# H-theorem
The **H-theorem** is a result of kinetic theory, published by [Ludwig Boltzmann](https://www.edgepedia.org/ludwig-boltzmann) in 1872, stating that for a dilute gas whose single-particle distribution function satisfies the [Boltzmann equation](https://www.edgepedia.org/boltzmann-equation), the quantity
\[ H[f_t] := \int f_t(\vec{v}) \ln f_t(\vec{v}) \, d^3\vec{v} \]
cannot increase in time, \( dH/dt \le 0 \), and is stationary if and only if the distribution is the Maxwell-Boltzmann equilibrium distribution[1](https://plato.stanford.edu/entries/statphys-statmech/). Because \( H \) behaves like a negative entropy, the theorem is often paraphrased as showing that entropy increases monotonically until the system reaches equilibrium, and it was the basis of Boltzmann's claim to have given a statistical foundation for the second law of thermodynamics[1](https://plato.stanford.edu/entries/statphys-statmech/) • [2](https://plato.stanford.edu/entries/statphys-boltzmann/).
| Key fact | Detail |
|---|---|
| Origin | Published in Boltzmann's 1872 paper (often abbreviated *Weitere Studien*), together with the Boltzmann equation; Boltzmann denoted the quantity \( E \)[2](https://plato.stanford.edu/entries/statphys-boltzmann/) |
| Statement | \( dH[f_t]/dt \le 0 \), with \( dH/dt = 0 \) exactly when \( f_t \) is the Maxwell-Boltzmann distribution \( A e^{-Bv^2} \)[1](https://plato.stanford.edu/entries/statphys-statmech/) |
| Entropy link | \( -kH \) can be identified with the non-equilibrium entropy of an ideal gas; the limit of \( H \) as \( t \to \infty \) equals the Gibbs entropy with opposite sign[3](https://encyclopediaofmath.org/wiki/Boltzmann_H-theorem) |
| Key assumption | The Stosszahlansatz (molecular chaos): velocities of colliding particles are uncorrelated and independent of position; this assumption is not invariant under time reversal[1](https://plato.stanford.edu/entries/statphys-statmech/) • [4](https://pmc.ncbi.nlm.nih.gov/articles/PMC5018848/) |
| Objections | Loschmidt's 1876 reversibility objection and Zermelo's 1896 recurrence objection show the theorem cannot follow from time-reversal-invariant dynamics alone[2](https://plato.stanford.edu/entries/statphys-boltzmann/) • [5](https://plato.sydney.edu.au/entries/time-thermo/) |
| Statistical reading | Boltzmann later conceded exceptions exist but are extremely improbable; entropy increase is overwhelmingly likely rather than necessary[6](https://web.stanford.edu/~peastman/statmech/thermodynamics.html) |
| Extensions | Quantum versions (von Neumann; quantum information theory), Tsallis-entropy versions, and rigorous validity results (Lanford's theorem)[4](https://pmc.ncbi.nlm.nih.gov/articles/PMC5018848/) • [7](https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.86.2938) • [1](https://plato.stanford.edu/entries/statphys-statmech/) |
## Statement and proof sketch
Boltzmann's 1872 paper was the first work on non-equilibrium gas theory, aimed at showing that whatever the initial state of a gas, it tends to evolve toward equilibrium[2](https://plato.stanford.edu/entries/statphys-boltzmann/). The gas is described by a distribution function \( f_t(\vec{v}) \) giving the fraction of molecules with velocity \( \vec{v} \) at time \( t \). Its derivation assumes hard-sphere particles of diameter \( D \), elastic collisions, a particle number large enough that the discrete distribution can be approximated by a continuous differentiable function, and a density low enough that only two-particle collisions matter[1](https://plato.stanford.edu/entries/statphys-statmech/).
The crucial assumption is the **Stosszahlansatz** (German for "collision number assumption"), which specifies how many collisions of a given type occur in a time interval. In modern terms it assumes the velocities of colliding particles are uncorrelated and independent of position[1](https://plato.stanford.edu/entries/statphys-statmech/) • [4](https://pmc.ncbi.nlm.nih.gov/articles/PMC5018848/). With this assumption, the collision contribution to \( dH/dt \) can be written as an integral of \( (f'f_1' - ff_1) \ln(f'f_1'/ff_1) \) times a positive collision kernel; since \( (f - g)(\ln f - \ln g) \ge 0 \) for positive \( f, g \), the integrand is nonpositive and \( H \) can only decrease[8](https://www.cmls.polytechnique.fr/perso/golse/Surveys/FGHdBookEvEq.pdf). Equality holds exactly when \( f'f_1' = ff_1 \) for almost all collisions, which forces \( f \) to be a Maxwellian \( \mathcal{M}_{(\rho,u,\theta)}(v) = \frac{\rho}{(2\pi\theta)^{3/2}} e^{-|v-u|^2/(2\theta)} \)[8](https://www.cmls.polytechnique.fr/perso/golse/Surveys/FGHdBookEvEq.pdf).
Boltzmann presented the result as a rigorous analytical proof of the second law, contrasting it with earlier proofs restricted to reversible cyclic processes[2](https://plato.stanford.edu/entries/statphys-boltzmann/). In the 1872 paper he denoted the quantity \( E \); the name "H-theorem" came later[2](https://plato.stanford.edu/entries/statphys-boltzmann/).
## Relation to entropy
The function \( H \) is defined with the opposite sign from thermodynamic entropy. For an ideal gas, \( -kH \), where \( k \) is the [Boltzmann constant](https://www.edgepedia.org/boltzmann-constant), can be identified with the non-equilibrium entropy; when the local Maxwell distribution is established, \( -h(t, \vec{r}) \) serves as an entropy density[3](https://encyclopediaofmath.org/wiki/Boltzmann_H-theorem). The theorem therefore generalizes the entropy concept to non-equilibrium states and asserts that this non-equilibrium entropy \( -kH \) increases monotonically as an isolated gas evolves toward equilibrium[2](https://plato.stanford.edu/entries/statphys-boltzmann/). The relationship with the second law is indirect: the second law speaks neither of non-equilibrium entropy nor of strictly monotonic increase, and Boltzmann's theorem covers an isolated gas in a fixed container, not general adiabatic processes[2](https://plato.stanford.edu/entries/statphys-boltzmann/).
## The reversibility and recurrence objections
**Loschmidt's objection (1876).** Josef Loschmidt argued that since the equations of (Hamiltonian) mechanics are time-reversal invariant, for every solution one can construct another by reversing all velocities and replacing \( t \) by \( -t \). Because \( H[f] \) is invariant under velocity reversal, if \( H \) decreases along the first solution it increases along the second, so the H-theorem cannot be a general theorem of mechanics[2](https://plato.stanford.edu/entries/statphys-boltzmann/). In his response of 1877, Boltzmann acknowledged that certain initial states lead to an increase of \( H \), but argued such states are extremely improbable and can be ignored[2](https://plato.stanford.edu/entries/statphys-boltzmann/). Commentators have found this rebuttal unsatisfactory, since any reasonable probability assignment is itself invariant under velocity reversal[2](https://plato.stanford.edu/entries/statphys-boltzmann/).
**Zermelo's objection (1896).** [Ernst Zermelo](https://www.edgepedia.org/ernst-zermelo) invoked Poincaré recurrence: a bounded system left to itself eventually returns arbitrarily close to its initial state, so \( H \) cannot decrease all the time[5](https://plato.sydney.edu.au/entries/time-thermo/) • [9](https://doi.org/10.1007/BF01807856). Both objections together establish that no function of the classical state can increase monotonically under conservative dynamics alone; some time-asymmetric ingredient had entered the theorem without being announced[5](https://plato.sydney.edu.au/entries/time-thermo/).
**Burbury and Condition A.** The logical situation was clarified in the British debate of 1894 to 1895, begun by Culverwell's question "Will anyone say exactly what the H-theorem proves?" Samuel Burbury argued that the theorem rests on an additional assumption, independent of mechanical theory, which he called "Condition A"; reversed motion would violate it, so the theorem would not apply[2](https://plato.stanford.edu/entries/statphys-boltzmann/). Bryan pointed out that the Stosszahlansatz itself is not invariant under time reversal[2](https://plato.stanford.edu/entries/statphys-boltzmann/). Boltzmann, participating in the debate, accepted that an additional assumption was needed and later stated that the minimum theorem and the second law are "only theorems of probability": from the equations of motion alone it can never be proved that \( H \) must always decrease[2](https://plato.stanford.edu/entries/statphys-boltzmann/).
## Statistical reading and rigorous status
The modern statistical reading holds that the H-theorem shows entropy increase to be likely rather than universal: Boltzmann admitted that initial states exist for which \( H \) decreases, but maintained they rarely, if ever, occur in nature[1](https://plato.stanford.edu/entries/statphys-statmech/). The molecular chaos assumption cannot be strictly true, but its net effect appears to cancel as the number of particles grows[10](https://ocw.mit.edu/courses/16-55-ionized-gases-fall-2014/939404b24861cd79e4809b09baf83d78_MIT16_55F14_Lecture9.pdf).
A century after Boltzmann, Oscar Lanford published a series of papers (1973 to 1981) culminating in Lanford's theorem, which addresses whether and in what sense the Boltzmann equation is consistent with the underlying Hamiltonian dynamics, using the Boltzmann-Grad limit to obtain a differentiable distribution function[1](https://plato.stanford.edu/entries/statphys-statmech/).
## Applications
The H-theorem is central to working kinetic theory. It supplies a priori entropy-production estimates on solutions of the Boltzmann equation, and it is so essential in deriving hydrodynamic equations from the Boltzmann equation that treatments of the subject give a complete proof of it[8](https://www.cmls.polytechnique.fr/perso/golse/Surveys/FGHdBookEvEq.pdf). From the standpoint of statistical mechanics, its principal significance is the mathematical expression of the assumptions of macroscopic thermodynamics: an isolated system spontaneously tends to thermodynamic equilibrium, the process being accompanied by an increase in entropy[3](https://encyclopediaofmath.org/wiki/Boltzmann_H-theorem).
## Quantum and generalized versions
**Quantum H-theorem.** John von Neumann, seeking to bypass the molecular chaos hypothesis, which is unjustified within classical mechanics, proposed a quantum-mechanical origin of entropy growth. He defined entropy through the density matrix and offered a proof of non-decreasing entropy resting on a final procedure of macroscopic measurement; because that proof invoked concepts beyond pure quantum mechanics, the search for a fully quantum foundation has continued[4](https://pmc.ncbi.nlm.nih.gov/articles/PMC5018848/). Within quantum information theory, a quantum H-theorem can be formulated as follows: the entropy gain during evolution is nonnegative if the system evolution can be described by a unital channel[4](https://pmc.ncbi.nlm.nih.gov/articles/PMC5018848/).
**Generalized entropies.** The theorem extends to other entropy measures. For a transport equation built on a nonextensive generalization of the molecular chaos hypothesis, an H theorem holds with [Tsallis entropy](https://www.edgepedia.org/tsallis-entropy), and the collisional equilibrium is given by Tsallis's q-nonextensive velocity distribution[7](https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.86.2938). Analogous H-theorems are statements of the same type formulated for statistical systems of a different or more general type, including non-ideal and quantized systems[3](https://encyclopediaofmath.org/wiki/Boltzmann_H-theorem).
## References
1. [Philosophy of Statistical Mechanics, Stanford Encyclopedia of Philosophy](https://plato.stanford.edu/entries/statphys-statmech/)
2. [Boltzmann's Work in Statistical Physics, Stanford Encyclopedia of Philosophy](https://plato.stanford.edu/entries/statphys-boltzmann/)
3. [Boltzmann H-theorem, Encyclopedia of Mathematics](https://encyclopediaofmath.org/wiki/Boltzmann_H-theorem)
4. [H-theorem in quantum physics, PMC](https://pmc.ncbi.nlm.nih.gov/articles/PMC5018848/)
5. [Thermodynamic Asymmetry in Time, Stanford Encyclopedia of Philosophy](https://plato.sydney.edu.au/entries/time-thermo/)
6. [Thermodynamics, Introduction to Statistical Mechanics (Stanford)](https://web.stanford.edu/~peastman/statmech/thermodynamics.html)
7. [Lima, Silva, Plastino (2001). Nonextensive Thermostatistics and the H Theorem. Physical Review Letters 86, 2938](https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.86.2938)
8. [Golse, F. The Boltzmann equation and its hydrodynamic limits](https://www.cmls.polytechnique.fr/perso/golse/Surveys/FGHdBookEvEq.pdf)
9. [Wu, Ta-You (1975). Boltzmann's H theorem and the Loschmidt and the Zermelo paradoxes. International Journal of Theoretical Physics 14, 289–294](https://doi.org/10.1007/BF01807856)
10. [Boltzmann's H-theorem, MIT OpenCourseWare 16.55 Lecture 9](https://ocw.mit.edu/courses/16-55-ionized-gases-fall-2014/939404b24861cd79e4809b09baf83d78_MIT16_55F14_Lecture9.pdf)
---
*Topic: Encyclopedia › Physical world and mathematics › Physics › Physics methods, practice, and community › History and philosophy of physics › Philosophy of physics › Philosophy of spacetime, thermodynamics, and statistical physics › Irreversibility and the approach to equilibrium*
*Initially written Sep 17, 2026 · Reviewed: Sep 30, 2026; Oct 15, 2026 · Edited: Sep 30, 2026; Oct 15, 2026 · Last review: Oct 15, 2026*
*Copyright 2026 EdgeChat AI, a subsidiary of Biostate AI.*
License: Edgepedia Community License 1.0, https://www.edgepedia.org/edgepedia/license