Boltzmann's H-Theorem: Irreversibility and the Arrow of Time

1. The H-Functional and Entropy

To track the system's evolution towards equilibrium, we need a quantity that measures its "distance" from that final state. Ludwig Boltzmann introduced such a quantity, which he called the H-functional:

This functional takes the entire distribution function $f$ at a given time $t$ and maps it to a single number, $H(t)$. You might recognize its form. In Module 1, we defined the Shannon information entropy of a probability distribution $p_i$ as $S=-k{\sum }_i{p}_i\ln p_i$. The H-functional is the continuous analogue for the phase-space distribution $f$. The thermodynamic entropy of the gas is directly related to H by:

(Up to an additive constant which depends on normalization). Thus, $H(t)$ is essentially the negative of the system's entropy. A state with a low value of H is a high-entropy, disordered state. A state with a high value of H is a low-entropy, more ordered state.

Boltzmann's stroke of genius was to connect this statistical definition of entropy, based on the distribution function $f$, to the older, thermodynamic definition related to the number of accessible microstates, famously encapsulated in the formula on his tombstone, $S=k\log W$. The Cambridge University video featuring mathematician Cédric Villani provides a beautiful explanation of how these two definitions are reconciled.

The H-theorem is the statement that $H(t)$ can only decrease over time. Let's now prove it.

2. The Proof of the H-Theorem

The H-theorem states that for any system evolving according to the Boltzmann equation,

The proof is a masterclass in applying the structure of the Boltzmann equation. We begin by differentiating the definition of $H(t)$ with respect to time:

The integral of the second term, $\int \partial f/\partial t{d}^{3}r{d}^{3}p$, is $\frac{d}{dt}\int f{d}^{3}r{d}^{3}p=\frac{dN}{dt}$, where $N$ is the total number of particles. Since collisions conserve particle number, $dN/dt=0$, and this term vanishes. We are left with:

Now, we substitute the full Boltzmann equation for $\partial f/\partial t$:

The contribution from the "streaming" terms (the first two on the right) is zero. This can be shown by integration by parts, assuming $f$ vanishes at the boundaries of our phase-space volume. For example, for the spatial term:

Integrating the inner integral by parts with respect to $\mathbf{r}$ gives $\int f{\nabla }_{\mathbf{r}}(\ln f\cdot \mathbf{v})d^{3}r=\int f(\frac{1}{f}{\nabla }_{\mathbf{r}}f\cdot \mathbf{v})d^{3}r=\int ({\nabla }_{\mathbf{r}}f\cdot \mathbf{v})d^{3}r$. The net result is zero when added to the original term. A similar argument holds for the force term.

Therefore, the entire change in H is due to collisions:

This is where the real work begins. The proof is laid out with exceptional clarity in David Tong's lecture notes. It involves a sequence of symmetrization arguments that are fundamental to working with the collision integral.

Let's summarize the key result from that derivation. After the symmetrization steps, one arrives at the expression:

where $f_1=f({\mathbf{p}}_1)$, $f_1^{\prime }=f({\mathbf{p}}_1^{\prime })$, etc., and $\omega (\dots )$ is the positive-definite scattering rate. The integrand has the form $-(\ln x-\ln y)(x-y)$. For any positive $x$ and $y$, this expression is always less than or equal to zero. If $x>y$, then $\ln x>\ln y$, and the product is positive. If $x<y$, then $\ln x<\ln y$, and the product is again positive. The leading minus sign ensures the whole integrand is non-positive.

Since the entire integrand is non-positive, the integral itself must be non-positive. We have thus proven Boltzmann's H-theorem:

The equality $dH/dt=0$ holds if and only if the integrand is zero for all possible collisions. This requires $f_1^{\prime }{f}_2^{\prime }=f_1{f}_2$, which is precisely the condition of detailed balance that defines the equilibrium state.

This proves that under the dynamics of the Boltzmann equation, the system evolves in such a way that its H-functional continuously decreases (and its entropy continuously increases) until it reaches the Maxwell-Boltzmann distribution, at which point H becomes constant.

3. Implications: The Arrow of Time

The H-theorem is a monumental achievement. It derives the second law of thermodynamics—the monotonic increase of entropy—from the microscopic dynamics of particles. In doing so, it establishes an "arrow of time": the direction in which $H$ decreases is the future.

However, this raises a profound conceptual problem. The underlying laws of motion (Newton's equations for classical particles) are perfectly time-reversible. If you were to watch a movie of two particles colliding, the movie played in reverse would also depict a perfectly valid physical collision. How can we derive irreversible behavior from reversible laws? As Cédric Villani puts it in the video, this was the start of many controversies.

The irreversibility was not conjured from thin air. It was subtly introduced when we formulated the Boltzmann equation. The crucial step was the Stosszahlansatz, or the assumption of molecular chaos. We assumed that the velocities of two particles are uncorrelated before they collide, i.e., $f_2({\mathbf{p}}_1,{\mathbf{p}}_2)=f_1({\mathbf{p}}_1)f_1({\mathbf{p}}_2)$. This is an assumption about the past. Had we assumed they were uncorrelated after the collision, we would have derived an "anti-Boltzmann equation" for which $dH/dt\ge 0$, and entropy would always decrease into the future.

Our physical intuition suggests that particles are correlated after they interact, but these correlations are "forgotten" before their next collision with other, random particles. The assumption of pre-collision chaos formalizes this intuition and is the very thing that breaks the time-reversal symmetry of the underlying dynamics, giving time its arrow.

This apparent contradiction sparked intense debate, leading to two famous objections:

1. Loschmidt's Reversibility Paradox: If we let the system evolve until some time $t$ and then magically reverse the velocity of every single particle, the time-reversed laws of mechanics dictate that the system should retrace its path, and H should increase back to its initial value. Why does the H-theorem forbid this?
2. Zermelo's Recurrence Paradox: Based on the Poincaré recurrence theorem, any isolated mechanical system in a finite volume must eventually return arbitrarily close to its initial state. This would imply that entropy must eventually decrease.

These are not trivial objections; they force us to confront the statistical and probabilistic nature of the second law.

Conclusion

In this lesson, we have reached a pivotal point in our study of kinetic theory. We have formally established the relentless march of a dilute gas towards equilibrium.

Key Takeaways:

• Boltzmann's H-functional, $H=\int f\ln fd\Gamma$, is equivalent to the negative of the system's entropy and serves as a measure of its distance from equilibrium.
• The H-theorem states that for a system governed by the Boltzmann equation, $dH/dt\le 0$. The entropy never decreases.
• The proof relies on the structure of the collision integral and, crucially, the symmetry properties of the collision process.
• Equality, $dH/dt=0$, holds if and only if the system satisfies detailed balance ($f_1{f}_2=f_1^{\prime }{f}_2^{\prime }$), which is the condition for equilibrium.
• The H-theorem provides a microscopic origin for the arrow of time, but this irreversibility is not a feature of the fundamental laws of motion. It is a consequence of the statistical assumption of molecular chaos (Stosszahlansatz).

We have established the theorem and identified the profound paradoxes it raises concerning reversibility. In our next lesson, we will dissect these paradoxes in detail and understand their resolution, which lies at the very heart of what statistical mechanics is all about.