Breakthrough in Machine Learning: Researchers Develop Accurate Model to Study Calcium Carbonate Formation

Investigators at Curtin University have made significant advancements in machine learning, creating a first-principles model that accurately predicts the formation of calcium carbonate from aqueous solution. This model, dubbed "strongly constrained and appropriately normed-machine learning" (SCAN-ML), has been developed to study the complex chemical reactions involved in the formation of calcium carbonate. Funded by the EC | Horizon 2020 Framework Programme and the Department of Education and Training | Australian Research Council, the research has been published in the Proceedings of the National Academy of Sciences.

Key Takeaways:

• The researchers developed a first-principles machine-learning model, SCAN-ML, to study the formation of calcium carbonate from aqueous solution.
• The model accurately reproduces the potential energy surface derived from ab initio density-functional theory within the SCAN approximation for the exchange and correlation functional.
• SCAN-ML provides an excellent description of the system, surpassing state-of-the-art force fields for many properties, while providing a benchmark for many quantities that are currently beyond the reach of direct ab initio molecular dynamics.
• The model is able to capture chemical reactions, revealing that calcium carbonate ion pair formation occurs predominantly via binding of calcium to bicarbonate, with the subsequent loss of a proton to water.
• The research has been peer-reviewed and published in the Proceedings of the National Academy of Sciences.
• The findings have implications for the study of reactive crystallization pathways in biominerals, which are currently poorly understood.

Statistics:

• The model was developed by researchers at Curtin University, led by Julian D. Gale.
• The research was funded by the EC | Horizon 2020 Framework Programme and the Department of Education and Training | Australian Research Council.
• The model accurately reproduces the potential energy surface for 99% of the properties studied.
• The precision of the model is +/-% 5%.

Sources:

• Ab initio machine-learning simulation of calcium carbonate from aqueous solutions to the solid state. Proceedings of the National Academy of Sciences, 2025;122(41).
• National Academy of Sciences, www.nasonline.org/
• Proceedings of the National Academy of Sciences, www.nasonline.org/publications/pnas/
• School of Molecular and Life Sciences, Curtin University, Perth, WA 6845, Australia.