Communication: Configuration interaction singles has a large systematic bias against charge-transfer states

Joseph E Subotnik

doi:10.1063/1.3627152

Abstract

We show that standard configuration interaction singles (CIS) has a systematic bias against charge-transfer (CT) states, wherein the computed vertical excitation energies for CT states are disproportionately too high (by >1 eV) as compared with non-CT states. We demonstrate this bias empirically for a set of chemical problems with both inter- and intra-molecular electron transfer, and then, for a small analytical model, we prove that this large difference in accuracy stems from the massive changes in electronic structure that must accompany long-range charge transfer. Thus far, the conclusion from this research is that, even in the context of wave function theory, CIS alone is insufficient for offering a balanced description of excited state surfaces (both CT and non-CT) and explicit electron-electron correlation must be included additionally (e.g., via CIS(D)) for charge-transfer applications.

Highlights

The simplest means for computing molecular excited states is configuration interaction singles (CIS), where the wave function is given the form | CIS = ia tia | a iFollowing standard protocol, in this communication, we denote occupied orbitals ij and virtual orbitals ab
The disadvantage of CIS is that, because the method makes no attempt to allow for electron-electron correlation in the excited state,3 it often does not recover accurate vertical excitation energies
time-dependent density functional theory4 (TD-DFT) fails miserably, for CT states because it does not recover the correct −1/r asymptotic behavior of charge-transfer states,2 which leads to CT excitation energies that are often many eV too low, and getting worse for larger systems

Summary

INTRODUCTION

The simplest means for computing molecular excited states is configuration interaction singles (CIS), where the wave function is given the form | CIS =. As a tool for computing excited state energies, CIS has several advantages: first, it is computationally cheap; second it recovers both non-charge-transfer (CT) and CT states, including the correct −1/r asymptotic behavior of CT states that comes about because of the Coulombic attraction between electron attachments and detachments.. TD-DFT fails miserably, for CT states because it does not recover the correct −1/r asymptotic behavior of charge-transfer states, which leads to CT excitation energies that are often many eV too low, and getting worse for larger systems.. We ask a very simple question: knowing that CIS recovers both non-CT and CT states (including the correct −1/r asymptotic form for CT states), can we say that the relative CIS energies are unbiased between CT and non-CT states? Verify these results with equation of motion coupled-cluster singles doubles (EOM-CCSD).

Water clusters

Waters

ANALYTICAL TREATMENT

FUTURE DIRECTIONS