Safeguarded Neurally-Accelerated Preconditioning (SNAP) framework for Nonlinear Optimization

DOI10.48550/ARXIV.2610.38790MaRDI QIDQ7409128arXivFDO

Authors Youngkyu Lee, George Em Karniadakis, Alena Kopaničáková

Publication date 28 October 2026

Published in arXiv

Tldr We propose the Safeguarded Neurally-Accelerated Preconditioning (SNAP) framework to accelerate nonlinear optimization methods by leveraging pre-trained neural operators for generating search directions; SNAP ensures global convergence and outperforms baselines in numerical experiments on various parametric optimization problems.

Full work available at URL https://doi.org/10.48550/ARXIV.2610.38790

arXiv classification

math.NAmath.OC

Cites work

• A Deep Conjugate Direction Method for Iteratively Solving Linear Systems
• A domain decomposition online-learning-enhanced nonlinear elimination preconditioner
• A fast neural hybrid Newton solver adapted to implicit methods for nonlinear dynamics
• A hybrid iterative method based on MIONet for PDEs: Theory and numerical examples
• A Learnable Multigrid Framework via Graph Convolutions
• A multigrid approach to discretized optimization problems
• A Neural Multigrid Solver for Helmholtz Equations with High Wavenumber and Heterogeneous Media
• A neural network multigrid solver for the Navier-Stokes equations
• A Neural-Operator Preconditioned Newton Method for Accelerated Nonlinear Solvers
• A Nonoverlapping Domain Decomposition Method for Extreme Learning Machines: Elliptic Problems
• A Recursively Recurrent Neural Network (R2N2) Architecture for Learning Iterative Algorithms
• Accelerating Algebraic Multigrid Methods via Artificial Neural Networks
• Accelerating conjugate gradient solvers for homogenization problems with unitary neural operators
• Accelerating hypersonic reentry simulations using deep learning-based hybridization (with guarantees)
• Accelerating multigrid solver with generative super-resolution
• Accelerating the convergence of Newton's method for nonlinear elliptic PDEs using Fourier neural operators
• Blending Neural Operators and Relaxation Methods in PDE Numerical Solvers
• Composing scalable nonlinear algebraic solvers
• Convergence Framework of Deep Learning-based Hybrid Iterative Methods and the Application to Designing a Fourier Neural Solver for Parametric PDEs
• D3M: A deep domain decomposition method for partial differential equations
• Data-driven multi-grid solver for accelerated pressure projection
• Deep Domain Decomposition Method: Elliptic Problems
• DeepOnet Based Preconditioning Strategies For Solving Parametric Linear Systems of Equations
• Elliptic partial differential equations of second order
• Fast meta-solvers for 3D complex-shape scatterers using neural operators trained on a non-scattering problem
• Fourier Neural Operator for Parametric Partial Differential Equations
• Fourier Neural Solver for large sparse linear algebraic systems
• Iterative Methods for Sparse Linear Systems
• Learning adaptive FETI-DP constraints for irregular domain decompositions
• Learning Interface Conditions in Domain Decomposition Solvers
• Learning Neural PDE Solvers with Convergence Guarantees
• Learning Singularity-Encoded Green's Functions with Application to Iterative Methods
• Learning-based local weighted least squares for algebraic multigrid method
• Linear Convergence of Gradient and Proximal-Gradient Methods Under the Polyak-\L{}ojasiewicz Condition
• Machine learning in adaptive domain decomposition methods -- predicting the geometric location of constraints
• MG-GNN: Multigrid Graph Neural Networks for Learning Multilevel Domain Decomposition Methods
• Model Problems for the Multigrid Optimization of Systems Governed by Differential Equations
• Multigrid-Augmented Deep Learning Preconditioners for the Helmholtz Equation
• Neural operators meet conjugate gradients: The FCG-NO method for efficient PDE solving
• Neural Preconditioning via Krylov Subspace Geometry
• Numerical Optimization
• Numerical solution of mixed-dimensional PDEs using a neural preconditioner
• On high-order multilevel optimization strategies
• On the Globalization of ASPIN Employing Trust-Region Control Strategies -- Convergence Analysis and Numerical Examples
• On the use of hybrid coarse-level models in multilevel minimization methods
• Properties of a class of multilevel optimization algorithms for equality-constrained problems
• Recursive Trust-Region Methods for Multiscale Nonlinear Optimization
• Some problems in the theory of quasilinear equations
• Spectral coarse spaces for the substructured parallel Schwarz method
• Spectrally Safe Neural Operator Warm-Starts for Large-Scale Newton Solvers
• Sur les équations intégrales non linéaires.

This page was built for publication: Safeguarded Neurally-Accelerated Preconditioning (SNAP) framework for Nonlinear Optimization

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7409128)