Speaker
Description
The projective quantum Monte Carlo (PQMC) algorithm provides a way to study the ground state properties of quantum many-body systems. Here, we employ a self-learning PQMC framework using an artificial neural network ansatz as a guiding wavefunction to study Rydberg square lattice systems. The accuracy of the PQMC simulations is improved by employing an accurate neural network ansatz, as evidenced by the greater performance of recurrent neural network (RNN) guiding wavefunctions compared to restricted Boltzmann machines (RBMs). We recover the phase transition from the disordered to the checkerboard phase of the Rydberg array system using the trained wavefunction. We further demonstrate that the efficiency of the self-learning PQMC simulation is significantly improved by training the guiding wavefunction using projected measurement data from a quantum simulator, despite imperfections in the data. Our findings suggest a robust pathway to utilize expressive neural networks and state-of-the-art quantum simulators for the efficient and accurate PQMC simulations.