In recent years there has been a
significant interest of DFT community
in the development of new (hybrid) meta-generalized gradient approximations
(meta-GGAs), including range-separated hybrids. The new DFT methods,
e.g., SCAN, SCAN0, ωB97M-V, and our own LC-PBETPSS-D3 promise
an improvement over well-established models, such as, PBE, PBE0, ωB97X-D,
and the M06-type functionals. However, the tests published to date
cover only main-group chemistry. This work fills this gap by examining
reactivity of model systems, such as gold-ligand complexes, Pd and
Ni insertion reactions into covalent bonds, and the pathway for olefin
metathesis by a model Grubbs system, all of which include late transition
metals. In the attempt to rationalize the performance of functionals,
we study the fractional charge and fractional spin errors of the Au
atom and the Au7 cluster. While we find the main qualitative
issues of DFT are not yet solved, the introduction of meta-GGA ingredients
yields a notable improvement makeing the new meta-GGAs the preferred
choice for transition-metal chemistry.