Exploring the accuracy and usefulness of semi-empirically scaled ADC schemes by blending second and third order terms.

Abstract

Different approaches to mixed-order algebraic-diagrammatic construction (ADC) schemes are investigated. The performance of two different strategies for scaling third-order contributions to the ADC secular matrix is evaluated. Both considered schemes employ a single tuning parameter and conserve general properties inherent to all ADC methods, such as hermiticity and size-consistency. The first approach, scaled-matrix ADC[(2) + x(3)], scales all contributions first occurring in ADC(3) equally and leads to an improvement of the accuracy of excitation energies compared to ADC(3) for x = 0.4-0.5. However, with respect to excited state dipole moments, this method provides lower accuracy than ADC(3). The second scaling approach, MP[(1) + x(2)] - ISR(3), scales the second order contributions of the ground-state wavefunction and derives a rigorous ADC scheme via the intermediate state representation formalism. Although the error in excitation energies is not improved, this method provides insight into the relevance of the individual terms of the ADC(3) matrix and indicates that the MP(2) wavefunction is, indeed, the optimal reference wavefunction for deriving a third-order single-reference ADC scheme.