[Pw_forum] vc-relax

[Timothy Mason]

Axel,

thanks for the insight.  That is outstanding information.   I expect many people are running into this problem (I've seen the force oscillation energy decrease phenomenom posted many times).   I am using what I thought were conservative ecuts (40 ryd) but perhaps not big enough for the stress tensor when using USPP.   This characteristic would be a great thing to warn people about in the wiki vc-relax section.   I'd offer to do it myself but I'd feel better leaving it to an expert.  I'll report back if this improves my calculation but with the higher ecuts it won't be for a while.   Is the issue confined to USPP?   would you recommend NCPP for VC-relax as a work around?




Tim Mason
University of Missouri St. Louis






On Sat, Jan 24, 2009 at 12:12 PM, Axel Kohlmeyer <akohlmey at cmm.chem.upenn.edu> wrote:

    On Sat, 24 Jan 2009, Timothy Mason wrote:

    TM> Dear Hania,
    TM>

    TM> as someone else struggling with VC-relax, I'd like to see a grep of
    TM> your cell volumes.  (just grep on volume).  I find that in these
    TM> cases where the forces oscillate but the engergies decrease (all
    TM> cases for me) the volumes continue to expand about linearly with
    TM> time.  I'm not sure what to make of it.  It even happens starting

    good point. that could very likely mean, that your stress
    tensor is not sufficiently converged wrt. to the wavefunction
    or density cutoff. one of the most problematic issues with
    vc-relax calculations is that the stress tensor need a much
    higher wavefunction cutoff (and very much highe density cutoff
    in case of ultra-soft pseudos) to be well converged. to make
    matters worse, in a vc-relax calculations, you do not keep
    those cutoffs constant, but the number of plane waves. so for
    an expanding cell you effectively decrease the cutoffs, and
    if you stress tensor is not well converged, it will become
    quickly not converged. there are some tricks to make this
    effect less of a problem, but the immediate solution is to
    start at a lattice constant that is definitely too large,
    i.e. the cell will never expand beyond it. you still have
    consider pulay stress, but that is relatively minor concern.

    TM> with experimental structures often increasing them by 50 percent
    TM> before convergence thresholds are met.

    a 50% increase sounds unphysical. have you tried restarting
    from the final configuration (resets the "effective" basis
    set size) and see if the results are consistent?

    cheers,
      axel.