Research Interest 2:
Develop new machine learning package and algorithms to accelerate ab initio calculations and catalyst design
As part of the long program “Complex High-Dimensional Energy Landscapes” hosted by the US NSF Institute for Pure & Applied Mathematics (IPAM, Los Angeles), we started developing a machine learning (ML) framework following the Behler-Parrinello artificial neural network principle. Our goal was to build a framework for the fitting of local and global potential energy surfaces (PES) to accelerate ab initio atomistic simulations. A number of atomistic systems (e.g., metal clusters, alloys, and surface-adsorbate systems) were tested using our framework and showed high fitting accuracy for both energy and forces. The current (PyTorch) version of this package is recently online: https://pyamff.gitlab.io/pyamff/index.html. Meanwhile, to accelerate the design of high-performance catalytic materials, we developed a number of machine learning models with rigorous feature analysis and modeling processes, which help to successfully predict the binding energies on a large number of bi-metallic alloy surfaces across the periodic table. An example of these works can be found in Ref. 1.
Representative Publications:
(1) H Li, S. Xu, M. Wang, Z. Chen, F. Ji, K. Cheng, Z. Gao, Z. Ding, and W. Yang. "Computational Design of (100) Alloy Surfaces for the Hydrogen Evolution Reaction", Journal of Materials Chemistry A, 2020, 8, 17987 (Cover Article & Editor’s Choice: Machine Learning for Materials Innovation).
Develop new machine learning package and algorithms to accelerate ab initio calculations and catalyst design
As part of the long program “Complex High-Dimensional Energy Landscapes” hosted by the US NSF Institute for Pure & Applied Mathematics (IPAM, Los Angeles), we started developing a machine learning (ML) framework following the Behler-Parrinello artificial neural network principle. Our goal was to build a framework for the fitting of local and global potential energy surfaces (PES) to accelerate ab initio atomistic simulations. A number of atomistic systems (e.g., metal clusters, alloys, and surface-adsorbate systems) were tested using our framework and showed high fitting accuracy for both energy and forces. The current (PyTorch) version of this package is recently online: https://pyamff.gitlab.io/pyamff/index.html. Meanwhile, to accelerate the design of high-performance catalytic materials, we developed a number of machine learning models with rigorous feature analysis and modeling processes, which help to successfully predict the binding energies on a large number of bi-metallic alloy surfaces across the periodic table. An example of these works can be found in Ref. 1.
Representative Publications:
(1) H Li, S. Xu, M. Wang, Z. Chen, F. Ji, K. Cheng, Z. Gao, Z. Ding, and W. Yang. "Computational Design of (100) Alloy Surfaces for the Hydrogen Evolution Reaction", Journal of Materials Chemistry A, 2020, 8, 17987 (Cover Article & Editor’s Choice: Machine Learning for Materials Innovation).