[QE-users] pw_readschema_file failed retrieving input info from xml file ph.x using grid in qe-6.3-bkp and 2D cutoff
Pietro Davide Delugas
pdelugas at sissa.it
Fri Mar 29 09:02:42 CET 2019
Hi
at a quick glance the reduce_io affects only io_level of the band
structure calculation started by run_nscf, so yes no problem in any case
for which band structure calculation is not performed.
I should have neutralized the problem in the delvopment version, now
reduce_io sets to 1 io_level of band computations, the issue is solved,
reduce_io is now kind of useless but doesn't hurt.
If you want to continue doing testing on the subject you are welcome,
but you should use the development version ( and yes update the subject )
thanks and regards - Pietro
On 03/28/2019 09:47 PM, Raphael Longuinhos Monteiro Lobato wrote:
> Hi,
>
> On 3/27/19 12:02 PM, Pietro Davide Delugas wrote:
>> Yes you are right.
>> reduce_io shouldn't be used anymore in ph.x input
>> Pietro
>>
> I'm doing some more tests and will report soon another time ... but in
> short:
>
> It seems that the flag reduce_io leads to error only in single-q,
> not-Gamma calculation. You can use it in single-q, at Gamma, and on
> any q when in dispersion mode, within grid approach).
>
> Should I open another thread (sorry if I'm mixing different subjects)
> ... or reduce_io will be "DEPRECATED" from now, and I should stop
> boring you guys.
>
> Best,
>
>> On 03/26/2019 05:11 PM, Raphael Longuinhos Monteiro Lobato wrote:
>>> Hi,
>>>
>>> On 3/26/19 1:02 PM, Pietro Davide Delugas wrote:
>>>> sorry are you still talking of ph.x with grid as in the beginning
>>>> of the thread or ph.x in general ?
>>> I guess (maybe, I wrong about it) it is for both cases, as the wfcs
>>> and bands computed when ph.x calls pw.x, data necessary for
>>> perturbation, are not saved in disk (in /scratch/_ph0/), nor in RAM
>>> (or maybe, it is in the RAM, but the codes looks for data in disk).
>>>
>>> Best,
>>>>
>>>>
>>>> On 03/26/2019 04:39 PM, Raphael Longuinhos Monteiro Lobato wrote:
>>>>> Hi,
>>>>>
>>>>>
>>>>> On 3/26/19 11:26 AM, Pietro Davide Delugas wrote:
>>>>>> I don't know whether the fact that the restart does not work with
>>>>>> reduce_io = .true. may be considered as a bug or as a feature.
>>>>> But as it is, I guess there is no way to use the reduce_io true.
>>>>> In my case, there is no recover = .true. (a restart) in the input:
>>>>>
>>>>> ph0
>>>>> &inputph
>>>>> prefix='test',
>>>>> reduce_io = .true.
>>>>> trans =.true.
>>>>> outdir='scratch',
>>>>> fildyn='test.dyn',
>>>>> /
>>>>> 0.009443 0.016356 0.000000
>>>>>
>>>>> And the output (main parts, I guess):
>>>>> Program PHONON v.6.4 starts on 26Mar2019 at 9:33: 7
>>>>> Reading data from directory:
>>>>> scratch/acnl1.save/
>>>>> ----2D----2D----2D----2D----2D----2D----2D----2D----2D----2D----2D----2D
>>>>>
>>>>> The code is running with the 2D cutoff
>>>>> Calculation of q = 0.0094430 0.0163560 0.0000000
>>>>> Subspace diagonalization in iterative solution of the
>>>>> eigenvalue problem:
>>>>> a serial algorithm will be used
>>>>> Title:
>>>>> ph0
>>>>> The potential is recalculated from file :
>>>>> scratch/_ph0/acnl1.save/charge-density
>>>>>
>>>>> Starting wfcs are 16 atomic wfcs
>>>>> End of band structure calculation
>>>>> ph0
>>>>> Computing dynamical matrix for
>>>>> q = ( 0.0094430 0.0163560 0.0000000 )
>>>>> Error in routine openfilq (1):
>>>>> file acnl1.wfc not found
>>>>>
>>>>> When commenting the reduce_io, all goes normal; it continues the
>>>>> "Computing dynamical matrix for" and computes the phonon frequencies.
>>>>>
>>>>> Best,
>>>>>>
>>>>>> In the second case we should warn the user about it in the
>>>>>> documentation
>>>>>>
>>>>>>
>>>>>> On 03/26/2019 02:58 PM, Raphael Longuinhos Monteiro Lobato wrote:
>>>>>>> Hi,
>>>>>>>
>>>>>>> On 3/26/19 10:32 AM, Pietro Davide Delugas wrote:
>>>>>>>> sorry but your message is painfully impervious ...
>>>>>>> Sorry ...
>>>>>>>>
>>>>>>>> do I understand correctly that at the moment, the only issue
>>>>>>>> is that at restart ph fails when the reduce_io flag is set to
>>>>>>>> true.
>>>>>>>> Is it right ?
>>>>>>> Yes, because when the reduce_io is true, the code do not save
>>>>>>> the bands and wfcs in the RAM, and the dfpt step ends with
>>>>>>>
>>>>>>> Error in routine openfilq (1):
>>>>>>> file acnl1.wfc not found
>>>>>>>
>>>>>>> After recalculate the bands, in the part
>>>>>>> The potential is recalculated from file :
>>>>>>> scratch/_ph0/acnl1.save/charge-density
>>>>>>>
>>>>>>> there is no:
>>>>>>> Writing output data file acnl1.save/
>>>>>>>
>>>>>>> As there is when the reduce io is false.
>>>>>>>
>>>>>>> Best,
>>>>>>>
>>>>>>>> Pietro
>>>>>>>>
>>>>>>>> On 03/26/2019 02:16 PM, Raphael Longuinhos Monteiro Lobato wrote:
>>>>>>>>> Hi,
>>>>>>>>>
>>>>>>>>> I just compiled the q-e-dev to try the changes from
>>>>>>>>> https://gitlab.com/QEF/q-e/issues/102
>>>>>>>>>
>>>>>>>>> It appears some things are missing (could be may fault). The
>>>>>>>>> scf (wf_collect false, 2D) wrote the wfcs, xml and bands on
>>>>>>>>> scratch. The ph0(reduce_io true, single-q out Gamma) created
>>>>>>>>> the phsave and save in _ph0, and copied only charge.dat to
>>>>>>>>> _ph0/save; no wfcs, nor xml in _ph0.In the phout, there is no
>>>>>>>>> "failed retrieving info from xml", it found the bands in
>>>>>>>>> scratch/save and printed (reading from /scratch/xml? as there
>>>>>>>>> is no xml in _ph0?) it is a 2D calc. It built the pot from
>>>>>>>>> _ph0/save (the charge.dat is there) did the scf, but forgot to
>>>>>>>>> write the wfcs from this scf (and apparently, did not stored
>>>>>>>>> it on RAM) ... going on, the dfpt step could not find the .wfc
>>>>>>>>> (is it a miss behave of the reduce io flag? previously, in a
>>>>>>>>> single calculation with reduce io flag it did not happen). I
>>>>>>>>> repeated the ph0 with the reduce io flag commented, and as
>>>>>>>>> results: the wfc are in _ph0, the xml in _ph0/save, after the
>>>>>>>>> phx calls pwx to calc the bands it wrote it, and the dfpt now
>>>>>>>>> finds the wfcs.
>>>>>>>>>
>>>>>>>>> Best,
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>
>
More information about the users
mailing list