Breakdown of ensemble equivalence

When microcanonical and canonical ensembles yield different macrostates, typically due to nonconcavity of microcanonical entropy, long-range interactions, or first-order transitions.
Breakdown of ensemble equivalence

“Ensemble equivalence” refers to the fact that different equilibrium ensembles (most commonly microcanonical vs canonical) give the same thermodynamic predictions in the thermodynamic limit. Breakdown means that, for some parameters, they do not.

This phenomenon is central in long-range systems and in models where the microcanonical entropy is not concave.

Microcanonical entropy and canonical free energy

Let e=E/Ve=E/V be energy density.

  • The microcanonical entropy density is

    s(e)  =  limV1VlogΩV(E),E=Ve, s(e) \;=\; \lim_{V\to\infty}\frac{1}{V}\log \Omega_V(E), \qquad E=Ve,

    as in and .

  • The canonical free energy density can be written in terms of the canonical partition function ZV(β)Z_V(\beta) (see ):

    f(β)  =  1βlimV1VlogZV(β). f(\beta) \;=\; -\frac{1}{\beta}\lim_{V\to\infty}\frac{1}{V}\log Z_V(\beta).

These are related by Legendre–Fenchel duality (see and ):

βf(β)  =  supe(s(e)βe),equivalentlyf(β)  =  infe(eβ1s(e)). -\beta f(\beta) \;=\; \sup_{e}\,\bigl(s(e)-\beta e\bigr), \qquad\text{equivalently}\qquad f(\beta) \;=\; \inf_{e}\,\bigl(e-\beta^{-1}s(e)\bigr).

What “equivalence” means (microcanonical vs canonical)

Informally, equivalence holds when:

  • the canonical energy concentrates at an energy density e(β)e(\beta), and
  • the microcanonical equilibrium at e(β)e(\beta) yields the same macroscopic observables as the canonical equilibrium at β\beta.

A clean sufficient condition is concavity of s(e)s(e) (plus mild regularity). When ss is strictly concave and differentiable, the microcanonical inverse temperature is

β  =  s(e), \beta \;=\; s'(e),

and the mapping eβe \leftrightarrow \beta is essentially one-to-one.

How breakdown happens: nonconcave entropy and “convex intruders”

If s(e)s(e) is not concave on an interval, then:

  • the canonical ensemble only “sees” the concave envelope s(e)s^{**}(e) (the biconjugate),
  • some energy densities that are admissible microcanonically are not realized canonically as typical energies.

This is one precise sense in which ensembles become inequivalent.

A geometric picture:

  • canonical equilibrium selects maximizers of s(e)βes(e)-\beta e,
  • if ss has a nonconcave region, the maximizer jumps between endpoints of that region as β\beta varies, producing phase coexistence and a first-order canonical transition (non-differentiability of ff).

Diagnostic: negative microcanonical heat capacity

Working in units where kB=1k_B=1, set T=1/βT=1/\beta with β=s(e)\beta=s'(e). Then

dTde  =  s(e)[s(e)]2,C(e)  =  (dTde)1  =  [s(e)]2s(e). \frac{dT}{de} \;=\; -\frac{s''(e)}{[s'(e)]^2}, \qquad C(e) \;=\; \left(\frac{dT}{de}\right)^{-1} \;=\; -\frac{[s'(e)]^2}{s''(e)}.

Therefore:

  • if s(e)>0s''(e)>0 (a local “convex” region), then C(e)<0C(e)<0.

Negative microcanonical heat capacity is incompatible with canonical equilibrium (where energy fluctuations enforce nonnegative heat capacity), and is a standard signature of nonequivalence; see .

Typical mechanisms and examples

  • Long-range interactions (mean-field-type spin systems, gravitational models) often generate nonconcave s(e)s(e); see .
  • Even for short-range models, finite-size systems can show apparent nonequivalence, but in many stable short-range systems equivalence is recovered as VV\to\infty.

Prerequisites