Partitioning the electronic wave function using deep variational Monte Carlo.

Area of Science:

• Quantum Chemistry
• Computational Chemistry
• Materials Science

Background:

• Accurate electronic wave function (WF) representation is crucial for predicting molecular properties.
• Current methods often struggle with scaling to larger and more complex systems.
• Decomposition of WFs into physically meaningful components can offer computational advantages.

Purpose of the Study:

• To introduce a novel wave function partitioning method.
• To integrate deep-learning variational Monte Carlo with generalized product function ansätze.
• To enable the separation of electronic WFs into partial components.

Main Methods:

• Developed a deep-learning variational Monte Carlo approach.
• Utilized ansätze based on generalized product functions for wave function partitioning.
• Applied the method to small molecules (Li to Mg atoms).

Main Results:

• Successfully separated electronic WFs into partial components (e.g., core and valence).
• Accurately reproduced key chemical properties like dissociation curves and ionization energies.
• Demonstrated that core WFs are transferable and reusable across different molecular geometries.

Conclusions:

• Core electrons can be effectively decoupled from valence electrons.
• The proposed framework offers potential for efficient computation and study of larger systems.
• This work may facilitate the ab initio development of effective core potentials.