Homogenisation of singular SPDEs

TL;DR

This paper develops a novel approach to homogenise a broad class of singular stochastic PDEs using regularity structures, establishing convergence results and renormalisation techniques for oscillatory equations like 2d g-PAM and ^4_3.

Contribution

It introduces a framework combining classical homogenisation with regularity structures, enabling analysis of singular SPDEs with oscillatory coefficients and establishing convergence and renormalisation results.

Findings

Proves convergence of solutions to homogenised equations under regularisation.

Identifies and separates small and large scale divergence terms.

Establishes renormalisation constants for joint limits of parameters.

Abstract

We introduce an approach to study homogenisation of a large class of singular SPDEs of the form $$ \partial_t u_\varepsilon - \nabla\cdot {A}(x/\varepsilon,t/\varepsilon^2) \nabla u_\varepsilon = F(x/\varepsilon , t/\varepsilon^2, u_\varepsilon , \nabla u_\varepsilon , \xi ) $$ which is based on the idea of importing (classical) homogenisation results into the framework of regularity structures and the insight that one can rewrite the SPDE under consideration in terms of a model, where the correctors (from homogenisation theory) are seen as further `abstract noises'. As applications, we establish periodic space-time homogenisation results for oscillatory generalisations of the 2d g-PAM and $\Phi^4_3$ equation proving that when the noise is regularised at scale $\delta\ll 1$ solutions to the equation with coefficient field ${A}(x/\varepsilon,t/\varepsilon^2)$, when appropriately renormalised, converge to solutions to the corresponding homogenised equation along any sequence $(\varepsilon,\delta)\to 0$. We make the observation that the unbounded divergences can be written as sums of two types of terms: `small scale' terms, the spatial dependence of which is an explicit local function of the coefficient field and `large scale' terms, which for logarithmic divergences are explicit involving the homogenised matrix and correctors. Furthermore, in order to recover the same solution to the corresponding homogenised equation along any joint limit $(\varepsilon, \delta)\to 0$ one has to subtract additional bounded renormalisation constants which appear due to oscillations at mesoscopic scales, as well as due to resonances between the coefficient field and the oscillations in the nonlinearity.