There’s another problem I forgot to mention, which is empty ‘dvscf’ file. But 
‘dvscf’ file is not empty if I split q only or don’t split the calculation.

I found this previous post [Q-e-developers] bug image parallelization in 
PHonon<https://lists.quantum-espresso.org/pipermail/developers/2016-September/001388.html>
 reporting same issue when using image parallelization, but it says works in 
GRID way (split q). So I’m wondering is splitting irr calculation similar to 
the image parallelization, which just doesn’t work for non-empty ‘dvscf’ file?

Looks like I can’t attach file here. I would just copy past part of 
input/output files I attached in my initial email below.

----------------<input.2.16>---------------------
phonons of GdAuGe q=2 irr=16
&inputph
  tr2_ph=1.0d-18,
  prefix='GdAuGe',
  outdir='./tmp/2.16',
  fildyn='GdAuGe.dyn',
  ldisp=.true.,
  nq1=3, nq2=3, nq3=1,
  start_q=2,
  last_q=2,
  start_irr=16,
  last_irr=16,
  lqdir=.true.,
  !recover=.true.,
  alpha_mix(1)=0.1,   ! Added: Much more conservative mixing for stability
  nmix_ph=10,         ! Added: More iterations for the mixing algorithm
/
----------------------------------------------
----------------<output.2.16>---------------
……
Waiting for input...
     Reading input from standard input

     Reading xml data from directory:

     ./tmp/3.14/GdAuGe.save/
……
---------------------------------------------
----------------<ph-collect.in>---------------------
phonons of GdAuGe at Gamma
&inputph
  tr2_ph=1.0d-12,
  prefix='GdAuGe',
  outdir='./tmp',
  fildyn='GdAuGe.dyn',
  fildvscf='dvscf',
  ldisp=.true.,
  nq1=3, nq2=3, nq3=1,
  recover=.true.
/
--------------------------------------------
----------------<ph-collect.out>---------------------
     The potential is recalculated from file :
     ./tmp/_ph0/GdAuGe.q_2/GdAuGe.save/charge-density

     Starting wfcs are  124 atomic +  132 random wfcs
     Checking if some PAW data can be deallocated...

Possibly too few bands at point    1   0.00000   0.00000   0.00000

Possibly too few bands at point    2   0.00000   0.38490   0.00000
--------------------------------------------

发件人: Yujia Teng
发送时间: 2026年1月7日 10:33
收件人: [email protected]
主题: Directory mismatch and too few bands problem in split irr calculation in 
ph.x

Dear QE forum,
I had the following 2 problems related to split irrep in ph.x calculation and 
wondering how to solve them. I’m following the GRID_example.


  1.  There’s always directory mismatch problem of some q and irr, when I split 
irrep. For example, for q=2,irr=1, the tmp dir is ‘./tmp/1.2’, but in its 
output file, it reads other tmp directory, like ‘./tmp/3.7’. I already set 
low_directory_check=.true., but this doesn’t work. Sometimes, it even reports 
namelist error in the input file but there’s no. My current solution is to 
write a check script to find these failed jobs and resubmit them.



  1.  There’s always ‘Too few bands at k point xxx’ warning when I collect the 
result, even if I increased nbnd a lot. My system has 124 wfcs, I increased 
nbnd up to 256 but the warning is still present. But there’s no such warning if 
I split q-point only or don’t split. I did a convergence test w.r.t nbnd on a 
small q-point grid, and in my system the phonon dispersion converges (or same 
as no split result) at nbnd=192, although the warning is still there. I’m not 
sure if the result is still correct for denser parameters.


Is there a way to fully avoid the problems above? I attach some input/outputs 
in zip file here.

Best,
Yujia
Rutgers University.


_______________________________________________________________________________
The Quantum ESPRESSO Foundation stands in solidarity with all civilians 
worldwide who are victims of terrorism, military aggression, and indiscriminate 
warfare.
--------------------------------------------------------------------------------
Quantum ESPRESSO is supported by MaX (www.max-centre.eu)
users mailing list [email protected]
https://lists.quantum-espresso.org/mailman/listinfo/users

Reply via email to