This is a somewhat random collection of notes that summarize discussions in Github issues and mailing lists to understand differences in results seen in the benchmarks.
Mailing list discussions:
- https://groups.google.com/forum/?fromgroups#!topic/astropy-dev/tbcpMQ1rOcY
- https://groups.google.com/forum/?fromgroups#!topic/astropy-dev/-ulzSY9qJKk
- https://groups.google.com/forum/?fromgroups#!searchin/astropy-dev/benchmark
Github issues with detailed discussion or results:
- #37
- ...
- 2012-11-20: Fixed
kapteyn
transforms to explicitly specify the epoch of observation for the B1950 conversion, and the equinox for the ecliptic conversion - 2012-11-21: Explicitly precess astropy.coordinates fk4 result to B1950
- 2015-01-14: Add first AltAz system benchmark
- 2015-01-15: Re-structure
coordinates-benchmark
(see GH #42) - 2020-08-20: Unify Python 2 and Python 3 (see GH #63)
> 1) The differences between kapteyn and (idl,pyast,tpm) on the J2000 to > B1950 tests is reduced seriously if you specify the epoch of > observation explicitly in kapteyn/convert.py. That is, change > > transform_to(skyout='fk4', tag='b1950') > > to > > transform_to(skyout='fk4,J2000_OBS', tag='b1950') > > This reduces the median difference from 0.217 arc-sec, to 0.0044 > arc-sec (this must mean that kapteyn is using a different default > epoch of observation to the other systems). > > 2) Improved agreement in J2000->ecliptic can be achieved by > specifying the equinox explicitly in kapteyn/convert.py. Change: > > transform_to(skyout='ecliptic', tag='ecliptic') > > to > > transform_to(skyout='ecliptic,J2000', tag='ecliptic') > > This causes kapteyn and IDL have zero difference. The kaptyn->pyast > median difference goes down from 0.04 to 0.036. The remaining > difference between kapteyn and pyast seems to be down to: > > a) DIfferent obliquity models. kaptyn uses the IAU 2000 model, and > pyast uses the IAU 2006 model. > b) Different precession models. kaptyn uses IAU 1976 and pyast uses IAU 2006. > > Reverting the models used by pyast gives zero difference between > kapteyn and pyast for the J2000->Ecliptic test