Fermi-Pasta-Ulam-Tsingou Equation Overview

• The Fermi-Pasta-Ulam-Tsingou equation is a discrete Hamiltonian model describing a 1D lattice with weak anharmonic nearest-neighbor interactions, central to studies of recurrence, metastability, and energy transfer.
• It encompasses various chains (α, β, α+β) and supports continuum limits reducing to integrable equations like KdV and mKdV, as well as kinetic and Burgers-type regimes, highlighting diverse asymptotic behaviors.
• Resonance analysis and divisor effects in finite systems reveal that energy redistribution and eventual thermalization depend on lattice arithmetic, with explicit links to nonlinear mode interaction and coherent state persistence.

The Fermi–Pasta–Ulam–Tsingou (FPUT) equation denotes, in its primary sense, the discrete Hamiltonian equations of motion for a one-dimensional lattice of particles coupled by weakly anharmonic nearest-neighbor springs; in modern usage it also denotes the continuum and kinetic equations that arise from this lattice in distinguished asymptotic regimes. The model is central to the study of recurrence, metastability, prethermalization, resonant energy transfer, and eventual equipartition. In current treatments, the term encompasses the $\alpha$-chain, the $\beta$-chain, and the combined $\alpha+\beta$ chain, together with their KdV, mKdV, Gardner, Burgers, Toda, and wave-kinetic descriptions (Gallone, 22 Sep 2025).

1. Discrete Hamiltonian formulation and normal-mode structure

In the periodic setting, the FPUT chain consists of $N$ identical masses with nearest-neighbor interaction and Hamiltonian

$H = H_0 + H_3 + H_4,$

$$H_0 = \sum_{j=1}^N \left[ \frac{p_j^2}{2m} + \frac{\kappa}{2}(q_j-q_{j+1})^2 \right],\qquad H_3 = \frac{\alpha}{3}\sum_{j=1}^N (q_j-q_{j+1})^3,\qquad H_4 = \frac{\beta}{4}\sum_{j=1}^N (q_j-q_{j+1})^4.$$

The corresponding Newton equation is

$$m \ddot{q}_j = \kappa(q_{j+1}+q_{j-1}-2q_j) + \alpha\big[(q_{j+1}-q_j)^2-(q_j-q_{j-1})^2\big] + \beta\big[(q_{j+1}-q_j)^3-(q_j-q_{j-1})^3\big].$$

This periodic formulation is the common starting point for resonance analysis and wave-kinetic theory (Bustamante et al., 2018).

A second standard realization uses fixed-end boundary conditions. For the dimensionless $\beta$-FPUT chain with $q_0=q_{N+1}=0$, the Hamiltonian is

$$H=\sum_{j=1}^{N}\left[\frac{1}{2}p_j^2+\frac{1}{2}(q_{j+1}-q_j)^2+\frac{\beta}{4}(q_{j+1}-q_j)^4\right],$$

and the equations of motion are

$\beta$0

For fixed ends, the linear frequencies are

$\beta$1

whereas for the periodic dimensionless chain they are

$\beta$2

The standard linear-mode energy is

$\beta$3

and the paradigmatic recurrence experiment excites the first mode, typically through

$\beta$4

for fixed ends, or its periodic traveling-wave analogue for the ring (Pace et al., 2019).

The distinction between periodic and fixed-end chains is not merely technical. It changes the Fourier basis, the dispersion relation, and the admissible exact resonances. At the same time, the two settings display the same broad phenomenology: low-dimensional initial excitations can remain organized for unexpectedly long times, and the approach to thermalization is highly structured rather than immediate. This suggests that the “FPUT equation” is best understood as a family of closely related lattice equations rather than a single normal form.

2. Exact resonances, divisibility effects, and finite-$\beta$5 energy transfer

In the weakly nonlinear regime, exact discrete resonances organize irreversible energy transfer in mode space. For the periodic FPUT chain with dispersion $\beta$6, exact $\beta$7-wave resonances satisfy modular momentum conservation together with exact frequency balance. A fundamental structural result is that $\beta$8-wave resonances, and more generally $\beta$9 processes, are forbidden by strict subadditivity of the dispersion. Consequently, $\alpha+\beta$0-wave resonances must be of $\alpha+\beta$1 type, and in the discrete finite chain they are integrable in the Birkhoff or resonant normal-form sense and do not produce spectrum-wide mixing (Bustamante et al., 2018).

Higher-order resonances are therefore decisive. The arithmetic classification based on cyclotomic polynomials shows that $\alpha+\beta$2-wave resonances always exist for any $\alpha+\beta$3, whereas $\alpha+\beta$4-wave resonances exist if $\alpha+\beta$5 is divisible by $\alpha+\beta$6 and $\alpha+\beta$7. The same analysis proves that the mode set can decompose into dynamically independent components whose number depends sensitively on the odd divisors of $\alpha+\beta$8 not divisible by $\alpha+\beta$9. This divisibility dependence explains why finite-$N$0 thermalization routes vary strongly with particle number: when $N$1, quintets are absent and sextuplets dominate; when $N$2, quintets enable faster component-wise mixing, while sextuplets bridge components and provide a restoring mechanism for full-scale thermalization (Bustamante et al., 2018).

This finite-$N$3 picture connects directly to kinetic theory. For the $N$4-FPUT chain in the kinetic limit $N$5, $N$6 with $N$7 and $N$8, the resonant $N$9-wave channel becomes the leading kinetic mechanism. A rigorous derivation for a reduced resonant evolution proves the wave kinetic equation up to the sub-kinetic time scale

$H = H_0 + H_3 + H_4,$0

after a deterministic phase renormalization removes dangerous self-interactions (Wu, 3 Jun 2025). A later rigorous treatment of the full $H = H_0 + H_3 + H_4,$1-FPUT system incorporates the non-resonant terms directly into the diagrammatic expansion and reaches times up to $H = H_0 + H_3 + H_4,$2, with

$H = H_0 + H_3 + H_4,$3

thereby validating the onset of thermalization in the weakly nonlinear large-box regime (Vassilev et al., 19 May 2026).

A common misconception is that “the” thermalization mechanism in FPUT chains is uniform across all system sizes. The finite periodic chain is arithmetic: its exact resonant manifolds depend on the divisor structure of $H = H_0 + H_3 + H_4,$4, whereas in the thermodynamic limit Fourier space becomes dense and the kinetic description becomes appropriate. The discrete resonance theory and the kinetic theory therefore describe different but compatible asymptotic layers of the same model.

3. Continuum reductions and integrable asymptotics

Long-wavelength asymptotics map the discrete FPUT lattice to several integrable or asymptotically integrable continuum equations. For quasi-unidirectional waves in the $H = H_0 + H_3 + H_4,$5 lattice, the first-order continuum reduction is KdV, and the second-order reduced dynamics is governed by a combination of the first two nontrivial equations in the KdV hierarchy. At third order, a combination of the first three nontrivial hierarchy equations appears only when

$H = H_0 + H_3 + H_4,$6

a condition satisfied by the Toda chain. This identifies a precise algebraic criterion for persistence of asymptotic integrability beyond leading order (Gallone et al., 2020).

For the $H = H_0 + H_3 + H_4,$7-FPUT chain, the continuum limit is the modified Korteweg–de Vries equation

$H = H_0 + H_3 + H_4,$8

Here $H = H_0 + H_3 + H_4,$9 yields focusing mKdV with bright solitons

$$H_0 = \sum_{j=1}^N \left[ \frac{p_j^2}{2m} + \frac{\kappa}{2}(q_j-q_{j+1})^2 \right],\qquad H_3 = \frac{\alpha}{3}\sum_{j=1}^N (q_j-q_{j+1})^3,\qquad H_4 = \frac{\beta}{4}\sum_{j=1}^N (q_j-q_{j+1})^4.$$0

whereas $$H_0 = \sum_{j=1}^N \left[ \frac{p_j^2}{2m} + \frac{\kappa}{2}(q_j-q_{j+1})^2 \right],\qquad H_3 = \frac{\alpha}{3}\sum_{j=1}^N (q_j-q_{j+1})^3,\qquad H_4 = \frac{\beta}{4}\sum_{j=1}^N (q_j-q_{j+1})^4.$$1 yields defocusing mKdV with kinks

$$H_0 = \sum_{j=1}^N \left[ \frac{p_j^2}{2m} + \frac{\kappa}{2}(q_j-q_{j+1})^2 \right],\qquad H_3 = \frac{\alpha}{3}\sum_{j=1}^N (q_j-q_{j+1})^3,\qquad H_4 = \frac{\beta}{4}\sum_{j=1}^N (q_j-q_{j+1})^4.$$2

The sign of $$H_0 = \sum_{j=1}^N \left[ \frac{p_j^2}{2m} + \frac{\kappa}{2}(q_j-q_{j+1})^2 \right],\qquad H_3 = \frac{\alpha}{3}\sum_{j=1}^N (q_j-q_{j+1})^3,\qquad H_4 = \frac{\beta}{4}\sum_{j=1}^N (q_j-q_{j+1})^4.$$3 therefore changes the solitary-wave content of the continuum description and, through it, the recurrence dynamics of the discrete chain (Pace et al., 2019).

When both quadratic and cubic nearest-neighbor nonlinearities are retained, the continuum reduction is the Gardner equation

$$H_0 = \sum_{j=1}^N \left[ \frac{p_j^2}{2m} + \frac{\kappa}{2}(q_j-q_{j+1})^2 \right],\qquad H_3 = \frac{\alpha}{3}\sum_{j=1}^N (q_j-q_{j+1})^3,\qquad H_4 = \frac{\beta}{4}\sum_{j=1}^N (q_j-q_{j+1})^4.$$4

Via Hirota bilinearization, this equation supports exact multi-soliton solutions and table-top soliton molecules generated by velocity resonance. The same framework recovers the cubic-only chain as mKdV and the quadratic-only chain as KdV through the stated Galilean and Miura interconnections (Kirane et al., 2023).

A different thermodynamic scaling suppresses dispersion and produces a Burgers regime. In that setting, a near-identity Hamiltonian normal form reduces the FPUT dynamics to a pair of generalized inviscid Burgers equations for dressed left- and right-moving fields. The resulting shock time

$$H_0 = \sum_{j=1}^N \left[ \frac{p_j^2}{2m} + \frac{\kappa}{2}(q_j-q_{j+1})^2 \right],\qquad H_3 = \frac{\alpha}{3}\sum_{j=1}^N (q_j-q_{j+1})^3,\qquad H_4 = \frac{\beta}{4}\sum_{j=1}^N (q_j-q_{j+1})^4.$$5

predicts when the Fourier spectrum develops the universal law

$$H_0 = \sum_{j=1}^N \left[ \frac{p_j^2}{2m} + \frac{\kappa}{2}(q_j-q_{j+1})^2 \right],\qquad H_3 = \frac{\alpha}{3}\sum_{j=1}^N (q_j-q_{j+1})^3,\qquad H_4 = \frac{\beta}{4}\sum_{j=1}^N (q_j-q_{j+1})^4.$$6

After approximately two shock times, the spectrum reaches a regime close to

$$H_0 = \sum_{j=1}^N \left[ \frac{p_j^2}{2m} + \frac{\kappa}{2}(q_j-q_{j+1})^2 \right],\qquad H_3 = \frac{\alpha}{3}\sum_{j=1}^N (q_j-q_{j+1})^3,\qquad H_4 = \frac{\beta}{4}\sum_{j=1}^N (q_j-q_{j+1})^4.$$7

which persists over an extensive time window before equipartition (Gallone et al., 2022). A full Hamiltonian perturbative derivation of this Burgers reduction, including the short-time law

$$H_0 = \sum_{j=1}^N \left[ \frac{p_j^2}{2m} + \frac{\kappa}{2}(q_j-q_{j+1})^2 \right],\qquad H_3 = \frac{\alpha}{3}\sum_{j=1}^N (q_j-q_{j+1})^3,\qquad H_4 = \frac{\beta}{4}\sum_{j=1}^N (q_j-q_{j+1})^4.$$8

has since been given for the periodic $$H_0 = \sum_{j=1}^N \left[ \frac{p_j^2}{2m} + \frac{\kappa}{2}(q_j-q_{j+1})^2 \right],\qquad H_3 = \frac{\alpha}{3}\sum_{j=1}^N (q_j-q_{j+1})^3,\qquad H_4 = \frac{\beta}{4}\sum_{j=1}^N (q_j-q_{j+1})^4.$$9 chain (Gallone et al., 2024).

These reductions clarify why distinct asymptotic equations coexist in the FPUT literature. KdV and mKdV describe weakly dispersive, long-wave, near-integrable dynamics; Burgers describes the zero-dispersion steepening regime; Toda provides the integrable neighbor governing adiabatic invariants and metastability. This suggests that continuum limits in FPUT theory are regime-selective rather than mutually exclusive (Gallone, 22 Sep 2025).

4. Recurrences, metastability, chaos, and ergodization

The classical recurrence problem remains one of the most quantitative parts of FPUT theory. For the fixed-end $$m \ddot{q}_j = \kappa(q_{j+1}+q_{j-1}-2q_j) + \alpha\big[(q_{j+1}-q_j)^2-(q_j-q_{j-1})^2\big] + \beta\big[(q_{j+1}-q_j)^3-(q_j-q_{j-1})^3\big].$$0 chain with first-mode excitation, the first recurrence time rescales as

$$m \ddot{q}_j = \kappa(q_{j+1}+q_{j-1}-2q_j) + \alpha\big[(q_{j+1}-q_j)^2-(q_j-q_{j-1})^2\big] + \beta\big[(q_{j+1}-q_j)^3-(q_j-q_{j-1})^3\big].$$1

For large $$m \ddot{q}_j = \kappa(q_{j+1}+q_{j-1}-2q_j) + \alpha\big[(q_{j+1}-q_j)^2-(q_j-q_{j-1})^2\big] + \beta\big[(q_{j+1}-q_j)^3-(q_j-q_{j-1})^3\big].$$2, $$m \ddot{q}_j = \kappa(q_{j+1}+q_{j-1}-2q_j) + \alpha\big[(q_{j+1}-q_j)^2-(q_j-q_{j-1})^2\big] + \beta\big[(q_{j+1}-q_j)^3-(q_j-q_{j-1})^3\big].$$3 depends only on $$m \ddot{q}_j = \kappa(q_{j+1}+q_{j-1}-2q_j) + \alpha\big[(q_{j+1}-q_j)^2-(q_j-q_{j-1})^2\big] + \beta\big[(q_{j+1}-q_j)^3-(q_j-q_{j-1})^3\big].$$4. In the nearly linear regime, it is linear in $$m \ddot{q}_j = \kappa(q_{j+1}+q_{j-1}-2q_j) + \alpha\big[(q_{j+1}-q_j)^2-(q_j-q_{j-1})^2\big] + \beta\big[(q_{j+1}-q_j)^3-(q_j-q_{j-1})^3\big].$$5 with positive slope on both sides of $$m \ddot{q}_j = \kappa(q_{j+1}+q_{j-1}-2q_j) + \alpha\big[(q_{j+1}-q_j)^2-(q_j-q_{j-1})^2\big] + \beta\big[(q_{j+1}-q_j)^3-(q_j-q_{j-1})^3\big].$$6, while in the highly nonlinear regime it satisfies

$$m \ddot{q}_j = \kappa(q_{j+1}+q_{j-1}-2q_j) + \alpha\big[(q_{j+1}-q_j)^2-(q_j-q_{j-1})^2\big] + \beta\big[(q_{j+1}-q_j)^3-(q_j-q_{j-1})^3\big].$$7

The prefactors differ for $$m \ddot{q}_j = \kappa(q_{j+1}+q_{j-1}-2q_j) + \alpha\big[(q_{j+1}-q_j)^2-(q_j-q_{j-1})^2\big] + \beta\big[(q_{j+1}-q_j)^3-(q_j-q_{j-1})^3\big].$$8 and $$m \ddot{q}_j = \kappa(q_{j+1}+q_{j-1}-2q_j) + \alpha\big[(q_{j+1}-q_j)^2-(q_j-q_{j-1})^2\big] + \beta\big[(q_{j+1}-q_j)^3-(q_j-q_{j-1})^3\big].$$9 because the negative-$\beta$0 continuum limit involves soliton–kink interactions that accelerate rephasing. The same study reports $\beta$1-dependent critical energies above which recurrences do not form within the measured times and argues that, in the thermodynamic limit with extensive energy, recurrences deteriorate and ultimately disappear (Pace et al., 2019).

Metastability can also be measured by observables tied to nearby integrable dynamics. In the FPUT-$\beta$2 chain, a Toda integral constructed from $\beta$3 acts as an adiabatic invariant and provides an operational equilibrium time. Its ergodization time is systematically longer than both the inverse maximum Lyapunov exponent and the Lyapunov saturation time. For large chains it becomes essentially system-size independent, whereas below a critical size it grows dramatically. The measured crossover is summarized by

$\beta$4

which the paper relates to a possible KAM-like regime (Christodoulidi et al., 11 Nov 2025).

A complementary viewpoint is mean-field rather than perturbative. For the quartic $\beta$5 chain with fixed ends, a mean-field Hamiltonian

$\beta$6

is constructed so that

$\beta$7

tends to zero in probability and almost surely under the Gibbs measure as $\beta$8. The resulting effective normal modes have renormalized frequencies

$\beta$9

and a phenomenological Langevin closure reproduces mode-energy relaxation and spectral line shapes across energies from the quasi-integrable regime to the strongly chaotic one (Ponno et al., 30 Nov 2025).

Chaos at low energy can itself be modeled as a random perturbation of the Toda chain. For the fixed-end $q_0=q_{N+1}=0$0 chain, the largest Lyapunov exponent is controlled by the interaction between a few soliton-like Toda modes and a background of radiative modes acting as an intrinsic bath. Retaining only one randomly perturbed Toda soliton-like mode reproduces the observed power law

$q_0=q_{N+1}=0$1

for large chains, and analogous exponents for quasi-Toda variants with higher-order matching to the Toda potential (Goldfriend, 2021).

Taken together, these results resolve a second misconception: positive Lyapunov exponents do not by themselves imply rapid ergodization. Weak chaos, metastable packets, adiabatic invariants, and very slow action diffusion can coexist over parametrically long windows before true thermal equilibrium is reached.

5. Nonequilibrium transport, damping, and finite-temperature dynamics

In the $q_0=q_{N+1}=0$2-FPUT chain, the microscopic lattice equations admit a mesoscopic kinetic closure in terms of a spatially nonhomogeneous wave kinetic equation for the occupation density $q_0=q_{N+1}=0$3: $q_0=q_{N+1}=0$4 with $q_0=q_{N+1}=0$5 and a $q_0=q_{N+1}=0$6-wave collision integral generated by the quartic nonlinearity. With thermostats imposed as incoming Rayleigh–Jeans spectra at $q_0=q_{N+1}=0$7 and $q_0=q_{N+1}=0$8, the stationary conductivity

$q_0=q_{N+1}=0$9

crosses over from ballistic scaling $$H=\sum_{j=1}^{N}\left[\frac{1}{2}p_j^2+\frac{1}{2}(q_{j+1}-q_j)^2+\frac{\beta}{4}(q_{j+1}-q_j)^4\right],$$0 at small $$H=\sum_{j=1}^{N}\left[\frac{1}{2}p_j^2+\frac{1}{2}(q_{j+1}-q_j)^2+\frac{\beta}{4}(q_{j+1}-q_j)^4\right],$$1 to anomalous scaling

$$H=\sum_{j=1}^{N}\left[\frac{1}{2}p_j^2+\frac{1}{2}(q_{j+1}-q_j)^2+\frac{\beta}{4}(q_{j+1}-q_j)^4\right],$$2

at large $$H=\sum_{j=1}^{N}\left[\frac{1}{2}p_j^2+\frac{1}{2}(q_{j+1}-q_j)^2+\frac{\beta}{4}(q_{j+1}-q_j)^4\right],$$3. The same kinetic analysis identifies a critical scale

$$H=\sum_{j=1}^{N}\left[\frac{1}{2}p_j^2+\frac{1}{2}(q_{j+1}-q_j)^2+\frac{\beta}{4}(q_{j+1}-q_j)^4\right],$$4

separating low-$$H=\sum_{j=1}^{N}\left[\frac{1}{2}p_j^2+\frac{1}{2}(q_{j+1}-q_j)^2+\frac{\beta}{4}(q_{j+1}-q_j)^4\right],$$5 ballistic phonons from high-$$H=\sum_{j=1}^{N}\left[\frac{1}{2}p_j^2+\frac{1}{2}(q_{j+1}-q_j)^2+\frac{\beta}{4}(q_{j+1}-q_j)^4\right],$$6 diffusive phonons. In this picture, anomalous conduction arises from the coexistence of nearly noninteracting low-wavenumber modes and locally equilibrated high-wavenumber modes. The same framework yields second-sound–like ballistic peaks with speeds $$H=\sum_{j=1}^{N}\left[\frac{1}{2}p_j^2+\frac{1}{2}(q_{j+1}-q_j)^2+\frac{\beta}{4}(q_{j+1}-q_j)^4\right],$$7 and a diffusive heat peak with effective diffusion coefficient $$H=\sum_{j=1}^{N}\left[\frac{1}{2}p_j^2+\frac{1}{2}(q_{j+1}-q_j)^2+\frac{\beta}{4}(q_{j+1}-q_j)^4\right],$$8 in the reported setup (Vita et al., 2022).

Finite temperature changes recurrence qualitatively. For the $$H=\sum_{j=1}^{N}\left[\frac{1}{2}p_j^2+\frac{1}{2}(q_{j+1}-q_j)^2+\frac{\beta}{4}(q_{j+1}-q_j)^4\right],$$9-FPUT chain with a spatially sinusoidal initial temperature profile, the continuum thermoelastic description coupled to the ballistic heat equation gives

$\beta$00

and a resonant mechanical response

$\beta$01

This “ballistic resonance” converts thermal energy into mechanical energy at early times, but at later times the mechanical energy decays monotonically. The study concludes that the well-known zero-temperature recurrence paradox is eliminated at finite temperatures in this setting (Kuzkin et al., 2019).

Weak dissipation also changes recurrence symmetry. In an optical NLSE analog of FPUT recurrence, small linear attenuation induces separatrix crossing at multiple critical losses,

$\beta$02

switching the dynamics between unshifted and shifted recurrence families. This does not suppress recurrence outright, but it destroys the conservative phase-space separation that protects the undamped dynamics (Vanderhaegen et al., 2022).

These nonequilibrium studies emphasize that “recurrence versus thermalization” is not a binary opposition. Boundary driving, ballistic heat transport, and weak damping all preserve recognizable FPUT structures while changing the relevant observables from return peaks to currents, symmetry classes, and transport exponents.

6. Generalizations: dimers, nonlocal solitary waves, and disorder

The FPUT equation extends naturally to polyatomic and heterogeneous lattices. In the dimer FPUT lattice, masses and/or spring potentials alternate with period $\beta$03, and the traveling-wave problem can be written as

$\beta$04

For nonsymmetric dimers, the existence of small-amplitude periodic traveling waves is proved by exploiting the gradient structure and translation invariance of the traveling-wave operator. The key point is that derivative orthogonality,

$\beta$05

eliminates one solvability condition and permits a bifurcation argument with a two-dimensional kernel (Faver et al., 2024).

Diatomic lattices also support nonlocal solitary waves. In the equal-mass limit, a monatomic solitary wave perturbs into a micropteron: a localized core together with a small periodic tail whose amplitude is

$\beta$06

rather than beyond all algebraic orders. The construction depends on a hidden solvability condition for an advance–delay operator and on asymptotically sinusoidal Jost solutions (Faver et al., 2019). Numerical continuation then reveals a connected landscape of micropterons, nanopterons, and solitary waves in the diatomic problem and gives evidence that selected diatomic solitary waves are stable under the full FPUT dynamics (Faver et al., 2020).

A different generalization introduces heterogeneity directly into the lattice couplings. In a non-Hamiltonian disordered $\beta$07-FPUT variant, variability is modeled by i.i.d. Gaussian factors $\beta$08 with tolerance $\beta$09. A two-mode multiple-scale reduction yields envelope equations whose effective coupling

$\beta$10

crosses zero at

$\beta$11

Below this threshold, energy transfer out of the first mode is strongly reduced and the chain exhibits $\beta$12-space localization; above it, the reduced dynamics acquires an unbounded sector and the full non-Hamiltonian system can blow up in finite time. Chaos diagnostics based on the maximum Lyapunov exponent and the Smaller Alignment Index show that when localization is strong, the probability of chaos increases with system size (Zulkarnain et al., 2022).

These extensions show that the FPUT equation is not restricted to a monatomic, homogeneous, conservative chain. Alternating masses, oscillatory tails, and disorder all preserve the core competition between discreteness, nonlinearity, and dispersion, while changing the spectrum of admissible coherent structures and the mechanisms by which energy remains trapped, leaks, or delocalizes.