Showing 1–2 of 2 results for author: Timm, C

Tunable Complexity Benchmarks for Evaluating Physics-Informed Neural Networks on Coupled Ordinary Differential Equations

Authors: Alexander New, Benjamin Eng, Andrea C. Timm, Andrew S. Gearhart

Abstract: In this work, we assess the ability of physics-informed neural networks (PINNs) to solve increasingly-complex coupled ordinary differential equations (ODEs). We focus on a pair of benchmarks: discretized partial differential equations and harmonic oscillators, each of which has a tunable parameter that controls its complexity. Even by varying network architecture and applying a state-of-the-art tr… ▽ More In this work, we assess the ability of physics-informed neural networks (PINNs) to solve increasingly-complex coupled ordinary differential equations (ODEs). We focus on a pair of benchmarks: discretized partial differential equations and harmonic oscillators, each of which has a tunable parameter that controls its complexity. Even by varying network architecture and applying a state-of-the-art training method that accounts for "difficult" training regions, we show that PINNs eventually fail to produce correct solutions to these benchmarks as their complexity -- the number of equations and the size of time domain -- increases. We identify several reasons why this may be the case, including insufficient network capacity, poor conditioning of the ODEs, and high local curvature, as measured by the Laplacian of the PINN loss. △ Less

Submitted 14 October, 2022; originally announced October 2022.

Comments: Accepted at KGML-AAAI-22: https://sites.google.com/vt.edu/kgml-aaai-22

What is the probability that a large random matrix has no real eigenvalues?

Authors: Eugene Kanzieper, Mihail Poplavskyi, Carsten Timm, Roger Tribe, Oleg Zaboronski

Abstract: We study the large-$n$ limit of the probability $p_{2n,2k}$ that a random $2n\times 2n$ matrix sampled from the real Ginibre ensemble has $2k$ real eigenvalues. We prove that, $$\lim_{n\rightarrow \infty}\frac {1}{\sqrt{2n}} \log p_{2n,2k}=\lim_{n\rightarrow \infty}\frac {1}{\sqrt{2n}} \log p_{2n,0}= -\frac{1}{\sqrt{2π}}ζ\left(\frac{3}{2}\right),$$ where $ζ$ is the Riemann zeta-function. Moreover,… ▽ More We study the large-$n$ limit of the probability $p_{2n,2k}$ that a random $2n\times 2n$ matrix sampled from the real Ginibre ensemble has $2k$ real eigenvalues. We prove that, $$\lim_{n\rightarrow \infty}\frac {1}{\sqrt{2n}} \log p_{2n,2k}=\lim_{n\rightarrow \infty}\frac {1}{\sqrt{2n}} \log p_{2n,0}= -\frac{1}{\sqrt{2π}}ζ\left(\frac{3}{2}\right),$$ where $ζ$ is the Riemann zeta-function. Moreover, for any sequence of non-negative integers $(k_n)_{n\geq 1}$, $$\lim_{n\rightarrow \infty}\frac {1}{\sqrt{2n}} \log p_{2n,2k_n}=-\frac{1}{\sqrt{2π}}ζ\left(\frac{3}{2}\right),$$ provided $\lim_{n\rightarrow \infty} \left(n^{-1/2}\log(n)\right) k_{n}=0$. △ Less

Submitted 26 March, 2015; originally announced March 2015.

Comments: 23 pages, 1 figure

MSC Class: 60B20

Journal ref: The Annals of Applied Probability 2016, Vol. 26, No. 5, 2733-2753