Towards Matrix-Free AD-Based Preconditioning of KKT Systems in PDE-Constrained Optimization
نویسندگان
چکیده
The presented approach aims at solving an equality constrained, finite-dimensional optimization problem, where the constraints arise from the discretization of some partial differential equation (PDE) on a given space grid. For this purpose, a stationary point of the Lagrangian is computed using Newton’s method, which requires the repeated solution of KKT systems. The proposed algorithm focuses on two topics: Firstly, Algorithmic Differentiation (AD) will be used to evaluate the necessary computations of gradients, Jacobian-vector products, and Hessian-vector products, so that only the objective f(y, u) and the PDE constraint e(y, u) = 0 have to be specified by the user. Secondly, we solve the KKT system iteratively using the QMR algorithm, with preconditioning provided by a multigrid approach. We wish to explore whether the Jacobian-vector products provided by AD are sufficient to construct suitable multigrid preconditioners. Our approach is then embedded into a globalized optimization routine. Numerical results for optimization problems involving a nonlinear reaction-diffusion model will be given.
منابع مشابه
Parallel Lagrange-Newton-Krylov-Schur Methods for PDE-Constrained Optimization. Part I: The Krylov-Schur Solver
Large scale optimization of systems governed by partial differential equations (PDEs) is a frontier problem in scientific computation. The state-of-the-art for such problems is reduced quasi-Newton sequential quadratic programming (SQP) methods. These methods take full advantage of existing PDE solver technology and parallelize well. However, their algorithmic scalability is questionable; for c...
متن کاملNumerical solution of KKT systems in PDE-constrained optimization problems via the affine scaling trust-region approach
A recently proposed trust-region approach for bound-constrained nonlinear equations is applied to the KKT systems arising from the discretization of a class of PDE-constrained optimization problems. Two different implementations are developed that take into account the large dimension and the special structure of the problems. The linear algebra phase is analyzed considering the possibility of ...
متن کاملPDE-constrained optimization: Preconditioners and diffuse domain methods
This thesis is mainly concerned with the efficient numerical solution of optimization problems subject to linear PDE-constraints, with particular focus on robust preconditioners and diffuse domain methods. Associated with such constrained optimization problems are the famous first-order KarushKuhn-Tucker (KKT) conditions. For certain minimization problems, the functions satisfying the KKT condi...
متن کاملA preconditioning technique for a class of PDE-constrained optimization problems
We investigate the use of a preconditioning technique for solving linear systems of saddle point type arising from the application of an inexact Gauss–Newton scheme to PDE-constrained optimization problems with a hyperbolic constraint. The preconditioner is of block triangular form and involves diagonal perturbations of the (approximate) Hessian to insure nonsingularity and an approximate Schur...
متن کاملInertia-Revealing Preconditioning For Large-Scale Nonconvex Constrained Optimization
Fast nonlinear programming methods following the all-at-once approach usually employ Newton’s method for solving linearized Karush-Kuhn-Tucker (KKT) systems. In nonconvex problems, the Newton direction is only guaranteed to be a descent direction if the Hessian of the Lagrange function is positive definite on the nullspace of the active constraints, otherwise some modifications to Newton’s meth...
متن کامل