Closed unageek closed 2 years ago
Do you have an example where arb_hypgeom_shi
performs poorly?
I'm aware that acb_hypgeom_shi
(imaginary argument) and arb_hypgeom_si
give bad results in the transition regions.
It would be best to just fix all the exponential/trigonometric/hyperbolic integrals to do the right thing, similarly to what acb_hypgeom_erf
already does (evaluate at midpoint, automatically increase precision to ensure that the result is accurate; use derivative for error bound).
Thank you for your response, and sorry for so late reply.
Do you have an example where
arb_hypgeom_shi
performs poorly?
I'm afraid not. Apparently, I was looking at a different function. So let me close this issue.
PS: I'm building a graphing calculator^1 that leverages Arb, and I'm very impressed by the quality of the library. Please keep up the great work!
arb_hypgeom_shi
callsacb_hypgeom_shi
which in turn callsacb_hypgeom_si
and which heuristically chooses an algorithm fromacb_hypgeom_si_1f2
andacb_hypgeom_si_asymp
. But of course, the chosen algorithm is not always the best one. So I want to try both algorithms and take the intersection of the results to add some robustness.The best way I can think of is to modify or copy-paste the code of the functions to add a parameter for specifying the algorithm. However, it would be hard for me to keep them in sync with the original code. Is there a more convenient way to achieve this?