COMPOL: A Unified Neural Operator Framework for Scalable Multi-Physics Simulations

ICLR 2026 Conference Submission5942 Authors

Keywords: operator learning, physical simulations, coupled and multi-physics simulations

TL;DR: A versatile multi-physics operator learning framwork

Abstract: Multi-physics simulations play an essential role in accurately modeling complex interactions across diverse scientific and engineering domains. Although neural operators, especially the Fourier Neural Operator (FNO), have significantly improved computational efficiency, they often fail to effectively capture intricate correlations inherent in coupled physical processes. To address this limitation, we introduce COMPOL, a novel coupled multi-physics operator learning framework. COMPOL extends conventional operator architectures by incorporating sophisticated recurrent and attention-based aggregation mechanisms, effectively modeling interdependencies among interacting physical processes within latent feature spaces. Our approach is architecture-agnostic and seamlessly integrates into various neural operator frameworks that involve latent space transformations. Extensive experiments on diverse benchmarks—including biological reaction-diffusion systems, pattern-forming chemical reactions, multiphase geological flows, and thermo-hydro-mechanical processes — demonstrate that COMPOL consistently achieves superior predictive accuracy compared to state-of-the-art methods.

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Submission Number: 5942