Saved in:
Bibliographic Details
Main Authors: Liu, Chenguang, Papapantoleon, Antonis, Rou, Jasper
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2512.25017
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914350572240896
author Liu, Chenguang
Papapantoleon, Antonis
Rou, Jasper
author_facet Liu, Chenguang
Papapantoleon, Antonis
Rou, Jasper
contents The aim of this article is to provide a firm mathematical foundation for the application of deep gradient flow methods (DGFMs) for the solution of (high-dimensional) partial differential equations (PDEs). We decompose the generalization error of DGFMs into an approximation and a training error. We first show that the solution of PDEs that satisfy reasonable and verifiable assumptions can be approximated by neural networks, thus the approximation error tends to zero as the number of neurons tends to infinity. Then, we derive the gradient flow that the training process follows in the ``wide network limit'' and analyze the limit of this flow as the training time tends to infinity. These results combined show that the generalization error of DGFMs tends to zero as the number of neurons and the training time tend to infinity.
format Preprint
id arxiv_https___arxiv_org_abs_2512_25017
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Convergence of the generalization error for deep gradient flow methods for PDEs
Liu, Chenguang
Papapantoleon, Antonis
Rou, Jasper
Numerical Analysis
Machine Learning
Computational Finance
68T07, 65M12
The aim of this article is to provide a firm mathematical foundation for the application of deep gradient flow methods (DGFMs) for the solution of (high-dimensional) partial differential equations (PDEs). We decompose the generalization error of DGFMs into an approximation and a training error. We first show that the solution of PDEs that satisfy reasonable and verifiable assumptions can be approximated by neural networks, thus the approximation error tends to zero as the number of neurons tends to infinity. Then, we derive the gradient flow that the training process follows in the ``wide network limit'' and analyze the limit of this flow as the training time tends to infinity. These results combined show that the generalization error of DGFMs tends to zero as the number of neurons and the training time tend to infinity.
title Convergence of the generalization error for deep gradient flow methods for PDEs
topic Numerical Analysis
Machine Learning
Computational Finance
68T07, 65M12
url https://arxiv.org/abs/2512.25017