Machine learning (ML) is transforming computational materials science, offering powerful tools for predicting and exploring complex potential energy landscapes with unprecedented efficiency. Quantum-mechanical atomistic simulations often become impractical for systems with many atoms, such as interfaces, surfaces, grain boundaries, and nanoparticles. Therefore, developing accurate and flexible general-purpose interatomic potentials is crucial for these investigations. This talk highlights advancements in neural network potentials (NNPs) using the charge equilibration technique (CENT), with a focus on their application to ionic and amorphous materials. The CENT method effectively captures the energy associated with charge transfer, a key bonding mechanism in ionic materials, making it a robust tool for predicting new material structures. The presentation will cover applications of CENT potentials to a range of ionic materials, including novel predictions for sheets, bulk, and surface phases of various stoichiometric and non-stoichiometric compounds. Combining CENT potentials with the Minima Hopping method enables efficient exploration beyond known structural funnels, leading to the discovery of new low-energy phases. Recent work includes the generation and study of amorphous molybdenum disulfide (MoS?), where CENT-based NNPs revealed the structural origins of its exceptional catalytic activity in hydrogen evolution. For systems with hundreds of atoms and long-time molecular dynamics (MD) simulations, density functional theory (DFT) computations are prohibitively expensive. Therefore, the ML-based approach offers a practical alternative, enabling detailed studies of large and complex systems.


Click the link to join this webinar


https://www.skyroom.online/ch/schoolofnanoscience/weeklyseminars