search

rodluger / Limbdark.jl

Analytical transit light curves for limb darkened stars
MIT License
14 stars 4 forks source link

Re-write #70

Closed rodluger closed 5 years ago

rodluger commented 5 years ago

Re-wrote most of section 5, changing a lot of the structure and notation to be more consistent with the starry paper. The a basis is now the "polynomial basis" p and the d basis is now the "Green's basis" g. I updated the Table, cleaned up some other stuff throughout the manuscript, and added some to-do items in red.

@ericagol and @dfm , can you check the paper and see if you like what I did? I think I/we still need to write a paragraph at the end of the introduction explaining what the s basis is and hinting to the reader where we're headed with our strange notation. I also need to write a short section summarizing the comparison to PyTransit. But I think we're getting very close!

@ericagol You asked me a while back if we should have a section convincing people they should use our new approach. @dfm, perhaps you could write a section on a system you've fit with exoplanet and how much faster it is with the gradient-based approach? We could add this during review, too, if it will take you more than a few hours to write up.

@ericagol My next task is to code up the faster M formalism in starry. I should have that done by tomorrow, and I'll add the starry comparison curves back in then.

rodluger commented 5 years ago

@ericagol I'm closing this momentarily while I resolve some Travis issues. Will reopen soon.

rodluger commented 5 years ago

OK, travis issues fixed! I'm still working on this branch, but you can merge this whenever you'd like and I'll submit a new PR once I'm done.

rodluger commented 5 years ago

@ericagol I added the starry comparison back in. I now have two curves: one for the "naive" implementation using spherical harmonics, which scales terribly because of the huge number of terms, and one for the dev version of my code, in which I finally implemented all of your equations. That one is comparable in speed to yours, although your gradient version is somewhat faster. There's still significant I/O overhead in getting from C++ to Python, so I'm still working on getting that time down. But I'm finally happy with my code's speed overall!