Abstract:Physics-informed deep learning often faces optimization challenges due to the complexity of solving partial differential equations (PDEs), which involve exploring large solution spaces, require numerous iterations, and can lead to unstable training. These challenges arise particularly from the ill-conditioning of the optimization problem, caused by the differential terms in the loss function. To address these issues, we propose learning a solver, i.e., solving PDEs using a physics-informed iterative algorithm trained on data. Our method learns to condition a gradient descent algorithm that automatically adapts to each PDE instance, significantly accelerating and stabilizing the optimization process and enabling faster convergence of physics-aware models. Furthermore, while traditional physics-informed methods solve for a single PDE instance, our approach addresses parametric PDEs. Specifically, our method integrates the physical loss gradient with the PDE parameters to solve over a distribution of PDE parameters, including coefficients, initial conditions, or boundary conditions. We demonstrate the effectiveness of our method through empirical experiments on multiple datasets, comparing training and test-time optimization performance.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is **the optimization challenges faced by physics - informed deep - learning methods when solving partial differential equations (PDEs)**. Specifically, these challenges include: 1. **Complexity of the optimization problem**: Due to the complexity of PDEs, the solution process needs to explore a large solution space, which usually requires a large number of iteration steps and may lead to unstable training. 2. **Condition number problem**: The condition number of the optimization problem is poor, especially when the loss function contains differential terms, which will cause the gradient descent algorithm in the optimization process to converge slowly and unstably. To solve these problems, the paper proposes a new method - **learning a solver**, that is, training a physics - informed iterative algorithm based on data to solve PDEs. The main features of this method include: - **Adaptive gradient descent**: The learned solver can automatically adapt to each PDE instance, significantly accelerating and stabilizing the optimization process, thereby achieving faster convergence. - **Solving parameterized PDEs**: Unlike traditional physics - informed methods that can only solve a single PDE instance, this method can handle parameterized PDEs, that is, distributions containing coefficients, initial conditions or boundary conditions. ### Method overview 1. **Problem definition**: - Consider a class of boundary value problems parameterized by the parameter \(\gamma\), in the form of: \[ N(u; \gamma)=f \quad \text{in} \quad \Omega, \] \[ B(u) = g \quad \text{on} \quad \partial\Omega, \] where \(N\) is a possibly nonlinear differential operator, \(B\) is a boundary operator, \(g\) is the initial/boundary condition, and \(f\) is the source term. 2. **Objective**: - Develop a general algorithm that can generate an approximate solution \(u\) given a PDE and its different input sets (\(\gamma, f, g\)). 3. **Training data**: - Assume there is a data set containing \(M\) problem instances, each instance consisting of PDE parameters \((\gamma_i, f_i, g_i)\) and the corresponding target solutions \((u_i)\) on grid points. 4. **Method framework**: - **Forward propagation**: Use a neural network \(F_\varrho\) to transform the gradient, thereby improving the performance of the baseline gradient descent algorithm (such as SGD). - **Backward propagation**: Update the parameters \(\varrho\) of the neural solver by minimizing the data loss \(L_{\text{DATA}}\). 5. **Inference process**: - For a new PDE instance, start from the initial parameter \(\Theta_0\), and iteratively update the parameter \(\Theta\) through the neural solver \(F_\varrho\) until the required accuracy is reached. ### Experimental results The paper verifies the effectiveness of this method through multiple data sets, including: - **Helmholtz equation** - **Poisson equation** - **Non - linear reaction - diffusion equation (NLRD)** - **Darcy flow equation** - **Heat conduction equation** The experimental results show that this method performs excellently in all evaluations. Compared with other baseline methods (such as fully - supervised methods, unsupervised physics - informed methods, hybrid methods, etc.), it improves the relative mean - squared error (Relative MSE) by at least 1 - 2 orders of magnitude. ### Conclusion The learning - solver method proposed in this paper effectively solves the optimization challenges of physics - informed deep - learning methods in solving PDEs, and has significant advantages especially for solving parameterized PDEs.

Learning a Neural Solver for Parametric PDE to Enhance Physics-Informed Methods

Meta-learning of Physics-informed Neural Networks for Efficiently Solving Newly Given PDEs

Physics Constrained Learning for Data-driven Inverse Modeling from Sparse Observations

Physics-informed learning of governing equations from scarce data

Physics-Guided Discovery of Highly Nonlinear Parametric Partial Differential Equations

Learning from Integral Losses in Physics Informed Neural Networks

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

Learning solutions of parametric Navier-Stokes with physics-informed neural networks

Solving Partial Differential Equations Using Deep Learning and Physical Constraints

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

Integration of physics-informed operator learning and finite element method for parametric learning of partial differential equations

Physics-informed machine learning

Enhancing neurodynamic approach with physics-informed neural networks for solving non-smooth convex optimization problems

Physics-constrained Deep Learning for Solving Seepage Equation

An extended physics informed neural network for preliminary analysis of parametric optimal control problems

An Analysis of Physics-Informed Neural Networks

Learning to Accelerate Partial Differential Equations via Latent Global Evolution

Discovering Nonlinear PDEs from Scarce Data with Physics-encoded Learning

Meta-PDE: Learning to Solve PDEs Quickly Without a Mesh

Semi-Implicit Neural Solver for Time-dependent Partial Differential Equations

On Physics-Informed Deep Learning for Solving Navier-Stokes Equations