Machine learning assisted investigation of the barocaloric performance in ammonium iodide

Abstract

Using the ab initio-based training database, we trained the potential function for ammonium iodide (NH4I) based on a deep neural network-based model. On the basis of this potential function, we simulated the temperature-driven β ⇒ α-phase transition of NH4I with isobaric isothermal ensemble via molecular dynamics simulations, the results of which are in good agreement with recent experimental results. As it increases near the phase transition temperature, a quarter of ionic bonds of NH4+-I− break so that NH4+ starts to rotate randomly in a disorderly manner, being able to store thermal energy without a temperature rise. It is found that NH4I possesses a giant isothermal entropy change (∼93 J K−1 kg−1) and adiabatic temperature (∼27 K) at low driving pressure (∼10 MPa). In addition, through partial substitution of I by Br in NH4I, it is found that the thermal conductivity can be remarkably improved, ascribed to the enhancement of lifetime of low frequency phonons contributed by bromine and iodine. The present work provides a method and important guidance for the future exploration and design of barocaloric material for practical applications.