The ABINIT Users Mailing List ( CLOSED )

["Anglade Pierre-Matthieu" <anglade@gmail.com>]

Hi,

K-point grid are just classical integration meshes. Just like the mesh
used in one dimension. But they have tree dimension. So instead of one
basis vector you have three. And you define the length between point.
There is two way to define this length. Either in real-space where
fine meshes will have large length or in reciprocal space where fine
meshes have small length between points.

The usual monkhorst pack scheme is to base the k-point mesh on the
axis of your cell. You impose the k-point to be placed at an integer
number of multiple of your real-sapce axis. This way you keep the
necesary periodicity of your cell for your k-poijt mesh. This is what
you do by setting "ngkpt" within Abinit. For instance
ngkpt 2 3 4 means that along x-axis in real space you have a k-point
every two basis vector; every 3 basis vector along y axis ...
Because of the relation between real and reciprocal space this means
that you have two non equivalent k-point position long x-axis, 3 along
y axis ...

Also, It is  possible to base the mesh on a linear combination of the
basis axis. This is what you do with "kptrlatt".

The rules to make a good k-point mesh are  the same as those used to
make a good integration mesh. For inknown functions you want something
where the point are placed as regularly as possible. If they are too
far appart in one direction or the other you may miss some properties
in this direction. That's why you aim at the most homogeneous sampling
of your reciprocal space when you define the k-point mesh.

From this point of view this is a simple geometrical problem.
The option "prtkpt" is doing all the analysis for you.

Obviously, knowing your system and its band properties you will be
able to find better meshes by hand-tuning the k-point grid. For
instance by defining more points  close to the ffermi surface.

regards

PMA


On 9/10/07, m_mousavi1980@yahoo.com <m_mousavi1980@yahoo.com> wrote:
>
>  Hi
>
>  I have problem in chosing the best kpoint.
>  I want to know the meaning of Monkhorst-Pack grids .
>
>  best regard
>


-- 
Pierre-Matthieu Anglade