no. 16 · Tuesday 17 November 2026
Moderate deviations for semilinear stochastic evolution equations in fast Markovian environments
Abstract
This work establishes moderate deviation principles for semilinear stochastic evolution equations with coefficients depending on fast Markov environments. The environment is either an exponentially ergodic Markov process on a Polish space, independent of the noise, or a finite-state Markov chain whose transition rates depend on time and on a finite-dimensional observation of the state. The rate function is quadratic in the residual of the linearized dynamics. Its covariance is the sum of the averaged covariance of the Wiener noise and the asymptotic covariance generated by the environment, given by a Green–Kubo formula in the independent case and by a Poisson equation in the switching case. The rate function is written with the Moore–Penrose inverse of the square root of this covariance and requires the residual to lie in its range, not merely in the closure. This allows degenerate covariances and nonclosed ranges. In the switching case, the dependence of the invariant distribution on the slow state adds an extra term to the linearized drift. Since the stochastic convolution is not a semimartingale, the proofs combine the regeneration split chain technique with Galerkin approximation. The assumptions are verified for stochastic reaction–diffusion equations, including space-time white noise in one spatial dimension.
- comments
- 68 pages, 2 figures, comments are welcome!
- fields
- Probability