power-gmres failed to find solution in 1000 iterations.

[Nicolas Gevers]

Hi all,

I am having one failed test while running "qutip.testing.run()".

It states "FAILED tests/test_steadystate.py::test_driven_cavity_power_gmres - Exception: power-gmres failed to find solution in 1000 iterations."

Any idea how I could solve it? Is this something I should be concerned about?

QuTiP Version:      4.6.2
Numpy Version:      1.21.2
Scipy Version:      1.7.1
Cython Version:     0.29.25
Matplotlib Version: 3.5.0
Python Version:     3.9.5
Number of CPUs:     4
BLAS Info:          INTEL MKL
OPENMP Installed:   False
INTEL MKL Ext:      False
Platform Info:      Windows (AMD64)
Installation path:  C:\ProgramData\Miniconda3\lib\site-packages\qutip

[Simon Cross]

Hi Nicolas,

Is this the same bug report as
https://github.com/qutip/qutip/issues/1750? If so, let's continue the
discussion there.

Given that you're using MKL Blas and numpy 1.21.2 there is a chance
that you're encountering this numpy bug
https://github.com/numpy/numpy/issues/20356. It only occurs on certain
Intel CPUs (those with AVX512 CPU extensions). I believe it is fixed
in numpy 1.22, but I haven't had a chance to confirm that yet.

A quick solution (if you installed with conda, which I assume you did
given that you have the MKL BLAS) is to try "conda install nomkl
qutip" and confirm that that installs a no-MKL Blas and that the tests
pass afterwards. It's also a bit odd that you have INTEL MKL BLAS but
no INTEL MKL Ext. Not sure if that is an issue or not, but it probably
means you won't lose anything but not using the MKL BLAS.

Thank you for reporting the bug and let me know if the suggestion works or not.

Regards,
Simon

Another option is to try numpy 1.22, but I'm not 100% sure how much
work that is (might be easy, might not).

I did try this