Scalable Quantum Monte Carlo with Direct-Product Trial Wave Functions

Area of Science:

• Computational Chemistry
• Quantum Monte Carlo Methods
• Electronic Structure Theory

Background:

• Phaseless auxiliary-field quantum Monte Carlo (AFQMC) methods face high computational demands, especially for molecules with strong electron correlations.
• Current multi-Slater determinant (MSD) trials can be computationally intensive, limiting their application to complex molecular systems.

Purpose of the Study:

• To propose and evaluate the use of direct-product wave functions as trials for MSD-AFQMC.
• To reduce the computational overhead associated with MSD-AFQMC by exploiting the compactness of direct-product multi-Slater determinant (DP-MSD) trials.
• To assess the accuracy and efficiency of DP-MSD trials for molecular systems with varying electronic structure complexity.

Main Methods:

• Developed a novel approach using direct-product wave functions as trials for MSD-AFQMC.
• Defined 'decomposable active space' as a condition where the active space can be divided into noncoupling subspaces.
• Employed localized-active space self-consistent field (LASCF) wave functions as DP-MSD trials and tested on various molecular systems.

Main Results:

• DP-MSD trials significantly reduced computational costs, up to 36 times for C2H6N4, when active spaces are decomposable.
• For weakly coupled systems like the benzene dimer, DP-MSD trials decreased computational cost while maintaining accuracy compared to complete active space trials.
• Accuracy decreased for systems with strong subspace coupling compared to complete active space approaches.

Conclusions:

• Direct-product multi-Slater determinant trials offer a computationally efficient alternative for MSD-AFQMC in specific molecular systems.
• The method is particularly beneficial for systems with noncoupling to weakly coupling subspaces requiring local multireference treatments.
• Careful consideration of subspace coupling is necessary to ensure accuracy.