Speaker
Description
We apply physics-informed neural networks (PINNs) to fully nonlinear relativistic spin hydrodynamics and study the dynamical conversion between spin and orbital angular momentum driven by the rotational viscosity, a dissipative effect intrinsic to spin hydrodynamics [1]. Such conversion is relevant to the phenomenology of relativistic heavy-ion collisions but has not been studied numerically in the fully nonlinear regime. Imposing total angular momentum conservation through the loss function, we achieve accurate numerical conservation, which is essential for reliably capturing the transfer between the two components. We demonstrate both orbital-to-spin and spin-to-orbital conversion, corresponding to the Barnett and Einstein-de Haas effects, respectively. We find that the conversion arises as friction between the rotation inside a fluid element and the surrounding vortical flow.
[1] H. Matsuda, K. Hattori, and K. Murase, arXiv:2512.17971.
