Understanding the density maximum of water with machine-learned potentials.

Area of Science:

• Physical Chemistry
• Computational Materials Science

Background:

• The anomalous density maximum of water at 4°C has been a long-standing puzzle.
• Previous explanations involved simplistic mixtures of water structures.

Purpose of the Study:

• To elucidate the microscopic origins of water's density anomaly.
• To develop a novel computational approach for simulating water's behavior.

Main Methods:

• Trained a machine-learned (ML) interatomic potential for water using density functional theory (DFT) data.
• Performed molecular dynamics (MD) simulations with the ML potential.
• Analyzed the hydrogen-bond network and structural order at various length scales.

Main Results:

• The ML potential accurately reproduced the experimental water density anomaly and thermal expansion.
• Simulations revealed an emergent liquid structure with short-range tetrahedral order and intermediate-range collapse.
• This structural behavior explains the density maximum at 4°C.

Conclusions:

• The conventional "mixture" model for water's density anomaly is insufficient.
• Water's density anomaly arises from a complex interplay of structural orderings across different length scales.
• ML potentials offer a powerful tool for investigating complex liquid behaviors.