High precision prediction of structure and thermal properties of ternary eutectic carbonates by machine learning potential for solar energy application

Abstract

Molten carbonates with high operating temperatures and excellent thermal properties are very promising phase change material for high temperature thermal energy storage. However, the structure and thermal properties of carbonates at high temperatures are lacking and difficult to measure accurately. Here, a deep potential model of ternary eutectic carbonates was developed by using first-principles molecular dynamics (FPMD) simulations as an initial dataset, and active learning using Deep Potential GENerator. The results indicate that the structure of carbonates becomes loose with increasing temperature, there is rotation of the CO₃^2- in motion, and there is a slight oscillation of the C-O bond. As the temperature increases from 700K to 1100K, the density linearly decreases from 2.01 g/cm³ to 1.86 g/cm³, and the viscosity exponentially decreases from 32.824 mPa⋅s to 3.806 mPa⋅s. The density, specific heat capacity, thermal conductivity and viscosity obtained from the simulation are in good agreement with the experimental values, where the minimum error in viscosity is only 2.45 %. This study opens a pathway to use machine learning potential to predict the melt structure and thermal properties of complex molten salt systems with high accuracy.