[Pw_forum] Band structure calculation

[Lorenzo Paulatto]

In data 05 ottobre 2009 alle ore 17:50:36, Bertrand SITAMTZE  
<siyouber at yahoo.fr> ha scritto:
> I which to calculate the Band structure for an ionic conductor. But it  
> seems as there is something wrong.The structure has 222 electrons and  
> 133 kohn-sham states. The problem is that I dont know if I have to  
> indicate the number of bounds (nbnd) during the band calculation.

Dear Bertrand,
the band structure calculation is a non-self-consistent calculation, where  
the ground-state charge density is used to compute the Hamiltonian  
eigenvalues (i.e. the bands) at the k-points you request. If you don't  
specify the number of *bands* using the nbnd keyword the default value  
will be used, its exact value depend on a few factors which you can find  
in the documentation.

> According to me, I should allow it and the code will consider it as the  
> half of electrons(Which could be very high in this case). But when I do   
> that, the code runs until the line:
>
> "Band structure calculation"
>
> and continue running without writing anything
>
> Please can somebody give me some hint about this?

Maybe the calculation is just slower than you expected! The current  
implementation of non-self-consistent calculation is not extremely  
efficient, although there are plans to improve it. Normally you can expect  
it to take as long as the first self-consistent iteration (not a full  
cycle, just the initial iteration).

If you increase the number of bands the required time will increase with  
its third power. Furthermore it will scale linearly with the number of  
k-points that describe the k-space path on which you want the bands to be  
computed. Finally, the empty bands are diagonalized to a higher precision  
in the band calculation than in the scf calculation, if you have many  
empty bands it will take even longer. Maybe a small increasing of these  
factor is enough to fill your system RAM memory and force the operating  
system to swap to disk, this would decrease performance dramatically!

Please have a look at your setup: how many bands and k-points have you  
requested in the band calculation with respect to the scf calculation?

best regards


-- 
Lorenzo Paulatto
SISSA  &  DEMOCRITOS (Trieste)
phone: +39 040 3787 511
skype: paulatz
www:   http://people.sissa.it/~paulatto/

     *** save italian brains ***
  http://saveitalianbrains.wordpress.com/