmath189 this allows us to study the effect of the truncation in a ho basis in terms of the more familiar ere parameters : any deviation from the value @xmath190 is due to truncation errors and we can calculate the induced range , shape parameter , and so on . in fig . [ phase_shift_unitarity_all_diag ] , we plot @xmath191 in a model space with @xmath192 as a function of @xmath193 . ( it is more natural to express @xmath15 in units of @xmath145 , but at unitarity @xmath6 provides the only length unit . ) at lo , @xmath194 starts off linear in @xmath193 : a linear fit ( indicated by the dashed line ) shows an induced effective range of about @xmath195 in this particular model space . at larger values of @xmath193 deviation from the linear behavior is seen , indicating the presence of further induced ere parameters . nlo and n@xmath9lo corrections reduce the size of the ere parameters , so that the results for the phase shifts improve order by order , getting closer and closer to the horizontal axis . since here @xmath196 , at @xmath197 , the errors are dominated by the higher orders , and therefore the lower orders do little to improve on the previous orders . however , at low momentum , the results systematically improve , as expected . -wave scattering phase shifts @xmath191 as function of the dimensionless squared momentum @xmath193 in the finite model space characterized by @xmath192 , at unitarity . the points at lo ( green circles ) , nlo ( red squares ) , and n@xmath9lo ( black diamonds ) are obtained from calculated energies via eq . ( [ transcendalloinv ] ) . the dashed line corresponds to a linear fit of the lo curve at small @xmath198 values . for comparison , we also show results with the nlo potential fully diagonalized ( black stars ) . ] our perturbative treatment thus provides a way to systematically reduce the effect of truncation to a reduced model space , demanded in the calculation of larger systems . emboldened by this success , one might be tempted to treat the subleading potentials exactly in the schrdinger equation . apart from the renormalization problems pointed out in ref . @xcite , we find no obvious numerical improvement when this is done here . in fig . [ phase_shift_unitarity_all_diag ] we also show for comparison results obtained when the nlo correction to the potential is not considered as a perturbation but fully diagonalized together with the lo potential . ( the same two lowest levels are used to fix the two coupling constants of the lo+nlo potential . ) one can see that by considering the lo+nlo potential this way , results are further away from the exact curve than by treating the nlo potential as a perturbation . this result is not particular to this example ; more generally , truncation errors get worse when subleading corrections are treated improperly . the reason is that doing so includes only part of the higher - order corrections ; it neglects the rest , needed to ensure systematic improvement . in the previous subsection we saw that the rate of approach to the asymptotic values of two - body energies improves systematically as the order increases at unitarity . we now show that there are no qualitative changes when we consider the case of a pseudopotential away from unitarity @xcite , where the scattering length @xmath1 is finite , with @xmath11 still vanishing . the corrections to the potential are taken into account as in the unitarity case , _ i.e. _ , the lo correction is iterated to all orders whereas higher corrections are treated as perturbations . the parameters at each order are adjusted so that the lowest levels satisfy eq . ( [ transcend ] ) exactly . the @xmath35 dependence of coupling constants is similar to unitarity , except for a markedly slower convergence for @xmath32 . this is particularly obvious for @xmath93 , when we compare the more general eq . ( [ c_0run ] ) , which applies here , with its unitarity version ( [ c_0rununit ] ) . as for energies , let us first consider @xmath199 . this situation corresponds to the case when the depth of the potential between the two particles is decreased starting from the fine - tuned value at unitarity . as an example , we show in fig . [ states_b_over_a_min1 ] results for @xmath200 for the same energy levels previously displayed at unitarity in fig . [ states_unitarity ] . the exact values are slightly higher than at unitarity . as before , convergence of the energies to the exact values is improved as more corrections to the potential are added , and the difference between the truncated - space energy and the exact result is mitigated as more corrections are included . , but for @xmath201 . ] for a weak enough potential , @xmath202 becomes small . the energies are then close to the ho energies , @xmath203 where @xmath204 is a small correction and @xmath205 . by using eq . ( [ transcend ] ) and expanding the gamma function around its poles one finds @xmath206 , \label{deltaeigv_loi}\ ] ] where @xmath207^{-1 } = \left ( 1+\frac{1}{n}\right ) ^n\prod_{m=1 , m\ne n}^{\infty } \left(1+\frac{1}{m}\right)^n \left(1-\frac{n}{m}\right ) . \label{pn}\ ] ] we have verified numerically that in this case convergence to the exact value can even be sped up by considering _ all _ interactions as perturbations , as in the `` natural '' continuum case discussed in ref . @xcite . we now turn to the case where @xmath208 . as the interaction between particles becomes stronger ( starting from the case at unitarity ) the scattering length @xmath1 decreases . the ground state can have negative energy . the plot of the energy for excited states ( starting from the second ) for @xmath209 is shown in fig . [ states_b_over_a_1 ] . the exact values are now slightly lower than at unitarity . again , there is no qualitative change in the pattern of convergence with respect to unitarity or negative @xmath1 . , but for @xmath209 . ] for a strong enough interaction , the absolute value of the ground - state energy , @xmath210 , is large and in the trapped system we have @xmath211.\ ] ] using eq . ( [ transcend ] ) , we find @xmath212 . \label{deutbe}\ ] ] thus , when @xmath213 , the ground - state energy @xmath214 converges to @xmath215 , the untrapped bound - state energy for the case @xmath216 . the excited levels are again close to the ho values @xmath217 where @xmath218 . following the same procedure as for large , negative @xmath41 , we obtain the corrections @xmath219 . \label{deltaeigv_loiii}\ ] ] a finite interaction range @xmath2 usually generates higher ere parameters of the same magnitude , @xmath23 , @xmath220 , _ etc . _ , even when there is fine - tuning that leads to @xmath5 . as a last example , we account for a finite effective range @xmath11 . regardless of the quantity chosen as input in lo , the existence of range introduces errors that are energy dependent and can only be accounted for in subleading orders . as before , we use as lo input the ground - state energy @xmath174 given by eq . ( [ transcendlo ] ) . this fixes the running of @xmath93 to the same values as in the previous subsection . however , at nlo , we obtain @xmath105 and @xmath107 from the first two states of eq . ( [ transcend ] ) with @xmath1 _ and _ @xmath11 non - vanishing . nlo is , thus , different from the previous subsection : it accounts not only for errors @xmath221 due to the explicit truncation to the model space but also for implicit ones , @xmath222 , in the potential . as discussed in sec . [ bush_formula_limit ] , the n@xmath9lo corrections are slightly more subtle @xcite . the introduction of range leaves an error that can be as big as @xmath223 . errors from the explicit truncation of the model space are now @xmath224 or @xmath225 , and , as in general , are smaller than errors from the truncation of the expansion once @xmath226 . these types of errors are one order in @xmath24 or @xmath227 from nlo , which requires @xmath53 for control . in contrast , errors from the shape parameter are only @xmath228 , two orders in @xmath24 down from nlo . thus , at n@xmath9lo we determine @xmath121 , @xmath229 , and @xmath123 from the lowest three levels of eq . ( [ transcend ] ) , still with non - vanishing @xmath1 and @xmath11 and neglecting all higher - order ere parameters . as an illustration , we take @xmath230 . the coupling constants for finite scattering length @xmath231 are plotted in figs . [ c0_lambda_b_a_1 ] , [ c2_lambda3_b_a_1 ] , and [ c4_lambda5_b_a_1 ] . the running of @xmath93 is identical to the one in the previous subsection , and as advertised is gentler than the one displayed in fig . [ c0_lambda_unitarity ] . as seen in eqs . ( [ c01asym ] ) and ( [ c21asym ] ) , the range changes the running of @xmath105 and @xmath107 dramatically with respect to the zero - range case , say figs . [ c0_lambda_unitarity ] and [ c2_lambda3_unitarity ] . the new runnings approach the limits given in eqs . ( [ c01asym ] ) and ( [ c21asym ] ) . at @xmath232 we find , for example , @xmath233 . the slow convergence is a consequence that at this cutoff @xmath234 is only @xmath235 , and the @xmath236 corrections are still not so small . for the other coupling constants , @xmath121 , @xmath229 , and @xmath123 , the running is also very different from the zero - range case . in agreement with our error estimates , these parameters are nearly flat when normalized with @xmath237 . qualitatively , the only change in energies with respect to previous subsections is the finite jump from lo to nlo due to the effective range . energies for a few excited states in the case of an infinite scattering length , @xmath169 , are plotted in fig . [ states_unitarity_point_one ] , and for a finite scattering length , @xmath209 , in fig . [ states_b_over_a_1_point_one ] . as evident , lo misses the correct asymptotic behavior by a little bit because it lacks the effective range . the nlo results converge , as @xmath98 , to the values given by eq . ( [ transcend ] ) with the range , as they should . however , the inclusion of n@xmath9lo corrections speeds up convergence considerably : for a fixed value of the cutoff @xmath68 the results at n@xmath9lo are much closer to the exact value . it is straightforward , for example , to generalize eqs . ( [ deutbe ] ) and ( [ deltaeigv_loiii ] ) to non - zero range . , but for @xmath238 . ] , but for @xmath209 and @xmath238 . we have extended the work initiated in ref . @xcite on the application of effective field theory to the trapped two - particle system to include subleading orders . in eft , the short - range interparticle potential is replaced by a delta function and its derivatives , as we presented in the main text , or equivalently by an energy - dependent delta function , as discussed in app . the singular nature of the interaction requires regularization and renormalization , which is naturally accomplished by the use of a finite model space , as required in actual calculations . once the parameters of the interaction are fixed in each model space using a finite number of levels , other levels can be calculated . the leading - order interaction is solved for exactly , whereas higher contributions are treated in perturbation theory . we have considered the corrections to the potential up to next - to - next - to - leading - order . we have shown explicitly that this method can systematically account for the physics of the effective - range expansion , while treating the subleading potential exactly gives worse results . in the limit of a large model space , the leading - order theory reproduces the pseudopotential result of ref . @xcite , while in subleading orders our approach gives range and higher ere corrections @xcite to the pseudopotential result . therefore we have provided an alternative derivation of these results . in turn , these asymptotic results allow us to incorporate the physics of scattering into the trapped system : by fitting the coupling constants to the asymptotic values of some levels , we can calculate other energies without necessarily resorting to fitting measured bound - state energies . this is important in the nuclear case where the scattering parameters are known . we have studied numerically the convergence to large model spaces in some detail . we have presented results at unitarity and for finite values of the scattering length @xmath1 as well as for finite values of the effective range @xmath11 . in all cases , we have observed convergence to the asymptotic values . moreover , for a fixed value of the cutoff , the difference between the exact value and the value obtained in a finite model space is mitigated as more corrections to the potential are taken into account . truncation errors can thus be reduced not only by increasing the size of the model space but also by increasing the order of the calculation . thus , we have a method to calculate the energies of few - body systems that is independent of the details of the short - range potential and can be improved systematically . although in the two - body system our method is simply an implementation of the effective - range expansion , it is well - suited for extension to larger systems where numerical calculations are required . in these cases , limited computational power restricts the size of accessible model spaces . we can fight this limitation by increasing the order of the calculation , thus accelerating convergence . results for systems with more particles will be presented in a future publication @xcite . we thank thomas papenbrock for pointing out some useful references and mike birse for interesting discussions . the work reported here benefited from hospitality extended to its authors by the national institute for nuclear theory at the university of washington during the program on effective field theories and the many - body problem ( int-09 - 01 ) , and to uvk by the kernfysisch versneller instituut at the rijksuniversiteit groningen . this research was supported in part by nsf grants phy-0555396 and phy-0854912 ( brb , jr ) , and by us doe grants de - fc02 - 07er41457 ( is ) and de - fg02 - 04er41338 ( jr , uvk ) . analogously to the multipole expansion in electrodynamics , the short - range two - body potential can be expanded in a power series in momenta @xcite , @xmath239 where @xmath240 ( @xmath241 ) is the initial ( final ) relative momentum , the @xmath242 are constants , and `` @xmath22 '' denote terms with six or more powers of momenta . here the @xmath32 , @xmath33 , and @xmath34 terms contribute to the @xmath0 wave , @xmath243 and @xmath244 to the @xmath13 wave , and @xmath245 to the @xmath246 wave . the contribution from @xmath247 to the on - shell two - body system vanishes , and thus it is only relevant in larger systems , where it can not be separated from few - body forces . if we were to include this term and set a similar linear system to eq . ( [ 3eqs ] ) ( based on the fit of the first four levels inside the ho ) , this system would have no solution : its zero determinant would indicate that the corrections @xmath34 and @xmath248 are not independent . in the situation of interest here , where the two - body @xmath0-wave scattering length @xmath1 is large , the @xmath0-wave constants are enhanced over the others by powers of @xmath1 , and as a result the @xmath32 term is lo , the @xmath33 term is nlo , the @xmath34 term is n@xmath9lo , while the others appear only at higher orders @xcite . in this paper we limit ourselves to n@xmath9lo , where only the @xmath0 wave is present . taking the fourier - transform with respect to both @xmath240 and @xmath241 , the potential is found to be given in coordinate space by eq . ( [ taylorcoord ] ) . it is non - local in the sense of involving derivatives . a completely equivalent formulation of the eft in the two - body system is achieved with an energy - dependent potential @xcite . it can be implemented through an auxiliary `` dimeron '' field @xcite . in lo , the dimeron is characterized by a mass @xmath249 and a coupling constant @xmath250 to a two - particle @xmath0-wave state . in the case of interest here , @xmath5 , the kinetic energy of the dimeron is an nlo effect @xcite . a subtlety is that the bare dimeron can be a ghost : the sign @xmath251 of the kinetic term can be positive or negative , depending on the sign of the effective range . denoting by @xmath27 is the total energy in the center - of - mass frame , the two - body potential is simply @xmath252 in momentum space , and @xmath253 in coordinate space . it is local , but energy dependent . this potential was considered in ref . an alternative treatment follows the same steps as that of the momentum - dependent potential ( [ taylorcoord ] ) in the main text , with the substitutions @xmath254 in particular , the lo is identical to that presented in the main text , with eq . ( [ sub1 ] ) . note that since only the ratio @xmath255 enters , the separation between @xmath256 and @xmath257 is arbitrary . at nlo , one finds equations that are somewhat simpler than eq . ( [ eq : corrnlo ] ) and the ones that follow . with eqs . ( [ sub2 ] ) and ( [ sub3 ] ) one can find the renormalization of @xmath258 and @xmath259 separately . the appearance of energy instead of momentum leads to a softening of the uv behavior . this results in a wavefunction at nlo of exactly the form ( [ wf2_nlo ] ) and ( [ a1 ] ) , but with eq . ( [ sub3 ] ) . in the infinite - cutoff limit the most singular term in eq . ( [ wf_nlo_inf ] ) is absent , and eq . ( [ expansion_nlo ] ) follows directly . any observable in the two - body system is identical for this potential and the momentum - dependent potential discussed in the text . the ho basis functions with the length parameter @xmath260 are the solutions of the three - dimensional schrdinger equation @xmath261 with energy ( in units of @xmath37 ) @xmath262 they are given by @xmath263 where @xmath264 are the usual spherical harmonics , and the radial parts @xmath265 can be shown to have the form @xmath266^{-1/2 } \left(\frac{r}{b}\right)^l \exp\left(-r^2/2b^2\right ) l_n^{(l+1/2)}\left(r^2/b^2\right ) . \label{defhowf}\end{aligned}\ ] ] the @xmath267 s are the generalized laguerre polynomials , which can be written as @xcite @xmath268 in terms of the confluent hypergeometric function @xmath269 . for contact interactions , it is useful to know the value of the radial wavefunction at the origin . because of the @xmath270 factor , only @xmath67 contributes , as expected . the @xmath0 wavefunction of energy @xmath271 is , omitting the @xmath272 labels , @xmath273^{-1/2 } \exp\left(-r^2/2b^2\right ) l_n^{(1/2)}\left(r^2/b^2\right ) . \label{hoswfprime}\end{aligned}\ ] ] in particular , @xmath274^{1/2},\ ] ] where @xmath275 use of stirling s formula yields @xmath276 for large @xmath277 , which leads to @xmath278 for large @xmath279 . the generalized laguerre polynomials satisfy @xcite @xmath280 in terms of the confluent hypergeometric function @xcite @xmath281.\ ] ] for small @xmath282 , @xmath283 . \label{smallx}\ ] ] also useful are sums involving the generalized laguerre polynomials at the origin @xcite : @xmath284 @xmath285 @xmath286 , \label{suml02}\end{aligned}\ ] ] and @xmath287 \right . \nonumber\\ & & \left . + \frac{\gamma(m+3/2)}{\gamma(m+2 ) } \ , \frac{_4f_3\left(1 , m+3/2,a+m+1,a+m+1 ; m+2,a+m+2,a+m+2;1\right)}{\pi(a+m+1)^2 } \right\ } , \nonumber\\ \label{suml04}\end{aligned}\ ] ] where @xmath89 and @xmath288 are generalized hypergeometric functions and @xmath289 is the digamma function @xcite . the correction at nlo is taken into account as a perturbation to the lo potential . we write the corresponding quantum state of the two - fermion system as @xmath290 where @xmath104 is the solution at lo and @xmath291 the correction at nlo . both of these vectors are expanded in a ho basis , as in eq . ( [ expansions ] ) . corrections at nlo are obtained by solving the schrdinger equation at first order , eq . ( [ sch1 ] ) . let us introduce the projection @xmath292 of the left and right sides of eq . ( [ sch1 ] ) , @xmath293 and define @xmath294 as @xmath295 . from the lhs of eq . ( [ r_n_l ] ) we can write the coefficient of the wave function at nlo as @xmath296 normalizing the wavefunction at lo to unity , @xmath297 keeping this normalization at nlo , @xmath104 and @xmath291 are orthogonal ; one then obtains , from eqs . ( [ cm ] ) for @xmath298 and ( [ cm1_app ] ) for @xmath299 , @xmath300 on the other hand , the rhs of eq . ( [ r_m2_app ] ) provides an expression for @xmath292 . using eq . ( [ cm ] ) , @xmath301_{r=0}= -2\mu \sum_{n } c_{n}^{(0 ) } e_n\phi_{n}(0 ) = 2 \mu \left [ -e^{(0 ) } \psi^{(0)}(0 ) + \kappa^{(0 ) } \sum_{n } \phi_{n}^{2}(0)\right ] . \label { laplace_app}\end{aligned}\ ] ] inserting this expression into the rhs of eq . ( [ r_m2_app ] ) and eliminating @xmath302 , @xmath303 . \label{r_m2_ter_app}\end{aligned}\ ] ] from eqs . ( [ ortho ] ) and ( [ r_m2_ter_app ] ) , we arrive at an expression for @xmath294 using eq . ( [ c_0_bis ] ) : @xmath304 \nonumber \\ & & + 2 \mu c_2^{(1 ) } \left[\left(\frac{\kappa^{(0)}}{c_0^{(0)}}\right)^2 + \frac{2e_0}{c_0^{(0 ) } } -\sum_{n } \phi_{n}^{2}(0)\right ] . \label{kappa1_app}\end{aligned}\ ] ] we can now finally obtain the coefficient @xmath305 by plugging eqs . ( [ r_m2_ter_app ] ) and ( [ kappa1_app ] ) into eq . ( [ cm1_app ] ) : @xmath306 \nonumber \\ & & + 2\mu \kappa^{(0 ) } \frac{c_2^{(1)}}{c_0^{(0)}}\phi_n(0).\end{aligned}\ ] ] the total wavefunction at nlo is then given by @xmath307 where @xmath308 is a first - order term defined as @xmath309.\end{aligned}\ ] ] at first order in @xmath308 and @xmath50 , the previous expression for the wavefunction is equal to @xmath310 upon insertion of the expression for the radial wavefunction of the ho basis we obtain eqs . ( [ wf2_nlo ] ) and ( [ a1 ] ) .

we have performed an ecliptic survey of the kuiper belt , with an areal coverage of 8.9 square degrees to a 50% limiting magnitude of , and have detected 88 kuiper belt objects , roughly half of which received follow - up one to two months after detection . using this survey data alone , we have measured the luminosity function of the kuiper belt , thus avoiding any biases that might come from the inclusion of other observations . we have found that the _ cold _ population defined as having inclinations less than has a luminosity function slope , and is different from the _ hot _ population , which has inclinations greater than and a luminosity function slope . as well , we have found that those objects closer than 38 au have virtually the same luminosity function slope as the _ hot _ population . this result , along with similar findings of past surveys demonstrates that the dynamically cold kuiper belt objects likely have a steep size distribution , and are unique from all of the excited populations which have much shallower distributions . this suggests that the dynamically excited population underwent a different accretion history and achieved a more evolved state of accretion than the cold population . as well , we discuss the similarities of the _ cold _ and _ hot _ populations with the size distributions of other planetesimal populations . we find that while the jupiter family comets and the scattered disk exhibit similar size distributions , a power - law extrapolation to small sizes for the scattered disk can not account for the observed influx of comets . as well , we have found that the jupiter trojan and _ hot _ populations can not have originated from the same parent popuation , a result that is difficult to reconcile with scattering models similar to the nice model . we conclude that the similarity between the size distributions of the _ cold _ population and the jupiter trojan population is a striking coincidence . kuiper belt centaurs trojan asteroids accretion

1008.1058

the size distribution is one of the most fundamental properties of a planetesimal population . as the size of an object is primarily determined from its accretion and collisional disruption histories , the size distribution can reveal important information on the accretion and collisional history of that population ( for a recent example , see * ? ? ? unlike the closer populations such as the asteroid belt , whose proximity has allowed the accurate measurement of their size distributions ( see for example * ? ? ? * ; * ? ? ? * ) , the distance to the kuiper belt has prevented an equally detailed determination of its size distribution . observations thus far have demonstrated that for the kuiper belt as a whole , the size distribution for objects with diameters , @xmath4 km , is well described by a power - law . at some size @xmath5 km , the size distribution rolls - over to a shallower distribution @xcite . smaller than the roll - over , the size distribution appears to remain shallow to objects as small as @xmath6 km @xcite . while the accuracy of current measurements prevents a detailed modelling of the history of objects in the region , some insight has already been gained . the steepness of the large object size distribution implies that for this population , accretion was a short - lived phenomenon , likely not more than a few 100 myr @xcite . the paucity of the observed belt , and the existence of the largest known members demonstrates that the belt has undergone significant mass depletion , losing as much or more than 99% of its primordial mass @xcite . the existence of the roll - over at sizes larger than @xmath7 km suggests that the belt has undergone significant collisional comminution in a region of significantly increased density compared to today @xcite . as well , recent observations have suggested that the size distribution of those dynamically cold ( low inclinations and eccentricities ) kuiper belt objects is different than that of the dynamically hot ( large inclinations and eccentricities ) population @xcite . @xcite found that the size distribution of objects with inclinations , @xmath8 was steeper than the size distribution of objects with @xmath9 . @xcite found similar results showing that the cold population had a steeper slope than the mixed populations as a whole . this result implies that the hot and cold populations are genetically separate populations , that have significantly different accretion and evolution histories . the results of @xcite were drawn from samples of objects compiled from many different surveys . this practice was necessitated by available data . all surveys had either shallow limiting magnitudes and a large areal coverage , or vice - versa . the result was that most surveys did not have a sufficient range of objects for which the size distribution could be accurately measured from that survey alone . the practice of using data from multiple surveys opens the results to the possibility of being affected by calibration issues and variations in sky density which could lead to an incorrect measurement of the size distribution . when these effects were properly accounted for , @xcite found that these data could not be used to reliably test differences in subpopulations of the kuiper belt . here we present the results of a new kuiper belt survey . by virtue of the survey s design , a large number of objects were discovered and followed to determine their inclinations , over a range of sizes sufficient to measure the size distribution for the hot and cold population without the use of other surveys . in section [ sec : observations ] we present our observations , and our data reductions and discovery techniques . in section [ sec : analysis ] we present the analysis of our results , in section [ sec : discussion ] we discuss the implications of our results , and we end with concluding remarks in section [ sec : conclusions ] . the discovery observations were made on october 18 , 2009 ( ut ) with suprime - cam on the 8.2 m subaru telescope @xcite . suprime - cam is 10-chip mosaic camera , with a field - of - view of roughly 34@xmath1027 with @xmath11 chip - gaps , and has a pixel scale of 0.2 " . the observations consisted imaging 34 fields , each visited three times in the r filter with the camera long axis oriented horizontally in ra , with each visit consisting of a single 200 s exposure . the first and last images of a field were separated by roughly 6.5 hours . the observed fields were all within @xmath12 of the ecliptic and within @xmath13 of opposition at the discovery epoch . the total areal coverage of our survey after accounting for chip - gaps was 8.93 square degrees . details of the individual fields are shown in table [ tab : pointings ] and figure [ fig : pointings ] . the observations included multiple images of the d1 and d4 fields of the canada - france - hawaii telescope supernova legacy survey @xcite at various airmasses to provide both photometric and astrometric calibrations of the discovery images . the images were pre - processed with standard techniques ; the bias levels were removed , and a master bias frame , set as the average of 10 bias images was removed from all science frames . a master sky - flat was produced from the science frames using a clipped median filter , and was removed from the science images producing background variations no larger than @xmath14 across a mosaic . photometric calibrations were done on a chip - by - chip basis using the mega - pipe source catalog of the d1 and d4 fields @xcite . the zeropoint was found to be identical for all chips within uncertainties , with a value , @xmath15 . the conversion from the suprime - cam instrumental magnitudes to those of mega - pipe was found to be @xmath16 atmospheric extinction was measured from the science frames . the airmass correction was found to be 0.08 magnitudes per unit increase in airmass . comparison of common background sources in each triplet revealed no large seeing or transparency variations throughout the night ( see figures [ fig : seeing ] and [ fig : airmass ] ) ; the night was photometric . the suprime - cam field of view is highly distorted , especially near field edges , requiring calibration to ensure accurate astrometry of moving sources in the observations . scamp @xcite was used to measure second order spatial distortions of the suprime - cam field , from the d1 and d4 fields on a chip - by - chip basis . little variance in the distortions was found over the airmass range of these observations . thus , a master distortion map for each chip was produced from all images of the d1 and d4 fields . absolute calibration of each science image was done using the usno - b catalog , and the master distortion map , resulting in residuals of roughly 0.3 " . relative astrometry between the three images of each field - a triplet - was performed by matching sources between an image , and the reference image , chosen as the image of the triplet with the lowest absolute residuals . this resulted in an excellent common astrometric solution for the triplet , with residuals of @xmath17 between all three images . this calibrations procedure ensured reliable astrometry to both maximize the discovery efficiency of moving sources , as well as secure the possibility of follow - up on future dates . the suprime - cam data were utilized to search for moving objects in the solar system , in and beyond the orbits of the gas - giant planets . to characterize the detection efficiency of our moving object detection method , artificial moving point sources were implanted in the observations . the stellar point - spread function ( psf ) was generated on an image - by - image and chip - by - chip basis from 10 - 15 hand - selected , visually inspected point sources for each chip . it was found that a spatially constant psf was sufficient to model the stellar shape across a chip . the psfs were created using the tools in the _ doaphot _ package of _ iraf _ and for each image consisted of an average moffat profile with look - up table of the psf stars of that chip . random artificial sources were generated on sun - bound orbits , with semi - major axes , @xmath18 au , eccentricities , @xmath19 , and inclinations , @xmath20 . anomaly , periapse , and nodal angles were chosen at random to ensure that the sources fell in the images and could be either approaching or receding from their nodes . 20 - 30 artificial sources were planted in each chip with apparent magnitudes in the range @xmath21 with image - to - image flux variations matching those measured from 30 - 40 bright stars common to each image of a triplet , thus ensuring that any extinction variations throughout the observations were accounted for . if an artificial source s motion would cause it to drift more than 0.2 pixels during an exposure , it was broken up into multiple fainter sources with 0.2 pixel spacing and a total flux equal to that of the artificial source . thus , trailing effects were fully accounted for in our search . moving objects were identified by their motions between images of a triplet . using sextractor @xcite , all sources in an image were tabulated . stationary sources common to each image in a triplet were flagged , and ignored from further consideration , leaving only those sources who were either moving , or had highly variable fluxes . the remaining sources of an image triplet were then searched for motion consistent with sun - bound objects . candidate sources were chosen as any three individual point - source detections , one per image , whose centroid moved at least one pixel ( 0.2 `` ) and at most @xmath225 '' between each image . these candidate sources were then further filtered by _ if the best - fit orbit to the candidate had a chi - squared larger than 7.5 , the candidate was rejected . the results of this were approximately 16000 candidate sources which were then visual inspected by an operator for final acceptance or rejection . the majority of visually inspected candidates were false , primarily caused by cosmic ray impacts , or extended sources which were identified by sextractor as more than one source in some images . of the 16,000 candidates , roughly a third were matched with implanted artificial sources . from these sources , the performance of the search routine was characterized . the results are shown in figure [ fig : efficiency ] . as can be seen , the detection efficiency is well described by the familiar functional form @xmath23 where @xmath24 , @xmath25 , and @xmath26 are the maximum efficiency , half - maximum magnitude , and half - width parameters . the survey achieved a 50% limiting magnitude for discovery of r=24.7 . the search was found to be sensitive to objects as distant as @xmath27 au , beyond which , insufficient motion was exhibited for reliable detection . no significant variation in search efficiency with inclination or eccentricity was found ( see figures [ fig : eff_i ] and [ fig : eff_r ] ) . the search resulted in the detection of 88 sources . the sources distances and inclinations were determined with _ fit_radec_. photometry was measured using standard aperture photometry techniques with corrections for airmass applied ( see figure [ fig : airmass ] ) . aperture corrections were measured from the bright stars from which the psfs were generated . photometric uncertainties were determined from the function @xmath28 where @xmath29 is a constant representing the extinction stability of the night in question , @xmath30 is a constant proportional to the parameters of the detector ( see * ? ? ? * for a derivation ) , and @xmath31 is the telescope photometric zeropoint . this function was fit to the difference in measured and planted magnitudes of the artificial sources . the best - fit parameters are @xmath32 . the best - fit @xmath29 is in agreement with the scatter in extinction exhibited in figure [ fig : airmass ] . along with the measured fluxes with uncertainties sampled per measurement from equation [ eq : delta - r ] , these results are presented in table [ tab : objects ] . one detected object was identified as previously known plutino 2004 vt75 which was not known to be in the observed fields prior to detection . this detection confirmed the accuracy of both the photometric and astrometric calibrations . the detected position of 2004 vt75 was @xmath33 from that predicted from the minor planet center . as well , this source was detected across different chips demonstrating that the astrometric fits for different chips of the same mosaic image are reliable in an absolute sense . for a subset of the objects , follow - up observations were performed with the low resolution imaging spectrometer ( lris ) on the keck - i telescope on the nights of november 21st , 22nd and december 21st , 2009 ( ut ) . lris is a dual channel ( blue and red ) camera , with each camera , a 2 ccd mosaic having approximately a 6x8 field of view , with a @xmath34 gap , and a pixel scale of @xmath35 . this particular instrument is convenient for our follow - up observations , as the positional uncertainty ellipse of our targets at the time of follow - up was slightly smaller than the areal coverage of a single detector ccd . the follow - up observations consisted of pairs of images at the positions of each target , predicted from _ fit_radec _ , with temporal spacing between each image sufficient to detect the object s motion . by virtue of the dual channel imaging abilities of the camera , images in both the g and r lris filters were taken simultaneously . exposure times were scaled from the flux of each source in the discovery images , and were chosen such that each source would have a photometric signal - to - noise ratio of @xmath36 in each image . images of the d1 and d4 supernova fields were taken for calibration purposes . due to time constraints , only those objects who s distance might place them beyond 39 au , and with magnitude in discovery images brightward of @xmath37 were followed . 41 objects satisfied these constraints and were pursued in the follow - up observations . due to the geometry of the follow - up observations , asteroid confusion was a concern . the predicted rate of motion of the follow - up targets was used as a diagnostic of whether or not a detected moving object was the correct source . the second ccd of the lris detector provided a measure of this asteroid confusion rate , as no kuiper belt objects were expected to fall within this region of the images . in the november follow - up data , 3 asteroids with brightnesses and rates of motion consistent with our kuiper belt targets were found in the second ccd for all follow - up exposures . with the exception of two follow - up pointings , our objects were not confused , as only one moving target per field was identified with a rate of motion consistent with the targeted source . this is consistent with our asteroid confusion rate determined from the other ccd . special attention for the two objects with confused follow - up was made during the december observations . from the december data , the interloper in the november observations was easily rejected achieving reliable links between the discovery and follow - up images for all follow - up targets . no confusion occurred during the december follow - up . our follow - up efforts were rewarded with 100% success rate for all 41 targets satisfying our distance and magnitude cuts . the conditions on the november nights were photometric and of moderate seeing , with full - width at half maximum @xmath38 . on the night in december , while it was photometric , the seeing conditions were mediocre , reducing the observing efficiency compared to the november nights . as such , all of the 41 objects for which we attempted follow - up have month - long arcs , while only a small subset of 10 have arcs longer than this . the results of the follow - up efforts are presented in table [ tab : objects ] . astrometric and photometric calibrations of the lris images were performed using the same techniques as applied to the suprime - cam images . the astrometric distortions of the lris field were small , but noticeable , and varied with airmass . distortion maps extracted from the calibration images at airmasses similar to the science images provided residual errors of @xmath39 with respect to the usno - b catalog . for the lris blue - channel , the photometric zeropoint was found to be @xmath40 with an airmass term @xmath41 . during the observations , the lris red - channel detectors were plagued by electronics difficulties causing charge transfer inefficiencies that were dependent on source flux . this effect was found to be as much as @xmath42 in engineering images . as such , we adopt a 10% uncertainty on the zeropoint of this detector , and find , @xmath43 . the red - airmass term was found to be @xmath44 . the conversions between the lris r and g magnitudes and the r and g mega - pipe magnitudes were found to be @xmath45 and @xmath46 the relatively large correction between @xmath47 and @xmath48 is caused by the lris beam - splitter configuration which , in the configuration we utilized , truncated the red side of the @xmath47 filter while preserving the throughput across the @xmath49 filter . to facilitate the combination of the discovery and follow - up images , all photometry needed to be converted to a single system . the sloan system @xcite was chosen . the conversion from megaprime magnitudes to the sloan system are @xmath50 and @xmath51 from equations [ eq : color_megasubaru ] , [ eq : r_megakeck ] , and [ eq : g_megakeck ] and the conversions presented above , we find that the conversions from the subaru and keck photometry to the sloan system are @xmath52 @xmath53 and @xmath54 using equations [ eq : r_kecksloan_r]-[eq : r_subarusloan_r ] , all photometry have been converted to the sloan system . the photometry presented in table [ tab : objects ] are the weighted averages of all measurements presented here . when no @xmath55 colour was available , the average color of all those objects that did receive follow - up , @xmath56 was used to convert to the sloan survey . as the range of keck colors for our sample is @xmath57 , we estimate the error in the conversion caused by using this average color is @xmath58 magnitudes , negligibly small . in figure [ fig : colour - inc ] we present the @xmath59 color versus inclination for those objects with follow - up observations . from this figure , it can be seen that our sample exhibits similar behaviour to the kuiper belt objects in the minor planet center analyzed by @xcite . that is , there is a correlation with colour and inclination ; low inclination objects are typically redder than higher inclination objects . as both our colours and inclinations are significantly more inaccurate than those in the sample analyzed by @xcite , a repeat of the same analysis on our data is not warranted , as no more insight would be gained than by visually comparing our sample to theirs ( see figure 1 of @xcite ) . as found by @xcite , their seems to be some variation in the kuiper belt size distribution with orbital excitation . more specifically , they defined two populations , the _ cold _ sample , which contains those objects with heliocentric distances , @xmath60 au and inclinations , @xmath61 , and the _ excited _ population as all those kuiper belt objects in the same distance range , but with large inclinations . @xcite found that for bright objects @xmath62 , the _ cold _ population had a much steeper size distribution than the _ excited _ population . this result was found from a sample of kuiper belt objects compiled from a large number of different surveys , providing a result possibly affected by calibration errors and sky density variations . @xcite analyzed the results of the deep ecliptic survey , and found similar results . that is , that the cold classical objects have a steeper size distribution than the resonant , or excited populations . this result however , was drawn from a survey without a calibrated detection efficiency , making the result untrustworthy . @xcite performed a similar analysis as @xcite , but with additional survey data . their findings were similar , ie . , that the hot population size distribution is shallower than that of the cold population . but much like @xcite , they considered data from multiple surveys and may have been affected by the same calibration errors and sky density variations . @xcite reanalyzed the survey data considered in @xcite along with new observations , and introduced a technique which accounted for possible errors induced from combining the results of different surveys when measuring the size distribution . they found that the apparent differences between the hot and cold populations could not be disentangled from possible effects caused by the combination of different surveys . when correctly accounting for these effects , no statistical difference between the hot and cold population was found . population differences as large as those inferred by @xcite might exist , but their presence could not be discerned from the available data sets . the survey we present here , has sufficient depth , and number of detections to probe the differences in size distributions of the hot and cold populations alone , without the need for other data . by doing so , we can ensure that the biases that may be introduced from the use of multiple datasets are avoided here providing a clean and reliable test for any differences between the hot and cold populations of the kuiper belt . we want to test the findings of @xcite . to that end , we define subsets of our detections , which are similar to those considered by @xcite . they are defined as follows : 1 . we define _ tno _ sample , as all the detections in our sample . @xcite we define the _ cold _ population as those objects with @xmath60 au and @xmath61 . we also further refine this , and define the _ _ cold__@xmath63 sample as only those sources from the _ cold _ sample who received follow - up observations in addition to the discovery observations . we define the _ _ hot__@xmath63 and _ hot _ samples as those objects with @xmath60 au and @xmath9 , who have and have not received follow - up observations respectively . we define the _ close _ population , as those objects with @xmath64 . we do not make any distinction about follow - up , as the majority of the _ close _ population was not observed beyond the discovery observations . we note here that the exact inclination division between the _ cold _ and _ hot _ populations is uncertain . measurements of the inclination distributions of the _ hot _ and _ cold _ samples suggest that the vast majority of cold classical objects will have inclinations less than @xmath65 @xcite . analysis of the optical colour distribution of kuiper belt objects reveals a correlation of colour with inclination ( * ? ? ? * for a recent example , see ) that suggests there is a division between low - inclination red objects , and low - inclination neutral objects at @xmath66 . given the uncertainty of the inclination division that separates the _ cold _ and _ hot _ populations , we consider the full range of inclination divisions testable by our observations and primarily discuss the historical division of @xmath67 unless otherwise stated . the _ close _ population is a mix of objects in mean - motion resonances with neptune , the centaurs , and the scattered disk objects . on the other hand , the _ hot _ population consists of hot classical , resonant , and scattered disk objects . while the exact fractional mix of each of these populations can not be determined from our observations , the minor planet center can guide us as to which populations dominate these subsamples . analysis of the objects in the minor planet center , near opposition in late october , within a few degrees of the ecliptic , reveals 13 objects satisfying @xmath64 au , 6 of which are plutinos , and 36 objects satisfying @xmath68 au and @xmath9 , of which only a few are plutinos , with the majority being hot classical and scattered disk objects . as the relative mixes of each dynamical class are different for the _ hot _ and _ close _ subsamples , any differences detected in the size distributions of the subsamples can provide insight into the size distributions of the underlying populations . as the kuiper belt objects we detect are not resolved , we have no measure of their sizes , and therefore can not measure the size distribution of our sample directly . rather , we must use an in - direct technique , and infer the size distribution from the luminosity function . brightward of @xmath69 , the luminosity function is well represented by a power - law . that is , the differential surface density of objects with magnitude @xmath70 per square - degree is given by @xmath71 where @xmath72 is the logarithmic slope of the power - law , and @xmath73 is the magnitude at which there is one object per square degree with that magnitude , or brighter . recent measurements suggest that , for the _ tno _ sample , @xmath74 and @xmath75 in the r - band @xcite . it is worth noting that the luminosity function has a break , or transition from the steep slope for large objects , to a shallower distribution for fainter sources . the transition occurs around @xmath76 , which is at the tail end of our survey sensitivity . as such , the break will have little to no effect on our observations , and we ignore it in our analysis . it can be shown that if the size distribution of a population is a power - law , with slope @xmath77 , ie . , @xmath78 , then the luminosity function of that population is given by equation [ eq : lf ] , with slope @xmath79 . as such , the slopes of the size distributions of each population can be determined from their luminosity functions ( see * ? ? ? * for a thorough discussion of this technique ) . the fits of equation [ eq : lf ] is done with a bayesian maximum likelihood technique . specifically , the likelihood @xmath80 is maximized over the luminosity function slope and normalization parameters , @xmath72 and @xmath73 . here @xmath81 is the areal coverage of our survey , @xmath82 is the set of magnitudes for our discovered objects , with @xmath83 the magnitude of object @xmath84 . @xmath85 is a functional representation of the uncertainty in the magnitude @xmath83 . as standard practice , we adopt a gaussian representation , with widths equal to the uncertainty in the observed magnitude of each source . to avoid potential errors induced at low detection efficiencies , we consider only those sources with probability of detection greater than 50% and truncate our detection efficiency below this point . @xcite has demonstrated that truncating the efficiency at 50% as we do here causes the best - fit slope @xmath72 to appear roughly 3 - 5% steeper than in reality . this effect however , is much smaller than the uncertainty in the best - fit slopes we present . for the _ _ cold__@xmath63 and _ _ hot__@xmath63 samples , our follow - up is limited to those sources with @xmath86 . thus @xmath37 is our magnitude cut for the _ _ cold__@xmath63 and _ _ hot__@xmath63 samples . in equation [ eq : lf ] we utilize only the magnitudes from discovery observations alone , as utilizing the data from other filters could possibly introduce biases into the results . we note however , that when utilizing all available flux measurements , the change in the results was significantly smaller than the uncertainties in the best - fit parameters . to test the quality of the best - fits , we utilize the anderson - darling statistic , @xmath87 where @xmath88 is the cumulative probability of detecting an object with magnitude @xmath89 , and @xmath90 is the cumulative distribution of detections . we calculate the probability , @xmath91 of finding a value , @xmath92 , larger than that of the observations , @xmath93 , given the best fit - parameters by bootstrapping the statistic , ie . , randomly drawing a subsample of objects from the best - fit power - law , with number equal to that detected in a particular sample , fitting equation [ eq : lf ] to that random sample , and computing @xmath92 . values of @xmath91 near 0 indicate that the functional form is a poor representation of the data . the luminosity functions of the _ tno _ , _ cold _ , _ hot _ , and _ close _ samples are presented in figure [ fig : lf ] . the results of the fits of equation [ eq : lf ] to the various subsamples are shown in figure [ fig : contours ] , and table [ tab : best - fits ] . as can be seen from figures [ fig : lf ] and [ fig : contours ] , the _ _ cold__@xmath63 exhibits a much steeper luminosity function , with slope @xmath94 than the _ _ hot__@xmath63 population which has slope @xmath95 . the slopes of these two populations differ at more than the 1-@xmath96 level . both of these samples are well described by power - laws , as exhibited by their anderson - darling statistics ( see table [ tab : best - fits ] ) . similar slopes within the uncertainties are found from the _ hot _ and _ cold _ samples . but these values are less trustworthy as the inaccurate inclinations of those objects which did not receive follow - up cause an uncertain amount of mixing between the two subsamples . in addition , the _ close _ sample exhibits virtually the same luminosity function slope , @xmath97 as the _ _ hot__@xmath63 ( and _ hot _ ) sample . as the _ close _ and _ hot _ subsamples are made up of different fractions of the centaur , resonant , hot classical , and scattered populations , these results suggests that all of these populations exhibit equally shallow luminosity functions . if one or more of the populations had a luminosity function as steep as the _ cold _ population , it would likely be detectable as a difference in @xmath72 between the _ hot _ and _ close _ populations , the steeper being the subsample with the greatest fraction of the steep dynamical population . this however , is not apparent . indeed , the slope does not change when considering the _ hot _ and _ close _ populations together , resulting in @xmath98 . the luminosity function of _ tno _ sample has a slope @xmath99 , which is significantly shallower than the slope @xmath100 found from other recent measurements in the same magnitude range @xcite . this surprising result can be understood relatively simply . the _ tno _ sample , is a mixture of the _ hot _ , and _ cold _ populations , which our results have shown have significantly different slopes . as such , we expect the _ tno _ sample to have a slope bounded by the slopes of the _ hot _ @xmath101 and _ cold _ @xmath102 populations . our observations were made at a ecliptic longitude where plutinos , which are primarily members of the hot population , are preferentially at perihelion ( see above ) . as such , we expect a greater fraction of plutinos detected in our observations , than at other longitudes where plutinos are not at perihelion . as the _ hot _ population has a shallower luminosity function than the _ cold _ population , we expect to see a flatter luminosity function for the _ tno _ sample at this longitude than at others . this simple reasoning can account for the variation in luminosity function slopes seen in past surveys ( see * ? ? ? * ) . along with the size distribution differences between the _ hot _ and _ cold _ populations , the colour distribution ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * and see figure [ fig : colour - inc ] ) , inclination distribution @xcite , binary fraction @xcite and lack of large objects on dynamically cold orbits @xcite suggest that these two populations are different . there is no clear evidence however , for a clean separation in inclination between the two . rather , it is more likely that a smooth gradient exists , and as such , the historical division of @xmath103 chosen to separate these two populations is relatively arbitrary . as such , we measured the luminosity function slopes while varying @xmath104 . the results are presented in figure [ fig : alpha - inc ] . as can be seen , the _ cold _ and _ hot _ population slopes become significantly different for @xmath105 . when considering all objects , not just those with follow - up , the slope of the _ hot _ population seems to become a constant , with @xmath106 for @xmath107 . this suggests that above @xmath108 the fraction of the population with a steep size distribution is insignificant , and the observed objects come almost entirely from the population with a shallow size distribution . as the uncertainty on @xmath72 is still quite large , these observations however , are insufficient to confirm such a hypothesis . to test the significance of the apparent difference in luminosity functions of the _ hot _ and _ cold _ populations , we utilize two different statistical tests . we first utilize the kuiper - variant of the kolmogorov - smirnov test as a non - parametric measure of the significance in the observed difference between two populations @xcite . we randomly bootstrapped from the cumulative luminosity function of the _ cold _ sample , a sample of objects equal in number to the _ hot _ sample . we then calculated the ks - statistic of this random sample compared to the _ cold _ sample . this process was repeated , and from this the probability of finding a simulated ks - statistic as large or larger than that found from the actual _ hot _ population was determined . we found that the two populations were drawn from separate parent populations at the 80 - 90% significance levels for inclination divisions @xmath109 . while the ks - test provides a non - parametric test , a better measure of the significance difference of the _ hot _ and _ cold _ populations can be found with the knowledge that the two populations are well described by power - laws . to that end , we also utilized the anderson - darling statistic , as done in @xcite . following similar procedures as above , by randomly sampling from the best - fit power - law of the _ _ cold__@xmath63 sample we determined the probability of finding a simulated anderson - darling statistic as large or larger than that found from the actual _ _ hot__@xmath63 population as compared to the observed _ _ cold__@xmath63 sample . we found that the two populations were drawn from separate parent populations at greater than the 3-@xmath96 significance for inclination divisions @xmath109 . the observed luminosity functions imply size distribution slopes @xmath110 , @xmath111 , and @xmath112 for the _ cold _ , _ hot _ , and _ close _ populations . the consistency of the _ hot _ and _ close _ populations leaves us to consider the two populations together , implying a slope for the combined population of @xmath113 . though the results presented by @xcite and @xcite were possibly affected by sky density variations and calibration errors , we find similar conclusions , ie . , the _ cold _ population size distribution _ is _ steeper than the _ hot _ population size distribution . given the similarity in these repeated measurements of the kuiper belt luminosity function , the results stand confident . while over the magnitude range of our observations , both the _ hot _ and _ cold _ populations are well described by power - law models , other works have put into question whether the power - law behaviour extends to all brightnesses . the fact that the luminosity function of the _ tno _ population exhibits a roll - over at magnitudes @xmath114 is well accepted . the results of @xcite however , suggest that the power - law behaviour of the _ cold _ population breaks down for the brightest objects . specifically they find that at the bright - end , the _ cold _ population luminosity function has a steeper slope than at fainter magnitudes . similarly , @xcite suggests that the _ hot _ and _ cold _ populations are intimate mixtures of two primordial populations , one with a steep size distribution , and one with a shallow size distribution . in such a scenario , both populations should exhibit shallow slopes for the largest objects , then turn - up to a steeper slope for smaller objects , before finally breaking to a shallow slope at some size where collisional processing has dominated . to that end , we test the power - law behaviour of both the _ hot _ and _ cold _ populations . for the _ cold _ population , a simple extrapolation of the best - fit luminosity function suggests that the brightest object on the sky in this population should have a magnitude of @xmath115 . the entire low - latitude kuiper belt has been surveyed for objects to a limiting magnitude of @xmath116 @xcite , and the brightest known _ cold _ object has @xmath117 , nearly a full magnitude fainter . this observation confirms the results of @xcite . namely , that the power - law behaviour of the _ cold _ population luminosity function can not extend to the brightest targets . rather the power - law must be truncated . the exact behaviour can not be determined from the observations we present here . the suggestion by @xcite that the cold population size distribution will have a shallow slope for the largest objects , and a turn - up to the observed steep is excluded by this result . similarly , extrapolating the best - fit power - law luminosity function of the _ hot _ population suggests that this population is well described by a power - law to its brightest objects . indeed , when excluding the largest objects which are known to have different albedos than the objects observed here @xcite the absolute magnitude distribution of the biggest objects has a slope compatible with our observations @xcite . to test the exactness of a power - law requires combination of other survey data sensitive to the brightest _ hot _ members . as well this analysis should include modelling of the latitude distribution of the _ hot _ population . the fact however , remains that the _ hot _ population has a flat size distribution over all observable sizes . the behaviour of the _ hot _ population luminosity function at the faint end is uncertain ; the shallow slope of @xmath118 for @xmath119 is compatible with the lack of faint detections of the _ hot _ population in fainter surveys , eliminating the need for a break at the faint end . if however , the _ hot _ population slope is as steep as the upper limits of our confidence interval , a break at magnitudes fainter than @xmath120 is still required . the existence of a break in the _ hot _ population will only be confirmed by an off - ecliptic survey sensitive to objects with @xmath121 . the turn - up , as proposed by @xcite would produce an abundance of faint objects in our observed _ hot _ population . we can test this utilizing the anderson - darling statistic . we generate a fake _ hot _ subsample from a shallow power - law with the observed _ hot _ population slope that turns up to the best - fit observed slope of the _ cold _ population , @xmath122 , at some magnitude @xmath123 . we then fit a power - law to that sample , and calculate the anderson - darling statistic . this is repeated , and the probability of finding a statistic worse than the observations is calculated versus the value of @xmath123 . the strength of this test is weakened by the sparse sampling of high - latitude sources in our dataset . we still however , eliminate such a turn - up to the slope of the _ cold _ population for @xmath124 ( r ) at the 2@xmath125 level and @xmath126 at the 3-@xmath96 level . this is incompatible with the assertion of @xcite who suggest that the _ hot _ population should have the same slope as the _ cold _ population over an absolute magnitude range of @xmath127 , corresponding to r magnitudes , @xmath128 . the significance of the test results over the entire magnitude range however , prevents this result from being iron clad . clearly , the possibility of a turn - up ( or down ) must be tested from a survey which detects a large sample of high - latitude objects . the difference in size distribution slopes of the _ hot _ and _ cold _ populations suggest very different histories for these two groups . interestingly , the shallow slope of the _ hot _ distribution is compatible with a heavily collisionally processed population , either one that has reached collisional equilibrium , or in which the largest objects are fragments of even larger , disrupted primordial bodies @xcite . for collisional evolution to completely reshape the size distribution of the _ hot _ population would require extremely high collision rates , seemingly incompatible with plausible protoplanetary disk densities , and formation scenarios @xcite . rather it is likely that this slope is the result of the accretionary processes that formed the _ hot _ population . unlike the _ hot _ population , the steep slope of the _ cold _ population is entirely incompatible with a collisionally evolved distribution . this slope must be the result of accretionary processes as well . the shallower slope of the _ hot _ population , and the fact that the largest objects of the _ hot _ population are larger than those of the _ cold _ population , implies that the _ hot _ population achieved a more advanced stage of accretion than the _ cold _ population @xcite . the incompatibility of their slopes implies that these two populations underwent different accretion scenarios . it is possible , that the _ hot _ population underwent a longer duration of accretion than the _ cold _ population . to halt accretion for the _ cold _ population would require excitation and mass depletion with some yet unseen perturber beyond the outer edge of the kuiper belt . the past existence of one or more planetary embryos in the kuiper belt region is a possible source of this excitation , and has been proposed to explain some of the kuiper belt dynamical features @xcite . another possibility is that the _ hot _ population underwent accretion in a more dense region of the protoplanetary nebula . this would result in more rapid accretion than compared to the _ cold _ population , allowing the _ hot _ population to grow to larger sizes before the process was halted . under this scenario , the _ hot _ and _ cold _ populations could then have similar accretion timescales . as currently favoured formation scenarios suggest the _ hot _ population was scattered from a region closer to the sun than their current locations @xcite , where the protoplanetary disk might have been more dense , this scenario seems likely . however the _ hot _ and _ cold _ populations came about , their incompatible size distributions imply different accretion scenarios . once accretion was finished , these objects were excited and emplaced onto their current orbits , creating the architecture of the current kuiper belt . given the dynamically cold nature of the _ cold _ population , it seems plausible that this population formed in - situ . what ever the mechanism(s ) ultimately responsible for the belt , our results show that little mixing between the _ hot _ and _ cold _ populations has occurred . there is a striking similarity in the size distributions of the _ cold _ population , and the jupiter trojans . both have similarly steep slopes , ( @xmath129 and @xmath130 ) , and both exhibit breaks to shallower slopes at roughly the same object diameter @xcite . this result argues strongly against the trojan formation mechanism of the so - called nice model in which the trojan populations and the hot and cold kuiper belt populations are all scattered into their current regions from the same primordial disk population @xcite . if this scenario were true , @xcite has shown that over the magnitude range of our observations , the hot population should have the same slope as that observed for large jupiter trojans . using the anderson - darling statistic , we tested the probability of the _ hot _ sample being drawn from a luminosity function with the slope of the jupiter trojan luminosity function , @xmath131 . we found that the _ hot _ sample could not be drawn from the jupiter trojan population at greater than the 3-@xmath96 significance . as the size distributions of these two populations in the size range considered have not evolved significantly since they were emplaced in their current regions @xcite , we must conclude that the hot and jupiter trojan populations must have different progenator populations . this result and the lack of mixing between the hot and cold populations is difficult to reconcile with the nice model . it seems likely the hot , cold , and trojan populations formed by separate means . if this is true , the similarities between the cold and trojan populations are quite a coincidence . another similarity is seen between the _ hot _ population ( @xmath132 ) and the jupiter family comets , who exhibit a size distribution slope of @xmath133 albeit over a smaller size range , @xmath134 km @xcite . this resemblance is interesting as one likely source of jupiter family comets are scattered disk objects - members of the _ hot _ population - which have fallen into the inner solar system under gravitational perturbations from the gas - giants , suggesting that the scattered disk size distribution might be a power - law for @xmath135 km . we consider this possibility here . while the observations we present do not measure the size distribution of the scattered disk directly , the similarity in slopes between the _ hot _ and _ close _ subsamples implies that the scattered disk can not have a size distribution significantly different than that of the _ hot _ sample as a whole . additionally we can not determine the total number scattered disk objects from our observations . rather , we turn to other estimates which suggest that , if their size distribution is a power - law with slope of @xmath136 , then there are roughly @xmath137 scattered disk objects with @xmath138 km ( * ? ? ? * ; * ? ? ? * schwamb , personal communication ) . simulations by @xcite suggest that if the scattered disk is the sole source of the jupiter family comets , then there must be at least @xmath139 objects in that population to account for the current flux of jupiter family comets through the inner solar system , implying that either the scattered disk is not the sole source of the jupiter family comets , or that the extrapolation of a power - law to @xmath140 km is unreasonable ; it must be that the size distribution of scattered disk objects steepens significantly in the @xmath141 km range . further observations are required before this feature can be detected . we have performed an ecliptic survey , and have detected 88 kuiper belt objects with magnitudes @xmath142 . a subset of these objects have received additional follow - up observations allowing us to accurately determine their inclinations and distances . using these data , we have measured the size distribution of the _ hot _ and _ cold _ subsamples , historically defined as those objects with inclinations above and below @xmath1 respectively . this measurement , which is independent of any previous observations has confirmed that the _ cold _ population has a much steeper luminosity function , with slope @xmath94 , than the _ hot _ population , with slope @xmath3 . the observed luminosity functions imply different size distributions for the _ hot _ and _ cold _ populations . the size distribution slopes of the the two subsamples are @xmath110 and @xmath111 . in addition , we have found that the _ close _ population , which is defined as those objects with heliocentric distance , @xmath143 have the same luminosity function slope as the _ hot _ subsample , demonstrating that these two share a similar size distribution . in addition , our findings suggest that the dynamical populations which make up both the _ hot _ and _ close _ populations must all have similar size distributions . the primary consequence of these findings is that the _ cold _ population , which consists primarily of cold classical kuiper belt objects , has a separate , and distinct accretion history from the _ hot _ population , requiring either potentially different accretion timescales for the two populations , or formation in different locations of the protoplanetary disk . these observations reveal the similarities in the size distributions between the subsamples of the kuiper belt and the jupiter family comets and trojans , suggesting a connection between the formation and subsequent evolution of the small body populations of the outer solar system . based in part on data collected at subaru telescope , which is operated by the national astronomical observatory of japan . some of the data presented herein were obtained at the w. m. keck observatory , which is operated as a scientific partnership among the california institute of technology , the university of california , and the national aeronautics and space administration . this research used the facilities of the canadian astronomy data centre operated by the national research council of canada with the support of the canadian space agency . the research upon which this paper is based was supported by national aeronautic and space administration ( nasa ) grant no . nasa.000261 54 natexlab#1#1url # 1`#1`urlprefix , p. , guy , j. , regnault , n. , pain , r. , aubourg , e. , balam , d. , basa , s. , carlberg , r. g. , fabbro , s. , fouchez , d. , hook , i. m. , howell , d. a. , lafoux , h. , neill , j. d. , palanque - delabrouille , n. , perrett , k. , pritchet , c. j. , rich , j. , sullivan , m. , taillet , r. , aldering , g. , antilogus , p. , arsenijevic , v. , balland , c. , baumont , s. , bronder , j. , courtois , h. , ellis , r. s. , filiol , m. , gonalves , a. c. , goobar , a. , guide , d. , hardin , d. , lusset , v. , lidman , c. , mcmahon , r. , mouchet , m. , mourao , a. , perlmutter , s. , ripoche , p. , tao , c. , walton , n. , feb . the supernova legacy survey : measurement of @xmath144 , @xmath145 and w from the first year data set . a&a 447 , 3148 . , p. g. , campo bagatin , a. , feb . 2009 . collisional evolution of trans - neptunian populations : effects of fragmentation physics and estimates of the abundances of gravitational aggregates . pss57 , 201215 . , g. , khushalani , b. , dec . 2000 . orbit fitting and uncertainties for kuiper belt objects . aj 120 , 33233332 . , g. m. , trilling , d. e. , allen , r. l. , brown , m. e. , holman , m. , malhotra , r. , sep . the size distribution of trans - neptunian bodies . aj 128 , 13641390 . , e. , jul . 2006 . automatic astrometric and photometric calibration with scamp . in : c. gabriel , c. arviset , d. ponz , & s. enrique ( ed . ) , astronomical data analysis software and systems xv . 351 of astronomical society of the pacific conference series . . 112+ . , e. , mellier , y. , radovich , m. , missonnier , g. , didelon , p. , morin , b. , 2002 . the terapix pipeline . in : bohlender , d. a. , durand , d. , handley , t. h. ( eds . ) , astronomical data analysis software and systems xi . vol . 281 of astronomical society of the pacific conference series . . 228+ . , w. f. , durda , d. d. , nesvorn , d. , jedicke , r. , morbidelli , a. , vokrouhlick , d. , levison , h. f. , dec . 2005 . linking the collisional history of the main asteroid belt to its dynamical excitation and depletion . icarus 179 , 6394 . , w. f. , nesvorn , d. , vokrouhlick , d. , morbidelli , a. , mar . the irregular satellites : the most collisionally evolved populations in the solar system . aj139 , 9941014 . , m. e. , may 2001 . the inclination distribution of the kuiper belt . aj 121 , 28042814 . , m. e. , 2008 . the largest kuiper belt objects . 335344 . , d. r. , durda , d. d. , marzari , f. , campo bagatin , a. , gil - hutton , r. , 2002 . collisional evolution of small - body populations . asteroids iii , 545558 . , j. l. , kern , s. d. , clancy , k. b. , gulbis , a. a. s. , millis , r. l. , buie , m. w. , wasserman , l. h. , chiang , e. i. , jordan , a. b. , trilling , d. e. , meech , k. j. , feb . 2005 . the deep ecliptic survey : a search for kuiper belt objects and centaurs . ii . dynamical classification , the kuiper belt plane , and the core population . aj 129 , 11171162 . , w. c. , nov . the collisional divot in the kuiper belt size distribution . apj706 , 119129 . , w. c. , kavelaars , j. j. , jan . 2009 . the size distribution of kuiper belt objects for d gsim 10 km . aj137 , 7282 . , w. c. , kavelaars , j. j. , holman , m. j. , pritchet , c. j. , gladman , b. j. , grav , t. , jones , r. l. , macwilliams , j. , petit , j .- m . , jun . the kuiper belt luminosity function from @xmath146 . icarus 195 , 827843 . , c. i. , george , m. r. , holman , m. j. , may 2009 . a subaru pencil - beam search for m@xmath147 ~ 27 trans - neptunian bodies . apj696 , 9195 . , c. i. , holman , m. j. , jul . a subaru archival search for faint trans - neptunian objects . aj136 , 8397 . , b. , chan , c. , jun . production of the extended scattered disk by rogue planets . apjl643 , l135l138 . , b. , kavelaars , j. j. , petit , j .- m . , morbidelli , a. , holman , m. j. , loredo , t. , aug . 2001 . the structure of the kuiper belt : size distribution and radial extent . aj 122 , 10511066 . , r. s. , feb . 2003 . the origin of the kuiper belt high - inclination population . icarus 161 , 404418 . , a. a. s. , elliot , j. l. , adams , e. r. , benecchi , s. d. , buie , m. w. , trilling , d. e. , wasserman , l. h. , aug . unbiased inclination distributions for objects in the kuiper belt . aj140 , 350369 . , s. d. j. , feb . 2008 . megapipe : the megacam image stacking pipeline at the canadian astronomical data centre . pasp120 , 212223 . , r. , metcalfe , t. s. , feb . 1998 . the orbital and absolute magnitude distributions of main belt asteroids . icarus 131 , 245260 . , d. c. , sheppard , s. s. , apr . physical properties of trans - neptunian object ( 20000 ) varuna . aj123 , 21102120 . , d. c. , trujillo , c. a. , luu , j. x. , aug . 2000 . population and size distribution of small jovian trojan asteroids . aj120 , 11401147 . , s. j. , mar . planet formation in the outer solar system . pasp 114 , 265283 . , s. j. , bromley , b. c. , oct the size distribution of kuiper belt objects . aj 128 , 19161926 . , s. j. , luu , j. x. , may 1998 . accretion in the early kuiper belt . i. coagulation and velocity evolution . aj115 , 21362160 . , h. f. , morbidelli , a. , vanlaerhoven , c. , gomes , r. , tsiganis , k. , jul . 2008 . origin of the structure of the kuiper belt during a dynamical instability in the orbits of uranus and neptune . icarus 196 , 258273 . , h. f. , stern , s. a. , mar . 2001 . on the size dependence of the inclination distribution of the main kuiper belt . aj121 , 17301735 . , p. s. , mukai , t. , apr an outer planet beyond pluto and the origin of the trans - neptunian belt architecture . aj135 , 11611200 . , r. , oct . the origin of pluto s peculiar orbit . nature365 , 819+ . , s. , komiyama , y. , sekiguchi , m. , okamura , s. , doi , m. , furusawa , h. , hamabe , m. , imi , k. , kimura , m. , nakata , f. , okada , n. , ouchi , m. , shimasaku , k. , yagi , m. , yasuda , n. , dec . subaru prime focus camera suprime - cam . pasj54 , 833853 . , a. , levison , h. f. , bottke , w. f. , dones , l. , nesvorn , d. , jul . considerations on the magnitude distributions of the kuiper belt and of the jupiter trojans . icarus 202 , 310315 . , a. , levison , h. f. , gomes , r. , 2008 . the dynamical structure of the kuiper belt and its primordial origin . the solar system beyond neptune , pp . 275292 . , a. , levison , h. f. , tsiganis , k. , gomes , r. , may 2005 . chaotic capture of jupiter s trojan asteroids in the early solar system . nature435 , 462465 . , d. , vokrouhlick , d. , jun . chaotic capture of neptune trojans . aj137 , 50035011 . , d. p. , greenberg , r. , aug . steady - state size distributions for collisional populations : analytical solution with size - dependent strength . icarus 164 , 334345 . , a. h. , kavelaars , j. , apr . pencil - beam surveys for trans - neptunian objects : limits on distant populations . arxiv e - prints . , n. , lacerda , p. , jewitt , d. , nov . 2008 . color - inclination relation of the classical kuiper belt objects . aj136 , 18371845 . , w. h. , 2002 . numerical recipes in c++ : the art of scientific computing . numerical recipes in c++ : the art of scientific computing by william h. press . xxviii , 1,002 p. : ill . ; 26 cm . includes bibliographical references and index . isbn : 0521750334 . , h. e. , ofek , e. o. , wenz , m. , sari , r. , gal - yam , a. , livio , m. , nelan , e. , zucker , s. , dec . 2009 . a single sub - kilometre kuiper belt object from a stellar occultation in archival data . nature462 , 895897 . , j. a. , tucker , d. l. , kent , s. , richmond , m. w. , fukugita , m. , ichikawa , t. , ichikawa , s .- i . , jorgensen , a. m. , uomoto , a. , gunn , j. e. , hamabe , m. , watanabe , m. , tolea , a. , henden , a. , annis , j. , pier , j. r. , mckay , t. a. , brinkmann , j. , chen , b. , holtzman , j. , shimasaku , k. , york , d. g. , apr . 2002 . the ugriz standard - star system . aj 123 , 21212144 . , j. , grundy , w. , brown , m. , cruikshank , d. , spencer , j. , trilling , d. , margot , j l . , 2008 . physical properties of kuiper belt and centaur objects : constraints from the spitzer space telescope . . 161179 . , d. c. , noll , k. s. , feb . detection of six trans - neptunian binaries with nicmos : a high fraction of binaries in the cold classical disk . aj131 , 11421148 . , s. a. , sep . on the collisional environment , accretion time scales , and architecture of the massive , primordial kuiper belt . aj112 , 1203+ . , s. a. , colwell , j. e. , aug . accretion in the edgeworth - kuiper belt : forming 100 - 1000 km radius bodies at 30 au and beyond . aj114 , 841+ . , g. , fernndez , j. a. , rickman , h. , licandro , j. , jun . nuclear magnitudes and the size distribution of jupiter family comets . icarus 182 , 527549 . , s. c. , romanishin , w. , oct extremely red kuiper - belt objects in near - circular orbits beyond 40 au . nature407 , 979981 . , c. a. , brown , m. e. , feb . 2002 . a correlation between inclination and color in the classical kuiper belt . apjl566 , l125l128 . , c. a. , brown , m. e. , jun . 2003 . the caltech wide area sky survey . earth moon and planets 92 , 99112 . , c. a. , jewitt , d. c. , luu , j. x. , feb . 2000 . population of the scattered kuiper belt . apjl529 , l103l106 . , k. , malhotra , r. , nov . the scattered disk as the source of the jupiter family comets . apj687 , 714725 . , p. r. , snodgrass , c. , fitzsimmons , a. , lowry , s. c. , sep the size distribution of jupiter - family comet nuclei . in : aas / division for planetary sciences meeting abstracts . 41 of aas / division for planetary sciences meeting abstracts . pp . 23.13+ . .suprime - cam discovery field centers . coordinates are presented in the j2000 epoch . [ tab : pointings ] [ cols="<,<",options="header " , ] [ fig : pointings ] . diagram of the fields observed for moving objects during the subaru observations . these fields were chosen to avoid bright stars . the field center coordinates are presented in table [ tab : pointings ] . fig [ fig : airmass ] . extinction versus airmass for the subaru discovery observations . each point is the average difference in magnitude of bright stars between common fields ; each of the 35 discovery triplets has two data points . the line is the best - fit linear extinction law , with slope 0.08 magnitudes per unit airmass . these data demonstrate the photometric nature of the discovery observations . fig [ fig : efficiency ] . detection efficiency versus artificial source @xmath148 magnitude . error - bars are 1-sigma poisson limits . best - fit parameters of the efficiency function given by equation [ eq : efficiency ] are presented at the top of the figure . fig [ fig : eff_i ] . detection efficiency versus artificial object inclination . from top to bottom , all planted sources , and those with r@xmath149 , @xmath150 , @xmath151 . this figure demonstrates that , given a particular object magnitude , our survey was equally sensitive to all inclinations . fig [ fig : eff_r ] . detection efficiency versus artificial object heliocentric distance . from top to bottom , all planted sources , and those with r@xmath149 , @xmath150 , @xmath151 . this figure demonstrates that , given a particular object magnitude , no significant variation of detection efficiency with distance is apparent . exceptions are at the extrema of the survey ; our search was only sensitive to objects with apparent motions @xmath152 roughly corresponding to distances @xmath153 au . fig [ fig : lf ] . the differential luminosity functions of the various samples defined in section [ sec : subsamples ] offset in units of 2 for clarity . diamonds : _ tno _ . squares : _ _ cold__@xmath63 . triangles : _ _ hot__@xmath63 . circles : _ close _ . error - bars are the 1-@xmath96 poisson limits for the number of objects in each bin . lines are the best - fit power - laws , with values given in table [ tab : best - fits ] . fig [ fig : contours ] . the 1 ( solid ) , 2 ( dashed ) and 3-@xmath96 ( dotted ) likelihood contours of the fits presented in figure [ fig : lf ] and table [ tab : best - fits ] . the corresponding subsample is labeled in the top right of each panel . fig [ fig : alpha - inc ] . best - fit slopes @xmath72 of the _ cold _ ( squares ) and _ hot _ ( circles ) populations versus the inclination division @xmath104 separating the two . left is using only those objects with follow - up . right is using all observations . the 1-@xmath96 range in @xmath154 is bracketed by the two horizontal lines .

magnetic reconnection in relativistic collisionless plasmas can accelerate particles and power high - energy emission in various astrophysical systems . whereas most previous studies focused on relativistic reconnection in pair plasmas , less attention has been paid to electron - ion plasma reconnection , expected in black - hole accretion flows and relativistic jets . we report a comprehensive particle - in - cell numerical investigation of reconnection in an electron - ion plasma , spanning a wide range of ambient ion magnetizations , from the semi - relativistic regime ( ultra - relativistic electrons but nonrelativistic ions , ) to the fully relativistic regime ( both species are ultra - relativistic , ) . we investigate how the reconnection rate , electron and ion plasma flows , electric and magnetic field structures , electron / ion energy partitioning , and nonthermal particle acceleration depend on . our key findings are : ( 1 ) the reconnection rate is about 0.1 of the alfvnic rate across all regimes ; ( 2 ) electrons can form concentrated moderately - relativistic outflows even in the semi - relativistic , small- regime ; ( 3 ) while the released magnetic energy is partitioned equally between electrons and ions in the ultra - relativistic limit , the electron energy fraction declines gradually with decreased and asymptotes to about 0.25 in the semi - relativistic regime ; ( 4 ) reconnection leads to efficient nonthermal electron acceleration with a-dependent power - law index , . these findings are important for understanding black - hole systems and lend support to semi - relativistic reconnection models for powering nonthermal emission in blazar jets , offering a natural explanation for the spectral indices observed in these systems . [ firstpage ] acceleration of particles accretion , accretion discs magnetic reconnection relativistic processes bl lacertae objects : general x - rays : binaries

1612.04493

magnetic reconnection is a basic plasma process often leading to the violent release of magnetic energy through rapid rearrangement of magnetic field geometry ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? relativistic magnetic reconnection can occur when , as in many high - energy astrophysical environments , the magnetic energy density exceeds the energy density of the ambient plasma ( including the rest - mass energy density ) . this situation is characterized by a large magnetization parameter @xmath4 , where @xmath5 is the ambient reconnecting magnetic field strength and @xmath6 is the relativistic enthalpy density , the sum of the relativistic energy density @xmath7 ( including rest mass ) and the plasma pressure @xmath8 . when @xmath9 , the ( relativistic ) alfvn velocity @xmath10 approaches the speed of light @xmath11 , and magnetic reconnection proceeds in the relativistic regime @xcite , leading to ultra - relativistic bulk flows , plasma heating to relativistic temperatures , nonthermal particle acceleration to ultra - relativistic energies , etc . there are many examples of high-@xmath12 ( @xmath13 ) and medium-@xmath12 ( @xmath14 ) astrophysical environments where relativistic magnetic reconnection is believed to take place , including pulsar magnetospheres , pulsar winds , and pulsar wind nebulae ( pwn ) @xcite ] ; gamma - ray bursts ( grbs ) @xcite ; magnetospheres of magnetars @xcite ; coronae of accreting black holes ( bhs ) in both galactic x - ray binaries ( xrbs ) and active galactic nuclei ( agn ) @xcite , as well as in relativistic jets powered by these systems , including blazars @xcite . many of these systems are observed to emit nonthermal broadband radiation ( characterized by a power - law energy spectrum ) . whether quiescent or in the form of bright , impulsive x - ray and @xmath15-ray flares with rapid time variability , such nonthermal radiation is thought to result from energetic electrons ( and positrons , if present ) emitting via synchrotron and/or inverse compton ( ic ) mechanisms ; the energy spectrum of emitting particles is therefore inferred to be nonthermal as well ( a power law , sometimes extending over orders of magnitude in energy ) . this highlights the importance of understanding the physical mechanisms of non - thermal particle acceleration operating in collisionless energy - conversion processes , of which magnetic reconnection is a leading candidate [ see , e.g. , @xcite for a review ] , taking place in relativistic astrophysical plasmas . nonthermal particle acceleration in plasma processes has been studied within fluid ( e.g. , magnetohydrodynamic ) frameworks by tracking distributions of passive test particles @xcite . however , such studies , though valuable in many respects , can not account for the effects of nonthermally accelerated particles acting back on the electromagnetic fields and fluid motions . these self - consistent effects are likely to be important , especially when the particle energy distributions @xmath16 have power - law indices @xmath17 ( as in ultra - relativistic electron - positron reconnection : * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , indicating that most of the particle kinetic energy resides in particles with energies much higher than average . because of this , a self - consistent study of nonthermal particle acceleration inherently requires a kinetic approach that evolves the entire particle distribution . a powerful computational tool often employed in numerical studies of kinetic plasma processes is particle - in - cell ( pic ) simulation . in particular , there have been a large number of pic studies of relativistic collisionless reconnection in electron - positron pair plasmas in the high-@xmath12 limit in the past few years @xcite . these studies have concentrated on pair plasmas for two main reasons . first , some of the most important astrophysical applications of relativistic reconnection involve pair plasmas , most notably pulsar magnetospheres , pulsar winds , and pwn . the second reason is the relative simplicity of pair plasma reconnection due to the absence of the scale separation , present in conventional electron - ion plasmas , between the important electron and ion kinetic scales such as the larmor radii , @xmath18 and @xmath19 , and the collisionless skin depths , @xmath20 and @xmath21 . computational advances have allowed the more recent of these pic studies ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) to add substantially to our understanding of relativistic pair - plasma reconnection by clearly demonstrating efficient nonthermal particle acceleration and mapping out its quantitative characteristics , such as the power - law index and high - energy cutoff , as functions of the system parameters . however , _ electron - ion _ ( _ ei _ ) plasmas predominate in many relativistic high - energy astrophysical environments in which magnetic reconnection is believed to be important ( throughout this paper , the ions are protons ) . for example , _ ei_-plasma reconnection has been invoked as the main mechanism responsible for electron heating and subsequent high - energy radiation in many of the systems mentioned above , e.g. , in accretion disk coronae ( adc ) in black hole systems , in gamma - ray flares in relativistic blazar jets ( and agn jets in general ) , including ultra - rapid tev flares , and for powering prompt emission in grbs . despite its importance , and in stark contrast to non - relativistic reconnection , relativistic reconnection in _ ei _ plasmas has so far received relatively little attention ; pic studies of _ ei _ reconnection in the relativistic regime have started only recently @xcite . when considering ultra - relativistic reconnection in electron - ion plasmas , it is important to note that the key fundamental plasma time- and length - scales , such as the plasma and cyclotron frequencies , the collisionless skin depth , and the larmor radius , become independent of the particle rest - mass and instead depend only on the average particle energy in this ultra - relativistic limit . therefore , if both electrons and ions become ultra - relativistic as a result of magnetic energy release in a reconnection process , then any differences in the dynamical behavior of the two species disappear ( in the absence of radiative losses ) . correspondingly , one _ a priori _ expects ultra - relativistic _ ei_-plasma reconnection to become essentially similar to ultra - relativistic pair - plasma reconnection , as was indeed confirmed in a recent study by @xcite . importantly , however , in many real astrophysical plasmas , one finds a more complex and interesting situation the _ semi - relativistic _ regime where typical electrons are ultra - relativistic ( with the average lorentz factor @xmath22 ) but ions are sub - relativistic ( @xmath23 ) ; here , @xmath24 is the average kinetic energy of electrons / ions , excluding the rest - mass energy . whether typical ions in the reconnection outflow are expected to be relativistic is governed by the upstream ` cold ' magnetization @xmath25 , where @xmath26 is the reconnecting magnetic field and @xmath27 is the ion density upstream of the reconnection region ; this quantity is basically twice the ratio of the available magnetic energy density to ion rest - mass energy density . magnetic reconnection in the semi - relativistic regime , characterized by @xmath28 , may be of great importance in , e.g. , accreting black - hole coronae and agn ( including blazar ) jets . because of the very large separation between the electron and proton rest masses , @xmath29 , the range of applicability of this regime is actually quite broad : it essentially covers three orders of magnitude in @xmath0 and hence in the expected average dissipated energy per particle , and correspondingly several orders of magnitude in the energy of emitted photons ( since @xmath30 ) . it is thus critical to understand this new , so far unexplored , reconnection regime , as well as the transition from this semi - relativistic regime to the pair - plasma - like regime of ultra - relativistic ions as @xmath0 increases from @xmath31 to @xmath2 . we note that , in contrast to our present study , @xcite focused solely on the ultra - relativistic regime , @xmath32 , while @xcite studied greatly reduced mass ratios ( @xmath33 ) with @xmath34 and mildly relativistic electrons , as well as with @xmath35 and barely relativistic electrons . here we report the results of our systematic investigation of collisionless _ ei_-plasma magnetic reconnection , using the real mass ratio @xmath36 , from the semi - relativistic through the ultra - relativistic regime covering the entire range from @xmath37 to @xmath38using the relativistic pic code vorpal / vsim @xcite . we focus on several important quantitative characteristics of the reconnection process for large system sizes @xmath39such as the reconnection rate , energy partitioning between electrons and ions , hall effect signatures , power - law slopes and cutoff energies of nonthermal particle populations and study how they scale with the upstream ion magnetization @xmath0 . this paper is organized as follows . in [ sec - sims ] we describe our numerical simulations , and then present the results in [ sec - results - dynamics][sec - results - kinetics ] . in [ sec - results - dynamics ] we discuss the basic fluid dynamics and electrodynamics of reconnection reconnection rate , plasma flows , and hall effect signatures . we then describe our findings regarding the energetics and kinetic aspects of the reconnection process in [ sec - results - kinetics ] , namely , the energy partitioning between electrons and ions , and nonthermal particle acceleration . after that , in [ sec - astro ] we discuss some of the most intriguing astrophysical applications of our results , in particular , as a possible natural explanation for ultra - rapid tev flares in agn / blazar jets [ with connections to the minijet model of @xcite ] , as well as the non - flaring emission at lower energies . finally , in [ sec - conclusions ] we summarize our conclusions . our simulations all two - dimensional ( 2d)of relativistic electron - proton plasma reconnection presented here are performed with the explicit electromagnetic particle - in - cell ( em - pic ) vorpal / vsim code @xcite , using a fully electromagnetic field solver @xcite and a relativistic boris particle - push @xcite with divergence - preserving current deposition @xcite . the simulations are initialized with an electron - proton plasma , using the real mass ratio @xmath40 , in a double - periodic relativistic harris equilibrium , which is described below in detail . our investigation focuses on the effects of varying ion magnetization @xmath0 ( defined below ) ; besides describing the setup below , we also relate the initial simulation parameters to @xmath0 . the fundamental aspects of the initial configuration are the reversing magnetic field and ambient ( background ) plasma . to allow the use of easily - implemented periodic boundary conditions , it is convenient to simulate two field reversals ( the so - called double - periodic configuration ) . as is commonly done , we add a relativistic harris current layer @xcite a non - uniform , drifting plasma component at each field reversal to provide current and pressure to balance the reversing magnetic field . the simulations are conducted in a 2d rectangular computational box with a regular cartesian grid and periodic boundary conditions in both directions . the initial double - reversing magnetic field configuration ( cf . figs . [ fig : aztimeevolutionsigp1 ] and [ fig : aztimeevolutionsig10 ] , top left panel ) , before perturbation ( described below ) , is : @xmath41 where @xmath26 is the asymptotic upstream reconnecting magnetic field strength , @xmath42 and @xmath43 are the mid - planes of the lower and upper current sheets where @xmath44 , and @xmath45 is the initial layer half - thickness , determined to be consistent with the other harris sheet equilibrium parameters . thus , our coordinate system places @xmath46 in the direction of the reconnecting magnetic field ( the plasma outflow direction ) , and @xmath47 in the direction of reconnected magnetic field ( the inflow direction , perpendicular to the current layers ) , with the ignorable @xmath48 direction parallel to the initial harris current . for simplicity , we limit this study to the case without a guide magnetic field component ( @xmath49 , initially ) . also , unlike most previous double - periodic simulations , our present simulations use a tall box , @xmath50 ( where @xmath39 and @xmath51 are the box sizes in the @xmath46 and @xmath47 directions , respectively ) to reduce interactions between the two layers . in addition to the two concentrated harris layers , which are described in more detail below , we initially fill the entire box with a low - density uniform and stationary background plasma , described by non - drifting relativistic maxwell - jttner electron and ion distributions with equal initial electron and ion densities , @xmath52 , and temperatures , @xmath53 . we then characterize the fundamental reconnection setup through the relationships between the upstream reconnecting magnetic field @xmath26 and the parameters of this ambient background plasma . in particular , we use the initial electron and ion ` cold magnetizations ' @xmath54 and @xmath55 to relate the magnetic field energy density @xmath56 to the background electron and ion densities , @xmath52 . in addition , we relate the background plasma temperature to the upstream magnetizations : for all simulations presented here we have chosen @xmath57 , @xmath58 . these background density and temperature choices yield an initial upstream plasma @xmath59 parameter of @xmath60 , so the upstream region is magnetically dominated . because @xmath61 is roughly the available magnetic energy per particle , the ` cold ' magnetizations set the basic scale for average particle energy gain during reconnection . it is convenient to define a corresponding length scale @xmath62 , where @xmath63 are nominal relativistic electron and ion larmor radii ; thus defined , @xmath64 is the larmor radius of an ultra - relativistic electron with energy @xmath65 ( or , if @xmath2 , of an ion with energy @xmath66 ) . the corresponding time - scale is @xmath67 . because the larmor radius @xmath64 and larmor period of typical electrons ( after being energized by reconnection ) scale with @xmath0 , simulation grid cell size and timestep are set basically proportional to @xmath0 , with small adjustments as @xmath0 decreases ( because the scale separation between electrons and ions becomes larger ) . we note that @xmath64 can also be interpreted as the half - thickness of the current sheet that balances ( via ampere s law ) the jump in @xmath68 equal to @xmath69 , assuming the sheet has density @xmath70 and drift velocity @xmath71 . that is , a uniform electron current density of @xmath72 in a layer of total thickness @xmath73 would support a magnetic field discontinuity @xmath74 . linking the initial upstream temperature to @xmath0 has both advantages and disadvantages . importantly , it helps to ensure that our simulations are always properly resolved , namely , that the debye length of the background plasma , @xmath75^{1/2}=\rho_c/20 $ ] , does not become small compared to the grid cell size ( which also scales with @xmath64see below ) . however , choosing @xmath76 to avoid extreme resolution requirements has the drawback of limiting the maximum @xmath77 . the ` hot ' @xmath12 with respect to electrons is defined as @xmath78 where @xmath79 is the relativistic enthalpy density of background electrons ; for ultra - relativistic electrons , @xmath80 , and so @xmath81 in our simulations . thus @xmath82 describes the ratio between the available magnetic energy and ( roughly ) the relativistic thermal energy ( plus the negligible electron rest - mass energy ) in the upstream region ; if @xmath82 is not much larger than one , then typical electrons will not gain significant energy compared to the average ( thermal ) energy they already have . for sub - relativistic ions , however , the rest - mass dominates the relativistic enthalpy , and @xmath83 . the total @xmath77 includes the enthalpies of both electrons and ions ; for @xmath84 , we have @xmath85 . figure [ fig : sigmas ] shows how all these magnetization parameters vary with @xmath0 ( given our choice of initial background plasma temperature ) . ( the ` cold ' @xmath12 with respect to background ions ) . circles mark the values of @xmath0 for simulations described in this paper . [ fig : sigmas ] ] with the upstream parameters settled , we now turn to the description of the two initial relativistic harris current sheets that are introduced to balance the magnetic field reversals @xcite . we choose the drifting electrons and ions to contribute equally to the initial layer pressure and to drift with equal but opposite speeds ( in the simulation , or ` lab ' frame ) : for all simulations , the drift speed is @xmath86 and the peak drift - to - background plasma density ratios are @xmath87 ( all densities are specified in the lab frame ) . the remaining harris - layer parameters ( layer half - thickness @xmath45 and co - moving temperatures @xmath88 of the relativistic maxwell - jttner distributions ) are determined by ampere s law [ @xmath89 , and pressure balance [ @xmath90 . this yields @xmath91 and @xmath92 . to be consistent with the magnetic profile , the drift plasma density varies with @xmath47 as @xmath93 $ ] for layers @xmath94 . to accelerate the onset of reconnection , we add a slight ( 1% ) initial magnetic perturbation in the @xmath48-component of the vector potential ( magnetic flux function ) , so that the total initial vector potential is ( e.g. , for the lower layer ) @xmath95 b_0 \delta \left[\ln \cosh \frac{y_{cs,2}-y_{cs,1}}{2\delta } -\ln \cosh \frac{y - y_{cs,1}}{\delta } \right].\end{aligned}\ ] ] this study focuses mainly on the dynamics , energetics , and kinetics ( i.e. , nonthermal particle acceleration ) of electron - ion magnetic reconnection as one transitions from the semi - relativistic regime ( non - relativistic ions but ultra - relativistic electrons : @xmath31 , @xmath96 ) to the fully ultra - relativistic regime ( both ions and electrons being ultra - relativistic , @xmath97 ) . to accomplish this goal , we conduct an extensive parameter - space study with respect to the upstream ion magnetization @xmath0 , specifically studying simulations with @xmath98 , for system size @xmath99 . as @xmath0 varies , the key initial energy - scales and length - scales namely , @xmath100 , @xmath101 , @xmath64 , and @xmath45change in proportion to @xmath0 , as described above . we note that the background electrons are always somewhat relativistic and usually very relativistic ; even for our lowest @xmath0 of @xmath102 , we have @xmath103 , yielding an average kinetic energy nearly equal to the electron rest mass for upstream electrons and these electrons become significantly more relativistic as they enter the reconnection layer . ions , on the other hand , range from sub- to ultra - relativistic as @xmath0 increases . our main @xmath0-comparison study is conducted using the same fiducial system size @xmath104 , which is chosen to be as large as possible while remaining computationally feasible . thus , unless specifically stated otherwise , all results shown in this work are for simulations with @xmath99 . this size is sufficiently large that the number and total energy of initially - drifting particles is small compared to background particles , and dynamical quantities , such as reconnection rate , are nearly independent of @xmath39 . in addition , however , in order to examine the extent to which these fiducial sizes reach the asymptotic ` large system regime , ' we explore the effects of varying @xmath39 for two values of @xmath105 , over a range @xmath106 . the simulations run until the magnetic energy essentially stops decreasing ; for @xmath99 , reconnection ends after roughly @xmath107 for large @xmath0 and after @xmath108 for @xmath109 . the discretization parameters of our simulations are as follows . the spatial resolution is represented by the ratio of grid cell size @xmath110 to the fundamental microphysical scale @xmath64 , and the timestep is set just below the usual ( 2d ) courant - friedrichs - lewy maximum stable step , @xmath111 . for the system size @xmath99 the resolution and number of initial particles per cell ( including all species ) are set to be @xmath112 ( 256 particles per cell ) for @xmath113 , @xmath114 ( 168 particles per cell ) for @xmath115 , and @xmath116 ( 128 particles per cell ) for @xmath32 . and 1 ) , @xmath117 , computational feasibility led us to use @xmath118 ( 128 particles per cell ) ; because of their size and fewer particles , they experienced energy non - conservation greater than 1% ( 1.5% and 1.1% ) . for smaller simulations we used the same resolution as for @xmath104 , but increased the number of particles per cell purely to improve the statistics of binning particles in energy distributions so that we could attribute , e.g. , a high - energy cutoff ( of an energy spectrum ) , to system size and not to a probability becoming small compared to the inverse of the total number of particles in the simulation . ] in this study we are able to adopt the real ion - to - electron rest mass ratio , @xmath119 , because relativistic effects decrease the scale separation between electrons and ions . essentially , the electron rest mass becomes irrelevant for highly relativistic electrons ; instead , the effective relativistic mass @xmath120 , or equivalently the electron energy @xmath121 , determines the characteristic electron length scales , such as the larmor radius @xmath122 . as we show in this study , both electrons and ions on average gain comparable amounts of energy as a result of reconnection , of order @xmath61 . as long as @xmath123 and hence @xmath96 ( which is satisfied in our study ) , so that typical electrons are ultra - relativistic , the characteristic electron larmor radius is @xmath124 . at the same time , the characteristic ion larmor radius scales as @xmath125 . thus the ion / electron scale separation , characterized by the ratio of the larmor radii , @xmath126 , is independent of the rest mass ratio @xmath127 and remains well below its non - relativistic value of @xmath128 for @xmath129 . in particular , the scale separation scales as @xmath130 in the semi - relativistic regime @xmath131 , diminishing to @xmath132 in the ultra - relativistic regime @xmath2 . thus , the traditional challenge created by the large disparity between electron and ion scales ( in non - relativistic reconnection ) is greatly reduced , allowing us to use the mass ratio @xmath36 . ) in - plane magnetic field lines ( green lines ) and out - of - plane current density @xmath133 ( red / blue shading ) at several different times , expressed in @xmath134 and also in terms of light - crossing times @xmath135 ; the magnetic field starts with a small perturbation ( upper left ) of reversing magnetic field and evolves through reconnection to a nearly steady state ( lower right ) . [ fig : aztimeevolutionsigp1 ] ] ( compare to fig . [ fig : aztimeevolutionsigp1 ] ) , the current layer is thinner ( relative to @xmath64 ) and more plasmoids form during reconnection . anti - reconnection layers can be seen forming between merging plasmoids ( e.g. , lower middle plot , upper current layer , @xmath136 ) [ fig : aztimeevolutionsig10 ] ] the typical evolution of reconnection is described by sequences of snapshots in figs . [ fig : aztimeevolutionsigp1 ] and [ fig : aztimeevolutionsig10 ] . these plots show magnetic flux contours ( i.e. , in - plane magnetic field lines ) superimposed on a color map showing the out - of - plane total electric current density @xmath137 , at several simulation snapshots for two simulations , @xmath109 and @xmath138 , both performed with our fiducial size @xmath139 . since the system is closed , the reconnection process is not stationary , and reconnection eventually ceases as the system asymptotically approaches the relaxed end state . the initially long and thin harris current layers ( @xmath140 , top left panel ) that support the ( slightly - perturbed ) reversing magnetic field @xmath68 quickly become unstable to the tearing instability , creating a chain of several first - generation , primary magnetic islands ( plasmoids ) , clearly visible in the top centre panel ( for @xmath138 there are about 7 primary plasmoids in the lower layer at @xmath141 ) . subsequently , the plasmoids undergo a complex and stochastic hierarchical evolution , as illustrated in figs . [ fig : aztimeevolutionsigp1 ] and [ fig : aztimeevolutionsig10 ] . secondary current sheets form between the plasmoids ; the plasmoids grow , fed by reconnection in the secondary current sheets and plasmoid mergers / coalescence ; meanwhile , new plasmoids form as a result of tearing instability in secondary current sheets . this process proceeds for a few global light crossing times , until the free magnetic energy for reconnection is exhausted ; eventually the number of plasmoids decreases as they merge with each other , and the system reaches the final tearing - stable equilibrium state with just one large magnetic island ( plasmoid ) and one significant x - point in each of the two initial layers ( lower - right panels of figs . [ fig : aztimeevolutionsigp1 ] and [ fig : aztimeevolutionsig10])these final major x - points are relaxed , with separatrices crossing at nearly 90 degree angles . at @xmath142 . by this time , the initially - drifting particles have been almost entirely swept up into the major plasmoid around @xmath143 , and are nearly absent from the region containing active x - points . [ fig : ndrift ] , title="fig : " ] + in plasmoid - dominated reconnection , such as we see in this study , there typically exist , except at the end of reconnection , more than one x- and o - point in each layer . it is useful to define the major x - point and o - point in each layer as the extrema of the flux function ; however , the major x - point is not necessarily the only or even the most active x - point in terms of the reconnection electric field @xmath144 at any given time ( although it is usually one of the most active x - points ) . for example , if we zoom in on the lower layer , as in fig . [ fig : ndrift ] , the major o - point is immediately apparent ( near @xmath143 ) , and there is another prominent o - point near @xmath145 flanked by two prominent x - points ; here ( and in subsequent figures ) we have drawn the magnetic field line ( separatrix ) that goes directly through the major x - point ( to the right of the minor o - point , at @xmath146 ) . after the onset of reconnection the drifting particles in the initial harris layers are quickly swept out of the initial x - points and trapped around the major o - points , as shown in fig . [ fig : ndrift ] at @xmath147 for @xmath109 , where one can see that even the prominent minor o - point contains no initially - drifting particles . meanwhile , background particles that flow into the layer are accelerated ( near x - points , where @xmath148 ) in the @xmath149 direction by the reconnection electric field @xmath144 to maintain the current sheets supporting the @xmath68 reversal . figure [ fig : nandjz ] zooms in on the background electron and ion currents and densities in the region near the major ( and also a prominent minor ) x - point in the lower layer at @xmath150 for @xmath109 ( cf . [ fig : aztimeevolutionsigp1 ] , upper right ) . with the initially - drifting particles already wrapped up in the major plasmoid , the background particles are entirely responsible for subsequent secondary current sheets and plasmoids , such as in fig . [ fig : nandjz ] . in time , the small plasmoid at the centre of fig . [ fig : nandjz ] ( and any other newly - created plasmoids ) will grow and merge with the major plasmoid , leaving a single x - point and o - point in the layer as reconnection winds down . \(a ) + ( b ) + ( c ) + ( d ) and @xmath99 , the above plots show the current ( @xmath133 ) and number density for background electrons and ions , in a narrow region ( @xmath151 ) around the right half of the lower layer at @xmath142 ( @xmath152 ) ; at this time all the particles in the original drifting layer are in the largest plasmoids ( not shown here ) . [ fig : nandjz ] , title="fig : " ] + the evolution of the basic energetics of the system is illustrated in figure [ fig : energyvstime ] , which shows different forms of energy magnetic , electric , and particle kinetic as functions of time . since the system is closed , the total energy in the box is conserved , simulations , which experienced energy non - conservation of 1.5% ( @xmath109 ) and 1.0% ( @xmath153 ) . ] as the magnetic energy is converted to particle kinetic energy . the electric energy indicates the active reconnection phase , decaying as the reconnection process slows down . ) , for @xmath109 and @xmath154 . [ fig : energyvstime ] ] the reconnection process reduces the unreconnected magnetic flux ( associated mostly with @xmath68 ) between the major x - points in the two current layers , while increasing the reconnected flux ( mostly @xmath155 ) between the major x- and o - points within each layer . mathematically , we define the full unreconnected flux @xmath156 ( which is twice the unreconnected flux per current layer ) as the total integrated magnetic flux between the two major x - points in the upper and lower layers ( or , equivalently , the difference in @xmath48 component of the vector electromagnetic potential at the two major x - points ) . figure [ fig : reconfluxvssigma ] , right panel , shows how @xmath156 decays during reconnection , for several different @xmath0 . since our closed system does not allow steady - state reconnection forever , we analyze reconnection dynamics during the period of ` mid - reconnection'after initial transients but long before the end of reconnection . specifically , we define mid - reconnection ( shown between dashed lines in fig . [ fig : reconrate ] ) as the time interval @xmath157 $ ] such that @xmath158 and @xmath159 , i.e. , starting when the unreconnected flux drops to 80% of its initial value and ending when it reaches 70% . because of the aspect ratio @xmath160 , the final system state retains a band of unreconnected flux between the layers around 50% of the original unreconnected flux ( fig . [ fig : reconfluxvssigma ] , right)and therefore mid - reconnection is approximately the middle fifth of active reconnection ( as measured by reconnected flux ) . the final state also retains about 6070% of the initial magnetic energy still in magnetic form ( fig . [ fig : reconfluxvssigma ] , left ) . the reason why we define mid - reconnection in terms of the initial flux ( rather than , e.g. , half - way between initial and final unreconnected flux ) is that it is difficult determine the precise final value of unreconnected flux . at the end of each simulation , a slow , relatively small - amplitude ` breathing ' mode alternately reconnects and anti - reconnects magnetic flux between the remaining x - point and o - point in each layer , resulting in a mild oscillation of unreconnected flux and magnetic energy . determining precise final values would thus require running simulations for several extra light - crossing times to determine the centre of the oscillation ; to save computation time , we therefore use well - defined initial values to determine the mid - reconnection time interval . we emphasize that the fiducial system size @xmath99 is large enough that most of the particles forming the initially - drifting harris sheet have been trapped around the major o - points by the beginning of mid - reconnection . this can be seen in fig . [ fig : ndrift ] , a snapshot of the initially - drifting particle density at the beginning of the mid - reconnection interval for @xmath109 , namely @xmath161 . therefore , we believe that , during and after the period of mid - reconnection , x - point and plasmoid dynamics ( except when involving the major plasmoid ) are controlled primarily by the upstream ( background ) environment , and are much less influenced by the initial harris sheets . one of the main quantities of interest in any reconnection study is the reconnection rate , a measure of how rapidly the magnetic flux is transferred from the upstream ( unreconnected ) region to the downstream ( reconnected ) region and , correspondingly , how rapidly the available magnetic energy is converted into plasma energy . the instantaneous reconnection rate is essentially the time - derivative of the unreconnected flux @xmath156 between the two layers ( times @xmath162 because reconnection proceeds in both layers and we want the rate to be defined as a positive quantity ) . the corresponding dimensionless instantaneous reconnection rate is then @xmath163 . however , @xmath164 exhibits both a complex chaotic short - time behavior due to multiple x - points and plasmoid dynamics , and a secular long - time behavior due the exhaustion of the unreconnected magnetic flux and magnetic free energy ( see fig . [ fig : reconrate ] , right panel ) . therefore , constructing a robust and simple measure of reconnection rate requires some care ; in this paper , we characterize ` the ' dimensionless reconnection rate for a given simulation as the normalized time - average @xmath165 during the mid - reconnection interval : @xmath166/(t_{70\%}-t_{80\%})$ ] . this reconnection rate corresponds to a reconnection inflow velocity of @xmath167 and a reconnection electric field @xmath168 ; with yee finite - difference electromagnetics @xcite , @xmath164 is exactly equivalent to the sum of out - of - plane electric fields at the two major x - points ( by faraday s law ) , and so @xmath169 is the average electric field @xmath144 at the major x - points during mid - reconnection . [ cols= " < , < " , ] at multiple snapshots in time , we analyze all background electrons in the simulation , keeping in mind that ( especially for our aspect ratio , @xmath160 ) at any time , a substantial number remain in the upstream region without having been energized by reconnection . the time - dependence of power - law index and high - energy cutoff is shown in fig . [ fig : pvstime ] for two runs , @xmath109 and @xmath170 . one can see that the index and cutoff reach levels close to their final values well before reconnection slows down due to the exhaustion of magnetic free energy ; we take the final values at the end of reconnection to be the representative values of @xmath8 and @xmath171 for each simulation . figure [ fig : alphavssigma ] shows the variation in the electron power - law index @xmath8 with @xmath0 ; for some values of @xmath0 , more than one simulation ( identical except for random initial particle velocities ) were run to gauge statistical variations in the power - law index . we find empirically that , for the range of @xmath0 studied , @xmath8 can be approximated by the simple formula ( which hopefully may find useful applications for astrophysical modeling , see [ sec - astro ] ) @xmath172 vs. @xmath0 , for electron energy spectra at the end of reconnection ; error bars showing the robustness of fit are not shown because they are not significantly larger than the @xmath173-symbols . ( for @xmath174 , however , the automatic fitting procedure failed to identify a single convincing power - law slope , with error bars covering @xmath175 ; we plot a value based on visual inspection ( see fig . [ fig : eleciondist1 ] ) , but consider it to be highly uncertain . ) [ fig : alphavssigma ] ] vs. @xmath0 . the error bars show variation due to different choices of parameters used to fit the power - law and cutoffs . results for @xmath174 are not shown because of the high uncertainty in the power - law index . [ fig : gcvssigma ] ] for large @xmath0 , where electron - ion reconnection should be identical to electron - positron reconnection , @xmath8 approaches a value around 1.9 , consistent with previous studies of pair - plasma reconnection . for these simulations , by virtue of our initial setup , @xmath176 as @xmath177 ; a much larger @xmath77 would yield @xmath8 closer to 1 than 2 @xcite . in the semirelativistic regime we find @xmath8 varying between 2 and 4 as @xmath0 decreases from about 10 to @xmath178 , reaching a value between @xmath179 and 3 at @xmath153 . for @xmath180 we have only one simulation , @xmath174 , and its power law appears steeper than for @xmath109 ; however , the steepness makes it hard to measure the index @xmath8 with much confidence . we note that the @xmath181 scaling of the electron power - law index in the semi - relativistic ( small-@xmath0 ) limit , expressed by equation [ eq : p_e ] , can be understood within the framework of stochastic 2nd - order fermi acceleration of ultra - relativistic electrons bouncing back and forth between plasmoids moving randomly along the reconnection layer . denoting the typical separation between plasmoids in the plasmoid chain as @xmath182 and the typical plasmoid speed as @xmath183 ( expected to be of order @xmath184 and hence @xmath185 in the semi - relativistic regime ) , we can estimate the average particle energy gain per bounce as @xmath186 . the typical time between bounces is simply @xmath187 , resulting in an effective acceleration time - scale of @xmath188 . thus , a typical energetic electron , bouncing between two plasmoids large enough to contain it , undergoes random - walk diffusion in energy space with a diffusion coefficient of @xmath189 , until the two plasmoids come together and merge . the particle then ` escapes ' from the inter - plasmoid acceleration region , e.g. , by becoming trapped inside the merged plasmoid , with a typical ` escape time ' that can be estimated simply as the characteristic time for two plasmoids to approach each other , @xmath190 ( omitting factors of order unity ) . thus , the 2nd - order fermi power - law index should scale as @xmath191 . since @xmath192 in the @xmath131 regime , we see that the power - law index has a @xmath0-dependence of the general form @xmath193 , where @xmath194 is a constant of order unity , consistent with our numerical findings . the high - energy cutoff @xmath171 depends mostly on @xmath0 ( fig . [ fig : gcvssigma ] ) ; as @xmath0 varies over more than 3 orders of magnitude , the cutoff energy scales as an @xmath195 prefactor times @xmath196 , or roughly @xmath197 . in the ultra - relativistic limit , the prefactor in this study is roughly twice that found in @xcite ; the difference may be due to different simulations setups . the normalized cutoff energy @xmath198 rises slowly with @xmath0 in the semi - relativistic regime , from around 2.5 to 4 or 4.5 as @xmath0 goes from @xmath178 to 10 . due to random fluctuations from simulation to simulation and the challenges of making precise measurements , as well as the difficulty of identifying convergence in the limit @xmath199 , we have not been able to conclude with confidence whether the computed power - law indices and cutoff energies are truly independent of @xmath39 for @xmath200 . we have specifically examined the @xmath39-dependence of indices and cutoffs for the cases of @xmath109 and @xmath153 by extending our studies to @xmath117 for these cases , as shown in fig . [ fig : alphaandgcvsl ] . first , however , we note that our system size @xmath99 is already squarely in the large - system regime described by @xcite ( for electron - positron reconnection in simulations with aspect ratio @xmath201 , the critical system size separating small- and large - system regimes , in the language of this paper , is reached when @xmath39 equals @xmath202 ) . indeed , the @xmath153 case yields an index and cutoff that look quite independent of @xmath39 for @xmath203 . it is much less clear whether the @xmath109 simulation has converged with respect to @xmath39 , although the trend suggests that @xmath204 for larger @xmath39 , and we suspect , based on the convergence of @xmath205 with @xmath39 and the apparent agreement between @xmath206 and @xmath207 , that @xmath8 may near its asymptotic limit at @xmath208 . that said , establishing an asymptotic limit is always difficult , especially since the computational resources needed to perform a ( 2d ) simulation scale as @xmath209 ( with our explicit algorithms ) , preventing us from exploring much larger @xmath39 . the ion spectra @xmath210 generally appear nonthermal , but exhibit obvious power laws only in the ultra - relativistic regime ( fig . [ fig : eleciondist1 ] ) . for @xmath211 , the ion spectra show nonthermal power laws that closely match the electron spectra at high energies ; however , for @xmath84 the ion spectra have a hard slope @xmath212 ( which appears horizontal in the compensated spectra of fig . [ fig : eleciondist1 ] ) that emerges immediately out of the cold ( upstream ) maxwellian distribution and extends only to medium energies . while not a really convincing power - law , @xmath210 in this region is flatter and broader than a maxwellian ( compare , e.g. , the maxwellian @xmath140 ion spectrum with late - time spectra in fig . [ fig : fvstime ] ) . at energies well below the _ electron _ cutoff energy , the ion spectra turn significantly downward , although at these high energies there are always more ions than electrons at the same energy ( consistent with the fact that ions collectively gain more energy than electrons ) . we note that the ion spectra even if described by power laws might be expected to show a break around @xmath213mev , where protons become trans - relativistic . indeed , the hard @xmath212 slope never extends beyond @xmath214mev ( most notably for @xmath215 in fig . [ fig : eleciondist1 ] ) . in summary , both electrons and ions are accelerated to nonthermal energy distributions , with the electron spectra forming pronounced power laws at high energies . the electron power - law index @xmath8 becomes steeper as @xmath0 decreases , and extends to a cutoff energy @xmath216 . in the ultrarelativistic limit , electron and ion spectra become similar , with a power - law slope @xmath217 that may depend on @xmath218 , which is 50 for this investigation . we know from electron - positron reconnection studies @xcite that in the ultrarelativistic limit a smaller @xmath82 results in a larger / steeper power - law index @xmath8 ; however , in the pair case , small @xmath82 means that electrons do not get accelerated very much ( compared to their initial thermal energy ) . in contrast , electron - ion reconnection can ( e.g. , for @xmath34 ) yield steeper slopes @xmath219 while maintaining @xmath220 ( allowing significant electron acceleration ) . the ion spectra are harder to characterize , but at high energies there are more ions than electrons with any given energy . the results obtained in this paper are relevant to a number of astrophysical systems , including accretion disk coronae ( adc ) and blazar jets , which are composed of electron - ion plasma likely undergoing reconnection in the semi - relativistic regime . the hard x - ray spectra of adc in black - hole x - ray binaries in the high - soft ( hs ) and steep power law ( spl ) states often exhibit relatively steep power - law tails that have spectral indices @xmath221 and extend to well beyond 1 mev @xcite . this high - energy radiation is generally believed to be produced by inverse - compton ( ic ) scattering of soft x - ray seed photons coming from the accretion disk by a nonthermal population of highly relativistic coronal electrons with an inferred power - law index @xmath222 . blazars often feature double - humped broad - band spectral energy distributions ( seds ) attributed to synchrotron radiation ( lower - energy component ) and ic scattering ( higher - energy component ) . there are two main spectral classes of blazars : flat - spectrum radio quasars ( fsrqs ) with synchrotron peaks in the far - infrared range and ic peaks in the mev range ; and high - frequency peaked bl lacs ( hbls ) with synchrotron peaks in uv and x - rays and ic peaks in the gev tev range ( see * ? ? ? * for a recent review ) . a systematic trend of increasing synchrotron luminosity and increasing compton dominance with decreasing synchrotron peak frequency is known as the blazar sequence @xcite . typical spectral indices of fsrqs in the x - rays are @xmath223 @xcite , indicating nonthermal distributions of ultra - relativistic electrons with @xmath224 . to explain the observed seds , these electron energy distributions should extend to @xmath225 in the case of fsrqs and to @xmath226 in the case of hbls . there are even a few ultra - high - frequency - peaked bl lacs ( uhbls ) that demand @xmath226 , possibly requiring hadronic models @xcite . the nonthermal electron power - law distributions inferred in these systems have often been attributed to first - order fermi acceleration in strong shocks @xcite , but such a model is viable only if the plasma involved is weakly magnetized . in contrast , adcs are thought to be strongly magnetized ; the buoyant rise of magnetic energy to form magnetically dominated coronae in accretion disk models with vertical stratification is well - documented in local ( shearing box ) simulations @xcite , suggesting that reconnection is the dominant dissipation and particle acceleration mechanism @xcite . as for blazar jets , the likely importance of magnetic energy out to large distances in these systems is due to the nature of the hydromagnetic acceleration process , which is efficient as long as the jet is magnetically dominated but becomes extremely inefficient once the kinetic energy and poynting fluxes become comparable @xcite . jet acceleration can be facilitated by dissipation of magnetic energy , e.g. , through reconnection triggered by current - driven instabilities @xcite . barring such dissipation , the conversion of poynting flux to kinetic energy would get ` stuck ' at values of @xmath227 . it is thus interesting to observe that the nonthermal electron distributions with @xmath228 , which emit radiation spectra with @xmath229 seen in adc and blazars , can be produced , according to our simulations , by magnetic reconnection with @xmath230 just the range where substantial dissipation should go hand - in - hand with gradual acceleration of the jet beyond approximate equipartition . we therefore suggest that reconnection in this semi - relativistic , near - equipartition regime , is a natural way to produce the observed nonthermal spectra in blazar jets , and perhaps in adc as well . furthermore , adopting our estimated cutoff lorentz factors for the accelerated electrons , @xmath231 , and assuming @xmath232 ( corresponding to @xmath233 and @xmath234 ) for both fsrqs and hbls , we obtain @xmath235 . this corresponds to a frequency of the observed synchrotron peak @xmath236hz , where @xmath237 is the comoving magnetic field in the emitting region , measured in gauss , and @xmath238 is the doppler boost due to the bulk lorentz factor of the jet , which typically lies in the range @xmath239 @xcite . fsrqs , with synchrotron peak frequencies extending from as low as @xmath240hz into the uv band , would therefore require low magnetic fields @xmath241 , consistent with the deduction that gamma - rays must be produced at rather large distances from the black hole , @xmath242 , where @xmath243 is the gravitational radius , to avoid excessive pair - production opacity @xcite . the gamma - rays from hbls , on the other hand , could be produced closer to the jet base ( at @xmath244 ) with corresponding magnetic field strengths @xmath245 . electron lorentz factors @xmath246 are also consistent with the tev photons being produced by the synchrotron - self - compton mechanism in hbls , in contrast to the mev gev gamma - rays in fsrqs , which most likely result from comptonization of ambient photon sources @xcite . in addition to the distributed , and relatively steady , blazar emission discussed above , blazars often exhibit intense , rapid flares in gamma - rays @xcite . to avoid excessive pair production , which would prevent the gamma - rays from escaping , the flaring regions must have local bulk lorentz factors that exceed the mean for the jet by a factor of a few @xcite . such ` minijets ' could represent outflow regions from relativistic reconnection sites @xcite , with local values of @xmath0 several times larger than the mean . we would then predict the spectral slopes to be correspondingly flatter and the cutoffs higher in proportion to @xmath247 ( assuming similar sources of seed photons for comptonization ) . an additional argument in support of the minijet model follows from our finding ( see [ sec : plasmaflows ] ) that relativistic reconnection - driven electron fluid flows do not actually require high @xmath0 but readily occur in the semi - relativistic , small-@xmath0 cases as well . since gamma - radiation is produced by the electrons , the minijet mechanism can thus operate successfully over a broad range of plasma magnetizations , including the @xmath248 regime favored by the observed spectral indices as described above . finally , we would like to stress the crucial role of the electron - ion plasma composition , in the semi - relativistic limit , for explaining the range of spectral indices typically seen in adc and blazar jets . kinetic simulations of relativistic reconnection in electron - positron pair plasmas that produce highly relativistic energy cutoffs necessary to explain the blazar observations yield nonthermal electron power - law indices close to @xmath249 @xcite . while such hard electron distributions may be necessary to explain the most extreme flaring events in blazars , characterized by very flat photon spectra with spectral indices near zero @xcite , they are much too flat to be consistent with most of the regular blazar emission observations , even if one took cooling into account . in contrast , in electron - ion reconnection , electrons are accelerated to ultra - relativistic energies even for modest magnetizations @xmath250 , which , according to our simulation results , are needed to explain the range of spectral indices observed in these systems . in this paper , we performed a comprehensive numerical investigation of antiparallel ( no guide magnetic field ) collisionless relativistic magnetic reconnection in an electron - ion plasma , relevant to many astrophysical systems , such as coronae of accreting black holes in x - ray binaries and active galactic nuclei , and their relativistic jets , including blazars . we used a series of large 2d relativistic particle - in - cell simulations with real proton / electron mass ratio @xmath251 to probe various fundamental aspects of reconnection , such as dynamics ( reconnection rate , electron and ion outflows , and hall - effect signatures ) , energetics ( e.g. , the energy partitioning between electrons and ions ) , and nonthermal particle acceleration in different astrophysically - relevant physical regimes . the main focus of our study was on exploring how these aspects are affected as one transitions from the semi - relativistic regime ( ultra - relativistic electrons but non - relativistic ions ) , which has not been systematically explored before , to the fully relativistic regime ( both ions and electrons are ultra - relativistic ) . since this transition is quantified by the initial ` cold ' upstream ion magnetization parameter @xmath252 ( basically , twice the average upstream magnetic energy per background ion normalized by the ion rest mass ) , our parameter - space study explored a broad range of @xmath0 , from very small ( semi - relativistic regime , @xmath253 ; down to @xmath174 in our simulations ) to very large ( ultra - relativistic regime , @xmath2 ; up to @xmath254 ) . we also explored the dependence of various reconnection characteristics on the system size @xmath39 ( while keeping the aspect ratio of our computational box fixed , @xmath160 ) , our largest production runs ( for all @xmath0 ) reaching well into the large - system , plasmoid - dominated regime . our main findings can be summarized as follows . first , the average dimensionless reconnection rate @xmath255 during the active reconnection phase approaches a constant value @xmath256 in the ultra - relativistic limit @xmath2 . this value is consistent with previous studies of relativistic electron - positron pair - plasma reconnection @xcite , which is in fact expected since any dynamical differences between ions and electrons disappear in the ultra - relativistic limit ( and in the absence of radiative losses ) and so electron - ion - plasma reconnection should become indistinguishable from the extensively studied pair case . our simulations also show that as @xmath0 is lowered below @xmath257 , @xmath258 decreases . this dependence of the reconnection rate on @xmath12 can be attributed almost entirely to the scaling of the upstream alfvn speed @xmath184 with @xmath0 : the reconnection inflow velocity properly normalized to @xmath184 , @xmath259 , remains essentially constant ( about 0.1 ) across the physical regimes , varying very little as @xmath0 is varied over 3 orders of magnitude . we also find that the reconnection rate is rather insensitive to the system size for large enough systems . our simulations also show that , just as in the non - relativistic case , reconnection diffusion regions around x - points develop a two - scale structure in the semi - relativistic regime , with narrow electron current layers embedded in broader ion layers . interestingly , we find that electron fluid outflows in these electron layers can locally become relativistic , even in the small-@xmath0 cases , when the overall ( ion ) reconnection outflows are sub - relativistic ; however , even for high-@xmath0 , these fluid outflows do not become more than mildly relativistic with lorentz factors of a few . because the potentially observable radiative signatures are dominated by electrons , this finding may have important astrophysical implications , especially in situations where relativistic motions of the emitting , flaring plasma fireballs are inferred , as in the minijets model for ultra - rapid tev flares in blazar jets @xcite . another interesting consequence of the different electron and ion flow patterns in the reconnection region in the semi - relativistic regime is the development of the classic hall - effect signatures near magnetic x - points ( again , similar to non - relativistic electron - ion reconnection ) : the out - of - plane magnetic field @xmath260 with the characteristic quadrupole structure and the corresponding dipolar ( directed into the layer and downstream ) in - plane electric field , which can be described by an electrostatic potential that forms potential wells inside the plasmoids . as one transitions into the ultra - relativistic regime , these signatures weaken , as expected . we also explored the question of the partitioning of the released magnetic energy between electrons and ions as a function of @xmath0 . this issue is important for interpreting astrophysical observations ( e.g. , for evaluating the energy budget of a high - energy flare ) and for computing observational signatures of global mhd simulations of , e.g. , coronae of accreting black holes , since only the electron fraction of the released energy can be promptly radiated away and hence directly observed as a flare . we found that while the electrons and ions get equal amounts of energy in the ultra - relativistic limit , as expected , the electron energy fraction decreases as @xmath0 is decreased , but eventually seems to asymptote at a finite value of about 1/4 in the semi - relativistic limit . we then proposed a simple empirical formula [ see eq . ( [ eq : energypartition ] ) ] that describes this dependence . finally , we characterized the resulting energy spectra of electrons and ions accelerated by reconnection . we found that both electrons and ions attain nonthermal distributions by the end of the reconnection process . in particular , we were able to obtain clear evidence for nonthermal particle acceleration of electrons , manifested as robust power - law energy distributions . we mapped out the dependence of the electron power - law index @xmath8 and high - energy cutoff @xmath171 on @xmath261 and @xmath39 , and found that the index becomes fairly insensitive to the system size but exhibits a relatively strong dependence on @xmath0 in the semi - relativistic regime , well - fit by the empirical relationship @xmath262 [ see eq . ( [ eq : p_e ] ) ] . the electron cutoff energy is found to scale approximately linearly with @xmath0 , namely , @xmath263 , over several orders of magnitude in @xmath0 , although its scaling with @xmath39 is not completely clear ; a conclusive detailed investigation of this issue is left for a future study . as for ions , based on our present simulations we can not claim a power - law ion energy distributions in the semi - relativistic regime , although in the ultra - relativistic case the ion distribution of course becomes very similar to that of the electrons , with a clear power law . we believe that our results have important astrophysical implications , e.g. , for high - energy emission from blazar jets . in particular , the electron power - law indices ( @xmath264 ) that correspond to the nonthermal radiation spectra typically observed in blazars ( @xmath223 ) can be produced by magnetic reconnection with modest ambient magnetization @xmath250 , in agreement with the values that are naturally expected in theoretical models of relativistic jet acceleration . our present study acts as a bridge connecting two previously extensively - investigated limits : non - relativistic electron - ion reconnection ( usually done in the context of , e.g. , solar flares and earth s magnetosphere ) and ultra - relativistic pair - plasma reconnection ( with applications to , e.g. , pulsar winds and pwn ) . this research provides baseline knowledge of magnetic reconnection in the astrophysically - important semi - relativistic electron - ion regime , describing the basic dynamics of reconnection as well as resulting particle acceleration , and offering a concrete way to connect theoretical models with astrophysical observations by providing specific fitting formulae for several key parameters of the accelerated particle populations ( such as @xmath265 , @xmath8 , and @xmath171 ) . our study also opens up a number of exciting new opportunities that should be explored in the near future ; it will serve as a departure point for more detailed studies of the semi - relativistic regime , including the examination of the effects of guide magnetic field , upstream plasma temperatures , initial configurations other than harris sheets , 3d effects , and synchrotron and ic radiative cooling . this work was supported by doe grants de - sc0008409 and de - sc0008655 , nasa grants nnx12ap17 g and nnx16ab28 g , and nsf grant ast-1411879 . d. a. uzdensky gratefully acknowledges the hospitality of the institute for advanced study and support from the ambrose monell foundation ; and k. nalewajko received support from the polish national science centre grant 2015/18/e / st9/00580 . this work required several supercomputer allocations . this work used the extreme science and engineering discovery environment ( xsede ) , which is supported by national science foundation ( nsf ) grant number aci-1053575 @xcite , and in particular we used the stampede supercomputer at the texas advanced computing center ( tacc ) at the university of texas at austin . this work also used the pleiades supercomputer , a resource provided by the nasa high - end computing ( hec ) program through the nasa advanced supercomputing ( nas ) division at ames research center . in addition , this work made use of an award of computer time provided by the innovative and novel computational impact on theory and experiment ( incite ) program ; in particular , we used the mira supercomputer and associated resources of the argonne leadership computing facility , which is a doe office of science user facility supported under contract de - ac02 - 06ch11357 . we also gratefully acknowledge the developers responsible for the vorpal / vsim simulation code . in this appendix , we describe how to calculate relativistic fluid flow velocities from particles in our kinetic simulations . for non - relativistic motion , bulk fluid flow is unambiguously defined as the average velocity ; in a local reference frame traveling at this velocity , the net particle current density and momentum density both vanish . for relativistic motion , however , these quantities may vanish in different frames and so the task / problem of defining the bulk flow velocity in terms of the moments of particle distribution function function becomes less trivial . we first describe the eckart flow velocity , defined to be the boost velocity that transforms to a reference frame in which the current density vanishes @xcite . given a particle distribution @xmath266 , where @xmath8 is the 4-momentum , we can define a density - current 4-vector @xmath267 where @xmath268 and @xmath269 ; @xmath266 and @xmath270 are invariant scalars , showing that @xmath271 transforms like the 4-vector @xmath272 . straightforward calculation shows that any ( time - like ) 4-vector @xmath271 can be boosted to a reference frame in which @xmath273 for @xmath274 by a velocity @xmath275 thus a boost by the average velocity transforms to a frame the eckart local rest frame in which the net current density is zero . on the other hand , it is equally valid to consider the landau fluid rest frame @xcite , in which the net momentum density vanishes ; in this frame , the components @xmath276 of the stress - energy tensor vanish . in a general frame , the stress - energy tensor is @xmath277 the components @xmath278 are proportional to the local average momentum @xmath279 . given @xmath280 , one can determine the boost that yields @xmath281 by first finding the boost along @xmath46 that reduces @xmath282 to zero , then boosting from that frame along @xmath47 to reduce @xmath283 to zero , and finally along @xmath48 . however , we can also determine the landau 4-velocity @xmath7 by considering the 4-vector @xmath284 . since boosting @xmath7 by itself yields @xmath285 , and @xmath286 in the boosted frame , @xmath287 thus @xmath288 is an eigenvector of the matrix @xmath289 with eigenvalue @xmath290 ; transforming this eigenvalue equation to the original frame , we see that @xmath7 is an eigenvector of @xmath291 with the same eigenvalue : @xmath292 in the boosted frame , the matrix @xmath289 is block diagonal , with the @xmath293 block consisting of @xmath294 , corresponding to the eigenvector @xmath295 , and the @xmath296 block , which is symmetric and therefore has 3 real eigenvalues and orthogonal eigenvectors , which are space - like because they are orthogonal to @xmath297 . therefore , the matrix @xmath291 also has 4 real eigenvalues , along with one time - like and three space - like eigenvectors . the time - like eigenvector , if normalized to @xmath298 ( and , if necessary , multiplied by @xmath299 so that @xmath300 ) , yields the landau 4-velocity @xmath7 . to be clear , we note that @xmath301 , where @xmath302 is the minkowski metric with diagonal @xmath303 . it is the non - symmetric matrix @xmath304 , and not the symmetric stress - energy tensor @xmath305 , that must be diagonalized . in the boosted frame , however , @xmath306 is symmetric because @xmath307 is symmetric and @xmath286 . thus , given a local stress - energy tensor @xmath308 , the landau 4-velocity @xmath7 can be found as follows : 1 . diagonalize the matrix @xmath304 and find its eigenvectors . only one eigenvector @xmath7 is time - like ; normalize it so that @xmath309 and @xmath310 ; @xmath7 is the landau fluid velocity . we have developed a robust procedure to fit the high - energy portion of particle energy spectra to a power - law ; subsequently we can determine the high - energy cutoff of the power - law . essentially , the procedure attempts to identify the longest section of the spectrum @xmath311 that resembles a power law @xmath312 , and then fits a power law to that section only . [ here , @xmath313 is the particle kinetic energy . ] because @xmath311 is generally an excellent power - law at very low energies ( the low - energy part of the initial maxwellian)and because we are mainly interested in the nonthermal spectrum we look only at @xmath314 where @xmath315 is around the average particle energy ( for this study , we used 1 , 2 , and 4 times the average energy ) . we begin by smoothing @xmath311 . empirically , we expect @xmath311 to decrease monotonically above @xmath316 , and so any non - monotonicity indicates noise . therefore , we find the monotonic function closest to @xmath311 using the pool - adjacent - violators algorithm @xcite , and use this to estimate noise ; that is , we subsequently approximate @xmath311 with the minimal number of cubic splines necessary so that the best spline fit is , on average , no further from @xmath311 than the monotonic fit . the spline fit yields a smoother @xmath311 . working with the smoothed @xmath311 , we calculate the local slope @xmath317 , and then , with brute - force , search for the longest interval [ in terms of @xmath318 over which @xmath319 remains approximately constant ( for example , @xmath8 varies at most within @xmath320 or @xmath321 of a central value ) . we estimate ` the ' power - law index @xmath8 as the median value of @xmath319 over that interval . after fitting a pure power - law over the same interval , we calculate the cutoff energy @xmath171 where the measured @xmath311 drops to a value that is @xmath322 times the pure power - law fit . this procedure for determining a power law has the advantage that , within the identified power - law section , the local slope @xmath319 does not vary much , and so guarantees that the identified section will really resemble a power - law ( although the procedure might fail to find such a section of reasonable length ) . however , this procedure depends on several ( somewhat arbitrary though reasonably - chosen ) parameters , such as @xmath315 , the amount of smoothing , and the allowed maximum variation in @xmath319 . therefore , for each particle spectrum obtained from simulation , we perform many fits with different sets of these parameters , taking the median @xmath8 ( over all fits ) as the final value . we estimate the robustness by determining the range of @xmath8 that encompasses the middle 68% of the fits , and express this range as ` error ' bars around the median .

this theoretical paper investigates the formation of ground state molecules from ultracold cesium atoms in a two - color scheme . following previous work on photoassociation with chirped picosecond pulses [ luc - koenig et al . , phys . rev . a * 70 * , 033414 ( 2004 ) ] , we investigate stabilization by a second ( dump ) pulse . by appropriately choosing the dump pulse parameters and time delay with respect to the photoassociation pulse , we show that a large number of deeply bound molecules are created in the ground triplet state . we discuss ( i ) broad - bandwidth dump pulses which maximize the probability to form molecules while creating a broad vibrational distribution as well as ( ii ) narrow - bandwidth pulses populating a single vibrational ground state level , bound by 113 . the use of chirped pulses makes the two - color scheme robust , simple and efficient .

physics0508090

the field of cold molecules is rapidly developing , since many research groups are working at the production of dense samples of cold ( @xmath11 k ) or ultracold ( @xmath2 100 @xmath3k ) molecules @xcite . moreover , the field of molecular condensates has opened recently , with experimental evidence for molecules formed in a degenerate gas , starting either from an atomic bose - einstein condensate @xcite , or from an atomic fermi gas @xcite . this intense activity is motivated by possible applications to very precise measurements ( lifetimes , checks on parity violation , determination of a possible dipole moment of the electron as a check for the standard model in elementary particles ) , bose - einstein condensation of complex systems ( molecule laser ) , or the emergence of a new ultracold chemistry . among the various routes to produce ultracold molecules , photoassociation has emerged as a purely optical technique starting from an assembly of laser - cooled atoms in a trap . the reaction takes place when a pair of ground state atoms interacting via the potential @xmath4 absorbs a photon red - detuned compared to the atomic line . at resonance , one vibrational level in an excited electronic potential @xmath5 of the molecule is populated . the reaction can be interpreted as a vertical transition at large distances , forming a long range molecule . we shall consider the example of two cold cesium atoms colliding in the lower triplet @xmath6 potential and forming a molecule in a vibrational level @xmath7 of the @xmath8 potential . the first experimental observation of ultracold cs@xmath9 molecules has made use of this reaction @xcite . in order to make molecules in the _ ground _ or _ lower triplet state _ , the photoassociation step must be followed by a stabilization step . the photoassociated molecules are short - lived and decay via spontaneous emission , most often through a vertical transition at large interatomic distances @xmath10 giving back a pair of atoms . however , specific mechanisms which favor radiative decay at shorter distances exist and allow for populating bound vibrational levels in the potential @xmath11 ( or possibly , for instance in case of heteronuclear dimers , to another potential correlated to the same asymptote @xcite ) . in the special case of the double well @xmath8 excited potential of cs@xmath9 , the smooth barrier between the two wells acts as a `` speed bump '' and slows down the vibrational motion in the external well such that vertical transitions can take place at intermediate distances @xmath12 . stable molecules formed by spontaneous emission through this mechanism have been detected in a mot @xcite . other efficient stabilization mechanisms such as resonant coupling @xcite exist . in a condensate , in order to retain coherence , spontaneous emission should be avoided and a stabilization step involving stimulated emission should be implemented within a two - color experiment . however , with continuous wave ( cw ) lasers such experiments are hardly feasible since the scheme is fully reversible : population is pumped back from bound levels of the ground state to the excited state , and then to continuum levels of the ground state , i.e. the molecule is dissociated . the use of laser pulses becomes thus unavoidable . in previous work @xcite , the possibility to increase the efficiency of the photoassociation reaction by use of chirped laser pulses was considered . the frequency of such pulses has a linear time - dependence . the parameters of the pulse were optimized to achieve a total transfer of population under adiabatic following conditions @xcite . it was shown @xcite that a significant improvement of the photoassociation rate compared to cw excitation can be obtained , since during the pulse the resonance condition is fulfilled for several vibrational levels @xmath7 . however , the quantity which must be improved is the formation rate of ground state molecules rather than the photoassociation rate . one advantage of excitation with chirped pulses is that the chirp parameter can be chosen to modify the shape of the vibrational wave packet formed in the potential @xmath5 . this may improve the efficiency of the stabilization step , and thus the formation rate of ground state molecules . the aim of the present work is therefore to suggest a two - color experiment using short laser pulses and to investigate how efficiently a second pulse can transfer the population to the bound vibrational levels of the ground state . starting from the vibrational wave packet formed in the excited state by photoassociation , we introduce a second pulse which is delayed in time from the first pulse . the time delay is chosen such that the wave packet has reached the vicinity of the inner turning point , where a vertical radiative transition to the ground state is expected to be efficient . the parameters of the second pulse , as well as the time delay between the two pulses are varied to improve the efficiency of the transfer , with two different goals : to either maximize the total population transfer , or to selectively populate one vibrational level . the paper is organized as follows : in sec . [ sec : previous ] we recall the main results of previous work on photoassociation with chirped laser pulses . the two - color scheme is introduced in sec . [ sec:2color ] , and the best choice for the dump pulse discussed by considering the franck - condon overlap between the vibrational wave packet in the excited state and wave functions of the bound levels in the ground state . the results for maximum population transfer are presented in sec . [ sec : many ] while sec . [ sec : single ] reports on the selective population of a single level . finally sec . [ sec : concl ] concludes . we briefly summarize the main results of refs . @xcite for the photoassociation of cold cesium atoms by chirped laser pulses . the main difference to cw excitation lies in two facts : several levels can be populated in the photoassociation step , and the vibrational wave packet can be shaped in order to allow for spatial `` focussing '' , as described below . we start from a transform - limited pulse with gaussian envelope , hereafter referred to as pump pulse , with maximum at @xmath13 , and with full width at half maximum of the temporal intensity profile ( fwhmi ) @xmath14=15 ps . chirping the pulse stretches the fwhmi to @xmath15=34.8 ps and decreases the maximum intensity @xmath16 ( @xmath17 = 120 kw @xmath18 ) , conserving the energy carried by the field . moreover , 98% of this energy is carried during the time window @xmath19 $ ] . we consider the reaction @xmath20 \nonumber & & \quad \quad \quad \mathrm{cs}_{2}(0_g^-(6s\,\,^{2}s_{1/2}+6p\,\,^{2}p_{3/2};v^{\prime},j'=0))\end{aligned}\ ] ] where the carrier frequency @xmath21 is chosen red - detuned relative to the d@xmath9 atomic line , at resonance with the @xmath7=@xmath22=98 level in the external well of @xmath23 which is bound by 2.65 @xmath0 . @xmath24 is the linear chirp rate in time domain , chosen negative and equal to -0.025 @xmath0 ps@xmath25 . the parameters of the pump pulse , labelled @xmath26 are recalled in table [ tab : pump ] . we will also consider other possibilities for the pump pulse , with positive or zero chirp , or with a smaller detuning : such pulses are respectively labelled @xmath27 , @xmath28 and @xmath29 . the latter pulse has a central frequency resonant with the excited level @xmath7=122 . the parameters for @xmath29 have been chosen to ensure the focussing condition without populating continuum levels in the photoassociation step . @xmath17 & @xmath30 & @xmath14 & @xmath15 & @xmath31 & @xmath32&@xmath33&10@xmath34@xmath35 + @xmath26 & 2.656 @xmath0 & 120 kw @xmath18&0.98 @xmath0&15 ps&34.8 ps&-0.025 @xmath0ps@xmath25&21.2 ps&2.32&3.2 + @xmath27 & 2.656 @xmath0 & 120 kw @xmath18&0.98 @xmath0&15 ps&34.8 ps&+0.025 @xmath0ps@xmath25&21.2ps&2.32&3.5 + @xmath28 & 2.656 @xmath0 & 120 kw @xmath18&0.98 @xmath0&15 ps&15 ps&0&21.2 ps&1&0.4 + @xmath36 & 0.675@xmath0 & 120 kw @xmath18&0.26 @xmath37&57.5 ps&110 ps&-0.002 @xmath0 ps@xmath25&81.3 ps&1.91 & 20 + note that in order to compare to earlier papers , we have also considered another the temporal width , @xmath38 where @xmath39 is now defined as the full width at half maximum of the electric field profile ( fwhme ) of the transform - limited pulse , while in refs . @xcite @xmath14 was defined with respect to intensity ( fwhmi ) . @xmath33 is the stretching factor due to chirping . during the time window , the central frequency is swept from @xmath40 to @xmath41 such that the resonance condition is satisfied for 15 vibrational levels around @xmath22=98 in the energy range @xmath42 @xmath0 . depending upon the sign of the chirp , the frequency is increasing ( @xmath43 ) or decreasing ( @xmath44 ) with time . the dynamics of the photoassociation process was studied by numerical solution of the time - dependent schrdinger equation describing the motion of wave packets in the ground state potential @xmath6 ( @xmath45 ) and in the @xmath46 excited potential ( @xmath23 ) coupled by the electromagnetic field . the initial state is represented by a stationary continuum wave function describing the relative motion ( @xmath47-wave ) of two colliding cs ( 6s , f=4 ) atoms , with an initial kinetic energy of 54 @xmath3k . the wave packet in the excited potential @xmath23 is analyzed by projection onto the vibrational wave functions of the @xmath7 levels in the external well . the main results of the calculations are : * during the pulse , many vibrational levels are populated . however , after the pulse , the population remains only in the 15 vibrational levels resonantly excited around @xmath22=98 . this defines a photoassociation window in the energy domain , for the levels located between @xmath48=106 and @xmath49 = 92 , bound by 1.74 @xmath0 and 3.57 @xmath0 , respectively . * in the impulsive approximation @xcite , assuming that the relative motion of the two atoms is frozen during the pulse , it is possible to analyze the numerical results by a two - state model at each internuclear distance @xmath10 . while the frequency of the laser is swept from @xmath50 to @xmath51 , the crossing point @xmath52 of the two field - dressed potential curves is moving in the @xmath53 $ ] range . assuming that the resonance condition corresponds to a vertical transition at @xmath54 , this defines a photoassociation window in the @xmath10 domain : in the example chosen , @xmath55 = 85 and @xmath56 = 107 a@xmath57 . within this range of distances , all the population of the initial continuum level can be adiabatically transferred to the excited state . * in the excited state , the levels @xmath58 which are populated have a classical outer turning point @xmath59 in the @xmath53 $ ] range , and the two - state model predicts that the population transfer occurs around the time @xmath60 when @xmath59= @xmath61 . for a negative chirp , the position of the crossing point varies from @xmath56 to @xmath55 , so that the upper levels are populated first ( see fig . [ fig : vib ] ) . for the pump pulse considered , the level @xmath48= 106 is populated at the beginning of the time window ( @xmath62 ) , the level @xmath49 = 92 is populated at the end ( @xmath63 ) , i.e. 70 ps later . it is the opposite for a positive chirp , choosing the @xmath64 pulse for photoassociation . * the total probability of population transfer to the excited state is reported in the last column of table [ tab : pump ] . the absolute value is discussed in refs . @xcite , the important result for the present paper being that the population obtained with @xmath28 is increased by a factor 8 when a chirped pulse @xmath26 is considered , and by a factor of 6 when the pulse @xmath29 , with a smaller detuning , is used . * after the end of the pulse , the vibrational wave packet @xmath65 in the excited state is moving from the outer classical turning point toward shorter distances ( see fig.[fig : vib ] ) . we analyze it through partial wave packets , with main components corresponding to the various levels @xmath7 . each one is reaching the inner turning point at a time which depends upon the excitation time @xmath60 and the vibrational period @xmath66 . in the photoassociation window considered here , the vibrational periods are in the range 350 - 196 ps , yielding @xmath67 = 175 ps and @xmath68 = 98 ps , respectively . for a negative chirp , we may choose the parameters of the pulse such as to create partial vibrational wave packets which reach the inner turning point approximately at the same time @xmath69 . + is populated at the beginning ( @xmath70 ) , while @xmath71 is populated at the end ( @xmath72 ) of the time window characterizing the photoassociation pulse . the magnitude of the chirp can be chosen such that , after half a vibration period @xmath73 , the partial wave packets arrive at the potential barrier at the same time @xmath74 . this time ( @xmath74 ) is optimal for radiative stabilization , with a second pulse , toward bound ground state levels . ] + for long range molecules , the vibrational period is known to obey a scaling law @xcite , and it is easy to choose the chirp parameter such that all the partial wave packets in the intermediate levels @xmath75 satisfy @xcite @xmath76 focussing of the wave packet is then obtained at time @xmath74 , after half a vibrational period . in contrast , for a positive chirp the lower levels are populated first , @xmath77 , and a spreading of the vibrational wave packet is expected . * several authors in the past have suggested ways of designing laser fields for the generation of spatially squeezed molecular wave packets @xcite . since they considered low vibrational levels , a numerical optimization procedure appeared to be necessary . here , we are dealing with long range molecules , where the motion is controlled by the asymptotic potential , and the availability of scaling laws makes the optimization easier , as for atomic rydberg wave packets @xcite . * we should note that in the chosen example , the vibrational progression and the scaling law of the vibrational period are slightly perturbed due to a tunneling effect for the levels in the vicinity of @xmath58=96 @xcite . this leads to a focussing time @xmath74=@xmath78 + 135 ps , slightly different from @xmath79 = 125 ps . moreover , the early population of levels which are perturbed by tunneling leads to population transfer to the inner well of the @xmath8 excited potential . this tunneling effect is not a general phenomenon , but very specific to the cesium @xmath80 molecular curve , and to the value chosen for the detuning . a detailed investigation would require consideration of two coupled channels in the region of the inner potential @xcite , beyond the scope of this work . in the present calculations , we start from the wave packet in the excited state and investigate how the population can efficiently be transferred to vibrational levels of the ground state by the second pulse . as in our previous work , this is achieved by studying the wave packet dynamics by numerical solution of the time - dependent schrdinger equation ( cf . @xcite for an overview over numerical methods and appendix [ sec : box ] for a discussion of the representation of continuum states ) . the scheme we suggest is illustrated in fig . [ fig : scheme ] ( left ) , where we have represented , at different times , the wave packet created by the negatively chirped picosecond pulse @xmath81 . we consider a vibrational wave packet @xmath82 , which has evolved on the excited state potential for time @xmath83 after the maximum of the pump pulse @xmath78 . in order to discuss directly transfer probabilities , we will renormalize the wave packet before applying the dump pulse , such that at any time @xmath84=1 . the variation of the photoassociation probability as reported in the last column of table [ tab : pump ] will be reconsidered in the conclusion . we first compute the franck - condon overlap matrix elements @xmath85 of the excited state wave packet with the bound ground state eigenfunctions ( cf . [ fig : scheme ] , right ) . the results markedly depend upon @xmath83 , and so does the probability to form stable molecules . for instance , just after the photoassociation pulse , at time @xmath78 + 40 ps @xmath86 ( circles in fig . [ fig : scheme ] , right ) , the franck - condon overlap factors are negligible for most bound levels of the ground state , except for the last level @xmath87 which is very loosely bound ( 5 @xmath88 10@xmath89 @xmath0 @xcite ) . in contrast , half a vibrational period later , at time @xmath90135 ps when the wave packet has moved to the inner turning point ( diamonds in fig . [ fig : scheme ] , right ) , a significant overlap is obtained with various deeply bound levels of the ground state , in particular with @xmath91 14 , 20 . ( b ) between the time - dependent wave packet in the excited state and all bound vibrational levels of the ground state . @xmath78 corresponds to @xmath92 . it is clear that the overlap with deeply bound ground state levels ( around @xmath93=14 ) is maximum when focussing of the excited state wave packet occurs , 135 ps or half a vibrational period after @xmath78 . in contrast , when the wave packet is at its outer turning point just after the pa pulse and again at about @xmath90250 ps ( one vibrational period @xmath73 ) , the overlap with only the last bound levels is significant . moreover , tunneling to and trapping of population in the inner potential well lead to overlap with ground state levels around @xmath94 : this is specific to the cesium @xmath95 state and to the value chosen for the detuning . note the stability of this population during half a vibrational period . ] the large overlap with @xmath96 can be explained by the coincidence of the inner classical turning point of the level @xmath58=92 and the outer turning point of @xmath93=14 , such that @xmath58=92 @xmath97 @xmath93=14 is a franck - condon transition . this large value is restricted to a short ( @xmath98 ps ) time window around @xmath99 ps as illustrated in fig . [ fig : schemenew ] , where the time - dependence of the franck - condon overlap is shown . note that the overlap with @xmath94 is due to population in the inner well of @xmath5 due to tunneling @xcite . from fig . [ fig : schemenew ] , we can indeed see that this population remains trapped at the considered timescale . the above mentioned large overlap with the last vibrational level of the ground state just after the photoassociation pulse ( @xmath100 ) is seen to reappear after one vibrational period , when the wave packet reaches again the outer turning point . however , we will not choose one of the last bound levels of the ground state as our target vibrational level @xmath93 , since a significant population is already achieved in a one - color experiment @xcite , making the implementation of a two - color scheme superfluous . in fact , the specific advantage of a pump - dump experiment is the possibility of stabilization into deeply bound levels in the ground state . we will therefore focus on populating the level @xmath101 with binding energy @xmath102 @xmath0 , and we will consider a second ( or dump ) pulse @xmath103 with parameters adapted to efficient population transfer toward @xmath101 , in particular with a time delay close to 135 ps . for the dump pulse @xmath104 , we define a central frequency @xmath105 , a maximum intensity @xmath17 , a duration ( fwhme ) of the transform - limited pulse @xmath106 , the time of maximum amplitude @xmath107 , and a stretching factor @xmath108 characterizing the strength of the chirp . the choice of the second color is dictated by the choice of the target levels . for @xmath109 ( cf . [ fig : scheme ] , right ) , we have chosen a central frequency @xmath110 @xmath0 , which is resonant with @xmath111 for a binding energy of the excited state wave packet equal to 2.1 @xmath0 . this binding energy , computed as @xmath112 , where @xmath113 is the kinetic energy operator , stays time - independent once the photoassociation pulse is over . it is found 0.5 @xmath0 smaller than the binding energy of the level @xmath114 which is resonant with the central frequency of the pump pulse . this difference reflects the non - symmetric population of the @xmath7 levels relative to @xmath22 , the photoassociation process being much more efficient for the upper levels . the spectral width of the second pulse @xmath115 is limited by the requirement that mostly bound levels should be populated in the ground state . since @xmath93=14 is bound by 113 @xmath0 , the energy spread of the vibrational wave packet in the excited state can be neglected and the maximum spectral width is estimated as 2@xmath88113 @xmath0 . therefore , for short pulses @xmath104 ( fwhme @xmath116 fs ) , population transfer to the continuum is to be expected leading to the formation of pairs of `` hot '' atoms . this conclusion relies on overlap integrals and needs to be corrected by the possible energy dependence of the transition dipole moment @xmath117 . finally , if we wish to selectively excite the level @xmath101 , the spectral width is limited by the vibrational energy splitting around @xmath101 which is about 7 @xmath0 . the typical pulse temporal width @xmath118 of the dump pulse should therefore be larger than 2 ps . in the following , we will discuss two schemes to form ground state molecules : 1 . for short pulses in the 10 - 300 fs range , important population transfer to several bound levels of the ground electronic state can be obtained ( cf . sec.[sec : many ] ) . the goal is then to maximize the total population transfer . the intensity , duration and possibly stretching factor , i.e. chirp , of the @xmath119 pulses will be systematically varied . 2 . for longer pulses in the picosecond domain , transfer of population to a single vibrational level , namely @xmath93=14 , is possible ( see sec.[sec : single ] ) . the goal is then to optimize selective population transfer to a chosen level . the third parameter is the time - delay between the two pulses , defined as @xmath120 . the discussion can be guided by the following arguments , illustrated in fig . [ fig : probability ] : , @xmath121 . ( bottom ) the upper limit to the probability of forming bound molecules is given by the sum over franck - condon overlap matrix elements between the time - dependent excited state wave packet ( assumed normalized to one ) and all the bound levels of the ground state . the highest probability of about 20% is expected for the focussed wave packet ( at time @xmath122 ps ) . ] * first , we can use a simple model for the choice of the time delay @xmath123 . considering the wave packet @xmath124 , it is possible to estimate an upper limit to the probability of forming molecules from the sum over the franck - condon overlap matrix elements @xmath125 . after a short and weak dump pulse at time @xmath107 , the wave packet created in the ground state is estimated by @xmath126 where @xmath127 denotes jointly the sum over bound levels with eigenenergies @xmath128 and the integral over continuum states with energy @xmath129 , while @xmath130 is the fourier transform of the field @xmath131 and @xmath3 the dipole moment , assumed to be @xmath132independent . ( [ eq : psig_perturb ] ) implies that vibrational levels and continuum states which are resonant within the bandwidth @xmath133 of the pulse are excited . the population transfer is proportional to the time - dependent franck - condon factors discussed above . assuming a @xmath134-pulse ( i.e. a pulse which leads to population inversion ) with bandwidth covering all ground state vibrational levels , the maximum probability is given by @xmath135 , restricting the sum in eq . ( [ eq : psig_perturb ] ) to the bound levels . the highest probability , 0.2 , is expected for a delay of 135 ps ( cf . [ fig : probability ] ) , when the wave packet @xmath136 is focussed after half a vibrational period . note the relatively short time window in which the probability is maximum . a secondary maximum is visible after one vibrational period ( 250 to 300 ps ) when the last levels are populated : much less population can then be transferred . note also the revival at @xmath137 ps , after one and a half vibrational period , with a larger width . the advantage of focussing is further manifested by comparing to the franck - condon overlaps of wave packets created by the pump pulses @xmath138 and @xmath121 . the maximum probability is now less than 10% , instead of 20% for the focussed wave packet , and it is spread during a larger time window . * next , we report time - dependent calculations for various dump pulses where the field is treated non - perturbatively and we start from a focussed wave packet . intensity and spectral width are varied ( see below sec . [ subsec : max ] ) in order to check the validity limits of the above mentioned simple model which relies on the impulsive approximation and weak fields . then a chirp of the dump pulse is introduced to obtain a robust scheme for population inversion ( cf . [ subsec : chirp ] ) . the dependence on the time delay is studied in sec . [ ssec : timeup ] . we furthermore present calculations starting from vibrational wave packets created with other photoassociation pump pulses such as @xmath139 , @xmath121 ( sec . [ ssec : signchi ] ) and considering a smaller detuning ( sec . [ ssec : detun2 ] ) . when the aim is to transfer as much population to the ground state as possible , the discussion of the pulse parameters is inspired by the arguments on population inversion in molecular electronic states developed in refs . @xcite . the basic idea is that for pulses which are short on the timescale of the vibrational dynamics , the wave packet does not move during the pulse ( impulsive approximation ) . the molecular system can then be reduced to an effective two - level system , and it is possible to find a @xmath134-pulse which leads to total population transfer . a @xmath134-pulse is defined as a pulse ( assumed here to be transform - limited and with gaussian envelope ) whose rabi angle , @xmath140 is equal to an odd multiple of @xmath134 . a pulse with rabi angle equal to an even multiple of @xmath134 induces zero population transfer . introducing a chirp enforces adiabatic following conditions for all rabi angles @xmath141 and therefore suppresses the oscillations in population transfer as a function of the rabi angle @xcite . the time - dependent schrdinger equation describing the evolution of the wave packets @xmath65 and @xmath142 in the excited and in the ground state , coupled by the radiative field of the dump pulse is solved numerically . our treatment of the field is non - perturbative ( cf @xcite ) and for intensities corresponding to rabi angles @xmath143 deviations from the estimate of eq . ( [ eq : psig_perturb ] ) are expected . deviations will also occur if the impulsive approximation ceases to be valid , i.e. for longer pulses . in this case , nuclear motion during the pulse can not be neglected and the wave packet is not simply projected to the electronic ground state . for all calculations presented in this section , the parameters of the short dump pulses are recalled in table [ tab : dump_short ] . the initial condition is given by the focussed excited state wave packet at @xmath122 ps , normalized to one such that population corresponds directly to probability . the propagation is carried out in the time interval @xmath144 $ ] with @xmath145 . the final population in the ground state @xmath146 is given by @xmath147 . @xmath148&@xmath118 & @xmath108&@xmath149 & delay @xmath120 & figs . + 11 843.5 @xmath0&2000 - 70 @xmath0 & 10 - 300 fs & 1 & 0 & 135 ps & [ fig : rabi],[fig : vibrabi ] + 11 843.5 @xmath0 & 200 , 70 @xmath0 & 100 , 300 fs & 2 , 5 , 10 & 120 , 21.7 , 5.52 ps@xmath150 & 135 ps & [ fig : rabichirp ] to [ fig : overlapsign ] + 11 845.0 @xmath0 & 200 , 70 @xmath0 & 100 , 300 fs & 1 & 0 & 135 ps & [ fig : overlapdetun ] + in fig . [ fig : rabi ] , we analyze the efficiency of the population transfer by varying the intensity of a transform - limited dump pulse for several durations @xmath118 , in the range of 10 - 300 fs . the approximation of an effective two - level system is shown to hold for all pulse durations , since oscillations as a function of the rabi angle are clearly manifested in the final state population @xmath146 and in the population of bound levels @xmath151 . the contrast of these oscillations is maximum for @xmath118=10 fs , where total population inversion is obtained with @xmath146 oscillating between 0 and 1 . for longer pulses ( @xmath118= 300 fs ) , only 10% of the population is transferred to the ground state . however , an analysis of the wave packet in the ground state shows that for the shorter pulses , the population is spread between continuum and bound levels . the probability to form bound molecules is at most 20% ( @xmath118=10 fs ) , so that 80% of the photoassociated molecules are dissociated by the dump pulse , giving pairs of hot atoms which usually leave the trap . with longer pulses , due to their narrower frequency bandwith , the transfer to continuum levels is no longer possible : for @xmath118=300 fs the 11% population transferred to the ground state is totally stabilized into bound levels . the maximum population transfer and its efficiency are reported in table [ tab : pop ] , demonstrating that in the intermediate range of pulse widths , up to 100 fs , a stabilization of @xmath152 20% of the photoassociated molecules is to be expected . ) as a function of the rabi angle , i.e. intensity . for all pulse durations shown , we find pronounced rabi oscillations , i.e. the approximation of an effective two - level systems holds . very short pulses ( 10 fs ) are able to induce complete population inversion , but the population which ends up in bound levels is limited to about 20% ; the dump pulse is then dissociating 80% of the photoassociated molecules . longer pulses ( 300 fs ) populate only bound levels and no continuum states , since , within their bandwidth , only bound levels are on resonance . ] . ground state populations after a dump @xmath134-pulse starting from a unity normalized wave packet in the excited state ; ratio of population in bound levels to overall ground state population after @xmath134-pulse ( cf . [ fig : rabi ] ) [ cols="<,^,^,^,^,^",options="header " , ] the possibility of stabilizing 20% of the photoassociated molecules is very appealing . however , the vibrational distribution of the population transferred to the ground state should also be discussed , since stable molecules should be formed in low vibrational levels . the wave packet created in the ground state after a @xmath134 pulse is analyzed through projection on the various @xmath153 vibrational wave functions , cf . fig . [ fig : vibrabi ] . for short enough pulses ( @xmath118 = 10 fs , 100 fs ) the distribution follows the variation of the franck - condon overlap discussed above ( cf . fig . [ fig : scheme ] , right ) . the difference between the 10 fs and 100 fs pulses affects mostly the population of continuum states ( not shown in the figure ) . due to the narrower bandwith , the longer pulse with @xmath118 = 300 fs is only populating levels up to @xmath93=25 , and no continuum levels . in all cases the levels around @xmath93=14 are preferentially populated . -pulses ( cf . fig . [ fig : rabi ] ) : the distributions after a 10 fs and 100 fs pulse are almost identical to the franck - condon overlap of the focussed excited state wave packet with ground state bound levels . the narrower bandwidth of a 300 fs pulse allows for less levels to be populated ( cut - off around @xmath154 and @xmath93=25 ) ; furthermore , no continuum states are excited ( cf . table [ tab : pop ] ) . ] the efficiency of the stabilization process would of course be reduced by considering rabi angles different from @xmath134 . therefore , for a given pulse duration , the scheme is very sensitive to the intensity of the dump pulse . this can be avoided by introducing _ chirped _ dump pulses @xcite . in fig . [ fig : rabichirp ] , we report numerical results obtained after chirping the dump pulse , starting from transform - limited pulses of 100 fs and 300 fs , respectively . the streching factor @xmath108 is varied from 1 to 10 , and the linear chirp parameter @xmath155 is chosen to be negative . note that the rabi angles in fig . [ fig : rabichirp ] only characterize the pulse intensities , they are those of the corresponding unchirped pulses ( the rabi angle becomes complex for chirped pulses ) . fig . [ fig : rabichirp ] clearly shows that chirping suppresses the variation of the final population as a function of the rabi angle . for @xmath156 100 fs and @xmath157 complete population inversion is reached at high intensities due to non - perturbative effects such as power broadening ( top left ) . however , as for the unchirped pulses , the population of bound levels of the ground state remains limited to about 20% ( bottom left ) . the advantage of chirping consists in this value of 20% being obtained for a wide range of rabi angles ( @xmath134 to 4 @xmath134 in our calculations ) . for @xmath118= 300 fs , nearly complete population inversion is achieved for stretching factors @xmath158 , 5 , and the probability of populating bound levels becomes close to 20% , as for the shorter pulses . the chirp parameter should , however , not be too large : for a strongly chirped pulse ( @xmath159 ) with corresponding transform - limited fwhm @xmath118= 300 fs , the population transfer decreases , and oscillations reappear . this is due to the breakdown of the effective two - level system approximation : with a pulse stretched to 3 ps , it is no longer justified to neglect the relative motion of the nuclei during the pulse . . the duration of the transform - limited pulses is @xmath118= 100 fs ( left ) and @xmath118 = 300 fs ( right ) . results for an unchirped pulse are plotted for comparison ( black circles ) . ] in fig . [ fig : rabichirp300fs ] , we compare the final ground state population after chirped dump pulses with @xmath160 for positive and negative chirp . the stretching factor is again varied from 1 to 10 . for pulses with corresponding transform - limited fwhm @xmath118= 100 fs , no difference between negative and positive chirp was observed ( data not shown ) . for @xmath118=300 fs , pulses with negative chirp perform slightly better than those with positive chirp . at high intensities ( @xmath161 ) , differences between open and filled symbols are seen , i.e. even fairly narrow - bandwidth pulses start to excite continuum levels . . the results for an unchirped pulse are plotted for comparison ( black circles ) . ] the effect of the chirp of the dump pulse on the final vibrational distribution in the ground state is analyzed in fig . [ fig : overlapchirp ] . , [ fig : rabichirp ] ) with @xmath162 fs ( top ) and @xmath163 fs ( bottom ) for two different intensities corresponding to @xmath164 ( red squares ) and @xmath165 ( blue triangles ) . the vibrational distribution after an unchirped @xmath134-pulse is shown for comparison ( black circles ) . ] a chirped pulse with intensity corresponding to @xmath166 ( red squares in fig . [ fig : overlapchirp ] ) leads to very a similar vibrational distribution as an unchirped @xmath134-pulse ( black circles ) for both pulse durations . for higher intensities ( @xmath167 , blue triangles ) , non - perturbative effects start to play a role , and the vibrational distribution is changed : for example , the maximum is not anymore observed for the level with the largest franck - condon factor ( @xmath101 ) . furthermore for @xmath163 fs , the approximation of an effective two - level system starts to break down due to the long pulse duration . while at higher intensity the dynamics becomes harder to interpret , the overall probability to form bound molecules is increased from 18% ( @xmath166 ) to 21% ( @xmath167 ) for @xmath162 fs and from 11% to 16% for @xmath163 fs . after excitation by the photoassociation pulse at long range , the wave packet is accelerated toward shorter distances until it reaches the inner barrier of the external potential well ( solid blue line in fig . [ fig : excdyn ] , top panel ) . while a small part of the wave packet tunnels into the inner potential well , most of it is reflected at the barrier and moves back toward larger distances ( dashed red line ) . ( solid blue line ) and @xmath168 ( dashed red line ) ( @xmath78 denotes the time where the pump pulse has its maximum ) . bottom : the franck - condon overlap of the excited state wave packets with the ground state bound ( left ) and the continuum ( right ) levels . ] the franck - condon factors reflect this time - dependence of the excited state wave packet ( fig . [ fig : excdyn ] , bottom panel , see also fig . [ fig : schemenew ] ) . short and sufficiently intense pulses ( @xmath162 fs , @xmath166 ) allow for inducing maximum population inversion in both cases , and the final ground state vibrational distributions completely mimic the franck - condon factors ( data not shown ) . however , only the wave packet at shorter distances has good franck - condon overlap with bound ground state levels , and ground state molecules are formed by the stabilization pulse . for the excited state wave packet at later times , only the last two bound levels of the ground state are populated ( except for tunneling ) , and a considerable amount of population is transferred to the continuum , i.e. two free atoms are created . fig . [ fig : excdyn ] confirms our initial assumption that stabilization to ground state molecules is optimal when the excited state wave packet is close to the inner turning point of the outer potential well . excited state wave packets after pump pulses with positive ( @xmath169 , red dashed lines ) and negative chirp ( @xmath170 , blue solid lines ) are compared in fig . [ fig : overlapsign ] at the time of maximum focussing at the inner barrier . ps for positive and negative sign of the pump pulse chirp ( @xmath170 and @xmath169 ) . ( bottom ) the franck - condon overlap of the excited state wave packet with the ground state bound ( left ) and the continuum ( right ) levels . ] the sign of the chirp determines how well the excited state wave packet can be focussed ( cf . top panel of fig . [ fig : overlapsign ] ) . for positive chirp , the wave packet is less focussed , and the maximum probability to form bound molecules is 6% compared to 21% for negative chirp . furthermore , almost half of these 6% are due to the specific topology of the @xmath80-potential of cs@xmath9 , i.e. tunneling to the inner well . focussing as obtained with @xmath170 on the other hand represents a general scheme . as in the previous sections the vibrational distributions after the dump pulse ( data not shown ) completely mimic the franck - condon overlap factors for pulses which induce maximum population transfer ( e.g. @xmath162 fs , @xmath166 ) . the population of bound levels is then 4.9% for @xmath169 to 18% for @xmath170 . [ fig : overlapdetun ] compares the excited state wave packet for two different detunings of the pump pulse ( @xmath170 and @xmath29 , respectively ) . for two different detunings of the pump pulse . bottom : the franck - condon overlap of the excited state wave packet with the ground state bound ( left ) and the continuum ( right ) levels . ] for smaller detuning ( @xmath29 ) , higher vibrational levels of the excited state potential get populated . due to the larger spread of vibrational periods , it is harder to fulfil the condition , eq . ( [ eq : focus ] ) , and the wave packet is less focussed ( cf . top panel fig . [ fig : overlapdetun ] ) . furthermore , no tunneling to the inner well of the @xmath80 state occurs for smaller detuning . hence , the peak in the franck - condon overlap factors around @xmath94 is missing . pulses which achieve maximum population inversion to the ground state can still be found ( e.g. @xmath162 fs , @xmath166 ) . however , the probability of populating bound ground state levels is at 3.4% considerably smaller . the efficiency of the complete two - color scheme is of course given by the efficiencies of pump _ and _ dump steps . one has to keep in mind that since the probability of the photoassociation step is higher for @xmath29 ( see table [ tab : pump ] ) , the two effects compensate and the final number of bound molecules is the same . in order to achieve selective population of the level @xmath101 , we consider longer pulses , with parameters reported in table [ tab : dump_long ] . @xmath30 & @xmath118 & @xmath108 & @xmath155 & delay @xmath120 & fig . + 11 843.5 @xmath0 & 20 - 7 @xmath0 & 1 , 2 , 3 ps & 1 & 0 & 135 ps & 13 + 11 843.5 @xmath0 & 7 @xmath0 & 3 ps & 1.5 , 2 & 1.84 , 1.2 ps@xmath150 & 135 ps & 14 + as explained in sec . [ ssec : param ] , pulses with corresponding transform - limited fwhm @xmath171 ps are expected to populate a single vibrational level . [ fig : single ] reports the final ground state population after dump pulses with @xmath172 ps ( black circles ) , 2 ps ( blue diamonds ) and 3 ps ( red triangles ) as a function of the rabi angle , i.e. pulse intensity on the left , and corresponding vibrational distributions on the right . ps ( black circles , 2 ps ( blue diamonds ) , and 3 ps ( red triangles ) as a function of the rabi angle . right : vibrational distributions of the final ground state wave packet corresponding to the points indicated by arrows in the left panel . ] a pulse with @xmath173 ps and moderate intensity indeed excites exclusively a single vibrational level ( filled red triangles in fig . [ fig : single ] , right ) . the population of this level is @xmath174 compared to the total population of @xmath175 . a broadening of the vibrational distribution is only observed at fairly high intensity ( @xmath176 , empty red triangles ) . for @xmath177 ps , broadening of the vibrational distribution sets in already at lower intensities ( @xmath178 , empty black circles ) . for @xmath172 ps , oscillations of the final population with the rabi angle still persist , albeit with much lower contrast than for the broad - bandwidth pulses of sec . [ sec : many ] . for longer pulses , the impulsive approximation completely ceases to be valid , and the behavior can not anymore be understood ( and employed for control ) in terms of an effective two - level system . chirping prolongs the pulses even further , and its effect on the final populations and vibrational distributions is investigated in fig . [ fig : single_chirp ] which compares moderately chirped pulses ( stretching factors @xmath179 , red diamonds and @xmath158 , blue triangles ) to the unchirped case ( black circles ) for @xmath173 ps . the final ground state population ( blue triangles , @xmath158 ) is decreased after a chirped pulse as compared to the unchirped case . however , a slight negative chirp ( filled red diamdonds , @xmath179 ) does not significantly influence the population transfer . in general , a range of intensities corresponding to rabi angles @xmath180 exists where 10% or more of the excited state wave packet are transferred to the ground state , almost exclusively to the vibrational level @xmath101 ( cf . [ fig : single_chirp ] , right ) . this result implies that some chirping ( which might always occur in an experimental setup ) does not perturb stabilization into a single level , i.e. the scheme is expected to be fairly robust . the formation of ultracold molecules is presently a very active field . previous work has demonstrated the advantage of using chirped laser pulses to create ultracold molecules by photoassociation in an assembly of cold atoms @xcite . the molecules are created in an electronically excited state and are short - lived . a stabilization step must therefore be implemented . the present theoretical work has discussed a two - color pump - dump experiment where a second pulse , delayed from the photoassociation ( pump ) pulse , transfers population to deeply bound levels of the ground state , creating stable molecules . following ref . @xcite , the pump pulse ( with picosecond duration and maximum at time @xmath78 ) photoassociates cs atoms populating , via an adiabatic transfer mechanism , about 15 vibrational levels in the external well of the @xmath181 ) potential curve . a vibrational wave packet is created which is bound by 2.1 @xmath0 and localized in the region of the outer turning point . after the pulse , the wave packet moves toward the inner barrier of the potential curve ( located at @xmath182 ) . the chirp parameter of the pump pulse can be chosen such that after half a vibrational period , at time @xmath78 + @xmath183 , this wave packet is focussed at the barrier @xcite . our analysis of the franck - condon overlap integrals has shown that the overlap with deeply bound levels @xmath93 of the ground triplet state is then maximum . a vertical transition toward the level @xmath93=14 , bound by 113 @xmath0 , is even possible , since the outer turning point of the corresponding wave function coincides with @xmath184 . we therefore have discussed how population can efficiently be transferred to the ground triplet state by a short dump pulse which should be in the femto- or picosecond domain and optimally delayed from the pump pulse by @xmath183 . the validity of this intuitive model has been checked with numerical calculations . we have considered two kinds of pulses : broadband pulses populating many vibrational levels of the ground state , with the aim of maximizing the population transfer , and longer pulses for selective population of a single level , @xmath93=14 . the first set of calculations has been performed for a time delay of @xmath183 . we have shown that for short pulses and a rabi angle equal to @xmath134 , total population inversion can be obtained , i.e. the vibrational wave packet in the excited state is completely transferred to the ground state . analysis of the wave packet shows that a maximum of 20% of this population goes to bound levels . the remaining 80% correspond to continuum states , i.e. the dump pulse also creates pairs of hot atoms . the vibrational distribution reflects the variation of the franck condon overlap . oscillations in the population transfer as a function of the rabi angle ( i.e. laser intensity for a fixed pulse duration ) can be suppressed by chirping the dump pulse . thus , total population inversion is obtained for a wide range of intensities ( corresponding to rabi angles of @xmath134 to @xmath185 in our calculations ) with an effective 20% population transfer to bound levels . employing a chirped dump pulse therefore yields a robust scheme in which the probability to create stable ground state molecules is 20% , independent of the exact value of laser intensity . by varying the time - delay between the two pulses , a strong variation in the population transfer which follows the variation of the franck - condon factors has been observed . only at shorter distance , the wave packet in the excited state has good franck - condon overlap with bound ground state levels , and ground state molecules are formed by the dump pulse . the highest efficiency is achieved in a narrow time window ( @xmath186ps ) during which the wave packet is maximally focussed at the barrier . for different time - delays , only the last two bound levels of the ground state are populated , and a considerable amount of population is transferred to the continuum , i.e. two free atoms are created . note that without a second pulse , stabilization would occur via spontaneous emission on a nanosecond timescale . the probability for creating ground state molecules is then given by a time average . since the wave packet spends more time at large distances than close to the barrier , times with unfavorable franck - condon overlaps dominate this average . _ it is therefore essential to employ a second pulse to fully take advantage of the time - dependent probability to create bound ground state molecules after photoassociation with picosecond pulses . _ the calculations have confirmed our intuition that stabilization to ground state molecules is optimal when the excited state wave packet is close to the inner turning point of the outer potential well . the best localization of the wave packet at the inner barrier is brought about by negatively chirping the pump pulse . there is yet another advantage when using a negative chirp of the pump pulse : the photoassociation probability is increased by a factor of 8 as compared to a transform - limited pulse . furthermore , the probability of creating bound molecules by the dump pulse is enhanced by roughly a factor of 3 for a negatively chirped pump pulse as compared to positively chirped or transform - limited ones . the number of stable molecules formed in a pump - dump sequence can be therefore be increased by 24 when appropriately choosing the pulse parameters . we have also considered a pump pulse with smaller detuning , for which the photoassociation probability is larger . however , due to a decrease of the franck - condon overlap , even for the maximally focussed wave packet , the stabilization step is much less efficient . therefore , the total number of stable molecules is roughly the same as in the previous case . finally , we have employed longer dump pulses with a duration of about 3 ps to selectively populate the @xmath93=14 level in the ground state . up to 12% of the photoassociated molecules can then be stabilized into a _ single _ level , which is fairly efficient . note that these molecules are created coherently by the pump - dump sequence . the full discussion of the definition of the optimal pulse should include the possibility of a repetition scheme , considering further pump and dump pulses . in treating a repetition scheme , we also have to take into account how the pump and dump pulses modify the continuum wave function describing two colliding atoms . this is beyond the scope of the present study . in conclusion , a two - color pump - dump scheme is rather promising for the creation of stable cold molecules . due to the choice of the cs@xmath9 0@xmath187 potential for the photoassociation step , the present work has analyzed formation of molecules in the @xmath93=14 level of the ground triplet state , bound by 113 @xmath0 . many other schemes , e.g. @xcite , are possible and should be considered to bring population into the @xmath93=0 level of the ground state which will open the way to the formation of stable molecular condensates . in this respect , we would like to point out the generality of the two - color pump - dump scheme which only requires efficient photoassociation and stabilization mechanisms involving either long range wells in the excited potentials @xcite or resonant coupling @xcite , but is not restricted to a particular system . we would like to acknowledge fruitful discussions with ronnie kosloff and to thank anne crubellier for computing independently overlap integrals with a numerov method in order to check our numerical results . this work has been supported by the deutsche forschungsgemeinschaft ( c.p.k . ) and the european union in the frame of the cold molecule research training network under contract hprn - ct-2002 - 00290 . the fritz haber center is supported by the minerva gesellschaft fr die forschung gmbh mnchen , germany . laboratoire aim cotton belongs to fdration lumire matire ( cnrs , universit paris xi ) . the fourier grid employs a finite number of grid points and periodic boundary conditions to represent a quantum wave function . for @xmath188 , this corresponds to the particle in a box problem . the scattering continuum is hence represented by a finite number of discrete box states . the highest momentum which is correctly represented on the grid is given in terms of the ( constant ) spacing @xmath189 between two grid points , @xmath190 . in order to model long - range interactions which require large grids but not small grid steps , a mapping procedure is introduced . the grid steps in the asymptotic region become very large , correspondingly high momenta are cut off . this alters the representation of continuum states on the grid . when diagonalizing the hamiltonian , two kinds of continuum eigenfunctions are obtained : true box levels and wave functions with an unphysical shape . the latter will be called `` cutoff - states '' in the following . the box levels can be easily identified by the scaling of their eigenenergies , @xmath191 ( except for the first few states just above the dissociation limit where the energies are perturbed by the threshold ) . the corresponding wave functions are standing waves of the box . the eigenenergies of the cutoff - states approximately scale with some high power ( @xmath192 ) of the box index , the highest eigenenergy being determined by the largest allowed grid step which is a parameter of the mapping . the corresponding eigenfunctions show very fast oscillations and do nt obey the boundary conditions at @xmath55 and @xmath56 . it can nevertheless be shown that these cutoff - states constitute an _ effective _ representation of the true , physical continuum for energies higher than the energy of the last true box level . to this end , the franck - condon overlaps of an excited state wave function with the eigenfunctions of all ground state levels ( i.e. bound , box and cutoff - levels ) are calculated and compared ( i ) for different mapping parameters and ( ii ) to franck - condon overlaps obtained by numerov integration @xcite . in the first set of calculations , the largest allowed grid step , @xmath193 , is varied . the hamiltonian is diagonalized for each @xmath193 and an increasing number of grid points , @xmath194 . for each @xmath193 , the number of cutoff - states turns out to be independent of @xmath194 , i.e. increasing @xmath194 leads to a larger number of true box states . in particular , for the smallest @xmath193 and largest @xmath194 used ( @xmath195 a@xmath57 , @xmath196 ) , a continuum representation is obtained which contains almost no cutoff - states ( 168 vs. 7969 true box states ) . the excited state wave function is then interpolated to these grids , and the franck - condon overlaps are computed . note , that the franck - condon overlaps for continuum states need to include the correct energy renormalization @xcite . the franck - condon overlaps turn out to be roughly independent of the grid parameters . we are particularly interested in the sum over the franck - condon overlaps which determines the probability of dissociating the excited state molecules into hot , free atoms with a short dump pulse ( cf . [ sec : many ] ) . this sum is roughly constant for all grids considered , varying between 0.21 for @xmath197 a@xmath57 and 0.17 for @xmath195 a@xmath57 . the variation is caused by the interpolation of the excited state wave function to the new grids which is not exact , the approximation getting worse for decreasing @xmath193 . in a second set of calculations , franck - condon overlaps with all ground state levels were calculated for selected excited state vibrational levels with both the mapped fourier grid and the numerov method . oscillations like those in fig . [ fig : excdyn ] are observed . the resolution of these oscillations in the mapped fourier grid method depends on the number of grid points , @xmath194 . good agreement between the two methods is obtained for @xmath198 , the value used throughout secs . [ sec : many ] and [ sec : single ] . 10 j. doyle , b. friedrich , r. krems , and f. masnou - seeuws , eur . j. d , * 31 * , 149 ( 2004 ) e. donley , n. claussen , s. thompson , and c. wieman , nature * 417 * ( 2002 ) n. r. claussen , e. a. donley , s. t. thompson , and c. e. wieman , phys . rev . lett . * 89 * , 010401 ( 2002 ) j. herbig , t. kraemer , m. mark , t. weber , c. chin , h .- c . ngerl , and r. grimm , science * 301 * , 1510 ( 2003 ) k. xu , t. mukaiyama , j. r. abo - shaeer , j. k. chin , d. e. miller , and w. ketterle , phys . lett . * 91 * , 210402 ( 2003 ) s. drr , t. volz , a. marte , and g. rempe , phys . 92 * , 020406 ( 2003 ) c. a. regal , c. ticknor , j. l. bohn , and d. s. jin , nature * 424 * , 47 ( 2003 ) m. w. zwierlein , c. a. stan , c. h. schunck , s. m. f. raupach , s. gupta , z. hadzibabic , and w. ketterle , phys . lett . * 91 * , 250401 ( 2003 ) k. e. strecker , g. b. partridge , and r. g. hulet , phys . * 91 * , 080406 ( 2003 ) j. cubizolles , t. bourdel , s. j. j. m. f. kokkelmans , g. v. shlyapnikov , and c. salomon , phys . * 91 * , 240401 ( 2003 ) s. jochim , m. bartenstein , a. altmeyer , g. hendl , c. chin , s. riedl , j. h. denschlag , and r. grimm , science * 302 * , 2101 ( 2003 ) s. jochim , m. bartenstein , a. altmeyer , g. hendl , c. chin , j. h. denschlag , and r. grimm , phys . lett . * 91 * , 240402 ( 2003 ) m. greiner , c. a. regal , and d. s. jin , nature * 426 * , 537 ( 2003 ) a. fioretti , d. comparat , a. crubellier , o. dulieu , f. masnou - seeuws , and p. pillet , phys . rev . lett . * 80 * , 4402 ( 1998 ) j. m. sage , s. sainis , t. bergeman , and d. demille . phys . * 94 * , 203001 ( 2005 ) c. m. dion , c. drag , o. dulieu , b. laburthe tolra , f. masnou - seeuws , and p. pillet , phys . lett . * 86 * , 2253 ( 2001 ) o. dulieu and f. masnou - seeuws , j. opt . b * 20 * , 1083 ( 2003 ) a. j. kerman , j. m. sage , s. sainis , t. bergeman , and d. demille , phys . lett . * 92 * , 153001 ( 2004 ) j. vala , o. dulieu , f. masnou - seeuws , p. pillet , and r. kosloff , phys . a , * 63*,013412 ( 2001 ) . e. luc - koenig , r. kosloff , f. masnou - seeuws , and m. vatasescu , phys . a * 70 * , 033414 ( 2004 ) e. luc - koenig , m. vatasescu , and f. masnou - seeuws , eur . j. d * 31 * , 239 ( 2004 ) , phys . lett . * 80 * , 1406 ( 1998 ) , j. chem . * 113 * , 1898 ( 2000 ) m. j. wright , s. d. gensemer , j. vala , r. kosloff , and p. l. gould . lett . * 95 * , 063001 ( 2005 ) , j. chem . * 101 * , 8461 ( 1994 ) w. c. stwalley , j. chem . phys . * 58 * , 3867 ( 1973 ) i. averbukh and m. shapiro , phys . a * 47 * , 5086 ( 1993 ) d. g. abrashkevich , i. sh . averbukh , and m. shapiro , j. chem . * 101 * , 9295 ( 1994 ) j. ahn , t. c. weinacht , and p. h. bucksbaum , science * 287 * , 463 ( 2000 ) m. vatasescu and f. masnou - seeuws , eur . j. d * 21 * , 191 ( 2002 ) , optics express * 8 * , 238 ( 2001 ) k. willner , o. dulieu , and f. masnou - seeuws , j. chem . phys . * 120 * , 548 ( 2004 ) c. p. koch , j. p. palao , r. kosloff , and f. masnou - seeuws . rev . a , * 70 * , 013402 ( 2004 ) f. masnou - seeuws and p. pillet . adv . in at . phys . * 47 * , 53 ( 2001 ) anne crubellier , private communication ( 2005 )

the unitary correlation operator method ( ucom ) is employed to treat short - range correlations in both , homogeneous liquid and small droplets of bosoniche atoms . the dominating short - range correlations in these systems are described by an unitary transformation in the two - body relative coordinate , applied either to the many - body state or to the hamiltonian and other operators . it is shown that the two - body correlated interaction can describe the binding energy of clusters of up to 6 atoms very well , the numerical effort consisting only in calculating one two - body matrix element with gaussian single - particle states . the increasing density of bigger droplets requires the inclusion of correlation effects beyond the two - body order , which are successfully implemented by a density - dependent two - body correlator . with only one adjusted parameter the binding energies and radii of larger droplets and the equation of state of the homogeneoushe liquid can be described quantitatively in a physically intuitive and numerically simple way .

cond-mat0107007

the quantum - mechanical description of ground state properties of interacting many - body systems is a long - standing challenge in theoretical physics . in general the problem is two - fold : the first task is to specify the interaction between the constituents in terms of a two - body potential . the second is to solve the quantum - mechanical many - body problem with that interaction . for many - body systems composed of helium atoms the first task , the construction of an appropriate interatomic potential , is solved rather precisely . the hfdhe2 potential of aziz _ et al . _ @xcite is able to reproduce experimental transport coefficients in a wide temperature range . a simpler and less precise description of the interaction is given by the lennard - jones potential @xcite @xmath1\ ] ] with @xmath2 and @xmath3 . since the emphasis of the investigations presented in this paper is more on novel concepts , which remarkably simplify the treatment of the many - body problem , we will restrict ourselves to this simple and widely used parameterization . a common property of many realistic microscopic interactions like van der waals type potentials between neutral atoms or the nucleon - nucleon interaction in atomic nuclei is a strong short - range repulsion , which induces strong short - range correlations in the quantum many - body state . the repulsive core prevents the particles to approach each other closer than the radius of this core such that the two - body density matrix is strongly depleted for particle distances smaller than the core radius . this effect can not be described in a mean - field like picture or in terms of a superposition of a finite number of symmetrized product states . in this work we treat these correlations with the unitary correlation operator method ( ucom ) , that has already been applied successfully in the framework of the nuclear many - body problem @xcite . in the following section we introduce the general formalism of the unitary correlation operator method and derive analytic expressions for correlated wave functions and the correlated hamiltonian . in the third section we determine the optimal correlation operator for liquid @xmath0he from the two - body system and discuss the structure of the effective interaction generated by the unitary transformation . in section iv we investigate the homogeneous @xmath0he liquid with that correlator and introduce a density - dependent two - body correlator that describes very well the influence of three- and more - body correlations . finally , we successfully apply the density - dependent correlator to describe the ground state of small @xmath0he droplets . considering the great numerical simplification only a few one - dimensional integrals have to be calculated compared to monte - carlo - type calculations @xcite , the agreement of the results is excellent . the goal of the unitary correlation operator method ( ucom ) is to describe short - range interaction - induced correlations by a unitary transformation of an uncorrelated many - body state with @xmath4 particles . the unitary operator @xmath5 associated with this transformation can be written as an exponential of a hermitian generator @xmath6 @xmath7 .\ ] ] the generator is chosen to be an irreducible two - body operator @xmath8 which describes genuine two - body correlations only . in principle we could add a three - body part to account for genuine three - body correlations . for the moment we omit this step to develop the basic formalism on a simple level . the particular structure of the unitary transformation which should reflect the properties of the short - range correlations is described by the two - body generator @xmath9 . the repulsive core of the interaction leads to a suppression of the probability density within the range @xmath10 of the core . uncorrelated many - body states that are symmetrized products of single - particle states lead , however , to a two - body density that does not show any depletion for relative distances @xmath11 between the pair of particles . therefore the unitary transformation should shift those particles that are in the forbidden region of the repulsion to larger distances . this is done by means of a distance - dependent radial shift in the relative coordinate of two - particles . the hermitian generator of this shift in two - body space reads @xmath12 , \ ] ] where @xmath13 is the operator of the relative momentum , @xmath14 is the relative coordinate , and @xmath15 is the associated unit vector . the function @xmath16 describes the size of the shift as function of the distance of the particles . first we want to apply the unitary correlation operator @xmath5 to an uncorrelated many - body state @xmath17 out of a low - momentum model space . the correlated many - body state @xmath18 is defined by @xmath19 here and in the following all correlated quantities are marked by a tilde . in order to demonstrate the generic effect of the correlation operator we discuss first a two - body system with an uncorrelated two - body state @xmath20 . using the definition of the correlation operator @xmath5 and the generator @xmath9 we can evaluate the correlated two - body wave function in coordinate representation explicitly @xcite @xmath21 the respective transformation with the hermitian adjoint correlation operator @xmath22 leads to @xmath23 the correlation operator acts in the relative coordinate @xmath24 of the two - body system only , the center of mass coordinate @xmath25 is invariant . according to the construction of the the generator the unitary transformation corresponds to a radial shift in the relative two - body coordinate @xmath26 the transformed relative distances are given by the correlation functions @xmath27 and @xmath28 , respectively . due the unitarity of the correlation operator @xmath29 the correlation functions @xmath27 and @xmath28 are inverse to each other @xmath30 = r .\ ] ] the conservation of the norm of the correlated two - body wave function is guaranteed by the metric factor @xmath31 in and @xmath32 where @xmath33 denotes the derivative of @xmath34 . it corresponds to the square root of the jacobi determinant associated with the transformation of the normalization integral . the correlation functions @xmath34 that describe the coordinate transformation are connected to the function @xmath16 , which enters in the generator , by the integral equation @xmath35 for small shift distances the approximate relation @xmath36 holds . since all correlated quantities can be expressed directly in terms of the correlation functions @xmath27 and @xmath28 we will use one of these as basic quantity to describe the detailed structure of the short - range correlations . the shift function @xmath16 is only used for formal considerations and never specified explicitly . complementary to the definition of a correlated state we can use the correlation operator to define correlated operators . for an arbitrary observable @xmath37 the correlated operator @xmath38 is given by the similarity transformation @xmath39 it is evident that the formulations in terms of correlated states and correlated operators are equivalent when expectation values or matrix elements are calculated @xmath40 thus we can choose the formulation that is technically or intuitively better suited for the specific application under consideration . in most cases the formulation in terms of correlated operators is easier to apply in the many - body system . moreover it provides a systematic approximation scheme that will be discussed in section [ sec : ucom_clusterexp ] . as for the correlated wave function we first discuss specific correlated operators for a two - body system . the generalization to many - body systems will be done in the following sections . the simplest operator of interest is the local two - body potential @xmath41 that depends on the relative distance @xmath42 only . to evaluate the correlated potential we use the results on the correlated two - body wave function to determine a general two - body matrix element of the form @xmath43 & \quad= \int\!\!{\ensuremath{\mathrm{d}}}^3r\,{\ensuremath{\mathrm{d}}}^3\!x\;{\ensuremath { \big < { \psi } \big| \,{{\ensuremath{{\ensuremath{{{{\ensuremath{\bm{c}}}}}^{\dag}}}}}}\ , \big| { { \ensuremath{\vec{x}}},{\ensuremath{\vec{r } } } } \big > } } \,v(r)\ , { \ensuremath { \big < { { \ensuremath{\vec{x}}},{\ensuremath{\vec{r } } } } \big| \,{{\ensuremath{{\ensuremath{\bm{c}}}}}}\ , \big| { \psi ' } \big > } } \\[-2pt ] & \quad= \int\!\!{\ensuremath{\mathrm{d}}}^3r\,{\ensuremath{\mathrm{d}}}^3\!x\;{\ensuremath { \big < { \psi } \big| { { \ensuremath{\vec{x}}},{\ensuremath{\vec{r } } } } \big > } } \;v[{\ensuremath{r_+}}(r)]\ ; { \ensuremath { \big < { { \ensuremath{\vec{x}}},{\ensuremath{\vec{r } } } } \big| { \psi ' } \big > } } \\[2pt ] & \quad= { \ensuremath { \big < { \psi } \big| \,{v[{\ensuremath{r_+}}({\ensuremath{{\ensuremath{\bm{r}}}}})]}\ , \big| { \psi ' } \big > } } . \end{split}\ ] ] in going from the second to the third line we use equations and and substitute the integration variable @xmath24 by @xmath44 . since the two - body states are arbitrary this relation is valid on the operator level @xmath45 .\ ] ] thus the correlated two - body potential @xmath46 is given by the original potential with transformed radial coordinate @xmath47 . the correlated operator of the kinetic energy in the two - body system is slightly more complicated due to its momentum dependence . first we decompose the operator of the two - body kinetic energy @xmath48 in a center of mass contribution @xmath49 and an relative part @xmath50 @xmath51 the operator of the center of mass momentum @xmath52 commutes with the correlation operator , i.e. , the operator of the center of mass kinetic energy is invariant under the unitary transformation . the correlation operator acts on the relative part @xmath50 of the kinetic energy only . a similar calculation as for the local potential leads to the following structure of the correlated relative kinetic energy @xcite @xmath53 besides @xmath50 an additional momentum dependent term appears in the correlated relative kinetic energy , which can be formulated in terms of a tensorial effective mass correction . it is conveniently split into a correction for the radial part @xmath54 and a different correction for the angular component @xmath55 @xmath56 where @xmath57 is the operator of the relative two - body angular momentum . the distance dependent effective mass corrections @xmath58 and @xmath59 depend on the correlation function @xmath28 only @xmath60 in addition to these momentum dependent terms a local contribution @xmath61 appears in @xmath62 . \end{split}\ ] ] this additional two - body potential is also determined by the correlation function @xmath28 and its derivatives . in the preceding section we evaluated correlated operators explicitly in the two - body system . however , correlated operators in the many - body system contain additional terms . the unitary transformation of an operator , e.g. , an one - body operator like the kinetic energy or a two - body operator like the potential , generates a correlated operator , which contains irreducible contributions to all particle numbers @xmath63 } + { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{o}}}}}}}}^{[2 ] } + { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{o}}}}}}}}^{[3 ] } + \dots . \end{split}\ ] ] we use the notation @xmath64}$ ] for the irreducible @xmath65-body part of the correlated operator @xmath38 . this decomposition according to irreducible particle number is called cluster expansion @xcite . the cluster expansion gains physical meaning by the so called cluster decomposition principle @xcite : two localized subsystems , which are separated beyond the range of interactions , are independent of each other . therefore the state of the total system decouples into a direct product of the states of the two subsystems . this implies an analogous decomposition property of the correlation operator and of correlated operators . because of the cluster decomposition principle the cluster expansion is a natural starting point for approximations of the correlated operators . for a selected particle there will be a certain number of other particles within the range of the correlation depending on the density . the number of particles in this cluster gives the maximum order of the cluster expansion that contributes . we can truncate the cluster expansion at low orders if the range of the correlations is sufficiently small compared to the average distance of the particles . the simplest nontrivial approximation results from the truncation of the cluster expansion beyond two - body order . the correlated operator in two - body approximation reads @xmath66 } + { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{o}}}}}}}}^{[2 ] } .\ ] ] this approximation requires that the system is sufficiently dilute such that contributions of the three - body order of the cluster expansion are small . or in other words , the probability for three and more particles to be in the range of the repulsive core simultaneously has to be small . the technical advantage of the two - body approximation is that closed analytic expressions for the correlated operators can be deduced by just considering the two - body system . the correlated hamiltonian in two - body approximation is of the form @xmath67 } + { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{t}}}}}}}}^{[2 ] } + { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{v}}}}}}}}^{[2 ] } \\ & = \sum_i^n { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{t}}}}}}}}^{[1]}_i + \sum_{i < j}^n { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{t}}}}}}}}^{[2]}_{ij } + \sum_{i < j}^n { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{v}}}}}}}}^{[2]}_{ij } . \end{split}\ ] ] the one - body part of the correlated kinetic energy is simply the uncorrelated kinetic energy operator . the two - body part consists of the effective mass corrections and the additional local potential as determined in the previous section @xmath68 } & = { \ensuremath{{\ensuremath{\bm{t}}}}}\\ { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{t}}}}}}}}^{[2 ] } & = { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{t}}}}}}}}_{\omega } + { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{t}}}}}}}}_r + { \ensuremath{\tilde{u}}}({\ensuremath{{\ensuremath{\bm{r } } } } } ) . \end{split}\ ] ] the correlated potential has only a two - body contribution that is given by @xmath69 } = { \ensuremath{\tilde{v}}}({\ensuremath{{\ensuremath{\bm{r } } } } } ) .\ ] ] in order to get a rough measure for the validity of the two - body approximation we define a smallness parameter @xmath70 as a product of the density @xmath71 of the system and a typical volume in which the correlations between a pair of particles change their relative wave function . the correlation volume @xmath72 is defined by the norm of the defect wave function , i.e. , the difference between the uncorrelated uniform wave function @xmath73 and its correlated companion @xmath74 ^ 2 = \int\!\!{\ensuremath{\mathrm{d}}}^3r\ ; [ { \ensuremath{\mathcal{r}_+}}(r)-1]^2 .\ ] ] the smallness parameter @xmath75 is a measure for the probability to find a third particle within the volume where the correlations between two particles change their relative wave function significantly . the two - body approximation is valid only if this probability is small such that three - body correlations are negligible . we have shown for different physical systems @xcite that the relative contribution of the three - body order to the energy exceeds 10% if the smallness parameter reaches a value of typically @xmath76 . if the two - body approximation is not sufficient one can include higher orders of the cluster expansion successively . however , already the calculation of a three - body correlated wave function starting from the many - body correlation operator in analogy to section [ sec : ucom_wavefct ] is not practicable . therefore we reverse the procedure and start from a general many - body coordinate transformation in analogy to the correlation function @xmath34 and determine correlated many - body wave functions and operators . in a second step we connect the explicit structure of the coordinate transformation with the many - body correlation operator , at least in an approximate way . consider a @xmath4-body system with a collective coordinate vector @xmath77 . the short - range correlations are described by a @xmath4-body coordinate transformation @xmath78 the transformation function @xmath79 corresponds to the correlation function @xmath27 in the two - body case . similar to @xmath28 we define an inverse transformation function @xmath80 by @xmath81 = x .\ ] ] the correlated @xmath4-body wave function @xmath82 is defined via the coordinate transformation @xmath83 with a metric factor @xmath84 that ensures norm conservation . the metric factor is given by the square root of the jacobi determinant of the transformation @xmath85 with the matrix elements @xmath86_{ij } = { \ensuremath{\frac{\partial{x'_i}}{\partial{x_j } } } } = { \ensuremath{\frac{\partial{[{\ensuremath{\mathcal{x}_-}}(x)]_i}}{\partial{x_j } } } } .\ ] ] formally we can construct a unitary many - body correlation operator @xmath5 that generates this particular coordinate transformation @xmath87 the check of the unitarity relation @xmath88 with this definition is straightforward . using this formulation we can evaluate many - body correlated operators in the same way as shown in section [ sec : ucom_operator ] for the two - body approximation . we will only present selected results that are needed for the following investigations . the full formalism is discussed in @xcite . a general local potential @xmath89 , which depends on the coordinates @xmath77 of all @xmath4 particles , is transformed as in the two - body case @xmath90 , \ ] ] i.e. , the correlated potential is given by the original potential with transformed coordinate dependence . the expression for the @xmath4-body correlated kinetic energy operator @xmath91 is more involved @xmath92_{jk } \ ; { \ensuremath{{\ensuremath{\bm{p}}}}}_k + { \ensuremath{\tilde{u}}}({\ensuremath{{\ensuremath{\bm{x } } } } } ) , \ ] ] where @xmath93 denotes the collective momentum vectors of all @xmath4 particles . as in the two - body case the correlated kinetic energy contains an effective mass tensor and an additional local potential . the effective mass tensor is given by the square of the jacobi matrix @xmath94_{jk } = \frac{1}{2 m } \sum_{i=1}^{3n } [ j_-(x)]_{ji}\ , [ j_-(x)]_{ik } .\ ] ] the additional local potential @xmath95 given by @xmath96_{x ' = { \ensuremath{\mathcal{x}_+}}(x ) } .\ ] ] with these expressions it is in principle possible to calculate expectation values of the correlated hamiltonian up to arbitrary order in the cluster expansion . in practice all extensions beyond two - body approximation are very costly , thus only the three - body order will be considered . we will use the general formulation of the many - body coordinate transformation to estimate the contribution of the three - body order of the cluster expansion . since we can not derive the many - body coordinate transformation directly from the correlation operator we will construct the transformation by generalization of the known one in the two - body system . it is useful to reformulate the two - body coordinate transformation discussed in section [ sec : ucom_wavefct ] in terms of the general transformation function @xmath97 , where the upper index indicates the number of particles involved . the transformation of the relative coordinate with the correlation function @xmath34 is equivalent to the following transformation of the single - particle coordinates @xmath98 with a shift vector @xmath99\ , { \ensuremath{\hat{r}}}.\ ] ] this notation reveals the intuitive picture behind the description of short - range correlations by means of a coordinate transformation . as a consequence of the short - range repulsion the particles are displaced along their connecting axis by the shift vector @xmath100 . the two - body transformation can be readily generalized to describe three - body correlations . the corresponding three - body coordinate transformation @xmath101 is given by @xmath102 thus the coordinate of the first particle is transformed with respect to the second and the third particle simultaneously . as a necessary prerequisite this transformation obeys the cluster decomposition principle . if one of the three particles is separated beyond the range of the correlations , i.e. , to distances where @xmath100 vanishes , then the transformation reduces to the two - body transformation for the remaining pair . we will use this three - body transformation to calculate the local three - body contributions of the correlated hamiltonian , i.e. , the correlated potential and the local part of the correlated kinetic energy . the irreducible three - body part @xmath103}$ ] of the correlated potential is given by @xmath104 } & = { \ensuremath{{\ensuremath{{{{\ensuremath{\bm{c}}}}}^{\dag}}}}}({\ensuremath{{\ensuremath{\bm{v}}}}}_{12 } + { \ensuremath{{\ensuremath{\bm{v}}}}}_{13 } + { \ensuremath{{\ensuremath{\bm{v}}}}}_{23})\ , { \ensuremath{{\ensuremath{\bm{c}}}}}\\ & - ( { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{v}}}}}}}}^{[2]}_{12 } + { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{v}}}}}}}}^{[2]}_{13 } + { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{v}}}}}}}}^{[2]}_{23 } ) , \end{split}\ ] ] where @xmath105 is the uncorrelated two - body potential and @xmath106}_{ij}$ ] is the two - body part of the correlated potential . the first term describes the fully correlated two - body potential in a three - body system , which is given by a three - body coordinate transformation according to . the second term is the sum of the two - body correlated potentials . an analogous expression results for the local three - body part of the correlated kinetic energy @xmath107}$ ] , which involves and . the only input needed to evaluate correlated quantities explicitly is the correlation function @xmath34 . since @xmath34 reflects the properties of the correlations induced by the two - body interaction it can be extracted from the interacting two - body system . one method to construct an optimal correlation function is based on the exact solution of the two - body schrdinger equation . the exact relative wave function @xmath109 contains all information on interaction induced two - body correlations . for the case of energy @xmath108 and relative angular momentum @xmath110 the solid line in figure [ fig : opt_e0solution ] shows the radial wave function ( solid line ) for the lennard - jones potential ( thin dotted line ) . the short - range part is dominated by a huge correlation hole , i.e. , a region where the strong repulsive core of the potential enforces a vanishing wave function . we can extract a correlation function that describes these short - range correlations by requiring that the exact solution is reproduced by a correlated ansatz wave function @xmath111 for all @xmath112 inside some maximum radius @xmath113 . thus the optimal correlation operator maps the short - range part of the uncorrelated wave function @xmath114 onto the exact solution @xmath115 . in coordinate representation this condition leads with and to @xmath116 which can be written as an implicit integral equation for the optimal correlation function @xmath28 . ] @xmath117 ^ 3 = 3 \int_0^r\!\ ! { \ensuremath{\mathrm{d}}}\xi\;\xi^2 \bigg [ \frac{\phi_{0}(\xi)}{\phi_{{\ensuremath{\mathrm{ex}}}}[{\ensuremath{r_+}}(\xi ) ] } \bigg]^2 .\ ] ] the uncorrelated wave function @xmath118 should be chosen in accordance with the uncorrelated many - body trial state . for example , if a condensate of @xmath4 uncorrelated bosons is described by a @xmath4-fold product of a gaussian - shaped single - particle wave functions the relative wave function for each pair of bosons is again of gaussian shape . the choice of the uncorrelated wave function @xmath118 determines , which structures of the exact solution are generated by the correlation operator and which have to be described by the ansatz state . in order to isolate short - range correlations the uncorrelated state should be able to describe the long - range behavior of the exact solution . according to this requirement we construct a hybrid wave function with a long - range behavior given by the exact zero - energy solution outside the range @xmath119 of the interaction : @xmath120 where @xmath121 is the s - wave scattering length of the interaction . for two @xmath122he atoms interacting via the lennard - jones potential we get @xmath123 . the short - range part of the ansatz state is described by a gaussian . at some radius @xmath124 both functions are matched with continuous derivative . the final form of the uncorrelated wave function reads @xmath125 & ; r \le \chi \\[3 mm ] 1 - a / r & ; r > \chi \end{cases } .\ ] ] the matching radius @xmath124 is chosen such that the correlation function @xmath126 determined from vanishes at some finite radius @xmath113 . this allows a natural separation of long- and short - range correlations in the two - body system . for the @xmath108 scattering problem ( solid line ) in comparison with the hybrid ansatz @xmath118 for @xmath127 ( dashed line ) . in addition the radial dependence of the lennard - jones potential @xmath128 is shown ( thin dotted line ) . ] resulting from the integral equation ( solid line ) and the fitted parameterization ( thin solid line ) . for comparison the parameterization with parameters resulting from energy minimization ( thin dotted line ) is depicted . ] in figure [ fig : opt_e0solution ] the hybrid ansatz with matching radius @xmath129 is shown in comparison with the exact @xmath108 solution . both wave functions agree nicely at large distances . at small distances the exact solution shows the pronounced depletion within the core and an enhancement in the attractive region of the potential . these short - range correlations are not contained in the uncorrelated wave function ( dashed curve ) . instead the unitary correlation operator @xmath5 , which maps the uncorrelated wave function onto the exact eigenfunction ( solid curve ) , takes over the task . as discussed in section [ sec : ucom_operator ] this has the great advantage to allow the definition of correlated interactions that act between simple uncorrelated many - body states like product states . the solution of the integral equation for the correlation function @xmath28 using these two wave functions is shown by the dashed line in figure [ fig : opt_correlator ] . the quantity @xmath126 corresponds to the net distance by which two particles with an initial separation @xmath112 are shifted away from each other . the shape of the curve shows that particles with distances smaller than the core radius ( @xmath130 ) are shifted out of the core . if the initial distance is already larger than the core radius , then the shift decreases rapidly and finally vanishes at @xmath131 . the small shift for larger radii ( @xmath132 ) originates from the minimal deviations of exact wave function and hybrid ansatz in this range . however these are long range correlations , which should not be described by the unitary correlation operator . therefore we will set the function @xmath126 to zero for @xmath133 in order to define the short - range correlation function . this intrinsic separation between long- and short - range correlations works only if the potential has a sufficiently strong attractive region in addition to the repulsive core . for this class of interactions the exact two - body wave function is depleted inside the core and enhanced in the attractive region . since the unitary correlation operator conserves the norm of the wave function it can generate this kind of structure very easily : the amplitude shifted out of the core is placed in the attractive region . for purely repulsive potentials the situation is different . the exact wave function of a scattering state is depleted inside the core and the excess probability is spread over a large volume . therefore the unitary mapping leads to a correlator of large range . in order to stay within the two - body approximation one has to restrict the range of the correlations and compensate by improving the uncorrelated trial states . an alternative method to determine the correlation function @xmath28 emerges from the variational principle . if we think of the energy expectation value to be calculated with correlated states and the uncorrelated hamiltonian then the correlation function introduces additional degrees of freedom into the trial state . by choosing a suitable parameterization for @xmath28 with few variational parameters we can minimize the correlated energy expectation value and obtain an optimal correlation function . according to the structure of the correlation function determined by the mapping procedure described in the previous section we use the following parameterization @xmath134\ ] ] with three free parameters . the parameter @xmath135 determines the range of the correlation function , @xmath136 is related the maximum shift distance and thus the radius of the core , and @xmath137 influences the slope of @xmath28 at small radii , i.e. , the `` hardness '' of the core . we used this parameterization successfully for the description of the structure of atomic nuclei and nuclear matter @xcite . in order to describe a homogeneous @xmath122he liquid we choose a constant wave function as two - body trial state @xmath138 . in that case only the local potentials @xmath139 and @xmath61 contribute to the expectation value of the correlated hamiltonian @xmath140 . \end{split}\ ] ] the expectation values of all momentum - dependent contributions @xmath141 in vanish . by inserting the parameterization of the correlation function into and we can express the integrand as function of the three variational parameters @xmath142 and @xmath137 . the integration as well as the minimization can easily be done numerically . we obtain a unique set of parameters for the optimal correlation function @xmath143 the solid line in figure [ fig : opt_correlator ] shows the resulting correlation function . it is in very good agreement with the correlation function determined by the mapping procedure ( dashed line ) . this demonstrates that the parameterization used in the variation is well suited for the correlation function . for numerical simplicity we will use the parameterized correlation function with parameters according to for the following investigations . ( thin dotted line ) , correlated potential @xmath144 ( dashed line ) and sum of correlated potential and local part of the correlated kinetic energy @xmath145 ( solid line ) . ] ( solid line ) and radial effective mass correction @xmath146 ( dashed line ) . ] with the explicit form of the correlation function we can study the structure of the correlated hamiltonian in detail . as shown in section [ sec : ucom_operator ] the unitary transformation generates a momentum - dependent effective interaction . the local part of this interaction consists of the correlated interaction potential @xmath46 and a local contribution @xmath147 from the correlated kinetic energy operator . figure [ fig : opt_corrint_local ] compares the radial dependencies of the uncorrelated lennard - jones potential , the correlated lennard - jones potential and the sum of all local contributions . the unitary transformation of the potential shifts the repulsive core to smaller radii . this leads to a substantial reduction of the strength of the core , only a moderate repulsion at very short ranges remains . if we add the local part of the correlated kinetic energy the repulsive part vanishes completely ; the local component of the correlated hamiltonian is purely attractive . actually , it can be shown analytically that mapping of an uncorrelated wave function , which is constant within the range of the repulsive core , onto the exact @xmath108 solution results in a local part @xmath148 of the correlated hamiltonian that vanishes identically inside the core @xcite . the small contributions of the correlated local potentials within the core range that show up in figure [ fig : opt_corrint_local ] result from the use of the restricted parameterization for the correlation function and the slight variation of the trial state . the momentum - dependent parts of the correlated hamiltonian , which are formulated in terms of distance - dependent effective mass corrections , are shown in figure [ fig : opt_corrint_effmass ] . the radial effective mass correction ( dashed line ) generates a repulsion if the uncorrelated relative wave function has non - vanishing radial derivatives . the angular effective mass correction ( solid line ) generates an attraction if the uncorrelated wave function has non - vanishing relative angular momentum . in general the contribution of these components is small compared to the total energy expectation value as the uncorrelated states have small gradients . formally we can combine all the two - body terms of the correlated hamiltonian to a momentum - dependent effective interaction @xmath149 } = { \ensuremath{\tilde{v}}}({\ensuremath{{\ensuremath{\bm{r } } } } } ) + { \ensuremath{\tilde{u}}}({\ensuremath{{\ensuremath{\bm{r } } } } } ) + { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{t}}}}}}}}_{\omega } + { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{t}}}}}}}}_r\ ] ] with a purely attractive local part given by and and a momentum dependence given by . with this effective interaction the correlated hamiltonian in two - body approximation has the standard form @xmath150}_{ij } .\ ] ] the next orders of the cluster expansion contribute additional three - body , four - body , and higher order effective interactions . the tamed effective two - body interaction @xmath151}$ ] has a different operator structure and completely different radial dependencies than the uncorrelated potential @xmath41 . however , both interactions are phase - shift equivalent , i.e. , in a two - body scattering process both interactions generate identical phase - shifts for any collision energy , by construction . this generic property of the effective interaction results from the finite range of the unitary transformation , which does not influence the asymptotic behavior of the scattering solutions . it is completely independent of the shape of the correlation function at short ranges . in that way the unitary correlator provides a recipe to generate an infinite manifold of phase - shift equivalent interactions . this aspect is of special interest in cases where the interaction is determined only on the basis of two - body scattering data like the nucleon - nucleon interaction . as a first application of the ucom formalism we want to investigate the equation of state , i.e. , the energy per particle as function of the density , for a homogeneous @xmath122he liquid at temperature @xmath152k . in this case all effects that the interaction has on the many - body state are described by the unitary correlation operator . the uncorrelated many - body state @xmath153 is a direct product of identical constant one - body states @xmath154 @xmath155 for the calculation of expectation values we assume a finite volume @xmath156 containing @xmath4 particles . accordingly the constant one - body states read @xmath157 the limit @xmath158 at constant density @xmath159 is performed in the final step of the calculation . first we calculate the energy per particle with the correlated hamiltonian in two - body approximation @xmath160 } + { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{u}}}}}}}}^{[2 ] } + { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{t}}}}}}}}_{\omega}^{[2 ] } + { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{t}}}}}}}}_r^{[2 ] } .\ ] ] the energy per particle in two - body approximation is defined by the expectation value of this hamiltonian calculated with the uncorrelated many - body state @xmath161 since the uncorrelated state is a product of constant one - body states only the local components @xmath162 and @xmath163 contribute ; the expectation values of all momentum - dependent terms vanish . a simple calculation for the expectation value of the local components leads to the following expression for the energy per particle in two - body approximation @xmath164 } + c_u^{[2]})\;\rho\ ] ] with constant coefficients given by @xmath165 } = \frac{1}{2 } \int\!\!{\ensuremath{\mathrm{d}}}^3r\ ; { \ensuremath{\tilde{v}}}(r ) , \qquad c_u^{[2 ] } = \frac{1}{2 } \int\!\!{\ensuremath{\mathrm{d}}}^3r\ ; { \ensuremath{\tilde{u}}}(r ) .\ ] ] the integrals over the correlated potential @xmath46 given by and the local part of the correlated kinetic energy @xmath147 according to are evaluated numerically @xmath166 } = -856.5 \ , \text{k\aa}^3 , \qquad c_u^{[2 ] } = 401.5 \ , \text{k\aa}^3 .\ ] ] obviously , for homogeneous bose fluids at zero temperature the two - body approximation is not able to describe saturation , i.e. , a minimum of the energy at some finite density . the energy in two - body approximation is always proportional to density and due to the negative sum of the coefficients drops with increasing density . he liquid as function of density calculated in two - body approximation @xmath167 ( dashed line ) and in three - body approximation @xmath168 ( dotted line ) . the vertical line marks the density at which the smallness parameter @xmath75 reaches the value @xmath169 . the circles show the results of an exact gfmc calculation @xcite ; statistical errors are smaller than the size of the symbols . ] the dashed line in figure [ fig : liq_energy_2b3b ] shows this behavior in comparison with results of a green s function monte carlo ( gfmc ) calculation @xcite , which are exact up to the statistical errors of the monte carlo sampling . despite the fact that the two - body approximation fails to generate a minimum , the predicted orders of magnitude for the energy are correct . this is remarkable , considering that the uncorrelated lennard - jones potential would give an infinite positive energy expectation value for all densities . the special unitary transformation that we apply tames the strong core of the potential and leads to a bound many - body system . we note that the two - body approximation does generate saturation in other systems like homogeneous fermi liquids . in these cases the mass corrections contribute , because of the fermi motion , and generate a positive contribution to the energy that grows with a higher power in density than the attractive local terms . a straightforward step to go beyond the two - body approximation is the inclusion of the three - body terms of the cluster expansion . in addition to the terms in the correlated hamiltonian in three - body approximation contains three - body contributions from the correlated potential and the correlated kinetic energy @xmath170 } + { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{u}}}}}}}}^{[3 ] } + { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{t}}}}}}}}^{[3 ] } .\ ] ] again only the local three - body terms @xmath171}$ ] and @xmath172}$ ] contribute to the energy expectation value with the uncorrelated states . the resulting energy per particle in three - body approximation reads @xmath173 } + c_u^{[2]})\;\rho + ( c_v^{[3 ] } + c_u^{[3]})\;\rho^2 .\ ] ] the coefficients are defined by the six - dimensional integrals @xmath174 } = \frac{1}{6}\int\!\!{\ensuremath{\mathrm{d}}}^3x_1{\ensuremath{\mathrm{d}}}^3x_2\;{\ensuremath{\tilde{v}}}^{[3]}({\ensuremath{\vec{x}}}_1,{\ensuremath{\vec{x}}}_2,{\ensuremath{\vec{x}}}_3=0)\ ] ] and the corresponding expression for @xmath175}$ ] . the integration is done with a standard monte carlo algorithm ( `` vegas '' from @xcite ) and yields @xmath176 } & = ( -2512 \pm 50)\ , \text{k\aa}^6 , \\ c_u^{[3 ] } & = ( 29739 \pm 1100)\ , \text{k\aa}^6 . \end{split}\ ] ] the local three - body terms of the correlated hamiltonian generate a positive contribution to the energy per particle which is proportional to @xmath177 . the dominant term is the local three - body contribution of the correlated kinetic energy . together with the negative density - proportional contribution of the local two - body potentials the energy in three - body approximation provides a minimum at finite density . the density dependence of the energy per particle in three - body approximation @xmath168 is shown by the dotted curve in figure [ fig : liq_energy_2b3b ] . obviously the three - body contribution is not a small correction to the two - body approximation for densities larger than @xmath178 , which is only one fourth of the saturation density . this demonstrates that many - body correlations have a strong influence in these systems and that the two - body approximation alone is not sufficient to describe ground state properties . a similar conclusion can be drawn from the smallness parameter @xmath179 . it reaches the phenomenological limit @xmath76 for the validity of the two - body approximation at about @xmath180 of the expected saturation density . however , the naive inclusion of the three - body terms of the cluster expansion of the correlated hamiltonian does not improve the result beyond a density of about @xmath178 . although we obtain saturation in three - body approximation , both , energy and density at the minimum of @xmath168 are substantially smaller than the results of the gfmc calculation . this discrepancy has several reasons : firstly , the optimal correlation function was determined in the two - body system , i.e. , in two - body approximation . for a consistent treatment on the level of the three - body approximation the optimal correlation function should be determined by minimizing the energy in three - body approximation . secondly , we expect that genuine three - body correlations play a very strong role @xcite . they are described by an irreducible three - body part @xmath181 in the generator that was not included in the present treatment . finally , in view of the large three - body contributions there is no good reason to neglect terms beyond three - body order . each of the points mentioned can in principle be explicitly included in the calculation . however , none of these corrections are expected to yield reasonably converged results in the particular case of the @xmath122he liquid . the density range where the description should be applied is far above the range where the two - body approximation is applicable , i.e. , where the smallness parameter @xmath75 is below the typical value of @xmath169 . the successive inclusion of higher orders of the cluster expansion will extend this range but one would have to include very high orders which are beyond the numerical possibilities to reach the expected saturation density . the only reasonable way to extend the treatment is a partial summation over all cluster orders . this is done in the framework of jastrow correlations with the so called hypernetted chain ( hnc ) summation schemes @xcite . however , this type of partial summation is not feasible for the unitary correlation operator because of its more complicated operator structure . we aim at an effective description of the higher orders of the cluster expansion that still allows a compact analytic formulation of correlated observables and does not require extensive numerical efforts . one general way to simulate the effect of higher cluster orders is to introduce density - dependent correlation functions on the level of the two - body approximation . at low densities the correlator shifts a pair of particles in an optimal way from the repulsive into the attractive region . when the density grows other particles will be nearby and obstruct the correlation that was optimal for a free particle pair . to simulate this effect the range of the correlation function @xmath28 should be more and more reduced for increasing density . this will reduce the binding as the particles will not be able any more to fully exploit the attractive region of the potential . in practice this is accomplished by scaling the parameters @xmath136 and @xmath135 , which have the dimension of a length , in the parameterization of the correlation function @xmath28 with some factor @xmath182 that depends on density . at very low densities we assume that the higher - order contributions are negligible such that @xmath183 . with growing density the effect of the higher cluster orders grows such that the scaling factor @xmath184 has to drop . the most simple ansatz for the density dependence of @xmath184 is @xmath185 with one free parameter @xmath186 . the density - dependent correlation function reads @xmath187^{\!\eta } \exp\!\!\big[\!\!-\exp\!\big(\frac{r}{\beta\xi(\rho)}\big)\!\big].\ ] ] using this correlation function all components of the correlated hamiltonian like correlated potential , local part of the correlated kinetic energy , and effective mass corrections become explicitly density - dependent . the energy per particle is given by @xmath188,\ ] ] where the subscript @xmath71 indicates that the density - dependent correlation function is used . due to the explicit density - dependence of the correlated local potentials the equation of state is not proportional to @xmath71 any more . from a taylor expansion of around @xmath189 we obtain terms to all powers of the density . thus all orders of the cluster expansion are represented by the density - dependent correlator in an effective way . the phenomenological parameter @xmath186 in the scaling factor is in general adjusted such that agrees with one experimental data point , or for some density with the result of a realistic calculation . we will use the result of the gfmc calculation @xcite for the density @xmath190 and energy @xmath191 at the saturation point and require @xmath192 this ensures that the equation of state calculated with the density - dependent correlator runs through that point . however , this does not imply that the minimum of coincides with the gfmc minimum . using the values @xcite @xmath193 and @xmath194 we obtain @xmath195 we have chosen to fix the density dependence , i.e. , @xmath186 , with a calculation instead of an experimental value , because that calculation uses the same potential and it is known that this simple lennard - jones potential does not reproduce exactly the experimental two - body data . he liquid calculated in two - body approximation with a two - body optimized correlation function @xmath167 ( dashed line ) and with density - dependent correlation function @xmath196 ( solid line ) . the vertical line marks the density at which the smallness parameter @xmath75 reaches the value @xmath169 . the gray circles show the results of an exact gfmc calculation @xcite , the minimum is indicated by the black circle . ] figure [ fig : liq_energy_2bdd ] shows the energy per particle @xmath196 calculated with the density - dependent correlation function . the phenomenological description of the higher order contributions by a density - dependent correlation function indeed generates a minimum of the energy with @xmath197 the position of the minimum ( indicated by the cross ) as well as the shape of the equation of state around the minimum agree very nicely the the gfmc result . as noted above this is not an immediate consequence of the adjustment of @xmath186 . the agreement shows that the phenomenological density - dependence is well suited to describe the major aspects of many - body correlations on a numerically very simple level . actually , all calculations have been performed within a simple _ mathematica _ notebook . once the density - dependence is fixed the correlation function can be used to evaluate any correlated observable of interest . the analytic expressions for many correlated observables in two - body approximation are rather simple , but the evaluation of higher cluster orders becomes complicated if not intractable . thus the inclusion of the phenomenological treatment of higher order effects by means of the density - dependent correlator is a very useful tool . interesting quantities are , e.g. , the radial two - body distribution function @xmath198 , the static structure factor @xmath199 or the one - body momentum distribution @xmath200 . the corresponding observables for fermionic liquids were discussed elsewhere @xcite . in the following section we will use the density - dependent correlation function to investigate the structure of small droplets of @xmath122he . this will demonstrate the universality of the concept . in a next step we want to use the unitary correlation operator method to investigate the ground state properties of small @xmath122he droplets . the many - body problem is treated in a simple variational framework . the many - body trial state @xmath201 is like for the homogeneous liquid assumed to be a direct product of identical one - body states . the one - body states are described by a gaussian wave function @xmath202\ ] ] with a common width parameter @xmath121 as the only variational degree of freedom . the gaussian trial state is applicable for small droplets with @xmath203 ; larger droplets show a saturation of the central density , which can not be modeled with this ansatz @xcite . an advantage of the gaussian ansatz is that the two - body wave function @xmath204 can be separated analytically into a center of mass component @xmath205 and a relative wave function @xmath206 @xmath207 , \\ { \ensuremath { \big < { { \ensuremath{\vec{r } } } } \big| { \phi } \big > } } & = ( 2\pi a)^{-3/4 } \exp[-{\ensuremath{\vec{r}}}^{\,2}\!/(4 a ) ] . \end{split}\ ] ] this simple trial state is able to describe mean - field like effects of the interaction , i.e. , the spatial localization of the atoms . however , any correlation effect beyond the mean - field level has to be described by the unitary correlation operator . first we use the two - body approximation with the correlation function obtained from the two - body scattering solution to investigate the ground state properties of small droplets . the correlated intrinsic hamiltonian @xmath208 in two - body approximation has the structure @xmath209 } + { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{u}}}}}}}}^{[2 ] } + { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{t}}}}}}}}_{\omega}^{[2 ] } + { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{t}}}}}}}}_r^{[2 ] } , \ ] ] where capital letters indicate summation over all particles , like @xmath210 } = \sum_{i < j}^n { \ensuremath{\tilde{{\ensuremath{{\ensuremath{\bm{v}}}}}}}}^{[2]}_{ij}$ ] . the uncorrelated intrinsic kinetic energy @xmath211 , i.e. , the total kinetic energy @xmath212 reduced by the center of mass contribution @xmath213 , can be expressed by the relative two - body kinetic energy @xmath50 @xmath214 where @xmath215 ^ 2 .\ ] ] due to the simple structure of the many - body trial state the expectation value of the correlated hamiltonian in the @xmath4-body system can be expressed by that of the two - body system @xmath216 moreover we can use that the intrinsic hamiltonian acts on the relative part @xmath206 of the two - body wave function only . this leads to a simple form of the correlated energy expectation value of the @xmath4-body system @xmath217 we used that the expectation value of the angular effective mass correction @xmath218 vanishes due to the spherical symmetry of the relative wave function ( relative s - wave state ) . the two - body expectation value of the relative kinetic energy is given by @xmath219 for the correlated two - body potential and the local part of the correlated kinetic energy we obtain @xmath220\;\big|{\ensuremath { \big < { r } \big| { \phi } \big > } } \big|^2 .\ ] ] finally , the expectation value of the radial effective mass correction reads @xmath221 for a fixed correlation function @xmath28 the radial dependencies of the correlated potential @xmath46 , the local part of the correlated kinetic energy @xmath147 , and the radial effective mass correction @xmath146 are known analytically ( see section [ sec : ucom_operator ] ) . the minimization of the energy @xmath222 as function of the width parameter @xmath121 as well as the calculation of the integrals in equations and is done numerically . another interesting observable is the mean - square radius of the droplet , defined by the expectation value of the operator @xmath223 where @xmath224 is the center of mass coordinate of the many - body system . the formulation in terms of the two - body operator @xmath225 reveals a similar structure like for the intrinsic kinetic energy . the many - body product state under consideration allows a direct calculation of the uncorrelated expectation value using the relative two - body wave function @xmath226^{1/2 } \\ & = \bigg[\frac{n-1}{2n } { \ensuremath { \big < { \phi } \big| \,{{\ensuremath{{\ensuremath{\bm{r}}}}}^2}\ , \big| { \phi } \big > } } \bigg]^{1/2 } = \bigg[\frac{n-1}{2n}\ , 3 a\bigg]^{1/2 } . \end{split}\ ] ] the correlated ms - radius @xmath227 in two - body approximation has a similar structure like the kinetic energy @xmath228 } .\ ] ] in addition to the uncorrelated mean - square radius @xmath229 the unitary transformation generates a two - body contribution @xmath230 } = \frac{1}{2n } \sum_{i < j}^n \big[{\ensuremath{r_+}}^2({\ensuremath{{\ensuremath{\bm{r}}}}}_{ij } ) - { \ensuremath{{\ensuremath{\bm{r}}}}}_{ij}^2\big ] .\ ] ] the expectation value of the correlated ms - radius can again be expressed by the relative two - body wave function alone @xmath231^{1/2}.\ ] ] in order to study the quality of the two - body approximation we first look at very small droplets with @xmath232 . for these droplet sizes an early variational monte carlo ( vmc ) calculation on the basis of the lennard - jones potential exists @xcite . the authors use a jastrow - type parameterization of two - body correlations with a long - range form adjusted to the behavior of the exact two - body solution . the calculation of the many - body expectation value with these trial states involves a monte carlo integration routine . the resulting energy expectation values show a statistical error of approximately 10% . our results of the minimization of @xmath222 with respect to @xmath121 using the system independent correlation function given in equations and for @xmath232 are summarized in figure [ fig : drop_2b ] . the upper panel shows the correlated energy per particle , the middle panel the correlated rms - radius of the droplet and the lower panel the smallness parameter @xmath233 versus the number of particles @xmath4 . ( lower panel ) for small @xmath122he droplets in two - body approximation with two - body optimized correlation function . the gray circles show results of a variational monte carlo calculation @xcite , statistical errors for the energy are indicated by the gray bars . the lines should guide the eye . ] as for the homogeneous liquid the expectation values of the uncorrelated lennard - jones potential would diverge . but the unitary transformation of the hamiltonian using the correlation function determined in the two - body system removes the strong short - range repulsion of the potential completely . with this tamed correlated potential and a simple product ansatz for the many - body wave function the energies of very loosely bound droplets actually agree very nicely for @xmath234 with the vmc result . for @xmath235 the energy expectation value is below the vmc value and the difference grows with increasing particle number or density . since expectation values are evaluated in two - body approximation , i.e. , the full cluster expansion is truncated above two - body order , a basic property of the ritz variational principle is lost : the variational minimum of the energy is not necessarily bounded from below by the exact energy eigenvalue . the observed overbinding indicates that the higher orders of the cluster expansion have to give a sizeable positive contribution to raise the expectation value above the exact eigenvalue . in direct connection with the overbinding in two - body approximation the correlated rms - radii of energy minimized states are systematically smaller than the vmc result . accordingly the densities are too high . in contrast to the homogeneous liquid the two - body approximation does not produce a pathological collapse towards high densities for small droplets . this is due to the positive contributions of the kinetic energy and the effective mass correction that are absent in the homogeneous case . as a measure for the validity of the two - body approximation we used the smallness parameter @xmath236 with correlation volume @xmath72 defined by . in order to specify the smallness parameter for inhomogeneous systems we have to define a measure for the average density . one possible definition is the value of the density for a step - like density profile with same ( uncorrelated ) rms - radius as the original distribution . for identical gaussian single - particle wave functions with width parameter @xmath121 this equivalent density is given by @xmath237^{3/2 } = \frac{3 n}{4 \pi } \bigg[\frac{2}{5 } \ ; \frac{n}{n-1}\ ; \frac{1}{a } \bigg]^{3/2}.\ ] ] the smallness parameter @xmath238 defined with this density is shown in the lower panel of figure [ fig : drop_2b ] as function of particle number . like for the homogeneous liquid the energy in two - body approximation starts to deviate from the exact result if the smallness parameter @xmath233 exceeds the value @xmath169 ( indicated by the dashed line ) . this again confirms that the two - body approximation is valid as long as the smallness parameter fulfills the condition @xmath239 . to account for the effect of many - body correlations in a simple but efficient way we employ the concept of density - dependent correlation functions introduced in section [ sec : liq_eos_densdep ] for the homogeneous liquid . to apply the density - dependent correlator in an inhomogeneous system we have to specify in which way the density - dependence should be evaluated . the most simple approach is to insert the equivalent density of the droplet , as defined in , into the density - dependent correlation function . that means we neglect that the effects from stronger many - body correlations in the center of the droplet do not exactly cancel the weaker ones at the surface . a more elaborate ansatz to account for the inhomogeneity would be to implement the density - dependence in local density approximation , i.e. , the correlation function which acts on the relative coordinate of a particle pair depends on the local density at their center of mass position . this is not done here . [ cols="^,^,^,^,^,^,^,^,^,^,^,^,^ " , ] the variational calculation of the ground state properties of small @xmath122he droplets including a correlation function that depends on the equivalent density is straightforward . the parameterization of the correlation function @xmath240 is used with the parameters determined in the two - body system . the parameter of the density - dependent scaling function is taken from the investigations of the homogeneous liquid . the energy expectation value in two - body approximation with density - dependent correlations is of the same form as for the density - independent correlation functions discussed in the preceding section . ( lower panel ) for small @xmath122he droplets in two - body approximation with the density - dependent correlation function fixed in the homogeneous liquid . the gray circles show results of a variational monte carlo calculation @xcite . the lines should guide the eye . ] the results of the energy minimization with respect to the width parameter @xmath121 are summarized in figure [ fig : drop_2bdd ] . the energy expectation values are in full agreement with the vmc calculation @xcite for all particle numbers . the overbinding observed with a density - independent correlation function is completely compensated by the density - dependence . the correlated rms - radii of droplets with @xmath241 are consistently larger by 0.2 than the vmc result . unfortunately the authors @xcite give no estimate for the statistical or systematic errors for their procedure to obtain the rms - radii . for the smallest droplets our result for the rms - radii are below the vmc results . this may be caused by the restriction to gaussian trial states , which are not able to describe the long - range exponential tail present in these extremely weakly bound systems . at the same time the insufficient trial state causes a slight underbinding of the smallest droplets . for completeness figure [ fig : drop_2bdd ] also shows the smallness parameter @xmath238 . since both , the correlation volume @xmath72 and the density of the droplets are reduced by the density - dependent correlator the product stays small . in any case @xmath233 has lost its meaning as a measure for three - body correlations as the density - dependence of the correlator already includes many - body correlations . table [ tab : drop ] summarizes the results of the energy minimization with two - body optimized and density - dependent correlator for droplet sizes up to @xmath242 . in addition to the correlated energy and the rms - radius the individual terms of the correlated hamiltonian are shown . the attractive correlated potential and the repulsive local part of the correlated kinetic energy are the major contributions and show a large cancellation . the intrinsic kinetic energy and the effective mass correction give rather small contributions . we presented a simple and elegant approach to describe strong short - range correlations in a many - body system that result from the strong repulsive core of the two - body interaction . these correlations induce high - momentum components into the relative wave function that can not be treated within a low - momentum model space . the correlation operator @xmath5 generates the short - range correlations by a unitary transformation in the relative coordinate of each pair of particles . it shifts the particles out of the repulsive core of the two - body interaction . the correlation operator is then used to define correlated observables , most notably the correlated hamiltonian . the unitary transformation yields a tamed effective two - body interaction , which consists of a purely attractive local part and a weak momentum dependence . in addition three - body and higher order terms are generated . for dilute systems , i.e. , if the range of the correlations is small compared to the average distance between the particles , it is sufficient to include the terms up to two - body order . in high - density systems the higher order terms can be included explicitly or simulated by a density - dependent correlation function . in order to study the applicability of the method in many - body systems , where short - range correlations play a dominant role , we investigated the ground state properties of homogeneous @xmath122he liquids and small droplets . the correlation function is fixed by energy minimization in the two - body problem with a constant uncorrelated wave function . alternatively the correlation function can be determined by mapping an uncorrelated wave function onto the exact energy @xmath108 two - body wave function . we use this correlation function to calculate the energy of the homogeneous @xmath122he liquid as function of density . the uncorrelated many - body state is a direct product of constant one - body states , thus the influence of the interaction on the state has to be described by the correlation operator alone . already in two - body approximation the strong short - range repulsion of the lennard - jones potential is tamed completely and we obtain a bound system . however , the two - body approximation does not generate saturation , i.e. , the energy per particle decreases proportional to density . to simulate the effect of many - body correlations we employ a density - dependent correlation function . the one parameter introduced by the density dependent scaling of the correlation function is fixed by adjusting the correlated energy for one value of the density to the result of a gfmc calculation . the resulting equation of state shows a minimum and is in very good agreement with the predictions of the gfmc calculation over the whole density range . in a second step we use the density - independent as well as the density - dependent correlation function to perform an _ ab initio _ calculation of the ground state structure of small @xmath122he droplets in a simple variational framework . it turns out that the two - body approximation with the density - independent correlation function gives a good description for small droplets with particle numbers @xmath234 . for larger droplets an overbinding compared to a quasi exact variational monte carlo calculation appears . due to the higher density in these droplets many - body correlations become relevant . if we use the density - dependent correlation function with one parameter adjusted in the homogeneous liquid then the energies and rms - radii are in very good agreement with the exact calculation for all particle numbers under consideration . we conclude that the unitary correlation operator method is a powerful and universal tool to handle short - range correlations in the many - body system . for sufficiently low densities one can use the two - body approximation to construct the correlated hamiltonian as well as other correlated observables in closed form . the resulting effective two - body interaction can be used for an ab initio description of the many - body system based on a low - momentum model space . a gross measure for the validity of the two - body approximation is the smallness parameter @xmath75 , which should be below a phenomenological limit of @xmath169 . for higher densities one can utilize density - dependent correlation functions to describe many - body correlations effectively within the two - body approximation . the additional parameter in the density - dependence can be fixed by one measured data point or by comparison with an exact calculation . nevertheless , the resulting density - dependent correlator can then be used for a broad range of applications , e.g. , _ ab initio _ calculations of correlated observables like energies , densities or momentum distributions of different systems without further adjustments . the great advantage of the unitary correlation operator method with density - dependent correlation functions is that besides its transparent physics it requires only minimal computational power compared to the huge computational resources needed for monte - carlo - type calculations . a set of _ mathematica _ notebooks that illustrate the application of the unitary correlation operator method to @xmath0he liquids can be found at the url : ` http://theory.gsi.de/~rroth/math/ ` .

we propose a new iterative unfolding method for experimental data , making use of a regularization function . the use of this function allows one to build an improved normalization procedure for monte carlo spectra , unbiased by the presence of possible new structures in data . we are able to unfold , in a dynamically stable way , data spectra which can be strongly affected by fluctuations in the background subtraction and simultaneously reconstruct structures which were not initially simulated . this method also allows one to control the amount of correlations introduced between the bins of the unfolded spectrum , when the transfers of events correcting the systematic detector effects are performed . unfolding , iterations lal 09 - 107 +

0907.3791

experimental distributions in high - energy physics are altered by detector effects . this can be due to limited acceptance , finite resolution , or other systematic effects producing a transfer of events between different regions of the spectra . provided that they are well controlled experimentally , all these effects can be included in the monte carlo simulations ( mc ) of the detector response . one can then use these simulations to correct the data for the same effects . several deconvolution methods for data affected by detector effects were described in the past ( see for example @xcite ) . the aim of this paper is to illustrate a new unfolding method allowing one to obtain a data distribution as close as possible to the `` real '' one for rather difficult , yet realistic , examples . this method is based on the idea that if the mc simulation provides a relatively good description of the data and of the detector response , one can use the transfer matrix to compute a matrix of unfolding probabilities . if the first condition is not fulfilled one can iteratively improve the transfer matrix . our method is using a first step , providing a good result if the difference between data and normalized reconstructed mc is relatively small , on the entire spectrum . if this is not the case , one should proceed with a series of iterations . the author learned the existence of the iterative methods present in the above list , when this study was close to completion . two of them ( @xcite ) use a direct comparison of data and reconstructed mc events to compute normalization coefficients for the true mc events . however , it has been shown in @xcite that this approach has a rather slow convergence . further on , in @xcite one can find two methods based on the comparison of the true mc with the unfolding result at a given iteration . this comparison allows the computation of the probability matrix to be used at the next step . for the parts of our procedure which are similar to the ones in these last two papers , we will emphasize which are the differences that make our method more general and better behaved . in section [ sec : importantconsiderations ] we present a series of considerations taken into account in the unfolding method developped in the present paper . we describe the ingredients of the method in sections [ sec : regf ] - [ sec : foldandunf ] . in section [ sec : iterunfstrategy ] , we assemble all these elements into an unfolding strategy . we provide two examples of applications of the method , with some possible tests , in sections [ sec : complexex ] and [ sec : simpleex ] . we introduce the notations systematically and we also recall them in appendix [ app : notation ] . in the unfolding procedure we will not concentrate on the correction of acceptance effects . it is straightforward to perform it on the distribution corrected for the effects resulting in a transfer of events between different bins of the spectrum . actually , these transfers have a physical meaning only for events which are in the acceptance of the detector and which passed all the cuts . the method described in @xcite computes at each iteration the spectrum corrected for the acceptance . this result is then used to improve the folding matrix ( i.e. the probabilities of the causes , as they are called in this reference ) . furthermore , the folding probabilities for each bin are normalized to the corresponding acceptance . it is straightforward to prove analytically that this method is completely equivalent to the one used in @xcite , where the acceptance is ignored in the unfolding and the corresponding correction is done at the end . a first , rather tricky point is the way how the unfolding deals with new structures , absent in the mc , but which can be present in the data . these structures are affected by the detector effects , and hence they need to be corrected . it seems that the singular values decomposition ( svd ) @xcite and the iterative @xcite methods provide a natural way of performing this correction . if the new structures in the data contain a relatively important number of events , they could also affect the normalization of mc spectra with respect to the data . the unfolding procedure described here makes use of a comparison method between data and mc spectra , which ignores significant shape differences when computing the relative normalization factor . this allows a coherent comparison of data and mc spectra and improves the convergence of the algorithm when the data contain structures which are not present in the mc simulation . it is also important to keep in mind that experimental spectra are generally obtained after background subtraction . this operation results in an increase of errors for the corresponding data points . due to bin - to - bin or correlated fluctuations of the subtracted background , these points can fluctuate within their errors , or they can hide systematic effects as large as the error components due to background subtraction . when computing the central values of the final distribution , the unfolding procedure has to take into account the size of the experimental errors , including the ones from background subtraction . not doing so could result in too large a transfer of events from / to different regions of the spectrum . this would affect the regions of important background subtraction , but could also bias the central values of more precise points in another region . such a systematic effect of the procedure is to be avoided . to our knowledge , none of the previous methods aim at dealing with this second type of problem , and at distinguishing it from the first one . we are going to describe a method which is be able to deal with the potential problems described above . the fluctuations which can be generated when using this type of method are dynamically reduced , in the sense that the way in which bin - by - bin corrections are performed avoids this problem . therefore , no additional smoothing by a model - dependent fitting of the unfolding result at each iteration step ( as proposed in @xcite ) , is required . the transfer of events performed by the unfolding introduces bin - to - bin correlations for the final spectrum . for further use of the data , these correlations are to be reduced as much as possible , if this can be done without introducing significant systematic biases of the result . actually , the correction of systematic effects can generally be obtained without performing on data the maximal transfer of events ( predicted by the transfer matrix ) , but only part of it . the method described in this paper allows one to control the amount of these correlations and an optimization can be done with respect to the potential systematic bias of the final result . this method is to be applied on binned , one dimensional data and the label of the physical quantities will be provided by the bin numbers . it can be directly generalized to multidimensional problems . for the sake of simplicity , we will consider only examples where the number of bins of the data and reconstructed mc is the same as the one for the unfolded distribution and true mc . however , it is rather easy to generalize the method to include a binning change in the unfolding operation . all the formulae in the following will be given for this general case . we suppose the statistics of the data and of the simulation high enough to allow a `` reasonable '' separation into a series of bins . if this is not the case , the size of the bins must be adapted to the available statistics . in order to dynamically reduce fluctuations and not to unfold events which could be due to fluctuations , in particular from the subtracted background , one can use a regularization function @xmath0 . this function provides an information on the significance of the absolute deviation @xmath1 between data and simulation in a given bin , with respect to the corresponding error @xmath2 . it must be a smooth monotonous function going from 0 , when @xmath3 , to 1 , when @xmath4 . @xmath5 is a scaling factor , used as a regularization parameter . as we will see in the following , changing the regularization function that is used in our method will change the way we discriminate between real deviations and statistical fluctuations . it can be seen as the complementary of a generalized ( by the use of @xmath5 ) `` p - value '' of a given deviation between two variables affected by errors , if one considers the hypothesis that the mc describes well the data in the given bin . actually , for @xmath6 this function is the complementary of the p - value defined in @xcite , computed for a given pdf . if the correlations between the errors of the values entering the computation of @xmath1 are not negligible , only the uncorrelated part must be kept when computing @xmath2 . for the unfolding procedure , we consider several types of functions of the relevant variable @xmath7 : @xmath8 their dependence on the relevant ratio is shown in fig . [ fig : functions ] . with respect to @xmath7.,width=529 ] in general , we will use different @xmath5 parameters for the regularization function for each step of the unfolding procedure . we will see however that some of these parameters can be unified or even dropped . in the unfolding procedure we will need to perform a comparison of data and reconstructed mc , as well as true mc and intermediate unfolded spectra . we must not forget that the data may contain structures , which were not ( well ) simulated in the mc . if the number of events corresponding to them is not negligible with respect to the total number of events in the data , the normalization of the mc spectra should be determined with care . one must actually ignore the `` unknown '' ( to the simulation ) events in these structures when computing the normalization factor . not doing so ( a standard normalization procedure would use the total number of events in the two spectra ) could result in generating fake differences between the two spectra , in regions where the data are well described by the simulation . for a good normalization , one needs to get an estimation of the number of events in the data / intermediate unfolding spectrum which correspond to features that were simulated in the reconstructed / true mc ( @xmath9 ) . dividing @xmath9 by the number of events in the true and reconstructed mc spectra ( @xmath10 ) one gets the normalization factor to be used for the mc . this is done as follows : + 1 ) a first estimation of this number is provided by the total number of events in the data and in the intermediate unfolding distribution ( @xmath11 ) , minus those which are potentially due to background subtraction : @xmath12 . here @xmath13 is the number of bins in the data and in the reconstructed mc distribution , @xmath14 is the number of bins in the unfolding result and in the true mc distribution , @xmath15 stands for the number of reconstructed data events in the bin k ( after background subtraction ) and @xmath16 for the number of events in the bin k of the data distribution which are potentially due to a fluctuation in the subtracted background , @xmath17 stands for the number of unfolded events in the bin j and @xmath18 for the number of events in the bin j of the unfolded distribution which are potentially due to a fluctuation in the subtracted background . + one can then improve this estimation by an iterative procedure : + 2 ) a better estimation is provided by @xmath19 with @xmath20 where @xmath21 is the number of reconstructed mc events and @xmath22 the number of true mc events , and @xmath23 if the errors of @xmath15 or @xmath24 ( @xmath25 ) are not correlated with the errors of @xmath21 or respectively @xmath22 ( @xmath26 ) . this hypothesis is true for the first - step comparison of data and reconstructed mc . the situation is different at further iteration steps and for the comparison of true mc and intermediate unfolding spectra . there , only their uncorrelated parts must be added quadratically to get @xmath27 . in practice , however , as the mc errors are generally negligible , considering only the data / unfolding errors at this level may be sufficient . + using eq . ( [ eq : neprdsmc ] ) , one adds to the previous estimate other events which have a low probability to belong to initially unknown structures . on the contrary , due to the factor in front of @xmath28 , there is only a relatively small contribution from events likely to belong to initially unknown structures . + in the special cases when @xmath29 or @xmath30 , the content of the corresponding bins will be neglected when computing the number of events for the data and mc . this kind of situation will also receive a special treatment in the unfolding procedure , the estimation of the background fluctuations and the improvement of the transfer matrix . + 3 ) replace @xmath9 by @xmath31 and iterate until the change in @xmath9 between two consecutive iterations gets very small . + in the examples presented in this paper we stopped the normalization iterations when the relative improvement of two consecutive steps was less than @xmath32 . it should be emphasized that a better estimation of @xmath9 improves the separation between `` known '' and `` unknown '' structures , which then improves the next estimation . at the same time , the values of the parameter @xmath33 ( for normalization ) , providing a good convergence of the normalization procedure , need to be studied , as described in section [ sec : optparam ] . the convergence properties of the normalization procedure also depends on the amount of background fluctuations which are not yet identified when the normalization is performed . in the normalization procedure , these fluctuations , corresponding to data points with larger error bars , and would more likely not be identified as new structures . as the estimation of the remaining background fluctuations gets improved , the normalization follows . at this level we made a distinction between @xmath16 and @xmath34 only to keep the method as general as possible . indeed , if the binning of the initial ( data ) and final ( unfolded ) distributions are not the same , one must build a prescription to convert one vector into another , by a rebinning ( in any case , no unfolding will be done for these events ) . using a rebinning transformation @xmath35 to pass from data to unfolding bins and the inverse rebinning transformation @xmath36 to pass from unfolding to data bins , one can convert one vector into the other , through : @xmath37 these two transformations , if valid for all the sets of possible spectra in the two binnings ( as defined before ) , must be non - linear . actually , if they were linear , as their two matrix would contain only positive elements , even if @xmath38 , their product could not be equal to the identity matrix . if the binning of the data and unfolding result are the same , the two vectors will always be identical ( i.e. @xmath39 ) . in the mc simulation of the detector one can directly determine the number of events which were generated in the bin @xmath40 and reconstructed in the bin @xmath41 ( @xmath42 ) . provided that the transfer matrix @xmath43 gives a good description of the detector effects , it is straightforward to compute the corresponding folding and unfolding matrix : @xmath44 the folding probability matrix , as estimated from the mc simulation , @xmath45 gives the probability for an event generated in the bin @xmath40 to be reconstructed in the bin @xmath41 . the unfolding probability matrix @xmath46 corresponds to the probability for the `` source '' of an event reconstructed in the bin @xmath41 to be situated in the bin @xmath40 . these matrix elements are well defined if the true mc bin j , and respectively the reconstructed mc bin i , are not empty . if that is not the case , one can set these values to zero and a special treatment must be provided for these bins in the unfolding algorithm . the folding matrix describes the detector effects , and one can only rely on the simulation in order to compute it . the quality of this simulation must be the subject of dedicated studies within the analysis . as a result of these studies , the transfer matrix can be improved , a systematic error is to be associated to it and must be propagated to the unfolding result . the unfolding matrix depends not only on the description of detector effects but also on the quality of the model which was used for the true mc distribution . it is actually this model which can ( and will ) be iteratively improved , using the comparison of the true mc and unfolded distributions . it is easy to see that the reconstructed and true mc distributions are related through : @xmath47 provided that the detector effects are well simulated , the folding equation ( [ eq : rmc ] ) will also stand for data , no matter what the true spectrum is in the real world . the unfolding equation ( [ eq : tmc ] ) can be written for data only in an approximate way , due to the differences between the model used for @xmath48 and the real physical signal which introduce an uncertainty on the initial unfolding matrix . however , @xmath48 and hence the unfolding matrix , can be iteratively improved and therefore it is an improved version of ( [ eq : tmc ] ) that will be used to compute the unfolding result from data . other effects will also be taken into account in the unfolding equation , as we shall see in the next subsection . equation ( [ eq : tmc ] ) can be seen as a `` basis for the unfolding '' , in the sense that it describes this operation performed on the mc . if the differences between data and reconstructed mc are relatively small , then those between the unfolding result and true mc will be small too . in any case , one can start with an unfolded distribution equal to the true mc , and add to it the unfolded data events which are not present in the reconstructed mc . for the initial true mc one can use either a previously measured spectrum ( if available ) or one inspired by the data and the transfer matrix . in order to perform the unfolding , one must first use the iterative procedure described in section [ sec : mcnorm ] to determine the mc normalization coefficient ( @xmath49 ) . at this level , it is the comparison of the reconstructed mc with the data that is used in the normalization procedure . one can then proceed to the unfolding , where , in the case of identical initial and final binnings , the result for @xmath50 $ ] is given by : @xmath51 where the same notation as in ( [ eq : neprdsmc ] ) is used . in the case of different binnings for the data and the unfolding , the kroneker symbol @xmath52 must be replaced by a rebinning transformation @xmath35 . the first two contributions to the unfolded spectrum are given by the normalized true mc and the events potentially due to a fluctuating background subtraction , which are not transferred from one bin to another . then one adds the events which are present in the data minus the potential effect from the background , and which are absent in the normalized reconstructed mc . a fraction @xmath53 of these events are unfolded using the estimate of the unfolding probability matrix @xmath54 , and for the rest , only a rebinning is done ( if necessary ) . with the description of the regularization functions given in section [ sec : regf ] , it is clear that reducing @xmath5 would result in increasing the fraction of unfolded events , and reducing the fraction for which only a rebinning is done . this unfolding step benefits from two regularizations : one through the use of the previously described functions , and a second ( implicit ) one , through the mixing of events in different bins , by transfer of events . changing @xmath5 can reduce one regularization and enhance the other . choosing an appropriate value for this coefficient provides one with a dynamical attenuation of spurious fluctuations , without reducing the efficiency of the unfolding itself . this is possible due to the fact that the fraction of unfolded events is larger when the deviations between the spectra are larger with respect to the corresponding errors . some precautions need to be taken if @xmath55 or @xmath56 . the first situation can typically occur if the background subtraction that was made for data , in the given bin , is too large , or if the binning is not well adapted to the data statistics . the second one indicates that the given data bin size is too small even for the mc statistics and the corresponding unfolding elements can not be computed . in these two situations , the events which are present in the data minus the potential remaining background and minus the normalized reconstructed mc will be simply rebinned ( or kept in their original bin if the data and unfolding binnings are identical ) . after having performed one unfolding procedure yielding to a result @xmath57 , as described before , one can directly compare it with the true mc . doing this , one can compute an estimate of the fluctuations from subtracted background events . as these fluctuations must be identical in the reconstructed data and in the unfolding result , it is important that the first unfolding does not affect them ( hence a constraint on the corresponding parameter @xmath58 ( for subtraction ) ) . just as for the unfolding , one must first use the iterative procedure described in section [ sec : mcnorm ] to determine the mc normalization coefficient ( @xmath49 ) , looking this time for structures present in the unfolding result , but not simulated in the true mc . then one can ( re-)estimate the background to be subtracted @xmath59 where @xmath60 . here , increasing @xmath58 would increase the fraction of not - simulated events , which will be assigned to the estimate of the remaining background from fluctuations in the subtraction . in principle one could perform this estimation using @xmath61 , with the right side of ( [ eq : imprbu ] ) increased by the previous estimate ( @xmath18 ) , but here we rather use @xmath62 so that we avoid an artificial increase of @xmath63 in regions where the spectra differ only due to a normalization effect . this allows one to use sufficiently large values of @xmath58 , to efficiently subtract real remaining background fluctuations . the use of large values of @xmath58 will also prevent the propagation and potential amplification of the fluctuations in the unfolding result . once this estimate is obtained , one can use it to get a better normalization and then reestimate the background fluctuations . this new information is however exploited very fast and one or two iterations are largely enough at this level . one can perform the ( re-)estimation ( [ eq : imprbu ] ) only in the usual case , when @xmath64 and @xmath65 . otherwise , the estimation of the remaining background is trivially modified @xmath66 implying a fixation of the corresponding `` pathological '' events at the next unfolding step . this solution seems natural if the unfolding result minus the remaining background has a negative number of events in the given bin . such a situation can occur due to an overestimation of the background that was subtracted to get the data spectrum to be unfolded . if the origin of the problem is the low mc statistics in the neighbourhood of the given bin , a more suitable solution could be the choice of an adapted binning . we feel that this problem should be dealt with in unfolding methods . as explained in the introduction , if the initial true mc distribution does not contain or badly describes some structures which are present in the data , one can iteratively improve it , and hence the transfer matrix . this can be done by using a better ( weighted ) true mc distribution , with the same folding matrix describing the physics of the detector , which will yield to an improved unfolding matrix . just as in the previous section , one must determine the mc normalization coefficient , looking for structures present in the unfolding result , but not simulated in the true mc . as the description of the new structures by the mc will get improved , the task of the normalization procedure will be easier . then the improvement is performed for one bin @xmath40 at the time . if however @xmath67 or @xmath68 , one does not modify the transfer matrix in the corresponding true bin . generally , this is not the case , and one can get an improved determination of all the elements of the column @xmath40 of the transfer matrix @xmath69 here , @xmath70 ( for modification ) stands for the regularization parameter used when modifying the matrix . increasing @xmath70 would reduce the fraction of events in @xmath71 used to improve the transfer matrix . this method allows an efficient improvement of the folding matrix , without introducing spurious fluctuations . this is due to the fact that the larger the difference between the spectra , with respect to the corresponding errors , the larger the fraction of events used for the improvement of the matrix will be . the amplification of small fluctuations can be prevented at this step of the procedure too . in this section we describe a quite general unfolding strategy , based on the elements presented before . it works for situations presenting all the difficulties listed before , even in a simplified form , where some parameters are dropped and the corresponding steps get trivial . the strategy can be simplified even more , for less complex problems . one will start with a null estimate of the background . a first unfolding , as described in section [ sec : unfold ] , is performed , with a relatively large value of @xmath72 . this step will not produce any important transfer of events from the regions with potential remaining background ( provided that @xmath73 is large enough ) . actually , it is the condition that this transfer must be very small that will impose a minimal value for @xmath73 . the first unfolding will perform the correction equal to the difference between the normalized true and reconstructed mc . this is generally the main correction that the unfolding has to provide . concerning the new structures in the data , depending on their relative size , on the size of the transfer corrections and on @xmath73 , the corresponding corrections at this level can be more or less important . one can even drop @xmath73 , by taking its value very large , which amounts to performing only the main correction discussed in the previous paragraph . this obviously guarantees the stability of the remaining background fluctuations for this step . at this level one can start the iterations : + 1 ) estimation of the remaining background fluctuations + an estimate of the remaining background fluctuations can be obtained using the procedure described in section [ sec : estbkg ] . the parameter @xmath58 used here must be large enough , not to underestimate the remaining background . as explained before , this can also prevent unfolding fluctuations from propagating . @xmath58 can however not be arbitrary large , as this operation must not bias initially unknown structures , by not allowing their unfolding . the ability of the method to simultaneously satisfy to these conditions clearly depends on the properties of the function @xmath74 , as it defines the degree of significance of deviations between spectra . + 2 ) improvement of the unfolding probability matrix + using the method described in section [ sec : impmatrix ] , one can improve the folding matrix @xmath43 and get a better estimate of the remaining background . a parameter @xmath70 , small enough for an efficient improvement of the matrix , yet large enough not to propagate spurious fluctuations ( if not eliminated at another step ) , must be used at this step . + 3 ) an improved unfolding + a parameter @xmath75 will then be used to perform an unfolding following section [ sec : unfold ] , exploiting the improvements done at the previous step . it must be small enough to provide an efficient unfolding , but yet large enough to avoid spurious fluctuations ( if not eliminated elsewhere ) . + these three steps will be repeated until one gets a good agreement between data and reconstructed mc plus the estimate of the background . another way of proceeding could consist in stopping the iterations when the improvement brought by the last one on the intermediate result is relatively small . if the parameters of the unfolding are well chosen , the two conditions yield similar results . provided that one has a good model of the true data , one can even use a toy simulation to estimate the needed number of steps . the values of the @xmath5 parameters are to be obtained from toy simulations , as will be shown in the following . as one does not have a prior lower limit for @xmath70 and @xmath76 , other than the one related to the regularization , one can discard them by taking their values to zero ( i.e. using all the data - reconstructed mc difference in the unfolding , and all the difference between the true mc and the intermediate unfolding , to improve the transfer matrix ) . this can be done if the higher limit on @xmath58 provides enough regularization at the level of the subtraction procedure . in principle , one could also perform an estimation of the remaining background fluctuations , when comparing the data with the reconstructed mc . this could be done as a straightforward adaptation of the method explained in section [ sec : estbkg ] or computing this estimate as an improvement of the previous one , with @xmath77 replaced by @xmath78 . this second method could be reliable if the normalization effect is small enough . however , when doing the estimation of the background fluctuations at the data - reconstructed mc level , one would not benefit from the iterative improvement of the true mc . this can be useful in the method of section [ sec : estbkg ] , as it improves the separation between real new structures and fluctuations , especially when the resolution effects are important . in the following we describe a rather complicated , yet realistic test , proving the robustness of the method . it exhibits all the features discussed previously , which are simultaneously taken into account by the unfolding . for the clarity of the presentation , the structures and dips of the spectrum are however separated . the true and reconstructed spectra for this test have 200 identical bins . the first step consists in building a model for the transfer matrix @xmath79 . the transfer matrix @xmath80 with @xmath81 , @xmath82 , @xmath83 and @xmath84 ; exhibits resolution effects and corresponds to a true spectrum which is mainly gaussian . in addition , this spectrum includes a gaussian resonance ( more narrow than resolution ) centred on the bin 130 and a dip around the bin 40 ( larger than resolution ) . one then includes a systematic transfer of events from high to lower bins , getting the final model matrix ( see fig . [ fig : ax4 ] ) : @xmath85 with @xmath86 and @xmath87 . in practice , one can find an example of this type of transfer of events , for measurements of the cross section @xmath88 with the radiative return method . indeed , not detecting a final state photon will result in an underestimation of the mass of the given event . of transfer matrix . here , for display reasons , the bins are four times larger than the real ones ( an average was computed inside squares of @xmath89 initial bins).,width=377 ] it is a fluctuation of the model that will be used as mc transfer matrix @xmath43 for the unfolding . the model itself provides the `` real '' detector matrix which will be crucial for the building and folding of the true distribution . one must then fix a normalisation ratio between data and mc simulation ( of the order of 1/10 for usual experiments ) . the true data distribution will be given by the normalised true distribution corresponding to @xmath79 , completed with structures unknown by the mc ( a `` bias '' ) : @xmath90 , \label{eq : bias1}\ ] ] with @xmath91 and @xmath92 . the resonance centred on the bin 90 is more narrow than resolution and will therefore be hard to correct , whereas the one centred in 170 is larger . the folding probability matrix is to be computed from @xmath79 . at this point , one can compute the known data ( @xmath93 ) , by folding the true data without the new structures . the folding of the true data yields to the reconstructed spectrum ( after fluctuation within statistical errors ) . the final data spectrum ( filled circles in fig . [ fig : unf1 ] ) is obtained after adding a remaining background component , where it can typically appear in practice , i.e. in the dip of the spectrum , with the average per bin : @xmath94 with @xmath95 . for the background , we attribute an error bar equal to its average value in each bin . this background can come with or without statistical fluctuations corresponding to its error bars . in order to simulate this feature , we fluctuate the background within its errors for @xmath96 only , and we add it to the data fluctuated within their statistical errors . the final errors of the data , which will enter the unfolding algorithm , are given by the quadratic sum of the errors of the background plus the statistical ones from the data without background . the unfolding result is to be compared with the true data plus the remaining background . this represents a test over a very large dynamic scale . the results that will be presented for this test were obtained using the function ( [ f:1mexp ] ) , for @xmath97 , but similar results were obtained with the other functions . the parameters to be used in the unfolding were studied with an independent toy simulation , with various parameters for the fluctuations in background subtraction , and the bias . in practice , one would need approximate values of the parameters for the spectrum and the potentially dangerous background fluctuations , in order to perform the further study , to determine the parameters of the unfolding . one can get a very preliminary estimate for the parametrization of the spectrum from the difference between data and reconstructed mc , together with the transfer matrix , and from the procedure of background subtraction . once the unfolding performed , the determination of the spectrum parameters is also drastically improved . in principle , at that point , one could try to re - estimate the optimal parameters of the unfolding and try to improve them with more accurate toy simulations . in practice however , excepting very extreme situations , this new estimate will be consistent with the first one . , without background or with the usual background value , obtained after , at most , from left to right , 30 , 50 and respectively 200 steps.,title="fig:",width=188 ] , without background or with the usual background value , obtained after , at most , from left to right , 30 , 50 and respectively 200 steps.,title="fig:",width=188 ] , without background or with the usual background value , obtained after , at most , from left to right , 30 , 50 and respectively 200 steps.,title="fig:",width=188 ] , without background or with a background twice less ( left ) , twice larger ( middle ) and four times larger ( right ) than usual . in the right figure , only the upper limit of the normalization improvement with background is visible . these improvement values were obtained after , at most , 50 steps.,title="fig:",width=188 ] , without background or with a background twice less ( left ) , twice larger ( middle ) and four times larger ( right ) than usual . in the right figure , only the upper limit of the normalization improvement with background is visible . these improvement values were obtained after , at most , 50 steps.,title="fig:",width=188 ] , without background or with a background twice less ( left ) , twice larger ( middle ) and four times larger ( right ) than usual . in the right figure , only the upper limit of the normalization improvement with background is visible . these improvement values were obtained after , at most , 50 steps.,title="fig:",width=188 ] the optimization of the parameter @xmath33 used in the normalization procedure , was studied for the comparison between data and reconstructed mc , with a null estimate of the remaining background fluctuations . as explained before , at this level , one would not take advantage on the corrections of the resolution effects on new resonances . actually , these corrections make easier the separation of new structures and remaining background fluctuations . a procedure that works properly at the data / reconstructed mc level , would work even better when comparing an intermediate unfolding result and the true mc . furthermore , the normalizations at these two levels are improved by a better knowledge of the remaining background fluctuations and by the improvement of the transfer matrix through the true mc . the first part of this study was performed with toys of data and reconstructed mc spectra , obtained as described in section [ subsec : building1 ] , with or without the remaining background given by eq . ( [ eq : bkgd ] ) . in both cases , the first estimate of @xmath9 is given by the number events in the data @xmath11 . for 100 toys , we plot the minimal and maximal improvement brought by the procedure ( @xmath98 ) , for a large interval of @xmath33 values ( see fig . [ fig : optlnub ] ) . we stop iterating when the relative improvement of two consecutive steps gets less than @xmath32 , or after a fixed number of steps ( if the first condition was not fulfilled before ) . fast convergence and good stability ( with respect to the background problem ) are exhibited for large values of the parameter of normalization . the procedure is clearly sensitive to the presence of background fluctuations , which are treated more like real structures , for small values of @xmath33 . it is only in this region of values of @xmath33 that the limit on the number of steps plays a role . too small a value of @xmath33 could be identified , even when running on fluctuating data , by the presence of instabilities in the sign and size of the correction ( their size could not converge for a reasonable number of iterations ) . one can conclude from these plots that a value of @xmath99 would provide a relatively good normalization improvement at the first steps of the unfolding , as well as when one gets an estimate of the background fluctuations . other amplitudes of background fluctuations ( see fig . [ fig : optlnob ] ) could yield to different conclusions . if these fluctuations are larger or not well known , one could prefer to use a large value of the parameter ( which could even come to perform just the standard normalization ) before the first background estimation , and a smaller one afterwards . it has been tested that a change of the centre of the background distribution ( eq . ( [ eq : bkgd ] ) ) by @xmath100 does not change the conclusions related to the normalization parameter . in practice , one can perform the previous study on a model similar to the data or directly using the data and reconstructed mc distribution . in this second case , only the black curve in fig . [ fig : optlnub ] would be available . the plots in the rest of this subsection were obtained for : * an amplitude of the resonance initially centred on the bin @xmath101 ( moved to the bin @xmath102 , with a width @xmath103 for this scan ) , reduced by @xmath104 with respect to eq . ( [ eq : bias1 ] ) ; * the amplitude of the resonance initially centred on the bin @xmath105 ( centred on the bin @xmath106 , with a width @xmath107 for this scan ) , increased by @xmath104 ; * the effect of the background subtraction reduced by @xmath104 with respect to the one in eq . ( [ eq : bkgd ] ) , and centred on the bin @xmath108 , with @xmath109 . the resulting optimal parameters are stable with respect to various changes of this type . the choice of the value @xmath73 ( to be used for the first unfolding step ) is made such that , for different relative amplitudes of the subtraction effect , it is not unfolded , and hence does not bias other regions of the spectrum . we have observed that the value @xmath110 , would be large enough , not to unfold fluctuating background effects for different relative values of their amplitude . in order to show that this parameter can be dropped , and the procedure simplified , we have however choosen a very large value for it , such that the correction done by the first unfolding is given only by the difference between the true and reconstructed mc . the two other parameters to be used and the number of necessary iterations , are analysed using a scan of their values , for test spectra with different amplitudes of the initially unknown resonances . for every set @xmath111 , we compute a distance of the @xmath112 type between the final reconstructed mc and the data minus the events to be subtracted , as well as between the unfolding result and the true spectrum plus background fluctuations , for a number of steps minimizing this last distance ( see fig . [ fig : optlulm ] ) . ( top line ) , @xmath113 ( middle line ) and @xmath114 ( bottom line).,title="fig:",width=188 ] ( top line ) , @xmath113 ( middle line ) and @xmath114 ( bottom line).,title="fig:",width=188 ] ( top line ) , @xmath113 ( middle line ) and @xmath114 ( bottom line).,title="fig:",width=188 ] + ( top line ) , @xmath113 ( middle line ) and @xmath114 ( bottom line).,title="fig:",width=188 ] ( top line ) , @xmath113 ( middle line ) and @xmath114 ( bottom line).,title="fig:",width=188 ] ( top line ) , @xmath113 ( middle line ) and @xmath114 ( bottom line).,title="fig:",width=188 ] + ( top line ) , @xmath113 ( middle line ) and @xmath114 ( bottom line).,title="fig:",width=188 ] ( top line ) , @xmath113 ( middle line ) and @xmath114 ( bottom line).,title="fig:",width=188 ] ( top line ) , @xmath113 ( middle line ) and @xmath114 ( bottom line).,title="fig:",width=188 ] these distances are computed using simply the statistical errors of the data , plus a small component for statistical fluctuations of the mc , for the first one . the anomalous values ( once divided by the number of degrees of freedom ) , obtained sometimes for the optimal parameters , can be explained by the fact that we ignore the change in the errors of the spectrum and they correlations due to the unfolding . however , as we are not interested in the absolute values of these distances , but only on the optimal parameters , this way to proceed is good enough . actually it is even to be preferred due to its inherent sensitivity to some potential instabilities of the iterations ( for inappropriate values of the parameters ) , generating large fluctuations . this sensitivity would be lost if one would compute the distances using for example the errors resulting from a toy simulation . asking for small values of the two distances ( at the unfolded / true mc and respectively data / reconstructed mc level ) , one can clearly discriminate the best region of values for @xmath70 and for @xmath76 . they are valid , not only for the parameters of the spectra used in this scan , but also for different ones used in the test of the method , as we ll see in the following . the corresponding plots indicate here small distance values for @xmath76 and @xmath70 close to zero ( or even null ) , indicating that the use of these parameters is not absolutely necessary . these plots also give even more information about potentially problematic parameterizations yielding for example badly unfolded ( not enough corrected ) new structures if @xmath70 is too large , or to fluctuations if one performs only small unfolding corrections in the iterations ( large @xmath76 ) and the other parameters preventing their propagation are not appropriate . this kind of study has some sensitivity to @xmath58 ( to be used in the subtraction procedure ) too , through the distance at the unfolded / true mc level . actually , if its value is too large , the subtraction is too strong preventing the good reconstruction of new structures . as we saw before , this distance is also enhanced if fluctuations propagate ( for example because of a too small @xmath58 ) . these distances provide however only global comparisons of the spectra . therefore , for complex situations like in this example , a more precise determination of the parameters is provided by local comparisons of the spectra within toy simulations ( i.e. local contributions to the previous distances ) . this kind of comparison performed for the spectra used in the previous scans , or other similar variations , indicates that a value @xmath115 allows one to fix background fluctuations and to prevent the ones of the procedure , without a bias of the reconstruction of new structures . as we will see , the choice of this value , as well as the other parameters , can be tested directly when unfolding the data . the plot of the optimal number of iterations provides a relatively good estimation of the number of steps which are needed in practice , when the corresponding unfolding parameters are used . the confidence one can make to this number depends however on the precision with which one knows the widths of the new resonances . one should keep in mind that this is just an estimation and , once the unfolding parameters are determined , a better suited strategy could be to stop the algorithm when the iterations start having a small impact on the result or yet when one gets a good agreement between data and reconstructed mc . here we present the way the unfolding behaves for the distributions described in section [ subsec : building1 ] . this tests the reliability of the unfolding method as well as the one of the procedure used to optimize the parameters . as explained in the previous subsection for the various test distributions , here also a parameter @xmath110 would be large enough to leave the background contribution in its initial position . the first unfolding step was however performed with a very large value for this parameter ( see fig . [ fig : unf1 ] and [ fig : test1 ] ) and it corrects all the elements of the spectrum which are known by the mc , for both kinds of transfer effects ( in spite of the fact that they are relatively important ) . this first step could also perform part of the needed corrections for the new structures , but we postpone these corrections to further unfolding steps , by taking a very large @xmath73 . a series of 65 iterations was then performed , with complete unfolding and matrix modification ( i.e. taking the parameters @xmath116 and @xmath117 ) . actually , in this case the regularization is provided by the use of @xmath113 , and the use of @xmath76 and @xmath70 can be dropped . a very good estimation of the remaining background fluctuations is obtained at the first iteration , and the further improvements are relatively small . however , for the sake of generality , it has been tested that good results can be obtained , with non - zero values of @xmath76 and @xmath70 . the main effect of these iterations is to correct the folding matrix , by introducing in the true mc distribution ( and implicitly in the reconstructed one ) , the structures which were unknown for the initial simulation ( see fig . [ fig : amx4nsteps ] ) . a first reason for which such a high number of iterations was needed in this test are the large resolution effects which must be corrected for the initially unknown structures , even for the narrow one . it has been tested that between 15 and 20 iterations are generally enough to reconstruct well the new structure centred on the bin 170 , which is slightly larger than the detector resolution , being however strongly affected by its effects too . the final estimation ( see fig . [ fig : data1-reconsteps ] ) reproduces well the background fluctuations added in the region of the dip , for the bins where it was fluctuating as well as for the ones where it was stable . the need to distinguish between potentially remaining background fluctuations and new structures , especially when having an important transfer of events in the unfolding , imposes important constraints on the parameter values that can be employed . this can be a the second reason which can increase the number of needed iterations . the difference between the data and the improved reconstructed mc is basically equal to the background estimate , which is not used when modifying the transfer matrix . the final unfolding result reconstructs well all the structures in the true data , without introducing important systematic effects due to the remaining background ( see fig . [ fig : toyunfoldingn1 ] and fig . [ fig : testnsteps ] ) . the error bars of the unfolding result(s ) were estimated using 100 mc toys , with fluctuated data and transfer matrix for the unfolding procedure . the corrected structures are affected by fluctuations coming from all the data bins contributing to them , which is particularly important for large folding / unfolding effects like the ones in this example . as expected , new structures are more affected than the ones initially known by the mc , because the corresponding data fluctuations can locally affect the improved mc transfer matrix . at this point one could perform a test of the parameterization of the procedure , directly on the data . actually one could perform a series of toy simulations of the unfolding , and compare the average of the unfolding results with the result found previously . a significant difference between the two , in any region of the spectrum , would indicate a problem in the parameterization , which should be reconsidered . an increase of the fluctuation of the unfolding result in the toys with respect to the ones of the data , comparable or larger than the correction performed on new structures ( i.e. by iterations if the first step is trivialized as before ) , would indicate the same kind of troubles . no such problems were observed in the previous example . if the study described in the previous subsection , using global distances between the spectra , is not enough sensitive to the values of the unfolding parameters , one can always build toy models for the data ( following [ subsec : building1 ] ) and perform the unfolding as explained before , in order to tune the method . one can use in the folding the matrix given by the mc , while a statistically fluctuated one will be used in this unfolding . the bias , used to get the true data , and the remaining background fluctuations , will be choosen such that the folded toy data are qualitatively close to the real ones . the local comparison of the unfolding result with the toy true data provides an excellent sensitivity to the parameters of the method . once the method is tuned , the possible ( small ) remaining difference between the unfolding result and the true data can be used to estimate the systematic error of the method . this difference could even be used to compute a correction of the result , provided that it is stable with respect to statistical fluctuations , the uncertainty of the data true model is smaller than this difference , and if one can not further improve the parameters of the method . among the functions listed in eq . ( [ f : regfunctions ] ) , ( [ f:1mexp ] ) and ( [ f:1mfrac ] ) , the ones with a null derivative at the origin seem to provide a better separation of fluctuations and real deviations . the small differences between the results obtained with the several ( well suited ) functions , as well as the ones related to some reasonable changes in the regularization parameter(s ) , can be used to estimate the corresponding systematic errors of the procedure . there is however at least a partial overlap between this estimation of the systematic error and the one proposed in the previous paragraph . the systematic errors related to the unfolding are not only dues to the procedure , but also to the knowledge of the folding matrix . the difference between data and mc at this level can be estimated in practice by detailed studies of events affected by effects that the unfolding procedure must correct . if the amount of some of these events is different in data and in the simulation , one can even correct the last one . this could also reduce the correction the unfolding has to bring at the level of the spectra . one can estimate the uncertainty on the transfer matrix coming from this ( or yet another ) kind of effects / corrections . performing the unfolding with the transfer matrix modified within these uncertainties allows one to compute the systematic error associated to the final spectrum . there can indeed also be an overlap between this estimation of the systematic error due to the transfer matrix and the ones previously associated to the unfolding procedure . we have seen that this unfolding method allows one to unfold structures having a width smaller than the detector resolution . however , these reconstructed structures are for sure affected by the fluctuations of the data bins where their events were initially situated , and by the precision of the knowledge of the transfer matrix . these two effects put a limit on the quality of the reconstructed structures . the use of the regularization functions in this method imply non - linear effects in the passage from data to the unfolded spectrum . this is actually a general feature of the regularization methods used in the unfolding procedures . it can slightly affect the shape of the error distribution of the unfolded spectrum . in this example , this type of effect was observed mainly at the edges of the new structures introduced in data , where it is more difficult to distinguish between statistical fluctuations and real significant deviations . in this section we provide an example that could be closer to the usual applications of this method , namely when the effects of the folding and the data - mc differences are relatively small . we indicate how the method can be considerably simplified for this kind of application and how one can easily build a test ( or a procedure to set the remaining parameter(s ) ) of the method . to build this second example we use the same procedure as the one described in subsection [ subsec : building1 ] , but we bring important changes in the spectra and transfer matrix . in order to provide a rather low statistics test of the method , we reduce the global statistics of mc and data by a factor of about 20 ( their relative ratio is equal to 10 , like in the previous example ) . concerning the transfer matrix , besides this change , we remove the dip around the bin @xmath118 in eq . ( [ eq : tmatrix ] ) and we reduce resolution effects by taking @xmath119 . the effect of systematic transfers of events from high index bins to the lower ones is also reduced by putting @xmath120 and @xmath121 , in eq . ( [ eq : amatrix ] ) . the `` bias '' of the true data with respect to the true mc is changed to a smaller and a smoother one : @xmath122 , \label{eq : bias2}\ ] ] and we do not introduce any remaining background fluctuations in this example . the errors associated to the data points are simply the statistical ones . given the fact that the complexity of this test is considerably reduced , with respect to the previous one , a simplified ( next to trivial ) version of the unfolding procedure allows one to perform the needed corrections . first of all , as in this example there are no significant new structures in the data , the normalization procedure looses its sensitivity and does not indicate any possible improvement at this level . indeed , one can see this by making a scan of the possible @xmath33 values with fluctuating data , as it was shown in the previous example . in the present case , the improvement limits interval contains the null value , for any value of the @xmath33 parameter . one can therefore drop this procedure here , for example by setting @xmath123 ( i.e. by using the standard normalization ) . the possibility of subtracting remaining background fluctuation is also dropped here . in practice one can take this decision by comparing the size of the data point errors with respect to the available statistics , or equivalently using the errors introduced by the background subtraction itself . the possible effect of this simplification in the unfolding strategy can also be tested directly on data , as explained in the previous example . we will finally use a simplified unfolding procedure using a first step with the initial transfer matrix and a parameter @xmath73 , and an iteration where the transfer matrix is improved and a new unfolding is performed . in this example , one iteration modifying the matrix is enough to correct the systematic data - reconstructed mc differences . the figure [ fig : simpexunf_data - recomc ] indicates this difference before the unfolding , and after one iteration . for the example shown here we chosed @xmath124 . although this choice is not crucial for the effect shown in this figure , its motivation will become clear after the following studies . in order to measure the full potential systematic effect of this iteration , we set the corresponding parameters : @xmath125 and @xmath126 ( i.e. we perform the maximal change of the transfer matrix and the maximal unfolding correction afterwards ) . we propose a direct test of the unfolding procedure and its parameterization , searching for potential systematic biases that could be introduced at this level . for this test we build a toy true and reconstructed data , inspired directly by the original data spectrum and the mc simulation ( which could be directly used in practice ) . the toy reconstructed data are then unfolded with a transfer matrix ( @xmath127 ) obtained after a fluctuation of the initial one available in practice ( @xmath43 ) . therefore , the mc for this test is identical to the one in the `` real '' unfolding , up to statistical fluctuations . we compare the result to the toy true data . we define the true data known ( @xmath128 ) by the mc as the normalized initial true mc ( obtained from the matrix @xmath43 , without any additional statistical fluctuations ) . we introduce the data `` known '' by mc , which are the result of the folding of the @xmath128 , using the transfer matrix @xmath43 . they are identical to the reconstructed mc in the first unfolding step , up to statistical fluctuations between @xmath43 and @xmath127 . as usual , we add a bias to the true data known by the mc , in order to get the `` true data '' ( @xmath129 ) . in order to build a test as close as possible to the real situation , the bias is obtained from the difference between data and the normalized initial reconstructed mc , multiplied by a constant factor . the true data are then folded using the transfer matrix @xmath43 . we define the reconstructed data as the result of this folding , with or without final statistical fluctuations . it is interesting to consider here these two variations of the test . the first one , where the data are fluctuated statistically , is closer to the real unfolding operation and tests the existence of potentially spurious effects due to the limited statistics in the data ( and mc ) . the second test , allows an easier search for potential systematic effects of the method , which one can look for in the difference between the unfolding result and the true data . if the multiplicative factor , used to obtain the bias , equals one , the data - `` known data '' difference in this test , is very close to the data - reconstructed mc difference in the real unfolding procedure ( up to final statistical fluctuations ( if any ) , transfer matrix fluctuations and smoothing folding effects ) ( see fig . [ fig : simpextp_data - recomc ] ) . we consider this factor equal to one in the following test ( if not stated otherwise ) . just as in the `` real unfolding '' ( see fig . [ fig : simpexunf_data - recomc ] ) all the systematic effects in the data - reconstructed mc difference are corrected after the first iteration . a test as the one build here is well suited , provided the data - mc difference and the unfolding corrections are relatively small . we measure the bias of the unfolding result with respect to the true data , after the first step and after one iteration . .,width=604 ] at this level one can not observe any systematic bias after the first unfolding step , and the impact of the iteration is very small . this study was performed for several sets of independently generated data . in practice , one could build several models starting from statistically fluctuated data samples . in order to measure the potential bias , we compute it in larger bins ( each of which contains 40 initial bins ) . for @xmath73 values up to at least @xmath130 we find maximal relative biases in the interval @xmath131\%$ ] . this interval becomes @xmath132\%$ ] for @xmath133 , @xmath134\%$ ] for @xmath135 and @xmath136\%$ ] for very large @xmath73 values ( i.e. when no unfolding correction is done on the data - rmc difference ) . this study shows that from the point of view of the potential bias remaining after the first unfolding step , the @xmath73 values up to @xmath130 provide very comparable results . however , the larger the @xmath73 value , the less smoothing is done on the data , as the fraction of their fluctuations kept in their initial bin for the final spectrum is larger , and the bins of the final spectrum are less correlated . this type of test could also be used to identify situations where the detector transfer matrix is ( almost)singular . such a situation would be encountered for example in the rather pathological situation of a detector with very bad resolution , in which case the data would be compatible with the @xmath137 no matter what @xmath138 distribution was used . therefore , the unfolding result would also be compatible with any other @xmath138 that was used to build the transfer matrix . it is at this point that the test described in this subsection comes into play . using various true data distributions for this test , one would assign a very large systematic error to the unfolding result , making it compatible with any spectrum preserving the number of events . this would of course indicate that the solution of the unfolding is not well defined . just as in the case of the previous test , for the `` real '' unfolding , the first step provides the main systematic correction , while the effect of the iteration is very small ( see fig . [ fig : simpexunf_impiter ] ) . .,width=604 ] the impact of the first iteration on the unfolding result can be seen as another measurement of the systematic error of the method . just as in the case of the toy test , we measured this effect in larger bins ( each of which contains 40 initial bins ) . we perform this study for several sets of independently generated data and transfer matrix . for @xmath73 values up to at least @xmath130 we find an averaged maximal ( over 5 large bins ) improvement of the order of @xmath139 with however important relative fluctuations between different data sets . when increassing @xmath73 , the potential improvement gets larger and more significant with respect to fluctuations . its value is approximatively @xmath140 for @xmath133 , @xmath141 for @xmath135 , and @xmath142 for very large @xmath73 values . this confirms the previous conclusion concerning this test , namely that for @xmath143 the first unfolding step provides already a good result , while the iteration(s ) could be useful for larger @xmath73 values . as explained before , using a larger @xmath73 prevents the unfolding and smoothing of the data fluctuations , keeping the bins less correlated . this effect can be observed at the level of the diagonal errors of the unfolding result , after one step ( see fig . [ fig : simpexunf_errors ] ) . ( dotted line ) , for @xmath144 ( solid line ) , and the data errors ( dashed line).,width=604 ] indeed , the diagonal errors after the first unfolding step for @xmath124 are larger than the ones obtained for @xmath144 , the two being smaller than the ones of the data , due to correlations . if , after the unfolding , the bin - to - bin correlations enhanced by the transfers of events are too large for the further use of the spectrum , one could decide to compute this spectrum in less correlated , larger bins ( if this would not imply a too large lost of information on the shape of the spectrum ) . this could be achieved through the use of a different ( smaller ) number of bins for the @xmath138 distribution or by a rebinning operation after the unfolding . all the functions listed in eq . ( [ f : regfunctions ] ) , ( [ f:1mexp ] ) and ( [ f:1mfrac ] ) yield very similar results for the unfolding in this example , although the suitable regularization parameters can be different for them . in this paper we introduced an unfolding procedure allowing one to deal with a series of problems that one can meet in practice , and which were not considered by the previous methods . using a regularization function , one can consistently compare mc with data spectra , even when the last one contain structures which were not initially simulated . it is the same regularization function that allows one to unfold spectra containing important fluctuations from the background subtraction , without generating biases by the transfer of these events in other regions of the spectrum . further more , one can dynamically avoid spurious fluctuations of the method when performing iterations to reconstruct structures that were not initially simulated . this method also allows the user to keep under control the bin - to - bin correlations of the final spectrum , while the needed systematic corrections are performed . we have described a general unfolding strategy that can be used for rather complicated situations , where all the problems discussed before are present . a simplified strategy has been studied and it can be applied for more common problems . we have described the way one can build reliable toy tests inspired by data , which can provide a good parameterization of the procedure . we also propose some tests for the intermediate or final results of the procedure , to be performed directly when unfolding the data . statistical errors can be directly propagated in this method . we have proposed several ways of estimating the systematic effects associated to the method , as well as the ones dues to the poor knowledge of transfer matrix . the c++ source code ( relying on root functionality ) for the unfolding procedure ( and the various described tests ) can be obtained upon request from the author . several variations of the unfolding strategy and/or intermediate steps , besides the ones mentioned in the text , are possible . in the general strategy , one could for example simultaneously compute the estimate of the remaining background fluctuations and improve the transfer matrix . this could be achieved by a straight - forward merging of the formulas [ eq : imprbu ] and [ eq : impra ] , with the use of the same @xmath145 parameter . one could even place the events from the estimate of the remaining background fluctuations on the diagonal of the transfer matrix . in this case , the corresponding events are not moved by the unfolding , just because they are `` predicted '' to be kept in place by the transfer matrix . further on , at the different steps of the general unfolding strategy described in this paper ( or of its possible variations ) , one could use different functions or parameters , which would yield an even more general method . the author would like to thank volker blobel , michel davier , sebastien descotes - genon and andreas hcker for very interesting and useful discussions on these topics , and for a careful reading of this manuscript . this work was supported in part by the eu contract no . mrtn - ct-2006 - 035482 , flavianet . as the acceptance correction will be performed at the very end , only events passing all the physical cuts will be implicitly considered when computing the quantities in the following definitions : * @xmath13 = number of bins in the data and in the reconstructed mc distribution * @xmath14 = number of bins in the unfolding result and in the true mc distribution * @xmath42 = the number of simulated events which were generated in the bin @xmath40 and reconstructed in the bin @xmath41 . * @xmath45 = the probability for an event generated in the bin j to be reconstructed in the bin i ( folding probability matrix , as estimated from the mc simulation ) * @xmath46 = the probability for the `` source '' of an event reconstructed in the bin i to be situated in the bin j ( unfolding probability matrix , as estimated from the mc simulation ) * @xmath35 = rebinning transformation , to pass from data to unfolding bins . this transformation can be non - linear . * @xmath36 = inverse rebinning transformation , to pass from unfolding to data bins . this transformation can be non - linear . * @xmath146 = number of reconstructed data events in the bin i ( after background subtraction ) * @xmath147 = number of events in the bin i of the data distribution which are potentially due to a fluctuation in the subtracted background * @xmath148 = number of reconstructed data events in the bin i , which do not correspond to structures which are not `` known '' by the mc . this number is exactly known in a simulation , but only an estimation of it can be computed when unfolding real data . * @xmath149 = number of true data events , `` known '' by the mc . this distribution can be computed for toy tests only . * @xmath150 = events corresponding to structures in the true data , which are not simulated in the mc . * @xmath151 = error of @xmath146 * @xmath17 = number of unfolded events in the bin j * @xmath18 = number of events in the bin j of the unfolded distribution which are potentially due to a fluctuation in the subtracted background * @xmath152 = error of @xmath17 * @xmath153 = number of reconstructed mc events in the bin i * @xmath154 = error of @xmath153 * @xmath155 = number of true mc events in the bin j * @xmath156 = error of @xmath155 * @xmath11 = number of events in the data and , by preservation , in the unfolding result * @xmath10 = number of events in the true and reconstructed mc spectra * @xmath9 = estimate of the number of events in the data , which were simulated in the mc * @xmath0 = regularization function , depending on @xmath1 , the absolute deviation between two values to be compared , @xmath2 , the estimate of the total error to be used for the comparison of the given bin of two spectra , and @xmath5 , a positive regularization parameter . a. hocker and v. kartvelishvili , nucl . instrum . a * 372 * , 469 ( 1996 ) [ arxiv : hep - ph/9509307 ] . v. blobel , desy 84 - 118 , [ arxiv : hep - ex/0208022 ] . a. kondor , jinr - e11 - 82 - 853 . l. lindemann and g. zech , nucl . instrum . meth . a * 354 * , 516 ( 1995 ) . g. dagostini , nucl . instrum . a * 362 * , 487 ( 1995 ) . d. acton _ et al . _ [ opal collaboration ] , z. phys . c * 59 * , 1 ( 1993 ) . h. n. multhei and b. schorr , math . methods appl . * 9 * , 137 ( 1987 ) . h. n. multhei and b. schorr , nucl . instrum . a * 257 * , 371 ( 1987 ) . c. amsler et al . , physics letters b667 , 1 ( 2008 ) available on the pdg www pages ( url : http://pdg.lbl.gov/ )

we describe techniques for comparing spectra extracted from cosmological simulations and observational data , using the same methodology to link lyman- properties derived from the simulations with properties derived from observational data . the eventual goal is to measure the coherence or clustering properties of lyman- absorbers using observations of quasar pairs and groups . we quantify the systematic underestimate in opacity that is inherent in the continuum fitting process of observed spectra over a range of resolution and . we present an automated process for detecting and selecting absorption features over the range of resolution and of typical observational data on the lyman- `` forest '' . using these techniques , we detect coherence over transverse scales out to 500 kpc in spectra extracted from a cosmological simulation at .

astro-ph0110450

the numerous lines of lyman-@xmath0 absorption that appear in the spectra of quasars are proving to be excellent cosmological probes . hydrodynamic simulations of the universe reveal an evolving network of sheets , filaments and halos caused by complex gravitational dynamics in the expanding universe . the lyman-@xmath0 `` forest '' only identifies the neutral component of the intergalactic medium , but the absorbers are well understood tracers of the overall mass distribution @xcite . the simulations show that the physical state of the diffuse gas causing lyman-@xmath0 absorption is relatively simple , enabling the use of analytic models to relate the absorption to the underlying mass and velocity fields @xcite . the combination of theory and simulation has allowed the lyman-@xmath0 forest to be used for measurements of the baryon fraction @xcite , the mass density @xcite , the amplitude of mass fluctuations @xcite , the mass power spectrum @xcite , the thermal history of the igm @xcite , the chemical evolution of the universe @xcite , and the metallicity of the igm @xcite . on the observational side , the detection of coincident absorption lines in the spectra of quasar pairs provided the first evidence that lyman-@xmath0 absorbers have a large transverse extent @xcite . in principle , this idea can be extended with the use of quasar groups to thread a contiguous volume and provide 3d `` tomography '' of the absorbers , revealing the variation of coherence and homogeneity with redshift . in practice , however , quasars of the appropriate brightness , redshift , and angular separation are hard to find . moreover , at @xmath4 , the opacity drops steadily and the lyman-@xmath0 forest thins out into a savannah , in agreement with numerical simulations @xcite . over most of the hubble time , the absorbers can only be observed in the vacuum ultraviolet with the twenty times smaller collecting area of the hubble space telescope ( hst ) . studies of the low redshift lyman-@xmath0 forest are still very much limited by the quality of the available data . figure [ fig-7 ] summarizes the state of observational capabilities for studies of the lyman-@xmath0 forest . the boxes give the approximate bounds in resolution and @xmath1 of several different instrument and telescope combinations . open boxes show examples of ground - based facilities that can only measure lyman-@xmath0 , at @xmath5 ; shaded boxes show past , present and future hst instruments . the horizontal bar shows the range of observed doppler parameters of the absorbing gas , which is independent of redshift @xcite . only echelle observations are sensitive to the thermal and hydrodynamic properties of the absorbing gas . the best quality data come from keck / hires and vlt / uves . at lower redshift , the stis echelle mode will be supplanted by the cosmic origins spectrograph ( cos ) . there is a large jump in resolution down to the level ( 1 - 2 ) where a larger number of targets is available and a larger redshift path length can be surveyed . the keck / lris region is typical of most moderate resolution ground - based data . the volume of existing data will be eclipsed by the @xmath6 high redshift quasar spectra that will emerge from the sloan digital sky survey @xcite . at lower redshift , the largest single data set comes from the hst quasar absorption line key project @xcite . the goals of this paper are to ( a ) define a set of automated procedures that can be used to analyze data ( with a wide range of spectral resolution , @xmath1 , and physical separations between paired lines of sight ) and spectra extracted from simulations , ( b ) form a bridge between the historically different methodologies of observers and simulators , and ( c ) facilitate the comparison between observations and simulations of the lyman-@xmath0 forest , in order to derive cosmological constraints . the major challenge to the first goal is the fact that absorption features in a quasar spectrum must be measured with respect to an unknown continuum , and in the presence of broad emission features . this is usually accomplished with a low - order function fitted to the continuum , but no procedure is fully robust in dealing with broad absorption features or the region just blue - ward of the lyman-@xmath0 emission line . even the comprehensive software of the hst key project can not automatically deblend complex spectral regions @xcite . by contrast , spectra extracted from the hydrodynamic simulations yield opacities directly with respect to a predetermined continuum . in addition , observers have traditionally treated absorption in terms of discrete features modeled by voigt profiles , which in turn must be convolved with the instrumental response function . this approach is observationally successful , particularly at low redshift where spectral features have low number density and appear isolated in the spectra . regardless of whether or not a voigt profile provides a suitable description of the absorbing gas , there is evidence from extracted simulation spectra that much of the opacity is associated with shallow and smoothly varying absorption . this is commonly referred to as the fluctuating gunn - peterson approximation @xcite . with regard to the second goal , there are substantial differences between the extraction of information from optical or uv spectra , and the the extraction of information from a spectrum created as a product of a cosmological simulation . observational spectra have finite @xmath1 and resolution , each one spans a large 1-dimensional redshift range , and the lyman-@xmath0 lines must be identified in spectral regions contaminated with metal lines from higher column density absorbers . details of the method for culling out metal lines from the lyman-@xmath0 forest will be given in paper ii . spectra extracted from hydrodynamic simulations have no noise and extremely high resolution , although the spectral features result from gas kinematics that map onto velocity space in a complex way . by contrast to the @xmath7 gpc or larger 1d path of the lyman-@xmath0 forest in a quasar spectrum , the simulation boxes span 20 - 40 @xmath2 mpc and are aliased for large scale structure measurements on scales larger than 5 - 10 @xmath2 mpc . however , the simulations are ideal for 3d measures of the gas distribution since @xmath8 independent spectra can be extracted , whereas 3d observations of absorber structure are limited by the rarity of suitable quasar pairs . the third goal of this project will be realized with the second and subsequent papers in this series . in the second paper , we will compare coherence measures of absorbers at @xmath9 with extractions from simulations . in future papers , we will extend the work to lower redshifts and investigate the dependence of the predicted absorber properties on cosmological parameters . this paper sets the stage by describing the techniques for automated and consistent comparisons between observations and simulations of the lyman-@xmath0 forest . in section 2 , we discuss the detection and selection of absorption features . in section 3 , we discuss the measurement of physical properties of the absorbers . in section 4 , we discuss results from single and multiple sight - lines , including a comparison between line counting and measures that use the entire information contained in the spectra . section 5 contains a brief summary . a numerical simulation of a @xmath3 cold dark matter ( cdm ) universe , with @xmath10 km s@xmath11 mpc@xmath11 , normalized to yield a present - day rms mass fluctuation of @xmath12 in spheres of radius 16 @xmath2 mpc , was performed by @xcite using a method based on smoothed - particle hydrodynamics , or sph @xcite . the simulation volume is a periodic cube of comoving size 22.2 @xmath2 mpc drawn randomly from a cdm universe with @xmath13 and baryon density @xmath14 . the simulation includes the effects of a uniform photoionizing radiation field , where radiative heating and cooling rates are computed assuming optically thin gas in ionization equilibrium with this field . there are @xmath15 sph particles and an equal number of dark matter particles ; the masses of the individual particles are @xmath16 and @xmath17 , respectively . the gas resolution varies from @xmath18 kpc in the highest density regions to @xmath19 kpc in the lowest density regions . although more recent simulations have been performed with higher spatial resolution and with the parameters of the currently favored cosmological model ( @xmath20 km s@xmath11 mpc@xmath11 , @xmath21 , @xmath22 ) , most of the issues that relate to comparisons with observations can be illustrated with this single simulation . spectra are extracted from the simulation cube by computing the imprint of the density , temperature , and velocity of the neutral gas fraction associated with each sph gas particle on a flat continuum ( for more details , see katz et al . one thousand sets of spectra were extracted from the simulation cube by randomly selecting 1000 positions from the x - projection . for each of the 1000 primary lines of sight ( plos ) , six adjacent spectra were extracted that are offset 33.3 , 100 , 233 , 400 , 667 , and 1000 @xmath2 kpc in proper distance from each plos . the azimuth angle of the adjacent spectra was varied randomly from one primary sightline to another . this strategy allows the examination of coherence of the absorption features in the extracted spectra over these ( and other intermediate ) transverse separations , in addition to enabling comparisons with observational data for quasar pairs having a range of separations . the transverse separations between the lines of sight range from less than the resolution of the simulation to a level where absorber coherence is expected to be very weak . each spectrum is sampled with 1000 data points spanning 1924.5 km s@xmath11 , which corresponds to a spectral range of 23.4 at @xmath3 . figure [ fig-9 ] shows six spectra extracted from the simulation illustrating the range of absorption features , with the mean opacity increasing going from panel @xmath23 to @xmath24 . panels @xmath25 and @xmath26 have a typical opacity for extracted spectra at this redshift , and less than 1% of the sight - lines are as heavily absorbed as the example in panel @xmath24 . one objective of this paper is to compare absorber properties measured from the observed spectra to those measured from the spectra extracted from the simulation . for a fair comparison to be made , the extracted spectra are processed to reflect the inevitable limitations of observational data due to instrumental effects of noise and resolution . the spectra are extracted from the simulation cube in terms of optical depth as a function of velocity , and are converted to transmission , @xmath27 , as a function of wavelength in angstroms , @xmath28 , where @xmath3 . as can be seen in figure [ fig-7 ] , the bulk of the uv data on the lyman-@xmath0 forest , which contains information about the igm over the last three - quarters of the age of the universe , is of rather modest @xmath1 and resolution . this means that the complex dynamical processes that imprint themselves on the absorber profile are not visible at low redshift , even in the few published echelle spectra @xcite . with non - echelle spectra , the spectral profile is dominated by the ( typically ) gaussian shape of the instrumental profile , and doppler parameters can not be measured . for the purposes of this paper and the following paper ii , we chose a range of values for the resolution , @xmath29 , and @xmath1 typical of most published quasar spectra . to span the parameter space of available data at any redshift we chose @xmath1 = 10 , 30 , and 100 , and @xmath29= 5 , 20 , 80 , and 300 km s@xmath11 , and form twelve realizations of a single spectrum extracted from the simulation , each with a different combination of @xmath1 and resolution . the extracted spectra are convolved with a gaussian profile having a fwhm , @xmath29 , with each of the four values chosen to represent the instrumental resolution of the available observational data . the velocity resolution of the extracted spectra ( @xmath30 km s@xmath11 ) is much higher than the resolution of observed quasar spectra , the range of which is represented by the four values 5 , 20 , 80 & 300 km s@xmath11 . we resample the extracted spectra to 3 per resolution element by spline interpolation . poisson noise is added to each spectrum by scaling the intensity at each pixel to the @xmath31 , adopting a poisson deviate with that value as the mean , and then renormalizing the spectrum . figure [ fig-8 ] shows two different extracted spectra each having an opacity similar to the mean at @xmath3 , exemplifying the typical absorption morphologies at that redshift . the twelve realizations for each individual extracted spectrum show the degraded appearance due to the various choices of @xmath1 and resolution . the highest resolution , highest @xmath1 spectrum ( top right panel in each case ) is a fairly close match to the original spectrum extracted from the simulation . the best published spectra from the keck telescopes appear like those in the top right panels , but there are a significant number of hst spectra that are similar in quality to the lower left panels . these plots clearly show the separate effects of @xmath1 and resolution on the appearance of the spectrum . the measurement of absorption features in quasar spectra , whether by line profile fitting or by using a type of flux statistic , must be performed relative to the quasar continuum . continuum fitting is generally done by iteratively fitting the data points in a spectrum , eliminating negative or downward excursions from the fit , and refitting . the position of the final adopted continuum depends on the @xmath1 of the spectrum and , more importantly , on the total flux absorption ( which is directly related to the line density ) and its distribution across the spectrum . such a procedure only approaches the `` true '' continuum in the limit of infinite signal to noise , and even then has the potential to miss weak or shallow absorption features . figure [ fig-8 ] demonstrates how features that can be seen clearly in the highest @xmath1 , highest resolution spectra are lost in lower @xmath1 , lower resolution spectra . given that there are no regions of truly zero absorption , it is essentially _ guaranteed _ that line and continuum fitting will miss some portion of the lyman-@xmath0 opacity . the systematic underestimation of the continuum level will also tend to underestimate the equivalent width and therefore the column density of fitted lines too . we use extracted spectra from the simulation to calibrate the size of this effect , and measure it as a function of @xmath1 and resolution . with many thousands of spectra to process , an automated procedure is required . after the simulation spectra are manipulated to mimic observational data , the continuum must be estimated as for the observed spectra . any algorithm must be simple conceptually and should be able to fit continua automatically for spectra with a range in absorber opacity corresponding to a redshift range of @xmath32 , and for spectra having a range of signal - to - noise ratios , @xmath33 . any method implicitly assumes that the data retains some of the quasar s emission continuum , and this implies that continua fit to highly absorbed regions will be less representative of the true continuum . mean absorption opacity and @xmath1 are the primary factors in determining how well the true ( or input ) continuum can be recovered . fitting a cubic spline is a standard method of estimating continua for observed spectra @xcite . splines offer the advantage of being able to fit a smooth , continuous function through a large number of data points . in quasar spectra , the large available wavelength range acts to define a smooth , slowly - varying continuum . however , the simulation volume offers only a short spectral range , and spline fitting to short extracted spectral segments with significant absorption can produce steeply - sloped continua . we suppress this effect in the case of the extracted spectra by `` tripling '' the spectrum , or lining up three identical spectra , taking advantage of the fact that the simulation cube is periodic . this is merely a practical mechanism for dealing with the edge effects in continuum and absorption line fitting for the analysis , only lines whose centers fall in the central third of the new spectrum ( the original extracted spectrum ) are included . to estimate a continuum computationally , representing what one might fit `` by eye , '' we implement a two step process . the first step is to iteratively fit a straight line to 23 sections of a spectrum ( corresponding to the independent spectral segment length for the extracted spectra ) , providing an estimate of the amplitude of the continuum over that wavelength range . this is a reasonable way to proceed because emission continua of quasars can be closely approximated by series of linear segments over @xmath34 scales , except in the vicinity of lyman-@xmath0 emission . all data points are weighted equally and an iterative process rejects points that deviate negatively by @xmath35 from the current fit , excluding them from subsequent fits . the rejection process converges after only four cycles . the method of continuum fitting used by the hst absorption line key project divides a spectrum into bins and fits a spline to the average flux in these bins . however , this bin size is larger than the length of our extracted simulation spectra , so the first step described above provides our first estimate to the continuum . the second step produces a smoothly varying continuum over the length of the spectrum using the amplitude of each of the fitted line segments from the first step . a single point is created for each of the three segments by evaluating the amplitude of the fitted straight line at the average wavelength of each segment . a cubic spline is fit through these points for the length of the spectrum . visual inspection of spectra with a range of flux decrements demonstrates that this two step method results in a more satisfactory fit to the continuum than the key project method , especially for heavily absorbed spectra , as the first step has the effect of `` floating '' the continuum to what appears `` by eye '' to be a more appropriate level . to evaluate how well the algorithm works for the observational spectra , we compare the final fitted continua using the new two - step process to the continuum fit using the method used by the hst absorption line key project . some manual adjustment of the continuum is necessary in the vicinity of the steeply sloped lyman-@xmath0 emission features . four of the six spectra to be analyzed in paper ii of this series pg 1343 + 264a , b , lb 9605 , and lb 9612 were studied using both methods , and the ratio of the two fitted continua is unity at all wavelengths to well within the errors . this demonstrates that no significant systematic effects are introduced by the continuum fitting algorithm . we can also infer that absorption line parameters returned by the two methods will not significantly differ . to show how @xmath1 and resolution affect the level of the fitted continuum , we fit continua using our method to 1000 randomly - selected spectra extracted from the simulation . we then compare the mean opacity computed for each of 1000 undegraded ( i.e. raw ) spectra extracted from the simulation to the mean opacity calculated relative to the fitted continuum for each degraded spectra in each of the 12 realizations of @xmath1 and @xmath29 . figure [ fig-11 ] shows the distribution of the difference of these two quantities for all 1000 spectra and for each realization . this difference is equivalent to the difference between the input continuum level ( unity , by definition ) and the fitted continuum level , converted into an opacity decrement , @xmath36 . an opacity decrement is the unavoidable consequence of any fitting procedure where the continuum is not known _ a priori _ as is the case for all observational data . the quantity @xmath36 therefore represents `` lost '' opacity , and it has two components . the first is due to the fact that strong absorbers will have their column densities slightly underestimated owing to the lower placement of the continuum . the second is due to the sum of broad and/or low level absorption that can not be recovered by fitting a noisy continuum ( the gunn - peterson approximation ) . see figure [ fig-9]@xmath37 for a good example . figure [ fig-16 ] shows how the severity of underestimating opacity depends on both @xmath1 and resolution by comparing the mean opacity measured for the undegraded spectra to the mean opacity measured relative to the fitted continuum for the degraded spectra . in figure [ fig-11 ] , we plot the distribution of the difference of these two quantities , @xmath36 . at the level of the highest quality echelle data ( top - right panel ) , 80 - 90% of the absorption opacity is recovered in almost every case . at the level of poor quality hst data ( lower - left panel ) , the opacity underestimation is typically 50% . the tail to high @xmath36 in each panel represents the relatively rare cases of heavily absorbed spectra , where no automated procedure can adequately recover the absorption . we have examined the general effects of a wide range of resolution and @xmath1 on the appearance of the spectra and on the ability to recover the true continuum level . for the phase of absorption line measurement , we home in on a narrower range of parameter space with the four combinations : @xmath38 and 30 , and resolution @xmath29@xmath39 and 300 km s@xmath11 . it can be seen from figure 1 that these values bracket most existing hst data and much of the anticipated data from the sloan quasar survey . this range also encompasses the data we will use in paper ii for a lyman-@xmath0 coherence measurement at @xmath3 . despite the conceptual limitations of the line / continuum fitting paradigm , we acknowledge the fact that a huge amount of observational data has been published using this type of procedure . a major goal of this paper is to form a bridge between the appropriate techniques used to analyze simulations and the traditional methods of quasar spectroscopy . available quasar pair data on the lyman-@xmath0 `` forest '' have @xmath1 and resolution generally no better than @xmath40 and @xmath29@xmath41 km s@xmath11 . this corresponds to the highest @xmath1 and resolution of the four parameter spaces for which we are measuring absorption lines in the extracted spectra . dectection of the weakest lines is limited by the @xmath1 and the resolution of the data , and measurement of the column density depends on the doppler parameter . however , integration of the evolution function for lyman-@xmath0 absorbers shows that less than 1% of absorbers are expected to have column densities higher than @xmath42 @xcite . the range of doppler parameters for lyman-@xmath0 absorbers at @xmath9 spans approximately 20 - 80 km s@xmath11 @xcite . for an absorber with a median doppler value @xmath43 km s@xmath11 , @xmath42 , and @xmath29@xmath41 km s@xmath11 , the line profile is just saturated or reaches zero flux at the line center . for higher resolution data , line - fitting methodologies represent individual absorbers with a voigt profile , but we demonstrate that for column densities less than @xmath42 and resolutions lower than @xmath29@xmath41 km s@xmath11 , the use of a gaussian profile is justified . we generate the flux profile for a lyman-@xmath0 absorption line ( with essentially infinite @xmath1 and resolution ) using subroutines from the program autovp @xcite for the case of @xmath44 and @xmath45 km s@xmath11 . the convolution of this `` intrinsic '' line profile with the instrumental line spread function ( a gaussian with @xmath29 ) is the expected flux line profile . sampling of the dispersion is chosen to mimic that of the degraded simulation spectra for this parameter space . one thousand absorption lines were created and poisson noise was added to represent @xmath46 . a gaussian profile was fitted to each monte carlo simulated line . this fitted gaussian profile was compared to the monte carlo line profile , yielding an average @xmath47 value of 0.011 . for 50 degrees of freedom ( the number of data points in the profile ) , this corresponds to a probability of less than @xmath48 that the two profiles are different . we also note that low resolution ( @xmath29@xmath49 km s@xmath11 ) renders undetectable all the complexities in the absorber profile caused by hydrodynamic effects . the use of a gaussian profile in fitting absorption features is therefore adequate for our immediate purposes , and for our initial science application in paper ii . absorption line selection and measurement is performed using software originally written by tom aldcroft @xcite that has since been substantially revised . the most recent version of the software retains the original code structure , the graphic interface tools , and the fundamental numerical subroutines . new methodology for fitting continua , and new algorithms for selecting and fitting absorption features have been implemented . the continuum - fitting process is described in the previous section . line selection and fitting is a two - step process where a preliminary list of lines is selected , which forms a first estimate to the simultaneous fit that is determined in a subsequent step . the line selection and fitting methodology was first described in @xcite . some refinements to the line fitting algorithm have been added to generalize finding the techniques for a range of data quality and to eliminate manual intervention . the range of data quality addressed in this work is @xmath50 to 30 and @xmath29@xmath41 to 300 km s@xmath11 . selecting a combination of absorption lines profiles to fit an absorption region is a difficult because of the effects of instrumental resolution and @xmath1 on the true features in the spectrum . the second extraction in figure [ fig-8 ] shows an example ( at @xmath51 ) where lines that are easily resolved at 20 km s@xmath11 can not be separated at a resolution 80 km s@xmath11 . the @xmath1 is not as important a factor in concealing the structure of an absorption feature until it becomes very low . no procedure can recover information that is lost due to limited instrumental resolution , but gaussian profile fitting does provide a fair comparison between the simulation spectra and observed spectra . even if a gaussian is a good approximation to the shape of strong absorption features , we do not expect that line - fitting can yield unique fits . the biggest danger is expected to be over - fitting , such as when a single broad feature is fit by several overlapping gaussians . however , it is also possible to under - fit , such as when noise causes a weak feature to fall below the significance threshold for a single unresolved line . in what follows , we describe a robust procedure for dealing with most of the situations presented by real quasar spectra . one simple way of fitting a number of gaussian profiles to a predefined absorption region is to fit the peak of a gaussian to the lowest flux point in the spectrum , subtract the profile from the spectrum , and then repeat the process at each local minimum . the method is considered to have converged when the result is consistent with the flux error of the original spectrum . however , with this method , a region which may be better fit in a @xmath47 sense by two barely resolved lines , will actually be fit by one strong line in the center , possibly with an unphysically large width , straddled by two smaller `` satellite '' lines . the sequential approach has the basic problem that lines are defined one at a time , which constrains the choice of subsequent lines , and does not ensure an optimal or unique fit . we have devised an iterative method to fit the maximum number of lines to a region and distribute the lines optimally across the region . unresolved lines are assumed , i.e. the absorber is well approximated by the instrument line profile ( a gaussian ) . subsequently , all combinations of lines from this preliminary list are fit simultaneously to the region using a marquardt minimization technique , and the fit that meets a preset criteria is chosen as the best fit . we consider that the best fit is the deepest depression in the @xmath47 surface of solutions . the marquardt minimization does not allow for large changes in the line center or line width ; by design , the preliminary step of finding lines and subsequent testing each combination of lines samples the surface of solutions well enough to find the best solution . this is justified in practice because the iterative procedure rarely leads to substantial changes in the centers and widths of the strongest features . in the first phase of line selection , an iterative search is made for all minima in the data array containing the flux values for each spectrum that has been convolved with the instrumental profile . primary _ line list is formed from minima found in the first pass and a _ secondary _ list is made from minima found in the convolved array after lines in the _ primary _ list are subtracted from the original data . primary _ and _ secondary _ lists combine to form the _ preliminary _ line list that defines an initial `` best guess '' at the final line list . the rationale for combining primary and secondary lists is as follows . the minima in the data array locate the strongest absorption features , but these inflection points do not account for all the absorption in a blended region . the secondary list serves to `` interleave '' the primary list , and combination of the two yields a fairly complete and robust map of the absorbers . the iterative method for initially locating and measuring lines in the spectrum of each component follows that described in detail by paper of the hst quasar absorption line key project @xcite . the equivalent width , @xmath52 at every pixel in each spectrum is computed by centering the spectral line profile on the @xmath53 pixel and performing the weighted sum over the 6@xmath54 limits of the gaussian forming the array @xmath55 . similarly , this is done for the error array , @xmath56 , and the interpolated error array , @xmath57 . the interpolated error array is calculated by replacing flux errors for data points that deviate negatively by more than 2@xmath54 by the average of the errors in up to five ( on each side ) of the adjacent continuum points . continuum in this sense is defined by points lying within a 2@xmath54 deviation from the fitted continuum . this array differs from the error array in that the errors at the centers of the absorption lines are increased slightly to reflect the noise in the adjacent true continuum . the @xmath1 of an unresolved line is conventionally defined as @xmath58 , whereas we choose to define `` significance '' as @xmath59 , which is used to select preliminary lines in the first phase . this correction for the drop in the errors at the centers of absorption lines , where the flux errors are smaller , allows for a more uniform comparison of relative line strengths . the primary line list is generated by identifying the minima in the @xmath55 array within a window of width one half the instrumental resolution that has a significance ( @xmath60 ) of three or more . a parabolic fit through the two adjacent points identifies the line center and the value of @xmath55 and @xmath56 corresponding to this wavelength give the equivalent width and error . the profiles of the lines are calculated and subtracted from the original spectra . this subtracted spectrum is then convolved and a secondary line list was derived in the same way . the secondary list is then subtracted from the original spectrum and a pass was made again that refits only the primary lines and may add new lines or drop existing lines . the number of lines added by the secondary list increases by a factor of 1.5 to 2 times the number of lines found in the primary list . it was found empirically that after approximately five iterations of finding lines for first the primary and then the secondary lists , the number of lines converged to a repeating series as the line centers shifted about slightly depending on the noise . the number of lines that are added or lost by all subsequent iterations is no more than a few percent of the total list . the preliminary line list was formed from the combination of the two lists , and if their profiles are subtracted from the original data array , the result has a mean flux of zero and variations entirely consistent with noise . the primary list of lines is input as a first guess to the simultaneous best fit of each absorbed region in a spectrum . for practical reasons the spectrum is broken up into regions . there are several ways of doing this and we chose to step through each spectrum and define the region to be fit as beginning where the flux array downward - crosses the continuum and ending where it upward - crosses the continuum . this method of defining the spectral regions is reasonable for data of resolution and @xmath1 under consideration here , because the number of upward- and downward - crossings divides the spectrum into sections that can be fit by a moderate number of preliminary lines , and because it limits the number of regions where no preliminary lines are found . fitting all combinations of the preliminary lines makes it computationally impractical to fit more than 11 lines per region , so in the rare occasions that this situation occurs the region is split at the highest flux data point in the region , and the fit is performed for the new region with a smaller number of lines . lines from the initial list that fall within the region to be fit were counted , @xmath61 , and the number of all possible combinations of these lines , @xmath62 , was computed by summing the binomial coefficient , @xmath63 , which gives the number of combinations possible for @xmath61 lines taken @xmath64 positions at a time , over the number of possible positions , @xmath65 @xmath66 the derived lines for each individual combination were worked out and saved , and each combination of lines was fit simultaneously to the region following a marquardt minimization technique that varies the amplitude , center and width of the lines . the line parameters and the reduced chi square ( @xmath67 ) , which is later used to select the best fit , for each combination were saved . the method of fitting all combinations of preliminary lines optimizes selection of the best fit to any absorption feature by maximally utilizing the information provided by the preliminary line list . however , in practice , @xmath67 is not a sensitive test of overfitting in selecting the best fit to a region @xcite . for example , it is possible for a fit to produce a @xmath67 that is improved from the prior fit but where the errors in the equivalent width are larger than the equivalent width itself . in our application of this procedure , the central wavelengths are fairly well determined and do not change more than a few standard deviations with subsequent fits . we therefore imposed secondary constraints to ensure that the final fit consisted of meaningful line parameters . the final `` best '' fit is chosen as the fit with the lowest @xmath67 that also fulfills the following criteria : ( a ) @xmath67 @xmath68 , ( b ) the errors in the fitted equivalent width and the fwhm are smaller than the measurement itself , ( c ) the minimum line separation for lines in a fit is @xmath69 , and ( d ) 0.65 @xmath70 fwhm @xmath70 4.5 . the first two of these constraints apply generally to the fitting procedure . the restriction on @xmath67 is an empirically determined practical limitation to the largest value expected to be associated with a reasonable fit . the best fit can have a @xmath67 that is rather high , simply due to the fact that there may be a portion of the fit region where no lines are fit but where the normalized flux is not exactly at the continuum level . failure of this criterion forces a redetermination of the fitted region by dividing it at the wavelength having the next highest intensity , and a refitting of the preliminary lines to obtain a fit having a lower @xmath67 . the restriction on the errors of fitted parameters to lines prevents over - fitting . the last two constraints depend on the range of @xmath1 and resolution of the data being considered . in this case , the choices were justified as follows . the minimum and maximum fwhm values are determined by convolving the range of expected doppler widths @xcite with the instrumental profile , and allowing for variation in the minimum fwhm due to noise . the minimum line separation also depends on resolution and noise , and was chosen empirically by forming the distribution of line separations and then examining the reliability of individual fits . instrumental resolution in this test case corresponds to 80 km s@xmath11 . there is a natural break in the distribution at @xmath71 km s@xmath11 and lines selected with smaller separations than this represent an over - fit with respect the features in the undegraded simulation spectrum . lines with separations of more than @xmath72 km s@xmath11 are reliably selected by the algorithm . for lines with separations between 70 and 90 km s@xmath11 it is not clear by eye whether adding the second line to the fit is justified , and in all cases the undegraded simulation spectra shows an asymmetry or two features close together . a monte carlo simulation to determine the recovery rate of two close lines shows that the software does not reliably recover the input lines until they are separated by @xmath7 100 km s@xmath11 . this motivates our minimum separation of 1.12 . however , we note that the statistical results of the analysis are not particularly sensitive to this exact choice . if the procedure fails one or more of the tests in any particular spectral region , the region is split at the next highest minimum intensity point and a new set of simultaneous fits is made using the smaller number of lines . if all of these fits fail , the last fit is chosen but is tagged . the tagged lines were examined for a subset of 200 spectra and were only 3% of the total number of lines selected . two thirds of these were moderately broad , weak features whose shape had been distorted with the addition of noise in general , a line was found in the vicinity of the true feature even if the addition of noise and resolution made it more difficult to detect . almost all of the lines from `` failed '' fits are weak enough that they will not meet the selection criteria to be included in the subsequent analysis . about one in 5 were strong broad features where the chosen fit appeared appropriate , but the @xmath67 was too high . a remaining tiny number , barely 0.01% of the whole sample , are weak features fit to the edge of broader absorption regions or are not real features at all . this same visual inspection of many of the fits showed that under - fitting or over - fitting is not a significant problem for the algorithms . we conclude that the fraction of strong regions of absorption that are not well fit by the line / continuum algorithm is negligible . the procedures for line - fitting are robust and can be applied in an automated way , but they are also somewhat complex . figure [ fig-12 ] shows how they work in practice , as applied to the two extracted spectra from figure [ fig-8 ] , and in each of the four combinations of @xmath38 and 30 , @xmath29@xmath41 and 300 km s@xmath11 . in other words , figure [ fig-12 ] shows actual line / continuum fits to the four parameter choices at the lower left of figure [ fig-8 ] . fitted continua and lines are overplotted in each case with the location of significant ( @xmath73 ) absorbers shown by tick marks . `` significance '' is defined in two ways : @xmath74 is related to the flux error and is used to describe the strength of a line in terms of the detection limit of the data ; @xmath75 is the error in the equivalent width returned by the software and is a measure of line reliability or goodness of fit . in practice , @xmath74 can be used to set a uniform detection threshold for absorbers and is directly related to the @xmath1 , where @xmath75 is an indicator of how well the equivalent width is determined lines with large @xmath75 might be found in noisy regions of the spectrum or in the wings of broader absorption features . all lines fit by the software are retained regardless of strength or reliability , but for the analysis we define two categories : _ secure _ and _ marginal_. the criterion for inclusion in either list is @xmath73 . lines that are @xmath76 make up the _ marginal _ list , and lines that have an equivalent width @xmath77 constitute the _ secure _ list . marginal lines may be reliable enough to be useful in the comparison with observed spectra , but in this work we only consider lines from the _ secure _ list in the analysis . in figure [ fig-12 ] , longer , thicker tick marks show the location of the secure lines , and shorter , narrower tick marks denote lines from the marginal category . absorption features were measured using the line - fitting algorithm for 300 primary lines of sight ( plos ) and for their adjacent sight - lines spanning separations from 33.3 to 1000 @xmath2 kpc a total of 2100 spectra extracted from the simulation . this was done for two different signal - to - noise ratios , @xmath78 and 30 , and two resolutions @xmath29@xmath79 and 300 km s@xmath11 , forming four realizations of the data . we compared the distribution of absorber properties measured from the simulated spectra to those found from observations of large samples of spectra published in the literature as a basic cross - check of our method . in figure [ fig-14 ] , we show the distribution of the line centers of absorbers in 300 plos of the @xmath46 , @xmath29@xmath41 km s@xmath11 realization for each of the three projection axes . the deviations from the mean ( shown as the dotted line ) indicate large scale , coherent structures within the simulation box , having filamentary and sheetlike properties . however , the average of the 3 projections converges toward the mean value indicating that the absorbers approach a uniform distribution over the full size of simulation box , and demonstrating that there are no artifacts introduced by edge effects . additionally , we examine the distribution of the number of absorbers per line of sight in figure [ fig-15 ] . this distribution is consistent with a gaussian distribution that is the expectation for a random distribution of absorbers . in particular , the high multiplicity of absorbers that would indicate substantial clustering is not seen . table [ tbl-1 ] lists the average number density ( @xmath80 ) of secure lines per line of sight above a limiting rest equivalent width threshold ( @xmath73 ) for each of the four realizations of the simulated spectra ( column 5 ) . for sample 1 ( column 1 ) we used 1000 plos for increased statistics on the highest @xmath1 and resolution realization , and for the other three samples we used 300 plos . the number of lines from observations is obtained in two ways from a study performed by @xcite . these values are listed in columns 6 and 7 of table [ tbl-1 ] . first , we simply count the number of absorbers found in the small wavelength range spanned by the simulated spectra at @xmath3 ( j. scott , private communication ) . however , a limited number of absorbers are found in this 23 wide region and so the error bars are large . since the evolution function ( described below ) is close to linear over @xmath81 , the poisson error can be reduced by using the lines counted in this five times larger spectral region as an estimate for the smaller spectral region actually spanned by the simulation . second , the average number of lines at @xmath3 can be computed from coefficients of an evolution function fitted to an appropriate sample of observed absorbers . this method is less direct but it has the merit of incorporating most of the available data . the number of lyman-@xmath0 lines per unit redshift per unit equivalent width can be described as follows : @xmath82 integrating equation 2 with respect to @xmath83 gives the number of lines per line of sight with a rest equivalent width greater than 0.16 for the chosen values @xmath84 , @xmath85 , and @xmath86 from @xcite : @xmath87 further integration with respect to @xmath88 gives the number of lines per spectrum in the range @xmath89 : @xmath90.\end{aligned}\ ] ] the number of lines in this redshift interval for a different equivalent width limit , @xmath91 , is obtained by scaling @xmath92 by the factor @xmath93.$ ] errors on @xmath92 are obtained using the @xmath94 error bars on @xmath95 . see column 7 in table [ tbl-1 ] . a comparison is made only for the @xmath29@xmath41 km s@xmath11 realizations ( samples 1 and 3 ) , because this resolution closely matches the data set used by @xcite . the error bars listed in table [ tbl-1 ] are merely reflect the poisson error and do not include systematic uncertainties due to differences in the line - fitting algorithms that can be as large as 10 - 20% . sample 3 most closely matches the sensitivity of the data set used by @xcite , and the number of lines found in the extracted spectra agree to within 20% with both methods for counting observed lines . sample 1 in table 1 agrees with the ( poorly - determined ) number predicted using a model for the evolution of the absorbers , and it also agrees to within 20% of the expected number from the direct line counting method . in addition to a check of aggregate line counts , it is important to check whether extracted spectra from the simulation return the observed distribution of line strength . we therefore compare the number of lines per equivalent width bin , @xmath96 , to a measurement that uses the data of @xcite . this recent study provides better statistics at @xmath3 than the hubble space telescope absorption line key project @xcite . figures [ fig-13 ] and [ fig-2 ] are plots of the number of secure ( @xmath77 ) absorbers counted in the spectra from the simulations , binned by rest equivalent width , for samples 1 and 3 . both distributions are well within the @xmath94 error bars on @xmath95 , showing that the absorber statistics obtained by line - fitting applied to the simulation spectra are consistent with those from observed samples . distributions formed that include the marginal lines ( @xmath97 ) show that the number of lines continues to increase in accordance with the exponential distribution down to @xmath98 , and inspection of figure [ fig-12 ] shows that these _ marginal _ lines always correspond to a real feature in the `` true '' spectrum . we conclude that absorber line counting from this hydrodynamic simulation provides an excellent match to the demographics of the lyman-@xmath0 forest at @xmath3 . the first indications that lyman-@xmath0 absorbers were larger than the halos of individual galaxies came from the detection of coincident lines in the spectra of quasar pairs @xcite . most paired lines of sight experiments define absorbers pairs by choosing a maximum separation in velocity , typically 50 - 300 km s@xmath11 , that lines must have in order to be labeled as coincident and therefore physically associated . the operational definition of a matching `` window '' depends on the line density , and it is chosen to minimize the probability of a chance match . a `` coincident '' pair is two lines that match in velocity closely enough that the probability of a chance match is small . however , the preselection of a velocity match window can potentially lose information , and it makes an implicit assumption about the kinematic state of the absorbing gas ( i.e. the amplitude of any velocity shear on that particular transverse scale ) . a very large number of paired lines of sight can be extracted from the simulations , with many potential absorber pairs at each of the transverse separations . we have designed a matching algorithm that does not depend on any prior definition of a velocity match . this allows an examination of how the coherence of the absorbers changes over transverse separations of 33 @xmath2 kpc to 1 @xmath2 mpc . coincident lines are defined such that coincidences are symmetric . in other words , if there is a line in sightline @xmath99 , its coincident line is the nearest line in velocity space in sightline @xmath100 . however , coincident lines are also selected starting with sightline @xmath100 and matching to sightline @xmath99 . an absorber is not identified as a coincident line unless it is a reversible match both from @xmath99 to @xmath100 and from @xmath100 to @xmath99 . this procedure will of course result in lines that are unmatched in both sight - lines . we account for the fact that the simulation cube is periodic by matching across the ends of each simulation spectrum . to discuss line matching statistics , we have chosen to use the single realization of the simulation spectra with @xmath1 = 30 and @xmath29@xmath41 km s@xmath11 . this most closely mimics the observational spectra of the three quasar pairs at @xmath3 that we will analyze in paper ii . absorber matches were found for 300 primary sight - lines seven extracted spectra make up each set . the fraction of matched lines is 87% , 73% , 65% , 61% , 59% and 58% for transverse separations 33 , 100 , 233 , 400 , 667 , 1000 @xmath2 kpc , respectively . the fraction of matched lines for random lines of sight formed by pairing lines of sight from the x - projection with lines of sight from the z - projection is 57% . figure [ fig-4 ] illustrates the line - matching procedure . figure [ fig-4]@xmath101 is merely a test of the algorithm . figure [ fig-4]@xmath102 shows the effect of noise and subsequently slight differences in the degraded spectra on the scatter between the equivalent widths of matched lines . at a @xmath46 , only 3% of the lines fail to match due to fluctuations added by random noise . figure [ fig-4]@xmath103 shows that there can be substantial differences in line strength even across a transverse scale of 33 @xmath2 kpc , which is below the resolution of the simulation where intrinsic differences must be small . nevertheless , the formal correlation is highly significant . figure [ fig-4]@xmath37 shows the null experiment , where absorbers detected in 300 primary sight - lines are matched against 300 randomly selected sight - lines from an orthogonal projection of the simulation . this ensures that any two features at similar wavelengths are separated by 10 - 20 @xmath2 mpc in space , and so are expected to be uncorrelated . in figure [ fig-4]@xmath102 , the scatter is larger than anticipated due to the addition of noise because the line selection procedure must deal with issues of deblending and continuum fitting . the very few outliers ( out of 924 line pairs ) are mostly caused by the situation where the addition of noise to a marginally resolved feature causes it to be fit with two components in one sightline but only one in the other . at first sight , the large scatter in figure [ fig-4]@xmath103 is more surprising , because the transverse separation is at the limit of resolution of the simulation . the scatter comes about because the mapping from real space to spectral space in complex . peculiar gas motions affect the optical depth both by shifting line centers and by changing line profiles due to velocity gradients ( bi & davidsen 1997 ) . the fact that our line fitting software is responding to real physical differences in opacity can also be seen from the analytic modelling of viel et al . they show calculated lyman-@xmath0spectra at @xmath104 with noticeable differences on transverse scales as small as 60 @xmath2 kpc , where a significant fraction of lines fit with voigt profiles have column densities that differ by more than a factor of two . taken together with the added effects of noise , the scatter in figure [ fig-4]@xmath103 can be readily understood . a major result of this paper is shown in figure [ fig-5 ] . this shows matched or coincident lines plotted against the velocity difference of the line pair for the six transverse separations with respect to the plos . ( of course , it is possible to generate many transverse separations between 0 and 1000 @xmath2 kpc using these spacings , but the choices in figure [ fig-5 ] illustrate the coherence phenomenon adequately . ) given the mean line spacing of 850 km s@xmath11 ( for a @xmath105 line ) , the line matching algorithm becomes aliased at 425 km s@xmath11 , so matches with separations above this value are not physically meaningful . the most sensitive test of absorber coherence uses the fact that truly coincident lines will have a small velocity separation between sight - lines , where random matches ( not physical associations ) will have a larger range of velocity differences . figure [ fig-5 ] suppresses the information on homogeneity by plotting the average of the equivalent widths of the paired absorbers against the velocity splitting of the pair . the distribution of the velocity splittings of matched pairs changes quite dramatically with transverse separation . to quantify this , we plot the cumulative distribution of the velocity splittings for the paired absorbers in figure [ fig-6 ] . to show the expectation for a random set of absorbers with no physical association between the sight - lines , we also form the cumulative distribution for pairings between spectra extracted from orthogonal projections through the simulation volume . any two paired absorbers in this case will have a physical separation of @xmath710 - 20 @xmath2 mpc and so are not anticipated to show coherence ( the dashed lines in figure [ fig-6 ] ) . a k - s test demonstrates that there is an excess of small velocity splittings , and therefore detectable coherence , up to about 500 @xmath2 kpc . a typical method of examining the homogeneity of lyman-@xmath0 absorbers over various transverse separations is to plot the equivalent widths of each line in a matched pair . however , noise and intrinsic differences in the absorbing regions imprint themselves on a spectrum and affect the line fitting algorithm in a complex way , so that apparently similar spectra can yield coincident line pairs with substantially different equivalent widths . to see if line matching depends on line strength , we formed distributions of equivalent widths for both the paired lines as well as the unpaired lines . a k - s test shows these distributions are drawn from the same distribution . even though there are many more weak ( lower column density ) lines than strong ( higher column density ) lines , the degree of coherence on any particular transverse scale does not depend on line strength . however , the fraction of matched lines decreases with increasing transverse scale ; a clear demonstration that coherence is being lost as the scale approaches 1 @xmath2 mpc . up until now we have been examining the coherence of the lyman-@xmath0 absorbers using the highest density peaks that are fitted as lines . clustering or coherence may also be analyzed through the use of continuous statistics that leverage the flux information in every independent point of the spectra . various types of correlation measures are well described in @xcite . we define the two - point correlation function ( tpcf ) of the normalized absorbed flux , @xmath106 , along neighboring lines of sight as @xmath107 where @xmath108 and @xmath109 refer to fluxes along the two lines of sight . figure [ fig-17 ] shows the tpcf for each of the transverse separations 33 , 100 , 233 , 400 , 667 , and 1000 @xmath2 kpc . we compute the auto - correlation function , which is the plos correlated with itself ( @xmath110 in equation 5 ) , and this is shown as the curve with the highest amplitude . we also compute the tpcf for lines of sight that are randomly associated by correlating the plos from spectra extracted from the x - projection with plos from the z - projection . the tpcf for discrete absorbers in a line - counting experiment is computed for all transverse separations by comparing the number of observed nearest neighbor pairs , @xmath111 , with the number of pairs formed from random lines of sight , @xmath112 as follows : @xmath113 the random lines of sight are the plos from the x- and z- projections . figure [ fig-17 ] shows the correlation functions for each of the six transverse separations . the amplitude of the tpcf for each method decreases to approach the curve for random demonstrating that coherence can be measured using line - fitting methods . in paper ii , we will illustrate the different types of information that are gleaned by using continuous flux statistics and line - counting techniques , and we will quantify the amount of data required to detect coherence using line counting . this paper has established techniques for the intercomparison of quasar spectra and spectra extracted from cosmological simulations , recognizing that observations are traditionally interpreted in terms of line - counting , while simulations offer a direct measure of neutral hydrogen opacity at every resolution element . this work anticipates a series of direct comparisons using space- and ground - based observations of quasar pairs and groups . initial tests have been carried out on the lyman-@xmath0 forest at @xmath114 , with simulation spectra degraded to span the range of observational quality of typical ground and space - based data . the main results are as follows : \(1 ) simulators measure opacity with respect to a pre - defined continuum , while observers must determine a quasar continuum that is not known _ a priori _ , in the presence of noise and emission features . software has been designed that robustly measures continua both for the small wavelength ( redshift ) range of simulations and for the long wavelength range of typical observations . the systematic underestimate of opacity ranges from @xmath115 for @xmath116 , @xmath29@xmath117 km s@xmath11 to @xmath118 for @xmath38 , @xmath29@xmath119 km s@xmath11 . \(2 ) a fully automated procedure has been developed for the detection and selection of absorption features that produces reliable results for observational data having @xmath50 to 30 and @xmath29@xmath41 to 300 km s@xmath11 . the techniques are robust even in extended regions of strong absorption . a direct comparison with published quasar surveys shows that the number density and equivalent width distribution of absorption lines at @xmath114 agrees with measurements of spectra extracted from an sph simulation . \(3 ) a technique has been established to match absorption lines between adjacent lines of sight . the match rate of coincident lines is used to search for coherence in the absorbing gas on transverse scales up to 1 @xmath2 mpc . the two - point correlation function of matched pairs reveals a significant excess out to transverse scales of @xmath7 500 @xmath2 kpc , indicating the detection of lyman-@xmath0 coherence on this scale . most of the signal of excess pairs occurs with velocity splittings of @xmath120 km s@xmath11 , indicating that the velocity field is quiet . the coherence signal measured with line counting and matching techniques agrees well with results from a two - point flux correlation analysis , which uses all the information in the simulation spectra . we are particularly grateful to tom aldcroft , who made available the code that formed the core of the software suite described in this paper , and jennifer scott , who shared her thesis data . we acknowledge useful discussions with romeel dav , craig foltz , jane charlton , and rupert croft . this work was supported by nasa astrophysical theory grants nag5 - 3922 , nag5 - 3820 , and nag5 - 3111 , by nasa long - term space astrophysics grant nag5 - 3525 , and by the nsf under grants asc93 - 18185 , ast-9803072 , and ast-9802568 . aldcroft , t. 1993 , ph.d . thesis , stanford university aguirre , a. et al . 2001 , , 556 , 11 aguirre , a. et al . 2001 , , submitted [ astro - ph/0006345 ] bahcall , j.n . 1993 , , 87 , 1 bechtold , j. , crotts , a.p.s . , duncan , r.c . , & fang , y. 1994 , , 437 , l83 bi , h. g. 1993 , , 405 , 479 bi , h. g. & davidsen , a. f. 1997 , , 479 , 523 cen , r. , miralda - escud , j. , ostriker , j.p . , & rauch , m. 1994 , , 437 , 9 cen , r. , phelps , s. , miralda - escud , j. , & ostriker , j.p . 1998 , , 496 , 577 cen , r. , & ostriker , j.p . 1999 , , 519 , l109 croft , r.a.c . , weinberg , d.h . , katz , n. , & hernquist , l. 1998 , , 495 , 44 croft , r.a.c . , weinberg , d.h . , pettini , m. , hernquist , l. , & katz , n. 1999 , , 520 , 1 croft , r.a.c . , et al . 2001 , submitted [ astro - ph/0012324 ] crotts , a.p.s . , & fang , y. 1998 , , 502 , 16 dav , r. , hernquist , l. , weinberg , d.h . , & katz , n. 1997 , , 477 , 2 dav , r. , hellsten , u. , hernquist , l. , katz , n. , & weinberg , d.h . 1998 , , 509 , 661 dav , r. , hernquist , l. , katz , n. & weinberg , d.h . 1999 , , 511 , 521 d.h . , & katz , n. 1997 , , 477 , 2 dinshaw , n. impey , c.d . , foltz , c.b . , weymann , r.j . , & chaffee , f.c . 1994 , , 437 , l87 dinshaw , n. , foltz , c.b . , impey , c.d . , & weymann , r.j . 1998 , , 494 , 567 gnedin , n.y . , & hui , l. 1996 , , 472 , l73 gnedin , n.y . 1998 , , 299 , 392 gunn , j.e . , & peterson , b.a . 1965 , , 142 , 1633 haehnelt , m , & steinmetz , m. 1998 , , 298 , l21 hellsten , u. et al . 1997 , , 487 , 482 hernquist , l. & katz , n. 1989 , , 70 , 419 hernquist , l. , katz , n. , weinberg , d.h . , & miralda - escud , j. 1996 , , 457 , 51 hu , e.m . , kim , t .- s . , cowie , l.l . songaila , a. , & rauch , m. 1995 , , 110 , 1526 jannuzi et al . 1998 , , 118 , 1 katz , n. , weinberg , d.h . , & hernquist , l. 1996 , , 105 , 19 katz , n. , weinberg , d.h . , hernquist , l. , & miralda - escud , j. 1996 , , 457 , 57 kirkman , d. , & tytler , d. 1997 , , 484 , 672 mcdonald , p. et al . 2000 , , 543 , 1 mcgill , c. 1990 , , 242 , 544 miralda - escud , j. , cen , r. , ostriker , j.p . , & rauch , m. 1996 , , 471 , 582 nusser , a. , & haehnelt , m. 2000 , , 313 , 587 penton , s.v . , stocke , j.t . , & shull , m.j . 2000 , , 130 , 121 petry , c.e . , impey , c.d . , & foltz , c.b . 1998 , , 494 , 6 penton , s.v . , shull , m.j . , & stocke , j.t . 2000 , , 544 , 150 rauch , m. , carswell , r.f . , chaffee , f.h . , foltz , c.b . , webb , j.k . , bechtold , j. , & green , r.f . 1992 , , 390 , 387 rauch , m. , et al . 1997 , , 489 , 7 rauch , m. et al . 1997 , , 481 , 601 rauch , m. 1998 , , 36 , 267 ricotti , m. , gnedin , n.y . , & shull , j.m . 2000 , , 534 , 41 schaye , j. et al . 2000 , , 318 , 817 schneider , d.p . , et al . 1993 , , 87 , 45 scott , j. , bechtold , j. , & dobrzycki , a. 2000 , , 130 , 37 tripp , t.m . , savage , b.d . , & jenkins , e.b . 2000 , , 534 , 1 viel , m. , matarrese , s. , mo , h. j. , haehnelt , m. g. , & theuns , t. 2001 , astro - ph/0105233 weinberg , d.h . , miralda - escud , j. , hernquist , l. , & katz , n. 1997 , , 490 , 564 weinberg , d. h. , katz , n. , & hernquist , l. 1998 , in asp conference series 148 , origins , eds . c. e. woodward , j. m. shull , & h. thronson , ( asp : san francisco ) , 21 weinberg , d.h . , croft , r.a.c . , hernquist , l. , katz , n. , & pettini , m. 1999 , , 522 , 563 weymann , r.j . , rauch , m. , williams , r. , morris , s. , & heap , s. 1995 , , 438 , 650 weymann , r.j , et al . 1998 , , 506 , 1 york , d.g . et al . 2000 , , 120 , 1579

in this work , the capacity of the gaussian full - duplex ( fd ) two - hop relay channel with self - interference ( si ) is studied . in the two - hop relay channel , the source transmits data to a relay , which forwards the data to the destination , and a direct link between the source and the destination is not present . since the relay is assumed to be fd , the relay is transmitting and receiving at the same time and in the same frequency band . we consider a practical fd relay , where the si caused by the transmitted signal to the received signal is taken into account . thereby , the si is modeled as an additive gaussian random noise whose variance is proportional to the instantaneous power of the transmit signal . we show that the capacity of the gaussian fd two - hop relay channel is achieved either by a gaussian or by a discrete input distribution at the relay . in contrast , the capacity achieving distribution at the source is gaussian with a varying variance in the form of a waterfilling power allocation across different symbols . we numerically compare the capacity of the gaussian fd two - hop relay channel with the achievable rates of conventional fd relaying , optimal half - duplex ( hd ) relaying , and conventional hd relaying . here , conventional fd relaying is shown to be nearly capacity achieving for all cases , where fd relaying significantly outperforms hd relaying .

1510.07394

in today s networks , relay - based transmission is adopted to bridge large distances between a source and a destination . in this work , we investigate the two - hop relay channel comprised of a source , a relay , and a destination , where a direct link between the source and the destination does not exist . there are two main modes of relay operation , the full - duplex ( fd ) mode and the half - duplex ( hd ) mode . in the fd mode , the relay transmits and receives at the same time and in the same frequency band . as a result , self - interference ( si ) caused by the relay s transmit signal impairs the received signal and can cause a severe degradation of the achievable rate of the relay channel . therefore , often the hd mode is preferred for practical implementations , where the relay transmits and receives either in different frequency bands or in different time slots to avoid si . information theoretical analyses of the capacity of the two - hop hd relay channel were provided in @xcite , @xcite . thereby , it was shown that the capacity of the hd relay channel can not be achieved by schemes using a fixed transmission / reception schedule for the relay . instead , for the gaussian two - hop hd relay channel , the optimal relay input symbol distribution is discrete and includes a zero symbol , while the source transmits gaussian distributed symbols when the relay transmits a zero symbol and is silent otherwise , see @xcite . hence , in each symbol interval either the source or the relay transmit a zero symbol . thereby , the positions of the zero symbols at the relay are dependent on the data received from the source . therefore , as the positions of the zero symbols in the codeword of the relay are random for the destination , the relay s zero symbols also carry information . on the other hand , the latest advances in hardware design have shown that it is in principle possible to suppress the si at the relay to such an extent that the fd mode outperforms the hd mode . as the transmit power of fd nodes is about 100 db higher than the noise floor , elaborate si mitigation techniques are necessary , see @xcite . if the si is canceled completely , the ideal fd relay channel , consisting of two successive additive white gaussian noise ( awgn ) channels , results . for the ideal fd relay channel , the capacity achieving input distributions at the source and the relay are gaussian , see @xcite , @xcite . however , in practice even after performing si mitigation , the residual si from the transmitted signal to the received signal is not negligible , and hence , its influence has to be taken into account . the achievable rates of the fd relay channel for gaussian input distributions at the source and the relay were already derived in @xcite . the authors of @xcite investigated the achievable rates of the fd multiple - input multiple - output relay channel using gaussian input distributions . in order to optimize the transmission rate , a hybrid hd / fd scheme was proposed in @xcite , where the transmission rates of conventional hd relaying and conventional fd relaying were compared in order to switch between hd and fd transmission depending on the instantaneous channel state . thereby , both the hd and the fd transmission schemes employed gaussian input distributions at the relay and at the source to transmit over the gaussian two - hop relay channel , which is not optimal in general , see @xcite . in this work , the capacity of the gaussian two - hop fd relay channel is studied , where the source - relay and relay - destination links are both affected by awgn . the residual si at the relay is modeled as a gaussian distributed additive noise whose variance is proportional to the instantaneous power of the symbol transmitted at the relay in the same symbol interval . we show that , in order to achieve the capacity of the considered channel , the source has to use gaussian signaling where the variance of the source symbols depends on the power of the si , and as a consequence , depends on the relay s transmit symbols . this dependence leads to a waterfilling solution , where the source decreases its transmit power for increasing power of the relay s transmit symbol . in fact , when the amplitude of the transmit symbol of the relay exceeds a certain threshold , the source is silent . the optimal distribution of the relay input symbols is shown to be gaussian as long as the relay - destination link is the bottleneck of the end - to - end transmission . otherwise , the relay employs a discrete input distribution in order to keep its influence on the source - relay channel small . we numerically compare the capacity of the gaussian fd two - hop relay channel with the capacity of the gaussian hd two - hop relay channel and the achievable rates of conventional fd and hd transmission . here , conventional fd relaying is shown to be nearly capacity achieving for all cases , where fd relaying significantly outperforms hd relaying . in this section , we present the models adopted for the gaussian fd two - hop relay channel and the si . the source - relay and the relay - destination channels are both assumed to be awgn channels . the source - relay channel is additionally impaired by si noise caused by the relay transmit symbols . there is no direct link between the source and the destination . therefore , the received symbols at the relay , @xmath0 $ ] , and the destination , @xmath1 $ ] , can be modeled as functions of the transmitted symbols at the source , @xmath2 $ ] , and the relay , @xmath3 $ ] , as follows @xmath4&=h_{sr}x_1[k]+h_{rr}x_2[k]+n_r[k]\label{r1}\\ r_2[k]&=h_{rd}x_2[k]+n_d[k],\label{r2}\end{aligned}\ ] ] where @xmath5 , @xmath6 , and @xmath7 denote the coefficients of the channel from the source to the relay , the si channel at the relay , and the channel from the relay to the destination , respectively . moreover , @xmath8\sim\mathcal{n}(0,\sigma_r^2)$ ] and @xmath9\sim\mathcal{n}(0,\sigma_d^2)$ ] are awgns at the relay and the destination , respectively . here , @xmath10 denotes a gaussian distribution with mean @xmath11 and variance @xmath12 . the signal part of @xmath0 $ ] is @xmath13 $ ] , whereas @xmath14+n_r[k]$ ] represents the impairments . as @xmath3 $ ] is known at the relay , si cancellation is employed to suppress @xmath14 $ ] , see @xcite . to this end , @xmath6 has to be estimated . the estimate , @xmath15 , can be modeled as @xmath16 , where @xmath17 denotes the estimation error . as the signal received at the relay for estimation of @xmath6 is corrupted by awgn , the estimate , @xmath15 , is also corrupted by gaussian noise , and therefore the estimation error , @xmath17 , is gaussian distributed with zero mean and variance @xmath18 . thereby , the value of @xmath18 depends on the employed estimation scheme , see @xcite . therefore , after removing the contribution of @xmath19 $ ] , the received symbols at the relay can be modeled as @xmath20&=h_{sr}x_1[k]-\tilde{h}_{rr}x_2[k]+n_r[k],\label{r3}\end{aligned}\ ] ] where @xmath21 . as @xmath17 is multiplied with the relay transmit symbol @xmath3 $ ] , the residual si in symbol interval @xmath22 is gaussian distributed with zero mean and variance @xmath23 $ ] . hence , if the relay does not transmit or transmits a zero symbol , there is no si noise , and therefore hd relaying is obtained as a special case . furthermore , the si noise power scales linearly with the instantaneous relay transmit power , which is in agreement with measurement results obtained for the residual si noise power , see e.g. @xcite . note that an alternative model used in @xcite , @xcite , and @xcite assumes that the si power depends on the average relay transmit power . we note that , in practice , all channel coefficients , @xmath5 , @xmath6 , and @xmath7 , have to be estimated . however , as the si link is much stronger than the source - relay link and the relay - destination link , i.e. , @xmath24 and @xmath25 , the estimation error of the si link is much stronger than the estimation errors of the source - relay and the relay - destination links . therefore , in the adopted channel model , we assume @xmath5 and @xmath7 are perfectly known and take only the estimation error of the si channel into account . without loss of generality , we divide @xmath26 $ ] in by @xmath5 and @xmath1 $ ] in by @xmath7 to obtain the normalized system model shown in fig . [ gaussianinstantinterference ] as @xmath27&=\frac{r_1[k]}{h_{sr}}=x_1[k]+z_{1}[k]+z_{3}[k]\label{channel123}\\ y_2[k]&=\frac{r_2[k]}{h_{rd}}=x_2[k]+z_{2}[k],\label{channel}\end{aligned}\ ] ] where @xmath28\sim\mathcal{n}(0,\sigma_1 ^ 2=\frac{\sigma_r^2}{h_{sr}^2})$ ] , @xmath29\sim\mathcal{n}(0,\sigma_2 ^ 2=\frac{\sigma_d^2}{h_{rd}^2})$ ] , and @xmath30\sim\mathcal{n}(0,\alpha x_2 ^ 2[k])$ ] , where @xmath31 is the si suppression factor of the adopted si cancellation technique . the channel model in , is memoryless , i.e. , @xmath32y_2[k]|x_1[k]x_2[k])\label{memoryless}\end{aligned}\ ] ] holds , where @xmath33 , @xmath34 , @xmath35 , and @xmath36 are sequences of symbols @xmath2 $ ] , @xmath3 $ ] , @xmath37 $ ] , and @xmath38 $ ] , respectively . additionally , a ` " ` per - node ` " ` power constraint @xmath39=p_i\le p_{i,\mathrm{max } } , \quad i=1,2 , \label{con3}\end{gathered}\ ] ] is assumed in this paper . in this section , we study the capacity of the gaussian fd two - hop relay channel with si . in order to study the capacity of the two - hop relay channel , we define random variables @xmath40 , @xmath41 , @xmath42 , and @xmath43 taking symbol values @xmath44 , @xmath45 , @xmath46 , and @xmath47 , respectively . the capacity of the memoryless degraded relay channel was derived in ( * ? ? ? * theorem 1 ) as @xmath48 as a direct link does not exist in the two - hop relay channel , the capacity expression in ( [ con1 ] ) simplifies to @xmath49 now , we study the optimal distribution @xmath50 , which maximizes the capacity in ( [ con2 ] ) . we note that @xmath51 can be written as @xmath52 where @xmath53 only influences @xmath54 in ( [ con2 ] ) . therefore , we first optimize @xmath53 for a given @xmath55 such that @xmath54 is maximized . then , we study the optimal distribution @xmath56 which maximizes ( [ con2 ] ) . the optimal distribution @xmath57 and the resulting mutual information of the source - relay hop are given in the following theorem . [ theo1 ] the optimal source input distribution @xmath57 , which maximizes ( [ con2 ] ) , is gaussian with zero mean and variance @xmath58 . moreover , the optimal @xmath58 is given by @xmath59^+ , \end{aligned}\ ] ] where @xmath60^+=\mathrm{max}\{0,x\}$ ] and @xmath11 is found from the following equation @xmath61 using @xmath62 and @xmath11 , the maximal mutual information @xmath54 is given by @xmath63 where @xmath64 . distribution @xmath53 only influences the term @xmath54 in ( [ con2 ] ) . as the random variable @xmath40 is impaired by additive gaussian noise with variance @xmath65 , @xmath54 is maximized by a gaussian @xmath53 , see @xcite . inserting a gaussian @xmath53 with zero mean and variance @xmath58 into @xmath54 and considering the power constraint leads to the following optimization problem @xmath66 as is a convex optimization problem , theorem [ theo1 ] can be obtained in a straightforward manner with the lagrangian method . the solution in theorem [ theo1 ] is similar to the waterfilling that is performed to optimally allocate power over parallel awgn channels . the waterfilling solution for the problem at hand is schematically illustrated in fig . [ waterfilling ] , where the total noise power at the relay is plotted as a function of the value of the transmitted symbol @xmath55 . in addition , an exemplary waterlevel is shown . as the noise at the relay increases quadratically with respect to @xmath67 , the source stops transmitting , if @xmath67 exceeds @xmath68 . this value reflects the integration limits in ( [ 36 ] ) and ( [ 35 ] ) . if @xmath69 holds , both the source and the relay transmit . thereby , the optimal scheme at the source facilitates a soft transition between hd and fd transmission unlike the hard transition in the hybrid hd / fd transmission protocol proposed in @xcite . , width=226,height=151 ] in this subsection , the optimal relay input distribution @xmath56 is studied . first , the relationship between @xmath54 and @xmath71 for the optimal distribution is investigated , which leads to the following lemma . [ lemma7 ] the capacity achieving relay input distribution @xmath56 yields for the source - relay link a mutual information , which is always equal to or bigger than the mutual information of the relay - destination link . assume an arbitrary distribution @xmath72 , which yields a higher mutual information for the relay - destination link than for the source - relay link . the mutual information of the source - relay link is found by the waterfilling solution in theorem [ theo1 ] . in this case , the ` " ` min ` " ` function in ( [ con2 ] ) is limited by the mutual information of the source - relay link . then , the distribution @xmath73 with @xmath74 $ ] also satisfies the power constraint in ( [ con3 ] ) . here , @xmath75 denotes the dirac delta function . for @xmath76 , @xmath77 holds . then , by increasing @xmath78 , @xmath54 increases , whereas @xmath71 is concave in @xmath78 , see @xcite . therefore , depending on @xmath72 , @xmath71 decreases or increases until it passes its maximum and then decreases . hence , by increasing @xmath78 such that the mutual information of the source - relay link and the mutual information of the relay - destination link are equal , we create a distribution @xmath79 outperforming @xmath72 . therefore , for the optimal distribution the first argument in the `` min''-function in ( [ con2 ] ) always needs to be bigger than or equal to the second argument in the ` " ` min`"`-function . using theorem [ theo1 ] and lemma [ lemma7 ] , the optimization problem in ( [ con2 ] ) can now be stated as @xmath80 where @xmath81 is the set of distribution functions . in the following , we state conditions under which the solution of ( [ 38 ] ) is gaussian . [ theo2 ] the gaussian distribution @xmath82 solves problem ( [ 38 ] ) if @xmath83 holds , where the waterlevel @xmath11 is found from @xmath84 where @xmath85 . in this case , the capacity is given by @xmath86 which is the capacity of the relay - destination channel under an average power constraint . the gaussian distribution @xmath72 with zero mean and variance @xmath87 is known to maximize the mutual information of the relay - destination channel under a power constraint and no additional constraints . hence , if for this gaussian @xmath72 , the mutual information of the source - relay link is equal to or higher than the mutual information of the relay - destination link , i.e. , c1 in is satisfied , it is overall optimal . the condition in is found by inserting a gaussian distributed @xmath72 into c1 in . furthermore , inserting a gaussian distributed @xmath72 into c2 in yields the equation for evaluating the waterlevel in . now , we need to study the distribution @xmath88 that solves problem ( [ 38 ] ) , if condition ( [ 39 ] ) is not satisfied . in this case , the optimal distribution @xmath89 satisfies the following theorem . [ theo3 ] if for a gaussian distributed @xmath90 , the mutual information of the source - relay link is lower than the mutual information of the relay - destination link , i.e. , condition ( [ 39 ] ) does not hold , then the optimal distribution @xmath89 has the following discrete form @xmath91 where @xmath92 $ ] reflects the probability of the discrete value @xmath93 , @xmath94 has to hold and @xmath95 . please refer to the appendix for a sketch of the proof . the full proof can be found in @xcite . according to lemma [ lemma7 ] , the capacity achieving distribution @xmath56 yields for the source - relay link a mutual information , which is always higher than or equal to the mutual information of the relay - destination link . if for a gaussian @xmath90 , the mutual information of the relay - destination link is higher than the mutual information of the source - relay link , the relay needs to modify its distribution in order to increase the mutual information of the source - relay link , such that lemma [ lemma7 ] is satisfied . this always results in a discrete distribution , which is symmetric with respect to @xmath96 . moreover , @xmath97 has to hold . for the optimal relay input distribution @xmath56 , the number of discrete points can be either finite or infinite . in particular , assume the case when a gaussian @xmath56 is optimal , i.e. , holds . then , by gradually increasing the si suppression factor @xmath98 at the relay , a transition occurs from the case where the gaussian distribution is optimal to the case when the discrete distribution is optimal . thereby , for a discrete distribution close to the boundary point where a gaussian distribution starts to become optimal , we expect an infinite number of points . in this section , we numerically evaluate the capacity of the gaussian fd relay channel . in order to have a point of reference for the capacity of the gaussian fd relay channel , we first define some benchmark schemes in section [ bench ] . subsequently , we present numerical results in section [ numres ] . in the conventional fd scheme , the input distributions at both the source and the relay are gaussian with constant variance @xcite . the source is transmitting with all its available power @xmath99 , whereas the power at the relay , @xmath100 , is adapted such that the mutual informations of the source - relay link and the relay - destination link are identical . the optimal hd scheme was proposed in @xcite , but can also be directly obtained from theorem [ theo3 ] when @xmath101 . for a gaussian distribution at the relay , the mutual information of the source - relay link becomes zero if @xmath101 . hence , the optimal relay input distribution , @xmath56 , is discrete , see theorem [ theo3 ] . as for all @xmath102 the noise at the relay becomes infinite , the source is only transmitting when the relay transmits a zero symbol . hence , the optimal distribution @xmath56 includes the zero symbol , i.e. @xmath103 . when the source is transmitting , it transmits gaussian distributed symbols with zero mean and constant variance . in the conventional hd scheme , the source and the relay are transmitting gaussian distributed symbols with constant power in an alternating manner , see @xcite . the fractions of time in which source and relay are transmitting are chosen such that the mutual informations of the source - relay link and the relay - destination link are identical . to compute the capacity of the gaussian fd relay channel , we need to search for the optimal discrete distribution solving ( [ 38 ] ) . as ( [ 38 ] ) is non - convex and additionally the optimal number of discrete points , which might be infinite , is not known , the presented results for the optimal distribution are approximations and the results presented for the corresponding capacity are lower bounds . in order to approximate the capacity , we search for the optimal distribution using a fixed finite number of discrete points . moreover , we use matlab s global optimization toolbox to solve the non - convex problem in ( [ 38 ] ) . the mutual information of the relay - destination link @xmath71 in can be evaluated very efficiently using gauss - hermite quadrature , see @xcite . for @xmath104 , @xmath105 , and @xmath106,width=302,height=166 ] for @xmath107 , @xmath105 , and @xmath106,width=302,height=166 ] in order to illustrate the properties of the optimal relay input distribution , @xmath56 , the best distributions found for @xmath105 and @xmath106 are shown for @xmath104 and @xmath107 in figs . [ dis1 ] and [ dis2 ] , respectively . the blue circle markers correspond to the discrete values of distribution @xmath56 , whereas the red dashed lines show the waterfilling limits of the source . the source does not transmit if the relay transmits a symbol whose absolute value exceeds the red dashed lines . for @xmath108 , the mutual information between the source and the destination is @xmath109 . from fig . 3 we observe that the highest absolute value of @xmath55 , at which the source is still transmitting , is @xmath110 . the smallest discrete values are at @xmath111 . in fig . [ dis2 ] , the si factor is increased to @xmath107 . hence , the source - relay link is impaired by a more severe si . therefore , the relay needs to adapt its distribution in order to keep the residual si noise for the source - relay link low . hence , it shifts its values towards @xmath96 . now , the smallest discrete values of @xmath56 are at @xmath112 . furthermore , the highest absolute value , at which the source is still transmitting , is @xmath113 . the mutual information between the source and the destination for @xmath107 is reduced to @xmath114 . for comparison , the optimal input distribution for hd transmission would yield a mutual information of @xmath115 ( for both @xmath104 and @xmath107 ) . for the optimal hd protocol , the two smallest discrete values are merging and yield a point at @xmath96 , as the source can only transmit when the relay transmits a zero symbol . , as a function of the suppression factor @xmath98 for @xmath106,width=302,height=168 ] in fig . [ varyalpha ] , we show the mutual information between the source and the destination as a function of the suppression factor @xmath98 . the powers at the source and the relay are chosen as @xmath105 and the noise powers at the relay and the destination are set to @xmath106 . for @xmath116 , the mutual informations between the source and the destination for the optimal fd scheme and the conventional fd scheme are identical . with increasing @xmath98 the mutual information of the optimal fd scheme decreases and approaches for high @xmath98 the mutual information of the optimal hd scheme , whereas the mutual information of the conventional fd scheme approaches zero . the hybrid hd / fd scheme using conventional fd or hd transmission proposed in @xcite , would switch for @xmath117 from fd to hd transmission . , as a function of @xmath118 for @xmath106 , width=302,height=168 ] in fig . [ plotivaryingp ] , we show the mutual information between the source and the destination versus the powers at the source @xmath99 and the relay @xmath87 , where @xmath99 and @xmath87 are identical . the noise powers at the relay and the destination are set to @xmath106 . we compare the capacities of fd relaying with @xmath104 and hd relaying . additionally , the capacity of ideal fd relaying and the achievable rates of conventional fd relaying with @xmath104 and conventional hd relaying are shown . as expected , the highest mutual information is obtained for ideal fd relaying . the capacity of the gaussian fd two - hop relay channel with si is significantly smaller than that for ideal fd relaying . furthermore , the achievable rate for conventional fd relaying is lower than its corresponding capacity . the gap between the capacity of the gaussian fd two - hop relay channel with si and achievable rate for conventional fd relaying is not significantly increasing with increasing power . for conventional hd relaying , the smallest achievable rate is obtained . from figs . [ varyalpha ] and [ plotivaryingp ] , we observe that for small @xmath98 , the gap between the capacity of the gaussian fd two - hop relay channel and the rate achievable with conventional fd relaying is small . on the other hand , for larger values of @xmath98 the capacity of the gaussian fd two - hop relay channel approaches the capacity of the gaussian hd two - hop relay channel . hence , depending on the value of @xmath98 , using either conventional fd relaying or optimal hd relaying is nearly capacity achieving . we studied the capacity of the gaussian fd two - hop relay channel with si at the relay . for this purpose , the si was modeled as gaussian distributed with zero mean and a variance , which linearly depended on the instantaneous transmit power of the relay . the optimal input distribution at the source was shown to be gaussian with zero mean and a variance that depended on the transmit symbol power of the relay . furthermore , we revealed that the optimal input distribution at the relay was either gaussian or discrete , depending on the channel state . moreover , the capacity of the gaussian fd two - hop relay channel was compared to the capacity of the gaussian hd two - hop relay channel and the rates achievable with conventional hd and fd relaying . thereby , the gap between conventional and optimal fd relaying was shown to be small except for the case of severe si . in this appendix , we prove that the optimal distribution @xmath56 is either gaussian with variance @xmath87 or discrete and symmetric with respect to @xmath96 . the proof is based on the approaches used in @xcite and @xcite . the proof presented in this appendix has the following outline . in section [ ccc ] , we present some preliminaries . in section [ cc0 ] , we discuss the existence of an optimal relay input distribution . in section [ cc1 ] , we formulate the lagrangian function of which is maximized by the optimal relay input distribution . in section [ cc2 ] , the maximization of the lagrangian is reformulated using the concept of weak differentiability . as the lagrangian is non - convex , a necessary condition for locally optimal relay input distributions , and hence , also the globally optimal relay input distribution is found . this leads to a condition that the marginal information density , see , of the optimal relay input distribution needs to satisfy for all points in the support of the optimal relay input distribution . as the marginal information density depends on the output distribution at the destination , we investigate in section [ cc3 ] what properties the output distribution at the destination has to have in order to satisfy the condition for local optimal points of the lagrangian . this leads to a contradiction for all continuous relay input distributions except the gaussian distribution . in section [ d ] , we exclude the gaussian input distribution @xmath119 with @xmath120 , and hence show that the optimal relay input distribution can only be gaussian with variance @xmath121 or discrete . in section [ e ] , we show that the optimal relay input distribution is symmetric . at last , some useful properties of hermitian polynomials , that are necessary in section [ cc3 ] , are presented in section [ f ] . in this section , we present some preliminaries for the proof . in order to simplify our notation , we perform the optimization with respect to the cumulative distribution function ( cdf ) @xmath122 of random variable ( rv ) @xmath41 instead of the distribution @xmath72 itself . additionally , we will omit the dependency of @xmath123 on @xmath55 wherever possible and write @xmath124 . the optimal cdf , which achieves the capacity , is denoted as @xmath125 . for simplicity and without loss of generality , the values of the noise powers @xmath126 and @xmath127 are assumed to be unity in the following , i.e. , @xmath106 . the mutual information @xmath71 which depends on the cdf , @xmath128 , is denoted by @xmath129 and can be written as @xmath130 with the marginal information density @xmath131dy_2,\label{marg}\end{aligned}\ ] ] where @xmath132 denotes the output distribution @xmath133 of the considered channel , when @xmath128 is the input cdf , see @xcite , @xcite . for the considered gaussian channel , @xmath134 , where @xmath135 is assumed to be gaussian distributed with zero mean and variance one , the marginal information density is given by @xmath136 let @xmath137 be a convex space , @xmath138 a functional from @xmath137 into the real line @xmath139 , @xmath125 an element of @xmath137 , and @xmath140 a number in @xmath141 $ ] . suppose there exists a mapping @xmath142 : @xmath143 such that @xmath144 then , @xmath138 is said to be weakly differentiable in @xmath137 at @xmath125 , and @xmath142 is the weak derivative in @xmath137 at @xmath125 . if @xmath138 is weakly differentiable in @xmath137 at @xmath125 for all @xmath145 , @xmath138 is said to be weakly differentiable , see @xcite . if the weak derivative @xmath147 is less than zero , then @xmath148 is smaller than @xmath149 for small @xmath140 . hence , if @xmath150 holds for all @xmath151 , then by going from @xmath125 into an arbitrary direction @xmath151 , @xmath148 decreases for small @xmath140 and therefore @xmath125 is a local maximum point of @xmath138 . [ defpoi ] a point @xmath55 is said to be a point of increase of @xmath123 , if @xmath123 increases at @xmath55 , i.e. , @xmath152 for an arbitrarily small @xmath140 . if @xmath55 is a point of increase of @xmath123 , then @xmath55 is in the support of @xmath72 . for the following discussion , the concept of weak continuity is necessary . a definition of weak continuity can be found in ( * ? ? ? * sec . 5.10 ) . in order to show the existence of an optimal distribution , we use the following lemma . let the set @xmath156 contain all cdfs satisfying the power constraint @xmath157 . in @xcite , it is shown that the set @xmath156 is compact and that , for an awgn channel , the mutual information is weakly continuous on @xmath156 . in the considered problem ( [ 38 ] ) , the feasible set @xmath158 is a subset of @xmath156 , as we need to satisfy additional constraints besides the power constraint . the subset @xmath158 is closed , as @xmath158 includes all its boundaries . moreover , a closed subset of a compact set is compact , see ( * ? ? ? additionally , as @xmath129 is weakly continuous on @xmath156 , see @xcite , it is also weakly continuous on subset @xmath158 . therefore , we conclude that the conditions of lemma [ lemma2 ] are satisfied and hence , there exists an optimal cdf @xmath125 that maximizes ( [ 38 ] ) . in this section , we formulate the lagrangian function of in order to study the optimal cdf @xmath159 . in order to use a lagrangian function for solving the non - convex problem in , we first need the following definition . let @xmath160 , be the @xmath161 equality constraints of an optimization problem . a point @xmath125 satisfying @xmath162 is said to be a regular point of these constraints if the @xmath161 linear functionals @xmath163 , @xmath164 , ... , @xmath165 are linearly independent , where @xmath166 denotes the gradient of @xmath167 at @xmath125 . [ lemma5 ] if @xmath125 is an extremum of the functional @xmath168 subject to the constraints @xmath169 , and @xmath125 is a regular point of these constraints , then there are @xmath161 scalars @xmath170 such that the functional @xmath171 possesses a saddle point at @xmath172 . [ lemma3 ] the optimal cdf @xmath125 maximizes the lagrangian @xmath173 for some @xmath174 , @xmath175 , @xmath176 with @xmath177^+}{1+\alpha x_2 ^ 2}\right ) df(x_2)\\ z(f)&=\int_{-\infty}^{\infty}[\mu_0 - 1-\alpha x_2 ^ 2]^+df(x_2)-p_{1,\mathrm{max}}\\ g(f)&=\int_{-\infty}^{\infty}x_2 ^ 2df(x_2)-p_{2 } , \end{aligned}\ ] ] where @xmath178 $ ] and @xmath179 denotes the optimal waterlevel . if the first constraint in ( [ 38 ] ) does not influence the optimal distribution , @xmath56 is gaussian distributed , see theorem [ theo2 ] . now , we consider the case when the optimal distribution @xmath56 is affected by the first constraint in ( [ 38 ] ) . as @xmath129 is a concave function , see @xcite , we can conclude that in this case the first constraint in ( [ 38 ] ) has to hold with equality . furthermore , we change the power constraint in ( [ 38 ] ) to @xmath180 with @xmath178 $ ] . hence , by proving that for an arbitrary @xmath100 only a discrete or a gaussian distribution can solve the new optimization problem , we can conclude that also for the optimal power @xmath100 , @xmath72 can only be discrete or gaussian . we assume the waterlevel @xmath11 in c1 and c2 of to be known and denote it by @xmath179 , as we do not want to find the optimal cdf @xmath125 itself , but want to study only the properties of @xmath125 . then , the second constraint in ( [ 38 ] ) does not determine the waterlevel @xmath11 anymore , but assures that all feasible distributions , including the optimal cdf , yield the same waterlevel @xmath179 . hence , we need to solve a modified optimization problem , which can be stated as @xmath181^+}{1+\alpha x_2 ^ 2}\right ) df(x_2)= i(f)\notag\\ & \int_{-\infty}^{\infty}[\mu_0 - 1-\alpha x_2 ^ 2]^+df(x_2)= p_{1,\mathrm{max}}\notag\\ & \int_{-\infty}^{\infty}x_2 ^ 2df(x_2)= p_{2}.\label{c2 } \end{aligned}\ ] ] the optimization problem in ( [ c2 ] ) is non - convex due to the first constraint . in order to use the lagrangian method for solving ( [ c2 ] ) we use lemma [ lemma5 ] . for that purpose , we need to ensure that the optimal cdf , @xmath125 , is a regular point of the optimization problem in ( [ c2 ] ) . according to ( * section 5.1 ) , if the optimal solution @xmath125 is not a regular point , i.e. , at least one gradient is a linear combination of the others , then we can remove constraints without affecting the optimal solution @xmath125 . however , as we assume in this appendix that the optimal solution @xmath125 is also influenced by the source - relay link , removing any constraint in ( [ c2 ] ) would affect the solution and therefore , the optimal cdf , @xmath125 , solving ( [ c2 ] ) maximizes the lagrangian in ( [ c3 ] ) . in this section , we maximize the lagrangian function presented in lemma [ lemma3 ] . this will yield an optimality condition for the marginal information density @xmath182 . the following lemma yields a necessary condition that @xmath125 has to satisfy in order to maximize the lagrangian in ( [ c3 ] ) . [ lemma4 ] the optimal cdf , @xmath125 , needs to satisfy @xmath183^+}{1+\alpha x_2 ^ 2}\right)\notag\\ & + \lambda_2[\mu_0 - 1-\alpha x_2 ^ 2]^++\lambda_3x_2 ^ 2\bigg]df(x_2)\notag\\ & \le i(f_0)+\lambda_2p_{1,\mathrm{max}}+\lambda_3p_{2 } , \quad \forall f(x_2),\label{c6 } \end{aligned}\ ] ] for some @xmath184 . the set of cdfs @xmath137 is convex , as the function @xmath185 with @xmath186 satisfies @xmath187 . hence , we can use the weak derivative in ( @xmath188 ) in order to maximize the lagrangian in ( [ c3 ] ) . the weak derivative of @xmath189 needs to satisfy @xmath190 as @xmath191 represents the condition for the cdf @xmath125 to be a local maximum of @xmath192 , see lemma [ lemma1 ] . we note that condition ( [ c4 ] ) is necessary but not sufficient for @xmath125 as @xmath191 is satisfied by all local optimal cdfs , whereas we search for the global optimal cdf . however , by proving that all distributions satisfying ( [ c4 ] ) can only be discrete or gaussian , we also prove that the optimal distribution has to be discrete or gaussian . to this end , the weak derivatives of @xmath129 , @xmath193 , @xmath194 , and @xmath195 have to be computed . the weak derivative of @xmath129 can be found for a channel with a power constrained input as @xcite , @xcite @xmath196 as @xmath193 , @xmath194 , and @xmath195 are affine , their weak derivatives are given by @xmath197 inserting the results for the weak derivatives and the constraints in ( [ c2 ] ) into ( [ c4 ] ) yields @xmath198^+}{1+\alpha x_2 ^ 2}\right)df(x_2 ) -i(f_0)\right)\notag\\ & + \lambda_2\left(\int_{-\infty}^{\infty}[\mu_0 - 1-\alpha x_2 ^ 2]^+df(x_2)-p_{1,\mathrm{max}}\right)\notag\\ & + \lambda_3\left(\int_{-\infty}^{\infty}x_2 ^ 2df(x_2)-p_2\right)\le 0.\label{c21 } \end{aligned}\ ] ] [ lemma6 ] the inequality in ( [ c6 ] ) is satisfied if and only if @xmath199^+}{1+\alpha x_2 ^ 2}\right)-\lambda_2\left[\mu_0 - 1-\alpha x_2 ^ 2\right]^+\notag\\ & -\lambda_3x_2 ^ 2 , \quad\forall x_2,\label{c7 } \end{aligned}\ ] ] holds . furthermore , let @xmath200 be the points of increase , see definition [ defpoi ] , of @xmath159 . then @xmath201^+}{1+\alpha x_2 ^ 2}\right)-\lambda_2\left[\mu_0 - 1-\alpha x_2 ^ 2\right]^+\notag\\ & -\lambda_3x_2 ^ 2 , \quad\forall x_2\in \mathcal{e}_0,\label{c8 } \end{aligned}\ ] ] holds . sufficiency of ( [ c7 ] ) for ( [ c6 ] ) is easily shown by inserting ( [ c7 ] ) into ( [ c6 ] ) . necessity is shown by contradiction . in particular , assume there exists a @xmath202 not satisfying ( [ c7 ] ) and hence satisfying @xmath203^+}{1+\alpha \tilde{x}_2 ^ 2}\right)-\lambda_2\left[\mu_0 - 1-\alpha \tilde{x}_2 ^ 2\right]^+\notag\\ & -\lambda_3\tilde{x}_2 ^ 2 . \end{aligned}\ ] ] then , by choosing @xmath123 to be a unit step function at @xmath202 , we obtain @xmath183^+}{1+\alpha x_2 ^ 2}\right)\notag\\ & + \lambda_2[\mu_0 - 1-\alpha x_2 ^ 2]^++\lambda_3x_2 ^ 2\bigg]df(x_2)\notag\\ & > i(f_0)+\lambda_2p_{1,\mathrm{max}}+\lambda_3p_{2 } , \end{aligned}\ ] ] which contradicts ( [ c6 ] ) . hence , ( [ c7 ] ) has to hold . now , the validity of ( [ c8 ] ) is shown . assume there exist some @xmath204 that do nt satisfy ( [ c8 ] ) but due to ( [ c7 ] ) satisfy @xmath205^+}{1+\alpha \tilde{x}_2 ^ 2}\right)-\lambda_2\left[\mu_0 - 1-\alpha \tilde{x}_2 ^ 2\right]^+\notag\\ & -\lambda_3\tilde{x}_2 ^ 2 .\label{c5 } \end{aligned}\ ] ] now , all points @xmath202 satisfying ( [ c5 ] ) are collected in the set @xmath206 , which is a subset of @xmath200 . hence , we obtain @xmath207^+\notag\\ -&\lambda_1\frac{1}{2 } \mathrm{ld}\left(1+\frac{\left[\mu_0 - 1-\alpha x_2 ^ 2\right]^+}{1+\alpha x_2 ^ 2}\right)-\lambda_3x_2 ^ 2\bigg ] df_0(x_2)\notag\\ + & \int_{\mathcal{e}_0-\mathcal{e}'}^{}\bigg[i(f_0)+\lambda_2p_{1,\mathrm{max}}+\lambda_3p_{2}-\lambda_2\left[\mu_0 - 1-\alpha x_2 ^ 2\right]^+\notag\\ -&\lambda_1\frac{1}{2 } \mathrm{ld}\left(1+\frac{\left[\mu_0 - 1-\alpha x_2 ^ 2\right]^+}{1+\alpha x_2 ^ 2}\right)-\lambda_3x_2 ^ 2\bigg ] df_0(x_2)\notag\\ \overset{(c)}{=}&\int_{-\infty}^{\infty}\bigg[i(f_0)+\lambda_2p_{1,\mathrm{max}}+\lambda_3p_{2}-\lambda_2\left[\mu_0 - 1-\alpha x_2 ^ 2\right]^+\notag\\ -&\lambda_1\frac{1}{2 } \mathrm{ld}\left(1+\frac{\left[\mu_0 - 1-\alpha x_2 ^ 2\right]^+}{1+\alpha x_2 ^ 2}\right)-\lambda_3x_2 ^ 2\bigg ] df_0(x_2 ) \notag\\ \overset{(d)}{=}&(1-\lambda_1)i(f_0),\label{c9 } \end{aligned}\ ] ] where @xmath208 and @xmath209 hold as @xmath210 , to obtain @xmath211 we insert ( [ c8 ] ) and ( [ c5 ] ) , and to obtain @xmath212 we take into account the constraints in ( [ c2 ] ) . as ( [ c9 ] ) is a contradiction , ( [ c8 ] ) has to be true . now , we investigate the implication that ( [ c8 ] ) has on the optimal distribution @xmath56 . the only pdfs , @xmath72 , that can satisfy the optimality condition in ( [ c8 ] ) are the gaussian distribution @xmath213 with @xmath178 $ ] and a discrete distribution of the form @xmath214 where @xmath92 $ ] denotes the probability of the discrete value @xmath215 , @xmath216 has to hold , @xmath95 , and @xmath75 denotes the dirac delta function . for investigating whether a continuous @xmath56 can satisfy ( [ c8 ] ) , we need to modify ( [ c8 ] ) such that both the left hand side and the right hand side of are analytic in @xmath55 . an analytic function is completely defined by its series expansion in one point , see @xcite . @xmath182 is analytic in @xmath55 , see ( [ c12 ] ) . the only non - analytic function in ( [ c8 ] ) is @xmath217^+$ ] . therefore , we split up condition ( [ c8 ] ) such that @xmath55 either satisfies @xmath218^+=0 $ ] or @xmath218^+=\mu_0 - 1-\alpha x_2 ^ 2 $ ] . we define @xmath219 to be the set of points @xmath55 satisfying @xmath220 . thereby , for the optimal cdf , @xmath125 , condition ( [ c8 ] ) can be stated as @xmath221 where @xmath222 and @xmath223 now , assume the optimal distribution @xmath56 is continuous for some @xmath55 . then , @xmath159 has a connected set @xmath224 of points of increase and therefore , depending on the location of @xmath224 , @xmath225 or @xmath226 for all points @xmath55 in @xmath224 . furthermore , both @xmath227 and @xmath228 are analytic in @xmath55 . thus , if we know that @xmath229 for all @xmath230 , we can conclude that the taylor series expansion in one point in the interior of @xmath224 is equal to zero and hence @xmath229 holds for all @xmath55 , see @xcite , @xcite . therefore , if the optimal distribution @xmath56 is continuous for any @xmath231 , @xmath225 holds for all @xmath55 . on the other hand , if the optimal distribution @xmath56 is continuous for any @xmath232 or @xmath233 , @xmath226 holds for all @xmath55 . the equations for @xmath227 and @xmath228 in and include the marginal information density @xmath234 . as the marginal information @xmath234 in involves the output of the optimal output distribution @xmath235 , we can check what output distribution @xmath235 satisfy @xmath236 , or @xmath237 . then , by examining whether a given output distribution @xmath235 satisfying @xmath238 , can be the output distribution of the relay - destination awgn channel , we can examine whether a continuous input distribution @xmath72 can be optimal . we have to distinguish between three cases of possible non - discrete distribution @xmath56 . @xmath56 can be continuous somewhere in @xmath219 , @xmath239 , or both . this yields cases a , b , c , and d in table [ t1 ] . by excluding case a and b and showing that the only possibility for case c is a gaussian @xmath56 , we will show that the optimal input distribution can only be gaussian or discrete . we start with case b , as it is the easiest to investigate . for this purpose , a fourier - hermite series decomposition is necessary , see appendix [ f ] for some properties of the fourier - hermite series decomposition . first , we decompose the integral in ( [ c12 ] ) using hermitian polynomials . to this end , we define @xmath240 where @xmath241 are the hermitian polynomials , see . note that @xmath242 is square integrable with respect to @xmath243 and hence can be decomposed using a fourier - hermite series decomposition , see @xcite . then , the integral in ( [ c12 ] ) can be written as @xmath244 where for @xmath208 , ( [ c13 ] ) and ( [ d2 ] ) are inserted and for @xmath211 , ( [ d1 ] ) is used . by inserting ( [ c14 ] ) into ( [ c12 ] ) and ( [ c12 ] ) into ( [ c11 ] ) , we obtain the hermitian coefficients for @xmath226 as @xmath245 comparing the exponents in yields @xmath246 inserting the hermitian coefficients into ( [ c13 ] ) , the output distribution is obtained as @xmath247 this solution can be a valid probability density function for @xmath248 , which yields a gaussian distribution . hence , we can conclude that if the optimal distribution @xmath56 is continuous somewhere in @xmath249 , the output needs to follow a gaussian distribution . therefore , also the input distribution @xmath56 has to be gaussian . this implies that @xmath56 is also continuous in @xmath219 and hence , we can exclude case b in table [ t1 ] . now , we evaluate what @xmath250 , implies for @xmath235 and exclude case a this way . for this purpose , we insert ( [ c14 ] ) into ( [ c12 ] ) and ( [ c12 ] ) into ( [ c10 ] ) , which yields for @xmath225 @xmath251 with @xmath252 inserting the hermitian coefficients into ( [ c13 ] ) , the output distribution is given by @xmath255 where @xmath256 are unknown constants . next , by showing that an input distribution @xmath72 , which is only continuous in @xmath219 , can not create the distribution @xmath235 in as the output of an awgn channel , we will exclude case a in table [ t1 ] . for that purpose , we need to distinguish the following three possibilities for @xmath256 . the first possibility is that there exists at least one @xmath257 with @xmath258 . the second possibility is that @xmath259 and @xmath260 hold . the last possibility is that @xmath261 holds for all @xmath262 . in order to exclude case a in table [ t1 ] , we eliminate all three possibilities for @xmath256 for an input distribution @xmath72 , that is only continuous in @xmath231 . first , we exclude the first possibility for @xmath256 , i.e. , there exists at least one @xmath257 with @xmath258 . as the optimal input distribution can only be continuous for all @xmath263 , the output pdf @xmath133 can be expressed as a function of @xmath72 as @xmath264 where @xmath265 and @xmath215 denote the weights and positions of the discrete values of @xmath72 in @xmath249 . the continuous part of @xmath72 yields a function which decays proportional to @xmath266 for high @xmath267 . the discrete part of @xmath72 just adds gaussian functions in . on the other hand , if there exist @xmath257 with @xmath258 then for high @xmath267 , @xmath235 is dominated by the corresponding factor @xmath268 , which can not be created by a distribution satisfying ( [ c16 ] ) . hence , we can exclude the first possibility for @xmath256 in case a. if the second possibility for @xmath256 holds , i.e. , @xmath269 and @xmath270 , @xmath235 is gaussian distributed , which implies that @xmath56 is gaussian distributed and hence does not satisfy case a. the last possibility for @xmath256 is that @xmath271 . then , @xmath235 is constant , which is not valid . therefore , we can exclude case a. the remaining non - discrete case c requires both @xmath225 and @xmath226 . as @xmath226 requires a gaussian distributed @xmath235 , see , @xmath225 has to hold for a gaussian distributed @xmath235 , too . this implies that @xmath270 , which can be achieved when @xmath272 , and hence @xmath273 . as @xmath174 is the lagrangian multiplier for the first condition in ( [ c2 ] ) , we conclude that for the case @xmath274 , the constraint for the mutual information of the source - relay link is non - active , and hence obviously a gaussian distribution maximizes @xmath71 . another possibility for @xmath272 and hence @xmath273 is @xmath116 . then , no si exists and we need to maximize the mutual information of two independent awgn channels under power constraints . in this case , the capacities are obviously achieved by gaussian input distributions . in the remaining case d , the optimal distribution @xmath56 is discrete . hence , we have shown that @xmath56 can only be discrete or gaussian distributed . in order to use the lagrangian function in lemma [ lemma5 ] , we chose the power constraint to hold with equality . hence , so far we have only shown that for every power @xmath275 $ ] the optimal distribution can only be gaussian or discrete . to continue the search for the optimal distribution , we show in section [ d ] that a gaussian distribution can only be optimal when its variance is @xmath87 . in this section , we prove that a gaussian distribution can only be optimal if its variance is @xmath87 , and hence if @xmath90 is not optimal the optimal distribution @xmath56 has to be discrete , as shown in appendices [ ccc ] - [ cc3 ] . assume a gaussian distributed @xmath119 with @xmath276 . the source only transmits when the relay transmits a symbol inside the limits of the waterfilling solution @xmath277 , otherwise the source is silent . hence , modifying the distribution @xmath72 outside these limits , does not change @xmath54 . as @xmath278 holds , the power @xmath279 can be used to increase @xmath280 . by decreasing the probability of the values next to the integration limits and increasing the probability of values farther from the integration limits , an alternative distribution @xmath281 of rv @xmath282 is created as @xmath283 with @xmath284 @xmath285 is chosen to satisfy @xmath286 hence @xmath287 is assured . the gaussian distribution @xmath72 , the waterfilling limits , and the alternative distribution @xmath281 are shown schematically in fig . [ gaus ] . assume the distribution function of a rv @xmath288 is limited to a finite interval . then , its differential entropy is maximized by a uniform distribution , see @xcite . therefore , @xmath289 holds . next , we show that @xmath290 holds , where @xmath291 . the entropy power inequality provides a lower bound on the differential entropy of the sum of two rvs , see @xcite . assume two random variables @xmath78 and @xmath292 with arbitrary distributions and differential entropies @xmath293 and @xmath294 , respectively . if @xmath295 and @xmath296 are normal distributed rvs with @xmath297 and @xmath298 , then @xmath299 holds . now assume a gaussian distributed rv @xmath300 with @xmath301 . hence , @xmath302 holds . then the inequality @xmath303 can be proved as @xmath304 where ( a ) follows from the entropy power inequality and ( b ) holds as @xmath305 and @xmath306 are both gaussian distributed , where @xmath307 . therefore , with @xmath276 it is possible to modify a gaussian distributed @xmath72 such that @xmath308 remains constant and @xmath71 increases . now assume a gaussian distribution @xmath309 with @xmath310 , with an arbitrarily small @xmath311 . then , @xmath308 increases and @xmath71 decreases , while the power @xmath312 is available at the relay to increase @xmath71 . as shown above , @xmath71 can be increased using @xmath313 without decreasing @xmath308 . therefore , using a non - gaussian distributed rv @xmath41 can increase both @xmath71 and @xmath308 and hence @xmath314 , when condition ( [ 39 ] ) is not satisfied . hence , a gaussian distributed @xmath119 with @xmath276 can not be optimal . however , as shown above , if the optimal distributed rv @xmath41 is not gaussian distributed , it has a discrete distribution . therefore , if the optimal distribution @xmath56 is not gaussian with variance @xmath87 , it is discrete . in this section , we prove that the optimal discrete distribution is symmetric with respect to @xmath96 . for that purpose , we show that for all non - symmetric distributions a corresponding outperforming symmetric distribution exists . assume an arbitrary distribution @xmath315 where @xmath316 $ ] denotes the probability of the discrete value @xmath317 , @xmath318 has to hold , and @xmath319 . defining the functions @xmath320 and @xmath321 as @xmath322 a new distribution @xmath281 satisfying @xmath323 , @xmath324 , can be constructed as @xmath325 where @xmath326 $ ] denotes the probability of the discrete value @xmath327 , @xmath328 holds , and @xmath95 . defining a third distribution @xmath329 of rv @xmath300 as @xmath330 with @xmath331 $ ] allows to vary the probability of the sign of a symbol @xmath332 via varying @xmath333 . as the probability of every possible value @xmath334 is not changed , the power constraint @xmath335 is still satisfied and the maximum mutual information of the source - relay link obtained by waterfilling , see theorem [ theo1 ] , remains constant . hence , by varying @xmath333 , only @xmath336 is modified , where the value of the rv @xmath337 is obtained as @xmath338 . now , by proving that @xmath339 , maximizes @xmath336 , we show that for every non - symmetric distribution , an equivalent symmetric distribution exists , which outperforms the non - symmetric distribution . the mutual information @xmath340 is concave in the input distribution @xmath88 , see @xcite . as @xmath329 is linearly dependent on @xmath341 , @xmath336 is concave in @xmath341 . hence , @xmath342 , is a sufficient condition for @xmath336 to achieve its maximum with respect to @xmath341 . with @xmath343 only @xmath344 has to be maximized with respect to @xmath341 . with @xmath338 , the output distribution @xmath345 is given by @xmath346 the derivative of @xmath344 with respect to @xmath333 is evaluated as @xmath347dy_2\notag\\ \overset{(a)}{=}&-\int_{-\infty}^{\infty}\left(c_n\frac{1}{\sqrt{2\pi}\sigma_2}e^{-\frac{(y_2-d_n)^2}{2\sigma_2 ^ 2}}-c_n\frac{1}{\sqrt{2\pi}\sigma_2}e^{-\frac{(y_2+d_n)^2}{2\sigma_2 ^ 2}}\right)\notag\\ \cdot&\mathrm{ld}(p_{y_2}^{**}(y_2))dy_2\label{f2}\end{aligned}\ ] ] where for @xmath208 we exploit that the equation in the third line of ( [ f2 ] ) is point - symmetrical with respect to @xmath348 , and hence its integral vanishes . now , assume @xmath349 , is inserted into ( [ f2 ] ) and therefore also into ( [ f1 ] ) . then , @xmath345 and hence @xmath350 are axially - symmetric with respect to @xmath348 . therefore , the integrand in ( [ f2 ] ) is the multiplication of a point - symmetric and axially - symmetric function with respect to @xmath348 , which yields a point - symmetric function with respect to @xmath348 . as the integral over this point - symmetrical function in ( [ f2 ] ) yields 0 , @xmath351 , holds when @xmath352 . hence , for all non - symmetric distributions @xmath72 a corresponding symmetric distribution @xmath329 exists , which outperforms @xmath72 . therefore , the optimal distribution @xmath56 has to be symmetric with respect to @xmath96 . in this appendix , some useful properties of hermitian polynomials are presented , see @xcite . the hermitian polynomials are defined as @xmath353 the first few polynomials @xmath354 are given by @xmath355 hermitian polynomials are orthogonal with respect to the weight function @xmath356 , i.e. , @xmath357 hermitian polynomials form a complete orthogonal system on the hilbert space @xmath358 of real functions on the real line that are square integrable with respect to @xmath359 , i.e. , @xmath360 hence , any function satisfying ( [ d3 ] ) may be uniquely decomposed using the fourier - hermite series decomposition . t. riihonen , s. werner , r. wichman , and z. b. eduardo , on the feasibility of full - duplex relaying in the presence of loop interference , " _ ieee 10th workshop on signal processing advances in wireless communications ( spawc ) _ , 2009 . b. p. day , a. r. margetts , d. w. bliss , and p. schniter , full - duplex mimo relaying : achievable rates under limited dynamic range , " _ ieee journal on selected areas in communication _ , vol . 8 , september 2012 .

we study a combined sample of 264 star - forming , 51 composite , and 73 active galaxies using optical spectra from sdss and mid - infrared ( mid - ir ) spectra from the spitzer infrared spectrograph . we examine optical and mid - ir spectroscopic diagnostics that probe the amount of star formation and relative energetic contributions from star formation and an active galactic nucleus ( agn ) . overall we find good agreement between optical and mid - ir diagnostics . misclassifications of galaxies based on the sdss spectra are rare despite the presence of dust obscuration . the luminosity of the [ neii ] 12.8 emission - line is well correlated with the star formation rate ( sfr ) measured from the sdss spectra , and this holds for the star forming , composite , and agn - dominated systems . agn show a clear excess of [ neiii ] 15.6 emission relative to star forming and composite systems . we find good qualitative agreement between various parameters that probe the relative contributions of the agn and star formation , including : the mid - ir spectral slope , the ratio of the [ nev ] 14.3 to [ neii ] 12.8 fluxes , the equivalent widths of the 7.7 , 11.3 , and 17 pah features , and the optical d " parameter which measures the distance a source lies from the locus of star forming galaxies in the optical bpt emission - line diagnostic diagram . we also consider the behavior of the three individual pah features by examining how their flux ratios depend upon the degree of agn - dominance . we find that the pah 11.3 feature is significantly suppressed in the most agn - dominated systems .

1206.5830

it has been established that a strong link exists between supermassive black holes ( smbhs ) and the galaxies in which they live @xcite and between accreting smbhs ( or active galactic nuclei , agn ) and star formation in particular ( e.g. * ? ? ? * ; * ? ? ? clues to this link can be revealed by detailed analysis of the spectra of galaxies and agn . star formation processes imprint their signatures on the spectrum . stellar populations of various ages produce strong continuum emission in the ultraviolet ( uv ) through near - ir region , while gas photo - ionized by the hot young stars produces prominent emission - lines in the uv , optical , and ir . the accretion disk surrounding the supermassive black hole in an agn produces its own strong uv and optical continuum emission ( the big blue bump @xcite ) . this emission can swamp that of the stellar population of the host galaxy when there is a direct view of the accretion disk ( i.e. in type 1 agn ) . dust in the interstellar medium of the host galaxy will absorb optical and uv light from both stars and the agn and re - radiate this in the ir . in agn , this reprocessing also occurs in the dusty obscuring torus @xcite . in type 2 agn , the direct view of the agn is blocked by the obscuring torus , and the presence of the agn can be inferred by prominent optical and infrared emission lines from highly ionized gas located beyond the torus and the strong mid - ir continuum emission from the torus . in such objects the signatures of host galaxys young stars and star formation are also present in optical and mid - ir spectra . these type 2 agn are therefore ideal laboratories for studying the connection between agn and star formation , provided that the tracers of this activity can be disentangled . multi - wavelength star formation indicators have been extensively studied in samples of quiescent star forming galaxies ( see * ? ? ? * for a review ) . star formation rates ( sfrs ) have been calibrated in the optical based on h@xmath1 emission , which results from recombination following ionization due to energetic photons from massive o and b stars , and in the uv based on the continuum which provides a window into emission from young stars . dust in the galaxy reprocesses optical and uv photons and re - radiates in the infrared ( ir ) , thus revealing obscured star formation . other ir indicators include emission from fine structure lines excited by starburst activity , such as [ neii ] 12.81@xmath2 m and [ neiii ] 15.56@xmath2 m @xcite , as well as spectral emission features associated with polycyclic aromatic hydrocarbons ( pahs ) , since these macro - molecules / small - grains can be ionized by single optical or uv photons @xcite and are prominent in ir spectra of star - forming galaxies @xcite . such host galaxy star formation diagnostics can be contaminated by the presence of an agn . photons emitted by the accretion disk enhance the optical and uv emission used to derive sfrs . the obscuring medium enshrouding the central engine contributes significantly to the infrared emission , constituting approximately 20% of the agn bolometric luminosity @xcite . disentangling the contribution from the agn vs. star formation then becomes a necessary intermediate step in using continuum ir emission as a tracer of star formation in active galaxies , and such a process can be plagued with uncertainties . previous studies have indicated that emission from pahs tend to be suppressed in galaxies hosting an active nucleus , which is sometimes attributed to the harsh radiation field produced by the agn accretion disk destroying pahs ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? a complementary view of the agn and starburst connection is to study spectroscopic signatures that parameterize the relative importance of these two processes . the optical bpt diagram ( a plot of [ oiii]@xmath35007/h@xmath4 vs. [ nii]@xmath36583/h@xmath1 ) provides a useful diagnostic to differentiate between star - forming galaxies , composite systems ( galaxies with comparable amounts of star - formation and agn activity ) and type 2 , or obscured , agn in the local universe @xcite . the optical `` d '' parameter is the distance a source lies from the locus of star forming galaxies on the bpt diagram : a higher d value indicates greater agn dominance @xcite . ratios of ir fine structure lines also parameterize the ionization field hardness . for instance , [ oiv]26@xmath2 m and [ nev]14.32@xmath2 m are primarily ionized by agn @xcite whereas [ neii]12.81@xmath2 m is excited by star formation @xcite . the ratio of these lines can then indicate the relative importance of these two processes @xcite . the mid - infrared ( mir ) spectral slope , @xmath5 , steepens as the amount of emission from cold dust heated by stars increases relative to the amount of emission from hotter dust heated by the agn . it is thus another potentially useful tool to assess the relative amount of star formation in active systems ( e.g. * ? ? ? * ; * ? ? ? the equivalent width ( ew ) of pah grains is another diagnostic to probe the relative amount of star formation to agn activity . an anti - correlation between ionization field hardness and the ratio of pah features at 7.7@xmath2 m and 11.3@xmath2 m has been reported and is interpreted as due to an increasing contribution to the mid - ir continuum emission by agn - heated dust . the selective destruction of smaller pah grains may also play a role ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , and so the flux ratios of the different pah features can then also potentially trace the relative contribution of agn and star formation in active galaxies . in this paper , we combine samples of star - forming galaxies , obscured agn ( seyfert 2 galaxies , or sy2s ) and composite systems to study the interplay between agn activity and star formation . using quiescent star forming galaxies as a base - line , we investigate the effects of agn activity on the following star formation diagnostics : optically derived sfrs from the sloan digital sky survey @xcite , the luminosities of the ir fine structure lines [ neii ] 12.81@xmath2 m and [ neiii ] 15.56@xmath2 m @xcite and the luminosities of the polycyclic aromatic hydrocarbons ( pahs ) at 7.7@xmath2 m , 11.3@xmath2 m and 17@xmath2 m @xcite . which of these star formation proxies agree the best among star - forming galaxies and agn and are thus least affected by the presence of an agn ? we expand upon the work of @xcite to test the agreement among diagnostics that parameterize the relative contributions of agn activity and star formation , including the equivalent width ( ew ) of pahs at 7.7 , 11.3 and 17@xmath2 m , ratios of pah fluxes @xcite , ratios of ir fine structure lines @xcite , mid - ir spectral slope ( @xmath5 * ? ? ? * ; * ? ? ? * ) and the optical d parameter @xcite . in @xcite we found that sy2s with stronger pah emission tend to have a softer ionization field and that the mir spectral slope was well correlated with pah ew . by expanding this parameter space into the regime of quiescent star forming galaxies , we test if pah features and the mid - ir spectral slope are dependent on the hardness of the radiation field , regardless of its source , or if ionization by an agn is necessary to appreciably affected observed features . finally , we use the results of this analysis to empirically decompose the mid - ir ( mir ) emission into a star - forming and agn component . to assess the efficacy of star formation diagnostics in active galaxies , we have combined samples of agn , quiescent star forming galaxies , and composite objects . sources were categorized by their bpt designations @xcite using the boundaries from @xcite and @xcite : galaxies above the @xcite delineation are classified as pure seyfert 2s , sources in between the @xcite and @xcite demarcations are classified as composites and the sources below the @xcite boundary are classified as quiescent star - forming galaxies . as discussed in @xcite , we selected an optical sample of sy2s from the sloan digital sky survey ( sdss ) . these sources were identified as pure agn on the basis of their [ oiii]@xmath35007/h@xmath4 and [ nii]@xmath36583/h@xmath1 flux ratios which place them above the theoretical starburst demarcation from @xcite on the bpt diagram . this sample , complete to an [ oiii ] flux limit of @xmath6 erg / s/@xmath7 ( hereafter `` [ oiii ] sample '' ) , comprises 20 sy2s . we complement these sy2s with a mid - infrared ( mir ) selected sample , the original 12@xmath2 m sample @xcite from the _ iras _ point source catalog , which contains 31 galaxies . using published or sdss measured values of [ oiii]@xmath35007/h@xmath4 and [ nii]@xmath36583/h@xmath1 , we have classified 27 12@xmath2 m sources as sy2s , 2 as composites ( f08572 + 3915 and ngc 7590 ) and 2 were dropped due to lack of published [ nii]/h@xmath1 values ( f04385 - 0828 and ngc 3982 ) . as both [ oiii ] and mir fluxes are indicators of intrinsic agn power ( e.g * ? ? * ; * ? ? ? * ; * ? ? ? * ) , these sy2 samples should to first order represent unbiased views of the agn population . the [ oiii ] sample has optical spectroscopy from sdss whereas the optical data for the majority of the 12@xmath2 m sample were based on literature values . both samples have low and high resolution _ spitzer _ infrared spectrograph ( irs * ? ? ? * ) data . our comparison samples of predominantly normal galaxies have sdss spectroscopy , _ spitzer _ irs data and ultraviolet photometry from the galaxy evolution explorer ( _ galex _ ) . the _ spitzer_-sdss-_galex _ spectroscopic survey ( ssgss , p.i . schiminovich ) contains 101 local ( 0.03 @xmath8 0.21 ) galaxies overlapping the lockman hole region @xcite . this sample is augmented by the _ sdss spectroscopic survey ( s5 , p.i . schiminovich ) , increasing the sample size of local ( in this case , 0.05 @xmath9 0.1 ) star - forming galaxies to 300 ( bertincourt et al . , in prep ) . both the ssgss and s5 samples have irs short - low ( sl ) and long - low ( ll ) coverage . the s5 sample also has irs short - high ( sh ) for all sources , but only 1/3 of the ssgss sources ( a sub - sample consisting of the brightest galaxies ) have high resolution infrared spectra . based on the bpt designations , there are 26 sy2s , 49 composites and 264 star - forming galaxies in the ssgss and s5 surveys . combining all four samples , we garner 264 star - forming galaxies , 51 composites , and 73 agn for this analysis . throughout this paper , we have adopted a cosmology of @xmath10 = 70 km / s / mpc , @xmath11 = 0.27 , and @xmath12 = 0.73 . optical spectra for the [ oiii ] , ssgss and s5 samples were drawn from sdss , using the mpa / jhu galaxy value - added catalog . this catalog includes ancillary derived data , such as star formation rates ( sfrs , * ? ? ? * ) , which we utilize in the subsequent analysis . only four of the 12@xmath2 m selected sources have sdss spectra ( 3 sy2s and 1 composite ) . optical data for the remainder of these sources were culled from the literature , as noted in @xcite . hence , these 24 sy2s and 1 composite are omitted from the sdss sfr analysis in section 4 . we emphasize here that all sdss - derived quantities are based on the sdss spectra alone . the details of the _ spitzer _ data reduction for the [ oiii ] and 12@xmath2 m samples were presented in @xcite , the ssgss sample in @xcite and the s5 sample in bertincourt et al . , in prep . , which follows the same procedures as @xcite . in short , the irs data were run through the standard _ spitzer _ calibration software . sky subtraction was performed for all low - resolution data and for the high resolution sources that had dedicated off source observations . all but 68 observations , from the ssgss sample , have high resolution data , leaving 223 star - forming galaxies , 33 composites and 64 sy2s with high resolution irs coverage . only eight sources , from the 12@xmath2 sample , did not have dedicated high resolution background observations , but since we utilize these data to measure emission line fluxes , the addition of the background to the underlying continuum has negligible effect on our derived results ( see * ? ? ? * ) . spectra were extracted with spice for the [ oiii ] , ssgss and s5 sources and with smart @xcite for the staring mode 12@xmath2 m observations . cubism @xcite was used to extract spectra from 12@xmath2 m spectral mapping observations . emission line fluxes used in this analysis ( [ neii ] 12.81@xmath2 m , [ nev ] 14.32@xmath2 m and [ neiii ] 15.56@xmath2 m ) were measured from the high resolution spectra , specifically the short - high ( sh ) module . the local continuum around the emission feature was fit with a local zero or first order polynomial . this local continuum fit was frozen and the emission feature was then fit with a gaussian . the error was estimated by smoothing the local continuum by the line width ( or the nominal minimum resolvable line width / r , where @xmath13 is the nominal wavelength of the emission line and r is the resolution of the sh module , @xmath14600 ] in the case of non - detections ) and calculating the rms around this smoothed continuum . in order for a line to be considered detected , it had to be significant at the 5@xmath15 level or above , have a line width at least on the order of the sh resolution ( i.e. full width half max @xmath16 0.01 @xmath2 m ) and have a line centroid near the nominal wavelength of the emission line ( i.e. @xmath17 0.03@xmath2 m ) . the [ neii ] and [ neiii ] fluxes and upper limits for the ssgss , s5 , [ oiii ] and 12@xmath2 m samples are listed in tables [ ssgss_flux ] - [ 12m_flux ] . since we consider [ nev ] upper limits in aggregate rather than individually ( see section 7.1 ) , we list only the [ nev ] detections in these tables . we measured the fluxes and equivalent widths ( ews ) of the polycyclic aromatic hydrocarbons ( pah ) at 7.7@xmath2 m , 11.3@xmath2 m and 17@xmath2 m with pahfit , a spectral decomposition routine that models dust continuum , starlight continuum , unresolved emission lines , pah emission features and silicate absorption / emission @xcite . pahfit requires a single stitched low - resolution spectrum . the long - low ( ll ) spectra was therefore shifted to match the short - low ( sl ) spectra in their overlapping regions , with typical adjustments @xmath1420% or lower ( see * ? ? ? * ; * ? ? ? this choice of normalization was made since the sl and sh apertures sizes are well - matched , better enabling us to compare quantities derived from the high and low resolution spectra . the routine was run with a mixed extinction model on all sources . we note that in @xcite , pahfit was run with a screen extinction , which assumes a foreground absorbing slab rather than dust mixed with gas , starlight and grains . we have re - run pahfit on the [ oiii ] and 12@xmath2 m sources using the default mixed extinction option to be consistent with the ssgss and s5 samples , though differences in results were slight . for these two sy2 samples , we also added a dust component at 1000k since this may better model strong agn @xcite , though we note the difference in the fluxes and ews between the default temperature array and this modified version is negligible for a majority of the sources and a factor of 2 or lower for most of the remaining sources . we consider a pah feature to be detected if it is present at greater than the 5@xmath15 level . in the subsequent analysis , we consider pah upper limits on the flux , but only include ews for detected sources . pah measurements are listed in tables [ ssgss_pah ] - [ 12m_pah ] . optical and ir fluxes have been utilized to calibrate sfrs in quiescent galaxies . below we investigate if emission from the agn contaminates these parameters and if composite systems are more similar to star - forming galaxies or seyfert 2s . here we consider optical sfrs derived from sdss using the methodology described in @xcite . we also explore ir star formation proxies , namely the fine structure [ neii ] 12.81@xmath2 m and [ neiii ] 15.56@xmath2 m emission lines as well as the luminosity of the polycyclic aromatic hydrocarbon ( pah ) features at 7.7 , 11.3 and 17 @xmath2 m . for star - forming galaxies , sdss sfr@xmath18 is calculated from the h@xmath1 luminosity , corrected for dust extinction , measured through the 3 sdss fiber . in these systems , h@xmath1 emission results from the recombination of ionizing photons from massive o stars , making this flux a useful parameterization of recent star formation . using a kroupa imf @xcite , h@xmath1 luminosity is then converted into a sfr @xcite . such a straightforward calculation fails for active systems due to emission powered by the agn contaminating the h@xmath1 flux . however , a sfr can be estimated in these systems using the strength of the dn(4000 ) feature , which is a break in the optical spectrum at 4000 that indicates the age of the host galaxy stellar population . the sfr@xmath18 for composites and agn in sdss is calculated starting with the strength of the dn(4000 ) break . this is then converted into a sfr by adopting the relationship between sfr@xmath18/m@xmath19 and dn(4000 ) measured for quiescent systems and then using measured value of m@xmath19 @xcite . in the infrared , [ neii ] at 12.81@xmath2 m has an ionization potential of 21.56 ev and is therefore excited by star formation activity . [ neiii ] at 15.56@xmath2 m has a higher ionization potential than [ neii ] at 41 ev , but it becomes more dominant in lower - metallicity star forming systems ( e.g. * ? ? ? @xcite have demonstrated that the sum of [ neii ] and [ neiii ] correlates better with total infrared luminosity ( a reliable proxy of sfr ) than just [ neii ] as this sum better accommodates the low metallicity systems . they subsequently calibrated a sfr based on [ neii ] + [ neiii ] . pah macro - molecules / small - grains are ionized by single ultraviolet or optical photons @xcite and are very prominent in the spectra of star forming galaxies @xcite . sfrs have consequently been calibrated on pah luminosities ( e.g. * ? ? ? however , the pah luminosity may be affected by agn activity , and the effect may be different for the diferent pah features ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in this analysis , we consider the sum of the pah luminosities at 7.7 , 11.3 and 17@xmath2 m ( @xmath20 ) . in section 5 , we examine each pah feature individually as a function of increasing ionization field strength . we compare the optical and ir sfr diagnostics pair - wise in figures [ ne2_sdss ] - [ pahsum_sfr ] as both scatter plots ( left hand panels ) and histograms of the ratios of the two proxies under consideration ( right hand panels ) to search for systematic offsets among galaxy type . threshold we required for detection . ] in these plots , blue represents star - forming galaxies , green denotes composites and red indicates agn . to take into account upper limits , we used survival analysis ( asurv , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) to determine the best - fit trends to the star - forming galaxies and sy2s ( shown as the blue dotted - dashed lines and red dashed lines , respectively , in the left hand panels of figures [ ne2_sdss ] - [ pahsum_sfr ] ) , to calculate the mean ratio of sfr proxies for each galaxy type ( table [ sfr_ratios ] ) and to quantify the significance of agreement / disagreement of the distributions of the sfr proxy ratios among star - forming galaxies , composites and agn ( table [ sfr_2samp ] ) . we use the gehan s generalized wilcoxon test to calculate the probability that the distributions of sfr proxy ratios among the sub - samples being considered are drawn from the same parent population . a probability value ( @xmath21 ) @xmath22 indicates that we can reject the null hypothesis and that the distributions differ at a statistically significant level . we note that not all sources are included in each calculation , either due to absence of data ( i.e. 25 12@xmath2 m sources have no sdss coverage ) or lack of significant detections in the diagnostics being considered . the number of sources used for each calculation are listed in tables [ sfr_ratios ] - [ sfr_2samp ] . as shown in figure [ ne2_sdss ] and table [ sfr_2samp ] , the [ neii ] luminosity in relation to sdss sfr@xmath18 is consistent among star - forming galaxies , composites and agn . however , the sum of [ neii ] and [ neiii ] does show signs of agn contamination as these sources are systematically offset to higher l@xmath23}$ ] + l@xmath24}$ ] values with respect to star - forming galaxies , as seen in figure [ nesum_sdss ] . indeed , the distribution of l@xmath23}$ ] + l@xmath24}$ ] normalized by sfr@xmath18 is significantly different between star - forming galaxies and agn ( table [ sfr_2samp ] ) . this result is due to the harder ionization field from the agn boosting [ neiii ] to higher values , biasing the sum as a sf indicator in active systems . this result is consistent with previous works ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) that demonstrated that [ neiii]/[neii ] is correlated with agn activity and thus a tracer of ionization field hardness . interestingly , the composite galaxies are not affected : their l@xmath23}$ ] + l@xmath24}$ ] sum differs significantly from agn , but not from star - forming galaxies . hence , the sfr calibrated on l@xmath23}$ ] and l@xmath24}$ ] is reliable for star - forming and composite galaxies , but not agn . consequently , to test the efficacy of pah luminosities in parameterizing star formation in agn , we compare pah luminosities to sdss sfr@xmath18 and l@xmath23}$ ] and not the sum of l@xmath23}$ ] and l@xmath24}$ ] . both [ neii ] and pah luminosities are upper limits for at least a handful of sources . since asurv is unable to accommodate doubly censored data points , we calculated the ratio means and performed two sample tests using upper limits on [ neii ] and pah luminosities separately . we report the results of both analyses in tables [ sfr_ratios ] and [ sfr_2samp ] , including the number of sources for each calculation . we find that the distribution of pah luminosities relative to sfr@xmath18 is consistent among star - forming galaxies , composites and agn , though the slope of the relation for sy2s differs markedly compared with star - forming galaxies ( figure [ pahsum_sfr ] ) . the difference in pah emission becomes more pronounced when compared to l@xmath23}$ ] , regardless of whether l@xmath25 or l@xmath23}$ ] is treated as the upper limit ( see table [ sfr_2samp ] ) . as table [ sfr_ratios ] demonstrates , there is a decrease of about 32% in the mean in the ratio of pah to [ neii ] luminosity in agn compared to both the quiescent star - forming galaxies and the composites . at face value we can not say whether this is caused by supression of the pah emission or enhancement in the [ neii ] emission . we will return to this issue later , when we explore the effects of ionization field hardness on the relative strengths of pah features at 7.7 , 11.3 and 17 @xmath2 m individually . expanding on the work of @xcite , we now investigate the agreement among multi - wavelength diagnostics of ionization field hardness to determine where `` normal '' star - forming galaxies live in this parameter space . as agn activity weakens and star formation becomes more dominant , do the trends found for agn smoothly extend into the regime of quiescent galaxies ? as discussed in @xcite , the [ oiv]26@xmath2m/[neii]12.81@xmath2 m flux ratio is a reliable tracer of the incident radiation field hardness @xcite since [ oiv ] is primarily ionized by agn activity @xcite whereas [ neii]12.81@xmath2 m is excited by star formation @xcite . however , we do not have long - high ( lh ) coverage for the ssgss and s5 samples , and thus lack high resolution data for the [ oiv ] line at 26@xmath2 m . we therefore use [ nev]/[neii ] as a diagnostic of the relative importance of agn versus starburst activity @xcite . we compare this ratio with other diagnostics studied in @xcite . first we use the optical d parameter ( @xmath26/h\alpha ) + 0.45]^2 + [ log([oiii]/h\beta ) + 0.5]^2}$ ] ) , which is the distance a source lies on the bpt diagram from the locus of star - forming galaxies , so that a higher d value indicates a greater dominance of agn activity @xcite . second we use @xmath5 ( mir spectral index ) which increases in the presence of cold dust associated with star formation activity @xcite . finally we use the pah equivalent widths ( ews ) which are expected to decrease as agn activity becomes more dominant ( e.g. * ? ? ? we also investigate if the ratios of the fluxes of the different pah features are affected by agn activity , as found by @xcite , @xcite and @xcite . due to its high ionization potential ( 97.1 ev ) , [ nev ] has been claimed to cleanly trace smbh accretion @xcite and to be an unambiguous signature of agn activity @xcite . also , as it is formed in the narrow line region , it should ideally be unaffected by toroidal obscuration and therefore a tracer of intrinsic agn flux , similar to the [ oiii ] 5007 and [ oiv ] 26@xmath2 m lines @xcite . however , in a majority of our spectra , this feature was not significant above the noise . to use [ nev]/[neii ] as a tracer of ionization field hardness , we consider the non - detections in aggregate , stacking the spectra among sub - groups of galaxy type , rather than employing individual upper limits . for this stacking exercise , we separated star - forming galaxies , composites and agn into two groups each , based on their location in the bpt diagram , in order of increasing nuclear activity . figure [ bpt_nev ] illustrates the method we used . the left hand plot is a bpt diagram for our full sample , with the unbroken black lines showing the @xcite ( upper ) and @xcite ( lower ) demarcations between star - forming galaxies , composites and agn . the dashed lines illustrate the additional empirical separations we used , approximately tracing the halfway point within each region . these lines correspond , in order of lowest to highest , to the following equations : @xmath27/h\beta ) = \frac{0.61}{log([nii]/h\alpha)+0.1 } + 1.3\ ] ] @xmath27/h\beta ) = \frac{0.61}{log([nii]/h\alpha)-0.2 } + 1.3\ ] ] @xmath27/h\beta ) = \frac{0.61}{log([nii]/h\alpha)-1 } + 1.19\ ] ] we also employed a cut on [ nii]/h@xmath1 to exclude low metallicity systems : log ( [ nii]/h@xmath1 ) @xmath28 -0.6 dex for star - forming galaxies and log ( [ nii]/h@xmath1 ) @xmath28 -0.2 dex for agn . the right - hand panel of figure [ bpt_nev ] shows our resulting sub - groups depicted in different colors with the @xcite ( upper ) and @xcite ( lower ) demarcations overplotted : star - forming group 1 - dark blue , star - forming group 2 - light blue , composite group 1 - dark green , composite group 2 - light green , agn group 1 - gold , agn group 2 - red . we stacked together the irs sh spectra for the sources in each color bin from figure [ bpt_nev ] b ) not detected in [ nev ] . each spectrum was shifted to the rest - frame , rebinned to a common wavelength grid of equal pixel spacing and normalized by the the average mir flux between 13.25@xmath2 m and 13.75@xmath2 m ; this window was chosen since it is free of strong emission features . figure [ stack_nev ] shows the stacked sh spectra for each galaxy bin . we used a 3@xmath15 clipping on the average flux in each pixel to reject outliers that would have otherwise introduced erroneous features into our stacked spectra due to noise . [ nev ] was detected above the 5@xmath15 level in just one of these bins , agn group 2 . for the other bins , we obtained [ nev ] upper limits using the procedure outlined in section 3 . [ nev]/[neii ] was calculated by normalizing the [ nev ] upper limit / detection by the stacked [ neii ] emission feature in each group , which was detected at high significance in all cases . in the following analysis , we plot the [ nev]/[neii ] detections and stacked [ nev]/[neii ] upper limits against the optical d value , pah ew ( for the sub - set of sources where the pah luminosity was detected above the 5@xmath15 level ) and @xmath5 . for the stacked [ nev]/[neii ] ratios , we calculate the average d , pah ew and @xmath29 value in each bin . the [ nev ] stacked detection is illustrated by the triangle in the subsequent figures . figures [ nev_ne2_d_alpha ] - [ nev_ne2_ew ] illustrate the trends between [ nev]/[neii ] and the other diagnostics of radiation field hardness . as expected , [ nev]/[neii ] is significantly correlated with d and significantly anti - correlated with @xmath5 and pah ews , as seen in the correlation coefficients ( @xmath30 ) and associated probabilities that the two quantities are uncorrelated ( @xmath31 ) reported in table [ rel_agn_sfr ] . we note that one galaxy classified as a quiescent star - forming galaxy based on its optical ( sdss ) spectrum has a [ nev ] detection . this suggests that it is harboring a `` hidden '' agn , i.e. an agn embedded in very dusty host galaxy that attenuates the optical signatures used to identify agn . however , we note that this source has other ir parameters ( i.e. @xmath5 , pah ews ) more consistent with those of typical star - forming galaxies . this may argue against `` hidden '' agn interpretation , which may then call into question using solely [ nev ] detection as an unambiguous signature of agn activity . _ in any case , we emphasize that the optical and mid - ir classifications of galaxies as being dominated by star formation or by a type 2 agn agree in the overwhelming majority of cases . _ this is good news , since it shows that sdss classifications ( which are available for roughly one million galaxies ) are highly reliable . we do not find a large population of optically hidden agn , as found by goulding & alexander ( 2009 ) . over half of the [ nev ] detected sources in their volume limited sample of nearby galaxies are not optically classified as agn . however , they do not include optically classifed liners or composites as `` optically classified agn , '' while we do consider these to contain agn . the fraction of [ nev ] detected sources in their sample that are classified optically as quiescent star - forming galaxies is 24% ( 4 of 17 ) , which is still a substantially larger fraction than what we find . the mid - ir spectral slope index , @xmath5 , increases with the relative amount of star formation @xcite . we plot @xmath5 as a function of the optical d parameter in figure [ alpha_d ] which shows that a wide range of @xmath5 values are present for the star - forming galaxies . consequently , there is no significant anti - correlation present in the full sample , with correlation coefficient ( @xmath30 ) of -0.195 . however , if we consider just the agn , a significant trend is apparent : @xmath30=-0.474 . this correspondence likely drives the significant anti - correlation found for [ nev]/[neii ] vs @xmath5 . the ir spectral index is thus useful in parameterizing the relative contribution of star formation in active galaxies but this appears to break down in the quiescent galaxy regime . consequently , we do not utilize @xmath5 as a tracer of ionization field hardness in the subsequent analysis . previous studies have indicated that the pah equivalent widths ( ews ) are anti - correlated with the presence of increasing agn activity . this can be due to two effects . first , as the agn becomes increasingly dominant energetically , the emission due to warm dust heated by the agn will rise . even if the intrinsic pah emission is unaffected , this increase in the mid - ir continuum flux will cause the pah ews to decrease . a second possible effect would be due to the actual destruction of dust grains by the hard ionization field from the agn central engine @xcite . our new expanded data set that spans the range from quiescent systems to strong agn provides the opportunity to test these processes over a full range of ionization field hardness and to test which features may be most sensitive to agn activity . figure [ ew_d ] plots the pah ew at 7.7 @xmath2 m , 11.3 @xmath2 m and 17@xmath2 m as a function of the optical d parameter . the relationship between d and the pah ews is significant , with @xmath32 - 0.65 ( table [ ew ] ) . however , if we focus on just the star - forming galaxies and composites , this significant anti - correlation disappears : @xmath30=-0.227 , -0.125 and -0.230 for the 7.7@xmath2 m , 11.3@xmath2 m and 17 @xmath2 m features , respectively . hence , it appears that only in the objects that are optically classified as agn is the pah ew appreciably affected , suggesting the agn contributes a significant fraction of the mid - ir continuum or perhaps destroys pahs . as noted above , the pah ew are sensitive to both the relative amount of mid - ir continuum from the agn as well as effects on the pah emission itself . a complementary approach that isolates the effect of the agn directly on the pah emission is to consider changes in the ratios of the different pah features as the agn becomes more dominant . thus , similar to the work of @xcite and @xcite , we investigate the relationship between radiation field hardness ( parameterized by [ nev]/[neii ] and d ) and the ratios of the pah fluxes in figures [ ratios_nev_ne2 ] and [ ratios_d ] . since the [ nev ] line is only detected in the optically - classifed agn ( with few exceptions ) , we compare d to the pah ratios for the agn subset alone , as well as for the full sample ( including star - forming and composite systems ) . qualitatively , it appears that the distribution of the agn are different in these figures from the star forming and composite objects . for the most agn - dominated sources ( d @xmath28 1.2 and log(l@xmath33}$]/l@xmath23}$ ] ) @xmath34 ) , there is a subset of the population that has abnormally low ratios of pah 11.3 @xmath0 to 17 @xmath0 and abormally high ratios of pah 7.7 @xmath0 to 11.3 @xmath0 . a similar but weaker effect is seen in the pah 7.7 @xmath0 to 17 @xmath0 ratios . these distributions do not appear to take the form of a gradual correlation across the whole galaxy population , but rather appear as a threshold effect at high values of d and log ( l@xmath33}$]/l@xmath23}$ ] ) . despite this caveat , we have quantified the statistical significances of overall trends that may be present . we emphasize that the correlations may be weak , but the trends found can be quantified as statistically significant as indicated by the correlation coefficients ( @xmath30 ) listed in table [ pah_ratio ] . l@xmath35/l@xmath36 displays no significant trend with ionization hardness , regardless of the radiation field parameterization used and whether the full sample or agn only are considered . using [ nev]/[neii ] as a probe of the ionization field , a significant trend becomes apparent with l@xmath37/l@xmath36 ( @xmath30=-0.564 ) and l@xmath35/l@xmath37 ( @xmath30=0.611 ) . however , with d as a diagnostic of the incident radiation , the l@xmath37/l@xmath36 relationship becomes less significant in the full sample ( @xmath30=-0.364 ) and even less correlated for the agn sub - sampple ( @xmath30=-0.247 ) . l@xmath35/l@xmath37 is not significantly correlated in the full sample ( @xmath30=0.179 ) , but becomes more signficant in the agn sample ( @xmath30=0.431 ) . taken together , these results suggest that the pah feature at 11.3@xmath2 m is suppressed in agn - dominated systems relative to the 7.7 and 17 @xmath0 features . this disagrees with the results of @xcite , @xcite , @xcite and @xcite who find that ratio of l@xmath35/l@xmath37 is smaller in agn than in star - forming galaxies . the disagreement with the first three studies listed above could result from the inclusion of a greater number of high - luminosity agn - dominated in our analysis . clearly , agn affect pah macro - molecules and this effect could be more pronounced in more luminous agn . the inconsistency with @xcite could be due to their seyfert sample being systematically closer . perhaps the agn effect on pahs are different in the region hundreds of parsecs around the central engine , as probed by the @xcite analysis , rather than the kiloparsec to galaxy - wide scale that we are investigating where there is a greater interplay between host galaxy star formation and agn activity . different screening effects at different physical scales may play an important role . more advanced theoretical work on the effect of the agn ionization field on pah macro - molecules would help to understand the physical causes of these inconsistencies . because we have combined a number of different samples selected in different ways , our agn systems ( mostly drawn from the sdss [ oiii ] and 12@xmath2 m samples ) are typically at lower redshift than most of the star - forming and composite systems ( drawn from ssgss and s5 ) . could any of the results above be influenced by the systematic differences in the physical sizes being probed ( i.e. smaller scale nuclear emission in closeby agn versus larger scale host galaxy emission for sources further away ) ? to test this , we impose a redshift cut of [email protected] , which corresponds to the lowest redshift source in the ssgss sample . this cut eliminates most of the 12@xmath2 m sources and almost half of the [ oiii ] sample . however , since we do retain a handful of [ oiii]-selected agn which are strong agn , we are able to test the relationship between pah ews and ionization field hardness over a wide range of nuclear galactic activity . figure [ ew_d_z_cut ] and [ ratios_d_z_cut ] illustrate that the trends found for the full sample still hold for this pared down sample . the agn at higher d values are associated with suppressed pah ews at a statistically significant level even when the host galaxy is more fully sampled ( see table [ ew ] ) . the only change in the pah ratio analysis is the significant anti - correlation of l@xmath37/l@xmath36 with respect to d in the agn sub - sample ( see table [ pah_ratio ] ) . we note that though the l@xmath35/l@xmath37 trend is driven by a handful of agn , the relationship is similar to the parent dataset where the parameter space of high l@xmath35/l@xmath37 values is more fully sampled . the thresholds in d and [ nev]/[neii ] for changes in the pah ratios noted above are still present . taken together , these results suggest that some care be used when using the pah luminosities to estimate star formation rates in galaxies with dominant agn . the 11.3 @xmath0 feature appears to be rather strongly suppressed in the most agn - dominated systems . mid - infrared emission in active galaxies results from both the toroidal obscuring medium , accounting for @xmath1420% of the bolometric agn flux @xcite , and host galaxy star formation . we define l@xmath38 as the rest - frame luminosity at 13.5@xmath2 m averaged over a 3@xmath2 m window : this region is free from strong pah features and provides a reliable estimate of the mir continuum emission . using quiescent galaxies as a baseline to estimate the amount of mir emission due to star formation , we can empircally separate the contributions of starburst activity and toroidal emission . to quantify the relative importance of these two effects , we solve : @xmath39 we set l@xmath40 = l@xmath23}$ ] and l@xmath41 = l@xmath42,agn}$ ] , the agn contribution to the observed [ oiii ] emission . to calculate l@xmath42,agn}$ ] , we use equation 3 in @xcite , which scales the observed [ oiii ] flux based on the source s position in the bpt diagram . for the agn where this correction failed ( i.e. l@xmath42,agn } > $ ] l@xmath42,obs}$ ] ) , we set l@xmath42,agn}$ ] = l@xmath42,obs}$ ] . for the composite and star - forming galaxies where this correction failed ( i.e. l@xmath42,agn } < 0 $ ] ) , we set l@xmath42,agn}$ ] = 0 . a vast majority of the star - forming galaxies had l@xmath42,agn}$ ] set to 0 . using the sources that had [ neii ] detections , we calculated @xmath1 and @xmath4 by using a least squares trimmed ( lts ) regression , a robust regression technique that removes outliers , on the above equation . as each execution of the lts regression results in different values for @xmath1 and @xmath4 , we performed this calculation 1000 times to determine the average values and spreads on the distributions . the resulting values , using a 3@xmath15 clipping mean , are @xmath1=89@xmath431 and @xmath4=111@xmath437 . we note that the difference between these values and using a straight average are negligible for @xmath1 and small for @xmath4 ( c.f . 109@xmath4315 ) . in figure [ mir_agn_sfr ] , we plot l@xmath38 vs. @xmath1l@xmath40 + @xmath4l@xmath41 ( i.e. 89 @xmath44 l@xmath23}$ ] + 124 @xmath44 l@xmath42,agn}$ ] ) , with the overplotted line indicating where the two quantities are equal . no systematic offsets among galaxy sub - type are present . this relationship can therefore be valuable for estimating the mir flux due to star - formation ( or conversely from the torus ) in active galaxies . to test this decomposition , we have plotted @xmath45l@xmath23}$]/l@xmath38 vs. the pah ew at 7.7@xmath2 m in figure [ aneii_mir_ew77 ] . as we have demonstrated above , this pah macro - molecule / small - grain is the least affected by agn activity ( i.e. not suppressed like the 11.3@xmath2 m feature nor mildly boosted like the 17@xmath2 m feature ) . hence , to first order the ew of the 7.7@xmath2 m pah feature is tracing the `` extra '' mir emission from agn heated dust . if our decomposition is correct , the relative amount of mir emission due to star formation activity , i.e. @xmath45l@xmath23}$]/l@xmath38 , should increase monotonically with increasing ew of the pah 7.7@xmath2 m feature . the dotted dashed lines have unity slope and were chosen to intersect the points where the mir emission is solely due to star formation ( e.g. log ( @xmath45 l@xmath23}$]/l@xmath38 ) = 0 dex ) and fiducial pah ew values that approximately enclose the star - forming galaxy locus . many of the agn fall within the region demarcated by these two lines . the dashed line is the best - fit line to the agn sources where the slope is held constant at unity . this fit neatly bisects the above region . these results suggest that our empirical decomposition works reasonably well in describing the relative contributions of star formation and agn activity to the mir emission . we have explored the infrared and optical parameter space where star - forming galaxies , composites and agn live to analyze diagnostics that parameterize host galaxy star formation . using star - forming galaxies as a control sample , we have investigated which star formation diagnostics are least affected by agn activity . we have studied diagnostics that trace the interplay between agn and starburst activity and tested whether a smooth transition exists over a range of radiation field hardness . finally , using the results of this analysis , we present an empirical decomposition of the mir flux into a star - forming and an agn component . _ our overall result is that the optical and mid - ir diagnostics of star formation and of the relative importance of young stars and the agn generally agree well . _ our more specific results are summarized as follows : * sfr diagnostics * the sdss derived sfr@xmath18 and the [ neii ] 12.81@xmath2 m luminosity agree well , and are the most reliable sf proxies we have considered for use in star - forming galaxies , composites and agn . the sum of [ neii ] and [ neiii ] 15.56@xmath2 m is systematically offset to higher values for agn due to the active nucleus boosting the [ neiii ] flux . the aggregate pah luminosity ( l@xmath25 ) relative to l@xmath23}$ ] is suppressed in agn compared with both quiescent and composite galaxies . though the ( l@xmath25)/sfr@xmath18 ratio agrees between star - forming galaxies and sy2s , analysis of pah flux ratios indicate abnormal behavior in some agn dominated systems , such as suppression of the 11.3@xmath2 m feature , suggesting diminished pah emission in strong agn rather than [ neii ] enhancement . the disparate slopes of ( l@xmath25)/sfr@xmath18 between sy2s and star - forming galaxies also hint that agn may contaminate the emission from pah macro - molecules . * ionization field hardness * we have used the optical d parameter , which indicates the distance a source lies from the star - forming galaxy locus on the bpt diagram @xcite , and l@xmath33}$]/l@xmath23}$ ] @xcite as probes of the incident radiation field hardness . the infrared diagnostics that parameterize the relative contributions of agn to star - forming activity that we have considered ( @xmath5 , pah ews and pah flux ratios ) only show significant trends with ionization field hardness in the agn population . this result suggests that rather than parameterizing incident radiation hardness in general , these ir diagnostics are only effective descriptors of ionization field hardness in agn . in agn , the spectral slope between 20 - 30 @xmath2 m ( measured by the spectral index @xmath5 ) has been shown to steepen in the presence of cold dust due to star formation @xcite . for the agn in our sample , @xmath5 is significantly anti - correlated with proxies of ionization field hardness , albeit with wide scatter , however this result does not hold when expanded into the regime of quiescent star forming galaxies . * pah ew * pah ews are significantly anti - correlated with the increasing hardness of radiation field , but this is only true when considering the agn . when considering just the star - forming galaxies , no trend exists between incident radiation hardness and the strength of the resulting pah feature , which agrees with the findings of @xcite and @xcite . this result is consistent with the hypothesis that the mid - ir continuum has a significant contribution from dust heated by the agn only in objects where the ionizing radiation field is dominated by the agn . for the 11.3@xmath2 m pah feature , our analysis of the pah flux ratios suggests that destruction of the pah macro - molecule in strong agn could also contribute to suppression of the pah 11.3@xmath2 m ew . * pah flux ratios * we find that the luminosity of pah 11.3 @xmath0 feature is signifcantly suppressed relative to the luminosities of the pah 7.7 @xmath0 and 17 @xmath0 features and relative to the optical and mid - ir derived star formation rates in the most agn - dominated systems . thus , some care is required in using the pah 11.3@xmath2 m luminosity as a proxy for the sfr in agn - dominated systems . * composite systems * galaxies that are optically classified as composite systems are more akin to quiescent star forming galaxies in terms of the mid - ir parameters that trace star formation : the smaller relative energetic importance of the agn in composites does not seem to affect [ neiii ] and pah luminosities . * [ nev ] as unambiguous agn signature * due to the high ionization potential of [ nev ] ( 97.1 ev ) , it is potentially an unequivocal signature of agn activity @xcite . one galaxy that is optically classied as a quiescent star - forming system in our sample has a [ nev ] detection , yet other ir parameters ( i.e. pah ew values ) are normal for a pure star forming galaxy . the nature of this galaxy as a strong `` hidden '' agn is not clear . * disentangling mir emission * using multiple linear regression , we fit the relation l@xmath38 = @xmath1l@xmath40 + @xmath4l@xmath41 to our full sample to determine @xmath1 and @xmath4 . we set l@xmath40 = l@xmath23}$ ] and l@xmath41=l@xmath42,agn}$ ] where in l@xmath42,agn}$ ] , we have subtracted out the estimated starburst contribution to the [ oiii ] flux using eq . 3 in @xcite . we find @xmath1=89@xmath431 and @xmath4=111@xmath437 . this decomposition can be useful in estimating the mir emission due to the circumnuclear agn obscuration or conversely host galaxy star formation in agn . lccccccc 33 & 160.291 & 56.932 & 0.185 & agn & 8.12@xmath43 0.70 & ... & 3.34@xmath43 0.57 + 61 & 160.952 & 58.197 & 0.073 & agn & 3.19@xmath43 0.08 & ... & 0.30@xmath43 0.05 + 11 & 161.720 & 56.252 & 0.047 & comp & 0.17@xmath43 0.01 & ... & @xmath46 0.06 + 12 & 162.268 & 56.224 & 0.072 & comp & 3.04@xmath43 0.09 & ... & 0.31@xmath43 0.04 + 15 & 161.758 & 56.307 & 0.153 & comp & 4.40@xmath43 0.30 & ... & @xmath46 1.64 + lcccccc 36.380 & -8.419 & 0.055 & agn & 1.16@xmath43 0.05 & ... & 1.02@xmath43 0.07 + 36.435 & -7.868 & 0.077 & agn & 2.72@xmath43 0.12 & ... & 1.03@xmath43 0.08 + 132.186 & 53.746 & 0.083 & agn & 4.72@xmath43 0.14 & 1.18@xmath43 0.09 & 2.54@xmath43 0.08 + 132.714 & 54.823 & 0.081 & agn & 3.17@xmath43 0.11 & ... & 2.25@xmath43 0.07 + 132.815 & 54.986 & 0.083 & agn & 3.64@xmath43 0.16 & ... & 1.68@xmath43 0.09 + ngc 291 & 13.375 & -8.768 & 0.019 & 1.61@xmath43 0.04 & 0.60@xmath43 0.02 & 1.12@xmath43 0.02 + mrk 609 & 51.356 & -6.144 & 0.034 & 5.51@xmath43 0.12 & ... & 1.80@xmath43 0.06 + ic 0486 & 120.087 & 26.614 & 0.027 & 1.10@xmath43 0.10 & 0.55@xmath43 0.06 & 1.04@xmath43 0.05 + 2masx j08035923 + 2345201 & 120.997 & 23.756 & 0.029 & ... & @xmath46 0.14 & @xmath46 0.42 + 2masx j08244333 + 2959238 & 126.180 & 29.990 & 0.025 & 0.54@xmath43 0.06 & 0.55@xmath43 0.07 & 0.70@xmath43 0.06 + iras 00198 - 7926 & 5.473 & -79.169 & 0.073 & agn & 35.65@xmath43 1.45 & 49.05@xmath43 1.53 & 60.06@xmath43 1.31 + ngc 0424 & 17.865 & -38.083 & 0.012 & agn & 3.06@xmath43 0.19 & 4.64@xmath43 0.27 & 5.67@xmath43 0.17 + ngc 1068 & 40.670 & -0.013 & 0.004 & agn & 5.34@xmath43 0.28 & 10.11@xmath43 0.48 & 15.02@xmath43 0.32 + ngc 1144 & 43.790 & -0.178 & 0.028 & agn & 10.79@xmath43 0.26 & 0.61@xmath43 0.10 & 2.88@xmath43 0.05 + ngc 1320 & 51.203 & -3.042 & 0.009 & agn & 0.47@xmath43 0.02 & 0.59@xmath43 0.02 & 0.78@xmath43 0.02 + lccccccc 13 & agn & 16.59@xmath43 0.82 & 1.23 & 7.61@xmath43 0.32 & 0.39 & 6.12@xmath43 1.09 & 0.27 + 21 & agn & 30.66@xmath43 1.73 & 2.95 & 10.80@xmath43 0.60 & 1.43 & 6.44@xmath43 0.49 & 1.09 + 23 & agn & ... & ... & ... & ... & ... & ... + 33 & agn & 157.18@xmath43 6.51 & 3.14 & 53.40@xmath43 1.08 & 1.01 & 37.98@xmath43 3.62 & 0.66 + 50 & agn & ... & ... & ... & ... & ... & ... + lcccccccc 36.380 & -8.419 & agn & 24.01@xmath43 2.34 & 5.74 & 7.92@xmath43 0.71 & 3.62 & 15.13@xmath43 2.76 & 2.28 + 36.435 & -7.868 & agn & 59.59@xmath43 4.64 & 1.44 & 16.41@xmath43 0.62 & 0.53 & 7.79@xmath43 1.54 & 0.28 + 132.186 & 53.746 & agn & 144.14@xmath43 5.69 & 4.81 & 41.27@xmath43 0.81 & 1.81 & 32.40@xmath43 1.82 & 1.38 + 132.714 & 54.823 & agn & 54.59@xmath43 5.05 & 4.51 & 21.61@xmath43 0.71 & 1.89 & 16.96@xmath43 1.69 & 1.04 + 132.815 & 54.986 & agn & 107.85@xmath43 3.30 & 5.05 & 29.54@xmath43 0.49 & 1.62 & 19.68@xmath43 1.23 & 1.17 + lcccccc ngc 0291 & 34.69@xmath43 0.95 & 10.07 & 7.44@xmath43 0.15 & 0.75 & 9.20@xmath43 0.15 & 0.50 + mrk 0609 & ... & ... & 52.35@xmath43 0.78 & 1.08 & 54.58@xmath43 0.54 & 0.94 + ic 0486 & 22.69@xmath43 0.47 & 0.75 & 4.15@xmath43 0.19 & 0.12 & 5.77@xmath43 0.12 & 0.16 + 2masx j08035923 + 2345201 & ... & ... & 1.49@xmath43 0.05 & 0.30 & 2.10@xmath43 0.13 & 0.40 + 2masx j08244333 + 2959238 & ... & ... & 2.80@xmath43 0.16 & 0.04 & 5.67@xmath43 0.29 & 0.10 + ngc 0424 & agn & 10.25@xmath43 0.65 & 0.09 & 4.46@xmath43 0.35 & 0.06 & 8.05@xmath43 0.61 & 0.19 + ngc 1144 & agn & 258.85@xmath43 1.37 & 2.21 & 92.63@xmath43 0.99 & 1.83 & 51.43@xmath43 0.73 & 1.56 + ngc 1320 & agn & 9.69@xmath43 0.38 & 0.52 & 2.45@xmath43 0.08 & 0.16 & 1.31@xmath43 0.12 & 0.11 + ngc 1386 & agn & 0.84@xmath43 0.02 & 0.71 & 0.16@xmath43 0.00 & 0.14 & 0.26@xmath43 0.01 & 0.27 + ngc 1667 & agn & 61.97@xmath43 1.14 & 3.85 & 18.76@xmath43 0.22 & 2.18 & 9.18@xmath43 0.13 & 1.63 + l@xmath23}$]/sfr@xmath18 & 40.97@xmath47 & 223 & 40.91@xmath48 & 32 & 40.92@xmath49 & 39 + ( l@xmath23}$]+l@xmath24}$])/sfr@xmath18 & 40.84@xmath50 & 223 & 40.74@xmath51 & 32 & 41.11@xmath52 & 39 + l@xmath25/sfr@xmath18 & 42.50@xmath53 & 261 & 42.59@xmath54 & 46 & 42.49@xmath55 & 35 + l@xmath25/l@xmath23}$ ] & 1.50 & 207 & 1.59@xmath56 & 31 & 1.33@xmath57 & 21 & pah uls + l@xmath23}$]/l@xmath25 & -1.61 & 141 & -1.62@xmath58 & 25 & -1.50@xmath59 & 17 & [ neii ] uls + l@xmath23}$]/sfr@xmath18 & 0.127 & 255 & 0.598 & 71 & 0.624 & 223 + ( l@xmath23}$]+l@xmath24}$])/sfr@xmath18 & 0.175 & 255 & 0.0005 & 71 & @xmath461@xmath60 & 262 + l@xmath25/sfr@xmath18 & 0.331 & 307 & 0.365 & 81 & 0.758 & 296 + l@xmath25/l@xmath23}$ ] & 0.213 & 238 & 0.0002 & 77 & 0.003 & 253 & pah uls + l@xmath23}$]/l@xmath25 & 0.945 & 166 & 0.006 & 42 & 0.0004 & 158 & [ neii ] uls + l@xmath33}$]/l@xmath23}$ ] vs. d & 0.562 & 4@xmath60 + l@xmath33}$]/l@xmath23}$ ] vs. @xmath5 & -0.568 & 6@xmath60 + l@xmath33}$]/l@xmath23}$ ] vs. pah ew 7.7 @xmath2 m & -0.702 & 1@xmath60 + l@xmath33}$]/l@xmath23}$ ] vs. pah ew 11.3 @xmath2 m & -0.779 & @xmath61 + l@xmath33}$]/l@xmath23}$ ] vs. pah ew 17 @xmath2 m & -0.628 & 2@xmath60 + d vs. pah ew 7.7 @xmath2 m & -0.679 & 344 + d vs. pah ew 11.3 @xmath2 m & -0.747 & 368 + d vs. pah ew 17 @xmath2 m & -0.658 & 281 + + + + d vs. pah ew 7.7 @xmath2 m & -0.227 & 291 + d vs. pah ew 11.3 @xmath2 m & -0.125 & 303 + d vs. pah ew 17 @xmath2 m & -0.230 & 221 + + + + d vs. pah ew 7.7 @xmath2 m & -0.566 & 318 + d vs. pah ew 11.3 @xmath2 m & -0.653 & 336 + d vs. pah ew 17 @xmath2 m & -0.549 & 249 + + + l@xmath35/l@xmath36 & -0.241 & 33 + l@xmath37/l@xmath36 & -0.564 & 37 + l@xmath35/l@xmath37 & 0.611 & 34 + + + + l@xmath35/l@xmath36 & -0.193 & 266 + l@xmath37/l@xmath36 & -0.364 & 278 + l@xmath35/l@xmath37 & 0.179 & 340 + + + + l@xmath35/l@xmath36 & 0.082 & 47 + l@xmath37/l@xmath36 & -0.247 & 58 + l@xmath35/l@xmath37 & 0.431 & 51 + + + + l@xmath35/l@xmath36 & -0.283 & 240 + l@xmath37/l@xmath36 & -0.356 & 246 + l@xmath35/l@xmath37 & 0.084 & 314 + + + + l@xmath35/l@xmath36 & -0.208 & 22 + l@xmath37/l@xmath36 & -0.554 & 27 + l@xmath35/l@xmath37 & 0.590 & 26 +

mess showed that the genus 2 torelli group is isomorphic to a free group of countably infinite rank by showing that genus 2 torelli space is homotopy equivalent to an infinite wedge of circles . as an application of his computation , we compute the homotopy type of the zero locus of any classical genus 2 theta function in , where denotes rank 2 siegel space . specifically , we show that the zero locus of any such function is homotopy equivalent to an infinite wedge of 2-spheres .

1509.08152

theta functions are classical holomorphic functions of several variables defined on @xmath3 , where @xmath4 denotes the rank @xmath5 siegel upper half - space . since the @xmath6th century , theta functions have played a central role in algebraic geometry , especially the geometry of curves and abelian varieties . although these functions have been the object of intense study for well over a century , it appears that essentially nothing is known about the specific topological properties of their zero loci once @xmath7 . in this paper , we will use geoffrey mess s explicit computation @xcite of the isomorphism type of the genus 2 torelli group to prove the following result . [ theorem1 ] let @xmath8 denote a theta function of genus 2 . then the zero locus of @xmath9 is homotopy equivalent to an infinite bouquet of 2-spheres . the outline of the paper is as follows . in section 2 , we present the necessary background in mapping class groups and the geometry of algebraic curves and their jacobians . in section 3 , we establish a link between zero loci of genus 2 theta functions and the universal family @xmath10 over the torelli space of genus 2 curves of compact type . let @xmath11 denote the zero locus of @xmath9 on the universal family @xmath12 of principally polarized abelian surfaces . [ proposition8 ] as a family of framed curves , the family @xmath13 is isomorphic to the universal family @xmath10 . we will also establish the geometric facts about the universal family needed to compute the integral homology groups of the zero locus of @xmath9 in @xmath14 . this information suffice to determine the homotopy type since , as we will show , @xmath11 is simply connected . proposition [ proposition8 ] is of crucial importance here , as it allows us to compute many of these homology groups in terms of the homology of the genus @xmath15 torelli group @xmath0 . sections 4 - 6 are technical in nature , and this is where the homological computations will take place . the key results here are propositions [ prop22 ] , [ prop31 ] and [ prop34 ] which together , along with the homological form of whitehead s theorem , imply that @xmath11 is homotopy equivalent to an infinite bouquet of 2-spheres . the paper concludes with section 7 , where we use commutator calculus in the fundamental group of a genus 2 surface to construct an infinite family of homotopically non - trivial maps @xmath16 . the main results of this paper were a part of my thesis . i would like to thank my advisor dick hain for suggesting this problem . fix a closed orientable surface of genus @xmath5 with @xmath17 marked points @xmath18 . the mapping class group @xmath19 is the group of isotopy classes of orientation - preserving diffeomorphisms of @xmath18 that fix the marked points . when @xmath20 , it will be omitted from the notation . note that the isomorphism type of the group @xmath21 depends only on @xmath5 and @xmath17 . the _ torelli group _ @xmath22 is the kernel of the natural map @xmath23 . in other words , the elements of @xmath22 are isotopy classes of orientation preserving diffeomorphisms of @xmath24 which act trivially on @xmath25 . the torelli group @xmath22 is not well - understood when @xmath26 . mess @xcite was , however , able to give a very explicit description of @xmath0 . to describe his result , we will need to introduce some terminology . let @xmath27 be a free abelian group endowed with a symplectic form . symplectic splitting _ of @xmath28 is a direct sum decomposition @xmath29 such that @xmath30 is orthogonal to @xmath31 with respect to @xmath32 . if @xmath27 is equal to @xmath25 equipped with an intersection form , such a decomposition is called a _ homology splitting_. let @xmath33 denote a separating simple closed curve ( sscc ) on @xmath24 . then @xmath33 induces a homology splitting on @xmath24 . conversely , every homology splitting on @xmath24 arises in this way . two sscc s @xmath34 on @xmath24 induce the same homology splitting if and only if @xmath35 is carried into @xmath36 by an element of @xmath22 . ( mess , @xcite)[theorem 9 ] the torelli group @xmath0 is isomorphic to a free group of countably infinite rank . there is precisely one free generator for each homology splitting on @xmath37 . the generator corresponding to a homology splitting @xmath38 is a dehn twist @xmath39 on a sscc @xmath40 realizing @xmath38 . _ it is important to note that the isotopy class of @xmath40 can not be chosen arbitrarily it is contained in a _ specific _ isotopy class of curves which induce the homology splitting @xmath38 . unfortunately , this isotopy class has not been determined explicitly . in @xcite , mess conjectures that the isotopy class of @xmath40 is represented by shortest - length closed geodesic inducing the homology splitting @xmath38 . _ the moduli space of genus @xmath5 curves @xmath41 is , as a variety , isomorphic to the quotient space @xmath42 . the torelli space @xmath43 is the quotient @xmath44 . it is the moduli space of genus @xmath5 curves with a symplectic basis for the first integral homology , i.e. a homology framing . because @xmath22 is torsion - free , @xmath43 is a @xmath45-space . a curve of compact type is a stable nodal curve all of whose irreducible components are smooth and whose dual graph is a tree . by deformation theory , the @xmath43 can be enlarged to a complex manifold @xmath46 whose points parametrize genus @xmath5 curves of compact type with a homology framing . the @xmath47-action on @xmath43 extends to @xmath46 , and the moduli space @xmath48 of curves of compact type is the quotient @xmath49 . a _ polarized abelian variety _ is an abelian variety along with a cohomology class @xmath50 ( called a polarization ) represented by a positive , integral , translation - invariant @xmath51 form . a _ principally polarized _ abelian variety ( ppav ) is a polarized abelian variety whose polarization , viewed as a skew - symmetric bilinear form on @xmath52 can be put into the form @xmath53 with respect to some integral basis for @xmath52 . a _ framing _ on a ppav @xmath54 is a choice of basis for @xmath52 which is symplectic with respect to @xmath55 . the symplectic basis will be denoted @xmath56 . given a framed ppav @xmath57 one can find a unique basis of holomorphic 1-forms for which the period matrix is of the form @xmath58 in this case , @xmath59 is called the _ normalized period matrix _ of @xmath60 . the space of normalized period matrices @xmath59 of @xmath5-dimensional ppav s is known as _ siegel space of rank @xmath5_. it will be denoted by @xmath4 . it is an open subspace of @xmath61 . the symplectic group @xmath47 of integral symplectic @xmath62 matrices holomorphically on @xmath4 by the following formula : @xmath63 the product of two ppav s @xmath64 is the ppav whose underlying torus is @xmath65 and whose polarization is the product polarization @xmath66 , where @xmath67 is projection onto the @xmath68 factor . a ppav which can be expressed as a product of two ppav s of positive dimension is called _ reducible_. the sublocus of @xmath4 parametrizing reducible ppav s will be denoted @xmath69 . there is an equality @xmath70}\bigcup_{\phi\in { \text{sp}}_g({\mathbb{z } } ) } \phi\left({\mathfrak{h}}_j\times { \mathfrak{h}}_{g - j}\right).\ ] ] [ remark11 ] the following notation shall be used throughout the paper . if a framed ( necessarily reducible ) ppav has period matrix @xmath71 for matrices @xmath72 with @xmath73 and @xmath74 , then we will write @xmath75 . there is a holomorphic map @xmath76 which sends the isomorphism class @xmath77 $ ] of a framed curve to its period matrix . it is called the _ period map_. the sharp form of the torelli theorem @xcite states that the period map is a double branched cover of its image , and that it is injective along the locus of hyperelliptic curves . the period map extends to a proper holomorphic map @xmath78 ( the _ extended period map _ ) that sends a framed curve of compact type to its period matrix . the torelli theorem no longer holds in this setting ; the fiber over a point in @xmath69 will have positive dimension . the boundary @xmath79 is equal to the preimage in @xmath46 of @xmath69 under the period map . a _ theta function of characteristic @xmath80 _ is a holomorphic function @xmath81 defined by the following series : @xmath82 where @xmath83 . the vector @xmath80 corresponds to a 2-torsion point of @xmath84 and @xmath85 is called a theta function _ of characteristic @xmath80_. by differentiating the series ( [ thetaseries ] ) , one shows that @xmath85 satisfies the _ heat equation _ : @xmath86 where @xmath87 and @xmath88 is the kronecker delta . the theta function @xmath85 is said to be _ even _ ( _ odd _ ) if for fixed @xmath59 it is even ( odd ) as a function of @xmath89 . equivalently , @xmath85 is even ( odd ) if the number @xmath90 is even ( odd ) . if @xmath85 is even ( odd ) , then the characteristic @xmath80 is also said to be even ( odd ) . equivalently , @xmath80 is even ( odd ) if and only if the number @xmath91 is equal to 0 ( or 1 ) . it should be noted that the theta function @xmath92 is a non - zero multiple of @xmath93 . if a characteristic @xmath94 can be written in the form @xmath95 for some characteristics @xmath96 with @xmath97 , then we write @xmath98 . if @xmath59 is a block matrix of the form @xmath99 , with @xmath100 , then there are characteristics @xmath101 such that @xmath102 and @xmath103 where @xmath104 . + + the restriction of the theta function @xmath85 to @xmath105 is the _ thetanull of characteristic @xmath80_. it is a holomorphic function @xmath106 the thetanull @xmath107 is called even or odd , respectively , if the corresponding theta function is . the odd thetanulls vanish identically . theta functions enjoy special transformation properties with respect to the @xmath47-action on @xmath3 given by the formula @xmath108 . more specifically , we have the function @xmath109 is equal to @xmath110 , where @xmath111 is a specific ( very complicated ) non - vanishing holomorphic function depending on @xmath112 and @xmath89 and @xmath113 is a characteristic depending on @xmath80 and @xmath114 given by another complicated formula . precise formulas are given on p.227 of @xcite . thetanulls obey similar transformation laws . let @xmath115 be a smooth curve of genus @xmath5 , and let @xmath116 denote the codimension 1 subvariety consisting of classes of effective divisors on @xmath115 of degree @xmath117 . choose a theta characteristic @xmath118 on @xmath115 . the map @xmath119 defined by @xmath120 sends @xmath121 to the locus @xmath122 in @xmath123 . by riemann - roch , the divisor @xmath122 is a symmetric theta divisor ; it will be denoted by @xmath11 . this is a geometric formulation of _ riemann s theorem_. the parity of @xmath11 matches that of the theta characteristic @xmath118 . let @xmath80 be a characteristic such that @xmath11 is the zero divisor of @xmath85 . then _ riemann s singularity theorem _ states that for @xmath124 , @xmath125 let @xmath115 denote a smooth curve of genus 2 . then there is a special interaction between the theta characteristics , weierstrass points , and the jacobian of @xmath115 . since the canonical linear system gives a 2:1 covering @xmath126 , the weierstrass point of @xmath115 coincide with the odd theta characteristics on @xmath115 . by abel s theorem , the choice of a weierstrass point @xmath127 on @xmath115 gives rise to an embedding @xmath128 defined by @xmath129 . the image is an odd theta divisor in @xmath130 . thus any theta divisor on @xmath131 is isomorphic to @xmath115 and is , in particular , a non - singular curve . the riemann singularity theorem gives a relation @xmath132 where @xmath133 is the period matrix of @xmath131 and @xmath85 is the theta function whose zero divisor is @xmath134 . this implies that any theta function on the jacobian of a smooth curve of genus 2 vanishes only to first order along its zero locus . + + if @xmath115 is a _ singular _ genus 2 curve of compact type , then it consists of two elliptic curves @xmath135 joined at a single node . the generalized jacobian of @xmath115 is the product @xmath136 . any theta divisor on a 1-dimensional ppav consists of a single point , and , from our description of the theta divisor on a reducible ppav , it follows that theta divisor of @xmath136 is of the form @xmath137 , for some points @xmath138 . in particular , @xmath115 is again isomorphic to the theta divisor of its ( generalized ) jacobian . any theta function @xmath139 on @xmath136 vanishes to first order along its zero locus , except at a single point where it vanishes to order 2 . let @xmath141 denote the universal family of framed principally polarized abelian surfaces over @xmath2 . this can be constructed as the quotient of @xmath14 by @xmath142 via the action @xmath143 since it has a global homology framing , it is a topologically trivial fiber bundle over @xmath2 with fiber a real 4-torus . any theta function @xmath144 can be viewed as a section of a holomorphic line bundle on @xmath145 . the zero locus @xmath11 of @xmath9 in @xmath145 will be referred to as a _ theta divisor on @xmath145_. it can be viewed as a family of curves over @xmath2 . the geometry of @xmath146 can be described explicitly using the special properties of theta functions . this will allow us to give a very concrete description of the geometry of the universal curve @xmath147 . we will be able to deduce results about the geometry and topology of the universal covering @xmath148 , since , as we will show , the universal cover of @xmath11 is simply the preimage of @xmath11 in @xmath14 , i.e. the zero locus of @xmath9 in @xmath14 . before we begin , we fix some notation . once and for all , define @xmath149 to be the theta function @xmath85 with characteristic @xmath150 . define @xmath151 to be the zero locus of @xmath149 in @xmath145 . this is done for technical reasons , and is only a matter of convenience . any other theta divisor on @xmath145 is a translate of @xmath151 by a 2-torsion section . most of the proofs that follows work equally well , with minor modifications , if an arbitrary theta function is used in place of @xmath149 . recall that a proper , surjective holomorphic map @xmath152 between complex analytic spaces is called a _ family of nodal curves _ if it is flat and its fibers are nodal curves @xcite . our first goal is to show that @xmath153 is a family of curves . this involves checking that the family is flat . there is a convenient test for flatness that applies when the source and target are both connected complex manifolds , namely that the fibers are equidimensional ( see , for example , @xcite , pp.113 - 114 ) . in order to apply this criterion , we will first show that @xmath149 defines a complex submanifold @xmath151 of @xmath145 . the zero locus @xmath154 is a non - singular subvariety . this condition can be checked locally . at any point @xmath155 , there is a defining function for @xmath151 of the form @xmath156 on an open subset @xmath157 of @xmath14 . by the heat equation , the gradient of this holomorphic function is @xmath158 this can never vanish , as that would imply that there is a period matrix @xmath159 theta function @xmath160 has a zero of multiplicity at least 3 . this would contradict the fact that @xmath161 has at worst nodal singularities , as explained in section [ section2.4.2 ] . the projection @xmath162 is a family of nodal curves . every fiber is a genus 2 curve of compact type , and the fiber over @xmath163 is naturally a framed curve with period matrix @xmath133 . since the map @xmath141 is proper and @xmath151 is a closed analytic subset of @xmath145 , it follows that @xmath164 is proper holomorphic map . the fiber @xmath165 over @xmath163 is isomorphic , as an analytic space , to the intersection @xmath166 . this is the analytic subspace of @xmath145 locally cut out by functions @xmath167 from here it follows that the fiber @xmath165 is isomorphic to the zero locus of @xmath168 inside the complex torus @xmath169 . this is a genus 2 curve of compact type ( hence a nodal curve ) , and it naturally inherits the homology framing on @xmath170 . with respect to this framing , the curve @xmath165 has period matrix @xmath133 . since @xmath151 and @xmath2 are connected complex manifolds and the fibers of @xmath171 are equidimensional , if follows that @xmath171 is flat . since framed curves of compact type have no automorphisms , the torelli space of compact type curves @xmath172 is a fine moduli space . as such , there is a universal family of framed curves of compact type @xmath173 . the restriction of @xmath174 to the preimage of @xmath175 in @xmath140 is the genus @xmath15 torelli curve , which we denote by @xmath176 @xcite . it is a @xmath177 fiber bundle over @xmath175 with standard fiber @xmath37 . the moduli map @xmath178 is the holomorphic map which sends a period matrix @xmath133 to the isomorphism class @xmath179 $ ] of framed curves with period matrix @xmath133 . this is simply the inverse map of the period map @xmath180 , hence it is a biholomorphism . since the family @xmath162 is recovered from @xmath10 via pullback along @xmath181 , there is an isomorphism of families of framed curves @xmath182 . the fact that @xmath140 is isomorphic to @xmath151 will be of considerable utility in the sequel . using facts about theta functions , we will be able to explicitly describe the geometry of @xmath151 and its universal cover , and therefore that of @xmath140 and @xmath183 . + + one of the first such results concerns the geometry of the locus @xmath184 of reducible fibers of @xmath10 . define @xmath185 to be the preimage of @xmath186 in @xmath151 under the projection @xmath153 . recall that a divisor with _ simple normal crossings _ is a normal crossings divisor all of whose irreducible components are smooth . the locus @xmath184 of singular fibers in @xmath140 is a divisor with simple normal crossings . the isomorphism @xmath187 induces a biholomorphism @xmath188 . we will prove that @xmath185 has simple normal crossings . our strategy is to prove the result for the locus @xmath189 lying over the connected component @xmath190 and then use the action of @xmath191 on @xmath145 to transplant the result to other components of @xmath186 . the locus @xmath192 is clearly an analytic subset of @xmath151 of codimension 1 ; it is defined by the single equation @xmath193 , where @xmath194 denotes the @xmath195 component of the projection @xmath196 , viewed as a 3-tuple of holomorphic functions . we may identify @xmath192 with the zero locus of @xmath149 inside of @xmath197 . when @xmath198 , there is a product decomposition @xmath199 now @xmath192 is the union of the zero loci of @xmath200 inside of @xmath197 . both components are smooth because , in local coordinates @xmath201 the gradients of the functions @xmath202 are non - vanishing along their zero loci . this is a consequence of the fact that the jacobi theta function @xmath203 has only simple zeros . thus @xmath192 is the union of two smooth components . we now show that the locus where these components intersect is locally the union of the hyperplanes @xmath204 and @xmath205 inside of @xmath14 . the two components meet along the locus @xmath206 in @xmath197 . since this is local problem , we will work in a sufficiently small neighborhood of @xmath207 , as this is a point in the preimage of the intersection of the two components . locally around @xmath208 we have an identity @xmath209 where @xmath111 is a non - vanishing holomorphic function . the claim now follows . now consider the component @xmath210 of @xmath185 lying over the component @xmath211 of @xmath186 . choose an element @xmath212 such that @xmath114 carries the component @xmath213 to @xmath211 . then the automorphism of @xmath145 defined by @xmath214\rightarrow [ m\cdot \omega , z\cdot(c\omega+d)^{-1}]\ ] ] carries @xmath197 to @xmath215 . the theta function @xmath156 pulls back to @xmath216 where @xmath217 is a specific non - vanishing holomorphic function , and @xmath218 is a specific characteristic derived from @xmath118 ( see , for example , @xcite ) . as we have argued , the zero locus of this function in @xmath219 is a divisor with simple normal crossings . it follows that @xmath210 is a divisor with simple normal crossings in @xmath151 . since @xmath185 is the union of all of the @xmath210 , all of which are disjoint , @xmath185 is a divisor with simple normal crossings . let @xmath220 denote a component of @xmath210 . the projection @xmath153 restricts to a projection @xmath221 . the projection @xmath221 is a family of elliptic curves . it is a topologically trivial bundle with fiber a 2-torus . the idea is to show that ehremann s theorem @xcite is applicable . first observe that @xmath222 is proper . since @xmath220 is a closed subset of @xmath210 , it follows that @xmath221 is proper . we now argue that @xmath221 is a submersion . for simplicity , we work with a component @xmath223 over @xmath213 of @xmath186 . the argument can be transferred to the other components of @xmath186 by symmetry . since the problem is local , we work with the component @xmath224 of the zero locus of the holomorphic function @xmath225 in @xmath226 all of whose points are of the form @xmath227 where @xmath228 . fix a point @xmath229 . define a map @xmath230 by @xmath231 this gives a section of the projection @xmath232 passing through the point @xmath233 . this implies that @xmath234 is a submersion , and hence that the projection @xmath221 is a submersion . observe that the fiber in @xmath220 over a point @xmath235 is the zero locus of the function @xmath236 in the complex torus @xmath237 . this is just the subtorus @xmath238 of @xmath239 consisting of points of the form @xmath240 . an application of ehresmann s theorem shows that @xmath241 is a trivial fiber bundle with standard fiber a 2-torus . _ it should also be noted that the singular locus @xmath242 of @xmath185 is precisely the locus along which the components of @xmath185 intersect . each irreducible component of @xmath242 is isomorphic to @xmath243 . the projection @xmath153 submerses the unique component of @xmath244 lying over @xmath211 onto @xmath211 . _ in a later section , we will invoke certain results from stratification theory ( see @xcite ) . as a consequence of our work so far we state the following proposition . the filtration @xmath245 induces a whitney stratification of @xmath151 . with respect to this stratification and the whitney stratification of @xmath2 induced by @xmath246 , the projection @xmath153 is a stratified submersion . very shortly , we will compute the fundamental group of @xmath151 ( and therefore also that of @xmath140 ) . we will make use of the following lemma , though it is strictly stronger than what we will need . [ sections ] the universal family @xmath247 admits six holomorphic sections . on each smooth fiber , these sections pass through the six weierstrass points . on the singular fibers , these sections pass through the six non - zero points of order 2 . we will prove this lemma by working with the isomorphic family @xmath248 . there are six theta functions @xmath156 of odd characteristic . these have the property that the corresponding thetanull @xmath249 vanishes identically . observe that there is an equality @xmath250 where @xmath251 is a certain non - vanishing holomorphic function and @xmath252 are elements of @xmath253 . it now follows that there are six holomorphic sections @xmath254 , @xmath255 , defined by @xmath256 , where @xmath257 is an index labeling the six odd theta nulls . there is an involution of the family @xmath141 defined by @xmath258 on each fiber . this induces an involution @xmath259 of @xmath153 , since the function @xmath260 is preserved up to non - zero scaling by the transformation @xmath261 . observe that @xmath259 fixes the six sections @xmath262 , since on each fiber of @xmath145 they take values in the points of order 2 . it follows that , on each smooth fiber of @xmath151 , the involution @xmath259 restricts to the hyperelliptic involution and the images of the @xmath262 pass through the six weierstrass points of the curve . similarly , @xmath259 restricts to the hyperelliptic involution on each singular fiber , and the images of the @xmath262 pass through the six non - zero points of order 2 on the curve . all of this is then transported back to @xmath263 via the isomorphism @xmath187 . by mess s work @xcite , @xmath175 is homotopy equivalent to an infinite bouquet of circles . the generators of @xmath264 have the following description . identify @xmath175 with @xmath265 via the period map . let @xmath266 be a small holomorphic disk in @xmath2 that is transverse to @xmath211 at @xmath267 . then for each @xmath38 there exists a continuous path @xmath268 from @xmath133 to a point on the boundary of @xmath266 such that the oriented loops @xmath269 freely generate @xmath270 . the torelli group @xmath0 is freely generated by the monodromy of @xmath271 around these loops . + + let us now fix some notation . these conventions will be in effect for the rest of the paper . let @xmath272 denote the fundamental group of the reference surface @xmath37 of genus 2 . let @xmath273 denote the commutator subgroup , and define @xmath274 denote the abelianization of @xmath275 . the embedding @xmath276 induces an isomorphism on fundamental groups @xmath277 . consider the torelli curve @xmath176 ( see @xcite , p. 461 ) . this is a @xmath177 fiber bundle @xmath278 . the fundamental group of @xmath279 is isomorphic to the pointed torelli group @xmath280 ( see @xcite ) . the only non - trivial segment of the associated long exact sequence of homotopy groups is @xmath281 by the preceding lemma , the torelli curve admits a section @xmath282 . this allows us to view @xmath0 as a subgroup of @xmath280 and @xmath280 as a semi - direct product . observe that the inclusion @xmath283 factors as @xmath284 . notice also that this inclusion induces a surjection @xmath285 . this latter statement follows from the fact that the fibers of @xmath279 are embedded in the fibers of @xmath145 as theta divisors . + + consider the homomorphism @xmath286 induced by inclusion . this is a surjection , since @xmath279 is the complement in @xmath140 of the normal crossings divisor @xmath184 . elements of @xmath0 , viewed as a subgroup of @xmath280 , lie in the kernel of @xmath287 because @xmath2 is contractible . elements in the commutator subgroup @xmath288 also lie in the kernel of @xmath287 because , by mess s description of the monodromy , any separating simple closed curve on any fiber of @xmath279 is freely homotopic to a vanishing cycle of @xmath140 . since the vanishing cycles are themselves nullhomotopic , it follows that any separating simple closed curve on a fiber of @xmath279 is as well . since any element of @xmath289 can be written as a product of separating simple closed curves , it follows that @xmath289 lies in the kernel . thus the subgroup @xmath290 of @xmath280 lies in the kernel of @xmath287 . there is a surjective homomorphism @xmath291 defined by @xmath292 , where @xmath293 and @xmath294 . the kernel of this homomorphism is precisely @xmath295 . this induces an isomorphism @xmath296 . thus we have obtained a sequence of group homomorphisms @xmath297 such that the composition @xmath298 is surjective . since @xmath299 is a finitely generated abelian group , this composition is actually an isomorphism . furthermore , since @xmath300 is surjective , the homomorphism @xmath301 is injective , hence it is an isomorphism . we shall call upon the following corollary in a later section . the inclusion @xmath302 induces an isomorphism @xmath303 , where @xmath304 is any fiber . we are now in a position to describe the geometry of the universal cover @xmath183 of @xmath140 . we will do this by examining the geometry of the universal cover @xmath305 of @xmath151 . as a corollary of the preceding proposition , we see that the inclusion @xmath306 induces an isomorphism of fundamental groups . from this it follows that the preimage of @xmath151 in @xmath14 , i.e. the zero locus of @xmath9 in @xmath14 , is a connected covering space of @xmath151 with covering group @xmath299 . in other words , it is the universal cover of @xmath151 . we will denote this space by @xmath307 . it follows that there is an isomorphism @xmath308 and that it respects the fibers of the projections @xmath309 and @xmath310 . we record this as there is a biholomorphism @xmath308 which preserves the fibers of the projections @xmath311 and @xmath312 . let us now describe the fibers of @xmath311 in some detail . since the inclusion of a fiber @xmath313 induces a surjection on the level of fundamental groups , it follows that the preimage of @xmath314 in @xmath183 is connected . since this preimage is precisely the fiber @xmath315 of @xmath183 over @xmath59 . it follows that the fibers of @xmath315 are connected . since the covering @xmath316 is an @xmath299-covering , it follows that the induced map @xmath317 is an @xmath299-covering . since the fundamental group @xmath318 is a finitely generated group with abelianization @xmath299 , it follows that @xmath315 is the universal abelian covering of @xmath314 . we record this in the following the fiber @xmath315 of @xmath183 over @xmath319 is the universal abelian cover of the fiber @xmath314 of @xmath140 over @xmath59 . concretely , the fiber @xmath315 is isomorphic to the zero locus of the theta function @xmath320 . now define @xmath321 to be the preimage of @xmath185 under the universal covering @xmath322 . let @xmath323 denote the preimage of @xmath210 in @xmath305 . then @xmath323 is a divisor in @xmath305 with simple normal crossings . it is easiest to see this by considering the locus @xmath324 over @xmath213 . this is isomorphic to the zero locus of @xmath325 inside of @xmath226 . this is the subvariety of @xmath326 given by the union of the components @xmath327 , where @xmath328 , with equations @xmath329 it is a divisor with simple normal crossings , and each irreducible component is isomorphic to @xmath330 . note that , for fixed @xmath257 , the components @xmath331 are acted upon freely and transitively by @xmath332 , where @xmath333 is the component of @xmath189 whose preimage is the union of the @xmath331 . by symmetry , the other loci @xmath323 are seen to have an essentially identical description . also note that , much in the same way that the projection @xmath153 restricts to a submersion @xmath221 , the projection @xmath312 submerses each irreducible component of @xmath323 onto @xmath211 . additionally , each component of the singular locus @xmath334 of @xmath323 is isomorphic to @xmath213 and mapped submersively onto @xmath211 by the projection @xmath312 . note that set of these components are acted on freely and transitively by the deck group @xmath299 . the locus @xmath321 of reducible fibers in @xmath305 is a divisor with simple normal crossings . its irreducible components are isomorphic to @xmath330 . the singular locus @xmath335 is smooth , and its irreducible components are isomorphic to @xmath213 . let @xmath336 denote the universal abelian cover of @xmath37 . as a corollary of the fact that @xmath337 is a fiber bundle over @xmath265 with fiber @xmath37 , we have that @xmath338 is a fiber bundle over @xmath265 with fiber @xmath336 . if @xmath339 denotes the preimage of @xmath184 in @xmath183 , then the same statement applies to @xmath340 as well . the preceding discussion is summarized in the following proposition . the filtration @xmath341 induces a whitney stratification of @xmath305 . with respect to this stratification and the whitney stratification of @xmath2 induced by @xmath246 , the projection @xmath312 is a stratified submersion . in this section , we will show that the integral homology groups @xmath344 vanish when @xmath345 . this will be done by considering the pl gysin sequence derived in hain s paper @xcite . in our case , this is an exact sequence which relates the homology groups of @xmath183 and @xmath340 to the compactly supported cohomology of @xmath339 . also in @xcite , a spectral sequence was derived which will allow us to compute @xmath346 , given certain combinatorial input that describes the way in which the components of @xmath339 intersect . + + we begin with the following statement . [ lemma26 ] the total space of the fiber bundle @xmath347 is a @xmath348 space . the homology groups @xmath349 vanish for @xmath350 . since the fiber and base of this fiber bundle are @xmath348 spaces , the total space is as well . the group fundamental group @xmath351 is an extension @xmath352 where @xmath289 denotes the commutator subgroup of @xmath272 . recall that @xmath289 is a free group . since @xmath289 and @xmath0 are free , and therefore have cohomological dimension 1 , the group @xmath353 has cohomological dimension at most 2 ( see @xcite , p. 188 ) . by the universal coefficients theorem , this implies that @xmath354 for @xmath350 . finally , we obtain isomorphisms @xmath355 for @xmath350 . we now describe the long exact sequence constructed in @xcite . let @xmath356 be an oriented pl manifold of dimension @xmath357 , and let @xmath358 be a closed pl subset ( see @xcite , p.2 ) . then there is a long exact sequence @xmath359 the groups @xmath360 can be computed , in principle , using the spectral sequence that was constructed in the same paper . we describe it now . suppose that @xmath361 is a locally finite union @xmath362 of closed pl subspaces of the manifold @xmath356 , where @xmath363 is a partially ordered set . define @xmath364 and set @xmath365 then there is a spectral sequence @xmath366 where @xmath367 . we will apply these tools in the case where @xmath368 and @xmath369 . at this point , we need to describe the combinatorics of the ways in which the components of @xmath321 intersect . note that the components of the form @xmath370 do not intersect each other ; they only intersect components of the form @xmath371 . similarly , the components of the form @xmath371 only intersect the components of the form @xmath370 . just as the components of @xmath324 fall into two types , so do the components of @xmath323 for an arbitrary component @xmath211 of @xmath186 . two components of @xmath323 are of the same type if and only if the deck group @xmath299 carries one to the other . arbitrarily label the components of @xmath323 of one type with a `` 1 '' ; label the components of @xmath323 of the other type with a 2 " . now @xmath323 is the union of the components @xmath372 and @xmath373 . [ lemma27 ] the groups @xmath346 vanish unless @xmath374 or @xmath375 . place any partial order on the set of irreducible components of @xmath339 . observe that two components @xmath376 and @xmath377 intersect if and only if @xmath378 and @xmath379 . this implies that no more than two components of @xmath339 can intersect non - trivially . otherwise two components of @xmath380 of the same type would intersect ; but this is impossible . therefore the subsets @xmath381 are all empty for @xmath382 . as has already been established , the components of @xmath383 are analytically isomorphic to @xmath330 , hence diffeomorphic to @xmath384 . the components of @xmath385 are analytically isomorphic to @xmath243 , hence diffeomorphic to @xmath386 . thus the @xmath387 page of the spectral sequence for @xmath388 is of the following form : @xmath389 because of the shape of this spectral sequence , all differentials vanish and this spectral sequence degenerates at the @xmath387 page . it follows that @xmath390 unless @xmath374 or @xmath391 . we now combine this result with the vanishing of the homology of @xmath340 in degrees at least 3 to obtain , by use of the pl gysin sequence , the following result . [ prop22 ] the group @xmath344 vanishes for @xmath345 . for each @xmath392 , we have exact sequences @xmath393 coming from the pl gysin sequence . by lemma [ lemma27 ] , when @xmath345 , the groups @xmath394 and @xmath395 are both zero , since then @xmath396 and @xmath397 are both @xmath398 . furthermore , @xmath399 for @xmath345 by lemma [ lemma26 ] . this implies that @xmath400 for @xmath343 by exactness . it will be convenient for our purposes to replace @xmath340 with a homotopy equivalent space . specifically , we would like to view it , up to homotopy equivalence , as a @xmath336 bundle over a 1-complex . we construct the 1-complex that will serve as the base as follows . view the real line as a cw complex having 0-cells at the even integers , and 1-cells joining the integer @xmath401 to @xmath402 for each @xmath403 . now attach a circle @xmath404 at each even integer @xmath402 . the resulting space is homotopy equivalent to an infinite bouquet of spheres . choose some indexing @xmath405 of the components @xmath211 by @xmath406 . now recall the description of mess s results given in section 1.1.2 above . define a continuous map @xmath171 from @xmath356 to @xmath265 in the following way . on the interval @xmath407 $ ] define @xmath171 by @xmath408 ; on the circle @xmath404 , define @xmath171 to be an orientation - preserving homeomorphism @xmath409 which sends the point @xmath410 to the point where @xmath268 meets @xmath411 ; on the interval @xmath412 $ ] , define @xmath171 by @xmath413 . the map @xmath414 is a homotopy equivalence . the bundle @xmath415 can now be pulled back to @xmath356 . call the pullback bundle @xmath416 . the total space of this bundle is homotopy equivalent to @xmath340 . let @xmath417 . denote the restriction of @xmath416 to @xmath418 by @xmath419 . away from the point @xmath410 , choose a fiber @xmath420 of @xmath421 and define @xmath422 . there is a direct sum decomposition @xmath423 where @xmath56 is a free abelian group . the inclusion @xmath424 factors through the inclusion @xmath425 , and induces a homotopy equivalence @xmath426 . by excision , there are natural isomorphisms @xmath427 for each @xmath392 . the long exact sequence of homology for the pair @xmath428 has a segment @xmath429 which we study in several steps . 1 . : first we show that @xmath430 . we will study the exact sequence of the pair @xmath431 . since @xmath432 is a fiber bundle over a contractible cw - complex , it is a trivial bundle . it therefore has the homotopy type of a single fiber , which we denote by @xmath433 . since @xmath434 is homotopy equivalent to a 1-complex , it follows that @xmath432 is homotopy equivalent to a 1-complex . similarly , each connected component of @xmath435 is a @xmath336-bundle over a contractible cw - complex . it follows that @xmath435 is homotopy equivalent to a 1-complex . the groups @xmath436 and @xmath437 vanish by step 1 . it follows that the group @xmath438 also vanishes . on the other hand , the group @xmath439 vanishes and the group @xmath440 is free abelian . this implies that @xmath441 is a subgroup of @xmath440 and is therefore also free abelian . finally , this implies that @xmath442 is free abelian . 3 . : the conclusions arrived at in step 2 imply that the sequence @xmath443 is exact and that the image of the rightmost map is a free abelian group , which we denote by @xmath444 . this implies that there is a split exact sequence @xmath445 + by functoriality of the long exact sequence of homology for a pair , there is a commutative diagram @xmath446 & h_2({\widetilde{{\mathcal{c}}}_e})\ar[r]\ar[d]^{\cong } & h_2({\widetilde{{\mathcal{c}}}_x } ) \ar[d]^{\cong}\ar[r ] & { \mathcal{f}}'\ar[r ] & 0\\ 0\ar[r ] & h_2(\coprod_{\beta } { \widetilde{\mathcal{c}}}_{\partial \delta_{\beta}})\ar[r ] & h_2({\widetilde{\mathcal{c}}}-{\widetilde{\mathcal{c}}}^{red})\ar[ru ] & } \ ] ] with exact rows . this implies that the corresponding sequence @xmath447 is exact . the result follows . our tool of choice in this section is the long exact sequence of homology for the pair @xmath448 . recall that we have already shown that @xmath400 when @xmath345 and that @xmath399 when @xmath350 . we are thus left to consider the sequence @xmath449 we will show that the boundary map @xmath450 factors in the following way : @xmath451^{\displaystyle\partial}\ar[rd]_{\cong } & & h_2({\widetilde{\mathcal{c}}}-{\widetilde{\mathcal{c}}}^{red})\\ & \bigoplus_{\beta}h_2({\widetilde{\mathcal{c}}}_{\beta})\ar@{^{(}->}[ru ] & } \ ] ] this will imply that @xmath452 vanishes and , coupled with the fact that @xmath453 is a free abelian group ( generated by small disks transverse to the irreducible components of @xmath339 ) , that @xmath454 is a free abelian group . + + in order to carry this out , we will first find a convenient direct sum decomposition for @xmath455 which is in some sense compatible with the boundary map . choose a closed tubular neighborhood @xmath456 of @xmath186 . this restricts to a tubular neighborhood @xmath457 of each component @xmath211 of @xmath186 . let @xmath458 . by excision , there is a natural isomorphism @xmath459 to prove the desired result , we will need to gain a thorough understanding of the terms @xmath460 . fix a component @xmath211 of @xmath186 . we will study the topology of the pair of spaces @xmath461 . our goal is to show that there is a homeomorphism @xmath462 here @xmath463 is the fiber of @xmath464 over the point @xmath465 , @xmath466 is its preimage in @xmath183 and @xmath467 is the preimage of @xmath468 in @xmath183 . we may assume that @xmath469 . + + in order to prove this result , it appears easier to first work with the pair of spaces @xmath470 . the immediate goal is to show that the sequence of projections @xmath471 fibers the pair @xmath472 locally trivially . we will use thom s first isotopy lemma ( see @xcite , p.41 ) to accomplish this . for the reader s convenience , we reproduce it here . let @xmath473 be a proper stratified submersion . then there is a stratum - preserving homeomorphism @xmath474 which is smooth on each stratum and commutes with the projection to @xmath475 . in particular the fibers of @xmath476 are homeomorphic by a stratum - preserving homeomorphism . by radially shrinking @xmath456 if needed , we may assume that @xmath464 is contained in a larger open tubular neighborhood @xmath477 of @xmath211 such that the projection @xmath478 agrees with the projection @xmath479 on @xmath457 . [ stratification ] the subset @xmath480 is closed . the subsets @xmath481 of @xmath482 are locally closed in @xmath483 and form strata which endow @xmath482 with the structure of a whitney stratified subset of @xmath483 . with respect to this stratification , the projection @xmath484 is a proper stratified submersion . the proofs of the first two statements are straightforward , but somewhat lengthy , verifications . the projection is proper because both of the maps @xmath485 and @xmath479 are . the final statement follows from the fact that @xmath153 is a stratified submersion , the fact that @xmath486 is a submersion , and the fact that the bundle projections @xmath487 are submersions . observe that the fiber of @xmath484 over @xmath267 is @xmath488 . this receives an induced stratification from the stratification on @xmath482 ; its strata are the intersections of @xmath488 with the strata of @xmath482 . [ prop31 ] there is a commutative diagram @xmath489\ar[rr]^{\cong\ \ \ \ } & & { \mathfrak{h}}_{2,\beta}^{red}\times \theta_{\overline{\delta}_{\beta}}\ar[ld]\\ & { \mathfrak{h}}_{2,\beta}^{red } & } \ ] ] where the top horizontal map is a stratum - preserving homeomorphism and the diagonal maps are the natural projections . by lemma [ stratification ] , we may apply thom s first isotopy lemma to the projection map @xmath484 . this gives the required result . there is a homeomorphism of pairs @xmath490 which commutes with projection to @xmath211 . this lifts to a homeomorphism @xmath491 which commutes with projection to @xmath211 . equivalently , there is a homeomorphism @xmath492 commuting with projection to @xmath211 . we now derive a few consequences of proposition [ prop31 ] . the total space of the family @xmath493 is homotopy equivalent to a 1-complex . we will first show that the total space of the family @xmath494 is homotopy equivalent to a 1-dimensional cw - complex . this will then be combined with corollary 23 with lemma 24 . we may assume that @xmath495 is sufficiently small that @xmath496 is homotopy equivalent to its central fiber @xmath497 , which is topologically the wedge @xmath498 of two 2-tori . consequently , it is an aspherical space with fundamental group @xmath499 . then @xmath500 is an @xmath299-covering of @xmath501 and so is an aspherical space with fundamental group isomorphic to the commutator subgroup @xmath502 $ ] . on the other hand , the central fiber @xmath503 of @xmath497 in @xmath504 is another such space and so there is a homotopy equivalence @xmath505 . but @xmath503 is a 1-dimensional stein space and so , by a theorem of hamm @xcite it has cw dimension 1 . the inclusion @xmath506 is a homotopy equivalence . thus @xmath507 has the homotopy type of a cw complex of dimension at most 2 . by corollary 23 , the inclusion @xmath508 is a homotopy equivalence . since the inclusion @xmath509 is also a homotopy equivalence , the result follows after we note that the fundamental group @xmath353 of @xmath510 is given by an extension @xmath511 and therefore has geometric dimension at most 2 ( see @xcite , p.188 ) . _ actually , @xmath507 has geometric dimension 2 . by the hochschild - serre spectral sequence , there is an isomorphism @xmath512 . it is readily verified that this latter group is non - zero . _ equipped with these results , let us now return to our analysis of the long exact sequence of the terms @xmath460 . an application of lemmas 24 and 25 reveals that the long exact sequence of homology for the pair @xmath513 has a segment @xmath514 piecing all of these isomorphisms together , we now have a commutative diagram @xmath515^{\partial } & h_2({\widetilde{\mathcal{c}}}-{\widetilde{\mathcal{c}}}^{red})\\ \bigoplus_{\beta}h_3({\widetilde{\mathcal{c}}}_{\overline{\delta}_{\beta}},{\widetilde{\mathcal{c}}}_{\overline{\delta}_{\beta}}- { \widetilde{\mathcal{c}}}^{red}_{\beta } ) \ar[r]^{\ \ \ \ \ \ \ \ \partial } _ { \ \ \ \ \ \ \cong}\ar[u]^{\cong } & \bigoplus_{\beta}h_2({\widetilde{\mathcal{c}}}_{\partial\overline{\delta}_{\beta}})\ar@{^{(}->}[u ] } \ ] ] this implies that there is an exact sequence @xmath516 where @xmath517 . since @xmath453 is a free abelian group ( it is freely generated by small discs transverse to the components of @xmath339 ) , the image of @xmath454 in @xmath453 is also a free abelian group . thus @xmath454 is an extension of a free abelian group by a free abelian group and is therefore also free abelian . putting all of this together , we have deduced [ prop31 ] the boundary map @xmath450 is injective , and its cokernel is a free abelian group . consequently , the group @xmath452 vanishes and @xmath454 is a free abelian group . we are actually in a position to deduce the homotopy type of @xmath183 . our strategy is to apply the homological form of whitehead s theorem ( see @xcite , p.367 ) . there is a homotopy equivalence @xmath518 for some index set @xmath363 whose cardinality matches the rank of @xmath454 . since @xmath183 is simply - connected , the hurewicz theorem gives an isomorphism @xmath519 . thus each generator of @xmath454 is represented by a continuous map @xmath520 . by taking the wedge of all these maps , we obtain a continuous map @xmath521 which induces isomorphisms @xmath522 on all homology groups . here @xmath363 is an index set whose cardinality matches the rank of @xmath454 . since @xmath523 and @xmath183 are both simply - connected and each admits the structure of a cw complex , whitehead s theorem implies that @xmath171 is a homotopy equivalence . we will now show that the rank of @xmath454 is infinite . the proof is not constructive , but in the next section we give a way of producing reasonably explicit elements . + + we begin with the following observation . since @xmath183 covers @xmath140 with covering group @xmath299 , there is a hurewicz fibration @xmath524 and therefore a spectral sequence @xmath525 converging to the homology of @xmath140 . by the computation of the homology of @xmath183 , the @xmath526 page of this spectral sequence is rather sparse . it has only two nonzero rows , the relevant parts of which are @xmath527 where @xmath528 . this spectral sequence degenerates at the @xmath529 page . consequently , there is an isomorphism @xmath530 where @xmath531 . since @xmath532 is finite - dimensional , the group @xmath454 has infinite rank if @xmath533 does . we will demonstrate that @xmath533 does not have finite rank by showing that it surjects onto a free abelian group of infinite rank . + + the long exact sequence of integral homology for the pair @xmath534 has a segment fitting into a commutative diagram @xmath535 & & \\ h_2({{\mathcal{c } } } ) \ar[r ] & h_2({{\mathcal{c } } } , { { \mathcal{c}}}-{{\mathcal{c}}}^{red } ) \ar[r]\ar[d ] & h_1({{\mathcal{c}}}-{{\mathcal{c}}}^{red})\ar[r]\ar@{^{}->>}[d ] & h_1({{\mathcal{c } } } ) \ar[r ] & 0 \\ & h_2({\mathfrak{h}}_2,{\mathfrak{h}}_2-{\mathfrak{h}}_2^{red } ) \ar[r]^{\ \ \cong } & h_1({\mathfrak{h}}_2-{\mathfrak{h}}_2^{red } ) & & } \ ] ] where @xmath536 is any smooth fiber . with the aid of this sequence , one proves the sequence @xmath537 is exact . this is a diagram chase . the crucial observation is that the composition @xmath538 is an isomorphism . our aim is to show that the projection @xmath539 has a kernel of infinite rank . we will make use of an integral basis for @xmath540 . to construct it , arbitrarily label the two components of @xmath541 with a 1 " and a 2 " , respectively ; denote these components by @xmath542 and @xmath543 . then it is possible to choose a sufficiently small disc @xmath266 which is transverse to @xmath211 at a single point such that there are two holomorphic sections @xmath544 with the property that @xmath545 intersects @xmath546 transversely in a single point . regarded as elements of @xmath540 , the sections @xmath547 form an integral basis for @xmath540 . at the same time , the discs @xmath266 form a basis for @xmath548 . it is clear that the image of @xmath545 in @xmath548 equal to @xmath266 for both choices of @xmath257 . thus the kernel of the projection @xmath549 is freely generated by the differences @xmath550 . this yields [ prop34 ] the group @xmath533 has infinite rank and , as a consequence , the group @xmath454 has infinite rank . it follows from mess s description @xcite of the monodromy of @xmath271 that any oriented sscc @xmath551 on a smooth fiber @xmath497 of @xmath271 is freely homotopic in @xmath271 to an oriented vanishing cycle @xmath552 in one of the subfamilies @xmath553 . the precise value of the index @xmath38 depends on the homology splitting of @xmath497 induced by @xmath551 . lifting to @xmath340 , this implies that any lift @xmath554 of @xmath551 to @xmath340 is freely homotopic to a lift @xmath555 of the vanishing cycle @xmath552 . assuming that @xmath266 is sufficiently small , one shows that @xmath555 is homologous in @xmath340 to a 1-cycle of the form @xmath556 , where @xmath557 denotes the lift of a section @xmath558 to @xmath500 and @xmath559 denotes its restriction to the boundary @xmath411 of @xmath266 . it follows that the image of the homology class of @xmath560 in @xmath561 is equal to the image of @xmath562 under the boundary map @xmath563 . certainly @xmath564 is a non - zero . now suppose we have a collection of pointed , oriented ssccs @xmath565 on @xmath497 , all with the same basepoint @xmath566 . further suppose the @xmath567 distinct homology splittings on @xmath497 and that the product @xmath568 is trivial in @xmath569 . let @xmath570 be the set of of pointed lifts of the @xmath567 to @xmath571 with basepoint @xmath572 . then there exists a non - zero element @xmath573 with the property that @xmath574 = 0\in h_1({\widetilde{\mathcal{c}}}-{\widetilde{\mathcal{c}}}^{red})$ ] . here @xmath575 indexes the component of @xmath186 corresponding to the homology splitting induced by @xmath567 , and @xmath576 $ ] denotes the homology class of @xmath577 in @xmath340 . + + very shortly , we will argue that there exist non - empty collections @xmath578 of such ssccs . in order to do so , we work with the standard reference surface @xmath37 . let @xmath579 denote the generating set for @xmath580 pictured below . the crucial tool here is the hall - witt identity ( see @xcite , p. 290 ) , which states that , for any elements @xmath581 in a group @xmath582 , @xmath583 = [ x , y]\cdot [ x , z]^{y^{-1 } } \ \ \ \ \ alternatively}\ \ \ \ \ [ xy , z ] = [ y , z]^{x^{-1}}[x , z]\ ] ] where @xmath584 = aba^{-1}b^{-1}$ ] and @xmath585 . [ lemma34 ] the commutators @xmath586 , [ b_1^{-1},a_2a_1]^{b_1^{-1}a_1^{-1}b_1^{-1 } } , [ a_1,b_1]$ ] and @xmath587^{b_1^{-1}}$ ] are represented by oriented ssccs on @xmath37 . they induce homology splittings that do not all coincide , and there is a multiplicative relation between them of the form @xmath588 = [ a_1,b_1][b_1^{-1},a_2a_1]^{b_1^{-1}a_1^{-1}b_1^{-1 } } [ a_1b_1,a_2a_1]^{b_1^{-1}}.\ ] ] it is easily seen that the homology splittingss appearing in figure 2 are all distinct . finally we verify the commutator identity in the statement of the lemma . it is a striaghtforward application of the hall - witt identities . we remark that the proof applies equally well in any group . start with the separating simple closed curve @xmath589 $ ] and expand using hall - witt : @xmath586 = [ a_1,b_1]\cdot [ a_1,a_2]^{b_1^{-1}}$ ] . using the identity @xmath590 = [ x , yx]$ ] we may write @xmath591 = [ a_1,a_2a_1]= [ a_1b_1b_1^{-1},a_2a_1 ] = [ b_1^{-1},a_2a_1]^{b_1^{-1}a_1^{-1}}\cdot [ a_1b_1,a_2a_1]\ ] ] so that finally we have @xmath586 = [ a_1,b_1]\cdot [ b_1^{-1},a_2a_1]^{b_1^{-1}a_1^{-1}b_1^{-1}}\cdot [ a_1b_1,a_2a_1]^{b_1^{-1}}$ ] . enrico arbarello , maurizio cornalba , phillip griffiths , and joseph daniel harris : geometry of algebraic curves : volume ii with a contribution by joseph daniel harris , 2011 , springer science & business media

we study the thermodynamical properties of compressed baryonic matter with strangeness within non - relativistic energy density functional models with a particular emphasis on possible phase transitions found earlier for a simple-mixture . the aim of the paper is twofold : i ) examining the phase structure of the complete system , including the full baryonic octet and ii ) testing the sensitivity of the results to the model parameters . we find that , associated to the onset of the different hyperonic families , up to three separate strangeness - driven phase transitions may occur . consequently , a large fraction of the baryonic density domain is covered by phase coexistence with potential relevance for ( proto)-neutron star evolution . it is shown that the presence of a phase transition is compatible both with the observational constraint on the maximal neutron star mass , and with the present experimental information on hypernuclei . in particular we show that two solar mass neutron stars are compatible with important hyperon content . still , the parameter space is too large to give a definitive conclusion of the possible occurrence of a strangeness driven phase transition , and further constraints from multiple - hyperon nuclei and/or hyperon diffusion data are needed .

1406.0395

in the effort of building more realistic equations of state ( eos ) on which the understanding of astrophysical issues as the structure and evolution of neutron stars ( ns ) or core - collapsing supernovae ( ccsn ) relies , special attention is presently paid to the behavior of baryonic matter at densities above nuclear matter saturation density . the subject is challenging as experimental data are too scarce to satisfactorily constrain the respective interactions , in particular if non - nucleonic degrees of freedom are involved . though , simple energetic arguments show that no reliable description can be conceived without considering strangeness @xcite . as such it is hoped that astrophysical observations can eventually supplement the missing knowledge so far attained in terrestrial laboratories . an example in this sense is the present debate about the measurement of very massive neutron stars , and the associated core composition . the early conclusions ruling out hyperons from the ns core seem to be refuted by recent relativistic and non - relativistic mean - field models showing that a sufficiently repulsive hyperon - nucleon ( @xmath1-@xmath2 ) and hyperon - hyperon ( @xmath1-@xmath1 ) interaction at high densities is able to reconcile the two solar mass measurements corresponding to psr j1614 - 2230 @xcite and psr j0348 + 0432 @xcite with the onset of strangeness @xcite without necessarily a very early deconfinement transition circumventing the hyperon puzzle @xcite . the presence of hyperons in dense stellar matter is expected to have important astrophysical consequences . we can recall for instance the modification of the neutron star cooling rate due to hyperonic urca processes @xcite leading to very fast cooling for stars with a mass high enough to allow for the onset of hyperons , a result , however , very sensitive to hyperonic pairing @xcite , and thus subject to large uncertainties . by allowing for weak non - leptonic reactions ( @xmath3 , @xmath4 ) , direct and modified hyperonic urca and strong interactions ( @xmath5 , @xmath6 ) hyperons are also shown to impact on bulk viscosity and , thus , damp r - mode instabilities @xcite . most of the predictions are done within mean - field models . however , because of generic attractive and repulsive couplings between the different baryonic species , phase transitions could , in principle , be faced . an example is the liquid - gas phase transition occurring in nuclear matter . if there is a phase transition the mean - field solutions should be replaced by the gibbs construction in the phase coexisting domains , thus modifying the equation of state . the occurrence of a phase transition in strange compressed baryonic matter has already been discussed in ref . @xcite , where a new family of neutron stars characterized by much smaller radii than usually considered was predicted . however , very attractive hyperon - hyperon couplings were considered in that study , which presently appear ruled out by the experimental information on the ground state energy of double - lambda hypernuclei . a detailed study of the phase diagram of dense baryonic matter was recently undertaken in ref . @xcite within a non - relativistic mean - field model based on phenomenological functionals . the models in refs . @xcite considered a simplified setup , taking only @xmath7 @xcite and @xmath8 @xcite baryon mixtures into account . it was shown that under these assumptions first- and second- order phase transitions exist , and are expected to be explored under the strangeness equilibrium condition characteristic of stellar matter . two astrophysically relevant consequences have been worked out . in ref . @xcite it has been demonstrated that in the vicinity of critical points the neutrino mean - free path is dramatically reduced , such that the neutrino transport can be considerably affected . ref . @xcite shows within a spherical simulation that , if during the proto - neutron star contraction after bounce the phase coexistence region is reached , a mini - collapse is induced , leading to pronounced oscillations of the proto - neutron star . our previous work suffers nevertheless of two major limitations . first , we considered the strangeness degree of freedom as fully exhausted by the @xmath9 hyperon . if the hyperonic couplings are such that @xmath10 or @xmath11 are more abundant than @xmath9 , as predicted by a series of models , the extension and localization of phase coexistence domains will be different . moreover , the possible dominance of a charged hyperon would impact on the direction of the order parameter and change the stability of the phase diagram with respect to the electron gas . in the extreme case , it could even make it disappear for neutron star matter . secondly , the previous study @xcite employs only one specific model , a phenomenological energy density functional , with one parameter set , assuming in particular a strongly attractive hyperon - hyperon interaction . the most recent analysis @xcite of double-@xmath9 hypernuclei now tend to suggest a very small attraction in the @xmath9-@xmath9 channel , though experimental data are still very scarce and extrapolations to infinite matter uncertain . moreover accurate fits exist of microscopic brueckner calculations @xcite providing functionals which are consistent with the available experimental data on nucleon-@xmath9 phase shifts . this can give some guidance on the functional in the nucleon-@xmath9 channel , although it is known that such models fail to reproduce the existence of very massive neutron stars . the aim of the present paper is thus twofold . firstly , to investigate the phase diagram of the whole baryonic octet . secondly , to test the sensitivity of the strangeness - driven phase transition on the hyperonic coupling constants . both @xmath12 and @xmath13 channels will be considered . since most experimental constraints concern the @xmath9-hyperon , to avoid proliferation of uncontrolled parameters , we will consider for this study only the simplest mixture @xmath14 which accounts for the three relevant densities : baryonic , charge / leptonic and strangeness . the paper is organized as follows . section [ section : themodel ] briefly presents the model and the phenomenology of the phase transition . the phase diagram of purely baryonic , as well as charge - neutral baryonic matter with leptons are spotted in sections [ section : uncharge ] and [ section : neutral ] . the model dependence of the results is analyzed in section [ section : modeldep ] by considering alternative density functionals for the @xmath2-@xmath1 and @xmath1-@xmath1-interactions and various values for the coupling constants . conclusions are drawn in section [ section : concl ] . in non - relativistic mean - field models , the total energy density of homogeneous baryonic matter is given by the sum of mass and kinetic energy densities of different particle species and the potential energy density : @xmath15 where @xmath16 ; @xmath17 ; @xmath18 represent the baryon , strangeness and charge number densities , respectively , corresponding to the three good quantum numbers assuming equilibrium with respect to strong interaction . the particle and kinetic energy densities can be expressed in terms of fermi - dirac integrals , @xmath19 with @xmath20 . @xmath21 , @xmath22 and @xmath23 denote , respectively , the inverse temperature , the effective @xmath24-particle mass and the effective chemical potential of the @xmath24-species self - defined by the single - particle density . the effective chemical potentials are related to the chemical potentials @xmath25 via @xmath26 where @xmath27 are the self - consistent mean field single - particle potentials . the potential energy density should in principle account for all possible couplings between nucleonic and hyperonic species , @xmath2-@xmath2 , @xmath2-@xmath1 and @xmath1-@xmath1 . the nuclear structure data constrain satisfactorily the @xmath2-@xmath2-interaction up to densities close to the normal nuclear saturation density and moderate isospin asymmetries , such that well constrained and reliable expressions for this functional , including isospin dependent effective masses and currents are available . the situation is much less clear for higher densities , strong isospin asymmetries as well as for channels containing hyperons . the most general expression of these potential energies can be expanded in a polynomial form @xmath28 as a guideline to characterize the couplings , the single particle potentials of baryon @xmath29 in pure @xmath30-matter are employed : @xmath31 . the coupling constants @xmath32 can then be adjusted to reproduce standard values of these potentials at some reference density , obtained within a ( model dependent ) analysis of the available experimental data . a skyrme - like expression has been frequently employed for the energy density , where the contribution of channel @xmath33 to the potential energy density is given by @xmath34 this form , which depends on only three parameters for each channel , is the simplest expression which corresponds to a controlled compressibility and fulfills the condition that @xmath35 vanishes at vanishing @xmath30-density @xmath36 , and becomes highly repulsive at @xmath30-high density @xmath37 . let us notice that this simple ( and probably simplistic ! ) form together with the fact that we fix it at one given density implies a coupling between short- and long - range behaviors of @xmath38 and @xmath39 potentials , @xmath40 , @xmath41 . concerning the channels including strangeness , the available experimental information is particularly scarce . hypernuclei experiments only provide information on @xmath42 , @xmath43 and @xmath44 potential well depths in symmetric nuclear matter at saturation densities and , to a less accurate extent , on the @xmath9-@xmath9 interaction potential . based on a wealth of @xmath9-hypernuclear data produced in @xmath45 reactions , the presently accepted value of @xmath46 is considered to be @xmath47 mev @xcite . @xmath48 is known to be attractive , too , @xmath49 mev , based on missing mass measurements in the @xmath50 reaction on carbon @xcite . the situation of @xmath51 is ambiguous . on the one hand @xmath52 reactions on medium - to - heavy nuclei point to a repulsive potential of the order of 100 mev or less @xcite . on the other hand , the observation of a @xmath53he bound state in a @xmath54he(@xmath55 ) reaction @xcite pleads in favor of an attractive potential . very few multi - hyperon exotic nuclei data exist so far and all of them correspond to double-@xmath9 light nuclei . the @xmath9-@xmath9 bond energy can be estimated from the binding energy difference between double-@xmath9 and single-@xmath9 hypernuclei , @xmath56 where @xmath57 measured bond energies are affected by huge error bars . double - lambda @xmath58be and @xmath59b data suggest @xmath60 mev @xcite while @xmath61he data point toward a lower value @xmath62 mev @xcite . the bond energy can be interpreted as a rough estimation of the @xmath63 potential at the average @xmath9 density @xmath64 inside the hypernucleus @xcite . the extrapolation of few body systems binding energies into mean - field quantities is , nevertheless , problematic . in this sense , the more attractive values extracted from larger nuclei could appear more appealing for the present application . though , it is presently considered that the most accurate experimental data correspond to the nagara event with a bond energy of@xmath65 mev . in this work we shall therefore consider the latter value , @xmath66 mev . in what regards the average @xmath9-density in light nuclei we shall use the value proposed in ref . @xcite , @xmath67 . it is clear that fixing the standard value of the potential at one specific density only constrains one parameter of the generic @xmath29-@xmath30-interaction . this shows that any phenomenological mean - field parameterization is subject to large uncertainties . in our previous studies @xcite we have employed the skyrme - based f unctional by balberg and gal @xcite , @xmath68 where @xmath69 are the isoscalar densities for nucleons , and @xmath9- , @xmath70- and @xmath71-hyperons . @xmath72 stands for the respective iso - vector densities and the values of @xmath73 are chosen identical for any @xmath74 for simplicity . as one may notice , the same functional form is employed in all channels and the potential energy proposed by eq . ( [ eq : epot_bg ] ) deviates from the simple polynomial form of eq . ( [ eq : epot_cc ] ) truncated at low order because of the @xmath75-dependence of the short - range term . the expressions of the single - particle potentials of baryon @xmath29 in pure @xmath29- and , respectively , @xmath30-matter are @xmath76 they show that fixing the potentials , introduces a correlation between the parameters governing repulsion and those governing attraction not present in the generic form of eq . ( [ eq : epot_pol ] ) . as we will see , such a correlation is also present in bhf microscopically based energy functionals proposed in ref . @xcite . a deeper anlaysis with a larger number of models would , however , be necessary to check whether this particularity possibly affects the generality of our results . a study in this sense with relativistic models is in progress @xcite . @xcite proposes three sets of parameters corresponding to different stiffnesses @xmath77 . for the sake of simplicity , a unique value is assumed for @xmath78 and @xmath79 , @xmath80 . in ref . @xcite , the stiffest interaction proposed in ref . @xcite , bgi , has been used . it is characterized by the values @xmath81 , @xmath82= - 784 mev @xmath83 , @xmath84=214.2 mev @xmath83 , @xmath85=1936 mev @xmath86 , @xmath87= - 340 mev @xmath83 , @xmath88=1087.5 mev @xmath89 , @xmath90= - 340 mev @xmath83 , @xmath91=214.2 mev @xmath92 , @xmath93=1087.5 mev @xmath89 , @xmath94= - 291.5 mev @xmath83 , @xmath95=0 , @xmath96=932.5 mev @xmath89 , @xmath97= - 486.2 mev @xmath83 , @xmath98=0 , @xmath99=0 , @xmath100=428.4 mev @xmath92 , @xmath101=1553.6 mev @xmath89 and leads to the following values of the different interaction potential depths in symmetric matter at normal nuclear saturation density : @xmath102 mev , @xmath103 mev , @xmath104=-19.4 mev , @xmath105= - 38 mev . for the reference value @xmath106 bgi provides @xmath107=-12.8 mev , @xmath108= -13.6 mev , meaning that it is too attractive than present double hypernuclei data indicate . we remind that out of parameter sets proposed in ref . @xcite , bgi produces the highest neutron star maximum mass : the maximum mass exceeds @xmath109 if only @xmath9 s are considered . however , if the full octet is accounted for , the maximum neutron star mass becomes too low , possiby due to a not sufficiently repulsive @xmath1-@xmath1 interaction at high densities . though , as one may notice by comparison with experimental data , the @xmath110 , and , probably , @xmath111 interactions are too attractive at low densitites . the same is true for the other two potentials proposed in ref . @xcite , too . it is interesting to notice that bg - like parameterizations in agreement with hypernuclei experimental data and able to reach two solar mass neutron stars have been proposed in ref . for the sake of easier comparison , to see how the phase diagram evaluated in @xcite evolves when the whole baryonic octet is accounted for , we will however stick to the original bgi in this paper . to see to what extent the existence of a strangeness driven phase transition is conditioned by the poorly - constrained @xmath1-@xmath1-interaction , in section [ section : modeldep ] we will return to the simple case of a @xmath112-mixture . we will compare the balberg and gal parameterization with an energy density functional , where the @xmath2-@xmath2 and @xmath9-@xmath2 interactions have been fitted to a microscopic brueckner - hartree - fock calculation ( bsl ) @xcite and vary the parameters of the @xmath9-@xmath9 interaction in both models . in particular we will show that qualitatively , in large regions of parameter space , the phase diagram does not depend on the exact parametrization employed . the phase diagram of a @xmath113-component system is , at constant temperature , a @xmath113-dimensional volume . the frontiers of the phase coexistence domain(s ) , @xmath114 , are determined by the @xmath115 conditions of thermodynamic equilibrium between @xmath116 different phases , @xmath117 where @xmath118 , @xmath119 and @xmath120 stand for the free energy density , entropy density and , respectively , pressure . for a system to present a phase coexistence , its mean - field solutions should be more expensive in terms of free energy than the state mixing given by eqs . ( [ eq : equil ] ) . mathematically , this is equivalent to the presence of a convexity anomaly of the thermodynamic potential in the density hyperspace . the number of coexisting phases is determined by the number of order parameters or , in terms of local properties , the number of spinodal instability directions . the last quantity is equal to the number of negative eigenvalues @xmath121 of the free energy curvature matrix @xmath122 , such that @xmath123 . the problem of phase coexistence in a @xmath113-component system can by reduced to a problem of phase coexistence in a one - component system by legendre transforming the thermodynamical potential @xmath124 with respect to the remaining @xmath125-chemical potentials @xcite . under the condition of equilibrium with respect to the strong interaction , baryonic matter is a three - component system with the densities @xmath126 as degrees of freedom . it is important to remark here that the use of @xmath127 as a good quantum number does not imply that strangeness density @xmath128 is conserved . in particular , along the strangeness equilibrium trajectory @xmath129 considered in this study @xmath128 obviously varies . this reduces the dimensionality of the phase space from 8(=number of species ) to 3 . to further reduce the dimensionality for studying phase coexistence , one may then perform the legendre transformation with respect to any set @xmath130 , @xmath131 and @xmath132 . formally the three choices should be considered in order to investigate all possible phase separation directions , as required by a complete study . within the simpler @xmath133 system studied in ref . @xcite , which has the same dimensionality as the full octet , we found that the order parameter is always one dimensional . this means that a single legendre transformation is enough to spot the thermodynamics provided that the order parameter is not orthogonal to the controlled density . the most convenient framework to easily access the physical trajectories is the one controlling the @xmath134-density : @xmath135 the coexisting phases , if any , will then be characterized by equal values of @xmath136 and @xmath120 and the phase instability regions will be characterized by a back - bending behavior of @xmath137 . within this section we will analyze the phase diagram of pure baryonic matter with strangeness following the lines exposed in the previous section . we employ the balberg and gal energy density @xcite , parameterization bgi , see section [ section : themodel ] . the upper panel of fig . [ fig : mu - rho ] illustrates the evolution of the baryonic chemical potential as a function of baryonic density at constant values of @xmath129 , @xmath139 and @xmath140=1 mev while the bottom panel depicts the abundances of different nucleonic and hyperonic species . three back - bending regions in @xmath141 exist . we can see that each back - bending is strictly correlated with the onset of new species , and a decrease of abundances of species already present . upon choosing @xmath129 , @xmath139 , the hyperonic production thresholds are exclusively determined by the particle s rest mass and the interaction potentials . since within the bgi parameterization , the @xmath1-@xmath1 and @xmath1-@xmath142 interaction depends only very weakly on the particular channel , the rest mass effects dominate . the first strange particle to appear , at about @xmath143 is therefore the less massive one , @xmath9 . the three quasi - degenerate @xmath70-particles whose masses are 74 mev higher than the @xmath9-mass , are produced starting from about @xmath144 . the most massive hyperons , the @xmath71-particles , are the last to be created , at about @xmath145 . at high densities hyperons become more abundant than nucleons . this shows that having accurate @xmath1-@xmath1 , @xmath1-@xmath142 , and @xmath2-@xmath1 interactions is as important as having reliable nucleonic ones . investigation of @xmath146 and @xmath147 ( not shown ) confirms that any back - bending can be cured by a maxwell construction and that the linear combination of stable phases has a lower value for @xmath148 than the mean - field solution , and corresponds thus to the energetically favored solution . this means that three distinct phase coexistence regions exist , induced by the onset of each hyperonic family . and @xmath139 at @xmath140=1 mev , employing the bgi parameterization @xcite . , title="fig : " ] and @xmath139 at @xmath140=1 mev , employing the bgi parameterization @xcite . , title="fig : " ] -system under strangeness equilibrium at @xmath140=1 mev as provided by the bgi parameterization @xcite in the @xmath134-@xmath149 ( a ) and @xmath128-@xmath149 ( b ) planes . the arrows indicate the directions of phase separation.,title="fig : " ] -system under strangeness equilibrium at @xmath140=1 mev as provided by the bgi parameterization @xcite in the @xmath134-@xmath149 ( a ) and @xmath128-@xmath149 ( b ) planes . the arrows indicate the directions of phase separation.,title="fig : " ] different thermodynamical conditions , i.e. different values of @xmath131 and @xmath140 , will obviously change particle production thresholds , abundances , and the location of phase coexistence regions . by correspondingly changing the values of @xmath131 , the whole 3-dimensional phase diagram for a given temperature can be explored . considering that in most astrophysically relevant situations the system is in equilibrium with respect to weak strangeness changing interactions , the most physically relevant part of the phase diagram is the cut corresponding to @xmath129 , which will be the only one considered within this work . the projections of the phase diagram of the @xmath138 mixture at the arbitrary temperature of 1 mev to the @xmath134-@xmath149 ( panel ( a ) ) and the @xmath128-@xmath149-plane ( panel ( b ) ) are represented in fig . [ fig : phd ] . the arrows indicate the directions of phase separation . roughly speaking , two large phase coexistence domains exist : the first one corresponds to the appearance of @xmath9- and @xmath70-hyperons , while the second one is due to cascades . they are well separated and extend over a significant total baryonic density range . at moderate values of @xmath150 , where particle production is mainly dictated by the rest mass , the thresholds for @xmath9- and @xmath70-hyperons are pulled apart , and the phase coexistence regions corresponding to the their respective onsets actually split up , as previously observed for @xmath151 , see fig . [ fig : mu - rho ] . at more important and negative @xmath150-values negatively charged particles are favored and consequently the @xmath10-threshold is shifted to lower densities and that for @xmath152 to higher ones . upon increasing the absolute value of @xmath150 finally the phase coexistence region triggered by the onset of @xmath70-hyperons merges with that for @xmath9-hyperons . the same happens for positive values of @xmath150 , but with the roles of @xmath10 and @xmath152 interchanged . the direction of the order parameter is not constant over the phase coexistence region . the phase transition induced by @xmath9-hyperons is always characterized by a very small component of the order parameter along @xmath149 , as the transition is mainly triggered by neutral @xmath9-hyperons , as already emphasized in ref . the @xmath70-induced phase transition has a small component along @xmath149 when the global @xmath70-charge is small , that is at low @xmath150-values , and a significant component at high @xmath150-values , i.e. for a high total @xmath70-charge . the @xmath71-induced phase transition has an order parameter with important contribution along @xmath149 whenever both @xmath153 and @xmath11 are created as their total charge can not vanish . at high-@xmath150-values the @xmath11 production threshold is beyond the density domain considered for this study such that only @xmath153 exist and consequently the charge dependence of the order parameter becomes very weak . the phenomenology of baryonic matter , as the one considered above , is purely academic . what is pertinent from the physical point of view is the phenomenology of electrically neutral matter , where the baryonic charge is compensated by leptonic charge . the net charge neutrality is a pre - requisite condition for the thermodynamic limit to exist and corresponds to matter that constitutes compact objects where baryons exist together with leptons and photons . it is commonly accepted that the different sectors are in thermal and chemical equilibrium with respect to strong and electromagnetic interactions . chemical equilibrium with respect to weak interactions can be satisfied or not depending on how fast the considered astrophysical system evolves compared with weak interaction rates . as such , @xmath155-equilibrium is reached in neutron stars while core - collapsing supernovae typically evolve out of @xmath155-equilibrium . to be as general as possible for the moment we shall not assume @xmath155-equilibrium . as mentioned before , we will , however , assume equilibrium with respect to strangeness changing weak interactions . in the mean - field approximation , the total thermodynamic potential can be written as the sum of a baryonic , leptonic and photonic contribution , @xmath156 . leptons and photons are well described by fermionic and , respectively , bosonic ideal gases @xcite . the introduction of leptonic degrees of freedom does not increase the dimensionality of the problem @xcite because the strict charge neutrality condition @xmath157 , imposed by thermodynamics , fixes @xmath149 in terms of leptonic density . thus the charge degree of freedom is removed and the associated chemical potential , @xmath150 becomes ill - defined . within this work , @xmath158 . the effect of other leptons , in particular muons , is considered beyond the scope of the present work and disregarded . technically , the only modification with respect to the analysis in the case of pure baryonic matter discussed in the previous section is the replacement of the charge density with the ( electron ) leptonic one . adding an ideal gas contribution to the free energy might change the convexity , i.e. the stability of the system . indeed , the thermodynamics of charge neutral matter can deeply differ from that of pure baryonic matter . as an example , the liquid - gas ( lg ) phase transition taking place in nuclear matter at sub - saturation densities is strongly quenched @xcite by the presence of electrons via the charge neutrality condition . this is due to the fact that a first order ( lg ) transition is associated with a macroscopic density fluctuation in direction of the order parameter . in the case of the nuclear lg transition , the order parameter has a strong component in charge direction , implying a macroscopic charge density fluctuation . this fluctuation is , however , strongly suppressed by the high incompressibility of the electron gas . @xcite shows that , for the @xmath8 system , the strangeness - driven phase transition is essentially not affected by the electrons . this is not surprising because @xmath9 fluctuations are poorly correlated to the electric charge , see previous section , too . the situation is different here , because of the presence of charged hyperons . as discussed before , we can see in fig . [ fig : phd ] that the order parameter has a significant component along the charge density especially for the @xmath71-induced transition at low @xmath150-values , and it is in this domain that we expect the most dramatic alteration of the phase coexistence region . the phase diagram of the @xmath154-system under strangeness equilibrium at @xmath140=1 mev is displayed in fig . [ fig : phd_+e ] in the plane @xmath134-@xmath159 . as before , the arrows mark the directions of the order parameter . the pattern of the phase diagram is roughly the same as for pure baryonic matter : depending on @xmath160 , @xmath9- and @xmath70-hyperons are responsible for one or two phase transitions which extend over @xmath161 @xmath162 and a @xmath71-induced phase transition occurs at higher baryonic densities . the most important shrinking of the phase coexistence is obtained at low @xmath159-values . the direction of phase separation is rotated in the sense that its component along @xmath159 gets smaller , which is expected since large electron density fluctuations are effectively suppressed . -system under strangeness equilibrium at @xmath140=1 mev as provided by bgi parameterization @xcite in the @xmath134-@xmath159-plane . the dotted curve marks the path corresponding to @xmath155-equilibrium . ] the predictions of a phenomenological density - functional model depend dramatically on the functional form assumed for the energy density and the employed values of the coupling constants . as discussed in section [ section : themodel ] , the functional form of the energy density in a non - relativistic phenomenological model is subject to large arbitrariness . the same is true for the coupling constants as the experimental data ( a ) correspond exclusively to low matter density , ( b ) are insufficient to constrain all the parameters of the potential energy functional and ( c ) are often subject to large uncertainties , especially for the @xmath1-@xmath163 channels . as a consequence , instead of one particular functional with one parameter set , one should rather consider different parameter sets and functional forms , satisfying the experimental conditions . for this reason , we will first examine the correlation between the existence of the phase transition and the parameters of the skyrme - based bg @xcite energy density functional . to avoid proliferation of unconstrained parameters , the issue is considered in the simple case of a @xmath8 mixture , which nevertheless satisfies the basic requirement of accounting for all relevant degrees of freedom , @xmath164 , @xmath127 , and @xmath165 the two interaction channels which can be responsible of the phase transition are the @xmath2-@xmath9 and the @xmath9-@xmath9 one . since the @xmath1-@xmath1 interactions are poorly known , we first consider the extreme situation where the @xmath9-@xmath9 coupling is completely absent . the parameters of the @xmath2-@xmath9-channel , @xmath166 , @xmath87 and @xmath88 , are considered as free variables which have to satisfy the unique condition @xmath167 mev , keeping for simplicity the reference value of bgi . we consider @xmath168 , @xmath169 mev @xmath170 mev @xmath83 and , in each case , calculate @xmath171 . the nuclear part remains the same as for bgi . -@xmath9-interaction : ( a ) limiting values of the coupling constant @xmath172 for which , at different @xmath166 , the symmetric @xmath112-system at @xmath140=0 manifests strangeness driven phase transition along @xmath129 . the star marks the @xmath173 values corresponding to the bsl @xmath12-interaction ( see text ) . ( b ) nucleonic density dependence of the @xmath9 potential in uniform symmetric nuclear matter @xmath174 for different @xmath175,@xmath176[mev @xmath177)$ ] sets : ( 1.72,-900 ) , ( 1.72,-300 ) , ( 2,-900 ) , ( 2,-300 ) , ( 3,-900 ) and ( 3,-300 ) . @xmath178 is determined such as @xmath46=-26.6 mev @xcite , see text ; ( c ) neutron star mass as a function of central baryon density for the @xmath175,@xmath179 sets considered in ( b ) ; ( d ) @xmath9 relative abundances as a function of normalized distance from the star center for the maximum mass configuration and the @xmath175,@xmath179 sets considered in ( b).,title="fig : " ] -@xmath9-interaction : ( a ) limiting values of the coupling constant @xmath172 for which , at different @xmath166 , the symmetric @xmath112-system at @xmath140=0 manifests strangeness driven phase transition along @xmath129 . the star marks the @xmath173 values corresponding to the bsl @xmath12-interaction ( see text ) . ( b ) nucleonic density dependence of the @xmath9 potential in uniform symmetric nuclear matter @xmath174 for different @xmath175,@xmath176[mev @xmath177)$ ] sets : ( 1.72,-900 ) , ( 1.72,-300 ) , ( 2,-900 ) , ( 2,-300 ) , ( 3,-900 ) and ( 3,-300 ) . @xmath178 is determined such as @xmath46=-26.6 mev @xcite , see text ; ( c ) neutron star mass as a function of central baryon density for the @xmath175,@xmath179 sets considered in ( b ) ; ( d ) @xmath9 relative abundances as a function of normalized distance from the star center for the maximum mass configuration and the @xmath175,@xmath179 sets considered in ( b).,title="fig : " ] -@xmath9-interaction : ( a ) limiting values of the coupling constant @xmath172 for which , at different @xmath166 , the symmetric @xmath112-system at @xmath140=0 manifests strangeness driven phase transition along @xmath129 . the star marks the @xmath173 values corresponding to the bsl @xmath12-interaction ( see text ) . ( b ) nucleonic density dependence of the @xmath9 potential in uniform symmetric nuclear matter @xmath174 for different @xmath175,@xmath176[mev @xmath177)$ ] sets : ( 1.72,-900 ) , ( 1.72,-300 ) , ( 2,-900 ) , ( 2,-300 ) , ( 3,-900 ) and ( 3,-300 ) . @xmath178 is determined such as @xmath46=-26.6 mev @xcite , see text ; ( c ) neutron star mass as a function of central baryon density for the @xmath175,@xmath179 sets considered in ( b ) ; ( d ) @xmath9 relative abundances as a function of normalized distance from the star center for the maximum mass configuration and the @xmath175,@xmath179 sets considered in ( b).,title="fig : " ] -@xmath9-interaction : ( a ) limiting values of the coupling constant @xmath172 for which , at different @xmath166 , the symmetric @xmath112-system at @xmath140=0 manifests strangeness driven phase transition along @xmath129 . the star marks the @xmath173 values corresponding to the bsl @xmath12-interaction ( see text ) . ( b ) nucleonic density dependence of the @xmath9 potential in uniform symmetric nuclear matter @xmath174 for different @xmath175,@xmath176[mev @xmath177)$ ] sets : ( 1.72,-900 ) , ( 1.72,-300 ) , ( 2,-900 ) , ( 2,-300 ) , ( 3,-900 ) and ( 3,-300 ) . @xmath178 is determined such as @xmath46=-26.6 mev @xcite , see text ; ( c ) neutron star mass as a function of central baryon density for the @xmath175,@xmath179 sets considered in ( b ) ; ( d ) @xmath9 relative abundances as a function of normalized distance from the star center for the maximum mass configuration and the @xmath175,@xmath179 sets considered in ( b).,title="fig : " ] the upper panel of fig . [ fig : instab_bg_noyy ] plots , as a function of the stiffness parameter @xmath166 , the maximum values of the coupling constant @xmath172 for which symmetric @xmath112-matter manifests phase coexistence along @xmath129 . as one may note , irrespective of @xmath166 , there is a wide range of values for the attractive @xmath1-@xmath2 coupling meaning that , in this model , phase coexistence in hyperonic matter is not conditioned by the @xmath1-@xmath1-interaction . the behavior of the @xmath9-potential in symmetric nuclear matter , @xmath180 , as a function of nucleonic density is illustrated in panel ( b ) of fig . [ fig : instab_bg_noyy ] for few representative @xmath166-values ( @xmath1811.72 , 2 and 3 ) and coupling constants situated at the extremities of the considered range ( @xmath182 -900 and -300 mev @xmath83 ) , both inside and outside the domain compatible with phase coexistence , as indicated on the figure . we can see that a wide variety of density behaviors are compatible with the presence of a phase transition . the new very precise astrophysical measurements of neutron star masses close to two solar masses @xcite represent a validity test for any astrophysical equation of state . as the rich recent literature testifies , this supplementary piece of information can neither confirm nor rule out the presence of hyperons in neutron stars . indeed , while it is true that in principle any extra degree of freedom softens the eos and , thus , lowers the maximum mass of the star , various models @xcite prove that hyperons are compatible with the two solar mass constraint . the predictions of the @xmath155-equilibrium eos at zero temperature for the neutron star mass as a function of central density obtained by solving the tolman - oppenheimer - volkoff ( tov ) @xcite equations for hydrostatic equilibrium of a spherical star , @xmath183 \left[m(r)+4 \pi r^3 \frac{p(r)}{c^2}\right ] \nonumber \\ & \cdot & \left [ 1-\frac{2 g m(r)}{c^2 r}\right]^{-1 } ; \nonumber \\ \frac{d m(r)}{d r}&=&4 \pi \epsilon(r ) r^2\end{aligned}\ ] ] are represented in panel ( c ) of fig . [ fig : instab_bg_noyy ] for the parameter sets considered in panel ( b ) . ( [ eq : coupling ] ) shows that , for the presently considered functional form of the energy density eq.([eq : epot_bg ] ) , a strong attraction at low density is associated with a strong repulsion at high density through the fixed @xmath46 . it is this peculiarity that makes possible to produce , by the most repulsive potentials , gravitational masses which largely exceed the reference @xmath184 limit and have an important @xmath9-hyperon fraction . it is , however , important to stress that these results have to be considered as qualitative , because of the artificial absence of other hyperons than @xmath9 s . the inclusion of the full octet will obviously influence the mass - radius relationship quantitatively . the bottom panel of fig . [ fig : instab_bg_noyy ] depicts , for the above considered @xmath12 interaction potentials and the maximum mass neutron star configuration , the @xmath9-relative abundances as a function of normalized distance from the star center . the reason why the curve corresponding to ( 3,-900 ) is missing is the extremely repulsive potential which prevent @xmath9s to appear . for the other interaction potentials one can see that hyperons not only exist , but they are abundant and populate most of the star s volume . the different central baryonic density values which correspond to the maximum mass configuration prevent a straightforward parallelism among the stiffness of the potential on one hand and the hyperonic relative density in the core and its extension on the other hand . this becomes obvious observing that the lowest and the highest fractions in the star core correspond to the softest considered potentials . we now turn to study the effect of the @xmath9-@xmath9 interaction , both on the existence of the phase transition and on the maximum ns mass . to keep the same framework we shall consider the original bgi parameterization for the @xmath2-@xmath9 channel , and the bg functional dependence in the @xmath9-@xmath9 channel . again , as in the case of the @xmath2-@xmath9 interaction , we vary the @xmath9-@xmath9 parameters keeping the @xmath9-potential in uniform @xmath9-matter at @xmath185 of nuclear saturation density fixed , @xmath186 , which leads to @xmath187 . the adopted value shall be @xmath188 - 0.67 mev , see the discussion in section [ section : themodel ] . we then consider different parameter sets @xmath189 in the ranges @xmath190 , @xmath191 mev @xmath192 mev @xmath83 and @xmath191 mev @xmath193 mev @xmath83 . -potential in uniform @xmath9-matter as a function of density corresponding to bgi parameterization with modified @xmath9-@xmath9 interaction : @xmath194=-900 , -480 , -300 and -200 mev @xmath83 and @xmath195 . ] abundances as a function of baryonic density along the @xmath155-equilibrium path for the effective interaction potentials considered in fig . [ fig : bgi+modifyy_ull].,title="fig : " ] abundances as a function of baryonic density along the @xmath155-equilibrium path for the effective interaction potentials considered in fig . [ fig : bgi+modifyy_ull].,title="fig : " ] adding the @xmath9-@xmath9 long - range attractive - short - range repulsive interaction , the @xmath8 toy system manifests phase coexistence in a much broader parameter range . more precisely , a strangeness - driven phase transition along the @xmath129 path is obtained for almost all considered sets . [ fig : bgi+modifyy_ull ] illustrates @xmath196 for @xmath81 and @xmath194=-900 , -480 , -300 , -200 mev . as in the case of the @xmath2-@xmath9-channel , a strong attraction leads to a steep rise at high densities and a deep minimum localized at low densities , due to the fact that we fix the value at @xmath106 and the correlation between attraction and repulsion for the bg functional form . [ fig : bgi+modif_yy_star ] depicts the predictions of these potentials for the ns gravitational mass as function of central density ( panel ( a ) ) together with the @xmath9-relative abundances as a function of baryonic density along the beta - equilibrium trajectory ( panel ( b ) ) . in all cases , for the @xmath2-@xmath9-channel the bgi parameter values have been employed . the relative ordering of the various curves is easily understandable for high central densities where the short range repulsion is effective : the stronger is the @xmath9-@xmath9 repulsion , the smaller is the relative @xmath9 density and the larger the obtained ns mass . equally predictable is the fact that small differences in the low density attractive part of the potential result in minor modifications of the @xmath9-production threshold , to a large extend dictated by the @xmath2-@xmath1 interaction . indeed , for the most attractive considered potential , @xmath9-hyperons emerge at a baryonic density only @xmath197 @xmath162 lower that than the one corresponding to the least attractive potential . the calculations presented so far were all obtained with a phenomenological non - relativistic functional , that proposed by balberg and gal @xcite , both for the @xmath2-@xmath9 and the @xmath9-@xmath9 channel . one can therefore wonder if the observed phase transition is not a pathology of the assumed and largely arbitrary functional form of the energy density . for more than fifteen years different microscopically motivated @xmath2-@xmath2 and @xmath2-@xmath1 interaction potentials have been proposed . these functionals have all been adjusted to brueckner - hartree - fock calculations hyper - nuclear matter based on different bare @xmath2-@xmath2 and @xmath2-@xmath1 interactions , and are designed to perform calculations of hypernuclei @xcite and more recently hyper - nuclear matter @xcite . in all of these potentials , @xmath1-@xmath1 interactions have been disregarded because of missing experimental constraints for the basic two - particle @xmath1-@xmath1 interaction and the difficulties observed with bhf calculations including the unsufficiently constrained bare @xmath1-@xmath1-interaction . the two parametrizations designed for hyper - nuclear matter , refs . @xcite , rely on the same energy density functional , @xmath198 where @xmath199 . they differ in the coupling constants values as the bhf calculations correspond to various treatments of the three - body forces and @xmath1-@xmath2-potentials and predict significantly different @xmath9- and @xmath10-abundances @xcite . for our application we have chosen to use the burgio - schulze - li parameterization @xcite because of its stiffer @xmath200 dependence . despite the functional dissimilarity between eqs . ( [ eq : epot_burgio ] ) and ( [ eq : epot_bg ] ) , the two parameterizations can be bridged via the @xmath9-potential in uniform symmetric nuclear matter , @xmath201 which , in both cases , is a polynomial in the baryonic density . similarly to eq . ( [ eq : coupling ] ) a correlation between short- and long - range interactions is present in eq . ( [ eq : ulnschulze ] ) . we note that eq . ( [ eq : ulnschulze ] ) can be mapped onto @xmath202 provided that @xmath203mev @xmath204 mev @xmath89 . the ( 1.72,-294.75 ) point is represented in fig . [ fig : instab_bg_noyy ] ( a ) by a star and sits outside the phase coexistence domain of a symmetric @xmath112 mixture at strangeness equilibrium . we note that these values are very close to ( @xmath205 , @xmath206 mev @xmath83 ) for which @xmath207 is depicted in fig . [ fig : instab_bg_noyy ] . -@xmath9-interaction : ( a ) limiting values of the coupling constant @xmath194 for which , at different @xmath166 , the symmetric @xmath112-system at @xmath140=0 manifests strangeness driven phase transition along @xmath129 ; @xmath208 is fixed via the condition @xmath209mev ; ( b ) @xmath9-density along the @xmath155-equilibrium path at @xmath140=0 for the original bsl potential and the cases in which bsl is supplemented with a @xmath9-@xmath9-interaction following the functional form proposed by balberg and gal @xcite ( see eq . ( [ eq : epot_bg ] ) ) and obeying the condition @xmath186=-0.67 mev , with @xmath166=2 and @xmath194=-1200 , -200 mev@xmath83 and , respectively , @xmath166=2 and @xmath194=0 and @xmath208=30000 mev@xmath89 . ( c ) gravitational mass as solution of the tov equations as a function of central baryon number density for the cases considered at ( b).,title="fig : " ] -@xmath9-interaction : ( a ) limiting values of the coupling constant @xmath194 for which , at different @xmath166 , the symmetric @xmath112-system at @xmath140=0 manifests strangeness driven phase transition along @xmath129 ; @xmath208 is fixed via the condition @xmath209mev ; ( b ) @xmath9-density along the @xmath155-equilibrium path at @xmath140=0 for the original bsl potential and the cases in which bsl is supplemented with a @xmath9-@xmath9-interaction following the functional form proposed by balberg and gal @xcite ( see eq . ( [ eq : epot_bg ] ) ) and obeying the condition @xmath186=-0.67 mev , with @xmath166=2 and @xmath194=-1200 , -200 mev@xmath83 and , respectively , @xmath166=2 and @xmath194=0 and @xmath208=30000 mev@xmath89 . ( c ) gravitational mass as solution of the tov equations as a function of central baryon number density for the cases considered at ( b).,title="fig : " ] -@xmath9-interaction : ( a ) limiting values of the coupling constant @xmath194 for which , at different @xmath166 , the symmetric @xmath112-system at @xmath140=0 manifests strangeness driven phase transition along @xmath129 ; @xmath208 is fixed via the condition @xmath209mev ; ( b ) @xmath9-density along the @xmath155-equilibrium path at @xmath140=0 for the original bsl potential and the cases in which bsl is supplemented with a @xmath9-@xmath9-interaction following the functional form proposed by balberg and gal @xcite ( see eq . ( [ eq : epot_bg ] ) ) and obeying the condition @xmath186=-0.67 mev , with @xmath166=2 and @xmath194=-1200 , -200 mev@xmath83 and , respectively , @xmath166=2 and @xmath194=0 and @xmath208=30000 mev@xmath89 . ( c ) gravitational mass as solution of the tov equations as a function of central baryon number density for the cases considered at ( b).,title="fig : " ] we have , however , to keep in mind that this functional gives an eos which is much too soft and fails to reproduce @xmath210 maximum neutron star mass @xcite . this is shown in the bottom panel of fig . [ fig : instab_schulze_yy ] , which depicts the ns mass as a function of central baryon number density . this is due to the lacking repulsion in the high density domain , meaning that probably the functional is not very reliable at the densities relevant for the phase transition . it is thus important at this point to stress that no firm conclusion can be drawn . it is certainly true that the phenomenological bg form is largely arbitrary ; however the description of the nucleon - hyperon interaction in the bhf theory can not be complete , neither . as we have already stressed , the @xmath1-@xmath1 interaction can not be neglected in hyperonic matter . it could well be the source of missing repulsion in microscopic models . due to the lack of information on this channel within microscopic calculations , for this channel we will adopt the simple polynomial form of bg , and supplement the bsl functional , eq . ( [ eq : epot_burgio ] ) , with it . the upper panel of fig . [ fig : instab_schulze_yy ] illustrates the maximum values of the coupling constant @xmath194 for which phase coexistence occurs in symmetric @xmath211 matter at various values of the stiffness parameter @xmath166 . the considered domains are @xmath212 and @xmath213 mev @xmath83 . as before , @xmath208 is obtained form the condition @xmath209mev . one can see that , in case of moderate @xmath2-@xmath1-repulsion as it is the case for bsl , a phase coexistence can still be obtained , but it requires a considerable attraction in the @xmath1-@xmath1 channel . the effect of the @xmath9-@xmath9-interaction on the ns mass - central density relation and , respectively , the @xmath9-hyperon abundances in @xmath155 equilibrium is represented in the bottom and middle panels of fig . [ fig : instab_schulze_yy ] . the two considered @xmath9-@xmath9-interactions correspond , respectively , to phase coexistence ( @xmath194=-1200 mev @xmath83 ) and stability with respect to phase separation ( @xmath194=-200 mev @xmath83 ) in symmetric @xmath112-matter . for the sake of the argument , the case of a purely repulsive and very strong @xmath214 interaction characterized by @xmath215 is considered , too . we can see that employing a strongly attractive coupling at low densities does only slightly shift the density threshold for @xmath9-production with respect to a weakly attractive coupling , but strongly enhances the equilibrium abundances just above threshold . as we have already noted several times , a strong attraction at low density is correlated to a strong repulsion at high densities , leading to a relative decrease of the @xmath9 abundances at higher densities with smaller values of @xmath216 . as a consequence , too , decreasing @xmath216 leads to an increase of the maximum ns mass . though , the 2 @xmath217 neutron star limit is not reached . in conclusion , we can say that the ad - hoc inclusion of an extra term in the bsl functional effectively accounting for the missing @xmath1-@xmath1 interaction does not solve the well - known neutron star maximum mass problem of the bhf theory . in this work , we have presented a complete study of the low temperature phase diagram of baryonic matter including hyperonic degrees of freedom within the phenomenological non - relativistic balberg and gal model @xcite . we have shown that the hyperon production thresholds are systematically associated with thermodynamic instabilities , leading to distinct first order phase transitions . these transitions can merge into a wide coexistence zone if the production thresholds of different hyperonic species are sufficiently close . as a consequence , a huge part of the phase diagram corresponds to phase coexistence between low - strangeness and high - strangeness phases . in contrast to the nuclear liquid - gas phase transition which is strongly quenched , this result is only slightly affected by adding electrons and positrons to fulfill the charge neutrality constraint . the only effect is a rotation of the direction of phase separation which reduces the electric charge density component of the order parameter . the reason is that this phase transition is driven mainly by the strangeness degree of freedom , such that the electric charge plays only a minor role . in the latter respect , we thus confirm the finding for the @xmath218-system of ref . @xcite even in the presence of charged hyperons . along the beta - equilibrium trajectory with @xmath219 the phase coexistence region corresponding to the pop up of @xmath9- and @xmath70-hyperons , as predicted by the parameterization bgi , extends over @xmath220 @xmath162 . physically this path is explored by neutron stars with untrapped neutrinos . following the study in the simple @xmath218-model in ref . @xcite , we expect that this phase coexistence region remains at higher temperatures and extends over density and lepton fraction @xmath221 domains explored by warm proto - neutron star matter . a more quantitative analysis is left for future work . the possible existence of such a phase transition is strongly conditioned by the @xmath2-@xmath1 and @xmath1-@xmath1 interaction . in the second part of the paper we have thus investigated on the one hand the dependence of the phase diagram on the interaction parameters within the phenomenological bg energy density functional and on the other hand , we have compared the results with an energy density functional based on microscopic bhf calculations by burgio , schulze and li @xcite . a complete parameter study of the energy functional would be very cumbersome and not very illuminating , because of the huge number of insufficiently constrained couplings . we have therefore considered the simplified situation of nuclear matter with @xmath9-hyperons . the phase diagram of this simple model is very similar to the one obtained including the full baryonic octet at densities below the threshold of appearance of more massive hyperons . we therefore believe that this simple model can give correct qualitative results for the full problem . both , @xmath2-@xmath1 and @xmath1-@xmath1-couplings are seen to play a role in determining the existence of an instability . within the bg model it is shown that an instability exists over a very large parameter domain and the two solar mass limit of ns is compatible with important hyperonic abundances . at variance , bsl is stable with respect to phase separation . though , phase instability can be reached when the original interaction potential is supplemented with a phenomenological @xmath1-@xmath1 interaction . we have considered both pure repulsive and attractive - repulsive potentials who fit the experimental data , the measurement of a positive bond energy in double-@xmath9 hypernuclei . the results show that an extra @xmath1-@xmath1 interaction always results in an enhanced maximum ns mass . in conclusion , we believe that there is no hyperon puzzle in the sense that the @xmath184 neutron star mass value is well compatible with important hyperonic content and , at the same time , the available constraints on the hyperon couplings . this work has been partially funded by the sn2ns project anr-10-blan-0503 and it has been supported by newcompstar , cost action mp1304 . r. r acknowledges partial support from the romanian national authority for scientific research under grants pn - ii - id - pce-2011 - 3 - 0092 and pn 09 37 01 05 and kind hospitality from lpc - caen and luth - meudon . 99 n. glendenning , phys . lett . * b114 * , 392 ( 1982 ) . p. demorest _ et al . _ , nature * 467 * 1081 ( 2010 ) . j. antoniadis , p.c.c . freire , n. wex _ et al . _ , science , 340 , 6131 ( 2013 ) . m. oertel , a. f. fantina , and j. novak phys . c * 85 * , 055806 ( 2012 ) . l. bonanno and a. sedrakian , astron . * 529 * , a16 ( 2012 ) . s. weissenborn , d. chatterjee and j. schaffner - bielich , nucl . phys . * a881 * , 62 ( 2012 ) . i. bednarek , p. haensel , j.l . zdunik , m. bejger , and r. manka , astron . astrophys . * 543 * , a157 ( 2012 ) . s. weissenborn , d. chatterjee and j. schaffner - bielich , phys . c * 85 * , 065802 ( 2012 ) . j. l. zdunik and p. haensel , astron . astrophys . * 551 * ( 2013 ) a61 . p. haensel and o.yu . gnedin , astron . astrophys . * 290 * , 458 ( 1994 ) . c. schaab , f. weber . weigel , n.k . glendenning , nucl . * a605 * ( 1996 ) 531 . c. schaab , s. balberg , j. schaffner - bielich , astrophys . j. 504 , l99 ( 1998 ) . d. chatterjee , d. bandyopadhyay , phys . d * 74 * , 023003 ( 2006 ) ; d. chatterjee , d. bandyopadhyay , astrophys . j. 680 , 686 ( 2008 ) ; m. sinha , d. bandyopadhyay , phys . d * 79 * , 123001 ( 2009 ) ; t.k . jha , h. mishra , v. sreekanth , phys . c * 82 * , 025803 ( 2010 ) . j. schaffner - bielich , m. hanauske , h. stocker and w. greiner , phys . * 89 * ( 2002 ) 171101 . f. gulminelli , ad . r. raduta , and m. oertel phys . c * 86 * , 025805 ( 2012 ) . f. gulminelli , ad . r. raduta , m. oertel , and j. margueron , phys . c * 87 * , 055809 ( 2013 ) . b. peres , m. oertel , j. novak , phys . d * 87 * , 043006 ( 2013 ) . et al . _ , a828 * , 191 ( 2009 ) . j. k. ahn _ et al . _ , c * 88 * , 014003 ( 2013 ) . g. f. burgio , h .- j . schulze , and a. li , phys . c * 83 * , 025804 ( 2011 ) . p. h. pile _ et al . _ , phys . rev . lett . * 66 * , 2585 ( 1991 ) ; t. hasegawa _ et al . c * 53 * , 1210 ( 1996 ) ; h. hotchi _ et al . c * 64 * , 044302 ( 2001 ) . p. khaustov _ et al . _ , phys . c * 61 * , 054603 ( 2000 ) . c * 70 * , 044613 ( 2004 ) ; m. kohno , y. fujiwara , y. watanabe , k. ogata , and m. kawai , phys . rev . c * 74 * , 064613 ( 2006 ) . et al . _ , lett . * 80 * , 1605 ( 1998 ) . g. b. franklin , nucl . phys . * a585 * , 83c ( 1995 ) . i. vidana , a. polls , a. ramos , and h .- j . schulze , phys . c * 64 * , 044301 ( 2001 ) . s. balberg , and a. gal , nucl a625 * , 435 ( 1997 ) . m. oertel , c. providencia _ et al . _ , in preparation . c. ducoin , ph chomaz , and f. gulminelli , nucl . phys . * a771 * , 68 ( 2006 ) . j. m. lattimer and f. d. swesty , nucl . phys . * a535 * , 331 ( 1991 ) . c. ducoin , ph . chomaz , f. gulminelli , nucl . phys . * a789 * , 403 ( 2007 ) . c. providencia , l. brito , s. s. avancini , d. p. menezes and ph . chomaz , phys . c * 73 * , 025805 ( 2006 ) . r. c. tolman , proc . usa 20,3 ( 1934 ) ; j. r. oppenheimer and g. m. volkoff , phys . 55 , 374 ( 1939 ) . j. cugnon , a. lejeune , and h .- j . schulze , phys . c * 62 * , 064308 ( 2000 ) . x .- zhou , h .- j . schulze , h. sagawa , c .- x . wu , and e .- zhao , phys . c * 76 * , 034312 ( 2007 ) , issn 0556 - 2813 . schulze , nucl . phys . * a835 * , 19 ( 2010 ) , issn 03759474 . schulze and t. rijken , phys . c * 84 * , 035801 ( 2011 ) .

we present a near - infrared extinction study of the dark globule b335 , a protostellar collapse candidate , using data from hst / nicmos and the w.m . keck observatory . these data allow a new quantitative test of the `` inside - out '' collapse model previously proposed to explain molecular line profiles observed toward this region . we find that the _ shape _ of the density profile is well matched by the collapse model , but that the _ amount _ of extinction corresponds to larger column densities than predicted . an unstable bonnor - ebert sphere with dimensionless outer radius provides an equally good description of the density profile , and is indistinguishable from the collapse model over the range in radius sampled by the extinction data . the bipolar outflow driven by the embedded young stellar object has an important effect on the extinction through the core , and modeling the outflow as a hollowed - out bipolar cone of constant opening angle provides a good match to the observations . the complete exinction map is well reproduced by a model that includes both infall and outflow , and an additional 20% dispersion that likely results from residual turbulent motions . this fitted model has an infall radius of ( pc for 250 pc distance ) , and an outflow cone semi - opening angle of . the fitted infall radius is consistent with those derived from molecular line observations and supports the inside - out collapse interpretation of the density structure . the fitted opening angle for the outflow is slightly larger than observed in high velocity co emission , perhaps because the full extent of the outflow cone in co becomes confused with ambient core emission at low velocities .

astro-ph0108470

the near - infrared color excess of extincted background stars can reveal the dust distribution in a dense core and provide information on physical structure . historically , this technique has been hampered in small , dense regions by limited instrumental sensitivity ( jones et al . 1980 , 1984 ) , but the development of large format array cameras has sparked renewed interest ( lada et al . 1994 , alves et al . 1998 , lada , alves & lada 1999 , alves , lada & lada 1999 , 2001 ) . the unprecedented sensitivity available with the near infrared camera and multi object spectrometer ( nicmos ) instrument on the hubble space telescope ( hst ) opens up a new regime . in particular , the high extinctions of dense cores that form the immediate environs of candidate low mass protostars are now accessible . molecular line surveys of nearby dark clouds have identified a large number of low mass dense cores ( myers , linke & benson 1983 ) of which roughly half are associated with young stellar objects detected by _ iras _ ( beichman et al . 1986 ) . these observations have contributed to the general paradigm that at least some low - mass stars like the sun form by the gravitational collapse of isolated , dense cores . in the standard theory , originating with shu ( 1977 ) , the dynamical collapse to a star / disk system occurs from the inside - out . first , an @xmath4 distribution is established prior to collapse as the cloud core loses magnetic and turbulent support through ambipolar diffusion and relaxes to a balance between gravity and thermal pressure . collapse is initiated at the center where the density is highest , and a wave of infall propagates outward at the sound speed . conditions inside the infall radius asymptotically approach free - fall , with @xmath5 and @xmath6 . modifications to the simple theory have been made to account for more realistic initial conditions , including slow rotation ( tereby , shu & cassen 1984 ) and residual magnetic fields ( galli & shu 1993 , li & shu 1997 ) . a similar radial density profile is also produced by the competing self - similar collapse theory of larson ( 1969 ) and penston ( 1969 ) . however , the larson - penston flow predicts far larger infall velocities than the shu theory , and has proved unsuccessful at explaining molecular line observations ( see e.g. zhou 1992 ) . unambigous identification of inside - out collapse motions has proved difficult for several reasons . first , the fact that collapse initiates in the center of a core means that it can be detected only with observations of sufficiently high angular resolution . further difficulty comes from the fact that the velocities associated with infall motions are small , comparable to intrinsic line widths and much smaller than the velocities of the bipolar outflows powered by young stellar objects . the presence of localized , redshifted self - absorption in spectral lines of moderate optical depth provides one signature of infall motions , but this signature is not unique . scenarios other than collapse can produce the same spectral line profiles . the most promising claims for collapse have been inferred from redshifted self - absorption , and consequently these claims are often controversial ( see the review of myers , evans & ohashi 2000 ) . the dense core in the b335 region is generally recognized as the best protostellar collapse candidate . the b335 region contains an isolated , roughly spherical globule at a distance of roughly 250 pc ( tomita , saito & ohtani 1979 ) . deeply embedded in the globule is a young stellar object discovered at far - infrared wavelengths by keene et al.(1983 ) and detected only at @xmath7 m by _ iras_. observations of high velocity molecular line emission have shown that the young stellar object drives a bipolar outflow with semi - opening angle of @xmath8 ( cabrit , goldsmith & snell 1988 , hirano et al . 1988 , 1992 ) . detailed radiative transfer models based on the theory of inside - out collapse provide good fits to many molecular line profiles of dense gas tracers observed at @xmath9@xmath10 resolution ( zhou et al . 1993 , choi et al . these models imply an infall radius of 0.03 pc ( @xmath11 ) , an age of @xmath12 yrs , and a central mass of @xmath13 m@xmath14 . while the agreement of the model with few parameters and the observed molecular line profiles strongly support inside - out collapse , the model is not unique . for example , in one extreme scenario discussed by zhou et al . ( 1993 ) , an unrelated foreground cloud of just the right excitation properties absorbs emission from a static core and exactly reproduces the infall signatures . recent molecular line observations of b335 made with the higher angular resolution of interferometry are also starting to show perhaps important discrepancies with the predictions of inside - out collapse ( wilner et al . 2000 ) . an extinction map of sufficient depth can probe the density structure of b335 and test the claim for inside - out collapse in a way that is free from many of the problems that plague molecular line studies , including excitation effects and spatially varying abundances of the molecular tracers . b335 is fortuitously placed for an extinction study , located at galactic coordinates @xmath15 , @xmath16 , which is a direction with a very high density of background stars with uniform properties . previous optical extinction studies of b335 were not sensitive to the structure of the dense core anywhere close to where the molecular line data suggest collapse ( see bok & mccarthy 1974 , dickman 1978 , tomita et al . 1979 and frerking et al . 1987 ) . in this paper , we present a near - infrared extinction study of b335 using data from hst / nicmos , supplemented by data from the w.m . keck observatory . section [ sec : obs ] briefly reviews the technique and describes the observations and data reduction . section [ sec : results ] presents an analysis of the extinction data in the context of several physical models for the b335 density structure , and comparisons with the results of previous studies . simulations of the observations are also presented , using a monte carlo approach detailed in the appendix . section [ sec : conc ] summarizes the main conclusions . measurements of the extinction of stars background to a dense core provides estimates of column density along many pencil beams . these estimates can be used to determine the overall structure of the obscuring core . the extinction measurements are best done at infrared wavelengths , where absorption due to dust is much less than in the optical , thereby probing regions of higher ( visual ) extinction . the basic method is to measure the near - infrared color excess for each background star : @xmath17 where @xmath18 is the intrinsic color of the star . the color excess represents the differential optical depth between the two near - infrared filters , which is directly proportional to dust column density . the observed color excess may be converted to an equivalent visual extinction using the standard reddening law ( e.g. rieke & lebosfsky 1985 ) , and then converted to gas column density by adopting a gas - to - dust ratio ( e.g. bohlin et al . there are some uncertainties in both of these conversions . the gas - to - dust ratio in protostellar cores is the subject of some debate ; at low extinction there is some evidence to suggest that it is constant along different lines of sight in the galaxy ( bohlin et al . 1978 ) , but at high extinction it is unclear whether the gas - to - dust ratio remains constant . for example , if grain growth occurs in the high extinction regions as a result of molecular depletion and agglomeration , then one might expect lower extinction for a given column density . the reddening law is probably less problematic and nearly universal at near - infrared wavelengths , given the small grain sizes present in protostellar envelopes ( mathis 1990 ) . in practice , the color excess method does not require a knowledge of the intrinsic colors of the individual extincted stars . a statistical correction to their observed colors may be obtained empirically from the background stellar population , provided it is sufficiently spatially uniform . for stars of spectral types a0 to late m , in the main sequence or the giant loci , the intrinsic colors lie in the narrow range @xmath19 ( koornneef 1983 ) . therefore , at high extinctions , little error is introduced by simply assuming a mean background star color derived from nearby control fields . the b335 globule was imaged with the nic3 camera on hst ( cycle 7nicmos , go-7843 , 1998 june 23 ) using two broadband filters , f160w and f222 m , similar to the usual h band and k band filters that are matched to the 1.6 @xmath20 m and 2.2 @xmath20 m atmospheric windows . the nic3 camera has @xmath21 pixels with size @xmath22 giving a @xmath23 field of view . the observations were obtained during the second of two periods of a few weeks when the hst secondary mirror was adjusted to allow the nic3 camera to be in focus ( the `` nic3 campaigns '' ) . the observing program had two parts : ( 1 ) a deep @xmath24 mosaic of the central region of high extinction , taking 6 orbits , and ( 2 ) a radial strip of images extending about @xmath25 from the center , taking 1 orbit , to reach the background stellar population and provide a sample of stars to transform nicmos magnitudes to a standard photometric system . the length of the strip was the maximum allowed before suffering the heavy overhead of guide star reaquisition . ( the strip included the center of b335 as insurance in the event that the deeper observations were unsuccessful . ) figure [ fig : skysurv ] shows the observed fields overlayed on a wide field optical view of the region from the digital sky survey . we note that the nic3 orientation angle on the sky and stepping direction of the radial strip were given by the spacecraft orientation on the day of observation . advance specification of a particular orientation would have created an untenable scheduling constraint ; the orientation obtained was based solely on where the spacecraft happened to be during the period when the nic3 camera was in focus and b335 was accessible . the central region was observed with a standard nine - point spiral dither pattern with @xmath9 offsets , and a total exposure time of 896 seconds per dither . the resulting square mosaic covers @xmath26 . the strip images were taken in _ one - chop _ mode , an exposure time of 56 seconds per chop , with an offset of @xmath27 between chops , and roughly @xmath28 overlap between successive images . this resulted in a strip of coverage @xmath29 wide that extends @xmath30 from the mosaic center . the data were taken in the nicmos _ multiaccum _ mode , for which an initial readout of the detector array is followed by non - destructive readouts during the course of a single integration . the intermediate readouts can be used to identify and to remove the effects of cosmic ray hits and saturated pixels . the nicmos data were obtained from the hst archive in fits format both unprocessed and already processed through the stsci pipeline ( calnic a & b ) . initial inspection showed the stsci pipeline did not adequately deal with a variety of instrumental effects , and so we reprocessed the data in iraf using the nicreduce software developed by brian mcleod ( mcleod 1997 , lehar et al . the nicreduce package provides superior and more versatile handling of sky and dark current subtraction , as well as cosmic ray hits and other defects . the bottom rows of the nicmos images suffered from a small amount of vignetting . the effect was most pronounced in the f222 m images , where about @xmath31 were unusable . the overlap of images in the dither pattern and in successive images of the strip was sufficient to ensure that the vignetting resulted in only a small loss in sensitivity in the affected regions , and no gaps in the spatial coverage . the dithered images of the central region of b335 suffered an additional complication due to the presence of a very bright star ( h magnitude of 9.0 ) on the west side of the mosaic center . the enormous count rate from the star caused the image reduction software to produce a small depression in the continuum level of the quadrant of each image that contained the star . this deviation was fixed by adding a constant to the relevant quadrant so that for each image the median pixel value was uniform in all four . ghost impressions of the star in successive exposures were interactively masked in the affected images and were not included in the final mosaic . stars were identified by visual inspection of the images , as several attempts at automatic star finding were compromised by the effects of diffraction spikes and rings around the brighter stars . photometry on each star was performed with the suite of routines in the _ apphot _ package . a series of apertures from @xmath32 to @xmath33 in radius was used for each star . aperture corrections were then performed using a minimum of four isolated and bright stars from the relevant image . each aperture was used to calculate a magnitude and associated error as a check to identify any unusual behavior . the measured count rates were corrected to a @xmath34 radius aperture , and then multiplied by 1.075 as detailed in the nicmos photometric calibration cookbook ( available from stsci ) . the instrumental magnitudes were then transformed to the vega scale using constants for each filter provided in the hst data handbook ( version 4 , dec 1999 ) . for stars that were observed in a region of overlap between strip images , weighted mean magnitudes were calculated . for stars observed in the overlap region of the central mosaic and the first strip image , the value from the more sensitive central mosaic was taken . the complete nicmos photometry list contains 1026 stars detected with the f160w filter and 471 stars detected with the f222 m filter with magnitude error @xmath35 . the nicmos limiting magnitudes are spatially dependent and are depicted in figure [ fig : limplot ] . in the central 32@xmath36 , with the most integration time , the limiting magnitudes are h=24.8 and k=21.5 ( in the cit system ) . observations of b335 were made on 1998 july 10 using nirc ( matthews & soifer 1994 ) on the keck i telescope . these observations were made for two reasons : ( 1 ) to obtain a better study of the stellar background population around b335 , and ( 2 ) to transform the nicmos f160w and f222 m magnitudes to a standard system . the nirc has @xmath21 pixels with size @xmath37 , giving a @xmath38 field of view . each basic observation consisted of nine dithered 5 second integrations , in a @xmath24 pattern with @xmath27 offsets , which were mosaiced together . observations were made with j , h , and k filters . the program had three parts : ( 1 ) nine images of the central region of b335 , approximately mimicking the sub - images of the nicmos central mosaic , ( 2 ) five images leading away from the center of b335 , at each position of a nicmos strip image , and ( 3 ) four additional fields @xmath39 away from b335 , to study the characteristics of the stellar background population . these regions are marked on figure [ fig : skysurv ] . the data reduction and calibration was done in iraf following the standard dark / flat / sky subtraction procedures , using a running flat . this procedure works well when the background emission does not change rapidly . for some periods during the observations , the sky emission in the h band did change relatively quickly , and the background subtraction for observations during these periods is not perfect . the typical seeing during the night was @xmath33 , though there were periods when it was better and also much worse . the majority of stars in the keck images of b335 could be identified with reference to the nicmos data , since the keck images are all less deep than the nicmos equivalents and the field of view slightly smaller . for the keck fields not observed with nicmos , all stars were again identified by visual inspection . photometry was performed for all non - saturated stars using the _ apphot _ package . again , a series of increasing radial apertures was used for each star , and each aperture was used to calculate a magnitude . this approach allowed a confidence level in the measurement to be assigned to each star , based on whether the distribution of measurements from different apertures was consistent with the predicted uncertainty . if the spread of the various magnitudes was greater than 0.2 magnitudes , then the measurement was identified as low quality ( 18% of the stars detected at h and 17% of the stars detected at k ) . these low quality stars have point - spread - functions that differ from the norm , mainly from those periods of poor and varying seeing conditions . the instrumental magnitudes were transformed to the cit system through observations of the hst infrared standards sj9164 and sj9182 ( persson et al . 1998 ) and the ukirt standard fs35 ( casali & hawarden 1992 ) . the internal dispersion in the zero points measured for these stars was 0.02 magnitudes , and no atmospheric extinction correction could be discerned . the keck photometry list contains a total of 297 stars detected at h , and 340 stars detected at k , in the overlap region between the keck and nicmos datasets . the four keck fields @xmath39 away from b335 contain typically 100 detected stars per square arcminute with h@xmath40 . the four fields display uniform properties and no discernable reddening ; the star counts in each image are identical to within poisson errors , and the mean @xmath41 color of the fields are all in the range @xmath42 to @xmath43 , consistent with the intrinsic colors of giant and main sequence stars . a detailed study of the background stellar population is presented in section [ sec : background ] . the limiting magnitudes of the keck observations are approximately h=19.7 and k=19.6 . for extinction calculations , the nicmos f160w and f222 m magnitudes are transformed to standard h and k magnitudes . linear least squares fits ( not including the lower quality keck data ) give slopes of @xmath44 for ( h@xmath45f160w ) vs. f160w , @xmath46 for ( k@xmath45f222 m ) vs. f222 m , and @xmath47 for ( h@xmath45k ) vs. ( f160w@xmath45f222 m ) , where the formal uncertainties in the fitted slopes are @xmath48 . the formal uncertainties are underestimates of the true uncertainties since the magnitude comparison slopes are strongly biased by the few brightest stars , and the color comparison slope is biased by the few reddest stars . experimentation with subsets of the stars resulted in variations in the fitted slopes at the @xmath49 level . we conclude that these data provide no evidence for non - zero slopes or color terms in the transformation . we calculated constant offsets for the magnitude conversions for each filter , using a weighted mean of ( h@xmath45f160w ) and ( k@xmath45f222 m ) values . the result is h = f160w@xmath50 , and k = f222m@xmath51 , where the uncertainties represent the standard error in the mean . figure [ fig : coltrans ] shows the scatter plot of @xmath41 vs. ( f160w@xmath45f222 m ) with the derived color transformation superimposed . for the @xmath41 colors of interest , in the range 0 to 3 , the differences between the two photometric systems are remarkably small . support for the accuracy of these transformations comes from published ground - based photometry of the one very bright star observed by krgel et al . ( 1983 ) , who measured johnson h and k fluxes of @xmath52 mjy and @xmath53 mjy , which correspond to magnitudes of h@xmath54 and k@xmath55 , and color @xmath56 . these values compare very favorably with the transformed nicmos magnitudes of h@xmath57 and k@xmath58 , and color @xmath59 . note that systematic errors limit the accuracy of the nic3 photometry to a few percent , and are not reflected in the random uncertainties quoted here . to check the uniformity of the background stellar population , the starcounts and mean colors of the four keck off fields were compared with those for the two nicmos strip images furthest away from the center of b335 . the colors in the different images are consistent within the observed dispersions . the fifth nicmos strip field has a mean color and standard deviation of @xmath60 , @xmath61 , very similar to that found for the combined keck off - fields : @xmath62 , @xmath63 . the fourth nicmos strip field shows a slightly redder mean color ( @xmath64 ) and higher dispersion ( @xmath65 ) than the other fields ; most likely , this field is not far enough away from the center of b335 to be completely free of its extinction , and this field was excluded from our characterization of the background stellar population . figure [ fig : backcolor ] presents a histogram of the @xmath41 colors of the `` background '' stars in 0.1 magnitude bins ( excluding the keck stars with lower quality photometry ) . the @xmath41 mean color is 0.13 magnitudes , with standard deviation 0.16 magnitudes . also shown for comparison is a gaussian distribution with the same mean and standard deviation ( normalized so that the area under each curve is equal ) . a color - magnitude diagram constructed for this sample shows no sign of any important systematic trends , e.g. there is no tendency for fainter stars to be redder . the starcounts in the fifth nicmos strip field are indistinguishable from those in the keck off - fields , taking into account the different different size fields of view . figures [ fig : images ] & [ fig : strip ] show the nicmos f160w and f222 m images of the b335 region . the central mosaic images in figure [ fig : images ] may be compared to h band and k band images of a smaller part of the central region of b335 obtained over three nights at keck by hodapp ( 1998 ) using nirc . in the region of overlap , the f222 m image shows the same stars as the k band image , but the overall sensitivity of the nicmos image is better and the point - spread - function much sharper , leading to better separation of nearby stars . the f160w image shows many more stars than the corresponding h band image . in contrast with the ground - based data , essentially all stars detected with the f222 m filter are also detected with the f160w filter because of the low background and greater sensitivity of the space - borne nicmos at the shorter wavelength . figure [ fig : nicmos ] conveys much of the information from the central mosaic of nicmos observations . the left plot is a pseudo - image in that the two axes are spatial , the apparent brightness of a star determines its `` size '' , and the value of its @xmath41 color determines its `` color '' . the right - hand plot displays the radial dependence of the @xmath41 colors out to @xmath66 from the center . the origin in both plots is given by the position of the embedded young stellar object , as determined by interferometric observations of millimeter continuum emission ( wilner et al . 2000 ) ; the uncertainty in the registration of the nicmos images and the millimeter peak is estimated to be about @xmath34 . based on an inspection of the extinctions , the stars in the right hand plot have been marked according to their spatial location : those that lie in the north or south quadrants are marked in red , and those that lie in the east or west quadrants are marked in blue ( the quadrant boundaries are marked in the pseudo - image as dashed lines ) . defining @xmath67 to be the angle subtended between the vector to a star and the axis of the outflow ( which runs very close to east - west ) , this division may be expressed simply as @xmath68 or @xmath69 . the nicmos central mosaic image contains 239 stars detected at h , 121 stars detected at k , and 119 stars that are detected in both filters . the detected star closest to the embedded young stellar object is at @xmath70 radius . the reddest detected star is at @xmath71 radius and has an @xmath41 color of 4.0 , corresponding to over 60 magnitudes of equivalent visual extinction . the major features of figure [ fig : nicmos ] are : ( 1 ) a strong gradient in the @xmath41 colors as one approaches the protostar ; ( 2 ) fewer stars are detected close to the center , despite better sensitivity ; ( 3 ) reddening of the stars turns off into background noise beyond the @xmath72 ( 0.15 pc ) outer radius of dense gas determined from molecular line data ( zhou et al . 1990 ) ; ( 4 ) there is a marked asymmetry in the stellar colors associated with the bipolar outflow stars viewed within @xmath73 of the outflow axis have systematically lower colors ; ( 5 ) the dispersion of the @xmath41 colors at a given radius and in the quadrants away from the outflow is @xmath74 greater than that of the background population . we describe several theoretical models for the density distribution of b335 and evaluate the success of these models in light of the extinction data . the models considered are not meant to comprise an exhaustive list . they include the inside - out collapse models previously proposed for b335 , bonnor - ebert spheres , and simple power law descriptions . we also discuss briefly the effects of the bipolar outflow and turbulent motions in b335 . there is a substantial literature of theoretical work by frank shu and colleagues describing the density structure of collapsing dense cloud cores , starting from the initial condition of a singular isothermal sphere . taking parameters appropriate for the b335 region , we briefly describe the pre - collapse state , the spherical inside - out collapse , and the modifications to spherical collapse expected for slow initial rotation or weak magnetic fields . _ pre - collapse configuration _ : the initial state is that of a singular isothermal sphere ; the density falls off as @xmath75 , where @xmath76 is the radius , with the normalization of the density determined by the effective sound speed . for b335 , zhou et al . ( 1990 ) derive a value for the effective sound speed of @xmath77 km s@xmath78 from molecular line data , which corresponds to a kinetic temperature of 13 k. the static initial condition is therefore given by : @xmath79 where the conversion to a molecular hydrogen number density assumes 0.7 for the the mass fraction in hydrogen . _ spherically symmetric collapse _ : the simplest scenario is the spherically symmetric collapse ( shu 1977 ) . in this model , a spherical wave of collapse propagates outwards from the center at the effective sound speed . the radial distance that the wave has traveled , sometimes called the infall radius , is the only additional parameter that defines the density distribution . inside the infall radius , conditions approach free fall , with the density taking the asymptotic form @xmath80 . by reproducing essential features of several molecular line profiles with this model , zhou et al . ( 1993 ) claimed evidence for inside - out collapse in b335 . choi et al . ( 1995 ) made more detailed radiative transfer models to match the zhou et al . ( 1993 ) observations of h@xmath81co and cs molecules , determining the infall radius to be @xmath82 pc ( @xmath11 ) . note that this derived infall radius falls entirely within the deep central mosaic of nicmos coverage . _ rotation _ : the effect of slow rotation on the collapse has been solved using a perturbative approach from the spherically symmetric case ( terebey , shu & cassen 1984 ) . the result is a distortion to the shape of the collapse wave of order @xmath83 , where @xmath84 is the number of times that the cloud has rotated since collapse initiated ( this is also equivalent to the ratio of the infall radius to the centrifugal radius ) . for b335 , frerking et al . ( 1987 ) found an angular velocity of @xmath85 s@xmath86 through c@xmath87o observations . this indicates a value of @xmath88 , and a distortion in the wave by only @xmath89 an entirely negligible effect . the density distribution is more strongly distorted , with the effect largest near the center of collapse . for radii greater than a tenth of the infall radius ( i.e.@xmath90 ) already well within the innermost radius where stars are detected the correction is always less than of order @xmath91 , which is also completely negligible . such slow rotation also has an unimportant effect on the model line profiles towards the center of b335 ( zhou 1995 ) . we therefore do not include rotation in our models . _ magnetic fields _ : the effect of a weak magnetic field on the inside - out collapse of a cloud has been studied by galli & shu ( 1993 ) and li & shu ( 1997 ) . the collapse wavefront and the density distribution are again distorted , in a similar way to the rotational model . in the magnetic case , the parameter @xmath84 is given by the ratio of the infall radius to the magnetic radius : @xmath92 galli & shu ( 1993 ) showed that the field effects only a small perturbation to the spherically symmetric solution outside a radius of @xmath93 . figure [ fig : nicmos ] shows that variation in color between the two spatial populations of stars can be as large as a factor of two , well beyond the infall radius of @xmath11 derived by choi et al . ( 1995 ) . since this certainly can not be construed as a `` small perturbation '' at these radii , we can place a lower limit on the magnetic field that would be needed to explain the asymmetry in the context of this model . taking @xmath94 , we find that : @xmath95 such a minimum magnetic field strength is implausibly large . for example , crutcher & troland ( 2000 ) measured a line - of - sight magnetic field of b@xmath96 @xmath20 g in the starless protostellar core l1544 using the zeeman effect in lines of oh , and this is much higher than the typical upper limits found for other dark clouds ( crutcher et al.1993 ) . in this context , it therefore seems very unlikely that a magnetically modified inside - out collapse could be responsible for the large asymmetry observed in the b335 extinction data . bonnor - ebert models are pressure confined isothermal spheres , for which the solution remains physical at the origin ( ebert 1955 , bonnor 1956 ) . in common with the singular isothermal sphere , the initial condition for inside - out collapse , they are solutions of a modified lane - emden equation ( chandrasekhar 1967 ) : @xmath97 where @xmath98 is the dimensionless radius , and @xmath99 is a logarithmic density contrast , with @xmath100 the ( finite ) central density . unlike the singular solution , the bonnor - ebert solutions do not diverge at the origin . instead , the boundary conditions are that the function @xmath101 and its first derivative are zero at the origin . the above equation can be solved by division into two first order equations which can then be tackled simultaneously using numerical techniques ( in our case a 4th order runge - cutta method ) . there is a family of solutions characterized by a single parameter the dimensionless outer radius of the sphere , @xmath102 . for a given sound speed and a particular choice of the shape of the density curve ( i.e. @xmath102 ) , there is one additional degree of freedom , the physical scale of the model . this additional degree of freedom may be removed by implementing an additional constraint , for example by fixing the outer radius , the total mass , or the external pressure . configurations with dimensionless outer radius @xmath103 are unstable to gravitational collapse ( bonnor 1956 ) . the gravitational collapse of bonnor - ebert spheres has been studied numerically by foster & chevalier ( 1993 ) . they find that the flow asymptotically approaches the larson - penston solution at the origin at the time of and prior to the formation of a central core , but emphasize that these large early accretion rates last for a very short time . if the cloud is initially very centrally condensed ( i.e.@xmath104 ) the later stages of infall closely resemble shu s inside - out collapse of a singular isothermal sphere . recently , alves , lada & lada ( 2001 ) used the color excess method to study b68 , a starless core , with data from the eso vlt . the b68 system is much less centrally condensed than b335 , and has much lower column density at the center ( only about 30 magnitudes of equivalent visual extinction ) . the alves et al . ( 2001 ) extinction map of b68 suggests that the density structure is well described by a bonnor - ebert sphere with dimensionless outer radius slightly in excess of critical : @xmath105 . although the b335 core is at a much later evolutionary stage than b68 , with an embedded protostar , we consider the bonnor - ebert sphere since it describes b68 so successfully . the outflows from young stellar objects can have a substantial impact on their surroundings . a bipolar outflow in b335 was discovered in co by frerking & langer ( 1982 ) . later studies in various tracers by cabrit , goldsmith & snell ( 1988 ) , hirano et al . ( 1988 , 1992 ) , chandler & sargent ( 1993 ) , and wilner et al . ( 2000 ) show that on arcsecond to arcminute scales , the outflow is well collimated with a semi - opening angle of about @xmath106 , and with the axis lying very nearly in the plane of the sky . the outflow orientation is close to east - west , with position angle roughly @xmath107 in the coordinate system of the nicmos images . inside the region affected by the outflow , the density structure is modified as the core material is swept out by the flow , but the exact form of the density field associated with the outflow is not known . the bipolar outflow may be an important driver of turbulent motions . zhou et al . ( 1990 ) determined a turbulent component to the effective sound speed in b335 of @xmath108 km s@xmath78 . coupling of the velocity and density fields via the continuity equation suggests that the perturbations to the velocity field should induce density contrast perturbations , i.e. @xmath109^{1/2 } \simeq 0.2 $ ] . the resulting perturbations to the column densities ( and extinctions ) will depend on the size scale of the velocity and density field perturbations . the perturbations to the column density will be largest when the size scale is a factor of a few smaller than the dense core . the extinction map is consistent with this possibility since the observed colors of neighboring stars are generally very similar , yet there exists a @xmath110% dispersion for stars that are not neighbors but are located at similar radial distances from the protostar ( see figure [ fig : nicmos ] ) . the effect of the column density perturbations are inevitably reduced by integration along the line - of - sight , but the observed residual turbulence most likely contributes significantly to the dispersion in the observed colors . we evaluate the goodness of fit for these various model density distributions by calculating a reduced @xmath111 , defined as : @xmath112 where @xmath113 is the number of free parameters in the model being fitted , and the sum extends over a particular subset of @xmath114 stars taken from the total number detected in both nicmos filters . the values of the observed color excess @xmath115 and the uncertainty @xmath116 are calculated using the properties of the unreddened stellar background population ( see section [ sec : background ] , and figure [ fig : backcolor ] ) , assuming each star to have an intrinsic color of @xmath117 , with an uncertainty of @xmath118 : @xmath119 \sigma_{i}^{\ , \mathrm{nic } } & = & \sqrt{\sigma_i^2 + \sigma_{\ , \mathrm{bg}}^2 } \nonumber \\ & = & \sqrt{\sigma_i^2 + 0.16 ^ 2 } \ , \end{aligned}\ ] ] where @xmath120 is the uncertainty in the observed @xmath41 color of a given star . this procedure implicitly assumes that the photometric errors ( @xmath120 ) and the intrinsic @xmath41 colors both have gaussian distributions . though the latter condition is not strictly true , monte carlo trials show that the @xmath121 values calculated with the above recipe agree very closely with those produced when the intrinsic color of each star is instead chosen randomly from the gaussian distribution with the same mean and standard deviation as the sample of background colors . we conclude that the results of the fitting are not sensitive to this approximation . since the models are in general non - linear in the fitting parameters , we analyze the uncertainty in the best - fit model parameters using a monte carlo technique known as the _ bootstrap _ method ( press et al . 1992 ) . to describe the method , let us represent the dataset that is being used in the fitting by @xmath122 . the first step is to construct a new dataset of equal size @xmath123 , by randomly selecting @xmath114 times from the original dataset . this new sample is then analyzed in the same way as the original dataset , and the fitted model parameters recorded . this process is then repeated @xmath124 times , where @xmath124 is sufficiently large that the resulting distribution of best - fit model parameters is insensitive to its exact value ( for our models , typically @xmath125 ) . the standard deviation of this distribution provides an estimate for the uncertainty of the parameters that best fit the original dataset . table [ tab : fits ] summarizes the results from the various @xmath126 analyses and these results are discussed below . figure [ fig : colschem ] shows a schematic diagram of the spatial coverage of the nicmos observations relative to the important size scales in b335 . the asymmetry in the observed colors appears to be associated with the bipolar outflow . since we do not know from theory the effect of the outflow on the material distribution , we initially restrict attention to the quadrants that are located away from the flow , @xmath68 , where @xmath67 is the angle subtended between the vector to a star and the axis of the outflow . to avoid noise near the boundary of the globule , we also restrict the fitting region to @xmath127 . in fit a of table [ tab : fits ] , we use a power law model for the density distribution . in fit b we use the shu inside - out collapse model , and in fit c we use a bonnor - ebert sphere . the best fitting single power law model ( fit a ) to the dense core region unaffected by the outflow has an index of @xmath128 . this value is well matched to the @xmath129 envelope of the isothermal sphere that represents the initial condition of the inside - out theory . the inside - out collapse solution ( fit b ) , which is similar to a broken power law with fixed indices , provides a somewhat better description than can a single power law model with arbitrary index . remarkably , the fitted infall radius of @xmath130 ( @xmath131 pc ) agrees within uncertainties with those calculated by zhou et al . ( 1993 ) and choi et al . ( 1995 ) by modeling molecular line profiles . that the minimum @xmath121 value for these fits are near 3 instead of unity reflects the @xmath74 greater dispersion in the observed colors than in the intrinsic colors of the background stars . a bonnor - ebert sphere provides an equally good description of the density structure in b335 ( fit c ) . figure [ fig : fitcomp ] shows a comparison of the @xmath41 radial profiles predicted by the bonnor - ebert sphere in fit c , and the inside - out collapse model in fit b. the plot shows that the two models are practically indistinguishable over the range in radius where there are stars to fit ( @xmath70@xmath132 ) . at large radii , the bonnor - ebert solution has slope close to a power law index of 2.0 , and at smaller radii , the slope flattens . in effect , the best fit bonnor - ebert sphere is one that mimics the inside - out collapse solution over the range in radius for which there is data . the models diverge only at such small radii that extinction data of sufficient depth to discriminate between them is currently impossible to obtain . one discriminant may be sensitive , high resolution observations of dust emission , which is strongest at these small radii where extinction data is lacking . regardless , the value of @xmath133 for the best fit bonnor - ebert sphere is well in excess of the critical value of 6.5 , and it is therefore an unstable configuration that should undergo gravitational collapse similar to shu s inside - out solution ( foster & chevalier 1993 ) . the simplest model we imagine to simulate the effects of the bipolar outflow on b335 is to assume that the outflow has completely cleared out the dense material in a bipolar cone of constant opening angle . figure [ fig : outflow ] shows the radial behavior of the @xmath41 colors from simulated observations we have made of an infalling model for four different values of the model outflow semi - opening angle @xmath134 . ( the appendix describes how the simulations are constructed . ) stars viewed through the model outflow cone are marked in red , and those not viewed through the outflow cone are marked in blue . the density distribution for @xmath135 in each of the simulations is the infalling model that best fits the @xmath136 data beyond @xmath137 from the outflow axis ( fit b ) . as the model opening angle increases , the stars begin to separate into two distinct populations . two effects are evident as the angle increases : ( 1 ) the population of detected stars with reduced colors becomes larger , and ( 2 ) the separation in the colors of the two populations grows larger . manifest in the simulation with the largest opening angle ( @xmath138 ) is that the first effect is not only caused by the outflow cone occupying a larger fraction of the sky , but is also due to there being less column density for lines - of - sight passing through the cone , which allows a greater number of the background stars to be detected . comparison of the radial behavior of the @xmath41 colors in these simulations with the observations in figure [ fig : nicmos ] shows that these simple models successfully reproduce the major features . the extent of the spread between the colors of the two populations of stars in the nicmos data seems to be reproduced best by the simulations with semi - opening angles of @xmath139 and @xmath137 . these angles also roughly reproduce the observed number of stars with significantly reduced colors . the simple hollow cone model of the bipolar outflow allows the stars viewed closer to the outflow axis to be included in the @xmath126 analyses . the fact that the predicted @xmath41 colors are a sensitive function of the model outflow opening angle means that we are able to fit for the opening angle of the flow . in both fits d & e , we fit for the entire region within @xmath127 , with a model which has both the infall radius and outflow opening angle as free parameters , and with the color excess normalization for each model chosen to minimize the @xmath121 in the region beyond @xmath137 from the outflow axis ( as in fit b ) . the only difference between fit d and fit e is that fit e includes an extra @xmath140 dispersion term in @xmath141 to account for additional structure in the cloud , perhaps from the residual turbulent fluctuations . table [ tab : fits ] lists the best - fit parameters for fit d and fit e , and they are nearly identical . in both fits d & e , @xmath121 is essentially an independent function of the infall radius and the outflow opening angle . figure [ fig : chisurf ] shows a contour plot of the surface of @xmath121 for fit d. the minimized value of @xmath121 is 4.02 , with best fit values of the infall radius and outflow opening angle of @xmath142 ( @xmath143 ) and @xmath144 , respectively . for fit e , the contour plots and fitted parameters are essentially identical infall radius @xmath145 ( @xmath146 ) and outflow opening angle @xmath147 but the minimized @xmath121 is @xmath148 due to the extra dispersion . that the two fits agree so closely , particularly with regard to the infall radius , is interesting . the extra dispersion in fit e reduces the weight of the most reddened ( and most central ) stars , which inevitably are those that constrain the fitting of the infall radius . the composite model with infall and bipolar outflow provides a consistent description of the density structure of b335 . when the regions affected by the outflow are excluded from consideration , the inside - out collapse solution provided a good description of the data . modeling the outflow as a hollowed - out bipolar cone allows inclusion of the stars from the outflow region in the @xmath126 analysis . despite the obvious crudeness of this model for the outflow , the fits to the entire dataset ( fits d & e ) constrain both the infall radius and outflow opening angle , and the fitted infall radii are entirely consistent with our earlier determination ( fit b ) and with the values determined by molecular line studies the fitted opening angle of the outflow is somewhat larger than indicated in studies of high velocity co emission . examination of the co velocity structure ( e.g. hirano et al . 1988 , 1992 ) suggests that the flow is more collimated at higher velocities , and that the opening angles dervied from co must be lower limits since the low velocity outflow becomes confused with the ambient globule emission ( and is resolved out by interferometers ) . this effect may explain why the opening angle appears wider in extinction than in co emission . of course , the one parameter model fitted here simply may not provide an adequate description of the outflow geometry at this level of detail . figure [ fig : models ] presents a simulation of the models for fit d and fit e to afford another way to examine the success of these models , as well as the differences between them ( the appendix describes the detailed recipe for constructing these simulations ) . the model for fit d has an infall radius of @xmath10 and an outflow opening angle of @xmath149 ( the normalization of the color excess is matched to the data by the @xmath126 analysis ) . this model reproduces the gross structure of the nicmos data very successfully . however , the radial plot shows that the dispersion in the @xmath41 colors at a given radius in the simulation is clearly significantly less than in the observations . the model for fit e is nearly identical to fit d except that the color excess has been multiplied by a gaussian random variable with mean of unity and standard deviation 20% . the simulated data from the model for fit e , with this additional dispersion to mimic the effects of residual turbulence , compares much more favorably with the observations in figure [ fig : nicmos ] . so far , we have been concerned only with the _ shape _ of the density profile and have ignored the free scaling factor applied to the color excess predicted by the various models during the fitting . for both the inside - out collapse models and the bonnor - ebert spheres , the normalization of the density distribution is entirely determined by the effective sound speed , measured to be @xmath150 km s@xmath78 ( zhou et al . this measurement , together with the distance to the cloud , gas - to - dust ratio , and near - infrared reddening law , make a unique prediction for the color excess normalization in each model . as listed in table [ tab : fits ] , a scaling factor @xmath15135 is required to match the models with the observed color excess . figure [ fig : modelb ] shows a simulation of the spherically symmetric inside - out collapse model with the effective sound speed of @xmath150 km s@xmath86 , the standard distance of 250 pc ( tomita et al . 1979 ) , and the infall radius of @xmath11 derived by choi et al . the simulation demonstrates the extent to which the model underestimates the observed reddening and extinction in b335 . if this model were correct , then the nicmos observations would have been sufficiently deep to see right through the center of the cloud , and the number of stars detected would have been far greater than observed . this therefore leads to the question : if the inside - out collapse model or the bonnor - ebert sphere model is correct , then what explains the need for this large scaling factor between the predicted extinction and the measured values ? the first possibility to consider is that the effective sound speed in b335 has been underestimated . to account for the @xmath15235 by a larger kinetic temperature in the initial state seems implausible a temperature of @xmath1534060 k would be needed . multi - transition observations of nh@xmath154 show that the bulk of the dense gas in b335 has a kinetic temperature of 1012 k ( menten et al . 1984 ) , consistent with a balance between heating and cooling processes . to account for the factor with non - thermal motions also seems implausible , given the measured widths of molecular lines . a non - thermal contribution of @xmath1550.350.45 km s@xmath78 would be required , which is clearly at odds with a multitude of molecular line data . for example , zhou et al.(1990 ) used observations of h@xmath81co to determine a small turbulent contribution of @xmath156 km s@xmath78 to the thermal sound speed . similar results from a variety of different molecules are documented by menten et al . ( 1984 ) and frerking , langer & wilson ( 1987 ) . one might envision a form of non - thermal support that is not reflected in the observed molecular line - widths , perhaps because of a special viewing geometry . if magnetic support were primarily from a static field and not waves , and this field were oriented so that the motions were not observable along the line - of - sight , then it would be effectively undetectable . although such a special situation seems contrived and unlikely , the orientation of b335 is special in that the outflow lies close to the plane of the sky , which may make such a configuration more plausible . another possibility that affects the overall scaling lies in the outer boundary , i.e. where the cloud ends . zhou et al . ( 1990 ) calculated the mass of b335 using an outer radius of 0.15 pc based on their radio map and the 250 pc distance estimate . the molecular line profiles of dense gas tracers are not particularly sensitive to this assumption . we have adopted this value for the outer radius throughout the extinction modeling . indeed , the column densities are also insensitive to this assumption since the density falls off steeply with radius . for a star at the infall radius of the best - fitting inside - out collapse model ( fit b ) , doubling the outer radius of the cloud increases the predicted extinction by only 7.5% , which is an unimportant amount with respect to the derived scaling factors . an overestimate of the distance to b335 would also result in predicted column densities that are too small . this distance uncertainty may be the dominant source of systematic error in the study . the distance of 250 pc was adopted from optical measurements of stellar reddening versus distance by tomita et al . ( 1979 ) , and this distance is uncertain at the 50% level . if b335 were closer , then the extinction would be larger at the same angular radius . to account for the full scaling factor would require that the distance be @xmath15735 times closer , only about 5080 pc , which seems unreasonable . also , for a given level of observed extinction , the inferred mass of the globule is proportional to the square of its distance ( since the angle subtended by b335 is fixed ) . for the outer boundary assumption of zhou et al . ( 1990 ) , the mass inferred from our extinction data is @xmath158 the 14 m@xmath14 for a distance of 250 pc may be large for a small globule , but reducing the distance by the whole scaling factor lowers the mass to less than 1 m@xmath14 for the inside - out collapse model , and a little over 1 m@xmath14 for the bonnor - ebert sphere . this seems small given that the globule must be forming a star with a substantial fraction of a solar mass , and losing a lot of mass through outflow in the process . furthermore , in the context of the inside - out collapse model , a reduction in the distance also results in a smaller mass for the protostellar core , as well as a shorter time since the onset of collapse ( both are proportional to the distance for a given sound speed and infall radius ) . in short , a closer distance for b335 would ameliorate the discrepancy between the observed and predicted extinction , but it is not clear if this could account for the entire difference . a final uncertainty is the conversion from column density to color excess using the gas - to - dust ratio and reddening law . significant variations in the reddening law have not been observed , and they are unlikely to explain much of the observed discrepancy in the color excess ( although we briefly return to this point in the appendix ) . by contrast , the gas - to - dust ratio has been observed to vary in the galaxy , with variations of @xmath159 not uncommon along different lines of sight ( bohlin et al . if the gas - to - dust ratio in b335 were lower than the standard value for the interstellar medium , or the reddening law steeper , then this would cause us to underestimate the extinction . potentially , the full scaling factor could be accounted for by this uncertainty , or perhaps by some combination of a closer distance and a modified gas - to - dust ratio . for example , if the true distance to b335 were actually 100 pc , and the gas - to - dust ratio were about one half of the standard value ( i.e. @xmath160 ) , then these two factors would fully account for the discrepancy between observed extinction and the predictions of either model , and also give a reasonable mass for the globule of @xmath161 . if molecular hydrogen were to deplete onto dust grains , this could cause such a decrease in the gas - to - dust ratio at high density . however , h@xmath81 is not thought to undergo significant freeze - out in this way ( sandford & allamandola 1993 ) , and a decrease in the gas - to - dust ratio is at odds with evidence that suggests an increase in high extinction regions ( e.g. bohlin et al . 1978 , de geus & burton 1991 , and ciolek & mouschovias 1996 ) . the density distribution in protostellar cores may also be studied through dust emission . if the emission is observed at long enough wavelength that the rayleigh - jeans approximation applies , then , in the optically thin regime , the intensity at impact parameter @xmath162 is proportional to the integral along the line - of - sight of the product of the density and the temperature . assuming that the intensity , temperature and density obey power law distributions in radius , then the power law index of the density distribution @xmath163 is simply given by @xmath164 , where @xmath113 is the observed power law index of the intensity profile , and @xmath165 the power law index of the temperature profile . the mass may also be derived from the radial intensity distribution , given an assumed mass opacity . unfortunately , the mass opacity of dust at millimeter and submillimeter wavelengths is uncertain , by a factor of five or more ( ossenkopf & henning 1994 ) . for b335 , dust emission at submillimeter and millimeter wavelengths has been studied recently by shirley et al . ( 2000 ) and motte & andr ( 2001 ) . shirley et al . ( 2000 ) constructed images at 850 @xmath20 m and 450 @xmath20 m at resolutions of @xmath166 and @xmath167 , respectively , using scuba at the jcmt . at both wavelengths , these observations are sensitive to emission out to an angular radius of roughly @xmath168 . they adopt a temperature profile with power law index @xmath169 for a region where the envelope is optically thin to the bulk of the infrared radiation ( e.g. emerson 1988 ; butner et al . they fit the circularly averaged maps at each wavelength , extending from inner cutoffs at angular radii of @xmath170 for the 850 @xmath20 m profile and @xmath171 for the 450 @xmath20 m profile . they derive power law indices of @xmath172 and @xmath173 from the 850 @xmath20 m and 450 @xmath20 m emission , respectively . they estimate a mass for b335 of 1.2 m@xmath14 using the total 850 @xmath20 m flux in a @xmath174 area and assuming a sphere of constant density . motte and andr ( 2001 ) constructed a 1.3 mm continuum image of b335 with @xmath175 resolution using the iram 30 m telescope , detecting emission out to an angular radius of @xmath174 . they adopt an isothermal temperature distribution and calculate a density power law index of @xmath176 . in this model , the total mass is @xmath177 . they note an asymmetry in the emission , with the emission being elongated perpendicular to the outflow axis . as in this extinction study , they perform fits to the off - outflow regions , and as a result increase the uncertainty in their fitted density index by 0.2 ( included in the quoted uncertainty ) . taking account the uncertainties , the density power law indices derived from the millimeter and submillimeter emission agree both with each other and with the results from dust extinction analysis ( fit a , @xmath178 ) . however , the extinction method provides a direct measurement of the distribution that suffers none of the added uncertainties associated with the temperature profile in the core . we present a near - infrared extinction study of the protostellar collapse candidate b335 using deep integrations from hst / nicmos supplemented by data from the w.m . keck observatory . in summary : 1 . we determine transformations between the nicmos filters f160w and f222 m and the standard h and k bands in the cit photometric system for a range of @xmath41 color from 0 to 3 . at the accuracy of our data , the derived transformations do not have significant slopes or color terms . 2 . the nicmos mosaic image of the central @xmath26 region of b335 shows a dramatic fall off in the number of stars detected towards the protostar location , where the extinction increases due to the central concentration of dense core material . the nicmos mosaic image contains a total of 119 stars detected in both filters , the innermost of which is at a radius of @xmath70 . the photometry shows a steep gradient in the @xmath41 colors toward the central protostar . in addition , there is a strong asymmetry in the colors superimposed on the overall gradient that coincides with the bipolar outflow . we compare a series of models of dense core stucture to the extinction data , including the previously proposed models of inside - out collapse derived from molecular line observations . the _ shape _ of the @xmath136 radial profile is well matched by the inside - out collapse model , in particular the density fall - off as @xmath129 for the envelope and a flattening within the purported infall zone . an unstable bonnor - ebert sphere with dimensionless outer radius @xmath0 provides an equally good description of the density profile , and is indistinguishable from the collapse model over the range in radius where there are stars to fit . the inside - out collapse model is not unique . the power law index for the density distribution derived from extinction is consistent with recent determinations from dust emission measurements at millimeter and submillimeter wavelengths . the bipolar outflow appears to be responsible for the dominant asymmetry in the extinction data . stars whose line of sight fall within @xmath137 of the outflow axis show lower extinction than those further away from the outflow . the effect of the outflow on the extinction data is reproduced very well by models that consider a hollowed out bipolar cone of constant opening angle . in the context of the inside - out collapse models , the observed asymmetry is too large to be explained by either rotation , or a weak magnetic field . 5 . the entire extinction dataset can be well fitted by a model that includes infall , bipolar outflow , and an additional @xmath140 dispersion that mimics residual turbulent motions . the best fit outflow semi - opening angle is @xmath3 and the best fit infall radius is @xmath1 . the fitted value of the infall radius is consistent with those derived from molecular line studies of b335 , and supports the inside - out collapse interpretation of the density structure . the fitted opening angle for the bipolar outflow is somewhat larger than has been observed in high velocity co emission , perhaps because the molecular line studies do not recover the full outflow width due to confusion with ambient emission at low velocities . the normalization of the observed color excess is a factor of @xmath179@xmath180 higher than the value predicted using the nominal 250 pc distance to b335 , and the standard gas - to - dust ratio and near - infrared reddening law . the discepancy can conceivably be explained by a combination of a closer distance and a lower gas - to - dust ratio . partial support for this work was provided by nasa through grant number hst - go-07843.01-a from the space telescope science institute , which is operated by aura , inc . , under nasa contract nas5 - 26555 we thank bob goodrich and randy campbell for invaluable assistance with the observations obtained at the w.m . keck observatory . dwah thanks brian mcleod for his help in running the nicreduce software , and alex wilson for his assistance with idl graphics . this appendix contains the tools and general recipe for constructing the simulated observations of b335 presented in sections [ sec : sims ] & [ sec : scaling ] . we construct a composite luminosity function for the background population by combining the data from fields away from the extinction of the b335 dense core . in deriving the luminosity function , the photometry for each field was cut at a very high completeness level , where the number of stars per unit magnitude range peaks . ( this is a very conservative estimate of limiting magnitude . ) to avoid any issues concerning how the stars should be binned , the cutoff was determined by finding where the gradient of the cumulative luminosity function was steepest . for the dataset from the fifth nicmos strip field , which includes 202 detected stars at h and 85 detected stars at k , the cut - off magnitudes are 20.7 at h , and 18.7 at k. for the combined dataset of the four keck off - fields , which includes 242 detected stars at h and 246 detected stars at k , the cut - off magnitudes were 18.8 for both filters . the empirical h - band luminosity function was calculated in three parts out to the nicmos high completeness limit ( h@xmath181 ) . for the bright end ( h@xmath182 ) , only stars from the nicmos fifth strip field are included , as stars brighter than this limit were saturated in the keck images . for the mid - range ( @xmath183 ) , stars from both datasets contribute to the function . for the faint end ( @xmath184 ) , above the keck high completeness limit , only stars in the nicmos fifth strip field were included . in each part , the number of stars was divided by the relevant observed area to build up the cumulative luminosity function . in the part where the nicmos and keck data overlap ( @xmath183 ) , the two datasets show good agreement . the k band luminosity function was constructed following the same procedure , and it is effectively the same as for the h band , offset by the mean color of @xmath185 magnitudes . the central regions of b335 were observed to a deeper limiting magnitude than the strip or the @xmath39 off fields , by about 4 magnitudes ( h band ) at the mosaic center , and about 3 magnitudes at the edge ( see figure [ fig : limplot ] ) . although the stars in these central regions of b335 are highly extincted , the empirical background luminosity function must be extrapolated in order to simulate the deepest observations . since the least extincted star in the nicmos mosaic has @xmath186 , corresponding to an extinction of @xmath187 , the luminosity function must be extrapolated by about 2 magnitudes at h in the simulations . the form of the h cumulative luminosity function suggests that a broken power law provides a good description . the function was fitted by minimizing the @xmath126 of the fit to the bright - end of the luminosity function . the resulting function is given by : @xmath188 simulated observations of b335 were constructed by superimposing a model cloud on a field of background stars with the appropriate characteristics . the column density was calculated at the position of each background star and converted into two observable quantities : ( 1 ) an extinction at h band , and ( 2 ) a color excess @xmath136 . these conversions are based on the gas - to - dust ratio , and reddening law . ( a brief discussion on the uncertainties of these quantities is given in section [ sec : method ] . ) 1 . generate a background star field by selecting randomly from the broken - power - law cumulative luminosity function in concert with the gaussian color distribution of the background population . the broken - power - law luminosity function is extended to h@xmath189 , just beyond the limiting magnitude at the center of the nicmos mosaic ( note that model background stars intrinsically fainter than h@xmath190 are not recovered in the simulated observations of the fitted models from table [ tab : fits ] the function is extended to h@xmath189 simply to ensure completeness ) . positions for each star are assigned randomly within a region @xmath191 , to encompass the area of the nicmos observations . 2 . select a model density distribution to describe b335 . 3 . calculate the column density for each star in the region . 4 . convert the column density to an extinction at h band , @xmath192 , and a color excess @xmath136 using an assumed gas - to - dust ratio , r , and a reddening law , @xmath193 . the `` standard '' prescription is ( bohlin et al . 1978 , rieke & lebofsky 1985 ) : @xmath194 { \mathrm{e}(\mathrm{h}-\mathrm{k})}\ : & = & a_{\mathrm{h } } \ : / \ : 2.78 \ .\end{aligned}\ ] ] ( in the simulations , the reddening law was allowed to vary up to 1525% steeper than the standard law in order to more exactly reproduce the observed starcounts . a fluctuation upward in the star counts at the one standard deviation level explains the inferred steeper reddening law . ) modify the magnitudes and colors of each background star appropriately . 6 . simulate the nicmos observations by applying the spatially dependent detection limits ( figure [ fig : limplot ] ) . llllll a & scaled power law : @xmath163 & @xmath195 & 146 & @xmath196 & 3.37 + b & scaled infall : @xmath197@xmath198 & @xmath199 & 146 & @xmath200 @xmath201 & 3.10 + c & scaled bonnor - ebert ( @xmath202 ) & @xmath203 & 146 & @xmath204 @xmath205 & 2.93 + d & scaled infall , hollow - cone outflow & @xmath206 & 209 & @xmath207 @xmath208 & 4.02 + e & as fit d with 20% dispersion & @xmath206 & 209 & @xmath209 @xmath210 & 1.04 +

we present observations of the uv absorption lines in the seyfert 1 galaxy ngc 3516 , obtained at a resolution of/ 40,000 with the space telescope imaging spectrograph ( stis ) on 2000 october 1 . the uv continuum was times lower than that observed during 1995 with the goddard high resolution spectrograph ( ghrs ) , and the x - ray flux from a contemporaneous _ chandra x - ray observatory ( cxo ) _ observation was a factor of below that observed with _ asca_. the stis spectra show kinematic components of absorption in ly , c iv , and n v at radial velocities of , , and km s ( components 1 , 2 , and 3 , respectively ) , which were detected in the earlier ghrs spectra ; the last of these is a blend of two ghrs components that have increased greatly in column density . four additional absorption components have appeared in the stis spectra at radial velocities of , , , and km s ( components 5 through 8) ; these may also have been present in earlier low - flux states observed by the _ international ultraviolet explorer ( iue)_. based on photoionization models , we suggest that the components are arranged in increasing radial distance in the order , 3 , 2 , 1 , followed by components 5 8 . we have achieved an acceptable fit to the x - ray data using the combined x - ray opacity of the uv components 1 , 2 and 3 . by increasing the uv and x - ray fluxes of these models to match the previous high states , we are able to match the ghrs c iv column densities , absence of detectable c iv absorption in components 5 through 8 , and the 1994 _ asca _ spectrum . we conclude that variability of the uv and x - ray absorption in ngc 3516 is primarily due to changes in the ionizing flux .

astro-ph0205389

it is now understood that intrinsic absorption is a common phenomenon in seyfert 1 galaxies , present in more than half of those observed with the spectrographs aboard the _ hubble space telescope ( hst ) _ ( crenshaw et al . 1999 ) . among those seyferts that show absorption , high ionization resonance lines such as n v @xmath81238.8 , 1242.8 and c iv @xmath81548.2 , 1550.8 are always present , along with ly@xmath4 , while lower ionization lines , such as si iv @xmath81393.8 , 1402.8 , and mg ii @xmath82796.3 , 2803.5 , are less common . the absorption lines can be blueshifted ( by up to 2100 km s@xmath6 ) with respect to the systemic velocities of the host galaxies , indicating net radial outflow . however , uv absorption lines can arise in dusty gas within the plane of the host galaxy ( crenshaw et al . 2001 ; crenshaw et al . 2002 ) ; these are often saturated and at velocities close to systemic . the ionic columns can be highly variable , which may be the result of changes in response to the ionizing continuum , as first suggested in case of ngc 4151 by bromage et al . ( 1985 ) and espey et al . ( 1998 ) , and discussed in detail in crenshaw et al . ( 2000a ) and kraemer et al . ( 2001a ) , or transverse motion , as appears to be the case for ngc 5548 and ngc 3783 ( crenshaw & kraemer 1999 ; kraemer , crenshaw , & gabel 2001b ) . in fact , both mechanisms may be at work in individual agn . in either case , the variability is indicative of the promixity of the absorbers to the central active nucleus of these galaxies ; in the case of ngc 4151 , constraints on the density and ionization of the strongest absorption system place it within 0.03 pc of the continuum source ( kraemer et al . 2001a ) . the presence of intrinsic absorption , typically in the form of bound - free edges of o vii and o viii , has been detected in the x - ray spectra of a similar fraction of seyfert 1 galaxies ( reynolds 1997 ; george et al . most recently , spectra obtained with the _ chandra x - ray observatory ( cxo ) _ have revealed that x - ray absorption lines associated with this material are also blue - shifted ( kaastra et al . 2000 ; kaspi et al . 2000 , collinge et al . the connection between the x - ray and uv absorption is complex and there is likely a range of physical conditions within the absorbers ( see kriss et al . 1996a ; kriss et al . 2000 ; krolik & kriss 2001 ) . furthermore , there is evidence that the x - ray absorbers in some sources must also be multi - zoned ( otani et al . 1996 ; reynolds et al . 1997 ) . ngc 3516 ( @xmath9 @xmath10 0.00875 ) is one of the few seyfert 1 galaxies with uv absorption lines , specifically n v , c iv , and si iv , strong enough to have been detected with the _ international ultraviolet explorer ( iue ) _ ( ulrich & boisson 1983 ) . these lines are known to vary on timescales as short as weeks ( voit , shull , & begelman 1987 : walter et al . 1990 ; kolman , halpern , & martin 1993 ) . walter et al . suggested that the c iv absorption consists of a narrow , apparently stable component in the core of the broad emission lines and a variable , broad , blue - shifted component . koratkar et al . ( 1996 ) found the blueshifted component present in _ spectra in 1978 , but it had disappeared between 1989 and 1993 , and has been absent in subsequent _ hopkins ultraviolet telescope _ ( hut ) ( kriss et al . 1996b ) and _ hst_/goddard high resolution spectrograph ( ghrs ) ( crenshaw , maran , & mushotzky 1998 ) observations . based on ghrs spectra , crenshaw et al . ( 1998 ) determined that the absorption in the core of the line ( walter et al.s `` stable '' component ) consisted of two broad components , with velocities of @xmath5380 and @xmath5 150 km s@xmath6 , and two narrow components at @xmath590 and @xmath530 km s@xmath6 . ngc 3516 also exhibits strong , variable x - ray absorption ( kolman et al . 1993 ; nandra & pounds 1994 ; kriss et al . 1996a ; mathur , wilkes , & aldcroft 1997 ) . although mathur et al . ( 1997 ) suggested that the x - ray and uv absorption arises in gas characterized by a single set of physical conditions , kriss et al . ( 1996a ) modeled the x - ray absorber as two zones , one with an ionization parameter ( the ratio of the density of photons with energies @xmath11 13.6 ev to the number density of hydrogen atoms at the illuminated face of the zone ) u @xmath10 1.66 and a total hydrogen column density of 1.4 x 10@xmath12 @xmath13 and a lower ionization zone with u @xmath10 0.32 and a total column density of 6.9 x 10@xmath14 @xmath13 , and argued that not all of the uv lines could form in these zones . in our previous paper ( netzer et al . 2002 ; hereafter paper i ) , we presented an analysis of the _ cxo_/low energy transmission grating spectrograph ( letgs ) spectrum obtained on 2000 october 6 . the x - ray continuum flux , flux@xmath15 @xmath10 ( 1.6 @xmath16 0.2 ) x 10@xmath17 photons @xmath13 s@xmath6 kev@xmath6 , was close to its historic minimum , roughly a factor of 8 10 below maximum ( observed in 1994 with _ asca _ ) . we found that the variations in observed flux and spectra at these epochs is consistent with an absorber of constant column density ( 7.9 x 10@xmath14 @xmath13 ) whose ionization state has changed in response to the variation in the ionizing continuum , specifically : for the current low flux state , u @xmath10 0.19 , while it would be a factor of 8 to 10 higher in the high flux state ( depending on the amplitude of the uv continuum variations ) , in rough agreement with the high ionization component described by kriss et al . ( 1996a ) . the model predicts large column densities ( @xmath3 10@xmath18 @xmath13 ) for n v and c iv , hence one or more of the observed kinematic components of the uv absorption should be associated with the x - ray absorber . we obtained echelle spectra of ngc 3516 with the space telescope imaging spectrograph ( stis ) on hst under a stis guaranteed time observer program . in this paper , we present our analysis of the physical conditions in the uv absorbers , nature of their variability , and the connection to the x - ray absorption detected in our contemporaneous _ cxo_/letgs observation . we obtained stis echelle spectra of the nucleus of ngc 3516 through a 02 x 02 aperture on 2000 october 1 . these observations and two previous observations of the c iv region with the goddard high resolution spectrograph ( ghrs ) are detailed in table 1 . the reduction of the ghrs spectra is described in crenshaw et al . ( 1998 , 1999 ) . we reduced the stis echelle spectra using the idl software developed at nasa s goddard space flight center for the stis instrument definition team . the data reduction includes a procedure to remove the background light from each order using a scattered light model devised by lindler ( 1999 ) . the individual orders in each echelle spectrum were spliced together in the regions of overlap . figure 1 compares the c iv regions from the high - resolution spectra obtained at the three epochs . as noted in crenshaw et al . ( 1998 , 1999 ) , the ghrs observations in 1995 showed four kinematic components of absorption in c iv ( which we will refer to as the `` core components '' ) . although there was a decrease in the underlying continuum plus broad c iv emission over the six - month interval in 1995 , there was no evidence for a change in the ionic column density ( or radial velocity coverage ) of any component . the new stis observations in 2000 october show a further decrease in the continuum plus broad c iv emission , plus a dramatic change in some of the c iv absorption components . in the stis spectrum , there is a strong absorption feature at the position of the weak ghrs components 3 and 4 . in addition , there is significant new absorption in the blue wing of the broad c iv profle , in the 1553 1559 region . this new absorption is reminiscent of the broad , variable blueshifted absorption seen in the c iv profile from early _ iue _ observations ( koratkar et al . 1996 ) , which will be discussed later in this section . there is no evidence for significant changes in components 1 and 2 , which are heavily saturated at all three epochs . figure 2 shows portions of the stis e140 m spectra where intrinsic absorption is detected in other ions . fluxes are plotted as a function of the radial velocity ( of the strongest member , for doublets ) , relative to the systemic redshift of z @xmath10 0.00875 from the optical emission lines ( crenshaw et al . 1999 ) , since h i 21-cm measurements are not available . each of the original 4 absorption components are present in ly@xmath4 and the doublets of n v @xmath8 1238.8 , 1242.2 and c iv @xmath0 1548.2 , 1550.8 . in addition , si iv @xmath8 1393.8 , 1402.8 absorption is present in components 1 and 2 . we have also detected c ii @xmath01334.5 , c ii * @xmath01335.7 ( fine - structure line ) , and mg ii @xmath8 2796.3 , 2803.5 intrinsic absorption , all in component 1 only . we have not reliably detected any other intrinsic absorption line in the e140 m or e230 m spectra . we have separated the new blueshifted absorption into four distinct kinematic components ( 5 8 ; which we will refer to as the `` blueshifted components '' ) . our criteria for identification of a component are that it must be distinct in radial velocity space ( i.e. , not severely blended with other components ) , and clearly present in n v or c iv . component 5 is by far the strongest and broadest new component , whereas components 6 8 are relatively weak . the procedures we used to measure the intrinsic absorption lines follow those of crenshaw et al . ( 1999 ) . to determine the shape of the underlying emission , we fit a cubic spline to regions on either side of the absorption , and then normalized the absorption profiles by dividing the observed spectra by the spline fits . we determined the covering factor c@xmath19 , which is the fraction of continuum plus emission that is occulted by the absorber in our line of sight , using the technique of hamann et al . ( 1997 ) for doublets . we have relied on the c iv doublets to determine these values , since many of the n v doublets are contaminated by galactic absorption or blending of kinematic components . due to the modest signal - to - noise of the spectra and blending of components in the wings , c@xmath19 can only be determined reliably in the cores of the lines . we are not able to accurately deconvolve components 3 and 4 in the stis spectra ( see figure 2 ) , so they are treated as one component ( 3@xmath74 ) in this paper . table 2 gives the radial velocity centroids , widths ( fwhm ) , covering factors , and associated one - sigma errors for each kinematic component . these velocity centroids represent the means and standard deviations determined from the measurements of the c iv , n v , si iv , and ly@xmath4 profiles ( when available for the latter two ) . compared to the ghrs measurements ( crenshaw et al . 1999 ) , there is no evidence for a change in the radial velocities of components 1 4 , except for perhaps a slight shift in component 2 from @xmath5148 km s@xmath6 ( ghrs ) to @xmath5183 km s@xmath6 ( possibly due to increased absorption in the blue wing of this component ) . the radial velocity centroid of component 3@xmath74 is very close to the position of the ghrs value for component 4 , suggesting ( but not proving ) that component 4 may be the one responsible for the large increase in absorption . our original suggestion that components 3 and 4 may be due to the interstellar medium or halo of the host galaxy ( crenshaw et al . 1998 ) is therefore still a possible explanation for one of these components ( probably component 3 ) , but not the other ( our models are based on the assumption that these components are essentially co - located , near the ionizing source ) . to determine the fwhm of the components , we used values from relatively unsaturated lines , to get an indication of the true velocity spread for each component . thus , we used the stis si iv lines for components 1 and 2 , the ghrs c iv lines for components 3 and 4 , and the stis ly@xmath4 , n v , and c iv lines for components 5 8 . for components 1 and 2 , the fwhms are therefore smaller than those measured from the ghrs c iv profiles ( crenshaw et al . 1999 ) , since the latter are heavily saturated . next , we examine the covering factors in table 2 . the covering factors for components 1 and 2 are _ consistent _ with a value of one , although we can not rule out the possibility that they are as small as 0.96 . component 3@xmath74 shows the strongest evidence for a non - unity covering factor . although the aperture for the stis observations is small , there is likely to be some emission from the narrow - line region ( nlr ) that is unocculted ( see kraemer et al . 2001 ; brotherton et al . this could explain the nonuntity covering factor for component 3@xmath74 since it is near the systemic redshift , where the nlr contribution is strongest . to illustrate this point , we have have simulated the nlr contribution by reproducing gaussians at the systemic velocity in figure 2 . we kept the doublet ratios fixed at 2:1 , chose a single value for the fwhm ( @xmath10 165 km s@xmath6 ) that provided a good overall fit , and scaled the intensities to obtain the best fit for each ion . given our assumptions , the nlr emission provides a reasonably accurate match to the troughs of the saturated lines ( note the contamination of n v @xmath01242.8 by galactic s ii ) . the error in covering factor for component 5 is too large to make a definite conclusion . we could not determine reliable values for components 6 8 , since the profiles are shallow and relatively noisy . lower limits from l@xmath4 and estimates from n v are consistent with the covering factors determined from c iv . to determine the ionic column densities , we converted each normalized profile to optical depth as a function of radial velocity , and integrated across the profile , as described in crenshaw et al . ( 1999 ) . for components 1 and 2 , the si iv profiles are much narrower than the profiles of ly@xmath4 , n v , and c iv , suggesting that the latter are heavily saturated ( assuming the same intrinsic velocity dispersion for all ions ) . by the same token , components 3 and 4 in the ghrs spectra are narrow in c iv compared to the stis component 3@xmath74 , again suggesting heavy saturation in the latter . furthermore , the possibility of nonunity covering factors for components 1 , 2 , and 3@xmath74 suggests that the column densities may be much larger than would be determined for c@xmath19 = 1 . thus , we can only determine lower limits to the column densties for these lines ( by assuming c@xmath19 = 1 ) , and the column densities for components 1 and 2 from the ghrs spectra ( crenshaw et al . 1998 , 1999 ) should be taken as lower limits . for component 5 , even if c@xmath19 = 0.6 , the depth of the lines are such that they are not heavily saturated . as we mentioned previously , we do not have reliable measurements of c@xmath19 for components 6 8 . since these components are shallow , they are unlikely to be saturated ( including ly@xmath4 in these components ) , unless the covering factors are extremely small , which seems unlikely . thus , we have assumed c@xmath19 @xmath10 1 for components 5 8 . table 3 gives our measurements of stis and ghrs column densities for all lines that have been detected in at least one component . we include lower limits for heavily saturated lines and upper limits for undetected lines to constrain the photoionization models . to pursue a possible connection between the continuum flux and the strength of the c iv absorption , we have examined all of the far - uv spectra that have been obtained of ngc 3516 . table 4 lists the additional far - uv spectra ( i.e. , those not listed in table 1 ) , which were obtained at lower spectral resolutions by the _ iue _ , _ hut _ , and the faint object spectrograph ( fos ) and space telescope imaging spectrograph ( stis ) on _ hst_. we have retrieved the most recently processed versions of these spectra from the _ iue _ and _ hst _ archives in order to measure their continuum fluxes and examine their c iv profiles . for the _ hut _ spectrum , we determined the continuum flux directly from the plot in kriss et al . ( 1996b ) . we measured the continuum fluxes by averaging the points in a bin centered at 1460 ( observed frame ) with a width of 20 . we determined the one - sigma flux errors from the standard deviations of the averages ; for the _ iue _ spectra , this technique is known to overestimate the errors ( clavel et al . 1991 ) , and we scaled the errors for these observations by a factor of 0.7 to ensure that observations taken on the same day agreed to within the errors ( on average ) . for the two ghrs spectra , the continuum regions adjacent to the broad c iv emission profile were not observed . in this case , we used separate fits to the continuum and broad emission in the stis c iv region and tried different linear combinations of these fits until an accurate match was obtained to the continuum plus broad emission in each ghrs spectrum . figure 3 shows that the far - uv continuum of ngc 3516 has varied dramatically over a span of 22 years . early _ iue _ observations were obtained when ngc 3516 was in low to moderate states , whereas later observations , including the intensive monitoring campaign of 1996 ( @xmath3jd 2,490,000 , see koratkar et al . 1996 ) , were obtained primarily at a high state . the last _ iue _ observations and subsequent observations by other instruments demonstrate the rapid variability of ngc 3516 ; in particular , the fos observations show a factor of 5 variation ( ratio of brightest to faintest ) over the course of 11 months ( see goad et al . 1999 ) . however , the trend is to lower flux levels in the most recent obervations . the new stis observations were obtained at very low ( but not unprecedented ) continuum flux levels . to examine the previous behavior of the c iv profile as a function of continuum luminosity , we have placed the _ iue _ spectra into three categories based on their continuum fluxes at 1460 : low ( 0 2.0 x 10@xmath20 ergs s@xmath6 @xmath13 @xmath6 ) , medium ( 2.0 4.0 x 10@xmath20 ergs s@xmath6 @xmath13 @xmath6 ) , and high ( 4.0 7.0 x 10@xmath20 ergs s@xmath6 @xmath13 @xmath6 ) . we averaged the spectra ( weighting by exposure time ) in each category . figure 4 shows the average spectra in the c iv profiles . for comparison , we binned the ghrs and stis spectra to the same wavelength sampling as the _ iue _ spectra ( 1.67 per bin ) , and smoothed slightly to match the _ iue _ resolution ( 6 fwhm ) . we overplot the binned high - state ghrs and low - state stis spectra in figure 4 . the _ iue _ spectra in figure 4 show a remarkable change in the c iv profile as a function of continuum luminosity . the blue emission peak just shortward ( in wavelength ) of the central absorption feature is stronger than the red peak at high luminosity , but becomes weaker and shifts to shorter wavelength at lower luminosities . the binned ghrs and stis spectra show the same effect . comparison with figure 1 shows that this effect is due to increased absorption from components 5 8 in the low - state stis spectrum . this effect has been noticed in previous _ iue _ studies . walter et al . ( 1990 ) characterized the c iv absorption in _ iue _ spectra as a blend of a narrow stable component in the core of the emission line plus a variable broad and blueshifted component . the similar appearances of the low - state _ iue _ and stis c iv profiles indicate that the variable blue - shifted absorption covers a similar range in radial velocity , and _ suggest _ that they originate from the same kinematic components . this would not be too surprising , since long - term stability in the radial velocities of absorption components is a common trait in seyfert 1 galaxies ( crenshaw et al . 1999 ) . koratkar et al . ( 1996 ) showed that the blueshifted component disappeared in _ spectra between 1989 october and 1993 february ( when ngc 3516 was in a high state ) . at the time , it was not clear if this effect was due to evolution of the absorbers ( e.g. , bulk motion of gas out of the line of sight ) or to a change in the ionization of the gas ( e.g. , a decrease in the c iv column due to an increase in the ionizing continuum , coincident with the increase in the uv continuum ) . there are several pieces of evidence that indicate that the latter ( variable ionization ) is certainly an important , and probably dominant , contributor to the c iv absorption variability . 1 ) our examination of individual _ iue _ spectra indicate that the c iv profile variability is more correlated with continuum flux than with time . in particular , the moderately high states at early times and the moderately low states at the end of the _ iue _ observations confirm the above trend . 2 ) walter et al . ( 1990 ) found an anticorrelation between continuum flux and c iv equivalent width in the _ iue _ spectra , consistent with this scenario . 3 ) the individual fos observations , which span a large range ( factor of 5 ) in continuum flux , show the same effect of decreasing blue peak with decreasing flux ( see the spectra in goad et al . 4 ) the high resolution stis and ghrs spectra confirm this effect , and identify the kinematic components responsible for the change in the c iv profile . we note that the same effect has been seen in the intrinsic o vi absorption in ngc 3516 ( hutchings et al . 2001 ) : high - velocity absorption ( consistent with components 5 8) appeared in a low - state _ fuse _ observation , but was not apparent in a moderately high - state _ hut _ observation obtained five years earlier . thus , we believe we have strong evidence for variable ionization of the absorption in ngc 3516 , such that the high - velocity uv absorption disappears at high continuum fluxes . photoionization models for this study were generated using the code cloudy90 ( ferland et al . we have modeled the absorbers as matter - bounded slabs of atomic gas , irradiated by the ionizing continuum radiation emitted by the central source . as per convention , the models are parameterized in terms of the ionization parameter , u. each separate kinematic component was initially modeled with one set of initial conditions , i.e. , u , n@xmath21 , and the total hydrogen column density , n@xmath21 ( @xmath10 n@xmath22 @xmath7 n@xmath23 to denote the ionic column density , where `` x '' is the atomic symbol and `` m '' is the ionization state . ] . for the models , we have assumed only thermal broadening , since 1 ) the absorption line widths could be due to the superposition of unresolved kinematic components and 2 ) comparison models assuming turbulent velocities of @xmath24 300 km s@xmath6 predict nearly identical ionic columns . since the strongest absorption lines in components 1 , 2 , and 3@xmath74 ( ly@xmath4 , n v , and c iv ) are saturated , we can only compare the model predictions to lower limits for the ionic column densities . however , the predictions for these components are constrained by the upper limits , or relatively small column densities , derived for lower ionization species , such as c ii and mg ii . furthermore , total x - ray opacity predicted by the uv absorbers models can not exceed that which we determined from the _ cxo_/letgs spectrum ( paper i ) . in paper i , we suggested that the total reddening in ngc 3516 may be slightly higher than e@xmath25 @xmath10 0.06 ( determined by kriss et al . [ 1996b ] ) . first , using combined stis and fos data from similar flux states , we dereddening the continuum using the galactic curve ( savage & mathis 1979 ) until we matched the ngc 4151 continuum . with the caveat that the ngc 3516 spectrum is from a low state , hence the fuv could be intrinsically depressed ( clavel et al . 1991 ; crenshaw et al . 1996 ) , while the ngc 4151 spectrum is from a moderately high state , we derived an upper limit of e@xmath25 @xmath24 0.15 . alternately , we found , using the average continuum fluxes at 1460 and 5510 determined by edelson et al . ( 2000 ) during their stis monitoring campaign that an underlying continuum of the form f@xmath26 @xmath27 @xmath28 , where @xmath29 , is consistent with an e@xmath25 @xmath2 0.10 . therefore , we have adopted a reddening value of e@xmath25 @xmath10 0.10 @xmath16 0.05 , of which 0.04 is galactic ( schlegel , finkbeiner , & davis 1998 ) . this is in rough agreement with kriss et al . ( 1996b ) , but suggests an uv power - law that is more consistent with typical values for low luminosity agn ( wills et al . 1985 ) . if the dust composition and dust - to - gas ratio are the same as in our galaxy ( see mathis , rumpl , & nordsieck 1977 ) , the intrinsic reddening , e@xmath25 @xmath2 0.06 , requires a neutral hydrogen column of @xmath3 3.0 x 10@xmath30 @xmath13 ( shull & van steenberg 1985 ) . note , however , that the gas within which the dust resides need not be neutral ( see brandt , fabian , & pounds 1996 ) , nor might we expect that the dust near an agn is the same as that found in the galactic ism ( pitman , clayton , & gordon 2000 ; kraemer et al . 2000 ; crenshaw et al . 2002 ) . as in paper i , we have assumed that the uv power - law continuum extends out into the euv , at which point it must steepen to meet the low - energy extension of the x - ray continuum . hence , we have approximated the spectral energy distribution ( sed ) , as a power - law , with the following indices : @xmath29 below 100 ev , @xmath31 over the range 100 ev @xmath32 500 ev , and @xmath33 above 500 ev . @xmath10 75 ks s@xmath6 mpc@xmath6 , the total lyman continuum photon luminosity is q @xmath2 2 x 10@xmath35 photons sec@xmath6 , roughly twice that for ngc 4151 when it was in a recent low luminosity state ( kraemer et al . it should be noted that the power - law break could occur at lower energies , in which case the ratio of the x - ray ionization parameter ( see george et al . 1998 ) to u would be somewhat larger , and the relative fractions of highly ionized species , such as o vii and o viii , would be somewhat higher than our predictions . we have also modeled the uv absorbers in the 1995 high state observed with the ghrs . in addition to a higher overall ionizing luminosity , we have assumed a slightly different sed , as discussed in section 4.3 . we have assumed roughly solar element abundances ( cf . grevesse & anders 1989 ) , which are , by number relative to h , as follows : he @xmath10 0.1 , c @xmath10 3.4 x 10@xmath17 , n @xmath10 1.2 x 10@xmath17 , o @xmath10 6.8 x 10@xmath17 , ne @xmath10 1.1 x 10@xmath17 , mg @xmath10 3.3 x 10@xmath36 , si @xmath10 3.1 x 10@xmath36 , s @xmath10 1.5 x 10@xmath36 , and fe @xmath10 4.0 x 10@xmath36 . the absorbing gas is assumed to be free of cosmic dust . as discussed in section 2 , the covering factor for each of the core components of absorption is @xmath37 0.9 . hence , the simplest geometric arrangement has the absorbers separated in radial position from the central source , but subtending the same solid angle , i.e. , `` screening '' each other from the ionizing continuum . as a result , the ionizing continuum that reaches the absorbers is diluted by the inverse square of its distance and filtered by intervening gas . this arrangement has important consequences for the response of the screened absorbers to changes in the luminosity of the central source , since they are affected by both changes in the ionizing flux _ and _ the opacity of the intervening gas . the core components were present in c iv in ghrs spectra taken while the continuum flux of ngc 3516 was @xmath11 4 times higher than during our stis observations , and components 1 and 2 were highly saturated ( crenshaw et al . 1998 ) . in the stis spectra each of the core components possesses saturated ly@xmath4 , n v , and c iv lines . component 1 has saturated si iv and lines from several low ionization species , hence must be the most optically thick at the he ii lyman limit . based on our photoionization models , the most dramatic changes in the transmitted ionizing continuum occur when there is a large drop in the he ii opacity , to the extent that a screened component which possessed a c iv column of @xmath37 10@xmath38 @xmath13 ( e.g. , a model with n@xmath21 @xmath10 10@xmath14 @xmath13 and a low - state ionization parameter u @xmath10 0.1 ) , would have essentially all of its carbon ionized out of c iv during the high state . since each of these absorbers was present while ngc 3516 was in a high flux state , it is unlikely that component 1 is closest to the ionizing source . expanding on this , we found that the best fit to the observed ionic columns is achieved when the three components are arranged in increasing order of he ii optical thickness , specifically : component 3@xmath74 lying closest , followed by 2 , then 1 . similarly , the absence of the blueshifted components during high flux states of ngc 3516 is consistent with them being screened by gas that is optically thick during low flux states , and hence we chose to have them positioned outside the core components . we have attempted to model each kinematic component as a single zone , adding additional zones only when necessary . as typical in such analyses , the resulting model parameters are a possible , but not necessarily unique , solution set . however , the dual constraints of screening and the total x - ray opacity significantly reduce the parameter space for the models . finally , the true test for this geometry and physical parameters is whether such a model can replicate the high - state column densities , as we discuss in section 4.3 . the model parameters for each component , including u , n@xmath21 , and @xmath39 ( the mean electron temperature ) , are listed in table 5 . implicit in our assumptions is that the distances between these components are significantly smaller than their average radial distance , i.e , they are roughly co - located . also , note that the predicted temperatures are sensitive to the choice of density and , hence , should be considered as very rough approximations . the predicted ionic columns for each component are compared to the measured values in table 6 ( for the full set of predicted column densities for the core components , see tables 7 through 10 ) . for component 3@xmath74 , n@xmath21 and u were constrained by the lower limits to n@xmath22 n@xmath40 and n@xmath41 , and the upper limit to n@xmath42 ( see table 3 ) . component 2 is ionized by the continuum transmitted by 3@xmath74 , and is constrained by the value of n@xmath42 , as well as the aforementioned lower limits . both of these components were successfully modeled as single - zoned absorbers . for component 1 , a single - zoned model underpredicts c ii , while overpredicting mg ii . this is a result of the ionization structure of the slab interior to the point where the gas becomes optically thick at the he ii lyman limit ( 54.4 ev ) . at this point most of the magnesium is in the form of mg iii , which has an ionization potential of 80 ev , little of which can be ionized to mg iv by the heavily absorbed ionizing continuum . on the other hand , most of the carbon is c iv , since c iii has an ionization potential of 47.7 ev , i.e. , below that of he ii . hence , while mg ii is populated by recombination of mg iii , there is little c ii , due to the relatively small fraction of carbon in the form of c iii . if the ionization parameter is significantly lower , the higher ionization states are less populated due to the paucity of ionizing photons , and the situation is mitigated , in the sense that the c iii and mg iii zones overlap more completely . hence , we used a two - zoned model to fit component 1 . the choice of parameters for the two components is somewhat arbitrary , however , for the sake of simplicity , we assumed that the n v , c iv , and si iv arise in one component ( 1a ) , while the c ii and mg ii are formed in the other ( 1b ) . note that , while the prediction for n@xmath43 is quite good , the model overpredicts n@xmath44 by a factor of @xmath3 3.5 . this may be the consequence of our assumption of solar abundances , as we discuss in section 5.2 . interestingly , components 1a and 1b predict a combined neutral column of n@xmath22 @xmath2 10@xmath18 @xmath13 , which is less than half the column determined from the 1995 hut spectrum ( kriss et al . 1996a ) . part of the difference may be due to the contribution from component 2 , since these components are not separable at low resolution . however , the continuum flux observed by hut was roughly a factor of two higher than in the stis spectra and , as we discuss in section 4.2 , the total neutral column should , therefore , have been lower . nevertheless , we do not detect any additional low ionization lines in the stis spectra that might be associated with an unmodeled neutral component . this apparent discrepancy may be due to the presence of a small additional column of neutral gas within our line - of - sight during the time of the hut observations . as shown in table 3 , the n@xmath41/n@xmath22 and n@xmath40/n@xmath22 ratios for the blueshifted components are unusual in the sense that both are large . although h i columns can be quite small in very highly ionized gas ( u @xmath37 0.5 ) , the fractional populations of the lithium - like ionic states tend to be similary small . in part , the unusual ionization balance in these components may be the result of illumination by a heavily absorbed continuum , which has very few photons with energies between 54.4 ev and 100 ev ( see figure 5 ) . in this case , n v is produced by x - ray ionization of n iv by photons with energies much greater than its ionization potential ; the photoionization cross - section at these energies is much lower than at the threshold energy . to overcome this inefficiency , the ionization parameter must be high ( @xmath37 0.5 ) , which in turn ionizes a greater fraction of the hydrogen than in gas exposed to an unfiltered continuum . while the n@xmath41/n@xmath40 ratios for the blueshifted components are well - fitted by the models , n@xmath22 is underpredicted for component 5 , and overpredicted for the other components . it must be noted that the overpredictions are _ much _ worse when an unfiltered continuum is used ( we did not include the effects of continuum filtering between these components , but they are all optically thin at the he ii lyman limit , so the effect is negligible ) . somewhat more curious , given the casual similarity of these absorption features , is the large range in n@xmath21 and u ( see table 5 ) . however , these components experienced an extreme drop in the ionizing flux , due to the drop in the intrinsic continuum luminosity and the increase in opacity of the intervening gas , over which time they may have undergone large temperature changes ( see section 5.2 ) . it is possible that the apparently peculiar ionization balance in these components is a result of this recent history . therefore , we only list the predicted columns for component 5 , which is the lowest ionization and most thermally stable ( which may explain why the predicted column densities are the most accurate ) . our models for the core components predict significant columns densities for c vi , n vi , o vi , and o vii , hence , significant x - ray opacities . as shown in figure 5 , the large he ii column supresses the soft - xray flux . finally , the total column density for these components is n@xmath21 @xmath10 7.7 x 10@xmath14 @xmath13 , which is nearly identical to the single zone model described in paper i. so , it is likely that the uv absorbers are responsible for the x - ray absorption . a full and detailed analysis of the x - ray spectra from ngc 3516 has already been presented in paper i. we do not consider it necessary to repeat such an analysis here . however , in order to test and illustrate the hypothesis that the uv absorbers and x - ray absorbers are dominated by the same components , we have performed direct comparisons of the models and data . the ionic column densities for each of the core components ( see tables 7 10 ) were used to predict the bound - free absorption edges in the x - ray band using the analytic expressions of verner & yakovlev ( 1995 ) and verner et al . we have excluded the blueshifted components since we have not constrained their physical states , although it is our belief that their column densities are too low to contribute much to the x - ray absorption . these were compared to the x - ray data assuming a galactic column density of @xmath45 appropriate for this line of sight ( murphy et al 1996 ) . it should be stressed that this direct and simple comparison does _ not _ include the additional spectral components discussed in the more detailed treatment of paper i. specifically , here we do not include the compton `` reflection '' from optically - thick material which is likely to dominate the observed x - ray spectrum @xmath46 kev , or or any emission from any other circumnuclear material . nevertheless assuming the sed described in section 3.1 , we find our column densities for the low - state spectrum produce a predicted x - ray spectrum similar to that of paper i. thus our model provides a comparable ( acceptable ) description of the _ cxo_/legts data to that shown in paper i , and is strong evidence that the uv and x - ray absorption arise in the same gas , as suggested by mathur et al . ( 1997 ) , although the absorber consists of multiple zones , characterized by different ionization parameters and column densities , as discussed in kriss et al . ( 1996a ) . also interesting is the lack of evidence for highly ionized gas ( e.g. , u @xmath3 1 , n@xmath21 @xmath37 10@xmath14 @xmath13 ) within the line - of - sight to the active nucleus in ngc 3516 , in contrast to ngc 3783 ( kaspi et al . 2000 ) , ngc 5548 ( kaastra et al . 2000 ) , and ngc 4051 ( collinge et al . 2001 ) . in paper i , we argued that the differences in the x - ray properties of ngc 3516 in the _ cxo_/letgs spectrum compared to earlier _ asca _ spectra were due to variations in the continuum flux and the corresponding changes in the ionization state and , therefore , opacity of the intervening absorber . since we have established that the uv and x - ray absorption occur in the same gas , it follows that the changes in uv absorbers must also be in response to variations in ionizing flux , rather than variations in total column density . as noted above , the continuum flux at 1460 was roughly 4.5 times greater in the ghrs observations than during the stis gto observations , and represents the agn in a high state . we assumed that the flux at the lyman limit scaled by the same factor . as there were no contemporaneous x - ray observations , we assumed that the 0.5 kev flux increased by a factor of 9 , as suggested in paper i. finally , we assumed that the @xmath4 remained @xmath10 @xmath51.0 between 13.6 ev and 100 ev , which requires a change in @xmath4 to @xmath51.87 from 100 ev to 500 ev ( although , clearly , we do not know how the sed may have changed in the euv to soft x - ray band ) . through an iterative analysis of the 1994 _ asca _ spectrum , we set the 0.5 to 10 kev spectral index to @xmath50.89 . in modeling the absorbers we held the column densities fixed and assumed the same relative positions . the high - state continuum increased q , and thus u for the unscreened gas , by a factor of @xmath2 4.6 . the predicted values of n@xmath40 for the core components are given in table 6 , and match the ghrs values well , although the prediction for component 2 , while still indicative of saturation , is slightly below the lower limit ( which may be indicative of a somewhat larger column density in the earlier epoch , as suggested by the limits on n@xmath22 determined from the hut spectrum [ kriss et al . 1996a ] ) . interestingly , in the low - state models , all the c iv associated with component 1 was from the higher ionization zone , while in the high - state both zones of component 1 contribute to it . the high - state model for component 5 predicts a negligibly small value for n@xmath40 , so the absence of the blueshifted components can also be explained by the higher ionizing flux . as with the low - state models , we have taken the predicted columns densities ( see tables 11 14 ) and used them to model the _ asca _ data obtained in 1994 april ( e.g. george et al . 1998 ) when the nucleus of ngc 3516 was a factor of 8 to 10 ( at @xmath47 kev ) brighter ( and the compton `` reflected '' continuum less important ) . we followed the standard procedures for the reduction and analysis of _ asca _ data , ignoring the 57 kev band ( where there is emission from fe k@xmath4 ) , and fitting the time - averaged spectral data from all four _ asca _ detectors simultaneously . only the spectral index and normalization of the underlying power law continuum was allowed to vary during the analysis ( the bound - free absorption appropriate for the `` high - state '' column densities were fixed at the values predicted by the uv models ) . as expected , the index of the continuum was in good agreement with our assumptions concerning the sed ( @xmath48 at 90% confidence ) . the resulting fit is shown in figure 6 , and it again matches the overall spectral characteristics . given the source is variable in both the x - ray band , and in the overall shape of the sed , and that there are no simultaneous high resolution uv measurements , we consider our predictions to be in remarkable agreement with the x - ray observations at this epoch . we have shown that our model for the uv absorbers , with the components arranged in increasing radial distance and each covering the continuum source , accurately predicts most of the ionic columns and the x - ray opacity . given the success of the model , we may use it as a starting point to unravel the geometry of the circumnuclear gas . in paper i , based on the recombination timescale for o vii , we placed a conservative lower limit on the density of x - ray absorber of n@xmath21 @xmath11 2 x 10@xmath49 @xmath50 . for component 1b , we predict that n@xmath51 @xmath10 1.87 x 10@xmath38 @xmath13 . the lack of detection of c iii@xmath52 1175 is indicative of n@xmath53 @xmath54 a few x 10@xmath55 @xmath13 . for t @xmath3 2.0 x 10@xmath56k , this ratio of n@xmath53/n@xmath51 corresponds to a density of @xmath54 10@xmath57 @xmath13 ( bromage et al . 1985 ; kriss et al . if the high - ionization core components are co - located with component 1b , their higher ionization parameters require that they have densities of @xmath54 10@xmath58 @xmath50 . alternatively , for component 1a , we predict that n@xmath51 @xmath10 9.43 x 10@xmath38 @xmath13 . based on the upper limit for n@xmath53 , the density of component 1a is @xmath59 10@xmath60 @xmath50 . hence , we derive a loose upper limit of n@xmath21 @xmath54 10@xmath61 @xmath50 for the density of the high - ionization core components . given our estimate of the continuum luminosity and the constraints on the densities of the absorbers , the core components must lie between 5.4 x 10@xmath62 cm and 1.2 x 10@xmath63 cm from the central source . this range in radial distance is consistent with full - covering of the broad c iv emission region , which has a radius of @xmath3 4.5 light - days ( koratkar et al . 1996 ) . by comparison , kraemer et al . ( 2001a ) estimated that the high density ( n@xmath21 @xmath11 10@xmath60 @xmath50 ) uv absorber detected in ngc 4151 , component d@xmath7e , must lie at a distance of @xmath54 6.2 x 10@xmath62 cm from the central source ( assuming h@xmath34 @xmath10 75 km s@xmath6 mpc@xmath6 ) . hence , the radial distance of the absorbing gas in ngc 3516 is roughly consistent with that in ngc 4151 , although possibly somewhat further from the central source and , apparently , less dense . on average , the absorbers are in a much higher ionization state than component d@xmath7e , for which u @xmath10 0.015 . since the ionizing luminosity of ngc 3516 is only a factor of @xmath3 2 greater than ngc 4151 , observed in similar low - flux states , the greater ionization state must be mostly due to the lower density . another interesting difference is the weakness of the scattered light in the troughs of the saturated absorption lines in ngc 3516 : @xmath59 5% of the low - state continuum flux , compared to @xmath11 50% of the low - state flux in ngc 4151 ( kraemer et al . for ngc 4151 , we argued that the scattered light consisted of continuum @xmath7 broad - line emission , reflected into our line - of - sight by free electrons in a component of highly ionized gas that extended beyond the region covered by the absorbers . either this component is weaker in ngc 3516 , the absorbers are covering more of the scattering region , or the scattering region is further away from the nucleus ( outside our projected aperture ) . the observed emission - line fluxes can be used to constrain the global covering factor of the absorbers . the observed c iv @xmath0 1550 flux is @xmath2 2.37 x 10@xmath64 ergs @xmath13 s@xmath6 , or l@xmath40 @xmath2 3.5 x 10@xmath65 ergs s@xmath6 . for a density of n@xmath21 @xmath10 10@xmath58 @xmath50 , the radial distance of component 3@xmath74 is r @xmath10 1.7 x 10 @xmath18 cm and the predicted c iv flux at the illuminated surface of the 4 core component models absorbers is 1.5 x 10@xmath49 ergs @xmath13 s@xmath6 , hence , the covering factor must be @xmath24 0.64 . interestingly , in its current low - flux state , approximately 50% of the c iv emission is from a component with fwhm of @xmath3 1500 km s@xmath6 ( see figure 2 and hutchings et al . [ 2001 ] ) , which is characteristic of the so - called intermediate line region ( ilr ; wills et al . [ 1993 ] ) . if the absorbers are the source of the ilr emission , the covering factor is @xmath3 0.3 . note , that although the core components cover both the broad and intermediate line contributions to the c iv emission , this does not imply that they must lie outside the ilr . for example , self - absorption by the ilr gas could occur in an expanding shell ( see hutchings et al . if the absorption arises in the ilr , the ilr gas has a significant radial velocity component , similar to the emission - line gas in the inner nlr of other seyfert galaxies ( crenshaw & kraemer 2000 ; kaiser et al . 2000 ; crenshaw et al . 2000b ; ruiz et al . 2001 ) . a connection between the ilr and nlr components has been suggested by sulentic & marziani ( 1999 ) , although our results suggest that the ilr could extend to the outer edge of the blr . from our estimate of the intrinsic reddening ( see section 4.1 ) , one would infer that there is a component of dusty gas of column density n@xmath21 @xmath3 3 x 10@xmath30 @xmath13 along our line - of - sight to the nucleus of ngc 3516 . note that this is based on assumption of a galactic dust / gas ratio ; if the ratio were lower , the column would be larger . as shown in kraemer et al . ( 2000 ) and crenshaw et al . ( 2001 ) , uv absorption lines can form in dusty gas surrounding an agn . typically , one would expect strong c iv lines from a column of this size . hence , either the absorption lines associated with the dusty column are buried in the troughs of the core components , or the dust exists in one or more of these absorbers . following barvainis ( 1987 ) , we can estimate the dust sublimation radius , i.e , the point interior to which dust grains would evaporate . during its low - flux state , the ionizing luminosity of the central source is @xmath3 1.4 x 10@xmath66 ergs s@xmath6 . assuming an average luminosity a factor of 3 greater ( see section 2.3 ) , and a dust sublimation temperature of 1500 k ( salpeter 1977 ) , the sublimation radius for ngc 3516 is @xmath54 2.6 x 10@xmath18 cm , or , roughly the same radial distance as the core components . although the absorbers are radially outflowing , we do not know their starting point , or whether they were well - shielded from the continuum source by intervening gas . hence it is possible that some grains have survived within them . alternatively , some dust could have been swept - up by the absorbers as they move outwards . if , for example , the dust were within component 1 , it may help explain the overprediction of mg ii ( section 4.1 ) . furthermore , since the column density of component 1 is roughly 5 times larger than our dusty column , the dust / gas ratio could be 20% the galactic ratio ( indicative of the harsh conditions close to the active nucleus ) , and still produce the observed reddening . in any case , although it is possible the dust is in a heretofore undetected component of gas , we suggest that the reddening may be due to a small amount of dust within one or more of the core components . perhaps the most unusual aspect of the intrinsic absorption in seyfert galaxies is the lack of evidence for radial acceleration . in the case of ngc 4151 , weymann et al . ( 1997 ) attributed this to the large radial distances of the absorbers . however , in kraemer et al . ( 2001a ) we determined that component d@xmath7e was close enough to the central source that it should undergo a detectable acceleration . to explain this , we suggested that the large column density and optical thickness of d@xmath7e would make radiative acceleration inefficient ( see williams 1972 ; mathews 1974 ) . the physical conditions in the core components of ngc 3516 are similar to d@xmath7e , hence the absence of detectable changes in radial velocities between 1995 and the present is understandable , although it is interesting that the blueshifted components , which only appear during lowstates , appear at generally the same velocities , and these are clearly optically thin ( of course , the radial acceleration of the absorbers may simply be too small to detect in the @xmath3 6 years between the ghrs and stis observations ) . another explanation is that the absorbers are not simply radially ( or even radiatively ) accelerated . disk - wind models ( e.g. , konigl & kartje 1994 ) require that the gas in the outflow must first be levitated off the accretion disk . bottorff , korista , & shlosman ( 2000 ) have modeled the absorbers as part of a magnetohydromagnetic wind , which has a strong non - radial component . elvis ( 2000 ) has also proposed an outflow scenario that includes a non - radial component . if the absorption arises in disk - driven wind , the gas would have a strong transverse component of velocity . however our analysis ( see also paper i ) suggests strongly that the variability of the absorption in ngc 3516 results from changes in ionization state due to the variable luminosity of the central source , rather than changes in column density due to transverse motion . one possibility is that the individual kinematic components are associated with different flow tubes or shells along our line - of - sight , which would originate at different radial positions at the base of the wind . since the total column density within the absorber does not appear to change , the amount of gas within these tubes must be fairly constant , although this does not imply confinement but , rather , a continuous flow . although the source luminosity changes dramatically , the flow density appears to be independent of the short - term changes in the flux , which may imply that it is not radiatively driven off the disk ( see , again , bottorff et al . 2000 ) . interestingly , in our model the radial velocities of components 1 through 5 increase with distance from the central source . this would be expected if they each originated at the roughly the same distance , and have been continuously accelerated , e.g. , via resonance line driving or some other radiative mechanism . however , the velocities can also depend on the radial distance at which the different components originated , hence a range in velocities may not be an indication of acceleration , but of different `` launch - pads '' for the components of the flow . another interesting aspect of the uv absorbers is their apparent stability ; the core components vary in ionization state but have shown evidence for significant c iv columns for two decades . as discussed in krolik , mckee , & tarter ( 1981 ) and , more recently , in krolik & kriss ( 2001 ) , there are two regimes of ionization ( characterized by the `` pressure '' ionization parameter , or u / t ) and temperature wherein a cloud of photoionized gas is stable to thermal perturbations : at low ionization / temperature , when line cooling is efficient , and at high temperature / ionization , when thermal balance is achieved via compton processes . in the intermediate region , small changes in ionization can result in large changes in temperature . although the exact relationship between u / t and t for a model nebula depends on the sed of the continuum radiation , including the non - ionizing continuum which is critical in the compton regime , and the atomic parameters , such as collision strengths , instability typically occurs between 5 x 10@xmath56 k and 10@xmath67 k. both our low- and high - state models for the core components predict that these absorbers remain thermally stable , which may help explain their continuing presence . on the other hand , the blueshifted components may not be stable . in figure 7 , we show the instability curve for the screened sed used for the models of the blueshifted components . in the current low - state , component 5 is in the low - temperature stable regime , while the more highly ionized components lie along the vertical , quasi - stable , region ( unstable regions are places where @xmath68 ) . this implies that these components have experienced large temperature fluctuations in response to the changes in ionizing flux , which may explain part of the discrepant model predictions . perhaps most of this gas is in the compton - cooled state , out of which uv absorbers condense when the continuum flux drops sufficiently . the variable opacity of the intervening core components amplifies the changes in ionizing flux and , hence , thermal perturbations of the blueshifted components . we have used medium - resolution echelle spectra obtained with _ hst_/stis to study the physical conditions in the uv absorbers in the seyfert 1 galaxy , ngc 3516 while it was in a recent low - continuum state . we have used contemporaneous _ cxo_/letg data ( discussed in paper i ) to examine the connection between the uv and x - ray absorbers , and have compared our model results to archival , high - state ghrs and _ asca _ spectra . we find that the bulk of the uv and x - ray absorption arises in the same gas , principally components 1,2 , and 3@xmath74 . the line - of - sight covering factors for these components are all @xmath37 0.9 , hence they must be screening one another from the ionizing source . we have demonstrated that the historical changes in the properties of the absorbers are due to changes in their ionization state in response to different levels of the ionizing flux . the changes are more dramatic for the screened components , which are also affected by the changes in opacity of the intervening gas . in particular , the disappearance of components 5 8 in c iv ( the variable blueshifted absorption seen in _ iue _ spectra ) is due to the much higher ionization parameters at high states , as a result of the much lower opacities of components 1 4 . although there are apparent similarities between ngc 3516 and ngc 4151 , such as linear radio structure ( nagar & wilson 1998 ) , nlr kinematics ( ruiz et al . in preparation ) , and central source luminosity , there are striking differences in the appearance of their uv spectra . first , unlike ngc 4151 ( see kraemer et al . 2001a ) , there is little evidence for scattered continuum radiation in a 02 x 02 aperture centered on the nucleus ; alternatively , the absorbers are _ covering _ the scattering region . second , in kraemer et al . ( 2001a ) we argued that the x - ray absorption in ngc 4151 was due to a component of highly ionized gas which we were unable to identify among the uv absorbers , which is clearly not the case for ngc 3516 . finally , the absorbers in ngc 3516 are both in a higher ionization state and of lower density than those in ngc 4151 . it will be interesting if these differences are intrinsic to these otherwise similar agn , or are a clue to the nature of mass outflow . s.b.k . and d.m.c . acknowledge support from nasa grant nag5 - 4103 . h.n . acknowledges support by isf and by umbc . t.j.t . acknowledges support for nasa grant nag5 - 7538 . thanks gary ferland and fred bruhweiler for illuminating discussions . we thank jay dunn for help with the tables . we thank an anonymous referee for their careful reading of the manuscript and useful comments . cccclll stis & e140 m & 1150 1730 & 46,000 & 5488 & 2000 october 1 & 2451819.5 + stis & e230 m & 2275 3120 & 30,000 & 2220 & 2000 october 1 2451819.5 + & & & & + ghrs & g160 m & 1529 1597@xmath69 & 20,000 & 7296@xmath70 & 1995 april 25 & 2449833.5 + ghrs & g160 m & 1529 1597@xmath69 & 20,000 & 6084@xmath70 & 1995 october 22 & 2449833.5 + crcr 1 & @xmath5376 ( @xmath167 ) & 70 ( @xmath168 ) & 0.98 ( @xmath160.02 ) + 2 & @xmath5183 ( @xmath168 ) & 44 ( @xmath1610 ) & 0.98 ( @xmath160.02 ) + 3@xmath74 & @xmath536 ( @xmath164 ) & 20 ( @xmath164 ) , 31 ( @xmath165)@xmath71 & 0.95 ( @xmath160.02 ) + 5 & @xmath51372 ( @xmath169 ) & 271 ( @xmath1644 ) & 0.81 ( @xmath160.21 ) + 6 & @xmath5994 ( @xmath1616 ) & 36 ( @xmath166 ) & + 7 & @xmath5837 ( @xmath167 ) & 99 ( @xmath1616 ) & + 8 & @xmath5692 ( @xmath164 ) & 35 ( @xmath165 ) & + cccccccc & & & stis & & & & + 1 & @xmath375.99 & @xmath3720.48 & @xmath3710.26 & @xmath371.86 & 0.79 @xmath160.25 & 1.44 @xmath160.30 & 0.05 @xmath160.02 + 2 & @xmath376.56 & @xmath3721.55 & @xmath3713.75 & 0.15 @xmath160.05 & @xmath590.25 & @xmath590.25 & @xmath590.02 + 3@xmath74 & @xmath373.34 & @xmath374.84 & @xmath376.20 & @xmath590.04 & @xmath590.25 & @xmath590.25 & @xmath590.02 + 5 & 0.97 @xmath160.18 & 1.65 @xmath160.39 & 2.11 @xmath160.30 & @xmath590.04 & @xmath590.25 & @xmath590.25 & @xmath590.02 + 6 & 0.09 @xmath160.04 & 0.29 @xmath160.10 & 0.10 @xmath160.04 & @xmath590.04 & @xmath590.25 & @xmath590.25 & @xmath590.02 + 7 & 0.31 @xmath160.05 & 1.35 @xmath160.24 & 0.31 @xmath160.07 & @xmath590.04 & @xmath590.25 & @xmath590.25 & @xmath590.02 + 8 & 0.26 @xmath160.04 & 0.48 @xmath160.14 & @xmath590.07 & @xmath590.04 & @xmath590.25 & @xmath590.25 & @xmath590.02 + & & & ghrs & & & & + 1 & & & @xmath378.0 & & & & + 2 & & & @xmath3713.7 & & & & + 3 & & & [email protected] & & & & + 4 & & & [email protected] & & & & + lcrll iue swp & 1150 1978 & @xmath3250 & 1978 june 20 1995 april 14@xmath69 & 2443679.5 2449821.5 + hut@xmath70 & 820 1840 & @xmath3450 & 1995 march 11 13 & 2449787.5 2449789.5 + fos g130h & 1150 1605 & @xmath31200 & 1995 december 30 & 2450081.5 + fos g130h & 1150 1605 & @xmath31200 & 1996 february 21 & 2450134.5 + fos g130h & 1150 1605 & @xmath31200 & 1996 april 13 & 2450186.5 + fos g130h & 1150 1605 & @xmath31200 & 1996 august 14 & 2450309.5 + fos g130h & 1150 1605 & @xmath31200 & 1996 november 28 & 2450415.5 + stis g140l & 1150 1715 & @xmath31000 & 1998 april 13 & 2450916.5 + stis g140l & 1150 1715 & @xmath31000 & 2000 april 17 & 2451651.5 + lccc & low - state models & & + 1a & 0.126 & 2.4e21 & 2.30e04 + 1b & 0.001 & 7.0e18 & 1.61e04 + 2 & 0.138 & 3.9e21 & 2.50e04 + 3@xmath74 & 0.185 & 1.4e21 & 2.78e04 + 5 & 1.58 & 1.5e19 & 4.25e04 + 6 & 8.50 & 1.5e20 & 1.05e05 + 7 & 10.7 & 2.0e21 & 1.39e05 + 8 & 10.7 & 5.5e20 & 1.26e05 + & high - state models & & + 1a & 0.793 & 2.4e21 & 4.74e04 + 1b & 0.008 & 7.0e18 & 1.89e04 + 2 & 0.669 & 2.4e21 & 4.86e04 + 3@xmath74 & 0.845 & 1.4e21 & 5.69e04 + 5 & 11.43 & 1.5e19 & 6.15e05 + lcccccc & & & low - state models ( stis epoch ) & & & + 1a & 7.52e02 & 6.85e01 & 2.03e03 & 1.96 & 0.09 & 0.05 + 1b & 3.06e02 & 0.03 & 2.72 & 0.46 & 2.23 & 0.14 + & ( @xmath375.99 ) & ( @xmath3720.48 ) & ( @xmath3710.26 ) & ( @xmath371.86 ) & ( 2.23 @xmath160.39 ) & ( 0.05 @xmath160.02 ) + 2 & 1.02e03 & 7.40e02 & 5.22e02 & 0.20 & 0.01 & + & ( @xmath376.56 ) & ( @xmath3721.55 ) & ( @xmath3713.75 ) & ( 0.15 @xmath160.05 ) & ( @xmath590.25 ) & ( @xmath590.02 ) + 3@xmath74 & 2.87e02 & 1.42e02 & 70.1 & & & + & ( @xmath373.34 ) & ( @xmath374.84 ) & ( @xmath376.20 ) & ( @xmath590.04 ) & ( @xmath590.25 ) & ( @xmath590.02 ) + 5 & 0.30 & 1.59 & 2.13 & & & + & ( 0.97 @xmath160.18 ) & ( 1.65 @xmath160.39 ) & ( 2.11 @xmath160.30 ) & ( @xmath590.04 ) & ( @xmath590.25 ) & ( @xmath590.02 ) + & & & high - state models ( ghrs epoch ) & & & + 1a & & & 4.58 & & & + 1b & & & 9.37 & & & + & & & ( @xmath378.0 ) & & & + 2 & & & 8.7 & & & + & & & ( @xmath3713.7 ) & & & + 3@xmath74 & & & 1.0 & & + & & & ( 0.9 @xmath160.14 ) & & & +

a family of classical integrable systems defined on a deformation of the two - dimensional sphere , hyperbolic and ( anti-)de sitter spaces is constructed through hamiltonians defined on the non - standard quantum deformation of a poisson coalgebra . all these spaces have a non - constant curvature that depends on the deformation parameter . as particular cases , the analogues of the harmonic oscillator and kepler coulomb potentials on such spaces are proposed . another deformed hamiltonian is also shown to provide superintegrable systems on the usual sphere , hyperbolic and ( anti-)de sitter spaces with a constant curvature that exactly coincides with . according to each specific space , the resulting potential is interpreted as the superposition of a central harmonic oscillator with either two more oscillators or centrifugal barriers . the non - deformed limit of all these hamiltonians can then be regarded as the zero - curvature limit ( contraction ) which leads to the corresponding ( super)integrable systems on the flat euclidean and minkowskian spaces . angel ballesteros , francisco j. herranz and orlando ragnisco _departamento de fsica , universidad de burgos , pza.misael bauelos s.n . , + e-09001 burgos , spain _ + e - mail : [email protected] , [email protected] _dipartimento di fisica , universit di roma tre and instituto nazionale di fisica nucleare sezione di roma tre , via vasca navale 84 , + i-00146 roma , italy _ + e - mail : [email protected] pacs : 02.30.lk 02.20.uw

math-ph0505081

one of the possible applications of quantum deformations of groups and algebras @xcite is the construction of classical and quantum integrable systems with an arbitrary number of degrees of freedom which was presented in @xcite . in this context , poisson coalgebras ( poisson algebras endowed with a compatible coproduct structure ) have been shown to generate in a systematic way certain ( super)integrable classical hamiltonian systems . in this construction , once a symplectic realization of the algebra is given , the generators of the poisson coalgebra play the role of dynamical symmetries of the hamiltonian , while the coproduct is used to ` propagate ' the integrability to arbitrary dimension . from this coalgebra approach , several well - known classical ( super)integrable systems have been recovered and some integrable deformations for them , as well as new integrable systems , have also been obtained @xcite . recently , this integrability - preserving deformation procedure has been used to introduce both superintegrable and integrable free motions on two - dimensional ( 2d ) spaces with curvature , either constant or variable , respectively @xcite . therefore one could expect that potential terms can also be considered , in such a way that the coalgebra approach should provide ( super)integrable potentials on curved spaces . the aim of the present paper is to prove this assertion through the construction of some relevant hamiltonians . in order to make these ideas more explicit , let us consider the non - standard quantum deformation of @xmath0 @xcite written as a poisson coalgebra @xmath7 with ( deformed ) poisson brackets , coproduct and casimir given by \{,}=2 z \{,}=-2\{,}=4 [ ba ] _ _ z(j_i)=j_i ^z + e^-z j_i i=+,3 [ bb ] _ z= -^2 [ bc ] where @xmath1 is a real deformation parameter . a two - particle symplectic realization of ( [ ba ] ) in terms of two canonical pairs of coordinates @xmath8 and momenta @xmath9 with respect to the usual poisson bracket \{f , g}=_i=1 ^ 2 ( - ) , [ ag ] and that depends on two real parameters @xmath10 , reads @xcite l + . [ be ] by substituting ( [ be ] ) in ( [ bc ] ) we obtain the two - particle casimir & & c_z = ( q_1p_2 - q_2p_1)^2 e^-z q_1 ^ 2e^z q_2 ^ 2 + ( _ 1 e^2z q_2 ^ 2 + _ 2 e^-2z q_1 ^ 2 ) & & + ( _ 1 + _ 2 ) e^-z q_1 ^ 2e^z q_2 ^ 2 [ bf ] which poisson - commutes with the generators ( [ be ] ) . the limit @xmath11 of such deformed generators leads to the well - known symplectic realization of @xmath0 : l . [ beeb ] the coalgebra approach @xcite ensures that _ any _ smooth function @xmath12 defined on ( [ be ] ) provides an integrable hamiltonian , for which @xmath13 is the constant of the motion . in this paper we shall study some choices for @xmath14 that lead to hamiltonians which are quadratic in the momenta and belong to the family _ z=12 f ( z ) + ( z ) [ ahaa ] where @xmath15 and @xmath16 are arbitrary smooth functions such that the @xmath17 is well defined and @xmath18 . therefore integrable deformations of the free motion of a particle on the 2d euclidean space are obtained from ( [ ahaa ] ) by setting @xmath19 and @xmath20 . two main representative cases appear @xcite : * the simplest _ hamiltonian + l h_z^i=12 + [ freei ] which defines the geodesic motion on a 2d riemannian space with metric s^2= e^-z q_2 ^ 2 q_1 ^ 2 + e^z q_1 ^ 2 q_2 ^ 2 [ cc ] and whose non - constant gaussian curvature @xmath21 is given by k(q_1,q_2)=- z ( z(q_1 ^ 2+q_2 ^ 2 ) ) . [ cd ] * the _ superintegrable _ hamiltonian + l h_z^s=12 e^ z + [ frees ] that leads to a riemannian metric of constant curvature which coincides with the deformation parameter , @xmath22 , namely s^2= e^-z q_1 ^ 2e^-2 z q_2 ^ 2 q_1 ^ 2 + e^-z q_2 ^ 2 q_2 ^ 2 . [ ec ] consequently , the ` classical ' limit @xmath2 corresponds to a zero - curvature limit . section 2 is devoted to the explicit solution of the geodesic flows on all these spaces , that complete the preliminary description given in @xcite and include deformations of the 2d sphere and hyperbolic spaces as well as of the @xmath23d ( anti-)de sitter spacetimes . the introduction of integrable potentials with coalgebra symmetry is then analysed by making use of the function @xmath16 and taking both parameters @xmath24 arbitrary . in fact , any @xmath16 such that @xmath25 can be interpreted as a deformation of the well - known 2d smorodinsky winternitz ( sw ) system @xcite formed by an isotropic harmonic oscillator with angular frequency @xmath26 plus two ` centrifugal terms ' governed by @xmath27 : ^sw = 12 ( p_1 ^ 2 + p_2 ^ 2 ) + + + _ 0 ( q_1 ^ 2+q_2 ^ 2 ) . [ ai ] on the other hand , analogues of the kepler coulomb ( kc ) potential can be obtained by considering any @xmath16 such that @xmath28 ( @xmath29 is another real constant ) : ^kc = 12 ( p_1 ^ 2 + p_2 ^ 2 ) + + - . [ aai ] in section 3 we shall propose the following _ integrable _ sw and kc systems , @xmath30 and @xmath31 , on the spaces of non - constant curvature previously defined through @xmath32 ( [ freei ] ) : & & h^_z=12 + _ 0 [ ahad ] + & & h^_z=12 - e^2 z . [ ahat ] the sw potential on the spaces of constant curvature with free motion given by @xmath33 ( [ frees ] ) will be introduced by means of the following choice for the hamiltonian ^_z=12 ^ z + _ 0 e^ z ^_z e^ z . [ bh ] we already know @xcite that this gives rise , under ( [ be ] ) , to a stckel - type system @xcite and so determines a _ superintegrable _ deformation of ( [ ai ] ) since , besides ( [ bf ] ) , there exists an additional constant of the motion given by _ z= e^z q_1 ^ 2 p_1 ^ 2 + e^z q_1 ^ 2+e^2 z q_1 ^ 2 . [ bjj ] note that this extra integral is _ not _ obtained from the coalgebra symmetry of the hamiltonian . parameters must vanishes and both the specific additional ` laplace - runge - lenz ' integral and the appropriate @xmath16 compatible with it should be previously obtained . ] section 4 is fully devoted to the study of @xmath34 , which is shown to provide a superintegrable system containing a ( curved ) harmonic oscillator together with two more potential terms ( either oscillators or centrifugal barriers ) on the usual sphere , hyperbolic and ( anti-)de sitter spaces . the explicit potentials are analysed in detail for each particular space . we stress that we recover known results on the riemannian spaces @xcite but also we obtain new ones on the relativistic spacetimes . finally , some remarks and comments close the paper . to start with , let us consider the metric ( [ cc ] ) defined by the free hamiltonian ( [ freei ] ) . let us call @xmath35 and introduce another parameter @xmath36 , such that @xmath37 and @xmath38 can be either real or pure imaginary numbers . then under the change of coordinates @xmath39 defined by @xcite ( _ 1 ) = \{z(q_1 ^ 2+q_2 ^ 2)}^z ^2(_2 ) = [ ce ] the metric ( [ cc ] ) and curvature ( [ cd ] ) are written as s^2= ( ^2 + _ 2 ^ 2 ^2 ) [ cg ] k()=-12 _ 1 ^ 2 . [ cj ] the product @xmath40 coincides with the metric of the 2d cayley klein ( ck ) spaces @xcite , all of them with constant curvature @xmath41 , provided that @xmath42 are identified with geodesic polar coordinates . hence @xmath37 , @xmath38 play the role of ( graded ) contraction parameters , determining the curvature and the signature of the metric , respectively . consequently , the metric ( [ cg ] ) can be interpreted as a deformation of the ck metric through the factor @xmath43 , which is responsible for the transition from the constant curvature to the non - constant one , or alternatively as a deformation of the _ flat _ euclidean ( @xmath38 real ) or minkowskian ( @xmath38 imaginary ) spaces , which are recovered under the classical limit @xmath2 . the expressions ( [ cg ] ) and ( [ cj ] ) are explicitly written for each particular ` deformed ' space in table [ table1 ] ; more details can be found in @xcite . from ( [ cg ] ) we compute the christoffel @xmath44 , riemann @xmath45 and ricci @xmath46 tensors @xcite ; their non - zero components turn out to be l + [ ch ] r^ _ = r _ = -12 _ 2 ^ 2 ^ 2(_1 ) ^2(_1 ) r^_=r_=-12 _ 1 ^ 2 ^2(_1 ) . [ ci ] the sectional and scalar curvatures are ( [ cj ] ) and @xmath47 , respectively . the connection ( [ ch ] ) allows us to write the geodesic equations + _ jk^i = 0 which , in our case , give rise to the following equations for @xmath48 where @xmath49 is the canonical parameter of the metric ( [ cg ] ) : & & -12 _ 1(_1 ) ( ) ^2 -(_1 ) ( 1+^2(_1 ) ) ( ) ^2=0 [ ck ] + & & + 2_1 ( ) = 0 . [ cl ] the latter equation has a first integral given by ( ) = 0 = -[cm ] where @xmath50 is a constant . by substituting this result in the metric ( [ cg ] ) we find the following velocities : l [ cn ] and the equation ( [ ck ] ) is fulfilled . thus the geodesic curve @xmath51 is obtained by eliminating the parameter @xmath49 and then integrating the equation ( ) ^2= ( - _ 2 ^ 2 ^ 2 ) [ co ] which finally leads to an involved solution that depends on elliptic functions . notice that all the expressions ( [ ck])([co ] ) are well defined under the limit @xmath52 . in this non - deformed / flat case we recover the known euclidean and minkowskian geodesics . explicitly , from ( [ cn ] ) and by taking the limit @xmath52 , we find that ^2=s^2+_2 ^ 2 ^ 2= [ cp ] where @xmath53 is the second integration constant . either from ( [ cp ] ) or from ( [ co ] ) with @xmath54 , we obtain the curve = . [ cq ] @xmath55 { \mbox { 2d deformed riemannian spaces}}&\quad{\mbox { $ ( 1 + 1)$d deformed relativistic spacetimes}}\\[4pt ] \hline \\[-6pt ] \mbox { $ \bullet$ deformed sphere $ { \bf s}^2_z$}&\quad\mbox { $ \bullet$ deformed anti - de sitter spacetime $ { \bf ads}^{1 + 1}_z$}\\[4pt ] z=-1;\ ( \la_1,\la_2)=({\rm i},1)&\quad z=-1;\ ( \la_1,\la_2)=({\rm i},{\rm i})\\[4pt ] \displaystyle{\d s^2 = \frac{1}{\cos \rr}\left ( \d \rr^2+\sin^2 \rr\,\d\te^2 \right ) } & \quad \displaystyle{\d s^2 = \frac{1}{\cos \rr}\left ( \d \rr^2-\sin^2 \rr\,\d\te^2 \right ) } \\[8pt ] \displaystyle{k = -\frac{\sin^2 \rr}{2\cos \rr } } & \quad \displaystyle{k = -\frac{\sin^2 \rr}{2\cos \rr } } \\[12pt ] \mbox { $ \bullet$ euclidean space $ { \bf e}^2$}&\quad\mbox { $ \bullet$ minkowskian spacetime $ { \bf m}^{1 + 1}$}\\[4pt ] z=0;\ ( \la_1,\la_2)=(0,1)&\quad z=0;\ ( \la_1,\la_2)=(0,{\rm i})\\[4pt ] \displaystyle{\d s^2 = \d \rr^2 + \rr^2\d\te^2 } & \quad \displaystyle{\d s^2 = \d \rr^2- \rr^2\d\te^2 } \\[2pt ] \displaystyle{k = 0 } & \quad \displaystyle{k = 0 } \\[6pt ] \mbox { $ \bullet$ deformed hyperbolic space $ { \bf h}_z^2$}&\quad\mbox { $ \bullet$ deformed de sitter spacetime $ { \bf ds}^{1 + 1}_z$}\\[4pt ] z=1;\ ( \la_1,\la_2)=(1,1)&\quad z=1;\ ( \la_1,\la_2)=(1,{\rm i})\\[4pt ] \displaystyle{\d s^2 = \frac{1}{\cosh \rr}\left ( \d \rr^2+\sinh^2 \rr\,\d\te^2 \right ) } & \quad \displaystyle{\d s^2 = \frac{1}{\cosh \rr}\left ( \d \rr^2-\sinh^2 \rr\,\d\te^2 \right ) } \\[8pt ] \displaystyle{k = -\frac{\sinh^2 \rr}{2\cosh \rr } } & \quad \displaystyle{k = -\frac{\sinh^2 \rr}{2\cosh \rr } } \\[8pt ] \hline \end{array}\ ] ] le us now consider now the metric ( [ ec ] ) that underlies the free superintegrable hamiltonian ( [ frees ] ) . under the change of coordinates ( [ ce ] ) we find that s^2= ( ^2 + _ 2 ^ 2 ^2 ) . [ ee ] by introducing a new radial coordinate @xmath56 defined by @xcite r=_0^ [ ef ] we obtain exactly the ck metric written in geodesic polar coordinates @xmath57 provided that now @xmath58 @xcite : s^2= r^2 + _ 2 ^ 2 ^2 . [ eh ] the non - zero connection and curvature tensors are given by ^r_=-_2 ^ 2 ( _ 1 r ) ^_r= [ ei ] r^r_r = r _ = _ 2 ^ 2 ^ 2(_1 r ) r^_rr = r_rr = _ 1 ^ 2z [ ej ] so that the sectional curvature is just @xmath22 . by applying the very same procedure described in section 2.1 we deduce the generic geodesic curve @xmath59 : = [ ek ] where @xmath50 and @xmath53 are integration constants . in the flat case @xmath2 the coordinate @xmath60 , so that the curve ( [ ek ] ) coincides with ( [ cq ] ) , as it should be . we specialize all this information for each space in table [ table3 ] . when comparing tables [ table1 ] and [ table3 ] , notice that the sign of @xmath1 for a given ` deformed ' space of non - constant curvature is the opposite to the corresponding one of constant curvature ( this is a consequence of the definition ( [ ef ] ) ) . @xmath55 { \mbox { 2d riemannian spaces}}&\quad{\mbox { $ ( 1 + 1)$d relativistic spacetimes}}\\[4pt ] \hline \\[-6pt ] \mbox { $ \bullet$ sphere $ { \bf s}^2 $ : $ ( \la_1,\la_2)=(1,1)$}&\quad\mbox { $ \bullet$ anti - de sitter spacetime $ { \bf ads}^{1 + 1}$ : $ ( \la_1,\la_2)=(1,{\rm i})$}\\[4pt ] \displaystyle{\d s^2 = \d r^2+\sin^2 r\,\d\te^2 } & \quad \displaystyle{\d s^2 = \d r^2-\sin^2 r\,\d\te^2 } \\[2pt ] \displaystyle { \gamma^r_{\te\te}=- \sin r \cos r \quad \gamma^\te_{\te r}= 1/\tan r } & \quad \displaystyle { \gamma^r_{\te\te}= \sin r \cos r \quad \gamma^\te_{\te r}= 1/\tan r } \\[2pt ] \displaystyle { \cte/ \tan r = \sin(\te+\te_0 ) } & \quad \displaystyle { \cte/ \tan r = \sinh(\te+\te_0 ) } \\[2pt ] \displaystyle{k = 1 } & \quad \displaystyle{k = 1 } \\[6pt ] \mbox { $ \bullet$ euclidean space $ { \bf e}^2 $ : $ ( \la_1,\la_2)=(0,1)$}&\quad\mbox { $ \bullet$ minkowskian spacetime $ { \bf m}^{1 + 1}$ : $ ( \la_1,\la_2)=(0,{\rm i})$}\\[4pt ] \displaystyle{\d s^2 = \d r^2 + r^2\d\te^2 } & \quad \displaystyle{\d s^2 = \d r^2- r^2\d\te^2 } \\[2pt ] \displaystyle { \gamma^r_{\te\te}=- r \quad \gamma^\te_{\te r}= 1/r } & \quad \displaystyle { \gamma^r_{\te\te}= r \quad \gamma^\te_{\te r}= 1/r } \\[2pt ] \displaystyle { \cte/ r = \sin(\te+\te_0 ) } & \quad \displaystyle { \cte/ r = \sinh(\te+\te_0 ) } \\[2pt ] \displaystyle{k = 0 } & \quad \displaystyle{k = 0 } \\[6pt ] \mbox { $ \bullet$ hyperbolic space $ { \bf h}^2 $ : $ ( \la_1,\la_2)=({\rm i},1)$}&\quad\mbox { $ \bullet$ de sitter spacetime $ { \bf ds}^{1 + 1}$ : $ ( \la_1,\la_2)=({\rm i},{\rm i})$}\\[4pt ] \displaystyle{\d s^2 = \d r^2+\sinh^2 r\,\d\te^2 } & \quad \displaystyle{\d s^2 = \d r^2-\sinh^2 r\,\d\te^2 } \\[2pt ] \displaystyle { \gamma^r_{\te\te}=- \sinh r \cosh r \quad \gamma^\te_{\te r}= 1/\tanh r } & \quad \displaystyle { \gamma^r_{\te\te}= \sinh r \cosh r \quad \gamma^\te_{\te r}= 1/\tanh r } \\[2pt ] \displaystyle { \cte/ \tanh r = \sin(\te+\te_0 ) } & \quad \displaystyle { \cte/ \tanh r = \sinh(\te+\te_0 ) } \\[2pt ] \displaystyle{k = -1 } & \quad \displaystyle{k = -1 } \\[6pt ] \hline \end{array}\ ] ] let @xmath61 be the canonical momenta corresponding to the new coordinates @xmath42 ( [ ce ] ) . the relationship between the initial phase space @xmath62 and the new one @xmath63 is found to be l + . [ dc ] therefore if we consider the generic integrable hamiltonian _ z^i=12 + ( z ) [ ahadgen ] and we perform the corresponding transformations we get & & h^i_z=12 ( _ 1 ) ( p_^2 + p_^2)+g ( ) + & & + ( + ) [ dd ] where @xmath64 and @xmath65 is an arbitrary smooth function . the corresponding constant of the motion follows from ( [ bf ] ) . if we define @xmath66 we find that _ z = p_^2++ [ de ] which does not depend on @xmath67 . furthermore this constant allows us to reduce ( [ dd ] ) to the 1d ( radial ) hamiltonian given by ^_z(,p_)=12 ( _ 1 ) p_^2 + c_z+ g ( ) . [ dede ] consequently , the hamiltonian ( [ dd ] ) defines a family of integrable systems which , for any @xmath68 , share the same constant of the motion ( [ de ] ) . we specialize in table [ table2 ] these expressions for each of the six spaces shown in table [ table1 ] . @xmath55 \mbox { $ \bullet$ deformed sphere $ { \bf s}^2_z$}&\quad\mbox { $ \bullet$ deformed anti - de sitter spacetime $ { \bf ads}^{1 + 1}_z$}\\[4pt ] \displaystyle { { h}^{\in}_z=\frac 12 \cos \rr \left(p_\rr^2 + \frac{1 } { \sin^2 \rr } \ , p_\te^2\right)+g(\rr ) } & \quad \displaystyle { { h}^{\in}_z=\frac 12 \cos \rr \left(p_\rr^2 -\frac{1 } { \sin^2 \rr } \ , p_\te^2\right)+g(\rr ) } \\[8pt ] \displaystyle{\qquad\quad + \frac{2 \cos \rr } { \sin^2 \rr}\left ( \frac{\otra_1}{\sin^2\te}+\frac{\otra_2}{\cos^2 \te } \right ) } & \quad \displaystyle{\qquad\quad -\frac{2 \cos \rr } { \sin^2 \rr}\left(\frac{\otra_1}{\sinh^2\te } -\frac{\otra_2}{\cosh^2 \te}\right ) } \\[8pt ] \displaystyle{\qquad\ , = \frac 12 \cos \rr \ , p_\rr^2 + \frac{\cos \rr}{2 \sin^2 \rr } \,{c}_z + g(\rr ) } & \quad \displaystyle{\qquad\ , = \frac 12 \cos \rr \ , p_\rr^2 -\frac{\cos \rr}{2 \sin^2 \rr } \,{c}_z + g(\rr ) } \\[10pt ] \displaystyle { { c}_z = p_\te^2+\frac{4 \otra_1}{\sin^2 \te } + \frac{4 \otra_2}{\cos^2 \te } } & \quad \displaystyle { { c}_z = p_\te^2+\frac{4 \otra_1}{\sinh^2 \te } -\frac{4 \otra_2}{\cosh^2 \te } } \\[14pt ] \mbox { $ \bullet$ euclidean space $ { \bf e}^2$}&\quad\mbox { $ \bullet$ minkowskian spacetime $ { \bf m}^{1 + 1}$}\\[4pt ] \displaystyle { { h}^{\in } = \frac 12 \left(p_\rr^2 + \frac{1}{\rr^2 } \ , p_\te^2\right)+g(\rr ) } & \quad \displaystyle { { h}^{\in } = \frac 12 \left(p_\rr^2 -\frac{1 } { \rr^2 } \ , p_\te^2\right)+g(\rr ) } \\[8pt ] \displaystyle{\qquad\quad + \frac{2 } { \rr^2}\left ( \frac{\otra_1}{\sin^2\te}+\frac{\otra_2}{\cos^2 \te } \right ) } & \quad \displaystyle { \qquad\quad -\frac{2 } { \rr^2}\left(\frac{\otra_1}{\sinh^2\te } -\frac{\otra_2}{\cosh^2 \te}\right ) } \\[8pt ] \displaystyle{\qquad\ , = \frac 12 \ , p_\rr^2 + \frac{1}{2 \rr^2 } \,{c}_z + g(\rr ) } & \quad \displaystyle{\qquad\ , = \frac 12 \ , p_\rr^2 -\frac{1}{2 \rr^2 } \,{c}_z + g(\rr ) } \\[10pt ] \displaystyle { { c } = p_\te^2+\frac{4 \otra_1}{\sin^2 \te } + \frac{4 \otra_2}{\cos^2 \te } } & \quad \displaystyle { { c } = p_\te^2+\frac{4 \otra_1}{\sinh^2 \te } -\frac{4 \otra_2}{\cosh^2 \te } } \\[14pt ] \mbox { $ \bullet$ deformed hyperbolic space $ { \bf h}_z^2$}&\quad\mbox { $ \bullet$ deformed de sitter spacetime $ { \bf ds}^{1 + 1}_z$}\\[4pt ] \displaystyle { { h}^{\in}_z=\frac 12 \cosh \rr \left(p_\rr^2 + \frac{1 } { \sinh^2 \rr } \ , p_\te^2\right)+g(\rr ) } & \quad \displaystyle { { h}^{\in}_z=\frac 12 \cosh \rr \left(p_\rr^2 -\frac{1 } { \sinh^2 \rr } \ , p_\te^2\right)+g(\rr ) } \\[8pt ] \displaystyle{\qquad\quad + \frac{2 \cosh \rr } { \sinh^2 \rr}\left ( \frac{\otra_1}{\sin^2\te}+\frac{\otra_2}{\cos^2 \te } \right ) } & \quad \displaystyle{\qquad\quad -\frac{2 \cosh \rr } { \sinh^2 \rr}\left(\frac{\otra_1}{\sinh^2\te } -\frac{\otra_2}{\cosh^2 \te}\right ) } \\[8pt ] \displaystyle{\qquad\ , = \frac 12 \cosh \rr \ , p_\rr^2 + \frac{\cosh \rr}{2 \sinh^2 \rr } \,{c}_z + g(\rr ) } & \quad \displaystyle{\qquad\ , = \frac 12 \cosh \rr \ , p_\rr^2 -\frac{\cosh \rr}{2 \sinh^2 \rr } \,{c}_z + g(\rr ) } \\[10pt ] \displaystyle { { c}_z = p_\te^2+\frac{4 \otra_1}{\sin^2 \te } + \frac{4 \otra_2}{\cos^2 \te } } & \quad \displaystyle { { c}_z = p_\te^2+\frac{4 \otra_1}{\sinh^2 \te } -\frac{4 \otra_2}{\cosh^2 \te } } \\[8pt ] \hline \end{array}\ ] ] amongst the possible choices for the deformed hamiltonian ( [ ahadgen ] ) , let us consider the expression ( [ ahad ] ) whose non - deformed limit is the sw system ( [ ai ] ) . this choice implies that the potential function @xmath68 appearing in ( [ dd ] ) readsg()=_0(_1 ) . [ df ] this gives ^_z=(_1 ) h^_z [ dg ] where & & h^_z = 12 ( p_^2 + p_^2)+_0 + & & + ( + ) . [ dh ] the physical interpretation of ( [ dg ] ) is the following . * when @xmath69 @xmath70 , the expression ( [ dg ] ) reduce to the sw hamiltonian on @xmath71 and @xmath72 written in polar coordinates ^^ = 12 ( p_^2 + ) + _ 0 ^2 + ( + ) . [ di ] the term @xmath73 is the usual harmonic oscillator potential , while those depending on @xmath74 and @xmath75 are two ` centrifugal barriers ' . * when @xmath76 , @xmath77 is well - known in the literature as the superintegrable sw system on spaces of constant curvature @xcite , where the @xmath78-term corresponds to a ` curved ' harmonic oscillator potential . in particular , if @xmath79 we recover the spherical oscillator or higgs potential @xcite , @xmath80 , on @xmath81 and @xmath82 , while @xmath83 leads to a hyperbolic oscillator @xmath84 on @xmath85 and @xmath86 . hence @xmath87 can properly be regarded as an integrable generalization of the superintegrable @xmath88 to spaces of non - constant curvature . we stress that @xmath88 can be reproduced from the hamiltonian ( [ bh ] ) in an exact way and it will be analysed in the next section . if we now consider the hamiltonian ( [ ahat ] ) whose non - deformed limit is the kc system ( [ aai ] ) , we get g()=-(_1 ) [ dj ] with @xmath89 . with this choice we find that ^ikc_z=(_1 ) h^kc_z [ dl ] where & & h^kc_z = 12 ( p_^2 + p_^2)- [ ddll ] + & & + ( + ) and the integrable ( but not superintegrable ) hamiltonian @xmath90 contains a kc potential in polar coordinates through the @xmath91-term . explicitly , * when @xmath69 we obtain the reduction ^ikc^kc = 12 ( p_^2 + ) - + ( + ) [ ddi ] which defines an integrable system formed by a composition of the kc potential , @xmath92 , with two centrifugal barriers on the flat spaces @xmath71 and @xmath72 . this hamiltonian is superintegrable whenever at least one of the parameters @xmath74 , @xmath75 is taken equal to zero ( see , e.g. , @xcite ) . * on the contrary , if @xmath76 then @xmath93 and the latter contains the kc potential on spaces of constant curvature @xcite in its spherical version , @xmath94 , on @xmath81 @xcite and @xmath82 @xmath95 , as well as in its hyperbolic one , @xmath96 , on @xmath85 and @xmath97 @xmath98 . therefore we can conclude that @xmath99 ( [ dl ] ) is an appropriate generalization of the kc potential to spaces of non - constant curvature , possibly supplemented by two more potential terms . now we can reproduce the same study in the case of the superintegrable hamiltonian ( [ bh ] ) defined on the spaces of constant curvature of section 2.2 . the relation between the initial phase space @xmath100 and the canonical geodesic polar coordinates @xmath57 and momenta @xmath101 turns out to be l + . [ fj ] by considering the realization ( [ be ] ) , the change of coordinates defined through the composition of ( [ ce ] ) and ( [ ef ] ) together with the above relations , it can be shown that the hamiltonian @xmath102 ( [ bh ] ) and its constants of the motion @xmath13 ( [ bf ] ) and @xmath103 ( [ bjj ] ) are expressed in the phase space @xmath104 as & & h^sw_z=12 ( p_r^2 + p_^2)+ _ 0 + & & + ( + ) [ fl ] + & & c_z = p_^2 + + [ fm ] + & & i_z=(_2(_2 ) p_r+p_)^2 + & & + 2_0_2 ^ 2 ^ 2(_2 ) + [ fn ] where we have rescaled these quantities in the following way : l + _ 1=2_2_2=2_1 . [ fk ] the constant of the motion @xmath105 allows us to reduce @xmath106 to a 1d radial system given by ^_z=12 p_r^2 + c_z + _ 0 . [ ffnn ] these results are displayed in table [ table4 ] for each space of constant curvature . the superintegrable hamiltonians written on the first column of table [ table4 ] are constructed on the three classical _ riemannian spaces _ ( @xmath107 ) and they are already known @xcite . we recall that all these results were obtained by applying different procedures that are not related to coalgebra symmetry . in geodesic polar coordinates these systems were also constructed in @xcite through a lie group approach , where the constants of the motion there denoted @xmath108 , @xmath109 are related to ( [ fm ] ) and ( [ fn ] ) by @xmath110 , @xmath111 . the hamiltonian @xmath106 on @xmath81 has been interpreted in @xcite as a superposition of three spherical oscillators ; the corresponding geometrical interpretation on @xmath85 can be found in @xcite . we stress that the second column of table [ table4 ] provides the coalgebraic generalization of the sw potential to the three ` classical ' _ relativistic spacetimes _ ( @xmath112 ) which , to our knowledge , was still lacking . in the following we shall describe the system @xmath106 on the six spaces , that includes a full description of the new cases ( on @xmath82 , @xmath113 and @xmath86 ) , as well as of the known potentials on @xmath114 , @xmath71 and @xmath85 . we point out that when the six spaces are considered altogether the interpretation becomes more comprehensible and transparent . let us firstly recall which is the ( physical ) geometrical role of the geodesic polar coordinates @xmath57 on the spaces of constant curvature of table [ table3 ] . these can be embedded in a 3d linear ambient space with coordinates @xmath115 given by @xcite x_0=(_1 r)x_1= ( _ 2 ) x_2= [ ga ] which fulfil the constraint @xmath116 ; the origin has ambient coordinates @xmath117 . consider a ( time - like ) geodesic @xmath118 , another ( space - like ) geodesic @xmath119 orthogonal at @xmath120 and a generic point @xmath121 with coordinates ( [ ga ] ) . then @xmath56 is the ( time - like ) distance along the geodesic @xmath122 that joins @xmath121 with @xmath120 , while @xmath123 is the ( rapidity ) angle of @xmath122 with respect to @xmath118 as shown in figure 1 . in the riemannian cases with @xmath107 the coordinates @xmath57 parametrize the complete space , while in the relativistic cases with @xmath112 , it is verified that @xmath124 , so that @xmath57 only cover the time - like lines limited by the isotropic lines @xmath125 on which @xmath126 . next , let @xmath127 ( @xmath128 ) be the intersection point of @xmath129 with its orthogonal geodesic through @xmath121 . then if @xmath130 denotes the ( time - like ) distance @xmath131 and @xmath132 the ( space - like ) distance @xmath133 , it can be shown that @xcite x_1= ( _ 2 ) x_2= . [ gb ] hence the potential of @xmath106 ( [ fl ] ) can be rewritten as ^_z=_0 + + [ gc ] which conveys a common interpretation on the six spaces : @xmath134 the @xmath78-term is a _ central _ harmonic oscillator , that is , with center at the origin @xmath120 . @xmath134 both @xmath135-terms @xmath136 are ` centrifugal barriers ' . alternatively , the @xmath135-terms may adopt a different interpretation in each particular space , depicted in figure 1 , that we proceed to study . all the trigonometric relations on these spaces that are necessary in our description can be found in @xcite . let @xmath137 @xmath136 be the intersection points of the geodesics @xmath129 with the equator of the sphere ( at a distance @xmath138 from @xmath120 ) and @xmath139 the distances along the geodesics joining @xmath121 and @xmath137 . by applying the cosine theorem for the two triangles @xmath140 we find that l r_1= 2 r+2r + r_2= 2 r+2r(2- ) [ gd ] that is r_1= rx r_2= ry . [ ge ] the same result follows by noticing that @xmath141 and that @xmath142 and @xmath143 lie on a same geodesic orthogonal to @xmath119 and @xmath118 , respectively . therefore the potential ( [ gc ] ) on @xmath81 can be expressed in two ways & & u^_z=_0 ^ 2 r++ [ gf ] + & & = _ 0 ^ 2 r+_1 ^ 2 r_1+_2 ^ 2 r_2+_1+_2 [ gg ] which show a superposition of the central spherical oscillator either with two spherical centrifugal barriers , or with two spherical oscillators with centers placed at @xmath137 @xcite . in this case , the analogous points to the previous ` centers ' @xmath137 would be beyond the ` actual ' hyperbolic space and so placed in the ` ideal ' one @xcite ( in the exterior region of @xmath85 ) . thus we only write the potential in the form ( [ gc ] ) : ^_z=_0 ^ 2 r++ [ gh ] which corresponds to a superposition of a hyperbolic oscillator centered at the origin with two hyperbolic ` centrifugal barriers ' @xcite . the limit @xmath2 of the potential @xmath144 on @xmath81 and @xmath85 can properly be performed on both expressions ( [ gf ] ) and ( [ gh ] ) leading to the usual harmonic oscillator @xmath145 and centrifugal potentials @xmath146 and @xmath147 , in such a manner that @xmath148 are cartesian coordinates on @xmath71 . thus the proper sw system is recovered @xcite . nevertheless , this contraction can not directly be applied to the potential on @xmath81 written in terms of @xmath139 ( [ gg ] ) ; in fact , when @xmath2 the points @xmath149 . we consider the intersection point @xmath150 between the time - like geodesic @xmath118 and the axis @xmath151 of the ambient space , which is at a time - like distance @xmath152 from the origin @xmath120 . if @xmath153 denotes the time - like distance @xmath154 , by applying the cosine theorem on the triangle @xmath155 , we find that l r_1= 2r + 2 r r_1= rx . [ gi ] that is , @xmath156 . therefore the potential becomes & & u^_z=_0 ^ 2 r+- [ gj ] + & & = _ 0 ^ 2 r+_1 ^ 2 r_1- + _ 1 . [ gk ] the former expression exhibits a superposition of a time - like ( spherical ) oscillator centered at @xmath120 with a time - like ( spherical ) centrifugal potential and a space - like ( hyperbolic ) one . under the latter form , the time - like centrifugal term is transformed into another spherical oscillator now with center at @xmath150 . recall that @xmath82 and @xmath86 are related by means of an interchanging between time - like lines and space - like ones ; the former are compact ( circular ) on @xmath82 and non - compact ( hyperbolic ) on @xmath86 , while the latter are non - compact on @xmath82 but compact on @xmath86 . so , conversely to previous case , we consider the intersection point @xmath157 between the space - like geodesic @xmath119 and the axis @xmath158 ( at a space - like distance @xmath138 from @xmath120 ) , such that @xmath159 is the space - like distance @xmath160 . the cosine theorem applied to the triangle @xmath161 ( with external angle @xmath123 ) gives rise to l r_2= 2r + 2 r r_2= ry . [ gl ] note that @xmath162 . hence we again find two ways to expressed the potential on @xmath86 & & u^_z=_0 ^ 2 r+- [ gm ] + & & = _ 0 ^ 2 r+-_2 ^ 2 r_2 -_2 [ gn ] that is , a superposition of a central time - like ( hyperbolic ) oscillator with a time - like ( hyperbolic ) centrifugal barrier , and either with another space - like ( spherical ) centrifugal barrier or with a space - like ( spherical ) oscillator centered at @xmath157 . the limit @xmath2 of ( [ gj ] ) and ( [ gm ] ) provides the corresponding sw potential on @xmath113 , which is formed by a time - like harmonic oscillator @xmath163 , one time - like centrifugal barrier @xmath146 together with another space - like one @xmath164 . the coordinates @xmath148 are the usual time and space ones . on the contrary , the expressions ( [ gk ] ) and ( [ gn ] ) are not well defined when @xmath2 since the points @xmath150 and @xmath157 go to infinity . throughout this paper we have constructed several hamiltonians that belong to the quite general family @xmath165 within the realization ( [ be ] ) . there are certainly many other possible choices that lead to integrable systems on spaces of non - constant curvature with deformed @xmath0 coalgebra symmetry . nevertheless , the explicit form of such a general hamiltonian can be restricted by taking into account the following requirements : ( i ) the kinetic energy is quadratic in the momenta , ( ii ) the limit @xmath2 of the underlying deformed spaces leads to @xmath166 , ( iii ) dimensions of the deformation parameter are @xmath167=[j_-]^{-1}=[q_i]^{-2}$ ] , and ( iv ) the potential only depends on the coordinates . under these assumptions the most general integrable hamiltonian is just ( [ ahaa ] ) ; notice that the case @xmath168 is transformed into a particular case of ( [ ahaa ] ) through the casimir @xmath13 . hence the ( deformed ) kinetic energy and potential of the resulting hamiltonian , @xmath169 , turn out to be & & t_z = 12 ( e^z q_2 ^ 2 p_1 ^ 2 + e^-z q_1 ^ 2 p_2 ^ 2 ) f(z(q_1 ^ 2+q_2 ^ 2 ) ) [ zb ] + & & v_z =( e^z q_2 ^ 2 + e^-z q_1 ^ 2 ) f(z(q_1 ^ 2+q_2 ^ 2 ) ) + ( z(q_1 ^ 2+q_2 ^ 2 ) ) . [ zc ] we remark that the general expression for the gaussian curvature of the 2d space associated with the kinetic energy ( [ zb ] ) can be found in @xcite . then it can be checked that in order to obtain a space of constant curvature from @xmath170 the choice is quite singular ( e.g. , @xmath171 ) , and in general one obtains very involved spaces of non - constant curvature for which the simplest choice is the one we have developed in this paper with @xmath172 . however other deformed spaces underlying ( [ zb ] ) and particular potentials contained in ( [ zc ] ) could be worth to be studied . other approaches to superintegrability on 2d spaces of variable curvature can be found in @xcite . on the other hand , we stress that by introducing a second parameter @xmath38 , that determines the signature of the metric , we have been able to obtain integrable potentials on relativistic spacetimes of non - constant curvature ; in this context , the deformed kc potential could be of interest in classical gravity . furthermore the known superintegrable sw potential on riemannian spaces have also been implemented on the three classical relativistic spacetimes of constant curvature . notice that we have avoided the contraction @xmath173 , which is well defined on both metrics ( [ cg ] ) and ( [ ee ] ) , since this would give rise to degenerate ( newtonian ) metrics , whose dynamical contents are not so interesting . finally , we recall that the existence of an underlying coalgebra symmetry for all these two - particle hamiltonians ensures that they can be generalized to @xmath174-dimensional spaces through the coproduct . in fact , the corresponding expressions in terms of the initial phase space @xmath100 can be found in @xcite . nevertheless the geometrical and physical description of the corresponding hamiltonians on @xmath174d curved riemannian and relativistic spaces ( thus including a proper study of sectional curvatures ) remains as an open problem which is currently under investigation . this work was partially supported by the ministerio de educacin y ciencia ( spain , project fis2004 - 07913 ) , by the junta de castilla y len ( spain , project bu04/03 ) , and by the infn - cicyt ( italy - spain ) . ballesteros a , herranz f j , musso f and ragnisco o 2004 in _ superintegrability in classical and quantum systems _ crm proceedings and lecture notes * 37 * ed p tempesta _ et al _ ( providence , ri : american mathematical society ) p 1 ( _ preprint _ math - ph/0412067 ) herranz f j , ballesteros a , santander m and sanz - gil t 2004 in _ superintegrability in classical and quantum systems _ crm proceedings and lecture notes * 37 * ed p tempesta _ et al _ ( providence , ri : american mathematical society ) p 75 ( _ preprint _ math - ph/0501035 ) \mbox { $ \bullet$ sphere $ { \bf s}^2$}&\quad\mbox { $ \bullet$ anti - de sitter spacetime $ { \bf ads}^{1 + 1}$}\\[4pt ] \displaystyle { { h}^{\sstc}_z=\frac 12 \left(p_r^2 + \frac{1 } { \sin^2 r } \ , p_\te^2\right)+\beta_0\tan^2 r } & \quad \displaystyle { { h}^{\sstc}_z=\frac 12 \left(p_r^2 -\frac{1 } { \sin^2 r } \ , p_\te^2\right)+\beta_0\tan^2 r } \\[8pt ] \displaystyle{\qquad\quad + \frac{1}{\sin^2 r}\left ( \frac{\beta_1}{\cos^2\te}+\frac{\beta_2}{\sin^2 \te } \right ) } & \quad \displaystyle{\qquad\quad + \frac{1}{\sin^2 r}\left ( \frac{\beta_1}{\cosh^2\te}-\frac{\beta_2}{\sinh^2 \te } \right ) } \\[8pt ] \displaystyle{\qquad\ \,= \frac 12 \,p_r^2 + \frac{1}{2\sin^2 r}\ , { c}_z+ \beta_0\tan^2 r } & \quad \displaystyle{\qquad\ \,= \frac 12 \,p_r^2 -\frac{1}{2\sin^2 r}\ , { c}_z+ \beta_0\tan^2 r } \\[10pt ] \displaystyle { { c}_z = p_\te^2 + \frac{2\beta_1 } { \cos^2 \te } + \frac{2 \beta_2}{\sin^2 \te } } & \quad \displaystyle { { c}_z = p_\te^2- \frac{2\beta_1 } { \cosh^2 \te } + \frac{2 \beta_2}{\sinh^2 \te } } \\[10pt ] \displaystyle { { i}_z=\left(\sin\te \,p_r+\frac{\cos\te}{\tan r}\,p_\te \right)^2 } & \quad \displaystyle { { i}_z=\left(\sinh\te \,p_r-\frac{\cosh\te}{\tan r}\,p_\te \right)^2 } \\[8pt ] \displaystyle { \qquad + 2\beta_0\tan^2 r\sin^2\te + \frac{2\beta_2}{\tan^2 r\sin^2\te } } & \quad \displaystyle { \qquad + 2\beta_0\tan^2 r\sinh^2\te + \frac{2\beta_2}{\tan^2 r\sinh^2\te } } \\[14pt ] \mbox { $ \bullet$ euclidean space $ { \bf e}^2$}&\quad\mbox { $ \bullet$ minkowskian spacetime $ { \bf m}^{1 + 1}$}\\[4pt ] \displaystyle { { h}^{\sstc}=\frac 12 \left(p_r^2 + \frac{1 } { r^2 } \ , p_\te^2\right)+\beta_0 r^2 } & \quad \displaystyle { { h}^{\sstc } = \frac 12 \left(p_r^2 -\frac{1 } { r^2 } \ , p_\te^2\right)+\beta_0 r^2 } \\[8pt ] \displaystyle{\qquad\quad + \frac{1 } { r^2}\left ( \frac{\beta_1}{\cos^2\te}+\frac{\beta_2}{\sin^2 \te } \right ) } & \quad \displaystyle{\qquad\quad + \frac{1 } { r^2}\left ( \frac{\beta_1}{\cosh^2\te}-\frac{\beta_2}{\sinh^2 \te } \right ) } \\[8pt ] \displaystyle{\qquad\ \,= \frac 12 \,p_r^2 + \frac{1}{2 r^2}\ , { c}_z+ \beta_0 r^2 } & \quad \displaystyle{\qquad\ \,= \frac 12 \,p_r^2 -\frac{1}{2 r^2}\ , { c}_z+ \beta_0 r^2 } \\[10pt ] \displaystyle { { c } = p_\te^2 + \frac{2\beta_1 } { \cos^2 \te } + \frac{2 \beta_2}{\sin^2 \te } } & \quad \displaystyle { { c } = p_\te^2- \frac{2\beta_1 } { \cosh^2 \te } + \frac{2 \beta_2}{\sinh^2 \te } } \\[10pt ] \displaystyle { { i } = \left(\sin\te \,p_r+\frac{\cos\te } { r}\,p_\te \right)^2 } & \quad \displaystyle { { i } = \left(\sinh\te \,p_r-\frac{\cosh\te } { r}\,p_\te \right)^2 } \\[8pt ] \displaystyle { \qquad + 2\beta_0 r^2\sin^2\te + \frac{2\beta_2 } { r^2\sin^2\te } } & \quad \displaystyle { \qquad + 2\beta_0 r^2\sinh^2\te + \frac{2\beta_2 } { r^2\sinh^2\te } } \\[14pt ] \mbox { $ \bullet$ hyperbolic space $ { \bf h}^2$}&\quad\mbox { $ \bullet$ de sitter spacetime $ { \bf ds}^{1 + 1}$}\\[4pt ] \displaystyle { { h}^{\sstc}_z=\frac 12 \left(p_r^2 + \frac{1 } { \sinh^2 r } \ , p_\te^2\right)+\beta_0\tanh^2 r } & \quad \displaystyle { { h}^{\sstc}_z=\frac 12 \left(p_r^2 -\frac{1 } { \sinh^2 r } \ , p_\te^2\right)+\beta_0\tanh^2 r } \\[8pt ] \displaystyle{\qquad\quad + \frac{1}{\sinh^2 r}\left ( \frac{\beta_1}{\cos^2\te}+\frac{\beta_2}{\sin^2 \te } \right ) } & \quad \displaystyle{\qquad\quad + \frac{1}{\sinh^2 r}\left ( \frac{\beta_1}{\cosh^2\te}-\frac{\beta_2}{\sinh^2 \te } \right ) } \\[8pt ] \displaystyle{\qquad\ \,= \frac 12 \,p_r^2 + \frac{1}{2\sinh^2 r}\ , { c}_z+ \beta_0\tanh^2 r } & \quad \displaystyle{\qquad\ \,= \frac 12 \,p_r^2 -\frac{1}{2\sinh^2 r}\ , { c}_z+ \beta_0\tanh^2 r } \\[10pt ] \displaystyle { { c}_z = p_\te^2 + \frac{2\beta_1 } { \cos^2 \te } + \frac{2 \beta_2}{\sin^2 \te } } & \quad \displaystyle { { c}_z = p_\te^2- \frac{2\beta_1 } { \cosh^2 \te } + \frac{2 \beta_2}{\sinh^2 \te } } \\[10pt ] \displaystyle { { i}_z=\left(\sin\te \,p_r+\frac{\cos\te}{\tanh r}\,p_\te \right)^2 } & \quad \displaystyle { { i}_z=\left(\sinh\te \,p_r-\frac{\cosh\te}{\tanh r}\,p_\te \right)^2 } \\[8pt ] \displaystyle { \qquad + 2\beta_0\tanh^2 r\sin^2\te + \frac{2\beta_2}{\tanh^2 r\sin^2\te } } & \quad \displaystyle { \qquad + 2\beta_0\tanh^2 r\sinh^2\te + \frac{2\beta_2}{\tanh^2 r\sinh^2\te } } \\[8pt ] \hline \end{array}\ ] ] figure 1 . geometrical description of the sw potential on the six spaces of tables [ table3 ] and [ table4 ] in a 3d linear ambient space . in the three spacetimes ( right ) space - like lines / distances are represented by dashed lines

a nonlinear electromagnetic scattering problem is studied in the presence of bound states in the radiation continuum . it is shown that the solution is not analytic in the nonlinear susceptibility and the conventional perturbation theory fails . a non - perturbative approach is proposed and applied to the system of two parallel periodic arrays of dielectric cylinders with a second order nonlinear susceptibility . this scattering system is known to have bound states in the radiation continuum . in particular , it is demonstrated that , for a wide range of values of the nonlinear susceptibility , the conversion rate of the incident fundamental harmonic into the second one can be as high as 40% when the distance between the arrays is as low as a half of the incident radiation wavelength . the effect is solely attributed to the presence of bound states in the radiation continuum .

1107.0468

a conventional approach to nonlinear electromagnetic scattering problems is based on the power series expansion in a nonlinear susceptibility @xmath0 . for example , for the 2nd order susceptibility , the physical parameter that determines nonlinear effects is @xmath1 where @xmath2 is the electric field at the scattering structure . the smallness of @xmath3 justifies the use of perturbation theory and the solution is analytic in @xmath0 . the situation is different if the scattering structure has resonances . planar periodic structures ( e.g. , gratings ) are known to exhibit sharp scattering resonances when illuminated by electromagnetic waves ( for a review see , e.g. , @xcite ) . furthermore , it is known ( see , e.g. , @xcite ) that in such structures a local electromagnetic field @xmath2 is amplified if the structure has narrow resonances : @xmath4 , where @xmath5 is the amplitude of the incident wave and @xmath6 is the width of the resonance . consequently , optical nonlinear effects are amplified if the system has a sufficiently narrow resonance . for example , an amplification of the second harmonics generation by a single periodic array of dielectric cylinders @xcite and by other single - array periodic systems @xcite have been reported , but no significant flux conversion rate , comparable to that in conventional methods of second harmonic generation , has been found . owing to the smallness of @xmath0 and a _ finite _ @xmath7 , the condition @xmath8 still holds for the studied structures . if two identical planar periodic structures are aligned parallel and separated by a distance @xmath9 , then it can be shown that for each resonance associated with the single structure , the combined structure has two close resonances whose width depends continuously on @xmath10 so that the width of one of the resonances vanishes , i.e. , @xmath11 as @xmath12 for some discrete set of distances @xmath13 , for sufficiently large @xmath10 @xcite . this means that the system has _ bound states in the radiation continuum_. their existence was first predicted in quantum mechanics by von neumann and wigner @xcite in 1929 and later they were discovered in some atomic systems @xcite ( see also @xcite for more theoretical studies ) . their analog in maxwell s theory has only attracted attention recently @xcite . in particular , for a system of two parallel arrays of periodically positioned subwavelength dielectric cylinders ( depicted in panel ( a ) of fig . [ fig:1 ] ) , the existence of bound states in the radiation continuum has been first established in numerical studies of the system @xcite . a complete classification of bound states as well as their analytic form for this system is given in @xcite for tm polarization . it is also shown @xcite that bound states exist in the spectral range in which more than one diffraction channel are open . from the physical point of view , bound states in the radiation continuum are localized solutions of maxwell s equations like waveguide modes , but in contrast to the latter their spectrum lies in the spectrum of scattering radiation ( diffraction ) modes . the perturbation theory parameter @xmath14 , as defined above , can no longer be considered small if bound states in the radiation continuum are present . this qualitative assessment should be taken with a precaution . in the present study , a rigorous analysis of the nonlinear scattering problem by means of the formalism of siegert states ( appropriately extended to periodic structures @xcite ) shows that no divergence of a local field occurs as @xmath12 . however , the conventional perturbative approach fails because the solution is not analytic in @xmath0 . the situation can be compared with a simple mechanical analog . consider a scattering problem for a particle on a line in a hard core repulsive potential @xmath15 , @xmath16 . no matter how small @xmath17 is , the particle never crosses the origin @xmath18 and a full reflection occurs , but it does so when @xmath19 ( a full transmission ) . so , the scattering amplitude is not analytic in @xmath17 . other , more sophisticated , examples of quantum systems with such properties are studied in @xcite . the purpose of the present study is twofold . first , the nonlinear scattering problem is studied in _ the presence of bound states in the radiation continuum_. a non - perturbative approach is developed to solve the problem . second , as an application of the developed formalism , the problem of the second harmonic generation is analyzed with an example of the system depicted in fig . [ fig:1 ] ( panel ( a ) ) . in section [ sec:1 ] , the nonlinear resonant scattering problem with bound states in the radiation continuum is transformed into a system of integral equations . a non - perturbative method is proposed to solve these equations in the approximation that takes into account two nonlinear effects : a second harmonic generation in the leading order of @xmath0 , and the fundamental harmonic generation by mixing the second and fundamental harmonics in the leading order in @xmath20 . this is the second order effect in @xmath0 known in the theory as the _ optical rectification_. the latter is shown to be necessary to ensure the energy flux conservation . the formalism is illustrated with an example of two parallel periodic arrays of dielectric cylinders shown in panel ( a ) of fig . [ fig:1 ] . the analysis is based on the subwavelength approximation ( section [ sec:2 ] ) when the incident wave length is larger than the radius @xmath21 of the cylinders . if @xmath22 is the magnitude of the wave vector , then the theory has three small parameters : @xmath23 where all the distances are measured in units of the structure period , in particular @xmath24 . @xmath25 is the scattering phase for a single cylinder , @xmath26 is the linear dielectric susceptibility , and the amplitude of the incident wave is set to one , @xmath27 . with this choice of units , all three parameters are dimensionless . the scattering amplitudes of the fundamental and second harmonics are explicitly found in section [ sec:3 ] . in section [ sec:4 ] , it is shown that the ratio of the flux of the second harmonic along the normal direction and the incident flux is @xmath28 where @xmath29 is the fundamental field on the cylinders , and @xmath30 is some function . the function @xmath2 is _ non - analytic _ in the vicinity of zero values of its arguments . the non - perturbative approach of section [ sec:1 ] is used to prove that the generated flux of the second harmonic attains its maximal value when the small parameters satisfy the condition @xmath31 where @xmath32 is a numerical constant , and @xmath33 is the magnitude of the wave vector of the bound state that occurs at @xmath34 . under this condition , @xmath35 becomes analytic in the scattering phase @xmath36 so that in the leading order , @xmath37 an interesting feature to note is the independence of the conversion efficiency on the nonlinear susceptibility @xmath0 ( in the leading order in the scattering phase @xmath36 ) . in other words , given a nonlinear susceptibility @xmath0 , by a fine tuning of the distance between the arrays one can always reach the maximal value which is only determined by the scattering phase at the wave length of a bound state . the lowest value of @xmath33 for the system considered occurs just below the first diffraction threshold ( the wavelength is slightly larger than the structure period ) @xcite , i.e. , @xmath38 . taking , for example , @xmath39 and @xmath40 ( so that @xmath41 ) , the conversion rate reads @xmath42 , that is , about @xmath43 of the incident flux is converted into the second harmonic flux , which is comparable with the conversion rate achieved in slabs ( crystals ) of optical nonlinear materials @xcite . from the physical point of view , the scattering structure plays the role of a resonator with the quality factor inversely proportional to @xmath6 . the field in the resonator is not uniform and has periodic peaks of the amplitude @xmath4 due to a constructive interference of the scattered fundamental harmonic . the second harmonic is produced by the induced dipole radiation of point scatters located at these peaks . the induced dipole strength is proportional to @xmath44 . the dipoles are excited by the incident wave and , due to their periodic arrangement , they radiate in phase producing a plane wave in the asymptotic region ( just like a phased array antenna ) . if the system has a resonance whose width @xmath6 can be continuously driven to zero by changing a physical parameter of the system , i.e. , the system has a bound state in the radiation continuum , then the strength of the induced dipoles radiating the second harmonics can be magnified as desired , but the resonator can not be excited by the incident radiation if @xmath45 ( a bound state is decoupled from the radiation continuum ) . so , the optimal width @xmath6 at which the second harmonic amplitude is maximal occurs for some @xmath46 , which explains the existence of conditions like ( [ condition ] ) . since the second harmonic is generated by point scatterers , the phase matching condition , needed for optically nonlinear crystals , is not required . the energy flux of the incident radiation is automatically redistributed and focused on the scatterers owing to the constructive interference . thanks to these physical features , an active length at which the conversion rate is maximal is close to @xmath47 whose smallest value for the system studied is roughly a half of the wave length of the incident light @xcite ( i.e. for an infrared incident radiation it is about a few hundreds nanometers ) . suppose that a scattering system has a translational symmetry along a particular direction and has non - dispersive linear and second - order nonlinear dielectric susceptibilities , @xmath48 and @xmath49 , respectively . when the electric field is parallel to the translational symmetry axis ( tm polarization ) , maxwell s equations are reduced to the scalar nonlinear wave equation @xmath50 let the coordinate system be set so that the functions @xmath51 and @xmath52 have support bounded in the @xmath53direction and the system has the translational symmetry along the @xmath54direction . in this case , @xmath48 , @xmath49 and @xmath55 are functions of @xmath56 and @xmath57 . in the asymptotic regions @xmath58 , eq . ( [ eq:1 ] ) becomes a linear wave equation . so , the scattering problem can be considered for a plane wave of the frequency @xmath59 that propagates from the asymptotic region @xmath60 to the region @xmath61 . furthermore , it is assumed that the functions @xmath48 and @xmath49 are piecewise constant , i.e. , @xmath62 and @xmath63 in regions occupied by the scattering system . a conventional treatment of the problem is based on the assumption that the solution @xmath55 is analytic in @xmath64 and , therefore , can be represented as a power series expansion , @xmath65 where @xmath66 is the amplitude of the fundamental harmonics in the zero order of @xmath64 , @xmath67 is the amplitude of the second harmonics in the first order of @xmath64 , and so on . this assumption is not true if the system has bound states in the radiation continuum . indeed , a general solution has the form @xmath68 , where @xmath69 is the solution when @xmath70 and @xmath71 is the correction due to nonlinear effects . let @xmath49 be written as @xmath72 , where @xmath73 is the indicator function of the region occupied by the scattering system , i.e. , its value is 1 in that region and 0 elsewhere . then , if @xmath74 is the green s function of the operator @xmath75 with appropriate ( scattering ) boundary conditions , the function @xmath71 satisfies the integral equation @xmath76 \label{eq:2.1}\ ] ] the power series expansion ( [ eq:2 ] ) can be obtained by the method of successive approximations for this integral equation , provided the series is proved to converge . according to scattering theory @xcite , the fourier transform of @xmath74 is meromorphic in @xmath77 . as is clarified shortly ( see discussion of eq.([eq:12 ] ) ) , its real poles correspond to bound states in the radiation continuum . hence , in the presence of a real pole @xmath78 , the kernel of @xmath74 is not summable and , therefore , the successive approximations produce a diverging series . this implies a non - analytic behavior of the solution in @xmath0 . thus , when a bound state in the radiation continuum is present , the conventional perturbative approach becomes inapplicable . here , a non - perturbative approach is developed to obtain the solution to the scattering problem that is valid in any small neighborhood of a real pole of the fourier transform of @xmath74 . suppose that the incident radiation is a plane wave @xmath79 where @xmath80 , @xmath81 , denote unit vectors along the @xmath57 , @xmath82 , and @xmath56 coordinate axes , respectively . a general solution to eq . ( [ eq:1 ] ) should then be of the form , @xmath83 where @xmath84 , and for all @xmath85 , @xmath86 is the complex conjugate of @xmath87 ( as @xmath55 is real ) . therefore it is sufficient to determine only @xmath88 . next , it is assumed that the scattering structure is periodic in the @xmath89direction ( e.g. , a grating ) . the units of length are chosen so that the period is one . then the amplitudes @xmath87 satisfy bloch s periodicity condition @xmath90 this condition follows from the requirement that the solution @xmath55 satisfies the same periodicity condition as the incident wave @xmath91 : @xmath92 by eq.([eq:1 ] ) , the amplitudes of the different harmonics satisfy the equations , @xmath93 for ease of notation , the parameter @xmath94 is often used in lieu of @xmath64 . since @xmath95 , it is a small parameter in the system . the scattering theory requires that for @xmath96 , the partial waves @xmath97 be outgoing in the spatial infinity ( @xmath98 ) . the fundamental waves @xmath99 are a superposition of an incident plane wave @xmath100 and a scattered wave which is outgoing at the spatial infinity . in all , the above boundary conditions lead to a system of lippmann - schwinger integral equations for the amplitudes @xmath87 : @xmath101+e^{i\kk\cdot \rr } % } \\ % \displaystyle { e_{l}&=&\hv((lk)^2)[e_{l}+\nu\sum_{p}e_{p}e_{l - p}],\quad \,l\geq 2 % } % \end{cases } \end{array } \right . \label{eq:7}\ ] ] and @xmath102 for @xmath103 , where @xmath104 is the integral operator defined by the relation @xmath105(\rr)=\frac{q^2}{4\pi}\int ( \varepsilon(\ro)-1)g_{q}(\rr|\ro)\psi(\ro ) d\ro \label{eq:8}\ ] ] in which @xmath106 is the green s function of the poisson operator , @xmath107 , with the outgoing wave boundary conditions . for two spatial dimensions , as in the case considered here @xmath108 and @xmath109 , the green s function is known @xcite to be @xmath110 where @xmath111 is the zero order hankel function of the first kind . when @xmath112 , the amplitudes of all higher harmonics @xmath113 vanish . therefore it is natural to assume that @xmath114 for a small @xmath94 . note that this does not generally imply that the solution , as a function of @xmath94 , is analytic at @xmath115 . under this assumption , the solution to the system ( [ eq:7 ] ) can be approximated by keeping only the leading terms in each of the series involved . in particular , the first equation in ( [ eq:7 ] ) is reduced to @xmath116 + 2\nu \hv(k^2)\left[\,\overline{e}_{1}e_{2}\right ] \label{eq:9}\ ] ] while the second equation becomes @xmath117+\nu \hv((2k)^{2})\left[e_{1}^{2}\right ] \label{eq:10}\ ] ] it then follows that a first order approximation to the solution of the nonlinear wave equation ( [ eq:1 ] ) may be found by solving the system formed by the equations ( [ eq:9 ] ) and ( [ eq:10 ] ) . to facilitate the subsequent analysis , the system is rewritten as @xmath118[e_{1}]=e^{i\kk\cdot\rr}+2\nu \hv(k^2)\left[\,\overline{e}_{1}e_{2}\right]\\ [ 1-\hv((2k)^2)][e_{2}]=\nu \hv((2k)^2)\left[e_{1}^{2}\right ] \end{cases } \label{eq:11}\ ] ] solving the first of eqs.([eq:11 ] ) involves inverting the operator @xmath119 , and therefore necessitates a study of the poles of the resolvent @xmath120^{-1}$ ] . such poles are eigenvalues in the generalized eigenvalue problem , @xmath121=e \label{eq:12}\ ] ] for fixed @xmath122 . the corresponding eigenfunctions @xmath123 are referred to as _ siegert states_. in contrast to siegert states in quantum scattering theory @xcite , electromagnetic siegert states satisfy the generalized eigenvalue problem ( [ eq:12 ] ) in which the operator is a nonlinear function of the spectral parameter @xmath124 . their general properties are studied in @xcite . eigenvalues have the form @xmath125 . if @xmath126 , then , according to scattering theory , such a pole is a resonance pole . in the case of the linear wave equation ( @xmath127 ) , the scattered flux peaks at @xmath128 indicating the resonance position , whereas the imaginary part of the pole @xmath6 defines the corresponding resonance width ( or a spectral width of the scattered flux peak ; a small @xmath6 corresponds to a narrow peak ) . if @xmath129 , the corresponding siegert state is a _ bound state_. this is a localized ( square integrable ) solution of eq.([eq:12 ] ) . suppose that the scattering system has a physical parameter @xmath10 such that a pole @xmath130 of the resolvent @xmath131^{-1}$ ] depends continuously on @xmath10 and that there is a particular value @xmath132 at which the pole becomes real , i.e. , @xmath133 . if @xmath134 , then the corresponding bound state lies in the radiation continuum . note that eq.([eq:12 ] ) may have solutions for real @xmath135 . these are bound states below the radiation continuum . such states are not relevant for the present study and , henceforth , bound states are understood as bound states in the radiation continuum . as noted in the introduction , two periodic planar scattering structures separated by a distance @xmath9 have bound states in the radiation continuum . suppose that for fixed @xmath136 , the set @xmath137 consists of isolated points ( which is generally true ) and the points @xmath138 do not belong to it ( which is fulfilled in a concrete example studied in the next section @xcite ) . consequently , the operator @xmath139 is regular in a neighborhood of @xmath140 and , for @xmath22 close to but not equal to @xmath141 , the operator @xmath119 is invertible . it then follows that @xmath66 satisfies the nonlinear integral equation , @xmath142\right]\right]\right ] \label{eq:13}\ ] ] where , in accord with the notation introduced in ( [ eq:8 ] ) , the function on which an operator acts is placed in the square brackets following the operator . the operator @xmath143 that determines the nonlinear " part of eq . ( [ eq:13 ] ) is not bounded as @xmath144 , no matter how small @xmath145 is . this precludes the use of a power series representation of the solution in @xmath146 . to find the solution of eq.([eq:13 ] ) when @xmath94 is small , its property under parity transformations is established first . suppose that the scattering system is such that the operator @xmath147 and the parity operator @xmath148 defined by @xmath149(x , z)=e(x ,- z)$ ] commute . this implies that the siegert states have a specific parity : @xmath150=pe_s$ ] where @xmath151 . consider then the ratio @xmath152}{e_1 } \label{eq:14}\ ] ] it will be proved that @xmath153 in the limit @xmath154 along a certain curve . indeed , it follows from the meromorphic expansion of @xmath120^{-1}$ ] that near a pole @xmath155 , @xmath156 where @xmath157 is some constant depending on @xmath10 , and @xmath158 is an appropriately normalized siegert state @xcite . consider then the curve of resonances @xmath159 in the @xmath160-plane . along @xmath161 , @xmath162 now , as @xmath163 , the width @xmath164 goes to @xmath165 , and the siegert state @xmath158 becomes a bound state @xmath166 in the radiation continuum . equation ( [ eq:16 ] ) shows that if @xmath157 does not go to zero faster than @xmath164 as @xmath167 , i.e. , the pole still gives the leading contribution to @xmath168 in this limit , then @xmath169}{e_b } = p= \pm 1 \label{eq:17}\ ] ] depending on whether the bound state @xmath166 is even or odd in @xmath56 . for the linear wave equation ( @xmath115 ) , the constant @xmath157 is shown to be proportional to @xmath170 @xcite . therefore , for a small @xmath94 , the assumption that @xmath157 does not go to zero faster than @xmath164 as @xmath167 is justified . based on the limit ( [ eq:17 ] ) , the following ( non - perturbative ) approach is adopted to solve eq.([eq:13 ] ) near a bound state . first , the curve of resonances @xmath161 in the @xmath160-plane is found . using the relation ( [ eq:14 ] ) in the right side of ( [ eq:13 ] ) , the field @xmath168 is expressed via the ratio @xmath171 with the pair @xmath160 being on the curve @xmath161 . next , the principal part of the amplitude @xmath66 relative to @xmath172 is evaluated near a critical point @xmath173 on @xmath161 by taking @xmath171 to its limit value ( [ eq:17 ] ) . this approach reveals a non - analytic dependence of the amplitude @xmath66 on the small parameters @xmath174 and @xmath64 of the system and allows to obtain @xmath168 and @xmath175 when a bound state is present in the radiation continuum . as the technicalities of the proposed non - perturbative approach depend heavily on peculiarities of the scattering system , the procedure is illustrated with a specific example . the system considered is sketched in fig . [ fig:1](a ) . it consists of an infinite double array of parallel , periodically positioned cylinders . the cylinders are made of a nonlinear dielectric material with a linear dielectric constant @xmath176 , and a second order susceptibility @xmath177 . the coordinate system is set so that the cylinders are parallel to the y - axis , the structure is periodic along the x - axis , and the z - axis is normal to the structure . the unit of length is taken to be the array period , and the distance between the two arrays relative to the period is @xmath9 . -axis , and is at a distance @xmath10 from the @xmath57-axis.panel ( b ) : the scattering process for the normal incident radiation ( @xmath178 ) . the scattered fundamental harmonic is symbolized by a single headed arrow while the ( generated ) second harmonic radiation is symbolized by a double headed arrow . the incident radiation wave length is such that only one diffraction channel is open for the fundamental harmonic while three diffraction channels are open for the second harmonic . the flux measured through the faces @xmath179 cancels out due to the bloch periodicity condition as explained in appendix [ asec:2 ] . panel ( c ) : the solid and dashed curves show the position ( frequency @xmath180 ) of scattering resonances as functions of the distance between the arrays , @xmath181 . the dots on the curves indicate positions of bound states in the radiation continuum ( i.e. , the values of @xmath10 at which a resonance turns into a bound state ) . the solid curve connects bound states symmetric relative to the reflection @xmath182 . the dashed line connects the skew symmetric bound states . the curves are realized for @xmath183 , and @xmath178 ( normal incidence ) . ] the solution of the integral equation ( [ eq:13 ] ) is obtained for @xmath124 near @xmath137 in the limit of subwavelength dielectric cylinders . the approximation is defined by a small parameter @xmath184 which is the scattering phase of a plane wave with the wavenumber @xmath185 on a single cylinder of radius @xmath21 . for sufficiently small @xmath21 , this approximation is justified . the integral kernel of @xmath186 is defined by ( [ eq:8 ] ) and has support on the region occupied by cylinders . the condition ( [ eq:18 ] ) implies that the wavelength is much larger than the radius @xmath21 , and therefore field variations within each cylinder may be neglected , so that @xmath187 where @xmath188 are the positions of the axes of the cylinders ( @xmath189 is an integer ) . the integration in @xmath190 $ ] yields then an infinite sum over positions of the cylinders . by bloch s condition , @xmath191 , so that the function @xmath192(x , z)$ ] is fully determined by the two values @xmath193 . in particular , @xmath105(0,\pm h ) \approx \alpha\psi(0,\pm h)+\beta\psi(0,\mp h ) \label{eq:19}\ ] ] where the coefficients @xmath194 and @xmath195 are shown to be @xcite [ eq:20 ] @xmath196 @xmath197 where @xmath198 with the convention that if @xmath199 , then @xmath200 . to obtain the energy flux scattered by the structure , the action of the operator @xmath186 on @xmath201 must be determined in the asymptotic region @xmath98 . it is found that for @xmath202 , @xmath105(x , z)\approx 2\pi i\dl(q ) \sum_{m=-\infty}^{\infty}\left(\psi(0,h)e^{i|z - h|q_{z , m}}+\psi(0,-h ) e^{i|z+h|q_{z , m}}\right ) \frac{e^{i x(q_{x}+2\pi m)}}{q_{z , m } } \label{eq:21}\ ] ] now that the action of the operator @xmath186 has been established in ( [ eq:19 ] ) and ( [ eq:21 ] ) , the amplitudes @xmath66 and @xmath67 of the fundamental and second harmonics can be determined by solving the system ( [ eq:11 ] ) . as noted earlier , this will be done along a curve @xmath161 in the @xmath203-plane defined by @xmath181 where @xmath204 is the real part of a pole of @xmath120^{-1}$ ] , or equivalently , when the incident radiation has the resonant wave number @xmath181 . to find the curve , the eigenvalue problem ( [ eq:12 ] ) is solved in in the approximation ( [ eq:19 ] ) : @xmath205\begin{pmatrix } e_{b+}\\ e_{b- } \end{pmatrix}=\begin{pmatrix } 0\\ 0 \end{pmatrix},\quad \hc=\begin{pmatrix } \al&\be\\ \be&\al \end{pmatrix } \label{eq:22}\ ] ] where @xmath206 and the functions @xmath207 and @xmath208 have been defined in the previous section . in particular , bound states occur at the points @xmath173 at which the determinant @xmath209 vanishes , @xmath210 it follows from eq.([eq:22 ] ) that the bound states for which @xmath211 are even in @xmath56 because @xmath212 in this case . similarly , the bound states for which @xmath213 are odd in @xmath56 . more generally , the poles of the resolvent @xmath120^{-1}$ ] are complex zeros of @xmath209 . they are found by the conventional scattering theory formalism . specifically , the resonant wave numbers @xmath214 are obtained by solving the equation @xmath215 for the spectral parameter @xmath124 . according to the convention adopted in the representation ( [ eq:15 ] ) , the corresponding resonance widths are defined by @xmath216 where @xmath217 denotes the derivative with respect to @xmath124 . this definition of the width corresponds to the linearization of @xmath218 near @xmath219 as a function of @xmath124 in the pole factor @xmath220^{-1}$ ] . the curves of resonances @xmath221 come in pairs . there is a curve connecting the symmetric bound states in the @xmath203-plane , and another curve that connects the odd ones . in what follows , only the curve connecting symmetric bound states will be considered . the other curve can be treated similarly . panel ( c ) of fig . [ fig:1 ] shows that the first symmetric bound state occurs when the distance @xmath9 is about half the array period , while the skew - symmetric bound states emerge only at larger distances . this feature is explained in detail in @xcite . so , the solution obtained near the first symmetric bound state corresponds to the smallest possible transverse dimension of the system ( roughly a half of the wave length of the incident radiation ) . thus , from now on the curve of resonances @xmath161 refers to the curve in the @xmath203-plane defined by the equation @xmath222 . to simplify the technicalities , it will be further assumed that only one diffraction channel is open for the fundamental harmonics , i.e. , @xmath223 . let @xmath181 be the solution of @xmath222 . by making use of the explicit form of the functions @xmath194 and @xmath195 for one open diffraction channel , one infers that along the curve @xmath181 , @xmath224 bound states in the radiation continuum occur when the distance @xmath10 satisfies the equation @xmath225 , i.e. , @xmath226 with @xmath189 being a positive integer . its solutions @xmath227 define the corresponding values of the wave numbers of the bound states , @xmath228 . so , the sequence of pairs @xmath229 indicates positions of the bound states on the curve @xmath161 . in the limit @xmath230 along @xmath161 , the function @xmath231 has the asymptotic behavior , @xmath232 the objective is to determine the dependence of the amplitudes @xmath66 and @xmath67 on the parameters @xmath174 and @xmath64 which are both small . to this end , let @xmath233 be the values of the field @xmath66 on the axes of the cylinders at @xmath234 , and @xmath235 be the values of the field @xmath67 on the same cylinders . in the subwavelength approximation , these values determine the scattered field because the latter is produced by the radiation of point dipoles induced by the incident wave on the scatterers and the strength of the dipoles is proportional to @xmath236 for the fundamental harmonic and @xmath237 for the second harmonic . applying the rule ( [ eq:19 ] ) to evaluate the action of the operator @xmath186 in the system ( [ eq:11 ] ) , the first equation of the latter becomes , [ eq:27 ] @xmath238 \begin{pmatrix } e_{1+}\\ e_{1- } \end{pmatrix } = 2\nu\hc \begin{pmatrix } \overline{e}_{1+}e_{2+}\\ \overline{e}_{1-}e_{2- } \end{pmatrix}+\begin{pmatrix } e^{+i h\kz}\\ e^{-i h\kz } \end{pmatrix}\ ] ] similarly , the second of eqs.([eq:11 ] ) yields , @xmath239 \begin{pmatrix } e_{2+}\\ e_{2- } \end{pmatrix } = \nu \hcd \begin{pmatrix } e_{1+}^{2}\\ e_{1-}^{2 } \end{pmatrix},\quad \hcd=\hc(2k,2\kx)\ ] ] as stated above , the resolvent @xmath240^{-1}$ ] is regular in a neighborhood of @xmath141 so that eq . ( [ eq:27.2 ] ) can be solved for @xmath237 , which defines the latter as functions of @xmath236 . the substitution of this solution into eq.([eq:27.1 ] ) gives a system of two nonlinear equations for the fields @xmath236 . adding these equations and replacing the field @xmath241 by its expression @xmath242 in terms of the field ratio of eq.([eq:14 ] ) yields the following implicit relation between the field @xmath243 and its amplitude @xmath244 : @xmath245 where @xmath246 and @xmath247 are small and , in terms of the field ratio @xmath248 , the values of @xmath249 and @xmath250 read , @xmath251 with @xmath252 and @xmath253 being defined by the relation , @xmath254^{-1}\hcd=\begin{pmatrix } a&b\\ b&a \end{pmatrix } \label{eq:29.1}\ ] ] in particular , @xmath249 and @xmath250 are continuous functions of @xmath171 and @xmath231 . in appendix [ asec:1 ] it is shown that if @xmath255 and @xmath256 are the respective limits of @xmath249 and @xmath250 as @xmath167 along the curve of resonances @xmath161 , then these limits are nonzero . it follows then that eq.([eq:28 ] ) for @xmath243 is singular in both @xmath94 and @xmath231 when these parameters are small , i.e. , in the limit @xmath257 . furthermore , there is no way to solve the said equation perturbatively in either of the parameters . a full non - perturbative solution can be obtained using cardano s method for solving cubic polynomials . indeed , by taking the modulus squared of both sides of the equation , it is found that , @xmath258 the solution to this cubic equation is obtained in appendix [ asec:3 ] . it is proved there that eq . ( [ eq:30 ] ) admits a unique real solution so that there is no ambiguity on the choice of @xmath243 . in the vicinity of a point @xmath173 along the resonance curve @xmath161 , the field @xmath243 is found to behave as , @xmath259 recall that @xmath260 . an explicit form of the function @xmath261 is given in appendix [ asec:3 ] ( see eq . ( [ eq : b8 ] ) ) . it involves combinations of the square and cube roots of functions in @xmath174 and @xmath64 and has the property that @xmath262 as @xmath263 ( in the sense of the two - dimensional limit ) . in the limit @xmath12 , the field ratio @xmath171 approaches @xmath264 for a symmetric bound state as argued earlier . therefore it follows from eqs.([eq:27.2 ] ) that @xmath265 because the matrix ( [ eq:29.1 ] ) exists at @xmath34 . since @xmath266 , relation ( [ eq:31 ] ) leads to the conclusion that @xmath267 thus , the approximation @xmath268 used to truncate the system ( [ eq:11 ] ) remains valid for @xmath10 close to the critical value @xmath269 despite the non - analyticity of the amplitudes at @xmath270 . for the nonlinear system considered , even though poynting s theorem takes a slightly different form as compared to linear maxwell s equations , the flux conservation for the time averaged poynting vector holds . the scattered energy flux carried across a closed surface by each of the different harmonics adds up to the incident flux across that surface . the flux conservation theorem is stated in appendix [ asec:2 ] . consider a closed surface that consists of four faces , @xmath271 and @xmath272 as depicted in fig . [ fig:1](b ) . as argued in appendix [ asec:2 ] , the scattered flux of each @xmath273-harmonics across the union of the faces @xmath274 vanishes because of the bloch condition ( and so does the incident flux for any @xmath136 ) . therefore only the flux conservation across the union of the faces @xmath275 has to be analyzed . if @xmath276 designates the ratio of the scattered flux carried by the @xmath273-harmonics across the faces @xmath275 to the incident flux across the same faces , then @xmath277 . thus , for @xmath278 , @xmath276 defines the _ conversion ratio _ of fundamental harmonics into the @xmath273-harmonics . in the perturbation theory used here , only the ratios @xmath279 and @xmath280 may be evaluated . by laborious calculations it can be shown that @xmath281 as one would expect ( see appendix [ asec:2 ] for details ) . hence , the efficiency of converting the fundamental harmonic into the second harmonic is simply determined by the maximum value of @xmath280 as a function of the parameter @xmath10 at a given value of the nonlinear susceptibility @xmath0 . the ratio @xmath35 is defined in terms of the scattering amplitudes of the second harmonic , i.e. , by the amplitude of @xmath175 in the asymptotic region @xmath98 : @xmath282 where @xmath283 is the wave vector of the second harmonic in the @xmath284 open diffraction channel . recall that the @xmath284 channel is open provided @xmath285 and in this case @xmath286 , while if the channel is closed , then @xmath287 . in the asymptotic region @xmath98 , the field in closed channels decays exponentially and , hence , the energy flux can only be carried in open channels to the spatial infinity . the summation in eqs.([eq:33 ] ) is taken only over those values of @xmath288 for which the corresponding diffraction channel is open for the second harmonic , which is indicated by the superscript `` @xmath289 '' in the summation index @xmath290 . note that there is more than one open diffraction channel for the second harmonic even though only one diffraction channel is open for the fundamental one . for instance , if the @xmath89component of the wave vector @xmath291 , i.e. , @xmath122 , is less than @xmath292 , there are 3 open diffraction channels for the second harmonic , the channels @xmath293 , and @xmath294 . these three directions of the wave vector of the second harmonic propagating in each of the asymptotic regions @xmath295 are depicted in fig . [ fig:1](b ) by double - arrow rays . thus , in terms of the scattering amplitudes introduced in eqs . ( [ eq:33 ] ) , the ratio of the second harmonic flux across @xmath275 to the incident flux is @xmath296 the scattering amplitudes @xmath297 and @xmath298 are inferred from eq.([eq:10 ] ) in which the rule ( [ eq:21 ] ) is applied to calculate the action of the operator @xmath299 in the far - field regions @xmath98 : @xmath300}\vspace{0.1 cm}\\ \displaystyle{t_{m}^{sh}=\frac{2\pi i\dl(2k)}{k_{z , m}^{sh}}\left[(e_{2+}+\nu e_{1+}^{2 } ) e^{-i h k_{z , m}^{sh}}+(e_{2-}+\nu e_{1-}^{2 } ) e^{i h k_{z , m}^{sh}}\right ] } \end{cases } \label{eq:35}\ ] ] since for @xmath301 , the amplitudes @xmath236 remain finite as @xmath167 along @xmath161 , and since @xmath302 in the said limit ; the principal part of @xmath280 in a vicinity of a bound state along the curve @xmath161 is obtained by setting @xmath303 in eq . ( [ eq:27.2 ] ) , solving the latter for @xmath237 , and substituting the solution into the expression for @xmath35 . the result reads @xmath304 where @xmath305 is a constant obtained by taking all nonsingular factors in the expression of @xmath280 to their limit as @xmath167 , which gives @xmath306_{(h , k)=(\hb,\kb ) } \label{eq:37}\ ] ] for @xmath252 and @xmath253 defined in eq.([eq:29.1 ] ) . using the identity @xmath307 , and substituting eq.([eq:28 ] ) into one of the factors @xmath308 , the conversion ratio @xmath280 is expressed as a function of a single real variable , @xmath309 where @xmath310 is a constant , and @xmath250 and @xmath249 in eq.([eq:29 ] ) have been taken at their limits as @xmath167 to obtain the principal part of @xmath280 . the function @xmath311 on @xmath312 is found to attain its absolute maximum at @xmath313 . this condition determines the distance @xmath9 between the arrays at which the conversion rate is maximal for given parameters @xmath314 and @xmath64 of the system . indeed , since @xmath315 should also satisfy the cubic equation ( [ eq:30 ] ) , the substitution of @xmath316 into the latter yields the condition [ eq:39 ] @xmath317 in particular , in the leading order in @xmath318 , the optimal distance @xmath9 between the two arrays is given by the formula , @xmath319 where as previously , @xmath320 . the maximum value @xmath321 of the conversion ratio @xmath280 is the sought - for _ conversion efficiency_. an interesting feature to note is that @xmath322 is independent of the nonlinear susceptibility @xmath20 because the constants @xmath323 , @xmath324 , and @xmath325 are fully determined by the position of the bound state @xmath326 . in other words , if the distance between the arrays is chosen to satisfy the condition ( [ eq:39.1 ] ) , the conversion efficiency @xmath321 is the same for a wide range of values of the nonlinear susceptibility @xmath20 . this conclusion follows from two assumptions made in the analysis . first , the subwavelength approximation should be valid for both the fundamental and second harmonics , i.e. , the radius of cylinders should be small enough . second , the values of @xmath327 and @xmath94 ( or @xmath0 ) must be such that the analysis of the existence and uniqueness of @xmath244 given in appendix [ asec:3 ] holds , that is , eq . ( [ eq:30 ] ) should have a unique real solution under the condition ( [ eq:39.1 ] ) . the geometrical and physical parameters of the studied system can always be chosen to justify these two assumptions as illustrated in fig . [ fig:2 ] . for the critical points @xmath328 , and @xmath329 and @xmath178 ( the normal incidence ) . the dashed parts of the curves indicate the regions where @xmath330 and , hence , @xmath331 , i.e. , the subwavelength approximation becomes inapplicable for the second harmonic . panel ( b ) : the conversion efficiency is plotted against @xmath122 for the critical points @xmath328 . the curves are realized for @xmath39 and @xmath329.panel ( c ) : the region of validity of the developed theory for the first bound state @xmath332 . the shadowed part of the @xmath333plane is defined by the condition @xmath334 under which , according to eq.([eq : b8 ] ) , the amplitude @xmath244 exists and unique as explained in appendix [ asec:3 ] . the plot is realized for @xmath335 and @xmath178 . the parabola - like curve is an actual boundary of the shadowed region ; the top horizontal line represents no restriction . so there is a wide range of the physical and geometrical parameters within the shadowed region which satisfy ( [ eq:39.1 ] ) . the regions of validity for other bound states looks similar . ] panels ( a ) and ( b ) of fig . [ fig:2 ] show the conversion efficiency @xmath321 for the first three symmetric bound states @xmath336 , as , respectively , a function of the cylinder radius @xmath21 when @xmath337 and of @xmath136 when @xmath39 . for all curves presented in the panels , @xmath338 . the values of @xmath321 are evaluated numerically by eq . ( [ eq:38 ] ) where @xmath339 . the solid parts of the curves in panel ( a ) correspond to the scattering phase @xmath340 with @xmath341 . note that the wavelength at which the second harmonic generation is most efficient is the resonant wavelength defined by @xmath342 where @xmath10 satisfies the condition ( [ eq:39.1 ] ) . for a small @xmath0 , the scattering phase at the resonant wavelength can well be approximated as @xmath343 . the condition @xmath344 ensures that the scattering phase for the second harmonic satisfies the inequality @xmath345 otherwise the validity of the subwavelength approximation can not be justified . the dashed parts of the curves in panels ( a ) of fig . [ fig:2 ] correspond to the region where @xmath346 . panels ( a ) and ( b ) of the figure show that the conversion efficiency can be as high as 40% for a wide range of the incident angles and values of the cylinder radius . such a conversion efficiency is comparable with that achieved in optically nonlinear crystals at a typical beam propagation length ( active length ) of a few centimeters , whereas here the transverse dimension @xmath9 of the system studied here can be as low as a half of the wavelength , i.e. , for an infrared incident radiation , @xmath9 is about a few hundred nanometers . indeed , as one can see in fig . [ fig:1](c ) , the first bound state occurs at @xmath347 and @xmath348 which corresponds to the wavelength @xmath349 . the stated conversion efficiency can be fairly well estimated in the leading order of @xmath318 : @xmath350 suppose that only one diffraction channel is open for the incident radiation . then @xmath351 in eq.([eq:41 ] ) ( three open channels for the second harmonics ) . let @xmath352 denote the term @xmath353 , i.e. @xmath352 is the second harmonic flux in which the contribution of the channels with @xmath354 is omitted . in particular , @xmath355 . one infers from eq . ( [ eq:41 ] ) that , @xmath356 as the pair @xmath173 at which a symmetric bound state is formed satisfies the equation @xmath357 , it follows that @xmath358 . hence , @xmath359 the wavenumbers @xmath141 at which the bound states occur lie just below the diffraction threshold @xmath360 , i.e. , @xmath361 ( see fig . [ fig:1](c ) and @xcite for details ) . so that in the case of normal incidence ( @xmath178 ) , the above estimate becomes , @xmath362 with @xmath363 . if , for instance , @xmath39 and @xmath364 , then , @xmath365 for @xmath366 . * appendices * here the limit values @xmath255 and @xmath256 of the functions @xmath249 and @xmath250 defined in eq . ( [ eq:29 ] ) are estimated as @xmath12 along the resonance curve @xmath161 . for @xmath256 this is immediate . indeed , in the aforementioned limit , the field ratio @xmath302 for a symmetric bound state and , therefore , @xmath367 for @xmath255 , the estimate follows from that wave numbers @xmath141 at which bound states exist are close to the diffraction threshold @xmath360 when only one diffraction channel is open for the fundamental harmonic , i.e. , @xmath223 @xcite . indeed , in the first order of @xmath318 , @xmath368 this proximity of the wavenumbers @xmath141 to the diffraction threshold @xmath360 allows one to determine the leading terms in the coefficients @xmath252 and @xmath253 defined in eq.([eq:29.1 ] ) , and hence @xmath255 . to proceed , the coefficients @xmath369 and @xmath370 of the matrix @xmath371 are rewritten by separating explicitly the real and imaginary parts : [ eq : a3 ] @xmath372 where @xmath373 and @xmath374 are defined by the relations , @xmath375 and the index @xmath290 indicates that the summations are to be taken over all open diffraction channels for the second harmonic . using the estimate ( [ eq : a2 ] ) , the functions @xmath376 are found to obey the estimates , @xmath377 these expressions are then used to estimate @xmath378 . in the first order of @xmath379 one infers that @xmath380 for the nonlinear wave equation ( [ eq:1 ] ) , the poynting theorem becomes , @xmath381=-\int_{\partial v } \mathbf{s}\cdot d\mathbf{n } \label{eq : c1}\ ] ] where @xmath382 is a closed region , and @xmath383 is its boundary . the vector @xmath384 is the poynting vector ( for simplicity , it is assumed that @xmath383 lies in the vacuum so that @xmath385 , and @xmath386 in a small neighborhood of @xmath383 ) . in the case of a monochromatic incident wave , the flux measured is the time - average of @xmath387 over a time interval @xmath388 . by averaging eq.([eq : c1 ] ) , it then follows that , @xmath389 this is the flux conservation . in terms of the different harmonics of eq.([eq:4 ] ) , the time averaged poynting vector becomes , @xmath390 of interest is the flux of the poynting vector across the rectangle depicted in fig . [ fig:1](b ) . by bloch s condition ( [ eq:5.1 ] ) , the contributions to the flux from the faces @xmath391 cancel out so that the flux measured is through the vertical faces @xmath392 . note that the vanishing of the flux across the faces @xmath274 is a consequence of the fact that the incident wave is uniformly extended over the whole @xmath89axis . for example , consider the normal incidence ( @xmath337 ) with one diffraction channel open for the incident radiation . then the poynting vector of the reflected and transmitted fundamental harmonic is normal to the structure and , hence , carries no flux across @xmath274 . the second harmonic @xmath393 has three forward and three backward scattering channels open , @xmath351 , relative to the @xmath53axis . the wave with @xmath353 propagates in the direction normal to the structure and does not contribute to the flux across @xmath274 . since the incident wave has an infinite front along the @xmath89axis , so do the scattered waves with @xmath354 . the waves with @xmath294 and @xmath394 carry opposite fluxes across each of the faces @xmath274 as the corresponding wave vectors have the same @xmath53components and opposite @xmath89components and , hence , the total flux vanishes . for a finite wave front ( but much larger than the structure period ) , the second harmonic would carry the energy flux in all the directions parallel to the corresponding wave vectors in each open diffraction channel . if @xmath276 is as defined in section [ sec:4 ] , then the flux conservation implies that @xmath395 . therefore , in the perturbation theory used , i.e. , when the system ( [ eq:7 ] ) is truncated to eqs.([eq:11 ] ) , the inequality @xmath281 must be verified to justify the validity of the theory . the conversion ratio @xmath280 is given in section [ sec:4 ] . if only one diffraction channel is open for the fundamental harmonic , then the ratio @xmath279 of the scattered and incident fluxes of the fundamental harmonic reads , @xmath396 where @xmath397 and @xmath398 are the transmission and reflection coefficients which are obtained from the far - field amplitude of @xmath66 as , @xmath399 where @xmath400 is the incident wave vector and @xmath401 is the wave vector of the reflected fundamental harmonic . it then follows from eqs.([eq:9 ] ) and ( [ eq:21 ] ) that @xmath402}\\ \displaystyle{t_{0}=i\frac{2\pi \dl(k)}{\kz } \bigl [ ( e_{1+}+2\nu e_{2+}\overline{e}_{1+})e^{-i h\kz}+(e_{1-}+2\nu e_{2-}\overline{e}_{1- } ) e^{i h\kz}\bigr ] } \end{cases } \label{eq : c6}\ ] ] in the vicinity of a critical point @xmath173 , the coefficients @xmath398 and @xmath397 obey the estimate , @xmath403 after some algebraic manipulations , it is found that , @xmath404 \label{eq : c8}\ ] ] where @xmath405 is the constant defined as , @xmath406_{(h , k)=(\hb,\kb ) } \label{eq : c9}\ ] ] and @xmath407 is introduced in eqs.([eq : a3 ] ) . expressing @xmath252 and @xmath253 defined by ( [ eq:29.1 ] ) via the coefficients @xmath408 and @xmath409 of the symmetric matrix @xmath371 , one also obtains @xmath410 since at the point @xmath173 the value of @xmath249 is @xmath411 , it follows that , @xmath412\bigg|_{(h , k)=(\hb,\kb ) } \label{eq : c11}\ ] ] for general complex numbers @xmath252 and @xmath253 , the expression in square brackets is always zero . therefore @xmath413 , and , @xmath414 by eq . ( [ eq : a6 ] ) , @xmath415 . in appendix [ asec:3 ] it is proved that @xmath416 . consequently , near the critical point @xmath173 , the right hand summand in eq.([eq : c12 ] ) is negative so that @xmath281 as required . the amplitude of the field @xmath243 is a root of the cubic polynomial in eq.([eq:30 ] ) which can be solved by cardano s method . put @xmath417 . for the new variable @xmath418 , eq . ( [ eq:30 ] ) assumes the standard form , @xmath419 where , @xmath420 as the amplitude @xmath66 is uniquely defined by the system ( [ eq:7 ] ) , it is therefore expected that the cubic in eq.([eq : b1 ] ) should have a unique real solution in order for the theory to be consistent . the latter holds if and only if the discriminant @xmath421 is nonnegative . to prove that @xmath422 , note first that @xmath423 in the vicinity of a critical point @xmath173 this follows from the estimates established in appendix [ asec:1 ] . indeed , in the first order of @xmath379 , @xmath424 next , consider the complex number @xmath425 \label{eq : b5}\ ] ] the positivity condition ( [ eq : b4 ] ) ensures that the coefficient of the complex number @xmath426 in the expression of @xmath427 is indeed real . after some algebraic manipulations , it can be shown that , @xmath428 thus @xmath429 as required . the only real solution @xmath418 to eq.([eq : b1 ] ) is then , @xmath430{\frac{-q+\sqrt{d_{3}}}{2}}+\sqrt[3]{\frac{-q-\sqrt{d_{3}}}{2 } } \label{eq : b7}\ ] ] it then follows that , @xmath431{\frac{1}{2}\left(\nu^2|\zeta|^2-\frac{1}{2}\vp^4\re\{\rho\}\pm \big||\zeta|^2\nu^2-\vp^4\rho\big|\right)}-\frac{\vp^{\frac{4}{3}}}{3}\re\{\zeta\xi\ } \label{eq : b8}\ ] ] provided @xmath432 . the latter condition imposes a limit on the validity of the perturbation theory developed in the present study , i.e. , the reduction of the system ( [ eq:7 ] ) to ( [ eq:11 ] ) is justified if @xmath432 this is to be expected because of the lack of analyticity in @xmath64 of the solution to the nonlinear wave equation ( [ eq:1 ] ) that can only occur at the critical points @xmath173 at which bound states in the radiation continuum exist . as one gets away from these critical points in the @xmath160-plane , the solution to the nonlinear wave equation becomes analytic in @xmath64 , meaning that all the terms that were neglected in finding the principal parts of the amplitudes must now also be taken into account to find a solution befitting the series of eq.([eq:2 ] ) . the shadowed region depicted in fig . [ fig:2](b ) shows the region of the @xmath160-plane in which the condition @xmath432 holds for the first symmetric bound . the presented analysis of the efficiency of the second harmonic generation is valid for any choice of the geometrical parameters , @xmath433 and @xmath21 , and the physical parameters , @xmath26 and @xmath434 , which satisfy the conditions ( [ eq:39.1 ] ) and @xmath432 . garca de abajo , rev . 79 , 1267 ( 2007 ) . s. v. shabanov , int . j. mod b 23 , 5191 ( 2009 ) . d. c. marinica , a. g. borisov and s. v. shabanov , phys . b 76 , 085311 ( 2007 ) . a. nahata , r. a. linke , t. ishi , k. ohashi , opt . 28 , 423 ( 2003 ) . d. c. marinica , a. g. borisov and s. v. shabanov , phys . 100 , 183902 ( 2008 ) . f. r. ndangali and s. v. shabanov , j. math . 51 , 102901 ( 2010 ) . j. von neumann and e. wigner , phys . z 30 , 465 ( 1929 ) . f. capasso , c. sirtori , j. faist , d. l. sivco , s .- n . g. chu and a. y. cho , nature ( london ) 358 , 565 ( 1992 ) . h. friedrich and d. wintgen , phys . a 32 , 3231 ( 1985 ) . j. okolowicz , m. ploszajczak and i. rotter , phys . 374 , 271 ( 2003 ) . a. z. devdariani , v. n. ostrovsky and yu . n. sebyakin , sov . jetp 44 , 477 ( 1976 ) . f. r. ndangali and s. v. shabanov , electromagnetic siegert states for periodic dielectric structures , ( submitted to jmp ) . klauder , beyond conventional quantization , ( cambridge , cambridge university press , 2000 ) . f. trger(ed . ) , springer handbook of lasers and optics , ( new york , springer+business media , 2007 ) ; p. 310 . r. g. newton , scattering theory of waves and particles , 2nd ed . ( springer - verlag , 1982 ) . m. reed and b. simon , methods of modern mathematical physics iii : scattering theory ( academic press , 1979 ) . m. morse and h. feshbach , methods of theoretical physics , ( new york , mcgraw - hill , 1953 ) .

we establish the inverse spectral transform for the conservative camassa holm flow with decaying initial data . in particular , it is employed to prove existence of weak solutions for the corresponding cauchy problem .

1510.04916

the _ inverse scattering transform _ @xcite is a powerful tool for solving initial value problems for certain nonlinear partial differential equations . originally developed for the korteweg de vries equation by gardner , greene , kruskal and miura @xcite , this method has since been successfully extended to various other completely integrable equations . it is the aim of the present article to establish the corresponding transform for the camassa holm equation @xmath0 with decaying initial data . due to the vast amount of literature on this equation , we only refer to a selection of articles @xcite containing further information . the relevance of the camassa holm equation stems from the fact that it constitutes a model for unidirectional wave propagation on shallow water @xcite . unlike the korteweg de vries equation , it allows for smooth solutions to blow up in finite time in a way that resembles wave - breaking this process has been described in detail @xcite and is known to only happen when the quantity @xmath1 changes sign . compared to the rather tame sign - definite case ( which shares a lot of similarities with the korteweg de vries equation @xcite ) , indefiniteness of @xmath2 causes serious complications ( for example noticeable from the discussions in @xcite ) . many of these problems are already apparent for the prototypical example of a peakon - antipeakon interaction @xcite , which is a special case of so - called multi - peakon solutions @xcite . on the other side , the camassa holm equation is known to be formally completely integrable in the sense that there is an associated isospectral problem @xmath3 where @xmath4 is a complex spectral parameter . solving the corresponding inverse problem is essentially equivalent to solving the initial value problem for the camassa holm equation . for this reason , it is not surprising that the encountered complications due to wave - breaking for indefinite @xmath2 reoccur when dealing with this inverse problem . and in fact , despite a large amount of articles , very little is known about the inverse problem for in the indefinite case and almost the entire literature on this subject restricts to strictly positive and smooth @xmath2 ( in which case the spectral problem can be transformed into a standard form that is known from the korteweg de vries equation @xcite ) . apart from the explicitly solvable finite dimensional case @xcite , only insufficient partial uniqueness results @xcite have been obtained so far for the inverse problem in the indefinite case . making use of recent results on the inverse spectral problem for indefinite strings in @xcite , we will be able to overcome these problems and establish the inverse spectral transform for the more general two - component camassa holm system @xmath5 where the auxiliary function @xmath6 satisfies @xmath7 for a class @xmath8 of decaying initial data to be defined below ; this should be compared to the definition of the set @xmath9 in ( * ? ? ? * section 6 ) , ( * ? ? ? * definition 3.1 ) , ( * ? ? ? * definition 4.1 ) . note that the two - component camassa holm system is equivalent to the more commonly used form in @xcite ( cf . * equation ( 4.2 ) ) ) , which is known to contain the camassa holm equation as a special case . the set @xmath9 consists of all pairs @xmath10 such that @xmath11 is a real - valued function in @xmath12 and @xmath13 is a non - negative finite borel measure on @xmath14 with @xmath15 for every borel set @xmath16 . associated with each pair @xmath17 is a distribution @xmath2 in @xmath18 defined by @xmath19 so that @xmath20 in a distributional sense , as well as a non - negative finite borel measure @xmath21 on @xmath14 defined such that @xmath22 for every borel set @xmath16 . let us point out that it is always possible to uniquely recover the pair @xmath10 from the distribution @xmath2 and the borel measure @xmath21 . now the class @xmath8 is defined by imposing an additional growth restriction on pairs in @xmath9 . the set @xmath23 consists of all pairs @xmath17 such that @xmath24 furthermore , the set @xmath8 is defined as the intersection of @xmath25 and @xmath26 . justified by the lax pair formulation of the two - component camassa holm system in @xcite and the results in @xcite , we will consider the spectral problem @xmath27 due to the low regularity of the coefficients , it is not clear how this differential equation has to be interpreted , which is why we clarify this matter in appendix [ apprelcan ] . basic properties of the spectral problem ( like realness and discreteness of the spectrum @xmath28 for example ) will be discussed in section [ secsp ] along with some further necessary conditions . in the following section , we will then solve the corresponding inverse problem for the class of coefficients corresponding to @xmath8 , giving a complete characterization of all possible spectral data . more precisely , we will establish a one - to - one correspondence between @xmath8 and a class of spectral data explicitly described by theorem [ thmip ] , which will be shown to be a homeomorphism with respect to suitable topologies . finally , in section [ seccch ] we will introduce _ the conservative camassa holm flow _ as a certain isospectral flow on @xmath8 and show that its integral curves define weak solutions of the two - component camassa holm system . as an immediate consequence of the solution of the inverse spectral problem , we will see that the phase space @xmath8 decomposes into a foliation of invariant isospectral sets @xmath29 , each of which can be parametrized by the set @xmath30 where @xmath31 is the entire function in . in terms of these coordinates , the conservative camassa holm flow on @xmath29 becomes the simple linear flow given by @xmath32 these facts are reminiscent of the fact that the conservative camassa holm flow can be viewed as a completely integrable infinite dimensional hamiltonian system . for integrals of a function @xmath33 which is locally integrable with respect to a borel measure @xmath34 on an interval @xmath35 , we will employ the convenient notation @xmath36 rendering the integral left - continuous as a function of @xmath37 . if the function @xmath33 is locally absolutely continuous on @xmath35 and @xmath38 denotes a left - continuous distribution function of the borel measure @xmath34 , then we have the integration by parts formula @xmath39 which will be used frequently throughout this article . given a discrete set @xmath28 of nonzero reals , we denote with @xmath40 the number of all @xmath41 with modulus not greater than @xmath42 . furthermore , we introduce the notation @xmath43 provided that the limit exists . similarly , subject to existence , we shall write @xmath44 where the limit is meant to be taken in the topology of locally uniform convergence . the limit in exists if and only if the limit in exists and the sum @xmath45 is finite . in this case , upon denoting the entire function in with @xmath31 , we have @xmath46 in the present section , we are going to introduce the spectral quantities that will linearize the conservative camassa holm flow on @xmath8 and derive their basic properties . to this end , we fix an arbitrary pair @xmath47 and first recall the definition of the distribution @xmath2 in as well as the one of the borel measure @xmath21 in . we introduce the spectrum @xmath28 associated with the pair @xmath10 as the set of all those numbers @xmath48 for which there is a solution @xmath33 of the differential equation @xmath49 that belongs to @xmath12 . regarding the precise meaning and some basic properties of this differential equation we refer to the discussion in appendix [ apprelcan ] . a first consequence of the growth restriction on @xmath10 in is the existence of particular solutions of the differential equation with prescribed asymptotics . [ thmthetaphi ] for every @xmath48 there is a unique solution @xmath50 of the differential equation with the asymptotics @xmath51 the derivative of @xmath50 is integrable and square integrable near @xmath52 . to begin with , let us introduce the diffeomorphism @xmath53 by @xmath54 where @xmath55 denotes the open negative / positive semi - axis . furthermore , let @xmath56 be a measurable function on @xmath55 such that ( we set @xmath57 as in appendix [ apprelcan ] ) @xmath58 for almost all @xmath59 ( note that the right - hand side is well - defined almost everywhere ) and define the non - negative borel measure @xmath60 on @xmath55 via setting @xmath61 for every borel set @xmath62 . the growth restriction on @xmath10 in implies that the function @xmath56 is square integrable and that the measure @xmath60 is finite ( thus @xmath60 can be extended to a borel measure on the closure of @xmath55 by setting @xmath63 ) . consequently ( see , for example , ( * ? ? ? * section 11.8 ) , ( * ? ? ? * theorem 1.1 ) , ( * ? ? ? * theorem a.2 ) ) , there is a unique matrix solution @xmath64 on @xmath55 of the integral equation @xmath65 for each @xmath48 . it follows from a version of gronwall s inequality ( see , for example , ( * ? ? ? * lemma 1.3 ) , ( * ? ? ? * lemma a.1 ) ) that the function @xmath64 satisfies the estimate @xmath66 where @xmath67 denotes the max norm and the non - negative function @xmath68 is given by @xmath69 plugging this estimate back into the integral equation , we furthermore get @xmath70 for the top - right entry of @xmath64 , we are actually able to sharpen the estimate to @xmath71 upon plugging this back into the integral equation one more time , we obtain @xmath72 next , we introduce the matrix function @xmath73 on @xmath14 by @xmath74 from the integral equation , a substitution ( use , for example , ( * ? ? ? * theorem 3.6.1 and corollary 3.7.2 ) ) as well as the definitions of @xmath56 and @xmath60 , one sees that @xmath75 for all @xmath76 , @xmath77 . upon employing the integration by parts formula , one sees that @xmath73 is a solution of the integral equation and thus the system . now under the additional assumption that @xmath4 is non - zero , let @xmath78 and @xmath50 be the solutions of the differential equation such that @xmath79}}(z , x ) & -\phi_\pm^{{[\pm]}}(z , x ) \end{pmatrix } , \quad x\in{{\mathbb r } } , \end{aligned}\ ] ] guaranteed to exist by lemma [ lemequcansys ] . for future purposes , let us note the bounds @xmath80}}(z , x ) { \mathrm{e}}^{\mp\frac{x}{2 } } \bigr|,\ , \bigl| \phi_\pm(z , x){\mathrm{e}}^{\pm\frac{x}{2 } } \bigr|,\ , \bigl| \phi_\pm^{{[1]}}(z , x ) { \mathrm{e}}^{\pm\frac{x}{2 } } \bigr| \\ & \qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad \leq2 { \mathrm{e}}^{4\lambda_\pm\left(\mp{\mathrm{e}}^{\mp x}\right ) |z| } , \quad x\in{{\mathbb r } } , \end{split}\end{aligned}\ ] ] that follow from the estimates and for @xmath64 , also using that @xmath81 furthermore , the inequalities in and turn into @xmath82}}(z , x ) { \mathrm{e}}^{\pm\frac{x}{2 } } \bigr|,\ , \bigl| \phi_\pm(z , x ) { \mathrm{e}}^{\pm\frac{x}{2 } } - 1\bigr|,\ , \bigl| \phi_\pm^{{[\pm]}}(z , x ) { \mathrm{e}}^{\pm\frac{x}{2 } } \pm 1\bigr| \\ & \qquad\qquad\qquad\qquad\qquad\qquad\qquad \leq 2 |z| \lambda_\pm\left(\mp{\mathrm{e}}^{\mp x}\right){\mathrm{e}}^{\lambda_\pm\left(\mp{\mathrm{e}}^{\mp x}\right ) |z| } , \quad x\in{{\mathbb r}}. \end{split } \end{aligned}\ ] ] in particular , we see that @xmath50 has the required asymptotics as well as @xmath83 which implies that @xmath50 is indeed uniquely determined by the asymptotics in . finally , the fact that the derivative of @xmath50 is integrable and square integrable near @xmath52 follows from the last bound in . it remains to set @xmath84 for all @xmath85 and note that the claim is obvious in the case when @xmath4 is zero . for every @xmath48 we introduce the complex number @xmath86 as the wronskian of the two solutions @xmath87 and @xmath88 , that is , in such a way that @xmath89 for almost all @xmath85 ; see corollary [ corwronskian ] . it follows readily that the set of zeros of @xmath31 coincides with the spectrum @xmath28 . thus , the next result implies that @xmath28 is a discrete set of nonzero reals with convergence exponent at most one . [ corevcar ] the functions @xmath90 and @xmath31 are real entire of cartwright class with only nonzero and real roots for each fixed @xmath85 . moreover , the function @xmath91 is a meromorphic herglotz nevanlinna function . since the entries of @xmath92 are real entire of cartwright class with only real roots ( cf . * section 1 ) ) , the same holds for @xmath90 upon noting that @xmath93 because the function @xmath94}}({\,\cdot\,},x)$ ] is real entire of cartwright class as well , so is @xmath31 as @xmath95}}(z , x ) - \phi_+^{{[1]}}(z , x ) \phi_-(z , x ) , \quad z\in{{\mathbb c}}. \end{aligned}\ ] ] next , we observe that the meromorphic function @xmath96}}(z , x)}{z\phi_\pm(z , x ) } , \quad z\in{{\mathbb c}}\backslash{{\mathbb r } } , \end{aligned}\ ] ] is a herglotz nevanlinna function . in fact , to this end one just needs to evaluate its imaginary part and use as well as the vanishing asymptotics of the solution @xmath50 near @xmath52 . since we may write @xmath97}}(z , x)}{z\phi_+(z , x ) } - \frac{\phi_-^{{[1]}}(z , x)}{z\phi_-(z , x ) } , \quad z\in{{\mathbb c}}\backslash{{\mathbb r } } , \end{aligned}\ ] ] this shows that the function in is a herglotz nevanlinna function as well . in particular , this guarantees that @xmath31 has only nonzero and real roots indeed . [ remetw ] although we will not prove this here , let us mention that it is possible to show that the exponential type of the entire function @xmath31 is simply given by @xmath98 where @xmath99 is the square root of the radon nikodm derivative of the absolutely continuous part of the borel measure @xmath21 ( with respect to the lebesgue measure ) . as the spectrum alone will not be enough , we need to introduce further spectral quantities . since the solutions @xmath100 and @xmath101 are linearly dependent for each fixed @xmath41 in view of corollary [ corwronskian ] , there is a unique nonzero @xmath102 such that @xmath103 which will be referred to as the coupling constant associated with @xmath104 . it will often be more convenient to work with the logarithmic coupling constant @xmath105 , defined by @xmath106 instead . we will see below that one can always recover ( the sign of ) the coupling constant @xmath107 from the quantity @xmath105 , provided the spectrum @xmath28 is known as well . furthermore , we introduce the right / left ( modified ) norming constant @xmath108 via @xmath109 upon employing and the asymptotics from theorem [ thmthetaphi ] , one sees that @xmath110 the following result gives a relation between all our spectral quantities . [ lemwderlam ] for each @xmath41 we have the relation @xmath111 where the dot denotes differentiation with respect to the spectral parameter . let us fix an arbitrary @xmath48 and introduce the function @xmath112}}(z , x ) - \dot{\phi}^{{[1]}}_\pm(z , x)\phi_\mp(z , x ) , \quad x\in{{\mathbb r}}.\end{aligned}\ ] ] from the bounds in and cauchy s integral formula , we obtain the estimates @xmath113}}(z , x ) { \mathrm{e}}^{\pm\frac{x}{2 } } \bigr| & \leq 2 ( |z|+1 ) \lambda_\pm\left(\mp{\mathrm{e}}^{\mp x}\right ) { \mathrm{e}}^{\lambda_\pm\left(\mp{\mathrm{e}}^{\mp x}\right ) ( |z|+1 ) } , \quad x\in{{\mathbb r}}.\end{aligned}\ ] ] upon writing @xmath114 as a linear combination of @xmath50 and @xmath78 , we see from the bounds in that the function @xmath115 tends to zero as @xmath116 . next , we note that from and one gets @xmath117}}(z , s ) \ , ds , \\ \phi_\pm^{{[1]}}(z,{\cdot\ , } ) \bigr|_x^y & = \int_x^y \left ( \frac{1}{4 } - z u(s ) - z^2 u'(s)^2\right ) \phi_\pm(z , s ) -z u'(s)\phi_\pm^{{[1]}}(z , s ) \ , ds \\ & \qquad - z^2 \int_x^y \phi_\pm(z , s ) d{\upsilon}(s ) , \end{aligned}\ ] ] for all @xmath76 , @xmath77 , and after differentiating with respect to @xmath4 also @xmath118}}(z , s ) + u'(s ) \phi_\pm(z , s ) \ , ds , \\ \dot{\phi}_\pm^{{[1]}}(z,{\cdot\ , } ) \bigr|_x^y & = \int_x^y \left ( \frac{1}{4 } - z u(s ) - z^2 u'(s)^2\right ) \dot{\phi}_\pm(z , s ) - z u'(s ) \dot{\phi}_\pm^{{[1]}}(z , s ) \ , ds \\ & \qquad - \int_x^y \left(u(s ) + 2 zu'(s)^2\right)\phi_\pm(z , s ) + u'(s)\phi_\pm^{{[1]}}(z , s ) \ , ds \\ & \qquad - 2 z \int_x^y \phi_\pm(z , s ) d{\upsilon}(s ) - z^2 \int_x^y \dot{\phi}_\pm(z , s ) d{\upsilon}(s).\end{aligned}\ ] ] in conjunction with the integration by parts formula , this gives @xmath119 upon letting @xmath120 in this equation , we obtain @xmath121 which readily yields the claimed identity ( one should also note that the functions @xmath11 and @xmath122 are integrable due to the growth restriction in and thus the bounds in guarantee that all integrals exist indeed ) . in particular , the previous result shows that all zeros of the entire function @xmath31 are simple . as a cartwright class function , the wronskian @xmath31 thus admits a product representation ( see , for example , ( * ? ? ? * section 17.2 ) ) of the form @xmath123 upon invoking the identities in , this fact allows us to read off trace formulas for the spectrum @xmath28 from the derivatives of the function @xmath31 at zero . [ proptrf ] the first two trace formulas are : @xmath124 in view of , we immediately obtain the first identity from and we are left to compute the second derivative of @xmath31 at zero . for this purpose , we first introduce the entire function @xmath125 by @xmath126}}(z , x ) - \theta_\pm^{{[1]}}(z , x ) \phi_\mp(z , x ) , \quad x\in{{\mathbb r}},~z\in{{\mathbb c } } , \end{aligned}\ ] ] so that we may write @xmath127 in conjunction with and , this allows us to estimate @xmath128}}(z , x ) { \mathrm{e}}^{\pm\frac{x}{2 } } \pm w(z)\bigr| \\ & \qquad \leq 4 { \mathrm{e}}^{5\lambda_\mp\left(\pm{\mathrm{e}}^{\pm x}\right ) |z|}\left(|zw(z)| \lambda_\mp\left(\pm{\mathrm{e}}^{\pm x}\right ) \left(|z|{\mathrm{e}}^{\pm x}+1\right ) + |v_\mp(z)| { \mathrm{e}}^{\pm x}\right ) \end{split } \end{aligned}\ ] ] for all @xmath85 and @xmath48 . from this we see that each of the three products @xmath129}}(z , x ) , & & \phi_+^{{[1]}}(z , x ) \phi_-(z , x ) , \end{aligned}\ ] ] is bounded uniformly in @xmath85 and locally uniformly in @xmath48 . thus , we may differentiate under the integral and in order to evaluate the derivative at zero , we first note that @xmath130 which follows from . moreover , for the top - right entry we even have @xmath131 after performing substitutions , this turns into the identities @xmath132}}(0,x ) & = \pm\frac{1}{2 } { \mathrm{e}}^{\mp\frac{x}{2 } } \int_x^{\pm\infty } \alpha_\pm(s)ds , \end{aligned}\ ] ] for all @xmath85 . plugging them into the differentiated integral in , we arrive at @xmath133 which proves the remaining claim in view of . before we are able to solve the inverse spectral problem , we need to derive one more necessary condition . more precisely , we will show that the growth restriction on the pair @xmath10 in implies certain asymptotic behavior of the spectral data . [ propmatz ] we have the identity @xmath134 to begin with , we introduce the function @xmath135 by @xmath136 from the estimates in , we see that @xmath137}}(z , x ) { \mathrm{e}}^{\pm\frac{x}{2 } } & \rightarrow \mp w(z ) , & x & \rightarrow\mp\infty , \end{aligned}\ ] ] where the convergence is locally uniform in @xmath48 . in much the same manner as above ( upon writing @xmath138 as in and using as well as to estimate ) , one obtains inequalities similar to for @xmath138 and concludes that @xmath139}}(z , x ) { \mathrm{e}}^{\pm\frac{x}{2 } } & \rightarrow - v_\pm(z ) , & x & \rightarrow\mp\infty , \end{aligned}\ ] ] where the convergence is again locally uniform in @xmath48 . thus , we may write @xmath140 that is , the function @xmath135 is the weyl titchmarsh function for ; cf . * equation ( 6.2 ) ) . in particular , it is a meromorphic herglotz nevanlinna function ( * ? ? ? * lemma 5.1 ) with simple poles at all points @xmath41 ( note that zero is not a pole since @xmath125 vanishes at zero ) with residues given by @xmath141 in view of lemma [ lemwderlam ] . after a substitution , we obtain from that @xmath142}}(0,x ) & = \frac{1}{2 } { \mathrm{e}}^{\pm\frac{x}{2 } } \int_x^{\pm\infty } \alpha_\mp(s)ds , \end{aligned}\ ] ] for every @xmath85 , which implies that @xmath143 . upon recalling the integral representation formula for herglotz nevanlinna functions ( see , for example , ( * ? ? ? * section 5.3 ) ) , this guarantees that the function @xmath135 admits the representation ( use also ( * ? ? ? * lemma 7.1 ) to conclude that there is no linear term present ) @xmath144 clearly , we may just as well write the function @xmath135 as the limit @xmath145}}(z , x)}{z\phi_\pm^{{[\pm]}}(z , x ) } = \mp \lim_{\xi\rightarrow\mp\infty } \frac{y_{\pm,21}(z,\xi)}{y_{\pm,22}(z,\xi)},\quad z\in{{\mathbb c}}\backslash{{\mathbb r}}. \end{aligned}\ ] ] now we observe that gives for every @xmath59 the expansion @xmath146 since the convergence in is uniform for all @xmath147 that lie in a small neighborhood of zero , we infer , after a substitution , that @xmath135 has the expansion @xmath148 upon differentiating and letting @xmath149 , we obtain the identity in . in this context , let us also mention the following result which characterizes the subclass of @xmath8 that gives rise to purely positive / negative spectrum . [ propdefinite ] the spectrum @xmath28 is positive / negative if and only if the borel measure @xmath21 vanishes identically and the distribution @xmath2 is non - negative / non - positive . in this case , the distribution @xmath2 can be represented by a non - negative / non - positive finite borel measure on @xmath14 ( for simplicity denoted with @xmath2 as well ) and @xmath150 if @xmath21 vanishes identically and @xmath2 is non - negative / non - positive , then and show that the spectrum is positive / negative . conversely , if the spectrum is positive / negative , then ( * ? ? ? * lemma 7.2 ) shows that the measures @xmath151 and @xmath152 vanish identically as well as that the functions @xmath153 and @xmath154 have non - decreasing / non - increasing representatives . clearly , the measures @xmath151 and @xmath152 vanish identically if and only if so does @xmath21 . given some @xmath155 with compact support , we set @xmath156 and note that @xmath157 this shows that @xmath2 is a non - negative / non - positive distribution and therefore can be represented by a non - negative / non - positive borel measure on @xmath14 . now for every @xmath158 let @xmath159 be the piecewise linear function such that @xmath160 is equal to one on @xmath161 $ ] , equal to zero outside of @xmath162 $ ] and linear in between . then from @xmath163 we see that @xmath2 is a finite measure as well as upon letting @xmath164 . we are now going to solve the corresponding inverse spectral problem for the class @xmath8 . let us point out that the given sufficient conditions on the spectral data are also necessary in view of corollary [ corevcar ] ( in conjunction with well - known properties of entire functions of cartwright class ; for example , see ( * ? ? ? * chapter 8) or ( * ? ? ? * theorem 17.2.1 ) ) as well as proposition [ propmatz ] and lemma [ lemwderlam ] . thus , we indeed obtain a complete characterization of all possible spectral data for the class @xmath8 . [ thmip ] let @xmath28 be a discrete set of nonzero reals such that the limit the number of all @xmath41 with modulus not greater than @xmath42 . ] @xmath165 exists in @xmath166 and such that the entire function @xmath31 is well - defined by , then this is the case if and only if the sum @xmath167 exists . ] @xmath168 moreover , for each @xmath41 let @xmath169 such that the sum @xmath170 is finite . then there is a unique pair @xmath47 such that the associated spectrum coincides with @xmath28 and the logarithmic coupling constants are @xmath105 for each @xmath41 . _ uniqueness . _ since the given spectral data uniquely determines the weyl titchmarsh function @xmath171 in view of and lemma [ lemwderlam ] , the uniqueness part in ( * ? ? ? * theorem 6.1 ) shows that the function @xmath154 and the borel measure @xmath152 are uniquely determined as well . it follows readily from the definition of @xmath152 in that this also uniquely determines the borel measure @xmath21 . furthermore , because of @xmath172 we conclude that the function @xmath11 is uniquely determined too and thus so is @xmath13 . _ existence . _ let us introduce the meromorphic herglotz nevanlinna function @xmath173 from the existence part of ( * ? ? ? * theorem 6.1 ) , we obtain a real - valued and locally square integrable function @xmath174 on @xmath166 and a non - negative borel measure @xmath175 on @xmath166 with @xmath176 such that the function @xmath177 is the weyl titchmarsh function for @xmath178 that is , if @xmath179 denotes the unique solution of the integral equation for each @xmath48 , then the function @xmath177 is given by @xmath180 in order to state a required fact from @xcite in a concise way , we introduce the set @xmath181 that consists of all @xmath182 such that the function @xmath174 is not equal to a constant almost everywhere near @xmath183 or such that @xmath183 belongs to the topological support of @xmath175 . let @xmath184 , @xmath185 be orthogonal to all @xmath186 that satisfy @xmath187 and functions @xmath188 , @xmath189 on @xmath166 be square integrable with respect to @xmath175 . upon setting @xmath190 we have the identity @xmath191 since the spectral transform introduced in ( * ? ? ? * section 5 ) is a partial isometry . motivated by the relation , we define the real - valued function @xmath11 on @xmath14 via @xmath192 so that @xmath11 belongs to @xmath193 and satisfies @xmath194 for almost all @xmath182 . furthermore , we define the borel measure @xmath13 on @xmath14 by , where @xmath21 is the borel measure on @xmath14 such that that we have @xmath195 for every borel set @xmath196 . since the function @xmath174 is square integrable near zero and the borel measure @xmath175 is finite near zero , it follows readily that @xmath197 for every @xmath198 . in particular , this guarantees that the function @xmath11 lies in @xmath12 near @xmath199 ( more precisely , one sees from that @xmath11 decays exponentially near @xmath199 and since @xmath200 is clearly square integrable near @xmath199 , we conclude that @xmath122 is as well ) and that the measure @xmath13 is finite near @xmath199 . we are now left to verify that the pair @xmath10 actually belongs to @xmath8 . in fact , in this case it is readily seen that @xmath174 coincides with the function @xmath154 as introduced by and that @xmath175 coincides with the borel measure @xmath152 as introduced by . from , we then see that the function @xmath177 coincides with the weyl titchmarsh function @xmath171 for this pair as introduced in the proof of proposition [ propmatz ] . upon taking and lemma [ lemwderlam ] into account , one then may conclude that the pair @xmath10 indeed gives rise to the desired spectral data . in order to show that the pair @xmath10 belongs to the set @xmath8 , let us first consider the special case when @xmath174 is equal to some constant @xmath201 almost everywhere near @xmath202 and @xmath175 vanishes near @xmath202 . from this we readily infer that @xmath203 as long as @xmath183 and @xmath204 are close enough to @xmath202 . thus , the function @xmath177 is given by @xmath205 and evaluating the limit as @xmath149 in this equation shows that @xmath206 has to be zero , which immediately implies that the pair @xmath10 belongs to @xmath8 . in particular , this proves the claim in the case when the set @xmath28 is finite ( since then the set @xmath207 is finite as well ) . thus , back in the general case , for every @xmath158 we may find a @xmath208 such that the corresponding weyl titchmarsh function @xmath209 is given by @xmath210 with the functions @xmath211 defined as in , we infer from ( * ? ? ? * proposition 6.2 ) that @xmath212 locally uniformly for all @xmath213 . in view of and , this shows that the functions @xmath214 converge pointwise to @xmath11 and since they are uniformly bounded in @xmath12 by proposition [ proptrf ] , a compactness argument shows that @xmath11 belongs to @xmath12 . to verify also the remaining growth restrictions on @xmath10 in the general case , we now may assume that there is an increasing sequence @xmath215 such that @xmath216 belongs to @xmath207 for every @xmath217 . we fix some @xmath217 and define the function @xmath218 by @xmath219 where @xmath220 is the weyl solution given by @xmath221 for every @xmath48 , and @xmath222 is the entire function defined in such a way that @xmath223 because the function @xmath220 is a weyl solution for every @xmath147 , we note that @xmath218 can be rewritten as ( see the proof of ( * ? ? ? * lemma 5.1 ) ) @xmath224 in particular , this implies that the entire function @xmath225 defined by @xmath226}}(z , x_n ) , \quad z\in{{\mathbb c } } , \end{aligned}\ ] ] is a de branges function , where @xmath227 is the solution of the differential equation so that ( compare the proof of theorem [ thmthetaphi ] ) @xmath228}}(z , x ) \end{pmatrix } = \begin{pmatrix } { \mathrm{e}}^{-\frac{x}{2 } } & 0 \\ 0 & { \mathrm{e}}^{\frac{x}{2 } } \end{pmatrix } z(z,{\mathrm{e}}^{x } ) , \quad x\in{{\mathbb r } } , \end{aligned}\ ] ] for each @xmath48 , with the quantities @xmath2 and @xmath21 defined as in and . the reproducing kernel @xmath229 in the corresponding de branges space @xmath230 is given by @xmath231 an integration by parts and the integral equation show that for every @xmath147 the function @xmath232 restricted to @xmath233 is orthogonal to all functions @xmath186 that satisfy . thus , the identity for the particular functions @xmath234 with @xmath235 and @xmath236 gives ( note that @xmath237 for all @xmath41 ) @xmath238 the values of @xmath222 on the set @xmath28 are readily evaluated and we obtain @xmath239 after employing simple linearity , continuity and density arguments . now let @xmath240 be a locally integrable , trace normed , real , symmetric and non - negative definite @xmath241 matrix function on @xmath166 such that if @xmath242 denotes the unique solution of the integral equation @xmath243 for every @xmath48 , then we have @xmath244 such a function @xmath240 is guaranteed to exist by the solution of the inverse problem for canonical systems due to de branges ; see ( * ? ? ? * theorem xii ) , ( * ? ? ? * theorem 2.4 ) . finiteness of the sum in implies ( see ( * ? ? ? * lemma 5.5 ) , ( * ? ? ? * theorem 6.17 ) and note that existence of and together establish ( * ? ? ? * condition ( c2 ) ) ; cf . @xcite ) that the entire function @xmath31 belongs to the cartwright class . moreover , since @xmath177 is a herglotz nevanlinna function , so does the function @xmath222 and thus also @xmath225 for every @xmath217 . in view of ( * theorem vii ) in conjunction with and upon employing ( * ? ? ? * theorem i ) , we obtain a @xmath245 so that @xmath246}}(z , x_n ) \end{pmatrix } = \gamma_n \begin{pmatrix } m_{12}(z , t_n ) \\ m_{22}(z , t_n ) \end{pmatrix } , \quad z\in{{\mathbb c } } , \end{aligned}\ ] ] for a real @xmath241 matrix @xmath247 with @xmath248 . evaluating at zero shows that @xmath249 for some @xmath250 . we note that the sequence @xmath251 may be chosen in such a way that it is non - increasing because the sequence of de branges spaces @xmath230 is non - increasing with respect to inclusion . thus , for every @xmath217 we have on the one side @xmath252}}(z , x_0 ) \end{pmatrix } & = \gamma_0 m(z , t_0 ) m(z , t_n)^{-1 } \gamma_n^{-1 } \begin{pmatrix } z\psi(z , x_n ) \\ -\psi^{{[+]}}(z , x_n ) \end{pmatrix } , \quad z\in{{\mathbb c } } , \end{aligned}\ ] ] and on the other side ( due to lemma [ lemequcansys ] ) also @xmath253}}(z , x_0 ) \end{pmatrix } = \omega_+(z , x_0,x_n ) \begin{pmatrix } z\psi(z , x_n ) \\ -\psi^{{[+]}}(z , x_n ) \end{pmatrix } , \quad z\in{{\mathbb c } } , \end{aligned}\ ] ] where @xmath254 denotes the matrix valued solution of the system such that @xmath255 is the identity matrix . in view of ( * problem 100 ) , this gives @xmath256 differentiating with respect to @xmath4 and evaluating at zero , we obtain @xmath257 since @xmath240 is integrable near zero , the top - right entry shows that the integrals @xmath258 are uniformly bounded for all @xmath217 . moreover , from the bottom - left entry we now infer that the quantities @xmath259 are uniformly bounded for all @xmath217 as well . finally , the top - left entry implies that the sum @xmath260 is uniformly bounded for all @xmath217 , which concludes the proof since @xmath215 . for a discrete set @xmath28 of nonzero reals such that the limit exists in @xmath166 and such that the entire function @xmath31 is well - defined by , we define the isospectral set @xmath29 as the set of all those pairs @xmath47 whose associated spectra coincide with the set @xmath28 . it is an immediate consequence of theorem [ thmip ] that the isospectral set @xmath29 is in one - to - one correspondence with the set is empty . ] @xmath261 by means of the bijection given by @xmath262 of course , the solution of the inverse spectral problem can easily be formulated in terms of the right / left norming constants as well . however , the picture ( that is , the condition corresponding to on the asymptotic behavior ) looks less symmetric . [ coripnc ] let @xmath28 be a discrete set of nonzero reals such that the limit exists in @xmath166 and such that the entire function @xmath31 is well - defined by . moreover , for each @xmath41 let @xmath108 such that @xmath263 and the sums @xmath264 are finite . then there is a unique pair @xmath47 such that the associated spectrum coincides with @xmath28 and the right / left norming constants are @xmath265 for each @xmath41 . in order to avoid misunderstandings , let us point out that this corollary has to be read as two separate statements according to the plus - minus alternative , that is , we are only able to prescribe either the right or the left norming constants . if the set @xmath28 is finite , then all conditions in theorem [ thmip ] and corollary [ coripnc ] are trivially satisfied . the solution of the inverse spectral problem in this case has a particular simple form and can be written down explicitly in terms of the spectral data ; see ( * ? ? ? * section 4 ) . we will see next that these kinds of solutions can be used to approximate solutions in the general case in a certain way . we conclude this section by establishing a continuity property for the inverse spectral transform . in order to state it , let @xmath266 belong to @xmath8 for every @xmath158 and denote all corresponding quantities in an obvious way with an additional subscript . [ propcont ] we have of the corresponding weyl titchmarsh functions given as in . ] @xmath267 as @xmath164 for all functions @xmath268 such that the limit of @xmath269 as @xmath270 exists and is finite if and only if the functions @xmath214 converge to @xmath11 pointwise and @xmath271 as @xmath164 for all @xmath85 . in this case , the functions @xmath214 converge to @xmath11 even locally uniformly and the borel measures @xmath272 converge to @xmath13 in the sense of distributions . it follows from ( * ? ? ? * proposition 6.2 ) that the first condition on convergence of and is equivalent to @xmath273 as @xmath164 for all @xmath59 , in which case the convergence is uniform as long as @xmath183 stays bounded . in conjunction with , one sees that this implies that the sequence @xmath274 converges to @xmath275 for all @xmath85 . moreover , upon employing a substitution in , we readily infer that holds as @xmath164 for all @xmath85 as well . both of these convergences are uniform as long as @xmath76 stays away from @xmath276 . in addition , for every smooth function @xmath277 on @xmath14 that vanishes near @xmath276 and such that @xmath278 has compact support we have the identity @xmath279 with @xmath280 given by . taking the limit @xmath164 on the right - hand side gives @xmath281 which shows that the borel measures @xmath272 converge to @xmath13 in the sense of distributions . for the converse direction , we now assume that the functions @xmath214 converge to @xmath11 pointwise and that holds as @xmath164 for every @xmath85 . as observed above , the latter condition is equivalent to the fact that holds as @xmath164 for all @xmath59 . upon expressing the integrals in in terms of @xmath214 and @xmath11 by using as well as a substitution , we infer ( use lebesgue s dominated convergence theorem and note that convergence of implies a uniform bound on the monotone integrands which yields a sufficient estimate for the functions @xmath214 ) that holds as @xmath164 for all @xmath59 , which establishes the equivalence in the claim . we endow @xmath8 with the unique first countable topology . ] such that the sequence @xmath266 converges to @xmath10 if and only if the functions @xmath214 converge to @xmath11 pointwise and both of the plus - minus alternatives in hold as @xmath164 for all @xmath85 . [ corcont ] if the sequence @xmath266 converges to @xmath10 in @xmath8 , then the functions @xmath214 converge to @xmath11 uniformly as well as weakly in @xmath12 , the borel measures @xmath272 converge to @xmath13 in the weak@xmath282 topology as the dual of the banach space @xmath283 . ] and the sequence @xmath284 converges to @xmath285 . we have seen in the first part of the proof of proposition [ propcont ] that the given assumption implies that the functions @xmath214 converge to @xmath11 uniformly . after choosing a suitable partition of unity ( two functions are sufficient ) we infer from that the sequence @xmath284 converges to @xmath285 . in particular , the sequence @xmath284 is bounded which allows to deduce the remaining claims in a straight forward manner . let us define _ the conservative camassa holm flow _ @xmath286 on @xmath8 as a mapping @xmath287 in the following way : given a pair @xmath47 with associated spectrum @xmath28 and logarithmic coupling constants @xmath105 for every @xmath41 , as well as a @xmath288 , the corresponding image @xmath289 under @xmath286 is defined as the unique pair in @xmath8 ( guaranteed to exist by theorem [ thmip ] ) for which the associated spectrum coincides with @xmath28 and the logarithmic coupling constants are @xmath290 of course , this definition is motivated by the well - known time evolution of spectral data for classical solutions of the camassa holm equation ; see ( * ? ? ? * section 6 ) . it is clear that the isospectral sets @xmath29 are invariant under the conservative camassa holm flow . moreover , the bijection in takes the conservative camassa holm flow on @xmath29 to a simple linear flow on @xmath291 whose solutions are explicitly given by . in conjunction with these facts , the trace formulas in proposition [ proptrf ] give rise to conserved quantities for the flow ; cf . @xcite . the two functionals @xmath292 on @xmath8 are invariant under the conservative camassa holm flow . since the wronskian @xmath31 is uniquely determined by the spectrum @xmath28 , it is invariant under the conservative camassa holm flow . therefore , we see that the functional @xmath293 on @xmath8 is invariant as well ( recall the notation from remark [ remetw ] ) . in particular , this shows that the property of @xmath99 vanishing almost everywhere ( or equivalent , the borel measure @xmath21 being singular with respect to the lebesgue measure ) is preserved . this case corresponds to the subclass of global conservative solutions of the camassa holm equation in @xcite , whereas the function @xmath99 is present for the general class of global conservative solutions of the two - component camassa holm system in @xcite . [ propflowcont ] the conservative camassa holm flow @xmath286 is continuous . let @xmath294 , @xmath295 for every @xmath158 such that @xmath296 as @xmath297 and suppose that the sequence @xmath266 converges to @xmath10 in @xmath8 so that holds as @xmath297 for all functions @xmath268 such that the limit of @xmath269 as @xmath270 exists and is finite . in conjunction with proposition [ proptrf ] , we infer from corollary [ corcont ] that there is an @xmath298 such that the intersections @xmath299 and @xmath300 are empty for every @xmath158 . given any continuous function @xmath301 on @xmath302 such that the limit of @xmath303 as @xmath270 exists and is finite , we choose a function @xmath304 such that @xmath305 as well as constants @xmath306 , @xmath307 such that @xmath308 for all @xmath309 with @xmath310 and such that @xmath311 for all @xmath158 . we then may estimate @xmath312 for every @xmath158 . since it follows readily from our assumptions that the right - hand side always converges to zero as @xmath164 , we infer that @xmath313 as @xmath164 for all functions @xmath268 such that the limit of @xmath269 as @xmath270 exists and is finite . in view of proposition [ propcont ] and the definition of the flow @xmath286 , this implies that the corresponding images @xmath314 converge to @xmath289 . for the global conservative solutions of the two - component camassa holm system in @xcite , continuity results similar to proposition [ propflowcont ] were obtained in ( * ? ? ? * theorem 5.2 ) and ( * ? ? ? * theorem 6.7 ) . however , the topologies used in @xcite are defined in a much more intricate and less explicit way ; cf . * lemma 6.4 and lemma 6.5 ) . as the main result of this section , we are now going to show that the integral curve @xmath315 for any fixed initial data @xmath316 defines a weak solution of the two - component camassa holm system . to this end , let us denote the pair @xmath317 with @xmath318 for every @xmath288 , so that @xmath11 can be regarded as a function on @xmath319 . note that the integral curve @xmath315 is continuous by proposition [ propflowcont ] , which guarantees that the function @xmath11 is at least continuous . [ thmweaksol ] the pair @xmath10 is a weak solution of the two - component camassa holm system in the sense that for every test function @xmath320 we have @xmath321 where the function @xmath6 is given by @xmath322 let us denote all the quantities corresponding to @xmath323 in an obvious way with an additional subscript . if the spectrum @xmath324 is a finite set , then it follows from ( * ? ? ? * section 5 ) that the pair @xmath10 is a weak solution of the two - component camassa holm system . otherwise , we define a pair @xmath325 for every @xmath158 in such a way that the associated spectrum coincides with @xmath326 $ ] and the right / left norming constants are @xmath327 for every @xmath328 $ ] . since the intersection @xmath326 $ ] is a finite set , the corresponding pairs @xmath266 obtained from the integral curves @xmath329 satisfy @xmath330 for every test function @xmath320 , where the function @xmath331 is given by @xmath332 moreover , our definitions and proposition [ proptrf ] guarantee the bounds @xmath333 and from proposition [ propcont ] we infer that the functions @xmath334 converge to @xmath335 locally uniformly and that the borel measures @xmath336 converge to @xmath337 in the weak@xmath282 topology ( to conclude this , we also used that the sequence @xmath338 is bounded ) for every fixed @xmath288 . in particular , this shows that the sequence of functions @xmath331 is bounded as well and converges to @xmath6 at least pointwise . finally , upon passing to the limit @xmath164 in and , we infer that the pair @xmath10 is a weak solution of the two - component camassa holm system . when the borel measure @xmath339 corresponding to the pair @xmath323 vanishes identically and the distribution @xmath340 is non - negative / non - positive , then this property is preserved by the conservative camassa holm flow in view of proposition [ propdefinite ] ; cf . * corollary 3.3 ) . in this case , the function @xmath11 is a weak solution of the camassa holm equation in the sense that for every test function @xmath320 we have @xmath341 where the function @xmath6 is given by @xmath342 under these assumptions , global weak solutions in this sense have been obtained before in @xcite . in the more challenging general case , existence of global weak solutions as in theorem [ thmweaksol ] has been established in @xcite by means of an elaborate transformation to lagrangian coordinates . if the spectrum @xmath324 associated with the pair @xmath323 is a finite set , then we recover the special class of conservative multi - peakon solutions @xcite , which can be written down explicitly in terms of the spectral data @xcite . we have seen in the proof of theorem [ thmweaksol ] that in the general case , our weak solution @xmath10 of the two - component camassa holm system can be approximated by a sequence of conservative multi - peakon solutions in a certain way ; cf . @xcite . we conclude this section with a comment on the long - time behavior of the conservative camassa holm flow : upon employing the method introduced in @xcite , it is possible to show that our weak solution @xmath10 of the two - component camassa holm system asymptotically splits into a ( in general infinite ) train of single peakons , each corresponding to an eigenvalue @xmath343 of the underlying spectral problem . for classical solutions of the camassa holm equation , this behavior was anticipated by mckean @xcite ( in accordance with numerical observations in @xcite ) and first proved in @xcite ( see also @xcite and @xcite for the multi - peakon case and @xcite for a particular class of low - regularity solutions ) . throughout this appendix , let @xmath11 be a real - valued function in @xmath344 and @xmath21 be a non - negative borel measure on @xmath14 . we define the distribution @xmath2 in @xmath345 by @xmath346 so that @xmath347 formally , and consider the ordinary differential equation @xmath348 where @xmath4 is a complex spectral parameter . of course , this differential equation has to be understood in a distributional sense in general ; cf . @xcite . in order to make this notion of solution more accessible , we will first show that the differential equation is equivalent to the first order system @xmath352 where we introduced @xmath353 . since @xmath21 is allowed to be a genuine borel measure , this system has to be understood as a measure differential equation @xcite in general . solutions of such an equation are not necessarily continuous but only locally of bounded variation and , in order to guarantee unique solvability of initial value problems , one has to impose a suitable normalization . here we will require solutions to be left - continuous , which is implicitly contained in the following definition : a solution of the system is a function @xmath354 with locally bounded variation such that @xmath355 for all @xmath76 , @xmath77 . in this case , the first component of @xmath356 is clearly locally absolutely continuous and the second component is left - continuous ; cf . . with this notion of solutions , initial value problems for the system are always uniquely solvable ; see , for example , ( * ? ? ? * section 11.8 ) , ( * ? ? ? * theorem 1.1 ) , ( * ? ? ? * theorem a.2 ) . [ lemequcansys ] if the function @xmath33 is a solution of the differential equation , then there is a unique left - continuous function @xmath357}}$ ] such that @xmath358}}(x ) = f'(x ) - z u'(x ) f(x ) \end{aligned}\ ] ] for almost all @xmath85 and the function @xmath359}}\pm\left(\frac{1}{2 } - z u\right)f \end{pmatrix } \end{aligned}\ ] ] is a solution of the system . conversely , if the function @xmath356 is a solution of the system , then its first component is a solution of the differential equation . first one notes that for every @xmath198 the equation takes the form @xmath360 upon integrating by parts . now from this we infer that a function @xmath361 is a solution of if and only if there is a @xmath198 and a constant @xmath362 such that @xmath363 for almost all @xmath85 . so if @xmath33 is a solution of , then this guarantees that there is a unique left - continuous function @xmath357}}$ ] such that holds for almost all @xmath85 . let us denote the second component in with @xmath364}}$ ] so that @xmath365}}(x ) = f'(x ) \mp \frac{1}{2 } f(x ) + z\alpha_\pm(x ) f(x ) \end{aligned}\ ] ] for almost all @xmath85 . then it is straight forward to show that @xmath366}}(s)ds + z\int_x^y \alpha_\pm(s)^2 zf(s ) - \alpha_\pm(s ) f^{{[\pm]}}(s)\,ds \\ & \qquad = \left.\pm \biggl(\frac{1}{2 } - z u \biggr)f\right|_{x}^y - \frac{1}{4 } \int_x^y f(s)ds + z \int_x^y u(s)f(s ) + u'(s)f'(s)\,ds \end{aligned}\ ] ] for all @xmath76 , @xmath77 . in combination with , we end up with @xmath367}}\right|_{x}^{y } = \pm\frac{1}{2 } \int_x^y f^{{[\pm]}}(s)ds & + z\int_x^y \alpha_\pm(s)^2 zf(s ) - \alpha_\pm(s ) f^{{[\pm]}}(s)\,ds + z \int_x^y zf \,d{\upsilon}\end{aligned}\ ] ] for all @xmath76 , @xmath77 , which shows that is a solution of the system . now suppose that @xmath356 is a solution of the system and denote the respective components with subscripts . the first component of shows that @xmath368 belongs to @xmath193 with @xmath369 for almost all @xmath85 . in combination with the second component of this gives @xmath370 for some @xmath198 and almost all @xmath85 . plugging twice into this equation , we see that holds with @xmath33 replaced by @xmath368 for some constant @xmath371 . as noted before , this guarantees that @xmath368 is a solution of the differential equation . [ coree2 ] for every @xmath198 and @xmath372 , @xmath373 there is a unique solution @xmath33 of the differential equation with the initial conditions @xmath374}}(c ) = d_2 . \end{aligned}\ ] ] if @xmath372 , @xmath375 and @xmath4 are real , then the solution @xmath33 is real - valued as well . if @xmath4 is zero , then the solutions of the differential equation are given explicitly and the claim is obvious . otherwise , it follows from unique solvability of initial value problems for the system in conjunction with lemma [ lemequcansys ] . we are now left to introduce the wronskian @xmath376 of two solutions @xmath33 , @xmath38 of the differential equation . although the following result is essentially a consequence of lemma [ lemequcansys ] and the fact that the traces of the matrices in vanish , we provide an independent direct proof . [ corwronskian ] for any two solutions @xmath33 , @xmath38 of the differential equation there is a unique complex number @xmath376 such that @xmath377 for almost all @xmath85 . this number is zero if and only if the functions @xmath33 and @xmath38 are linearly dependent . first of all , note that if @xmath33 is a solution of the differential equation and @xmath378 , then and the integration by parts formula show that @xmath379}}h\right|_{x}^y & = \int_x^y f'(s)h'(s)ds + \frac{1}{4 } \int_x^y f(s)h(s)ds \\ & \qquad - z\int_x^y u(s)f(s)h(s ) + u'(s)(fh)'(s)\,ds - z^2 \int_x^y fh\ , d{\upsilon}\end{split}\end{aligned}\ ] ] for all @xmath76 , @xmath77 . if @xmath38 is another solution of the differential equation , then @xmath380}}(x ) - f^{{[1]}}(x ) g(x ) \end{aligned}\ ] ] for almost all @xmath85 . employing , one sees that the right - hand side of is constant , which proves existence of the number @xmath376 . the remaining claim follows in a standard way upon utilizing the uniqueness part in corollary [ coree2 ] . let us mention that one can easily get rid of the constant potential term in the first order system by scaling solutions with the matrix function @xmath381 at the cost of rescaling the remaining matrix coefficients in as well . it is even possible to further transform into a canonical system in standard form @xmath382 with a locally integrable , non - negative and trace normed matrix function @xmath383 . since this is of no need for our purposes , we will not do this here but refer to ( * ? ? ? * proof of theorem 6.1 ) , where such a transformation has been employed . f. gesztesy and r. weikard , _ some remarks on the spectral problem underlying the camassa holm hierarchy _ , in _ operator theory in harmonic and non - commutative analysis _ , 137188 , oper . theory adv . , 240 , birkhuser / springer , cham , 2014 . k. grunert , h. holden and x. raynaud , _ lipschitz metric for the two - component camassa holm system _ , in _ hyperbolic problems : theory , numerics , applications _ , 193207 , aims on applied mathematics * 8 * , aims , 2014 . h. holden and x. raynaud , _ a numerical scheme based on multipeakons for conservative solutions of the camassa holm equation _ , in _ hyperbolic problems : theory , numerics , applications _ , 873881 , springer , berlin , 2008 .

liouville function , multiplicative function , factorization into primes , mobius function , riemann hypothesis this paper investigates the analytic properties of the liouville function s dirichlet series that obtains from the function , where is a complex variable and is the riemann zeta function . the paper employs a novel method of summing the series by casting it as an infinite number of partial sums over sub - series that exhibit a certain symmetry and rapid convergence . in this procedure , which heavily invokes the prime factorization theorem , each sub - series has the property that it oscillates in a predictable fashion , rendering the analytic properties of the dirichlet series determinable . with this method , the paper demonstrates that , for every integer with an even number of primes in its factorization , there is another integer that has an odd number of primes ( multiplicity counted ) in its factorization . furthermore , by showing that a sufficient condition derived by littlewood ( 1912 ) is satisfied , the paper demonstrates that the function is analytic over the two half - planes and . this establishes that the nontrivial zeros of the riemann zeta function can only occur on the critical line . [ firstpage ]

1609.06971

this paper investigates the behaviour of the liouville function which is connected to riemann s zeta function , @xmath3 , defined by@xmath8 where @xmath9 is a positive integer and @xmath2 is a complex variable , with the series being convergent for @xmath10 . this function has zeros ( referred to as the trivial zeros ) at the negative even integers @xmath11riemann s hypothesis claims that the nontrivial zeros of the zeta function all occur on the critical line which occurs at @xmath12 it has been shown that there are an infinite number of zeros on the critical line . the claim that these are the only nontrivial zeros has eluded proof to date , and this paper demonstrates that the riemann hypothesis ( rh ) is resolvable by tackling the liouville function s dirichlet series generated by @xmath13 , which is readily rendered in the form@xmath14 where @xmath15 is the liouville function defined by @xmath16 , with @xmath17 being the total number of prime numbers in the factorization of @xmath9 , including the multiplicity of the primes . we would also need the summatory function @xmath18 , which is defined as the partial sum upto @xmath19 terms of the following series : @xmath20 since the function @xmath4 will exhibit poles at the zeros of @xmath21 , we seek to identify where @xmath3 can have zeros by examining the region over which @xmath4 is analytic . by demonstrating that a sufficient condition , derived by littlewood ( 1912 ) , for the rh to be true is indeed satisfied , we show that all the nontrivial zeros of the zeta function occur on the critical line @xmath7 . briefly , our method consists in judiciously partitioning the set of positive integers into sets of sub - series and couching the infinite sum in ( 2 ) into an infinite number of sums over these sub - series , with each such series being uniformly convergent . this method of considering a slowly converging series as a sum of many sub - series was used by the author in problems where neumann series were involved.(eswaran ( 1990 ) ) . in this paper we were able to break up the sum of the liouville function into sums over many sub - series whose behaviour is predictable . it so turns out that one prime number @xmath22 ( and its powers ) which is associated with a particular sub - series controls the behaviour of the sub series . each sub - series is in the form of rectangular functions ( waves ) of unit amplitude but ever increasing periodicity and widths -we call these as ` harmonics ' , so every prime number is thus associated with such harmonic rectangular functions which then play a role in contributing to the value of @xmath18 . it so turns out that if n goes from n to n+1 , the new value of @xmath23 solely depends upon the factorization of n+1 , and the particular harmonic that contributes to the change in @xmath18 is completely determined by this factorization . since factorizations of numbers involve primes which are uncorrelated , we are to deduce that the statistical distribution of @xmath18 when n is large is like that of the cumulative sum of n coin tosses , ( a head contributing + 1 and a tail contributing -1 ) , and thus logically lead to the final conclusion of this paper . we found a new method of factorizing every integer and placing it in an exclusive subset , where it and its other members form an increasing sequence which in turn factorize alternately into odd and even factors ; this method exploited the inherent symmetries of the problem and was very useful in the present context . once this symmetry was recognized , we saw that it was natural to invoke it in the manner in which the sum in ( 2 ) was performed . we may view the sum as one over subsets of series that exhibit convergence even outside the domain of the half - plane @xmath10 . we were rewarded , for following the procedure pursued in this paper , with the revelation that the liouville function ( and therefore the zeta function ) is controlled by innumerable rectangular harmonic functions , whose form and content are now precisely known and each of which is associated with a prime number and all prime numbers play their due role . and in fact all harmonic functions associated with prime numbers below or equal to a particular value n determine the present value of l(n ) . when we are oblivious to the underlying symmetry being alluded to here , we render the summation in ( 2 ) less tractable than necessary . this is precisely what happens when we perform the summation in the usual manner , setting @xmath24 in sequence . the liouville function @xmath15 is defined over the set of positive integers @xmath9 as @xmath25 , where @xmath26 is the number of prime factors of @xmath9 , multiplicities included and by definition @xmath27 . thus @xmath28 when @xmath9 has an even number of prime factors and @xmath29 when it has odd . it is a completely arithmetical function obeying @xmath30 for any two positive integers @xmath31 . we shall consider subsets of positive integers such as @xmath32 arranged in increasing order and are such that their values of @xmath33 alternate in sign : @xmath34 it turns out that we can label such subsets with a triad of integers , which we now proceed to do . to construct such a labeling scheme , consider an example of an integer @xmath9 that can be uniquely factorized into primes as follows:@xmath35 where @xmath36 are prime numbers and the @xmath37 are the integer exponents of the respective primes , and @xmath38 is the largest prime with exponent exceeding @xmath39 . integers of this sort , with at least one multiple prime factor are referred to here as class i integers . in contrast , we shall refer to integers with no multiple prime factors as class ii integers . a typical integer , @xmath40 , of class ii may be written@xmath41 where , once again , the prime factors are written in increasing order . we now show how we construct a labelling scheme for integer sets that exhibit the property in ( 2.3 ) of alternating signs in their corresponding @xmath42 s . first consider class i integers . with reference to ( 2.4 ) , we define integers @xmath43 as follows:@xmath44 in ( 2.6 ) , @xmath45 is the product of all primes less than @xmath38 , @xmath22 is the largest multiple prime in the factorization , and @xmath46 is the product of all prime number larger than @xmath38 in the factorization . thus the class i integer @xmath9 can be written@xmath47hence we will label this integer @xmath9 as @xmath48,using the triad of numbers@xmath49 and the exponent @xmath50 . it is to be noted that @xmath46 will consist of prime factors all larger than @xmath22 and @xmath46 can not be divided by the square of a prime number . consider the infinite set of integers , @xmath51 , defined by@xmath52 the class i integer @xmath9 necessarily belongs to the above set because @xmath53 . since the consecutive integer members of this set have been obtained by multiplying by @xmath22 , thereby increasing the number of primes by one , this set satisfies property ( 3 ) of alternating signs of the corresponding @xmath33 s . note that the lowest integer of this set @xmath54 of class i integers is @xmath55 . we may similarly form a series for class ii integers . the integer @xmath40 in ( 5 ) may be written @xmath56 , with @xmath57 , @xmath58 , and @xmath59 . this class ii integer is put into the set @xmath51 defined by@xmath60 the set containing class ii integers is distinguished by the facts that @xmath59 for all of them , their largest prime factor is always @xmath22 and none of them can be divided by the square of a prime number @xmath61 , s.t . @xmath62 , in other words the factor @xmath45 can not be divided by the square of a prime . in this set , too , the @xmath33 s alternate in sign as we move through it and so property ( 2.3 ) is satisfied . again , note that the lowest integer of this set @xmath63 is class ii integer : @xmath64 , all the others are class i. in what follows , we shall find it handy to refer to the set of ascending integers comprising @xmath51 as a ` tower ' . it is important to distinguish between a tower ( or set ) described by a triad like @xmath49 and an integer belonging to that set . the set or tower of class i integers described by the label @xmath49 is the infinite series @xmath65 , the first element of which is @xmath55 and all other members of which are @xmath66 , where @xmath67 . a set or tower containing a class ii integer described by @xmath68 is the infinite sequence @xmath69 , eq.(2.9),of which only the first element @xmath64 is a class ii integer and all other members , @xmath70 , where @xmath71 , are class i , because the latter have exponents greater than 1 . for convenient reference , we shall refer to the first member of a tower as the base integer or the base of the tower . it is also worth noting that when we refer to a triad like @xmath72 , where @xmath73 , we are invariably referring to the integer @xmath66 and not to any set or tower . labels for sets do not contain exponents ; only those for integers do . of course , the particular integer @xmath74 belongs to set or tower @xmath49 two simple examples illustrate the construction of the sets denoted by @xmath54 : ex . 1 : the integer @xmath75 , which factorizes as @xmath76 , is clearly a class i integer since it is divisible by the square of a prime number in fact there are two such numbers , @xmath77 and @xmath78but we identify @xmath22 with @xmath78 as it is the larger prime . it is a member of the set @xmath79 . 2 : the integer @xmath80 , which factorizes as @xmath81 , is a class ii integer because it is not divisible by the square of a prime number . it belongs to the set @xmath82 . note that two different integers can not share the same triad . is the product @xmath83 , which obviously can not take on two distinct values . ] and two different triads can not represent the same integer . and @xmath84 represent the same integer , say @xmath9 . then we must have @xmath85it follows that at least two numbers of the tetrad @xmath86 must differ from their counterparts in the tetrad @xmath87 . since the factorization of @xmath9 is unique , this is impossible . ] thus the mapping from a triad to an integer is one - one and onto . a formal proof is in the appendix . the following properties of the sets @xmath51 may be noted : \(a ) the factorization of an integer @xmath9 immediately determines whether it is a class i or a class ii type of integer . \(b ) the factorization of integer @xmath9 also identifies the set @xmath51 to which @xmath9 is assigned . \(c ) the procedure defines all the other integers that belong to the same set as a given integer . \(d ) every integer belongs to some set @xmath51 ( allowing for the possibility that @xmath59 ) and only to one set . this ensures that , collectively , the infinite number of sets of the form @xmath51 exactly reproduce the set of positive integers @xmath88 , without omissions or duplications . our procedure , taking its cue from the deep connection between the zeta function and prime numbers , has constructed a labeling scheme that relies on the unique factorisation of integers into primes . in what follows , we shall recast the summation in ( 2 ) into one over the sets @xmath51 . the advantage of breaking up the infinite sum over all positive integers into partial sums over the @xmath51 sets will soon become clear . we shall now implement the above partitioning of the set of all positive integers to examine the analytic properties of @xmath4 in ( 1.2 ) . we shall rewrite the sum in ( 1.2 ) into an infinite number of sums of sub - series , ensuring that each sub - series is uniformly convergent even as @xmath89 . we begin , however , by assuming that @xmath10 , which makes the series in ( 1.2 ) absolutely convergent . we write the right hand side in sufficient detail so that the implementation of the partitioning scheme becomes self - evident:@xmath90 we have explicitly written out a sufficient number of terms of the right hand side of ( 1.2 ) so that those corresponding to each of the first 30 integers are clearly visible as a term is included in one ( and only one ) of the sub - series sums in ( 3.10 ) . on the right hand side , the second term contains the integers @xmath91 ; the third contains @xmath92 ; the fourth contains @xmath93 ; the fifth contains @xmath94 ; sixth contains @xmath95 ; the seventh contains @xmath96 ; the eighth contains @xmath97 ; the ninth contains @xmath98 ; and so on . note that in the ninth , fifteenth , and twentieth terms the running index is deliberately switched from @xmath99 to @xmath100 to alert the reader to the fact that the summation starts from @xmath77 and not from @xmath101 as in all the other sums . ( note that , in the ninth term , the class i integer @xmath102 is assigned to the set @xmath103 and not to the set @xmath104 , because the first term identifies @xmath22 as @xmath77 and @xmath46 as @xmath78 where as the second term onwards @xmath78 has exponents , which violates our rules of precedence and would be an illegitimate assignment given our partitioning rules . ) the sub - series in ( 3.10 ) have one of two general forms:@xmath105 \notag\ ] ] or @xmath106\ ] ] the above geometric series occurring within square brackets in the above two equations can actually be summed ( because they are convergent),(see whittaker and watson ) but we will refrain from doing so , and ( 1.2 ) can be rewritten as @xmath107 where the first group of summations pertain to class i integers @xmath9 characterized by the triad @xmath108 and the second group pertain to those integers which are characterized by set @xmath109 the first member in the set is a class ii integer andothers class i. in the above we have defined the function @xmath110 of the complex variable @xmath2 which is a sub - series involving terms over only the tower @xmath49 for a class i integer as follows@xmath111 and the function @xmath112 of the complex variable @xmath2 which is a sub - series involving terms over only the tower @xmath113 whose 1st term is a class ii integer as@xmath114 with the understanding that when @xmath59 we use the function in ( 3.14 ) instead of ( 3.13 ) , we may write @xmath4 as@xmath115 comparing the above eq.(3.15 ) with eq(3.10 ) one can easily see that each term which appears as a summation in ( 3.10 ) is actually a sub - series over some tower which we denote as @xmath116 in ( 3.15 ) . so we see that @xmath4 has been broken up into a number of sub - series . the important point to note that the @xmath117 value of each term in the sub - series changes its sign from + 1 to -1 and then back to + 1 and -1 alternatively . therefore if the starting value of @xmath117 at the base was + 1 then the cumulative contribution of this tower ( sub series ) to @xmath18 as n , the upper bound , increases from @xmath19 to @xmath118 will fluctuate between be 0 and 1 . for some other tower whose base value of @xmath117 is @xmath119 then its cumulative contribution to l(n ) will fluctuate between or 0 and @xmath119 , these cumulative contributions can be represented as in the form of a rectangular wave.(figure 1 ) we have arrived at a critical point in our paper . we have cast the original function @xmath13 as a sum of functions of @xmath2 . since the triad @xmath74 uniquely characterises all integers , the summations over @xmath120 and @xmath46 above are equivalent to a summation over all positive integers @xmath9 , as in ( 1.2 ) , though not in the order @xmath121 . the manner in which the triads were defined ensures that there are neither any missing integers nor integers that are duplicated . although we did not explicitly do it , we mentioned in passing that the sum over @xmath100 in ( 3.13)and ( 3.14 ) is readily performed since it is a geometric series ( see ( 3.11 ) that rapidly converges . this is true not merely for @xmath10 but also as @xmath122 . whether @xmath4 converges when the summation is carried out over all the towers @xmath49 and , if so , over what domain , of @xmath2 is the central question that we seek to answer in the next section . the answer to which as we shall see determines the analyticity of f(s ) and thus resolves the riemannian hypothesis . we can recast ( 3.15 ) , still in the domain @xmath10 , in the form@xmath123 where @xmath124 is a function appropriately defined below . by construction , every @xmath9 in the above summation can be written as@xmath125 where @xmath126 , @xmath127 , and @xmath128 are positive integers , @xmath127 is the largest prime in the factorization of @xmath9 , with either ( _ i _ ) an exponent @xmath129 , and @xmath130 is the product of primes larger than @xmath131 but with exponents equal to @xmath101 ( for class i integers ) or ( _ ii _ ) it is the largest prime factor with @xmath132 and @xmath133 ( for class ii integers ) . we define @xmath124 as follows:@xmath134 the factors @xmath135 and @xmath136 in the denominators of ( 3.13 ) and ( 3.14 ) are simply @xmath137 , where @xmath9 is the integer characterized by the @xmath72 triad ( with @xmath59 in the latter case ) . we are now in a position to examine the summatory liouville function @xmath18 by actually summing up the individual contribution from each sub - series . to do all this systematically , we will explicitly illustrate the process starting from @xmath138 up to @xmath139 . each of these numbers is factorized and expressed uniquely as a triad . the n=1 is a constant term , which is the trivial @xmath140 , then the next number @xmath141 , is contained in the tower shown below the one corresponding to @xmath142 ; and @xmath143 , is the tower below the previous ; @xmath144 however @xmath145 is already contained in the tower @xmath146 as its second member ; the next n s : @xmath147 , give rise to the new towers @xmath148 of course is the third member of the old tower @xmath146 similarly @xmath149 is the 2nd member of @xmath150 . after this the new towers which make their appearance are : @xmath151 and @xmath152 . see figure 1 . now each tower @xmath49 contributes to @xmath18 ( consider n fixed in the following ) according to the following rules : \(i ) a particular tower will contribute only if its base number is less than or equal to n i.e @xmath153 \(ii ) and the contribution @xmath154 , to @xmath18 from this particular tower will be exactly as follows : case a ; class ii integer @xmath155 @xmath156 , where r is the largest integer such that @xmath157 case b ; class i integer @xmath158 @xmath159 , where k is the largest integer such that @xmath160 now since each successive @xmath33 changes sign from @xmath161 to @xmath119 or vice a versa , the contributions of each tower can be thought of as a rectangular wave of ever increasing width but constant amplitude @xmath119 or @xmath161 , see figure . to find the value of l(n ) ( n fixed),all we need to do is count the jumps of each wave , as we move from n=0 a jump upwards is called a positive peak a jump downwards is a negative peak . draw a vertical line at n , we are assured that it will hit one and only one peak ( positive or negative ) in one of the sub - series ; then count the total number of positive peaks p(n ) and negative peaks q(n ) , of the waves on and to the left of this vertical line , then @xmath162 ; the reason for this rule will be clear after the next section . let us take an example let s take @xmath163 , there is a positive peak for the constant term ( 1,1,1),the next wave ( 1,2,1 ) contributes one negative peak ( at 2)and a positive peak ( at 4),the wave ( 1,3,1 ) contributes a @xmath119 peak ( at 3)and ( 5,1 ) contributes a @xmath119 ( at 5)thus a total of three negative peaks and two positive peaks adding up we see that @xmath164 , which is of course correct . now if we take @xmath165,let us now draw a vertical line at n=10 , looking at this line and to its left we see that there are additionally three positive and two negative peaks thus adding this contribution of+1 to the previously calculated value @xmath166 we get @xmath167.(two red vertical lines just just beyond n=5 and n=10 are drawn for convenience ) now if wish to compute l(15 ) we see that there are three more negative peaks and two positive peaks thus giving a value @xmath168 . counting the peaks further on it is easy to check that @xmath18 is correctly predicted for every value of n up to 30 and in particular : @xmath169 ; @xmath170 and @xmath171 . in summary to calculate l(n ) we merely need to count the negative and positive peaks of the waves on n and to the left of n. in the figure we have drawn a number of waves and labeled the tower to which each belongs using a triad of numbers . they are sufficient for one to easily calculate l(n ) up to n=30 and check them out by comparing the numbers with the plot of l(n)shown on the top of the figure . we turn to a more fundamental point : we now show the distribution of value l(n ) as equivalent to the distribution obtained by summing n coin tosses . we shall now examine the analyticity of the function @xmath172 utilizing a technique introduced by littlewood ( 1912 ) . in this , we follow the treatment of edwards ( 1974 , pp . 260 - 261 ) . the series in ( 16 ) can be couched as the integral@xmath173 where @xmath174 is a step function that is zero at @xmath175 and is constant except at the positive integers , with a jump of @xmath124 at @xmath9 . the value of @xmath176 at the discontinuity , at an integer @xmath9 , is defined as @xmath177 $ ] , which is equal to @xmath178 . assuming @xmath10 , integration by parts yields@xmath179-\int\limits_{0}^{\infty } g(x)d[x^{-s } ] \quad\quad \quad\quad ( 5.20)\notag\\ & = & \underset{x\rightarrow \infty } { \lim } \quad \lbrack x^{-s}g(x)+s\int\limits_{0}^{x}g(x)x^{-s-1}dx \notag \\ & = & s\int\limits_{0}^{\infty } g(x)x^{-s-1}dx , \quad\quad(5.21 ) \notag\end{aligned}\ ] ] where the last step follows from the fact that @xmath180 , which implies that @xmath181 as @xmath182 . we further observe , following littlewood ( 1912 ) , that as long as @xmath183 grows less rapidly than @xmath184 for some @xmath185 it follows that both terms in the line preceding ( 21 ) converge for all @xmath2 in the half - plane @xmath186 , that is , for @xmath187 . by analytic continuation , @xmath188 converges in this half - plane . since this result will be important in what follows , we record it here . _ theorem 1 [ littlewood ( 1912 ) ] : when _ @xmath183 _ _ grows less rapidly with _ _ @xmath189 _ _ than _ _ @xmath184 _ _ for some _ _ @xmath185 _ _ , _ _ @xmath4__is analytic in the half - plane _ _ @xmath187__. _ _ we shall now demonstrate that the sufficient condition stated in theorem 1 is satisfied for a specific value of @xmath190 that settles the riemann hypothesis . using the definition of @xmath191@xmath192 we may rewrite @xmath191 as@xmath193 , \tag{5.23}\ ] ] where @xmath194 and @xmath195 are kronecker deltas ( e.g. @xmath196 if @xmath197 and @xmath198 otherwise ) . the summations over @xmath45 , @xmath22 , @xmath100 , and @xmath46 in ( 5.23 ) are undertaken with the understanding that the triads @xmath74 will only include integers @xmath199 . since the summation over @xmath100 is over an individual tower(if we keep ( m , p , u ) fixed we can write(5.23 ) as @xmath200 this is nothing but eq.(3.15 ) evaluated from each subseries @xmath201 by making @xmath202 . of course , what we have called @xmath191 is really the summatory liouville function,@xmath18 , defined earlier by ( 1.2b):@xmath203 expression ( 5.23 ) is crucial because , in the light of theorem 1 , its behaviour will determine the validity of the riemann hypothesis . every term in the summation in ( 5.23 ) is either @xmath161 or @xmath119 . we need to determine , for given @xmath19 , how many terms contribute @xmath161 and how many @xmath119 , and then determine how the sum @xmath191 varies with @xmath19 . as we go through the list @xmath204 , we are assigning the integers to various sets of the sort @xmath51 . to use our terminology of towers , we shall be ` filling up ' slots in various towers from the bottom up until we have exhausted all @xmath19 integers . ( when @xmath19 increases , in general , we shall not only be filling up more slots in existing towers but also adding new towers that were previously not included . ) so the behaviour of @xmath191 is determined by how many of the numbers that do not exceed the upperbound @xmath19 contribute @xmath161 and how many @xmath119 . it is convenient to identify the @xmath33 of an integer by the triad which uniquely defines that integer . to avoid abuse of notation , we shall denote the value @xmath15 in terms of the @xmath33 value of the base integer of the tower to which @xmath9 belongs . we will define the @xmath33 of the base of a tower in uppercase , as @xmath205 . in other words if @xmath206 then it will belong to a tower whose base number is @xmath207 , where @xmath208 @xmath209and @xmath210 @xmath211 now we define @xmath212 , since the @xmath42 of a product of integers is the product of the @xmath33 of the individual integers . of course , once we know @xmath213 we will know the @xmath33 of all other numbers belonging to the tower because they alternate in sign . to determine the behaviour of @xmath191 , the following theorem is important . _ theorem 2 : for every integer that is the base integer of a tower labeled by the triad _ @xmath49 _ _ , and therefore belonging to the set _ _ @xmath54 , _ there is another unique tower labeled by the triad _ @xmath214 _ and therefore belonging to the set _ @xmath215 _ _ with a base integer for which _ _ @xmath216 . : let us write the integers at the base of a tower in the form @xmath217 described by the triad @xmath49 , where we shall assume that @xmath218 if @xmath219 and @xmath132 if @xmath59 . these correspond to the lowest members of sets of class i and class ii integers , respectively , which are the integers of concern here . in the constructions below , we shall multiply ( or divide ) @xmath220 by the integer @xmath77 . since @xmath77 is the lowest prime number , such a procedure does not affect either the value of @xmath22 or @xmath46 in an integer and so we can hold these fixed . we begin by excluding , for now , triads of the form @xmath221 , integers which are single prime numbers . we allow for this in case 3 below . : suppose @xmath45 is odd . it is clear that if we choose @xmath222 , then @xmath223 . we may say that @xmath49 and @xmath224 are ` twin ' pairs in the sense that their @xmath225s are of opposite sign . note that @xmath49 and @xmath224 are integers at the base of two different towers ; they are not members of the same tower . ( recall that the members of a given tower are constructed by repeated multiplication of @xmath22 . ) : suppose @xmath45 is even . in this case , we need to ascertain the highest power of @xmath77 that divides @xmath45 . if @xmath45 is divisible by @xmath77 but not by @xmath226 , assign @xmath227 . ( so @xmath228 gets assigned to @xmath229 , and @xmath230 , by case 1 above , gets assigned to @xmath231 . ) more generally , suppose the even @xmath45 is divisible by @xmath232 but not by @xmath233 , where @xmath100 is an integer . then , if @xmath100 is even , assign @xmath234 ; and if @xmath100 is odd , assign @xmath227 . ( so @xmath235 gets assigned to @xmath236 . and , in reverse , @xmath237 gets assigned to @xmath238 . ) thus for odd @xmath45 the following sequence of pairs ( twins ) hold : @xmath49 and @xmath239 are twins at bases of different towers having @xmath42s of opposite signs , @xmath240 and @xmath241 are twins at bases of different towers having @xmath33s of opposite signs , @xmath242 and @xmath243 are twins at bases of different towers having @xmath33s of opposite signs , and so on . : now consider the case where the triad describes a prime number , that is , it takes the form @xmath221 . for the moment , suppose this prime number is not @xmath244 in this case , where @xmath245 , we simply assign @xmath246 . clearly , @xmath247 , and the numbers @xmath248 and @xmath249 are at the bases of different towers . : finally , consider the case where the prime number is 2 , that is , the integer @xmath146 , for which @xmath250 . we match this prime to the integer @xmath101 . by definition @xmath251 . thus the first two integers have opposite signs for their values of @xmath42 . @xmath252 so , in partitioning the entire set of positive integers , the number of towers that begin with integers for which @xmath253 is exactly equal to those that begin with integers for which @xmath254 . the consequences of the above theorem is that each integer has a unique twin whose @xmath33 value is of the opposite sign . this is because if the base of two towers are twins the next higher number in the first tower is the twin of the next higher number in the second tower , and so on . theorem 2 has the immediate implication that , we believe , has never been established to date ( see borwein et al ( 2006 ) ) : _ theorem 3 : in the set of all positive integers , for every integer which has an even number of primes in its factorization there is another unique integer ( its twin ) whose factorization has an odd number of factors . _ in the first part of this section we detailed littlewood s proof of his criteria ( theorem 1 ) for the condition that must hold for a function such as @xmath4 , to be analytically continuable to the line @xmath255 , when it is known that it is analytic in the region @xmath10 . the need to do this was because we wished to use his criteria for our function @xmath4 which is given in the form ( 3.10)or ( 3.15 ) above . so we have to find whether @xmath256 the crucial value is the exponent @xmath190 which needs to be found,(of course r.h predicts @xmath257 , but this has to be confirmed ) . in other words we need to study the growth of @xmath18 as @xmath19 becomes very large . in this section we also derived two more theorems , viz theorem 2 and theorem 3 , which we will be needing to find the value of this exponent @xmath190 . we are nearing the final leg of our journey . however , to come to a proper conclusion regarding the riemann hypothesis we need some more investigations to determine the proper behavior of l(n ) in order to find the true value of the exponent @xmath190 and thereby settle the riemann hypothesis . hence this will be our task in the next sections . now we will investigate the properties of the summatory liouville function @xmath258 . we perform the investigation under two separate ( exclusive cases ) case 1 , when the sequence is expected to be non - random and case 2 , when the sequence is expected to be random . we are forced to resort to this because the literature is ambivalent about the random nature of prime numbers and prime factorization . and as there is no telling ab initio whether the lioville sequence behaves like case 1 or as case 2 . so we investigate both the cases and report our findings . we need to define precisely what we mean by a ` non - random behavior ' so that we can proceed with our mathematical investigation . here we follow the criteria of many computer scientists , who believe that a sequence can be considered truly random if the sequence does not repeat and any sequence that repeats no matter how much later , should not be considered random . ( see press.etal or donald knuth ) we will adopt this definition because it seems appropriate and precisely definable . so in this subsection we assume that the sequence of lambdas is inherently cyclic and repeats after some @xmath259 consecutive numbers . our first step is to verify if this scenario is really possible at all . we do this by using the fact that every @xmath260 can be calculated if we know which tower the particular number @xmath9 belongs to . then we check out if two numbers belonging to the sequence but occurring in different cycles can really be members of the same tower(we are really checking for consistency as the foregoing will clarify ) . after that we estimate the contribution to l(n ) from such a hypothetical sequence and then examine the consequences . * a non - random sequence * is a sequence of @xmath259 consecutive integers @xmath261 which are such that the related sequence : generated by the liouville function viz : @xmath262 repeats . since the integers are consecutive @xmath263 and the @xmath117 seuence repeats : @xmath264 for ever positive integer @xmath100 , @xmath265 . now if we denote @xmath266 for all integer @xmath267 , therefore the entire sequence after @xmath268 , has the form : @xmath269 @xmath270 we then call the set of consecutive numbers @xmath271 , these are of course a collection of @xmath161 s and @xmath119 s in random order . and in ( 4 ) we call the first @xmath259 numbers as the first cycle , the next @xmath259 numbers as the 2nd cycle and so on , of course the overall sequence has a period @xmath259 . * a.1 investigation : * we will now investigate belonging to a cycle the probability of @xmath272 , is exactly equal to the probability of @xmath273 . but this already follows from theorem 3 which is for the general case , so if @xmath259 is finite this conclusion is inescapable and one can skip all this and goto the paragraph final conclusion: preceding eq . 6.11 . the only reason why the author has included this subsection in this paper is because of the curious appearance of fermat s little theorem into the argument to satisfy consistency . ] the conditions under which a tower @xmath274 can generate numbers which belong to @xmath275 , ( note we do not assume a single tower can generate all numbers in the cycle ) . let us suppose @xmath275 as defined above , is really such a sequence . let us take some prime p which belongs to a tower @xmath49 , and consider the terms in the tower of class i integers @xmath276 now since @xmath259 is finite it is obvious that @xmath277 must be a number somewhere in some cycle let us say it is equal to the @xmath278 @xmath279 term in the @xmath280 cycle . but the integer value of the @xmath278 @xmath281 term is : @xmath282 therefore we must have the value of @xmath277 equal the former : @xmath283 @xmath284 of course by the very definition of the sequence and its periodicity we will have @xmath285 now the list of numbers in ( 5 ) are infinite but the cycle length @xmath259is finite ; also all the numbers in ( 5 ) occur as a term in some cycle , so there is bound to be some cycle n where the term : @xmath286 where @xmath287 again is equal to the @xmath278 @xmath279 term : so we then have : @xmath288 subtracting ( 7 ) from ( 8) we have the following requirement for consistency : @xmath289 there is no loss in generality if we assume , for the time being , the length of a cycle is a prime , is not a multiple of a prime say @xmath290 , b an integer , we can just define @xmath9 as @xmath291 , so the eq . following has the same form . ] so let us put @xmath292 , and @xmath293 and @xmath294 we then have from ( 9 ) : @xmath295 but we have assumed @xmath22 to be an arbitrary prime and equation ( 6.10 ) tells us that all primes @xmath22 are such that there exists two exponents @xmath296 such that they are multiples of a fixed prime q. at first sight this seems to be a rather remote possibility , but we will now prove that this will actually happen not for just some rare primes but actually for every prime p and any given prime q which we consider as fixed . this consequence is due to fermat s little theorem . there is no loss if we consider that p and q are prime to each ( other wise we can write q = gq,and the gcd g can be absorbed in n ) . fermat s little theorem states that under these conditions we can always choose @xmath297 then @xmath298 will always be divisible by q , this permits us to divide ( 6.10 ) out by q then choose @xmath9 such that the consistency test ( 6.10 ) is satisfied . so we have arrived at our following theorem . * theorem : * given a repeating sequence of type ( 6.4 ) with fixed periodicity , @xmath259 , then every tower of type @xmath49 , for every prime p , is contained in the sequence in the sense given by eq(6.8 ) . now we argue using the following facts : \(i ) the number of towers are infinite and \(ii ) the length @xmath259 is finite , further we know that \(iii ) the sequence(6.4 ) contains identical cycles each of which is actually repetitions of : @xmath299 , but we have already proved ( theorem 3)that \(iv ) the total number of @xmath300 which are equal to @xmath161 is exactly equal to the total number which are equal to@xmath119 . this means that \(v ) the number of + 1 s within @xmath301 is exactly equal to the number of -1s.,but we have proved that ( vi)each of these members can be generated by some or other tower of type @xmath49 , but \(vii ) these towers are infinite in number and the members belonging to a cycle are only @xmath259 ( finite ) in number . @xmath302 * final conclusion : therefore the only minimal conclusion * that we can come to is that given some fixed number @xmath303 belonging to a sequence of consecutive integers @xmath304 the chances of @xmath305 being equal to @xmath161 is exactly equal to the probability that @xmath305 is equal to @xmath119 . and this is true for all members @xmath303 in the cycle and since we have assumed that the whole sequence is cyclic , this statement is true for every integer @xmath9 . ( except of course , for the finite number which are less than @xmath306 - the beginning of the sequence . we ignore these because they are only finite in number ) . so we can then assume as we compute @xmath307 starting from @xmath308 up to @xmath19 , step by step , then the answer that we will get is like computing a sum of @xmath309 random variables defined as : @xmath310 where @xmath189 is a random variable and its @xmath311 value takes on the values @xmath161 or @xmath119 with equal probability . this is exactly the same as summing up to n , however since we are dealing with a cyclic sequences whose @xmath117 values have exactly @xmath312 @xmath161 s and @xmath119 s within one cycle . therefore if we sum up to @xmath313 , we would get@xmath314 as exactly zero ( ie if we stop summing @xmath19 exactly after @xmath315 cycles ) , but this will not happen often so we would usually find that after summing to a random @xmath19 a value which varies from 0 to @xmath312 , since every cycle has @xmath312 , @xmath161 s and @xmath312 , @xmath119 s , in random order within the length @xmath312 , therefore if you add consecutive values of say @xmath316 terms you will get a minimum value of 0 when all of them cancel and a maximum value of @xmath312 , if all the @xmath161 occur together in the first half of a cycle and also @xmath317 , @xmath100 : an integer . @xmath318 though we have gone about deriving the above in a rather long - winded manner , this final result is not surprising because it is what one would expect to get when we sum a cyclic sequence which has equal number of @xmath161 s and @xmath119 s terms in one cycle ( albeit randomly distributed within the length @xmath319 ) . but now if this value is substituted to littlewood s condition , and we take the limit @xmath320 we get zero . this means that we should be able to analytically continue @xmath321 left - words from @xmath322 to @xmath323 . but it is common knowledge ( see hardy)that there are very many zeros at @xmath324 and these will appear as poles in @xmath4 , hence we really can not continue @xmath4 to @xmath325 , this fact indicates that the period @xmath319 can not be a finite number . in other words it is the nature of the number system that a sequence of consecutive lambdas can never be cyclic with a finite sequence of length @xmath259 . so in this section we have concluded the only viable case to be examined is the case when the sequence of lambdas is noncyclic or `` random '' . this is done in the next section . we now again consider the series @xmath326 it has already been proved in this paper that , over the set of all positive integers , the respective probabilities that an integer @xmath9 has an odd or even number of prime factors are equal . that is , the probability of @xmath260 being + 1 or -1 are both 1/2 . so , @xmath327 where @xmath328 , can with equal probability , be either + 1 or -1 . if , in addition , it is assumed that the probabilities affecting the values of @xmath329 and @xmath330 , @xmath331 , are independent of each other , then we could deduce @xmath332 and then through an application of littlewood s argument , rh can be proved . we will show below that @xmath329 and @xmath330 are indeed _ independent _ random variables that are _ uncorrelated _ with each other as @xmath19 tends to infinity . + we have seen in a previous section that each tower makes it appearance , while computing @xmath18 , from @xmath333 the criteria was the tower with the base value @xmath334 makes its appearance when @xmath335 . now the summation in the figure is actually performed as : @xmath336 to appreciate the relevance of the independence of the @xmath329 s in eq . ( 7.14 ) , consider the series : @xmath337 where again @xmath9 is a positive integer , and the summation ( partial ) is being taken over a single tower and @xmath338 is + 1 or -1 depending on whether @xmath9 is even or odd . therefore @xmath339 it is evident that the probabilities that @xmath338 is + 1 or -1 are equal , as was the case with @xmath260 . however , here for any @xmath19 , @xmath340 can either be 0 or 1,(we assume for convenience the base value of the tower has @xmath341 else the oscillations will be between 0 and -1 ) and certainly does not scale as @xmath342 . this is because the @xmath343 s are perfectly correlated , and we are certain that if @xmath344 is 1 then @xmath345 and vice versa . here we see that the _ correlation interval _ , which we will define as the shortest distance along the ordered set of integers for which by knowing one value of a random variable we can predict ( or predict with better than even probability ) the value of another . to invalidate the equivalence of ( 7.15 ) , it matters little if the correlation interval is 1 or 100 , or @xmath346 , as any finite correlation interval would negate the independence of the random numbers making up the sequence in ( 7.17 ) or ( 7.16 ) + we have earlier demonstrated that all positive integers can be partitioned into subsets @xmath347 such that @xmath348 where the prime factorisation of @xmath349 with @xmath350 , is used to define @xmath351 and @xmath352 , where @xmath353 is the highest prime in the factorization with multiplicity greater than one . we have also shown that every consecutive pair in either of these types of subsets are twins " with opposite values , and thus are perfectly correlated by knowing the value of one we know the value of the other . + now consider the correlation interval of a pair of twins ( @xmath354 ) in an arbitrary subset @xmath347 . we see that this is @xmath355 now consider that every subset @xmath347 , except the trivial single - element one containing the integer 1 , is infinite and is ordered by @xmath100 which increases to infinity . as @xmath356 and @xmath357 , the correlation interval will increase without bound , as @xmath100 increases , so @xmath358 in _ every _ subset @xmath347 . as the sum in ( 7.16 ) is over all such subsets , it follows that the correlation interval for the entire set of @xmath260 s tends to infinity , when @xmath359 . this implies that the @xmath260 s are not correlated with each other and are independent random variables with equal probability of being + 1 or -1 , and so should sum to @xmath342 , ( see chandrasekhar ( 1943)),thus we have at last established that the exponent @xmath257.invoking littlewood s theorem ( sec.5 ) , we deduce that @xmath360 is analytic in the region @xmath361 which implies @xmath362 has no zeros in the same region . but riemann had shown by using symmetry arguments is the euler gamma function , then showed that this xi function has the symmetry property @xmath363 which in turn implied that that the zeros of @xmath362 ( if any ) which are not on the critical line will be symmetrically placed about the point s=1/2 , ie.if @xmath364 is a zero then @xmath365 ) is a zero see whittaker and watson page 269 . ] that if @xmath362 has no zeros in the latter region then it will have no zeros in the region @xmath366 , taking both these results together we are lead to the inevitable conclusion : that all the zeros can only lie on the critical line @xmath7 , thus proving the riemann hypothesis . in this section we give a reason as to why things worked so well and and we could actually prove what we have proved . we have already seen that as @xmath19 tends to infinity there will be an ever increasing number of rectangular waves , see figure 1 ( or eq.(5.23b)for details ) , participating in the summatory liouville function @xmath18 and each of these contributions make @xmath18 completely uncorrelated with l(n-1),l(n-2),etc and rh follows . in the following we will be arguing by using an analogy from information theory first discovered by shannon,(shannon ( 1948 ) ) we will call a tower which contributes @xmath161 s and @xmath119 s to @xmath18 as a broadcaster of bits and the contribution @xmath367 , increases or decreases the collection of bits @xmath18 collated by the listener according as @xmath161 arrives first or @xmath119 arrives first at @xmath368 . essentially , each tower is a broadcaster of rectangular waves of ever increasing periods . according to shannon every sequence is information . and the ` message ' in a sequence of bits can be said to contain information if there is at least some relationship between the present group of bits to the preceding group of bits ( like words in a sentence ) . we show in the argument following that the nature of the towers in ( 5.23 b ) destroy this information as @xmath19 tends to infinity . the situation is describable as follows : \(i ) it has been revealed that the contributions to @xmath18 , is coming from towers . if we keep @xmath19 fixed then the contribution @xmath367 to the previous summed value @xmath369 is exactly equal to @xmath370 , so @xmath371 . but we know that the integer @xmath19 is a member of a unique tower which we denote as the set @xmath49 and some member in this is exactly equal to @xmath19 , i.e. @xmath372 , in fact the triad @xmath49 and exponent @xmath373 are uniquely obtained by factorizing @xmath19 . we then have @xmath374 , which is completely determined . we had represented these contributions from a particular tower as rectangular waves in fig 1 . \(ii ) the contributions from the rectangular wave associated with a fixed @xmath49 change with the sign of the positive or negative peak arriving at @xmath19 and as @xmath373 increases from @xmath375 their duration exponentially increases thus delaying their arrival times and drastically increasing their correlation lengths . ( iii)but each tower contributes to different @xmath19 s but the distance along @xmath19 increases exponentially as @xmath373 increases i.e.the arrival times at the listener increases . \(iv ) the same situation as ( i),(ii ) and ( iii ) happens for every other time @xmath368 from 1 to infinity , while @xmath18 takes contributions from other towers . \(vi ) each tower @xmath49 comes into play as soon as @xmath368 reaches the base value of the tower i.e.@xmath376 after this @xmath19 , this particular tower broadcasts bits in ever increasing time intervals forever . \(vii ) as @xmath19 tends to infinity in - numerable towers come into play and each broadcasted bit received at time @xmath19 becomes completely uncorrelated to the bit received at time @xmath377 , so @xmath367 has no relation to @xmath378 essentially because @xmath367 and @xmath378 come from different towers . @xmath283 @xmath283 and according to shannon s information theory there will be no pattern in the bits and the entropy of the signal approaches infinity . at this point @xmath367 behaves like a coin toss and @xmath18 as the cumulative summation of @xmath19 coin tosses , ( see chandrasekhar ( 1943)),thus we have @xmath379 as @xmath19 tends to infinity proving rh . so we have seen that the rh is because of inherent randomness of factorization of @xmath19 into primes . in this paper we have investigated the analyticity of the dirichlet series of the liouville function by constructing a novel way to sum the series . the method consists in splitting the original series into an infinite sum over sub - series , each of which is convergent . it so turns out each sub - series is a rectangular function of unit amplitude but ever increasing periodicity and each along with its harmonics is associated with a prime number and all of them contribute to the summatory liouville function and to the zeta function . a number of arithmetical properties of numbers played a role in the proof of our main theorem , these were : the fact that each number can be uniquely factorized and then placed in an exclusive subset , where it and its other members form an increasing sequence and factorize alternately into odd and even factors ; and each subset can be labelled uniquely using a triad of integers which in their turn can be used to determine all the integers which belong to the subset , even fermat s little theorem played a role in our proof . in this paper we have also demonstrated that for every integer that has an even number of primes as factors ( multiplicity included ) , there is an integer that has an odd number of primes . this provides a proof for the long - suspected but unproved conjecture until now that the summatory liouville function and therefore the riemann hypothesis bears an analogy with the coin - tossing problem , denjoy had long suspected this as far back as 1931 . further , it has now been revealed that the randomness of the occurrence of prime numbers plays an important role in determining the analyticity of the zeta function , and in establishing the riemann hypothesis : the zeta function has zeros only on the critical line:@xmath12 . truth to tell even this connection of the role of the randomness of the primes to the rh problem was long suspected and even a book called the music of the primes " by marcus du sautoy , had appeared in 2003 ( harper collins ) , i could not help but recall the title of his book when i saw the rhythms of the harmonic functions , generated by prime numbers , that are depicted in fig 1 . @xmath283 @xmath380 i dedicate this paper to my teachers : mr john wright of bishop s school poona , prof . mookerjee of st . aloysius college jabalpur , prof . p.m.mathews of univerity of madras , mr . vishnu of bhel r & d hyderabad and to my first teachers - my parents . all of them lived selfless lives and nearly all are now long gone : may they live in evermore . @xmath283 @xmath283 * acknowledgments * : i thank my wife suhasini for her unwavering faith and encouragement . i thank my brothers mukesh and vinayak eswaran for carefully going over the manuscript and for their suggestions . i also thank the support of the management of snist , viz . p. narsimha reddy and dr . mahi , for their constant support , and my departmental colleagues and the hod dr . aruna varanasi for providing a very congenial environment for my research . @xmath283 1 . apostol , t.,(1998 ) _ introduction to analytic number theory _ , chapter 2 , pp . 37 - 38 , springer international , narosa publishers , new delhi . \2 . borwein , p. , choi , s . , rooney , b . , and weirathmueller , a . , ( 2006 ) _ the riemann hypothesis _ , springer . chandrasekhar s. , ( 1943)stochastic problems in physics and astronomy,rev.of modern phys . vol 15 , no 1 , pp1 - 87 \4 . denjoy , a. , ( 1931 ) l hypothese de riemann sur la distribution des zeros .. c.r . paris 192 , 656 - 658 \5 . edwards , h.m . , ( 1974 ) , _ riemann s zeta function _ , academic press , new york . eswaran , k. , ( 1990)`on the solution of dual integral equations occurring in// diffraction problems ' , proc . of roy . london , a 429 , pp . 399 - 427 \7 . hardy , g.h . , ( 1914 ) , ` sur les zeros de la fonction @xmath3 de riemann , ' _ comptes rendus de lacad . des sciences ( paris ) _ , 158 , 10121014 . knuth d.,(1968 ) ` art of computer programming ' , vol 2,chap 3 . ( 1968)addison wesley \9 . landau , e . , 1899 ph . d thesis , friedrich wilhelm univ . berlin \10 . littlewood , j.e . ( 1912 ) , quelques consequences de lhypoth`ese que la fonction @xmath3 de riemann na pas de zeros dans le demi - plan @xmath381 , ' _ comptes rendus de lacad . des sciences ( paris ) _ , 154 , pp . transl . version , in edwards op cit . riemann , b . , ( 1892)in ` gessammeltz werke ' , liepzig , 1892 , trans . ` on the number of primes less than a given magnitude ' available in edwards book , opcit pp299 - 305 \12 . sautoy , marcus du , ( 2003 ) music of the primes ( harper collins ) \13 . shannon , c.e . , ( 1948)`a mathematical theory of communication',bell systems technical journal , vol 27 , pp 379 - 423 , 623 - 656 , july , october 1948 . \14 . press , w.h , teukolsky , s.a . , vetterling , w.t . and flannery , b.p . , press etal ( 1986)`on numerical recipes in fortran',chap 7 , cambridge univ . press , 1986 \15 . whittaker , e.t . and watson , g.n . , ( 1989)`a course of modern analysis ' , universal book stall new delhi * i. scheme of partitioning numbers into sets * : let us say @xmath382 , then it will have at least one prime which has an exponent of 2 or above and among these there will a largest prime @xmath353 whose exponent is atleast 2 or above . such a prime will always exist for a class i number . then by definition the number to the right of @xmath38 is either 1 or is a product of primes with exponents only 1 . now multiply all the numbers to the left of @xmath38and call it @xmath45 i.e. @xmath383 and the product of numbers to the right of @xmath38as @xmath46 i.e. @xmath384now this triad of numbers @xmath385 will be used to label a set , note @xmath386 let us define the set @xmath387 : @xmath388 obviously @xmath386 which has @xmath389 belongs to the above set . also notice the factor involved in each number increases by a single factor of @xmath38therefore the @xmath117values of each member alternate in sign : @xmath390 in this paper all sets defined as @xmath51will have the property of alternating signs of @xmath117 eq . note in the above set containing only class i integers @xmath45 will have only prime factors which are each less than @xmath391 . let us consider the integer @xmath392 this is factorized as @xmath393 and since this is a class i integer , and @xmath394 because 7 is the highest prime factor whose exponent is greater than one . @xmath395 @xmath396 and @xmath397 and therefore @xmath398 is a member of the set @xmath399 @xmath400 ex 2 . now let us consider the simple integer : @xmath401 this is a class i integer and belongs to @xmath402 ex 3 . let us consider the integer @xmath80 this is factorized as : @xmath403 and is a class ii integer as there no exponents greater than 1 , and @xmath404 and since 17 is the highest prime number we put this in the set : @xmath405 note : if a tower has a class ii integer then it will appear as the first ( base ) member , all other numbers will be class i numbers . let the integer be @xmath408 this is factorized as@xmath409 since this is a class ii integer we see @xmath410 and the set which it belongs is @xmath411 * ii . theorems on representation of integers and their partitioning into sets . * let @xmath190 and @xmath413 be two integers which when factorized according to our convention as : @xmath414 and @xmath415 ( let us consider only class i integers @xmath416and @xmath417 are all @xmath418 ) if they are both equal to the same triad(say ) @xmath419 . then @xmath420 . consider the first two equalities @xmath421 this means @xmath22 is the largest prime with @xmath73 , on the l.h.s . similarly @xmath40 is the largest prime with exponent @xmath422 on the r.h.s . now if @xmath423 this means @xmath424 must divide @xmath425 , but this can not happen since @xmath425 can not contain a prime greater than @xmath40 with an exponent @xmath73 , now if @xmath426 then @xmath427 must divide @xmath428 but this again can not happen since @xmath46 can not contain an exponent @xmath422.so we see @xmath429,and @xmath430.but once again unique factorization would imply since @xmath46 contains all prime factors above @xmath22 and @xmath425 must contain only prime factors above @xmath431 the only possibility is @xmath432 but this also makes @xmath433that is the triad of @xmath190 is @xmath412 . similarly equating the second and third equalities @xmath434 and using similar arguments we see @xmath435,and @xmath436 that is @xmath437 . the same logic can be used to prove the theorem for class ii integers when @xmath438 qed if there are two triads @xmath439 and @xmath440 and represent the same integer say @xmath441 which can be factorized as @xmath442 . where the factorization is done as per our rules then we must have @xmath443 by using exactly similar arguments as above(in theorem a ) we conclude that we must have @xmath444 and @xmath432 ; similarly imposing the condition on the second triad @xmath445 , we conclude @xmath446 and @xmath447 ; thus obtaining @xmath448 this means the two triads are actually identical.qed

we study the effects of finite proton mass on the energy levels of hydrogen atoms moving transverse to a superstrong magnetic field with generalized pseudomomentum . field strengths of order gauss are typically found on the surfaces of neutron stars , but we also study the regime gauss , where the landau excitation energy of the proton is large . we adopt two different approaches to the two - body problem in strong magnetic field , and obtain an approximate , but complete solution of the atomic energy as a function of and . we show that , for , there is an orthogonal set of bound states , which do not have any landau excitation contribution in their energies . the states with very large have small binding energy and small transverse velocity , but are nevertheless distinct from the fully ionized states . the final results for the excitation energies are given in the form of analytical fitting formulae . the generalized saha equation for the ionization - recombination equilibrium of hydrogen gas in the presence of a superstrong magnetic field is then derived . although the maximum transverse velocity of a bound atom decreases as increases , the statistical weight due to transverse motion is actually increased by the strong magnetic field . for astrophysically interesting case of relatively low density and temperature , we obtain analytic approximations for the partition functions . the highly excited bound states have a smaller statistical weight than the fully ionized component .

astro-ph9510123

it is well known that a superstrong magnetic field of order @xmath5 gauss , typically found on the surfaces of neutron stars , can dramatically change the structure of neutral atoms and other bound states ( see , e.g. , ref.@xcite for an early review and ref.@xcite for a recent text on atoms in strong magnetic fields ) . the atomic unit @xmath6 for the magnetic field strength and a dimensionless parameter @xmath7 are b_o = m_e^2e^3 c^3=2.35 10 ^ 9 g ; b. when @xmath8 , the cyclotron energy of the electron @xmath9 , where @xmath10 is the magnetic field strength in units of @xmath11 , is much larger than the typical coulomb energy , thus the coulomb forces act as a perturbation to the magnetic forces on the electrons , and at most temperatures the electrons settle into the ground landau level . because of the extreme confinement of electrons in the transverse direction , the coulomb force becomes much more effective for binding electrons in the parallel direction . the atom has a cigar - like structure . moreover , it is possible for these elongated atoms to form molecular chains by covalent bonding along the field direction . hydrogen atoms in strong magnetic field have been studied extensively@xcite . we have also recently completed a study of the electronic structure of hydrogen molecules and chains in the strong field regime ( @xmath8)@xcite . however , in many of these studies , the center - of - mass ( cm ) effects of the proton motion have been neglected . in the case of the hydrogen atom , although significant effort has been devoted to calculating the energy levels of an electron in the static coulomb potential of a _ fixed _ proton ( infinite mass ) to a high precision and for arbitrary magnetic field strength@xcite , the two - body problem in strong magnetic field , including the effects of finite proton mass , has been studied in detail only recently ( e.g.,@xcite ) . in this paper we focus on the hydrogen atom , but discuss those aspects which are important for application to molecules in very strong fields ( we shall study molecular excitation levels in a later paper@xcite ) . a free electron confined to the ground landau level , the usual case for @xmath8 , does not move perpendicular to the magnetic field . such motion is necessarily accompanied by landau excitations . when the electron ( still in the landau ground state ) combines with a proton , the mobility of the neutral atom across the field depends on the ratio of the atomic excitation energy and the landau excitation energy @xmath12 for the proton . as the typical coulomb excitation is @xmath13 in atomic units , it is convenient to define a critical field strength @xmath14 via b_critb_crit=1.8010 ^ 4 ; b_crit = b_critb_o=4.2310 ^ 13 g. for @xmath15 ( terrestrial conditions ) the landau energies of both electron and proton are minor perturbations and one can construct wave packets that mimic the classical motion of a neutral atom across the field . in this case the internal structure of the atom or molecule is coupled to the center - of - mass motion only through a `` motional stark effect '' ( see sec . ii ) . in the opposite limit of @xmath16 , the landau energies are larger than the coulomb excitation energy , and both electron and proton ( in the atom ) are in the landau ground state at reasonable temperatures . in this case , quantum mechanics can not mimic classical motion . the canonical field strengths of radio pulsars , with @xmath0 slightly less than @xmath14 , present an intermediate case . however , at all field strengths one can introduce a pseudomomentum @xmath17 perpendicular to the field and , in principle , find the bound - state energy @xmath18 as a function of @xmath17 . one question of interest is the range of @xmath1 for which @xmath18 increases linearly with @xmath19 ( as it does for ordinary kinetic energy ) ; but in any case @xmath20 is needed to evaluate the saha equation for the equilibrium between neutral and ionized hydrogen ( sec . previous treatments@xcite of the ionization equilibrium in strong magnetic fields did not properly take account of the non - trivial effects of atomic motion . states with large @xmath1 , where velocity decreases with increasing @xmath1 ( see ref.@xcite and our eq . [ 3.29 ] ) are of particular interest , especially for @xmath0 larger than @xmath14 . the separation of the center - of - mass ( cm ) motion of a bound state in the presence of a magnetic field has been studied previously based on the conserved pseudomomentum ( e.g.,@xcite ) . we briefly review and clarify this pseudomomentum scheme in sec . ii . when @xmath21 , perturbation calculations for hydrogen atom motion ( e.g.,@xcite ) are valid over a wide range of @xmath1 values and lead to interesting phenomena such as bent trajectories@xcite . model atmospheres can be affected by details of the transverse motion ( e.g.,@xcite ) . some accurate numerical calculations for general @xmath1 and for several values of @xmath0 ( but all below @xmath14 ) are now available @xcite , but we concentrate on the @xmath22 regime in sec . a different approach to the two - body problem ( for positronium atom ) in the superstrong field regime @xmath8 has been developed in@xcite . our purpose in this paper is not to obtain accurate energy spectra of a moving hydrogen atom in certain limited regimes , as have been done in most of the papers mentioned above ; rather , we seek complete ( though approximate ) solutions of the two - body problem for a wide range of parameter space , including @xmath0 just below @xmath14 ( a common case for neutron stars ) , but especially for @xmath23 ( in case @xmath24 g exists in some neutron stars , as has been suggested recently@xcite ) . our emphasis is on finding physically meaningful approximate fitting formulae for the atomic energy of the moving atom over all relevant values of @xmath1 ( sec . iii ) , in order to determine the equilibrium between neutral and ionized hydrogen ( sec . however , in practice we shall be interested mainly in the regimes where the thermal energy @xmath25 is much less than the ground - state binding energy of the atom , while the gas density is much smaller than the internal density of the atom , so that the neutral and ionized fractions are of the same order of magnitude . in such cases , we are most interested in two kinds of excited bound states : ( i ) those with excitation energies up to a few atomic units ( comparable to @xmath25 but a small fraction of the binding energy ) and ( ii ) states that are only barely bound ( e.g. , those with extremely large @xmath1 ) , for which one has to check whether phase - space factors make them unimportant relative to ionized hydrogen . for applications to molecules@xcite and multi - electron atoms@xcite with @xmath4 , a controversy arises regarding the `` coupling '' of the electron s orbital quantum number with the landau level of the proton ( or nuclei ) . this is discussed in sec . our calculations in this paper are based on non - relativistic quantum mechanics . for @xmath26 g ( note that @xmath27 is close to @xmath14 only by coincidence ) , i.e. , @xmath28 , the transverse motion of the electron becomes relativistic . however , the relativistic correction to the atomic binding energy is small as long as the electron remains non - relativistic along field direction@xcite . except as otherwise noted , we shall use atomic units ( a.u . ) throughout the paper , in which mass and length are expressed in units of the electron mass @xmath29 and the bohr radius @xmath30 cm , energy in units of @xmath31 ev ; field strength is in units of @xmath6 ( eq . [ 1.1 ] ) and temperature in units of @xmath32 k. to set the scene , we briefly review the pseudomomentum approach to the two - body problem of a hydrogen atom in a strong magnetic field@xcite . however , the physical meaning of the pseudomomentum of the atom needs some clarification . for a free particle of charge @xmath33 and mass @xmath34 in a constant magnetic field ( assumed to be aligned in the @xmath35-axis ) , there are three momentum - like vectors : the _ canonical momentum _ @xmath36 ; the _ mechanical momentum _ @xmath37 where @xmath38 is the vector potential and @xmath39 is the velocity ; and the _ pseudomentum _ ( or the generalized momentum ) , as defined by = + e_i*b*. that @xmath40 is a constant of motion can be easily seen from the classical equation of motion for the particle @xmath41 . the parallel component @xmath42 is simply the linear momentum , while the constancy of the perpendicular component @xmath43 is the result of the fact that the guiding center of the circular orbit of the particle does not change with time . the position vector @xmath44 of this guiding center is related to @xmath43 by _ c=_e_ib^2 . mathematically , the conservation of @xmath40 is the result of the invariance of the hamiltonian under a spatial translation plus a gauge transformation @xcite . the existence of the integration constant @xmath43 or @xmath45 implies infinite degeneracy of a given landau energy level . we can use @xmath40 to classify the eigenstates . however , since two components of @xmath43 do not commute , @xmath46=-ie_i b$ ] , only one function of @xmath47 can be diagonalized for stationary states . this means that the guiding center of the particle can not be specified accurately . if we use @xmath48 to classify the states , then the wavefunction has the well - known form @xmath49 @xcite , where the function @xmath50 is centered at @xmath51 , as can be inferred from eq . the landau degeneracy in an area @xmath52 is thus given by @xmath53 , where we have used @xmath54 . on the other hand , if we choose to diagonalize @xmath55 , we obtain the landau wavefunction @xmath56 in cylindrical coordinates@xcite , where @xmath57 is the `` orbital '' quantum number ( denoted by @xmath58 in some references ) . for the ground landau level this is w_0 m ( _ ) w_m(,)=1(2m!)^1/2 ( ) ^m ( ^24 ^ 2)(-im ) . the distance of the guiding center of the particle from the origin is given by _ m=(2m+1)^1/2 , m=0,1,2 , , where @xmath59 is the cyclotron radius = ( ceb)^1/2= a_o(b_ob)^1/2=b^-1/2 ( a.u . ) = 2.5710 ^ -10b_12 ^ -1/2 ( cm ) . the corresponding value of @xmath60 is given by @xmath61 . note that @xmath19 assumes discrete values since @xmath57 is required to be an integer in order for the wavefunction to be single - valued . the degeneracy @xmath62 of the landau level in an area @xmath63 is then determined by @xmath64 , which again yields @xmath65 . we also note that @xmath66 is related to the @xmath35-angular momentum @xmath67 , as is evident from the @xmath68 factor in the cylindrical wavefunction ( eq . [ 2.3 ] ) . in general , we can show that j_z = xp_y - yp_x=12e_ib(_^2-_^2 ) = ( m - n ) |e_i|e_i , where we have used @xmath69 , and @xmath70 is the quantum number for the landau excitations . we now consider the electron - proton system . it is easy to show that even with the coulomb interaction between the particles , the total pseudomomentum is a constant of motion = _ 1+_2 , where the subscripts @xmath71 refer to electron ( charge @xmath72 ) and proton . moreover , unlike the single particle case , here all components of @xmath40 commute . thus it is natural to separate the cm motion from the internal degree of freedom , using the vector pseudomomentum @xmath40 as an explicit constant of motion ( although we shall discuss an alternative set of basic states in sec . iv ) . from eq . ( 2.2 ) we have _ k=_c1-_c2=-b^2 . thus we see that @xmath43 _ is directly related to the separation @xmath73 between the guiding center of the landau orbit of the electron and that of the proton_. consider the energy eigenstate with a fixed @xmath40 . introduce the center - of - mass coordinate @xmath74 and the relative coordinate @xmath75 . writing the two - body wavefunction as ( , ) =( ) , so that @xmath76 has a well - defined value of @xmath40 , while @xmath77 $ ] is a gauge factor , the schrdinger equation reduces to h()=(h_o+h)()= ( ) , with @xmath78 where @xmath79 , @xmath80 ( a.u . ) , @xmath81 . we will make no distinction between @xmath82 and @xmath83 , nor @xmath84 and @xmath29 in our following calculations . ( 2.10)-(2.12 ) have already been derived in refs.@xcite . clearly , the cm motion is coupled to the internal motion through the second term in @xmath85 , which has the form of the potential in a motion - induced electric field @xmath86 . this term represents the so called `` motional stark effect '' ( although such a description is not exactly accurate , since @xmath87 does not correspond to the cm velocity@xcite ) . for small @xmath1 , this effect can be considered by treating @xmath85 as a perturbation ( sec . iii ) , but the eigenstates of @xmath88 can in principle be used as a set of basic states for developing eigenstates of the full hamiltonian for any value of @xmath1 . however , the following transformed version of the hamiltonian is more convenient for large @xmath1 . motivated by the fact that @xmath43 measures the separation of the guiding centers of the electron and the proton , we can remove the `` stark term '' by introducing a displaced coordinate @xmath89 , where @xmath73 is given by eq . ( 2.8 ) . after a gauge transformation , with ( ) ( im_p - m_e2m_)( ) , the hamiltonian becomes h = k_z^22m+12(+12)^2 -1m_p()-1|+_k| , where @xmath90 . this expression has been obtained in refs.@xcite . we shall see in sec . iii that this alternative form of the hamiltonian is useful in the regime where @xmath7 is much larger than @xmath91 , defined in eq . we consider first the hamiltonian formulation in terms of eqs . ( 2.11)-(2.12 ) . for the zeroth order hamiltonian @xmath88 the quantum numbers for the basic states are @xmath42 , the number of nodes in the @xmath35-wavefunction @xmath92 , the electron landau level integer @xmath70 and the `` orbital '' quantum number @xmath93 . in this paper we only consider @xmath8 and thus restrict ourselves to @xmath94 . the energy eigenvalues of @xmath88 for the @xmath95 states can be written as _ m^(0)(k_z)=k_z^22m+e_m+mbm , where @xmath96 is the energy of a bound electron in the fixed coulomb potential of an infinitely massive positive charge . the last term in eq . ( 3.1 ) for @xmath97 represents landau energy excitations for the proton , but @xmath57 is merely the `` orbital '' quantum number for the electron wavefunction and measures the relative @xmath98angular momentum @xmath99 . thus , there is no separate quantum number for the proton in this formulation . the `` coupling '' between the electron quantum number @xmath57 and the proton landau excitation @xmath100 in eq . ( 3.1 ) results from the conservation of total pseudomomentum . the term @xmath101 has the form ( e.g.,@xcite ) e_m-0.16 al_m^2 , l_m , where @xmath102 is a coefficient which varies slowly with @xmath7 and @xmath57 ( e.g. , @xmath103 for @xmath104 when @xmath105 , and @xmath106 for @xmath104 when @xmath107 ) . in most formulae below , we replace @xmath102 by unity ; the numerical values can be found in@xcite . the atom has a cigar - like shape , with size @xmath108 ( cf . [ 2.4 ] ) perpendicular to the field and @xmath109 along the field direction . note that the correction due to the reduced mass @xmath84 could be easily incorporated by a simple scaling : e_m-0.16am_e^2 . however , this is a small correction ( of order @xmath110 ) , and will be neglected hereafter . equations ( 3.2)-(3.3 ) refer to the `` tight - bound '' states for which the number of nodes @xmath92 of the @xmath35-wavefunction @xmath111 of the electron is zero . for @xmath112 , the energy eigenvalues are approximately given by@xcite : e_m=-12_1 ^ 2(1 - 4_m_1 a_o ) , _ 1=1,2,3,for the odd states ( @xmath113 ) , and e_m=-12_1 ^ 2 , _ 1=1,2,3,for the even states ( @xmath114 ) . the sizes of the wavefunctions are @xmath115 perpendicular to the field and @xmath116 ( a.u . ) along the field . these states have much lower binding energies compared to the `` tight - bound '' states . we now consider the energies and eigenstates of the atom for finite @xmath1 . the two different hamiltonian forms are discussed in sec . iii.b - c , and the general approximate expressions for the energies are then given in sec . we focus on the `` tight - bound '' states only , since finite @xmath1 will make the weakly - bound @xmath112 states even less bound ( although in sec . v we will include an estimate of the statistical weight of these states in the partition function of bound atom ) . for sufficiently small @xmath1 , we can use standard perturbation theory to calculate the correction of energy @xmath18 due to @xmath85 given by eq . ( 2.12 ) ( see also@xcite ) . let @xmath17 be along the @xmath117-axis , then the @xmath118-dependent part of @xmath85 is @xmath119 . we consider only @xmath94 and @xmath95 , so the exact eigenstates of @xmath120 are superpositions of the @xmath88-eigenstates with @xmath93 . the only non - zero matrix elements of @xmath121 are of the form w_m|x|w_m+1=(m+12)^1/2 , and the energy differences of adjacent @xmath88-eigenstates are approximately given by _ m^(0)=_m+1^(0)-_m^(0 ) 0.32l(2m+32m+1)+bm , where @xmath122 and the factor @xmath123 is an approximation to a slowly varying function of @xmath57 and @xmath7 . we first consider the ground state @xmath124 . using @xmath125 and eq . ( 3.6 ) we note that perturbation theory is justified if @xmath1 is much smaller than a `` perturbation limit '' @xmath126 defined as k_p = m_0^(0)b^1/2 = b^1/2(1+mlb ) b^1/2(1+b_critb ) , where @xmath127 is a slowly varying function of @xmath7 ( e.g. , @xmath128 for @xmath129 ) . for @xmath130 , the energy @xmath131 to be added to @xmath132 in eq . ( 3.1 ) is given by second order perturbation theory ( plus a diagonal term ) as _ 0^(2)=k_^22m(1-bm_0^(0 ) ) = k_^22 m _ , where @xmath133 is the effective mass for the `` transverse motion '' of the atom , m_= m(1+blm ) m(1+bb_crit ) . thus the effective mass @xmath133 increases with increasing @xmath7 . similar calculations for the @xmath134 states yield @xmath135 , with 1-mm_m . a convenient ( but approximate ) expression for the effective mass @xmath136 is given by m_mm+_m ( 2m+1)bl_m , where @xmath137 is of the same order of magnitude as @xmath138 , but different ( by a factor of a few ) for different @xmath57-states . the important feature in eqs . ( 3.11)-(3.12 ) is that the effective mass is larger for the higher-@xmath57 state . the quadratic form of the effective `` transverse kinetic energy '' in eq . ( 3.9 ) is valid only when it is much less than the `` perturbation limit '' , reached when @xmath139 . using eq . ( 3.8 ) and the approximation in eq . ( 3.10 ) , this kinetic energy limit becomes = l2(1+mlb ) 1.7(1+b_critb ) ( a.u . ) . for @xmath140 ( even if @xmath8 ) this limit is large compared with @xmath141 a.u . , so that the quadratic perturbation energy ( or transverse kinetic energy ) in eq . ( 3.9 ) is valid for the most important ( low energy ) states . moreover , from eq . ( 3.10 ) the effective mass is close to the actually proton mass @xmath82 . for superstrong fields , @xmath142 , on the other hand , the effective mass ( for @xmath143 ) is much larger than @xmath82 and the perturbation formalism already breaks down when the transverse kinetic energy is only of order @xmath141 a.u . at least for @xmath4 , we have to consider values of @xmath1 large enough so that the perturbation treatment for the formalism in eqs . ( 2.10)-(2.12 ) is unsuitable . for any magnetic field strength there are still eigenstates for arbitrarily large values of @xmath1 , but the transformed hamiltonian in eqs . ( 2.13)-(2.14 ) is now more suitable for calculating the energy . as mentioned , eq . ( 2.14 ) gives an alternative formulation for the hamiltonian where @xmath144 does not appear explicitly , but the displacement of the electron - proton guiding centers does with @xmath145 ( in atomic units)@xcite . we again focus only on the landau ground state for the electron ( @xmath94 ) , but in principle we must include all @xmath57-values ( with proton landau excitation energy @xmath100 ) and mixing between these states . we consider first the approximation where we omit mixing , i.e. , we use the diagonal matrix element of @xmath146 for a fixed @xmath57-value , and restrict ourselves to @xmath95 , @xmath147 and @xmath148 , so that @xmath149 . we can estimate the size @xmath150 of the atom along the @xmath35-axis and the energy @xmath151 for two different regimes of the values of @xmath152 . \(i ) for @xmath153 ( but not necessarily @xmath154 ) and with @xmath150 as a variational parameter we have _ m~1l_z^2 - 1l_z+mbm , where the first term is the kinetic energy along the @xmath35-axis , and the second term is the potential energy of the electron . the logarithmic factor in eq . ( 3.14 ) comes from an integration over the `` cigar - shaped '' electron cloud in the displaced coulomb potential @xmath155 . minimizing @xmath151 with respect to @xmath150 , we obtain l_z~()^-1 , _ m ~-()^2+mbm . the mixing between different @xmath57-states is unimportant when @xmath156 . this can be seen from the order of magnitude estimate of the off - diagonal matrix element between @xmath124 and @xmath157 states : 0|1|+_k|| 1~0|_kr^3|1 ~r_kl_z(^2+r_k^2 ) l , as compared with @xmath158 . for @xmath159 the mixing is non - negligible , especially when @xmath160 ; some results are given in ref.@xcite . when @xmath142 , the mixing can be neglected for all @xmath152 . ( ii ) for @xmath161 , the coulomb logarithm in eq . ( 3.14 ) disappears , and we have ( for @xmath124 ) : _ 0~1l_z^2 - 1(l_z^2+r_k^2)^1/2 . in the limit of @xmath162 , minimization of @xmath163 with respect to @xmath150 yields l_z~r_k^3/4<<r_k , _ 0~-1r_k =- bk_. thus for @xmath164 , the atom is very weakly bound ( @xmath165 ) . the limiting scaling relations in eq . ( 3.18 ) have been identified in@xcite . we can calculate the energy eigenvalue more accurately . substituting @xmath166 into @xmath167 with @xmath168 given by eq . ( 2.14 ) , and averaging over the transverse direction , we obtain a one - dimensional schrdinger equation -12d^2dz^2f_m(z)+v_m(z , r_k)f_m(z ) = _ mf_m(z ) , where the averaged potential is given by v_m(z , r_k)=-w_m(_)|1|+_k| |w_m( _ ) . the function @xmath169 can be evaluated using an integral representation ( e.g.,@xcite ) v_m(z , r_k)=-1_0^dq ( -q^22-q|z|)j_0(qr_k/ ) l_m(q^22 ) , where @xmath170 is the bessel function of zeroth order and @xmath171 is the laguerre polynormial of order @xmath57@xcite . we solve for @xmath172 by integrating eq . ( 3.19 ) numerically from @xmath173 to @xmath174 subject to the boundary conditions @xmath175 at @xmath174 and @xmath176 as @xmath177 . the energy eigenvalue @xmath151 as a function of @xmath152 are shown in figure 1 for @xmath107 and @xmath178 . similar numerical results have also been obtained in ref.@xcite . for @xmath179 , we can fit the energy to a form similar to eq . the total energy of the @xmath124 state is then given by _ 0(k_z , k _ ) -0.16()^2 , and the atomic size @xmath180 . equation ( 3.22 ) reduces to eq . ( 3.2 ) for @xmath181 . from the numerical results we find @xmath182 . for @xmath161 , the binding energy of the atom is much smaller , and eq . ( 3.22 ) should be replaced by the order of magnitude relation ( 3.18 ) , while the actual numerical values of the energies are not important in practice ( see sec . v ) . in the @xmath4 regime , the maximum value @xmath126 for the perturbation treatment of sec . iii.b to be valid becomes @xmath183 ( cf . [ 3.8 ] ) . thus for @xmath184 , eq . ( 3.22 ) should be consistent with eq . ( 3.9 ) . indeed , when @xmath185 , eq . ( 3.22 ) reduces to _ 0(k_z , k _ ) -0.16l^2 + 0.32clbk_^2 . the dependence on @xmath1 is again quadratic , and the corresponding effective transverse mass is @xmath186 , in agreement with eq . ( 3.10 ) . the @xmath134 states can be similarly calculated using eqs . ( 3.19)-(3.21 ) . some numerical results are again shown in figure 1 . the energy can be expressed approximately as _ m(k_z , k_)+ mbm-0.16()^2 . from the numerical results we again obtain @xmath187 . comparison with the numerical results of potekhin@xcite , who included the mixing of different @xmath188states , indicates that eqs . ( 3.22 ) and ( 3.24 ) are accurate to within @xmath189 in the relevant regime of @xmath1 ( @xmath190 ) when @xmath191 . the agreement becomes better as @xmath7 increases . for smaller @xmath7 , however , the perturbative results of sec . 3.b should be adequate ( see sec . 3.d ) . finally , if we consider sufficiently strong magnetic field so that not only @xmath192 ( or @xmath191 ) but also @xmath193 is satisfied , then eq . ( 3.24 ) implies that all the @xmath134 states are unbound , as has already been noted in ref.@xcite . however , this does _ not _ mean that there is no other bound state except a single non - degenerate @xmath124 state . indeed , eq . ( 3.22 ) indicates that there are many states for which the guiding centers of proton and electron are separated by a small @xmath152 , and these states have similar energies compared to the ground state ( @xmath194 ) . in the pseudomomentum scheme discussed here , these states occupy a continuum @xmath1-space . as we shall see in sec . iv , these closely - packed energy levels can be made discrete if we use a different set of eigenstates . consider for the moment the cases with @xmath195 , i.e. , @xmath196 , but with no other restrictions on @xmath1 or @xmath7 . in sec . iii.b - c , we have obtained reasonably accurate ground - state ( @xmath124 ) energy of a hydrogen atom in two limiting regimes : ( i ) for @xmath21 , where eqs . ( 3.9)-(3.10 ) are applicable up to adequately large values of @xmath1 , and ( ii ) for @xmath4 , where the energy is given by eq . we write the total ground state energy in the form _ 0(k_z , k_)-0.16l^2+e_(k _ ) , and want to find a general fitting formula for the `` transverse kinetic energy '' @xmath197 with @xmath195 , but @xmath198 otherwise arbitrary . with the inequality @xmath196 , the second term in eq . ( 3.22 ) can be approximated by @xmath199 . we propose the following fitting formula e_(k_)(1+k_^2 ) , 0.64k_p^2 , where @xmath133 is given by eq . recall that the parameter @xmath128 for @xmath129 , and a typical number to use is @xmath200 . this formula reduces to eq . ( 3.9 ) when @xmath143 and to the above approximation of eq . ( 3.22 ) when @xmath4 . we expect eq . ( 3.26 ) to be accurate to within @xmath201 for @xmath195 . in the regime @xmath142 and @xmath202 ( i.e. , @xmath203 ) , eq . ( 3.26 ) reduces to @xmath204(a.u . ) , which is large compared with the thermal energy ( temperature ) @xmath205 for the astrophysical applications of interest . for evaluating the integral over @xmath1 extending from zero to infinity in the atomic partition function ( sec . v ) , we shall advocate using eq . ( 3.26 ) for all @xmath1 even though this expression tends to infinity,^2,\ ] ] with @xmath206 . this expression can be applied to @xmath207 and gives the correct limiting result for @xmath208 , but it is not convenient to use in practice . for the applications discussed in sec . v , eqs . ( 3.25)-(3.26 ) are adequate . ] whereas the correct expression should approach the finite limit @xmath209 for @xmath208 . the difference is appreciable only where @xmath210 ( so that the boltzman factor @xmath211 is very small ) and our prescription amounts to `` cutting off '' the integral , i.e. , omitting the states with @xmath162 from the integral . this omission is advantageous , since these states should be treated together with the ionized states , and both turn out to be unimportant ( sec . v ) . our fitting formulae for the energies of the @xmath134 states are less accurate . in the small-@xmath1 limit , the @xmath1-dependent term in eq . ( 3.24 ) reduces to the quadratic form @xmath212 , with the effective mass given by eq . similar to eqs . ( 3.25)-(3.26 ) we fit @xmath151 to the analytical form : _ m(k_z , k_)+mbm -0.16l_m^2+e_m(k _ ) , where e_m(k _ ) ( 1+k_^2_m ) , _ m0.64_m ( 2m+1)b^2 , so that @xmath213 , @xmath214 . although @xmath137 can vary by a factor of a few for different values of @xmath0 and @xmath57 , equation ( 3.28 ) has the correct approximate functional dependence for a wide range of @xmath0 and @xmath1 . as noted in the introduction , one can not construct a wave packet to mimic the classical behavior of a moving atom when both electron and proton are confined to the ground landau level . therefore when @xmath16 , the notion of `` motion across the magnetic field '' does not have a unique meaning . nevertheless , one can ask two questions about the transverse pseudomomentum : ( i ) is there an upper limit to @xmath1 beyond which there is no bound state ? ( ii ) when a bound state exists , what is the value of @xmath215 , the analog to the classical center - of - mass transverse velocity of the atom ? from our discussion in this section , we have seen that there exist bound states for all values of @xmath0 and @xmath1 , although the states with large @xmath1 are very weakly bound . from eq . ( 3.26 ) , we have @xmath216 ( note that for @xmath208 , eq . [ 3.18 ] should be used and we have @xmath217 instead ) . clearly , in general @xmath218 is smaller than its field - free counterpart : @xmath219 ( so that the effective mass description is valid ) only for @xmath220 , and @xmath221 ( classical behavior ) only for @xmath140 . as @xmath222 , the center - of - mass velocity approaches zero . for a given magnetic field strength , the maximum @xmath223 is given by @xmath224 which occurs when @xmath225 . for @xmath191 , the states with @xmath226 have not only small velocity but also large electron - proton separation @xmath227 . nevertheless , these states are quite distinct from an electron - proton pair with separation @xmath152 in fully ionized hydrogen , because the relative @xmath35-coordinate satisfies a bound - state wavefunction ( localized , although with large scale - length and small binding energy ) . the partition function for these states is smaller than that for the ionized component , because the sum over the relative momentum in the @xmath35-direction , @xmath228 , is absent . the basic states used in sec . iii for the e - p two - body problem are explicit eigenstates of the transverse pseudomomentum @xmath17 . for @xmath181 , this formulation has the advantages that the electron s `` orbital '' number @xmath57 is a good quantum number , the wavefunction can be related to that for a fixed positive charge and the energy is given explicitly by eq . the @xmath181 states with different values of @xmath57 are orthogonal to each other and could be used to satisfy the pauli principle for the electrons in a hydrogen molecule@xcite or a multi - electron atom@xcite . the last term @xmath229 in eq . ( 3.1 ) is unimportant when @xmath140 , but it is very large when @xmath142 ( and @xmath97 ) . the simplest wavefunction for the h@xmath230-ground state in this formulation would use one @xmath124 and one @xmath157 electron@xcite , and the energy would include the positive term @xmath231 . one therefore might conclude that h@xmath230 has a positive energy relative to two h atoms when @xmath142 . in a future paper@xcite we will show that this is not the case . in a molecule , only the total pseudomomentum is conserved , not that for individual atoms . thus for @xmath142 one would have to perform an integral over @xmath1 of individual atoms to form a molecular eigenstate ( similar situation exists for multi - electron atoms ) . to our knowledge this complicated task has not yet been carried out . we propose here an alternative set of basic states which are not eigenstates of the pseudomomentum . this formulation is more complex for single h atom , but can be generalized more easily to molecule . moreover , this formulation makes the definition of eigenstates clearer ( unlike the pseudomomentum approach , where @xmath57 is a good quantum only for @xmath181 ) , and therefore offers an intuitive understanding of the degeneracy of states . let @xmath232 , @xmath233 be the position vectors of electron and proton in a h atom . we introduce coordinates @xmath234 and @xmath235 , but stick to @xmath236 and @xmath237 . in this mixed coordinate system , the hamiltonian for the electron - proton pair becomes @xmath238 , with ( cf . footnote 1 ) : h_o = k_z^22m+p_z^22+_i = e , p12m_i ( _ i - e_i_i)^2 , where @xmath239 , and the interaction potential is v=-1 ^ 1/2 . we set @xmath147 without loss of generalality . consider the transverse part of the wavefunction for a bound state . the landau wavefunctions of the electron and proton form a complete set w_n_1m_1(_1)w_n_2m_2^*(_2 ) , n_1,n_2,m_1,m_2=0,1 , 2,(@xmath240 specify the landau excitations , @xmath241 are the `` orbital '' quantum numbers as in eqs . [ 2.3]-[2.4 ] ) . in general , an eigenstate of @xmath168 can be constructed as ( z,_1,_2)= _ n_1,m_1,n_2,m_2w_n_1m_1(_1)w_n_2m_2^*(_2 ) f_n_1m_1n_2m_2(z ) , where we have restricted ourselves to @xmath35-wavefunctions without a node ( @xmath95 ) . substituting eq . ( 4.4 ) into the schrdinger equation @xmath242 and using the orthogonal relations for the functions @xmath243 to average over the transverse direction , we obtain a set of coupled differential equations from which the functions @xmath111 s and the eigenvalue @xmath18 can be calculated , at least in principle . this set of equations is greatly simplified as a result of the conservation of total @xmath35-angular momentum @xmath67 . form eq . ( 2.6 ) , we have j_z=-(m_1-n_1)+(m_2-n_2 ) . indeed , since the basis function @xmath244 , while the interaction potential @xmath245 depend only on @xmath246 , we readily see that only the states with the same @xmath67 are coupled . we shall use the formalism of this section only when @xmath4 , in which case the landau energy of the proton is large compared to the atomic coulomb energy , so that both @xmath247 and @xmath248 become good quantum numbers . for astrophysical applications we are then interested in the ground landau levels , @xmath249 . the energy eigenstate with a fixed @xmath35-angular momentum @xmath250 can be constructed as ^(m_t)(z,_1,_2)= _ m_1w_m_1(_1)w_m_t+m_1^*(_2 ) f^(m_t)_m_1(z ) . the equations for the functions @xmath251 s are then given by -12d^2dz^2f^(m_t)_m_1(z ) -_m_1g^(m_t)_m_1m_1(z ) f^(m_t)_m_1(z)=^(m_t)f^(m_t)_m_1(z ) , m_1=0,1,2,where g^(m_t)_m_1m_1(z)= w_m_1(_1)w_m_t+m_1^*(_2 ) |1|_1-_2| essentially forms a differential - integral equation system ( the sum over @xmath252 can be considered as an integration ) . some mathematical formulae for evaluating the function @xmath253 in eq . ( 4.8 ) are given in appendix a. since the states with different @xmath254 are orthogonal , we can use the variational principle for each value of @xmath254 separately to find a rigorous upper bound to @xmath255 . to this end , we choose as a simple trial wavefunction the first term in eq . ( 4.6 ) , i.e. , we include only @xmath256 . equation ( 4.7 ) then reduces to a single differential equation which is straightforward to solve numerically . for @xmath257 , we find that the upper bound can be written in the form : ^(m_t ) < -0.16a()^2 , with @xmath258 ( depending on the values of @xmath254 ) , and @xmath102 is a coefficient close to unity ( as in eq . [ 3.2 ] ) . this expression is equivalent to eq . ( 3.22 ) , with @xmath259 playing the role of @xmath152 , but here @xmath254 is an integer . the form in eq . ( 4.9 ) is expected since @xmath260 measures the difference between the distances to the origin of the guiding center of the electron and that of the proton . although the @xmath261 state is not an exact @xmath1-eigenstate , in a qualitative sense the separation between electron and proton increases with increasing @xmath254 . thus we expect g^(m_t)(z ) ~1 ^ 1/2 , which then naturally leads to the form in eq . the decrease from @xmath7 in eq . ( 3.2 ) to @xmath262 in eq . ( 4.9 ) arises from the fact that both the electron and proton wavefunctions have finite spread around the same axis ( unlike the usual pseudomomentum approach , where the relative coordinate is used ) . the actual energy eigenvalue @xmath263 can be obtained by solving iteratively the series of equations in ( 4.7 ) using the standard shooting algorithm@xcite . we have carried out the calculations for the @xmath264 and @xmath265 states . our numerical results are given in figure 2 for three different values of field strength @xmath266 and @xmath267 . typically , more than @xmath268 terms in the sum of eq . ( 4.6 ) are needed in order to attain convergence of the energy to within @xmath269 . we find that the ground - state energy eigenvalue @xmath270 agrees with the standard value ( eq . [ 3.2 ] with @xmath124 ) in the limit of @xmath4 . this is expected because the @xmath264 ground state is also a @xmath181 eigenstate ( note that @xmath271 implies @xmath272 and hence @xmath264 ) . also , we see that as @xmath0 increases , the upper bound given by eq . ( 4.9 ) becomes asymptotically closer to the actual energy @xmath263 . for general @xmath254 , we can not give a rigorous lower bound to @xmath255 , but it is approximately given by eq . ( 4.9 ) with @xmath273 replaced by @xmath7 , so that the fractional uncertainty decreases as @xmath0 increases . note that although the contribution of the of @xmath274 terms in eq . ( 4.6 ) to the correct energy becomes smaller as @xmath0 increases , the contribution to the correct wavefunction is always non - negligible . it is instructive to consider the degeneracy of an energy eigenstate . without the coulomb interaction between electron and proton , there is a double landau degeneracy in @xmath252 and @xmath275 . when the coulomb interaction is included , the degeneracy in @xmath260 is removed , but a single degeneracy remains in @xmath252 , i.e. , for a given @xmath254 , the eigenfunction corresponding to @xmath263 is not unique : the eigenstate in eq . ( 4.6 ) with real @xmath276 s is presumably the state where the proton is centered at the origin ; but there must be other states with the same energy , centered at different positions . we can demonstrate this degeneracy explicitly as follows . a landau wavefunction centered at the origin of the coordinate can be expanded in terms of wavefunctions centered at some point @xmath277 as w_nm(_)=_me^i(m-m)_o i_n+m , n+m(r_o^22 ^ 2 ) e^i_o_/2w_nm(_-_o ) , where @xmath278 , @xmath279 is given by @xmath280 ( see eq . [ 2.2 ] ) , @xmath281 is the polynomial as defined in@xcite , and @xmath282 is the gauge factor . we consider only the ground landau level @xmath94 , and write w_m(_)= e^i_o_/2_mc_m^m w_m(_-_o ) , where @xmath283 is a ( complex ) function of @xmath284 and @xmath285 . the energy eigenstate in eq . ( 4.6 ) can then be written as ^(m_t)(_1,_2)=e^i_o(_1-_2)/2 _ m_t , where we have defined f_m_1^(m_t m_t)(z)_m_1 c_m_1^m_1 c_m_1+m_t^m_1+m_t f_m_1^(m_t)(z ) . however , the term inside @xmath286 $ ] in eq . ( 4.13 ) is exactly an energy eigenstate @xmath287 with @xmath288 based on the coordinate system centered at @xmath285 . since the state represented by the left - hand - side of eq . ( 4.13 ) has a definite energy @xmath263 , while states with different @xmath289 have different energies , we must have @xmath290 in eq . ( 4.13 ) . thus _ i^(m_t)(_1,_2)=e^i_o(_1- _ 2)/2_j^(m_t)(_1-_o,_2-_o ) , where we have added the subscripts `` i , j '' to indicate that there are many states associated with a given @xmath254 , i.e. , the states with the same energy @xmath263 is not unique . clearly , the degeneracy ( per unit area ) for a given @xmath254 is @xmath291 , i.e. , a single landau degeneracy ( see the discussion following eq . [ 2.5 ] ) . the above discussions demonstrate that there is a discrete set of states with @xmath292 , all having similar energies as the ground state ( @xmath264 ) , and do not have any positive contribution @xmath293 in their energies . this has important consequences for the binding of hydrogen molecules in the @xmath4 regime . in a forthcoming paper@xcite we shall use one @xmath264 and @xmath265 atom to construct the wavefunction for the h@xmath230 ground state , which also does not involve any landau excitation of the proton . we now consider the ionization - recombination equilibrium of hydrogen atoms , e@xmath294p@xmath295 h , given by the generalized saha equation in the presence of a strong magnetic field . previous treatments of this problem ( e.g.,@xcite ) either ignored the coupling between the center - of - mass motion and the internal atomic structure , or did not have available our generalized formula for the `` transverse kinetic energy '' as a function of the pseudomomentum @xmath1 . let @xmath205 be the gas temperature in atomic units ( about @xmath296 k ) , so that the boltzmann constant @xmath297 is set equal to unity , and @xmath298 be the number density ( also in a.u . ) of protons ( either free or bound ) in the gas . we write @xmath299 and @xmath300 , so that a `` wigner - seitz cylinder '' of radius and length @xmath301 contains one proton on the average . some of the partition function integrals can be simplified if the density and temperature satisfy three inequalities : ( i ) the density is low in the sense that @xmath302 is much larger than the largest dimension ( i.e. , the @xmath35-dimension ) @xmath303 of the ground - state atom ; ( ii ) the temperature is much smaller than the ground - state binding energy @xmath304 ; ( iii ) the coulomb attraction between a proton and an electron at typical separation @xmath301 is of order @xmath305 ; we assume @xmath306 so that the `` imperfect gas corrections '' are small . the saha equation involves the bound - atom partition function @xmath307 compared with the product @xmath308 of the two free - particle partition functions . each of these systems has six discrete or `` nominally continuous '' quantum numbers . two of these for the free e - p system refer to the @xmath35-motion , which can be represented by @xmath42 , the center - of - mass @xmath35-momentum and @xmath228 , the relative @xmath35-momentum . for the bound system ( h atom ) , the @xmath42-partition function is identical , but instead of @xmath228 we have the quantum number @xmath92 . in both systems the electron landau excitations have the same quantum number @xmath70 and energy @xmath309 , so that the @xmath70-partition functions are the same ( in practice , @xmath310 , so we only need to consider the ground landau state of electron and the @xmath70-partition function is essentially unity ) . for the bound system , the remaining three quantum numbers are @xmath57 and the two cartesian components of the transverse pseudomomentum @xmath17 . for the free e - p system the three quantum numbers are @xmath248 , the proton landau level integer , and the two transverse parameters @xmath311 and @xmath312 . we first consider only the ground state of the h atom , with @xmath313 . using eq . ( 3.25 ) , the canonical partition function in the volume @xmath314 of a `` wigner - seitz cylinder '' can be written as z(h ) r_g(mt2)^1/2 ( |e(h)|t)z _ , where the factor @xmath315 results from the free center - of - mass @xmath35-motion and @xmath316 is the ground - state binding energy . the partition function @xmath317 associated with the `` transverse motion '' of the atom has the form z_=_g(2)^2_0^k_max 2k_dk _ , where @xmath197 is the generalized `` transverse kinetic energy '' . the upper limit @xmath318 of the integral in eq . ( 5.2 ) is determined by the condition @xmath319 so that the pressure - ionized states are excluded in the bound state partition function . it thus has the density - dependent form @xmath320 . as discussed in sec . iii.d , @xmath321 is well approximated by eq . ( 3.26 ) for @xmath1 up to @xmath322 . however , for @xmath208 ( or @xmath162 ) the correct expression for @xmath321 is close to @xmath209 ( almost independent of @xmath1 ) , whereas the approximate @xmath321 in eq . ( 3.26 ) increases with @xmath1 . at very low density ( @xmath162 ) , there are highly excited states with @xmath1 between @xmath7 and @xmath320 , whose contribution to @xmath317 is proportional to @xmath323 . however , these states cover only a narrow range of binding energies ( of order @xmath324 ) and can be neglected compared with the ionized components , in view of the inequality @xmath325 . we can therefore omit these states entirely , but also make only a small error if we merely replace @xmath321 by the approximation in eq . ( 3.26 ) , which is an overestimate for @xmath326 . this approximation has the advantage that the extension of the integral in eq . ( 5.2 ) to infinity is not only finite but also small . we therefore get a convenient expression for @xmath317 by extending the integration to @xmath327 . we have z__0^ k_dk_=a_gm_t2 , with m_=m_(1 - 2m_t)^-1 . thus @xmath317 is proportional to @xmath328 , which is larger than @xmath133 ( or equal to it when @xmath329 , so that the effective mass approximation is valid throughout the regime of interest ) , and @xmath133 is larger than the actual mass @xmath82 . while the transverse motion is `` slowed down '' in the sense that @xmath330 is smaller than the zero - field result , the @xmath331 states still exist and their statistical weight @xmath332 is actually increased over the zero - field result by the strong magnetic field . we now consider the internal partition functions associated with the @xmath112 and @xmath134 excited states , i.e. , we write the total bound - state partition function as z(h)=v_g(mt2)^1/2m_t2 ( |e(h)|t)z_(h ) z_m(h ) . start with the quantum number @xmath92 . the internal partition function relative to the ground state is z_(h)1+(-|e_0|t)_=1^_max 2(|e_|t ) . here @xmath333 is given by eqs . ( 3.4)-(3.5 ) but can be approximated by @xmath334 , and the factor of @xmath335 comes from the near - degeneracy of the even and odd states . the sum in eq . ( 5.6 ) is thus similar to that for a field - free atom except that the usual weight - factor @xmath336 is missing as the atom is one - dimensional . since @xmath337 for the temperature regime of interest , the individual term in the sum is very small , but the @xmath338 would diverge if @xmath339 were allowed to become infinite . as for the high rydberg states in the field - free h atom , the divergence is avoided by including only states which fit inside the wigner - seitz cylinder . the size of the @xmath340 state is @xmath341 ( a.u . ) , so we should choose @xmath342 with energy @xmath343 , i.e. , we omit states with such extended wavefunctions that they would be pressure - ionized . with this prescription the sum in eq . ( 5.6 ) becomes finite but density - dependent , i.e. , it increases as @xmath344 with decreasing @xmath298 . making use of the inequality @xmath345 , we split the sum into two parts : one extends from @xmath346 to @xmath347 , where @xmath348 ; the other from @xmath347 to @xmath339 ( which exceeds @xmath347 because of the inequality ) . the first part contains only a few terms , each with @xmath337 , so they can be neglected ( compared with unity , coming from the @xmath95 term ) . the second part represents the highly excited states and would contribute approximately @xmath349 , which could be large for very small density @xmath298 . however , these states should be considered separately and compared with the ionized components : these states have negative energies of order @xmath350 , while the ionized components have positive energies of order @xmath205 . because of the inequality @xmath306 , we neglect these states entirely in the rest of the paper and set @xmath351 . the contribution of the @xmath134 states to the bound - state partition function can be considered in the same manner as that of the @xmath124 state discussed before . the internal partition function associated with the @xmath57-states is given by in eq . ( 5.7 ) results from the proton spin term , which is not explicitly included in our calculations ( see footnote 1 ) . note that when the proton spin term ( but not the abnormal magnetic moment ) is taken into account in eq . ( 2.11 ) or ( 2.13 ) , the energy of the @xmath57-th atomic state is given by @xmath352 , where the proton spin @xmath353 ( compare with eq . [ 3.27 ] ) . ] z_m(h)(1+e^-b / mt ) _ m=0^m_mm_ , where m_m=m_m(1 - 2m_mt_m)^-1 , so that @xmath354 ( cf . [ 5.4 ] ) , and @xmath136 , @xmath355 are given in eqs . ( 3.12 ) and ( 3.28 ) . the sum in eq . ( 5.7 ) simplifies in two extreme regimes of the field strength : for @xmath140 one may need to include a number of terms in the sum , but then @xmath356 is close to @xmath82 ; for @xmath142 the effective mass @xmath328 is much larger than @xmath82 , but since @xmath357 one need to include only the @xmath124 ground state , and hence @xmath358 . finally , it is instructive to consider the partition function based on the alternative scheme discussed in sec . iv . using eq . ( 4.9 ) ( and relacing @xmath262 by @xmath7 ; see the discussion following eq . [ 4.10 ] ) , we see that @xmath307 can still be written as eq . ( 5.1 ) , but the transverse partition function @xmath317 is now given by z _ ~_gb2_m_t=0^m_t , max ~_gb2_0^m_t , maxdm_t ( 1+cm_t)^-l / t , where the factor @xmath359 comes from the landau degeneracy of the @xmath254-th state , and @xmath360 . clearly , eq . ( 5.9 ) has the same form as eq . ( 5.3 ) in the limit of @xmath361 , demonstrating the equivalence of the two energy level schemes of sec . ii - iii and sec . the partition function of the free electrons in the ground landau level is given by z(e)r_g(t2)^1/2 ( _ gb2 ) , where the factor @xmath362 represents the free @xmath35-motion and the factor @xmath363 is the landau degeneracy . for the free protons ( see footnote 1 ) , we have @xmath364^{-1},\end{aligned}\ ] ] where the sum extends over all landau levels of the proton , and @xmath365 is the spin degeneracy : @xmath366 for @xmath367 and @xmath368 for @xmath369 . given @xmath370 , @xmath371 and @xmath307 , the ionization - recombination equilibrium can be obtained using the condition @xmath372 for the chemical potentials . in the density and temperature regimes of interest , with @xmath373 , we have n_g(b2)^-2 m_(1 - 2m_t)^-1 ( t2)^1/2(b2mt ) ( |e(h)|t)z_m(h ) , where @xmath374 , @xmath375 , @xmath376 are the number density fraction of different species , @xmath377 ( with @xmath378 ) , and @xmath379 is given by eq . this is the generalized saha equation in the presence of a superstrong magnetic field . more details on the applications of this result to neutron star atmospheres will be presented elsewhere@xcite . the effects of center - of - mass motion of neutral hydrogen atom in a strong magnetic field are rather intricate , mainly due to the high degree of degeneracy associated with the quantum states . using the usual pseudomomentum scheme ( sec . ii - iii ) , we have obtained approximate solutions for the energy of the atom as a function of the field strength and conserved pseudomomentum for a wide range of parameter regimes . in particular , we have considered field strengths @xmath380 g , when the landau excitation energy of proton is considerable . states with large transverse pseudomomentum have small binding energy and transverse velocity , but are nevertheless quite distinct from fully ionized states . we have concentrated on convenient analytic fitting formulae which give at least a reasonable approximation over various parameter regimes ( see particularly eq . ( 3.26 ) for the `` transverse kinetic energy '' ) . since there may be neutron star atmospheres with @xmath381 g , we are particularly interested in the cases with @xmath4 , where the proton landau energy @xmath382 is very large . by considering an alternative scheme to the usual pseudomomentum formulation ( sec . iv ) , we have shown that there are atomic states with orbital wavefunctions orthogonal to that of the ground state , but without any landau excitation appearing in their energies . we have also derived the generalized saha equation for the equilibrium between neutral hydrogen atoms and the ionized component . we focused on the cases of astrophysical interest , where the density is relatively low and the thermal energy @xmath25 is small compared to the atom s ground state binding energy . although the maximum transverse velocity of bound atoms is small in strong magnetic fields , the statistical weight due to transverse motion is actually increased by the strong fields , not decreased ( sec . the statistical weight of highly excited bound states is smaller than that of the fully ionized component . our results are important for determining the physical conditions of magnetic neutron star atmospheres as well as the soft x - ray ( or euv ) radiation spectra from them . some of these issues will be studied in a future paper@xcite . we thank lars bildsten , george pavlov and ira wasserman for useful discussions . this work has been supported in part by nsf grants ast 9119475 , ast 9315375 , and nasa grant nagw666 to cornell university , and nagw-2394 to caltech . in this appendix , we derive some mathematical formulae needed for evaluating the function @xmath253 defined by eq . ( here the lengthscale is in units of the cyclotron radius @xmath59 . ) using the identity @xmath383 we write eq . ( 4.8 ) as g_m_1m_1^(m_t)(z)=g_m_1m_1^(m_t)(z)= 12 ^ 2e^iq_zz m_t+m_1|e^i*q*__1|m_t+m_1m_1|e^-i*q*__2|m_1 . using the general result@xcite for the matrix element @xmath384 , and integrating over @xmath385 , we obtian ( assuming @xmath386 without loss of generality ) : g_m_1m_1^(m_t)(z)=_0^ dq e^-q|z|-q^2 ( q^22)^m_1-m_1 l_m_t+m_1^m_1-m_1(q^22 ) l_m_1^m_1-m_1(q^22 ) , where @xmath387 is the laguerre polynomial of order @xmath70@xcite . we now define constant coefficients @xmath388 via ( x2)^m_1-m_1 l_m_t+m_1^m_1-m_1(x2 ) l_m_1^m_1-m_1(x2 ) = _ n=0^m_1+m_1+m_t g_n^(m_t)(m_1,m_1)l_n(x ) . using the relation v_n(z)=_0^dq e^-q^2/2-q|z|l_n(q^22 ) = 1 n!_0^ dx x^n e^-x(x+z^2/2)^1/2 , where the second equality can be used to evaluate the function @xmath389 , we obtain g_m_1m_1^(m_t)(z)=_n=0^m_1+m_1+m_tg_n^(m_t)(m_1,m_1 ) v_n(z ) . we calculate the coefficients @xmath388 using the orthogonal relations of laguerre polynomials . we can identify two special cases : ( i ) when @xmath264 , we have @xmath390 and @xmath391 , where @xmath392 and @xmath393 are given in@xcite ; ( ii ) when @xmath394 , we have @xmath395 and @xmath396 , where @xmath397 and @xmath398 are again given in@xcite . early study of essentially the same problem dealt with excitons in semiconductors , where the strong field condition can be mimicked in virtue of the small effective mass and the presence of a dielectric medium . see r. j. elliot and r. loudon , j. phys . solids * 15 * , 196 ( 1960 ) ; h. hasegawa and r. e. howard , _ ibid . _ * 21 * , 179 ( 1961 ) . for calculations in the superstrong field regime ( @xmath399 g ) , see r. cohen , j. lodenquai , and m. ruderman , phys . lett . * 25 * , 467 ( 1970 ) ; v. canuto and d. c. kelley , astrophys . space sci . * 17 * , 277 ( 1972 ) ; j. simola and j. virtamo , j. phys . b : at . mol . * 11 * , 3309 ( 1978 ) , and references therein .

the paper studies closed queueing networks containing a server station and client stations . the server station is an infinite server queueing system , and client stations are single - server queueing systems with autonomous service , i.e. every client station serves customers ( units ) only at random instants generated by a strictly stationary and ergodic sequence of random variables . the total number of units in the network is . the expected times between departures in client stations are . after a service completion in the server station , a unit is transmitted to theth client station with probability , and being processed in theth client station , the unit returns to the server station . the network is assumed to be in a semi - markov environment . a semi - markov environment is defined by a finite or countable infinite markov chain and by sequences of independent and identically distributed random variables . then the routing probabilities and transmission rates ( which are expressed via parameters of the network ) depend on a markov state of the environment . the paper studies the queue - length processes in client stations of this network and is aimed to the analysis of performance measures associated with this network . the questions risen in this paper have immediate relation to quality control of complex telecommunication networks , and the obtained results are expected to lead to the solutions to many practical problems of this area of research .

math0612224

we consider closed queueing networks containing a server station and @xmath0 client stations . the server station is an infinite server queueing system with identical servers . client stations are single - server queueing systems with an autonomous service mechanism , where customers ( units ) are served only at random instants generated by a strictly stationary and ergodic sequence of random variables . queueing systems with an autonomous service mechanism were introduced and originally studied by borovkov @xcite , @xcite . the formal definition of these systems in the simplest case of single arrivals and departures is as follows . let @xmath6 denote an arrival point process , let @xmath7 denote a departure point process , and let @xmath8 be a queue - length process , and all of these processes are started at zero ( @xmath9 ) . then the autonomous service mechanism is defined by the equation : @xmath10 the queueing systems with an autonomous service mechanism have been studied in many papers ( e.g. abramov @xcite , @xcite , @xcite , fricker @xcite , @xcite , gelenbe and iasnogorodski @xcite ) . the structure of queueing systems or networks with autonomous service and their analysis is much easier than those structure and analysis of usual " systems with generally distributed service times . queueing systems and networks with autonomous service , because of their simple construction , are studied under general settings on dependent inter - arrival and inter - departure times , and their analysis is often based on methods of stochastic calculus and the theory of martingales . the corresponding results for usual markovian queueing systems or networks follow as a particular case of the corresponding results for queueing systems or networks with poisson input and autonomous service . for different applications of queueing systems ( networks ) with autonomous service see e.g. @xcite , @xcite , @xcite and @xcite . the assumption that the queueing mechanism is autonomous substantially simplify the analysis . however , according to sample - path results of @xcite this assumption can be removed . so , all of the main results remain valid for quite general client / server networks without the special assumption that the service mechanism is autonomous . in the present paper we study client / server networks in a semi - markov environment . there has been an increasing attention to queueing systems in a random environment in the literature ( e.g. @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite and others ) . however , most of these papers mainly develop the theory and remain far from real - world applications . the aim of the present paper is twofold . first , we establish new theoretical results for client / server networks in semi - markov environment describing the behavior of queue - length processes of this network . second , we show how these theoretical results can be applied to solve real - world problems . some of these problems are solved in the present paper . other ones will be solved in the future . the model of the network , which is considered in this paper , is very close to the models considered in @xcite and @xcite ( see figure 1 ) . for a more general construction of network with two types of node and multiple customer classes see @xcite . the description of the present model is based on that of the model of @xcite . for other papers studying the models of client / server networks see also @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite and other papers . the previous assumptions of @xcite , that are also used in the present paper , are repeated below . the departure instants in the @xmath3th client station ( @xmath11 ) are denoted by @xmath12 , @xmath13 , @xmath14 , , and @xmath15 each sequence @xmath16 forms a strictly stationary and ergodic sequence of random variables . the corresponding point process associated with departures from the client station @xmath3 is denoted @xmath17 the total number of units in the network is @xmath1 . the number @xmath1 is a large parameter , and we assume that @xmath1 increases indefinitely . this means that we assume that @xmath1 is a series parameter , and the series of models ( with different @xmath1 ) are considered on the same probability space . it was assumed in @xcite that the service time of each unit in the server station is an exponentially distributed random variable with a _ given _ parameter @xmath18 . in the present paper , the assumption is another . under the assumption that an environment is random ( markov or semi - markov ) , the parameter @xmath18 is not longer a _ constant value_. it is a random variable , taking values in dependence of environment states . ( by environment states we mean the states of the corresponding markov chain , which is described later . ) the same is noted regarding the routing probability matrix . there were the standard assumptions in @xcite that after a service completion at the server station , a unit was transmitted to the client station @xmath3 with probability @xmath4 , @xmath19 , and @xmath20=1 . these assumptions are not longer valid in the case of the system considered in this paper . the routing probabilities are assumed to be random , taking the values in dependence of the environment states as well . let us first describe _ markov states _ of semi - markov environment , and then latter we describe the notion of semi - markov environment as well . the network is assumed to be complete in the following sense . the server station and @xmath0 client stations communicate by links . the number of links is @xmath0 . therefore , the markov chain of the environment states is defined as follows . the number of possible environment states is assumed to be finite or countable infinite . these states are denoted by @xmath21 , @xmath22 . for the markov or semi - markov environment considered later we shall also use the notation @xmath23 . the meaning of this notation is the state of markov chain in time @xmath24 . the initial state is denoted by @xmath25 or @xmath26 . the completeness of the network means that @xmath15 for any @xmath3=1,2 , , @xmath0 there exists @xmath27 , such that @xmath28 . this assumption is not used in our proofs explicitly . nevertheless , it must be mentioned . if @xmath29 for all @xmath27 , then the station @xmath3 becomes isolated and therefore not representative . in the above assumption by @xmath30 we mean the value of probability @xmath31 when the network is in state @xmath21 . it is assumed additionally that for any @xmath22 , the sum of probabilities @xmath32 = 1 . the notation with similar meaning is used for parameter @xmath18 . namely , @xmath33 is assumed to be strictly positive for any state @xmath21 , however @xmath34 = @xmath35 can be equal to 0 ( because @xmath30 need not be strictly positive in general ) . the results of @xcite are associated with the asymptotic analysis of a closed client / server network with a bottleneck station as @xmath1 increases indefinitely . one of the main results of @xcite was then developed in @xcite for networks containing two types of node and multiple customer classes , where one of client stations was bottleneck . the results obtained in @xcite , @xcite , @xcite as well as in preceding paper @xcite all are a theoretical contribution to the theory of client / server computer networks with bottlenecks . the aim of the present paper is another . we follow towards performance analysis , and are aimed to use the known theoretical results of @xcite and @xcite . however , the known theoretical results obtained in these papers are scanty for their immediate application to real telecommunication systems , whose parameters can change in time . therefore , for the purpose of performance analysis , a substantial development of the earlier results related to this model is required . therefore , before studying the behaviour of queueing networks in semi - markov environment , we first study the behaviour of this network in its special case of piece - wise deterministic environment . then the results are extended to the more general situation of semi - markov environment . the paper is organized as follows . in section [ motivation ] , the paper is motivated by formulating the concrete real - world problems , where the results of the present paper can be applied . in section [ definitions ] , necessary notions of piece - wise deterministic , markov and semi - markov environments as well as classification of bottleneck stations are defined . in section [ semi - martingale ] , the semi - martingale decomposition for the queue - length process in semi - markov environment is derived . then in section [ piece - wise deterministic ] , the queue - length processes in piece - wise deterministic environment are studied . in fact we study usual " closed client / server networks under general assumptions . in section [ semi - markov ] , the results are extended to the case of semi - markov environment . in section [ application ] , an example of an application of the main theorems of the present paper to client / server computer network with failing client stations is considered . in section [ discussion ] , new problems and monotonicity conditions associated with these problems for the case of markov environment are discussed . in section [ numerical example ] , the example of numerical study for the simplest network is considered . the results of the paper are concluded in section [ concluding remarks ] . in this section we motivate our study by real - world problems associated with performance analysis of the network , its separate stations or a subset of those stations . the main results of the present paper are the subject for many concrete practical problems having significant value . two of them are considered in @xcite , @xcite . article @xcite assumes that client stations all are identical and subject to breakdowns . article @xcite discusses the similar problem for not identical client stations . as a client station is breakdown the parameters of the network are changed . one or other breakdown leads to bottleneck and risk of a destruction of the network performance . the aforementioned articles @xcite , @xcite establish confidence intervals helping to avoid with a given high probability that performance destruction . some results of these articles are outlined in section [ application ] of this paper . another circle of problems is as follows . if a computer network operates for a given fixed time interval ( 0 , @xmath36 ) , then its performance characteristic ( for example , payment for the increasing a given level of queue in a given client station , or in a subset of client stations ) depends on the characteristics of this network at the initial time @xmath37 . using the terminology of the present paper , these characteristics at the initial time @xmath37 specify the initial condition of the environment . for one initial condition , the measure of time that the queue - length is greater than a given fixed level @xmath38 is @xmath39 , and we must pay , say @xmath40 . for other initial condition , this measure of time is @xmath41 and the corresponding payment is @xmath42 . on the other hand , the cost for initialization the first initial condition related to the above client station of the network is @xmath43 , and that cost for initialization the second initial condition is @xmath44 . so , the total expenses in the first case are @xmath43 + @xmath40 , and in the second case @xmath44 + @xmath42 . if @xmath43 + @xmath45 + @xmath42 , then we say that the first _ strategy _ is more profitable than the second one . in practical context , the first strategy can mean _ the first type repair _ of the network , or a failing client station , or a subset of failing client stations of the network , and the second strategy - _ the second type repair _ of that network , or a failing client station , or a subset of failing client stations correspondingly . for example , the first type repair can contain an additional prophylactical service and therefore to be more expensive that the second type repair ( i.e. @xmath46 ) . let us now define a semi - markov environment in mathematical terms . for each state @xmath47 of the markov chain let us define the sequence of independent and identically distributed random variables : @xmath48 [ defn - semimi]let @xmath49 be a markov chain with finite or countable infinite states space @xmath50 . let @xmath51 be a random process , and let @xmath52 be an increasing sequence of random points , i.e. @xmath53 . suppose that the process @xmath51 is defined as follows : @xmath54 , and @xmath55 for all @xmath56 . suppose also that the increments @xmath57 , @xmath58=0,1,2 , , coincide in distribution with @xmath59 , where the equality @xmath60 defines the index @xmath61 . then the process @xmath51 is called _ semi - markov environment_. the above semi - markov environment @xmath51 is assumed to be given on special probability space @xmath62 , which in turn is contained in the common filtered probability space @xmath63 , @xmath64 , @xmath65 ) , @xmath66 . the two significant special cases of semi - markov environment are as follows . [ defn - piece - wise - det ] a semi - markov environment is called piece - wise deterministic , if @xmath15 the sequences of all are deterministic , i.e. @xmath67 , where @xmath68 is the same constant for all @xmath69 . @xmath15 the parameters of networks are not randomly dependent of the environment states , i.e. @xmath70 , @xmath71 , , @xmath72 , , for all @xmath3=1,2 , , @xmath0 , where @xmath73 , @xmath74 , all are non - random constants . in other words , a piece - wise deterministic environment defines time dependent closed client / server network with fixed piece - wise constant parameters depending on time . [ defn - mi ] in the case where the sequences consist of independent and exponentially distributed random variables with rates @xmath68 , @xmath61=1,2 , , then the semi - markov environment @xmath51 is called _ markov environment_. another equivalent definition of markov environment and independent of definition [ defn - semimi ] is as follows . let @xmath51 be a homogeneous markov process with initial state @xmath26 and transition probabilities @xmath75 from the state @xmath76 to the state @xmath77 ( @xmath78 ) during a small time interval ( @xmath24 , @xmath79 ) , and there is probability @xmath80 to stay in the same state @xmath76 during the same time interval ( @xmath24 , @xmath81 ) . in many papers on queueing theory , telecommunication systems , inference of stochastic processes , statistics and other areas , the above markov environment is often associated with _ markov modulated poisson process_. the phrases _ markov environment _ , _ piece - wise deterministic environment _ and _ semi - markov environment _ are more appropriate in the context of the present paper . there was mentioned that @xmath23 denotes the state of the semi - markov environment in time @xmath24 . for example , the equality @xmath82 means that in time @xmath83 the associated markov chain is in state @xmath84 . we also use the following notation : @xmath85 . in the sequel we shall also use the notation @xmath86 , @xmath87 and correspondingly , @xmath88 . it is worth noting , that the definition of @xmath86 and @xmath89 and consequently @xmath90 and @xmath91 have the different meaning . in general @xmath89 @xmath92 and @xmath91 @xmath93 . @xmath88 and @xmath91 both are random parameters , and @xmath94 where @xmath95 , while @xmath96 where @xmath97 . ( in the sequel this dependence upon @xmath98 will be always implied , but shown explicitly only in the cases where it is necessary . ) more specific explanation of the above difference is as follows . if at the initial time moment @xmath37 all of units are in the server station , then the input rate to the @xmath3th client station is @xmath99 . ( @xmath100 is the individual rate of each unit arriving to the client station @xmath3 , and therefore the rate between arrivals is @xmath99 . ) the time parameter 0 in parentheses is associated with the state of markov environment in time 0 . specifically , if at the initial time moment @xmath37 all of units are in the server station , then @xmath101 . if the network is considered without markov or semi - markov environment , then that initial arrival rate is @xmath102 which is associated with the individual rate @xmath103 of each unit of the server station . if at the initial time moment there are @xmath104 units in the server station , @xmath105 , then for arrival process to any client station @xmath3 of a standard network ( without random environment ) we also use the notation @xmath100 . however , the meaning of @xmath100 is not longer the individual rate of each unit at time @xmath37 arriving to the station @xmath3 . more specifically , @xmath106 , where @xmath103 is the individual rate of each unit at time @xmath37 arriving to the station @xmath3 . the meaning of @xmath90 is similar . the only difference that it is said about an arbitrary time @xmath24 . for example , if there are @xmath107 units in the server station in time @xmath24 , @xmath108 , then @xmath109 . resuming the above , @xmath91 is associated with individual service rate of each unit , while @xmath90 is a recalculated ( relative ) rate depending on the state of the queue - length processes in time @xmath24 , as explained above . regarding the departure ( service ) rates our assumption in the paper is as follows . the departure rate of the @xmath3th client station is assumed to be independent of semi - markov environment as well as of input rates @xmath90 . specifically , @xmath15 it is assumed that the the expectation of service ( inter - departure ) time in the @xmath3th client station is @xmath110 for all @xmath111=1,2 , therefore , if at the initial time moment @xmath37 all of units are in the server station , then the load parameter of the @xmath3th client station is @xmath112 , and in the case where there is no semi - markov environment , @xmath113 . in the case of network in markov environment , the meaning of the notation @xmath90 is the same as well . if there are @xmath114 units in the server station in time @xmath24 , @xmath108 , and the rate of arrival of a unit from the server to the client station @xmath3 is @xmath91 , then @xmath90 = @xmath115 . then the load of the @xmath3th client station in time @xmath24 is @xmath116 . now introduce necessary definitions . the first two definitions are related to both standard client / server networks and client / server networks in semi - markov environment . [ defn1 ] the client station @xmath3 is called _ locally non - bottleneck _ in time @xmath24 if @xmath117 . otherwise , the @xmath3th client station is called _ locally bottleneck _ in time @xmath24 . a client station locally ( non- ) bottleneck in time 0 will be also called _ initially ( non- ) bottleneck_. [ defn2 ] the client station @xmath3 is called _ non - bottleneck _ in time interval [ @xmath118 if it is locally non - bottleneck in all points of this interval otherwise , if there is a point @xmath119 such that @xmath120 , then the client station is called _ bottleneck _ in time interval [ @xmath118 . a client station is called _ ( non- ) bottleneck _ if it is ( non- ) bottleneck for all @xmath24 . the special definition for standard client / server networks ( without semi - markov environment ) is as follows . [ defn3 ] a client station is called _ absolutely non - bottleneck _ if it is a locally non - bottleneck station at the moment when all of units are in the server station . otherwise , a client station is called _ absolutely bottleneck_. clearly , that absolutely non - bottleneck client station @xmath3 is a non - bottleneck client station , because then for all @xmath24 we have @xmath121 . in the next section we prove that absolutely bottleneck client station is a bottleneck client station as well . that is , if a client station is currently locally bottleneck station , then it never can become a locally non - bottleneck . this means that the client station is forever bottleneck . definition [ defn3 ] can be extended to network stations in semi - markov environment for an arbitrary time @xmath24 . specifically , we have the following definition . [ defn4 ] the client station @xmath3 of a network in random environment is called absolutely non - bottleneck in time @xmath24 if @xmath122 . in other words , the client station @xmath3 is absolutely non - bottleneck in time @xmath24 if in that time @xmath24 the network belongs to some state @xmath47 of the environment , i.e. @xmath23 = @xmath47 , and in this state @xmath123 . otherwise , this client station is called absolutely bottleneck in time @xmath24 . the last notion enables us to judge on the behavior of client stations in random intervals [ @xmath124 ) , where the network is in given state @xmath76 . recall that @xmath125 is a time instant when the state of semi - markov environment is changed . so , during the random interval [ @xmath124 ) the network is in a fixed state of the semi - markov environment . consider a client station @xmath3 ( @xmath11 ) . let @xmath126 denote a queue - length there in time @xmath24 . assume that at the initial time instant @xmath37 , all of units are in the server station , i.e. @xmath127 for all @xmath11 . this is the simplest case , and we start from its study . for @xmath128 , @xmath129 where @xmath130 is the arrival process to client station @xmath3 , and @xmath131 is the departure process from that client station @xmath3 . the equation for departure process is as follows . let @xmath132 be a point process associated with consecutive departures from the @xmath3th client station . then , @xmath133 the definition of the departure process given by is as in @xcite . however , the construction of arrival process is more difficult . specifically , @xmath134 the processes @xmath135 , @xmath136 , appearing in relation are a collection of conditionally independent poisson processes with parameters depending on @xmath98 . this means the following . assume that for a given realization @xmath98 we have a sequence : 0@xmath137 , and for some @xmath58 , @xmath138 . assume that @xmath139 . then @xmath135 , @xmath136 , is the sequence of poisson processes with the parameter @xmath140 . that is the rate of poisson process depends on the state of semi - markov environment in time @xmath141 . this is just the main difference between the consideration of @xcite , where @xmath103 , @xmath11 , were non - random constants . relation can be then rewritten @xmath142 where @xmath143 is an associated sequence of ( conditionally ) independent poisson process with parameter depending on the state of the semi - markov environment @xmath144 . ( here in and later we use the standard notation for a minimum of two numbers : @xmath145 . ) relations and enable us to write the following representation for the queue - length process @xmath126 : @xmath146 this implies that @xmath126 is the normal reflection of the process @xmath147 at zero . more accurately , @xmath126 is a nonnegative solution of the skorokhod problem ( see @xcite as well as @xcite , @xcite , @xcite and others ) on the normal reflection of the process @xmath148 at zero ( for the detailed arguments see @xcite ) . recall that according to the skorokhod problem , @xmath149 and @xmath126 has the representation @xmath150 in the sequel , it is convenient to use the notation : @xmath151 for any cdlg function @xmath152 satisfying @xmath153 ( see e.g. @xcite ) . according to this notation , can be rewritten @xmath154 . next , we take into account that the process @xmath130 is a semimartingale adapted with respect to the filtration @xmath155 . let @xmath156 denote the compensator of @xmath130 and @xmath157 denote the square integrable martingale of @xmath130 in the doob - meyer semimartingale decomposition : @xmath130 = @xmath156 + @xmath157 , @xmath11 . then , the process @xmath148 given by can be represented @xmath158 where @xmath159 the details for last formula can be obtained from @xcite or @xcite , theorem 1.6.1 . let us now study asymptotic properties of the normalized queue - lengths @xmath160 = @xmath161 as @xmath162 . for normalized processes we will use small latin letters . for example , @xmath163 , @xmath164 , and so on . then , from we have @xmath165 let us derive a relation for @xmath166 . ( @xmath167 denotes the limit in probability . ) from lenglart - rebolledo inequality ( e.g. liptser and shiryayev @xcite ) , we have : @xmath168 where @xmath169 = @xmath170 vanishes such that @xmath171 . then , by virtue of , the term @xmath172 = @xmath173 vanishes as well , and for any small @xmath174 the fraction @xmath175 vanishes . therefore , @xmath176 for all @xmath11 and @xmath177 . next , according to the assumption above , @xmath178 . therefore , @xmath179\mbox{d}s . \end{aligned}\ ] ] representation is the extension of the similar result of @xcite for queue - length processes in client stations of standard client / server networks . recall that the representation obtained in @xcite and @xcite is @xmath180\mbox{d}s.\ ] ] as we can see , the representations given by and are similar . the only difference is in the presence of infinite sum containing the integrals with random upper and lower bounds , and is a particular case of . in this section we discuss the behavior of the queue - length processes in a large closed client / server network in the case of piece - wise constant environment . for this case relation reduces to @xmath181\mbox{d}s\\ & & + \int_{\sigma_{r}}^{t } \left[\lambda_j\left\{1-\sum_{i=1}^k \phi_s(x_i)\right\}-\mu_j\right]\mbox{d}s,\nonumber\end{aligned}\ ] ] where @xmath182 is the number of the state changes before time @xmath24 . therefore , the bottleneck analysis of the network in piece - wise constant environment reduces to the analysis of the network in traditional formulation ( without random environment ) in some given time intervals such as [ @xmath183 ) , @xmath111=1,2, ,@xmath182 , or [ @xmath184 ) . notice , that the bottleneck analysis of the markovian client / server model has been originally studied by kogan and liptser @xcite . then these results were extended for the case of autonomous service mechanism in client stations in @xcite . however , the results obtained in both of these papers are related to a single special case and are not enough for the purpose of our performance analysis . therefore we will study all possible cases including the behaviour of the network under different initial lengths of queues in client stations and several absolutely bottleneck and absolutely non - bottleneck client stations . in @xcite and @xcite there has only been considered the case where in the initial time moment @xmath37 all of units are in the server station ( i.e. the client stations all are empty ) and only one ( the @xmath0th ) client station is a bottleneck station . ( in this particular case the notions of ( non-)bottleneck and absolutely ( non-)bottleneck client station coincide . ) specifically , there has been proved the following result in @xcite . [ prop1 ] let @xmath185 , and @xmath126 denotes the queue - length in the @xmath3th client station in time @xmath24 . under the assumption that the @xmath0th client section is bottleneck , for @xmath186 and for any @xmath128 we have : @xmath187=0\}=1-\rho_j(t),\label{ba1 - 1}\\ & & \lim_{n\to\infty}\int_0^t\rho_j(s)\mathbb{p}\{q_{j , n}(s)=l\}\mbox{d}s\nonumber\\ & & = \lim_{n\to\infty}\int_0^t \mathbb{p}\{q_{j , n}[s_{j , n}^*(s)]=l+1\}\mbox{d}s,\label{ba1 - 2}\\ & & \ \ \ l=0,1,\ldots,\nonumber\end{aligned}\ ] ] where @xmath188,\label{ba2 - 1}\\ q(t)&=&\left(1-\frac{1}{\rho_k(0)}\right)(1-\mbox{e}^{-\rho_k(0)\mu_kt}).\label{ba2 - 2}\end{aligned}\ ] ] in the case of markovian network , relations , can be resolved explicitly , and queue - length distribution in the non - bottleneck stations is the time - dependent geometric distribution @xcite , @xcite . the meaning of @xmath189 in is the limiting fraction of units remaining at the server station in time @xmath24 as @xmath162 . for example , in the case where @xmath190 this fraction remains the same at any time @xmath24 as initially , that is , as @xmath1 large , the number of units in the server station remains asymptotically equivalent to @xmath1 . however , if @xmath191 , then the number of units in the server station in time @xmath24 is asymptotically equivalent to @xmath192 $ ] . then , the number of units remaining in bottleneck station in time @xmath24 is asymptotically equivalent to @xmath193 . let us study various cases of the client / server network with bottlenecks . these cases will be studied in order of increasing complexity . we possibly shorten the proofs . assuming that the initial condition of the network is the same as in proposition [ prop1 ] ( i.e. at the initial time @xmath37 all of units are at the server station ) , let us study the case where the client stations 1,2, ,@xmath194 are non - bottleneck , while the rest client stations @xmath195 are bottleneck . in this case we have the following result . [ prop2 ] under the assumption that the client sections @xmath195 are bottleneck , for @xmath196 and for any @xmath128 we have : @xmath187=0\}=1-\rho_j(t),\label{ba3 - 1}\\ & & \lim_{n\to\infty}\int_0^t\rho_j(s)\mathbb{p}\{q_{j , n}(s)=l\}\mbox{d}s\nonumber\\ & & = \lim_{n\to\infty}\int_0^t \mathbb{p}\{q_{j , n}[s_{j , n}^*(s)]=l+1\}\mbox{d}s,\label{ba3 - 2}\\ & & \ \ \ l=0,1,\ldots,\nonumber\end{aligned}\ ] ] where @xmath188,\label{ba4 - 1}\\ q(t)&=&\left(1-\frac{\sum_{v = k_0 + 1}^k\mu_v}{\sum_{v = k_0 + 1}^k\lambda_v}\right ) \left(1-\exp\left[-t\sum_{v = k_0 + 1}^k\lambda_v\right]\right).\label{ba4 - 2}\end{aligned}\ ] ] note , that equations and are the same as and , and the main difference between proposition [ prop1 ] and proposition [ prop2 ] is only in the expression for @xmath197 . the difference between expressions , and , can be easily explained in the framework of the proof of proposition [ prop2 ] . considering all of the bottleneck client stations as a separate subsystem , one can notice that the arrival rate to this subsystem is @xmath198 , and the service rate ( the sum of reciprocals of the expected inter - departure times ) is @xmath199 . this subsystem can be thought as a bottleneck station with the load @xmath200 to be specific , note the method of sections 2 , 3 , and 4 of @xcite leads to the same equations , and the equations for normalized queue - lengths in the bottleneck stations all are an elementary extension of the case considered in @xcite . therefore proposition [ prop2 ] is an elementary extension of proposition [ prop1 ] and its proof is the same as in @xcite or @xcite . notice , that the number of units remaining at the bottleneck station @xmath69 , @xmath201 in time @xmath24 is asymptotically equal to @xmath202t-\lambda_v\int_0^tq(s)\mbox{d}s\right).\ ] ] recall the main elements of the known proof for the representation @xmath197 given by , and consequently explain the proof of . we use the notation similar to that of the earlier papers @xcite and @xcite . the difference between arrival and departure processes in the @xmath3th client station is denoted @xmath203 and its normalization @xmath163 . let @xmath204 , @xmath11 , denote the limit in probability of @xmath205 , as @xmath162 . the queue - length in the @xmath3th client station in time @xmath24 is denoted @xmath126 , its normalization is denoted @xmath206 , and the limit of @xmath160 in probability , as @xmath162 , is denoted @xmath207 . next , let @xmath208 for any cdlg function @xmath43 satisfying the condition @xmath153 . the functional @xmath209 has been introduced in @xcite . it characterizes a solution of the skorokhod problem on normal reflection at zero . therefore @xmath210 , being the functional @xmath211 applied to the function @xmath204 , describes the dynamic of normalized queue - length in the @xmath3th client station under the usual " initial conditions given in propositions [ prop1 ] and [ prop2 ] . under these usual " conditions , the functions @xmath204 , @xmath11 satisfy the system of equations : @xmath212-\mu_j\right\}\mbox{d}s , \ \ \ j=1,2,\ldots , k.\ ] ] note , that the normalized functions @xmath204 , @xmath11 are usual ( non - random ) continuous functions , and characterizes a usual system of linear differential equations . the statement of proposition [ prop1 ] is based on the solution of the system of these equations . ( it is proved in @xcite that there is a unique solution of the system of equations . ) more specifically , in the case where the node @xmath0 is bottleneck , @xmath213 , and we therefore have @xmath214 . in other words , @xmath215 can be replaced by @xmath216 in these equations of . the solution of the system of equations is @xmath204=0 for the non - bottleneck stations @xmath186 , and @xmath217 for the bottleneck station @xmath218 , where @xmath197 is given by . in the case of several bottleneck stations , we write the similar equation for the dynamic of the normalized cumulated queue - length process in all of the bottleneck stations . specifically , @xmath219-\sum_{v = k_0 + 1}^k\mu_v\right\}\mbox{d}s,\ ] ] gives solution for @xmath197 in the similar statement of proposition [ prop2 ] . then , the solution of system for the bottleneck client stations @xmath69=@xmath220,@xmath221 , @xmath0 is given by @xmath222 where @xmath197 is defined by , and the queue - lengths in the bottleneck stations are asymptotically evaluated by relation . now , we discuss the behavior of the network , in which the client stations are not initially empty . this is the next step of the extension of the original result of proposition [ prop1 ] . the analysis of cases related to initially not empty queues is much more difficult . therefore , we start from the simplest case of the network containing only one client station , i.e. @xmath223 . let @xmath224 be a positive real number , and let us assume that the initial number of units in this client station is asymptotically equivalent to @xmath225 . consider the following two cases : ( i ) the client station is initially bottleneck , i.e. @xmath226 , and ( ii ) the client station is initially non - bottleneck , i.e. @xmath227 . case ( i ) is relatively simple . it is a simple extension of the cases considered above . specifically , we have the following system of equations : @xmath228-\mu_1\right\}\mbox{d}s . \end{aligned}\ ] ] from we have the following solution : @xmath229 the normalized queue - length @xmath230 in this client station is @xmath231 . case ( i ) can be easily extended to a more general case of @xmath0 initially bottleneck client stations . let @xmath232 , @xmath11 , denote nonnegative real numbers , and @xmath233 + @xmath234 + + @xmath235 . assume then that the initial number of units in the @xmath3th client station is asymptotically equivalent to @xmath236 . let @xmath197 denote the cumulated normalized queue - length in all of client stations . [ prop3 ] assume that all client stations are initially bottleneck , and the initial queue - lengths in client stations are asymptotically equivalent to @xmath225 , @xmath237 , , @xmath238 correspondingly ( @xmath239 ) , as @xmath162 . then , @xmath240\right ) , \end{aligned}\ ] ] and the normalized queue - length in the @xmath3th client station is defined as @xmath241t-\lambda_j(0)\int_0^t r(s)\mbox{d}s\right).\ ] ] let us now discuss case ( ii ) . this case is also described by system of equations , and the dynamic of the process @xmath242 is therefore similar to the case considered above . however , this case is more delicate . the client station is initially non - bottleneck , i.e. @xmath227 , and the function @xmath242 is therefore decreasing in the right side of 0 . according to the convention , the initial value of queue is asymptotically equivalent to @xmath243 , and therefore @xmath244 . then the meaning of @xmath245 is a maximally possible rate of units arriving from the server station to client station , when all of units are in the server station and the client station is empty . then the client station is absolutely non - bottleneck if @xmath246 , and it is absolutely bottleneck if @xmath247 . consider first the case of an absolutely bottleneck station , i.e. @xmath247 . in this case , from the solution given by we have @xmath248 the meaning of the last result is the following . let @xmath249 be an asymptotic value of the queue - length in the client station , when this queue station ( at the first time ) becomes a locally bottleneck station . then , according to the normalized queue - length in the client station approaches to this level @xmath250 as @xmath251 . notice , that the same level for normalized queue - length is achieved for an initially bottleneck station in case ( i ) . the result of is also supported by results and of proposition [ prop1 ] . thus , for a bottleneck station , the same level is asymptotically achieved independently of an initial queue - length . for this reason for any absolutely bottleneck station in which @xmath247 we do not distinguish between two cases @xmath252 and @xmath253 , and absolutely bottleneck client station is always a bottleneck station . consider now the case of an absolutely non - bottleneck client station ( and therefore non - bottleneck client station ) where @xmath246 . then , according to the same calculation as in we have @xmath254 therefore , there exists the time instant @xmath255 when the normalized queue - length becomes at the first time empty . for this time instant we have the equation @xmath256 let us extend the result of case ( ii ) for a network with @xmath0 initially non - bottleneck client stations , all satisfying the condition @xmath252 , @xmath11 . it is assumed that the initial number of units in the @xmath3th client station is asymptotically equivalent to @xmath236 ( @xmath233 + @xmath234 + + @xmath235 ) as @xmath162 . assume also that the first @xmath194 client stations are ( absolutely ) non - bottleneck , i.e. @xmath257 = @xmath103(0)(1@xmath258 ) @xmath259 , @xmath196 , while the rest @xmath260 client stations are ( absolutely ) bottleneck , i.e. @xmath261 = @xmath262(0)(1@xmath263 ) @xmath264 , @xmath265 . similarly to we have the following system of equations @xmath266 and for @xmath267 we have the solution @xmath268\right).\ ] ] then for @xmath204 , @xmath11 , we have the following solutions : @xmath269 ( recall that @xmath270 , @xmath11 . ) however , since the first @xmath194 client stations are non - bottleneck , then the equality @xmath271 for the normalized queue - lengths in client stations is valid only for the values @xmath24 of the interval @xmath272 , where the value @xmath255 can be found from as @xmath273 let @xmath274 = @xmath275 . then the normalized queue - length process @xmath276 is as follows . for @xmath272 , @xmath277 , and for @xmath278 it satisfies the equation @xmath279\mbox{d}s,\ ] ] where @xmath280 . together with for all remaining @xmath3 = 1,2 , , @xmath281 , @xmath282 , , @xmath0 and @xmath278 we have : @xmath283 and similarly to for @xmath267 we have the solution @xmath284\right).\ ] ] therefore for @xmath3 = 1,2 , , @xmath274 - 1 , @xmath274 + 1 , , @xmath0 and @xmath278 we obtain @xmath285 and since @xmath286 is nonnegative , for @xmath287 and @xmath278 we obtain @xmath288 thus the dimension of the system is decreased by 1 , and the procedure can be repeated similarly . specifically , again since the client stations @xmath3 = 1,2 , , @xmath281 , @xmath282 , , @xmath194 all are non - bottleneck stations , then the equality @xmath271 for the normalized queue - length processes is valid for all @xmath289 , where @xmath290 is defined as @xmath291 setting now @xmath292 one can continue this procedure to find @xmath293 , , @xmath294 and then to know the behaviour of the queue - length processes in all of ( non - bottleneck and absolutely bottleneck ) client stations for all @xmath24 . the considered extension of case ( ii ) is in fact the general case in which amongst @xmath0 client stations there are @xmath194 non - bottleneck client stations , and the rest @xmath260 stations are absolutely bottleneck ( i.e. part of them can be initially non - bottleneck ) , and all @xmath0 client stations are with arbitrarily large initial queue - lengths . the result can be formulated as follows . [ thm1 ] assume that there are @xmath0 client stations , where @xmath194 client stations ( non necessarily the first ones ) are absolutely non - bottleneck , and the initial normalized queue - length in all of these @xmath0 client stations are @xmath232 in limit as @xmath162 correspondingly ( @xmath11 ) . then there are time instants @xmath295 when the normalized queue - lengths in these @xmath194 absolutely non - bottleneck stations correspondingly achieves zero and remains then to stay at zero . these points as well as the normalized queue - lengths in time @xmath24 in the rest @xmath260 stations are defined according to scheme [ alg1 ] below . [ alg1 ] consider the system @xmath296 where @xmath297 , and @xmath298\right).\ ] ] then , @xmath299 and the argument @xmath24 in and belongs to the interval [ 0,@xmath255 ] . in this case the normalized queue - length at the @xmath3th client station , @xmath271 for all @xmath3=1,2, ,@xmath0 . let @xmath274 = @xmath301 . then @xmath276=0 for all @xmath278 . in the next step , we consider the system of @xmath302 equations , and @xmath278 ( the equation for @xmath286 , @xmath278 , is excluded ) : @xmath303 where @xmath304\right).\ ] ] then , @xmath305 and the argument @xmath24 in and belongs to the interval [ @xmath255,@xmath290 ] , and subscript @xmath3 in these equations is @xmath307 . in this case @xmath271 , and @xmath308 for all @xmath309 . consequently excluding equations one - by - one , then the @xmath111th instant @xmath310 , @xmath311 , is defined from the system @xmath312 where @xmath313\right ) . \end{aligned}\ ] ] then , @xmath314 and the argument @xmath24 in and belongs to the interval [ @xmath315,@xmath310 ] , and subscript @xmath3 in these equations takes the values from 1 to @xmath0 but @xmath274 , @xmath317 , , @xmath318 . in this case the normalized queue - lengths @xmath271 , and @xmath319=0 for all @xmath320 . theorem [ thm1 ] containing scheme [ alg1 ] is easily applied to networks with piece - wise deterministic environment . specifically , the theorem is applied to each of the intervals [ @xmath183 ) , @xmath321 and [ @xmath184 ) . the statement of theorem [ thm1 ] is easily adapted to client / server networks in semi - markov environment . the difference between the approach to the piece - wise deterministic environment and semi - markov environment is only that the points @xmath322 are random , and general relation rather than must be used . for example , considering the first term of , we have @xmath323\mbox{d}s\\ & = \int_{0}^{\sigma_{1}\wedge t}\left[\lambda_j(\mathcal{e}(0))\left\{1-\sum_{i=1}^k \phi_s(x_j)\right\}-\mu_j\right]\mbox{d}s , \end{aligned}\ ] ] and the extension the above theory of section [ piece - wise deterministic ] is elementary . for example one can make the assumption about arbitrary initial conditions in client stations and arbitrary number of bottleneck stations . the same theory of section [ piece - wise deterministic ] can be developed for any interval with random lower and upper bounds as well , such that @xmath324\mbox{d}s,\ ] ] resulting in adaptation of the theory of section [ piece - wise deterministic ] to client / server networks in semi - markov environment . the theorem below is the adaptation of theorem [ thm1 ] to the case of a random interval [ @xmath325 , @xmath326 ) . ( the extension of theorem [ thm1 ] to the other intervals [ @xmath327 , @xmath328 ) , @xmath111=1,2 , , is similar . ) [ thm2 ] assume that amongst @xmath0 client stations , there are @xmath194 absolutely non - bottleneck in time 0 , and the rest @xmath260 client stations are absolutely bottleneck in time 0 . assume that the initial normalized queue - lengths in all of these @xmath0 client stations are @xmath232 in limit as @xmath162 correspondingly ( @xmath11 ) . then there are time instants @xmath295 which are defined recurrently by scheme [ alg1 ] of theorem [ thm1 ] . we only take into account the values @xmath329 satisfying the inequality @xmath330 , i.e. we set @xmath331 = @xmath332 . then , the only instants @xmath333 @xmath334 @xmath335 are taken into account in this theorem . the main relation of scheme [ alg1 ] now looks @xmath336 where @xmath297 , and @xmath337\right).\ ] ] the other relations of this theorem are defined similarly to the corresponding relations of scheme [ alg1 ] where only argument @xmath24 is replaced by @xmath338 in the corresponding places . specifically , now looks @xmath339 where @xmath267 is defined by , and now looks @xmath340 where @xmath267 is defined by , and @xmath341 . in this section we discuss application of the above main theorem for symmetric large closed client server computer networks . we consider a network with @xmath0 identical servers . we assume that all of the processes started at zero , i.e. @xmath342 , and the following other assumptions and notation are used . the service time of each unit in the server station is exponentially distributed with parameter @xmath18 . there are @xmath0 client stations in total , and each of client stations is a subject to breakdown . a lifetime of each client station has the probability distribution @xmath343 . in this case the moments of breakdown are associated with change states of semi - markov environment , and this example is one of applications of the above theory . the process @xmath344 satisfies the condition @xmath345 ( the equivalent form of condition is considered in section [ definitions ] . it is assumed here that @xmath346 . ) the relations between parameters @xmath18 , @xmath347 and @xmath0 are assumed to be @xmath348 and @xmath349 condition means that all of the client stations are initially non - bottleneck . condition means that after one of other breakdown all of the client stations become bottleneck . denote @xmath350 the maximum number of available client stations when the client stations all are bottleneck . then for all @xmath351 the rest @xmath111 client stations will be bottleneck as well . let @xmath105 be a given positive number . we say that the network is _ at risk _ if the total number of units in client stations exceeds the value @xmath104 . assuming that at the initial time moment all of the units are in the server stations , the aim of this section is to find a confidence interval [ 0 , @xmath352 ) such that with a given high probability @xmath353 ( say @xmath354 ) the network will not be at risk during that time interval [ 0 , @xmath352 ) . the parameter @xmath1 is assumed to be large . therefore we study asymptotic solution of this problem as @xmath162 , that is we study limiting ( as @xmath162 ) normalized cumulative queue - length process in client stations . this problem is hard in general . here we consider a particular case of @xmath355 . the limiting normalized queue - length process is denoted @xmath197 . at the initial time moment @xmath37 there are @xmath0 available client stations . let @xmath255 , @xmath290, ,@xmath356 be the moments of their breakdown , @xmath357 . the above moments of breakdown are associated with the behavior of the time dependent network , which can be considered as a network in semi - markov environment . let us then apply theorems [ thm1 ] and [ thm2 ] . ( last theorem [ thm2 ] is used consequently from one random interval to another . here we do not explain the details of this application assuming that they are clear . ) the random time interval [ 0 , @xmath356 ] is the lifetime of the entire system . therefore @xmath197 is to be considered during the aforementioned random interval [ 0 , @xmath356 ] . we assume that @xmath355 . therefore , according to theorem [ thm1 ] and theorem [ thm2 ] we obtain that in the random interval [ 0 , @xmath359 ) , @xmath360 , while in the random interval [ @xmath359 , @xmath356 ) the equation for @xmath197 is @xmath361 where @xmath362 is given by @xmath363 in the last endpoint @xmath356 we set @xmath364 we have the following relationships : @xmath365^i[g(t)]^{k - i},\ ] ] @xmath366\sum_{i=1}^{k-1}\binom{k-1}{i}[1-g(t - t_\gamma)]^i[g(t - t_\gamma)]^{k - i-1 } , \end{aligned}\ ] ] where @xmath367 is such the value of @xmath24 under which @xmath368 the value @xmath367 is found from the relation @xmath369\sum_{i=1}^{k-1}\binom{k-1}{i}[1-g(t - t_\gamma)]^i[g(t - t_\gamma)]^{k - i-1}\mbox{d}t } { \int_0^\infty\sum_{i=2}^{k}\binom{k}{i}[1-g(t - t_\gamma)]^{i}[g(t - t_\gamma)]^{k - i}\mbox{d}t}=p,\ ] ] and then from one can find the corresponding value @xmath370 . if the value of @xmath370 is not greater than @xmath371 , then the value @xmath352 of the interval [ @xmath359 , @xmath352 ) is to be taken @xmath372 . otherwise , if @xmath373 , then the value @xmath352 is to be taken @xmath374 . in the particular case of @xmath375 we have the following results . relations , and reduces correspondingly to @xmath376 ^ 2,\ ] ] @xmath377[1-g(t - t_\gamma ) ] , \end{aligned}\ ] ] and @xmath378[1-g(t - t_\gamma)]\mbox{d}t } { \int_0^\infty[1-g(t - t_\gamma)]^2\mbox{d}t}=p.\ ] ] we consider the following example for @xmath375 : @xmath379 , @xmath380 , @xmath381 , @xmath354 , @xmath382 . from we have : @xmath383 solution of the equation @xmath384 yields @xmath385 . from we obtain : @xmath386 this value of @xmath370 is less than @xmath381 , and therefore this value @xmath387 is a required value for parameter , which defines a desired confidence interval . theorem [ thm2 ] looks very complicated , and its further analysis is very difficult to make a conclusion on the behavior of queue - length processes . for example , it seems very difficult to obtain any numerical characteristics of normalized queue - length processes analytically , @xmath388 for example . therefore , the numerical work should be based on simulation of semi - markov environment in order to obtain required performance characteristics of the process . ( by simulation of semi - markov environment we mean a multiple realizations for a markov process in order to calculate required numerical performance characteristics of the process . ) for the purpose of performance analysis we also should restrict the class of networks and processes describing the behavior of queue - length processes in client stations . this restriction is related to application of the results rather than development of the theory . in many practical examples the quality characteristics of networks are changed monotonically resulting in one or other strategy of repair mentioned in section [ motivation ] . the aforementioned comparing of two different strategies for fixed interval ( 0 , @xmath36 ) requires application of theorem [ thm2 ] , and the problem can be solved without any additional assumption requiring monotonicity . however , under general settings we can not answer to many significant questions . one of them is _ how behave this criteria when the considered time interval is changed ? _ for example , we have two strategies corresponding two different initial conditions of semi - markov environment , and suppose we concluded that the first strategy is more profitable than the second one for specific time interval ( 0 , @xmath36 ) . _ is this conclusion remains correct ( or becomes not correct ) for another time interval ( 0 , @xmath389 ) ? _ another typical question is as follows . again , we have two strategies corresponding two different initial conditions of semi - markov environment . suppose we established that for an interval ( 0 , @xmath36 ) the both strategies are equivalent . let @xmath390 be a new time instant , and @xmath391 . _ which one of the strategies is now more profitable in the new time interval ( 0 , @xmath390 ) , the first or second one ? _ the same question can be asked under the opposite inequality @xmath392 . these questions can be answered in the case when the class of the processes studied numerically has a monotone stricture and is described by the properties listed below . then in certain cases the behavior of queue - length processes in client stations and consequently a conclusion about better strategy for other time intervals can be established as well . assuming for simplicity that the environment is _ markov _ , then aforementioned properties are as follows . \(1 ) for any two positive integers @xmath393 assume that @xmath394 . recall that @xmath395 + @xmath396 ( @xmath78 ) are the transition probabilities from the state @xmath76 to the state @xmath77 of a homogeneous markov process for a small time interval ( @xmath24 , @xmath79 ) . \(2 ) @xmath397 for all @xmath11 , and @xmath393 . property ( 1 ) means that the markov process @xmath51 is an increasing process in the following sense : for two time moments @xmath398 and @xmath399 we have @xmath400 , which means that the state of a markov process in time @xmath398 is not greater ( in stochastic sense ) than that state in time @xmath399 for any integer positive @xmath111 . the above property remains correct for any @xmath401 , i.e. @xmath402 . property ( 1 ) also means that for two markov processes @xmath403 and @xmath404 having the same transition probabilities , but different initial conditions satisfying @xmath405 , we also have @xmath406 , @xmath407 . ( for details of the proof of these properties see e.g. kalmykov @xcite . ) consequently , from property ( 2 ) we have @xmath408 for all @xmath11 and any integer positive @xmath111 . moreover , for all @xmath11 and any @xmath401 we have @xmath409 . thus the rates @xmath91 are increasing in time . as a result , the queue - length processes in client stations increase sharper than in the case of fixed @xmath103 of usual " network , and more extended problems mentioned in this section seem can be solved as well . we however do not provide their solutions in the present paper . in the next section , numerical investigation for concrete client server networks in a given markov environment , satisfying the above two properties is provided . in this section we do not intend to challenge a problem of comparing two different strategies or finding an optimal strategy . we only show ( step - by - step ) how to study the behavior of queue - lengths in client stations numerically . however , the detailed explanations of the given example can help to solve some of the aforementioned problems of section [ motivation ] . the example models a markov environment , i.e. in our example we simulate exponentially distributed random variables describing the state changes in the markov environment . ( in the given case by simulation we mean one realization of the process in order to study numerically a specific sample path of normalized queue - length process . ) we consider the simplest case of markov transition matrix of the order 4 @xmath410 associated with the continuous markov process @xmath51 . in this matrix @xmath411 , @xmath412 , and @xmath413=1 . this means that if the initial state of the process is @xmath414 , then the next state is @xmath415 . consequently , @xmath416 , and @xmath417 . then @xmath418 for all @xmath419 . assume also that @xmath420 , @xmath412 , so that @xmath421 = 1 , @xmath111=1,2,3 . for simplicity , the network contains only 2 client stations . assume that @xmath422=1 , @xmath423=2 , and for @xmath111=3,4 , @xmath424=@xmath425 . we also assume that @xmath424 = @xmath426 for @xmath427 . the values @xmath428 = @xmath429 = 2 . @xmath233 = @xmath234 = 0.1 , i.e. at the initial time moment each client station contains 10% of all units in the queue . we set @xmath36=3 , and study behavior of queue - length processes in client station in the time interval ( 0 , 3 ) . by simulation we obtained the following exponentially distributed random variables : 0.5488 , 1.0892 and 1.8734 . the sum of these 3 random variables is greater than 3 , so this quantity of exponentially distributed random variables is enough for our experiment . note , that @xmath430 . there three time intervals : [ 0 , 0.5488 ) , [ 0.5488 , 1.6380 ) , [ 1.6380 , 3 ) . for the time interval [ 0 , 0.5488 ) the two client stations are absolutely non - bottleneck because @xmath431=0.5 , and we have the following equations : @xmath432 therefore , from we obtain : @xmath433 substituting 0.5488 for @xmath24 in one can see that @xmath434 the endpoints are negative , therefore we are to find such the values @xmath255 and @xmath290 such that @xmath435=0 and @xmath436=0 . in our case @xmath255 = @xmath437 . this means that @xmath438 for all @xmath439 of the given interval [ 0 , 0.5488 ) . therefore , in the endpoint of this interval @xmath440 . consider now the time interval [ 0.5488 , 1.6380 ) . in point 0.5488 we now set @xmath441=0 . therefore , @xmath442 = @xmath443 = 2 , and the both client stations are bottleneck in [ 0.5488 , 1.6380 ) . since @xmath444=1 , @xmath445 , then @xmath242 and @xmath446 are equal to zero in this interval , and @xmath230 = @xmath447 = 0 in this interval as well . we arrive at the last time interval [ 1.6380 , 3 ) . similarly to the above , we have @xmath448=0 , and @xmath449=3 , and the both client stations are bottleneck in [ 1.6380 , 3 ) . however , in the both client stations we have @xmath450=1.5 , @xmath445 . therefore , after a little algebra we have the following equations : @xmath451 for all @xmath24 from the interval [ 1.6380 , 3 ) . in the present paper we introduced a class of client / server networks in order to study performance measures of real client / server networks . our analysis was based on the results of earlier papers related to closed queueing networks with bottleneck . however , for purpose of real applications , we developed the earlier results and provided complete analysis of standard bottleneck client / server networks . we then extended our results for client / server networks in semi - markov environment . the results obtained in this paper are then used for analysis of confidence intervals of client / server networks with failing client stations . numerical study given in this paper will help to clearly understand solution for many related problems . the future work can be related to application of the theoretical results of this paper to concrete technological problems similar to those formulated in section [ motivation ] . the research was supported by the australian research council , grant # dp0771338 . krichagina , e.v . , liptser , r.sh . and puhalskii , a.a . diffusion approximation for a system that an arrival stream depends on queue and with arbitrary service . _ theory of probability and its applications _ , 33 , 114 - 124 .

we report new detailed surface and photometry of 81 stellar structures in the disk of the barred galaxy ngc 3367 . the images show many different structures indicating that star formation is going on in the most part of the disk . ngc 3367 is known to have a very high concentration of molecular gas distribution in the central regions of the galaxy and bipolar synchrotron emission from the nucleus with two lobes ( at 6 kpc ) forming a triple structure similar to a radio galaxy . we have determined the and magnitudes and and colors for the central region ( nucleus ) , a region which includes supernovae 2003 aa , and 79 star associations throughout ngc 3367 . estimation of ages of star associations is very difficult due to several factors , among them : filling factor , metallicity , spatial distribution of each structure and the fact that we estimated the magnitudes with a circular aperture of 16 pixels in diameter , equivalent to kpc . however , if the colors derived for ngc 3367 were similar to the colors expected of star clusters with theoretical evolutionary star tracks developed for the lmc and had a similar metallicity , ngc 3367 show 51% of the observed structures with age type swb i ( few tens of myrs ) , with seven sources outside the bright surface brightness visible disk of ngc 3367 .

0704.1450

it is a continuous topic of study how a galaxy evolves with time @xcite . integrated photometry of star clusters and star associations in a galaxy can provide fundamental insight on the star - formation history @xcite . @xmath5 colors of galaxies depend mainly on their history of star formation @xcite . the fact that morphologically peculiar galaxies show bluer @xmath6 colors in the two color diagram suggests that galaxies with peculiar appearance have experienced anomalous star formation histories characterized by recent burst . models with recent bursts of star formation ( sfr ) predict colors that in general lie off the normal relation for models with monotonically decreasing sfr @xcite , for example , a very recent strong burst results in an excessively blue @xmath6 for a given @xmath7 @xcite . integrated photometry of star clusters and star associations in a galaxy would provide fundamental insight on the star - formation history @xcite . ngc 3367 , the disk galaxy with a stellar bar of our study , is noteworthy because of observational characteristics , some of them are : * 1 ) * it shows a bipolar synchrotron outflow from the nucleus with a compact unresolved source with deconvolved diameter less than 65 pc , and two large lobes straddling the nucleus with a total , projected extent on the sky , from the north - east to the south - west lobe of about 12 kpc ( an image like a radio galaxy , but from , a nearby late disk galaxy ) ; only the south - west lobe shows polarized emission , suggesting that this lobe is out of the plane of the disk and is closer to the observer than the other lobe @xcite ; * 2 ) * the axis of the ejected outflow is highly inclined with respect to the axis of rotation of the disk of the galaxy @xcite ; * 3 ) * there is a lot of molecular gas mostly concentrated in the central 9@xmath8 with an unresolved source with m(h@xmath9 m@xmath10 ; this molecular mass suggests an optical extinction value a@xmath11 magnitudes toward the nucleus @xcite ( as it has been detected through its co(j=1 - 0 ) emission with the ovro mm array@xmath12 ) ; its total molecular ( h@xmath13 ) mass within 11.4 kpc diameter is m(h@xmath14 m@xmath10 @xcite ; * 4 ) * it shows a bright southwest optical structure ( resembling a `` bow shock '' ) with a collection of h@xmath15 knots at a galactocentric radius of about 10 kpc from the nucleus @xcite , * 5 ) * it has _ crooked _ arms in its northwest side , and * 6 ) * it seems to be isolated with no similar - diameter galaxy in its proximity ; ngc 3367 lies , behind the leo group of galaxies , at a distance of 43.6 mpc , with a closest companion lying at a projected distance of @xmath16 563 kpc or 18 optical diameters away @xcite . ngc 3367 is an sbc(s ) barred spiral galaxy with a stellar bar structure of diameter @xmath17 kpc oriented at a position angle ( p.a . ) @xmath18 . the disk is inclined with respect to the plane of the sky at @xmath19 @xcite . ngc3367 has an x ray luminosity much larger than any normal spiral galaxy @xcite . its radio continuum emission at a @xmath20 angular resolution shows weak radio continuum emission extended throughout the disk @xcite . from fabry perot observations of the h@xmath15 emission it is deduced that the kinematic major axis lies projected on the disk at a p.a . of 51@xmath21 @xcite . its optical spectrum shows moderately broad h@xmath15+[nii ] lines with fwhm@xmath22 km s@xmath23 , but its emission of h@xmath24 is stronger than its emission of [ oiii]@xmath25 and there is weak emission of he ii @xmath26 suggesting the existence of wr stars @xcite . ngc 3367 has been clasiffied as an hii nucleus ( from the optical line ratios ; see @xcite and normal content of atomic hydrogen with m@xmath27 m@xmath10 @xcite . table 1 lists relevant information for ngc 3367 coming from the literature . previous @xmath28 phometry of ngc 3367 has been obtained for the whole galaxy with magnitudes reported in rc3 @xcite and compiled in the nasa extragalactic data base ( ned)@xmath29 as listed in table 1 ; previously reported colors are as follows : @xmath30 = 1.45 . larson & tinsley ( 1978 ) have discussed different factors that would affect the @xmath5 colors , among them are _ reddening _ ( galactic and intrinsic ) , _ gaseous emission lines and nonthermal emission _ , _ chemical composition _ and _ initial mass function _ ( imf ) . the color @xmath7 of ngc 3367 agrees with the color expected from a late type disk galaxy @xcite ; however the @xmath6 color of ngc 3367 certainly lie above the _ mean _ value of normal disk galaxies and the position in the @xmath6 vs @xmath7 @xcite is more in agreement with the statistical colors of galaxies from the arp atlas of peculiar galaxies @xmath31 . these colors of ngc 3367 suggest that this late type barred galaxy is currently having many regions with recent star formation . although galactic reddening is a major source of uncertainity in the @xmath5 colors , for this galaxy it is not a major concern given its lalitude , @xmath32 . we report our magnitude estimates for the different structures corrected for galactic extinction . gaseous emission lines , different chemical composition and different initial mass function may contribute to the uncertainities of the expected colors @xcite . this paper reports new @xmath33 photometry of 81 individual structures throughout ngc 3367 using the 84 cm optical telescope in san pedro mrtir , baja california , mxico @xcite . for the data reduction and analysis we used the iraf@xmath34 task _ phot _ with an aperture radius of 8 pixels ( @xmath35 pc ) . the existence of a close connection between violent galactic dynamical phenomena is consistent with theoretical expectations that high velocity collisions and shock fronts should be effective in compressing gas to high densities and triggering rapid star formation . section 2 reports the observations , 3 reports the results and 4 gives the discussion and finally 5 describes the conclusions . the observations were carried out using the 84 cm optical telescope at the observatorio astronmico nacional at san pedro mrtir , baja california , mxico on 2005 , march 8 and 9 . we used a ccd camera site3 with a 1024@xmath36 pixels ; each pixel size is equivalent to @xmath37 and we did not use any binning . attached to the camera was a filter box . a journal of the photometric observations is given in table 2 . column ( 1 ) give a characteristic ; columns ( 2)-(6 ) give the filters ; rows indicate the data for each column , seeing , central wavelength and total width of the filter used , and the number of frames times the integration time ( in seconds ) of each of the images ( m67 and ngc 3367 ) in each filter . images were debiased , trimmed , and flat - fielded using standard iraf procedures . first , the bias level of the ccd was subtracted from all exposures . a run of 10 bias images was obtained per night , and these were combined into a single bias frame which was then applied to the object frames . the images were flat - fielded using sky flats taken in each filter at the beginning and at the end of each night . in the case of estimating the magnitudes for the structures throughout the disk in ngc 3367 , we have made careful analysis utilizing different iraf tasks , in particular , _ qphot _ and _ phot _ using different aperture diameters for bright stars in the field of ngc 3367 in order to estimate the uncertainities in the final magnitudes at different filters before estimating the final magnitudes reported for the different structures in ngc 3367 . the magnitudes reported here for the structures within ngc 3367 were estimated using an aperture radius of 8 pixels ( @xmath38 pc ) . photometric calibration was achieved by nightly observations of standard stars of known magnitudes from the `` dipper asterism '' m67 star cluster @xcite . a total of 9 standard stars with a color interval @xmath39 and a similar range in @xmath40 were observed . the main extinction coefficients in @xmath41 , @xmath42 , @xmath43 and @xmath1 as well as the color terms were calculated according to the following equations @xcite : @xmath44 @xmath45 @xmath46 @xmath47 @xmath48 where @xmath41 , @xmath42 , @xmath43 and @xmath1 will be the estimated magnitudes , @xmath49 , @xmath50 , @xmath51 and @xmath52 are the instrumental ( and airmass - corrected ) magnitudes . @xmath15 and @xmath24 are the transformation coefficients for each filter . in a first iteration , a constant value associated with the sky background was subtracted from the images by using an interactive procedure that allows the user to select regions on the frame free of galaxies and bright stars . notice that uncertainities in determining the sky background may be the dominant source of uncertainities in the estimation of the magnitudes , color and surface brightness profiles . the most energetic cosmic - ray events were automatically masked using the iraf task _ cosmicrays _ and field stars outside the bright visible disk of ngc 3367 were removed using the iraf task _ imedit _ when necessary . within the galaxy itself , care was taken to identify superposed stars . a final step in the basic reduction involved registration of all available frames for each galaxy and in each filter to within @xmath53 pixel . this step was performed by measuring centroids for foreground stars on the images and then performing geometric transformations using iraf tasks _ geomap _ and _ geotran_. all final images were convolved with a two dimensional gaussian function in such a way as to have a similar final resolution equivalent to the largest one given by the @xmath42 filter seeing , that is , full width at half maximum , fwhm@xmath54 . an estimation of the uncertainities in our photometry involves two parts : 1 ) the procedures to obtain instrumental magnitudes and 2 ) the uncertainty when such instrumental magnitudes are transformed to a standard system . for 1 ) , notice that the magnitudes produced at the output of the iraf routines _ qphot _ and _ phot _ differ from each other as a result of different methods used in those procedures . since we also have applied extinction corrections to the instrumental magnitudes in this step , our estimation of the uncertainities are mainly concerned with these corrections and the estimation of the airmass . after a least square fitting , the associated uncertainities to the slope for each principal extinction coefficient are ; @xmath55 , @xmath56 , @xmath57 , @xmath58 and @xmath59 . an additional uncertainity @xmath60 from the airmass routines in iraf was also considered . for 2 ) , the zero point and first order color terms are the most important to consider . after transforming to the standard system , by adopting our best - fit coefficients , the formal uncertainities from the assumed relations for @xmath15 were 0.05 , 0.05 , 0.04 , 0.04 and 0.04 in @xmath61 , @xmath41 , @xmath42 , @xmath43 and @xmath1 and 0.04 , 0.03 , 0.03 and 0.04 for @xmath24 . to estimate the total uncertainity in each band , it is necessary to use the transformation equations and then propagate the errors . total typical uncertainties are 0.1 , 0.1 , 0.1 , 0.09 and 0.1 in @xmath61 , @xmath41 , @xmath42 , @xmath43 and @xmath1 bands , respectively . the estimated total magnitudes in this work were compared against other estimations reported in the literature . this has been done for the standard stars and for ngc 3367 in common with other works . our estimated ccd magnitudes for the nine stars in m67 are listed in table 3 along with magnitudes and cross reference identification numbers reported in the literature @xcite . a plot of our ccd estimated magnitudes versus those reported by guilliland et al . , ( 1991 ) and chevalier & ilovaisky ( 1991 ) are shown in figure 1 with no significant deviations , a continuous line with slope 1 is also plotted . these results suggest , a @xmath62 , as the typical magnitude uncertainity . this is in agreement with our prior uncertainity estimations . in this study , we have estimated total magnitudes for @xmath33 computed from different circular apertures chosen interactively to assure that they are large enough to contain the whole galaxy and still small enough to limit the uncertainities due to bad estimation of the flux from the background and any field stars . finally we chose an aperture with a diameter of 140@xmath8 equivalent to 2@xmath63 ; this is slightly larger than the 2@xmath64 major diameter as given in ned . there is , however , a foreground star within the disk of ngc 3367 which we did not remove before estimating the global magnitudes and colors for ngc 3367 . the star is object 82 in tables 7 and 9 and it lies about 29@xmath8 to the north - west of the nucleus ; we did not attempt to remove it because it is surrounded with emission from stellar structures in the disk and thus it makes it difficult to estimate the intensity level to replace the star with ; star i d 82 turns out to be a weak star with magnitudes @xmath65 . total magnitudes from ngc 3367 were estimated in each band by using the _ phot _ routines in iraf . our final estimated magnitudes for ngc 3367 were corrected for galactic extinction ( but not for intrinsic extinction ) ; they are @xmath66 , @xmath67 , @xmath68 , @xmath69 , and @xmath70 . the colors of ngc 3367 are @xmath71 = 1.48 . [ color color diagrams ] since the aperture to estimate the magnitudes for the stellar structures throughout ngc 3367 must be small in order to include only star associations or small structures , it was decided to use an aperture with radius 8 pixels equivalent to @xmath72 pc and use iraf task _ phot_. one compared the estimated magnitudes , with an aperture of 8 pixels , for the bright stars in the field south west of ngc 3367 with those listed in table 5 ( with aperture radius of 26 pixels ) . the magnitude differences vary according to the star and the filter . for star i d 86 the difference in empirical magnitudes are @xmath73 , @xmath74 , @xmath75 , @xmath76 , and @xmath77 ; while for star i d 91 the differences in magnitudes are @xmath78 , @xmath79 , @xmath80 , @xmath81 , and @xmath82 , fair is to say that the magnitudes with aperture radius of 8 pixels have been corrected for galactic extinction . magnitudes differences for each filter for both stars are similar indicating indeed that using a smaller aperture radius gives an artificial weaker object by the amount given in each filter but the differences are within the values for the uncertainities of our method . the estimated magnitudes for the different structures throughout ngc 3367 ( stellar associations ) can be taken as reliable and are listed in table 6 . table 7 lists the estimated magnitudes for the bright stars and a background galaxy in the field of ngc 3367 . table 8 lists the different colors @xmath6 , @xmath7 and @xmath3 for the stellar structures throughout ngc 3367 , while table 9 lists the colors of the stars and the small galaxy in the field of ngc 3367 . see tables 6 for comments on individual objects . our estimated magnitudes for the whole galaxy in @xmath61 , @xmath41 , @xmath42 and @xmath1 bands and those reported in the rc3 catalogue @xcite are in good agreement . the galactic extinction towards ngc 3367 in the different filters are : a@xmath83 , a@xmath84 , a@xmath85 , a@xmath86 , and a@xmath87 @xcite . it is a complicated problem to carry out photometry of stellar structures located within the disk of a galaxy , partly because of the strongly varying background , and partly because the structures are not perfect point sources ( in other words , there must be a varying filling factor ) nor circles ( as it was our aperture ) . for disk galaxies , the colors @xmath6 , @xmath7 , @xmath3 are closely related to spectral type and show that there exist remarkable differences in stellar content @xcite . the colors of structures in a given galaxy are most certainly affected by internal reddening and chemical enrichment . in a particular galaxy , as in our study , the colors of different structures would indicate the contributions of the component stars in star clusters , very young star clusters embedded in hii region or compact hii regions , stellar associations without gas , associations embedded in hii regions and star clusters and associations with traces of emission [ e.g. @xcite ] . it was decided in our photometric study of structures in ngc 3367 to estimate the color magnitudes of different structures using iraf task _ phot _ with an aperture with radius 8 pixels ( 3@xmath89 ) which translates , in ngc 3367 , to a radius with linear scale of 715 pc . no aperture correction were applied to our estimated magnitudes of the different structures . the photometry of all structures was obtained with the iraf task _ phot _ which takes an area of the sky , well off the extention of the disk of the galaxy in study , as a background to be compared with . in order to have an estimation of how good were our derived magnitudes for the stellar structures throughout ngc 3367 , we first estimated the magnitudes of bright field stars to the south west of ngc 3367 ( objets i d s 85 and 90 in table 7 ) utilizing two iraf tasks , _ qphot _ , and _ phot _ ; each of these stars is relatively isolated and thus we could use apertures with different radii in order to estimate their magnitudes in the five filters . table 4 gives our estimated magnitudes for the different filters for two bright stars in the south west of the field of ngc 3367 utilizing iraf task _ qphot _ using an aperture radius of 23 pixels , equivalent to @xmath90 , and a width of ring ( in order to estimate the brightness of the sky ) of pixels , equivalent to width of ring of @xmath91 . additionally iraf task _ phot _ was used with three different circular apertures , one with a radius of 20 pixels , equivalent to @xmath92 , other with radius 23 pixels , equivalent to @xmath90 and another one with 26 pixels , equivalent to @xmath93 . the estimated magnitudes for the five filters , @xmath61 , @xmath41 , @xmath42 , @xmath43 , and @xmath1 , are listed in table 5 . the largest uncertainity observed in table 5 using three different apertures amounts to @xmath94 mag in @xmath61 , @xmath41 , @xmath42 , and @xmath43 and it is @xmath95 mag in @xmath1 for both stars . comparison between estimated magnitudes using iraf tasks _ qphot _ and _ phot _ for these field stars may be estimated comparing the results listed in tables 4 and 5 . the difference amounts in the worst case to 0.12 mag indicating that indeed iraf task _ phot _ obtains a brighter value in each filter with the largest aperture ; however the difference in magnitudes using different apertures with iraf task _ phot _ is at worst @xmath96 in the @xmath1 filter between using an aperture s radius of 20 pixels and an aperture s radius of 26 pixels . the structures were chosen from the images in the @xmath1 and @xmath61 filters by eye at several locations of the disk of the galaxy with emphasis in choosing regions at different positions within ngc 3367 , for example : regions at larger distances than the bright south - west semicircle ( more than 10 kpc from the center , easily observed in images from the palomar sky surveys ) , from the stellar bar s ends , from structures along and on spiral arms , from structures just beyond the stellar bar , etc . in order to have enough of them as to be able to compare their colors and estimate their ages . it included structures on the outside of the intense semicircle structure with bright surface brightness to the west of the disk . table 6 lists the positions , @xmath15(j2000.0 ) and @xmath97(j2000.0 ) @xmath98 of the different structures throughout ngc 3367 , where we have estimated their magnitudes in the @xmath99 and @xmath1 broadband filters _ corrected _ for galactic extinction . table 8 is similar to table 6 and lists the colors @xmath6 , @xmath7 , and @xmath3 corrected for galactic extinction of each stellar structure in ngc 3367 along with their estimated swb type . we have estimated the colors of the area in front of the nucleus ; the estimated magnitudes were not corrected for intrinsic extinction ; a visual extinction has been estimated , from blue spectroscopy , to be only a@xmath100 @xcite . true magnitudes must be brighter since it is known that there is a lot of molecular gas in the central region of ngc 3367 @xcite . at the distance of ngc 3367 ( 43.6 mpc ) the absolute magnitude is m@xmath101 = -13.24 taking into account the galactic optical extinction . as a comparison , the compact nucleus in the barred spiral galaxy ngc 4314 , a liner with a circumnuclear structure @xcite , has the following magnitude and colors within an aperture of 6@xmath8 , 290 pc : @xmath102 , @xmath103 , and @xmath104 @xcite , m@xmath101 = -10.93 ( at 10 mpc taking into account the galactic optical extinction ) ; the compact nucleus of ngc 4314 is reported to have a blue magnitude of @xmath105 mag per square arc second @xcite . figure 2 shows the @xmath6 versus @xmath7 color color diagram for the different structures throughout the disk of ngc 3367 . as a comparison , figure 3 , shows the corresponding @xmath6 versus @xmath7 diagram for the large magellanic cloud @xcite stellar associations embedded in hii regions ( na in the nomenclature of bica et al . ) , star clusters ( c ) , very young clusters embedded in hii region or compact hii region ( nc ) , stellar associations without gas ( a ) and a supernovae remanent . previuos @xmath5 photometry of the smc and lmc star clusters indicated that young clusters in the smc were concentrated along its bar , while old clusters showed no concentration to the bar @xcite ; while the young clusters in the lmc were in and around the bar @xcite . the difference between the color color diagrams between the clusters in the magellanic clouds and the structures in ngc 3367 is noteworthy : the lmc has many very young clusters with colors equivalent to the swb 0 type @xcite @xmath106 ( see section on ages below ) and might also be the result of filling factor ( radius of the aperture used , since the lmc is much closer than ngc 3367 ) , intrinsic extinction and different chemical enrichment in ngc 3367 . this section shows various images of ngc 3367 . figure 4 is the image in the @xmath61 filter in a grey scale and contours , proportional to surface brightness , showing the different structures throughout ngc 3367 . figures 5 , 6 , 7 , and 8 show the four different quadrants of ngc 3367 with a circle indicating the position of the structure observed and its number , according to table 6 , with the circle having the diameter of the aperture utilized for the photometry , that is @xmath107 which at the distance of ngc 3367 translates into a linear scale of @xmath108 kpc . figure 9 shows the @xmath1 image of the galaxy with a larger area in order to indicate the field stars and the background spiral galaxy . figures 10 , 11 and 12 show the @xmath6 , @xmath7 and @xmath3 color images of ngc 3367 . there are 42 stellar structures with age type swb i @xmath109 out of 81 observed which correspond to 51% . a natural question is , where is their spatial location ? structure i d 1 corresponds to the central innermost stellar association which includes the nucleus of the galaxy ; i d 2 and 3 are two stellar associations embedded in bright hii regions on the western bright surface brightness edge of the disk at about 10 kpc from the center ; i d 4 and 5 lie on an inner eastern bright surface brightness area which seems to be a spiral arm or part of an inner ring ( just beyond the edge of the stellar bar ) at about 4.5 kpc from the center ; i d 6 and 48 lie on another structure on an eastern spiral arm , parallel to the arm where i d 4 and 5 lie , at about 8 kpc from the center ; i d 7 and 8 lie on the inner most spiral arm north - west of the center ; i d 14 lies on the crooked spiral arm north west of the center at about 6 kpc ; i d 18 lies at about 10 kpc east of the center just outside of the second spiral arm ; i d 19 - 22 , 25 , 27 all lie on the southern spiral arm ; i d 30 lies on a circular surface brightness structure south - west of the south - west end of the stellar bar ; i d 31 - 35 and 40 all lie on the western spiral arm ; i d 43 lies on an interarm on the eastern region of the disk at about 8.5 kpc ; i d 57 lies on an interarm region west of the nucleus at about 8 kpc . structures i d s 52 - 55 , 58 , 59 , 62 - 74 , and 80 lie outside the bright surface brightness semicircle at distances more than about 13 kpc . the features are detected and shown in figs . 4 , 5 , 6 , 7 , and 8 . it is important to emphasize that in the central region ( i d 1 ) , the western sources ( i d 4 - 6 ) and the eastern sources ( i d 2 , 3 ) , there is a large velocity dispersion of ionized gas , detected through h@xmath15 @xcite and in the molecular gas co @xcite . in summary , most of the structures with age type swb i lie close or on spiral arms . perhaps the most important result of the spatial distribution of structures with age type swb i is that several lie outside of the eastern bright surface brightness disk , i d 58 and 59 to the south , i d 65 , 66 , 69 , 70 , to the east while i d 53 , 71 - 74 lie to the western side and i d 75 lies well outside the disk to the north - eastern end . the colors of structures i d 58 , 59 , 71 - 74 and 75 suggest that they are newly formed stellar structures and yet they lie at distances beyond the bright visible disk of ngc 3367 , that is beyond 10 kpc from the center . it is difficult to speak about ages of the different structures throughout ngc 3367 mainly because , as we have seen , different structures could indicate star associations embbeded in hii regions , very young star cluster embedded in hii regions , star associations not associated with gas , star associations in or around the nucleus , and other combinations . among some of the problems with estimating the magnitudes of different structures are : filling factor , spatial distribution of structure within the aperture , and the fact that we used a circular aperture . we realize that with our circular aperture diameter , of @xmath4 kpc , for estimating magnitudes and colors one can not detect individual star clusters in this galaxy . however , the study of star clusters with @xmath0 and @xmath1 photometry and colors , and their derived ages in the lmc @xcite might be taken as an excellent reference to compare the @xmath0 and @xmath1 colors of the different structures throughout ngc 3367 and _ estimate _ their ages if only one were to assume that the colors were equivalent , the chemical abundance were similar , have the same metallicity as in the lmc ( which of course there is no reason to be an absolute valid assumption ) and more important we might _ use _ the swb ages if one ignores the internal reddening . i d 42 , which includes supernova 2003aa , ( asterisk in figs . 2 and 4 ) has colors similar to swb ii . structures i d 52 and i d 53 , beyond the bright western high surface brightness rim of ngc 3367 ( filled triangles in fig . 2 ) have equivalent colors to clusters swb ivb and v. i d 37 and i d 38 are the structures north - east and south - west respectively of the stellar bar ( filled squares in fig . 2 ) have colors similar to swb vi , indicating old stellar populations , but this result needs to be taken with caution since there the colors have not been corrected for intrinsic extinction as a result of the probable existence of molecular gas ( co , h@xmath13 ) and dust at least to the southwest side of the bar @xcite . figure 12 shows the histogram of the number of structure throughout ngc 3367 versus the color @xmath7 where there is a large number of structures with blue colors , @xmath110 . figure 13 shows the histogram of age groups according to swb types ( assuming that the models for star clusters in the lmc also apply to the structures and angular resolution observed in ngc 3367 ) where 64% of the structures observed have age types smaller or equal than swb type ii or less than 70 myrs . figs . 13 and 14 suggest that a large number of structures throughout ngc 3367 are relatively blue and young structures . this number of young structures would probably increase as one were able to somehow correct for intrinsic extinction , and different metallicity . to conclude , we have obtained several images in the broadband filters @xmath0 and @xmath1 in order to estimate the optical magnitudes and colors of 81 different stellar structures throughout the disk of ngc 3367 . the structures were chosen from the images in the @xmath1 and @xmath61 filters by eye at different locations of the disk of the galaxy in order to have enough of them as to be able to compare their colors as a function of their spatial location . it included structures to the north - east , south and west , well outside of the semicircle structure with bright surface brightness . our estimated magnitudes for the ngc 3367 , after correcting for galactic extinction are @xmath66 , @xmath67,@xmath68 , @xmath69 , and @xmath70 . ngc 3367 global colors , after being corrected for galactic extinction are @xmath71 = 1.48 . all images except the image in @xmath42 were convolved with an appropiate two dimensional gaussian function in order to have a final of 2@xmath111 full width at half maximum , since the seeing was different for each filter and was about 1@xmath112 to @xmath113 . the 81 stellar structures reported in this paper were selected mainly from the @xmath61 and the @xmath1 images based on the location in the disk in order to have as many as possible from different positions in ngc 3367 . we have made careful analysis of the calibration methods to estimate magnitudes and have compared different iraf tasks , in particular , _ qphot _ and _ phot _ utilizing different apertures in order to estimate the uncertainities in the final magnitudes at different filters with 9 stars in m 67 as our standards . in the case of estimating the magnitudes for the structures throughout the disk in ngc 3367 , we have made careful analysis utilizing different iraf tasks , in particular , _ qphot _ and _ phot _ using different aperture diameters for the bright stars in the field of ngc 3367 in order to estimate the uncertainities in the final magnitudes at different filters before estimating the final magnitudes reported for the different structures in ngc 3367 . the magnitudes reported here for the structures in ngc 3367 were estimated using an aperture radius of 8 pixels ( @xmath38 pc ) . we were able to estimate the optical magnitudes @xmath0 and @xmath1 and colors @xmath114 and @xmath3 for 81 different stellar structures throughout the disk in ngc 3367 . it is difficult to estimated the ages of the structures in ngc 3367 given our circular aperture diameter , equivalent to @xmath4 kpc . however if their colors were taken as representative as the colors of star clusters ( studied by many other authors with better angular resolution , appropiate chemical abundance and current theoretical star evolutionary tracks ) in the lmc many have colors equivalent to cluster types swb i , and swb ii suggesting that many are young associations with only tens of millions of years of age . we found 41 structures out of 81 ( 50% ) with age type swb i. several of those structures lie near or in spiral arms ; some lie outside of the bright surface brightness visible disk ( for example the poss image ) . in order to understand better the large scale environment of ngc 3367 it would also be very important to determine the large scale spatial distribution and kinematics of atomic hydrogen ( hi ) and new observations with high angular resolution of the emission of the molecular gas ( co , h@xmath13 ) in the center of ngc 3367 . jagb and hmht thank the staff of the observatorio astronomico nacional at san pedro mrtir , baja california , mxico for the help in the observations . hmht acknowledges support from grant conacyt ( mxico ) 42810 . jagb acknowledges partial financial support from dgapa ( unam , mxico ) grant in107806 - 2 . the authors would like to thank the anonimous referee for his ( her ) comments and suggestions on how to improve this paper . allen , c.w . 1976 , astrophysical quantities , 3rd . edition ( london : the athalone press ) anders , p. & fritze - v . alvensleben , u. 2003 , , 401 , 1063 benedict , g. f. 1980 , , 85 , 513 bica , e. , clara , j. j. , & dottori , h. , 1992 , , 103 , 1859 bica , e. , claria , j. j. , dottori , h. , santos , j. f. c. , jr . , & piatti , a. e. 1996 , 102 , 57 chevalier , c. , & ilovaisky , s.a . , 1991 , , 90,225 chiosi , c. , bertelli , g. , & bressan , a. 1988 , , 196 , 84 cid fernandes , r. , gonzlez delgado , r. m. , storchi - bergmann , t. , martins , l. pires & schmitt , h. 2005 , , 356 , 270 condon , j. j. , cotton , w. d. , greisen , e. w. , yin , q. f. , perley , r. a. , taylor , g. b. & broderick , j. j. 1998 , , 115 , 1693 fabbiano , g. , kim , d .- w . & trinchieri , g. 1992 , , 80 , 531 filippenko , a. v. , & chornock , r. 2003 , iau circular 8084 garca - barreto , j.a . , carrillo , r. & vera - villamizar , n. 2003 , , 126 , 1707 garca - barreto , j.a . , downes , d. , combes , f. , gerin , m. , magri , c. , carrasco , l. & cruz - gonzalez , i. 1991 , , 244 , 257 garca - barreto , j.a . , franco , j. & carrillo , r. 1996a , , 469 , 138 garca - barreto , j.a . , franco , j. , carrillo , r. , venegas , s. & escalante - ramirez , b. 1996b , rev.mexicana astron . astrofis . , 32 , 89 garca - barreto , j.a . & rosado , m. , 2001 , , 121 , 2540 garca - barreto , j.a . , rudnick , l. , franco , j. & martos , m. 1998 , , 116 , 111 garca - barreto , j.a . , franco , j. & rudnick , l. 2002 , , 123 , 1913 garca - barreto , j.a . , scoville , n.z . , koda , j. & sheth , k. 2005 , , 129 , 125 giogia , i.m . , maccacaro , t. schild , r.e . , wolter , a. stocke , j.t . , morris , s.l . & henry , j.p . 1990 , , 72 , 567 girardi , l. , & bica , e. 1993 , , 274 , 279 grosbl , p. j. 1985 , , 60 , 261 guilliland , r.l . , brown , t.m . , duncan , d.k . , suntzeff , n. b. , lockwood , g.w . , thompson , d.t . , schild , r.e . , jeffrey , w.a . & penprase , b.e . 1991 , , 101 , 541 ho , l.c . , filippenko , a.v . , & sargent , w. l. w. 1997a , , 112 , 315 ho , l.c . , filippenko , a.v . , sargent , w. l. w. & peng , c.y . 1997b , , 112 , 391 huchtmeier , w.k . & seiradakis , j.h . 1985 , , 143 , 216 larsen , s.s . , 1999 , , 139 , 393 larson , r. b. , 1977 , american scientist , 65 , 188 larson , r.b . , & tinsley , b.m . 1978 , , 219 , 46 lynds , b , t. , furenlid , i. & rubin , j. 1973 , , 182 , 659 maloney , p. 1990 in the insterstellar medium in galaxies , ed . thronson jr . & j.m . schull ( dordrecht : kluwer academic publishers ) , p. 493 mihalas , d. & binney , j. 1981 galactic astronomy , ( san francisco : w.h . freeman and co. ) monet , d. g. , levine , s. e. , canzian , b. , ables , h. d. , bird , a. r. , dahn , c. c guetter , h. h , et al . 2003 , , 125 , 984 pfenniger , d. 2000 , in asp conf . s. 197 xvth iap meeting dynamics of galaxies : from the early universe to the present , ed . f. combes , g. a. mamon & v. charmandaris , p 413 richter , m. , watson , a. , montes , g. & benitez - de - la - mora , g. 2005 , annual report on ccd imaging at the oan - spm , instituto de astronomia , ensenada , b.c . mexico sandage , a. 1961 , the hubble atlas of galaxies ( washington : carnegie institution of washington ) schlegel , d. j. , finkbeiner , d. p. & davis , m. 1998 , , 500 , 525 schweizer , f. , miller , b. w. , whitmore , b. c. & fall , s. m. 1996 , , 112 , 183 searle , l. , sargent , w. l. w. , & bagnuolo , w. g. 1973 , , 179 , 427 searle , l. , wilkinson , a. , & bagnuolo , w. g. 1980 , , 239 , 80 stocke , j.t . , morris , s.l . , giogia , i. m. , maccacaro , t. schild , r. , wolter , a. fleming , t.a . , & henry , j.p . 1991 , , 76 , 813 swift , b. , pugh , h. & li , w. 2003 , iau circular 8063a tully , r.b . 1988 , nearby galaxy catalog , ( cambridge : cambridge univ . press ) van den bergh , s. 1981 , , 46 , 79 van der kruit 1989 in the world of galaxies , ed . h.g . corwin jr . & l. bottinelli , ( new york : springer verlag ) p. 256 vaucouleurs , g. de , vaucouleurs , a. , corwin , h. g. , buta , r. , fouque , r. , & paturel , g. , 1993 , third reference catalog of galaxies , rc3 , ( new york : springer verlag ) . vron - cetty , m .- , & vron , p. 1986 , , 66 , 335 vron , p. , gonalves , a.c . & vron - cetty , m .- , 1997 , , 319 , 52 cccccc & @xmath61 & @xmath41 & @xmath42 & @xmath43 & @xmath1 seeing & @xmath115 & @xmath116 & @xmath54 & @xmath117 & @xmath118 @xmath119 ( ) & 600 & 950 & 1400 & 400 & 1600 @xmath120 ( ) & 3540 & 4330 & 5750 & 6340 & 8040 m67 & 8@xmath121 & 8@xmath121 & 8@xmath122 & 8@xmath123 & 8@xmath123 ngc 3367 & 5@xmath124 & 5@xmath124 & 5@xmath124 & 8@xmath125 & 8@xmath125 [ tabjournal ] lllccccc + i d & @xmath15(j2000.0 ) & @xmath97(j2000.0 ) & u & b & v & r & i 85 & 10@xmath126 & + 13@xmath127 & 13.91 & 13.51 & 12.87 & 12.47 & 12.2290 & 10@xmath128 & + 13@xmath129 & 13.13 & 13.00 & 12.68 & 12.37 & 12.12 rrrrrrrr 1 & 34@xmath130 & 45@xmath131 02@xmath132 & 14.84 & 15.39 & 14.92 & 14.39 & 13.92 + 2 & [email protected] & 45@xmath131 00@xmath134 & 16.49 & 17.28 & 17.05 & 16.82 & 16.45 + 3 & [email protected] & 44@xmath131 54@xmath135 & 16.45 & 17.24 & 17.00 & 16.79 & 16.43 + 4 & [email protected] & 45@xmath131 09@xmath111 & 16.65 & 16.91 & 16.47 & 16.03 & 15.50 + 5 & [email protected] & 45@xmath131 05@xmath134 & 16.64 & 16.92 & 16.47 & 16.02 & 15.48 + 6 & [email protected] & 45@xmath131 11@xmath112 & 16.89 & 17.31 & 17.10 & 16.79 & 16.31 + 7 & [email protected] & 45@xmath131 16@xmath136 & 16.98 & 17.27 & 16.84 & 16.42 & 15.82 + 8 & [email protected] & 45@xmath131 10@xmath111 & 16.73 & 17.05 & 16.62 & 16.20 & 15.62 + 9 & [email protected] & 45@xmath131 18@xmath137 & 17.09 & 17.24 & 16.79 & 16.36 & 15.84 + 10 & [email protected] & 45@xmath131 15@xmath136 & 17.16 & 17.24 & 16.72 & 16.26 & 15.67 + 11 & [email protected] & 45@xmath131 22@xmath111 & 17.58 & 17.71 & 17.24 & 16.77 & 16.18 + 12 & [email protected] & 45@xmath131 28@xmath89 & 17.42 & 17.63 & 17.22 & 16.83 & 16.27 + 13 & [email protected] & 45@xmath131 28@xmath134 & 17.56 & 17.72 & 17.31 & 16.91 & 16.36 + 14 & [email protected] & 45@xmath131 20@xmath132 & 17.51 & 17.83 & 17.48 & 17.08 & 16.51 + 15 & [email protected] & 45@xmath131 13@xmath111 & 17.82 & 17.93 & 17.49 & 17.06 & 16.43 + 16 & [email protected] & 45@xmath131 05@xmath111 & 17.75 & 18.02 & 17.68 & 17.30 & 16.71 + 17 & [email protected] & 44@xmath131 39@xmath138 & 17.22 & 17.49 & 17.09 & 16.69 & 16.14 + 18 & [email protected] & 45@xmath131 03@xmath136 & 18.58 & 19.09 & 18.70 & 18.29 & 17.64 + 19 & [email protected] & 44@xmath131 29@xmath132 & 18.05 & 18.53 & 18.33 & 17.93 & 17.48 + 20 & [email protected] & 44@xmath131 26@xmath134 & 17.52 & 17.90 & 17.49 & 17.08 & 16.58 + 21 & [email protected] & 44@xmath131 27@xmath135 & 17.76 & 18.02 & 17.57 & 17.15 & 16.60 + 22 & [email protected] & 44@xmath131 24@xmath136 & 17.32 & 17.83 & 17.47 & 17.07 & 16.63 + 23 & [email protected] & 44@xmath131 23@xmath137 & 17.42 & 17.66 & 17.32 & 16.95 & 16.43 + 24 & [email protected] & 44@xmath131 23@xmath112 & 17.38 & 17.62 & 17.30 & 16.93 & 16.42 + 25 & [email protected] & 44@xmath131 25@xmath138 & 17.57 & 17.84 & 17.43 & 17.06 & 16.50 + 26 & [email protected] & 44@xmath131 26@xmath137 & 17.66 & 17.94 & 17.59 & 17.22 & 16.68 + 27 & [email protected] & 44@xmath131 38@xmath137 & 17.35 & 17.62 & 17.25 & 16.87 & 16.28 + 28 & [email protected] & 44@xmath131 44@xmath139 & 17.54 & 17.67 & 17.24 & 16.81 & 16.21 + 29 & [email protected] & 44@xmath140 & 17.45 & 17.59 & 17.15 & 16.75 & 16.12 + 30 & 32@xmath141 & 44@xmath142 & 17.53 & 17.88 & 17.50 & 17.15 & 16.54 + 31 & 31@xmath143 & 45@xmath144 & 17.78 & 18.26 & 17.99 & 17.66 & 17.10 + 32 & 32@xmath145 & 45@xmath146 & 17.45 & 17.89 & 17.58 & 17.27 & 16.79 + 33 & [email protected] & 45@xmath147 & 17.43 & 17.87 & 17.58 & 17.27 & 16.80 + 34 & [email protected] & 45@xmath148 & 17.47 & 17.91 & 17.68 & 17.36 & 16.88 + 35 & [email protected] & 45@xmath149 & 17.49 & 18.04 & 17.78 & 17.46 & 17.02 + 36 & [email protected] & 45@xmath150 & 17.61 & 17.57 & 16.91 & 16.43 & 15.76 + 37 & [email protected] & 45@xmath151 & 17.61 & 17.25 & 16.40 & 15.84 & 15.20 + 38 & [email protected] & 45@xmath152 & 17.50 & 17.29 & 16.48 & 15.93 & 15.25 + 39 & [email protected] & 45@xmath153 & 18.69 & 18.96 & 18.63 & 18.24 & 17.75 + 40 & [email protected] & 45@xmath154 & 18.35 & 18.72 & 18.45 & 18.12 & 17.70 + 41 & [email protected] & 45@xmath155 & 18.16 & 18.31 & 17.96 & 17.62 & 17.12 + 42 & [email protected] & 45@xmath156 & 18.14 & 18.29 & 17.85 & 17.42 & 16.91 + 43 & [email protected] & 45@xmath157 & 17.88 & 18.16 & 17.72 & 17.28 & 16.77 + 44 & [email protected] & 45@xmath158 & 18.14 & 18.32 & 17.99 & 17.62 & 17.09 + 45 & [email protected] & 45@xmath159 & 17.73 & 17.83 & 17.45 & 17.07 & 16.52 + 46 & [email protected] & 45@xmath160 & 17.64 & 17.82 & 17.41 & 17.02 & 16.48 + 47 & [email protected] & 44@xmath161 & 17.67 & 17.62 & 17.05 & 16.59 & 15.94 + 48 & [email protected] & 44@xmath162 & 17.58 & 17.86 & 17.44 & 17.00 & 16.46 + 49 & [email protected] & 44@xmath163 & 17.82 & 17.88 & 17.37 & 16.90 & 16.31 + 50 & [email protected] & 44@xmath164 & 17.65 & 17.75 & 17.20 & 16.72 & 16.09 + 51 & [email protected] & 44@xmath165 & 17.38 & 17.49 & 16.97 & 16.52 & 15.89 + 52 & [email protected] & 44@xmath166 & 19.42 & 19.60 & 19.01 & 18.58 & 17.79 + 53 & [email protected] & 44@xmath167 & 19.21 & 19.67 & 19.05 & 18.61 & 17.81 + 54 & [email protected] & 44@xmath168 & 19.04 & 19.34 & 19.01 & 18.61 & 17.92 + 55 & [email protected] & 44@xmath169 & 19.32 & 19.56 & 19.12 & 18.66 & 18.02 + 56 & [email protected] & 44@xmath170 & 18.51 & 18.63 & 18.18 & 17.76 & 17.13 + 57 & 32@xmath171 & 44@xmath172 & 17.96 & 18.23 & 17.83 & 17.42 & 16.84 + 58 & 36@xmath173 & 44@xmath174 & 19.70 & 19.93 & 19.48 & 19.07 & 18.30 + 59 & 36@xmath175 & 44@xmath176 & 19.42 & 19.81 & 19.43 & 19.00 & 18.29 + 60 & 37@xmath177 & 45@xmath178 & 18.90 & 19.02 & 18.71 & 18.28 & 17.77 + 61 & 37@xmath179 & 45@xmath180 & 18.62 & 18.84 & 18.63 & 18.23 & 17.86 + 62 & 35@xmath181 & 46@xmath182 & 19.68 & 18.84 & 18.63 & 18.23 & 17.86 + 63 & 34@xmath183 & 46@xmath182 & 19.61 & 19.80 & 19.63 & 19.27 & 18.96 + 64 & 33@xmath184 & 46@xmath185 & 19.81 & 19.90 & 19.85 & 19.36 & 19.12 + 65 & 32@xmath186 & 45@xmath187 & 19.50 & 19.69 & 19.46 & 19.18 & 18.70 + 66 & 39@xmath188 & 44@xmath189 & 19.57 & 19.87 & 19.32 & 18.96 & 18.21 + 67 & 39@xmath183 & 45@xmath182 & 20.07 & 20.12 & 19.63 & 19.08 & 18.49 + 68 & 39@xmath190 & 45@xmath191 & 20.10 & 20.23 & 19.86 & 19.29 & 18.60 + 69 & 38@xmath192 & 45@xmath193 & 19.96 & 20.07 & 19.56 & 19.16 & 18.18 + 70 & 39@xmath194 & 44@xmath195 & 19.62 & 19.85 & 19.35 & 18.97 & 18.23 + 71 & 31@xmath194 & 44@xmath196 & 20.42 & 20.70 & 20.31 & 19.89 & 18.76 + 72 & 30@xmath197 & 44@xmath198 & 20.51 & 20.92 & 20.48 & 20.17 & 18.88 + 73 & 31@xmath199 & 44@xmath200 & 20.14 & 21.24 & 20.66 & 20.33 & 19.16 + 74 & 30@xmath201 & 45@xmath202 & 20.61 & 21.01 & 21.34 & 20.58 & 19.43 + 75 & 43@xmath203 & 46@xmath204 & 20.90 & 21.30 & 20.79 & 20.02 & 20.73 + 76 & 36@xmath205 & 45@xmath206 & 16.73 & 16.99 & 16.55 & 16.12 & 15.62 + 77 & 35@xmath207 & 45@xmath208 & 17.46 & 17.31 & 16.59 & 16.09 & 15.44 + 78 & 33@xmath209 & 44@xmath210 & 18.01 & 18.02 & 17.36 & 16.84 & 16.15 + 79 & 36@xmath171 & 45@xmath211 & 18.25 & 18.28 & 17.93 & 17.56 & 17.04 + 80 & 39@xmath199 & 45@xmath212 & 20.31 & 20.21 & 19.53 & 19.13 & 18.27 + 81 & 35@xmath199 & 45@xmath213 & 18.27 & 18.20 & 17.60 & 17.12 & 16.41 + rrrrrrrr 82 & 33@xmath194 & 45@xmath214 & 17.91 & 17.94 & 17.37 & 16.86 & 16.21 + 83 & 28@xmath181 & 45@xmath215 & 19.66 & 19.04 & 18.13 & 17.52 & 17.04 + 84 & 42@xmath216 & 44@xmath217 & 18.20 & 17.94 & 17.19 & 16.75 & 16.36 + 85 & 27@xmath181 & 43@xmath218 & 13.93 & 13.54 & 12.91 & 12.51 & 12.26 + 86 & 29@xmath219 & 44@xmath220 & 21.85 & 21.73 & 20.29 & 19.40 & 17.93 + 87 & 29@xmath221 & 43@xmath222 & 21.64 & 20.99 & 20.09 & 19.29 & 18.45 + 88 & 39@xmath223 & 44@xmath224 & 20.63 & 20.34 & 19.63 & 18.65 & 17.32 + 89 & 28@xmath225 & 42@xmath226 & 20.63 & 20.34 & 19.13 & 18.43 & 17.56 + 90 & 30@xmath175 & 42@xmath227 & 13.16 & 13.03 & 12.73 & 12.41 & 12.17 + rrrrr 1 & -0.55 & 0.47 & 1.00 & i + 2 & -0.79 & 0.23 & 0.60 & i + 3 & -0.79 & 0.24 & 0.57 & i + 4 & -0.26 & 0.44 & 0.97 & i + 5 & -0.28 & 0.45 & 0.99 & i + 6 & -0.42 & 0.21 & 0.79 & i + 7 & -0.29 & 0.43 & 1.02 & i + 8 & -0.32 & 0.43 & 1.00 & i + 9 & -0.15 & 0.45 & 0.95 & ii + 10 & -0.08 & 0.52 & 1.05 & iii + 11 & -0.13 & 0.47 & 1.06 & ii + 12 & -0.21 & 0.41 & 0.95 & ii + 13 & -0.16 & 0.41 & 0.95 & ii + 14 & -0.32 & 0.35 & 0.97 & i + 15 & -0.11 & 0.44 & 1.06 & iii + 16 & -0.27 & 0.34 & 0.97 & i + 17 & -0.27 & 0.40 & 0.95 & i + 18 & -0.51 & 0.39 & 1.05 & i + 19 & -0.48 & 0.20 & 0.84 & i + 20 & -0.38 & 0.41 & 0.91 & i + 21 & -0.26 & 0.51 & 0.97 & i + 22 & -0.51 & 0.36 & 0.84 & i + 23 & -0.24 & 0.34 & 0.89 & i + 24 & -0.24 & 0.32 & 0.88 & i + 25 & -0.27 & 0.41 & 0.93 & i + 26 & -0.28 & 0.35 & 0.91 & i + 27 & -0.27 & 0.37 & 0.97 & i + 28 & -0.13 & 0.43 & 1.03 & ii + 29 & -0.14 & 0.44 & 1.03 & ii + 30 & -0.35 & 0.38 & 0.96 & i + 31 & -0.48 & 0.27 & 0.89 & i + 32 & -0.44 & 0.31 & 0.79 & i + 33 & -0.44 & 0.29 & 0.78 & i + 34 & -0.44 & 0.23 & 0.80 & i + 35 & -0.55 & 0.26 & 0.76 & i + 36 & + 0.04 & 0.66 & 1.15 & vii + 37 & + 0.36 & 0.85 & 1.20 & vi + 38 & + 0.21 & 0.81 & 1.23 & vii + 39 & -0.27 & 0.33 & 0.88 & i + 40 & -0.37 & 0.27 & 0.75 & i + 41 & -0.15 & 0.35 & 0.84 & ii + 42 & -0.15 & 0.44 & 0.94 & ii + 43 & -0.28 & 0.44 & 0.95 & i + 44 & -0.18 & 0.33 & 0.88 & ii + 45 & -0.10 & 0.38 & 0.93 & iii + 46 & -0.18 & 0.41 & 0.93 & ii + 47 & + 0.05 & 0.57 & 1.11 & ivb + 48 & -0.28 & 0.42 & 0.98 & i + 49 & -0.06 & 0.51 & 1.06 & iva + 50 & -0.10 & 0.55 & 1.11 & iii + 51 & -0.11 & 0.52 & 1.08 & iii + 52 & -0.18 & 0.59 & 1.22 & i + 53 & -0.44 & 0.60 & 1.24 & i + 54 & -0.30 & 0.33 & 1.09 & i + 55 & -0.24 & 0.44 & 1.10 & i + 56 & -0.12 & 0.45 & 1.05 & iii + 57 & -0.27 & 0.40 & 0.99 & i + 58 & -0.23 & 0.45 & 1.18 & i + 59 & -0.39 & 0.38 & 1.14 & i + 60 & -0.12 & 0.31 & 0.94 & iii + 61 & -0.22 & 0.21 & 0.77 & ii + 62 & -0.01 & 0.17 & 0.75 & iii + 63 & -0.19 & 0.17 & 0.67 & ii + 64 & -0.09 & 0.05 & 0.73 & iii + 65 & -0.19 & 0.23 & 0.76 & ii + 66 & -0.30 & 0.55 & 1.11 & i + 67 & -0.05 & 0.49 & 1.14 & iva + 68 & -0.13 & 0.37 & 1.26 & iii + 69 & -0.11 & 0.51 & 1.38 & iii + 70 & -0.23 & 0.50 & 1.12 & i + 71 & -0.28 & 0.38 & 1.55 & i + 72 & -0.41 & 0.44 & 1.60 & i + 73 & -1.10 & 0.58 & 1.50 & i + 74 & -0.40 & -0.33 & 1.91 & ii + 75 & -0.40 & 0.51 & 0.36 & i + 76 & -0.26 & 0.44 & 0.94 & i + 77 & + 0.15 & 0.72 & 1.15 & vii + 78 & -0.01 & 0.66 & 1.21 & vii + 79 & -0.03 & 0.35 & 0.89 & iva + 80 & + 0.10 & 0.68 & 1.26 & vii + 81 & + 0.07 & 0.60 & 1.19 & ivb + rrrrr 82 & -0.03 & + 0.57 & 1.16 & f7v + 83 & + 0.62 & + 0.91 & 1.09 & k3v + 84 & + 0.26 & + 0.75 & 0.83 & g8v + 85 & + 0.39 & + 0.63 & 0.65 & g2iii + 86 & -0.88 & + 1.44 & 2.36 & k5iii + 87 & + 0.65 & + 0.90 & 1.64 & g7iii + 88 & + 0.29 & + 0.71 & 2.31 & k3iii + 89 & + 0.29 & + 1.21 & 1.57 & spiral + 90 & + 0.13 & + 0.30 & 0.56 & f1iii + +

we compute next - to - next - to - leading order qcd corrections to the gluon - induced production cross section of higgs boson pairs in the large top quark mass limit using the soft - virtual approximation . in the limit of infinitely - heavy top quark we confirm the results in the literature . we add two more expansion terms in the inverse top quark mass to the result . since the expansion converges poorly , we try to improve on it by factorizing the exact leading order cross section . we discuss two ways of doing that and conclude that the finite top quark mass effects shift the cross section at most by about 10% at next - to - leading order and by about 5% at next - to - next - to - leading order . pacs numbers : 14.80.bn , 12.38.bx , 14.65.ha

1508.00909

in the coming years , one of the main tasks in particle physics is the understanding of the mechanism of the electroweak symmetry breaking . after the experimental determination of the higgs boson mass , the higgs potential is fully fixed in the standard model . however , it is very important to independently measure the self - coupling of the higgs boson , which can be obtained from studying the production of higgs boson pairs . since the corresponding cross section is @xmath2 times smaller than the one for single higgs boson production , higgs boson pair production poses a challenging problem for the lhc , even after the luminosity upgrade around 2020 . there are a number of phenomenological analyses which investigate the possibility to extract the self coupling from cross section measurements . first studies have been performed more than 15 years ago @xcite . since the discovery of the higgs boson there has been an increasing interest in this topic and a number of refined analyses have been performed , see , e.g. , refs . @xcite . higgs boson pairs can be produced via the fusion of two partons or vector bosons , via the radiation off vector bosons , or in association with heavy quarks . similar to single higgs boson production , the numerically dominant mechanism is gluon fusion although the leading order ( lo ) contribution is loop - suppressed . due to the larger yukawa coupling , the dominant contribution comes from top quark loops in the standard model . for this reason we concentrate in this paper on such contributions . for the lo order corrections the exact dependence on the top quark mass and the center - of - mass energy is known @xcite . at next - to - leading ( nlo ) qcd corrections have been computed for the first time more than 15 years ago @xcite in the infinite top quark mass limit using an effective theory . finite top quark mass effects have been investigated in ref . @xcite where a systematic expansion in the inverse top quark mass has been applied and a quantitative estimate of the quark mass effects has been provided . it has been estimated that they do not exceed @xmath3 of the nlo contribution . finite top quark mass effects have also been considered in ref . @xcite where the exact real radiation contribution is combined with the effective - theory virtual corrections . as a result , a reduction of about @xmath4 of the cross section is obtained . we will comment in section [ sec::nlo ] on this issue . within the effective theory also next - to - next - to - leading ( nnlo ) contributions are available @xcite . in this context it is interesting to note that the three - loop matching coefficient of the effective operator for two higgs bosons and two , three or four gluons is different from the one for single higgs boson production @xcite . the results for the virtual corrections obtained in ref . @xcite have been cross - checked in ref . @xcite where the calculation has been performed without reference to the effective theory . the resummation of threshold - enhanced logarithms to next - to - next - to - leading logarithmic ( nnll ) accuracy has been performed in refs . @xcite . the remainder of this paper is organized as follows : in the next section we review the construction of the soft - virtual approximation for the production cross section and discuss two options to factorize the exact lo result from the higher order contributions . we argue that a factorization at the level of the differential cross section w.r.t . the higgs boson pair invariant mass leads to more stable results . afterwards we reconsider in section [ sec::nlo ] the top mass corrections at nlo . virtual nnlo corrections including finite top quark mass effects are computed in section [ sec::nnlo ] . they are used in section [ sec::impr_nnlo ] to present phenomenological results for higgs pair production up to nnlo . section [ sec::con ] contains our conclusions . we write the perturbative expansion of the partonic cross section for the production of higgs boson pairs in the form @xmath5 where @xmath6 , @xmath7 is the partonic center - of - mass energy and @xmath8 . since the quark - induced channels are numerically small @xcite we consider in this paper only the @xmath9 channel . we use the variable @xmath10 to parametrize the dependence of the cross section on the higgs boson and top quark mass . we renormalize the top quark mass in the on - shell scheme . furthermore , we set the factorization and renormalization scale equal to each other and write @xmath11 . for later convenience we introduce for @xmath12 @xmath13 and denote the sum of the first two terms on the right - hand side by @xmath14 . finite top quark mass effects to @xmath15 at nlo have been considered for the first time in ref . the applied method is based on `` reversed unitarity '' @xcite where , with the help of the optical theorem , the imaginary part of forward scattering amplitudes are computed to obtain the total cross section . the virtual corrections have also been computed by directly considering the @xmath15 amplitude . expansion terms up to order @xmath16 of the nlo contribution to @xmath17 have been computed @xcite . the factorization of the exact lo corrections has then been implemented at the partonic level for the total cross section using @xmath18 where @xmath19 contains the exact dependence on @xmath20 and @xmath21 . in ref . @xcite , where the nlo result has been computed for the first time in the heavy top quark limit , a different approach has been applied . actually , the exact lo result has been factorized before the integration over the higgs pair invariant mass . in this approach the ( total ) nlo cross section reads @xmath22 where exact and exp reminds whether exact or ( in @xmath21 ) expanded results are used . @xmath23 is the invariant mass squared of the higgs boson pair . expressed in terms of @xmath24 it is possible to re - write eq . ( [ eq::nlo_fact1 ] ) as @xmath25 from the comparison of eqs . ( [ eq::nlo_fact2 ] ) and ( [ eq::nlo_fact3 ] ) one expects that ( [ eq::nlo_fact2 ] ) leads to better results since the differential factorization ( df ) in eq . ( [ eq::nlo_fact2 ] ) results in a better - behaved integrand , in particular for @xmath26 . for the virtual corrections , which are proportional to @xmath27 , one has immediate access to the @xmath23 dependence and the df of eq . ( [ eq::nlo_fact2 ] ) can be applied . ) and ( [ eq::nlo_fact2 ] ) are equivalent . ] the real corrections , however , are obtained with the help of the optical theorem which directly leads to the total cross section and thus eq . ( [ eq::nlo_fact2 ] ) can not be used . for the construction of the soft - virtual ( sv ) approximation , which is discussed below , we need in addition to the virtual term also the contributions from soft gluon emission . since the soft contributions are proportional to the lo born cross section eq . ( [ eq::nlo_fact2 ] ) can immediately be applied to the sv cross section . in the following we discuss the construction of the sv approximation for the cross section ( see also ref . @xcite ) . for simplicity we consider for the following schematic reasoning the total cross section @xmath28 . note , however , that the arguments also hold for @xmath29 . in a first step we split @xmath28 according to @xmath30 where the two terms on the right - hand side correspond to the virtual corrections ( including ultra - violet counterterms ) and the real corrections ( including the contributions from the factorization of initial - state singularities ) . they are individually divergent and only the sum is finite . in the next step we re - organize the terms on the right - hand side such that @xmath28 can be written as @xmath31 where `` sv '' refers to `` soft - virtual '' and `` h '' to `` hard '' . this is achieved by splitting @xmath32 into a divergent and a finite term and separating @xmath33 into a ( divergent ) soft and ( finite ) hard contribution . the soft contribution is combined with @xmath32 to obtain @xmath34 such that @xmath35 @xmath34 and @xmath36 are separately finite . @xmath37 is constructed following ref . @xcite ; explicit expressions can be found in refs . note that the finite part is constructed solving the equation @xmath38 for @xmath39 . in this paper @xmath32 is computed including top quark mass effects . mass effects are automatically taken into account in @xmath37 and @xmath40 since these contributions are proportional to the exact lo cross section @xcite . note that eqs . ( [ eq::sig_split ] ) , ( [ eq::sig_sv_h ] ) , ( [ eq::sig_sv_h_2 ] ) and ( [ eq::sig_fin ] ) also hold for the differential cross section @xmath41 and thus a factorization as suggested in eq . ( [ eq::nlo_fact2 ] ) can be performed for the soft - virtual contribution . to be more specific we present the nlo and nnlo differential version of eq . ( [ eq::sig_fin ] ) which reads @xcite @xmath42 \frac{{\rm d}\sigma^{(0)}}{{\rm d}t } \ , , \nonumber\\ & \frac{{\rm d}\sigma_v^{(2)}}{{\rm d}t } = \frac{{\rm d}\sigma_{v,\rm fin}^{(2)}}{{\rm d}t } + 2 \mbox{re}\left[i_g^{(1)}\right ] \frac{{\rm d}\sigma_{v,\rm fin}^{(1)}}{{\rm d}t } + \left\ { \left|i_g^{(1)}\right|^2 + 2 \mbox{re}\left[\left(i_g^{(1)}\right)^2\right ] + 2 \mbox{re}\left[i_g^{(2)}\right ] \right\ } \frac{{\rm d}\sigma^{(0)}}{{\rm d}t } \ , . \label{eq::sig_fin_2}\end{aligned}\ ] ] where @xmath43 with @xmath44 ( @xmath45 ) being the incoming ( outgoing ) momentum of a gluon ( higgs boson ) , @xmath46 and explicit expressions for the operators @xmath47 can be found in ref . @xcite . we adopt the notation from ref . @xcite and parametrize radiative corrections to the partonic cross section via @xmath48 with @xmath49 and @xmath50 where the renormalization scale dependence in the strong coupling constant @xmath51 is suppressed . from eq . ( [ eq::g ] ) one obtains for the total cross section @xmath52 with @xmath53 in the second line of eq . ( [ eq::g_2 ] ) we split @xmath54 into soft - virtual and hard contribution . note that in our approach we do not have access to @xmath55 . actually , at nnlo we only have @xmath56 at hand and at nlo only the heavy top expansion of @xmath57 is available to us . explicit results for @xmath56 can be found in ref . @xcite including higher order terms in @xmath58 specifying , however , the renormalization and factorization scale to @xmath59 . for completeness we provide the nlo and nnlo results for @xmath56 for generic @xmath60 in the limit @xmath61 . the results read @xmath62 - 4c_ald_0 + 8c_ad_1 \ , , \nonumber\\ g_\textrm{sv}^{(2 ) } & = d_{-1 } \bigg\{-\frac{4\pi^2}{3}c_a^2l^2 + \frac{11\pi^2}{108}n_l^2 + { \frac{\sigma^{(2)}_\textrm{fin}}{\sigma^{\rm lo}}}+ l\bigg[-\frac{1}{3}{\frac{\sigma^{(1)}_\textrm{fin}}{\sigma^{\rm lo}}}n_l \nonumber\\ & \quad + c_a\left(\left(-\frac{1}{3 } - \frac{2\pi^2}{9}\right ) n_l + \frac{11}{6 } { \frac{\sigma^{(1)}_\textrm{fin}}{\sigma^{\rm lo}}}\right)+ c_a^2\left(\frac13 + \frac{11\pi^2}{9 } - 38\zeta_3\right)\bigg ] \nonumber\\ & \quad + c_a^2\left(\frac{607}{81 } + \frac{517\pi^2}{108 } - \frac{\pi^4}{80 } - \frac{407\zeta_3}{36}\right ) + c_a\left[\frac{2\pi^2}{3}{\frac{\sigma^{(1)}_\textrm{fin}}{\sigma^{\rm lo}}}+ n_l\left(-\frac{82}{81 } - \frac{11\pi^2}{8 } + \frac{37}{18}\zeta_3\right)\right]\bigg\}\nonumber\\ & \quad + d_0 \bigg\{l^2\left(-\frac{11}{3}c_a^2 + \frac{2}{3}c_an_l\right ) + c_an_l \left(\frac{56}{27 } - \frac{4\pi^2}{9}\right ) + l\bigg[c_a^2 \left(-\frac{134}{9 } + \frac{10\pi^2}{3}\right ) \nonumber\\ & \quad + c_a \left(\frac{20}{9}n_l - 4{\frac{\sigma^{(1)}_\textrm{fin}}{\sigma^{\rm lo}}}\right)\bigg ] + c_a^2\left(-\frac{404}{27 } + \frac{22\pi^2}{9 } + 78\zeta_3\right ) \bigg\ } + d_1 \bigg[16c_a^2l^2\nonumber\\ & \quad + l\left(\frac{44}{3}c_a^2 - \frac{8}{3 } c_an_l\right ) + c_a^2\left(\frac{268}{9 } - \frac{20\pi^2}{3}\right ) + c_a\left(-\frac{40}{9}n_l + 8{\frac{\sigma^{(1)}_\textrm{fin}}{\sigma^{\rm lo}}}\right)\bigg ] \nonumber\\ & \quad + d_2 \bigg[-\frac{44}{3}c_a^2 - 48c_a^2l + \frac{8}{3}c_an_l\bigg ] + 32c_a^2d_3 \ , , \label{eq::g_sv}\end{aligned}\ ] ] where @xmath63 are qcd colour factors , @xmath64 is the number of massless quarks , @xmath65 and @xmath66_+ \,.\end{aligned}\ ] ] the quantities @xmath67 and @xmath68 are evaluated for @xmath69 . they are obtained from @xmath70 and @xmath71 in eq . ( [ eq::sig_fin_2 ] ) after integration over @xmath72 . in the next section we apply the formalism described in this section at nlo and compare to the results obtained in ref . the numerical effects at nnlo are presented in section [ sec::impr_nnlo ] using the mass corrections computed in section [ sec::nnlo ] . in this section we restrict ourselves to nlo and compare the results of ref . @xcite with the alternative factorization discussed in the previous section . to obtain numerical results we employ mstw2008 parton distribution functions ( pdfs ) @xcite and consistently use n@xmath73lo pdfs to compute n@xmath73lo ( @xmath74 ) cross sections . we assume the energy of the lhc to be @xmath75 and adopt the values of the strong coupling constants @xmath76 that we use in our computation from the mstw pdf fit : @xmath77 we renormalize the top quark in the on - shell scheme and use @xmath78 gev @xcite . for the higgs boson we use @xmath79 gev @xcite . for numerical results shown in this section the renormalization and factorization scales have been set to @xmath80 . partonic cross section as a function the center - of - mass energy including various orders in the inverse top quark mass . the dotted and dashed curves show the breakup of the complete result ( solid lines ) into soft - virtual and hard contribution.,title="fig : " ] + in fig . [ fig::sig_part ] we show the results for the partonic cross section as a function of @xmath7 from fig . 6 of ref . @xcite , see solid lines . the splitting of these results into soft - virtual and hard contributions is shown as dotted and dashed curves , respectively . for the complete result and soft - virtual approximation the infinite - top quark result corresponds to the ( red ) second line from below . the lowest line includes in addition the @xmath81 corrections . the curves including higher order mass corrections are above the infinite - top quark result . the topmost curves include @xmath16 terms . the hard contribution shows a quite flat behaviour above @xmath82 gev and exhibits smaller shifts when including top mass corrections . furthermore , the infinite top mass result corresponds to the topmost curve and the lowest curve includes @xmath16 terms . from fig . [ fig::sig_part ] it is evident that the hard contribution is numerically much smaller than the soft - virtual one . partonic cross section as a function the center - of - mass energy including various orders in the inverse top quark mass . the dashed and solid curves correspond to the factorization for the total and differential cross section , respectively . the colour coding is taken over from fig . [ fig::sig_part].,title="fig : " ] + in fig . [ fig::sig_part_2 ] we compare the solid curves from fig . [ fig::sig_part ] ( which are dotted in this plot ) with the results obtained with the help of df applied to the sv approximation and adding the hard contribution as given by the dashed lines in fig . [ fig::sig_part ] . the relative position of the lines and the colour coding is as in fig . [ fig::sig_part ] . for lower values of @xmath7 the two approaches lead to comparable results . however , the df curves have their maximum at lower values of @xmath7 and lead to a smaller cross section for larger values of @xmath7 . furthermore , one observes a drastic improvement in the convergence behaviour when including higher order mass corrections . in particular , for @xmath83 gev the difference between the infinite top mass result and the one including @xmath16 terms amounts to only @xmath84 fb to be compared with @xmath85 fb for the dashed curves . partonic nlo @xmath86 factor for the factorization performed at the level of the total ( dashed ) and differential ( solid ) cross section . , title="fig : " ] + at this point we want to stress that the splitting between the soft - virtual and hard contribution in eq . ( [ eq::sig_sv_h ] ) is arbitrary . in fact , the soft - virtual contribution of @xmath54 , @xmath56 , is constructed for the limit @xmath87 and thus it is possible to replace @xmath56 by @xmath88 with @xmath89 ( see , e.g. , discussions in refs . the hard contribution is modified accordingly such that the sum of @xmath90 does not change . one observes that for @xmath91 the soft - virtual contribution approximates the partonic nlo contribution to the cross section very well . ] such that at the hadronic level the deviations are below @xmath92 . based on this observation we use @xmath91 for the nnlo cross section where we only have the soft - virtual approximation at hand . furthermore , better results are obtained by replacing @xmath93 in eq . ( [ eq::g_sv ] ) by @xmath94 which is justified since @xmath95 in the soft limit . note that in ref . @xcite it has been observed that the soft - virtual approximation constructed in mellin space approximates the full ( effective - theory ) result with an accuracy of 2% . it is interesting to look at the partonic @xmath86 factor which is defined via @xmath96 results for the two methods to factorize the exact lo term are plotted in fig . [ fig::sig_k_part ] as a function of @xmath7 where the dashed curves are already shown in ref . one observes that df leads to a lower @xmath86 factor and that the spread among the various @xmath21 orders is smaller . furthermore , it is interesting to note that for df the top quark pair threshold behaviour of the lo term is not washed out in contrast to the dashed curves . it is common to both factorization methods that there is a strong raise when approaching the threshold for higgs boson pair production ( see also discussion in ref . @xcite ) . nlo hadronic cross section and @xmath86 factor as a function of @xmath97.,title="fig : " ] + fig . [ fig::sig_hadr ] shows the hadronic cross section @xmath98 for higgs boson pair production including nlo corrections as a function of @xmath99 which is a technical upper cut on the partonic center - of - mass collision energy . it is introduced via @xmath100 where the luminosity function is given by @xmath101 @xmath102 are the gluon distribution functions in the @xmath103 scheme . note that in the soft limit @xmath97 is a good approximation to @xmath23 . the various lines in fig . [ fig::sig_hadr ] correspond to the inclusion of different orders in @xmath21 at nlo . for convenience we show on the right end of fig . [ fig::sig_hadr ] the total cross section for @xmath104 tev . note that the approximation used for the computation of the @xmath105 terms is not valid for large values of @xmath106 ( neither is the effective - theory result ) . however , it can be used as an estimate of the mass correction terms . using the spread as an estimate for the uncertainty we conclude that a finite top mass induces a @xmath107% uncertainty on top of the infinite top quark mass result . the lower panel of fig . [ fig::sig_hadr ] shows the hadronic @xmath86 factor which is obtaind from eq . ( [ eq::k_nlo ] ) by replacing @xmath28 by @xmath98 and using nlo pdfs in the numerator and lo pdfs in the denominator . @xmath108 raises close to threshold , however , for @xmath109 gev one observes a flat behaviour of @xmath110 ( for @xmath111 ) . splitting of partonic cross section ( solid lines ) into real ( uppser dashed lines ) and virtual ( lower dotted lines ) contributions ( see text for details).,title="fig : " ] + top quark mass effects to double higgs boson production have also been considered in ref . @xcite . in the approximation used in that reference the real corrections are treated exactly , however , the infinite top quark mass approximation is used for the virtual corrections . a decrease of the cross section by about 10% due to finite top quark mass is reported . in fig . [ fig::sig_part_real_virt ] we split poles which cancel in the sum . in fig . [ fig::sig_part_real_virt ] the finite contributions are plotted . ] the partonic results for the solid lines of fig . [ fig::sig_part ] ( which corresponds to the factorization according to eq . ( [ eq::nlo_fact1 ] ) ) into virtual corrections ( including the ultra - violet counterterms ; dotted lines ) and the real - radiation part which include the contributions from the splitting functions ( dashed lines ) . we observe that for @xmath112 gev , the region where our approximation is valid , the top quark mass corrections to the real radiation part ( upper dashed curves ) reduce the infinite top result by about 10% , in agreement with the observations of ref . @xcite . on the other hand , the top mass effects to the virtual contribution leads to a positive shift as compared to the effective theory result . summing real and virtual corrections leads to an overall positive effect from top mass corrections as can be seen by the solid curves , see also fig . [ fig::sig_part_2 ] . note that up to @xmath113 gev top mass corrections are dominated by the @xmath81 terms . in this section we compute the virtual corrections to the nnlo cross section including top quark mass corrections . afterwards we construct @xmath39 as described in section [ sec::fac ] and use eq . ( [ eq::g_sv ] ) to evaluate the partonic and hadronic cross sections including @xmath114 correction terms . we have applied two methods to compute the virtual corrections . in the first one we consider the amplitudes for @xmath15 up to three - loop order and in the second one the forward scattering amplitude is considered , which , after taking the imaginary part , directly leads to the total cross section . in both cases we perform an asymptotic expansion for large top quark mass . the first approach has the advantage that it is straightforward to introduce higgs boson decays whereas the second approach can immediately be applied to real corrections . + + + nnlo calculations require corrections up to three - loop order to the process @xmath15 . typical contributing feynman diagrams at lo , nlo and nnlo are shown in fig . [ fig::gghh ] . they are generated with the help of qgraf @xcite . note that in this approach no contributions with external ghosts have to be considered since we project on physical states . the transformation to form @xcite code is done with the program q2e @xcite and the asymptotic expansion for large top quark masses is realized with the help of exp @xcite . after expanding the identified hard subgraphs in the small quantities one arrives at one - scale vacuum integrals up to three loops and massless one- and two - loop four - point diagrams with two massless and two massive external momenta . the vacuum integrals are computed using matad @xcite . in the case of the four - point integrals we apply fire @xcite to express them as linear combinations of master integrals . the latter are shown in fig . [ fig::gghh_ml ] where we have @xmath115 , and @xmath116 . analytic results for all master integrals can be found in the literature ( see , e.g. , refs . @xcite ) . in this paper we only show results for the triangle graph in the second line of fig . [ fig::gghh_ml ] since for our purpose the representations given in refs . @xcite are less suited . we use instead @xmath117 + \mathcal{o}(\epsilon^2 ) \bigg\}\ , , { \addtocounter{equation}{1}\tag{\theequation}}\label{eq::i14}\end{aligned}\ ] ] with @xmath118 @xmath119 are goncharov polylogarithms @xcite with weight @xmath120 and argument @xmath121 defined through @xmath122 with @xmath123 and @xmath124 the functions of the @xmath125 term in eq . ( [ eq::i14 ] ) can be expressed in terms of logarithms and dilogarithms via @xmath126 we have cross checked the numerical result for the @xmath125 and @xmath127 terms of @xmath128 in eq . ( [ eq::i14 ] ) against fiesta @xcite . using this method we have computed terms up to order @xmath16 at nlo @xcite and terms up to order @xmath114 at nnlo . as an important check we have computed the @xmath81 corrections for general qcd gauge parameter which drops out in the final expression . one- and two - loop master integrals needed after applying asymptotic expansion to the amplitude @xmath15 . all internal lines are massless , @xmath115 , and @xmath116.,title="fig : " ] one- and two - loop master integrals needed after applying asymptotic expansion to the amplitude @xmath15 . all internal lines are massless , @xmath115 , and @xmath116.,title="fig : " ] one- and two - loop master integrals needed after applying asymptotic expansion to the amplitude @xmath15 . all internal lines are massless , @xmath115 , and @xmath116.,title="fig : " ] + one- and two - loop master integrals needed after applying asymptotic expansion to the amplitude @xmath15 . all internal lines are massless , @xmath115 , and @xmath116.,title="fig : " ] one- and two - loop master integrals needed after applying asymptotic expansion to the amplitude @xmath15 . all internal lines are massless , @xmath115 , and @xmath116.,title="fig : " ] one- and two - loop master integrals needed after applying asymptotic expansion to the amplitude @xmath15 . all internal lines are massless , @xmath115 , and @xmath116.,title="fig : " ] + one- and two - loop master integrals needed after applying asymptotic expansion to the amplitude @xmath15 . all internal lines are massless , @xmath115 , and @xmath116.,title="fig : " ] one- and two - loop master integrals needed after applying asymptotic expansion to the amplitude @xmath15 . all internal lines are massless , @xmath115 , and @xmath116.,title="fig : " ] one- and two - loop master integrals needed after applying asymptotic expansion to the amplitude @xmath15 . all internal lines are massless , @xmath115 , and @xmath116.,title="fig : " ] + one- and two - loop master integrals needed after applying asymptotic expansion to the amplitude @xmath15 . all internal lines are massless , @xmath115 , and @xmath116.,title="fig : " ] + from the calculation of @xmath15 one obtains in a first step results for @xmath129 . integration over phase space then leads to @xmath24 . for the results in section [ sec::impr_nnlo ] this integration is performed numerically . the second method is based on the use of the optical theorem in analogy to the nlo calculation performed in ref . this method serves as an important cross check . in the following we provide some of the technical details * the amplitudes for @xmath130 are generated with the help of qgraf @xcite . * in a first step about 17 million diagrams are generated . however , most of them do not contain a cut through exactly two higgs bosons . for this reason we post - process the qgraf output and filter @xcite the amplitudes describing the virtual corrections to @xmath15 . typical feynman diagrams are shown in fig . [ fig::gg2gg ] . * form @xcite code is then generated by passing the output via q2e @xcite , which transforms feynman diagrams into feynman amplitudes , to exp @xcite . * our in - house form code applies projectors @xmath131 for each pair of external gluons which includes also non - physical degrees of freedom in the sum . thus also contributions with ghosts in the initial state have to be considered . note that this is in contrast to single higgs boson production which has no virtual contributions with ghosts in the initial state . lo , nlo and nnlo feynman diagrams needed for the forward scattering amplitude @xmath130 . solid lines refer to top quarks , curly lines to gluons and dashed lines to higgs bosons . at nnlo only virtual contributions are shown . wavy lines denote the cuts.,title="fig : " ] lo , nlo and nnlo feynman diagrams needed for the forward scattering amplitude @xmath130 . solid lines refer to top quarks , curly lines to gluons and dashed lines to higgs bosons . at nnlo only virtual contributions are shown . wavy lines denote the cuts.,title="fig : " ] + lo , nlo and nnlo feynman diagrams needed for the forward scattering amplitude @xmath130 . solid lines refer to top quarks , curly lines to gluons and dashed lines to higgs bosons . at nnlo only virtual contributions are shown . wavy lines denote the cuts.,title="fig : " ] + lo , nlo and nnlo feynman diagrams needed for the forward scattering amplitude @xmath130 . solid lines refer to top quarks , curly lines to gluons and dashed lines to higgs bosons . at nnlo only virtual contributions are shown . wavy lines denote the cuts.,title="fig : " ] lo , nlo and nnlo feynman diagrams needed for the forward scattering amplitude @xmath130 . solid lines refer to top quarks , curly lines to gluons and dashed lines to higgs bosons . at nnlo only virtual contributions are shown . wavy lines denote the cuts.,title="fig : " ] + lo , nlo and nnlo feynman diagrams needed for the forward scattering amplitude @xmath130 . solid lines refer to top quarks , curly lines to gluons and dashed lines to higgs bosons . at nnlo only virtual contributions are shown . wavy lines denote the cuts.,title="fig : " ] lo , nlo and nnlo feynman diagrams needed for the forward scattering amplitude @xmath130 . solid lines refer to top quarks , curly lines to gluons and dashed lines to higgs bosons . at nnlo only virtual contributions are shown . wavy lines denote the cuts.,title="fig : " ] + lo , nlo and nnlo feynman diagrams needed for the forward scattering amplitude @xmath130 . solid lines refer to top quarks , curly lines to gluons and dashed lines to higgs bosons . at nnlo only virtual contributions are shown . wavy lines denote the cuts.,title="fig : " ] lo , nlo and nnlo feynman diagrams needed for the forward scattering amplitude @xmath130 . solid lines refer to top quarks , curly lines to gluons and dashed lines to higgs bosons . at nnlo only virtual contributions are shown . wavy lines denote the cuts.,title="fig : " ] + resulting feynman diagrams after shrinking the top quark loops to a point according to the rule of asymptotic expansion . the blob represents one - loop vacuum integrals.,title="fig : " ] + ( a ) ( b ) ( c ) the application of the asymptotic expansion for large top quark mass leads to a factorization of the five - loop integrals into the following contributions : 1 . four one - loop vacuum integrals times one - loop integrals which are already present at lo , see fig . [ fig::gg2gg_ae](a ) ; 2 . three one - loop vacuum integrals times two - loop integrals , see fig . [ fig::gg2gg_ae](b ) ; 3 . two one - loop vacuum integrals times three - loop integrals , see fig . [ fig::gg2gg_ae](c ) ; 4 . two - loop times one - loop vacuum integrals times two - loop integrals ; 5 . three - loop times one - loop vacuum integrals times one - loop integrals . the mass scale in the vacuum integrals is given by the top quark . they are again computed with the help of matad @xcite . for the remaining integrals , which depend on @xmath132 , we use the in - house programs rows @xcite and topoid @xcite to perform the reduction to master integrals . the latter are depicted in fig . [ fig::gg2gg_mi ] . phase space master integrals occurring in the amplitude for @xmath130 . the first line contains lo and nlo integrals . the integrals in line two and three are needed for the nnlo virtual corrections . single and double lines represent massless and higgs propagators , respectively . double lines with gray - shaded interspace ( last two diagrams in first row ) correspond to higgs boson propagators which shall not be cut . a cross marks an inverse propagator.,title="fig : " ] phase space master integrals occurring in the amplitude for @xmath130 . the first line contains lo and nlo integrals . the integrals in line two and three are needed for the nnlo virtual corrections . single and double lines represent massless and higgs propagators , respectively . double lines with gray - shaded interspace ( last two diagrams in first row ) correspond to higgs boson propagators which shall not be cut . a cross marks an inverse propagator.,title="fig : " ] phase space master integrals occurring in the amplitude for @xmath130 . the first line contains lo and nlo integrals . the integrals in line two and three are needed for the nnlo virtual corrections . single and double lines represent massless and higgs propagators , respectively . double lines with gray - shaded interspace ( last two diagrams in first row ) correspond to higgs boson propagators which shall not be cut . a cross marks an inverse propagator.,title="fig : " ] phase space master integrals occurring in the amplitude for @xmath130 . the first line contains lo and nlo integrals . the integrals in line two and three are needed for the nnlo virtual corrections . single and double lines represent massless and higgs propagators , respectively . double lines with gray - shaded interspace ( last two diagrams in first row ) correspond to higgs boson propagators which shall not be cut . a cross marks an inverse propagator.,title="fig : " ] phase space master integrals occurring in the amplitude for @xmath130 . the first line contains lo and nlo integrals . the integrals in line two and three are needed for the nnlo virtual corrections . single and double lines represent massless and higgs propagators , respectively . double lines with gray - shaded interspace ( last two diagrams in first row ) correspond to higgs boson propagators which shall not be cut . a cross marks an inverse propagator.,title="fig : " ] phase space master integrals occurring in the amplitude for @xmath130 . the first line contains lo and nlo integrals . the integrals in line two and three are needed for the nnlo virtual corrections . single and double lines represent massless and higgs propagators , respectively . double lines with gray - shaded interspace ( last two diagrams in first row ) correspond to higgs boson propagators which shall not be cut . a cross marks an inverse propagator.,title="fig : " ] + phase space master integrals occurring in the amplitude for @xmath130 . the first line contains lo and nlo integrals . the integrals in line two and three are needed for the nnlo virtual corrections . single and double lines represent massless and higgs propagators , respectively . double lines with gray - shaded interspace ( last two diagrams in first row ) correspond to higgs boson propagators which shall not be cut . a cross marks an inverse propagator.,title="fig : " ] phase space master integrals occurring in the amplitude for @xmath130 . the first line contains lo and nlo integrals . the integrals in line two and three are needed for the nnlo virtual corrections . single and double lines represent massless and higgs propagators , respectively . double lines with gray - shaded interspace ( last two diagrams in first row ) correspond to higgs boson propagators which shall not be cut . a cross marks an inverse propagator.,title="fig : " ] phase space master integrals occurring in the amplitude for @xmath130 . the first line contains lo and nlo integrals . the integrals in line two and three are needed for the nnlo virtual corrections . single and double lines represent massless and higgs propagators , respectively . double lines with gray - shaded interspace ( last two diagrams in first row ) correspond to higgs boson propagators which shall not be cut . a cross marks an inverse propagator.,title="fig : " ] phase space master integrals occurring in the amplitude for @xmath130 . the first line contains lo and nlo integrals . the integrals in line two and three are needed for the nnlo virtual corrections . single and double lines represent massless and higgs propagators , respectively . double lines with gray - shaded interspace ( last two diagrams in first row ) correspond to higgs boson propagators which shall not be cut . a cross marks an inverse propagator.,title="fig : " ] phase space master integrals occurring in the amplitude for @xmath130 . the first line contains lo and nlo integrals . the integrals in line two and three are needed for the nnlo virtual corrections . single and double lines represent massless and higgs propagators , respectively . double lines with gray - shaded interspace ( last two diagrams in first row ) correspond to higgs boson propagators which shall not be cut . a cross marks an inverse propagator.,title="fig : " ] phase space master integrals occurring in the amplitude for @xmath130 . the first line contains lo and nlo integrals . the integrals in line two and three are needed for the nnlo virtual corrections . single and double lines represent massless and higgs propagators , respectively . double lines with gray - shaded interspace ( last two diagrams in first row ) correspond to higgs boson propagators which shall not be cut . a cross marks an inverse propagator.,title="fig : " ] + phase space master integrals occurring in the amplitude for @xmath130 . the first line contains lo and nlo integrals . the integrals in line two and three are needed for the nnlo virtual corrections . single and double lines represent massless and higgs propagators , respectively . double lines with gray - shaded interspace ( last two diagrams in first row ) correspond to higgs boson propagators which shall not be cut . a cross marks an inverse propagator.,title="fig : " ] phase space master integrals occurring in the amplitude for @xmath130 . the first line contains lo and nlo integrals . the integrals in line two and three are needed for the nnlo virtual corrections . single and double lines represent massless and higgs propagators , respectively . double lines with gray - shaded interspace ( last two diagrams in first row ) correspond to higgs boson propagators which shall not be cut . a cross marks an inverse propagator.,title="fig : " ] phase space master integrals occurring in the amplitude for @xmath130 . the first line contains lo and nlo integrals . the integrals in line two and three are needed for the nnlo virtual corrections . single and double lines represent massless and higgs propagators , respectively . double lines with gray - shaded interspace ( last two diagrams in first row ) correspond to higgs boson propagators which shall not be cut . a cross marks an inverse propagator.,title="fig : " ] phase space master integrals occurring in the amplitude for @xmath130 . the first line contains lo and nlo integrals . the integrals in line two and three are needed for the nnlo virtual corrections . single and double lines represent massless and higgs propagators , respectively . double lines with gray - shaded interspace ( last two diagrams in first row ) correspond to higgs boson propagators which shall not be cut . a cross marks an inverse propagator.,title="fig : " ] phase space master integrals occurring in the amplitude for @xmath130 . the first line contains lo and nlo integrals . the integrals in line two and three are needed for the nnlo virtual corrections . single and double lines represent massless and higgs propagators , respectively . double lines with gray - shaded interspace ( last two diagrams in first row ) correspond to higgs boson propagators which shall not be cut . a cross marks an inverse propagator.,title="fig : " ] phase space master integrals occurring in the amplitude for @xmath130 . the first line contains lo and nlo integrals . the integrals in line two and three are needed for the nnlo virtual corrections . single and double lines represent massless and higgs propagators , respectively . double lines with gray - shaded interspace ( last two diagrams in first row ) correspond to higgs boson propagators which shall not be cut . a cross marks an inverse propagator.,title="fig : " ] all three - loop master integrals factorize into a two - loop form factor contribution and the one - loop master integral of the lo calculation . from the latter only the imaginary part is needed which is well known . the results for the two - loop form factor integrals can , e.g. , be found in refs . @xcite . the two - loop master integrals are more involved . a numerical calculation would probably be possible , however , we follow the approach outlined in ref . @xcite for the nlo master integrals and perform an expansion in @xmath133 [ cf . ( [ eq::delta ] ) ] . all integrals contain a massless sub - loop for which analytic expressions are known . the massless two - point function can be expressed in terms of @xmath134 functions and results for the triangle with two massless external legs and the ( crossed ) box can be found in ref . analytic expressions for the triangle diagram with squared external momenta @xmath135 are given in eq . ( [ eq::i14 ] ) . after expanding in @xmath132 the remaining phase - space integration can be performed analytically . we have computed expansion terms up to order @xmath136 and found agreement with the results obtained in the previous subsection which for this purpose have also been integrated analytically after performing the expansion in @xmath132 . the approach based on the optical theorem requires special care in the treatment of the imaginary parts originating from the two - loop form factor diagrams . such contributions either correspond to @xmath137 or @xmath138 . in the former case the two - loop integrals originate from the product of two one - loop contributions containing factors @xmath139 and @xmath140 , respectively , which finally leads to @xmath141 . in the other case one has @xmath142 . the corresponding discussion for single higgs production can be found in ref . @xcite . note that the approach based on the @xmath15 amplitudes leads to simpler intermediate expressions . thus , it is possible to allow for a general qcd gauge parameter @xmath143 when computing the @xmath81 terms . furthermore , also the @xmath114 corrections could be evaluated ( for @xmath144 ) whereas in the optical theorem approach only @xmath81 terms could be computed in feynman gauge . however , let us mention that this approach can be used in a straightforward way to compute nnlo top quark mass effects to the hard contribution of the total cross section whereas in the approach of the previous subsection this is less obvious . to conclude this section let us summarize our procedure to obtain the sv corrections at nnlo . we compute virtual corrections to @xmath15 including @xmath114 corrections . note that we have @xmath145 . using eq . ( [ eq::sig_fin ] ) we construct @xmath146 which enters @xmath147 in eq . ( [ eq::g_sv ] ) . the differential and total cross section is then obtained with the help of eqs . ( [ eq::g ] ) and ( [ eq::sigma_tot ] ) . in this section we discuss the effect of the @xmath81 and @xmath114 terms on the nnlo cross section for the production of higgs boson pairs . the cross section in the infinite top quark mass limit has been computed in ref . @xcite . in ref . @xcite the three - loop matching coefficient has been added , completing the nnlo prediction , and the virtual corrections from ref . @xcite have been cross checked . comparison of the lo , nlo and nnlo contributions to the partonic cross section . at lo the exact result is shown and at nlo and nnlo the first three terms in the large-@xmath148 expansion are shown . for all curves the nnlo - value for @xmath51 is used . for the renormalization and factorization scale we use @xmath111 . , title="fig : " ] + the results for the virtual corrections computed in section [ sec::nnlo ] are inserted in the formalism described in section [ sec::fac ] to construct the quantity @xmath146 which enters eq . ( [ eq::g_sv ] ) . the result for the partonic cross section is shown in fig . [ fig::sig_part_lo_nlo_nnlo ] as a function of the partonic center - of - mass energy where the exact lo result is compared with nlo and nnlo . at nlo and nnlo three terms in the mass expansion are shown . at nlo to have a direct comparison with nnlo . ] furthermore , at nnlo the sv approximation is shown for @xmath91 ( cf . discussion in section [ sec::nlo ] ) . note that the nnlo curves peak for smaller values of @xmath7 than at nlo and lo . as far as the top quark mass corrections are concerned the same pattern is observed as at nlo : the correction term of order @xmath21 decreases the infinite top quark mass result which is overcompensated by the @xmath149 term resulting in a small positive correction . nnlo partonic cross section for different scales and for @xmath150 ( dotted ) and @xmath91 ( solid ) . only the @xmath151 result is shown . , title="fig : " ] + in fig . [ fig::sig_part_nnlo ] we compare the nnlo - sv contribution for two different scales , @xmath111 and @xmath152 . this plot furthermore shows the effect of @xmath150 and @xmath91 . note that the choice @xmath91 , which we expect to better approximate the complete result , leads to an increase of the cross section . hadronic lo , nlo and nnlo - sv ( with @xmath91 ) cross sections as a function of @xmath97 . for their evaluation the respective value of @xmath51 is used . at lo the exact result and at nlo only the @xmath151 term is shown . at nnlo the @xmath151 , @xmath153 and @xmath149 results are plotted . the results in the right panel with `` @xmath154 '' at the bottom correspond to the prediction of the total cross section . for this plot @xmath111 has been used . , title="fig : " ] + the hadronic cross section as a function of @xmath97 is shown in fig . [ fig::sig_hadr_lo_nlo_nnlo ] for @xmath111 . at lo the exact result is used and the nlo curve is based on the @xmath151 results . using instead the @xmath155 terms leads to an upwards shift of about 5% as can be seen in fig . [ fig::sig_hadr ] . at nnlo three curves are shown which include terms up to @xmath151 , @xmath153 and @xmath149 . as at nlo one observes good convergence up to @xmath156 gev . for higher values of @xmath99 the @xmath153 curve is below and the @xmath149 curve above the infinite top quark mass result leading to a @xmath157 effect for the total cross section on the rightmost part of the plot . to be conservative , we thus estimate that the nnlo top quark mass effects lead to an uncertainty of @xmath158 . note that the nnlo corrections amount to about 20% of the lo result . hadronic nlo ( dotted ) and nnlo ( solid ) @xmath86 factor as a function of @xmath97.,title="fig : " ] + fig . [ fig::sig_k_hadr_nnlo ] shows the hadronic @xmath86 factor at nnlo which is defined by @xmath159 as a function of @xmath106 . for comparison also the nlo result from fig . [ fig::sig_hadr ] is shown as dotted curve using the @xmath0 result . one observes that the various @xmath21 orders lead to similar results for @xmath160 . furthermore , there is a strong raise close to threshold which is due to the steeper behaviour of the nnlo cross section as can be seen in the inlay of fig . [ fig::sig_part_lo_nlo_nnlo ] . for higher values of @xmath97 , in particular for the total cross section , @xmath160 approaches @xmath161 . .[tab : nnlo]total hadronic cross section at lo , nlo and nnlo - sv including top quark mass effects using @xmath162 and @xmath91 . [ cols="<,<,<,^",options="header " , ] results for the total cross section at lo , nlo and nnlo are summarized in table [ tab : nnlo ] for @xmath111 . at nlo only @xmath151 terms are included in the analysis whereas at nnlo @xmath151 , @xmath153 and @xmath149 terms are considered . in this way we can estimate the top mass effects of the nnlo term . besides the cross section also the @xmath86 factor is shown . at nnlo we use @xmath91 and we furthermore show the relative deviation due to @xmath163 terms defined through @xmath164 the two known mass correction terms lead to a change of the cross section by about @xmath165 . assuming a similar pattern as at nlo we thus estimate that nnlo top quark mass corrections change the effective - theory result by at most @xmath166 . we compute nlo and nnlo corrections to double higgs boson production in gluon fusion beyond the effective - theory approach . the starting point of the calculation are full - theory feynman diagrams . we perform an asymptotic expansion in the limit where the top quark mass is large and compute at nnlo three terms in the @xmath1 expansion for the virtual corrections . they are used to construct a soft - virtual approximation for the production cross section . in the limit @xmath0 the effective - theory result of ref . @xcite is confirmed and @xmath81 and @xmath114 terms are added . the main result of this paper is illustrated in fig . [ fig::sig_hadr_lo_nlo_nnlo ] where the hadronic cross section is shown as a function of @xmath97 ( a technical cut on the partonic center - of - mass energy ) . the curves including @xmath1 corrections deviate from the infinite mass result only by a few per cent which leads us to the estimate that the effective - theory result is accurate to @xmath166 . analog results for the mass corrections at nlo are shown in fig . [ fig::sig_hadr ] which constitutes an update of ref . @xcite . here we estimate the uncertainty to @xmath167 . we want to stress that the results obtained in this paper provide excellent approximations for small values of the partonic center - of - mass energy , say below @xmath168 gev . although in this region the cross section is small it is of interest since there the cross section has a characteristic behaviour . furthermore , it is possible to use our result in this region as a benchmark for future calculations taking into account the exact dependence on @xmath148 . the methods described in section [ sec::nnlo ] can also be used to compute top mass corrections to the real radiation part . however , the simplifications used in ref . @xcite where results have been obtained for @xmath169 do not apply once finite mass effects are considered . the calculation is much more challenging since significantly more feynman diagrams contribute and more complicated master integrals have to be computed . in this paper for the first time the effect of a finite top quark mass has been examined for the nnlo cross section for double higgs boson production . whereas at nlo an exact calculation is within reach this is certainly not the case at nnlo . thus our results become particular important once our nlo approximations are compared to an exact calculation which increases the confidence in the uncertainty estimate . furthermore , one probably can obtain a prescription to tune the approximation procedure and hence reduce the uncertainty at nnlo . we would like to thank kirill melnikov for providing to us the analytic result for @xmath128 in eq . ( [ eq::i14 ] ) , for many useful discussions and for carefully reading the manuscript . this work is supported by the deutsche forschungsgemeinschaft through grant ste 945/2 - 1 and by kit through its distinguished researcher fellowship program . parts of this work were supported by the european commission through contract pitn - ga-2012 - 316704 ( higgstools ) . j.g . would like to express a special thanks to the mainz institute for theoretical physics ( mitp ) for its hospitality and support . a. djouadi , w. kilian , m. mhlleitner and p. m. zerwas , eur . j. c * 10 * ( 1999 ) 45 [ hep - ph/9904287 ] . u. baur , t. plehn and d. l. rainwater , phys . d * 67 * ( 2003 ) 033003 [ hep - ph/0211224 ] . u. baur , t. plehn and d. l. rainwater , phys . d * 69 * ( 2004 ) 053004 [ hep - ph/0310056 ] . m. j. dolan , c. englert and m. spannowsky , jhep * 1210 * ( 2012 ) 112 [ arxiv:1206.5001 [ hep - ph ] ] . a. papaefstathiou , l. l. yang and j. zurita , phys . d * 87 * ( 2013 ) 011301 [ arxiv:1209.1489 [ hep - ph ] ] . j. baglio , a. djouadi , r. grber , m. m. mhlleitner , j. quevillon and m. spira , jhep * 1304 * ( 2013 ) 151 [ arxiv:1212.5581 [ hep - ph ] ] . f. goertz , a. papaefstathiou , l. l. yang and j. zurita , jhep * 1306 * ( 2013 ) 016 [ arxiv:1301.3492 [ hep - ph ] ] . m. gouzevitch , a. oliveira , j. rojo , r. rosenfeld , g. p. salam and v. sanz , jhep * 1307 * ( 2013 ) 148 [ arxiv:1303.6636 [ hep - ph ] ] . e. w. n. glover and j. j. van der bij , nucl . b * 309 * ( 1988 ) 282 . t. plehn , m. spira and p. m. zerwas , nucl . b * 479 * ( 1996 ) 46 [ erratum - ibid . b * 531 * ( 1998 ) 655 ] [ hep - ph/9603205 ] . s. dawson , s. dittmaier and m. spira , phys . d * 58 * , 115012 ( 1998 ) [ hep - ph/9805244 ] . j. grigo , j. hoff , k. melnikov and m. steinhauser , nucl . b * 875 * ( 2013 ) 1 [ arxiv:1305.7340 [ hep - ph ] ] . f. maltoni , e. vryonidou and m. zaro , jhep * 1411 * ( 2014 ) 079 [ arxiv:1408.6542 [ hep - ph ] ] . d. de florian and j. mazzitelli , phys . b * 724 * ( 2013 ) 306 [ arxiv:1305.5206 [ hep - ph ] ] . d. de florian and j. mazzitelli , phys . rev . 111 , * 201801 * ( 2013 ) [ arxiv:1309.6594 [ hep - ph ] ] . j. grigo , k. melnikov and m. steinhauser , nucl . b * 888 * ( 2014 ) 17 [ arxiv:1408.2422 [ hep - ph ] ] . d. y. shao , c. s. li , h. t. li and j. wang , arxiv:1301.1245 [ hep - ph ] . d. de florian and j. mazzitelli , arxiv:1505.07122 [ hep - ph ] . c. anastasiou and k. melnikov , nucl . phys . b * 646 * , 220 ( 2002 ) [ hep - ph/0207004 ] . j. grigo , j. hoff , k. melnikov and m. steinhauser , pos radcor * 2013 * ( 2013 ) 006 [ arxiv:1311.7425 [ hep - ph ] ] . j. grigo and j. hoff , arxiv:1407.1617 [ hep - ph ] . d. de florian and j. mazzitelli , jhep * 1212 * ( 2012 ) 088 [ arxiv:1209.0673 [ hep - ph ] ] . s. catani , phys . b * 427 * ( 1998 ) 161 [ hep - ph/9802439 ] . a. d. martin , w. j. stirling , r. s. thorne and g. watt , eur . j. c * 63 * ( 2009 ) 189 [ arxiv:0901.0002 [ hep - ph ] ] . k. a. olive _ et al . _ [ particle data group collaboration ] , chin . c * 38 * ( 2014 ) 090001 . g. aad _ et al . _ [ atlas and cms collaborations ] , phys . * 114 * ( 2015 ) 191803 [ arxiv:1503.07589 [ hep - ex ] ] . c. anastasiou , c. duhr , f. dulat , e. furlan , t. gehrmann , f. herzog and b. mistlberger , phys . b * 737 * ( 2014 ) 325 [ arxiv:1403.4616 [ hep - ph ] ] . f. herzog and b. mistlberger , arxiv:1405.5685 [ hep - ph ] . p. nogueira , j. comput . * 105 * ( 1993 ) 279 . j. kuipers , t. ueda , j. a. m. vermaseren and j. vollinga , comput . commun . * 184 * ( 2013 ) 1453 [ arxiv:1203.6543 [ cs.sc ] ] . r. harlander , t. seidensticker and m. steinhauser , phys . b * 426 * ( 1998 ) 125 [ hep - ph/9712228 ] . t. seidensticker , hep - ph/9905298 . m. steinhauser , comput . commun . * 134 * ( 2001 ) 335 [ hep - ph/0009029 ] . a. v. smirnov and v. a. smirnov , arxiv:1302.5885 [ hep - ph ] . a. v. smirnov , comput . commun . * 189 * ( 2014 ) 182 [ arxiv:1408.2372 [ hep - ph ] ] . t. g. birthwright , e. w. n. glover and p. marquard , jhep * 0409 * ( 2004 ) 042 [ hep - ph/0407343 ] . r. k. ellis and g. zanderighi , jhep * 0802 * ( 2008 ) 002 [ arxiv:0712.1851 [ hep - ph ] ] . f. chavez and c. duhr , jhep * 1211 * ( 2012 ) 114 [ arxiv:1209.2722 [ hep - ph ] ] . a. b. goncharov , math . lett . * 5 * ( 1998 ) 497 [ arxiv:1105.2076 [ math.ag ] ] . a. v. smirnov , comput . commun . * 185 * ( 2014 ) 2090 [ arxiv:1312.3186 [ hep - ph ] ] . j. hoff , `` methods for multiloop calculations and higgs boson production at the lhc '' , dissertation , kit , 2015 . j. a. m. vermaseren , math - ph/0010025 . t. gehrmann , t. huber and d. maitre , phys . b * 622 * ( 2005 ) 295 [ hep - ph/0507061 ] . a. pak , m. rogal and m. steinhauser , jhep * 1109 * ( 2011 ) 088 [ arxiv:1107.3391 [ hep - ph ] ] .

in a rindler - type coordinate system spanned in a region outside of a black hole horizon , we have nonvanishing classical holographic charges as soft hairs on the horizon for stationary black holes . taking a large black hole mass limit , the spacetimes with the charges are described by asymptotic rindler metrics . we construct a general theory of gravitational holographic charges for a ( 1 + 3)-dimensional linearized gravity field in the minkowski background with rindler horizons . although matter crossing a rindler horizon causes horizon deformation and a time - dependent coordinate shift , that is , gravitational memory , the supertranslation and superrotation charges on the horizon can be defined during and after its passage through the horizon . it is generally proven that holographic states on the horizon can not store any information about absorbed perturbative gravitational waves . however , matter crossing the horizon really excites holographic states . by using gravitational memory operators , which consist of the holographic charge operators , we suggest a resolution of the no - cloning paradox of quantum information between matter falling into the horizon and holographic charges on the horizon from the viewpoint of the contextuality of quantum measurement . + + + _ graduate school of science , tohoku university , + sendai 980 - 8578 , japan _ +

1606.02443

recently hawking , perry , and strominger ( hps ) @xcite @xcite proposed an interesting scenario that may resolve the information loss problem @xcite . they suggest that quantum information about collapsing matter is stored in an infinite number of conserved noether currents having asymptotic symmetries , including supertranslation on a horizon . this is expected to maintain the unitarity of quantum gravity , and all of the information may be accessible in the region outside of the horizon . by incorporating superrotation symmetry on a horizon , this symmetry - based scenario may also provide a possibility of revealing the statistical mechanical origin of the bekenstein hawking entropy @xmath0 , as already pointed out in previous papers @xcite and @xcite by one of the authors of this paper . the nonvanishing holographic charges of these asymptotic symmetries can yield a huge number of different physical states with the same adm energy and angular momentum . the degeneracy is so large that it may account for the order of @xmath0 @xcite . in holographic charge arguments , the physical degrees of freedom emerge from would - be gauge degrees of freedom of the general covariance . this can be grasped easily by recalling the poincar covariance as a simple example . in fact , the lorentz transformation is a subgroup of the poincar group and generates an infinite number of physical states with different values of the momentum . the poincar transformation is a subgroup of general coordinate transformations . because general relativistic theories have the general covariance as the gauge symmetry , the poincar transformation can be regarded as a would - be gauge transformation , which actually causes a transition between physical states . on a horizon , a similar mechanism works , and an infinite - dimensional asymptotic symmetry appears . in the spirit of brown and henneaux @xcite , strominger first suggested in @xcite that a three - dimensional black hole entropy is derived using a virasoro algebra as an asymptotic symmetry at spatial infinity . carlip @xcite proposed an asymptotic virasoro symmetry on a black hole horizon and argued that the black hole entropy is derived using the cardy formula with a macroscopically large central charge . however , the original argument of carlip encountered various types of criticism @xcite@xcite@xcite@xcite@xcite and remains controversial . the existence of supertranslation and superrotation as a consistent asymptotic symmetry on a stationary horizon of a four - dimensional schwarzschild black hole was first reported in a paper @xcite by one of the authors of the present paper . subsequently , it was explicitly demonstrated that falling matter really excites charged states of the asymptotic symmetry in a three - dimensional black hole spacetime @xcite . it is not known yet whether a gravitational wave excites charged states of supertranslation and superrotation on a horizon . hps recently revisited supertranslation and superrotation on the horizon @xcite @xcite by using a different coordinate system from that in @xcite and @xcite . using a coordinate system , hps introduced asymptotic metrics near a horizon as @xmath1 where the horizon is located at @xmath2 and uppercase roman letters run over spatial coordinates on the horizon . based on the above form , hps argue that stationary black holes do not carry classical supertranslation hair because the holographic charge vanishes in the classical level @xcite . it is worth noting that the coordinate system in eq . ( [ 001 ] ) covers both regions inside of the horizon and outside of the horizon , and can be physically implemented by free - falling block - numbered clocks distributed in the space near the horizon , as depicted in the left panel of figure 1 . because of the clock free motion , it is very natural from the viewpoint of equivalence principle that nonzero holographic charges on the horizon can not be observed . apart from hps s setup , if we adopt a rindler - type coordinate system in which an asymptotic near - horizon metric is given by@xmath3 where @xmath4 is rindler acceleration and the horizon is located at @xmath5 , we indeed have nonvanishing classical holographic charges on the horizon even for stationary black holes @xcite . the coordinate system in eq . ( [ 002 ] ) is implemented by accelerating block - numbered clocks distributed in the space near the horizon , as depicted in the right panel of figure 1 . the appearance of the charges @xmath6 $ ] in the accelerated coordinate system is reminiscent of that of a thermal bath in the unruh effect @xcite . $ ] on the horizon vanish for stationary black holes . ( right ) a coordinate system implemented by accelerating block - numbered clocks distributed in the space near the horizon . in this coordinate system , non - zero classical holographic charges @xmath6 $ ] appear on the horizon as soft black hole hair even for stationary black holes . the appearance of the charges in the accelerated coordinate system is reminiscent of that of a thermal bath in the unruh effect . , height=207 ] as in the black hole complementarity scenario @xcite , the hps scenario requires some mechanism to avoid the no - cloning paradox . if all of the information about collapsing matter is stored in the conserved charges on the horizon , we may make a precise copy of the quantum information of the matter inside of the horizon by using the charge information . although this naively seems to contradict the no - cloning theorem of quantum mechanics @xcite , hps have not yet provided any plausible resolution of this paradox . let us consider the gravitational collapse depicted in the left panel of figure 2 . in this scenario , information about the collapsing matter is imprinted in the holographic charge states of asymptotic symmetries on the horizon . if we throw additional matter into the black hole , as depicted in the right panel of figure 2 , a new horizon appears and encloses the old horizon . then new holographic charge states have to carry all of the information about the original collapsing matter and additional matter . naively , this appears strange . the new holographic charges must remember quantum information about the behavior of the original collapsing matter before the additional matter arrived . thus , we potentially have duplicate quantum information about the collapsing matter on the two different horizons . this challenges the no - cloning theorem again . this situation remains unchanged even if we take a large mass limit on the black hole . in this limit , the near - horizon geometry is merely a minkowski spacetime , and each horizon coincides with one of the rindler horizons . the situation is depicted in figure 3 . thus , even if a linearized theory of quantum gravity is considered in a minkowski background with rindler horizons , the no - cloning problem should be resolved properly . this implies that the investigation of holographic charges on rindler horizons is valuable . to analyze the above problems of asymptotic rindler spacetimes , we construct a general theory of gravitational holographic charges for a ( 1 + 3)-dimensional linearized gravity field in section 2 of this paper . although matter crossing a rindler horizon generates horizon deformation and a time - dependent coordinate shift , that is , gravitational memory , the charges of supertranslation and superrotation on the horizon can be defined during and after its passage through the horizon . in particular , it is verified that the charges become time - independent after matter absorption by the horizon . it should be emphasized that the time independence of the charge is very nontrivial for a general coordinate transformation . for instance , a noether current for an infinitesimal coordinate transformation @xmath7 is given in einstein gravitational theory by @xmath8 and is locally conserved : @xmath9 . the charge @xmath10 is defined by the integration of @xmath11 in a bulk region @xmath12 as@xmath13 even though @xmath9 holds , the time independence of @xmath10 is not generally ensured . this is because the surface integral of the flux @xmath14 does not disappear for many @xmath15 even if we have no gravitational wave and matter on the surface . from the local balance relation of conserved currents , @xmath16 @xmath10 can vary in time because the right - hand side of the above equation is generally nonzero . thus , for a surface of interest such as a horizon , the time independence of the holographic charges has to be checked independently by confirming that @xmath9 . in section 2 , we show the time independence of the holographic charges on rindler horizons . a general formula for conserved holographic charges after the horizon absorbs the matter is provided . it is also proven that holographic states on the horizon can not store any information about absorbed perturbative gravitational waves . to show the memory effect of horizons , a measure of the classical gravitational memory @xmath17 on a future horizon at @xmath18 with a rindler wedge located at @xmath19 is introduced by time integration of the holographic charges . it differs from the standard gravitational memory that appears in @xcite @xcite @xcite @xcite @xcite and provides a more profound insight because it consists of conserved charges . in section 3 , we introduce a quantum gravitational memory operator @xmath20 for a rindler horizon and propose a possible resolution of the no - cloning paradox from the viewpoint of quantum measurement contextuality . the energy - momentum tensor of quantum matter inside of a horizon does not commute with @xmath21 on the horizon . this also leads to noncommutativity between holographic charges of different horizons . thus , a measurement of @xmath21 on a horizon affects other @xmath22 and quantum states of matter inside of the horizon . they are not independent observables , and the reality of @xmath21 is subject to the contextuality of quantum measurement . on the basis of this fact , we propose a conjecture that the holographic charge reality is conditioned to measurements by appropriate physical detectors . then no cloning paradox arises , at least in the first order of perturbative quantum gravity . in a method similar to that of unruh dewitt particle detectors @xcite @xcite for unruh radiation in a rindler spacetime , physical detectors accumulating information for evaluation of a quantum metric near the horizon may observe the holographic charges as a reality . owing to gravitational interaction between infalling matter and the metric detectors distributed in the space , the detectors share quantum entanglement with the matter inside of the horizon . if we place no detector to measure the quantum metric near the horizon , the absorbed matter is not decohered by the measurement . all of the quantum information about the infalling matter remains carried by the matter itself . hence , the holographic noether charges are not physical objects that share quantum entanglement with the matter inside but merely a gauge freedom of general covariance in this case . therefore , even in quantum theory , the horizon can not be regarded as a real holographic screen spanning the space in the absence of measurement devices . this conjecture avoids the no - cloning paradox and also supports a conservative conjecture for the information loss problem in @xcite , @xcite and @xcite , which is quite different from that in the firewall conjecture @xcite . in this paper , the natural units are adopted : @xmath23 . in this section , we construct a general theory of gravitational holographic charges for a weak gravity field . first , a useful gauge condition , referred to as the rindler gauge , is introduced for an arbitrary configuration of the field . next , using the rindler gauge , a regge teitelboim canonical theory is formulated for holographic charges on a rindler horizon . using the gravitational constant @xmath24 , let us introduce the planck length constant @xmath25 . then a perturbative gravitational field is written as @xmath26 where @xmath27 is the minkowski metric , and @xmath28 is a weak field in an arbitrary gauge . to consider a future rindler horizon @xmath29 , let us introduce light - cone coordinates @xmath30 for convenience . consider the conformal rindler coordinates @xmath31 , @xmath32 in the @xmath33 plane . in the four - dimensional coordinates @xmath34 , the background metric @xmath35 takes the standard rindler form as @xmath36{cccc}\bar{g}_{\tau\tau } & \bar{g}_{\tau\rho } & \bar{g}_{\tau y } & \bar{g}_{\tau z}\\ \bar{g}_{\rho\tau } & \bar{g}_{\rho\rho } & \bar{g}_{\rho y } & \bar{g}_{\rho z}\\ \bar{g}_{y\tau } & \bar{g}_{y\rho } & \bar{g}_{yy } & \bar{g}_{yz}\\ \bar{g}_{z\tau } & \bar{g}_{z\rho } & \bar{g}_{zy } & \bar{g}_{zz}\end{array } \right ] = \left [ \begin{array } [ c]{cccc}-\delta & 0 & 0 & 0\\ 0 & \delta & 0 & 0\\ 0 & 0 & 1 & 0\\ 0 & 0 & 0 & 1 \end{array } \right ] , \label{22}\ ] ] where@xmath37 under a coordinate transformation , the weak field @xmath28 in eq . ( [ e1 ] ) is transformed as@xmath38 and the total metric is given by @xmath39 under an infinitesimal general coordinate transformation @xmath40 for the metric , @xmath41 changes as follows : @xmath42 here @xmath43 stands for metric perturbation in a new gauge in the @xmath44 coordinates . by using this gauge freedom of @xmath45 , we can always impose on @xmath43 the gauge condition @xmath46 which is referred to as the rindler gauge in this paper . in the flat coordinates @xmath47 , the rindler gauge imposes @xmath48 = 0,\label{e6}\\ \varphi_{\tau\rho}^{(r ) } & = -\frac{1}{4\kappa^{2}}\left [ \left ( x^{+}\right ) ^{2}h_{++}^{(r)}-\left ( x^{-}\right ) ^{2}h_{--}^{(r)}\right ] = 0 , \label{e7}\]]@xmath49 where @xmath50 is defined as @xmath51 the @xmath52 component given by @xmath53\ ] ] is rewritten as @xmath54 using eqs . ( [ e6 ] ) and ( [ e7 ] ) . similarly , the @xmath55 component defined as @xmath56 is computed as@xmath57 from eq . ( [ e9 ] ) . these equations can be summarized as the following asymptotic condition around @xmath58 : @xmath36{cccc}\varphi_{\tau\tau}^{(r ) } & \varphi_{\tau\rho}^{(r ) } & \varphi_{\tau y}^{(r ) } & \varphi_{\tau z}^{(r)}\\ \varphi_{\rho\tau}^{(r ) } & \varphi_{\rho\rho}^{(r ) } & \varphi_{\rho y}^{(r ) } & \varphi_{\rho z}^{(r)}\\ \varphi_{y\tau}^{(r ) } & \varphi_{y\rho}^{(r ) } & \varphi_{yy}^{(r ) } & \varphi_{yz}^{(r)}\\ \varphi_{z\tau}^{(r ) } & \varphi_{z\rho}^{(r ) } & \varphi_{zy}^{(r ) } & \varphi_{zz}^{(r)}\end{array } \right ] = \left [ \begin{array } [ c]{cccc}o\left ( \left ( x^{-}\right ) ^{2}\right ) & 0 & o\left ( x^{-}\right ) & o\left ( x^{-}\right ) \\ 0 & 0 & 0 & 0\\ o\left ( x^{-}\right ) & 0 & o\left ( \left ( x^{-}\right ) ^{0}\right ) & o\left ( \left ( x^{-}\right ) ^{0}\right ) \\ o\left ( x^{-}\right ) & 0 & o\left ( \left ( x^{-}\right ) ^{0}\right ) & o\left ( \left ( x^{-}\right ) ^{0}\right ) \end{array } \right ] . \label{e10}\ ] ] thus , in the rindler gauge , any weak field takes the form of eq . ( [ e10 ] ) around the rindler horizon of @xmath58 . in the flat coordinates @xmath59 , the transformation to the rindler gauge is expressed as @xmath60 where @xmath61 for @xmath62 , the @xmath63 component gauge condition @xmath64 becomes @xmath65 . \label{11}\ ] ] the second gauge condition , @xmath66 , yields @xmath67 , \label{12}\ ] ] and the third gauge condition , @xmath68 , gives @xmath69 the above equations determine the gauge parameters @xmath62 for given @xmath28 . next , let us solve the equations explicitly using a variable @xmath70 defined as @xmath71 eq . ( [ 11 ] ) reads@xmath72 , \end{aligned}\ ] ] where @xmath73 is the partial derivative with respect to @xmath74 for a fixed @xmath70 . integration with respect to @xmath74 gives the following equation . @xmath75 dq , \label{14}\ ] ] where @xmath76 is an unfixed integration function of @xmath77 . subtracting eq . ( [ 12 ] ) from eq . ( [ 11 ] ) gives @xmath78 \\ & = -\frac{1}{2}\left [ \left ( x^{+}\right ) ^{2}h_{++}+2x^{+}x^{-}h_{+-}+\left ( x^{-}\right ) ^{2}h_{--}\right ] \\ & + \frac{1}{2}\left [ \left ( x^{+}\right ) ^{2}h_{++}-\left ( x^{-}\right ) ^{2}h_{--}\right ] .\end{aligned}\ ] ] using the @xmath79 variables , this is expressed as @xmath80 substituting eq . ( [ 14 ] ) into the above equation yields @xmath81 dq\\ & + \frac{1}{x^{+}}\int_{0}^{x^{+}}\left [ h_{+-}\left ( q , sq , y , z\right ) + sh_{--}\left ( q , sq , y , z\right ) \right ] dq.\end{aligned}\ ] ] by integrating the equation with respect to @xmath74 , we have a general solution of @xmath82 such that @xmath83 dq\nonumber\\ & -\frac{1}{2}\int_{0}^{x^{+}}q\left [ \partial_{-}h_{++}\left ( q , sq , y , z\right ) + 2s\partial_{-}h_{+-}\left ( q , sq , y , z\right ) + s^{2}\partial_{-}h_{--}\left ( q , sq , y , z\right ) \right ] dq\nonumber\\ & + \frac{x^{+}}{2}\int_{0}^{x^{+}}\left [ \partial_{-}h_{++}\left ( q , sq , y , z\right ) + 2s\partial_{-}h_{+-}\left ( q , sq , y , z\right ) + s^{2}\partial_{-}h_{--}\left ( q , sq , y , z\right ) \right ] dq , \label{17}\ ] ] where @xmath84 is an unfixed integration function . from this result and eq . ( [ 14 ] ) , @xmath85 is solved as @xmath86 dq\nonumber\\ & + s\int_{0}^{x^{+}}\left [ h_{+-}\left ( q , sq , y , z\right ) + sh_{--}\left ( q , sq , y , z\right ) \right ] dq\nonumber\\ & + \frac{s}{2}\int_{0}^{x^{+}}q\left [ \partial_{-}h_{++}\left ( q , sq , y , z\right ) + 2s\partial_{-}h_{+-}\left ( q , sq , y , z\right ) + s^{2}\partial_{-}h_{--}\left ( q , sq , y , z\right ) \right ] dq\nonumber\\ & -\frac{x^{+}s}{2}\int_{0}^{x^{+}}\left [ \partial_{-}h_{++}\left ( q , sq , y , z\right ) + 2s\partial_{-}h_{+-}\left ( q , sq , y , z\right ) + s^{2}\partial_{-}h_{--}\left ( q , sq , y , z\right ) \right ] dq . \label{16}\ ] ] similarly , eq . ( [ 13 ] ) is rewritten as @xmath87 and the gauge parameter @xmath88 can be solved as follows . @xmath89 dq\nonumber\\ & -\frac{1}{2}\partial_{a}\int_{0}^{x^{+}}q\left [ h_{++}\left ( q , sq , y , z\right ) + 2sh_{+-}\left ( q , sq , y , z\right ) + s^{2}h_{--}\left ( q , sq , y , z\right ) \right ] dq\nonumber\\ & + \frac{x^{+}}{2}\partial_{a}\int_{0}^{x^{+}}\left [ h_{++}\left ( q , sq , y , z\right ) + 2sh_{+-}\left ( q , sq , y , z\right ) + s^{2}h_{--}\left ( q , sq , y , z\right ) \right ] dq , \label{18}\ ] ] where @xmath90 is an unfixed integration function . next , let us focus on a case in which an incoming matter field or gravitational wave takes nonzero values in a region @xmath91 $ ] with @xmath92 and vanishes outside of it . at the initial time @xmath93 , the gravity field is in the vacuum state . thus , we can set a boundary condition as @xmath94 the matter crossing the horizon at @xmath58 induces nonzero values of @xmath95 @xmath96 after the pass of the matter ( @xmath97 ) such that @xmath98 note that the above coordinate transformations yield a time - dependent metric even after matter passes across the horizon . because eq . ( [ 33 ] ) implies that @xmath99 , @xmath100 generates horizon deformation after the matter absorption . from eqs . ( [ 17 ] ) , ( [ 16 ] ) , and ( [ 18 ] ) , they are determined as @xmath101 where the gauge conditions of @xmath28 remain arbitrary except that @xmath28 vanishes outside of @xmath91 $ ] . it is easy to check that eqs . ( [ 19 ] ) , ( [ 20 ] ) , ( [ e19 ] ) , and ( [ e20 ] ) are invariant under a gauge transformation such that @xmath102 , with @xmath103 for @xmath104 $ ] . here @xmath105 generates a superrotation charge , and @xmath106 generates a supertranslation charge , as seen in the next subsection . @xmath107 corresponds to a gauge freedom from the viewpoint of the entire minkowski spacetime and is associated with a generalized spatial translation of a rindler region . however , it turns out that the corresponding charge ( generator ) is not well defined in the canonical formulation . thus , @xmath107 should be regarded as just one of the dynamical variables , which controls horizon deformation , and is not associated with any asymptotic symmetry in this formulation . let us consider the regge teitelboim canonical theory in the conformal rindler coordinates @xmath108 . the adm decomposition of the metric is given by @xmath109 where the lowercase roman letters run over @xmath110 and @xmath111 . the conjugate momentum of @xmath112 is defined as @xmath113,\ ] ] where @xmath114 is the extrinsic curvature , @xmath115 where @xmath116 denotes the three - dimensional covariant derivative using @xmath112 . the hamiltonian density @xmath117 and momentum density @xmath118 are defined as @xmath119 -\frac{\sqrt{h}}{\kappa^{2}}r^{(3)},\\ & \mathcal{h}_{a}=-2\pi_{ab}|^{b},\end{aligned}\ ] ] and the einstein equation in pure gravity imposes the following constraints : @xmath120 in order to analyze generators of coordinate transformation in the canonical theory , let us consider an infinitesimal transformation @xmath121 and introduce @xmath122 the generator @xmath123 $ ] for the transformation @xmath124 is the sum of the bulk term @xmath125 $ ] and surface term @xmath6 $ ] : @xmath126=h[\xi]+q[\xi].\ ] ] the bulk term is given by @xmath127=\int d^{3}\sigma\left [ \hat{\xi}^{\tau}\mathcal{h}+\hat{\xi}^{a}\mathcal{h}_{a}\right]\ ] ] and becomes zero if the equation of motion is satisfied . the surface term is obtained by integration of @xmath128 & = \int d^{3}\sigma\partial_{c}\left ( g^{abcd}\left [ \hat{\xi}^{\tau}\delta h_{ab}|_{d}-\hat{\xi}^{\tau}|_{d}\delta h_{ab}\right ] \right ) \nonumber\\ & + \int d^{3}\sigma\partial_{c}\left ( 2\hat{\xi}_{a}\delta\pi^{ac}\right ) \nonumber\\ & \ + \int d^{3}\sigma\partial_{c}\left ( \left [ \hat{\xi}^{a}\pi^{bc}+\hat{\xi}^{b}\pi^{ac}-\hat{\xi}^{c}\pi^{ab}\right ] \delta h_{ab}\right ) , \label{21}\ ] ] where @xmath129 in general , the integrability of eq . ( [ 21 ] ) is nontrivial , and integration is possible only for limited transformations . near an event horizon of a schwarzschild black hole , the integrability is proven for supertranslation and superrotation @xcite . similarly , a rindler spacetime admits the integrability of the following asymptotic transformation on a horizon with @xmath5 . the coordinate transformation of the asymptotic symmetry is given by @xmath130 where @xmath131 is an arbitrary function of @xmath132 and generates supertranslation on the future rindler horizon at @xmath58 and on the past rindler horizon at @xmath93 . @xmath133 generates a general coordinate transformation in the @xmath134 plane and corresponds to superrotation on the same horizon . under this transformation , the rindler metric in eq . ( [ 22 ] ) is transformed into a stationary asymptotic metric given by @xmath36{cccc}g_{\tau\tau } & g_{\tau\rho } & g_{\tau y } & g_{\tau z}\\ g_{\rho\tau } & g_{\rho\rho } & g_{\rho y } & g_{\rho z}\\ g_{y\tau } & g_{y\rho } & g_{yy } & g_{yz}\\ g_{z\tau } & g_{z\rho } & g_{zy } & g_{zz}\end{array } \right ] = \left [ \begin{array } [ c]{cccc}-\delta+o\left ( \delta^{2}\right ) & o\left ( \delta^{2}\right ) & o\left ( \delta\right ) & o\left ( \delta\right ) \\ o\left ( \delta^{2}\right ) & \delta & o\left ( \delta^{2}\right ) & o\left ( \delta^{2}\right ) \\ o\left ( \delta\right ) & o\left ( \delta^{2}\right ) & o\left ( \delta ^{0}\right ) & o\left ( \delta^{0}\right ) \\ o\left ( \delta\right ) & o\left ( \delta^{2}\right ) & o\left ( \delta ^{0}\right ) & o\left ( \delta^{0}\right ) \end{array } \right ] , \label{23}\ ] ] where @xmath135 . the supertranslation charge is defined by eq . ( [ 21 ] ) with @xmath136 the superrotation charge corresponds to @xmath137 by use of @xmath138 ( y , z)}=0 $ ] and the above asymptotic form in eq . ( [ 23 ] ) , @xmath139 in eq . ( [ 21 ] ) is computed as@xmath140 } \right ) + 2\xi^{a}(y , z)\delta\pi_{a}^{\rho } \right ] _ { \rho=\infty } \label{010}\ ] ] and can be integrated as @xmath141 = \int dydz\left [ -\frac{\xi^{\tau}(y , z)}{\kappa^{3}}\left ( \sqrt{\det\left [ h_{ab}\right ] } -1\right ) + 2\xi^{a}(y , z)\pi _ { a}^{\rho}\right ] .\ ] ] by using eqs . ( [ 24 ] ) and ( [ 25 ] ) , the charges are evaluated as @xmath141 = -\frac{1}{\kappa^{3}}\int dydz\left [ \begin{array } [ c]{c}\xi^{\tau}(y , z)\left ( \sqrt{\det\left [ \partial_{a}x^{c}(y , z)\partial _ { b}x^{c}(y , z)\right ] } -1\right ) \\ + \xi^{a}(y , z)\partial_{a}t(y , z)\sqrt{\det\left [ \partial_{b}x^{d}(y , z)\partial_{c}x^{d}(y , z)\right ] } \end{array } \right ] . \label{42}\ ] ] note that the transformation in eqs . ( [ 24 ] ) , ( [ 241 ] ) , and ( [ 25 ] ) is fixed only near the horizon . actually , we can extend it into the bulk region outside of the horizon and give an arbitrary @xmath142 dependence to @xmath143 . by using this gauge freedom , it is possible to assume without loss of generality that @xmath144 therefore , the holographic symmetry on the horizon is independent of the asymptotic symmetry at spatial infinity ( @xmath145 ) . the metrics generated by the symmetry on the horizon give the same asymptotic gravitational field at spatial infinity . returning to the case of a black hole with finite mass , this means that the black hole has infinite degeneracy near the horizon with the same adm energy and angular momentum as at spatial infinity . this degeneracy is so large that state counting may give the same order of the entropy as @xmath0 @xcite , as will be seen again in the next subsection . if we consider an asymptotic metric with @xmath146 } \neq0 $ ] , which includes the effects of incoming matter and a gravitational wave across a rindler horizon at @xmath58 , @xmath139 has an additional @xmath147-derivative term of @xmath148 , and the nonperturbative integrability is broken for supertranslation . in this case , the first term on the right - hand side of eq . ( [ 21 ] ) is computed as @xmath149 \right ) \nonumber\\ & = -\int d^{3}\sigma\partial_{c}\delta\left [ \frac{\xi^{\tau}}{\kappa^{2}}\sqrt{\det\left [ h_{ab}\right ] } \right ] \nonumber\\ & + \int d^{3}\sigma\partial_{c}\left ( \xi^{\tau}\frac{\sqrt{\det\left [ h_{ab}\right ] } } { 2\kappa}\left [ \left ( \partial_{\rho}h^{ab}\right ) \delta h_{ab}-2\partial_{\rho}\left ( h^{ab}\delta h_{ab}\right ) \right ] \right ) . \label{02}\ ] ] the first term on the right - hand side of eq . ( [ 02 ] ) is integrable and has already appeared in eq . ( [ 010 ] ) . the second term is new and not always integrable for supertranslation . however , if we concentrate on the weak gravity field @xmath50 in the rindler gauge , integration of @xmath150 remains achievable . the charge on the horizon is computed as @xmath151 & = \frac{1}{2\kappa^{2}}\lim_{x^{-}\rightarrow0}\int dydz\xi^{\tau } \left [ x^{+}\partial_{+}+x^{-}\partial_{-}-1\right ] h_{aa}^{(r)}\left ( x^{+},x^{-},y , z\right ) \nonumber\\ & -\frac{2}{\kappa x^{+}}\lim_{x^{-}\rightarrow0}\int dydz\xi^{a}\left [ x^{+}\partial_{+}+x^{-}\partial_{-}+1\right ] h_{-a}^{(r)}\left ( x^{+},x^{-},y , z\right ) , \label{45}\ ] ] where @xmath152 . it is possible to obtain the same result using the wald zoupas covariant current formulation in @xcite . in the region with @xmath97 with some positive @xmath153 , @xmath28 vanishes . thus , @xmath50 is given simply by @xmath154 , where @xmath155 are given by eqs . ( [ 17 ] ) , ( [ 16 ] ) , and ( [ 18 ] ) . the charge is calculated as @xmath156=-\frac{1}{\kappa^{2}}\int dydz\xi^{\tau}\partial_{a}r_{a}\left ( 0,y , z\right ) + \frac{4}{\kappa^{2}}\int dydz\xi^{a}\partial_{a}t\left ( 0,y , z\right ) , \label{50}\ ] ] where @xmath157 is given by eq . ( [ 19 ] ) , and @xmath158 is given by eq . ( [ 20 ] ) . this result establishes the time independence of the holographic charges in the future region . if the vector field @xmath159 tends to zero at spatial infinity in the @xmath160plane , it can be decomposed into @xmath161 using this decomposition , we obtain the following expression for the superrotation charge : @xmath162=\frac{1}{\kappa}\int dydz\left [ \partial_{a}\partial_{a}\xi ^{(2)}(y , z)\right ] \left [ \int_{0}^{\infty}\partial_{-}h_{++}\left ( q,0,y , z\right ) dq\right ] . \label{300}\ ] ] from this result , the superrotation charge is found to vanish for the area - preserving component generated by @xmath163 . only the rotationless component generated by @xmath164 can take nonzero values of the superrotation charge . so far , we have not yet used the einstein equation , @xmath165 substituting eq . ( [ e1 ] ) into the above equation and taking the first - order terms in terms of @xmath4 yields the following equation of motion for @xmath28 : @xmath166 from this equation , the energy momentum conservation of matter , @xmath167 , is automatically satisfied . if the standard harmonic gauge for @xmath28 , @xmath168 is adopted , the equation of motion reads@xmath169 for example , in the coordinates @xmath47 , @xmath170 obeys @xmath171 h_{++}=-\kappa t_{++ } , \label{e3}\ ] ] and @xmath172 obeys @xmath171 h_{+-}=\frac{\kappa}{4}t_{aa } , \label{e4}\ ] ] where the index @xmath173 takes @xmath174 and @xmath111 . from eq . ( [ e3 ] ) , we obtain @xmath175 dq . \label{27}\ ] ] by using the gauge condition in eq . ( [ e2 ] ) , the following relation holds . @xmath176 substituting the above equation into eq . ( [ 27 ] ) gives @xmath177 therefore , the supertranslation charge is computed as @xmath178=-\frac{1}{2\kappa}\int dydz\xi^{\tau}(y , z)\left [ \int _ { 0}^{\infty}qt_{++}\left ( q,0,y , z\right ) dq\right ] .\nonumber\ ] ] to evaluate the superrotation charge , we adopt the green function that obeys @xmath171 g\left ( x^{+},x^{-},y , z\right ) = \delta\left ( x^{+}\right ) \delta\left ( x^{-}\right ) \delta^{2}\left ( x^{a}\right ) .\ ] ] this green function takes the fourier form @xmath179 } { 4k_{+}k_{-}+k_{a}k_{a}}e^{i\left ( k_{y}y+k_{z}z\right ) } dk_{+}dk_{-}dk_{y}dk_{z}.\ ] ] using this expression , we obtain a convenient formula : @xmath180 where @xmath181 is a two - dimensional green function in the @xmath182 plane satisfying @xmath183 and given by @xmath184 } } { k_{y}^{2}+k_{z}^{2}}\frac{dk_{y}dk_{z}}{\left ( 2\pi\right ) ^{2}}=\frac { 1}{4\pi}\ln\left [ \frac{\left ( y - y^{\prime}\right ) ^{2}+\left ( z - z^{\prime}\right ) ^{2}}{\kappa^{2}}\right ] .\ ] ] using eq . ( [ 31 ] ) , we obtain @xmath185 \left [ \int_{-\infty}^{\infty}\partial_{-}t_{++}\left ( q,0,y^{\prime},z^{\prime } \right ) dq\right ] dy^{\prime}dz^{\prime}.\end{aligned}\ ] ] on the basis of these results , we obtain a general formula for the charges . @xmath186\nonumber\\ & = -\frac{1}{2\kappa}\int dydz\xi^{\tau}(y , z)\left [ \int_{0}^{\infty}x^{+}t_{++}\left ( x^{+},0,y , z\right ) dx^{+}\right ] \nonumber\\ & + \frac{1}{4\pi}\int dydz\int dy^{\prime}dz^{\prime}\xi^{a}(y , z)\partial _ { a}\nonumber\\ & \times\ln\left [ \frac{\left ( y - y^{\prime}\right ) ^{2}+\left ( z - z^{\prime}\right ) ^{2}}{\kappa^{2}}\right ] \left [ \int_{0}^{\infty } \partial_{-}t_{++}\left ( x^{+},0,y^{\prime},z^{\prime}\right ) dx^{+}\right ] . \label{4}\ ] ] this is the main result of this paper . this expression is invariant under changes in the cutoff @xmath187 in the two - dimensional green function . first , it should be stressed that in eq . ( [ 4 ] ) , the holographic charges of both supertranslation and superrotation vanish if we have no matter energy - momentum tensor . as seen in eq . ( [ 27 ] ) , supertranslation charges of incoming gravitational wave vanish due to the einstein equation . it is also verified that the superrotation charges vanish by considering an incoming gravitational wave field as @xmath188 + h.c.\right ] , \label{302}\ ] ] where the upper end of @xmath189 integration is zero and describes the incoming wave condition . when this is substituted into eq . ( [ 300 ] ) , using the fact that @xmath190 support is only in the region @xmath191 , it is noticed that @xmath192 so @xmath193 $ ] certainly vanishes for gravitational waves . this means that the charges do not store gravitational wave information , at least to the first order of @xmath4 . this result for gravitational waves is similar to the result for electromagnetic waves . on a rindler horizon , the hps charge in @xcite is given by @xmath194 where @xmath195 is the @xmath196 component of the electric field on the horizon at @xmath58 , and @xmath197 is a @xmath198 gauge parameter depending on @xmath199 . an electromagnetic wave can yield nontrivial time evolution of @xmath200 during its passage across the horizon . however , in the late - time region with @xmath97 , the wave is already located inside of the horizon , and no electric field is on the horizon . therefore , the charge tends to zero , and no information about the electromagnetic wave is stored in holographic states on the horizon boundary with infinite radius at null future infinity enclosing interior electric charges , electromagnetic waves crossing the boundary can shift the values of the hps charges on the boundary . ] . the reason that the electromagnetic holographic charge vanishes for radiation is essentially that the charge arises from a gauge - invariant electric field . however , it should be stressed that the situation for gravitational waves may differ from that for electromagnetic waves . in general relativity , gauge - dependent variables such as the metric are also physical observables . by using many clocks that are distributed in space and exchanging signals between them , the metric is determined experimentally . as shown in eq . ( [ 45 ] ) , the holographic charges are not determined by the curvature tensors @xmath201 . they consist of the metric and its first derivatives . thus , the charges are gauge - dependent objects , although they can be observed by physical detectors . further , a stationary asymptotic rindler spacetime without any matter in it has nontrivial charges of supertranslation and superrotation @xcite . hence , even in pure gravity , it is possible that a gravitational wave generates a shift in the values of the charges in the future region . to investigate this possibility , a long calculation is required . after the computation is complete , we find , as a nontrivial result , that gravitational waves never shifts the charges of supertranslation and superrotation . note that the gravitational holographic states store the information about the electromagnetic waves , as seen in eq . ( [ 4 ] ) with@xmath202 where @xmath203 is a tensor of electric field and magnetic field . this may tempt us to imagine that the same scenario applies for gravitational waves by taking account of higher - order correction terms in terms of @xmath4 , which induces a pseudo energy - momentum tensor for gravitational waves . unfortunately , we encounter serious trouble . the integrability of the supertranslation charge in eq . ( [ 02 ] ) breaks down because of the higher - order corrections in the second term . thus , it remains a crucial open question whether holographic charges on a horizon can be defined for strong gravitational waves . this will be discussed again in section 4 . four more comments regarding the main result in eq . ( [ 4 ] ) are listed below . ( i ) the classical energy condition ensures positivity of the generator of the lorentz boost ( rindler energy ) on the horizon in the supertranslation charge : @xmath204 thus , the bulk rindler energy associated with @xmath205 in eq . ( [ 4 ] ) always decreases when incoming matter crosses the horizon . ( ii ) the superrotation charge in eq . ( [ 4 ] ) for matter propagating in parallel in the @xmath196 direction , which obeys @xmath206 , is equal to zero . in the superrotation charged states , mainly interference information about waves propagating in different directions is stored . ( iii ) the local supertranslation charge @xmath207/\delta\xi^{\tau}(y , z)$ ] retains only the information about the total amount of rindler energy @xmath208 that passes through the point @xmath134 . however , the superrotation charge stores information about matter nonlocally . the influence of @xmath209 at some point propagates widely in the @xmath134 plane via the logarithmic long - range behavior of the two - dimensional green function @xmath181 . this resembles the behavior of the black hole s matrix of t hooft @xcite . ( iv ) in a minkowski background , we have an infinite number of rindler horizons . thus , we can define a measure of gravitational memory at different horizons using the supertranslation and superrotation charges . let us define @xmath210 for a future rindler horizon at @xmath18 and a rindler wedge at @xmath211 as @xmath212 \nonumber\\ & + \frac{1}{4\pi}\int dydz\int dy^{\prime}dz^{\prime}\xi^{a}(y , z)\partial _ { a}\nonumber\\ & \times\ln\left [ \frac{\left ( y - y^{\prime}\right ) ^{2}+\left ( z - z^{\prime}\right ) ^{2}}{\kappa^{2}}\right ] \left [ \int_{0}^{\infty } \partial_{-}t_{++}\left ( x^{+}+x_{h}^{+},x_{h}^{-},y^{\prime},z^{\prime } \right ) dx^{+}\right ] . \label{5}\ ] ] using this memory , each horizon acts as a holographic screen that stores matter information , as depicted in figure 4 . in the next section , @xmath213 is quantized , and the no - cloning paradox is discussed . here let us apply the state counting argument in @xcite to stationary asymptotic rindler spacetimes . this counting is just a rough estimation . however , it strongly suggests that the horizon states can supply an entropy of @xmath214 and implies that superrotation and supertranslation on a horizon may make it possible to create black holes with thick hair . let us consider the following generators of supertranslation and superrotation on a rindler horizon in the @xmath44 coordinates : @xmath215 & = \xi^{\tau}\left ( y , z\right ) \partial_{\tau},\\ g_{sr}\left [ \xi^{a}\right ] & = \xi^{a}\left ( y , z\right ) \partial_{a}.\end{aligned}\ ] ] they form a closed algebra such that @xmath216 , g_{st}\left [ \xi^{\prime\tau } \right ] \right ] & = 0,\label{36}\\ \left [ g_{sr}\left [ \xi^{a}\right ] , g_{st}\left [ \xi^{\tau}\right ] \right ] & = g_{st}\left [ \xi^{a}\partial_{a}\xi^{\tau}\right ] , \label{38}\\ \left [ g_{sr}\left [ \xi^{a}\right ] , g_{sr}\left [ \xi^{\prime a}\right ] \right ] & = g_{sr}\left [ \xi^{b}\partial_{b}\xi^{\prime a}-\xi^{\prime b}\partial_{b}\xi^{a}\right ] . \label{37}\ ] ] assume that the @xmath134 plane has a finite area @xmath217 by imposing periodic boundary conditions in the @xmath174 and @xmath111 directions . then the fourier expansion of @xmath218 is given by @xmath219 to regularize the ultraviolet divergence , the cutoffs @xmath220 and @xmath221 are introduced , and the corresponding momenta are on the order of the planck scale . this yields the following relation.@xmath222 the generators of each fourier component are defined as@xmath223 and satisfy @xmath224 = 0.\ ] ] the algebraic structure in eqs . ( [ 36 ] ) , ( [ 38 ] ) , and ( [ 37 ] ) is very similar to that of poincar algebra , although we have an infinite number of generators on the horizon . @xmath225 $ ] corresponds to the momentum , and @xmath226 $ ] corresponds to the angular momentum and lorentz boost . to analyze the irreducible representation of the algebra in eqs . ( [ 36 ] ) , ( [ 38 ] ) , and ( [ 37 ] ) , wigner s little group technique @xcite may play a crucial role , as it does in the poincar group case . the simplest nontrivial irreducible unitary representation derived from little group analysis is given by a hilbert space spanned by simultaneous eigenstates of all the supertranslation charges @xmath227 , which are hermitian operators associated with @xmath228 and obey @xmath229 = 0 . \label{41}\ ] ] the eigenvalues of @xmath227 are identified as classical values of @xmath230 , which is obtained by rescaling the supertranslation charge in eq . ( [ 42 ] ) by the black hole scale @xmath231 as @xmath232 } \exp\left ( 2\pi in_{y}\frac { y}{l_{y}}\right ) \exp\left ( 2\pi in_{z}\frac{z}{l_{z}}\right ) , \ ] ] where @xmath233 is a regular coordinate transformation in the @xmath134 plane . under the above hypothesis , we can evaluate @xmath234 as @xmath235 } \exp\left ( 2\pi in_{y}\frac{y}{l_{y}}\right ) \exp\left ( 2\pi in_{z}\frac{z}{l_{z}}\right ) \right\vert \\ & = \left\vert \int\frac{dydz}{l_{y}l_{z}}\exp\left ( 2\pi in_{y}\frac { y(y , z)}{l_{y}}\right ) \exp\left ( 2\pi in_{z}\frac{z(y , z)}{l_{z}}\right ) \right\vert \\ & \leq\left\vert \int\frac{dydz}{l_{y}l_{z}}\right\vert = 1.\end{aligned}\ ] ] hence , the range of eigenvalues is independent of @xmath231 . in quantum gravity , let us imagine that the number of quantum analogs of @xmath236 is such that the number of eigenvalues becomes finite and of order one with respect to @xmath231 : @xmath237 then , using the cutoff in eq . ( [ 44 ] ) , the degeneracy of the supertranslation charge is estimated as @xmath238 and yields an entropy of the same order as the black hole entropy . @xmath239 therefore , it is possible that a huge number of charged states on the horizon accounts for the statistical mechanical origin of @xmath0 . even though exact state counting of the charges on the horizon in nonperturbative quantum gravity has not been achieved yet and is a crucial open question , we no longer need to believe that black holes are hairless . in this section , we discuss the no - cloning problem of rindler horizons . let us consider quantization of the weak gravity field . its quantum field is given by @xmath240 the graviton component , @xmath241 , is a canonically quantized weak gravitational wave and obeys @xmath242 . the other component , @xmath243 , comes from quantized matter sources and is given by @xmath244 where @xmath245satisfies @xmath246 for the vacuum state @xmath247 in the standard canonical quantization . both components can contribute to supertranslation and superrotation quantum charges defined as @xmath248 & = \frac{1}{2\kappa^{2}}\int dydz\xi^{\tau } ( y , z)\left [ x^{+}\partial_{+}-1\right ] \hat{h}_{aa}^{(r)}(x^{+},0,y , z)\nonumber\\ & -\frac{2}{\kappa x^{+}}\int dydz\xi^{a}(y , z)\left [ x^{+}\partial _ { + } + 1\right ] \hat{h}_{-a}^{(r)}(x^{+},0,y , z ) , \label{47}\ ] ] where @xmath249 in the rindler gauge is defined from @xmath250 using eqs . ( [ 17 ] ) , ( [ 16 ] ) , and ( [ 18 ] ) as in classical theory . in this paper , we concentrate on the memory effect of @xmath243 . the quantum effect of @xmath251 on the holographic charges will be reported elsewhere . let us define the quantum gravitational memory operators on a future horizon at @xmath18 with a rindler wedge located at @xmath252 as @xmath253 \nonumber\\ & + \frac{1}{4\pi}\int dydz\int dy^{\prime}dz^{\prime}\xi^{a}(y , z)\partial _ { a}\nonumber\\ & \times\ln\left [ \frac{\left ( y - y^{\prime}\right ) ^{2}+\left ( z - z^{\prime}\right ) ^{2}}{\kappa^{2}}\right ] \left [ \int_{0}^{\infty } \partial_{-}\hat{t}_{++}\left ( x^{+}+x_{h}^{+},x_{h}^{-},y^{\prime } , z^{\prime}\right ) dx^{+}\right ] . \label{35}\ ] ] consider a massless matter field @xmath254 that is initially in the vacuum state @xmath247 . for the field @xmath254 , let us consider a local unitary operator @xmath255 that includes some information to be measured by observers . applying @xmath255 to @xmath247 generates a quantum wavepacket in an excited state , @xmath256 . the wavepacket crosses a future rindler horizon , as depicted in figure 5 . from the viewpoint of the entire minkowski spacetime , the information about @xmath255 is continuously carried by the wavepacket . however , from the viewpoint of the rindler spacetime , we have holographic charges on the horizon that store , at least partially , the information about @xmath255 , as depicted in figure 6 . then , an important problem is how much information about @xmath255 is stored on the horizon . in the black hole complementarity scenario @xcite of nonperturbative quantum gravity , all of the information can be copied and stored on the horizon . although this sounds incompatible with the no - cloning theorem @xcite , it may be possible to reconcile this discrepancy for realistic black hole cases because there is a singularity inside of the horizon that might completely delete quantum information that was carried by matter colliding with the singularity @xcite . however , in the minkowski spacetime , no singularity exists . thus , the black hole complementarity approach does not succeed . further , we have an infinite number of rindler horizons in the minkowski case . in principle , the information about @xmath255 can be simultaneously shared by many horizons , as depicted in figure 7 . of course , the classical component of the information about @xmath255 is replicable and easy to share . what happens if the holographic charges of two or more horizons store all of the purely quantum information about @xmath255 ? this looks very troublesome at a deep level of the full quantum gravity theory , and the no - cloning paradox may be inevitable if the holographic charges have reality for an observer who passes the horizons and observes the holographic charges for each horizon . to avoid that , we propose a simple conjecture about this question from the viewpoint of quantum measurement contextuality . first , it is worth noting that the energy - momentum tensor of quantum matter inside of a horizon does not commute with @xmath21 . this also leads to noncommutativity between the holographic charges of different horizons . for instance , let us consider a free massless scalar field @xmath254 . its @xmath257 component of the energy - momentum tensor is given by@xmath258 , where @xmath259 stands for normal ordering of the operators . a commutator for @xmath260 and @xmath261 is computed as @xmath262 = \langle0|\left [ \partial _ { + } \hat{\varphi}(x),\partial_{+}\hat{\varphi}(y)\right ] |0\rangle\left ( \partial_{+}\hat{\varphi}(x)\partial_{+}\hat{\varphi}(y)+\partial_{+}\hat{\varphi}(y)\partial_{+}\hat{\varphi}(x)\right ) . \label{003}\ ] ] here @xmath263 |0\rangle$ ] is proportional to @xmath264 . the integration in the definition of @xmath265 in eq . ( [ 35 ] ) includes @xmath261 , which has a nonvanishing commutation relation with @xmath260 in eq . ( [ 003 ] ) . hence , @xmath266 \neq0,\ ] ] and this results in the noncommutativity of the memory operators : @xmath267 \neq0.\ ] ] when we measure @xmath21 on a horizon , other @xmath268 values and quantum states of matter inside the horizon are affected because of the wavefunction collapse induced by the measurement . this suggests that the holographic charge reality is conditioned to measurements by appropriate physical detectors . as is well established , electric charge takes a universal value for a particle independently of observers and measurements . so it can be treated as reality . however , the holographic charge is not . rather , it emerges via measurements by appropriate physical detectors for measurements of the near - horizon metric . as depicted in figure 8 , the charge becomes physical only when we distribute the metric measurement devices in the space . without measurement devices , the charge is merely a gauge freedom of the general covariance in the minkowski background . then no cloning paradox arises , at least in the first order of perturbative quantum gravity . this is very similar to the case of unruh dewitt particle detectors @xcite @xcite for unruh radiation in a rindler spacetime . for hawking unruh particles in the minkowski vacuum state as well , quantum metric measurements result in the reality of the holographic charges . , created by applying a local unitary operator @xmath255 comes across a future rindler horizon.,height=207 ] is carried by the wavepacket @xmath269 . however , from rindler spacetime we have a holographic charge on the horizon , which stores the information of @xmath255.,height=207 ] can be simultaneously shared by an infinite number of rindler horizons.,height=207 ] here we comment on the duration of holographic charge measurements . as seen in eq . ( [ 35 ] ) , the charges are evaluated from time integrations of @xmath270 , which passes through the horizon . thus , the measurement is not achieved by any instantaneous measurement of the energy - momentum tensor of infalling matter on the horizon . however , one might expect that instantaneous measurements of the charges are still possible in order to directly detect the near - horizon metric in the rindler gauge at a later time . actually , in eq . ( [ 47 ] ) , the charges at a later time can be computed from an equal - time metric on the horizon in the rindler gauge . thus , if the metric in the rindler gauge at that time can be physically observed , the charges are fixed during an arbitrarily short time . however , this does not work because the metric in the rindler gauge can not be fixed uniquely . ambiguity exists because there are four arbitrary functions , @xmath271 , and @xmath272 in eqs . ( [ 17 ] ) , ( [ 16 ] ) , and ( [ 18 ] ) , so the values of the charges are not determined by instantaneous measurements . what we can do is to measure how much the charges increase during evolution of the infalling matter . a change in the charge is not observed by any instantaneous measurement at a fixed time . a measurement of the holographic charge outputs meaningful data only when the metric evolution is continuously monitored by metric detectors during the entire evolution . again , owing to gravitational interaction , the detectors inevitably interact with infalling matter during its evolution and share quantum entanglement , as depicted in figure 9 . thus , infalling matter is decohered . in this paper , a general theory of gravitational holographic charges for a ( 1 + 3)-dimensional linearized gravity field was formulated . the main result appears in eq . ( [ 4 ] ) . as a lemma , it is found that holographic states on the horizon can not store any information about absorbed perturbative gravitational waves . when we take into account second - order weak gravitational waves , the integrability of holographic charges in eq . ( [ 21 ] ) is broken for supertranslation . this raises the natural question of whether holographic charges are defined nonperturbatively for strong gravitational waves . a naive guess would be that it is impossible . because asymptotic symmetry on a horizon appears owing to the background metric isometry , it is natural to expect that a large departure from the background metric no longer has any symmetry and breaks the integrability of holographic charges for no specific reason . in section 3 , we proposed a conjecture to resolve the no - cloning paradox between infalling matter and holographic charges . if the reality concept of holographic charges is abandoned , no paradox occurs . the holographic charges are merely an emergent concept , and near - horizon metric measurement devices make observations as if the charges had some reality via entanglement between matter inside of the horizon and the detectors . as a discussion , an interesting question can be posed . do hawking unruh particles in realistic gravitational collapse , which propagate toward null future infinity , act as a metric detector near the horizon ? it is certainly true that the mode functions of the particles flush through a near - horizon region early in the gravitational collapse . then do the particles store any information about events on the horizon ? the possibility that the particles remember such information has been seriously discussed for a long time by many researchers , especially by t hooft @xcite and page @xcite . however , in the minkowski background with a rindler horizon , the hawking unruh particles observed by unruh de witt particle detectors seem to completely forget the information on the horizon . for instance , let us consider an aichelburg sexl shock wave @xcite passing through a rindler horizon . initially , the quantum fields of the hawking unruh particles are in the minkowski vacuum . after the shock wave passes , the quantum states of the fields are not excited and remain the vacuum state because the lorentz invariance of the spacetime prohibits particle creation by the shock wave . this means that no information about the shock wave is stored in the quantum fluctuation of the fields . thus , the hawking unruh particles observed later also remember nothing about the shock wave . contrary to the expectation of hps @xcite , the information loss problem seems to remain elusive even if we take account of the asymptotic symmetry on the horizon , although it may reveal the statistical mechanical origin of the bekenstein hawking entropy . we would like to thank song he and w.g . unruh for useful discussions . this research is partially supported by jsps kakenhi grant number 16k05311 , the foundational questions institute and silicon valley community foundation . a. almheiri , d. marolf , j. polchinski , and j. sully , j. high energy phys . * 02 * , 062 ( 2013 ) ; c.f . s. l. braunstein , arxiv:0907.1190 , published as s. l. braunstein , s. pirandola , and k. zyczkowski , `` better late than never : information retrieval from black holes , '' phys . rev . lett . * 110 * , 101301 ( 2013 ) . d. n. page , `` black hole information , '' in _ proceedings of the 5th canadian conference on general relativity and relativistic astrophysics _ , university of waterloo , 1315 may , 1993 , ed . by r. b. mann and r. g. mclenaghan ( world scientific , singapore , 1994 ) , pp . 141 ; d. n. page , phys . lett . * 71 * , 3743 ( 1993 ) .

we apply a generalized spin - basis analysis to associated higgs production and gauge boson pair production at lep . this framework allows us to identify a choice of spin axes for the processes which leads to strikingly different correlations among the decay products , even well above threshold . this spin basis optimizes the difference in the angular correlations for these two processes . in contrast , the same distributions display little contrast when the helicity basis is used . we also apply this technique to the case of boson pair production . 00b$]-maximized # 1#2 # 1#2 # 1 # 1 # 1#2

hep-ph9803410

the search for a light higgs boson and the study of gauge boson pair production @xcite are important physics goals of the lep2 upgrade at cern . if kinematically accessible to lep2 , the higgs boson will be primarily produced via the bjorken process @xmath3@xcite . an irreducible physics background to this processes is the production of @xmath4 boson pairs , which is most troublesome when @xmath5 . the clear separation of these two processes is required for a convincing higgs boson discovery or lower bound mass limit . kinematics as well as characteristic angular correlations will be useful in this separation . the study of @xmath1 boson pair production is also interesting in its own right@xcite , as it involves vertices coupling three gauge bosons ( especially @xmath6 ) , vertices which to date are poorly - probed experimentally . these couplings may be probed with varying degrees of sophistication by testing the predictions for the total cross section , production angle distribution , and correlations among the vector boson decay products . hence for all of these processes , @xmath0 and @xmath7 , we are keenly interested in the angular correlations among the final state particles . to disentangle these correlations it is very useful to understand the spin correlations of the heavy particles produced . these spin correlations as well as the subsequent correlations of the decay products are the focus of this paper . until recently , most spin - related studies were carried out within the framework provided by the zero momentum frame helicity basis ( see , for example refs . @xcite ) . for particles which are ultra - relativistic , such as at the next generation of linear colliders , this is appropriate . however , for particles which are only moderately relativistic , there is no reason to expect that the helicity basis will produce the best description of the physics involved . indeed , for the case of @xmath8 production at low energy @xmath9 and @xmath10 colliders , it was shown in refs . @xcite that the helicity basis is far from optimal . in this paper we use the generalized spin basis introduced in ref . @xcite to describe @xmath11 , @xmath12 , and @xmath13 production and decay at lep2 energies . in particular , we write the polarized production amplitudes in terms of the most general @xmath14-conserving choice of spin axes , and from these expressions determine the choice of axes which leads to the most distinctive correlations among the final state particles . we argue that it is advantageous at lep2 energies to tune the choice of spin axis to the experimental issue being investigated , rather than to simply use the helicity basis and try to unravel the contributions from what are in many cases nearly - equally populated spin states . for all three processes , @xmath0 and @xmath7 , much effort in the literature has been devoted to the analysis of various anomalous trilinear couplings@xcite , employing , of course , the helicity basis . rather than attempt to rework all of those studies in terms of the generalized spin - basis , we have chosen to concentrate on the picture within the standard model . it is clear from our results , however , that such a study would be worthwhile , as we expect different spin bases to be optimal for the extraction / limitation of different anomalous couplings . such an analysis would be a major extension of this work . the remainder of this paper is organized as follows . after a brief discussion of our notation and conventions in sec . [ xibasis ] , we consider the @xmath15 system within the generalized spin basis framework ( sec . [ zhandzz ] ) . for each process , we consider in detail the polarized production cross sections , and construct spin bases suggested by the expressions for the amplitudes . we consider the decay of polarized @xmath4 s , and then link the production and decay to study the correlations among the decay products . we are able to construct a basis in which these correlations differ significantly for the @xmath11 and @xmath12 cases , whereas in the helicity basis the correlations are virtually non - existent . sec . [ wwsection ] contains the corresponding analysis for the @xmath16 system . in this case , matters are less clear - cut , and we present several suggestions for bases which may be useful under different circumstances . finally , sec . [ conc ] contains our conclusions . explicit expressions for the polarization vectors we employ for the massive vector bosons in the generalized spin basis appear in the appendix . to describe the polarized production cross sections for @xmath11 , @xmath12 , and @xmath13 discussed in this paper , we employ the generic spin basis introduced by parke and shadmi in ref . @xcite , as illustrated in fig . [ xidef ] . for @xmath12 and @xmath13 , this is the most general basis which conserves @xmath14 . we label the two particles produced in the collision by @xmath17 and @xmath18 . the zero momentum frame ( zmf ) production angle @xmath19 is defined as the angle between the electron and @xmath17 directions . the spin states for the first particle are defined in the its rest frame , where we decompose the @xmath17 spin along the direction @xmath20 , which makes an angle @xmath21 with the @xmath18 momentum in the clockwise direction . likewise , the @xmath18 spin states are defined in its rest frame along the direction @xmath22 , which makes the _ same _ angle @xmath21 with the @xmath17 momentum , also in the clockwise direction . we denote the two transverse polarization states by @xmath23 and @xmath24 and the state by 0 . throughout this paper we use the terms `` transverse '' and `` longitudinal '' to refer to directions relative to the spin axis , _ not _ the direction of motion of the particle . a generic vector boson spin will be designated by @xmath25 . if we sum over all of the polarizations of the vector boson(s ) , then the dependence on @xmath21 drops out of the result . note that although the higgs boson is a scalar , we may still define a spin axis for it as if it were a vector . the spin zero character of the higgs will be reflected in a lack of any dependence on the choice of this axis . within the generic spin basis framework , specific spin bases are defined by stating the relationship between @xmath21 and @xmath19 ( and any other relevant event parameters ) . for example , the ubiquitous helicity basis is defined by fixing . [ helicitydef ] in this case , the spins are defined along the directions of motion of the particles as seen in the zmf . another interesting basis is the beamline basis @xcite , which is defined by = ; = - . [ bmldef ] here @xmath26 is the zmf speed of @xmath17 . in this basis , the spin axis for @xmath17 is the electron direction . furthermore , if @xmath17 and @xmath18 have identical mass , then the spin axis for @xmath18 is the positron direction . later in this paper , we will encounter additional bases , whose definitions are inspired by the form of the matrix elements for the processes under consideration . except for the fermion masses , which we set equal to zero , all input masses and coupling constants used in the computations presented in this paper are the central values as reported in the 1996 review of particle properties@xcite . furthermore , we also neglect the coupling between the electron and higgs . if the higgs is light enough to be observed at lep2 , then its production will be dominated by the process @xmath28 , for which the largest background is @xmath29 . this suggests the potential for difficulties should the higgs mass lie too close to the mass of the @xmath4 . in particular , at @xmath30 , the tree - level cross section for @xmath11 is only 0.5 pb when @xmath31 , compared to 1.2 pb for @xmath12 . as noted by brown @xcite , this situation improves if @xmath32-tagging is employed , as the higgs decays mainly to @xmath33 whereas the @xmath4 decays to this final state only 15% of the time . rather than rely on @xmath32-tagging , kunszt and stirling @xcite have considered the distribution of the angle that the decay leptons in @xmath34 . make with the direction of the beam in the zmf . they find a significant difference in this distribution in the @xmath11 and @xmath12 cases . finally , summers @xcite has added the effect of polarizing the beams on the distribution of ref . @xcite . in this section we examine the @xmath11 and @xmath12 processes within the generalized spin basis framework . we begin in sec . [ polzhprod ] by analyzing the polarized production amplitudes for the process @xmath35 in great detail . the expressions for these amplitudes will show us how to construct a basis in which only transversely - polarized @xmath4 s are produced in association with the higgs . in sec . [ polzzprod ] we present polarized production amplitudes for @xmath36 . after describing the decay of polarized @xmath4 s ( sec . [ polzdec ] ) , we combine the production and decay information to discuss the angular correlations among the decay products in the @xmath15 system ( sec . [ zzcorrel ] ) . our analysis shows that the angular variable suggested by kleiss and stirling @xcite , while close to optimal , may be improved upon , especially as the @xmath11 pair is produced further and further above threshold . in particular , we find that in the basis constructed in sec . [ polzhprod ] , certain decay angular distributions are very different for @xmath11 versus @xmath12 . in contrast , these same distributions are nearly featureless in the helicity basis . we conclude this part of the paper in sect . [ pseudo ] with a brief look at how the production of a pseudoscalar higgs would differ from the standard model higgs . although we focus on the case where @xmath37 , our results are more general , since the existence of our optimized basis for @xmath28 depends only on that process being kinematically allowed . it is also potentially useful at the tevatron , because the amplitude for @xmath38 has the same spin structure as the amplitude for @xmath35 . to describe the process @xmath35 , we take particle @xmath17 in fig . [ xidef ] to be the @xmath4 , and particle @xmath18 to be the higgs . in general , the higgs and @xmath4 have different masses ( @xmath39 and @xmath40 respectively ) , leading to the following connection between the center - of - mass energy @xmath41 and speed @xmath42 of the @xmath4 boson : = . [ betaz ] rather than use eq . ( [ betaz ] ) to eliminate one of @xmath43 or @xmath42 from our expressions , it is more convenient to use both quantities . if we neglect the electron mass , there is but a single diagram for @xmath35 . it leads to the differential cross section = g_f^2 m_w^2 & & m_z^3 ( s+m_z^2-m_h^2 ) ^2 & & \ { ^2 2_w [ _ l^(,,)]^2 + 4 ^4 _ w [ _ r^(,,)]^2 } [ zhsigma ] where @xmath44 is the fermi coupling constant , @xmath45 is the mass of the @xmath1 boson , @xmath46 is the weinberg angle , and @xmath47 is the usual relativistic factor , @xmath48 . the two terms in the curly brackets come from the two possible chiralities of the initial electron line . all of the spin information is contained in the spin functions , @xmath49 , which are given by _ l^ ( , , ) = _ r^ ( , , ) = 1 [ zhtranss ] and _ l^0 ( , , ) = _ r^0 ( , , ) = - . [ zhlongs ] summing over the three possible spins of the @xmath4 we obtain the total ( unpolarized ) differential cross section _ d^ = g_f^2 m_w^2 & & m_z^3 ( s+m_z^2-m_h^2 ) ^2 & & \ { 1 - 4 ^ 2_w+8 ^ 4_w } \ { 2(1-^2 ) + ^2 ^ 2}. [ zhsigmatotal ] eq . ( [ zhsigmatotal ] ) is independent of the spin axis angle @xmath21 , as it must be . since , in general , at each value of @xmath19 we are allowed to choose a different value of @xmath21 , it is useful to plot these amplitudes in the @xmath50-@xmath51 plane . consequently , we define the quantity f^ ( , , ) , [ zhfraction - def ] which is the fraction of the total cross section coming from the spin state @xmath25 . in fig . [ zh - prodcontours ] we have plotted @xmath52 for a collider energy @xmath30 and higgs mass @xmath53 . these plots illustrate the features of the amplitude discussed below , and are an indispensable aid when the amplitude becomes complicated , as in the @xmath12 and @xmath13 cases . for @xmath54 , we recover the usual helicity basis expressions from eqs . ( [ zhsigma])([zhlongs ] ) : _ l^ ( , , ) = _ r^ ( , , ) = & & 1 ( 1 - ) ; _ l^0 ( , , ) = _ r^0 ( , , ) = & & . at high energy , the transverse amplitudes die off , leaving only the longitudinal amplitude . an examination of eq . ( [ zhtranss ] ) reveals that it is not possible to make both of the transverse spin functions @xmath55 and @xmath56 vanish simultaneously . consequently , it is impossible to make either of the transverse amplitudes of eq . ( [ zhsigma ] ) vanish . there is , however , a zero in the longitudinal contribution , eq . ( [ zhlongs ] ) , which results from choosing = ; = , [ zhideal ] @xmath57 . the basis corresponding to this choice , the basis , will turn out to be very useful in the study of @xmath11 events . the nonvanishing spin functions in the basis are _ l^(,,^-1 ( ) ) & = & _ r^(,,^-1())&= & 1 . these functions are completely flat in @xmath50 at threshold , and become proportional to @xmath58 for @xmath59 . note that the existence of this basis does not depend on either the machine energy or the higgs mass : eqs . ( [ betaz ] ) and ( [ zhideal ] ) remain well - defined so long as @xmath60 . in particular , for @xmath61 , we have 1 ; 0 , [ zhhighe ] that is , @xmath62 . this is clearly _ not _ the helicity basis . the basis is also the basis in which the @xmath63 and @xmath64 components are each maximized . for completeness , we mention the basis , defined by @xmath65 . as its name suggests , the basis maximizes the fraction of @xmath4 s . this basis is potentially useful for large values of @xmath66 since the fraction of @xmath4 s with increasing energy approaches unity faster than in the helicity basis . in table [ zh - breakdowns ] , we list the contributions to the total amplitude for each of the three possible @xmath4 spins as measured in the helicity , beamline , , and bases . we employ a center - of - mass energy @xmath67 and a higgs mass @xmath53 , the value which makes separation of @xmath11 and @xmath12 events the most difficult . note that in the helicity basis , the three spin states are populated nearly equally . as we shall see below , this feature makes the helicity basis a very poor choice for studying the @xmath15 system . because the higgs mass is still unknown , we might worry that the predictions for the spin compositions would depend strongly on @xmath39 , rendering this discussion pointless . for any particular machine energy and higgs mass , eq . ( [ betaz ] ) tells us the appropriate value of @xmath42 to use . in fig . [ zhbetaplot ] we have plotted the fractional contribution of each of the three spin components as a function of @xmath42 for the helicity , beamline , , and bases . we see that in the basis , the spin composition is remarkably flat as a function of @xmath42 , except near @xmath68 . in fact , for @xmath69 , @xmath70 when @xmath67 . so , for the values of @xmath39 and @xmath41 accessible at lep , the spin composition in the basis is effectively constant . contrast this to the beamline basis , which actually coincides with the basis at @xmath71 and the helicity basis for @xmath72 ! in this case , the spin composition is not particularly stable . a second reason for preferring a basis in which the spin composition is stable with respect to changes in @xmath42 relates to the effects of initial state radiation ( isr ) . although we have neglected these effects in our computations , it is apparent what should happen qualitatively : some of the observed events will be produced from @xmath73 pairs with a center - of - mass energy lower than the machine energy . therefore , the smaller the dependence on @xmath42 , the less sensitive the breakdowns will be to the effects of isr . finally , in fig . [ zh - cth ] we compare the production angular distributions of the polarized cross sections in the helicity and bases . we see that in the helicity basis , the identity of the dominant component depends on @xmath50 , whereas in the basis , the ratio of the two non - vanishing amplitudes is essentially constant . we now turn to the process @xmath36 . with our comparison to @xmath11 production in mind , we identify the @xmath4 which decays leptonically with @xmath17 and the @xmath4 which decays to @xmath33 with @xmath18 in fig . [ xidef ] . the only diagrams for @xmath12 production are @xmath74-channel electron exchange diagrams , one for each ordering of the two @xmath4 s . for an initial fermion pair with left - handed chirality we have & & = g_f^2 m_w^4 ^2 ^2 , [ prodzzleft ] while the result for right - handed chirality reads & & = g_f^2 m_w^4 ^2 ^2 . [ prodzzright ] these expressions contain the common zmf speed of the two @xmath4 s , which is connected to the center - of - mass energy by = . [ plainbeta ] we have not combined the pairs of terms appearing in the square brackets in order to make manifest the similarities to the @xmath1-pair amplitudes presented in sec . [ wwsection ] . all of the spin information is contained in the spin functions , @xmath75 , which may be written as _ l^ ( , , ) = 2_1(,,)_2 ( , , ) [ funcab ] _ l^ ( , , ) = 2_3(,,)_4 ( , , ) [ funcaa ] _ l^ ( , , ) = _ l^0 ( , , ) = [ funcaoob ] _ l^00 ( , , ) = 2(1 - 2+^2 ) -4_3(,,)_4 ( , , ) [ funcoo ] where _ 1 ( , , ) & = & ( 1++ ) - ( + ) _ 2 ( , , ) & = & ( - ) + ( + ) _ 3 ( , , ) & = & ( - ) + _ 4 ( , , ) & = & - ( - ) . [ gfuns ] since the replacement @xmath76 has the effect of interchanging the @xmath23 and @xmath24 states , expressions for the remaining spin combinations may be obtained from the relation _ l^| ( , , ) = _ l^-,-| ( , , ) . [ leftother ] the total differential cross section is remarkably simple . summing over the spins of both @xmath4 s and including both fermion chiralities we obtain _ , |,c d_c^| = & & g_f^2 m_w^4 ( ^42_w + 16 ^ 8 _ w ) & & ( 1-^2 ) . [ zztotal ] of the nine spin configurations , there are only six which are independent , since @xmath14 invariance forces the equality of the @xmath77 and @xmath78 components , as well as the equality of the @xmath79 and @xmath80 components @xcite . therefore , for the rest of the paper we will refer to @xmath81 $ ] , @xmath82 $ ] , and @xmath83 $ ] as single components . one of the ways which the @xmath84 and @xmath85 violation present in the weak interactions manifests itself as the different prefactors in eqs . ( [ prodzzleft ] ) and ( [ prodzzright ] ) . the spin functions , on the other hand , obey these symmetries , and are connected by _ r^| ( , , ) = _ l^-,-| ( , , ) . [ leftright ] because of the somewhat complicated form of the amplitudes , it is instructive to plot the amplitudes for each of the six independent spin configurations in the @xmath50-@xmath51 plane . in fig . [ zz - prodcontours ] , we have plotted the quantity f^| ( , , ) [ fraction - def ] ( _ i.e. _ the fraction of the total amplitude in a given spin configuration ) for a machine center - of - mass energy of 192 gev . visible in the plots is the fact that the @xmath86 contribution is exactly twice the @xmath81 $ ] contribution for all values of @xmath19 and @xmath21 [ see eqs . ( [ prodzzleft ] ) , ( [ funcaa ] ) and ( [ funcoo ] ) ] . also noteworthy are the broad minima in the @xmath87 , @xmath88 , @xmath86 , and @xmath81 $ ] components in the vicinity of the diagonal . the @xmath82 $ ] and @xmath89 $ ] contributions have corresponding maxima in this region . in table [ zz - breakdowns ] we give the fraction of the total amplitude coming from each of the six independent spin configurations in the four bases tabulated for @xmath11 . in two of the bases ( helicity and ) , the high degree of symmetry present in eqs . ( [ prodzzleft ] ) and ( [ prodzzright ] ) resulting from the two identical final state particles manifests as the equalities @xmath90 and @xmath82=[(0+)+(-0)]$ ] . it is important to note that we _ define _ the basis in terms of eqs . ( [ betaz ] ) and ( [ zhideal ] ) , even when we are dealing with @xmath12 production . thus , the spin breakdowns in this basis depend on the higgs mass via the value assigned to @xmath66 . for @xmath91 ( the interesting region ) , the curve traced out by the basis in the @xmath50-@xmath51 plane lies near the diagonal , an area in the @xmath50-@xmath51 plane where the contributions from the various spins are relatively stable . thus , it is not surprising to find that for @xmath91 and @xmath67 , there are no significant deviations from the entries for the basis in table [ zz - breakdowns ] . in fig . [ zzbetaplot ] we have have plotted the fraction of the total amplitude in each of the spin components using the helicity and bases as a function of @xmath26 . in the helicity basis , we find that the @xmath92 and @xmath93 contributions are always equal , as are the @xmath94 $ ] and @xmath95 $ ] contributions . consequently , if we look at the population of the spin states for one of the @xmath4 s while summing over the other @xmath4 , we find that the @xmath23 and @xmath24 states are populated equally . furthermore , at threshold , this inclusive breakdown is precisely @xmath96 for each spin , and this changes relatively slowly with increasing @xmath26 . the situation is considerably better in the basis , where no two of the six spin contributions are equal , and where there is a reasonable difference in the inclusive contributions . while we have used @xmath97 ( _ i.e. _ @xmath31 ) in preparing these plots , we have verified that for the range accessible to lep , the dependence on the higgs mass is inconsequential . in fig . [ zz - cth ] we compare the angular distributions of the polarized production cross sections in the helicity and bases at @xmath67 . in the helicity basis there is a complicated interplay among the six spin components , whereas in the basis the two dominant components appear in approximately the same ratio , independent of @xmath50 . the @xmath98 coupling violates both parity and flavor universality . thus , the angular distributions for the decay of polarized @xmath4 bosons are forward - backward asymmetric , and depend on which fermions appear in the final state . neglecting the mass of the final state fermions , ) and ( [ zl ] ) . these effects are greatest for @xmath32-quarks , where they are less than a few percent . ] we obtain the following angular distributions in the rest frame of the decaying @xmath4 : = 38[zt ] for the transverse polarizations and = 34 ^2[zl ] for the polarization . these distributions have been normalized to unit area by inclusion of the partial width @xmath99 for the decay @xmath100 . we take @xmath101 to be the angle between the direction of motion of the fermion and the spin axis as seen in the @xmath4 rest frame , and absorb the dependence on the couplings between left- and right - handed chirality fermions to the @xmath4 into the factor @xmath102 . for convenience , we have collected the values of @xmath102 in table [ zdecaytable ] . the corresponding distributions are plotted in fig . [ zdecaydist ] . unfortunately , the decays with the most distinct distributions , @xmath103 are invisible . furthermore , charge and flavor identification for decays to quarks ( which appear as two jets ) is virtually impossible , except perhaps for @xmath32 quarks . for @xmath104 , we have a fairly large overlap between the @xmath63 and @xmath64 distributions . all of these features lead us to conclude that it may not be possible to distinguish between the @xmath63 and @xmath64 polarizations . however , as we shall see below , the difference between the @xmath105 and @xmath106 states is sufficient for separating @xmath11 and @xmath12 , provided that the appropriate spin basis is used . naturally , since the higgs is a scalar , its decay angle distributions will be completely flat . in this section we will put everything together to discuss the correlations among the @xmath15 decay products . for concreteness , we take the final state to be @xmath107 , the @xmath108 coming from a decaying @xmath4 , and the @xmath33 from the higgs or other @xmath4 . the results we present do not change very much if we consider a @xmath109 final state : since it is not possible to tell which light jet is the @xmath110 and which is the @xmath111 , all distributions in the @xmath4 decay angle @xmath112 defined below would be symmetric about @xmath113 . on the other hand , @xmath114 is only slightly larger than @xmath115 , leading to small asymmetries between positive and negative values of @xmath116 . in order to display the advantages of using the basis defined by eq . ( [ zhideal ] ) , we consider a scatter plot in the decay angles @xmath112 and @xmath117 , defined as follows . on the @xmath118 side of the event , let @xmath112 be the angle between the @xmath119 emission direction and the @xmath4 spin axis in the @xmath4 rest frame . on the @xmath120 side of the event , let @xmath117 be the angle between the @xmath32 emission direction jets , then the distributions presented here should be folded about @xmath121 . ] and the @xmath122 spin axis in the @xmath122 rest frame . in fig . [ zz - zhdecaycontours ] we present this distribution for @xmath11 and @xmath12 production for the helicity and bases for @xmath123 and @xmath31 . the difference between the two plots is dramatic : in the helicity basis , the two distributions are nearly flat , while there is significant structure when the basis is used . this difference is underscored by the plots in fig . [ zzvszh1d ] , which compare the 1-dimensional distributions in @xmath116 and @xmath124 . clearly , the angular information contained in an analysis employing the basis provides a useful handle in the separation of @xmath11 and @xmath12 events , even when @xmath5 . in contrast , the helicity angles are virtually useless in this respect ! kunszt and stirling @xcite have noted significant angular correlations in the zmf between the direction of the final state muons and the beam direction . effectively this is the beamline basis . at threshold , the beamline basis coincides with the optimal ( ) basis , and the zmf coincides with the @xmath4 rest frame . at lep2 , @xmath125 , _ i.e. _ not that far above threshold . thus , even though they do nt use the optimal basis , kunszt and stirling do see non - trivial angular correlations . the largest possible correlations , however , are obtained using the muon direction relative to the basis spin axis in the @xmath4 rest frame . furthermore , the gap between their distribution and the optimal one widens as we move further above threshold . an issue which may be raised concerns the effect of finite detector resolution on the ability of the experimentalists to accurately measure the decay angles we have proposed , especially since they have been defined in the rest frames of the parent particles , not the lab frame . to get an estimate of the size of these effects , we introduced gaussian energy smearing with a width set by @xmath126 into the energies of the charged leptons and the @xmath32 quarks , and then compared the decay angles reconstructed solely from the smeared momenta to the actual decay angles . we take the root mean square deviation of this difference to be an estimate of the error introduced by the smearing . since the definition of @xmath21 depends both on the speed of the parent boson as well as the production angle , the detailed values of our error estimate will vary with the machine energy , choice of basis , and the process ( @xmath11 or @xmath12 ) in which the bosons were produced . typical values at @xmath30 and @xmath31 are @xmath127 , for all three bases . to see the effects of this angular smearing , an experimentalist would need to use bins of approximately this size or smaller . the total number of events observed will probably not be large enough to warrant such a large number of bins in the @xmath128 plots . although a more detailed study using the full - blown detector simulation is clearly needed , it would appear that it is possible to measure the indicated angles with sufficient precision . the amplitudes for the processes @xmath129 can be written as a sum over the spin states of the intermediate massive particles of the production times the decay amplitudes for each spin component . in the classical limit there is no interference between these amplitudes . however , in general , these amplitudes do interfere . the size of these interference terms can be used as a measure of how well the classical result for a particular basis describes the full process : if the interference terms are small , then the classical picture captures nearly all of the physics . on the other hand , if the interference terms are large , a significant amount of the physics is omitted from the classical picture . to quantify these considerations , at each kinematic point for a given process , we may define the quantity @xmath130 , the value of the interference terms for a given spin basis divided by the total @xmath131 matrix element squared . for a process containing @xmath132 independent intermediate spin configurations , @xmath130 can range from @xmath133 ( total destructive interference ) to @xmath134 ( total constructive interference ) . a plot of @xmath135 displays the relative importance of the various values of @xmath130 . clearly , in the ideal case where classical intuition captures all of the physics , this distribution will be a delta function centered at @xmath136 . [ zhinf ] compares the helicity , beamline ant bases for @xmath11 production and decay , whereas fig . [ zzinf ] does the same for @xmath12 . for the signal ( @xmath11 ) , we see that the peaking in @xmath130 is about an order of magnitude greater in the basis than in the beamline basis . the helicity basis shows essentially no peaking . for the background ( @xmath12 ) , the difference is not as dramatic , but it is still clear that the basis gives the smallest interference terms , and hence the superior description . although the higgs boson of the standard model is unambiguously @xmath14-even , for the sake of comparison , it is useful to consider how the spin correlations differ for a @xmath14-odd higgs @xmath137 ( commonly referred to as a pseudoscalar ) , a feature of two - doublet higgs models . although a tree - level @xmath138 coupling is forbidden , such a coupling can be generated at the one - loop level @xcite . following ref . @xcite , we write the ( effective ) @xmath138 vertex as - igm_z k_1^ k_2^ _ , where @xmath139 and @xmath140 are the 4-momenta of the two @xmath4 s , @xmath141 is a dimensionless coupling constant , and @xmath142 is the mass scale at which this vertex is generated . this coupling leads to the differential cross section = g_f^2 m_w^2 & & ( m_z ) ^4 ( s+m_z^2-m_h^2 ) ^2 & & \ { ^2 2_w [ _ l^(,)]^2 + 4 ^4 _ w [ _ r^(,)]^2 } . [ zasigma ] in this case the spin functions are independent of energy : _ l^ ( , , ) = _ r^ ( , , ) = & & 1 ( ) ; _ l^0 ( , ) = _ r^0 ( , ) = & & [ zaspinfun ] thus , the total ( unpolarized ) differential cross section is _ d^ = g_f^2 m_w^2 & & ( m_z ) ^4 ( s+m_z^2-m_h^2 ) ^2 & & ( 1 - 4 ^ 2_w+8 ^ 4_w ) ( 2 - ^2 ) , [ zasigmatotal ] in agreement with ref . @xcite . it is clear from the especially simple form of the spin functions , eq . ( [ zaspinfun ] ) , that the optimal basis for @xmath143 is the helicity basis , independent of the machine energy . only the helicity basis has the property that one of the three amplitudes ( the longitudinal one ) vanishes . because the @xmath21 dependence of the @xmath11 and @xmath143 amplitudes differs so greatly , a measurement of the @xmath4 spin composition of a higgs signal for different choices of @xmath21 could potentially differentiate between the two cases . in particular , it would be useful to measure the fraction of longitudinally polarized @xmath4 s in a sample of @xmath4-higgs candidates in both the helicity and bases . for the signal events , a scalar higgs should show no longitudinal @xmath4 s in the basis , while for a pseudoscalar higgs this would occur only in the helicity basis . with the crossing of the @xmath1-pair production threshold at lep2 , it is now possible to probe the electroweak sector of the standard model in new ways . the tree - level diagrams for @xmath145 contain triple gauge boson vertices ( @xmath146 and @xmath147 ) , allowing direct studies of the yang - mills structure of the theory . the precise form of these couplings predicted by the @xmath148 nature of the underlying theory is reflected in the various angular distributions associated with the @xmath13 events @xcite . if we allow the strengths of the existing @xmath146 and @xmath147 couplings to vary , or if we introduce new ones , then not only will the differential distribution in the production angle @xmath19 be affected , but the shares of the total cross section coming from the different spin states of the @xmath1 s will also vary . this second effect will show up in the distributions in the angles between the decay products and suitably defined spin axes . as mentioned in the introduction , a complete reworking of the anomalous trilinear couplings analysis using our generalized spin basis , while worthwhile , would be a major extension of this work . nevertheless , the general means of proceeding should be clear : for each anomalous coupling a computation of the polarized differential cross sections as a function of @xmath21 is required . by examining the functional form of these cross sections ( with the aid of contour plots in the @xmath50-@xmath51 plane ) , a basis may be constructed which highlights the deviations introduced by the anomalous coupling . the decay angular distributions in this basis then encode the consequences of the new physics being investigated . in the standard model , three diagrams contribute to the total @xmath149 cross section . we regroup these three diagrams into three contributions as follows . first , we note that the two diagrams involving an @xmath43-channel photon or @xmath4 boson contribute for both possible chiralities of the initial fermion line . thus , the first two contributions to the total cross section will be referred to as the @xmath150 and @xmath151 pieces . is trivial in the sense that it shows up as a ( @xmath1 ) spin - independent prefactor : no choice of spin basis for the @xmath1 s can make these two contributions easier or harder to separate . note that the contribution from the photon alone may be determined from eqs . ( [ prodwwleft ] ) and ( [ prodwwright ] ) by taking the @xmath4 mass to infinity , and dropping the last term of eq . ( [ prodwwleft ] ) . ] the third diagram involves the @xmath74-channel exchange of a neutrino and contributes only when the initial fermion line is left - handed . it will form the third contribution , which we will refer to as the neutrino piece . before presenting the differential cross sections for polarized production , let us consider the relative size of these three contributions to the total cross section . in table [ diagcontribs ] , we have tabulated the values for each piece at tree - level for a machine energy of 192 gev . we see that the @xmath151 contribution is insignificant . furthermore , there is a non - trivial interference term , which we will denote by @xmath152 , between the @xmath150 and neutrino contributions . contribution and the left - handed pieces . ] in fig . [ infzplot ] , we have plotted the squares of the three contributions as well as the interference term @xmath152 as a function of the zero momentum frame production angle @xmath19 , which is taken to be the angle between the momentum of the electron and the @xmath153 . although it is tempting to attribute the well - known peaking of the cross section in the forward direction to the @xmath74-channel pole in the neutrino diagram , fig . [ infzplot ] indicates that this is not the case . the square of the neutrino diagram is not only well - behaved near @xmath154 , but even _ decreases _ slightly in that region . instead , the shape of the differential cross section is dominated by the interference between the @xmath150 and neutrino pieces , which changes sign from @xmath64 to @xmath63 for increasing @xmath50 . these general features become even more pronounced at higher center - of - mass energies . to describe the polarized production cross sections for @xmath155 , we choose to identify the @xmath153 with particle @xmath17 and the @xmath156 with particle @xmath18 in fig . [ xidef ] . we now present the differential cross sections for polarized @xmath157 production . when the initial fermion pair has left - handed chirality , we have & & = ^2 & & \ { 2(m_w / m_z)^2 - ( 1-^2)^2_w & & - 2 _ l^| ( , , ) } ^2 . [ prodwwleft ] here @xmath25 ( @xmath158 ) is the spin of the @xmath153 ( @xmath156 ) . the first term in the curly brackets is the @xmath159 contribution , while the second term comes from the neutrino . for the other initial fermion chirality ( right - handed ) , we have simply = g_f^2 m_w^2 ^4_w ^2 . [ prodwwright ] all of the spin information in contained in the @xmath160 , which are the same spin functions defined in connection with @xmath12 production , namely eqs . ( [ funcab])([leftright ] ) . naturally , we should re - interpret @xmath26 as the common zmf speed of the two @xmath1 s . the presence of the neutrino contribution to eq . ( [ prodwwleft ] ) makes the expression for the total differential cross section somewhat messy : _ , |,c d_c^| = & & & & & & ^3\ { 16 + ^2 } + g_f^2 m_w^2 & & \ { 2 + 12 ^ 2 ^ 2 ^ 2 + 2 ^ 2(1-^2)^2 } - & & & & \ { ( 1+^2)[4+^2 ^ 2 ^ 2 ] + 2(1-^2)[^2 ^ 2- 2(1-^2 ) ] } . [ wwtotal ] the successive terms in eq . ( [ wwtotal ] ) come from the square of the @xmath159 contributions ( summed over both @xmath73 chiralities ) , the square of the neutrino contribution , and the interference term @xmath152 . as in the @xmath12 case , the individual spin - dependent amplitudes are complicated . therefore , in fig . [ prodcontours ] , we have plotted the fractional contribution to each of the six independent spin configurations for a center - of - mass energy of 192 gev . from these plots , we see that the @xmath161 $ ] and @xmath86 contributions never dominate the total amplitude at 192 gev , while the @xmath162 $ ] or @xmath163 contributions can be made large with the proper choice of @xmath21 . from the contour plots and the expressions for the spin components we have identified four interesting spin bases : helicity , , beamline , and 00b , which we have studied in some detail . in fig . [ wwsolns ] , we have plotted the connection between @xmath21 and @xmath19 at 192 gev for these four bases . in tables [ ww - breakdowns][infz - breakdowns ] , we indicate the fraction of the total cross section , @xmath159 , @xmath164 and @xmath152 contributions broken down into the six independent spin configurations for an @xmath73 collider running at 192 gev . we begin our survey with the helicity basis . this basis has one noteworthy feature : the @xmath163 and @xmath165 components of both @xmath159 contributions vanish at all @xmath41 . this is not true for the neutrino diagram ; therefore , a measurement of these spin components is a direct measurement of the neutrino contribution . at @xmath166 gev the sum of these two spin components is 56% of the total cross section . another useful basis is the basis , which is defined by choosing @xmath21 so that the @xmath92 component is as large as possible . in principle , it is a straightforward exercise to derive an analytic expression connecting @xmath21 and @xmath19 by differentiating @xmath167 with respect to @xmath21 and setting the result equal to zero . in practice , however , such an expression would be too complicated to be illuminating . furthermore , it turns out that we must allow negative values of @xmath168 in order to get the largest possible contribution . this is equivalent examining both the @xmath163 and @xmath93 amplitudes , and taking the larger of the two at each point . selecting @xmath169 whenever the @xmath93 amplitude is chosen converts it to a @xmath92 amplitude , according to eq . ( [ leftother ] ) . thus , it is expedient to use a numerical procedure to determine @xmath21 in this basis : the results are plotted in fig . [ wwsolns ] for @xmath123 . the discontinuity seen in the @xmath51 versus @xmath50 curve for this basis near @xmath170 is caused by the presence of two competing maxima . at @xmath171 the @xmath92 component is 64% of the total in this basis . in the beamline basis , defined by eq . ( [ bmldef ] ) , the direction of the spin axis for the @xmath153 ( @xmath156 ) viewed in its rest frame coincides with the direction of the electron ( positron ) in that frame . with the choice ( [ bmldef ] ) for @xmath21 , the functions @xmath172 and @xmath173 vanish . this means that the neutrino does not contribute to the @xmath174 $ ] , @xmath95 $ ] , and @xmath165 spin configurations @xcite . unfortunately , the fraction of the total in these spin configurations is small , less than 5% at @xmath166 gev . however , the @xmath175 $ ] component is 80% of the total at this @xmath41 . finally , we come to the 00b basis , which is defined by choosing the value of @xmath21 which maximizes the fraction of the amplitude coming from the @xmath94 $ ] component . once again the maximization condition leads to a complicated expression ( it is a quartic equation for @xmath176 ) . therefore , we present the numerically derived solution in fig . [ wwsolns ] . in this basis , the @xmath175 $ ] component is more than 92% of the total at @xmath177 gev . for these last three bases , , beamline , and 00b , there are large negative interference terms between the @xmath164 and @xmath150 contributions in all of the dominant spin components . [ wwcomposcth ] shows the angular dependence of the polarized production cross sections . it is clear from these plots that in the helicity basis , there is a complicated structure , with four different amplitudes being the largest , depending on the value of @xmath50 . on the other hand , in the basis , the @xmath87 component dominates at all angles . likewise , in the beamline and 00b bases , the @xmath82 $ ] component dominates at all angles . finally , because the relative contributions of the polarized cross sections depend upon the center - of - mass energy , we display this dependence in the range from threshold to 210 gev in fig . [ wwcomposroots ] . as mentioned near the end of sec . [ polzhprod ] , the effect of including initial state radiation in the results is to produce events at lower @xmath41 . from fig . [ wwcomposroots ] , we see that the addition of such events would slightly increase the fraction of the dominant spin component in the beamline and 00b bases and slightly decrease the dominant component in the helicity and bases . the 00b basis is slightly less sensitive to small shifts in @xmath41 than the beamline basis , while the basis is slightly less sensitive to small shifts in @xmath41 than the helicity basis . the presence of maximal parity violation in the coupling between the @xmath1 boson and fermions results in distinctly different angular distributions for the @xmath1 decay products for all three @xmath1 polarizations . in fact , the decay distributions in the rest frame of the decaying @xmath153 ( or @xmath156 ) are the same as for @xmath4 decay [ _ i.e. _ eqs . ( [ zt ] ) and ( [ zl ] ) ] with @xmath178 for all possible final states , leptonic and hadronic . thus , they follow the @xmath179 curves presented in fig . [ zdecaydist ] . note , however , in the case of @xmath180 , it is more convenient to use the distribution in the _ antilepton_angle @xmath181 : this distribution may be generated by replacing @xmath128 with @xmath182 in eqs . ( [ zt ] ) and ( [ zl ] ) . because the distributions for transversely - polarized @xmath1 s are narrower than the one for ly - polarized @xmath1 s , it is easier to identify the @xmath63 and @xmath64 polarization states than the @xmath106 state . to demonstrate this , suppose we `` tag '' the spin of the parent @xmath153 based on the value of @xmath128 as follows : the spin is taken to be @xmath23 if @xmath183 , @xmath24 if @xmath184 , and @xmath106 if @xmath185 . the probabilities that these assignments are correct may be computed from eqs . ( [ zt ] ) and ( [ zl ] ) , yielding ( 7 + 4y+y^2 ) , & & ( 3-y^2 ) , & & . [ wtagging ] no matter what value is chosen for @xmath186 , the probability of correct identification in the central region is always less than 50% . on the other hand , the correct identification rate for the @xmath63 and @xmath64 states in their respective regions can be of order 75% . we now consider correlations between the decay products of the @xmath157 pair . for the sake of concreteness , let us consider the case where the @xmath153 decays to @xmath187 and the @xmath156 decays to @xmath188 . let @xmath101 be the angle between the @xmath189 emission direction and the @xmath153 spin axis in the @xmath153 rest frame , and @xmath181 be the angle between the @xmath190 emission direction and the @xmath156 spin axis in the @xmath156 rest frame . for simplicity we are assuming one can determine the direction of the two neutrinos . if one of the @xmath1-bosons decays into two jets , then the correlations of the down - quark jet are the same as that of the charged lepton . since determining which jet is the down - type jet is problematic , the distributions given here will have to be folded such that they are symmetric when @xmath191 . the first set of correlations we wish to discuss may be conveniently displayed as a scatter plot in @xmath128 and @xmath192 . in fig . [ wwdecaycontours ] we have plotted the predictions for this distribution using the helicity , , beamline , and 00b bases . on one hand , the 00b plot is very nearly the pure distribution for the decay of the @xmath82 $ ] spin state . however , the broadness of the decay distributions for the longitudinal @xmath1 s cause the events to be rather spread out in this plot . on the other hand , there is a nice contrast between the maximum and minimum values in the plot . the functional dependence is less simple , however , as fully three of the six spin configurations contribute at or above the @xmath193 level . although a precision measurement of the complete 2-dimensional distribution requires a large number of events , it is still possible to perform some interesting tests with fewer events . for example , if @xmath14 is conserved , then the scatter plot will be symmetric about the line @xmath194 . so a measure of the asymmetry in the number of events on either side of this line provides a simple test of @xmath14 symmetry . in fig . [ wwinf ] , we have plotted the distribution of the interference terms @xmath135 ( as described in sec . [ zzcorrel ] ) for @xmath157 production and decay using the helicity , , beamline , and 00b bases . none of the bases under consideration ( nor any basis which we have found ) performs particularly well in this respect . the basis contains the smallest interference terms on average , but there is still a non - negligible component away from @xmath136 . this bump / peak at positive @xmath130 for all spin bases is caused by the strong azimuthal correlations associated with the pure @xmath195 coupling of the @xmath1-boson to its decay products . a similar structure appears in the @xmath130 plots for @xmath36 in the situation where both @xmath4 s decay to @xmath179 . although there is no obvious spin basis which is ideally suited to the study of @xmath1 pair production at lep , the basis is clearly better than the helicity basis in many ways . the same can be said of the 00b basis compared to the beamline basis . a detailed study of the correlations in more than one basis is required to disentangle the spin correlations in this process . if the higgs mass turns out to be in the vicinity of the @xmath4 mass , then the basis , in which the @xmath4 produced in association with he higgs is purely in the @xmath105 polarization states , provides a useful handle with which to distinguish between @xmath11 and @xmath12 events , based upon the angular distributions of the decay products . in this situation , the helicity basis supplies no useful information . even if the higgs mass is significantly different from the @xmath4 mass , the basis still exists and leads to distinctive angular correlations among the decay products . a comparison between the and helicity descriptions of the data is useful for distinguishing a scalar from pseudoscalar higgs . for the study of the @xmath13 events there is no ideal basis . however , the basis is better than helicity , and the 00b basis is better than the beamline basis at describing the correlations . not only is there a greater fraction of the total cross section concentrated in the dominant components in these preferred bases , but the interference distributions are somewhat narrower . the lack of an ideal basis implies that study of the @xmath13 system is best carried out using more than one basis , depending upon which quantity is being tested . furthermore , simply checking to see that the correlations change in the correct manner as the basis is varied tests the standard model in ways which can not be accomplished with the helicity basis alone . the fermi national accelerator laboratory ( fnal ) is operated by universities research association , inc . , under contract de - ac02 - 76cho3000 with the u.s . department of energy . high energy physics research at mcgill university is supported in part by the natural science and engineering research council of canada . we would like to thank peter fisher and philip bambade for illuminating discussions . gm would like to thank the fnal theory group for their kind hospitality during visits to initiate and wrap - up this work . sp would like to thank the aspen center for physics where a portion of this work was completed . following kuijf@xcite , we consider a massive vector boson of momentum @xmath196 and mass @xmath197 such that @xmath198 . let @xmath199 be the spin vector associated with this vector boson such that @xmath200 and @xmath201 . we need the three polarization vectors @xmath202 such that in the vector boson rest frame the boson has spin projection @xmath203 with rest to the spatial part of the spin vector @xmath199 . these three polarization vectors are conveniently written in terms of the two vectors v_1 = ( v+ms ) v_2 = ( v - ms ) . note @xmath204 , @xmath205 and @xmath206 . the decay products of a transverse vector boson are directly correlated with @xmath207 and @xmath208 , not @xmath196 and @xmath199 . using the spinor notation of mangano and parke@xcite , where |u(v_1 ) ( 1_5 ) ( 1_5 ) v(v_1 ) , we have _ ^ & = & ^ + _ 0^ & = & ^ - ^ ( = v_1 ^ -v_2^ ) . the phase factors have been chosen such that @xmath209 is real , @xmath210 , and @xmath211 if we interchange @xmath207 and @xmath208 . our convention is that these polarization vectors are for outgoing vector bosons whereas for incoming vector bosons we use @xmath212 . the helicity basis is obtained by choosing the spatial part of the spin vector @xmath199 to be in the same direction as the spatial part of the momentum vector @xmath196 of the vector boson . for example , if v = m ( 1 , ) using an obvious notation then choose v_1 = 12 m ( 1 + ) ( 1 , ) v_2 = 12 m ( 1- ) ( 1 , - ) . bjorken , proc . summer institute on particle physics , slac report 198 ( 1976 ) ; j. ellis , m.k . gaillard , and d.v . nanopoulos , nucl . phys . * b106 * , 292 ( 1976 ) ; b.w . lee , c. quigg , and h.b . thacker , phys . rev . * d16 * , 1519 ( 1977 ) ; b.l . ioffe and v.a . khoze , sov . * 9 * , 50 ( 1978 ) . gaemers and g.j . gounaris , z. phys . * c1 * , 259 ( 1979 ) ; g. gounaris , j. layssac , g. moultaka , and f.m . renard , int . j. mod a8 * , 3285 ( 1993 ) ; m. bilenky , j.l . kneur , f.m . renard , and d. schildknecht , nucl . phys . * b409 * , 22 ( 1993 ) ; k. hagiwara , t. hatsukano , s. ishihara and r. szalapski , nucl . phys . * b496 * , 66 ( 1997 ) . for a recent review of trilinear gauge boson couplings see : h. aihara , _ et . al . _ , in : _ electroweak physics and beyond the standard model _ , world scientific publishing co. , singapore , eds . t. barklow , s. dawson , h. haber , and j. siegrist , hep - ph/9503425 . ccddddc & spin & helicity & beamline & & & + & configuration & basis & basis & basis & basis & + & @xmath213 & 35.6% & 3.5% & 0.0% & 51.8% & + & @xmath214 & 32.2% & 55.4% & 57.4% & 24.1% & + & @xmath215 & 32.2% & 41.2% & 42.6% & 24.1% & + ccddddc & spin & helicity & beamline & & & + & configuration & basis & basis & basis & basis & + & @xmath86 & 11.7% & 5.6% & 0.0% & 0.1% & + & @xmath87 & 22.6% & 4.2% & 0.3% & 25.7% & + & @xmath88 & 22.6% & 2.4% & 0.2% & 25.7% & + & @xmath216 & 5.8% & 2.8% & 0.0% & 0.0% & + & @xmath217 & 18.7% & 54.6% & 64.1% & 24.2% & + & @xmath218 & 18.7% & 30.4% & 35.4% & 24.2% & + & @xmath219 & 30.4% & 48.1% & 49.7% & 24.4% & + & @xmath220 & 34.8% & 32.9% & 32.4% & 37.8% & + & @xmath221 & 34.8% & 19.0% & 17.9% & 37.8% & + ccccdcc & & fermion & @xmath102 & value & & + & & @xmath222 & @xmath223 & 0.834 & & + & & @xmath224 & @xmath225 & 0.968 & & + & & @xmath226 & @xmath227 & 0.574 & & + & & @xmath164 & @xmath228 & 1.000 & & + & & @xmath229 & @xmath230 & 0.500 & & + cccddcc & & diagram(s ) & contribution & fraction & & + & & @xmath151 & 0.2 pb & 1.0% & & + & & @xmath150 & 13.8 pb & 70.6% & & + & & @xmath164 & 36.7 pb & 188.0% & & + & & inf[@xmath150,@xmath164 ] & @xmath6431.1 pb & @xmath64159.6% & & + & & total & 19.5 pb & & & + ccddddc & spin & helicity & @xmath87 & beamline & @xmath217 & + & configuration & basis & maximized & basis & maximized & + & @xmath86 & 8.9% & 10.4% & 11.6% & 2.2% & + & @xmath87 & 9.9% & 64.6% & 2.1% & 3.6% & + & @xmath88 & 46.2% & 2.9% & 0.6% & 0.6% & + & @xmath216 & 4.0% & 7.5% & 3.6% & 0.6% & + & @xmath217 & 9.9% & 0.1% & 79.7% & 92.5% & + & @xmath218 & 21.1% & 14.6% & 2.4% & 0.6% & + ccddddc & spin & helicity & @xmath87 & beamline & @xmath217 & + & configuration & basis & maximized & basis & maximized & + & @xmath86 & 26.8% & 11.0% & 1.9% & 2.1% & + & @xmath87 & 0.0% & 16.5% & 30.6% & 42.1% & + & @xmath88 & 0.0% & 1.2% & 0.6% & 0.6% & + & @xmath216 & 3.6% & 1.9% & 3.6% & 7.3% & + & @xmath217 & 20.6% & 36.6% & 32.5% & 18.6% & + & @xmath218 & 20.6% & 4.4% & 2.4% & 0.8% & + ccddddc & spin & helicity & @xmath87 & beamline & @xmath217 & + & configuration & basis & maximized & basis & maximized & + & @xmath86 & 43.2% & 6.3% & 10.8% & 0.1% & + & @xmath87 & 9.9% & 110.2% & 24.5% & 63.9% & + & @xmath88 & 46.2% & 5.6% & 0.0% & 0.0% & + & @xmath216 & 7.5% & 7.6% & 0.0% & 6.0% & + & @xmath217 & 32.0% & 36.5% & 152.7% & 118.0% & + & @xmath218 & 49.2% & 21.8% & 0.0% & 0.8% & + ccddddc & spin & helicity & @xmath87 & beamline & @xmath217 & + & configuration & basis & maximized & basis & maximized & + & @xmath86 & @xmath6461.1% & @xmath646.9% & @xmath641.1% & @xmath640.0% & + & @xmath87 & 0.0% & @xmath6462.2% & @xmath6453.1% & @xmath64102.5% & + & @xmath88 & 0.0% & @xmath643.9% & 0.0% & 0.0% & + & @xmath216 & @xmath647.1% & @xmath642.0% & 0.0% & @xmath6412.7% & + & @xmath217 & @xmath6442.8% & @xmath6473.0% & @xmath64105.4% & @xmath6444.1% & + & @xmath218 & @xmath6448.7% & @xmath6411.6% & 0.0% & @xmath640.3% & +

a new numerical framework , based on the use of a simple first order strongly hyperbolic evolution equations , is introduced and tested in case of-dimensional spherically symmetric gravitating systems . the analytic setup is chosen such that our numerical method is capable to follow the time evolution even after the appearance of trapped surfaces , more importantly , until the true physical singularities are reached . using this framework , the gravitational collapse of various gravity - matter systems are investigated , with distinguished attention to the evolution in trapped regions . it is justified that in advance to the formation of these curvature singularities , trapped regions develop in all cases , thereby supporting the validity of the weak cosmic censor hypothesis of penrose . various upper bounds on the rate of blow - up of the ricci and kretschmann scalars and the misner - sharp mass are provided . in spite of the unboundedness of the ricci scalar , the einstein - hilbert action was found to remain finite in all the investigated cases . in addition , important conceptual issues related to the phenomenon of topology changes are also discussed .

0911.2373

the diffeomorphism invariance of einstein s theory of gravity is in an intimate relation with the fact that there is a significant redundancy in the representation of the true physical degrees of freedom . it can therefore be a great challenge to carry out a faithful investigation of dynamical processes even in case of spherically symmetric spacetimes , in spite of the simplifications offered by the symmetries . correspondingly , the selection of the most appropriate variables , done by applying a suitable gauge fixing , i.e. , the most effective framework to carry out the study of a given dynamical system , is considered to be a sort of art . in the numerical investigations of spherically symmetric dynamical systems , the method of choptuik that had been applied first by him in @xcite while exploring the critical phenomenon in the gravitational collapse of various gravity - matter systems turned to be the most successful in the sense that it is still widely used . however , choptuik s choice has both advantages and disadvantages . perhaps the most important advantage is that one has to solve only two first order partial differential equations ( pdes ) for the basic metric variables , and @xmath1 in terms of which the spacetime metric can be given as @xmath2 ] which , along with the matter field equations , determine the full evolution of the associated gravity matter system . on the other hand , the following objections may also be raised . first of all , the radial coordinate @xmath3 is chosen so that the area @xmath4 of the @xmath5-invariant @xmath6-spheres are given as @xmath7 . as it is well - known , coordinate systems of this type are not suitable to follow evolution in regions where `` trapped surfaces '' are formed because a coordinate singularity also develops ( for a short discussion see , e.g. , the last paragraph of @xcite ) . another objection is that one of the first order pdes is hyperbolic while the other is elliptic . this means that one of the metric equations which is a constraint equation has to be integrated on succeeding time level surfaces repeatedly . this process slows down time integration and makes it hard to apply the powerful tool of adaptive mesh refinement ( amr ) which ensures high numerical accuracy in strongly dynamical processes @xcite . there are two side - remarks in order . first of all , although it is possible to implement a variant of amr for the numerical integration of mixed hyperbolic and elliptic equations @xcite , the precision is partly lost because the applicability of amr necessitates the extrapolation of some variables in time , something which is better to be avoided in a time integration process . thereby , the use of a fully hyperbolic system is preferable in case of numerical integration of the field equations based on a finite difference schema . secondly , it was shown by one of the present authors in @xcite that by making use of the kodama vector field as the time evolutions vector field , the mixed elliptic - hyperbolic system can be replaced by a fully hyperbolic one . however , we have found the inevitable formation of a coordinate singularity in numerical simulations , the appearance of which was anticipated in the discussion below equation ( 4.15 ) of @xcite . the above findings motivate the search for a better analytic set - up with a more appropriate choice of the basic variables . before proceeding and presenting our proposal for such a choice we would like to mention that several attempts have also been made to use reduced versions of the adm and bssn @xcite formalisms in spherical symmetry , see @xcite . these models have been yielded by the reduction of a more complicated evolutionary system , therefore they do not optimally fit the spherically symmetric setup . either they apply coordinates that are not suitable to cover regions with trapped surfaces , as it happens , e.g. , in @xcite , or they are simply too complicated see , e.g. , the basic set of field equations ( 9a)(9f ) and ( 10a)(10c ) in @xcite . they may also suffer from numerical instabilities at the origin because negative powers of @xmath3 appear in the evolutionary equations . thereby , it is of considerable interest to single out a simple and general enough framework within which time evolution can be investigated on the largest possible part of the physical spacetime up to the appearance of true geometrical singularities . to match this requirement , we shall start by choosing a fully hyperbolic evolutionary system that is automatically applicable to describe the evolution in trapped region(s ) . this choice should be such that the equations are free from the numerical instabilities that used to appear in the origin in spherically symmetric spacetimes . an analytic framework fitting the above outlined expectations may be chosen as follows . based on the results of earlier investigations in spherically symmetric ( see , e.g. , @xcite ) and also in generic ( see , e.g. , @xcite ) dynamical spacetimes , the metric of the four dimensional spherically symmetric spacetime @xmath8 will be assumed to possess the form @xmath9 where the coordinates @xmath10 and @xmath11 label the points of the two - dimensional timelike surfaces transverse to the transitivity surfaces of the rotation group , and @xmath12 , @xmath13 , @xmath3 are smooth functions of ( @xmath10 , @xmath11 ) . note that @xmath11 generally differs from the area - radial coordinate . this condition , as we have already mentioned , is necessary to extend the domain of time evolution to include trapped surfaces when they exist . by making use of this geometrical framework , gravitational collapse have been investigated in some simple gravity - matter systems . in the simplest possible case of asymptotically flat configurations , the associated time evolution of the system is qualitatively expected @xcite to be as indicated on fig.[coll ] where the event horizon$ ] . ] , and the apparent horizon are also shown . the latter is foliated by marginally trapped surfaces and it is represented by a curve connecting the two ends of the zigzag line depicting the singularity . a portion of matter either falls into the singularity or reaches future timelike or null infinity , @xmath14 or @xmath15 . the asymptotic structure of the spacetime is expected to approach that of the schwarzschild solution as we get closer to @xmath14 along the null generators of @xmath15 @xcite . if the collapsing matter has no considerable radiative degrees of freedom , the mass of the developing black hole is expected to be close to the total mass on the initial data surface . note that spherical symmetry makes it possible to define mass and energy inside an invariant metric sphere ( see section[trap ] ) . if the mass inside the marginally outer trapped surface is found to approach the total mass while moving outwards along the apparent horizon , then we have a strong indication that our spacetime grid covers the truly dynamical part of the collapse . moreover , as it will be demonstrated in section[dyn ] , the lapse function @xmath13 may always be chosen such that the cauchy surfaces can get arbitrarily close to the singularity . we were also interested in investigating the time evolution of more exotic initial data specifications . likewise in the standard friedman - robertson - walker cosmological models these are known to be spherically symmetric around any of their spacetime events there is a freedom in choosing the topology of the initial data surfaces . the base manifold @xmath16 of the investigated spacetimes coincides with the future cauchy development of some three - dimensional achronal hypersurface @xmath17 . thereby , @xmath18 $ ] and it possesses the product space structure @xmath19 @xcite . since the spacetime is spherically symmetric , the @xmath20 time level surfaces these are diffeomorphic to @xmath17can also be foliated by the transitivity surfaces of the rotation group . in virtue of the particular form of the applied line element ( [ the_metric.eq ] ) the metric induced on the transitivity surfaces of the rotation group is given as @xmath21 . thereby , the vanishing of @xmath3 , which is , in fact , the area - radius function , is directly related to the existence of an origin . to see how many origins we might have , let us consider some simple choices for the topology of the initial data surface @xmath17 . whenever @xmath17 is a connected geodesically complete spacelike hypersurface possessing a trivial product bundle structure , its topology is either @xmath22 , @xmath23\times\mathbb{s}^2 $ ] , @xmath24 or @xmath25 ( see as an illustration the bottom line of fig.[top ] ) . accordingly , there may be one origin , two origins or no origin at all on our initial hypersurface . = 4.3 cm = 4.3 cm = 4.3 cm = 4.3 cm during time evolution , the geometrical properties of the time level surfaces may change such that new origins are produced , possibly indicating a change in the topology , as illustrated on fig.[top ] . at this point it is important to clear up the some of the related notions and potential misconceptions . while in the above sentences the terminology of topology change have been used , it should be kept in mind that it is the topology of the limit of time level surfaces comprising the future boundary of the pertinent `` domain of dependence''what may differ from that of the time level surfaces . a careful approach is needed in applying notions such as _ domain of dependence , cauchy development _ and _ cauchy surface _ which were introduced in @xcite by geroch . we would like to emphasise first that there is a significant distinction between the notions `` initial data surface '' and `` cauchy surface '' . we may chose an arbitrary achronal hypersurface @xmath17 as an initial data surface . the initial data together with the field equations can be used to determine the solution everywhere in the domain of dependence @xmath26 $ ] . as opposed to this , the use of the notion of `` cauchy surface '' tacitly requires additional knowledge about the global properties of the underlying spacetime . in particular , see theorem 11 of @xcite , a spacetime is known to be globally hyperbolic if and only if it possesses a cauchy surface . the confusion might arise and , in fact , apparently it does arise in many circumstances in consequence of the fact that the domain of dependence @xmath26 $ ] of any initial data surface @xmath17 , is a globally hyperbolic spacetime itself on its own right , i.e. , @xmath17 is a cauchy surface for @xmath26 $ ] . in many cases it is obvious to recognise that there exists a globally hyperbolic global extension @xmath27 of @xmath28,g_{ab})$ ] such that the boundary of @xmath26 $ ] is not empty in @xmath29 , while the cauchy horizon , @xmath30=h^+[\sigma ] \cup h^-[\sigma]$ ] of @xmath26 $ ] is obviously empty with respect to @xmath26 $ ] . however , a generic method that could be applied in all the possible cases does not exist . although the concept of `` maximal cauchy development '' was introduced in @xcite , it is based on zorn s lemma , which makes its use difficult in practice . as a simple example , consider the maximal analytic extension of the schwarzschild spacetime . in kruskal coordinates @xmath31 the line element is given by @xmath32 where the schwarzschild coordinates @xmath3 and @xmath33 are determined by the following implicit relations ( see , e.g. , @xcite ) : @xmath34 by introducing suitable new coordinates @xmath35 and @xmath36 , the line element of the schwarzschild spacetime preserves the form ( [ the_metric.eq ] ) . the time slicings two of which are indicated on fig.[sch]and the corresponding lapse functions vary accordingly . in the first case ( left panel ) , two points on the @xmath37 hypersurface hit the @xmath38 schwarzschild singularity . the appearance of these points can be interpreted as the formation of two origins , since @xmath39 in all other points of @xmath37 . correspondingly , the topology @xmath24 of @xmath40 is replaced by the disjoint union of @xmath41 , @xmath42 and @xmath43 at @xmath44 ( see the third column of fig . [ top ] ) . it is not hard to modify @xmath40 by inserting suitably located further `` steps '' such that the number of @xmath42 components of the pertinent @xmath45 may take an arbitrary integer value . although the time level surfaces on the left panel do not cover the future domain of dependence , @xmath46 $ ] , of @xmath40 , it is worthwhile to keep in mind that @xmath40 is a cauchy surface for the schwarzschild spacetime . as it is indicated on the right panel of fig.[sch ] , the entire of @xmath46 $ ] can be covered by time level surfaces if the new coordinates @xmath10 and @xmath11 , and in turn , the lapse @xmath13 are chosen properly . there is no topology change and the section of the cauchy surfaces within the black hole region uniformly converge to the @xmath38 singularity . in section[dyn ] , the application of this type of dynamically determined @xmath13 will be demonstrated via various examples . we would like to mention that the type of topology changes indicated by the left panel of fig.[sch ] showed up in many of our numerical simulations . we have found that , in consequence of the developing inhomogeneities , certain parts of the time level surfaces may get closer to the singularity much faster than others . nevertheless , as shown by the above example , the change of the topology refers to the limit of the applied time level surfaces rather than the underlying spacetimes . as it will be demonstrated in the succeeding sections , the @xmath13 function can be chosen such that the time evolution is slowed down in spacetime regions close to the singularity . this way we were able to enlarge the domain of dependences significantly and to get arbitrarily close to the developing spacetime singularities in our simulations . as a consequence , we were able to investigate the rate of blowing up of the curvature while approaching the singularity . the structure of this paper is the following . in section[intr ] , with the help of a simple - minded approach , we demonstrate the main technical issues in solving the einstein - scalar field equations numerically . these difficulties are resolved in section[intr2 ] , where the misner - sharp mass is introduced as an auxiliary variable to stabilise the applied numerical representation near the origin . here we also describe the `` trick '' which makes our framework capable to follow the time evolution within the trapped region . section[stabor ] outlines the analytic procedure that guarantees the required stability of the time evolution in the neighbourhood of an origin . in section[energybal ] , a method is presented which provides the opportunity to monitor the energy transfer processes . trapped surfaces , trapped regions and apparent horizons are defined in section[trap ] . dynamical investigations of gravitational collapse are presented in simple and in certain less obvious circumstances are presented in section[dyn ] . the results of the convergence and accuracy tests are also presented in this section . the paper is closed by summarising our results and by providing a short discussions concerning the interpretation and some of the immediate consequences of our findings . to have a truly dynamical spacetime , some matter fields also have to be specified in addition to the already introduced geometrical framework . for simplicity , we shall restrict our considerations to a real self - interacting scalar field with lagrangian @xmath47 where the potential possesses the form relevant to the klein - gordon field , @xmath48 with mass parameter @xmath49 . note , however , that @xmath50 could be chosen to be any sufficiently regular function of the @xmath51 field . the field equations relevant for the gravity - matter system can be given as @xmath52 where the energy - momentum tensor reads as @xmath53\,.\end{aligned}\ ] ] it is straightforward to verify , by a direct calculation , that our basic variables @xmath12 , @xmath13 , @xmath3 and @xmath51 satisfy the evolution equations @xmath54 @xmath55 @xmath56 where the abbreviating notation @xmath57 is applied ( for @xmath58 = @xmath59 , @xmath13 , @xmath3 and @xmath51 ) . similarly , the constraint equations are @xmath60 where @xmath61 the number of independent equations is in accordance with the fact that the evolution equation ( [ psi_tautau.eq ] ) for @xmath51 is simply the actual form of ( [ psi.eq ] ) , and that in a generic four - dimensional spherically symmetric spacetime there may only be four independent equations derived from the einstein s equation ( [ einst ] ) . in particular , equations ( [ omega_tautau.0.eq ] ) and ( [ r_tautau.0.eq ] ) can be seen to be equivalent to the combinations @xmath62 and @xmath63 , whereas the constraints are @xmath64 and @xmath65 . notice that we have no evolution equation for @xmath13 . according to the terminology introduced by friedrich in @xcite , this function can be seen to play a role analogous to the `` gauge source function '' . it is freely specifiable , provided that it is sufficiently regular on the domain of time evolution . by making use of this freedom , we were able to slow down evolution in our numerical simulations at the parts of the time level surfaces whenever they have got close to the singularity . defining the coordinate speed of light , @xmath66}={\mathrm{d}\rho}/{\mathrm{d}\tau}$ ] along the radial null geodesics in the @xmath67 part requiring the vanishing of the pertinent part of the line element , it is straightforward to see that @xmath66}=\beta$ ] . thereby , as far as @xmath13 is guaranteed to be bounded so is @xmath66}$ ] . in our simulations , @xmath13 was guaranteed to be less than or equal to 1.2 everywhere . it is also remarkable that the system of evolution equations may be viewed as a system of coupled non - linear wave equations for the variables @xmath12 , @xmath3 and @xmath51 . these type of equations are known to possess a well - posed initial value problem ( see , e.g. , @xcite ) . in addition , it can be checked by a direct calculation that whenever the evolution equations are satisfied and , in turn , their derivatives also vanish the following set of first order strongly hyperbolic evolution equations can be derived @xmath68+\frac{({\alpha}_\tau\beta+{\alpha}\beta_\tau ) e^\tau_{\ \rho}-(2{\alpha}_\rho\beta+{\alpha}\beta_\rho ) e^\tau_{\ \tau}}{{\alpha}\beta } \ = \ 0,\label{cev1}\\ & & \hskip-.6cm\partial_\tau e^\tau_{\ \tau}-\beta^2\partial_\rho e^\tau_{\ \rho}+\frac2r\left[r_\tau e^\tau_{\ \tau}-\beta^2r_\rho e^\tau_{\ \rho}\right]+\frac{{\alpha}_\tau e^\tau_{\ \tau}- \beta({\alpha}_\rho\beta+3{\alpha}\beta_\rho ) e^\tau_{\ \rho}}{{\alpha}}\ = \ 0\ , . \label{cev2}\end{aligned}\ ] ] since these equations are linear and homogeneous in the variables @xmath69 and @xmath70 , they ensure that the constraints propagate with the evolution , therefore it is enough to impose them only on the initial data hypersurface . in spite of the attractive features of the above outlined evolutionary system see equations ( [ omega_tautau.0.eq]-[psi_tautau.eq ] ) and ( [ constraints.eq ] ) along with ( [ cev1])([cev2]) , it has been found to be unstable in numerical simulation because truncation errors grow rapidly near the origin(s ) , i.e. , where @xmath3 tends to zero . the instability is caused by terms of the form @xmath71 appearing in ( [ r_tautau.0.eq ] ) with @xmath72 and in ( [ omega_tautau.0.eq ] ) with @xmath73 . the nominator of ( [ crashing_term.eq ] ) consists of three independent variables , each of which contributes to its numerical error . although it should vanish rapidly while approaching the origin , it does not exactly do that in practice and , in turn , its slightest error is amplified enormously via the division by @xmath74 . increasing the resolution does not help in such a circumstance , the simulations crash even earlier . we tried to cure this instability in various ways such as extrapolation from outer points , decreasing the order of finite differences near the origin , increasing the numerical dissipation term , and also by using different parametrisations of the metric which were supposed to behave better close to the critical points . neither of these attempts turned out to be successful . by the inspection of the term ( [ crashing_term.eq ] ) responsible for the above mentioned instabilities and of the misner - sharp mass @xmath75 @xcite defined by the relation @xmath76 it seems to be rewarding to introduce @xmath75 as an auxiliary variable . by deriving it with respect to @xmath10 and @xmath11 and by using the pertinent form of einstein s equations ( [ einst ] ) to eliminate the second derivatives of @xmath3 , an evolution and a constraint equation can be deduced : @xmath77 what makes the introduction of @xmath75 even more preferable is related to the following observations . first of all , einstein s equations @xmath78 can be seen to hold whenever @xmath79 it is important to emphasise that instead of solving the constraint equations @xmath65 and @xmath64 on the initial data surfaces , it suffices to solve ( [ m_rho.2.eq ] ) , provided that the evolution equations ( [ omega_tautau.2.eq])-([m_tau.2.eq ] ) hold . to see this , recall first that ( [ r_tautau.2.eq ] ) is equivalent to @xmath80 in virtue of ( [ m.prop1.eq ] ) , ( [ m.prop2.eq ] ) , ( [ m_tau.2.eq ] ) and ( [ m_rho.2.eq ] ) , we have that @xmath81 equations ( [ seg1.eq])-([seg2.eq ] ) along with the relation @xmath82 , imply then that for the variables @xmath69 , @xmath70 , the homogeneous linear equations @xmath83 hold . the algebraic sub - determinant of this system is @xmath84 , which will be shown in section [ trap ] to vanish only at isolated marginally trapped surfaces . therefore the system ( [ seg3.eq ] ) possesses only the trivial solution which justifies our claim . restricting our considerations to the case of a self - interacting scalar field again , the components of the energy - momentum tensor can be given as @xmath85 in this case , the complete set of evolution equations consists of ( [ omega_tautau.2.eq])-([m_tau.2.eq ] ) and ( [ psi_tautau.eq ] ) . note that the introduction of @xmath75 makes the use of ( [ omega_tautau.2.eq ] ) superfluous since @xmath12 can be determined from ( [ m.def.eq ] ) once the fields @xmath75 and @xmath3 are known , however , this way we get a 0/0 type formula for @xmath59 at ` marginally trapped surfaces ' ( for their precise definition see section[trap ] ) and an instability in the numerical simulation , thereby making it impossible to study the evolution inside the black hole region . in order to overcome these difficulties , we kept ( [ omega_tautau.2.eq ] ) as an evolution equation for @xmath12 . then , as it will be demonstrated in section[dyn ] ( see fig.[coll1 ] ) we could guarantee that our numerical framework does possess the desired capabilities without the loss of numerical accuracy anywhere in the domain of dependence . by making use of standard procedures ( described in details , e.g. , in @xcite ) , a first order evolutionary system for our spherically symmetric dynamical system can be deduced as follows . in addition to @xmath75 , @xmath12 , @xmath3 , and @xmath51 , the first derivatives @xmath86 , @xmath87 , @xmath88 , @xmath89 , @xmath90 and @xmath91 are also considered as being independent variables , and the evolution equations are complemented by the relations @xmath92 note that the last three relations are , in fact , the integrability conditions for @xmath12 , @xmath3 , and @xmath51 . as a byproduct of this reduction process , in addition to the true dynamical constraint ( [ m_rho.2.eq ] ) , we also have to take care of the trivial ones @xmath93 the yielded first order system can then be put into the form @xmath94 where the associated @xmath95-dimensional vector variable @xmath96 , the @xmath97 matrix @xmath98 and the @xmath95-dimensional source vector @xmath1 are given as @xmath99 @xmath100 since the eigenvectors of the matrix @xmath98 comprise a complete system and its eigenvalues are all real , this first order system is strongly hyperbolic @xcite . , @xmath101 and @xmath102 , the above set of field equations could be put into the form of a first order symmetric hyperbolic system for the vector valued variable @xmath103 . ] it is known that both the analytic and numerical well - posedness of the initial value problem is guaranteed for this type of evolution equations @xcite . an additional preferable property of the investigated dynamical system is that for the coordinate speed of light , @xmath66}\leq \beta_\mathrm{max}$ ] holds , where @xmath104 denotes the maximum value of @xmath13 . the behaviour of the basic field variables @xmath105 close to an origin located , say , at @xmath106 can be explored by substituting the form of taylor expansion @xmath107 into the field equations for the functions @xmath105 , and also by requiring the solutions to be at least of class @xmath108 for some @xmath109 . assuming that the solutions are at least of class @xmath110 in a neighbourhood of @xmath111 , the resulting relations are @xmath112 by making use of the indicated parity properties of the functions @xmath113 and @xmath51 , the grid boundary at the origin can be treated exactly the same way as it was done in @xcite . in particular , the method described in details in section d of @xcite makes it possible to determine all the spatial derivatives in a neighbourhood of an origin , by making use of a symmetric stencil . note that the factor @xmath114 in the first order term in ( [ reg2_r ] ) has a simple geometrical explanation . in order to have a regular origin , i.e. , to avoid the appearance of conical singularities , we have to guarantee that while approaching to the origin , @xmath115 , the ratio of the circumference of an infinitesimal , origin - centred circle and of the corresponding proper radius tends to @xmath116 , i.e. , @xmath117 where lhospital s rule is also applied . having the relations ( [ reg2_m ] ) and ( [ reg2_r ] ) , it is also straightforward to see why the system introduced in section[intr2 ] has been found stable in our numerical simulations . one of the reasons for this unexpected stability is related to the fact that the term ( [ crashing_term.eq])which was responsible for the numerical instabilities in case of the simple - minded system reads as @xmath118 since @xmath119 takes the values @xmath120 and @xmath6 , the term ( [ crashing_term.eq ] ) is replaced by a completely regular one in our new setup . it is also important to note that the misner - sharp mass @xmath75 is subject to a first order pde ( [ m_tau.2.eq ] ) which does not involve a singular term at all . if there is an origin , it is useful to have the limiting form of the source vector @xmath1 appearing in ( [ u_tau.0.eq ] ) . by making use of ( [ reg2_m])-([reg2_psi ] ) , it is straightforward to verify that the non - trivial components @xmath121 , @xmath122 and @xmath123 of @xmath1 , as given in ( [ b ] ) , tend to the following regular limits @xmath124 as it is well - known , energy transfer processes can not be investigated in general relativity , since appropriate quasi - local quantities have not been found as yet . however , this can be done in spherically symmetric dynamical spacetimes , using the kodama vector field @xcite @xmath125 where @xmath126 and @xmath127 denotes the volume form associated with the metric @xmath128 induced on the two - dimensional surface transverse to the @xmath5 group orbits . it was shown in @xcite that @xmath129 is divergence free , i.e. , @xmath130 , and also that @xmath131 vanishes , where @xmath132 denotes the einstein tensor @xmath133 . it follows then that the vector field @xmath134 is also divergence free . accordingly , even though our spherically symmetric spacetime is fully dynamical , @xmath135 behaves as a locally conserved energy flux type vector field . then , by making use of stokes theorem and by choosing @xmath136 to be a four dimensional spacetime region with boundary @xmath137 , we have that @xmath138 where @xmath139 denotes the 4-volume element while @xmath140 is the 3-volume element induced on the boundary @xmath137 of @xmath136 , which can be given as @xmath141 , where @xmath142 is the ( outward pointing ) unit normal 1-form field on @xmath137 . with our particular choice of coordinates , the components of the kodama vector field can be given as @xmath143 which , in virtue of ( [ kv00 ] ) , implies that the non - zero components of the divergence free vector field @xmath135 read as @xmath144\ \ \ { \rm and } \ \ \ j^\rho=\frac{1}{{\alpha}\beta}\left[t^\rho_{\ \tau}r_\rho - t^\rho_{\ \rho}r_\tau\right]\,.\ ] ] to characterise the energy transfer processes in a spacetime domain @xmath136 bounded by sections of @xmath20 and @xmath145 hypersurfaces , as it is depicted on fig.[enbal ] , we introduce the following auxiliary notations . = 6.5 cm denote the origin - centred ball of radius @xmath146 on the @xmath147 hypersurface by @xmath148 , and the part of the cylindrical hypersurface @xmath149 between @xmath150 and @xmath151 by @xmath152 . then the boundaries @xmath153 ( @xmath154 ) , as shown on fig.[enbal ] , are represented by the differences @xmath155 of the balls @xmath156 and @xmath157 , while @xmath158 are the cylinders @xmath159 connecting the boundaries of the balls @xmath160 and @xmath161 . then the energy contained in @xmath162 can be given as @xmath163 where @xmath164 is the volume element on the @xmath165 hypersurface and @xmath166 is its unit normal @xmath120-form field , i.e. , @xmath167 similarly , the energy transported through the cylindrical hypersurface @xmath152 is @xmath168 where @xmath169 is the volume element on the @xmath170 hypersurface and @xmath171 is its unit normal @xmath120-form field , i.e. , @xmath172 using these notations , the energy balance equation ( [ stokes ] ) takes the form @xmath173 the following relations can be verified by straightforward calculation : @xmath174 thus @xmath175 and @xmath176 can be given as @xmath177 it is important to mention here that , in virtue of the relations ( [ m_rho.2.eq ] ) , ( [ jt ] ) and ( [ vol_tau.eq])-([vol_rho.eq ] ) , the misner - sharp mass function @xmath75 can be given on a @xmath147 time level surface as @xmath178 accordingly , in virtue of ( [ vol_tau.eq])-([vol_rho.eq ] ) and ( [ mtau ] ) , the expression @xmath179 could also be interpreted as the combined energy density of our composed gravity - matter system . to define trapped surfaces , start with a smooth orientable @xmath6-dimensional compact manifold @xmath180 with no boundary in a @xmath0-dimensional spacetime @xmath8 . let @xmath181 be a smooth future directed non - vanishing null vector field on @xmath180 which is normal to @xmath180 , i.e. , @xmath182 for any vector field @xmath183 tangent to @xmath180 . consider then the null hypersurface @xmath184 generated by geodesics starting on @xmath180 with tangent @xmath181 . by parallel propagating @xmath181 along these generators , it can be extended onto @xmath184 . denote by @xmath96 the synchronised affine parameterisations of these null geodesics . the hypersurface @xmath184 is smooth in a neighbourhood of @xmath180 , and it is smoothly foliated by the @xmath96-level surfaces . denote by @xmath185 the volume element associated to the metric @xmath128 induced on these @xmath6-dimensional surfaces . then the null expansion @xmath186 with respect to @xmath181 is defined as @xmath187 where @xmath188 denotes the lie derivative with respect to the null vector field @xmath181 . it is straightforward to see that the sign of @xmath186 remains intact under a positive rescaling of @xmath181 , i.e. , whenever @xmath181 is replaced by @xmath189 , where @xmath58 is a sufficiently smooth positive function on @xmath180 . therefore , as far as the sign of @xmath186 is concerned , @xmath181 and @xmath190 may be considered to be equivalent . it is well - known that there always exist two equivalence classes of future directed non - vanishing null vector fields on @xmath180 which are normal to @xmath180 . let @xmath191 and @xmath192 be two such vector fields and denote the corresponding null expansions by @xmath193 and @xmath194. and @xmath195 signs have nothing to do with outward and inward pointing directions . although these directions can not be defined for generic spacetimes , it is worthwhile to mention that an adequate quasi - local notion of outward and inward direction can be introduced without referring to global properties of the underlying spacetimes provided that attention is restricted to untrapped or marginally trapped surfaces ( see , e.g. @xcite ) . ] according to the original definition of penrose @xcite , a @xmath6-dimensional surface @xmath180 in a @xmath0-dimensional spacetime is _ future trapped _ or _ untrapped _ if both of the future directed null geodesic congruences orthogonal to @xmath180 are converging , or one of them is diverging while the other is converging at @xmath180 , i.e. , @xmath196 or @xmath193 and @xmath194 have opposite signs . if one of the expansions vanishes identically while the other is everywhere non - positive , then @xmath180 is called future _ marginally trapped _ _ past trapped _ and past _ marginally trapped _ surfaces can be defined analogously by reversing the time orientation applied tacitly in the above definitions . choose now @xmath180 to be a @xmath6-sphere of radius @xmath3 , invariant under the @xmath5 symmetry of our spherically symmetric spacetime . in virtue of the @xmath197 part of the metric ( [ the_metric.eq ] ) , an obvious choice for the non - vanishing future directed null vector fields @xmath191 and @xmath192 on @xmath180 is @xmath198 while the @xmath6-volume element @xmath185 on the @xmath6-spheres @xmath199 foliating the @xmath200 null hypersurfaces is @xmath201 where @xmath202 denotes the levi - civita symbol . it is straightforward to check that the pertinent null expansions are @xmath203 equations ( [ m.def.eq ] ) and ( [ thpm ] ) imply that the following relation holds on @xmath180 : @xmath204 since all the functions @xmath12 , @xmath13 and @xmath205 are positive , apart from origin(s ) or the spacetime singularity , an @xmath5 invariant @xmath6-sphere @xmath180 of radius @xmath3 is untrapped or trapped ( future or past ) if and only if @xmath206 is positive or negative , respectively . we shall call a spacetime region @xmath207 future trapped if each point of it belongs to an @xmath5 invariant trapped 2-sphere . the past boundary , @xmath208 $ ] , of such a future trapped region will be referred as _ future apparent horizon _ @xmath209.-invariant 2-spheres @xcite . ] the notion of past trapped region , @xmath210 , and _ past apparent horizon _ , @xmath211 , may be introduced analogously . the first order strongly hyperbolic system described in section[intr2 ] and given in the form ( [ u_tau.0.eq ] ) was solved numerically with our finite difference code called gridripper amr @xcite ( see also @xcite ) . the time integration process uses the method of lines based on a fourth order runge - kutta scheme @xcite for a detailed description of this method , for the case of fixed spatial resolutions , see , e.g. , @xcite and as it is indicated by its name , adaptive mesh refinement techniques are also incorporated . before determining the time evolution of our system using equation ( [ u_tau.0.eq ] ) , we have to specify suitable initial data on a spacelike hypersurface @xmath212which hereafter will always be chosen as the @xmath213 hypersurface so that all the constraint equations are satisfied . recall that the components of the vector variable @xmath96 in ( [ u_tau.0.eq ] ) are @xmath214@xmath215 and @xmath90 and , as it follows from the discussion in section[intr2 ] , we apparently have only the constraint @xmath216\ ] ] beside the three trivial ones ( [ co])([cp ] ) . however , the price to be paid for using the `` superfluous '' evolution equation ( [ omega_tautau.2.eq ] ) is the appearance of an additional constraint ( [ m.def.eq ] ) , which implies that only two of the four variables , @xmath217 and @xmath218 may be chosen freely . to satisfy all constraints on the initial hypersurface , we have applied either of the following two selection procedures . * fixing of @xmath3 and @xmath88 from among the functions @xmath217 and @xmath88 on @xmath212 . start by specifying the sufficiently regular but otherwise arbitrary functions @xmath219 and @xmath90 , then choose @xmath89 and @xmath91 so as to satisfy the trivial constraints ( [ cr ] ) and ( [ cp ] ) . by substituting the relation @xmath220 into ( [ cm ] ) , the initial data for @xmath75 can be determined numerically . then the initial @xmath12 and @xmath87 are determined by ( [ om ] ) and the trivial constraint ( [ co ] ) , respectively . finally , @xmath86 is the @xmath10-derivative of ( [ om ] ) , @xmath221 * fixing of @xmath12 , @xmath3 , @xmath51 and @xmath90 on @xmath212 . choose @xmath222 and @xmath91 so as to satisfy the trivial constraints ( [ co])-([cp ] ) . from ( [ m.def.eq ] ) , we get @xmath223 where the sign depends on the specific physical problem we intend to investigate . by substituting the right hand side of ( [ rtau ] ) into ( [ cm ] ) , the yielded equation can be solved ( numerically ) for @xmath75 on the initial data surface @xmath212 . then the initial data for @xmath88 is determined by the relation ( [ rtau ] ) . finally , @xmath86 can be determined similarly as in case a. procedure a. is simpler but b. provides a convenient control of both the temporal geometrical setup and the energy distribution of the matter field . the latter is also more appropriate to investigations of expanding cosmological models . other choices may also be possible , however , the use of these two cases was found to be completely satisfactory . it is worth to be emphasised that either of these initial data specifications is suitable to host all the possible einstein - scalar field systems in the selected setup . to summarise , we have free control only of the four dynamical variables @xmath219 and @xmath90 or @xmath224 and @xmath90 in the two cases , respectively . in addition , the specification of @xmath13 is also required . once these functions are specified , they determine all other functions by either of the outlined procedures on @xmath212 . in our numerical simulations , we always used @xmath225 on @xmath212 . as it was emphasised earlier , the lapse function @xmath13 is freely specifiable . also , as it was indicated by the simple example in the introduction , the extent of the spacetime domain covered by the time level surfaces is influenced significantly by the chosen lapse function . to achieve the largest possible domain , we use a dynamically determined @xmath13 satisfying the evolution equation @xmath226\left(\frac{r_\tau}{r}\right)^3\beta \,,\label{beta_tau.eq}\ ] ] where @xmath227 is a positive real parameter , starting with the initial value @xmath228 . the value of the parameter was chosen to be @xmath229 in all of our simulations . equation ( [ beta_tau.eq ] ) was motivated as follows . to slow down the evolution close to the developing singularities , we need a function which tends to zero there rapidly enough . for this reason , first we applied @xmath230 . besides decaying fast enough , this function is also regular at the origin since , in virtue of ( [ reg2_m ] ) and ( [ reg2_r ] ) , @xmath231 remains finite while @xmath232 . the @xmath10-derivative of this @xmath13 is `` almost '' ( [ beta_tau.eq ] ) , but without the factor @xmath233 on the right hand side . the inclusion of this factor was found to be necessary in stabilising the time evolution close to the singularity . its use is supported by the observation that in a neighbourhood of a point where the function @xmath3 takes its minimum the factor @xmath234 gets to be sufficiently small providing thereby some additional slowing down to the evolution of @xmath13 there . let us start by investigating the gravitational collapse of a massive real scalar field in the `` conventional '' case , i.e. , with the assumption that the initial data is specified on a hypersurface @xmath212 possessing the topology of @xmath43 . according to procedure a , we specify two metric functions on @xmath212 as @xmath235 the scalar field @xmath51 is chosen to be a smooth hunch with compact support @xmath236 with vanishing @xmath10-derivative on @xmath212 @xmath237 the self - interaction potential is assumed to possess the form @xmath238 . the parameter values are @xmath239 in the numerical simulation , amr was used with @xmath240 spatial points on the base grid and five refinement levels ( refinement ratio 2 ) . to demonstrate the significance of the freedom in specifying @xmath13 , we shall compare the result yielded in two cases , @xmath241 and the dynamical lapse determined by ( [ beta_tau.eq ] ) . the initial state is the same in these two cases , as shown on fig.[ic ] = 6.cm where the initial matter and gravity - matter energy density distributions associated with a shell of radius @xmath11 , @xmath242 and @xmath243 are plotted . the @xmath11 integrals of these quantities are , respectively , the matter and gravity - matter energy contained in an origin - centred ball of radius @xmath11 on the @xmath244 time slices , @xmath245 and @xmath246 . the time evolution of this initial state , with @xmath247 , yields the gravitational collapse of the scalar field until a spacetime singularity is formed at @xmath248 . the change of the metric function @xmath3 during the pertinent time evolution is depicted on fig.[r1 ] , where the appearance of the second origin is clearly visible at @xmath249 . fig.[coll1 ] , = 5.cm shows @xmath12 , @xmath250 , @xmath251 and @xmath252 on an intermediate time level surface where trapped surfaces are already present . the value of @xmath12 is calculated two ways : by solving the evolution equation ( [ omega_tautau.2.eq ] ) and by using the algebraic relation ( [ om ] ) . at the first sight , the two calculations yield approximately the same result . however , there is a tiny difference close to the marginally trapped surfaces which , in virtue of ( [ thpm ] ) and ( [ thpm2 ] ) , are associated with the zeros of @xmath251 and @xmath252 in the present case . this difference is enlarged by @xmath11-derivation , thereby @xmath253 and @xmath254 differ significantly here . whenever we started to evolve the system using ( [ om ] ) , the error became enormous in a couple of time steps , making it impossible to carry on with the time evolution in the trapped region . to reach long - term stability even in regions with trapped surfaces , it turned out to be essential to apply ( [ omega_tautau.2.eq ] ) as an evolution equation . fig.[ec ] is to demonstrate that the combined gravity - matter energy is preserved with high precision during the entire history . more precisely , = 6.4 cm the energy balance relation ( [ stokes2 ] ) , defined by making use of the kodama vector field in the previous section , is found to hold during the entire evolution . the wiggly ends of the curves at @xmath255 are associated with the very appearance of the scalar curvature singularity at @xmath256 . in the evaluation of the energy balance relation ( [ stokes2 ] ) , the particular choices @xmath257 and @xmath258 have been made . moreover , @xmath259 and @xmath260 are defined as @xmath261 notice that doubling the number of grid points shifts the curve downward by about factor @xmath262 . therefore fig . [ ec ] provides a justification that our numerical method is fourth order accurate even with respect to @xmath259 , which is a complicated function of the elementary field variables . note that the experienced convergence order is in accordance with the applied fourth order finite difference schemes , both for the spatial derivatives and for the time integration . [ constraints.beta1.fig ] shows that the constraints ( [ m_rho.2.eq ] ) and ( [ om ] ) are preserved with high numerical accuracy even on the last time slice before reaching the singularity .- dependent values of the constraint violations , the @xmath11-dependence of their averaged values are shown , in order to smooth out the noise , on figs . [ constraints.beta1.fig]-[constraints.dynbeta.fig ] . the averaging for each @xmath11 is done by using the interval @xmath263 $ ] around @xmath11 . ] the misner - sharp mass @xmath75 , its derivative , @xmath264 , and the @xmath59 metric component are also plotted for comparison . note that @xmath75 appears to blow up at the point where the singularity is about to form , its maximum value reaches @xmath265 on the time slice shown . the error of the constraints ( thick solid curves ) also have peaks in this point . in case of dynamical lapse , the value of the ricci scalar is of the order @xmath266 within the interval @xmath267 at @xmath268 . constraint preservation on this time level surface is shown by fig.[constraints.dynbeta.fig ] . on fig.[ec2 ] the time evolution of the gravity - matter energy density distribution associated with a shell of radius @xmath11 , @xmath269 , is shown . this spacetime diagram depicts the variation of @xmath269 above the @xmath270 coordinate plane , where @xmath271 denotes the conformal time defined , along the constant @xmath11-coordinate lines , as @xmath272 . the use of @xmath271 is advantageous since , as it is indicated by fig.[tr1 ] , the radial null geodesics ( see the thin dotted lines ) are well approximated . this , along with @xmath273{\mathrm{d}\rho}$ ] and that the @xmath10-integral of @xmath274 is negligible , implies that in the @xmath270 coordinate plane they satisfy the differential equation @xmath275\approx \mp 1 $ ] . ] by @xmath276 lines . on top and bottom , the orthogonal projections of the world sheet of @xmath277 can be seen . the curvy boundary represents the singularity . it is remarkable that @xmath277 is positive everywhere outside the apparent horizon . however , beyond this boundary , it starts to oscillate such that the amplitude of this oscillation is increasing towards the intersection of the singularity and the @xmath278 line . it is important to note that in spite of this oscillatory behaviour of @xmath277 , the misner - sharp mass is always nonnegative . on fig.[tr1 ] , the location of the apparent horizon is plotted on the @xmath11-@xmath271 coordinate plane . in accordance with the generic expectations which are also justified by figs.[coll]and[coll1 ] , there are two marginally trapped surfaces on each @xmath20 time level surface . the @xmath11-coordinate of the inner one is decreasing while that of the outer one is increasing . the plotted radial null geodesics indicate that the apparent horizon on the left panel is everywhere achronal as it is expected . it is also remarkable that the outer part of the apparent horizon tends to become almost null very rapidly . on the right panel of fig.[tr1 ] the misner - sharp mass is plotted as a function of @xmath11 along the future apparent horizon , @xmath279 . as it is clearly visible , @xmath75 is increasing monotonously along @xmath279 and it tends to the value of the adm mass @xmath280 which is equal to the misner - sharp mass @xmath281 in the present case since the matter field is of compact support . as the area of the apparent horizon is @xmath282 and @xmath283 , this plot is also in accordance with the penrose inequality @xmath284 . as mentioned above , a true scalar curvature singularity develops by the end of the time evolution inside the black hole region when both the kretschmann scalar , @xmath285 , and the ricci scalar curvature , @xmath286 , tend to infinity . as it is justified by the log - log plots on left panel of fig.[r3 ] , the blow up rate of @xmath285 , @xmath286 and @xmath75 , at fixed @xmath287 location , are @xmath288 , @xmath289 and @xmath290 , respectively . on the right panel of fig . [ r3 ] , the @xmath11-dependence of the critical exponents @xmath291 , @xmath292 and @xmath293 of the blow up rates @xmath294 and @xmath295 are shown . it is visible that these curvature scalars , along with the misner - sharp mass , diverge faster close to the origin . we would like to point to the fact that the blow - up of the kretschmann scalar is completely consistent with the relation @xmath296 derived by christodoulou for the case of the collapse of a massless scalar field in @xcite . interestingly , in spite of the divergence of the ricci scalar , the quantity @xmath297 remains finite . thereby the einstein - hilbert action @xmath298 remains finite as well , for any particular choice of @xmath299 with @xmath300 in @xmath301\times\mathbb{s}^2 $ ] . whenever the topology of the initial data surface , @xmath212 , is chosen to be @xmath23\times\mathbb{s}^2 $ ] , there has to be two origins on @xmath212 in the considered spherically symmetric setting . motivated by and mimicking the simplicity of the geometric setup of the frw cosmological model , we used procedure b ( see section [ inidata ] ) and chose the functional form of the initial @xmath12 and @xmath3 as @xmath303 to have a nearly central symmetric time evolution , the initial data for the scalar field was chosen to be slightly asymmetric by requiring @xmath51 and @xmath90 to take the form @xmath304 the parameters @xmath305 , @xmath306 , @xmath307 , @xmath308 , @xmath309 , @xmath119 and @xmath310 in the above functional expressions were fixed as @xmath311 while the number of the spatial grid points @xmath312 was chosen to be 1,000 . on fig.[s3-energy ] the time evolution of the gravity - matter energy density distribution associated with a shell of radius @xmath11 , @xmath269 , is shown . the left panel of fig.[s3-mt - action ] shows a somewhat exotic distribution of the marginally trapped surfaces produced by the time evolution . this plot justifies that the formation of the curvature singularity is censored , i.e. , a connected future trapped region develops in advance to the appearance of the singularity . there are two apparently smooth curves starting at the bottom and ending at the opposite upper corner where @xmath313 and @xmath314 are identically zero , respectively . these curves intersect at @xmath315 . the corresponding @xmath6-sphere is a maximal surface . the points on the thick part of the curves represent future marginally trapped @xmath6-surfaces while the thin parts depict the location of the past marginally trapped @xmath6-surfaces . thereby the associated past apparent horizon may also be referred as the boundary of the dynamical white hole region . on the right panel of fig.[s3-mt - action ] the time dependence of the spatial @xmath316-volume , @xmath317 , is shown . it increases exponentially up to @xmath318 after which the tendency is reversed and an even faster decay of @xmath50 is experienced . an interesting new feature of the evolution is that the metric function @xmath12 appears to blow up while approaching the singularity . however , its blow - up is compensated by the decay of @xmath13 ; the proper time @xmath319 was found to be finite along any constant @xmath11 world - line meaning that the singularity is in finite proper time distance from the initial data surface . the @xmath3-dependence of the square root of the kretschmann scalar and the misner - sharp mass , along the world - line @xmath287 , are indicated on the left panel of fig.[s3baby - r ] . on the right the @xmath11-dependence of the critical exponents @xmath292 , @xmath291 and @xmath293 are given . interestingly , christodoulou s relation @xmath320derived only for the case of the collapse of a massless scalar field in a spacetime with spatial topology @xmath43 in @xcite does apply here as well . the value of the einstein - hilbert action , @xmath321 remains finite as it is indicated by the @xmath10-dependence of the lagrangian @xmath322 , shown on the left panel of fig.[s3-mt - action ] . whenever the topology of the initial data surface @xmath212 is @xmath324 , the simplest initial configuration to start with is a homogeneous ` torus ' with constant initial radius and with constant initial expansion . accordingly , we used scheme a and chose the initial @xmath3 and @xmath88 as @xmath325 while the number of the spatial grid points @xmath312 was chosen to be 4,000 . for the the initial @xmath51 and @xmath90 , we used the relations in ( [ homogen_psi_wave1.initial.eq ] ) with the parameter values in ( [ s3par ] ) , while the values of the metric parameters were chosen as @xmath326 it is important to keep in mind that now @xmath11 is a periodic coordinate , along the @xmath327 factor . this periodicity length was chosen to be @xmath116 . notice also that according to the above choice of initial data , there is no origin at all on @xmath212 . on fig.[s1xs2-energy ] the time evolution of the gravity - matter energy density distribution associated with a shell of radius @xmath11 , @xmath269 , is shown . as opposed to the other cases , @xmath269 tends to zero at the singularity . the kretschmann scalar blows up in the singularity such that christodoulou s relation @xmath328 holds . the location of the future and past apparent horizons are also indicated on the left panel of fig.[s1xs2-mt - action ] . the appearance of curvature singularities are preceeded by the formation of trapped regions . as in the previous case , both future and past apparent horizons are formed . they intersect at the @xmath6-surfaces with @xmath329 . interestingly enough , while the @xmath6-surface with @xmath330 and @xmath331 are maximal the other two @xmath6-surfaces have no definite character . more precisely , they are locally maximal in the spatial @xmath11-direction whereas they are locally minimal in the timelike @xmath10-direction . the spacetime region with untrapped surfaces is very limited in the present case . it is represented by the points between the future ( thick solid line ) and the past ( thin solid line ) apparent horizons on the left panel of fig.[s1xs2-mt - action ] . the @xmath10-dependence of the lagrangian @xmath332 indicates ( see the right panel of fig.[s1xs2-mt - action ] ) that the einstein - hilbert action , @xmath333 remains finite as in the previous cases . finally , the @xmath10-dependence of the spatial @xmath316-volume , @xmath334 , is also shown on the right . the volume starts to increase rapidly , then a short oscillation followed by an extremely fast contraction can be seen . in most of the former numerical simulations , one of the aims was to minimise the extent of the trapped region . this was achieved by making use of techniques like singularity ( and trapped region ) avoiding slicings and black hole excision the latter was originally suggested by unruh in 1984 ( for its first adaptation in numerical simulations see , e.g. , @xcite ) . the unsatisfactory aspect of this trouble avoiding attitude , and the reason for choosing here the simplest possible framework of spherically symmetric dynamical configurations are justified by the following comments of david hilbert ( 1902 ) : the key technical achievements that , as we believe , have not been applied before in numerical simulations , are the following : * a strongly hyperbolic ( symmetrisable ) first order system of evolution equations was singled out for @xmath0-dimensional spherically symmetric gravitating systems . * the analytic setup ensures that time evolution can be studied on equal footing in trapped and untrapped regions . * the numerical framework applies this analytic setup and incorporates the techniques of amr . by making use of these technical developments we achieved the following results : * by introducing a suitable evolution equation for the lapse function @xmath13 , the extent of the investigated spacetime domain was enlarged significantly and the physical singularities were approached arbitrarily closely everywhere . * the location of the future and past apparent horizons has been determined . * detailed investigation of the rate of curvature blow - up while approaching the singularity . * the einstein - hilbert action remained finite in all the investigated cases , in spite of the blow up of the ricci scalar . as one of our aims was to clear up some conceptual issues of topology change , let us close this paper by related comments . in dealing with the problem of topology changes , one should start by recalling the result of geroch @xcite which asserts that if topology change develops then there must exist either closed causal curves beyond the cauchy horizon or a spacetime singularity has to appear . due to tipler s theorem @xcite , the latter case manifests itself if einstein s equations are imposed . based on these results , it has been widely held that no indication of topology changes will ever show up in classical general relativity , therefore the quantum theory should be investigated to see whether they may occur . the idea that the topology may change in quantum gravity was originally proposed by wheeler @xcite . since then , one of the most important questions in any quantum theory of gravity is whether there is a non - zero probability for the topology of space to change . there have been a number of classical investigations aiming to demonstrate the feasibility of topology change by making use of the techniques of differential geometry and topology ( see , e.g. , @xcite ) . nevertheless , no quantitative investigations have been carried out yet . therefore it is important to emphasise that some of our findings provide the first definite quantitative support to all the former speculations concerning the existence of topology changes . in summarising our pertinent observations , we can say , in accordance with the above recalled results of geroch and tipler , that instead of having a regular cauchy horizon , along with a causality violating region beyond , a spacetime singularity develops at the `` new origins '' . it was also found that the kretschmann scalar always blows up there . it is important to emphasise that the einstein - hilbert action @xmath335 was found to be bounded in all of our investigations . the existence of time developments with apparent topology changes may be significant in quantum theoretical considerations . for instance , in the sum over histories approach to quantum gravity , the transition amplitude for the topology change between the two riemannian manifolds @xmath336 and @xmath337 is supposed to be given by the formula @xmath338 \,\mathcal{d}g_{ab}\ , , \ ] ] where the boundary of @xmath16 is the disjoint union of @xmath339 and @xmath340 , @xmath341 denote the einstein - hilbert action and @xmath342 is the space of @xmath0-dimensional lorentzian geometries on @xmath16 , while the sum is over all @xmath0-manifolds whose boundary is the disjoint union of @xmath339 and @xmath340 with riemannian metrics @xmath343 and @xmath344 ( see , e.g. , @xcite ) . admittedly , the right hand side of ( [ fint ] ) is completely formal and is far from being defined as yet . nevertheless , regardless of the specific form of the measure on the space of metrics , @xmath345 , it is widely held that not only the smooth lorentzian metrics but all the metrics with finite einstein - hilbert action should be taken into account in evaluating the functional integral . therefore the spacetimes with topology changes investigated in this paper , might be of interest in quantum theoretical considerations . it is of obvious importance to know whether the methods introduced in this paper could be adopted in more generic geometrical setup allowing the presence of gravitational waves and with the inclusion of binary black holes . we would like to mention that by making use of the analytic setup suggested in @xcite such a generalisation seems to be possible . the results of our corresponding investigations will be published elsewhere . this research was supported in part by otka grant k67942 .

we present a unified scaling description for the dynamics of monomers of a semiflexible chain under good solvent condition in the free draining limit . we consider both the cases where the contour length is comparable to the persistence length and the case . our theory captures the early time monomer dynamics of a stiff chain characterized by dependence for the mean square displacement(msd ) of the monomers , but predicts a first crossover to the rouse regime of for , and a second crossover to the purely diffusive dynamics for the entire chain at . we confirm the predictions of this scaling description by studying monomer dynamics of dilute solution of semi - flexible chains under good solvent conditions obtained from our brownian dynamics ( bd ) simulation studies for a large choice of chain lengths with number of monomers per chain n = 16 - 2048 and persistence length lennard - jones ( lj ) units . these bd simulation results further confirm the absence of gaussian regime for a 2d swollen chain from the slope of the plot of which around changes suddenly from , also manifested in the power law decay for the bond autocorrelation function disproving the validity of the wlc in 2d . we further observe that the normalized transverse fluctuations of the semiflexible chains for different stiffness as a function of renormalized contour length collapse on the same master plot and exhibits power law scaling at extreme limits , where for extremely stiff chains ( ) , and for fully flexible chains . finally , we compare the radial distribution functions obtained from our simulation studies with those obtained analytically .

1401.8247

macromolecules adsorbed on substrate surfaces occur in many different contexts , from materials science to biophysics , _ e.g. _ , biomolecules interacting with cell membranes . hence the understanding of conformations and dynamics of macromolecules in such a ( quasi- ) two - dimensional geometry has been of long - standing interest @xcite . note also that many methods to characterize polymer conformations experimentally , _ e.g. _ , electron microscopy @xcite , atomic force microscopy ( afm ) @xcite , fluorescence microscopy of suitably labelled biopolymers @xcite , require that these macromolecules are attached to a substrate . in this context , considerations of macromolecules confined to a strictly two - dimensional geometry are of interest , at least as a limiting case . the same statement holds when one considers macromolecules confined to nanoslits with non - adsorbing walls , a topic that also has found much recent interest . while the statistical mechanics of completely flexible polymers in @xmath17 dimensions has been studied extensively since a long time and is well understood @xcite , under many circumstances it should be taken into account that macromolecules are stiff and not flexible on small scales @xcite . this is true both for simple synthetic polymers _ e.g. _ , polystyrene , alkane chains , _ etc . _ , and for various biopolymers , _ e.g. _ , double - stranded ( ds ) and single stranded ( ss ) dna , polysaccharides , proteins , _ etc . apart from very stiff polymers ( _ e.g. _ , actin , titin , microtubules , _ etc . _ ) the persistence length @xmath1 @xcite characterizing the stiffness typically is much less than the contour length @xmath0 of a macromolecule , and the crossover from rod - like behavior to the behavior of flexible polymers needs to be considered . as is well - known , the scales of interest range from the sub - nanometer scale to the micrometer scale @xcite , and hence in the theoretical modeling coarse - grained models must be used @xcite . in the present work , we wish to address the problem of polymer conformation and dynamics for semi - flexible polymers in two dimensions , using molecular dynamics simulations of a bead - spring type model with a bond angle potential by which we can control the stiffness of the chains over a wide range . we note that the standard analytical coarse - grained description in terms of the kratky - porod @xcite model for wormlike chains in @xmath17 dimensions is not very useful , since it neglects excluded volume effects completely , although they are known to be very important in @xmath17 @xcite . a study of semiflexible polymers in terms of a lattice model @xcite is expected to yield valid results for universal properties of semiflexible polymer , _ i.e. _ , on length scales much larger than the persistence length ; but on smaller scales it can describe only stiffness of the type similar to that of alkane chains , where @xmath1 is of the order of the typical length of `` all trans '' sequences , in between monomers taking a gauche ( @xmath18 ) minimum in the torsional potential . for such cases _ e.g. _ , dsdna we expect that the ( small ) flexibility of the macromolecules is due to fluctuations in bond lengths and bond angles , rather than disorder in the population of states in the torsional potential , and then the present off - lattice model is more realistic . in addition , the lattice work @xcite applying the pruned enriched rosenbluth method ( perm ) @xcite could not address the dynamics of the chains at all . previous work on the dynamics of single semiflexible chains in dilute solution @xcite has focused on the case @xmath19 almost exclusively , and most of the work @xcite has studied the effect of hydrodynamic ( hd ) interactions mediated by the solvent . assuming that the substrate surface provides a stick boundary condition with respect to solvent fluid flow , one can show @xcite that hd interactions are essentially screened , and hence are ignored here ( as well as in our preliminary communication where a small part of our results were presented @xcite ) from the outset . while the kratky - porod worm - like - chain ( wlc ) model has been found to be grossly inadequate to describe a semiflexible chain in 2d , vast amounts of analytical and numerical work have been accumulated using the wlc model as the starting point @xcite . recent experimental results of confined biopolymers on a 2d substrate are also analyzed using the well known results of wlc model @xcite . therefore , in the following section we summarize the main results of the wlc model which we will revisit in the subsequent sections to compare our simulation results . the organization of the paper is as follows . in the next section we introduce the wlc model . next in sec . [ scaling_section ] a scaling theory is derived where we show that monomer dynamics of a semiflexible polymer exhibits a double crossover as a function of time . we then introduce the bead spring model for a semiflexible chain in sec . [ model_section ] . the results of bd simulation are presented in sec . [ bd_section ] , which is divided into two sub - sections : the equilibrium properties are presented in sec . [ eqlb_section ] ; in sec . [ dynamics_section ] among other results we validate the predictions of the scaling theory using ( bd ) simulation for chains of different length and stiffness . the hamiltonian corresponding to the bending energy for the wlc model is given by @xmath20 where @xmath21 is the position vector of a mass point , @xmath0 is the inextensible contour length , @xmath22 is the bending rigidity , and the integration is carried out along the contour @xmath23 @xcite . using symmetry arguments for the free energy it can be shown @xcite that the chain persistence length @xmath1 for a wlc in 2d and 3d are given by + @xmath24 @xmath25 + + the model has been studied quite extensively applying path integral and other techniques @xcite and exact expressions of various moments of the distribution of monomer distances along the chain have been worked out . the end - to - end distance in the wlc model is given by @xcite @xmath26\right ) . \label{rn_wlc}\ ] ] in the limit @xmath27 one gets @xmath28 and the chain behaves like a gaussian coil ; for @xmath29 , @xmath30 and the chain behaves like a rod . evidently the model neglects the excluded volume ( ev ) ( eqn . [ lj ] of the bead - spring model in sec.-[model_section ] ) interaction and hence interpolates between rod and gaussian limits only . the wlc model can be viewed as a limiting case of a freely rotating chain @xcite , where the correlation between bond vectors @xmath31 and @xmath32 is assumed to follow @xmath33 where @xmath34 and the characteristic length is defined as the persistence length @xmath1 . from eqn . [ bexp ] it then immediately follows that @xmath1 can be calculated from the bond angle @xmath35 , where @xmath36 is the unit vector of the corresponding bond vector @xmath37 as follows : @xmath38 likewise , dynamics of the wlc model have been explored using langevin type of equation @xcite . one can expect that the dynamics of a stiff chain will be dominated by transverse fluctuations ( bending modes ) @xcite and that the short time dynamics will be governed by the chain persistence length . indeed a relaxation dynamics using the wlc hamiltonian ( eqn . [ wlc ] ) approach yields an expression for fluctuation @xmath39 which crosses over to simple diffusion at late time @xcite . as we will see later from our results that even for a gaussian chain a more `` complete theory '' should have captured an intermediate regime characterized by a growth law @xmath40 for a fully flexible chain for an intermediate time when the fluctuation becomes of the order of radius of gyration of the chain . however , the langevin theories for the wlc chain did not describe this regime . we first develop the scaling description for the dynamics of a two dimensional semiflexible chain in the free draining limit . the free draining limit is of particular interest in 2d because it often satisfies the experimental conditions , such as dna confined in a 2d substrate where the effect of hydrodynamics is negligible . for a wlc several theories based on langevin dynamics have been developed most of which indicate a @xmath41 dependence of the transverse fluctuation of the msd @xmath42 ( see eqn . [ g1 ] ) with time @xmath43 for the stiff chain . therefore , we start with the eqn . [ granek ] below derived by granek @xcite and farge and maggs @xcite using a langevin dynamics framework for the wlc hamiltonian@xmath44 where @xmath45 is the monomer reorientation rate . prefactors of order unity are omitted throughout @xcite . for early time the monomer dynamics will be independent of the chain length @xmath46 until the fluctuations in monomer position become of the order of @xmath1 . therefore , denoting the time when the first crossover occurs as @xmath47 and substituting @xmath48 and @xmath49 in eqn . [ granek ] we immediately get @xmath50 for @xmath51 the monomer dynamics is described by @xmath52 according to eqn . [ granek ] until @xmath53 at time @xmath54 . the width of this region is independent of @xmath46 and solely a function of @xmath1 ( see fig . [ phase ] ) . for @xmath55 the dynamics is governed by the rouse relaxation of monomers of a fully flexible ev chain in 2d characterized by @xmath56 . @xmath57 characterizes the onset of the purely diffusive regime when @xmath58 @xcite . recall that the exponent @xmath59 that describes the scaling of the end - to - end distance @xmath60 according to @xmath61 is @xmath62 in 2d . , @xmath63(a ) and @xmath64 , @xmath65(b ) . blue ( red ) symbols , solid and dashed lines correspond to @xmath52(@xmath66 ) , black dashed , solid lines correspond to @xmath67 . here the power laws eqn . [ granek ] , [ gmid ] and [ g3_scaling ] are plotted , using units of @xmath68 , @xmath69 , @xmath70 . the width of each region shows how these regimes depend on @xmath1 and @xmath46 . note that in reality we expect a very gradual change of slope on the log - log plot at both crossover times , rather than sharp kinks.,title="fig : " ] + , @xmath63(a ) and @xmath64 , @xmath65(b ) . blue ( red ) symbols , solid and dashed lines correspond to @xmath52(@xmath66 ) , black dashed , solid lines correspond to @xmath67 . here the power laws eqn . [ granek ] , [ gmid ] and [ g3_scaling ] are plotted , using units of @xmath68 , @xmath69 , @xmath70 . the width of each region shows how these regimes depend on @xmath1 and @xmath46 . note that in reality we expect a very gradual change of slope on the log - log plot at both crossover times , rather than sharp kinks.,title="fig : " ] note that when we increase the stiffness of the chain at fixed chain length , the times @xmath47 and @xmath57 can be made to coincide ; this happens for @xmath71 , as expected : when the contour length and the persistence length are of the same order , the regime described by eqn . [ gmid ] is no longer present . we will verify the time dependence of @xmath72 at various regimes from bd simulations . we then obtain @xmath57 as follows : @xmath73 substituting @xmath47 from eqn . [ tau1 ] in above @xmath74 at @xmath75 @xmath76 substituting eqn . [ gmid ] for @xmath77 we get @xmath78 we also note that the dynamics of the center of mass is given by @xmath79 the `` phase diagram '' for the crossover dynamics in terms of @xmath46 , and @xmath1 are shown in fig . [ phase ] . notice that for a stiffer chain the region for @xmath80 for which we predict @xmath81 requires to study very long chains and therefore , is hard to see in simulation for a stiffer chain . we have used a bead spring model of a polymer chain with excluded volume , spring and bending potentials as follows @xcite . the excluded volume interaction between any two monomers is given by a short range lennard - jones ( lj ) potential with cut off and shifted in its minimum . @xmath82+\epsilon \;\;\;\;\;\;\;\ ; \mathrm{for~~ } r\le 2^{1/6}\sigma \nonumber \\ & = & 0 \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\ ; \;\;\;\;\;\;\;\;\;\;\;\ \mathrm{for~~ } r > 2^{1/6}\sigma\;. \label{lj}\end{aligned}\ ] ] here , @xmath83 is the effective diameter of a monomer , and @xmath84 is the strength of the potential . the connectivity between neighboring monomers is modeled as a finitely extensible nonlinear elastic ( fene ) spring with @xmath85 where @xmath86 is the distance between consecutive monomers , @xmath87 is the spring constant and @xmath88 is the maximum allowed separation between connected monomers @xcite . the chain stiffness is introduced by adding an angle dependent interaction between successive bonds as ( fig . [ model ] ) @xmath89 here @xmath90 is the complementary angle between the bond vectors @xmath91 and @xmath92 , respectively , as shown in fig . [ model ] . the strength of the interaction is characterized by the bending rigidity @xmath22 . introducing unit tangent vector @xmath93 ( @xmath94 ) we note that the discretized version of eqn . [ wlc ] can be written as @xmath95 where @xmath96 is the bond length in our simulation . therefore , for fixed bond length eqn . [ ebend ] represents the discrete version of the wlc model of eqn . thus the ev effect introduced through eqn . [ lj ] are completely absent in the wlc model . subtended by the vectors @xmath91 and @xmath92 . ] -0.5truecm we use the langevin dynamics with the following equation of motion for the i@xmath97 monomer @xmath98 here @xmath99 is the monomer friction coefficient and @xmath100 , is a gaussian white noise with zero mean at a temperature t , and satisfies the fluctuation - dissipation relation : @xmath101 the reduced units of length , time and temperature are chosen to be @xmath83 , @xmath102 , and @xmath103 respectively . for the spring potential we have chosen @xmath104 and @xmath105 , the friction coefficient @xmath106 , the temperature is kept at @xmath107 . the choice of the fene potential along with the lj interaction parameters ensures that the average bond - length in the bulk @xmath108 . with the choice of these parameters the probability of chain crossing is very low . we also find that the average bond - length @xmath109 is almost independent of the range of chain stiffness parameter ( @xmath110 ) used in our simulation . strictly speaking , the contour length @xmath0 is @xmath111 . the equation of motion is integrated with the reduced time step @xmath112 following the algorithm proposed by van gunsteren and berendsen @xcite . we have carried out brownian dynamics ( bd ) simulation for a wide range of chain length ( @xmath46=16 - 2048 ) and bending constant ( @xmath22 = 0 - 320 ) . because of the argument given in sec . [ scaling_section ] very long chains were needed to clearly identify the crossover regimes . first we present the equilibrium properties of the chains in section [ eqlb_section ] followed by the dynamical quantities presented in sec . [ dynamics_section ] . from the bd simulation we have monitored the average @xmath113 and replacing @xmath114 in eqn . [ lp1 ] calculated the chain persistence length for various values of @xmath22 . one expects that eqns . [ lp2d ] and [ lp1 ] must give results that agree with each other when the persistence length is much larger than the range of the excluded volume interaction , but that the two results agree even for small value of @xmath115 we believe is a nontrivial result . the comparison of calculated @xmath1 by different methods is shown in table [ table ] . we also observe that the @xmath1 calculated using eqn . [ lp1 ] practically has no dependence on chain length @xmath46 . we have used eqn . [ lp1 ] for further analysis of our data in the subsequent sections . note that this behavior differs from the result found by hsu _ for a lattice model @xcite , where a renormalization of @xmath1 by excluded volume was shown to occur . thus on length scales of order @xmath1 there is no strict universality between different models . we emphasize , however , that the persistence length , when it is supposed to measure the local intrinsic stiffness of the chain ( as supposed in the kratky - porod model ) , can not be estimated from the asymptotic decay of the bond vector correlation function @xmath116 with the `` chemical distance '' @xmath23 along the chain , that is the conventional definition given in all the polymer physics textbooks : as will be shown below ( sec . [ bond_section ] ) , we verify the predicted @xcite power law behavior for very long chains and large @xmath23 , previously seen already for a lattice model of semi - flexible chains by hsu _ et al . _ although lattice and continuum models have different statistical properties when one considers lengths of the scale @xmath1 , for much larger scales the behavior should be universal , and hence this power law decay is expected . we also note that a definition of the persistence length dating back to flory , where one considers the correlation of the first bond vector @xmath117 with the end - to - end vector @xmath60 , @xmath118 which has been advocated by cifra @xcite as an `` exact expression '' , must similarly be refuted : in 2d , redner and privman @xcite have shown that @xmath119 for large @xmath46 is logarithmically divergent with @xmath46 already for a simple self - avoiding walk ( saw ) . for completeness , we mention that an analogous definition for inner bond vectors @xmath120 even shows a power - law divergence , @xmath121 , both in 2d and 3d ( see hsu et al . @xcite and @xcite ) . thus we urge that the results and discussion presented in this section need to be taken seriously in writing future review articles and newer edition of the existing textbooks . the extension of flory theory for a semi - flexible chain has been done by schaefer , joanny , and pincus @xcite and nakanishi @xcite which states that the rms of the end - to - end distance @xmath122 in @xmath123 spatial dimensions exhibits the following scaling relation @xmath124 where @xmath125 is the bond length ( @xmath109 in our simulation ) . for @xmath17 this reduces to @xmath126 . in other words if the end - to - end distance is scaled by the appropriate power of the persistence length @xmath1 , then this renormalized end - to - end distances @xmath127 for different values of the chain stiffness parameter @xmath22 will fall onto the same master plot . for a large combination of chain length @xmath46 and stiffness parameter @xmath22 we observe excellent fit to our equilibrium data for @xmath128 to eqn . [ scaling ] as shown in fig . [ r1n ] . versus @xmath129 for various values of the chain stiffness parameter . all the data for different stiffness parameter collapse on the same master plot . the solid line is a fit to a straight line . only data points for which the contour length exceeds the persistence length were included in this plot . ] -0.5truecm it is worth noting that the persistence length calculated from eqn . [ lp1 ] using the formula from wlc model uses the local correlation , namely the angle between the subsequent bond vectors and hence is expected to provide a decent value of the persistence length when ev is also included . the excellent collapse of the data for @xmath130 for various values of the stiffness parameter ( @xmath131 ) on the same master curve indicates that eqn . [ lp1 ] _ can be used as the standard definition of persistence length even in presence of the ev effect_. having established the definition of persistence length which validated eqn . [ scaling ] , we now use eqn . [ rn_wlc ] presented in section [ wlc_section ] to analyze the bd simulation results for the end - to - end distance . please note that limiting cases of eqn . [ rn_wlc ] are either a gaussian coil ( @xmath28 for @xmath132 ) or a rod ( @xmath30 for @xmath133 ) . we have used simulation results to plot @xmath134 for a large number of values(@xmath135 ) of @xmath12 ( @xmath136 ) . as a function of @xmath12 obtained from different combinations of chain length @xmath46 and stiffness parameter @xmath22 ( log - log scale ) . the solid ( maroon ) line is a fit to the formula @xmath137 for @xmath138 . the inset shows the same for small values of @xmath139 which clearly indicates that limiting slope of unity ( @xmath140 ) for @xmath141 . ] -0.5truecm we have also taken additional care that a given value of @xmath12 is generated for different combinations of @xmath0 and @xmath1 . these results are shown in fig . for @xmath142 we observe that @xmath143 while for @xmath15 the data very nicely fit with @xmath144 . this is consistent with prior mc results using a lattice model by hsu _ et al . _ . however , since our studies are done in continuum we are able to get data that is for much shorter length scales . the fact that in the rod - like regime ( @xmath145 ) the best fit " exponent is 0.95 rather than the asymptotic value 1.0 is due to the fact that for @xmath22=32 and 64 the rods " still exhibit nonnegligible transverse fluctuation unlike truly stiff rods . the gaussian behavior that eqn . [ rn_wlc ] implies @xmath146 for large @xmath0 would mean a horizontal straight line in fig . [ rn ] , but no indication of such a behavior is seen . the simulation data then implies the strict absence of a gaussian limit for 2d swollen semi - flexible chains due to severe dominance of the ev interaction . this result should be contrasted with the simulation results in 3d , where one sees a gradual crossover from rod limit to the ev limit ( in 3d ) passing through a gaussian regime @xcite . it is reasonable to define an average axis for a polymer chain in the rod limit ( @xmath147 ) . in this limit using wlc chain hamiltonian the transverse fluctuation with respect to this average axis has been shown @xcite to obey the following equation @xmath148 the above equation implies that the roughness exponent @xcite @xmath149 ( @xmath150 ) for a weakly bending rod . starting from an extremely stiff chain where the transverse fluctuations are expected to be governed by the roughening exponent @xmath149 , if we approach the limit of fully flexible chain , then the fluctuations become isotropic and in this limit one can expect that @xmath151 , so that in 2d @xmath152 . in order to calculate the transverse fluctuation , in our simulation , for each configuration of the polymer chain , we choose @xmath153 as the longitudinal axis and calculate transverse fluctuations as follows : @xmath154 where @xmath155 is the perpendicular distance of the @xmath156 monomer with respect to the instantaneous direction @xmath60 . we have repeated this calculation for several chain lengths from extremely stiff chains to fully flexible chains . this is shown in fig . [ tfl](a ) . if one does not analyze the data carefully , one might be misled to conclude that the exponent @xmath157 increases gradually with the chain stiffness @xmath22 . however , as we will see that the proper interpretation requires a scaling description in terms of @xmath12 as scaling variable . the interesting aspect of this rescaled dimensionless transverse fluctuation is shown in fig . [ tfl](b ) as a function of rescaled length @xmath12 where the rescaled fluctuations collapse on the same master curve . this plot exhibits a maximum and then decreases gradually for large value of @xmath12 . it is worth noting that analytical results do not exist for chains with intermediate stiffness . the physical origin of this maximum can be interpreted as follows . starting from the stiff chain limit when @xmath142 it increases for a more flexible chain when the chain undergoes shape changes from rod to ellipsoid - like blob , and finally to isotropic spherical blobs . naturally , when chain flexibility is defined in units of the persistence length for some value of @xmath12 ( @xmath158 3 from fig . [ tfl](b ) ) the fluctuation becomes maximum before it becomes isotropic . to the best of our knowledge this result is new and in principle can be used to measure the persistence length of a semiflexible polymer by measuring the transverse fluctuations using fluorescence probes . this would require numerically analyzing a large number of images of semiflexible chains ( of a given kind of polymer ) with varying contour length . for each image of a chain one can extract @xmath0 as well as the end - to - end vector @xmath60 and then use eqn . [ lbot ] to extract @xmath159 . plotting then @xmath11 versus @xmath0 one would find @xmath160 from the position of the maximum of this plot . the orientational correlation between successive bonds decays exponentially in a wlc model according to eqn . however , recent monte carlo ( mc ) studies by hsu , paul , and binder have verified that a swollen semiflexible chain in 2d exhibits a power law decay as a function of the separation @xmath23 between the beads @xcite given by @xmath161 for a fully flexible chain @xmath162 in 2d so that @xmath163 . it is then expected that a semiflexible chain will exhibit the same behavior when @xmath15 . while for very stiff chain one needs to have extremely long chain to see this asymptotic regime for large @xmath15 , yet satisfying the condition @xmath164 , from our simulation we clearly see this trend for moderate values of @xmath22 . we calculated the bond correlation function from its definition and tested both eqn . [ bexp ] and [ bpower ] . first in fig . [ semilog ] we show the semi - log plot of eqn . [ bexp ] . is calculated from the slope of the fitted line , which is very close to the values ( correspondingly 105.8 , 52.6 , 25.9 , 12.6 , 6.05 , 3.31 ) from @xmath165 ( see table [ table ] ) . ] the straight lines are fitted only with the first several data points in order to get values of @xmath1 which are close to those calculated from @xmath165 . the deviation from the initial slope ( only after few points ) clearly shows that eqn . [ bexp ] predicted by the wlc model does not hold good for a 2d swollen chain as expected from the result of fig . [ rn ] . fig . [ loglog ] shows the log - log plot of eqn . [ bpower ] where we have also included the graph for a fully flexible chain for reference . as a function of @xmath23 for various combinations of @xmath46 and @xmath22 . while for large @xmath22 the asymptotic slope of @xmath166 is preempted by finite size effect , for @xmath167 and @xmath168 the slope of @xmath169 is clearly visible ; ( b ) same as in ( a ) but using rescaled variable @xmath170 which shows excellent data collapse for @xmath171 . the inset shows the same but only for the cases to emphasize that in the limit @xmath172 there is a perfect data collapse for the power law scaling . , title="fig : " ] + as a function of @xmath23 for various combinations of @xmath46 and @xmath22 . while for large @xmath22 the asymptotic slope of @xmath166 is preempted by finite size effect , for @xmath167 and @xmath168 the slope of @xmath169 is clearly visible ; ( b ) same as in ( a ) but using rescaled variable @xmath170 which shows excellent data collapse for @xmath171 . the inset shows the same but only for the cases to emphasize that in the limit @xmath172 there is a perfect data collapse for the power law scaling . , title="fig : " ] please note that even for a fully flexible chain it requires a rather long chain length ( @xmath173 ) to clearly see the regime with slope with @xmath174 over an appreciably broad range of abscissa values . for comparison we put the graph for a shorter fully flexible chain of @xmath175 . here the curve starts to decrease faster before it reaches the slope corresponding to @xmath174 . naturally for stiffer chains ( which could be thought of a flexible chain of length @xmath12 for this purpose ) for the maximum chain length @xmath176 considered in this paper we only see the regime characterized by @xmath163 only for @xmath177 and 4.0 . it is expected that the bond vector correlation also exhibits a scaling behavior when studied as a function of @xmath170 , as even for moderate values of @xmath178 ( see fig . [ semilog ] ) , this length rescaling @xmath179 overrides the exponential decay for small @xmath170 . with a choice of distance between monomers in units of @xmath1 all chains are expected to behave as fully flexible chains . therefore , if we use the renormalized distance @xmath170 to replot fig [ loglog](a ) then one expects that the power law correlation for chains with different stiffness will collapse on the same master plot . this is shown in fig . [ loglog](b ) . we observe excellent data collapse for @xmath171 . deviations from this collapse occur at a progressively larger value of @xmath170 as the ratio @xmath12 increases either by increasing the contour length @xmath0 for a fixed @xmath1 or for the same contour length @xmath0 and lowering the value of @xmath1 . this is expected , since eqn . [ bpower ] can hold only for @xmath164 . of course , a numerical study of the regime @xmath180 for large @xmath1 is prohibitively difficult . however , the inset shows both the data collapse and the @xmath163 regimes for two chain lengths ( @xmath173 and 2048 ) and for two values of @xmath22 ( @xmath115 and @xmath181 ) which proves beyond doubt that for @xmath182 the bond autocorrelation exhibits a power law decay of a fully flexible chain . the hamiltonian for the wlc chain model has been studied by many analytic technics assuming that for moderate chain lengths and stiff enough chains the ev effect will not dominate . since we already established the severe dominance of the ev effect , gaussian regime is absent for a 2d swollen chain . here we compare the radial distribution functions for chains with different stiffness . in particular , we compare the results from our simulation with analytic results of wilhelm and frey @xcite in 2d . for the wlc model wilhelm and frey @xcite have derived expressions for the radial distribution functions @xmath183 ( both in 2d and 3d ) in terms of infinite series . in 2d the expression for @xmath183 is given as follows . \times \nonumber \\ \mathrm{exp}\left [ -\frac{(l+1/4)^2}{2t(1-r)}\right ] d_{3/2}\left [ \frac{2l+1/2}{\sqrt{2t(1-r ) } } \right ] , \label{frey}\end{aligned}\ ] ] where @xmath185 , @xmath186 and @xmath187 is a parabolic cylinder function . for the whole range of values of @xmath22 in our simulation , when we plot the analytic result of eqn . [ frey ] we find that this series is fully dominated by the first term and we did nt see visual differences from the plot that includes the first four terms . therefore , in fig . [ distri ] we plot only the first term of this series ( @xmath188 ) term given by @xmath189^{5/4 } \times \nonumber \\ \mathrm{exp}\left [ -\frac{(1/4)^2}{2t(1-r)}\right ] d_{3/2}\left [ \frac{1/2}{\sqrt{2t(1-r ) } } \right ] \label{freyl0}\end{aligned}\ ] ] please note that for the radial distribution function wilhelm and frey @xcite does not take into account excluded volume effects . in fig . [ distri ] we have also included the distribution for a completely flexible chain ( @xmath190 ) for comparison purposes . for @xmath181 and 8 ( _ i.e. _ , @xmath191 ) the simulated distribution still is intermediate between the behavior of fully flexible chains ( which we expect to result for @xmath192 for all @xmath22 ! ) and the distribution of the kratky - porod model . we now look at the dynamics of a swollen semi - flexible chain . we are using a brownian dynamics ( bd ) scheme and therefore , hd effects are not included in our studies . however , for polymers adsorbed on a flat surface simulation studies have shown that the hd effects are negligible for no - slip boundary conditions @xcite . thus we expect the results are relevant for a fair comparison of experimental studies . also we would like to point out that unlike equilibrium properties , computational time increases dramatically to study time dependent properties for the same chain length . thus the diffusion and relaxation studies for the longest chain length reported in this paper took significant time for well converged runs . we verified that the diffusion constant @xmath193 is independent of the persistence length @xmath1 and depends only on the chain length @xmath46 and scales as @xmath194 with very good accuracy , as expected in a bd formalism . next we consider monomer relaxation dynamics of the chain . following previous work for the relaxation of a fully flexible chain @xcite we have monitored various quantities pertaining to a single monomer relaxation . these quantities have been studied in the past using bd algorithm @xcite , and dynamical monte carlo algorithms ( dmc ) @xcite , including the bond - fluctuation model ( bfm ) @xcite . the dynamics of the individual monomers and the collective dynamics for the whole chain have been characterized by the functions @xmath42 , @xmath195 , @xmath196 , @xmath197 , and @xmath198 . they are defined as follows : @xmath199 ^ 2\rangle , \label{g1}\\ & g_2(t ) = \nonumber \\ & \langle \left [ \left(\mathbf{r}_{n/2}(t ) - \mathbf{r}_{cm}(t ) \right ) - \left ( \mathbf{r}_{n/2}(0 ) - \mathbf{r}_{cm}(0)\right ) \right]^2 \rangle,\label{g2}\\ & g_3(t ) = \langle \left [ \mathbf{r}_{cm}(t ) - \mathbf{r}_{cm}(0)\right]^2 \rangle , \label{g3}\\ & g_4(t ) = \langle \left [ \mathbf{r}_{end}(t ) - \mathbf{r}_{end}(0)\right]^2 \rangle , \label{g4}\\ & g_5(t ) = \nonumber \\ & \langle \left [ \left ( \mathbf{r}_{end}(t ) - \mathbf{r}_{cm}(t)\right ) - \left(\mathbf{r}_{end}(0 ) - \mathbf{r}_{cm}(0)\right ) \right]^2 \rangle . \label{g5}\end{aligned}\ ] ] the quantities @xmath42 , @xmath196 , and @xmath197 reflect the time dependence of the position of the middle monomer , the center of mass , and the end monomer of the chain respectively . at late times for distances greater than the gyration radius the functions @xmath42 , @xmath196 and @xmath197 will describe the motion of the entire chain . consequently , @xmath200 for @xmath201 where @xmath57 is given by eqn . [ tau2 ] . the quantities @xmath195 and @xmath198 on the contrary measure the relative displacement of the middle and the end monomer with respect to the center of mass of the chain . the functions @xmath195 and @xmath198 saturate at finite static values @xmath202 and @xmath203 respectively , since for @xmath204 the orientations of the vectors @xmath205 , @xmath206 are uncorrelated with their counterparts at @xmath207 . to study monomer dynamics , we have carried out bd simulation for various chain lengths @xmath208 and for chain stiffness @xmath209 . we only show a limited set of data . as a reference and for comparison with the data for chains with @xmath210 , we first show data for a fully flexible chain where we expect to see a single crossover dynamics from @xmath81 at an early time for @xmath211 to a purely diffusive dynamics for the entire chain ( @xmath212 ) . this is shown in fig . [ kappa0 ] for chain length @xmath175 and 1024 respectively . ( @xmath213 ) for a fully flexible chain ( @xmath214 ) of length @xmath175 . ( b ) @xmath215 ( @xmath216 ) for a fully flexible chain of length @xmath173.,title="fig : " ] + ( @xmath213 ) for a fully flexible chain ( @xmath214 ) of length @xmath175 . ( b ) @xmath215 ( @xmath216 ) for a fully flexible chain of length @xmath173.,title="fig : " ] we have checked that the graphs for other chain length are similar and have the @xmath217 dependence for the fully flexible chain . at late times the functions @xmath218 and @xmath219 saturates and the functions @xmath72 , @xmath220 and @xmath221 grow linearly as a function of time , becoming practically indistinguishable from each other . similar studies have been reported earlier by grest and kremer @xcite and by gerroff , milchev , paul , and binder @xcite . however , our studies are much more exhaustive and quantitatively captures the crossover from @xmath217 to the purely diffusive regime which were only qualitative in previous studies and for shorter chains . we now show data for the double crossover to support our scaling analysis for @xmath210 . in particular we show data for chain length @xmath175 with @xmath2222.0 and @xmath173 with @xmath2222.0 , 4.0 respectively . unlike fig . [ kappa0 ] for @xmath190 , plots for @xmath42 , @xmath195 shown in fig . [ kappa_2_4 ] are characterized by a @xmath41 slope which then crosses over to the regime characterized by @xmath223 and that @xmath42 eventually merges with @xmath196 . when we compare fig . [ kappa_2_4](a ) and fig . [ kappa_2_4](b ) consistent with the prediction of scaling theory we observe that the extent of the @xmath41 region in both the graphs are the same as they have the same value of @xmath224 , although the chain lengths are different . likewise , comparing fig . [ kappa_2_4](b ) and fig . [ kappa_2_4](c ) we note that since both plots have the same chain length the beginning of the second crossover occur almost at the same time ( @xmath225 ) but since the latter chain is twice as stiff it has a wider @xmath41 regime resulting in a narrower span of @xmath217 regime . thus fig . [ kappa_2_4 ] unambiguously confirms predictions from the scaling theory . we observe that at early time @xmath226 ; however the width of the region @xmath227 is narrower than that of @xmath81 and it exhibits a slightly lower value of slope before saturation . considering that the crossovers are broad , we believe that this is due to finite size effect as it is evident if we compare fig . [ kappa_2_4](a ) and fig . [ kappa_2_4](b ) . in the latter case , which is for a larger chain length , the difference between @xmath42 and @xmath195 are much smaller and we notice that the difference begins at a later time . ideally for very large system the point when @xmath42 would change its slope towards @xmath212 , @xmath195 also tend to saturate . the sharpness of this feature would require a much larger system . we also note the same feature in fig . [ kappa0 ] . ( black ) , @xmath195 ( red ) and @xmath196 ( green ) as a function of time on a log - log scale for chain length @xmath175 and @xmath228 . the blue and magenta dashed lines correspond to straight lines @xmath229 , and @xmath230 , respectively , where @xmath231 and @xmath232 are constants . ( b ) same but for @xmath173 and @xmath228 . ( c ) same but for @xmath173 and @xmath233 . note that for a fully flexible chain the slope of the curve @xmath234 would monotonously increase with time , unlike the present case.,title="fig : " ] + ( black ) , @xmath195 ( red ) and @xmath196 ( green ) as a function of time on a log - log scale for chain length @xmath175 and @xmath228 . the blue and magenta dashed lines correspond to straight lines @xmath229 , and @xmath230 , respectively , where @xmath231 and @xmath232 are constants . ( b ) same but for @xmath173 and @xmath228 . ( c ) same but for @xmath173 and @xmath233 . note that for a fully flexible chain the slope of the curve @xmath234 would monotonously increase with time , unlike the present case.,title="fig : " ] + ( black ) , @xmath195 ( red ) and @xmath196 ( green ) as a function of time on a log - log scale for chain length @xmath175 and @xmath228 . the blue and magenta dashed lines correspond to straight lines @xmath229 , and @xmath230 , respectively , where @xmath231 and @xmath232 are constants . ( b ) same but for @xmath173 and @xmath228 . ( c ) same but for @xmath173 and @xmath233 . note that for a fully flexible chain the slope of the curve @xmath234 would monotonously increase with time , unlike the present case.,title="fig : " ] + but the plots for @xmath42 quite clearly show three distinct scaling regimes of @xmath52 crossing over to @xmath81 and then merging with @xmath67 at late times . we have further confirmed the existence of this double crossover by plotting @xmath235 as shown in fig . [ g1_min ] . the existence of an initial plateau ( @xmath236 ) , followed by a decay ( @xmath237 ) , and of a minimum ( near @xmath57 ) before the diffusion ( @xmath238 ) starts further demonstrates quite conclusively that the exponent changes from @xmath239 . we would like to mention that because of the width of the @xmath217 regime becomes narrower for stiffer chains we were unable to see this regime unambiguously in simulation of shorter chains and/or larger @xmath22 ( _ e.g. _ , for @xmath175 and @xmath240 ) . while for @xmath175 the double crossover is clear for @xmath177 , but becomes ambiguous for @xmath241 which required an increased chain length of @xmath173 . as expected , the crossovers are rather gradual , spread out over a decade in time @xmath43 each , and hence for chains that are not long enough the existence of these regimes is missed in previous work . as a function of @xmath43 corresponding to the plots of fig . [ kappa_2_4](a ) ( red ) , ( b ) ( blue ) and ( c ) ( green ) respectively . in each graph the minimum occurs in the intermediate regime characterized by @xmath217 . , title="fig : " ] + the interplay of rouse modes and the bending modes with respect to the monomer dynamics of semiflexible chains was considered in early work by harnau _ et al . _ @xcite , in the framework of a rouse model generalized by higher order terms to account for chain stiffness . they ignored excluded volume , and considered a single chain length , attempting to model @xmath242 alkanes in a melt . they found that their results were neither consistent with the rouse behavior ( @xmath243 ) nor with the power law due to bending modes ( @xmath244 ) . in our view , the chain length studied in this work was too small to observe both power laws separately , rather all their data fall in a regime of smooth crossover . having established the double - crossover we now further investigate the consequence of scaling prediction that the first crossover occurs at time @xmath245 when @xmath246 . [ g1g2 ] shows a plot of @xmath247 as a function of rescaled time @xmath248 which shows data collapse for @xmath249 and @xmath250 for various combinations of chain length @xmath46 and @xmath22 confirming the length and the time scales for these crossovers . note that the scaling theory of section [ scaling_section ] implies @xmath251 where @xmath47 , @xmath57 are the times defined in eqn . [ tau1 ] and [ tau2 ] . if the first argument of the scaling function @xmath252 , @xmath253 is small in comparison to unity , we can approximate eqn . [ gtilde ] as @xmath254 , which reduces to eqn . [ granek ] . for @xmath255 we can rewrite the scaling function as @xmath256 note that @xmath257 remains constant when we increase @xmath46 and @xmath1 by the same factor : this observation explains that the scaling of the data in fig . [ g1g2 ] encompasses the full range of times . and @xmath258 as a function of @xmath259 on a log - log scale for chain lengths @xmath260 , 1024 with @xmath177 , 4.0 respectively . the dot - dashed and dashed lines correspond to slopes 0.75 ( maroon ) and 0.6 ( magenta ) respectively . ] in conclusion , we have studied conformations , fluctuations , and crossover dynamics of a swollen semiflexible chain in 2d . we first developed a scaling theory which generalizes early time monomer dynamics of a fully flexible chain for a semiflexible chain characterized by its contour length and the persistence length . we predict a double crossover which arises due to the presence of an additional length scale introduced through the chain persistence length . monomer dynamics up to a length scale @xmath1 is independent of chain length and is characterized by the @xmath41 power law . at a later time @xmath261 when the size of the fluctuations becomes bigger than @xmath1 the dynamics begin to look like that of a fully flexible chain and characterized by the well known @xmath262 growth . both of these exponents have been discussed in the literature separately but have not been emphasized that before the entire chain reaches purely diffusive regime , there ought to be two and not one crossover , the first crossover differentiates chains of different stiffness . previously the dynamics of monomer msd of semiflexible polymers has also been studied by harnau _ et al . _ @xcite , using a rouse - type model generalized to include chain stiffness . they saw a gradual crossover , in between bending modes and rouse modes , but did not consider the scaling description of the crossover . note that excluded volume effects were absent in their model , and hence it is not applicable in @xmath17 dimensions . motivated by recent lattice mc results for a swollen chain in 2d predicting the absence of a gaussian regime we undertook similar studies in 2d continuum using bd simulation . while checking our data for the rms end - to - end distance for chain of different contour length and persistence length we discovered that we regain the well known results for the end - to - end distance due to schaefer , joanny , and pincus @xcite and nakanishi @xcite provided we use the definition of the persistence length given by either the lattice or continuum version of the kratky - porod wlc model . we explain this by noting that the persistence length being a local property of a chain does not depend on the ev interaction . this is further reassured when we note that the persistence length calculated this way does not depend on the chain length unlike well used textbook definition of persistence length where the projection of the end - to - end vector on the first bond is used as the definition and does depend on the chain length . therefore , we emphasize that the latter definition needs to be discarded . we also confirm the absence of the gaussian regime in the continuum bead - spring model where the swollen chain for @xmath142 behaves like a rod and thereafter always behaves like a swollen chain . considering that there are increased number of activities to explore the properties of biomolecules on a surface , our result ( fig . [ rn ] ) will be extremely valuable to analyze the experimental data correctly for stiff molecules on flat surfaces . it has not escaped our attention that many such reported analyses are still done using the wlc model and/or calculating the chain persistence length from projected end - to - end to the first bond . transverse fluctuations in a stiff chain has been addressed analytically in the literature only in the extremely stiff chain limit where one finds that it is described by the roughening exponent . analytic calculations for moderately stiff chains are hard to carry out , and to date there are no results for transverse fluctuations spanning the entire regime from a stiff to a fully flexible chain . we have numerically obtained this result and pointed out that the appropriate length variable to analyze the data is to use the persistence length as the unit of length . when we use @xmath12 as the length scale to plot transverse fluctuation we discover the non - monotonic behavior of this fluctuation reaching a maximum for some @xmath263 . we point out that this universal scaling of the transverse fluctuation can be used to measure the persistence length of the chain . another accompanying consequence of the absence of gaussian regime for a swollen chain in 2d is the decay of the bond correlation function which exhibits a power law decay . again , by choosing the normalized contour segment @xmath170 as the appropriate variable we regain the exponent @xmath163 which describes the decay of bond autocorrelation for a fully flexible chain . we must point out that many of these results and analyses on chain conformations and equilibrium fluctuations point to a common theme . in the limit @xmath15 we recover the expected behavior of a fully flexible chain and chains with different stiffness exhibit universal scaling behavior when persistence length is chosen as the unit of length . this general idea extends to our study of monomer dynamics as well . we have provided a new scaling theory of monomer dynamics for semiflexible polymers in 2d . our theory predicts novel crossover dynamics at an intermediate time when the fluctuations of the monomers @xmath264 . around this time the monomer dynamics become the same as that of a fully flexible swollen chain characterized by @xmath265 in 2d . the theory expands the existing scaling theory for monomer dynamics for a wlc and that of a fully flexible chain to include the effect of the chain persistence length . fully flexible swollen chains are self - similar objects , while a polymer segment up to its own persistence length is not . therefore , it is expected that for length scales up to @xmath1 the dynamics will have different characteristics due to bending modes arising out of the chain stiffness . the ev effect is almost negligible for the @xmath41 regime and therefore , our result is the same as that of previous studies using wlc hamiltonian @xcite . for the @xmath217 regime originating from the ev effect , where the monomer dynamics are governed by rouse relaxation of a fully flexible chain , our theory elucidates the exact role of chain persistence length neither contained in wlc model nor studied before . we also validate our new scaling theory by extensive bd simulation results . a subtle issue concerns the limit @xmath22 towards infinity while keeping the contour length @xmath0 fixed . then transverse motions of the monomers relative to each other , in a coordinate system where the @xmath266-axis is fixed along the rod - like polymer , are completely suppressed . in the laboratory coordinate system " , however , the rod still can make random transverse motions , namely rigid body rotations and translations . however , in addition to those motions still motions of the monomers relative to each other along the axis of the rod are possible . these motions may give rise to a transient @xmath267 behavior , as a model calculation for a one - dimensional harmonic chain shows @xcite . however , our data for @xmath268 for large @xmath22 due to the smoothness of crossovers did not allow to clearly separate this mode of motions from the displacements due to transverse fluctuations . in the present manuscript we have ignored hd interactions as they are not significant for 2d swollen chains on a substrate . however we now present simple estimates of generalization of our results in 3d and/or in presence of hd interactions which will be relevant for a 3d swollen chain . in the free draining limit , the results @xmath41 will remain the same in 3d @xcite , but the intermediate rouse relaxation regime will be characterized by @xmath269 ( @xmath270 in 3d ) , for the case where the ev is relevant ( _ i.e. _ , msds exceeding @xmath271 ) . for msd in between @xmath272 and @xmath273 gaussian behavior prevails , @xmath274 , and hence @xmath275 in that regime . the crossover between flexible and stiff chain dynamics in 3d in the free draining limit was studied by steinhauser _ et al . _ @xcite , but no scaling analysis is done . replacing rouse relaxation by zimm relaxation one immediately sees that in presence of hd interaction the intermediate regime is characterized by @xmath276 @xcite . notice that in this case @xmath59 cancels out and this relaxation should be the same in 2d and 3d . however , as shown by hinczewski and netz @xcite , very complicated crossovers occur in this case . we now make some comments about some recent experiments to study monomer dynamics . this is typically done using fcs where a tagged monomer can be directly watched in real time . however , as has been mentioned by petrov _ et al . _ @xcite that since the @xmath41 regime or the intermediate regimes ( either @xmath40 or @xmath277 ) occur at much shorter time scales compared to the longest relaxation time , unless extreme caution is taken for the measurement of msd of a labeled particle , the interpretation can be misleading , especially for shorter dna fragments @xcite . for dsdna of length @xmath278 base pairs ( which is equivalent to @xmath279 fluorescence correlation spectroscopy ( fcs ) studies of petrov _ et al . _ observed the zimm regime characterized by a @xmath277 power law . however , a clear demonstration of the double crossover is lacking in recent experiments with biopolymers @xcite . this is partly due to lack of resolution of the experiments and partly due to the fact that lacking any theoretical predictions for this phenomenon , researchers did not specifically investigate the precise behavior of msd before the onset of overall chain diffusion very carefully . we will provide some physical arguments why the experimental detection can be hard : a simple calculation for fig . [ phase ] shows that in order for the width of the @xmath41 and @xmath280 to be equal ( in logarithmic scale ) one needs @xmath281 in 2d . in other words for a stiffer chain one needs a very long chain to see the @xmath280 regime . indeed in our simulation we found ( not shown here ) that for @xmath282 , 32 , and 64 , the results with chain length up to @xmath175 are largely dominated by the @xmath41 regime and we did not clearly see the @xmath280 regime . it is only after we lowered the value of @xmath22 and used longer chain ( @xmath173 ) , we identified these two regimes quite conclusively ( fig . [ kappa_2_4 ] ) . we suspect that the same might happen in experiments @xcite . for extremely stiff chains the @xmath217 ( or @xmath283 in 3d ) region can be extremely narrow and could either be missed or the rather smooth double crossover might be mistakenly interpreted as a single crossover ( with @xmath277 in 2d ) . therefore , we believe that these results will not only promote new experiments but will be extremely valuable in identifying and interpreting different scaling regimes for the monomer dynamics of semiflexible polymers . the research has been partially supported ( ah ) by the ucf office of research & commercialization and the ucf college of science seed grant . ab acknowledges travel support from deutsche forschungsgemeinschaft at the institut fr physik , johannes gutenberg - universitt , mainz . we thank both the referees for their constructive criticism and comments . de gennes , adv . colloid interface sci . * 27 * , 189 ( 1987 ) g.j . fleer , m.a . cohen - stuart , j.m.h.m . scheutjens , t. cosgrove , and b. vincent , _ polymers at interfaces _ ( chapman & hall , london , 1993 ) e. eisenriegler , _ polymers near surfaces _ ( world scientific , singapore , 1993 ) r.r . netz and d. andelman , phys . rep . * 380 * , 1 ( 2003 ) t. takabayashi , y. morija and f. oosawa , biochim . biophys . acta * 492 * , 357 ( 1977 ) b.t . stokke , a. elgsaeter , g. skjak - break , and d. smidsrod , carbohydr . 160 * , 13 ( 1987 ) s. trachtenberg and i. hammel , j. struct . biol . * 109 * , 18 , ( 1992 ) j. bednar , p. furrer , v. kalritch , a.z . stasiak , j. dubochet and a. stasiak , j. mol . biol . * 254 * , 579 ( 1995 ) r. schnauer , p. bertoncini , g. machaidze , u. aebi , j.c . perriard , m. hegner , and i. agaskova , j. mol . biol . * 349 * , 367 ( 2005 ) p. papadopoulos , g. floudas , i. schnell , i. lieberwirth , t.q nguyen , and h .- a . klok , biomacromolecules * 7 * , 618 ( 2006 ) x.e . li , w. lehman , and s. fischer , j. mol . biol . * 395 * , 327 ( 2010 ) h.g . hansma and h. hoh , annu . biomol . struct . * 23 * , 115 ( 1994 ) n. yoshinaga , k. yoshikawa , and s. kidoaki , j. chem 116 * , 9926 ( 2002 ) f. valle , m. favre , p. de los rios , a. rosa , and g. dietler , phys . 95 * , 158105 ( 2005 ) j. moukhtar , e. fontaine , c. faivre - moskalenko , and a. arneodo , phys . lett . * 98 * , 178101 ( 2002 ) k. rechendorff , g. witz , j. adamcik , and g. dietler , j. chem . phys . * 131 * , 095103 ( 2009 ) n. mcke , k. klenin , r. kirmse , m. bussiek , h. herrmann , m. hafner , and j. langowski , plos one * 4 * , issue 11 , e7756 ( 2009 ) j. moukhtar , c. faivre - moskalenko , p. milani , b. audit , c. vaillant , e. fontaine , f. mongelard , g. lavorel , p. st .- jean , p. bouvet , f. argout , and a. arneodo , j. phys . b * 114 * , 5125 ( 2010 ) b. maier and j.o . rdler , phys . lett . * 82 * , 1911 ( 1999 ) b. maier and j.o . rdler , macromolecules * 33 * , 7185 ( 2000 ) p.g . de gennes , _ scaling concepts in polymer physics _ ( cornell - university press , ithaca , 1979 ) a.d . sokal , in _ monte carlo and molecular dynamics simulations in polymer science _ , ed . by k. binder , chapter 2 ( oxford univ . press , new york , 1995 ) p.j . flory , _ statistical mechanics of chain molecules _ ( hanser , new york , 1989 ) a.y . grosberg , a.r . khokhlov , _ statistical physics of macromolecules _ ( aip press , new york , 2004 ) m. rubinstein and r.h . colby , _ polymer physics _ ( oxford univ . press , 2003 ) d.a.d . parry and e.n . baker , rep . . phys . * 47 * , 1133 ( 1984 ) h .- p . hsu , w. paul and k. binder , macromolecules * 43 * , 3094 ( 2010 ) k. binder , in _ monte carlo and molecular dynamics simulations in polymer science _ ed . by k. binder , chapter 1 ( oxford univ . press , new york , 1995 ) c. hyeon and d. thirumalai , nature commun . * 2 * , 487 ( 2011 ) h .- p . hsu , w. paul , and k. binder , macromol . theory simul . * 20 * , 510 ( 2011 ) o. kratky and g. porod , j. colloid sci . * 4 * , 35 ( 1949 ) r. harris and j. hearst , j. chem . phys . * 44 * , 2595 ( 1966 ) h .- p . hsu , w. paul . and k. binder , epl * 95 * , 68004 ( 2011 ) h .- p . hsu and k. binder , j. chem . phys . * 136 * , 024901 ( 2012 ) h .- p . hsu , w. paul , and k. binder , j. chem . phys . * 137 * , 1749002 ( 2012 ) h .- p . hsu and k. binder , soft matter * 9 * , 10512(2013 ) p. grassberger , phys . e * 56 * , 3682 ( 1997 ) h .- p . hsu and p. grassberger , j. stat . phys . * 144 * , 597 ( 2011 ) d.w . schaefer , j.f . joanny and p. pincus , macromolecules * 13 * , 1280 ( 1980 ) l. harnau , r.g . winkler , and p. reineker , j. chem . phys . * 104 * , 6355 ( 1996 ) k. kroy and e. frey , phys . e * 55 * , 3092 ( 1997 ) r. shusterman , s. alon , t. gavrinyov , and o. krichevsky , phys . lett . * 92 * , 048303 ( 2004 ) e.p . petrov , t. ohrt , r.g . winkler and p. schwille , lett . * 97 * , 258101 ( 2006 ) m. hinczewski and r.r . netz , epl * 88 * , 18001 ( 2009 ) m. hinczewski , x. schlagberger , m. rubinstein , o. krichevsky , and r.r . netz , macromolecules * 42 * , 860 ( 2009 ) m. hinczewski and r. r. netz , physica a * 389 * , 2993 ( 2010 ) . steinhauser , j. schneider , and a. blumen , j. chem . phys . * 130 * , 164902 ( 2009 ) a. winkler , p. virnau , k. binder , r.g . winkler , and g. gompper , j. chem . * 138 * , 054901 ( 2013 ) a. huang , r. adhikari , a. bhattacharya and k. binder , epl * 105 * , 18002 ( 2014 ) . l. harnau , r. g. winkler , and p. reineker , epl * 45 * , 488 ( 1999 ) .

to shed light onto the circumnuclear environment of 22 ghz ( 1.3 cm ) ho maser galaxies , we have analyzed some of their multi - wavelength properties , including the far infrared luminosity ( fir ) , the luminosity of the\lambda5007 $ ] emission line , the nuclear x - ray luminosity , and the equivalent width of the neutral iron emission line ( ew ( k ) ) . our statistical analysis includes a total of 85 sources , most of them harboring an active galactic nucleus ( agn ) . there are strong anti - correlations between ew ( ) and two `` optical thickness parameters '' , i.e. the ratios of the x - ray luminosity versus the presumably more isotropically radiated $ ] and far infrared ( fir ) luminosities . based on these anti - correlations , a set of quantitative criteria , ew ()ev ,}$ ] and can be established for compton - thick nuclear regions . 18 ho maser galaxies belong to this category . there are no obvious correlations between the ew ( k ) , the $ ] luminosity and the isotropic ho maser luminosity . when comparing samples of seyfert 2s with and without detected ho maser lines , there seem to exist differences in ew ( k ) and the fraction of compton - thick nuclei . this should be studied further . for agn masers alone , there is no obvious correlation between fir and ho maser luminosities . however , including masers associated with star forming regions , a linear correlation is revealed . overall , the extragalactic fir - ho data agree with the corresponding relation for galactic maser sources , extrapolated by several orders of magnitude to higher luminosities .

0912.2159

thanks to a large number of dedicated surveys during the last 15 years , the number of galaxies known to host 22ghz ( @[email protected] cm ) h@xmath12o masers has increased tenfold and 85 sources at distances larger than those of the magellanic clouds have been reported so far to exhibit h@xmath12o maser emission ( e.g. , braatz & gugliucci 2008 , darling et al . 2008 , greenhill et al . . some of them are agn - related , while others are found in off - nuclear star forming regions . among the masers clearly identified as agn - related , a large fraction ( @xmath140% ) have been identified as `` disk - maser '' candidates ( kondratko et al . their maser spots are associated with central , typically parsec sized molecular accretion disks and maser line spectra show high velocity features " ( red - shifted and blue - shifted features ) , in addition to the systemic velocity components . the study of disk - maser sources has become a very important subject , permitting mass estimates of supermassive black holes , distance estimates of galaxies , and providing a perspective to improve the accuracy of the hubble constant and to constrain the equation of state for the elusive dark energy ( e.g. braatz et al . 2009 ) . observations show that among agn masers , h@xmath2o maser spots locate preferentially in the nuclear regions of seyfert 2 or liner galaxies and most of them are heavily obscured ( @xmath13@xmath710@xmath14@xmath15 ; braatz et al . 1997 , madejski et al . 2006 , zhang et al . 2006 , greenhill et al . the nuclear x - ray source is generally believed to heat the gas to temperatures suitable for 22ghz h@xmath2o maser emission ( neufeld et al . 1994 ) , which is supported by a relation between maser luminosity , the unabsorbed intrinsic nuclear x - ray luminosity , and the mass of the black hole ( kondratko et al . 2006 , su et al . 2008 ) . for the obscured nuclear regions h@xmath2o masers and x - rays provide , unlike optical data , deeply penetrating views . the x - ray absorption along the line - of - sight to the nucleus can provide through spectral model fitting important information on the nature of the circumnuclear environment . with modern x - ray telescopes , high quality x - ray spectra have been obtained for more than 30 h@xmath2o maser galaxies . based on an analysis of such spectra , the nuclear column densities of maser host galaxies were investigated by zhang et al . ( 2006 ) and greenhill et al . while these studies have shown that nuclear h@xmath2o masers are mostly found in environments with high column density , it is still open whether h@xmath2o masers preferentially arise from compton - thick ( @xmath13@xmath710@xmath16@xmath15 ) agn . to reduce limitations and uncertainties when modeling x - ray spectra , here we try to provide additional constraints to evaluate line - of - sight column densities . the iron @xmath6 emission line ( @xmath16.4kev ) is the most prominent line in the x - ray spectra of agn . it is believed to be produced either by transmission through the absorbing material ( leahy & creighton 1993 ) or via the process of x - ray scattering / reflection by the cold iron in the nuclear accretion disk ( e.g. , lightman & white 1988 ; fabian et al . 2000 ) or torus ( ghisellini et al . 1994 ) . so it provides useful information to constrain the column density that absorbs the continuum and , in some cases , to distinguish between compton - thick and -thin sources . flat hard x - ray spectra and high equivalent widths ( ew ( k@xmath5 ) ) of the iron line were found in highly absorbed sources and are generally used to identify compton - thick nuclei ( e.g. , matt 1997 ) . however , currently there are no quantitative criteria on ew ( k@xmath5 ) yet as a probe of the gas absorption . unlike the x - ray absorption , @xmath3\lambda 5007 $ ] and fir emission have often been used as comparatively isotropic indicators of the intrinsic nuclear power ( e.g , mulchaey et al . 1994 , alonso - herrero et al . 1997 , shu et al . 2007 ) . the forbidden @xmath3\lambda 5007 $ ] line emission originates in the narrow line region ( nlr ) , which is assumed to be isotropic for both type 1 and type 2 systems . due to possible shielding effects in the torus , netzer et al . ( 2006 ) proposed that the @xmath17\lambda 3727 $ ] line is a better tracer than @xmath3 $ ] for its considerably larger emission region . however , this still needs to be checked , since the @xmath17 $ ] line can also arise from high mass star formation . and their results are based on a high - luminosity high - redshift agn sample . for the less luminous , relatively nearby agn as those of our maser sample , the @xmath3\lambda 5007 $ ] luminosity has been extensively used as an indicator of the intrinsic agn power ( e.g. , mulchaey et al . 1994 , alonso - herrero et al . 1997 , maiolino & rieke 1995 , heckman et al . 2005 , risaliti et al . 1999 , bassani et al . 1999 , panessa et al . 2006 , lamastra et al . the fir emission , produced on linear scales much larger than the nuclear torus , should be free of biases caused by the viewing angle and should be highly isotropic ( e.g. , mulchaey et al . so even without a detailed knowledge of the x - ray spectrum , a comparison of the observed x - ray luminosity with those of the more isotropic tracers of nuclear power can help us to determine the gas absorption along the line of sight toward the agn ( e.g. , bassani et al . 1999 ) . in the following , x - ray , @xmath3 $ ] , and fir data are collected for those h@xmath2o maser galaxies which are located farther than the magellanic clouds and which are known to host 22ghz h@xmath2o masers . the data are analyzed to determine gas column densities along the line of sight toward the agn and to derive properties related to activity in highly obscured and therefore particularly elusive compton - thick nuclear environments . data related to the galaxies with detected h@xmath2o maser emission , located at larger distances than the magellanic clouds , are compiled in table1 . among the 85 published h@xmath2o maser sources , there are 66 agn masers ( 63 `` megamasers '' with isotropic luminosities @xmath18 and 3 `` kilomasers '' with @xmath19 ) . the megamasers are classified as such because of their luminosity . while most of them have not yet been studied in detail , all thoroughly investigated megamasers are not separated by more than a few pc from the line of sight to the agn of their parent galaxy . ten kilomasers are related to star formation regions and another nine kilomasers are still awaiting interferometric observations to investigate their nature . all activity types of these h@xmath2o maser galaxies are listed in table1 and their dominant types ( following bennert et al . 2009 ) are used for the statistics ( for details , see table2 ) . most of the agn maser sources are seyfert 2 galaxies or liners . llcccccccc * ngc17 * & sy2,lirg , hii & & & & & 1.33 & & & + ngc23 & liner , lirg , hi & & & & & 1.09 & 5.89 & 37.9 & ho97 + _ ic10 _ & & & & & & 3.65 & 12.5 & 442.85 & ho97 + ngc235a & sy1 & & & & & 1.1 & & & + ngc253 & sy2,sbg , hii & & & & & 73.56 & & & + ngc262(mrk348 ) & sy2 & a & @xmath20 & 482 & awa00 & 0.34 & 6.02 & 177 & dah88 + & & g & @xmath21 & 1270 & bas99 & & & & + irasf01063 - 803 & & & & & & 0.37 & & & + ngc449(mrk1 ) & sy2 & & & & & 0.32 & 5.86 & 236.51 & dah88 + ngc520 & sbg , hii & & & & & 2.86 & & & + _ ngc598 _ & hii & & & & & 67.71 & 4.55 & 0.24 & ho97 + * ngc591 * & sy2 & x & @xmath22 & 20 & gua05b & 0.28 & & 178 & whi92 + ngc613 & sy2,hii & & & & & 2.6 & & & + ic0184 & sy2,hii & & & & & & & & + ngc1052 & sy2,liner & b & @xmath23 & 400 & ter02 & 0.14 & 2.82 & 13.3 & dah88 + * ngc1068 * & sy2,sy1 & x & @xmath24 & 462 & cap06 & 25.01 & 7.00 & 6780 & dah88 + & & a & @xmath25 & 350 & bas99 & & & & + ngc1106 & sy2 & & & & & 0.19 & & & + mrk1066 & sy2 & c & @xmath26 & 23 & shu07 & 1.14 & 8.51 & 514 & whi92 + * ngc1320 * & sy2 & x & @xmath27 & 496.2 & gre08 & 0.29 & & & + * ngc1386 * & sy2 & x & @xmath28 & @xmath29 & gua05b & 0.72 & 5.7 & 1020 & sto89 + & & a & @xmath30 & 20 & bas99 & & & & + iras03355 + 0104 & sy2 & & & & & 0.13 & & & + _ ic342 _ & sy2,hii & & & & & 8.29 & 7.69 & 3.4 & ho97 + mg j0414 + 0534 & qso1 & & & & & & & & + * ugc3193 * & & & & & & 0.36 & & & + ugc3255 & sy2 & & & & & 0.19 & & & + mrk3 & sy2 & x & @xmath31 & 590 & bia05 & 0.57 & 6.67 & 4610 & whi92 + & & b & @xmath32 & 650 & cap99 & & & & + & & a & @xmath33 & 650 & bas99 & & & & + _ ngc2146 _ & hii & & & & & 12.54 & 11.1 & 30.47 & ho97 + vii zw73 & sy2 & & & & & 0.21 & & & + ngc2273 & sy2 & x & @xmath34 & @xmath35&gua05b & 0.7 & 6.92 & 164 & whi92 + & & a & @xmath36 & 125 & ter02 & & & & + * ugc3789 * & & & & & & 0.26 & & & + mrk78 & sy2 & & & & & 0.15 & 6.50 & 242.76 & dah88 + j0804 + 3607 & qso2 & & & & & & & & + _ he2 - 10 _ & sbg & c & & 9.56@xmath37&jr05 & 2.40 & & & + * 2masxj08362280*&sy2 & & & & & 0.17 & & & + mrk1210 & sy2,sy1 & c & @xmath1188 & 840 & zha09 & 0.36 & 5.20 & 580 & ter91 + & & x & 130 & 970 & gua02 & & & & + & & b & @xmath38 & 930 & ohn04 & & & & + & & a & @xmath39 & 160 & awa00 & & & & + * ngc2639 * & sy1.9 & a & @xmath40 & 25.3 & ter02 & 0.43 & 4.16 & 4 & ris99 + ngc2782 & sy1,sbg & c & 990 & @xmath130 & zha06 & 0.91 & 6.67 & 62.19 & ho97 + ngc2824(mrk394)&sy ? & & & & & 0.12 & & & + sbs0927 + 49 & liner & & & & & 0.29 & & & + * ngc2960 * & liner & & & & & 0.24 & & & + ugc5101 & sy1.5,line r , lirg&x&@xmath41 & 8.1 & ima03 & 1.06 & & 191 & kim95 + ngc2979 & sy2 & & & & & 0.2 & & & + ngc2989 & hii & & & & & & & & + _ ngc3034 _ & sbg , hii & & & & & 96.49 & 25.0 & 1615.17 & ho97 + * ngc3079 * & sy2,liner & x & @xmath42 & 33 & cap06 & 5.5 & 25.0 & 92 & ho97 + & & b & @xmath43 & 37@xmath448 & iyo01 & & & & + * mrk34 * & sy2 & & & & & 0.43 & 10.5 & 204.19 & dah88 + _ ngc3359 _ & hii & & & & & 0.70 & & & + * ic2560 * & sy2 & x & 2320@xmath45&38.8@xmath46&til08 & 4.29 & & @xmath47 & ris99 + & & c & 2770@xmath44490 & 38.4@xmath48&mad06 & & & & + * ngc3393 * & sy2 & x & @xmath49 & @xmath50 & gu05a & 0.33 & 4.12 & 316 & dia88 + & & a & @xmath51 & 40 & bas99 & & & & + ngc3556 & hii & & & & & 7.29 & 7.14 & 2.24 & ho97 + arp299(ngc3690 ) & & x , c&@xmath52 & 43.7 & bal04 & 9.43 & 5.88 & 35.6 & ho97 + & & b & @xmath53 & & del02 & & & & + ngc3735 & sy2 & & & & & 1.02 & 6.31 & 33 & ho97 + _ antennae _ & sbg & c & & & & & & & + * ngc4051 * & sy1.5 & x & @xmath54 & 627 & cap06 & 1.32 & 3.33 & 59.99 & ho97 + ngc4151 & sy1.5 & x & @xmath55 & 4510 & cap06 & 1.11 & 3.45 & 1695.39 & ho97 + & & a & 101@xmath445 & @xmath120000 & wea01 & & & & + _ ngc4214 _ & sbg & c & & 243 & har04 & 1.81 & & & + * ngc4258 * & sy1.9,liner & x & @xmath56 & 837 & cap06 & 5.08 & 9.12 & 262 & hec80 + & & a & 250@xmath44100 & 300 & bas99 & & & & + ngc4293 & liner & & & & & 0.58 & 7.69 & 5.95 & ho97 + * ngc4388 * & sy2 & x & @xmath57 & 762 & cap06 & 1.44 & 5.50 & 374@xmath4450&bas99 + & & a & @xmath58 & 1200 & bas99 & & & & + ngc4527 & liner , hii & & & & & 3.80 & & & + * eso269-g012*&sy2 & & & & & 0.23 & & & + ngc4922 & sy2,liner & & & & & 0.58 & 7.14 & 33.03 & kim95 + * ngc4945 * & sy2 & c & 1300 & 500 & don03 & 41.38 & & @xmath47 & ris99 + & & b & @xmath11300 & 540 & gua00 & & & & + & & a & @xmath59 & 350 & bas99 & & & & + ngc5194 & sy2,hii & x & @xmath60 & 48 & cap06 & 6.62 & 8.33 & 228 & ho97 + & & a & @xmath61 & 91.9 & ter02 & & & & + _ ngc5253 _ & sbg , hii & c & & 29.9 & jr05&3.16 & & & + mrk266(ngc5256)&sy2,lirg , sbg & b & 575 & 56 & ris00 & 0.84 & 5.92 & 44.33 & dah88 + ngc5347 & sy2 & c & @xmath62 & 22 & lev06 & 0.27 & & 114 & tra01 + * ngc5495 * & sy2,hii ? & & & & & 0.27 & & & + * circinus * & sy2 & c & @xmath63 & 1400 & smi01 & 26.11 & 19.1 & 6970 & bas99 + ngc5506(mrk1376)&sy1.9 & x & @xmath64 & 5800 & bia03 & 1.06 & 7.20 & 333 & lum01 + & & a & @xmath65 & 8380 & bas99 & & & & + ngc5643 & sy2 & x & 500 & 84 & gua04 & 2.59 & 6.40 & 662 & whi92 + & & a & @xmath66 & 130 & bas99 & & & & + * ngc5728 * & sy2,hii & c & @xmath67 & 133 & shu07 & 0.96 & 5.96 & 761 & sto95 + & & c & 1130 & & zha06 & & & & + * ugc09618ned02*&liner , hii & & & & & 0.92 & & & + * ngc5793 * & sy2 & & & & & 0.6 & & & + ngc6240 & sy2,liner & c & @xmath68 & 170 & pta03 & 2.18 & 17.2 & 135@xmath4420&kim95 + & & a & @xmath69 & 190 & bas99 & & & & + * ngc6264 * & sy2 & & & & & & & & + ngc6300 & sy2 & x & @xmath70 & @xmath71 & mat04b & 2.34 & & 320 & lum01 + * ngc6323 * & sy2 & & & & & & & & + eso103-g035 & sy2,sy1 & a & @xmath72 & 907 & tur97 & 0.37 & 6.31 & 112 & po196 + irasf19370 - 013 & sy2,hii & & & & & 0.26 & & & + * 3c403 * & fr@xmath73 & c & 244@xmath4420 & & kra05 & 0.1 & & & + * ngc6926 * & sy2,hii & & & & & 0.65 & 15.6 & 21.98 & kim95 + am2158 - 380ned02 & sy2,rg & & & & & & & & + * txs2226 - 184*&liner & & & & & & & & + ngc7479 & sy2,liner & & & & & 0.83 & 10.0 & 37.9 & kim95 + ic1481 & liner & & & & & 0.12 & & & + .activity types of h@xmath12o maser host galaxies ( 79 sources with available type ) [ cols="<,^,^,^,^",options="header " , ] \a ) types from ned : inter sy . , intermediate seyfert types ; liner , low - ionization nuclear emission line region ; sbg , starburst galaxy ; frii , fanaroff - riley type ii radio galaxy ; hii , classified as hii region galaxy ; b ) sf - masers , h@xmath2o masers associated with off - nuclear star formation regions ; c ) kilomasers without known type ; d ) agn - masers , h@xmath2o masers associated with agn . besides the type of nuclear activity , we collected multi - wavelength data and parameters for all of the 85 extragalactic h@xmath2o maser sources , including the ew ( @xmath6 ) of the iron emission line , the observed x - ray ( 2 - 10kev ) flux f@xmath74 , the fir flux f@xmath75 , the observed h@xmath76 line intensity ratio and the @xmath3\lambda 5007 $ ] flux density , f@xmath77}$ ] . these data are also presented in table1 . the x - ray data of our sample are collected from the literature , based on observations from _ asca _ , _ bepposax _ , _ chandra _ , and the _ xmm - newton _ satellite . for some sources more than one value were reported in the literature for parameters such as the ew ( k@xmath5 ) of the iron line and the observed x - ray flux , either due to intrinsic variability of the source or due to a different modeling of the spectra . for comparison , all results available from the literature are listed in table1 . for our statistical analyses , observational results were taken from _ xmm - newton _ and _ chandra _ whenever possible . otherwise the most recent measurements were used . for the @xmath3\lambda 5007 $ ] emission line , the fluxes are mainly compiled from dahari & de robertis ( 1988 ) , ho et al . ( 1997 ) , kim et al . ( 1995 ) and whittle ( 1992 ) . again the most recent result was adopted when two or more observations were found . since the nlr may be obscured by material in the host galaxy , the observed @xmath3\lambda 5007 $ ] flux should be corrected to determine its intrinsic flux . the narrow - line balmer decrement was used to estimate this extinction . assuming an intrinsic balmer decrement @xmath78 , the intrinsic @xmath3 $ ] line fluxes were derived from the formula @xmath79,cor}=f_{[\rm o\,{iii}],obs}[(h_{\alpha}/h_{\beta})_{obs}/(h_{\alpha}/ h_{\beta})_{0}]^{2.94}$ ] ( bassani et al . 1999 ) . the @xmath3\lambda 5007 $ ] fluxes were obtained for 46 maser sources and their luminosities were calculated from their extinction - corrected fluxes . the iras ( infrared astronomical satellite ) point source catalogue was used to obtain the infrared fluxes ( 12 , 25 , 60 , 100@xmath80 m ) for our h@xmath2o maser host galaxies . following the method of wouterloot & walmsley ( 1986 ) , the infrared flux ( @xmath81@xmath80 m ) was derived by extrapolating flux densities beyond 12 and 100@xmath80 m and assuming a grain emissivity proportional to frequency @xmath82 . in this way , infrared luminosities could be determined for 76 h@xmath2o maser galaxies . throughout the paper , the luminosity distance was derived using calculators i provided by the nasa extragalactic database ( ned ) , assuming @xmath83 = 0.270 , @xmath84 = 0.730 , and @xmath85=70kms@xmath86mpc@xmath86 ( e.g. , spergel et al . 2007 ) . , 76 sources , left panel ) and the @xmath87\lambda 5007 $ ] line luminosity ( @xmath88 $ ] , 46 sources , right panel ) for h@xmath2o maser sources ( solid lines ) , in units of ergs@xmath86 . for comparison , the distributions are also given for a seyfert 2 sample without known h@xmath2o maser emission ( dashed lines , 38 sources from mulchaey et al . all numbers are plotted on a percent scale ( % ) .,width=377 ] $ ] line luminosity , corrected for extinction , of h@xmath2o maser galaxies on a logarithmic scale . for comparison , the distribution is also given for a seyfert 2 sample without known h@xmath2o maser emission ( dashed lines , 38 sources from mulchaey et al . right panel : the fir versus extinction - corrected @xmath3 $ ] luminosity . the straight line shows a linear fit to the unweighted data.,width=453 ] for our h@xmath2o maser galaxies with available data , the luminosity distributions of both isotropic indicators , the @xmath3 $ ] line and the fir luminosities , are presented in fig . 1 . the left panel shows the histogram of the infrared luminosities for all 76 h@xmath2o maser galaxies with available data ( @xmath89 , hereafter luminosity in logarithmic scale and in units of ergs@xmath86 ) . @xmath90 ranges from 41.8 to 45.5 , with a mean value of [email protected] ( the error denotes the standard deviation of the mean ) . for the sub - sample of 10 masers associated with massive star formation , the mean value of @xmath9@xmath91@[email protected] is slightly fainter than that of the agn masers . however , the difference between both distributions ( not shown here ) is not significant . the histogram shown in the right panel presents the number of the h@xmath2o maser sources as a function of the @xmath3 $ ] line luminosity . it gives the range of @xmath9@xmath92}$]@xmath7 from 39 to 43.2 and a mean value of [email protected] , comparable to results of seyfert 2 galaxies not associated with detected h@xmath2o maser ( mulchaey et al . 1994 ) . for comparison , the latter sample is also presented in fig . 1 ( dashed lines ) . for the two tracers @xmath90 and @xmath93}$ ] , the ratio was calculated and the distribution of the ratio is shown as a histogram in fig . 2 ( left panel ) . no significant difference can be found for the distribution of the ratio between our h@xmath2o maser galaxies and the seyfert 2 galaxy sample not containing galaxies with known masers ( mulchaey et al . this is in agreement with the fact that most h@xmath2o maser sources have been found in seyfert 2 systems . the good agreement between seyfert 1 and seyfert 2 galaxies ( mulchaey et al . 1994 ) with respect to this parameter , in spite of different viewing angles , lends further support to an approximately isotropic emission of both tracers . in order to further compare those two isotropic tracers , the fir luminosity is plotted against the extinction - corrected @xmath3 $ ] luminosity of our maser sample in the right panel of fig.2 . a correlation is found , logl@xmath75=([email protected])+ ( [email protected])logl@xmath77}$ ] with a spearman s rank correlation coefficient r=0.66 and a chance probability @xmath9 5@xmath9410@xmath95 . assuming the corrected @xmath3 $ ] line luminosity to be a good isotropic tracer ( bassani et al . 1999 ) , the strong correlation between fir luminosity and @xmath3 $ ] line luminosity may suggest that the fir luminosity is to some extent also an indicator of intrinsic nuclear activity , even though the fir flux may be contaminated by a spatially extended starburst component . ) of the iron line versus the thickness parameters l@xmath96/l@xmath97}$ ] ( t1 , left panel ) and l@xmath96/l@xmath75 ( t2 , right panel ) . square and pentacle symbols represent compton - thick and -thin sources as identified by modeling their x - ray spectra ( e.g. , zhang et al . the bold squares show those compton - thick sources , which have high ew ( k@xmath5 ) and low t values . these are located in the upper - left regions of both panels . three _ exceptional _ sources ngc2639 , mrk3 , and ngc4945 are marked . for ngc4945 , the lower limit of its [ oiii ] luminosity is taken . details are given in sect.3.2.,width=604 ] as mentioned above , nuclear absorbing columns can be obtained by analyzing x - ray spectra . these analyses are model dependent . comparing the observed 2 - 10kev x - ray emission from the nuclear region , absorbed by the obscuring material along the line - of - sight , with the intrinsic nuclear power provides another method for evaluating the absorbing column density . the ratio of the observed 2 - 10kev luminosity and the luminosity of the nuclear isotropic indicators was assumed to represent the optical thickness parameter , which allows us to diagnose the gas absorption ( bassani et al . 1999 ) . here the l@xmath74/l@xmath97}$ ] and l@xmath74/l@xmath75 ratios are used as optical thickness parameters . in addition , high ew ( @xmath6 ) values of the iron emission line are considered as a qualitative feature indicating a heavily obscured nucleus ( e.g. , maiolino & risaliti 2007 ) . combining the ew ( @xmath6 ) and the optical thickness parameters , we probe the circumnuclear environment of the maser sources . for the maser galaxies with available data ( ew ( @xmath6 ) , l@xmath74 , l@xmath97}$ ] and l@xmath75 , in total 31 agn maser sources ) , the ew ( k@xmath5 ) of the iron lines is plotted against the optical thickness parameters in fig.3 . the prominent feature of the figure is the existence of an anti - correlation , which is similar to that obtained for the seyfert 2 sample of bassani et al . ( 1999 ) , which is predominantly containing targets without known maser lines . in the left panel ( ew ( @xmath6 ) v.s . l@xmath74/l@xmath97}$ ] ) , a least - squares fit shows log(ew ( @xmath6))=([email protected])+ ( [email protected])logl@xmath74/l@xmath97}$ ] , with spearman s rank correlation coefficient r=-0.57 and a chance probability of p @xmath10.001 . the possibly compton - thick sources ( shown by squares in fig.3 ) , determined by modeling their x - ray spectra , cluster in the upper left region , the region with high ew ( k@xmath5 ) and low l@xmath74/l@xmath97}$ ] values . this area is marked by dashed lines . the approximate boundaries between compton - thick and -thin sources , the latter shown by pentacles , are log(l@xmath74/l@xmath97}$])@xmath10.25 ( i.e. , @xmath8@xmath12@xmath10}$ ] ) and logew ( @xmath6)@xmath12.5 ( i.e , ew@xmath1300ev ) . for comparison with seyfert 2 galaxies without detected h@xmath2o maser emission , the seyfert 2 sample of bassani et al . ( 1999 ) is used , excluding those objects with maser emission . the trend is the same for this sample ( here not shown ) . compton - thick sources fall again into the upper left region with high ew ( k@xmath5 ) and low l@xmath74/l@xmath97}$ ] values , while compton - thin sources are located in the lower right . the right panel of fig.3 shows a similarly clear anti - correlation between ew ( k@xmath5 ) and l@xmath74/l@xmath75 . linear fitting results in log(ew ( @xmath6))=(1.71 @xmath44 0.26 ) + ( -0.36@xmath44 0.08)logl@xmath74/l@xmath75 , with r=-0.61 and p @xmath1 2 @xmath9410@xmath98 . most of those sources with high ew ( k@xmath5 ) and low l@xmath74/l@xmath75 value ( upper - left region , in square symbols ) are compton - thick sources , while those sources with low ew ( k@xmath5 ) and high l@xmath74/l@xmath75 value ( in pentacles ) are compton - thin as determined from x - ray spectral fitting . the approximate boundaries between compton - thick and -thin environments are in this case logl@xmath74/l@xmath75@xmath1 - 2.75 ( i.e. , @xmath11@xmath1600@xmath8 ) and logew ( @xmath6)@xmath12.5 , i.e. again ew@xmath1300ev . we like to emphasize that these boundaries are not arbitrary . among the 20 compton - thick candidate sources from our sample , classified by conventional x - ray spectroscopy , we find 19 in each of our limited `` compton - thick '' regions , related either to l@xmath74/l@xmath97}$ ] or to l@xmath74/l@xmath75 . 18 of these sources are identical ( see also sect.4 ) . the results obtained so far could be affected by systematic errors in the measurements . assuming for the [ oiii ] values uncertainties of @xmath120% ( dahari & de robertis 1988 ) results in l@xmath74/l@xmath97}$ ] errors of order 0.08dex . obviously , this does not affect the robust fundamental trend in our diagnostic fig.3 . compton - thick sources are still placed in the upper left and compton - thin sources in the lower right . the anti - correlation between the ew ( @xmath6 ) and the optical thickness parameters are readily explained . with an increase of the absorbing column density , the x - ray luminosity will decrease so that l@xmath74/l@xmath97}$ ] and l@xmath74/l@xmath75 are reduced with respect to l@xmath97}$ ] and l@xmath75 . on the other hand , the ew ( @xmath6 ) values will increase , since these are measured against a reduced 6.4kev continuum level . with compton - thick galaxies being located in the upper left part of the panels in fig . 3 , we find for our h@xmath2o maser galaxies three criteria hinting at a compton - thick nuclear environment : ew ( @xmath6)@xmath7300ev , @xmath8@xmath92@xmath10}$ ] and @xmath11@xmath7600 @xmath8 . these are independent of the detailed shape of the x - ray spectrum . the search for new extragalactic h@xmath2o masers is ongoing and important with respect to several key aspects of modern astrophysics ( see , e.g. , sect.1 and braatz et al . 2009 ) . here we therefore investigate possible indicators of h@xmath2o maser emission . we analyze relationships between h@xmath2o maser luminosity and the iron line ew ( k@xmath5 ) , fir , and @xmath99}$ ] luminosity . ) ( logarithmic scale , in units of ev ) for seyfert 2s with detected h@xmath2o maser emission ( 19 sources , solid lines ) and non - masing seyfert 2s ( 34 sources from bassani et al . 1999 , dashed lines).,width=302 ] for 33 maser sources with available ew ( @xmath6 ) of the iron line in table1 , we obtain an ew ( k@xmath5 ) mean value of @xmath100ev . x - ray observations show that strong iron line emission is common in the spectra of agn hosting h@xmath2o masers . it is interesting to check if the 6.4kev line can be used as a criterion to search for agn masers . have seyfert 2s with detected maser emission higher ew ( k@xmath5 ) values than non - maser seyfert 2s ? our statistical results show that the iron line ew ( @xmath6 ) of masing seyfert 2 galaxies ( mean value @xmath101ev and median value @xmath1800ev for our 19 maser sources ) is higher than that of the non - maser seyfert 2 sample ( mean value @xmath102ev and median value @xmath1200ev for 34 sources from bassani et al . figure 4 shows the distributions of both samples . while the difference seems to be obvious at first sight , we should cautiously avoid a definite conclusion due to the large scatter , the still too small number of sources , and the incompleteness of the studied samples . potential differences in sensitivity have also to be addressed . for our maser host seyfert 2s , the ew ( @xmath6 ) values were taken almost exclusively ( except four sources ) from _ xmm - newton _ or _ chandra _ observations . the results for seyfert 2s without known masers ( bassani s sample ) come mostly from asca data of lower sensitivity , which might lead to an increase in the real average ( only sources with a strong iron line could be detected ) . this strengthens our result and amplifies the difference between seyfert 2s with and without detected 22ghz h@xmath2o maser . nevertheless , we consider our result as tentative . ) of the iron line ( logarithmic scale , in units of ev ) versus isotropic luminosity of the h@xmath2o maser emission ( logarithmic scale , in @xmath103 ) . center : [ oiii ] line luminosity against @xmath104 ; bottom : fir luminosity against @xmath104 , empty circles show galactic h@xmath2o masers from jaffe et al . ( 1981 ) and the line marks the correlation of @xmath104/@xmath105}$]@xmath110@xmath106 . squares and triangles represent agn - masers and masers in star formation regions respectively . disk - masers , as a subsample of agn - masers , for which maser emission is anticipated to be particularly well connected with indicators of the intrinsic nuclear power , are marked by crosses over squares.,width=377 ] in order to investigate possible correlations between h@xmath2o maser and iron emission lines , the ew ( @xmath6 ) of the fe line was plotted against the isotropic h@xmath2o luminosity in fig.5 ( upper panel ) . as already mentioned , h@xmath2o maser emission can be produced by collisional pumping in a dense molecular layer , which is heated by irradiated x - rays from the nucleus ( neufeld et al . strong fe @xmath6 emission is believed to be produced via x - ray reflection by the cold iron in the circumnuclear region ( e.g. , fabian et al . 2000 ) . in those cases where h@xmath2o maser and iron line emission are detected , the nuclear x - ray emission plays a key role . correlations between maser and iron line emission are therefore expected . however , our results show no apparent trend between the iron line ew ( k@xmath5 ) and the isotropic h@xmath2o maser luminosity , neither for the entire sample nor the two subsamples , agn- and star - forming masers ( squares and triangles , respectively , in fig.5 ) . in view of alternative h@xmath2o excitation mechanisms ( see , e.g. , lo 2005 ) , the subsample of possible disk - masers ( with detected high velocity maser features ) was analyzed separately ( see the crosses in fig.5 ) . however , even in this case no significant correlation can be found . the [ oiii ] and fir luminosities were also plotted against h@xmath2o maser luminosity in fig.5 ( central and bottom panel respectively ) . there is no significant correlation between @xmath107}$ ] and @xmath104 , although maser sources related to star formation seem to have lower [ oiii ] luminosities than agn masers . for fir versus maser luminosities , there appears to be a correlation , similar to that previously found by henkel et al . ( 2005 ) , castangia et al . ( 2008 ) , bennert et al . ( 2009 ) and surcis et al . the relation was first found for galactic star forming regions by jaffe et al . 1981 , i.e. , luminous h@xmath2o masers form in star formation regions with high fir luminosity . for comparison , values of galactic h@xmath2o masers are also plotted in fig.5 ( empty circles ) and the line shows the correlation of @xmath104/@xmath105}$]@xmath110@xmath106 from jaffe et al . apparently , there exists a correlation between fir and h@xmath2o maser luminosity over many orders of magnitude . when considering agn - masers only ( squares in fig.5 , lowest panel ) , the strongest masers appear to be overluminous with respect to the @xmath108-@xmath109 correlation . this is likely caused by the different properties of agn versus star - forming masers . compton - thick nuclei are known to contribute a significant fraction of the hard x - ray background . their density as a function of redshift is also a relevant parameter when studying the evolution of the universe . thus our newly defined criteria identifying such nuclei may be helpful when trying to reach this goal . based on our new approach introduced in sect.3.2 , 18 h@xmath2o maser sources in the upper left part of the panels in fig.3 ( see the solid squares ) can be considered to be compton - thick . these are ngc591 , ngc1068 , mrk1066 , ngc1386 , ngc2273 , ugc5101 , ngc2782 , ngc3079 , ic2560 , ngc3393 , arp299 , ngc5194 , mrk266 , ngc5347 , circinus , ngc5643 , ngc5728 and ngc6240 . comparing this with zhang et al . ( 2006 ) , there are five new sources , ngc591 , ugc5101 , ngc3072 , ic2560 , mrk266 . about 60% of the 31 agn masers turn out to be compton - thick . this is consistent with the result found by greenhill et al . ( 2008 ) . for maser sources associated with seyfert 2 nuclei , 60% ( 15/25 ) are compton - thick , which is , however , not significantly , higher than that of seyfert 2 objects without detected maser emission ( @xmath145% , 9/20 from risaliti et al . 1999 ; 19/42 from bassani et al . 1999 ) . our sample also contains a few _ exceptional _ sources , which show the limits of our selection criteria . for ngc4945 , its low [ oiii ] line flux places the source outside our compton - thick border ( ew ( @xmath6)-@xmath110/@xmath10}$ ] , see left panel in fig.3 ) . however , the [ oiii ] flux from risaliti et al . ( 1999 ) only gives a very stringent lower limit to the actual intrinsic [ oiii ] emission . the lack of a reliable [ oiii ] flux is thought to be due to high absorption in the edge - on galactic disk , instead of its intrinsic weakness . this is supported by its hard x - ray spectrum , which indicates that ngc4945 hosts one of the brightest agn in the hard x - ray range ( @xmath7100kev ) and can therefore be considered to contain a ` bona - fide ' compton - thick nucleus ( e.g. , guainazzi et al . similar to ngc4945 , mrk3 is also considered to be a ` bona - fide ' compton - thick seyfert 2 from its large brightness in the hard x - ray range ( e.g. , cappi et al . however , the source is found outside our limited ew ( @xmath6)-l@xmath74/l@xmath75 region ( right panel in fig.3 ) for its relatively low value of the infrared flux . more constraints are desirable to probe its circumnuclear environment . the disk - maser galaxy ngc2639 is located inside the required ew ( @xmath6)-l@xmath74/l@xmath75 region and outside the ew ( @xmath6)-l@xmath74/l@xmath97}$ ] area . however , large uncertainties in the x - ray results ( asca observations ) have to be noted . ngc2639 is a weak x - ray source with an asca count rate @xmath90.01countss@xmath86 . the asca data were analyzed by wilson et al . ( 1998 ) and terashima et al . ( 2002 ) and the source was considered to be compton - thin ( @xmath13@xmath15@xmath9410 @xmath14@xmath15 ) . low photon statistics lead to uncertainties in the fitting models and chandra observation are therefore needed to investigate its highly obscured nucleus . we conclude with some cautionary notes . first , uncertainties are involved when using the @xmath3\lambda 5007 $ ] line and fir luminosity as nuclear isotropic indicators . although assumed to be isotropic , the @xmath3 $ ] line luminosity might depend on the geometry of the system , for example on the opening angle of the torus and the inclination of the large scale disk . shielding effects may affect the ionizing radiation seen by the nlr . fir emission of h@xmath2o megamaser galaxies may mainly arise from the agn environment , but there is possible contamination from a starburst component and it is unclear how much it contributes . second , the ew ( k@xmath5 ) of the iron line can be affected by other factors , such as the geometry of the accretion disk and the inclination angle at which the reflecting surface is viewed ( e.g. , fabian et al . 2000 , bianchi et al . a high ew ( k@xmath5 ) of the iron line can also appear if the radiation is anisotropic or if there is a time lag between a drop in the continuum and the line emission ( bassani et al . 1999 ) . with future advanced x - ray telescopes , sensitive observations of more h@xmath2o maser host galaxies at higher energies ( above 10kev ) will provide important complementary information , further constraining nuclear column density . in this paper , multi - wavelength data from the complete sample of galaxies ( @xmath111@xmath7100kpc ) so far reported to host 22ghz h@xmath2o masers are analyzed , including the equivalent width ( ew ) of the iron @xmath6 line , the @xmath3\lambda 5007 $ ] line , and the x - ray and fir emission . the latter two are considered to be the isotropic tracers of intrinsic nuclear power . the observed nuclear x - ray luminosity , compared with the luminosities of these two isotropic tracers ( our optical thickness parameters ) , can be used as a measure of the circumnuclear absorption . the ew ( @xmath6 ) and the optical thickness parameters ( @xmath110/@xmath112}$ ] , @xmath110/@xmath108 ) are combined here to probe the obscuration of maser host agn . the main results are summarized below : \(1 ) our statistical analysis shows obvious anti - correlations between the ew ( @xmath6 ) of the fe emission line and the two optical thickness parameters . without requiring a full x - ray spectrum , compton - thick nuclear environments can be identified with these parameters and are found to be characterized approximately by ew ( @xmath6)@xmath7300ev , @xmath8@xmath92@xmath113}$ ] and @xmath11@xmath7600@xmath8 ; \(2 ) 18 h@xmath2o maser sources matching these criteria are identified to be compton - thick . a comparison with zhang et al . ( 2006 ) shows , that among these there are five newly identified h@xmath2o maser galaxies which are compton - thick , i.e. , ngc591 , ugc5101 , ngc3072 , ic2560 and mrk266 . masers associated with seyfert 2 nuclei may be more likely compton - thick ( 60% ) than seyfert 2s without detected maser emission ( @xmath145% ) . however , this still has to be confirmed with larger samples ; \(3 ) in an attempt to guide future h@xmath2o maser surveys , new ways to find extragalactic h@xmath2o sources are also explored . maser sources may show larger ew ( k@xmath5 ) values than non - maser seyfert 2s , which , however , also needs further support . no significant correlations have been found between ew ( k@xmath5 ) , @xmath112}$ ] and @xmath104 . there appears a linear correlation between @xmath108 and @xmath104 , which is consistent with the correlation found for galactic h@xmath2o masers . however , the strongest h@xmath2o masers appear overluminous with respect to their @xmath108 . this may be related to their different origin when compared with masers associated with sites of massive star formation well outside the nuclear region of their parent galaxy . we wish to thank the anonymous referee for many detailed and constructive comments as well as p. castangia for critically reading the manuscript . this work is supported partly by the national natural scientific foundation of china ( 10633010 ) and guangdong province natural science foundation ( 8451009101001047 ) . we made use of the nasa / ipac extragalactic database ( ned ) , high - energy astrophysics science and research center ( heasarc ) and the nasa astrophysics data system bibliographic services ( ads ) . 2 alonso - herrero , a. , ward , m. j. & kotilainen , j. k. 1997 , mnras , 288 , 947 awaki , h. , ueno , s. , taniguchi , y. & weaver , k. a. 2000 , apj , 542 , 175 ballo , l. , braito , v. , della ceca , r. , maraschi , l. , tavecchio , f. & dadina , m. 2004 , apj , 600 , 634 bassani , l. , dadina , m. , maiolino , r. , salvati , m. , risaliti , g. , della ceca , r. , matt , g. & zamorani , g. 1999 , apjs , 121 , 473 beckmann , v. , gehrels , n. , favre , p. , walter , r. , courvoisier , t. j .- l . , petrucci , p .- o . & malzac , j. 2004 , apj , 614 , 641 bennert , n. , barvanis , r. , henkel , c. & antonucci , r. 2009 , apj , 695 , 276 bianchi , s. , balestra , i. , matt , g. , guainazzi , m. & perola , g. c. 2003 , a&a , 402 , 141 bianchi , s. , miniutti , g. , fabian , a. c. , & iwasawa , k. 2005 , mnras , 360 , 380 braatz , j. a. , wilson , a. s. , & henkel , c. 1997 , apjs , 110 , 321 braatz , j. a. & gugliucci , n. e. 2008 , apj , 678 , 96 braatz , j. a. , condon , j. j. , henkel , c. , lo , k .- y . & reid , m.j . 2009 , astro2010 , 23 cappi , m. et al . 1999 , a&a , 344 , 857 cappi , m. et al . 2006 , a&a , 446 , 459 castangia , p. , tarchi , a. , henkel , c. & menten , k. m. 2008 , a&a , 479 , 111 dahari , o. & de robertis , m. m. 1988 , apjs , 67 , 249 darling , j. , brogan , c. & johnson , k. 2008 , apjs 685 , 39 della ceca , r. et al . 2002 , apj , 581 , 9 diaz , a. i. , prieto , m. a. & wamsteker , w. 1988 , a&a , 195 , 5 done , c. , madejski , g. m. , zycki , p. t. , & greenhill , l. j. 2003 , apj , 588 , 763 fabian , a. c. , iwasawa , k. , reynolds , c. s. & young , a. j. 2000 , pasp , 112 , 1145 ghisellini , g. , haardt , f. & matt , g. 1994 , mnras , 267 743 greenhill , l. j. , tilak , a. & madejski , g. 2008 , apj , 686 , l13 guainazzi , m. , matt , g. , brandt , w. n. , antonelli , l. a. , barr , p. & bassani , l. 2000a , a&a , 356 , 463 guainazzi , m. , matt , g. , fiore , f. , & perola , g. c. , 2002 , a&a , 388 , 787 guainazzi , m. , rodriguez - pascual , p. , fabian , a. c. , iwasawa , k. & matt , g. 2004 , mnras , 355 , 297 guainazzi , m. , fabian , a. c. , iwasawa , k. , matt , g. & fiore , f. 2005a , mnras , 356 , 295 guainazzi , m. , matt , g. & perola , g. c. 2005b , a&a , 444 , 119 heckman , t. m. , balick , b. & crane , p. c. 1980 , a&as , 40 , 28 heckman , t. m. , ptak , a. , hornschemeier , a. & kauffmann , g. , 2005 , apj , 634 , 161 henkel , c. , peck , a. b. , tarchi , a. , nagar , n. m. , braatz , j. a. , castangia , p. & moscadelli , l. 2005 , a&a , 436 , 75 ho , l. c. , filippenko , a. v. & sargent , w. l. w. 1997 , apjs , 112 , 315 imanishi , m. , terashima , y. , anabuki , & n. , nakagawa , t. 2003 , apj , 596 , 167 impellizzeri , c. m. v. , mckean , j. p. , castangia , p. , roy , a. l. , henkel , c. , brunthaler , a. & wucknitz , o. 2008 , nature , 456 , 927 iwasawa , k. , maloney , p. r. & fabian , a. c. 2002 , mnras , 336 , 71 iyomoto , n. , fukazawa , y. , nakai , n. & ishihara , y. 2001 , apj , 561 , 69 jaffe , d. t. , guesten , r. & downes , d. 1981 , apj , 250 , 621 jenkins , l. f. , roberts , t. p. , ward , m. j. & zezas , a. 2004 , mnras , 352 , 1335 kim , d .- c . , sanders , d. b. , veilleux , s. , mazzarella , j. m. & soifer , b. t 1995 , apj , 98 , 129 kondratko , p.t . , greenhill , l. j. & moran , j. m. 2006 , apj , 652 , 136 kraft , r. p. , hardcastle , m. j. , worrall , d. m. & murray s. s. 2005 , apj , 622 , 149 lamastra , a. , bianchi , s. , matt , g. , perola , g. c. , barcons , x. & carrera , f. j. 2009 , a&a , 504 , 73 leahy , d. a. , & creighton , j. 1993 , mnras , 263 , 314 levenson , n. a. , heckman , t. m. , krolik , j. h. , krolik , j. h. , weaver , k. a. & zycki , p. t. 2006 , apj , 648 , 111 lightman , a. p. & white , t. r. 1988 , apj , 335 , 57 lo , k. y. 2005 , ara&a , 43 , 625 lumsden , s. l. & alexander , d. m. 2001 , mnras , 328 , l32 madejski , g. , done , c. , zycki , p. t. & greenhill , l. 2006 , apj , 636 , 75 maiolino , r. & rieke , g. h. , 1995 , apj , 454 , 95 maiolino , r. & risaliti , g. 2007 , in asp conf . , 373 , the central engine of activ e galactic nuclei , ed . l. c. ho & j .- m.wang ( sanfrancisco , ca : asp ) , 447 matt , g. , guainazzi , m. , perola , g. c. , fiore , f. , nicastro , f. , cappi , m. & piro , l. 2001 , a&a , 377 , 31 matt , g. 1997 , mem . ital . , 68 , 127 matt , g. , bianchi , s. , guainazzi , m. & molendi , s. , 2004 , a&a , 414 , 155 matsumoto , c. , nava , a. , maddox , l. a. , leighly , k. m. , grupe , d. , awaki , h. & ueno , s. 2004 , apj , 617 , 930 mulchaey , m. et al . 1994 , apj , 436 , 586 netzer , h. , mainieri , v. , rosati , p. & trakhtenbrot , b. 2006 , a&a , 453 , 525 neufeld , d. a. , maloney , p. r. , & conger , s. 1994 , apj , 436l , 127 ohno , m. , fukazawa , y. & iyomoto , n. 2004 , pasj , 56 , 425 panessa , f. , bassani , l. , cappi , m. , dadina , m. , barcons , x. , carrera , f. j. , ho , l. c. & iwasawa , k. 2006 , a&a , 455 , 173 polletta , m. , bassani , l. , malaguti , g. , palumbo , g. g. c. & caroli , e. 1996 , apj , 106 , 399 ptak , a. , heckman , t. , levenson , n. a. , weaver , k. & strickland , d. 2003 , apj , 592 , 782 risaliti , g. , maiolino , r. & salvati , m. 1999 , apj , 522 , 157 risaliti , g. , gilli , r. , maiolino , r. & salvati m. , 2000 , a&a , 357 , 13 shu , x. w. , wang , j. x. , jiang , p. , fan , l. l. & wang , t , g . , 2007 , apj , 657 , 167 smith , d. a. & done , c. , 1996 , mnras , 280 , 355 smith , d. a. & wilson , a. s. , 2001 , apj , 557 , 180 spergel , d. n. et al . 2003 , apjs , 148 , 175 storchi - bergmann , t. & pastoriza , m. g. 1989 , apj , 347 , 19 storchi - bergmann , t. , kinney , a. l. & challis , p. 1995 , apjs , 98 , 10 su , j. b. , zhang , j. s. & fan , j.h . , . j. astron . astrophys . , 8 , 547 surcis , n. , tarchi , a. , henkel , c. , ott , j. , lovell , j. & castangia , p. 2009 , a&a , 502 , 529 terlevich , r. , melnik , j. , masegosa , j. , moles , m. & copetti , m. v. f. 1991 , a&as , 91 , 285 terashima , y. , iyomoto , n. , ho , l. c. & ptak , a. f. , 2002 , apjs , 139 , 1 tran , h. d. 2001 , apj , 554 , l19 turner , t. j. , george , i. m. , nandra , k. & mushotzky , r. f. 1997 , apjs , 113 , 23 weaver , k. a. , gelbord , j. & yaqoob , t. 2001 , apj , 550 , 261 whittle , m. 1992 , apjs , 79 , 49 wilson , a. s. et al . 1998 , apj , 505 , 587 wouterloot , j. g. a. & walmsley , c. m. 1986 , a&a , 168 , 237 zhang , j. s. , henkel c. , kadler , m. , greenhill , l. j. , nagar , n. , wilson , a. s. & braatz , j. a. 2006 , a&a , 450 , 933 zhang , j. s. & fan , j. h. 2009 , sci chin . g , 6 , 960

polarimetry is extensively used as a tool to trace the interstellar magnetic field projected on the plane of sky . moreover , it is also possible to estimate the magnetic field intensity from polarimetric maps based on the chandrasekhar - fermi method . in this work , we present results for turbulent , isothermal , 3-d simulations of sub / supersonic and sub / super - alfvenic cases . with the cubes , assuming perfect grain alignment , we created synthetic polarimetric maps for different orientations of the mean magnetic field with respect to the line of sight ( los ) . we show that the dispersion of the polarization angle depends on the angle of the mean magnetic field regarding the los and on the alfvenic mach number . however , the second order structure function of the polarization angle follows the relation , being dependent exclusively on the alfvenic mach number . the results show an anti - correlation between the polarization degree and the column density , with exponent , in agreement with observations , which is explained by the increase in the dispersion of the polarization angle along the los within denser regions . however , this effect was observed exclusively on supersonic , but sub - alfvenic , simulations . for the super - alfvenic , and the subsonic model , the polarization degree showed to be intependent on the column density . our major quantitative result is a generalized equation for the cf method , which allowed us to determine the magnetic field strength from the polarization maps with errors . we also account for the role of observational resolution on the cf method .

0801.0279

it is believed that giant molecular clouds in the interstellar medium ( ism ) are threaded by large scale magnetic fields @xcite . however , it is still not completely clear what is the role of the magnetic field in the dynamics of the ism and what is its effect on the star formation process . also , the ratio of the magnetic and turbulent energy in these environments is a subject of controversy @xcite . magnetic fields can influence the injection and evolution of turbulence bringing more complexity to this issue ( see lazarian & cho [ 2004 ] for review ) . as an example , simulations have shown that strongly magnetized turbulent media develop structures with lower density contrasts when compared to pure hydrodynamic turbulence @xcite . observationally , different techniques can be used to measure the ism magnetic field and determine its intensity and topology . zeeman splitting of spectral lines provides a direct and precise derivation of the magnetic field component along the line of sight ( los ) , mainly for clouds presenting strong spectral lines @xcite . however , it can not be applied to the clouds where the line intensities are too weak . typically , when observed , zeeman measurements of molecular clouds give @xmath4 g , and suggest the correlation with density @xmath5 , which is consistent to the expected relation for compressions of magnetic fields frozen into plasma . spectral line broadening show that molecular clouds present supersonic , but critically alfvenic motions @xcite . this fact shows that the turbulent motions may be excited by mhd modes instead of being purely hydrodynamical . one of the most readily available methods of studying the perpendicular component of the magnetic field is based on the polarization of dust thermal emissions at infrared and submillimetric wavelengths @xcite . the alignment of grains in respect to the magnetic field is a hot research topic ( see lazarian [ 2007 ] for review ) . radiative torques ( rats ) can promote alignment of irregular dust particles , resulting in different intensities for polarized radiation parallel and perpendicular to the local magnetic field @xcite . grains with long axis aligned perpendicular to the magnetic field induce polarization parallel to the magnetic field for transmitted star light , and perpendicular to the field lines for the dust emission . @xcite showed that rats are very efficient on the grain alignment process in molecular clouds , even for the very dense regions ( up to @xmath6 ) . they also showed that the alignment efficiency strongly depends on the grain size , being practically perfect for large grains ( @xmath7 @xmath8 m ) . more detailed studies of the rats efficiency by @xcite confirmed this claim . therefore , for a range of @xmath9 it is acceptable to assume that the grains are well - aligned . for a given polarization map of an observed region , the mean polarization angle indicates the orientation of the large scale magnetic field . on the other hand the polarization dispersion gives clues on the value of the turbulent energy . this , as a consequence , can be used to determine the magnetic field component along the plane of sky . @xcite introduced a method ( cf method hereafter ) for estimating the ism magnetic fields based on the dispersions of the polarization angle and gas velocity . simply , it is assumed that the magnetic field perturbations are alfvenic and that the rms velocity is isotropic . a promising approach to test this method is to create two - dimensional ( plane of sky ) synthetic maps from numerically simulated cubes . @xcite performed 3d - mhd simulations , with @xmath10 resolution , in order to obtain polarization maps and study the validity of the cf method on the estimation of the magnetic field component along the plane of sky . they showed that the cf method gives reasonable results for highly magnetized media , in which the dispersion of the polarization angle is @xmath11 . however , they did not present any other statistical analysis or predictions that could be useful for the determination of the ism magnetic field from observations . @xcite presented a complementary work , with a more detailed analysis regarding the limited observational resolution on the cf method , and presented a modified equation to account for the differences obtained previously . they concluded that lower observational resolution leads to an overestimation of the magnetic field from the cf equation . they also showed a good agreement between the cf technique and the expected magnetic field of their simulations , except for the weak field models . polarization maps from numerical simulations can also be used in the study of the correlation between the polarization degree and the total emission intensity ( or dust column density ) . observationally , the polarization degree in dense molecular clouds decreases with the total intensity as @xmath12 , with @xmath13 @xcite . @xcite studied the role of turbulent cells in the @xmath14 relation using supersonic and super - alfvenic self - graviting mhd simulations . they found a decrease of polarization degree with total dust emission within gravitational cores , in agreement with observations , if grains are assumed to be unaligned for @xmath15 . when the alignment was assumed to be independent on @xmath9 , the anti - correlation was not observed . recently , @xcite extended this work and refined the calculation of polarization degree introducing the radiative transfer properly . in that work , the decrease in the alignment efficiency arises without any _ ad hoc _ assumption . the alignment efficiency decreases as the radiative torques become less important in the denser regions . however , it is still not clear the role of the magnetic field topology and the presence of multiple cores intercepted by the line of sight on the decrease of polarization degree . in this work we attempt to extend the previously cited studies improving and applying the cf method for different situations . for that , we studied both sub and super - alfvenic models , to study the role of the magnetic field topology in the observed polarization maps . we simulate different observational resolutions in the calculations of polarization maps and provide combined statistical analysis for both dust absorption and emission maps . we also present statistics based methods to characterize the turbulence and magnetic properties from polarization maps . we performed numerical simulations of magnetized turbulent plasma with higher resolution , which are described in sec . 2 . from the data , we computed observable " polarization maps , as shown in sec . 3 . we then present the statistics and spatial distributions of angle and polarization degree for different models in sec . 4 . in sec . 5 we propose the generalized equation for the cf method and compare it with the expected values to study its validity . in sec . 6 , we discuss the improved procedures of polarization vector statistics , which allow observers to characterize the mean and fluctuating magnetic field of the cloud . we also discuss the applicability of our approach for polarized molecular and atomic lines , and compare the results with previous works . ou summary is provided in sec . the simulations were performed solving the set of ideal mhd equations , in conservative form , as follows : @xmath16 @xmath17 = \rho{\bf f},\ ] ] @xmath18 with @xmath19 , where @xmath20 , @xmath21 and @xmath22 are the plasma density , velocity and pressure , respectively , @xmath23 is the magnetic field and @xmath24 represents the external acceleration source , responsible for the turbulence injection . for molecular clouds , we may assume that the ratio of dynamical to radiative timescales is very large . under this assumption , the set of equations is closed by an isothermal equation of state @xmath25 , where @xmath26 is the speed of sound . the equations are solved using a second - order - accurate and non - oscillatory scheme , with periodic boundaries , as described in @xcite . initially , we set the intensity of the x - directed magnetic field @xmath27 and the gas thermal pressure @xmath22 . this allows us to obtain sub - alfvenic or super - alfvenic , and subsonic or supersonic models . the turbulent energy is injected using a random solenoidal function for @xmath24 in fourier space . this , in order to minimize the influence of the forcing in the formation of density structures . we inject energy at scales @xmath28 , where @xmath29 is the box size and @xmath30 is the eddy size of the injection scale . the rms velocity @xmath31 is kept close to unity , therefore @xmath21 and the alfvn speed @xmath32 will be measured in terms of the rms @xmath31 . also , the time @xmath33 is measured in terms of the dynamical timescale of the largest turbulent eddy ( @xmath34 ) . we performed four computationally extensive 3d mhd simulations , using high resolution ( @xmath35 ) , for different initial conditions , as shown in table 1 . we simulated the clouds up to @xmath36 , i.e. 5 times longer than the dynamical timescale , to ensure a full development of the turbulent cascade . we obtained one subsonic and three supersonic models . one of the supersonic models is also super - alfvenic . each data cube contains information about parameterized density , velocity and magnetic field . as noted from eqs . ( 1 ) and ( 2 ) , the simulations are non self - gravitating and , for this reason , the results are scale - independent . regarding the gas distribution in each model we found an increasingly contrast for increasing sonic mach number , independent on the alfvenic mach number . this result was also obtained , and studied with more details , in @xcite . subsonic turbulence show a gaussian distribution of densities , while the increased number and strnght of shocks in supersonic cases create more smaller and denser structures . in these cases , the density contrast may be increased by a factor of 100 - 10000 compared to the subsonic case . the magnetic field topology , on the other hand , depends on the alfvenic mach number . sub - alfvenic models show a strong uniformity of the field lines , while the super - alfvenic case shows a very complex structure . both effects , the density contrast and the magnetic field topology , may play a role on the polarimetric maps , as shown further in the paper . here we focus on the determination of observable quantities from our synthetic maps . from the density and magnetic field cubes we created plane of sky " maps for column density and the linear polarization vectors . from the velocity field cubes it was possible to obtain the rms velocity , which is necessary to test the cf method . to create the polarization maps we assumed that the radiation is originated exclusively by thermal emission from perfectly aligned grains . the dust abundance is supposed to be linearly proportional to the gas density and , in this case , the total intensity may be assumed to be proportional to the column density . we also assumed that all dust particles emit at the same temperature . in this work we assume the dust polarization to be completely efficient ( @xmath37 ) , and perfect grain alignment . under these assumptions , the local angle of alignment ( @xmath38 ) is determined by the local magnetic field projected into the plane of sky , and the linear polarization stokes parameters @xmath39 and @xmath40 are given by : @xmath41 where @xmath20 is the local density and @xmath42 is the inclination of the local magnetic field and the line of sight . we then obtain the integrated @xmath39 and @xmath40 , as well as the column density , along the los . notice that , for the given equations the total intensity ( stokes @xmath43 ) is assumed to be simply proportional to the column density . the polarization degree is calculated from @xmath44 and the polarization angle @xmath45 . previous works ( e.g. @xcite ) obtained the polarization maps integrating all cells along the line of sight , in spite of the fact that the local density may be too low to present an observable dust component . in reality , dense dust clouds are surrounded by warmer and rarefied regions , in which the dust component is negligible . to simulate this effect , we neglect any contribution for cells with density lower than an specific threshold , which depends on the model . the threshold for each model is arbitrarily chosen to keep the minimum column density , i.e. intensity , as 0.3 of its maximum . in figs . 1 and 2 , we show the obtained maps of column density and the polarization vectors for model 3 . we used the two extreme orientations of the magnetic field regarding the los ( 0 and @xmath46 ) . clearly , as shown in fig . 1 , the external magnetic field oriented in the x - direction dominates the polarization process . fluctuations on the polarization angle are seen within the condensations , which are dense enough to distort the magnetic field lines . in fig . 2 , since the magnetic field is oriented along the los , the polarization is due exclusively by the random field components . the dispersion of the polarization angle is large and the polarization degree is , in average , lower than obtained in fig . 1 . similar results were obtained for models 1 and 2 . even though presenting different magnetic to gas pressure fraction , all the sub - alfvenic models present similar maps . in fig . 3 , we show the column density and polarization maps of the super - alfvenic case ( model 4 ) , assuming the magnetic field perpendicular to the los . here , the kinetic energy is larger than the magnetic pressure . as a consequence the gas easily tangles the magnetic field lines . the angular dispersion is larger and the polarization degree is smaller when compared to the sub - alfvenic case . for the super - alfvenic case , the orientation of the magnetic field regarding the los is irrelevant to the polarization maps . the histograms of polarization angles are shown in fig . 4 . in the upper panel we show the histograms for the sub - alfvenic ( models 1 , 2 and 3 ) and the super - alfvenic ( model 4 ) cases , with the mean magnetic field lines perpendicular to the los . the polarization angles present very similar distributions and almost equal dispersion for the sub - alfvenic cases . this happens mainly because they do not depend on the density structures , but on the magnetic topology . strongly magnetized turbulence creates more filamentary and smoother density structures ( i.e. low density contrast ) if compared to weakly magnetized models and , most importantly the magnetic field lines are not highly perturbed . for model 4 , the distribution is practically homogeneous , which means that the polarization is randomly oriented in the plane of sky . it occurs because the turbulent / kinetic pressure is dominant and the gas is able to easily distort the magnetic field lines . in the bottom panel of fig . 4 we show the polarization angle histograms obtained for model 3 but for different orientations of the magnetic field . the dispersion of the polarization angle is very similar for inclination angles @xmath47 , and increases for larger inclinations . it may be understand if noted that the projected magnetic field @xmath48 is of order of the random component @xmath49 . it shows that the dominant parameter that differ the distributions of @xmath50 is the uniform magnetic field projected in the plane of sky , and not the intensity of the global magnetic field . furthermore , the distribution pf polarization angles may be used to determine if a sample of clouds in a given region of the ism present sub or super - alfvenic turbulence . since it is very unlikely to have all clouds with mean magnetic field pointed towards the observer , one non - homogeneous distribution of @xmath50 would reveals a sub - alfvenic turbulence . in turbulence studies it is useful to calculate the density and velocity power spectra . it allows a better understanding and characterization of the energy cascade process and the correlation between different scales . in fig . 5 we present the power spectra of the polarization angle for model 3 with different inclination angles @xmath51 ( _ upper panel _ ) , model 4 with different @xmath51 ( _ middle panel _ ) and for the different models ( _ bottom panel _ ) . the spectra were obtained for sizes smaller than @xmath52 to eliminate contaminations from the forcing at large scales . as the inclination angle @xmath51 increases , more power is found in smaller scales ( larger @xmath53 ) and the spectrum becomes flatter . interestingly , spectrum slopes could be used for the determination of the magnetic field inclination . however , the same trend is found by increasing the sonic and alfvenic mach numbers , as seen in the bottom panel . the degeneracy between alfvenic mach number and the magnetic field inclination makes it impossible to correctly estimate @xmath54 ( or the mass - to - flux ratio ) of a given cloud from polarimetric map spectra unless additional information regarding the orientation of the magnetic field is given . possibly , a different statistical analysis should be used to bypass this problem . the study of the decorrelation between different scales may show more sensitivity to @xmath51 and @xmath55 than spectra , as shown below . the second order structure function ( sf ) of the polarization angles is defined as the average of the squared difference between the polarization angle measured at 2 points separated by a distance @xmath30 : @xmath56 the structure functions calculated for the different models are shown in fig . 6 . in the upper panel we present the sfs obtained for model 3 with different values of @xmath51 . as expected , all curves present a positive slope showing the increase in the difference of polarization angle for distant points . however , the small scales part of the sf presents a plateau extending up to @xmath57 pix . this range corresponds to the dissipation region and may also be related to the smallest turbulent cells . as shown in the polarization maps , we should expect an increase in the values of sf as @xmath51 increases because of the increase in the dispersion of polarization angles . from fig . 6 ( upper panel ) it is noticeable the increasing profiles of the sfs . however , surprisingly , the obtained slopes are very similar . for an assumed relation @xmath0 we obtained @xmath58 , for @xmath59 pix , independently on @xmath51 . in the middle panel we show the structure functions for model 4 with different @xmath51 . here , the sfs are almost equal , as noticed by the spectra . we also obtained a very similar slope for the different values of @xmath51 . it shows that the slope is independent on @xmath51 . in the bottom panel we show the sfs calculated for the different models with @xmath60 . it is noticeable the increase in the sf for higher mach numbers . however , the slopes are notably different . the maximum slope is @xmath61 , 0.8 , 0.5 and 0.3 for models 1 , 2 , 3 and 4 , respectively . observationally , the molecular cloud m17 shows @xmath58 up to @xmath62pc @xcite , which would be in agreement with a cloud excited by supersonic and sub(or critically)-alfvenic turbulence . from these results we could possibly indicate that sfs of polarization maps may be used for the determination of the magnetic field intensity . associated to other analysis , as spectra and polarization angle histograms , it would be possible to determine also the magnetic field inclination regarding the los . however , a more detailed study , using more numerical simulations considering a large range of parameters , is needed to support these results . as discussed above , the structure functions of polarization angles present a plateau at small scales . possibly , if we had pencil beam " observations , its range could reflect the size of the smallest turbulent cells ( @xmath63 ) . infinite resolution observations would measure a non - zero ( due to the neighboring eddies discontinuity ) and flat sf up to a scale @xmath64 . for @xmath65 the sf would present a positive slope . however , could the flat part of the sf also be dependent on the observational resolution , instead of the turbulent structures exclusively ? to study this effect , we calculated the polarization maps considering different observational resolutions . from the original @xmath66 polarization maps , which is assumed to be the real cloud , we create the observed " maps considering beam sizes @xmath67 , @xmath68 and @xmath69 pix . the polarization angle is obtained from the @xmath39 and @xmath40 integrated over the neighboring cells , and the sfs for each resolution were obtained from eq . the obtained sfs are shown in fig . 7 , for @xmath70 perpendicular ( upper panel ) and parallel to the los ( bottom panel ) . in both plots , we show that the sf is dependent on the observational resolution . as a common result , lower observational resolution results in higher sf at small scales and lower sf at large scales . this is a result of the averaging of the polarization map in the beam size . in the small boxes we show the logarithm of the structure function , and the slopes for the lowest and highest resolutions , 0.35 and 0.50 , respectively therefore , the observational resolution may influence the determination of the magnetic field from the slopes of sfs , and this method should be used carefully . an interesting feature is the extended plateau at small scales . the sf present a plateau up to a limiting scale , and a positive slope at larger scales . observationally , similar profiles were obtained by @xcite , which shows that the current results may be implemented by high resolution observations . the limiting scale is approximately the beam size @xmath71 . therefore , the obtained results show that the smallest turbulent cells are only detectable if the condition @xmath72 is satisfied in the observations . @xcite estimated @xmath73 for mhd turbulence , considering the viscous and ion - neutral collision damping , as : @xmath74 where @xmath75 is the mean - free - path for ion - neutral collisions , @xmath76 is the eddy velocity at the injection scale @xmath29 and @xmath77 is the fraction of neutral atoms . considering @xmath78 cm , @xmath79 , @xmath80 , @xmath81 and @xmath82pc , we obtain @xmath83pc . this represents @xmath84 arcsec for the orion molecular cloud . considering the instruments available , the required resolution is a little larger than obtained using sharp and sofia ( @xmath85 arcsec ) , but could be reached by the sub - millimeter array ( sma ) ( @xmath86 arcsec ) . observations at high resolutions could then also help us to better understand the process occurring at scales smaller than @xmath73 . however , the outcome of observations is yet uncertain since @xcite showed that the magnetic field structures may be complex even below this scale . another interesting analysis is related to the correlation between the column density and the polarization degree . as noticeable from figs . 1 , 2 and 3 , the polarization degree is smaller within high column density regions for all models . this result is supported by observations , and was detected for several objects @xcite . typically , the polarization degree follows the relation @xmath87 , where @xmath88 @xcite and @xmath43 is the total intensity . a possible explanation could be the change on dust size and geometry at denser regions . in this case they would be less effectively aligned in respect to the magnetic field @xcite . another possibility could be an increase on the thermal collisions with gas and other dust particles in high density clumps @xcite . however , our numerical simulations does not take into account such processes and , therefore , those could not be influencing our results . in fig . 8 we show the correlation between the polarization degree and the column density for model 3 , considering orientations of magnetic field regarding the los ( _ upper panel _ ) . also , we show the correlations for the different models , assuming the magnetic field at the plane of the sky ( _ bottom panel _ ) . for all angles , the polarization of high column densities tend to decrease to the minimum value ( @xmath89 ) , which is the value obtained for the case of purely random magnetic component ( @xmath90 ) . this minimum polarization degree should be zero for homogeneous density and random magnetic field . in inhomogeneous media it depends on the number of dense structures intercepted by the line of sight . the major contribution for the polarized emission comes from dense clumps , which are few along the los . this poor statistics results in a non - zero polarization degree . for the super - alfvenic case , the contrast in density is larger as well as the number of high density structures . in this case , the polarization degree is smaller , as seen in fig . we obtain , as best fit for the plots , a correlation exponent @xmath91 . @xcite , studying the radiative torque efficiency in the grain alignment process , found larger values for @xmath92 . if grain alignment is implemented properly , the value of @xmath92 should increase ( see cho & lazarian [ 2005 ] ) , but it is out of the scope of this work . from the bottom panel , it is noticeable that for sub - sonic turbulence the polarization degree is large even for the higher column densities . it occurs because in the sub - sonic models the contrast in density is small and the simulated domain is more homogeneous . also , the number of dense clumps , which are able to tangle the field lines , is reduced in the sub - sonic case . on the contrary , for the super - alfvenic case we obtain a correlation similar to @xmath90 , i.e. purely random magnetic field components . these results indicate that the decrease in the polarization degree observed in molecular clouds may be partially due to an increase in the random to uniform ratio of the magnetic field components . the higher density flows are able to easily tangle the magnetic field lines . on the other hand , in the low density streams outside the clumps the magnetic field tends to be more uniform and the polarization degree higher . @xcite proposed a method for estimating the ism magnetic fields based on the dispersion of polarization angles and the rms velocity . basically , assuming that the magnetic field perturbations are alfvenic , i.e. @xmath93 , and that the rms velocity is isotropic we have : @xmath94 where @xmath95 is the observational rms velocity along the los . using the small angle approximation @xmath96 , it reduces to : @xmath97 where @xmath50 is measured in radians and @xmath98 is a correction factor ( @xmath84 ) @xcite , which depends on medium inhomogeneities , anisotropies on velocity perturbations , observational resolution and differential averaging along the los . @xcite noticed from their numerical simulations that the cf method ( eq . [ 7 ] ) was a good approximation for the cases where @xmath99 , i.e. when the uniform component of the magnetic field is much larger than the random components . this conclusion is expected from eq . ( 7 ) since it is applicable only for small values of @xmath100 , due to the angular approximation . if one wants to expand the applicability of the cf method for cases where the random component of the magnetic field is comparable to the uniform component , or for larger inclination angles , it is necessary to take into account two corrections in eq . firstly , we must introduce the total magnetic field projected in the plane of sky @xmath101 , where @xmath102 represents the mean field component projected on the plane of sky . we assume here , for the sake of simplicity , that @xmath49 is isotropic . @xcite substituted @xmath100 in the cf equation by @xmath103 , where @xmath104 was calculated locally , to provide a correction for the small agle approximation . however , they showed that this case lead to an underestimation ( by a factor of 100 ) of the magnetic field in super - alfvenic cases . it occurred because , as @xmath105 it gives @xmath106 . to avoid this , they introduced a correction , which was the geometric average of the standard @xmath107 and the modified value @xmath108 . here , we implement the correction of the small angle approximation in a simpler way . we assume that the @xmath109 is a global relation and , in this case , we may firstly obtain the dispersion of @xmath50 and then calculate its tangent . substituting @xmath100 in eq . ( 7 ) by @xmath110 , we obtain the modified cf equation : @xmath111 which is a generalized form of eq . ( 7 ) . as an example , if polarization maps give @xmath112 , eq . ( 8) gives @xmath113 and @xmath114 . this is expected for @xmath115 or @xmath116 . here we assume the obtained cubes as the real clouds subject to observational studies . in the previous sections we presented the expected results considering infinite observational resolution . however , observational data analysis may be biased by the limited instrumental resolution . therefore , we must understand its role on the statistical analysis of the measured parameters . we applied eq . ( 8) to our simulated clouds , taking into account the effects of finite resolution . here , we intended to determine the role of the resolution on the determination of the magnetic field strength from the cf method . we calculated the average of the density weighted rms velocity along the los ( @xmath117 ) and the dispersion of the polarization angle ( @xmath100 ) within regions of @xmath118 pixels . to simulate a realistic cloud we chose the mean magnetic field intensities given in table 2 . in fig . 9 we show the averaged values of the obtained magnetic field for different map resolutions ( @xmath119 , @xmath120 and @xmath121 pixels ) for model 3 with different inclinations of the magnetic field . for all inclinations , coarser resolution calculations from the cf method tend to overestimate the magnetic field intensity . finer resolutions result in the convergence to the actual values @xmath122 . this trend is seen for different inclinations and models . the following equation seems to best fit this behavior : @xmath123 where @xmath124 represents the observational resolution ( total number of pixels ) , @xmath125 and @xmath126 are constants obtained from the best fitting . @xmath126 represents the value of @xmath107 for infinite resolution observations , i.e. the best magnetic field estimation from the cf method . ( 10 ) is shown as the dotted lines in fig . 9 . in fig . 10 we show the magnetic field obtained from eq . ( 8) for the different models with @xmath60 . the dotted lines represent the best fitting using eq . ( 10 ) . the fit parameters , as well as the expected values of the magnetic field from the simulations for all models , are shown in table 2 . here , the magnetic fields are given in units of the mean field @xmath70 . since the simulations are scale independent , one could choose values of @xmath70 to represent a real cloud , in accordance with the parameters of table 1 . as an example , assuming a cloud with @xmath127@xmath128 and @xmath129k , and @xmath130 ( model 3 ) , we get @xmath131 g . choosing differently the density , temperature or the model given by the simulations , i.e. @xmath54 , we obtain a different mean magnetic field . the obtained parameter @xmath125 is very similar for the different inclinations , but are different depending on the model , mainly because it is related to the scale on which the dispersion of the polarization angle changes . since @xmath125 seems to depend on the model and not on the inclination it could also be used by observers to infer the physical properties of clouds from polarization maps . it is shown that @xmath126 decreases as @xmath51 increases but do not reach zero as would be expected from @xmath102 . however , if we compare the magnetic field strength obtained from the cf method with the total magnetic field ( last column of table 2 ) as proposed in eq . ( 8) , the convergence between the estimative and the actual values is much better . the error ( @xmath132/log(b_{\rm cf}^0)$ ] ) using this method is @xmath133 considering all cases , validating the cf method under the assumptions used for eq . ( 8) . as a practical use , observers could obtain polarimetric maps of a given region of the sky for different observational resolutions ( e.g. changing the resolution via spatial averaging ) . using the cf technique for each resolution and , then apply eq . ( 9 ) to determine the asymptotic value of the magnetic field projected into the plane of sky @xmath126 . dense cloud envelopes and diffuse clouds typically present very weak or no far - ir and sub - mm dust emission . in these cases , it is very difficult to obtain polarimetric maps from dust emission and other methods are necessary . some of these clouds are known to intercept rich clusters of stars . ir , optical and ultraviolet ( uv ) emissions from these stars suffer extinction by the dust component of the intercepting clouds and , as a consequence , the detected stellar radiation may be polarized . absorption polarization maps are considered infinite resolution " measurements of polarization vectors and can be very useful on the study of the magnetic field of the ism . however , since it depends on the stellar background , detections are rare and the polarization maps are sparse . to test if the current absorption polarization maps are statistically relevant , as well as the applicability of the cf technique for these type of observation , we simulated the polarization of background stars in our cubes . assuming our cubes to be 1 arcmin@xmath134 of the sky , and using the density estimative of @xmath135/arcmin@xmath134 for our galaxy @xcite , we recalculated the polarization maps of @xmath135 randomly positioned stars . each star is assumed to originate an unpolarized total intensity @xmath136 , which is absorbed by the dust component intercepting the line of sight . at each cell , we compute the absorbed intensities @xmath137 and @xmath138 , which depend on the local magnetic field orientation . again , as in the calculation of the emission maps , we assume maximum efficiency in the polarization by the dust . after integration , the total absorptions @xmath139 and @xmath140 are used to obtain @xmath39 and @xmath40 . the obtained polarization angle is then rotated by @xmath141 ( opposite polarization ) in order to be compared with the emission polarimetric maps . in fig . 11 we illustrate the obtained results . we exemplify the obtained maps with a zoomed clumpy region ( 100 x 100 pixels ) of model 3 with @xmath142 ( upper panel ) . here , it is shown that just a few stars ( @xmath143 ) are detected . to test the predictions of the magnetic field proposed in this work , we calculated the dispersion and the structure function of the polarization angles . in the bottom panel of fig . 11 we show the histogram of @xmath50 , and its structure function ( squares ) , for model 3 with @xmath144 . we compared it with the dust emission sf ( solid line ) . the structure functions seem very similar but , due to the small number of stars , the dispersion in the sf for absorption is large . in this case , the sfs from absorption maps may present too large uncertainties , which make difficult the analysis of the magnetic field from sf slopes . an alternative would be the use of the improved cf technique presented in sec . 5 . we applied the cf method to the absorption polarization maps of model 3 , with @xmath144 , which resulted in @xmath145 @xmath8 g . this value is comparable to the result obtained from the emission polarimetric maps ( @xmath146 @xmath8 g ) . even with a small number of stars , the obtained result is similar to the finest resolution case of polarized emission . it occurs because the stars act as single pixel measurements and there is no averaging of @xmath50 within the observational beam size . the down - side of this technique is its higher noise . currently there are few observed polarization maps available from extinction of background stars . fortunately , some projects are being implemented and promise to bring us complete sets of polarization maps of background stars , which would be compared to the presented results , like the galactic plane infrared polarization survey @xcite . however , we believe that joint analysis of both , emission and absorption , polarization maps can provide a more complete understanding of the magnetic field in dense and diffuse regions of the ism . emission and extinction polarimetric measurements provide an unique technique for the study of the magnetic field , projected into the plane of sky , in molecular clouds . synthetic extinction maps depend on additional assumptions about the stellar population and may be more explored in a future work . in this work we focused on providing synthetic emission polarimetric maps , as well as different statistical analysis that could be used in the future by observers to infer the physical properties of the studied region . the physical interpretations of our results , as well as the comparisons with previous theoretical works , are given as follows . in this work we presented four different models : ( 1 ) @xmath147 , @xmath148 and @xmath149 , ( 2 ) @xmath150 , @xmath151 and @xmath149 , ( 3 ) @xmath130 , @xmath152 and @xmath149 and ( 4 ) @xmath150 , @xmath152 and @xmath153 . similar studies provided by @xcite characterized different models by their pressure ratio . however , our results show completely different polarization maps for the two coincident @xmath54-value models . this because the super - alfvenic flows tend to tangle the magnetic field lines , what is not seen in the sub - alfvenic models ( model 2 ) , even with similar pressure ratio . the super - alfvenic case shows a randomly distributed column density maps , with high constrast between the denser and rarefied regions . on the other hand , sub - alfvenic cases are more filamentary , with contrasts increasing with the sonic mach number . this general picture is independent on the angle between the external mean magnetic field and the los @xmath51 . however , for @xmath154 nearly parallel to the los , the observed polarization will mostly be due to the random fluctuation component @xmath49 . this effect is noticeable comparing figs . 1 and 2 . we see that for sub - alfvenic turbulence the large scale density enhancements are mostly parallel to the mean magnetic fields , with exception to the very dense cores , which can easily change the orientation of the magnetic field . as a consequence , polarization maps will present dense structures mostly aligned with the mean magnetic field . this effect also play a role on the generation of the polarization maps . since we integrate the polarization vectors along the los , the low density cells will systematically increase the homogeneous contribution , as well as the resulting polarization degree . to avoid this effect , we disregarded the contribution from low density cells using a threshold , which depends on the model used . for the models where the magnetic field is oriented parallel to the los , the polarization maps will show polarization vectors randomly oriented in respect to the density structures . it reveals the degeneracy on the polarimetric maps between the super - alfvenic models with those with @xmath155 nearly parallel to the los . the polarization maps showed that the polarization degree is anti - correlated to the column density , in exception to the subsonic case . this result is in agreement with the observations , which revealed polarization holes " associated to the dense cores for most of the regions observed . observationally , @xcite showed that the polarization maps of molecular clouds follow the relation @xmath87 , with @xmath88 . the same trend is observed from polarized extinction of background stars @xcite . it was proposed that this behavior occurs due to changes on dust properties inside denser cores , or even by an increase in thermal pressure , causing depolarization . @xcite studied the role of the radiative torques on the grain alignment at dense cores and obtained @xmath156 , depending on the dust size distribution . @xcite obtained a similar behavior from their numerical simulations of protostellar cores , though for only three dense cores of one single simulation . they assumed a cut - off on grain alignment efficiency for @xmath15mag . for their model , if the alignment efficiency is independent on @xmath9 , the polarization degree was shown to be independent on the column density . @xcite extended this work , improving the radiative transfer . they naturally obtained a decrease in the grain alignment at denser regions , explaining the lower degree of polarization . in our models , we assumed perfect grain alignment ( independent on @xmath9 ) . therefore , the depolarization is exclusively due to the dispersion increase of the polarization angles in denser regions . we obtained @xmath157 for models ( 2 ) and ( 3 ) , but no correlation ( i.e. @xmath158 ) was found for models ( 1 ) and ( 4 ) . for model ( 1 ) , even the densest cores are unable to tangle the magnetic field lines and the polarization degree is homogeneously large . for model ( 4 ) , we have the opposite situation . the super - alfvenic turbulence causes a strong dispersion of the magnetic field even at the less dense regions . for this case , the polarization degree is low everywhere . for models ( 2 ) and ( 3 ) , the turbulence is unable to destroy the magnetic field structure , but is able to create the dense cores by shocks . the cores are dense enough to drag the magnetic field lines and to increase the local @xmath159 . possibly , our results differ from the obtained by @xcite because of : i - numerical resolution , ii - @xmath159 , and iii - self - gravity . we used @xmath35 simulations ( instead of a @xmath160 ) and , as a result , our magnetic field structure is less homogeneous and the density constrast is higher . the larger complexities present in our cubes increase the effect described in the previous paragraph . @xcite and @xcite used a single , super - alfvenic , model . we showed that the polarization maps for @xmath159 present a flat @xmath161 correlation . finally , self - gravity causes the colapse of the denser regions compressing the magnetic field within these cores . as a consequence , if no strong diffusion takes place , the polarization degree tends to grow . as a future work , we plan to study properly the depolarization at dense cores considering the grain alignment process and self - gravity . we found that the distributions of polarization angles of sub - alfvenic models are similar , even for different magnetic to gas pressure ratios . however , the dispersion of angles increases with @xmath55 and with the inclination of the external magnetic field regarding the line of sight ( @xmath51 ) . actually , we noticed that the critical parameter is the alfvenic mach number considering the magnetic field component projected into the plane of sky ( i.e. @xmath162 ) . we can compare these results with @xcite and @xcite . the first used one single super - alfvenic model , and obtained irregular ( flat ) distributions of polarization angle . the latest analysed supersonic models for @xmath163 and 1.0 . they obtained clearly gaussian distributions for @xmath130 , with increasing dispersion for larger inclinations . also , they obtained flatter distributions as @xmath54 increases ( i.e. as the alfvenic mach number increases ) . these are all in agreement with our results . on the other hand , the power spectra analysis was showed to depend on the sonic mach number . the spectra of the polarization angles show an increase in the power of small scales for increasing @xmath164 . this occurs due to the amplification on the perturbations of the smallest scales for stronger turbulence . however , the same behavior is seen varying the inclination of the mean magnetic field . in this sense , there is a degeneracy between the alfvenic mach number and the orientation of the magnetic field . in this sense , structure functions of the polarization angle showed to be useful to avoid this degeneracy . the sub - alfvenic models presented sfs with slope @xmath58 ( @xmath0 ) , independent on the magnetic field orientation . on the other hand , the super - alfvenic model presented flatter sfs ( @xmath165 ) . therefore , we could conjecture that it could be possible to obtain the mean magnetic field intensity , independently on its orientation regarding the los , from the sf slopes . needless to say that more models are needed to confirm this possibility . besides , sfs can potentially be used to study the turbulence eddies . its flat profile at small scales may provide informations about the amplitude and size of the smallest turbulence cells . however , we showed that the results are affected by the observational resolution . the cf method has been proven an useful tool for the determination of the magnetic field in the ism . we studied its validity using the obtained polarization maps from our models . @xcite proposed that the cf method would only be applicable in restricted cases , in which the dispersion of the polarization angle is small ( @xmath11 ) . it is consistent with the original approximations involving the derivation of the cf equation . we derived a generalized formula for the cf method , based on the same assumptions of the original work @xcite , but that accounts for larger dispersion models . basically , we assume that the perturbations are alfvenic and that we have an isotropic distribution of @xmath31 . we , for the first time , successfully applied this equation to the super - alfvenic and large inclination models . we also studied dependency of the cf method and the observational resolution . as also shown by @xcite and @xcite , the cf method overestimates the magnetic field for coarser resolutions . therefore , we propose a general equation to fit the observational data considering maps with different resolution . the asymptotic value of the given procedure provides the infinite resolution " measurement from the cf method and is consistent with the expected values from the simulations . as stated before , a possible limitation in the presented model is the absence of self - gravity effects . at the denser regions , the magnetic field configurations may possibly be different as the cloud collapses and drags the field lines . this process is responsible for the hour - glass structures observed in several gravitationally unstable clouds ( e.g. valle & fiege 2007 ) . as a consequence , self - gravity increases the magnetic field locally and reduces the dispersion of the polarization angles within the dense clumps . to test the stability of the dense clumps in our simulations we may estimate the jeans length ( @xmath166 ) . for the parameters chosen in section 5.3 , the denser structures , considering all models , are characterized by @xmath167pc , @xmath168@xmath128 and @xmath169k , resulting in @xmath170pc . therefore , since @xmath171 , the denser structures may be unstable , at least for the given parameters . on the other hand , self - gravity plays a role at small regions and may not be statistically important for the previous results ( except for the @xmath161 correlation ) as we studied regions much larger than the very dense clumps . we plan to study the effects of self - gravity on the obtained results in a further work . the ratio of thermal gas to magnetic pressures ( @xmath54 ) is typically used as the dominant parameter on the characterization of the degree of magnetization of a cloud . in this sense , systems with similar @xmath54 values should present similar distribution of structures and dynamics . however , we showed that the sonic ( @xmath164 ) and alfvenic ( @xmath55 ) mach numbers , which quantify the ratio of the kinetic to the thermal and magnetic pressures , respectively , divide the models in different regimes . for the case of polarization vectors , our simulations showed that @xmath55 is decisive . for clouds with @xmath172 , the gas motions excited by turbulence are confined by the magnetic field and are not able to change its configuration . actually , the perturbations in the magnetic field occur , but are small compared to the mean field ( @xmath173 ) . in this case , the polarization vectors are uniform , as shown in fig . 4 . for @xmath174 , the magnetic pressure is small compared to the kinetic energy of the turbulent gas and the mean magnetic field can be easily distorted . as a consequence , the polarization maps would show large dispersion of @xmath50 . obviously , a large dispersion of @xmath50 can also be related to a projection effect . if the mean magnetic field is projected along the line of sight ( @xmath175 ) , only the small perturbations @xmath49 will be seen as the polarization vectors . however , considering a large number of clouds , there is a very low probability for all to present @xmath175 . in this case , if observations systematically show very large dispersions of @xmath50 , it means that the turbulence in the ism may be typically super - alfvenic . otherwise , the ism then presents sub(quasi)-alfvenic turbulence . observations of a given cloud could then be compared to our fig . 4 to determine under which regime the turbulence is operating . it is particularly interesting since the ratio of magnetic to turbulent energy in the ism is still subject of controversy @xcite . the number of simulations presented in this work , as well as the numerical resolution , must be increased in future works . in any case , from the models we have , we intend to provide observers with a straightforward procedure to characterize the magnetic field and turbulence properties of molecular clouds . firstly , from the polarization maps of a given region one should obtain the second order structure function of the polarization angle . from the sf , it is possible to characterize the turbulent cascade and the magnetic field . the extension of the flat profile at small scales give the turbulence cut - off scale . on the other extreme , the flat profile at large scales indicate the energy injection lengths . from the maximum slope of @xmath0 , it is possible to determine the averaged alfvenic mach number and , as a consequence , the magnetic field intensity . another method to obtain the magnetic field intensity is based on the cf technique . from the observed velocity dispersion it is possible to estimate the amplitude of the random component of the magnetic field from eq . the total magnetic field is then obtained from eq . ( 9 ) , using the dispersion of the polarization angle . to avoid the dependence on the observational resolution , it is suggested to evaluate the dispersion of the polarization angle for different resolutions ( which may be simulated by averaging neighboring vectors of the polarization maps ) and determine the asymptotic total magnetic field from eq . finally , subtracting the total field by the random component , it is possible to determine the mean magnetic field projected in the plane of sky . also , combining the mean magnetic field obtained from both methods , it is possible to estimate the angle between the mean magnetic field and the los . although it was not considered in the present calculations , a correct treatment of grain alignment is needed for a full understanding of the polarization in molecular clouds . for instance , we showed that the polarization degree is anti - correlated with the column density with slope @xmath2 , while observations sometimes give @xmath176 @xcite . this difference is related with the alignment efficiency at different regions of the cloud @xcite . the theory of grain alignment has developed fastly during the past decade ( see lazarian [ 2007 ] ) . it is currently believed that radiative torques play a major role on the alignment process and it strongly depends on @xmath9 ( see lazarian & hoang [ 2007 ] ) . with increasing extinction ( @xmath177 ) , the radiative torques are less effective and only large grains are aligned . all in all , both observations @xcite and theory @xcite , suggest that there is a range of @xmath9 for which our assumptions are correct . it might happen that subsonic mechanical alignment of irregular grains , introduced in @xcite , extends the range of @xmath9 over which grains are aligned when compared to the estimates based on radiative torques only . the observed band is also selective regarding the dust sizes and different bands reveal the polarization of different dust components . all these effects will be included in a future work , and a more realistic study of the polarization intensity distribution will be obtained . in the present work we focused on calculating synthetic polarization maps of fir emission from dust particles , which were assumed to be perfectly aligned with the magnetic field . unfortunately , due to inefficient grain alignment at the dense cloud cores , the dust polarization degree may decrease and different methods have to be used . the polarization of molecular lines have been shown to be an additional tool for the study of the magnetic fields in the ism @xcite . molecules are present in the dense and cold cores of the clouds and may be detected by thermal line emissions . polarimetric maps of molecular emission can be used on the study of regions with @xmath178 , and can be directly associated with the zeeman measurements . based on the goldreich - kylafis effect @xcite , the molecular sublevel populations will present imbalances due to the magnetic field , generating polarized rotational transitions . however , the survival of molecules depend on restrict conditions , as for @xmath178 molecules may be frozen into dust particles . another difficulty regarding this method is the fact that the gk effect generates polarization either parallel or perpendicular to the magnetic field lines , making the polarization maps . on the other hand , polarized scattering and absorption from atoms and ions provide information about the magnetic field in warm and rarefied regions , like the diffuse ism and the intergalactic medium . polarization arising from aligned atoms and ions is a new method @xcite . unlike molecular lines that live in the excited state long enough to be imprinted by the magnetic field , the atomic exited states are short lived and tend to decay in timescales shorter than the larmor precession of the atom . however , species with fine and/or hyperfine structure of ground or metastable states can be aligned . this fact opens new horizons for polarimetric studies of magnetic fields @xcite . it is useful to comment here that the results shown in this work are also valid for the observed polarization maps of molecular and atomic emission lines . this because the assumptions made for the calculations disregard any special consideration about the emitting species , which could be atoms , molecules or dust particles . also , it is worth mentioning that these techniques are either a substitute , for regions where no fir dust emission is detectable , or complementary to the dust polarized emission but at different wavelengths ( e.g. optical and uv radiation ) . depending on the @xmath9 range considered , polarization of atoms and molecules may complement the dust emission and absorption maps , which are usually much more detailed . in this work we studied of polarization maps and its applicability on the determination of magnetic fields in molecular clouds based on numerical simulations . here we compare the obtained results with the previous theoretical works . @xcite performed numerical simulations , with @xmath10 resolution , considering plasma @xmath54 values 0.01 , 0.1 and 1.0 . their results showed homogeneous polarization maps for @xmath130 ( strongly magnetized turbulence ) , and a complex distribution of polarization vector for @xmath150 and 1.0 ( weakly magnetized turbulence ) . they also obtained an increase in the dispersion of polarization angles with the increase of the external magnetic field inclination regarding the line of sight . these results are in agreement with our models , except for the fact that two of our models with equal @xmath54 presented completely different polarization maps . this because the @xmath54 value does not reveal how the magnetic field lines respond to the turbulence . the alfvenic and sonic mach numbers reflect how strong is the turbulent pressure compared to the magnetic and thermal pressures , respectively . they obtained a higher polarization degree for @xmath54 = 0.01 , obviously because of the magnetic field intensity , but larger values of @xmath179 for larger column densities ( i.e. for larger @xmath43 ) , in disagreement with observations . we believe that this was caused by their method for obtaining the polarization degree . they obtained the integrated stokes parameters , weighted by local density , for all cells along the los . we , on the other hand , used a threshold on density to avoid the contribution of very rarefied regions ( where the magnetic field structure is systematically more uniform ) . @xcite focused their work on the polarization of dust emission from dense cores . they implemented a more realistic calculation of the polarization degree based on the efficiency of the alignment for different values of @xmath180 . in this case , they were able to obtain a decreasing polarization degree with the total intensity related to the grain properties , and not to the statistics of polarization vectors along the line of sight . we also believe that the numerical resolution may be playing a role on the polarization degree . more refined simulations systematically result in more complex structures for density and magnetic field lines . as a consequence , the alignment vectors along the line of sight present larger dispersion resulting in a lower polarization degree . we found no previous theoretical work presenting an extended statistical anaylsis considering all the pdf , spectra and structure function of polarization angle , and therefore no comparison can be made . @xcite and @xcite also tested the cf technique using their simulations , considering eq . they obtained good agreement , with a correction factor of @xmath84 , between the calculated estimations and the expected values only for the models with @xmath181 . @xcite presented an extended analysis using a larger number of models , with different physical parameters and numerical resolutions ( including 1 model with @xmath35 resolution ) . they also studied the effects of observational resolution on the obtained maps . they concluded that coarser resolutions result in more uniform polarization vectors . as a consequence , the cf method overestimates the magnetic field intensity . this is in full agreement with our results . they also tested the reliability of the cf technique in weakly magnetized clouds . they proposed the modified equation @xmath182[1 + 3\delta(\tan \phi)^2]^{1/2}$ ] , is the value obtained using the standard cf equation , and @xmath183 is the corrected value proposed by @xcite . ] which gave good results compared with the expected values for their models , with discrepancies of a factor @xmath184 . we tested their equation to our models 3 and 4 with @xmath60 , representing a strong and weakly magnetized cloud , respectively . for model 3 , the obtained value is in agreement with that shown in table 2 . the ratio between the two measurements is @xmath185 . for model 4 , the proposed equation underestimates the magnetic field , and compared to with our method it gives @xmath186 . despite of the few simulations available for compariron , their method seems to systematically underestimate the magnetic field intensity . more tests are needed to determine which method may give the best results . furthermore , even though not addressed by @xcite , we studied the dependence of the polarization angles and the cf technique with the inclination of the magnetic field regarding the los . we showed that there is a degeneracy between the results of weakly magnetized clouds and strongly magnetized clouds with high @xmath51 . the modified cf formula presented in this work gave good results for all cases . as discussed before , even though not taking into account the self - gravity in our simulations it mostly induces changes at small scales , as noted by @xcite . as a result , they showed the cf technique to be insensitive to self - gravity . in this work we presented turbulent 3-d high resolution mhd numerical simulations in order to study the polarized emission of dust particles in molecular clouds . we obtained synthetic dust emission polarization maps calculating the stokes parameters @xmath39 , @xmath40 and @xmath43 assuming perfect grain alignment and that the dust optical properties are the same at all cells . under these conditions , we were able to study the polarization angle distributions and the polarization degree for the different models and for different inclinations of the magnetic field regarding the los . as main results , we : - presented a generalization of the cf method , which showed useful for : i- the determination of the total magnetic field projected in the plane of sky , and ii- the separation of the two components @xmath122 and @xmath49 ; these results represent important tools for present and future polarimetric fir observational studies . in the future it would be necessary to increase the number of simulated models , with different physical parameters and with better resolutions , to test these conclusions . it would also be interesting to implement the grain alignment processes properly and study their effects on the obtained results . using more computational models it will be possible to test the proposed method to remove the degeneracy between the alfvenic mach number and the angle between the magnetic field and the los . this would also help us to provide the observers a large number of simulated clouds that could be used as benchmarks . d.f.g . , a.l . and g.k . thank the financial support of the nsf ( no . ast0307869 ) , the center for magnetic self - organization in astrophysical and laboratory plasmas and the brazilian agencies fapesp ( no . 06/57824 - 1 and 07/50065 - 0 ) and capes ( no . 4141067 ) . arce , h. g. , goodman , a. a. , bastien , p. , manset , n. & sumner , m. 1998 , , 499 , 93 cortes , p. c. , crutcher , r. m. & watson , w. 2005 , , 628 , 780 cho , j. & lazarian , a. 2005 , , 631 , 361 chandrasekhar , s. & fermi , e. 1953 , , 118 , 113 crutcher , r. m. 1999 , , 520 , 706 crutcher , r. m. , roberts , d. a. , troland , t. h. & goss , w. m. 1999 , , 515 , 275 dolginov , a. z. & mytrophanov , i. g. 1976 , , 43 , 257 dotson , j. 1996 , , 470 , 566 draine , b. t. & weingartner , j. c. 1996 , , 470 , 551 draine , b. t. & weingartner , j. c. 1997 , , 480 , 633 esquivel , a. & lazarian , a. 2005 , , 631 , 320 garwood , r. & jones , t. j. 1987 , , 99 , 453 girart , j. m. , crutcher , r. m. & rao , r. 1999 , , 525 , 109 girart , j. , greaves , j. m. , crutcher , r. m. & lai , s .- 2004 , , 292 , 119 girart , j. m. , rao , r. & marrone , d. p. 2006 , science , 313 , 812 goldreich , p. & kylafis , n. d. 1981 , , 243 , 75 goldreich , p. & kylafis , n. d. 1982 , , 253 , 606 gonalves , j. , galli , d. & walmsley , m. 2005 , , 430 , 979 greaves , j. s. , holland , w. s. & dent , w. r. f. 2002 , , 578 , 224 heiles , c. & troland , t. h. 2005 , , 624 , 773 heitsch , f. , zweibel , e. , mac low , m. m. , li , p. s. , & norman , m. l. 2001 , , 561 , 800 hildebrand , r. h. , davidson , j. a. , dotson , j. l. , dowell , c. d. , novak , g. , & vaillancourt , j. e. 2000 , , 112 , 1215 hildebrand , r. h. , dotson , j. l. , dowell , c. d. , schleuning , d. a. & vaillancourt , j. e. 1999 , , 516 , 834 hoang , t. & lazarian , a. 2008 , , in press kritsuk , a. g. , norman , m. l. , padoan , p. & wagner , r. 2007 , , 665 , 416 kowal , g. , lazarian , a. & beresniak , a. 2007 , , 658 , 423 lai , s .- p . , girart , j. m. & crutcher , r. m. 2003 , , 598 , 392 lazarian , a. , goodman , a. a. & myers , p. c. 1997 , , 490 , 273 lazarian , a. & pogosyan , d. 2000 , , 537 , 720 lazarian , a. & cho , j. 2004 , , 292 , 29 lazarian , a. , vishniac , e. t. & cho , j. 2004 , , 603 , 180 lazarian , a. 2007 , , 106 , 225 lazarian , a. & hoang , t. 2007a , , 378 , 910 lazarian , a. & hoang , t. 2007b , , 669 , 77 matthews , b. c. & wilson , c. d. 2002 , , 574 , 822 myers , p. c. & goodman , a. a. 1991 , , 373 , 509 ostriker , e. c. , stone , j. m. & gammie , c. f. 2001 , , 546 , 980 padoan , p. , goodman , a. , draine , b. t. , juvela , m. , nordlund , a. & rognvaldsson , o. e. 2001 , , 559 , 1005 padoan , p. & norlund , a. 2002 , , 576 , 870 pelkonen , v .- m . , juvela , m. & padoan , p. 2007 , , 461 , 551 schleuning , d. a. 1998 , , 493 , 811 shiode , joshua h. , clemens , d. p. , janes , k. a. , pinnick , a. & taylor , b. 2006 , , 209 , 156 valle , j. p. & fiege , j. d. 2007 , , 134 , 628 whittet , d. c. b. , hough , j. h. , lazarian , a. & hoang , t. 2008 , , in press wolf , s. , launhardt , r. & henning , t. 2003 , , 592 , 233 yan , h. & lazarian , a. 2006 , , 653 , 1292 yan , h. & lazarian , a. 2007a , , 657 , 618 yan , h. & lazarian , a. 2007b , astro - ph:0711.0926 zweibel , e. g. 1990 , , 362 , 545 ( _ up _ ) , and for model 3 and different magnetic field orientations in respect to the line of sight ( angles @xmath51 ) ( _ bottom_).,title="fig:",width=491 ] + ( _ up _ ) , and for model 3 and different magnetic field orientations in respect to the line of sight ( angles @xmath51 ) ( _ bottom_).,title="fig:",width=491 ] ) ( _ up _ ) , model 4 with different @xmath51 ( _ middle _ ) and for the different models with magnetic field perpendicular to the line of sight ( @xmath60 ) ( _ bottom_).,title="fig:",width=415 ] + ) ( _ up _ ) , model 4 with different @xmath51 ( _ middle _ ) and for the different models with magnetic field perpendicular to the line of sight ( @xmath60 ) ( _ bottom_).,title="fig:",width=415 ] + ) ( _ up _ ) , model 4 with different @xmath51 ( _ middle _ ) and for the different models with magnetic field perpendicular to the line of sight ( @xmath60 ) ( _ bottom_).,title="fig:",width=415 ] ) ( _ up _ ) , model 4 with different @xmath51 ( _ middle _ ) and for the different models with magnetic field perpendicular to the line of sight ( @xmath60 ) ( _ bottom_).,title="fig:",width=377 ] + ) ( _ up _ ) , model 4 with different @xmath51 ( _ middle _ ) and for the different models with magnetic field perpendicular to the line of sight ( @xmath60 ) ( _ bottom_).,title="fig:",width=377 ] + ) ( _ up _ ) , model 4 with different @xmath51 ( _ middle _ ) and for the different models with magnetic field perpendicular to the line of sight ( @xmath60 ) ( _ bottom_).,title="fig:",width=377 ] ( _ bottom _ ) . @xmath188 is 100% , 98% , 97% and 85% for model 1 , 2 , 3 and 4 , respectively.,title="fig:",width=415 ] + ( _ bottom _ ) . @xmath188 is 100% , 98% , 97% and 85% for model 1 , 2 , 3 and 4 , respectively.,title="fig:",width=415 ] . _ bottom _ : histogram of polarization angle for model 3 with @xmath144 , and its structure function ( squares ) compared to the emission sf ( solid line).,title="fig:",width=377 ] + . _ bottom _ : histogram of polarization angle for model 3 with @xmath144 , and its structure function ( squares ) compared to the emission sf ( solid line).,title="fig:",width=453 ]

in these proceedings , i shall review the basic concepts of perturbative qcd in its high - energy limit . i shall concentrate on the approach to the unitarity limit , usually referred to as _ saturation _ , as well as on the gluon - number fluctuations the importance of which has recently been discovered . i shall explain the basic framework showing the need for those phenomena , first , from a simple picture of the high - energy behaviour , then , giving a short derivation of the equation driving this evolution . in the second part , i shall exhibit an analogy with statistical physics and show how this allows to derive _ geometric scaling _ in qcd with saturation . i shall finally consider the effects of gluon - number fluctuations on this picture and draw the physical consequences , _ i.e. _ a new scaling law , arising from those results .

hep-ph0610436

the quest for the high - energy behaviour of perturbative qcd started thirty years ago , soon after qcd was proposed as the fundamental theory of strong interactions . to grasp this problem , it has been realised that saturation effects were to be considered in order to satisfy unitarity constraints . throughout these proceedings , i shall summarise the present status of our understanding of this limit , which basically means finding an equation giving the evolution of amplitudes towards high energy and finding the properties of its solutions . in order to motivate the physical picture we want to reproduce , let us consider the problem of deep inelastic scattering ( dis ) _ i.e. _ @xmath0 collisions . this process depends on two kinematical variables : the virtuality @xmath1 of the photon , imposing @xmath2 as the ( transverse ) resolution scale at which the photon scans the proton , and the bjorken @xmath3 , given by the fraction of the proton momentum carried by the parton struck by the virtual photon in a frame where the proton is moving fast . this last variable is related to the centre - of - mass energy @xmath4 , meaning that the high - energy limit corresponds to the small-@xmath3 limit ( at fixed @xmath1 ) . for practical purposes , we also introduce the _ rapidity _ defined through @xmath5 . in figure [ fig : dis ] , we have represented the typical configuration of the proton in different domains of this phase space . we start with a proton at low @xmath1 and low energy , represented as three partons ( bottom - left part of the picture ) . if we increase the virtuality of the photon , we are able to resolve its partons into smaller ones . this type of evolution is described in perturbative qcd by a renormalisation group equation known as the dokshitzer - gribov - lipatov - altarelli - parisi ( dglap ) equation @xcite . in that limit , the number of partons rises logarithmically while their typical size decreases like @xmath6 so that the proton becomes more and more dilute . in what follows , we shall be interested in the other type of evolution , namely the evolution towards smaller values of @xmath3 for constant @xmath1 . if one boosts the proton , what typically happens is that we create new partons of a size comparable to the parents ones . obviously , during this evolution , known as the balitsky - fadin - kuraev - lipatov ( bfkl ) evolution @xcite , the proton approaches the black - disc limit for which partons start to overlap . at that level , they start to interact among themselves and unitarity corrections are to be considered @xcite . this blackening of the proton is called saturation and we shall describe it in details from perturbative qcd in these proceedings . in addition , it has recently been discovered that gluon - number fluctuations play a crucial role in the evolution towards high energy . as we shall see later on , those fluctuations , dominating the evolution in the dilute tail ( _ i.e. _ at large momentum ) of the system , are amplified by bfkl evolution and thus influence greatly the whole evolution . in the first part of these proceedings , we shall derive the equations describing the evolution towards high - energy . we shall begin our analysis with a short introduction to the dipole picture which provides a very intuitive framework to deal with high - energy qcd at large-@xmath7 . this will allow for a derivation of the bfkl equation . the solution of the bfkl equation naturally leads to the introduction of unitarity corrections and to new evolution equations - the balitsky / jimwlk hierarchy @xcite and the balitsky - kovchegov ( bk ) equation @xcite . we shall then derive and discuss the contribution due to gluon - number fluctuations @xcite . in the second part , we shall consider the solutions to those evolution equations . first , we shall show that , within relevant approximations , the bk equation reduces @xcite to the fisher - kolmogorov - petrovsky - piscounov ( f - kpp ) equation @xcite , well studied in statistical physics . this allows for a description @xcite of the asymptotic behaviour of the scattering amplitudes in terms of travelling waves , translating in qcd into the _ geometric scaling _ property @xcite . we shall also show that geometric scaling extends to nonzero momentum transfer @xcite . in the last section , we shall come back to the effect of gluon - number fluctuations , which have recently proven to have important consequences @xcite on the approach to saturation . in order to clearly emphasise the need for a resummation at high energy , let us start with a fast - moving quark of momentum @xmath8 . this quark emits bremsstrahlung gluons characterised by their transverse momentum @xmath9 and their longitudinal momentum @xmath10 . the probability for emitting such a gluon is @xmath11 when we consider large transverse momentum ( high @xmath1 and fixed @xmath3 in dis ) , the collinear divergence @xmath12 needs to be resummed and this gives rise to a renormalisation group equation known as the dglap equation . throughout this proceedings , we shall instead consider the limit of high energy , in which , we are sensitive to the gluons of small longitudinal momentum @xmath13 at fixed @xmath1 . this leaves a large phase - space to allow for successive soft gluon emissions satisfying @xmath14 which gives a contribution of order @xmath15 even when the coupling @xmath16 is small enough to ensure applicability of perturbation theory , for sufficiently small values of @xmath3 ( _ i.e. _ at sufficiently high energy ) , we have @xmath17 and all these contributions have to be resummed . in practice , it is more convenient to start with a quark - antiquark pair of transverse coordinates @xmath18 and @xmath19 . successive soft gluon emissions can be considered as independent and are depicted in figure [ fig : dip_wf ] . in order to simplify the discussion , we shall consider the large-@xmath7 limit ( @xmath7 is the number of colours ) . in that limit , a gluon of transverse coordinate @xmath20 can be considered as a quark - antiquark pair at point @xmath20 . this means that , instead of considering a wavefunction made of quarks , antiquarks and gluons , it is sufficient to consider an _ onium _ which is an evolving system of colourless @xmath21 dipoles . we can thus think only in terms of dipoles . this is the dipole picture introduced by mueller @xcite . at high - energy , all possible gluon emissions from one dipole are thus tantamount to this dipole splitting into two child dipoles as depicted in figure [ fig : dip_split ] . we see that the probability density for a dipole of transverse coordinates @xmath22 to split into two child dipole @xmath23 and @xmath24 can be computed from gluon emissions from the quark and antiquark lines . the computation is performed in old - fashioned light - cone perturbation theory but we shall simply give the final result : @xmath25 with @xmath26 and @xmath27 ^ 2 = \frac{({\mathbf{x}}-{\mathbf{y}})^2}{({\mathbf{x}}-{\mathbf{z}})^2({\mathbf{z}}-{\mathbf{y}})^2}.\ ] ] let us denotes by @xmath28 the scattering amplitude denotes the average over all possible realisations of the target colour field . it can be understood as the expectation value for the @xmath29-matrix operator over the target wavefunction . ] for a dipole made of a quark at transverse coordinate @xmath18 and an antiquark at @xmath19 . getting the evolution of @xmath30 from the dipole picture is pretty straightforward . indeed , if one boosts the dipole @xmath22 from a rapidity @xmath31 to a rapidity @xmath32 , this dipole shall split into two dipoles @xmath23 and @xmath24 . each of those dipoles can interact with the target which leads to the following evolution equation @xmath33 where we have explicitly used the probability density for dipole splitting . equation is the bfkl equation , derived in the mid - seventies @xcite by balitsky , fadin , kuraev and lipatov . note that here we have derived the equation by putting the evolution in the projectile dipole . it is also possible to write the evolution equation for the dipole density @xmath34 in the target and to obtain by letting the projectile dipole interact with the evolved target . we sketch out the derivation of the bfkl equation for the dipole wavefunction of the target in appendix [ app : bfkln ] . this equation being linear , one expects its solution to grow exponentially . to be more explicit , if one neglects impact - parameter dependence , _ i.e. _ assumes that @xmath35 , the solution of is found by mellin transform : @xmath36 where @xmath37 is the bfkl eigenvalues plotted on figure [ fig : chi ] , and @xmath38 describes the initial condition . for a fixed dipole size , one finds the high - energy behaviour by expansion around the saddle point @xmath39 . one obtains @xmath40,\ ] ] with @xmath41 being the bfkl pomeron intercept . this expression suffers from two major problems : * first , even if one start with an initial condition peaked around a small value of @xmath42 , when rapidity increases , the amplitude starts to diffuse to the large dipole sizes _ i.e. _ to the non - perturbative domain ; * second , the solution of the bfkl equation grows exponentially with rapidity . hence it violates the unitarity constraint @xmath43 obtained from first principles . this unitarity problem is precisely the point where we meet the requirement to take into account multiple interactions . the most straightforward way to see this is to come back to the splitting of a dipole @xmath22 into @xmath23 and @xmath24 . when the target becomes dense enough , both dipoles can scatter on it . we thus have to take into account this contribution , represented in figure [ fig : multiple ] . this leads to a quadratic suppression term in the evolution equation which becomes @xmath44 the first thing to remark is that this new term is of the same order as the previous ones when @xmath45 or , equivalently , when @xmath46 . it is thus , as expected , a mandatory contribution near the unitarity limit . however , equation involves a new object , namely @xmath47 which probes correlations inside of the target . in a general framework , one should then write down an equation for @xmath48 , which will involve @xmath48 through bfkl - like contributions and @xmath49 from unitarity requirements . this ends up with a complete hierarchy , giving the evolution for each @xmath50 , known as the ( large-@xmath7 ) balitsky hierarchy @xcite . if the target is sufficiently large and homogeneous , one can simply assume @xmath51 , ending up with a closed equation for @xmath30 @xmath52 this last expression is the balitsky - kovchegov ( bk ) equation @xcite , which is the most simple equation one can obtain from perturbative qcd by including both the bfkl contributions at high - energy and the corrections from unitarity . remembering that the dipole splitting is equivalent to gluon emission , we see from the diagram in figure [ fig : multiple ] that the bk equation resums _ fan diagrams _ in addition to bfkl ladders . it is easy to check that @xmath53 is an unstable fix point of the bk equation , while @xmath54 is a stable fix point , ensuring unitarity is satisfied . for @xmath55 . the dashed lines correspond to the solution of the bfkl equation while the solid lines correspond to bk evolution . ] in addition , this new term also solves the problem of infrared diffusion . this can be seen on figure [ fig : dipdens ] which shows the dipole occupation number for numerical solutions of the bfkl and bk equations in momentum space ( @xmath56 is the dipole momentum , canonically conjugated to its size ) . the bfkl solution exhibits the gaussian shape predicted from , extending both to the infrared and to the ultraviolet . in the bk solution , we clearly see that the infrared evolution is cut and the maximum of the distribution provides a natural scale @xmath57 , called the _ saturation momentum _ , increasing with rapidity . in general , one can say that saturation corrections cut the emission of dipoles of sizes larger that the inverse of the saturation momentum . we shall come back with more details to the study of the solution of the bk equation in the next section . . they correspond , from left to right , to bfkl ladders , saturation corrections and fluctuations effects . ] very recently , it has been realised that the question of the high - energy limit of qcd , even at the leading logarithmic accuracy , is not fully described by the balitsky hierarchy . to see this more precisely , let us consider in more details the evolution equation for @xmath48 . it corresponds to the scattering of two dipoles off a target and the diagrams contributing to one step of rapidity evolution are shown in figure [ fig : t2 ] . the first graph on the left of this figure is simply the bfkl - ladder contribution for which @xmath58 . the diagram in the middle corresponds to multiple scattering ( see also figure [ fig : multiple ] ) . it accounts for unitarity corrections and gives a contribution @xmath59 . if those two graphs were the only relevant ones in the evolution , one would obtain the ( large-@xmath7 ) balitsky equation . nonetheless , it has been argued that a third contribution , represented by the rightmost diagram of figure [ fig : t2 ] , has to be included . this graph describes multiple scattering off the projectile or , equivalently , gluon - number fluctuations inside the target . after the pomeron - merging term , contributing to saturation , one thus now also includes the pomeron - splitting contribution , which gives rise to a new term in the evolution equation @xmath60 . by combining pomeron mergings and pomeron splittings it is possible to build _ pomeron loops_. for that reason , in what follows we shall refer to those equations as the _ pomeron - loop equations_. for large-@xmath7 and at the two - gluon - exchange level , this new term has been computed @xcite . we shall only give the main steps hereafter . we start with the observation that gluon - number fluctuations are expected to be important when the target is still dilute , which is always the case for sufficiently small dipole sizes . in that case , it can be understood as a superposition of dipoles and the @xmath61-channel gluon in the fluctuation diagram ( last diagram of figure [ fig : t2 ] ) can be considered as emitted through a dipole splitting in the target , as represented on figure [ fig : fluct ] . since we work in the two - gluon - exchange approximation , the scattering amplitude and the dipole density are related through @xmath62 with @xmath63 being the dipole - dipole amplitude at the two - gluon - exchange level @xmath64.\ ] ] this relation can be inverted to obtain @xmath65 the new contribution to the evolution of @xmath66 is thus @xmath67 the correction becomes of the same order as the bfkl contribution when @xmath68 , or @xmath69 , _ i.e. _ in the dilute regime where , as expected , fluctuations should lead to important effects . although this may at first sight seems irrelevant for the physics of saturation , one has to realise that , due to colour transparency , the amplitude becomes arbitrarily small for small dipole size , hence , even for dense targets , there is always some region of the phase - space where the system is dilute and its evolution governed by fluctuation effects . let us give an additional argument in favour of the importance of the fluctuation term on high - energy evolution . once we have a pomeron splitting coming from a fluctuation , it grows , from bfkl evolution , like two pomerons , _ i.e. _ @xmath70 . this has to be compared with the one - pomeron exchange @xmath71 . we see that at very large energies @xmath72 , bfkl evolution compensates the extra factor of @xmath73 coming from the initial fluctuation . we shall see more precisely in the last section that those fluctuation effects in the dilute tail have important consequence on the saturation physics . so far in this section , we have only discussed the evolution equation for @xmath48 . what we really get is an infinite hierarchy , where the evolution of @xmath30 is given by , as in the balitsky hierarchy , and the evolution of @xmath50 contains three contributions : a linear bfkl term proportional to @xmath50 , a ( negative ) saturation term proportional to @xmath74 as it appears in the balitsky hierarchy , and the new fluctuation contribution proportional to @xmath75 , similar to . remarkably enough , it is possible to show that this infinite hierarchy for the average amplitudes can be rewritten as a single langevin equation , for a simpler of langevin equation encountered in section 4 . ] @xmath76\\\nonumber & + & \frac{1}{2 } \frac{{\bar\alpha}}{2\pi}\frac{\alpha_s}{2\pi } \int_{{\mathbf{u}}{\mathbf{v}}{\mathbf{z } } } { \cal{a}}_0({\mathbf{x}}{\mathbf{y}}|{\mathbf{u}}{\mathbf{z } } ) \frac{|{\mathbf{u}}-{\mathbf{v}}|}{({\mathbf{u}}-{\mathbf{z}})^2 } \sqrt{\nabla^2_{\mathbf{u}}\nabla^2_{\mathbf{v}}t_{{\mathbf{u}}{\mathbf{v}}}}\ , \nu_{{\mathbf{u}}{\mathbf{v}}{\mathbf{z}};y},\end{aligned}\ ] ] where @xmath77 is a gaussian white noise satisfying @xmath78 different realisations of the noise term lead to different events @xmath79 , from which one can compute @xmath50 using standard statistical techniques . however , dealing with equation remains very difficult since the noise term is off - diagonal and non - local . finally , let us conclude this section by quoting that the equation , including linear , saturation and fluctuation effects is the most complete equation known in perturbative high - energy qcd so far . its extension beyond the large-@xmath7 approximation is still a challenging problem . let is make a short digression and consider a _ reaction - diffusion _ system in which we have a set of particles evolving through splitting at a rate @xmath80 and merging at a rate @xmath81 : @xmath82 for simplicity , we shall not consider any spatial dimension , though it only introduces technical difficulties without altering the conclusions . if we want to study the time evolution of such a system , the best way is to start with the _ master equation _ which gives the evolution of @xmath83 , the probability to have @xmath84 particles . it contains four terms , corresponding to gain and loss terms for splitting and merging weighted by adequate combinatorial factors : @xmath85 in practice , this equation is only helpful to find evolution of observable , averaged , quantities . the most simple example is the average particle number @xmath86 which can be generalised to a subset of @xmath56 particles @xmath87 using the master equation , one finds the evolution of the particle number and its correlators after a bit of algebra @xmath88 more interesting is the evolution of the scattering amplitude for this system off a generic target , defined by @xmath89 where we evolve the particle system up to a time @xmath90 and the scattering matrix for the target from @xmath90 to @xmath91 . to be consistent , this definition should not depend on any specific choice for @xmath90 . this gives a renormalisation group equation from which one can infer the evolution of the scattering amplitude : @xmath92}_{\text{fluct.}}.\ ] ] we clearly recognise in this expression the bfkl , saturation and fluctuations contributions obtained in qcd . inserting the correct spatial degrees of freedom to account for transverse coordinates , we find @xcite that the high - energy qcd evolution hierarchy obtained in the previous section can be seen as an _ effective dipole reaction - diffusion process _ where rapidity plays the role of time and with the vertices @xmath93 for splitting and merging respectively . one recognise in these expressions that the splitting is given by the bfkl kernel while the merging comes from the computation in the previous section . this interesting analogy is however to be taken with care . indeed , the merging vertex is not positive defined . physically , this comes from the fact that fluctuations really involve the gluonic degrees of freedom . their interpretation as a reaction - diffusion system in terms of dipoles can hence only be effective . in this section , we shall sketch out the main properties of the solutions of the evolution equations towards high energy . as we shall explain in details in the next lines the major source of information comes from the analogy between the bk equation and the fisher - kolmogorov - petrovsky - piscounov ( f - kpp ) equation which is well studied in statistical physics . this section shall be devoted to derive this analogy and its consequences in the case of the bk equation . we shall first consider the impact - parameter - independent bk equation and then show how the arguments extend to the full bk equation . we leave the discussion concerning the effect of fluctuations for the next section . so , let us start with the bk equation . we shall first restrict ourselves to the impact - parameter - independent version of the equation , for which one can easily go to momentum space by using quotations as they do not play any role in the mean - field approximation . ] @xmath94 and where the equation becomes @xmath95 - \bar\alpha { \tilde{t}}^2(k).\ ] ] it is actually possible to rewrite the integral in is equation in term of a differential operator : @xmath96 with @xmath97 , @xmath98 being a soft reference scale and @xmath99 is the bfkl kernel . in order to simplify the problem , we shall work in the saddle point approximation ( often referred to as the _ diffusive approximation _ ) _ i.e. _ expand the bfkl kernel to second order around @xmath100 , transforming the complicated differential operator in equation into a second - order operator @xmath101 up to a linear change of variable switching from @xmath31 and @xmath102 to time @xmath103 and space @xmath104 and a renormalisation of the amplitude @xmath105 , the bk equation in the saddle point approximation becomes @xmath106 this equation is nothing but the f - kpp equation studied in statistical physics since forty years and applying to many problem . it describes reaction diffusion processes in the mean - field approximation where one can have creation and annihilation of particles locally and diffusion to neighbouring site . this equation has many applications _ e.g. _ in chemistry and biology . one knows that the f - kpp equation admits travelling waves as asymptotic solutions _ i.e. _ there exists a critical velocity @xmath107 and a critical slope @xmath108 , determined only from the knowledge of the linear kernel of the f - kpp equation @xmath109 , such that ( see figure [ fig : bkfront ] for a pictorial representation coming from the bk equation which we shall explain a bit later ) @xmath110 although working in the diffusive approximation makes the discussion easier due to the direct matching with the f - kpp equation , this assumption is not required . indeed , for an equation with a more complicated kernel ( than @xmath109 ) , one can prove the existence of travelling waves provided the three following conditions are satisfied : 1 . the amplitude has @xmath111 as unstable fix point and @xmath112 as stable fix point , 2 . the initial condition decreases faster than @xmath113 at large @xmath102 ( see below for the general definition of @xmath108 ) , 3 . the equation obtained by neglecting the nonlinear terms admits superposition of waves as solution : @xmath114,\ ] ] where the wave of slope @xmath80 travels at a speed @xmath115 . then , travelling waves are formed during evolution with the critical parameters @xmath108 and @xmath107 obtained , from the linear kernel only , through the relation @xmath116 this corresponds to the selection of the wave with the minimal speed @xmath117 and the point where group and phase velocities are equal . before going any further , let us give a qualitative argument explaining how the minimal speed is selected . to simplify the discussion , we shall consider the initial front at the left of figure [ fig : schem ] . at an initial time , it captures the main features of the spatial dependence : saturation ( first condition here - above ) followed by a decrease with the critical exponent @xmath108 and finally a steeper decrease ( second condition here - above ) . after one step of time evolution , each of the decreasing fronts shall evolve with its own speed ( those fronts are in a dilute regime so we can use the third condition ) . the result of this evolution is shown on figure [ fig : schem ] and we clearly see that the region over which the critical slope , _ i.e. _ the one with the minimal speed , is growing during the evolution . this selection mechanism is very general . it mainly relies on the properties of the linear equation and saturation only ensures that the slope is @xmath108 at some point in the initial condition . from the leftmost amplitude and evolve to higher @xmath31 using . amplitude is shown for @xmath118 and clearly exhibits a travelling - wave pattern . ] . more precisely , the speed of the wave , _ i.e. _ the exponent of the saturation scale , is plotted . it goes to a constant as it should . ] in the case of the bk equation , the first condition is satisfied due to bfkl growth and saturation , the second comes from colour transparency ( @xmath119 ) and the last one is equivalent to the solution ( with @xmath120 replaced by @xmath121 ) of the bfkl equation obtained by neglecting nonlinear terms in . the critical parameters ( for the complete bfkl kernel _ i.e. _ for ) , are found to be @xmath122 and @xmath123 . written in terms of the qcd variables @xmath31 and @xmath124 , one can then write the asymptotic solution for the impact - parameter - independent bk equation under the form @xmath125^{-\gamma_c}\log\left[\frac{k^2}{q_s^2(y)}\right],\ ] ] with the saturation scale given by @xmath126.\ ] ] the formation of a travelling - wave pattern when energy increases can easily be seen on numerical simulations of the bk equation . as displayed in figure [ fig : bkfront ] , if one start with a steep enough initial condition ( leftmost curve ) , the amplitude increases with energy and a wave moving into the dilute domain gets formed . from that simulation , one can extract the saturation scale by solving @xmath127 at each values of @xmath31 and for a fixed threshold @xmath128 . in figure [ fig : bkqs ] , we have plotted @xmath129 which goes to a constant value as it is expected from . equation has a remarkable property : it proves that at high energy the amplitude , _ a priori _ a function of both @xmath31 and @xmath56 , depends only on the ratio between @xmath56 and the saturation momentum . this property , known as _ geometric scaling _ , has been observed @xcite in the hera measurements of the proton structure function . the fact that geometric scaling can be derived from the bk equation is one of the most important indication for the experimental observation of saturation . the saturation scale obtained from the structure function data is of the order of 1 gev for @xmath130 . this means that high - energy qcd can indeed be studied from the point of view of perturbation theory . up to now , we have only discussed the impact - parameter - independent bk equation . we might therefore ask whether or not these arguments extend to the full equation including all phase - space degrees of freedom . a dipole of transverse coordinates @xmath22 is then represented through its size @xmath131 and impact parameter @xmath132 . the problem is then that dipole splitting is non - local in impact parameter . this means that the bk equation couples different values of @xmath133 and we can not apply directly the previous arguments for each value of @xmath133 . again , the solution consists in moving to momentum space and replace the impact parameter by the momentum transfer @xmath134 @xmath135 the bk equation then takes a form @xcite for which the bfkl kernel is local in @xmath134 ( the non - locality of the non - linear term is not important for our purposes as its only role is to ensure saturation ) @xmath136 we can now proceed with this equation in a similar way as for the the impact - parameter - independent case . the existence of travelling waves requires the three conditions stated previously to be satisfied . the two first ones are trivially satisfied . for the third condition to be valid , we need to find solutions of the linear part of , _ i.e. _ the complete bfkl equation , which can be expressed as a superposition of waves . this consists in a careful treatment of the solutions of the bfkl equation @xcite including all phase - space variables . it turns out that , using the momentum transfer @xmath134 instead of the impact parameter @xmath133 , leads to a powerful factorisation between the target and the projectile ( see appendix [ app : nonzero ] for more details ) . then , we can show @xcite that when the dipole momentum @xmath56 is much larger than the momentum transfer @xmath137 and the typical scale of the target @xmath98 , the solutions of the bfkl equation are a superposition of waves and therefore , we obtain travelling waves for the full bk equation . more precisely , the high - energy behaviour of the amplitude takes the form @xmath138^{-\gamma_c}\log\left[\frac{k^2}{q_s^2(q;y)}\right].\end{aligned}\ ] ] this expression is exactly the same as for the previous case except that the saturation scale now depends on momentum transfer @xmath139\ ] ] with @xmath140 we explain in appendix [ app : nonzero ] how the exponential behaviours required for travelling - waves formation appear and how the reference scale @xmath98 or @xmath137 emerges . naively , we can expect the saturation scale at large @xmath137 to be determined by @xmath137 as , in the tail of the amplitude , we are dominated by the hard part of the scattering . in addition , when @xmath141 we should recover the @xmath142-independent result and some soft reference scale should come back . it is very interesting to notice that the critical slope @xmath108 and speed @xmath107 , obtained from the bfkl kernel only , are the same as for the impact - parameter - independent case . the important point at this stage is that this study _ predicts _ geometric scaling at non - zero momentum transfer . experimental measurements of the dvcs cross - sections or diffractive @xmath143-meson production are good candidates and can provide more evidence for saturation . let us now consider the effect of the fluctuation contribution . this new term , added to the full balitsky hierarchy , turns a simple non - linear equation into an infinite hierarchy with a complicated transverse - plane dependence but the integration over the internal variable @xmath20 in is unknown . ] . we can however simplify the equation to a tractable problem by performing a local - fluctuations approximation . this amounts to simplify the dipole - dipole scattering amplitude @xmath63 used to relate the dipole density @xmath84 with the scattering amplitude @xmath29 . we assume that two dipoles interact if they are of the same size and if their centre - of - mass are sufficiently close to allow for an overlap of the two dipoles . within this approximation equation is replaced by @xmath144 where @xmath145 is an unknown fudge factor of order 1 . once this is done , the remaining steps are as follows : one fourier - transform the dipole size @xmath146 into momentum @xmath56 and perform a coarse - graining approximation to get rid of the impact - parameter dependence of the fluctuation term . this computation results into the following simplified form of the hierarchy ( we give only the expression for @xmath30 and @xmath48 for simplicity ) @xmath147 where , as previously , @xmath148 . again , this infinite hierarchy has the advantage that it can be rewritten under the form of a langevin equation @xmath149,\ ] ] where @xmath150 is a gaussian white noise satisfying the following commutation relations : @xmath151 & & \\[-3 mm ] { \left\langle\eta(l_1,y_1)\eta(l_2,y_2)\right\rangle } & = & \frac{4}{\bar\alpha}\delta(l_1-l_2)\delta(y_1-y_2 ) . \nonumber\end{aligned}\ ] ] the langevin equation is a stochastic equation describing an event - by - event picture . each realisation of the noise term corresponds to a particular evolution of the target and one can show ( see appendix [ app : langevin ] ) that once we average over those realisations using the correlations the complete hierarchy for the evolution of @xmath50 is recovered . it is interesting to notice that equation is formally equivalent to the bk equation with an additional noise term . as for the case of the bk equation , we can first restrict ourselves to the diffusive approximation . this leads to the stochastic fkpp ( sfkpp ) equation which amounts to add a noise term to the fkpp equation . it has found many applications in various fields , _ e.g. _ in the description of reaction - diffusion systems . the noise term appears once we have to consider discreteness effects ( _ e.g. _ for a reaction - diffusion system on a lattice , only an integer number of particles are admitted per site ) . when the number of particles involved goes to infinity , the mean - field approximation seems justified and the ( non - stochastic ) fkpp equation is valid . . the bk speed has been added for comparison . one sees that the speed decreases when @xmath152 increases . ] the basic effects of the additional noise term in the sfkpp equation are known , at least qualitatively . despite the fact that the fluctuations are only expected to have large consequences on the dilute tail of the wavefront , these modifications change the picture at saturation . in order to test the validity of those results in the case beyond the diffusive approximation , _ i.e. _ with the full bfkl kernel , we have performed @xcite numerical studies ( see also @xcite ) of the qcd equation . moreover , many analytical results are only known in the limit where the noise strength @xmath152 is ( irrealistically for qcd ) small while we have concentrated our numerical work of physically acceptable values . in the following paragraphs , we review the main effects of the fluctuation contribution and show that they are observed in the numerical analysis . first of all , for a single event , the evolved amplitude shows a travelling - wave pattern ( up to small fluctuations in the far tail which , at least a this level , are irrelevant for the wave pattern at saturation ) . this means that each single realisation of the noise leads to geometric scaling , as it is the case for the bk equation . at this level , the major difference comes from a decrease of the speed of the wave . analytically , this can only be computed in the limit @xmath153 _ i.e. _ when the fluctuations are a small perturbation around the mean - field behaviour . it has been found @xcite that @xmath154 unfortunately , this expression gives only reliable results for extremely small values of @xmath153 such as @xmath155 , which is not sufficient for realistic situations . the numerical simulations we have performed clearly exhibits this decrease of the speed when the noise strength @xmath152 increases ( see figure [ fig : speed ] ) . the second noticeable effect of the noise term is to introduce dispersion between different events . different events have the same shape but their position @xmath156 ( @xmath157 $ ] in physical variables ) fluctuates . at a given rapidity , one can compute the dispersion of those events . this diffusion process being very similar to a random walk , one expects that the dispersion of the events behaves like @xmath158 : @xmath159 the parametric dependence of the diffusion coefficient @xmath160 has been computed numerically @xcite and behaves . ] like @xmath161 for small @xmath152 . but few dispersion is obtained in early stages of the evolution . ] and @xmath162 . ] concerning the numerical simulations , the dispersion of the wavefront between different events and its increase with rapidity are manifest on figure [ fig : events ] . one can then extract the dispersion as a function of rapidity . the resulting curve , shown on figure [ fig : disp ] for different values of the noise strength , calls for two remarks . first , the expected asymptotic behaviour ( dispersion @xmath163 ) is observed and the diffusion coefficient @xmath160 increases with rapidity and with the noise strength . however , for small values of the rapidity , we do not observe a significant dispersion . this lack of dispersion may come from the fact that we have to wait for the travelling front to get formed before dispersion becomes significant . however , those effects in the beginning of the evolution are perhaps not universal and deserve more detailed studies . this dispersion of the events has an important physical consequence : although each single event displays geometric scaling , once we compute the average amplitude , dispersion induces geometric scaling violations . this effect is best seen on figure [ fig : front - avg ] where we have compared the evolution including the fluctuations to the bk results . while in the bk equation , a fixed travelling - wave pattern is formed , once fluctuations are taken into account , we see a broadening of the average amplitude as rapidity increases . those violations of geometric scaling can be studied @xcite in a model which encompasses all the physical arguments presented here - above and which shall prove to be very efficient for phenomenological studies . in order to reconstruct physical amplitudes , we need two building ingredients : * a front - by - front amplitude which displays geometric scaling . this amplitude has to satisfy unitarity in the infrared and must decrease exponentially in the ultraviolet . the simplest choice , which is actually sufficient for what follows , is @xmath164 & \text{if } \rho\ge\rho_s,\\ 1 & \text{if } \rho < \rho_s , \end{cases}\ ] ] where we have introduced the notations @xmath165 and @xmath166 $ ] . this expression shows indeed geometric scaling since it is a function of @xmath167 only . the precise value of @xmath168 is not relevant for what follows , although it might be adequate to adopt @xmath169 or @xmath170 . * a dispersion between different events . physically , this dispersion of the position of the events , i.e. of their saturation scale , means that the saturation scale becomes a random variable . we thus need a probability distribution for @xmath171 . we shall take a gaussian form @xmath172.\ ] ] the parameters of the gaussian ( the average saturation scale @xmath173 and the dispersion @xmath174 ) are expected to grow linearly with rapidity , but we shall treat them as parameters . the average amplitude can be constructed from and @xmath175 \exp\left(-\gamma_0z+\frac{\gamma_0 ^ 2\sigma^2}{2}\right),\end{aligned}\ ] ] with @xmath176 . in the last line , the first term comes from the event - by - event fronts which are at saturation ( @xmath177 ) while the second one comes from the exponential tail . practically , there are two interesting limits to analyse in more details : 1 . + when the dispersion is small ( compared to @xmath178 which is the typical decay length of the exponential tail ) , the dispersion can be neglected . we thus recover @xmath179 and geometric scaling is satisfied . since @xmath180 , this is what happens in early stages of the evolution . 2 . + for large dispersion , _ i.e. _ at very high energies , we find @xmath181 this expression is valid for @xmath182 which is arbitrarily large at high energy . from this analysis , we deduce two very important physical consequences for the high - energy behaviour of the amplitudes : * a new scaling law , called _ diffusive scaling _ , appears . the amplitude scales like @xmath183 , which , in more physical variables means that it only depends on the ratio @xmath184 . it is possible to show that this property of diffusive scaling extends to physical observables such that dis , diffractive dis @xcite and gluon production @xcite . * the error function in fully comes from the event - by - event fronts which are at saturation . in other words , the high - energy behaviour of the amplitudes can be obtained by taking @xmath185 . thus , at high energy , the scattering amplitudes are dominated by _ black spots _ , @xmath186 or @xmath177 . let us now summarise the main points raised in these proceedings . as this document is in itself some kind of a summary , i shall only pick up the most important points , refer to various references for detailed approaches and discuss open questions . first , we have shown that it is possible to describe the evolution to high energy in pqcd by an infinite hierarchy of equations ( see ) . this gives the evolution of the @xmath30 matrix ( averaged over the target wave - function ) and its higher - order correlations @xmath50 . in this hierarchy , the evolution of @xmath50 contains three types of contribution : 1 . the linear bfkl growth , proportional to @xmath187 . this is the usual high - energy contribution computed thirty years ago and corresponding to the exchange of @xmath56 pqcd pomerons . it leads to a fast ( exponential ) increase of the scattering amplitude . 2 . the saturation corrections , proportional to @xmath188 . this negative term becomes important when the amplitude reaches unitarity and it allows for the constraint @xmath189 to be satisfied . 3 . the fluctuations term , proportional to @xmath190 . these fluctuations correspond to gluon - number fluctuations in the target . they play an important role when the amplitude is of order @xmath73 or , equivalently , when the dipole density is of order one , _ i.e. _ in the dilute regime where one indeed expects fluctuation effects to appear . if one neglects the fluctuation contributions , we recover the balitsky hierarchy ( in its large-@xmath7 formulation ) and , in the mean field approximation , the bk equation is obtained . this last equation , much simpler than the infinite hierarchy captures many important points concerning the physics of saturation . we also have to point out that , as long as we do not take into account fluctuations , the evolution at all orders in @xmath191 exists under the form of the balitsky / jimwlk equation . the balitsky equations take the form of a hierarchy , similar to the one introduced in these proceedings but involving , in addition to dipoles , more complicated objects like quadrupoles , sextupoles , ... the jimwlk equation is an equivalent formulation based on the colour glass condensate approach , in which the probability density for the colour field of the target evolves through a functional equation . further information concerning the balitsky ( resp . cgc ) approach can be found in @xcite ( resp . @xcite ) . many additional things can be said concerning those different contributions , _ e.g. _ concerning the equivalence between descriptions from the target and projectile point of view ( see _ e.g. _ @xcite ) , or the attempts to describe the evolution beyond its large@xmath192 limit @xcite . those questions are active fields of research and the interested reader is forwarded to the corresponding references for further information . in the second part of this overview of high - energy qcd , we have concentrated ourselves on the physical consequences arising from saturation and fluctuation effects . we have investigated how those contributions manifest themselves on the scattering amplitude and change the exponential behaviour obtained from the bfkl evolution . again , the most important points can be summarised in two important steps which have their respective physical consequences : 1 . considering bfkl with saturation effects ( mean - field picture ) , one can show that the bk equation lies in the same universality class as the fkpp equation . this implies that travelling waves are formed during the evolution towards high energy . from the bfkl kernel one obtains the critical parameters which are the anomalous dimension observed in the large-@xmath1 tail and the speed of the wave , equivalent to the exponent of the saturation scale . in physical terms , these travelling waves correspond to geometric scaling . the experimental observation of geometric scaling in dis at hera can thus be seen as a consequence of saturation . it is extremely important to realise that geometric scaling is a prediction from saturation physics which extends to large values of @xmath1 , far beyond the saturation scale itself beyond the saturation scale . ] . this geometric scaling has also been predicted at nonzero momentum transfer , a result which may be applied for example to vector - meson production or dvcs . if one includes the effects of fluctuations , the evolution becomes a langevin equation equivalent to the stochastic fkpp equation , including a noise term . for each realisation of this noise , we observe geometric scaling with a decrease of the speed w.r.t . the mean - field case . if we consider a bunch of events , we observe that , although they all have the same shape , their position is diffused with diffusion increasing like @xmath158 . this dispersion of the saturation scale for a set of events implies that geometric scaling is violated for the averaged amplitude . recent studies indicate that geometric scaling is still valid at small rapidities . at higher energies , a new ( diffusive ) scaling is predicted @xcite in which the amplitude scales w.r.t . the scaling variable @xmath193/\sqrt{y}$ ] . although these two pictures endow most of the physical effects , they are far from being completely understood . among the improvements , one can quote phenomenological applications like geometric and diffusive scaling predictions for vector - meson productions , for diffraction @xcite and for lhc physics @xcite . on more theoretical grounds most of the work that still need to be done concerns the effects of fluctuations . most of the results known so far are derived within the local approximation for the noise term and neglecting the impact parameter dependence . in addition , apart from numerical studies , analytical results can only be applied to irrealistically small values of @xmath16 and a better understanding for more physical values is still lacking . although the existing picture is believed to contain all the qualitative effects , a more precise treatment is certainly an interesting and challenging problem to address . within the framework presented here , the link between the qcd evolution equations and equations from statistical physics has led to a large number of interesting ideas and results . even in statistical physics , many questions are yet opened , hence , extending the qcd picture including more sophisticated treatments of the bfkl kernel and more precise forms of the noise term is certainly not a straightforward task . the questions of making clear predictions for phenomenology , of understanding the effects of fluctuations beyond the local - noise approximation and of including impact - parameter dependence are hence expected to give interesting work in the near future . among particle physics , high - energy qcd is thus one of the most active fields . because of those open questions and because of the requirement for an excellent knowledge of qcd at the lhc , we expect that high - energy qcd is going to be even more active in the near future . * acknowledgements * i would like to thank warmly michal praszalowicz and andrzej bialas , for the invitation to give a course in the cracow school of theoretical physics . i also thank robi peschanski for a careful reading of this proceedings . is funded by the national funds for scientific research ( fnrs ) , belgium . let us show that it is also possible to obtain the bfkl equation by looking at the evolution of the dipole densities in the target . we hence need to compute how , within a small increase of rapidity , the density of dipoles of coordinates @xmath22 evolves . the answer is @xmath194 the three contribution in this equation has well determined origin : the first term correspond to the situation where , at rapidity @xmath31 , we had a dipole @xmath23 which splits into @xmath22 and @xmath195 hence the positive contribution to the evolution of @xmath34 . the second term has similar explanation with an original dipole @xmath24 and the third one corresponds to the situation where we had one dipole @xmath22 at rapidity @xmath31 which disappears through splitting into @xmath23 and @xmath24 , leading to a loss term coming with a minus sign . getting to the the evolution of the scattering amplitude when we probe that system with a dipole @xmath196 required a few technical manipulations . first , we shall use eq . to construct the scattering amplitude from the dipole density by convolution with the fundamental dipole - dipole interaction ( see ) is not really important , we shall only ask it to be conformal invariant . ] : @xmath197.\ ] ] the trick to recover is to use conformal properties . for example , for the first term in the equation , we perform the following change of variables is equivalent to the complex number @xmath198 . ] for the integration over @xmath19 : @xmath199 which has the effect to interchange @xmath3 with @xmath200 and @xmath201 with @xmath202 . it thus implies @xmath203 carrying on with the @xmath18 and @xmath20 integration will give @xmath204 renaming @xmath205 into @xmath20 and performing a similar treatment for the remaining terms , one recovers the bfkl equation . in this appendix , we show that a langevin equation with a noise term proportional to @xmath206 ( with @xmath150 a gaussian white noise ) , turns into the fluctuation term in the hierarchy for averaged amplitudes . to stay as close as possible to the problem we are concerned with , let us consider a sfkpp - like equation with one spatial dimension and a generic linear kernel : @xmath207 with @xmath208 and @xmath209 . as usual , the langevin equation has to be taken with the ito prescription . this means that we consider discrete time steps @xmath210 and write @xmath211 with the r.h.s . taken at time @xmath212 and the following noise correlators : @xmath213 we can then consider the time evolution of any functional @xmath214 $ ] of the event - by - event amplitude @xmath215 . by series expansion and using the langevin equation , we have @xmath216 & = & f[u_i(x)]\\ & + & \int dx\ , \frac{\delta f[u_i(x)]}{\delta u_i(x ) } \ , \delta [ \chi(\partial_x ) u_i(x ) - u_i^2(x ) + \sqrt{2\sigma\,u_i(x)}\,\eta_i(x)]\\ & + & \frac{1}{2 } \int dx_1 dx_2\ , \frac{\delta^2 f[u_i(x)]}{\delta u_i(x_1)\,\delta u_i(x_2 ) } \\ & & \phantom{1intdx_1dx_2\ , } \delta [ \chi(\partial_{x_1 } ) u_i(x_1 ) - u_i^2(x_1 ) + \sqrt{2\sigma\,u_i(x_1)}\,\eta_i(x_1)]\\ & & \phantom{1intdx_1dx_2\ , } \delta [ \chi(\partial_{x_2 } ) u_i(x_2 ) - u_i^2(x_2 ) + \sqrt{2\sigma\,u_i(x_2)}\,\eta_i(x_2)].\end{aligned}\ ] ] we want to obtain the evolution of the average amplitude @xmath217\right\rangle}$ ] . for that , we take the average in the previous equation and use the correlators of the noise term : @xmath218\right\rangle}-{\left\langlef[u_i(x)]\right\rangle}}{\delta}}\\ & = & \int dx\ , { \left\langle\frac{\delta f[u_i(x)]}{\delta u_i(x ) } \ , [ \chi(\partial_x ) u_i(x ) - u_i^2(x)]\right\rangle}\\ & + & \frac{1}{2 } \int dx_1 dx_2\ , { \left\langle\frac{\delta^2 f[u(x)]}{\delta u(x_1)\,\delta u(x_2 ) } \ , \delta \,2\sigma\,\sqrt{u_i(x_1)u_i(x_2)}\,\frac{1}{\delta}\delta(x_1-x_2)\right\rangle},\end{aligned}\ ] ] where we have only kept the leading terms in @xmath210 . one can finally take the continuum limit for the time variable and obtain @xmath219\right\rangle}}\\ & = & \int dx\ , { \left\langle\frac{\delta f[u(x , t)]}{\delta u(x , t ) } \ , [ \chi(\partial_x ) u(x , t ) - u^2(x , t ) ] + \frac{\delta^2 f[u(x , t)]}{[\delta u(x , t)]^2 } \ , \sigma\,u(x , t)\right\rangle}.\end{aligned}\ ] ] in particular , one can consider the case @xmath220 = u^k(x)$ ] , from which one recovers the infinite hierarchy of equations @xmath221 in this expression , we clearly see that the last term , responsible for fluctuations , directly comes from the noise term in the langevin equation . the two remaining ones correspond to the balitsky hierarchy _ i.e. _ linear evolution nd saturation corrections . let us show in more details how the reference scale for the saturation scale at nonzero momentum transfer moves from @xmath222 at large @xmath91 to a soft scale @xmath223 when @xmath91 becomes small . from the condition for travelling - waves formation , we need to satisfy the requirement that the amplitude in the dilute tail ( described by the linear equation ) can be expressed as a superposition of waves . hence , our starting point is the generic solution for projectile - target scattering in the bfkl limit : @xmath224 where @xmath225 is the momentum of the projectile dipole , @xmath134 is the momentum transfer and @xmath226 is a typical scale for the target . the crucial point at this stage of the analysis is that , in this momentum space representation , we have a factorisation between a target and a projectile contribution . since we are interested in the dilute tail of the scattering amplitude , the scale of the projectile is bigger than @xmath227 , @xmath91 and @xmath228 . in that limit , we know the asymptotic behaviour of the bfkl eigenfunctions @xmath229 on the target side , two limits are possible depending on the ordering between @xmath91 and @xmath228 . again , we can expand the bfkl eigenfunctions to get appearing when @xmath230 instead of @xmath80 when @xmath231 comes from conformal properties of the bfkl equation . ] @xmath232 reinserting everything back into the expression for the amplitude , we obtain the required exponential behaviour in the two limits : @xmath233 this means that the reference scale for the saturation scale is fixed by the hardest between the target soft scale @xmath228 and the momentum transfer @xmath91 as anticipated from the beginning . l. n. lipatov , _ sov . * 23 * , ( 1976 ) 338 ; e. a. kuraev , l. n. lipatov and v. s. fadin , _ sov . . jetp _ * 45 * , ( 1977 ) 199 ; i. i. balitsky and l. n. lipatov , _ sov . j. nucl _ * 28 * , ( 1978 ) 822 . i. i. balitsky , nucl . * b463 * , 99 ( 1996 ) [ arxiv : hep - ph/9509348 ] ; j. jalilian - marian , a. kovner , a. leonidov and h. weigert , nucl . b * 504 * ( 1997 ) 415 [ arxiv : hep - ph/9701284 ] ; e. iancu , a. leonidov and l. d. mclerran , nucl . a * 692 * ( 2001 ) 583 [ arxiv : hep - ph/0011241 ] ; h. weigert , nucl . a * 703 * ( 2002 ) 823 [ arxiv : hep - ph/0004044 ] . y. v. kovchegov , phys . rev . * d60 * , 034008 ( 1999 ) [ arxiv : hep - ph/9901281 ] ; y. v. kovchegov , phys . rev . * d61 * , 074018 ( 2000 ) , [ arxiv : hep - ph/9905214 ] . a. h. mueller and a. i. shoshi , nucl . phys . b * 692 * ( 2004 ) 175 [ arxiv : hep - ph/0402193 ] ; e. iancu , a. h. mueller and s. munier , phys . b * 606 * ( 2005 ) 342 [ arxiv : hep - ph/0410018 ] . e. iancu and d. n. triantafyllopoulos , nucl . phys . a * 756 * ( 2005 ) 419 [ arxiv : hep - ph/0411405 ] . b * 610 * ( 2005 ) 253 [ arxiv : hep - ph/0501193 ] . a. h. mueller , a. i. shoshi and s. m. h. wong , nucl . b * 715 * ( 2005 ) 440 [ arxiv : hep - ph/0501088 ] . s. munier and r. peschanski , phys . * 91 * , 232001 ( 2003 ) [ arxiv : hep - ph/0309177 ] ; phys . rev . * d69 * , 034008 ( 2004 ) [ arxiv : hep - ph/0310357 ] ; phys . rev . * d70 * , 077503 ( 2004 ) [ arxiv : hep - ph/0310357 ] . a. m. stato , k. golec - biernat , and j. kwiecinski , phys * 86 * , 596 ( 2001 ) [ arxiv : hep - ph/0007192 ] . c. marquet and g. soyez , nucl . a * 760 * ( 2005 ) 208 [ arxiv : hep - ph/0504080 ] . c. marquet , r. peschanski and g. soyez , nucl . a * 756 * ( 2005 ) 399 [ arxiv : hep - ph/0502020 ] . g. soyez , phys . d * 72 * ( 2005 ) 016007 [ arxiv : hep - ph/0504129 ] . r. enberg , k. golec - biernat and s. munier , phys . d * 72 * ( 2005 ) 074021 [ arxiv : hep - ph/0505101 ] . c. marquet , r. peschanski and g. soyez , phys . d * 73 * ( 2006 ) 114005 [ arxiv : hep - ph/0512186 ] . e. brunet and b. derrida , phys . * e57 * , 2597 ( 1997 ) [ arxiv : cond - mat/0005362 ] ; comp . * 121 * 376 ( 1999 ) [ arxix : cond - mat/0005364 ] e. brunet , b. derrida , a. h. mueller and s. munier , phys . e * 73 * ( 2006 ) 056126 [ arxiv : cond - mat/0512021 ] . y. hatta , e. iancu , c. marquet , g. soyez and d. n. triantafyllopoulos , nucl . a * 773 * ( 2006 ) 95 [ arxiv : hep - ph/0601150 ] . c. marquet , g. soyez and b. w. xiao , phys . b * 639 * ( 2006 ) 635 [ arxiv : hep - ph/0606233 ] . e. iancu , c. marquet and g. soyez , arxiv : hep - ph/0605174 . a. h. mueller , _ nucl . phys . _ * b415 * ( 1994 ) 373 ; a. h. mueller and b. patel , _ nucl . _ * b425 * ( 1994 ) 471 ; a. h. mueller , _ nucl . phys . _ * b437 * ( 1995 ) 107 . e. iancu , g. soyez and d. n. triantafyllopoulos , nucl . a * 768 * ( 2006 ) 194 [ arxiv : hep - ph/0510094 ] . l. n. lipatov , sov . jetp * 63 * ( 1986 ) 904 [ zh . * 90 * ( 1986 ) 1536 ] ; h. navelet and r. peschanski , nucl . b * 507 * ( 1997 ) 353 [ arxiv : hep - ph/9703238 ] . c. marquet , contribution to these proceedings . i. balitsky , arxiv : hep - ph/0101042 . e. iancu , a. leonidov and l. d. mclerran , nucl . phys . a * 692 * ( 2001 ) 583 [ arxiv : hep - ph/0011241 ] ; e. ferreiro , e. iancu , a. leonidov and l. mclerran , nucl . a * 703 * ( 2002 ) 489 [ arxiv : hep - ph/0109115 ] . a. kovner and m. lublinsky , phys . d * 72 * ( 2005 ) 074023 [ arxiv : hep - ph/0503155 ] . a. kovner and m. lublinsky , phys . d * 71 * ( 2005 ) 085004 [ arxiv : hep - ph/0501198 ] , phys . * 94 * ( 2005 ) 181603 [ arxiv : hep - ph/0502119 ] ; y. hatta , e. iancu , l. mclerran and a. stasto , nucl . a * 762 * ( 2005 ) 272 [ arxiv : hep - ph/0505235 ] ; y. hatta , e. iancu , l. mclerran , a. stasto and d. n. triantafyllopoulos , nucl . a * 764 * ( 2006 ) 423 [ arxiv : hep - ph/0504182 ] .

in ref . [ physica a * 296 * , 405 ( 2001 ) ] , starting from the one parameter deformation of the exponential function , a statistical mechanics has been constructed which reduces to the ordinary boltzmann - gibbs statistical mechanics as the deformation parameter approaches to zero . the distribution obtained within this statistical mechanics shows a power law tail and depends on the non specified parameter , containing all the information about the temperature of the system . on the other hand , the entropic form , which after maximization produces the distribution and reduces to the standard boltzmann - shannon entropy as , contains the coefficient whose expression involves , beside the boltzmann constant , another non specified parameter . in the present effort we show that is the unique existing entropy obtained by a continuous deformation of and preserving unaltered its fundamental properties of concavity , additivity and extensivity . these properties of permit to determine unequivocally the values of the above mentioned parameters and . subsequently , we explain the origin of the deformation mechanism introduced by and show that this deformation emerges naturally within the einstein special relativity . furthermore , we extend the theory in order to treat statistical systems in a time dependent and relativistic context . then , we show that it is possible to determine in a self consistent scheme within the special relativity the values of the free parameter which results to depend on the light speed and reduces to zero as recovering in this way the ordinary statistical mechanics and thermodynamics . the novel statistical mechanics here presented preserves unaltered the mathematical and epistemological structure of the ordinary statistical mechanics and is suitable to describe a very large class of experimentally observed phenomena in low and high energy physics and in natural , economic and social sciences . finally , in order to test the correctness and predictability of the theory , as working example we consider the cosmic rays spectrum , which spans 13 decades in energy and 33 decades in flux , finding a high quality agreement between our predictions and observed data .

cond-mat0210467

the following one - parameter deformations of the exponential and logarithm functions @xmath13 which reduce to the standard exponential and logarithm respectively as the real deformation parameter @xmath1 approaches to zero , has been introduced recently in ref . the above functions have many very interesting properties @xcite ( some being identical to the ones of the undeformed functions ) that permit to construct a statistical mechanics ( and thermodynamics ) which generalizes the standard boltzmann - gibbs one . this @xmath1-deformed statistical mechanics preserves unaltered the structure of the ordinary one and can be used to explain a very large class of experimentally observed phenomena described by distribution functions exhibiting power law tails . the areas where this formalism can be applied include among others , low and high energy physics , astrophysics , econophysics , geology , biology , mathematics , information theory , linguistics , etc . @xcite . in ref . @xcite it has been shown that the statistical distribution @xmath14\ , ) \ , \label{i3}\end{aligned}\ ] ] which generalizes the boltzmann - gibbs distribution , can be obtained also by maximizing , after properly constrained , the entropy @xmath15 which reduces to the standard entropy @xmath6 as the deformation parameter approaches to zero . the coefficient @xmath8 , which also absorbs the boltzmann constant @xmath16 , depends on a free parameter @xmath9 ( see eq . ( 65 ) of ref . @xcite ) which remains to be determined together with the parameter @xmath3 which contains the information about the temperature @xmath17 of the system . a first question which arises naturally is if it is possible and how to find any criterion which allows us to fix the parameters @xmath3 and @xmath9 or at least express these in terms of the deformation parameter @xmath1 , in order to reduce the free parameters of the theory . a second question regards the properties of the entropy @xmath10 . it is well known that the boltzmann - shannon entropy @xmath6 is concave , additive and extensive . we know that @xmath10 is concave with respect to the variable @xmath5 , but what happens about its additivity and extensivity ? more in general , beside the boltzmann - shannon entropy other concave , additive and extensive entropies do exist ? a third question is related to the physical mechanism which originates the deformation introduced by the parameter @xmath1 . in other words , does a more fundamental theory exist where this deformation emerges , or is it simply a purely mathematical tool ? a fourth question is if it is possible to extend the theory originally developed in the framework of a classical kinetics to treat statistical systems in the context of a relativistic kinetics . a fifth question regards the deformation parameter @xmath1 . this parameter will continue to remain free or is it possible to determine its value self consistently within the theory ? the present paper deals with the above questions and its purpose is double . firstly , we will show that @xmath10 is the unique existing , concave , additive and extensive entropy , beside the boltzmann - shannon entropy . as we will see these properties of @xmath10 are sufficient to determine unequivocally the values of parameters @xmath3 and @xmath9 . secondly , we will show that the deformation introduced by @xmath1 is a purely relativistic effect and then we will explain the deformation mechanism within the einstein special relativity . then , we will formulate a relativistic @xmath1-kinetics and we will calculate the value of @xmath1 . finally , in order to test the predictability and correctness of theory here proposed we will consider two sets of experimental data . in particular we will analyze the cosmic rays spectrum which spans 13 decades in energy and 33 decades in particle flux that , as it is widely known , violates the boltzmann - gibbs statistics . as we will see , we have a high quality agreement between the theory and the observed data . the paper is organized as it follows . in sect . ii , generalizing the approach proposed in ref.s @xcite , we introduce a class of one parameter deformed structures and study their mathematical properties . in sect . iii , within this context , starting from the jaynes maximum entropy principle we consider the most general class of deformed statistical mechanics preserving the main features of the standard boltzmann - gibbs one . in sect . iv , we show that the entropy @xmath10 introduced in ref . @xcite is the only one existing beside the boltzmann - shannon entropy @xmath6 which is simultaneously concave , additive and extensive . then the statistical mechanics and thermodynamics based on @xmath10 can be viewed as a natural extension of the boltzmann - gibbs one , recovering this last as the deformation parameter @xmath1 approaches to zero . in sect . v , we consider the mean properties of @xmath1-exponential and @xmath1-logarithm which have a fundamental role in the formulation of the new statistical mechanics . in sect . vi , we extend the formalism to a time dependent and relativistic context . in particular after introducing the relativistic @xmath1-kinetic evolution equation we study its stationary state and prove the h - theorem . in sect . vii , we explain the origin of the @xmath1-deformation and show that it emerges naturally within the einstein special relativity . in sect . viii , we propose an approach which permits to determine within the special relativity the value of the parameter @xmath1 . in sect . ix , we compare two sets of experimental data with the predictions of the present theory . finally in sect . x , some concluding remarks are reported . let @xmath18 be an arbitrary real function of the real variable @xmath19 , that we call generator of the deformation , having the following properties \i ) @xmath20 ; \ii ) @xmath21 ; \iii ) @xmath22 ; \iv ) @xmath23 ; \v ) @xmath24 , for @xmath25 . starting from the generator @xmath18 , we construct the real function @xmath26 of the real variable @xmath19 and depending on the real parameter @xmath1 , as follows @xmath27 its properties descent directly from the ones of the generator @xmath18 : \i ) @xmath28 ; \ii ) @xmath29 ; \iii ) @xmath30 ; \iv ) @xmath31 ; \v ) @xmath32 , for @xmath25 and then @xmath33 ; \vi ) @xmath32 , for @xmath34 and then @xmath35 ; \vii ) @xmath36 . together with the function @xmath26 one can introduce the inverse function @xmath37 , through @xmath38 , which assumes the form @xmath39 _ proposition 1 : _ the composition law @xmath40 defined through @xmath41 which reduces to the ordinary sum as @xmath42 , namely @xmath43 , is a deformed sum and the algebraic structure @xmath44 forms an abelian group . _ proof : _ indeed from the definition of @xmath45 the following properties of @xmath40 follow \1 ) associativity property : @xmath46 ; \2 ) neutral element : @xmath47 ; \3 ) opposite element : @xmath48 ; \4 ) commutativity property : @xmath49 . of course the @xmath1-difference indicated with @xmath50 is defined as @xmath51 . _ proposition 2 : _ the composition law @xmath52 defined through @xmath53 which reduces to the ordinary product as @xmath42 , namely @xmath54 , is a deformed product and the algebraic structure @xmath55 forms an abelian group . _ proof : _ indeed from the definition of @xmath45 we have the following properties \1 ) associative law : @xmath56 ; \2 ) neutral element : @xmath57 ; \3 ) inverse element : @xmath58 ; \4 ) commutative law : @xmath59 ; being the neutral element @xmath60 while the inverse element of @xmath19 is @xmath61 . of course the @xmath1-division @xmath62 is defined as @xmath63 . _ proposition 3 : _ the deformed sum @xmath40 and product @xmath52 obey the distributive law @xmath64 and then the algebraic structure @xmath65 forms an abelian field . _ proof : _ this proposition follows from the ( [ ii3 ] ) and ( [ ii4 ] ) . we remark that the field @xmath66 is isomorphic with the field @xmath67 . moreover @xmath68 and then the structure @xmath69 it is not a field . _ proposition 4 : _ the function @xmath37 has the two following properties @xmath70 _ proof : _ these properties follow directly from the definitions given by eq.s ( [ ii3 ] ) and ( [ ii4 ] ) . _ proposition 5 : _ the function @xmath45 and its inverse @xmath37 obey the following scaling laws : @xmath71 with @xmath72 _ proof : _ these laws follow from the definitions of the functions @xmath26 and @xmath37 . _ proposition 6 : _ the pseudodistributive law @xmath73 holds and then the structure @xmath69 is a pseudofield . _ proof : _ indeed by using the propositions @xmath74 and @xmath75 one obtains @xmath76 consider the two algebraic structures @xmath77 and @xmath78 with @xmath79 and @xmath80 . let us introduce the set of the functions @xmath81 with @xmath82 . the @xmath1-differential @xmath83 is defined as @xmath84 and results to be @xmath85 we define the @xmath1-derivative for the functions of the set @xmath86 through @xmath87 with @xmath88 and @xmath89 . we observe that the @xmath90-derivative , which reduces to the usual one as the deformation parameter @xmath91 , can be written in the form @xmath92 from which clearly it appears that the @xmath90-derivative is governed by the same rules of the ordinary one . the @xmath1-exponential @xmath93 is defined as eigenstate of the @xmath1-derivative @xmath94 and is given by @xmath95 it results that @xmath96 the @xmath1-exponential , just as the ordinary exponential , has the properties @xmath97 furthermore , the @xmath1-exponential has the two properties @xmath98 with @xmath99 , and can be expressed in terms of the generator @xmath18 as @xmath100^{1/\kappa}\ \ . \label{ii31}\ ] ] the @xmath1-logarithm @xmath101 is defined as the inverse function of the of @xmath1-exponential , namely @xmath102 , and is given by @xmath103 it results that @xmath104 the @xmath1-logarithm , just as the ordinary logarithm , has the properties @xmath105 furthermore , the @xmath1-logarithm has the two properties @xmath106 with @xmath99 , and can be expressed in terms of the generator @xmath18 as @xmath107 eq . ( [ ii45 ] ) defines a very large class of deformed logarithms varying the arbitrary function @xmath18 . these deformed logarithms can depend on many other parameter ( through the generator @xmath18 ) besides the parameter @xmath1 . we recall briefly that in literature one can find other one @xcite or two @xcite parameter deformations of the exponential and logarithm functions . anyway in following we will consider the deformed logarithms defined through eq . ( [ ii45 ] ) and depending only on the parameter @xmath1 . we define the @xmath1-hyperbolic sine and cosine @xmath108 \ \ , \label{ii46 } \\ \cosh_{_{\{{\scriptstyle \kappa}\}}}(x ) = \frac{1}{2}\left[\exp_{_{\{{\scriptstyle \kappa}\}}}(x ) + \exp_{_{\{{\scriptstyle \kappa}\}}}(-x)\right ] \ \ , \label{ii47}\end{aligned}\ ] ] starting from the @xmath1-euler formula @xmath109 it is straightforward to introduce the @xmath1-hyperbolic trigonometry , which reduces to the ordinary one as @xmath7 . for instance , the formulas @xmath110 still hold true . all the formulas of the ordinary hyperbolic trigonometry still hold true after properly deformed . the deformation of a given formula can be obtained starting from the corresponding undeformed formula , and then by making in the argument of the hyperbolic trigonometric functions the substitutions @xmath111 , and obviously @xmath112 ( n times ) . for instance it results @xmath113 @xmath114 and so on . the @xmath1-de moivre formula involving hyperbolic trigonometric functions having arguments of the type @xmath115 with @xmath116 , assumes the form @xmath117^{r}\!\!=\!\cosh_{_{\{{\scriptstyle \kappa}/{\scriptstyle r}\}}}\!(r x)\!\pm\!\sinh_{_{\{{\scriptstyle \kappa}/{\scriptstyle r}\}}}\!(r x ) \ . \label{ii52}\ ] ] also the formulas involving the derivatives of the hyperbolic trigonometric function still hold , after properly deformed . for instance we have @xmath118 ^ 2 } \label{ii54 } \ \ , \end{aligned}\ ] ] and so on . the @xmath1-cyclic trigonometry can be constructed analogously . the @xmath1-sine and @xmath1-cosine is defined as @xmath119 we remark that it results : @xmath120 and @xmath121 . the @xmath1-inverse hyperbolic or cyclic trigonometric functions can be introduced starting from the corresponding direct functions just as in the case of the undeformed mathematics . it is trivial to verify that @xmath1-inverse functions are related to the @xmath1-logarithm by the usual formulas of standard mathematics . for instance we have @xmath122 and so on . let us consider the set of the non negative real functions @xmath123 . _ proposition 7 : _ the composition law @xmath124 defined through @xmath125 which reduces to the ordinary product as @xmath42 , namely @xmath126 , is a deformed product and the algebraic structure @xmath127 forms an abelian group . _ proof : _ indeed this product has the following properties + 1 ) associative law : @xmath128 ; + 2 ) neutral element : @xmath129 ; + 3 ) inverse element : @xmath130 ; + 4 ) commutative law : @xmath131 . of course the division @xmath132 can be defined through @xmath133 . the deformed @xmath1-power @xmath134 is defined through @xmath135 and generalizes the ordinary power @xmath136 . in particular , when @xmath137 is integer one has @xmath138 , ( r times ) . _ proposition 8 : _ the algebraic structure @xmath139 forms an abelian monoid . _ proof : _ indeed the element @xmath140 does not admit an inverse element . furthermore , just as in the case of the ordinary product , it results @xmath141 . _ proposition 9 : _ the composition law @xmath142 defined through @xmath143\right\ } \ , \label{ii59}\ ] ] which reduces to the ordinary sum as the deformation parameter approaches to zero , namely @xmath144 , is a deformed sum and the algebraic structure @xmath145 forms an abelian monoid . _ proof : _ indeed this sum has the following properties + 1 ) associative law : @xmath146 ; + 2 ) neutral element : @xmath147 ; + 3 ) commutative law : @xmath148 . we remark that the product @xmath149 and sum @xmath142 are distributive operations @xmath150 . the product @xmath149 allows us to write the following property of the @xmath1-exponential @xmath151 equivalently eq . ( [ ii60 ] ) can be written also in the form @xmath152 eq . ( [ ii61 ] ) gives a relevant property for the @xmath1-logarithm . finally , starting from the definition of the @xmath1-power @xmath134 , we obtain the following relation @xmath153 the relations given by eq.s ( [ ii61 ] ) and ( [ ii62 ] ) , which express two mathematical properties of the @xmath1-logarithm , will be very useful in the next section in defining a new additive and extensive entropy . let us consider the following non normalized statistical distribution involving the @xmath154exponential @xmath155 we write the real non specified parameter @xmath3 as @xmath156 being @xmath157 a new real parameter , @xmath158 the boltzmann constant and @xmath17 the temperature of the system . in the following it will be useful to introduce the distribution @xmath159 which is related with @xmath5 through @xmath160 , being @xmath9 another new real parameter which will be determined together with @xmath157 in the following . we recall that in the ordinary statistical mechanics the mean value of a given physical quantity @xmath161 depending on the variable @xmath162 and the distribution @xmath163 , is defined as @xmath164 analogously , in the case where @xmath165 depends on two independent variables @xmath166 , @xmath167 and on the two independent distribution functions @xmath168 and @xmath169 , we have that the mean value is given by @xmath170 it is easy to verify that the stationary distribution @xmath171 can be obtained as solution of the following variational equation @xmath172=0\ \ . \label{iii6}\ ] ] then the distribution @xmath171 can be viewed as maximizing the information content @xmath173 @xmath174 under the constraints @xmath175 @xmath176 imposing the conservation of the mean energy and of the particle number respectively . note that the chemical potential @xmath177 should be chosen in such a way to set the particle number equal to unity namely @xmath178 @xcite . we observe that when @xmath179 it results @xmath180 and the information content @xmath181 is the mean value of the ordinary logarithm . in this case the above variational equation expresses the jaynes maximum entropy principle which conducts to the boltzman - gibbs statistical mechanics . in the following , in analogy with the standard statistical mechanics , we require that @xmath173 must be expressed as the mean value of @xmath182 . to do so we must consider the subclass of the deformed logarithms obeying the condition @xmath183 the above condition , as we will see in the following , allows the simultaneous determination of both the form of @xmath1-logarithm and the values of the free parameters @xmath9 and @xmath157 . firstly we observe that this class contains the standard logarithm @xmath184 , for which results @xmath179 , @xmath185 , @xmath186 . it will be the task of the next section to investigate on the existence of new additional solutions of eq . ( [ iii9 ] ) , beside the standard logarithm . taking into account this condition we can write the variational equation ( [ iii6 ] ) in the form @xmath187 we define the @xmath1-entropy through @xmath188 so that @xmath10 can be viewed as proportional to the mean value of the @xmath182 , namely @xmath189 in this definition of @xmath10 we have a perfect analogy with the shannon entropy @xmath6 which is the proportional to the mean value of the @xmath190 . it is remarkable that in both the definitions of @xmath10 and @xmath6 appears the standard mean value given by eq . ( [ iii7 ] ) . ( [ iii10 ] ) assumes the form @xmath191 and then @xmath192 the above variational equation can be viewed as defining a maximum entropy principle analogous of the jaynes principle of the standard boltzmann - gibbs statistical mechanics @xcite . we remark that this maximum entropy principle , in the form given by eq . ( [ iii10 ] ) , holds only and exclusively for the subclass of @xmath1-logarithms , which are solutions of the integral equation ( [ iii9 ] ) . we show now that the families of entropies , defined through ( [ iii12 ] ) and involving the @xmath1-logarithms which are solutions of eq . ( [ iii9 ] ) , have two important properties typical of the shannon entropy . to do so we consider the properties ( [ ii61 ] ) and ( [ ii62 ] ) of @xmath1-logarithm , which rearrange as @xmath193 with @xmath194 and @xmath195 . when the systems @xmath74 and @xmath196 described through @xmath197 and @xmath198 respectively are statistically independent , and after taking into account the definitions of the mean values ( [ iii7 ] ) , ( [ iii8 ] ) and of the @xmath1-entropy ( [ iii13 ] ) , the two above properties of @xmath1-logarithm transform into the following properties for the @xmath1-entropy @xmath199 + s_{\kappa}[n_2]= s_{\kappa}[n_{12 } ] \ \ , \label{iii16 } \\ & & r\,s_{\kappa}[n]= s_{\kappa}[n^ { * } ] \ \ \label{iii17 } .\end{aligned}\ ] ] eq.s ( [ iii16 ] ) and ( [ iii17 ] ) say that the entropies @xmath10 defined starting from the @xmath1-logarithms which are solutions of eq . ( [ iii9 ] ) are additive and extensive just as the shannon entropy . the distribution @xmath200 describes the composite system obtained starting from the systems @xmath74 and @xmath196 while @xmath201 describes the scaled system related to the system described through @xmath171 . note that the state described through the distribution @xmath200 is different with respect to the state described through the distribution @xmath202 resulting @xmath203\leq s_{\kappa } [ n_1 n_2]$ ] . finally , from the concavity property of the deformed @xmath1-logarithm the concavity of @xmath10 , follows @xmath204\geq t\,s_{\kappa } [ n_1]+ ( 1-t ) s_{\kappa } [ n_2 ] \ \ , \label{iii18}\end{aligned}\ ] ] with @xmath205 . as we have already noted , the ordinary logarithm is solution of eq . ( [ iii9 ] ) and then the shannon entropy @xmath206=-k_{_b}\int d^3 p \ , \,\ , n\,\ln n \ \ , \label{iii19}\ ] ] which is additive and extensive ( @xmath207 and @xmath208 ) , is admitted within the present formalism . in the next section we will show that eq . ( [ iii9 ] ) admits a new ( only one ) less evident solution . then beside the shannon entropy we have a new concave , additive and extensive entropy which is the @xmath1-entropy proposed in ref . @xcite . we consider eq . ( [ iii9 ] ) which , after performing a derivation with respect to @xmath171 , assumes the form @xmath209 in the following we will determine the explicit form of @xmath182 by solving this differential - functional equation . we recall that @xmath182 can be expressed in terms of the generator function according to eq . ( [ ii45 ] ) so that eq . ( [ iv1 ] ) becomes @xmath210 in the above equation the function to be determined is now the generator function @xmath211 . to do so we make the following changes of variables @xmath212 so that eq . ( [ iv2 ] ) assumes the following simple form @xmath213 the property ( v ) of the generator @xmath18 imposes that @xmath214 obeys the two conditions @xmath215 and @xmath216 . these conditions , if combined with eq . ( [ iv6 ] ) , can be equivalently written in the form @xmath217 it is more convenient to take into account these conditions and write eq . ( [ iv6 ] ) under the form @xmath218 after recalling the property ( ii ) of @xmath18 , which imposes that @xmath219 , eq . ( [ iv9 ] ) can be written as @xmath220 let us introduce the new function @xmath221 . we can see that eq . ( [ iv10 ] ) , after deriving with respect to @xmath11 , transforms into the following functional equation @xmath222 finally the nonlinear transformation defined through @xmath223 permits us to write the last equation under the form @xmath224+\cosh \big[\xi(c)-\xi(t)\big ] \nonumber \\ & & = \cosh \big[\xi(c+t)\big]+\cosh \big[\xi(c - t)\big ] \ \ . \label{iv12}\end{aligned}\ ] ] it is trivial to verify that the most general solution of eq . ( [ iv12 ] ) is given by @xmath225 with @xmath137 an arbitrary real parameter . _ shannon solution _ : we note that in the case @xmath226 we obtain @xmath227 and then @xmath228 . this is the well known standard logarithm which , inserted in eq . ( [ iii12 ] ) , produces the shannon entropy . _ the new solution _ : we consider now the case @xmath229 . it is easy to realize that in this case the generator @xmath18 assumes the form @xmath230 \ \ . \label{iv14}\end{aligned}\ ] ] for simplicity of the exposition we firstly discuss the case @xmath231 for which @xmath232 . after some simple calculations we obtain that only when @xmath233 it is possible to determine the real constants @xmath157 and @xmath9 obtaining @xmath234 in this case the generator @xmath232 imposes the following expressions for the deformed logarithm and exponential @xmath235 for the general case where @xmath236 we obtain the same solution given by eq.s ( [ iv15])-([iv19 ] ) with the only difference that in place of @xmath1 now the scaled parameter @xmath237 appears . then we can set @xmath231 without loosing the generality of the theory . moreover for @xmath179 we obtain the shannon solution as a particular and limiting case of the new solution . finally , after inserting the expression of the @xmath1-logarithm given by eq . ( [ iv18 ] ) into eq . ( [ iii12 ] ) , we can write the new additive and extensive entropy in the following simple form @xmath238=-\,k_{_b}\int d^3p\,\,\,\frac{n^{1+\kappa}-n^{1-\kappa}}{2\kappa } \ \ . \label{iv20}\end{aligned}\ ] ] we can write the entropy @xmath10 also in terms of the distribution @xmath5 obtaining @xmath239=-k_{_b}\int d^3p\,\,\,\left ( c_{\kappa}f^{1+\kappa } + c_{-\kappa}f^{1-\kappa}\right ) \ \ \label{iv21}\end{aligned}\ ] ] where the coefficient @xmath240 depends exclusively on the deformation parameter @xmath1 and is given by @xmath241 the above entropy is contained , as a particular case , in the class of entropies introduced previously in ref . @xcite ( in eq . ( 65 ) of this reference it appears the non specified parameter @xmath9 , while the boltzmmann constant is absent because setted to have @xmath242 ) . we recall that the entropy given by eq . ( [ iv20 ] ) is different from the nonextensive entropy introduced in ref.s @xcite . of course the statistical distribution defined through eq.s ( [ iii1 ] ) and ( [ iv19 ] ) , introduced previously in ref . @xcite , is also different from the distribution of the nonextensive statistics @xcite and of the plasmas physics @xcite . let us report here the main mathematical properties , some of these reported in ref.s @xcite , of the functions @xmath243 and @xmath101 defined through eq.s ( [ iv18 ] ) and ( [ iv19 ] ) , respectively . we start by observing that the generator of the deformation is the function @xmath232 and then from ( [ ii1 ] ) and ( [ ii2 ] ) we obtain @xmath244 we have also ( @xcite , p. 58 ) @xmath245 the following function @xmath246=\frac{q^x - q^{-x } } { q - q^{-1 } } \ \ , \label{v4}\end{aligned}\ ] ] which has a central role in quantum group theory @xcite , results to be proportional to @xmath247 , namely we have the relation @xmath246=\frac{1 } { \ln_{_{\{{\scriptstyle \kappa}\}}}\!(e ) } \,\,x^{\scriptscriptstyle{\{}{\scriptstyle\kappa}\scriptscriptstyle{\ } } } \ \ , \label{v5}\end{aligned}\ ] ] which can be written also in the form @xmath246=\frac{\ln_{_{\{{\scriptstyle \kappa}\}}}\!(e^x ) } { \ln_{_{\{{\scriptstyle \kappa}\}}}\!(e ) } \ \ , \label{v6}\end{aligned}\ ] ] with @xmath248 . we note that the well known symmetry of quantum group theory @xmath249 is related with the symmetry @xmath250 of the present theory . we also observe that , exploiting eq.s ( [ ii6 ] ) and ( [ ii7 ] ) , we can obtain the two following properties of the function @xmath251 $ ] @xmath252=[\,x\,]\stackrel{\kappa ' } { \oplus}[\,y\ , ] \ \ , \label{v7}\\ & & [ \,x\,y\,]=[\,x\,]\stackrel{\kappa ' } { \otimes}[\,y\ , ] \ \ , \label{v8}\end{aligned}\ ] ] with @xmath253 . the definitions of @xmath1-sum and @xmath1-product given through ( [ ii3 ] ) and ( [ ii4 ] ) respectively transform as follows @xmath254 @xmath255 in particular the @xmath1-sum assumes a very simple form @xmath256 starting directly from the @xmath1-sum given by ( [ v11 ] ) , one obtains the following expressions for the @xmath1-differential and @xmath1-derivative @xmath257 we consider now the functions @xmath258 and @xmath101 which can be written also as @xmath259 we remark the following concavity properties @xmath260 a very interesting property of these functions is their power law asymptotic behavior @xmath261 the taylor expansion of @xmath1-exponential is given by @xmath262 ( @xcite , p. 26 ) , with the coefficients @xmath263 defined as @xmath264 it results that @xmath265 and @xmath266 . we note that the first three terms in the above taylor expansion are the same as the ordinary exponential , namely @xmath267 in fig.1 , the function @xmath268 for a fixed value of @xmath1 is plotted . we note that the bulk of this function is very close to the standard exponential . indeed the taylor expansion of @xmath269 is the same , up to second order , of the one of @xmath270 . the tail of @xmath268 behaves as a power law . between the bulk and the tail an intermediate region whose extension depends on the value of @xmath1 exists . in fig . 2 , the function @xmath268 for some different values of @xmath1 is plotted . we note that when @xmath34 the @xmath1-exponential approaches the ordinary one . the taylor expansion of @xmath271 converges if @xmath272 and assumes the form @xmath273 ( @xcite , p. 25 ) , with @xmath274 while for @xmath275 , @xmath276 is given by @xmath277 it results @xmath278 and @xmath279 . the first terms of the expansion are @xmath280 another expansion involving the @xmath1-exponential @xmath281 is the following @xmath282 ( @xcite , p. 58 ) , being @xmath283 exploiting this expansion , we can write the @xmath1-exponential as an infinite product of standard exponentials @xmath284 on the other hand the @xmath1-exponential can be viewed as a continuous linear combination of an infinity of standard exponentials . namely for @xmath285 it results ( @xcite , p. 1108 ) @xmath286 the following two integrals can be useful @xmath287\,|2\kappa|^{-r}}{\big[1-(r\!-\!1)\ , \frac{\gamma\left(\frac{1}{|2\kappa|}-\frac{r}{2 } \right)}{\gamma\left(\frac{1}{|2\kappa|}+\frac{r}{2 } \right)}\,\,\gamma\left(r\right)\,\,\,\,\,\,\,\,\ , \label{v31}\end{aligned}\ ] ] @xmath288 we conclude the present section focusing our attention to another interesting property of @xmath1-logarithm . we consider the following eigenvector equation @xmath289 and examine the case of the eigenvector @xmath290 . it is trivial to verify that when @xmath291 we obtain that the eigenvalue @xmath292 of this operator is the standard logarithm @xmath293 . we pose now the question if it is possible to determine the operator @xmath294 associated to the same eigenvector @xmath290 and having as eigenvalue the @xmath1-logarithm , namely @xmath295 we obtain that this operator is the finite difference operation @xmath296 which reduces to the standard derivative as the increment @xmath297 of the independent variable approaches to zero . one can find many other elegant and useful mathematical properties for the @xmath1-functions which obviously we can not report here . in the present section we treat the statistical system , considered previously in stationary conditions , within a relativistic and kinetic framework . the new relativistic kinetics here presented , in the limit @xmath298 , reduces to the classical kinetics considered in @xcite . by using the standard notation of the relativistic theory we denote with @xmath299 the four - vector position and with @xmath300 the four - vector momentum , being @xmath301 and employ the metric @xmath302 @xcite . let us consider the following relativistic kinetic equation @xmath303 , & & \label{vi1}\end{aligned}\ ] ] where the distribution @xmath5 is a function of the four - vectors @xmath19 and @xmath162 , namely @xmath304 . we note that the left hand side of eq . ( [ vi1 ] ) is the same of the standard relativistic boltzmann equation but the collision integral in the right hand side results to be more complicated , containing the deformed product @xmath305 and the arbitrary function @xmath306 which we suppose to be positive and increasing . the factor @xmath307 is the transition rate which depends only on the nature of the two body particle interaction . the above equation in the case @xmath179 and @xmath308 reduces to the already known relativistic boltzmann equation describing the standard relativistic kinetics @xcite . clearly in the case @xmath309 and @xmath310 , the above equation describes a new relativistic kinetics , radically different from the standard one . we anticipate that this new relativistic kinetics , which we will consider here , defines a statistics resulting to be independent on the particular form of the function @xmath306 . _ steady states : _ we consider now the steady states of eq . ( [ vi1 ] ) for which the collision integral becomes equal to zero . then we have @xmath311 and after taking into account the property ( [ ii61 ] ) of the @xmath1-logarithm , we obtain @xmath312 this last equation represents a conservation law and then we can conclude that @xmath313 is a summational invariant ; in the most general case it is a linear combination of the microscopic relativistic invariants , namely a constant and the four - vector momentum . in ref . @xcite it is shown that in presence of external electromagnetic fields the more general microscopic relativistic invariant has the form @xmath314 constant , being @xmath315 the hydrodynamic four - vector velocity with @xmath316 . then we can pose @xmath317 consequently we obtain the following stationary distribution @xmath318 in the case @xmath179 this distribution reduces to the already known relativistic distribution @xcite . the above equilibrium distribution , in the global rest frame where @xmath319 and in absence of external forces ( @xmath320 ) , simplifies as @xmath321 and assumes the same form of the distribution ( [ iii1 ] ) . we remark that for @xmath322 this distribution presents a power law behavior , namely @xmath323 being @xmath324 . in order to introduce explicitly the dependence on the velocity variable in the distribution ( [ vi6 ] ) , we consider the expression of the relativistic kinetic energy @xmath325 @xmath326 with @xmath327 and the lorentz factor given by @xmath328 after defining @xmath329 we write eq . ( [ vi6 ] ) as follows @xmath330 \right ) \ \ . \label{vi10}\ ] ] note that in the region @xmath331 this distribution assumes the form @xmath332 which , after setting @xmath333 , coincides with the non relativistic statistical distribution , proposed in ref . @xcite . _ h - theorem : _ in the standard relativistic kinetics it is well known from the h - theorem that the production of entropy is never negative and in equilibrium conditions there is no entropy production . in the following we will demonstrate the h - theorem for the system governed by the kinetic equation ( [ vi1 ] ) . for simplicity of the notation , hereafter we omit the letter @xmath1 in the symbol of @xmath1-entropy . we define the four - vector entropy @xmath334 , in terms of the distribution @xmath335 , as follows @xmath336 and note that @xmath337 coincides with the @xmath1-entropy defined previously through eq . ( [ iv20 ] ) while @xmath338 is the entropy flow . if we take into account the relation ( [ iii9 ] ) , the above four - vector entropy can be written in terms of the distribution @xmath5 as @xmath339 it is trivial to verify that the entropy production @xmath340 can be calculated starting from the definition of @xmath341 and the evolution equation ( [ vi1 ] ) , obtaining @xmath342 \ , \ln_{_{\{{\scriptstyle \kappa}\}}}\!f \nonumber \\ & & - k_{_b}\lambda\,\alpha \,m \int \frac{d^3p}{p^{0}}\,(\ln_{_{\{{\scriptstyle \kappa}\}}}\!f)\,f^{\nu}\frac{\partial f}{\partial p^{\,\nu}}. \ \ \ \label{vi14}\end{aligned}\ ] ] since the lorentz force @xmath343 has the properties @xmath344 and @xmath345 the last term in the above equation involving @xmath343 is equal to zero @xcite , namely @xmath346 \ , \ln_{_{\{{\scriptstyle \kappa}\}}}\!f . \ \ \ \label{vi15}\end{aligned}\ ] ] given the particular symmetry of the integral in eq . ( [ vi15 ] ) we can write the entropy production as follows @xmath347 \nonumber \\ & & \times \ , [ \ln_{_{\{{\scriptstyle \kappa}\}}}\!f + \ln_{_{\{{\scriptstyle \kappa}\}}}\!f_1-\ln_{_{\{{\scriptstyle \kappa}\}}}\!f ' - \ln_{_{\{{\scriptstyle \kappa}\}}}\!f'_1 ] . \ \ \ \ \ \label{vi16}\end{aligned}\ ] ] finally , we set this equation in the form @xmath348 \nonumber \\ & & \times \left[\,\ln_{_{\{{\scriptstyle \kappa}\}}}\ ! ( f'\otimes\mbox{\raisebox{-2mm}{\hspace{-3.3mm}$\scriptstyle \kappa$ } } \hspace{1mm}f'_1)- \ln_{_{\{{\scriptstyle \kappa}\}}}\!(f\otimes\mbox{\raisebox{-2mm}{\hspace{-3.3mm}$\scriptstyle \kappa$ } } \hspace{1mm}f_1)\right ] . \label{vi17}\end{aligned}\ ] ] after imposing that @xmath349 is an increasing function , it results @xmath350\,[\ln_{_{\{{\scriptstyle \kappa}\}}}\!(h_1)-\ln_{_{\{{\scriptstyle \kappa}\}}}\!(h_2)]\geq 0 $ ] @xmath351 and then we can conclude that @xmath352 this last relation is the local formulation of the relativistic h - theorem which represents the second law of the thermodynamics for the system governed by the evolution equation ( [ vi1 ] ) . concerning the arbitrary positive and increasing function @xmath306 appearing in the collision integral of the evolution equation , we note that , if we suppose that obeys to the following condition @xmath353 we recover the expression @xmath354 proposed in ref . in the present section we will show that the deformation introduced by the parameter @xmath1 emerges naturally within the einstein s special relativity , so that one can see the @xmath1-deformation as a purely relativistic effect . let us consider in the one - dimensional frame @xmath355 two identical particles of rest mass @xmath356 . we suppose that the first particle moves toward right with velocity @xmath357 while the second particle moves toward left with velocity @xmath358 . the relativistic momenta of the particles are given by @xmath359 and @xmath360 respectively , being @xmath361 we consider now the same particles in a new frame @xmath362 which moves at constant speed @xmath358 toward left with respect to the frame @xmath355 . in this new frame the particles have velocities given by @xmath363 and @xmath364 respectively , being @xmath365 the well known relativistic additivity law for the velocities . in the same frame @xmath362 the particle relativistic momenta are given by @xmath366 and @xmath367 , respectively . up to now , we have simply recalled some well known concepts of the special relativity @xcite . let us pose the following question : if it is possible and how to obtain the value of the relativistic momentum @xmath368 starting directly from the values of the momenta @xmath166 and @xmath167 in the frame @xmath355 . the answer to this apparently innocent question is affirmative . one , after straightforward calculations ( see the theorem in this section ) , arrives to the following surprising result @xmath369 in words , the relativistic momentum @xmath368 of the first particle in the rest frame of the second particle is the @xmath1-deformed sum , with @xmath370 , of the momenta @xmath166 and @xmath167 of the particles in the frame @xmath355 . unexpectedly we discover that the @xmath1-sum is the additivity law for the relativistic momenta . eq.([vii3 ] ) which we write in the form @xmath371 says that the @xmath1-deformed sum and the relativistic sum of the velocities are intimately related and reduce both , to the standard sum as the velocity @xmath11 approaches to infinity . the deformations in both the cases are relativistic effects and are originated from the fact that @xmath11 has a finite value . eq.([vii4 ] ) follows as a particular case from the following theorem : _ theorem : _ let @xmath372 be the relativistic momenta of two particles @xmath373 of rest mass @xmath374 and @xmath375 which move in the one dimensional frame @xmath355 with speed @xmath357 and and @xmath358 respectively . if we indicate with @xmath376 the @xmath1-sum defined through eq . ( [ v11 ] ) and with @xmath377 the velocity relativistic additivity law defined through eq . ( [ vii2 ] ) , it results that @xmath378 _ proof : _ we start by using the definition of the @xmath1-sum , subsequently we use the explicit form of the relativistic momentum and finally we use the definition of the velocity relativistic additivity law @xmath379 ^ 2}+ \frac{p_2\,(v_2)}{m_2 } \sqrt{1 + \left[\frac{p_1\,(v_1)}{m_1\,c}\right ] ^2 } \nonumber \\ & & = \frac{v_1}{\sqrt{1-\left(v_1/c \right)^2 } } \,\sqrt{1 + \frac{(v_2/c)^2}{1-\left(v_2/c\right)^2 } } \nonumber \\ & & + \frac{v_2}{\sqrt{1-\left(v_2/c \right)^2 } } \,\sqrt{1 + \frac{(v_1/c)^2}{1-\left(v_1/c\right)^2 } } \nonumber \\ & & = \frac{v_1+v_2 } { \sqrt{\left[1-(v_1/c)^2 \right ] \left[1-(v_2/c)^2 \right ] } } \nonumber \\ & & = \left(v_1 \oplus^c v_2\right ) \ , \frac{1+v_1 v_2/c^2 } { \sqrt{\left[1-(v_1/c)^2 \right ] \left[1-(v_2/c)^2 \right ] } } \nonumber \\ & & = \frac{v_1 \oplus^c v_2}{\sqrt{1-\displaystyle{\frac{1}{c^2}\left(\frac{v_1+v_2 } { 1+v_1v_2/c^2}\right)^2 } } } \nonumber \\ & & = \frac{v_1 \oplus^c v_2 } { \sqrt { 1-\displaystyle{\frac{\left(v_1 \oplus^c v_2\right)^2}{c^2 } } } } \nonumber \\ & & = \frac{p_i\,(v_1\oplus^c v_2)}{m_i } \ \ \ . \label{vii6a}\end{aligned}\ ] ] trivially from eq . ( [ vii6 ] ) one obtains eq.([vii4 ] ) as particular case , when @xmath380 . note that the parameter @xmath1 has different values in these two equations because the summed quantities in the two cases are different . we can easily explain the meaning of the deformed derivative . we indicate with @xmath162 the relativistic momentum in the frame @xmath355 , and with @xmath381 the derivative with respect to @xmath162 of the lorentz invariant scalar @xmath307 . the same quantities in the frame @xmath362 are indicated with @xmath382 and @xmath383 , respectively . it is trivial to verify that @xmath384 and then we can conclude that the @xmath1-deformed derivative can be viewed as a standard derivative in an appropriate frame . in the next section we will consider , in the framework of the special relativity , the @xmath1-statistics of @xmath385-identical particles , where @xmath1 is a dimensionless parameter and we will determine its value . to do so , it is more convenient to write eq . ( [ vii4 ] ) in the form @xmath386 which holds for any @xmath1 . in the present section we calculate the value of the parameter @xmath1 which , due to the symmetry @xmath387 of the theory , we consider positive , namely @xmath388 . clearly if we start from the @xmath1-statistical distributions @xmath389 and @xmath390 describing two independent statistical systems , we can construct the distribution @xmath391 which describes a new composite system . this system in the case @xmath179 reduces to the one described through the distribution @xmath392 . in the following we will assume that _ the distribution @xmath392 describes a state also in the case @xmath393_. obviously this state is different from the one described by @xmath391 . as we will see , this simple but meaningful hypothesis is sufficient to determine the value of @xmath1 in the case of relativistic systems . taking into account the form of the distributions @xmath389 and @xmath390 given by eq . ( [ vi6 ] ) ( for simplicity we pose @xmath394 ) , and the property ( [ ii30 ] ) of the @xmath1-exponential one can write immediately @xmath395 with @xmath396 . after some simple algebra we rewrite eq . ( [ viii1 ] ) as follows @xmath397 with @xmath398 and @xmath399 we assume now that the @xmath1-exponential in the right hand side of the eq . ( [ viii2 ] ) has the same structure of the one given by eq . ( [ vi6 ] ) . clearly we must impose that @xmath400 be exclusively expressed in terms of non statistical parameters . in the following we will show that @xmath401 . to do this we exploit eq . ( [ vii7 ] ) obtained within the special relativity . starting from this equation and taking into account the property ( [ ii30 ] ) of the @xmath1-exponential we obtain @xmath402 recall that we wish to calculate the parameter @xmath1 which has a value that does not depend on the particle energy . then , without loosing generality , we can consider eq . ( [ viii4 ] ) in the ultrarelativistic region @xmath403 where it results @xmath404 : @xmath405 after comparing eq.s ( [ viii2 ] ) and ( [ viii5 ] ) , one obtains the relation @xmath406 which is the same as the one given by eq . ( [ viii3 ] ) , only if we impose that @xmath401 . eq.([viii6 ] ) can be also written as @xmath407 and results to be formally similar @xmath408 to the relation defining the relativistic momentum given by eq . ( [ vii1 ] ) . we can extract finally the value of @xmath1 obtaining @xmath409 it is important to emphasize that this expression of the parameter @xmath1 holds only under the above mentioned hypothesis and imposes that @xmath410 . this condition on the value of @xmath1 coincides with the one expressed by eq . ( [ iv15 ] ) and obtained in a completely different way . we have @xmath179 only if @xmath411 or if @xmath412 . the limiting case @xmath413 is obtained if @xmath414 or if @xmath415 . at this point one can write the distribution ( [ vi6 ] ) in the form @xmath416 note that the statistical information of the system , namely the temperature is hidden exclusively in the parameter @xmath1 . when @xmath417 the distribution ( [ viii9 ] ) shows a power law asymptotic behavior @xmath418 the distribution ( [ viii9 ] ) viewed as a function of the velocity becomes @xmath419 concerning its derivative one obtains @xmath420 and then for @xmath421 it results @xmath422 then for @xmath423 one has both @xmath424 and @xmath425 in @xmath426 . for @xmath427 results @xmath424 and @xmath428 in @xmath426 . finally for @xmath429 results @xmath424 and @xmath430 in @xmath426 . in the non relativistic region for which @xmath331 we have @xmath431 while in the limit @xmath12 one recovers the standard maxwellian distribution @xmath432 the explicit form of the distribution ( [ viii11 ] ) when @xmath433 simplifies as follows @xmath434 and in the limit @xmath12 becomes @xmath435 in fig . 3 , the distribution function given by eq . ( [ viii16 ] ) ( after normalization ) versus @xmath436 for different values of @xmath1 and then for different values of @xmath437 according to eq . ( [ viii8 ] ) , is plotted . since long time it is known that the cosmic rays spectrum , which extends over @xmath438 decades in energy , from a few hundred of mev ( @xmath439 ev ) to a few hundred of eev ( @xmath440 ev ) and spans @xmath441 decades in particle flux , from @xmath442 to @xmath443 @xmath444^{-1}$ ] , is not exponential and then it violates the boltzmann equilibrium statistical distribution @xmath445 @xcite . approximately this spectrum follows a power law @xmath446 and the spectral index @xmath447 is near 2.7 below @xmath448 ev , near 3.1 above @xmath449 ev and again near 2.7 above @xmath450 ev . on the other hand it is known that the particles composing the cosmic rays are essentially the normal nuclei as in the standard cosmic abundances of matter . then the cosmic rays can be viewed as an equivalent statistical system of identical relativistic particles with masses near the mass of the proton ( @xmath451 mev ) . these above characteristics ( relativistic particles with a very large extension both for their flux and energy ) yield the cosmic rays spectrum an ideal physical system for a preliminary test of the correctness and predictability of the theory here proposed . we consider the statistical distribution @xmath452 given by eq . ( [ vi6 ] ) or ( [ viii9 ] ) . the particle flux @xmath453 can be calculated trivially if we take into account the relativistic expression linking @xmath454 and @xmath162 obtaining @xmath455\ , \exp_{_{\{{\scriptstyle \kappa}\}}}\!\big(\!\!-\beta\,\,(e-\mu)\,\big ) \ \ . \label{ix1}\ ] ] note that this particle flux , in agreement with the observational data , decays following the power law @xmath456 with @xmath457 analogously the particle flux obtained starting from the boltzmann - gibbs statistical distribution is given by @xmath458\ , \exp\left(-\frac{e}{k_{_{b}}t}\right ) \ \ . \label{ix4}\ ] ] we use these two theoretical distributions of particle flux to fit the cosmic rays data reported in ref . @xcite . in fig . 4 , we show the observed data together with the theoretical curves @xmath459 ( solid line ) and @xmath460 ( dotted line ) . the curve @xmath459 corresponding to @xmath461 @xmath462^{-1}$ ] , @xmath463 , @xmath464 , and @xmath465 provides a high quality agreement with the observed data . this agreement over so many decades is quite remarkable . from the value of @xmath1 and @xmath466 and adopting eq . ( [ viii7 ] ) we obtain that @xmath467 . ) with @xmath461 @xmath462^{-1}$ ] , @xmath463 , @xmath464 , @xmath465 and @xmath467 . the dotted line represents the theoretical curve obtained within the standard boltzman - gibbs statistics given by eq . ( [ ix4 ] ) with @xmath468 @xmath469^{-1}$ ] , @xmath463 and @xmath467 . the observational data are collected by s.p . swordy @xcite . ] in the same figure the curve @xmath460 ( with @xmath470 @xmath462^{-1}$ ] , @xmath463 and @xmath467 ) which decays exponentially and can not fit the observed data violating the boltzmann - gibbs statistics , is reported . a short remark must be made at this point . clearly , the power law asymptotic behavior of the spectrum @xmath459 is imposed by the @xmath1-exponential whose origin is the @xmath1-sum . but the @xmath1-sum emerges naturally within the special relativity as the composition law of the relativistic momenta . then we can conclude that the power law asymptotic behavior of the cosmic rays flux is simply the signature of the particle relativistic nature . it is widely known that the boltzmann - gibbs distribution @xmath471 originally proposed to describe a classical particle gas in thermodynamic equilibrium can be adopted to describe an enormous amount of phenomena in nature . on the other hand the power law tails have been observed experimentally in several fields of science . some times in particular fields , this power law has a name ( e.g. pareto law in econophysics , gutenberg - richter law in seismology etc ) . furthermore the power law tail is preceded by an exponential region and between the two regions exists a third intermediate region . it is worth remarking that the @xmath1-exponential defines a distribution which can describe simultaneously the three above regions ( see fig.1 ) and then is particularly suitable to describe the above mentioned phenomena . as a working example we analyze the experimental data reported in ref . @xcite related to the rain events in meteorology . in fig . 5 is plotted the number density @xmath385[events/(year mm ) ] of rain events versus the event size @xmath472[mm ] on a double logarithmic scale . we note that the data have a large extension ( the abscissa spans 7 decades and the ordinate 5 ) and remark that its behavior is typical of a class of experimental data which we find in several areas of science . in order to fit the experimental data we adopt the distribution @xmath473 and , as one can see in fig . 5 , a remarkable agreement is obtained with @xmath474 [ events/(year mm ) ] , @xmath475 [ mm@xmath476 and @xmath477 . clearly , one can hunt other mechanisms , different from the relativistic one , leading to @xmath1-statistics . beside the important problem of the agreement between the theoretical curve and the observational data one does not neglect the epistemological problem concerning the structure of the theory which must be able both to explain the origin and to determine the value of any parameter appearing within the theory . we summarize briefly the results obtained in the present effort . we have shown that beside the boltzmann - shannon entropy , the quantity @xmath478 with @xmath479 and @xmath480 , is the only existing entropy , simultaneously concave , additive and extensive . starting from this entropy it is possible to construct a generalized statistical mechanics ( and thermodynamics ) having the same mathematical and epistemological structure of the boltzmann - gibbs one , which is recovered when the deformation parameter @xmath1 approaches to zero . within this generalized statistics the distribution function assumes the form @xmath481 with @xmath0 while the constants @xmath9 and @xmath3 are given by @xmath482^{1/2\kappa}$ ] , @xmath483 . the chemical potential @xmath177 can be fixed by the normalization condition . this distribution has a bulk very close to the exponential one while its tail decays following a power law @xmath484 . we have shown that , within the special relativity , it is possible to determine the value of @xmath1 , obtaining in this case @xmath485 so that the relativistic statistical meechanics does not contain free parameters . the theory can describe observational data in many fields . in particular we find a high quality agreement in analyzing the spectrum of the cosmic rays which violates manifestly the boltzmann - gibbs statistics . this is an important test for the theory because the cosmic rays are relativistic particles and their spectrum has a very large extension ( @xmath438 decades in energy and @xmath441 decades in flux ) . g. kaniadakis , physica a * 296 * , 405 ( 2001 ) . g. kaniadakis , phys . a * 288 * , 283 ( 2001 ) . g. kaniadakis and a.m. scarfone , physica a * 305 * , 69 ( 2002 ) . j. naudts , physica a , in press ( 2002 ) , [ cond - mat/0203489 ] . s. abe and y. okamoto , _ nonextensive statistical mechanics and its applications _ , lectures notes in physics * 560 * , spring - verlag berlin ( 2001 ) . g. kaniadakis , m. lissia and a. rapisarda eds . , special issue of physica a * 305*,(2002 ) . o. peters , c. hertlein , and k.christensen , phys . 88 * , 018701 - 1 ( 2002 ) o. peters , and k.christensen , phys . e * 66 * , 036120 ( 2002 ) . stanley , l.a.n . amaral , x. gabaix , p. gopikrishnan , v. plerou , physica a * 299 * , 1 ( 2001 ) . c. tsallis , quimica nova * 17 * , 468 ( 1994 ) . s. abe , phys . lett . a * 224 * , 326 ( 1997 ) . borges and i. roditi , phys . a * 246 * , 399 ( 1998 ) . jaynes , phys . 106 * , 620 ( 1957 ) ; * 108 * , 171 ( 1957 ) . j. harvda and f. charvat , kybernetica * 3 * , 30 ( 1967 ) . c. tsallis , j. stat . phys . * 53 * , 479 ( 1988 ) v.m . vasyliunas , j. geophys . res . * 73 * , 2839 ( 1968 ) . i.s . gradshteyn and i.m . ryzhik , _ table of integrals , series , and product _ , academic press london ( 2000 ) . tuszynaki , j.l . rubin , j. meyer , and m. kibler , phys . lett . a * 175 * , 173 ( 1993 ) . m.r . ubriaco , phys . e * 57 * , 179 ( 1998 ) . de groot , w.a . van leeuwen , and ch.g . van weert , _ relativistic kinetic theory : principle and applications _ , north - holland , amsterdam ( 1980 ) . libboff , _ kinetic theory _ , prentice - hall international , englewoord cliffs , nj , ( 1990 ) . biermann , and g. sigl , _ physics and astrophysics of ultra - hight - energy cosmic rays _ , lectures notes in physics * 576 * , spring - verlag berlin ( 2001 ) , [ astro - ph/0202425 ] . c. tsallis , j.c . anjos , and e.p . borges , arxiv : astro - ph/0203258 . swordy , cosmic rays data , chicago university , http://astroparticle.uchicago.edu/announce.htm .

we consider the wireless two - way relay channel , in which two - way data transfer takes place between the end nodes with the help of a relay . for the denoise - and - forward ( dnf ) protocol , it was shown by koike - akino et . al . that adaptively changing the network coding map used at the relay greatly reduces the impact of multiple access interference at the relay . the harmful effect of the deep channel fade conditions can be effectively mitigated by proper choice of these network coding maps at the relay . alternatively , in this paper we propose a distributed space time coding ( dstc ) scheme , which effectively removes most of the deep fade channel conditions at the transmitting nodes itself without any csit and without any need to adaptively change the network coding map used at the relay . it is shown that the deep fades occur when the channel fade coefficient vector falls in a finite number of vector subspaces of , which are referred to as the singular fade subspaces . dstc design criterion referred to as the _ singularity minimization criterion _ under which the number of such vector subspaces are minimized is obtained . also , a criterion to maximize the coding gain of the dstc is obtained . explicit low decoding complexity dstc designs which satisfy the singularity minimization criterion and maximize the coding gain for qam and psk signal sets are provided . simulation results show that at high signal to noise ratio , the dstc scheme provides large gains when compared to the conventional exclusive or network code and performs slightly better than the adaptive network coding scheme proposed by koike - akino et . al .

1203.3128

we consider the two - way wireless relaying scenario shown in fig.[dnf_protocol ] . two - way data transfer takes place between the nodes a and b with the help of the relay r. it is assumed that all the three nodes operate in half - duplex mode , i.e. , they can not transmit and receive simultaneously in the same frequency band . the idea of physical layer network coding for the two way relay channel was first introduced in @xcite , where the multiple access interference occurring at the relay was exploited so that the communication between the end nodes can be done using a two phase protocol . a protocol called denoise - and - forward ( dnf ) was proposed in @xcite , which consists of the following two phases : the _ multiple access _ ( ma ) phase ( fig . [ dnf_mac ] ) , during which a and b simultaneously transmit to r and the _ broadcast _ ( bc ) phase ( fig . [ dnf_bc ] ) during which r transmits to a and b. network coding map , which is also referred to as the denoising map , is chosen at r in such a way that a ( b ) can decode the messages of b ( a ) , given that a ( b ) knows its own messages . during the ma phase , the transmissions from the end nodes were allowed to interfere at r , but the harmful effect of this interference was mitigated by a proper choice of the network coding map used at r. information theoretic studies for the physical layer network coding scenario were reported in @xcite,@xcite . a differential modulation scheme with analog network coding for bi - directional relaying was proposed in @xcite . the design principles governing the choice of modulation schemes to be used at the nodes for uncoded transmission were studied in @xcite . an extension for the case when the nodes use convolutional codes was done in @xcite . a multi - level coding scheme for the two - way relaying scenario was proposed in @xcite . power allocation strategies and lattice based coding schemes for bi - directional relaying were proposed in @xcite . it was observed in @xcite that the network coding map used at the relay needs to be changed adaptively according to the channel fade coefficients , in order to minimize the impact of the multiple access interference ( mai ) . a computer search algorithm called the _ closest - neighbour clustering _ ( cnc ) algorithm was proposed in @xcite to obtain the adaptive network coding maps resulting in the best distance profile at r. an adaptive network coding scheme for mimo two - way relaying based on the cnc algorithm was proposed in @xcite . an alternative procedure to obtain the adaptive network coding maps , based on the removal of deep channel fade conditions using latin squares was proposed in @xcite . a quantization of the set of all possible channel realizations based on the network code used was obtained analytically in @xcite . an extension of the adaptive network coding scheme for mimo two - way relaying using latin rectangles was made in @xcite . as an alternative to the adaptive network coding schemes in @xcite and @xcite@xcite , in this paper , we propose a distributed space time coding ( dstc ) scheme , which mitigates the effect of mai to the fullest extent possible at the transmitting nodes itself without any csit . for the proposed dstc scheme the network coding map used at r need not be changed adaptively according to channel conditions which reduces the complexity at r to a great extent and also eliminates the need for overhead bits from r to a and b to indicate the choice of the network coding map . a distributed space time coding scheme for a wireless two - way relay network with multiple relay nodes was proposed in @xcite , in which the dstc was constructed at the relay nodes . in the proposed scheme , the dstc is constructed at the end nodes a and b. throughout , a quasi - static fading scenario is assumed with the channel state information ( csi ) available only at the receivers . let @xmath1 and @xmath2 denote the fade coefficients associated with a - r and b - r links and @xmath3 and @xmath4 denote the fade coefficients associated with r - a and r - b links . all the fading coefficients are assumed to follow rician distribution . let @xmath5 denote the unit energy @xmath6 point constellation used at the end nodes . let @xmath7 denote the mapping from bits to complex symbols used at a and b. in the sequel , we briefly describe the adaptive network coding schemes based on the dnf protocol proposed in @xcite , @xcite @xcite . throughout the paper , by dnf protocol , we refer to the schemes proposed in @xcite and @xcite@xcite . in the dnf protocol , transmission occurs in two phases : multiple access ( ma ) phase during which a and b simultaneously transmit to r and broadcast ( bc ) phase during which r transmits to a and b. let @xmath8 , @xmath9 @xmath10 denote the complex symbols transmitted by a and b respectively , where @xmath11 . the received signal at @xmath12 is given by , @xmath13 the additive noise @xmath14 is assumed to be @xmath15 where @xmath16 denotes the circularly symmetric complex gaussian random variable with mean zero and variance @xmath17 let @xmath18 denote the maximum likelihood ( ml ) estimate of @xmath19 at r based on the received complex number @xmath20 depending on the value of @xmath1 and @xmath21 r chooses a many - to - one map @xmath22 where @xmath23 is the signal set ( of size between @xmath24 and @xmath25 ) used by r during the @xmath26 phase . in order to ensure that a ( b ) is able to decode b s ( a s ) message , the map @xmath27 should satisfy the exclusive law @xcite , i.e. , @xmath28 the cnc algorithm proposed in @xcite obtains the map @xmath27 which results in the best distance profile during the ma phase at r. the cnc algorithm is run for all possible channel realizations and a partition of the set of all channel realizations is obtained depending on the chosen network coding map . for a given channel realization , the choice of the network coding map is indicated to a and b using overhead bits . during the bc phase r transmits @xmath29 the received signals at a and b during the bc phase are respectively given by , @xmath30 where @xmath31 and @xmath32 are independent and @xmath33 since the map @xmath27 satisfies the exclusive law and a ( b ) knows its own message @xmath34 ( @xmath35 ) , it can decode @xmath35 ( @xmath34 ) by decoding @xmath36 the cnc algorithm optimizes the entire distance profile instead of maximizing only the minimum distance . in some cases , this results in the use of signal sets with a larger cardinality during the bc phase . to solve this problem , an algorithm called the nearest neighbour clustering ( nnc ) algorithm was proposed in @xcite which maximizes the minimum distance alone , instead of optimizing the entire distance profile . the choice of the network coding map obtained depends only on the ratio @xmath37 and not the individual values of @xmath1 and @xmath2 @xcite . in @xcite , the values of @xmath37 for which deep channel conditions occur were identified and network coding maps which remove the harmful effect of these deep channel conditions were obtained by the completion of partially filled latin squares . for the proposed dstc scheme , transmission occurs in four phases : two ma phases during which a and b simultaneously transmit to r followed by two bc phases during which r transmits to a and b. two independent complex symbols each from a to b and b to a get exchanged at the end of the four phases and hence the information rate in bits per channel use for the proposed scheme is same as that of the dnf protocol . let @xmath38 denote two independent complex symbols a wants to communicate to b. similarly , b wants to communicate two independent complex symbols @xmath39 to a. during the @xmath40 ma phase @xmath41 a transmits @xmath42 a function of @xmath43 and @xmath44 and similarly b transmits @xmath45 a function of @xmath46 and @xmath47 the received signal at r during the two ma phases can be written as , @xmath48 & = \left [ h_a \ : h_b \right ] \begin{bmatrix } f_a^1(x_{a_1},x_{a_2 } ) & f_a^2(x_{a_1},x_{a_2 } ) \\ f_b^1(x_{b_1},x_{b_2 } ) & f_b^2(x_{b_1},x_{b_2 } ) \end{bmatrix } + \left [ \begin{array}{c } z_{r_1 } z_{r_2 } \end{array } \right],\end{aligned}\]]where @xmath49 denotes the received signal at r during the @xmath40 ma phase , @xmath50 and @xmath51 are independent and @xmath52 let @xmath53 $ ] and @xmath54.$ ] the matrix , @xmath55represents a _ dstc codeword matrix_. note that in the dstc codeword matrix , @xmath43 and @xmath56 can occur only in the first row and , @xmath46 and @xmath57 can occur only in the second row . in this way the dstc differs from space time codes for the conventional 2@xmath581 multiple antenna system with two collocated antennas at the transmitter in which the complex symbols can occupy any entry in the codeword matrix . for a complex number @xmath59 let @xmath60 and @xmath61 denote the real and imaginary parts of @xmath62 a dstc is said to be linear if the entries of the first row of the codeword matrices are complex linear combinations of @xmath63 and the entries of the second row are complex linear combinations of @xmath64 any codeword matrix @xmath65 of a linear dstc can be written as , @xmath66 the matrices @xmath67 and @xmath68 are referred to as the _ weight matrices _ of the dstc . note that the entries of the second ( first ) row are zeros in the matrices @xmath69 and @xmath70 ( @xmath71 and @xmath72 ) . a linear dstc is said to be over the signal set @xmath5 if the entries of the first ( second ) row of the codeword matrices are complex linear combinations of @xmath43 and @xmath56 ( @xmath46 and @xmath57 ) , where @xmath73 @xmath44 @xmath46 and @xmath57 belong to the signal set @xmath74 for a linear dstc over @xmath5 , codeword matrix @xmath65 is of the form @xmath75 where @xmath76 and @xmath77 are @xmath78 complex matrices referred to as the _ generator matrices _ at node a and b respectively . throughout the paper , we consider only linear dstcs over a signal set @xmath74 let @xmath79 denote the maximum likelihood estimate of @xmath80 at r. the relay r transmits @xmath81 and @xmath82 during the first and second bc phases respectively , where @xmath83 denotes the bit - wise xor operation . the received signals at the end nodes during the two bc phases are given by , @xmath84 and @xmath85 where @xmath86 since a ( b ) knows its own messages and the xor map satisfies the exclusive law , a ( b ) can decode @xmath87 ( @xmath88 ) @xmath41 by decoding @xmath89 note that for the proposed dstc scheme the signal set used during the bc phase is of the minimum cardinality @xmath90 ( the cardinality of the signal set should be at least @xmath90 to convey @xmath91 information bits ) . in contrast , for the scheme proposed in @xcite , depending on channel conditions unconventional signal sets with cardinality greater than the minimum cardinality are required . minimum cardinality signal set is used during the bc phase and throughout the paper the focus is on optimizing the performance during the ma phase . some of the advantages of the proposed dstc scheme over the schemes proposed in @xcite , @xcite@xcite are summarized below : * unlike the schemes proposed in @xcite,@xcite@xcite , for the proposed dstc scheme , the network coding map used at r need not be changed adaptively according to channel conditions . any network coding map satisfying the exclusive law will give the same performance and for simplicity , the conventional bit - wise exclusive or ( xor ) map itself can be used . this reduces the complexity at r to a great extent and also eliminates the need for overhead bits from r to a and b to indicate the choice of the network coding map . * for the scheme proposed in @xcite , for certain channel conditions the adaptive network coding map necessitates the use of unconventional signal sets with cardinality greater than the minimum cardinality required during the bc phase , which results in a degradation in performance . for the proposed scheme , the relay always uses a conventional signal set with minimum cardinality . * the adaptive network coding maps were obtained in @xcite , by exhaustive computer search . for the proposed scheme no such computer search is required , since the same network code is used irrespective of channel conditions . the contributions and organization of the paper are as follows : * for a classical @xmath92 mimo system with collocated antennas , deep channel fade conditions occur when the channel fade coefficient vector belongs to a finite number of vector subspaces of @xmath93 referred to as the singular fade subspaces . the way in which transmit diversity schemes ( space time codes ) remove the harmful effect of these singular fade subspaces is discussed . the connection between the dimension of these singular fade subspaces and the transmit diversity order is explained ( section ii ) . * the mac phase of the dnf protocol for the two - way relaying scenario can be viewed as a virtual @xmath94 miso system . the singular fade subspaces for the classical @xmath94 miso system , are singular fade subspaces for the two - way relaying scenario as well . the connection between dimension of these singular fade subspaces and the diversity order for the adaptive network coding schemes proposed in @xcite and @xcite-@xcite is discussed ( section iii a ) . * the singular fade subspaces for the proposed dstc scheme are identified . the goal of minimizing the number of singular fade subspaces results in a new design criterion referred as the _ singularity minimization criterion _ for dstcs . it is shown that for a properly chosen dstc , most of the vector subspaces which were singular fade subspaces for the dnf protocol , are no longer singular fade subspaces for the dstc scheme . also , a criterion to maximize the coding gain of the proposed dstc scheme is obtained ( section iii b ) . * it is shown that for dstcs which are over @xmath95 where @xmath5 is a square qam or @xmath90-psk signal set , the coding gain is maximized when the generator matrices @xmath76 and @xmath77 at nodes a and b are unitary matrices . explicit construction of dstcs over qam and psk signal sets which satisfy the singularity minimization criterion and maximize the coding gain are provided . it is shown that for all dstcs over @xmath5 with unitary generator matrices @xmath76 and @xmath77 , the ml decoding complexity at r is @xmath96 for any arbitrary signal set and is @xmath97 for square qam signal sets . note that the brute force ml decoding complexity is @xmath98 ( section iv ) . * simulation results presented in section v show that at high snr , the dstc scheme provides large gains when compared to the conventional xor network code based on the dnf protocol and performs slightly better than the adaptive network coding scheme proposed in @xcite . * _ notations _ : * the complex number @xmath99 is denoted by @xmath100 the set of integers , gaussian integers , rational , real and complex numbers are respectively denoted as @xmath101 , \mathbb{q } , \mathbb{r}$ ] and @xmath102 all the vector spaces and vector subspaces considered in this paper are over the complex field @xmath103 unless explicitly mentioned otherwise . throughout , vectors are denoted by bold lower case letters and matrices are denoted by bold capital letters . let @xmath104 denote the circularly symmetric complex gaussian random vector with zero mean and covariance matrix @xmath105 where @xmath106 denotes the @xmath107 identity matrix . let @xmath108 denote the vector subspace over @xmath109 spanned by the complex vectors @xmath110 for a matrix @xmath111 @xmath112 and @xmath113 denotes its transpose and conjugate transpose respectively . for a vector subspace @xmath114 of a vector space , @xmath115 denotes the vector subspace @xmath116 and @xmath117 denotes the dimension of @xmath118 the all zero vector of length @xmath119 is denoted by @xmath120 for a square matrix @xmath111 let @xmath121 denote its rank and let @xmath122 denote its determinant . for a complex number @xmath59 @xmath60 and @xmath61 denote the real and imaginary parts of @xmath59 @xmath123 denotes its conjugate and @xmath124 denotes its absolute value . for a vector @xmath125 @xmath126 denotes its euclidean norm . for a matrix @xmath111 @xmath127 and @xmath128 respectively denote the row space and column space of @xmath129 @xmath130 denotes the expectation of @xmath131 in this section , to explain the notion of singular fade subspaces , we digress from the two - way relaying scenario and focus on the classical mimo system with collocated antennas . consider the classical mimo system with @xmath132 transmit antennas at the transmitter tx and @xmath133 receive antennas at the receiver rx , with @xmath134 being the @xmath135 complex fade coefficient matrix . the entries of the matrix @xmath134 are assumed to be i.i.d . and rician distributed . consider the spatial multiplexing of independent complex symbols at tx , i.e. , the received complex vector at rx is given by @xmath136 where @xmath137 is the transmitted message vector of length @xmath132 whose components independently take values from the signal set @xmath5 and @xmath138 is @xmath139 let @xmath140 denote the effective signal set at rx , i.e. , @xmath141 let @xmath142 denote the difference constellation of the signal set @xmath95 i.e. , @xmath143 the distances between two points in the effective constellation @xmath144 are of the form @xmath145 where @xmath146 for an @xmath147 mimo system , the channel fade coefficient matrix @xmath134 is said to be a @xmath148 if the minimum distance of the effective constellation @xmath144 is zero . the row space of a deep fade matrix is said to be a _ deep fade space_. let @xmath149 denote the @xmath150 row of @xmath151 since @xmath152 for the minimum distance of the effective constellation @xmath144 to be zero , all the vectors @xmath153 should fall in a vector subspace of the form @xmath154 for some @xmath155 in other words , for @xmath156 to be zero , the row space of @xmath134 should be a subspace of the vector subspace of @xmath93 of the form @xmath154 for some @xmath155 the vector subspaces of the form @xmath154 are referred to as the _ singular fade subspaces_. formally , a singular fade subspace can be defined as follows : a vector subspace @xmath157 of @xmath93 is said to be a singular fade subspace if all the vector subspaces of @xmath157 are deep fade spaces . note that @xmath158 is always a singular fade subspace referred to as the _ trivial singular fade subspace_. consider the @xmath94 miso system with spatial multiplexing with 4-psk signal set @xmath159 the difference constellation of 4 psk signal set has 9 points @xmath160 for this case , the set of fourteen singular fade subspaces , which are of the form @xmath161 where @xmath162 are given by , @xmath163 the fade coefficient matrix ( which is a row vector for this example ) is a deep fade matrix ( vector ) if the row space of the fade coefficient vector is a subspace of one of these 14 singular fade subspaces , i.e. , the fade coefficient vector should belong to one of these 14 vector subspaces . for example , @xmath164^t$ ] belongs to the vector subspace @xmath165 and is a deep fade matrix . note that the singular fade subspaces depend only on the number of transmit antennas @xmath132 and the signal set @xmath74 they are independent of the number of receive antennas @xmath166 as illustrated in the following example . consider the 2 @xmath58 2 mimo system with 4-psk signal set @xmath159 the set of 14 singular fade subspaces for this case is the same as that of @xmath94 miso system given in . for a fade coefficient matrix to be a deep fade matrix , both the rows should belong to one of these 14 vector subspaces . for example , @xmath167 is a deep fade matrix since @xmath168 and @xmath169 belong to the vector subspace @xmath170 the dimension of the singular fade subspace @xmath161 and the transmit diversity order of the pair - wise error event @xmath171 are inherently connected , where @xmath172 and @xmath173 . with spatial multiplexing , the transmit diversity order of the pair - wise error event @xmath174 is 1 while @xmath175 it is the presence of these @xmath176 dimensional singular fade subspaces that results in a transmit diversity order of 1 . the receive diversity order @xmath133 comes due to the fact that for a fade coefficient matrix to be a deep fade matrix , all the @xmath133 rows of the fade coefficient matrix should belong to the same singular fade subspace . the use of full diversity space times space time codes results in the maximum transmit diversity order @xmath177 in the next subsection , the connection between the singular fade subspaces of space time codes and transmit diversity order will be established . consider the case when tx uses a space time code @xmath178 of size @xmath179 where @xmath180 let @xmath181 denote a codeword matrix of the space time code , where @xmath182 where @xmath183 denotes the number of independent complex symbols transmitted . similar to the spatial multiplexing case , the effective constellation at rx which is a subset of @xmath184 can be defined . it is easy to verify that the minimum distance of the effective constellation at rx becomes zero when @xmath185 is a subspace of the vector subspace @xmath186 for some @xmath187 note that @xmath188 denotes the vector subspace @xmath189 the vector subspaces @xmath190 are the singular fade subspaces for the @xmath132 transmit antenna system with the space time code @xmath191 even though the probability that @xmath185 is a subspace of one of the singular fade subspaces is zero , with a non - zero probability @xmath185 falls in the neighbourhood of a subspace of one of the singular fade subspaces , which results in low values of the minimum distance of the effective constellation . the dimension of the singular fade subspace @xmath190 is equal to @xmath192 while the transmit diversity order for the pair - wise error event @xmath193 equals @xmath194 @xcite , where @xmath195 with every pair - wise error event @xmath171 we can associate a singular fade subspace @xmath196 among all the pair - wise error events , those error events for which the codeword difference matrix has the least rank determine the overall system transmit diversity order . equivalently , among all the pair - wise error events , those error events for which the associated singular fade subspace has the largest dimension will dominate the overall error probability . this is expected since among all the singular fade subspaces , the probability that @xmath185 falls in the neighbourhood of a subspace of the singular fade subspace , will be the largest for those singular fade subspaces which have the largest dimension . if the space time code is such that @xmath197 is full rank for all @xmath198 all the singular fade subspaces @xmath190 collapse to be the zero - dimensional trivial singular fade subspace @xmath199 thereby ensuring a transmit diversity order of @xmath132 for all the pair - wise error events . consider the @xmath94 miso system with alamouti space time code whose design matrix is given by @xmath200 since the design matrix is full rank for all choices of @xmath201 and @xmath202 the column space of the codeword difference matrix is always @xmath0 and hence all the singular fade subspaces collapse to be the zero dimensional trivial singular fade subspace @xmath203 equivalently , all the pair - wise error events @xmath204 have a transmit diversity order 2 . the full rank alamouti space - time code removed the effect of all the vector subspaces which were non - trivial singular fade subspaces for the spatial multiplexing system , thereby increasing the diversity order of all the pair - wise error events from 1 to 2 . for a @xmath205 generalized linear complex orthogonal design ( gcod ) @xcite , the design matrix @xmath206 constructed iteratively is given by , @xmath207 the codeword difference matrix for the gcod is full rank for any signal set . hence , irrespective of the signal set , the trivial singular fade subspace @xmath208 is the only singular fade subspace for the gcod . + consider the @xmath209 quasi - orthogonal design ( qod ) @xcite , whose codeword matrix is given by @xmath210 let @xmath211 irrespective of the signal set used , the minimum rank of the codeword difference matrix for the @xmath209 qod is 2 . for example , when @xmath212 and @xmath213 the rank of the codeword difference matrix is 2 . equivalently , there exists a non - trivial singular fade subspace , @xmath214 note that the 2 @xmath215 alamouti code removes the effect of the harmful non - trivial singular fade subspaces for any signal set . on the other hand , for the @xmath209 qod there exists non - trivial singular fade subspaces for any signal set . in general , a space time code can offer full transmit diversity for some but not all signal sets . in other words , for some signal set , a space time code might have only the trivial singular fade subspace , while for some other signal set , the same space time code might have non - trivial singular fade subspaces . for a space time code which does not offer full transmit diversity for a signal set , there would exist non - trivial singular fade subspaces . these are illustrated in the following example . [ example_2_ciod ] consider the @xmath78 co - ordinate interleaved orthogonal design ( ciod ) @xcite whose codeword matrices are of the form @xmath216 where @xmath217 let @xmath211 the codeword difference matrix is not full rank in the following two cases : + _ case 1 : _ @xmath218 and at least one out of @xmath219 and @xmath220 is non - zero . for this case , the singular fade subspace is given by @xmath221 + _ case 2 : _ @xmath222 and at least one out of @xmath223 and @xmath224 is non - zero . for this case , the singular fade subspace is given by @xmath225 + hence , there exists the following two non - trivial singular fade subspaces : @xmath226 and @xmath221 however , when the signal set is @xmath227 where @xmath228 is not a multiple of @xmath229 the @xmath78 ciod offers full transmit diversity . equivalently , there are no singular fade subspaces other than the trivial singular fade subspace for the @xmath78 ciod with the signal set @xmath227 when @xmath228 is not a multiple of @xmath230 [ example_4_ciod ] consider the @xmath209 ciod @xcite whose codeword matrices are of the form @xmath231 where @xmath232 for the 4-psk signal set considered , this stc does not offer full transmit diversity and there are pair - wise error events which have a transmit diversity order less than 2 . the determinant of the codeword difference matrix for this stc is given by , @xmath233 hence the code - word difference matrix is not full rank in the following two cases : + _ case 1 : _ @xmath234 and at least one out of @xmath235 is non - zero . for this case , the first two columns of the codeword difference matrices are zeros . the column span of the codeword difference matrix is @xmath236 and hence the corresponding singular fade subspace is given by @xmath237 + _ case 2 : _ @xmath238 and at least one out of @xmath239 is non - zero . similar to _ case 1 _ , it can be shown that the singular fade subspace for this case is given by @xmath240 hence , for the @xmath209 ciod , with 4-psk signal set @xmath241 there exists two non - trivial singular singular fade subspaces . similar to the @xmath78 ciod , when the signal set is a rotated 4-psk signal set , @xmath227 where @xmath228 is not a multiple of @xmath229 the @xmath209 ciod offers full transmit diversity and there are no non - trivial singular fade subspaces . + in general , the @xmath78 ciod given in example [ example_2_ciod ] and the @xmath242 ciod given in example [ example_4_ciod ] , offer full diversity for those signal sets for which the co - ordinate product distance ( cpd ) and @xmath243 is defined to be @xmath244 the cpd of a signal set is defined to minimum among all cpds between pairs of points in the signal set @xcite . ] is non - zero @xcite . equivalently , there are non - trivial singular fade subspaces for the @xmath78 and @xmath209 ciod , for signal sets whose cpd is non - zero . in fact , this is true for any generalized co - ordinate interleaved orthogonal design ( gciod ) , as illustrated in the next example . + consider the @xmath205 generalized co - ordinate interleaved orthogonal design ( gciod ) @xcite whose codeword design matrix is given by , @xmath245 the complex number @xmath246 where @xmath247 denotes @xmath248 modulo @xmath249 and @xmath250 is the codeword matrix of the gcod @xcite of size @xmath251 the determinant of the codeword difference matrix is given by @xmath252 the determinant is non - zero for those signal sets for which the cpd is non - zero and there are no non - trivial singular fade subspaces . for those signal sets for which the cpd is zero , the determinant becomes zero under the following two cases : _ case 1 : _ @xmath253 and at least one of the elements of the set @xmath254 @xmath255 is non - zero . it can be verified that the singular fade subspace for this case is given by @xmath256 where @xmath257 denotes the @xmath258 length vector whose @xmath40 component is one and all other components are zeros . + _ case2 : _ @xmath259 and at least one of the elements of the set @xmath260 @xmath255 is non - zero . for this case , the singular fade subspace is given by @xmath261 in the previous subsection , the notion of singular fade subspaces was introduced and its connection to the transmit diversity order of the mimo system with collocated antennas was established . since the ma phase of the two - way relaying scenario can be viewed as a virtual @xmath94 miso system , there exists singular fade subspaces for this case as well . in subsection [ sing_dnf ] , the singular fade subspaces for the two - way relaying scenario are identified . the reason why the adaptive network coding schemes based on the dnf protocol proposed in @xcite and @xcite- @xcite mitigate the effect of these harmful singular fade subspaces is discussed . in subsection [ sing_dstc ] , it is shown that minimizing the harmful effect of these singular fade subspaces can also be achieved by a proper choice of the dstc , without any need to adaptively change the network code at r according to channel conditions . let @xmath262 and @xmath263 from the discussion in section [ sing_mimo ] , it follows that the singular fade subspaces for the dnf protocol are of the form @xmath264 the ratio @xmath265 determines all the singular fade subspaces for the dnf protocol . in @xcite-@xcite , the ratio @xmath265 was called the _ singular fade state_. as mentioned earlier in section [ sing_mimo ] , @xmath266 and the diversity order for the pair - wise error event that a pair @xmath267 is wrongly decoded at r as @xmath268 ( denoted as @xmath269 ) are inherently connected . the diversity order for the error event @xmath269 is equal to @xmath270)=1 $ ] while @xmath271)=1.$ ] let @xmath272 denote the effective constellation at r. let @xmath273 denote the minimum distance of @xmath274 when @xmath275^t$ ] falls in one of the singular fade subspaces , @xmath273 becomes zero . even though the probability that the vector @xmath276^t$ ] belongs to a singular fade subspace is zero , @xmath273 is greatly reduced when @xmath276^t$ ] falls close to a singular fade subspace , a phenomenon referred as _ distance shortening_. for @xmath277 and @xmath278 the cnc algorithm @xcite avoids the distance shortening occurring in the neighbourhood of a singular fade subspace @xmath279 by ensuring that @xmath280 i.e. , r does not distinguish the pairs @xmath267 and @xmath281 which are said to be clustered together . in fact , for every realization of @xmath282 $ ] ( not necessarily in the neighbourhood of singular fade subspaces ) , the cnc algorithm chooses the network coding map which results in the best distance profile at r by appropriate clustering of the signal points . the scheme proposed in @xcite-@xcite avoids distance shortening in the neighbourhood of singular fade subspaces by proper choice of clustering for only the singular fade subspaces and not for every realization of the channel fade coefficients . consider the two singular fade subspaces : @xmath283 and @xmath284 the distance shortening which occurs in the neighbourhood of these singular fade subspaces is unavoidable , since the pairs @xmath267 and @xmath285 ( and also the pairs @xmath267 and @xmath286 ) which result in these singular fade subspaces can not be clustered together without violating the exclusive law . such singular fade subspaces are referred as the _ non - removable singular fade subspaces_. the dimension of these singular fade subspaces is one or equivalently , the error events @xmath287 and @xmath288 always result in diversity order one . the singular fade subspaces other than the non - removable singular fade subspaces are referred as the _ removable singular fade subspaces_. the removable singular fade subspaces are of the form @xmath289 which are dependent on the signal set @xmath5 used . the non - removable singular fade subspaces are @xmath290 and @xmath291 which are independent of the signal set used . owing to the presence of non - removable singular fade subspaces , the overall diversity order of the dnf protocol can not exceed one . from the discussion above , it is clear that there are two classes of singular fade subspaces : removable and non - removable . the non - removable singular fade spaces are created by the channel and is independent of the signal set used . whatever may be the choice of the network code , the harmful effects of these non - removable singular fade subspaces can not be mitigated . the harmful effect of the removable singular fade subspaces , which are created by the signal set , can be removed by a proper choice of the adaptive network coding map at r , as in @xcite and @xcite@xcite . to sum up , in the dnf protocol , the transmissions from the nodes a and b are allowed to interfere at r and the effect of mai is effectively mitigated by adaptively changing the network coding map , thereby removing the harmful effect of all the removable singular fade subspaces . let @xmath292 and @xmath293 then @xmath294 denotes a codeword difference matrix of the dstc , where @xmath295 is the codeword matrix of the dstc defined in . from the discussion in section [ sing_mimo ] , it follows that the singular fade spaces for the proposed dstc scheme are of the form @xmath296 consider the singular fade subspaces of the form @xmath297 and @xmath298 where @xmath299 the first row of the matrix @xmath300 has both the entries to be zero . hence , @xmath301 and the singular fade subspace @xmath302 by a similar reasoning , @xmath303 if the dstc codeword matrices are such that @xmath304 @xmath305 and @xmath306 all the singular fade subspaces @xmath307 collapse to be the trivial singular fade subspace @xmath308 equivalently , all the pair - wise error events @xmath309 have diversity order 2 . hence , for a properly chosen dstc , other than the trivial singular fade subspace , the singular fade subspaces are only the two non - removable singular fade subspaces , while for the dnf protocol , in addition , we had the removable singular fade subspaces . in this way , by a proper choice of dstc , the occurrence of the removable singular fade subspaces is avoided at the transmitting nodes itself , without any csit . hence , we have the following design criterion referred as the _ singularity minimization criterion _ for dstcs for two - way relaying : _ the dstc codeword difference matrices @xmath310 need to be full rank for all @xmath311 and @xmath312 _ to minimize the number of singular fade subspaces . dstcs satisfying the above criterion are referred as the _ singularity minimal _ dstcs . hence for a dstc which is singularity minimal , the only error events which result in diversity order 1 are of the form @xmath313 and @xmath314 hence , the overall coding gain is equal to minimum among all the non - zero singular values of the codeword difference matrices which are of the form @xmath315 and @xmath316 @xcite . note that the matrices @xmath315 and @xmath316 are of rank 1 and have only one non - zero singular value . we have the following coding gain criteria for singularity minimal dstcs : _ the minimum among all the non - zero singular values of the codeword difference matrices which are of the form @xmath315 and @xmath316 needs to be maximized_. consider the dstc @xmath317 this dstc is nothing but the scheme where a and b transmit in separate time slots , making sure that their transmissions do not interfere at the relay . even though this dstc avoids all the removable singular fade subspaces , the end - to - end rate in complex symbols per channel use is less than that of the dnf protocol . a signal set is said to be algebraic if all the signal points of the signal set are algebraic numbers over @xmath318 . if there exists a polynomial with coefficients from @xmath318 of which the number is a root . if there does not exist a polynomial with coefficients from @xmath318 of which the number is a root , the number is said to be transcendental @xcite . ] all the commonly used signal sets like qam and psk are algebraic signal sets . in this subsection , a class of dstcs which are singularity minimal for algebraic signal sets is provided . let @xmath319 be a full rank complex matrix . consider the class of dstcs whose codeword matrices are of the form @xmath320 the class of dstcs whose codeword design matrices are of the form given above are singularity minimal for all algebraic signal sets . the proof is as follows : for @xmath311 and @xmath312 at least one of the two components of @xmath321 as well as @xmath322 should be non - zero . hence , @xmath323 and @xmath324 since @xmath325 is transcendental is transcendental for all @xmath326 . ] whereas @xmath327 and @xmath328 are algebraic over @xmath329 the codeword difference matrix @xmath330 is full rank for all @xmath331 and @xmath332 since @xmath333 [ dstc_1 ] consider the case when @xmath334 let 4-psk be the signal set used at a and b. the dstc codeword matrix for this case is given by , @xmath335 a and b are made to transmit in two different time slots which results in low decoding complexity at r , since a s and b s transmissions can be decoded independently . it can be verified that the coding gain for this dstc is approximately 0.6877 . [ dstc_2 ] consider the case when @xmath336 let 4-psk be the signal set used at a and b. the dstc codeword matrix for this case is given by , @xmath337 the scaling factor of @xmath338 is to ensure unit average energy per symbol per time slot . it can be verified that the coding gain for this dstc is approximately 0.6877 , same as that of the dstc given in example [ dstc_1 ] . the coding gain of the dstcs given in examples [ dstc_1 ] and [ dstc_2 ] is approximately 0.6877 , which is less than the minimum distance of the unit energy 4-psk signal set , which is @xmath339 in the next section , it is shown that for dstcs over square qam and @xmath90-psk signal sets , the coding gain is upper bounded by the minimum distance of the signal set and explicit dstc constructions which achieve this bound with equality are provided . in this section , it is shown that the coding gain of the dstcs over square qam and @xmath90-psk signal sets are upper - bounded by the minimum distance of the signal set . in subsection [ dstc_unitary ] , a condition under which a singularity minimal dstc over square qam and @xmath90-psk signal set meets the upper bound with equality is obtained and explicit constructions of dstcs are provided . in subsection [ dstc_unitary_dc ] , the constructed dstc s are shown to be fast ml decodable , i.e. , the ml decoding complexity of the constructed dstcs is shown to be less than the brute - force decoding complexity which is @xmath340 note that the generator matrices @xmath76 and @xmath77 at a and b should be such that the average energy per time slot is unity , i.e. , @xmath341 and @xmath342 for singularity minimal dstc over @xmath95 where @xmath5 is a square qam or @xmath90-psk signal set , the coding gain is upper bounded by the minimum distance of the signal set @xmath74 see appendix a. in the following subsection , the condition under which the upper - bound given in the previous lemma is satisfied with equality is identified and explicit construction of dstcs are provided . the following proposition states that for dstcs over @xmath95 choosing @xmath76 and @xmath77 to be unitary matrices ensures that the upper - bound on the coding gain is satisfied with equality , for qam and psk signal sets . for singularity minimal dstcs over square qam or @xmath90-psk signal sets , the coding gain is maximized when the generator matrices @xmath76 and @xmath77 at a and b are unitary matrices . when @xmath76 and @xmath77 are unitary matrices , @xmath343 and also @xmath344 @xmath345 hence , @xmath346 and similarly + @xmath347 where @xmath348 denotes the minimum distance of @xmath74 the coding gain of the dstc is the minimum among all the non - zero singular values of the codeword difference matrices which are of the form @xmath315 and @xmath349 i.e. , the coding gain is equal to @xmath350 which is equal to @xmath351 in the following examples , constructions of singularity minimal dstcs whose generator matrices are unitary are provided . [ dstc_ldc1 ] consider the dstc over @xmath5 for which @xmath352 and @xmath353 where @xmath354 @xmath355 @xmath356 and @xmath357 the dstc codeword matrix is of the form @xmath358 the codeword difference matrix @xmath310 is full rank for all @xmath359 and @xmath360 when the signal points belong to @xmath361 $ ] @xcite . hence the dstc is singularity minimal for all signal sets whose signal points belong to @xmath361.$ ] also , since @xmath76 and @xmath77 are unitary , for square qam signal set , the dstc maximizes the coding gain . the dstc given in construction [ dstc_ldc1 ] was constructed in @xcite towards satisfying the design criterion formulated in @xcite for the two - user non - cooperative multiple access channel ( mac ) . in @xcite , the dstc given in the above example was shown to be dmt optimal for two - user mac . [ dstc_ldc2 ] consider the dstc for which @xmath362 and @xmath363 where @xmath364 the dstc codeword matrix @xmath365 is given by , @xmath366 for a complex number @xmath367 let @xmath368 denote the smallest field containing @xmath318 and @xmath369 it is shown in lemma [ const2_proof ] below that choosing @xmath370 ensures singularity minimality for signal sets ( for example qam ) whose signal points belong to @xmath371 and choosing @xmath372 ensures singularity minimality for signal sets ( for example @xmath90-psk ) whose signal points belong to @xmath373 also , since @xmath76 and @xmath77 are unitary , this dstc maximizes the coding gain , for square qam and @xmath90-psk signal sets . the advantage of this construction over construction [ dstc_ldc1 ] is that encoding at node a is simple , since it does not involve any linear combination of @xmath43 and @xmath374 [ const2_proof ] for the dstc given in construction [ dstc_ldc2 ] , choosing @xmath370 ensures singularity minimality for signal sets whose points belong to @xmath371 and choosing @xmath372 ensures singularity minimality for signal sets whose signal points belong to @xmath373 the proof is given for the case when the signal points belong to @xmath375 the proof for the case when the signal points belong to @xmath376 is exactly similar and is omitted . + let @xmath377 and @xmath378 where @xmath379 and @xmath86 to prove singularity minimality , it needs to be shown that when at least one out of @xmath380 and @xmath381 ( @xmath382 and @xmath328 ) is non - zero , the codeword difference matrix is full rank . the ratios @xmath383 and @xmath384 belong to @xmath371 while @xmath385 does not belong to @xmath375 hence , @xmath386 and @xmath387 since @xmath388 and @xmath389 @xmath390 and @xmath391 belong to @xmath392 where @xmath393 denotes the smallest filed containing @xmath394 and @xmath395 the determinant of the codeword difference matrix is given by , @xmath396 the determinant is non - zero since the ratio @xmath397 belongs to @xmath392 while @xmath398 does not belong to @xmath399 after the two ma phases , r jointly decodes for the two message vectors @xmath400 and @xmath401 of a and b respectively . in general , the complexity of this joint ml decoding at r is @xmath402 where @xmath24 is the cardinality of the signal set @xmath74 the choice of the generator matrices @xmath76 and @xmath77 being unitary not only maximizes the coding gain for qam and psk signal sets , but also results in a reduced decoding complexity at r. the following proposition states that when conditional ml decoding @xcite , @xcite is employed , the decoding complexity of the dstcs constructed in the previous section for which the generator matrices @xmath76 and @xmath77 are unitary is @xmath96 for any arbitrary signal set and is @xmath97 for square qam signal set . note that the brute force decoding complexity is @xmath340 when the generator matrices of the singularity minimal dstc over @xmath5 are unitary , the decoding complexity using conditional ml decoding is @xmath96 when the signal set @xmath5 is arbitrary and is @xmath97 when the signal set @xmath5 is square qam . see appendix b. compared with the dnf protocol , the decoding complexity is more for singularity minimal coding gain maximal dstcs over @xmath74 for the dnf protocol , the decoding complexity is @xmath97 for non square qam signal sets while it is @xmath403 for square qam signal set . @xmath35 can be decoded with constant decoding complexity by rounding off to the nearest integer , which results in an overall decoding complexity of @xmath404 as indicated by the simulation results in the next section , the proposed dstc offers slightly better performance than the adaptive network coding scheme and eliminates the need for adaptive switching of network coding maps at r. but this comes at the cost of increased decoding complexity at r. all the simulation results presented are for the case when the end nodes use 4-psk signal set . by ` dstc 1 ' and ` dstc 2 ' we refer to the dstcs given in construction [ dstc_ldc1 ] and construction [ dstc_ldc2 ] respectively . as a reference scheme , we consider the scheme in which xor network code is used irrespective of channel conditions and no dstc is employed , which is referred as ` xor n / w code ' . assuming unit noise variances at all the nodes , the average energies of the transmissions at the nodes , which are assumed to be equal , is defined to be the signal to noise ratio ( snr ) . the proposed dstc scheme is also compared with the adaptive network coding schemes proposed in @xcite and @xcite-@xcite . since for 4-psk signal set , the adaptive network coding scheme based on the nearest neighbour clustering ( nnc ) algorithm proposed in @xcite and the scheme based on latin squares proposed in @xcite-@xcite turn out to be the same , without distinguishing them we refer to both as ` adaptive n / w code ' . [ fig : ber_comp1 ] shows the snr vs ber performance for different schemes for the case when all the fading coefficients are i.i.d . and rayleigh distributed . in fig . [ fig : ber_comp2 ] and fig . [ fig : ber_comp3 ] similar plots are shown for a rician fading scenario with rician factors of 0 db and 5 db respectively . from fig . [ fig : ber_comp1]-[fig : ber_comp3 ] , it can be seen that the diversity order is one for all the schemes . also , it can be seen that at high snr , both ` dstc 1 ' as well as ` dstc 2 ' offer nearly the same performance and they perform better than the ` xor n / w code ' as well as the ` adaptive n / w code ' . for a rayleigh fading scenario , at high snr , the dstcs offer a gain of 2 db over ` xor n / w code ' while the ` adaptive n / w code ' offers a gain of about 0.5 db over the ` xor n / w code ' . for a rician factor of 0 db , at high snr , the dstcs offer a gain of 2 db over ` xor n / w code ' while the ` adaptive n / w code ' offers a gain of about 1.2 db over the ` xor n / w code ' . for a rician factor of 5 db , at high snr , the dstcs offer a gain of 5.5 db over xor n / w code while the ` adaptive n / w code ' offers a gain of about 4 db over the ` xor n / w code ' . the reason why the dstc based scheme performs better than the adaptive n / w coding scheme is as follows : during the bc phase always a 4 point signal set is used for the dstc based scheme , while depending on channel conditions 4 point or 5 point signal set is used for the adaptive network coding scheme @xcite,@xcite . a dstc scheme was proposed for the two - way relaying scenario . it was shown that deep channel fades occur when the channel fade coefficient vector falls in a finite number of vector subspaces called the singular fade subspaces . the connection between the dimension of these vector subspaces and the transmit diversity order was established . design criterion to minimize the number of singular fade subspaces for the dstc scheme and maximize the coding gain were obtained . explicit low decoding complexity constructions of dstcs were provided . the problem of constructing singularity minimal dstcs with decoding complexity same as that of the dnf protocol , without sacrificing the coding gain , remains open . extending the dstc scheme for two - way relaying with multiple antennas and multi - way relaying are possible directions for future work . this work was supported partly by the drdo - iisc program on advanced research in mathematical engineering through a research grant as well as the inae chair professorship grant to b. s. rajan . 160 s. zhang , s. c. liew and p. p. lam , `` hot topic : physical - layer network coding '' , acm mobicom 06 , pp . 358365 , sept . p. popovski and h. yomo , `` the anti packets can increase the achievable throughput of a wireless multi hop network '' , ieee icc 2006 , istanbul , turkey , june 2006 . s. j. kim , p. mitran and v. tarokh , `` performance bounds for bidirectional coded cooperation protocols '' , ieee trans . theory , vol . 54 , pp.52355241 , nov . 2008 . p. popovski and h. yomo , `` physical network coding in two - way wireless relay channels '' , ieee icc , glasgow , scotland , june 2007 . l. song , y. li , a. huang , b. jiao and a. v. vasilakos , `` differential modulation for bidirectional relaying with analog network coding '' , ieee trans on signal processing , vol . 58 , no . 7 , july 2010 . t. koike - akino , p. popovski and v. tarokh , `` optimized constellation for two - way wireless relaying with physical network coding '' , ieee journal on selected areas in comm . 773787 , june 2009 . t. koike - akino , p. popovski and v. tarokh , `` denoising strategy for convolutionally - coded bidirectional relaying '' , ieee icc 2009 , dresden , germany , june 2009 . b. hern and k. narayanan , `` multilevel coding schemes for compute - and - forward '' , ieee isit , st . petersburg , russia , july 2011 . m. p. wilson and k. narayanan , `` power allocation strategies and lattice based coding schemes for bi - directional relaying '' , ieee isit , seoul , korea , july 2009 . t. koike - akino , `` adaptive network coding in two - way relaying mimo systems '' , ieee globecom 2010 . vishnu namboodiri , vijayvaradharaj muralidharan and b. sundar rajan , `` wireless bidirectional relaying and latin squares , '' ieee wcnc 2012 , paris , france , april 2012 ( available online at arxiv : 1110.0084v2 [ cs.it ] , 16 nov . vijayvaradharaj muralidharan , vishnu namboodiri , and b. sundar rajan , `` channel quantization for physical layer network - coded two - way relaying , '' ieee wcnc 2012 , paris , france , april 2012 ( available online at arxiv : 1109.6101v2 [ cs.it ] , 16 nov . vijayvaradharaj t. muralidharan and b. sundar rajan , `` wireless network coding for mimo two - way relaying using latin rectangles '' , available online at arxiv : 1201.4477v1 [ cs.it ] , 21 jan . 2012 . t. cui , f. gao , t. ho and a. nallanathan , `` distributed space time coding for two - way wireless relay networks , '' ieee trans . on signal processing , 2 , feb . 2009 . v. tarokh , n. seshadri and a. r. calderbank , `` space time codes for high data rate wireless communication : performance criterion and code construction , '' ieee trans . on info . theory , vol . 2 , march 1998 . w. su and x .- g.xia , `` on space time block codes from complex orthogonal designs , '' wireless pers . 1 , pp . 1 - 26 , apr . 2003 . hamid jafarkhani , `` a quasi - orthogonal space time block code , '' ieee trans . 49 , pp . 1 - 4 , jan 2001 . zafar ali khan and b. sundar rajan , `` single - symbol maximum likelihood decodable linear stbcs , '' ieee trans . theory , vol . 5 , may 2006 . n. jacobson , basic algebra i , @xmath405 edition , new york : w . h. freeman , 1985 . maya badr and jean - claude belfiore , `` distributed space - time block codes for the non - cooperative multiple access channel , '' int . zurich seminar on communications ( izs ) , march 12 - 14 , 2008 . m. gartner and h. bolcskei , `` multiuser space - time / frequency code design , '' isit 2006 , seattle , pp . 28192823 , 2006 . e. biglieri , y. hong and e. viterbo , `` on fast - decodable space - time block codes , '' ieee trans . theory , vol . 524 - 530 , feb . 2009 . k. pavan srinath and b. sundar rajan , `` low ml decoding complexity , large coding gain , full rate , full - diversity stbcs for 2 @xmath58 2 and 4 @xmath58 2 mimo systems , '' ieee journal of selected topics in signal processing , vol . 3 , no.6 , december 2009 . since @xmath406 is hermitian , it is unitarily diagonalizable , i.e. , @xmath407 where @xmath408 is a unitary matrix and @xmath409 is a diagonal matrix with diagonal entries denoted as @xmath410 and @xmath411 note that @xmath410 and @xmath412 are non - negative since @xmath406 is positive semi - definite . let @xmath413 we have , @xmath414 since @xmath415 for square qam and @xmath90-psk signal sets . since @xmath416 we have @xmath417 the coding gain of the dstc is the minimum among all the non - zero singular values of the codeword difference matrices which are of the form @xmath315 and @xmath349 i.e. , the coding gain is equal to @xmath418 let @xmath348 denote the minimum distance of the signal set @xmath74 consider @xmath419 where @xmath420.$ ] let @xmath421 $ ] and @xmath422 $ ] denote the rows of @xmath423 for @xmath424,$ ] @xmath425 hence , we have , @xmath426 similarly , we have , + @xmath427 since @xmath408 is unitary @xmath428 and @xmath429 therefore , we have , @xmath430 since @xmath408 is unitary , @xmath431 for a given @xmath410 and @xmath432 the upper - bound in is maximized over all @xmath433 when the two terms inside @xmath434 are equal , i.e. , @xmath435 for which @xmath436 and this maximum value is equal to @xmath437 since , @xmath438 the maximum value of the upper - bound in is less than or equal to @xmath439 hence , @xmath440 similarly , it can be shown that @xmath441 is also upper - bounded by @xmath351 hence , the coding gain of the dstc over square qam or @xmath90-psk signal set is upper - bounded by @xmath351 this completes the proof . to prove the proposition , we adopt a procedure similar to the one used in @xcite . let @xmath442^t,$ ] @xmath443^t$ ] and @xmath444^t.$ ] the vector @xmath445 can be written as @xmath446 where @xmath447 is a @xmath448 real matrix whose entries are functions of @xmath1 and @xmath21 determined by the dstc . using @xmath449 decomposition , the matrix @xmath447 can be decomposed as @xmath450 where @xmath451 is a orthogonal matrix and @xmath452 can be written as @xmath453,$ ] with @xmath454 @xmath455 being an upper - triangular matrix . the joint ml decoding metric at r is given by @xmath456 where @xmath457 for a singularity minimal dstc over @xmath95 let the generator matrices be @xmath458 and @xmath459 where @xmath408 and @xmath460 are unitary matrices . let @xmath461 and @xmath462 denote the @xmath40 rows of @xmath408 and @xmath460 respectively . then the weight matrices of the dstc defined in are given by , @xmath463 and @xmath464 we have , @xmath465 and similarly , @xmath466 hence , @xmath467 where @xmath468 denotes the @xmath78 null matrix . also , @xmath469 since @xmath470 and @xmath471 are orthogonal vectors . hence , @xmath472 . similarly , using the fact that @xmath408 and @xmath460 are unitary matrices , it can be shown that the following pairs of matrices are also hurwitz - radon orthogonal and @xmath473 are said to be hurwitz - radon orthogonal if @xmath474 : @xmath475 @xmath476 @xmath477 @xmath478 @xmath479 @xmath480 @xmath481 @xmath480 @xmath482 @xmath480 @xmath483 @xmath484 @xmath485 @xmath486 @xmath487 let @xmath488 denote the @xmath40 component of the vector @xmath489 the @xmath40 and @xmath490 columns of @xmath447 are orthogonal and hence the @xmath491 entry of @xmath492 ( @xmath493 ) is zero for all realizations of @xmath1 and @xmath21 if and only if the weight matrices of the dstc corresponding to the symbols @xmath488 and @xmath494 are hurwitz - radon orthogonal ( follows from theorem 2 , @xcite ) and @xmath494 need to be hurwitz - radon orthogonal , for the @xmath491 entry of @xmath492 ( @xmath493 ) to be zero for all realizations of @xmath1 and @xmath21 and hence the ` only if ' part also holds . ] . hence the matrix @xmath492 is of the form given below . note that @xmath496 denotes possible non - zero entries . the claim is that all the entries denoted by @xmath496 are non - zeros . it is clear that all the diagonal entries are non - zeros . for the @xmath497 entry in to be a zero , @xmath498 which implies that @xmath470 and @xmath499 are orthogonal vectors . then the vector @xmath499 should belong to the one - dimensional subspace which is orthogonal to @xmath500 since @xmath471 also belongs to this one - dimensional subspace and both @xmath499 as well as @xmath471 are of unit norm , @xmath501 for some angle @xmath502 in that case , the dstc codeword difference matrix is of the form @xmath503 which is not full rank when @xmath504 @xmath505 and hence the singularity minimization criterion is violated . hence , @xmath497 entry shown by @xmath496 in is non - zero . by a similar argument , it can be shown that the other non - diagonal entries denoted by @xmath496 in are non - zeros . from the matrix @xmath492 given in , it can be seen that conditioning on the variables @xmath46 and @xmath506 the symbols @xmath43 and @xmath56 can be decoded independently @xcite . since the total number of choices for @xmath46 and @xmath57 is @xmath25 and independently decoding @xmath43 and @xmath56 requires @xmath507 computations , the decoding involves @xmath508 computations and hence the decoding complexity at r is @xmath509 for square qam signal sets , the decoding complexity can be further reduced , since the real and imaginary parts independently take values . from , it can be seen that conditioning on @xmath46 and @xmath506 the real and imaginary parts of @xmath43 as well as @xmath56 can be decoded independently . since decoding the real and imaginary points of a signal point in qam signal set is of constant complexity independent of @xmath24(decoding can be done by rounding off to the nearest integer @xcite ) , the ml decoding complexity is @xmath97 for square qam signal sets . this completes the proof .

we construct triples of commuting real structures on the moduli space of higgs bundles , whose fixed loci are branes of type , and . we study the real points through the associated spectral data and describe the topological invariants involved using , and equivariant-theory . = 1.1

1309.1195

the moduli space @xmath6 of @xmath7-higgs bundles on a compact riemann surface @xmath8 is the space of solutions to the gauge theoretic hitchin equations on the surface , where @xmath7 is a complex reductive lie group . the smooth locus of @xmath6 is a hyperkhler manifold , so there are complex structures @xmath9 obeying the same relations as the imaginary quaternions . this paper is concerned with the study of several naturally defined real structures on this moduli space . in fact , as it is impossible for an involution to be anti - holomorphic in all three complex structures it is natural to consider not one but three real structures simultaneously . we introduce a naturally defined triple of commuting real structures @xmath10 on @xmath6 . we give geometric interpretations for these real structures and use spectral data to build up a detailed picture of their fixed point sets . along the way we encounter various forms of @xmath5-theory as a convenient tool for studying the connected components of these fixed point sets . + the real structures @xmath10 are defined in section [ sec : real ] . the involution @xmath11 is defined by taking a real form @xmath12 of @xmath7 . amongst the fixed points of @xmath11 are solutions to the hitchin equations with holonomy in the real form @xmath12 . moreover , when @xmath12 is the split real form , we prove in theorem [ proporder2 ] that the fixed points of @xmath11 are points of order @xmath13 in the fibres of the hitchin fibration , as seen in @xcite . the involution @xmath14 , introduced in @xcite is defined by choosing a real structure on @xmath8 , an anti - holomorphic involution @xmath15 . amongst the fixed points of @xmath14 are representations of the orbifold fundamental group of the action of @xmath16 on @xmath8 . combining these two involutions we obtain a third involution @xmath17 . amongst the fixed points of @xmath18 are pseudo - real higgs bundles , introduced in @xcite . the involutions @xmath19 are well - behaved with respect to the hitchin fibration and by restriction we find that their fixed point sets are real integrable systems . we also give a description of the fixed point sets in terms of orbifold representations in section [ secrep ] . + section [ sec : gl ] considers in detail the case of the general linear group @xmath20 , with real form @xmath21 . in this case a higgs bundle @xmath22 is a rank @xmath23 holomorphic vector bundle @xmath24 and a holomorphic @xmath25-form valued endomorphism @xmath26 of @xmath24 . if @xmath27 , then as we recall in the paper polystable higgs bundles @xmath22 correspond to bundles with flat connection @xmath28 , where @xmath29 has reductive holonomy . we say that @xmath29 is simple if the only constant endomorphisms of @xmath24 are multiples of the identity . restricting to simple , reductive holonomy we find : * fixed points of @xmath11 are flat bundles with holonomy in @xmath30 or @xmath31 . * fixed points of @xmath14 are flat bundles with involution @xmath32 covering @xmath16 and preserving @xmath29 . * fixed points of @xmath18 are flat bundles with anti - linear isomorphism @xmath32 , covering @xmath16 , preserving @xmath29 and with @xmath33 . as a first step towards identifying the connected components of the fixed point sets , we may consider the topological data associated to the underlying bundle as follows : * for @xmath11 , the bundle @xmath24 carries a real or quaternionic structure , thus defines a class @xmath34 $ ] in @xmath35 or @xmath36 , real or quaternionic @xmath5-theory . * for @xmath14 , the bundle @xmath24 carries a lift of the @xmath37-action on @xmath8 , thus defines a class @xmath34 \in k^0_{\mathbb{z}_2}(\sigma)$ ] in @xmath37-equivariant @xmath5-theory . * for @xmath18 , the bundle @xmath24 carries a real or quaternionic structure in the sense of atiyah @xcite , hence a class @xmath34 $ ] in @xmath38 or @xmath39 . in section [ sec : gl ] we determine these @xmath5-theory groups and show that the @xmath5-theory classification is sharp in the sense that one can recover the bundle plus additional topological data up to isomorphism from the @xmath5-theory class . + as an example of the utility of these @xmath5-theory classes , we consider the following construction . let @xmath40 $ ] with involution @xmath41 . the quotient @xmath42 is a @xmath43-manifold with boundary @xmath44 . then , we show the following : + * theorem [ 3man ] * _ let @xmath28 be a fixed point of @xmath18 with simple holonomy . then @xmath29 extends over @xmath45 as a flat connection if and only if the class @xmath34 \in \tilde{k}^0_{\mathbb{z}_2}(\sigma)$ ] in reduced equivariant @xmath5-theory is trivial . _ + in section [ sec : spectral ] we study the spectral data associated to fixed points . as we recall , spectral data consists of a branched cover @xmath46 called the spectral curve and a line bundle @xmath47 . the bundle @xmath24 and higgs field @xmath26 are recovered by pushing down @xmath48 to @xmath8 . for fixed points , the additional structure on @xmath24 determines similar structure on the spectral line @xmath48 and we recover the @xmath5-theory classes as push - forwards from the spectral curve to @xmath8 . the push - forward in the @xmath3-theory case was recently used by hitchin for this purpose @xcite . we complement this by determining explicit expressions for the push - forwards in equivariant @xmath5-theory and @xmath4-theory in theorems [ prop : eqpush ] and [ prop : krpush ] . the image of the push - forward maps @xmath49 and @xmath50 describe which topological classes of bundle with real structure can be given a reductive flat connection ( with smooth spectral curve ) . this is the analogue of the milnor - wood type inequalities for the topological invariants obtained through @xmath5-theory . + along the paper we adopt the physicists language in which a lagrangian submanifold is called an _ a - brane _ and a complex submanifold a _ b - brane_. a submanifold of a hyperkhler manifold may be of type @xmath51 or @xmath52 with respect to each of the complex structures and we may speak of branes of type @xmath53 and @xmath2 . under this classification the fixed point sets of @xmath10 are branes of type @xmath54 and @xmath2 respectively . the main reason for considering branes is the connection to mirror symmetry and the geometric langlands program . this program asserts that the moduli spaces @xmath55 are in duality , where @xmath56 is the langlands dual group of @xmath7 . according to this duality , specifically homological mirror symmetry , there should be an equivalence of categories of branes on @xmath6 and @xmath57 under which there are exchanges @xmath58 , @xmath59 and @xmath60 . we conclude our work with section [ sec : duality ] , in which we speculate on how this duality acts on the fixed point sets of @xmath10 . * acknowledgements : * the authors would like to thank s. bradlow , o. garca - prada , n. hitchin , and f. schaffhauser for helpful comments . we review higgs bundles , the hyperkhler structure of their moduli space and recall the hitchin fibration . let @xmath8 be a riemann surface of genus @xmath61 with canonical bundle @xmath5 , and @xmath7 a complex lie group with lie algebra @xmath62 . we shall assume throughout the paper that @xmath7 is reductive . given a principal @xmath7-bundle @xmath63 on @xmath8 , we let @xmath64 denote the corresponding adjoint bundle . a @xmath7-higgs bundle on a riemann surface @xmath8 is a pair @xmath65 , where @xmath66 is a holomorphic structure on a principal @xmath7-bundle @xmath63 and @xmath26 is a holomorphic section of @xmath67 . in the case of @xmath20 a @xmath7-higgs bundle is equivalent to a classical higgs bundle @xmath22 , consisting of a rank @xmath23 holomorphic bundle @xmath24 and a holomorphic map @xmath68 . + in order to define a moduli space of such pairs we shall recall the notions of stability and @xmath69-equivalence . let @xmath70 be the slope of the vector bundle @xmath24 . we say that a @xmath71-higgs bundle @xmath22 is semi - stable if for every subbundle @xmath72 such that @xmath73 we have @xmath74 , and it is stable if one has a strict inequality . if one can write @xmath75 for @xmath76 stable pairs such that @xmath77 , then we say the higgs bundle is polystable . to define @xmath69-equivalence consider a strictly semi - stable higgs bundle @xmath78 . as it is not stable , @xmath24 admits a subbundle @xmath72 of the same slope which is preserved by @xmath26 . if @xmath79 is a subbundle of @xmath24 of least rank and same slope which is preserved by @xmath26 , it follows that the pair @xmath80 is stable . then , by induction one obtains a flag of subbundles @xmath81 where @xmath82 for @xmath83 , and where the induced higgs bundles @xmath84 are stable . this is the _ jordan - hlder filtration _ of @xmath24 , and it is not unique . however the graded object @xmath85 is unique up to isomorphism . two semi - stable higgs bundles @xmath22 and @xmath86 are said to be @xmath69-equivalent if @xmath87 . for a stable pair @xmath22 the associated graded object coincides with @xmath22 and the @xmath69-equivalence class is just the isomorphism class of the original pair . more generally each @xmath69-equivalence class contains a unique polystable object . through the above definitions , one may construct the moduli space @xmath88 of @xmath69-equivalence classes of classical semi - stable higgs bundles of rank @xmath23 and degree @xmath89 , or equivalently , the moduli space of polystable rank @xmath23 degree @xmath89 higgs bundles . this space is a quasi - projective scheme of complex dimension @xmath90 and contains an open subscheme @xmath91 corresponding to the moduli scheme of stable pairs @xcite . when @xmath23 and @xmath89 are coprime semi - stable implies stable and the moduli space @xmath88 is smooth . + by extending the stability notions to @xmath7-higgs bundles , one can define stable , semi - stable and polystable @xmath7-higgs bundles ( see e.g. , @xcite , @xcite ) . then one can construct a corresponding moduli space of polystable @xmath7-higgs bundles , denoted @xmath92 or @xmath6 . the dimension of @xmath93 is @xmath94 . the connected components of @xmath93 are in bijection with isomorphism classes of principal @xmath7-bundles and these are parametrised by @xmath95 @xcite . for @xmath96 we let @xmath97 denote the corresponding connected component of @xmath93 . we shall briefly recall here the construction of a hyperkhler metric on @xmath93 , obtained by an infinite dimensional hyperkhler reduction . for simplicity we consider the case where @xmath7 is semi - simple . the reductive case requires only minor alterations such as modifying the hitchin equations to allow for projectively flat connections . fix a hermitian metric @xmath98 on @xmath8 and an anti - holomorphic involution @xmath99 whose fixed point set @xmath100 gives the compact real form of @xmath7 . let @xmath63 be a principal @xmath7-bundle and fix a reduction of structure to @xmath100 . the reduction of structure determines a corresponding anti - linear involution @xmath101 on the adjoint bundle . given a section @xmath102 of @xmath64 we write @xmath103 for @xmath104 . we shall denote by @xmath105 the killing form on @xmath62 . + the space @xmath106 of holomorphic structures on @xmath63 is an affine space over @xmath107 , hence the cotangent bundle @xmath108 is an infinite dimensional flat hyperkhler manifold . the tangent space to @xmath108 at any point can be naturally identified with the direct sum @xmath109 and we shall denote by @xmath110 tangent vectors to this space . in terms of this identification the metric on @xmath108 is given by @xmath111 moreover there are compatible complex structures @xmath9 defined by @xmath112 satisfying the usual quaternionic relations . this defines the hyperkhler structure on @xmath108 . we shall denote by @xmath113 the corresponding khler forms @xmath114 for a pair @xmath115 let @xmath116 be the associated unitary connection and @xmath117 the curvature of @xmath118 . the group @xmath119 of unitary gauge transformations acts on @xmath108 preserving the hyperkhler structure . this action has a hyperkhler moment map @xmath120 , \overline{\partial}_a \phi ) $ ] . the hyperkhler quotient @xmath121 of this action is then the moduli space of solutions to the hitchin equations @xmath122 = 0 , \ ; \ ; \ ; \ ; \overline{\partial}_a \phi = 0,\ ] ] modulo unitary gauge transformations . from this we obtain a hyperkhler structure on the smooth points of the moduli space of solutions to the hitchin equations . + for @xmath7 semi - simple it is a result of hitchin @xcite and simpson @xcite that a @xmath7-higgs bundle @xmath65 is gauge equivalent to a solution of the hitchin equations ( [ heq ] ) if and only if it is polystable . this is used to establish an isomorphism between the moduli space @xmath93 of polystable @xmath7-higgs bundle and the moduli space of solutions to the hitchin equations . in particular this gives a hyperkhler structure on the smooth points of @xmath93 . + a solution to the above hitchin equations ( [ heq ] ) defines an associated flat @xmath7-connection @xmath123 . from the results of donaldson @xcite and corlette @xcite , the mapping @xmath124 gives an isomorphism between the higgs bundle moduli space @xmath92 and the character variety @xmath125 of reductive representations of @xmath126 in @xmath7 ( the definition of reductive representations and further details are recalled in section [ secrep ] ) . we say that a polystable higgs bundle @xmath127 is simple if the only covariant constant gauge transformations of the associated connection @xmath29 are those valued in the centre @xmath128 of @xmath7 . in particular , one has that simple higgs bundles give smooth points on the moduli space @xcite . + the hyperkhler quotient construction carries over to the case of a reductive group requiring only a small modification . consider for example the case @xmath20 . the hitchin equations for a higgs bundle pair @xmath22 should be modified to @xmath129 = -2 \pi i \mu(v ) vol_\sigma , \ ; \ ; \ ; \ ; \overline{\partial}_a \phi = 0,\ ] ] where @xmath130 is the slope of @xmath24 and @xmath131 the volume form on @xmath8 . the associated connection @xmath123 is now only projectively flat and we obtain a representation of a central extension of @xmath126 . the moduli space @xmath132 has a natural complex lagrangian fibration over a vector space @xmath133 . to define this fibration let @xmath134 be a homogeneous basis for the algebra of invariant polynomials on @xmath62 , of degrees @xmath135 . following @xcite , the hitchin fibration is given by @xmath136 the map @xmath137 , referred to as the hitchin map , is a proper map for any choice of basis ( see ( * ? ? ? * section 4 ) for details ) . given @xmath96 consider the restricted hitchin map @xmath138 . for each component @xmath97 the smooth fibres of @xmath137 are connected @xcite complex lagrangian submanifolds with respect to the holomorphic symplectic form @xmath139 . we have @xmath140 , and the hitchin map gives each component @xmath97 the structure of an algebraically completely integrable system @xcite . in particular @xmath137 is generically a submersion and the generic fibres are abelian varieties . having defined the moduli space @xmath141 of @xmath7-higgs bundles on a compact riemann surface @xmath8 , we shall now consider the three natural involutions @xmath142 and @xmath18 on it and study their fixed point sets . the moduli space @xmath141 admits several distinct real structures , a phenomenon related to its hyperkhler geometry . first consider a real form of @xmath7 , given by the fixed point set @xmath12 of an anti - holomorphic involution @xmath143 . for any real form @xmath144 we can find an anti - holomorphic involution @xmath99 commuting with @xmath144 , whose fixed point set defines the compact real form of @xmath7 ( for details , see @xcite ) . the cartan involution of the real form @xmath12 is the holomorphic involution @xmath145 . from @xmath144 we obtain an involution @xmath11 on the moduli space of higgs bundles , given by @xmath146 the involution @xmath11 is an isometry of the hyperkhler metric , holomorphic in @xmath147 and anti - holomorphic in @xmath148 , @xmath5 . therefore its fixed point set is a brane of type @xmath0 . + a second way to obtain a real structure on @xmath141 is to consider real structures on the surface @xmath8 . let @xmath15 be an anti - holomorphic involution , a real structure on @xmath8 . viewing @xmath141 as a moduli space of connections , the action of @xmath16 on @xmath8 induces an involution @xmath14 by pullback of connections . in terms of higgs bundles , @xmath14 is given by @xmath149 we have seen in @xcite that @xmath14 is an isometry which is holomorphic in @xmath148 and anti - holomorphic in @xmath147 , @xmath5 . therefore its fixed point set is a brane of type @xmath1 . lastly , by combining a real structure on the group @xmath7 with a real structure on the surface @xmath8 , we obtain a third class of involution @xmath18 , given by @xmath150 since @xmath17 we have that @xmath18 is an isometry , holomorphic in @xmath5 and anti - holomorphic in @xmath147 , @xmath148 . therefore its fixed point set is a brane of type @xmath2 . the fixed point sets of @xmath10 meet the smooth points of @xmath93 in complex lagrangian submanifolds . the case of @xmath14 was established in @xcite and the same argument applies to the other involutions . the fixed point sets have interpretations in terms of the corresponding holonomy representations . the simplest case are fixed points of @xmath14 , which in terms of connections are simply those connections @xmath29 which are isomorphic to their pullback @xmath151 . for the involution @xmath18 we have fixed points given by pseudo - real higgs bundles as defined in @xcite . to discuss fixed points of @xmath11 , recall that the involutions @xmath152 induce corresponding involutions on the lie algebra @xmath62 . we shall denote by @xmath153 the fixed point set of @xmath144 , the lie algebra of @xmath12 . we obtain a decomposition @xmath154 of @xmath153 into the @xmath155-eigenspaces of @xmath156 , the cartan decomposition of @xmath153 . let @xmath157 and @xmath158 be the complexifications of @xmath159 and @xmath160 . let @xmath161 be the maximal compact subgroup of @xmath12 and @xmath162 its complexification . higgs bundles with holonomy in the real form @xmath12 are pairs @xmath127 , where @xmath66 is a holomorphic structure on a principal @xmath162-bundle @xmath163 and @xmath26 is a holomorphic section of @xmath164 . clearly such higgs bundles are fixed points of @xmath11 . in general , there are fixed points of @xmath11 and @xmath18 other than those we have just described . we address the problem of classifying all fixed points in section [ secrep ] in terms of special classes of representations . + we consider the action of the real structures on the connected components of @xmath93 . for this note that the cartan involution @xmath165 induces an involution @xmath166 . the action of the involutions @xmath10 on the space of connected components @xmath167 is given by @xmath168 where @xmath96 . let @xmath63 be a principal @xmath7-bundle on @xmath8 . we define @xmath169 to be the principal bundle diffeomorphic to @xmath63 but with @xmath7-action @xmath170 . at the level of isomorphism classes this gives the map @xmath166 . from the definition of @xmath11 it is clear that it sends a principal @xmath7-bundle @xmath63 to @xmath169 . it was shown in @xcite that the action of @xmath14 on principal @xmath7-bundles is @xmath171 . the result for @xmath18 follows since @xmath17 . when @xmath172 is the compact real form , we find a close relationship between the fixed point sets of @xmath14 and @xmath18 . [ propmultbyi ] let @xmath172 be the compact real form . then the fixed point sets of @xmath19 are diffeomorphic . let @xmath173 denote the action of @xmath174 sending @xmath175 to @xmath176 . then if @xmath172 we see that @xmath177 . it follows that @xmath178 exchanges the fixed point sets of the involutions @xmath19 . by restriction @xmath148 defines a complex structure on the fixed point set of @xmath14 . similarly @xmath5 defines a complex structure on the fixed point set of @xmath18 . moreover we have @xmath179 , so the isomorphism in proposition [ propmultbyi ] is in fact a complex analytic isomorphism . in section [ secrep ] , we see that the fixed point sets for @xmath19 correspond to quite different classes of representations . it is then unexpected to find a natural bijection between these spaces . + when @xmath144 is the split real form the involution @xmath11 admits a very simple description in terms of the hitchin fibration . for @xmath96 , we have that @xmath11 sends @xmath97 to @xmath180 . thus fixed points can only occur when @xmath181 . suppose now that @xmath181 and let @xmath182 be a non - singular fibre of @xmath138 corresponding to a point @xmath183 . from ( * ? ? ? * chapter 4 ) we have the following : [ proporder2 ] the action of @xmath11 preserves the hitchin fibration , @xmath184 . let @xmath96 with @xmath181 and let @xmath185 be a non - singular fibre of @xmath97 . the restriction @xmath186 of @xmath11 to @xmath185 has fixed points . let @xmath187 be a fixed point . then the action of @xmath188 on @xmath185 is given by @xmath189 with respect to the origin defined by @xmath23 . for the split real form the map @xmath190 preserves the invariant polynomials , hence @xmath184 . then for any @xmath96 with @xmath181 , we have that @xmath11 acts on the fibres of @xmath191 . let @xmath182 be a non - singular fibre of @xmath137 over a point @xmath183 . as the fibres are connected @xcite we have that @xmath185 is a complex torus and the restriction @xmath192 is a complex automorphism , since @xmath11 is holomorphic in @xmath147 . + recall that the hitchin fibration is lagrangian with respect to @xmath139 . let @xmath193 be a complex basis for @xmath194 and define vector fields @xmath195 on @xmath185 by requiring @xmath196 . these vector fields are commuting and integrate to an action of @xmath197 on @xmath185 by translation . the stabilizer of this action is a lattice @xmath198 giving an identification @xmath199 . using the fact that @xmath11 is an isometry which is anti - holomorphic in @xmath200 we have @xmath201 . on the other hand since @xmath184 we have @xmath202 . it follows that @xmath203 for each @xmath178 . this is enough to ensure that @xmath11 has a fixed point @xmath187 . using @xmath23 as an origin , we have by exponentiation that the action of @xmath11 on @xmath185 is given by @xmath189 . in particular , for each @xmath96 with @xmath181 , the intersection of the fixed points of @xmath11 on @xmath97 with the smooth fibres of the hitchin fibration @xmath204 is given by the elements of order two in those fibres . the involutions @xmath19 are well - behaved with respect to the hitchin fibration @xmath205 and give rise to real integrable systems . in @xcite we saw that there exists an anti - linear involution @xmath206 such that @xmath207 . the details of this proof carry over without difficulty to the case @xmath18 and we obtain a second anti - linear involution @xmath208 such that @xmath209 . let @xmath210 be the fixed point set of @xmath14 and @xmath211 the fixed point set of @xmath212 . the restriction of @xmath213 to @xmath214 gives a symplectic structure and @xmath215 is a lagrangian fibration @xcite . the non - singular fibres are a disjoint union of tori , though the number of components of the fibres generally varies as one moves around the base @xmath216 . the proof applies just as well to the case @xmath18 . let @xmath217 be the fixed point set of @xmath18 and @xmath218 the fixed point set of @xmath219 . the restriction of @xmath220 to @xmath217 gives a symplectic structure and @xmath221 is a lagrangian fibration . again , the non - singular fibres are unions of tori . the situation for @xmath11 is different . in this case the action of @xmath11 on @xmath93 covers the involution @xmath222 . note that @xmath223 is linear and thus its fixed point subspace need not be half dimensional . moreover it is possible that the fixed point subspace of @xmath223 could lie entirely within the singular locus of the hitchin fibration , so that the restriction of @xmath137 to the fixed point set of @xmath11 may be poorly behaved ( for example this is the case when @xmath224 , @xmath225 and @xmath226 , @xmath227 respectively , studied in @xcite ) . in the case of the involution @xmath18 , the set of fixed points in the moduli space of classical higgs bundles for which @xmath228 was thoroughly studied in @xcite . in this section we consider the case where @xmath20 with real structure @xmath229 and cartan involution @xmath230 . here a higgs bundle @xmath22 is a rank @xmath23 holomorphic vector bundle @xmath24 and holomorphic map @xmath68 . we will see that @xmath5-theory in various forms helps to distinguish between connected components of the fixed point sets of the real structures @xmath10 . fixed points of @xmath11 may be obtained from solutions to the hitchin equations with holonomy in the corresponding real form @xmath231 . for higgs bundles with holonomy in @xmath30 we may take @xmath24 to be a holomorphic @xmath232-bundle with reduction to the maximal compact @xmath233 and @xmath26 to be symmetric . such a higgs bundle @xmath22 is a fixed point of @xmath11 since the orthogonal structure gives an isomorphism between @xmath22 and @xmath234 . since the vector bundle @xmath24 has a real structure it defines a class @xmath34 \in ko^0(\sigma)$ ] in the @xmath3-theory of @xmath8 . we have an isomorphism @xmath235 given by @xmath34 \mapsto ( { \rm rank}(v ) , w_1(v ) , w_2(v ) ) $ ] , so the relevant topological invariants are the stiefel - whitney classes @xmath236 of @xmath24 . + when @xmath237 is even a second class of fixed points of @xmath11 are given by higgs bundles with holonomy in @xmath238 . for this @xmath24 is a holomorphic bundle with @xmath239-structure and @xmath240 using the symplectic transpose @xcite . the symplectic structure then gives an isomorphism @xmath241 so that such higgs bundles are fixed points of @xmath11 . the symplectic structure on @xmath24 defines a class @xmath34 \in ksp(\sigma)$ ] , the grothendieck group of bundles with symplectic structure on @xmath8 . however since @xmath242 is simply connected all symplectic bundles are trivial and @xmath243 , the only invariant being the rank of the bundle . we will see in section [ secrep ] that a fixed point of @xmath11 with simple holonomy must belong to one of these two classes . let @xmath22 be a fixed point of @xmath14 with simple holonomy . this requires @xmath24 to have degree zero , so the associated connection @xmath123 is flat . fixed points of @xmath14 correspond to connections @xmath29 such that @xmath244 , for @xmath245 an anti - holomorphic involution as introduced in section [ sec : real ] . equivalently there is a bundle isomorphism @xmath32 covering @xmath16 which respects @xmath29 . thus @xmath246 is covariantly constant , so assuming @xmath29 is simple we have that @xmath247 for some constant @xmath248 . choosing a square root @xmath249 and replacing @xmath250 by @xmath251 we obtain an involution @xmath32 lifting @xmath16 and preserving @xmath29 . by simplicity the lift @xmath250 is unique up to an overall sign . the @xmath37-action of @xmath250 gives @xmath24 the structure of a @xmath37-equivariant vector bundle and thus defines a class @xmath34 \in k^0_{\mathbb{z}_2}(\sigma)$ ] in equivariant @xmath5-theory . the class @xmath34 $ ] is determined from @xmath22 up to a sign ambiguity which corresponds to replacing @xmath250 by @xmath252 . + if @xmath16 acts without fixed points , the quotient @xmath253 is a compact non - orientable surface . equivariant bundles are obtained by pullback from @xmath254 and we have @xmath255 this defines a @xmath37-invariant with the property that it is non - zero for the unique non - trivial rank @xmath23 bundle on @xmath254 . when @xmath16 has @xmath256 fixed components we have , by the mayer - vietoris sequence @xmath257 over each fixed component @xmath258 of @xmath16 we may decompose @xmath259 into the @xmath155-eigenspaces @xmath260 of @xmath250 . taking the dimension @xmath261 gives a homomorphism @xmath262 and taking the sum of these homomorphisms over the fixed components of @xmath16 gives the isomorphism ( [ equkz2fix ] ) . the group of @xmath37-equivariant line bundles is given by @xmath263 . for each assignment of @xmath155 to each fixed component of @xmath16 , we can therefore find a corresponding equivariant line bundle unique up to isomorphism . this shows that exactly @xmath264 classes in @xmath265 are represented by rank @xmath23 equivariant bundles . + the equivariant @xmath5-theory classification is sharp in the following sense : let @xmath266 be equivariant vector bundles on @xmath8 . then @xmath266 are isomorphic as equivariant vector bundles if and only if @xmath34 = [ v ' ] \in k^0_{\mathbb{z}_2}(\sigma)$ ] . in the case that @xmath16 has no fixed points this is trivial . suppose now that @xmath16 has fixed points and let @xmath24 be an equivariant vector bundle . let @xmath32 be the involution on @xmath24 and let @xmath267 be the fixed circle components of @xmath16 . suppose that for @xmath268 we are given sections @xmath269 of @xmath270 such that @xmath271 , where @xmath272 . we claim that there exists a rank @xmath273 subbundle @xmath274 such that @xmath275 and the restriction of @xmath48 to @xmath276 is spanned by @xmath269 . let @xmath48 be the equivariant line bundle with eigenvalue @xmath277 over @xmath276 . considering @xmath278 it suffices to assume @xmath279 for all @xmath178 . this is enough to ensure the existence of a section @xmath280 of @xmath24 which restricts to @xmath269 on @xmath276 and such that @xmath281 . this proves our claim . choosing a @xmath250-invariant hermitian metric and taking the complement of @xmath48 in @xmath24 we have that @xmath24 is the sum of @xmath48 and a lower rank equivariant bundle . the proposition now follows by induction on rank . we give another description of equivariant @xmath5-theory , which though unconventional is well suited to spectral curves . let @xmath24 be a complex vector bundle with anti - linear isomorphism @xmath282 covering @xmath16 . we say @xmath283 is symmetric if @xmath284 for all @xmath285 . let @xmath137 be a hermitian metric on @xmath24 , taken anti - linear in the first factor . we can also view @xmath137 as an anti - linear isomorphism @xmath286 . we say @xmath137 is compatible with @xmath283 if @xmath287 . setting @xmath288 , we have that @xmath32 is a linear involution covering @xmath16 and @xmath289 becomes a class in @xmath290 . it is clear that this class does not depend on the choice of compatible hermitian metric @xmath137 . therefore we may identify @xmath290 with the grothendieck group of vector bundles @xmath24 equipped with a symmetric anti - linear isomorphism @xmath282 . note that over the fixed points of @xmath16 , the map @xmath283 defines an indefinite hermitian form @xmath291 . the signature of this form over each component corresponds to the dimensions of the @xmath155-eigenspaces of the restriction of @xmath250 . + consider the product @xmath292 $ ] with involution @xmath41 . the quotient @xmath293 is a 3-manifold with boundary @xmath294 . from @xmath45 we obtain a distinguished subspace of representations of @xmath126 , those representations which extend as flat connections from @xmath8 to @xmath45 . such representations are fixed points of @xmath14 @xcite . moreover , we can use equivariant @xmath5-theory to characterise which fixed points of @xmath14 arise in this manner . [ 3man ] let @xmath22 be a solution of the hitchin equations with a lift of @xmath16 to an involution @xmath32 preserving the associated flat connection @xmath29 . suppose also that @xmath29 has simple holonomy . then @xmath29 extends over the 3-manifold @xmath45 if and only if the class @xmath34 \in \tilde{k}_{\mathbb{z}_2}^0(\sigma)$ ] is trivial ( possibly on replacing @xmath250 by @xmath252 ) . let @xmath295 be the quotient which is a surface with boundary . there is a homotopy equivalence @xmath296 , so we can restate the problem in terms of finding a local system on @xmath254 such that the pullback to @xmath8 gives the local system of constant sections of @xmath29 . if such a local system exists we must have that @xmath250 or @xmath252 acts as the identity on each fixed component by simplicity . conversely suppose that @xmath250 acts as the identity on each fixed component . then the local system of constant sections of @xmath29 can be factored by @xmath250 giving the desired local system on @xmath254 . let @xmath22 be a fixed point of @xmath18 with simple holonomy . again it is easiest to describe fixed points in terms of the associated connection @xmath29 . in this case fixed points correspond to the condition that there is an anti - linear isomorphism @xmath32 covering @xmath24 and preserving @xmath29 . since @xmath246 is covariantly constant we have , assuming @xmath29 is simple , that @xmath247 for some non - zero constant @xmath297 . then since @xmath298 we find @xmath297 is real . after rescaling we may assume that @xmath297 is either @xmath273 or @xmath299 . since @xmath29 is simple it follows that one and only one of these two cases can occur . + when @xmath300 we have that @xmath16 lifts to an anti - linear involution @xmath32 preserving @xmath29 . this gives @xmath24 the structure of a real vector bundle in the sense of atiyah @xcite and a class @xmath34 \in kr^0(\sigma)$ ] . the map @xmath250 is determined only up to rescaling by @xmath301 , but this action is trivial on @xmath38 so that the class @xmath34 $ ] is independent of the choice of @xmath250 . let @xmath302 be the number of components of the fixed point set of @xmath16 . from the mayer - vietoris sequence we have an isomorphism @xcite @xmath303 for a bundle @xmath24 real with respect to @xmath16 , let @xmath89 be the degree . for each fixed circle component @xmath258 of @xmath16 the restriction @xmath259 is a real line bundle over the circle which has a first stiefel - whitney class @xmath304 . let @xmath305 be the first stiefel - whitney classes of @xmath24 restricted to the fixed components . then @xmath34 $ ] corresponds to @xmath306 . conversely every class in @xmath307 can be represented by a real bundle of rank @xmath23 for any @xmath23 . + when @xmath308 we have that @xmath16 lifts to an anti - linear isomorphism @xmath32 preserving @xmath29 and @xmath309 . this gives @xmath24 a quaternionic structure . let @xmath39 denote the grothendieck group of quaternionic vector bundles on @xmath8 . then , from @xcite there is an isomorphism @xmath310 . using this one finds @xmath311 when the anti - holomorphic involution @xmath16 on @xmath8 has no fixed points there exists a quaternionic line bundle @xmath48 of degree @xmath312 . tensoring by @xmath48 gives the isomorphism @xmath313 . when @xmath16 has fixed points any quaternionic bundle must have even rank @xmath314 and even degree @xmath315 @xcite . moreover every such pair @xmath316 occurs for if @xmath24 is a complex bundle of rank @xmath317 and degree @xmath89 , the bundle @xmath318 admits a quaternionic structure . in all cases we see that the @xmath319-class of a quaternionic bundle is classified by rank and degree . + from @xcite it can be deduced that the @xmath4-theory class of a real bundle @xmath320 on @xmath8 completely determines @xmath24 as a real bundle . the same is true for quaternionic bundles and @xmath319-theory . from @xcite , principal @xmath7-higgs bundles have associated some spectral data on which we can study the induced action of the involutions @xmath142 and @xmath18 . in this section we study these fixed point sets in terms of the corresponding spectral data . the fibres of the hitchin system are most easily seen using spectral curves , introduced in @xcite . given a classical higgs bundle @xmath22 of rank @xmath23 , the spectral curve @xmath69 is the set of points @xmath321 satisfying the characteristic equation @xmath322 where @xmath323 . from bertini s theorem @xmath69 is smooth for generic @xmath324 . for a smooth spectral curve @xmath69 the projection @xmath325 restricts to @xmath69 giving a branched @xmath23-fold cover @xmath46 and the canonical bundle @xmath326 of @xmath69 is given by @xmath327 . the eigenspaces of @xmath26 define a line bundle @xmath161 on @xmath69 . more precisely , @xmath328 is the cokernel of @xmath329 in @xmath330 , where @xmath331 is the tautological section of @xmath332 @xcite . the bundle @xmath24 is recovered as the direct image sheaf @xmath333 and we recover @xmath26 by pushing forward the endomorphism @xmath334 . conversely , given a generic point @xmath335 and a line bundle @xmath161 over the corresponding spectral curve @xmath69 , we obtain a higgs bundle by the above construction . this identifies the fibre of the hitchin map over @xmath336 with the picard variety @xmath337 of @xmath69 . let @xmath338 be a spin structure on @xmath8 . it is convenient to define a new line bundle @xmath48 such that @xmath339 . then if @xmath340 , we have by grothendieck - riemann - roch @xmath341 . the map sending a line bundle @xmath342 to the vector bundle @xmath343 with corresponding higgs field identifies the fibre of @xmath344 over @xmath336 with @xmath345 , the space of degree @xmath89 line bundles on @xmath69 . in particular the fibre of @xmath346 may be identified with the jacobian @xmath347 of @xmath69 . for this section we take @xmath46 to be an @xmath23-sheeted branched cover , not necessarily given as a spectral curve and let @xmath348 denote the corresponding canonical bundles . let @xmath349 be spin structures on @xmath8,@xmath69 . given a holomorphic vector bundle @xmath79 on @xmath69 , set @xmath350 and @xmath351 . note that @xmath352 and @xmath353 . + for @xmath3-theory we take @xmath79 to be a rank @xmath354 holomorphic bundle with complex orthogonal structure . choosing a compatible hermitian structure on @xmath79 we have a reduction to @xmath355 , so @xmath79 defines a class @xmath356 \in ko^0(s)$ ] . by relative duality we obtain an isomorphism @xmath357 , which is symmetric . thus @xmath24 has an orthogonal structure and defines a class @xmath34 \in ko^0(\sigma)$ ] . this construction defines a push - forward homomorphism @xmath358 in real @xmath5-theory . the choice of spin structures @xmath359 , @xmath360 are the @xmath3-orientations required to define the push - forward . + to consider the other @xmath5-theory groups suppose that @xmath16 lifts to an anti - holomorphic involution @xmath361 . we must choose spin structures compatibly . for this recall that to any compact riemann surface @xmath8 with anti - holomorphic involution @xmath16 , there exists a spin structure @xmath359 such that @xmath362 @xcite . in fact we have that @xmath16 lifts to an anti - linear map @xmath363 such that @xmath364 is the map @xmath365 . if @xmath16 has fixed points then @xmath366 , and if @xmath16 has no fixed points then @xmath367 if @xmath98 is odd and @xmath299 if @xmath98 is even @xcite . in other words , @xmath368 admits a real or quaternionic square root @xmath359 . similarly choose such a spin structure @xmath360 on @xmath69 with anti - linear isomorphism @xmath369 covering @xmath370 . if @xmath258 is a fixed component of @xmath16 then @xmath371 defines a real structure on the restriction of @xmath372 to @xmath373 . by squaring we obtain a real non - vanishing section of @xmath368 along the circle and hence an orientation of the circle . similarly we can use @xmath374 to orient circles in @xmath69 fixed by @xmath370 . + for equivariant @xmath5-theory we suppose that @xmath79 is a holomorphic vector bundle on @xmath69 with a symmetric anti - linear isomorphism @xmath375 covering @xmath370 . recall that this defines a class @xmath356 \in k_{\mathbb{z}_2}^0(s)$ ] . for the push - forward construction we require @xmath283 to be holomorphic in the sense that if @xmath376 is the germ of a holomorphic section of @xmath79 at @xmath377 , then @xmath378 is the germ of a holomorphic section of @xmath379 as @xmath380 . set @xmath350 . combining @xmath283 with @xmath374 , @xmath371 we obtain an isomorphism @xmath381 . setting @xmath340 we have that @xmath382 descends to a symmetric anti - linear isomorphism @xmath383 covering @xmath16 and hence defines an equivariant vector bundle on @xmath8 . sending @xmath79 to @xmath384 defines a push - forward homomorphism @xmath385 . when @xmath16 has fixed points the following theorem completely determines the push - forward in equivariant @xmath5-theory : [ prop : eqpush ] let @xmath258 be a fixed component of @xmath16 and @xmath386 the dimensions of the @xmath155-eigenspaces of the @xmath37-action on @xmath24 over @xmath373 . if @xmath370 has no fixed points then @xmath387 . if @xmath370 has fixed points then @xmath349 are real with respect to @xmath388 . let @xmath389 be the fixed circle components of @xmath370 lying over @xmath373 and @xmath390 the dimensions of the @xmath155-eigenspaces of the @xmath37-action on @xmath79 over @xmath276 . let @xmath391 be the degree of @xmath392 using the orientations induced by @xmath393 . then @xmath394 let @xmath395 be a point of @xmath373 which is not a branch point . consider the points in the inverse image @xmath396 . pairs of points in @xmath396 exchanged by @xmath370 will push down to give a @xmath37-action with @xmath155-eigenspaces of equal dimension , so will not contribute to @xmath397 . therefore we need only consider points in @xmath396 which are fixed by @xmath370 . let @xmath398 be the points of @xmath396 lying on @xmath276 . let @xmath399 be a real non - vanishing section of @xmath368 over @xmath373 compatible with the orientation of @xmath373 determined by @xmath371 and similarly let @xmath400 be a real non - vanishing section of @xmath326 over @xmath276 compatible with @xmath374 . let @xmath401 according to whether @xmath402 is a positive or negative multiple of @xmath400 , so @xmath403 . let @xmath404 be the @xmath37-action , @xmath405 a compatible hermitian metric and @xmath406 . then @xmath350 and we obtain a map @xmath407 . choose @xmath408 such that @xmath409 and @xmath410 with @xmath411 . now let @xmath412 and write @xmath413 , @xmath414 with @xmath415 . then , as @xmath416 is defined through relative duality , we have @xmath417 . if @xmath418 , this is @xmath419 . this contributes a sign of @xmath420 to @xmath397 . summing we obtain the result . for the cases of real and quaternionic @xmath5-theory suppose @xmath79 is a holomorphic bundle on @xmath69 with real or quaternionic structure @xmath404 . to define the push - forward we require @xmath250 to be holomorphic meaning the map @xmath421 send holomorphic germs to holomorphic germs . since @xmath422 carry real or quaternionic structures it follows that the vector bundle @xmath423 is similarly real or quaternionic and this structure descends to @xmath340 . if @xmath424 are both real or both quaternionic we obtain homomorphisms @xmath50 and @xmath425 . on the other hand if one of @xmath424 is real and the other quaternionic , we obtain push - forwards of the form @xmath426 and @xmath426 . the following theorem , which can be proved similarly to theorem [ prop : eqpush ] , completely determines the push - forward in @xmath4-theory : [ prop : krpush ] suppose @xmath79 has a real structure with respect to @xmath370 . let @xmath258 be a fixed component of @xmath16 and @xmath427 the first stiefel - whitney class of @xmath24 over @xmath373 . if @xmath370 has no fixed points then @xmath428 . if @xmath370 has fixed points then @xmath429 have real square roots . choose real square roots @xmath424 such that their restriction over any fixed component of @xmath430 are trivial as real bundles . let @xmath389 be the fixed circle components of @xmath370 lying over @xmath373 and @xmath431 the first stiefel - whitney class of @xmath79 over @xmath276 . then @xmath432 in the case where we start with a quaternionic bundle on @xmath69 the push - forward follows from grothendieck - riemann - roch . let @xmath22 be a higgs bundle with @xmath30-holonomy and @xmath48 the corresponding spectral line bundle , hence @xmath433 . by theorem [ proporder2 ] we have @xmath434 , so @xmath435 is trivial and @xmath48 is a real line bundle on @xmath69 . then @xmath48 defines a class @xmath436 \in ko^0(s)$ ] and we see that the @xmath3-theory class of @xmath24 is given by the push - forward @xmath34 = p_*[l]$ ] . this observation was recently made by hitchin @xcite and was used to relate the second stiefel - whitney class of @xmath24 to the mod @xmath13 index of @xmath48 . note that the fixed points of @xmath11 with holonomy in @xmath31 always have singular spectral curves @xcite , so on restricting to smooth fibres we do not see these fixed points . + next consider fixed points of the involution @xmath14 . let @xmath437 be the hitchin fibration , where for @xmath71 we have @xmath438 . let @xmath439 be the map sending a holomorphic differential @xmath427 to @xmath440 . we show in @xcite that @xmath441 . therefore the fixed points of @xmath14 lie over fixed points of @xmath16 . let @xmath46 be a non - singular spectral curve associated to a fixed point @xmath442 of @xmath439 . we extend @xmath16 to an involution @xmath443 on the total space of @xmath5 by setting @xmath444 for @xmath445 . then since @xmath69 corresponds to a fixed point @xmath442 , we find that @xmath370 restricts to an anti - holomorphic involution on @xmath69 covering @xmath16 . choosing a spin structure @xmath338 , we identify the fibre of the hitchin map over @xmath336 with the picard variety @xmath337 . assume that the chosen spin structure @xmath338 is preserved by @xmath16 , so we obtain a real or quaternionic structure @xmath446 . the action of @xmath14 on the fibre @xmath337 is then given by @xmath447 @xcite . choosing a hermitian structure on @xmath48 which is @xmath370-invariant , we have @xmath448 , so the involution @xmath370 lifts to @xmath48 defining a class @xmath436 \in k^0_{\mathbb{z}_2}(s)$ ] . the class of @xmath24 in @xmath290 is then given by the push - forward @xmath34 = p_*[l]$ ] . as discussed previously the class of @xmath34 $ ] in equivariant @xmath5-theory is only well - defined modulo replacing the involution @xmath32 with @xmath252 . this corresponds to the ambiguity in lifting @xmath370 to an involution on @xmath48 . using theorem [ prop : eqpush ] we can determine which classes in @xmath290 lie in the image of the push - forward . note that the push - forward depends on the topology of @xmath449 and the maps @xmath430 , but not the complex structures . the classes in the image of @xmath450 are the isomorphism classes of equivariant bundles which admit an equivariant flat connection with reductive holonomy and smooth spectral curve . the condition @xmath34 \in { \rm im}(p_*)$ ] may be interpreted as an analogue of the milnor - wood inequality for equivariant flat connections . + finally , consider fixed points of the involution @xmath18 . for a higgs bundle @xmath22 we have @xmath451 . it follows that @xmath441 , so that fixed points of @xmath18 must lie over fixed points of @xmath439 . let @xmath442 be a fixed point corresponding to a smooth spectral curve @xmath46 . as before we have an anti - holomorphic involution @xmath361 . choosing an @xmath16-invariant spin structure @xmath338 we have that the action of @xmath18 on the fibre @xmath347 is given by @xmath452 . fixed points correspond to line bundles with @xmath453 . thus @xmath48 carries either a real or quaternionic structure which pushes down to a real or quaternionic structure on @xmath454 . the class of @xmath24 in @xmath4- or @xmath319-theory is then the push - forward @xmath34 = p_*[l]$ ] . given a spin structure @xmath338 , recall that in @xcite hitchin constructed a section @xmath455 of the hitchin fibration which is invariant under @xmath11 . choosing @xmath338 with @xmath456 , we have that @xmath280 is fixed by all three involutions @xmath10 . this shows that the fixed point sets of @xmath10 are non - empty and contain smooth points . using theorem [ prop : krpush ] we may determine which classes in @xmath4-theory lie in the image of the push - forward . this gives a constraint on the topology of a real bundle @xmath34 $ ] to admit a real flat connection with reductive holonomy and smooth spectral curve . in this case the constraint for @xmath34 $ ] to lie in the image of @xmath450 is simple to state : for each fixed component @xmath258 of @xmath16 for which @xmath457 contains no fixed point of @xmath370 , we require @xmath259 to be trivial as a real bundle . in this section we give a representation theoretic description of the fixed point set of the involutions for any complex semisimple group @xmath7 . the case of reductive groups is similar but one must replace the fundamental group by a central extension . a representation @xmath458 is called reductive if the induced action of @xmath126 on @xmath62 by the adjoint representation decomposes into a direct sum of irreducible representations . let @xmath459 be the set of all reductive representations of @xmath126 into @xmath7 given the compact - open topology . the group @xmath7 acts on this space by conjugation and the quotient @xmath460 is hausdorff @xcite . as discussed in section [ sec : hyper ] there is a homeomorphism between the moduli space of polystable higgs bundles and the character variety @xmath125 of reductive representations . the correspondence sends a solution @xmath175 of the hitchin equations to the monodromy representation of the associated connection @xmath123 . we say that a reductive representation @xmath458 is simple if for all @xmath461 such that @xmath462 , we have @xmath463 . + consider a basepoint @xmath464 and write @xmath126 for @xmath465 . let @xmath466 be a path from @xmath467 to @xmath468 and define @xmath469 by @xmath470 . for @xmath471 \in \pi_1(\sigma)$ ] one has @xmath472 and @xmath473 . in terms of conjugacy classes of representations @xmath458 , the involutions @xmath10 take the form @xmath474 this description of the involutions makes the relation @xmath475 especially clear . + let @xmath476 be anti - involutions of @xmath7 . following @xcite , we say that @xmath477 are inner equivalent if @xmath478 for some @xmath479 . note that it is possible for distinct real forms to be inner equivalent . let @xmath480 be a simple fixed point of @xmath11 . then there exists an anti - involution @xmath481 inner equivalent to @xmath144 such that @xmath482 . conversely if @xmath481 is inner equivalent to @xmath144 and @xmath483 is a reductive representation with @xmath482 , then @xmath483 is a fixed point of @xmath11 . if @xmath483 is a fixed point of @xmath11 then @xmath484 for some @xmath461 . applying @xmath144 twice and using simplicity gives @xmath485 . let @xmath486 . then @xmath481 is an anti - holomorphic involution inner equivalent to @xmath144 . clearly we have @xmath482 . the proof of the converse is immediate . for example if we consider @xmath21 then aside from @xmath30 itself the only inner equivalent real form is @xmath31 which occurs when @xmath23 is even . properties of inner equivalent involutions in relation with higgs bundles appear in @xcite and are also currently being studied in @xcite . the fixed points of @xmath19 can be described in terms of the orbifold fundamental group @xmath487 of @xmath8 . this group is related to the usual fundamental group by a short exact sequence @xmath488 to describe this group we consider separately the cases where @xmath16 has fixed points or not . if @xmath16 acts freely then the quotient @xmath295 is a compact non - orientable surface . the orbifold fundamental group is simply the fundamental group of @xmath254 and ( [ equofg ] ) becomes the usual exact sequence for a double cover . when @xmath16 has fixed points , we take our basepoint @xmath467 to be a fixed point and @xmath466 to be the constant path . in this case @xmath489 is an involution and @xmath490 is the semi - direct product @xmath491 , where @xmath37 acts on @xmath126 by @xmath489 . in either case we may take the orbifold fundamental group to be @xmath492 with product given by @xmath493 , @xmath494 , @xmath495 , @xmath496 . + let @xmath497 be given by @xmath498 , and @xmath499 otherwise . then @xmath297 is a @xmath13-cocycle representing the non - trivial class in @xmath500 . let @xmath501 be the central extension of @xmath487 by @xmath502 corresponding to @xmath503 . observe that the extension class @xmath504 is trivial when restricted to @xmath126 so we obtain a short exact sequence @xmath505 explicitly @xmath501 is the group generated by @xmath506 together with an element @xmath507 modulo the relations @xmath508 , @xmath509 , @xmath510 , where @xmath511 and @xmath512 is a generator of @xmath502 . there is a bijection between conjugacy classes of representations @xmath513 and equivalence classes of pairs @xmath514 , where @xmath458 is a representation , and @xmath515 such that @xmath516 . two pairs @xmath517 are equivalent if @xmath518 and @xmath519 for some @xmath479 . given @xmath520 , we let @xmath483 be the restriction of @xmath283 to @xmath521 and take @xmath522 . this gives the desired bijection . the above establishes that fixed points of @xmath14 correspond to representations of @xmath501 . on the other hand there is an obstruction to representing a fixed point of @xmath14 as a representation of @xmath487 . suppose that @xmath483 is a simple point fixed by @xmath14 . then @xmath523 for some @xmath515 . since @xmath483 is simple any two such @xmath524 differ by an element of @xmath128 . applying @xmath489 twice and using simplicity we find @xmath525 . in order to extend @xmath483 to a representation of @xmath487 we need to find such a @xmath524 with @xmath526 . replacing @xmath524 by @xmath527 for @xmath528 we get @xmath529 . hence the obstruction to extending @xmath483 to an orbifold representation is a class in @xmath530 . + the case of fixed points of @xmath18 is similar . for this let @xmath531 be the semi - direct product @xmath532 , where @xmath37 acts on @xmath7 by @xmath144 . let @xmath533 be the projection . there is a bijection between conjugacy classes of representations @xmath534 such that @xmath535 and equivalence classes of pairs @xmath514 , where @xmath458 is a representation , and @xmath515 such that @xmath536 . two pairs @xmath517 are equivalent if @xmath518 and @xmath537 for some @xmath479 . it follows that every fixed point of @xmath18 extends to a representation of @xmath501 . as in the case of @xmath14 we can consider the problem of extending a fixed point of @xmath18 to a representation of @xmath487 . suppose @xmath483 is a simple fixed point of @xmath18 , so @xmath536 for some @xmath515 . applying the involutions @xmath144 and @xmath489 twice and using simplicity we obtain an element @xmath538 such that @xmath539 . replacing @xmath524 by @xmath527 for @xmath528 we have @xmath540 , so we obtain a class in @xmath541 . this class is the obstruction to lifting @xmath483 to a representation @xmath542 with @xmath543 . for example when @xmath20 and @xmath144 is conjugation we find @xmath544 . in this case the trivial class in @xmath545 corresponds to real bundles and the non - trivial class to quaternionic bundles . let @xmath56 be the langlands dual group of @xmath7 . there is a correspondence between invariant polynomials for @xmath7 and @xmath56 giving an identification @xmath546 . the moduli spaces @xmath547 are then torus fibrations over a common base and their non - singular fibres are dual abelian varieties @xcite . kapustin and witten give a physical interpretation of this in terms of s - duality , using it as the basis for their approach to the geometric langlands program @xcite . in this approach a crucial role is played by the various types of branes and their transformation under mirror symmetry . this duality exchanges branes according to @xmath58 , @xmath59 , @xmath60 . we consider here the question of how this duality acts on the fixed point sets of the real structures @xmath10 . we make some conjectures without attempting to give rigorous justifications . we let @xmath548 denote the fixed point set of @xmath549 for @xmath550 . + the simplest case is the fixed point set of @xmath14 , which is of type @xmath1 . since the definition of @xmath14 requires only the choice of anti - holomorphic involution @xmath16 , we have a corresponding involution @xmath551 on @xmath552 with fixed point set an @xmath1-brane @xmath553 . we conjecture that @xmath554 is the dual brane to @xmath555 . we give some evidence for this in @xcite . + consider next the @xmath2-brane @xmath556 . since the dual brane @xmath557 is also of type @xmath2 one might conjecture that it is the fixed point set of a corresponding involution @xmath558 . to define @xmath558 we need to choose a real structure @xmath559 on @xmath56 . if @xmath7 is a simple group not of type @xmath560 , the lie algebras of @xmath7 and @xmath56 coincide and we have a natural choice for @xmath559 . the @xmath560 cases however remain a mystery . + the most interesting case is the @xmath0-brane @xmath561 . the dual @xmath562 must be of type @xmath563 , a submanifold which is complex with respect to @xmath9 . one natural way of constructing @xmath563-branes in @xmath552 is to take a complex subgroup @xmath564 and to let @xmath565 be the space of @xmath56-higgs bundles with holonomy in @xmath566 . it remains to find a natural choice of subgroup @xmath566 . in @xcite a correspondence between real structures on @xmath7 and complex subgroups of the dual group @xmath56 is given . we conjecture that the correspondence given in @xcite determines the correct dual brane to @xmath567 . some evidence for this duality in the case @xmath568 has been shown by hitchin in @xcite . o. garca - prada , involutions of the moduli space of @xmath570-higgs bundles and real forms , in `` vector bundles and low codimensional subvarieties : state of the art and recent developments '' _ quaderni di matematica _ , editors : g. casnati , f. catanese and r. notari ( 2007 ) .

below we discuss the existence of a motherbody measure for the exterior inverse problem in potential theory in the complex plane . more exactly , we study the question of representability almost everywhere ( a.e . ) in of ( a branch of ) an irreducible algebraic function as the cauchy transform of a signed measure supported on a finite number of compact semi - analytic curves and a finite number of isolated points . firstly , we present a large class of algebraic functions for which there ( conjecturally ) always exists a positive measure with the above properties . this class was discovered in our earlier study of exactly solvable linear differential operators . secondly , we investigate in detail the representability problem in the case when the cauchy transform satisfies a quadratic equation with polynomial coefficients a.e . in . several conjectures and open problems are posed .

1406.1972

the study of local and global properties of the cauchy transform and the cauchy - stieltjes integral was initiated by a. cauchy and t. stieltjes in the middle of the 19th century . large numbers of papers and several books are partially or completely devoted to this area which is closely connected with the potential theory in the complex plane and , especially , to the inverse problem and to the inverse moment problem , see e.g. @xcite , @xcite , @xcite , @xcite , @xcite . during the last decades the notion of a motherbody of a solid domain or , more generally , of a positive borel measure was discussed both in geophysics and mathematics , see e.g. @xcite , @xcite , @xcite , @xcite . it was apparently pioneered in the 1960 s by a bulgarian geophysicist d. zidarov @xcite and later mathematically developed by b. gustafsson @xcite . although a number of interesting results about motherbodies was obtained in several special cases , @xcite , @xcite , @xcite there is still no consensus about its appropriate general definition . in particular , no general existence and/or uniqueness results are known at present . below we use one possible definition of a motherbody ( measure ) and study a natural exterior inverse motherbody problem in the complex plane . ( in what follows we will only use borel measures . ) _ main problem . _ given a germ @xmath1 of an algebraic ( or , more generally , analytic ) function near @xmath2 , is it possible to find a compactly supported signed measure whose cauchy transform coincides with ( a branch of ) the analytic continuation of @xmath3 a.e . in @xmath0 ? additionally , for which @xmath3 it is possible to find a positive measure with the above properties ? if such a signed measure exists and its support is a finite union of compact semi - analytic curves and isolated points we will call it _ a real motherbody measure _ of the germ @xmath3 , see definition [ def : main ] below . if the corresponding measure is positive then we call it _ a positive motherbody measure _ of @xmath3 . an obvious necessary condition for the existence of a positive motherbody measure is @xmath4 since @xmath5 is the total mass . a germ ( branch ) of an analytic function @xmath6 with @xmath4 will be called _ positive_. if @xmath7 then @xmath3 is called a _ probability _ branch . ( necessary and sufficient conditions for the existence of a probability branch of an algebraic function are given in lemma [ lm : probbr ] below . ) the formal definition of a motherbody measure that we are using is as follows . [ def : main ] given a germ @xmath8 of an analytic function near @xmath2 , we say that a signed measure @xmath9 is its _ motherbody _ if : \(i ) its support @xmath10 is the union of finitely many points and finitely many compact semi - analytic curves in @xmath0 ; ( ii ) the cauchy transform of @xmath9 coincides in each connected component of the complement @xmath11 with some branch of the analytic continuation of @xmath3 . _ remark . _ notice that by theorem 1 of @xcite if the cauchy transform of a positive measure coincides a.e . in @xmath0 with an algebraic function @xmath3 then the support of this measure is a finite union of semi - analytic curves and isolated points . therefore it is a motherbody measure according to the above definition . whether the latter result extends to signed measures is unknown at present which motivates the following question . _ problem 1 . _ is it true that if there exists a signed measure whose cauchy transform satisfies an irreducible algebraic equation a.e . in @xmath0 then there exists , in general , another signed measure whose cauchy transform satisfies a.e . in @xmath0 the same algebraic equation and whose support is a finite union of compact curves and isolated points ? does there exist such a measure with a singularity on each connected component of its support ? classically the inverse problem in potential theory deals with the question of how to restore a solid body or a ( positive ) measure from the information about its potential near infinity . the main efforts in this inverse problem since the foundational paper of p. s. novikov @xcite were concentrated around the question about the uniqueness of a ( solid ) body with a given exterior potential , see e.g. recent @xcite and @xcite and the references therein . p. s. novikov ( whose main mathematical contributions are in the areas of mathematical logic and group theory ) proved uniqueness for convex and star - shaped bodies with a common point of star - shapeness . the question about the uniqueness of contractible domains in @xmath0 with a given sequence of holomorphic moments was later posed anew by h. s. shapiro , see problem 1 , p. 193 of @xcite and answered negatively by m. sakai in @xcite . a similar non - uniqueness example for non - convex plane polygons was reported by geophysicists in @xcite , see also @xcite . observe that the knowledge of the cauchy transform of a compactly supported ( complex - valued ) measure @xmath12 near @xmath2 is equivalent to the knowledge of its ( holomorphic ) moment sequence @xmath13 where @xmath14 ( these moments are important conserved quantities in the theory of hele - shaw flow , @xcite . ) more precisely , if @xmath15 is the logarithmic potential of @xmath12 and @xmath16 is its cauchy transform then the taylor expansion of @xmath17 at @xmath2 has the form : @xmath18 observe also the important relations : @xmath19 which should be understood as equalities of distributions . it turns out that the existence of a compactly supported positive measure with a given cauchy transform @xmath3 near @xmath2 imposes no conditions on a germ except for the obvious @xmath4 , see theorem [ th : prop ] below . on the other hand , the requirement that the cauchy transform coincides with ( the analytic continuation ) of a given germ @xmath3 a.e . in @xmath0 often leads to additional restrictions on the germ @xmath3 which are not easy to describe in terms of the defining algebraic equation , see [ sq - roots ] . below we study two classes of algebraic functions of very different origin and our results for these two cases are very different as well . for the first class the obvious necessary condition @xmath7 seems to be sufficient for the existence of a positive motherbody measure . ( at present we can prove this fact only under certain additional restrictions . ) for the second class the set of admissible germs has a quite complicated structure . these results together with a number of conjectures seem to indicate that it is quite difficult , in general , to answer when a given algebraic germ @xmath3 admits a motherbody measure and if it does then how many . several concluding remarks are in place here . our interest in probability measures whose cauchy transforms are algebraic functions a.e . in @xmath0 was sparked by the pioneering work @xcite . since then the class of interesting examples where such situation occurs has been substantially broadened , see @xcite , @xcite , @xcite . some general local results when one considers a collection of locally analytic functions instead of a global algebraic function were obtained in @xcite and later extended in @xcite . * acknowledgements . * the authors are obliged to t. bergkvist , j. e. bjrk , late j. borcea , b. gustafsson , a. martnez - finkelshtein , and h. rullgrd for numerous discussions of the topic over the years . studies of quadrature domains by h. s. shapiro were in many ways an important source of inspiration for the present project . it is our pleasure to acknowledge this influence . the first rather simple result of the present paper ( which apparently is known to the specialists ) is as follows . [ th : prop ] given an arbitrary germ @xmath20 of an analytic function near @xmath2 there exist ( families of ) positive compactly supported in @xmath0 measures whose cauchy transform near @xmath2 coincides with @xmath3 . given a branch @xmath21 of an analytic function near @xmath2 we first take a germ of its logarithmic potential , i.e. a germ @xmath22 of harmonic function such that @xmath23 satisfying the relation @xmath24 in a punctured neighborhood of @xmath2 where @xmath25 stands for a germ of harmonic function near @xmath2 . for any sufficiently large positive @xmath26 the connected component @xmath27 of the level curve @xmath28 near infinity is closed and simple . it makes one turn around @xmath2 . to get a required positive measure whose cauchy transform coincides with @xmath3 near @xmath2 take @xmath27 for any @xmath26 large enough and consider the complement @xmath29 where @xmath30 is the interior of @xmath27 containing @xmath2 . consider the equilibrium measure of mass @xmath5 supported on @xmath31 . by frostman s theorem , this measure is in fact supported on @xmath27 ( since @xmath27 has no polar subsets ) , its potential is constant on @xmath27 and this potential is harmonic in the complement to the support . ( for definition and properties of equilibrium measures as well as frostman s theorem consult @xcite . ) thus it should coincide with @xmath22 in @xmath30 since the total mass is correctly chosen . then its cauchy transform coincides with @xmath3 in @xmath30 . if we choose @xmath32 as the branch at @xmath2 , then @xmath33 and @xmath27 is the circle @xmath34 . the equilibrium measure in this case is the uniform probability measure on this circle , and its cauchy transform @xmath35 will be @xmath36 inside the circle and @xmath37 outside . since @xmath36 is not the analytic continuation of @xmath37 , the uniform measure on the circle is not a motherbody measure . however the unit point mass @xmath12 at the origin , will have cauchy transform @xmath38 in @xmath39 and therefore will be a motherbody measure for the germ @xmath37 at @xmath2 . we now give a necessary condition and a slightly stronger sufficient condition for an algebraic curve given by the equation @xmath40 to have a probability branch at @xmath2 . here every @xmath41 and @xmath42 is an arbitrary finite subset of pairs of non - negative integers , i.e an arbitrary set of monomials in @xmath43 variables . in other words , @xmath42 is the set of all monomials appearing in @xmath44 with non - vanishing coefficients . ( in what follows @xmath45 stands for the variable corresponding to the cauchy transform . ) [ lm : probbr ] if the algebraic curve given by equation has a probability branch at @xmath2 then @xmath46 in particular , there should be at least two distinct monomials in @xmath42 whose difference of indices @xmath47 equals @xmath48 . if equation is satisfied , and additionally @xmath49 then there is a unique probability branch at @xmath2 satisfying equation . substituting @xmath50 in we get @xmath51 assuming that the algebraic curve given by @xmath52 has a branch @xmath53 where @xmath54 are undetermined coefficients , we substitute the latter expression in the above equation and get @xmath55 collecting the entries of the minimal degree which equals @xmath56 , we obtain an obvious necessary condition for the solvability of given by @xmath57 let us now show that , together with , are sufficient for the solvability of with respect to the sequence of coefficients @xmath58 . indeed , due to algebraicity of @xmath59 it suffices to prove the formal solvability of . let @xmath60 and rewrite the equation as @xmath61 which is now a polynomial in @xmath62 and @xmath63 . assume that @xmath64 satisfies @xmath65 the fact that this equation holds for @xmath66 is exactly the relation ( [ eq : firststep ] ) which gives the basis of the inductive construction of @xmath67 , as follows . letting @xmath68 be the partial derivative of @xmath69 with respect to the first variable , we have the following relation for the undetermined coefficient @xmath70 : @xmath71 since we have assumed that @xmath72 is non - zero , and that , by the induction assumption , @xmath73 , we can solve the latter equation for @xmath70 . thus , by induction , we have proven that there is a formal series solution of . therefore conditions and are sufficient for the existence of a probability branch which is also unique in this case . _ remark . _ note that for an irreducible algebraic curve defined by the second condition ( [ eq : branchpoint ] ) in the lemma says that @xmath74 is not its branching point . it is clearly possible , though cumbersome , to give necessary and sufficient conditions for the existence of a probability branch in terms of algebraic relations between the coefficients of the equation . an example of an equation that does not satisfy both conditions in the lemma , but still has a probability branch as a solution , is @xmath75 . in fig . 1 the necessary condition of lemma [ lm : probbr ] for the polynomials with newton polygons shown on each of three pictures is that the sum of the coefficients at the monomials labelled by the letters a , b , ... vanishes . since we are working with irreducible algebraic curves ( obtained as the analytic continuation of given branches at @xmath2 ) we will need the following statement . let @xmath76 be an arbitrary set of monomials in @xmath43 variables . denote by @xmath77 the linear span of these monomials with complex coefficients , and denote by @xmath78 the _ newton polygon _ of @xmath76 , i.e. its convex hull . [ lm : irred ] a generic polynomial in @xmath77 is irreducible if and only if : \(i ) @xmath78 is two - dimensional , i.e. not all monomials in @xmath76 lie on the same affine line ; \(ii ) @xmath76 contains a vertex on both coordinate axes , i.e. pure powers of both variables . notice that ( ii ) is already satisfied if @xmath76 contains the origin ( i.e. polynomials in @xmath77 might have a non - vanishing constant term ) . observe that the property that a generic polynomial from @xmath77 is irreducible is inherited , i.e. if @xmath76 contains a proper subset @xmath79 with the same property then it automatically holds for @xmath76 . the necessity of both conditions ( i ) and ( ii ) is obvious . if ( i ) is violated then any polynomial in @xmath77 can be represented as a polynomial in one variable after an appropriate change of variables . if ( ii ) is violated that any polynomial in @xmath77 is divisible by a ( power of a ) variable . to prove sufficiency we have to consider several cases . if @xmath76 contains @xmath80 and @xmath81 where both @xmath82 and @xmath83 are positive then already a generic curve of the form @xmath84 is irreducible unless @xmath80 , @xmath85 and @xmath86 lie on the same line . if @xmath76 contains @xmath80 and @xmath87 for some positive @xmath82 and no other pure powers of @xmath88 ( or , similarly , @xmath81 and @xmath87 for some positive @xmath89 and no other pure powers of @xmath45 ) then @xmath76 should contain another monomial @xmath90 with @xmath91 and @xmath89 positive . but then a generic curve of the form @xmath92 is irreducible . finally , if the only monomial on the axes is @xmath87 but there exists two monomials @xmath93 and @xmath94 with @xmath95 then a generic curve of the form @xmath96 is irreducible . [ cor : irredpos ] an irreducible polynomial @xmath59 having a probability branch has a non - negative @xmath97 . if @xmath98 then the set @xmath42 of monomials of @xmath59 must contain the origin , i.e. , a non - vanishing constant term . if we denote by @xmath99 the newton polygon of @xmath59 then the geometric interpretation of the latter corollary is that @xmath100 should ( non - strictly ) intersect the bisector of the first quadrant in the plane @xmath101 . the case @xmath98 will be of special interest to us , see [ minalgfun ] . we say that a polynomial @xmath59 ( and the algebraic function given by @xmath102 ) is _ balanced _ if it satisfies the condition @xmath98 . for polynomials with @xmath103 having a positive branch the problem of existence of a motherbody measure seems to be dramatically more complicated than for @xmath98 , see [ sq - roots ] . for @xmath103 a rather simple situation occurs when @xmath45 is a rational function , i.e. @xmath45 satisfies a linear equation . [ lm : rational ] a ( germ at @xmath2 of a ) rational function @xmath104 with coprime numerator and denominator admits a motherbody measure if and only if it has all simple poles with real residues . if all residues are positive the corresponding motherbody measure is positive . recall the classical relation @xmath105 between a measure @xmath12 and its cauchy transform @xmath106 where the derivative is taken in the sense of distributions . by assumption @xmath17 coincides almost everywhere with a given rational function . therefore , by the above formula , @xmath12 is the measure supported at the poles of the rational function that gives a weight to each pole coinciding up to a factor @xmath107 with the residue at that pole . since by assumption @xmath12 has to be a real measure this implies that all poles of the rational function should be simple and with real residues . the above statement implies that the set of rational function of degree @xmath108 admitting a motherbody measure has dimension equal to half of the dimension of the space of all rational functions of degree @xmath108 . the case when @xmath106 satisfies a quadratic equation is considered in some detail in [ sq - roots ] . in the case of balanced algebraic functions our main conjecture is as follows . [ conj : m=0 ] an arbitrary balanced irreducible polynom @xmath59 with a positive branch admits a positive motherbody measure . appropriate newton polygons are shown on the central and the right - most pictures in fig . 1 . the next result essentially proven in @xcite strongly supports the above conjecture . further supporting statements can be found in @xcite . [ th : main ] an arbitrary balanced irreducible polynomial @xmath59 with a unique probability branch which additionally satisfies the following requirements : \(i ) @xmath42 contains a diagonal monomial @xmath109 which is lexicographically bigger than any other monomial in @xmath42 ; \(ii ) the probability branch is the only positive branch of @xmath59 ; admits a probability motherbody measure . by lexicographically bigger we mean that the pair @xmath110 is coordinate - wise not smaller than any other pair of indices in @xmath42 as shown on the right - most picture in fig . 1 . condition ( i ) is the only essential restriction in theorem [ th : main ] compared to conjecture [ conj : m=0 ] since condition ( ii ) is generically satisfied . observe also that an irreducible balanced polynomial @xmath59 must necessarily have a non - vanishing constant term , see lemma [ lm : irred ] . balanced polynomials satisfying assumptions of theorem [ th : main ] are called _ excellent balanced _ polynomials . an interesting detail about the latter theorem is that its proof has hardly anything to do with potential theory . our proof of theorem [ th : main ] is based on certain properties of eigenpolynomials of the so - called exactly solvable differential operators , see e.g. @xcite . we will construct the required motherbody measure as the weak limit of a sequence of root - counting measures of these eigenpolynomials . a linear ordinary differential operator @xmath111 where each @xmath112 is a polynomial of degree at most @xmath82 and there exists at least one value @xmath113 such that @xmath114 is called _ exactly solvable_. an exactly solvable operator is called _ non - degenerate _ if @xmath115 . the symbol @xmath116 of the operator is , by definition , the bivariate polynomial @xmath117 observe that @xmath118 is exactly solvable if and only if @xmath116 is balanced . ( notice that @xmath116 , by definition , has no constant term . ) given an exactly solvable @xmath119 , consider the following _ homogenized spectral problem _ : @xmath120 where @xmath121 is called the _ homogenized spectral parameter_. given a positive integer @xmath122 we want to determine all values @xmath123 of the spectral parameter such that equation has a polynomial solution @xmath124 of degree @xmath108 . using notation @xmath125 one can easily check that the eigenvalues @xmath123 satisfy the equation : @xmath126 ( if @xmath118 is non - degenerate , i.e. @xmath127 , then there typically will be @xmath89 distinct values of @xmath123 for @xmath108 large enough . ) introducing the normalized eigenvalues @xmath128 we get that @xmath129 satisfy the equation : @xmath130 if @xmath131 exists then it should satisfy the relation : @xmath132 if @xmath118 is a degenerate exactly solvable operator , then let @xmath133 be the maximal @xmath82 such that @xmath134 . by definition , @xmath135 vanish for all @xmath136 . thus , the first @xmath137 terms in vanish as well implying that has @xmath133 non - vanishing and @xmath138 vanishing roots . [ lm : sim ] given an exactly solvable operator @xmath118 as above , * for the sequence of vanishing eigenvalues @xmath139 there exists a finite upper bound of the degree of a non - vanishing eigenpolynomial @xmath140 . * for any sequence @xmath141 of eigenvalues such that @xmath142 where @xmath143 is some non - vanishing simple root of the sequence of their eigenpolynomials is well - defined for all @xmath144 , i.e. for all sufficiently large @xmath108 . item ( i ) is obvious since reduces to @xmath145 when @xmath146 implying @xmath147 which is impossible for any polynomial of degree exceeding @xmath148 . to prove ( ii ) , notice that defines @xmath149 as an algebraic function of the ( complex ) variable @xmath108 , and that is then the equation for the value at @xmath150 . by assumption there is a unique branch with the value @xmath143 , and then the corresponding branch of the algebraic function will intersect other branches in at most a finite number of points . hence for @xmath108 very large we may identify @xmath151 as belonging to this branch . the next result is central in our consideration . [ th : loc][see theorem 2 , @xcite ] for any non - degenerate exactly solvable operator @xmath118 , such that has no double roots , there exists @xmath152 such that the roots of all eigenpolynomials of the homogenized problem whose degree exceeds @xmath152 are bounded , i.e. there exists a disk centered at the origin containing all of them at once . unfortunately the existing proof of the latter theorem is too long and technical to be reproduced in the present paper . the next local result is a keystone in the proof of theorem [ th : main ] , comp . proposition 3 of @xcite . [ th : ct ] given an exactly solvable @xmath153 consider a sequence @xmath154 of the eigenpolynomials of such that the sequence @xmath141 of their eigenvalues satisfies the condition @xmath142 where @xmath143 is some non - vanishing root of . let @xmath155 be the normalized logarithmic derivative of @xmath124 . if the sequence @xmath156 converges to a function @xmath157 in some open domain @xmath158 , and the derivatives of @xmath159 up to order @xmath89 are uniformly bounded in @xmath160 then @xmath157 satisfies in @xmath161 the symbol equation : @xmath162 note that each @xmath163 is well - defined and analytic in any open domain @xmath161 free from the zeros of @xmath164 . choosing such a domain @xmath161 and an appropriate branch of the logarithm such that @xmath165 is defined in @xmath161 , consider a primitive function @xmath166 which is also well - defined and analytic in @xmath161 . straightforward calculations give : @xmath167 @xmath168 , and @xmath169 . more generally , @xmath170 where the second term @xmath171 is a polynomial of degree @xmath172 in @xmath173 . the equation @xmath174 gives us : @xmath175 or , equivalently , @xmath176 letting @xmath177 and using the boundedness assumption for the first @xmath148 derivatives we get the required equation ( [ algfunc ] ) . take an excellent balanced irreducible polynomial @xmath178 having a probability branch at @xmath2 . since @xmath59 has a non - vanishing constant term , we can without loss of generality assume that it is equal to @xmath179 . consider its corresponding differential operator @xmath180 , that is the operator whose symbol @xmath116 equals @xmath181 . notice that since @xmath59 has a probability branch at @xmath2 then there exists a root of which is equal to @xmath87 . therefore , there exists a sequence @xmath141 of eigenvalues of @xmath182 satisfying the condition @xmath183 . to settle theorem [ th : main ] , notice that by theorem [ th : loc ] for any @xmath184 as above the roots of all its eigenpolynomials of all sufficiently large degrees are localized . choose a sequence @xmath141 of eigenvalues of @xmath182 satisfying the condition @xmath183 which exists for any excellent @xmath59 . consider the corresponding sequence @xmath140 of eigenpolynomials of @xmath182 , and the sequence @xmath185 of the root - counting measures of these eigenpolynomials , together with the sequence @xmath186 where @xmath187 are their cauchy transforms . ( observe that the cauchy transform of a polynomial @xmath44 of degree @xmath108 equals @xmath188 . ) by theorem [ th : loc ] all zeros of all these @xmath124 are contained in some fixed disk . therefore the supports of all @xmath189 are bounded implying that there exists a weakly converging subsequences @xmath190 of the sequence @xmath185 . chosing an appropriate further subsequence we can guarantee that the corresponding subsequence @xmath191 converges almost everywhere in @xmath0 . ( the rigorous argument for the latter claim is rather technical , see details in @xcite . ) taking a further subsequence , if necessary , we get that the subsequences @xmath192 will be bounded a.e . on any compact subset of @xmath0 . finally , notice that if the limit @xmath193 exists in some domain @xmath161 then also @xmath194 exists in @xmath161 and equals @xmath195 where @xmath196 and the sequence @xmath141 is chosen as above . since it satisfies the condition @xmath197 we have settled the existence of a probability measure whose cauchy transform coincides a.e . in @xmath0 with a branch of the algebraic function given by the equation @xmath198 which is the same as @xmath102 . in this section we discuss which quadratic equations of the form : @xmath199 with @xmath200 admit motherbody measure(s ) . for the subclass of with @xmath201 identically vanishing , such results were in large obtained in @xcite . very close statements were independently and simultaneously obtained in @xcite and @xcite . to go further , we need to recall some basic facts about quadratic differentials . a ( meromorphic ) quadratic differential @xmath202 on a compact orientable riemann surface @xmath203 without boundary is a ( meromorphic ) section of the tensor square @xmath204 of the holomorphic cotangent bundle @xmath205 . the zeros and the poles of @xmath202 constitute the set of _ singular points _ of @xmath202 denoted by @xmath206 . ( non - singular points of @xmath202 are usually called _ regular_. ) if @xmath202 is locally represented in two intersecting charts by @xmath207 and by @xmath208 resp . with a transition function @xmath209 , then @xmath210 any quadratic differential induces a canonical metric on its riemann surface @xmath203 , whose length element in local coordinates is given by @xmath211 the above canonical metric @xmath212 on @xmath203 is closely related to two distinguished line fields given by the condition that @xmath213 and @xmath214 resp . the integral curves of the first field are called _ horizontal trajectories _ of @xmath202 , while the integral curves of the second field are called _ vertical trajectories _ of @xmath202 . in what follows we will mostly use horizontal trajectories of quadratic differentials and reserve the term _ trajectories _ for the horizontal ones . here we only consider rational quadratic differentials , i.e. @xmath215 . any such quadratic differential @xmath202 will be globally given in @xmath0 by @xmath216 , where @xmath217 is a complex - valued rational function . trajectories of @xmath202 can be naturally parameterized by their arclength . in fact , in a neighborhood of a regular point @xmath218 on @xmath0 , one can introduce a local coordinate called a _ canonical parameter _ and given by @xmath219 one can easily check that @xmath220 implying that horizontal trajectories in the @xmath88-plane correspond to horizontal straight lines in the @xmath62-plane , i.e they are defined by the condition @xmath221 . a trajectory of a meromorphic quadratic differential @xmath202 given on a compact @xmath203 without boundary is called _ singular _ if there exists a singular point of @xmath202 belonging to its closure . for a given quadratic differential @xmath202 on a compact surface @xmath222 denote by @xmath223 the union of all its singular trajectories and singular points . in general , @xmath224 can be very complicated . in particular , it can be dense in some subdomains of @xmath203 . we say that a singular trajectory is _ double singular _ if a ) it approaches a singular point in both directions , i.e. if one moves from a given point on this trajectory in one direction or the opposite ; b ) each of these singular points is either a zero or a simple pole . we denote by @xmath225 ( the closure of ) the set of double singular trajectories of . ( one can easily show that @xmath226 is an imbedded ( multi)graph in @xmath203 . here by a _ multigraph _ on a surface we mean a graph with possibly multiple edges and loops . ) finally , denote by @xmath227 the subgraph of @xmath226 consisting of ( the closure of ) the set of double singular trajectories whose both ends are zeros of @xmath202 . a non - singular trajectory @xmath228 of a meromorphic @xmath202 is called _ closed _ if @xmath229 such that @xmath230 for all @xmath231 . the least such @xmath232 is called the _ period _ of @xmath233 . a quadratic differential @xmath202 on a compact riemann surface @xmath203 without boundary is called _ strebel _ if the set of its closed trajectories covers @xmath203 up to a set of lebesgue measure zero . in this subsection we find the relation of the question for which triples of polynomials @xmath234 equation admits a real motherbody measure to a certain problem about rational quadratic differentials . we start with the following necessary condition . [ pr : stand ] assume that equation admits a real motherbody measure @xmath12 . then the following two conditions hold : \(i ) any connected curve in the support of @xmath12 coincides with a horizontal trajectory of the quadratic differential @xmath235 \(ii ) the support of @xmath12 should include all branching points of . _ remark . _ observe that if @xmath236 and @xmath201 are coprime , the set of all branching points coincides with the set of all zeros of @xmath237 . in particular , in this case requirement proposition [ pr : stand ] implies that the set @xmath238 for the differential @xmath239 should contain all zeros of @xmath237 . the fact that every curve in @xmath240 should coincide with some horizontal trajectory of is well - known and follows from the plemelj - sokhotsky s formula , see e.g. @xcite . it is based on the local observation that if a real measure @xmath241 is supported on a smooth curve @xmath242 , then the tangent to @xmath243 at any point @xmath244 should be perpendicular to @xmath245 where @xmath246 and @xmath247 are the one - sided limits of @xmath45 when @xmath248 , see e.g. @xcite . ( here @xmath249 stands for the usual complex conjugation . ) solutions of are given by @xmath250 their difference being @xmath251 the tangent line to the support of the real motherbody measure @xmath12 satisfying at its arbitrary smooth point @xmath218 , being orthogonal to @xmath252 is exactly given by the condition @xmath253 . the latter condition defines the horizontal trajectory of @xmath239 at @xmath218 . finally the observation that @xmath254 should contain all branching points of follows immediately from the fact that @xmath106 is a well - defined function in @xmath255 . in many special cases statements similar to proposition [ pr : stand ] can be found in the literature , see e.g. recent @xcite and references therein . proposition [ pr : stand ] allows us , under mild nondegeneracy assumptions , to formulate necessary and sufficient conditions for the existence of a motherbody measure for which however are difficult to verify . namely , let @xmath256 with coordinates @xmath257 be the algebraic curve given by ( the projectivization of ) equation . @xmath258 has bidegree @xmath259 and is hyperelliptic . let @xmath260 be the projection of @xmath258 on the @xmath88-plane @xmath261 along the @xmath45-coordinate . from we observe that @xmath262 induces a branched double covering of @xmath261 by @xmath258 . if @xmath236 and @xmath201 are coprime and if @xmath263 , the set of all branching points of @xmath264 coincides with the set of all zeros of @xmath237 . ( if @xmath265 then @xmath2 is also a branching pont of @xmath262 of multiplicity @xmath266 . ) we need the following lemma . [ lm : poles ] if @xmath236 and @xmath201 are coprime , then at each pole of i.e. at each zero of @xmath236 , only one of two branches of @xmath258 goes to @xmath2 . additionally the residue of this branch at this zero equals that of @xmath267 . indeed if @xmath236 and @xmath201 are coprime , then no zero @xmath218 of @xmath236 can be a branching point of since @xmath268 . therefore only one of two branches of @xmath258 goes to @xmath2 at @xmath218 . more exactly , the branch @xmath269 attains a finite value at @xmath218 while the branch @xmath270 goes to @xmath2 where we use the agreement that @xmath271 . now consider the residue of the branch @xmath247 at @xmath218 . since residues depend continuously on the coefficients @xmath272 it suffices to consider only the case when @xmath218 is a simple zero of @xmath236 . further if @xmath218 is a simple zero of @xmath236 then @xmath273 by proposition [ pr : stand ] ( besides the obvious condition that has a real branch near @xmath2 with the asymptotics @xmath274 for some @xmath275 ) the necessary condition for to admit a motherbody measure is that the set @xmath238 for the differential contains all branching points of , i.e. all zeros of @xmath237 . consider @xmath276 . since @xmath238 contains all branching points of @xmath262 , @xmath277 consists of some number of open sheets each projecting diffeomorphically on its image in @xmath278 . ( the number of sheets in @xmath277 equals to twice the number of connected components in @xmath279 . ) observe that since we have chosen a real branch of at infinity with the asymptotics @xmath274 , we have a marked point @xmath280 over @xmath2 . if we additionally assume that @xmath281 then @xmath2 is not a branching point of @xmath262 and therefore @xmath282 . [ lm : cut ] if @xmath263 , then any choice of a spanning ( multi)subgraph @xmath283 with no isolated vertices induces the unique choice of the section @xmath284 of @xmath258 over @xmath285 which : \a ) contains @xmath286 ; b ) is discontinuous at any point of @xmath287 ; c ) is projected by @xmath262 diffeomorphically onto @xmath285 . here by a spanning subgraph we mean a subgraph containing all the vertices of the ambient graph . by a section of @xmath258 over @xmath285 we mean a choice of one of two possible values of @xmath258 at each point in @xmath285 . obvious . observe that the section @xmath284 might attain the value @xmath2 at some points , i.e. contain some poles of . denote the set of poles of @xmath284 by @xmath288 . now we can formulate our necessary and sufficient conditions . [ th : necsuf ] assume that the following conditions are valid : \(i ) equation has a real branch near @xmath2 with the asymptotic behavior @xmath274 for some @xmath275 , comp . lemma [ lm : probbr ] ; \(ii ) @xmath236 and @xmath201 are coprime , and the discriminant @xmath289 of equation has degree @xmath290 ; \(iii ) the set @xmath238 for quadratic differential @xmath239 given by contains all zeros of @xmath237 ; \(iv ) @xmath239 has no closed horizontal trajectories . then admits a real motherbody measure if and only if there exists a spanning ( multi)subgraph @xmath291 with no isolated vertices , such that all poles in @xmath292 are simple and all their residues are real , see notation above . indeed assume that @xmath293 satisfying ( ii ) admits a real motherbody measure @xmath12 . assumption ( i ) is obviously neccesary for the existence of a real motherbody measure and the necessity of assumption ( iii ) follows from proposition [ pr : stand ] if ( ii ) is satisfied . the support of @xmath12 consists of a finite number of curves and possibly a finite number of isolated points . since each curve in the support of @xmath12 is a trajectory of @xmath239 and @xmath239 has no closed trajectories then the whole support of @xmath12 consists of double singular trajectories of @xmath239 connecting its zeros , i.e. belongs to @xmath238 . moreover the support of @xmath12 should contain sufficently many double singular trajectories of @xmath239 such that they include all the branching points of . by ( ii ) these are exactly all zeros of @xmath237 . therefore the union of double singular trajectories of @xmath239 belonging to the support of @xmath12 is a spanning ( multi)graph of @xmath294 without isolated vertices . the isolated points in the support of @xmath12 are necessarily the poles of . observe that the cauchy transform of any ( complex - valued ) measure can only have simple poles ( as opposed to the cauchy transform of a more general distribution ) . since @xmath12 is real the residue of its cauchy transform at each pole must be real as well . therefore existence of a real motherbody under the assumptions ( i)(iv ) implies the existence of a spanning ( multi)graph @xmath287 with the above properties . the converse is also immediate . _ observe that if ( i ) is valid then assumptions ( ii ) and ( iv ) are generically satisfied . notice however that ( iv ) is violated in the special case when @xmath201 is absent considered in subsection [ strb ] . additionally , if ( iv ) is satisfied then the number of possible motherbody measures is finite . on the other hand , it is the assumption ( iii ) which imposes severe additional restrictions on admissible triples @xmath272 . at the moment the authors have no information about possible cardinalities of the sets @xmath288 introduced above . thus it is difficult to estimate the number of conditions required for to admit a motherbody measure . theorem [ th : necsuf ] however leads to the following sufficient condition for the existence of a real motherbody measure for . [ cor : suf ] if , additionally to assumptions ( i)(iii ) of theorem [ th : necsuf ] , one assumes that all roots of @xmath236 are simple and all residues of @xmath295 are real then admits a real motherbody measure . indeed if all roots of @xmath236 are simple and all residues of @xmath295 are real then all poles of are simple with real residues . in this case for any choice of a spanning ( multi)subgraph @xmath287 of @xmath238 , there exists a real motherbody measure whose support coincides with @xmath287 plus possibly some poles of . observe that if all roots of @xmath236 are simple and all residues of @xmath295 one can omit assumption ( iv ) . in case when @xmath239 has no closed trajectories then all possible real motherbody measures are in a bijective correspondence with all spanning ( multi)subgraphs of @xmath238 without isolated vertices . in the opposite case such measures are in a bijective correspondence with the unions of a spanning ( multi)subgraph of @xmath238 and an arbitrary ( possibly empty ) finite collection of closed trajectories . although we at present do not have rigorous results about the structure of the set of general equations admiting a real motherbody measure , we think that generalizing our previous methods and statements from @xcite , @xcite and @xcite , one would be able to settle the following conjecture . [ conj : positive ] fix any monic polynomial @xmath236 of degree @xmath296 and an arbitrary polynomial @xmath201 of degree at most @xmath297 . let @xmath298 denote the set of all polynomials @xmath299 of degree at most @xmath108 such that admits a probability measure , i.e a positive motherbody measure of mass @xmath87 . then @xmath298 is a real semi - analytic variety of ( real ) dimension @xmath108 . in this subsection we present in more detail the situation when the middle term in is vanishing , i.e. @xmath301 . in this case one can obtain complete information about the number of possible motherbody signed measures and also a criterion of the existence of a positive motherbody measure . this material is mainly borrowed from @xcite . the following statement can be easily derived from results of ch . 3 , @xcite . [ lm : quadrpoles ] [ lemma1 ] if a meromorphic quadratic differential @xmath202 is strebel , then it has no poles of order greater than 2 . if it has a pole of order 2 , then the coefficient of the leading term of @xmath202 at this pole is negative . in view of this lemma let us introduce the class @xmath302 of meromorphic quadratic differentials on a riemann surface @xmath203 satisfying the above restrictions , i.e. their poles are at most of order @xmath43 and at each such double pole the leading coefficient is negative . it is known that for a meromorphic strebel differential @xmath202 given on a compact riemann surface @xmath203 without boundary the set @xmath224 has several nice properties . in particular , it is a finite embedded multigraph on @xmath203 whose edges are double singular trajectories . in other words , for a strebel quadratic differential @xmath202 , one gets @xmath303 . our next result relates a strebel differential @xmath202 on @xmath261 with a double pole at @xmath2 to real - valued measures supported on @xmath224 . [ th : charac ] ( i ) given two coprime polynomials @xmath236 and @xmath299 of degrees @xmath296 and @xmath108 respectively where @xmath236 is monic and @xmath299 has a negative leading coefficient , the algebraic function given by the equation @xmath304 admits a motherbody measure @xmath305 if and only if the quadratic differential @xmath306 is strebel . \(ii ) such motherbody measures are , in general , non - unique . if we additionally require that the support of each such measure consists only of double singular trajectories , i.e. is a spanning subgraph of @xmath307 then for any @xmath202 as above there exists exactly @xmath308 real measures where @xmath309 is the total number of connected components in @xmath310 ( including the unbounded component , i.e. the one containing @xmath2 ) . these measures are in @xmath311-correspondence with @xmath308 possible choices of the branches of @xmath312 in the union of @xmath313 bounded components of @xmath310 . to prove theorem [ th : charac ] we need some information about compactly supported real measures and their cauchy transforms . ( 440,270)(0,0 ) ( 170,240)(1,0)60 ( 170,240 ) ( 230,240 ) ( 200,275)(250,270)(230,240 ) ( 200,275)(150,270)(170,240 ) ( 200,205)(250,210)(230,240 ) ( 200,205)(150,210)(170,240 ) ( 180,255 ) ( 220,255 ) ( 180,255)(200,265)(220,255 ) ( 180,225 ) ( 220,225 ) ( 180,225)(200,215)(220,225 ) ( 20,110)(1,0)60 ( 20,110 ) ( 80,110 ) ( 55,110)@xmath314 ( 50,150 ) ( 0,-1)6 ( 50,137 ) ( 0,-1)6 ( 50,123 ) ( 0,1)6 ( 50,97 ) ( 0,-1)6 ( 50,83 ) ( 0,1)6 ( 50,70 ) ( 0,1)6 ( 30,125 ) ( 70,125 ) ( 30,125)(50,135)(70,125 ) ( 55,128)@xmath315 ( 30,95 ) ( 70,95 ) ( 30,95)(50,85)(70,95 ) ( 55,90)@xmath315 ( 15,30)1st measure ( 120,110 ) ( 180,110 ) ( 150,145)(200,140)(180,110 ) ( 155,128)@xmath314 ( 150,145)(100,140)(120,110 ) ( 155,90)@xmath315 ( 150,152 ) ( 0,-1)6 ( 150,138 ) ( 0,1)6 ( 150,120 ) ( 0,-1)6 ( 150,97 ) ( 0,-1)6 ( 150,83 ) ( 0,1)6 ( 150,70 ) ( 0,1)6 ( 130,125 ) ( 170,125 ) ( 130,125)(150,135)(170,125 ) ( 130,95 ) ( 170,95 ) ( 130,95)(150,85)(170,95 ) ( 155,140)@xmath315 ( 115,30)2nd measure ( 220,110 ) ( 280,110 ) ( 250,75)(300,80)(280,110 ) ( 255,128)@xmath315 ( 250,75)(200,80)(220,110 ) ( 255,90)@xmath314 ( 250,150 ) ( 0,-1)6 ( 250,137 ) ( 0,-1)6 ( 250,123 ) ( 0,1)6 ( 250,98 ) ( 0,1)6 ( 250,84 ) ( 0,-1)6 ( 250,68 ) ( 0,1)6 ( 230,125 ) ( 270,125 ) ( 230,125)(250,135)(270,125 ) ( 230,95 ) ( 270,95 ) ( 230,95)(250,85)(270,95 ) ( 255,75)@xmath315 ( 215,30)3rd measure ( 320,110)(1,0)60 ( 320,110 ) ( 380,110 ) ( 350,145)(400,140)(380,110 ) ( 355,128)@xmath314 ( 350,145)(300,140)(320,110 ) ( 355,90)@xmath314 ( 350,75)(400,80)(380,110 ) ( 355,143)@xmath315 ( 350,75)(300,80)(320,110 ) ( 355,110)@xmath315 ( 350,152 ) ( 0,-1)6 ( 350,136 ) ( 0,1)6 ( 350,122 ) ( 0,-1)6 ( 350,98 ) ( 0,1)6 ( 350,84 ) ( 0,-1)6 ( 350,68 ) ( 0,1)6 ( 330,125 ) ( 370,125 ) ( 330,125)(350,135)(370,125 ) ( 355,75)@xmath315 ( 330,95 ) ( 370,95 ) ( 330,95)(350,85)(370,95 ) ( 315,30)4th measure ( 00,00)figure 2 . the set @xmath303 of a strebel differential @xmath202 and 4 related real measures . [ fignew ] [ comp . ii , @xcite ] [ lemma4 ] suppose @xmath316 and that @xmath317 as @xmath318 and let @xmath12 be a compactly supported measure in @xmath319 such that @xmath320 in the sense of distributions . then @xmath321 almost everywhere , where @xmath322 is the cauchy transform of @xmath12 . it is clear that @xmath323 is locally integrable , analytic off the closure of the support of @xmath12 and vanishes at infinity . consider @xmath324 and notice that @xmath325 is a locally integrable function vanishing at infinity and satisfying @xmath326 in the sense of distributions . we must show that @xmath327 almost everywhere . let @xmath328 be an approximate to the identity , i.e. @xmath329 , @xmath330 and @xmath331 , and consider the convolution @xmath332 it is well - known that @xmath333 and that @xmath334 in @xmath335 for any compact set @xmath336 . moreover @xmath337 this shows that @xmath338 is an entire function which vanishes at infinity , implying that @xmath339 . hence @xmath327 a.e . lemma [ lemma4 ] shows how given a strebel differential @xmath340 to construct real measures whose support lies in @xmath224 and whose cauchy transforms satisfy by specifying branches of @xmath312 in @xmath341 . [ proof of theorem [ th : charac ] ] to settle part ( i ) we first show that for a given strebel differential @xmath306 one can construct real measures supported on @xmath224 with the required properties . to do this choose arbitrarily a branch of @xmath312 in each bounded connected component of @xmath342 . ( in the unbounded connected component we have to choose the branch which behaves as @xmath343 at infinity with @xmath344 ; such a choice is possible by our assumptions on the leading coeffcients of @xmath236 and @xmath299 . ) define now @xmath345 in the sense of distributions . the distribution @xmath12 is evidently compactly supported on @xmath346 . by lemma [ lemma4 ] we get that @xmath347 satisfies a.e in @xmath0 . it remains to show that @xmath12 is a real measure . take a point @xmath218 in the support of @xmath12 which is a regular point of @xmath348 and take a small neighborhood @xmath349 of @xmath218 which does not contain roots of @xmath299 and @xmath236 . in this neighborhood we can choose a single branch @xmath350 of @xmath312 . notice that @xmath349 is divided into two sets by the support of @xmath12 since it by construction consists of singular trajectories of @xmath12 . denote these sets by @xmath351 and @xmath352 resp . choosing @xmath351 and @xmath352 appropriately we can represent @xmath323 as @xmath353 in @xmath349 up to the support of @xmath12 , where @xmath354 denotes the characteristic function of the set @xmath355 . by theorem 2.15 in @xcite , ch.2 we have @xmath356 for any test function @xmath357 with compact support in @xmath349 . notice that the last equality holds because @xmath357 is identically zero in a neighborhood of @xmath358 , so it is only on the common boundary of @xmath203 and @xmath359 that we get a contribution to the integral given in the last equality . this common boundary is the singular trajectory @xmath233 intersected with the neighborhood @xmath349 . the integral @xmath360 is real since the change of coordinate @xmath361 transforms the integral to the integral of @xmath357 over a part of the real line . this shows that @xmath12 is locally a real measure , which proves one implication in part ( i ) of the theorem . to prove the converse implication , i.e. that a compactly supported real measure @xmath12 whose cauchy transform satisfies everywhere except for a set of measure zero produces the strebel differential @xmath362 consider its logarithmic potential @xmath363 , i.e. @xmath364 the function @xmath363 is harmonic outside the support of @xmath12 , and subharmonic in the whole @xmath319 . the following relation mentioned in the introduction @xmath365 connects @xmath366 and @xmath323 . it implies that the set of level curves of @xmath363 coincides with the set of horizontal trajectories of the quadratic differential @xmath367 . indeed , the gradient of @xmath363 is given by the vector field with coordinates @xmath368 in @xmath0 . such a vector in @xmath0 coincides with the complex number @xmath369 . hence , the gradient of @xmath366 at @xmath88 equals to @xmath370 ( i.e. the complex conjugate of @xmath347 ) . but this is the same as saying that the vector @xmath371 is orthogonal to ( the tangent line to ) the level curve of @xmath363 at every point @xmath88 outside the support of @xmath12 . finally , notice that at each point @xmath88 one has @xmath372 which means that the horizontal trajectories of @xmath373 are the level curves of @xmath363 outside the support of @xmath12 . notice that @xmath363 behaves as @xmath374 near @xmath2 and is continuous except for possible second order poles where it has logarithmic singularities with a negative leading coefficient . this guarantees that almost all its level curves are closed and smooth implying that @xmath348 is strebel . let us settle part ( ii ) of theorem [ th : charac ] . notice that if a real measure whose cauchy transform satisfies a.e . is supported on the compact set @xmath224 ( which consists of finitely many singular trajectories and singular points ) all one needs to determine it uniquely is just to prescribe which of the two branches of @xmath312 one should choose as the cauchy transform of this measure in each bounded connected component of the complement @xmath375 . ( the choice of a branch in the unbounded component is already prescribed by the requirement that it should behave as @xmath376 near @xmath2 where @xmath344 . ) any such choice of branches in open domains leads to a real measure , see the above proof of part ( i ) . concerning possible positive measures we can formulate an exact criterion of the existence of a positive measure for a rational quadratic differential @xmath306 in terms of rather simple topological properties of @xmath224 . to do this we need a few definitions . notice that @xmath224 is a planar multigraph with the following properties . the vertices of @xmath224 are the finite singular points of @xmath202 ( i.e. excluding @xmath2 ) and its edges are singular trajectories connecting these finite singular points . each ( open ) connected component of @xmath375 is homeomorphic to an ( open ) annulus . @xmath224 might have isolated vertices which are the finite double poles of @xmath202 . vertices of @xmath224 having valency @xmath87 ( i.e. hanging vertices ) are exactly the simple poles of @xmath202 . vertices different from the isolated and hanging vertices are the zeros of @xmath202 . the number of edges adjacent to a given vertex minus @xmath43 equals the order of the zero of @xmath202 at this point . finally , the sum of the multiplicities of all poles ( including the one at @xmath2 ) minus the sum of the multiplicities of all zeros equals @xmath377 . by a _ simple cycle _ in a planar multigraph @xmath224 we mean any closed non - selfintersecting curve formed by the edges of @xmath224 . ( obviously , any simple cycle bounds an open domain homeomorphic to a disk which we call the _ interior of the cycle_. ) [ pr : crit ] a strebel differential @xmath362 admits a positive motherbody measure if and only if no edge of @xmath224 is attached to a simple cycle from inside . in other words , for any simple cycle in @xmath224 and any edge not in the cycle but adjacent to some vertex in the cycle , this edge does not belong to its interior . the support of the positive measure coincides with the forest obtained from @xmath224 after the removal of all its simple cycles . _ remark . _ notice that under the assumptions of proposition [ pr : crit ] all simple cycles of @xmath224 are pairwise non - intersecting and , therefore , their removal is well - defined in an unambiguous way . in particular , the compact on the right part of fig . 3 admits no positive measure since it contains an edge cutting a simple cycle ( the outer boundary ) in two smaller cycles . the left picture on fig . 4 has no such edges and , therefore , admits a positive measure whose support consists of the four horizontal edges of @xmath224 . ( 440,160)(0,0 ) ( 85,120 ) ( 80,120 ) ( 80,120)(1,0)10 ( 90,120 ) ( 125,120 ) ( 120,120 ) ( 120,120)(1,0)10 ( 130,120 ) ( 97,120 ) ( 97,120)(1,0)16 ( 113,120 ) ( 60,120)(75,195)(150,120 ) ( 60,120)(75,45)(150,120 ) ( 150,120 ) ( 150,120)(1,0)20 ( 170,120 ) ( 230,120)(1,0)60 ( 230,120 ) ( 290,120 ) ( 260,145)(310,175)(290,120 ) ( 260,145)(210,175)(230,120 ) ( 260,85)(310,90)(290,120 ) ( 260,85)(210,90)(230,120 ) ( 260,145 ) ( 260,145)(0,-1)12 ( 260,133 ) ( 280,130 ) ( 240,130 ) ( 240,130)(260,122)(280,130 ) ( 240,105 ) ( 280,105 ) ( 240,105)(260,95)(280,105 ) ( 00,40)figure 3 . two examples of @xmath224 admitting and not admitting a positive measure . [ proof of proposition [ pr : crit ] ] notice that given a finite measure supported on a finite union of curves with continuous density we get that its logarithmic potential will be a continuous function . as we have shown in the proof of part ( i ) of theorem [ th : charac ] , in our situation the one - sided limits of the gradient of the logarithmic potential are orthogonal to the tangent lines of the curves in the support . in other words , the logarithmic potentials of our real measures attain constant values on each connected component of the support . ( this phenomenon is characteristic for the so - called equilibrium measures of a given collection of curves , see @xcite . ) moreover , if the considered measure @xmath12 is positive ( resp . negative ) on a given curve in its support then its potential attains a local maximum ( resp . minimum ) on this curve . thus for a positive measure as above its potential has no local minima except at @xmath2 . as a direct corollary of the latter observation we get that the support of a positive measure as above can not contain cycles . indeed , let it contain a cycle . without loss of generality we can assume that this cycle is simple ( i.e. does not have self - intersections ) since every cycle consists of simple cycles . consider the interior of this simple cycle . on its boundary the potential is constant and is locally decreasing in the direction pointing inside this cycle . therefore , the potential must have a local minimum in the interior which is impossible . also notice that if a required positive measure exists then its potential should increase on each simple cycle in the direction of its interior . let us now show that the existence of a positive measure implies that no edge of @xmath224 is attached to a simple cycle from inside . indeed , assume that such an edge exists . the potential should be constant on each connected component of @xmath224 and , in particular , on the one containing the considered cycle and the extra edge attached to it . finally , it should increase in the direction of the interior of the cycle . but this immediately implies that the potential attains a local minimum on this edge which means that the ( density of the ) measure on this extra edge is negative , see fig . 2 where the arrows show the directions of the gradient . let us show the converse implication , i.e. that the absence of such edges implies the existence of a positive measure . notice that the assumptions of proposition [ pr : crit ] are equivalent to the fact that any connected component of the graph @xmath224 has the following property . no edge belonging to a component is located inside the interiors of the cycles belonging to this component ( if they exist ) . thus we can uniquely define the branch of @xmath312 in each connected component of @xmath375 so that on the boundary of each simple cycle in @xmath224 the gradient of the logarithmic potential points inside the cycle . ( recall that the gradient coincides with the complex conjugate of the chosen branch of @xmath312 and is orthogonal to the edges of @xmath224 at each point except the vertices . ) since no simple cycles belonging to the same connected component of @xmath224 lie within each other this choice is unique and well - defined . it leads to the positive measure supported on the complement to the union of all simple cycles of @xmath224 . let us finish the paper with the following important observation . [ prop : density ] for any monic @xmath236 of degree @xmath296 the set of all polynomials @xmath299 of degree @xmath108 and with leading coefficient @xmath179 such that the differential @xmath378 is strebel is dense in the space of all polynomials of degree @xmath108 with leading coefficient @xmath179 . in fact , this set is the countable union of real semi - analytic varieties of positive codimension . this circumstance illustrates the difficulty of the general problem to determine for which algebraic equations a real motherbody measure exists . * 1 . * the natural question about which algebraic functions of degree bigger than @xmath43 whose newton polygon intersects the diagonal in the @xmath379-plane nontrivially admit a real motherbody measure is hard to answer . some steps in this direction can be found in @xcite . this topic is apparently closely related to the ( non - existing ) notion of strebel differential of higher order which we hope to develop in the future . in any case , it is clear that no results similar to theorem [ th : main ] are possible and one needs to impose highly non - trivial additional conditions on such functions to ensure the existence of a probability measure . problem . given a finite set @xmath76 of monomials satisfying the assumptions of lemma [ lm : probbr ] and lemma [ lm : irred ] , consider the linear space @xmath77 of all polynomials @xmath59 whose newton polygon is contained in @xmath76 . what is the ( hausdorff ) dimension of the subset @xmath380 of polynomials admitting a motherbody measure ? bjrk , j. borcea , r.bgvad , _ subharmonic configurations and algebraic cauchy transforms of probability measures . _ notions of positivity and the geometry of polynomials trends in mathematics 2011 , pp 3962 . j. borcea , r. bgvad and b. shapiro , homogenized spectral pencils for exactly solvable operators : asymptotics of polynomial eigenfunctions , publ . rims , vol 45 ( 2009 ) 525568 . corrigendum : homogenized spectral pencils for exactly solvable operators : asymptotics of polynomial eigenfunctions " , publ . rims , vol 85 ( 2012 ) 229233 . l. hrmander , the analysis of linear partial differential operators . i. distribution theory and fourier analysis . reprint of the second ( 1990 ) edition . classics in mathematics . springer - verlag , berlin , 2003 . x+440 pp . e. b. saff , v. totik , logarithmic potentials with external fields . appendix b by thomas bloom . grundlehren der mathematischen wissenschaften [ fundamental principles of mathematical sciences ] , 316 . springer - verlag , berlin , ( 1997 ) , xvi+505 pp .

we present a generic condition for lorentzian manifolds to have a barrier that limits the reach of boundary - anchored extremal surfaces of arbitrary dimension . we show that any surface with nonpositive extrinsic curvature is a barrier , in the sense that extremal surfaces can not be continuously deformed past it . furthermore , the outermost barrier surface has nonnegative extrinsic curvature . under certain conditions , we show that the existence of trapped surfaces implies a barrier , and conversely . in the context of ads / cft , these barriers imply that it is impossible to reconstruct the entire bulk using extremal surfaces . we comment on the implications for the firewall controversy .

1312.3699

despite the successes of the ads / cft correspondence , a full understanding of the duality is still lacking . since the original conjecture @xcite , much of the research effort has been aimed at developing a precise dual dictionary connecting geometric quantities in the bulk to observables in the boundary field theory . however , this dictionary remains poorly understood , hampering our ability to reconstruct the bulk geometry from the field theory , compute field theory quantities from the classical bulk theory , and assess whether or not the correspondence is complete . this last effort has been placed under scrutiny by the recent controversy over the black hole interior ( see e.g.@xcite ) , and direct efforts to construct or demonstrate an inability to construct a full dual dictionary , may serve to address it . to resolve such questions , a good starting point is the identification of bulk probes that depend primarily on the bulk geometry and are dual by the ads / cft dictionary to known field theory observables . the existence of such bulk probes is not guaranteed ; however , spacelike extremal surfaces , which are covariantly - defined and depend exclusively on the bulk geometry , are dual to well - understood field theory observables . in fact , spacelike extremal surfaces constitute most of the probes used in ads / cft . the dual observables include correlators @xcite , entanglement entropy @xcite , and wilson loops @xcite . the length of spacelike geodesics with boundary endpoints allows one to compute , in the wkb approximation , the two - point correlator of a high conformal dimension field operator at the endpoints . the area of a codimension 2 spacelike extremal surface anchored on some region @xmath0 on the boundary of an asymptotically locally anti - de sitter ( alads ) spacetime is associated with the entanglement entropy of @xmath0 within the boundary field theory . a field theory observable which is dual to an extremal surface must in some way encode information about the bulk geometry at the location of the extremal surface . if there is a limit on how far into the bulk such surfaces can reach , this also limits our ability to reconstruct the geometry from the corresponding dual observables . the natural question that arises is how much of the bulk can be recovered from extremal surface probes . extremal surface probes and their maximal reach were studied in @xcite for the case of static and translationally or spherically symmetric asymptotically ads spacetimes , and in a large amount of literature ( e.g. @xcite and @xcite ) for eternal ads black holes . studies have also been done for extremal surfaces in time - dependent geometries ( e.g. @xcite ) . in this paper , we establish a general constraint on the reach of probe spacelike extremal surfaces in spacetimes which are not necessarily alads . we find that many generic geometries admit a certain kind of surface which acts as a barrier for probe surfaces anchored to one boundary . we furthermore show that trapped surfaces with nonpositive null extrinsic curvature give rise to extremal surface barriers , and that on spacetimes which admit a totally geodesic spatial slice the existence of barriers implies the existence of either a singularity or a marginally trapped surface . more explicitly , we consider a codimension 1 `` splitting surface '' @xmath1 which divides the spacetime into two regions . @xmath1 may be spacelike , timelike , or null . we take some class @xmath2 of spacelike extremal surfaces of codimension 1 or greater , which are anchored to some boundary @xmath3 on one side of @xmath1 . then , given two simple assumptions , we can show that no members of this class of extremal surfaces cross ( or even touch ) @xmath1 . in other words , the splitting surface acts as an extremal surface barrier . the two assumptions are as follows : 1 . all of the extremal surfaces in the class @xmath2 can be continuously deformed ( while remaining extremal ) to surfaces which lie on the exterior of @xmath1 . @xmath1 has nonpositive extrinsic curvature ( as measured by normal vectors pointing outward , towards the exterior of @xmath1 ) . in other words , @xmath1 can only bend outward relative to its tangent plane . given these assumptions , @xmath1 acts as a barrier to any of the extremal surfaces lying in the class @xmath2 . this generalizes some previously known result in riemannian manifolds , that surfaces with an inward - pointing mean curvature vector can sometimes act as local barriers to minimal surfaces , and also act as global barriers in certain circumstances ( see e.g. @xcite ) . our proof uses elements from the approach of @xcite and @xcite , and shows that spacelike extremal surfaces can not approach surfaces with negative extrinsic curvature . they can , however , approach arbitrarily close to totally geodesic surfaces . we do not need to make any assumptions about causality or an energy condition ; the restriction is purely geometrical in nature . the first criterion is often satisfied : it simply requires that there be no obstruction to deforming the extremal surface ( while maintaining extremality ) away from @xmath1 . ( e.g. geodesics in the btz black hole geometry @xcite ) . this is not an issue for our proof , since we do not require that the deformed extremal surface be minimal . ] in many cases , this criterion holds for all extremal surfaces due to the topology and geometry of the spacetime , so the class @xmath2 consists of all extremal surfaces . for example , this is true in vacuum ads spacetimes . more generally , if the extremal surfaces are anchored to just one connected component of the boundary @xmath3 , and the bulk spacetime is homotopically trivial , then all surfaces may be continuously deformed towards @xmath3 . however , it may be an additional constraint on the geometry that the surfaces can remain extremal while this happens . for example , if there is another region @xmath4 which acts as a barrier to extremal surfaces anchored anywhere on the boundary of the spacetime @xmath5 , then @xmath4 may be an obstruction to deforming extremal surfaces past @xmath1 . the second criterion may be satisfied by a totally geodesic surface , i.e. a surface with vanishing extrinsic curvature . can also be equivalently defined as a surface @xmath1 in @xmath6 such that every geodesic on @xmath1 ( with respect to the induced connection on @xmath1 ) is also a geodesic on @xmath7 ( with respect to the connection on @xmath7 . ] an example of such a surface is a non expanding black hole horizon ( which , given the area theorem @xcite , is a stationary black horizon for spacetimes obeying the null energy condition ) . however , totally geodesic surfaces do not appear in generic spacetimes . fortunately , the result also applies to surfaces with negative extrinsic curvature , which can appear in generic spacetimes . examples will be given in section [ examples ] . in fact , if there exists a surface @xmath1 whose extrinsic curvature is negative everywhere , we will show that it is not actually the tightest bound on extremal surfaces . there will be some other `` outermost '' barrier @xmath4 some distance away , which has at least partly nonnegative extrinsic curvature ( in section [ singularities ] , we will prove a theorem relating compact outermost barriers to trapped surfaces and singularities ) . the existence of such extremal surface barriers and their link to singularities is a curiosity which is of relevance to the firewall controversy . the fact that many bulk probes do not extend beyond the barrier may suggest that the firewall , if it exists , may be at or behind the outermost barrier . at the very least , it shows that field theory observables that are dual to extremal surfaces anchored at one boundary do not directly reveal any information about the interior . any information about this region in the bulk must come from probes that are not extremal surfaces . in some cases , the presence of a barrier might even indicate that the boundary field theory does not have sufficient information to reconstruct the geometry behind the barrier @xcite . in that case , additional factors of the hilbert space may be needed to describe that region . an interpretation involving a loss of determinism may at first seem implausible in the case where @xmath1 is spacelike : one could use the bulk equations of motion to evolve the information in @xmath1 forwards or backwards in time and thus reconstruct some or all of the interior of @xmath1 . however , the presence of a firewall at @xmath1 would result in a breakdown of the classical bulk equations of motion . in this case , our hope of reconstructing the interior of @xmath1 relies entirely on the ads / cft correspondence dictionary . this dictionary , however , is primarily composed of extremal surface probes , which we prove below can not probe past @xmath1 . this paper is structured as follows . in section [ theorem ] , we prove our sufficient condition for a spacetime to have a `` barrier surface '' past which extremal surfaces can not reach ; we further prove that trapped surfaces form extremal surface barriers . we will also show that the outermost barrier surface must have partly nonnegative extrinsic curvature . section [ examples ] provides examples of bulk spacetimes with extremal surface barriers . in section [ singularities ] , we prove that in a certain a class of geometries , barriers occur only in the presence of singularities or trapped surfaces . we first prove in two parts that extremal surfaces do not probe past surfaces with a nonpositive extrinsic curvature tensor , and further , that barriers with everywhere negative extrinsic curvature are not the outermost barrier surfaces . @xmath6 will be a lorentzian manifold with at least one boundary @xmath3 for the rest of this paper . we start with two definitions . define a codimension 1 surface @xmath1 in @xmath7 to be a _ splitting surface _ if it separates @xmath7 into two open regions , ext@xmath8 and int@xmath8 , where we define the interior and exterior relative to the normal which defines the extrinsic curvature , and @xmath9 . this normal is taken to point towards ext@xmath8 . let @xmath1 now be a splitting surface . let @xmath10 be a family of spacelike extremal surfaces in @xmath7 of codimension @xmath11 such that all the @xmath12 can all be continuously deformed from some initial surface @xmath13 anchored at ext@xmath14 , and all the @xmath12 are anchored at ext@xmath15 ( see , e.g. fig . [ def ] ) . , so that the extrinsic curvature @xmath16 is well - defined . presumably this assumption can be weakened if care is taken in dealing with distributional extrinsic curvatures . ] then any surface in the family @xmath12 shall be called _ @xmath1-deformable_. * theorem 2.1 : * let @xmath1 be a codimension 1 splitting surface in @xmath7 such that for any vector field @xmath18 on @xmath1 , @xmath19 , where @xmath20 is the extrinsic curvature of @xmath1 . , the induced metric on @xmath1 , the extrinsic curvature on @xmath1 is given by @xmath21 . for a surface @xmath22 of codimension @xmath23 , the extrinsic curvature carries an additional index , and is defined in terms of @xmath24 , the induced metric on @xmath22 : @xmath25 . ] any @xmath1-deformable spacelike extremal surface which is anchored within @xmath26 remains in ext@xmath8 . moreover , no such spacelike extremal surface ever touches @xmath1 ( or even comes arbitrarily close to touching @xmath1 ) . , we could prove , using precisely the same formalism used in the proof above , that surfaces anchored in int@xmath8 remain in int@xmath8 . ] the proof closely follows that of lemma b in @xcite . let @xmath12 be the family of surfaces all deformable from @xmath27 as defined above . if all surfaces @xmath12 are in ext@xmath8 , we are done , since that implies that surfaces which are anchored at @xmath28 can never be deformed past @xmath1 . suppose there exists a surface @xmath29 in @xmath10 such that @xmath30 . then , since @xmath29 is linked to @xmath27 via a series of smooth deformations , there exists a `` midway '' surface @xmath22 in the family of surfaces which coincides with @xmath1 at a set of points @xmath31 and is tangent to @xmath1 at those points . [ def ] illustrates the deformation family . we focus on one coincident point @xmath32 of @xmath1 and @xmath22 . as explained in @xcite , @xmath33 is a measure of how much @xmath1 curves away from the tangent plane normal to @xmath34 with motion away from @xmath32 in the @xmath18 direction , where @xmath34 is the normal to @xmath1 when @xmath1 is not null , and we take @xmath35 to be the null generator for null @xmath1 . similarly , @xmath36 measures how much @xmath22 curves away from the plane normal to @xmath1 at @xmath32 as one moves in the @xmath18 direction away from @xmath32 ( restricted to motion tangent to @xmath22 ) . @xmath22 is outside of @xmath1 , so @xmath1 curves away from its tangent plane at least as much as @xmath22 does : @xmath37 in particular : @xmath38 where @xmath24 is the induced metric on @xmath22 . the equality in eq . [ firstcond ] follows from the fact that @xmath22 is extremal . @xmath22 is spacelike , so @xmath24 is riemannian and has no temporal components , which implies that @xmath39 , since @xmath40 . we therefore have a contradiction with eq . [ negcond ] , so the surfaces in the family @xmath10 never reach @xmath1 , which in turn implies that @xmath29 does not exist . moreover , it is impossible for @xmath22 to approach arbitrarily close to @xmath1 , because in the limit eq . [ negcond ] would still hold , and we would again arrive at a contradiction . may allow us to prove the existence of a barrier with a weaker condition than in theorem 2.1 . for example , to prove that a spacelike codimension 1 surface is a barrier for @xmath41-dimensional extremal surfaces , it is sufficient if the sum of the largest @xmath41 eigenvalues of @xmath20 are negative . however , the analysis is more complicated for timelike or null barriers . ] * theorem 2.2 : * let @xmath1 be a splitting surface as above , but we now take it to be totally geodesic , i.e. @xmath43 . any @xmath1-deformable spacelike extremal surface which is anchored in the intersection of the boundary @xmath3 with one component of @xmath7 ( as divided by @xmath1 ) remains in that component has some nontrivial topology or geometrical defect that creates an obstruction for @xmath1-deformability . if we can find another spacetime @xmath44 which agrees with @xmath7 everywhere inside @xmath1 and also obeys the following conditions : ( a ) the extremal surface @xmath22 is the same in @xmath7 and @xmath44 , ( b ) the splitting surface @xmath1 is the same in @xmath7 and @xmath44 , and ( c ) @xmath22 is @xmath1-deformable in @xmath44 , then @xmath1 is still a barrier in @xmath7 , by the proof of theorem 2.2 . ] . note that in the case where @xmath1 is totally geodesic , it provides a barrier both for surfaces in int@xmath8 and ext@xmath8 . the proof is identical for @xmath45 and @xmath46 . for concreteness we assume @xmath47 . we take @xmath22 to be the midway surface as above , and we show that if @xmath22 agrees with @xmath1 at @xmath32 and is tangent to it , then @xmath48 . this would imply that @xmath22 is no longer anchored at @xmath28 , and therefore it is impossible to deform a surface to reach and cross @xmath1 while maintaining boundary conditions in ext@xmath8 . the saturated equation is : @xmath49 this shows that the two surfaces curve away in the normal direction equally , and is an indication that the two surfaces must agree everywhere on @xmath22 if they coincide at a point . we now show this in more detail . suppose @xmath1 is null , and let @xmath34 be the null generator of @xmath1 directed towards @xmath22 . the corresponding covector is defined : @xmath50 , as illustrated in fig . consider a small neighborhood @xmath51 of @xmath32 of length scale @xmath52 . let @xmath53 and let @xmath54 . define the coordinate @xmath55 : @xmath56 where @xmath57 is obtained by parallel transporting @xmath58 along @xmath1 , and @xmath59 is a vector pointing from @xmath60 to @xmath32 in the small neighborhood around @xmath32 . note that because @xmath1 is totally geodesic , @xmath61 is null . within @xmath51 , we can always find an @xmath62 such that @xmath63 . for such a choice of @xmath60 , we therefore obtain : @xmath64 where @xmath65 is obtained by parallel transporting @xmath66 along @xmath22 . the null extrinsic curvatures therefore obey : @xmath67 let @xmath68 , defined on @xmath1 , be the proper distance from @xmath32 . we can use @xmath68 as a coordinate labeling points on @xmath1 . let @xmath69 be the volume element of a constant @xmath68 slice . define @xmath70 to be a green s function on the ball of points @xmath71 with @xmath72 , where @xmath73 where @xmath74 is defined on @xmath1 . for sufficiently small @xmath75 , the metric on @xmath1 is approximately a flat euclidean - signature metric , so @xmath76 or @xmath77 in @xmath78 . since @xmath79 ( regardless of what @xmath80 is ) , we find @xmath81 . when @xmath75 is small , these inequalities must continue to hold if the metric is slightly deformed by nonzero curvature . integrating over a ball @xmath82 of radius @xmath83 : @xmath84 since @xmath85 is never negative ( that would contradict the assumption that @xmath22 does not cross @xmath1 ) , we conclude that @xmath86 everywhere on a neighborhood of @xmath32 . but this implies that @xmath86 everywhere , since it solves an elliptic equation , and thus @xmath48 . this completes the proof for null @xmath1 . this reasoning applies directly to the cases where @xmath1 is timelike or spacelike by simply substituting the @xmath22-pointing spacelike or timelike normal of @xmath1 , respectively , for the null generator in the @xmath22 direction . this proof allows us to generalize a previously - known result for totally geodesic surfaces : + * corollary 2.3 : * let @xmath1 be a totally geodesic codimension @xmath87 surface in @xmath7 . let @xmath22 be a codimension @xmath88 spacelike extremal surface satisfying the following conditions : 1 . @xmath22 coincides with @xmath1 at some point(s ) @xmath32 , 2 . @xmath22 does not cross @xmath1 anywhere ( and therefore @xmath22 is tangent to @xmath1 at @xmath32 ) , then @xmath48 . + let @xmath89 be the outwards - pointing normals to @xmath1 as above ( null generators in the @xmath22 direction , if @xmath1 is null ) , where @xmath90 runs from @xmath91 . let @xmath92 . let @xmath51 be a neighborhood of @xmath32 . we define @xmath55 for @xmath53 as above : @xmath93 where @xmath54 is chosen so that @xmath94 . then : @xmath95 where for the case of null @xmath1 , @xmath96 , and for timelike and spacelike cases , we just require them to be equal . as before , we obtain @xmath97 by parallel transporting along @xmath1 and @xmath98 by parallel transporting along @xmath22 . the remaining reasoning in the proof of theorem 2.2 above yields @xmath99 , and we are done . note that theorems 2.1 and 2.2 , and corollary 2.3 together also give a complete description of the behavior of extremal surfaces in the presence of barriers , provided that all disconnected components are @xmath1-deformable . corollary 2.3 also leads to a natural corollary of theorems 2.1 and 2.2 , which combines the assumptions made in those theorems : * corollary 2.4 : * let @xmath1 be a splitting surface in @xmath7 such that @xmath100 for some vector fields @xmath101 on @xmath1 and @xmath102 for all other vector fields @xmath103 on @xmath1 . then any @xmath1-deformable spacelike extremal surface which is anchored in ext@xmath8 remains in ext@xmath8 . moreover , although such extremal surfaces might conceivably come arbitrarily close to @xmath1 if they only propagate along the @xmath104 directions , they can not approach @xmath1 if they propagate along the @xmath105 directions . as above , let @xmath10 be a family of @xmath1-deformable surfaces , all deformable from @xmath13 , and let @xmath22 be the midway surface between @xmath29 and @xmath27 . we again focus on a coincident point @xmath106 . [ negcond ] , we have @xmath107 , where @xmath24 is the induced metric on @xmath22 . we can decompose the metric into the components that lie along the @xmath108 directions and the @xmath109 directions : @xmath110 where @xmath111 and @xmath112 for some constants @xmath113 and @xmath114 . . [ negcond ] can be decomposed as follows : @xmath115 if @xmath22 has no components in the @xmath105 directions , then @xmath116 , and eq . [ decomposition ] reduces to @xmath117 . this is simply the case of theorem 2.2 , so @xmath22 could potentially get arbitrarily close to @xmath1 while remaining on ext@xmath8 . if @xmath22 has at least one component in @xmath105 directions , then eq . [ decomposition ] yields @xmath118 . this is simply the case of theorem 2.1 , so @xmath22 can not approach @xmath1 , and must always remain in ext@xmath8 . we conclude that if @xmath22 propagates in the directions in which @xmath1 has vanishing extrinsic curvature , @xmath22 can approach @xmath1 , but if @xmath22 propagates in the directions in which @xmath1 has negative extrinsic curvature , @xmath22 can not approach @xmath1 . one consequence of corollary 2.4 is that any null splitting surface @xmath1 which is foliated by surfaces where _ all _ components of the null extrinsic curvature are nonpositive is an extremal surface barrier . but if we are only interested in whether or not @xmath1 is a barrier to codimension 2 extremal surfaces , it turns out the we can do better : it is sufficient if the expansion @xmath119 , i.e. the _ trace _ of the null extrinsic curvature is nonpositive . this is because when a codimension 2 extremal surface @xmath22 touches ( but does not cross ) @xmath1 , we can take the trace of eq . [ 2.1 ] over all @xmath120 spacelike directions to obtain : @xmath121[thetas ] as in the case of corollary 2.3 , the above equation can only be saturated if @xmath22 lies on @xmath1 . this means that @xmath1 is a barrier to @xmath1-deformable spacelike extremal codimension 2 surfaces . if we assume the null curvature condition @xmath122 , we can additionally show how to use a codimension 2 extremal surface to construct a barrier to other codimension 2 surfaces . * theorem 2.5 : * let @xmath123 be a codimension 2 spacelike extremal surface . let @xmath1 be the union of null congruences shot outwards from @xmath123 towards both the future and the past directions . assuming the null curvature condition , @xmath1 is a barrier to @xmath1-deformable codimension 2 surfaces anchored to ext(@xmath1 ) . by the standard construction , the raychaudhuri equation implies that the two null congruences shot out from @xmath123 converge , so that @xmath124 , where @xmath125 is defined as the expansion moving outwards away from @xmath123 . as in the proof of theorems 2.1 and 2.2 , we assume for contradiction that @xmath22 is a midway surface touching @xmath1 but not crossing it . ( it does not matter which of the two null congruences @xmath22 touches first , but whichever one it touches first , we are only interested in the @xmath125 of that congruence . ) at the point of coincidence , eq . [ thetas ] implies that @xmath126 and @xmath127 both vanish . therefore , by the same reasoning as in corollary 2.3 , @xmath22 must lie entirely on @xmath1 . but then it can not be anchored at ext(@xmath1 ) . even without assuming the null curvature condition , the following corollary follows immediately : * corollary 2.6 : * if @xmath1 is a null surface foliated by ( marginally ) outer trapped surfaces ( i.e. @xmath119 ) , then @xmath1 is a barrier to @xmath1-deformable codimension 2 surfaces anchored to ext(@xmath1 ) . thus , if we are only interested in codimension 2 extremal surfaces ( e.g. for purposes of holographic entanglement entropy @xcite ) , we can prove the existence of barriers in more circumstances . we proved that the presence of the surface @xmath1 with nonpositive spatial extrinsic curvature is a sufficient condition for an extremal surface barrier . this raises the question of whether this barrier is the closest one to the extremal surfaces . consider some region on the boundary of the spacetime , @xmath129 . let @xmath130 denote the spacelike extremal surfaces anchored at @xmath0 . we define the outermost barrier @xmath1 for surfaces anchored at @xmath0 to be given by : @xmath131 ( for some ( non - alads ) spacetimes , @xmath0 may be a proper subset of ext@xmath8 , as shown in fig . [ outermost ] . ) the above definition is partly motivated by @xcite , although the case under discussion there involved only minimal area surfaces , whereas we consider all extremal surfaces . note that we can similarly define the outermost barrier for a subset @xmath132 of all spacelike extremal surfaces anchored at @xmath0 by defining @xmath133 . we now prove some properties of the outermost barrier . + * theorem 2.7 : * let @xmath1 be an outermost barrier in @xmath7 , and let @xmath10 be a family of extremal surfaces that can all be smoothly deformed from some initial surface @xmath13 such that some surface @xmath134 is arbitrarily close to @xmath1 . then @xmath107 , where @xmath135 is the induced metric on @xmath22 . by definition of outermost barrier , for any point @xmath32 , spacelike extremal surfaces in ext@xmath8 can be deformed to either come arbitrarily close to @xmath1 or coincide with and be tangent to @xmath1 at @xmath32 . in the latter case , we can simply take the limiting surface that touches @xmath1 . in any small neighborhood where @xmath1 and one of these spacelike extremal surfaces touch or nearly touch , eq . [ negcond ] yields @xmath107 , where @xmath24 is the induced metric on the extremal surface , and we are done . in particular , the extremal surface is spacelike and @xmath24 is riemannian , so in order for all of the components of @xmath20 to add up to a nonnegative number , at least one eigenvalue of @xmath136 must be nonnegative . whenever we can find a barrier of entirely negative extrinsic curvature , we can therefore find a barrier of partly nonnegative curvature . this is in agreement with theorem 2.1 , which implies that extremal surfaces can not come arbitrarily close to a surface with strictly negative curvature . the existence of a barrier is therefore quite generic , since we expect it to occur for any surface @xmath1 which splits the spacetime into 2 regions and has negative or vanishing extrinsic curvature components . many of the known analytic solutions that are alads admit a totally geodesic surface that acts an as extremal surface barrier , as well as multiple surfaces with negative extrinsic curvature . totally geodesic surfaces are quite special and therefore spacetimes containing them are not representative of general alads spacetimes . we expect , however , that small perturbations of spacetimes that admit splitting surfaces with negative extrinsic curvature do again result in spacetimes with splitting surfaces of nonpositive extrinsic curvature . the existence of such surfaces is thus stable under small perturbations of the metric . we therefore conclude that _ many generic alads spacetimes admit an extremal surface barrier_. moreover , we show in the examples below that the barrier often separates a region of proximity to a singularity from the boundary of the spacetime . in section [ singularities ] , we further prove that for certain spacetimes , the existence of a compact outermost barrier implies the existence of singularities . the implications of this barrier in the context of ads / cft are disturbing : as rule of thumb , the best - understood aspect of the duality dictionary is generically limited in the scope of information it contains about the bulk . this at best suggests that we must change our approach towards extracting bulk information by using probes that are not extremal surfaces . at worst indicates that complete information of a large class of bulk geometries may simply not be contained in the dual field theory @xcite . in this section , we provide some examples of spacetimes with barriers . some of these barriers are totally geodesic , others have negative extrinsic curvature . the latter condition is stable under small perturbations , although the former is not . the simplest example is pure ads itself . if we consider the metric on the poincar patch : @xmath137 surfaces of constant @xmath138 or surfaces of constant @xmath139 are totally geodesic . since extremal surfaces in pure ads remain on constant time slices , it is clear that they do not cross constant @xmath138 totally geodesic surfaces . similarly , a surface which is anchored between @xmath140 and @xmath141 will not propagate to @xmath142 or @xmath143 . the next example we consider is an isotropic ads cosmology . the metric on global ads can be written , via a coordinate transformation ( see e.g. @xcite ) , as an open friedmann - robertson - walker ( frw ) universe in the interior of the lightcone , and foliated into de sitter slices in the exterior of the lightcone : @xmath144 where the two are related via an analytic continuation ( see fig . [ frw ] ) . any small perturbation to this metric results in a curvature singularity replacing the coordinate singularity . if we couple the metric to , say , a scalar field , as in @xcite , the resulting geometry features a curvature singularity ( the first indication of some sort of an accumulation surface for codimension 2 surfaces in this geometry was found in @xcite ) : @xmath145 where @xmath146 and @xmath147 at the lightcone and @xmath148 is a big - crunch collapse . it was found in @xcite that this spacetime contains a totally geodesic barrier : there is a point @xmath149 at which @xmath150 reaches a maximum . the extrinsic curvature of any constant @xmath138 surface with respect to the past - pointing timelike normal is given by : @xmath151 where @xmath152 is the frw metric inside the lightcone . the constant @xmath153 surface @xmath1 is therefore totally geodesic , and surfaces with @xmath154 have negative extrinsic curvature . we demonstrate the barrier at work here using spacelike radial geodesics as an example of extremal surface probes . this is easily generalized to extremal surfaces with fewer symmetries , but for brevity we limit ourselves to the simplest case . consider a geodesic with endpoints at @xmath155 and @xmath156 , where @xmath125 is one of the suppressed angles in eq . the length functional for this geodesic within the frw region is : @xmath157 where we have chosen @xmath158 to parametrize the geodesic . the boundary conditions @xmath155 and @xmath156 imply @xmath159 . we solve for the geodesic from the lightcone interior outwards , starting from @xmath160 to the lightcone and the exterior region , using the reflection symmetry in @xmath125 along the geodesic path . since @xmath158 is a cyclic coordinate , the hamiltonian is conserved : @xmath161 solving for @xmath162 yields : @xmath163 since the scale factor vanishes at the lightcone and at the singularity , and reaches its maximum at some intermediate time @xmath153 , it follows from eq . [ tprime ] that any geodesic with @xmath164 must propagate towards progressively larger @xmath138 . geodesics with @xmath165 stay on @xmath166 for all @xmath158 . this is precisely what theorem 2.2 states : extremal surfaces in the interior of @xmath1 ( i.e. @xmath154 ) remain in the interior , extremal surfaces in the exterior of @xmath1 ( i.e. @xmath167 ) remain in the exterior , and extremal surfaces that coincide with @xmath1 ( i.e. @xmath153 ) lie on @xmath1 , where @xmath1 is the totally geodesic surface given by @xmath153 . one implication of this result is that , in the isotropic ads cosmology , only geodesics on the exterior of @xmath1 make it to the lightcone at @xmath168 and from there to the boundary . the totally geodesic surface prevents boundary - anchored probes from getting arbitrarily close to the singularity . another prominent example of a spacetime with a totally geodesic barrier surface as prescribed in section [ theorem ] includes any black hole with a stationary horizon @xcite . this category includes alads spacetimes such as schwarzschild - ads , rn - ads , kerr - ads , the btz black hole , the planar ads black hole ( the horizon in this case was separately proven to be a barrier surface in @xcite ) , and the hyperbolic ads black hole . in particular , in ads - schwarzschild , there are two totally geodesic surfaces : the past- and future - directed horizons , denoted @xmath169 and @xmath170 in fig . spacelike extremal surfaces anchored at @xmath171 must therefore always remain in region 1 , while those anchored at @xmath172 must remain in region 3 of the spacetime . extremal surfaces anchored at both @xmath171 and @xmath172 are not constrained by our theorems to remain in any part of the spacetime . in fact , @xcite found that spacelike geodesics with one endpoint at @xmath171 and one endpoint at @xmath172 can probe arbitrarily close to the singularity , and in particular , can cross both @xmath169 and @xmath170 . we emphasize that this does not contradict our results in the previous section because these geodesics are anchored to two boundaries . as a point of interest , @xcite found a barrier surface , or accumulation surface , for codimension 2 extremal surfaces anchored to both boundaries , while , as noted above , radial geodesics observe no such barrier . , has just one boundary , does not have a barrier for all extremal surfaces because spacelike geodesics which enter at early times can leave at arbitrarily late times . however , there is an effective barrier for surfaces that are anchored to the boundary at late times , and there may be a barrier for codimension 2 surfaces . following the first version of this paper , an analysis of the behavior of extremal surfaces in some vaidya spacetimes was done in @xcite . ] it is clear at this point that spacetimes with singularities seem to have a particular proclivity for admitting barriers . this is not a coincidence : corollary 2.6 states that null surfaces foliated by ( marginally ) trapped surfaces act as barriers at least for codimension 2 extremal surfaces . if we additionally assume the stronger statement that _ all _ components of the null extrinsic curvature are nonpositive , we found in corollary 2.4 . that it is a barrier to extremal surfaces of every dimension . in particular , when the null barrier @xmath1 is spatially compact , i.e. the entire boundary is in ext@xmath8 , then @xmath1 is ruled by trapped surfaces . if the spacetime is furthermore globally ( or ads ) hyperbolic , spatially noncompact , and obeys the null curvature condition , the existence of trapped surfaces guarantees the existence of singularities and horizons @xcite . in the next section , we will prove a partial converse of our conclusions from corollary 2.4 : barriers imply the existence of trapped surfaces or singularities , at least for spacetimes which admit a totally geodesic spacelike slice . it is worthwhile here to comment on any bearing this might have on the recent controversy over the completeness of the ads / cft correspondence . the question of whether the black interior is fully described by the boundary field theory is particularly relevant to this discussion . in @xcite , it was shown that there exist operators in the static black hole interior that can not exist in the dual field theory . the fact that a well - used probe of the bulk geometry can not reach into a bulk region which is not completely described by the boundary field theory may not be coincidental . since the limited reach of extremal surfaces in the static black hole geometry is simply a special case of the barrier theorems presented in section [ theorem ] , we may expect that other black hole alads geometries admitting barriers may manifest similar incompleteness on the dual field theory side . one may also speculate that field theories dual to other geometries , not necessarily black holes , may exhibit the same behavior near a singularity . the examples in the previous section suggest that there is some connection between singularities ( particularly singularities masked by horizons ) , and extremal surface barriers . in this section , we will prove this result in some special cases . consider a spacetime manifold @xmath7 containing a totally geodesic spacelike slice @xmath173 . such slices can be found , for instance , in static spacetimes , or more generally , spacetimes with a moment of time reflection symmetry . since we intend to drop in extremal surfaces from the boundary of @xmath173 , we must assume that @xmath173 is noncompact . because @xmath173 is totally geodesic , these extremal surfaces remain on @xmath173 . suppose further that there is an outermost barrier @xmath1 such that the intersection between @xmath1 and @xmath173 is nonempty and compact . then we prove below that @xmath173 either admits a singularity , or else a marginally trapped surface . if , in addition , we assume global hyperbolicity ( including alads spacetimes with the appropriate generalization of global hyperbolicity ) , the null curvature condition ( @xmath174 for all null vectors @xmath175 ) and the generic condition ( roughly , that each null ray encounters at least a little bit of null curvature or shear ) then the existence of a marginally trapped surface itself implies the existence of a singularity somewhere on @xmath7 @xcite . we first prove a lemma : + * theorem 4.1 : * no compact extremal surface barrier @xmath1 in @xmath7 ever touches extremal surfaces anchored to @xmath3 . we prove this by contradiction . let @xmath1 be a barrier , and suppose there exists a spacelike extremal surface @xmath22 which is anchored to @xmath3 . by assumption , @xmath176 everywhere except where it coincides . in any neighborhood of such a coincident point @xmath32 , there exists a point such that : @xmath177 where @xmath178 is the null generator of @xmath1 for null @xmath1 , and the @xmath22-directed normal to @xmath1 otherwise . @xmath22 can not lie on @xmath1 for a continuous neighborhood ( otherwise it would lie on @xmath1 everywhere , which would violate the boundary conditions on @xmath22 ) . @xmath22 can therefore touch @xmath1 while remaining anchored on the boundary . but then we can slightly deform @xmath22 in a neighborhood of @xmath32 , to make a new surface @xmath179 which crosses @xmath1 . using the elliptical equation of extremality , we can solve for @xmath179 outside the neighborhood as well . because the perturbation is small , @xmath179 must still be anchored to the boundary . this shows that @xmath1 is not a barrier , and we have arrived at a contradiction . theorem 4.1 may at first seem to indicate that compact outermost barriers can not exist , since we have defined outermost barriers as the boundary of the union of some set of extremal surfaces : one might naturally ask how the outermost barrier is constructed , if extremal surfaces can not touch it . the barrier must be constructed as a limit of extremal surfaces that come arbitrarily close to touching it . we assume therefore , that one can find sequences of extremal surfaces approaching any point @xmath32 , whose limit is an extremal surface tangent to @xmath1 at @xmath32 . however , the limiting extremal surface must not be anchored anymore to the boundary , or it would contradict theorem 4.1 . we will make use of this limit construction of the outermost barrier in the proof of theorem 4.2 below , where we prove a condition relating the existence of barriers to the presence of singularities and trapped surfaces . because we are taking a limit of extremal surfaces , one may worry that this limit will be related to the boundary in a bizarre way , perhaps by spiraling around , getting arbitrarily close to the boundary without actually being anchored to it . we will deal with these pathological cases by including them in the following definition : a surface @xmath22 is _ weakly anchored _ if there exists a @xmath180 such that all points @xmath181 are less than distance @xmath182 away from the boundary @xmath3 along @xmath22 ( after compactification of @xmath3 ) . with a timelike barrier @xmath1 , projected onto a spacelike codimension 1 totally geodesic slice @xmath173 yields a compact barrier @xmath183 for extremal surfaces on @xmath173.,width=188 ] * theorem 4.2 : * let @xmath7 admit a totally geodesic codimension 1 spacelike slice @xmath173 . are a finite distance away from the boundary after compactification . ] given a compact , nonempty , outermost barrier @xmath183 for weakly - anchored , codimension 2 , @xmath183-deformable ( i.e. where the deformations are restricted to @xmath173 ) spacelike extremal surfaces on @xmath173 ( e.g. fig . [ full ] ) , then one of the following is true : 1 . there exists a singularity outside or on @xmath183 , or 2 . @xmath183 is a marginally trapped surface ( in either time direction ) . since @xmath183 is an outermost barrier , there exist extremal surfaces that either coincide with it and are tangent to it at coincident points , or come arbitrarily close to coinciding with it . the former case is ruled out by theorem 4.1 . let @xmath10 be a family of spacelike extremal surfaces that approach arbitrarily close to some point @xmath184 while remaining weakly anchored on @xmath185 . in a neighborhood of @xmath186 , we take the limit of these surfaces @xmath179 of @xmath187 that reaches @xmath1 at @xmath186 . since we are operating at the `` physics '' standard of rigor , we assume without proof that ( a ) this limit must exist for some sequence of extremal surfaces , and that ( b ) the resulting extremal surface may be extended outside the neighborhood of @xmath186 by solving the elliptical equation of extremality . when we solve this equation , by theorem 4.1 , @xmath179 must no longer be weakly anchored on @xmath188 . since @xmath179 can not be weakly anchored to @xmath188 , part of it must terminate somewhere in the interior of the spacetime . it can do it in one of the following two ways ( fig . [ thm3 ] ) : ( a ) : : @xmath179 terminates in a singularity on or outside @xmath183 , or ( b ) : : @xmath179 spirals outside @xmath183 forever , without ever coming to an end . in that case , we can define @xmath189 to be the limit set of the extremal surface as it gets arbitrarily far away from the points where it is anchored to the boundary . is not ( weakly ) anchored to the boundary anywhere , then @xmath190 . ] @xmath189 must be bounded by some compact innermost surface @xmath191 , where @xmath192 . can not come arbitrarily close to the boundary , since then it would be boundary - anchored by the generalized definition above . ] if ( a ) is true , we get the first part of the claim . we now show that ( b ) results in the second part of the claim . to do so , we need to show that @xmath193 . we do this by contradiction . suppose @xmath194 . then @xmath195 . this means that it is _ not _ a barrier for extremal surfaces weakly anchored at @xmath3 , and therefore by theorem 2.1 there must exist a point @xmath196 such that @xmath197 on the other hand , by construction , @xmath198 can come arbitrarily close to @xmath191 at any point @xmath199 . this implies that the following holds everywhere : @xmath200 ( the proof is essentially the same as theorem 2.7 , but reversing the roles of the interior and exterior . ) we have arrived at a contradiction , so instead we must assume that @xmath201 . this tells us that @xmath179 must in fact coincide with @xmath183 . since @xmath179 is extremal , it follows that @xmath183 satisfies @xmath202 this is the same as saying that @xmath183 is a marginally trapped surface , which proves the second part of the claim . the assumption that there exists an outermost barrier @xmath183 for codimension 2 extremal surfaces living on @xmath173 is reasonable : if we take @xmath7 to have a totally geodesic slice @xmath173 and a compact outermost barrier @xmath1 such that @xmath203 , then in general , we expect that @xmath204 will be compact , and be a barrier for extremal surfaces on @xmath173 . there should then exist some outermost @xmath183 which , at least for alads spacetimes , will still be compact . ( if there is a barrier for _ all _ extremal surfaces , then obviously there must be a barrier for those which are codimension 2 and restricted to @xmath173 . ) if we further assume global hyperbolicity , the null curvature condition , and the generic condition , we are able to say that marginally trapped surfaces only occur in geodesically incomplete spacetimes , so that barriers in these spacetimes exist only in the presence of singularities @xcite . singularities are therefore intrinsically linked to extremal surface barriers . assuming some form of cosmic censorship , these singularities must be hidden behind horizons , thus outermost barriers are also linked to horizons . in particular , the direct application of theorem 4.2 to the ads / cft correspondence suggests that in general , extremal surface probes are not good probes for spacetimes with singularities . we expect that theorem 4.2 can be further generalized to both the noncompact barrier case , as section [ cosmology ] suggests , and to any time - dependent geometry , although this may require assuming the null curvature condition ( implied by the null energy condition and einstein s equations ) , even to prove the existence of the trapped surface . we proved in theorem 2.1 , theorem 2.2 , and corollary 2.4 that a splitting surface with nonpositive extrinsic curvature acts as a barrier to extremal surfaces ( at least those which are @xmath1-deformable ) . this can include spacelike , timelike , or null barrier surfaces . if we are only interested in barriers to codimension 2 surfaces ( as used in the holographic entanglement entropy conjecture @xcite ) , then we have shown in theorem 2.5 and corollary 2.6 that null surfaces foliated by ( marginally ) trapped surfaces are barriers . conversely , for spacetimes with a totally geodesic slice , the existence of a barrier guarantees the existence of either singularities or else trapped surfaces for spacetimes with a totally geodesic slice @xmath173 , so long as the intersection of the barrier and @xmath173 is compact . we also showed that outermost barriers have at least one nonnegative extrinsic curvature component ( theorem 2.7 ) , and we argued that extremal surface barriers occur somewhat generically in alads spacetimes . this presents a setback to attempts to reconstruct the bulk geometries from field theory operators using extremal surface probes . these results are particularly troubling in light of how pervasive extremal surface probes are in applications of the ads / cft correspondence . if there are regions which can not be reconstructed from the entire boundary , than either those regions do not really exist as semiclassical regions ( e.g. because there is a firewall at their boundary ) , or else if they do exist , the information inside of them must be contained in a new factor of the hilbert space which is in addition to the boundary @xcite . we have also studied cases in which there is a barrier to extremal surfaces located on only _ part _ of the boundary . in this case it is not surprising that the entire bulk can not be reconstructed . thus the location of the barrier might give clues about how much of the bulk can be reconstructed from a cft region . see @xcite for discussion of this question . one should bear in mind that extremal surfaces are not the only probe used in ads / cft . if there are other bulk probes besides extremal surfaces , it is possible that a firewall ( if it exists at all ) might be located somewhere behind the outermost barrier . conversely , if there are some extremal surfaces which do not actually correspond to dual observables ( for example , because they are not the _ minimum _ area extremal surface ) , then it could conceivably be that the firewall is actually outside of the barrier . it may also be that there are regions of space where _ some _ kinds of extremal surfaces can probe , but not enough to fully reconstruct the geometry . it is unclear what the status of these regions would be . in the case of a vaidya spacetime where a black hole forms from the collapse of a shell , there is no extremal surface barrier . in particular , there are spacelike geodesics which pass through any point of the black hole interior . the amps argument for firewalls @xcite applies to sufficiently old black holes that form from collapse , yet the absence of a barrier might seem to suggest that there can be no firewall . however , the spacelike geodesics in question are anchored at very large time separation on the boundary . in fact , @xcite argued that the wkb stationary phase approximation is not dominated by spacelike geodesics passing through the interior of the black hole , but rather by complex geodesics . it is unclear how to use the barrier results proven here in the case where the extremal surfaces are complex , as was found in @xcite for two - point correlators in the case of an ads - schwarzschild bulk . however , even for the vaidya spacetime , we still find that there is a partial barrier , for spacelike surfaces anchored sufficiently far to the future of the collapsing shell . this is because the spacetime is just schwarzschild at late times . thus the extremal surface barrier might still give clues about the location of the firewall , although it does not necessarily tell us when the firewall would first appear . in general , our knowledge of the ads / cft dictionary comes from a bootstrapping procedure where ( a ) proposed new ingredients to the dictionary can be checked using our knowledge of bulk physics , while ( b ) the behavior of the bulk can also be predicted using the dictionary . but if at step ( b ) , our current understanding of the dictionary predicts a firewall resulting in a breakdown of the bulk equations of motion used at step ( a)then it is unclear how to proceed . should we modify the bulk equations of motion or the dictionary ? perhaps it is the local bulk equations of motion which should be kept sacrosanct , and other things we think we know about ads / cft should be modified in order to preserve them . ultimately , the question is what choice of dictionary leads to the most consistent , complete , and interesting form of the correspondence . there are several remaining questions worth noting . first , we have yet to find a complete description of the most general barrier , in particular for the outermost barrier surface . such a description would facilitate an understanding of precisely how generic extremal surface barriers are . we expect that , assuming the null curvature condition , it should be possible to show that compact outermost barriers can only occur in spacetimes with singularities . we note that ( aside from theorem 2.5 ) the main results of this paper are purely geometrical in nature they apply to any lorentzian spacetime without needing to use of any energy condition restricting the sign of the curvature . ( although given the existence of trapped surfaces as shown in theorem 4.2 , the null curvature condition would say that there must be singularities . ) this suggests that our results are still valid even in semiclassical spacetimes ( e.g. evaporating black holes ) where local energy conditions are violated . however , the significance of these geometrical results for ads / cft is not the same , since in the semiclassical ( finite @xmath22 ) regime , there are quantum corrections to the dictionary relating extremal surfaces to boundary physics . for example , the entanglement entropy of a cft region is no longer given just by the area of an extremal surface . instead it receives corrections due to the bulk entanglement entropy @xcite . thus we should really be interested in surfaces which extremize the area _ plus _ the bulk entanglement entropy . to find barriers to these surfaces , it is natural to look for null surfaces foliated by quantum trapped surfaces , where the sum of the area and the bulk entropy is decreasing . one can then prove quantum generalizations of the types of classical theorems proven here @xcite . + we thank dalit engelhardt , sebastian fischetti , ahmed almheiri , gary horowitz , don marolf , william kelly , and mudassir moosa for helpful discussions . this work is supported in part by the national science foundation under grant no . phy12 - 05500 . ne is also supported by the national science foundation graduate research fellowship program under grant no . aw is also supported by the simons foundation . a. almheiri , d. marolf , j. polchinski and j. sully , `` black holes : complementarity or firewalls ? , '' jhep * 1302 * , 062 ( 2013 ) [ arxiv:1207.3123 [ hep - th ] ] . a. almheiri , d. marolf , j. polchinski , d. stanford and j. sully , `` an apologia for firewalls , '' jhep * 1309 * ( 2013 ) 018 [ arxiv:1304.6483 [ hep - th ] ] . d. marolf and j. polchinski , `` gauge / gravity duality and the black hole interior , '' phys . lett . * 111 * , 171301 ( 2013 ) [ arxiv:1307.4706 [ hep - th ] ] . k. papadodimas and s. raju , `` state - dependent bulk - boundary maps and black hole complementarity , '' arxiv:1310.6335 [ hep - th ] . s. b. giddings , `` black holes , quantum information , and unitary evolution , '' phys . rev . d * 85 * , 124063 ( 2012 ) [ arxiv:1201.1037 [ hep - th ] ] . j. maldacena and l. susskind , `` cool horizons for entangled black holes , '' arxiv:1306.0533 [ hep - th ] . r. bousso , `` complementarity is not enough , '' phys . d * 87 * , 124023 ( 2013 ) [ arxiv:1207.5192 [ hep - th ] ] . k. papadodimas and s. raju , `` an infalling observer in ads / cft , '' jhep * 1310 * , 212 ( 2013 ) [ arxiv:1211.6767 [ hep - th ] ] . s. g. avery , b. d. chowdhury and a. puhm , `` unitarity and fuzzball complementarity : alice fuzzes but may not even know it! , '' jhep * 1309 * , 012 ( 2013 ) [ arxiv:1210.6996 [ hep - th ] ] . p. kraus , h. ooguri and s. shenker , `` inside the horizon with ads / cft , '' phys . d * 67 * , 124022 ( 2003 ) [ hep - th/0212277 ] . s. ryu and t. takayanagi , `` aspects of holographic entanglement entropy , '' jhep * 0608 * , 045 ( 2006 ) [ hep - th/0605073 ] . v. e. hubeny , m. rangamani and t. takayanagi , `` a covariant holographic entanglement entropy proposal , '' jhep * 0707 * , 062 ( 2007 ) [ arxiv:0705.0016 [ hep - th ] ] . j. m. maldacena , `` wilson loops in large n field theories , '' phys . * 80 * , 4859 ( 1998 ) [ hep - th/9803002 ] . v. e. hubeny , `` extremal surfaces as bulk probes in ads / cft , '' jhep * 1207 * , 093 ( 2012 ) [ arxiv:1203.1044 [ hep - th ] ] . s. s. pal , `` extremal surfaces and entanglement entropy , '' arxiv:1312.0088 [ hep - th ] . l. fidkowski , v. hubeny , m. kleban and s. shenker , `` the black hole singularity in ads / cft , '' jhep * 0402 * , 014 ( 2004 ) [ hep - th/0306170 ] . t. hartman and j. maldacena , `` time evolution of entanglement entropy from black hole interiors , '' jhep * 1305 * , 014 ( 2013 ) [ arxiv:1303.1080 [ hep - th ] ] . j. maldacena and g. l. pimentel , `` entanglement entropy in de sitter space , '' jhep * 1302 * , 038 ( 2013 ) [ arxiv:1210.7244 [ hep - th ] ] . n. engelhardt and g. t. horowitz , `` entanglement entropy near cosmological singularities , '' jhep * 1306 * , 041 ( 2013 ) [ arxiv:1303.4442 [ hep - th ] ] . w. fischler , s. kundu and j. f. pedraza , `` entanglement and out - of - equilibrium dynamics in holographic models of de sitter qfts , '' arxiv:1311.5519 [ hep - th ] . w. w. meeks , iii and s. t. yau , `` the existence of embedded minimal surfaces and the problem of uniqueness , '' _ math . 179(2 ) : 151 - 168 , 1982 . d. marolf and a. c. wall , `` eternal black holes and superselection in ads / cft , '' class . grav . * 30 * , 025001 ( 2013 ) [ arxiv:1210.3590 [ hep - th ] ] . b. czech , j. l. karczmarek , f. nogueira and m. van raamsdonk , `` the gravity dual of a density matrix , '' class . * 29 * , 155009 ( 2012 ) [ arxiv:1204.1330 [ hep - th ] ] . t. hertog and g. t. horowitz , jhep * 0407 * , 073 ( 2004 ) [ hep - th/0406134 ] . unpublished manuscript , n. engelhardt , t. hertog , and g. horowitz . m. mars , `` stability of mots in totally geodesic null horizons , '' class . * 29 * , 145019 ( 2012 ) [ arxiv:1205.1724 [ gr - qc ] ] . p. hjicek , `` exact models of charged black holes . i. geometry of totally geodesic null hypersurface '' , commun . phys . * 34 * , 37 - 52 ( 1973 ) . v. e. hubeny and h. maxfield , `` holographic probes of collapsing black holes , '' arxiv:1312.6887 [ hep - th ] . s. w. hawking and g. f. r. ellis , `` the large scale structure of spacetime , '' cambridge university press , cambridge , 1973 . m. headrick and t. takayanagi , `` a holographic proof of the strong subadditivity of entanglement entropy , '' phys . d * 76 * , 106013 ( 2007 ) [ arxiv:0704.3719 [ hep - th ] ] . a. c. wall , `` maximin surfaces , and the strong subadditivity of the covariant holographic entanglement entropy , '' arxiv:1211.3494 [ hep - th ] . v. e. hubeny and m. rangamani , `` causal holographic information , '' jhep * 1206 * , 114 ( 2012 ) [ arxiv:1204.1698 [ hep - th ] ] . r. bousso , s. leichenauer and v. rosenhaus , `` light - sheets and ads / cft , '' phys . d * 86 * , 046009 ( 2012 ) [ arxiv:1203.6619 [ hep - th ] ] . j. louko , d. marolf and s. f. ross , phys . d * 62 * , 044041 ( 2000 ) [ hep - th/0002111 ] . t. faulkner , a. lewkowycz and j. maldacena , `` quantum corrections to holographic entanglement entropy , '' jhep * 1311 * , 074 ( 2013 ) [ arxiv:1307.2892 [ hep - th ] ] .

we investigate the coherent manipulation of interacting rydberg atoms placed inside a high - finesse optical cavity for the deterministic preparation of strongly coupled light - matter systems . we consider a four - level diamond scheme with one common rydberg level for interacting atoms . one side of the diamond is used to excite the atoms into a collective ` superatom ' rydberg state using either-pulses or stimulated raman adiabatic passage ( stirap ) pulses . the upper transition on the other side of the diamond is used to transfer the collective state to one that is coupled to a field mode of an optical cavity . due to the strong interaction between the atoms in the rydberg level , the rydberg blockade mechanism plays a key role in the deterministic quantum state synthesis of the atoms in the cavity . we use numerical simulation to show that non - classical states of light can be generated and that the state that is coupled to the cavity field is a collective one . we also investigate how different decay mechanisms affect this interacting many - body system . we also analyze our system in the case of two rydberg excitations within the blockade volume . the simulations are carried out with parameters corresponding to realizable high - finesse optical cavities and alkali atoms like rubidium . pacs numbers : : 32.80.ee , 42.50.ex , 37.30.+i , 42.50.

1504.04898

we investigate using rydberg atom interactions to deterministically synthesize collective quantum states that can be strongly coupled to an optical cavity for applications in quantum information science , quantum optics and the production of quantum light fields . the method enables a single or several collective excitations to be prepared in a high - finesse optical cavity . the number of excitations is precisely controlled so that photon number states can be produced using the coupled atoms - cavity system . the system we describe in this paper differs from typical jaynes - cummings ( jc ) and tavis - cummings ( tc ) physics in that strong atom - atom interactions play a key role in the initial state preparation , which is the focus of this paper . rydberg atoms have many exotic properties . for example , rydberg atoms are large , interact strongly with one another , have relatively long lifetimes , and can be manipulated with external electric and magnetic fields @xcite . particularly important to the work presented in this paper , the interactions between ultracold rydberg atoms can be larger than typical frequency stabilized laser spectral bandwidths and kinetic energies at distances of over @xmath2 m @xcite . the interactions can be manipulated easily with electric fields , relative to valence states @xcite . there is a broad range of behavior that can be induced when the multi - level character of the interactions is taken into account . rydberg atom blockade is a result of the strong interactions between rydberg atoms . in rydberg atom blockade , the long range interactions between rydberg atoms suppress multiple excitations in a volume in which the interactions are large enough to shift the excitation laser out of resonance with the rydberg atom transition frequency @xcite . it has been realized over the last 15 years that rydberg atom blockade is useful for a host of applications , particularly in quantum optics and quantum information science . rydberg atom blockade can be used to control quantum dynamics and prepare interesting quantum states @xcite . rabi oscillations have been observed in effective two - level ultracold rydberg atoms @xcite demonstrating long coherence times and collective behavior . collective rabi oscillations on rydberg transitions have even been observed in a thermal rubidium ( rb ) vapor on a time scale below @xmath3ns @xcite . atoms contained in cavities , both optical and microwave , have been important systems for investigating the quantum - classical boundary in physics . the ability to entangle atoms with light presented by cavity quantum electrodynamics has also served as a model system for applications in quantum information science @xcite . a cavity quantum electrodynamical system with a single excitation or controlled number of excitations is useful both for fundamental studies @xcite and applications that use quantum behavior , particularly those involving quantum entanglement @xcite . entangled states can be created between distant atoms in one cavity or separate cavities @xcite . trapping of a single atom in a high - finesse optical cavity has been achieved @xcite . however , it is technically challenging to couple the light in the cavity to a single atom because of the small absorption cross - section @xcite . the problem of small absorption cross - sections can be handled by using an atomic ensemble , where light more strongly interacts with collective atomic states . in this paper , we theoretically show that one can excite collective @xmath0 atom superposition states using rydberg atom interactions that can be strongly coupled to a high - finesse cavity . we observe @xmath4 enhancement of the rabi oscillations within the cavity showing that the collective state can be realized @xcite and analyze the decay mechanisms that affect this interacting many - body system . collectively enhanced cooperativity factors help to improve the efficiency of photon generation out of the cavity . the efficiency can be improved by increasing the number of atoms . we also show that by chirping the laser excitation pulses in time , the single atom rydberg blockade effect can be suppressed and we can create two rydberg excitations within the blockade radius that are coupled to the cavity . we use a four - level diamond type atomic scheme as shown in fig . [ fig:1](b ) . one side of the diamond is used to collectively excite the atoms to the rydberg level using a pair of @xmath1-pulses or a pair of counter - intuitive stimulated raman adiabatic passage ( stirap ) pulses with rabi frequencies @xmath5 and @xmath6 . by using a laser pulse , @xmath7 , and the single atom cavity coupling strength , @xmath8 , the state of the collective atoms - cavity system can be controlled . rydberg atom blockade plays an important role in this process , since we are interested in cases where a single or controlled finite number of coherently shared excitation(s ) are produced in a collection of @xmath9 state rb atoms . the @xmath0 interacting atoms - cavity system is within the reach of experimental realization @xcite . the system can be used to prepare single photons using a single excitation and more complex quantum light fields by preparing multiple excitations in the cavity . recently , a jc model in the optical domain has been proposed for investigating rydberg blockaded atomic ensembles in a cavity . the transmission through the cavity was studied numerically by monte - carlo simulation @xcite . in contrast to this work and others , our studies resemble a pumping scheme for inverting the population of a laser , except the upper state of the lasing transition is prepared in a collective state . the pumping scheme is similar to a raman or superradiant laser and work done on low atom number masers and micromasers . however , for these types of lasers and masers , the atoms are coherent but non - interacting @xcite . our approach is useful for single photon generation @xcite , as well as the generation of other types of quantum light fields , and coherent optical manipulation for quantum information processing @xcite . the collective atoms - cavity system can also be important for generating unique quantum states such as superposition states of the atomic ensemble @xcite . the approach does not require phase matching and it may be easier to use this system than those based on phase matching to create more elaborate quantum light fields @xcite . creating arbitrary quantum light fields remains an important challenge for quantum optics , so it is useful to explore a wide range of different strategies . the method that we explore in this paper may have technological and experimental advantages because the collective state is prepared on a set of transitions which have significantly different excitation wavelengths , @xmath10 nm for all transitions in rb , than the cavity mode to which it is coupled . the cavity output can be filtered and detected with low background . fig . [ fig:1 ] shows a conceptual drawing of the experimental setup . the setup consists of a collection of atoms trapped in a high - finesse optical cavity . the atomic energy level scheme is a four - level diamond scheme with atomic states denoted by @xmath11 , @xmath12 , @xmath13 and @xmath14 . @xmath15 is the energy of the @xmath16 atomic level , therefore the transition frequencies @xmath17 where @xmath18 and @xmath19 are labels corresponding to the respective states @xmath11 , @xmath12 , @xmath13 and @xmath14 . the rabi frequencies are taken to be time dependent since we are concerned with applying pulses for the preparation of a collective state that we can couple to the cavity as well as with controlling the photons emitted from the cavity . the decay processes that we consider explicitly in our model are also shown . we envision the atoms to be held in the cavity with a far off resonance dipole trap ( fort ) or a magnetic trap . the dipole trapping wavelength can ideally be set to a magic wavelength so that the frequency of the @xmath20 transition is fixed in the cavity . we discuss example design parameters in this section as well as the following 2 . additional decay processes and the consequences of the main approximations used are also discussed later in the paper . and @xmath21 with rabi frequencies @xmath22 and @xmath7 , respectively . the transition @xmath20 is coupled to an optical cavity with coupling strength @xmath8 . @xmath23 is the decay rate of the rydberg state , @xmath24 is the transverse spontaneous decay rate of the @xmath12 state and @xmath25 is the cavity decay rate . energy levels are shown for rb . [ fig:1],width=576 ] one side of the diamond is used to collectively excite the atoms to a rydberg level using a pair of @xmath1-pulses or counter - intuitive stirap pulses with rabi frequencies @xmath5 and @xmath6 . @xmath5 drives the @xmath26 transition , frequency @xmath27 , while @xmath6 drives the @xmath28 transition , frequency @xmath29 . if the interaction between the rydberg atoms is strong enough so that the @xmath30 transition is shifted out of resonance after the first rydberg excitation occurs , further population of @xmath14 is blocked . the other side of the diamond system is used to produce a collective state that is close to resonance with a field mode of a high - finesse optical cavity . the optical cavity mode is coupled to the @xmath20 transition since the resonant frequency of the cavity , @xmath31 , is assumed to be close to the atomic transition frequency , @xmath32 . in fact , @xmath33 for most of the calculations that follow . the control laser is treated classically and has a frequency @xmath34 that couples to the @xmath35 transition at frequency @xmath36 with a rabi frequency @xmath7 . the cavity field is treated as a quantum field . the single atom states of the coupled atoms - cavity field system are products of @xmath37 or @xmath38 , each coupled to a ladder of photon number states . the density matrix equations are not easily solvable for an arbitrary number of atoms because the number of states increases dramatically with @xmath0 . one way to reduce the basis set size in a many - body system is to use a mean - field approximation . for an ultracold rydberg gas this means each highly excited atom shifts the rydberg levels of surrounding atoms out of resonance with the laser field @xcite . in this approach , atomic coherence is damped out and the quantum master equations reduce to many - body rate equations . this method is useful for understanding excitation suppression due to rydberg atom blockade in cold gases @xcite . a second way to practically address the increasing basis set size is to utilize the idea of local blockade and truncate the basis set when the number of excitations in the system exceeds some limit @xcite . the second method is appropriate when the system dynamics are constrained to a subspace of the complete basis . rydberg atom dipole blockade naturally justifies this approach . in the spirit of the latter approach , we carry - out our calculations in a truncated hilbert space , similar to that used by saffman and walker @xcite . in our calculations , we assume that the interaction volume is less than the blockade volume . this allows for the modeling of the blockade effect through a straightforward detuning of states with more than a single rydberg excitation . the level shift between a pair of atoms labeled @xmath18 and @xmath19 each in a rydberg state can be estimated using @xmath63 where two types of simplified interatomic interactions can be considered : ( a ) dipole - dipole interaction where @xmath64 , and ( b ) van der waals interaction where @xmath65 . @xmath66 are the dispersion coefficients , and @xmath67 and @xmath68 are the positions of the @xmath16 and @xmath69 atoms , respectively . here , we are primarily interested in repulsive potentials due to the atom - atom interaction . more realistic rydberg atom interaction potentials have a complicated dependence on internuclear separation , including many avoided crossings that can be manipulated with electric and magnetic fields @xcite . we use eq . [ interaction ] to estimate the magnitude of the interaction whose exact value and radial dependence is not critical provided it is large enough for rydberg blockade for the present discussion . the hamiltonian for atom - atom interaction can be written as @xmath70 the condition that all atoms in the interaction volume lie within the blockade volume translates to @xmath71 , \label{blockadecond}\ ] ] where @xmath72 is the spontaneous decay rate of the intermediate state @xmath13 . the situation where all atoms are within a blockade volume can be realized by using sufficiently narrow bandwidth lasers to form an interaction volume with rydberg atom interactions strong enough to satisfy eq . [ blockadecond ] . in this work , the atoms are also assumed to be cold enough that they can be approximated as stationary , or frozen in space . these ideas are extendable to countable numbers of excitations by increasing the interaction volume or decreasing the blockade volume . we also consider the case where we want to compensate the shift due to the one rydberg excitation blockade , @xmath73 . this shift can be compensated by changing the time dependent detuning @xmath42 of the excitation fields , so that we can excite two rydberg atoms in the blockade volume . more elaborate states can be created by , for example , applying more pulses to create several collective excitation states . note that for 2 excitations , rather flat avoided crossings that occur in rydberg atom interactions can be used to create 2 excitation states with minimized spatial correlation @xcite . the total hamiltonian of the @xmath0 atom system coupled to the cavity is @xmath74 the density matrix operator @xmath75 of the @xmath0 interacting atoms and all the fields obeys liouville s equation @xmath76 + \mathcal{l}\hat{\rho } , \label{evolution}\end{aligned}\ ] ] where the decays are due to spontaneous emission from @xmath77 and @xmath14 and loss from the cavity . these effects are included in the second term of the equation using the lindblad operator @xmath78 . @xmath79 can be written as @xmath80 with @xmath81 \big ) , ~~\end{aligned}\ ] ] @xmath82 \big ) , ~~\end{aligned}\ ] ] @xmath83 \big ) , ~~\end{aligned}\ ] ] @xmath84 @xmath85 and @xmath86 where @xmath87 is the rydberg state decay rate , @xmath88 is the transverse spontaneous emission rate into the free space modes and @xmath25 is the cavity decay rate . a dummy state @xmath89 is used to model decay into non - cavity modes and loss from the collective state manifold due to dephasing . here the ` 0 ' includes all other modes of the light field to simplify the notation . in particular , we use the dummy state to model the effect of additional dephasing mechanisms described in the discussion section and to estimate their effect . we define the parameter @xmath90mhz , the free space spontaneous decay rate corresponding to @xmath91rb , as a scale for the decay rates used for the calculations that follow . the total rydberg state decay rate is @xmath92 khz for @xmath93 . the black body radiation decay rate of the rydberg state is very small , @xmath94 khz for @xmath93 , compared to the total decay rate . consequently , we ignore black body decay which would cause atoms to leave our model system @xcite . at large principal quantum numbers , @xmath46 , where the interactions are strongest , which are most appropriate for the scheme described in this paper , the neglect of black body decay is best justified . @xmath95 is small but we include it for completeness in the liouville equations . the time variation of the electric field of the laser pulses is of the form , @xmath96.\ ] ] @xmath97 is the amplitude , @xmath98 is the time when the pulse reaches its maximum value and @xmath99 is the pulse width . @xmath100 is taken to be smaller than the shift due to the interaction of rydberg atoms @xmath73 . the key parameter for determining the emission rate into the cavity for a single atom is @xmath101 , which is the ratio of the rate of emission into the cavity mode to the loss due to spontaneous emission into other electromagnetic field modes . this ratio is an expression for the purcell factor , @xmath102 , described in more detail in the next section , and is related to the single atom cooperativity factor , @xmath103 . the cavity decay rate @xmath25 is determined by the mirror reflectivities . the single atom coupling strength @xmath8 is determined by the atomic transition dipole moment , @xmath104 , for the transition near the cavity resonance and the mode volume of the cavity , @xmath105 $ ] . realistic cavity parameters limit @xmath106 . for collective excitations of @xmath0 atoms , the effective purcell factor and collectively enhanced cooperativity increase , since @xmath104 increases like @xmath107 . the blockade radius depends on the states of the rydberg atoms . the blockade radius for principal quantum number @xmath123 , corresponding to the @xmath124mhz excitation rabi frequency , is @xmath125 m . at @xmath123 , the shifts due to the interaction of rydberg atoms @xmath126 mhz at @xmath127 m @xcite . for densities of @xmath128 @xmath129 , there are @xmath130 atoms in the blockade volume with @xmath107 shot to shot atom number fluctuations of @xmath131 atoms . the possible number of atoms in the blockade volume far exceeds the atom numbers considered in this paper . small atom numbers , as studied here , can be generated by decreasing the trap density . in the purcell regime , the coupling to the cavity , accounted for by @xmath8 , can be replaced with an enhanced spontaneous emission rate into the cavity mode . we explore this regime by replacing the direct coupling of the atoms to the cavity with an incoherent decay term into one mode of the electromagnetic field @xmath146 with @xmath147 \big).~~\end{aligned}\ ] ] effectively , the cavity is explicitly removed from the problem and the picture becomes one where an atom , or collective excitation , in free space preferentially emits into a specific mode of the electromagnetic field . the purcell picture provides physical insight into preparing and detecting a countable number of excitations created in a cavity . first , for low @xmath0 and a single rydberg excitation as a target , we consider the possibility of 2 rydberg excitations occurring in our system . in these calculations , we show that , for @xmath148 and @xmath149 , there is insignificant leakage out of the hilbert space bounded by a single rydberg excitation . for calculations with @xmath150 , we use these results to justify the truncation of the hilbert space at the 2 rydberg excitation level . these results are also supported by numerical results of many other works done on rydberg blockade . in these studies it is demonstrated that a large probability ( @xmath151 ) for single rydberg excitation of a superatom can be achieved while the probability of multiple excitations is negligible @xcite . in fig . [ fig:2 ] , we show the results for @xmath152 atoms with one rydberg excitation . the model is efficient for investigating the case of @xmath153 , but here we focus our attention on a few interacting atoms within the cavity . the time dependence and peak amplitudes of the rabi frequencies are shown in fig . [ fig:2](a - c ) , while the probabilities for various states are shown in fig . [ fig:2](d - i ) . for @xmath154 and @xmath155 , the results are exact within the bounds of the model . to evaluate the solution for @xmath156 , we ignore the higher order excitation states . the population in those states are @xmath157 0.0015 justifying the truncation at the 2 excitation level for the work addressed in this paper . to further verify the model , we numerically investigate the preparation of a single collective excitation and the observation of a @xmath107 enhancement of the rabi oscillations of the superatom in the cavity . collective states coupled to the cavity will exhibit a @xmath107 enhancement in their effective atoms - cavity coupling strength , @xmath158 . interacting atoms . the effective rabi frequency @xmath159 is represented by the black solid curve . the red dotted curve shows the rabi frequency of the control laser pulse @xmath7 and the blue dashed curve shows the cavity coupling strength @xmath8 . panels ( d - i ) show the corresponding probabilities for population of states @xmath160 ( black solid ) , @xmath161 ( red dashed ) , @xmath162 ( blue solid ) , @xmath163 ( magenta dot - dashed ) and @xmath164 ( green dotted ) . parameters for the calculation are @xmath165khz , @xmath166mhz , @xmath167mhz @xmath168 , @xmath169 , @xmath170mhz , @xmath171mhz and @xmath172mhz . [ fig:2],width=604,height=566 ] for @xmath173 , we used two @xmath1 pulses , @xmath39 and @xmath40 , applied at the same time to prepare an atom in the rydberg state . subsequently , a control laser pulse with rabi frequency @xmath7 is turned on to transfer the collective state to one that is near to resonance with the cavity . the population oscillates within the cavity between @xmath162 @xmath174 @xmath163 . due to the cavity decay rate , @xmath175 , these oscillations decay with time @xcite . for @xmath155 and @xmath176 , two counter - intuitive stirap pulses @xcite with pulse areas of @xmath1 were used to create a collective state with a single rydberg excitation . we also increased the amplitude of the control laser pulse to optimize the transfer probability . in fig . [ fig:2](e ) and fig . [ fig:2](f ) , the frequency of the rabi oscillations are enhanced by @xmath107 in the cavity . the decay rate and detuning of the cavity were the same as in fig . [ fig:2](d ) . we also obtained the same results as shown in fig . [ fig:2](e ) and fig . [ fig:2](f ) using two superposed @xmath1 pulses ( i.e. without using a stirap sequence ) . the plots in fig . [ fig:2 ] demonstrate that only a single excitation is prepared in the cavity and it collectively interacts with the cavity field . this result shows our numerical approach is consistent with expectations . in fig . [ fig:2](g - i ) , we can see that the probability of populating the cavity ground state , @xmath160 , is increased with increasing number of atoms . for @xmath177 , due to imperfect @xmath1-pulses , the population returning to the rydberg state increases compared to fig . [ fig:2](d - f ) where there is no decay @xcite . these results suggest that the probability of populating the cavity ground state depends on the number of atoms and the spontaneous emission rate out of the cavity . as we increase the coupling to the cavity mode , it , not surprisingly , makes the photon production from the cavity more efficient . this is the result of increasing the effective ( collectively enhanced ) cooperativity . [ fig:3 ] shows the time variation of the rabi oscillations for @xmath0 atoms in a resonant cavity , with @xmath169 [ fig . [ fig:3](a - c ) ] , and @xmath178 [ fig . [ fig:3](d - f ) ] , for different cavity decay rates @xmath179 . as the cavity decay rate is increased , the rabi oscillations are more quickly damped out @xcite . however , increasing the number of atoms enhances the frequency of the oscillations by the @xmath107 factor as shown in fig . [ fig:3](a - c ) . these oscillation frequencies are further enhanced by detuning the cavity @xcite at the expense of a decreasing probability amplitude for photon production as shown in fig . [ fig:3](d - f ) . these results are consistent with the jc model . interacting atoms in the cavity for cavity detuning ( a - c ) @xmath169 and ( d - f ) @xmath180 4.3 mhz @xmath181 . three different cavity decay rates are shown . the black solid curve corresponds to @xmath182 1.4 mhz @xmath183 , the red dotted curve corresponds to @xmath182 2.85 mhz @xmath184 , the blue dashed curve corresponds to @xmath182 4.3 mhz @xmath185 , and @xmath186 . other parameters are the same as in fig . [ fig:2 ] . [ fig:3],width=537 ] the rate of single photon emission for a given number of atoms versus time is shown in fig . [ fig:4 ] with cavity decay rate @xmath175 and @xmath187 . in this case , we observe oscillations in the rate of photon emission that change with @xmath0 . the most interesting thing here is that the rate of photon emission is more strongly peaked at early times by increasing @xmath0 . these results demonstrate that preparing the collective excitation helps to have the photon emitted at a specific time , due to an increase in the effective cooperativity from the collective state . atoms with one rydberg excitation vs time . the black solid curve represents the photon emission with @xmath154 , the red dotted curve corresponds to @xmath155 and the blue dashed curve corresponds to @xmath176 . the cavity decay rate is @xmath182 1.4 mhz @xmath183 and @xmath188 3 mhz @xmath189 . other parameters are the same as in fig . [ fig:2 ] . [ fig:4],width=288 ] interacting atoms . the effective rabi frequency @xmath159 is represented by the black solid curve . the red dotted curve shows the rabi frequency of the control laser pulse @xmath7 and the blue dashed curve shows the cavity coupling strength @xmath8 . panels ( d - f , j - l ) show the corresponding probabilities for the population of states @xmath160 ( black solid ) , @xmath162 ( red dot ) , @xmath161 ( blue dashed ) , @xmath163 ( magenta dot - dashed ) and @xmath164 ( green dotted ) . the cavity decay rate is @xmath182 72.6 mhz . other parameters are the same as in fig . [ fig:2 ] . [ fig:5],width=537 ] in fig . [ fig:5 ] , we show the results for @xmath152 atoms for the full model and with an incoherent decay based on an enhanced spontaneous emission into the cavity mode as discussed in section iv . the time dependence and amplitudes of the rabi frequencies for the incoherent decay associated with the purcell effect are shown in fig . [ fig:5](a - c ) , while the probabilities for various states are shown in fig . [ fig:5](d - f ) . rabi frequencies for the full model are shown in fig . [ fig:5](g - i ) with the associated probabilities for each state in fig . [ fig:5](j - l ) . in both cases , we observe that the photon emission efficiency is enhanced with number of atoms because the cooperativity is effectively increasing due to the collectively enhanced @xmath104 . the efficiency , for a given number of atoms , for an incoherent decay for the purcell effect based calculation is greater than in the full model because there is significant transient population of the cavity excited state in the full calculation . population of @xmath162 can spontaneously decay into a free space electromagnetic field mode , degrading the probability that the photon is emitted into the cavity mode . interacting atoms with different cavity length , @xmath114 . the mirror radius of curvature @xmath113 = 10 cm . @xmath25 is decreasing to increase c for fixed @xmath8 . other parameters are the same as in fig . [ fig:2 ] . [ fig:6],width=614 ] in fig . [ fig:6 ] , the photon emission efficiency is plotted as a function of the cooperativity , c , for @xmath0 interacting atoms with different cavity length @xmath114 . as we decrease @xmath114 , the photon emission efficiency can be improved because @xmath114 increases the single atom @xmath8 . the efficiency of photon emission also depends on the control laser pulse which transfers the rydberg population into the cavity . for large cavity lengths , we can see that the photon emission efficiency first increases with c and then decreases . although c is increasing , @xmath25 is decreasing . the efficiency decreases because @xmath25 becomes smaller compared to @xmath88 . as we increase the number of atoms , the photon emission efficiency can be further improved by increasing the effective cooperativity but only when cavity loss beats out spontaneous emission . interacting atoms with ( a - c ) @xmath190 72.6 mhz @xmath191 , and ( g - i ) @xmath190 8 mhz @xmath192 . the effective rabi frequency @xmath159 is represented by the black solid curve . the red dotted curve shows the rabi frequency of the control laser pulse @xmath7 and the blue dashed curve shows the cavity coupling strength @xmath8 . panels ( d - f , j - l ) show the corresponding probabilities for the population of states @xmath160 ( black solid ) , @xmath162 ( red dotted ) , @xmath161 ( blue dashed ) , @xmath163 ( magenta dot - dashed ) and @xmath164 ( green dotted ) with ( d - f ) @xmath193 , and ( j - l ) @xmath194 . other parameters are the same as in fig . [ fig:2 ] . [ fig:7],width=537 ] in fig . [ fig:7 ] , we show the results for @xmath152 atoms in the weak ( @xmath195 ) and strong ( @xmath196 ) coupling regime . the time dependence and peak amplitudes of the rabi frequencies for the weak coupling regime are shown in fig . [ fig:7](a - c ) and for the strong coupling regime in fig . [ fig:7](g - i ) . the probabilities for various states are shown in fig . [ fig:7](d - f ) and fig . [ fig:7](j - l ) . in fig . [ fig:7](d - f ) , the rabi oscillations between @xmath162 and @xmath163 can not be observed due to the comparatively large cavity decay rate @xmath25 . @xmath163 decays and produces a photon that escapes the cavity . if we increase the value of the single atom cooperativity , for a given number of atoms , the population of the @xmath164 state decreases and the population of the cavity ground state is enhanced . if we increase the number of atoms , we observe that the ground state population is also increased . thus , the increase in effective cooperativity resulting from an increase in the number of atoms plays a constructive role in enhancing the efficiency of photon production . in fig . [ fig:7](j - l ) , the rabi oscillations between @xmath162 and @xmath163 can not be observed because there is a coherent population transfer from the rydberg state @xmath161 to the cavity ground state @xmath197 . the transient population in @xmath162 is approximately zero . this reduces the loss due to spontaneous emission into the dummy state @xmath164 and improves the photon emission efficiency into the cavity mode @xcite . we can also investigate the more complicated situation where two rydberg atoms are excited within the blockade radius . the level scheme for this case is shown in fig . [ fig:8 ] . using a sequence of pulses with effective rabi frequencies @xmath198 and @xmath199 while changing the one photon rydberg level detuning @xmath42 to compensate the blockade shift @xmath73 , we can excite two rydberg atoms into the state @xmath200 @xcite . then we have two channels : ( 1 ) @xmath201 and ( 2 ) @xmath202 by which @xmath203 can be produced . by applying a control laser pulse with rabi frequency @xmath7 and the cavity coupling strength @xmath8 , we can observe emission of two photons from different atomic dipoles out of the cavity mode . by using eq . [ evolution ] , we can calculate the photon correlations . interacting atoms coupled to a cavity with coupling strength @xmath8 [ blue dashed line ] . @xmath198 [ black solid line ] and @xmath199 [ brown solid line ] are the effective rabi frequencies of the laser pulses for one and two rydberg excitations , respectively . @xmath7 [ red dotted line ] is the rabi frequency of the control laser pulse . ( b ) pictorial view of two rydberg excited atoms coupled with the cavity . the excitation is effectively a macrodimer state.[fig:8],width=576 ] interacting atoms with ( a ) @xmath182 72.6 mhz @xmath191 , and ( b ) @xmath182 8 mhz @xmath204 . rabi frequencies @xmath205 and @xmath206 are represented by the black solid and red dotted curves , respectively , and generate the first and second rydberg excitations , respectively . the blue dashed curve shows @xmath7 and the magenta dot - dashed curve shows @xmath8 . ( c , d ) probability of finding atoms in @xmath160 ( black solid ) , @xmath161 ( red dotted ) , @xmath200 ( blue dashed ) , @xmath207 ( magenta dot - dashed ) and @xmath164 ( green dotted ) . ( e , f ) probability of finding atoms in @xmath208 ( black line ) , @xmath209 ( red line ) , @xmath210 ( blue line ) , @xmath211 ( magenta line ) , @xmath162 ( green line ) , @xmath163 ( brown line ) . parameters for the calculation are @xmath165khz , @xmath212 mhz @xmath213 , @xmath169 , @xmath170mhz , @xmath172mhz and @xmath42 changes with time from @xmath214 to @xmath215 mhz . [ fig:9],width=576 ] in fig . [ fig:9](a , b ) , we show the rabi frequencies versus time for the system of @xmath216 interacting atoms . rabi frequency @xmath205 drives the first excitation while @xmath206 drives the second excitation . the cavity decay rates are @xmath193 and @xmath194 in fig . [ fig:9](a , b ) respectively . in fig . [ fig:9](c , d ) , we can see that a 1 rydberg excitation state is generated by using an effective rabi pulse @xmath198 . subsequently , by applying another effective rabi pulse @xmath199 , we can generate a 2 rydberg excitation state . this state is similar to a rydberg atom macrodimer @xcite . a single off - resonant excitation of the macrodimer can also be used , equivalent to a photoassociation of the molecule @xcite . by using control laser pulse @xmath7 with cavity coupling strength @xmath8 , we observe that more transient population builds up in @xmath162 and @xmath208 compared to @xmath217 and @xmath163 . this situation increases the spontaneous emission from @xmath162 and @xmath208 to the dummy state @xmath89 . the photon emission efficiency in the weak cavity coupling regime is reduced due to the spontaneous emission . if we increase the cooperativity the photon emission efficiency can be improved . [ fig:9](d , f ) corresponds to the strong coupling regime . in this case , we can populate the 2 rydberg excitation state and observe photons emitted from the cavity . we see that the photons are quickly emitted from the cavity and photon emission is improved because the transient population of @xmath162 and @xmath208 is negligible compare to @xmath217 and @xmath163 . the population is coherently transferred from the rydberg state to the cavity ground state , similar to the single excitation case . interacting atoms with ( a ) @xmath182 72.6 mhz @xmath191 , and ( b ) @xmath182 8 mhz @xmath204 . rabi frequencies @xmath205 and @xmath206 are represented by the black solid and red dotted curves , respectively , and generate the first and second rydberg excitations , respectively . the blue dashed curve shows @xmath7 and the magenta dot - dashed curve shows @xmath8 . ( c , d ) probability of finding atoms in @xmath160 ( black solid ) , @xmath161 ( red dotted ) , @xmath200 ( blue dashed ) , @xmath207 ( magenta dot - dashed ) and @xmath164 ( green dotted ) . ( e , f ) probability of finding atoms in @xmath208 ( black line ) , @xmath209 ( red line ) , @xmath210 ( blue line ) , @xmath211 ( magenta line ) , @xmath162 ( green line ) , @xmath163 ( brown line ) . other parameters are the same as in fig . [ fig:9 ] . [ fig:10],width=576 ] in fig . [ fig:10 ] , for @xmath176 interacting atoms , the detunings and blockade shift are the same as in fig . [ fig:9 ] . similar to fig . [ fig:9 ] , the photons are quickly emitted from the cavity and some population decays into the dummy state @xmath164 . the photon emission probability is larger in this case because the effective cooperativity is larger . although small , as we increase the number of atoms the photon emission probability is enhanced , due to the effective increase in @xmath8 . 72.6 mhz @xmath191 , and ( b ) @xmath182 8 mhz @xmath204 . other parameters are the same as in fig . [ fig:9 ] . [ fig:11],width=460 ] the expectation value of the dipole - dipole correlation for the case of two rydberg excited atoms is shown in fig . [ fig:11 ] . this quantity is given by @xmath218 and measures the cross - correlation existing between the dipoles belonging to the two atoms @xcite . @xmath219 are the raising and lowering operators for the cavity transition which can be defined as @xmath220 and @xmath221 , where subscript @xmath18 represents the @xmath16 atom . in fig . [ fig:11](a , b ) , the normalized dipole - dipole correlation function as a function of time is shown . we see that the dipoles are correlated . the laser pulses and cavity coupling time dependencies are the same as shown in fig . [ fig:9](a , b ) . in fig . [ fig:11](a ) , with spontaneous emission , the prominent dip observed is due to the transient population in the cavity excited state that results from spontaneous emission outside the cavity . [ fig:12](a , b ) shows the rate of photon emission efficiency for two interacting rydberg atoms with two excitations for cavity decay rates @xmath193 and @xmath194 respectively . in fig . [ fig:12](a , b ) , the weak cavity coupling case has larger photon emission amplitude than the strong coupling regime . with the addition of spontaneous emission , the rate of photon emission decreases significantly in the weak cavity coupling case , again due to the transient population in the cavity excited state . the second order correlation function of the photons emitted by the two different atoms is given by @xmath222 where subscripts @xmath223 represent the @xmath16 and @xmath69 atoms respectively . if log[g@xmath224 > 0 $ ] , radiated photons tend to be bunched while if log[g@xmath224 < 0 $ ] , the photons are anti - bunched . this function is plotted in fig . [ fig:12](a , b ) with and without spontaneous emission as a function of time for @xmath193 and @xmath194 . in these cases , we observe that log[g@xmath225 is smaller than zero for short times when spontaneous emission is neglected . this indicates that the photons are anti - bunched at short times when the photons are preferentially emitted . the value of log[g@xmath225 is greater than zero for longer times , indicating the emission of photons is bunched after the initial photon pulse . if we consider the spontaneous emission into the dummy state @xmath226 then log[g@xmath225 is always greater than zero . log[g@xmath224 > 0 $ ] implies the emission of photons is always bunched for the more realistic case . the additional dephasing destroys the interferences of the 2 oscillating dipoles in this case . ] of photons emitted by two different atoms with two rydberg excitations versus time . ( e , f ) variation of total entropy of the system with time . ( a , c , e ) @xmath182 72.6 mhz @xmath191 and ( b , d , f ) @xmath182 8 mhz @xmath204 . other parameters are the same as in fig . [ fig:9 ] . [ fig:12],width=441 ] to define the entanglement and purity of the quantum systems , there are several measures . one of them is the von neumann entropy of the quantum system . this is a very sensitive measure of the purity of the quantum system and it determines the degree of correlation between the collective excitations and photon states @xcite . the von neumann entropy of the atoms - cavity system can be defined as @xmath227 with density operator @xmath228 where the @xmath229 s are the diagonalized eigenvalues of the system and @xmath19 is the number of states in the hilbert space . since our system is multi - partite , and there is no general measure to study the entanglement for multi - partite system , we approximate our system as a bipartite system of atoms and photons . then , the sum of the entropies of the 2 sub - systems atoms , @xmath230 , and photons , @xmath231 , should be larger than the total entropy of the system , @xmath232 . the bound for the entropy for these two interacting sub - systems can be defined by the araki and lieb @xcite triangle inequality , @xmath233 the total entropy of the system is plotted in fig . [ fig:12 ] taking into account and also neglecting spontaneous emission as a function of time . [ fig:12](e , f ) shows the variation of total entropy with cavity decay rates @xmath193 and @xmath194 . as photons are emitted from the cavity , the entropy of the system first increases but then decreases due to the loss of the photons . in fig . [ fig:12](e , f ) , the entropy of the system is larger for the realistic case where spontaneous emission out of the cavity is considered . in both cases , after the complete emission of the photons , the entropy of the photonic sub - system has reached zero with time and system is in a pure entangled atomic state @xcite . in the literature , there are several different ways to generate single photons @xcite . one way is to use four - wave mixing @xcite . the approach we have investigated in this work is superficially similar to the four - wave mixing method because of the atomic level scheme . one fundamental difference is that the method described here does not require phase matching . in the four - wave mixing method , the phase matching condition needs to be fulfilled to observe the photon which will be emitted in a diffraction - limited solid angle . the approach described in this paper uses an alternative way to generate single photons by preparing a single collective state using rydberg atom blockade in a cavity . the lasers are time dependent in such a way that the preparation , and control are not present at the same time . technologically , rydberg blockade allows one to deterministically control the number of excitations in the cavity and the cross - section for coupling to the cavity is enhanced by the collective nature of the initial state . this addresses some of the problems inherent in trapping single atoms in an optical cavity to generate single photons . the diamond configuration enables one to isolate the quantum state preparation from the single photon output . the output is in a single mode determined by the properties of the optical cavity . a similar atomic level scheme has been studied for collective excitation for raman and superradiant lasers @xcite . in these systems , @xmath0 atoms are excited collectively without considering atom - atom interactions . our model shares the structure of the excitation scheme , but rydberg blockade prepares an @xmath0 atom entangled state with a controlled number of excitations or controlled number of singly excited superatoms . in a more general case , spatial correlations can be written into the sample . this is quite different from these laser schemes as our approach produces a state similar to a micromaser but with deterministic control over the number of excitations in the cavity . the system we described can be useful for studying superradiant behavior because a collective state can be deterministically prepared as the initial state . the cavity coupling can be used to analyze the collective quantum state by detecting the emitted radiation @xcite . most of the models discussed previously for rydberg atoms in a cavity use ground state atoms as an initial state . the jc model for the atoms initially in the ground state in an optical cavity with a single rydberg excitation as an essential piece of the dynamics has been proposed in ref . this model is different from our approach . in our model , we first excite a single collective state by using the rydberg blockade mechanism and then by manipulating a control laser field , generate photons from the cavity . in addition , the preparation lasers are all far detuned from the cavity field frequency . now , if we consider the rydberg excited state as an initial state of the collective atoms - cavity system , then the photon emission efficiency can be increased due to the collectively enhanced cooperativity and can be further improved once the strong coupling regime is achieved due to a negligible transient population in the cavity excited state @xcite . we have neglected several dephasing processes in our analysis because we anticipate these effects are small , or can be made small with a judicious choice of parameters . the effects that are of most concern are the spatial dependence of the atoms - cavity coupling , dependence of the rydberg atom interactions on internuclear separation , motion of the atoms , and interaction between rydberg excitations when there are 2 excitations simultaneously in the cavity . taking these effects into account with the straightforward liouville equations we have used in our calculations is involved , so we estimate their effect on our current results . light pressure forces on the atoms in the cavity are negligible because we have assumed the cavity is on resonance with the @xmath234 transition for times after the atoms are excited to the @xmath235 state , the exception being fig . [ fig:3 ] where we show a detuned cavity to demonstrate that our solution is producing expected results . there is no asymmetry in the absorption of the photons in the cavity under these conditions and the photon recoil is small . the recoil temperature of @xmath108rb is @xmath236nk . additionally , for a large fraction of the time the atoms are interacting with the cavity , there are no photons inside it . we have run our calculations with additional dephasing rates by coupling the singly and doubly excited rydberg states to the dummy state . the results of these calculations with additional dephasing provide an estimate of the effects of dephasing on the collective excitations . in these calculations , we used the same rates for each transition . both processes were modeled with the lindblad operator . we examined additional dephasing rates up to @xmath237khz . the results for the photon production probability are reduced by around @xmath238 for @xmath237khz of dephasing on each transition for the excitation timescales used in this work in the 2 excitation case . the reduction in the overall photon production scales linearly as these dephasing rates are increased . practically , ultracold samples at temperatures @xmath239nk can be generated and used for an experiment to reduce the decoherence resulting from thermal motion , although photon recoil can limit the temperature to @xmath236nk . at temperatures of @xmath240k , the atoms only move @xmath241 nm in @xmath2s . because this is a small fraction of a blockade radius , free evolution of the atoms can be neglected at @xmath240k . the corresponding doppler shifts of the atoms at @xmath240k are @xmath242khz . doppler shifts on this order can lead to reductions in the photon production rate of @xmath243 . the laser spectral bandwidths and the inverse of the times over which the collective states freely evolve are broader than the doppler widths of the atoms that make up the collective excitations . in our case , for very low , but experimentally achievable temperatures , the frozen gas approximation is valid to around @xmath244 in the calculated photon production probability . in general , multiple excitation states within the blockade volume and states consisting of more than 1 superatom will have spatial correlations @xcite because the interaction between the rydberg atoms depends on their internuclear separation and the orientation of the internuclear axis relative to any fields . however , as mentioned earlier in the paper , 2 rydberg atoms can be excited around an avoided crossing where there exists a broad region where the energy has weak dependence on the internuclear separation . these types of potentials can be observed when the multi - level nature of the rydberg atom interactions and external magnetic and electric fields are considered . in fact , the flat regions of the rydberg potentials , so called ` stationary regions ' , are regions where rydberg atom pair excitations occur most readily . the acceleration of the atoms after excitation is determined by the gradient of the interaction potential and has been observed experimentally @xcite . the acceleration on these types of interaction potentials can be negligible as demonstrated in these works - in these experiments , we had to wait @xmath245s to observe a difference between a thermally expanding cloud of cold atoms at @xmath246k and the rydberg atom collision partners @xcite . scaling this time to a @xmath247k temperature gives a time of @xmath248s . similarly , if we assume the atoms constantly accelerate to the final velocity measured in @xcite , @xmath249@xmath250s@xmath251 , in @xmath252s , it takes @xmath253s to reach a thermal velocity that corresponds to @xmath247k . the time over which the 2 rydberg excitations exist in our case is only @xmath240s , so this effect can be negligible with an optimized choice of interacting rydberg state , or use of electric and magnetic fields to engineer the interaction potentials @xcite . recall that the acceleration will cease when the 2 rydberg excitation state is projected onto the non - interacting atom collective state . for interaction potentials where there is significant radial dependence , the efficiency of the 2 excitation dynamics will be reduced and there will be spatial correlations in the state that is excited . changes in the efficiency can depend strongly on the rydberg atom interaction potentials . for the calculations in this paper on 2 excitation states , the collective excitations are not independent of one another but are analogous to a doubly excited rydberg molecule or macrodimer . in this case , the dephasing rates of these entities should be approximately equal to those occurring in a 2 independent excitation case where 2 superatoms are present , provided there are no additional open molecular decay channels . in many cases , the decay time for the macrodimer is determined by the radiative decay of the rydberg atoms that make up the molecule @xcite . ground state - rydberg atom molecules are not important because the internuclear spacings are greater than the size of the rydberg orbital @xcite . the cavity coupling strength also has spatial dependence which must be considered . the size of the collective excitation is very large compared to the wavelength of the cavity transition . the relative sizes can be quantified by comparing the blockade radius for @xmath123 to the wavelength corresponding to the cavity resonance , @xmath254 nm . the cavity coupling strength is effectively averaged over the spatial dependence of the cavity field mode , reducing the average cavity coupling strength by @xmath255 in this simple argument . a rough estimate places half the atoms at nodes and the other half at antinodes . using this approximation to estimate the spread of atomic energy shifts for the atom - cavity coupling , yields a spread of atom energies of @xmath256 . it is interesting to explore the dephasing caused by the interaction of a collective excitation that is spatially larger than the modulation in the cavity coupling in more detail , but such an investigation is outside the scope of the work here . numerically , this can be approached using quantum monte carlo methods which are computationally intensive but possible @xcite . for a collective state spread over a large number cavity resonant wavelengths , the overall effect is primarily to reduce the number of atoms participating in the atoms - cavity dynamics since atoms in the antinodes of the cavity are affected by the cavity coupling while those at nodes essentially behave as if in free - space . the atoms located at antinodes dominate the dynamics because they are coupled more strongly to the cavity . the spatial dependence will lead to fluctuations of the atoms - cavity coupling parameters that will decrease as the number of atoms in the collective state increases . although our calculations are for low atom number collective states , the results scale to higher atom number collective states , at least with respect to the spatial dependence of the cavity mode . the discussion of the dephasing mechanisms here points to a common factor that is important for a broad range of rydberg atom quantum optics experiments . the product of the time the atoms exist in a particular collective state and the dephasing rate due to atomic motion , including heating mechanisms that arise in an experimental scheme , needs to be small . as a consequence , our method requires cold samples , @xmath240k , and fast excitation dynamics , @xmath257s . in addition , to create multiple excitation states , we require potentials that are engineered to reduce the acceleration of atom pairs after excitation and provide a broad range of internuclear separations over which atom pairs can be excited , if spatial correlations are undesirable . speeding up the excitation processes used in our scheme can compensate for some of the losses due to the additional dephasing mechanisms if , as in the case investigated , the rydberg atom interaction strength is large enough to accommodate an increase in the rabi frequencies of the laser pulses . the purcell factors and cooperativities we have used can be justified on the grounds that some experiments have achieved cavities with sufficient finesse and mode volumes to at least approach , or realize , the parameter values ( @xmath106 , @xmath8 , @xmath25 ) used here @xcite . lower cooperativity factors , of course , will reduce the efficiency of the approach that we have presented , however , a smaller cooperativity factor does not preclude the qualitative observation of the proposed scheme to deterministically synthesizing quantum states in a resonant , high - finesse cavity . realizing this goal is interesting and can motivate improvements in the cavity design . to summarize , we showed that rydberg blockade allows one to deterministically prepare a fixed number of collective excitations inside a cavity using our model . this is impossible for the case of direct excitation of atoms inside a cavity , since , for noninteracting atoms , the excitation is probabilistic . in the case of 2 excitation states generated by the blockade mechanism , we can deterministically observe interesting dynamics and analyze the correlations between two , or more , atoms and the photons emitted from the cavity . the 2 excitation states we examined in this paper can be understood as macrodimers excited on flat relatively featureless interatomic potentials that are subsequently coherently transferred to noninteracting , collective excitation states . both the excitation and transfer processes produce collective states as the excitations are shared by the atoms in the sample . using macrodimer states that are not featureless can be used to introduce spatial correlations into the collective excitations . a detailed treatment of the spatial correlations introduced by different interatomic potentials lies outside the scope of this paper , but is a future goal . the engineering of spatial correlations by using rydberg atom interactions in a collective ensemble is interesting and could be useful for experiments in quantum optics , for example in studying superradiance . in this paper , we showed that the collective state dynamics can be controlled deterministically by demonstrating results for collective dynamics that are similar to the tc model for two excited atoms . we have observed anti - bunched photon emission when we neglect spontaneous emission out of the cavity . with spontaneous emission the emitted light is bunched in the weak ( @xmath195 ) and strong ( @xmath196 ) coupling regime . similarly , the entropy of the system decreases with time , while the second order correlation function of the photons emitted by the 2 collective excitations is enhanced . the efficiency of photon production is enhanced by increasing the effective cooperativity by increasing @xmath8 through the cavity parameters or the number of atoms , as would be expected . we have investigated a four - level system of @xmath0 entangled atoms placed inside a high - finesse cavity . the rydberg blockade mechanism plays an important role in preparing the atoms in a collective state . we have used numerical simulations to show that one can deterministically excite collective @xmath0 atom superposition states using rydberg blockade and laser pulses . rydberg blockade allows one to deterministically prepare a fixed number of collective excitations inside a cavity . this is impossible for the case of direct excitation of atoms inside a cavity since that process is probabilistic . in the case of 2 collective excitation states generated by the blockade mechanism , we can deterministically observe bunched photon emission from the cavity similar to the case of 2 excited atoms initially prepared in a cavity . the scheme is useful for the generation of single photons as well as other types of quantum light fields . it could also be useful for coherent optical manipulation in quantum information processing and the study of interesting quantum optical phenomena such as superradiance @xcite . the rates used for the calculations are realistic and experimentally achievable . the most important dephasing mechanisms were included or addressed . this system merits further theoretical and experimental study because it demonstrates flexibility for deterministically synthesizing collective quantum states that can be analyzed by interrogating the photons leaving the cavity . this work was supported by afosr ( fa9550 - 12 - 1 - 0282 , fa9550 - 15 - 1 - 0381 ) and nsf ( phy-1104424 ) . 93ifxundefined [ 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 http://dx.doi.org/10.1017/cbo9780511524530[__ ] ( , ) link:\doibase 10.1103/physreva.74.020701 [ * * , ( ) ] @noop * * , ( ) http://stacks.iop.org/0953-4075/44/i=18/a=184007 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.93.063001 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.99.163601 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.87.037901 [ * * , ( ) ] link:\doibase 10.1038/nphys1178 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.112.043602 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.100.253001 [ * * , ( ) ] link:\doibase 10.1088/1367 - 2630/10/4/045026 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.100.113003 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.112.183002 [ * * , ( ) ] link:\doibase 10.1038/nphys1183 [ * * , ( ) ] link:\doibase 10.1038/nphys2413 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.107.243001 [ * * , ( ) ] link:\doibase 10.1038/nature07127 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.73.565 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/0370-1573(73)90011-2 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.85.2392 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.106.210503 [ * * , ( ) ] http://stacks.iop.org/1367-2630/14/i=5/a=053022 [ * * , ( ) ] link:\doibase 10.1103/physreva.59.1025 [ * * , ( ) ] link:\doibase 10.1103/physreva.78.013808 [ * * , ( ) ] link:\doibase 10.1103/physreva.85.052317 [ * * , ( ) ] link:\doibase 10.1038/nature11023 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.110.083603 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.83.4987 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.90.133602 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.82.3791 [ * * , ( ) ] link:\doibase 10.1126/science.287.5457.1447 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.98.233601 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.106.160501 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.103.083601 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.105.193603 [ * * , ( ) ] link:\doibase 10.1103/physreva.57.3084 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.111.113602 [ * * , ( ) ] link:\doibase 10.1103/physreva.82.053832 [ * * , ( ) ] link:\doibase 10.1209/0295 - 5075/19/1/002 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.71.995 [ * * , ( ) ] link:\doibase 10.1103/physreva.54.4440 [ * * , ( ) ] link:\doibase 10.1103/physreva.87.063805 [ * * , ( ) ] link:\doibase 10.1103/physreva.87.053412 [ * * , ( ) ] link:\doibase 10.1038/451664a [ * * , ( ) ] link:\doibase 10.1103/physrevlett.100.170504 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.82.2313 [ * * , ( ) ] link:\doibase 10.1126/science.1127676 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.91.060401 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.82.1041 [ * * , ( ) ] link:\doibase 10.1103/physreva.78.063830 [ * * , ( ) ] link:\doibase 10.1103/physreva.85.063821 [ * * , ( ) ] link:\doibase 10.1103/physreva.76.013413 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.93.163001 [ * * , ( ) ] link:\doibase 10.1103/physreva.80.033418 [ * * , ( ) ] link:\doibase 10.1103/physreva.72.063403 [ * * , ( ) ] link:\doibase 10.1103/physreva.66.065403 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/s0030-4018(99)00679-3 [ * * , ( ) ] link:\doibase 10.1038/nphys1307 [ * * , ( ) ] link:\doibase 10.1103/physreva.76.011403 [ * * , ( ) ] link:\doibase 10.1103/physreva.79.052504 [ * * , ( ) ] http://dx.doi.org/[__ ] ( , ) link:\doibase 10.1103/physreva.84.023413 [ * * , ( ) ] link:\doibase 10.1103/physreva.87.033416 [ * * , ( ) ] @noop _ _ ( , ) http://dx.doi.org/10.1017/cbo9780511524530[__ ] ( , ) link:\doibase 10.1103/physreva.33.1757 [ * * , ( ) ] hhttp://www.tandfonline.com / doi / abs/10.1080/09500349708231869 [ * * , ( ) ] http://www.tandfonline.com/doi/abs/10.1080/00107511003602990 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.108.023602 [ * * , ( ) ] link:\doibase 10.1103/physreva.82.033412 [ * * , ( ) ] link:\doibase 10.1080/09500340701584076 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/0003-4916(88)90006-1 [ * * , ( ) ] link:\doibase 10.1103/physreva.44.6023 [ * * , ( ) ] http://iopscience.iop.org/1464-4266/6/8/016/cites [ * * , ( ) ] http://link.springer.com/article/10.1007%2fbf01646092#page-1 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.65.3385 [ * * , ( ) ] link:\doibase 10.1103/physreva.59.2468 [ * * , ( ) ] http://www.nature.com/nature/journal/v407/n6803/abs/407491a0.html [ * * , ( ) ] http://dx.doi.org/10.1063/1.3610677 [ * * , ( ) ] link:\doibase 10.1038/nature11361 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.107.103001 [ * * , ( ) ] http://dx.doi.org/10.1038/nature11596 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.113.023006 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.114.203002 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.112.013002 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.104.043002 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.113.123003 [ * * , ( ) ] link:\doibase 10.1038/nature07945 [ * * , ( ) ] link:\doibase 10.1126/science.1260722 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.58.1320 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1103/physrevlett.113.143601 [ * * , ( ) ] \doibase " http://dx.doi.org/10.1016/0370-1573(82)90102-8 " [ * * , ( ) ]

individual atoms in optical cavities can provide an efficient interface between stationary qubits and flying qubits ( photons ) , which is an essentiel building block for quantum communication . furthermore , cavity assisted controlled - not ( cnot ) gates can be used for swapping entanglement to long distances in a quantum repeater setup . nonetheless , dissipation introduced by the cavity during the cnot may increase the experimental difficulty in obtaining long - distance entanglement distribution using these systems . we analyse and compare a number of cavity - based repeater schemes combining various entanglement generation schemes and cavity assisted cnot gates . we find that a scheme , where high - fidelity entanglement is first generated in a two - photon detection scheme and then swapped to long distances using a recently proposed heralded cz - gate exhibits superior performance compared to the other schemes . the heralded gate moves the effect of dissipation from the fidelity to the success probability of the gate thereby enabling high - fidelity entanglement swapping . as a result , high - rate entanglement distribution can be achieved over long distances even for low cooperativities of the atom - cavity systems . this high - fidelity repeater is shown to outperform the other cavity - based schemes by up to two orders of magnitude in the rate for realistic parameters and large distances ( km ) .

1504.03703

distribution of entanglement is an essential task in quantum communication @xcite . entanglement can be used to make highly secure communication channels due to the sensitivity of entangled quantum systems to external influences @xcite . while this sensitivity makes it possible to detect any attack from an eavesdropper , it also makes it hard to distribute entanglement over large distances since any noise from the enviroment quickly destroyes the entanglement . direct transmission of a quantum signal suffers from loss and decoherence from the transmission channel , which results in an exponential decrease of the rate with distance @xcite . to overcome this problem , it has been proposed to use quantum repeaters , where entanglement is first created over short distances by direct transmission and then stored in quantum memories until it can be swapped to larger distances @xcite ( see fig . [ fig : figure1 ] ) . much effort has been devoted to the construction of quantum repeaters based on atomic ensembles , where the large number of atoms , in principle , enables highly efficient quantum memories @xcite . nonetheless , the limited efficiencies demonstrated in current experiments with atomic ensembles @xcite prevents the construction of a practical quantum repeater based on currently existing setups . single emitter systems such as color centers and trapped ions have also been considered for quantum repeaters @xcite . the long coherence times demonstrated with , e.g. trapped ions make them desirable as quantum memories . nonetheless , entanglement needs to be created non - locally between two memories in the initial step of a repeater . this requires efficient transfer of information from the quantum memories onto light in the form of single photons . to this end , it is an advantage to place the emitter inside a cavity , which can greatly enhance the light - emitter coupling @xcite . entanglement swapping can then be performed with a cavity mediated cnot gate @xcite but in this case , the detrimental effect of cavity loss and spontaneous emission from the emitters may prevent obtaining efficient entanglement swapping . the parameter characterizing the effect of dissipation in the emitter - cavity system is the cooperativity , @xmath1 . it has been argued that a direct implementation of gates in a cavity will make the gate fidelity , @xmath2 , have a poor scaling of @xmath3 @xcite . to overcome this problem for current cavities with limited @xmath1 , it has been suggested to employ entanglement purification after each swap operation to boost the entanglement but this either requires a large number of resources or a time consuming sequential generation of purification pairs @xcite . here we analyze and compare a number of cavity - based quantum repeaters which combines various proposals for entanglement generation and cavity - assisted cnot gates . we find that the best scheme is where high - fidelity entanglement is generated using a two - photon detection scheme similar to ref . @xcite and swapped to large distances using the heralded cz - gate proposed in ref . the heralded gate enables nearly perfect entanglement swapping when successful allowing for many swaps without the need of entanglement purification . as a result , high - rate entanglement distribution is achieved even for low cooperativities . compared to the other cavity - based repeaters , this high - fidelity repeater achieves up to two orders of magnitude higher secret key rate ( see below ) for realistic parameters and large distances ( @xmath0 km ) . specifically , we have compared to repeaters where entanglement is generated using a single - photon detection scheme similar to ref . @xcite , which allows for a better rate at the expense of fidelity . furthermore we have considered schemes where entanglement swapping is achieved using the deterministic cnot gate suggested in ref . @xcite , combining it with the local entanglement generation scheme of ref . the advantage of this gate is that the fidelity scales as @xmath4 , which is a significant improvement of the @xmath5 scaling characterizing the performance of a direct implementation of gates in a cavity . as a result , long - distance entanglement distribution can also be achieved with this gate but it requires cooperativities above @xmath6 , which might be challenging to achieve experimentally . furthermore , we include the possibility of initial purification in repeaters based on the single - photon detection scheme in order to allow for the higher rate of this scheme to compensate for the lower fidelity compared to the two - photon detection scheme . to reflect a realistic near - term approach to quantum repeaters , we only consider scenarios with 2 or 4 qubits per repeater station . for the same reason , we also do not consider the possibility of intermediate entanglement purification . here , initial purification refers to purification in the elementary links ( see fig . [ fig : figure1 ] ) while intermediate purification refers to purification in the subsequent stages of a repeater . we have numerically optimized all the considered repeater schemes for a range of cooperativities and distances to find the highest achievable secret key rate ( see below ) . note that a similar optimization of repeater schemes based on dynamical programming was described in ref . @xcite . in that work , both initial and intermediate entanglement purification was considered assuming high - fidelity operations . our optimization is less detailed since we do not consider intermediate purification . on the other hand , we include how the errors of the operations depend on the physical parameters such as the cooperativity and investigate concrete physical implementations . finally , we compare the high - fidelity repeater considered here to both an ion - trap repeater and one of the best repeaters based on atomic ensembles @xcite . for a distance of 1000 km , the high - fidelity repeater outperforms both of these schemes for @xmath7 . we will first describe the details of the high - fidelity quantum repeater , which we find to have the best performance and later discuss and compare with the various other schemes . the first step in any quantum repeater is to create non - local entanglement in the elementary links ( see fig . [ fig : figure1 ] ) . connected by repeater stations pictured as cavities containing single emitters . after creating entanglement in the elementary links the entanglement is swapped to larger distances by combining the elementary links . the numbers to the right in the figure refers to the swap level of the repeater . in the first swap level , the four elementary links are connected pairwise to make two longer links . in the second swap levels these two links are connected to create entanglement over the total distance . the total number of swap levels is thus 2 for this depicted setup . , width=288,height=144 ] to this end , a two - photon detection scheme , as proposed in ref . @xcite , is considered . the basic setup is shown in fig . [ fig : figure2](a ) . denotes the cavity coupling . for the two - photon scheme the levels @xmath8 and @xmath9 are assumed to have equal coupling of @xmath10 to the excited state @xmath11.,scaledwidth=50.0% ] both emitters are initially prepared in the excited state @xmath11 by a strong excitation pulse and the cavity is assumed to couple both the @xmath12 and @xmath13 transitions with equal coupling strength @xmath10 ( see fig . [ fig : figure2](b ) ) . the two transitions are , however , assumed to produce photons with different polarizations such that the emission of a cavity photon creates an entangled state between the photon and the emitter of the form @xmath14 where @xmath15 ( @xmath16 ) is the single photon state with polarization 1 ( 2 ) . the probability of one of the emitters to emit a photon of either polarization through the cavity , into an optical fiber transmitting it to the detection stage during a time interval @xmath17 $ ] is @xmath18 assuming perfect outcoupling to the fiber and that the decay rate of the cavity , @xmath19 , is much larger than the cavity coupling , @xmath20 . we have here introduced the cooperativity @xmath21 , where @xmath22 is the spontaneous emission rate of the emitters into modes other than the cavity . this is the key parameter characterizing the performance of the cavity - based repeaters . the photons are sent from the cavities to a central polarizing beam splitter ( pbs ) . if two photons of the same polarizations are incident on the pbs , they leave in different output ports , while photons of different polarization leave in the same output port . the outputs are then sent to a second set of polarizing beam splitters and all four outputs of these are finally measured with single photon detectors ( spd ) . a click in a detector in each arm heralds the creation of the bell state @xmath23 between the emitters up to a local qubit rotation . neglecting dark counts of the detectors , the heralded fidelity is unity ( see app . [ two ] ) while the success probability of the scheme is @xmath24 with @xmath25 being the total detection efficiency including inefficient outcoupling of the cavity light , losses in the transmission fibers and imperfect detectors . compared to schemes based on single - photon detection ( see below ) the rate of this two - photon detection scheme decreases rapidly with decreasing @xmath25 . on the other hand it has a high fidelity , which is desirable for the subsequent stages of entanglement swapping as we will show below . for entanglement swapping we find that the best performance is achieved using the heralded cz - gate described in ref . the gate was described in detail for @xmath26rb atoms in ref . @xcite but it can be easily generalized to any set of emitters , which have the appropriate level structure ( see fig . [ fig : figure3 ] ) . note that the gate operation relies on only qubit state @xmath9 coupling to the cavity while the states @xmath8 and @xmath9 had equal cavity couplings in the entanglement generation scheme . to achieve this change in couplings , the state @xmath8 should be mapped to another level in between the entanglement generation and the gate operation . for a realization with alkali atoms where the qubit states would be realized in the hyperfine grounds states , this could be achieved by applying a magnetic field to resolve the hyperfine states and applying a microwave pulse resonant with only the @xmath8 state . in the heralded gate , the cavity is assumed to contain two qubit atoms and one auxiliary atom to facilitate the gate . the auxiliary atom is initialized in a state @xmath27 that does not couple to the cavity and it would therefore not interfere with the entanglement generation scheme . by addressing the auxiliary atom with a weak laser pulse , an ac stark shift is introduced , which gives a phase that depends on the state of the qubit atoms . together with single qubit rotations , this enables a cz - gate between the two qubit atoms . furthermore , the auxiliary atom can function as an error detector in the sense that any cavity decay or spontaneous emission changes the state of the atom . performing a heralding measurement of the state of the auxiliary atom at the end of the driving pulse removes all dissipative errors . as a consequence , the gate gets limited only by non - adiabatic effects . as shown in ref . @xcite , a heralded error below @xmath28 is possible with a gate time of @xmath29 , where @xmath22 is the atomic linewidth . the failure probability of the gate scales as @xmath5 and the high fidelity thus comes at the cost of a finite but possibly low failure probability . a cz gate combined with single qubit rotations is sufficient to perform direct entanglement swapping . for simplicity , we assume perfect single qubit rotations and 100% efficient measurement of atomic states for all schemes considered . relaxing this assumption will in general decrease the rate of all the considered repeater schemes but schemes with a high number of swap levels like the high - fidelity repeater will be influenced more on the rate than schemes with a low number of swap levels . the advantage of the high - fidelity repeater can be understood by considering the requirement for reaching a certain threshold fidelity , @xmath30 of the distributed pair . in this case , the maximum number of swap levels is @xmath31 , where @xmath32 are the errors of the initial entanglement generation and the entanglement swapping respectively . the combination of the high - fidelity two - photon detection scheme and the heralded gate thus makes it possible to have a repeater with many elementary links while maintaining a high fidelity of the final distributed pair even for low cooperativities since the error of the heralded gate is still high in this regime . we imagine that the distributed entanglement is used to generate a secret key between two parties referred to as alice and bob . there exist various quantum key distribution schemes @xcite , however , the general idea is that alice and bob can exclude that an eavesdropper has any information about the key by measuring their qubits and compairing results . we will assume that a six - state version of the bb84 protocol described in ref . @xcite is used to generate the secret key . this protocol consists of three main steps . first alice and bob picks a basis according to some probability distribution and measure the state of their qubits thereby producing two bit strings referred to as the _ raw key_. afterwards they compare their choice of basis and only keep the bits where they chose the same measurement basis thereby producing a _ sifted key_. finally , alice and bob estimate the information that some eavesdropper could possibly have obtained about their key and perform privacy amplification @xcite . if the errors are not too big , they can obtain a shorter but completely secure key . for the six - state protocol , the secret key rate , @xmath33 can be defined as @xmath34 where @xmath35 is the distribution rate of the entangled pairs , @xmath36 is the probability that alice and bob choose the same measurement basis and @xmath37 is the secret key fraction , which depends on the fidelity of the distributed pairs . we assume a worst case scenario where the distributed pairs are werner states of the form @xmath38 for such states , it is shown in ref . @xcite that the secret key fraction in the six - state protocol can be estimated in the limit of infinitely long raw keys to be @xmath39 where @xmath40 and @xmath41 is the binary entropy . is valid in the limit of perfect sifting and privacy amplification , which we assume to be the case . furthermore , we assume an asymmetric version of the six state protocol , where one basis is used almost all the time such that @xmath42 @xcite . fig . [ fig : figure5 ] shows how the secret key fraction depends on the fidelity of the distributed pairs . ) as a function of the infidelity , @xmath43 , of the final entangled pair . for @xmath44 it is no longer possible to extract a secret key from the raw keys.,scaledwidth=45.0% ] as shown in the figure , high - fidelity pairs are required in order to have a non - vanishing secret key fraction . again this points to the high - fidelity two - photon detection scheme and the nearly error - free heralded entanglement swapping as the best choice for the repeater . the main goal of the quantum repeater is to overcome the effect of fiber losses . we model the fiber losses with a transmission efficiency @xmath45 , where @xmath46 is the length of the elementary links of the repeater and @xmath47 is the fiber attenuation length . @xmath48 enters in the total detection efficiency @xmath25 as described above . for a given resource of @xmath49 repeater stations , one can either use all stations in a single repeater with @xmath50 swap levels or one can construct a number of parallel , independently operated chains of repeaters with less swap levels . increasing the number of swap levels , decreases the fiber losses in the elementary links and thus increases the rate of entanglement generation . if , however , the length of the elementary links is already small , such that , e.g. imperfect spd dominates the rate , then increasing the number of swap levels does not lead to any improvement . in this case it is advantageous to use the extra repeater stations to make another repeater with less swap levels , which runs in parallel with the already existing one . to make a proper assessment of the performance of repeater one should therefore include that adding swap levels costs resources in the form of additional repeater stations . in our comparison , we therefore consider a normalized secret key rate , @xmath51/ ( # of stations ) , which is the secret key rate divided by the total number of repeater station instead of the bare secret key rate . to evaluate the performance of the repeater we calculate the achievable rate @xmath52 as described in appendix [ app : rate ] with the assumptions summarized in table [ tab : parameter ] about fiber losses etc .. the resulting rate for various swap level used in the repeater is shown in fig . [ fig : figurex4 ] as a function of distance . as seen in the figure , the optimal number of swap level changes with distance while considering the normalized secret key rate . the rate was calculated as described in app . [ app : rate ] for a cooperativity of 100 with the assumptions summarized in table [ tab : parameter ] about fiber losses , detection efficiencies etc . for distances @xmath53 km , only a single swap level is needed since the fiber losses are relatively small while more swap levels are needed as the distance increases . ) as a function of the distribution distance for a high - fidelity repeater consisting of the two - photon entanglement generation scheme and the heralded gate for entanglement swapping . we have considered @xmath54 and 4 swap levels and have assumed a cooperativity of @xmath55100 and two qubits per repeater station . the secret key rate was calculated as described in app . [ app : rate ] with the assumptions summarized in table [ tab : parameter ] . , scaledwidth=48.0% ] in most repeater schemes , the qubits in each repeater station are assumed to be operated simultaneously with half of the qubits being used to generate entanglement with the neighbouring station to each side such that entanglement attempts in all the elementary links are done simultaneously . we will refer to this as a _ parallel _ repeater . we will , however , also consider another sequential way of operating the qubits , where all qubits in a station are first used to make entanglement in one elementary link . after this has been obtained , all but one qubit are then used to make entanglement over the neighboring link in the opposite direction . this is referred to as a _ sequential _ repeater . the advantage of the sequential repeater is that the rate of the lowest level in the repeater , the entanglement generation , is increased . this comes at the cost of a waiting time between entanglement attempts in neighboring links . as the number of qubits per repeater station increases the sequential repeater will start to outperform the parallel repeater . we find that this happens with 4 qubits per repeater station ( see sec . [ sec : optim ] ) . we have found that the high - fidelity repeater that we have described above outperforms a number of other cavity - based repeater schemes , which can be contructed with different schemes of entanglement generation and cnot gates . below , we describe the constituents of these other schemes and compare them to those of the high - fidelity repeater it has also been suggested to use single - photon detection schemes similar to ref . @xcite to generate entanglement in the elementary links . the setup of a single - photon detection scheme is also shown in fig . [ fig : figure2 ] . we assume that the two emitters are initially prepared in a state @xmath56 by a weak excitation pulse such that the excitation probability is @xmath57 . an emitter can then go from state @xmath11 to state @xmath9 by emitting a cavity photon . the emitted photons are collected from the cavities and combined on a balanced beam splitter ( bs ) on a central station between the two cavities . neglecting losses , the detection of a single photon after the bs will project the state of the emitters into the bell state @xmath58 up to a single qubit rotation in the limit @xmath59 where we can neglect the possibility of double excitations . the probability of an emitter to go from @xmath11 to @xmath9 , by emission of a cavity photon during a time interval @xmath17 $ ] , is @xmath60 ( see eq . ) under similar assumptions as in the two - photon scheme described above . neglecting dark counts but including losses , the total probability of a single click at the central station is @xmath61 with @xmath25 being the total detection efficiency as for the two - photon scheme . the first term is the probability to emit and detect a single photon while the second term is the probability of emitting two cavity photons but only getting a single click ( we assume that we do not have access to number - resolving detectors ) . the probability , to have a single click and have created the state @xmath62 , is @xmath63 . the average heralded fidelity conditioned on a single click is thus @xmath64 . to lowest order in @xmath65 , @xmath66 while the success probability is @xmath67 . there is thus a tradeoff set by @xmath57 between the success probability and the fidelity for the single - photon detection scheme . this is in contrast to the two - photon detection scheme where @xmath68 regardless of success probability . the success probability of the single - photon detection scheme is not as sensitive to the detection efficiency @xmath25 as the two - photon detection scheme as shown in fig . [ fig : figurex3 ] . if the detection efficiency @xmath25 is large , the two - photon scheme is desirable since it will have both a high success probability and a high fidelity . however , if @xmath25 is small , the single - photon scheme will be advantageous since it has a relatively high success probability . due to the possible high success probability but limited fidelity of the single - photon scheme , it might be desirable to combine it with entanglement purification to increase the final fidelity . in this way the higher success probability of the single - photon scheme may compensate the lower fidelity . we have therefore considered the possibility of initial entanglement purification in repeaters based on the single - photon detection scheme as described below . based on a detailed analysis of the various errors that limit the fidelity for the single - photon scheme including dark counts of the detectors ( see app . [ single ] for details ) , we find that the purification protocol of ref . @xcite effectively corrects for the errors in the single - photon scheme and we assume that this is used for the initial purification . however , as pointed out in ref . @xcite an improved fidelity , at the expense of a factor of @xmath69 in the success probability , can be obtained by only accepting outcomes where the two heralding qubits are found in state @xmath70 instead of also accepting @xmath71 outcomes . we will also consider this modification to the purification protocol in ref . the protocol relies on a cnot operation , which we assume to be made with the same gate used to perform the subsequent entanglement swapping ( see below ) . to reflect the most realistic near - term quantum repeaters we consider at most 4 qubits per repeater stations . we therefore assume that the purification is performed in a pumping scheme @xcite , where the fidelity of a single pair is pumped by combining it with pairs of lower fidelity since this requires the lowest number of qubits per station . the effect of combining the single - photon scheme with initial purification is shown in fig . [ fig : figurex3 ] where , for simplicity , the purification is assumed to be performed with a deterministic gate with perfect fidelity and without the modification of ref . if high fidelity pairs are desired for , e.g. , a repeater with many swap levels , entanglement purification can increase the rate of the entanglement generation . for high collection efficiencies it is , however , desirable to use the two photon scheme since this has a higher rate . in particular the two photon scheme becomes desirable if high fidelity pairs are required . and @xmath72 both without purification ( @xmath73 ) and with one round of purification ( @xmath74 ) . the rate is shown as a function of the total detection efficiency @xmath25 . we have neglected dark counts and assumed that the cnot gate is deterministic and have perfect fidelity . the rate has been calculated as described in app . [ app : rate ] . furthermore , we have assumed that each repeater station contains 4 qubits , which are either used for purification or to increase the rate of the entanglement generation . , scaledwidth=50.0% ] in our analysis , both the initial purification and the subsequent entanglement swapping involves a cavity - based cnot gate . besides the heralded cnot gate used in the high - fidelity repeater , which we will refer to as _ gate 1 _ , a deterministic cavity - based gate proposed in ref . @xcite could be used . combining the gate scheme of ref . @xcite with the local entanglement generation scheme of ref . @xcite results in a deterministic cnot gate with an error scaling as @xmath75 . we will refer to this gate as _ gate 2_. this gate does not require an auxiliary atom as gate 1 but rather two auxiliary levels in the qubit atoms as shown in fig . [ fig : figure4a ] . in this scheme , the quantum information is stored in the horizontal / qubit degree of freedom ( subscripts @xmath76 and @xmath77 ) and the vertical / level degree of freedom ( denoted @xmath20 and @xmath78 ) is used to make an entanglement assisted cnot gate between the atoms . separating the qubit degree of freedom from the level degree of freedom , the gate works by ideally making the transformation @xmath79 where @xmath80 denote the vertical states and @xmath81 ( @xmath82 ) is the qubit state of the first ( second ) qubit , which could be entangled with atoms at neighboring repeater stations . the entanglement between the levels @xmath27 and @xmath83 can be used to make a cnot gate , if the levels of the atoms can be measured non - destructively , i.e. without revealing any information about the qubit state as described in ref . @xcite . both the transformation shown in eq . and the non - destructive measurements can be obtained by sending a weak coherent pulse onto a two - sided cavity and detecting any reflected light ( see fig . [ fig : figure4b ] and app . [ app : cnot ] ) . if the light is resonant with the empty cavity mode , atoms in @xmath83 will shift the cavity resonance . consequently , photons will be reflected and constitutes a qnd measurement of the presence of atoms in @xmath83 . spontaneous emission from the atoms will limit the fidelity of the gate to @xmath84 , where @xmath85 is the detection efficiency and @xmath21 is the cooperativity . the gate time is limited by the time of the single qubit rotations and the coherent pulses . we assume that this gives a gate time on the order of 10 @xmath86s . as a benchmark , we also consider a naive approach where a direct gate between two qubits is made in a cavity without the use of an auxliary atom or auxliary atomic states . to characterize such a gate we consider a situation where the setup of gate 1 is used to make a deterministic gate by simply ignoring the heralding condition . we will refer to this gate as _ gate 3_. for such a gate , we find that the gate fidelity will scale as @xmath87 and the time of the gate will be limited by the time of the single qubit rotations which we assume to be @xmath88 @xmath86s . the characteristics of the three gates we consider are summarized in table [ tab : table2 ] and illustrated in fig . [ fig : figurex2 ] . .the characteristics of the three gates considered for the cavity - based repeaters . @xmath1 is the cooperativity of the atom - cavity system and @xmath85 is the single photon detection efficiency in gate 2 . the time of the single qubit rotations is assumed to be 10 @xmath86s . [ cols="^,^,^,^",options="header " , ] the free parameters in the optimizations are the number of swap levels , the number of purifications with / without the modification of ref . @xcite and whether a parallel or sequential repeater protocol is used . in the optimizations , we calculate the secret key rate on a grid of all these parameters and pick the combination giving the highest rate . [ fig : figurex6 ] shows a specific example where the combination of the single - photon scheme with gate 2 is investigated for a parallel repeater and a cooperativity of 100 . ) as a function of the distribution distance ( distance ) for a parallel repeater based on the single photon generation scheme and gate 2 . the cooperativity was assumed to be 100 and we assumed 4 qubits per repeater station . the optimal number of swap levels ( @xmath50 ) and purification rounds ( @xmath89 ) for a given distance can be directly read off from the plot as the combination giving the highest rate . note that because the gate fidelity is limited , curves with @xmath90 and @xmath91 are not shown since they result in a much lower secret key rate . the purification schemes was considered to be without the modification of ref . @xcite.,scaledwidth=46.0% ] the number of swap levels and purifications , giving the highest rate for a specific distance , can be directly read off from the figure . the same calculations are then done for a sequential repeater protocol and compared to the parallel repeater protocol with / without the modified purification in order to find the highest rate for this specific combination of entanglement generation scheme and cnot gate . this is done for all combinations of entanglement generation schemes and cnot gates . the optimal evolution time , @xmath92 , and excitation probability , @xmath65 , in the entanglement generation schemes are found for each grid point using a built - in numerical optimization in the program matlab . the key parameter , determining the performance of the cnot gates , is the cooperativity ( see table [ tab : table2 ] ) . we therefore optimize for cooperativities @xmath93 $ ] and distances between 100 km and 1000 km . finally , the optimizations are performed for both 2 qubits per repeater station and 4 qubits per repeater station . note that the auxiliary atom used in gate 1 is not counted as a qubit and schemes based on this gate thus in principle contain an additional atom per repeater station . we model the effect of the non - perfect gates , as depolarizing channels such that the output of a gate operation described by a unitary @xmath94 working on a set @xmath95 of two qubits is @xmath96 where @xmath97 is the fidelity of the gate , @xmath98 is the identity matrix of the set , @xmath99 is the trace over the set and @xmath100 is the initial density matrix describing the system before the gate operation . we use eq . to propagate the density matrix from the entanglement generation ( see app . [ single]-[two ] ) through the steps of initial purification and entanglement swapping and calculate the average fidelity of the distributed pairs . to calculate the secret key fraction , we treat the distributed pairs as werner states as described in sec . [ sec : secret ] . [ h ] + [ h ] + the secret key rates per repeater station of the high - fidelity repeater and the other cavity - based repeaters are shown in fig . [ fig : figurex7 ] for distances @xmath101 $ ] km and a cooperativity of 100 and in fig . [ fig : figurex8 ] for a distance of 1000 km and cooperativities in the interval @xmath102 $ ] . as shown in fig . [ fig : figurex7 ] , the repeaters based on gate 3 are simply not able to distribute entanglement over large distances for realistic cooperativities . as a consequence , repeaters based on gate 3 do not appear on fig . [ fig : figurex8 ] since their secret key rate is simply too low . in general , the high - fidelity repeater ( 2-photon , gate 1 ) achieves the highest secret key rate for a broad range of cooperativities and long distances @xmath103 km . this reflects both that this protocol allows for a higher number of swap levels and that the secret key rate favors the distribution of high - fidelity pairs since these gives the highest secret fraction ( see fig . [ fig : figure5 ] ) . it is also apparent from fig . [ fig : figurex8 ] that while repeaters based on gate 2 need cooperativities above 100 for a distance of 1000 km , repeaters based on gate 1 are able to function with much lower cooperativities around @xmath104 . this is because the heralded gate has nearly unit fidelity independent of the cooperativity . for high cooperativities and/or low distances , a repeater based on the two - photon detection scheme and gate 2 can give a slightly higher secret key rate than the high - fidelity repeater . this improvement is , however , less than a factor of 2 in the secret key rate . the steps in the rates of the schemes based on gate 2 in fig . [ fig : figurex8 ] originate from the fact that as the cooperativity increases , the fidelity of gate 2 increases and at some point the fidelity is high enough to allow for another swap level , which makes the rate increase abruptly . from the optimizations , we find that the sequential repeater architecture achieves slightly higher rates ( less than a factor 2 ) than the parallel repeater architecture for 4 qubits per repeater station , while the opposite is the case for 2 qubits per repeater station . in general , repeaters based on the two - photon detection schemes outperform repeaters based on the single - photon detection scheme except for repeaters based on gate 3 . this reflects that repeaters based on gate 3 can not perform many swap levels since the fidelity simply decreases too rapidly with the number of swap levels . the result of the optimization was that no swap levels were actually preferred for repeaters based on gate 3 for @xmath105 . as a result , the elementary links in these repeaters are long and fiber losses therefore significantly decrease the total detection probability @xmath25 in the entanglement generation schemes . in the limit of very low @xmath25 , the one - photon scheme is advantageous since the success probability only depend linearly on @xmath25 . for the optimizations , we have assumed that the combined efficiency of the spd detectors and outcoupling of light from the cavities is @xmath10650% . if this efficiency is smaller , repeaters based on single - photon detection may be desirable . we also find that the purification protocol , in general , performs better with than without the modification of ref . the improvement is , however , limited to a factor of @xmath107 for the parameters considered in figs . [ fig : figurex7]-[fig : figurex8 ] . it is important to stress that the rates plotted in figs . [ fig : figurex7]-[fig : figurex8 ] , are the secret key rates divided by the total number of repeater stations . the actual distribution rate can thus be obtained by multiplying with the number of repeater stations . for the high - fidelity repeater , we find a secret key rate of @xmath10816 hz over 1000 km for 33 repeater stations and a cooperativity of 1000 assuming 2 qubits per repeater station . for a more modest cooperativity of 100 , a secret key rate of @xmath1081.5 hz over 1000 km can be obtained . we can compare the rate found here to the rate obtainable with repeaters based on atomic ensembles . in ref . @xcite an efficient repeater based on atomic ensembles is described , which achieves one of the highest distribution rates for repeaters based on atomic ensembles @xcite . the fidelity of the distributed pair and the distribution rate are derived in ref . @xcite for a repeater with four swap levels corresponding to 17 repeater stations . based on this , we have calculated the secret key rate assuming an optimistic , basic repetition rate of the ensembles of 100 mhz and memory and spd efficiencies of 90% . the rate of the ensemble - based repeater is also shown in figs . [ fig : figurex7]-[fig : figurex8 ] for similar assumptions about fiber losses etc . as for the cavity - based repeaters . we have assumed that the repeater uses four swap levels for all distances even though a smaller number of swap levels may be desirable for smaller distances ( @xmath109 km ) @xcite . for a distance of 1000 km , we find a rate of @xmath110 hz for 33 repeater stations . this shows that repeaters based on individual atoms in cavities may be very promising candidates for realizing efficient quantum repeaters with rates exceeding those obtainable with atomic ensembles . the main reason for this is that very efficient entanglement swapping can be realized in the cavity - based repeaters which greatly enhances the distribution rates for long distances . on the contrary , repeaters based on atomic ensembles and linear optics have an upper limit on the swapping efficiency of 50% . for comparison we have also considered a repeater based on ion traps where there is no cavity to collect the light . non - local entanglement can still be created by collecting the emitted light with a lens as demonstrated in ref . @xcite where a collection efficiency of 10% was reported . the entanglement swapping can be realized using a gate , which has been demonstrated experimentally with a fidelity of 99.3% and a gate time of @xmath111 @xmath86s @xcite . note that this fidelity was measured for the generation of a single state in ref . @xcite but we will assume it to be the fidelity of the entanglement swap . the rate of such a ion - trap repeater is shown in figs . [ fig : figurex7]-[fig : figurex8 ] with assumptions about fiber losses etc . summarized in tab . [ tab : parameter ] . we have assumed a collection efficiency of 10% and as a result , the one photon scheme with modified purification performs better than the two - photon scheme for 4 qubits per repeater station . however for two qubits , where purification is not possible , the two - photon scheme is in general advantageous except for small distances ( @xmath112 200 km ) . we have assumed a gate fidelity of 99.3% and have plotted the highest rate obtainable with either the one - photon or two - photon scheme . it is seen that the high - fidelity repeater outperforms the ion - trap repeater for @xmath7 , which is mainly due to the low collection efficiency in the entanglement generation . the ultimate rate obtainable with a repeater with perfect deterministic entanglement swapping and entanglement generation based on the two - photon detection scheme with a collection efficiency set by @xmath113 is also shown in figs . [ fig : figurex7]-[fig : figurex8 ] under our assumptions about fiber loss , detection efficiency etc .. a similar repeater was considered in ref . @xcite to demonstrate the feasibility of repeaters based on trapped ions . for @xmath114 , the high - fidelity repeater achieves only a factor of @xmath69 slower rate than this ultimate limit for a distance of 1000 km . in conclusion , we have performed a detailed analysis of quantum repeaters based on individual emitters in optical cavities . we have found that a high - fidelity repeater based on the heralded gate described in ref . @xcite combined with a two - photon detection scheme is the best option over a large parameter regime and enables high secret key rates over large distances even for limited cooperativities @xmath115 . compared with a number of other cavity based repeaters it achieves rates , which are up to two orders of magnitude faster for long distances ( 1000 km ) and cooperativities @xmath115 . for small distances or higher cooperativities , a repeater based on the deterministic cnot gate described in ref . @xcite combined with a two - photon detection scheme can achieve rates which are slightly higher than the high - fidelity repeater but the improvement is less than a factor of 2 . we have also compared the high - fidelity repeater to the repeater in ref . @xcite , which is based on atomic ensembles . for a distance of 1000 km and @xmath116 the high - fidelity repeater begins to outperform the ensemble - based repeater and an improvement of more than two orders of magnitude in the secret key rate is possible for @xmath117 . the main reason for the advantage of the high - fidelity repeater is that entanglement can be swapped very efficiently using the heralded cnot gate described in ref . consequently , the number of swap levels in the repeater can be increased without the need of intermediate purification which greatly enhances the rate for large distances . a similar advantage could in principle be achieved by resorting to a trapped ion system where efficient gates can be implemented . for current systems , the collection efficiencies are , however , so low that a trapped ion system could be outperformed by a cavity system with a limited finesse of @xmath118 . if the collection efficiency could be overcome , e.g. by placing the ions in a cavity with a high cooperativity , the rate can be substantially improved , but with @xmath119 the high fidelity repeater investigated here is within a factor of two of this ideal repeater . it should , however , be noted that we have compared schemes with strong physical differences in our analysis . the high - fidelity repeater requires an extra auxiliary atom while auxiliary atomic levels are required to decrease the error of the deterministic cavity - based cnot gates . the ensemble - based and ion - trap repeaters are also very different physical systems compared to the cavity - based repeaters with individual atoms . the different experimental difficulties in realizing the physical requirements for the various repeater schemes should be included in a more advanced assessment . finally , we note that while we have investigated a number of different possible repeater protocols there may be even more advantages procedures . hence the results that we have derived here should be seen as lower limits to the achievable communication rates . a particular interesting possibility could be to investigate proposals along the lines of ref . @xcite , which also rely on heralding measurements to detect errors during entanglement generation and two qubit operations . possibly some of the ideas from these schemes could be used to improve the communication beyond what we have found here . we gratefully acknowledge the support from the lundbeck foundation , the carlsberg foundation , nsf , cua , darpa , afosr muri , and arl . the research leading to these results has also received funding from the european research council under the european union s seventh framework programme ( fp/2007 - 2003 ) through siqs ( grant no . 600645 ) and erc grant qios ( grant no . the setup of the single - photon scheme is described in sec . [ sec:1phot ] . the single photon detectors are assumed to have a dark count probability of @xmath120 and an efficiency of @xmath85 while the transmission efficiency of the fibers is denoted @xmath48 . as described in sec . [ sec : generation ] , the probability of an emitter to go from the excited state , @xmath11 to the ground state @xmath9 is @xmath60 while the excitation probability is @xmath57 . the scheme is conditioned on a single click at the central station . depending on which detector gave the click , a single qubit rotation can be employed such that ideally the state @xmath62 is created . going through all the possibilities of obtaining a single click at the central station , we find that the density matrix following a single click , and possible subsequent single qubit rotations , is @xmath121 with coefficients @xmath122 \nonumber \\ \alpha_{1}&=&\frac{1}{p_{\text{1click}}}\big[\frac{1}{2}\epsilon^{2}(1-p_{\text{phot}})^{2}p_{\text{dark}}(1-p_{\text{dark}})\big ] \nonumber \\ \beta_{1}&=&\frac{1}{p_{\text{1click}}}\big[\frac{1}{2}\eta_{\text{d}}\eta_{\text{f}}p_{\text{phot}}\epsilon^{4}(1-p_{\text{phot}})(1-p_{\text{dark}})\nonumber \\ & & + 2(1-\eta_{\text{f}})p_{\text{phot}}\epsilon^{2}(1-\epsilon^{2})p_{\text{dark}}(1-p_{\text{dark } } ) \nonumber \\ & & + ( 1-\eta_{\text{d}}\eta_{\text{f}})p_{\text{phot}}\epsilon^{4}(1-p_{\text{phot}})p_{\text{dark}}(1-p_{\text{dark}})\nonumber \\ & & + ( 1-\epsilon^{2})\epsilon^{2}(1-p_{\text{phot}})p_{\text{dark}}(1-p_{\text{dark } } ) \nonumber \\ & & + \frac{1}{2}\epsilon^{2}(1-p_{\text{phot}})^{2}p_{\text{dark}}(1-p_{\text{dark } } ) \nonumber \\ & & + 2\eta_{\text{f}}(1-\eta_{\text{d}})p_{\text{phot}}\epsilon^{2}(1-\epsilon^{2})p_{\text{dark}}(1-p_{\text{dark}})\big ] \nonumber \\ \tilde{\alpha}_{1}&=&\frac{1}{p_{\text{1click}}}\big[2(1-\epsilon^{2})^{2}p_{\text{dark}}(1-p_{\text{dark}})\nonumber \\ & & + 2(1-\epsilon^{2})\epsilon^{2}(1-p_{\text{phot}})p_{\text{dark}}(1-p_{\text{dark}})\big ] \nonumber \\ \tilde{\beta}_{1}&=&\frac{1}{p_{\text{1click}}}\big[\eta_{\text{d}}\eta_{\text{f}}p_{\text{phot}}\epsilon^{4}(1-p_{\text{phot}})(1-p_{\text{dark}})\nonumber \\ & & + 2(1-\eta_{\text{d}}\eta_{\text{f}})p_{\text{phot}}\epsilon^{4}(1-p_{\text{phot}})p_{\text{dark}}(1-p_{\text{dark } } ) \nonumber \\ & & + 2(1-\eta_{\text{d}}\eta_{\text{f}})^{2}p_{\text{phot}}^{2}\epsilon^{4}p_{\text{dark}}(1-p_{\text{dark}})\nonumber \\ & & + 2(1-\eta_{\text{d}}\eta_{\text{f}})\eta_{\text{d}}\eta_{\text{f}}p_{\text{phot}}^{2}\epsilon^{4}(1-p_{\text{dark}})^{2}\big ] . \end{aligned}\ ] ] here we have assumed that with probability @xmath123 , an emitter is excited but spontaneously decay to the ground states instead of emitting a cavity photon . furthermore , we have assumed that the decay rates to the two ground states are equal such that the emitter ends up in @xmath124 . note that the detectors are not assumed to be number resolving . @xmath125 is the total success probability given by @xmath126 assuming @xmath127 , the dominant error is where both qubits are excited but only a single click is detected at the central station . this leaves the qubits in the state @xmath128 and this error is efficiently detected by the purification scheme described in ref . for the two photon scheme described in sec . [ sec : generation ] , we condition on a click in two detectors . once again we assume that appropriate single qubit rotations are employed depending on which detector combination clicked such that ideally the state @xmath129 is created . we find that the density matrix describing the qubit state after a successful event is @xmath130 where we have defined @xmath131 \nonumber \\ \alpha_{2}&=&\frac{(1-p_{\text{dark}})^{2}}{p_{\text{2click}}}\big[\eta_{d}(1-\eta_{\text{f}})\eta_{\text{f}}p_{\text{dark}}p_{\text{phot}}^{2}\nonumber \\ & & + p_{\text{dark}}^{2}(1-p_{\text{phot}})^{2}+\eta_{\text{d}}\eta_{\text{f}}p_{\text{dark}}p_{\text{phot}}(1-p_{\text{phot}})\nonumber \\ & & + ( 1-\eta_{\text{f}})^{2}p_{\text{phot}}^{2}p_{\text{dark}}^{2}\nonumber \\ & & + 2\eta_{\text{f}}(1-\eta_{\text{d}})(1-\eta_{\text{f}})p_{\text{phot}}^{2}p_{\text{dark}}^{2}\nonumber \\ & & + 2(1-\eta_{\text{d}}\eta_{\text{f}})p_{\text{phot}}(1-p_{\text{phot}})p_{\text{dark}}^{2}\nonumber \\ & & + \eta_{\text{d}}(1-\eta_{\text{d}})\eta_{\text{f}}^{2}p_{\text{dark}}p_{\text{phot}}^{2}\nonumber \\ & & + \eta_{\text{f}}^{2}(1-\eta_{\text{d}})^{2}p_{\text{phot}}^{2}p_{\text{dark}}^{2}\big ] \nonumber \\ \beta_{2}&=&\alpha_{2}+\frac{(1-p_{\text{dark}})^{2}}{p_{\text{2click}}}\big[\eta_{\text{d}}(1-\eta_{\text{d}})\eta_{\text{f}}^{2}p_{\text{dark}}p_{\text{phot}}^{2}\nonumber \\ & & + \eta_{\text{f}}^{2}(1-\eta_{\text{d}})^{2}p_{\text{phot}}^{2}p_{\text{dark}}^{2}\big ] . \end{aligned}\ ] ] the success probability @xmath132 is @xmath133\end{aligned}\ ] ] as in the single - photon scheme , we have not assumed number resolving detectors and we have assumed that with probability ( @xmath134 ) , an emitter spontaneously decay to one of the ground states resulting in the state @xmath124 . here we describe the local entanglement generation scheme presented in ref . @xcite , which can be used to make a deterministic cnot gate as described in sec . [ sec : cnotgate ] . we assume that weak coherent light is continuously shined onto the cavity such that at most one photon is in the cavity at all times . a single - photon detector continuously monitors if any photons are reflected from the cavity and the coherent light is blocked if a click is recorded before @xmath135 photons on average have been sent onto the cavity . if no click was recorded during this time , both atoms are interpreted as being in the @xmath20 levels . the steps of the entangling scheme are the following . 1 . both atoms are initially prepared in the superposition @xmath136 by e.g. a @xmath137-pulse . coherent light is sent onto the cavity . if a click is recorded before on average @xmath138 photons have been sent onto the cavity , the levels of the atoms are flipped ( @xmath139 ) . if no click is recorded , the atoms are interpreted as being in @xmath140 and the procedure is repeated from step 1 . 3 . conditioned on the first click , another coherent light pulse is sent onto the cavity after the levels of the atoms have been flipped . if a click is recorded before @xmath141 photons on average have been sent onto the cavity , the entangling scheme is considered to be a success . here @xmath142 is the average number of photons that had been sent onto the cavity before the first click . if no click is recorded , the atoms are interpreted as being in @xmath140 and the procedure is repeated from step 1 . as described above , the entangling scheme is repeated until it is successful leading to a deterministic creation of entanglement in the end . as described in ref . @xcite a series of non - destructive measurements of the atoms together with single qubit rotations can be used to make a cnot operation after the entanglement has been created . the non - destructive measurements can be performed using the same technique of monitoring reflected light as in the entangling scheme and we assume that we can effectively tune the couplings to the cavity such that possibly only a single atom couples . here we analyse the rate of entanglement distribution for the different repeater architectures considered in the main text . the total rate of the repeater is set by the average time of entanglement creation , initial purification and entanglement swapping . assuming that entanglement generation has a success probability , @xmath143 , we estimate the average time @xmath144 it takes to generate @xmath145 entangled pairs in one elementary link using @xmath146 qubits , which can be operated in parallel , as @xmath147 here @xmath148 is the speed of light in the fibers and @xmath149 is the time of local operations such as initialization of the qubits . the factor @xmath150 can be thought of as the average number of coin tosses needed to get at least @xmath145 tails if we have access to @xmath146 coins , which we can flip simultaneously and the probability of tail is @xmath143 for each coin @xcite . it is furthermore assumed that coins showing tail after a toss are kept and only the coins showing head are tossed again until @xmath145 tails are obtained . in the repeater context , the coins are entanglement generation attempts and tail is successful entanglement generation . the time it takes per `` toss '' is @xmath151 . to calculate the expressions for @xmath152 , we follow the lines of ref . @xcite where similar factors are derived . the expression for @xmath153 is already derived in ref . @xcite and their result is stated below @xmath154 for @xmath155 where @xmath156 , we only need to find expressions for @xmath157 with @xmath158 , @xmath159 with @xmath160 and @xmath161 since we have a maximum of 4 qubits pr . repeater station . for @xmath162 , we have that @xmath163[(q^{2})^{l-1}pq ] \nonumber \\ & & + \binom{3}{1}\binom{2}{2}\sum_{k=1}^{\infty}\sum_{l=1}^{\infty}(k+l)[(q^{3})^{k-1}pq^{2}]\nonumber \\ & & \times[(q^{2})^{l-1}p^{2}],\end{aligned}\ ] ] where @xmath164 . the first term in eq . describes the situations where three tails are obtained in a single toss after a given number of tosses , where all coins showed head . we will refer tosses where all coins show tail as failed tosses . the second term describes the situation where we get two tails in the same toss after a given number of failed tosses . the third and fourth terms are where we get a single tail after a given number of failed tosses . the coin showing tail is then kept and the two remaining coins are tossed until we obtain another tail ( third term ) or two tails simultaneously ( fourth term ) . the geometric series in eq . can be solved to give @xmath165 where the approximate expression is for @xmath166 . note that the factor of @xmath167 corresponds to a simple picture where it on average takes @xmath168 attempts to get the first tail using 3 coins and @xmath169 attempts to get the second using the remaining @xmath170 coins . in a similar manner , we find that @xmath171 here the approximate expressions are all for @xmath166 and they correspond to the expressions one would get using simple pictures similar to the one described above in the discussion of @xmath162 . after creating a number of entangled pairs in an elementary link of the repeater , they may be combined to create a purified pair of higher fidelity . as previously mentioned , we assume an entanglement pumping scheme since this requires less qubit resources than a cascading scheme . let @xmath172 denote the success probability of the purification operation , which depends on the fidelity of the two initial pairs ( @xmath173 ) and the fidelity of the cnot gate used in the purification operation . note that @xmath174 also contains the success probability of the cnot gate used in the purification for the heralded gate . we estimate the average time @xmath175 , it takes to make one purified pair from two initial pairs of fidelity @xmath173 , using @xmath146 qubits in parallel in the entanglement generation step , as @xmath176 where @xmath177 is the time of the purification operation including the classical comunication time required to compare results . here @xmath178 is the time of the cnot operation and @xmath179 is the communication time between the two repeater stations sharing the entangled pairs . to further pump the entanglement of the purified pair , a new entangled pair is subsequently created using @xmath180 qubits operated in parallel . the average time it takes to make @xmath89 rounds of purification is thus estimated as @xmath181 with @xmath182 . here @xmath183 is the fidelity of the purified pair after @xmath184 purifications . the total rate of a repeater , consisting both of purification and entanglement swapping , depends on the specific repeater achitecture . we will first consider the case of both a parallel and sequential repeater operated with deterministic gates and afterwards the same situations with probabilistic gates . for a parallel repeater with @xmath50 swap levels and deterministic gates , we first estimate the average time it takes to generate @xmath185 purified pairs , i.e. a purified pair in each elementary link . we assume that each pair is purified @xmath89 times such that the time to generate one purified pair is @xmath186 where we have solved the recurrence in eq . . we now wish to estimate the total time , @xmath187 it takes to make purification in every elementary link , i.e. the time it takes to make @xmath185 pairs . a lower limit of @xmath187 is simply @xmath188 but this is a very crude estimate if the purification have a limited success probability since the time is not determined by the average time but by the average time for the last link to succeed . we therefore make another estimate of the average time by treating @xmath188 as consisting of @xmath189 independent binomial events with probabilities @xmath190 we then estimate the average time , @xmath191 it takes to make @xmath185 purified pairs as @xmath192 eq . is a better estimate for the average time than @xmath188 in the limit of small success probabilities since it takes into account that we need success in all links . this is contained in the factors @xmath193 . however , it overestimates the average distribution time when the purification has a large success probability . how much it overestimates depends on @xmath50 and @xmath89 . comparing @xmath188 to eq . we find numerically that for @xmath194 and @xmath195 there is a factor @xmath107 between the two estimates , in the limit of large success probability for the purification operation . as discussed in sec . [ sec : optim ] we never consider more than 5 swap levels in our optimization and since we have a limited number of qubits pr . repeater station , we will never have to consider more than 2 rounds of purification . we can therefore use the estimate for @xmath187 given in eq . . to get the average time it takes to distribute one entangled pair over the total distance , @xmath196 , of the repeater , we need to add the time of the entanglement swapping , @xmath197 to @xmath187 . we estimate @xmath198 as @xmath199 where the first term is the time of the classical communication and @xmath178 is the time of the cnot operation involved in the swap procedure . the average distribution rate , of a parallel repeater with deterministic gates , is thus @xmath200 . for a repeater using sequential entanglement creation with deterministic gates , we estimate the time it takes to generate purified pairs in all @xmath185 pairs as @xmath201 here we have indicated that the number of qubits , which can be operated in parallel is @xmath202 for the first @xmath203 pairs and @xmath204 for the next @xmath203 pair compared to the parallel repeater , where only @xmath146 qubits can be used in all @xmath185 pairs . note , that we have assumed that first entanglement is established in half of the links and only when this is completed , entanglement is created in the remaining half of the links . this is clearly not the fastest way of operating the repeater but it gives an upper limit of the average distribution time and hence a lower limit on the rate . the entanglement swapping of the sequential repeater is exactly the same as for the parallel repeater and the average total rate , of the sequential repeater with deterministic gates , is thus @xmath205 . to estimate the total , average distribution time of a repeater using parallel entanglement creation with @xmath50 swap levels and probabilistic gates , we will again treat @xmath188 as consisting of @xmath189 independent binomial events , as we did for the deterministic gates . the time it takes to make a single swap can be estimated as @xmath206 where @xmath207 is the probability of the swap operation , i.e. the probability of the cnot gate . can be iterated such that the average time it takes to make @xmath50 swap levels is estimated as @xmath208 where @xmath209 here @xmath210 is either @xmath211 or @xmath212 . in the limit of @xmath213 and assuming no initial purification , eq . reduces to the well - known fomula @xcite @xmath214 since @xmath215 and the time of local operations in the swaps can be neglected in this limit . however , for higher success probabilities , eq . more accurately estimates the average distribution time . the average rate of a parallel repeater with probabilistic gates and @xmath50 swap levels is then @xmath216 . from a numerical study , we again find that for @xmath217 and @xmath218 , eq . underestimates the average distribution rate with a factor that increases with the number of swap levels , @xmath50 . however , for @xmath219 and @xmath195 , we find that this factor is @xmath107 . the operation of a sequential repeater with probabilistic gates is not straightforward since it is unclear how the sequential generation of entanglement should take place after a failed swap operation . we therefore choose to assume that initially , entanglement is generated in all @xmath185 links sequentially . when this is completed the first round of entanglement swapping is performed . if a swap fails , entanglement is restored in a parallel manner in this section , i.e. the sequential operation is only employed in the initial generation of entanglement . thus if @xmath220th swaps fail in the first swap level an extra waiting time of @xmath221 is needed to restore entanglement in the @xmath222th links in a parallel manner and swap them successfully . is very similar to eq . , which estimates the time needed for a single swap at the first swap level . nonetheless , the functions @xmath223 , which appears in eq . takes into account that we need @xmath220 successful swaps instead of only a single swap . furthermore , we assume that the swap operations of a swap level is only initiated when all swap operations in the preceeding level have been successful . the average time , it takes for all swap operations in the first level to succeed , is then estimated as @xmath224,\end{aligned}\ ] ] where @xmath225 is the kronecker delta symbol and @xmath226 . it is seen that in the limit @xmath227 , eq . correctly reduces to @xmath228 , which simply is the time of the classical communication of the results of the bell measurements and the time of the local operations . can be generalized such that the time it takes to perform the @xmath145th swap level is @xmath229,\end{aligned}\ ] ] which can be compared to eq . which estimates the time to make a successful swap at the @xmath50th level ( let @xmath230 for comparison ) . once again the functions @xmath223 takes into account that we need @xmath220th successful swaps instead of just a single successful swap . the total rate of a sequential repeater with probabilistic gates and @xmath50 swap levels can then be estimated as @xmath231 . 37ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty link:\doibase 10.1038/nature07127 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.78.3221 [ * * , ( ) ] \doibase doi:10.1088/0034 - 4885/76/9/096001 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.81.1301 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.81.5932 [ * * , ( ) ] link:\doibase 10.1038/35106500 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.83.33 [ * * , ( ) ] @noop ( ) link:\doibase 10.1103/revmodphys.82.1041 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.96.070504 [ * * , ( ) ] link:\doibase 10.1103/physreva.79.042340 [ * * , ( ) ] link:\doibase 10.1038/nature11023 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.75.3788 [ * * , ( ) ] link:\doibase 10.1103/physreva.52.3554 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.106.090502 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.91.097905 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.76.722 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.77.2818 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.84.4002 [ * * , ( ) ] link:\doibase 10.1038/35074041 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.90.253601 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.114.110502 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.91.067901 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.90.127903 [ * * , ( ) ] link:\doibase 10.1073/pnas.0703284104 [ * * , ( ) ] @noop ( ) link:\doibase 10.1103/physrevlett.67.661 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.81.3018 [ * * , ( ) ] link:\doibase 10.1103/physrevx.4.041041 [ * * , ( ) ] link:\doibase 10.1088/0034 - 4885/70/8/r03 [ * * ( ) , 10.1088/0034 - 4885/70/8/r03 ] link:\doibase 10.1038/ncomms4376 [ * * ( ) , 10.1038/ncomms4376 ] link:\doibase 10.1103/physreva.77.062301 [ * * , ( ) ] link:\doibase 10.1038/nphys3150 [ * * , ( ) ] link:\doibase 10.1038/nphys961 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.79.5178 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.78.4293 [ * * , ( ) ] link:\doibase 10.1103/physreva.83.012323 [ * * , ( ) ]

we report new results on the complexity of the valued constraint satisfaction problem ( vcsp ) . under the unique games conjecture , the approximability of finite - valued vcsp is fairly well - understood . however , there is yet no characterisation of vcsps that can be solved exactly in polynomial time . this is unsatisfactory , since such results are interesting from a combinatorial optimisation perspective ; there are deep connections with , for instance , submodular and bisubmodular minimisation . we consider the min and max csp problems ( _ i.e. _ where the cost functions only attain values in ) over four - element domains and identify all tractable fragments . similar classifications were previously known for two- and three - element domains . in the process , we introduce a new class of tractable vcsps based on a generalisation of submodularity . we also extend and modify a graph - based technique by kolmogorov and ivn ( originally introduced by takhanov ) for efficiently obtaining hardness results in our setting . this allow us to prove the result without relying on computer - assisted case analyses ( which otherwise are fairly common when studying the complexity and approximability of vcsps . ) the hardness results are further simplified by the introduction of powerful reduction techniques . * keywords * : constraint satisfaction problems , combinatorial optimisation , computational complexity , submodularity = [ draw , circle , fill = black , text = white , minimum size=4pt , inner sep=0pt ] = [ ] = [ dotted ]

1102.2880

this paper concerns the computational complexity of an optimisation problem with strong connections to the _ constraint satisfaction problem _ ( csp ) . an instance of the constraint satisfaction problem consists of a finite set of variables , a set of values ( the domain ) , and a finite set of constraints . the goal is to determine whether there is an assignment of values to the variables such that all the constraints are satisfied . csps provide a general framework for modelling a variety of combinatorial decision problems @xcite . various optimisation variations of the constraint satisfaction framework have been proposed and many of them can be seen as special cases of the valued constraint satisfaction problem ( ) , introduced by schiex et al . this is an optimisation problem which is general enough to express such diverse problems as max csp , where the goal is to maximise the number of satisfied constraints , and the minimum cost homomorphism problem ( ) , where all constraints must be satisfied , but each variable - value tuple in the assignment is given an independent cost . to accomplish this , instances of the assign costs ( possibly infinite ) to individual tuples of the constraints . it is then convenient to replace relations by _ cost functions _ , _ i.e. _ functions from tuples of the domain to some set of costs . this set of costs can be relatively general , but much is captured by using @xmath1 , where denotes the set of non - negative rational numbers . we arrive at the following formal definition . let @xmath2 be a finite domain , and let @xmath3 be a set of functions @xmath4 . by vcsp(@xmath3 ) we denote the following minimisation problem : instance : : : a set of variables @xmath5 , and a sum @xmath6 , where @xmath7 , @xmath8 , and @xmath9 is a list of @xmath10 variables from @xmath5 . solution : : : a function @xmath11 . measure : : : @xmath12 , where @xmath13 is the list of elements from @xmath2 obtained by applying @xmath14 component - wise to @xmath9 . the set @xmath3 is often referred to as the _ constraint language_. we will use @xmath3 as our parameter throughout the paper . for instance , when we say that a class of vcsps @xmath15 is polynomial - time solvable , then we mean that vcsp@xmath16 is polynomial - time solvable for every @xmath17 . finite - valued functions , _ i.e. _ functions with a range in , are sometimes called _ soft constraints_. a prominent example is given by functions with a range in @xmath0 ; they can be used to express instances of the well - known min csp and max csp problems ( which , for instance , include max @xmath18-cut , max @xmath18-sat , and nearest codeword as subproblems ) . on the other side we have _ crisp constraints _ which represent the standard type of csp constraints . these can be expressed by cost functions taking values in @xmath19 . a systematic study of the computational complexity of the was initiated by cohen et al . @xcite ; for instance , they prove a complexity dichotomy for @xmath16 over two - element domains . this was the starting point for an intensive research effort leading to a large number of complexity results for vcsp : examples include complete classifications of conservative constraint languages ( _ i.e. _ languages containing all unary cost functions ) @xcite , @xmath0 languages on three elements @xcite , languages containing a single @xmath0 cost function @xcite , and arbitrary languages with @xmath19 cost functions @xcite . we note that some of these results have been proved by computer - assisted search something that drastically reduces the readability , and insight gained from the proofs . we also note that there is no generally accepted conjecture stating which vcsps are polynomial - time solvable . the picture is clearer when considering the approximability of finite - valued vcsp . raghavendra @xcite have presented algorithms for approximating any finite - valued vcsp . these algorithms achieve an optimal approximation ratio for the constraint languages that can not be solved to optimality in polynomial time , given that the unique games conjecture ( ugc ) is true . for the constraint languages that can be solved to optimality , one gets a ptas from these algorithms . furthermore , no characterisation of the set of constraint languages that can be solved to optimality follows from raghavendra s result . thus , raghavendra s result does not imply the complexity results discussed above ( not even conditionally under the ugc ) . the goal of this paper is to prove a dichotomy result for vcsp with @xmath0 cost functions over four - element domains : we show that every such problem is either solvable in polynomial time or np - hard . such a dichotomy result is not known for csps on four - element domains ( and , consequently , not for unrestricted vcsps on four - element domains ) . our result proves that , in contrast to the two - element , three - element , and conservative case , submodularity is not the only source of tractability . in order to outline the proof , let @xmath3 denote a constraint language with @xmath0 cost functions over a four - element domain @xmath2 . we will need two tractability results in our classification . the first one is well - known : if every function in @xmath3 is submodular on a chain ( _ i.e. _ a total ordering of @xmath2 ) , then vcsp@xmath16 is solvable in polynomial time . the second result is new and can be found in section [ sec : tractable ] : we introduce _ 1-defect chain multimorphisms _ and prove that if @xmath3 has such a multimorphism , then vcsp@xmath16 is tractable . a multimorphism is , loosely speaking , a pair of functions such that @xmath3 satisfies certain invariance properties under them . the algorithm we present is based on a combination of submodular and bisubmodular minimisation @xcite . the hardness part of the proof consists of four parts ( sections [ sec : endo][sec : full ] ) . we begin by introducing some tools in section [ sec : endo ] and [ sec : graph ] . section [ sec : endo ] concerns the problem of adding ( crisp ) constant unary relations to @xmath3 without changing the computational complexity of the resulting problem . the main tool for doing this is using the concept of _ indicator problems _ introduced by jeavons et al . @xcite ( see also cohen et al . section [ sec : graph ] introduces a graph construction for studying @xmath3 . in principle , this graph provides information about the complexity of vcsp@xmath16 based on the two - element sublanguages of @xmath3 . similar graphs has been used repeatedly in the study of vcsp , _ cf . _ @xcite . equipped with these tools , we determine the complexity of vcsp@xmath16 over a four - element domain in section [ sec : binary ] . the graph introduced in section [ sec : graph ] allows us to prove that , when @xmath3 is a _ core _ ( _ cf . _ section [ sec : endo ] ) , vcsp@xmath16 is polynomial - time solvable if and only if @xmath3 is submodular on a chain or @xmath3 has a 1-defect chain multimorphism ( theorem [ thm : fourclass ] ) . some proofs of intermediate results are deferred to section [ sec : full ] . throughout this paper , we will assume that @xmath3 is a finite set of \{0,1}-valued functions . by @xmath16 we denote the problem @xmath16 . it turns out to be convenient to introduce a generalisation of this problem in which we allow additional constraints on the solutions . from a perspective , this means that we allow _ crisp _ as well as @xmath0-valued cost functions . to make the distinction clear , and since we will not be using any _ mixed _ cost functions , we represent the crisp constraints with relations instead of @xmath19-valued cost functions . let @xmath3 be a set of @xmath0-valued functions on a domain @xmath2 , and let @xmath20 be a set of finitary relations on @xmath2 . by ( @xmath21 ) we denote the following minimisation problem : instance : : : a @xmath16-instance @xmath22 , and a finite set of constraint applications @xmath23 , where @xmath24 and @xmath25 is a matching list of variables from @xmath5 . solution : : : a solution @xmath14 to @xmath22 such that @xmath26 for all @xmath27 . measure : : : the measure of @xmath14 as a solution to @xmath22 . we will generally omit the parenthesis surrounding singletons in unary relations , as in the following definition : let @xmath28 be the set of constant unary relations over @xmath2 . it is often possible to enrich a set of functions @xmath3 without changing the computational complexity of . in this paper , we will make use two distinct , but related notions aimed at this purpose . [ def : wpp ] let @xmath29 be an instance of @xmath30 , and let @xmath31 be a sequence of distinct variables from @xmath32 . let @xmath33 _ i.e. _ the projection of the set of optimal solutions onto @xmath34 . we say that such a relation _ has a weighted pp - definition in @xmath30_. let @xmath35 denote the set of relations which have a weighted pp - definition in @xmath36 . for an instance @xmath37 of , we define @xmath38 to be the optimal value of a solution to @xmath37 , and to be undefined if no solution exists . the following definition is a variation of the concept of the _ expressive power _ of a valued constraint language , see for example cohen et al . @xcite . [ def : express ] let @xmath22 be an instance of @xmath30 , and let @xmath39 be a sequence of distinct variables from @xmath32 . define the function @xmath40 by letting @xmath41 . we say that @xmath42 is _ expressible over _ @xmath36 . let @xmath43 denote the set of _ total functions _ expressible over @xmath36 . [ prop:1 ] let @xmath44 and @xmath45 be finite sets . then , @xmath46 is polynomial - time reducible to @xmath30 . the reduction from @xmath47 to @xmath48 is a special case of theorem 3.4 in @xcite . we allow weights as a part of our instances , but this makes no essential difference . for the remaining part , we will assume that @xmath49 contains a single relation @xmath50 . the case when @xmath51 , for @xmath52 can be handled by eliminating one relation at a time using the same argument . let @xmath53 be an instance of @xmath48 . for each application @xmath54 , @xmath55 , we create a copy @xmath56 of @xmath37 in which the variables @xmath34 have been replaced by @xmath57 . we now create an instance @xmath22 of @xmath36 as follows : let @xmath58 , @xmath59 , and let the set of constraint applications of @xmath22 consist of all applications from @xmath60 apart from those involving the relation @xmath61 , and all applications from @xmath56 , @xmath55 . we will choose @xmath62 large enough , so that if @xmath53 is satisfiable , then in any optimal solution @xmath14 to @xmath29 , the restriction of @xmath14 to the set @xmath63 is forced to be an optimal solution to the instance @xmath64 . it then follows that @xmath65 , so we can recover an optimal solution to @xmath66 from @xmath14 . the value of @xmath62 is chosen as follows : if all solutions to @xmath67 have the same measure , we can let @xmath68 . otherwise , let @xmath69 be the minimal difference in measure between a sub - optimal solution , and an optimal solution to @xmath67 . assume that @xmath70 , and let @xmath71 . note that if @xmath14 is any solution to the instance obtained from @xmath60 by removing all constraint applications , then @xmath72 . we can then let @xmath73 ; the representation size of @xmath62 is linearly bounded in the size of the instance @xmath60 . it is easy to check that if @xmath22 is unsatisfiable , or if @xmath74 , then @xmath60 is unsatisfiable . otherwise @xmath75 . we now turn our attention to _ multimorphisms _ and tractable minimisation problems . let @xmath2 be a finite set . let @xmath76 be a function , and let @xmath77 , with components @xmath78 . then , we let @xmath79 denote the @xmath80-tuple @xmath81 . a ( binary ) _ multimorphism _ of @xmath3 is a pair of functions @xmath82 such that for any @xmath83 , and matching tuples @xmath34 and @xmath84 , @xmath85 the concept of multimorphisms was introduced by cohen et al . @xcite as an extension of the notion of _ polymorphisms _ to the analysis of the problem . let @xmath15 be a finite set of triples @xmath86 , where @xmath87 is a finite set and @xmath88 are functions mapping @xmath89 to @xmath87 . mfm@xmath90 is a minimisation problem with instance : : : a positive integer @xmath80 , a function @xmath91 \to [ |x|]$ ] , and a function @xmath92 where @xmath93 . furthermore , @xmath94 for all @xmath95 . the function @xmath96 is given to the algorithm as an oracle , _ i.e. _ , for any @xmath97 we can query the oracle to obtain @xmath98 in unit time . solution : : : a tuple @xmath97 . measure : : : the value of @xmath98 . for a finite set @xmath15 we say that mfm@xmath90 is _ oracle - tractable _ if it can be solved in time @xmath99 for some constant @xmath100 . it is not hard to see that if @xmath101 is a multimorphism of @xmath3 , and mfm@xmath102 is oracle - tractable , then @xmath16 is tractable . we now give two examples of oracle - tractable problems . a partial order on @xmath2 is called a _ lattice _ if every pair of elements @xmath103 has a greatest lower bound @xmath104 ( meet ) and a least upper bound @xmath105 ( join ) . a _ chain _ on @xmath2 is a lattice which is also a total order . for @xmath106 , let @xmath107 be a lattice on @xmath87 . the _ product lattice _ @xmath108 is defined on the set @xmath109 by extending the meet and join component - wise : for @xmath110 and @xmath111 , let @xmath112 , and let @xmath113 . a function @xmath114 is called _ submodular _ on the lattice @xmath115 if @xmath116 for all @xmath117 . a set of functions @xmath3 is said to be submodular on @xmath118 if every function in @xmath3 is submodular on @xmath118 . this is equivalent to @xmath119 being a multimorphism of @xmath3 . it follows from known algorithms for submodular function minimisation that mfm@xmath90 is oracle - tractable for any finite set @xmath15 of finite _ distributive lattices _ ( _ e.g. _ chains ) @xcite . the second example is strongly related to submodularity , but here we use a partial order that is not a lattice to define the multimorphism . let @xmath120 , and define the functions @xmath121 by letting @xmath122 , @xmath123 if @xmath124 , and @xmath125 otherwise . we say that a function @xmath126 is _ bisubmodular _ if @xmath96 has the multimorphism @xmath127 . it is possible to minimise a @xmath18-ary bisubmodular function in time polynomial in @xmath18 , provided that evaluating @xmath96 on a tuple is a primitive operation @xcite . in this section , we introduce a new multimorphism which ensures tractability for ( and more generally for ) . let @xmath128 and @xmath100 be two distinct elements in @xmath2 . let @xmath129 be a partial order which relates all pairs of elements except for @xmath128 and @xmath100 . assume that @xmath82 are two commutative functions satisfying the following conditions : * if @xmath130 , then @xmath131 and @xmath132 . * if @xmath133 , then @xmath134 , and @xmath135 . we call @xmath136 a _ 1-defect chain _ ( over @xmath137 ) , and say that @xmath138 is the _ defect _ of @xmath136 . if a function has the multimorphism @xmath101 , then we also say that @xmath101 is a _ 1-defect chain multimorphism_. three types of 1-defect chains are shown in fig . [ fig:1](a c ) . note this is not an exhaustive list , _ e.g. _ for @xmath139 , there are 1-defect chains similar to fig . [ fig:1](b ) , but with @xmath140 . when @xmath141 , type ( b ) is precisely the product lattice shown in fig . [ fig:1](d ) . we denote this lattice by @xmath142 ( -1.5,-1.75 ) rectangle ( 1.25,1.7 ) ; ( ) at ( 0,-2 ) ( a ) ; ( ua1 ) at ( 0,-1.2 cm ) [ ] ; ( ua ) at ( 0,-0.6 cm ) [ ] edge[edged](ua1 ) ; ( ual ) at ( -0.6,-0.6 cm ) [ ] @xmath143 ; ( ud ) at ( 0,0 ) [ ] edge[edged](ua ) ; ( udl ) at ( -0.6,0 ) [ ] @xmath144 ; ( ud1 ) at ( 0,0.6 cm ) [ ] edge[edged](ud ) ; ( ub ) at ( -0.7cm,1.2 cm ) [ label = left:@xmath128 ] edge[edge](ud1 ) ; ( uc ) at ( 0.7cm,1.2 ) [ label = right:@xmath100 ] edge[edge](ud1 ) ; ( -1.5,-1.75 ) rectangle ( 1.25,1.7 ) ; ( ) at ( 0,-2 ) ( b ) ; ( ua1 ) at ( 0,-1.5 cm ) [ ] ; ( ua ) at ( 0,-0.9 cm ) [ ] edge[edged](ua1 ) ; ( ual ) at ( -0.6,-0.9 cm ) [ ] @xmath143 ; ( ua2 ) at ( 0,-0.3 cm ) [ ] edge[edged](ua ) ; ( ub ) at ( -0.7cm,0 ) [ label = left:@xmath128 ] edge[edge](ua2 ) ; ( uc ) at ( 0.7cm,0 ) [ label = right:@xmath100 ] edge[edge](ua2 ) ; ( ud1 ) at ( 0,0.3 cm ) [ ] edge[edge](ub ) edge[edge](uc ) ; ( ud ) at ( 0,0.9 cm ) [ ] edge[edged](ud1 ) ; ( udl ) at ( -0.6,0.9 cm ) [ ] @xmath144 ; ( ud2 ) at ( 0,1.5 cm ) [ ] edge[edged](ud ) ; ( -1.5,-1.75 ) rectangle ( 1.25,1.7 ) ; ( ) at ( 0,-2 ) ( c ) ; ( ua1 ) at ( 0,1.2 cm ) [ ] ; ( ua ) at ( 0,0.6 cm ) [ ] edge[edged](ua1 ) ; ( ual ) at ( -0.6,0.6 cm ) [ ] @xmath144 ; ( ud ) at ( 0,0 ) [ ] edge[edged](ua ) ; ( udl ) at ( -0.6,0 ) [ ] @xmath143 ; ( ud1 ) at ( 0,-0.6 cm ) [ ] edge[edged](ud ) ; ( ub ) at ( -0.7cm,-1.2 cm ) [ label = left:@xmath128 ] edge[edge](ud1 ) ; ( uc ) at ( 0.7cm,-1.2 ) [ label = right:@xmath100 ] edge[edge](ud1 ) ; ( -1.5,-1.75 ) rectangle ( 1.25,1.7 ) ; ( ) at ( 0,-2 ) ( d ) ; ( ua ) at ( 0,-1 cm ) [ label = below:@xmath147 ; ( ub ) at ( -0.7cm,0 ) [ label = left:@xmath128 ] edge[edge](ua ) ; ( uc ) at ( 0.7cm,0 ) [ label = right:@xmath100 ] edge[edge](ua ) ; ( ud ) at ( 0,1 cm ) [ label = above:@xmath148 edge[edge](ub ) edge[edge](uc ) ; [ ex:1defect ] let @xmath149 , and assume that @xmath136 is a 1-defect chain , with defect @xmath138 , and that @xmath150 . if @xmath151 , then @xmath152 and @xmath153 are the meet and join of @xmath142 , _ cf . _ fig . [ fig:1](d ) . when @xmath154 we have the situation in fig . [ fig:1](a ) , and when @xmath155 we have the situation in fig . [ fig:1](c ) . in the two latter cases , @xmath152 and @xmath153 are given by the two following multimorphisms ( rows and columns are listed in the order @xmath156 , _ e.g. _ @xmath157 ) : @xmath158 the proof of tractability for languages with 1-defect chain multimorphisms is inspired by krokhin and larose s @xcite result on maximising supermodular functions on maltsev products of lattices . first we will need some notation and a general lemma on oracle - tractability of mfm problems . for an equivalence relation @xmath159 on @xmath2 we use @xmath160 $ ] to denote the equivalence class containing @xmath161 . the relation @xmath159 is a _ congruence _ on @xmath136 , if @xmath162 = f(x_2,y_2)[\theta]$ ] and @xmath163 = g(x_2,y_2)[\theta]$ ] whenever @xmath164 = x_2[\theta]$ ] and @xmath165 = y_2[\theta]$ ] . we use @xmath166 to denote the set @xmath167 \mid x \in d\}$ ] and @xmath168 to denote the function @xmath169 , y[\theta ] ) \mapsto f(x , y)[\theta]$ ] . [ lem : maltsev ] let @xmath170 be two functions that map @xmath171 to @xmath2 . if there is a congruence relation @xmath159 on @xmath102 such that 1 ) mfm@xmath172 is oracle - tractable ; and 2 ) mfm@xmath173 is oracle - tractable , then mfm@xmath102 is oracle - tractable . let @xmath174 be the function we want to minimise . we define a new function @xmath175 by @xmath176 it is clear that @xmath177 . by assumption 2 in the statement of the lemma we can compute @xmath178 given @xmath179 . to simplify the notation we let @xmath180 and @xmath181 . we will now prove that @xmath178 is an instance of mfm@xmath182 . let @xmath183 and choose @xmath184 $ ] and @xmath185 $ ] so that @xmath186 ) = h({\boldsymbol{x}}')$ ] and @xmath187 ) = h({\boldsymbol{y}}')$ ] . we then have @xmath188)+h'({\boldsymbol{y}}[\theta ] ) & = h({\boldsymbol{x } } ' ) + h({\boldsymbol{y } } ' ) \label{eq:1 } \\ & \geq h(f({\boldsymbol{x}}',{\boldsymbol{y } } ' ) ) + h(g({\boldsymbol{x}}',{\boldsymbol{y } } ' ) ) \label{eq:2 } \\ & \geq h'(f({\boldsymbol{x}}',{\boldsymbol{y}}')[\theta ] ) + h'(g({\boldsymbol{x}}',{\boldsymbol{y}}')[\theta ] ) \label{eq:3 } \\ & = h'(f({\boldsymbol{x}},{\boldsymbol{y}})[\theta])+h'(g({\boldsymbol{x}},{\boldsymbol{y}})[\theta ] ) ) \label{eq:4 } \\ & = h'(u({\boldsymbol{x}}[\theta],{\boldsymbol{y}}[\theta]))+h'(v({\boldsymbol{x}}[\theta],{\boldsymbol{y}}[\theta ] ) ) . \label{eq:5}\end{aligned}\ ] ] here follows from our choice of @xmath189 and @xmath190 , follows from the fact that @xmath96 is an instance of mfm@xmath102 , follows from the definition of @xmath178 , and finally and follows as @xmath159 is a congruence relation of @xmath152 and @xmath153 . hence , @xmath178 is an instance of mfm@xmath182 and can be minimised in polynomial time by the first assumption in the lemma . armed with this lemma and the oracle - tractability of submodular and bisubmodular functions described in the previous section , we can now present a new tractable class of -problems . [ prop:1defecttract ] if @xmath3 has a 1-defect chain multimorphism , then @xmath16 is tractable . assume that @xmath3 has a 1-defect chain multimorphism @xmath101 over @xmath137 with defect @xmath138 . we prove that mfm@xmath102 is oracle - tractable . assume that @xmath128 and @xmath100 are maximal elements , _ @xmath191 for all @xmath192 . in this case the equivalence relation @xmath159 with classes @xmath193 , @xmath194 , @xmath195 is a congruence relation of @xmath102 furthermore , mfm@xmath196 and mfm@xmath197 are oracle - tractable @xcite . it now follows from lemma [ lem : maltsev ] that mfm@xmath102 is oracle - tractable . the same argument works for the case when @xmath128 and @xmath100 are minimal elements . if @xmath198 , but @xmath128 and @xmath100 are not maximal , then we can use the congruence relation @xmath199 with classes @xmath200 and @xmath201 . here , @xmath202 and @xmath203 are chains , and @xmath197 is a 1-defect chain of the previous type . one can show that when mfm(@xmath15 ) and mfm(@xmath204 ) are both oracle - tractable , then so is mfm@xmath205 . combining this with the technique used above , we can now solve the minimisation problem . an analogous construction works in the case when @xmath206 , using the congruence consisting of the class @xmath207 and its complement . finally , when @xmath208 , we can use the congruence relation @xmath209 with classes @xmath210 and @xmath211 . here , @xmath212 , @xmath213 , and @xmath214 are all chains and thus the mfm problem for these triples is oracle - tractable @xcite we now turn to prove a different property of functions with 1-defect chain multimorphisms . it is based on the following result for submodular functions on chains , which was derived by queyranne et al . @xcite from earlier work by topkis @xcite ( see also burkard et al . this formulation is due to deineko et al . @xcite : [ lem : burkard ] a function @xmath114 is submodular on a chain @xmath215 if and only if the following holds : every binary function obtained from @xmath152 by replacing any given @xmath216 variables by any constants is submodular on this chain . it is straightforward to extend this lemma to products of chains , such as @xmath142 . here , we outline the proof of the corresponding property for arbitrary 1-defect chains , which will be needed in section [ sec : binary ] . we will use the following observation . a binary operation @xmath217 is called a _ 2-semilattice _ if it is idempotent , commutative , and @xmath218 for all @xmath219 . [ prop : fg ] let @xmath136 be a 1-defect chain with a defect on @xmath138 . 1 . if @xmath220 , then @xmath152 is a 2-semilattice and @xmath221 for @xmath219 . if @xmath222 , then @xmath153 is a 2-semilattice and @xmath223 for @xmath219 . 3 . for @xmath224 , we have @xmath225 and @xmath226 . for @xmath130 , the equalities @xmath218 and @xmath221 follow from the underlying partial order . assume instead that @xmath133 , and that @xmath227 . since @xmath228 , we have that @xmath229 is the greatest lower bound of @xmath230 and @xmath231 , which is @xmath230 . we also have that @xmath232 is the lowest upper bound of @xmath230 and @xmath231 , which is @xmath231 . an analogous argument proves ( 2 ) . the first equality of ( 3 ) follows from ( 1 ) if @xmath220 , and the second equality follows from ( 2 ) if @xmath222 . at least one of @xmath220 and @xmath222 holds . if both holds , there is nothing to prove , so assume that @xmath220 , but @xmath233 . we then have @xmath234 and @xmath235 for @xmath133 , so the second equality of ( 3 ) also holds . the remaining case follows similarly . [ lem:1defectburkard ] a function @xmath126 , @xmath236 , has the 1-defect chain multimorphism @xmath101 if and only if every binary function obtained from @xmath96 by replacing any given @xmath216 variables by any constants has the multimorphism @xmath237 . let @xmath138 be the defect of @xmath101 . we prove the statement for the case @xmath220 . the other case follows analogously . every function obtained from @xmath96 by fixing a number of variables is clearly invariant under every multimorphism of @xmath96 . for the opposite direction , assume that @xmath96 does not have the multimorphism @xmath237 . we want to prove that there exist vectors @xmath183 such that @xmath238 with @xmath239 , where @xmath240 denotes the _ hamming distance _ on @xmath241 , _ i.e. _ the number of coordinates in which @xmath34 and @xmath84 differ . assume to the contrary that the result does not hold . we can then choose a function @xmath96 of minimal arity such that @xmath242 the arity of @xmath96 must in fact be equal to the least @xmath243 ; otherwise , we could obtain a function @xmath178 of strictly smaller arity by fixing the variables in @xmath96 on which @xmath34 and @xmath84 agree . this would contradict the minimality in the choice of @xmath96 . we will first show that it is possible to choose @xmath34 and @xmath84 so that @xmath244 for all @xmath245 . let @xmath246 so that @xmath247 , and let @xmath248 be two vectors with @xmath249 , satisfying ( [ eqn : hineq ] ) . now , assume that @xmath250 . we then have @xmath251 since both @xmath252 and @xmath253 are strictly less that the arity of @xmath96 , we have by assumption @xmath254 @xmath255 by combining these inequalities , we get @xmath256 @xmath257 let @xmath258 , @xmath259 , @xmath260 , and @xmath261 . by proposition [ prop : fg](3 ) , we have @xmath262 @xmath263 hence , we can rewrite the previous inequality : @xmath265 it follows that either the pair @xmath34 and @xmath84 , or the pair @xmath266 and @xmath267 satisfies condition ( [ eqn : hineq ] ) . furthermore , @xmath244 and @xmath268 , for all @xmath245 . if instead we have vectors @xmath34 and @xmath84 satisfying ( [ eqn : hineq ] ) such that @xmath244 for _ some _ , but not all @xmath245 , then we proceed as follows . note that @xmath244 implies @xmath269 . without loss of generality , we may therefore assume that @xmath270 , with @xmath271 for @xmath272 , are such that @xmath273 and @xmath274 , possibly by first exchanging @xmath34 and @xmath84 . for these vectors , condition ( [ eqn : hineq ] ) now reads : @xmath275 due to the minimality of @xmath96 s arity , we must have @xmath276 we therefore have @xmath277 let @xmath278 and @xmath279 . by proposition [ prop : fg](1 ) , @xmath152 is a 2-semilattice , so we have @xmath280 , and thus @xmath281 furthermore , @xmath282 , so @xmath283 we therefore conclude that @xmath284 so that condition ( [ eqn : hineq ] ) holds for @xmath34 and @xmath84 with @xmath244 for all @xmath245 . from now on , we assume that @xmath34 and @xmath84 are chosen in this way . let @xmath285 . for each @xmath245 , let @xmath286 be an injection which fixes @xmath287 , and sends @xmath288 to @xmath128 or @xmath100 in such a way that @xmath289 . let @xmath290 be the chain defined by @xmath291 if @xmath292 and @xmath293 , @xmath294 if @xmath191 , and @xmath295 if @xmath296 . then , @xmath297 , and @xmath298 , for all @xmath245 . let @xmath299 , and let @xmath300 be such that @xmath301 and @xmath302 . define @xmath303 . then , @xmath304 @xmath305 it follows that @xmath178 is not submodular on @xmath306 . by lemma [ lem : burkard ] , there are elements @xmath307 with @xmath308 such that @xmath309 . hence , @xmath310 @xmath311 and @xmath312 . this contradicts the original choice of @xmath96 . in this section , we show that under a natural condition , it is possible to add constant unary relations to @xmath3 without changing the computational complexity of the corresponding -problem . let @xmath313 . a function @xmath314 is called an _ endomorphism of @xmath96 _ if for every @xmath18-tuple @xmath315 , it holds that @xmath316 . the function @xmath153 is an endomorphism of @xmath3 if it is an endomorphism of each function in @xmath3 . the set of all endomorphisms of @xmath3 is denoted by ( @xmath3 ) . a bijective endomorphism is called an _ automorphism_. the automorphisms of @xmath3 form a group under composition . [ def : core1 ] a set of functions , @xmath3 , is said to be a _ core _ if all of its endomorphisms are injective . the idea is that if @xmath3 it not a core , then we can apply a non - injective endomorphism to every function in @xmath3 , and obtain a polynomial - time equivalent problem on a strictly smaller domain . we can then use results previously obtained for smaller domains @xcite . thus , we can restrict our attention to the case when @xmath3 is a core . jeavons et al . @xcite defined the notion of an _ indicator problem of order @xmath18 _ for csps . we will exploit indicator problems of order 1 here , adapted to the setting of . [ def : indicator ] let @xmath3 be a finite set of @xmath0-valued functions over @xmath2 . let @xmath317 denote the set containing a variable @xmath318 for each @xmath319 , and for @xmath320 , let @xmath321 . the indicator problem @xmath322 is defined as the instance of @xmath16 with variables @xmath317 , and sum @xmath323 , where @xmath10 is the arity of the function @xmath324 . let @xmath325 be the function defined by @xmath326 . theorem 3.5 in @xcite implies the following property of @xmath322 : for any finite set of functions , @xmath3 , the set of optimal solutions to @xmath322 is equal to @xmath327 . the proof of the following result follows the lines of similar results for related problems , such as the csp decision problem . [ prop : constants ] let @xmath3 be a core over @xmath2 . then , @xmath328 is polynomial - time reducible to @xmath16 . let @xmath37 be an instance of @xmath328 . the only way for @xmath37 to be unsatisfiable is if it contains two contradicting constraint applications @xmath329 and @xmath330 , with @xmath331 . this is easily checked in polynomial time . otherwise , let @xmath34 be a list of the variables @xmath317 , and let @xmath332 . now modify @xmath37 to an instance @xmath333 of @xmath334 as follows . add the variables in @xmath317 to @xmath335 , and add the constraint application @xmath336 . furthermore , remove each constraint @xmath329 , and replace @xmath337 by @xmath338 throughout the instance . let @xmath339 be an optimal solution to @xmath333 . since @xmath3 is a core , @xmath340 is an automorphism of @xmath3 , and so is its inverse , @xmath341 . hence , @xmath342 is also an optimal solution to @xmath333 . from @xmath14 we easily recover a solution to @xmath37 of equal measure , and conversely , any solution to @xmath37 can be interpreted as a solution to @xmath333 . it follows that we have a reduction from @xmath343 to @xmath344 . by proposition [ prop:1 ] , we finally have a reduction from @xmath344 to @xmath16 . for @xmath345 , let @xmath346 denote the function @xmath347 and @xmath348 for @xmath349 . [ lem : noend ] if @xmath350 , then @xmath351 contains a unary @xmath0-valued function @xmath352 such that @xmath353 and @xmath354 . let @xmath313 be a function in @xmath3 , and @xmath355 be elements such that @xmath356 , but @xmath357 . let @xmath37 be the instance of @xmath328 with variables @xmath358 , sum @xmath359 , and constraint applications @xmath360 for @xmath361 . then , @xmath362 is a unary function in @xmath351 , with @xmath353 and @xmath354 . let @xmath3 be a core over @xmath2 . in this section , we define a graph @xmath363 which encodes either the -hardness of @xmath328 or provides a multimorphism for the binary functions in @xmath351 . the graph is a variation of a graph defined by kolmogorov and ivn @xcite , with changes made to accommodate for additional multimorphisms . let @xmath5 be the set of partial functions @xmath364 such that * @xmath152 and @xmath153 are defined on a subset @xmath365 ; * @xmath152 and @xmath153 are idempotent and commutative ; and * @xmath366 or @xmath367 . we do allow @xmath368 in the definition of @xmath5 , _ i.e. _ there is precisely one vertex for each singleton in @xmath2 . for @xmath345 , we let @xmath369 $ ] denote the graph induced by the set of vertices defined on @xmath370 . let @xmath371 $ ] and @xmath372 $ ] . there is an edge in @xmath373 between @xmath374 and @xmath375 if there is a binary function @xmath376 such that @xmath377 the following lemma shows how @xmath378 can be used to construct multimorphisms of binary functions in @xmath351 : [ lem : indset ] let @xmath379 be an independent set in @xmath378 with precisely one vertex @xmath380 from each subgraph @xmath381 $ ] . then , every binary function @xmath376 has the multimorphism @xmath237 defined by @xmath382 and @xmath383 . assume to the contrary that @xmath101 is not a multimorphism of @xmath96 . then , there are tuples @xmath384 such that @xmath385 but this would imply that @xmath386 , which is a contradiction since @xmath387 is an independent set . for distinct @xmath103 , let @xmath388 denote the vertex @xmath389 $ ] such that @xmath390 and @xmath391 . we say that such a vertex is _ conservative_. let @xmath392 denote the set of all conservative vertices , and let @xmath393 $ ] be the subgraph of @xmath378 induced by @xmath392 . let @xmath394 be the set of vertices @xmath395 such that @xmath396 . for conservative vertices @xmath397 and @xmath398 , condition ( [ eq : basic ] ) reduces to : @xmath399 for a vertex @xmath400 , we let @xmath401 denote the vertex @xmath402 . it follows immediately from @xmath403 that @xmath404 iff @xmath405 . next , we prove a number of basic properties of the graph @xmath378 . [ lem : niceh ] if @xmath406 , then there exists a function @xmath376 such that @xmath407 . by definition of @xmath378 , we can find @xmath408 such that @xmath409 since @xmath3 is assumed to be a core , lemma [ lem : noend ] is applicable for all choices of @xmath410 and @xmath128 . using the unary functions obtained from this lemma , it is possible to ensure that the inequality in ( [ eqn : a1b2 ] ) holds for a function @xmath152 with @xmath411 . we will also assume that @xmath412 so that @xmath413 . let @xmath414 and @xmath415 be unary functions such that @xmath416 and @xmath417 , and let @xmath418 and @xmath419 , and note that @xmath420 . now , define @xmath421 the function @xmath96 satisfies the inequality @xmath422 , and furthermore , @xmath423 and @xmath424 the lemma follows . [ gprop ] 1 . [ item : xyz ] let @xmath231 , @xmath337 , and @xmath425 be conservative vertices , with @xmath426 , @xmath427 , and assume that @xmath428 . then , @xmath429 . [ item : paths ] for @xmath430 , let @xmath431 be a path of conservative vertices in @xmath378 , with @xmath432 . if @xmath80 is even , then @xmath433 , otherwise @xmath434 . [ item : loop ] for @xmath435 , let @xmath436 be an odd cycle of conservative vertices in @xmath378 , with @xmath437 . then , there is a loop on @xmath438 . [ item : prop1 ] if @xmath406 , then for each element @xmath439 , either @xmath440 or @xmath441 . [ item : trans2 ] if @xmath442 and @xmath443 , then @xmath444 [ item : abac ] if there is a loop on @xmath445 , but @xmath395 and @xmath446 are loop - free , then @xmath447 . properties ( 1)(3 ) are minor variations of lemma 11@xmath448 and @xmath449 in @xcite . we include the proofs here for completeness . ( [ item : xyz ] ) let @xmath450 , @xmath451 , and @xmath452 . by lemma [ lem : niceh ] , we have @xmath453 such that @xmath454 and @xmath455 . let @xmath456 , which is in @xmath351 since @xmath457 . now , @xmath458 . we also have @xmath459 . it follows that @xmath460 , so @xmath461 . ( [ item : paths ] ) and ( [ item : loop ] ) these two properties follow by repeated application of ( [ item : xyz ] ) , keeping in mind that @xmath404 iff @xmath405 . ( [ item : prop1 ] ) by definition there exists a function @xmath376 such that @xmath462 . if @xmath463 , then we are in the first case . otherwise , @xmath464 , so @xmath465 , which shows that we are in the second case . ( [ item : trans2 ] ) by ( [ item : prop1 ] ) , @xmath466 implies @xmath447 or @xmath467 . in the first case , we are done , so we assume that the latter holds . again by ( [ item : prop1 ] ) , @xmath468 implies @xmath469 or @xmath470 . in the latter case , we are done , hence it follows that if @xmath471 , then we have both @xmath472 and @xmath473 in @xmath373 . ( [ item : abac ] ) by ( [ item : prop1 ] ) , @xmath474 implies @xmath475 or @xmath476 . in the first case , this in turn implies either @xmath477 or @xmath478 . in the second case , it implies either @xmath479 or @xmath480 . hence , if both @xmath395 and @xmath446 are loop - free , then @xmath447 . we are now ready to derive a classification of the computational complexity of over a four - element domain . from here on , we assume that @xmath2 is the domain @xmath481 . first , we prove a result which describes the structure of the unary functions in @xmath482 , when @xmath3 is a core . let @xmath483 , @xmath484 , @xmath485 , and let @xmath486 . for distinct @xmath219 , let @xmath487 if @xmath488 , and @xmath489 otherwise . [ prop : corecases ] let @xmath3 be a core over @xmath481 and assume that @xmath490 . then , @xmath491 and for all unary functions @xmath492 , we have @xmath493 . if @xmath494 , then @xmath495 . let @xmath496 be the set of unary functions in @xmath351 . in order to simplify notation we will denote a unary function @xmath352 by the vector @xmath497 . to exclude the functions @xmath498 , @xmath499 , @xmath500 , and @xmath501 from the endomorphisms of @xmath3 , lemma [ lem : noend ] states that @xmath496 must contain certain unary @xmath0-valued functions . the following table lists the possibilities , provided that @xmath490 , so in particular @xmath502 . [ cols="^,^,^,^",options="header " , ] for each of the four functions @xmath503 , it is necessary that at least one of the three functions in the corresponding column is in @xmath496 . first assume that @xmath504 . we note that @xmath505 , so we conclude that @xmath506 . since @xmath501 is not an endomorphism of @xmath3 , we must therefore either have @xmath507 or @xmath508 in @xmath496 . in the former case , we can add @xmath509 to obtain @xmath508 , so we know that @xmath510 . by a similar argument , considering the function @xmath500 , we conclude that @xmath511 . since @xmath512 , we have reached a contradiction . a similar argument shows that @xmath506 . assume instead that @xmath513 , @xmath511 . as noted above , we must have @xmath514 , and consequently @xmath515 . but @xmath516 so again we have a contradiction . thus , the only possibility is that @xmath517 , so @xmath491 . it is not hard to see that one can write every unary function @xmath352 such that @xmath495 as a linear combination of functions from @xmath518 with non - negative coefficients . we show that if @xmath519 is a unary function in such that @xmath520 , then @xmath521 . the full statement follows similarly . let @xmath522 , and let @xmath523 . define @xmath524 if @xmath525 , and @xmath526 otherwise . then , @xmath527 , and @xmath528 can be written as a linear combination of functions from @xmath529 with non - negative coefficients . hence @xmath530 , and @xmath521 . we need the following two propositions in order to prove theorem [ thm : fourclass ] . their proofs are deferred to the next section . [ prop:6 ] assume that @xmath491 , and that @xmath531 is bipartite . then , the set of binary functions in @xmath351 is submodular on a chain . [ prop : fullprop ] assume that @xmath491 , that @xmath531 is not bipartite , but that @xmath532 $ ] is . then , the set of binary functions in @xmath351 has a 1-defect chain multimorphism . [ thm : fourclass ] let @xmath3 be a core over @xmath2 with @xmath533 . if @xmath3 is submodular on a chain , or if @xmath3 has a 1-defect chain multimorphism , then @xmath16 is tractable . otherwise , it is -hard . assume that @xmath532 $ ] has a loop on a vertex @xmath395 . it then follows from lemma [ lem : niceh ] that there is a function @xmath376 such that @xmath534 , and @xmath396 . by proposition 5.1 in @xcite , @xmath328 is -hard . by proposition [ prop : constants ] , @xmath328 reduces to @xmath16 . hence , the latter is also -hard . if instead @xmath532 $ ] is loop - free , then it is bipartite , by lemma [ gprop]([item : loop ] ) . we may assume that @xmath491 : this is trivial if @xmath535 . if @xmath536 is strictly contained in @xmath537 , then up to an automorphism we may assume that @xmath490 , and the inclusion follows by proposition [ prop : corecases ] . for a @xmath18-ary function @xmath83 , let @xmath538 be the set of binary which can be obtained from @xmath96 by fixing at least @xmath216 variables , and let @xmath539 be the union of @xmath538 over all @xmath83 . now , if @xmath531 is bipartite , then by proposition [ prop:6 ] , the set of binary functions in @xmath351 is submodular on a chain . since this set contains @xmath539 , we may conclude , by lemma [ lem : burkard ] , that @xmath3 is submodular on this chain as well . it follows that @xmath16 is tractable @xcite . otherwise , @xmath531 is not bipartite , and by proposition [ prop : fullprop ] , the set of binary functions in @xmath351 have a 1-defect chain multimorphism . since this set contains @xmath539 , we may conclude , by lemma [ lem:1defectburkard ] this time , that @xmath3 has a 1-defect chain multimorphism . it now follows from proposition [ prop:1defecttract ] that @xmath16 is tractable . [ lem : noniso ] if @xmath491 , and @xmath540 is not isolated in @xmath531 , then @xmath541 . by assumption , there is an edge @xmath542 . if @xmath543 , then @xmath544 since @xmath491 . if instead @xmath545 , then it follows from lemma [ gprop]([item : prop1 ] ) that either @xmath546 or @xmath547 , and @xmath548 due to @xmath491 . in either case , @xmath541 follows from lemma [ gprop]([item : xyz ] ) . for an independent set @xmath387 in @xmath531 , let @xmath549 denote the binary relation on @xmath2 defined by @xmath550 iff @xmath551 . let @xmath552 be a 2-colouring of the subgraph of @xmath531 induced by the non - isolated vertices . we first show that @xmath549 is a partial order on @xmath2 . let @xmath553 . then , @xmath395 and @xmath446 have the same colour in @xmath387 , and it follows that @xmath554 . hence , by lemma [ gprop]([item : trans2 ] ) , we have @xmath555 . by lemma [ lem : noniso ] , @xmath556 , so @xmath557 and @xmath558 . now , let @xmath137 be a linear extension of @xmath549 , and let @xmath559 be the corresponding subset of @xmath392 . the set @xmath560 is independent since @xmath387 is independent and @xmath561 is a set of isolated vertices in @xmath531 . since there are no edges from @xmath392 to the singleton vertices in @xmath378 , we can add all of these to @xmath560 as well . thus , by lemma [ lem : indset ] , every binary function in @xmath351 is submodular on the chain @xmath562 , where @xmath563 and @xmath564 are defined with respect to the total order @xmath137 . in the following , we will let @xmath101 denote the vertex in @xmath378 given by @xmath565 and @xmath566 . [ lem : fgz ] assume that @xmath567 and that there is an edge @xmath568 , @xmath569 . then , @xmath570 or @xmath571 , and @xmath572 or @xmath573 . let @xmath574 . by definition , there exists a function @xmath376 such that @xmath575 . if @xmath576 , then @xmath577 since @xmath578 by proposition [ prop : corecases ] . thus , @xmath579 . if @xmath580 , then we obtain @xmath581 following a similar argument , and the remaining two cases can be deduced in the same way . [ lem : u0 ] if @xmath582 , and there is a loop on @xmath583 or @xmath584 , then there is a loop on at least one of the vertices @xmath388 , @xmath585 , @xmath586 , @xmath587 . assume , without loss of generality , that there exists an @xmath376 such that @xmath588 . by proposition [ prop : corecases ] , @xmath582 implies the relations @xmath589 , @xmath590 , @xmath591 , @xmath592 , and @xmath593 . it follows that @xmath594 , which implies that the inequality @xmath595 holds in at least one of the cases @xmath596 . we follow a strategy similar to that of proposition [ prop:6 ] . however , instead of using @xmath531 we now consider the graph @xmath597 $ ] , where @xmath598 . first , we show that @xmath599 $ ] is bipartite . if @xmath600 , then @xmath599 = g[{v'}_\gamma]$ ] is bipartite by assumption . otherwise , @xmath601 . since @xmath532 = g[{v'}_0]$ ] is loop - free , we know from lemma [ lem : u0 ] that there is no loop on @xmath583 , nor on @xmath584 . thus , by lemma [ gprop]([item : loop ] ) , @xmath599 $ ] is bipartite . assume for the moment that the following holds : @xmath602 $ from $ { \overrightarrow{by}}$ to $ { \overrightarrow{yc}}$\,}}.\ ] ] let @xmath552 be a 2-colouring of the subgraph of @xmath599 $ ] induced by the non - isolated vertices . we claim that @xmath549 is a partial order on @xmath2 . let @xmath553 and observe that ( [ eqn : property ] ) implies @xmath603 . as in the proof of proposition [ prop:6 ] , we can argue that @xmath445 is connected by even paths to both @xmath395 and @xmath446 . since @xmath603 , it follows that @xmath604 . now take a transitive extension of @xmath549 which orders all pairs of elements except for @xmath128 and @xmath100 , and let @xmath559 be the corresponding subset of @xmath605 . we can assume ( possibly by swapping @xmath387 and @xmath606 ) that @xmath607 . next we show that @xmath608 is independent . this will ensure that @xmath609 holds in the constructed multimorphism . if @xmath101 is not connected to any vertex in @xmath605 , then @xmath608 is trivially independent . otherwise , by lemma [ lem : fgz ] , ( [ eqn : property ] ) , and lemma [ lem : noniso ] , we can show that from any @xmath610 such that @xmath568 , there are odd paths in @xmath599 $ ] to each vertex in the set @xmath611 . since @xmath599 $ ] is bipartite , it follows that @xmath612 , so @xmath613 by lemma [ gprop]([item : trans2 ] ) . hence , @xmath614 , and @xmath615 . it remains to verify that @xmath608 together with the singleton vertices in @xmath378 also form an independent set , _ i.e. _ that there is no edge between a singleton and @xmath101 . but by condition ( [ eq : basic ] ) this is equivalent to saying that each row and column of every binary function in @xmath351 is submodular on @xmath142 , which follows from proposition [ prop : corecases ] . by lemma [ lem : indset ] , every binary function in @xmath351 has the 1-defect chain multimorphism corresponding to @xmath608 . finally , we prove property ( [ eqn : property ] ) . if @xmath600 , then by lemma [ gprop]([item : loop ] ) , and the fact that @xmath531 contains an odd cycle , we have a loop on @xmath583 . since @xmath616 and @xmath617 are loop - free for @xmath618 , we have @xmath619 by lemma [ gprop]([item : abac ] ) . otherwise , @xmath601 . we argued above that @xmath531 does not contain any loop in this case . thus , by lemma [ gprop]([item : loop ] ) , every odd cycle @xmath620 in @xmath531 must intersect both @xmath621 and @xmath622 . now , by repeatedly applying lemma [ gprop]([item : paths ] ) to @xmath620 , we obtain a triangle on a subset of @xmath623 . by lemma [ lem : noniso ] , we can conclude that @xmath531 in fact contains the complete graph on these four vertices . in particular , we have both @xmath624 and @xmath625 . by lemma [ gprop]([item : prop1 ] ) , we therefore have either @xmath626 or @xmath627 , and furthermore , either @xmath628 or @xmath629 . since there is no loop on @xmath584 , we conclude that either the path @xmath630 or the path @xmath631 is in @xmath599 $ ] . in the same way , we find an odd path from @xmath586 to @xmath632 . we have presented a complete complexity classification for over a four - element domain . more importantly , we have compiled a powerful set of tools which will allow further systematic study of this problem . in particular , we have shown that it is possible to add ( crisp ) constants to an arbitrary core , without changing the complexity of the problem . this result holds in the more general case of as well ( although this requires a slightly different definition of endomorphisms ) , thus answering question 4 in ivn @xcite . we have also demonstrated that the techniques used by krokhin and larose @xcite for lattices can be used effectively in the context of arbitrary algebras as well , and in doing so , we have given the first example of an instance where submodularity does not suffice as an origin of tractability for . we hope that this insight will inspire an interest in the search for more tractable cases which are not explained by submodularity . finally , we have shown that graph representations such as the one defined by kolmogorov and ivn @xcite can be used to great effect , even in non - conservative settings . the curious readers may ask themselves several questions at this point , and the following one is particularly important : do 1-defect chain multimorphisms define genuinely new tractable classes ? there is still a possibility that the tractability can be explained in terms of submodularity . we answer this question negatively with the following example . consider the language @xmath633 where @xmath634 is defined such that @xmath635 if and only if @xmath636 or @xmath637 . then , @xmath3 is a core on @xmath481 but it is not submodular on any lattice . however , @xmath3 has the 1-defect chain multimorphisms @xmath374 and @xmath375 from example [ ex:1defect ] . a related question is why bisubmodularity does not appear in the classification of min csp over domains of size three @xcite . the reason is that for any cost function @xmath638 which is bisubmodular , the tuple @xmath639 minimises @xmath96 . it follows that any @xmath0 constraint language over three elements which is bisubmodular is not a core . there are several ways of extending this work , and one obvious way is to study instead of . it is known that the _ fractional polymorphisms _ of the constraint language , introduced by cohen et al . @xcite , characterise the complexity of this problem ( see also @xcite ) . multimorphisms are a special case of such fractional polymorphisms . as in the case of , it is currently not known if submodularity over every finite lattice implies tractability for . distributive lattices imply tractability , and certain constructions on lattices preserve tractability ( homomorphic images and maltsev products ) @xcite . furthermore , the five element modular non - distributive lattice ( also known as the diamond ) implies tractability for _ unweighted _ @xcite . finally , it is known that submodularity over finite modular lattices implies containment in * np * @xmath640 * conp * @xcite . it is thus clear that in order to approach further classification of either or , it will be necessary to study the complexity of minimising submodular cost functions over new finite lattices . as a last note , we mention that it seems to be possible to adapt proposition [ prop:6 ] to the classification in @xcite of for conservative finite - valued languages . this would yield a simpler description of those tractable cases .

a new non - conservative stochastic reaction - diffusion system in which two families of random walks in two adjacent domains interact near the interface is introduced and studied in this paper . such a system can be used to model the transport of positive and negative charges in a solar cell or the population dynamics of two segregated species under competition . we show that in the macroscopic limit , the particle densities converge to the solution of a coupled nonlinear heat equations . for this , we first prove that propagation of chaos holds by establishing the uniqueness of a new bbgky hierarchy . a local central limit theorem for reflected diffusions in bounded lipschitz domains is also established as a crucial tool .

1311.2325

the original motivation of this project is to study the transports of positive and negative charges in solar cells . we model a solar cell by a domain in @xmath0 that is divided into two disjoint sub - domains @xmath1 and @xmath2 by an interface @xmath3 , a @xmath4-dimensional hypersurface , which can be possibly disconnected . @xmath1 and @xmath2 represent the hybrid medium that confine the positive and the negative charges , respectively . at microscopic level , positive and negative charges are initially modeled by @xmath5 independent reflected brownian motion ( rbm ) with drift on @xmath1 and on @xmath2 respectively . ( in this paper , they are actually modeled by @xmath5 independent random walks on lattices inside @xmath1 and @xmath2 that serve as discrete approximation of rbm with drifts . ) these random motions model the transport of positive ( respectively negative ) charges under an electric potential ( see figure [ fig : interface ] ) . is the interface of @xmath6 and @xmath2 ] these two types of particles annihilate each other at a certain rate when they come close to each other near the interface @xmath3 . this interaction models the annihilation , trapping , recombination and separation phenomena of the charges . the interaction distance is of microscopic order @xmath8 where @xmath9 is comparable to 1 , and the intensity of annihilation per pair is of order @xmath10 where @xmath11 is a given parameter . this means that , intuitively and roughly speaking , according to a random time clock which runs with a speed proportional to the number of pairs ( one particle of each type ) of distance @xmath8 , we annihilate a pair ( picked uniformly among those pairs of distance less than @xmath8 ) with an exponential rate of parameter @xmath10 . the above scaling guarantees that in the limit , a nontrivial proportion of particles are annihilated in any open time interval . we investigate the scaling limit of the empirical distribution of positive and negative charges ; that is , the hydrodynamic limit of this interacting diffusion system . we show that in the macroscopic level , the empirical distribution converges to a deterministic measure whose density satisfies a system of partial differential equations that has non - linear interaction at the interface . the study of _ hydrodynamic limits _ of particle systems with stochastic dynamics is of fundamental importance in many areas . this study dates back to the sixth hilbert problem in year 1900 , which concerns the mathematical treatment of the axioms of physics , and to boltzmann s work on principles of mechanics . proving hydrodynamic limits corresponds to establishing _ the law of large number _ for the empirical measure of some attributes ( such as position , genetic type , spin type , etc . ) of the individuals in the systems . it contributes to our better understanding of the asymptotic behavior of many phenomena , such as chemical reactions @xcite , population dynamics @xcite , super - conductivity @xcite , quantum dynamics @xcite , fluid dynamics @xcite , etc . it reveals fascinating connections between the microscopic stochastic systems and deterministic partial differential equations that describe the macroscopic pictures . it also provides approximations via stochastic models to some partial differential equations that are hard or impossible to solve directly . since the work of boltzmann and hilbert , there have been many different lines of research on stochastic particle systems . various models were constructed and different techniques were developed to establish hydrodynamic limits . among those techniques , the entropy method @xcite and the relative entropy method @xcite are considered to be general methods . unfortunately these methods do not seem to work for our model due to the singular interaction near the interface . many models studied in literature are conservative , for example exclusion processes @xcite and fleming - viot type systems @xcite . reaction - diffusion systems ( r - d system ) constitute a class of models that are typically non - conservative . these are systems which have hydrodynamic limits of the form @xmath12 ( a reaction - diffusion equation , or r - d equation in short ) , where @xmath13 is a function in @xmath14 which is thought of as the reaction term . r - d systems arise from many different contexts and have been studied by many authors . for instance , for the case @xmath13 is a polynomial in @xmath14 , these systems contain the schlgl s model and were studied in @xcite on a cube with neumann boundary conditions , and in @xcite on a periodic lattice . recently , perturbations of the voter models which contain the lotka - volterra systems are considered in @xcite . in addition to results on hydrodynamic limit , @xcite also established general conditions for the existence of non - trivial stationary measures and for extinction of the particles . our model is a non - conservative stochastic particle system which consists of two types of particles . in @xcite , burdzy and quastel studied an annihilating - branching system of two types of particles , for which the total number of particles of each type remains constant over the time . its hydrodynamic limit is described by a linear heat equation with zero average temperature . in contrast , besides being non - conservative , our model gives rise to a system of nonlinear differential equations that seems to be new . moreover , the interaction between two types of particles is singular near the interface of the two media , which gives rise to a boundary integral term in the hydrodynamic limit . the approach of this paper provides some new tools that are potentially useful for the study of other non - equilibrium systems . we now give some more details on the discrete approximation of the spatial motions in our modeling . we approximate @xmath15 by square lattices @xmath16 of side length @xmath8 , and then approximate reflected diffusions on @xmath15 by continuous time random walks ( ctrws ) on @xmath16 . the rigorous formulation of the particle system is captured by the operator @xmath17 in ( [ e : generator0 ] ) . let @xmath18 be the position of the particle with index @xmath19 in @xmath20 at time @xmath21 . we prescribe each particle a mass @xmath22 and consider the normalized empirical measures @xmath23 here @xmath24 stands for the dirac measure concentrated at the point @xmath25 , while @xmath26 if and only if the particle @xmath27 is alive at time @xmath21 , and @xmath28 if and only if the particle @xmath29 is alive at time @xmath21 . for fixed positive integer @xmath5 and @xmath30 , @xmath31 is a random measure on @xmath20 . we want to study the asymptotic behavior , when @xmath32 ( or equivalently @xmath33 ) , of the evolution in time @xmath21 of the pair @xmath34 . our first main result ( theorem [ t : conjecture ] ) implies the following . suppose each particle in @xmath15 is approximating a rbm with gradient drift @xmath35 , where @xmath36 is strictly positive . then under appropriate assumptions on the initial configuration @xmath37 , the normalized empirical measure @xmath38 converges in distribution to a deterministic measure @xmath39 for all @xmath30 , where @xmath40 is the solution of the following coupled heat equations : @xmath41 and @xmath42 where @xmath43 is the inward unit normal vector field on @xmath44 of @xmath15 and @xmath45 is the indicator function on @xmath3 . note that @xmath46 corresponds to the particular case when there is no drift . the above result tells us that for any fixed time @xmath30 , the probability distribution of a randomly picked particle in @xmath16 at time @xmath21 is close to @xmath47 when @xmath5 is large , where @xmath48 is a normalizing constant . in fact , the above convergence holds at the level of the path space . that is , the full trajectory ( and hence the joint law at different times ) of the particle profile converges to the deterministic scaling limit described by ( [ e : coupledpde:+ ] ) and ( [ e : coupledpde:- ] ) , not only its distribution at a given time . * question * : how about the limiting joint distribution of more than one particles ? our second main result ( theorem [ t : correlation ] ) answers this question . it asserts that * propagation of chaos * holds true for our system ; that is , when the number of particles tends to infinity , their positions appear to be independent of each other . more precisely , suppose @xmath49 and @xmath50 unlabeled particles in @xmath51 and @xmath52 , respectively , are chosen uniformly among the living particles at time @xmath21 . then , as @xmath32 , the probability joint density function for their positions converges to @xmath53 uniformly for @xmath54 and for @xmath21 in any compact time interval , where @xmath55 is a normalizing constant . a key step in our proof of propagation of chaos ( theorem [ t : correlation ] ) is theorem [ t : uniqueness_hierarchy ] . the latter establishes uniqueness of solution for the infinite system of equations satisfied by the correlation functions of the particles in the limit @xmath32 . such infinite system of equations is sometimes called _ bbgky - hierarchy _ in statistical physics . our bbgky hierarchy involves boundary terms on the interface , which is new to the literature . our proof of uniqueness involves a representation and manipulations of the hierarchy in terms of trees . this technique is related to but different from that in @xcite which used feynman diagrams . it is potentially useful in the study of other stochastic models involving coupled differential equations . to establish hydrodynamic limit result ( theorem [ t : conjecture ] ) , we employ the classical tightness plus finite dimensional distribution approach . tightness of @xmath56 in the skorokhod space is proved in theorem [ t : tight ] . this together with the propagation of chaos result ( theorem [ t : correlation ] ) establishes the hydrodynamic limit of the interacting random walks . two new tools for discrete approximation of random walks in domains are developed in this article . namely , the local central limit theorem ( local clt ) for reflected random walk on bounded lipschitz domains ( theorem [ t : lclt_ctrw ] ) and the ` discrete surface measure ' ( lemma [ l : discreteapprox_surfacemea ] ) . we believe these tools are potentially useful in many discrete schemes which involve reflected brownian motions . weak convergence of simple random walk on @xmath16 to rbm has been established for general bounded domains in @xcite and @xcite . however , we need more for our model ; namely a local convergence result which guarantees that the convergence rate is uniform up to the boundary . for this , we establish the local clt . we further generalize the weak convergence result and the local limit theorem to deal with rbms with gradient drift . there are two reasons for us to consider gradient drift . first , it is physically natural to assume the particles are subject to an electric potential . second , the maximal extension theorem , ( * ? ? ? * theorem 6.6.9 ) , which is a crucial technical tool used in @xcite and @xcite , has established only in symmetric setting . the proof of the local clt is based on a ` discrete relative isoperimetric inequality ' ( theorem [ t : isoperimetric_discrete ] ) which leads to the poincar inequality and the nash inequality . the crucial point is that these two inequalities are uniform in @xmath8 ( scaling of lattice size ) and is invariant under the dilation of the domain @xmath57 . the paper is organized as follows . in section 2 , we introduce the stochastic model and some preliminary facts that will be used later . we then prove the existence and uniqueness of solution for the coupled pde . the main results , theorem [ t : conjecture ] and theorem [ t : correlation ] , will be rigorously formulated . we also mention various extensions of our main results in remark [ rk : generalizationresults ] . section 3 and section 4 contains the proof of theorem [ t : correlation ] and theorem [ t : conjecture ] respectively . section 5 is devoted to the proofs of the discrete relative isoperimetric inequality and the local clt . for the reader s convenience , we list our notations here : [ cols= " < , < " , ] in the last row , if @xmath58 is even , then @xmath59 and @xmath60 ; if @xmath58 is odd , then @xmath61 and @xmath62 . we can now group the terms in each level @xmath63 as a sum of @xmath63 terms to obtain @xmath64 in the last term , we have used the observation that when @xmath65 , the smallest element in @xmath66 is at least 2 and so the sum stops before reaching @xmath67 . from this and the simple estimates like @xmath68 we have derived the following [ l : i_n_j_n ] @xmath69 where @xmath70 our goal in this section is show that @xmath72 for some @xmath73 . our proof relies on the following recursion formula pointed out to us by david speyer : @xmath74 we assume ( [ e : recursion_j_n(t ) ] ) for now and use it to establish the following lemma . the proof of ( [ e : recursion_j_n(t ) ] ) will be given immediately after it . [ l : rate_j_n ] @xmath75 is homogeneous in the sense that @xmath76 moreover , @xmath77 @xmath78 is obvious from after a change of variable . let @xmath79 be the collection of functions @xmath80 satisfying @xmath81 . we can rewrite @xmath82 as @xmath83 this is a sum of @xmath84 terms . when we put @xmath85 , the smallest term is @xmath86 hence we have the lower bound @xmath87 . unfortunately , the largest term is exactly @xmath88 which grows faster than @xmath89 for any @xmath90 . hence for the upper bound , we will employ the recursion formula ( [ e : recursion_j_n(t ) ] ) . we apply the homogeneity to the right hand side of ( [ e : recursion_j_n(t ) ] ) to obtain @xmath91 the integrals are now simple one dimensional and can be evaluated : @xmath92 since @xmath93 , we have @xmath94 , where @xmath95 is defined by the recursion @xmath96 with @xmath97 . the generating function @xmath98 of @xmath95 clearly satisfies @xmath99 . we thus see that @xmath100 , where @xmath101 is the lambert @xmath101 function . by lagrange inversion theorem ( see theorem 5.4.2 of @xcite ) , @xmath102 ( for @xmath103 ) . hence by comparing coefficients in the series expansion of @xmath104 , we have @xmath105 as desired . by stirling s formula , @xmath106 ( where @xmath107 means @xmath108 ) . hence @xmath109 for some @xmath73 . monte carlo simulations suggests that @xmath110 . the recursion ( [ e : recursion_j_n ] ) also makes it clear that @xmath71 s are all in @xmath111 $ ] ( polynomials in @xmath112 with rational coefficients ) and makes it easy to compute them recursively . this is because @xmath113 is rational if @xmath114 is odd and is a rational multiple of @xmath112 if @xmath114 is even . for example , @xmath115 , @xmath116 and @xmath117 . we now turn to the proof the recursion formula ( [ e : recursion_j_n(t ) ] ) which is restated in the following lemma . [ l : recursion_j_n(t ) ] @xmath118 provided that we set @xmath119 . the proof is based on standard combinatorial methods for working on sums over planar rooted trees ( see @xcite ) . * step 1 : summing over labeled trees . * recall ( [ e : recursion_j_n(t)_1 ] ) . there are @xmath84 elements in @xmath79 . we can visualize each of them as a rooted tree with vertex set @xmath120 and a directed edge from @xmath19 to @xmath121 for each @xmath19 . for example , the 6 elements of @xmath122 can be represented by [ edge from parent/.style = draw , < - , thick , node distance=18 mm , level distance=14 mm ] = [ sibling distance=12 mm ] ( a ) 1 child node 2 child node 3 child node 4 ; 0.2truein \(b ) [ right = of a ] 1 child node 2 child node 4 child node 3 ; \(c ) [ right = of b ] 1 child node 2 child node 3 child node 4 ; \(d ) [ right = of c ] 1 child node 2 child node 3 child node 4 ; ( e ) [ right = of d ] 1 child node 2 child node 3 child node 4 ; \(f ) [ right = of e ] 1 child node 2 child node 3 child node 4 ; the trees are drawn so that arrows point upwards and the children of a given vertex are listed from left to right . note that the second and forth tree of the list are the same up to relabeling the vertices . the idea is to group terms in ( [ e : recursion_j_n(t)_1 ] ) like this together . first , we rewrite ( [ e : recursion_j_n(t)_1 ] ) in terms of trees . let @xmath123 be the set of ` decreasing trees ' , which are trees whose vertices are labeled by @xmath120 and such that @xmath124 whenever there is an edge @xmath125 . then @xmath126 * step 2 : summing over unlabeled trees . * a planar tree is a rooted unlabeled tree where , for each vertex , the children of that vertex are ordered . we draw a planar tree so that its children are ordered from left to right . here are the 5 planar rooted trees on 3 + 1 vertices : [ edge from parent/.style = draw , < - , thick , node distance=20 mm , level distance=14 mm ] = [ sibling distance=12 mm ] ( a ) @xmath127 child node @xmath127 child node @xmath127 child node @xmath127 ; 0.2truein \(b ) [ right = of a ] @xmath127 child node @xmath127 child node @xmath127 child node @xmath127 ; ( c ) [ right = of b ] @xmath127 child node @xmath127 child node @xmath127 child node @xmath127 ; ( d ) [ right = of c ] @xmath127 child node @xmath127 child node @xmath127 child node @xmath127 ; ( e ) [ right = of d ] @xmath127 child node @xmath127 child node @xmath127 child node @xmath127 ; let @xmath128 be the set of planar rooted trees with @xmath63 vertices . in general , there are @xmath129 ( the catalan number ) elements in @xmath130 , see exercise 6.19 in @xcite . we now group all the integrals in ( [ e : recursion_j_n(t)_2 ] ) with the same planar tree . for example , two different labeled trees ( the second and forth in our list of labeled trees ) both give the same unlabeled planar tree ( which is the second in the above list ) . we redraw this unlabeled planar tree @xmath131 below and attach letters @xmath132 to @xmath131 for later use . [ edge from parent/.style = draw , < - , thick ] child node b child node d child node c ; the integrands corresponding to the second and forth labeled trees are @xmath133 they are the same to @xmath134 once we relabel the variables by the vertices of @xmath131 . that is , @xmath135 for the first term and @xmath136 for the second . we now go back and keep track of the bounds of integration . in the first integral , they are @xmath137 and , in the second integral , they are @xmath138 . we can group these together as @xmath139 which is the same as @xmath140 and @xmath141 . in general , the inequality constraints we have are of two types . first , whenever we have an edge @xmath142 , we get the inequality @xmath143 . second , if @xmath144 and @xmath145 are children of @xmath14 with @xmath144 to the left of @xmath145 , then @xmath146 . let @xmath147 be the polytope cut out by these inequalities where @xmath148 is the variable at the root . we have proved @xmath149 * step 3 : grouping terms for which the root has degree @xmath63 . * we abbreviate @xmath150 if @xmath151 has more than one vertex ( otherwise @xmath152 ) . then ( [ e : recursion_j_n(t)_3 ] ) translates into @xmath153 fix an integer @xmath63 and let @xmath151 be a tree whose root has degree @xmath63 . removing the root leaves behind @xmath63 children , denoted in chronical order by @xmath154 , and @xmath63 planar subtrees @xmath155 having @xmath156 as its root for @xmath157 . then @xmath158 hence , group together the terms where the root has degree @xmath63 , we have @xmath159 here the summation conditions include that @xmath160 and each different ordering of @xmath161 are considered to be different , where @xmath162 is the number of vertices in @xmath155 . this abbreviation applies whenever @xmath163 appears . on other hand , we have by applying to each @xmath164 below that @xmath165^{k+1 } } \prod_{j=2}^{k+1 } \frac{dt_j}{\sqrt { t_1 - t_j } } \int_{p(t_j , t_j ) } \omega(t_j , t_j).\end{aligned}\ ] ] * step 4 : identifying the integrals . * it remains to show that ( [ e : recursion_j_n(t)_4 ] ) is equal to ( [ e : recursion_j_n(t)_5 ] ) . let @xmath166 denote the space of permutations of @xmath167 . then the right hand side of is equal to @xmath168 which is @xmath75 by . this completes the proof of the lemma . _ proof of theorem [ t : uniqueness_hierarchy]_. by lemma [ l : beta_i_n ] , lemma [ l : i_n_j_n ] and lemma [ l : rate_j_n ] , we have @xmath169 for all @xmath170 $ ] and @xmath171 . this implies that there is a constant @xmath172 so that @xmath173 for @xmath174 and for all @xmath175 . note that @xmath176 also satisfies the hierarchy ( [ e : hierarchy_2 ] ) , and that @xmath177 . using the hypothesis @xmath178 , we can extend to obtain @xmath173 for @xmath170 $ ] . this section is devoted to the proof of theorem [ t : conjecture ] . throughout this section , we assume the conditions of theorem [ t : conjecture ] hold . since @xmath179 has a finite state space , we know that for all bounded function @xmath180 that is smooth in the first coordinate with @xmath181 , we have two @xmath182-martingales below : @xmath183 and @xmath184 where @xmath185 is the generator defined in . see lemma 5.1 ( p.330 ) of @xcite or proposition 4.1.7 of @xcite for a proof . we will use this fact to construct some important martingales in lemma [ l : oderofm ] below . in general , suppose @xmath186 is a ctrw in a finite state space @xmath187 , whose one step transition probability is @xmath188 and mean holding time at @xmath189 is @xmath190 . its infinitesimal generator of @xmath191 is the discrete operator @xmath192 the formal adjoint @xmath193 of @xmath194 is defined by @xmath195 it can be easily checked that @xmath196 we denote by @xmath197 the generator of the ctrw @xmath198 on @xmath199 , respectively , and by @xmath200 the corresponding formal adjoint . in our case , @xmath201 for all @xmath189 . we can check that if @xmath202 , then @xmath203 [ l : oderofm ] for any @xmath204 , @xmath205 is an @xmath182-martingale for @xmath206 , where @xmath207 is the measure - valued process defined by . moreover , if @xmath208 , then there is a constant @xmath73 independent of @xmath5 so that for every @xmath209 , @xmath210}m^2(t)\big]\leq \frac{c\,t}{n}.\ ] ] similar statements hold for @xmath211 . the lemma follows by applying ( [ e : mtg_mf ] ) and ( [ e : mtg_nf ] ) to the function @xmath212 we spell out the details here for completeness . observe that @xmath213 . fix @xmath214 , and define @xmath215 to be the function from @xmath216 to @xmath217 which maps @xmath218 to @xmath219 . then by the definition of @xmath220 in ( [ e : generator0 ] ) , @xmath221 similarly , for @xmath222 , we have @xmath223 hence , by linearity of @xmath224 , ( [ e : generator1 ] ) and then ( [ e : dual_discrete_generator ] ) , we have @xmath225 hence @xmath226 and @xmath227 is an @xmath182-martingale by ( [ e : mtg_mf ] ) . next , we compute @xmath228 $ ] . note that @xmath229 where @xmath230 can be computed explicitly using ( [ e : generator0 ] ) . hence from ( [ e : mtg_nf ] ) , we can check that @xmath231=\e[\<m\>_t]=\e\left[\int_0^t{\mathfrak{l}(f^2)(\eta_s)-2f(\eta_s)\mathfrak{l}f(\eta_s)}\,ds\right]=\int_0^t{\e[g(\eta_r)]}dr,\ ] ] where @xmath232 after taking expectation for @xmath233 , the first term is the last display is of order at most @xmath22 since @xmath208 , while the second term inside the bracket is of order at most @xmath22 , uniformly in @xmath234 $ ] , by ( [ e : jump_secondmoment ] ) . hence @xmath235\leq \frac{c}{n}$ ] for some @xmath236 . doob s maximal inequality then gives . from the second term of ( [ e : formula_for_generator ] ) , we see that if the parameter of the killing time is of order @xmath10 , then we need @xmath9 to be comparable to 1 . the following simple observation is useful for proving tightness when the transition kernel of the process has a singularity at @xmath237 . it says that we can break down the analysis of the fluctuation of functionals of a process on @xmath238 $ ] into two parts . one part is near @xmath237 , and the other is away from @xmath237 where we have a bound for a higher moment . its proof , which is based on the prohorov s theorem , is simple and is omitted ( detail can be found in @xcite ) . [ l : tightness_criteria ] let @xmath239 be a sequence of real valued processes such that @xmath240 is continuous on @xmath238 $ ] a.s . , where @xmath241 . suppose the following holds : * there exists @xmath242 such that @xmath243<\infty$ ] for any @xmath244 , + * @xmath245 for any @xmath246 . then @xmath247\big\}_{n\in\mathbb{n}}$ ] is tight in @xmath248,\r)$ ] . here is our tightness result for @xmath249 . we need lemma [ l : tightness_criteria ] in the proof mainly because we do not know if @xmath250 is finite or not . [ t : tight ] the sequence @xmath249 is relatively compact in @xmath251,\mathfrak{e})$ ] and any subsequential limit of the laws of @xmath249 carries on @xmath248,\mathfrak{e})$ ] . moreover , for all @xmath252 , @xmath253 are all tight in @xmath248,\r)$ ] . we write @xmath254 in place of @xmath255 for convenience . by stone - weierstrass theorem , @xmath256 is dense in @xmath257 in uniform topology . it suffices to check that @xmath258 is relatively compact in @xmath251,\r^2)$ ] for all @xmath259 ( cf . proposition 1.7 ( p.54 ) of @xcite ) for this weak tightness criterion ) . by prohorov s theorem ( see theorem 1.3 and remark 1.4 of @xcite ) , @xmath258 is relatively compact in @xmath251,\r^2)$ ] if ( 1 ) and ( 2 ) below holds : * for all @xmath170 $ ] and @xmath246 , there exists a compact set @xmath260 such that @xmath261 * for all @xmath246 , @xmath262 we first check ( 1 ) . since @xmath263 is bounded on @xmath20 and @xmath264 for all @xmath265 , we can always take @xmath266\times[-\|\phi^{-}\|_{\infty},\,\|\phi^{-}\|_{\infty}]$ ] . to verify ( 2 ) , since @xmath267 , we only need to focus on @xmath268 . by lemma [ l : oderofm ] , @xmath269 so we only need to verify ( 2 ) with @xmath270 replaced by each of the 3 terms on rhs of the above equation ( [ e : tight ] ) . for the first term of ( [ e : tight ] ) , we apply lemma [ l : tightness_criteria ] for the case @xmath271 and @xmath272 . since @xmath273 , we have @xmath274 for some constant @xmath275 which only depends only on @xmath276 . using lemma [ l : q_near_i ] , we have @xmath277 \leq \frac{1}{n}\sum_{i=1}^n\,\sum_{d^{\eps}}|\a^+_{\eps}\phi(\cdot)|p^{\eps,+}(r , x_i,\cdot)m_{\eps}(\cdot)\leq c_1(d , d_+,\phi)+\frac{c_2(d , d_+,\phi)}{\eps\vee r^{\frac{1}{2}}},\ ] ] which is in @xmath278 $ ] as a function in @xmath279 . this implies hypothesis ( ii ) of lemma [ l : tightness_criteria ] , via the chebyshev s inequality . hypothesis ( i ) of lemma [ l : tightness_criteria ] can be verified easily using the upper bound ( [ e : uniformbound ] ) for the correlation function , or by direct comparison to the process without annihilation : @xmath280 \\ & \leq & \left(\frac{1}{n}\sum_{i=1}^n\e[\a_{\eps}\phi(x^i_r)]\right)^2 + \frac{1}{n^2}\sum_{i=1}^n\e \left [ ( \a_{\eps}\phi)^2(x^i_r ) \right]-\frac{1}{n^2}\sum_{i=1}^n\left(\,\e[\a_{\eps}\phi(x^i_r)]\,\right)^2 \\ & \leq & c(d , d,\phi)\left(1+\frac{1}{\sqrt{r}}+\frac{1}{r}\right ) . \end{aligned}\ ] ] for the second term of ( [ e : tight ] ) , by ( [ e : jump_secondmoment ] ) we have @xmath281<\infty$ ] . hence ( 2 ) holds for this term by lemma [ l : tightness_criteria ] . for the third term of ( [ e : tight ] ) , by chebyshev s inequality , doob s maximal inequality and lemma [ l : oderofm ] , we have @xmath282\\ & \leq & \frac{1}{\eps_0 ^ 2}\,\e\left[\left(2\sup_{t\in[0,t]}|m_{\phi}(t)|\right)^2\right]\\ & \leq & 16\e[m_{\phi}(t)^2]\leq \frac{c}{n } . \end{aligned}\ ] ] we have proved that ( 2 ) is satisfied . hence @xmath249 is relatively compact . using ( 2 ) and the metric of @xmath283 , we can check that any subsequential limit @xmath284 of the laws of @xmath249 concentrates on @xmath285 . in general , to prove tightness for @xmath286 in @xmath251 , a\times b)$ ] , it is not enough to prove tightness separately for @xmath287 and @xmath288 in @xmath251 , a)$ ] and @xmath251 , b)$ ] respectively . ( however , the latter condition implies tightness in @xmath251 , a)\times d([0,t ] , b)$ ] trivially ) . see exercise 22(a ) in chapter 3 of @xcite . for example , @xmath289 converges in @xmath290 but not in @xmath291 . the reason is that the two processes can jump at different times ( @xmath292 and @xmath293 ) that become identified in the limit ( only one jump at @xmath292 ) ; this can be avoided if one of the two processes is @xmath294-tight ( i.e. has only continuous limiting values ) , which is satisfied in our case since @xmath295 and @xmath211 turns out to be both @xmath294-tight . even without condition ( ii ) of theorem [ t : conjecture ] for @xmath296 , we can still verify hypothesis ( i ) of lemma [ l : tightness_criteria ] . actually , applying ( [ e : discrete ] ) to suitable test functions , we have @xmath297 = 0 .\ ] ] suppose @xmath298 is a subsequential limit of @xmath56 , say the convergence is along the subsequence @xmath299 . by the skorokhod representation theorem , the continuity of the limit in @xmath21 and ( * ? * theorem 3.10.2 ) , there exists a probability space @xmath300 such that @xmath301 } \big\| ( \x^{+,n'}_t,\,\x^{-,n'}_t)-(\x^{\infty,+}_t,\,\x^{\infty,-}_t ) \big\|_{\mathfrak{e } } = 0 \quad \p \hbox{-a.s},\ ] ] hence we have for any @xmath30 and @xmath302 , @xmath303= \e[\<\mathfrak{x}^{+,\infty}_t,\phi\ > ] \quad \text { and } \quad \lim_{n'\to\infty}\,\e[(\<\mathfrak{x}^{+}_t,\phi\>)^2]= \e[(\<\mathfrak{x}^{+,\infty}_t,\phi\>)^2].\ ] ] combining with corollary [ cor : correlation ] , we have @xmath304 here we have used the simple fact that if @xmath305= ( \e[x^2])^{1/2}=a$ ] , then @xmath306 a.s . suppose @xmath307 is a countable dense subset of @xmath308 . then for every @xmath206 , @xmath309 since @xmath310 , we can pass to rational numbers to obtain @xmath311 hence , @xmath312 similarly , @xmath313 in conclusion , any subsequential limit is the dirac delta measure @xmath314 this together with theorem [ t : tight ] completes the proof of theorem [ t : conjecture ] . suppose @xmath315 is a bounded lipschitz domain and @xmath316 is strictly positive . suppose @xmath191 is a @xmath317-reflected diffusion and @xmath318 be an @xmath8-approximation of @xmath191 , described in the subsection [ underlyingmotion ] . in this section , we prove the local central limit theorem ( theorem [ t : lclt_ctrw ] ) , the gaussian upper bound ( theorem [ t : upperhke ] ) and the hlder continuity ( theorem [ t : holdercts ] ) for @xmath319 . the proofs are standard once we establish a discrete analogue of a relative isoperimetric inequality ( theorem [ t : isoperimetric_discrete ] ) for bounded lipschitz domains . note that any lipschitz domain enjoys the uniform cone property and any bounded @xmath4-dimensional manifold of class @xmath320 has finite perimeter . hence , by corollary 3.2.3 ( p.165 ) and theorem 6.1.3 ( p.300 ) of @xcite , we have the following relative isoperimetric inequality . [ prop : relative_iso_ineq ] let @xmath315 be a bounded lipschitz domain and @xmath321 $ ] . then @xmath322 where @xmath323 is the collection of open subsets @xmath324 such that @xmath325 and @xmath326 is a manifold of class @xmath320 ( i.e. each point @xmath327 has a neighborhood which can be represented by the graph of a lipschitz function ) . moreover , @xmath328 for all @xmath329 . recall the @xmath331-symmetric ctrw @xmath318 on @xmath332 defined in the subsection [ underlyingmotion ] . the dirichlet form @xmath333 of @xmath318 in @xmath334 is given by @xmath335 where @xmath336 are the conductance on the graph @xmath332 defined in the subsection [ underlyingmotion ] . the stationary measure @xmath337 of @xmath318 is given by @xmath338 , where @xmath339 . we now consider the scaled graph @xmath340 , which is an approximation to the bounded lipschitz domain @xmath341 by square lattice @xmath342 . clearly the degrees of vertices are given by @xmath343 . define the function @xmath344 on @xmath341 by @xmath345 . then define the ctrw @xmath346 using @xmath344 as we have done for @xmath318 using @xmath347 . the mean holding time of @xmath346 is @xmath348 . clearly , the symmetrizing measure @xmath349 and the stationary probability measure @xmath350 have the scaling property @xmath351 and @xmath352 . let @xmath353 be the transition density of @xmath346 with respect to the symmetrizing measure @xmath349 . then @xmath354 for every @xmath30 , @xmath355 , @xmath329 and @xmath356 . following the notation of @xcite , we let @xmath357 be a finite set , @xmath358 be a markov kernel on @xmath357 and @xmath112 the stationary measure of @xmath359 . note that a markov chain on a finite set induces a natural graph structure as follows . let @xmath360 for any @xmath361 . define the set of directed edges @xmath362 . [ def : isoperimetricconstant ] for any @xmath367 , define @xmath368 we call @xmath369 an * isoperimetric constant * of the chain @xmath370 . it provides rich information about the geometric properties of @xmath357 and the behavior of the chain ( cf . @xcite ) . in our case , @xmath371 , @xmath372 and @xmath373 in @xmath330 , where @xmath374 is the one - step transition probabilities of @xmath346 defined in subsection [ underlyingmotion ] . for @xmath375 and @xmath376 , we have @xmath377\subset d \ } , \\ \partial a & = & \{x\in a:\ , \exists y\in d^{\eps}\setminus a \text { such that } |x - y|=\eps \text { and the line segment } [ x , y]\subset d\ , \ } , \\ \tilde{\partial } a & : = & \{x\in a:\,\exists y\in \eps\z^d \text { such that } |x - y|=\eps \text { and } ( x , y]\cap \partial d \neq \emptyset\ } , \\ \delta a & : = & \tilde{\partial}a \setminus \partial a.\end{aligned}\ ] ] in this notation , we have @xmath378 , @xmath379 , @xmath380 and @xmath381 . see figure [ fig : extenddomain2_1 ] for an illustration . the following is a key lemma which allows us to derive the relative isoperimetric inequality for the discrete setting from that in the continuous setting , and hence leads us to theorem [ t : isoperimetric_discrete ] . ( extension of sub - domains)[l : extenddomain ] let @xmath389 be the stationary measure of the simple random walk ( srw ) on @xmath332 . for any @xmath367 , there exist positive constants @xmath390 , @xmath391 and @xmath392 such that if @xmath393 , then for all grid - connected @xmath376 with @xmath394 , we can find a connected open subset @xmath324 which contains @xmath384 and satisfies : for @xmath400 , let @xmath401 be the cube in the dual lattice which contains @xmath189 . since @xmath384 is grid - connected , we have @xmath402 is connected in @xmath0 , where @xmath402 is the interior of @xmath403 . ( see figure [ fig : extenddomain2_1 ] for an illustration . ) note that we can not simply take @xmath404 because ( d ) may fail , for example when @xmath405 contributes too much to @xmath326 , i.e. , when @xmath406 is large . however , @xmath407 is close to @xmath408 and so we can fill in the gaps between @xmath405 and @xmath408 to eliminate those contributions . in this process , we may create some extra pieces for @xmath326 , but we will show that those pieces are small enough . following this observation , we will eventually take @xmath409 where @xmath410 for some small enough @xmath244 . since @xmath411 is a bounded lipschitz domains , we can choose @xmath244 small enough so that @xmath412 . moreover , @xmath413 implies @xmath414 . so we can choose @xmath8 small enough so that @xmath415 . hence @xmath416 satisfies ( b ) . by lipschitz property again , there exists @xmath417 such that @xmath418 for any @xmath419 . hence ( c ) is satisfied . it remains to construct @xmath420 in such a way that @xmath410 for some small enough @xmath244 ( more precisely , for @xmath421 small enough so that @xmath412 ) and that ( a ) and ( d ) are satisfied . we will construct @xmath420 in 3 steps : step 1 : ( construct @xmath422 to seal the opening between @xmath408 and the subset of @xmath423 which are close to @xmath408 . see figure [ fig : extenddomain2_2 ] . ) write @xmath424 where @xmath425 and @xmath426 points in @xmath427 are marked in solid black in figure [ fig : extenddomain2_2 ] . for @xmath428 , consider the following closed cube centered at @xmath189 : @xmath429 let @xmath430 be the union of all connected components of @xmath431 whose closure intersects @xmath432 and define @xmath433 step 2 : ( fill in the gaps between @xmath408 and @xmath434 near @xmath435 . see figure [ fig : extenddomain2_3 ] ) note that @xmath436 does not contribute to @xmath437 . let @xmath438 be the union of all connected components of @xmath439 whose closure intersects @xmath432 for some @xmath440 . it is clear that @xmath416 is connected and @xmath443 is piecewise linear , so ( a ) is satisfied . for any @xmath444 , we have @xmath445 , where @xmath446 therefore , @xmath447 whenever the corresponding surface measures are defined . it is clear that by construction we have now @xmath453 . moreover , each @xmath454 is adjacent to at most @xmath455 points in @xmath456 , and for each @xmath428 , there are at most @xmath457 cubes in @xmath458 . so we have @xmath459 hence ( d ) is satisfied . since @xmath460 , we have @xmath461 . to complete the proof , it suffices to show that @xmath462 . this is equivalent to show that any curve in @xmath463 starting from any point in @xmath450 must lie in @xmath464 . let @xmath465 $ ] be an arbitrary continuous curve starting at an arbitrary point @xmath466 such that @xmath467 and @xmath468 for some @xmath469 $ ] . define @xmath470 . since @xmath471 and @xmath472 , the time when @xmath473 first exits @xmath474 must be less than @xmath21 by continuity of @xmath473 . that is , @xmath475 it suffices to show that @xmath476 \cap \theta_{x}\neq\emptyset$ ] for some @xmath428 . we do so by constructing a continuous curve @xmath477 which is close to @xmath473 and passes through @xmath478 for some @xmath428 . since @xmath480}dist(\gamma(s),\,\partial d ) < 2r\eps$ ] , we can choose @xmath8 small enough ( depending only on @xmath411 ) and split @xmath481 $ ] into finitely many disjoint intervals @xmath3 s so that the @xmath482-tube of each @xmath483 lies in a coordinate ball @xmath484 of @xmath411 . for @xmath485 , project @xmath486 vertically upward ( along the @xmath487-th coordinate of @xmath484 ) onto @xmath488 to obtain @xmath489 . note that @xmath490 maybe discontinuous even in the interior of @xmath3 . however , it is continuous on @xmath491 $ ] except possibly for finitely many points . let @xmath492 be the collection of discontinuities for @xmath493)$ ] . then @xmath494 and we can connect @xmath495 to @xmath496 by a continuous curve @xmath497\longrightarrow \partial ( \omega_{d^{\eps}})\cap d\ , \cap b(\widehat \gamma(s_j-),\,8r\eps ) \cap b(\widehat \gamma(s_j+),\,8r\eps)$ ] . define @xmath498\longrightarrow \partial ( \omega_{d^{\eps}})\cap d$ ] to be the continuous curve obtained by concatenating @xmath490 and @xmath499 ( see figure [ fig : extenddomain_curve ] ) . then @xmath500 and @xmath501 . by the continuity of @xmath477 , there is some @xmath502 such that @xmath503 . ( roughly speaking , on @xmath504 , @xmath505 is separated from @xmath506 by @xmath427 . ) we can also assume that @xmath384 is grid - connected . this is because @xmath518 whenever @xmath519 with @xmath520 . from lemma [ l : extenddomain ] and proposition [ prop : relative_iso_ineq ] , we have @xmath521 we thus have the first inequality since @xmath522 . the second inequality follows from the first since @xmath523 . for the ctrw @xmath346 on @xmath330 , we let @xmath112 be the stationary measure . observe that , because @xmath524 and @xmath525 , we have @xmath526 for all @xmath329 and @xmath527 . hence we only need to consider the case @xmath375 . in view of theorem [ t : isoperimetric_discrete ] and ( [ e : approximate_m(x ) ] ) , we have ( taking @xmath399 ) [ cor : poincare](poincar inequality ) there exist @xmath532 such that @xmath533 for all @xmath534 , @xmath535 , @xmath329 . here @xmath536 , @xmath537 and @xmath538 are the same constants in corollary [ cor : isoperimetric_discrete ] , and @xmath539 is the dirichlet form in @xmath540 of the ctrw @xmath346 ( see ( [ e : dirichletform_ctrw ] ) ) . the above poincar inequality already tells us a positive lower bound for the spectral gap of @xmath346 and hence gives us an estimate for the mixing time . however , we will state a stronger result in proposition [ prop : mixing ] in the next subsection . the discrete relative isoperimetric inequality leads to the following two functional inequalities ; namely , a poincar inequality and a nash inequality that are uniform in @xmath8 and in scaling @xmath543 . the uniformity in scaling helps proving the near diagonal lower bound for @xmath319 . ( nash s inequality and poincar inequality uniform in @xmath8 and in scaling ) there exist @xmath532 and @xmath544 such that @xmath545 @xmath546 for every @xmath547 , @xmath548 and @xmath549 , where @xmath537 and @xmath538 are the same constants in corollary [ cor : isoperimetric_discrete ] ; @xmath536 and @xmath539 is the dirichlet form in @xmath540 of the ctrw @xmath346 ( see ( [ e : dirichletform_ctrw ] ) ) . note that @xmath550 is the dirichlet form of the unit speed ctrw with the same one - step transition probabilities as that of @xmath551 . hence ( [ e : nash_a1 ] ) follows directly from ( * ? ? ? * theorem 3.3.11 ) and ( [ hats ] ) . for ( [ e : nash_a2 ] ) , let @xmath552 with @xmath553 . for any nonempty subset @xmath554 , @xmath555 hence , @xmath556 by ( * ? ? ? * theorem 3.3.10 ) , @xmath557 [ prop : mixing](mixing time estimate ) there exists @xmath73 which depends only on @xmath487 such that @xmath560 for every @xmath30 , @xmath561 , @xmath535 and @xmath329 . here @xmath537 and @xmath538 are the constants in corollary [ cor : isoperimetric_discrete ] . by ( [ e : nash_a1 ] ) and theorem 2.3.1 of @xcite , we have @xmath562 after simplification and using ( [ e : isoperimetric_discrete2 ] ) , we obtain the upper bound which is of order @xmath563 . on other hand , by corollary [ cor : poincare ] and ( * ? ? ? * lemma 2.1.4 ) , we obtain the exponential term on the right hand side . equipped with the nash inequality ( [ e : nash_a2 ] ) and the poincar inequality ( [ e : nash_a1 ] ) , one can follow a now standard procedure ( see , for example , @xcite or @xcite ) to obtain two sided gaussian estimates for @xmath319 . in the following , @xmath564 , @xmath565 and @xmath566 are positive constants which depends only on @xmath567 and @xmath151 . [ t : upperhke_a ] there exist constants @xmath568 , @xmath569 , and @xmath570 $ ] such that @xmath571 for every @xmath572 , @xmath573 , @xmath329 and @xmath574 . moreover , the following weaker bound holds for all @xmath30 : @xmath575 in particular , this implies the upper bound in theorem [ t : upperhke ] which is the case when @xmath375 . the gaussian lower bound for @xmath319 in theorem [ t : lowerhke ] then follows from the lipschitz property of @xmath411 and a well - known chaining argument ( see , for example , page 329 of @xcite ) . therefore , we have the two - sided gaussian bound for @xmath319 as stated in theorem [ t : upperhke ] and theorem [ t : lowerhke ] . it then follows from a standard ` oscillation ' argument ( cf . theorem 1.31 in @xcite or theorem ii.1.8 in @xcite ) that @xmath319 is hlder continuous in @xmath580 , _ uniformly in @xmath8_. more precisely , [ t : weakconvergence_rbmdrift ] let @xmath315 be a bounded domain whose boundary @xmath408 has zero lebesque measure . suppose @xmath411 also satisfies : @xmath585 suppose @xmath316 is strictly positive . then for every @xmath209 , as @xmath586 , 1 . @xmath587 converges weakly to the stationary process @xmath588 in the skorokhod space @xmath251,{\overline}{d})$ ] . @xmath589 converges weakly to @xmath590 in the skorokhod space @xmath251,{\overline}{d})$ ] whenever @xmath591 converges to @xmath592 . for ( i ) , the proof follows from a direct modification of the proof of ( * ? ? ? * theorem 3.3 ) . recall the definition of the one - step transition probabilities @xmath188 , defined in the paragraph that contains ( [ e : conductance_baisedrw+ ] ) and ( [ e : conductance_baisedrw- ] ) . observe that , since @xmath593 , approximations using taylor s expansions in the proofs of ( * ? ? ? * lemma 3.1 and lemma 2.2 ) continue to work with the current definition of @xmath188 . thus we have @xmath594 the process @xmath318 has a lvy system @xmath595 , where for @xmath419 , @xmath596 following the same calculations as in the proof of ( * ? ? ? * theorem 3.3 ) , while noting that ( * ? ? ? * theorem 6.6.9 ) ( in place of ( * ? ? ? * theorem 1.1 ) ) can be applied to handle general symmetric reflected diffusions as in our present case , we get part ( i ) . part ( ii ) follows from part ( i ) by a localization argument ( cf . * remark 3.7 ) ) . _ proof of theorem [ t : lclt_ctrw]_. for each @xmath355 and @xmath30 , we extend @xmath597 to @xmath598 in such a way that @xmath319 is nonnegative and continuous on @xmath599 , and that both the maximum and the minimum values are preserved on each cell in the grid @xmath600 . this can be done in many ways , say by the interpolation described in @xcite , or a sequence of harmonic extensions along the simplexes ( described in @xcite ) . consider the family @xmath601 of continuous functions on @xmath602 . theorem [ t : upperhke ] and theorem [ t : holdercts ] give us uniform pointwise bound and equi - continuity respectively . by arzela - ascoli theorem , it is relatively compact . i.e. for any sequence @xmath603 $ ] which decreases to @xmath604 , there is a subsequence @xmath605 and a continuous @xmath606 such that @xmath607 converges to @xmath608 locally uniformly . on other hand , by part ( ii ) of theorem [ t : weakconvergence_rbmdrift ] , if the original sequence @xmath609 is a subsequence of @xmath610 , then @xmath611 . more precisely , the weak convergence implies that for all @xmath30 , @xmath612 then by the continuity of both @xmath613 and @xmath608 in the second coordinate , we have @xmath611 on @xmath614 . since @xmath615 and @xmath616 are continuous on @xmath598 ( cf . @xcite ) , we obtain @xmath617 on @xmath618 . in conclusion , we have @xmath319 converges to @xmath613 locally uniformly through the sequence @xmath619 . * acknowledgements * : we thank krzysztof burdzy , rekhe thomas and tatiana toro for helpful discussions . in particular , we are grateful to david speyer for pointing out the recurrence relation to us . we also thank guozhong cao , samson a. jenekhe , christine luscombe , oleg prezhdo and rudy schlaf for discussions on solar cells . financial support from nsf solar energy initiative grant dmr-1035196 as well as nsf grant dms-1206276 is gratefully acknowledged . r.f . bass and p. hsu . some potential theory for reflecting brownian motion in hlder and lipschitz domains . * 19 * ( 1991 ) , 486 - 508 . r.f . bass and t. kumagai . symmetric markov chains on @xmath620 with unbounded range . * 360 * ( 2008 ) , 2041 - 2075 . comparison of stochastic and deterministic models of a linear chemical reaction with diffusion . * 19 * ( 1991 ) , 1440 - 1462 . c. boldrighini , a. de masi and a. pellegrinotti . nonequilibrium fluctuations in particle systems modelling reaction - diffusion equations . * 42 * ( 1992 ) , 1 - 30 . c. boldrighini , a. de masi , a. pellegrinotti and e. presutti . collective phenomena in interacting particle systems . appl . _ * 25 * ( 1987 ) , 137 - 152 . k. burdzy and z .- q . discrete approximations to reflected brownian motion . probab . _ * 36 * ( 2008 ) , 698 - 727 . k. burdzy and z .- q . chen . reflected random walk in fractal domains . * 41 * ( 2011 ) , 2791 - 2819 . k. burdzy , r. holyst and p. march . a fleming - viot particle representation of dirichlet laplacian . * 214 * ( 2000 ) , 679 - 703 . k. burdzy and j. quastel . an annihilating - branching particle model for the heat equation with average temperature zero . _ * 34 * ( 2006 ) , 2382 - 2405 . carlen , s. kusuoka and d.w . upper bounds for symmetric markov transition functions . henri poincar - probab . * 23 * ( 1986 ) , 245 - 287 . mr 898496 . _ from markov chains to non - equilibrium particle systems . _ world scientific , 2003 . on reflecting diffusion processes and skorokhod decompositions . theory relat . fields _ * 94 * ( 1993 ) , 281 - 316 . g. david . morceaux de graphes lipschitziens et intgrales singulires sur une surface . _ rev . mat . iberoamericana . _ * 4 * ( 1988 ) , 73114 ( french ) . mr 1009120 . g. david and s. semmes . singular integrals and rectifiable sets in @xmath621 : beyond lipschitz graphs . _ astrisque . _ * 193 * ( 1991 ) , 152 . mr 1113517 ( 92j:42016)9120 . p. dittrich . a stochastic model of a chemical reaction with diffusion . theory relat . fields . _ * 79 * ( 1988 ) , 115 - 128 . p. dittrich . a stochastic partical system : fluctuations around a nonlinear reaction - diffusion equation . _ stochastic processes . appl . _ * 30 * ( 1988 ) , 149 - 164 . l. erds , b. schlein and h. t. yau . derivation of the cubic non - linear schrdinger equation from quantum dynamics of many - body systems . _ inventiones mathematicae _ * 167*(3 ) ( 2007 ) , 515 - 614 . ethier and t.g . _ markov processes . characterization and convergence . _ wiley , new york , 1986 . systems of reflected diffusions with interactions through membranes . phd thesis . in preparation , 2014 . hydrodynamic limits . soc . _ * 1 * ( 2005 ) , 699 - 717 . guo , g.c . papanicolaou and s.r.s . nonlinear diffusion limit for a system with nearest neighbor interactions . phys._. * 118 * ( 1988 ) , 31 - 59 . p. gyrya and l. saloff - coste . _ neumann and dirichlet heat kernels in inner uniform domains . _ paris : socit mathmatique de france , 2011 . c. kipnis and c. landim . _ scaling limits of interacting particle systems . _ springer , 1998 . c. kipnis , s. olla and s.r.s . hydrodynamics and large deviations for simple exclusion process . _ comm . pure appl . * 42 * ( 1989 ) , 115 - 137 . p. kotelenez . law of large numbers and central limit theorem for linear chemical reactions with diffusion . * 14 * ( 1986 ) , 173 - 193 p. kotelenez . high density limit theorems for nonlinear chemical reactions with diffusion . theory relat . fields _ * 78 * ( 1988 ) , 11 - 37 . limit theorems for sequences of jump markov processes approximating ordinary differential processes . _ j. appl . * 8 * ( 1971 ) , 344 - 356 . _ approximation of population processes . _ siam , philadelphia , 1981 . r. lang and n.x . smoluchowski s theory of coagulation in colloids holds rigorously in the boltzmann - grad - limit . _ zeitschrift fr wahrscheinlichkeitstheorie und verwandte gebiete . _ * 54 * ( 1980 ) , 227 - 280 . r. m. may and m. a. nowak . evolutionary games and spatial chaos . * 359.6398 * ( 1992 ) , 826 - 829 . v.g . _ sobolev spaces . _ springer verlag , 1985 l. saloff - coste . _ lectures on finite markov chains . _ lecture notes in math . , springer , 1997 . stanley . _ enumerative combinatorics ii . _ cambridge university press , cambridge , 1999 . stroock . diffusion semigroups corresponding to uniformly elliptic divergence form operators . _ sminaire de probabilits xxii . _ lecture notes in math . , vol . 1321 , pp . 316 - 347 , springer , berlin , 1988 . stroock and w. zheng . markov chain approximations to symmetric diffusions . poincar - probab . statist . _ * 33 * ( 1997 ) , 619 - 649 . mr 1473568 .

it has been shown that the anomalies observed in and decays can be resolved by adding a single scalar or vector leptoquark to the standard model , while constraints from other precision measurements in the flavour sector can be satisfied without fine - tuning . to further explore these two interesting scenarios , in this paper , we study their effects in the semi - leptonic decay . using the best - fit solutions for the operator coefficients allowed by the current data of mesonic decays , we find that ( i ) the two scenarios give the similar amounts of enhancements to the branching fraction and the ratio , ( ii ) the two best - fit solutions in each of the two scenarios are also indistinguishable from each other , ( iii ) both scenarios give nearly the same predictions as the standard model for the longitudinal polarizations of and as well as the lepton - side forward - backward asymmetry . with future measurements of these observables in decay at the lhcb , the two leptoquark scenarios could be further tested , and even differentiated from the other np explanations to the anomalies .

1611.01635

while no direct evidences for physics beyond the standard model ( sm ) have been found at the lhc so far , there are some interesting indirect hints for new physics ( np ) in the flavour sector @xcite . it is particularly interesting to note that intriguing effects of lepton favour universality violation ( lfuv ) have been observed in rare @xmath8-meson decays . to be more specific , the ratios of charged - current decays , @xmath9 and @xmath10 , with @xmath11 , have been measured by the babar @xcite , belle @xcite and lhcb @xcite collaborations . the latest averages by the heavy flavor average groups ( hfag ) @xcite , @xmath12 exceed the sm predictions , @xmath13 by @xmath14 and @xmath15 , respectively . once the measurement correlations between @xmath16 and @xmath17 are taken into account , the deviation will be at @xmath18 level . another hint of lfuv has also been reported in the @xmath19 process by the lhcb experiment @xcite : @xmath20\,{\rm gev}^2}=0.745_{-0.074}^{+0.090}({\mathrm{stat.}})\pm0.036({\mathrm{syst.}})\,,\ ] ] which is about @xmath21 lower than the corresponding sm prediction @xmath22 @xcite . the observed @xmath7 and @xmath23 anomalies , if confirmed with future more precise data , would be clear signs for np beyond the sm , and have already inspired lots of studies ; for a recent review , the readers are referred to refs . @xcite and references therein . here we are interested in the possible np solutions with a single scalar or vector leptoquark ( lq ) scenario @xcite . in ref . @xcite , it has been shown that the anomalies @xmath7 , @xmath23 and @xmath24 could be addressed by adding to the sm just one tev - scale scalar lq transforming as @xmath25 under the sm gauge group . on the other hand , as shown in ref . @xcite , the @xmath7 , @xmath23 and the angular observable @xmath26 in @xmath27 decay could be explained by just one vector lq transforming as @xmath28 under the sm gauge group . under the constraints from both the ratios @xmath7 and the @xmath29 spectra of @xmath0 decays provided by the babar @xcite and belle @xcite collaborations , four best - fit solutions are found for the operator coefficients induced by the scalar lq @xcite , two of which are , however , already excluded by the purely leptonic @xmath30 decay @xcite . at the same time , two best - fit solutions are also found for the operator coefficients induced by the vector lq @xcite . to further explore the two interesting lq scenarios , in this paper , we shall study their effects in the semi - leptonic @xmath2 decay , which is induced by the same quark - level transition as the @xmath0 decays . while the @xmath31 baryons are not produced at an @xmath32 @xmath8-factory , they account for about @xmath33 of the @xmath34-hadrons produced at the lhc @xcite . remarkably , the produced number of @xmath31 baryons is comparable to that of @xmath35 or @xmath36 mesons , and is significantly higher than the one of @xmath37 mesons @xcite . due to the spin - half nature of @xmath31 , its decay may provide complementary information compared to the corresponding mesonic one . motivated by the @xmath7 anomalies , the semi - leptonic @xmath38 decay has been studies recently in refs . @xcite . in this paper , besides the total and differential branching fractions , we shall also discuss the ratio @xmath39 , the longitudinal polarizations of @xmath5 and @xmath6 , as well as the lepton - side forward - backward asymmetry . for the @xmath40 transition form factors , we use the results both from qcd sum rules @xcite , which satisfy the heavy quark effective theory ( hqet ) relations @xcite , and from the latest lattice calculations @xcite . using the best - fit solutions for the operator coefficients allowed by the current data of mesonic decays , we find that the two scenarios give the similar amounts of enhancements relative to the sm predictions for the branching fraction @xmath3 and the ratio @xmath41 , and the two best - fit solutions in each of the two scenarios are also indistinguishable from each other based only on these two observables . on the other hand , both of these two scenarios give nearly the same predictions as the sm for the longitudinal polarizations of @xmath5 and @xmath6 as well as the lepton forward - backward asymmetry . with future precise measurements of these observables at the lhcb , the two scenarios could be further tested and even differentiated from the other explanations to the @xmath7 anomalies @xcite . this paper is organized as follows : in section [ sec : framework ] , we recapitulate briefly both the scalar and vector lq scenarios @xcite . in section [ sec : analytic calculation ] , we calculate the helicity amplitudes and list the relevant physical observables for the semi - leptonic @xmath38 decays . in section [ sec : numerical results ] , the scalar and vector lq effects on the branching fraction @xmath3 , the ratio @xmath41 , the @xmath5 and @xmath6 longitudinal polarizations , as well as the lepton - side forward - backward asymmetry are discussed . we finally conclude in section [ sec : conclusions ] . the @xmath40 transition form factors and the helicity - dependent differential decay rates are collected in appendices [ sec : form factors ] and [ sec : helicity rates ] , respectively . in this section , we recapitulate the scalar and vector lq models , where a single tev - scale scalar or vector lq is added to the sm to address the aforementioned anomalies @xcite . for a recent comprehensive review of lq models , the readers are referred to ref . @xcite . firstly , we consider the scalar lq @xmath42 transforming as @xmath43 under the sm gauge group , in which its couplings to sm fermions are described by the lagrangian @xcite @xmath44 where @xmath45 are the yukawa coupling matrices in flavour space , and @xmath46 denote the left - handed quark and lepton doublet , while @xmath47 the right - handed up - type quark and lepton singlet , respectively . the charge - conjugated spinors are defined as @xmath48 , @xmath49 ( @xmath50 ) . such a scalar @xmath42 mediates the @xmath51 decay at tree level , and the resulting effective weak hamiltonian including the sm contribution is given as @xcite @xmath52\,,\end{aligned}\ ] ] where @xmath53 , @xmath54 , @xmath55 are the wilson coefficients of the corresponding four - fermion operators and , at the matching scale @xmath56 , are given explicitly as @xmath57 & c_s(m_\phi)=c_t(m_\phi)=-\frac{\lambda_{b\nu_{\tau}}^l\lambda_{c\tau}^{r\ast } } { 4\sqrt{2}g_fv_{cb}m_{\phi}^2}\,.\label{eq : wcmuphi}\end{aligned}\ ] ] in order to resum potentially large logarithmic effects , the wilson coefficients @xmath54 and @xmath55 should be run down to the characteristic scale of the process we are interested in , _ i.e. _ , @xmath58 , while @xmath53 is not renormalized because of the conservation of vector currents . the explicit evolution equations could be found , for example , in ref . @xcite . as shown in ref . @xcite , such a scalar lq could explain the @xmath7 , @xmath23 and @xmath24 anomalies , while constraints from other precision measurements in the flavour sector can be satisfied without fine - tuning . especially , under the constraints from both the ratios @xmath7 and the measured @xmath29 spectra in @xmath59 decays , four best - fit solutions are found for the operator coefficients induced by the scalar lq @xcite , two of which are , however , already excluded by the purely leptonic @xmath30 decay @xcite . consequently , in this paper , we shall consider only the remaining two solutions denoted by @xmath60 and @xmath61 in ref . @xcite . we now introduce the second scenario in which the sm is extended by a vector @xmath62 triplet @xmath63 transforming as @xmath28 under the sm gauge group . the coupling of the vector multiplet @xmath63 to a lepton - to - quark current with @xmath64 structure is given by @xcite @xmath65 where @xmath66 ( @xmath67 ) are the pauli matrices in the @xmath62 space , and @xmath68 and @xmath69 ( @xmath70 ) the left - handed lepton and quark doublets , respectively . the lagrangian eq . is written in the fermion mass basis , with @xmath71 defined as the couplings of the @xmath72 component of the triplet , @xmath73 , to @xmath74 and @xmath75 . expanding the @xmath62 components , we get explicitly @xmath76\,\nonumber\\[0.2 cm ] & \quad + u_{3\mu}^{(5/3)}(\sqrt2\mathcal vg)_{ij}\bar u_i\gamma^\mu p_l\ell_j\,\nonumber\\[0.2 cm ] & \quad + u_{3\mu}^{(-1/3)}(\sqrt2g\mathcal u)_{ij}\bar d_i\gamma^\mu p_l\nu_j+{\rm h.c.}\ , , \end{aligned}\ ] ] where @xmath77 and @xmath78 represent the cabibbo - kobayashi - maskawa ( ckm ) @xcite and the pontecorvo - maki - nakagawa - sakata ( pmns ) @xcite matrix , respectively . here we assume the neutrinos to be massless and , therefore , the pmns matrix can be rotated away through field redefinitions . the vector multiplet @xmath63 can also mediate the @xmath51 transitions at tree level , and the resulting effective weak hamiltonian including the sm contribution can be written as @xcite @xmath79 where @xmath80 is the wilson coefficient at the matching scale @xmath81 and is given by @xmath82 unlike in the scalar lq case , the vector lq only generates @xmath64 couplings and , therefore , the wilson coefficient @xmath80 need not be renormalized . as shown in ref . @xcite , the vector lq scenario could also accommodate the @xmath7 , @xmath23 as well as the angular observable @xmath26 in @xmath83 decay . fitting to the measured ratios @xmath7 , along with acceptable @xmath29 spectra , two best - fit solutions , denoted as @xmath84 and @xmath85 , respectively , are found in this scenario @xcite : @xmath86 where @xmath87 is taken as a benchmark . it should be noted that the triplet nature of @xmath88 also leads to various charged lepton - flavour - violating decays , such as the @xmath89 and @xmath90 decays , which have been discussed in refs . @xcite . in this subsection , we give the helicity amplitudes for the process @xmath38 both within the sm and in the two lq scenarios . following refs . @xcite and starting with the effective weak hamiltonian given by eqs . and , one can get the helicity amplitudes of the decay . since all types of the leptonic helicity amplitudes can be found in ref . @xcite , we give only the hadronic helicity amplitudes . for the @xmath64-type current , we have @xcite @xmath91 where @xmath92 and @xmath93 denote the helicities of the daughter baryon @xmath5 and the effective ( axial-)vector - type current , respectively . the explicit expressions of @xmath94 in terms of the hadronic matrix elements defined by eqs . and could be found in ref . @xcite . for the @xmath95-type current , the corresponding helicity amplitudes are given by @xcite @xmath96 h^{sp}_{\pm\frac{1}{2},0}=&\frac{\sqrt{q_+}}{m_b - m_c}\left(f_1^vm_-+f_3^v\frac{q^2}{m_1}\right)\pm\frac{\sqrt{q_- } } { m_b+m_c } \left(f_1^am_+-f_3^a\frac{q^2}{m_1}\right)\,,\end{aligned}\ ] ] where we use the abbreviations @xmath97 and @xmath98 . the hadronic helicity amplitudes of the ( pseudo-)tensor - type current are defined as @xmath99 and their explicit expressions , in terms of the hadronic matrix elements defined by eqs . and , are given by @xmath100 h_{\frac{1}{2},+,-}^t=&-f_t\sqrt{q_+}-g_t\sqrt{q_-}\,,\nonumber\\[0.2 cm ] h_{\frac{1}{2},+,t}^t=&-\sqrt{\frac{2}{q^2}}\left(f_t\sqrt{q_-}m_++g_t\sqrt{q_+}m_-\right ) + \sqrt{2q^2}\left(f_t^v\sqrt{q_-}-g_t^v\sqrt{q_+}\right)\,,\nonumber\\[0.2 cm ] h_{\frac{1}{2},0,t}^t=&-f_t\sqrt{q_-}-g_t\sqrt{q_+}+f_t^v\sqrt{q_-}m_+-g_t^v\sqrt{q_+}m_- + f_t^s\sqrt{q_-}q_++g_t^s\sqrt{q_+}q_-\,,\nonumber\\[0.2 cm ] h_{-\frac{1}{2},+,-}^t=&f_t\sqrt{q_+}-g_t\sqrt{q_-}\,,\nonumber\\[0.2 cm ] h_{-\frac{1}{2},0,-}^t=&\sqrt{\frac{2}{q^2}}\left(f_t\sqrt{q_+}m_--g_t\sqrt{q_-}m_+\right)\,,\nonumber\\[0.2 cm ] h_{-\frac{1}{2},0,t}^t=&-f_t\sqrt{q_-}+g_t\sqrt{q_+}+f_t^v\sqrt{q_-}m_+ + g_t^v\sqrt{q_+}m_-+f_t^s\sqrt{q_-}q_+-g_t^s\sqrt{q_+}q_-\,,\nonumber\\[0.2 cm ] h_{-\frac{1}{2},-,t}^t=&-\sqrt{\frac{2}{q^2}}\left(f_t\sqrt{q_-}m_+-g_t\sqrt{q_+}m_-\right ) + \sqrt{2q^2}\left(f_t^v\sqrt{q_-}+g_t^v\sqrt{q_+}\right)\,.\end{aligned}\ ] ] the helicity amplitudes satisfy the relations @xmath101 and @xmath102 , while all the others are found to be zero . using the hqet relations given by eq . , we can further simplify these helicity amplitudes . here we follow the conventions used in refs . @xcite , and write the two - fold differential angular decay distribution as @xmath103\,,\ ] ] with @xmath104 a_1=&c_v^2\left[2\sin^2\theta_\ell\big(h_{\frac{1}{2},0}^2+h_{-\frac{1}{2},0}^2\big)+(1-\cos\theta_\ell)^2 h_{\frac{1}{2},+}^2+(1+\cos\theta_\ell)^2h_{-\frac{1}{2},-}^2\right]\,,\nonumber\\[0.2 cm ] a_2^v=&c_v^2\big[2\cos^2\theta_\ell\big(h_{\frac{1}{2},0}^2+h_{-\frac{1}{2},0}^2\big)+\sin^2\theta_\ell \big(h_{\frac{1}{2},+}^2+h_{-\frac{1}{2},-}^2\big)+2\big(h_{\frac{1}{2},t}^2+h_{-\frac{1}{2},t}^2\big)\nonumber\\[0.1 cm ] & \,\,\,\,-4\cos\theta_\ell \big(h_{\frac{1}{2},0}h_{\frac{1}{2},t}+h_{-\frac{1}{2},0 } h_{-\frac{1}{2},t}\big)\big ] \,,\nonumber\\[0.2 cm ] a_2^t=&\frac{c_t^2}{4}\big[2\sin^2\theta_\ell\big(h_{\frac{1}{2},+,-}^{t2}+h_{\frac{1}{2},0,t}^{t2}+ h_{-\frac{1}{2},+,-}^{t2}+h_{-\frac{1}{2},0,t}^{t2}+2h_{\frac{1}{2},+,-}^th_{\frac{1}{2},0,t}^t+ 2h_{-\frac{1}{2},+,-}^th_{-\frac{1}{2},0,t}^t\big)\nonumber\\[0.1 cm ] & \,\,\,\,+(1+\cos\theta_\ell)^2\big(h_{-\frac{1}{2},0,-}^{t2}+h_{-\frac{1}{2},-,t}^{t2}+ 2h_{-\frac{1}{2},0,-}^th_{-\frac{1}{2},-,t}^t\big)\nonumber\\[0.1 cm ] & \,\,\,\,+(1-\cos\theta_\ell)^2\big(h_{\frac{1}{2},+,0}^{t2}+h_{\frac{1}{2},+,t}^{t2}+ 2h_{\frac{1}{2},+,0}^th_{\frac{1}{2},+,t}^t\big)\big]\,,\nonumber\\[0.2 cm ] a_3=&\frac{c_t^2}{8}\big[2\cos^2\theta_\ell\big(h_{\frac{1}{2},+,-}^{t2}\!+\!h_{\frac{1}{2},0,t}^{t2 } + h_{-\frac{1}{2},+,-}^{t2}\!+h_{-\frac{1}{2},0,t}^{t2}+ 2h_{\frac{1}{2},+,-}^th_{\frac{1}{2},0,t}^t+2h_{-\frac{1}{2},+,-}^th_{-\frac{1}{2},0,t}^t\big)\nonumber\\[0.1 cm ] & \,\,\,\,+\sin^2\theta_\ell\big ( h_{\frac{1}{2},+,0}^{t2}\!+\!h_{\frac{1}{2},+,t}^{t2}+h_{-\frac{1}{2},0,-}^{t2}+h_{-\frac{1}{2},-,t}^{t2 } + 2h_{\frac{1}{2},+,0}^th_{\frac{1}{2},+,t}^t+2h_{-\frac{1}{2},0,-}^th_{-\frac{1}{2},-,t}^t\big)\big]\nonumber\\[0.1 cm ] + & c_s^2\big(h^{sp2}_{\frac{1}{2},0}+h^{sp2}_{-\frac{1}{2},0}\big)\,,\nonumber\\[0.2 cm ] a_4=&c_vc_s\big[-\cos\theta_\ell\big(h_{\frac{1}{2},0}h^{sp}_{\frac{1}{2},0}+h_{-\frac{1}{2},0}h^{sp}_{-\frac{1}{2},0}\big ) + \big(h_{\frac{1}{2},t}h^{sp}_{\frac{1}{2},0}+h_{-\frac{1}{2},t}h^{sp}_{-\frac{1}{2},0}\big)\big]\nonumber\\[0.1 cm ] + & c_vc_t\big[\frac{\cos^2\theta}{2}\big(h_{\frac{1}{2},0}h^t_{\frac{1}{2},+,-}+ h_{\frac{1}{2},0}h^t_{\frac{1}{2},0,t}+h_{-\frac{1}{2},0}h^t_{-\frac{1}{2},+,-}+ h_{-\frac{1}{2},0}h^t_{-\frac{1}{2},0,t}\big)\nonumber\\[0.1 cm ] & \hspace{0.9cm}-\frac{\cos\theta}{2}\big(h_{\frac{1}{2},t}h^t_{\frac{1}{2},+,-}+ h_{\frac{1}{2},t}h^t_{\frac{1}{2},0,t}+h_{-\frac{1}{2},t}h^t_{-\frac{1}{2},+,-}+ h_{-\frac{1}{2},t}h^t_{-\frac{1}{2},0,t}\big)\nonumber\\[0.1 cm ] & \hspace{0.9cm}+\frac{(1-\cos\theta)^2}{4}\big(h_{\frac{1}{2},+}h^t_{\frac{1}{2},+,0 } + h_{\frac{1}{2},+}h^t_{\frac{1}{2},+,t}\big)\nonumber\\[0.1 cm ] & \hspace{0.9cm}+\frac{(1+\cos\theta)^2}{4}\big(h_{-\frac{1}{2},-}h^t_{-\frac{1}{2},0,- } + h_{-\frac{1}{2},-}h^t_{-\frac{1}{2},-,t}\big)\nonumber\\[0.1 cm ] & \hspace{0.9cm}+\frac{\sin^2\theta}{4}\big(h_{\frac{1}{2},+}h^t_{\frac{1}{2},+,0}+ h_{\frac{1}{2},+}h^t_{\frac{1}{2},+,t}+h_{-\frac{1}{2},-}h^t_{-\frac{1}{2},0,-}+ h_{-\frac{1}{2},-}h^t_{-\frac{1}{2},-,t}\nonumber\\[0.1 cm ] & \hspace{1.9cm}+2h_{\frac{1}{2},0}h^t_{\frac{1}{2},+,-}+2h_{\frac{1}{2},0}h^t_{\frac{1}{2},0,t } + 2h_{-\frac{1}{2},0}h^t_{-\frac{1}{2},+,-}+2h_{-\frac{1}{2},0}h^t_{-\frac{1}{2},0,t}\big)\big]\,,\nonumber\\[0.2 cm ] a_5=&-2c_sc_t\cos\theta\big(h^{sp}_{\frac{1}{2},0}h^t_{\frac{1}{2},+,-}+h^{sp}_{\frac{1}{2},0 } h^t_{\frac{1}{2},0,t}+h^{sp}_{-\frac{1}{2},0}h^t_{-\frac{1}{2},+,- } + h^{sp}_{-\frac{1}{2},0}h^t_{-\frac{1}{2},0,t}\big)\,,\end{aligned}\ ] ] where @xmath105 is the @xmath5 momentum in the @xmath31 rest frame , @xmath29 the momentum transfer squared , and @xmath106 the polar angle of the lepton , as shown in fig . 1 of ref . @xcite . integrating out @xmath107 in eq . , one can then obtain the differential decay rate @xmath108 . with eqs . and at hand , we can get the following physical observables : * the differential and total branching fractions @xmath109 where @xmath110 is the lifetime of @xmath31 baryon , and @xmath111 the lepton mass . * the differential and integrated ratios @xmath112 * the lepton - side forward - backward asymmetry @xmath113 defined as the relative difference between the differential decay rates where the angle @xmath106 is smaller or greater than @xmath114 . once the individual helicity - dependent differential decay rates are calculated , which are collected in appendix [ sec : helicity rates ] , we can obtain another two observables , the @xmath29-dependent longitudinal polarizations of @xmath5 baryon and @xmath6 lepton , which are defined , respectively , as @xmath115 the above results are given for the scalar lq scenario . for the vector lq case , we need only to replace @xmath53 by @xmath80 given by eq . , while setting @xmath54 and @xmath55 to zero . in this section , we investigate the scalar and vector lq effects on the aforementioned observables , to see if their effects are large enough to cause sizable deviations from the corresponding sm predictions . firstly , we collect in table [ tab : inputs ] all the input parameters used in this paper . 0.25 in .[tab : inputs ] input parameters used in our numerical analyses . [ cols="^,^,^",options="header " , ] from the numerical results given in tables [ tab : branching ratios ] and [ tab : lattice results ] , we can draw the following conclusions : * the branching fractions are very sensitive to the form - factor parameterizations used in qcd sum rules @xcite . the triangular region ( with @xmath116 or @xmath117 ) for the continuum model gives more reliable predictions compared to the rectangular one , because , within the sm , the former leads to consistent results with that obtained using the lattice - based form factors , and also with the experimental data @xmath118 @xcite . thus , from now on , we consider only the triangular continuum model with two different values of @xmath119 . the ratio @xmath41 , on the other hand , is insensitive to the form - factor parameterizations , as is generally expected . it is also noted that the predicted @xmath41 using the lattice - based form factors is a little bit larger than that obtained from qcd sum rules , both within the sm and in the two lq scenarios . * in the scalar lq scenario , the predicted branching fraction @xmath120 is enhanced by about @xmath121 ( @xmath122 ) in the @xmath60 ( @xmath61 ) case , no matter the form factors are taken from qcd sum rules or from the lattice qcd calculations . the slight difference between the two solutions results from the @xmath123 numerical difference in the dominated coefficient @xmath124 , which has been discussed in ref . as the decay modes with light leptons ( @xmath11 ) are assumed to be free from the scalar lq contribution , the ratio @xmath41 is also enhanced by the corresponding percentages relative to the sm prediction . * in the vector lq scenario , compared to the sm prediction , the branching fraction @xmath120 is found to be enhanced by about @xmath121 in the @xmath84 and by about @xmath125 in the @xmath85 case , respectively . to understand this , we should note that there is only one @xmath64 coupling in this scenario , and the resulting effective coefficients @xmath80 , corresponding to the two solutions @xmath84 and @xmath85 ( cf . eq . ) , are given , respectively , as @xmath126 one can see clearly that , just like @xmath127 in scalar lq scenario , @xmath128 also has nearly the same absolute values for the two solutions @xmath84 and @xmath85 , both enhancing the sm result by @xmath129 , but the sign of solution @xmath85 is flipped relative to the sm part . * as the two effective couplings @xmath124 and @xmath130 are both enhanced by about @xmath131 , compared to the sm part , they would give quite similar predictions for the other observables in @xmath2 decay . the @xmath29 dependences of the differential branching fraction @xmath132 and the ratio @xmath133 are displayed in fig . [ fig : dbr ] , both within the sm and in the two lq scenarios . as the results based on the form - factor parametrizations with @xmath116 are similar to that with @xmath117 , we show only the case with @xmath116 . one can see that these two observables present the same features as the corresponding @xmath29-integrated ones discussed above : the predicted @xmath132 using the lattice - based form factors are a little bit larger than that based on qcd sum rules , and are enhanced at @xmath134 at most in both the two lq scenarios . however , the ratio @xmath133 is insensitive to the choices of the form factors . + finally , we show in fig . [ fig : polarizations ] the @xmath29 dependences of the @xmath5 and @xmath6 longitudinal polarizations as well as the lepton - side forward - backward asymmetry . since the resulting effective coefficients @xmath53 and @xmath135 of the dominated @xmath64 couplings appear both in the numerator and in the denominator of these ratios , the np effects are cancelled exclusively . at the same time , the form - factor dependences of these observables are reduced to a large extent , and all the four cases in qcd sum rules ( @xmath136 ) give almost the same curves for each observable , while being only slightly different from that obtained with the lattice - based form factors , as shown in fig . [ fig : polarizations ] . as a consequence , all these three observables are insensitive to the two lq scenarios and behave nearly the same as in the sm . as demonstrated in refs . @xcite , both the scalar and vector lq scenarios could explain the anomalies observed in @xmath0 and @xmath1 decays and , for each scenario , there exist two best - fit solutions for the operator coefficients , because the other two solutions for scalar lq are already excluded by the @xmath137 decay @xcite . to further explore these two interesting scenarios , in this paper , we have studies their effects in the semi - leptonic @xmath2 decay , which is induced by the same quark - level transition as in @xmath0 decays . besides the branching fraction @xmath3 and the ratio @xmath41 , we have also discussed the @xmath29 distributions of these two observables , as well as the @xmath5 and @xmath6 longitudinal polarizations and the lepton - side forward - backward asymmetry , using the @xmath40 transition form factors from both the qcd sum rules and the latest lattice qcd calculations . using the best - fit solutions for the operator coefficients allowed by the current data of mesonic decays , we have found that the two lq scenarios give the similar amounts of enhancements relative to the sm predictions for the branching fraction @xmath3 and the ratio @xmath41 . the two best - fit solutions in each of the two scenarios are still found to be indistinguishable from each other based only on these two observables . on the other hand , both of the two lq scenarios give nearly the same predictions as the sm for the @xmath5 and @xmath6 longitudinal polarizations , as well as the lepton - side forward - backward asymmetry . as a consequence , we conclude that , while the two lq scenarios could be distinguished from the sm , it is quite difficult to distinguish between them using the semi - leptonic @xmath2 decay . with the large number of @xmath31 produced at the lhc , we expect that , with future measurements of the observables in @xmath2 decay at the lhcb , the two lq scenarios could be further tested , and even differentiated from the other np explanations to the @xmath7 anomalies . the work is supported by the national natural science foundation of china ( nsfc ) under contract nos . 11675061 , 11435003 , 11225523 and 11221504 . xl is also supported by the scientific research foundation for the returned overseas chinese scholars , state education ministry , and by the self - determined research funds of ccnu from the colleges basic research and operation of moe ( ccnu15a02037 ) . xz is supported by the ccnu - qlpl innovation fund ( qlpl2015p01 ) . xl acknowledges the hospitality of the munich institute for astro- and particle physics ( miapp ) of the dfg cluster of excellence `` origin and structure of the universe '' , where this work was finalized . the hadronic matrix elements of the vector and axial - vector currents between the two spin - half baryons @xmath31 and @xmath5 can be parameterized in terms of three form factors , respectively , as @xcite @xmath139u_1(p_1,\lambda_1)\,,\label{eq : vector ffs}\\[0.2 cm ] \langle \lambda_c,\lambda_2|\bar c\gamma_\mu\gamma_5 b|\lambda_b,\lambda_1\rangle=&\bar u_2(p_2,\lambda_2)\left[f_1^a(q^2)\gamma_\mu-\frac{f_2^a(q^2)}{m_{\lambda_b}}i\sigma_{\mu\nu}q^\nu+\frac{f_3^a(q^2 ) } { m_{\lambda_b}}q_\mu\right]\gamma_5u_1(p_1,\lambda_1)\,,\label{eq : axial - vector ffs}\end{aligned}\ ] ] where @xmath140 , @xmath141 is the four - momentum transfer , and @xmath142 denote the helicities of the @xmath31 and @xmath5 baryons , respectively . using the equations of motion , we can then obtain the hadronic matrix elements of the scalar and pseudo - scalar currents between the two baryons , which are given , respectively , as @xmath143u_1(p_1,\lambda_1)\,,\label{eq : scalar ffs}\\[0.2 cm ] \langle \lambda_c,\lambda_2|\bar c\gamma_5b|\lambda_b,\lambda_1\rangle=&\frac{1}{m_b+m_c}\bar u_2(p_2,\lambda_2)\left[f_1^a(q^2)(m_1+m_2)-\frac{f_3^a(q^2 ) } { m_{\lambda_b}}q^2\right]\gamma_5u_1(p_1,\lambda_1)\,,\label{eq : pseudo - scalar ffs}\end{aligned}\ ] ] where @xmath144 and @xmath145 are the current quark masses evaluated at the scale @xmath146 . the hadronic matrix elements of the tensor and pseudo - tensor currents between the @xmath31 and @xmath5 baryons can be generally parameterized as @xcite @xmath147 u_1(p_1,\lambda_1)\,,\label{eq : tensor ffs}\\ \langle \lambda_c,\lambda_2|\bar ci\sigma_{\mu\nu}\gamma_5b|\lambda_b,\lambda_1\rangle=&\bar u_2(p_2,\lambda_2 ) \left[g_t i\sigma_{\mu\nu } + g_t^v ( \gamma_\mu q_\nu-\gamma_\nu q_\mu ) + g_t^s ( p_\mu q_\nu - p_\nu q_\mu)\right ] \gamma_5 u_1(p_1,\lambda_1)\,,\label{eq : pseudo - tensor ffs}\end{aligned}\ ] ] where @xmath148 . the @xmath138 transition form factors have also been studies based on the hqet @xcite , and the following relations among the form factors can be found @xcite : @xmath149 & f_2^v = f_2^a =- f_3^v =- f_3^a\,,\\[0.2 cm ] & f_t^v\;=g_t^v = f_t^s = g_t^s=0\,.\end{aligned}\ ] ] an alternate helicity - based definition of the @xmath138 form factors can be found in ref . @xcite , and the explicit relations between these two sets of form factors are also given in ref . @xcite . in this paper , we use the results obtained both in the qcd sum rules @xcite and in the most recent lattice qcd calculation with @xmath150 dynamical flavours @xcite . however , since there are currently no lattice results for the ( pseudo-)tensor form factors yet , we still use the following hqet relations to relate them to the corresponding ( axial-)vector ones , @xmath151 in order to discuss the @xmath5 and @xmath6 polarizations , we need the helicity - dependent differential decay rates , which are collected below ( normalized by the prefactor @xmath152 defined in eq . ): @xmath153 & + 2h^t_{\frac{1}{2},+,-}h^t_{\frac{1}{2},0,t}+2h^t_{\frac{1}{2},+,0}h^t_{\frac{1}{2},+,t}\big)\big ] + \frac{8}{3}c_v^2\big(h_{\frac{1}{2},0}^2+h_{\frac{1}{2},+}^2\big)+4c_s^2h^{sp2}_{\frac{1}{2},0}\nonumber\\[0.1 cm ] & + \frac{c_t^2}{3}\big(h^{t2}_{\frac{1}{2},+,-}+h^{t2}_{\frac{1}{2},0,t}+h^{t2}_{\frac{1}{2},+,0 } + h^{t2}_{\frac{1}{2},+,t}+2h^t_{\frac{1}{2},+,-}h^t_{\frac{1}{2},0,t}+2h^t_{\frac{1}{2},+,0}h^t_{\frac{1}{2},+,t } \big)\nonumber\\[0.1 cm ] + & \frac{4m_\ell}{\sqrt{q^2}}\big[c_vc_t\big(h_{\frac{1}{2},0}h^t_{\frac{1}{2},+,- } + h_{\frac{1}{2},0}h^t_{\frac{1}{2},0,t}+h_{\frac{1}{2},+}h^t_{\frac{1}{2},+,0 } + h_{\frac{1}{2},+}h^t_{\frac{1}{2},+,t}\big)\nonumber\\[0.1 cm ] & + 2c_vc_s\big(h_{\frac{1}{2},t}h^{sp}_{\frac{1}{2},0}\big)\big]\,,\nonumber\\[0.2 cm ] \frac{{\rm d}\gamma^{\lambda_2=-1/2}}{{\rm d}q^2}=&\frac{m_\ell^2}{q^2}\big[\frac{4}{3}c_v^2\big(h_{-\frac{1}{2},-}^2\!+\!h_{-\frac{1}{2},0}^2 + 3h_{-\frac{1}{2},t}^2\big)\!+\!\frac{2}{3}c_t^2\big(h^{t2}_{-\frac{1}{2},+,-}\!+\!h^{t2}_{-\frac{1}{2},0,t}\ ! + \!h^{t2}_{-\frac{1}{2},0,-}\!+\!h^{t2}_{-\frac{1}{2},-,t}\nonumber\\[0.1 cm ] & + 2h^t_{-\frac{1}{2},+,-}h^t_{-\frac{1}{2},0,t}+2h^t_{-\frac{1}{2},0,-}h^t_{-\frac{1}{2},-,t}\big)\big ] + \frac{8}{3}c_v^2\big(h_{-\frac{1}{2},-}^2+h_{-\frac{1}{2},0}^2\big)+4c_s^2h^{sp2}_{-\frac{1}{2},0}\nonumber\\[0.1 cm ] & + \frac{c_t^2}{3}\big(h^{t2}_{-\frac{1}{2},+,-}\!+\!h^{t2}_{-\frac{1}{2},0,t}\!+\ ! h^{t2}_{-\frac{1}{2},0,-}\!+\!h^{t2}_{-\frac{1}{2},-,t}\ ! + \!2h^t_{-\frac{1}{2},+,-}h^t_{-\frac{1}{2},0,t}\!+\!2h^t_{-\frac{1}{2},0,-}h^t_{-\frac{1}{2},-,t}\big)\nonumber\\[0.1 cm ] + & \frac{4m_\ell}{\sqrt{q^2}}\big[c_vc_t\big(h_{-\frac{1}{2},0}h^t_{-\frac{1}{2},+,- } + h_{-\frac{1}{2},0}h^t_{-\frac{1}{2},0,t}+h_{-\frac{1}{2},-}h^t_{-\frac{1}{2},0,- } + h_{-\frac{1}{2},-}h^t_{-\frac{1}{2},-,t}\big)\nonumber\\[0.1 cm ] & + 2c_vc_s\big(h_{-\frac{1}{2},t}h^{sp}_{-\frac{1}{2},0}\big)\big]\,,\nonumber\\[0.2 cm ] \frac{{\rm d}\gamma^{\lambda_\tau=1/2}}{{\rm d}q^2}=&\frac{m_\ell^2}{q^2}c_v^2\big[\frac{4}{3}\big(h_{\frac{1}{2},+}^2\!+\!h_{\frac{1}{2},0}^2\!+\ ! h_{-\frac{1}{2},-}^2\!+\!h_{-\frac{1}{2},0}^2\big)\!+\!3\big(h_{\frac{1}{2},t}^2\!+\!h_{-\frac{1}{2},t}^2\big)\big ] + 4c_s^2\big(h^{sp2}_{\frac{1}{2},0}+h^{sp2}_{-\frac{1}{2},0}\big)\nonumber\\[0.1 cm ] & + \frac{c_t^2}{3}\big[h^{t2}_{\frac{1}{2},+,-}\!+\!h^{t2}_{\frac{1}{2},0,t}+h^{t2}_{\frac{1}{2},+,0}+ h^{t2}_{\frac{1}{2},+,t}+h^{t2}_{-\frac{1}{2},+,-}+h^{t2}_{-\frac{1}{2},0,t}+h^{t2}_{-\frac{1}{2},0,-}\!+ h^{t2}_{-\frac{1}{2},-,t}\nonumber\\[0.1 cm ] & + 2\big(h^t_{\frac{1}{2},+,-}h^t_{\frac{1}{2},0,t}+h^t_{\frac{1}{2},+,0}h^t_{\frac{1}{2},+,t}+ h^t_{-\frac{1}{2},+,-}h^t_{-\frac{1}{2},0,t}+h^t_{-\frac{1}{2},0,-}h^t_{-\frac{1}{2},-,t}\big)\big]\nonumber\\[0.1 cm ] + & \frac{4m_\ell}{3\sqrt{q^2}}\big[6c_vc_s\big(h_{\frac{1}{2},t}h^{sp}_{\frac{1}{2},0}+h_{-\frac{1}{2},t } h^{sp}_{-\frac{1}{2},0}\big)+c_vc_t\big(h_{\frac{1}{2},0}h^t_{\frac{1}{2},+,-}+ h_{\frac{1}{2},0}h^t_{\frac{1}{2},0,t}\nonumber\\[0.1 cm ] & + h_{\frac{1}{2},+}h^t_{\frac{1}{2},+,0}+h_{\frac{1}{2},+}h^t_{\frac{1}{2},+,t}+ h_{-\frac{1}{2},0}h^t_{-\frac{1}{2},+,-}+h_{-\frac{1}{2},0}h^t_{-\frac{1}{2},0,t}+ h_{-\frac{1}{2},-}h^t_{-\frac{1}{2},0,-}\nonumber\\[0.1 cm ] & + h_{-\frac{1}{2},-}h^t_{-\frac{1}{2},-,t}\big)\big]\,,\nonumber\\[0.2 cm ] \frac{{\rm d}\gamma^{\lambda_\tau=-1/2}}{{\rm d}q^2}=&\frac{8c_v^2}{3}\big(h_{\frac{1}{2},+}^2\!+\!h_{\frac{1}{2},0}^2\!+\!h_{-\frac{1}{2},-}^2\!+ \!h_{-\frac{1}{2},0}^2\big)+\!\frac{2m_\ell^2}{3q^2}c_t^2\big[h^{t2}_{\frac{1}{2},+,-}\!+\!h^{t2}_{\frac{1}{2},0,t}\!+ \!h^{t2}_{\frac{1}{2},+,0}\!+\!h^{t2}_{\frac{1}{2},+,t}\!\nonumber\\[0.1 cm ] & + \!h^{t2}_{-\frac{1}{2},+,-}\!+\!h^{t2}_{-\frac{1}{2},0,t } + h^{t2}_{-\frac{1}{2},0,-}\!+\!h^{t2}_{-\frac{1}{2},-,t}+2\big(h^t_{\frac{1}{2},+,-}h^t_{\frac{1}{2},0,t } + h^t_{\frac{1}{2},+,0}h^t_{\frac{1}{2},+,t}\nonumber\\[0.1 cm ] & + h^t_{-\frac{1}{2},+,-}h^t_{-\frac{1}{2},0,t}+h^t_{-\frac{1}{2},0,- } h^t_{-\frac{1}{2},-,t}\big)\big]\nonumber\\[0.1 cm ] + & \frac{8m_\ell}{3\sqrt{q^2}}c_vc_t\big(h_{\frac{1}{2},0}h^t_{\frac{1}{2},+,-}+ h_{\frac{1}{2},0}h^t_{\frac{1}{2},0,t}+h_{\frac{1}{2},+}h^t_{\frac{1}{2},+,0}+ h_{\frac{1}{2},+}h^t_{\frac{1}{2},+,t}\nonumber\\[0.1 cm ] & + h_{-\frac{1}{2},0}h^t_{-\frac{1}{2},+,-}+h_{-\frac{1}{2},0}h^t_{-\frac{1}{2},0,t}+ h_{-\frac{1}{2},-}h^t_{-\frac{1}{2},0,-}+h_{-\frac{1}{2},-}h^t_{-\frac{1}{2},-,t}\big)\,.\end{aligned}\ ] ] a. crivellin , _ new physics in the flavour sector _ , in _ 51st rencontres de moriond on qcd and high energy interactions la thuile , italy , march 19 - 26 , 2016 _ , 2016 . http://arxiv.org/abs/1605.02934[arxiv:1605.02934 ] . z. ligeti , _ flavor constraints on new physics _ , in _ 27th international symposium on lepton photon interactions at high energy ( lp15 ) ljubljana , slovenia , august 17 - 22 , 2015 _ , 2016 . http://arxiv.org/abs/1606.02756[arxiv:1606.02756 ] . collaboration , j. p. lees et al . , _ measurement of an excess of @xmath155 decays and implications for charged higgs bosons _ , _ phys . * d88 * ( 2013 ) , no . 7 072012 , [ http://arxiv.org/abs/1303.0571[arxiv:1303.0571 ] ] . collaboration , m. huschle et al . , _ measurement of the branching ratio of @xmath156 relative to @xmath157 decays with hadronic tagging at belle _ , _ phys . * d92 * ( 2015 ) , no . 7 072014 , [ http://arxiv.org/abs/1507.03233[arxiv:1507.03233 ] ] . collaboration , a. abdesselam et al . , _ measurement of the branching ratio of @xmath158 relative to @xmath159 decays with a semileptonic tagging method _ , http://arxiv.org/abs/1603.06711[arxiv:1603.06711 ] . collaboration , r. aaij et al . , _ measurement of the ratio of branching fractions @xmath161 _ , _ phys . * 115 * ( 2015 ) , no . 11 111803 , [ http://arxiv.org/abs/1506.08614 [ arxiv:1506.08614 ] ] . [ addendum : phys . lett.115,no.15,159901(2015 ) ] . collaboration , h. na , c. m. bouchard , g. p. lepage , c. monahan , and j. shigemitsu , _ @xmath162 form factors at nonzero recoil and extraction of @xmath163 _ , _ phys . * d92 * ( 2015 ) , no . 5 054510 , [ http://arxiv.org/abs/1505.03925[arxiv:1505.03925 ] ] . m. bauer and m. neubert , _ minimal leptoquark explanation for the @xmath168 , @xmath23 , and @xmath24 anomalies _ , _ phys . * 116 * ( 2016 ) , no . 14 141802 , [ http://arxiv.org/abs/1511.01900 [ arxiv:1511.01900 ] ] . li , y .- d . yang , and x. zhang , _ revisiting the one leptoquark solution to the @xmath170 anomalies and its phenomenological implications _ , _ jhep _ * 08 * ( 2016 ) 054 , [ http://arxiv.org/abs/1605.09308[arxiv:1605.09308 ] ] . t. gutsche , m. a. ivanov , j. g. korner , v. e. lyubovitskij , p. santorelli , and n. habyl , _ semileptonic decay @xmath173 in the covariant confined quark model _ _ * d91 * ( 2015 ) , no . 7 074001 , [ http://arxiv.org/abs/1502.04864 [ arxiv:1502.04864 ] ] . [ erratum : phys . rev.d91,no.11,119907(2015 ) ] . w. detmold , c. lehner , and s. meinel , _ @xmath174 and @xmath175 form factors from lattice qcd with relativistic heavy quarks _ , _ phys . * d92 * ( 2015 ) , no . 3 034503 , [ http://arxiv.org/abs/1503.01421 [ arxiv:1503.01421 ] ] . r. s. marques de carvalho , f. s. navarra , m. nielsen , e. ferreira , and h. g. dosch , _ form - factors and decay rates for heavy lambda semileptonic decays from qcd sum rules _ , _ phys * d60 * ( 1999 ) 034009 , [ http://arxiv.org/abs/hep-ph/9903326[hep-ph/9903326 ] ] . a. datta , m. duraisamy , and d. ghosh , _ diagnosing new physics in @xmath187 decays in the light of the recent babar result _ * d86 * ( 2012 ) 034027 , [ http://arxiv.org/abs/1206.3760[arxiv:1206.3760 ] ] . s. fajfer , j. f. kamenik , i. nisandzic , and j. zupan , _ implications of lepton flavor universality violations in b decays _ , _ phys . _ * 109 * ( 2012 ) 161801 , [ http://arxiv.org/abs/1206.1872 [ arxiv:1206.1872 ] ] . y. sakaki and h. tanaka , _ constraints on the charged scalar effects using the forward - backward asymmetry on @xmath192 _ , _ phys . rev . _ * d87 * ( 2013 ) , no . 5 054002 , [ http://arxiv.org/abs/1205.4908[arxiv:1205.4908 ] ] . a. crivellin , a. kokulu , and c. greub , _ flavor - phenomenology of two - higgs - doublet models with generic yukawa structure _ , _ phys . * d87 * ( 2013 ) , no . 9 094031 , [ http://arxiv.org/abs/1303.5877 [ arxiv:1303.5877 ] ] . r. dutta , a. bhol , and a. k. giri , _ effective theory approach to new physics in @xmath194 and @xmath195 leptonic and semileptonic decays _ , _ phys . * d88 * ( 2013 ) , no . 11 114023 , [ http://arxiv.org/abs/1307.6653[arxiv:1307.6653 ] ] . fan , w .- f . wang , s. cheng , and z .- j . xiao , _ semileptonic decays @xmath198 in the perturbative qcd factorization approach _ , _ chin . * 59 * ( 2014 ) 125132 , [ http://arxiv.org/abs/1301.6246[arxiv:1301.6246 ] ] . b. bhattacharya , a. datta , d. london , and s. shivashankara , _ simultaneous explanation of the @xmath23 and @xmath199 puzzles _ , _ phys . _ * b742 * ( 2015 ) 370374 , [ http://arxiv.org/abs/1412.7164 [ arxiv:1412.7164 ] ] . y. sakaki , m. tanaka , a. tayduganov , and r. watanabe , _ probing new physics with @xmath29 distributions in @xmath203 _ , _ phys . * d91 * ( 2015 ) , no . 11 114028 , [ http://arxiv.org/abs/1412.3761[arxiv:1412.3761 ] ] . r. barbieri , g. isidori , a. pattori , and f. senia , _ anomalies in @xmath8-decays and @xmath204 flavour symmetry _ , _ eur . j. _ * c76 * ( 2016 ) , no . 2 67 , [ http://arxiv.org/abs/1512.01560 [ arxiv:1512.01560 ] ] . s. bhattacharya , s. nandi , and s. k. patra , _ optimal - observable analysis of possible new physics in @xmath205 _ , _ phys . _ * d93 * ( 2016 ) , no . 3 034011 , [ http://arxiv.org/abs/1509.07259[arxiv:1509.07259 ] ] . l. calibbi , a. crivellin , and t. ota , _ effective field theory approach to @xmath206 , @xmath207 and @xmath208 with third generation couplings _ , _ phys * 115 * ( 2015 ) 181801 , [ http://arxiv.org/abs/1506.02661 [ arxiv:1506.02661 ] ] . a. crivellin , j. heeck , and p. stoffer , _ a perturbed lepton - specific two - higgs - doublet model facing experimental hints for physics beyond the standard model _ , _ phys . * 116 * ( 2016 ) , no . 8 081801 , [ http://arxiv.org/abs/1507.07567[arxiv:1507.07567 ] ] . r. alonso , b. grinstein , and j. martin camalich , _ lepton universality violation and lepton flavor conservation in @xmath8-meson decays _ , _ jhep _ * 10 * ( 2015 ) 184 , [ http://arxiv.org/abs/1505.05164 [ arxiv:1505.05164 ] ] . c. hati , g. kumar , and n. mahajan , _ @xmath211 excesses in alrsm constrained from @xmath8 , @xmath212 decays and @xmath213 mixing _ , _ jhep _ * 01 * ( 2016 ) 117 , [ http://arxiv.org/abs/1511.03290[arxiv:1511.03290 ] ] . m. a. ivanov , j. g. krner , and c .- t . tran , _ analyzing new physics in the decays @xmath218 with form factors obtained from the covariant quark model _ , http://arxiv.org/abs/1607.02932[arxiv:1607.02932 ] . j. zhu , h .- wang , y .- y . fan , q. chang , and y .- g . xu , _ probing the r - parity violating supersymmetric effects in the exclusive @xmath222 decays _ * d93 * ( 2016 ) , no . 9 094023 , [ http://arxiv.org/abs/1602.06491[arxiv:1602.06491 ] ] . d. a. faroughy , a. greljo , and j. f. kamenik , _ confronting lepton flavor universality violation in b decays with high-@xmath228 tau lepton searches at lhc _ , http://arxiv.org/abs/1609.07138[arxiv:1609.07138 ] . b. bhattacharya , a. datta , j .- guvin , d. london , and r. watanabe , _ simultaneous explanation of the @xmath23 and @xmath168 puzzles : a model analysis _ , http://arxiv.org/abs/1609.09078[arxiv:1609.09078 ] . c. bourrely , i. caprini , and l. lellouch , _ model - independent description of @xmath233 decays and a determination of @xmath234 _ , _ phys . _ * d79 * ( 2009 ) 013008 , [ http://arxiv.org/abs/0807.2722 [ arxiv:0807.2722 ] ] . [ erratum : phys . rev.d82,099902(2010 ) ] . t. feldmann and m. w. y. yip , _ form factors for @xmath236 transitions in scet _ , _ phys . * d85 * ( 2012 ) 014035 , [ http://arxiv.org/abs/1111.1844[arxiv:1111.1844 ] ] . [ erratum : phys . rev.d86,079901(2012 ) ] .

let be a hamiltonian vector field defined on a symplectic manifold , a nowhere vanishing smooth function defined on an open dense subset of . we will say that the vector field is _ conformally hamiltonian_. we prove that when is complete , when is hamiltonian with respect to another symplectic form defined on , and when another technical condition is satisfied , then there is a symplectic diffeomorphism from onto an open subset of , which maps each orbit to itself and is equivariant with respect to the flows of the vector fields on and on . this result explains why the diffeomorphism of the phase space of the kepler problem restricted to the negative ( resp . positive ) values of the energy function , onto an open subset of the cotangent bundle to a three - dimensional sphere ( resp . two - sheeted hyperboloid ) , discovered by gyrgyi ( 1968 ) , re - discovered by ligon and schaaf ( 1976 ) , is a symplectic diffeomorphism . cushman and duistermaat ( 1997 ) have shown that the gyrgyi - ligon - schaaf diffeomorphism is characterized by three very natural properties ; here that diffeomorphism is obtained by composition of the diffeomorphism given by our result about conformally hamiltonian vector fields with a ( non - symplectic ) diffeomorphism built by a variant of moser s method . infinitesimal symmetries of the kepler problem are discussed , and it is shown that their space is a lie algebroid with zero anchor map rather than a lie algebra . charles - michel marle ( communicated by the associate editor name )

1011.5731

i am very glad to submit a paper for the special issue of the journal of geometric mechanics in honour of tudor ratiu . i followed his scientific work for several years ; i specially praise the superb book he wrote with juan - pablo ortega @xcite . i hope that he will find some interest in the present work . the kepler problem is a completely integrable hamiltonian system with important applications in the physical world : it provides a very accurate model of the motion of planets in the solar system , and its quantized equivalent provides a good model of the hydrogen atom . moreover , several features of the kepler problem make it very interesting for the mathematician : some of its motions ( those in which there is a collision of the moving point with the attractive centre ) are not defined for all times , but the system can be regularized , _ i.e. _ mapped into a new hamitonian system whose motions are defined for all times ; the lie algebra of infinitesimal symmetries of a given energy level of the phase space depends on that energy level . in section [ conformallyfields ] some results in symplectic geometry are presented . these results will be used in section [ kepler ] to explain why a diffeomorphism between the phase space of the kepler problem , restricted to negative values ( resp . positive values ) of the energy , and an open subset of the cotangent space to a 3-dimensional sphere ( resp . , a 3-dimensional two - sheeted hyperboloid ) is symplectic . that diffeomorphism was discovered by gyrgyi @xcite ( 1968 ) , re - discovered by ligon and schaaf @xcite ( 1976 ) and discussed by cushman and duistermaat @xcite ( 1997 ) , who have shown that it is characterized by the three very natural properties : 1 . it maps the set of points in the phase space of the kepler problem where the energy is negative ( resp . , positive ) onto the tangent bundle of the @xmath9-sphere ( resp . , the two - sheeted @xmath9-dimensional revolution hyperboloid ) with its zero section removed . 2 . it intertwines the kepler and delaunay vector fields ( a rescaling of the geodesic vector field on the @xmath9-sphere , or on the @xmath9-dimensional hyperboloid ) . 3 . it intertwines the @xmath10-momentum mappings of the kepler and delauney vector fields . we will see that the remarkable properties of that diffeomorphism appear as very natural consequences of the results presented in section [ conformallyfields ] . we will also discuss the weak infinitesimal symmetries of the kepler problem , and we will show that their set is a lie algebroid with a zero anchor map , rather than a lie algebra . after completion and submission of the present paper , the work of gert heckman and tim de laat @xcite , recently posted on arxiv , was indicated to us . the method used by these authors to explain the properties of the gyrgyi - ligon - schaaf diffeomorphism rests on the same ideas as ours . we also learnt that conformally hamiltonian vector fields were used , in the theory of bi - hamiltonian vector fields , by a.j . maciejewski , m. prybylska and a.v . tsiganov @xcite . unless another assumption is explicitly stated , all manifolds , functions , applications , vector fields and differential forms considered in this work will be assumed to be smooth , _ i.e _ differentiable of class @xmath11 . [ flow ] let @xmath0 be a vector field on a manifold @xmath4 . the _ differential equation determined by _ @xmath0 is the ordinary differential equation @xmath12 the _ flow _ of that equation is the map @xmath13 , defined on an open subset @xmath14 of @xmath15 , with values in @xmath4 , such that , for each @xmath16 , the parametrized curve @xmath17 is the maximal solution @xmath18 of the above differential equation which takes the value @xmath19 for @xmath20 . it means that @xmath21 and that the open subset @xmath14 of @xmath15 is such that , for each @xmath16 @xmath22 is the largest open interval of @xmath23 on which a solution @xmath18 of the differential equation determined by @xmath0 satisfying @xmath24 can be defined . [ partial flow ] the map @xmath13 is sometimes called _ partial flow _ to distinguish it from the _ full flow _ @xmath25 , which is the map defined on an open subset of @xmath26 , with values in @xmath4 , such that , for each @xmath27 and @xmath28 , the parametrized curve @xmath29 is the maximal solution of the above differential equation which takes the value @xmath30 for @xmath31 . the vector field @xmath0 being time - independent , we have @xmath32 , so the full flow is determined by the partial flow . it is no longer true for ordinary differential equations determined by a time - dependent vector field ; the full flow should then be used instead of the partial flow . on a manifold @xmath4 , let @xmath0 be a vector field and @xmath2 a nowhere vanishing function . consider the two ordinary differential equations @xmath33 let us assume that there exists a smooth function @xmath34 such that for each solution @xmath18 of @xmath35 @xmath36 for each solution @xmath37 of @xmath35 defined on the open interval @xmath38 of @xmath23 , let @xmath39 then the function @xmath40 is a diffeomorphism from @xmath38 onto another open interval @xmath41 of @xmath23 , and the map @xmath42 is a solution of the ordinary differential equation @xmath43 . the derivative of the function @xmath44 , at each poit @xmath45 , is @xmath46 , which never vanishes . therefore @xmath44 is a diffeomorphism . the chain rule shows that @xmath47 is a solution of @xmath43 . for a general smooth vector field @xmath0 and a general smooth nowhere vanishing function @xmath2 given on @xmath4 , there may be no globally defined function @xmath34 verifying @xmath48 . in the kepler problem , that function exists , as we will see in subsection [ levi - civitaparameter ] , and is affine in the variable @xmath49 . the following proposition gives some information about the existence of the map @xmath50 . on a manifold @xmath4 , let @xmath0 be a vector field and @xmath51 a nowhere vanishing function . we denote by @xmath52 the flow of @xmath0 ( [ flow ] ) . for each @xmath53 , there exists a smooth function @xmath50 , defined on an open neighbourhood @xmath54 of @xmath55 in @xmath15 , such that for each solution @xmath37 of the differential equation determined by @xmath0 defined on an open interval @xmath38 containing @xmath56 , and each @xmath45 , @xmath57 the function @xmath50 is not unique : any smooth function defined on @xmath4 can be chosen for its restriction to @xmath55 . the function @xmath50 is affine with respect to the variable @xmath49 if and only if its restriction @xmath58 , @xmath59 is such that the lie derivative with respect to the vector field @xmath0 of the function @xmath60 is constant along each integral curve of @xmath0 . when the preceding condition is satisfied , the affine extension of @xmath50 to the whole @xmath15 , still denoted by @xmath50 , satisfies @xmath35 for all solutions @xmath61 of the differential equation determined by @xmath0 and all @xmath62 . let us choose any @xmath27 . the map @xmath63 is a diffeomorphism of @xmath15 onto itself , which sends the open subset @xmath14 on which the flow @xmath13 is defined onto the subset @xmath64 . therefore @xmath54 is an open subset of @xmath15 which contains @xmath55 . let @xmath65 be any smooth function . the formula @xmath66 defines a smooth function @xmath67 whose restriction to @xmath55 is @xmath68 . for each @xmath69 such that @xmath70 , we have @xmath71 therefore @xmath72 since @xmath73 is the maximal solution of the differential equation determined by @xmath0 which takes the value @xmath30 for @xmath31 , we see that the map @xmath50 satisfies condition @xmath35 of the statement above . the map @xmath50 is affine with respect to @xmath49 if and only if its partial derivative with respect to @xmath49 does not depend on @xmath49 . we have @xmath74 taking into account @xmath75 we obtain @xmath76 setting @xmath77 , we see that @xmath50 is affine with respect to @xmath49 if and only if , for all @xmath78 , @xmath79 which can also be written as @xmath80 where @xmath81 is the lie derivative of the function @xmath60 with respect to the vector field @xmath0 . this equality expresses the fact that @xmath81 is constant along each integral curve of @xmath0 . when the preceding condition is satisfied , the function @xmath50 can be uniquely extended into a function , affine with respect to the variable @xmath49 , defined on @xmath15 , and one easily check that it satisfies condition @xmath35 of the statement above for all solutions @xmath61 of the differential equation determined by @xmath0 and all @xmath62 . let @xmath1 be a smooth symplectic manifold and @xmath82 a smooth function . the unique vector field @xmath83 such that @xmath84 is called the _ hamiltonian vector field _ associated to @xmath85 , and @xmath85 is called a _ hamiltonian _ for @xmath83 . let @xmath86 be a nowhere vanishing function . the vector field @xmath87 will be called a _ conformally hamiltonian vector field _ , with @xmath85 as _ hamiltonian _ and @xmath2 as _ conformal factor_. the vector field @xmath6 satisfies @xmath88 [ result1 ] let @xmath89 be a symplectic manifold , @xmath82 a smooth hamiltonian , @xmath0 the associated hamiltonian vector field . we assume that @xmath0 is complete ; in other words , its flow @xmath13 is defined on the whole of @xmath15 . let @xmath3 be an open dense subset of @xmath4 , @xmath90 be a smooth , nowhere vanishing function and @xmath5 be the conformally hamiltonian vector field on @xmath3 , with hamiltonian @xmath85 and conformal factor @xmath2 . its flow will be denoted by @xmath91 . let @xmath92 be a smooth function such that for each maximal solution @xmath18 of the differential equation determined by @xmath6 , @xmath93 we assume that there exists on @xmath3 another symplectic form @xmath7 such that @xmath94 in other words , the vector field @xmath6 is both hamiltonian with respect to @xmath7 with @xmath85 as hamiltonian and conformally hamiltonian with respect to @xmath95 with the same @xmath85 as hamiltonian and with @xmath2 as conformal factor . under these assumptions , the map @xmath96 is a symplectic diffeomorphism from @xmath8 onto an open subset of @xmath89 , equivariant with respect to the flow of @xmath6 on @xmath3 and the flow of @xmath0 on @xmath4 , that is @xmath97 and for each @xmath98 in the open subset of @xmath99 on which @xmath91 is defined @xmath100 the maximal integral curve of the differential equation determined by @xmath6 , which takes the value @xmath101 for @xmath31 , is @xmath102 the same geometric curve in @xmath3 , parametrized by @xmath103 instead of @xmath49 , is an integral curve of the differential equation determined by the vector field @xmath0 . the values of the parameter @xmath104 which correspond to @xmath105 and to @xmath106 are , respectively , @xmath107 since @xmath13 is the flow of the vector field @xmath0 , we have @xmath108 let @xmath109 be the map @xmath110 we are going to prove that for all @xmath111 such that @xmath112 is defined , @xmath113 the above formula expresses the fact that @xmath114 maps the graph of the integral curve @xmath115 of the vector field @xmath6 which takes the value @xmath30 for @xmath31 , into the graph of the integral curve @xmath116 which takes the value @xmath30 for @xmath117 . observe that @xmath114 associates , to the point of the integral curve of @xmath6 reached for the value @xmath49 of the parameter , the point of the integral curve of @xmath0 reached for the same value @xmath118 of the parameter . replacing @xmath19 by @xmath119 in the formula which defines @xmath114 , we get @xmath120 but we have shown that @xmath121 therefore @xmath122 we have proven that @xmath123 . since @xmath6 is a hamiltonian vector field on @xmath8 , with @xmath85 as hamiltonian , the kernel of the closed @xmath124-form on @xmath99 @xmath125 is the rank - one completely integrable distribution generated by the nowhere vanishing vector field @xmath126 where @xmath49 denotes the coordinate function on the factor @xmath23 . we have indeed @xmath127 similarly , since @xmath0 is a hamiltonian vector field on @xmath89 , with @xmath85 as hamiltonian , the kernel of the closed @xmath124-form on @xmath15 @xmath128 is the rank - one completely integrable distribution generated by the nowhere vanishing vector field @xmath129 where the coordinate function on the factor @xmath23 is now denoted by @xmath104 . we have indeed @xmath130 we recall that @xmath109 maps injectively each leaf of the foliation of @xmath99 into a leaf of the foliation of @xmath15 . on the manifold @xmath99 , the 2-forms @xmath131 and @xmath132 both have the same kernel , since their kernels determine the same foliation . each of these @xmath124-forms is therefore the product of the other by a nowhere vanishing function . this function is in fact the constant @xmath133 , because @xmath134 and we have @xmath135 . restricted to @xmath136 , the map @xmath114 becomes @xmath137 since @xmath7 and @xmath95 are the forms induced , respectively , by @xmath131 on @xmath136 and by @xmath138 on @xmath139 , we have @xmath140 the @xmath124-forms @xmath95 and @xmath7 being nondegenerate , that proves that the map @xmath141 is open . but its geometric interpretation proves that @xmath141 is injective : for each @xmath142 , @xmath143 is the point of @xmath4 reached , for the value @xmath144 of the parameter @xmath104 , by the integral curve @xmath47 of @xmath0 whose value for @xmath145 is @xmath146 ; it always exists , because @xmath0 is assumed to be complete . being open and injective , @xmath141 is a diffeomorphism of @xmath3 onto an open subset of @xmath4 , and more precisely , since @xmath147 , a symplectic diffeomorphism of @xmath8 onto an open subset of @xmath89 . we have seen that @xmath114 is equivariant with respect to the flows of the vector fields @xmath148 on @xmath99 and @xmath149 on @xmath15 . by projection on the second factor , we see that @xmath141 is equivariant with respect to the flows of @xmath6 on @xmath3 and @xmath0 on @xmath4 . in the physical space , mathematically modelled by an euclidean affine 3-dimensional space @xmath150 , with associated euclidean vector space @xmath151 , let @xmath152 be a material point of mass @xmath153 subjected to the gravitational field created by an attractive centre @xmath154 . the physical time is mathematically modelled by a real independent variable @xmath49 . let @xmath155 the force @xmath156 exerted on @xmath152 is @xmath157 where @xmath158 is the constant which characterizes the acceleration field created by @xmath154 . the equations of motion are @xmath159 the second equation above becomes singular when @xmath160 , so we will assume that the massive point @xmath152 evolves in @xmath161 . that space will be called the _ configuration space _ of the kepler problem . the above equations of motion are those determined by a hamiltonian , time - independent vector field on a symplectic 6-dimensional manifold , called the _ phase space _ of the kepler problem : it is the cotangent bundle @xmath162 to the configuration space . we will identify the tangent and cotangent bundles by means of the euclidean scalar product . the hamiltonian , whose physical meaning is the total energy ( kinetic plus potential ) of @xmath152 , is @xmath163 the energy @xmath164 is a first integral of the motion , _ i.e. _ , it is constant along each integral curve . the group @xmath165 acts on the configuration space @xmath161 by rotations around @xmath154 ; the canonical lift of that action to the cotangent bundle leaves invariant the liouville @xmath133-form and its exterior derivative , the canonical symplectic form of @xmath162 . it also leaves invariant the hamiltonian @xmath164 . the lifted @xmath165-action is hamiltonian , and the corresponding momentum map is an @xmath166-valued first integral . since the euclidean vector space @xmath151 is of dimension @xmath9 , once an orientation of that space has been chosen , we can identify the lie algebra @xmath167 and its dual space @xmath166 with the vector space @xmath151 itself . with that identification the bracket in the lie algebra @xmath167 becomes the _ vector product _ , denoted by @xmath168 ; the coupling between spaces in duality becomes the _ scalar product _ @xmath169 ; the momentum map of the hamiltonian action of @xmath165 is ( up to a sign change ) the well known _ angular momentum _ @xmath170 in addition to the energy @xmath164 and the angular momentum @xmath171 , the equations of motion of the kepler problem have as a first integral the _ eccentricity vector _ @xmath172 discovered by _ jakob herman ( 16781753 ) _ @xcite , improperly called the _ laplace vector _ , or the _ runge - lenz vector _ , whose origin remained mysterious for a long time : @xmath173 using the well - known fact ( the _ first kepler s law _ ) that the orbit ( the curve described by @xmath152 as a function of time ) is a conic section with @xmath154 as one of its foci , the eccentricity vector has a very simple geometric meaning : it is a dimensionless vector , parallel to the major axis of the orbit , directed from the attracting centre towards the perihelion , of length numerically equal to the eccentricity of the orbit . we will not recall in full how the equations of motion of the kepler problem can be solved , since it is done in several excellent texts @xcite . let us however briefly indicate the proof , due to william rowan hamilton ( 18051865 ) @xcite , of an important fact : the hodograph is the curve , drawn in the associated vector space @xmath151 , by the velocity vector @xmath174 of the particle , as a function of time . ] of each motion of the kepler problem is a circle or an arc of a circle . let us first look at solutions for which at the initial time @xmath56 , @xmath175 and @xmath176 are not collinear , so @xmath177 . since @xmath171 is time - independent , @xmath175 and @xmath176 are never collinear . we choose an orthonormal frame , positively oriented , with @xmath154 as origin and unit vectors @xmath178 , @xmath179 and @xmath180 with @xmath171 parallel to @xmath180 . we have @xmath181 the vectors @xmath175 et @xmath176 remain for all times parallel to the plane @xmath182 . let @xmath183 be the polar angle made by @xmath175 with @xmath178 . we have @xmath184 therefore @xmath185 this is the _ second kepler law _ , also called _ law of areas _ , since @xmath186 is the area swept by the straight segment @xmath187 during an unit time , with sign @xmath188 if @xmath183 increases with time and @xmath189 if it decreases . observe that @xmath183 s variation is strictly monotonic . so we can take @xmath183 as independent variable , instead of time @xmath49 . we may write @xmath190 this ordinary differential equation for the unknown @xmath176 , which no more involves @xmath175 , can be readily integrated : @xmath191 where @xmath192 is a ( vector ) integrating constant . we will choose @xmath179 such that @xmath193 where @xmath194 is a numeric constant of the same sign as @xmath195 . with @xmath154 as origin let us draw two vectors in the plane @xmath182 , the first one ( constant ) being equal to @xmath192 , and the second one ( which varies with @xmath183 ) equal to @xmath176 . the end point of that second vector moves on a circle whose centre is the end point of the vector equal to @xmath192 , and whose radius is @xmath196 this circle ( or part of a circle ) is ( up to multiplication by @xmath153 ) the _ hodograph _ of the kepler problem . a short calculation leads to the following very simple relation between the energy @xmath164 of a motion , the radius @xmath197 of its hodograph and the distance @xmath198 from the attracting centre @xmath154 to the centre of the hodograph : @xmath199 observe that the right - hand side @xmath200 is the _ _ power _ _ of a point @xmath154 with respect to a circle @xmath201 is the real number @xmath202 , where @xmath203 and @xmath204 are the two intersection points of @xmath201 with a straight line @xmath205 through @xmath154 . that number does not depend on @xmath205 and is equal to @xmath206 , where @xmath207 is the centre and @xmath197 the radius of @xmath201 . ] of @xmath154 with respect to the hodograph . let @xmath50 be the function , defined on the product with @xmath23 of the phase space of the kepler problem , @xmath208 a short calculation using the equations of motion shows that for any solution @xmath209 of the kepler problem , @xmath210 with @xmath211 as the new independent variable , instead of the time @xmath49 , the equations of motion become @xmath212 i will call @xmath104 the _ levi - civita parameter _ : it was introduced , as far as i know for the first time , by tullio levi - civita in @xcite , by the differential relation which expresses @xmath213 as a fuction of @xmath214 and @xmath215 . the integrated formula which gives @xmath216 is in the paper @xcite by jean - marie souriau , and in exercise 8 , chapter ii of the book by r. cushman and l. bates @xcite . this formula has probably been known before for a long time in the celestial mechanics community , but i do not know who found it for the first time . with the levi - civita parameter as new independent variable , the system is no longer hamiltonian , but rather _ conformally hamiltonian_. the levi - civita parameter will be used in subsection [ symplecticdiff ] in a slightly different context : we will define a diffeomorphism @xmath217 from the phase space of the kepler problem restricted to negative ( resp . positive ) values of the energy , onto an open dense subset of the cotangent bundle to a three - dimensional sphere ( resp . , to one sheet of a two - sheeted three - dimensional hyperboloid ) . on this new phase space equipped with its canonical symplectic form , the image of the vector field which determines the equations of motion of the kepler problem will be conformally hamiltonian instead of hamiltonian , while the image of the vector field transformed by the use of the levi - civita parameter as independent variable will be hamiltonian . g. gyrgyi @xcite gave the expression of a symplectic diffeomorphism from the phase space of the kepler problem restricted to the negative ( resp . positive ) values of the energy , onto an open subset of the cotangent bundle to a three - dimensional sphere ( resp . two - sheeted hyperboloid ) . he obtained this diffeomorphism in two steps . first , following the ideas of fock @xcite , employed two years later by moser in his well known paper @xcite , he arranged the components of the first integrals @xmath164 , @xmath171 and @xmath218 of the kepler problem into a @xmath219 matrix , and used that matrix to define two @xmath220-dimensional vectors @xmath221 and @xmath222 ( formulae 2.15 and 2.16 of gyrgyi s 1968 paper @xcite ) . i believe that the map @xmath223 obtained by gyrgyi after difficult to follow calculations , is the map , maybe rescaled , i have called @xmath224 in subsection [ smap ] below . second , in section entiteled getting back the time @xmath49 ; bacry s generators of his 1968 paper , gyrgyi composes the map @xmath223 with a suitably chosen map deduced from the flow of the kepler vector field . he did not prove that the map he finally obtained is symplectic . this result was proven by ligon and schaaf @xcite who , eight years later , rediscovered the same diffeomorphism . twenty years later , r. h. cushman and j. j. duistermaat @xcite discussed the properties of that diffeomorphism , gave new proofs of its remarkable properties and stated the nice characterization indicated in the introduction . for negative values of the energy , the gyrgyi - ligon - schaaf diffeomorphism is the map which , to each element @xmath225 of the phase space of the kepler problem such that @xmath226 , associates the element @xmath227 of @xmath228 @xmath229 in these formulae @xmath18 is the angle given , as a function of @xmath225 , by @xmath230 the quantities @xmath231 and @xmath232 are vectors of @xmath233 , while @xmath234 and @xmath235 are vectors of @xmath236 , such that @xmath237 in other words , @xmath238 is a point of the sphere @xmath239 of radius 1 centered at the origin , embedded in @xmath236 ( endowed with its usual euclidean metric ) , and @xmath240 is a non - zero vector tangent to that sphere at point @xmath238 . using the usual scalar product of @xmath236 to identify vectors and covectors , we may consider @xmath241 as an element of the cotangent bundle to the sphere @xmath239 minus its zero section . for positive values of the energy , the gyrgyi - ligon - schaaf diffeomorphism is given by similar formulae , the trigonometric functions @xmath242 and @xmath243 being replaced by the hyperbolic functions @xmath244 and @xmath245 and the euclidean metric on @xmath236 by the lorentz pseudo - euclidean metric . we will show in what follows that the gyrgyi - ligon - schaaf symplectic diffeomorphism can be easily obtained , and that all its properties can be proven , by an adaptation of moser s method for the regularization of the kepler problem and application of theorem [ result1 ] . that method rests on the fact that hodographs of the kepler problem are circles or arcs of circles , and that the stereographic projection maps circles into circles . we do not need to know in advance that the eccentricity vector @xmath218 is a first integral of the kepler problem , since this property is an easy consequence of moser s method . in 1935 , v. a. fock @xcite used an inverse stereographic projection for the determination of the energy levels of hydrogen atoms in quantum mechanics . more recently j. moser @xcite used the same idea for the regularization of the kepler problem . his method yields a correspondence between oriented hodographs of motions of the kepler problem with a fixed , negative energy , and great oriented circles of a three - dimensional sphere . as explained by d. v. anosov @xcite and j. milnor @xcite , the method can easily be adapted for motions with a fixed , positive energy : instead of a three - dimensional sphere , one should use one sheet of a two - sheeted three - dimensional hyperboloid of revolution , and one obtains a correspondence between oriented hodographs of motions of the kepler problem with the chosen positive energy , and connected components of great hyperbolas drawn on that sheet of hyperboloid ( that means intersections of that sheet of hyperboloid with a plane containing the symmetry centre ) . to deal with cases @xmath246 and @xmath247 simultaneously , we introduce the auxiliary quantity @xmath248 let @xmath249 be an orthonormal frame of @xmath150 . with @xmath154 as origin . we add to the basis @xmath250 of @xmath151 a fourth vector @xmath251 and we denote by @xmath252 the corresponding coordinate . so we get a 4-dimensional affine space @xmath253 . the physical space @xmath150 will be identified with the affine subspace of @xmath253 determined by the equation @xmath254 . in @xmath253 , for each @xmath255 , let @xmath256 be the quadric defined by @xmath257 it is a sphere if @xmath258 , a two - sheeted hyperboloid if @xmath259 . we will see that there is a value of @xmath260 particularly suited for each value of the energy @xmath164 . let @xmath261 be the point with coordinates @xmath262 . the _ stereographic projection _ ( usual if @xmath258 , generalized if @xmath259 ) of the quadric @xmath256 minus point @xmath261 on the space @xmath150 is the map which associates , with each point @xmath263 , the intersection point @xmath153 of the straight line which joins @xmath261 and @xmath4 , with @xmath150 . see figure [ projstereo ] . if @xmath258 that map is a diffeomorphism from @xmath264 onto @xmath150 . if @xmath259 it is a diffeomorphism from @xmath265 onto the open subset of @xmath150 complementary to the 2-sphere of centre @xmath154 and radius @xmath260 the upper sheet ( @xmath266 ) of the hyperboloid ( minus point @xmath261 ) is mapped onto the outside of that sphere and the lower sheet ( @xmath267 ) onto its inside . + since the ( maybe generalized ) stereographic projection is a diffeomorphism , it can be uniquely lifted to the cotangent bundles in such a way that the pull - back of the liouville form on @xmath268 is equal to the liouville form on @xmath269 . we apply this construction with @xmath270 , the linear momentum of a keplerian motion of energy @xmath271 . each @xmath133-form on @xmath150 can be written @xmath272 , where @xmath175 is a vector field on @xmath150 , or , since @xmath270 , @xmath273 recall that @xmath258 if @xmath246 and @xmath274 if @xmath247 . the equation of @xmath256 is @xmath275 each @xmath133-form on @xmath256 can be written @xmath276 where @xmath277 is a vector field defined on @xmath256 , tangent to that quadric . therefore , @xmath278 the cotangent lift of the ( maybe generalized ) stereographic projection will be denoted by @xmath279 . it maps each pair @xmath280 made by a point @xmath263 and a vector @xmath277 tangent to @xmath256 at that point , onto a pair @xmath225 of two vectors of @xmath151 , in such a way that @xmath281 the formulae which give @xmath282 as functions of @xmath283 are easily obtained after some calculations : @xmath284 where we have set @xmath285 the formulae for the inverse transform are : @xmath286 these formulae prove that if @xmath287 and @xmath288 are two collinear vectors in @xmath151 such that @xmath289 their images , by the inverse stereographic projection , are two points @xmath290 and @xmath291 symmetric of each other with respect to @xmath154 . that property may also be proven without calculation as a consequence of properties of the transformation by inversion . we recall that if a straight line through @xmath154 meets the hodograph of a keplerian motion at two points @xmath292 and @xmath293 , @xmath294 is the power ( see the second footnote in subsection [ hodograph ] ) of @xmath154 with respect to the hodograph , and is equal to @xmath295 , where @xmath164 is the energy of the motion . therefore the above results shows that the inverse streographic projection maps the hodographs of motions of energy @xmath296 on curves , drawn on the quadric @xmath256 , which are symmetric with respect to @xmath154 . these curves are great circles of the sphere @xmath256 if @xmath258 , and great hyperbolas of the hyperboloid @xmath256 if @xmath259 . the name great hyperbola is used by analogy with great circle : it means the intersections of @xmath256 with a plane containing the symmetry centre @xmath154 . as a consequence , the cotangent lift @xmath297 of the inverse stereographic projection maps the hodographs of motions of the kepler problem of energy @xmath298 on geodesics of the quadric @xmath256 . the hodographs of motions with a different value of the energy are mapped on circles ( if @xmath258 ) or hyperbolas ( if @xmath259 ) drawn on the quadric @xmath299 , but which are not _ great _ circles , or _ great _ hyperbolas . so we see that for each value of the energy @xmath164 , there is a particularly well suited value of @xmath260 : it is the unique @xmath255 such that @xmath298 , with @xmath258 if @xmath246 and @xmath259 if @xmath247 . [ antisymp ] we have defined the mapping @xmath300 so that @xmath301 since the symplectic form of @xmath268 is @xmath302 and that of @xmath269 @xmath303 , the map @xmath279 is an _ anti - symplectic _ diffeomorphism , rather than a symplectic diffeomorphism . the transformed hamiltonian @xmath304 is @xmath305 where we have set @xmath306 the expression of @xmath307 shows that @xmath308 if and only if @xmath309 is such that @xmath310 . by differentiation we obtain @xmath311 for a motion of energy @xmath298 , the second term of the right hand side of the above equality vanishes , since @xmath312 , and we see that @xmath297 transforms the hamiltonian vector field of the kepler problem into a conformally hamiltonian vector field on @xmath269 minus the zero section , with hamiltonian @xmath313 and conformal factor @xmath314 . we introduced a minus sign in the conformal factor to account for the fact that @xmath297 is anti - symplectic ( see remark [ antisymp ] ) . for keplerian motions of energy zero the stereographic projection should be replaced by an inversion with pole @xmath154 and ratio @xmath315 . formulae for its cotangent lift are @xmath316 the correspondence @xmath317 being involutive , the formulae for the inverse transformation are @xmath318 denoting by @xmath319 the cotangent lift of the inversion of pole @xmath154 and ratio @xmath315 , we easily obtain the transformed hamiltonian @xmath320 now @xmath277 is a vector of @xmath151 , with 3 components , so we have set @xmath321 as above this result proves that @xmath322 if and only if @xmath323 is such that @xmath324 . by differentiation we obtain @xmath325 for a motion of energy @xmath326 , the second term of the right hand side of the above equality vanishes , since @xmath327 , and we see that @xmath328 transforms the hamiltonian vector field of the kepler problem into a conformally hamiltonian vector field on @xmath329 minus the zero section , with hamiltonian @xmath313 and conformal factor @xmath330 . we introduced a minus sign in the conformal factor to account for the fact that @xmath319 is anti - symplectic ( remark [ antisymp ] applies to @xmath319 as well as to @xmath279 ) . let us first recall some properties about infinitesimal symmetries . consider a vector field @xmath0 on a manifold @xmath4 . another vector field @xmath331 on @xmath4 is said to be an _ infinitesimal symmetry _ of the differential equation determined by @xmath0 ( or , in short , of @xmath0 ) if @xmath332=0 $ ] ; it is said to be a _ weak infinitesimal symmetry _ of @xmath0 if at each point @xmath16 , @xmath332(x)$ ] and @xmath333 are collinear . if @xmath331 is an infinitesimal symmetry or a weak infinitesimal symmetry of @xmath0 and @xmath252 a smooth function , @xmath331 is a weak infinitesimal symmetry of @xmath334 , since @xmath335= h[z , x]+\bigl({\mathcal l}(z)h\bigr)\,x\,,\ ] ] whichs shows that at each point @xmath16 , @xmath336(x)$ ] and @xmath333 are collinear . when @xmath337 is a hamiltonian vector field on a symplectic manifold @xmath1 , with a function @xmath85 as hamiltonian , a vector field @xmath331 on @xmath4 such that @xmath338 and @xmath339 is an infinitesimal symmetry of @xmath83 and a weak infinitesimal symmetry of @xmath340 for any smooth function @xmath252 . when in addition @xmath331 is hamiltonian , its hamiltonian is a first integral of @xmath83 and of @xmath341 . conversely , if @xmath342 and @xmath2 are two smooth first integrals of the hamiltonian vector field @xmath83 , the vector field @xmath343 , @xmath344 being the hamiltonian vector field with hamiltonian @xmath342 , is an infinitesimal symmetry of @xmath83 and a weak infinitesimal symmetry of @xmath340 for any smooth function @xmath252 . we have indeed @xmath345&=gh[x_f , x_h]+g\bigl(i(x_f)h\bigr)\,x_h - h\bigl(i(x_h)g\bigr)\,x_f\\ & = ghx_{\{f , h\}}+g\{f , h\}\,x_h - h\{h , g\}\,x_f = g\{f , h\}\,x_h\ , , \end{aligned}\ ] ] since @xmath346 . for the vector field which determines the equations of motion of the kepler problem , we already know a three - dimensional vector space of infinitesimal symmetries : it is made by the canonical lifts to the cotangent bundle of infinitesimal rotations of the configuration space @xmath161 around the attractive centre @xmath154 . these canonical lifts are hamiltonian vector fields , and their hamiltonians ( linear combinations of the components of the angular momentum @xmath195 ) are first integrals of the equations of motion . the results obtained in subsections [ cotangentlift ] and [ zeroenergy ] show that other weak infinitesimal symmetries exist . in these subsections , for each value @xmath347 of the energy @xmath164 , we have built a symplectic diffeomorphism of the phase space of the kepler problem onto an open subset of another symplectic manifold : the cotangent space to a three - dimensional sphere when @xmath348 , the cotangent space to @xmath150 when @xmath349 and the cotangent space to a three - dimensional hyperboloid when @xmath350 . that sphere , affine space @xmath150 or hyperboloid will be called the _ new configuration space _ , and its cotangent bundle ( with the zero section removed ) the _ new phase space_. for each energy level @xmath347 , the new phase space has a 6-dimensional lie group of global symmetries : @xmath351 when @xmath348 , the group @xmath352 of orientation - preserving isometries ( rotations and translations ) of @xmath150 when @xmath349 and the lorentz group @xmath353 when @xmath350 . the direct image of the hamiltonian vector field of the kepler problem , restricted to each energy level of the phase space , is equal to the restriction , to the corresponding submanifold of the new phase space , of a conformally hamiltonian vector field , whose hamiltonian is invariant by the action of the @xmath354-dimensional lie group of symmetries of the new phase space . therefore , on each energy level of the phase space of the kepler problem , six linearly independent weak infinitesimal symmetries exist , of which three are the already known linearly independent infinitesimal symmetries associated with the cotangent lifts of infinitesimal rotations of @xmath150 around three non - coplanar axes through @xmath154 . the other three correspond to the additional symmetries of the considered energy level of the new phase space . we will see that they depend smoothly on the value of the energy . therefore , on the phase space of the kepler problem considered as a whole , there exist @xmath354 linearly independent vector fields , three of them being infinitesimal symmetries and the other three weak infinitesimal symmetries . we will see that these six vector fields , which of course are tangent to each energy level , are hamiltonian and are all infinitesimal symmetries of the kepler vector field . so their hamiltonians are first integrals of the equations of motion . in subsection [ energymomentum ] we will see that these infinitesimal symmetries do not span a lie algebra , but rather a fibered space in lie algebras @xcite over the base @xmath23 , the value of the energy running over that base , in other words a _ lie algebroid _ @xcite of a special kind , with a zero anchor map . let us recall @xcite that if @xmath355 is a left action of a lie group @xmath356 on a manifold @xmath4 , its canonical lift to the cotangent bundle is a hamiltonian action @xmath357 with an equivariant momentum map @xmath358 given by @xmath359 in this formula , @xmath360 is the canonical projection ; @xmath361 , the lie algebra of @xmath356 ; @xmath362 , the dual space of @xmath363 ; and @xmath364 is the fundamental vector field on @xmath4 associated to @xmath0 , defined by @xmath365 the momentum map @xmath366 is a first integral of any hamiltonian vector field @xmath83 or conformally hamiltonian vector field @xmath341 on @xmath367 whose hamiltonian @xmath85 is invariant under @xmath368 . for each @xmath369 the hamiltonian vector field on @xmath367 whose hamiltonian is @xmath370 is an infinitesimal symmetry of @xmath83 . we apply these results , with for @xmath371 the action on the new phase space of its symmetry group . as a vector space , the lie algebra of the symmetry group is canonically isomorphic to @xmath372 , but with a lie algebra bracket @xmath373_e$ ] which depends on the energy level @xmath347 . that bracket will be determined in subsection [ energymomentum ] . two kinds of infinitesimal symmetries of the new phase space exist . the first kind is made by the fundamental vector fields which correspond to elements @xmath374 of @xmath372 . these vector fields are the canonical lifts to the cotangent bundle of the infinitesimal rotation of @xmath150 around the axis through @xmath154 parallel to @xmath375 . the second kind is made by the fundamental vector fields which correspond to elements @xmath376 of @xmath372 . these vector fields are the canonical lifts to the cotangent bundle of the following infinitesimal transformations of the new configuration space : * when @xmath348 , the restriction to the sphere @xmath256 , with @xmath377 , of the infinitesimal rotation of @xmath378 in which the plane spanned by @xmath379 and @xmath251 rotates , while the vector subspace of @xmath151 orthogonal to @xmath379 remains fixed ; * when @xmath349 , the infinitesimal translation of @xmath150 parallel to @xmath379 ; * when @xmath350 , the restriction to the hyperboloid @xmath256 , with @xmath380 , of the infinitesimal lorentz transformation of @xmath378 in which the plane spanned by @xmath379 and @xmath251 is transformed , while the vector subspace of @xmath151 orthogonal to @xmath379 remains fixed . the space @xmath372 will be identified with its dual space by using the scalar product on each factor . the momentum map , which will be denoted by @xmath381 , can be written @xmath382 , each component taking its values in @xmath151 . in the three cases @xmath348 , @xmath349 and @xmath350 , the hamiltonian @xmath85 has the same expression @xmath383 it is equivalent to the classical hamiltonian @xmath384 of a particle moving freely on @xmath256 when @xmath385 or on @xmath150 when @xmath349 . it is indeed invariant under the action @xmath368 . the first component @xmath386 of the momentum map is @xmath387 and its second component is @xmath388 as above , we have set when @xmath385 @xmath389 let us now come back to the phase space of the kepler problem . the composed map @xmath390 , which has two components @xmath391 and @xmath392 , is the momentum map of the lie algebra action of infinitesimal symmetries . using the formulae which give @xmath393 and @xmath277 as functions of @xmath175 and @xmath176 , we obtain for the first component , in the three cases @xmath348 , @xmath349 and @xmath350 @xmath394 its expression does not depend on @xmath347 . up to a change of sign , it is the angular momentum @xmath171 . for the second component we obtain @xmath395 using the value of the energy @xmath396 and , when @xmath385 , the equality @xmath397 , these formulae become @xmath398 where @xmath399 is the eccentricity vector . since the energy @xmath400 is a first integral , we see that by choosing @xmath401 and , for @xmath385 , by multiplying the weak infinitesimal symmetries by the smooth function @xmath402 , we can arrange things so that the second component of the momentum map becomes @xmath403 which no longer depends on the energy level @xmath347 . therefore , although we have defined them separately on each energy level , the three additional weak infinitesimal symmetries smoothly depend on the energy level , since they are the hamiltonian vector fields whose hamiltonians are the components of @xmath399 ; they are true ( not only weak ) infinitesimas symmetries since they are hamiltonian vector fields whose hamiltonians are first integrals of the kepler equations of motion . the poisson brackets of the components of the angular momentum @xmath171 and of the eccentricity vector @xmath218 in an orthonormal frame @xmath250 of @xmath151 are @xmath404 the family of fuctions on the phase space of the kepler problem spanned by the components of @xmath171 and of @xmath218 is not a lie algebra , because the function @xmath164 appears in the right hand sides of equalities in the last line . of course , if we replace @xmath218 by @xmath405 , we get a lie algebra isomorphic to @xmath10 when @xmath246 and to @xmath406 when @xmath247 . but in our opinion , this is not a good idea , because if we do this the energy level @xmath407 is lost and the true geometric nature of the family of infinitesimal symmetries , on the whole phase space of the kepler problem , is hidden . the above formulae make up the _ bracket table _ of smooth functions defined on the dual of a lie algebroid ( with zero anchor map ) , whose base is @xmath23 ( spanned by the coordinate @xmath164 ) and standard fibre @xmath408 ( spanned by the components of @xmath171 and @xmath218 as coordinate functions ) . the _ energy - momentum map _ is the map @xmath409 , defined on the phase space of the kepler problem , with values in @xmath410 , @xmath411 the space @xmath410 will be called the _ energy - momentum space_. as seen above , that space is the dual of a lie algebroid ; it should be considered as fibered over its first factor @xmath23 , the fibre @xmath412 over each point @xmath413 being equipped with a linear poisson structure which smoothly depends on @xmath347 . with this structure , that fibre is the dual space of a lie algebra which , as a vector space , can be identified with @xmath372 , with a lie algebra bracket @xmath373_e$ ] which smoothly depends on @xmath347 . that bracket is easily deduced from the poisson brackets of the components of @xmath171 and @xmath218 indicated above : for each energy level @xmath347 , the map which associates to @xmath414_e\right)$ ] the hamiltonian vector field with hamiltonian @xmath415 restricted to the energy level @xmath347 of the phase space , must be a lie algebras homomorphism . this remark leads to the formula @xmath416_e= \left(-\vect{u_1}\times\vect{v_1}+\frac{2e}{m^3k^2}\vect{u_2}\times\vect{v_2}\,,\,-\vect{u_1}\times\vect{v_2 } + \vect{u_2}\times\vect{v_1}\right)\,.\ ] ] by gluing together all the fibres @xmath372 over all points @xmath417 , we get on @xmath418 a lie algebroid structure , whose dual is the energy - momentum space . the above formula gives the bracket of two smooth sections @xmath419 and @xmath420 of that lie algebroid . for each @xmath413 , @xmath421(e)$ ] only depends on the values taken by @xmath422 and @xmath423 at point @xmath347 , because the considered lie algebroid has a zero anchor map . let @xmath424 be a lie algebroid @xcite whose base is a smooth manifold @xmath204 and whose anchor map is denoted by @xmath425 . the bracket of two smooth sections @xmath426 and @xmath427 of @xmath428 is denoted by @xmath429 . let @xmath430 be a surjective submersion of a smooth manifold @xmath4 onto the base @xmath204 . action _ of the lie algebroid @xmath424 on the fibered manifold @xmath430 is a map @xmath431 which associates , to each smooth section @xmath432 of @xmath428 , a vector field @xmath433 on @xmath4 , in such a way that for each smooth section @xmath104 of @xmath428 and each smooth function @xmath434 , @xmath435 and that for each pair @xmath436 of smooth sections of @xmath428 , @xmath437=x_{\{s_1,s_2\}}\,.\ ] ] let us now assume that the manifold @xmath4 is endowed with a symplectic form @xmath438 . the vector fields @xmath433 on @xmath4 associated to smooth sections @xmath104 of the lie algebroid @xmath424 can not all be hamiltonian , because if for some choice of the section @xmath104 , @xmath433 is hamiltonian , the vector field @xmath439 , associated to the section @xmath440 , where @xmath434 is a smooth function , will not in general be hamiltonian . so one may wonder what should be a reasonable definition of a lie algebroid hamiltonian action on a symplectic manifold . the answer is suggested by the example of the kepler problem . in the last subsection , we described the lie algebroid @xmath424 of infinitesimal symmetries of the kepler problem : @xmath441 , @xmath442 , @xmath424 is the projection on the first factor , the anchor map is the zero map @xmath443 . the vector bundle @xmath424 being trivial , each pair of vectors @xmath444 can be considered as a ( constant ) section @xmath445 of our lie algebroid . we defined the action of our lie algebroid on the phase space of the kepler problem by taking , as vector field @xmath446 associated to the section @xmath445 , the hamiltonian vector field with hamiltonian @xmath447 this condition unambiguously defines our lie algebroid action ( and , simultaneously , the bracket composition law of its sections ) since the module of its smooth sections is spanned by the constant sections . it would be interesting to see whether such a construction can be extended for more general lie algebroids , with a nonzero anchor map . the main disadvantage of moser s regularization method is that it handles separately each energy level . this disadvantage may be partially removed by the following procedure , which allows to handle together all negative ( resp . , all positive ) energy levels . however , negative , positive and zero energy levels still can not be handled together with that procedure . let us consider the quadric @xmath256 , defined by @xmath449 where @xmath260 may take any positive value , and let @xmath450 be the quadric defined by @xmath451 where @xmath452 is a fixed positive quantity . to each point @xmath453 , we associate the point @xmath454 such that @xmath455 we lift this diffeomorphism @xmath456 to the cotangent bundles , and we get a symplectic diffeomorphism @xmath457 . this symplectic diffeomorphism associates to each pair @xmath458 made by a point @xmath453 and a vector @xmath459 tangent to @xmath450 at that point , the pair @xmath460 made by a point @xmath454 and a vector @xmath461 tangent to @xmath256 at that point : @xmath462 for each @xmath255 , we compose the symplectic diffeomorphism @xmath463 with the symplectic diffeomorphism @xmath464 built in subsection [ cotangentlift ] . we obtain a family , indexed by @xmath255 , of symplectic diffeomorphisms @xmath465 : @xmath466 where we have set @xmath467 the formulae for the inverse transformation are @xmath468 the diffeomorphism @xmath469 sends the subset of @xmath162 on which the energy is @xmath470 into the subset of @xmath471 on which @xmath472 . therefore , if @xmath473 and @xmath474 are two distinct possible values of @xmath260 , the images by @xmath475 of the energy level @xmath476 , and by @xmath477 of the energy level @xmath478 , are disjoint . by restricting each map @xmath469 to the subset of @xmath162 on which the energy is @xmath470 , and by gluing together these restricted maps for all possible values of @xmath260 , we obtain a unique diffeomorphism @xmath224 from the open subset of the cotangent bundle @xmath162 minus the zero section on which the energy @xmath400 is negative if @xmath258 , positive if @xmath259 , onto @xmath479 minus the zero section . this diffeomorphism , built by gluing together pieces of the symplectic diffeomorphisms @xmath469 for all values of @xmath255 , is no longer symplectic ! it is given by the formulae , in which we no longer write the subscript @xmath452 , @xmath480 we have set @xmath389 the transformed hamiltonian @xmath481 is @xmath482 up to the constant factor @xmath483 , it is the _ delaunay hamiltonian _ @xcite , _ i.e. _ , the kepler hamiltonian in delaunay coordinates . this property is related to the fact that delaunay variables are action - angle variables for the kepler problem . in the rest of the paper we write @xmath484 for @xmath450 . the north pole of @xmath484 is denoted by @xmath261 instead of @xmath485 . let @xmath486 be the hamiltonian vector field on @xmath329 with hamiltonian @xmath164 , _ i.e. _ the hamiltonian vector field of the kepler problem , and let @xmath83 be the hamiltonian vector field , on @xmath487 minus the zero section , whose hamiltonian @xmath488 is given by the above formula . the direct image @xmath489 of the vector field @xmath486 by the diffeomorphism @xmath224 is not equal to @xmath83 , since @xmath224 is not symplectic . a short calculation leads to its expression , @xmath490 the function @xmath2 is smooth on @xmath487 minus the zero section , and becomes singular when @xmath491 , _ i.e. _ on the fibre over the north pole @xmath261 . we see that the map @xmath224 sends the hamiltonian vector field of the kepler problem to a conformally hamiltonian vector field , with @xmath85 as hamiltonian and with @xmath2 as conformal factor . unlike the conformal factor @xmath2 , which becomes singular on the cotangent space to @xmath484 at the north pole , the hamiltonian @xmath85 is smoothly defined on the whole cotangent bundle @xmath493 minus the zero section . the associated hamiltonian vector field @xmath83 is complete . its flow , defined on @xmath494 , @xmath495 has slightly different expressions for @xmath258 ( negative energy ) and for @xmath259 ( positive energy ) . observe that @xmath496 is a first integral of @xmath83 . @xmath497 @xmath498 we have set @xmath499 . at the end of subsection [ smap ] , we have seen that the image by @xmath224 of the hamiltonian vector field @xmath486 of the kepler problem is a conformally hamiltonian vector field @xmath341 defined on an open dense subset of @xmath500 , whose hamiltonian @xmath85 is smoothly defined on the whole @xmath500 ; the corresponding hamiltonian vector field @xmath83 is complete . moreover the conformally hamiltonian vector field @xmath341 is also hamiltonian , with hamiltonian @xmath85 , not for the canonical symplectic form of @xmath493 , but for the pull - back by @xmath217 of the canonical symplectic form of @xmath268 . we have seen in subsection [ levi - civitaparameter ] that for every solution @xmath501 of the equations of motion of the kepler problem , @xmath502 the conformal factor @xmath2 , whose expression is given at the end of subsection [ smap ] , may also be expressed as @xmath503 therefore @xmath504 all the integral curves of the conformally hamiltonian vector field @xmath341 can be written @xmath505 where @xmath501 is a solution of the equations of motion of the kepler problem . we see that the pull - back by @xmath506 of the function , defined on the product with @xmath23 of the open subset of the phase space of the kepler problem on which the energy @xmath164 is negative if @xmath258 , positive if @xmath259 , @xmath507 has the properties of the function @xmath50 of theorem [ result1 ] of subsection [ confhamfields ] . that theorem can therefore be applied . it proves that by composing the symplectic diffeomorphism @xmath224 with the flow @xmath508 of the hamiltonian vector field @xmath83 , for suitably chosen values of the independent variable @xmath104 , we get a symplectic diffeomorphism from the phase space of the kepler problem , restricted either to negative , or to positive energies , onto an open subset of @xmath493 . the explicit expression of this symplectic diffeomorphism is @xmath509 the quantities @xmath510 are given , as functions of @xmath511 , by the formulae at the end of subsection [ flowofxh ] . we have to set @xmath512 the formulae at the end of subsection [ smap ] give the expressions of @xmath513 , @xmath252 , @xmath514 and @xmath515 as functions of @xmath516 and @xmath517 . finally we see that the symplectic diffeomorphism so obtained is that of gyrgyi , ligon and schaaf . r. cushman and l. bates , in chapter ii of the book @xcite , offer a detailed discussion of its properties , notably its behaviour with respect to the liouville @xmath133-forms of the cotangent spaces of the kepler configuration manifold and of the @xmath9-dimensional sphere ( or hyperboloid ) . the properties of the diffeomorphism from the phase space of the kepler problem to the cotangent bundle of a sphere or of an hyperboloid are explained , in subsection [ smap ] , from a point of view other than that used by cushman and duistermaat @xcite . our explanation is founded on a very natural property of conformally hamiltonian vector fields . however , the procedure we used , as well as those used by gyrgyi @xcite , ligon and schaaf @xcite remains unsatisfactory in the fact that it handles separately negative and positive energy levels . the space of motions of the kepler problem is connected , since by varying slowly the energy level , an elliptic motion can be transformed into a parabolic , then into an hyperbolic motion . by completely different methods , j .- souriau @xcite proposed a global description of the manifold of motions of the kepler problem , of its regularization and of its global and infinitesimal symmetries . it should be possible to do something similar at the level of the phase space instead of at the level of the manifold of motions . the fact that the kepler vector field is hamiltonian with respect to a symplectic form and conformally hamiltonian with respect to another syplectic form is remarkable ; it is probably related to the complete integrability of that vector field , as suggested by an anonymous referee . in @xcite , a. maciejewski , m. prybylska and a. tsiganov have used conformally hamiltonian vector fields within the theory of bi - hamiltonian systems , to build completely integrable systems . we have shown that the infinitesimal symmetries of the kepler problem form a lie algebroid rather than a lie algebra . actions of lie algebroids on symplectic manifolds is a subject which seems to us interesting ; it should be nice to have other examples of hamiltonian vector fields whose infinitesimal symmetries form a lie algebroid , maybe with a non - zero anchor map . a. douady and m. lazard @xcite have shown that lie algebroids with a zero anchor map are integrable into lie groupoids ; it would be interesting to write down explicitly the action of a lie groupoid with the groups @xmath351 , @xmath518 and @xmath353 as isotropy groups on a regularized phase space of the kepler problem . i am much indebted towards my colleagues and friends of the centre international de thories variationnelles , the international workshop on differential geometric methods in mechanics and the seminar of hamiltonian geometry for helpful discussions , towards alan weinstein who , after listening to a presentation of a preliminary version of this work , made several enlighting observations , and towards alain guichardet and alain albouy who communicated me their unpublished papers on the kepler problem . i address my thanks to the two anonymous referees whose judicious observations and suggestions were very helpful . j. bernoulli , _ extrait de la rponse de m. bernoulli m. herman , date de basle le 7 octobre 1710 _ , histoire de lacadmie royale des sciences , anne m.dcc.x , avec les memoires de mathmatiques et de physique pour la mme anne , pages 521533 ( 1710 ) . available on line : http://gallica.bnf.fr/ark:/12148/bpt6k34901/f707.image.pagination . j. herman , _ extrait dune lettre de m. herman m. bernoulli , date de padoe le 12 juillet 1710 _ , histoire de lacadmie royale des sciences , anne m.dcc.x , avec les memoires de mathmatiques et de physique pour la mme anne , pages 519521 ( 1710 ) . available on line : http://gallica.bnf.fr/ark:/12148/bpt6k34901/f709.image.pagination . ( mr0773497 ) j .- m . souriau , _ gomtrie globale du problme deux corps _ , ( french ) [ global geometry of the two - body problem ] , in modern developments in analytical mechanics supplemento al vol . * 117 * , atti della accademia della scienze di torino , ( 1983 ) , 369418 .

by molecular dynamics simulation of a coarse - grained bead - spring type model for a cylindrical molecular brush with a backbone chain of effective monomers to which with grafting density side chains with effective monomers are tethered , several characteristic length scales are studied for variable solvent quality . side chain lengths are in the range , backbone chain lengths are in the range , and we perform a comparison to results for the bond fluctuation model on the simple cubic lattice ( for which much longer chains are accessible , , and which corresponds to an athermal , very good , solvent ) . we obtain linear dimensions of side chains and the backbone chain and discuss their-dependence in terms of power laws and the associated effective exponents . we show that even at the theta point the side chains are considerably stretched , their linear dimension depending on the solvent quality only weakly . effective persistence lengths are extracted both from the orientational correlations and from the backbone end - to - end distance ; it is shown that different measures of the persistence length ( which would all agree for gaussian chains ) are not mutually consistent with each other , and depend distinctly both on and the solvent quality . a brief discussion of pertinent experiments is given .

1111.0481

macromolecules which consist of a `` backbone '' polymer , to which flexible or stiff side chains are grafted , the so - called `` bottle brush polymers '' , find very great interest recently ( see refs @xcite for reviews ) . varying the chemical nature of both backbone chain and side chains , their chain lengths @xmath6 and the grafting density @xmath1 , the structure of these cylindrical molecular brushes can be widely varied . i.e. , their local thickness " as measured by the cross - sectional radius @xmath7 or linear dimensions of individual side chains can be varied as well as their local `` stiffness '' , traditionally measured by `` the '' persistence length @xmath8 @xcite , and their effective contour length @xmath9 . we here use quotation marks with respect to `` the '' persistence length , because there is evidence , at least for the case of very good solvent conditions , that a unique persistence length measuring the `` intrinsic '' stiffness of a polymer can not be defined in the standard fashion @xcite . now an intriguing observation @xcite is the sensitivity of the large - scale structure of these bottle brush polymers to solvent quality : one finds a thermally induced collapse of single macromolecules from cylindrical brushes to spheres , in a very small temperature range , and it is speculated that these bottle brushes could be useful as building blocks of `` soft nanomachines '' @xcite . we also note that biopolymers with comb - like architecture are ubiquitous in nature ( such as proteoglycans @xcite , or the aggrecane molecules that play a role in the soft lubricating layers in human joints @xcite , etc . ) , and probably in this context temperature and/or solvent quality ( or ph value ) are relevant parameters as well . in view of these facts , a comprehensive clarification of how the properties of bottle brush polymers depend on solvent quality clearly would be interesting . although there are occasional experimental reports , how particular linear dimensions of these polymers scale in various solvents ( e.g. @xcite ) , we are not aware of a systematic study of this problem . while work based on self - consistent field theory ( scft ) predicted already very early on @xcite that the side chain gyration radius @xmath10 scales as @xmath11 for good solvents and @xmath12 for theta solvents , there is now evidence from experiment , simulation and theory that these power laws apply if at all only for side chain length @xmath2 of the order of @xmath13 , which are of no practical relevance : experiments have only studied the range @xmath14 , and the range accessible in simulations @xcite is similarly restricted . numerical modeling applying the scheutjens - fleer version of scft has given clear evidence @xcite that even within this mean - field approach @xmath15 is needed to reach the regime where the predicted power laws @xcite apply . while the investigation of the scaling ( self - similar ) properties of bottle brushes with extremely long backbone chains and very long side chains may be a challenging theoretical problem , it is of little relevance for the experimentally accessible systems , and clearly not in the focus of the present paper . both scft theories @xcite and scaling theories @xcite consider ideal chains aiming to describe the behavior at the theta temperature , while previous simulations have almost exclusively considered the good solvent case only . notable exceptions are studies of globule formation of bottle brushes with very short side chains ( @xmath16 ) under poor solvent conditions @xcite and the study of microphase separation of bottle brushes with straight backbones in poor solvent @xcite . the present work intends to make a contribution to close this gap , by presenting a simulation study of a coarse - grained bead - spring type model of bottle brush polymers where the solvent quality is varied from very good solvent conditions to the theta point regime . in the next section , we shall describe the model and simulation technique , while in section iii we present our numerical results . in section iv , a summary is given , as well as an outlook on pertinent experiments . extending our previous work on the simulation of bottle brush polymers with rigid backbones @xcite , we describe both the backbone chain and the side chains by a bead - spring model @xcite , where all beads interact with a truncated and shifted lennard - jones potential @xmath17 and nearest neighbors bonded together along a chain also experience the finitely extensible nonlinear elastic potential @xmath18 , @xmath19 being the distance between the beads . thus @xmath20+c \ , , \qquad r\le r_c \ , , \label{eq1}\ ] ] while @xmath21 , and where @xmath22 . the constant @xmath23 is defined such that @xmath24 is continuous at the cutoff . henceforth units are chosen such that @xmath25 , @xmath26 , the boltzmann constant @xmath27 , and in addition also the mass @xmath28 of beads is chosen to be unity . the potential eq . ( [ eq1 ] ) acts between any pair of beads , irrespective of whether they are bonded or not . for bonded beads additionally the potential @xmath18 acts , @xmath29 \ , , \qquad 0 < r \le r_0 \ , , \label{eq2}\ ] ] while @xmath30 , and hence @xmath31 is the maximal distance that bonded beads can take . we use the standard choice @xcite @xmath32 and @xmath33 . related models have been used with great success to study the glassification of polymer melts formed from short chains @xcite and to study the effects of solvent quality of polymer brushes on flat planar substrates @xcite . for such models of brushes on a planar substrate , the implicit solvent model ( eq . ( 1 ) ) has been compared with models using explicit solvent molecules , and it was found that the results are very similar . note that in our model there is no difference in interactions , irrespective of whether the considered beads are effective monomers of the backbone or of the side chains , implying that the polymer forming the backbone is either chemically identical to the polymers that are tethered as side chains to the backbone , or at least on coarse - grained length scales as considered here the backbone and side chain polymers are no longer distinct . there is also no difference between the bond linking the first monomer of a side chain to a monomer of the backbone and bonds between any other pairs of bonded monomers . of course , our study does not address any effects due to a particular chemistry relating to the synthesis of these bottle brush polymers , but , as usually done @xcite we address universal features of the conformational properties of these macromolecules . there is one important distinction relating to our previous work @xcite on bottle brush polymers with rigid backbones : following grest and murat @xcite , there the backbone was taken as an infinitely thin straight line in continuous space , thus allowing arbitrary values of the distances between neighboring grafting sites , and hence the grafting density @xmath1 could be continuously varied . for the present model , where we disregard any possible quenched disorder resulting from the grafting process , of course , the grafting density @xmath1 is quantized : we denote here by @xmath34 the case that every backbone monomer carries a side chain , @xmath35 means that every second backbone monomer has a side chain , etc . chain lengths of side chains were chosen as @xmath36 , @xmath37 , @xmath38 , and @xmath39 , while backbone chain lengths were chosen as @xmath40 , @xmath41 , and @xmath42 , respectively . it is obvious , of course , that for such short side chain lengths any interpretation of characteristic lengths in terms of power laws , such as @xmath43 , is a delicate matter , @xmath44 being an `` effective exponent '' and characterizes only the specified range of rather small values of @xmath2 , and not the limit @xmath45 considered by most theories @xcite . thus , the actual value of @xmath44 is of limited interest , it only gives an indication to which part of an extended crossover region the data belong . however , we emphasize that ( i ) our range of @xmath2 nicely corresponds to the range available in experiments @xcite . ( ii ) the analysis in terms of power laws with effective exponents is a standard practice of experimentalists in this context ( e.g. @xcite ) . we recall that for linear chains the theta temperature for the present ( implicit solvent ) model has been roughly estimated @xcite as @xmath46 ( note , however , that there is still some uncertainty about the precise value of @xmath47 : for a similar model @xcite the correct value of @xmath47 , @xmath48 in this case , could only be established for chain lengths exceeding @xmath49 ) . thus , in the present work we have thoroughly studied the temperature range @xmath50 . from previous work @xcite on rather long chains in polymer brushes on flat surfaces , using the same model ( eqs . ( [ eq1 ] ) , ( [ eq2 ] ) ) to describe the interactions , it is known that for @xmath51 one finds a behavior characteristic for ( moderately ) good solvents . very good solvent conditions could be obtained from a slightly different model that has extensively been studied for standard polymer brushes @xcite , where in eq . ( [ eq1 ] ) the cutoff is chosen to coincide with the minimum of the potential , @xmath52 ( and then also @xmath53 can be chosen for this essentially athermal model ) . rather than carrying out simulations for bottle brushes using this model , we found it more appropriate to compare to the athermal version of the bond fluctuation model on the simple cubic lattice , which describes very good solvent conditions @xcite and has been used in our earlier work @xcite . the use of this model has several advantages : ( i ) due to the fact that excluded volume constraints can be monitored via the occupancy of lattice sites , a very efficient implementation of the pivot algorithm in this monte carlo approach has become possible @xcite . therefore , very large bottle brush polymers can be equilibrated , up to @xmath54 , a task that is very difficult to achieve by molecular dynamics ( md ) methods ( ii ) the extent to which the bond fluctuation model and md results agree ( for comparable choices of parameters ) yields some insight to what extent the preasymptotic regime that we study is model dependent . of course , a truly universal behavior ( apart from amplitude prefactors ) can only be expected for the asymptotic regime where the side chain length @xmath45 , that is not accessible in simulations or experiments . one could ask why we do not use versions of the bond fluctuation model where an effective attraction between monomers is included @xcite to study the effect of variable solvent quality in the framework of this model . the reason is that for temperatures slightly below the theta temperature already practically frozen configurations of monomers occur , with several monomers next to each other blocking any possibility to move . thus the convergence towards equilibrium then is extremely slow . however , since the application of the athermal version of the bond fluctuation model ( bfm ) to the simulation of bottle brush polymers is well documented ( hsu and paul @xcite have given a careful discussion of the effort needed for the bfm to sample equilibrium properties . ) in the recent literature @xcite , we do not give any details here . in the md simulation , the positions @xmath55 of the effective monomers with label @xmath56 evolve in time @xmath57 according to newton s equation of motion , amended by the langevin thermostat @xcite @xmath58 where @xmath59 is the total potential acting on the @xmath56th bead due to its interactions with the other beads at sites @xmath60 , @xmath61 is the friction coefficient , and @xmath62 the associated random force . the latter is related to @xmath61 by the fluctuation - dissipation relation @xmath63 following previous work @xcite we choose @xmath64 , the md time unit @xmath65 being also unity , for our choice of units . ( [ eq3 ] ) was integrated using the leap frog algorithm @xcite , with a time step @xmath66 , and utilizing the gromacs package @xcite . for the calculation of properties of the bottle brushes typically @xmath67 statistically independent configurations are averaged over . of course , for bottle brushes with large @xmath0 equilibration of the polymer conformations is a difficult problem . since we expect that end - to - end distance @xmath68 and gyration radius @xmath69 of the whole molecule belong to the slowest relaxing quantities , we studied the autocorrelation function of @xmath70 to test for equilibration , @xmath71 note that @xmath72 means an average of @xmath73 over the time @xmath74 in the md trajectory ( and the average @xmath75 is computed by averaging over @xmath67 statistically independent runs ) . despite a substantial investment of computer time , @xmath76 still exhibits significant fluctuations ( remember that quantities such as @xmath68 and @xmath69 are known to exhibit a `` lack of self - averaging @xcite '' , irrespective of how large @xmath0 is ) . [ fig1 ] gives some examples for @xmath76 . if the number of samples would be infinite and @xmath77 , we should expect a monotonous decay of @xmath76 towards zero as @xmath57 becomes large . due to the fact that the number of samples is not extremely large , and @xmath78 not extremely small , fig . [ fig1 ] gives clear evidence for noise that is still correlated . we see that for small @xmath2 the noise amplitude starts out at @xmath79 , and the time scale on which the fluctuations of @xmath76 change sign is at @xmath80 in ( a ) , @xmath81 in ( b ) , and @xmath82 in ( c ) . while in cases ( a)(b ) the ( statistically meaningful ) initial decay of @xmath76 occurs so fast that it can only be seen on a magnified abscissa scale ( inserts ) , we see that in ( c ) the initial decay is also much slower , and the associated time scale is of the same order as the time over which fluctuations are correlated . we have carefully considered @xmath76 for all cases studied , and we have concluded that for our largest system studied ( shown in fig . [ fig1](c ) , with a total number of @xmath83 effective monomers ) the statistical effort is not yet sufficient to allow meaningful conclusions on the overall linear dimensions of the bottle brush , while in all other cases the effort was judged to be sufficient . the damped oscillatory character of the relaxation seen particularly in fig . [ fig1](c ) could be a matter of concern ; we attribute this relaxation behavior to particular slow collective motions ( breathing type modes ) of the chain . [ fig2 ] shows a small selection of snapshot pictures of equilibrated bottle brush polymers . from these snapshot pictures it is already clear that the side chains cause a significant stiffening of the backbone , at least on a coarse - grained scale , and that bottle brushes where @xmath0 is not very much larger than @xmath2 look like wormlike chains this conclusion corroborates pictures generated experimentally ( by atomic force microscopy or electron microscopy techniques , e.g. @xcite ) , but this observation should not mislead one to claim that the kratky - porod wormlike chain model @xcite often employed to analyze such micrographs provides an accurate description of bottle brush polymers , as we shall see . of course , both the bead - spring model and the bfm are idealizations of realistic comb - branched polymers , where also bond - angle potentials and torsional potentials are present and contribute to the local chain stiffness . thus it is gratifying to note that nevertheless measured structure factors of real bottle brush polymers @xcite can be mapped almost quantitatively to their simulated bfm counterparts @xcite , if the lattice spacing is fixed at a length of a few angstroms , and about 3 chemical monomers are mapped onto two effective monomers of the bfm . residual minor discrepancies may to some extent be due to solvent quality effects @xcite . fig . [ fig3 ] presents log - log plots of the normalized mean square gyration radius of the side chains @xmath84 as a function of side chain length @xmath2 ( @xmath85 is the bond length between successive monomers ) , comparing data for grafting densities @xmath35 and @xmath86 , and two temperatures ( @xmath87 and @xmath51 , respectively ) . three different backbone lengths ( @xmath40 , @xmath41 , and @xmath42 ) are included , but the dependence of the data on backbone length is not visible on the scale of the graph . such a dependence on @xmath0 would be expected due to effects near the chain ends of the backbone : in ref . @xcite it was shown that there is less stretching of the side chains in the direction perpendicular to the backbone near the chain ends of the backbone than in the central part of the backbone . however , this effect is almost completely compensated by the fact that side chains near the backbone chain ends are more strongly stretched in the direction of the backbone ( side chains then are oriented away from the backbone chain ends ) . this fact has already been seen for bottle brushes with rigid backbone but free ends @xcite . we have also checked that for the bead spring model average side chain linear dimensions for bottle brushes with rigid and flexible backbones are identical for the bond fluctuation model , at least in the accessible parameter regime . for the bond fluctuation model a similar equivalence has been found also for the radial density profile of the monomers @xcite . [ fig4 ] shows then the temperature dependence of @xmath88 for the different choices of @xmath2 , demonstrating that the side chain extension is independent of @xmath0 and agrees well between rigid and flexible backbones . the small deviations observable between the rigid and the flexible backbone cases for large @xmath2 are due to residual systematic errors of the simulations for the flexible backbone . the present work therefore suggests that the equivalence between bottle brushes with rigid and flexible backbones carries over to chains in variable solvent quality , down to the theta point ( but we caution the reader that this equivalence will break down for poor solvents , in the regime of intermediate grafting densities where for rigid backbones pearl - necklace structures occur @xcite ) . the straight lines on the log - log plots in fig . [ fig3 ] illustrate the empirical power law @xmath89 consistent with corresponding experiments ( see @xcite and references quoted therein ) and we find that @xmath44 decreases with decreasing solvent quality and with decreasing grafting density . in the good solvent regime , @xmath90 @xcite , the established value of the exponent describing the swelling of linear polymers under good solvent conditions , @xmath91 for @xmath45 @xcite . as expected , @xmath44 is still distinctly smaller than the value predicted from scaling @xcite , @xmath92 or scft @xcite , @xmath93 . but the slight enhancement of @xmath44 with respect to its value for free chains ( @xmath94 at the theta point , @xmath87 ; @xmath95 for good solvents ) is evidence that the side chains interact with each other , which is a prerequisite for the expected induced stiffening of the backbone chain . we now turn to the backbone linear dimensions . here we first caution the reader that in good solvent conditions we clearly also expect that the mean square end - to - end distance of a bottle - brush polymer satisfies the standard scaling relation @xmath96 but the longer the side chain length the larger @xmath0 must be chosen such that eq . ( [ eq8 ] ) can be verified . we start the discussion with data from the bond fluctuation model , where ( for @xmath97 ) data up to @xmath54 are available @xcite . [ fig5 ] hence shows a log - log plot of @xmath98 ) versus @xmath2 . the factor @xmath99 in the denominator is arbitrarily chosen , since for gaussian chains ( @xmath94 ) the result would simply be the persistence length @xmath8 in units of the bond length @xcite . thus , it was suggested @xcite that in the excluded volume case one could introduce an effective persistence length @xmath100 via the definition @xmath101 then fig . [ fig5 ] can be interpreted as a plot of @xmath102 versus @xmath2 in the region where it is basically independent of @xmath0 . the quantity @xmath100 defined in this way is compatible with an effective power law , @xmath103 but the effective exponent @xmath104 clearly depends on @xmath0 , if @xmath0 is not chosen large enough ( fig . [ fig6 ] ) . thus , fig . [ fig5 ] clearly shows that @xmath100 is not a quantity characterizing the intrinsic stiffness of a bottle - brush polymer . for large @xmath0 , the exponent @xmath104 seems to saturate at a value close to @xmath105 and @xmath106 for @xmath35 and @xmath86 , respectively , but for chain lengths @xmath0 in the range from @xmath107 to @xmath108 it is only in the range from @xmath109 to @xmath110 and from @xmath111 to @xmath112 for @xmath35 and @xmath86 , respectively . thus , it is no surprise that for the bead - spring model choosing @xmath40 , @xmath41 , or @xmath42 values of this effective exponent @xmath104 in a similar range are found ( fig . [ fig7 ] ) . of course , one must be aware that there is no strict one - to - one correspondence @xcite between the meaning of chain lengths @xmath0 and @xmath2 in different models : actually there may be the need for conversion factors @xmath113 and @xmath114 between the bead spring ( bs ) model and the bond fluctuation model ( bfm ) . the same fact is true when we compare simulations to experiments ; e.g. , the data of rathgeber et al . @xcite could be mapped to the bfm @xcite implying an equivalence between @xmath115 and @xmath116 and between @xmath117 and @xmath118 , for instance . we also note that with decreasing temperature @xmath119 decreases , irrespective of @xmath2 ( and also @xmath104 decreases ) . again we remind the reader that this parameterization of our data in terms of effective exponents ( also used in related experimental work @xcite only can serve to indicate the place in an extended crossover region to which the data belong . when we analyze the gyration radius of the backbone as a function of backbone chain length for the bs model ( figs . [ fig8 ] ) we find that over a restricted range of @xmath0 one can fit the data by effective exponents again , @xmath120 , with @xmath121 , and it is seen that @xmath44 increases systematically with side chain length @xmath2 , and @xmath44 for @xmath51 is larger than for @xmath87 , for the same choice of @xmath2 . of course , this variation of the effective exponent is just a reflection of a gradual crossover from the rod - like regime ( @xmath122 ) to the self - avoiding walk - like behavior of swollen coils ( @xmath123 , with @xmath124 ) in the good solvent regime @xmath125 or gaussian - like coils ( @xmath126 ) for the theta point ( @xmath87 ) , respectively . similar plots as in fig . [ fig8 ] but for the bfm are shown in fig . [ fig9](a)(b ) . due to the availability of equilibrated data for much longer backbone chain lengths @xmath0 for the bfm , the plot of @xmath127 versus @xmath0 ( fig . [ fig9](c ) ) indeed shows that the data settle down to a horizontal plateau implying that the asymptotic region where eq . ( [ eq9 ] ) can be applied indeed is reached . instead of @xmath128 in eq . ( [ eq9 ] ) , we use @xmath129 such that the estimate of the @xmath2-dependent effective persistence length @xmath130 is given by the horizontal line in fig.[fig9](c ) . actually , for fixed side chain length @xmath2 the different choices of @xmath0 can be collapsed on a master curve , when @xmath0 is rescaled by @xmath131 @xcite , which can be interpreted as an estimate for the number of segments per persistence length . in our previous work @xcite , we have shown that bottle brushes under very good solvent conditions can be described as self - avoiding walks of effective blobs " , having a diameter that is just twice the cross - sectional radius of the bottle - brush , and containing a number @xmath131 of backbone monomers . note that hence @xmath131 is not a fit parameter , but has been determined independently . this data collapse on a master curve is demonstrated for the bfm for the case @xmath34 , @xmath132 , @xmath133 , @xmath134 , and @xmath135 ( fig , [ fig9](d ) ) . since `` the '' persistence length of polymers in the standard textbooks @xcite is traditionally defined from the decay of bond orientational correlations along the chains , we next focus on this quantity . defining the bond vectors @xmath136 in terms of the monomer positions @xmath137 as @xmath138 , @xmath139 , this bond orientational correlation is defined as @xmath140 note that @xmath141 and hence @xmath142 , of course . considering the limit @xmath143 , and assuming gaussian chain statistics , one obtains an exponential decay , since then @xmath144 $ ] , and thus @xmath145\ , , \qquad l_p^{-1}=-\ln \langle \cos \theta ( 1 ) \rangle/\ell_b \ , . \label{eq12}\ ] ] however , it has been shown by scaling and renormalization group arguments @xcite and verified by simulations @xcite that in the good solvent case there actually occurs a power law decay @xmath146 while for chains at the theta point @xcite or in melts @xcite one has @xmath147 as far as bond orientational correlations are concerned , gaussian chain statistics hence is misleading for polymers , under all circumstances . however , shirvanyants et al . @xcite suggested that for semiflexible polymers ( at the theta point ) one can use still eq . ( [ eq12 ] ) but only for @xmath148 , where @xmath149 controls the crossover from the simple exponential decay , eq . ( [ eq12 ] ) , to the power law , eq . ( [ eq14 ] ) . indeed , for a simple self - avoiding walks ( saw ) model on the simple cubic lattice , where chain stiffness was controlled by an energy cost @xmath150 when the chain makes a @xmath151 kink on the lattice @xcite , it was shown that this suggestion @xcite works qualitatively , also in the good solvent case , with @xmath152 for @xmath153 , although the crossover between eqs . ( [ eq12 ] ) and ( [ eq13 ] ) is not sharp but rather spread out over a decade in the variable @xmath154 . motivated by this finding @xcite , figs . [ fig10 ] , [ fig11 ] hence present a few examples where @xmath155 is plotted vs. @xmath154 on a semi - log plot , to test for a possible applicability of eq . ( [ eq12 ] ) for not too large @xmath154 . however , we find that in fact eq . ( [ eq12 ] ) does never hold for small @xmath154 ( @xmath156 , @xmath99 , @xmath157 ) , unlike the semiflexible saw model of ref . @xcite , rather there occurs a very fast decay of @xmath155 following a strongly bent curve ( only two successive values of @xmath154 can always be fit to a straight line , of course , but there is never an extended regime where a straight line through the origin , @xmath158 , would be compatible with the data ) . it is also remarkable that this initial behavior is almost independent of @xmath1 and @xmath2 ; an obvious interpretation is that the stiffening of the backbone caused by the presence of long side chains is effective only on mesoscopic length scales along the backbone , but not on the scale of a few subsequent backbone bonds , which maintain a high local flexibility . only for @xmath159 the data are compatible with a relation @xmath160 where @xmath161 and @xmath162 are phenomenological constants , and @xmath163 depends on both @xmath2 , @xmath1 , and @xmath164 distinctly ( but can not be accurately obtained from our simulations , because for @xmath165 the statistical accuracy of the data deteriorates . ) obviously , the relation ( eq . ( [ eq12 ] ) ) @xmath166 fails , but it seems tempting to identify an effective persistence length @xmath167 as @xmath168 . however , when we would define the persistence length in this way , we obtain the result that @xmath167 depends on both @xmath0 and on @xmath164 , not only on the side chain length @xmath2 and grafting density @xmath1 ( fig . [ fig12 ] and table [ table1 ] ) . for the bead - spring model , one often estimates that the length unit @xmath169 physically corresponds to about @xmath170 . the data for @xmath171 at @xmath34 and good solvent conditions for @xmath172 then would span a range from about @xmath173 to about @xmath174 , i.e. a similar range as proposed in recent experiments@xmath175 . but already our previous work on the athermal bond fluctuation model @xcite has given some evidence , that defining a persistence length from a fit of the data for @xmath155 to eq . ( [ eq15 ] ) is not suitable to obtain a measure of the local intrinsic stiffness , since @xmath167 depends on @xmath0 . the present data show that @xmath167 depends on @xmath164 as well , and hence is not just controlled by the chemical architecture of the bottle brush ( via the parameters , @xmath1 , @xmath2 , and @xmath0 ) , but depends on solvent conditions as well . thus , we argue that the physical significance of a persistence length @xmath167 extracted from bond orientational correlations in this way is very doubtful , even at the theta point . we also note that sometimes , due to curvature on the semi - log plot ( e.g. fig . [ fig11](d ) ) such fits are ill - defined . finally , we consider the possible validity of the kratky - porod result for the end - to - end distance of the chains at the theta temperature , where for @xmath143 we have ( apart from logarithmic corrections ) @xmath176 , i.e. for this property a formula analogous to gaussian chains holds . the kratky - porod result describes the crossover from rods to gaussian chains , @xmath177 \ , . \label{eq16}\ ] ] we can define an effective exponent @xmath178 in terms of the logarithmic derivative of this function , fig . [ fig13 ] ( a),(b ) , @xmath179 } { d \ln ( l / l_p)}=2\nu_{\rm eff}(l / l_p)-1 \ , . \label{eq17}\ ] ] using the data for @xmath44 at @xmath180 from fig . [ fig8](a)(c ) ( and similar data not shown for @xmath119 ) we hence obtain estimates for the ratios @xmath181 for the various choices of @xmath2 for both @xmath35 and @xmath34 . note that the errors in our estimation of @xmath44 translate into rather large errors in our estimates for @xmath182 ( both these errors of our data are indicated in fig . [ fig13](a)(b ) ) . since we expect that the actual variation of @xmath119 with @xmath0 exhibits slight curvature ( although this is hardly detected in fig . [ fig8](a)(c ) ) , we use only this intermediate value of @xmath180 to estimate the relation between @xmath2 and @xmath182 , and we use neither @xmath40 nor @xmath183 for this purpose . if the contour length @xmath184 of the kratky - porod model , eq . ( [ eq16 ] ) , would simply be the `` chemical '' contour length @xmath185 , where the bond length @xmath85 connecting two neighboring monomers along the backbone is @xmath186 , our results for @xmath182 would readily yield explicit results for @xmath8 . however , using then eq . ( [ eq16 ] ) to compute @xmath187 yields an overestimation by about a factor of five for @xmath34 . this discrepancy proves that @xmath184 is significantly smaller than @xmath188 . since eq . ( 17 ) implies that there is a unique correspondence between @xmath44 and the ratio @xmath189 , the factor of five discrepancy for @xmath190 ( which is proportional to the product of @xmath191 and @xmath184 ) means that both @xmath191 and @xmath184 must be smaller by the same factor ( at about @xmath192 ) . thus we define @xmath193 , where @xmath194 is interpreted as the average projection of a backbone bond on the direction of the coarse - grained contour of the wormlike chain . thus we also can use the data shown in fig . [ fig13](a)(b ) to obtain explicit estimates for @xmath195 , taking into account @xmath193 instead of @xmath196 . interestingly , the mechanisms leading to @xmath197 is also evident in the presence of the constant @xmath198 in the fits of @xmath199 in figs . [ fig10 ] , [ fig11 ] . using then the estimates @xmath200 and @xmath201 ( @xmath202 , the data for @xmath203 at @xmath87 are roughly compatible with eq . ( [ eq16 ] ) , when we use @xmath204 from fig . [ fig13](a)(b ) and take @xmath205 , as is shown in fig . [ fig13](c)(d ) . in this way , we have defined a correction factor @xmath206 in fig . [ fig13 ] , which was assumed to depend neither on @xmath2 nor on @xmath0 . note that while only @xmath180 was used in fig . [ fig13](a)(b ) , data for @xmath40 , and @xmath42 are included in fig . [ fig13](c)(d ) . the resulting values of @xmath195 which hence by construction do not depend on @xmath0 ( table [ table1 ] ) are considered to be the most reliable estimates for the considered model . however , we emphasize that eq . ( [ eq16 ] ) for bottle brushes is useful only if the solution is at theta conditions , but not in the good solvent regime . using the data for @xmath44 at @xmath180 from fig . [ fig8](a)(c ) ( and similar data not shown for @xmath119 ) we hence obtain estimates for the ratios @xmath181 for the various choices of @xmath2 for both @xmath35 and @xmath34 . in the present paper , a coarse - grained bead - spring model for bottle brush polymers was studied by molecular dynamics methods , varying both the chain length @xmath0 of the backbone and of the side chains ( @xmath2 ) , for two values of the grafting density , under variable solvent conditions . the main emphasis of the present work was a study of the various characteristic lengths describing the conformation of the macromolecule , contrasting the behavior under theta conditions with the behavior in the good solvent regime . also a comparison with corresponding results for the bond fluctuation model has been performed ; this athermal model represents very good solvent conditions , and while for corresponding parameters it yields results that are rather similar to the results for the bead spring model in the good solvent regime , it has the advantage that much larger values of @xmath0 can be studied . among the quantities that have been studied are the end - to - end distances of side chains and of the backbone , as well as the respective gyration radii , but also mesoscopic lengths that are particularly popular when a description in terms of the kratky - porod wormlike chain model is attempted , namely `` the '' persistence length and `` the '' contour length of the wormlike chain . we have shown that for the range of rather short side chain lengths @xmath2 that is only accessible , either in simulation or experiment , one often finds that a description in terms of power laws with effective exponents is adequate . specifically , the side chain radii vary as @xmath207 , and @xmath208 ( at the theta temperature @xmath209 ) or @xmath210 to @xmath211 ( in the good solvent regime ) , see fig . these effective exponents ( for the present models with flexible backbones ) are the same as for otherwise equivalent models with stiff backbones , and also are practically independent of the backbone chain length . while the resulting effective exponents are systematically smaller than the asymptotic values predicted theoretically for the limit @xmath45 , @xmath212 ( good solvents ) or @xmath213 ( theta solvents ) , respectively , they are comparable to the results of pertinent experiments : zhang et al . @xcite report @xmath214 for a bottle brush where the chemical nature of the main chain and the side chains is identical . for other systems exponents in the range @xmath215 are reported @xcite , again values comparable to our findings . some experimental studies also report that somewhat different results occur when different solvents are used @xcite , but a systematic variation of solvent quality to our knowledge has not yet been performed . our results have shown that for @xmath216 the side chain radii vary only very weakly with temperature ( which in our model causes the change in solvent quality ) , see fig . [ fig4 ] . also the backbone end - to - end distance shows a power law variation with the side chain length , @xmath217 ( fig . [ fig5 ] - [ fig7 ] ) , reflecting thus the systematic backbone stiffening with increasing side chain length . in this case , however , the effective exponent depends on the backbone chain length @xmath0 , for @xmath218 , before a saturation at @xmath219 ( @xmath35 ) or @xmath220 ( @xmath34 ) occurs , in the case of the bond fluctuation model . even somewhat larger exponents are observed for the bead - spring model ( up to @xmath221 , fig . [ fig7 ] ) . a further effective exponent describes the variation of the backbone end - to - end distance and gyration radius with backbone chain length ( figs . [ fig8 ] , [ fig9 ] ) . we have interpreted this variation as a crossover from rod - like behavior at small @xmath0 to self - avoiding walk - like behavior ( under good solvent conditions ) or random walk - like behavior ( under theta conditions ) with increasing @xmath0 . finally , `` the '' persistence length has been estimated in various ways , and demonstrating that the different estimates are not mutually consistent with each other it was argued that a unique persistence length does not exist . while a regime of exponential decay of the bond autocorrelation function often can be observed over some intermediate range of contour distances ( fig . [ fig10 ] , [ fig11 ] ) , the resulting estimates for a persistence length do not only depend on side chain length @xmath2 and grafting density @xmath1 , but also on backbone chain length @xmath0 and on temperature @xmath164 ( fig . [ fig12 ] ) . however , when one studies the variation of the end - to - end distance of the backbone for @xmath209 an analysis in terms of the kratky - porod wormlike chain model \{eq . ( [ eq16 ] ) } becomes feasible ( fig . [ fig13 ] ) . but one must not identify the contour length @xmath184 implied by this model with the `` chemical '' contour length @xmath222 , where @xmath85 is the actual bond length , but rather one has @xmath193 with @xmath201 distinctly smaller than @xmath85 . this effect results from the flexibility of the backbone on small scales ; only on the scale of several backbone bonds does the stiffening due to the mutual side chain repulsions come into play . thus , at the theta point both an effective contour length @xmath184 and a persistence length @xmath223 are well - defined quantities , in terms of a fit of the data to the kratky - porod model , while under good solvent conditions such an analysis is not appropriate . we hope that the present work stimulates some more systematic experimental work on these issues . of course , for real bottle brushes where the chemical nature of the backbone often differs from the side chains , the interesting possibility arises that the backbone and the side chains have rather different theta temperatures . another subtle issue concerns the precise location of theta conditions . while we expect that for the limit @xmath143 at fixed ratio @xmath224 the theta conditions coincide with the theta conditions for simple linear polymers ( @xmath143 , @xmath225 ) , it is not clear whether the location of the theta point for the limit @xmath143 at fixed ( small ) @xmath2 depends significantly on @xmath2 . thus , also more theoretical work clearly is needed . thanks for financial support by the austrian science foundation within the sfb vicom ( grant f41 ) , and also the max planck fellowship during the time he was in mainz . h. received funding from the deutsche forschungsgemeinschaft ( dfg ) , grant no sfb 625/a3 . we are grateful for extensive grants of computer time at the juropa under the project no hmz03 and softcomp computers at the jlich supercomputing centre ( jsc ) . information about algorithms and implementation details for the gromacs package can be found at url= http://www.gromacs.org ; h.j.c . berendsen , d. van der spoel and r. van drunen comp . . comm . * 91 * , 43 ( 1995 ) ; e. lindahl , b. hess and d. van der spoel j. mol . mod . * 7 * , 306 ( 2001 ) . .estimates of the effective persistence length @xmath226 shown in fig . [ fig12 ] are listed for the bs model with the grafting densities @xmath35 and @xmath227 at temperatures @xmath87 and @xmath228 . various values of the backbone length @xmath0 and the side chain length @xmath2 are chosen here . all length quoted in this table are given in units of @xmath229 ( with @xmath230 ) . note that @xmath231 depends not only on @xmath2 and @xmath1 , but also on @xmath0 and @xmath164 , and hence does not seem as a characteristic of intrinsic chain stiffness . estimates of @xmath195 shown in fig . [ fig13 ] are also listed for comparison . [ cols= " > , > , > , > , > , > " , ] ( a ) ( see eq . ( [ eq6 ] ) ) versus the time @xmath57 in units of @xmath232 . several cases are shown : @xmath35 , @xmath180 , @xmath233 , @xmath87 ( a ) @xmath86 , @xmath180 , @xmath234 , @xmath87 ( b ) ; and @xmath86 , @xmath183 , @xmath235 , @xmath87 ( c ) . the fluctuations of @xmath76 change sign initially at @xmath236 ( a ) , @xmath81 , and @xmath237 ( c ) , as shown in the inserts.,title="fig : " ] ( b ) ( see eq . ( [ eq6 ] ) ) versus the time @xmath57 in units of @xmath232 . several cases are shown : @xmath35 , @xmath180 , @xmath233 , @xmath87 ( a ) @xmath86 , @xmath180 , @xmath234 , @xmath87 ( b ) ; and @xmath86 , @xmath183 , @xmath235 , @xmath87 ( c ) . the fluctuations of @xmath76 change sign initially at @xmath236 ( a ) , @xmath81 , and @xmath237 ( c ) , as shown in the inserts.,title="fig : " ] + ( c ) ( see eq . ( [ eq6 ] ) ) versus the time @xmath57 in units of @xmath232 . several cases are shown : @xmath35 , @xmath180 , @xmath233 , @xmath87 ( a ) @xmath86 , @xmath180 , @xmath234 , @xmath87 ( b ) ; and @xmath86 , @xmath183 , @xmath235 , @xmath87 ( c ) . the fluctuations of @xmath76 change sign initially at @xmath236 ( a ) , @xmath81 , and @xmath237 ( c ) , as shown in the inserts.,title="fig : " ] ( a ) between successive monomers , @xmath238 , versus side chain length @xmath2 , for @xmath35 ( a ) and @xmath86 ( b ) . data for the bead spring model at two temperatures ( @xmath87 and @xmath51 ) , as indicated in the figure are included , as well as three backbone chain lengths ( @xmath40 , @xmath41 , and @xmath42 , respectively ) . for comparison , also data for the athermal bond fluctuation model for comparable backbone chain length are included . straight lines indicate effective exponents @xmath239 ( @xmath87 ) or @xmath240 ( @xmath51 ) in case ( a ) , and @xmath208 ( @xmath87 ) or @xmath241 ( @xmath51 ) in case ( b ) . for the bond fluctuation model under very good solvent conditions the slightly larger effective exponent ( @xmath242 for @xmath35 , and @xmath243 for @xmath86 ) than for the off - lattice model with @xmath51 results.,title="fig : " ] ( b ) between successive monomers , @xmath238 , versus side chain length @xmath2 , for @xmath35 ( a ) and @xmath86 ( b ) . data for the bead spring model at two temperatures ( @xmath87 and @xmath51 ) , as indicated in the figure are included , as well as three backbone chain lengths ( @xmath40 , @xmath41 , and @xmath42 , respectively ) . for comparison , also data for the athermal bond fluctuation model for comparable backbone chain length are included . straight lines indicate effective exponents @xmath239 ( @xmath87 ) or @xmath240 ( @xmath51 ) in case ( a ) , and @xmath208 ( @xmath87 ) or @xmath241 ( @xmath51 ) in case ( b ) . for the bond fluctuation model under very good solvent conditions the slightly larger effective exponent ( @xmath242 for @xmath35 , and @xmath243 for @xmath86 ) than for the off - lattice model with @xmath51 results.,title="fig : " ] + for four chain lengths ( @xmath36 , @xmath37 , @xmath38 , and @xmath39 ) , and three backbone lengths ( @xmath40 , @xmath41 , and @xmath42 ) , respectively . all data are for the bead spring model and @xmath34 . data for bottle - brush polymers with flexible backbone are connected by curves to guide the reader s eyes . data are only shown for @xmath87 and @xmath228 for the rigid backbone case . ] ( a ) versus side chain length @xmath2 , for the bond fluctuation model with @xmath35 ( a ) and @xmath86 ( b ) including different choices of @xmath0 , as indicated.,title="fig : " ] ( b ) versus side chain length @xmath2 , for the bond fluctuation model with @xmath35 ( a ) and @xmath86 ( b ) including different choices of @xmath0 , as indicated.,title="fig : " ] + ( a ) , extracted from the slope of the straight lines in fig . [ fig5 ] , versus the backbone chain length @xmath0 for @xmath35 ( a ) and @xmath86 . all data are for the athermal bond fluctuation model.,title="fig : " ] ( b ) , extracted from the slope of the straight lines in fig . [ fig5 ] , versus the backbone chain length @xmath0 for @xmath35 ( a ) and @xmath86 . all data are for the athermal bond fluctuation model.,title="fig : " ] ( a ) versus @xmath2 . all data are for the bead spring model with @xmath86 at both @xmath51 ( a ) and @xmath87 ( b ) . three backbone lengths are shown as indicated.,title="fig : " ] ( b ) versus @xmath2 . all data are for the bead spring model with @xmath86 at both @xmath51 ( a ) and @xmath87 ( b ) . three backbone lengths are shown as indicated.,title="fig : " ] + ( a ) versus backbone chain length @xmath0 , for @xmath35 ( a)(b ) and @xmath86 ( c)(d ) , including both @xmath87 ( a)(c ) and @xmath51 ( b)(d ) , for the bead - spring model . several side chain lengths @xmath2 are included as indicated . straight lines indicate a fit with effective exponents @xmath44.,title="fig : " ] ( b ) versus backbone chain length @xmath0 , for @xmath35 ( a)(b ) and @xmath86 ( c)(d ) , including both @xmath87 ( a)(c ) and @xmath51 ( b)(d ) , for the bead - spring model . several side chain lengths @xmath2 are included as indicated . straight lines indicate a fit with effective exponents @xmath44.,title="fig : " ] + ( c ) versus backbone chain length @xmath0 , for @xmath35 ( a)(b ) and @xmath86 ( c)(d ) , including both @xmath87 ( a)(c ) and @xmath51 ( b)(d ) , for the bead - spring model . several side chain lengths @xmath2 are included as indicated . straight lines indicate a fit with effective exponents @xmath44.,title="fig : " ] ( d ) versus backbone chain length @xmath0 , for @xmath35 ( a)(b ) and @xmath86 ( c)(d ) , including both @xmath87 ( a)(c ) and @xmath51 ( b)(d ) , for the bead - spring model . several side chain lengths @xmath2 are included as indicated . straight lines indicate a fit with effective exponents @xmath44.,title="fig : " ] + ( a ) for the bond fluctuation model versus backbone chain length @xmath0 , for @xmath244 , and several side chain lengths @xmath2 , for @xmath35 ( a ) and @xmath86 ( b ) . part ( c ) shows the plot of rescaled radius of gyration @xmath245 of the bottle brush polymers versus @xmath0 for @xmath86 . the persistence length @xmath130 is determined by the values of plateau . part ( d ) shows a crossover scaling plot , collapsing for @xmath132 , @xmath133 , @xmath134 , and @xmath135 , @xmath86 , and all data for @xmath246 on a master curve , that describes the crossover from rods @xmath247 to swollen coils @xmath248 . for this purpose , @xmath0 is rescaled with the blob diameter @xmath131 , which has been determined to be @xmath249 , @xmath37 , @xmath133 , and @xmath250 for @xmath132 , @xmath133 , @xmath134 , and @xmath135 , respectively @xcite.,title="fig : " ] ( b ) for the bond fluctuation model versus backbone chain length @xmath0 , for @xmath244 , and several side chain lengths @xmath2 , for @xmath35 ( a ) and @xmath86 ( b ) . part ( c ) shows the plot of rescaled radius of gyration @xmath245 of the bottle brush polymers versus @xmath0 for @xmath86 . the persistence length @xmath130 is determined by the values of plateau . part ( d ) shows a crossover scaling plot , collapsing for @xmath132 , @xmath133 , @xmath134 , and @xmath135 , @xmath86 , and all data for @xmath246 on a master curve , that describes the crossover from rods @xmath247 to swollen coils @xmath248 . for this purpose , @xmath0 is rescaled with the blob diameter @xmath131 , which has been determined to be @xmath249 , @xmath37 , @xmath133 , and @xmath250 for @xmath132 , @xmath133 , @xmath134 , and @xmath135 , respectively @xcite.,title="fig : " ] + ( c ) for the bond fluctuation model versus backbone chain length @xmath0 , for @xmath244 , and several side chain lengths @xmath2 , for @xmath35 ( a ) and @xmath86 ( b ) . part ( c ) shows the plot of rescaled radius of gyration @xmath245 of the bottle brush polymers versus @xmath0 for @xmath86 . the persistence length @xmath130 is determined by the values of plateau . part ( d ) shows a crossover scaling plot , collapsing for @xmath132 , @xmath133 , @xmath134 , and @xmath135 , @xmath86 , and all data for @xmath246 on a master curve , that describes the crossover from rods @xmath247 to swollen coils @xmath248 . for this purpose , @xmath0 is rescaled with the blob diameter @xmath131 , which has been determined to be @xmath249 , @xmath37 , @xmath133 , and @xmath250 for @xmath132 , @xmath133 , @xmath134 , and @xmath135 , respectively @xcite.,title="fig : " ] ( d ) for the bond fluctuation model versus backbone chain length @xmath0 , for @xmath244 , and several side chain lengths @xmath2 , for @xmath35 ( a ) and @xmath86 ( b ) . part ( c ) shows the plot of rescaled radius of gyration @xmath245 of the bottle brush polymers versus @xmath0 for @xmath86 . the persistence length @xmath130 is determined by the values of plateau . part ( d ) shows a crossover scaling plot , collapsing for @xmath132 , @xmath133 , @xmath134 , and @xmath135 , @xmath86 , and all data for @xmath246 on a master curve , that describes the crossover from rods @xmath247 to swollen coils @xmath248 . for this purpose , @xmath0 is rescaled with the blob diameter @xmath131 , which has been determined to be @xmath249 , @xmath37 , @xmath133 , and @xmath250 for @xmath132 , @xmath133 , @xmath134 , and @xmath135 , respectively @xcite.,title="fig : " ] + ( a ) vs. @xmath154 for the bead spring model with @xmath35 . data are chosen for @xmath40 ( a)(b ) and @xmath183 ( c)(d ) at @xmath87 ( a)(c ) and @xmath51 ( b)(d ) , respectively . several choices of @xmath2 are shown as indicated . straight lines illustrate fits to @xmath251 , with constants @xmath161 , @xmath162 quoted in the figure.,title="fig : " ] ( b ) vs. @xmath154 for the bead spring model with @xmath35 . data are chosen for @xmath40 ( a)(b ) and @xmath183 ( c)(d ) at @xmath87 ( a)(c ) and @xmath51 ( b)(d ) , respectively . several choices of @xmath2 are shown as indicated . straight lines illustrate fits to @xmath251 , with constants @xmath161 , @xmath162 quoted in the figure.,title="fig : " ] + ( c ) vs. @xmath154 for the bead spring model with @xmath35 . data are chosen for @xmath40 ( a)(b ) and @xmath183 ( c)(d ) at @xmath87 ( a)(c ) and @xmath51 ( b)(d ) , respectively . several choices of @xmath2 are shown as indicated . straight lines illustrate fits to @xmath251 , with constants @xmath161 , @xmath162 quoted in the figure.,title="fig : " ] ( d ) vs. @xmath154 for the bead spring model with @xmath35 . data are chosen for @xmath40 ( a)(b ) and @xmath183 ( c)(d ) at @xmath87 ( a)(c ) and @xmath51 ( b)(d ) , respectively . several choices of @xmath2 are shown as indicated . straight lines illustrate fits to @xmath251 , with constants @xmath161 , @xmath162 quoted in the figure.,title="fig : " ] + ( a ) vs. @xmath154 for the bead spring model with @xmath86 and @xmath180 at the temperatures @xmath87 ( a ) and @xmath51 ( b ) , respectively . several choices of @xmath2 are shown as indicated . straight lines illustrate fits to a @xmath252 , with constants @xmath161 , @xmath162 as quoted in the figure.,title="fig : " ] ( b ) vs. @xmath154 for the bead spring model with @xmath86 and @xmath180 at the temperatures @xmath87 ( a ) and @xmath51 ( b ) , respectively . several choices of @xmath2 are shown as indicated . straight lines illustrate fits to a @xmath252 , with constants @xmath161 , @xmath162 as quoted in the figure.,title="fig : " ] + ( a ) as a function of side chain length @xmath2 for @xmath87 ( a)(c ) and @xmath51 ( b)(d ) for several choices of @xmath0 . results for @xmath86 and @xmath35 are shown as indicated and the exponent estimates @xmath104 extracted from the slope are listed . values of @xmath167 are also listed in table [ table1].,title="fig : " ] ( b ) as a function of side chain length @xmath2 for @xmath87 ( a)(c ) and @xmath51 ( b)(d ) for several choices of @xmath0 . results for @xmath86 and @xmath35 are shown as indicated and the exponent estimates @xmath104 extracted from the slope are listed . values of @xmath167 are also listed in table [ table1].,title="fig : " ] + ( c ) as a function of side chain length @xmath2 for @xmath87 ( a)(c ) and @xmath51 ( b)(d ) for several choices of @xmath0 . results for @xmath86 and @xmath35 are shown as indicated and the exponent estimates @xmath104 extracted from the slope are listed . values of @xmath167 are also listed in table [ table1].,title="fig : " ] ( d ) as a function of side chain length @xmath2 for @xmath87 ( a)(c ) and @xmath51 ( b)(d ) for several choices of @xmath0 . results for @xmath86 and @xmath35 are shown as indicated and the exponent estimates @xmath104 extracted from the slope are listed . values of @xmath167 are also listed in table [ table1].,title="fig : " ] + ( a ) versus @xmath182 as predicted from the kratky - porod model , eqs . ( [ eq16 ] ) , ( [ eq17 ] ) ( full curves ) . the numbers for @xmath182 extracted for @xmath35 ( a ) and @xmath86 ( b ) are quoted in the figure for various @xmath2 at the abscissa . the average of the exponents @xmath44 from fig . [ fig8](a)(c ) and from @xmath128 ( not shown ) are quoted on the ordinate . log - log plot of @xmath253 vs. @xmath204 for @xmath35 ( c ) and @xmath86 ( d ) . several choices of @xmath2 are shown as indicated . data plotted by the same symbol correspond to @xmath40 , @xmath41 , and @xmath42 from left to right . the prediction for the kratky - porod model \{eq . ( [ eq16 ] ) } is also shown for comparison . approximate data collapse is obtained by introducing a factor @xmath254 ( @xmath255 and @xmath256 for @xmath35 and @xmath227 , respectively ) , cf . values of @xmath195 are also listed in table [ table1].,title="fig : " ] ( b ) versus @xmath182 as predicted from the kratky - porod model , eqs . ( [ eq16 ] ) , ( [ eq17 ] ) ( full curves ) . the numbers for @xmath182 extracted for @xmath35 ( a ) and @xmath86 ( b ) are quoted in the figure for various @xmath2 at the abscissa . the average of the exponents @xmath44 from fig . [ fig8](a)(c ) and from @xmath128 ( not shown ) are quoted on the ordinate . log - log plot of @xmath253 vs. @xmath204 for @xmath35 ( c ) and @xmath86 ( d ) . several choices of @xmath2 are shown as indicated . data plotted by the same symbol correspond to @xmath40 , @xmath41 , and @xmath42 from left to right . the prediction for the kratky - porod model \{eq . ( [ eq16 ] ) } is also shown for comparison . approximate data collapse is obtained by introducing a factor @xmath254 ( @xmath255 and @xmath256 for @xmath35 and @xmath227 , respectively ) , cf . values of @xmath195 are also listed in table [ table1].,title="fig : " ] + ( c ) versus @xmath182 as predicted from the kratky - porod model , eqs . ( [ eq16 ] ) , ( [ eq17 ] ) ( full curves ) . the numbers for @xmath182 extracted for @xmath35 ( a ) and @xmath86 ( b ) are quoted in the figure for various @xmath2 at the abscissa . the average of the exponents @xmath44 from fig . [ fig8](a)(c ) and from @xmath128 ( not shown ) are quoted on the ordinate . log - log plot of @xmath253 vs. @xmath204 for @xmath35 ( c ) and @xmath86 ( d ) . several choices of @xmath2 are shown as indicated . data plotted by the same symbol correspond to @xmath40 , @xmath41 , and @xmath42 from left to right . the prediction for the kratky - porod model \{eq . ( [ eq16 ] ) } is also shown for comparison . approximate data collapse is obtained by introducing a factor @xmath254 ( @xmath255 and @xmath256 for @xmath35 and @xmath227 , respectively ) , cf . values of @xmath195 are also listed in table [ table1].,title="fig : " ] ( d ) versus @xmath182 as predicted from the kratky - porod model , eqs . ( [ eq16 ] ) , ( [ eq17 ] ) ( full curves ) . the numbers for @xmath182 extracted for @xmath35 ( a ) and @xmath86 ( b ) are quoted in the figure for various @xmath2 at the abscissa . the average of the exponents @xmath44 from fig . [ fig8](a)(c ) and from @xmath128 ( not shown ) are quoted on the ordinate . log - log plot of @xmath253 vs. @xmath204 for @xmath35 ( c ) and @xmath86 ( d ) . several choices of @xmath2 are shown as indicated . data plotted by the same symbol correspond to @xmath40 , @xmath41 , and @xmath42 from left to right . the prediction for the kratky - porod model \{eq . ( [ eq16 ] ) } is also shown for comparison . approximate data collapse is obtained by introducing a factor @xmath254 ( @xmath255 and @xmath256 for @xmath35 and @xmath227 , respectively ) , cf . values of @xmath195 are also listed in table [ table1].,title="fig : " ] +

a novel approach is developed to find the three - body breakup amplitudes and cross sections within the modified faddeev equation framework . the method is based on the lattice - like discretization of the three - body continuum with a three - body stationary wave - packet basis in momentum space . the approach makes it possible to simplify drastically all the three- and few - body breakup calculations due to discrete wave - packet representations for the few - body continuum and simultaneous lattice representation for all the scattering operators entering the integral equation kernels . as a result , the few - body breakup can be treated as a particular case of multi - channel scattering in which part of the channels represents the true few - body continuum states . as an illustration for the novel approach , an accurate calculations for the three - body breakup process with non - local and local interactions are calculated . the results obtained reproduce nicely the benchmark calculation results using the traditional faddeev scheme which requires much more tedious and time - consuming calculations .

1205.5730

the last decades have inaugurated great success in precise _ ab - initio _ calculations for few - body scattering processes @xcite . these calculations made it possible to describe accurately the results of numerous recent experiments on elastic @xmath2 scattering at energies up to 350 mev and also the three - body breakup @xmath0 at low and moderate energies @xmath3 mev . however some problems remain unsettled even at such low energies . these are the so - called @xmath4-puzzle ( as well as other puzzles for various tensor and vector analyzing powers ) in elastic scattering , the problems with an adequate description of the pairwise @xmath5-channel contribution to three - body breakup at low energies @xcite and breakup cross section in some particular three - particle configurations such as the quasi - free scattering @xcite and the space - star @xcite configurations . the most plausible reason for the visible discrepancies with experimental data in this area is likely not insufficient accuracy of numerical calculations but rather some deficiency in the input @xmath6- and @xmath7-interactions . at the same time , the progress in the field of precise few - nucleon calculations , particularly in testing of new models for @xmath7-interactions , is restrained strongly by a high number of very complicated few - nucleon calculations , especially above the three - body threshold . because of these complications of traditional computational schemes for the direct solution of the faddeev yakubovsky equations , one observes a rise of interest in recent years in alternative approaches @xcite to calculate the scattering observables by simpler methods . among such alternative approaches one can note a preference for the so called @xmath8-methods . these methods are based on expansions of the scattering solution into a basis of square - integrable functions @xcite . such @xmath8-methods proved to be very well suited and quite efficient for numerous applications . one of the most successful approaches of this type is the continuum - discretized coupled - channel ( cdcc ) method developed three decades ago for treatment of breakup processes in direct nuclear reactions @xcite . the cdcc approach in its traditional form was unable to treat other channels than elastic scattering and projectile ( or target ) breakup . recently a few groups generalized the traditional cdcc approach to scattering of three - fragment projectiles by a stable target @xcite . however this generalized approach can be considered as a hybrid method : @xmath8 discretization of inner motion in the three - body projectile and the traditional treatment of a coupled - channel problem on the other hand , the present authors have developed some alternative @xmath8-technique @xcite which is based on the idea of complete continuum discretization with a special stationary wave - packet basis in momentum space ( three - body lattice basis ) . the basic distinction of such an approach from the traditional cdcc - scheme for the three - body systems is that the wave - packet approach is dealing with a full discretization of the three - body continuum . in other words , the discretization on _ both _ jacobi coordinates is used here rather than the discretization on the alone coordinate of the projectile inner motion as in the cdcc approach2 scattering and also the breakup @xmath9 process using the laguerre polynomial basis . unfortunately this prospective approach has not received any further development . ] . our approach with the global discretization over all valence coordinates leads immediately to a few important advantages in the accurate treatment of few - body scattering . among those the following are the most important : * the few - body scattering problem is consistently formulated in a hilbert space of three - body _ normalized _ states , similarly to the bound - state problem . * the approach employs the integral equation framework of scattering theory instead of the differential equation approach ( e.g. in the cdcc ) where the boundary conditions in few - body scattering channels are not easy to formulate , especially in terms of the @xmath8 basis used . contrary to this , the integral equation formulation allows to avoid any explicit account of the boundary conditions . * when working within the wave - packet formulation of the scattering problem one can derive explicit formulas for some scattering operators ( e.g. channel resolvents ) . such fully analytical finite - dimensional approximations , being substituted into integral equation kernels , lead immediately to their algebraic matrix analogues . thus , our final equations are simple matrix linear equations with regular matrices . in our previous works @xcite we have demonstrated how to find the elastic @xmath10 scattering amplitudes in lattice representation . in the present paper , we generalize the technique to the three - body breakup treatment . so we present here the complete formalism for determination of three - body breakup amplitudes . it is important to emphasize in this connection that the accurate treatment of three - body breakup within the @xmath8 type approach is much less obvious than that of elastic ones and thus requires some additional delicate theoretical studies . in particular , the matrix elements in the breakup amplitude are not truncated over all spatial coordinates ( in contrast to the elastic and rearrangement amplitudes ) , so the validity of the @xmath8 scheme in the treatment of the breakup processes should be studied carefully . as some substantiation for such approach one can consider the three - body breakup calculations within the cdcc - approach where the discretization of the continuum in the projectile inner subhamiltonian has been used for the description of the breakup amplitudes @xcite . so , the natural generalization of such a partial continuum discretization to the case of full three- and few - body continuum within the faddeev equation approach is an important next step . moreover , this fully discretized approach studied in the present work allows to simplify drastically all calculations and makes it more universal and elegant . the present work has the following structure . in the section ii , a three - body lattice - like free wave - packet basis is described in detail together with a similar basis for the channel hamiltonian . here we also discuss the properties of these bases . the complete formalism for elastic scattering and breakup , as applied to the @xmath2-system in the packet representation is presented in section iii . in the section iv , a few useful numerical illustrations and their comparison with the standard faddeev benchmark calculations are given . our results are summarized in the section v. for the sake of convenience for the reader we add three appendices . in appendix a we describe the detailed scheme for calculation of the three - body overlap matrix in the three - body lattice basis for recoupling of different jacobi coordinates . in appendix b we give the convenient wave - packet formalism for the solution of three - body scattering problem with separable pairwise interactions . in the last appendix c we discuss some features of our numerical calculations . we consider here the problem of a scattering of three identical particles 1 , 2 and 3 ( nucleons ) with mass @xmath11 , interacting via pairwise short - range potentials @xmath12 @xmath13 . it is convenient to use three jacobi momentum sets @xmath14 corresponding to three channel hamiltonians @xmath15 @xmath13 which define the asymptotic states of the system . for example , the channel hamiltonian @xmath16 has the form of the direct sum of two - body subhamiltonians @xmath17 where subhamiltonian @xmath18 describes @xmath1 subsystem consisting of particles 2 and 3 with interaction @xmath19 and subhamiltonian @xmath20 corresponds to the free motion of nucleon 1 relative to center of mass of the subsystem \{23}. as we study the identical particle system , we will omit further , where it is possible , the jacobi index @xmath21 . we start from the free - motion three - body hamiltonian defined in the given jacobi momentum set ( * p , q * ) @xmath22 where the subhamiltonian @xmath23 defines the free motion of two nucleons with the relative momentum @xmath24 and the subhamiltonian @xmath25 defines the free motion of the third nucleon with the momentum @xmath26 relative to the center of mass of given @xmath1 subsystem . now we will construct our three - body @xmath8 basis using discretization of the continua of the two above subhamiltonians . in doing this we will employ the complete sets of continuum wave functions @xmath27 and @xmath28 ( for every partial wave ) normalized according to the conditions : @xmath29 when discretizing , we truncate the continuum of @xmath23 and @xmath25 by maximal values @xmath30 and @xmath31 respectively , so that the continuous spectra above these values can be neglected . further , the selected energy regions @xmath32 $ ] and @xmath33 $ ] are divided onto non - overlapping bins @xmath34_{i=1}^m\}$ ] and @xmath35_{j=1}^n\}$ ] . such energy bins correspond to momentum bins @xmath36 $ ] and @xmath37 $ ] , so that the end - points of both sets are interrelated by conventional formulas @xmath38 and @xmath39 . to further simplify the notation , we will denote the intervals in the variable @xmath40 ( both the energy and momentum ones ) as @xmath41 and in the variable @xmath42 as @xmath43 . we use also the following notations for the widths of the corresponding momentum intervals : @xmath44 now let s define a set of _ free _ stationary wave packets ( wps ) as integrals of the plane waves ( corresponding to the _ free _ motion ) over the above momentum bins for both subhamiltonians : @xmath45 where @xmath46 and @xmath47 are some known weight functions and @xmath48 and @xmath49 are normalization factors , directly related to the weight functions @xmath50 so that the wp states are normalized to unity : @xmath51 it is important to stress , these wp states belong to _ a hilbert space _ ( similarly to the bound state functions ) and wp functions are square - integrable : in configuration space they vanish at infinity in contrast to the initial plane waves . but in the relevant restricted range of configuration space the wp states still resemble quite closely the exact scattering states taken at the bin center energy ( or momentum ) @xcite . the sets of such wp states @xmath52 and @xmath53 form an orthonormalized bases in hilbert space which can be used as normal @xmath8 bases , e.g. also for variational calculations . in our previous papers @xcite we have discussed the properties of wp s in detail . a distinctive feature of wp bases is that the matrices of the subhamiltonians found in such bases are diagonal : @xmath54 where values @xmath55 and @xmath56 are defined via corresponding end - points of bins @xmath41 and @xmath43 @xcite . the most useful property of wps is that the matrices of the resolvents @xmath57^{-1}$ ] and @xmath58^{-1}$ ] are diagonal in the corresponding wp bases and their elements have explicit analytical forms @xcite . different choices of weight functions lead to different sets of wps . in practical calculations in this work we use the momentum wave packets with the unit weight functions : @xmath59 it is easy to see that the overlap of such free momentum wp with a plane wave , i.e. the momentum representation of packet state ( [ ip ] ) itself has the form : @xmath60 where we have introduced a function @xmath61 which is equal to unity if the momentum @xmath40 belongs to the interval @xmath36 $ ] and vanishes in the other case . so , the wave packet @xmath62 takes a form of simple step - like function in the momentum representation . the sets of the constructed free wp states can be used to find two - body bound - states and also to solve a two - body scattering problem , e.g. for finding the two - body off - shell @xmath63-matrix @xcite . in the present paper we will use these two - body @xmath8 bases to construct three - body wps for solution of the three - body scattering problem . three - body wave packet states are built as direct products of two - body ones . however , here one should take into account the spin and angular parts of the functions . the total three - body wp basis function can be written as : latexmath:[\[\label{xij } the spin - angular state of the @xmath1 pair , @xmath65 is the spin - angular state of third nucleon , while @xmath66 is the set of the three - body quantum numbers . the state ( [ xij ] ) is a wp analog of the exact state of the three - body continuum @xmath67 for the free hamiltonian @xmath68 . the properties of such three - body wp s are very similar to those of two - body wave packets @xcite . in particular , the matrix of the three - body free hamiltonian @xmath68 and its resolvent @xmath69^{-1}$ ] are diagonal in the so constructed basis . in other words , such a wp basis defines _ an `` eigen '' wave - packet subspace _ for the free three - body hamiltonian @xmath68 . since the basis functions are the products of both step - like functions in variables @xmath40 and @xmath42 , the solution of the three - body scattering problem in such a basis corresponds to a formulation of the scattering problem _ on a two - dimensional momentum lattice_. therefore we will refer to such a basis as a _ lattice basis_. let us denote the two - dimensional bins ( i.e. the lattice cells ) as @xmath70 . in the few - body case , the lattice basis functions are constructed as direct products of the two - body free wps , so the basis space corresponds to a multi - dimensional lattice . in principle , using the above lattice basis , one can solve a general three - body scattering problem by projecting all the scattering operators onto such a basis . in particular , the matrix of the three - body free resolvent @xmath71 can be expressed in the above lattice representation fully analytically @xcite . let us consider the particle permutation operator @xmath72 which enters in the faddeev equation for three identical particles and is defined as @xmath73 the matrix of the operator @xmath72 in the lattice basis corresponds to the overlap between basis functions defined in different jacobi sets : @xmath74 where the argument 1 ( or 2 ) in the basis functions means a corresponding jacobi set . such matrix element can be calculated with the definition of the basis functions in momentum space ( [ proj_rule ] ) : @xmath75_{ij , i'j'}^{{\alpha}{\beta},{\alpha}'{\beta}'}\equiv\langle x_{ij}^{\gamma{\alpha}{\beta}}|p|x_{i'j'}^{\gamma{\alpha}'{\beta}'}\rangle=\nonumber\\ \label{perm}\int_{{\mathfrak{d}}_{ij}}dpdq\int_{{\mathfrak{d}}'_{i'j'}}dp'dq ' \times\frac{p^{\gamma}_{{\alpha}{\beta},{\alpha}'{\beta}'}(p , q , p',q')}{\sqrt{d_id_{i'}\bar{d}_j\bar{d}_{j'}}},\end{aligned}\ ] ] where the prime at the lattice cell @xmath76 indicates that the cell belongs to the other jacobi set while the @xmath77 is the kernel of particle permutation operator in a momentum space . this kernel , as is well known @xcite , is proportional to the product of the dirac delta and heaviside theta functions . however , due to `` packetting '' ( i.e integration over momentum bins in eq . ( [ perm ] ) ) these singularities get averaged over the cells of the momentum lattice and , as a result , the elements of the permutation operator matrix in the wp basis are finite . using the above `` packetting '' procedure and the hyperspherical momentum coordinates , the calculation of the matrix element in eq . ( [ perm ] ) can be done using only a one - dimensional numerical integration over the hypermomentum @xmath78 . the technique of this calculation for the s - wave basis functions is given in appendix a of the present paper . the generalization for higher partial waves is straightforward . it should be emphasized here that the fixed lattice - like form for the permutation operator matrix makes it possible to avoid the complicated and time consuming multi - dimensional interpolations of the current solution when solving the faddeev equations ( in momentum space ) by iterations in conventional approach @xcite . such numerous multi - dimensional interpolations at each iteration step take a big portion of the computational time in practical numerical procedure . when solving the four - body yakubovsky equations the dimension for these interpolations increase and thus the computational efforts get even higher . so , avoiding the very numerous multidimensional interpolations in each step of the iterations leads to a tremendous acceleration for all three - body calculations in momentum space . thus , the two - dimensional momentum lattice basis constructed above can be applied directly to solving the faddeev equations for the conventional transition operator @xmath79 . however , by using the very convenient form for the spectral representation of the resolvent operators in the wp basis one can employ some alternative ( but equivalent ) form of the faddeev equation , which makes it possible to avoid the time - consuming calculation of the fully off - shell @xmath63-matrix at many energies ( which requires to solve very often the lippmann - schwinger equations at every energy and for different spin - orbit channels ) and to replace it by calculating the resolvent of the @xmath1 subhamiltonian @xmath18 in the corresponding _ scattering _ wp - representation . the latter can be made easily by straightforward one - fold diagonalization of the @xmath18 subhamiltonian matrix . as has been demonstrated earlier @xcite the stationary wave packets can be built not only for free hamiltonians but also for perturbed two - body @xmath80 and three - body channel hamiltonians @xmath16 . in case of the @xmath18 subhamiltonian , its continua @xmath81 $ ] for every spin - angular configuration @xmath82 are divided into separate bins @xmath83_{k=1}^k\}$ ] and one can build the scattering wave packet for every such bin @xmath84 $ ] in the form @xmath85 i.e. as an integral over the exact scattering wave function @xmath86 on the energy interval @xmath87 . here we use the unit weight function and @xmath88 is the width of interval @xmath87 . it is easy to show that such packet states have the same properties with respect to their `` eigen '' hamiltonian @xmath18 as free wps with respect to the free hamiltonian @xmath23 . the only difference is that the set of scattering wp s should be accomplished with the bound state functions of @xmath18 ( if they exist ) . jointly with the possible bound state wave functions for the @xmath18 subhamiltonian , the scattering wp s form an orthonormalized basis in which both the matrix of the hamiltonian @xmath18 and the matrix of its resolvent @xmath89^{-1}$ ] are diagonal @xcite . the projection properties for the wp of @xmath18 will be similar to those for @xmath23 ( [ proj_rule ] ) , viz . : @xmath90 at first glance it may appear that the exact scattering wp basis is useless because its construction would require the knowledge of exact scattering wave functions of the hamiltonian @xmath18 . however , as has been demonstrated @xcite , the properties of the exact scattering wp s for @xmath18 are quite similar to those of respective pseudostates obtained by the diagonalization of the hamiltonian matrix in some complete @xmath8-basis . so that , in actual calculations one can replace the set of wp s @xmath91 by the set of respective pseudostates @xcite . such an @xmath8-basis can be used as a very good approximation for the free wp - basis ( [ ip ] ) . as a result of such hamiltonian matrix diagonalization , one gets a set of pseudostates @xmath92 together with a set of their eigenvalues @xmath93 . in this paper , we restrict ourselves to @xmath94-wave spin - dependent pair interactions only . we assume that there is a single bound state @xmath95 ( deuteron ) with binding energy @xmath96 in the @xmath1 spin - triplet channel and there are no bound states in the @xmath1 spin - singlet channel . in case of @xmath94-wave scattering with @xmath94-wave @xmath1 interactions , the indices @xmath82 , @xmath97 and @xmath98 in eq . ( [ xij ] ) include only the spin quantum numbers . so , below we will use the value of the spin of @xmath1-pair @xmath99 instead of index @xmath82 , while the index @xmath97 , which indicates the spin value of the third nucleon ( i.e. @xmath100 ) , will be omitted everywhere . the index @xmath98 defining the set of quantum numbers for three - body states is reduced to the total spin of the three - body system @xmath101 which in the @xmath94-wave case is equal to the total angular momentum of the system and therefore is conserved . after the above diagonalization in the spin - triplet channel ( @xmath102 ) one gets a set of pseudostate functions , the first of which @xmath103 with the energy @xmath96 , is an approximation for the deuteron wave function , while the other @xmath104 pseudostates with energies @xmath105 are localized in the continuum spectrum and correspond to scattering wp states for @xmath18 . in the spin - singlet channel there are no @xmath1 bound states , so that all functions @xmath106 in eq . ( [ exp_z ] ) are approximated by scattering wp s . it is important to note that as a by - product of our diagonalization procedure one gets simultaneously the discrete representation for @xmath1 partial phase shifts @xmath107 for all pseudo - states energies ( i.e. in one step ! ) see the detail in ref . @xcite . since the free wp basis functions ( in the momentum space ) are step - like functions , the momentum dependence of all functions expressed via such a basis have a histogram - like form . an example of the momentum dependence for the bound state ( deuteron ) function in such step - like basis in comparison with the exact function for yamaguchi triplet @xmath1 potential ( see appendix b ) is displayed in fig . [ fig0 ] . the fig . [ wpac ] displays the functions of two pseudostates ( with @xmath108 ) obtained in the lattice basis in comparison with the corresponding exact scattering wave packets which can be calculated exactly for the separable yamaguchi potential . it is interesting to see that although functions of the exact scattering wp s ( [ zi ] ) ( dashed lines in the figure ) have the logarithmic singularities at the boundaries of the `` on - shell '' interval ( i.e. the one which the state energy belongs to ) they are square - integrable as well as the free - motion wp s . it is clear from the comparison that the pseudostates composed from step - like wave packets reproduce very well the structure of the exact scattering wave packets `` on average '' . now having at our disposal the two - body bases for subhamiltonians @xmath18 and @xmath25 one can construct the three - body wp basis for the channel hamiltonian @xmath16 which defines the asymptotic motion in the system . the three - body wp states corresponding to the channel hamiltonian @xmath16 can be defined similar as the wp - states for the three - body free hamiltonian @xmath68 , i.e. as direct products of two - body wp states for @xmath25 and @xmath18 subhamiltonians ( jointly with the bound - state ) multiplied by the spin functions of the system : @xmath109 are the @xmath1 subsystem and the total three - body spins correspondingly only . since in case of @xmath94-wave @xmath1 pairwise interaction the spin- and isospin quantum numbers are interrelated uniquely by the pauli principle we can omit further the isospin parts of the wave functions and corresponding quantum numbers . ] . when using the above pseudostate approximation , these three - body states , similarly to two - body scattering wps , are related to the three - body lattice basis states by a simple rotation transformation ( similar to eq . ( [ exp_z ] ) ) : latexmath:[\[\label{rotation } hence , starting from the free wp bases for every pair subsystem one gets a set of basis states both for the three - body free @xmath68 and the channel @xmath16 hamiltonians . the basis defined in eq . ( [ si ] ) defines _ an `` eigen '' wp - subspace _ for the channel hamiltonian @xmath16 . this allows us to construct an analytical finite - dimensional approximation for the channel resolvent @xmath111^{-1}$ ] . indeed , the exact three - body channel resolvent is the convolution of the two - body subresolvents @xmath112 and @xmath113 : @xmath114 using further the spectral expansions for the two - body resolvents and integrating over @xmath115 , one gets an explicit expression for the channel resolvent @xmath116 as a sum of two terms @xmath117 . here the bound - continuum part @xmath118 is the spectral sum over the three - body states corresponding to the free motion of the deuteron relatively to the third nucleon . so , the imaginary part of @xmath118 is related to a discontinuity on the two - body cut of the riemann surface of the three - body energy @xmath119 . the continuum - continuum part @xmath120 of the channel resolvent includes the channel three - body states with the @xmath1 pair with interaction in the continuum and @xmath121 is defined by a discontinuity on the three - body cut on the energy surface ( see the details in ref . @xcite ) . projecting the exact channel resolvent onto the three - body channel wp basis defined in eq . ( [ si ] ) , one can find analytical formulas for the matrix elements of the @xmath116 operator . the respective matrix is diagonal in all wave - packet and spin indices : @xmath122 here the diagonal matrix elements @xmath123 are defined as integrals over the respective momentum bins and depend in general on the spectrum partition parameters ( i.e. the @xmath124 and @xmath125 values ) and the total energy @xmath119 only . they do not depend explicitly on the interaction potential @xmath19 . if the solution of scattering equations in the finite - dimensional wp basis converges with increasing the basis dimension , the final result turns out to be _ independent _ on the particular spectral partition parameters . we have found @xcite the explicit formulas for the resolvent matrix elements ( [ dg1 ] ) when one uses the energy wp s , i.e. wp s with the weight functions @xmath126 , @xmath127 . the representation ( [ dg1 ] ) for the channel resolvent is the basic expression for our wave - packet approach , since it gives explicit analytical formulas for the three - body resolvent and thus it allows to simplify drastically the solution of general three - body scattering problem . this expression can be used directly to solve the finite - dimensional analog of the faddeev equations for the three components of the total scattering wave function @xcite . alternatively the very convenient representation ( [ dg1 ] ) can also be used to solve some particular three - body scattering problems using the three - body lippmann schwinger equations @xcite . now let us proceed with solving the @xmath2 elastic and breakup scattering problems . the elastic scattering observables can be found from the faddeev equation ( fe ) for the transition operator @xmath128 , e.g. in the form @xcite : @xmath129 where @xmath63 is two - body @xmath63-matrix in three - body space and @xmath72 is the particle permutation operator . the equivalent form of fe for the transition operator @xmath79 has the form : @xmath130 where @xmath116 is the resolvent of the channel hamiltonian @xmath16 . since @xmath131 the operators @xmath128 and @xmath79 coincide on - shell and half - shell . since the determination of the off - shell channel resolvent in three - body space is a rather time consuming solution of the fe in the form ( [ pvg ] ) , it is very seldom employed for practical solutions . actually a similar form of the equations is associated with formalisms of the configuration space faddeev equations , where quite different numerical approaches have been employed @xcite than for the momentum space fe . however , since in the lattice approach one has explicit analytical formulas for the three - body channel resolvent @xmath116 the form ( [ pvg ] ) of fe turns out to be very appropriate for the numerical solution in a wp basis . the elastic @xmath2 scattering amplitude ( for a given value of total spin @xmath132 ) can be defined as matrix element of the solution of the eq . ( [ pvg ] ) taken in the initial state @xmath133 : @xmath134 the breakup amplitude for one faddeev component of the three - body wave function ( the so - called single - component amplitude ) can be found from the elastic transition operator @xmath79 after applying the operator @xmath135 from the left : @xmath136 to obtain the differential breakup cross sections , the total breakup amplitudes can be found by the contributions of all three single - component amplitudes . thus , we change the conventional treatment of the breakup process @xcite and consider the three - body asymptotic states as scattering states for the channel hamiltonian @xmath16 rather than as the states of the three - body continuum for the free hamiltonian @xmath68 . it is a quite natural when treating the elastic scattering amplitudes because the initial state wave functions correspond to the channel hamiltonian @xmath16 . in full analogy with this , one can treat the deuteron breakup as its excitation into a continuum @xmath1-state in the two - body subsystem governed by the @xmath18 subhamiltonian . as was already indicated above such a treatment of breakup processes is close to the configuration space approach . in fact , when solving the three - body faddeev equations in the configuration space @xcite one finds the breakup amplitude @xmath137 which determines the asymptotic behavior of the three - body wave function in hyperspherical coordinates @xmath138 and @xmath139 as follows @xmath140{}\frac{{\cal a}(\theta)}{(k\rho)^{5/2}},\quad k=\sqrt{p^2+{{\frac{3}{4}}}q^2},\ ] ] where @xmath141 and @xmath142 are two jacobi coordinates and @xmath143 is the hypermomentum . this breakup amplitude is defined for every spin - angular configuration and interrelated to the partial single - component breakup amplitudes ( [ abr ] ) by the following formula @xmath144 where @xmath145 is the hyperangle in momentum space . now if one transforms the formulas for the breakup amplitudes @xmath146 from @xcite to the integral form one receives the following definition for the breakup amplitudes in the momentum hyperspherical representation @xmath147 where @xmath148 is scattering function for the hamiltonian @xmath18 corresponding to the outgoing boundary condition . these functions are distinguished from the real - valued functions @xmath149 used in our approach by only a phase factor : @xmath150 where @xmath151 is the s - wave phase shift of the @xmath1 scattering in the channel with spin @xmath94 . using formulas ( [ acal_t1 ] ) and ( [ acal_t ] ) one can derive an alternative to formula ( [ abr ] ) for the _ single - component _ breakup amplitude via the scattering functions of the channel hamiltonian @xmath16 @xmath152 summarizing this derivation one can conclude that the breakup amplitudes can be defined quite similar to a matrix element for the elastic scattering transition operator @xmath79 with replacement of the the @xmath1 bound - state wavefunction with the exact scattering functions for the @xmath1 subhamiltonian . having now the required representations for both the elastic and breakup amplitudes , we will proceed in solving the faddeev equation in `` eigen '' wp subspace of the channel hamiltonian @xmath16 . in our wave - packet approach , all the operators in eq . ( [ pvg ] ) are projected onto a three - body wave - packet basis corresponding to the channel hamiltonian @xmath16 . in other word , every operator , e.g. @xmath79 , is replaced with its finite - dimensional wp representation : @xmath153 finally one gets the matrix analog for the eq . ( [ pvg ] ) ( for the given value of @xmath132 ) @xmath154 here @xmath155 and @xmath156 are the matrices of the pair interaction and the channel resolvent respectively , the matrix elements of which can be found in an explicit form . the matrix @xmath155 of the potential @xmath19 is diagonal in the indices @xmath157 of the wave - packet basis ( [ iq ] ) for the free subhamiltonian @xmath158 and has the block form : @xmath159^{ss'}_{kj , k'j'}={\delta}_{jj'}{\delta}_{ss ' } \langle z_k^s|v^s_1|z_{k'}^{s}\rangle.\ ] ] these matrix elements do not depend on in index @xmath160 and can be written with the usage of the rotation matrix @xmath161 defined in eq . ( [ exp_z ] ) as : @xmath162 in the last expression , the potential matrix elements in the free wp basis are used which have the form @xmath163 where @xmath164 is the momentum representation for the interaction potential . it implies that the matrix elements ( [ vks ] ) can be found analytically for a wide variety of the potential forms . an important ingredient of our lattice approach presented here is the representation of the permutation operator @xmath72 as an overlap matrix @xmath165 between the channel wp basis functions for different sets of the jacobi coordinates . using further the approximation ( [ exp_z ] ) for the scattering wave packets @xmath166 , these matrix elements can be expressed through the overlap matrix @xmath167 for the free lattice basis function of eq . ( [ perm ] ) with the help of the rotation matrices @xmath161 @xmath168 now let s replace the exact operator @xmath79 in the formula for the elastic @xmath2 scattering amplitude ( [ ael ] ) with its lattice counterpart @xmath169 , and employ further the projection rule for the free wp states ( [ proj_rule ] ) . then one gets that the on - shell elastic amplitude in the wave - packet representation can be calculated as a a diagonal ( on - shell ) matrix element of the @xmath170-matrix : @xmath171 where @xmath172 is the wp basis state corresponding to the initial state : index 0 denotes the bound state of the @xmath1 pair ( deuteron ) and index @xmath173 denotes the `` on - shell '' @xmath42-bin @xmath174 with the on - shell momentum @xmath175 : @xmath176 . it has been shown above that the breakup amplitude is proportional to a half - shell matrix element latexmath:[$\langle z_0,q_0,\sigma calculation of the elastic scattering amplitude , the finite - dimensional operator @xmath169 into the expression for the breakup amplitude and utilizing the projection rules for the free - motion and scattering wave - packets , one gets : @xmath178 where @xmath179 and @xmath180 $ ] are momenta corresponding to @xmath181 and @xmath43 bins respectively and the @xmath182 is the momentum width of the @xmath181 bin . thus , we just have found that the elastic and breakup amplitudes can be calculated directly using the diagonal ( `` on - shell '' ) and non - diagonal ( `` half - shell '' ) matrix elements of _ the same operator _ @xmath169 . however , some problem still arises here : how to define correctly which of the basis states @xmath183 correspond to the on - shell states of the @xmath16 . due to the discretization of the spectrum every wp basis state corresponds to the energy @xmath184 and thus one does not get the exact coincidence of these energies for different `` on - shell '' three - body wp states with the energy @xmath119 of the initial state . in other words , the energy conservation for three - body wp states is fulfilled only approximately within the corresponding bin widths . to avoid this difficulty we apply some energy averaging procedure to the transition matrix elements which is quite natural for the lattice - like representation . let us to rewrite expression ( [ t_sig ] ) for the breakup amplitude via wave functions of pair subsystems normalized to @xmath185-functions in energy . then , when calculating the breakup amplitudes one will need the transition operator matrix elements of the form ( we omit for the sake of brevity all the spin labels ) : @xmath186 where @xmath187 is the @xmath1 scattering state with energy @xmath115 , @xmath188 is the wave function of the subhamiltonian @xmath25 describing the free motion of the third nucleon relative to the @xmath1 subsystem , and @xmath119 is the total energy in the c.m . system . in the framework of the fully - discretized representation it is quite natural to make an energy averaging for the transition matrix elements @xmath189 over the excitation energy @xmath115 , which leads to the following integrals : @xmath190 here @xmath191\}$ ] is some set of intervals , in general independent of the initial partition of the excitation energy @xmath115 . @xmath192 are the corresponding widths . further , by replacing the exact @xmath79 operator with its wave packet counterpart ( [ umatr ] ) and using the projection rules for the scattering and free wp s , one can define a new approximation for the breakup amplitudes : @xmath193.\nonumber\end{aligned}\ ] ] here the sum runs over all possible indices @xmath194 and @xmath160 for which the difference in the square brackets is positive . then , for the single - component breakup amplitudes , one obtains the following approximate expression with the elements @xmath195 : @xmath196 . \end{array}\ ] ] now , using the energy averaged wp amplitudes one can get easily the following formula for the breakup amplitude in the hyperspherical representation : @xmath197 after the determination of the single - component breakup amplitudes , the total breakup amplitude is derived using contributions of all three faddeev components and can be written with the help of the particle permutation operator @xmath72 in the following way : @xmath198 the differential breakup cross section is related to this total breakup amplitudes as follows @xcite : @xmath199 here @xmath200 and @xmath201 are momenta of particles to be registered , @xmath202 is the arclength of the kinematical curve and @xmath203 is the phase space volume defined as : @xmath204 where @xmath205 is the center of mass momentum in the laboratory system . in the case of @xmath94-wave @xmath1 interactions , the total breakup amplitude is defined as sum of spin - quartet ( @xmath206 ) and spin - doublet ( @xmath207 ) single - component terms defined for all different jacobi sets . e. g. the @xmath202-wave amplitude of the two - neutron emission can be represented as a sum of three terms @xmath208 where @xmath209 are the total amplitudes for the quartet and doublet channels . in accordance with ref . @xcite , they are expressed through single - component breakup amplitudes defined for different sets of jacobi momenta @xmath210 as follows ( we assume here that neutrons are particles 1 and 2 and the proton is the particle 3 ) : @xmath211-\\ & { { \frac{3}{4}}}[t^{{{\frac{1}{2}}}1}(p_1,q_1)+t^{{{\frac{1}{2}}}1}(p_2,q_2)]+t^{{{\frac{1}{2}}}0}(p_3,q_3)\big\},\\ m^{{{\frac{1}{2}}}1}=&\frac12\big\{t^{{{\frac{1}{2}}}0}(p_1,q_1)-t^{{{\frac{1}{2}}}0}(p_2,q_2)+\\ & t^{{{\frac{1}{2}}}1}(p_1,q_1)-t^{{{\frac{1}{2}}}1}(p_2,q_2)\big\}.\\ \end{array } \label{mdif}\ ] ] finally the differential cross section of the @xmath212 breakup is expressed through the partial total amplitude ( [ abr_p ] ) as : @xmath213 to determine the differential cross section for the two - neutron emission one has to calculate the elements @xmath195 ( [ fpe ] ) for every spin component @xmath214 of the total breakup amplitude and then substitute them into the explicit formulas ( [ mdif ] ) and ( [ abr_p ] ) . thus , it has been demonstrated above that in our wp approach one can find quite naturally all breakup amplitudes together with the elastic scattering amplitude . this gives a very nice _ universal _ and unified calculation scheme . some details of the numerical procedure for solving the matrix equation ( [ m_pvg ] ) are discussed in appendix c. because we have developed here a novel approach for the treatment of the three - body breakup processes we would need precise and reliable tests to check our new procedure . in all the tests below we employ as a convenient universal wp basis the free momentum wave packets for jacobi momenta @xmath42 and @xmath40 constructed using the generalized tchebyshev grid : @xmath215^t , i=1,\ldots , n,\ ] ] where @xmath216 is the common scale parameter and the @xmath63-parameter determines the `` sparseness degree '' of the bin set . a similar grid with the size @xmath217 and common scale @xmath218 is introduced for discretization of the momenta @xmath40 . as a first extremely convenient test we have chosen the three - body model with a separable @xmath1-potential . this model , which can be treated numerically very accurately in various kinematical breakup situations seems to be very appropriate for such a test ( see below ) . we consider here the @xmath2 breakup with pairwise separable @xmath1-interactions in the form : @xmath219 as it is well known , the faddeev equations for such potentials can be reduced to one - dimensional integral equations in momentum space . such equations , as demonstrated in our previous work @xcite , can be solved quite accurately using the two - body free wp basis . for convenience of the reader we describe all the details of this procedure in appendix b. here we will refer to the results found in such approach as to the `` exact ones '' . we will compare the latter with the results of the solution of the general three - body wp scheme with the eq . ( [ m_pvg ] ) for the separable potential . the fig . [ fig02 ] shows the unaveraged ( i.e. having a histogram form ) and energy averaged breakup amplitudes for the yamaguchi potential obtained from the general matrix fe ( [ m_pvg ] ) in the lattice basis . the energy averaged amplitudes have been calculated using the averaging procedure described above . it is clearly seen from fig . [ fig02 ] the method developed makes it possible to find rather smooth energy dependence for the breakup amplitudes . using the averaged amplitudes , we have found the single - component breakup amplitudes @xmath220 as functions of hyperangle @xmath145 for both quartet and doublet three - body spin channels . now , we can compare the approximated wp breakup amplitudes ( [ fpe ] ) found within our general formalism for a separable model with the exact amplitudes derived from directly solving the one - dimensional faddeev equations . in the figs . [ fig1]-[fig3 ] such a comparison between approximated and `` exact '' results is presented . the two - body free wp bases with size @xmath221 and @xmath222 have been used in the calculation of the approximated amplitudes . we observe in the figs . [ fig1]-[fig3 ] quite a very good agreement between the lattice - approximations and the `` exact '' amplitudes . the corresponding curves are almost indistinguishable in the figures . the only differences are seen at the hyperangle region @xmath223 90@xmath224 for the spin - doublet @xmath225 amplitude . it is clear also that the wp amplitudes calculated in a finite - dimensional @xmath8 basis converge to the exact ones with increasing basis size . having tested our novel approach using the simple separable model for the @xmath1 force one can move to a more realistic case of a local @xmath1-interaction . for this , we have chosen the so - called mt i - iii @xmath1 central potential which was frequently used in the past for the test of few - body calculations . so , we can compare our results for this model with very accurate benchmark calculations @xcite . for the present wp calculations we use again the three - body lattice basis constructed on a tchebyshev two - dimensional grid . the results of such a comparison are presented in fig . [ fig4]-[fig6 ] for our single - component hyperspherical amplitudes @xmath226 . one can observe in the figs . [ fig4]-[fig6 ] a quite satisfactory general agreement with the results of the benchmark calculations @xcite except in the region @xmath223 90@xmath224 , similarly to the case of the separable @xmath1 interaction . it should be mentioned , that the from the amplitude @xmath227 eq . ( [ ca_mt ] ) has an additional factor inversely proportional to the relative momentum @xmath40 compared to the amplitude @xmath228 discussed in the previous subsection . so , the differences to the exact solution of @xmath227 are more visible at the region corresponding to small values of @xmath40 . similar difficulties at @xmath223 90@xmath224 are also observed in other works @xcite in which the breakup calculations have been done in the configuration space . in ref . @xcite it has been demonstrated that for the correct calculation of the breakup amplitudes in the area of small relative @xmath1 momenta one has to employ the explicit integral form for the breakup amplitude @xmath229 . in our approach the minor disagreement of the wp breakup amplitudes and the exact benchmark results at very low relative momenta can be related to some uncertainties in the determination of the widths and the corresponding values of momenta for the wp scattering states of the pair continuum very close to the threshold . so the case of very low relative momenta in the lattice approach deserves a separate study which is under way . using further the single - component amplitudes we have found the total ( i.e. with inclusion of all three faddeev components ) breakup amplitudes as well as differential cross sections of two - neutron emission for different kinematical configurations . in figs . [ fig7 ] and [ fig8 ] the differential cross sections for two - neutron emission with our wp technique are presented for two configurations and compared to the results of ref . one configuration includes the fsi peak while the second one is related to the so called `` space - star '' breakup kinematics . for derivation of the faddeev cross sections we used an interpolation of the data in the table presented in ref . @xcite for the single - component amplitudes . in summary , the agreement between conventional and lattice results is generally very good . the curves are mostly hardly distinguishable in figs . [ fig7],[fig8 ] . in the present work we generalized the wave - packet method developed by the present authors earlier for the discretization of the three - body continuum and used it for finding the three - body breakup amplitudes . as far as the present authors are aware , the study in the work is the first where the faddeev breakup amplitudes are obtained completely in the three - body @xmath8-basis . thus , it would be appropriate to enumerate some important distinctive features of our lattice - like approach . \1 . due to projection of the scattering integral equations onto the wave - packet @xmath8 basis corresponding to the three - body channel hamiltonian @xmath16 , we get an explicit analytical representation for the three - body channel resolvent @xmath116 that is used in all further calculations . for this we employ the version of the integral faddeev equations with the kernel @xmath230 instead of the conventional form @xmath231 . this simplifies drastically the whole calculations scheme as compared to the conventional one , because first we do not need to know the full off - shell pair @xmath63-matrix at many different energies and in addition we get the matrix kernel in a very convenient finite - dimensional form . as an input information for the @xmath1-interaction , we use only the results of _ a single diagonalization _ ( for every spin channel ) of the @xmath1 hamiltonian matrix . from such a diagonalization we get immediately the whole set of pseudo - states ( the scattering wp s ) and partial phase shifts at many energies corresponding to these pseudo - states @xcite . \2 . for the matrix of the transition operator , one gets an _ universal _ linear matrix equation with finite matrix elements . the diagonal ( on - shell ) elements of this solution determines the elastic scattering amplitudes while the non - diagonal ( half - shell ) elements determines the single - component breakup amplitudes up to some known phase factor . the structure of the kernel for the matrix equation obtained is very convenient for numerical realization . due to the fact that the kernel is a product of a diagonal matrix , two block matrices and a very sparse matrix , it is possible to greatly reduce requirements for the ram storage size and noticeably decrease the computation time . the effect of the particle permutation operator in the faddeev kernel is represented now with the help of the universal matrix of basis functions overlapping for different jacobi coordinate sets . it allows to avoid very time - consuming numerous re - interpolations of the current solutions ( at iterations ) from one set of jacobi coordinates to another one at every iteration step @xcite . \5 . due to an averaging of the integral kernels over the cells in momentum space the very complicated energy singularities of the kernel above the breakup threshold ( e.g. the moving branching points etc . ) are smoothed and one can solve the few - body scattering equations directly at real energies , i.e. without any contour deformation to the complex - energy plane . this fact also facilitates enormously the practical solution of few - body equations above the breakup threshold . the comparison of the results obtained in our approach with those for the model for the separable @xmath1 potential and with benchmark breakup calculations ( with a semi - realistic local @xmath1-potential ) has demonstrated that the wp method allows to get quite accurate three - body breakup amplitudes and cross sections . still , the region of very low relative @xmath1 momenta requires some additional study . some inaccuracy of our results in this area can be related to two factors : ( i ) a slow convergence of the wp amplitudes in this region , and ( ii ) some uncertainty of the wp representation of the two - body continuum in the region of very low relative momenta . we plan to devote a special study for the solution of this problem . in summary one can conclude that the total lattice - like @xmath8 discretization of the three - body continuum allows to find an accurate solution for the three - body faddeev equations for breakup amplitudes and simplifies enormously the calculation together with a noticeable reduction of the computational cost . * acknowledgments * the authors appreciate a partial financial support from the rfbr grants 10 - 02 - 00096 , 10 - 02 - 00603 , and 12 - 02 - 00908 . in our approach we employ a lattice basis , i.e. a basis built by free wp s in momentum space . the two - dimensional ( three - body ) wave packet in momentum space are step - like functions of variables @xmath40 and @xmath42 : @xmath232 which are nonzero only at the intervals @xmath233 $ ] and @xmath234 $ ] ( @xmath235 and @xmath236 are the widths of corresponding intervals ) . such wave packets are normalized to unity with the weight @xmath237 and form an orthonormal basis ( it is assumed that the intervals are not overlapping ) . the matrix element of the permutation operator @xmath72 between plane waves has a simple form for the @xmath94-wave : @xmath238 where @xmath239 is a spin - channel coupling matrix . the @xmath185-function guaranties energy conservation and @xmath141 is the cosine of the angle between vectors @xmath26 and @xmath240 which ( with taking into account the @xmath185-function ) can be expressed as a function of three momenta , e.g. @xmath40 , @xmath42 , @xmath241 : @xmath242 the condition @xmath243 in ( [ p0 ] ) restricts the allowed values of momenta to a region , where the overlap is nonzero . to find the matrix elements of the permutation operator @xmath72 over the free wps ( [ fwp ] ) , one has to integrate the function @xmath244 over rectangular cells @xmath70 , @xmath245 ( where the upper prime at the interval symbol denote that it refers to a different set of jacobi coordinates ) : @xmath246 actually , the integral in eq . ( [ p_pack ] ) is reduced to an area of two overlapping rectangular areas @xmath247 and @xmath76 . hyperspherical ( polar in the s - wave case ) coordinates are most convenient to calculate such overlaps . let us introduce the reduced ( rescaled ) momentum variable @xmath248 : @xmath249 then , the energy conservation takes the `` homogeneous '' form @xmath250 . the hyperspherical coordinates @xmath251 are introduced as usually : @xmath252 in these hyperspherical coordinates the integral in eq . ( [ p_pack ] ) takes the following form : @xmath253 where we define the overlapping square : @xmath254 thus we get that the permutation matrix element is directly interrelated to this square : @xmath255 the condition @xmath243 can be expressed through the hyperangular variables @xmath256 as follows : @xmath257 determined by the condition @xmath243 is a rectangle in the plane @xmath258 restricted by four straight lines ( see fig . [ fig9 ] ) : therefore , the integral in eq . ( [ p00 ] ) can be evaluated as the external ( numerical ) integral over @xmath259 in the range between @xmath260 and @xmath261 from the area of intersection of the rectangle @xmath202 and the rectangle @xmath262 whose vertices depend on @xmath263 ( see fig . [ fig9 ] ) : @xmath264 the integration limits over @xmath259 are equal : @xmath265 if @xmath266 then the cells do not overlap and the matrix element is equal to 0 . the coordinates of vertices of the rectangle @xmath267 are computed directly ( see the fig . [ fig2n ] for further explanations ) : @xmath268 here , if @xmath269 then @xmath270 should be replaced by 0 , and if @xmath271 then @xmath272 should be replaced by @xmath273 ; the same rule should be applied also to primed values . the area of intersection @xmath274 in the plane @xmath256 is evaluated analytically by the formulas of elementary geometry . we consider here a system of three nucleons with equal masses @xmath11 interacting by a separable @xmath1 force in s - wave spin - singlet ( @xmath275 ) and spin - triplet ( @xmath102 ) states correspondingly : @xmath276 the two - body @xmath63-matrices are also separable : @xmath277 where @xmath278 are known functions @xmath279 and @xmath280 is the free two - particle resolvent . function @xmath281 for triplet state has the pole at the deuteron binding energy @xmath96 . the corresponding bound state wavefunction @xmath95 is defined as follows : @xmath282 where @xmath283 is the residue of @xmath284 at the pole . we use here the two - parameter yamaguchi potentials with the form factors : @xmath285 in this case @xmath286 and @xmath283 take the form : @xmath287 the potential parameters @xmath288 and @xmath289 are taken from ref . @xcite . the faddeev equation for the elastic transition operator @xmath79 in this case reduces to the system of one - dimensional integral equations of the lippmann schwinger type @xcite for the elastic scattering amplitudes corresponding to the total spin @xmath290 and orbital momentum @xmath291 ( in the case of the @xmath94-wave pair interactions the @xmath132 and @xmath291 are conserved separately ) : @xmath292 where @xmath293 the kernels @xmath294 in eq . ( [ vecfe ] ) are defined as follows : @xmath295 where @xmath296 , @xmath297 are the legendre polynomials , and @xmath239 is a spin - channel coupling matrix . in case of quartet scattering one has the single equation with @xmath102 and @xmath298 , while there are two coupled equations with @xmath99 in the case of doublet scattering and @xmath299 , @xmath300 after solving eq . ( [ vecfe ] ) , the partial wave elastic on - shell amplitude can be defined from the relation @xmath301 where @xmath302 . now we turn to the determination of the breakup amplitude . substituting the explicit formulas for the @xmath63-matrix in ( [ sepp ] ) and the deuteron wave function ( [ psisep ] ) into ( [ abr ] ) , one can express the partial `` faddeev '' breakup amplitudes via the elastic amplitudes @xmath303 : @xmath304 let s now proceed with the lattice version for elastic amplitude . we introduce the free wp basis ( [ iq ] ) and project eq . ( [ vecfe ] ) to this basis . finally , we find the matrix equation : @xmath305 where the letters with double lines denote matrices of corresponding operators in the wp subspace for the given values of total spin @xmath132 and orbital momentum @xmath291 ( below we shall omit @xmath132 and @xmath291 for brevity ) . more definitely , @xmath306 the elastic on - shell amplitude in wp representation is defined from the diagonal `` on - shell '' matrix elements of the lattice transition matrix @xmath307 : @xmath308 similarly , the packet approximation for the breakup amplitude is determined by off - diagonal matrix elements of @xmath309 : @xmath310 where @xmath311 and @xmath312 is the midpoint of bin @xmath43 . the wp representation for the breakup amplitude @xmath137 which determines the asymptotics of the breakup wave function in hyperspherical coordinates has the form : @xmath313 here we will discuss some details of the numerical procedure for solving the matrix equation ( [ m_pvg ] ) . the main difficulty is its large dimension . quite satisfactory results can be obtained with a basis size @xmath314 . it is means that in the simplest one - channel ( quartet ) case one gets a kernel matrix with dimension @xmath315 which occupies @xmath316 gb ( at single precision ) of ram or external memory of the computer . in the two - channel ( doublet ) case the required amount of memory increases by a factor 4 . where @xmath319 the matrix of the channel resolvent @xmath320 is diagonal and its elements are defined by simple explicit formulas . the matrix of the potential @xmath321 has a block - type structure ( [ bv1 ] ) : in fact , it is the direct product of the @xmath322-matrix of the two - body interaction and the unit @xmath323-matrix . the rotation matrix @xmath161 has a similar form and the actual dimension @xmath322 . the free permutation matrix @xmath324 is very sparse due to the energy conservation condition . as a rule , only about @xmath325 of its elements are distinguished from zero , and the sparsity increases when the basis dimension increases . so , if to summarize all these details one finds that the kernel in the matrix equation ( [ m_pvg ] ) is the product of four matrices : a diagonal one , a very sparse one and two block matrices with actual dimension @xmath322 . if , instead of storing the entire matrix @xmath317 , we store its factors only ( for the sparse matrix @xmath324 we store the nonzero elements only ) , then we can save a huge portion of physical memory : for the above example we shall need only @xmath326 mb instead of initial 6.4 gb . such an enormous reduction of the required memory allows us to perform calculations without using an external memory , which , in its turn , reduces the calculation time by approximately one order . this possibility to avoid storing a very large amount of data is related to the specific procedure used by us for the solution of the equation ( [ m_pvg ] ) . as a matter of fact , to find the elastic and breakup amplitudes one needs not all but only on - shell matrix elements of the transition operator . so , each of these elements can be found without complete solving the matrix equation ( [ m_pvg ] ) but by means of a simple iteration procedure with subsequent summing the iterations via the pade - approximant technique . if we do not store the entire matrix of kernel @xmath317 , to perform each iteration we need only three additional matrix - vector multiplications with the matrices of a special form . abc w. glckle , h. witaa , d. hber , h. kamada , j. golack , phys . rep . * 274 * , 107 ( 1996 ) . h. witaa , w. glckle , j. golak , a. nogga , h. kamada , r. skibinski , j. kuros - zolnierczuk phys . c * 63 * , 024007 ( 2001 ) . elster , t. lin , w. gloeckle , s. jeschonnek , phys . c * 78 * , 034002 ( 2008 ) . a. deltuva , a.c . fonseca , phys . c * 76 * , 021001(r ) ( 2007 ) . r. lazauskas , j. carbonell , phys . c * 70 * , 044002 ( 2004 ) . a. deltuva , a.c . fonseca , p.u . sauer , phys . c * 71 * , 054005 ( 2005 ) . a. kievsky _ et al . _ , j. phys . g * 35 * , 063101 ( 2008 ) . h. witaa , w. gloeckle , phys . c * 83 * , 034003 ( 2011 ) . x. c. ruan _ et al . _ , phys . rev . c * 75 * , 057001 ( 2007 ) . space star r. lazauskas , j. carbonell , phys . c * 84 * , 034002 ( 2011 ) . s. quaglioni , w. leidemann , g. orlandini , n. barnea , v.d . efros , phys . c * 69*,044002 ( 2004 ) . w. glckle , g. rawitscher , nucl . a * 790 * , 282c ( 2007 ) . r. kozack , f.s . levin , phys . c * 36 * , 883 ( 1987 ) . kurouglu , phys . a * 44 * , 7307 ( 1991 ) . i. bray , d.a . konovalov , i.e. mccarthy , phys . rev . a * 43 * , 1301 ( 1991 ) . bang , a.i . mazur , a.m. shirokov , yu.f . smirnov , s.a . zaytsev , ann . phys . * 208 * , 299 ( 2000 ) . z. papp , c-.y . hlousek , b. knya , s.l . yakovlev , phys . a * 63 * , 062721 ( 2001 ) ; p. doleschall , z. papp , phys . c * 72 * , 044003 ( 2005 ) . piyadasa et al . , phys . c * 60 * , 044611 ( 1999 ) . rasoanaivo , g.h . rawitscher , phys . c * 39 * , 1709 ( 1989 ) . j. a. tostevin , f.m . nunes , i.j . thompson , phys . c * 63 * , 024617 ( 2001 ) . thompson , in _ scattering _ , edited by e.r . pike and p.c . sabatier ( academic , new - york , 2001 ) , p.1360 . rubtsova , v.i . kukulin , a.m.m . moro , phys . c * 78 * , 034603 ( 2008 ) . rubtsova , v.i . kukulin , v.n . pomerantsev , phys . c * 79 * , 064602 ( 2009 ) . pomerantsev , v.i . kukulin , o.a . rubtsova , phys . c * 79 * , 034001 ( 2009 ) . kukulin , v.n . pomerantsev and o.a . rubtsova , theor . math 150 * , 403 ( 2007 ) . schmid and h. zeigelmann , _ the quanum mechanical three - body problem _ , braunschweig ( 1974 ) . rubtsova , v.i . kukulin , v.n . pomerantsev , a. faessler , phys . c ( 2010 ) . w. glckle , g.l . payne , phys . c * 45 * , 974 ( 1992 ) . v.m . suslov and b. vlahovic , phys . c * 69 * , 044003 ( 2004 ) . payne , w. glckle , j.l . friar , phys . c * 61 * , 024005 ( 2000 ) . cahill , i.h . sloan , nucl . * a165 * , 161 ( ( 1971 ) . j.l . friar _ et al . _ , c * 51 * , 2356 ( 1995 ) .

we have performed measurements of scintillation light in liquid neon , observing a signal yield in our detector as high as ( 3.5 0.4 ) photoelectrons / kev . we measure pulse shape discrimination efficiency between electronic and nuclear recoils in liquid neon from 50 and 300 kev nuclear recoil energy . we also measure the parameter in liquid neon between 30 and 370 kev nuclear recoil energy , observing an average above 50 kev . we observe a dependence of the scintillation time distribution and signal yield on the pressure and temperature of the liquid neon .

1111.3260

the deap / clean collaboration has proposed to use liquid argon and liquid neon as scintillating targets in large scale detectors to observe both @xmath3-solar neutrinos and dark matter in the form of weakly interacting massive particles ( wimps ) @xcite . in particular , the miniclean experiment is designed to do a dark matter search with first liquid argon and then liquid neon as both a systematic check on backgrounds and as a precursor to a multi - tonne scale detector @xcite . because of the expected @xmath4-dependence of the cross section for wimp - nucleon scattering @xcite , filling the same detector alternately with argon and neon will change the expected dark matter signal in a known way while maintaining an identical level of external backgrounds . here , we extend previous studies of the scintillation properties of liquid neon @xcite . like the other liquefied noble gases , liquid neon is relatively inexpensive , is easily purified of radioactive contaminants , scintillates brightly when exposed to ionizing radiation and is dense enough to self - shield , reducing the background level in the center of a larger volume of liquid . the key for dark matter detection is to be able to suppress the electronic recoils that make up most of the backgrounds from the nuclear recoils that would make up a wimp signal by use of some combination of self - shielding and discrimination . because @xmath5 scattering also produces electronic recoils , discrimination does not improve the performance of a neutrino detector . however , neon has no long - lived radioactive isotopes and is more readily purified of such contaminants as @xmath6ar and @xmath7kr @xcite than the heavier noble gases , rendering it an ideal target for low - energy neutrinos . when ionizing radiation interacts in liquid neon or any other liquefied noble gas , ultraviolet light is produced via scintillation . the incoming radiation collides with an electron or nucleus in the liquid and deposits energy . the resulting electronic or nuclear recoil then excites or ionizes other neon atoms . the excited atoms quickly combine with surrounding ground state atoms to form dimer states , ne@xmath8 . the ions also bond with ground state atoms to form ionized molecules , ne@xmath9 , which can in turn recombine with free electrons to also form the excited dimer states . finally , the dimer states decay emitting scintillation photons with wavelength 80 nm . because these photons have insufficient energy to excite neon atoms @xcite , the liquid does not absorb its own scintillation light and large light yields can be attained from liquid noble gas detectors . the metastable molecules are produced in both singlet and triplet states . the singlet state decays in nanoseconds , while the triplet state undergoes a forbidden spin flip before decaying , extending the triplet lifetime to 15 @xmath10s in neon @xcite . because nuclear and electronic recoils produce different ratios of singlets to triplets , pulse timing can be used to identify the source of the initial excitation via pulse shape discrimination ( psd ) @xcite . as the psd depends directly on the scintillation timing , we have attempted to re - measure the time constants of liquid neon scintillation in this work . we discovered two additional time components in the light signals produced by ionizing radiation , and we also found that both the intensity and lifetime of each component changes significantly with the temperature and pressure of the liquid , a new feature that is not observed in the heavier noble gases . a second property of scintillation in liquefied noble gases is that nuclear recoils produce less light than electronic recoils of the same energy . the ratio of signal yields at zero electric field for the two event classes is known as the nuclear recoil scintillation efficiency , or @xmath1 . the value of @xmath1 in combination with the discrimination efficiency sets the nuclear recoil energy analysis threshold of a liquefied noble gas dark matter detector . because in general @xmath11 , two energy scales may be employed , denoted `` kevee '' and `` kevr . '' the actual energy deposited by a nuclear recoil is expressed in units of kev recoil or kevr . most calibration sources used in liquid noble gas detectors are @xmath12 or @xmath13 sources that produce electronic recoils in the liquid . therefore , the energy calibration gleaned from these sources refers to the energy deposited by an electronic recoil , or kev electron equivalent ( kevee ) . in practice , we often have no direct energy calibration for nuclear recoils and so use electron equivalent energies . the conversion factor is @xmath1 . at zero electric field @xmath14 = e [ \mathrm{kevr } ] \times \mathcal{l}_\mathrm{eff}.\ ] ] psd has been studied in argon @xcite and xenon @xcite . our group performed a measurement of psd in neon as well , but nuclear recoils were observed at only one energy @xcite . similarly , @xmath1 has been measured in argon @xcite and xenon @xcite , but at only one energy in neon . in this paper , we report measurements of scintillation light in liquid neon over a wide energy range . we measure psd from 50 to 300 kevr and @xmath1 from 30 to 370 kevr . we also present results on the time dependence of liquid neon scintillation light as a function of temperature . the apparatus used in these measurements is named microclean and has been described in detail elsewhere @xcite . a schematic is shown in fig . [ fig : cell ] . the detector has an active volume of 3.14 liters viewed by two 200-mm - diameter hamamatsu r5912 - 02mod photomultiplier tubes ( pmts ) , which are specifically designed for use in cryogenic liquids . the active volume is defined by a polytetrafluoroethylene ( ptfe ) cylinder 200 mm in diameter and 100 mm in height , with two 3-mm - thick fused - silica windows at top and bottom . all inner surfaces of the ptfe and windows are coated with @xmath15 mg/@xmath16 of tetraphenyl butadiene ( tpb ) @xcite , which shifts the wavelength of the ultraviolet light to approximately 440 nm . the tpb thickness on the ptfe cylinder was reduced by @xmath17 relative to that used in @xcite to mitigate tpb absorption of the wavelength - shifted light . the microclean vessel is contained within a stainless steel vacuum dewar for thermal insulation purposes . the gas handling system is shown schematically in fig . [ fig : gashandle ] . ultra - high - purity neon flows through a nupure omni 1000 @xcite getter and an external charcoal trap at 77 k ( not shown ) before passing into a vacuum cryostat through a tube at the top . previous studies have shown that activated charcoal provides a very effective purifier of many impurities in neon , including radon @xcite . once inside the vacuum cryostat , the gas flows into a copper liquefier cell attached to the second stage of a cryomech model pt805 cryorefrigerator before dripping into the microclean vessel . there is a circulation loop inside the vacuum dewar for continuous purification of the neon , again using activated charcoal as the purifier . liquid neon flows out of the bottom of the microclean vessel into a nearby plumbing volume containing a heater . the heater ( labeled `` circ . pump '' in fig . [ fig : gashandle ] ) acts as a pump by boiling the liquid ; the resulting pressure differential between the heater and the liquefier drives gas up a tube through a charcoal trap mounted inside the cryostat before reentering the top of the liquefier , where gravity pulls the liquid back into the main chamber . the `` @xmath18kr@xmath19 generator '' shown in fig . [ fig : gashandle ] refers to the setup for introducing @xmath18kr@xmath19 into the liquid neon for calibration purposes as described in sec . [ sec : calibrations ] and @xcite . the @xmath18kr@xmath19 generator is connected to the gas inlet line just outside the vacuum dewar . the incoming gas can be diverted through the trap on its way into the detector to introduce @xmath18kr@xmath19 atoms into the neon . the data acquisition ( daq ) system consists of a 250 mhz , 12-bit caen v1720 waveform digitizer ( wfd ) . the two pmt channels are recorded with a record length of 64 @xmath10s to collect as much of the long - lived triplet light as possible , and a sample trace is shown in fig . [ fig : sampletrace ] . the pmt signals are passed through a lecroy model 612am dual - output amplifier , with one copy passed to the wfd and the other copy sent to triggering electronics . the trigger is set in several different configurations . in most cases , a trigger is generated when each pmt signal crosses a threshold of 1/2 times the mean height of a single photoelectron within a 100 ns coincidence window . for some low energy runs when an external detector is used as a coincidence tag , the trigger is generated when the sum of the two pmt signals crosses a threshold of roughly 1/2 the size of a photoelectron in coincidence with a signal in the external tagging detector . in each 64 @xmath10s data record , approximately 8 @xmath10s of presamples are collected to measure the baseline of each pulse . the data collected by the daq software are saved in a root - based file structure for analysis @xcite . the data processing is similar to that described in @xcite . the presamples are averaged to obtain a baseline and a baseline root - mean - square , and the baseline is subtracted from each trace . we integrate each trace from 50 ns before the trigger ( defined as the time at which the voltage first rises above @xmath20 of its maximum value ) to 44 @xmath10s after the trigger to determine the total area of each pulse . in performing this integral , we restrict the range of integration to 100 ns windows around locations where the trace voltage crosses a threshold of three times the baseline root - mean - square . this integration method is a hybrid of single photoelectron counting and pulse integration . we reject data for which the trigger times of the two pmts differ by greater than 20 ns . we define an asymmetry parameter , @xmath21 where @xmath22 and @xmath23 are the signal areas in the top and bottom pmts . we reject data for which the asymmetry parameter is farther than @xmath24 from the mean value as determined by fitting a gaussian function to the asymmetry data . as the triplet component of liquid neon scintillation is spread out over tens of microseconds , there are many single photoelectron pulses well separated in time in the tails of events . to obtain a single photoelectron spectrum for each pmt , the last 30 @xmath10s of each trace is evenly divided into 150 ns windows . if the voltage spread in the first and last 30 ns of each 150 ns window is less than three times the size of the baseline root - mean - square , the peak height and pulse area of the window are collected in a 2-d histogram ; this restriction prevents inclusion of partial single photoelectrons in the data set . a gaussian is fit to the 1-d histogram of the pulse areas to determine the mean single photoelectron size . in previous tests of the r5912 - 02mod , it was found that the gain dropped substantially at 30 k @xcite . for the data presented here , we observe a sharp drop in pmt gain upon cooling to liquid neon temperatures , followed by a slow decrease in gain over several weeks before finally reaching a plateau . the low resulting gain necessitates the application of higher voltages to the pmts and amplification from the lecroy 610am amplifier ; for most of the data presented here , the pmts are biased at 1750 v , the absolute gain of the pmts is @xmath25 , and the pmt signals are further amplified by a factor of 10 with the lecroy amplifier . a single photoelectron peak is resolved in the bottom pmt but not in the top pmt . to improve pmt performance , we use two high voltage lines to bias the pmts , with a separate high voltage running directly to the first dynode . by tuning the first dynode voltage relative to the total voltage , we can achieve some improvement in the single photoelectron spectra but not enough to recover a resolvable single photoelectron peak in the top pmt . we use several radioactive sources for calibration purposes : the @xmath18kr@xmath19 generator described in @xcite , a 10-@xmath10ci @xmath26na source for 511-kev and 1275-kev gamma rays , and a d - d neutron generator producing a monoenergetic beam of 2.8-mev neutrons @xcite . the @xmath18kr@xmath19 generator produces metastable @xmath18kr@xmath19 nuclei that decay via two electromagnetic transitions with energy 32.1 kev and 9.4 kev . the second transitions has a half life of 154 ns , and given the long time scale of liquid neon scintillation , both transitions are observed together as a single waveform of total energy 41.5 kev . the @xmath26na source produces positrons that immediately annihilate into two 511-kev gamma rays with opposite momenta ; we tag the second gamma ray using a nai crystal placed back - to - back with our apparatus to improve the data quality from the @xmath26na source and reduce backgrounds . we reject data where the trigger times in the nai crystal and the liquid neon differ by more than 30 ns , and we also make a cut to select events within 2@xmath27 of the 511-kev peak of the nai crystal . we also use a tagging detector when investigating nuclear recoils with the neutron generator . we require a pmt viewing bc501a organic scintillator to record an event within 200 ns of an event in the liquid neon . the organic scintillator is contained in a cylinder with a diameter of 127 mm and a depth of 127 mm . we also make a psd cut in the organic scintillator and a cut on the time - of - flight ( tof ) between the event in the liquid neon and the tagging detector . these cuts will be discussed in more detail in sec . [ sec : leff ] . both the neutron generator and the organic scintillator are located approximately 1.63 m from microclean , and the experimental setup can be seen schematically in fig . [ fig : nuc_scatt ] . by changing the scattering angle @xmath28 , we can choose the energy of the nuclear recoils that scatter just once in the liquid , @xmath29 , using simple kinematics : @xmath30,\end{aligned}\ ] ] where @xmath31 is the incident neutron energy and @xmath32 is the atomic mass of the target . we measure the angle by running a string marked at both ends and at its midpoint from the generator around the dewar and back to the generator . we mark the location of the midpoint of the string on the dewar and then repeat the process for the organic scintillator . the ratio of the arc length between the two points on the dewar to the circumference ( 175.6 cm ) provides the angle , @xmath28 . the accuracy is limited by our knowledge of the midpoint of the string , and by repeated measurements we estimate the uncertainty on the arc length to be 6.4 mm , for an uncertainty on the angle of 1.3 degrees . we obtain an energy calibration by comparing @xmath26na data with a simulation of the detector performed with the reactor analysis tool ( rat ) @xcite . combining geant4 and root @xcite in a single simulation package , the rat simulation of microclean contains a complete optical model of the inner detector and pmts and has been used previously in studies of liquid argon scintillation @xcite . in general , there are two features in the @xmath26na data that can be used for calibration purposes : a 511-kev photoabsorption `` hump '' and the compton edge . the analysis begins by first obtaining a rough estimate for the signal yield from the 511 kev hump , without using the simulation . we tune the parameters of the simulation to achieve the estimated light yield before simulating the @xmath26na source including a flat background contribution to account for the events observed above 600 kev from accidental coincidence of background events in the argon and 511 kev gamma rays in the nai detector . we then re - fit the data to the simulated spectrum , and the scale factor between the integrated charge from both pmts to the energy scale of the simulation provides the energy calibration ; an example of this fit is shown in fig . [ fig : lysim ] . while the energy resolution is not an explicit parameter of the fit , we do smear the photon counts from the simulated spectrum with a gaussian kernel having a standard deviation of @xmath33 , as this smearing provides better agreement between simulation and data . this additional smearing suggests a source of noise in the real experiment that is not being accurately modeled in the simulation , and a likely source of this noise is the poor gain dispersion characteristics of the pmts . it could also be a property of the liquid neon . the simulation does not accurately model the data below about 200 kev , predicting about @xmath34 fewer counts than observed down to the low energies where thresholding effects begin ( about 40 kev for the run shown in fig . [ fig : lysim ] ) . this data excess is seen for a variety of different gain and trigger threshold settings , and it is currently unexplained , although asymmetry effects between the pmts smearing out very low energy events is one possibility , as is an additional source of background in the detector . given the discrepancy between data and simulation and the fact that that the trigger and gain settings were varied throughout the course of the run , the energy calibration fit is performed between 200 and 550 kev . there is some tension between the best fit at the compton edge and the 511 kev hump , and the average @xmath35ndf is 2.5 for all runs with the nominal fit range . if the fit is restricted to the range between 350 kev and 550 kev , the agreement is much better , with an average @xmath35ndf of 1.4 . we estimate the systematic uncertainty caused by this tension to be @xmath36 by fitting each feature independently and taking the difference in the resulting energy scale relative to the nominal fit as the uncertainty . performing the fit without the flat background contribution in the simulation changes the calibration by less than @xmath37 . an additional source of uncertainty in this calibration are the statistics of fitting the simulated and observed spectra , but this source of uncertainty is estimated to be @xmath37 . we cross check the energy calibration obtained from the @xmath26na source with both higher and lower energy references . for our lower energy reference , we use the @xmath18kr@xmath19 source described in the next section which is the primary reason we believe the energy calibration is accurate to within our uncertainties . for the high energy reference , we look at the compton edges of the 1.4 mev and 2.6 mev gamma rays produced in the decay of @xmath38k and @xmath39tl , as there are a number of these events in background data . figure [ fig : bgcomp ] shows background data with the energy scale derived from the @xmath26na calibration compared to a simulation of 1.4 mev and 2.6 mev gamma rays using the rat simulation . as the purpose of this simulation is solely to cross - check the accuracy of the energy scale derived from the @xmath26na at higher energies , we do not attempt to simulate any other sources of background ( i.e. lower energy gamma rays in the detector materials , cosmic rays , or other backgrounds ) ; therefore , we do not expect the simulation to agree with the data below 1 mev . we find evidence for a slight non - linearity in either the energy response of the liquid neon or the pmts at these energies , as the background data do not agree with the simulation if we assume a linear response of the detector up to 2.6 mev . the fit is improved if the observed energy ( as calibrated by the @xmath26na source ) is actually representing the following function of the deposited energy in the simulation : @xmath40 it is likely that this dependence is caused by a non - linearity of the top pmt , as a comparison using just the signal in the bottom pmt requires no energy scaling to provide agreement . we also find that the data are best fit with an energy resolution of @xmath41 . between 1 mev and 2.7 mev , we find a @xmath35ndf for this fit of 1.27 . na source by comparing background data to simulations containing @xmath38k and @xmath39tl components . we clearly see the two compton edges of the 1.4 mev and 2.6 mev gammas produced by these sources . we do not attempt to simulate any other backgrounds , so the divergence of data and mc below 1000 kev is not surprising . there is evidence for a non - linearity in the energy scale at these high energies that is likely related to a the response of the top pmt , and we correct for this using eq . [ eq : energy].,width=302 ] previously , we reported on use of a @xmath18kr@xmath19 generator as a source of low energy electronic recoils in liquid argon and neon @xcite . this kr generator became available at the end of the experiment described here to provide a second , independent energy calibration at 41.5 kev , with a clear peak appearing in the data as shown in fig . [ fig : kr ] . when the energy calibration is derived from the @xmath26na spectrum as described above , the energy of the @xmath18kr@xmath19 peak agrees with 41.5 kev within statistical uncertainties , suggesting both that the energy calibration is accurate and that the response of the microclean detector is linear between 9.4 kev and 511 kev ( recall that the @xmath18kr@xmath19 atoms emit two electrons at 9.4 kev and 32.1 kev to provide the full energy peak ) . the resolution of our detector at 41.5 kev is @xmath42 ( @xmath27/e of the gaussian fit shown in fig . [ fig : kr ] ) . the energy calibration described above uses the total signal from both pmts . given the absence of a well - defined mean single photoelectron in the top pmt , the absolute signal yield is estimated from the bottom pmt alone by matching the energy calibration to the integrated charge of the bottom pmt divided by the size of the single photoelectron of the bottom pmt . after ending the liquid neon run described in this report , we filled the detector with liquid argon without warming up to room temperature or opening the system ; at the higher temperatures of liquid argon , we were able to resolve the single photoelectron response of both pmts . from these measurements , we determined that the top tube observes @xmath43 more signal than the bottom tube @xcite , due to some combination of photon collection efficiency and photocathode quantum efficiency . therefore , we extrapolate the total light yield in the liquid neon to include both tubes based on the observed pmt efficiencies . using this method , we determine a photoelectron yield of @xmath44 photoelectrons / kevee from the @xmath18kr@xmath19 data . the total uncertainty includes an estimate of the uncertainty in the relative efficiency of the two tubes introduced by changing the temperature from 85 k to 25 k , but it is dominated by the single photoelectron response of the bottom pmt . we next use the @xmath18kr@xmath19 data to validate the @xmath26na calibration method . neglecting the systematic uncertainties from the pmt photoelectron response and efficiency , as they apply equally to both methods for determining the light yield , the standard @xmath26na energy calibration taken at the same time as the @xmath18kr@xmath19 data returns a light yield of @xmath45 photoelectrons / kevee , in agreement with the @xmath18kr@xmath19 calibration . this agreement provides confidence that we can trust the @xmath26na calibration for the data taken before the the @xmath18kr@xmath19 source became available . as will be discussed in sec . [ sec : detstab ] , the signal yield was not stable throughout the course of the run ; the largest signal yield observed in this study was @xmath46 photoelectrons / kevee , a factor of about 3.5 larger than we achieved in our previous measurement ( 0.9 photoelectrons / kevee @xcite ) . we attribute the increase to immersing the pmts in the liquid instead of viewing through pressure windows , improved photocathode coverage , and a change in the thickness of the wavelength shifter . kr@xmath19 run in neon . as discussed in the text , the observed light yield for these data is @xmath44 photoelectrons / kevee . the resolution is @xmath42 ( @xmath27/e of the gaussian fit shown in red ) at 41.5 kev.,width=302 ] the presence of impurities can reduce the amount of light detected both by quenching excimer states non - radiatively or by absorbing the vuv scintillation light , and both processes have been studied in liquid argon @xcite . we monitored both the light yield and the triplet lifetime over the course of the experiment to gauge the effect of impurities . initially , the signal yield of the detector was only 1.9 photoelectrons / kevee . we then engaged the circulation pump and flowed neon through the charcoal trap shown in the schematic of fig . [ fig : gashandle ] to remove impurities . the light yield subsequently increased , reaching a peak at @xmath46 photoelectrons / kevee as shown in fig . [ fig : circulation ] , providing evidence that impurities were being contained by the charcoal . because we could not purge the internal charcoal trap while running the experiment , it is not clear whether all the impurities in the neon were removed or whether the trap became saturated . these data were taken for a liquid temperature of 28.7 k. na source . the light yield reached a maximum at @xmath46 photoelectrons / kevee . the error bars are statistical only and do not include the uncertainty introduced by the single photoelectron response of the bottom pmt.,width=302 ] both the light yield and the triplet lifetime varied with time . as will be discussed in sec . [ sec : temp ] , both of these parameters are strongly and unexpectedly affected by the temperature and pressure of the liquid neon . to stabilize the temperature and pressure , we disengaged the circulation pump for the remainder of the run , potentially allowing contaminants to build up in the neon and decreasing the observed light yield in the @xmath18kr@xmath19 data taken at the very end of the experiment to the @xmath44 photoelectrons / kevee reported above . to account for the temperature effects , the psd and @xmath1 data were collected at three different temperatures , 26.7 k , 27.8 k and 28.7 k , where each data set was taken over a roughly two week period . we measured the light yield approximately every other day during each run to determine the stability of the detector for each temperature set point . the observed signal yields were @xmath47 , @xmath48 and @xmath49 photoelectrons / kevee at 26.7 k , 27.8 k and 28.7 k respectively , where the error bars represent the root - mean - square error of all the calibration runs taken at the operating temperature . we therefore conclude that the detector signal yield was stable to better than 3@xmath50 for each temperature setting . we define the prompt fraction @xmath51 as @xmath52 where @xmath53 is the voltage trace from the pmt , @xmath54 ns is an integration period determined to optimize the psd @xcite , @xmath55 ns , @xmath56 @xmath10s , and @xmath57 is the trigger time . as mentioned in the previous section , we collected data at 26.7 k , 27.8 k and 28.8 k. in energy bins of 5 kevee , we fit a gaussian function to the empirical distributions to determine the mean prompt fraction , @xmath58 , for nuclear and electronic recoils , and figure [ fig : fp ] shows @xmath58 as a function of energy for both classes of events . the error bars include both statistical uncertainties and a systematic uncertainty of @xmath59 stemming from differences in the two pmts , with the systematic error generally dominating ; the total uncertainty is smaller than the size of the markers for the electronic recoil data . there is also a small systematic error associated with fitting an asymmetric distribution with a symmetric gaussian function that we do not include in this analysis . the nuclear recoil data were acquired in coincidence with a bc501a organic scintillator module as described in sec . [ sec : calibrations ] . two effects may be observed : first , the mean prompt fractions for nuclear and electronic recoils converge at low energies . this effect has been observed in liquid argon @xcite and is likely due to the fact that @xmath60 for nuclear and electronic recoils also converge as the energy decreases . also , the mean prompt fraction decreases with increasing temperature and pressure for both electronic and nuclear recoils at all energies . this will be discussed further in sec . [ sec : temp ] . uncertainty stemming from differences in the two pmts.,width=302 ] we define the electronic recoil contamination ( erc ) as the probability of mistaking an electronic recoil for a nuclear recoil given a particular level of nuclear recoil acceptance . we estimate the erc as the number of tagged electronic recoil events with @xmath61 , setting the nuclear recoil acceptance level to approximately @xmath62 . the observed erc at each temperature is shown in fig . [ fig : psd ] . our observed erc is about a factor of 5 - 10 better than that observed in @xcite . theoretically , we might have expected a stronger improvement given that we collect at least three times as much light . the smaller than expected improvement is in part attributable to the use of a @xmath63 that changes with energy , instead of the flat cut value assumed previously in @xcite . as shown in fig . [ fig : fp ] , the mean prompt fractions for electronic and nuclear recoils converge at lower energies , reducing the discrimination power . = 0.24 ( see sec . [ sec : leff]).,width=302 ] a second cause of weak psd is the electronic noise in the experiment . we fit the data using a statistical model for @xmath51 described in @xcite , and figure [ fig : psdmodel ] shows the psd for the 26.7 k data with a fit to this model . in @xcite , we described a method for estimating the size of the additional noise for comparison between detectors , modeling the prompt and late noise as a constant multiplied by the mean prompt and late signals : @xmath64 and @xmath65 . as discussed in @xcite , we are wary of interpreting these fit parameters as a literal measurement of the noise ; however , we do believe they can be used as points of comparison . with that method , the noise parameters in the liquid neon are a factor of two to three larger than those observed in argon . the probable cause for such a large amount of noise is the poor gain characteristics of the pmts and the use of an additional amplifier physically far away from the pmt itself , in addition to the factor of four longer integration period required for liquid neon scintillation . the solid line in fig . [ fig : psdmodel ] shows the statistical model prediction in the ideal case where the additional noise is set to zero . we note that the effect of the additional noise is most relevant for energies above 100 kevr . to test the hypothesis that the poor pmt response negatively affected the psd results , the signals in each pmt were analyzed separately . as expected given the single photoelectron distributions , the bottom pmt provided as much as an order of magnitude better psd compared to the top pmt above 150 kevr , dropping to a factor of about 2 at lower energies , although the combination of the two pmts still provides an improvement over just the bottom pmt alone . to achieve better overall psd , more effort will be needed to improve the response of the pmts at liquid neon temperatures ; as the two pmts exhibited different behaviors when cold , all pmts intended for use in a liquid neon detector may need to be cryogenically tested . another potential area of improvement would be a more sophisticated discrimination statistic . given the complicated nature of scintillation in liquid neon ( see sec . [ sec : temp ] ) , a psd analysis using more of the timing information along the lines of the multi - bin method described in @xcite might prove rewarding . as mentioned in the introduction , it is well known that nuclear recoils produce less scintillation light in liquid noble gases than electronic recoils do . there are several processes related to the linear energy transfer ( let ) of the incoming radiation that are known to contribute to the lower scintillation efficiency of nuclear recoils . the first of these is the lindhard effect @xcite , where some of the energy deposited goes into heat instead of the creation of molecules . the lindhard effect has been shown to be sufficient to explain quenching in germanium @xcite , but it is inadequate to explain @xmath1 for the noble gases . a second mechanism is known as bi - excitonic or hitachi quenching @xcite , whereby @xmath66 when such a reaction occurs , the potential for creating two photons from the two original exciton states has been reduced to one photon via recombination of the single resulting ion . this reaction is more likely to occur in high density tracks like those produced by nuclear recoils . recently , a model has been proposed to phenomenologically tie this quenching to the stopping power @xcite . lastly , some fraction of ionized electrons never recombine ; if these `` escape electrons '' are more likely for nuclear recoils than electronic recoils , this process would contribute to a smaller @xmath1 . this effect has been observed to lower @xmath1 in liquid xenon @xcite . the experimental setup for measuring the nuclear recoil scintillation efficiency has been described in @xcite in the context of a measurement in liquid argon . one salient difference between the current analysis and the argon analysis is the absence of mc simulations of the neutron scattering . there are few experimental data on neon - neutron scattering , with geant4 switching discretely from data to a model for neutrons of energy @xmath67 mev . there are efforts underway to improve this situation for future studies @xcite . instead of fitting mc spectra to the data to determine the scintillation efficiency , a simple gaussian function is fit to the data to determine the mean of the observed peak in units of kevee . given the non - gaussian nature of the single photoelectron distribution , we expect some skewing of the observed spectra towards high energies . in a simple mc of a similar situation in a larger detector , the contribution of this skew is @xmath68 for the lowest energy point , below the level of our dominant uncertainties . hence , our scientific judgment is that the uncertainty introduced by this skew does not contribute significantly to our final uncertainty estimate . the final expression for @xmath1 is the observed mean of the signal yield divided by the expected energy of the recoil as predicted by eq . [ eq : neutronscattering ] . data were taken at nine different scattering angles corresponding to energies ranging from @xmath69 kevr to @xmath70 kevr . each point , with the exception of the point at 178 kevr , was taken at the three different operating temperatures . the results at each temperature are consistent with each other to within the uncertainties of the measurement , so for the final determination of @xmath1 , all the data from each temperature are combined into one data set . figure [ fig : peaks ] shows the final gaussian fits at each energy , and figure [ fig : neonleff ] shows the observed nuclear recoil scintillation efficiency , along with the result from the previous measurement at 387 kevr @xcite . to obtain @xmath1 we assume that the signal yield from electronic recoils scales linearly with energy ( consistent with our observations between 40 kevee and 511 kevee ) and compare to the corresponding signal yield at each nuclear recoil energy . the mean value of @xmath1 thus obtained is 0.24 . because we expect there to be some contamination from multiple - scattering and gamma backgrounds into the distribution , as observed in argon data @xcite , we limit the fit range in the region of the peak in the distribution ; the fit range is iteratively chosen to encompass plus and minus one sigma from the central value representing our estimate of the mean energy of the peak . the uncertainty associated with this choice of fit range will be discussed in sec . [ sec : neonlefferrors ] . the average @xmath35ndf for the fits shown in fig . [ fig : peaks ] is 1.07 ; the largest value of @xmath35ndf is 2.5 for the 132.2 kevr peak , with all others below 1.3 . one data selection cut requires that the two pmts trigger within 40 ns of each other and that the asymmetry parameter is within one sigma of the mean asymmetry value for the run . to select neutron - like events , a psd cut was applied to the organic scintillator data as shown in fig . [ fig : oscpsd ] . we also apply a tof cut between an event in the neon and the tagging event in the organic scintillator . a final cut is applied based on the psd described in the previous section ; this is a very loose cut , eliminating events farther than two sigma from the measured mean value of @xmath58 , for nuclear recoils at 26.7 k. the tof cut appears to be quite effective in selecting single - scatter nuclear recoil events in liquid neon . for all runs , neutron events formed a clear peak in the tof spectrum . as single - scattering neutrons are more likely to be found at the very beginning of the tof peak , the tof cut was located at the front side of the peak . first , the total number of counts in the neutron peak was found , and then the tof distribution was scanned until the number of counts in a 4-ns - wide bin exceeded @xmath36 of the total integral of the neutron peak ; this point set the left boundary of the tof window . the width of the tof acceptance window was chosen to be 10 ns . an example of the tof cut is shown in fig . [ fig : tofwindow ] for 368.7 kev events . figures [ fig : tofhe ] and [ fig : tofle ] show the change in the observed energy spectra produced by sliding the location of the tof cut window to earlier and later times for both the lowest and highest energy runs . if the window is moved 10 ns earlier , very few events pass the cut and no peak in energy is visible . at the standard cut location , clear peaks in the energy spectra for both the low energy and high energy points are visible . as the window moves to accept events with longer times - of - flight , the energy peak loses definition due to the inclusion of multiple - scattering backgrounds , becoming an almost flat background when the tof cut is moved to the center of the tof peak ( at 20 ns beyond the standard cut ) . these results provide confidence that the standard tof cut at the front side of the tof peak is selecting single - scattering events and eliminating multiple - scattering backgrounds . there are a number of sources of uncertainty in this measurement of @xmath1 relating to both experimental parameters as well as data selection cuts . there is a general uncertainty of @xmath36 from the energy calibration described above . studies of the analysis cuts were performed independently to assess the size of the systematic error associated with each . in each case , we took the difference between the value of @xmath1 determined by the standard analysis and that determined with the changed cut value as our estimate of the systematic error associated with that cut . the size of the coincidence window between the trigger times of the two pmts was narrowed to 20 ns , and the cut was dropped entirely to study its effect . both the individual trigger cuts and the asymmetry cut were expanded to encompass the region within two sigma of the mean for the entire run . the range in pulse area of the organic scintillator psd cut was first narrowed and then scanned across the entire distribution of nuclear recoils ( the lower band in fig . [ fig : oscpsd ] ) . to assess the uncertainty associated with the tof cut , the width of the tof window was both narrowed to 6 ns and expanded to 16 ns , without changing the location ; as already discussed , moving the tof cut window away from the front of the tof peak eliminated any observable peak , so the location of the cut was not included as a systematic uncertainty . finally , the analysis was repeated with no psd cut applied to the neon data . to estimate the size of the uncertainty associated with the fit range in fig . [ fig : peaks ] , the range was expanded to include the region within two sigma of the central value . the uncertainty associated with the electron equivalent energy calibration was determined to be 2@xmath50 . a source of uncorrelated error is uncertainty in the location of the organic scintillator affecting the predicted value of the nuclear recoil energy . an uncertainty of 0.64 cm in the location of the organic scintillator around the detector corresponds to an angular uncertainty of 1.3@xmath71 at each position . a final source of uncertainty is the trigger efficiency . data for the three lowest energy positions were taken at different hardware trigger thresholds such that the difference between the lowest and highest threshold was larger than the average height of a single photoelectron in each pmt . the statistical uncertainties in the fit results of fig . [ fig : peaks ] were generally below the level of the other uncertainties in the measurement . the main source of uncertainty for the lower energies found in the experiment was the choice of fit range , producing an uncertainty of as large as @xmath72 in the 28.9 kevr data set . in all cases , the effect of increasing the fit range was to increase the observed value of @xmath1 . the results from changing the hardware threshold were consistent with each other at each of the lowest three energy points . at the highest energies , the dominant source of uncertainty was the width of the tof window . all sources of uncertainty were assumed to be independent and combined in quadrature to produce the final uncertainty values shown in fig . [ fig : neonleff ] . table [ table : leff ] lists the central values and the uncertainties associated with each identified source of error . [ cols="^,^,^,^,^,^,^,^,^,^,^,^,^,^,^,^ " , ] the first point to note about our results for @xmath1 in liquid neon is an upturn at low energies similar to one observed in our previous measurement of @xmath1 in liquid argon @xcite , although the uncertainties on the lowest energy points are also the most significant . both hardware and software trigger effects were examined in attempts to explain this upturn , with neither producing the observed result . as with the argon , we do not know whether this is a physically real effect . in the introduction , two main processes were listed as causes of the lower scintillation efficiency of nuclear recoils in liquid noble gases . the first was the lindhard effect , where some energy goes into heat instead of the creation of molecules . a second effect was hitachi quenching , where two excited atomic states interact to produce one ground state atom and one ion , reducing the total number of ne@xmath73 molecules available to produce scintillation photons . et al . _ have developed a phenomenological model to account for these two effects @xcite , and we apply that model here to the current results drawing values for the stopping power from the tables of srim @xcite ( `` srim '' in fig . [ fig : neonleff ] ) . we also use stopping powers obtained from mei et al . from a second model for stopping power as described in @xcite ( `` dm '' in fig . [ fig : neonleff ] ) . the result of fitting the lindhard+birks model to our data is shown in fig . [ fig : neonleff ] . the birks parameters , @xmath74 , for the srim and dm stopping powers are determined to be @xmath75 and @xmath76 mev@xmath77 g @xmath78 , respectively . this can be compared to the value of @xmath79 mev@xmath77 g @xmath78 determined in @xcite from the single data point marked as `` picoclean result '' in fig . [ fig : neonleff ] . the kink that appears around 40 kevr in the srim curve is potentially mirrored in the upturn at low energies but is not present in the dm model . in developing the lindhard+birks model for argon , mei _ et al . _ found a similar kink from the srim data that was not present in alternative models @xcite . the established model for scintillation in liquefied noble gases consists of two exponentially decaying components representing the contribution from singlet and triplet molecules . in our measurements of scintillation in liquid neon , although fast and slow exponential behavior is observed , neither a two - component nor a three - component model is sufficient to describe the time dependence . in addition , the observed time constants and intensities of each component vary significantly with the temperature and pressure of the liquid . in an attempt to understand the various components of liquid neon scintillation and the temperature dependence , we examine average traces from electronic recoil events with energies between 300 kevee and 400 kevee for nine different pressure and temperature settings . all data are derived from tagged @xmath26na events ; the trigger times of each event in the data sets are aligned and the events are summed together to produce the average traces . figure [ fig : neon3exp ] shows an example trace for data taken at 27.4 k , along with fits to a two- and three - exponential model . while the three - exponential model improves on the quality of the overall fit , it still does not accurately model the data between about 50 ns and 1 @xmath10s . we therefore adopt a four - exponential mixture model @xcite . the exponential decay constants are denoted @xmath80 , and the relative weights , which are non - negative and sum to 1 , are denoted @xmath81 . the mixture model for the probability density function of the creation times of scintillation photons created by an energy deposition at @xmath82 is @xmath83 we model the expected voltage trace produced by the pmts as a convolution of an experimentally determined impulse response function of the pmt , @xmath84 , and the pdf model : @xmath85 where @xmath57 is the time of the energy deposit . we perform the fit by minimizing the matusita distance@xcite between the model and the data . example fits to the 26.7 k and 28.8 k data are shown in fig . [ fig : kevinexample ] . we associate @xmath86 and @xmath87 with the longest and shortest time components of the scintillation , respectively , and they are shown in fig . [ fig : tau12 ] as a function of neon temperature . the short time constant , @xmath87 , should not be strictly associated with the decay of the singlet state , as the fast component is also affected by the timing characteristics of the tpb wavelength shifter and the pmts . the remaining scintillation time constants , representing intermediate time scales , are shown in fig . [ fig : tau34 ] . the weights of each component as a function of temperature are shown in fig . [ fig : ps ] , and all fit parameters are reported in table [ table : neonparameters ] . and @xmath87 , for neon scintillation as a function of neon temperature . the error bars represent the combined estimated statistical and systematic uncertainties , derived as described in the text . , width=302 ] and @xmath88 , for neon scintillation as a function of neon temperature . the error bars represent the combined estimated statistical and systematic uncertainties , derived as described in the text . , width=302 ] we estimate one sigma random uncertainties using a bootstrap resampling scheme @xcite . we determine approximate estimates of the systematic uncertainties associated with choice of fit window by varying the end time of the integral in eq . [ eq : model ] between 44 and 52 @xmath10s . we estimate the systematic uncertainty associated with our measurement of the single photoelectron response of the pmts by using several different estimates of @xmath84 taken from different data sets throughout the run . we also perform the fit for each pmt individually to estimate the uncertainties associated with the different pmts . in figs . [ fig : tau12]-[fig : ps ] , the error bars represent the combination in quadrature of the statistical and systematic uncertainties . in many cases , the error bars are smaller than the size of the marker . as a further consistency check , the analysis was repeated for events with energy between 50 and 150 kevee , and all the trends discussed below were observed . we make a few general observations regarding the data before continuing to a discussion of the physics . first , the observed time constants decrease with increasing temperature and pressure ( figs . [ fig : tau12]-[fig : tau34 ] ) . second , the relative intensity of the long - lived component increases with temperature , primarily at the expense of the intermediate components with a small contribution taken from the short - lived part ( fig . [ fig : ps ] ) . the total signal yield also changes with temperature . figure [ fig : absoluteintensity ] shows the contribution to the total signal yield by each of the four exponential decay processes in the model ( @xmath89 where @xmath90 is the total signal yield ) at four temperatures taken within three days to minimize the effect of changing detector conditions ; the total signal yield increases with increasing temperature , suggesting that not all of the increase in @xmath91 is due to the decreases in the other components . cccccccccccccccccc t & p & @xmath86 & @xmath92 & @xmath87 & @xmath93 & @xmath94 & @xmath95 & @xmath88 & @xmath96 & @xmath91 & @xmath97 & @xmath98 & @xmath99 & @xmath100 & @xmath101 & @xmath102 & @xmath103 + ( k ) & ( mbar ) & & & & & & & & & & & + + 26.7 & 816 & 18.37 & 1.13 & 17.93 & 0.70 & 3.00 & 0.12 & 0.31 & 0.03 & 0.56 & 0.02 & 0.16 & 0.01 & 0.23 & 0.02 & 0.049 & 0.002 + 27.4 & 1048 & 18.34 & 1.12 & 17.05 & 0.75 & 2.72 & 0.18 & 0.26 & 0.03 & 0.66 & 0.02 & 0.13 & 0.01 & 0.17 & 0.02 & 0.039 & 0.003 + 27.7 & 1160 & 16.14 & 1.10 & 16.22 & 1.02 & 2.23 & 0.43 & 0.20 & 0.04 & 0.73 & 0.03 & 0.11 & 0.01 & 0.12 & 0.02 & 0.032 & 0.004 + 28.0 & 1281 & 16.40 & 1.05 & 16.50 & 0.86 & 2.32 & 0.36 & 0.22 & 0.04 & 0.71 & 0.03 & 0.12 & 0.01 & 0.13 & 0.02 & 0.034 & 0.004 + 28.3 & 1380 & 13.77 & 0.15 & 15.72 & 0.74 & 1.46 & 0.37 & 0.15 & 0.02 & 0.81 & 0.02 & 0.097 & 0.005 & 0.07 & 0.01 & 0.025 & 0.004 + 28.5 & 1467 & 13.60 & 0.25 & 14.97 & 0.78 & 1.48 & 0.29 & 0.14 & 0.02 & 0.81 & 0.02 & 0.10 & 0.01 & 0.07 & 0.01 & 0.025 & 0.003 + 28.6 & 1520 & 12.33 & 0.26 & 14.88 & 0.51 & 0.89 & 0.12 & 0.10 & 0.01 & 0.85 & 0.01 & 0.089 & 0.004 & 0.04 & 0.01 & 0.020 & 0.002 + 28.7 & 1575 & 12.09 & 0.38 & 13.77 & 0.67 & 0.75 & 0.08 & 0.09 & 0.01 & 0.86 & 0.01 & 0.08 & 0.01 & 0.037 & 0.004 & 0.020 & 0.002 + 28.9 & 1662 & 12.07 & 0.39 & 13.95 & 0.73 & 0.73 & 0.08 & 0.09 & 0.01 & 0.86 & 0.01 & 0.08 & 0.01 & 0.035 & 0.004 & 0.019 & 0.002 + + 27.7 & 1160 & 13.80 & 0.75 & 18.38 & 0.88 & 2.29 & 0.34 & 0.31 & 0.05 & 0.43 & 0.01 & 0.35 & 0.01 & 0.142 & 0.005 & 0.08 & 0.01 + 28.3 & 1380 & 10.99 & 0.51 & 17.49 & 1.02 & 1.26 & 0.31 & 0.175 & 0.04 & 0.52 & 0.02 & 0.32 & 0.01 & 0.105 & 0.004 & 0.05 & 0.01 + there are two final observations that can be made from the data . first , the long time constant is affected by the presence of impurities , as the long time constant at constant temperature increased during the purification cycle described in sec . [ sec : detstab ] . second , we perform a similar analysis on untagged nuclear recoil data at two different temperatures ( unfortunately , the tagged nuclear recoil data are limited by statistics ) and present the results at the bottom of table [ table : neonparameters ] . the parameter trends between the two nuclear recoil data sets are consistent with the electronic recoil data , although the long time constant for nuclear recoil events appears to be @xmath104 @xmath10s shorter than the corresponding time constants for electronic recoils . besides the very prompt light produced by the decay of singlet states , there are three additional , distinct scintillation components . the triplet lifetime in liquid neon has previously been measured to be 15.4 , 2.9 and 3.9 @xmath10s @xcite , with the large discrepancy between the first and second two measurements attributed to impurities in the liquid in the latter two . in solid neon , the long exponential time constant has been measured to be 3.3 and 3.9 @xmath10s @xcite . a very slow tail in the solid has also been observed with a lifetime of 560 @xmath10s , attributed to excitations of the 3s@xmath105p@xmath73 atomic state @xcite ; in one measurement of the liquid , an excited atomic triplet state was observed to decay non - exponentially , disappearing within several microseconds @xcite . measurements in gaseous neon suggest that long - lived @xmath106p@xmath107 and @xmath108p@xmath107 atomic states can react via three - body collisions to form radiative molecular states @xcite ; the lifetime of the long - lived molecular state in gas , presumably equivalent to the triplet state , was measured to be 6.6 @xmath10s @xcite . lastly , as the temperature of the gas is reduced , the emission due to molecular states disappears , leaving only the atomic spectra behind @xcite . before further discussing the current neon data , we compare the results to the situation in argon and helium . first , the data presented in @xcite for scintillation in liquid argon were taken at many different temperatures covering a larger absolute temperature range than that examined in liquid neon , and there was no change in the observed triplet lifetime or signal yield in argon . in liquid helium , the triplet lifetime is 13 s @xcite . two other characteristic time scales were observed in scintillation in liquid helium . first , metastable he(2@xmath108s ) atoms created by ionizing radiation have been observed to last for 8000 s in vacuum @xcite ; in liquid , these states only last for 15 @xmath10s , and their disappearance is accompanied by the appearance of vibrationally excited triplet molecules , implying the creation of molecules by collisions of the excited atomic states @xcite . second , a @xmath109 component has been observed over a finite time window and attributed to collisions of long - lived triplet states leading to the production of the shorter - lived singlet states which promptly decay @xcite . king and voltz developed a model to describe slow emission light produced by scintillators via triplet - triplet collisions , neglecting the effects of the radiative decay of triplet molecules @xcite . the kv model was adapted by mckinsey et al . to describe the @xmath109 component in helium over a finite time window , although the exact form of the power law depends on the assumed track geometry @xcite . finally , an important similarity between liquid helium and liquid neon that sets them apart from the heavier noble gases is that free electrons form bubbles or `` localized '' states in both liquids @xcite . these bubble states occur because the electron - atom repulsion due to pauli exclusion is strong relative to the weak polarization of helium and neon atoms , as evidenced by a positive scattering length for low energy electrons in helium and neon @xcite . in the heavier noble gases , the polarization term becomes stronger and overwhelms the repulsive pseudopotential . while electron bubbles would influence recombination rates , affecting the timing of scintillation on short time scales as well as the number density of singlet and triplet molecules , it is unlikely that they are important for the late time behavior . however , in liquid helium , he@xmath110 molecules also form bubble states @xcite . similar studies and calculations have not been performed for liquid neon , but it is certainly possible that ne@xmath110 could also form bubble states , affecting the long - lived triplet molecules and providing another difference between scintillation in liquid neon and liquid argon . the comparison to liquid helium seems to provide the most illumination , particularly as free electrons form bubbles in both liquids . as in the current measurements , four time components have been observed in liquid helium scintillation : a prompt decay from singlet molecules , a slow exponential decay from triplet molecules , an intermediate exponential decay from atomic states that can also react to form molecules , and a @xmath109 component from triplet molecules reacting to produce singlet molecules . in neon , we likewise attribute the slowest and fastest ( @xmath86 and @xmath87 ) components to singlet and triplet molecules . the number of promptly created singlet molecules decreases with increasing temperature and pressure , while the number of triplet molecules greatly increases . we know that there is a connection between the recoil track density and the production of singlet and triplet molecules , and a possible contributor to the changing intensities of each component is the changing density of excitations with temperature . from measurements of the electron mobility and the einstein relation , the diffusion constant for localized electrons increases by @xmath111 between 26.7 and 28.8 k @xcite . recalling that one channel for the production of metastable molecules is ion - electron recombination , in a simplified picture one expects a larger diffusion constant for electrons to result in a greater maximal separation between ions and recombining electrons . this could produce a greater proportion of triplet molecules given the necessity of an electron spin flip , a mechanism that has been proposed to explain the behavior of xenon scintillation under an electric field @xcite . however , the increase in @xmath91 with increasing temperature is much greater than the corresponding decrease in @xmath98 , requiring some additional explanation that may be found in the intermediate components . continuing to examine the triplet states , we must also conclude that the apparent lifetime of the triplet molecules changes as a function of the temperature or pressure of the liquid environment . impurities in liquid noble gases can shorten the apparent triplet lifetime via absorption or non - radiative collisions . however , in the data of table [ table : neonparameters ] , the shorter lifetime observed at higher temperature and pressure is not accompanied by a decrease in @xmath91 ; in fact , the contribution of the triplet component greatly increases with decreasing @xmath86 , contrary to what one would expect if decreasing @xmath86 was caused by impurities . one possible mechanism that could explain the changing triplet lifetime without an accompanying reduction in @xmath91 is the presence of molecular bubble states analogous to those found in liquid helium ; one could easily imagine that the radiative lifetime of the molecule changes depending on whether it resides within a bubble or not . if ne@xmath8 molecules are weakly bound in bubbles at these temperatures , then some fraction of molecules could be localized in bubbles with an additional fraction freely existing in the liquid , leading to the following model for the decay rate of triplet molecules , @xmath112 : @xmath113\frac{<n_t>}{\tau_{\bar{b}}}.\ ] ] here , @xmath114 is a function of temperature and pressure representing the fraction of molecules contained in bubbles , @xmath115 is the lifetime of molecules in bubbles and @xmath116 is the lifetime of molecules outside of bubbles . although such a model could account for the observed behavior , it is purely speculative given our lack of theoretical guidance for whether neon molecules truly do form stable bubbles . there are two intermediate components , one with a lifetime of order @xmath117 ns and the other with a lifetime of order @xmath118s . we can again draw a parallel to the situation in liquid helium and attribute some of this component to collisions between triplet molecules seeding the creation of singlet molecules that immediately decay as in the king and voltz theory . however , the kv theory can not entirely account for the intermediate exponential behavior , even if one modifies the kv model to include the decay of the triplet state . therefore , we attribute the longer - lived intermediate exponential component to either excited atomic states producing singlet molecules that immediately decay or to the decay of the long - lived atomic states themselves . in the first case , the observed lifetime @xmath94 is equivalent to the reaction rate of excited atomic states with ground - state ne atoms ; if this collision rate increases with temperature or pressure ( perhaps because of increasing diffusion or bubble fluctuations ) , the observed lifetime would decrease . in the latter case , @xmath94 would be some combination of the true radiative lifetime of the atomic excitation and a quenching reaction . in contrast to the very long - lived exponential channel , the decreasing @xmath94 _ is _ accompanied by a decreasing @xmath100 , suggesting the existence of a non - radiative channel . taking the argument one step further , if we combine the two scenarios with the modification that three - body collisions of excited atomic states create _ triplet _ molecules , the non - radiative channel could result in the production of triplet molecules and the observed increase in @xmath91 . this interpretation could be consistent with the disappearance of a triplet atomic spectrum over several microseconds as observed by suemoto and kanzaki @xcite . in the future , understanding the effects of an applied electric field on liquid neon scintillation might illuminate some of the temperature / pressure dependencies observed in this work . a second study that would be useful in assessing whether molecules reside in bubble states would be further laser spectroscopy ; in liquid helium , molecular bubbles were first discovered in the absence of wavelength shifts for rovibrational structure in emission spectra of he@xmath8 between gaseous and liquid helium . these studies could be accompanied by theoretical work on the potential existence of molecular bubbles in liquid neon . third , the experimental conditions of microclean required that the liquid follow the saturation line , rendering it impossible to separate the temperature effects from those associated with pressure . it is likely that bubble interactions depend on the pressure more strongly than on the liquid temperature , and an experiment capable of probing the liquid phase space away from the saturation line and at higher temperatures could prove interesting . in this paper , we have reported measurements of scintillation light . with pmts immersed directly in liquid neon , we have observed a substantial signal yield of up to ( @xmath119 ) photoelectrons / kevee . the observed signal yield is greater than what would be required for a @xmath3-neutrino measurement in clean and much larger than the signal yield assumed in previous simulations of a tonne - scale clean detector @xcite . we have demonstrated the use of activated charcoal as a purifier to remove light reducing impurities in the liquid . we have quantified the performance of the prompt ratio psd method for liquid neon by measuring the erc for a nuclear recoil acceptance probability of approximately 0.5 between 50 kevr and 300 kevr , although we believe our measurement to be an upper limit on the discrimination power achievable with liquid neon due to noise in our system . as in previous studies of liquid argon , we observe a convergence of the @xmath51 parameter at low energies . we have also made a measurement of @xmath1 in liquid neon . finally , we have observed a very interesting time dependence of the scintillation of liquid neon , consisting of four distinct components and a clear temperature / pressure dependence in both intensity and timing . we gratefully acknowledge dongming mei for useful discussion and data on the stopping power of neon . we thank joshua klein , stan seibert , franco giuliani and thomas caldwell for useful discussion and for their work on the rat simulation package . this work was supported by a packard science and engineering fellowship , yale university , and the u.s . department of energy . computer simulations were supported in part by the facilities and staff of the yale university faculty of arts and sciences high performance computing center .

in this paper , we consider multistopping problems for finite discrete time sequences . -stops are allowed and the aim is to maximize the expected value of the best of these stops . the random variables are neither assumed to be independent not to be identically distributed . the basic assumption is convergence of a related imbedded point process to a continuous time poisson process in the plane , which serves as a limiting model for the stopping problem . the optimal-stopping curves for this limiting model are determined by differential equations of first order . a general approximation result is established which ensures convergence of the finite discrete time-stopping problem to that in the limit model . this allows the construction of approximative solutions of the discrete time-stopping problem . in detail , the case of i.i.d . sequences with discount and observation costs is discussed and explicit results are obtained . . .

1201.0083

in this paper , we consider multistopping problems for discrete time sequences @xmath2 . in comparison to the usual stopping problem , there are @xmath1 stops @xmath3 allowed . the aim is to determine these stopping times in such a way that @xmath4 = e[x_{t_1}\vee\cdots \vee x_{t_m } ] = \mathrm{sup}.\ ] ] thus , the gain of a stopping sequence @xmath5 is the expected maximal value of the @xmath1 choices @xmath6 . in the case @xmath7 , this stopping problem reduces to the classical moser problem [ @xcite ] . we will see that optimal @xmath1-stopping times exist and are determined by a recursive description . our aim is to obtain _ explicit _ approximative solutions of the @xmath1-stopping problem in ( [ eq:1.1 ] ) under general distributional conditions . in particular , we do not assume that the random variables @xmath8 are independent or identically distributed or are even of specific i.i.d . form with @xmath9 as assumed in several papers in the literature . our basic assumption is convergence of the imbedded planar point process ( [ eq:1.2 ] ) of rescaled observations to some poisson point process @xmath10 in the plane , @xmath11 here @xmath12 is a nomalization of the @xmath8 induced typically from the central limit theorem for maxima respectively related point process convergence results . our aim is to prove that under some regularity conditions the optimal @xmath1-stopping problem of @xmath0 can be approximated by a suitable formulated @xmath1-stopping problem for the continuous time poisson process @xmath10 which serves as a limiting model for the discrete time model . furthermore , we want to show that the stopping problem in the limit model can be solved in explicit form . the solution is described by an increasing sequence of stopping curves with their related threshold stopping times . these curves solve usual one - stopping problems for transformed poisson processes and are characterized by differential equations of first order , which can be solved either in exact form or numerically . the solution for the limit model also allows us to construct approximative optimal stopping times for the discrete time model . we apply this approach in detail to the @xmath1-stopping of sequences @xmath13 _ with discount _ and observation costs and i.i.d . sequences @xmath14 . it has been observed in several papers in the literature that optimal stopping may have an easier solution in a related form for a poisson number of points or for imbedded homogeneous poisson processes as for instance in the classical house selling problem or in best choice problems . for @xmath7 [ see , e.g. , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite ] . for general reference , we refer to @xcite , chapter 2 . for @xmath15 , multistopping problems were introduced in @xcite who derived some structural results corresponding roughly to theorem [ theo:2.2 ] ; compare also some extensions in @xcite . the two stopping problem has been considered in the case of poissonian streams in @xcite . in this paper , differential equations were derived corresponding to those for the one - stopping problems in @xcite , @xcite and @xcite . multiple buying selling problems were studied in @xcite based on a vector valued formulation with pay - off given by the sum of the @xmath1-choices instead of the max as in ( [ eq:1.1 ] ) ; see also the extension in @xcite . in @xcite ( @xcite ) the case of 2-stopping problems for i.i.d . sequences was treated based on the approximative approach in @xcite ( @xcite ) . the results in this paper were rederived in @xcite ( @xcite ) and in @xcite . in case @xmath7 based on this approximation for several classes of independent and dependent sequences optimal solutions have been found in explicit form [ see @xcite(@xcite ) and @xcite ] . the present paper establishes an extension of the approximative approach as described above to @xmath1-stopping problems as in ( [ eq:1.1 ] ) . it is based on the dissertation of @xcite to which we refer for some technical details in the proofs . the program to establish this approximation approach in general is based on the following steps . in section [ sec:2 ] , we formulate the necessing recursive characterization of the optimal solutions of the @xmath1-stopping problem corresponding to bellman s optimality equation . section [ sec:3 ] is devoted to solve the @xmath1-stopping problem for the limit model of an inhomogeneous poisson process . a particular difficulty arises from the fact that in the limit model the intensity function is typically infinite along a lower boundary curve , in consequence , known stationary markovian techniques as for homogenous poisson processes do not apply . the main result , theorem [ theo:3.3 ] , shows that the optimal @xmath1-stopping problem can be reduced to @xmath1 1-stopping problems for transformed poission processes . the optimal stopping curves are characterized by a sequence of differential equations of first order . in section [ sec:4 ] , we are able to derive explicit solutions for some classes of differential equations , as appearing in the description of the optimal stopping curves . this part is based on developments in @xcite for the case @xmath7 . section [ sec:5 ] gives the basic approximation theorem ( theorem [ theo:5.2 ] ) allowing to approximate the finite discrete problems by @xmath1-stopping in the limit model . the proof of this result needs to develop a new technique . it also uses essentially the extension of the convergence of multiple stopping times in proposition [ prop:5.1 ] in @xcite for @xmath7 to @xmath16 . we restrict our presentation to the essential new part of this proof . finally in section [ sec:6 ] we obtain as application solutions in _ explicit _ form for optimal @xmath1-stopping problems for sequences @xmath13 with @xmath14 i.i.d . and with discount and observation costs @xmath17 , @xmath18 . it is remarkable that we get detailed results including the asymptotic constants as well as approximative optimal stopping sequences in explicit form . our aim is to extend these results in subsequent papers to further classes of stopping problems as to selection problems , to the sum cost case as well as to some classes of dependent sequences . it seems also possible as done in the case @xmath7 , to extend this approach to the case where cluster processes arise in the limit . in this section , we give a formulation of the optimality principle for the @xmath1-stopping of discrete recursive sequences . given a discrete time sequence @xmath19 in a probability space @xmath20 with filtration @xmath21 the @xmath1-stopping problem ( @xmath22 ) is to find stopping times @xmath23 w.r.t . the filtration @xmath24 such that @xmath25= e [ x_{t_1}\vee\cdots \vee x_{t_m } ] = \mathrm{sup}.\ ] ] in case @xmath7 , is identical to the usual ( one-)stopping problem . a well - known recursive solution of this problem [ see @xcite , theorem 3.2 ] is based on the threshold curves @xmath26 of the optimal stopping time defined by @xmath27 \\[-8pt ] w_i & : = & e [ x_{i+1}\vee w_{i+1 } { |}\mathcal{f}_{i } ] \qquad\mbox{for } i = n-1,\ldots , 0 . \nonumber\end{aligned}\ ] ] we need a version of this classical result for stopping times larger than a given stopping time @xmath28 . [ prop:2.1 ] for any time point @xmath29 , the @xmath30-stopping time @xmath31 is optimal in the sense that for any @xmath32-stopping time @xmath33 we have @xmath34 = w_k\ge e[x_t{|}\mathcal f_k ] \qquad p \mbox{-a.s.}\ ] ] for any @xmath32-stopping time @xmath28 , the @xmath32-stopping time @xmath35 is optimal in the sense that for any @xmath32-stopping time @xmath36 with @xmath37 on @xmath38 and @xmath39 on @xmath40 we have @xmath41 = w_s\ge e [ x_t { |}\mathcal{f}_{s } ] \qquad p \mbox{-a.s.}\ ] ] [ rem:2.1 ] for @xmath1 stopping problems , the following variant of proposition [ prop:2.1 ] will also be needed [ for details of the proof , see @xcite ] . let @xmath42 be random variables taking values in a measurable space @xmath43 and @xmath44 a filtration in @xmath45 such that @xmath46 for all @xmath47 . let @xmath28 be an @xmath32-stopping time , let @xmath48 be @xmath49-measurable and @xmath50 be measurable with @xmath51 . also define recursively for @xmath52 @xmath53 \\[-8pt ] w_i(z ) & : = & e[h(y_{i+1},z)\vee w_{i+1}(z ) { |}\mathcal{f}_{i } ] \qquad\mbox{for } i = n-1,\ldots , 0 . \nonumber\end{aligned}\ ] ] then the @xmath54 stopping time @xmath55 where @xmath56 is optimal in the sense that for any further @xmath54-stopping time @xmath36 with @xmath37 on @xmath38 and @xmath39 on @xmath40 we have @xmath57 = w_s(z_s ) \ge e [ h(y_t , z)|\mathcal{f}_{s } ] \qquad p \mbox{-a.s.}\ ] ] similar as for the one - stopping problems the idea of solving is simple . the @xmath58th stopping time @xmath59 should be @xmath60 if the @xmath61-stopping value past @xmath60 with guarantee value @xmath8 is in expectation larger than the @xmath62-stopping value past @xmath60 and with guarantee value reached before time @xmath60 . this idea leads to the following construction . define @xmath63 for @xmath64 and inductively for @xmath65 , @xmath66 define thresholds @xmath67 by @xmath68 \\[-8pt ] \qquad w^m_i(x)&:= & e [ w^{m-1}_{i+1 } ( x_{i+1 } ) \vee w^m_{i+1}(x ) { |}\mathcal { f}_{i } ] \qquad\mbox{for } i = n - m,\ldots , 0 . \nonumber \ ] ] the related threshold stopping times are defined recursively for @xmath69 by @xmath70 for @xmath71 and @xmath72 . equation corresponds to a sequence of @xmath1 one - stopping problems for ( more complicated ) transformed sequences of random variables . the following result extends the classical recursive characterization of optimal stopping times for one - stopping problems in proposition [ prop:2.1 ] to the case @xmath73 . related structural results can be found in the papers of @xcite , @xcite , @xcite , @xcite , @xcite and @xcite ( @xcite ) . [ theo:2.2 ] the @xmath32-stopping times @xmath74 are optimal in the sense that for all @xmath32-stopping times @xmath75 with @xmath76 we have @xmath77 \\ & & \qquad = e \bigl [ w^{m-1}_{t^{m}_{1}(k , x ) } \bigl(x\vee x_{t^{m}_{1}(k , x)}\bigr ) { |}\mathcal{f}_k\bigr ] = w^m_k(x ) \\ & & \qquad \ge e [ x\vee x_{t_1}\vee\cdots \vee x_{t_m } { |}\mathcal{f}_k ] \qquad p \mbox{-a.s.}\end{aligned}\ ] ] the proof of theorem [ theo:2.2 ] follows by induction in @xmath1 based on proposition [ prop:2.1 ] and remark [ rem:2.1 ] similarly as in the case @xmath7 . for details , see @xcite , satz 2.1or @xcite , proposition 2.1 . in general , the recursive characterization of optimal @xmath1-stopping times and values is difficult to evaluate . our aim is to prove that one can construct optimal @xmath1-stopping times and values approximatively by considering related limiting @xmath1-stopping problems for poisson processes in continuous time . in this section , we deal with the optimal @xmath1-stopping problem for the limit model given by a poisson point process @xmath10 . we consider a poisson process @xmath79 in the plane restricted to some set @xmath80\times\overline{\mathbb{r } } ; x > f(t)\},\ ] ] where @xmath81\to{\mathbb{r}}\cup\{-\infty\}$ ] is a continuous lower boundary function of @xmath10 . the intensity of @xmath10 may be ( and in typical cases is ) infinite along the lower boundary @xmath82 . as in @xcite , respectively , @xcite who consider the case @xmath7 , we assume that the intensity measure @xmath83 of @xmath10 is a radon measure on @xmath84 with the topology on @xmath84 induced by the usual topology on @xmath85\times\overline{\mathbb{r}}$ ] . thus any compact set @xmath86 has only finitely many points . by convergence in distribution `` @xmath87 on @xmath84 , '' we mean convergence in distribution of the restricted point processes . this is the basic assumption made in this paper . we generally assume the boundedness condition @xmath88<\infty.\ ] ] let @xmath89\times\overline{\mathbb{r}}\cap m_f))$ ] , @xmath90 $ ] , denote the relevant filtration of the point process @xmath10 . a stopping time for @xmath10 or @xmath10-stopping time is a mapping @xmath91 $ ] with @xmath92 for each @xmath93 $ ] . denote by @xmath94 the reward w.r.t . stopping time @xmath36 . let @xmath95 be a continuous transformation of the points of @xmath10 such that @xmath96 define @xmath97 and for any guarantee value @xmath98 and @xmath99 the _ optimal stopping curve _ @xmath100 of the transformed poisson process by @xmath101 \dvtx t > t \mbox { is an } n\mbox{-stopping time } \ } , \nonumber \\[-8pt ] \\[-8pt ] \hat u(1,x ) & : = & v(1,x ) . \nonumber\end{aligned}\ ] ] it will be shown in the following proposition that the treshold stopping time corresponding to @xmath100 is an optimal stopping time for the poisson process . for the basic notions of stopping of point processes ; see @xcite , respectively , @xcite . the following proposition is the analogue of proposition [ prop:2.1 ] for continuous time poisson processes . it is essential for the solution of the @xmath1-stopping problem of @xmath10 . [ prop:3.1 ] let @xmath10 satisfy the boundedness condition ( ) , let @xmath102 satisfy condition and assume the following _ separation condition _ for the optimal stopping boundary @xmath100 : @xmath103 then : @xmath100 is continuous on @xmath85\times[c,\infty]$ ] and for all @xmath104\times[c,\infty]$ ] holds @xmath105 \nonumber \\[-8pt ] \\[-8pt ] & = & e \bigl [ v\bigl(t(t , x),\overline{y}_{t(t , x)}\vee c\bigr)\vee v(1,x ) \bigr ] \nonumber\end{aligned}\ ] ] with the optimal stopping time @xmath106 @xmath107 is for @xmath108 $ ] the optimal stopping curve of the transformed poisson process @xmath109 for the guarantee value @xmath110 . let @xmath28 be an @xmath10-stopping time , let @xmath111 be real @xmath112-measurable with @xmath113 and @xmath114 the set of all @xmath10-stopping times @xmath36 with @xmath115 on @xmath116 and @xmath117 on @xmath118 . then @xmath119 is optimal in the sense that @xmath120 & = & \hat{u}(s , z ) \nonumber \\[-8pt ] \\[-8pt ] & \ge & e [ v(t,\overline{y}_{t}\vee z ) { |}\mathcal{a}_s ] \qquad p\mbox{-a.s . } \nonumber\end{aligned}\ ] ] for all @xmath121 . \(a ) the statement in ( a ) is proved by discretization . since @xmath122 is continuous and @xmath123 is right continuous there exists a monotonically nonincreasing , continuous function @xmath124\to[\hat c,\infty)$ ] , @xmath125 such that @xmath126 on @xmath127 . thus , for @xmath128 , the sets @xmath129\times\overline{\mathbb{r}}\cap m_{\hat f_2}$ ] are compact in @xmath130 . for @xmath98 , @xmath131 and @xmath132 define @xmath133 } v(\tau_k , y_k\vee x).\ ] ] consider the filtration @xmath134 . then @xmath135 is @xmath136 measurable and @xmath137 , @xmath138 are independent . we define @xmath139\times[c,\infty)\to\overline{\mathbb{r}}$ ] by @xmath140 \dvtx t > t \mbox { an } \mathcal{a}^n \mbox{-stopping time}\ } \qquad\mbox{for } t\in[0,1),\hspace*{-32pt}\nonumber \\[-8pt ] \\[-8pt ] w_n(1,x ) & : = & v(1,x).\hspace*{-32pt } \nonumber\end{aligned}\ ] ] then for @xmath99 by proposition [ prop:2.1 ] , we have @xmath141 = v^n_{\lfloor2^nt\rfloor}(x)\ ] ] with the optimal @xmath142-stopping time @xmath143 and @xmath144 \\[-8pt ] v^n_i(x ) & : = & e \bigl[m^n _ { ( { i+1})/{2^n}}(x)\vee v^n_{i+1}(x)\bigr ] , \qquad i=2^n-1,\ldots , 0 . \nonumber\end{aligned}\ ] ] the function @xmath145 is monotonically nonincreasing and constant on the intervals @xmath146 . we also have @xmath147,\\ ( 2 ) & & \hspace*{5pt } w_n(t , x ) \ge w_{n+1}(t , x ) \qquad \forall t\in[0,1 ] . \ ] ] for the proof of ( 1 ) note that for any stopping time @xmath148 , @xmath149 is an @xmath150-stopping time with @xmath151 and @xmath152 . therefore , @xmath153}v(\tau_k , y_k\vee x ) \ge v(t,\overline y_t\vee x).\ ] ] this implies @xmath154 \dvtx t > t \ n \mbox{-stopping time}\ } = \hat{u}(t , x)$ ] . the proof of ( 2 ) is similar . if @xmath148 is an @xmath155-stopping time , then @xmath156 is an @xmath157-stopping time with @xmath158 and @xmath159 . thus , as above , we obtain @xmath160 . relations ( 1 ) and ( 2 ) imply the existence of a monotonically nonincreasing function @xmath161 \to{\mathbb{r}}\cup\{-\infty\}$ ] with @xmath162 and @xmath163 pointwise . it can be shown by our assumptions on @xmath102 and @xmath10 that @xmath164 is continuous [ see @xcite ] . for @xmath165 with @xmath166 for all compact @xmath167 and for @xmath104\times[c,\infty]$ ] and @xmath168 , we have the convergence @xmath169 with the stopping time @xmath170 \\[-8pt ] & \stackrel{(*)}{= } & \inf\{\tau_k > t \dvtx v(\tau_k , y_k ) > w(\tau_k , x)\ } , \qquad\inf\varnothing:= 1 . \nonumber\end{aligned}\ ] ] for the proof , note that monotone convergence of @xmath171 and continuity of the limit @xmath172 implies uniform convergence from above . thus , for @xmath98 points of @xmath10 on the graph of @xmath173 are ignored by all stopping times @xmath174 and @xmath175 . the second equality @xmath176 holds since @xmath177 and since by assumption @xmath178 is strictly monotonically increasing . this implies by fatou s lemma the following sequence of inequalities : @xmath179\\ & \le & e \bigl[v\bigl(t(t , x),\overline{y}_{t(t , x)}\vee x\bigr)\bigr ] \le\hat{u}(t , x).\end{aligned}\ ] ] thus , @xmath180 is continuous and @xmath181 $ ] . as @xmath182 implies that @xmath183 for @xmath184 , we have @xmath185 $ ] , which means that @xmath186 is the optimal stopping curve of the poisson process @xmath187 with guarantee value @xmath110 . \(b ) to prove optimality of the stopping time @xmath188 , set @xmath189 . then @xmath190 is an @xmath191-stopping time and by holds @xmath192 let @xmath193 be the set of all @xmath191-stopping times @xmath194 with @xmath195 on @xmath196 and @xmath197 on @xmath198 . let @xmath121 . by discretization @xmath115 in general does not imply @xmath199 . thus , we modify the discretization and define @xmath200 . then analogously to @xmath201 this implies the inequalitites @xmath202 \\ & & \qquad \le e [ m^{n}_{t_n}(z ) { |}\mathcal{a}_{s_n } ] \chi_{\ { { \lceil{t2^n}\rceil}/{2^n } > s_n\ } } \\ & & \qquad \quad { } + e [ v(t,\overline{y}_{t}\vee z ) { |}\mathcal{a}_{s_n } ] \chi_{\ { { \lceil{t2^n}\rceil } /{2^n } = s_n\ } } \\ & & \qquad \stackrel{(*)}{\le } \underbrace{e \bigl[m^{n}_{t_n(s_n , z)}(z ) { |}\mathcal{a}_{s_n } \bigr]}_{= w_n(s_n , z ) } \chi_{\ { { \lceil { t2^n}\rceil}/{2^n } > s_n\ } } \\ & & \qquad \quad { } + e [ v(t,\overline{y}_{t}\vee z ) { |}\mathcal{a}_{s_n } ] \chi_{\ { { \lceil{t2^n}\rceil } /{2^n } = s_n\}}.\end{aligned}\ ] ] @xmath176 holds by remark [ rem:2.1 ] . since we have @xmath203 , where @xmath204\times\overline{\mathbb{r}}\cap m_f)$ ] and with @xmath205 , @xmath206 . as @xmath207 we conclude @xmath208 \\ & & \qquad \le e \bigl [ m^{n}_{t_n(s_n , z)}(z ) \chi_{\ { { \lceil { t2^n}\rceil}/{2^n } > s_n\ } } { |}\mathcal{a}_{s } \bigr ] \\ & & \qquad \quad { } + e \bigl [ v(t,\overline{y}_{t}\vee z ) \chi_{\ { { \lceil { t2^n}\rceil}/{2^n } = s_n\ } } { |}\mathcal { a}_{s } \bigr ] \\ & & \qquad = w_n(s_n , z ) e \bigl[\chi_{\ { { \lceil{t2^n}\rceil}/{2^n } > s_n\ } } { |}\mathcal{a}_{s } \bigr ] + e \bigl [ v(t,\overline{y}_{t}\vee z ) \chi_{\{{\lceil { t2^n}\rceil}/{2^n } = s_n\ } } { |}\mathcal { a}_{s } \bigr],\end{aligned}\ ] ] and by the lemma of fatou we have by @xmath209 \le e \bigl [ v\bigl(t(s , z),\overline{y}_{t(s , z)}\vee z\bigr ) { |}\mathcal { a}_{s } \bigr ] = \hat{u}(s , z).\ ] ] as @xmath115 was chosen arbitrary this implies ( b ) . in the sequel , we need the following _ differentiability condition _ to be fulfilled . ( d ) assume that there is a version of the density @xmath210 of @xmath83 on @xmath84 such that the intensity function @xmath211 is continuous on @xmath212\times{\mathbb{r}}$ ] . furthermore , we assume that @xmath213 for all @xmath90 $ ] . the following proposition determines the intensity function of transformed poisson processes . [ prop:3.2 ] let @xmath214 be a poisson process with intensity function @xmath215 satisfying the boundedness condition ( ) . let @xmath95 , @xmath216 be a @xmath217-function monotonically nonincreasing in @xmath218 and monotonically nondecreasing in @xmath219 with @xmath220 for all @xmath90 $ ] . define @xmath221 and @xmath222 . then @xmath223 , @xmath224 with a @xmath217-function @xmath225 . @xmath226 is a poisson process on @xmath227 with intensity measure @xmath228 and intensity fuction @xmath229 , @xmath230 . by @xcite [ ( @xcite ) , proposition 3.7 ] , @xmath231 is a poisson process with intensity measure @xmath232 . the transformation formula implies that the density @xmath233 of @xmath234 is given by @xmath235 after this preparation , we now consider the @xmath1-stopping problem for poisson processes . the aim is to solve @xmath236 = \mathrm{sup},\ ] ] where the supremum is over all @xmath10-stopping times signifies that @xmath237 for each @xmath60 on @xmath238 and @xmath239 on @xmath240 . ] @xmath241 . this problem has been considered for poisson processes on @xmath85\times ( c,\infty)$ ] already in @xcite in the special case of intensity functions of the form @xmath242 with @xmath243 and @xmath244 a continuous distribution function with @xmath245 . equation models the case of i.i.d . random variables arriving at poisson distributed arrival times . @xcite derive for this case differential equations for the optimal stopping curves . explicit solutions are however not given in any case . in the following , we extend these results to the case of general intensities . we subsequently also identify classes of examples of intensity functions which allow essentially explicit solutions . in order to guarantee the existence of optimal @xmath1-stopping times , we restrict ourselves in the following to the case where the lower boundary is constant , @xmath246 . define optimal @xmath1-stopping curves for guarantee value @xmath98 , @xmath247 , and @xmath99 by@xmath248 @xmath249 \dvtx t < t_1<\cdots < t_m\le1\nonumber\\ & & \hspace*{150pt } n\mbox{-stopping times } \ } , \\ u^m(1,x)&:= & x.\nonumber \ ] ] further let @xmath250 for @xmath104\times[c,\infty]$ ] and @xmath251 for @xmath90 $ ] . @xmath252 is called _ optimal @xmath1-stopping curve _ of @xmath10 for guarantee value @xmath219 . define the inverse function @xmath253 by @xmath254\times [ c,\infty].\ ] ] further define @xmath255\times[c,\infty]\to\overline{\mathbb{r}}$ ] by @xmath256 as well as @xmath257 then @xmath258 iff @xmath259 and further @xmath260 the optimal @xmath1-stopping for poisson processes can be reduced by the previous structural results to @xmath1 1-stopping problem for transformed poisson processes . the transformations are given by the optimal stopping curves @xmath261 or equivalently by the inverses @xmath262both sequences of curves are defined recursively . thus , we consider the transformed poisson processes @xmath263 define the ( optimal ) stopping times @xmath264 with guarantee value @xmath219 by @xmath265 \\[-8pt ] \quad t_\ell^m(t , x ) & : = & \inf \bigl\{\tau_k > t_{\ell-1}^m(t , x)\dvtx y_k > \gamma ^{m-\ell+1}\bigl(\tau_k,\overline{y}_{t_{\ell-1}^m(t , x ) } \vee x\bigr ) \bigr\}. \nonumber\end{aligned}\ ] ] the following theorem characterizes the optimal stopping time as threshold stopping time based on the optimal stopping curves . these are given by a system of @xmath1 differential equations of first order . [ theo:3.3 ] let @xmath246 and @xmath10 satisfy the boundedness condition ( ) and the separation condition ( ) , that is , @xmath266 for @xmath99 . let @xmath267 \dvtx \mu((t,1]\times(x,\infty])=0 \}$ ] . then for @xmath247 , @xmath268 holds @xmath269 \\ & = & e \bigl [ u^{m-1}\bigl(t^{m}_1(t , x ) , \overline{y}_{t^{m}_1(t , x)}\vee x\bigr ) \bigr ] \ ] ] with optimal stopping times @xmath270 defined in . for @xmath271\times[c,\infty)\dvtx t < t_0(x)\}$ ] holds @xmath259 while @xmath272 else . in particular , @xmath273 for @xmath99 and @xmath252 is the optimal stopping curve of the transformed poisson process @xmath274 . under the differentiability condition , ( ) @xmath252 solves the differential equation @xmath275 \\[-8pt ] u^m(1,x ) & = & x. \nonumber\end{aligned}\ ] ] for @xmath276 , has a unique solution . if @xmath277 and if @xmath278 where @xmath279 $ ] , then also in this case @xmath280 for @xmath281 is uniquely determined by . the proof is by induction in @xmath1 . our induction hypothesis is that the statement of theorem [ theo:3.3 ] holds and moreover that for any @xmath282-stopping time @xmath28 and any @xmath283-measurable @xmath111 with @xmath113 we have @xmath284-a.s . @xmath285 = u^m(s , z ) \ge e [ z\vee\overline{y}_{t_1}\vee\cdots \vee\overline{y}_{t_m } { |}\mathcal{a}_s]\ ] ] for all @xmath10-stopping times @xmath286 . further , @xmath287\times[c,\infty)\dvtx u^m(t , x)>u^{m-1}(t , x)\}.\ ] ] for the one - stopping problem @xmath7 the statement of theorem [ theo:3.3 ] is contained in @xcite . proposition [ prop:3.1 ] with @xmath288 implies the first part of the induction hypothesis while the second part follows from @xcite , lemma 2.1(c ) . for the induction step @xmath289 , we obtain for all stopping times @xmath290 and @xmath111 @xmath283-measurable by the induction hypothesis ( note that @xmath291 ) : @xmath292 \nonumber\\ & & \qquad \le e \bigl [ ( z\vee\overline{y}_{t_1 } ) \vee\overline { y}_{t_1^{m}(t_1,z\vee\overline{y}_{t_1 } ) } \vee\cdots \vee\overline { y}_{t_{m}^{m}(t_1,z\vee\overline{y}_{t_1 } ) } { |}\mathcal{a}_s\bigr ] \\ & & \qquad = e [ u^m(t_1,z\vee\overline{y}_{t_1 } ) { |}\mathcal { a}_s].\nonumber\end{aligned}\ ] ] this expression is maximized by proposition [ prop:3.1 ] by @xmath293 where @xmath294 the maximizing value is given by @xmath295 . for the proof , we need to show that @xmath296 for @xmath297 . we next establish this and at the same time show ( [ eq:3.23a ] ) for @xmath298 . note that for @xmath98 @xmath299 \dvtx t > t \ n\mbox{-stopping time } \}\\ & \ge & e \bigl[u^m\bigl(t_1^m(t , x),\overline{y}_{t_1^m(t , x)}\vee x\bigr)\bigr]\\ & \stackrel{(*)}{\ge } & e\bigl [ u^{m-1}\bigl(t_1^m(t , x),\overline { y}_{t_1^m(t , x)}\vee x\bigr)\bigr]\\ & = & u^m(t , x ) \qquad\mbox{by induction hypothesis}.\end{aligned}\ ] ] by , we have strict inequality in @xmath176 if and only if @xmath300 . using lemma 2.4 in @xcite , we see that this is equivalent to @xmath301\times{\mathbb{r}})>0 $ ] . this in turn is equivalent to @xmath302\times{\mathbb{r}}\not=\varnothing\ ] ] [ since @xmath303 is monotonically nonincreasing and by definition of @xmath304 . we are going to show that this is fulfilled for all points @xmath305 . so let @xmath305 and thus by induction hypothesis @xmath259 or equivalently @xmath258 . under the assumption that @xmath306\times{\mathbb{r}}\subset a^c$ ] , we obtain that also @xmath307 since @xmath308 is closed . this implies that @xmath309 since @xmath310 is strictly increasing , it follows that @xmath311 , which is a contradiction . thus , holds true . with the choice @xmath312 , @xmath313 further , we obtain @xmath314 = u^{m+1}(t , x).\end{aligned}\ ] ] finally , in holds @xmath315 by proposition [ prop:3.1 ] @xmath316 is the optimal stopping curve of the poisson process @xmath317 on @xmath318 at the guarantee value @xmath219 . we already proved that the separation condition is fulfilled for the stopping of @xmath319 and by proposition [ prop:3.2 ] @xmath319 has the intensity function @xmath320 . the existence and uniqueness results for the differential equation therefore follow with our assumption from the corresponding result in @xcite for the case @xmath7 . for the case of one - stopping problems , some classes of intensity functions @xmath322 have been introduced in @xcite which allow to determine optimal stopping curves in explicit form . solving the optimality equations in for the sequence of optimal stopping curves for the @xmath1-stopping problem is in general much more demanding . however , for some of the classes considered in @xciteexplicit solutions can be given also in the @xmath1-stopping case . we consider intensity functions @xmath322 of the form @xmath323 or @xmath324 as in @xcite with @xmath325 or @xmath326 in case and @xmath327 in case . for the general motivation of these classes and these conditions , we refer to @xcite . in particular , we will see that the main application considered in this paper to @xmath1-stopping of i.i.d . sequences with discount and observation costs is covered by these classes . we first state the results in the three cases mentioned and then give the proof . _ case 1 : @xmath215 satisfies with @xmath102 monotonically nonincreasing , @xmath325 . _ here @xmath328 . @xmath329\to[0,\infty)$ ] is monotonically nonincreasing continuous , @xmath330 and we assume that @xmath331\to[0,\infty)$ ] is a @xmath217-function with @xmath332 on @xmath127 . we define @xmath333 and assume that there exists some @xmath334 with @xmath335 . define @xmath336 , . then for @xmath73 by induction holds : the function @xmath337 given by @xmath338 has exactly one zero @xmath339 and the optimal @xmath1-stopping curves are given for @xmath340 $ ] by @xmath341 where @xmath342\to[r_m,\infty]$ ] is the inverse function of @xmath343 \to[0,\infty]$ ] , @xmath344 the system of functions @xmath345 , respectively , @xmath346 is by recursively defined . in particular , it holds that @xmath347 and thus determination of the optimal stopping curves is reduced to finding a zero point of @xmath348 . _ case 2 : @xmath215 satisfies with @xmath102 monotonically nondecreasing , @xmath326 . _ here @xmath349 . @xmath350\to[0,\infty)$ ] is monotonically nonincreasing continuous , @xmath351 , @xmath352 and @xmath353 for @xmath354 . further , we assume that @xmath331\to[0,\infty]$ ] is a @xmath217-function with @xmath355 on @xmath127 . we define @xmath356 and assume that there exists some @xmath357 with @xmath335 . define @xmath358 , @xmath359 . then for @xmath73 by induction holds : the function @xmath360 defined by @xmath361 has exactly one zero @xmath362 and the optimal @xmath1-stopping curves are given for @xmath363 by @xmath364 where @xmath365\to[r_m,0]$ ] is the inverse of @xmath366 \to[-\infty,0]$ ] , @xmath367 in particular , @xmath368 . _ case 3 : @xmath215 satisfies with @xmath102 monotonically nonincreasing @xmath327 . _ then @xmath349 . @xmath350\to[0,\infty)$ ] is monotonically nonincreasing continuous , @xmath369 and @xmath370 for @xmath371 . further , we assume that @xmath331\to[-\infty,\infty)$ ] is a @xmath217-function with @xmath372 on @xmath127 . we define @xmath373 and assume that there exists some @xmath374 such that @xmath335 . define @xmath358 , @xmath359 . then for @xmath73 by induction holds : the function @xmath375 defined by @xmath376 has exactly one zero @xmath339 . the optimal @xmath1-stopping curves are given for @xmath377 by @xmath378 where @xmath379 $ ] is the inverse of @xmath380\to\overline{\mathbb{r}}$ ] , @xmath381 we have @xmath382 . we only give the proof of case 2 . the proof of both other cases is similar . the proof is by induction in @xmath1 where we additionally include that @xmath383 and thus @xmath384 . in the case @xmath7 , the statement has been shown in @xcite [ with @xmath358 , @xmath359 , @xmath385 . induction step @xmath386 is the optimal stopping curve of @xmath319 at the guarantee value @xmath219 . @xmath319 has the intensity function @xmath387 thus , @xmath388 again is of type and we have to check the conditions of case 2 in @xcite , who deal with optimal one - stopping w.r.t . this type of intensity functions . first , we note that @xmath389 has a zero in @xmath390 since @xmath391 and thus @xmath392 . further by substitution , we have @xmath393 as @xmath394 and @xmath395 . thus , the conditions hold true and the result follows . for intensity functions @xmath215 not of the form as in , the optimality differential equations in theorem [ theo:3.3 ] typically can only be solved numerically . in some cases , however , one can derive bounds for the optimal stopping curves @xmath396 which can be used to derive necessary uniform integrability and separation conditions [ see @xcite , pages 6062 ] for the following approximation result . in this section , an extension of the approximation results in @xcite [ ( @xcite , theorem 2.1 ] and @xcite [ ( @xcite ) , theorem 4.1 ] , for optimal one - stopping problems for dependent sequences is given to the class of @xmath1-stopping problems . for the special case of i.i.d . sequences with distribution function @xmath244 in the domain of the gumbel extreme value distribution @xmath397 a corresponding approximation result was given in the case @xmath398 in @xcite . the following result concerns the dependent case and needs a new technique of proof which is based on discretization . the main result of this section states that under some conditions convergence of the finite imbedded point processes @xmath399 to a poisson process @xmath10 implies approximation of the stopping behavior . we use the same general assumptions as in section 4 of @xcite as well as the notation in section 2 for the poisson process @xmath10 . in particular , @xmath400 are the functions defined in . further , the lower boundary curve @xmath82 of @xmath10 is given by @xmath246 , @xmath10 is a poisson process on @xmath85\times(\overline{\mathbb{r}}\setminus\{c\})$ ] and @xmath401 are the canonical filtrations induced by the imbedded point process @xmath399 and we assume the convergence condition @xmath402 on @xmath84 as throughout this paper [ see ( [ eq:1.2 ] ) and the introduction of section [ sec:3 ] ] . the first result is an extension of proposition 2.4 in @xciteon the convergence of threshold stopping times to the case @xmath73 . for the technically involved proof , we refer to @xcite , lemma 2.6 . [ prop:5.1 ] let @xmath104\times[c,\infty)$ ] be fixed and let @xmath403\to \overline{\mathbb{r}}$ ] be functions such that @xmath404 uniformly on any interval @xmath405 $ ] with @xmath406 . define the corresponding threshold stopping times @xmath407 for @xmath71 . if @xmath87 on @xmath408 , we obtain convergence @xmath409 let now @xmath410 be the stopping thresholds for the @xmath1 stopping of @xmath411 and the filtration @xmath401 ( see section [ sec:2 ] ) . the optimal _ @xmath1-stopping curves _ w.r.t . @xmath401 are defined as follows . for @xmath412 and @xmath64 let @xmath413 and @xmath414 for @xmath415 $ ] . more explicitly , we have for @xmath416 ( see theorem [ theo:2.2 ] ) @xmath417 \dvtx tn < t_1<\cdots < t_m\le n \nonumber\\ & & \hspace*{196.5pt}\mathcal{f}^n\mbox{-stopping times } \bigr\ } \\ & = & e\bigl [ x^n_{t^{n , m}_1(t , x ) } \vee\cdots \vee x^n_{t^{n , m}_m(t , x ) } \vee x { |}\mathcal{f}^n_{\lfloor tn\rfloor}\bigr]\qquad p\mbox{-a.s . } \nonumber\end{aligned}\ ] ] the corresponding optimal @xmath1-stopping times are given by @xmath418 for @xmath71 , where @xmath419 . @xmath420 is right continuous and a piecewise constant curve in the space of random variables . we have the iterative representation ( see theorem [ theo:2.2 ] ) @xmath421 \qquad p\mbox{-a.s.}\ ] ] further , @xmath422 are monotone in the sense that for @xmath423 @xmath424 \qquad p\mbox{-a.s.}\ ] ] in the opposite direction , we obtain for @xmath423 @xmath425 \qquad p\mbox{-a.s.}\ ] ] this follows inductively from the recursive definition of the thresholds @xmath426 . we also need the following further conditions [ for motivation , see @xcite ] : _ asymptotic independence condition . _ for @xmath427 @xmath428 _ uniform integrability condition . _ @xmath429 , with @xmath430 , is uniformly integrable and @xmath431 < \infty.$ ] _ uniform integrability from below . _ for some sequence @xmath432 of monotonically nonincreasing functions @xmath433\to{\mathbb{r}}\cup\{-\infty\}$ ] with @xmath434 pointwise , for all @xmath99 and the corresponding threshold stopping times @xmath435 holds @xmath436 = 0.\ ] ] a modified version of ( l ) is the condition ( l@xmath437 ) : for @xmath247 , there exists some sequence of monotonically nonincreasing functions @xmath438\to\overline{\mathbb{r}}$ ] such that @xmath439 pointwise and further the corresponding threshold stopping times @xmath440 satisfy @xmath441 = 0.\ ] ] condition ( l@xmath437 ) in combination with ( u ) implies uniform integrability of @xmath442 . denote @xmath443 [ theo:5.2 ] assume that @xmath444 on @xmath85\times ( \overline{\mathbb{r}}\setminus\{c\})$ ] and also assume conditions ( ) and ( ) . in case @xmath349 also assume the modified uniform integrability condition ( @xmath445 ) . for all @xmath104\times[c,\infty)$ ] holds @xmath446 if @xmath447 assume @xmath448 . then we have in particluar @xmath449 \to u^m(0).\ ] ] in case @xmath450 are independent random variables and if for @xmath447 we assume that @xmath451 or @xmath452 for @xmath453 , then we obtain @xmath454 if @xmath447 and @xmath448 , then @xmath455 defines an asymptotically optimal sequence of @xmath1-stopping times , that is , convergence as in holds for these stopping times . in case @xmath349 , @xmath456 \eqntext{2\le\ell\le m , } \ ] ] are asymptotically optimal stopping times , where @xmath457 are the threshold functions from condition @xmath458 . since we use point process convergence on @xmath85\times(\overline{\mathbb{r}}\setminus\{c\})$ ] and canonical filtrations , we can apply the skorohod theorem and hence we assume w.l.o.g . @xmath284-a.s . convergence of the point processes . \(a ) consider at first the case @xmath447 . let @xmath99 be a fixed element . we introduce at first discrete majorizing stopping problems . for @xmath73 and @xmath459 , define the discrete time points @xmath460 and discrete time random variables @xmath461 } x^n_j\vee c \qquad\mbox{for } 1\le i\le k,\ ] ] and consider the filtration @xmath462 with @xmath463 . the corresponding @xmath1-stopping curves are given inductively for @xmath73 by backward induction for @xmath464 by @xmath465 \nonumber\\ \eqntext{\mbox{for } i= k - m,\ldots , 0.}\end{aligned}\ ] ] these stopping problems majorize the original @xmath1-stopping problem , @xmath466 \dvtx \\ & & \hphantom{{\operatorname{ess\sup}}\ { } 0<t'_1<\cdots < t'_m\le k \ \mathcal{f}^{n , k}\mbox{-stopping times } \ } \\ & \stackrel{(*)}{= } & { \operatorname{ess\sup}}\ { e [ x^{n , k}_{t'_1}\vee\cdots \vee x^{n , k}_{t'_m}\vee x \dvtx \mathcal{f}^{n , k}_{0 } ] \dvtx \\ & & \hphantom{{\operatorname{ess\sup}}\ { } 0<t'_1\le\cdots \le t'_m\le k \ \mathcal { f}^{n , k}\mbox{-stopping times } \ } \\ & \ge & { \operatorname{ess\sup}}\bigl\ { e\bigl [ x^n_{t_1}\vee\cdots \vee x^n_{t_m}\vee x { |}\mathcal{f}^{n}_{\lfloor t n\rfloor}\bigr ] \dvtx \\ & & \hphantom{{\operatorname{ess\sup}}\ { } tn < t_1<\cdots < t_m\le n \ \mathcal{f}^{n}\mbox { -stopping times } \bigr\ } \\ & = & u^m_n(t , x)\qquad p\mbox{-a.s.},\end{aligned}\ ] ] since for all @xmath467-stopping times @xmath468 it holds that @xmath469 are @xmath470-stopping times with @xmath471 , thus @xmath472 . for the proof of @xmath176 define for @xmath470-stopping times @xmath473 the @xmath470-stopping times @xmath474 by @xmath475 we will prove convergence as @xmath476 to the stopping problem of @xmath477 } y_l\vee c \qquad\mbox { for } 1\le i\le k,\ ] ] with filtrations @xmath478 , @xmath479 and optimal thresholds @xmath480 \qquad\mbox{for } i= k - m,\ldots , 0.\end{aligned}\ ] ] by definition for @xmath481 holds @xmath482 \dvtx i < t\le k - m+1 \ \mathcal{a}^k\mbox{-stopping times}\ } \\ & = & { { } ^m{u}^k } ( a^k_i , x),\end{aligned}\ ] ] where @xmath483 are the optimal stopping curves of the processes @xmath484 at guarantee value @xmath219 . at first we establish that for any @xmath60 the random variable @xmath485 is independent of the @xmath486-algebra @xmath487 . for the proof , note that by condition ( a ) @xmath488 thus , we obtain by the continuous mapping theorem that for any continuous @xmath489 $ ] we have @xmath490 this implies using uniform integrability that @xmath491 \stackrel { l^1}{\longrightarrow } e [ f(y^k_{i+1})].\ ] ] on the other hand , by point process convergence it holds that @xmath492-a.s . and thus also @xmath493 . this implies @xmath494-convergence of conditional expectations : @xmath491 \stackrel { l^1}{\longrightarrow } e [ f(y^k_{i+1 } ) { |}\mathcal{f}^{k}_i].\ ] ] in consequence , we obtain @xmath495 = e [ f(y^k_{i+1 } ) { |}\mathcal{f}^{k}_i]$ ] @xmath284-a.s . for all continuous functions @xmath496 $ ] , and thus independence of @xmath497 and @xmath498 . the next point to establish is proved by induction in @xmath1 . the induction hypothesis is : for all @xmath459 , @xmath98 and @xmath499 @xmath500 for all @xmath501 $ ] and all @xmath98 , we further have @xmath502 we do the induction step for @xmath503 : assertion ( 1 ) we shall prove by backward induction on @xmath60 : for @xmath504 the assertion is trivial . we now consider the induction step from @xmath505 to @xmath60 : from the induction hypothesis , we know that @xmath506 for all @xmath98 . from this , the monotonicity of @xmath507 in @xmath219 and the continuity of @xmath508 in @xmath219 we can conclude that @xmath509 for details , see @xcite . by the induction hypothesis for @xmath60 , we also know that @xmath510 for @xmath98 , implying @xmath511 from this , we get @xmath512 \\ & & \qquad \stackrel{l^1}{\longrightarrow } e [ { { } ^{m-1}{u}^k_{i+1}}(y^k_{i+1})\vee{{}^m{u}^k_{i+1}}(x){|}\mathcal{f}^{k}_i]\end{aligned}\ ] ] as @xmath476 . the expression on the left - hand side equals @xmath513 , and since @xmath514 and @xmath497 are independent as shown above , the right - hand side equals @xmath515 . this completes the induction on @xmath60 and the proof of assertion ( 1 ) . for the proof of assertion ( 2 ) , observe that the process @xmath516 converges on @xmath517\times(c,\infty]$ ] to @xmath518 . further , by induction hypothesis we have uniform convergence of @xmath519 to @xmath520 as @xmath521 . from this , we obtain convergence of the transformed point processes @xmath522 on @xmath523\times\overline{\mathbb{r}}$ ] and thus convergence of the optimal stopping curves of these processes , which proves ( 2 ) . based on ( 1 ) and ( 2 ) , we obtain the estimate @xmath524 the right - hand side converges for @xmath476 and @xmath521 to @xmath525 . thus , we have shown @xmath526 to obtain convergence in probability , we next establish that @xmath527 . this however is implied by the inequality @xmath528\ ] ] holding true for all @xmath467-stopping times @xmath529 , and in particular for @xmath530 for @xmath71 . proposition [ prop:5.1 ] then implies the above statement . for @xmath349 , we obtain similarly the convergence @xmath531 for @xmath276 . then the convergence of @xmath532 results as follows : @xmath533 this implies that @xmath534 for all @xmath535 . let @xmath536 be the stopping times from condition ( l@xmath437 ) and let @xmath537 for @xmath71 . then we obtain by proposition [ prop:5.1 ] and uniform integrability of @xmath538 that @xmath539 \\ & & \qquad \stackrel{n\to\infty}{\longrightarrow } e\bigl[\overline { y}_{t^m_1(t,-\infty ) } \vee\cdots \vee\overline{y}_{t^m_m(t,-\infty ) } \bigr ] = u^m(t).\end{aligned}\ ] ] thus , @xmath540 . as consequence , we obtain @xmath541 which was to be shown . \(b ) for the proof of ( b ) , see @xcite . \(c ) for @xmath349 , we obtain the statement using uniform integrability and proposition [ prop:5.1 ] . for @xmath447 holds @xmath542 \\ & & \qquad = e [ x^n_{\hat{t}^{n , m}_1 } \vee\cdots \vee x^n_{\hat { t}^{n , m}_m}\vee c ] \\ & & \qquad \quad { } - \int_{\{x^n_{\hat{t}^{n , m}_1 } \vee\cdots \vee x^n_{\hat{t}^{n , m}_m } < c\ } } ( c - x^n_{\hat{t}^{n , m}_1}\vee\cdots \vee x^n_{\hat{t}^{n , m}_m } ) \,dp.\end{aligned}\ ] ] the first term converges by proposition [ prop:5.1 ] to the stated limit . the modulus of the second term can be estimated from above by @xmath543 the reason for restricting in ( b ) to independent sequences is the necessity to give estimates of @xmath544 from above [ cf . the case @xmath7 in @xcite ] . in the dependent case , this amounts to . for @xmath281 in contrast to the case @xmath7 one has to consider terms of the form @xmath545 . it seems however difficult to establish the necessary point process convergence of @xmath546 in the general dependent case . as application , we study in this section the optimal @xmath1-stopping of i.i.d . sequences with discount and observation costs . in the case @xmath7 , this problem has been considered in various degree of generality in @xcite , @xcite , @xcite and @xcite . let @xmath548 be an i.i.d . sequence with d.f . @xmath244 in the domain of attraction of an extreme value distribution @xmath215 , thus for some constants @xmath549 , @xmath550 @xmath551 consider @xmath552 the sequence with discount and observation factors , , @xmath553 and both sequences monotonically nondecreasing or nonincreasing . for convergence of the corresponding imbedded point processes @xmath554 the following choices of @xmath555 , @xmath556 turn out to be appropriate : @xmath557 \\[-8pt ] \hat a_n&:= & c_n a_n , \hat b_n:=c_nb_n+d_n \qquad \mbox{for } f\in d(\lambda ) , \nonumber\end{aligned}\ ] ] where @xmath558 , @xmath559 , @xmath397 are the frchet , weibull , and gumbel distributions and @xmath560 , @xmath561 are the corresponding normalizations in . we give further conditions on @xmath17 , @xmath18 to establish point process convergence in . related conditions are given in @xcite in the treatment of i.i.d . sequences with trends , respectively , in @xcite ( @xcite ) . unlike before , @xmath562 denotes here a general constant and not the guarantee value . the guarantee value of @xmath10 is in case @xmath558 given by 0 and in cases @xmath559 , @xmath397 given generally by @xmath563 . this application shows in particular the importance of treating the case with lower boundary @xmath563 as in sections [ sec:2 ] and [ sec:3 ] of this paper , respectively , in @xcite . we state the optimality results for all three cases . it turns out that in all of the following examples the intensity functions of the transformed poisson processes are of the form studied in section [ sec:4 ] . hence , we obtain an explicit form of the solutions and optimal stopping curves . [ theo:6.1 ] let @xmath564 with @xmath565 and @xmath566 ( i.e. , @xmath567 ) . we assume that @xmath568 and also convergence @xmath569\ ] ] with constants @xmath562 , @xmath570 , as well that @xmath571 does not converge to 0 . assume that @xmath572 and that the function @xmath573 , @xmath574 has no zero point . then it holds : let @xmath582 be the functions defined in for @xmath583 . then @xmath584 for @xmath71 are asmptotically optimal sequences of @xmath1-stopping times , that is , the limit in is attained also for these sequences . [ theo:6.2 ] let @xmath585 with @xmath586 and @xmath587 ( i.e. , @xmath588 @xmath284-a.s . ) . further let @xmath589 and @xmath568 , and @xmath590\ ] ] for constants @xmath562 , @xmath570 . if @xmath591 , then assume that either @xmath592 is monotonically nondecreasing or @xmath593 does not converge to 0 . if @xmath597 and the function @xmath598 , @xmath599 has no zero point then holds with @xmath600 . here @xmath601 is the @xmath1-stopping curve of the poisson process @xmath602 . @xmath603 are the corresponding inverse functions defined in and . let @xmath604 be an increasing sequence @xmath605 such that @xmath606 [ e.g. , @xmath607 . define functions @xmath457 by @xmath608 where @xmath609 . then the @xmath1-stopping times defined by @xmath610 for @xmath71 , are asymptotically optimal , that is , convergence as in does also hold for them . if @xmath614 , then @xmath615 -\hat{b}_n}{\hat { a}_n } \to u^m(0),\ ] ] where @xmath576 is the @xmath1-stopping curve of the poisson process @xmath577 with intensity function @xmath616\times{\mathbb{r}}$}.\ ] ] let @xmath582 be the inverse functions defined in and , let @xmath617 be an increasing sequence with @xmath618 ( e.g. , @xmath619 ) . let @xmath457 be defined as @xmath620 then @xmath621 define an asymptotic optimal sequence of @xmath1-stopping times , that is , convergence as in holds for them .

the thermodynamical properties of a generalized dicke model are calculated and related with the critical properties of its energy spectrum , namely the quantum phase transitions ( qpt ) and excited state quantum phase transitions ( esqpt ) . the thermal properties are calculated both in the canonical and the microcanonical ensembles . the latter deduction allows for an explicit description of the relation between thermal and energy spectrum properties . while in an isolated system the subspaces with different pseudo spin are disconnected , and the whole energy spectrum is accesible , in the thermal ensamble the situation is radically different . the multiplicity of the lowest energy states for each pseudo spin completely dominates the thermal behavior , making the set of degenerate states with the smallest pseudo spin at a given energy the only ones playing a role in the thermal properties , making the positive energy states thermally inaccesible . their quantum phase transitions , from a normal to a superradiant phase , are closely associated with the thermal transition . the other critical phenomena , the esqpts occurring at excited energies , have no manifestation in the thermodynamics , although their effects could be seen in finite sizes corrections . a new superradiant phase is found , which only exists in the generalized model , and can be relevant in finite size systems . _ keywords _ : quantum phase transitions .

1605.05357

the description of non - equilibrium dynamics and thermalization in isolated quantum many - body systems has received renewed interest during the last years thanks mainly to the development of novel and powerful numerical approaches to study these problems , as well as , to the advance in sophisticated experimental techniques to control quantum systems with many degrees of freedom @xcite . while the connection between thermodynamics and statistical physics is clear in the context of microscopic laws governed by classical mechanics , for quantum mechanics is not the case because of the difficulties in defining important concepts like chaos , integrability and analytic solvability . we lack a complete framework able to explain how equilibrium and thermal states described by the ensembles of statistical mechanics arise from a microscopic description , and how a quantum phase transition ( qpt ) and its dynamics could be interpreted from the thermodynamic point of view . furthermore , in the case of many particles and degrees of freedom , the techniques developed to deal with the zero temperature case can not easily be extended for solving finite temperature problems @xcite . at zero temperature , quantum systems occupy only the ground - state . however , with finite temperature , a quantum system has enough thermal energy to occupy excited states . therefore , in order to study finite temperature problems in quantum many - body systems the information of their spectra is significant . this leads us to relate the point of view of statistical ensambles , where the temperature plays an important role , with the perspective of isolated systems , where the spectrum is relevant . one interesting feature of the spectrum in quantum many - body systems is the excited - state quantum phase transition ( esqpt ) . an esqpt is a singularity in the density of states which takes place along the energy spectrum for fixed values of the hamiltonian parameters @xcite and has a strong semi - classical connection @xcite . the esqpts have been studied in several nuclear physics models @xcite and it has been suggested they could have important effects in decoherence @xcite and the temporal evolution of quantum quenches @xcite . the relationship between the ground state qpts and esqpts is not completely clear , neither are the dynamical properties of the latter , so these issues are open to current research . the aim of this work is to relate some critical features of the quantum spectrum in atom - field systems , specifically the qpt and esqpt , with their thermodynamics . we study these features in a generalized dicke model , including the dicke and tavis - cummings hamiltonians @xcite , which describe a system of @xmath0 two - level atoms interacting with a single monochromatic electromagnetic radiation mode within a cavity . in the language of quantum computing and quantum information , they also describe a set of @xmath0 qubits from quantum dots , bose - einstein condensates in optical lattices , and circuit qed @xcite interacting through a bosonic mode . both hamiltonians are paradigmatic examples of quantum collective behavior in quantum optics . while the dicke hamiltonian is non - integrable , the tavis - cummings hamiltonian is its integrable version due the rotating - wave approximation ( rwa ) . the dicke model is interesting not only because its experimental realizations , but also thanks to its critical phenomena : the superradiant thermal phase transition @xcite , the well - known superradiant qpt , related with quantum chaos and entanglement @xcite , and the presence of ( _ dynamic _ and _ static _ ) esqpts @xcite . the dicke model is a suited toy model to explore the connection between thermodynamics and the spectrum of quantum systems . in order to address both hamiltonians at once , and at the same time to have the possibility to go from integrability to non - integrability , we put together the dicke and tavis - cummings hamiltonians in one expression introducing a control parameter . we call it the generalized dicke hamiltonian . originally , the thermodynamic analysis for the tavis - cummings model was presented by k. hepp and e. h. lieb in 1973 @xcite . their method was simplified trough the laplace s integral method and extended for the multimode case by wang and hioe @xcite . later , the counter - rotating terms were included in @xcite . in the following years , several authors developed different methods and approaches to study the thermodynamics of the dicke model @xcite . all these approaches rely on the canonical ensemble . the only analysis in the microcanonical ensamble we are aware of , presents some partial results employing non - normalized gaussian distributions @xcite . in this work , we calculate the thermodynamic properties of the generalized dicke model . after a review of the well - known procedure to calculate the canonical partition function employing the laplace s integral method , used as reference , we make the calculation in the micro - canonical ensemble , building a natural link between the thermodynamics and the properties of the quantum spectrum . as the generalized dicke hamiltonian commutes with the total pseudo spin operator @xmath1 , the hamiltonian is block - diagonal in subspaces labeled with @xmath2 running from @xmath3 to @xmath4 , where @xmath5 is the eigenvalue of the total pseudo spin operator . in the last years most of the studies about the dicke model and its qpt have been restricted to the symmetric representation i. e. the subspace with maximum pseudo - spin sector @xmath6 , where the ground - state lies . nevertheless , in order to describe the thermodynamic properties of the full spectrum it is necessary to include all the @xmath2 sectors . as mentioned above , a satisfactory framework for this is still missing , so we employ a semi - classical approach to calculate the microcanonical ensamble . this article is organized as follows . in section 2 we calculate the thermodynamics of the generalized dicke model in the canonical ensemble . the results of this section are recovered in section 3 , but from a microcanonical approach . in order to obtain the number of states for a given energy , a semi - classical approximation to the ground - state energies and to the density of states is obtained for each sector @xmath7 . likewise , the thermodynamical limit of the multiplicities @xmath8 is discussed and obtained . finally , we give our conclusions . besides , we present several appendixes with a detailed discussion of the calculations and other considerations about the thermal phases . in this section , we calculate the thermodynamics of the generalized dicke model following the traditional procedure for calculating the canonical partition function @xcite . the generalized dicke hamiltonian is , @xmath9.\ ] ] with @xmath10 $ ] . when @xmath11 we recover the integrable tavis - cummings hamiltonian @xcite , meanwhile when @xmath12 we have the non - integrable dicke hamiltonian @xcite . the pseudo - spin collective operators are defined in terms of the pauli operators as @xmath13 ( with @xmath14 ) . with this we can write the hamiltonian as , @xmath15 we want to calculate the canonical partition function , @xmath16 with @xmath17 , being @xmath18 the boltzmann s constant . in order to obtain the trace , we chose glauber coherent states for the field part and single pseudo - spin states for the atomic sector , @xmath19 we calculate the canonical partition function as @xmath20 the result is ( for details see appendix a ) @xmath21 where @xmath22 , and , for convenience , we have defined new functions @xmath23\right\}\ ] ] with @xmath24}. \label{chi_delta}\ ] ] the free energy @xmath25 , entropy @xmath26 , and energy @xmath27 per particle are @xmath28,\\ % \frac{\mathcal{s}_{\delta}(t)}{k_{b}}&=\lim_{\mathcal{n}\rightarrow\infty } \frac{1}{\mathcal{n } } \left\{\ln\left[\mathcal{z}_{\delta}(t,\mathcal{n})\right]+\frac{1}{k_{b}\beta\mathcal{z}}\frac{\partial \mathcal{z}_{\delta}(t,\mathcal{n})}{\partial t}\right\},\\ % \mathcal{u}_{\delta}(t)&=\lim_{\mathcal{n}\rightarrow\infty } \frac{1}{\mathcal{n}}\left[\frac{}{}f(t,\mathcal{n})+t\,s(t,\mathcal{n})\right].\end{aligned}\ ] ] to calculate the thermodynamic expectation values of some observables of interest like the number of photons and the pseudo - spin collective operators , we proceed as follows . to calculate the number operator and its powers , we use the same formalism employed to find the partition function . because the number operator only affects the boson trace we have , @xmath31 employing the variables @xmath32 and expanding the newton binomial we have the final expression , @xmath33 now , we find an expression for the thermal average of the collective pseudo - spin operators @xmath30 with @xmath14 . in this case , the operator does not affect the photon trace , only the atomic one , so we come back a few steps in order to calculate it . we start from the photon trace , @xmath34 using that @xmath35 , we see it is possible to reorder the index to obtain the following expression @xmath36 the final result is ( for details see appendix b ) @xmath37\right\}.\end{aligned}\ ] ] the form of the partition function is specially suitable for using the steepest descents method or laplace s integral method @xcite to calculate it . this method consists in approximating the exponential integrand by a gaussian function around the global maximum of the function @xmath38 @xmath39 where @xmath40 are the extremal values of @xmath38 , and @xmath41 is the determinant of the hessian matrix . in order to calculate the extremal values of @xmath42 , we obtain the derivatives of @xmath43 , @xmath44 where we have defined @xmath45 therefore , the first derivatives of @xmath42 are @xmath46\right\},\end{aligned}\ ] ] with @xmath47 the second derivatives are @xmath48\right.\\ \nonumber % \fl & - \left.\frac{4\omega}{\omega_{0}}\left(\frac{\gamma}{\gamma_{\pm}}\right)^{4}\frac{u_{\pm}^{2}}{\chi^{3}_{\delta}(u_{+},u_{-})}\left[\eta\,\chi_{\delta}(u_{+},u_{-})\,\mbox{sech}^{2}\left[\frac{}{}\eta\ , \chi_{\delta}(u_{+},u_{-})\right]+\right.\right . \\ \nonumber % \fl & \left.\left . -\tanh\left[\frac{}{}\eta\ , \chi_{\delta}(u_{+},u_{-})\right]\right]\right\},\\ % \fl \frac{\partial^{2 } \phi_{\delta}}{\partial u_{+}\,\partial u_{-}}&=-2\beta\omega \frac{u_{+}\,u_{-}}{\chi^{3}_{\delta}(u_{+},u_{-})}\frac{4\omega}{\omega_{0}}\left(\frac{\gamma}{\gamma_{+}}\right)^{2}\left(\frac{\gamma}{\gamma_{-}}\right)^{2}\times \\ \nonumber \fl & \times \left\{\tanh\left[\frac{}{}\eta\ , \chi_{\delta}(u_{+},u_{-})\right]-\eta\,\chi_{\delta}(u_{+},u_{-})\,\mbox{sech}^{2}\left[\frac{}{}\eta\ , \chi_{\delta}(u_{+},u_{-})\right]\right\}.\end{aligned}\ ] ] if we look for the extremal points by making @xmath49 , we have three possibilities * a ) * both @xmath50 , * b ) * @xmath51 and @xmath52 , * c ) * @xmath53 and @xmath54 . the case @xmath55 implies that the two equations ( @xmath56 ) @xmath57 must hold simultaneously for the same function @xmath43 , which is impossible , therefore there is not a maximum with both @xmath55 and this case can be discarded . in the following we analyze the remaining cases separately . this first case , @xmath58 , corresponds clearly to a _ normal phase _ due to the fact @xmath59 , i.e. there is an average of zero photons in the field . by noting that @xmath60 , the function and its second derivatives become , @xmath61\right\},\\ \nonumber % & \left.\frac{\partial^{2 } \phi_{\delta}}{\partial u_{+}\,\partial u_{-}}\right|_{(0,0)}=0.\end{aligned}\ ] ] therefore , the determinant of the hessian matrix is @xmath62\right\}\left\{1 - \left(\frac{\gamma}{\gamma_{-}}\right)^{2}\,\tanh\left[\eta\right]\right\}.\end{aligned}\ ] ] to ensure that this point corresponds to a maximum , and consequently that the laplace s method around this point can be used to approximate the partition function , the conditions @xmath63 and @xmath64 must hold simultaneously . since @xmath65 , these conditions are satisfied if and only if @xmath66 the previous condition defines , consequently , the _ normal phase _ region in the parameters - temperature ( @xmath67 ) space . observe that , for @xmath68 the previous condition is satisfied for any value of the temperature , whereas for @xmath69 the condition defines a range @xmath70 $ ] ( @xmath71 $ ] in terms of temperature ) for the normal phase , where @xmath72 using the laplace s method around the point @xmath73 gives @xmath74\right\},\ ] ] where @xmath75 is a function of order @xmath76 which is negligible in the thermodynamic limit @xmath77\right)^{-1/2}\left(1 - \left(\frac{\gamma}{\gamma_{-}}\right)^{2}\tanh\left[\eta\right]\right)^{-1/2}.\ ] ] the free energy per particle is @xmath78.\ ] ] we calculate the partition function s first derivative , @xmath79 to obtain the entropy per particle @xmath80-\eta\,\tanh(\eta).\ ] ] therefore the energy per particle is @xmath81 also , we can calculate the heat capacity , which is @xmath82 similarly , by using the laplace s method , after both integrations , the thermal average of the photon number is @xmath83 and the thermal average of the collective pseudo - spin operator is @xmath84\right\}\right|_{0,0}=\\ \nonumber \fl \sigma_{\delta,\mu}&=-\frac{1}{2}\tanh(\eta)\delta_{\mu , z}.\end{aligned}\ ] ] finally , as we are interested on the energy diagram , we can express entropy in terms of energy . we note that @xmath85 then @xmath86+\left(-\frac{2\mathcal{u}_{\delta}}{\omega_{0}}\right)\,\mbox{arctanh}\left(-\frac{2\mathcal{u}_{\delta}}{\omega_{0}}\right).\ ] ] if we define @xmath87 we finally have @xmath88-\mathcal{e}_{\delta}^{n.ph.}\,\mbox{arctanh}\left(\mathcal{e}_{\delta}^{n.ph.}\right).\ ] ] next , we calculate the remaining cases . the second and third cases corresponds to consider one @xmath89 and the other @xmath90 . these cases are related to a _ superradiant phase _ due to the fact @xmath91 . the condition for having an extreme point ( [ eq1 ] ) for @xmath92 and @xmath93 , reduces to @xmath94 because of the definition of @xmath95 , eq.([chi_delta ] ) , we are only interested in solutions of the previous equation with @xmath96 . to guarantee the existence of a such solution the following condition must be satisfied @xmath97 note that the previous condition holds only for @xmath98 , and , since @xmath99 , if the condition with @xmath100 holds the condition with @xmath101 also does , but not conversely . therefore , for @xmath102 we have only extremal points @xmath103 , whereas for @xmath104 , in addition to the previous ones , a second pair of extremal points appears at @xmath105 . the nature of these extremal points is unveiled for calculating the second derivatives . from eq.([chi_delta ] ) , if @xmath106 then @xmath107 evaluating the second derivatives at these extremal points gives @xmath108 - 1\right\},\\ \nonumber % \left . \frac{\partial^{2 } \phi_{\delta}}{\partial u_{\mp}^{2}}\right|_{(u_{\pm}^{m},0)}= -2\beta\omega\left[1 - \left(\frac{\gamma_{\pm}}{\gamma_{\mp}}\right)^{2}\right].\\ \nonumber % \left.\frac{\partial^{2 } \phi_{\delta}}{\partial u_{+}\,\partial u_{- } } \right|_{(u_{\pm}^{m},0)}=0.\end{aligned}\ ] ] from these expressions the hessian s determinant is @xmath109\left(\frac{\chi_{\delta}^{2}-1}{\chi^{2}_{\delta}}\right)\left\{\eta\,\left[\left(\frac{\gamma}{\gamma_{\pm}}\right)^{2}-\left(\frac{\gamma_{\pm}}{\gamma}\right)^{2}\chi_{\delta}^{2}\right]-1\right\}.\ ] ] as @xmath110 , for the points @xmath111 we have @xmath112 with @xmath113 , whereas for @xmath114 @xmath115 , then they correspond to maximal and saddle points , respectively . in the following we will deal with the maximal , @xmath103 , whose existence is guaranteed by the condition @xmath116 this condition defines the region in the parameters - temperature space corresponding to a first superradiant phase . on the other hand , there is a pair of saddle points that appears at @xmath104 . it defines a second superradiant phase , which coexists with the first one in the same parameters - temperature region . however , this second superradiant phase can not be an equilibrium state from the thermodynamic point of view because it corresponds to a saddle - point . it does not have effects in the thermodynamics properties of the model , but its finite size corrections could make it detectable . the corresponding expressions for this second superradiant phase are very similar to the ones presented below for the thermally relevant superradiant phase . they are shown in appendix c. the transition between the normal and the first superradiant phase is given by the critical temperature defined in eq.([betacrit ] ) . in the region @xmath104 , two saddle points at @xmath105 appear , but the maximal of @xmath117 are still given by @xmath103 . in figure [ fig:1 ] , the normal and superradiant phases are shown in the @xmath118 space for given values of the other parameters ( @xmath119 and @xmath120 ) . the surfaces of the function @xmath117 are also plotted , where it can be seen that the maximal points change from @xmath73 in the normal phase to @xmath103 in the superradiant one . the region where two saddle points appear is also indicated in the diagram and illustrated by a representative @xmath117 surface . it is important to emphasize what happens when we have the tavis - cummings case ( @xmath11 ) and the dicke case ( @xmath12 ) . this is reflected on the critical values of the coupling @xmath121 . in the first case , with @xmath11 we have @xmath122 then , only the normal and the superradiant phase with @xmath123 exist . the integrable tavis - cummings case is the only one which has this feature . for every other value of @xmath124 , which corresponds to non - integrable cases , there are two superradiant phases which can be distinguished . as @xmath124 tends to @xmath125 the critical value for the second superradiant goes to infinity , making this phase unobservable . therefore , again we have only two phases , the normal and the superradiant phases , the latter marked , at @xmath126 , by @xmath127 . [ cols= " < , < , < " , ] for @xmath128 , two different energy regimes exist , one for @xmath129 and a second one for @xmath130 , at these large energies the whole bloch sphere becomes available . the sdos is @xmath131 where @xmath132 and we have used the following definitions @xmath133},\ ] ] @xmath134 with @xmath135.\ ] ] for @xmath136 , the normal to superradiant qpt has already occurred , and a new energy regime appears , which is defined by the interval @xmath137 . the other two intervals are the same as in the previous case ( @xmath129 and @xmath130 ) . the sdos is @xmath138 finally , for @xmath139 a new intertwined energy interval appears , yielding four intervals * 1 ) * @xmath140 , * 2 ) * @xmath141 , * 3 ) * @xmath129 , and 4 ) @xmath130 . the sdos is @xmath142 the energies separating two contiguous energy intervals define the so called excited - state quantum phase transitions ( esqpts ) , because at these energies critical changes in the properties of the sdos and in the topology of the available phase space ( see fig.[fig:7 ] ) are observed . for @xmath128 only one esqpt is observed at @xmath143 . for @xmath136 , two esqpts are present , one at @xmath144 , called _ dynamical _ @xcite , and other at the same energy as before @xmath143 , which was called _ static_. for the last case @xmath139 , in addition to the two previous esqpts , a third esqpt at critical energy @xmath145 , occurs . in fig.[fig:7 ] the sdos for every pseudospin , @xmath146 , are plotted for three different couplings ( @xmath68 , @xmath147 , and @xmath148 ) . the colors of the curves indicate the different energy regimes in the sdos . the number of colors in the curves gives the number of esqpts in the corresponding sdos : * a ) * two colors ( blue and black ) one esqpt , * b ) * three colors ( red , blue and black ) two esqpts and * c ) * four colors ( red , orange , blue and black ) three esqpts . observe that , for the last case ( @xmath148 ) , the three kind of sdos are obtained when the pseudo - spin @xmath2 is varied , 3-esqpts sdos for large @xmath2 , 2-esqpts sdos for intermediate @xmath2 and 1-esqpt sdos for small @xmath2 . it is interesting to observe ( see fig.[fig:6 ] ) that the region in the energy vs coupling space corresponding to @xmath149 ( the orange part of the sdos curves ) coincides with the region where the function @xmath38 of the canonical ensemble present two saddle points at @xmath150 , i.e. the region we have , already identified as the second superradiant phase that could be relevant in the finite size ( @xmath151 ) case . with this , we have obtained the first necessary ingredient in order to calculate the number of states @xmath152 . aside from finding the thermal meaning of the qpt , we have deduced the sdos for each pseudo - spin @xmath153 . while the sdos results are interesting by themselves , we will show below that their contribution , including their critical properties ( the esqpts ) , are completely negligible in the thermodynamic limit . however , they could give rise to physical manifestations when finite number corrections are considered . in the next section we study the properties of the multiplicities @xmath8 , whose properties allow to justify the previous assertions . now , we explore the behavior of the multiplicities @xmath8 , we will show that its thermodynamical ( @xmath154 ) properties completely determine the behavior of the number of states for a given energy , and consequently the entropy of the system . we consider a set of @xmath0 distinguishable particles or qubits , then the degeneracy of each pseudospin @xmath2 , the number of _ physically distinguishable states _ , is given by @xcite @xmath155.\ ] ] it is interesting to observe that , if we consider a set of @xmath0 bosons , only the completely symmetric representation of the collective pseudospin , which corresponds to the maximum value of @xmath156 , has to be considered . consequently , the number of states is given by @xmath157 , and the entropy goes zero in the thermodynamic limit giving no thermodynamic observables @xcite . in fig . [ fig:8 ] we plot the multiplicities as a function of @xmath158 $ ] , for different number of two - level atoms @xmath0 . we can observe that there is a dominant pseudospin whose ratio @xmath159 goes to zero as @xmath154 . the multiplicities represent the statistical weight for a given @xmath2-sector and , as we are interested on thermal observables , we want to know the behavior of @xmath8 in the thermodynamical limit . in order to do so we employ the stirling approximation @xmath160 if we express the previous formula in terms of variable @xmath161 $ ] , we obtain @xmath162.\end{aligned}\ ] ] neglecting terms of order less than @xmath0 , we finally have , @xmath163 in the thermodynamical limit , the multiplicity is a monotone decreasing function of the variable @xmath164 ( see fig.[fig:9 ] ) , whose maximum value , at @xmath165 ( corresponding to lowest @xmath166 pseudospin ) , is @xmath167 . furthermore , in the same limit , all the multiplicities of the larger pseudo - spins @xmath168 ( with @xmath169 ) are negligible respect to the multiplicity of the given pseudospin @xmath170 @xmath171 therefore , since the sdos calculated before grows linearly with @xmath2 , for a fixed energy @xmath172 the number of states will be entirely dominated by the multiplicity of the lowest pseudo - spin compatible with that energy . from the study of the lowest energy for every @xmath2 , we know that , for any coupling , the minimal energy of the pseudo - spin @xmath2 increases as @xmath2 decreases ( as it can be seen in fig . [ fig:5 ] ) . consequently , for a given coupling and energy ( @xmath173 ) only the largest pseudo - spins ( those that satisfy @xmath174 ) are available and the multiplicity of the smallest one , @xmath175 , determines the thermodynamics of the system . then , in the thermodynamic limit , the leading contribution to the number of states for a given energy @xmath172 is given by the multiplicity @xmath176 @xmath177=\ln\left[\sum_{j=0}^{\mathcal{n}/2}y(\mathcal{n},j)\,\nu(e,\mathcal{n},j)\delta e\right]\simeq \ln\left[y(\mathcal{n},z_m)\right],\ ] ] where @xmath178 is the multiplicity ( [ eqn : mul ] ) of the smallest @xmath2-pseudospin available for the energy @xmath172 . the smallest pseudo - spin compatible with a given energy @xmath173 , can be obtained by solving , for @xmath2 , the equation @xmath179 where @xmath180 is the lowest energy for given @xmath2 , eq.([eqn : gse ] ) . the solution , @xmath181 , to the previous equation is given and discussed in the following subsection . with the results of the previous subsections , we have all the necessary ingredients to calculate the thermodynamics of the generalized dicke model from a microcanonical ensemble approach . as it was discussed before , the number of states for a given energy @xmath172 , and consequently the entropy is determined , in the thermodynamic limit , by the multiplicity of the minimal pseudo - spin @xmath182 , compatible with @xmath172 @xmath183,\ ] ] where @xmath184 is given by the solution of @xmath185 . from eq . ( [ eqn : gse ] ) and using fig.[fig:5 ] as a guide , it is clear that the previous equation takes two different forms depending on the values of coupling and energy . for the case @xmath68 or @xmath186 with @xmath187 [ the already identified normal phase obtained in the canonical ensemble approach , with @xmath188 the critical energy given by eq.([eqpt ] ) ] , the equation @xmath189 takes the simple form @xmath144 , whose solution is clearly @xmath190 , which implies @xmath191 for the case @xmath186 with @xmath192 ( the already identified superradiant phase ) the equation @xmath189 becomes @xmath193=-\frac{j\omega_o}{2}\left[\left(\frac{\mathcal{n}}{2j}\right)\left(\frac{\gamma_{+}}{\gamma}\right)^2+\left(\frac{2j}{\mathcal{n}}\right)\left(\frac{\gamma}{\gamma_{+}}\right)^2\right],\ ] ] whose solution is given by @xmath194 the minimal pseudospin @xmath184 is equal to the function @xmath195 ( [ funu ] ) defined in the canonical ensemble approach to express the entropy as a function of the energy . gathering all the previous results , is very easy to prove that the entropy from the microcanonical approach is exactly the same obtained previously from the canonical ensemble . the entropy per particle is given by @xmath196= \frac{1}{\mathcal{n}}\ln\left[\left(\frac{4}{(1-z_m)^{1-z_m}(1+z_m)^{1+z_m}}\right)^{\mathcal{n}/2}\right]\nonumber\\ & = & \ln(2 ) -\frac{1}{2}\left(1+z_{m}\right)\,\ln\left(1+z_{m}\right)-\frac{1}{2}\left(1-z_{m}\right)\,\ln\left(1-z_{m}\right),\end{aligned}\ ] ] which is equal to the canonical result of eq.([entrocan ] ) remembering that @xmath197 and that the internal energy @xmath198 . therefore , we have solved the thermodynamics of the generalized dicke model in the micro - canonical ensemble . let us discuss briefly the case @xmath199 . as the lowest energies for every @xmath2 satisfy @xmath200 , all the pseudo - spins are available for these energy values . therefore , for @xmath199 the number of states is approximated by the multiplicity of the smallest pseudospin @xmath166 , which , as it was noted before , is approximated by @xmath201 . this multiplicity is equal to the dimension of the atomic subspace and independent on energy , consequently the entropy becomes constant for @xmath202 and , since @xmath203 , this energy region is unreachable at finite temperature . the same result obtained previously from the canonical ensemble approach to the thermodynamics . previously , in @xcite , a gaussian approximation to the micro - canonical ensemble was used to study the dicke hamiltonian . there it was concluded that the internal energy @xmath204 is another thermal phase transition . from our results is clear , however , that @xmath204 is simply the infinite temperature limit . other thermodynamical observables can be obtained form the microcanonical approach . we calculate the temperature from the entropy , @xmath205 where the derivative of @xmath184 is @xmath206^{-1/2 } & \mbox{for the superradiant phase } \\ \nonumber & \mbox { ( $ \gamma\geq \gamma_+$ with $ e < e_j^{qpt}$ ) } \\ -1 & \mbox{for the normal phase } \\ \nonumber & \mbox{($\gamma<\gamma_+$ or $ \gamma\geq \gamma_+$ with $ e\geq e_j^{qpt}$ ) } \\ 0 & \mbox{for}\,\,\ , e\geq 0 , \end{array}\right.\ ] ] with @xmath207 . then , the temperature as a function of the energy is , @xmath208^{-1/2}\times & \mbox{for the superradiant phase}\\ \times \mbox{arctanh}\left\{\left(\frac{\gamma_{+}}{\gamma}\right)\left[-\left(2\epsilon+\left(\frac{\gamma_{+}}{\gamma}\right)^{2}\right)\right]^{1/2 } \right\ } & \\ \frac{2}{\omega_{0}}\,\mbox{arctanh}(-\epsilon ) & \mbox{for the normal phase } \\ \end{array}\right.\ ] ] by evaluating the temperature at the critical energy , @xmath209 , we obtain the critical temperature , which separates the normal and superradiant thermal phases @xmath210 we have recovered the critical temperature of the finite - temperature super radiant phase transition obtained previously from the canonical ensemble . besides , as it was already mentioned , we observe that for @xmath211 the derivative @xmath212 , consequently , @xmath213 and this energy range is unavailable for finite temperatures . finally , from ( [ eqn : temmic ] ) we can recover the internal energy in the micro canonical ensemble . for example , in the normal phase the energy is @xmath214 which agrees with our calculations in the canonical ensemble . before concluding , let us discuss the critical properties of the generalized dicke energy spectrum and their relation with the thermodynamical critical phenomena , under the light of our microcanonical results . as it was shown , the thermodynamical properties of the model are entirely given by the properties of the lowest energy state at each pseudospin @xmath2 . particularly , the thermal phase transition line between the normal and superradiant phase , is the aggregated of the qpts of each pseudospin @xmath215 . on the other hand , the other critical phenomena observed in the energy spectrum , the so called esqpts , since they occur at energies larger than the minimal energy of each pseudo - spin @xmath2 have no effect or manifestation in the thermal properties . one of them , the called _ static _ esqpt occuring at energy @xmath216 , belongs to a forbidden thermodynamic energy region ( @xmath199 ) . the other two esqpts , the one occurring at @xmath217 and that at @xmath145 , even if they are in an energy range thermally available , their contributions to the thermodynamics are negligible , and their effects disappear completely in the thermodynamical limit . nonetheless , it would be interesting to determine whether the esqpts have any effect or manifestation in the finite size corrections to the thermodynamical ( @xmath154 ) limit , which goes beyond scope of the present contribution . other interesting topic in the study of the finite size corrections , would be the comparison between a canonical and a microcanonical approach . in the thermodynamical limit , this work has explicitly shown that both descriptions give exactly the same results . we have solved the thermodynamics of the generalized dicke model both in the canonical and , for the first time , in the microcanonical ensemble . in order to calculate the microcanonical ensemble we employed a semi - classical approximation for the density of states . we showed results for all the relevant observables , which let us in a simple way recover the results for interesting temperature limits , @xmath218 , @xmath219 and @xmath220 . we have demonstrated that the results for both ensembles agree in the thermodynamical limit , and , in this way we linked the point of view of the canonical statistical ensambles and the perspective of isolated quantum systems . besides , we obtained expression for the semiclassical dos for the extended dicke model . all of these calculations could help to study problems with a tunable parameter between an integrable system ( tavis - cummings ) and a non - integrable one ( dicke ) . like the dicke and tavis - cummings models , the generalized dicke model studied here exhibits only two thermal phases in the thermodynamical limit , a normal and a superradiant phase . however , unlike the dicke or tavis - cummings models , we identify a region which potentially could give rise to a second superradiant phase . however , this second superradiant phase can not be an equilibrium state . from the thermodynamic point of view , it corresponds to a saddle point , i. e. a non stable phase . from the semi - classical point of view , it is not the minimum of the classical energy surface . the minimum corresponds to the first superradiant phase . the second superradiant phase could have observable effects only in the finite size case . we connected in a simple way the thermal phase transition in the generalized dicke model with the qpts of the lowest energy states of each subspace of pseudospin @xmath2 , by calculating , for each subspace , their degeneracies and semi - classical lowest energies and densities of states . we found that the curve in the energy - coupling space where the qpt for each subspace @xmath2 takes place reproduces the curve of the critical energy corresponding to the critical temperature in the phase diagram , i.e. , to the thermal phase transition . then , the superradiant qpt is the thermal superradiant phase transition . so , we have related these critical phenomena in the spectrum with its thermal counterpart . the excited states of every @xmath2 are thermodynamically inaccessible , and the thermodynamical properties of the model are entirely given by the lowest energy states of each pseudospin @xmath2 . consequently the critical phenomena observed in the excited energy spectrum , the so called esqpts , have no effect and manifestation in the thermal properties . however they could be of interest in a finite size study of the model @xcite . the region of positive energies was also shown to be thermally inaccesible . the formalism presented here is immediately applicable to other systems formed by a set of @xmath0 identical few level atoms whose hamiltonians are expressed in terms of collective operators satisfying a given algebra . example of this is the lipkin - meshkov - glick model whose hamiltonian is expressed in terms of @xmath221 operators . we hope the formalism presented here could help to understand the relationship between the critical properties of the quantum spectrum with the thermal critical phenomena . also , this approach could help the developing of techniques to study finite temperature problems in isolated quantum many - body systems . we thank a. relao , p. prez - fernndez , p. stranksy and m. kloc for fruitful discussions which lead to the development of this work and its enrichment . also , we show our gratitude to p. cejnar for the support and insight , as well as , for the reading of the manuscript . m. bastarrachea - magnani wants to thank r. rossignoli for his interest on the subject and his pertinent advices . this work has received partial economical support from sep - conacyt , and from redtc - conacyt , mexico . in this appendix we calculate the canonical partition function , eq.([zfunc ] ) , @xmath222 following @xcite the field trace is , @xmath223\right\}.\end{aligned}\ ] ] then , we arrange the sum taking advantage of the independence between the atoms , @xmath224^{\mathcal{n}}.\end{aligned}\ ] ] where , @xmath225.\ ] ] in order to calculate the atomic trace we find the eigenstates and egenvalues of @xmath226 . in the @xmath227 basis we have , @xmath228 \\ \frac{\gamma}{\sqrt{\mathcal{n}}}\left[(1+\delta ) re(\alpha)-i(1-\delta)i im(\alpha)\right ] & -\frac{\omega_{0}}{2 } \\ \end{array } \right)\end{aligned}\ ] ] the eigenvalue equation is , @xmath229=0.\ ] ] the eigenvalues @xmath230 are , @xmath231},\ ] ] and the eigenstates , @xmath232 we introduce the function @xmath233 , defined through @xmath234 in order to express ( [ zint ] ) as @xmath235 where @xmath236\right\}\ ] ] with @xmath237}. \label{chi}\ ] ] in order to calculate the partition function , we write @xmath22 and the integral in terms of these scaled variables @xmath238 and @xmath239 @xmath21 where the functions @xmath240 and @xmath233 become @xmath23\right\},\ ] ] and @xmath24}.\ ] ] as the pauli operator @xmath241 only acts over one of the spins we can separate that expectation value from the rest . @xmath242 for all the spins @xmath243 the expectation values are solved in the same way than in appendiz a. now we need to pay attention to @xmath244 . in order to evaluate it we need to look at the form of the pauli matrices in the basis @xmath245 i. e. we are interested on evaluating the following expression , @xmath246 writing the pauli matrices using the kronecker delta as , @xmath247 and employing the eigenvectors @xmath248 in eqn . [ ch3eigen ] expressed in the @xmath164 basis , after multiplying the matrices we have , @xmath249\right\}\ ] ] now , we evaluate the matrix elements realizing we have the same element for all @xmath250 , @xmath251=\\ \nonumber \fl & & = -\frac{2\mathcal{n}\,\sinh\left(\frac{\beta\omega_{0}}{2}\chi_{\delta}(\alpha)\right)}{\omega_{0}\,\chi_{\delta}(\alpha)}\left\{\omega_{0}\delta_{\mu , z}+\frac{\gamma}{\sqrt{\mathcal{n}}}\left[(1+\delta)re(\alpha)\delta_{\mu , x}-(1-\delta)im(\alpha)\delta_{\mu , y}.\right]\right\}\end{aligned}\ ] ] therefore , the expression of the thermal averages for @xmath30 is , using eqn . [ ch3eqjs ] , @xmath252\right\}.\end{aligned}\ ] ] which is finally , @xmath253\right\}.\end{aligned}\ ] ] and in the @xmath32 [ @xmath254 variables @xmath255\right\}.\end{aligned}\ ] ] as explained in the main text , there are two superradiant phases for certain hamiltonian parameters and energies , but they can not coexist thermodynamically and the second superradiant phase is discarded because it corresponds to a non - stable thermal state . for completeness , we present in this appendix the expressions in the canonical ensemble related to this phase , which are analogous to the first superradiant phase . the function @xmath256 evaluated @xmath257 is , @xmath258,\ ] ] the partition function becomes , @xmath259\right\}\right].\ ] ] where @xmath260\right\}^{-1/2}_{\mbox{.}}\ ] ] the free energy would be @xmath261.\ ] ] from where we can obtain the energy and heat capacity @xmath262 , \\ \fl c_{\delta}(t)&=k_{b}\eta^{2}\,\mbox{sech}^{2}\left(\eta\chi_\delta\right ) \left(\frac{\gamma}{\gamma_{-}}\right)^{4}\frac{\tanh^{2}\left(\eta\,\chi_{\delta}\right)}{1-\left(\frac{\gamma}{\gamma_{-}}\right)^{2}\eta\,\mbox{sech}^{2}\left(\eta\,\chi_{\delta}\right)}.\end{aligned}\ ] ] in this appendix we derive the leading order expression of the quantum density of states in the semi - classical approximation , using coherent states . we begin with the quantum density of states @xmath263 with @xmath264 the energy spectrum . we rewrite the dirac delta function in eq . [ a ] in its integral form @xmath265 then , in this way it is possible to write the exponential eq . [ b ] in terms of the eigenstates of the hamiltonian @xmath266 d\tau , \nonumber\end{aligned}\ ] ] and it can be observed we have the trace of the evolution operator multiplied by a phase @xmath267 d\tau.\ ] ] given that the trace of the evolution operator eq . [ e ] is independent of the basis , we can rewrite it in terms of the coherent state basis @xmath268 d\tau,\ ] ] where @xmath269 represents a coherent state associated to the field , @xmath270 represents a coherent state associated to the pseudo - spin , @xmath271 , and the differentials are @xmath272 d i m [ \alpha]}{\pi } \mbox { \ \ \ \ and \ \ \ \ \ } d^2 z=\frac{2j+1}{\pi}\frac{d re[z ] d im[z]}{\left(1+|z|^2\right)^2}.\ ] ] expanding the exponential eq . [ f ] in a taylor series , @xmath273 |{\alpha , z}\rangle,\ ] ] @xmath274 where we have used the normalization of the coherent states @xmath275 . if we assume @xmath276 , we approximate eq . [ g ] by the first two terms and obtain the leading order semi - classical approximation @xmath277 replacing eq . [ h ] in eq . [ f ] , we obtain @xmath278 d\tau\nonumber\\ & = & \int \left[\frac{1}{2\pi } \int_{-\infty}^{\infty } e^{i\left(e - h_{cl}\left(\alpha , z \right)\right)\tau } d\tau \right ] d^{2}\alpha~d^{2}z\nonumber\\ & = & \int \delta\left[e - h_{cl}\left(\alpha , z \right)\right ] d^{2}\alpha~d^{2}z . \label{i}\end{aligned}\ ] ] if we write equation eq . [ i ] in canonical variables , we have an interesting interpretation . the complex numbers @xmath279 and @xmath164 in terms of canonical variables ( quadratures for @xmath279 and a projection over the bloch sphere for @xmath164 ) , eq.([paracoh ] ) , are @xmath280 where @xmath281 , and @xmath282 and @xmath283 are the zenith and azimuthal angles of the bloch sphere . to express the differentials in terms of these variables we need the determinant of the jacobian matrix , @xmath284 for the bosonic variables it is easy @xmath285 the calculation for the spin variables is not straightforward . however , if we know how they are projected over the bloch sphere whose axes are given by @xmath286 , @xmath287 , and @xmath288 , then it is easy to obtain the real ( @xmath289 ) and imaginary ( @xmath290 ) parts of @xmath164 in terms of the canonical variables @xmath291 y @xmath283 @xmath292 with this we have @xmath293 substituting eq . [ n ] in eq . [ o ] in order to calculate the determinant of the jacobian matrix , we have @xmath294 we note that @xmath295 . therefore , eq . [ p ] is simplified to @xmath296 in the limit @xmath297 we obtain @xmath298 finally , replacing the results eq . [ qqq ] and eq . [ r ] in eq . [ i ] we obtain @xmath299 dq_+~dq_-~dj_{z}~d\phi.\ ] ] this equation represents the lowest order semiclassical approximation to the density of sates . we calculate in this appendix the extremal values of the semiclassical hamiltonian , eq.([hclass ] ) , @xmath300.\end{aligned}\ ] ] its derivatives are @xmath301,\\ % \fl \frac{\partial h_{cl}}{\partial \phi}&=-\frac{\gamma\,j}{\sqrt{\mathcal{n}/2}}\sqrt{1-\left(\frac{j_{z}}{j}\right)^{2}}\left[(1+\delta)q_{+}\,\sin(\phi)+(1-\delta)q_{-}\,\cos(\phi)\right].\end{aligned}\ ] ] from @xmath302 and @xmath303 , we obtain @xmath304 now , from @xmath305 we have to possibilities , @xmath306 in the first case , it immediately follows from eqs.([aq+ ] ) and ( [ aq- ] ) that @xmath307 and @xmath308 . as @xmath309 ( north and south poles of the bloch sphere respectively ) the azimuthal angle @xmath283 is completely irrelevant . then , we have two extremal points for any value of the coupling constant @xmath310 by evaluating the hamiltonian at these points we obtain @xmath311 the other case , @xmath312 , implies , from @xmath305 that @xmath313 on the other hand , from ( [ aq+ ] ) and ( [ aq- ] ) it is easy to find @xmath314 in order the two eqs.([aeqn:1 ] ) and ( [ aeqn:2 ] ) hold for arbitrary @xmath124 we have the following possibilities : 1 . @xmath315 and @xmath316 , which gives us the already found extremal points ( [ coornormal ] ) . 2 . @xmath317 , then @xmath318 , @xmath319 , and @xmath316 with @xmath320 . 3 . @xmath321 , then @xmath322 , @xmath323 and @xmath324 with @xmath325 . note that for the tavis - cummings case ( @xmath11 ) both eqs . ( [ aeqn:1 ] ) and ( [ aeqn:2 ] ) become the same , and the condition holds for every @xmath326 $ ] . this case is solved in @xcite . for the second of the latter possibilities , by substituting ( [ aq+ ] ) in @xmath327 , we obtain @xmath328 , where @xmath329 then , by substituting this result in ( [ aq+ ] ) , we obtain @xmath330 . the resulting extremal points are @xmath331 since @xmath332 , the previous point is valid if and only if @xmath333 . by evaluating the hamiltonian at these points we obtain the energy @xmath334.\ ] ] for the third of the previously enumerated possibilities , by substituting ( [ aq- ] ) in @xmath327 , we obtain @xmath335 . then , we substitute this result back in ( [ aq- ] ) to obtain @xmath336 . giving the following extremal points @xmath337 which is a valid point , provided that @xmath338 . the energy of this extremal points is obtained by evaluating the hamiltonian , the result is @xmath339.\ ] ] clearly , as @xmath10 $ ] it is easy to see that @xmath340 . we summarize our findings as follows ( see equally table 1 of the main text ) 1 . for @xmath341 , the energies of the extremal points in ascending order are @xmath342 ( ground - state ) and @xmath343 ( local maximum ) . 2 . for @xmath344 , the energies of the extremal points in ascending order are @xmath345 ( ground state ) , @xmath342 ( saddle point ) and @xmath343 ( local maximum ) . 3 . for @xmath139 , the energies of the extremal points in ascending order are @xmath345 ( ground state ) , @xmath346 ( saddle point ) , @xmath342 ( local maximum ) and @xmath343 ( local maximum ) . we derive in this appendix analytic expressions for the sdos , which are determined by the bounds of variables @xmath347 and @xmath283 . we follow closely the appendix a of ref.@xcite . in this reference , the study was limited to the maximal pseudospin case @xmath156 with critical coupling given by @xmath348 . in order to extend the results of that reference to every pseudospin , the following simple substitution has to be made @xmath349 . in the above mentioned appendix , it was demonstrated that , after integration of the bosonic variables , @xmath350 and that this integral gives real values if and only if the following condition is fulfilled @xmath351 where @xmath352 , @xmath353 , and we have used @xmath354 defined in eq.([defgamm ] ) . the previous condition determines the bounds of variables @xmath347 and @xmath283 . to see more clearly these bounds we rewrite the previous condition as @xmath355.\end{aligned}\ ] ] clearly , if @xmath356 the condition is satisfied for every @xmath357 , whereas if @xmath358 the condition is never fulfilled . for @xmath359 the condition is fulfilled for two intervals of @xmath283 , @xmath360 $ ] and @xmath361 $ ] , with @xmath362.\ ] ] therefore , to determine the bounds of the variables @xmath283 and @xmath352 , it is necessary to study the behavior of function @xmath363 in the interval @xmath364 $ ] for different couplings and energies . the different behaviors of function @xmath363 are classified according to three couplings intervals @xmath365 , @xmath366 and @xmath367 . for each of these coupling intervals , different energy regimes can be identified . in figure [ fig:10 ] a graphical summary of the behavior of function @xmath363 is presented . three different coupling were selected representing the three coupling intervals mentioned above . for the first coupling , @xmath368 , the function @xmath363 takes values below to @xmath125 only for @xmath369 . in the energy interval @xmath370 [ blue lines in panel a ) of fig.[fig:10 ] ] , the function is less than @xmath3 ( implying that @xmath283 can take values from @xmath3 to @xmath371 ) only for @xmath372 , where @xmath373 is the largest root of @xmath374 whose roots are given by @xmath375 with @xmath376 defined by @xmath377.\ ] ] for @xmath378 , with @xmath379 defined below , the function @xmath363 takes values in the interval @xmath380 $ ] , consequently the angle @xmath283 is limited to two intervals around @xmath3 and @xmath381 , as explained before eq.([limphi ] ) . the value @xmath379 is the largest root of equation @xmath382 with roots given by @xmath383 for this energy interval , the values @xmath384 are forbidden , and the density of states is given by @xmath385 for @xmath386 [ black lines in panel a ) of fig.[fig:10 ] ] , the function @xmath356 for all @xmath387 $ ] , consequently neither @xmath283 nor @xmath347 are limited ( the whole bloch sphere become available ) . hence , the density of states is @xmath388 in this case , corresponding to panel b ) of fig.[fig:10 ] ( @xmath389 ) , the two energy intervals of the previous case \{@xmath390 and @xmath391 } remain together the corresponding expressions for the density of states . but a new energy interval appears @xmath392 . for this energy interval , the function @xmath363 [ red lines in panel b ) of fig . [ fig:10 ] ] takes values less than @xmath125 and greater than @xmath3 , in the interval @xmath393 , with the roots @xmath394 defined above . consequently , the angle @xmath283 is bounded as explained before eq.([limphi ] ) . hence , the density of states for this energy interval is @xmath395 finally , the behavior of @xmath363 for this case is depicted in panel c ) of fig . [ fig:10 ] ( @xmath396 ) . the energy intervals and corresponding expression for the density of states are similar to the previous cases with the following changes : the interval @xmath397 changes to @xmath398 , and a new intertwined interval emerges , that given by @xmath399 . in this latter interval , the function @xmath363 [ orange lines in panel c ) of fig . [ fig:10 ] ] takes values in the interval @xmath380 $ ] for @xmath400 $ ] and @xmath401 $ ] , where the values of @xmath283 are bounded by eq.([condi ] ) . on the other hand , for @xmath402 $ ] the function @xmath363 is less than @xmath3 and , consequently , there the angle @xmath283 can take values in the whole interval @xmath403 . therefore , for this new energy interval , @xmath404 , the density of states is @xmath405 99 polkovnikiv a , sengupta k , silva a and vengalattore m , 2011 _ rev . phys . _ * 83 * 863 eisert j , friesdorf m and gogolin c , 2015 _ nat . phys . _ * 11 * 124 gogolin c and eisert j , 2016 _ rep . phys . _ * 79 * 056001 wilms j , vidal j , verstraete f and dusuel s , 2012 _ j. stat . _ p01023 caprio m a , cejnar p and iachello f , 2008 _ ann . phys . _ * 323 * 1106 strnsk p , macek m and cejnar , 2014 _ annals of physics _ * 345 * 73 ; strnsk p , macek m , leviatan a and cejnar p , 2015 _ annals of physics _ * 356 * 57 cejnar p , strnsk p and kloc m , 2015 _ phys . _ * 90 * 114015 cejnar p , macek m , heinze s , jolie j and dobes j , 2006 _ j. phys . a _ * 39 * l515 relao a , arias j m , dukelsky j , garca - ramos j e , and prez - fernndez p , 2008 _ phys . . a _ * 78 * 060102 ; prez - fernndez p , relao a , arias j m , dukelsky j , and garca - ramos j e , 2009 ibid . * * 032111 prez - fernndez p , cejnar p , arias j m , dukelsky j , garca - ramos j e , and relao a , 2011 _ phys . rev . a _ * 83 * 033802 santos l f and prez - bernal f , 2015 _ phys . * 92 * 050101(r ) tavis m and cummings f w , 1968 _ phys . * 170 * 379 dicke r h , 1954 _ phys . rev . _ * 93 * 99 baumann k , guerlin c , brennecke f and esslinger t , 2010 _ nature _ * 464 * 1301 nagy d , knya g , szirmai g and domokos p , 2010 _ phys . lett . _ * 104 * 130401 schneble d , torii y , boyd m , streed e w , pritchard d e and ketterle w , 2003 _ science _ * 300 * 475 scheibner m , schmidt t , worschech l , forchel a , bacher g , passow t and hommel d , 2007 _ nat . _ * 3 * 106 blais a , huang r - s , wallraff a , girvin s m and schoelkopf r j , 2004 _ phys . rev . a _ * 69 * 062320 fink j m , bianchetti r , baur m , gppl m , steffen l , filipp s , leek p j , blais a , and wallraff a , 2009 _ phys . rev . _ * 13 * 083601 hepp k and lieb e h , 1973 _ ann . ( n.y . ) * 76 * 360 ; 1973 _ phys . rev . a _ * 8 * ( 5 ) 2517 emary c and brandes t , 2003 _ phys . * 67 * 066203 ; 2003 _ phys . _ * 90 * 044101 prez - fernndez p , relao a , arias j m , cejnar p , dukelsky j and garca - ramos j e , 2011 _ phys . e _ * 87 * 046208 brandes t , 2013 _ phys . e _ * 88 * 032133 bastarrachea - magnani m a , lerma - hernndez s and hirsch j g , 2014 _ phys . a _ * 89 * 032101 ; 2014 _ phys . a _ * 89 * , 032102 bastarrachea - magnani m a , lpez - del - carpio b , lerma - hernndez s and hirsch j g , 2015 _ phys . _ * 90 * 068015 wang y k , and hioe f t , 1973 _ phys . rev . a _ * 7 * ( 3 ) 831 hioe f t , 1973 _ phys . rev . _ a * 8 * 1440 carmichael h j , gardiner c w and walls d f , 1973 _ phys . lett . _ * 46 a * 47 comer duncan g , 1974 _ phys . rev . a _ * 9 * ( 1 ) 418 brankov i g , 1975 _ n. s. theor . _ * 22 * 16 gibberd r w , 1974 _ aust . j. phys . _ * 27 * 241 gilmore r and bowden c m , 1976 _ phys . * 13 * 1898 lee b s , 1976 _ j. phys . a math . gen . _ * 9 * 573 vertogen g and de vries a s , 1974 _ phys . lett . a_. * 48a * 451 orszag m , 1977 _ j. phys . a : math . gen_. * 10 * l21 orszag m , 1977 _ j. phys . a : math . gen _ * 10 * 1995 aparicio - alcalde m , bucher m , emary c and brandes t , 2012 _ phys . e _ * 86 * 012101 gilmore r , 1981 _ catastrophe theory for scientists and engineers_. ( n. y. : dover ) . liberti g and zaffino r l , 2005 _ eur . j. b. _ * 44 * 535 jaworski w , 1985 _ z. phys . _ 59 _ 483 bender c m and orzag s a , 1978 _ advanced mathematical methods for scientists and engineers . _ ( usa : mcgraw - hill ) gutzwiller m c , 1990 _ chaos in classical and quantum mechanics _ ( new york : springer ) rossignoli r and plastino a , 1984 _ phys . c _ * 30 * 1360 prez - fernndez p and relao a ( unpublished ) .

maxwell s equations are considered with transparent boundary conditions , for initial conditions and inhomogeneity having support in a bounded , not necessarily convex three - dimensional domain or in a collection of such domains . the numerical method only involves the interior domain and its boundary . the transparent boundary conditions are imposed via a time - dependent boundary integral operator that is shown to satisfy a coercivity property . the stability of the numerical method relies on this coercivity and on an anti - symmetric structure of the discretized equations that is inherited from a weak first - order formulation of the continuous equations . the method proposed here uses a discontinuous galerkin method and the leapfrog scheme in the interior and is coupled to boundary elements and convolution quadrature on the boundary . the method is explicit in the interior and implicit on the boundary . stability and convergence of the spatial semidiscretization are proven , and with a computationally simple stabilization term , this is also shown for the full discretization .

1605.04086

maxwell s equations on the whole three - dimensional space are considered with initial conditions and inhomogeneity having support in a bounded domain that is not required to be convex ( or in a finite collection of such domains ) . the study of such problems leads to transparent boundary conditions , which yield the restriction of the solution to the domain . such boundary conditions are nonlocal in space and time , for both acoustic wave equations and maxwell s equations . there is a vast literature to tackle this problem in general for wave equations : fast algorithms for exact , nonlocal boundary conditions on a ball @xcite , local absorbing boundary conditions @xcite , perfectly matched layers , which were originally considered for electromagnetism in @xcite , and numerical coupling with time - dependent boundary integral operators @xcite . all the above approaches , except the last one , are inadequate for non - convex domains . the local methods fail because waves may leave and re - enter a non - convex domain . inclusion of a non - convex domain in a larger convex domain is computationally undesirable in situations such as a cavity or an antenna - like structure or a far - spread non - connected collection of small domains . the main objective of the present work is to transfer the programme of @xcite from acoustic wave equations to maxwell s equations : to propose and analyze a provably stable and convergent fully discrete numerical method that couples discretizations in the interior and on the boundary , without requiring convexity of the domain . like abboud _ et al . _ @xcite ( and later also @xcite ) for the acoustic wave equation , we start from a symmetrized weak first - order formulation of maxwell s equations . in the interior this is discretized by a discontinuous galerkin ( dg ) method in space @xcite together with the explicit leapfrog scheme in time @xcite . the boundary integral terms are discretized by standard boundary element methods in space and by convolution quadrature ( cq ) in time @xcite . this yields a coupled method that is explicit in the interior and implicit on the boundary . the choice of a cq time discretization of the boundary integral operators is essential for our analysis , and to a lesser extent also the choice of the leapfrog scheme in the interior . however , our approach is not specific to the chosen space discretizations which could , in particular , be replaced by conformal edge elements @xcite . while the general approach of this paper is clearly based on @xcite , it should be emphasized that the appropriate boundary integral formulation requires a careful study of the time - harmonic maxwell s equation . this is based on @xcite , with special attention to the appropriate trace space on the boundary and to the corresponding duality . due to the analogue of green s formula for maxwell s equations , the duality naturally turns out to be an anti - symmetric pairing . the calderon operator for maxwell s equation , which arises in the boundary integral equation formulation of the transparent boundary conditions , differs from the acoustic case to a large extent , and therefore the study of its coercivity property is an important and nontrivial point . similarly to the acoustic case , the continuous - time and discrete - time coercivity is obtained from the laplace - domain coercivity using an operator - valued version , given in @xcite , of the classical herglotz theorem @xcite . both the second and first order formulation of maxwell s equations are used . the spatial semi - discretization of the symmetrized weak first - order formulation of maxwell s equations has formally the same matrix vector formulation as for the acoustic wave equation studied in @xcite , with the same coercivity property of the calderon operator . because of this structural similarity , the stability results of @xcite , which are shown using the matrix vector setting , remain valid for the maxwell case without any modification . on the other hand , their translation to the functional analytic setting differs to a great extent . therefore further care is required in the consistency analysis . in section [ section : recap helmholtz ] we recapitulate the basic theory for maxwell s equation in the laplace domain . based on buffa and hiptmair @xcite , and further on @xcite , we describe the right boundary space , which allows for a rigorous boundary integral formulation for maxwell s equations . then the boundary integral operators are obtained in a usual way from the single and double layer potentials . in section [ section : calderon ] we prove the crucial technical result of the present work , a coercivity property of the calderon operator for maxwell s equation in the laplace domain . this property translates to the continuous - time maxwell s equations later , in section [ subsection : calderon op for maxwell s eqn ] , via an operator - valued herglotz theorem . in section [ section : boundary int form ] we study the interior exterior coupling of maxwell s equations , resulting in an interior problem coupled to an equation on the boundary with the calderon operator . we derive a first order symmetric weak formulation , which is the maxwell analogue of the formulation of @xcite for the acoustic wave equation . together with the continuous - time version of the coercivity property of the calderon operator , this formulation allows us to derive an energy estimate . later on this analysis is transfered to the semi - discrete and fully discrete settings . section [ section : discretization ] presents the details of the discretization methods : in space we use discontinuous galerkin finite elements with centered fluxes in the domain @xcite , coupled to continuous linear boundary elements on the surface . time discretization is done by the leapfrog scheme in the interior domain , while on the boundary we use convolution quadrature based on the second - order backward differentiation formula . an extra term stabilizes the coupling , just as for the acoustic wave equation @xcite . the matrix vector formulation of the semidiscrete problem has the same anti - symmetric structure and the same coercivity property as for the acoustic wave equation , and therefore the stability results shown in @xcite can be reused here . in sections [ section : semidiscrete results ] and [ section : fully discrete results ] we revise the parts of the results and proofs of @xcite where they differ from the acoustic case , which is mainly in the estimate of the consistency error . finally , we arrive at the convergence error bounds for the semi- and full discretizations . to our knowledge , the proposed numerical discretizations in this paper are the first provably stable and convergent semi- and full discretizations to interior exterior coupling of maxwell s equations . we believe that the presented analysis and the techniques , which we share with @xcite , can be extended further : to other discretization techniques for the domain , such as edge element methods @xcite , higher order discontinuous galerkin methods , and different time discretizations in the domain , together with higher order runge kutta based convolution quadratures on the boundary @xcite . for ease of presentation we consider only constant permeability and permittivity . however , it is only important that the permeability and permittivity are constant in the exterior domain and in a neighbourhood of the boundary . in the interior these coefficients may be space - dependent and discontinuous . in the latter case the equations can be discretized in space with the dg method as described in @xcite . in this paper we focus on the appropriate boundary integral formulation and on the numerical analysis of the proposed numerical methods . numerical experiments are intended to be presented in subsequent work . concerning notation , we use the convention that vectors in @xmath0 are denoted by italic letters ( such as @xmath1 ) , whereas the corresponding boldface letters are used for finite element nodal vectors in @xmath2 , where @xmath3 is the ( large ) number of discretization nodes . hence , any boldface letters appearing in this paper refer to the matrix vector formulation of spatially discretized equations . functions defined in the domain @xmath4 are denoted by letters from the roman alphabet , while functions defined on the boundary @xmath5 are denoted by greek letters . let us consider the _ time - harmonic maxwell s equation _ , obtained as the laplace transform of the second order maxwell s equation ( with constant permeability @xmath6 and permittivity @xmath7 ) : @xmath8 where @xmath9 is the boundary of a bounded piecewise smooth domain ( or a finite collection of such domains ) @xmath10 , not necessarily convex , with exterior normal @xmath11 . we shortly recall some useful concepts and formulas regarding the above problem , based on @xcite and @xcite . for the usual trace we will use the notation @xmath12 . tangential _ and _ magnetic _ traces are defined , respectively , as @xmath13 these traces are also often called _ dirichlet trace _ and _ neumann trace _ , motivated by the analogue of green s formula for maxwell s equations ( for sufficiently regular functions ) : @xmath14 we introduce an important notation , the @xmath15}_{\gamma}= \int_{\gamma}(\operatorname{\gamma}w \times \nu ) \cdot \operatorname{\gamma}v \,{\textrm{d}}\sigma , \ ] ] which appears on the right - hand side of . we note that the relation @xmath16}_{\gamma}= { [ } \operatorname{\gamma}_t w , \operatorname{\gamma}_t v { ] } _ { \gamma}$ ] holds , cf . @xcite . let us now set @xmath17 , which provides @xmath18}_{{\gamma}}.\ ] ] moreover , if @xmath19 satisfies and @xmath20 , then @xmath21}_{{\gamma}}.\ ] ] we collect some results on function spaces , which will play an important role later on . all of the results in the present subsection can be found in section 2 of @xcite . let us start by recalling the usual definition of the sobolev space corresponding to the @xmath22 operator : @xmath23 with corresponding norm @xmath24 . clearly , the above integral relations hold for functions @xmath25 . now we are turning to trace spaces . however , even though @xmath26 is a continuous mapping , @xmath27 is not the right choice for boundary integral operators . as it was emphasized by buffa and hiptmair @xcite : the study of the continuous mapping @xmath28 is _ _ `` actually sufficient for the understanding of homogeneous boundary conditions for fields in the hilbert space context . however , to impose meaningful non - homogeneous boundary conditions or , even more important , to lay the foundations for boundary integral equations we need to identify a proper trace space''__. in the following , we briefly summarize the definition of such a trace space , together with some related results . the hilbert space @xmath29 collects the @xmath30 traces of @xmath31 functions , for @xmath32 , i.e. , @xmath33 . this space is equipped with an inner product such that @xmath34 is continuous . in particular , the space @xmath35 has the dual space @xmath36 , defined with respect to the ( extended ) duality @xmath37}_{{\gamma}}$ ] . then , the above mentioned _ proper trace space _ is given as : @xmath38 with norm @xmath39 the tangential trace satisfies the following analogue of the trace theorem . [ lemma : trace ineq ] the trace operator @xmath40 is continuous and surjective . the following lemma clarifies the role of the anti - symmetric pairing @xmath41}_{{\gamma}}$ ] . [ lemma : hdiv - dual ] the pairing @xmath41}_{{\gamma}}$ ] can be extended to a continuous bilinear form on @xmath42 . with this pairing the space @xmath42 becomes its own dual . the above results clearly point out that a natural choice of trace space is @xmath43}_{{\gamma}}\big)$ ] , which fits perfectly to the analogue of green s formula and to the boundary integral formulation of maxwell s equations . this trace space is appropriate for the analysis of boundary integral operators . on potentials and boundary integral operators we follow buffa and hiptmair @xcite , and we also refer to @xcite . the usual boundary integral potentials for the time - harmonic maxwell s equation @xmath44 are obtained , based on @xcite and @xcite : the ( electric ) _ single layer potential _ is given , for @xmath45 , as @xmath46 while the ( electric ) _ double layer potential _ is given , for @xmath45 , as @xmath47 where the fundamental solution is given , for @xmath48 , as @xmath49 the solution then has the representation @xmath50 where @xmath51\ ! ] } = { [ \![\operatorname{\gamma}_t ( s{^{-1}}\operatorname{curl}u)]\ ! ] } { \qquad \hbox { and } \qquad } \psi = { [ \![\gamma_t u]\!]}.\ ] ] here @xmath52\ ! ] } = \gamma^- v - \gamma^+ v$ ] denotes the jumps in the boundary traces . a further notation is the average of the inner and outer traces on the boundary : @xmath53 . on vectors both operations are acting componentwise . for every @xmath54 and @xmath55 , formula defines @xmath56 . because of the jump relations @xmath57\ ! ] } = & \ \hbox{id } , & { [ \![\gamma_n \circ { { \mathcal d}}(s)]\ ! ] } = & \ 0 , \\ { [ \![\gamma_t \circ { { \mathcal s}}(s)]\ ! ] } = & \ 0 , & { [ \![\gamma_n \circ { { \mathcal d}}(s)]\ ! ] } = & \ \hbox{id},\end{aligned}\ ] ] @xmath58 and @xmath59 are reconstructed from @xmath19 by . let us now define the boundary integral operators . as opposed to the general second order elliptic case , due to additional symmetries of the problem , they reduce to two operators @xmath60 and @xmath61 , see ( * ? ? ? * section 5 ) . they satisfy @xmath62 in ( * ? ? ? * section 5 ) the continuity of these operators was proven , without giving an explicit dependence on @xmath63 . such bounds are crucial in the analysis later , therefore we now show @xmath63-explicit estimates for the boundary integral operators . our result is based on @xcite . [ lemma : int operator s bounds ] for @xmath64 the boundary integral operators @xmath65 are bounded as @xmath66 these estimates can be shown by adapting the arguments of ( * ? ? ? * section 4.2 ) . in particular , by using the anti - symmetric pairing @xmath37}_{{\gamma}}$ ] instead of the usual @xmath67 inner product , the results of ( * ? ? ? * theorem 4.4 ) transfer from @xmath68 to the estimates stated here . @xmath69 furthermore , using the potential representation of the solution , the averages of the traces can be expressed using the operators @xmath60 and @xmath61 in the following way : @xmath70 { \{\!\!\{\gamma_n u\}\!\!\ } } = & \ { \{\!\!\ { \gamma_n { { \mathcal s}}(s){\varphi}\}\!\!\ } } + { \{\!\!\ { \gamma_n { { \mathcal d}}(s)\psi \}\!\!\ } } \\ = & \ k(s){\varphi}+ v(s)\psi . \end{aligned}\ ] ] an important role will be played by the following operator on @xmath71 , to which we refer as a _ calderon operator _ : @xmath72 the extra factor @xmath73 appears unmotivated here , but will turn out to be convenient later . this operator satisfies the following coercivity result , which is the key lemma of this paper . [ lemma : coercivity ] there exists @xmath74 such that the calderon operator satisfies @xmath75}_{{\gamma}}\geq \beta \,{m(s ) } \big ( ( { \varepsilon}\mu){^{-1}}\|s{^{-1}}{\varphi}\|_{{\mathcal{h}_\gamma}}^2 + \|s{^{-1}}\psi\|_{{\mathcal{h}_\gamma}}^2 \big ) \end{aligned}\ ] ] for @xmath76 and for all @xmath77 , with @xmath78 the proof has a structure similar to the proof of the corresponding result for the acoustic helmholtz equation ( * ? ? ? * lemma 3.1 ) , although it now uses a different functional - analytic setting . the structural similarity becomes possible thanks to the anti - symmetric duality pairing that replaces the symmetric duality pairing of the acoustic case . for given @xmath77 , we define @xmath79 by the representation formula . we can then express @xmath58 and @xmath59 in terms of @xmath19 by . we note that yields @xmath80 now , using the properties of the anti - symmetric pairing @xmath37}_{{\gamma}}$ ] ( acting componentwise on @xmath81 ) , the analogue of green s formula and using the definition of the traces , we obtain @xmath82}_{{\gamma}}=&\ \ { [ } { [ \![\operatorname{\gamma}_n u]\ ! ] } , { \{\!\!\{\operatorname{\gamma}_t u\}\!\!\ } } { ] } _ { { \gamma}}+ { [ } { [ \![\operatorname{\gamma}_t u]\ ! ] } , -{\{\!\!\{\operatorname{\gamma}_n u\}\!\!\ } } { ] } _ { { \gamma}}\\ = & \ { [ } \operatorname{\gamma}_n^- u , \operatorname{\gamma}_t^- u { ] } _ { { \gamma}}- { [ } \operatorname{\gamma}_n^+ u , \operatorname{\gamma}_t^+ u { ] } _ { { \gamma}}\\ = & \ s \ \big(\|s{^{-1}}\operatorname{curl}u\|_{l^2({\mathbb{r}}^3\setminus{\gamma})}^2 + { \varepsilon}\mu \|u\|_{l^2({\mathbb{r}}^3\setminus{\gamma})}^2 \big).\end{aligned}\ ] ] we further obtain @xmath83\ ! ] } \big\|_{{\mathcal{h}_\gamma}}^2 \\ & \leq \ c \big ( \|\operatorname{curl}u\|_{l^2({\mathbb{r}}^3\setminus{\gamma})^3}^2 + \|u\|_{l^2({\mathbb{r}}^3\setminus{\gamma})^3}^2 \big)\\ & = \ c |s|^2 \big ( \|s{^{-1}}\operatorname{curl}u\|_{l^2({\mathbb{r}}^3\setminus{\gamma})^3}^2 + |s|^{-2 } \|u\|_{l^2({\mathbb{r}}^3\setminus{\gamma})^3}^2 \big)\\ & \leq \ c |s|^2 \max\{1,|s|^{-2}({\varepsilon}\mu){^{-1}}\ } \big(\|s{^{-1}}\operatorname{curl}u\|_{l^2({\mathbb{r}}^3\setminus{\gamma})^3}^2 + { \varepsilon}\mu \|u\|_{l^2({\mathbb{r}}^3\setminus{\gamma})^3}^2 \big ) , \end{aligned}\ ] ] and for @xmath84 we use the fact that @xmath85 : @xmath86\ ! ] } \big\|_{{\mathcal{h}_\gamma}}^2 \\ & \leq \ c ( { \varepsilon}\mu){^{-1}}\big(\|s{^{-1}}\operatorname{curl}\operatorname{curl}u\|_{l^2({\mathbb{r}}^3\setminus{\gamma})^3}^2 + \|s{^{-1}}\operatorname{curl}u\|_{l^2({\mathbb{r}}^3\setminus{\gamma})^3}^2 \big ) \\ & = \ c \big ( { \varepsilon}\mu \|s u\|_{l^2({\mathbb{r}}^3\setminus{\gamma})^3}^2 + ( { \varepsilon}\mu){^{-1}}\|s{^{-1}}\operatorname{curl}u\|_{l^2({\mathbb{r}}^3\setminus{\gamma})^3}^2 \big ) \\ & \leq \ c |s|^2 \max\{1 , |s|^{-2}({\varepsilon}\mu){^{-1}}\ } \big(\|s{^{-1}}\operatorname{curl}u\|_{l^2({\mathbb{r}}^3\setminus{\gamma})^3}^2 + { \varepsilon}\mu\|u\|_{l^2({\mathbb{r}}^3\setminus{\gamma})^3}^2 \big)\end{aligned}\ ] ] where , for the first inequalities in both estimates , we used the trace inequality of lemma [ lemma : trace ineq ] . extraction of factors and dividing through completes the proof . let us consider the first order formulation of maxwell s equations , in the following form : @xmath87 with appropriate initial and boundary conditions . if the initial conditions satisfy the last two equations , then they hold for all times , see @xcite , therefore these conditions are assumed to hold . the permeability and permittivity is denoted by @xmath6 and @xmath7 , respectively , and they are assumed to be positive constants , while @xmath88 denotes the electric current density . using the relation @xmath89 , the above equation can be written as the second order problem @xmath90 with @xmath91 . setting @xmath92 , applying laplace transformation , and writing @xmath19 instead of @xmath93 , we obtain the time - harmonic version . we recall an operator - valued continuous - time herglotz theorem from ( * ? ? ? * section 2.2 ) , which is crucial for transferring the coercivity result of lemma [ lemma : coercivity ] from the maxwell s equation in the laplace domain to the time - dependent maxwell s equation . we describe the result in an abstract hilbert space setting . let @xmath94 be a complex hilbert space , with dual @xmath95 and anti - duality @xmath96 . let @xmath97 and @xmath98 be both analytic families of bounded linear operators for @xmath64 , satisfying the uniform bounds : @xmath99 for any integer @xmath100 , we define the integral kernel @xmath101 for a function @xmath102,{{\mathcal h}})$ ] with vanishing initial data , @xmath103 , we let @xmath104 that is , @xmath105 is the distributional convolution of the inverse laplace transform of @xmath106 with @xmath107 . the operator - valued version of the classical herglotz theorem from @xcite yields the following result . [ lemma : herglotz ] in the above setting , the following two statements are equivalent : 1 . @xmath108 , for any @xmath109 , @xmath110 ; + 2 . @xmath111 , for all @xmath102,{{\mathcal h}})$ ] , with @xmath103 , and for all @xmath112 . consider the second order formulation of maxwell s equations in three dimensions : @xmath113 , \\ { e}(x,0 ) = & \ { e}_0 & { \qquad \hbox { in } } & { \mathbb{r}}^3 , \\ { \partial}_t { e}(x,0 ) = & \ { h}_0 & { \qquad \hbox { in } } & { \mathbb{r}}^3.\end{aligned}\ ] ] let @xmath114 be a bounded lipschitz domain , with boundary @xmath9 , and further assume that the initial values and @xmath115 are supported within @xmath116 . we rewrite this problem as an interior problem over @xmath116 : @xmath117 , \\ { e}^-(x,0 ) = & \ { e}_0 & { \qquad \hbox { in } } & { \omega } , \\ { \partial}_t { e}^-(x,0 ) = & \ { h}_0 & { \qquad \hbox { in } } & { \omega},\end{aligned}\ ] ] and as an exterior problem over @xmath118 : @xmath119 , \\ { e}^+(x,0 ) = & \ 0 & { \qquad \hbox { in } } & { \omega}^+ , \\ { \partial}_t { e}^+(x,0 ) = & \ 0 & { \qquad \hbox { in } } & { \omega}^+.\end{aligned}\ ] ] the two problems are _ coupled _ by the transmission conditions : @xmath120 using the temporal convolution operators of section [ subsec : herglotz ] , the solution of the exterior problem is given as @xmath121 with boundary densities @xmath122 which satisfy the equation @xmath123 here @xmath124 is the temporal convolution operator with the distribution whose laplace transform is the calderon operator @xmath106 defined in . from now on , we use maxwell s equations in their first order formulation on the interior domain @xmath116 ( and we omit the omnipresent superscript @xmath125 ) : @xmath126 , \ ] ] with the coupling through the calderon operator as @xmath127 where @xmath128 and @xmath129 . in addition , by @xmath130 we obtain @xmath131 where we also used . hence , @xmath132 and @xmath133 in the same way as in ( * ? ? ? * lemma 4.1 ) for the acoustic wave equation , the coercivity of the calderon operator @xmath106 for the time - harmonic maxwell s equation as given by lemma [ lemma : coercivity ] together with the operator - valued continuous - time herglotz theorem as stated in lemma 2.3 yields coercivity of the time - dependent calderon operator @xmath134 . [ lemma : time - cont coercivity ] with the constant @xmath74 from lemma [ lemma : coercivity ] we have that @xmath135}_{{\gamma}}{\textrm{d}}t \\ & \ \geq \beta c_t \int_0^t e^{-2t / t } \big ( ( { \varepsilon}\mu){^{-1}}\|{\partial}_t{^{-1}}{\varphi}(\cdot , t)\|_{{\mathcal{h}_\gamma}}^2 + \|{\partial}_t{^{-1}}\psi(\cdot , t)\|_{{\mathcal{h}_\gamma}}^2 \big ) { \textrm{d}}t \end{aligned}\ ] ] for arbitrary @xmath136 and for all @xmath137,{\mathcal{h}_\gamma})$ ] and all @xmath138,{\mathcal{h}_\gamma})$ ] with @xmath139 and @xmath140 , and with constant @xmath141 . a gronwall argument then yields the following energy estimate ; see ( * ? ? ? * lemma 4.2 ) . [ lemma : general energy - like est ] let the functions @xmath142 \to [ 0,\infty)$ ] , @xmath143 \to { \mathbb{r}}$ ] , and @xmath144 , { \mathcal{h}_\gamma})$ ] with @xmath139 , @xmath140 , be such that for all @xmath145 $ ] @xmath146}_{{\gamma}}= { { \mathcal f}}(t).\ ] ] then , with @xmath141 , @xmath147 analogously to @xcite , a symmetric weak form of is obtained on using @xmath148}_{{\gamma}},\ ] ] and using for the boundary term . here @xmath149 denotes the standard @xmath150 inner product . the coupled weak problem then reads : find @xmath151 and @xmath77 such that @xmath152}_{{\gamma}}+ ( { j},w ) \\ & \ \hphantom{({\varepsilon}{\partial}_t{e},w ) } = { { \textstyle \frac12}}(\operatorname{curl}{h } , w ) + { { \textstyle \frac12}}({h } , \operatorname{curl}w ) - { { \textstyle \frac12}}{[}\mu{^{-1}}{\varphi } , \operatorname{\gamma}_t w { ] } _ { { \gamma}}+ ( { j},w ) , \\ & \ ( \mu { \partial}_t{h},z ) = -{{\textstyle \frac12}}(\operatorname{curl}{e } , z ) - { { \textstyle \frac12}}({e } , \operatorname{curl}z ) + { { \textstyle \frac12}}{[}\operatorname{\gamma}_t { e } , \operatorname{\gamma}_t z { ] } _ { { \gamma}}\\ & \ \hphantom{(\mu { \partial}_t{h},z ) } = -{{\textstyle \frac12}}(\operatorname{curl}{e } , z ) - { { \textstyle \frac12}}({e } , \operatorname{curl}z ) - { { \textstyle \frac12}}{[}\psi , \operatorname{\gamma}_t z { ] } _ { { \gamma}},\\ & \ \biggl{[}{\binom{\xi}{\eta } } , b({\partial}_t ) { { \binom{{\varphi}}{\psi}}}\biggr{]}_{{\gamma}}= { { \textstyle \frac12}}\big ( { [ } \xi,\mu{^{-1}}\operatorname{\gamma}_t { e}{]}_{{\gamma}}+ { [ } \eta,\operatorname{\gamma}_t { h}{]}_{{\gamma}}\big ) \end{aligned}\ ] ] hold for arbitrary @xmath153 , and @xmath154 . while this weak formulation is apparently non - standard for maxwell s equations , we will see that it is extremely useful , in the same way as the analogous formulation proved to be for the acoustic case in @xcite . testing with @xmath155 , @xmath156 and @xmath157 , @xmath158 in , by using we obtain @xmath159}_{{\gamma}}+ ( { j},{e } ) , \\ & \ ( \mu { \partial}_t{h},{h } ) = -{{\textstyle \frac12}}(\operatorname{curl}{e } , { h } ) - { { \textstyle \frac12}}({e } , \operatorname{curl}{h } ) - { { \textstyle \frac12}}{[}\psi , \operatorname{\gamma}_t { h}{]}_{{\gamma}},\\ & \ \biggl{[}{\binom{{\varphi}}{\psi } } , b({\partial}_t ) { { \binom{{\varphi}}{\psi}}}\biggr{]}_{{\gamma}}= { { \textstyle \frac12}}\big ( { [ } { \varphi},\mu{^{-1}}\operatorname{\gamma}_t { e}{]}_{{\gamma}}+ { [ } \psi,\operatorname{\gamma}_t { h}{]}_{{\gamma}}\big ) , \end{aligned}\ ] ] and summing up the three equations yield @xmath160}_{{\gamma}}= ( { j},{e } ) .\ ] ] for @xmath161 , the coercivity of the continuous - time calderon operator , as stated in lemmas [ lemma : time - cont coercivity ] and [ lemma : general energy - like est ] , yields that the electromagnetic energy @xmath162 satisfies the energy estimate ( with @xmath163 ) for arbitrary @xmath136 . for the spatial discretization we use , as an example , the central flux discontinuous galerkin ( dg ) discretization from @xcite ( see also @xcite ) in the interior and continuous linear boundary elements on the surface . we triangulate the bounded polyhedral domain @xmath116 by simplicial triangulations @xmath164 , where @xmath165 denotes the maximal element diameter . for our theoretical results we consider a quasi - uniform and contact - regular family of such triangulations with @xmath166 ; see e.g. @xcite for these notions . we adopt the following notation from ( * ? ? ? * section 2.3 ) : the faces @xmath167 of @xmath164 , decomposed into boundary and interior faces : @xmath168 . the normal of an interior face @xmath169 is denoted by @xmath170 . it is kept fixed and is the outward normal of one of the two neighbouring mesh elements . we denote by @xmath171 that neighbouring element into which @xmath170 is directed . the outer faces of @xmath164 are used as the triangulation of the boundary @xmath5 . the dg space of vector valued functions , which are elementwise linear in each component , is defined as @xmath172 the boundary element space @xmath173 is taken as @xmath174 the corresponding nodal basis functions are denoted by @xmath175 and @xmath176 , respectively . jumps and averages over faces @xmath177 are denoted analogously as for trace operators on @xmath9 , see section [ subsection : boundary integral op ] : @xmath178\!]}_f = \operatorname{\gamma}_f^-w - \operatorname{\gamma}_f^+w { \qquad \hbox { and } \qquad } { \{\!\!\{w\}\!\!\}}_f = { { \textstyle \frac12}}(\operatorname{\gamma}_f^-w + \operatorname{\gamma}_f^+w),\ ] ] where @xmath179 is the usual trace onto the face @xmath180 . we often omit the subscript as it will always be clear from the context . the discrete @xmath22 operator with centered fluxes was presented in ( * ? ? ? * section 2.3 ) : @xmath181\ ! ] } , { \{\!\!\{w_h\}\!\!\ } } { ] } _ { f } . \end{aligned}\ ] ] by the arguments of the proof of lemma 2.2 in @xcite , we obtain that the discrete curl operator satisfies the discrete version of green s formula , @xmath182}_{\gamma}.\ ] ] the @xmath183 operator is well defined on @xmath184 , with the broken sobolev space @xmath185 which is a hilbert space with natural norm and seminorm @xmath186 and @xmath187 , respectively . using the above discrete @xmath22 operator , the semidiscrete problem reads as follows : find @xmath188 and @xmath189 such that for all @xmath190 and @xmath191 , [ eq : semidiscrete problem ] @xmath192}_{{\gamma}}+ ( { j},w_h ) , \\[1 mm ] \label{eq : dg - bem } & \ ( \mu { \partial}_t{h}_h , z_h ) = -{{\textstyle \frac12}}(\operatorname{curl}_h { e}_h , z_h ) - { { \textstyle \frac12}}({e}_h , \operatorname{curl}_h z_h ) - { { \textstyle \frac12}}{[}\psi_h , \operatorname{\gamma}_t z_h { ] } _ { { \gamma}},\\[2 mm ] \nonumber & \ \biggl{[}{\binom{\xi_h}{\eta_h } } , b({\partial}_t ) { \binom{{\varphi}_h}{\psi_h } } \biggr{]}_{{\gamma}}= { { \textstyle \frac12}}\big ( { [ } \xi_h,\mu{^{-1}}\operatorname{\gamma}_t { e}_h{]}_{{\gamma}}+ { [ } \eta_h,\operatorname{\gamma}_t { h}_h { ] } _ { { \gamma}}\big ) . \end{aligned}\ ] ] all expressions are to be interpreted in a piecewise sense if necessary . we collect the nodal values of the semidiscrete electric and magnetic field into the vectors @xmath193 , and similarly the nodal vectors of the boundary densities are denoted by @xmath194 and @xmath195 . upright boldface capitals always denote matrices of the discretization . we obtain the following coupled system of ordinary differential equations and integral equations for the nodal values : @xmath196 the matrix @xmath197 denotes the symmetric positive definite mass matrix , while the other matrices are defined as @xmath198 which happens to be a symmetric matrix , and @xmath199}_{{\gamma } } , \qquad { { \mathbf c}}_0=\mu{^{-1}}{{\mathbf c}}_1 .\ ] ] the matrix @xmath200 is given by @xmath201 where the blocks have entries @xmath202}_{{\gamma}}{\qquad \hbox { and } \qquad } { { \mathbf k}}(s)|_{kk ' } = { { \textstyle \frac12}}{[}b_{k'}^{\gamma } , k(s ) b_k^{\gamma}{]}_{{\gamma}}.\ ] ] for this matrix we have the following coercivity estimate . [ lemma : coercivity - discrete ] with @xmath74 and @xmath203 from lemma [ lemma : coercivity ] , the matrix @xmath200 satisfies @xmath204 for @xmath76 and for all @xmath205 , where the mass matrix @xmath206 , for the inner product @xmath207 corresponding to the norm on @xmath42 , is defined by @xmath208 . the result follows from lemma [ lemma : coercivity ] on noting that for the vectors @xmath209 and @xmath210 and the corresponding boundary functions in @xmath211 @xmath212 we have @xmath213}_{{\gamma}}\ ] ] and @xmath214 . let us emphasize the following observation : _ the above matrix vector formulation is formally the same as the one for the acoustic wave equation in ( * ? ? ? * section 5.1 ) , with the same coercivity estimate for the boundary operator @xmath215 by lemmas [ lemma : coercivity - discrete ] and [ lemma : time - cont coercivity ] . as an important consequence , the stability results proven in @xcite hold for the present case as well . _ the choice of a dg method in the interior and of continuous boundary elements for the spatial discretizations is not necessary for our analysis . other space discretization methods , for instance the ones going back to raviart and thomas @xcite , ndlec @xcite , and many others , detailed in the excellent survey article @xcite , or locally divergence - free methods such as @xcite , could also be used as long as they yield a matrix vector formulation of the form and a coercivity estimate as in lemma [ lemma : coercivity - discrete ] . following ( * ? ? ? * section 2.3 ) we give a short recap of convolution quadrature and introduce some notation . for more details see @xcite and @xcite . convolution quadrature ( cq ) discretizes the convolution @xmath216 by the discrete convolution @xmath217 where the weights @xmath218 are defined as the coefficients of @xmath219 in the present paper we choose @xmath220 which corresponds to the second - order backward difference formula . from @xcite , it is known that the method is of order two , @xmath221 for functions @xmath107 that are sufficiently smooth including their extension by @xmath222 to negative values of @xmath223 . an important property of this discretization is that it preserves the coercivity of the continuous - time convolution in the time discretization . we have the following result . [ @xcite , lemma 2.3 ] [ lemma : cq coercivity ] in the setting of lemma [ lemma : herglotz ] condition ( i ) implies , for @xmath224 small enough and with @xmath225 , @xmath226 for any function @xmath227 with finite support . combining lemma [ lemma : coercivity ] and lemma [ lemma : cq coercivity ] yields the following coercivity property of the cq time - discretization of the time - dependent calderon operator considered in lemma [ lemma : time - cont coercivity ] . [ lemma : time - discrete coercivity ] in the situation of lemma [ lemma : time - cont coercivity ] , we have for @xmath228 and @xmath229 that @xmath230}_{{\gamma}}\\ & \ \geq \beta c_t { { \mathit{\delta}t}}\sum_{n=0}^n e^{-2t_n / t } \big ( ( { \varepsilon}\mu){^{-1}}\|({\partial}_t^{{\mathit{\delta}t}}){^{-1}}{\varphi}(\cdot , t_n)\|_{{\mathcal{h}_\gamma}}^2 + \|({\partial}_t^{{\mathit{\delta}t}}){^{-1}}\psi(\cdot , t_n)\|_{{\mathcal{h}_\gamma}}^2 \big ) \end{aligned}\ ] ] for all sequences @xmath231 and @xmath232 in @xmath42 , with @xmath233 for a @xmath234 ( which depends only on @xmath235 and tends to 1 as @xmath235 goes to zero ) . similarly to @xcite , we use the leapfrog or strmer verlet scheme ( see , e.g. , @xcite ) in the interior : @xmath236 where the last substep of the previous step and the first substep can be combined to a step from @xmath237 to @xmath238 when no output at @xmath239 is needed : @xmath240 this is coupled with convolution quadrature on the boundary@xmath241^{n+1/2 } = { \binom{{{\mathbf c}}_0^t \bar{\boldsymbol{e}}^{n+1/2}}{{{\mathbf c}}_1^t { \boldsymbol{h}}^{n+1/2 } } } + { \binom{0}{- \alpha { { \mathit{\delta}t}}^2 \mu{^{-1}}{{\mathbf c}}_1^t { { \mathbf m}}{^{-1}}{{\mathbf c}}_1 \dot{{\boldsymbol \psi}}^{n+1/2 } } } , \ ] ] where the operation @xmath242 is averaging in time and @xmath243 . the second term on the right - hand side is a stabilizing term , with a parameter @xmath244 . the role of this extra term becomes clear from the proof of the stability result for the acoustic wave equation ( * ? ? ? * lemma 8.1 ) , which applies to the maxwell case as well . like for the acoustic case , the choice @xmath245 yields a stable scheme under the cfl condition @xmath246 . up to a factor 2 this is the cfl condition for the leapfrog scheme for the equation with natural boundary conditions . in each time step , a linear system with the matrix @xmath247 needs to be solved for @xmath248 and @xmath249 , where @xmath250 and @xmath251 by the coercivity lemma [ lemma : coercivity ] , @xmath252 is positive definite . moreover , @xmath253 is symmetric positive definite . using that the obtained discrete system is of the same form and with the same coercivity property as for the acoustic wave equation , the stability results carry over from section 6 of @xcite . only minor technical modifications are needed , such as using the appropriate energy and norms . the only point where the analysis of the semidiscrete problem deviates from the acoustic case is the consistency error estimates , which require special care . we consider a system with additional inhomogeneities @xmath254\to l^2(\omega)^3 $ ] and @xmath255\to { \mathcal{h}_\gamma}$ ] , which will later be obtained as the system of error equations with the defects of an interpolation of the exact solution . the coupled system @xmath256}_{{\gamma}}+ ( j_h , w_h ) , \\[1 mm ] \label{eq : dg - bem - stab } & \ ( \mu { \partial}_t{h}_h , z_h ) = -{{\textstyle \frac12}}(\operatorname{curl}_h { e}_h , z_h ) - { { \textstyle \frac12}}({e}_h , \operatorname{curl}_h z_h ) - { { \textstyle \frac12}}{[}\psi_h , \operatorname{\gamma}_t z_h { ] } _ { { \gamma}}+(g_h , w_h ) , \\[2 mm ] \nonumber & \ \biggl{[}{\binom{\xi_h}{\eta_h } } , b({\partial}_t ) { \binom{{\varphi}_h}{\psi_h } } \biggr{]}_{{\gamma}}= { { \textstyle \frac12}}\big ( { [ } \xi_h,\mu{^{-1}}\operatorname{\gamma}_t { e}_h{]}_{{\gamma}}+ { [ } \eta_h,\operatorname{\gamma}_t { h}_h { ] } _ { { \gamma}}\big ) \\ & \ \hphantom{\biggl{[}{\binom{\xi_h}{\eta_h } } , b({\partial}_t ) { \binom{{\varphi}_h}{\psi_h } } \biggr{]}_{{\gamma}}= } + ( \xi_h,\rho_h)_{\gamma}+(\eta_h,\sigma_h)_{\gamma}\end{aligned}\ ] ] where @xmath207 denotes the inner product on @xmath42 , has the matrix - vector formulation @xmath257 where @xmath206 is the boundary mass matrix with entries @xmath258 . the solution of this system can be bounded in terms of @xmath259 by the stability results proven in lemma 6.16.3 in @xcite . we immediately translate the stability lemmas of @xcite into the functional analytic setting . the energy estimate of lemma 6.1 of @xcite becomes the following . [ lemma : semidiscr stability - energy estimate ] the semidiscrete energy @xmath260 satisfies the bound , for @xmath261 , @xmath262 provided that @xmath263 and @xmath264 . the estimates for the boundary functions of lemma 6.3 of @xcite now translate into the following . [ lemma : semidiscr stability - boundary functions ] for @xmath261 , the boundary functions are bounded as @xmath265 provided that @xmath266 , @xmath267 , @xmath263 and @xmath264 . we consider the projection of functions on @xmath4 and @xmath5 to continuous piecewise linear finite element functions by interpolation : let @xmath268 denote the operator of piecewise linear ( with respect to the triangulation @xmath164 ) and continuous interpolation in @xmath4 , and let @xmath269 denote the operator of piecewise linear continuous interpolation on @xmath5 . since the normal vector @xmath11 is constant on every face of @xmath5 , we then have @xmath270 which implies that @xmath269 maps @xmath271 into @xmath42 . moreover , this yields the very useful relation @xmath272 as is seen by noting that @xmath273 it is because of that we work in the following with interpolation operators rather than orthogonal projections . we recall the standard results for the interpolation errors . [ lemma : interpolation error ] there exists a constant @xmath274 , independent of @xmath165 , such that for all @xmath275 , @xmath276 the following interpolation error estimate is a standard result for boundary element approximations , see @xcite . [ lemma : interpolation error - boundary ] there exists a constant @xmath274 , independent of @xmath165 , such that for all @xmath277 , @xmath278 we remark that for piecewise smooth boundaries just the piecewise @xmath279 regularity is needed . for the boundary functions we have the following interpolation error bounds . [ lemma : interpolation error - calderon ] there exists a constant @xmath280 , increasing at most polynomially in @xmath223 and independent of @xmath165 , such that for any @xmath261 @xmath281 for all @xmath282 , { \mathcal{h}_\gamma}\cap h^{3/2}({\gamma})^3)$ ] with @xmath283 and @xmath284 . the proof is similar to that of lemma 7.2 in @xcite : first we bound the action of the blocks of @xmath106 , then we use plancherel s formula to bound the action of the convolution operator @xmath124 . by the boundedness of the boundary integral operators lemma [ lemma : int operator s bounds ] , for @xmath285 we obtain @xmath286 then , lemma [ lemma : interpolation error - boundary ] yields @xmath287 a similar estimate holds for the blocks @xmath288 , and so we obtain @xmath289 using plancherel s formula and causality then yields the stated bound . @xmath69 we study the defects ( or consistency errors ) obtained on inserting the interpolated solution @xmath290 into the semidiscrete variational formulation . these defects are defined by @xmath291}_{{\gamma}}\\ & ( d_h^h , z_h ) = ( \mu { \partial}_t{i_h}{h},z_h ) - { { \textstyle \frac12}}(\operatorname{curl}_h { i_h}{e},z_h ) - { { \textstyle \frac12 } } ( { i_h}{e},\operatorname{curl}_h z_h ) \\ & \hphantom{(d_h^h , z_h ) = } + { { \textstyle \frac12}}{[}\pi_h \psi,\operatorname{\gamma}_t z_h{]}_{{\gamma}}\\ & ( \xi_h , d_h^\psi)_{\gamma}+ ( \eta_h , d_h^{\varphi})_{{\gamma}}= \biggl{[}{\binom{\xi_h}{\eta_h } } , b({\partial}_t ) { \binom{\pi_h{\varphi}}{\pi_h\psi } } \biggr{]}_{{\gamma}}\\ & \hphantom{(\xi_h , d_h^\psi)_{\gamma}+ ( \eta_h , d_h^{\varphi})_{{\gamma}}= } - { { \textstyle \frac12}}\big ( { [ } \xi_h,\mu{^{-1}}\operatorname{\gamma}_t { i_h}{e}{]}_{{\gamma}}+ { [ } \eta_h,\operatorname{\gamma}_t { i_h}{h}{]}_{{\gamma}}\big)\end{aligned}\ ] ] for all @xmath292 and @xmath293 . these defects are bounded as follows . [ lemma : defect bounds ] if the solution of maxwell s equations is sufficiently smooth , then the defects satisfy the first - order bounds , for @xmath261 , @xmath294 the constant @xmath280 grows only polynomially with @xmath223 . we begin with the defect @xmath295 . we have for @xmath296 , @xmath297}_{{\gamma}},\end{aligned}\ ] ] where we used the discrete green s formula . since @xmath298 , the boundary term vanishes by the relation . we further note that @xmath299 and @xmath300 because @xmath301 is a continuous function and so has no jumps on inner faces . the exact solution satisfies maxwell s equation and hence @xmath302 subtracting the two equations therefore yields @xmath303 with the interpolation error bounds of lemma [ lemma : interpolation error ] the right - hand terms are estimated as @xmath304 times the @xmath305 norm of @xmath306 . we thus conclude that @xmath307 similarly we estimate the defect @xmath308 for the magnetic equation . for the boundary defects @xmath309 we have for all @xmath310 , using the boundary equation , @xmath311}_{{\gamma}}\ ] ] where @xmath312 are given by @xmath313 which is bounded by @xmath304 in the @xmath314 norm by lemmas [ lemma : interpolation error - calderon ] and [ lemma : interpolation error ] . it then follows that also the defects @xmath309 , which are interpolated by @xmath173 , are bounded in the same way , using lemma [ lemma : hdiv - dual ] : @xmath315 if we differentiate twice with respect to time before estimating and commute interpolations and time derivatives , this yields the stated bound for the boundary defects . @xmath69 [ theorem : semidiscrete error bound ] assume that the initial data @xmath316 and @xmath317 have their support in @xmath116 . let the initial values of the semidiscrete problem be chosen as the interpolations of the initial values : @xmath318 and @xmath319 . if the solution of maxwell s equations is sufficiently smooth , then the error of the dg bem semidiscretization satisfies , for @xmath261 , the first - order error bound @xmath320 where the constant @xmath280 grows at most polynomially in @xmath223 . we insert the interpolated solution @xmath290 into the semidiscrete variational formulation and apply the stability lemmas , lemmas [ lemma : semidiscr stability - energy estimate ] and [ lemma : semidiscr stability - boundary functions ] , to the error equations that have the defects in the role of the inhomogeneities . we then use the defect bounds of lemma [ lemma : defect bounds ] to arrive at a first - order error bound for @xmath321 . the interpolation error estimates of lemma [ lemma : interpolation error ] and [ lemma : interpolation error - calderon ] together with the triangle inequality then complete the proof . similarly to the semidiscrete case , the stability analysis of the full discretization only depends on the formulation of the fully discrete problem and , which again coincides with the acoustic case in form and relevant properties . hence , the analysis of the full discretization can be carried over directly from ( * ? ? ? * section 8) . the original results are again translated into the current functional analytic setting . we show stability results under the cfl condition @xmath322 the fully discrete electric and magnetic field satisfies the inequality below . [ lemma : fully discr stability - interior ] under the cfl condition and for a stabilization parameter @xmath323 , the discrete energy @xmath324 is bounded , at @xmath325 , by @xmath326 where @xmath327 is independent of @xmath165 , @xmath328 and @xmath329 . using @xmath330 , the above result also yields a bound on @xmath331 . for the boundary densities we have the following fully discrete estimate . [ lemma : fully discr stability - boundary ] under the cfl condition and for a stabilization parameter @xmath323 , the discrete boundary functions are bounded , at @xmath325 , by @xmath332 where @xmath327 is independent of @xmath165 , @xmath328 and @xmath329 . the following convergence estimate for the full discretization is then shown in the same way as in the proof of theorem 9.1 of @xcite , using the consistency errors of the spatial discretization given in section [ subsection : consistency ] , using known error bounds of the leapfrog scheme and convolution quadratures , and applying lemmas [ lemma : fully discr stability - interior ] and [ lemma : fully discr stability - boundary ] . [ theorem : fully discrete error bound ] assume that the initial conditions @xmath316 and @xmath317 , and the inhomogeneity @xmath88 have their supports in @xmath116 . let the initial values of the semidiscrete problem be chosen as the interpolations of the initial values : @xmath333 and @xmath334 . if the solution of maxwell s equations is sufficiently smooth , and under the cfl condition and with a stabilization parameter @xmath323 , the error of the dg bem and leapfrog convolution quadrature discretization and is bounded , at @xmath325 , by @xmath335 where the constant @xmath280 grows at most polynomially in @xmath223 . we have given a stability and error analysis of semi- and full discretizations of maxwell s equations in an interior non - convex domain coupled with time - domain boundary integral equations for transparent boundary conditions . a key result for the analysis of this problem is the coercivity estimate of the calderon operator proved in lemma [ lemma : coercivity ] , which is preserved under trace - space conforming boundary discretizations and translates from the laplace domain to the continuous - time domain by the operator - valued version of herglotz theorem as restated in lemma [ lemma : herglotz ] and to the convolution quadrature time discretization by lemma [ lemma : time - discrete coercivity ] . another important aspect is that the symmetrized weak formulation , as first proposed in @xcite for the acoustic wave equation , is preserved under space discretization to yield a finite - dimensional system of the form . in this paper the space discretization is exemplified by a dg discretization in the interior and continuous boundary elements . other interior discretizations that are commonly used for maxwell s equations , such as edge elements , could equally be used as long as they lead to a matrix formulation . similarly , other trace - space conforming boundary elements such as raviart thomas elements could be used , since they also preserve the coercivity of lemma [ lemma : coercivity - discrete ] . once the matrix formulation has the structure with the coercivity of lemma [ lemma : coercivity - discrete ] , the analysis in @xcite shows stability of the spatial semi - discretization and of the full discretization with a stabilized leapfrog method . together with estimates for the consistency error , which we derive in sections [ subsection : interpolation error bounds ] and [ subsection : consistency ] in an exemplary way for the particular space discretization considered , we then obtain error bounds for the semi - discretization . moreover , using known error bounds for the consistency error of the leapfrog method and of the convolution quadrature time discretization on the boundary , we finally obtain error bounds of the full discretization . we claim no originality on the constituents of the discretization of maxwell s equation in space and time , in the interior and on the boundary . the novelty of this paper is the stability and error analysis of their coupling . it is remarkable that , in spite of the fundamentally different functional - analytic framework , the stability analysis extends directly from the acoustic to the maxwell case . this becomes possible because we show here that the coercivity and the matrix formulation of the discretization are of the same type for both maxwell and the acoustic case . on the other hand , the analysis of the consistency errors depends strongly on the functional - analytic setting and is different for discretizations of maxwell s equations and the acoustic wave equation . we thank two anonymous referees for their helpful comments . we are grateful for the helpful discussions on spatial discretizations with ralf hiptmair ( eth zrich ) during a birs workshop ( 16w5071 ) in banff . this work was supported by the deutsche forschungsgemeinschaft ( dfg ) through sfb 1173 . approximation of integral equations by finite elements . error analysis . in r. dautray and j .- l . lions , editors , _ mathematical analysis and numerical methods for science and technology _ , volume 4 , chapter xiii , pages 359370 . springer , 1990 .

we consider the statistical deconvolution problem where one observes replications from the model , where is the unobserved random signal of interest and is an independent random error with distribution . under weak assumptions on the decay of the fourier transform of we derive upper bounds for the finite - sample sup - norm risk of wavelet deconvolution density estimators for the density of , where is assumed to be bounded . we then derive lower bounds for the minimax sup - norm risk over besov balls in this estimation problem and show that wavelet deconvolution density estimators attain these bounds . we further show that linear estimators adapt to the unknown smoothness of if the fourier transform of decays exponentially and that a corresponding result holds true for the hard thresholding wavelet estimator if decays polynomially . we also analyze the case where is a `` supersmooth''/analytic density . we finally show how our results and recent techniques from rademacher processes can be applied to construct global confidence bands for the density . .

1103.1489

consider the statistical deconvolution model @xmath9 where @xmath2 is a real - valued random variable with unknown probability density @xmath10 and @xmath3 is an error term independent of @xmath2 that is distributed according to the probability measure @xmath4 on @xmath11 . the law @xmath12 of @xmath13 equals the convolution @xmath14 and we denote its density by @xmath15 . let @xmath16 be i.i.d . replications of @xmath13 in the model ( [ deconvolution - model ] ) and denote by @xmath17 the associated empirical measure . the _ deconvolution problem _ is about recovering the unknown density @xmath7 from the noisy observations @xmath18 . it has been extensively studied : we refer to carroll and hall @xcite , stefanski @xcite , stefanski and carroll @xcite , fan @xcite , diggle and hall @xcite , goldenshluger @xcite , pensky and vidakovic @xcite , delaigle and gijbels @xcite , hesse and meister @xcite , johnstone et al . @xcite , johnstone and raimondo @xcite , bissantz et al . @xcite , bissantz and holzmann @xcite , meister @xcite , butucea and tsybakov @xcite and pensky and sapatinas @xcite , also to the monograph meister @xcite , as well as to cavalier @xcite for a survey of the literature on general inverse problems in statistics , of which deconvolution is a special case . one key lesson from the aforementioned literature is that a lower bound on the regularity of the signal @xmath3 is necessary to be able to estimate @xmath7 with reasonable accuracy . this lower bound is often quantified by a lower bound on the decay of the fourier transform @xmath19 $ ] of @xmath4 and fourier inversion techniques are applied to construct estimators for @xmath7 . most of the literature on this problem ( with some notable exceptions , to be discussed below ) deals with the @xmath20-theory , that is , involves the loss function @xmath21 and is often restricted to the case of periodic and hence compactly supported @xmath7 . these restrictions are theoretically convenient , in particular since fourier analysis - based methods can be used without too much difficulty , using the parseval plancherel isometry . however , a sound understanding of the local behavior of deconvolution estimators seems to be of significant statistical importance . in particular a theory that could deal with sup - norm loss @xmath22 could be used in the construction of confidence bands for the object @xmath7 of statistical interest . a fortiori it is not at all clear whether the intuitions from @xmath20-theory carry over to pointwise and uniform loss functions in generality , bearing in mind that @xmath20-convergence properties of fourier series can give a completely inadequate picture of their pointwise or uniform behaviour . in the present article , we use methods from empirical process theory to derive finite - sample sup - norm risk bounds for deconvolution density estimators based on fourier inversion with meyer ( or similar band - limited ) wavelets . these estimators were studied in pensky and vidakovic @xcite and johnstone et al . @xcite , and have since been successfully used in inverse problems . our results hold under minimal assumptions on the density @xmath7 and the distribution @xmath4 : we require @xmath7 to be bounded , which is unavoidable if one considers sup - norm loss , and we assume that the fourier transform of @xmath4 is nonzero on the intervals of support of the meyer wavelet , which is necessary to define any estimator based on fourier inversion and which also makes @xmath7 identifiable . our risk bounds imply rates of convergence for the deconvolution density estimator that are optimal in global sup - norm loss , without any moment or support restrictions whatsoever , both in the severely ill - posed case ( where linear methods suffice ) , as well as in the moderately ill - posed case ( where we propose a suitable thresholding method ) . to be more precise , given the law @xmath4 of the error term and a density @xmath7 belonging to some besov body @xmath23 with unknown @xmath24 , we devise purely data - driven estimators @xmath25 such that , for every @xmath26 , @xmath27 where @xmath28 is the minimax rate of convergence in sup - norm loss over the given besov body and given the error law @xmath4 . we also obtain a result of this kind for the case where @xmath7 is `` supersmooth , '' that is , has an exponentially decaying fourier transform . to the best of our knowledge , the minimax lower bounds derived in this article are also new . we should note that the main delicate mathematical point in this work is to link the @xmath20-based procedure of fourier inversion to a pointwise , or even uniform , control of the random fluctuations of the centered linear density estimator ; this problem is already implicit in the conditions on @xmath19 $ ] and @xmath7 imposed by stefanski and caroll @xcite , fan @xcite and goldenshluger @xcite , who considered pointwise loss . even stronger assumptions were imposed in the nice paper bissantz et al . @xcite , wherein the limiting ( extremal - type ) distribution of the uniform deviations over compact sets of certain kernel deconvolution density estimators for @xmath7 is derived this is the only result that we are aware of in the literature on deconvolution estimation that deals with sup - norm loss in the moderately ill - posed case ( stefanski @xcite deals only with the simpler severely ill - posed case ) . our empirical process approach gives results under minimal conditions and also yields the relevant concentration inequalities that allow for a satisfactory treatment of adaptation , which the results in bissantz et al . @xcite do not address . we should note that applying empirical process tools in this setting is not at all straightforward : the usual approach would be to show that certain kernels are of bounded variation and thus the associated sets of translates and dilates are of vapnik chervonenkis type ( e.g. , nolan and pollard @xcite , einmahl and mason @xcite , gin and guillou @xcite ) , but this does not seem viable in the deconvolution problem , due to the fact that the bounded variation norm does not possess a nice fourier - analytical characterization . we can , however , solve this problem by combining recent results on vc properties of functions of quadratic variation in gin and nickl @xcite with littlewood paley theory and the fact that wavelet bases are compatible with both the @xmath20- and @xmath29-structure simultaneously ; see lemma [ entropy ] for this key result . our results can be used to construct confidence bands in the deconvolution problem and we discuss this in some detail below , as well as relations to work in @xcite . we suggest a new approach to nonparametric confidence bands based on rademacher symmetrization , in a similar vein as in recent work of koltchinskii @xcite . while these confidence bands may be conservative , they allow for an explicit finite - sample analysis under minimal assumptions . let us finally remark that this article also contains new results for the standard density estimation problem ( where @xmath4 equals dirac measure @xmath30 at @xmath31 ) . in this field , our results contribute in several respects : first , vapnik chervonenkis properties of wavelet projection kernels have thus far only been derived for daubechies wavelets @xcite and battle lemari wavelets @xcite , and the present article achieves the same for wavelets with compactly supported fourier transform ( e.g. , meyer wavelets ) . furthermore , our main adaptation result , theorem [ result - threshold ] , is completely free of any moment conditions and thus shows , as may have been suspected , that the moment conditions imposed in theorem 8 in @xcite are not necessary . finally , the confidence bands we suggest can also be used for regular wavelet density estimators and we are not aware of any other results on global confidence bands in density estimation , except for the rather technical ones in @xcite . we start with some preliminary definitions and facts . for any lebesgue integrable function @xmath32 the fourier transform @xmath33 $ ] of @xmath34 is defined as @xmath33(t ) = \int_{\mathbb{r}}h(x)e^{-itx}\,dx$ ] , @xmath35 and we use the natural extension of @xmath36 to @xmath37 . we further denote by @xmath38 the inverse fourier transform so that @xmath39 for @xmath40 . the fourier transform of the density @xmath15 from ( [ deconvolution - model ] ) is then given by @xmath41(t ) = f[f](t)f[\varphi](t)\ ] ] for every @xmath42 . another standard property of the fourier transform we shall frequently use is its scaling property : for @xmath43 and @xmath44 the function @xmath45 has fourier transform @xmath46(t ) = \alpha^{-1}f[h](\alpha^{-1}t)$ ] . let @xmath47 and @xmath48 be , respectively , a scaling function and the associated wavelet function of a multiresolution analysis . we refer to @xcite for the basic theory of wavelets that we shall use freely in this article . the dilated and translated scaling and wavelet functions at resolution level @xmath49 and scale position @xmath50 are defined as @xmath51 , @xmath52 , @xmath53 . now , denote by @xmath54 the inner product in the hilbert space @xmath55 . the density @xmath7 can be formally expanded into its wavelet series @xmath56 where the coefficients are given by @xmath57 , @xmath58 , . as is well known , the regularity properties of a function @xmath7 can be measured by the decay of their wavelet coefficients . we define besov spaces as follows . [ besov ] let @xmath59 , @xmath24 or let @xmath60 and @xmath61 . let @xmath47 and @xmath48 be the meyer scaling function and mother wavelet , respectively ( see , e.g. , section 2 of @xcite for a definition ) . the besov space @xmath62 is defined as the set of functions @xmath63 where @xmath64 are the norms of the sequence spaces @xmath65 , and with the usual modification in the case @xmath66 . moreover , for any @xmath67 , the besov ball of radius @xmath68 is defined as @xmath69 . before we construct explicit estimators for the density @xmath7 of @xmath2 in the deconvolution model ( [ deconvolution - model ] ) , we derive a result that gives a benchmark for the best performance of _ any _ estimator @xmath70 . more precisely , we derive lower bounds for the minimax rate of convergence of @xmath71 in sup - norm loss , uniformly over besov bodies of densities @xmath7 under various assumptions on the error law @xmath4 . we will subsequently show that these lower bounds can be attained by certain wavelet - based estimators and are thus optimal . to this end , define the minimax @xmath72-risk over the hlder class @xmath73 as @xmath74 where the infimum is taken over all possible estimators @xmath75 . note that an estimator in the deconvolution problem means any measurable function of a sample @xmath76 from density @xmath14 that takes values in the space of bounded functions on @xmath11 . we shall make the following assumption on @xmath19 $ ] to establish the lower bounds . [ lb - cond ] there exist constants @xmath77 , @xmath78 and @xmath79 such that @xmath19(t)$ ] is differentiable for every @xmath80 satisfying @xmath81 and @xmath82(t)|\leq c(1+t^2)^{-w/2}e^{-c_0|t|^{\alpha}},\ ] ] as well as @xmath83)'(t)|\leq c'(1+t^2)^{-w'/2}e^{-c_0|t|^{\alpha}}.\ ] ] this condition is weaker than the standard ones employed in deconvolution problems to establish lower bounds ( cf . @xcite ) , where an additional condition is imposed on the second derivative of @xmath19 $ ] . it covers the usual candidates for @xmath5 including the case @xmath84 which corresponds to classical density estimation ( @xmath85 ) . the following theorem distinguishes the `` moderately ill - posed '' case , where @xmath19 $ ] decays only polynomially , from the `` severely ill - posed '' case , where @xmath19 $ ] decays exponentially fast , and shows that the optimal rates of estimation in the sup - norm depend both on the smoothness of @xmath7 and the decay of @xmath19 $ ] . [ theo - minmax ] let condition be satisfied . then , for any @xmath86 there exists a constant @xmath87 such that for every @xmath88 we have @xmath89 one may be interested in replacing the hlder class @xmath23 by a more general besov body , @xmath90 , @xmath91 of densities . it follows from the proof of theorem [ theo - minmax ] that the minimax rate over @xmath90 equals the one for @xmath23 with @xmath92 and the sobolev embedding @xmath93 will imply that our upper risk bounds derived in the following sections attain this rate . we thus restrict ourselves to @xmath23 without loss of generality . recall the model ( [ deconvolution - model ] ) . we now show , following @xcite , how one can estimate @xmath7 from a sample of @xmath12 by `` deconvolving '' @xmath12 or , rather , a suitable approximation of it , on a wavelet basis @xmath94 that satisfies the following condition . [ wav ] assume @xmath95 for every @xmath96 , and for some @xmath97 we have @xmath98)\subset[-a , a]$ ] , as well as @xmath99)\subset[-a ,- a]\setminus[-a',a']$ ] . assume , further , that @xmath100 this condition is satisfied for meyer wavelets with @xmath101 and @xmath102 ( these choices are not optimal , but feasible)see , for instance , section 2 in @xcite but other band - limited wavelet bases are also admissible . if @xmath103 then the functions @xmath104 , @xmath105 , are the kernels of the orthogonal projections of @xmath55 onto the closed subspaces @xmath106 spanned by @xmath107 . we write , for @xmath108 , @xmath109 possibly real - valued , @xmath110 where the second equality holds pointwise , in view of ( [ abs ] ) . suppose the fourier transform of the error law @xmath4 satisfies @xmath111| > 0 $ ] on @xmath98(2^{-j}(\cdot)))$ ] . we then have , from plancherel s theorem , that @xmath112(2^{-j}t)}f[f](t)\,dt\nonumber \\[-8pt]\\[-8pt ] & = & \sum_k\phi(2^jx - k)\frac{1}{2\pi}\int_\mathbb r\overline{f[\phi_{0k}](2^{-j}t)}f[g](t)(f[\varphi](t))^{-1}\,dt\nonumber \\ & = & 2^j\sum_k\phi(2^jx - k)\int_\mathbb r\tilde\phi_{jk}(y)g(y)\,dy=\int_\mathbb r k_j^*(x , y)g(y)\,dy,\nonumber\end{aligned}\ ] ] where the ( nonsymmetric ) kernel @xmath113 is given by @xmath114 with @xmath115(2^{-j}\cdot)}{\overline{2^jf[\varphi]}}\biggr](x ) = \phi_{0k}(2^j\cdot)\ast f^{-1}\biggl[\frac{1_{[-2^ja,2^ja]}}{\overline{f[\varphi]}}\biggr](x).\ ] ] we should note that young s inequality for convolutions implies , for fixed @xmath49 , that @xmath116 , and then also @xmath117 , which justifies the above operations . since we have a sample @xmath118 from the density @xmath15 , the identity ( [ dec ] ) suggests a natural estimator of @xmath7 , namely the wavelet deconvolution density estimator @xmath119 we start with some results for the uniform deviations @xmath121 where the inequality follows from ( [ abs ] ) . this suggests to study the empirical process indexed by the class of functions @xmath122 . in fact , some further scaling depending on the error distribution @xmath4 will be useful to obtain a class with constant envelope . the rather intricate fourier - analytical definition of @xmath123 in ( [ tilde ] ) makes it difficult to apply standard results from empirical process theory . what is needed is that @xmath124 be a vapnik chervonenkis ( vc - type ) class of functions . in the classical density estimation case ( where @xmath19 = 1 $ ] ) , this follows from results in nolan and pollard @xcite for translates of a fixed function of bounded variation . we could , however , not control the bounded variation norm of @xmath123 for general @xmath4 in a way that would be useful , mainly because the bounded variation norm does not interact well with fourier transforms . recent results by gin and nickl @xcite show that the bounded variation condition in nolan and pollard @xcite can be replaced by @xmath125-variation for general @xmath126 , and the case @xmath127 , which corresponds to `` quadratic variation , '' can be linked in a more efficient way to fourier analysis by using littlewood paley theory . the following key lemma shows that @xmath124 , suitably normalized , is indeed a vc - type class of functions , under minimal conditions on @xmath19 $ ] . denote by @xmath128 the @xmath129-covering numbers of a class of functions @xmath124 with respect to the @xmath130-distance . [ entropy ] suppose that @xmath131 satisfy condition and that @xmath111(t)|>0 $ ] on @xmath132 $ ] . define @xmath133}|f[\varphi](t)|\ ] ] ( which exists and is positive for every @xmath49 since @xmath4 is a probability measure ) . then the class @xmath134 , @xmath109 , is uniformly bounded by the constant @xmath135 and satisfies , for every @xmath136 , @xmath137 for finite positive constants @xmath138 depending only on @xmath131 , and where the supremum extends over all probability measures @xmath139 on @xmath11 . combining this lemma with moment bounds for empirical processes indexed by vc - type classes of functions in @xcite , as well as with talagrand s @xcite inequality , we obtain the following result . [ suprat ] suppose that @xmath131 satisfy condition , that @xmath111(t)|>0 $ ] on @xmath132 $ ] , let @xmath140 be as in and define @xmath141 . let @xmath142 be the deconvolution wavelet density estimator from and assume that @xmath2 has a bounded density @xmath143 . then there exists a constant @xmath144 , depending only on @xmath145 , such that for every @xmath146 , every @xmath109 and @xmath147 , @xmath148 where @xmath149 . in addition , there exists a constant @xmath150 depending only on @xmath131 , such that for every @xmath151 and @xmath152 @xmath153\\[-8pt ] & & \qquad\le e^{-(1+u)j'}.\nonumber\end{aligned}\ ] ] the constant @xmath154 is unspecified here , although it could be computed explicitly . obtaining realistic constants is an intricate matter , but one can use symmetrization techniques to circumvent this problem ; see proposition [ rad ] below . the techniques from the previous section allow us to establish similar uniform estimates for the deviations of the empirical wavelet deconvolution coefficients @xmath155 from their means . such results are particularly interesting for nonlinear thresholding procedures that we shall study below . we have , for @xmath48 satisfying condition [ wav ] , @xmath156(2^{-l}\cdot)}}{f[\varphi]}(t)f[g](t)\,dt \\ & = & 2^{l/2}\int_{{\mathbb{r}}}f^{-1}\biggl[2^{-l}\frac{f[\psi_{0k}](2^{-l}\cdot)}{\overline{f[\varphi]}}\biggr](x)g(x)\,dx = : 2^{l/2}\int_{{\mathbb{r}}}\tilde\psi_{lk}(x)g(x)\,dx.\end{aligned}\ ] ] a natural unbiased estimator of @xmath157 is therefore @xmath158 and the object of interest in this subsection is the random variable @xmath159 . we should note that for wavelets satisfying condition [ wav ] ( e.g. , meyer wavelets ) , and even if @xmath15 has compact support , the last supremum is over an infinite set , so empirical process techniques are particularly useful . lemma [ entropy ] and proposition [ suprat ] have the following analogs for @xmath160 . [ entropy1 ] suppose that @xmath131 satisfy condition , that @xmath111(t)|>0 $ ] on @xmath161 $ ] and let @xmath162 be as in . then the class @xmath163 , @xmath164 is uniformly bounded by a fixed constant @xmath135 and satisfies , for every @xmath136 , @xmath165 for constants @xmath166 depending only on @xmath131 . [ suprat - beta ] suppose that @xmath94 satisfy condition , that @xmath111(t)|>0 $ ] on @xmath161 $ ] , let @xmath162 be as in and define @xmath167 . assume that @xmath2 has a bounded density @xmath168 . then , for every @xmath146 , for every @xmath169 and @xmath126 , we have @xmath170 where @xmath171 depends only on @xmath172 and where @xmath173 is as in proposition [ suprat ] . in addition , there exists a constant @xmath174 depending only on @xmath175 such that for every @xmath169 and @xmath176 , @xmath177 we now show how the risk bounds from the previous section imply optimal rates of convergence for densities @xmath178 in the deconvolution problem , under the standard decay conditions on @xmath19 $ ] from the inverse problem literature . we first consider the case where the error law @xmath4 decays exponentially fast . in this `` severely ill - posed '' case , one can find a universal choice of @xmath49 for which the linear estimator attains the exact minimax rate , even without having to know the value @xmath179 . [ uni - band ] suppose that @xmath94 satisfy condition and assume that@xmath111(t)| \ge ce^{-c_0|t|^\alpha}$ ] for every @xmath180 and some @xmath181 . let @xmath182 be the estimator defined in , where @xmath183 for some @xmath184 satisfying @xmath185 . then there exists a constant @xmath186 , depending only on @xmath187 , such that for every @xmath188 , we have @xmath189 we now turn to the case where @xmath19 $ ] decays polynomially , the so - called `` moderately ill - posed '' case . here , the linear estimator @xmath6 is only minimax optimal if one knows the value of @xmath179 . [ mod - ill - theo ] suppose that @xmath131 satisfy condition and assume that @xmath111(t)|\ge c(1+|t|^{2})^{-w/2}$ ] for every @xmath190 and some @xmath191 , @xmath192 . let @xmath193 be the estimator defined in with @xmath194 satisfying @xmath195 . then there exists a constant @xmath196 , depending only on @xmath197 , such that for every @xmath188 , we have @xmath198 the question arises as to whether we can achieve this rate of convergence without having to know the value of @xmath179 in our choice of @xmath199 so that we can adapt to the unknown smoothness @xmath179 of @xmath7 . this can be done using the wavelet thresholding deconvolution estimator proposed in johnstone et al . @xcite in the periodic setting , defined as follows : for @xmath200 positive integers , to be specified below , the hard thresholding estimator equals @xmath201 where @xmath202 was introduced in section [ linsup ] . the threshold @xmath203 is chosen such that @xmath204 where @xmath205 , with @xmath173 from proposition [ suprat ] and @xmath206 a `` large enough '' constant that depends only on @xmath207 . if @xmath173 is unknown , then it can be replaced by an estimate , as in @xcite . [ result - threshold ] suppose that @xmath131 satisfy condition . suppose that @xmath4 is such that @xmath111(t)|\geq c(1+|t|^{2})^{-w/2}$ ] for every @xmath190 and some @xmath191 , @xmath192 . let @xmath208 be the thresholded estimator in with @xmath209 we then have , for every @xmath210 and every @xmath24 , that @xmath211 where @xmath212 depends only on @xmath213 . in the last sections , we established the minimax rate of estimation of a density in @xmath214 for the sup - norm error , both in the moderately and severely ill - posed cases , and constructed estimators that attain this rate . it was pointed out in @xcite for the @xmath215-error that the linear and thresholded estimators attain faster rates of convergence if we consider classes of supersmooth densities instead of the usual besov spaces . in this section , we investigate this phenomenon for the sup - norm error . we show that the minimax rate of convergence for the sup - norm is the same as that obtained for the @xmath215-error up to an additional @xmath216 factor and that wavelet estimators can attain this rate . for simplicity , and to highlight the main ideas , we only consider the nonadaptive case . assume that @xmath7 belongs to the class of supersmooth densities , @xmath217(t)|^{2}\exp(2\tilde{c}_0|t|^{s})\,dt\le 2\pi l\biggr\},\ ] ] where @xmath218 . in the moderately ill - posed case , we have the following result . [ supsm ] let @xmath131 satisfy condition . assume that @xmath219 for some @xmath220 and that @xmath111(t)|\ge c(1+|t|^2)^{-w/2}$ ] for every @xmath221 and some @xmath191 , @xmath192 . let @xmath193 be the estimator defined in with @xmath194 satisfying @xmath222 then there exists a constant @xmath223 depending only on @xmath224 such that for every @xmath225 we have @xmath226 the rates we obtained for the sup - norm error are similar to those obtained by @xcite and @xcite for the @xmath215-error , up to the presence of the additional factor @xmath216 . this additional factor can be heuristically explained by the presence of the quantity @xmath227 in the deviation term @xmath228 derived in proposition [ suprat ] . the next theorem implies that this @xmath216 factor is indeed necessary . [ minm2 ] fix @xmath229 and @xmath230 . assume that @xmath4 satisfies condition with @xmath231 and @xmath232 . there then exists a positive constant @xmath233 such that @xmath234 we can also obtain a faster rate of convergence in the severely ill - posed case for supersmooth densities , balancing the bias bound from proposition [ smooth - bias ] below with the variance bound from proposition [ suprat ] above . we can then obtain similar results as in @xcite , with additional logarithmic terms in the rate of convergence , due to the fact that we consider sup - norm loss instead of @xmath20-loss . one of the main statistical challenges in the nonparametric deconvolution problem is the construction of confidence bands for @xmath7 ( cf . @xcite ) . in @xcite , the exact uniform ( over compact subsets of @xmath11 ) limit distribution of certain linear kernel - based deconvolution estimators for @xmath7 is derived , assuming that @xmath7 satisfies @xmath235(u)||u|^r\,du<\infty$ ] for @xmath236 and that @xmath15 is once differentiable with bounded derivative , and if the fourier transform of the error variable decays exactly like a polynomial , that is , @xmath111(t)|\asymp c|t|^{-w}$ ] for some @xmath191 , @xmath192 . if the underlying smoothness @xmath237 of @xmath7 is known , then these results can be used to construct asymptotic confidence bands for @xmath7 that shrink at certain rates of convergence . we suggest here an alternative approach to confidence bands in the nonparametric deconvolution problem . instead of extreme value theory , we use concentration inequalities and rademacher processes . this allows for almost assumption - free results and has the advantage that the confidence band can be shown to be valid on the whole real line and for every sample size @xmath0 . on the downside , these bands are likely to be too conservative in the limit . one fundamental problem of using concentration inequalities ( as in proposition [ suprat ] ) in practice is that often , no reasonable values for the leading constant @xmath154 are available . to circumvent this problem , we use here an idea that goes back to koltchinskii @xcite and bartlett , boucheron and lugosi @xcite ; see also gin and nickl @xcite , where this approach was introduced in density estimation . define a rademacher process and the associated supremum , @xmath238 with @xmath239 an i.i.d . rademacher sequence , independent of the @xmath240 s ( and defined on a large product probability space ) . @xmath241 can be computed in practice by first simulating @xmath0 i.i.d . random signs , applying these signs to the summands @xmath242 of the wavelet deconvolution density estimator ( [ linear - estimator ] ) and maximizing the resulting function . similarly , one can consider @xmath243 , the expectation of @xmath244 with respect to the rademacher variables only , which is a stochastically more stable quantity . we shall use the fact that this is the supremum of a centered process which can be shown to concentrate around @xmath245 . to describe the concentration property , recall @xmath140 from ( [ del ] ) and define the random variable @xmath246 where @xmath247 , @xmath248 and @xmath249 as in ( [ abs ] ) . if @xmath250 is unknown , it can be replaced by @xmath251 in practice so that @xmath252 is completely data - driven . we start with a confidence band @xmath253 $ ] for the mean @xmath254 of @xmath6 . [ rad ] let @xmath142 be the estimator from and suppose that @xmath111|>0 $ ] on @xmath132 $ ] . assume that @xmath2 has a bounded density @xmath255 . we then have , for every @xmath146 , every @xmath256 and every @xmath257 that @xmath258 moreover , the band @xmath259 has expected diameter @xmath260 for every @xmath261 , every @xmath262 , every @xmath263 and some constant @xmath154 depending only on @xmath264 . proposition [ rad ] still holds true when @xmath244 is replaced by @xmath265 , the expectation of @xmath244 with respect to the rademacher variables only . ( this follows from combining the proof of proposition [ rad ] with the arguments in the proof of proposition 2 in @xcite . ) we did not try to optimize the constants in the choice of @xmath252 and they are likely to be suboptimal , as they depend on the constants in the lower - deviation version of talagrand s inequality , where sharp constants are not yet known . a `` practical '' choice may be to replace the @xmath266 in front of @xmath241 by @xmath267 and to ignore the third `` poissonian '' term in ( [ sigr ] ) . we again emphasize that we simply need @xmath111(t)|$ ] to be bounded from below on the fixed interval @xmath132 $ ] for our results to hold and we do not need any support or moment assumptions on @xmath7 . in particular , this nonasymptotic result can even be used in principle when @xmath19 $ ] equals zero eventually , by choosing @xmath49 small enough . if @xmath268 , with @xmath179 known , then the last proposition can be readily applied for the construction of confidence bands @xmath269 for the unknown density @xmath7 using undersmoothing ( just as in @xcite ) and these bands can be shown to shrink at the optimal rate of convergence depending on the smoothness of @xmath7 . we do not detail this here , nor do we address the more difficult problem of _ adaptive _ confidence bands : using proposition [ rad ] , such results can be obtained in the same way as in the case of density estimation considered in @xcite . instead , and for sake of illustration , let us construct a nonasymptotic confidence band in the supersmooth case @xmath270 , @xmath271 known , with moderately ill - posed error distribution . [ band ] let @xmath272 and @xmath199 be as in corollary . let @xmath273 be as in above and define the confidence band @xmath274,\qquad x\in\mathbb r,\ ] ] where @xmath275 is any positive real number . then , for every @xmath261 and every @xmath262 , @xmath276 where [ @xmath277 , as in proposition ] @xmath278 satisfies @xmath279 as @xmath280 . moreover , if @xmath281 is the maximal diameter of @xmath282 then @xmath283 where @xmath154 depends on @xmath284 . since @xmath285 this confidence band has asymptotic coverage for @xmath286 arbitrary , but more is true : @xmath287 equals zero from some @xmath0 onward and one can , in principle , even obtain coverage for every fixed sample size @xmath0 by choosing @xmath275 in dependence of @xmath68 ( and of the constants that define @xmath252 ) . our proof adapts to the present situation standard lower bound techniques as in @xcite . we recall that the kullback leibler divergence between two distributions @xmath12 and @xmath139 is defined by @xmath288 to establish lower bounds for the minimax risk ( [ minmax ] ) , we use the following lemma ( see theorem 2.5 on page 99 of @xcite)actually , an adaptation of it to the deconvolution problem at hand . [ lemlb ] let @xmath289 be a metric on @xmath23 . let @xmath290 be a sequence of positive real numbers and let @xmath291 be a finite set of probability densities such that @xmath292 and @xmath293 , @xmath294 . further , let @xmath4 be a fixed probability measure and let @xmath295 be the product probability measure corresponding to a sample of size @xmath0 from the law @xmath14 , @xmath296 , and assume that the kl divergences satisfy , for every @xmath296 and some @xmath297 , @xmath298 then , @xmath299 where @xmath300 denotes the infimum over all estimators based on a sample of size @xmath0 from the density @xmath14 and where @xmath301 is a constant that depends only on @xmath302 . we use this lemma to prove theorem [ theo - minmax ] . let @xmath48 be the meyer wavelet . fix @xmath303 and let @xmath256 be arbitrary ( to be chosen later ) . define the set of functions @xmath304 as follows : consider the standard cauchy density @xmath305 , set @xmath306 for @xmath307 and for any @xmath308 , @xmath309 where @xmath310 for some integer @xmath311 specified below . we show that the constants @xmath312 can be chosen such that @xmath313 is a density on @xmath11 and , in fact , belongs to @xmath23 for every @xmath314 and every integer @xmath315 . clearly , @xmath313 integrates to 1 since @xmath48 is orthogonal on constants . we next prove @xmath316 for all @xmath317 and suitable @xmath318 . first , we have @xmath319 for @xmath320 large enough and depending only on @xmath321 , in view of @xmath322(u)=e^{-\eta|u|}$ ] , definition [ besov ] , @xmath323(u)|\le 2^{-l/2}\|\psi\|_1 e^{-|2^la'|\eta}$ ] with @xmath102 and a similar estimate for @xmath324 . thus , we have , for @xmath325 @xmath326 having chosen @xmath327 , we can choose @xmath328 suitably small but positive and depending on @xmath327 and @xmath48 so that @xmath329 on @xmath11 for any @xmath317 . this is easily established by using the fact that the meyer wavelet decays faster at infinity than any polynomial [ i.e. , the estimate @xmath330 for every @xmath331 and every @xmath332 , whereas @xmath333 decays at infinity like @xmath334 . to proceed with the proof , we set @xmath335 . we first prove the separation property in sup - norm for the @xmath313 s . for any distinct @xmath336 , we have @xmath337 by definition of the meyer wavelet , we have , for any @xmath338 , @xmath339 for some @xmath340 . by the decay property of the meyer wavelets mentioned above , there exists a numerical constant @xmath311 , large enough but finite , such that for any @xmath341 satisfying @xmath342 , we have @xmath343 . thus , we have , for any @xmath338 , @xmath344 we now check the second condition of lemma [ lemlb ] . let @xmath18 be an i.i.d . sample with distribution @xmath345 admitting the density @xmath346 with respect to the lebesgue measure on @xmath347 . fubini s theorem and the fact that @xmath48 is orthogonal on constants give , for @xmath348 , that @xmath349 . thus , by definition of the kullback leibler divergence and the inequality @xmath350 for @xmath351 , we obtain , for any @xmath314 , that @xmath352\\[-8pt ] & \le & n\gamma_{j}\int_{\mathbb r}(\psi_{jk_m}\ast\varphi)(y)\biggl(1+\gamma_j\frac{\psi_{jk_m}\ast\varphi}{f_0\ast\varphi}(y)\biggr)\,dy\nonumber \\ & \le & n\gamma_{j}^{2}\int_{\mathbb r}\frac{(\psi_{jk_m}\ast\varphi)^{2}}{f_0\ast\varphi}(y)\,dy.\nonumber\end{aligned}\ ] ] to proceed , we observe that @xmath353 being cauchy implies that @xmath354 for some @xmath301 and every @xmath355 . this is obviously true for @xmath356 in any compact set @xmath357 $ ] , and for @xmath358 , it follows from @xmath359 ^ 2}\,d\varphi(x ) = \frac{\eta}{\pi},\ ] ] in view of fatou s lemma . consequently , we have @xmath360 let us first consider the quantity @xmath361 . plancherel s theorem gives @xmath362(t)|^{2}|f[\varphi](t)|^{2}\,dt\nonumber \\[-8pt]\\[-8pt ] & \le & c_3 2^{-j}\|\psi\|_{1}^{2}\int_{\operatorname{supp}(f[\psi_{jk_m}])}(1+t^{2})^{-w}e^{-2c_0|t|^{\alpha}}\,dt\nonumber\end{aligned}\ ] ] for some constants @xmath363 depending only on @xmath364 . for the quantity @xmath365 , we obtain similarly , using in addition the spectral representation of the differential operator , that @xmath366(t)f[\varphi](t))'|^{2}\,dt\nonumber \\ & & \qquad= c_2\int_{\mathbb r}\bigl|\bigl(f[\psi_{jk_m}]'(t)f[\varphi](t)+f[\psi_{jk_m}](t)f[\varphi]'(t)\bigr)\bigr|^{2}\,dt \\ & & \qquad= c_2\int_{\mathbb r}\bigl|\bigl(2^{-3j/2}f[\psi_{0k_m}]'(2^{-j}t)f[\varphi](t)+f[\psi_{jk_m}](t)f[\varphi]'(t)\bigr)\bigr|^{2}\,dt\nonumber \\ & & \qquad\le 2c_4 2^{-3j}\biggl(\int_{{\mathbb{r}}}|x\psi(x)|\,dx\biggr)^{2}\int_{\operatorname{supp}(f[\psi_{jk_m}])}(1+t^{2})^{-w}e^{-2c_0|t|^{\alpha}}\,dt\nonumber \\ & & \qquad\quad{}+ 2c_42^{-j}\|\psi\|_{1}^{2}\int_{\operatorname{supp}(f[\psi_{jk_m}])}(1+t^{2})^{-w'}e^{-2c_0|t|^{\alpha}}\,dt,\nonumber\end{aligned}\ ] ] where @xmath367 depends only on @xmath368 . combining ( [ interm - minmax-1])([minmax - interm3 ] ) and the explicit formula for the support of the meyer wavelet , we can bound @xmath369 by @xmath370 where @xmath371 depends only on @xmath372 . it remains to estimate the size of these integrals and select @xmath49 appropriately and we distinguish the moderately and severely ill - posed cases . in the moderately ill - posed case ( @xmath373 , @xmath374 ) , we have @xmath375 for some constant @xmath376 independent of @xmath377 . taking @xmath378 and @xmath379 small enough ( independent of @xmath0 and @xmath49 ) in the definition of @xmath380 gives @xmath381 , where we recall that @xmath382 . the separation rate @xmath290 for this choice of @xmath199 becomes , for any @xmath383 distinct , @xmath384 for some constant @xmath385 independent of @xmath0 . this proves theorem [ theo - minmax ] for the moderately ill - posed case . for the severely ill - posed case ( @xmath386 ) , we similarly obtain that @xmath387 with @xmath388 and constants @xmath389 , @xmath390 independent of @xmath377 . taking @xmath391 with @xmath392 large enough gives @xmath393 where @xmath394 , @xmath395 are nonnegative constants independent of @xmath396 . for this choice of @xmath397 the separation rates @xmath290 become , for any @xmath383 distinct , @xmath398 where @xmath399 is independent of @xmath0 . this concludes the proof of the theorem . proof of lemma [ entropy ] set @xmath400}\frac{1}{\overline{f[\varphi]}}\biggr)(x),\ ] ] which is bounded and continuous , and rewrite @xmath401 so that it is sufficient to study the class consisting of translates of the fixed function @xmath402 . first , note that @xmath403 , @xmath348 , is uniformly bounded in view of the last estimate and since @xmath404 where we have used young s convolution inequality and plancherel s theorem . to prove the entropy bound , we will show that @xmath405 has finite quadratic variation ( i.e. , @xmath406-variation ) . in fact , to obtain a bound on the quadratic variation that is independent of @xmath49 , we renormalize and show that the function @xmath407 has quadratic variation bounded by a constant @xmath408 that depends only on @xmath47 . this will complete the proof of the lemma by using lemma 1 in @xcite , which states that the set of dilates and translates of a fixed function @xmath34 of bounded @xmath125-variation , @xmath147 , is of vc - type with constants @xmath409 depending only on @xmath125 and the @xmath125-variation norm of @xmath34 . we will prove that @xmath407 has bounded quadratic variation by showing that it is contained in the ( homogeneous ) besov space @xmath410 , which is sufficient , in view of the continuous embedding of @xmath410 into the space @xmath411 of functions of quadratic variation ( a result due to peetre see theorem 5 in @xcite for a proof , also the proof of theorem 2 in @xcite , which applies to @xmath127 as well ) . the seminorm @xmath412 of @xmath410 has the following littlewood paley characterization : @xmath413\|_2,\ ] ] where @xmath414 is a dyadic partition of unity with @xmath414 supported in @xmath415 $ ] ( see , e.g. , theorem 6.3.1 and lemma 6.1.7 in @xcite ) . we bound the littlewood paley norm : using the fact that @xmath416 = 2^{-j } f[\phi](2^{-j}\cdot)$ ] and plancherel s theorem , introducing the notation @xmath417 and in view of the support of @xmath414 , we have the bound @xmath418\bigr]\bigr\|_2 \\ & & \qquad= \frac{1}{2\pi } 2^{-j}\delta_j\sum_l 2^{l/2}\biggl\|\gamma_l f[\phi](2^{-j}\cdot)1_{[-2^ja,2^ja]}(\overline{f[\varphi]})^{-1}\frac{\langle u\rangle^{1/2}}{\langle u\rangle^{1/2}}\biggr\|_2 \\ & & \qquad\le c2^{-j}\delta_j\sum_l\bigl\|\gamma_l f[\phi](2^{-j}\cdot)1_{[-2^ja,2^ja]}(\overline{f[\varphi]})^{-1}\langle u\rangle^{1/2}\bigr\|_2 \\ & & \qquad\le c 2^{-j}\sum_l\sqrt{\int_{-2^ja}^{2^ja}\gamma^2_l(u)|f[\phi](2^{-j}u)|^2\langle u\rangle\,du } \\ & & \qquad\le c(a)2^{-j/2}\sum_l\|f^{-1}[\gamma_l f[\phi](2^{-j}\cdot)]\|_2 \\ & & \qquad= c(a)\sum_l\|f^{-1}[\gamma_l f[\phi_{j0}]]\|_2\le c(a)\|\phi_{j0}\dot{\|}_{0,2,1}.\end{aligned}\ ] ] to bound the last quantity , we use the inequality @xmath419 ( which follows from definition [ besov ] and results in @xcite , section 6.10 ) . by orthogonality of the wavelet basis ( @xmath256 , without loss of generality ) , @xmath420 the first term on the right - hand side is bounded by @xmath421 since @xmath422 is an @xmath20-projection . for the second term , we note , writing @xmath423 for @xmath424 and using the change of variables @xmath425 and condition [ wav ] , that @xmath426 so that @xmath427 this shows that @xmath428 is bounded by a fixed constant that depends only on @xmath131 , which completes the proof of the entropy bound . the proof of lemma [ entropy1 ] is the same ( in fact , it is simpler since , in the last step , by orthogonality , only the resolution level @xmath429 has to be considered ) . proof of proposition [ suprat ] we recall ( [ sup ] ) and observe that @xmath430 is bounded by the fixed constant @xmath135 . we prove @xmath431 ; the case @xmath432 is the same , except for notation . using the moment inequality ( 57 ) in @xcite and lemma [ entropy ] , we obtain @xmath433 where @xmath434 is obtained as follows : using plancherel s theorem , @xmath435(2^{-j}u)|^2|f[\varphi](u)|^{-2}\,du \\ & \le & \frac{1}{2\pi}2^{-2j}\|g\|_\infty\int_{-2^ja}^{2^ja}|f[\phi_{0k}](2^{-j}u)|^2\,du \\ & \le & \frac{1}{2\pi}2^{-j}\|g\|_\infty\int_{-a}^a|f[\phi_{0k}](v)|^2\,dv \\ & = & 2^{-j}\|g\|_\infty\le 2^{-j}g^2\equiv\sigma^2,\end{aligned}\ ] ] a bound which does not depend on @xmath34 . the claim for general @xmath125 follows from standard arguments for uniformly bounded empirical processes , using , for instance , proposition 3.1 in @xcite . we now prove the second statement . for every @xmath436 , talagrand s inequality in bousquet s version @xcite applied to @xmath437 yields @xmath438 now , the first statement of the proposition and taking @xmath439 imply , after some elementary computations , that @xmath440 which completes the proof . proof of proposition [ suprat - beta ] the proof is the same as that of proposition [ suprat ] ( up to some obvious modifications ) . first , consider theorem [ uni - band ] . the bias is @xmath441 where @xmath442 depends only on @xmath443 ( see theorem 9.4 in @xcite ) . for the `` variance '' term , proposition [ suprat ] and our choice for @xmath444 give @xmath445 using proposition [ suprat ] and the above bias - variance decomposition , the proof of theorem [ mod - ill - theo ] is similar to that of theorem [ uni - band ] and is left to the reader . for simplicity of notation , we suppress the suprema over @xmath23 in most of what follows uniformity of the bound follows from tracking all of the constants involved and noting that any density in @xmath23 is bounded by a fixed constant @xmath135 that depends only on @xmath302 . we have @xmath446 the first term in the right - hand side is treated in proposition [ suprat ] , which implies that @xmath447 ) . for the third , `` deterministic , '' term , we have , from standard approximation results for wavelets ( theorem 9.4 in @xcite ) , @xmath448 which is again of smaller order than the right - hand side in ( [ maint ] ) . the quantity inside the expectation of the supremum of the second term can be decomposed , for any @xmath449 , as @xmath450 and we denote these terms ( i)(iv ) . we first treat the `` large deviation '' terms ( ii ) and ( iii ) . for ( ii ) , using ( [ abs ] ) and the cauchy schwarz inequality , we have @xmath451 \\ & & \qquad\leq \sum_{l=0}^{j_{1}-1}2^{l/2}c(\psi)\bigl[e\sup_{k}|\hat{\beta}_{lk}-\beta_{lk}|^{2}\bigr]^{1/2}\bigl[e\sup_{k}1_{|\hat{\beta}_{lk}|>\tau,|\beta_{lk}|\leq\tau/2}\bigr]^{1/2}.\nonumber\end{aligned}\ ] ] we have , using the second part of proposition [ suprat - beta ] , choosing @xmath206 large enough depending only on @xmath452 and using the fact that @xmath453 is bounded by a fixed constant independent of @xmath429 , @xmath454\\[-8pt ] & & \qquad\le \pr\biggl(\sup_{k } |\hat{\beta}_{lk}-\beta_{lk}|>c(a , w , c)\kappa ' g\frac{1}{\delta_l}\sqrt{\frac{\log n}{n}}\biggr)\nonumber \\ & & \qquad\le \pr\biggl(\sup_{k } |\hat{\beta}_{lk}-\beta_{lk}|>\frac{c(a , w , c)\kappa'}{\delta_l}g\sqrt{\frac{[1+(\log n / l')-1]l'}{n}}\biggr)\nonumber \\ & & \qquad\le e^{-2\log n}.\nonumber\end{aligned}\ ] ] now , combining ( [ interm1 ] ) and ( [ interm2 ] ) with the first part of proposition [ suprat - beta ] yields the bound @xmath455 for ( ii ) . for term ( iii ) , using ( [ interm2 ] ) , as well as @xmath456 for any density @xmath7 , we have @xmath457 we now bound ( i ) . let @xmath458 be such that @xmath459 and @xmath460 [ such @xmath458 exists by the definitions ] . proposition [ suprat - beta ] and ( [ abs ] ) give @xmath461 where @xmath462 depends only on @xmath463 . for the second part of ( i ) , using the fact that definition [ besov ] implies @xmath464 for @xmath465 , the definition of @xmath203 and proposition [ suprat - beta ] , we obtain @xmath466 where @xmath467 depends only on @xmath468 . to complete the proof , we control the term ( iv ) . again using ( [ biascof ] ) , we have @xmath469 since the antagonistic terms in the minimum are strictly monotone in @xmath429 , the @xmath470 for which they are maximal is the one where they are equal so that @xmath471 [ cf . ( [ opt ] ) ] . if we denote by @xmath472 $ ] the integer part of @xmath473 then the last sum is bounded by @xmath474}2^{l(w+1/2)}\sqrt{\frac{\log n}{n}}+c'\sum_{l=[l^*]+1}^{j_1 - 1 } 2^{-ls } \le c''\biggl(\frac{\log n}{n}\biggr)^{s/(2w+2s+1)}.\ ] ] the following proposition is the wavelet - analog of a similar result in proposition 1 in @xcite for kernel regularizations . [ smooth - bias ] let @xmath131 satisfy condition . let @xmath219 for some @xmath220 . we then have , for every @xmath475 that @xmath476 where the constant @xmath477 depends only on @xmath478 . using ( [ abs ] ) , plancherel s theorem and the fact that @xmath479 we have @xmath480(u)}ff(u)\,du\biggr| \\ & \le & c ' \sum_{l\ge j}\sup_{k\in \mathbb z } \int_{{\mathbb{r } } } |f[\psi](2^{-l}u)| |ff(u)|\,du \\ & \le & c'\|\psi\|_1 \sum_{l\ge j}\int_{\mathbb{r}\setminus[-2^la',2^la']}|ff(u)|e^{\tilde c_0 |u|^s}e^{-\tilde c_0|u|^s}\,du \\ & \le & c''\|\psi\|_1 \sqrt{l}\sum_{l\ge j}\sqrt{\int_{2^la'}^{\infty}e^{-2\tilde c_0 u^s}\,du}\end{aligned}\ ] ] and the result follows from the inequality @xmath481 for @xmath482 . proof of corollary [ supsm ] decomposing the sup - norm error of the linear estimator into `` bias '' and `` variance '' terms and applying propositions [ suprat ] and [ smooth - bias ] , we have , for any @xmath109 , @xmath483 where @xmath484 depends only on @xmath485 . the result follows immediately for @xmath486 and for @xmath487 in view of the fact that @xmath488 for all @xmath24 . proof of theorem [ minm2 ] the proof of this theorem follows the one of theorem [ theo - minmax ] up to the following modifications . let @xmath125 be the standard cauchy density . fix @xmath489 . since @xmath490(u)=e^{-|u|}$ ] , we see from the scaling property of fourier transforms and since @xmath491 that there exists a constant @xmath492 large enough such that @xmath493 . as in the proof of theorem [ theo - minmax ] , we consider the functions @xmath494 , @xmath495 , @xmath496 , @xmath497 with @xmath4982^{js}}.$ ] we have @xmath499 for every @xmath317 if @xmath379 is a constant taken small enough and depending only on @xmath500 since @xmath501(t)|^2e^{2\tilde{c}_0|t|^s}\,dt \\ & & \qquad\le 2\int_{\mathbb r}|f[f_0](t)|^2e^{2\tilde{c}_0|t|^s}dt+2\gamma_j^2\int_{\mathbb r}|f[\psi_{j , k}](t)|^2e^{2\tilde{c}_0|t|^s}\,dt \\ & & \qquad\le 4\pi\nu^{2}l+2\gamma_j^22^{-j}\|\psi\|_1 ^ 2\int_{a'2^{j}}^{a2^{j}}e^{2\tilde{c}_0|t|^s}\,dt \\ & & \qquad\le 4\pi\nu^{2}l+2(c')^2lj2^{2jw}e^{-2\tilde{c}_0[a^s+1]2^{js}}\|\psi\|^2_1 a2^je^{2\tilde{c}_0a^s2^{js } } \\ & & \qquad\le 2\pi l.\end{aligned}\ ] ] take @xmath502}\log n$ ] . the proof of theorem [ theo - minmax ] then implies , @xmath503 , @xmath504 for some constant @xmath505 independent of @xmath0 . next , for any @xmath506 the kullback leibler divergence between @xmath345 and @xmath507 satisfies @xmath5082^{js}}2^{-2jw}\le c_{4}(c')^{2}l j.\ ] ] this and lemma [ lemlb ] together yield the result for @xmath379 chosen small enough independently of @xmath509 . proof of proposition [ rad ] we use proposition [ rad0 ] below . note that @xmath510 the class @xmath511 has envelope @xmath512 in view of ( [ abs ] ) and ( [ enve ] ) . since proposition [ rad0 ] deals with classes of functions bounded by @xmath513 we have to rescale , that is , we consider the class @xmath514 which is uniformly bounded by @xmath515 . furthermore , the upper bound for the weak variances @xmath516 can be taken to be @xmath517 in view of the estimate @xmath518 which uses young s inequality ( and the definition of @xmath519 from the proof of lemma [ entropy ] ) . to prove the inequality , set @xmath520 and @xmath521 so that @xmath522 but this quantity equals the probability in proposition [ rad0 ] below for @xmath523 . for the second claim of the proposition , we only have to show that @xmath524 has , up to constants , the required order as a function of @xmath396 . but this follows readily from the usual desymmetrization inequality for rademacher processes ( cf . , e.g. , expression ( 23 ) in @xcite ) , as well as from proposition [ suprat ] . proof of corollary [ band ] the result follows from standard arguments ( combining propositions [ rad ] and [ smooth - bias ] ) . we start with the following inequality , which is a bernstein - type version of similar inequalities in @xcite and complements the results in @xcite . let @xmath525 for any set @xmath124 and functions @xmath526 . [ rad0 ] let @xmath527 be i.i.d . with law @xmath12 on a measurable space @xmath528 . let @xmath124 be a countable class of real - valued measurable functions defined on @xmath529 , uniformly bounded by @xmath515 , and let @xmath530 . we have , for every @xmath262 and @xmath531 , that @xmath532 is greater than or equal to @xmath533 we first recall the lower - deviation version of talagrand s inequality , as given in @xcite , and a simple consequence of it . using the notation @xmath534 , we have , using the inequalities @xmath535 and @xmath536 , that @xmath537 and one likewise proves , using the upper - deviation version of talagrand s inequality @xcite , @xmath538\\[-8pt ] & & \hphantom{e^{-x}\ge\pr\biggl\{}\ge 1.5e\biggl\|\frac{1}{n}\sum_{i=1}^n\bigl(f(x_i)-pf\bigr)\biggr\|_\mathcal f+\sqrt{\frac{2x\sigma^2}{n}}+\frac{7x}{3n}\biggr\}.\nonumber\end{aligned}\ ] ] to prove the proposition , observe that @xmath539 where we have used the standard rademacher symmetrization inequality ( e.g. , ( 23 ) in @xcite ) . the first quantity on the right - hand side of the last inequality is less than or equal to @xmath532 , by ( [ bq ] ) . for the second term , note that the first displayed inequality in this proof also applies to the randomized sums @xmath540 , by taking @xmath541 , @xmath542 , instead of @xmath124 and the probability measure @xmath543 instead of @xmath12 . it is easy to see that @xmath544 can be taken to be the same as for @xmath124 . this gives the overall bound @xmath545 and a change of variables in @xmath341 gives the final bound . both authors would like to thank two anonymous referees for their remarks , and the second author gratefully acknowledges the hospitality of the cafe braeunerhof in vienna . bousquet , o. ( 2003 ) . concentration inequalities for sub - additive functions using the entropy method . in _ stochastic inequalities and applications . progress in probability _ * 56 * 213247 . birkhuser , basel . gin , e. , lataa , r. and zinn , j. ( 2000 ) . exponential and moment inequalities for @xmath135-statistics . in _ high dimensional probability , ii . progress in probability _ * 47 * 1338 . birkhuser , boston , ma .