Phase Equilibrium of Water with Hexagonal and Cubic Ice Using the SCAN Functional

Area of Science:

• Computational chemistry
• Materials science
• Machine learning in physics

Background:

• Machine learning models are increasingly used for accurate physicochemical simulations.
• Density Functional Theory (DFT) is a common method, but computationally expensive for complex phenomena like ice nucleation.
• Investigating phase equilibrium and nucleation requires advanced simulation techniques.

Purpose of the Study:

• To investigate the phase equilibrium of water, hexagonal ice (Ih), and cubic ice (Ic) using machine learning and DFT.
• To calculate properties related to ice nucleation that are beyond the reach of direct DFT simulations.
• To assess the accuracy of a deep neural network model trained on SCAN DFT data.

Main Methods:

• Utilized a deep neural network machine learning model trained on SCAN DFT data.
• Employed enhanced sampling simulations driven by the machine learning model.
• Applied a reweighting procedure to compute properties for the SCAN functional.
• Performed direct Density Functional Theory (DFT) calculations.

Main Results:

• Calculated melting temperatures, nucleation driving forces, heats of fusion, and densities for ice polymorphs.
• Achieved correct qualitative predictions for all investigated properties.
• Demonstrated quantitative agreement with experimental data, surpassing some semiempirical potentials.
• Confirmed that SCAN correctly predicts hexagonal ice (Ih) is more stable than cubic ice (Ic).

Conclusions:

• Machine learning models, trained on DFT data, offer a powerful approach for simulating complex physicochemical phenomena.
• This method enables the calculation of properties crucial for understanding ice nucleation.
• The SCAN functional, when used with machine learning, provides accurate predictions for water-ice phase equilibria.