{"id":"arxiv/2604.15617","url":"https://arxiv.org/abs/2604.15617","tags":["physics.comp-ph","cs.LG","math.NA"],"score":0,"title":"A Structure-Preserving Graph Neural Solver for Parametric Hyperbolic Conservation Laws","domain":"physics.comp-ph","source":"arxiv","authors":["Jiamin Jiang","Shanglin Lv","Jingrun Chen"],"reviews":0,"subject":"physics.comp-ph","summary":"Hyperbolic conservation laws govern a wide range of transport-driven dynamics featuring shocks, contact discontinuities, and complex wave interactions, posing distinct challenges for deep-learning-based surrogate modeling. While classical numerical methods provide robust and physically admissible solutions, their computational cost restricts applicability in many-query tasks such as parametric studies and design optimization. Conversely, existing neural surrogates offer rapid inference but often fail to respect intrinsic PDE structures, leading to non-physical artifacts, rollout instability,…","category":"research","comments":0,"disputed":false,"division":"physics","mentions":0,"discussed":0,"published":"2026-04-17","reactions":{}}