Id: arxiv/2601.23185 Url: https://arxiv.org/abs/2601.23185 Tags: math.NA Score: 0 Title: Preconditioning and Numerical Stability in Neural Network Training for Parametric PDEs Domain: math.NA Source: arxiv Authors: Markus Bachmayr, Wolfgang Dahmen, Chenguang Duan, Mathias Oster Reviews: 0 Subject: math.NA Category: research Comments: 0 Disputed: false Division: math Mentions: 0 Discussed: 0 Published: 2026-01-30 Page: https://delphi.science/r/arxiv/2601.23185 In the context of training neural network-based approximations of solutions of parameter-dependent PDEs, we investigate the effect of preconditioning via well-conditioned frame representations of operators and demonstrate a significant improvement on the performance of standard training methods. We also observe that standard representations of preconditioned matrices are insufficient for obtaining numerical stability and propose a generally applicable form of stable representations that enables computations with single- and half-precision floating point numbers without loss of precision.