week ending

PRL 116, 120601 (2016) PHYSICAL REVIEW LETTERS 25 MARCH 2016

Dissipation Bounds All Steady-State Current Fluctuations

Todd R. Gingrich,* Jordan M. Horowitz, Nikolay Perunov, and Jeremy L. England
Physics of Living Systems Group, Department of Physics, Massachusetts Institute of Technology,
400 Technology Square, Cambridge, Massachusetts 02139, USA
(Received 15 December 2015; revised manuscript received 4 February 2016; published 21 March 2016)
Near equilibrium, small current fluctuations are described by a Gaussian distribution with a linear-
response variance regulated by the dissipation. Here, we demonstrate that dissipation still plays a dominant
role in structuring large fluctuations arbitrarily far from equilibrium. In particular, we prove a linear-
response-like bound on the large deviation function for currents in Markov jump processes. We find that
nonequilibrium current fluctuations are always more likely than what is expected from a linear-response
analysis. As a small-fluctuations corollary, we derive a recently conjectured uncertainty bound on the
variance of current fluctuations.

DOI: 10.1103/PhysRevLett.116.120601

One of the most useful insights into thermodynamics has Markov jump process with rates rðy; zÞ [22]. It is conven-
been that fluctuations near equilibrium are completely ient to picture these dynamics occurring on a graph (as in
characterized by just one principle: the fluctuation- Fig. 1), with vertices denoting states and edges (or links)
dissipation theorem [1]. Far from equilibrium, however, symbolizing possible transitions. We assume the dynamics
fluctuations exhibit less universal structure. As such, to be ergodic and that rðz; yÞ > 0 whenever rðy; zÞ > 0, so
characterizing the rich anatomy of nonequilibrium fluctua- the system’s probability density relaxes to a unique steady
tions has been handled on a case by case basis, with few state πðxÞ in the long-time limit. Thermodynamics enters
universal nonequilibrium principles. Notable exceptions by requiring the transitions to satisfy local detailed balance;
are the fluctuation theorems [2–7], as well as fluctuation- the ratio of rates on each edge can then be identified with a
dissipation theorems for nonequilibrium steady states generalized thermodynamic force
[8–12]. Recently, Barato and Seifert have proposed a

new kind of nonequilibrium principle, a thermodynamic πðyÞrðy; zÞ
Fðy; zÞ ¼ ln ; ð1Þ
uncertainty relation that expresses a trade-off between the πðzÞrðz; yÞ
variance of current fluctuations and the rate of entropy
production [13]. It reveals that away from equilibrium, which quantifies the dissipation in each transition [23].
dissipation continues to regulate small fluctuations. While For example, if a transition were mediated by a thermal
the thermodynamic uncertainty relation was not proven in reservoir at inverse temperature β, we have F ¼ Δs þ βq,
general, analytical calculations and numerical evidence where Δs ¼ − ln½πðzÞ=πðyÞ is the change in the system’s
support its validity [13]. Applications appear myriad, stochastic entropy [24] and βq ¼ ln½rðy; zÞ=rðz; yÞ is the
and already include insights into chemical kinetics as well
as biochemical sensing [14,15].
In this Letter, we demonstrate that dissipation in fact
constrains all current fluctuations. In particular, we prove a
pair of general thermodynamic inequalities for the large
deviation function of the steady-state empirical currents in
Markov jump processes. Such processes model a variety of
scenarios, including molecular motors [16], chemical
reaction networks [17,18], and mesoscopic quantum devi-
ces [19]. Our analysis reveals that far from equilibrium,
current fluctuations are always more probable than would
be predicted by linear response [20,21]. Remarkably, our
relationship bounds even rare fluctuations (large devia-
tions), and by specializing to small deviations we obtain the FIG. 1. Current fluctuations illustration: for a four-state model
thermodynamic uncertainty relation. (inset) in a nonequilibrium steady state, the integrated current
We have in mind a system with N mesoscopic states (or J T —the net number of hops between pairs of states—along each
configurations), x ¼ 1; …; N. Transitions between pairs of edge is plotted as a function of time. Each integrated current
states, say from y to z, are modeled as a continuous-time displays an average rate perturbed by stochastic fluctuations.

0031-9007=16=116(12)=120601(5) 120601-1 © 2016 American Physical Society

 week ending
PRL 116, 120601 (2016) PHYSICAL REVIEW LETTERS 25 MARCH 2016

heat dissipated into the reservoir. Here and through- The subscript weakened linear response (WLR) connotes a
out, kB ¼ 1. weakening of Eq. (3). Our analysis reveals this bound is
Now we imagine watching the system evolve for a long tightest, and indeed as strong as the linear-response bound,
time from time t ¼ 0 to T as it jumps along a sequence of when d ∝ F. Under this condition,
P the generalized current
states xðtÞ, and we measure the integrated empirical current is the dissipation rate Σ ¼ y<z jðy; zÞFðy; zÞ:
through all the links by counting the net number of
transitions along each edge ðΣ − Σπ Þ2
IðΣÞ ≤ I WLR ðΣÞ ¼ : ð5Þ
Z T 4Σπ
JT ðy; zÞ ≡ dt ðδxðt− Þ;y δxðtþ Þ;z − δxðt− Þ;z δxðtþ Þ;y Þ; ð2Þ
0 Derivations of these inequalities appear at the end of this
Letter; here, we examine their meaning and explore their
where xðt Þ denotes the state of the system just before and consequences.
after a jump. Its rate jT ðy; zÞ ¼ J T ðy; zÞ=T, the empirical Foremost, we stress that these bounds are not quadratic
current, asymptotically converges in the infinite-time truncations of I for small currents, but originate in a linear-
limit to the average steady-state value limT→∞ jT ðy; zÞ ¼ response expansion about small force or entropy produc-
jπ ðy; zÞ ≡ πðyÞrðy; zÞ − πðzÞrðz; yÞ. For long finite times, tion. We can see this, roughly, by analyzing linear-response
fluctuations in the vector of empirical currents j away from fluctuations in the entropy production rate σðy; zÞ ¼
the typical value jπ are possible, but exponentially rare, jðy; zÞFðy; zÞ. Near equilibrium, the typical fluctuations
with a probability density that satisfies a large deviation are known to be Gaussian with a variance that is twice the
principle PðJT ¼ TjÞ ∼ e−TIðjÞ [25]. The large deviation mean σ π [20,30]
rate function IðjÞ yields an extension of the central limit
theorem, quantifying not just the Gaussian fluctuations X (σðy; zÞ − σ π ðy; zÞ)2
I LR ðσÞ ¼ ð6Þ
about the typical value (jπ , which is the minimum of I), but 4σ π ðy; zÞ
y<z
also the relative likelihood of rare fluctuations. In general,
the determination of this large deviation function is X (jðy; zÞ − jπ ðy; zÞ)2
challenging and analytical expressions are limited to ¼ Fðy; zÞ: ð7Þ
particular models (e.g., Refs. [16,26–28]). y<z
4jπ ðy; zÞ
Our main result is a pair of general thermodynamic
bounds on the rate function of empirical currents. The This quadratic rate function indicates that in linear response
first is a bound for the current fluctuations in terms of the each edge supports Gaussian current fluctuations with a
rate of steady-state entropy production along each link variance regulated by the thermodynamic force. Now
σ π ðy; zÞ ¼ jπ ðy; zÞFðy; zÞ: imagine turning up the force, pushing the system away
from linear response. We would expect the typical currents
X (jðy; zÞ − jπ ðy; zÞ)2 to grow, due to both an increased bias and more activity in
IðjÞ ≤ I LR ðjÞ ¼ σ π ðy; zÞ; ð3Þ the number of jumps. Remarkably, Eq. (3) implies that
y<z 4jπ ðy; zÞ2
these effects are accompanied by an increase in the like-
lihood of rare current fluctuations, in excess of the linear-
where the sum extends over all edges. As we will argue,
response prediction.
such Gaussian fluctuations are what one would have
As a consequence of these bounds, we have a general
expected from a linear-response analysis; as such, this
proof of the conjectured thermodynamic uncertainty rela-
bound is tightest within linear response. The subscript
tion [13]. Namely, the relative uncertainty in a generalized
linear response (LR) reinforces this observation. Further,
current ϵ2d ¼ Varðjd Þ=ðjπd Þ2 , which is the variance normal-
the inequality is saturated not only at the minimum of I,
ized by the mean, verifies the inequality
j ¼ jπ , but also at the symmetric point j ¼ −jπ , as our
bound also satisfies the fluctuation theorem for cur-
ϵ2d Σπ ≥ 2: ð8Þ
rents [5,29].
Our second inequality is a weakened form of Eq. (3) for
Thus, controlling current fluctuations by reducing their
any generalized
P current expressed as a linear combination
relative uncertainty ϵ2d costs a minimal dissipation. The
jd ¼ y<z dðy; zÞjðy; zÞ. The key benefit of this weakened
inequality follows from Eq. (4) by noting that the current’s
form is that now the current fluctuations
Pare constrained by variance is obtained from the rate function as I 00 ðjπd Þ ¼
the average dissipation rate Σπ ¼ y<z jπ ðy; zÞFðy; zÞ,
1=Varðjd Þ, and that the large-deviation inequality translates
which is often easier to measure than the individual entropy to the second derivative, since I and I LR have the same
production rates σ π : minimum. A similar argument applied to Eq. (3) leads to an
uncertainty relation for the current fluctuations along each
ðjd − jπd Þ2 π
Iðjd Þ ≤ I WLR ðjd Þ ¼ Σ : ð4Þ transition, Var(jðy; zÞ) ≥ 2jπ ðy; zÞ2 =σ π ðy; zÞ, which is
4ðjπd Þ2 tighter than the bound predicted using dðu; vÞ ¼ δðu;vÞ;ðy;zÞ
120601-2
 week ending
PRL 116, 120601 (2016) PHYSICAL REVIEW LETTERS 25 MARCH 2016

FIG. 2. Generalized-current fluctuations in the four-state FIG. 3. ASEP current fluctuations: the total current rate
model: generalized currents jd were constructed randomly by function Iðjρ Þ (black dashed) is bounded by I WLR ðjρ Þ (blue)
choosing dðy; zÞ ∈ ½−1; 1Þ. All the Iðjd Þ (colored) and IðΣÞ for the L ¼ 15 ASEP. Inset: schematic of the L ¼ 15 ASEP with
(dashed black) fall in the blue-shaded region that satisfies the open boundaries. Particles (shaded circles) jump to neighboring
bound (4), with the differences Δ ¼ Iðjd Þ − I WLR ðjd Þ plotted in sites with rates p, q in the bulk, and with rates given by Greek
the inset. Rates: rð1; 2Þ ¼ 3, rð1; 3Þ ¼ 10, rð1; 4Þ ¼ 9, letters in and out of the boundary reservoirs (shaded semicircles).
rð2; 1Þ ¼ 10, rð2; 3Þ ¼ 1, rð2; 4Þ ¼ 2, rð3; 1Þ ¼ 6, rð3; 2Þ ¼ 4, Blocked transitions are marked by a red “x.” Results are shown
rð3; 4Þ ¼ 1, rð4; 1Þ ¼ 7, rð4; 2Þ ¼ 9, rð4; 3Þ ¼ 5. using the parameters α ¼ 1.25, β ¼ 0.5, γ ¼ 0.5, δ ¼ 1.5, p ¼ 1,
and q ¼ 0.5, corresponding to a high-density phase [33].

in Eq. (8). Furthermore, our derivation shows that these edges. For even moderately large L, it is impractical to
uncertainty relations are tightest in linear response and when record the average entropy production rates across all the
d ∝ F, as predicted by Barato and Seifert [13]. edges, but we may more easily measure the mean dissipation
To illustrate the generalized-current bound (4) we Σπ . Indeed, the ASEP only has one generalized force
numerically evaluate the rate function for two toy models: conjugate to the total particle current jρ across the system:
a four-state model and the 1D asymmetric exclusion Σπ ¼ jρ f ρ with f ρ ¼ ln½αβpL−1 =ðγδqL−1 Þ=ðL þ 1Þ. This
process (ASEP) with open boundary conditions. proportionality of Σπ and jρ ensures that the WLR bound
Our first example is the multicyclic four-state graph is equivalent to the stronger LR bound, explaining the
depicted in Fig. 1. The rate functions for the dissipation Σ tightness observed in Fig. 3.
as well as for a collection of random generalized currents jd To summarize, dissipation constrains near-equilibrium
were numerically computed using standard methods [31] current fluctuations, which in turn bound far-from-
and plotted in Fig. 2. As required by our bound, all the Iðjd Þ equilibrium fluctuations. Thus, reducing current fluctua-
fall below I WLR ðjd Þ (within the blue-shaded region). tions carries a fundamental energetic cost. This observation
Interestingly, some generalized currents lie much closer suggests a design principle: for fixed average dissipation
to the bound than others. In particular, the rate function for and current, the fluctuations are most suppressed in a near-
the dissipation IðΣÞ (black dashed) saturates the bound at equilibrium process. Such a principle may aid in engineer-
Σ ¼ Σπ . Consequently, the bound is significantly tighter ing complex systems and understanding energy or accuracy
for dissipation than for the other generalized currents, trade-offs in biological physics [34]. For instance, suppose
illustrating that the tightness of the bound is quite sensitive we seek to construct a precise nonequilibrium process to
to the choice of d. reliably generate a current, e.g., a biochemical reaction
While the generalized-current bound (4) is not our network that produces a target molecule at a desired rate.
strongest, we emphasize the important benefit that it avoids One can introduce energy-consuming metabolic cycles in
the computation of edgewise entropy production rates. This an attempt to attenuate fluctuations. However, with a fixed
advantage is especially profitable in many-particle dynam- energy budget, it is impossible to surpass the linear-
ics. To illustrate this point, we consider the current response bound, no matter how complex the design.
fluctuations in a canonical model of nonequilibrium par- Derivation.—To obtain Eqs. (3) and (4), we begin with
ticle transport, the 1D ASEP [33]. The model consists of L the level 2.5 large deviations for continuous-time Markov
sites, occupied by at most one particle. Particles hop into processes [35]. This rate function describes the joint
unoccupied neighboring sites with rates p, q, and enter or fluctuations for theRempirical current jT with the empirical
leave from two boundary particle reservoirs with rates α, β, density pT ðyÞ ≡ T1 0T dt δxðtÞ;y :
γ, δ, as drawn in Fig. 3. The many-particle dynamics could X
be cast as a single-body dynamics on a graph, as in our Iðp; jÞ ¼ Ψ(jðy; zÞ; jp ðy; zÞ; ap ðy; zÞ); ð9Þ
first example, but the graph has 2L vertices and ðL þ 3Þ2L−2 y<z

120601-3
 week ending
PRL 116, 120601 (2016) PHYSICAL REVIEW LETTERS 25 MARCH 2016

where
≤ inf fI LR ðjÞjhd; ji ¼ jd ; hhk ; ji ¼ 0 ∀ kg; ð15Þ

 j
j j̄
Ψðj; j̄; aÞ ¼ j arcsinh − arcsinh since the infimum respects inequality. As I LR is a quadratic
a a
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi form with N þ 1 linear constraints, the minimization can be
− a2 þ j2 − a2 þ j̄2 ; ð10Þ performed analytically; the solution is complicated and is
presented below. Inequality (4), however, follows readily
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi once we observe that an upper bound on the infimum in
ap ðy; zÞ ≡ 2 pðyÞpðzÞrðy; zÞrðz; yÞ, and jp ðy; zÞ ≡ Eq. (15) can be obtained by evaluating I LR at any j ¼ j
pðyÞrðy; zÞ − pðzÞrðz; yÞ is the mean current associated that satisfies the constraints. We choose j ¼ ðjd =jπd Þjπ ,
with the empirical density [29,36,37]. This expression which being proportional to the steady-state current must
for Iðp; jÞ applies only for conservative currents, where
P be conservative and trivially satisfies hd; j i ¼ jd .
z jðy; zÞ ¼ 0 for all y; otherwise, Iðp; jÞ is infinite. Evaluating I LR ðj Þ gives the weaker quadratic bound (4).
We obtain Eq. (3) in two steps: an application of the Finally, we demonstrate that Eq. (4) is indeed tightest
large-deviation contraction principle followed by a simple when d ∝ F by minimizing Eq. (15) directly with N þ 1
inequality. First, we turn Iðp; jÞ into a rate function for just Lagrange multipliers to impose the constraints. The sol-
j by using the contraction principle [25], which states that ution can be expressed compactly in terms of the pseu-
IðjÞ ¼ inf p Iðp; jÞ. We can then bound the infimum by doinverse ofP a symmetric square matrix with dimension
evaluating Iðp; jÞ at any normalized density pðxÞ. The N þ1, Bkl ¼ y<z ½jπ ðy;zÞ2 =σ π ðy;zÞhk ðy;zÞhl ðy;zÞ (where
interesting choice is the steady state πðxÞ: for notational convenience we have set h0 ≡ d), as
X
IðjÞ ≤ Ψ(jðy; zÞ; jπ ðy; zÞ; aπ ðy; zÞ): ð11Þ ðjd − jπd Þ2 −1
y<z Iðjd Þ ≤ I LR ðjd Þ ¼ B00 : ð16Þ
4
Next, we bound each Ψ with a quadratic This inequality represents the LR bound contracted to the

 generalized scalar current jd . B, which determines the values
ðj − j̄Þ2 j̄
Ψðj; j̄; aÞ ≤ 2j̄arcsinh ; ð12Þ of the Lagrange multipliers associated with the constrained
4j̄2 a minimum, depends on the choice of generalized current.
Generally, B−100 is onerous to compute, but it is simple in the
which can be verified by confirming that the difference special case when d ∝ F. Using the result B−1 π
00 ¼ 1=Σ , we
between the right- and left-hand sides reaches its minimal find that the WLR and LR bounds for the entropy production
value of zero at j ¼ j̄. We arrive at Eq. (3) by finally fluctuations coincide, confirming the tightness of the WLR
recognizing that σ π ¼ 2jπ arcsinhðjπ =aπ Þ. bound for entropy production.
Armed with this derivation, we can more clearly identify This derivation sets the stage for exploring related
the physical interpretation of I LR as a linear-response aspects of nonequilibrium fluctuations, such as empirical
bound. We expand the contribution of each edge to density fluctuations [38], the impact of dynamical phase
Eq. (11) in terms of a small force F by utilizing the transitions [39], and the role of the activity [36,40].
relationship aπ ¼ jπ = sinhðF=2Þ:
We gratefully acknowledge Robert Marsland for helpful
ðj − jπ Þ2 (j2 − ðjπ Þ2 )2 2 conversations. We thank Raphael Chetrite and an anony-
Ψðj; jπ ; aπ Þ ¼ π F− F þ OðF3 Þ:
4j 192ðjπ Þ3 mous referee for identifying an error in an earlier draft of
this work. This research is funded by the Gordon and Betty
ð13Þ
Moore Foundation to T. R. G. as a Physics of Living
The first order term describes exactly the predicted Systems Fellow through Grant No. GBMF4513, as well
Gaussian linear response fluctuations. All higher order as to J. M. H. and J. L. E. through Grant No. GBMF4343.
corrections must have a negative sum.
The generalized current inequality in Eq. (4) follows Note added.—Recently, we became aware of the preprint
from Eq. (3) by contraction. First, let us introduce a by Pietzonka, Barato, and Seifert [40], which conjectures
notation for the inner Eq. (4) as a universal bound. Their work complements ours
P product on the vector space of in that it provides additional analytical calculations for
currents, hf; gi ¼ y<z fðy; zÞgðy; zÞ. With this notation
the generalized current is jd ¼ hd; ji, and the current special cases and extensive numerical support.
conservation constraints can be expressed by defining
hk ða; bÞ ¼ δka − δkb , so that hhk ; ji ¼ 0. We thus have,
*
by the contraction principle, toddging@[Link]
[1] R. Kubo, M. Toda, and N. Hashitsume, Statistical Physics
Iðjd Þ ¼ inf fIðjÞjhd; ji ¼ jd ; hhk ; ji ¼ 0 ∀ kg ð14Þ II: Nonequilibrium Statistical Mechanics, Vol. 31 (Springer
j Science & Business Media, 2012).

120601-4
 week ending
PRL 116, 120601 (2016) PHYSICAL REVIEW LETTERS 25 MARCH 2016

[2] J. Kurchan, Fluctuation theorem for stochastic dynamics, [22] N. van Kampen, Stochastic Processes in Physics and
J. Phys. A 31, 3719 (1998). Chemistry, Vol. 1 (Elsevier, New York, 1992).
[3] G. E. Crooks, Entropy production fluctuation theorem and [23] M. Esposito and C. Van den Broeck, Three faces of the
the nonequilibrium work relation for free energy second law. I. Master equation formulation, Phys. Rev. E 82,
differences, Phys. Rev. E 60, 2721 (1999). 011143 (2010).
[4] J. L. Lebowitz and H. Spohn, A Gallavotti–Cohen-type [24] U. Seifert, Stochastic thermodynamics, fluctuation theorems
symmetry in the large deviation functional for stochastic and molecular machines, Rep. Prog. Phys. 75, 126001
dynamics, J. Stat. Phys. 95, 333 (1999). (2012).
[5] D. Andrieux and P. Gaspard, Fluctuation theorem for [25] H. Touchette, The large deviation approach to statistical
currents and Schnakenberg network theory, J. Stat. Phys. mechanics, Phys. Rep. 478, 1 (2009).
127, 107 (2007). [26] R. J. Harris, A. Rákos, and G. M. Schütz, Current fluctua-
[6] R. Chetrite and K. Gawędzki, Fluctuation relations for tions in the zero-range process with open boundaries, J. Stat.
diffusion processes, Commun. Math. Phys. 282, 469 (2008). Mech.: Theory Exp. (2005) P08003.
[7] C. Jarzynski, Equalities and inequalities: Irreversibility and [27] B. Derrida, Non-equilibrium steady states: fluctuations and
the second law of thermodynamics at the nanoscale, Annu. large deviations of the density and of the current, J. Stat.
Rev. Condens. Matter Phys. 2, 329 (2011). Mech.: Theory Exp. (2007) P07023.
[8] T. Harada and S.-i. Sasa, Equality Connecting Energy [28] M. Gorissen, A. Lazarescu, K. Mallick, and C. Vanderzande,
Dissipation with a Violation of the Fluctuation-Response Exact Current Statistics of the Asymmetric Simple Exclusion
Relation, Phys. Rev. Lett. 95, 130602 (2005). Process with Open Boundaries, Phys. Rev. Lett. 109, 170601
[9] T. Speck and U. Seifert, Restoring a fluctuation-dissipation (2012).
theorem in a nonequilibrium steady state, Europhys. Lett. [29] L. Bertini, A. Faggionato, and D. Gabrielli, Flows, currents,
74, 391 (2006). and cycles for Markov chains: Large deviation asymptotics,
[10] M. Baiesi, C. Maes, and B. Wynants, Fluctuations and Stoch. Proc. Appl. 125, 2786 (2015).
Response of Nonequilibrium States, Phys. Rev. Lett. 103, [30] T. Speck and U. Seifert, Distribution of work in isothermal
010602 (2009). nonequilibrium processes, Phys. Rev. E 70, 066112
[11] J. Prost, J.-F. Joanny, and J. M. R. Parrondo, Generalized (2004).
Fluctuation-Dissipation Theorem for Steady-State Systems, [31] Iðjd Þ is calculated as the Legendre transform of the scaled
Phys. Rev. Lett. 103, 090601 (2009). cumulant generating function ψðzÞ ¼ limT→∞ ð1=TÞ×
[12] U. Seifert and T. Speck, Fluctuation-dissipation theorem in lnhezTjd i, which was obtained numerically as the
nonequilibrium steady states, Europhys. Lett. 89, 10007 (2010). maximum (real) eigenvalue of a tilted rate matrix
[13] A. C. Barato and U. Seifert, Thermodynamic Uncertainty [4,25,32].
Relation for Biomolecular Processes, Phys. Rev. Lett. 114, [32] V. Lecomte, C. Appert-Rolland, and F. van Wijland,
158101 (2015). Thermodynamic formalism for systems with Markov dy-
[14] A. C. Barato and U. Seifert, Universal bound on the Fano namics, J. Stat. Phys. 127, 51 (2007).
factor in enzyme kinetics, J. Phys. Chem. B 119, 6555 [33] T. Chou, K. Mallick, and R. Zia, Non-equilibrium statistical
(2015). mechanics: from a paradigmatic model to biological trans-
[15] A. C. Barato and U. Seifert, Dispersion for two classes of port, Rep. Prog. Phys. 74, 116601 (2011).
random variables: General theory and application to infer- [34] G. Lan, P. Sartori, S. Neumann, V. Sourjik, and Y. Tu, The
ence of an external ligand concentration by a cell, Phys. Rev. energy–speed–accuracy trade-off in sensory adaptation,
E 92, 032127 (2015). Nat. Phys. 8, 422 (2012).
[16] B. Altaner, A. Wachtel, and J. Vollmer, Fluctuating currents [35] A. C. Barato and R. Chetrite, A formal view on level 2.5
in stochastic thermodynamics. II. Energy conversion and large deviations and fluctuation relations, J. Stat. Phys. 160,
nonequilibrium response in kinesin models, Phys. Rev. E 1154 (2015).
92, 042133 (2015). [36] C. Maes and K. Netočný, Canonical structure of dynamical
[17] H. Qian and D. A. Beard, Thermodynamics of stoichio- fluctuations in mesoscopic nonequilibrium steady states,
metric biochemical networks in living systems far from Europhys. Lett. 82, 30003 (2008).
equilibrium, Biophys. Chem. 114, 213 (2005). [37] L. Bertini, A. Faggionato, and D. Gabrielli, Large
[18] T. Schmiedl and U. Seifert, Stochastic thermodynamics of deviations of the empirical flow for continuous time Markov
chemical reaction networks, J. Chem. Phys. 126, 044101 chains, Ann. Inst. H. Poincaré Probab. Statist. 51, 867
(2007). (2015).
[19] M. Esposito, K. Lindenberg, and C. Van den Broeck, [38] M. D. Donsker and S. S. Varadhan, Asymptotic Evaluation
Thermoelectric efficiency at maximum power in a quantum of certain Markov process expectations for large time, I,
dot, Europhys. Lett. 85, 60010 (2009). Commun. Pure Appl. Math. 28, 1 (1975).
[20] C. Maes and K. Netočný, Minimum entropy production [39] J. P. Garrahan, R. L. Jack, V. Lecomte, E. Pitard, K. van
principle from a dynamical fluctuation law, J. Math. Phys. Duijvendijk, and F. van Wijland, Dynamical First-Order
(N.Y.) 48, 053306 (2007). Phase Transition in Kinetically Constrained Models of
[21] C. Maes, K. Netočný, and B. Wynants, On and beyond Glasses, Phys. Rev. Lett. 98, 195702 (2007).
entropy production: the case of Markov jump processes, [40] P. Pietzonka, A. C. Barato, and U. Seifert, Universal bounds
Markov Proc. Relat. Fields 14, 445 (2008). on current fluctuations, arXiv:1512.01221.

120601-5