{"id":"arxiv/2607.07506","url":"https://arxiv.org/abs/2607.07506","tags":["math.NA"],"score":0,"title":"Structure-Guided Gauss-Newton Method: Linear Advection-Reaction Equation","domain":"math.NA","source":"arxiv","authors":["Zhiqiang Cai","César Herrera"],"reviews":0,"subject":"math.NA","summary":"The least-squares neural network (LSNN) method introduced in [5] for linear advection-reaction equations is capable of accurately approximating discontinuous solutions without a priori knowledge of the interface location. However, the resulting discretization is a non-convex optimization problem that is computationally intensive and complex. In this paper, we propose a structure-guided Gauss-Newton (SgGN) method that alternates between the linear (output) and the nonlinear (hidden layer) parameters. At each outer iteration, the linear parameters are computed by a linear solver, and the…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-07-08","reactions":{}}