Additional operator integrals and selection of gauge origin

[frieschneider]

Operator integrals:

• In pyscf.py and veloxchem.py the selection of the gauge origin is possible for gauge-origin dependent operator integrals
• pyscf.py includes additional operator integrals (diamagnetic magnetizability, electric quadrupole (traced, traceless and velocity)

Reference State

• added nuclear quadrupole (gauge_origin dependent)

Testing:

• properties_test.py add gauge origin dependency (transition magnetic dipole moments and rotatory strengths) and test transition_quadrupole_moments
• ReferenceState_refdata_test.py add nuclear quadrupoles
• backends_crossref_test.py add some crossref checks between pyscf and veloxchem to verify gauge origin selection
• pyscf_test.py and veloxchem_test.py extend operators test to gauge origin dependent tests.

ToDo:

• new referencedata (see dump_pyscf.py: dump nuclear quadrupole moments)
• rename nabla to electric_dipole_velocity in DataHfProvider.py and dump_pyscf.py when generating new testdata

[maxscheurer]

Thanks a lot! Some initial comments from a first quick look 😁
Can you please have a look at this, too, @apapapostolou @Drrehn?

[frieschneider]

The selection of the origin is now an argument of the respective property, i.e. state.reference_state.operators.magnetic_dipole('mass_center').

I changed the tests so that I can reproduce some reference data I previously generated. However, I think my tests and how I generated this reference data can still be improved.
(So far I still need to change the generation of the reference data for the new implementation. )

Thus, @apapapostolou , @Drrehn , @maxscheurer maybe we should talk about how I should generate reference data and how to change the tests?

[apapapostolou]

As for the reference data, I think you should not regenerate all the reference data for each gauge origin but only have different operator integrals.

[frieschneider]

Concerning the new tests:

I updated the test and in order to do that I also needed to generate new hf data with pyscf.
I adapted the dump_pyscf.py script (from the submodule adcc-testdata).

You can find the modified file in my HeiBox:
https://heibox.uni-heidelberg.de/d/7d2e9c3c6dbc4f8eb217/

I needed to regenerate new adcc reference data, I changed the file generate_adcc_reference.py, however, this changed file can be found in this pull request.

The new reference files can be generated with the provided scripts. Should I add the updated reference data to the HeiBox as well?

[maxscheurer]

Probably the reference data (adcc ref, that is) need to be re-generated once #175 is merged because that affects the data as well.

[jonasleitner]

I don't really like the idea to return callables just to make mark_excitation_property work. It also leads to (in my opinion) non-intuitive behavior for ExcitedStates.excitation_property_keys since getattr(ExcitedStates, key) may now also return a callable, while otherwise e.g. a numpy array is returned. For this reason you might also have to adjust the ExcitedStates.to_dataframe export - the only other part of the code except Excitation where ExcitedStates.excitation_property_keys is used.
Maybe it is also fine to not decorate the gauge dependent member functions with mark_excitation_propertyfor now and add them at a later point? I think we need to iterate over ExcitedStates and Excitation anyway for #168.

[frieschneider]

I don't agree with that. I think it makes sense to return them as callables:

• It would be unintuitive to exclude them, since they are actually excitation properties as well, and transition_magnetic_dipole_moments were included before as well.
• However, it will not work to return them as numpy arrays, since they are no longer a constant value, but depend on the choice of gauge origin.

Therefore, I don't want to simply exclude them. I think it would make more sense to adjust ExcitedStates.to_dataframe. It could also take a gauge_origin parameter. I also need to extend it so that it can handle two-dimensional operators as well, which are needed for transition_electric_quadrupole_moments for example.

[apapapostolou]

Everything looks good to me. I agree with @frieschneider and think that we can leave the gauge-dependent excitation properties as they are for now. If ExcitedStates and Excitation are changed in the future, we can reconsider this as well.

If @jonasleitner also gives his approval, I think this is ready to be merged.