Extended Second-Order Multireference Algebraic Diagrammatic Construction Theory for Charged Excitations

TL;DR

This paper introduces a new implementation of multireference algebraic diagrammatic construction theory (MR-ADC) for charged excitations, including second-order and extended approximations, enabling efficient and accurate simulations of strongly correlated molecular systems.

Contribution

The paper presents the first implementation of second-order EA-MR-ADC(2) and extended EA/IP-MR-ADC(2)-X methods with low computational scaling, improving accuracy and efficiency in modeling strongly correlated molecules.

Findings

EA- and IP-MR-ADC(2)-X achieve accuracy similar to NEVPT2

Implementation has O(M^5) scaling with basis set size

Efficient algorithm avoids four-particle RDM computations

Abstract

We report a new implementation of multireference algebraic diagrammatic construction theory (MR-ADC) for simulations of electron attachment and ionization in strongly correlated molecular systems (EA/IP-MR-ADC). Following our recent work on IP-MR-ADC [J. Chem. Theory Comput. 2019, 15, 5908], we present the first implementation of the second-order MR-ADC method for electron attachment (EA-MR-ADC(2)), as well as two extended second-order approximations (EA- and IP-MR-ADC(2)-X) that incorporate a partial treatment of third-order electron correlation effects. Introducing a small approximation for the second-order amplitudes of the effective Hamiltonian, our implementation of EA- and IP-MR-ADC(2)-X has a low O(M^5) computational scaling with the basis set size M. Additionally, we describe an efficient algorithm for solving the first-order amplitude equations in MR-ADC and partially-contracted second-order N-electron valence perturbation theory (NEVPT2) that completely avoids computation of the four-particle reduced density matrices without introducing any approximations or imaginary-time propagation. For a benchmark set of eight small molecules, carbon dimer, and a twisted ethylene, we demonstrate that EA- and IP-MR-ADC(2)-X achieve accuracy similar to that of strongly-contracted NEVPT2, while having a lower computational scaling with the active space size and providing efficient access to transition properties.