Fermi--Pasta--Ulam--Tsingou problems: Passage from Boltzmann to $q$-statistics

TL;DR

This study investigates how long-range interactions in a generalized FPU system lead to a transition from Boltzmann to q-statistics, with molecular dynamics confirming the shift in statistical behavior as interaction range varies.

Contribution

It demonstrates, through first-principle simulations, the transition from Boltzmann to q-statistics in FPU systems with long-range interactions, identifying the critical interaction decay parameter.

Findings

Boltzmann statistics holds for short-range interactions (α > 1).

q-statistics describes long-range interactions (α < 1).

The q-parameter decreases from ~5/3 to 1 as α increases.

Abstract

The Fermi-Pasta-Ulam (FPU) one-dimensional Hamiltonian includes a quartic term which guarantees ergodicity of the system in the thermodynamic limit. Consistently, the Boltzmann factor $P(\epsilon) \sim e^{-\beta \epsilon}$ describes its equilibrium distribution of one-body energies, and its velocity distribution is Maxwellian, i.e., $P(v) \sim e^{- \beta v^2/2}$. We consider here a generalized system where the quartic coupling constant between sites decays as $1/d_{ij}^{\alpha}$ $(\alpha \ge 0; d_{ij} = 1,2,\dots)$. Through {\it first-principle} molecular dynamics we demonstrate that, for large $\alpha$ (above $\alpha \simeq 1$), i.e., short-range interactions, Boltzmann statistics (based on the {\it additive} entropic functional $S_B[P(z)]=-k \int dz P(z) \ln P(z)$) is verified. However, for small values of $\alpha$ (below $\alpha \simeq 1$), i.e., long-range interactions, Boltzmann statistics dramatically fails and is replaced by q-statistics (based on the {\it nonadditive} entropic functional $S_q[P(z)]=k (1-\int dz [P(z)]^q)/(q-1)$, with $S_1 = S_B$). Indeed, the one-body energy distribution is q-exponential, $P(\epsilon) \sim e_{q_{\epsilon}}^{-\beta_{\epsilon} \epsilon} \equiv [1+(q_{\epsilon} - 1) \beta_{\epsilon}{\epsilon}]^{-1/(q_{\epsilon}-1)}$ with $q_{\epsilon} > 1$, and its velocity distribution is given by $P(v) \sim e_{q_v}^{ - \beta_v v^2/2}$ with $q_v > 1$. Moreover, within small error bars, we verify $q_{\epsilon} = q_v = q$, which decreases from an extrapolated value q $\simeq$ 5/3 to q=1 when $\alpha$ increases from zero to $\alpha \simeq 1$, and remains q = 1 thereafter.