The ABINIT Users Mailing List ( CLOSED )

[Xavier Gonze <gonze@pcpm.ucl.ac.be>]

Dear Joe,

On 30 Dec 2005, at 20:54, jzwanzig@dal.ca wrote:

> Dear Colleagues,
>
> When using the "prtkpt = 1" option to view suggestions for the kpt  
> grid, the ending summary includes the value of the "merit_factor".  
> I've read through the source code where this is calculated but am  
> still a little unclear as to how to intrepret it. Does the  
> merit_factor indicate that any grid with a higher merit factor will  
> be more accurate than any grid with a lower merit factor,  
> regardless of the associated values of nkpt or kptrlen,

No

> or does it mean that, for a two grids with similar values of  
> kptrlen, the one with the higher merit factor will be better?

Yes

> Somehow I would have thought that physically kptrlen would be the  
> most important indicator of grid quality,

Yes

> but that some grids, due to symmetry, achieve a given kptrlen more  
> efficiently than others.

Exactly ...

> Any comments on these issues would be helpful. Thanks much--

Well, this analysis is performed at the end of the routine  
testkgrid.F90 .
The merit factor is precisely the analysis of the actual number of k  
points needed for some
kptrlen, with respect to the volume of the sphere based on kptrlen,  
divided by the number of the
symmetries. It is more an auxiliary tool for the understanding of the  
behaviour of the
grids, than a basis on which to select a k point grid ...

Happy new year !
Xavier