Ab initio machine-learning simulation of calcium carbonate from aqueous solutions to the solid state

Area of Science:

• Computational Chemistry
• Materials Science
• Biomineralization

Background:

• Understanding calcium carbonate formation is crucial for biominerals.
• Accurate simulation of aqueous systems and interfaces remains challenging.
• First-principles methods offer high accuracy but are computationally expensive.

Purpose of the Study:

• To develop a machine-learning model for accurate simulation of calcium carbonate formation.
• To enable first-principles molecular dynamics studies of reactive crystallization.
• To investigate the ion-pair formation mechanism in calcium carbonate precipitation.

Main Methods:

• Developed a strongly constrained and appropriately normed-machine learning (SCAN-ML) model.
• Used density-functional theory (DFT) with SCAN approximation for training.
• Performed molecular dynamics simulations of ions, solid phases, and the calcite/water interface.

Main Results:

• SCAN-ML accurately reproduces the potential energy surface from ab initio DFT.
• The model surpasses state-of-the-art force fields in describing system properties.
• Identified calcium carbonate ion pair formation via calcium-bicarbonate binding and proton transfer.

Conclusions:

• SCAN-ML provides a benchmark for simulating reactive crystallization pathways.
• The model enables ab initio studies of poorly understood biominerals.
• This approach advances the computational study of mineral formation processes.