[AstroPy] Dependencies on large data files

[Peter Williams]

Bundling my replies ...

On Wed, 2014-07-23 at 17:34 -0400, Scott Ransom wrote:
> As for barycenter calculations, we are implementing this to very-high 
> precision using the JPL ephemerides for the new high-precision pulsar 
> timing package PINT.  For that package we are pulling the JPL ephemeris 
> files and then using SPICE for most of the ephemeris table work.  We use 
> the astropy Time objects (which are sweet) for all the time calculations.
> 
> The software is in its early stages of development (and hasn't been 
> updated recently because of other time commitments), but barycentering 
> is already in there to few nano-sec precision, including all the 
> relativistic effects in our Solar System as well.
> 
> https://github.com/nanograv/PINT

That's great to hear! I know you guys are the pros at this kind of thing.

On Wed, 2014-07-23 at 13:34 -0700, Tim Jenness wrote:
> ERFA is in astropy already so there are some astrometry routines that
> exist without having python visibility at the moment.
> 
> Which routines do you have?

I've mainly used the NOVAS routines that do things like calculate
horizon coordinates and astrometric positions accounting for proper
motion, aberration, etc. A lot of these calculations are basic spherical
trig for most uses, but NOVAS does them right to VLBI-level precision. I
personally don't need anything that accurate (so I'm probably doing a
bunch of stuff wrong...), but it seems good to build on a foundation
that *can* achieve that level of precision.

If I recall correctly, SOFA/ERFA doesn't include the high-level routines
to do these kinds of calculations. And as far as I can tell, AstroPy
doesn't currently provide tools for things like calculating horizon
coordinates, rise/set times, astrometry, etc ... correct?

On Thu, 2014-07-24 at 00:21 +0200, Thomas Robitaille wrote:
> To answer this original question about large data files, we have indeed
> put some thought into this and the ``get_pkg_data`` and related routines
> in ``astropy.utils.data`` do have the ability to fetch data from a
> data.astropy.org server automatically. We haven't really used it until
> now because we haven't had the need, but this sounds like a good
> application.

Thanks for the pointer -- I had some dim memory that there was
infrastructure for this kind of thing.

Overall, I think it'd be great for AstroPy to include infrastructure for
BJDs and high-quality positional astronomy routines. But there's a
question of how "high-quality" one can get before the APIs and
dependencies become too cumbersome for a general-use package.

So, for the experts in the room: what level of precision do you think
can be achieved in a general-use library like AstroPy? Is that something
where the JPL ephemeris will be a useful piece of infrastructure, or are
analytic calculations going to be good enough?

Cheers,

Peter