Heterogeneous
photocatalytic processes, in which a photocatalyst
absorbs light to produce redox-active electron–hole pairs,
have strong prospects of application in light energy conversion and
environmental remediation. Although considerable efforts have been
dedicated to the development of new photocatalysts, nanoparticulated
anatase TiO2 continues to be the reference system. Its
limited light absorption properties have been addressed in different
ways, and among these, doping with heteroatoms (e.g., C or N) is a
simple and efficient (although poorly understood) strategy. In the
case of C-doping, although a significant number of computational works
have described its major features, there are still conflicting reports
on the local coordination of C, on the thermodynamic feasibility of
the doping process, and on the optical properties of the doped material.
Here, by considering surfaces instead of bulk anatase, we demonstrate
that the C-doped structures are stabilized by up to 5 eV, indicating
a spontaneous doping process across a broad range of oxygen pressures.
Furthermore, we show that the calculated absorption spectrum for the
most stable configuration, Ti-by-C substitution, is strongly dependent
on the theory level: while semiempirical calculations indicate a red
shift with respect to pristine TiO2, calculations at the
highest achievable level (qpGW
0-BSE [Bethe–Salpeter
equation] without the Tamm–Dancoff approximation [TDA]) show
no visible light absorption. In addition, we find that the commonly
used TDA introduces a significant shift to the calculated spectra
of the doped (but not the pristine) material. These results have two
important implications that can be generalized to other systems: (i)
a correct estimation of changes in optical properties upon doping
may require theory levels higher than G
0
W
0-BSE(TDA) and (ii) thermodynamic parameters
determined from doping bulk structures may significantly deviate when
considering surfaces, of relevance for nanoscaled materials.