A first-principles machine-learning force field for heterogeneous ice nucleation on microcline feldspar

Area of Science:

• Atmospheric Chemistry
• Materials Science
• Computational Physics

Background:

• Ice formation in the atmosphere influences precipitation and climate.
• Feldspar minerals are efficient ice nuclei, but the formation mechanism is unclear.
• Understanding ice nucleation on mineral surfaces is key to atmospheric science.

Purpose of the Study:

• To develop a first-principles machine-learning model for studying ice nucleation on microcline feldspar.
• To accurately simulate ice formation processes at the atomic level.
• To investigate the role of feldspar surfaces in atmospheric ice nucleation.

Main Methods:

• Developed a machine-learning potential energy surface model.
• Trained the model using density-functional theory (DFT) with the SCAN functional.
• Applied the model to simulate water and ice interactions with microcline feldspar surfaces.

Main Results:

• The machine-learning model accurately reproduces DFT energies and forces.
• Identified preferred microcline surface terminations in a vacuum.
• Observed water density correlations extending ~10 Å from microcline surfaces.
• Demonstrated high accuracy in simulating ice formation on large microcline systems.

Conclusions:

• The developed machine-learning force field is a reliable tool for studying ice nucleation on feldspar.
• Further research will focus on calculating nucleation barriers and rates.
• This work advances the understanding of feldspar's role in atmospheric ice formation and climate.