[Pw_forum] graphene on BN calculation

Lorenzo Paulatto paulatto at sissa.it
Wed Dec 16 18:22:28 CET 2009


Replying to message "[Pw_forum] graphene on BN calculation" from mohamed  
sabri majdoub (16/12/09):

> I am not sure that I am doing it the right way. I am using 40 processors  
> and I the calculations look very slow to me.

Well, it's a big system. You can speed it up by using tqr=.true.  
(real-space augmentation charge). There is also an experimental real-space  
algorithm, but I don't know how reliable it is at the moment (you can  
contact baris _at_ sissa.it for additional info).

>           tstress  = .t.

there is no need to print the stress, and it takes time to compute (not  
much)

>     diagonalization = 'cg'

For better performance you should use the default (david). If you have  
some problem with it, you should ask for help on the forum before falling  
back to cg, which is non much more robust and quite a bit slower. Also,  
diagonalization errors are most of the time input errors/bad atomic  
positions.


> K_POINTS automatic
> 30 30 1 1 1 0

this is a huge number of k-points for such a cell, are you sure you need  
all of them? Remember that every time you increase the cell size by N you  
can reduce the number of k-points by the same factor. (e.g. a 2x2x2  
super-cell only need on eighth of the k-points of the original cell).  
Computation time scales linearly with the number of points, e.g. "10 10 1  
1 1 0" will be about 9 times faster.

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/


More information about the Pw_forum mailing list