Understanding the Difficulty of Solving Cauchy Problems with PINNs

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Wang, Tao, Zhao, Bo, Gao, Sicun, Yu, Rose
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866909225749315584
author Wang, Tao
Zhao, Bo
Gao, Sicun
Yu, Rose
author_facet Wang, Tao
Zhao, Bo
Gao, Sicun
Yu, Rose
contents Physics-Informed Neural Networks (PINNs) have gained popularity in scientific computing in recent years. However, they often fail to achieve the same level of accuracy as classical methods in solving differential equations. In this paper, we identify two sources of this issue in the case of Cauchy problems: the use of $L^2$ residuals as objective functions and the approximation gap of neural networks. We show that minimizing the sum of $L^2$ residual and initial condition error is not sufficient to guarantee the true solution, as this loss function does not capture the underlying dynamics. Additionally, neural networks are not capable of capturing singularities in the solutions due to the non-compactness of their image sets. This, in turn, influences the existence of global minima and the regularity of the network. We demonstrate that when the global minimum does not exist, machine precision becomes the predominant source of achievable error in practice. We also present numerical experiments in support of our theoretical claims.
format Preprint
id arxiv_https___arxiv_org_abs_2405_02561
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Understanding the Difficulty of Solving Cauchy Problems with PINNs
Wang, Tao
Zhao, Bo
Gao, Sicun
Yu, Rose
Machine Learning
Numerical Analysis
Physics-Informed Neural Networks (PINNs) have gained popularity in scientific computing in recent years. However, they often fail to achieve the same level of accuracy as classical methods in solving differential equations. In this paper, we identify two sources of this issue in the case of Cauchy problems: the use of $L^2$ residuals as objective functions and the approximation gap of neural networks. We show that minimizing the sum of $L^2$ residual and initial condition error is not sufficient to guarantee the true solution, as this loss function does not capture the underlying dynamics. Additionally, neural networks are not capable of capturing singularities in the solutions due to the non-compactness of their image sets. This, in turn, influences the existence of global minima and the regularity of the network. We demonstrate that when the global minimum does not exist, machine precision becomes the predominant source of achievable error in practice. We also present numerical experiments in support of our theoretical claims.
title Understanding the Difficulty of Solving Cauchy Problems with PINNs
topic Machine Learning
Numerical Analysis
url https://arxiv.org/abs/2405.02561