Orbital-free density functional theory powered by machine learning

Hohenberg and Kohn have proven that the exact electronic ground state energy and electron density can be obtained by minimizing an (unfortunately unknown) energy functional with respect to the density. Using atomistic machine learning, we have obtained for the first time an approximate density functional that, when applied to the organic molecules in QM9, yields energies with chemical accuracy relative to the Kohn–Sham reference while also converging to meaningful electron densities. Augmenting the training data with densities obtained from perturbed potentials proved key to these advances. Going beyond the results published in [1] and available open source [2], this talk will show latest results on non-equilibrium geometries and for open-shell systems.

Joint work with Roman Remme, Tobias Kaczun, Tim Ebert, Christof A. Gehrig, Dominik Geng, Gerrit Gerhartz, Marc K. Ickler, Manuel V. Klockow, Peter Lippmann, Johannes S. Schmidt, Simon Wagner, Andreas Dreuw.  

[1] Remme et al. "Stable and accurate orbital-free density functional theory powered by machine learning." Journal of the American Chemical Society (2025). https://pubs.acs.org/doi/abs/10.1021/jacs.5c06219

[2] https://github.com/sciai-lab/structures25