Do physics-informed neural networks (PINNs) need to be deep? Shallow PINNs using the Levenberg–Marquardt algorithm

Muhammad Luthfi Shahab, Imam Mukhlash, Hadi Susanto · 2026

Read Do physics-informed neural networks (PINNs) need to be deep? Shallow PINNs using the Levenberg–Marquardt algorithm by Muhammad Luthfi Shahab, Imam Mukhlash, Hadi Susanto (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-2602-08515.txt.

Description

Physics-informed neural networks are usually deep, but depth may not be necessary. Muhammad Luthfi Shahab and two co-authors test shallow networks trained with the Levenberg-Marquardt algorithm on nonlinear partial differential equations. Their arXiv paper recasts training as nonlinear least squares and derives exact derivative formulas for the network.

Those formulas yield an explicit Jacobian for the optimizer and clarify how outputs relate to spatial and temporal derivatives. Tests cover four classic equations, and the method beats BFGS, L-BFGS, and Adam on speed, accuracy, and final loss. Shallow networks even beat deeper ones with far fewer parameters. Readers get evidence that architecture and optimizer deserve joint attention.

Catalog record

CREATORMuhammad Luthfi Shahab, Imam Mukhlash, Hadi Susanto
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE60 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-08515

Browse the full library · About the Tower of Babel