aiqbench analysis
This research introduces NeuraLSP, a neural multigrid preconditioner that replaces graph-based aggregation with low-rank spectral representations to solve large-scale sparse linear systems from PDEs. By employing a novel subspace loss function, the method effectively suppresses rank inflation and achieves up to 53% faster convergence compared to current state-of-the-art approaches across various physics problems. The work advances scientific machine learning by providing theoretical guarantees for robustness in high-performance computing workflows.
NeuraLSP: A Neural Spectral Preconditioner for Accelerating PDE Solvers
arXiv:2601.20174v3 Announce Type: replace-cross Abstract: Solving large-scale sparse linear systems originating from partial differential equations (PDEs) is a fundamental topic in high-performance scientific computing, where preconditioners are crucial. Multigrid methods are among the most effective preconditioners, yet their performance is dictated by the accurate construction of grid transfer operators. Current neural multigrid methods learn such operators with graph neural networks (GNNs), typically by extracting connectivity from discretized system matrices. While effective, these graph-based constructions suffer from rank inflation, resulting in unnecessarily large coarse spaces and slower convergence. To ameliorate, this paper advocates NeuraLSP, a new neural multigrid preconditioner that replaces graph aggregation with a fixed low-rank spectral representation derived from the left singular subspace of near-nullspace components. At the network design level, NeuraLSP is trained with a novel subspace loss function, which preserves the error modes most relevant to multigrid convergence while suppressing rank inflation. This paper's grand innovation hinges upon both theoretical guarantees and empirical robustness to rank inflation, affording up to a 53% speedup over SOTA neural preconditioners across a variety of PDE families.
Abstract:Solving large-scale sparse linear systems originating from partial differential equations (PDEs) is a fundamental topic in high-performance scientific computing, where preconditioners are crucial. Multigrid methods are among the most effective preconditioners, yet their performance is dictated by the accurate construction of grid transfer operators. Current neural multigrid methods learn such operators with graph neural networks (GNNs), typically by extracting connectivity from discretized system matrices. While effective, these graph-based constructions suffer from rank inflation, resulting in unnecessarily large coarse spaces and slower convergence. To ameliorate, this paper advocates NeuraLSP, a new neural multigrid preconditioner that replaces graph aggregation with a fixed low-rank spectral representation derived from the left singular subspace of near-nullspace components. At the network design level, NeuraLSP is trained with a novel subspace loss function, which preserves the error modes most relevant to multigrid convergence while suppressing rank inflation. This paper's grand innovation hinges upon both theoretical guarantees and empirical robustness to rank inflation, affording up to a 53% speedup over SOTA neural preconditioners across a variety of PDE families.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2601.20174 [cs.LG]
(or arXiv:2601.20174v3 [cs.LG] for this version)
Full text under the publisher's open licence · 185 words
Related coverage
BandPC: Learning Residual-Band Preconditioner Combinations for Flexible Conjugate Gradient Solvers
arXiv · Oct 14, 2026
Are Deep Learning Based Hybrid PDE Solvers Reliable? Why Training Paradigms and Update Strategies Matter
arXiv · Jun 3, 2026
One-shot acceleration of transient PDE solvers via online-learned preconditioners
arXiv · Jul 21, 2026
DFSC: Error-Controlled Differentiable Mittag-Leffler Propagation for Fractional Scientific Machine Learning
arXiv · Aug 3, 2026