Chemical Reactions in Molten Lithium Carbonates and Hydroxides with Deep Potential Molecular Dynamics

D Kussainova and AZ Panagiotopoulos, JOURNAL OF PHYSICAL CHEMISTRY B, 129, 7282-7292 (2025).

DOI: 10.1021/acs.jpcb.5c02163

We generated a machine learning potential model to study chemical reactions in the liquid phase and under vapor-liquid equilibrium conditions for molten lithium carbonate (Li2CO3), lithium hydroxide (LiOH), and their mixtures. The model was trained on ab initio data representing pure and mixed systems at different concentrations across a broad range of temperatures and pressures. The deep potential model was then used in molecular dynamics (MD) simulations, allowing the observation of formation and dissociation reactions over time scales accessible to classical MD simulations. Simulations of pure and mixed systems revealed the decomposition of Li2CO3 into vapor-phase carbon dioxide (CO2) and LiOH into liquid-phase water (H2O), which can partially vaporize into the vapor phase. In the presence of Li2CO3 and H2O, a small amount of bicarbonate (HCO3 -) ions was observed at equilibrium. Reactions were analyzed in terms of equilibrium and Henry's law constants, with results that are thermodynamically consistent across different concentrations. We also showed that vapor and dissolved CO2 can react with liquid LiOH to produce Li2CO3 and H2O. Clear signs of immiscibility were observed when a large amount of CO2/H2O was added to LiOH at lower temperatures. This work demonstrates how machine learning potentials can be used to predict thermodynamic properties and chemical reactions in multiphase systems with the capabilities and accuracy of quantum chemical methods.

Return to Publications page