Studying excited-state-specific perturbation theory on the Thiel set

TL;DR

This paper evaluates the performance of a new excited-state-specific perturbation theory (ESMP2) on the Thiel set, showing that regularization improves accuracy and reduces sensitivity to system size, outperforming several other methods.

Contribution

The study demonstrates that regularized ESMP2 provides a more accurate and size-insensitive approach for excited-state calculations on the Thiel set, with potential for low-cost characterization of doubly excited states.

Findings

Regularization reduces ESMP2 sensitivity to π system size.

Regularized ESMP2 outperforms CC2, EOM-CCSD, CC3, and DFT approaches.

ESMP2 doubles norm helps identify doubly excited states efficiently.

Abstract

We explore the performance of a recently-introduced $N^5$-scaling excited-state-specific second order perturbation theory (ESMP2) on the singlet excitations of the Thiel benchmarking set. We find that, without regularization, ESMP2 is quite sensitive to $\pi$ system size, performing well in molecules with small $\pi$ systems but poorly in those with larger $\pi$ systems. With regularization, ESMP2 is far less sensitive to $\pi$ system size and shows a higher overall accuracy on the Thiel set than CC2, EOM-CCSD, CC3, and a wide variety of time-dependent density functional approaches. Unsurprisingly, even regularized ESMP2 is less accurate than multi-reference perturbation theory on this test set, which can in part be explained by the set's inclusion of some doubly excited states but none of the strong charge transfer states that often pose challenges for state-averaging. Beyond energetics, we find that the ESMP2 doubles norm offers a relatively low-cost way to test for doubly excited character without the need to define an active space.