Design patterns: elements of reusable object-oriented software
Design patterns: elements of reusable object-oriented software
Adaptive Finite Element Methods for Optimal Control of Partial Differential Equations: Basic Concept
SIAM Journal on Control and Optimization
SIAM Journal on Scientific Computing
SC '05 Proceedings of the 2005 ACM/IEEE conference on Supercomputing
Function, Gradient, and Hessian Recovery Using Quadratic Edge-Bump Functions
SIAM Journal on Numerical Analysis
A Framework for the Adaptive Finite Element Solution of Large-Scale Inverse Problems
SIAM Journal on Scientific Computing
Hi-index | 0.00 |
Optimization problems constrained by complex dynamics can lead to computationally challenging problems especially when high accuracy and efficiency are required. We present an approach to adaptively control numerical errors in optimization problems approximated using the finite element method. The discrete adjoint equation serves as a key tool to efficiently compute both parameter sensitivities and goal-oriented error estimates at the same discretized levels. By using a recovery method for the error estimators, we avoid expensive higher order adjoint calculations. We nest the adaptivity of the mesh within the optimization algorithm, which is responsible for converging both the state and optimization algorithms and thereby allowing the reuse of state, parameters, and reduced Hessian in subsequent optimization iterations. Our approach is demonstrated on a parameter estimation problem for contamination transport in a contact tank reactor. Significant efficiency and accuracy improvements are realized in comparison to uniform grid refinement strategies and black-box optimization methods. A flexible and maintainable software interface was developed to provide access between the underlying linear algebra of a production simulator and advanced numerical algorithms such as optimization and error estimation.