[Wien] LAPW0 hangs in mBJ calculation
John McLeod
john.mcleod at usask.ca
Fri Dec 17 05:01:14 CET 2010
Hi Dr. Blaha,
Attached please find the .struct files for PbO and HgO, and the updated
BRJ ("brj.f_new") and original BRJ ("brj.f_original") routines.
In all cases the regular GGA or LDA calculation converges normally.
When the updated BRJ routine is used, LAPW0 in the mBJ cycle (vxc = 28,
50 for .in0 and in0_grr, respectively, PRATT 0.2 mixing in .inm) hangs
for both cases (there are no errors or warnings, and the day file stops
updating). I tried:
- serial and k-point parallelization
- changing the R0 value for Hg (and Pb) in the struct file from
0.00000500 to 0.00000300 and 0.00001000
- spin polarized and non-spin polarized
In all cases, LAPW0 -grr completes but the regular LAPW0 seems to get
stuck in an infinite loop.
With the original BRJ routine, the mBJ calculation converges normally.
Thanks,
John
On 12/16/2010 03:35 AM, Peter Blaha wrote:
> Please send me one of your struct files to my private email for testing.
>
> Am 15.12.2010 09:14, schrieb John McLeod:
>> Hi Dr. Blaha,
>>
>> I did read through the mailing list. As per earlier suggestions, I
>> was using a modified brj.f routine (sent out by Dr. Jalali on October
>> 25th, 2010 in response to problems with
>> the large rho in heavy elements and small rho in light light
>> elements, in this post:
>> http://zeus.theochem.tuwien.ac.at/pipermail/wien/2010-October/013886.html),
>> attached. (I don't
>> know if this is the most recent brj.f routine though.)
>>
>> Anyway it seems that the new (as of 25/10/2010) brj.f is causing the
>> problems. It seems I inadvertently ran the spin-polarized calculation
>> using the old (the one with the original
>> WIEN2k_10.1) brj.f file.
>>
>> Sorry for not checking this before posting to the mailing list.
>>
>> Anyway, now LAPW0 hangs with both HgO and PbO using the newer brj.f
>> routine, while the calculations converge properly with the old brj.f
>> routine - so it seems pretty clear that
>> the problem is with heavy elements.
>>
>> Now I am confused: I thought the updated brj.f routine was supposed
>> to handle heavy elements better than the old brj.f routine. Should I
>> roll back to the original brj.f routine,
>> or is there a more recently updated version I should use?
>>
>> (I did search the mailing list but the most recent post mentioning
>> "brj.f" was from October 30th)
>>
>> Thanks again,
>> John McLeod
>>
>> Ph.D. candidate,
>> Department of Physics and Engineering Physics,
>> University of Saskatchewan
>>
>> On 12/14/2010 11:59 PM, Peter Blaha wrote:
>>> Did you check the mailinglist postings ? Similar reports were made
>>> several weeks ago
>>> and modifications of the mbj routines were posted. Are you using the
>>> modified routines
>>> or the original ones from WIEN2k_10 ??
>>>
>>> Am 14.12.2010 23:28, schrieb John McLeod:
>>>> Hello WIEN2k users,
>>>>
>>>> I am running version 10.1 on a quad core x86_64 architecture with
>>>> Intel 11.1.072 IFC and MKL.
>>>> I was calculating the DOS for a variety of binary oxides, and I ran
>>>> into a problem with HgO.
>>>>
>>>> For the attached .struct file, a GGA calculation (PBE96, with all
>>>> default parameters: -6 Ry cutoff, 1000 k-points, non-spin
>>>> calculation) converges quickly and the results
>>>> look reasonable.
>>>>
>>>> For the mBJ calculation, the single .in0 (NR2V --> R2V) and
>>>> .inm_vresp calculation cycle finishes quickly. For the full mBJ
>>>> cycle (with xc = 28 in .in0, xc = 50 in
>>>> .in0_grr, and 0.2 PRATT mixing in .inm) the first LAPW0 cycle
>>>> (lapw0 -grr) finishes quickly but the main LAPW0 cycle seems to
>>>> freeze (it did not complete after 14 hours).
>>>> Below is the day file. I suppose the `tauwrong' parameter is an
>>>> indication of what the error is.
>>>>
>>>> This problem was persistent after restarting the calculation with
>>>> different parameters (in1new, non-parallel, etc.) and on different
>>>> computer systems. I did not, however,
>>>> try adjusting RMTs, k-points, or the core/semi-core cutoff.
>>>>
>>>> Anyway, when I started from scratch with a spin-polarized
>>>> calculation of HgO (PBE96, all other parameters default) the mBJ
>>>> extension converges after several cycles at 0.5
>>>> PRATT mixing. Looking at the DOS, the spin up and spin down states
>>>> are identical (as I would expect).
>>>>
>>>> A non-spin polarized PBE96+mBJ calculation converges for the
>>>> related oxides CdO, PbO, and Au2O3 (with the appropriate crystal
>>>> structures). If I switch the Hg for Pb in the
>>>> attached struct file (creating a fictitious phase of PbO) the
>>>> non-spin polarized PBE96+mBJ calculation converges. Further, if I
>>>> switch Pb for Hg in litharge-phase PbO
>>>> (creating a fictitious phase of HgO) the calculation converges.
>>>>
>>>> I am curious if anyone has any insight into what is going on here.
>>>>
>>>> I am happy with the spin polarized calculation, but would like to
>>>> figure out why the spin polarized calculation worked while the
>>>> regular calculation did not.
>>>>
>>>> Thanks,
>>>> John McLeod
>>>>
>>>> Ph.D. candidate,
>>>> Department of Physics and Engineering Physics,
>>>> University of Saskatchewan
>>>>
>>>> ---------------- case.dayfile start ----------------
>>>> using WIEN2k_10.1 (Release 7/6/2010) in /usr/local/share/WIEN2k/10.1
>>>> start (Mon Dec 13 11:40:15 CST 2010) with lapw0 (40/99 to go)
>>>>
>>>> cycle 1 (Mon Dec 13 11:40:15 CST 2010) (40/99 to go)
>>>>
>>>> > lapw0 -grr -p (11:40:15) starting parallel lapw0 at Mon Dec 13
>>>> 11:40:15 CST 2010
>>>> -------- .machine0 : processors
>>>> running lapw0 in single mode
>>>> 2.235u 0.262s 0:02.49 100.0% 0+0k 0+8624io 0pf+0w
>>>> > lapw0 -p (11:40:18) starting parallel lapw0 at Mon Dec 13
>>>> 11:40:18 CST 2010
>>>> -------- .machine0 : processors
>>>> running lapw0 in single mode
>>>> int:rho,tauw,grho,g2rho 3906392.62874521 5.552066580361303E+015
>>>> 294540672667.298 -3.693753715914154E+016 tauwrong=
>>>> -1.834378635950119E+016
>>>> [snip: more lines of "int:rho,..." as above]
>>>> int:rho,tauw,grho,g2rho 3906392.62878114 5.552066580175691E+015
>>>> 294540672663.729 -3.693753716133459E+016 tauwrong=
>>>> -1.834378636059656E+016
>>>> ---------------- case.dayfile end ----------------
>>>>
>>>>
>>>>
>>>> _______________________________________________
>>>> Wien mailing list
>>>> Wien at zeus.theochem.tuwien.ac.at
>>>> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>>>
>>
>>
>>
>> _______________________________________________
>> Wien mailing list
>> Wien at zeus.theochem.tuwien.ac.at
>> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: brj.f_new
URL: <http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20101216/90681727/attachment.bat>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: brj.f_original
URL: <http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20101216/90681727/attachment-0001.bat>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: HgO.struct
URL: <http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20101216/90681727/attachment-0002.bat>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: PbO.struct
URL: <http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20101216/90681727/attachment-0003.bat>
More information about the Wien
mailing list