A Hard-Constrained PINN Framework for Variable-Coefficient Identification in the Two-Dimensional Sine-Gordon Equation

This paper proposes a hard-constrained dual-branch physics-informed neural network (PINN) framework for identifying unknown variable coefficients in the two-dimensional sine-Gordon equation. Three prescribed classes of coefficients are considered, including time-dependent, space-dependent, and spatiotemporally varying coefficients. The framework uses a state network and a coefficient network to reconstruct the state field and the unknown coefficient simultaneously. A hard-constrained trial solution is constructed to satisfy the initial displacement, initial velocity, and Dirichlet boundary conditions analytically. Residual-based adaptive refinement and coefficient smoothness regularization are further incorporated to improve the empirical inversion behavior in the tested settings. Numerical experiments are conducted on a unit-square domain with known source terms and prescribed initial-boundary data. The results show that the time-dependent and space-dependent coefficients can be recovered with relative coefficient errors mainly on the order of 10−2, whereas the fully spatiotemporal coefficient case is more challenging, with relative errors on the order of 10−1. Ablation studies, noise and observation-sensitivity tests, inverse-PINN baseline comparisons, and an independent forward-solve benchmark are provided to assess the behavior of the method. The study is restricted to controlled numerical settings with known forcing and prescribed Dirichlet data. Extensions to irregular geometries, mixed boundary conditions, model-form errors, real measurements, and large-scale implementations remain topics for future work.

Authors

Institutions

Publication Details

Journal
Mathematics
Published
2026-11-16
DOI
https://doi.org/10.3390/math14224207
Primary Topic
Model Reduction and Neural Networks
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A Hard-Constrained PINN Framework for Variable-Coefficient Identification in the Two-Dimensional Sine-Gordon Equation

Jinyao Xie
4 links
Mathematics
Model Reduction and Neural Networks
article

A Hard-Constrained PINN Framework for Variable-Coefficient Identification in the Two-Dimensional Sine-Gordon Equation

Jinyao Xie
article en

Abstract

This paper proposes a hard-constrained dual-branch physics-informed neural network (PINN) framework for identifying unknown variable coefficients in the two-dimensional sine-Gordon equation. Three prescribed classes of coefficients are considered, including time-dependent, space-dependent, and spatiotemporally varying coefficients. The framework uses a state network and a coefficient network to reconstruct the state field and the unknown coefficient simultaneously. A hard-constrained trial solution is constructed to satisfy the initial displacement, initial velocity, and Dirichlet boundary conditions analytically. Residual-based adaptive refinement and coefficient smoothness regularization are further incorporated to improve the empirical inversion behavior in the tested settings. Numerical experiments are conducted on a unit-square domain with known source terms and prescribed initial-boundary data. The results show that the time-dependent and space-dependent coefficients can be recovered with relative coefficient errors mainly on the order of 10−2, whereas the fully spatiotemporal coefficient case is more challenging, with relative errors on the order of 10−1. Ablation studies, noise and observation-sensitivity tests, inverse-PINN baseline comparisons, and an independent forward-solve benchmark are provided to assess the behavior of the method. The study is restricted to controlled numerical settings with known forcing and prescribed Dirichlet data. Extensions to irregular geometries, mixed boundary conditions, model-form errors, real measurements, and large-scale implementations remain topics for future work.

MathematicsVol. 14(22)
North Minzu University (CN)
North Minzu University
Model Reduction and Neural Networks
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