Physics-informed neural networks for solving nonlinear Bloch equations in atomic magnetometry

Abstract

In this study, we address the challenge of analyzing spatial spin distribution based on the nonlinear Bloch equations in atomic magnetometry through the use of physics-informed neural networks (PINNs). Atomic magnetometry plays a crucial role in the field of biomagnetism, where it is used to detect weak magnetic fields produced by the human brain, heart, and other organs. The Bloch equations describe the spin polarization of atomic clusters in an external magnetic field, but their nonlinearity can make the analysis of the spin distribution in spatial domain difficult. By utilizing PINNs, we provide a numerical solution to the nonlinear Bloch equations, examining the effect of different pump light schemes and wall collisions. Additionally, we propose a easily executed system identification method for the Bloch equations through the use of PINNs in a data-driven discovery mode, expanding the design space of atomic magnetometry beyond traditional simulation methods.