Nightly actions failing

[woodsp-ibm]

It appears nightly actions have been failing for the past couple of nights. First is an error, that appears in multiple tests causing them to fail

 _slsqplib.error: Input array gradx must be 1D.

and then a mismatched result compared to expected

Arrays are not almost equal to 2 decimals

Mismatched elements: 5 / 5 (100%)
Max absolute difference among violations: 1.96224533
Max relative difference among violations: 1.96224533
 ACTUAL: array([-0.96,  0.94,  0.88,  0.78,  0.61])
 DESIRED: array([1., 1., 1., 1., 1.])

From an initial quick look it seems scipy 1.16.0 released when the tests started failing (Jun 22). Note the same tests that are presently passing are under Python 3.10. which is still using the earlier 1.15.3 as it seems 1.16.0 is now only for 3.11 and above

[edoaltamura]

I noticed these (and other) errors also in local installs of scipy over the past few days, but no related GH issue on the scipy repo yet.

[woodsp-ibm]

FYI I did this qiskit-community/qiskit-algorithms#234 to sort issues there that arose after scipy 1.16.0 came out. One or two changes, since optimizers were copied here, may be more directly relevant. The issues were all around when using COBYLA where the implementation was changed in 1.16.0 - the PR has quite a few comments that may be of help here,