[CPMD-list] forces, convergence and gga

[De Dobbelaere Christopher]

Hi cpmd users,

Another few additions :)

> CD> Hi,
> CD> 
> CD> Thanks for the replies (Axel also).
> CD> 
> CD> Referring to question 2:
> CD> Maybe i didn't formulate the problem as it should. What i did is the following:
> CD> I want to use the VDB uspp and BLYP functional (because B3LYP describes the molecule relatively good in gaussian). So i tried different cutoff values (20 to 45) and looked for convergence for some 'important' bond lengths. I think the convergence for S-N and S-C is not really good, but C=C and C-H seem to converge nicely (cutoff values going from 20 Ry to 45 Ry).
>
> please note, that there is a significant difference between BLYP
> and B3LYP, the former does not include the hartree-fock exchange.
> so you should first compare (in gaussian) how the values change
> when going from B3LYP to BLYP.


I know, but I'm not worried about the fact that the values are 'wrong', it's just that they do not converge when increasing cutoff values that worries me.


> CD> Values for S-N going from 20 to 45 Ry (steps of 5 Ry) in Angstrom:
> CD> 1.72176      1.71772         1.71836         1.71851         1.71658         1.71772       
> CD> Values for C-N:
> CD> 1.81367      1.82001         1.81453         1.82000         1.81888         1.82267       
> CD> Values for C=C:
> CD> 1.42900      1.41553         1.41521         1.41571         1.41532         1.41554       
> CD> 
> 
> CD> The question now is either or not the S-N and C-N values may be
> CD> considered as converged or not? And maybe someone knows why the S-N
> CD> and C-N values are not as good converged as the C=C and C-H values.
> CD> We do know from experience that C-N and S-N (in our molecule) are
> CD> difficult to converge in gaussian calculations also, so maybe it's
> CD> just an functional issue, but maybe i'm just doing something wrong
> CD> with cpmd..
>
> what kind of _density_ cutoff did you use here? also what was
> the convergence for the geometry optimization and the
> corresponding wavefunction optimization convergence parameter?


I used the following parameters:

 CONVERGENCE ORBITALS
  1.0d-7
 CONVERGENCE GEOMETRY
  4.5d-4

CONVERGENCE INITIAL
  1.0d-4
 CONVERGENCE ADAPT
  0.02
 CONVERGENCE ENERGY
  0.05

and

 GC-CUTOFF
  1.0d-06


> it may simply be, that you are a victim of 'ripples'.
> please note, that the pseudopotentials you are using were
> 'optimized' to give good results at a plane wave cutoff
> of 25ry.


I thought the best way to get a 'good' cutoff value is to try several ones and get use the one that gives convergence (more or less)?

 

Thanks in advance,
Christopher
Undergrad. Student
University of Antwerp

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://cpmd.org/pipermail/cpmd-list/attachments/20051207/eea0d3e0/attachment.html