ReLU^k Neural de Rham Complexes

DOI10.48550/arXiv.2607.22478MaRDI QIDQ7284390arXivFDO

Authors Jindong Wang, Jinchao Xu, Kaibo Hu

Publication date 24 July 2026

Published in arXiv

Tldr We construct finite-dimensional de Rham subcomplexes generated by fixed-neuron shallow ReLUk neural networks and introduce spaces of neural differential forms compatible with the exterior derivative. We prove exactness in arbitrary dimension under a linear independence assumption on the lowest-order family and provide numerical evidence for stable discretizations and convergence rates consistent with the underlying approximation theory.

Full work available at URL https://doi.org/10.48550/arXiv.2607.22478

arXiv classification

math.NA

Cites work

• Approximation by superpositions of a sigmoidal function
• BGG sequences with weak regularity and applications
• Bernstein-Gelfand-Gelfand sequences
• Complexes from complexes
• Deep learning in high dimension: neural network expression rates for generalized polynomial chaos expansions in UQ
• Divergence-free Linearized Neural Networks: Integral Representation and Optimal Approximation Rates
• Finite element exterior calculus, homological techniques, and applications
• Finite element exterior calculus: From Hodge theory to numerical stability
• Integral Representations of Sobolev Spaces via ReLU$^k$ Activation Function and Optimal Error Estimates for Linearized Networks
• Mixed finite element methods and applications
• Multilayer feedforward networks are universal approximators

This page was built for publication: ReLU$^k$ Neural de Rham Complexes

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7284390)