Refine
Document Type
- ZIB-Report (5)
- Doctoral Thesis (1)
Language
- English (6)
Is part of the Bibliography
- no (6)
Keywords
- optimal control (6) (remove)
Institute
In optimal control problems with nonlinear time-dependent 3D PDEs, full 4D discretizations are usually prohibitive due to the storage requirement. For this reason gradient and quasi-Newton methods working on the reduced functional are often employed. The computation of the reduced gradient requires one solve of the state equation forward in time, and one backward solve of the adjoint equation. The state enters into the adjoint equation, again requiring the storage of a full 4D data set. We propose a lossy compression algorithm using an inexact but cheap predictor for the state data, with additional entropy coding of prediction errors. As the data is used inside a discretized, iterative algorithm, lossy coding maintaining an error bound is sufficient.
This paper presents concepts and implementation of the finite element toolbox Kaskade 7, a flexible C++ code for solving elliptic and parabolic PDE systems. Issues such as problem formulation, assembly and adaptivity are discussed at the example of optimal control problems. Trajectory compression for parabolic optimization problems is considered as a case study.
In high accuracy numerical simulations and optimal control of time-dependent processes, often both many time steps and fine spatial discretizations are needed. Adjoint gradient computation, or post-processing of simulation results, requires the storage of the solution trajectories over the whole time, if necessary together with the adaptively refined spatial grids. In this paper we discuss various techniques to reduce the memory requirements, focusing first on the storage of the solution data, which typically are double precision floating point values. We highlight advantages and disadvantages of the different approaches. Moreover, we present an algorithm for the efficient storage of adaptively refined, hierarchic grids, and the integration with the compressed storage of solution data.
This paper presents efficient computational techniques for solving an optimization problem in cardiac defibrillation governed by the monodomain equations. Time-dependent electrical currents injected at different spatial positions act as the control. Inexact Newton-CG methods are used, with reduced gradient computation by adjoint solves. In order to reduce the computational complexity, adaptive mesh refinement for state and adjoint equations is performed. To reduce the high storage and bandwidth demand imposed by adjoint gradient and Hessian-vector evaluations, a lossy compression technique for storing trajectory data is applied. An adaptive choice of quantization tolerance based on error estimates is developed in order to ensure convergence. The efficiency of the proposed approach is demonstrated on numerical examples.
For the solution of optimal control problems governed by nonlinear parabolic PDEs, methods working on the reduced objective functional are often employed to avoid a full spatio-temporal discretization of the problem. The evaluation of the reduced gradient requires one solve of
the state equation forward in time, and one backward solve of the ad-joint equation. The state enters into the adjoint equation, requiring the storage of a full 4D data set. If Newton-CG methods are used, two additional trajectories have to be stored. To get numerical results which are accurate enough, in many case very fine discretizations in time and space are necessary, which leads to a significant amount of data to be stored and transmitted to mass storage. Lossy compression methods were
developed to overcome the storage problem by reducing the accuracy of the stored trajectories. The inexact data induces errors in the reduced gradient and reduced Hessian. In this paper, we analyze the influence of such a lossy trajectory compression method on Newton-CG methods for optimal control of parabolic PDEs and design an adaptive strategy for choosing appropriate quantization tolerances.
Optimal control problems governed by nonlinear, time-dependent PDEs on three-dimensional spatial domains are an important tool in many fields, ranging from engineering applications to medicine. For the solution of such optimization problems, methods working on the reduced objective functional are often employed to avoid a full spatio-temporal discretization of the problem. The evaluation of the reduced gradient requires one solve of the state equation forward in time, and one backward solve of the adjoint equation. The state enters into the adjoint equation, requiring the storage of a full 4D data set. If Newton-CG methods are used, two additional trajectories have to be stored. To get numerical results that are accurate enough, in many cases very fine discretizations in time and space are necessary, leading to a significant amount of data to be stored and transmitted to mass storage.
This thesis deals with the development and analysis of methods for lossy compression of such finite element solutions.
The algorithms are based on a change of basis to reduce correlations in the data, combined with quantization. This is achieved by transforming the finite element coefficient vector from the nodal to the hierarchical basis, followed by rounding the coefficients to a prescribed precision. Due to the inexact reconstruction, and thus inexact data for the adjoint equation, the error induced in the reduced gradient, and reduced Hessian, has to be controlled, to not impede convergence of the optimization. Accuracy requirements of different optimization methods are analyzed, and computable error estimates for the influence of lossy trajectory storage are derived. These tools are used to adaptively control the accuracy of the compressed data.
The efficiency of the algorithms is demonstrated on several numerical examples, ranging from a simple linear, scalar equation to a semi-linear system of reaction-diffusion equations. In all examples considerable reductions in storage space and bandwidth requirements are achieved, without significantly influencing the convergence behavior of the optimization methods.
Finally, to go beyond pointwise error control, the hierarchical basis transform can be replaced by more sophisticated wavelet transforms. Numerical experiments indicate that choosing suitable norms for error control allows higher compression factors.