Abstract (eng)
Due to its role in many natural processes as well as its anomalous properties, water remains the center of focus for a number of researches. The nuclei of hydrogen atoms are formed by a single proton, whose low mass leads to water being prominently affected by nuclear quantum effects. To include these effects in first principles simulations, a computational method based on the quantum mechanical Feynman path integral formulation is used. Known as path integral molecular dynamics, each particle is treated as a ring polymer to approximate their quantum nature, which further increases the already high computational cost. However, through the use of neural network methods, the speed of calculations can be improved to make the size and time scale needed for meaningful evaluations feasible while retaining high accuracy. In this work, a neural network potential is trained based on reference energies and forces of the density functional theory potential energy surface of water in the Revised Perdew–Burke–Ernzerhof approximation corrected for dispersion. To gauge the impact of nuclear quantum effects on the water/vapor interface, that neural network potential is used both with and without employing path integral molecular dynamics to simulate an interfacial water system. Structural and thermodynamic properties are calculated along the liquid/vapor coexistence line to garner insight of the performance of the combination of path integral molecular dynamics and neural network potentials. The results show the expected shift of the critical point upon inclusion of the nuclear quantum effects, as well as changes in the preferential structure at the water’s surface. The computational results presented have been achieved in part using the Vienna Scientific Cluster (VSC).