Structure and Dynamics of Supercritical Water Determined With Neural Network Quantum Molecular Dynamics

Water subjected to very high temperatures and pressures inside the Earth's Mantle exists in its supercritical form. It exhibits extraordinary properties such as having a low dielectric constant, which stems from the breakdown of hydrogen bonds at supercritical temperatures. This makes supercritical water a non-polar solvent and the basis for many innovative technologies. In this study we investigate the hydrogen bonds, its lifetime in supercritical water and its role in controlling the dielectric constant using Neural Network Quantum Molecular Dynamics (NNQMD). Two deep neural networks are constructed. The first to produce long trajectories using neural network quantum molecular dynamics (NNQMD) and the second to predict the locations of maximally localized Wannier functions (MLWF) and calculate the dielectric constant from NNQMD trajectories.