Global convergence of the Newton method in the inverse problems of memory reconstruction (Q1375038): Difference between revisions

Two inverse problems of memory reconstruction are considered in the article. Problem 1. Determine functions \(u\: [0,T] \rightarrow X\) and \(h\:[0,T] \rightarrow \mathbb R\) such that \[ u_t -Au = \int \limits_0^t h(t-\tau) Bu(\tau) d \tau + f(t), \quad t \in [0,T], \] \[ u(0) =u_0, \] \[ \Phi[u(t)]=\varphi(t), \quad t \in [0,T]. \] Here \(A\) is the infinitesimal generator of a strongly continuous semigroup \(V(t)\) with \(||V(t)||\leq M e^{t \beta}\); \(A^{-1}\) exists and is bounded; \(u_0 \in X\), \(f\: [0,T] \rightarrow X\), \(\varphi\: [0,T] \rightarrow \mathbb R\), and \(\Phi \in X^*\) are known, \(B\) is a closed linear operator, \(X\) is a Banach space, \(D(A)\) and \(D(B)\) are dense in \(X\). Problem 2. Determine functions \(u\: [0,T] \rightarrow X\) and \(h\:[0,T] \rightarrow \mathbb R\) such that \[ u_{tt} -Au = \int \limits_0^t h(t-\tau) Bu(\tau) d \tau + f(t), \quad t \in [0,T], \] \[ u(0) =u_0, \quad u'(0)=u_1 \] \[ \Phi[u(t)]=\varphi(t), \quad t \in [0,T]. \] Here \(A\) is the generator of a strongly continuous family of cosine-functions \(C(t)\) with \(||C(t)||\leq M e^{t |\beta|}\); \(A^{-1}\) exists and is bounded; \(u_0, u_1 \in X\), \(f\: [0,T] \rightarrow X\), \(\varphi\: [0,T] \rightarrow \mathbb R\), and \(\Phi \in X^*\) are known, \(B\) is a closed linear operator, \(X\) is a Banach space, \(D(A)\) and \(D(B)\) are dense in \(X\). Four theorems are proven which assert that, under fulfilment of \(\Phi[B u_0] \neq 0\) and certain smoothness and agreement conditions on functions entering the formulations of the problems, solutions to Problems 1 and 2 exist, are unique, and determined via the Newton method. The problems are first reduced to systems of Volterra equations of the second order, then the Newton method is applied to the derived systems. Global convergence of the Newton method is proven in the corresponding weight spaces.