iteration goes on after total energy converge

Elk Users
Hongbin
2013-05-29
2013-06-11
  • Hongbin

    Hongbin - 2013-05-29

    Dear ELK users,

    For one of my calculations, the code is keeping running even the self-consistency has
    achieved. Here is the variation of total energies from the INFO.OUT for the last tens
    iterations before I killed the job:

    …..
    …..
    Absolute change in total energy (target)   :   0.1532493625E-03 (  0.1000000000E-03)
    Absolute change in total energy (target)   :   0.2901032567E-03 (  0.1000000000E-03)
    Absolute change in total energy (target)   :   0.8126953617E-06 (  0.1000000000E-03)
    Absolute change in total energy (target)   :   0.5617889110E-04 (  0.1000000000E-03)
    Absolute change in total energy (target)   :   0.5063213757E-04 (  0.1000000000E-03)
    Absolute change in total energy (target)   :   0.8587929187E-05 (  0.1000000000E-03)
    Absolute change in total energy (target)   :   0.2882082481E-04 (  0.1000000000E-03)
    Absolute change in total energy (target)   :   0.1258071279E-04 (  0.1000000000E-03)
    Absolute change in total energy (target)   :   0.1337088179E-05 (  0.1000000000E-03)
    Absolute change in total energy (target)   :   0.1337405411E-04 (  0.1000000000E-03)
    Absolute change in total energy (target)   :   0.2738874173E-05 (  0.1000000000E-03)
    Absolute change in total energy (target)   :   0.1956016058E-04 (  0.1000000000E-03)
    Absolute change in total energy (target)   :   0.6740447134E-06 (  0.1000000000E-03)
    Absolute change in total energy (target)   :   0.3041911987E-04 (  0.1000000000E-03)
    Absolute change in total energy (target)   :   0.2724811202E-04 (  0.1000000000E-03)
    Absolute change in total energy (target)   :   0.1746026101E-04 (  0.1000000000E-03)

    It seems that all the other quantities are also converged, e.g., magnetic moments.
    Could you please tell me how to fix it?

    Thank you very much!

    Regards

    Hongbin

     
  • Markus

    Markus - 2013-05-29

    Hi Hongbin,

    I guess the code is waiting for convergence of the potential RMS difference. Grep INFO.OUT for RMS. You can change the convergence criteria with epspot and epsengy.

    Markus

     
  • Hongbin

    Hongbin - 2013-05-29

    Hi Markus,

    You are right! Here is the last few output of the variation of the effective potential:

    RMS change in effective potential (target) :   0.3380620440E-05 (  0.1000000000E-05)
    RMS change in effective potential (target) :   0.1695786469E-05 (  0.1000000000E-05)
    RMS change in effective potential (target) :   0.2038628452E-05 (  0.1000000000E-05)
    RMS change in effective potential (target) :   0.3877500202E-05 (  0.1000000000E-05)
    RMS change in effective potential (target) :   0.1209726682E-05 (  0.1000000000E-05)
    RMS change in effective potential (target) :   0.3685530669E-05 (  0.1000000000E-05)
    RMS change in effective potential (target) :   0.1180698804E-05 (  0.1000000000E-05)
    RMS change in effective potential (target) :   0.3390023538E-05 (  0.1000000000E-05)

    It never hits the target though it gets very close to it.

    So, which value should I use as the criteria for convergence in general for this quantity?

    I was a little bit worried about the total energy convergence since the default convergence distance
    is 10^-4 Hartree. If I would like to calculate any quantity which is in the meV range, such as magnetocrystalline
    anisotropy and exchange constant by mapping, I guess one has to use much smaller value for the
    total energies, right?

    Thank you!

    Regards,

    Hongbin

     
  • Markus

    Markus - 2013-06-03

    Hi Hongbin,

    you are right, for MAE you need to do better than the default convergence criteria. I found MAE calculations extremely difficult to converge and finally I skipped this stuff. I've played with the mixer settings but never got really good convergence. I can not tell you why elk has these difficulties.

    Markus

     

Log in to post a comment.