We develop a machine-learning model based on ab initio potentials to study essential thermodynamic properties of aqueous electrolyte solutions such as activity coefficients and solubilities.
Aqueous electrolyte solutions have wide-ranging applications from biological systems to carbon sequestration and energy production.
Molecular modeling is crucial to understand properties of such systems. Classical molecular dynamics simulations using empirical forcefields have been widely employed to model electrolyte solutions, but these models result in significant tradeoffs between accuracy and computational efficiency.
Recent advances in machine-learning techniques, however, have made possible the development of models based on ab initio potentials that can improve the accuracy of the empirical forcefields while retaining their computational efficiency.
In the current project, we aim to develop a deep learning-based potential to model the behavior of aqueous NaCl solutions. We are interested in studying activity coefficients of electrolyte solutions and solubility of ions in water for broad ranges of temperatures and concentrations.
For this purpose, we seek to establish the thermodynamically viable path to calculate activity coefficients and solubilities, using free energy calculations and machine-learning models. Our studies may improve previous molecular simulations that could not accurately predict thermodynamic properties and their temperature dependence.
This project is part of the ERC advanced grant project “New Paradigm in electrolyte Thermodynamics” with Prof. Georgios Kontogeorgis as PI.
Main supervisor:
Professor Athanassios Z. Panagiotopoulos