[Wien] band in Hybrid functunal sumhfpara'
tran at theochem.tuwien.ac.at
tran at theochem.tuwien.ac.at
Wed Jan 8 15:55:44 CET 2014
Hi,
Is it really "sumhfpara" which is not finishing? If this is the case
I don't understand why. Can you send me the files :log and case.dayfile.
Beside this, there is a problem that I am aware of since very recently:
In this article
http://prb.aps.org/abstract/PRB/v87/i16/e165122
it is shown that the method used in WIEN2k to treat the Coulomb
singularity in the unscreened Hartree-Fock exchange (the spherical
truncated method) does not work for cells whose lattice constants are very
different from each other (1 or 2 dimensional systems are the extrem
cases). Look at the green "x" in Figs. 7 and 8 in this article.
So, I suspect that this problem concerns your calculation on
nanoribbon. In this case it would be better to use a screened hybrid
functional for which there is no problem.
One last thing: Is it necessary to use a 16x16x1 k-mesh for the scf
calculations? I would say that less k-points is probably enough
(and less expensive),
F. Tran
On Wed, 8 Jan 2014, ali ghafari wrote:
> Dear Tran
> Thank you very much for your advice.
> I finish the calculations by B3LYP on Nanoribbon with 16 16 1 k-points. But for band structure calculation I have used 20 kpoints at
> the in-plane of BZ (i.e. M-->Gamma--->K--->Gamma). now after 2 months the 'sumhfpara' hasn't finished yet! Which is very surprising
> because for SCF it just took about less than one week (on a pc with cpu i7 930). In the following, you will see the case.outputhf_
> which is for scf.
> I think, cpu time for scf should be to much more that cpu time for run_bandplothf_lapw -p. is it right?
> what is your advice?
> Best Regards
> Ali
>
>
> cpu time (in sec.) for calc_h:
> ------------------------------
> total : 3812
> all before loop ik1 : 0 0%
> FFT 1 : 0 0%
> alm, blm, clm : 0 0%
> ftmp3a, ftmp3b, ftmp3c : 17 0%
> FFT 2 : 33 1%
> qlm, tdd : 12 0%
> FFT 3 : 592 16%
> pseudocharge, potential : 552 14%
> FFT 4 : 624 16%
> Hamiltonian in interstitial : 165 4%
> Hamiltonian in spheres 1 : 2 0%
> Hamiltonian in spheres 2 : 139 4%
> diagonalization and other things : 0 0%
>
> wall time (in sec.) for calc_h:
> -------------------------------
> total : 3986
> all before loop ik1 : 0 0%
> FFT 1 : 0 0%
> alm, blm, clm : 0 0%
> ftmp3a, ftmp3b, ftmp3c : 18 0%
> FFT 2 : 34 1%
> qlm, tdd : 13 0%
> FFT 3 : 621 16%
> pseudocharge, potential : 577 14%
> FFT 4 : 653 16%
> Hamiltonian in interstitial : 173 4%
> Hamiltonian in spheres 1 : 2 0%
> Hamiltonian in spheres 2 : 145 4%
> diagonalization and other things : 0 0%
>
> cpu time (in sec.) for hf:
> --------------------------
> total : 3821
> Hamiltonian (calc_h) : 3812 100%
>
> wall time (in sec.) for hf:
> ---------------------------
> total : 3995
> Hamiltonian (calc_h) : 3986 100%
>
>
> On Thursday, November 14, 2013 8:54 AM, "tran at theochem.tuwien.ac.at" <tran at theochem.tuwien.ac.at> wrote:
> So, I suppose these are the FFTs which take most of the time because of the
> large unit cell used for the vacuum. At the end of the files case.outputhf_*
> the timing is shown and the values for "FFT 3" and "FFT 4" should be
> much larger than the others.
>
> On Wed, 13 Nov 2013, ali ghafari wrote:
>
> > both structures have 3 atoms. Furthermore, I have selected the same values
> > in case.inhf for both structures.
> >
> >
> >
> > On Wednesday, November 13, 2013 8:27 PM, "tran at theochem.tuwien.ac.at"
> > <tran at theochem.tuwien.ac.at> wrote:
> > I suppose that it's because there are more atoms in the unit cell for
> > nanoribbon than for bulk. How many atoms do contain these unit cells?
> >
> > In general, since hybrid functionals are very expensive it is
> > recommended to do some tests to search for the lowest values of
> > the parameters in case.inhf which still leads to the desired accuracy.
> >
> > On Wed, 13 Nov 2013, ali ghafari wrote:
> >
> > > Dear Tran
> > > Thank you very much for reply.
> > > I'm using 50 K-point for ploting band structure of ZrSe2 bulk
> > > and nanoribbon. But in the bulk structure "run bandplothf lapw" is more
> > than 100 times faster than nanoribbon. the question is why?
> > > Best Regards
> > > Ali
> > >
> > >
> > > On Wednesday, November 13, 2013 5:00 PM, "tran at theochem.tuwien.ac.at"
> > <tran at theochem.tuwien.ac.at> wrote:
> > > sumhfpara does not run in parallel simply because this is the program
> > > which merges the vector files case.vectorhf_1, case.vectorhf_2,
> > > etc. created by hf on processors 1, 2, etc.
> > > Anyway, sumhfpara is very fast (a few seconds). This is hf which is very
> > > expensive. How many k-points are you using for plotting band structure?
> > > Maybe you should choose less k-points.
> > >
> > > F. Tran
> > >
> > > On Tue, 12 Nov 2013, ali ghafari wrote:
> > >
> > > > Dear Prof. Blaha
> > > > I'm calculating the band structure by hybrid functional B3LYP. for
> > polting band structure as
> > > > discussed on page 53 of UG, we should run "run_bandplothf_lapw -p" which
> > will takes about two weeks
> > > > for a structure with 2 atoms.
> > > > I see,
> > > > lapw1 -band -p
> > > > hf -band -p
> > > > sumhfpara -band -d
> > > >
> > > > my question is: why 'sumhfpara' can not run on parallel?
> > > > Best Regards
> > > > Ali
> > > >
> > > >
> > >
> > > _______________________________________________
> > > Wien mailing list
> > > Wien at zeus.theochem.tuwien.ac.at
> > > http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> > > SEARCH the MAILING-LIST at:
> > http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html
> > >
> > >
> > >
> > >
> >
> > _______________________________________________
> > Wien mailing list
> > Wien at zeus.theochem.tuwien.ac.at
> > http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> > SEARCH the MAILING-LIST at:
> > http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html
> >
> >
> >
> >
>
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> SEARCH the MAILING-LIST at: http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html
>
>
>
>
More information about the Wien
mailing list