In 1955, Fermi, Pasta, Ulam, and Tsingou simulated a chain of nonlinear oscillators expecting it to thermalize — to share energy equally among all modes. It didn't. Energy cycled among a few modes without spreading, launching the study of nonlinear dynamics and eventually chaos theory. Seventy years later, the question remains: when does a nonlinear lattice thermalize, and why?
Fu et al. (arXiv: 2603.23347) identify two universal classes. In lattices with extended normal modes — the standard case, where modes span the entire system — thermalization always occurs, with a time that scales as the inverse square of the nonlinear coupling strength. Weak nonlinearity means slow thermalization, but thermalization is inevitable. The system will equilibrate; it just takes longer than Fermi expected.
The second class has localized normal modes — modes trapped in finite regions by disorder. Here, thermalization fails completely. Arbitrarily weak perturbations cannot induce it. These are “thermal insulators”: systems that never reach equilibrium regardless of how long you wait.
The through-claim: the geometry of normal modes determines the fate of energy. If modes are extended, they form a connected resonance network that eventually distributes energy everywhere. If modes are localized, the network is fragmented, and energy stays trapped. Thermalization is a connectivity problem, not a timescale problem.
The twist: disorder accelerates thermalization in the first class by breaking translational symmetry and creating new resonance channels. The same disorder that localizes modes in the second class speeds up equilibration in the first. Disorder isn't universally helpful or harmful — its effect depends on the connectivity of the network it modifies.
Fu, Wang, Lin, He, Wang, Zhang & Zhao, 2603.23347. Statistical mechanics / nonlinear dynamics / thermalization / FPUT problem. Note (2026-11-06; this replaces a correction I wrote on 2026-11-05, which blamed the essay for differences that come from a revision): the first version of arXiv:2603.23347 (Fu and co-authors, on the Fermi-Pasta-Ulam-Tsingou problem) stated that classical lattice systems fall into two universal classes, one with extended normal modes where the thermalization time scales as g to the minus 2 so that thermalization is inevitable in the thermodynamic limit, and one with all normal modes localized where the relaxation time is nearly independent of system size and the exponent diverges as g goes to zero; the essay follows that abstract. The current version (v2) describes itself as a focused review (8 figures) of developments in near-integrable dynamics, wave resonances and phonon kinetic theory. It keeps the extended-mode case (with a sufficiently connected resonance network, thermalization time proportional to g to the minus 2) but says that for localized reference eigenmodes thermalization does not fail completely: low-order resonance networks fragment and thermalization is controlled by higher-order processes, with numerical regimes where the time scales as g to the minus 2, 4 and 6 and weak system-size dependence. So the essay's picture of two sharply separate classes, one of which never thermalizes, matches the first version and not the current one. I did not re-read the full paper. Added 2026-11-06: the abstract of arXiv:2603.23347 (now v2; first and current abstracts share about 20 percent of their wording) changed substantially after its first version, so some differences described in the notes above may come from that revision and not from an error in the essay (I may have worked from the earlier version). This was found by a later mechanical comparison of the first and current abstracts; I have not re-checked the essay claim by claim against the first version.