Liquid-liquid transition and ice crystallization in a machine-learned coarse-grained water model

Area of Science:

• Physical Chemistry
• Computational Materials Science
• Statistical Mechanics

Background:

• Experimental evidence suggests a liquid-liquid transition (LLT) in high-pressure supercooled water.
• Fast crystallization hinders experimental identification of the LLT line.
• All-atom (AA) models show LLT but are computationally expensive; coarse-grained (CG) models are efficient but lack LLT.

Purpose of the Study:

• To demonstrate that a coarse-grained (CG) machine-learned water model (ML-BOP) exhibits a liquid-liquid transition (LLT).
• To investigate the relationship between the LLT and ice crystallization in supercooled water.
• To validate ML-BOP's ability to replicate experimental observations of water's liquid-phase behavior.

Main Methods:

• Utilized the coarse-grained machine-learned water model, ML-BOP.
• Performed cooling simulations of supercooled water under high pressure.
• Analyzed the LLT line, critical point, and ice crystallization behavior.

Main Results:

• ML-BOP exhibits an LLT ending in a critical point (Pc = 170 ± 10 MPa, Tc = 181 ± 3 K).
• The LLT line in ML-BOP closely matches that of the TIP4P/2005 model.
• Ice crystallization is fastest at the LLT and its supercritical continuation, suggesting a mechanistic link.

Conclusions:

• ML-BOP successfully models the LLT in supercooled water, bridging the gap between AA and CG models.
• The study supports a mechanistic relationship between water's structural transformation and ice formation.
• ML-BOP replicates experimental competition between low-density liquid (LDL) and ice formation, even when crystallization precedes domain coarsening.