Page 1 of 1

possible bug for wannierize=.false. ?

Posted: Tue Dec 06, 2016 4:49 pm
by AgentZero
Dear all,

Recently, I tried to calculate electron-phonon couplings using previously generated MLWFs.
So I set wannierize=.false. and filukk= './XXXX.ukk'. But the code crashes, with last few lines as follows in output (please see file-1).
In the error output file of my computer, it seems to me that epw.x tried to write to 'patterns' file in phsave folder (please see
file-2). And I am sure that I have permission to edit these 'patterns' files.
Could you please give me some advice to fix the problem when using wannierize=.false.? Thanks in advance!

Best wishes,
Miao Gao

---------------------------- file-1 -------------------------
-------------------------------------------------------------------
Using ./XXXX.ukk from disk
-------------------------------------------------------------------


Dipole matrix elements calculated


Calculating kmap and kgmap
Progress kmap: ########################################
Progress kgmap: ########################################
kmaps : 11.47s CPU 14.84s WALL ( 1 calls)
Symmetries of bravais lattice: 24
Symmetries of crystal: 24


===================================================================
irreducible q point # 1
===================================================================

Symmetries of small group of q: 24
in addition sym. q -> -q+G:

Number of q in the star = 1
List of q in the star:
1 0.000000000 0.000000000 0.000000000
Imposing acoustic sum rule on the dynamical matrix


----------------------- file-2 ---------------------------

forrtl: severe (47): write to READONLY file, unit 6, file ./save/XXXX.phsave/patterns.1.xml
Image PC Routine Line Source
epw.x 0000000000B998CD Unknown Unknown Unknown
epw.x 000000000046B773 elphon_shuffle_wr 297 elphon_shuffle_wrap.f90
epw.x 00000000004335B6 MAIN__ 139 epw.f90
epw.x 0000000000432916 Unknown Unknown Unknown
libc.so.6 0000003921A1ED1D Unknown Unknown Unknown
epw.x 0000000000432809 Unknown Unknown Unknown

Re: possible bug for wannierize=.false. ?

Posted: Tue Dec 06, 2016 7:34 pm
by sponce
Hello Miao Gao,

Did you generated the wannier run with a previous version of the code or is it the same?

If its the same, then wannierize=.false. should work.

Can you try removing the filukk= './XXXX.ukk' input variable but keeping wannierize=.false. and try again.

Otherwise, the code does not write on the patterns.1.xml but only reads them.

Are you sure the path to the file is correct? This means that the save folder is in the same directory as your EPW calculation.

I mean, in the directory where you run the EPW calculation, can you do
vim ./save/XXXX.phsave/patterns.1.xml ?

Best,

Samuel

Re: possible bug for wannierize=.false. ?

Posted: Wed Dec 07, 2016 2:35 am
by AgentZero
Dear Samuel,

Thank you very much for your quick response and advice!

Both the previous wannier run and the new calculation with wannierize=false are carried out using EPW v4.0 and qe-5.4.0.
Following your suggestion, I remove the filukk='./XXXX.ukk' in the input file, but the epw.x crashes again with the same errors.
I also checked the path to patterns.1.xml, vim ./save/XXXX.phsave/patterns.1.xml is ok. As you mentioned, I am also confused about writing to READONLY file patterns.1.xml.
Do we need to change other variables, such as epbwrite/read, or epwwrite/read?
Besides XXXX.ukk, is other file needed when setting wannierize=.false.?
Look forward to your reply.

Best wishes,
Miao Gao

Re: possible bug for wannierize=.false. ?

Posted: Wed Dec 07, 2016 7:04 am
by AgentZero
Dear Samuel,

I further test wannierize=.false. with EPW v4.1 and QE-6.0. Now it works fine.

I'm not familiar with the parellel conmmunication inside epw code. I have another issue for your comments.
Is it possible to run epw.x with the number of processors (NPS) larger than the total number of coarse k points (NKP)?
The reason for raising this question is that sometime NKP is minor, especially for compounds with large unit cell. So if NPS
must be equal to NKP, the speed of epw calculation will be slow.

At the beginning, I thought this restriction (NPS=NKP) is imposed by Wannier90. So I split the whole epw run into
a wannier run with small NPS and a subsequent run with wannierize=false and large NPS.
But I also encountered the error message, 'some nodes have no k-points'.

Thanks again!

Best wishes,
Miao Gao

Re: possible bug for wannierize=.false. ?

Posted: Wed Dec 07, 2016 12:37 pm
by sponce
Glad to hear that it work with the new EPW version 4.1

To answer your question, yes it is possible in the new version but only after the XX.epmatwp1 has been written.

Therefore you first need to do a run with NPS=NKP (coarse k mesh) and have epwrite = .true.

You can use small fine k/q grids.

Then you can do a restart using

kmaps = .true.
epbwrite = .false.
epbread = .false.
epwwrite = .false.
epwread = true.

In that case you can use as many processor as FINE k-grid points (so much more NPS).

Cheers,

Samuel

Re: possible bug for wannierize=.false. ?

Posted: Wed Dec 07, 2016 4:57 pm
by AgentZero
Dear Samuel,

Thank you very much! I will try your advice!

Best wishes,
Miao Gao