Id: arxiv/2611.13231 Url: https://arxiv.org/abs/2611.13231 Tags: math.NA, cs.LG Score: 0 Title: Block-Sparse Gauss-Newton Optimization of Finite Basis Physics-Informed Neural Networks: Sparse Automatic Differentiation Approach Domain: math.NA Source: arxiv Authors: Denis Korolev, Adrian Hill, Jeeeun Lee Reviews: 0 Subject: math.NA Category: research Comments: 0 Disputed: false Division: math Mentions: 0 Discussed: 0 Published: 2026-11-09 Page: https://delphi.science/r/arxiv/2611.13231 In this work, we develop an efficient Gauss-Newton optimization approach for finite basis physics-informed neural networks (FBPINNs). The key ingredient is the overlapping domain decomposition architecture of FBPINNs, which localizes the influence of each subnetwork and induces block sparsity in both the residual Jacobian and the corresponding Gauss-Newton matrix. We characterize this structure and exploit it through sparse automatic differentiation, using graph coloring to partition structurally orthogonal Jacobian columns or rows. This reduces differentiation costs and memory requirements…