On discrete regularization of nonlinear equations and problems of optimal control over singular systems (Q1390394): Difference between revisions

We consider the following problem of optimal control \[ \min \bigl\{J(u): u\in U\}, \quad J(u)= ||| u||| + \|z(u)-x_0\|, \tag{1} \] in which the state of the system \(z=z(u)\in X\) is related to the control \(u\in U\subset W\) via the operator equation \[ A(u,z) =y. \tag{2} \] Here \(X\) and \(W\) are real Hilbert spaces with the norms \(\|\cdot \|\) and \(||| \cdot |||\); \(A:U \times X\to X^*\) is a nonlinear operator, \(U\) is a convex bounded set in \(W\), the elements \(x_0\in X\) and \(y\in X^*\) are fixed. We investigate a method for searching the solution of problem (1)--(2), which is a modification of the method of discrete operator regularization for problems of the form (1)--(2) with coercive operator \(A(u, \cdot)\). The necessity to modify the method is due to possible absence of convergence of the method even in the case where this operator is coercive. We give conditions for the consistency of regularization and discretization parameters with the error, which provide the construction of a sequence of approximations strongly converging to the solution. The difference of these conditions from known requirements, which provide convergence of the methods of discrete regularization of equations and extremal problems, is in removing the a priori assumptions on the rate of convergence of discrete approximations of a solution to this solution. In this situation, the convergence is provided due to making consistent the parameter of discretization with the estimate of the norms of solutions of the corresponding model regularized equations.