Physics-Informed Residuals for Adaptive Mesh Refinement in Finite-Difference PDE Solvers

Henry Kasumba, Ronald Katende · 2026

Read Physics-Informed Residuals for Adaptive Mesh Refinement in Finite-Difference PDE Solvers by Henry Kasumba, Ronald Katende (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-02475.txt.

Description

Kasumba and Katende use neural networks as a probe, not a solver, for partial differential equations. Classic finite-difference solvers stay reliable. Their cost depends on where mesh detail sits. Uniform refinement wastes effort where the solution is smooth. Sharp gradients, fronts, and oscillations need the detail instead.

Their hybrid strategy samples a physics-informed neural network residual over the domain. The residual becomes cellwise indicators for mesh refinement. The final answer still comes from a finite-difference solver. On the viscous Burgers equation, PINN-threshold refinement hits error 0.021067 with 60 degrees of freedom. Uniform refinement needs 192 for error 0.022617. At matched mesh size, error falls by about 67.5 percent. A gradient indicator stays slightly more accurate. Two- and three-dimensional proxy tests agree. PINN-guided refinement is useful, not universally best.

Catalog record

CREATORHenry Kasumba, Ronald Katende
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE72 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.02475

Browse the full library · About the Tower of Babel