Statistical Mechanics
Jarzynski Equality: Free Energy From Non-Equilibrium Work
Pull on a single RNA hairpin with optical tweezers, yank it apart in a few milliseconds, and every time you repeat the experiment you spend a different amount of work — sometimes more, sometimes less, scattered over a spread of several kBT. In 1997 Christopher Jarzynski proved something that looked impossible: the exponential average of that irreversible, fluctuating work equals the equilibrium free-energy difference exactly, no matter how violently fast you pull.
The Jarzynski equality states that ⟨exp(−W/kBT)⟩ = exp(−ΔF/kBT), where W is the work done on a system in a driven, non-equilibrium process, ΔF is the difference in Helmholtz free energy between the initial and final equilibrium states, kB is Boltzmann's constant, and the angle brackets denote an average over infinitely many repetitions. It is an exact identity — not an inequality, not an approximation — that recovers the reversible thermodynamic answer from arbitrarily irreversible measurements.
- TypeExact non-equilibrium work relation (identity)
- DiscoveredChristopher Jarzynski, 1997
- Key equation⟨exp(−W/kBT)⟩ = exp(−ΔF/kBT)
- RegimeAny driving speed, from reversible to arbitrarily fast
- Typical scaleWorks ~ a few kBT; kBT ≈ 4.1 pN·nm ≈ 0.026 eV at 300 K
- First tested inRNA hairpin pulling, Liphardt et al., 2002
Interactive visualization
Press play, or step through manually. The visualization is yours to drive — try it before reading on.
Watch the 60-second explainer
A condensed visual walkthrough — narrated, captioned, under a minute.
What the Jarzynski Equality Actually Says
Consider a system in contact with a heat bath at temperature T, initially sitting in thermal equilibrium. An external agent then changes a control parameter λ — the position of an optical trap, the length of a stretched molecule, the volume of a gas — along a fixed protocol from λ0 to λ1 over a finite time. The work W done during this drive is a random variable: because the system starts from a thermally sampled microstate and the bath jostles it, each realization yields a different W.
The equality asserts:
- ⟨e−βW⟩ = e−βΔF, with β = 1/(kBT)
- ΔF = F(λ1) − F(λ0) is the equilibrium free-energy difference between the two endpoint states
- The average runs over the full ensemble of trajectories
The remarkable point is that ΔF is an equilibrium quantity, yet W is measured in a process that never comes to rest. Fast, dissipative pulling and slow, reversible pulling give the same exponential average. Convexity of the exponential recovers the familiar second law ⟨W⟩ ≥ ΔF as a corollary.
How It Is Derived
The cleanest derivation follows a single Hamiltonian trajectory. Let the system evolve under time-dependent Hamiltonian H(x, λ(t)) as λ is driven from λ0 to λ1. Work along a trajectory is W = ∫ (∂H/∂λ) λ̇ dt, which by energy bookkeeping equals the change in energy minus the heat exchanged with the bath.
- Start the ensemble from the canonical distribution ρ0(x) = e−βH(x,λ0)/Z0.
- Average e−βW over initial conditions and stochastic bath noise.
- The key step: e−βW weighting exactly cancels the trajectory's phase-space compression, so the ensemble average collapses to Z1/Z0.
Since F = −kBT ln Z, the ratio Z1/Z0 = e−βΔF. The result holds for Hamiltonian, Langevin, and Markov-jump dynamics alike, provided the dynamics preserves the instantaneous equilibrium distribution when λ is held fixed. Crucially, the system need not stay in equilibrium during the drive — only the starting state and the reference endpoint free energy are equilibrium objects.
Key Quantities and a Worked Example
The natural energy unit is kBT. At T = 300 K, kBT ≈ 4.14 × 10−21 J ≈ 4.14 pN·nm ≈ 0.0257 eV. Molecular pulling experiments deposit works of order 1–100 kBT, so the exponential average is dominated by rare trajectories where W happens to fall below ΔF.
- Dissipated work: Wdiss = ⟨W⟩ − ΔF ≥ 0. In a near-equilibrium Gaussian regime, ⟨e−βW⟩ = e−β(⟨W⟩ − βσ²/2), giving Wdiss = βσ²/2 where σ² is the work variance.
- Example: Unfold an RNA hairpin, ΔF ≈ 60 kBT ≈ 0.25 aJ. Pull fast and ⟨W⟩ ≈ 65 kBT with σ ≈ 3 kBT. The trajectories that matter for the average sit near W ≈ ⟨W⟩ − βσ² ≈ 56 kBT — improbable events, so you need many pulls.
This exponential sensitivity is the practical curse: the number of trajectories needed grows roughly as eβWdiss, so keeping dissipation below ~4 kBT is essential for convergence.
How It Is Measured and Used
The first clean experimental test came in 2002 when Liphardt, Dumont, Smith, Tinoco, and Bustamante used optical tweezers to repeatedly unfold and refold a single P5abc RNA hairpin, recovering ΔF within a fraction of a kBT even from strongly irreversible pulls. Later work by Collin et al. (2005) verified the tighter Crooks fluctuation theorem on RNA.
- Single-molecule biophysics: optical/magnetic tweezers and AFM measure force–extension curves; integrating force over extension gives W for each trajectory.
- Free-energy computation: molecular dynamics uses "fast-switching" or steered-MD to estimate ΔF and ligand-binding free energies without full equilibrium sampling.
- Estimator care: naive exponential averaging is biased; practitioners use the Bennett acceptance ratio combining forward and reverse work, or cumulant/Gaussian corrections.
The equality also underlies measurements of the free-energy landscape of proteins, DNA, and molecular motors, and connects to information-thermodynamics tests using colloidal particles and electronic single-electron boxes.
Relation to the Second Law and Fluctuation Theorems
The Jarzynski equality sits inside a family of exact non-equilibrium results. Taking a Jensen's-inequality bound of the exponential gives ⟨W⟩ ≥ ΔF — the second law of thermodynamics re-derived, now with a mechanism for the occasional "violation": individual trajectories can have W < ΔF, and they must, or the average could never equal e−βΔF.
- Crooks (1999): the fluctuation theorem PF(+W)/PR(−W) = eβ(W−ΔF) is more detailed; integrating it reproduces Jarzynski exactly.
- Reversible limit: as driving slows, the work distribution narrows onto W = ΔF and the equality becomes a plain equality.
- Steady-state fluctuation theorems (Evans–Searles, Gallavotti–Cohen): govern entropy production in driven steady states rather than transitions between equilibria.
Unlike the coarse second-law inequality, these are trajectory-level statements: they quantify precisely how often and how far the second law is transiently 'broken' at the nanoscale.
Significance, Subtleties, and Open Questions
The Jarzynski equality reframed a century-old view: thermodynamics is not only about slow, reversible idealizations but also encodes exact constraints on wild, fast processes. It launched the field of stochastic thermodynamics and made free energies experimentally accessible one molecule at a time.
- Convergence problem: exponential averages are dominated by rare small-W events; poor sampling systematically overestimates ΔF. This remains the central practical limitation.
- Definition of work: subtle debates (Jarzynski vs. inclusive/exclusive work, and the quantum case) hinge on how W is defined when the coupling to the drive is strong.
- Quantum extensions: the two-projective-measurement scheme (Tasaki, Kurchan) reproduces the equality; measuring it directly is hard because it requires energy measurements that disturb coherence.
Landmark verifications include Hummer & Szabo's 2001 potential-of-mean-force reconstruction and dozens of colloidal-bead experiments. The equality is now a textbook cornerstone and a routine tool in computational drug design, yet efficient, unbiased estimation of ΔF from finite data is still an active research frontier.
| Relation | Statement | Type | Regime of validity |
|---|---|---|---|
| Second law (Clausius) | ⟨W⟩ ≥ ΔF | Inequality | Any process; average work |
| Jarzynski equality | ⟨e^(−W/kBT)⟩ = e^(−ΔF/kBT) | Exact identity | Any driving rate; start in equilibrium |
| Crooks fluctuation theorem | P_F(W)/P_R(−W) = e^((W−ΔF)/kBT) | Detailed identity | Microscopically reversible dynamics |
| Reversible limit | W = ΔF (no dissipation) | Equality | Quasi-static, infinitely slow driving |
| Bennett acceptance ratio | Combines forward + reverse work | Estimator | Free-energy computation from both directions |
Frequently asked questions
What is the Jarzynski equality in simple terms?
It says that if you repeatedly do work on a system by driving it from one equilibrium state to another — however fast and irreversibly — the exponential average of that work, ⟨e^(−W/kBT)⟩, exactly equals e^(−ΔF/kBT). In plain language, you can recover the reversible free-energy difference ΔF from measurements taken far out of equilibrium, as long as you average over many repetitions.
Does the Jarzynski equality violate the second law of thermodynamics?
No — it contains and refines the second law. Averaging the exponential with Jensen's inequality gives ⟨W⟩ ≥ ΔF, exactly the second law. What is new is that individual trajectories can transiently have W < ΔF; these rare 'violations' are required for the equality to hold and vanish on average, so the macroscopic second law is never broken.
Who discovered the Jarzynski equality and when?
Christopher Jarzynski published it in 1997 in Physical Review Letters while at Los Alamos National Laboratory. Gavin Crooks generalized it into the Crooks fluctuation theorem in 1998–1999, and the two results together form the backbone of modern non-equilibrium work relations.
How is the Jarzynski equality tested experimentally?
The classic test uses optical tweezers to pull single RNA or DNA molecules, measuring the work for each stretch-and-release cycle. Liphardt et al. verified it on an RNA hairpin in 2002, recovering the folding free energy to within a fraction of kBT even from fast, dissipative pulls. Colloidal beads in laser traps and single-electron devices provide other tests.
Why is the exponential average hard to compute in practice?
Because e^(−W/kBT) is dominated by rare trajectories where W dips below ⟨W⟩. The number of samples needed grows roughly exponentially in the dissipated work Wdiss = ⟨W⟩ − ΔF. If Wdiss exceeds a few kBT, finite data systematically overestimates ΔF, so experimenters keep pulling slow enough that dissipation stays small or use bidirectional estimators like the Bennett acceptance ratio.
How does the Jarzynski equality relate to the Crooks fluctuation theorem?
Crooks is more detailed: it relates the full work distribution of a forward process to that of the time-reversed process, PF(W)/PR(−W) = e^((W−ΔF)/kBT). Integrating the Crooks relation over all W immediately yields the Jarzynski equality, so Jarzynski is a corollary. Crooks also gives a convenient graphical way to find ΔF: it is the work value where forward and reverse distributions cross.