Machine-Learning Force Field Based Computer Simulation of Rich Physical Phase Behaviour of Two-Dimensional Water/Ice in Nano-Confinement

Project: Research

View graph of relations

Description

Water and ice in nanoscale confinement are ubiquitous in geological and biological environment, and engineering settings. Understanding the properties of nano-confined water/ice is closely relevant to diverse practical phenomena such as boundary lubrication in nanofluidic devices and synthesis of antifreeze proteins for ice-growth inhibition. On scientific front, many previous simulation studies have shown that new and anomalous phase behavior can arise in nano-confined two-dimensional (2D) water/ice that are not seen in bulk water/ice.To date, classical molecular dynamics (MD) simulations have been dominantly used for predicting new phases of 2D ices, for which the classical water force fields are generally employed. Experimentally, a few 2D ices, originally predicted from MD simulations, havebeen confirmed. However, recent first-principles Born-Oppenheimer MD (BOMD) simulations showed that some newly predicted 2D ices, e.g., the zigzag monolayer ice formed at 2.5 GPa pressure and 250 K, do not obey the rules of bulk ice. This intriguingevidence indicates that the classical water force fields are inadequate to describe certain phase behavior of nano-confined water/ice even at modest 1-5 GPa pressure range, not to mention that in ultrahigh-pressure range, dissociation of water molecules and formation of superionic ice could occur, for which the classical force field cannot be applicable at all. On the other hand, although the first-principles BOMD simulations can be reliable to predict physical behaviour of 2D ices, their high computational cost limits the system size (< 100 molecules) and time of simulation (< 100 ps) for systematic investigation of the phase behaviour. To determine comprehensive physical phase behaviour of 2D water/ice, machinelearning force fields (MLFFs) at first-principles accuracy should bring to bear, while MLFF based MD simulations are needed to implement larger system sizes and much longer time ofsimulation, to compute the phase boundary crystalline phases, as well as to account for the nuclear quantum effects when combined with the path integral molecular dynamics (PIMD) simulations.In this project, we will develop MLFFs that can describe nano-confined water/ice up to 50 GPa pressure and 2000 K temperature. We will establish a state-of-the-art computational scheme consisting of the first-principles method, random structure search, and PIMD method together with the MLFF-based MD simulations, in order to explore the rich physical phase behaviour of nano-confined water/ice over broad pressure/temperature range, from negative pressure of -1 GPa to 50 GPa, and from 0 K to 2000 K. At negative pressure, we will investigate the phase behaviour of low-density ice, particularly, guest-free ice hydrate, while at ultrahigh pressure, we will search for novel forms of 2D ice with diverse physical features, particularly, 2D plastic ice and 2D ice with partial ionicity and superionicity. The nanoconfinement effect and the nuclear quantum effects will be examined via the PIMD simulations. This comprehensive computational effort will not only offer real-time dynamic evidence on the spontaneous formation of novel 2D ices, but also provide atomic-level insights into a variety of physical phase behaviour of the 2D monolayer/bilayer/trilayer icesand superionic ices, with the ultimate goal for stimulating experimental efforts to detect any of these novel 2D ices at extreme conditions.

Detail(s)

Project number9043515
Grant typeGRF
StatusActive
Effective start/end date1/07/23 → …