General remarks

• The phrase non-local in the discussion of density functionals does not mean that non-local potentials are involved.
  The potentials are perfectly local, but when you go beyond LDA and include gradient corrections,
  the value of the density functional potential in a point r is evaluated not only from the local value
  of the charge density, but also from the gradient of the charge density.
• The Stoll formula is considered to be a correlation correction to the Local Density Approximation.
  It is conceptually not correct to use the Stoll correction and apply non-local gradient (GGA) corrections to the correlation.
  It is the user's responsibility, in general and also here, to avoid using options that are not solidly justified theoretically.
• It is questionable to apply gradient corrections to the correlation, while not doing so at the same time for the exchange.
  Therefore the program will check this and stop with an error message.
  This check can be overruled with the key allow.
• The issue of the 'best' density functional is a subject of extensive and widespread research.
  It is generally recognized that applying gradient corrections to the simplest Local Density Approximation
  usually gives better results for comparison with experimental data, especially as regards bond energies
  and the spectra computed from one-electron energies.
• The incorporation of gradient corrections during the SCF significantly increases the computing effort.
  In this respect it makes no difference which specific GGA formula is applied.
  The Energy (PostSCF) feature is therefore an alternative worthwhile considering:
  it saves a lot of time and the effects of this approximation are often small as regards the SCF solution,
  so the non-self-consistent aspect hardly shows up in the computed bond energy.
  In Geometry Optimizations, however, the Post-SCF option implies that the energy gradients are computed from
  the LDA energy expression and hence the resulting optimized geometry corresponds to the LDA functional.
  In such a case, including the GGA term may make a substantial difference to the computed equilibrium geometry.