← All papers
First page of Stratonovich Waves Driven by Space-Time Gaussian Noise

Stratonovich Waves Driven by Space-Time Gaussian Noise

Le Chen

math.PR Nov 17, 2026 · v1 math.AP
A Lean 4 companion repository fully checks some results and the finite and algebraic steps of others. Two of the four main theorems lie outside the Lean checks.
Motivated by Xia Chen's moment conjecture, we determine the exact second-moment asymptotics of the Stratonovich hyperbolic Anderson model in dimensions $d\in\{1,2,3\}$, driven by a centered Gaussian noise with temporal covariance $|s-t|^{-α_0}$ and an even, nonnegative, positive-definite tempered Radon spatial covariance $Γ$ homogeneous of order $-α$, where $0<α_0<1$, $0<α<d$ or $α=d=1$, and $α+α_0<2$. For initial displacement one and velocity zero, our main result is that $T^{-κ}\log\mathbb{E}u^2(T,x)$ converges, for every fixed $x$, to a constant $C_2(Γ,α_0)$, where $κ=(4-α-α_0)/(3-α)$. The constant is given by an intrinsic variational formula: a supremum over nonnegative smooth compactly supported space-time controls of twice the time integral of the square root of a Rayleigh value built from the covariance mean, minus half of the covariance variance. The variational class includes controls that do not separate in time and space, and the formula retains the original covariance amplitude, kinetic coefficient one and Gaussian cost. The constant is positive and finite when $Γ\neq0$. An independent upper bound establishes finiteness at the scale $T^κ$. The identification of the constant combines Gaussian phase operators, convergence from the regularized noise, estimates uniform in the spatial localization radius, and finite optimization with nonnegative Gram entries. We also bound the constant through spatial energies and occupation measures, with explicit bounds for spatial white noise in one dimension. The work was developed with AI assistance; the proofs have not yet been checked line by line by a human reader, and a public verification ledger records the status of each result.

The work determines the exact second-moment asymptotics of the Stratonovich hyperbolic Anderson model in dimensions 1–3, driven by Gaussian noise with fractional temporal covariance and homogeneous spatial covariance. It is motivated by Xia Chen's moment conjecture.

The constant is identified through Gaussian phase operators, convergence from regularized noise, estimates uniform in the localization radius, and finite optimization with nonnegative Gram entries. The work was developed with an AI research system (Amy) using several LLM families. A public verification ledger tracks checking status, and a Lean 4 companion verifies some results and the finite and algebraic steps of others.

T^{-κ} log E u²(T,x) converges to a positive, finite constant given by an intrinsic variational formula, with κ=(4-α-α0)/(3-α). Explicit bounds are given for one-dimensional spatial white noise. The full upper scale and the second-moment constant theorems are not covered by Lean.