Signatures of a liquid–liquid transition in an ab initio deep neural network model for water

Abstract

The possible existence of a metastable liquid–liquid transition (LLT) and a corresponding liquid–liquid critical point (LLCP) in supercooled liquid water remains a topic of much debate. An LLT has been rigorously proved in three empirically parametrized molecular models of water, and evidence consistent with an LLT has been reported for several other such models. In contrast, experimental proof of this phenomenon has been elusive due to rapid ice nucleation under deeply supercooled conditions. In this work, we combined density functional theory (DFT), machine learning, and molecular simulations to shed additional light on the possible existence of an LLT in water. We trained a deep neural network (DNN) model to represent the ab initio potential energy surface of water from DFT calculations using the Strongly Constrained and Appropriately Normed (SCAN) functional. We then used advanced sampling simulations in the multithermal–multibaric ensemble to efficiently explore the thermophysical properties of the DNN model. The simulation results are consistent with the existence of an LLCP, although they do not constitute a rigorous proof thereof. We fit the simulation data to a two-state equation of state to provide an estimate of the LLCP’s location. These combined results—obtained from a purely first-principles approach with no empirical parameters—are strongly suggestive of the existence of an LLT, bolstering the hypothesis that water can separate into two distinct liquid forms.

Machine learning potentials for complex aqueous systems made simple

“…3 D , Inset ). Such atomic force errors are similar or even smaller than those reported for other developed MLPs, e.g., for pure water ( 8 – 10 , 44 ). At the same time, the error estimate obtained for the AIMD configurations features essentially the same distance resolved profile, which reveals that our C-NNP simulations are able to conserve their predictive power, while substantially extending both time and length scales of the simulations.…”

“…In recent years, machine learning potentials (MLPs) have become a promising alternative, bypassing expensive ab initio calculations and extending length and time scales in molecular simulations ( 3 – 7 ). This is exemplified in studies on the understanding of the unique properties of water ( 8 – 10 ), structural and electronic transitions in disordered silicon ( 11 ), and phase transitions of hybrid perovskites ( 12 ) to name but a few. The success of MLPs is grounded in a number of distinct approaches that have been introduced over the years, notably, using artificial neural networks ( 13 – 17 ) or kernel-based methods ( 18 – 23 ).…”

Assessing the Accuracy of the SCAN Functional for Water through a Many-Body Analysis of the Adiabatic Connection Formula

“…studied the isotope effects in liquid H 2 O and D 2 O using a deep neural network potential (NNP) trained on the SCAN functional. 35 A similar NNP was used to investigate the properties of supercooled water, 36 as well as the equilibrium between liquid water and ice Ih and Ic, 37 and the ice Ih/XI transition. 38 Building upon these previous studies, particularly the many-body analysis reported in ref 32, we provide here an assessment of the accuracy of the SCAN functional in predicting the properties of water.…”

Manifestations of metastable criticality in the long-range structure of model water glasses

“…As a result, much work aimed at probing the phase behavior of supercooled water has been done via simulation 13 . Simulations of several classes of water models of varying accuracy and complexity have yielded evidence consistent with an LLT [24][25][26][27] . Some researchers have suggested the first-order-like transition between LDA and HDA provides evidence of the LLT in water, with LDA considered to be the structurally arrested analog of LDL and HDA the corresponding analog of HDL 5,28 .…”