Microcanonical ensemble

Intuition [Beginner]

An isolated system has fixed energy $E$, particle number $N$, and volume $V$. Different arrangements of all particle positions and momenta — different microstates — are consistent with these constraints. The microcanonical ensemble postulates that every accessible microstate is equally likely.

The number of accessible microstates is $\Omega (E,V,N)$. Boltzmann's entropy is

$$S=k_B\ln \Omega .$$

This is entropy at its root — not "disorder" but a count of possibilities. A system with $\Omega =1$ has $S=0$. A system with $\Omega =10^{10^{23}}$ has enormous entropy. More microstates means higher entropy, and high-$\Omega$ macrostates dominate: that is why entropy increases.

Visual [Beginner]

Consider two coins. Each can be heads (H) or tails (T). There are four microstates:

Microstate	Coin 1	Coin 2
1	H	H
2	H	T
3	T	H
4	T	T

The macrostate "both same" (HH or TT) has $\Omega =2$. The macrostate "one heads, one tails" (HT or TH) also has $\Omega =2$. With two coins every macrostate is equally likely.

Now scale to $N$ coins. The macrostate "all heads" has $\Omega =1$. The macrostate "half heads" has $\Omega =\left(\genfrac{}{}{0ex}{}{N}{N/2}\right)\approx 2^N/\sqrt{N}$, which is astronomically larger. For $N=10^{23}$ particles the "roughly half heads" macrostate overwhelms everything else — its $\Omega$ is so large that any other macrostate is never observed.

Worked example [Beginner]

A room of $N=4$ students, each holding either a red or blue card. Count the microstates for each macrostate.

Macrostate: 4 red. Only one arrangement: RRRR. So $\Omega =1$ and $S=k_B\ln 1=0$.

Macrostate: 3 red, 1 blue. The blue card can be with student 1, 2, 3, or 4. So $\Omega =4$ and $S=k_B\ln 4$.

Macrostate: 2 red, 2 blue. Choose any 2 of 4 students for blue: $\Omega =\left(\genfrac{}{}{0ex}{}{4}{2}\right)=6$ and $S=k_B\ln 6$.

The 2-red-2-blue macrostate is the most likely — it has the most microstates and the highest entropy. Scale to $10^{23}$ particles and this bias becomes absolute.

Check your understanding [Beginner]

Formal definition [Intermediate+]

The microcanonical ensemble describes an isolated system with fixed total energy $E$, fixed particle number $N$, and fixed volume $V$. The fundamental postulate of statistical mechanics (the equal a priori probability hypothesis) states: every microstate compatible with these constraints is equally probable.

For a classical system of $N$ particles in three dimensions, each microstate is a point $({\mathbf{q}}_1,\dots ,{\mathbf{q}}_N,{\mathbf{p}}_1,\dots ,{\mathbf{p}}_N)$ in the $6N$-dimensional phase space $\Gamma$. The accessible region is the energy shell

$${\Gamma }_E=\{(\mathbf{q},\mathbf{p})\in \Gamma :H(\mathbf{q},\mathbf{p})=E\},$$

where $H$ is the Hamiltonian 09.04.02 pending. The phase-space average of any observable $A$ is

$$⟨A⟩=\frac{1}{\Omega (E)}{\int }_{{\Gamma }_E}A(\mathbf{q},\mathbf{p})\frac{d\Sigma }{|\nabla H|},$$

where $d\Sigma$ is the surface element on ${\Gamma }_E$ and $\nabla H$ is the gradient in the $6N$-dimensional phase space. The normalisation factor is the density of states:

$$\Omega (E)={\int }_{{\Gamma }_E}\frac{d\Sigma }{|\nabla H|}.$$

An equivalent (and for many calculations more convenient) definition uses the phase-space volume enclosed by the energy surface:

$$\Sigma (E)={\int }_{H\le E}{d}^{3N}q{d}^{3N}p.$$

The density of states is then $\Omega (E)=\partial \Sigma /\partial E$.

Boltzmann entropy. The entropy of the microcanonical ensemble is

$$\boxed{S(E,V,N)=k_B\ln \Omega (E,V,N).}$$

The logarithm makes entropy extensive (additive for independent subsystems) while $\Omega$ is multiplicative: if system $A$ has ${\Omega }_A$ microstates and independent system $B$ has ${\Omega }_B$, the combined system has ${\Omega }_{AB}={\Omega }_A{\Omega }_B$ and $S_{AB}=k_B\ln ({\Omega }_A{\Omega }_B)=S_A+S_B$.

Temperature. The thermodynamic temperature is defined by

$$\frac{1}{T}={\left(\frac{\partial S}{\partial E}\right)}_{V,N}.$$

Pressure.

$$\frac{P}{T}={\left(\frac{\partial S}{\partial V}\right)}_{E,N}.$$

Chemical potential.

$$\frac{\mu }{T}=-{\left(\frac{\partial S}{\partial N}\right)}_{E,V}.$$

These three relations are the bridge from the microcanonical ensemble to thermodynamics. Given $S(E,V,N)$, all thermodynamic quantities follow by differentiation.

The ideal gas. For $N$ non-interacting particles of mass $m$ in volume $V$, the phase-space volume enclosed by energy $E$ is

$$\Sigma (E,V,N)=\frac{V^N}{N!h^{3N}}\frac{(2\pi mE{)}^{3N/2}}{\Gamma (3N/2+1)},$$

where $h$ is a constant with dimensions of action (setting the phase-space measure; its physical value is Planck's constant) and $N!$ corrects for indistinguishable particles (Gibbs' correction). Differentiating:

$$\Omega (E)=\frac{\partial \Sigma }{\partial E}=\frac{V^N}{N!h^{3N}}\frac{(2\pi m{)}^{3N/2}}{\Gamma (3N/2)}{E}^{3N/2-1}.$$

The entropy is

$$S=k_B\ln \Omega =k_B\left[\frac{3N}{2}\ln E+N\ln V+\text{const}\right],$$

where "const" absorbs all the $N$- and $m$-dependent terms that do not vary with $E$ or $V$.

Equipartition theorem (microcanonical derivation). Consider the average of $p_1^2/(2m)$ in the microcanonical ensemble. By the spherical symmetry of the momentum integral in $\Sigma (E)$, each of the $3N$ quadratic momentum degrees of freedom contributes equally to the total kinetic energy. Since the total kinetic energy equals $E$ (to leading order in $N$), each degree of freedom carries average energy $E/(3N)$. Setting $E=\frac{3}{2}N{k}_{B}T$ from $1/T=\partial S/\partial E$ gives the equipartition result: each quadratic degree of freedom has average energy $\frac{1}{2}{k}_{B}T$.

The virial theorem. For a system with Hamiltonian $H={\sum }_i{p}_i^2/(2m_i)+V({\mathbf{q}}_1,\dots ,{\mathbf{q}}_N)$, the virial theorem states

$$2⟨T⟩=-⟨\sum _{i=1}^N{\mathbf{q}}_i\cdot {\mathbf{F}}_i⟩,$$

where ${\mathbf{F}}_i=-{\nabla }_{{\mathbf{q}}_i}V$ is the force on particle $i$. The proof follows from integrating ${\sum }_i{\mathbf{q}}_i\cdot {\mathbf{p}}_i$ over the energy shell and applying integration by parts — the boundary term vanishes because $\Omega (E)$ is computed on a closed surface. For power-law potentials $V\propto r^n$, the virial theorem gives $⟨T⟩=(n/2)⟨V⟩$.

Counterexamples to common slips

• $\Omega (E)$ is not dimensionless. The density of states has dimensions of (energy)${}^{-1}$ $\times$ (phase-space volume). The entropy $S=k_B\ln \Omega$ requires $\Omega$ to be dimensionless, achieved by the factor $h^{3N}$ in the denominator — $h$ sets the phase-space unit. Different choices of $h$ shift $S$ by an additive constant; in the thermodynamic limit ($N\to \infty$) this constant is negligible compared to the $N\ln N$ terms. The physical value $h=6.626\times 10^{-34}$ J$\cdot$s (Planck's constant) makes $S$ match the experimentally measured Third-Law entropy.

• The energy shell is not a volume. The density of states $\Omega (E)$ is a surface integral on $H=E$, not a volume integral. The equivalent formulation $\Omega =\partial \Sigma /\partial E$ uses the enclosed volume but the derivative converts it to a surface quantity. Confusing the two leads to wrong powers of $E$.

• The equal-probability postulate is not derived. It is an axiom of statistical mechanics, justified by the ergodic hypothesis (Master tier) but not provable from mechanics alone. Systems with exactly solvable dynamics (e.g., integrable systems) may violate it.

• $S=k_B\ln \Omega$ is not the same as the thermodynamic entropy of Clausius. They agree in the thermodynamic limit, but for small systems the Boltzmann entropy can behave differently (e.g., it may be non-concave).

Key theorem with proof [Intermediate+]

Theorem (Ideal gas equation of state from the microcanonical ensemble). For a monatomic ideal gas of $N$ particles of mass $m$ in volume $V$, the microcanonical entropy

$$S(E,V,N)=k_B\ln \Omega (E,V,N)$$

implies the equation of state $PV=N{k}_{B}T$ and the energy relation $E=\frac{3}{2}N{k}_{B}T$.

Proof. Start from the density of states derived above:

$$\Omega (E)=\frac{V^N}{N!h^{3N}}\frac{(2\pi m{)}^{3N/2}}{\Gamma (3N/2)}{E}^{3N/2-1}.$$

The entropy is

$$S=k_B\ln \Omega =k_B[N\ln V+(\frac{3N}{2}-1)\ln E+\text{const}(N,m)].$$

For $N$ large, $3N/2-1\approx 3N/2$, so to leading order in $N$:

$$S\approx k_B[N\ln V+\frac{3N}{2}\ln E+\text{const}(N,m)].$$

Temperature. Differentiate with respect to $E$ at fixed $V,N$:

$$\frac{1}{T}={\left(\frac{\partial S}{\partial E}\right)}_{V,N}=k_B\cdot \frac{3N}{2}\cdot \frac{1}{E},$$

giving

$$E=\frac{3}{2}N{k}_{B}T.\text{\hspace{1em}(∗)}$$

Pressure. Differentiate with respect to $V$ at fixed $E,N$:

$$\frac{P}{T}={\left(\frac{\partial S}{\partial V}\right)}_{E,N}=k_B\cdot \frac{N}{V},$$

giving

$$PV=N{k}_{B}T.\text{\hspace{1em}(∗∗)}$$

Equations ($\ast$) and ($\ast \ast$) are the ideal gas law and the equipartition-based energy formula, derived entirely from counting microstates. No equations of motion were solved; no molecular trajectories were computed. The thermodynamic behaviour of the ideal gas is a consequence of how many microstates exist at each energy and volume. ∎

Corollary (Sackur-Tetrode equation). Evaluating the constant terms in the entropy gives the Sackur-Tetrode formula for the entropy of a monatomic ideal gas:

$$S=N{k}_B\left[\ln \left(\frac{V}{N}{\left(\frac{4\pi mE}{3N{h}^2}\right)}^{3/2}\right)+\frac{5}{2}\right].$$

This is the exact (leading-order-in-$N$) entropy of a classical ideal gas. The $5/2$ term combines the Stirling approximation for $N!$ and the volume of a $3N$-dimensional sphere. The Sackur-Tetrode entropy is extensive (depends on $V/N$ and $E/N$), confirming that the Gibbs $N!$ correction is necessary for consistency with thermodynamics.

Worked example: two-level system

A system of $N$ non-interacting particles, each of which can be in a ground state (energy 0) or an excited state (energy $\epsilon$). If $n$ particles are excited, the total energy is $E=n\epsilon$. The number of microstates is

$$\Omega (n)=\left(\genfrac{}{}{0ex}{}{N}{n}\right)=\frac{N!}{n!(N-n)!}.$$

The entropy is $S=k_B\ln \left(\genfrac{}{}{0ex}{}{N}{n}\right)$. Using Stirling's approximation for large $N$:

$$S\approx k_B[-n\ln \frac{n}{N}-(N-n)\ln \frac{N-n}{N}].$$

The temperature follows from $1/T=\partial S/\partial E=(1/\epsilon )\partial S/\partial n$:

$$\frac{1}{T}=\frac{k_B}{\epsilon }\ln \frac{N-n}{n}.$$

Solving for the excitation fraction:

$$\frac{n}{N}=\frac{1}{1+e^{\epsilon /(k_{B}T)}}.$$

At low temperature ($k_{B}T\ll \epsilon$), nearly all particles are in the ground state ($n\approx 0$). At high temperature ($k_{B}T\gg \epsilon$), both levels are equally populated ($n\approx N/2$). The negative-temperature regime $n>N/2$ (more particles excited than not) corresponds to $T<0$, which is hotter than any positive temperature — it occurs in laser gain media and nuclear-spin systems.

Exercises [Intermediate+]

Ergodic hypothesis and its limitations [Master]

The microcanonical ensemble rests on the equal a priori probability postulate: every microstate on the energy surface $H=E$ is equally likely. The standard justification is the ergodic hypothesis: a trajectory in phase space visits every point on the energy surface, so time averages equal ensemble averages.

Definition. A Hamiltonian system on the energy surface ${\Gamma }_E$ is ergodic if the Hamiltonian flow is metrically transitive on ${\Gamma }_E$: every invariant subset of ${\Gamma }_E$ has either zero or full Liouville measure.

Birkhoff's ergodic theorem (1931). Let $({\Gamma }_E,{\varphi }_t,\mu )$ be a measure-preserving dynamical system where $\mu$ is the Liouville measure on ${\Gamma }_E$. For any $L^1$ observable $A$, the time average

$$\bar{A}(\mathbf{x}):=\underset{T\to \infty }{\lim }\frac{1}{T}{\int }_0^{T}A({\varphi }_t(\mathbf{x}))dt$$

exists for almost every $\mathbf{x}\in {\Gamma }_E$. If the system is ergodic, then $\bar{A}(\mathbf{x})=⟨A{⟩}_{\mu }$ for almost every $\mathbf{x}$ — the time average equals the microcanonical ensemble average, independent of the initial condition.

The ergodic theorem does not prove that physical systems are ergodic. It says: if the system is ergodic, then time averages equal ensemble averages. Proving ergodicity for a specific system is extraordinarily difficult. The first rigorous example was Sinai's proof (1963) that the hard-sphere gas is ergodic. For realistic potentials, ergodicity is generally assumed rather than proved.

Limitations.

1. Integrable systems are not ergodic. A system with $n$ independent conserved quantities in involution on a $2n$-dimensional phase space is confined to an $n$-dimensional torus, not the full $(2n-1)$-dimensional energy surface. The harmonic oscillator, the Kepler problem, and the Toda lattice are integrable and hence non-ergodic.

2. KAM systems are "mostly" non-ergodic. The Kolmogorov-Arnold-Moser theorem 05.09.01 guarantees that sufficiently irrational invariant tori persist under small perturbations. The phase space of a near-integrable system contains both regular (torus-confined) and chaotic regions, so the system is not ergodic on the full energy surface.

3. The time scale matters. Even if a system is ergodic, the time to explore the energy surface may exceed the age of the universe. Observables on laboratory timescales need not agree with ensemble averages.

4. Quantum mechanics changes the question. The energy surface is replaced by a finite-dimensional subspace of Hilbert space, and the quantum ergodic hypothesis (quantum chaos, random-matrix theory) takes a different form 12.07.01 pending.

The thermodynamic limit and equivalence of ensembles [Master]

The microcanonical ensemble is the most fundamental of the three standard ensembles (microcanonical, canonical, grand canonical), but it is also the most technically difficult to work with. The canonical and grand canonical ensembles are far more convenient for calculations. Their physical equivalence rests on the thermodynamic limit.

Definition. The thermodynamic limit is the limit $N\to \infty$, $V\to \infty$, $N/V=\rho$ fixed (and, for the microcanonical ensemble, $E/N=\epsilon$ fixed).

Equivalence of ensembles. In the thermodynamic limit, the microcanonical, canonical, and grand canonical ensembles give identical results for all thermodynamic quantities. The relative fluctuations of energy in the canonical ensemble scale as $1/\sqrt{N}$, and the relative fluctuations of particle number in the grand canonical ensemble scale as $1/\sqrt{N}$. These fluctuations vanish as $N\to \infty$, and the three ensembles become indistinguishable.

The equivalence breaks down for:

• Small systems (proteins, nanoclusters): fluctuations are large and ensemble choice matters.
• Systems with long-range interactions (gravitational systems, plasma): the thermodynamic limit is non-extensive and ensemble inequivalence can occur.
• First-order phase transitions in the microcanonical ensemble: the microcanonical entropy $S(E)$ can be non-concave, leading to negative specific heat, which is impossible in the canonical ensemble. The ensembles disagree in the transition region.

The surface-volume problem and coarse-grained entropy [Master]

The density of states $\Omega (E)$ is defined as an integral on the energy surface $H=E$. A common alternative is to define

$$\bar{\Omega }(E)=\frac{1}{\delta E}{\int }_{E\le H\le E+\delta E}d\Gamma ,$$

the number of states in a thin shell of width $\delta E$. The two definitions agree when $\delta E$ is small but large enough to contain many microstates — a physically necessary coarse-graining. The entropy $S=k_B\ln \bar{\Omega }$ is independent of the choice of $\delta E$ to leading order in $N$, because the volume of the shell is dominated by the outer surface.

For continuous classical systems the density of states must be coarse-grained: without a phase-space unit $h^{3N}$, $\Omega$ has dimensions and $S=k_B\ln \Omega$ is undefined. The Planck constant $h$ provides the natural unit. The resulting entropy is the coarse-grained Boltzmann entropy, and its independence from the coarse-graining scale (in the thermodynamic limit) is a consistency check.

An alternative definition, the Gibbs entropy (also called the volume entropy),

$$S_{\text{Gibbs}}=k_B\ln \Sigma (E)=k_B\ln {\int }_{H\le E}d\Gamma ,$$

uses the enclosed volume rather than the surface density. For the ideal gas, $S_{\text{Gibbs}}=k_B[\frac{3N}{2}\ln E+N\ln V+\text{const}]$, which differs from the Boltzmann entropy $S=k_B[(\frac{3N}{2}-1)\ln E+N\ln V+\text{const}]$ by $-k_B\ln E$. The difference vanishes in the thermodynamic limit relative to the leading $N\ln N$ terms, but for small systems it is measurable. The Gibbs entropy is always concave (non-decreasing $1/T$), while the Boltzmann surface entropy need not be. The correct choice remains debated in the foundations literature.

Jaynes' maximum-entropy principle [Master]

Jaynes (1957) recast statistical mechanics as an inference problem. Given partial knowledge of a system (fixed $E$, $V$, $N$), assign probabilities to microstates so as to be maximally non-committal about missing information. The unique solution is the distribution maximising $S=-k_B{\sum }_i{p}_i\ln p_i$ subject to the constraints.

For the microcanonical ensemble the constraint is: $p_i=0$ unless microstate $i$ has energy $E$. Maximising the Shannon entropy under this constraint gives $p_i=1/\Omega$ for all accessible microstates — the equal-probability postulate is derived rather than assumed.

Jaynes' framework extends immediately to the canonical ensemble (constraint: fixed mean energy $⟨E⟩$) and the grand canonical ensemble (constraints: fixed mean $⟨E⟩$ and $⟨N⟩$). The Lagrange multipliers enforcing the constraints are $1/T$ and $\mu /T$. This information-theoretic approach bypasses the ergodic hypothesis entirely: statistical mechanics is not about dynamics but about inference under incomplete information. The debate between the dynamical (ergodic) and information-theoretic (Jaynesian) foundations of statistical mechanics remains open.

Lean formalization [Intermediate+]

Mathlib has measure theory (MeasureTheory.Integral.Bochner, MeasureTheory.Measure.Lebesgue) and integration on manifolds. It does not formalise:

• The density of states $\Omega (E)$ as a surface integral on an energy shell.
• Boltzmann entropy $S=k_B\ln \Omega$ as a physical concept.
• The microcanonical ensemble average $⟨A⟩=\frac{1}{\Omega }{\int }_{{\Gamma }_E}Ad\Sigma /|\nabla H|$.
• The thermodynamic limit ($N\to \infty$) and equivalence of ensembles.
• Birkhoff's ergodic theorem in the specific form used in physics.

The formalisation pathway would start from the Liouville measure on $T^{\ast }Q$ (built on existing Mathlib cotangent-bundle machinery) and proceed to the surface integral on $\{H=E\}$ via the coarea formula. lean_status: none reflects this gap.

Connections [Master]

• Hamilton's equations 09.04.02 pending define the phase-space flow whose energy surface $H=E$ is the arena for the microcanonical ensemble. Liouville's theorem (phase-space volume conservation) is what makes the Liouville measure on ${\Gamma }_E$ the natural one.

• Maxwell-Boltzmann distribution 11.02.01 pending is recovered from the microcanonical ensemble by integrating out all but one particle's degrees of freedom — the marginal distribution of a single particle's momentum in the microcanonical ensemble is Maxwell-Boltzmann.

• First and second laws 11.01.01 pending acquire microscopic foundations: the first law is energy conservation under $H$; the second law is the statement that macroscopic evolution proceeds toward higher-$\Omega$ macrostates.

• Canonical ensemble 11.04.01 pending is derived from the microcanonical ensemble by coupling the system to a large heat bath and tracing out the bath's degrees of freedom. The partition function $Z={\sum }_i{e}^{-\beta E_i}$ replaces $\Omega (E)$ as the central object.

• Ising model (math side) 08.06.01 counts microstates of a discrete spin system; the microcanonical density of states $\Omega (E)$ for the Ising model is a combinatorial object studied in the math-side treatment.

• KAM theorem 05.09.01 limits the validity of the ergodic hypothesis: near-integrable systems retain KAM tori that prevent exploration of the full energy surface.

• Quantum chaos 12.07.01 pending replaces the classical ergodic hypothesis with the quantum eigenstate thermalisation hypothesis (ETH), which asserts that individual energy eigenfunctions reproduce microcanonical expectations.

Historical & philosophical context [Master]

Boltzmann introduced the formula $S=k\ln W$ (using $W$ for Wahrscheinlichkeit, probability) in 1877, extending his earlier work on the kinetic theory of gases (1868, 1872). The 1877 paper, "Über die Beziehung zwischen dem zweiten Hauptsatze der mechanischen Wärmetheorie und der Wahrscheinlichkeitsrechnung", established that the second law of thermodynamics is a statistical statement: entropy increases not by mechanical necessity but because high-entropy macrostates overwhelm low-entropy ones in number.

Gibbs, in Elementary Principles in Statistical Mechanics (1902), systematised the ensemble approach — the microcanonical, canonical, and grand canonical ensembles — and introduced the phase-space density formalism. Gibbs' "microcanonical" (his term) ensemble was the foundation on which the others were built. The Gibbs paradox (the non-extensivity of entropy for mixing identical gases) was resolved by the $N!$ correction, which Gibbs introduced ad hoc and which receives its natural justification from quantum indistinguishability.

Ehrenfest and Ehrenfest's 1911 review article, "Begriffliche Grundlagen der statistischen Auffassung in der Mechanik", subjected the ergodic hypothesis to its first rigorous critique, distinguishing the original ergodic hypothesis (the trajectory passes through every point of ${\Gamma }_E$, which is impossible for a flow on a continuous space) from the quasi-ergodic hypothesis (the trajectory is dense in ${\Gamma }_E$). Birkhoff's 1931 ergodic theorem provided the precise mathematical framework; the Ehrenfests' critique made clear that the ergodic hypothesis is a strong assumption, not a theorem.

Sinai proved ergodicity of the hard-sphere gas in 1963 (published 1970), the first physically relevant ergodicity result. The quantum counterpart — the eigenstate thermalisation hypothesis — was proposed by Deutsch (1991) and Srednicki (1994) and remains an active research frontier.

Jaynes' 1957 papers, "Information theory and statistical mechanics" (Phys. Rev. 106, 620 and 108, 171), recast the entire foundation in information-theoretic terms. The Jaynesian programme has been influential but controversial: critics argue it conflates epistemic uncertainty with physical randomness, while proponents argue it provides the cleanest derivation of the ensembles.

Bibliography [Master]

Primary literature (cite when used; not all currently in reference/):

• Boltzmann, L., "Über die Beziehung zwischen dem zweiten Hauptsatze der mechanischen Wärmetheorie und der Wahrscheinlichkeitsrechnung", Wiener Berichte 76 (1877), 373–435.
• Gibbs, J. W., Elementary Principles in Statistical Mechanics (Charles Scribner's Sons, 1902).
• Ehrenfest, P. & Ehrenfest, T., "Begriffliche Grundlagen der statistischen Auffassung in der Mechanik", Encyklopädie der mathematischen Wissenschaften IV 2 II (1911).
• Birkhoff, G. D., "Proof of the ergodic theorem", Proc. Natl. Acad. Sci. USA 17 (1931), 656–660.
• Sinai, Ya. G., "On the foundation of the ergodic hypothesis for a dynamical system of statistical mechanics", Dokl. Akad. Nauk SSSR 153 (1963) [Sov. Math. Dokl. 4 (1963), 1818–1822].
• Jaynes, E. T., "Information theory and statistical mechanics", Phys. Rev. 106 (1957), 620–630; 108 (1957), 171–190.
• Schroeder, D. V., An Introduction to Thermal Physics (Addison-Wesley, 2000).
• Reichl, L. E., A Modern Course in Statistical Physics, 2nd ed. (Wiley, 1997).
• Landau, L. D. & Lifshitz, E. M., Statistical Physics, Part 1, 3rd ed. (Course of Theoretical Physics Vol. 5, Pergamon, 1980).
• Huang, K., Statistical Mechanics, 2nd ed. (Wiley, 1987).
• Susskind, L. & Hrabovsky, G., The Theoretical Minimum: Statistical Mechanics (Basic Books, 2013).
• Tong, D., Statistical Physics (DAMTP Cambridge lecture notes, §3 "Statistical ensembles").

---

Wave 2 physics unit, produced 2026-05-18 per PHYSICS_PLAN §5. Status: draft pending Tyler's review and the §11 Next-Actions retro.