[Wien] convergence problem :: runsp_c_lapw

[Laurence Marks]

It is safe, although it may not be efficient to do this. I would
recommend changing it to 0.2 and doing "echo 0.05 > .msec" to give the
algorithm a slight push if it has stagnated.

N.B., in the case sent pure PRATT mixing was being used for some
reason, so this won't matter.

On Thu, Apr 24, 2014 at 9:25 AM, Luis Ogando <lcodacal at gmail.com> wrote:
> Dear Prof. Marks,
>
>    Just a quick question: if someone reduce the greed to a very low value
> but go back to the default values in case.inm for the last iterations, is it
> safe ?
>    All the best,
>            Luis
>
>
> 2014-04-24 9:15 GMT-03:00 Laurence Marks <L-marks at northwestern.edu>:
>>
>> Why did you change case.inm away from the defaults?
>>
>> a) Set the greed to 0.2 not 0.005. As I have said before, this does
>> not do what you think it does and concepts such as "reduce the mixing
>> factor if it does not converge" are wrong. Yes, wrong. Very wrong.
>> Completely wrong. Always wrong. To understand more please read Journal
>> of Chemical Theory and Computation 9 (2013) 2786.
>>
>> b) Use MSR1, and in extreme cases perhaps MSEC3. Do not use BROYD.
>> Perhaps reduce the greed to 0.1, almost never less than this.
>>
>> A little explanation. The factor of "0.005" is called the mixing
>> greed.in the spirit of algorithm greed (e.g.
>> http://en.wikipedia.org/wiki/Greedy_algorithm). If it is too large the
>> self-consistent iterations can become unstable. However, you rarely
>> need to deal with this and the mixer does. If you set it too a small
>> value (which you did) you are making the algorithm take such tiny
>> steps that it can starve to death. The code "honors" the user setting
>> of a very small value (at least in 12.1) under the assumption that the
>> user knows what he/she is doing.
>>
>> On Thu, Apr 24, 2014 at 5:25 AM, Martin Gmitra <martin.gmitra at gmail.com>
>> wrote:
>> > Dear Wien2k users,
>> >
>> > I am doing calculations of a hexaboride films invoking: runsp_c_lapw -so
>> > -eece
>> > My problem is two fold:
>> >
>> > (i) SCF calculations successfully converged for thin films (9 atomic
>> > units)
>> > and bulk systems. I do have a problem to converge a thicker film (11
>> > atomic
>> > units) even TEMP broadening of 0.02 is used. The observations after 800
>> > iterations are following:
>> >
>> > (1) fermi energy is jumping (roughly) between two values
>> > :FER  : F E R M I - ENERGY(FERMI-SM.)=   0.3682855520
>> > :FER  : F E R M I - ENERGY(FERMI-SM.)=   0.3682855520
>> > :FER  : F E R M I - ENERGY(FERMI-SM.)=  -0.7779431003
>> > :FER  : F E R M I - ENERGY(FERMI-SM.)=  -0.7779431003
>> > :FER  : F E R M I - ENERGY(FERMI-SM.)=   0.3683943150
>> > :FER  : F E R M I - ENERGY(FERMI-SM.)=   0.3683943150
>> > :FER  : F E R M I - ENERGY(FERMI-SM.)=  -0.7767634505
>> > :FER  : F E R M I - ENERGY(FERMI-SM.)=  -0.7767634505
>> >
>> > (2) charge distance non uniformly oscillates between 0.1 and 0.001
>> > :DIS  :  CHARGE DISTANCE       ( 1.1067771 for atom    4 spin 1)
>> > 0.1064425
>> > :DIS  :  CHARGE DISTANCE       ( 0.0059078 for atom    6 spin 1)
>> > 0.0013328
>> > :DIS  :  CHARGE DISTANCE       ( 0.0092322 for atom    6 spin 1)
>> > 0.0015769
>> > :DIS  :  CHARGE DISTANCE       ( 0.0092664 for atom    6 spin 1)
>> > 0.0015106
>> > :DIS  :  CHARGE DISTANCE       ( 0.5478509 for atom    4 spin 1)
>> > 0.0466613
>> > :DIS  :  CHARGE DISTANCE       ( 0.0078766 for atom    6 spin 1)
>> > 0.0014700
>> > :DIS  :  CHARGE DISTANCE       ( 0.9814094 for atom    4 spin 1)
>> > 0.0939173
>> > :DIS  :  CHARGE DISTANCE       ( 0.0070829 for atom    6 spin 1)
>> > 0.0015672
>> > :DIS  :  CHARGE DISTANCE       ( 0.0411319 for atom    4 spin 2)
>> > 0.0053433
>> > :DIS  :  CHARGE DISTANCE       ( 0.0072009 for atom    6 spin 1)
>> > 0.0014308
>> > :DIS  :  CHARGE DISTANCE       ( 0.1357701 for atom    4 spin 2)
>> > 0.0150618
>> >
>> > (3) magnetic moments are zero within spheres whereas there is
>> > :MMTOT: SPIN MAGNETIC MOMENT IN CELL   =    -0.00010
>> > :MMINT: MAGNETIC MOMENT IN INTERSTITIAL =   -0.00003
>> >
>> > (4) my case.inm
>> > BROYD  0.0   YES  (BROYD/PRATT, extra charge (+1 for additional e),
>> > norm)
>> > 0.005           mixing FACTOR for BROYD/PRATT scheme
>> > 1.00  1.00      PW and CLM-scaling factors
>> > 9999  8         idum, HISTORY
>> >
>> > :MIX  :   MSEC3  REGULARIZATION: 2.50E-04  GREED: 0.005  Newton 1.000
>> > with PRATT after 100 iterations
>> > MIX  :   PRATT  REGULARIZATION: 2.50E-04  GREED: 0.005
>> >
>> > Do you have any hints how to converge the calculations?
>> >
>> > (ii) I am running the calculations with version 12.1. The runsp_c_lapw
>> > script has been updated (debugged) in version 13.1 but SCF diverges in
>> > my
>> > cases (thin and thicker films).
>> >
>> > Does anybody experience similar problem?
>> >
>> > Best regards,
>> > Martin Gmitra
>> > Uni Regensburg
>> >
>> >
>> >
>>
>>
>>
>> --
>> Professor Laurence Marks
>> Department of Materials Science and Engineering
>> Northwestern University
>> www.numis.northwestern.edu 1-847-491-3996
>> "Research is to see what everybody else has seen, and to think what
>> nobody else has thought"
>> Albert Szent-Gyorgi
>> _______________________________________________
>> Wien mailing list
>> Wien at zeus.theochem.tuwien.ac.at
>> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>> SEARCH the MAILING-LIST at:
>> http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html
>
>



-- 
Professor Laurence Marks
Department of Materials Science and Engineering
Northwestern University
www.numis.northwestern.edu 1-847-491-3996
"Research is to see what everybody else has seen, and to think what
nobody else has thought"
Albert Szent-Gyorgi