Reconstructing Nanoparticle Structural Dynamics from Molecular Simulations and Machine Learning

(a) The brief workflow of generating the neural networks potential based on the atom-centered descriptors. (b) The overall accuracy of training for both energy and forces within 5111 structures (927 clusters, 1717 bulk and 2467 surfaces), (c) The preliminary results of applying the trained MLFF to simulate the diffusion path of ad-atom between two adjacent fcc sites. It should be emphasized that our MLFF correctly reproduced the transition mechanism at DFT level as described in the literature.


  1. Yanxon H., Zagaceta D., Wood B.C., Zhu Q. (2020). Neural Networks Potential from the Bispectrum Component: A Case Study on Crystalline Silicon. (Open Access: PDF)
  2. Zagaceta D., Yanxon H., Zhu Q. (2020). Spectral Neural Network Potentials for Binary Alloys. J. Appl. Phys., 128, 045113. (Open Access: PDF)
  3. Yanxon H., Zagaceta D., Tang B., Matteson D., Zhu Q. (2020). PyXtal FF: a Python Library for Automated Force Field Generation. (Open Access: PDF)

Get new content delivered directly to your inbox.

Create your website with
Get started
%d bloggers like this: