Search papers, labs, and topics across Lattice.
A machine-learned interatomic potential for water was trained using deep neural networks and advanced density functional theory data. Molecular dynamics simulations using this potential accurately reproduce the experimental density anomaly and thermal expansion coefficient of water. Structural analysis reveals the density anomaly arises from a liquid structure with tetrahedral coordination at short range but collapse at intermediate range, suggesting a more nuanced mechanism than previously understood.
Water's density anomaly isn't just about mixed structures; it's a delicate balance of short-range order and intermediate-range collapse, revealed by a new machine-learned potential.
After melting, at ambient pressure, the density of water continues to increase with temperature until it reaches a maximum around 4 掳C. For nearly a century, this phenomenon has been qualitatively attributed to a mixture of ordered and disordered structures. Herein, we employ a deep neural network to train a machine learned (ML) interatomic potential for water using electronic structure data from advanced density functional theory. Notably, molecular dynamics simulations with the ML potential reproduce both the experimental water density anomaly and the thermal expansion coefficient. Detailed structural analysis of the computed hydrogen-bond network reveals that the density anomaly arises from an emergent liquid structure that retains nearly ideal tetrahedral coordination at short range but collapses at intermediate range. Our findings point to a more delicate mechanism causing the density maximum than the conventional picture, emphasizing the collective roles of structural orderings at different length scales.
