A Feynman-Kac result via Markov BSDEs with generalised drivers (Q2278678)

The paper contributes to the understanding of backward stochastic differential equations (BSDEs) with distributional drivers and their link to rough non-linear partial differential equations (PDEs). More precisely, the authors consider a multi-dimensional Markovian BSDE of the form \[ Y_t = \Phi(W_T) +\int_t^T Z_r b(r, W_r) \mathrm d r + \int_t^T f(r, W_r, Y_r, Z_r) \mathrm d r - \int_t^T Z_r \mathrm d W_r, \] for \(t\in [0,T]\), where the distributional coefficient \(b\in C([0,T]; H^{-\beta}_{q})\) for some \(\beta\in (0, 1/2)\), \(f\) is Lipschitz continuous and \(W\) is a Brownian motion. Here \(H^{-\beta}_{q}\) denotes a fractional Sobolev space. The authors give an intrinsic meaning to the distributional term \(Z_r b(r,W_r)\) in order to define and solve the above BSDE. The main result states that, under suitable conditions, \(Y_t= u(t, W_t)\) is a (generalized) solution to above BSDE, where \(u\) is the unique mild solution to the PDE \[ \partial_t u(t) +\frac12 \Delta u(t) =- \nabla u^*(t) \, b(t)- f(t,\cdot,u(t), \nabla u(t)),\quad u(T) = \Phi, \] where the terms \(\Delta u \) and \(\nabla u^*\, b\) are defined componentwise. As a consequence, a Feynman-Kac (implicit) representation for the solution \(u\) to the above PDE is provided.