The Fermi-Pasta-Ulam Paradox: When Energy Refuses to Spread

A question about how heat spreads

In 1955, Enrico Fermi, John Pasta, Stanislaw Ulam and Mary Tsingou ran one of the first ever numerical experiments on a computer (the MANIAC at Los Alamos) to test a basic assumption of statistical mechanics: that a chain of coupled nonlinear oscillators, given enough time, should thermalize — its energy should spread out and equilibrate across all of its normal vibrational modes, the way heat mixes in a gas. They started with all the energy in the single lowest-frequency mode and expected it to gradually leak into the higher modes and stay spread out.

What actually happened: recurrence, not thermalization

It didn't equilibrate. Energy did leak from mode 1 into a handful of neighbouring low-order modes — but then, instead of continuing to spread, it flowed back, and after a surprisingly short time almost all of the energy had returned to mode 1, close to the system's original state. This near-return kept happening in a roughly periodic pattern, now called FPU recurrence, and it directly contradicted the expectation of ergodic thermalization. The result was so unexpected the team's report was essentially a puzzle for the physics community, not a confirmation.

H = Σᵢ [ ½pᵢ² + ½(qᵢ₊₁-qᵢ)² + α(qᵢ₊₁-qᵢ)³ ]
// linear chain (α=0): energy in each normal mode is separately conserved forever
// weak nonlinearity (α small): modes exchange energy... then mostly give it back
// this near-return, not full mixing, is the FPU recurrence

Why weak nonlinearity almost preserves the linear modes

A purely linear chain of coupled oscillators (α = 0) decomposes exactly into independent normal modes — sinusoidal standing waves, each with its own conserved energy and no way to exchange energy with any other mode, ever. Adding a small cubic nonlinearity couples the modes together, so energy can leak between them — but for weak enough coupling and appropriately chosen initial conditions, the system stays close to an underlying integrable structure for a very long time, and the trajectory in mode-energy space nearly retraces itself instead of wandering off to full equipartition. This connects directly to the KAM (Kolmogorov-Arnold-Moser) theorem, which guarantees that many quasi-periodic orbits of a nearly-integrable Hamiltonian system survive a small perturbation rather than immediately dissolving into chaos.

Solitons and the road to a different kind of physics

The real explanation for FPU recurrence came a decade later, in 1965, when Norman Zabusky and Martin Kruskal took the continuum limit of the FPU chain and found it was governed by the Korteweg-de Vries (KdV) equation — the same equation that describes shallow-water waves. They discovered that its solutions include solitons: solitary waves that pass through each other in collisions and re-emerge with their shape and speed completely unchanged, as if they hadn't interacted at all. That near-elastic, non-dispersive collision behaviour is exactly what makes the FPU chain's energy nearly retrace its initial state — the solitons carrying the energy return close to their starting configuration instead of dispersing it irreversibly. Zabusky and Kruskal's paper coined the very word "soliton," and the FPU experiment is now recognised as the discovery event for an entire branch of nonlinear wave physics, plus one of the first demonstrations that computer experiments could reveal real mathematical phenomena nobody had predicted analytically.

Why full thermalization does eventually happen

The FPU recurrence is not permanent immunity from thermalization — it is a very long transient. Push the nonlinearity strength above a threshold (studied extensively since the 1960s, notably by Izrailev and Chirikov) and the system's near-integrable structure breaks down: chaotic mixing takes over, the recurrences become less and less clean, and given enough time the energy does eventually spread toward equipartition across all modes, consistent with the original statistical-mechanics expectation. The FPU chain is a case study in a broader lesson: whether a nonlinear system thermalizes quickly, slowly, or only after an astronomically long recurrence-dominated transient depends sensitively on how strong the nonlinearity is relative to the coupling and how special the initial condition is.

What the simulation shows

The demo integrates a chain of masses connected by springs with a weak cubic (or quartic, the β-FPU variant) nonlinear term added to Hooke's law, starting with all the energy in the lowest mode, and plots each mode's energy over time. Watch the energy climb in modes 2 and 3, peak, and then — instead of continuing to spread to modes 4, 5, 6 — flow back down toward mode 1 again. That almost-but-not-quite return, repeating in a loosely periodic pattern, is precisely the anomaly Fermi, Pasta, Ulam and Tsingou found on the MANIAC in 1955.

Frequently asked questions

Try it live

Everything above runs in your browser — open FPU Recurrence and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.