[Wien] (no subject)

[Laurence Marks]

The number of k-points needed depends upon:
a) Whether you are doing a pre-convergence, e.g. MSR1a or a final
convergence.
b) The size of the cell
c) Whether you have a metal or an insulator (partially occupied states at
E_F)
d) The atomic number -- high Z can converge worse -- and whether you have
d/f electrons.

I would estimate that for a metal something like a k-point density of 50
nm^3 is good for reasonable precision. Divide by 2-4 for an insulator,
divide by 2-4 for pre-convergence and perhaps multiply by 2-4 for phonons
or high precision.

So long as memory is not an issue, you should try and divide the k-points
across cores/nodes in an even fashion. It will also depend upon whether mpi
or threading is better on your system, this seems to vary. It is hard to
make clean statements without more information.

For a small problem (e.g. RKM=200-500) I would tend to distribute with 1-2
cores per k-point, e.g. use
1:node1
1:node1
1:node1
1:node1

In .machines. (Add a :2 after the node1 to use mpi.) For a larger problem
(e.g. RKM+12000) you probably want to just run one process. For a very
large problem, e.g. RKM=30000 -- get access to a larger cluster!

Experiment.


On Tue, Mar 12, 2019 at 8:34 AM umbreenrasheed <umbreenrasheed at yahoo.com>
wrote:

> How many k points should be selected for 8 parallel processors selection
> in wein2k..
> What is general relation between the two?
>


-- 
Professor Laurence Marks
"Research is to see what everybody else has seen, and to think what nobody
else has thought", Albert Szent-Gyorgi
www.numis.northwestern.edu ; Corrosion in 4D: MURI4D.numis.northwestern.edu
Partner of the CFW 100% program for gender equity, www.cfw.org/100-percent
Co-Editor, Acta Cryst A
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20190312/2a17fb32/attachment.html>