COMPOL: Neural Operator Framework for Coupled Multi-Physics Simulations

Researchers have introduced COMPOL, a novel coupled multiphysics operator learning framework designed to improve the accuracy of simulations involving multiple interacting physical processes. While existing neural operators like the Fourier Neural Operator (FNO) have enhanced computational efficiency, they often fail to capture the intricate correlations in coupled systems. COMPOL extends conventional architectures by incorporating recurrent and attention-based aggregation mechanisms, enabling effective modeling of interdependencies among physical processes within latent feature spaces. The framework is architecture-agnostic and can integrate into various neural operator frameworks that involve latent space transformations. Extensive experiments on benchmarks, including biological reaction-diffusion systems, demonstrate its effectiveness.