Improved Optimization for the Cluster Jastrow Antisymmetric Geminal Power and Tests on Triple-Bond Dissociations

TL;DR

This paper introduces a specialized variational Monte Carlo optimization method for the cluster Jastrow antisymmetric geminal power ansatz, significantly improving efficiency and accuracy in modeling complex molecular bonds.

Contribution

It develops a lower-cost, more effective optimization technique for the cluster Jastrow antisymmetric geminal power ansatz, outperforming previous methods in accuracy and computational efficiency.

Findings

Achieved superior accuracy in triple-bond dissociation tests.

Demonstrated improved optimization performance over quasi-Newton methods.

Validated on N₂ and [ScO]⁺, outperforming traditional theories.

Abstract

We present a novel specialization of the variational Monte Carlo linear method for the optimization of the recently introduced cluster Jastrow antisymmetric geminal power ansatz, achieving a lower-order polynomial cost scaling than would be possible with a naive application of the linear method and greatly improving optimization performance relative to the previously employed quasi-Newton approach. We test the methodology on highly multi-reference triple-bond stretches, achieving accuracies superior to traditional coupled cluster theory and multi-reference perturbation theory in both the typical example of N$_2$ and the transition-metal-oxide example of [ScO]$^+$.