ABSTRACT
Computational physics is a new discipline that uses computers and computer science as tools and approaches, applies appropriate mathematical methods, conducts numerical analysis of physical problems, and performs numerical simulation of practical processes. This chapter reviews the applications of various emerging deep learning technologies in computational physics in recent years, and focuses on the specific implementation of several special methods. Common encountered physics-informed neural network (PINN) is adopted in the rectangular coordinate system. This chapter expands it to any orthogonal curvilinear coordinate system and resolves some typical physics problems. PINN can also be employed in spherical coordinate systems. The chapter investigates the electrostatic field in spherical coordinates. The convolutional PINN is progressively becoming prevalent for the merits of parameter sharing kernels, spatial feature extractions, and light network weights. A deep learning network based on the GNN is proposed to predict the electrical potential with the known boundaries and the geometries.
