[QE-users] ph.x diverges when restart
Yujia Teng
yujia.teng at physics.rutgers.edu
Mon Feb 9 15:02:49 CET 2026
Dear QE community,
I'm using QE v7.3.1. I found that when restarting from phonon calculation, it often diverges. I'm wondering is there a way to avoid this?
For example, 'thresh' was on the order of 1E-09 at the end of previous job. When restart, it jumps from 1E-09 to 1E-03 or 02. and |ddv_scf|^2 changed from 2E-04 to 2E+01. Due to the walltime limit on HPC cluster, I've set 'max_seconds' tag to ensure ph.x ends cleanly. In my initial run, I already lowered alpha_mix to 0.1 and increased nmix_ph to 10 to speed up convergence.
---------------------------ph.out-------------------------------------------
iter # 19 total cpu time : 417.8 secs av.it.: 22.5
thresh= 1.794E-09 alpha_mix = 0.100 |ddv_scf|^2 = 2.307E-04
iter # 20 total cpu time : 440.0 secs av.it.: 1.0
thresh= 1.519E-03 alpha_mix = 0.100 |ddv_scf|^2 = 2.301E-04
.
.
.
iter # 51 total cpu time : 3263.7 secs av.it.: 20.8
thresh= 1.000E-02 alpha_mix = 0.100 |ddv_scf|^2 = 1.605E+01
iter # 52 total cpu time : 3403.6 secs av.it.: 21.0
thresh= 1.000E-02 alpha_mix = 0.100 |ddv_scf|^2 = 2.770E+01
------------------------------------------------------------------------
---------------------------ph.in-------------------------------------------
phonons of GdAuGe at q=3
&inputph
tr2_ph=1.0d-16,
prefix='GdAuGe',
outdir='./tmp/3',
fildyn='GdAuGe.dyn',
fildvscf='dvscf',
ldisp=.true.,
nq1=6, nq2=6, nq3=3,
start_q=3,
last_q=3,
recover=.true.,
alpha_mix(1) = 0.1,
nmix_ph=10,
max_seconds=13500,
/
------------------------------------------------------------------------
Best,
Yujia
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20260209/ca70a5ee/attachment.html>
More information about the users
mailing list