VDW: option 7 No 3rd body interaction???

Total energy, geometry optimization, DFT+U, spin....

Moderator: bguster

Locked
johnbrehm
Posts: 12
Joined: Mon Sep 25, 2017 9:30 pm

VDW: option 7 No 3rd body interaction???

Post by johnbrehm » Mon Apr 23, 2018 8:19 pm

Hi,

I am very confused by the Abinit literature.

It says for vdw-xc: 7: apply vdw-DFT-D3(BJ) as proposed by Grimme (based on Becke-Jonhson method J. Chem. Phys. 2004-2006)
Available only for ground-state calculations and response functions; see vdw_tol variable to control convergency

I go to vdw_tol.
there it says:
Only relevant if vdw_xc==5

Ok. Great.
So I do a bunch of relaxations and I notice while using vdw == 7 the following: --------------------------------------------------------------
Van der Waals DFT-D3 semi-empirical dispersion potential
with Becke-Jonhson (BJ) refined by Grimme et al. J.
Comput. Chem. 32, 1456 (2011)
---------------------------------------------------------------
Scaling factors: s6 = 1.000, s8 = 0.787
Damping parameters: a1 = 0.429, a2 = 4.441
Cut-off radius = 1.20806E+02 Bohr
Number of pairs contributing = 2051514
DFT-D3 (no 3-body) energy contribution = -4.52233E-01 Ha
----------------------------------------------------------------

AND NOW I AM REALLY WORRIED:
THe last line says (no 3-body)!

I go back to the literature, and see another variable: vdw_tol_3bt
But it says:
Only relevant if vdw_xc == 6
AND more importantly:
Default is -1 (Comment: Do include the 3-body term in the correction). --- [Note:This comment is not mine -- it is in the literature]

BUT, then in contradiction:
Control the computation of the 3-body correction inside DFT-D3 dispersion correction (Grimme approach) to the total energy:
-If vdw_tol_3bt<0, no 3-body correction.
-If vdw_tol_3bt>0, the 3-body term is included with a tolerance = vdw_tol_3bt


So, now totally confused but disturbed that my output has been yielding "(no-3 body)"
I decide to set it and compare to my earlier runs --I set it to
vdw_tol_3bt == 1d-10 (just like the default value for vdw_tol).

And now I get an output with 3 body energy corrections -- I think:
--------------------------------------------------------------
Van der Waals DFT-D3 semi-empirical dispersion potential
with Becke-Jonhson (BJ) refined by Grimme et al. J.
Comput. Chem. 32, 1456 (2011)
---------------------------------------------------------------
Scaling factors: s6 = 1.000, s8 = 0.787
Damping parameters: a1 = 0.429, a2 = 4.441
Cut-off radius = 1.20806E+02 Bohr
Number of pairs contributing = 2051500
DFT-D3 (no 3-body) energy contribution = -4.52234E-01 Ha

---------------------------------------------------------------
3-Body Term Contribution:
Number of shells considered = 12
Additional 3-body contribution = 2.68812201945E-02 Ha
Total E (2-body and 3-body) = -4.25353118753E-01Ha
---------------------------------------------------------------


My question to ABINIT:
Which is the correct way to use vdw == 7????

thanks,
JB

cespejo
Posts: 21
Joined: Fri Feb 26, 2010 8:12 pm

Re: VDW: option 7 No 3rd body interaction???

Post by cespejo » Fri May 04, 2018 6:28 pm

Dear JB,
DFT-D1, DFT-D2 and DFT-D3 stands for different versions of the so called Grimme method to include van der Waals eenrgies in DFT. In DFT-D3 a three-body term can be included (however, this is not reason why it is called DFT-D3). In the literature, people have used the denomination DFT-D3 both when they were including, or not this three-body term. Its use is generally given in the computational method. On the other hand the difference between DFT-D3 and the DFT-D3(BJ) method relies in the type of damping function used for two-body contributions to the vdW energy for small interatomic distances. With this in mind vdw_xc=7 corresponds to the DFT-D3(BJ) method.

Concerning the tolerance (vdw_tol), it is in fact used in the three DFT-D methods; it will be effectively used in the computation in DFT-D3(BJ). I recommand to use vdw_tol 1d-12 [Ha].

If you want to include three-body terms you should set the variable vdw_tol_3bt as well, but to the best of our knowledge it has never been used in DFT-D3(BJ), only in DFT-D3. While, it would be in principle feasible to implement a damping function for this three-body term in DFT-D3(BJ), the three-body is costly to compute and gives a relatively small contribution to the total dispersion energy (~5%) and not necessarily in the good direction. Finally, while forces and stresses are available with this three-body term inside abinit, phonons and elastic constants computed in DFPT are not.

Cheers,
Camilo and Benoît

Locked