Neural-Network-Based Selective Configuration Interaction Approach to Molecular Electronic Structure.

Area of Science:

• Quantum Chemistry
• Computational Physics
• Materials Science

Background:

• Accurate quantum chemistry calculations are crucial for understanding molecular properties.
• Traditional configuration interaction (CI) methods struggle with computational cost for large systems.
• Neural network potentials offer a promising avenue for accelerating quantum simulations.

Purpose of the Study:

• To develop and implement a novel selective neural-network configuration interaction (NNCI) method.
• To evaluate the performance of NNCI for molecular calculations.
• To demonstrate the efficiency and applicability of NNCI for larger and extended systems.

Main Methods:

• Combined Hartree-Fock with a neural-network-supported quantum-cluster solver.
• Implemented selective neural-network configuration interaction (NNCI) with flexible basis sets and boundary conditions.
• Evaluated NNCI performance on small molecules, comparing results with full CI calculations.

Main Results:

• NNCI accurately reproduced correlation energy for N2 using significantly fewer determinants (4x10^5) compared to full CI (10^10).
• Increasing the number of included orbitals showed a clear advantage over approaching full CI with fewer orbitals.
• The method demonstrated high efficiency and scalability for molecular calculations.

Conclusions:

• NNCI offers a computationally efficient and accurate approach for electronic structure calculations.
• The method's implementation in condensed matter simulation software broadens the scope of CI calculations.
• NNCI is applicable to a wider range of problems, including extended systems via embedding.