CCL:G: D3 in Gaussian (previous message G: b3lyp*)

 Sent to CCL by: "Stefan  Grimme" [grimme,thch.uni-bonn.de]
 Dear all,
 because there was a recent post claiming that our D3 method now has
 found its way into the Gaussian code, please allow me some comments
 to this (I quote the following from a previous post):
 >In newest revision of Gaussian (D.01) which has just arrived (released on
 >10 May 2013) there is new keyword implemented for dispersion correction
 >which include two DFT-D3 correction variants. It is called
 >EmpiricalDispersion=PFD, GD3 or GD3BJ and quote from release notes
 >"explicitly request Petersson-Frisch dispersion [Austin12] or
 >Grimme’s D3 dispersion [Grimme10] or D3BJ dispersion
 [Grimme11].".
 1. GD3BJ is the recommended, physically more sound version based
    on the Becke-Johnson damping scheme.
    GD3 can be used for checking and clarifying the more short-ranged
    contributions which is smaller than in GD3BJ due to the different
    damping (the dispersion coefficients are the same).
    For some functionals like those from the M0X class
    only GD3 is available because they already include
    the short/medium-range part in the functional and only require
    correction for the long-range, London part.
 2. This London part of D3 dispersion is NON-EMPIRICAL
    (not fitted). Its comes out of the TDDFT box.
    What is fitted is the connection to the short range DFT part and
    this involves three global parameters similar to most other
    dispersion corrections (e.g. XDM, vdW-TS, VV10) where
    this is done similarly. And by the way: if this is called
 "empirical",
    then consistently every fitted density functional (e.g. B3LYP) should
    be called "empirical". The better way to clarify its theoretical
 basis
    and to distinguish D3 from other methods is to call it
    "atom pairwise" correction (or even more
    precisely: "system-dependent, density-independent, atom pairwise";
    in that sense XDM or vdW-TS differs from D3 by being density dependent
    and VV10 is just "non-local, density dependent").
 3. The "*" in B3LYP* and D3 are not related (although D3 in its B3LYP
    form can be used for B3LYP* giving a B3LYP*-D3).
 Hope this helps
 Stefan