FluxNet: Learning Capacity-Constrained Local Transport Operators for Conservative and Bounded PDE Surrogates
arXiv:2602.01941v2 Announce Type: replace-cross Abstract: Autoregressive learning of time-stepping operators provides an effective approach to data-driven partial differential equation (PDE) simulation, yet for conservation laws, they face a fundamental challenge: learned updates may violate global conservation over long rollouts. For the important subclass of mass-conservation-type equations, the problem is compounded by inherent physical bounds (e.g., nonnegativity or concentrations in [0,1]) whose violation further destabilizes predictions. We introduce FluxNet, which learns cumulative tran
The proliferation of data-driven PDE simulations highlights the urgent need for robust methods to ensure physical consistency, a gap FluxNet aims to address via advancements in autoregressive learning.
This development improves the reliability and stability of AI models used for simulating complex physical systems, which is critical for scientific discovery and engineering applications across various fields.
AI models for PDE simulation can now more effectively maintain fundamental physical laws like conservation of mass and boundedness, leading to more accurate and trustworthy long-term predictions.
- · Scientific computing researchers
- · Material science engineers
- · Climate modeling institutions
- · AI/ML developers
- · Traditional numerical simulation methods (long term)
- · Less robust data-driven PDE surrogates
More accurate and stable AI-driven simulations accelerate research and development in fields relying on PDEs such as fluid dynamics, materials science, and climate modeling.
Reduced computational cost and improved reliability of simulations could lead to faster innovation cycles and design optimization in various industries.
Enhanced predictive capabilities from these models might enable the discovery of new materials, more precise climate interventions, or advanced engineering solutions not feasible with current methods.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG