Abstract:Physics Informed Neural Networks is a numerical method which uses neural networks to approximate solutions of partial differential equations. It has received a lot of attention and is currently used in numerous physical and engineering problems. The mathematical understanding of these methods is limited, and in particular, it seems that, a consistent notion of stability is missing. Towards addressing this issue we consider model problems of partial differential equations, namely linear elliptic and parabolic PDEs. We consider problems with different stability properties, and problems with time discrete training. Motivated by tools of nonlinear calculus of variations we systematically show that coercivity of the energies and associated compactness provide the right framework for stability. For time discrete training we show that if these properties fail to hold then methods may become unstable. Furthermore, using tools of $\Gamma-$convergence we provide new convergence results for weak solutions by only requiring that the neural network spaces are chosen to have suitable approximation properties.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper primarily explores the issues of stability and convergence of Physics Informed Neural Networks (PINNs) when solving Partial Differential Equations (PDEs). Specifically: 1. **Introduction of the concept of stability**: - The current PINNs methods have limited mathematical understanding, particularly lacking a consistent definition of stability. - The paper attempts to introduce a new concept of stability and validate its effectiveness through models of linear elliptic and parabolic PDEs. 2. **Different types of PDEs and their stability**: - It considers PDEs with different stability properties, including linear elliptic and parabolic PDEs. - For the case of time-discrete training, it demonstrates that if certain properties do not hold, the method may become unstable. 3. **Framework of stability and convergence**: - Using tools from nonlinear variational methods, it systematically proves that coercivity of energy and related compactness provide the correct framework for stability. - For time-discrete training, if these properties do not hold, the method may become unstable. - By utilizing tools of Γ-convergence, it provides new convergence results for weak solutions, requiring only that the neural network space has appropriate approximation properties. ### Summary This paper aims to provide a systematic framework for the analysis of stability and convergence of the PINNs method when solving PDEs, thereby addressing the current lack of stability in existing methods. By studying models of linear elliptic and parabolic PDEs, the paper proposes a new concept of stability and proves its effectiveness. Additionally, the paper explores the impact of time-discrete training on the stability and convergence of the algorithm.

On the Stability and Convergence of Physics Informed Neural Networks

Some Results on Neural Network Stability, Consistency, and Convergence: Insights into Non-IID Data, High-Dimensional Settings, and Physics-Informed Neural Networks

An Analysis of Physics-Informed Neural Networks

Stability Analysis of Physics-Informed Neural Networks for Stiff Linear Differential Equations

Transient Stability Analysis with Physics-Informed Neural Networks

Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations

On instabilities in neural network-based physics simulators

Physics Informed Deep Learning (Part I): Data-driven Solutions of Nonlinear Partial Differential Equations

Convergence and error analysis of PINNs

Invariant Physics-Informed Neural Networks for Ordinary Differential Equations

Critical Investigation of Failure Modes in Physics-informed Neural Networks

Physics informed neural networks for continuum micromechanics

Control of Partial Differential Equations via Physics-Informed Neural Networks

Convergence of physics-informed neural networks applied to linear second-order elliptic interface problems

Physics Informed Neural Networks (PINNs) as intelligent computing technique for solving partial differential equations: Limitation and Future prospects

Physics-Informed Neural Network Policy Iteration: Algorithms, Convergence, and Verification

Evaluating Error Bound for Physics-Informed Neural Networks on Linear Dynamical Systems

A Rate of Convergence of Physics Informed Neural Networks for the Linear Second Order Elliptic PDEs

Characterizing and Mitigating the Difficulty in Training Physics-informed Artificial Neural Networks under Pointwise Constraints

Physics informed neural networks for elliptic equations with oscillatory differential operators

A Stabilized Physics Informed Neural Networks Method for Wave Equations