[XtalOpt-devel] Xtalopt considers OK runs "failures"

[David Lonie]

On Wed, May 1, 2013 at 4:14 PM, Cohen, Ron <rcohen at carnegiescience.edu> wrote:
> I have been wondering why so many runs seem to fail using pwscf in
> xtalopt, and it is hard to tell because the output files disappear
> after a failed run. However, I caught one and it seems it did not fail
> and converged fine. In the middle of the run there was this warning
> which is why I guess it called it a failure:
>
>      WARNING: info= 41 for Hessian
>
>      WARNING: bfgs curvature condition failed, Theta= 0.922
>      uphill step: resetting bfgs history
>
> It ran much longer after that but ultimately xtalopt rejected it.

Does the output file open in Avogadro? It sounds like a bug in the
OpenBabel parser.

> Is it easy to fix this behavior?

If it's a parsing error, yes. XtalOpt will also parse the xtal.out
file searching for the string "Final" (as in final energy or so), and
if this is not present, it marks the jobs as failure. See
src/xtalopt/optimizers/pwscf.cpp -- more strings can be added to the
m_completionStrings list if there's a better one you come across.

> Also, as a TODO it would be good to have
> the reason for failure logged.
>
> Another useful feature would be a stop button, with an option to kill
> jobs or not. For example I would have to kill this run if I wanted to
> change the code. These runs take weeks.

These won't happen until a new maintainer comes along, but I agree,
they'd be useful features.

Dave