The classic density-functional theory (DFT) formalism introduced by Hohenberg, Kohn, and Sham in the mid-1960s is based on the idea that the complicated N-electron wave function can be replaced with the mathematically simpler 1-electron charge density in electronic structure calculations of the ground stationary state. As such, ordinary DFT cannot treat time-dependent (TD) problems nor describe excited electronic states. In 1984, Runge and Gross proved a theorem making TD-DFT formally exact. Information about electronic excited states may be obtained from this theory through the linear response (LR) theory formalism. Beginning in the mid-1990s, LR-TD-DFT became increasingly popular for calculating absorption and other spectra of medium- and large-sized molecules. Its ease of use and relatively good accuracy has now brought LR-TD-DFT to the forefront for this type of application. As the number and the diversity of applications of TD-DFT have grown, so too has our understanding of the strengths and weaknesses of the approximate functionals commonly used for TD-DFT. The objective of this article is to continue where a previous review of TD-DFT in Volume 55 of the Annual Review of Physical Chemistry left off and highlight some of the problems and solutions from the point of view of applied physical chemistry. Because doubly-excited states have a particularly important role to play in bond dissociation and formation in both thermal and photochemistry, particular emphasis is placed on the problem of going beyond or around the TD-DFT adiabatic approximation, which limits TD-DFT calculations to nominally singly-excited states.

Under the usual assumption of noninteracting v-representability, density-functional theory (DFT) together with time-dependent DFT (TDDFT) provide a formally exact single-reference method suitable for the theoretical description of the electronic excited-states of large molecules, and hence for the description of excited-state potential energy surfaces important for photochemistry. The quality of this single-reference description is limited in practice by the need to use approximate exchange–correlation functionals. In particular it is far from clear how well approximations used in contemporary practical TDDFT calculations can describe funnel regions such as avoided crossings and conical intersections. These regions typically involve biradical-like structures associated with bond breaking and conventional wisdom would seem to suggest the need to introduce explicit double excitation character to describe these structures. Although this is lacking in ordinary spin-preserving (SP) TDDFT, it is present to some extent in spin-flip (SF) TDDFT. We report our tests of Wang–Ziegler noncollinear SF-TDDFT within the Tamm–Dancoff approximation for describing the avoided crossing in the C_2v CC ring-opening reaction of oxirane and for describing the conical intersection relevant for the more physical asymmetric CO ring-opening reaction of oxirane. Comparisons are made with complete active space self-consistent field and quantum Monte Carlo benchmark results from two previous papers on the subject [J. Chem. Phys., 2007, 127, 164111; ibid129, 2008, 124108]. While the avoided crossing in the C_2v pathway is found to be reasonably well described, the method was found to be only partially successful for the conical intersection (CX) associated with the physically more important asymmetric pathway. The origin of the difficulties preventing the noncollinear SF-TDDFT method from giving a completely satisfactory description of the CX was traced back to the inability of SF-TDDFT based upon a single triplet reference state to correlate all potentially relevant configurations involving not just two but three nearly degenerate orbitals (n, σ_CO , and ). This article is also the first report of our implementation of SF-TDDFT within the DEMON2K program.

It has long been known that only one-electron excitations are available from adiabatic time-dependent density functional theory (TDDFT). This is particularly clear in Casida’s formulation of TDDFT linear response theory [M. E. Casida, in Recent Advances in Density Functional Methods, Part I, edited by D. P. Chong (World Scientific, Singapore, 1995), p. 155]. Nevertheless the explicit inclusion of two- and higher-electron excitations is necessary for an adequate description of some excited states, notably the first excited singlet states of butadiene and quartet excited states of molecules with a doublet ground state. The equation-of-motion superoperator approach is used here to derive a Casida-like propagator equation which can be clearly separated into an adiabatic part and a nonadiabatic part. The adiabatic part is identified as corresponding to Casida’s equation for adiabatic TDDFT linear response theory. This equivalence is confirmed by deriving a general formula which includes the result that Gonze and Scheffler derived to show the equivalence of TDDFT and Görling-Levy adiabatic connection perturbation theory for the exchange-only optimized effective potential [X. Gonze and M. Scheffler, Phys. Rev. Lett. 82, 4416 (1999)]. The nonadiabatic part explicitly corrects adiabatic TDDFT for two- and higher-electron excitations. The “dressed TDDFT” of Maitra, Zhang, Cave, and Burke is obtained as a special case where the ground state is closed shell [N. T. Maitra, F. Zhang, R. J. Cave, and K. Burke, J. Chem. Phys. 120, 5932 (2004)]. The extension of dressed TDDFT to the case where the ground state is an open-shell doublet is presented, highlighting the importance of correctly accounting for symmetry in this theory. The extension to other ground state spin symmetries is a straightforward consequence of the present work.

Ionization potentials (IPs) and electron affinities (EAs) are important quantities input into most models for calculating the open-circuit voltage (V_oc) of organic solar cells. We assess the semi-empirical density-functional tight-binding (DFTB) method with the third-order self-consistent charge (SCC) correction and the 3ob parameter set (the third-order DFTB (DFTB3) organic and biochemistry parameter set) against experiments (for smaller molecules) and against first-principles GW (Green’s function, G, times the screened potential, W) calculations (for larger molecules of interest in organic electronics) for the calculation of IPs and EAs. Since GW calculations are relatively new for molecules of this size, we have also taken care to validate these calculations against experiments. As expected, DFTB is found to behave very much like density-functional theory (DFT), but with some loss of accuracy in predicting IPs and EAs. For small molecules, the best results were found with ΔSCF (Δ self-consistent field) SCC-DFTB calculations for first IPs (good to ± 0.649 eV). When considering several IPs of the same molecule, it is convenient to use the negative of the orbital energies (which we refer to as Koopmans’ theorem (KT) IPs) as an indication of trends. Linear regression analysis shows that KT SCC-DFTB IPs are nearly as accurate as ΔSCF SCC-DFTB eigenvalues (± 0.852 eV for first IPs, but ± 0.706 eV for all of the IPs considered here) for small molecules. For larger molecules, SCC-DFTB was also the ideal choice with IP/EA errors of ± 0.489/0.740 eV from ΔSCF calculations and of ± 0.326/0.458 eV from (KT) orbital energies. Interestingly, the linear least squares fit for the KT IPs of the larger molecules also proves to have good predictive value for the lower energy KT IPs of smaller molecules, with significant deviations appearing only for IPs of 15–20 eV or larger. We believe that this quantitative analysis of errors in SCC-DFTB IPs and EAs may be of interest to other researchers interested in DFTB investigation of large and complex problems, such as those encountered in organic electronics.