In this work we get insight into the impact of reduced density matrix functionals on the quality of removal/addition energies obtained using the extended Koopmans' theorem (EKT). Within the reduced density matrix functional theory (RDMFT) the EKT approach reduces to a matrix diagonalization, whose ingredients are the one- and two-body reduced density matrices. A striking feature of the EKT within RDMFT is that it opens a band gap, although it is too large, in strongly correlated materials, which are a challenge for state-of-the-art methods such as $GW$. Using the one-dimensional Hubbard model and the homogeneous electron gas as test cases, we find that (i) with exact or very accurate density matrices the EKT systematically overestimates the band gap in the Hubbard model and the bandwidth in the homogeneous electron gas and (ii) with approximate density matrices, instead, the EKT can benefit from error cancellation. In particular we test a new approximation which combines random-phase approximation screening with the power functional approximation to the two-body reduced density matrix introduced by Sharma et al. [Phys. Rev. B 78, 201103 (2008)]. An important feature of this approximation is that it reduces the EKT band gap in the studied models; it is hence a promising approximation for correcting the EKT band-gap overestimation in strongly correlated materials.
Read full abstract