The ABINIT Users Mailing List ( CLOSED )

[paul.boulanger@umontreal.ca]

Dear Yaroslav,

                I work with one dimensional systems (polymers) and I often 
find that 1x1xN is enough convergence. This is not a surprise, since I 
calculate the properties of one isolated polymer strand. To do this in 
ABINIT, I have to choose a cell that will have huge paramaters in the 
directions perpendiculare to the polymer. Hence, a huge cell paramater in 
real space signifies a small reciproqual cell parameter. So my Brillouin zone 
is so small in those direction 1x1xN is enough to sample it. But make a 
convergence test for MxMxN. You should see if it is necessary are ot for your 
system. 

                Like Xavier said, there is no variational behavior for the K 
points and one often observes oscillations in the total energy with respect 
to k point mesh. But, you must be carefull not to forget the precision of the 
calculation. By this I mean that the pseudopotentiel approche already sets an 
error in the calculation, often about 1mHa per atom (depends on the pseudo), 
so convergence of the total energy past this point is a waste of time.   

                    If we take a look to the energies you sared with us, we 
see that your energy is being converged to:

  Energy (Ha)           diff(E-E(1x1x36))
-159,74610701       -0,00531214
-159,74100236   -0,00020749
-159,74076391    0,00003096
-159,74091974   -0,00012487
-159,74082211   -0,00002724
-159,74086755   -0,00007268
-159,74084928   -0,00005441
-159,74079487   

We see that your convergence in total energy is pretty good. I would say 
converged already at 1x1x8. but this is only in total energy. Like Xavier 
warned, you should check this convergence with the properties you want to 
calculate.

paul