---
title: Boltzmann-Type Entropy Structure
url: https://www.emergentmind.com/topics/boltzmann-type-entropy-structure
type: topic
---

# Boltzmann-Type Entropy Structure

Searching arXiv for recent and foundational papers on Boltzmann-type entropy structure.
arXiv search query: "Boltzmann entropy structure entropy microcanonical kinetic generalized entropy"
Boltzmann-type entropy structure denotes a family of constructions in which entropy is tied to the size of the microscopic set compatible with a prescribed macrodescription, and, in kinetic or reduced descriptions, to convex entropy functionals whose evolution reproduces a Boltzmann \(H\)-theorem, a balance law, or an equivalent dissipation principle. In its elementary microcanonical form this structure appears as \(S(E)=k\ln \Omega(E)\); in phase-space language as \(S_B(X)=k\log \mathrm{vol}\,\Gamma(X)\); and in quantum macrospace language as \(S_{qB}(\nu)=k\log\dim H_\nu\). Across classical, quantum, kinetic, generalized, and computational settings, the persistent issue is which entropy functional preserves extensivity, macroscopic accessibility, equilibrium selection, and physically meaningful nonequilibrium entropy production [1404.2760; 1903.11870].

## 1. Foundational microcanonical and macrostate formulations

In the classical Boltzmannian formulation, entropy is assigned to an individual system through the macrostate containing its actual microstate. If \(X\) is the phase point of the system and phase space is partitioned as \(X=\bigcup_\nu \Gamma_\nu\), then \(\Gamma(X)\) denotes the unique macro set containing \(X\), and the entropy is
\[
S_B(X)=k\log \mathrm{vol}\,\Gamma(X), \qquad S_B(\nu)=k\log \mathrm{vol}\,\Gamma_\nu.
\]
This construction presupposes coarse-grained macrovariables \(M_j\) and corresponding macrostates
\[
\Gamma_\nu=\{x\in X: M_j(x)=\nu_j\ \forall j\},
\]
typically inside an energy shell \(X_{mc}\). In dilute-gas settings, the familiar continuum expression \(-k\int \tilde f\log \tilde f\) arises as a special case in which the macrostate is specified by the empirical one-particle distribution derived from the actual microstate, not by an ensemble density [1903.11870].

The shell-counting form
\[
S(E)=k\ln \Omega(E)
\]
is the corresponding microcanonical expression when \(\Omega(E)\) denotes the number of microstates with energy exactly \(E\). Here entropy is attached to the degeneracy of a single energy shell rather than to a cumulative count of states below that shell. In the macroscopic limit this distinction is decisive whenever the density of states is nonmonotone, particularly for bounded spectra and negative-temperature regimes [1404.2760].

Quantum versions preserve the same structural pattern. If the Hilbert space decomposes as
\[
H=\bigoplus_\nu H_\nu,
\]
with each \(H_\nu\) a macrospace, then the quantum Boltzmann entropy is
\[
S_{qB}(\nu)=k\log\dim H_\nu,
\]
and for \(\psi\in H_\nu\), one sets \(S_{qB}(\psi)=S_{qB}(\nu)\). Classically the “size” of the compatible macrostate is phase-space volume; quantum mechanically it is Hilbert-space dimension. This preserves the core Boltzmann idea that entropy is the logarithm of the size of the macrostate supporting the actual state [1903.11870].

## 2. Thermodynamic order, extensivity, and definitional controversies

A central thermodynamic question is whether Boltzmann entropy, after suitable regularization, determines adiabatic accessibility in the same way as thermodynamic entropy. For equilibrium states specified by extensive variables \(\vec a\), the regularized Boltzmann entropy is introduced as an extensive large-\(X\) limit,
\[
\tilde S_B[\vec a]:=\lim_{X\rightarrow\infty}\frac{1}{X}\log D^{(X)\downarrow}_{\vec a},
\]
and, when \(\tilde S_B[\vec a]\) exists and is convex and strictly increasing in each component of \(\vec a\), it satisfies
\[
\tilde S_B[\vec a]\le \tilde S_B[\vec a'] \Leftrightarrow \vec a\prec_{ad}\vec a'.
\]
In this sense, the regularized Boltzmann entropy provides a necessary and sufficient condition for macroscopic adiabatic transformation, so its order-theoretic role coincides with that of thermodynamic entropy [1601.00487].

A different controversy concerns the relation between Boltzmann entropy and Gibbs entropy. For macroscopic equilibrium states, the Gibbs entropy of the microcanonical ensemble and the Boltzmann entropy of the dominant equilibrium macrostate agree to leading order, but outside equilibrium they need not coincide. The Boltzmannian position is that thermodynamic entropy pertains to the actual system and its macroregion, whereas fine-grained Gibbs entropy is a functional of an ensemble density and is invariant under Hamiltonian evolution. This makes the Boltzmann notion the candidate that tracks the second law for individual systems, while Gibbs ensembles remain computationally useful for equilibrium and typicality arguments [1903.11870].

The most pointed definitional dispute in the present literature is the proposal to replace
\[
S(E)=k\ln\Omega(E)
\]
with the cumulative expression
\[
S_G(E)=k\ln \sum_{E'<E}\Omega(E').
\]
For bounded-spectrum systems, where \(\Omega(E)\) rises to a maximum at some \(E^*\) and then decreases, \(S_G\) ceases to follow the actual shell multiplicity beyond the maximum. In that regime \(S_G(E)\to S(E^*)\) in the macroscopic limit, its finite differences vanish exponentially fast with system size, and the associated temperature \(T_G\) scales as \(e^{cN}\), so it is not intensive. The same analysis shows that, in a two-level model, \(S_G\) can assign temperatures that reverse the operational hot–cold ordering of two systems. The resulting conclusion is that \(S_G\) is either asymptotically identical to \(S\) or unrealistically independent of the energy [1404.2760].

## 3. Kinetic balance laws, irreversibility, and geometric dissipation

In kinetic theory, Boltzmann-type entropy structure appears as a local balance law. For a state variable obeying autonomous dynamics, kinetic entropy exists when one can construct a local entropy density \(\rho_s\), entropy current \(\mathbf j_s\), and nonnegative production term \(\mathrm{RHT}_s\) such that
\[
\partial_T \rho_s(R,T)+\nabla_R\cdot \mathbf j_s(R,T)=\mathrm{RHT}_s(R,T)\ge 0.
\]
For the classical Boltzmann equation this gives
\[
\rho_s(R,T)=\int dp\, f(p,R,T)\ln(1/f(p,R,T)),
\]
and detailed balance yields a Boltzmann \(H\)-theorem. The same structure extends to Landau’s Fermi-liquid kinetic equation, where the local entropy density becomes
\[
\mathcal S(f)= -f\ln f+\varsigma(1+\varsigma f)\ln(1+\varsigma f),
\]
giving the standard Fermi-Dirac and Bose-Einstein mode entropies. In the full non-equilibrium Green’s-function setting, however, this perfect-integral structure generally fails outside local equilibrium, because the entropy equation acquires extra terms that cannot be cast as either a local flux divergence or a manifestly sign-definite production term [1403.6162].

A distinct information-theoretic reinterpretation of Boltzmann irreversibility replaces coarse-graining in phase-space cells by deletion of correlations. If a composite state \(W\) is mapped to the product of its marginals \(PW=\rho_1\otimes\cdots\otimes\rho_n\), then
\[
S(PW)-S(W)=I_{1\ldots n}\ge 0,
\]
where \(I_{1\ldots n}\) is mutual information. In the deterministic relaxation model
\[
\frac{dW}{dt}=LW-\gamma(W-PW),
\]
the entropy production becomes
\[
\dot S=\gamma\,D(W\Vert PW)+\gamma\,D(PW\Vert W)\ge 0.
\]
This reformulates the Stoßzahlansatz as repeated deletion of interpartition mutual information rather than ordinary coarse-graining, and extends the Boltzmann logic to arbitrary classical or quantum partitions [2407.17322].

For the spatially homogeneous Boltzmann equation, the same dissipation mechanism has been recast as a gradient-flow structure. The driving functional is the Boltzmann–Shannon entropy
\[
H(\mu)=\int_{\mathbb R^d} f(v)\log f(v)\,dv,
\]
and the equation evolves as the steepest descent of \(H\) in a collision-adapted geometry built from binary-collision rates rather than Wasserstein transport. The entropy dissipation
\[
D(f)=\frac14\int \log\frac{f'f_*'}{ff_*}\,(f'f_*'-ff_*)\,B\,dv\,dv_*\,d\omega
\]
plays the role of squared metric slope, and the energy–dissipation identity
\[
H(f_T)-H(f_0)+\frac12\int_0^T D(f_t)\,dt+\frac12 A_T(f)\ge 0
\]
is saturated exactly by solutions of the homogeneous Boltzmann equation [1603.00540].

## 4. Quantum statewise entropy and nonequilibrium pure-state dynamics

Quantum formulations preserve the Boltzmannian idea that entropy can be assigned to an individual pure state, but they make explicit that the value depends on the chosen macrovariables. For a freely expanding one-dimensional quantum ideal gas, two macrostate choices have been analyzed. The \(U\)-macrovariables are local conserved fields in position space; the \(f\)-macrovariables are expectation values of occupation operators in a localized wavepacket basis that coarse-grains both position and momentum. For both choices the corresponding entropies \(s_B^U\) and \(s_B^f\) grow and eventually saturate. When the gas is initially at equilibrium in half the system and then expands to occupy twice the initial volume, the paper finds
\[
\log 2\le \Delta s_B^f\le 2\log 2 \quad \text{for fermions},
\]
and
\[
0\le \Delta s_B^f\le \log 2 \quad \text{for bosons}.
\]
In the quantum regime where the final stationary state is not at thermal equilibrium, \(\Delta s_B^U\) is greater than \(\Delta s_B^f\), showing directly that different macrostates retain different amounts of information about the same pure-state evolution [2303.12330].

For interacting quantum fields, a statewise Boltzmann entropy can be constructed from a pure state \(|\psi\rangle\) by averaging over unitary operators that shift the field in wavevector space. The resulting density operator
\[
f(\psi)=\frac{1}{|K'|}\sum_{\kappa\in K'}\sum_j P_j\,U_\kappa |\psi\rangle\langle\psi|U_\kappa^\dagger P_j
\]
defines
\[
S(\psi)=-\operatorname{tr}[f(\psi)\ln f(\psi)].
\]
Under the paper’s assumption on the thermodynamic-limit behavior of the shifted canonical state, \(S(\psi)\) coincides with thermodynamic entropy for almost all pure states in the canonical pure-state ensemble corresponding to a thermodynamic state, while still evolving nontrivially under Hamiltonian dynamics for general self-interacting fields [1909.00961].

A separate quantum controversy arises in open systems under strong coupling and non-Markovian dynamics. In numerically “exact” HEOM simulations, the entropy derived from a reduced partition function and quasi-static Helmholtz free energy—the paper’s “Boltzmann entropy”—and the reduced-system von Neumann entropy show similar dependence on system–bath coupling at the level of system entropy change, but the corresponding total entropy productions differ qualitatively. The Boltzmann total entropy production remains positive, whereas the von Neumann-based total entropy production becomes negative when the initial state is the unfactorized thermal equilibrium state of the total system. The stated reason is that the von Neumann construction does not properly take into account the contribution of the entropy from the system–bath interaction [2012.09546].

## 5. Nonadditivity, composability, and generalized Boltzmann principles

Boltzmann-type structure is not restricted to strictly additive entropy. Einstein’s likelihood principle requires that for probabilistically independent subsystems \(A\) and \(B\),
\[
W(A+B)=W(A)W(B).
\]
Boltzmann–Gibbs entropy satisfies this because it is additive, but additivity is sufficient rather than necessary. Tsallis entropy
\[
S_q=k\frac{1-\sum_i p_i^q}{q-1}=k\sum_i p_i\ln_q\frac{1}{p_i}
\]
obeys the nonadditive composition law
\[
\frac{S_q(A+B)}{k}= \frac{S_q(A)}{k}+\frac{S_q(B)}{k}+(1-q)\frac{S_q(A)}{k}\frac{S_q(B)}{k},
\]
yet still yields likelihood factorization because its generalized exponential satisfies
\[
e_q^{x\oplus_q y}=e_q^x e_q^y.
\]
The same logic extends to broader generalized entropies \(S_G\), so likelihood factorization does not uniquely single out Boltzmann–Gibbs entropy [1407.6052].

A more elaborate generalization replaces additive or polynomial composability by a rational group law,
\[
\Phi(x,y)=\frac{x+y+a xy}{1+bxy}.
\]
The associated trace-form entropy \(S_{a,b,r}\) is
\[
S_{a,b,r}^{(\pm)}[p]=\sum_i p_i\,\Log_{a,b,r}^{(\pm)}(1/p_i),
\]
reduces to Tsallis for \(b=0,\ a=r=1-q\), and reduces to Boltzmann–Gibbs in the further limit \(r\to 0\). In the equiprobable case it becomes
\[
S_{a,b,r}^{(\pm)}[1/W]=\Log_{a,b,r}^{(\pm)}(W),
\]
so the logarithm of multiplicity survives, but the logarithm is now the one associated with the rational composition group. The same paper distinguishes exponential-growth regimes, power-law-growth regimes, and saturating or “freezing” regimes of accessible phase-space growth, arguing that different entropy functionals become appropriate in each case [1507.05058].

A finite-level reformulation of the maximum-entropy principle gives a further structural refinement. Solving the normalization and mean-energy constraints explicitly, one finds that entropy stationarity forces the log-probability vector to lie in the span of the constant vector and the energy vector, hence
\[
p_i=e^{-\alpha-\beta e_i}.
\]
The same result can be written as the cocycle relation
\[
p_j^{e_{lk}}\,p_l^{e_{kj}}\,p_k^{e_{jl}}=1,
\]
which encodes the affine dependence of \(-\ln p_i\) on the reduced energies \(e_i\) without invoking Lagrange multipliers [2512.22871].

## 6. Gravity, cosmology, and the entropy of the universe

Self-gravity modifies Boltzmann-type entropy structure in ways that are sharply unlike short-range systems. For a closed Friedmann–Lemaître type universe with matter modeled as \(N\) self-gravitating particles on \(S_R^3\), the matter entropy is treated Boltzmannially, either through a coarse-grained single-particle density
\[
S_{\mathrm B}(f;N)
\]
or through the equilibrium phase-space count
\[
S_{\mathrm B}^{\mathrm{erg}}(E,N)=\ln\!\Big[\frac{1}{N!}mc^2\Phi'(E)\Big].
\]
In the classical Newtonian model with
\[
H^{(N)}=\sum_{k=1}^N \frac{|p_k|^2}{2m}-\sum_{k<l}\frac{Gm^2}{|q_k-q_l|},
\]
the microcanonical phase-space integral diverges for \(N\ge 3\), so the Boltzmann entropy is unbounded above and no maximum-entropy state exists. For nonrelativistic quantum fermions, Pauli exclusion restores entropy maxima and yields core–halo equilibria and white-dwarf-like condensed states. In pseudo-special-relativistic quantum gravity, entropy maxima are lost again above a Chandrasekhar threshold. The same analysis argues that Bekenstein–Hawking entropy does not always compensate for the Boltzmann entropy of matter swallowed by a black hole, so a single all-swallowing black hole is not automatically the maximum-entropy state of a closed universe [1905.12727].

A distinct Newtonian cosmology constructs all entropies explicitly as Boltzmann entropies from a cosmological wavefunction written in Madelung form,
\[
\psi=A\exp\!\left(\frac{i}{\hbar}\mathcal I\right)=\exp(S+iI),\qquad S=\frac{\mathcal S}{2k_B}.
\]
The matter entropy is then
\[
\mathcal S_m=2k_B\ln A,
\]
while gravitational entropy is tied to the expectation value of the Hubble potential through
\[
\mathcal S_g=\mathcal N\,\frac{k_B mH_0}{\hbar}R^2.
\]
Within this Newtonian model, the estimated gravitational entropy is of order \(10^{123}k_B\), in compliance with the holographic upper bound, whereas the matter entropy scales only logarithmically with the cosmic radius [1703.08082].

## 7. Structure-preserving closures and computational realizations

In kinetic computation, Boltzmann-type entropy structure has become a design principle. For entropy-based moment closures of transport or Boltzmann systems, the moment entropy
\[
h(u)=\langle \eta(f_u)\rangle
\]
is induced by the minimum-entropy reconstruction
\[
\min_{g\in F_m}\ \langle \eta(g)\rangle \quad \text{s.t.}\quad u=\langle m g\rangle,
\]
with entropy variables \(\alpha_u=\nabla_u h(u)\). Neural surrogates preserve this structure by learning either the convex map \(u\mapsto h(u)\) with an input-convex neural network or the monotone map \(u\mapsto \alpha_u\). The aim is not merely regression of fluxes, but preservation of convex entropy, entropy dissipation, hyperbolicity, and realizability through the exponential ansatz \(f_u=\eta_*'(\alpha_u\cdot m)\) [2201.10364].

A more recent regularized framework replaces the singular closure by a partially regularized entropy \(h^\gamma\), then approximates its reduced normalized form \(\hat h^\gamma\) with a convex neural network, yielding the two-stage approximation
\[
h \;\longrightarrow\; h^\gamma \;\longrightarrow\; h^p.
\]
If the closure uses the entropy-gradient-based ansatz
\[
f_u^\gamma=\eta_*'(\nabla_u h^\gamma(u)\cdot m),
\]
then the paper proves the entropy inequality
\[
\partial_t h^\gamma(u)+\nabla_{\mathbf x}\cdot j^\gamma(u)\le 0
\]
and symmetrizable hyperbolicity for the closed moment system. The same entropy structure survives under the learned approximation \(h^p\), while the method yields a much lower memory footprint than traditional methods with competitive computation times and simulation accuracy [2404.14312].

Entropy structure also constrains reduced kinetic models such as ES-BGK. For
\[
\partial_t f+v\cdot \nabla_x f=A_\nu(\mathcal M_\nu(f)-f),
\]
the entropy production satisfies the Cercignani-type estimate
\[
D_\nu(f)\ge \min\{1+2\nu,\,1-\nu\}\,A_\nu\,H(f\mid \mathcal M_0), \qquad -\frac12<\nu<1,
\]
so the only equilibria are local Maxwellians despite the anisotropic ellipsoidal Gaussian target. The same analysis identifies an additional remainder term \(R_\nu\) with the sign-definite property
\[
\nu\,R_\nu\ge 0,
\]
showing that the anisotropic correction affects entropy production in a controlled one-sided way throughout the physical parameter range [1707.09882].

Taken together, these developments show that Boltzmann-type entropy structure is not a single formula but a persistent architecture: entropy as logarithmic state multiplicity for individual macrostates, as extensive regularization governing adiabatic order, as local or geometric dissipation in kinetic equations, as a statewise quantity for pure quantum dynamics, as a generalized composability principle beyond strict additivity, and as a convex invariant underpinning structure-preserving numerical closure.

Source: https://www.emergentmind.com/topics/boltzmann-type-entropy-structure