A new study shows how Deep Potential can be used to enable the prediction of a phase diagram for water over a vast range of temperatures and pressures.
Introduced in 2017, Deep Potential (DP) is an end-to-end deep neural network representation of the potential energy surface for atomic and molecular systems. DP has been attracting increasing attention as a solution to the accuracy-efficiency dilemma in computational physics and materials sciences by combining the strength of first-principles quantum mechanical calculations and large-scale molecular dynamics (MD) simulations.
The new study, The Phase Diagram of a Deep Potential Water Model, from researchers at Princeton University, the Institute of Applied Physics and Computational Mathematics and the Beijing Institute of Big Data Research, has been published as an “Editor’s Selection” in the leading physics journal Physical Review Letters.
The team uses the DP method to predict the phase diagram of water from ab initio quantum theory, from low temperature and pressure to about 2400 K and 50 GPa, representing an important milestone in the application of the DP approach.
This work is the first to apply first-principles accuracy to study the behaviour of water over a large range of temperatures and pressures using a single universal model. The phase diagram of water is so rich that in the temperature and pressure domain with T less than 400 K and P less than 50 GPa, there are ten stable phases — nine solid phases (ice Ih, II, III, V, VI, VII, VIII, XI and XV) and one liquid phase — as well as five metastable phases (ice IV, IX, XII, XIII and XIV).
Molecular dynamics (MD) simulations can provide microscopic insight into the many water phases. MD’s key component, potential energy surface (PES), can be constructed either by fitting a physically motivated force field or from quantum theory (ab initio MD aka AIMD). However, the prohibitively high computational cost of AIMD makes study of the water phase diagram possible only with empirical force fields, which struggle with the ionic phases.
Advances in machine learning have made MD simulations possible for ab initio quality. The researchers note however that no attempts have been made to describe water in a wide thermodynamic range that includes ordered and disordered ice, superionic ice, and molecular and ionic fluid phases.
In this work, the team leverages DP to make it possible to predict the phase diagram of water from ab initio quantum theory over a vast range of temperatures and pressures. The researchers also show that with further training, the approach could potentially be extended to other thermodynamic conditions, including the vapour and phases at higher temperatures and pressures. With the help of a DP Generator (DP-gen) tool, the approach completes the process of data acquisition and model training almost automatically, and requires very little ice phase and liquid water configuration data.
The data and models from the study have been published in the DP Library. These resources are helpful not only in the study of the water phase diagram, but also in understanding various strange behaviours of water, such as its phase transition behaviour and thermal conductivity under different temperatures and pressures. The data and models can also serve as a valuable reference for further research into nuclear quantum effects and more quasi-functional, wider temperature-pressure ranges.
The paper The Phase Diagram of a Deep Potential Water Model is onPhysical Review Letters and arXiv.
Author: Hecate He | Editor: Michael Sarazen, Chain Zhang
We know you don’t want to miss any news or research breakthroughs. Subscribe to our popular newsletter Synced Global AI Weekly to get weekly AI updates.
0 comments on “New Milestone for Deep Potential Application: Predicting the Phase Diagram of Water”