Enforcing governing equation constraints in neural PDE solvers via training-free projections

Λεπτομέρειες βιβλιογραφικής εγγραφής
Τίτλος: Enforcing governing equation constraints in neural PDE solvers via training-free projections
Συγγραφείς: Rochman Sharabi, Omer, Louppe, Gilles
Πηγή: Machine Learning and the Physical Sciences Workshop (NeurIPS 2025), San Diego, United States - California [US-CA], 06/12/2025
Έτος έκδοσης: 2025
Θεματικοί όροι: Computer Science - Learning, Engineering, computing & technology, Computer science, Ingénierie, informatique & technologie, Sciences informatiques
Περιγραφή: Neural PDE solvers used for scientific simulation often violate governing equation constraints. While linear constraints can be projected cheaply, many constraints are nonlinear, complicating projection onto the feasible set. Dynamical PDEs are especially difficult because constraints induce long-range dependencies in time. In this work, we evaluate two training-free, post hoc projections of approximate solutions: a nonlinear optimization-based projection, and a local linearization-based projection using Jacobian-vector and vector-Jacobian products. We analyze constraints across representative PDEs and find that both projections substantially reduce violations and improve accuracy over physics-informed baselines.
Τύπος εγγράφου: conference poster not in proceedings
http://purl.org/coar/resource_type/c_18co
conferencePoster
peer reviewed
Γλώσσα: English
Relation: https://arxiv.org/abs/2511.17258
DOI: 10.48550/arXiv.2511.17258
Σύνδεσμος πρόσβασης: https://orbi.uliege.be/handle/2268/340544
Rights: open access
http://purl.org/coar/access_right/c_abf2
info:eu-repo/semantics/openAccess
Αριθμός Καταχώρησης: edsorb.340544
Βάση Δεδομένων: ORBi