The Boltzmann Equation — A Mathematical Proof of Why Time Cannot Flow Backward

Boltzmann's Move: Count Heads, Ignore Identities

In 1872, Ludwig Boltzmann faced a tangle: a lump of gas contains 10²³ molecules, each at a different position and moving at a different velocity. Track the exact trajectory of every molecule? Impossible — and unnecessary.

His idea was elegant: slice the space holding the gas into tiny cells. Each cell is small enough, on the macroscopic scale, to be treated as a point, yet large enough, on the microscopic scale, to contain countless molecules. Then, rather than asking what each molecule is actually doing, ask only this: in this cell, how many molecules have velocities falling within a given range?

This statistical quantity is written f(t, x, v) — the density of molecules at time t, near position x, with velocity v.

With this f in hand, the macroscopic quantities fall out: particle density, temperature (the greater the molecules' kinetic energy — the faster they move — the hotter it is), and pressure (the sum of the impulses delivered as molecules strike the walls of the container). Even the ideal gas law can be derived from f.

Boltzmann wrote down an equation — the left-hand side describing the free flight of particles between collisions, the right-hand side describing the changes of velocity produced by collisions. In plain terms, it tallies the particles' "debits and credits": how many drift in from elsewhere, and how many leave because a collision with another particle has changed their velocity.

The H-Theorem and Its Name: Entropy

With an equation in hand, one can work out the direction of evolution. Boltzmann defined a quantity H:

H = ∫ f · log f  d(over all velocities and positions)

He found that whenever collisions occur, H must decrease. No matter how the collisions proceed — fast molecules striking slow ones, or slow ones striking fast — the arithmetic yields only a decrease in H, never an increase.

Attach a minus sign, and this quantity becomes the information entropy of information theory. A falling H says this: the system is running toward a more disordered, more uniform state. A tank of hot water will never spontaneously sort itself into one scalding half and one icy half — not because physics forbids it, but because the number of possible arrangements of 10²³ molecules is staggeringly large, and the uniform arrangements vastly outnumber the non-uniform ones. Flip 100 fair coins, and getting roughly fifty heads and fifty tails is far more likely than getting all heads — the same principle, scaled up to astronomical numbers.

Loschmidt's Challenge: Try Reversing the Velocities

In 1876, Josef Loschmidt posed a thought experiment: once the gas has filled its container, instantly reverse the velocity of every molecule.

Every molecule's direction of motion is flipped, yet Newton's laws have no objection — play a recording of billiard balls backward, and every frame of collision still obeys the physics. So from this moment on, the system will retrace its path: the gas contracts back into the left half. Entropy decreases, and H rises.

This collides head-on with Boltzmann's H-theorem. If the velocity-reversed state of the particles is a physically legitimate one, then the proof of the H-theorem must be hiding an extra assumption somewhere, quietly ruling out the possibility of such a retreat.

Molecular Chaos — An Unassuming but Pivotal Assumption

Boltzmann's reply was a little sly: "Very well — go ahead and try reversing the velocities of all 10²³ molecules." Indeed, to pull this off you would have to set every molecule's velocity with absurd precision. The gravitational disturbance caused by shifting a single gram of matter several light-years away would be enough to wreck such a delicate arrangement.

But the real problem lies in one step of the H-theorem's proof. Boltzmann assumed that there is no special correlation between the velocities of two colliding particles — that is, the probability that a particle with velocity v₁ strikes a particle with velocity v₂ is simply the product of their individual probabilities. This assumption is called Molecular Chaos (also known as the Stosszahlansatz).

Why is this assumption needed? Because with correlations, the situation is completely different. Imagine a swarm of particles whose velocities all point toward a common center — after colliding, they all converge on that center, until in the end every particle but one sits still, while a single "lone survivor" flies off carrying all the momentum. In reverse, that lone survivor shoots out and precisely blows apart a neatly ordered pile — a coincidence that could never arise naturally among 10²³ particles.

Return to Loschmidt's experiment: at the very instant the gas has just filled the container, subtle correlations do exist among the molecules — many still carry the "memory" of having just rushed over from the left. Let things develop normally forward in time, and these correlations do not affect the rise of entropy. But run them backward, and the correlations reveal themselves: the system retreats along a carefully engineered path.

The trouble is this: Boltzmann merely asserted that molecular chaos holds, without proving it mathematically. Hilbert's sixth problem, posed in 1900, was precisely a demand that someone turn this "assertion" into a mathematical theorem.

Lanford's Half-Step: A Proof That Lasted a Ten-Billionth of a Second

The road from Newtonian mechanics to the Boltzmann equation stalled for a full century.

In 1958, the American mathematician Harold Grad proposed a route: let the number of particles tend to infinity while their radius tends to zero, but hold a delicate balance so that the gas as a whole remains dilute. On average, each particle collides with only a constant number of companions. This is called the dilute-gas limit, or the Boltzmann–Grad limit.

In 1975, the American mathematician Oscar Lanford III took the first step within this framework. Starting from the Newtonian mechanics of hard spheres, he genuinely derived the Boltzmann equation — but only over an extremely short span of time: roughly a fraction of the mean interval between collisions, which in real time amounts to just 10^-10 seconds.

Lanford's tool was the collision diagram. The flight paths of the particles are drawn as lines — with the time axis running upward, each collision is a node where four lines meet. If a diagram contains no loops, then every probability can be factored into an independent product, and molecular chaos holds. Each tree corresponds to one term in the expansion of the Boltzmann equation; add up all the trees, and you have the equation itself.

The trouble lies precisely in the loops. When a particle that A has already struck comes back, after a chain of collisions, to strike A again, the two particles are no longer passing strangers: the cause A planted has returned as its effect. Lanford proved that in the dilute limit the probability of such loops tends to zero. But this is essentially the summation of a geometric series, which requires the absolute value of the common ratio to be less than 1. This means his proof can only hold over short times — once time grows long, loops proliferate, and the series no longer converges.

The Web-Pruning Algorithm: Severing Correlations, Following Only the Trunk

For the next fifty years, no one could extend Lanford's "valid for short times" result to arbitrarily long times. The longer the collisions go on, the more numerous and intricate the correlations among particles become.

In 2025, Yu Deng, Zaher Hani, and Xiao Ma crossed this barrier.

Their core idea sounds natural, yet its execution is extraordinarily delicate: slice the entire time axis into many layers. At each new time layer, "prune away" the parts that can already be computed independently — these trunks are already so close to the solution of the Boltzmann equation that there is no need to trace them further into the past. Only the parts that still carry correlations are pursued backward in time: how was their upper layer formed? Which particles does the layer above that one implicate?

The paper settles the matter precisely: to strike again a target particle of radius merely ε, the probability must be multiplied by a factor as small as ε^γ. The contribution of the loop structures in the collision diagram tends to zero at a rate of log(ε)^C × ε^γ — no logarithmic growth can outpace the decay of a positive power. Loops may be numerous, but the geometric coincidences they demand are ever more stringent.

And so, within the classical framework of a dilute hard-sphere gas and smooth solutions, the proof from Newtonian mechanics to the Boltzmann equation is at last complete. Moreover, the chain of derivation holds for exactly as long as the smooth solution of the Boltzmann equation itself persists. The time t can be any finite value fixed in advance — no longer a proof that runs for merely 10^-10 seconds.

A 125-Year Road

The three-step roadmap that Hilbert sketched in Paris in 1900 — Newtonian mechanics → Boltzmann equation → fluid equations — was completed, one step at a time, across three different eras.

• The second step (Boltzmann equation → fluid equations) was finished first. Hilbert himself opened the way in 1912; Chapman and Enskog each refined the expansion method during the First World War; and it was only after the 1980s that a generation of mathematicians supplied the rigorous proof.
• The first step (Newtonian mechanics → Boltzmann equation), from Lanford's half-step in 1975 to the full journey completed by Yu Deng's team in 2025, took exactly 50 years.

From a single small ball obeying reversible laws, to a glass of water, a gust of wind, and an irreversible macroscopic world — there is no magic in between, only a bridge built of probability, geometry, combinatorics, and analysis.