Author Topic: DFT+D geometry optimization  (Read 11052 times)

lyzhu

  • Newbie
  • *
  • Posts: 9
  • Karma: +0/-0
DFT+D geometry optimization
« on: October 05, 2009, 04:02:31 PM »
Dear All,

I tried to use b3lyp+D to optimize geometry.  However, the energy start to oscillate at last. For example:
   -3597.245407632   
   -3597.245407684     
   -3597.245407604     
   -3597.245407589     
   -3597.245407479   
   -3597.245408160     
   -3597.245407654     
   -3597.245407547     
   -3597.245406674

And from the convergence information, only Max gradient not converged:

                               Converged?     Value      Criterion
             Energy change         yes      0.0000009   0.0000010
             RMS of displacement   yes      0.0000237   0.0005000
             RMS of gradient       yes      0.0004777   0.0005000
             MAX displacement      yes      0.0001510   0.0010000
             MAX gradient          no       0.0029975   0.0001000

During the optimization, I used the following keyword:
$scfconv        8
$scfdamp   start=0.800  step=0.050  min=0.05

Any comments are very appreciated! Thanks!





christof.haettig

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 291
  • Karma: +0/-0
    • Hattig's Group at the RUB
Re: DFT+D geometry optimization
« Reply #1 on: October 07, 2009, 06:17:28 PM »
It looks like a problem with the threshold and/or the grid.

The integral screening and the grid for the numerical integration in DFT lead to
a certain numerical noise in the gradient. With the default setting, the DFT gradients in
TURBOMOLE also neglect the derivatives w.r.p. to the grid parameter, since
they are usually negligible for geometry optiomizations.

For difficult cases or if you use thight convergence thresholds for the geometry, the default settings
can be insufficient. You can try to
    - increase scfconv further .e.g to 9
    - use a large grid
    - request that the derivatives w.r.p. the grid parameters are taken into account in the
      gradient calculation by specifying "weight derivatives" in the $dft data group.

If this does not help you should contact your TURBOMOLE reseller and ask for support.

Christof

lyzhu

  • Newbie
  • *
  • Posts: 9
  • Karma: +0/-0
Re: DFT+D geometry optimization
« Reply #2 on: October 19, 2009, 04:49:19 PM »
Dear Dr. Christof,

According to your suggestions, I increased scfconv to 9 and used a large gridsize 5. They works and the geometry optimization is converged.

Thanks you very much!

Best regards!

Lingyun




dlchen

  • Newbie
  • *
  • Posts: 6
  • Karma: +0/-0
Re: DFT+D geometry optimization
« Reply #3 on: November 04, 2009, 07:42:30 PM »
Hi,

Just curious, how to perform B3LYP-D optimization? From manual, it seems only B97-D has the dispersion term. Can other DFT method also be combined with D in Turbomole (version 6.0)? If so, how to set the parameters with define?

Thanks,

Deli

uwe

  • Global Moderator
  • Hero Member
  • *****
  • Posts: 558
  • Karma: +0/-0
Re: DFT+D geometry optimization
« Reply #4 on: November 05, 2009, 11:52:51 AM »
Hi,

B97-D has explicitly been re-parametrized, and B97 is not implemented in Turbomole. You have to add $disp to the control file also in case that you are using B97-D!

The functionals b-lyp, b97-d, b3-lyp, tpss, b-p, pbe, and b2-plyp-d are supported by DFT+D. You just have to add $disp to the input file, the name of the functionals themselves does not have to be changed.

Regards,

Uwe

dlchen

  • Newbie
  • *
  • Posts: 6
  • Karma: +0/-0
Re: DFT+D geometry optimization
« Reply #5 on: November 05, 2009, 08:36:33 PM »
Hi,

I see. Thanks very much!

Best,

Deli