The performance of multilevel quantum chemical approaches, which utilize an atom-based
system partitioning scheme to model various electronic excited states, is studied.
The considered techniques include the mechanical-embedding (ME) of "our own N-layered
integrated molecular orbital and molecular mechanics" (ONIOM) method, the point charge
embedding (PCE), the electronic-embedding (EE) of ONIOM, the frozen density-embedding
(FDE), the projector-based embedding (PbE), and our local domain-based correlation
method. For the investigated multilevel approaches, the second-order algebraic-diagrammatic
construction [ADC(2)] approach was utilized as the high-level method, which was embedded
in either Hartree-Fock or a density functional environment. The XH-27 test set of
Zech et al. [ J. Chem. Theory Comput., 2018, 14, 4028] was used for the assessment,
where organic dyes interact with several solvent molecules. With the selection of
the chromophores as active subsystems, we conclude that the most reliable approach
is local domain-based ADC(2) [L-ADC(2)], and the least robust schemes are ONIOM-ME
and ONIOM-EE. The PbE, FDE, and PCE techniques often approach the accuracy of the
L-ADC(2) scheme, but their precision is far behind. The results suggest that a more
conservative subsystem selection algorithm or the inclusion of subsystem charge-transfers
is required for the atom-based cost-efficient methods to produce high-accuracy excitation
energies.
