search

sudarshanv01 / coeffnet

Equivariant network to predict activation barriers and molecular orbitals through coefficients of molecular orbitals.
MIT License
7 stars 0 forks source link

Projecting fixed dimensional minimal basis regardless of the full basis set used in DFT calculations #18

Open hpatel1567 opened 1 year ago

hpatel1567 commented 1 year ago

Determining a methodology to project the same size matrix regardless of the basis set used in DFT calculations. A way to accurately/scientifically aggregate full basis set functionals into the minimal basis set (which determines the matrix size) is at the core of the problem. Given a molecule computed at different basis sets, at minimum HOMO/LUMO coefficient ordering should be the same and ideally coefficient should be similar numerically?

sudarshanv01 commented 1 year ago

It is true that currently we are defining the minimal basis in a very ad-hoc way (i.e. just removing the d or higher orbitals). I imagine the easiest way to do it (for small organic molecules) would be

  1. Remove the d-orbitals
  2. Remove all s and p-orbitals except those that are "required". How I would define "required" would be that they are formally the highest s and p orbitals expected for the molecule. For example, for carbon you would take 2s and 2p. For Cl it would be 3s and 3p and so on. So basically each atom would have exactly 1s and 3p functions. The only catch is that it would be somewhat tedious to construct such a matrix.