SpecialFunctions.jl icon indicating copy to clipboard operation
SpecialFunctions.jl copied to clipboard

bessely() sometimes returns bogus

Open hsgg opened this issue 8 years ago • 2 comments

I have found that bessely() with very large argument returns various results. For example,

julia> using SpecialFunctions

julia> bessely(1000.5, 7.745186719805684e8)
-1.65447992609e-312

julia> x = bessely(1000.5, 7.745186719805684e8)
1.3586779285536575e24

The assignment to x enhances the effect, though different answers are returned without assignment as well.

I am using julia-v0.6.1 on x86_64 archlinux. The effect is less severe, but also present, on my macosx machine, which is running 0.6.1-pre3. Finally, on archlinux, I've also seen it with v0.6.2-pre.2, commit 247087d94c.

hsgg avatar Nov 07 '17 20:11 hsgg

Well that's odd. Any ideas, @simonbyrne?

ararslan avatar Nov 07 '17 20:11 ararslan

It seems like the fortran code is returning an error code of 3, which corresponds to precision loss. My guess is that the variation is due to different values being provided in the initialised Ref objects.

simonbyrne avatar Nov 07 '17 22:11 simonbyrne