mirror of
https://github.com/boostorg/math.git
synced 2026-02-11 11:52:21 +00:00
2a0f5f111c
* Bessel approximation is now good for b-2a <= 0 via Tricomi's relations. * Selection condition based on convergence is now more rigorous, or at least more conservative. But... * Error rates are much higher because a poorly converging bessel approximation has slightly less cancellation than the generic series. Hopefully this can be fixed via https://dlmf.nist.gov/13.13. And... * We need to rewrite these new versions to use recurrence relations on the Bessel functions (but these need to be backwards, so we would calculate and cache say 50 evaluations, then the next 50 and so on). And... * Bessel approximations can underflow spuriously when b is large.... really need a log_bessel_j to fix this as we will be multiplying by tgamma(b). [CI SKIP]