Equivariant machine learning of Electric Field Gradients -- Predicting the quadrupolar coupling constant in the MAPbI$_3$ phase transition
We present a strategy combining machine learning and first-principles calculations to achieve highly accurate nuclear quadrupolar coupling constant predictions. Our approach employs two distinct machine-learning frameworks: a machine-learned force field to generate molecular dynamics trajectories and a second model for electric field gradients that preserves rotational and translational symmetries. By incorporating thermostat-driven molecular dynamics sampling, we enable the prediction of quadrupolar coupling constants in highly disordered materials at finite temperatures. We validate our method by predicting the tetragonal-to-cubic phase transition temperature of the organic-inorganic halide perovskite MAPbI$_3$, obtaining results that closely match experimental data.