[Wien] SECLR4 error in large supercells / high number of k points

[Kurt Lejaeghere]

I have tried out all of your ideas:

> Assuming that there is nothing wrong with your LAPACK (sometimes there
> is), normally I run across this type of problem when something has
> gone wrong with the potential due to a bad initial density (from
> dstart)

The problem mostly occurs only after a few iterations and at a charge  
distance :DIS that is already reasonably small ( < 0,01). Sometimes  
the calculations work without a glitch, but very small changes (e.g. a  
change in the internal positions) can lead to a SECLR4 error.

> a ghost-band that is not quite big enough to crash lapw2 (but
> gives a warning)

Sometimes warnings occur in this series of calculations, because of a  
small QTL-B error (2 to 3). However not all cases with a warning give  
rise to a SECLR4 error and not all cases with a SECLR4 error cause  
such a warning.

> -- In case.scf0 -- do the fits (at the end) and the VZER terms look
> reasonable compared to what you have for a smaller case?

As far as I can see they do. :FIT is constantly of the order of  
magnitude of 1E-2 or 1E-3 and the :VZER are, both for a problematic  
and a correct case, either simultaneously zero or of the same order of  
magnitude.

> -- If it ran for a cycle or so, look in case.scfm (also in case.scf)
> and see if there are very large differences in the QMT's, indicating
> that you are far from a solution.

If you mean the values of :BIG, they are smaller than 1E-1.

> -- Similarly look in case.outputm at the core integrals -- they should
> be close to an integer. If not you may need to change the lower energy
> limit because you have semi-core states which should be in the valence
> band but are in the core.

All core integrals are at the most 1E-6 away from an integer value.

> -- Similarly look in case.scf(m) at the :NEC terms to see if you are
> losing electrons.

All terms are at the most 1E-3 away from an integer value, so this  
should be good enough.

Nothing from this list seems to solve the problem. As far as I can see  
there is nothing wrong with the calculations, since similar cells with  
less (but still enough) k points don't cause SECLR4 errors.

With kind regards

Kurt Lejaeghere
Ghent University