Possible bug when deltaRhoOut generated

[bhourahine]

As the work array is then not used, so nothing to copy. Would impact excited state use of resulting density matrix.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (91794c2) 70.63% compared to head (57d8281) 70.63%.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1367   +/-   ##
=======================================
  Coverage   70.63%   70.63%           
=======================================
  Files         230      230           
  Lines       43961    43963    +2     
=======================================
+ Hits        31051    31053    +2     
  Misses      12910    12910           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[vanderhe]

Your fix seems fine, although I'm not sure if we should simultaneously store $\mathrm{P}(\mathbf{k})$ and $\Delta\mathrm{P}(\mathbf{k})$ at all. At this point they are of course still identical, but I don't see a reason why we could not store e.g. $\Delta\mathrm{P}(\mathbf{k})$ only and calculate $\mathrm{P}(\mathbf{k})$ on the fly in the linresp routines.

[bhourahine]

Agreed, but for the moment lets just keep things working and then refactor later.

[vanderhe]

@bhourahine As far as I can see this bug only shows up for colinear spin-pol. + excited gradients, but there is no such regression test, e.g. Benzene-TD-LC-Spinpol and Benzene-TD-LC-Spinpol-Iter do not test the gradients (not included in _autotest.tag)

[bhourahine]

@vanderhe suggests we need a regression for this, either by extending those tests or adding a new case. The Casida routines only handle $\Gamma$-point and then only for excitation energies at the moment.