The restricted open-shell Kohn-Sham (ROKS) approach is a
highly accurate method for estimating core-excitation energies of closed-shell
molecules,^{Hait:2020a}
as described in Section 7.6. Here, we briefly recapitulate
the key aspects and refer the reader to Ref. Hait:2020a for details
ROKS with the SCAN functional is found to reproduce
40 experimental core excitation energies (from the 1s orbital, *i.e.*, K-edge) of
second period elements (C, N, O, and F) to an RMS error of 0.2 eV and a maximum
absolute error of 0.5 eV. The $\omega $B97X-V functional provides
similar (if a little worse) accuracy as well. Similar behavior is observed for
the L${}_{2,3}$ edges of third-period elements Si, P, S and Cl. Other widely used
functionals like PBE fare somewhat worse, but still predict much lower error
than TDDFT with corresponding functionals.

That said, the ROKS approach is state-specific in
that it can only predict a single state at a time and needs to be told which
state to target (via the $reorder_mo section, as shown in
example 7.7). This makes it less black-box than TDDFT as the final
orbital needs to be identified *a priori*, perhaps via a pilot TDDFT job if
no other information is available. (For core-level excitations, the intial
orbital is usually intuitively obvious.) ROKS can also
be used for two-site doubly core-ionized states, or other systems with one
broken electron pair in total.

The accuracy of ROKS stems from three factors: choice of density functional (SCAN or $\omega $B97X-V), excited state orbital optimization (only available via SGM for core excitations, as described in Section 4.5.11) and a sufficiently flexible basis set. The last is key, as the split-core functions (as provided by basis sets like cc-pCV$n$Z) are needed instead of standard basis sets like cc-pV$n$Z that only have split valence functions. Indeed, a basis of triple-zeta quality like cc-pCVTZ is necessary to fully account for the core-hole relaxation and smaller basis sets lead to systematic overestimation of excitation energies. However, the highly local nature of the core-hole ensures that a large basis is only needed for the target atom of the ROKS calculation, and a smaller basis (of double-zeta quality, like cc-pVDZ) is adequate for all other atoms. An example of this mixed basis strategy is given below in example 7.12.3. Details about using mixed basis sets in general can be found in Section 8.5.

The number of cycles needed for ROKS calculations can also be considerably reduced by decoupling the cole-hole relaxation from the rest of the orbital optimization. This entails a restricted open-shell $\mathrm{\Delta}$SCF calculation of the core-ionized state first, and use of those orbitals as guess for ROKS. Example 7.7 is a representative case for how such calculations should proceed.

The conjunction of high accuracy and low computational cost (due to cheapness of the SCAN meta-GGA and the mixed basis strategy) makes ROKS a very attractive approach for computing core spectra of large, closed-shell systems (where more expensive wave function theories are unaffordable). Users are requested to cite Ref. Hait:2020a when using ROKS for core excitations.

$molecule 0 1 C 0.0000 0.0000 0.0000 O 0.0000 0.0000 1.1282 $end $rem method scan basis gen basis2 aug-cc-pVDZ symmetry false $end $basis C aug-cc-pCVTZ **** O aug-cc-pVDZ **** $end @@@@@@@@@@@@ $molecule 1 2 C 0.0000 0.0000 0.0000 O 0.0000 0.0000 1.1282 $end $rem method scan basis gen unrestricted false scf_guess read symmetry false scf_algorithm sgm $end $reorder_mo 1 3 4 5 6 7 2 1 3 4 5 6 7 2 $end $basis C aug-cc-pCVTZ **** O aug-cc-pVDZ **** $end