UniFluids: Unified Neural Operator Learning with Conditional Flow-matching

TL;DR

UniFluids introduces a unified neural operator learning framework using conditional flow-matching and diffusion Transformers, enabling scalable, accurate PDE solutions across diverse dimensions and variables with improved efficiency.

Contribution

It is the first to apply flow-matching for parallel sequence generation in unified PDE operator learning, integrating heterogeneous datasets with a novel 4D spatiotemporal representation.

Findings

Achieves high prediction accuracy across 1D, 2D, and 3D PDE datasets.

Demonstrates strong scalability and cross-scenario generalization.

Employs x-prediction to significantly enhance prediction accuracy.

Abstract

Partial differential equation (PDE) simulation holds extensive significance in scientific research. Currently, the integration of deep neural networks to learn solution operators of PDEs has introduced great potential. In this paper, we present UniFluids, a conditional flow-matching framework that harnesses the scalability of diffusion Transformer to unify learning of solution operators across diverse PDEs with varying dimensionality and physical variables. Unlike the autoregressive PDE foundation models, UniFluids adopts flow-matching to achieve parallel sequence generation, making it the first such approach for unified operator learning. Specifically, the introduction of a unified four-dimensional spatiotemporal representation for the heterogeneous PDE datasets enables joint training and conditional encoding. Furthermore, we find the effective dimension of the PDE dataset is much lower than its patch dimension. We thus employ $x$-prediction in the flow-matching operator learning, which is verified to significantly improve prediction accuracy. We conduct a large-scale evaluation of UniFluids on several PDE datasets covering spatial dimensions 1D, 2D and 3D. Experimental results show that UniFluids achieves strong prediction accuracy and demonstrates good scalability and cross-scenario generalization capability. The code will be released later.