Hi everyone,
I have found what I can only conclude is a disturbing what I can only conclude is an error in the DFT-D corrections in Dmol3, and I assume in Castep as well.
In Grimme's 2006 paper (J. Comput Chem, 27, 1787, 2006) which details the DFT-D2 scheme, he specifies a geometric mean for the C6_ij parameters, such that:
C6_ij = sqrt(C6_i * C6_j)
In his earlier 2004 (J. Comput Chem, 25, 1463, 2004) paper for the original DFT-D scheme he specifies a more arithmetic mean of the form:
C6_ij = 2 * C6_i * C6_j / (C6_i+C6_j)
I have been doing some comparison calculations between Dmol solid and the DFT package Siesta. Siesta implements the DFT-D2 scheme if you put in all of the parameters yourself for each atom pair. In doing that I have discovered that it appears as though the earlier 2004 combination rule is the one used in Dmol, not the later 2006 one. Since all of the parameters for DFT-D2 assume the geometric mean, this means that the results from these calculations are incorrect. The absolute differences in energy between the two methods is also quite large, of the order of tens or even hundreds of kJ/mol. I would consider this a huge problem for any production calculations in Dmol as it any calculations done will not agree with other calculations done with the DFT-D2 scheme correctly implemented from the literature.
Is this assessment correct, and are the Dmol/Castep developers aware of the issue?