Eric Agol

@Tusay Thanks for the summary! The `rstar` is extremely arbitrary, and needs to be looked into further.

@langfzac I'm still unclear on whether this is an issue, however, since there are enough transit times for TRAPPIST-1 that you would think at least one transit might have been...

I think this has been fixed - there is a maximum number of iterations here: > https://github.com/ericagol/NbodyGradient.jl/blob/cc7ba5d1e7eaaa57313396627a8aa05ba5040019/src/ics/kepler.jl#L33-L35

This is still an issue in the current version of the code: https://github.com/ericagol/NbodyGradient.jl/blob/85631d64cd22125820474fc9acd8f6da2f29c78f/src/transits/timing.jl#L14 The only reason this doesn't affect the transit-finding is that usually we pass a very large value...

We compared the integrator with the C implementation of Hernandez & Bertschinger. The Julia version is ~5-10 times slower. This is pretty surprising - in my experience - and based...

@giordano Yes, we've done that - but maybe we missed something, or could possibly could do a better job!

@dmhernan Thanks!

The comparison with universal.c was useful - thanks to a few small changes, we can now match the C speed. See this pull request (an older version of NbodyGradient was...

(These two codes are currently not in the repository - please let me know if you would like to see them. The version of universal.c which was used for this...

So, to improve the speed of the ah18/ttv code, we need to avoid array allocations (allocation of x0 & v0 in kepler_step was causing the biggest slow down in the...