Many thanks for this.
Concerning the previous comment, it seems that the tolerance can be adjusted in cbesselj.m ('tol'). Still needs some tweaking to allow one to call the function with independent arrays for nu and x (i.e. both not scalar).
20 Oct 2008
It's a very good initiative but a precision of two or so digits is not good enough for my purposes, and I would think this goes for many other users too.
24 Oct 2006
If you compare the result with mathematica which can set the precision digits. This code has very big error. Thanks for sharing though.