Artificial intelligence applied to electronic and dynamic simulations of nanomaterials

• Grant number: 24/18416-3
• Support Opportunities: Scholarships in Brazil - Program to Stimulate Scientific Vocations
• Start date: January 08, 2025
• End date: February 15, 2025
• Field of knowledge: Physical Sciences and Mathematics - Physics - Condensed Matter Physics
• Principal Investigator: Gabriel Ravanhani Schleder
• Grantee: Muanangani Manuel Garcia Lutundisa
• Host Institution: Centro Nacional de Pesquisa em Energia e Materiais (CNPEM). Ministério da Ciência, Tecnologia e Inovação (Brasil). Campinas , SP, Brazil

Abstract

Machine Learning Interatomic Potentials (MLIPs) combine the accuracy of quantum methods with the computational efficiency of classical force fields, enabling simulations of atoms, molecules, biosystems, solids, surfaces, and nanomaterials. Recently, advanced MLIPs that utilize equivariant representations and deep graph neural networks, known as "universal models," have been prominent. We will evaluate the universality of available UIPs, such as MACE and CHGNet, in generalization tasks, validating their efficiency for fine-tuning specialized models and expanding the material space coverage in the training dataset. We will apply these potentials to molecular dynamics simulations of interest to LNNano, including dynamic structural properties of nanomaterials. Examples include simulations of experiments at the National Nanotechnology Laboratory, such as the breakup processes of 2D material nanoflakes, twisted 2D systems, and oxide nanoclusters, visualized by high-resolution transmission electron microscopy (HRTEM) in situ. Our aim is methodological development and validation, in addition to applied results to the mentioned physical systems. (AU)