End-to-end symmetry preserving potential energy models for molecules and materials
Machine learning models are changing the paradigm of molecular modeling. Of particular interest is the inter-atomic potential energy surface (PES). In this talk, I would introduce deep learning-based PES models that we recently developed [1-3]. I would focus on how to construct a PES model that is end-to-end, accurate, and scalable, and preserves all the natural symmetries. Further, I would show the performance of these models when describing finite and extended systems including organic molecules, metals, semiconductors, and insulators. The corresponding open-source software  is efficiently parallelized and interfaced with popular machine learning and molecular dynamics packages.
1. L. Zhang, et al, arxiv: 1805.09003
2. L. Zhang, et al, Phys. Rev. Lett. 120, 143001
3. J. Han, et al, Comm. Comp. Phys. 23.3 (2018): 629-639.
4. H. Wang, et al. Comp. Phys. Comm., 2018: 0010-4655 (https://github.com/deepmodeling/deepmd-kit)
Contact: Lei Wang, 9853