SEARCH

SEARCH BY CITATION

Keywords:

  • bond dissociation energy;
  • density functional calculations;
  • heat of formation;
  • neural networks;
  • thermochemistry

Abstract

Previously, we have put forward the X1 method that combines B3LYP with neural network correction for an accurate yet efficient prediction of thermochemistry. Without paying additional computational cost, X1 reduces B3LYP’s mean absolute deviation (MAD) for a set of 92 bond dissociation energies (BDEs) from 5.5 to 2.4 kcal mol−1. In this work, we extend X1 and propose the X1s method by including the spin change from molecules to atoms during atomization as an additional descriptor. X1s further reduces the MAD for BDEs to 1.4 kcal mol−1, thus showing substantial improvement.