DSO: Dual-Scale Neural Operators for Stable Long-term Fluid Dynamics Forecasting

2026-03-25

AI Summary

Existing neural operators often suffer from instability and degraded accuracy in long-term fluid dynamics forecasting due to blurred local details and drift in global trends. This work proposes the Dual-Scale Operator (DSO), which explicitly decouples and independently models the evolution of fine-grained local structures and large-scale motion tendencies in fluid systems for the first time. DSO employs depthwise separable convolutions to capture high-resolution local features while leveraging an MLP-Mixer to aggregate long-range global information, forming a dual-path collaborative prediction architecture. Evaluated on standard turbulence benchmarks, DSO substantially enhances long-term prediction stability and achieves state-of-the-art accuracy, reducing prediction error by over 88% compared to current methods.

Abstract

Long-term fluid dynamics forecasting is a critically important problem in science and engineering. While neural operators have emerged as a promising paradigm for modeling systems governed by partial differential equations (PDEs), they often struggle with long-term stability and precision. We identify two fundamental failure modes in existing architectures: (1) local detail blurring, where fine-scale structures such as vortex cores and sharp gradients are progressively smoothed, and (2) global trend deviation, where the overall motion trajectory drifts from the ground truth during extended rollouts. We argue that these failures arise because existing neural operators treat local and global information processing uniformly, despite their inherently different evolution characteristics in physical systems. To bridge this gap, we propose the Dual-Scale Neural Operator (DSO), which explicitly decouples information processing into two complementary modules: depthwise separable convolutions for fine-grained local feature extraction and an MLP-Mixer for long-range global aggregation. Through numerical experiments on vortex dynamics, we demonstrate that nearby perturbations primarily affect local vortex structure while distant perturbations influence global motion trends, providing empirical validation for our design choice. Extensive experiments on turbulent flow benchmarks show that DSO achieves state-of-the-art accuracy while maintaining robust long-term stability, reducing prediction error by over 88% compared to existing neural operators.

Problem

• long-term fluid dynamics forecasting
• neural operators
• local detail blurring
• global trend deviation
• stability

Innovation

• Dual-Scale Neural Operator
• long-term stability
• fluid dynamics forecasting
• local-global decoupling
• neural PDE solvers