[Wien] Problem Energy Sep LAPW2 ACTUALLY MKL PROBLEM

Laurence Marks L-marks at northwestern.edu
Fri Jan 8 15:30:20 CET 2010


Can you please provide a way of testing/documenting this. I might have
it (or might not), and more information would be useful particularly a
way to reproduce it.

On Fri, Dec 11, 2009 at 6:45 AM, David Tompsett <dat36 at cam.ac.uk> wrote:
> Dear All,
>
> I recently wrote to the list facing problems with strange eigenvalues
> leading to errors in lapw2. I have solved the issue and thought I would
> write to the list as the actual cause is due to errors in MKL 10.1 and early
> 10.2 documented at:
>
> http://software.intel.com/en-us/articles/dgemm-and-sgemm-accuracy/
>
> I am not sure if these problems have been mentioned on the list before, but
> may affect anyone who has built Wien2k with any of the affected versions of
> Intel MKL. The errors only occur when the system size is such that the
> matrix sizes in lapw1 satisfy the criteria listed at the link above. Thus it
> may appear that one's code is working well and then odd errors will occur
> for particular cases.
>
> After rebuilding my Wien2k with either old MKL 9.1 or the latest 10.2.2.025
> the problems disappeared.
>
> Best wishes,
> David Tompsett.
>
> --
> David A. Tompsett
> Quantum Matter Group
> Cavendish Laboratory
> J. J. Thomson Avenue
> Cambridge CB3 0HE
> U.K.
> Tel: +44 7907 566351 (mobile)
> Fax: +44 1223 768140
> http://www-qm.phy.cam.ac.uk/
>
>
> Peter Blaha wrote:
>>
>> You said, this happens already in the first iteration ?
>>
>> What are your RMT values of all atoms ?   RKmax ?
>>
>> Are you sure, your structure is correct ?
>>
>> The message 15 eigenvalues below -9. indicates that much more is incorrect
>> than just the energy parameters of atom 24.
>>
>> I don't think you should put the E-parameters to -8 Ry. Probably you do
>> not expect eigenvalues in this range.
>>
>> Hard to guess what you could have made wrong.
>>
>> David Tompsett schrieb:
>>>
>>> Dear Peter and All,
>>>
>>> Thank you for the help. From the case.output2 files it seems that the
>>> problem is with very low lying p-states:
>>>  QTL-B VALUE .EQ. 1876.88467   in Band of energy   -8.53431   ATOM=   24
>>>   L=  1
>>>   Check for ghostbands or EIGENVALUES BELOW XX messages
>>>   Adjust your Energy-parameters or use -in1new switch, check RMTs  !!!
>>>
>>> My linearization energies for this atom are:
>>> 0.60    5  0      (GLOBAL E-PARAMETER WITH n OTHER CHOICES, global
>>> APW/LAPW)
>>> 0    0.60      0.000 CONT 1
>>> 0   -4.00      0.005 STOP 1
>>> 1   -2.21      0.010 CONT 1
>>> 1    0.60      0.000 CONT 1
>>> 2    0.60      0.010 CONT 1
>>>
>>> My first thought is to alter -2.21 to -8.00 for the p-state LO, but I am
>>> concerned that since -8.53431 is so low that somehow my calculation window
>>> is including multiple p-states ie. more than two sets of p-states.
>>> Also, the case.output1 file says that there are:
>>>          15 EIGENVALUES BELOW THE ENERGY   -9.00000
>>>      ********************************************************
>>>
>>> Please advise me as to how to proceed.
>>>
>>> Thank you,
>>> David.
>>>
>>> Peter Blaha wrote:
>>>>
>>>> You are doing k-parallel calc !
>>>>
>>>> What says     lse   ?
>>>> Check the individual case.output2_xx files !
>>>>
>>>> David Tompsett schrieb:
>>>>>
>>>>> Dear Peter and All,
>>>>>
>>>>> the end of my case.output2 looks as follows and confirms that there is
>>>>> some error in finding the separation of semi-core and valence states, though
>>>>> from the userguide it seems that this should not be a major issue without
>>>>> using -in1new.
>>>>>
>>>>> eseper below EF  0.500000000000000    eseper minimum gap
>>>>>  5.000000000000000E-002
>>>>> Energy to separate semicore and valence states:  -999.000000000000
>>>>>  :NOE  : NUMBER OF ELECTRONS          =1313.000
>>>>>
>>>>> :FER  : F E R M I - ENERGY(TETRAH.M.)=   0.94199
>>>>> NbFe2Super222NM.weigh
>>>>>
>>>>>
>>>>>
>>>>>                                8
>>>>> writing
>>>>> NbFe2Super222NM.weigh_1
>>>>>
>>>>>
>>>>>
>>>>>                    writing
>>>>> NbFe2Super222NM.weigh_2
>>>>>
>>>>>
>>>>>
>>>>>                    writing
>>>>> NbFe2Super222NM.weigh_3
>>>>>
>>>>>
>>>>>
>>>>>                    writing
>>>>> NbFe2Super222NM.weigh_4
>>>>>
>>>>>
>>>>>
>>>>>                    writing
>>>>> NbFe2Super222NM.weigh_5
>>>>>
>>>>>
>>>>>
>>>>>                    writing
>>>>> NbFe2Super222NM.weigh_6
>>>>>
>>>>>
>>>>>
>>>>>                    writing
>>>>> NbFe2Super222NM.weigh_7
>>>>>
>>>>>
>>>>>
>>>>>                    writing
>>>>> NbFe2Super222NM.weigh_8
>>>>>
>>>>>
>>>>>
>>>>>                    NK           8
>>>>>
>>>>> Otherwise the file seems OK and I copy the rest of it below (excluding
>>>>> the KVEC list). The error is occurring on the very first SCF cycle each
>>>>> time.
>>>>>
>>>>> Many thanks for any more help,
>>>>> David.
>>>>>
>>>>
>>>
>> _______________________________________________
>> 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
>



-- 
Laurence Marks
Department of Materials Science and Engineering
MSE Rm 2036 Cook Hall
2220 N Campus Drive
Northwestern University
Evanston, IL 60208, USA
Tel: (847) 491-3996 Fax: (847) 491-7820
email: L-marks at northwestern dot edu
Web: www.numis.northwestern.edu
Chair, Commission on Electron Crystallography of IUCR
www.numis.northwestern.edu/
Electron crystallography is the branch of science that uses electron
scattering and imaging to study the structure of matter.


More information about the Wien mailing list