trust_region

pymor-demo trust_region OUTPUT-NUMBER GRID-INTERVALS TRAINING-SAMPLES

Error aware trust-region method for PDE-constrained parameter optimization problems.

This demo compares three different approaches for solving PDE-constrained parameter optimization problems with linear and quadratic output functionals. As optimization method we compare:

1. The reference method: Projected BFGS, only based on the FOM.

2. The classical RB method: First standard Greedy for building a ROM, then projected BFGS with ROM.

3. The error aware trust-region method proposed in [KMO+21].

The methods are compared in terms of computational time, iterations, optimization error, and FOM/ROM evaluations.

Parameters:

OUTPUT-NUMBER, --output-number

Selects type of output functional, where 0 stands for linear and 1 for a quadratic output. [Required, Choices: 0, 1]

GRID-INTERVALS, --grid-intervals

Grid interval count. [Required]

TRAINING-SAMPLES, --training-samples

Number of samples used for training the reduced basis. [Required]