Dear Prof. Blaha, dear All,
Thank you for your comments!
/SCRATCH directories on the older cluster that I am using are usually
quite full (in some cases perhaps can be solved by writing some emails
asking people to delete old stuff). New harddisk solution is considered,
hopefully will be possible. In any case it's always good to have a
solution to use less space.
In the meantime (after SCF and doing save_lapw) I reduced emin/emax
ranges in in1c and inso and managed to calculate 100x100 mesh overnight
on the /SCRATCH space that was available.
Best,
Lukasz
On 2018-08-19 07:16, Peter Blaha wrote:
By choosing a local scratch directory (like /scratch or /tmp or ...,
this depends on your computers, you can then add -scratch /scratch to
all wien2k run or x commands) and using massively k-parallelism (on as
many different nodes as possible, you can distribute these large
case.vector_xx files on many computers. (PS: otherwise a 2 or 4TB
harddisk costs less than 100 $ !!??).
The case.vectorso_xx files are not needed for fermi surfaces (but of
course they are needed for scf or optics, or ...) and can be reduced
in size by selecting emin/emax in case.inso very narrow.
Another simple trick to save disk space would be:
Start x lapw1 -p -band -up/dn and kill this after a few seconds. Do
the same with lapwso.
It will prepare a couple of uplapw1_xx.def and case.klist_xx files.
Now you run in a small loop (csh syntax):
foreach i (1 2 3 4)
lapw1c uplapw1_$i.def # or lapw1 , depending on in1 or in1c)
lapw1c dnlapw1_$i.def
lapwso uplapwso_$i.def
rm case.vectorup_$i case.vectordn_$i
end
This should produce only one vector file at the time and by choosing
many cores, you can make them arbitrarily small.
Of course, this does not work when you need the vector files for other
programs (lapw2, optics, ...)
Am 18.08.2018 um 20:18 schrieb pluto:
Dear Gavin,
Thank you for your comments. This is what I was afraid.
I can glue output files by myself, no problem, hopefully they don't
change the order of k-points. But one way to solve the disk-space
issue would be to write a script that calculates SOC eigenvalues
k-point by k-point. Then there would be no large vector files.
I am not feeling competent to modify and recompile WIEN2k, but perhaps
I am able to solve it by the script.
Best,
Lukasz
On 8/18/2018 3:43 PM, Gavin Abo wrote:
From WIEN2k 18.2 UG page 126:
WFFIL: standard option, writes wave functions to file case.vector
(needed in lapw2)
SUPWF: suppresses wave function calculation (faster for testing
eigenvalues only)
WFPRI: prints eigenvectors to case.output1 and writes case.vector
(produces long outputs!)
With SUPWF from above, it looks like case.vector* files are not
written by lapw1 unless WFFIL or WFPRI are used instead. Since Table
4.3 shows case.vector* files are required for lapwso, it may be that
you cannot workaround not using them.
As far as I know, there is no built in WIEN2k program to combine the
.output* files. That is why the multiple recent and past posts about
it, such as:
https://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg17847.html
https://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg17829.html
https://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg00508.html
https://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg02057.html
https://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg02055.html
(awk)
http://www.democritos.it/pipermail/xcrysden/2015-September/001731.html
http://www.democritos.it/pipermail/xcrysden/2015-December/001788.html
Though, I don't see why you couldn't automate it yourself either
using a job script (e.g., by adding the "cat case.outputso_1* >
case.output1up" to it), by writing your own script (e.g., bash script
containing your run[sp]_lapw and "cat case.outputso_1* >
case.output1up" commands), or by modifying the WIEN2k source code.
On 8/18/2018 5:07 AM, pluto wrote:
Dear All, dear Prof. Blaha,
I used SUPWF in case.in1c to limit the size of output files.
Then I used:
x lapw1 -band -up -p -c
x lapw1 -band -dn -p -c
x lapwso -up -p -c
There are no errors, but lapwso does not seem to calculate
eigenvalues.
I use 4 parallel cores, so I have 4 output files. Output files
without SOC look OK, they contain eigenvalues (files
case.output1up_1 to 4, same for dn). But SOC output files
(outputso_1 to 4) do not contain any eigenvalues, and I don't see
any other output files.
Am I doing something wrong? Manual in Table 4.3 says that lapwso
needs vector files, which are of course missing here. Is there a way
out?
Also, is there a way to automatically "glue" output files to have
just one large output file?
Best,
Lukasz
___
Wien mailing list
Wien@zeus.theochem.tuwien.ac.at
http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
SEARCH the MAILING-LIST at:
http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/
