A Girsanov approach to slow parameterizing manifolds in the presence of noise (1903.08598v2)

Abstract: We consider a three-dimensional slow-fast system arising in fluid dynamics with quadratic nonlinearity and additive noise. The associated deterministic system of this stochastic differential equation (SDE) exhibits a periodic orbit and a slow manifold. We show that in presence of noise, the deterministic slow manifold can be viewed as an approximate parameterization of the fast variable of the SDE in terms of the slow variables, for certain parameter regimes. We exploit this fact to obtain a two dimensional reduced model from the original stochastic system, which results into a Hopf normal form with additive noise. Both, the original as well as the reduced system admit ergodic invariant measures describing their respective long-time behaviour. It is then shown that for a suitable Wasserstein metric on a subset of the space of probability measures on the phase space, the discrepancy between the marginals along the radial component of each invariant measure is controlled by a parameterization defect measuring the quality of the parameterization. An important technical tool to arrive at this result is Girsanov's theorem that allows us to derive such error estimates in presence of an oscillatory instability. This approach is finally extended to parameter regimes for which the variable to parameterize is no longer evolving on a faster timescale than that of the resolved variables. There also, error estimates involving the Wasserstein metric are derived but this time for reduced systems obtained from stochastic parameterizing manifolds involving path-dependent coefficients to cope with such challenging regimes.