A neural network protocol for predicting molecular bond energy
- 13 Downloads
Molecular bond energy is a key parameter for analyzing the properties of chemical activity, stability and flexibility. Calculating bond energy is a challenge due to the cost of first-principles simulations and unsatisfactory prediction using empirical formula. Here we show that a neural network (NN) machine-learning method can achieve quick prediction of bond energies of organic molecules. Using atomic species and charge information as descriptors, we trained a NN protocol and applied it to predict the bond energy in a certain chemical bond that agreed with density functional theory calculations. This protocol also provided a way to evaluate the effects of different methods of atomic charge analysis on NN training. Trained to accurately estimate bond energies, this NN protocol provides a cost-effective tool for optimizing chemical reactions, accelerating molecular design, and other important applications.
Keywordsbond energy neural network density functional theory
Unable to display preview. Download preview PDF.
This work was supported by the National Key Research and Development Program of China (2017YFA0303500, 2018YFA0208603), the National Natural Science Foundation of China (21633006, 21633007, 21790350), the Fundamental Research Funds for the Central Universities (WK2340000072). The numerical calculations in this paper have been done on the supercomputing system in the Supercomputing Center of University of Science and Technology of China and the National Supercomputing Center in Changsha.