Smooth approximation of stochastic differential equations

Consider an It\^o process $X$ satisfying the stochastic differential equation $dX=a(X)\,dt+b(X)\,dW$ where $a,b$ are smooth and $W$ is a multidimensional Brownian motion. Suppose that $W_n$ has smooth sample paths and that $W_n$ converges weakly to $W$. A central question in stochastic analysis is to understand the limiting behaviour of solutions $X_n$ to the ordinary differential equation $dX_n=a(X_n)\,dt+b(X_n)\,dW_n$. The classical Wong-Zakai theorem gives sufficient conditions under which $X_n$ converges weakly to $X$ provided that the stochastic integral $\int b(X)\,dW$ is given the Stratonovich interpretation. The sufficient conditions are automatic in one dimension, but in higher dimensions the correct interpretation of $\int b(X)\,dW$ depends sensitively on how the smooth approximation $W_n$ is chosen. In applications, a natural class of smooth approximations arise by setting $W_n(t)=n^{-1/2}\int_0^{nt}v\circ \phi_s\,ds$ where $\phi_t$ is a flow (generated for instance by an ordinary differential equation) and $v$ is a mean zero observable. Under mild conditions on $\phi_t$ we give a definitive answer to the interpretation question for the stochastic integral $\int b(X)\,dW$. Our theory applies to Anosov or Axiom~A flows $\phi_t$, as well as to a large class of nonuniformly hyperbolic flows (including the one defined by the well-known Lorenz equations) and our main results do not require any mixing assumptions on $\phi_t$. The methods used in this paper are a combination of rough path theory and smooth ergodic theory.