On backward stochastic evolution equations in Hilbert spaces and optimal control (Q884510)

Let \((\Omega,\mathfrak F_T,\mathbf P)\) be a probability space together with a normal filtration \(\{\mathfrak F_t, 0\leq t\leq T\}\), \(X\) and \(U\) be two separable Hilbert spaces, \(W\) be a \(\mathcal Q\)-Wiener process on \((\Omega,\mathfrak F_T,\mathbf P)\) with a linear bounded covariance operator such that \(\operatorname{tr}\mathcal Q <\infty\). The authors study the existence and uniqueness of solutions to the following class of backward stochastic evolution equations in a Hilbert space \(X\) \[ \left\{\begin{aligned} dy(t)&= -[Ay(t) + F(t,y(t), z(t))]\,dt - [G(t,y(t)) + z(t)]\,dw(t),\\ y(T) &=\xi, \end{aligned}\right.\tag{1} \] where \(A : D (A)\subset X\to X\) is a linear operator which generates a \(C_0\)-semigroup \(\{S(t),0\leq t\leq T\}\) on \(X\), \(F : [0, T] \times X\times L^0_2 \to X\) and \(G : [0, T]\times X\to L^0_2\) are given measurable mappings, and \(\xi\in L^2(\Omega,\mathfrak F_T,X)\). A new result on the existence and uniqueness of the adapted solution to a backward stochastic evolution equation (1) in Hilbert spaces under a non-Lipschitz condition is established. A stochastic maximum principle for optimal control problems of stochastic systems governed by backward stochastic evolution equations in Hilbert spaces is obtained. The applicability of this result is then illustrated in a discussion of a concrete quadratic problem: \[ J(u)= {\mathbf E}\langle Gy(0), y(0)\rangle + {\mathbf E}\int_0^T\langle\varGamma(t)u(t), u(t)\rangle\,dt\to \text{min} \] subject to \[ \left\{\begin{aligned} dy(t)&= [Ay(t) + Bu(t) + Cz(t)]\,dt + z(t)\,dw(t),\\ y(T)&=\xi, \end{aligned}\right. \] where \(B : U \to X\), \(C : L^0_2\to X\), \(\varGamma : [0, T]\to L(U)\), \(G = G^\ast\), \(\varGamma(t) =\varGamma^\ast(t)\).