Enhancing the Computational Efficiency of the DoNOF Program through a New Orbital Sorting Scheme

TL;DR

This paper introduces a new orbital sorting scheme for natural orbital functionals that enhances computational efficiency by gradually expanding orbital subspaces, reducing cost and improving convergence in the DoNOF program.

Contribution

The novel orbital sorting approach allows flexible subspace expansion, enabling more efficient and accurate calculations compared to previous fixed sorting methods.

Findings

Two-step strategy achieves lowest computational cost.

New sorting scheme improves convergence.

Validated on H2O, H2O2, and NH3 benchmarks.

Abstract

This work presents a novel approach to distribute orbitals into subspaces within electron-pairing-based natural orbital functionals (NOFs). This approach modifies the coupling between weakly and strongly occupied orbitals by applying an alternating orbital sorting strategy. In contrast to the previous orbital sorting that enforced electron pairing within subspaces of contiguous orbitals, the new approach provides greater flexibility, enabling a calculation scheme where the size of the subspaces can be gradually expanded. As a consequence, one can start using subspaces of only one weakly occupied orbital (perfect pairing) and progressively enlarge their size by incorporating more weakly occupied orbitals (extended pairing) up to the maximum size allowed by the basis set. In this way, the alternate orbital sorting allows solving first a simpler problem with small subspaces and leverage its orbital solution for the more intensive problem with larger subspaces, thereby reducing the overall computational cost and improving convergence, as we observed in the DoNOF program. The efficiency provided by the new sorting approach has been validated through benchmark calculations in H2O, H2O2, and NH3. In particular, we compared three strategies: i) solving directly the calculation with the largest subspaces (one-shot strategy), as was usually done before this work, ii) starting with perfect pairing and stepwise increasing the number of orbitals in the subspaces one by one until reaching the maximum size (incremental strategy), and iii) starting with perfect pairing and transitioning directly to the maximum subspace size (two-step strategy). Our results show that the two-step approach emerges as the most effective strategy, achieving the lowest computational cost while maintaining high accuracy.