Gravity-Informed Neural Networks for Post-Newtonian Binary Dynamics

This paper introduces GravINNs, a gravity-informed neural network framework designed to learn post-Newtonian binary orbital dynamics. The approach directly incorporates equations of motion into the training phase, enabling physics-informed learning. The framework is developed at three complementary levels: first constructing single-orbit surrogates for conservative post-Newtonian dynamics, then extending to a parametric model spanning a four-dimensional parameter space. This work targets gravitational physics and relativistic astrophysics applications.