Re: [Kwant] QZ Iteration fails

Hi everyone, I wasn't yet able to work on figuring out what is the origin of the issue, but it's not forgotten. We recently migrated to use an issue tracker, so you can keep up with the update on the investigation over here: https://gitlab.kwant-project.org/kwant/kwant/issues/22. If you have any new observations, do share them. César: the differing numbers of modes is a different issue, see https://gitlab.kwant-project.org/kwant/kwant/issues/23 and should be much easier to fix. Best, Anton On Thu, Aug 6, 2015 at 3:21 PM, cesar nunez wrote:

Hi Anton I have another error with zigzag ribbons. When I run graphene lattice with a complex next-nearest-neighbors (Kane-Mele model). I have the next error for values "t2<0.0005j":

RuntimeError: Numbers of left- and right-propagating modes differ, possibly due to a numerical instability.

If somebody know how fix it, please let me know. Best Regards César

2015-08-05 11:50 GMT-03:00 Anton Akhmerov :

...

Hi Maximilian, César,

Thanks for reporting this, looks quite interesting. I don't have an immediate answer yet, but I'll check it out. Can you try and search for the simplest/smallest system that shows the instability?

Best, Anton

On Wed, Aug 5, 2015 at 4:45 PM, cesar nunez wrote:

...

Dear All

I have the same problem when used the Kane-Mele model. The problem disappears when the next-nearest-neighbor hopping is zero or lower than 0.0005j. Some ideas for fix it?

Best Regards César

2015-08-05 6:08 GMT-03:00 Maximilian Trescher :

...

Dear kwant users and developers,

this is my first post to this list, I'm a PhD student from Berlin. Currently I'm trying to implement a lattice model of a pyrochlore slab in kwant.

I found a (more or less straightforward) way to introduce periodic boundary conditions in the system in the maillinglist archive. This works very nicely as long as I keep the periodicity along a lattice vector.

But I run into problems when I choose different directions: Most of the time all solvers fail with a "LinAlgError: QZ iteration failed to converge in zgges".

I know about this post http://comments.gmane.org/gmane.comp.science.kwant.user/372 but in my case it occurs not only for a band bottom, but for a range of energies.

To give you all the details I prepared a Ipython notebook that you can view here: http://nbviewer.ipython.org/gist/faucon/729050da5e3f2eae1bb8#

(Note that the inclusion of complex next-nearest-neighbor hopping is crucial to produce the error, I tried to create the shortest example which reproduces my problem).

Do you have any idea how to circumvent this problem? Is it somehow related to http://netlib.org/lapack/Errata/vrac/lapack_known_issues.html#QZ this problem in Lapack?

I would be thankful for any hint, Best, Maximilian

-- César Núñez Ramírez Doctor(C) Física U.T.F.S.M.

-- César Núñez Ramírez Doctor(C) Física U.T.F.S.M.