A Comparative Study of the Physics-Informed Neural Networks and the Nonstandard Finite Difference Methods for the Solution of the Two-Dimensional Fisher-KPP Reaction-Diffusion Equation

This study presents a comparative investigation of a Nonstandard Finite Difference (NSFD) method and a Physics-Informed Neural Network (PINN) for the numerical solution of the two-dimensional nonlinear Fisher–KPP reaction–diffusion equation. The NSFD scheme incorporates a nonstandard temporal denominator, a nonlocal treatment of the nonlinear reaction term, and mixed time levels, while the PINN approximates the solution by minimizing a physics-informed loss constructed from the governing equation and the prescribed initial and boundary conditions. The reference traveling-wave solution is not used as interior training data for the PINN or in the NSFD interior update, but is employed independently for numerical validation. Theoretical analysis of the NSFD scheme establishes positivity, preservation of the invariant interval [0, 1], consistency, linearized stability, and convergence. In particular, first-order max-norm convergence is established along the refinement path ∆x = ∆y = h and ∆t = h3. Numerical grid-refinement and invariant-region experiments are consistent with these theoretical results. The PINN is trained using a two-stage Adam–L-BFGS optimization procedure and subsequently evaluated independently against the reference solution. The numerical comparison shows that both methods successfully approximate the spatial and temporal behavior of the reference solution. However, under the computational settings considered, the PINN produces smaller RMSE, MAE, maximum absolute error, and relative L2 error than the NSFD method, thereby providing a more accurate approximation of the benchmark. The computational comparison reveals a complementary trade-off: the NSFD method provides a substantially less expensive direct time-marching solution, whereas the PINN incurs a considerably higher initial training cost but enables rapid post-training evaluation. These results demonstrate that, for the present Fisher–KPP benchmark, the PINN performs better in terms of numerical accuracy, while the NSFD method offers lower direct computational cost together with analytically established structure-preserving and convergence properties.

Authors

Publication Details

Journal
Preprints.org
Published
2026-11-18
DOI
https://doi.org/10.20944/preprints202611.1939.v1
Primary Topic
Model Reduction and Neural Networks
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

A Comparative Study of the Physics-Informed Neural Networks and the Nonstandard Finite Difference Methods for the Solution of the Two-Dimensional Fisher-KPP Reaction-Diffusion Equation

Kizito Muzhinji, Ndivhuwo Ndou, Lekaba Thabo Maomela
Preprints.org
Model Reduction and Neural Networks
preprint

A Comparative Study of the Physics-Informed Neural Networks and the Nonstandard Finite Difference Methods for the Solution of the Two-Dimensional Fisher-KPP Reaction-Diffusion Equation

Kizito Muzhinji, Ndivhuwo Ndou, Lekaba Thabo Maomela
preprint en

Abstract

This study presents a comparative investigation of a Nonstandard Finite Difference (NSFD) method and a Physics-Informed Neural Network (PINN) for the numerical solution of the two-dimensional nonlinear Fisher–KPP reaction–diffusion equation. The NSFD scheme incorporates a nonstandard temporal denominator, a nonlocal treatment of the nonlinear reaction term, and mixed time levels, while the PINN approximates the solution by minimizing a physics-informed loss constructed from the governing equation and the prescribed initial and boundary conditions. The reference traveling-wave solution is not used as interior training data for the PINN or in the NSFD interior update, but is employed independently for numerical validation. Theoretical analysis of the NSFD scheme establishes positivity, preservation of the invariant interval [0, 1], consistency, linearized stability, and convergence. In particular, first-order max-norm convergence is established along the refinement path ∆x = ∆y = h and ∆t = h3. Numerical grid-refinement and invariant-region experiments are consistent with these theoretical results. The PINN is trained using a two-stage Adam–L-BFGS optimization procedure and subsequently evaluated independently against the reference solution. The numerical comparison shows that both methods successfully approximate the spatial and temporal behavior of the reference solution. However, under the computational settings considered, the PINN produces smaller RMSE, MAE, maximum absolute error, and relative L2 error than the NSFD method, thereby providing a more accurate approximation of the benchmark. The computational comparison reveals a complementary trade-off: the NSFD method provides a substantially less expensive direct time-marching solution, whereas the PINN incurs a considerably higher initial training cost but enables rapid post-training evaluation. These results demonstrate that, for the present Fisher–KPP benchmark, the PINN performs better in terms of numerical accuracy, while the NSFD method offers lower direct computational cost together with analytically established structure-preserving and convergence properties.

Preprints.org
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