Menu

Difficulty converging for Yb

kanato
2008-04-26
2013-06-05
  • kanato

    kanato - 2008-04-26

    Hi, I am trying to calculate the Pressure-Volume for Yb in GGA.  I am having lots of issues with the total energy curve vs. volume being not smooth, especially around the experimental equilibrium volume. 

    I get several messages about "valence eigenvalues below evalmin" and "not enough empty states."  I have played around with both the evalmin setting and the nempty setting but I haven't had much luck getting good results. 

    Sometimes too the code takes a very long time to complete a single iteration, while the same input file with a different "scale" parameter runs very quickly.

    My settings are:
    rgkmax  11
    sctype  20
    spinorb .true.
    lmaxapw 14
    gmaxvr 14
    ngridk  24 24 24
    nempty 15
    evalmin -15

    So my questions are:

    1) Are there special basis considerations that I have not addressed here to deal with f-orbital systems?

    2) Is there some way to tell what setting I should use for evalmin?  I have not changed it to, say, -1000 because I do not know how it will affect the computation time.  (I feel the evalmin issue is also the cause of the nempty issue, but I may be wrong here.)

    3) Is the default smearing adequate for this sort of calculation with 24^3 k-points, or do I need to change it to get started?  I would normally do a convergence study wrt the smearing, but I need to get the SCF cycle to finish first.

    Note:
    The point here is not to get the "right" answer, because I know LDA or GGA will not be accurate in the equation of state for f-orbitals.  The point here is to calculate GGA equation of state to compare with other results I have from LDA+DMFT.

     
    • kanato

      kanato - 2008-04-26

      Oh I forgot to mention... I have played around with the beta0 and betamax parameters, this helps convergence but I get a "bumpy" energy vs. volume curve.

       
    • ckoe

      ckoe - 2008-04-28

      Dear kanato,

      does the problem with evalmin go away when you (strongly) reduce rgkmax ?
      If so, this might indicate that gmaxvr is too small. IMHO rgkmax=11 is a
      hugh value, and you might run into overcompleteness issues anyway.

      Regarding the bumpy energy vs. volume curve: If beta0 and/or betamax are
      too small I would expect that you might run into pseudo-convergence as
      with usual simple mixing schemes. That could cause bumps.

      On the other hand there seem to be a number of issues generally connected
      to the use of plane-wave basis sets with changing cell size. If I recall
      right the number of plane waves and the quality of IBZ integration
      changes due to the changed cell, but I have no first hand experience with
      this. You might want to check FAQs/Manuals of usual PW pseudopotential
      codes.

      Regards

      Christof

       
    • kanato

      kanato - 2008-04-29

      Christof,

      Thanks for the suggestion.  I think it was as you suspected, that gmaxvr was too small.  I have reduced rgkmax some and increased gmaxvr and now my calculations converge.

      I am still having problems with bumpy energy vs. volume however.  I have betamax = 0.7 and beta0 = 0.07 (the calculation will not converge for defaults) and the energy curve is still bumpy around the equilibrium volume.  I have decreased epspot to 1e-7 that that seems to have helped some.  I will try lowering it some more to see if that's the issue.  If you or anyone else has any other suggestions for improving it, I would be interested.

      I have some experience with total energy calculations vs. volume with PW/pseudopotential codes.  In my experience it is best to keep the same parameters for k-points and basis set over the entire range of volumes.  Making changes to these will cause small shifts in the total energy curves, which will create artifacts at the volume where the parameters are changed when the E vs. V curve is differentiated to get pressure.  So I think it is best to choose a single set of parameters that are "good enough" for the whole calculation range; then you can follow the results through the entire range without having to worry about artificial changes to total energy due to changing the basis set cutoff or BZ integration.

      Of course, the above is based on PW/pseudopotential methods where the energy cutoff for the basis set is determined by the pseudopotential core, which is insensitive to volume changes.  The LAPW method treats atomic cores differently, so I do not have any intuition as to whether or not extra care needs to be taken for pressure calculations with this basis set.

       

Log in to post a comment.