Abstract:
Simulating the real-time dynamics of strongly correlated electrons remains a major computational challenge. I will present our approach based on transformer-based neural quantum states (NQS), where many-body wavefunctions are represented by neural networks and their time evolution is learned by minimizing the residual of the time-dependent Schrödinger equation. I will discuss recent insights into the tNQS framework, particularly regarding its accuracy and variational bias, and demonstrate its application to quantum quenches in the t-J model, a paradigmatic effective model of doped Mott insulators relevant to cuprate superconductors. Finally, I will present preliminary results on extending this approach to the three-band Emery model, providing a first step towards studying real-time dynamics in a more microscopic description of cuprates.