Neural Network Localized Orthogonal Decomposition for Numerical Homogenization of Diffusion Operators with Random Coefficients

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Kröpfl, Fabian, Peterseim, Daniel, Ullmann, Elisabeth
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866915496971993088
author Kröpfl, Fabian
Peterseim, Daniel
Ullmann, Elisabeth
author_facet Kröpfl, Fabian
Peterseim, Daniel
Ullmann, Elisabeth
contents This paper presents a neural network--enhanced surrogate modeling approach for diffusion problems with spatially varying random field coefficients. The method builds on numerical homogenization, which compresses fine-scale coefficients into coarse-scale surrogates without requiring periodicity. To overcome computational bottlenecks, we train a neural network to map fine-scale coefficient samples to effective coarse-scale information, enabling the construction of accurate surrogates at the target resolution. This framework allows for the fast and efficient compression of new coefficient realizations, thereby ensuring reliable coarse models and supporting scalable computations for large ensembles of random coefficients. We demonstrate the efficacy of our approach through systematic numerical experiments for two classes of coefficients, emphasizing the influence of coefficient contrast: (i) lognormal diffusion coefficients, a standard model for uncertain subsurface structures in geophysics, and (ii) hierarchical Gaussian random fields with random correlation lengths.
format Preprint
id arxiv_https___arxiv_org_abs_2509_12896
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Neural Network Localized Orthogonal Decomposition for Numerical Homogenization of Diffusion Operators with Random Coefficients
Kröpfl, Fabian
Peterseim, Daniel
Ullmann, Elisabeth
Numerical Analysis
65N25, 60G60, 68T07
This paper presents a neural network--enhanced surrogate modeling approach for diffusion problems with spatially varying random field coefficients. The method builds on numerical homogenization, which compresses fine-scale coefficients into coarse-scale surrogates without requiring periodicity. To overcome computational bottlenecks, we train a neural network to map fine-scale coefficient samples to effective coarse-scale information, enabling the construction of accurate surrogates at the target resolution. This framework allows for the fast and efficient compression of new coefficient realizations, thereby ensuring reliable coarse models and supporting scalable computations for large ensembles of random coefficients. We demonstrate the efficacy of our approach through systematic numerical experiments for two classes of coefficients, emphasizing the influence of coefficient contrast: (i) lognormal diffusion coefficients, a standard model for uncertain subsurface structures in geophysics, and (ii) hierarchical Gaussian random fields with random correlation lengths.
title Neural Network Localized Orthogonal Decomposition for Numerical Homogenization of Diffusion Operators with Random Coefficients
topic Numerical Analysis
65N25, 60G60, 68T07
url https://arxiv.org/abs/2509.12896