Mean-field density matrix decompositions

Area of Science:

• Quantum Chemistry
• Computational Chemistry
• Materials Science

Background:

• Mean-field theories like Hartree-Fock and Density Functional Theory are fundamental in quantum chemistry.
• Accurate prediction of molecular properties relies on robust theoretical frameworks.
• Understanding electron distribution is key to rationalizing chemical phenomena.

Purpose of the Study:

• To introduce novel, robust decompositions for mean-field Hartree-Fock and Kohn-Sham density functional theory.
• To enable partitioning of one-electron reduced density matrices into atomic and bond contributions.
• To enhance machine learning models in quantum chemistry through improved data granularity.

Main Methods:

• Utilizing localized molecular orbitals for robust decompositions.
• Implementing physically sound charge population protocols.
• Partitioning one-electron reduced density matrices into bond-wise or atomic contributions.
• Comparing new decompositions against literature alternatives for molecular energies and dipole moments.

Main Results:

• The new decompositions are lossless, providing accurate partitioning of electronic properties.
• Demonstrated ability to expose and amplify compositional features for machine learning.
• Showcased improvements in granularity of data for quantum chemistry machine learning.
• Preliminary results suggest enhanced structure-property inferences.

Conclusions:

• Decomposed mean-field theory offers a powerful interpretative tool for electronic phenomena.
• The developed methods significantly improve data granularity for machine-learned quantum chemistry.
• Future work can refine structure-property inferences by leveraging increased dataset complexity.