A high-accuracy machine-learning water model for exploring water nanocluster structures.

Area of Science:

• Physical Chemistry
• Computational Chemistry
• Materials Science

Background:

• Water's unique properties are linked to its microscopic structures, but these are difficult to identify.
• Understanding water nanocluster structures is crucial for relating macroscopic properties to microscopic details.
• Accurate and efficient simulation methods are needed for large-scale (nm, ns) water nanocluster studies.

Purpose of the Study:

• To develop and validate an efficient machine-learning (ML) water model for exploring water nanocluster structures.
• To assess the ML model's accuracy compared to quantum chemistry methods for large-scale simulations.
• To investigate water's structural evolution and phase transitions using the ML model.

Main Methods:

• Utilized a machine-learning (ML) water model to simulate water nanoclusters up to several nanometers in size.
• Employed ML model combined with velocity autocorrelation function analysis to simulate vibrational spectra.
• Applied the ML model to simulate structural evolution during crystal-liquid transitions.

Main Results:

• The ML water model efficiently predicts nanocluster structures with accuracy comparable to density functional theory.
• Simulated vibrational spectra using the ML model matched experimental results, verifying predicted atomic structures.
• The ML model precisely predicted phase transition temperatures for water clusters of varying sizes.

Conclusions:

• The ML water model offers a computationally efficient and accurate tool for studying water nanostructures.
• This model facilitates large-scale simulations of water's structural evolution and physical properties.
• The validated ML model advances research on the relationship between water's microscopic structure and macroscopic behavior.