Menu

unable to get convergent SCF

Elk Users
pipidog
2016-11-22
2016-11-24
  • pipidog

    pipidog - 2016-11-22

    Dear all,

    I tried to pefrom DFT+U in the examples folder. It contains several examples such as NiO or FeAl, etc.
    However, only few of them can get convergent results. The energy difference between successive loops
    can be up tp several thousands Ha. Even if I tuned beta0 and betamax to very small values (0.001 and
    0.005) doesn't resolve the issue. The covergence seems worse when there is magnetization.

    I checked the log file, I noticed there are several warning message, like:
    (This calculation involves Eu, so the required total charge is 63)

    Warning(rhonorm): total charge density incorrect for s.c. loop 19
    Calculated : 935.0090463
    Required : 63.00000000

    also, other warning, like:

    Warning(linengy): could not find 3 linearisation energies in s.c. loop 19

    I think the huge difference between the calculated and required total charge makes the problem.
    Also the problem 3 linearisation will lead to unphysical energies.
    Did anyone run magnetic DFT+U case before? Is there any suggestion to make magnetic DFT+U
    problem converge?

    Thanks!

     

    Last edit: pipidog 2016-11-22
  • mfechner

    mfechner - 2016-11-23

    Dear pip,

    Most of my computations are done with dft+u, so I may help you. Could you state first which calculation are not working for you ?

    And moreover general advice for magnetic insulators and convergence of dft+u computations using elk use a big (means > 2-3) initial field in combination with a reducebf of 0.9, helps in most cases. Futher you should use linear mixing !!

    best
    Michael

     
  • martin_frbg

    martin_frbg - 2016-11-23

    Just to confirm - do you see the same extreme behaviour with the provided FeAl and NiO
    examples, or is it primarily your own Eu-containing case that has the problem ? (With such extreme fluctuations I would check the validity of the crystallographic data first: correct spacegroup symmetry,
    lattice constants given in the correct unit, no
    errors in atomic positions

     
  • pipidog

    pipidog - 2016-11-23

    Hello,

    @martin_frbg, @mfechner :

    Yes, the same situations for NiO and FeAl. For NiO, I got the following warning messages:
    (Note: I didn't change even a word in the example. It is exactly the file that comes
    with the installation. )

    Warning(rhonorm): total charge density incorrect for s.c. loop 22
    Calculated : 113261.0667
    Required : 72.00000000

    the calculated total charge is up to 10^5, ridiculous, right? How can it be possible?
    and other warning message like this:

    Warning(linengy): could not find 3 linearisation energies in s.c. loop 21

    As a result, the total energy fluctation is riduculously high (DTOTENERGY.OUT):

    0.1145948069
    2.690398307
    3.995987812
    3.220265166
    2.260247789
    1.659961949
    0.3422696704
    0.9553809841
    0.7653877444E-01
    0.1757704570
    0.1009939176
    0.2070372118
    0.5353451980E-01
    0.9757245653E-03
    0.5220048191E-02
    0.6339468378E-01
    0.6467444852
    446.0472136
    309985.6309
    232228.8903
    218959.7012

    Below are parameters of the NiO-AFM-LDA+U input file:
    (Yes, just copy and paste the built-in example. Nothing changed)

    tasks
    0
    10
    dft+u
    1 2 : dftu,inpdftu
    1 2 0.29399 0.30078 0.18799 : is, l, f0, f2, f4
    spinpol
    .true.
    scale
    7.893389
    avec
    1.0 0.5 0.5
    0.5 1.0 0.5
    0.5 0.5 1.0
    atoms
    2 : nspecies
    'Ni.in' : spfname
    2 : natoms; atpos, bfcmt below
    0.0 0.0 0.0 0.0 0.0 -2.0
    0.5 0.5 0.5 0.0 0.0 2.0
    'O.in' : spfname
    2 : natoms; atpos, bfcmt below
    0.25 0.25 0.25 0.0 0.0 0.0
    0.75 0.75 0.75 0.0 0.0 0.0
    reducebf
    0.5
    nempty
    8
    sppath
    '/home/pipidog/program/elk-4.0.15/species/'
    ngridk
    4 4 4

    So, can anyone suggest me what's wrong? Is there something wrong with the code?

    //========
    Shu-Ting Pi
    Postdoctoral Fellow
    University of California, Davis

     

    Last edit: pipidog 2016-11-23
  • mfechner

    mfechner - 2016-11-24

    Dear Shu-Ting,

    I think you have a problem with your elk verision. I tested again the NiO+DFT+U+FLL example and it runs through without problem. Did you try other computations with your version like Fe in the magnetism example folder. If this also fails you may have a problem with the parallism on you cluster. I would thus suggest compiling elk first without any openmp flags !!! Then rerun the calculation just using mpi for the beginning

    best regards
    Michael

     

Log in to post a comment.