Physics-Informed Neural Networks (PINNs) are a new family of numerical methods, based on deep learning, for modeling boundary value problems. They offer an advantage over traditional numerical methods for high-dimensional, parametric, and data-driven problems. However, they perform poorly on problems where the solution exhibits high frequencies, such as discontinuities or sharp gradients. In this work, we develop a PINN-based solver for modeling three-dimensional, transient and static, parametric electromagnetic problems in discontinuous media. We use the first-order Maxwell's equations to train the neural network. We use a level-set function to represent the interface with a continuous function, and to enrich the network's inputs with high-frequencies and interface information. Finally, we validate the proposed methodology on multiple 3D, parametric, static, and transient problems.
opPINN: Physics-Informed Neural Network with operator learning to approximate solutions to the Fokker-Planck-Landau equation
Korea Institute for Advanced Study (KIAS)(韩国高等研究院);University of Southern California(南加利福尼亚大学);Pohang University of Science and Technology (POSTECH)(浦项科技大学)
Resolving positive semi-definiteness in physics-informed kernels for scientific machine learning
Institute of Theoretical & Computational Physics, Graz University of Technology(格拉茨工业大学理论与计算物理研究所);Courant Institute, New York University(纽约大学柯朗数学科学研究所);Division of Applied Mathematics, Brown University(布朗大学应用数学系);School of Mechanical Engineering, Purdue University(普渡大学机械工程学院)
Physics-Informed Neural Networks with Fourier Features and Attention-Driven Decoding
High Technology High School(高科技高中);Institute for Gravitation and the Cosmos(引力与宇宙研究所);The Pennsylvania State University(宾夕法尼亚州立大学);Department of Physics(物理系);Princeton University(普林斯顿大学)
School of Interdisciplinary Science, Beijing Institute of Technology(北京理工大学交叉科学学院);School of Mechatronical Engineering, Beijing Institute of Technology(北京理工大学机电工程学院);Beijing Institute of Technology (Zhuhai)(北京理工大学(珠海))