A locking-free enriched linearized neural network for linear elasticity

Neural-network Galerkin methods suffer from volumetric locking in nearly incompressible elasticity. We propose the locking-free enriched linearized neural network (LF-ENN), to the best of our knowledge, the first neural-network Galerkin method proved to be locking-free. Following the enriched Galerkin (EG) method of Yi, Lee and Zikatanov (2022), for finite $λ$ it admits a displacement-only formulation whose trial space is a linearized neural network enriched by one dilation field per element of an auxiliary mesh. The parameter $λ$ enters only through a piecewise constant discrete divergence, whose action on the dilation fields is the graph Laplacian of the dual graph of the mesh. Hence any conforming field can be made discretely divergence-free, and the error is bounded independently of $λ\in(0,\infty]$. The network needs no boundary factor: the Dirichlet condition is imposed weakly by Nitsche's method, and a Galerkin least-squares term on the shear part, as in ENN, together with a Hessian penalty on the elements along the traction boundary, replaces the trace inequality that networks lack. Under the stated approximation and quasi-uniformity assumptions, the discrete stability and error-estimate constants are independent of network width, mesh size, and $λ$, for both clamped and mixed boundary conditions. For solutions with reduced regularity, we establish an error estimate of order $h^s$ under the assumptions $u\in H^{1+s}(Ω)^d$ and $p\in H^s(Ω)$, with $s>1/2$. Numerical experiments, including Cook's membrane, demonstrate the effectiveness of the method in the presence of boundary singularities.

Authors

Publication Details

Journal
arXiv (Cornell University)
Published
2026-11-17
DOI
https://doi.org/10.48550/arxiv.2611.22398
Primary Topic
Advanced Numerical Methods in Computational Mathematics
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

A locking-free enriched linearized neural network for linear elasticity

Bonyong Kim, Young Ju Lee
arXiv (Cornell University)
Advanced Numerical Methods in Computational Mathematics
preprint

A locking-free enriched linearized neural network for linear elasticity

Bonyong Kim, Young Ju Lee
preprint en

Abstract

Neural-network Galerkin methods suffer from volumetric locking in nearly incompressible elasticity. We propose the locking-free enriched linearized neural network (LF-ENN), to the best of our knowledge, the first neural-network Galerkin method proved to be locking-free. Following the enriched Galerkin (EG) method of Yi, Lee and Zikatanov (2022), for finite $λ$ it admits a displacement-only formulation whose trial space is a linearized neural network enriched by one dilation field per element of an auxiliary mesh. The parameter $λ$ enters only through a piecewise constant discrete divergence, whose action on the dilation fields is the graph Laplacian of the dual graph of the mesh. Hence any conforming field can be made discretely divergence-free, and the error is bounded independently of $λ\in(0,\infty]$. The network needs no boundary factor: the Dirichlet condition is imposed weakly by Nitsche's method, and a Galerkin least-squares term on the shear part, as in ENN, together with a Hessian penalty on the elements along the traction boundary, replaces the trace inequality that networks lack. Under the stated approximation and quasi-uniformity assumptions, the discrete stability and error-estimate constants are independent of network width, mesh size, and $λ$, for both clamped and mixed boundary conditions. For solutions with reduced regularity, we establish an error estimate of order $h^s$ under the assumptions $u\in H^{1+s}(Ω)^d$ and $p\in H^s(Ω)$, with $s>1/2$. Numerical experiments, including Cook's membrane, demonstrate the effectiveness of the method in the presence of boundary singularities.

arXiv (Cornell University)
Advanced Numerical Methods in Computational Mathematics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon