On measure solutions of backward stochastic differential equations (Q841478)

The authors consider the backward stochastic differential equation (BSDE) \(dY_t=-f(t,Z_t)dt+Z_tdW_t,\in [0,T],\) \(Y_T=\xi,\) where \(T>0\) is a finite time horizon and \(W=(W_t)_{t\in[0,T]}\) a Brownian motion. The nonlinear drift coefficient \(f\) of the BSDE is allowed to have quadratic growth in \(z\). In contrast with the classical solution \((Y,Z)\) of the BSDE, the authors introduce the notion of measure solution \((Y,Z,Q)\), where \(Q\) is a probability measure on the underlying measurable space and \(Y,Z\) are stochastic processes adapted to the process \(W\) s.t. i) \(Q\{\int_0^TZ_t^2<+\infty\}=1,\) ii) \(\widetilde{W}=W-\int_0^.f(s,Z_s)/Z_s\, ds\) is a \(Q\)-Brownian motion, iii) \(\xi\) is \(Q\)-integrable and iv) \(Y_t=E_Q[\xi|{\mathcal F}_t^W],\, t\in[0,T].\) They begin by a discussion of the case of a bounded terminal condition \(\xi\). In this case they show that the measure solutions \((Y,Z,Q)\) which probability measure \(Q\) is equivalent to the underlying one, is just a reinterpretation of the classical solution \((Y,Z)\). In the case where the assumption of boundedness on \(\xi\) is replaced by that of integrable exponential moments, a series of examples is discussed by the authors. These examples show that in the case of non-uniqueness, classical solutions that fail to be measure solutions can coexist with different measure solutions. In the last section the authors consider BSDEs whose coefficient is Lipschitz continuous with time dependent constant. Without using knowledge about strong solutions in their algorithm they prove the existence of a measure solution. This is done by iterating the successive application of martingale representation property and Girsanov transformation w.r.t. the drift gotten from the previous step. For the such obtained sequence of probability measures tightness in the week topology is shown.