Physics Informed Deep Learning (Part I): Data-driven Solutions of Nonlinear Partial Differential Equations
By Maizar Raissi, Paris Perdikaris, G. Em Karniadakis. 2017
1,171 citations recorded by the source corpus.
We introduce physics informed neural networks -- neural networks that are trained to solve supervised learning tasks while respecting any given law of physics described by general nonlinear partial differential equations. In this two part treatise, we present our developments in the context of solving two main classes of problems: data-driven solution and data-driven discovery of partial differential equations. Depending on the nature and arrangement of the available data, we devise two distinct classes of algorithms, namely continuous time and discrete time models. The resulting neural networks form a new class of data-efficient universal function approximators that naturally encode any underlying physical laws as prior information. In this first part, we demonstrate how these networks can be used to infer solutions to partial differential equations, and obtain physics-informed surrogate models that are fully differentiable with respect to all input coordinates and free parameters.
Subfields: Artificial Intelligence, Computational Mechanics, Statistical and Nonlinear Physics, AI for Physical Sciences
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- G. Em Karniadakis
- Paris Perdikaris
- Liu Yang
- Yibo Yang
- Maizar Raissi
- Lu Lu
- Dongkun Zhang
- Stefania Fresca
In Computational Mechanics
- G. Em Karniadakis
- Andrea Manzoni
- Paris Perdikaris
- Stefania Fresca
- Timon Rabczuk
- Maizar Raissi
- Yibo Yang
- Zhiping Mao
In Statistical and Nonlinear Physics
- G. Em Karniadakis
- Lu Lu
- Paris Perdikaris
- Zhiping Mao
- Andrea Manzoni
- Maizar Raissi
- Gianluigi Rozza
- Liu Yang
In AI for Physical Sciences
- G. Em Karniadakis
- Paris Perdikaris
- Lu Lu
- Maizar Raissi
- Stefania Fresca
- Andrea Manzoni
- Zhiping Mao
- Yibo Yang
Citations in the corpus
- On the Art and Science of Machine Learning Explanations (2018, background): “Many of the breakthroughs in ANN explanation stem from the straight-forward calculation of accurate derivatives of the trained ANN response function w.r.t. to input features made possible by the proliferation of deep learning toolkits such as tensor-flow [22].”
- Differentiable Physics-informed Graph Networks (2019, background): “Raissi et al. (2017a;b) introduced a concept of physics-informed neural networks, which utilize physics equations explicitly to train neural networks.”
- Neural-net-induced Gaussian process regression for function approximation and PDE solution (2018, methodology): “We also solve the same equation using the NN approach proposed in [8].”
- Deep Hidden Physics Models: Deep Learning of Nonlinear Partial Differential Equations (2018, methodology): “One could avoid the first drawback concerning numerical differentiation by assigning prior distributions in the forms of Gaussian processes [38, 39, 40, 41] or neural networks [35, 34] to the unknown solution u.”
- Multistep Neural Networks for Data-driven Discovery of Nonlinear Dynamical Systems (2018, background): “Inspired by recent developments in physics-informed deep learning [19, 20], we construct structured nonlinear regression models that can discover the dynamic dependencies in a given set of temporal data-snapshots, and return a closed form model that can be subsequently used to forecast future states”
- Adversarial Uncertainty Quantification in Physics-Informed Neural Networks (2018, methodology): “[6, 7] have also pointed out cases in which the training phase of these algorithms faces severe difficulties for reasons that are cur-”
- DGM: A deep learning algorithm for solving partial differential equations (2017, background): “Recently, [40, 41] develop physics informed deep learning models.”
- Machine Learning of Space-Fractional Differential Equations (2018, methodology): “Neural networks were used to solve and discover (integer-order) PDEs in [28, 29].”
- The role of artificial intelligence in achieving the Sustainable Development Goals (2019, neutral): “…much less energy than what is used to train AI models, the available knowledge introduced in the model (see, for instance, physics-informed deep learning 18 ) does not need to be learnt through data-intensive training, a fact that may signi fi cantly reduce the associated energy consumption.”
- Machine learning for metal additive manufacturing: predicting temperature and melt pool fluid dynamics using physics-informed neural networks (2020, methodology): “In particular, a widely used approach in SciML is to train a conventional deep learning (DL) model such as Gaussian process regression (GPR) [43,44] or deep neural network (DNN) [45–49] with physical principle constraints.”
- Learning Parameters and Constitutive Relationships with Physics Informed Deep Neural Networks (2018, methodology): “Also, our method can be easily extended to timedependent PDEs by defining DNNs in Eq (5) as functions of both x and t [18, 19].”
- Physics-informed neural networks (PINNs) for fluid mechanics: a review (2021, background): “First proposed in [19,20], see also [15], PINNs have attracted a lot of attention in the scientific computing community as well as the fluid mechanics community.”
- Multiscale modeling meets machine learning: What can we learn? (2019, methodology): “Naturally, this brings up the question that, provided we know the underlying physics, can we integrate our prior knowledge to constrain the the space of admissible solutions to a manageable size [100]? Recent trends in computational physics suggest that exactly this approach [14]: to create data-eff”
- Weak Adversarial Networks for High-dimensional Partial Differential Equations (2019, methodology): “Solving general high-dimensional partial differential equations (PDEs) has been a long-standing challenge in numerical analysis and computation [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12].”
- Physics-aware Difference Graph Networks for Sparsely-Observed Dynamics (2020, background): “Since Raissi et al. (2017a;b) use the coordinates as input and compute derivatives based on the coordinates to represent the equation, the setting is only valid when the data are densely observed over spatial and temporal space.”