CMSP Atomistic Simulations Tutorial
Speaker: Cesare Malosso (SISSA)
Abstract:
The last decade has seen the rise of machine-trained potentials, exemplified by deep-neural networks or Gaussian processes, as potent instruments for atomistic simulations. These potentials boast nearly quantum mechanical precision while incurring only marginally higher costs compared to classical force fields. This advancement paves the way for broadening the scope of AIMD simulations to systematically explore bulk thermodynamic properties, dynamic behavior, and transport properties that were once beyond reach. Among the array of available machine-learning frameworks, one of the most successful has been developed by Car and his colleagues. Their DeePMD software, coupled with DPGEN software, facilitates the training and preparation of datasets to create precise models suitable for atomistic simulations. In this presentation, we'll delve into the key components of these methods along with a pedagogical overview elucidating how to practically use these powerful and increasingly essential tools.