The ABINIT Users Mailing List ( CLOSED )

[hiva m <hiva.ms@gmail.com>]

Thanks both of your emails were useful for me. I hope you won't punish yourself more!

On Wed, Feb 11, 2009 at 6:23 AM, matthieu verstraete <matthieu.jean.verstraete@gmail.com> wrote:

I stand humbly corrected by the grand master (I had not checked the type of system in tutorial 3.3):

What I wrote about k-point and ecut convergence is always true, but for insulators (as is the case for Si in 3.3) there is no need to add unoccupied bands for the ground state. You know how many are needed to give the full GS density. For metals this is not the case, and you need to include enough bands to account for the full Fermi surface, and a few more to improve SCF convergence (see the tutorial on Al etc...). Also, because there is no Fermi surface, the k-point convergence will be much faster.

I will recompile and test abinit 3 times for my penance. Oh yeah, I already did that yesterday... Then I'll give up answering the forum, for Lent.

Matthieu

On Sun, Feb 8, 2009 at 8:16 PM, matthieu verstraete <matthieu.jean.verstraete@gmail.com> wrote:

Yes, you should converge nband as well.

As to the criterion for stopping convergence, as is mentioned in several of the tutorials this depends what you are calculating. The correct way to do things is to do a convergence test on the final quantity you are interested in: atomic positions, lattice parameters, GW energy gap, or phonon frequency etc... These will all converge slightly differently wrt nband and ecut, and you should be sure your quantity of interest is converged. In particular, integrated quantities (scalars, like total energy) converge faster than individual matrix elements or kpoint dependent quantities (like the wavefunction or potential matrix elements tested by tolwfr/tolvrs).

Another issue is the monotonic character of the convergence: when you increase ecut or nband, in principle the energy should always decrease. When you increase the kpoint grid, the convergence oscillates around the correct value, depending on the exact positioning of the kpoints in the Brillouin Zone (eg inside or outside the Fermi surface, at the zone edge or not, etc...). This is also described in part of a tutorial on the kpoints.

On Sun, Feb 8, 2009 at 8:00 PM, hiva m <hiva.ms@gmail.com> wrote:

In the part that the convergence tests are done on k-points, lesson 3.3 I didn't get finally what caused the choosing of 4 4 4 for the suitable k-point grid? the difference of 0.0001 Ha in etot?

Yes, this is the chosen criterion.
 
Matthieu

--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Dr. Matthieu Verstraete

European Theoretical Spectroscopy Facility (ETSF) 
Dpto. Fisica de Materiales, 
U. del Pais Vasco,            
Centro Joxe Mari Korta, Av. de Tolosa, 72,   Phone: +34-943018393
E-20018 Donostia-San Sebastian, Spain        Fax  : +34-943018390

Mail : matthieu.jean.verstraete@gmail.com
http://www-users.york.ac.uk/~mjv500

--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Dr. Matthieu Verstraete

European Theoretical Spectroscopy Facility (ETSF) 
Dpto. Fisica de Materiales, 
U. del Pais Vasco,            
Centro Joxe Mari Korta, Av. de Tolosa, 72,   Phone: +34-943018393
E-20018 Donostia-San Sebastian, Spain        Fax  : +34-943018390

Mail : matthieu.jean.verstraete@gmail.com
http://www-users.york.ac.uk/~mjv500