[Pw_forum] eigenvectors failed to converge in development pw.x

[Peter Scherpelz]

Hi,

I encountered a strange error that I wanted to report in case it affects
current development, in which an scf calculation fails in the nightly
development (#19173) but succeeds in 5.1.2. Both are compiled on a
multi-core linux desktop.

When noncollinear calculation is specified, and a potential from the sg15
database is used ( http://fpmd.ucdavis.edu/qso/potentials/sg15_oncv/ ), in
the newest development version only I get an error right after "starting
wfc are random" is printed:
     Error in routine cdiaghg (7):
     eigenvectors failed to converge

The calculation proceeds normally with a separate NC pseudopotential from
D. Ceresoli's library, and proceeds normally in 5.1.2.
I've created a tarball with stripped-down inputs and outputs that exhibit
the successful and unsuccessful runs, in case that helps:
https://dl.dropboxusercontent.com/u/2715514/qe_conv_test.tar.gz

Best,
Peter
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20150409/9f220cfb/attachment.html>