[DFTB-Plus-User] Error: Failure in diagonalisation routine ZHEEVR, unable to complete Cholesky factorization of B 46

[Lachlan Oberg]

Hello Bálint,


Thanks for your quick reply, your solution works! Could I also ask a question regarding a paragraph in the SpinOrbit section of the manual (section 2.4.3, page 29):


"For periodic systems, use of this keyword automatically prevents the folding by

inversion normally used in SupercellFolding{}, but manually specified KPointsAndWeights should
not be reduced by inversion."

I took this to mean that manual k-point sampling should be used in spin orbit calculations, and this is the reason I used 2x2x2 sampling as it was quick to write out. Is this not a correct interpretation of that paragraph?

Kind regards
Lachlan

________________________________
From: DFTB-Plus-User <dftb-plus-user-bounces at mailman.zfn.uni-bremen.de> on behalf of Bálint Aradi <aradi at uni-bremen.de>
Sent: Monday, May 8, 2017 6:58:13 PM
To: User list for DFTB+ related questions
Subject: Re: [DFTB-Plus-User] Error: Failure in diagonalisation routine ZHEEVR, unable to complete Cholesky factorization of B 46

Dear Lachlan,

> Regardless of the spin coupling constants I enter (with the exception of
> non-zero values for the s orbital), I obtain the following error after
> 10 geometry steps:
>
> -> Failure in diagonalisation routine ZHEEVR, unable to complete
> Cholesky factorization of B     46
>
> Any ideas on what I can do to fix this? For reference, I am using the
> si-band parameter set.

Note, the siband parameter set is not meant for geomertry optimisation
at all. It only contains the electronic parameterisation and a dummy
spline. So, if you do geometry optimisation, it definitely ends up in a
completely unphysical geometry, where the overlap becomes non-positive
definite and therefore the diagonalization fails.

By the way, a side note: You should use a better k-point set as 2x2x2
for the primitive cell. Actually, if you k-point sampling is decent
enough, you should see no forces on the atom, as those cancel out in a
cubic structure...

> Changing to the pbc parameter set, geometry convergence DOES occur.
> However, after sampling along the k-lines L - G - X and using dp_bands,
> I obtain the error:
>
>   File
> "/apps/dftbp/1.3.1/lib/python2.7/site-packages/dptools/bandout.py", line
> 69, in fromfile
>     nkpt = len(eigvalarrays) / nspin
> ZeroDivisionError: integer division or modulo by zero
>
> I assume this means I have to set nspin /= 0 somewhere, but I cannot
> figure out how to do so.

This is a bug, I am sorry for it. The dp_tools package is apparently not
aware about the non-colinear output format. While we are fixing it, you
can convert the band.out file using the following shell-command:

sed 's/KPT[ ]*\([^ ]*\)/KPT \1 SPIN 1/g' band.out > band_fixed.out

If you run dp_tools on band_fixed.out, it should work.

  Best regards,

  Bálint

--
Dr. Bálint Aradi
Bremen Center for Computational Materials Science, University of Bremen
http://www.bccms.uni-bremen.de/cms/people/b-aradi/


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.zfn.uni-bremen.de/pipermail/dftb-plus-user/attachments/20170508/610b1521/attachment.html>