Deep Potential Molecular Dynamics Study of PMN Relaxor Ferroelectricity

We investigate the relaxor ferroelectric Pb(Mg_1/3Nb_2/3)O₃ (PMN) crystal through all-atom molecular dynamics simulations, employing a neural-network potential energy model (Deep Potential) trained on density functional theory (DFT) data with the SCAN approximation. With consistent DFT data on maximally localized Wannier functions, another neural network model (Deep Wannier) is trained to predict the bulk polarization of PMN. The static and frequency-dependent susceptibility of PMN predicted by our models are found to qualitatively agree with experiments.

To gain insight into the mechanism of relaxor ferroelectricity, we examine the temperature behavior of the local order parameter, i.e., a soft pseudospin identified by the local dipole associated to each PbMgO₃ or PbNbO₃ unit. By analyzing the Edwards-Anderson order parameter and the correlation of the dipoles with the local chemical composition, we establish a connection between PMN and conventional spin glasses.