about FFTGvecs

Run-time issues concerning Yambo that are not covered in the above forums.

Moderators: Daniele Varsano, andrea marini, Conor Hogan, myrta gruning

about FFTGvecs

Postby yping » Wed Jun 17, 2009 7:19 pm

Hi, developers
I have a question about FFTGvecs to expand the wavefunctions. I found if I used a larger cutoff (20Ry around 11300 plane waves) to do the ground state calculation then I cut the plane waves to 5000 (FFTGvecs=5000) for the expansion of the wavefunctions to compute BSE spectrum, it's better than if I directly use around 10Ry(5600 plane waves) to do ground state calculation then compute the BSE spectrum when I compare these two cases with 20Ry calculation without cuting plane waves for expansion of the wavefunction in BSE calcution. I thought if you use the same cutoff to do ground state calculation then compute the spectra is equivalent to cut the plane waves to do BSE calculation. How to understand this?
yuan ping
University of California, Davis
yping
 
Posts: 22
Joined: Sat Apr 04, 2009 4:35 am

Re: about FFTGvecs

Postby andrea marini » Thu Jun 18, 2009 9:33 am

yping wrote:I found if I used a larger cutoff (20Ry around 11300 plane waves) to do the ground state calculation then I cut the plane waves to 5000 (FFTGvecs=5000) for the expansion of the wavefunctions to compute BSE spectrum, it's better than if I directly use around 10Ry(5600 plane waves) to do ground state calculation


Dear yuan ping,

I am sorry but I do not fully understand your question. What do you mean as it is better ? What is better ? The comparison with the experiment, the speed ... ?

In general the ground state calculation should be done by using an optimal energy cutoff that can be deduced, for example, from the structural properties (in solids one can check the total energy as a function of the lattice constant). This is because using a too small energy cutoff could lead to a wrong spatial dependence of the wavefunctions, even if the energy levels seem unchanged.

When you do excited state calculations, in general, you can reduce the number of G vectors by using FFTGvecs. Also in this case you should check the the final result is not affected by the value of FFTgvecs.

In conclusion, if you see a difference in the Yambo results using the same value of FFTGvecs, but using two sets of core databases obtained from a ground state calculation performed with two different energy cutoffs, I think you should carefully converge your ground state calculation first.

I hope I was helpful

Andrea
Andrea MARINI
Istituto di Struttura della Materia, CNR, (Italy)
User avatar
andrea marini
 
Posts: 447
Joined: Mon Mar 16, 2009 4:27 pm

Re: about FFTGvecs

Postby myrta gruning » Thu Jun 18, 2009 9:36 am

Dear Yuan Ping,

What do you mean by "it' s better"? What is your benchmark? Or are you observing some strange/unphysical feature?
Is your ground state calculation converged with 10 or 20 Ry?

About the cutoff it holds the general rule, that you have to choose the smallest for which your results are converged.
About the FFTGvecs I tend not to change the default value except for the cases where I have problem with the memory and I am forced to lower it. This will in general lead to less accurate results, so one has to check whether and how this is going to affect the results (i.e. the wavefunction normalization) in order to see which value gives sufficient accuracy for your purposes.

The two procedures are not suppose to give the same result in general.
If the GS calculation is not converged at 10 Ry, then this calculation will give e.g. different KS eigenvalues, so starting with one or the other ground state will lead to different results.
On the other hand, by lowering the number of FFTGvecs, you are affecting e.g. the orthonormality of your basis, and this can as well lead to differences.

Cheers,
m
Dr Myrta GrĂ¼ning
Atomistic Simulation Centre
Queen's University Belfast - Northern Ireland

http://www.researcherid.com/rid/B-1515-2009
User avatar
myrta gruning
 
Posts: 303
Joined: Tue Mar 17, 2009 11:38 am

Re: about FFTGvecs

Postby myrta gruning » Thu Jun 18, 2009 9:37 am

Ooops Andrea :?

you had already answered
I did not see it.

Well sorry...

cheers ;)
Dr Myrta GrĂ¼ning
Atomistic Simulation Centre
Queen's University Belfast - Northern Ireland

http://www.researcherid.com/rid/B-1515-2009
User avatar
myrta gruning
 
Posts: 303
Joined: Tue Mar 17, 2009 11:38 am


Return to Other issues

Who is online

Users browsing this forum: No registered users and 1 guest