Banach neural operator for Navier-Stokes equations

Classical neural networks are known for their ability to approximate mappings between finite-dimensional spaces, but they fall short in capturing complex operator dynamics across infinite-dimensional function spaces. Neural operators, in contrast, have emerged as powerful tools in scientific machine learning for learning such mappings. However, standard neural operators typically lack mechanisms for mixing or attending to input information across space and time. In this work, we introduce the Banach neural operator (BNO) – a novel framework that integrates Koopman operator theory with deep neural networks to predict nonlinear, spatiotemporal dynamics from partial observations. The BNO approximates a nonlinear operator between Banach spaces by combining spectral linearization (via Koopman theory) with deep feature learning (via convolutional neural networks and nonlinear activations). This sequence-to-sequence model captures dominant dynamic modes and allows for mesh-independent prediction. Numerical experiments on the Navier-Stokes equations demonstrate the method’s accuracy and generalization capabilities. In particular, BNO achieves robust zero-shot super-resolution in unsteady flow prediction and consistently outperforms conventional Koopman-based methods and deep learning models.

💡 Research Summary

This paper introduces the Banach Neural Operator (BNO), a novel framework designed to learn nonlinear operators between infinite-dimensional function spaces, with a primary application in solving complex spatiotemporal partial differential equations (PDEs) like the Navier-Stokes equations. The core innovation of BNO lies in its integration of Koopman operator theory with deep convolutional neural networks (CNNs) within a sequence-to-sequence learning architecture.

Traditional neural networks struggle with infinite-dimensional mappings and mesh-dependent solutions. While neural operators like DeepONet and the Fourier Neural Operator (FNO) have made significant strides, they often lack mechanisms for interpretable temporal dynamics or struggle with transient, non-periodic phenomena. BNO addresses these limitations by leveraging the Koopman theory, which posits that nonlinear dynamics can be represented by a linear operator acting on a high-dimensional space of observables. BNO approximates this Koopman operator using a spectral approach inspired by Dynamic Mode Decomposition (DMD).

The BNO architecture operates in three key stages: First, a CNN-based encoder projects the high-dimensional spatiotemporal input data into a lower-dimensional latent space. Second, a learned linear evolution operator (approximating the Koopman operator) propagates the latent state forward in time. Third, a CNN-based decoder maps the evolved latent state back to the high-dimensional physical space, producing the prediction for the next time step. This encoder-evolve-decoder block is iteratively applied in an autoregressive manner, sharing parameters across time steps, enabling long-term forecasting.

The methodology is rigorously tested on the 2D unsteady Navier-Stokes equations, modeling viscous fluid flow past a cylinder. The experiments demonstrate BNO’s superior capabilities in several challenging scenarios: accurate long-horizon prediction, extrapolation to higher Reynolds numbers not seen during training, and most notably, “zero-shot super-resolution.” This means a BNO model trained exclusively on low-resolution spatial grids can accurately predict high-resolution flow fields without any retraining, showcasing its true mesh-independent nature. The results show that BNO consistently outperforms conventional Koopman-based methods (like standard DMD) and other deep learning models in terms of prediction accuracy and generalization. The authors attribute this success to BNO’s ability to distill the dominant, coherent spatiotemporal structures (Koopman modes) of the nonlinear dynamics into a linear, evolvable latent subspace, which generalizes across different spatial discretizations. In conclusion, BNO presents a powerful, interpretable, and generalizable framework for operator learning in scientific machine learning, effectively bridging data-driven methods with physics-based spectral analysis.