Abstract:Physics informed neural networks (PINNs) have emerged as a powerful tool to provide robust and accurate approximations of solutions to partial differential equations (PDEs). However, PINNs face serious difficulties and challenges when trying to approximate PDEs with dominant hyperbolic character. This research focuses on the development of a physics informed deep learning framework to approximate solutions to nonlinear PDEs that can develop shocks or discontinuities without any a-priori knowledge of the solution or the location of the discontinuities. The work takes motivation from finite element method that solves for solution values at nodes in the discretized domain and use these nodal values to obtain a globally defined solution field. Built on the rigorous mathematical foundations of the discontinuous Galerkin method, the framework naturally handles imposition of boundary conditions (Neumann/Dirichlet), entropy conditions, and regularity requirements. Several numerical experiments and validation with analytical solutions demonstrate the accuracy, robustness, and effectiveness of the proposed framework.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is the difficulties and challenges encountered by existing Physics - Informed Neural Networks (PINNs) when dealing with Partial Differential Equations (PDEs) with dominant hyperbolic characteristics, especially when these equations may develop shock waves or discontinuities. Specifically: 1. **Limitations of existing methods**: - PINNs face serious difficulties when dealing with solutions with sharp gradients or discontinuities, especially in the absence of artificial viscosity, and cannot provide reasonable approximations. - The quality of the solution, the convergence speed of the model, and the landscape of the loss function are highly dependent on the choice of the viscosity parameter. - PINNs show poor accuracy when dealing with inviscid scalar conservation laws, resulting in large generalization errors. - In the discontinuous region, the classical solutions of PDEs may not hold, further increasing the convergence challenges of current deep - learning - based methods. - Many existing methods need to know the discontinuity location in advance in order to predict solutions with sharp gradients. 2. **Research objectives**: - Develop a deep - learning framework based on the Discontinuous Galerkin (DG) method that can capture any shock waves or discontinuities without prior knowledge of the discontinuity location. - By combining the advantages of the DG method, enable the framework to handle discontinuities on a global scale while maintaining the smoothness and continuity of the solution at the discrete level. - Ensure the consistency of satisfying the entropy inequality and the zero - viscosity limit without adding extra penalty terms in the objective function. - Provide a method for naturally handling boundary conditions (Neumann/Dirichlet) without introducing extra terms in the composite loss function. 3. **Technical contributions**: - Propose a new deep - learning framework based on the Discontinuous Galerkin method for solving nonlinear PDEs with discontinuous solutions. - By choosing an appropriate function space, a global solution with specific regularity (continuity and differentiability) can be constructed. - Based on the DG - FEM discretization technique, the framework can capture sharp jumps or discontinuities in the solution. - Use the weak form to reduce the regularity requirements for the solution and use convolution operations to numerically approximate the integrals in the weak form. - Naturally handle initial conditions and Dirichlet/Neumann boundary conditions without introducing extra terms in the composite loss function. In summary, this paper aims to overcome the limitations of existing PINNs when dealing with PDEs with dominant hyperbolic characteristics, especially in dealing with shock waves or discontinuities, by developing a new deep - learning framework based on the Discontinuous Galerkin method.

A Deep Learning Framework for Solving Hyperbolic Partial Differential Equations: Part I

Solving Partial Differential Equations Using Deep Learning and Physical Constraints

A hands-on introduction to Physics-Informed Neural Networks for solving partial differential equations with benchmark tests taken from astrophysics and plasma physics

A physics-informed neural network framework for modeling obstacle-related equations

Physics-informed attention-based neural network for hyperbolic partial differential equations: application to the Buckley-Leverett problem

Solving High-Dimensional Partial Differential Equations Using Deep Learning.

A Physics-Informed Neural Network Framework For Partial Differential Equations on 3D Surfaces: Time-Dependent Problems

An Efficient Framework for Solving Forward and Inverse Problems of Nonlinear Partial Differential Equations Via Enhanced Physics-Informed Neural Network Based on Adaptive Learning

Partial‐differential‐algebraic equations of nonlinear dynamics by physics‐informed neural‐network: (I) Operator splitting and framework assessment

Solving Partial Differential Equations with Point Source Based on Physics-Informed Neural Networks

Partial-differential-algebraic equations of nonlinear dynamics by Physics-Informed Neural-Network: (I) Operator splitting and framework assessment

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

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

DeepXDE: A Deep Learning Library for Solving Differential Equations

Solving differential equations using physics informed deep learning: a hand-on tutorial with benchmark tests

An Overview on Machine Learning Methods for Partial Differential Equations: from Physics Informed Neural Networks to Deep Operator Learning

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

PinnDE: Physics-Informed Neural Networks for Solving Differential Equations

Data-Guided Physics-Informed Neural Networks for Solving Inverse Problems in Partial Differential Equations

Improved Deep Neural Networks with Domain Decomposition in Solving Partial Differential Equations

Operator Learning Enhanced Physics-informed Neural Networks for Solving Partial Differential Equations Characterized by Sharp Solutions