Backward stochastic differential equations with singular terminal condition (Q860713)

Let \(T>0\) be a finite time horizon and \(W\) denote a \(d\)-dimensional Brownian motion. For \(q>0\) the author studies the backward stochastic differential equation (for short: BSDE) \(dY_t=Y_t| Y_t|^q dt+Z_tdW_t,\, t\in[0,T[\), related to an \({\mathcal F}_T\)-measurable random variable \(\xi\) with values in \(\overline{R}=R\cup\{-\infty,+\infty\}\). This type of BSDE is new in the literature, it allows the function \(f(y)=-y| y|^q\) to have a more than quadratic growth and leads to very interesting blow up studies for \(Y\). The author proves for \(\xi\geq 0\) the existence of a minimal solution such that \(\lim_{t\uparrow T}Y_t\geq \xi\) and characterizes the blow up behavior of \(Y\). He gets namely that on \(\{\xi =+\infty\}\) \(Y_t\) behaves like \((T-t)^{-1/q}\), for \(t\uparrow T\). In a Markovian framework, where \(\xi=g(X_T)\) (for \(g:R^d\rightarrow \overline{R}^+\) and \(X\) is a driving diffusion process) sufficient conditions are established under which \(Y_t\rightarrow \xi\) \((t\uparrow T)\), P-a.s. This result is used to give to the minimal viscosity solution of the associated PDE a stochastic interpretation. In Section 5 the results for the BSDE are extended to a terminal condition \(\xi\) without sign restriction.