28/04/2026
The next talk at the AUB Mathematics Seminar will be on Tuesday, May 5, from 11:30 to 12:30 in Bliss Hall 205. Everyone interested is warmly welcome.
Speaker: Joe Germany (Department of Mathematics, AUB)
Title: Approximation Theory for Exactly Periodic, Divergence-Free Neural Networks and Application to the Navier-Stokes and Nonlocal Advection-Diffusion PDEs
Abstract: Many physically relevant models have properties such as periodic boundary conditions or incompressibility. Standard neural network architectures do not enforce these properties exactly, creating difficulties in carrying out high-regularity analysis and energy estimates.
In this talk, we introduce neural network architectures that are exactly periodic and/or exactly divergence-free, by construction, and we establish approximation theories for these methods.
Our networks display abundant analytical advantages for carrying out energy estimates and allow for the approximation of nonlocal, periodic operators like the fractional Laplacian and the Riesz transform. We apply our techniques to study the convergence analysis of such networks for the incompressible Navier-Stokes equations and a family of fractional nonlocal advection-diffusion equations.
Finally, we present an a posteriori framework to rigorously translate empirically observed decay rates of neural network approximations into decay rates for the true PDE solution, providing a novel bridge between physics-informed neural networks and rigorous numerics.
This is joint work with Elie Abdo (AUB), Xu Yang (UCSB), Ruimeng Hu (UCSB), and Lihui Chai (Sun Yat-sen University).
About the speaker: Joe Germany is an undergraduate student in Math and Physics at AUB. His interests lie at the intersection of machine learning, physics, and partial differential equations, particularly in the study of structure-preserving neural architectures and the mathematical analysis of machine learning algorithms.
Up-to-date information regarding the upcoming talks can be found on the website of the seminar:
https://www.aub.edu.lb/fas/math/Pages/ags.aspx
This seminar series is supported by the FAS Dean's Office.