rv_offset parameters and "solar units"

[dmrowan]

Hi there,

I'm working with a binary where I have a light curve and radial velocities. I have multiple spectrographs, and I'd like to include an RV offset to account for systematic offsets between the instruments.

say my two instruments are alpha and beta. I have more RV measurements from alpha, so I want to fit for an RV offset that is applied to measurements from beta.

I want the RV offset to be applied to both the primary and the secondary RVs so I want to add a constraint to ensure the rv_offset of the secondary just points to the rv_offset for the primary.

In the past, I've done something like this:

lhs = self.bundle.get_parameter('rv_offset@secondary@rv_beta')
rhs = self.bundle.get_parameter('rv_offset@primary@rv_beta')

self.bundle.add_constraint(lhs, rhs)

I last did this a few PHOEBE versions ago, and this isn't successful on 2.4.22. When adding the constraint, the error is something like:

[ERROR] [bundle:run_constraint:7760]: Constraint 'rv_offset@secondary@rv_beta@rv@constraint' raised the following error while attempting to solve for 'rv_offset@secondary@rv_beta@rv@dataset'.  Consider flipping the constraint or changing the value of one of ['rv_offset@primary@rv_beta@rv@dataset', 'rv_offset@secondary@rv_beta@rv@dataset'] until the constraint succeeds.  Original error: cannot convert '0.7526634758875196 km / s' with physical_type='speed/velocity' to solar units

The constraint gets made, but doesn't seem to be enforced. printing

print(bundle['secondary@rv_offset@rv_beta@constraint'])

gives a more informative error message. Here is the relevant portion

File phoebe/parameters/parameters.py:11951, in ConstraintParameter.get_result.<locals>.get_values.<locals>._single_value(quantity, string_safe_arrays)
  11949 if isinstance(quantity, u.Quantity):
  11950     if self.in_solar_units:
> 11951         v = u.to_solar(quantity).value
  11952     else:
  11953         v = quantity.si.value

File phoebe/dependencies/unitsiau2015/__init__.py:118, in to_solar(object)
    115 physical_type = _get_physical_type(object)
    117 if physical_type not in _physical_types_to_solar.keys():
--> 118     raise NotImplementedError(f"cannot convert '{object}' with physical_type='{physical_type}' to solar units")
    120 return object.to(_physical_types_to_solar.get(physical_type))

NotImplementedError: cannot convert '0.7526634758875196 km / s' with physical_type='speed/velocity' to solar units

It looks like by default every constraint is set to have constraint.in_solar_units = True. Modifying

list_of_constraints_requiring_si = []

to

list_of_constraints_requiring_si = ['custom']

in constraint.py seems to fix it, but I have a feeling this is not the intended approach. Any advice here? More generally, is there a reason why the rv_offsets are applied to components separately and not applied at the dataset level? In other words, why would one want to add an RV offset to the primary but not the secondary?

Cheers,
Dom

[kecnry]

Hi Dom -

Can you let me know what version of astropy you are using? Was that different than the version of astropy you had when it was working? This might be a change in how astropy is handling units that we need to handle in phoebe as well. Would you mind opening an issue with those details and we will try to reproduce and get it fixed?

Although setting a manual constraint like this should work and do what you want, there is also a systemic physical velocity you can set (vgamma@system) that would effectively do the same thing but without needing the constraint. The rv_offset parameters are intended to be per-component and mostly useful for hot stars where the observed lines (used to measure the RVs) are created in different levels of the atmospheres. See Underhill & Hill 1994; Harmanec et al. 2002; Shenar et al. 2018, and the brief explanation (basically the same as I included here) in the 2.3 release paper, section 3.4.5.

Hope this helps!
Kyle

[dmrowan]

Hi Kyle,

Thanks -- just created the issue. The astropy version I am using is '7.1.0'. I can pretty much guarantee I wasn't using this version when I last did this kind of thing, but I don't know what version it was, unfortunately.

Thanks for sharing the reference for a physical motivation for rv offsets! But I think my question is about a slightly different thing, which is systematics in instrument calibration. In short, different spectrographs will have different RV zero-points, so unless you are getting an RV standard each night, we need to account for this in the fit. We want an offset applied to a specific instrument. This is something that is commonly done for single-lined spectroscopic binaries (see TheJoker docs or a real-world example), but we also do this for SB2s (for example :)). In these cases, the default assumption is that the RV offset of the primary star is the same as the RV offset of the secondary star. This is different from the system velocity vgamma.

[kecnry]

ok, in that case this might be the best solution for you. In the future, #902 may provide more flexibility (more code-overhead, but you could define the adjustment to be however you need without having to add constraints to existing parameters). Let me know if you think that would be helpful for your use-case.

[aprsa]

What you really want here is per-RV-dataset vgamma value. This of course is not a physical circumstance but instrumental; you could arguably correct this outside of phoebe (by, say, median-normalizing the datasets), but I can see the appeal in having this done in phoebe so that you get proper posteriors and cross-correlations with other parameters.

[dmrowan]

Yes, I think this would be an example of a DatasetFeature where we have a vgamma+offset for the second instrument.

I'll just keep my modified list_of_constraints_requiring_si = ['custom'] and proceed!

Thanks again,
Dom