Dear Isaiah Thank you for reporting.
Part of this bug comes from the lack of the initialization of the tetrahedron
method.
I attached a patch file to fix it. However the recover-mode has another bug.
I tested the calculation of phonon of Al. as
$ cat scf.in
&CONTROL
calculation = 'scf' ,
prefix='al',
pseudo_dir = '../pseudo/',
outdir='./'
/
&SYSTEM
ibrav = 2,
celldm(1) = 7.628216862d0,
nat = 1,
ntyp = 1,
ecutwfc = 40.0 ,
ecutrho = 150 ,
occupations = "smearing"
smearing = "m-p"
degauss = 0.05
/
&ELECTRONS
/
ATOMIC_SPECIES
Al 26.98000 Al.pbe-n-rrkjus_psl.0.1.UPF
ATOMIC_POSITIONS alat
Al 0.000000000 0.000000000 0.000000000
K_POINTS automatic
8 8 8 0 0 0
$ pw.x -in scf.in
$ cat ph1.in
Al Phonon
&INPUTPH
outdir = "./"
prefix = "al"
start_irr = 1
last_irr = 1
recover = .true.
/
0.5 0.5 0.5
$ ph.x -in ph1.in
$ cat ph2.in
Al Phonon
&INPUTPH
outdir = "./"
prefix = "al"
start_irr = 2
last_irr = 2
recover = .true.
/
0.5 0.5 0.5
$ ph.x -in ph2.in
Then I found standard outputs as
Representation # 2 modes # 2 3
Self-consistent Calculation
kpoint 1 ibnd**** solve_linter: root not converged NaN
kpoint 1 ibnd**** solve_linter: root not converged NaN
kpoint 2 ibnd**** solve_linter: root not converged NaN
:
By the way, you can perform successfully the distributed computation as
$ cat ph1.in
:
start_irr = 1
last_irr = 1
recover = .false.
:
$ ph.x -in ph1.in
$ cat ph2.in
:
start_irr = 2
last_irr = 2
recover = .false.
:
$ ph.x -in ph2.in
$ cat ph_collect.in
:
!start_irr =
!last_irr =
recover = .true.
:
$ ph.x -in ph_collect.in
Best regards,
Mitsuaki Kawamura
--
------------------------------------------------------
Dr. Mitsuaki Kawamura
Software Advancement Team
Supercomputer Section
Materials Design and Characterization Laboratory
The Institute for Solid State Physics, Kashiwa, Japan
e-mail : [email protected]
------------------------------------------------------
From: [email protected] [mailto:[email protected]] On Behalf
Of Isaiah Moses
Sent: Tuesday, November 28, 2017 8:53 AM
To: PWSCF Forum <[email protected]>
Subject: Re: [Pw_forum] qe-6.2 Phonon recover=.true. Not Working
More precisely,
For phonon calculation to resume after stopping, the following errors were
obtained:
Representation # 13 mode # 13
Self-consistent Calculation
kpoint 1 ibnd 22 solve_linter: root not converged NaN
kpoint 2 ibnd 22 solve_linter: root not converged NaN
kpoint 3 ibnd 22 solve_linter: root not converged NaN
.........
.........
.........
kpoint 286 ibnd 22 solve_linter: root not converged NaN
kpoint 287 ibnd 22 solve_linter: root not converged NaN
kpoint 288 ibnd 22 solve_linter: root not converged NaN
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Error in routine broyden (3):
factorization
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
stopping ...
And I would have to remove _ph0 directory for the calculation to start afresh
again.
Thanks,
Isaiah
On Mon, Nov 27, 2017 at 8:17 PM, Isaiah Moses <mailto:[email protected]> wrote:
Dear all,
I've tried to split phonon calculation among the irreps, collect the results to
obtain the dynamical matrix using qe-6.2 (optimized tetrahedron method) as I
would usually do with qe-6.1 and qe-5.4.
With the recover=.true., the calculation would not use the results of the
different irreps that had already been computed but instead started all over
the whole calculation again.
I've tried to just stop phonon calculation (without splitting) and resubmit
again and the same thing occurred, the whole calculation started all over.
It seems there is a problem with the recover=.true. option in qe-6.2 release.
Any assistant in getting out of this shall be appreciated.
I appreciate your anticipated response,
Isaiah
--
Isaiah Abu Moses
Graduate Student,
Physics Department,
University of Ibadan,
Nigeria
--
Isaiah Abu Moses
Graduate Student,
Physics Department,
University of Ibadan,
Nigeria
recover.diff
Description: Binary data
_______________________________________________ Pw_forum mailing list [email protected] http://pwscf.org/mailman/listinfo/pw_forum
