Rough stochastic differential equations

We build a hybrid theory of rough stochastic analysis which seamlessly combines the advantages of both It\^o's stochastic and Lyons' rough differential equations. This gives a direct and intrinsic understanding of multidimensional diffusion with Brownian noise $(B,\tilde B)$ $$ dY_t(\omega)=b(\omega,t,Y_t(\omega))dt+\sigma(\omega,t,Y_t(\omega))dB_t(\omega)+f({\omega,t},Y_t(\omega))d\tilde B_t \,, $$ in the annealed form, when conditioned on its environmental noise $\tilde B$. This situation arises naturally e.g. in filtering theory, for Feynman--Kac representations of solutions to stochastic partial differential equations, in Lions--Souganidis' theory of pathwise stochastic control, and for McKean--Vlasov stochastic differential equations with common noise. In fact, we establish well-posedness of rough stochastic differential equations, with $\tilde B$ replaced by a genuine rough path. As consequence, the `annealed' process $Y$ is a locally Lipschitz function of its environmental noise in rough path metrics. There is also interest in taking $\tilde B=\tilde B^H$, a fractional Brownian motion which fits our theory for $H>1/3$. Our assumptions for $b,\sigma$ agree with those from It\^o theory, those for $f$ with rough paths theory, including an extension of Davie's critical regularity result for deterministic rough differential equations. A major role in our analysis is played by a new scale of stochastic controlled rough paths spaces, related to a $(L^m,L^n)$-variant of stochastic sewing.Comment: Differential equations driven by Brownian and rough nois