Search papers, labs, and topics across Lattice.
This paper addresses the challenges of solving elliptic PDEs with rough and high-contrast inputs, which are common in multiscale problems across various scientific fields. By introducing LOD-MSNO, a hybrid approach that integrates the Localized Orthogonal Decomposition (LOD) method with neural operator learning, the authors effectively enhance the accuracy of neural operators while maintaining computational efficiency. The results show that LOD-MSNO significantly outperforms existing neural operator baselines in accuracy for complex multiscale scenarios, while also providing theoretical error estimates for its framework.
LOD-MSNO achieves superior accuracy in multiscale problems by combining traditional numerical methods with cutting-edge neural operator techniques, challenging the limits of current approaches.
Multiscale problems are notoriously difficult to tackle using traditional numerical methods, as accurately resolving fine-scale features often requires prohibitively fine discretizations. This challenge is particularly pronounced in applications such as materials science, fluid dynamics, climate systems, chemical processes, and complex networks. Recent neural operator models provide a promising data-driven alternative, but frequently struggle to achieve sufficient accuracy in the presence of strongly heterogeneous or oscillatory coefficients. In this work, we focus on the solution of elliptic PDEs with rough and high-contrast inputs. The Localized Orthogonal Decomposition (LOD) method is a well-established numerical approach for such problems, but it comes, however, at a substantial computational cost. We investigate the performance of popular neural operator architectures on these challenging multiscale problems and identify key limitations in their ability to resolve fine-scale structure. To overcome these challenges, we introduce LOD-MSNO (LOD-Multiscale Neural Operator), a hybrid approach that leverages the LOD method as a strong multiscale prior by building on its representation of the solution as a linear combination of problem-adapted basis functions, while addressing its main computational bottlenecks through data-driven operator learning. We further provide theoretical error estimates for the proposed coefficient-learning framework. Lastly, we demonstrate the potential of our proposed method to outperform current neural operator baselines in terms of accuracy for challenging multiscale inputs, while mainly retaining the computational efficiency of neural operator models.
