Abstract:Data-driven discovery of governing equations in computational science has emerged as a new paradigm for obtaining accurate physical models and as a possible alternative to theoretical derivations. The recently developed physics-informed neural networks (PINNs) have also been employed to learn governing equations given data across diverse scientific disciplines. Despite the effectiveness of PINNs for discovering governing equations, the physical models encoded in PINNs may be misspecified in complex systems as some of the physical processes may not be fully understood, leading to the poor accuracy of PINN predictions. In this work, we present a general approach to correct the misspecified physical models in PINNs for discovering governing equations, given some sparse and/or noisy data. Specifically, we first encode the assumed physical models, which may be misspecified, then employ other deep neural networks (DNNs) to model the discrepancy between the imperfect models and the observational data. Due to the expressivity of DNNs, the proposed method is capable of reducing the computational errors caused by the model misspecification and thus enables the applications of PINNs in complex systems where the physical processes are not exactly known. Furthermore, we utilize the Bayesian PINNs (B-PINNs) and/or ensemble PINNs to quantify uncertainties arising from noisy and/or gappy data in the discovered governing equations. A series of numerical examples including non-Newtonian channel and cavity flows demonstrate that the added DNNs are capable of correcting the model misspecification in PINNs and thus reduce the discrepancy between the physical models and the observational data. We envision that the proposed approach will extend the applications of PINNs for discovering governing equations in problems where the physico-chemical or biological processes are not well understood.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the model mismatch problem caused by incomplete or incorrect understanding of certain physical processes when using Physics - Informed Neural Networks (PINNs) to discover the governing equations of physical systems. Specifically, when the physical model is mismatched in a complex system, the prediction accuracy of PINNs will be affected. To address this challenge, the author proposes a general method to correct the model mismatch problem in PINNs and reduces the computational error caused by model mismatch by introducing other Deep Neural Networks (DNNs) to model the difference between the observed data and the imperfect model. In addition, the author also uses Bayesian Physics - Informed Neural Networks (B - PINNs) and ensemble PINNs to quantify the uncertainty caused by noise and/or missing data. ### Main contributions: 1. **Correcting model mismatch**: By adding an additional DNN to model the difference between the observed data and the assumed physical model, the error caused by model mismatch is reduced. 2. **Uncertainty quantification**: Use B - PINNs and ensemble PINNs to quantify the uncertainty caused by noise and/or missing data. 3. **Symbolic regression**: Combine the symbolic regression method to extract explicit governing equations from the trained DNN. ### Numerical experiments: The paper verifies the effectiveness of the proposed method through four numerical experiments: 1. **ODE system**: Study the influence of different prior knowledge on the reaction model. 2. **One - dimensional reaction - diffusion equation**: Deal with the reaction model that may be mismatched. 3. **Two - dimensional channel flow**: Assume that the constitutive relation of non - Newtonian fluid is mismatched as Newtonian fluid. 4. **Two - dimensional cavity flow**: Also assume that the constitutive relation of non - Newtonian fluid is mismatched as Newtonian fluid. ### Experimental results: - **Correct physical model**: PINNs can accurately infer the reaction term \(\phi\) and the surrogate model \(u_\theta\), while fitting the data and satisfying the ODE. - **Mismatched physical model**: PINNs cannot accurately infer \(\phi\), and the prediction results are completely wrong. - **Correcting the mismatched physical model**: By adding DNN to correct the mismatched model, the prediction accuracy of \(\phi\) is significantly improved, and the computational accuracy of \(u\), \(f\) and \(\phi\) is similar to the case where the physical model is correctly specified. ### Conclusion: The method proposed in the paper can effectively correct the model mismatch problem in PINNs, and by introducing additional DNN and uncertainty quantification techniques, the prediction accuracy in complex systems is improved. These results indicate that the proposed method has broad application prospects in the case of incomplete understanding of physical processes.

Correcting model misspecification in physics-informed neural networks (PINNs)

Correcting model misspecification in physics-informed neural networks (PINNs)

Physics-informed neural networks (PINNs) for numerical model error approximation and superresolution

B-PINNs: Bayesian physics-informed neural networks for forward and inverse PDE problems with noisy data

Convergence and error analysis of PINNs

Recipes for when physics fails: recovering robust learning of physics informed neural networks

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

SetPINNs: Set-based Physics-informed Neural Networks

Physics-informed Neural Networks for Encoding Dynamics in Real Physical Systems

Error-Aware B-PINNs: Improving Uncertainty Quantification in Bayesian Physics-Informed Neural Networks

Machine Learning Through Physics–Informed Neural Networks: Progress and Challenges

Binary structured physics-informed neural networks for solving equations with rapidly changing solutions

Scientific Machine Learning Through Physics–Informed Neural Networks: Where we are and What's Next

Enhancing PINNs for solving PDEs via adaptive collocation point movement and adaptive loss weighting

Densely Multiplied Physics Informed Neural Networks

Loss-attentional physics-informed neural networks

Characterizing possible failure modes in physics-informed neural networks

Theory-guided physics-informed neural networks for boundary layer problems with singular perturbation

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

A novel discretized physics-informed neural network model applied to the Navier-Stokes equations

Numerical analysis of physics-informed neural networks and related models in physics-informed machine learning