Molecular dynamics simulation and machine learning for predicting hydrogen solubility in water: Effects of temperature, pressure, finite system size and choice of molecular force fields

Abstract

Understanding hydrogen solubility in groundwater is critical for predicting the long-term performance of underground hydrogen storage in geological formations. Nevertheless, the effect of finite-size, force fields on hydrogen solubility were not investigated in previous molecular dynamics(MD) studies and available experimental data were limited to a narrow temperature range. This paper proposed a combined MD and machine learning(ML) methods to study hydrogen solubility over a wide temperature range of 273–433 K. Comparison of different models revealed that TIP4P/2005 water model and hydrogen model of Yang & Zhong (J.Phys.Chem.B,109(2005),11862–11864) was successful in reproducing the experimental results. A minimum value of hydrogen solubility presents at 330 K. Finite-size effects were quantified and suitable hydrogen and water models were identified. Eight ML models were developed based on density, temperature, pressure, and excess chemical potential. They justified that the pressure is the most influential variable, whereas temperature has a non-linear effect on hydrogen solubility.