Learning Robust Marking Policies for Adaptive Mesh Refinement

DOI10.1137/22M1510613MaRDI QIDQ6189171zbMATH OpenOpenAlexWikidataFDO

Authors Andrew Gillette, Brendan Keith, Socratis Petrides

Publication date 8 February 2024

Published in SIAM Journal on Scientific Computing (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/2207.06339

zbMATH Keywords

\(hp\)-refinementadaptive mesh refinementfinite element methodmachine learningreinforcement learning

Mathematics Subject Classification ID

Laplace operator, Helmholtz equation (reduced wave equation), Poisson equation (35J05) Finite element, Rayleigh-Ritz and Galerkin methods for boundary value problems involving PDEs (65N30) Mesh generation, refinement, and adaptive methods for boundary value problems involving PDEs (65N50) Computational learning theory (68Q32) Learning and adaptive systems in artificial intelligence (68T05)

Abstract: In this work, we revisit the marking decisions made in the standard adaptive finite element method (AFEM). Experience shows that a na"{i}ve marking policy leads to inefficient use of computational resources for adaptive mesh refinement (AMR). Consequently, using AFEM in practice often involves ad-hoc or time-consuming offline parameter tuning to set appropriate parameters for the marking subroutine. To address these practical concerns, we recast AMR as a Markov decision process in which refinement parameters can be selected on-the-fly at run time, without the need for pre-tuning by expert users. In this new paradigm, the refinement parameters are also chosen adaptively via a marking policy that can be optimized using methods from reinforcement learning. We use the Poisson equation to demonstrate our techniques on $h$- and $hp$-refinement benchmark problems, and our experiments suggest that superior marking policies remain undiscovered for many classical AFEM applications. Furthermore, an unexpected observation from this work is that marking policies trained on one family of PDEs are sometimes robust enough to perform well on problems far outside the training family. For illustration, we show that a simple $hp$-refinement policy trained on 2D domains with only a single re-entrant corner can be deployed on far more complicated 2D domains, and even 3D domains, without significant performance loss. For reproduction and broader adoption, we accompany this work with an open-source implementation of our methods.

Recommendations

• Deep reinforcement learning for adaptive mesh refinement
• Recurrent neural networks as optimal mesh refinement strategies
• A remark on the optimality of adaptive finite element methods
• Quasi-optimal convergence of AFEM based on separate marking. I
• Adaptive algorithms. Abstracts from the workshop held September 18--24, 2016

Cites work

• A collection of 2D elliptic problems for testing adaptive grid refinement algorithms
• A comparison of hp-adaptive strategies for elliptic partial differential equations
• A Convergent Adaptive Algorithm for Poisson’s Equation
• A fully automatic \(hp\)-adaptivity
• A note on the design of \(hp\)-adaptive finite element methods for elliptic partial differential equations
• A posteriori error estimation in finite element analysis
• A systematic strategy for simultaneous adaptive hp finite element mesh modification using nonlinear programming
• Aspects of an adaptive \(hp\)-finite element method: Adaptive strategy, conforming approximation and efficient solvers
• Axioms of adaptivity
• Convergence of Adaptive Finite Element Methods
• Efficient Time-Stepping for Numerical Integration Using Reinforcement Learning
• Lower bounds of the discretization error for piecewise polynomials
• Machine learning based refinement strategies for polyhedral grids with applications to virtual element and polyhedral discontinuous Galerkin methods
• MFEM: a modular finite element methods library
• Moving Mesh Generation Using the Parabolic Monge–Ampère Equation
• On an h-type mesh-refinement strategy based on minimization of interpolation errors
• On residual-based a posteriori error estimation in hp-FEM
• Optimality of a standard adaptive finite element method
• Output-based adaptive aerodynamic simulations using convolutional neural networks
• Recurrent neural networks as optimal mesh refinement strategies
• Refinement of polygonal grids using convolutional neural networks with applications to polygonal discontinuous Galerkin and virtual element methods
• Reinforcement learning. An introduction
• Sobolev regularity estimation for hp-adaptive finite element methods
• The finite element methods for elliptic problems.
• The h-p version of the finite element method. I. The basic approximation results
• The superconvergent patch recovery anda posteriori error estimates. Part 1: The recovery technique
• The superconvergent patch recovery anda posteriori error estimates. Part 2: Error estimates and adaptivity
• Thepandh-pVersions of the Finite Element Method, Basic Principles and Properties
• Toward a universal h-p adaptive finite element strategy. I: Constrained approximation and data structure
• Toward a universal h-p adaptive finite element strategy. II: A posteriori error estimation
• Toward a universal h-p adaptive finite element strategy. III: Design of h-p meshes

Cited in

(2)

• Deep reinforcement learning for adaptive mesh refinement
• DynAMO: multi-agent reinforcement learning for dynamic anticipatory mesh optimization with applications to hyperbolic conservation laws

This page was built for publication: Learning Robust Marking Policies for Adaptive Mesh Refinement

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6189171)