Electron Density-Based Machine Learning for Accelerating Quantum Calculations

Joshua Lansford, Shodor Education Foundation, Inc.

Usage Details

The electronic density distribution completely specifies a chemical system’s state and can be calculated using density functional theory (DFT). We are developing an atom-centered machine learning (ML) algorithm, trained on electron density, which can generate catalytic reaction mechanisms and kinetic models at reduced computational cost. The ML-based model will accelerate the three most computationally intensive components of the reaction energy profile: transition states, global minima, and entropy. This methodology will be the first to combine structural and density data to enhance ML convergence. Due to the atom-centered nature, the representation of the molecular systems will be invariant to rotations, translations, and reordering of atoms. Because we will use distances, partial charges, and atomic dipoles, the model will be generalizable to any size system.