Elk

Hi everyone! I want to perform a calculation of the phonon spectrum using elk on a 444 mesh, but to perform it one needs a cluster which consists at least of 10-20 processors. So the issue of implementing a MPI compiler appears. As I can see there is only one version of MPI Exciting located at http://lapw.wikidot.com/exct-mpi but it's written there that forces (which of course are needed for phonon spectrum) won't work in parallel. So the question is - are there some other possibilities to run Elk whith some MPI compiler? And a parallel question exclusively for developers of Elk: when (at least approximately) a pseudopotencial version of Elk will be released (where phonon calculations are believed to be faster than in direct method)?

Thanks beforehand, Anton

Hi Anton!

Currently we're working on the "pseudopotential" approach (actually it's morphed into a mixed basis approach, but the efficiency is the same). This will take a few more months, but should be very powerful when finished.

Sometime next year we may also add MPI natively. The reason this hasn't been done until now is that it complicates the code to such a degree, that adding new physics becomes a lot harder (and adding new physics is our main motivation). But the main parts of the code don't change much now between releases, so including MPI may not be too much of an inconvenience.

In the meantime, for phonons you may be better off using ABINIT or PWSCF.

Cheers, Kay.

Hi Kay! Thanks for the info, I also was thinking about trying Abinit. Hope, it would be enough 1-2 machines for it...