Empirical dispersion corrections

Find answers or ask questions regarding Dalton calculations.
Please upload an output file showing the problem, if applicable.
(It is not necessary to upload input files, they can be found in the output file.)

Post Reply
Carravetta
Posts: 6
Joined: 14 Feb 2014, 13:42
First name(s): Vincenzo
Last name(s): Carravetta
Affiliation: CNR-IPCF
Country: Italy

Empirical dispersion corrections

Post by Carravetta » 20 Nov 2015, 12:54

I wonder if any user had (good? bad?) experience with the option .DFTD3 that switches on Grimmes DFT-D3 empirical dispersion correction to a density functional.
I am running (DALTON 2015.0) calculations for the gs of Si2 (R=2.3 Angstrom, basis set cc-pVQZ ) with BLYP and the mentioned .DFTD3 option. The dispersion
correction is computed by Dalton as E_disp : -0.004622463941 au, correspondig to about -2.9 kcal/mol.
Using the same Grimme approach (and same functional and basis set) with the program NWChem I get Dispersion correction = -0.000038821955 au (-0.024 kcal/mol).
Similar (to NWChem) results are obtained by Quantum-ESPRESSO (plane-waves + pseudopotentials).
A difference of two orders of magnitude seems excessive to me even for a tiny quantity as the dispersion energy. An answer from the main author (Andrew Teale)
of that part of the code would be, of course, very much appreciated.
Let me add that the gs of Si2 is a quartet, but Dalton doesn't complain about the presence of open shells running the Grimme calculation.

Vincenzo Carravetta

a_teale
Posts: 9
Joined: 02 Sep 2013, 14:38
First name(s): Andy
Last name(s): Teale

Re: Empirical dispersion corrections

Post by a_teale » 20 Nov 2015, 15:25

Could you post or send me (andrew.teale@nottingham.ac.uk) the input .dal .mol and output file for your example.

taylor
Posts: 589
Joined: 15 Oct 2013, 05:37
First name(s): Peter
Middle name(s): Robert
Last name(s): Taylor
Affiliation: Tianjin University
Country: China

Re: Empirical dispersion corrections

Post by taylor » 20 Nov 2015, 15:37

I know nothing about DFTD3, but I had a couple of suggestions/comments. First, the Dalton number seems unrealistically large for such a system. Second, I am not sure that merely running to completion guarantees that the treatment of open-shell systems (I assume you meant "triplet" rather than "quartet"?) is correct. And third, is it worth doing a similar comparison for (say) N2 or P2 (although the latter is messily multiconfigurational in reality) where you will have a closed-shell system, and comparing this with another program?

Best regards
Pete

a_teale
Posts: 9
Joined: 02 Sep 2013, 14:38
First name(s): Andy
Last name(s): Teale

Re: Empirical dispersion corrections

Post by a_teale » 20 Nov 2015, 16:12

Ok ... I ran a quick test with Grimme's definitive code...

i) I think the number of -0.004622463941 au you quote is from a .DFD3BJ run (D2 corrections with the recommended Becke-Johnson (BJ) damping).

ii) Below are checks with DFT-D3 and DFT-D3-BJ

DFT-D3

./dftd3 si2.xyz -func b-lyp -zero

_________________________________

| DFTD3 V3.0 Rev 2 |
| S.Grimme, University Bonn |
| Aug 2013 |
| see dftd3 -h for options |
_________________________________

Please cite DFT-D3 work done with this code as:
S. Grimme, J. Antony, S. Ehrlich and H. Krieg,
J. Chem. Phys. 132 (2010), 154104
If used with BJ-damping cite also
S. Grimme, S. Ehrlich and L. Goerigk,
J. Comput. Chem. 32 (2011), 1456-1465
For DFT-D2 the reference is
S. Grimme, J. Comput. Chem., 27 (2006), 1787-1799

files read :
si2.xyz
C6 coefficients used:
5 C6 for element 14
Z= 14 CN= 0.000 C6(AA)= 317.86
Z= 14 CN= 0.951 C6(AA)= 270.26
Z= 14 CN= 1.944 C6(AA)= 233.13
Z= 14 CN= 2.941 C6(AA)= 221.50
Z= 14 CN= 3.868 C6(AA)= 149.77

# XYZ [au] R0(AA) [Ang.] CN C6(AA) C8(AA) C10(AA) [au]
1 0.00000 0.00000 0.00000 si 1.896 0.964 270.5 19349.6 1695414.9
2 0.00000 0.00000 4.34637 si 1.896 0.964 270.5 19349.6 1695414.9

molecular C6(AA) [au] = 1082.09

DFT-D V3
DF b-lyp
parameters
s6 : 1.0000
s8 : 1.6820
rs6 : 1.0940
rs18 : 1.0000
alpha6 : 14.0000
alpha8 : 16.0000
k1-k3 : 16.0000 1.3333 -4.0000
Cutoff : 94.8683 a.u.
CN-Cutoff: 40.0000 a.u.

Edisp /kcal,au: -0.0101 -0.00001605

E6 /kcal : -0.0011
E8 /kcal : -0.0090
% E8 : 89.18
normal termination of dftd3

This appears to match well with the DALTON output...



DFT-D Empirical Dispersion Correction
-------------------------------------

Running DFT-D Version : 3
Reference : S. Grimme, J. Antony, S. Ehrlich and H. Krieg,
J. Chem. Phys. 132 (2010), 154104
Becke-Johnson Damping : F
3-body terms : F


DFT-D3 Functional Dependent Parameters
--------------------------------------

s_6 = 1.000000000000 sr_6 = 1.094000000000
s_8 = 1.682000000000 sr_8 = 1.000000000000
alp = 14.000000000000


+------------------------------+
! Dispersion Energy Correction !
+------------------------------+

E_disp : -0.000016045038

DFT-D3-BJ

_________________________________

| DFTD3 V3.0 Rev 2 |
| S.Grimme, University Bonn |
| Aug 2013 |
| see dftd3 -h for options |
_________________________________

Please cite DFT-D3 work done with this code as:
S. Grimme, J. Antony, S. Ehrlich and H. Krieg,
J. Chem. Phys. 132 (2010), 154104
If used with BJ-damping cite also
S. Grimme, S. Ehrlich and L. Goerigk,
J. Comput. Chem. 32 (2011), 1456-1465
For DFT-D2 the reference is
S. Grimme, J. Comput. Chem., 27 (2006), 1787-1799

files read :
si2.xyz
C6 coefficients used:
5 C6 for element 14
Z= 14 CN= 0.000 C6(AA)= 317.86
Z= 14 CN= 0.951 C6(AA)= 270.26
Z= 14 CN= 1.944 C6(AA)= 233.13
Z= 14 CN= 2.941 C6(AA)= 221.50
Z= 14 CN= 3.868 C6(AA)= 149.77

# XYZ [au] R0(AA) [Ang.] CN C6(AA) C8(AA) C10(AA) [au]
1 0.00000 0.00000 0.00000 si 0.962 0.964 270.5 19349.6 1695414.9
2 0.00000 0.00000 4.34637 si 0.962 0.964 270.5 19349.6 1695414.9

molecular C6(AA) [au] = 1082.09

DFT-D V3(BJ)
DF b-lyp
parameters
s6 : 1.0000
s8 : 2.6996
a1 : 0.4298
a2 : 4.2359
k1-k3 : 16.0000 1.3333 -4.0000
Cutoff : 94.8683 a.u.
CN-Cutoff: 40.0000 a.u.

Edisp /kcal,au: -2.9006 -0.00462246

E6 /kcal : -0.6943
E8 /kcal : -2.2063
% E8 : 76.06
normal termination of dftd3

Also seems to line up with DALTON,


DFT-D Empirical Dispersion Correction
-------------------------------------

Running DFT-D Version : 3
Reference : S. Grimme, J. Antony, S. Ehrlich and H. Krieg,
J. Chem. Phys. 132 (2010), 154104
Reference : S. Grimme, S. Ehrlich and L. Goerigk,
J. Comput. Chem. 32 (2011), 1456-1465
Becke-Johnson Damping : T
3-body terms : F


DFT-D3 Functional Dependent Parameters
--------------------------------------

s_6 = 1.000000000000 sr_6 = 0.429800000000
s_8 = 2.699600000000 sr_8 = 4.235900000000
alp = 14.000000000000


+------------------------------+
! Dispersion Energy Correction !
+------------------------------+

E_disp : -0.004622463941

So I think you have run .DFD3BJ in your calculation. Could you check your inputs...

Clearly there is a large difference between DFT-D3 and DFT-D3-BJ. Although the former is closer to the other number you quote of -0.000038821955 au it doesn't line up.

Since the empirical dispersion correction is a simple contribution that is independent of details of the actual electronic structure the numbers should line up exactly. So I suspect you are not getting DFT-D3 in your other codes.

a_teale
Posts: 9
Joined: 02 Sep 2013, 14:38
First name(s): Andy
Last name(s): Teale

Re: Empirical dispersion corrections

Post by a_teale » 20 Nov 2015, 16:26

I think your other number is actually DFT-D2, just for completeness :-).

./dftd3 si2.xyz -func b-lyp -old

_________________________________

| DFTD3 V3.0 Rev 2 |
| S.Grimme, University Bonn |
| Aug 2013 |
| see dftd3 -h for options |
_________________________________

Please cite DFT-D3 work done with this code as:
S. Grimme, J. Antony, S. Ehrlich and H. Krieg,
J. Chem. Phys. 132 (2010), 154104
If used with BJ-damping cite also
S. Grimme, S. Ehrlich and L. Goerigk,
J. Comput. Chem. 32 (2011), 1456-1465
For DFT-D2 the reference is
S. Grimme, J. Comput. Chem., 27 (2006), 1787-1799

files read :
si2.xyz
loading DFT-D2 parameters ...
C6 coefficients used:
1 C6 for element 14
Z= 14 CN=****** C6(AA)= 160.10

# XYZ [au] R0(AA) [Ang.] CN C6(AA) C8(AA) C10(AA) [au]
1 0.00000 0.00000 0.00000 si 1.560 0.964 160.1 11451.2 1003354.7
2 0.00000 0.00000 4.34637 si 1.560 0.964 160.1 11451.2 1003354.7

molecular C6(AA) [au] = 640.39

DFT-D V2
DF b-lyp
parameters
s6 : 1.2000
alpha6 : 20.0000
Cutoff : 94.8683 a.u.
CN-Cutoff: 40.0000 a.u.

Edisp /kcal,au: -0.0244 -0.00003884

E6 /kcal : -0.0244
normal termination of dftd3

and from DALTON,

DFT-D Empirical Dispersion Correction
-------------------------------------

Running DFT-D Version : 2
Reference : S. Grimme, J. Comput. Chem., 27 (2006), 1787-1799


DFT-D2 Functional Dependent Parameters
--------------------------------------

s_6 = 1.200000000000 sr_6 = 1.100000000000
s_8 = 0.000000000000 alp = 20.000000000000


+------------------------------+
! Dispersion Energy Correction !
+------------------------------+

E_disp : -0.000038840549


I am happy to look into this more if you go through your input / output files and still think there is a discrepancy.

Carravetta
Posts: 6
Joined: 14 Feb 2014, 13:42
First name(s): Vincenzo
Last name(s): Carravetta
Affiliation: CNR-IPCF
Country: Italy

Re: Empirical dispersion corrections

Post by Carravetta » 03 Dec 2015, 15:02

Dear Andrew

sorry for my late reply (I could not check the Dalton forum until today) and many thanks for all your efforts.

I agree with your conclusions:
1) my calculation was done with the .DFD3BJ option that, according to the Dalton manual is the "presently recommended version" for the Grimmes DFT empirical dispersion correction,
2) the calculations with other codes very probably used the D2 correction, as your calculations suggest.
I also agree with comments from Pete (the gs of Si2 is a triplet of course) that -0.004622463941 au is an unrealistical value for Si2 and that the completion of a calculation does not guarantees about its correctness (that's, indeed, why I run other calculations).

I think that the correctness of Dalton code is proved, but then the question moves to the "quality" of the Grimme correction in the different versions. As an expert in this field would you still say that DFD3BJ is the recommended version? Or should we accept that, as it happens sometimes with the choise of the density functional, also the empirical dispersion correction is a "matter of taste"? ;)

Best regards

Vincenzo

Post Reply

Who is online

Users browsing this forum: No registered users and 13 guests