Fluctuation Theorems

Fluctuation theorems are exact statements about the probability of thermodynamic fluctuations in small systems driven away from equilibrium. They explain how the second law emerges from reversible microscopic dynamics even though individual runs can temporarily move entropy in the “wrong” direction.

The unifying idea is a comparison between a forward experiment and a suitably defined reverse experiment. If $\gamma$ denotes a trajectory or measurement record and $\tilde\gamma$ denotes its time-reversed counterpart, many fluctuation relations have the form

$$\frac{P_F[\gamma]} {P_R[\tilde\gamma]} = e^{\sigma[\gamma]},$$

where $\sigma[\gamma]$ is dimensionless entropy production. When every reverse event has a forward counterpart with nonzero probability, averaging the inverse ratio gives an integral fluctuation theorem,

$$\left\langle e^{-\sigma}\right\rangle_F = 1,$$

and Jensen’s inequality gives

$$\langle\sigma\rangle_F \ge 0.$$

Thus the ordinary second-law inequality is a consequence of a stronger probability statement. The theorem does not say negative entropy-production events are impossible. It says their probabilities are constrained relative to the corresponding reverse events.

Forward and Reverse Fluctuations

“Fluctuation theorem” is a family name, not a single formula. The right theorem depends on the variables and dynamics:

Setting	Typical random variable	Standard relation
closed driven system	work $W$	Jarzynski equality and Crooks relation
open thermal trajectory	entropy production $\sigma$	detailed and integral fluctuation theorems
nonequilibrium steady state	exchanged heat or entropy flow	steady-state fluctuation theorem
transport junction	transferred charge, particles, or energy	counting-statistics fluctuation symmetry
feedback-controlled process	work plus information	feedback-modified fluctuation relation

The shared structure is microscopic reversibility plus a clear operational definition of the measured quantity. In quantum mechanics the word “measured” is not cosmetic: projective energy measurements, weak measurements, quantum jumps, homodyne records, and full counting statistics define different probability distributions.

Entropy Production Form

A compact classical-looking statement uses dimensionless entropy production,

$$\sigma = \frac{\Delta s_{\mathrm{sys}}[\gamma]}{k_B} - \sum_\alpha \beta_\alpha Q_\alpha,$$

Here $Q_\alpha[\gamma]$ is trajectory heat entering the system, and $\Delta s_{\mathrm{sys}}/k_B=\ln(p_n^0/p_m^\tau)$ is a stochastic endpoint entropy change for the specified initial and final outcomes. It is not the change of the ensemble von Neumann entropy on each run. The corresponding reservoir entropy term is $-\beta_\alpha Q_\alpha$.

When a forward protocol and a reverse protocol are related by time reversal, a detailed fluctuation theorem has the schematic form

$$\frac{P_F(\sigma)} {P_R(-\sigma)} = e^\sigma.$$

Multiplying by $P_R(-\sigma)e^{-\sigma}$ and integrating over $\sigma$ gives

$$\left\langle e^{-\sigma}\right\rangle_F=1.$$

Jensen’s inequality then implies

$$\langle\sigma\rangle_F\ge0.$$

The support condition matters: if the reverse experiment has total probability $\lambda$ on events with no forward counterpart, then $\langle e^{-\sigma}\rangle_F=1-\lambda$. This is called absolute irreversibility. It arises, for example, in some error-free measurement and feedback limits. The forward average samples only its own support, so it cannot recover that missing reverse probability.

Work Relations

For a closed driven system initially in $e^{-\beta H_0}/Z_0$, the two-point measurement protocol assigns work $W=E_m^\tau-E_n^0$. Unitarity gives

$$\langle e^{-\beta W}\rangle_F=e^{-\beta\Delta F}, \qquad \Delta F=-\beta^{-1}\ln(Z_\tau/Z_0).$$

This Jarzynski equality needs no assumption of slow driving and no reverse experiment for its proof. At positive temperature it implies $\langle W\rangle_F\ge\Delta F$.

With the physical reverse protocol, including reversal of time-odd fields, Crooks’ relation further compares forward and reverse work probabilities:

$$P_F(W)=e^{\beta(W-\Delta F)}P_R(-W).$$

For discrete spectra this is a relation between probability masses at corresponding work values. See Jarzynski Equality and Crooks Relation for the derivations and examples.

The Quantum Measurement Issue

The two-point measurement scheme is standard because it gives a positive work distribution and a clean connection to energy conservation. It also has a cost: the first projective energy measurement removes initial coherences in the energy basis.

This is harmless when the initial state is Gibbs and therefore commutes with $H_0$. It becomes conceptually important when the initial state has energy coherence or when one wants to assign work without a strong initial measurement. Alternative quantum formulations include:

• weak-measurement and quasiprobability approaches;
• full counting statistics with counting fields;
• interferometric measurements of characteristic functions;
• quantum-jump and continuous-measurement trajectory definitions;
• inclusive system-plus-bath energy measurements.

These formulations answer different operational questions. A theorem should always state which record defines the random variable.

Open Quantum Trajectories

For a weakly coupled open system with energy-resolved jumps, usually obtained after secularization, a trajectory is a sequence of no-jump evolutions and jumps. A jump associated with bath $\alpha$ and Bohr frequency $\omega$ transfers energy

$$Q_\alpha = \hbar\omega$$

into the system when the jump raises the system energy, and $Q_\alpha=-\hbar\omega$ when it lowers the system energy. The frequency here labels the energy transferred in the microscopic reservoir coupling. A general driven or Floquet jump can exchange drive quanta as well, so it cannot automatically be identified with a bare instantaneous energy gap.

Thermal rates satisfy local detailed balance. For a transition $n\to m$ caused by bath $\alpha$,

$$\frac{W_{m\leftarrow n}^{(\alpha)}} {W_{n\leftarrow m}^{(\alpha)}} = e^{-\beta_\alpha(E_m-E_n)}$$

when $E_m-E_n$ is the energy gained by the system. This microscopic ratio is the seed of the trajectory-level entropy production.

For a trajectory beginning in state $n$ and ending in state $m$, one often writes

$$\sigma[\gamma] = \ln \frac{p_n^0}{p_m^\tau} - \sum_\alpha \beta_\alpha Q_\alpha[\gamma],$$

where the first term is the stochastic system entropy change and the second is the bath entropy change. Under the right reverse process,

$$\frac{P_F[\gamma]} {P_R[\tilde\gamma]} = e^{\sigma[\gamma]}.$$

This trajectory statement is the open-system analogue of the closed-system work relation. It relies on thermally consistent rates and a well-defined monitoring scheme. A master equation can be useful for state evolution while still failing to define a trustworthy thermodynamic trajectory ensemble if its jumps do not correspond to physical reservoir exchanges.

Counting Statistics View

In mesoscopic transport and quantum optics, fluctuation relations are often written for generating functions. If $P(N)$ is the probability of transferring $N$ particles, or $P(Q)$ the probability of transferring energy $Q$, one introduces a counting field $\chi$ and a generating function

$$Z(\chi) = \sum_N P(N)e^{i\chi N}.$$

Thermodynamic consistency appears as a symmetry of $Z(\chi)$ or of the long-time cumulant-generating function. For example, a two-reservoir particle current has an affinity built from chemical potentials and temperatures, and the fluctuation symmetry encodes the probability ratio between current flowing with and against the thermodynamic bias.

This language is especially useful when individual projective energy measurements on the system are not the most natural experimental record. It also connects fluctuation theorems to noise, response, and higher cumulants.

Relation to Linear Response

Fluctuation theorems are nonlinear nonequilibrium statements. Linear response and fluctuation-dissipation relations emerge when one expands near equilibrium.

For work, the Jarzynski equality gives a simple illustration. If $W$ is approximately Gaussian with mean $\mu$ and variance $v$, then

$$\left\langle e^{-\beta W}\right\rangle = e^{-\beta\mu+\beta^2v/2}.$$

Jarzynski’s equality therefore implies

$$\mu-\Delta F = \frac{\beta v}{2}.$$

Near equilibrium, dissipated work is tied to work fluctuations. The full theorem is stronger than this Gaussian relation, but the relation shows why equilibrium fluctuations and irreversible response are deeply linked. See Fluctuation–Dissipation Relation for the linear-response side.

Feedback and Information

When a measurement outcome is used to choose the subsequent protocol, the reverse experiment and entropy budget must include information. For a classical noninvasive measurement with errors, $I=\ln[p(y|x)/p(y)]$ compares the likelihood of outcome $y$ given microscopic state $x$ with its marginal likelihood. With initially thermal dynamics, compatible reverse protocols, common final free energy, and the support condition above, a feedback relation is

$$\left\langle e^{-\beta(W-\Delta F)-I} \right\rangle = 1,$$

For a general quantum instrument, backaction and the chosen reverse process can require different information terms. The displayed equality is not a universal identity obtained by inserting the classical mutual information into any quantum protocol. Under its stated assumptions, Jensen’s inequality gives

$$\langle W\rangle \ge \Delta F - k_B T\,\langle I\rangle.$$

This does not mean information creates free energy from nowhere. It means the controller’s record is a thermodynamic resource. A complete cycle must include the physical memory, the measurement backaction, and the cost of resetting or maintaining the controller.

For a concrete engine, see Maxwell Demon and Information. The open-system ingredients are developed in Quantum Instruments and Stochastic Master Equations.

Exercises

Detailed Implies Integral

Assume

$$\frac{P_F(\sigma)} {P_R(-\sigma)} = e^\sigma.$$

Assume the reverse distribution has no weight outside the paired forward support and derive $\langle e^{-\sigma}\rangle_F=1$.

Solution

Start from the forward average:

$$\left\langle e^{-\sigma}\right\rangle_F = \int d\sigma\, P_F(\sigma)e^{-\sigma}.$$

Using $P_F(\sigma)e^{-\sigma}=P_R(-\sigma)$,

$$\left\langle e^{-\sigma}\right\rangle_F = \int d\sigma\,P_R(-\sigma).$$

Changing variables to $\sigma'=-\sigma$ gives

$$\int d\sigma'\,P_R(\sigma')=1.$$

Jarzynski Implies the Work Bound

Use Jensen’s inequality to show that

$$\left\langle e^{-\beta W}\right\rangle = e^{-\beta\Delta F}$$

implies $\langle W\rangle\ge\Delta F$.

Solution

For the convex function $e^x$,

$$e^{\langle x\rangle} \le \langle e^x\rangle.$$

Set $x=-\beta W$. Then

$$e^{-\beta\langle W\rangle} \le e^{-\beta\Delta F}.$$

Taking logarithms and multiplying by $-1/\beta$ reverses the inequality because $-1/\beta\lt0$:

$$\langle W\rangle \ge \Delta F.$$

Missing Reverse Events

Suppose a forward experiment always gives record $a$. Its reversed record has reverse probability $0.8$; the remaining reverse probability $0.2$ has no forward counterpart. Compute $\sigma(a)$ and $\langle e^{-\sigma}\rangle_F$.

Solution

$\sigma(a)=\ln(1/0.8)$, so $\langle e^{-\sigma}\rangle_F=0.8=1-\lambda$. The missing $0.2$ cannot appear in a forward average. This is a probability example of the support restriction; interpreting $\sigma$ thermodynamically additionally requires a physical reverse process.

Feedback Bound

Assume the feedback-modified relation

$$\left\langle e^{-\beta(W-\Delta F)-I} \right\rangle = 1.$$

Use Jensen’s inequality to derive the corresponding average-work bound.

Solution

Let

$$x = -\beta(W-\Delta F)-I.$$

Jensen’s inequality gives

$$e^{\langle x\rangle} \le \left\langle e^x\right\rangle = 1.$$

Thus $\langle x\rangle\le0$, or

$$-\beta(\langle W\rangle-\Delta F) - \langle I\rangle \le 0.$$

Rearranging,

$$\langle W\rangle \ge \Delta F - k_B T\,\langle I\rangle.$$

References

• M. Campisi, P. Hänggi, and P. Talkner, “Colloquium: Quantum fluctuation relations: Foundations and applications,” Reviews of Modern Physics 83, 771–791 (2011). arXiv:1012.2268.
• G. E. Crooks, “Entropy production fluctuation theorem and the nonequilibrium work relation for free energy differences,” Physical Review E 60, 2721-2726 (1999).
• M. Esposito, U. Harbola, and S. Mukamel, “Nonequilibrium fluctuations, fluctuation theorems, and counting statistics in quantum systems,” Reviews of Modern Physics 81, 1665-1702 (2009).
• C. Jarzynski, “Nonequilibrium equality for free energy differences,” Physical Review Letters 78, 2690-2693 (1997).
• J. Kurchan, “A quantum fluctuation theorem,” arXiv:cond-mat/0007360 (2000).
• Y. Murashita, K. Funo, and M. Ueda, “Nonequilibrium equalities in absolutely irreversible processes,” Physical Review E 90, 042110 (2014). DOI:10.1103/PhysRevE.90.042110.
• T. Sagawa and M. Ueda, “Generalized Jarzynski equality under nonequilibrium feedback control,” Physical Review Letters 104, 090602 (2010).
• U. Seifert, “Stochastic thermodynamics, fluctuation theorems and molecular machines,” Reports on Progress in Physics 75, 126001 (2012).
• P. Talkner, E. Lutz, and P. Hänggi, “Fluctuation theorems: Work is not an observable,” Physical Review E 75, 050102(R) (2007).
• H. Tasaki, “Jarzynski relations for quantum systems and some applications,” arXiv:cond-mat/0009244 (2000).