[Pw_forum] Problem in vc-relax, radial FFT

[Lorenzo Paulatto]

On Fri, 20 Apr 2012 10:09:20 +0200, Ali ALLAM <ali.allam2 at hotmail.com>
wrote:
> mixing_beta = 0.1d0,diagonalization = 'cg',

These 2 parameters are making you waste a lot of time: use the default
values unless you have some serious reason to change them.

> I try to decrease ecutwfc for 25 Ry and ecutrho = 200 Ry,

Decrease? Why do you think that making convergence worst would improve? I  
assume you tested the pseudopotentials for convergence with respect to  
ecutwfc and ecutrho. If you did not, do it now: set up a small calculation  
for bulk Nb (BCC, alat~6.23bohr) and Silicon (diamond structure,  
alat~10.20bohr) and increase the cutoff until they converge to 0.001Ry or  
best. Then take the higher cutoff of the 2 calculations (very likely the  
Niobium one).

> and to change the 'smearing' ,

Your smearing is very small, especially compared with a number of k-points
which is not so large. I assume you tested it properly for convergence; if  
you did not take any reasonable volume, set smearing='mv' and degauss=0.01  
or 0.02 and try to increase the number of kpoints until the energy  
converges to ~0.001Ry

> but always i converge the structure with a big difference between  
> cell_parameters.the difference between cell-parameters is about 20%.

Why is this a problem? Maybe your initial guess was not very good, where
did you get it from? Why are you relaxing at constant volume instead of  
letting the volume change?

> So my question is : What is the real problem in my input ?

To be honest, I donnot see which is your problem. Maybe your initial
crystal structure in just not very close to the optimal one. Or your  
calcution is not properly converged.

     		


-- 
Lorenzo Paulatto IdR CNRS @ IMPMC/UPMC - Paris 6
phone: +33 (0)1 44275 084 / skype: paulatz
www:   http://www.impmc.upmc.fr/~paulatto/
mail:  23-24/4é14 Boîte courrier 115, 4 place Jussieu 75252 Paris Cédex 05