Allow negation of representations with differentials #10987

Activity

1. mhvk

   added

   Feature Request

   coordinates

   on Nov 3, 2020
2. ayshih commented on Nov 5, 2020

   

   ayshih

   on Nov 5, 2020

   Contributor

   More actions

   While I believe that the negation of a Cartesian representation/differential is uncontroversial, I suspect that it may be controversial for other representations, e.g., cylindrical:

   >>> import astropy.coordinates.representation as r
   >>> import astropy.units as u
   
   >>> pos = r.CartesianRepresentation(1, 0, 0)*u.m
   >>> vel = r.CartesianDifferential(0, 1, 0)*u.m/u.s
   
   # The velocity converted to cylindrical components
   >>> vel.represent_as(r.CylindricalDifferential, pos)
   <CylindricalDifferential (d_rho, d_phi, d_z) in (m / s, rad / s, m / s)
       (0., 1., 0.)>
   
   # If cylindrical velocity is directly negated, it negates the cylindrical components
   # The sign for `d_phi` has changed ==> the circulation is reversed
   >>> -vel.represent_as(r.CylindricalDifferential, pos)
   <CylindricalDifferential (d_rho, d_phi, d_z) in (m / s, deg / s, m / s)
       (-0., -1., -0.)>
   
   # The natural negation of the Cartesian representation/differential is {-pos, -vel}
   # We then convert the new velocity to cylindrical components
   # Note that the sign for `d_phi` has not changed ==> the circulation is preserved
   >>> (-vel).represent_as(r.CylindricalDifferential, -pos)
   <CylindricalDifferential (d_rho, d_phi, d_z) in (m / s, rad / s, m / s)
       (1.2246468e-16, 1., -0.)>

   That is, the "right" answer for a cylindrical representation/differential depends on whether you think that the circulation should be preserved/reversed. Note that expecting the circulation to be reversed when working in cylindrical could be weird because walking through Cartesian would always preserve the circulation.
3. mhvk commented on Nov 5, 2020

   

   mhvk

   on Nov 5, 2020

   ContributorAuthor

   More actions

   @ayshih - agreed one has to be a bit careful, but I think one can make things clear and consistent by insisting that for any operation, the answer does not matter in what representation it is done, i.e., there is a default that converts to cartesian, applies the operation, and the converts back, and there are also shortcuts that give the same result. Indeed, this is how all the operations are implemented right now on representations without differentials. I think the problem with differentials is that for higher order ones, it might be less clear on how to proceed - but that is basically because we cannot yet transform them properly to cartesian in the first place.
4. mhvk

   closed this as completedin #11470on Jul 19, 2021