NO scattering from Au(111) and LiF(001) represent contrasting benchmark systems for understanding energy transfer at gas-surface interface. We reveal an interesting mechanical origin of their difference by comparative molecular dynamics simulations with high-dimensional neural network potentials of both systems. We find that the vibrational energy of highly-vibrating NO can largely transfer to translation/rotation and further dissipate into substrate phonons, when approaching the high dissociation barrier on Au(111). This mechanical energy transfer channel is however forbidden in the purely repulsive NO/LiF(001) system or for low-vibrating NO on Au(111), where molecular vibration is barely coupled to other degrees of freedom. Phys. Rev. Lett.,126(15), 156101, 2021.
We propose an embedded atom neural network approach with simple piecewise switching function based descriptors, resulting in a favorable linear scaling with the number of neighbor atoms. Numerical examples validate that this piecewise machine learning model can be over an order of magnitude faster than various popular machine learned potentials with comparable accuracy for both metallic and covalent materials, approaching the speed of the fastest embedded atom method (i.e. several μs/atom per CPU core). Phys. Chem. Chem. Phys. 23, 1815, 2021(Front Cover).
High-dimensional machine learning representations of potential energy surface and electronic friction tensor have allowed us to perform a comprehensive quantitative analysis of the performance of nonadiabatic molecular dynamics in describing vibrational state-to-state scattering of NO on Au(111), revealing the applicability and limitation of electronic friction theory. JACS Au, 1(2), 164-173, 2021.
Machine learned potential energy surfaces enabled dynamical calculations with unprecedented accuracy and extraordinary efficiency, leading to new insights and discoveries. J. Phys. Chem. Lett. 11(13), 5120, 2020.
Efficient intramolecular vibrational energy redistribution between stretching modes in H2O scattering on Cu(111) is revealed by state-to-state quantum scattering calculations. Phys. Rev. Lett., 123(10), 106001, 2019
Our group at USTC aims to understand reaction mechanisms and dynamics of elementary chemical reactions at atomic level, from a theoretical perspective. We have been studying important elementary reactions in gas phase, photochemistry, and particularly reactions of molecules at interfaces that are relevant to heterogeneous catalysis. We focus especially on the development of efficient and accurate machine learning methods for fitting first-principles potential energy surfaces (and also other molecular properties), which stand at the heart of chemical dynamics and spectroscopic simulations of complex chemical systems. We also develop quantum, quasi-classical, and approximate path integral based methods for studying reaction dynamics of molecular scattering at metal surfaces. With these fundamental efforts, we hope to gain in-depth insights into various physiochemical processes at gas-surface interfaces and offer valuable foundations for catalyst design.
Influence of supercell size on Gas-Surface Scattering: A case study of CO scattering from Au(111) , is published on Chem. Phys.. Congratulations!
Accurate Machine Learning Prediction of Protein Circular Dichroism Spectra with Embedded Density Descriptors, is published on JACS Au.. Congratulations!
Infrared Activities of Adsorbed Species on Metal Surfaces: The Puzzle of Adsorbed Methyl (CH3）, is published on J. Phys. Chem. Lett.. Congratulations!