`bilby.gw.conversion` module for waveforms which use a custom reference frequency

@serguei.ossokine writes:

I was recently thinking of doing some test runs with SEOBNRv4PHM and bilby/parallel bilby and I think that there might be currently a bug/misconfiguration possible when running with this waveform. In particular, as far as I can tell, currently the code that transforms from the spin angles (so tilts, \phi_{jl}, etc) to Cartesian coordinates of the spin, always uses the reference frequency. For SEOB the reference frequency is actually ignored and it's always the starting frequency. Therefore if the reference frequency is different from the starting frequency, the code will do the transformation from spin angles to Cartesian coordinates of the spin at the "wrong" frequency - i.e. not at the starting frequency of EOB waveform.

I don't think this affects any results so far, but I think it's important to fix this. This does not affect RIFT - RIFT samples directly in Cartesian spin components. (It also doesn't affect LALInference which handles SEOB in a special way for this reason)

I think I've seen this effect when doing test runs.

The solution here is to tell the waveform generator the correct reference frequency. AFAIK, there is no way from lalsimulation to know if the waveform uses the starting frequency or something passed in (do we pass it in and it is ignored.. or is it never passed it?).

I think this is a priority issue for O4 development.