[gpaw-users] dense k point sampling

[Georg Kastlunger]

Hello,

I have a somehow related question which might also help Sylvain i think.

Im trying to calculate an fcc Au surface with a very dense kpoint mesh. 
To do so I started with a kpts=(4,4,2) calculation to converge the 
density and tried to fix this density (fixmagmom=True in calculator) to 
calculate something like a band structure, as also shown in the tutorial 
on band structure calculations. Unfortunately I always get an error when 
I try to do so, which is related to the domain decomposition done in the 
parallelization process:

File 
"/home/lv70174/georg/gpaw_inst_10428/lib64/python2.6/site-packages/gpaw/paw.py", 
line 542, in initialize
GPAW CLEANUP (node 173): <type 'exceptions.RuntimeError'> occurred.  
Calling MPI_Abort!
RuntimeError: Density reinitialization conflict with 'fixdensity' - 
specify domain decomposition.
Density reinitialization conflict with 'fixdensity' - specify domain 
decomposition.
     raise RuntimeError("Density reinitialization conflict "

In a next step I tried to define the domain decomposition, which was 
used for the inital calculation, by hand (parallel= {'domain': 
(x,y,z)}), which led to a memory overflow in all cases.
Before I now start to use a huge number of cores to get enough memory, I 
wanted to ask if what I am doing is the right thing. I would be very 
grateful for some feedback on and experiences with this.

Best wishes,
Georg



On 11/07/2013 05:09 PM, Kristian Sommer Thygesen wrote:
> Hi
>
> You should not do at sc DFT calculation using 200 k points. It should be possible to specify the k points in the transmission calculator based on an LCAO Hamiltonian obtained from a coarse sampling.
>
> Best regards
> Kristian
> ________________________________________
> From: gpaw-users-bounces at listserv.fysik.dtu.dk [gpaw-users-bounces at listserv.fysik.dtu.dk] on behalf of Sylvain Tricot [sylvain.tricot at univ-rennes1.fr]
> Sent: Thursday, November 07, 2013 4:14 PM
> To: gpaw-users at listserv.fysik.dtu.dk
> Subject: [gpaw-users] dense k point sampling
>
> Dear gpaw-users list,
>
> We are currently computing transmission coefficient maps on Fe(001) (12
> layers including both principal layers and scattering region).
> Calculations are done at no bias voltage. Results are fine for now and
> transmission maps very similar to what is reported in the litterature
> are found.
>
> BUT, very fine structures are expected in the transmission maps. Those
> structures could only be observed by significantly improving the number
> of k points.
> Typical k points meshes found in the litterature are about 200x200x1.
> But I cannot figure how such a dense k-grid could be possible since when
> setting kpts=(24,24,1) in our example, the cluster computer is asking
> for the whole 40GB of ram.
>
> So my questions are:
> Is there an implemented way in GPAW to achieve such dense k-point
> sampling using the transport calculator ?
> Is it possible to split the calculation in smaller parts which would
> compute only a given set of k-points, store them, and then reconstruct
> the whole brillouin zone ?
>
> Thanks in advance for any thoughts and help.
>
> --
> Sylvain Tricot
> Ingénieur de Recherche
> Institut de Physique de Rennes - UMR 6251
> Département Matériaux/Nanosciences
> Université de Rennes I
> Campus de Beaulieu
> Bat 11E
> 35042 Rennes Cedex - France
>
> tel: +33 (0)2 23 23 69 06
>
> _______________________________________________
> gpaw-users mailing list
> gpaw-users at listserv.fysik.dtu.dk
> https://listserv.fysik.dtu.dk/mailman/listinfo/gpaw-users
>
> _______________________________________________
> gpaw-users mailing list
> gpaw-users at listserv.fysik.dtu.dk
> https://listserv.fysik.dtu.dk/mailman/listinfo/gpaw-users
>