The ABINIT Users Mailing List ( CLOSED )

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

Dear MTStorr,

M.T.Storr wrote:
> Hi,
>
> I was wondering if anyone could help me with a method to select 
> appropriate k-point grids for a system. I have just got hold of the paper 
> by Monkhorst and Pack but fear it may be a while before things become 
> clear.
> I would like to know how to work out a set of k-point grids to try faced 
> with a particular system. 

A very rough estimation of the TOTAL number of k points in the
Brillouin Zone, for "moderately" converged properties :
for insulators, it will be on the order of 500 divided by the number of atoms 
in the cell
for metals, it will be easily twice or five times larger.

> I am aware that Abinit has an automated procedure to 
> try some grids, but should one just assume that the grid suggested by 
> Abinit is the best one to use, or should one try grids with more k-points 
> aswell.

You should never take a convergence parameter as granted once for all.
After a first, careful convergence study for your property of
interest, please check regularly by using better or less
k points, to see the effect such a change has on the final result(s).

> Further to this once one has a grid for a partciluar system how deos one 
> work out the actual number of k-points in that grid.

ABINIT announce the number of k points it effectively uses (in the
reduced BZ). The total number of k points in the FULL zone is the
product ngkpt(1)*ngkpt(2)*ngkpt(3)*nshiftk . For grid setting using
kptrlatt, use the determinant of kptrlatt instead of the product
of the ngkpt's .

> I apologise if this is a somewhat simple question, but I am struggling 
> with this concept at the moment.

Good continuation,
Xavier Gonze