An Approximation Scheme for Reflected Stochastic Differential Equations

Lawrence Christopher Evans, Daniel W. Stroock

Published 2010-08-20Version 1

In this paper we consider the Stratonovich reflected stochastic differential equation $dX_t=\sigma(X_t)\circ dW_t+b(X_t)dt+dL_t$ in a bounded domain $\O$ which satisfies conditions, introduced by Lions and Sznitman, which are specified below. Letting $W^N_t$ be the $N$-dyadic piecewise linear interpolation of $W_t$ what we show is that one can solve the reflected ordinary differential equation $\dot X^N_t=\sigma(X^N_t)\dot W^N_t+b(X^N_t)+\dot L^N_t$ and that the distribution of the pair $(X^N_t,L^N_t)$ converges weakly to that of $(X_t,L_t)$. Hence, what we prove is a distributional version for reflected diffusions of the famous result of Wong and Zakai. Perhaps the most valuable contribution made by our procedure derives from the representation of $\dot X^N_t$ in terms of a projection of $\dot W_t^N$. In particular, we apply our result in hand to derive some geometric properties of coupled reflected Brownian motion in certain domains, especially those properties which have been used in recent work on the "hot spots" conjecture for special domain.