AI and the World's Hard Problems · Chemistry & materials · arXiv · 2026-10-20
Machine Learning-Assisted Discovery of Carbon-Related Defect Formation Mechanisms in Diamond
The authors fine-tuned a universal machine learning interatomic potential on a small dataset from density functional theory to study carbon-related defects in diamond. They obtained a migration barrier of 1.84 eV for a mono-interstitial and found that two interstitials form a di-interstitial over a lower barrier. The model also identified an additional di-interstitial pathway and a stepwise divacancy dissociation mechanism beyond the training data.