Preconditioners Based on "Parareal" Time-Domain Decomposition for Time-Dependent PDE-Constrained Optimization [chapter]

Stefan Ulbrich
2015 Contributions in Mathematical and Computational Sciences  
We consider optimization problems governed by time-dependent parabolic PDEs and discuss the construction of parallel preconditioners based on the parareal method for the solution of quadratic subproblems which arise within SQP methods. In the case without control constraints, the optimality system of the subproblem is directly reduced to a symmetric PDE system, for which we propose a preconditioner that decouples into a forward and backward PDE solve. In the case of control constraints we apply
more » ... a semismooth Newton method and apply the preconditioner to the semismooth Newton system. We prove bounds on the condition number of the preconditioned system which shows no or only a weak dependence on the size of regularization parameters for the control. We propose to use the parareal time domain decomposition method for the forward and backward PDE solves within the PDE preconditioner to construct an efficient parallel preconditioner. Numerical results show the efficiency of the approach.
doi:10.1007/978-3-319-23321-5_8 fatcat:ijyrbwtwwjebzhhe3rirtwvm6i