A reduced-cost third-order algebraic diagrammatic construction based on state-specific frozen natural orbitals: Application to the electron-attachment problem

TL;DR

This work introduces a reduced-cost, non-Dyson EA-$ADC(3)$ method based on state-specific frozen natural orbitals (SS-$FNO$) for electron-attachment problems. By combining density fitting (DF), natural auxiliary functions (NAF), and a state-specific truncation scheme with a perturbative correction, the authors achieve substantial reductions in virtual-space size and computational cost while maintaining high accuracy. Benchmark results on the EA24 set and non-valence correlation-bound (NVCB) states show that SS-$FNO$-EA-$ADC(3)$, particularly with the corrected truncation, rivals advanced EOM-CCSD-based methods and outperforms some local approximations in challenging cases. The approach scales to large systems (e.g., Zn-protoporphyrin) and offers a promising path toward accurate, scalable electron-attachment calculations in complex, sizable molecular systems, with planned extensions to relativistic regimes.

Abstract

We have developed a reduced-cost non-Dyson third-order algebraic diagrammatic construction theory for the electron-attachment problem based on state-specific frozen natural orbitals. Density fitting and truncated natural auxiliary functions were employed to enhance computational efficiency. The use of state-specific frozen natural orbitals significantly decreases the virtual space and provides a notable speedup over the conventional EA-ADC(3) method with a systematically controllable accuracy. A perturbative correction for the truncated natural orbitals significantly reduces the error in the calculated electron affinity values. The method also shows sufficient accuracy in the case of non-valence correlation-bound anions, where the local approximation-based methods fail. The efficiency of the method is demonstrated by performing an EA-ADC(3) calculation with more than 1300 basis functions.

Figures (10)

• Figure 1:
• Figure 2: The comparison of the percentage of absolute errors of EA values (in eV) of O$_3$ molecule in aug-cc-pVQZ basis set for the FNO and SS-FNO versions of EA-ADC(3) with respect to their respective canonical analogues (a) across the percentage of active virtual orbitals and (b) across different truncation thresholds.
• Figure 3: The convergence of error (in eV) in SS-FNO-EA-ADC(3) results in aug-cc-pVQZ basis and aug-cc-pVQZ auxiliary basis with respect to full NAF values. The SS-FNO truncation threshold has been kept at $10^{−4}$.
• Figure 4: The molecules in the EA24 test set.
• Figure 5: The distribution of error in EA using SS-FNO-EA-ADC(3) method in aug-cc-pVDZ basis and aug-cc-pVDZ/C auxiliary basis with respect to the canonical EA-ADC(3) values (in eV) for the EA24 test set at different SS-FNO truncation thresholds.