Crooks fluctuation theorem

Relation between forward and reverse nonequilibrium work distributions: P_F(W)/P_R(-W)=exp(β(W-ΔF)), implying Jarzynski’s equality and identifying ΔF from a crossing point.
Crooks fluctuation theorem

Setup (forward and reverse protocols)

Consider a system with control parameter λ\lambda (e.g. a trap position, field, volume) driven by a time-dependent protocol λt\lambda_t for t[0,τ]t\in[0,\tau].

  • Forward (F): start in equilibrium at λ0\lambda_0 with inverse temperature β=1/(kBT)\beta=1/(k_B T) (with TT the ), typically the .
  • Reverse (R): start in equilibrium at λτ\lambda_\tau and drive with the time-reversed protocol λ~t=λτt\tilde\lambda_t=\lambda_{\tau-t}.

For each realization, define the work WW (see ). Let PF(W)P_F(W) and PR(W)P_R(W) be the corresponding work probability densities (with respect to the relevant ).

Let ΔF=F(λτ)F(λ0)\Delta F = F(\lambda_\tau)-F(\lambda_0) denote the equilibrium free-energy difference at the same β\beta (see and ).

Theorem (Crooks)

Under microreversible dynamics (time-reversal symmetric Hamiltonian dynamics, or Markov dynamics satisfying microscopic reversibility such as with a heat bath), the work distributions satisfy

PF(W)PR(W)  =  exp ⁣(β(WΔF)). \frac{P_F(W)}{P_R(-W)} \;=\; \exp\!\bigl(\beta(W-\Delta F)\bigr).

Immediate consequences

  1. Crossing point identifies ΔF\Delta F.
    The value WW^\star where PF(W)=PR(W)P_F(W^\star)=P_R(-W^\star) satisfies W=ΔFW^\star=\Delta F.

  2. Jarzynski follows by integration.
    Multiply both sides by PR(W)eβWP_R(-W)e^{-\beta W} and integrate over WW to obtain

    eβWF  =  eβΔF, \left\langle e^{-\beta W}\right\rangle_F \;=\; e^{-\beta \Delta F},

    i.e. .

  3. Second-law inequality.
    Using Jensen’s inequality,

    WF    ΔF, \langle W\rangle_F \;\ge\; \Delta F,

    consistent with the .

Information-theoretic form (dissipation as relative entropy)

At the level of trajectories ω\omega (paths), Crooks’ theorem is equivalent to a path-space likelihood ratio

lnPF(ω)PR(ω~)=β(W(ω)ΔF), \ln\frac{\mathbb{P}_F(\omega)}{\mathbb{P}_R(\tilde\omega)} ={} \beta\bigl(W(\omega)-\Delta F\bigr),

so the mean dissipated work WΔFF\langle W-\Delta F\rangle_F is proportional to a between forward and reverse path measures.

Practical note

Crooks’ relation underlies statistically efficient estimators of ΔF\Delta F (e.g. Bennett-type methods) built from both PFP_F and PRP_R; see .