High-accuracy electronic structures of condensed phases at Hartree-Fock cost with machine-learnt, correlated 2-body reduced density matrices

The field of machine learning has opened the door to significant scientific advances. In particular, it has opened the possibility to predict molecular electronic structures [1,2,3,4,5] at a very reduced computational cost. However, handling the complexity of condensed-phase systems has remained challenging. Leveraging machine learning models for the correlated part of the electronic two-body reduced density matrix (2-cRDM), total electronic energies, atomic forces, scattering factors, and critical molecular properties are predicted at coupled cluster level with an algorithm scaling like the Hartree-Fock method. Additionally, a 1-RDM embedding method coupled with a novel 1- and 2-RDM purification scheme delivers accurate condensed-phase electronic structures of water clusters, droplets, and solvated species.

[1] Yuanming Bai, Leslie Vogt-Maranto, Mark E. Tuckerman, and William J. Glover. Machine learning the Hohenberg-Kohn map for molecular excited states. Nat. Commun., 13(1):7044, November 2022.

[2] Stefano Battaglia. Chapter 25 - Machine learning wavefunction. In Pavlo O. Dral, editor, Quantum Chemistry in the Age of Machine Learning, pages 577–616. Elsevier, January 2023.

[3] Felix Brockherde, Leslie Vogt, Li Li, Mark E. Tuckerman, Kieron Burke, and Klaus-Robert Müller. Bypassing the Kohn-Sham equations with machine learning. Nat. Commun., 8(1):872, October 2017.

[4] Jan Hermann, James Spencer, Kenny Choo, Antonio Mezzacapo, W. M. C. Foulkes, David Pfau, Giuseppe Carleo, and Frank No´e. Ab initio quantum chemistry with neural-network wavefunctions. Nat. Rev. Chem., 7(10):692–709, August 2023.

[5] Frank Noé, Alexandre Tkatchenko, Klaus-Robert Müller, and Cecilia Clementi. Machine Learning for Molecular Simulation. Annu. Rev. Phys. Chem., 71(1):361–390, April 2020.