Thermosalient Phase Transitions from Machine Learning Interatomic Potential

Abstract

We developed an accurate machine learning interatomic potential for the thermosalient molecular crystal N-2-propylidene-4-hydroxybenzohydrazide. This crystal exhibits one of the largest mechanical responses during its thermosalient phase transition. Leveraging the speed of our developed potential, we performed Gibbs free energy calculations that successfully predict phase transitions in good agreement with experimental observations. Additionally, our model accurately captures the phenomenon of negative linear thermal expansion preceding the thermosalient phase transition. We show that the energy barrier exists at phase transition temperature and that this energy is purely elastic, elucidating the physical reasons for the thermosalient effect.