AI for Molecular Simulations: From Irradiated Metals to Interstellar Ices
Samuel Del Fré, a postdoctoral researcher at PHLAM and former postdoctoral researcher at UMET will speak at a conference taking place on October 14 at 10:30 AM in the Chevreul Institute auditorium.
During this conference, entitled “AI for Molecular Simulations: From Irradiated Metals to Interstellar Ices” Samuel Del Fré will present, in English, his work on machine learning and deep learning applied to computational chemistry topics:
Machine learning (ML) and deep learning (DL) are reshaping computational chemistry, ranging for instance, from large-scale data analysis to the development of accurate potential energy surfaces (PES). In this presentation, we highlight two examples.
- The first addresses irradiated metals (Ni, FeNiCr, Zr), where an unsupervised ML pipeline (SOAP + autoencoder + UMAP + HDBSCAN) reveals radiation-induced defects in displacement cascades without requiring prior knowledge of defect structures.
- The second explores astrochemistry: a DL-based PES for CO interstellar ices reproduces vibrational energy redistribution in the photodesorption mechanism with near-ab initio accuracy but at a fraction of the cost, enabling near-perfect agreement with experiments.
These cases exemplify how ML broadens the scale and scope of molecular simulations.