stephenberry
commented
3 years ago Thanks for pointing out heyoka, it looks like a neat library. There are a number of higher precision libraries (up to 128bit quad precision) that I don't think would be difficult to use with Ascent, but I haven't tried, and I'm not sure which one would be best to implement.
That is an interesting competition. I have run high order Taylor solvers (like heyoka) in the past and I can see their appeal for this problem, but I feel like a custom implicit solver would be a better approach for this kind of problem where the equations are all known.
If you try to tackle this problem with Ascent I would be happy to give feedback. Another option is looking into boost's odeint implicit solvers.
Do you have any idea of how heyoka implements 128bit floating point support?
Would be nice if Ascent could support 80-bit precision as https://github.com/bluescarni/heyoka already does.
From https://arxiv.org/abs/2105.00800 :
"Although traditionally double precision has been considered enough for high-precision applica- tions in astrodynamics and celestial mechanics, in recent times our understanding of the physics of the Solar System has progressed to the point that the use of double-precision arithmetic can become the limiting factor in achieving the desired level of accuracy"
There is an ongoing competition https://kelvins.esa.int/space-debris-the-origin/challenge/ where I think Ascent is a good alternative to Heyoka, but possibly requires high accuracy (30 years orbit propagation).