BSDE on an infinite horizon and elliptic PDEs in infinite dimension (Q2474201)

The present paper studies the existence and the uniqueness of the mild solution of the partial differential equation (PDE) \({\mathcal L}u(x)+\psi(x, u(x),\nabla u(x)G(x))=0,\) \(x\in H\), where \(H\) is a Hilbert space, and the second order differential operator \({\mathcal L}\) is of the form \({\mathcal L}=\frac12\)tr\((GG^*(x)D^2)+\langle Ax,\nabla\rangle + \langle F(x),\nabla\rangle\), \(A\) is the generator of a \(C_0\)-semigroup of bounded linear operators in \(H\), \(F\) and \(G\) are functions with an appropriate Lipschitz property, and \(\lambda>0\). In difference to earlier works which investigated the above PDE for constant \(G\) with the help of a fixed point argument, \textit{M. Fuhrman} and \textit{G. Tessitore} [Ann. Probab. 32, No. 1B, 607--660 (2004; Zbl 1046.60061)] followed the idea, which is widely spread in the finite-dimensional case, to interpret the PDE stochastically with the help of a Markovian forward-backward stochastic differential equation (SDE) with infinite time horizon, in which the usual final condition of the backward equation is replaced by an appropriate growth condition. This allowed them to obtain the existence and the uniqueness for \(\lambda>0\) great enough. In the present paper the authors show that, under suitable assumptions, one has the existence and uniqueness for all \(\lambda>0\), even if one still allows \(G\) to depend on \(x\) or to be degenerate. For this the authors use recent results by \textit{P. Briand} and \textit{Y. Hu} [J. Funct. Anal. 155, No. 2, 455--494 (1998; Zbl 0912.60081)] and \textit{M. Royer} [Stochastics Stochastics Rep. 76, No. 4, 281--307 (2004; Zbl 1055.60062)] on the existence and the uniqueness of bounded solutions of backward SDEs with infinite or random time horizon. The main technical point consists in proving the differentiability of such bounded solutions with respect to the initial datum \(x\) of the forward SDE. This is done by the authors by either supposing that \(G\) is non-degenerate or by assuming that \(G\) is constant and \(A+\nabla F\) is dissipative. In the second part of the paper the authors apply their results to the study of a stochastic control problem.