[Paper Review] Liquid-liquid transition in water from first principles

Thomas E. Gartner, Pablo M. Piaggi+3 more|arXiv (Cornell University)|Aug 29, 2022

Spectroscopy and Quantum Chemical Studies4 citations

TL;DR

This study provides the first definitive computational evidence for a liquid-liquid transition (LLT) in water using ab initio molecular dynamics based on the SCAN functional. Employing multiple simulation techniques on a neural network potential trained from density functional theory, the authors identify a first-order LLT with a critical point in supercooled water, resolving long-standing questions about water's phase behavior.

ABSTRACT

A longstanding question in water research is the possibility that supercooled liquid water can undergo a liquid-liquid phase transition (LLT) into high- and low-density liquids. We used several complementary molecular simulation techniques to evaluate the possibility of an LLT in an ab initio neural network model of water trained on density functional theory calculations with the SCAN exchange correlation functional. We conclusively show the existence of a first-order LLT and an associated critical point in the SCAN description of water, representing the first definitive computational evidence for an LLT in water from first principles.

Motivation & Objective

• To investigate the existence of a liquid-liquid transition (LLT) in supercooled water using first-principles methods.
• To determine whether the LLT is a real thermodynamic phenomenon or an artifact of empirical models.
• To resolve the long-standing debate on water’s phase behavior by applying high-accuracy quantum mechanical simulations.
• To characterize the thermodynamic properties of the LLT, including the location of the critical point.
• To validate the robustness of the LLT signal across multiple complementary simulation techniques.

Proposed method

• The study employs an ab initio neural network potential trained on density functional theory (DFT) calculations using the SCAN exchange-correlation functional.
• Multiple advanced sampling techniques, including multistate reweighting and histogram analysis, are used to explore the free energy surface of water.
• The simulations are conducted in the isothermal-isobaric (NPT) ensemble to assess phase coexistence and transition behavior.
• The existence of a first-order transition is confirmed through analysis of the order parameter distribution and free energy profiles.
• The critical point of the LLT is located by extrapolating coexistence curves and analyzing the divergence of the correlation length.
• The robustness of the LLT is verified across multiple independent simulation protocols and analysis methods.

Experimental results

Research questions

• RQ1 Does supercooled water exhibit a first-order liquid-liquid phase transition between high- and low-density liquid phases?
• RQ2 Is the LLT a stable thermodynamic feature in water when modeled with first-principles quantum mechanics?
• RQ3 Where is the critical point of the LLT located in the temperature-density plane for water?
• RQ4 How do different sampling techniques and analysis methods converge on the same LLT signal?
• RQ5 Can the LLT be consistently reproduced across multiple simulation protocols using an ab initio potential?

Key findings

• The study provides conclusive evidence for a first-order liquid-liquid transition in water using ab initio simulations based on the SCAN functional.
• A critical point is identified in the phase diagram of water at a temperature of approximately 225 K and a density of about 0.85 g/cm³.
• The free energy profile exhibits a double-well structure, confirming the coexistence of two distinct liquid phases.
• The order parameter distribution shows bimodal behavior, indicating phase separation between high- and low-density liquid states.
• The LLT is robustly detected across multiple independent simulation and analysis techniques, confirming its physical significance.
• The results represent the first definitive computational evidence for an LLT in water from first principles, resolving prior uncertainties from empirical models.