Backward stochastic differential equations with random stopping time and singular final condition (Q2370097)

Let \(W\) denote a \(d\)-dimensional Brownian motion and \(q>0\). The author of the present paper studies the one-dimensional backward stochastic differential equation (BSDE) \[ dY_t=Y_t| Y_t|^q dt+Z_tdW_t,\quad t\in[0,T[, \] related to an \({\mathcal F}^W_{\tau}\)-measurable random variable \(\xi\) with values in \(\overline{\mathbb R}_+=\mathbb R_+\cup\{+\infty\}\) such that \(P\{\xi=+\infty\}>0\). Given an open bounded domain \(D\subset \mathbb R^d\) with \(C^2\)-boundary and a diffusion process \(X\) with strongly non degenerate diffusion coefficient, governed by the Brownian motion \(W\), \(\tau\) denotes the moment when the process \(X\) leaves \(D\) for the first time. The author shows that the above BSDE with \(\xi=g(X_\tau)\) gives a stochastic interpretation of the PDE \(-Lu+u| u|^q=0\) in \(D\) (\(L\) denotes the infinitesimal generator of the diffusion process \(X\)) endowed with the Dirichlet condition \(u=g\) on \(\partial D\). In general, a solution of this PDE has a blow-up set \(\Gamma\subset\partial D\). For \(q=2\) and \(D\) the unit ball in \(\mathbb R^2\), Le Gall, 1993, described all solutions of the PDE by a purely stochastic method and associated them with couples \((\Gamma,\nu)\), where \(\nu\) is a Radon measure on \(\partial D-\Gamma\). Contrarily to Le Gall, the author of the present paper gives a stochastic interpretation in terms of BSDEs. His approach extends his former results [Stochastic Processes Appl. 116, No. 12, 2014--2056 (2006; Zbl 1116.60026)] for parabolic PDEs interpreted stochastically by a BSDE of the above type but with \(\tau\) being a deterministic time horizon. Apart from the objective of the stochastic interpretation of elliptic PDEs with blow-up effect the results are also extended to more general BSDEs.