Arxiv:0911.3929V2

arXiv:0911.3929v2 [cond-mat.stat-mech] 22 Feb 2010 e.Cnie ytmwihi ntal ta equilibrium an at initially is state which system arbitrarily a parame- Consider driven external an ter. system varying involve by a equilibrium which from on away [17], far done theorem work Crooks of the fluctuation equal- and Jarzynski [14–16] the ity called relations, remarkable two are [10]. processes dynam- jump Langevin Markovian as and such [9] one, ics stochastic for dynamics later orig- thermostatted and [3] was deterministic law for theorem 2nd derived fluctuation the inally The into av- insight on thermodynamics. an are of give These [9–13], and ‘entropy process non-negative etc. irreversible erage 8], as an [7, termed in contact in production’ usually being quantities, body a between nonequilibrium pro- cold exchanged a heat heat and e.g., [1–3], fluid hot system, sheared driven a a in distributions duced in probability quantities the irreversibility various in of of asymmetries of measure terms quantitative fluctuation in recently a The the give [1–6]. through theorems theorems especially has fluctuation progress then, discovered of since amount founda- fair made of A time been the physics. has since statistical interest dynamics of great tion microscopic of subject reversible a been time from cesses eae oa qiiru state equilibrium an to relaxes protocol fixed work a called for process, this process In forward interval. time this outside constant reeeg ieec ∆ difference energy equilibrium free to trajectories nonequlibrium over performed varying rameter nta qiiru state equilibrium initial T n vnulytesse eae oteeulbimstate equilibrium the A to relaxes system the eventually and ntervrepoes h ytmi rvnfo the from driven is system the process, reverse the In . h aznk qaiyrltsaeaeo exp( of average relates equality Jarzynski The . lsl once oteflcuto hoes there theorems, fluctuation the to connected Closely pro- irreversible of thermodynamics Understanding A W ttemperature at λ λ sdn ntesse.Tesse eventually system The system. the on done is ( ( oeulbimflcuto hoesi h rsneo tim a of presence the in theorems fluctuation Nonequilibrium t t natm nevl0 interval time a in ) .Tesse sdie u feulbimby equilibrium of out driven is system The ). oasaehvn nfr rbblt esr naconstant 05.40.-a a 05.20.-y, 05.70.Ln, on numbers: measure PACS probability equilib system. isolated of uniform swi an is out a of driving ensemble having goes the microcanonical bath state when that, heat a syst is the to combined validity the the when called for even bath, requirement valid heat a are and theorems system sto a a consider in We forces nonconservative and classical field for symmetry-breaking theorems fluctuation nonequilibrium study We h exp( INTRODUCTION − βW hsc eatet ehin-Ire nttt fTechnol of Institute Israel - Technion Department, Physics ymtybekn edadnnosraieforces nonconservative and field symmetry-breaking T F B ) n ope oa xenlpa- external an to coupled and i = o ees protocol reverse a for exp( = F ( B B ) ≤ − − tsm temperature same at β t F λ ∆ ≤ ( ( A t F ,a mutof amount an ), τ as ) (1) ) where uybaaPradhan Punyabrata λ − ( λ τ ( βW t − is ) t ) ) rcs n htfrtervreprocess, reverse the for that done and work process of probabilities the the relates of theorem ratio Crooks The constant. Boltzmann the evtv ocshv o encniee.Rcnl,the Recently, considered. been noncon- but not [31], or have dynamics forces fields dy- servative stochastic symmetry-breaking Hamiltonian as time-reversal well classical any as for 30] before stud- [29, been line namics have this relations along fluctuation ied work with pre- The dynamics dynamics Newtonian served. microscopic stochastic the obeying of New- symmetries or microscopic obeying dynamics either and tonian combined, system a bath consider heat explicitly a we a where follow path we However different a [26–28]. with constraint trajectories flux nonequilib- macroscopic nonequilibrium a entropy- a to of modeling principle [25] with Jaynes’ of maximization applying contact by the prescription bath, in heat a of rium possibly is effect system, there nonequilibrium driven Recently the account bath. into take equilibrium to from away in goes 24]. bath [23, system heat driving the the during of of heat vicinity portion the generated to the the heat dissipated consequently continuously and scenario, is alwaysbath, realistic force is be driving a bath a to In heat by assumed the is equilibrium. cases, and in these explicitly, in mea- considered However canonical not to 22]. respect [17, sat- with sure dynamics balance Markov detailed or isfying 19–21] [9, fluctuation- theorem the dissipation satisfying 3], dynamics [2, Langevin dynamics stochastic by as, thermostatted such ways, deterministic various in employing modeled fluctuation is and the [18], theorems and phenomena irreversible derstanding processes, reverse the and forward the respectively. for work of tions where where ytma state at system ti hrfr motn o n mly etbath heat a employs one how important therefore is It un- for crucial is bath heat a of mechanism Dissipative ytm ntepeec fatime-reversal a of presence the in systems F P hsi swl sadtriitcstup. set deterministic a as well as chastic ce ff h obndsse relaxes system combined the off, tched F ( A ( W and ) m n hwta h fluctuation the that show and em, and ) imdrn rvn.Teonly The driving. during rium nrysrae ossetwith consistent surface, energy A P F g,Hia Israel Haifa, ogy, and P R ( F P B ( − ( R r qiiru reeeg fthe of energy free equilibrium are ) W B ( W W epciey and respectively, ) ) r h rbblt distribu- probability the are ) = e β ( W e-reversal − ∆ F W ) β o h forward the for 1 = /k B T , (2) k B 2

fluctuation theorem involving particle-current has been the classical systems consisting of particles which do not studied in a case of quantum mechanical transport of have any intrinsic magnetic moment. electrons across a quantum dot in the presence of a time- Here is a brief outline of the paper. In section II, we independent magnetic field [32]. study systems obeying Newtonian dynamics in the pres- In this paper, we generalize the fluctuation theorems ence of an external magnetic field and some other non- for classical systems in the presence of a time-reversal conservative force fields. In section III, we give a general symmetry-breaking field, such as an external magnetic proof of the fluctuation theorems for stochastic systems field, and nonconservative forces which cannot be derived in the absence of detailed balance in a microcanonical set from gradient of scalar potentials. We consider a system up, in section IV we then illustrate the ideas using two and a heat bath, combined, in a deterministic as well as a simple stochastic models. In section V, we generalize the stochastic set up and we show that the fluctuation theo- results for other intensive thermodynamic variables, e.g., rems are valid in the presence of time-reversal symmetry- pressure and chemical potential which determine the ini- breaking fields and nonconservative forces, even when the tial and final equilibrium states of the system in contact heat bath goes out of equilibrium during the driving. The with a heat bath. validity only requires that (1) in the absence of driving the system and the heat bath, combined, relax to a state with a uniform probability measure on a constant en- NEWTONIAN DYNAMICS ergy surface, and (2) there exists a time-reversal operation under which work performed on the system is odd. First, we study the nonequilibrium fluctuation theo- Although we specifically consider an external magnetic rems in a general deterministic framework for a system field in the paper, the results are also applicable to other and a heat bath combined, called the combined system time-reversal symmetry-breaking fields, e.g., a Coriolis (CS). We consider the CS, which is governed by micro- force present in a rotating system. scopic Newtonian dynamics, in the presence of an ex- In a deterministic set up, we consider a system obeying ternal magnetic field B~ (~r, t) and a nonconservative force Newtonian dynamics and we prove the fluctuation theo- field f~(~r, t). Force fields in general may be dependent rems using the fact that Liouville’s theorem is valid even both on position ~r and time t. The nonconservative force in the presence of an external time-dependent magnetic f~(~r, t) cannot be derived from gradient of a scalar po- field as well as other nonconservative forces. We also ex- tential. In addition, there may be conservative forces tend our analysis to stochastic dynamics in the presence present in the CS which can be derivable from gradient of a time-reversal symmetry-breaking field in a micro- of scalar potentials. A microstate of the CS is denoted canonical set up. We consider an isolated system gov- by a variable Y which contains positions and velocities erned by Markovian dynamics where there is violation of of all particles, i.e., Y (~r1, ~r2,...,~v1, ~v2,... ) ~ri, ~vi ≡ ≡{ } detailed balance with respect to a uniform measure, i.e., where ~ri and ~vi are position and velocity of i-th particle forward and corresponding reverse transition probabili- in the CS respectively. Newton’s equations of motion for ties are not equal in general. Reversing the time-reversal i-th particle can be written as symmetry-breaking field results in dynamics where all ~r˙ = ~v , (3) forward and corresponding reverse transition probabili- i i ˙ ~ ~ mi~vi = ~ri V (~ri, λ(t)) + qi~vi B(~ri,t) ties are interchanged with each other. Although detailed −∇ × balance is violated, we prove the fluctuation theorems ∂A~(~ri,t) qi + f~(~ri,t) (4) only requiring that the steady state measure of an iso- − ∂t lated system is uniform on a constant energy surface. where mi and qi are mass and charge of the i-th particle, We primarily focus on the work fluctuation relations, V (~r, λ(t)) is total scalar potential at position ~r due to the i.e., the Jarzynski equality and the Crooks fluctuation inter-particle interaction potentials as well as an external theorem. The system is driven by varying a control pa- potential with a time-dependent control parameter λ(t), rameter of an external potential or (and) by nonconser- ~ ~ri the gradient operator with respect to coordinate ~ri, vative forces. For systems obeying Newtonian dynam- ∇~ ics, the driving force may be due to a nonconservative A(~r, t) is the vector potential at position ~r and time t ~ electric field induced by an external time varying mag- due to the external magnetic field B(~r, t) which can be netic field in addition to other nonconservative forces. written as curl of the vector potential, i.e., B~ (~r, t) = The work fluctuation theorems have recently been stud- A~(~r, t). The third term in the r.h.s. of Eq. 4 is due ∇× ied for a few specific cases of a single Brownian particle to the time varying magnetic field B(~r, t) which induces in the presence of both time-independent [33, 34] and a nonconservative electric field, ∂A/∂t~ . The induced time-dependent [35] magnetic field. However we formu- electric field, like f(~r, t), cannot be− derived from gradient late the problem in a more general setting, taking into of a scalar potential. account the system and the heat bath degrees of freedom It is important to note that, when the nonconserva- explicitly. Note that our analysis is applicable only to tive force field f(~r, t) is present, there is no Hamiltonian 3 for the CS and consequently Eqs. 3, 4 cannot be de- since (dE/dt) = 0 outside the time interval 0 t τ. rived using a familiar Hamiltonian prescription of classi- The rate of change of energy is clearly odd under≤ time-≤ cal mechanics [38]. However the microscopic Newtonian reversal, i.e., (dE/dt) (dE/dt) as t t, ~vi ~vi →− →− →− equations of motion are still invariant under time-reversal and A~ A~ because (∂A/∂t~ ) is even and dλ/dt is with the direction of the magnetic field also reversed, odd under→ time-reversal. − In other words, total work per- i.e., as t t, ~vi ~vi and B~ (~r) B~ (~r) (equiva- formed equals to the difference in total energy between → − → − → − lently A~(~r) A~(~r)), Eqs. 3 and 4 remain unchanged. the final and the initial point of a trajectory and there- → − Therefore, for any trajectory Y(t) ~ri(t), ~vi(t) with fore total work is odd under time-reversal, F [Y(t)] = ≡ { } W a fixed protocol λ(t), B~ (~r, t), f~(~r, t) in a time range R[Y˜ (t)]. −W t , there{ exists a reverse} trajectory Y˜ (t) Let us now consider time evolution of phase space den- −T ≤ ≤ T ≡ Y Y ~ri( t), ~vi( t) for the corresponding reverse proto- sity ρ( ) at a phase space point ~ri, ~vi . From the { − − − } ≡{ } col λ( t), B~ (~r, t), f~(~r, t) . Note that in the time- equation of continuity, one obtains that the rate of change Y reversal{ − operation− − mentioned− above,} direction of the non- of phase space density ρ( ) equals to the divergence of Y˙ conservative forces, f as well as ∂A/∂t, are unchanged. local phase space current density ρ , i.e., one gets the In the subsequent discussions,− we consider a process local conservation equation [36] in a time interval t where is very large −T ≤ ≤ T T ∂ρ ∂ Y˙ compared to any other time scales. We assume that the + Y (ρ )=0, (8) magnetic field B~ (~r, t), the external parameter λ(t) and ∂t ∂ the nonconservative force field f~(~r, t) couple only to the where we have denoted the divergence of the phase Y˙ Y system. The field B~ (~r, t) (or equivalently the vector po- space current as ∂(ρ )/∂ = i,β[∂(ρr˙i,β)/∂ri,β + ~ ~ ∂(ρv˙i,β)/∂vi,β] where ri,β and vi,β are β-th Cartesian tential A(~r, t)), λ(t) and f(~r, t) are varied according to P a fixed protocol only in time interval 0 t τ where component (β =1, 2, 3 in three dimension) of the position ≤ ≤ τ . Otherwise B~ , λ are kept constant and f~ = 0 vector ~ri and the velocity vector ~vi respectively. Taking outside≪ T the interval 0 t τ. derivative explicitly with respect to the phase space point Y Although there is no≤ Hamiltonian≤ in the presence of , Eq. 8 can be rewritten as nonconservative forces, energy function of the CS, in ∂ρ ∂ρ ∂Y˙ terms of positions and velocities, can be defined as + Y˙ . + ρ =0, (9) ∂t ∂Y ∂Y 1 2 E( ~ri, ~vi )= mi~vi + V ( ~ri , λ(t)), (5) Y˙ Y { } 2 { } where .(∂ρ/∂ ) = i,β[r ˙i,β(∂ρ/∂ri,β) + i ! X v˙i,β(∂ρ/∂vi,β)] and the phase space compression where the first term is the total kinetic energy and the Y˙ Y P factor ∂ /∂ = i,β[(∂r˙i,β/∂ri,β) + (∂v˙i,β/∂vi,β)]. second term V ( ~ri , λ(t)) is the total potential energy, { } Since the r.h.s. of Eq. 3 is independent of ~ri, taking containing both the interaction pair-potentials dependent partial derivative ofP Eq. 3 with respect to the position on relative position ~ri ~rj between any pair of particles | − | coordinate, one gets (∂r˙i,β/∂ri,β) = 0. Note that the 2nd i, j and an external potential with a control parameter λ. term in the r.h.s. of Eq. 4 depends on ~vi only through The total energy of the CS, defined in terms of positions the cross product with the external magnetic field vector does not depend and velocities, on the external magnetic B~ and therefore partial derivative ∂(~vi B~ )β/∂vi,β = 0 field. However a time varying magnetic field does change × where (~vi B~ )β denotes β-th Cartesian component of the energy of the CS because of the work performed by an × the vector (~v B~ ). Since all other terms in the r.h.s. induced nonconservative electric field ∂A~(~r ,t)/∂t. This i i of Eq. 4 are× independent of ~v , taking derivative of can be seen as following: Using Eqs. 3, 4, the rate of i Eq. 4 with respect to the velocity coordinates, one gets change of total energy E of the CS can be written as (∂v˙i,β/∂vi,β) = 0. This implies that the phase space d compression factor ∂Y˙ /∂Y = 0, and therefore, from Eq. [E( ~ri, ~vi )] = ~vi. ~i + (∂V /∂λ)λ,˙ (6) dt { } F i 9, one arrives at Liouville’s theorem, X where ~i = [ qi∂A~(~ri,t)/∂t+f~(~ri,t)] is sum of all the ex- dρ ∂ρ ∂ρ F − = + Y˙ . =0. (10) ternal nonconservative forces acting on i-th particle of the dt ∂t ∂Y CS. This implies that the rate of change of total energy of the CS equals to the rate of work W done by all the The above equation is an important statement which says external forces on the system, i.e., (dE/dt) = (dW/dt). that, even in the presence of a time-dependent external Total work W performed on the system can be calculated magnetic field and other time-dependent nonconservative as W = T (dE/dt)dt, or forces, a set of phase space points flow like an incom- −T pressible fluid under microscopic Newtonian time evolu- R τ dE tion equations. Given that the phase space is incom- W = dt (7) dt pressible, the CS at t = , with f~ =0= (∂A/∂t~ ) Z0 ±T − 4 and λ = constant, can be considered to have a uniform entropy. We set the Boltzmann constant kB = 1 after- (microcanonical) measure on a constant energy surface, wards. Partitioning the CS into two parts, the system E(Y, λ)= constant. and the heat bath with energies ǫ and (E ǫ) respec- − SB(E−ǫ)+S(ǫ,λ) −1 One can now prove the Crooks theorem by using Li- tively, one can write Pst(Y, λ)= e dǫ ouville’s theorem that the phase space is incompress- where SB (E ǫ) and S(ǫ, λ) are entropy of the heat bath ible and the property that total work performed on and the system− respectively. WeR have here assumed the the CS is odd under simultaneous reversal of time interaction energy between the system and the bath to and the magnetic field. Let us denote the probabil- be much smaller than energy of either the system or the ity distributions of work W , P (W ; α(t)) PF (W ) and bath. Now introducing inverse temperature β of the heat ≡ P (W ;˜α(t)) PR(W ), respectively for a forward proto- bath, β = ∂SB(E)/∂E and expanding SB(E ǫ) in lead- ≡ − col α(t) λ(t), f~(~r, t), B~ (t) and corresponding reverse ing order of ǫ/E, SB(E ǫ) = SB(E) βǫ + (ǫ/E) ≡{ } − − −SB(EO) βF (λ) protocolα ˜(t) λ( t), f~(~r, t), B~ ( t) . Now follow- in the limit ǫ E, one gets Pst(E, λ) = e e ≪ ing the arguments≡ { − along the− line− of− Ref.} [30, 31], we where the Helmholtz free energy of the system F (λ) = consider a set A of initial phase space points at time (1/β) ln e−βǫeS(ǫ,λ)dǫ with eS(ǫ,λ) density of states − t = which evolve from a constant energy surface of the system with energy ǫ. R with−T energy E to a set of points A′ of the final phase From conservation of energy, we have E( ) = T space points of a constant energy surface with energy [E( ) + W ] where E( ) = E(Y( ), λ0), −T −T −T E + W at time t = for driving under the forward pro- E( ) = E(Y( ), λτ ) and W is total work per- T T tocol α(t). Total workT performed on the system in this formed for the forward protocol. Writing proba- process is W and the probability distribution of work bilities of the initial and final microstates respec- Y −SB(E(−T )) βF (λ(0)) PF (W ) = ω(A)/Ω(E) where ω(A) is the phase space tively as Pst( ( ), λ0) = e e and −T −SB (E(−T )) −βW βF (λ(τ)) volume of the set A and Ω(E) is the phase space volume Pst(Y( ), λ(τ)) = e e e , one gets T of the constant energy surface with total energy E. Now the ratio of probabilities of the final and initial equilib- for any trajectory with the forward protocol α(t), there rium microstates of the CS as exists a unique time reversed trajectory with the reverse Y Pst( ( ), λ(0)) β(W −∆F ) protocolα ˜(t), and work performed along a time reversed −T = e (12) Pst(Y( ), λ(τ)) trajectory, initially starting from one of the set of phase T ′ space points AR obtained by velocity-reversal of the set where β is inverse equilibrium tempera- A′, is negative of the work performed for the correspond- ture of the heat bath. Note that, writing ing forward trajectory. Therefore, for the reverse trajec- Pst(Y( ), λ(0))/Pst(Y( ), λ(τ)) = exp(∆SCS) in −T T tories, the phase space transforms from the energy sur- the l.h.s of Eq. 12, one obtains the thermodynamic face with energy E +W to an energy surface with energy relation T ∆SCS = W ∆F where ∆SCS is change − E. Then, the probability distribution PR( W ) can be in total entropy SCS of the CS, ∆F is change in free ′ − written as PR( W )= ω(AR)/Ω(E + W ). Now using Li- energy of only the system and temperature T =1/β [37]. ouville’s theorem− that phase space is incompressible, we Now substituting the above ratio of the probabilities ′ ′ ′ have ω(A)= ω(A ), and then using ω(A )= ω(AR) that into Eq. 11, one obtains the Crooks theorem in the phase space volume does not change under reversal of ve- presence of a time-dependent external magnetic field locities, one obtains the ratio of the probabilities of work and a nonconservative force, W and W as PF (W )/PR( W ) = Ω(E + W )/Ω(E). − − P (W ; λ(t), B~ (t)) The ratio can also be written as = eβ(W −∆F ). (13) P ( W ; λ( t), B~ ( t)) PF (W ) Pst(Y( ), λ(0)) − − − − = −T (11) PR( W ) Pst(Y( ), λ(τ)) The Jarzynski equality exp( βW ) = exp( β∆F ) fol- − T lows by integrating the Crooksh − theoremi [17].− where Pst(Y, λ)=1/Ω(E(Y, λ)) is the initial or the final The Crooks theorem has a simpler form when λ is kept equilibrium probability distribution of the CS. Note that constant (implying ∆F = 0), f~ = 0 throughout and only Pst(Y, λ) is independent of the external magnetic field B~ the magnetic field B~ (t) varies in a time-symmetric cycle as the total energy E given in Eq. 5, and therefore Ω(E), where B~ (t) = B~ (τ t) with initial and final values of − does not depend on B~ . B~ = 0. Consider an electrical circuit which is symmet- At this point one can separate the system from the ric with respect to B~ (t), e.g., see Fig. 1 where a ring heat bath by defining entropy and temperature of the is placed in an uniform time-dependent magnetic field in CS which has a uniform probability measure on a con- the direction perpendicular to the ring. The time vary- stant energy surface. The probability Pst(Y, λ) of a mi- ing magnetic field induces an oscillating electric field and crostate of the CS, at t = , is inverse of phase space an electric current in the circuit. For any finite number volume Ω(E) of a constant±T energy surface with energy of such cycles, the induced electric field ∂A/∂t~ per- − E, i.e., Pst = 1/Ω = exp( S/kB) where S is defined as forms work W on the system and thus generates heat − 5

Time dependent magnetic field Transition probabilities are chosen so that they obey B=Curl A(r,t) symmetries and conservation laws of underlying mi- A(r, t) croscopic dynamics. The degrees of freedom Y(t) Y , Y may be identified as two sets of stochastic≡ { + −} variables, Y+ (e.g., position) and Y− (e.g., velocity), and there exists a Y¯ Y , Y for any given Y. In the ≡ { + − −} t=0 t=τ t presence of a time-reversal symmetry-breaking field, such as an external magnetic field B~ , we impose a condition on the transition probabilities as given below,

w(Y′ Y; B~ )= w(Y¯ Y¯ ′; B~ ). (14) FIG. 1: Electric current flows in a ring due to a time- | | − dependent magnetic field B~ (~r, t)= ∇× A~(~r, t), perpendicular The above condition can be taken as definition of a to the ring, where the vector potential A~(~r, t) at position ~r magnetic field in a stochastic set up where reversing varies with time t. the magnetic field results in interchanging forward and corresponding reverse transition probabilities with each other. Indeed, under suitable assumptions, the tran- in the circuit. In this case, due to the geometric sym- sition probabilities chosen above can be derived for a metry, the probability distribution of work is same for closed isolated classical system governed by a micro- B~ (t) and B~ (t) and only depends on the magnitude of scopic Newtonian dynamics [39, 40]. Note that Eq. 14 − B~ , i.e., PF (W )= PR(W ) P (W ; B~ ). Since B~ varies in equates transition probabilities of two different systems, time-symmetric cycle, one≡ finally| arrives| at the Crooks one with a magnetic field B~ and the other with a mag- theorem P (W ; B~ )/P ( W ; B~ ) = exp(βW ) which gives netic field B~ . Time-reversal of a trajectory Y(t), in − an estimate of| irreversibility| − | of| the heat produced in an a symmetric time range t , is defined as −T ≤ ≤ T alternating electric current-carrying circuit. Y˜ (t) = Y¯ ( t) Y+( t), Y−( t) when t t. − ≡ { − − − } → − The fluctuation theorems can be similarly extended to The variables Y and Y¯ transform to each other under the cases where there are a Coriolis force 2m(~ω ~v) and a time-reversal where time-reversal operation is ensured by centrifugal force m~ω (~ω ~r) acting on a particle× of mass the condition in Eq. 14. m [38], ~ω being angular× velocity× of the rotating system, In the absence of a magnetic field, Eq. 14 (with in addition to an external magnetic field. The fluctuation B~ = 0) implies extended detailed balance condition ′ ′ theorems are still valid provided that one reverses the di- w(Y Y) = w(Y¯ Y¯ ). When Y Y+ contains only | | ≡ { } ′ rection of the angular velocity ~ω as well as the magnetic position-like variables, Eq. 14 becomes w(Y+ Y +; B~ )= ~ ′ | field B. This is because even if one adds the centrifu- w(Y + Y+; B~ ) which, for B~ = 0, implies the condition gal and Coriolis forces in Eq. 4, the phase space is still of detailed| balance− w(Y Y′ )= w(Y′ Y ) [39]. ˙ +| + +| + incompressible, i.e., ∂Y/∂Y = 0. Note that, under the condition of Eq. 14, an isolated CS in general does not satisfy detailed balance. To stress the violation of detailed balance, later in section III.A, STOCHASTIC DYNAMICS we would specifically consider a case where the reverse transition is not allowed and any reverse transition prob- In this section, we consider a system and a heat ability corresponding to a forward one is set to be zero. bath combined (CS), in a general stochastic framework. However choice of the transition probabilities cannot Stochasticity may arise due to incomplete knowledge of be arbitrary and one has to put some constraints so that some of the degrees of freedom in the original deter- (a) all the states are connected to each other ensuring ministic system [40]. Due to incomplete knowledge of the Markov process is ergodic, and (b) steady state con- the degrees of freedom, a system is described by some figurations are all equally probable. In this paper, we coarse-grained variables and is governed by a stochastic consider a class of stochastic models which satisfy con- dynamics. We consider a Markovian dynamics of the CS, straints (a) and (b). For a network of discrete states specified by transition probability w(Y′ Y)dY′, from a in a configuration space, a sufficient condition for such configuration Y at time t to any configuration| in the vol- a class, which we call loopwise balance condition, can ume element dY′ around Y′ at time t + ∆t, where the easily be formulated (see appendix for details). Even degrees of freedom of the CS are denoted as Y(t) at time if the CS relaxes to a state having a uniform measure, t. Transitions are allowed on a constant energy surface of the state would be a nonequilibrium steady state due to the CS. In subsequent discussion, we consider a class of the violation of detailed balance. It is important to note models where, in the absence of driving, a uniform (mi- that, provided there exists a unique and uniform steady crocanonical) measure is realized on a constant energy state measure ρ(Y) = constant for a Markov process surface of the isolated CS. with a magnetic field B~ , the Markov process with the re- 6 verse magnetic field B~ is well defined, i.e., the transition the trajectory Y˜ t, λ˜t, B~ , without reversing B~ , may not probabilities are still− normalized dY′w(Y′ Y; B~ ) = 1, even be realizable{ if some} of the reverse transition prob- and the same steady state measure ρ(Y) =| constant− is abilities are zero. From Eq. 14, the probabilities of R guaranteed for the Markov process with B~ . In other a trajectory from a given initial configuration with the words, dYw(Y′ Y; B~ )ρ(Y) = ρ(Y′) =−constant magnetic field B~ and that of the corresponding reverse | ⇒ ~ dY¯ ′w(Y¯ Y¯ ′; B~ )ρ(Y¯ ′) = ρ(Y¯ ) = constant, i.e., uni- trajectory with B are equal, R − form steady| state− measure is invariant under reversal of R [ Yt, λt, B~ ]= [ Y˜ t, λ˜t, B~ ] (15) the magnetic field. This is because the normalization P { } P { − } condition dY′w(Y′ Y; B~ ) = 1 implies the steady state | where [.] denotes respective probability of a trajectory. condition dY¯ ′w(Y¯ Y¯ ′; B~ ) = 1, which can be shown P R We call the above equation as the microscopic reversibil- by using Eq. 14 and| the transformation− Y Y¯ . This is R → ity (MR) condition hereafter. As a special case, when discussed and illustrated in the appendix. B~ = 0, the above condition can be written simply as We stress that the assumption of Markovian dynamics of a system and a heat bath, combined, is weaker than [ Yt, λt ]= [ Y˜ t, λ˜t ]. (16) that of Markovian dynamics of only the system. Even P { } P { } if the combined system obeys Markovian dynamics, dy- We define F [ Yt, λt, B~ ] as work performed along W { } namics of the system, in lower dimensional configuration a trajectory Yt, λt, B~ where F = [E(Y , λτ ) { } W T − space, is non-Markovian, in contrast to the system con- E(Y−T , λ0)], the difference in total energy of the final sidered in Ref. [22]. and initial point of the trajectory. For a forward proto- The total energy of the CS is denoted as E(Y, λ) where col λt , we define the probability distribution of work { } λ is an external parameter coupled only to the system. W , as P (W ; λt , B~ ) PF (W ) which can be written as When the CS is not driven, E(Y, λ) is conserved. Im- { } ≡ portantly, total energy E depends explicitly only on Y PF (W )= Pst(Y , λ , B~ ) [ Yt, λt, B~ ] −T 0 0 P { } and λ, not on the magnetic field B~ [41] and it is an even Yt {X} function of Y so that E(Y, λ)= E(Y¯ , λ). For simplic- − δ( F W ), (17) ity we assume time t changes in discrete step of ∆t and × W − we consider a Markov chain in a time range t where P (Y , λ , B~ ) is the initial steady state distribu- −T ≤ ≤ T st −T 0 where is very large. The parameter λ is changed from Y ~ T tion at time t = , and [ t, λt, B ] is the probability λ = λ0 to λ = λτ according to a deterministic protocol −T P { } of the trajectory Yt, λt, B~ . For the reverse protocol in a finite time interval 0 t τ where τ and ˜ { } ~ ≤ ≤ ≪ T λt with reversed magnetic field B, the probability otherwise kept constant. We call it a forward protocol. { } ~ − ˜ distribution P (W ; λ−t , B) PR(W ) of work W can A reverse protocol is defined as λt λ−t . { } − ≡ { }≡{ } be written as An amount of work δWt at time step t may be performed on the system in two ways. One may usually PR(W )= Pst(Y¯ T , λτ , B~ τ ) [ Y˜ , λ,˜ B~ ] change the external parameter from λt to λt + δλt, keep- − P { − } {Y˜ t} ing Y fixed, and the work performed is δWt = E(Yt, λt + X Y δ( R W ), (18) δλt) E( t, λt). Now we introduce here the second way × W − of performing− work on the system. One may also change where work performed along the trajectory Y˜ , λ˜ , B~ the degrees of freedom of the CS, at a time step t, deter- t t is = [E(Y¯ , λ ) E(Y¯ , λ )]. { − } ministically from Y to Y′ = S∆t(Y ), keeping λ fixed, R −T 0 T τ t t t ThroughoutW the paper,− we use two symmetry relations where St is a time-reversal symmetric evolution opera- as following. tor, i.e., if Y Y′ under influence of a nonconservative (1) [ Y , λ , B~ ]= [ Y˜ , λ˜ , B~ ], i.e., the work force, Y¯ ′ Y¯→under influence of the same force. For ex- F t t R t t performedW { is odd under} −W simultaneous{ reversal− } of time and ample, St→may simply be the Newtonian time evolution the magnetic field. operator. We will illustrate this by using a simple model (2) P (Y, λ, B~ )= P (Y, λ)= P (Y¯ , λ), i.e., the steady in section IV.B. Work performed in this case is calculated st st st state distribution is independent of the magnetic field as δW = E(Y + δY , λ ) E(Y , λ ) which is the work t t t t t t and invariant when velocities are reversed. performed by a nonconservative− force when the evolu- To show the symmetry relation 1, one should note that tion operator St contains such a force. The total work work done along a trajectory is, by definition, the differ- W performed on the system is written as W = δW . t t ence in total energy of the CS at the final and the initial A trajectory is denoted by Yt, λt, B~ where Yt, λt are Y P point of the trajectory, implying F = [E( T , λτ ) Y { }~ W − respective values of , λ at time t and B is the external E(Y−T , λ0)] and R = [E(Y¯ −T , λ0) E(Y¯ T , λτ )]. Now magnetic field. Given a trajectory Yt, λt, B~ , there is Y W Y¯ − { } using E( , λ) = E( , λ), i.e., energy is invariant when a unique reverse trajectory Y˜ t, λ˜t, B~ with reversed velocities are reversed, one obtains the symmetry relation { − } magnetic field B~ and reverse protocol λ˜t . Note that 1. The symmetry relation 2 holds because the steady − { } 7 state distribution of the CS is uniform on a constant en- P (W ; λt , B~ ) P (W ; λt , B~ ) (similarly for the { } − ≡ { } | | ergy surface where energy of the CS is independent of work probability distribution with reverse protocol λ˜t ). the magnetic field and the uniform steady state distri- This implies that, in this case, the Crooks theorem{ holds} bution does not change for the reversed velocities. In- even when the magnetic field is same for the forward dependence of the total energy on the magnetic field has and the reverse protocol. Replacing the index B~ by already been manifested in Eq. 5 where energy of a deter- B~ in Eq. 21, one can now write the Crooks theorem− ministic system has been expressed in terms of positions as P (W ; λt , B~ )/P ( W ; λ˜t , B~ ) = exp[β(W ∆F )]. and velocities [41]. Note that{ in} this case− one{ does} not have to reverse− the Using microscopic reversibility condition of Eq. 15 and direction of the magnetic field in the reverse protocol, the symmetry relation 2, changing summation indices and therefore the Crooks theorem expresses symmetries Yt Y˜ t , and then using the symmetry relation in the probability distributions of work for a system with 1,{ Eq.}→{ 17 can} be rewritten as same dynamics for the forward and the reverse protocol. This type of symmetry would be illustrated in an Y Pst( −T , λ0) example given in section IV.A. PF (W )= Pst(Y¯ T , λτ ) Pst(Y , λτ ) ˜ T {XYt} [ Y˜ t, λ˜t, B~ ]δ( R + W ). (19) STOCHASTIC DYNAMICS: ILLUSTRATION ×P { − } W Now defining entropy and temperature of the CS, In this section, we illustrate the ideas developed in the as done before in the case of Newtonian dynam- previous section by constructing two simple stochastic ics in section II, one can write the probabilities models. First we consider the effect of an external mag- of the initial and final microstates respectively as netic field where, to ensure violation of detailed balance, Y −SB(E−T ) βF (λ0) Y Pst( −T , λ0) = e e and Pst( T , λτ ) = we specifically choose reverse transition probability to be −SB(E−T ) −βWF βF (λτ ) e e e where SB entropy of the heat zero for any nonzero forward transition probability. Sec- bath and F (λ) the Helmholtz free energy of the system ond we consider a nonconservative force in a stochastic with the external parameter λ. So the ratio of the prob- set up. Although the two models considered in this sec- abilities can be written as tion are just toy models for a system and a heat bath, they nevertheless demonstrate the dissipative and equi- Pst(Y−T , λ0) = eβ(WF −∆F ), (20) librating mechanism of a heat bath, and subsequently P (Y , λ ) st T τ show the validity of the Crooks theorem even when the heat bath goes out of equilibrium during driving. where ∆F = F (λτ ) F (λ0) the difference in the Helmholtz free energy.− Substituting the above ratio of probabilities into Eq. 19, one arrives at the Crooks the- Time-reversal symmetry-breaking field orem in the presence of an external magnetic field,

~ We take a one dimensional ring of L + 1 sites where P (W ; λt , B) β(W −∆F ) { } = e (21) site i = 0 is considered as the system and all other sites, P ( W ; λ˜t , B~ ) − { } − 1 i L, are considered to be the heat bath (see Fig. ≤ ≤ 2). At any site i there is an energy variable ei 0. where the probability distributions of work in gen- ≥ eral depend on the magnetic field B~ as the transition The energy at site i = 0 is given by e0 = λx where probabilities depend on B~ . The Jarzynski equality, the external parameter λ couples only to the system via exp( βW ) = exp( β∆F ), is derived straightforwardly an internal degree of freedom x > 0. A configuration h − i − of the CS is thus specified by Y x, e ,...,eL . The by integrating the Crooks theorem [17]. Note that Eq. 21 ≡ { 1 } relates the probability distributions of work for two sys- dynamics is the following: a site i is chosen randomly and a fixed amount of energy δ ( 1) is transferred only in tems with different microscopic dynamics, i.e., one sys- ≪ tem with a magnetic field B~ and the other with a mag- one direction (say, anti-clockwise) to the nearest neighbor netic field B~ . Importantly, unlike the Crooks theorem, site, i.e., the Jarzynski− equality is written without any reference ei ei δ ; ei+1 ei+1 + δ. (22) to the magnetic field and so the Jarzynski equality is a → − → L statement regarding a system with a particular dynam- The total energy E = i=0 ei is conserved in this pro- ics. cess. Whenever energy e0 at i = 0 is changed, the vari- P ′ ′ The Crooks theorem takes an interesting form, if ge- able x is updated accordingly: e0 e0 x x = ′ → ⇒ → ometry of a system is symmetric with respect to the mag- e0/λ. For ei <δ, the energy transfer is not allowed. netic field B~ . Given this symmetry, the work probability There is a mean energy current in anti-clockwise di- distributions do not depend on the direction of B~ , but rection which may be considered to be due to an exter- depend only on the magnitude B~ : P (W ; λt , B~ ) = nally applied field in this direction (analogous to B~ ). The | | { } 8

i=L δ i=0 δ 100 δ

δ (-W) i=1 R 10 (W)/P

FIG. 2: Schematic diagram of the system and the heat bath. F 1 The system is the site i = 0, and rest of the sites, 1 ≤ i ≤ L, P constitute the heat bath. 0.1 1 3 5 7 reverse field corresponds to the dynamics where energy W is transferred in clockwise direction (analogous to B~ ). Since there are no velocity-like variables, we have Y−= Y¯ FIG. 3: The ratio PF (W )/PR(−W ) is plotted versus work W and time-reversal is simply defined as Y(t) Y( t) as − → − in semi-Log scale and fitted with exp[β(W ∆F )] for L = 100, t t in a symmetric time interval t . Note β = 1.0, λ0 = 1.0, λτ = 11.0 where τ = 100 and ∆F = →− −T ≤ ≤ T that, in this case, reverse transition probability is zero for (1/β)ln(λτ /λ0). any nonzero forward transition probability because energy is transferred only in one direction (anti-clockwise), i.e., w(Y Y′; B~ ) = 0 for any nonzero w(Y′ Y; B~ ) = 0. the forward and the reverse protocol. The probability Therefore| a time reversed trajectory is possible| only6 for distributions of work W for the forward and the re- the dynamics where energy is transferred in the reverse verse protocol are denoted as P (W ; λ) P (W ) and direction (i.e., clockwise). Clearly, the model satisfies the F P (W ; λ˜) P (W ) respectively. Due to≡ symmetriy of microscopic reversibility w(Y′ Y; B~ ) = w(Y Y′; B~ ) as R the ring geometry,≡ the work distributions do not depend given in Eq. 14 (with Y = Y|¯ ), and also satisfies| − the on the direction of the energy transfer. We verify nu- symmetry relations 1 and 2. merically that the Crooks theorem is indeed satisfied, When λ is kept constant, total energy E is conserved i.e., PF (W )/PR( W ) = exp[β(W ∆F )] with ∆F = and the dynamics is a totally asymmetric zero range pro- − − F (λτ ) F (λ0). In Fig. 3, we plot PF (W )/PR( W ) as cess [42] on a ring with a constant hopping rate where − − a function of W where λ0 = 1.0, λτ = 11.0, τ = 100, number of particles at a site is ei/δ. With total number β = 1.0, L = 100. The parameter λ is increased in a of particles fixed in the process, steady state configura- specific way: first λ is increased in 5 equal discrete steps tions are all equally probable. This can be understood by upto t = 5, then held constant upto t = 95, and again mapping the zero range process to a totally asymmetric increased in 5 equal discrete steps upto t = 100. The pa- simple exclusion process [42] where all possible states are rameter λ is varied in this particular way to ensure that equally probable in the steady state. In the limit of large the energy fluctuations travel around the ring and can L, probability distribution of energy at any site i is given perturb the system at site i = 0 within the measurement by the Boltzmann distribution, P (e ) = βe−βei where i time τ = 100. In Fig. 3, the ratio PF (W )/PR( W ) fits L −1 − β = [ i=0 ei/(L+1)] is inverse temperature of the CS. well with exp[β(W ∆F )] where ∆F = (1/β) ln(λτ /λ0) The partition function of the system, for a fixed value of is the theoretical of− the difference in free energy. P ∞ −βλx −1 λ, can be calculated as (λ) = 0 e dx = (βλ) and the free energy is givenZ by F (λ)= β−1 ln . R − Z The system is driven by changing the external parame- Nonconservative force ter, in discrete step of δλt at t-th time step, from an initial value λ to a final value λτ in time interval 0 t τ. For When a nonconservative force is present in the CS, e.g., 0 ≤ ≤ each increment δλt, an amount of energy δWt is added a system of particles in a ring in contact with a heat bath to the system (i = 0) where δWt = (∂e0/∂λ)δλt = x.δλt and with a force acting in anti-clockwise direction as in is defined as work performed at t-th time step. Total Fig. 1, the force field cannot be derived from the gradient work performed is W = t δWt. We set a unit of time of a scalar potential and therefore cannot be absorbed in such that all sites are updated with rate one per unit the expression of the total energy of the CS (e.g., see Eq. Monte Carlo time. For theP reverse protocol, the exter- 5). In this case, unlike changing an external parameter nal parameter is varied as λ˜(t) = λ(τ t) in time in- λ, the system is driven by changing Y Y + δY as − → terval 0 t τ, from λτ to λ0. Note that energy is discussed in section III. To illustrate this, we consider a always transferred≤ ≤ in anti-clockwise direction both for CS which consists of L + 1 lattice sites in one dimension. 9

2 The site i = 0 has energy e = p , with an internal because, given a trajectory [p(t), ei(t) ], the trajectory 0 { } variable p, and any other site i has energy ei 0. The [p( t), ei( t) ] is not realizable as p only increases un- site i = 0 is considered to be the system and≥ the rest der− the{ driving.− } However the microscopic reversibility is the heat bath. A configuration of the CS is specified condition in Eq. 16 is satisfied as the transition probabil- ′ ′ by Y p,e1,...,eL where, for any given Y, there is a ities have an additional symmetry, w(p , ei p, ei ) = ≡{ } ′ ′ ′ { }| ′ { } Y¯ p,e ,...,eL . The dynamics is chosen as follows. w( p, ei p , ei ), i.e., w(Y Y)= w(Y¯ Y¯ ). ≡ {− 1 } − { }|− { } | | For 1 i L we choose a site at random and exchange Following the general proof given in section III, ≤ ≤ energy between sites i and i + 1 randomly, one can see that the Crooks theorem is satisfied, P (W )/P ( W ) = exp(βW ), where the free energy ei q(ei + ei ); ei (1 q)(ei + ei ), (23) − → +1 +1 → − +1 change ∆F = 0. Since the forward and reverse protocol of driving is same in the above example, we have where q [0, 1] is a uniform random number. The total ∈ used PF (W ) PR(W ) P (W ) in the Crooks theorem. energy E = L is constant in this process. We update i=0 For a time-dependent≡ external≡ nonconservative force, the the site i = 0 slightly differently where we consider that increment δ of the internal variable p at a time step t the site i = 0P can interchange energy only with site i = 1. t will be δ = f .dt, where f is the force at time step t. Say, energy of the two sites, before update, are e = p2 t t t 0 In this case, the reverse protocol should be f for a and e respectively. We generate a random number ξ −t 1 given forward protocol f , and one should{ distinguish} uniformly distributed in the range [ p ,p ] where t max max between the work probability{ } distributions P (W ) and p = √e + e . We then update the− internal variable F max 0 1 P (W ). Then the Crooks theorem can be written in the p and energy of the site i =0, 1 as given below, R more general form as PF (W )/PR( W ) = exp(βW ). − p ξ; e ξ2; e (e + e ξ2). (24) → 0 → 1 → 0 1 − The update rule ensures that detailed balance is satis- GENERALIZATION fied with respect to a uniform measure on a constant energy surface of the CS. Consequently, while the CS The fluctuation theorems can be generalized to the is not driven, the site i = 0 has the Boltzmann prob- cases where a system is in contact with a heat bath with ability distribution P (p) = exp( βp2)/ where β = pressure P and (or) chemical potential µ. Let us consider L −1 − Z [ i=0 ei/(L + 1)] is inverse temperature of the CS and the combined system with total energy E, volume V and ∞ 2 = −∞ dp exp( βp ) is the partition function. number of particles N which are globally conserved. En- ZPThe system is driven− by changing the internal variable ergy, volume and number of particles ǫ, v and n of the p as follows:R p p + δ where δ > 0 is a constant (choice system fluctuate due to interaction with the heat bath. of the sign of δ is→ arbitrary). Now two following steps per- Pressure P and chemical potential µ can be defined, sim- formed repeatedly: Step.1 - random sequential update of ilar to temperature, as given below, L bonds of the CS using Eq. 23, 24 and Step.2 - update ∂S (E,V,N) of the site i = 0 by changing the internal variable from βP = B , (25) p to p + δ. The second step may be thought of, as if the ∂V ∂S (E,V,N) internal variable p is like momentum of a particle and βµ = B . (26) it is updated due to effect of an external constant non- ∂N conservative force f which changes p by a fixed amount Now using the expansion of the heat bath entropy S (E δ = f.dt in a small time interval dt. Note that, under the B ǫ, V v,N n)= S (E,V,N) βǫ βP v βµn in the− driving, the transition p p + δ (also (p + δ) p) B limit− of ǫ −E, v V and n −N,− one can− rewrite the is allowed, but the transition→ p + δ p−is not allowed.→ − ratio of the≪ probabilities≪ of microstates≪ at t = , as In other words, internal variable p only→ changes in one given in Eq. 20, as ±T direction, i.e., either increases (for δ > 0) or decreases (for δ < 0), under the driving. Work δWt done on the Y Pst( −T , λ0) β(WF [{Yt,λt,B~ t}]−∆G) system is the change in energy of the system (i.e., site Y = e (27) 2 2 Pst( T , λτ ) i = 0) where δWt = (p + δ) p = δ(2p + δ) and total − work W = t δWt. where (β,P,µ,λ) the grand potential of the system in The dynamics considered above is similar to the G P equilibrium with a heat bath of inverse temperature β, Langevin dynamics of a Brownian particle in a thermal pressure P and chemical potential µ, λ an external pa- environment where an external nonconservative force is rameter and ∆ = (λτ ) (λ0) with the grand poten- acting on the particle. One should note that, given a tial defined asG G − G trajectory [p(t), ei(t) ], one can define time-reversal op- G { } eration in two ways, i.e., as t t,[p(t), ei(t) ] 1 → − { } → (λ)= ln dǫ dv dne−β(ǫ+P v+µn)eS(ǫ,v,n,λ) [ p( t), ei( t) ] or [p(t), ei(t) ] [p( t), ei( t) ]. G −β But− − only{ the− first} way of time-reversal{ } → is− relevant{ − here} Z Z Z (28) 10 where S(ǫ,v,n,λ) is entropy of the system. Then the w5 Crooks theorem can be written as given below, w6 P (W ; λ(t), B~ (t)) = eβ(W −∆G) (29) P ( W ; λ( t), B~ ( t)) − − − − w C 8 C 4 which is obtained by replacing the Helmholtz free energy 7 w 2 C F in Eq. 21 by the grand potential . w w 2 4 2 w1 G w3 w w w w6 w 2 w 4 5 C3 1 1 SUMMARY w3 w 4 w C 2 C w 4 In this paper, we have studied the fluctuation theo- 1 3 w6 rems for a classical system in contact with a heat bath in C w 5 6 w5 the presence of a time-reversal symmetry-breaking field w C6 w and nonconservative forces, in a deterministic as well as 5 6 a stochastic set up. We have shown that the fluctuation theorems are valid under the condition that, in the FIG. 4: (Color online) Schematic diagram of a network in absence of any driving, the system and the heat bath, a configuration space: Configurations are denoted as nodes combined, relax to a state having a uniform probability C1, C2, C3, ... C8. Nodes are connected by various closed measure on a constant energy surface. The fluctuation loops, each of which is assigned a transition rate. Transition theorems have been proved in a very general setting by rates assigned to the dotted arrows should be added to get the corresponding total transition rate. using the time-reversal symmetry and the conservation laws, and accordingly defining the intensive thermodynamic variables like temperature, pressure, chemical potential obtained from a microcanonical ensemble. In the w5+w6 deterministic case of Newtonian dynamics, we have first shown that Liouville’s theorem holds even in the presence C of a time-dependent external magnetic field and other w4 8 time-dependent nonconservative forces and then, using C7 w2 w +w C Liouville’s theorem, we have proved the Crooks Theorem 2 4 w 2 and the Jarzynski equality in the presence of such forces. 1 w w +w 3 In the stochastic case, where the combined system obeys 4 5 +w C w1+w2 Markovian dynamics, the work fluctuation theorems have 6 3 + w3 been shown to be valid even when the reverse transition w w1+w2 4 C probabilities are not equal to the corresponding forward C1 4 w3 transition probabilities, thus violating detailed balance w6 condition. C5 w6 C w 6 5 w5 +w6 ACKNOWLEDGMENTS FIG. 5: Network with the magnetic field B~ : Transition rates The author thanks J. Robert Dorfman, Dov Levine and assigned to the dotted arrows in Fig. 4 are added to get the Yariv Kafri for many useful discussions and acknowledges actual transition rate (denoted by the thick arrows in this figure). Transition rates in general depend on the magnetic a fellowship of the Israel Council for Higher Education. field.

APPENDIX Although we now specifically consider the case where For an equilibrium system, detailed balance with re- all reverse transition rates to be zero for corresponding spect to a uniform (microcanonical) probability measure nonzero forward transition rates (similar to the exam- is a sufficient condition for all states to be equally prob- ple considered in section IV.A), the following discussion able in the final equilibrium state. Here we formulate can be straightforwardly generalized to cases where a a sufficient condition for having equally probable steady forward and corresponding reverse transition rate both states for a nonequilibrium system with finite number of may be nonzero. Also we only consider here the case states. where there is no velocity-like variables, however the gen- 11 eralization to such cases is straightforward. In Fig. 4, w +w a network in a configuration space is shown schemati- 5 6 cally. A configuration C is denoted by a node in the graph. Nodes are connected by drawing closed loops, w C 8 where each loop is assigned a transition rate, e.g., see C 4 7 w Fig. 4 where loops are assigned transition rates w1, 2 C w2+w4 2 w2, w3, etc. If two configurations are connected by w1 w3 more than one loop, each assigned with different tran- w4+w5 +w C w1+w2 sition rates, the total transition rate from one configura- 6 3 + w tion to another is given by sum of the transition rates. 3 w w1+w2 For example, in Fig 4, the total transition rate from 4 C C1 w 4 C2 to C1 is w(C1 C2) = (w1 + w2 + w3). Similarly, w 3 w(C C ) = (w +| w ), w(C C ) = (w + w + w ), 6 3| 1 1 2 5| 7 4 5 6 C5 w6 etc. The resulting network is shown in Fig. 5. We call C this way of assigning a transition rate (or transition prob- w 6 5 w5 +w6 ability) to a closed loop of configurations in a graph as loopwise balance. Only constraint for drawing such loops FIG. 6: Network with the reverse magnetic field −B~ : Reversal is that all nodes must be connected to each other along of the direction of the magnetic field B~ corresponds to reversal some path so that the system is ergodic. Apart from this, of all transition rates, i.e., interchanging forward and reverse loops are otherwise drawn arbitrarily. Note that since transition rate. The network for the reverse magnetic field the Markov process is ergodic, it has a unique steady −B~ results from the network of Fig. 5 by reversing all the state solution. There are several ways to connect nodes arrows. satisfying the constraint of having uniform steady state measure and, since the Markov process is ergodic, all configurations always have equal steady state probabilities. To see this, consider the Master equation for the Markov direction, i.e., just by reversing the arrows on a network process defined on a network in Fig. 4, as done in Fig. 6. Note that the transition rates assigned to loops in general depend on the magnetic field. dP (C1) = (w + w + w + w )P (C ) However the steady state distribution remains uniform dt − 1 2 3 6 1 and thus independent of the magnetic field, which is the +(w + w + w )P (C )+ w P (C ), 1 2 3 2 6 5 symmetry relation 2 considered in section III. Note that, dP (C2) = (w1 + w2 + w3)P (C2) although the Master equation changes under reversal of dt − the magnetic field, the steady state solution is still un- +w1P (C3)+ w2P (C8)+ w3P (C4), changed, i.e., all steady states are still equally probable...... dP (C7) = (w4 + w5 + w6)P (C7) dt − [1] D. J. Evans, E. G. D. Cohen and G. P. Morris, Phys. +(w5 + w6)P (C4)+ w4P (C8), Rev. Lett. 71, 2401 (1993). dP (C ) [2] D. J. Evans and D. J. Searles, Phys. Rev. E 50, 1645 8 = (w + w )P (C ) dt − 2 4 8 (1994). [3] G. Gallavotti and E. G. D. Cohen, Phys. Rev. Lett. 74, +(w + w )P (C ). (30) 2 4 3 2694 (1995). From above set of equations it is clear that all steady [4] C. Bustamante, J. Liphardt and F. Ritort, Phys. Today states have equal probabilities, i.e., P (C ) = P (C ) = 58, 43 (2005). 1 2 [5] R. J. Harris and G. M. Schütz, J. Stat. Mech.: Theor. P (C ) = = P (C ) = constant is the steady state 3 ··· 8 and Exp. P07020 (2007). solution of the Master equation. Since the network is [6] U. Seifert, Eur. Phys. J. B 64, 423 (2008). ergodic, the steady state is also unique. Therefore loop- [7] C. Jarzynski, and D. K. Wojcik, Phys. Rev. Lett. 92, wise balance is a sufficient condition for having a uniform 230602 (2004). steady state measure in an ergodic Markov process with [8] K. Saito, and A. Dhar, Phys. Rev. Lett. 99, 180601 finite number of states. (2007). If the Markov process, as defined on the network in Fig. [9] J. Kurchan, J. Phys. A 31, 3719 (1998). [10] J. L. Lebowitz and H. Spohn, J. Stat. Phys. 95, 333 5, is considered to be in the presence of a magnetic field ~ (1999). B, then the Markov process with the reverse magnetic [11] T. Hatano and S. Sasa, Phys. Rev. Lett. 86, 3463 (2001). field B~ is defined on the same network by assigning [12] C. Maes and K. Netocny, J. Stat. Phys. 110, 269 (2003). − transition rates from one node to another in the reverse [13] U. Seifert, Phys. Rev. Lett. 95, 040602 (2005). 12

[14] C. Jarzynski, Phys. Rev. Lett. 78, 2690 (1997). [32] K. Saito and Y. Utsumi, Phys. Rev. B 78, 115429 (2008). [15] S. Park and K. Schulten, J. Chem. Phys. 120, 5946 [33] A. M. Jayannavar and M. Sahoo, Phys. Rev. E 75, (2004). 032102 (2007). [16] H. Oberhofer, C. Dellago, and P. L. Geissler, J. Phys. [34] D. Roy and N. Kumar, Phys. Rev. E 78, 052102 (2008). Chem. B 109, 6902 (2005). [35] A. Saha and A. M. Jayannavar, Phys. Rev. E 77, 022105 [17] G. E. Crooks, Phys. Rev. E 60, 2721 (1999). (2008). [18] R. A. Blythe, Phys. Rev. Lett. 100, 010601 (2008). [36] R. C. Tolman, The Principles of Statistical Mechanics [19] O. Narayan and A. Dhar, J. Phys. A: Math. Gen. 37, 63 (Clarendon Press, Oxford, 1938). (2004). [37] Since some amount of work done on the system gets dis- [20] T. Mai and A. Dhar, Phys. Rev. E 75, 061101 (2007). sipated to the heat bath, there is a slight change dT in [21] T. Speck and U. Seifert, J. Stat. Mech: Theor. and Exp. temperature of the bath (although dT is inﬁnitesimal due L09002 (2007). to large size of a bath). The net free energy change of the [22] G. E. Crooks, Phys. Rev. E 61, 2361 (2000). CS equals ∆FCS = ∆FB +∆F where ∆FB = −SBdT [23] E. G. D. Cohen and D. Mauzerall, J. Stat. Mech: Theor. is a ﬁnite change in free energy of the bath and SB is and Exp. P07006 (2004). entropy of the bath [19]. [24] E. G. D. Cohen and D. Mauzerall, Molecular Physics [38] H. Goldstein, C. Poole, and J. Safko, Classical Mechan- 103, 2923 (2005). ics, (Addison-Wesley, Third Edition, 2000). [25] E. T. Jaynes, Phys. Rev. 106, 620 (1957). E. T. Jaynes, [39] N. G. van Kampen, Stochastic processes in Physics and Phys. Rev. 108, 171 (1957). Chemistry (Amsterdam: Elsevier, 1992), page 114-117. [26] R. M. L. Evans, Phys. Rev. Lett. 92, 150601 (2004). [40] S. R. de Groot and P. Mazur, Non-equilibrium Thermo- [27] R. M. L. Evans, J. Phys. A: Math. Gen. 38, 293 (2005). dynamics (Dover Publications, New York, 1884), page [28] A. Simha, R. M. L. Evans, and A. Baule, Phys. Rev. E 92-100. 77, 031117 (2008). [41] J. H. Van Vleck, The Theoty of electric and Magnetic [29] C. Jarzynski, J. Stat. Phys. 98, 77 (2000). Susceptibilities (The Clarendon Press, Oxford, 1932), [30] B. Cleuren, C. Van den Broeck and R. Kawai, Phys. Rev. page 22, page 97-99. Lett. 96, 050601 (2006). [42] M. R. Evans and T. Hanney, J. Phys. A: Math. Gen. 38, [31] P. Pradhan, Y. Kafri, and D. Levine, Phys. Rev. E 77, R195 (2005). 041129 (2008).