Fluctuation theorem for currents in open quantum systems

A quantum-mechanical framework is set up to describe the full counting statistics of particles flowing between reservoirs in an open system under time-dependent driving. A symmetry relation is obtained which is the consequence of microreversibility for the probability of the nonequilibrium work and the transfer of particles and energy between the reservoirs. In some appropriate long-time limit, the symmetry relation leads to a steady-state quantum fluctuation theorem for the currents between the reservoirs. On this basis, relationships are deduced which extend the Onsager-Casimir reciprocity relations to the nonlinear response coefficients.


I. INTRODUCTION
Quantum systems can be driven out of equilibrium by time-dependent perturbations, by interaction with reservoirs at different chemical potentials or temperatures, or by combining both. In the latter cases, the quantum system is open and currents of energy and particles are flowing across the system. Such processes take place in mesoscopic electronic conductors as well as in chemical reactions. These nonequilibrium processes can be characterized by the relations linking their currents to the differences of chemical potentials. These relations may be linear in the case of Ohm's law, but are typically nonlinear, which defines the nonlinear response coefficients.
Alternatively, the full counting statistics of the particles transferred between the reservoirs can be considered. This statistics aims to characterize the transfers of particles in terms of the functions generating all the statistical moments of the fluctuating numbers of particles. The knowledge of this generating function gives access not only to the conductance and the noise power but also to higher-order moments and thus to the properties of nonlinear response. The full counting statistics has attracted considerable theoretical interest and is also envisaged in experimental measurements [1,2,3]. After the pioneering work of Ref. [4], several methods have been developed in order to obtain the full counting statistics in mesoscopic conductors. One of them is based on Keldysh Green's function formalism, in which an expression for the generating function has been obtained within a semiclassical approximation [5,6,7,8,9]. The full counting statistics can also be obtained on the basis of quantum Markovian master equations describing the fluctuations of the currents [10], as well as in terms of stochastic path integrals [11]. The generating function obtained in the approaches using the semiclassical approximation or the Markovian master equation has been shown to obey a symmetry relation as the consequence of time reversibility [12]. In nonequilibrium statistical mechanics, this relation is known as the fluctuation theorem which has been established for several classes of systems. These latter are either time-independent deterministic [13,14,15] or stochastic systems sustaining nonequilibrium steady states [16,17,18,19,20,21,22,23], or time-dependent Hamiltonian or stochastic systems, in which case the fluctuation theorem is closely related to Jarzynski nonequilibrium work theorem [24,25,26]. Similar symmetry relations have been considered for continuous-time random walks [27]. Quantum versions of the fluctuation theorem and Jarzynski nonequilibrium work theorem have also been obtained [28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43]. Moreover, a further relationship has recently been proved for time-dependent quantum Hamiltonian systems [44], allowing the derivation of Green-Kubo formulas and Onsager-Casimir reciprocity relations for the linear response coefficients [45,46,47,48].
An open question is to bridge the gap separating the time-dependent situations from the nonequilibrium steady states which are expected to be reached in the long-time limit. The problem is to deal with nonequilibrium steady states without relying on the semiclassical or Markovian approximations, or on the neglect of the energy or particle content of the subsystem coupling the reservoirs.
In the present paper, our purpose is to prove the fluctuation theorem for the currents in open systems obeying Hamiltonian quantum dynamics and sustaining nonequilibrium steady states in the long-time limit. We start by considering a time-dependent quantum system in contact with energy and particle reservoirs at different temperatures and chemical potentials. The amounts of energy and particles which are exchanged between the initial and final times are determined by quantum measurements. This framework is similar to the one considered by Kurchan to obtain a fluctuation theorem for quantum systems [28]. Here, this framework is extended by taking the initial states as grand-canonical instead of canonical equilibrium states, which allows us to deal with transfers of particles between the reservoirs. In this way, we obtain an exact relationship which is the consequence of microreversibility for the probability of a certain exchange of energy and particles between the reservoirs during the time-dependent external drive. An equivalent symmetry relation is obtained for the generating function of all the fluctuating variables. However, these symmetry relations are expressed in terms of the temperatures and chemical potentials of the reservoirs. The problem is that we need a symmetry relation in terms of the differences of temperatures and chemical potentials, which define the thermodynamic forces (also called the affinities) driving the currents across the system. The importance of this point has recently been discussed in the review [40].
The central contribution of the present paper is the proof that, in the long-time limit, the aforementioned generating function only depends on the differences between the parameters of the reservoirs. This proof is carried out by obtaining lower and upper bounds on the generating function in terms of a new generating function which only depends on the differences of parameters and further functions which are bounded in the long-time limit. Combining this fundamental result with the previously established symmetry relation of the generating function, the fluctuation theorem is proved for nonequilibrium steady states in open quantum systems. Thanks to this quantum fluctuation theorem, the Onsager-Casimir reciprocity relations and their generalizations to the nonlinear response coefficients can be inferred [21,49].
The plan of the paper is the following. The protocols for the forward and reversed drives of the open system are introduced in Sec. II. The symmetry relations for the probability and the generating function are proved in Sec. III. In Sec. IV, we obtain the quantum fluctuation theorem for the currents in the steady state reached in the long-time limit. In Sec. V, the consequences of the fluctuation theorem on the linear and nonlinear response coefficients are deduced. The conclusions are drawn in Sec. VI.

A. The total Hamiltonian
We consider a total quantum system composed of a subsystem in contact with several reservoirs of energy and particles. Initially, the reservoirs are decoupled. During a time interval T , the reservoirs are put in contact by some time-dependent interaction which has the effect of changing the energy and the particle numbers in each reservoir. The total Hamiltonian of the system is thus given by where H j denotes the Hamiltonian of the j th isolated reservoir before the interaction is switched on. During the time interval 0 ≤ t < T , the system is submitted to the time-dependent interaction V (t), which describes the coupling of the reservoirs by the subsystem. Beyond the final time t = T , the reservoirs are decoupled into the HamiltoniansH j . We suppose that the whole system is placed in a magnetic field B.
The observables of the reservoirs are the Hamiltonian operators H j andH j , as well as the numbers of particles of several species N jα with α = 1, 2, ..., c. The total number of particles of species α is given by where n α denotes the number of particles of species α in the subsystem between the reservoirs. Since the numbers of particles of species α are conserved within each isolated reservoirs before and after their coupling, we have that for all j, j ′ = 1, 2, ..., r and α = 1, 2, ..., c.
We suppose that the Hamiltonian operator H(t; B) has the symmetry under the time-reversal operator Θ. This latter is an antilinear operator such that Θ 2 = I and which has the effect of changing the sign of all odd parameters such as magnetic fields. Equation (6) expresses the microreversibility in an external magnetic field. The numbers of particles are symmetric under time reversal: In the following, we consider protocols with two quantum measurements at the initial and final times separated by a unitary time evolution (see Ref. [40] for a review).

B. The forward protocol
The forward time evolution is defined as with the initial condition U F (0; B) = I. In the Heisenberg representation, the observables evolve according to which also concerns the time-dependent Hamiltonian The average of an observable is thus obtained from We note that the dependence on the magnetic field is implicit in these expressions. The initial state of the system is taken as the following grand-canonical equilibrium state of the decoupled reservoirs at the different inverse temperatures β j = 1/(k B T j ) and chemical potentials µ jα : where Φ j (B) = −k B T j ln Ξ j (B) denotes the thermodynamic grand-potential of the j th reservoir in the initial equilibrium state. A quantum measurement is performed at the initial time. The system is observed in the eigenstate |Ψ k of the reservoir operators of energy and particle numbers: At the final time, another quantum measurement is performed and the system is observed in the eigenstate |Ψ l of the reservoir operators of energy and particle numbers: Accordingly, during the forward time evolution, the following change of energy in the j th reservoir is observed: while the number of particles of species α in the j th reservoir changes by C. The reversed protocol The evolution operator of the reversed process is defined as with the initial condition U R (0; B) = I, and is related to the one of the forward process by the following

Lemma: The forward and reversed time evolution operators at the final time T are related to each other by
This lemma is proved by noting that the forward time evolution in the magnetic field B, followed by the operation of time reversal, by the reversed time evolution in the magnetic field −B, and finally by time reversal again is equal to the identical operator: from which we deduce Eq. (20).
The reversed protocol is supposed to start with the following grand-canonical equilibrium state of the final decoupled reservoirs: at the same inverse temperatures β j = 1/(k B T j ) and chemical potentials µ jα as in the forward protocol and wherẽ Φ j (−B) = −k B T j lnΞ j (−B) denotes the grand-canonical thermodynamic potential of the j th reservoir in the final equilibrium state and reversed magnetic field. Similarly to the forward protocol, quantum measurements are performed at the initial and final times to determine the changes of energies and particle numbers in the reservoirs.

III. CONSEQUENCES OF MICROREVERSIBILITY
A. The symmetry relation for the probability of the fluctuations The probability distribution function to observe the energy (17) and particle transfers (18) during the forward protocol is defined as We notice that this function is a probability density because the quantities δ(·) are Dirac distributions for both the energy and the particle numbers. Inserting the expression of the initial density matrix (12), using the Dirac delta distributions to replace the initial energies and numbers into the final ones, we find that where we have introduced the difference of the thermodynamic grand-potential of the j th reservoir as According to the lemma (20), the probability of the transition k → l during the forward process is equal to the probability of the transition l → k in the reversed process and magnetic field: Substituting this identity in Eq. (24) and introducing the probability of negative changes in the energies and particle numbers during the reversed process as with the final density matrix (22), we obtain the following symmetry relation: If this relation is restricted to the energy change in a single system, this fluctuating quantity is the work W performed on the system and we recover the quantum version of Crooks' fluctuation theorem with the corresponding difference of free energy ∆F =F (−B) − F (B) [25]. The relation (28) extends this result to the transfer of particles under the effect of the differences of chemical potentials driving the system out of equilibrium.

B. The symmetry relation for the generating function
The generating functions of the statistical moments of the exchanges of energy and particles are defined by for the forward and reversed processes. The knowledge of these generating functions provides the full counting statistics of the process. We notice that the generating function of the forward protocol is alternatively defined as and the generating function of the reversed protocol as with Taking the Laplace transforms of the symmetry relation (28), we obtain an equivalent symmetry relation in terms of the generating functions: in terms of the temperatures and chemical potentials of the reservoirs. As mentioned in the introduction, this symmetry relation has not yet the appropriate form because the thermodynamic forces or affinities do not appear.

IV. QUANTUM FLUCTUATION THEOREM FOR THE CURRENTS
In this section, we prove that, in the long-time limit, the generating functions entering into the symmetry relation (37) only depend on the differences of the parameters ξ j and η jα , leading to the requested symmetry. In the long-time limit, a nonequilibrium steady state can be reached between the reservoirs if the coupling remains constant over the whole time interval except finite transients.

A. The theorem
We consider a situation where two large quantum systems interact through a bounded time-dependent perturbation described by V (t). Then, the generator of time evolution of the whole system is given by Hereafter, we assume that Let N jα (j = 1, 2) be the number of α-particles in the j th large system and assume that [N 1α + N 2α + n α , H 1 + H 2 + V 0 ] = 0 with a bounded n α (imagine that a quantum dot is located between the two electrodes).
Since the interaction is symmetric under time reversal V (t) = V (T − t), the evolution operator of the forward and reversed protocols are identical and therefore solutions of one and the same equation: with the initial condition U (0; B) = I and = 1. For the same reason, the initial and final reservoir Hamiltonians are the same, H j =H j for all j = 1, 2, ..., r, so that the initial and final density matrices have the same definition where Φ j (B) is the thermodynamic grand-potential of the j th reservoir in magnetic field B. Accordingly, the forward and reversed generating functions also have the same definition with Eqs. (32) and (33) and where the average · is carried out with respect to the density matrix (41). According to Eq. (37), this generating function has the symmetry in terms of the temperatures and chemical potentials of the reservoirs. Our purpose in this section is to prove the Proposition. Assume that the limit exists, it is a function only of ξ 1 − ξ 2 and η 1α − η 2α : The interpretation of this proposition is that, because of the finiteness of the subsystem and the interaction V 0 , the energy and particles lost by the left (respectively right) reservoir are transferred to the right (respectively left) reservoir within the overwhelming duration t 0 ≤ t ≤ T − t 0 and, as a result, Q becomes a function Q depending only on the differences ξ 1 − ξ 2 and η 1α − η 2α . We remark that the explicit form of the generating function Q(ξ 1 − ξ 2 , η 1α − η 2α ; B) is given by Eq. (90).
The above proposition implies that where we have introduced the affinities: driving respectively the heat current and the α-particle currents from reservoir 2 to reservoir 1. The result (49) is obtained by using the definition (45) at the line (46), the symmetry (43) and the independency of the quantities ∆Φ j on the time interval T at the line (47), again the definition (45) at the line (48), and finally the definitions of the affinities (50) and (51). Hence, we have the Fluctuation theorem. The generating function of the independent currents satisfies the symmetry In the particular case where the two systems have the same temperature, β 1 = β 2 , the generating function has the symmetry: and we recover the symmetry of the generating function of the independent particle currents, which has already been proved elsewhere for stochastic processes [23]. We notice that the fluctuation theorem (52) which is here proved thanks to the proposition (44)- (45) reduces to the steady-state fluctuation theorem presented as Eq. (104) in the review [40] for vanishing magnetic field, B = 0. Accordingly, the proposition (44)-(45) also provides a rigorous proof of such steady-state fluctuation theorems.

B. Setting
In order to demonstrate the above proposition, the time evolution is decomposed into different pieces corresponding to the short initial transient over 0 < t < t 0 , the long steady interaction over t 0 < t < T − t 0 , and the final short transient over T − t 0 < t < T . We introduce the lapse of time of the steady interaction In addition to U (t; B) defined by Eq. (8), we introduce U 1 (t; B) as the solution of It is then easy to show . We further note that where where Γ λ (τ ) = e iH0τ Γ λ e −iH0τ (λ = i or f ). For later purpose, we introduce where This is our starting point. Note that C and D are Hermitian for real ξ j and η jα and that D is the function only of ξ 1 − ξ 2 and η 1α − η 2α .

C. Some Inequalities
Here, for the sake of self-containedness, well-known equalities and inequalities [50] necessary for the following proof are summarized. For an operator X, the operator norm X is defined by and it satisfies: Equality 1: For any unitary U , U † XU = X . Indeed, we find where we have set |ψ = U |ϕ .
where we have used Equality 1 for the norm. Similarly, e −C−D Γ f (τ ) † e C+D = e −C−D Γ † f e C+D . In short, in terms of we have Step 2: In terms of X 2 = e −C−D Γ i (−t 0 )e C+D , one has from Inequality 2 Conversely, in terms of In short, let G 2 (ξ j , η jα ; B) ≡ e C+D Γ † τ e −2C−2D Γ τ e C+D , then (76) Step 3: We set whereH = H 1 + H 2 + V 0 andN α = N 1α + N 2α + n α . Then, in terms ofΓ τ = U τ e iH0τ , we have Furthermore, Inequality 2 gives Conversely, we have where we have used Inequality 2. Let Y 3 = e iH0τ e −D e −C e C e D e −iH0τ , then Inequality 1 and Equality 1 lead to Thus, Step 4: From Steps 1 to 3, in terms of we have Note that the constants L and K are independent of τ and that G 3 (ξ j , η jα ; B) is a function only of ξ 1 − ξ 2 and η 1α − η 2α since the operator D depends only on them.
Step 5: If exists, one has In short, we have shown: The left-most term only contains D, which is a function only of ξ 1 − ξ 2 and η 1α − η 2α , or Q.E.D.
We would like to remark that, even when the system has very long but finite recurrent times, the quantities Q and Q can be introduced and the Proposition is expected to hold with errors of order of 1/T . Firstly, in such a case, the ratio − ln G/T would converge to a definite value Q provided T is sufficiently longer than the relaxation time but shorter than the recurrent time. The same would be valid for Q. Secondly, even in such a case, the quantities appearing in (85) and (86) are bounded by constants independent of the reservoir volumes Ω if the interaction V (t) and the subsystem particle numbers n α have finite norms. Indeed, where 'T exp' is the time-ordered exponential, and its norm is bounded by exp [t 0 sup t V (t) ] [50]. Then, the product e C+D Γ i (−t 0 )e −C−D is Ω-independent and has a finite norm because e −C−D is O(e ∓Ω ) if e C+D is O(e ±Ω ). By similar arguments, all the norms in Eqs. (85) and (86) are found to be Ω-independent and, thus, the difference Q − Q is of the order of 1/T instead of Ω/T . Accordingly, the equality (91) is obtained in the limit T → ∞.

E. Generalization
The previous results can be generalized to the case of r > 2 reservoirs. In this case, the proposition (45) is that the generating function is a function depending only on the independent parameters:ξ with j = 1, 2, ..., r − 1. The proof is similar as in the case r = 2 with the operators: replacing Eqs. (61)-(62), where H 0 = r k=1 H k and N 0α = r k=1 N kα . In the general case, the fluctuation theorem should read in terms of the independent affinities A jα ≡ −β j µ jα + 1 r r k=1 β k µ kα , for α = 1, 2, ..., c, with j = 1, 2, ..., r − 1.

A. Fluctuation theorem and response coefficients
If we gather the independent parameters and affinities in the case of r = 2 reservoirs as or in the general case of r > 2 reservoirs as with α = 1, 2, ..., c and j = 1, 2, ..., r − 1, the fluctuation theorem (52) reads where we have explicitly written the dependence of the generating function on the affinities defining the nonequilibrium steady state. The idea is to differentiate successively the fluctuation theorem with respect to both λ λ λ and A A A to obtain symmetry relations for the linear and nonlinear response coefficients as well as further coefficients characterizing the statistics of the current fluctuations [21].
On the one hand, the mean currents can be obtained from the generating function and, on the other hand, expanded in powers of the affinities: which defines the response coefficients: · · · around the state of thermodynamic equilibrium. We notice that, if we set λ λ λ = 0 0 0 in the fluctuation theorem (52), we obtain the identities The former is a condition of normalization and the latter a condition of global detailed balancing which is a consequence of the fluctuation theorem (52) but may be assumed for itself as a weaker property than the fluctuation theorem [51].
On the other hand, the generating function of the cumulants of the fluctuating currents at equilibrium satisfies obtained from the fluctuation theorem (52) at the equilibrium A A A = 0 0 0. We start by differentiating the fluctuation theorem with respect to the generating parameters {λ α } and also the affinities {A α } to get which shows in particular that the mean currents vanish at equilibrium.

B. Onsager-Casimir reciprocity relations and Green-Kubo formulas
Now, we differentiate Eq. (112) with respect to λ β to obtain Setting λ λ λ = A A A = 0 0 0, we find the identity for the second-order cumulant of the current fluctuations at equilibrium. If we differentiate Eq. (112) with respect to A β , we get which reduces to for λ λ λ = A A A = 0 0 0. We recover the formulas of Green-Kubo type in the case α = β: The differentiation of Eq. (113) with respect to A β leads to Using Eqs. (109) and (110) in the limit λ λ λ = A A A = 0 0 0, we have Combining with Eq. (119), we finally find the Onsager-Casimir reciprocity relations: We notice that Eq. We end with the differentiation of Eq. (121) with respect to A γ to obtain Taking λ λ λ = A A A = 0 0 0 therein, we deduce We notice that this relation is the consequence of the weaker condition of global detailed balancing (110) alone and could hold even though the fluctuation theorem (52) does not as recently shown in Ref. [51] where the versions of Eq. (132), which are (anti)symmetrized with respect to the magnetic field, appears with the notations M α,βγ = αβ,γ , and whereupon we finally find the symmetry relations: If the magnetic field vanishes B = 0, Eqs. (135) reduces to the symmetry relations which were previously deduced as consequences of the fluctuation theorem [22,49].

D. Symmetry relations at 3rd order
Beside the third-order response coefficient (108), we introduce the coefficients with characterizing the nonequilibrium responses of respectively the second and third cumulants of the current fluctuations.
We continue the deduction by further differentiating the relations of the previous subsection with respect to λ δ or A δ at λ λ λ = A A A = 0 0 0. We find successively from Eq. (124): which has been obtained elsewhere as a consequence of the fluctuation theorem [49].

VI. CONCLUSIONS
In this paper, a fluctuation theorem for the currents has been proved for open quantum systems reaching a nonequilibrium steady state in the long-time limit [52]. In the considered protocol, the heat and particle currents are defined in terms of the exchanges of energy and particles between reservoirs, as measured at the initial and final times when the reservoirs are decoupled.
We start from a general symmetry relation for the generating function of the exchanges of energy and particles, which is the consequence of the microreversibility guaranteed by the measurement protocol. This symmetry relation is expressed in terms of the temperatures and chemical potentials of the reservoirs. However, the fluctuation theorem for the currents requires a symmetry with respect to the thermodynamic forces or affinities which are given in terms of the differences of temperatures and chemical potentials.
We show that this symmetry indeed holds by proving that, in the long-time limit, the generating function only depends on the differences between the parameters corresponding to the different reservoirs. Accordingly, the generating function of the currents has the symmetry of the fluctuation theorem with respect to the affinities. A rigorous proof is thus established for such steady-state fluctuation theorems as considered in the review [40].
As a consequence, the Onsager-Casimir reciprocity relations can be obtained for the linear response coefficients from the fluctuation theorem. Furthermore, generalizations of the reciprocity relations to the nonlinear response coefficients can also be deduced.