Convergence analysis of multigrid methods with collective point smoothers for optimal control problems (Q434223): Difference between revisions

The analysis presented in this paper is discussed for the following elliptic distributed control model problem of tracking type: Minimize the cost functional \(J\) given by \[ J(y,u)=(1/2)\|y-y_d\|^2_{L_2(\Omega)} + (\alpha/2)\|u\|^2_{L_2(\Omega)}, \] subject to the elliptic boundary value problem \[ -\Delta y + y = u \text{ in } \Omega \text{ and } (\partial y / \partial n)=0 \text{ on }\partial \Omega, \] where \(y \in H^1(\Omega)\) is the state variable and \(u \in L^2(\Omega)\) is the control variable. The function \(y_d \in L^2(\Omega)\) is given and \(\alpha >0\) is some regularization or cost parameter. Here \(\Omega\) is a bounded domain in \(\mathbb{R}^d\) for \(d \in \{1,2,3\}\) with Lipschitz boundary \(\partial \Omega\). The authors construct and analyze numerical methods that produce an approximate solution to the optimization problem, where the computational time can be bounded by the number of unknowns times a constant which is independent of the parameter \(\alpha\), in particular for small values of \(\alpha\). The solution of the optimization problem is characterized by the Karush-Kuhn-Tucker (KKT) system. The authors derive the KKT system for the model problem. The discretization is done by standard techniques. For the model problem the authors use a family of meshes which is obtained based on some coarsed triangular mesh (grid level \(K=0\)) and uniform refinement. The space of discretized functions \(X_K = Y_K \times P_K\) is constructed by the Courant element: \(Y_K = P_K\) is the set of continuous and piecewise linear functions. A convergence proof for multigrid methods with special collective point smoothers for general grids, based on the classical splitting of the analysis into smoothing and approximation property, is presented. Finally, numerical results which confirm the theoretical results and illustrate the efficiency of the method (even in cases which are not governed by the convergence theory) are proposed.