Statistical inference for rough Burgers-type stochastic partial differential equations

Abstract

We consider the Burgers-type stochastic PDE \[ du = θ\,\partial_x^2 u\,dt + f(u)\,dt + g(u)\,\partial_x u\,dt + σ\,dW(t), \] driven by space--time white noise on the one-dimensional torus, where \(θ>0\) is the diffusivity parameter and \(σ>0\) the noise intensity. The transport term \(g(u)\,\partial_x u\) admits no classical interpretation, and we work with the mild solutions defined via rough paths following~\cite{Hairer2011}, for which the solution decomposes as \(u=ψ+R\) with \(ψ\) a stationary centred Gaussian process solving a damped stochastic heat equation and \(R\) a remainder. We first show that, away from the initial time, \(R\) is Lipschitz in space and Hölder in time of order strictly greater than \(1/2\), hence strictly smoother than \(ψ\) in both variables. We then compute exactly the spatial quadratic variation and the temporal quartic variation of \(ψ\), almost surely and not merely in \(L^{2}(Ω)\), and prove that the corresponding variations of \(u\) are unaffected by the remainder. Combined with an exact time-scaling identity, this separation of regularity scales yields strongly consistent pathwise estimators of \(θ\). To our knowledge this is the first inference procedure for an SPDE whose solution is defined only pathwise, as a measurable function of a rough path, rather than as an Itô process.