bigfloats of special functions (e.g. Bessel).. wasRe: integrate bessel_j errcatch?



On 12/8/2011 12:58 PM, Raymond Toy wrote:
>
>     For example, evaluation of a Bessel function near a zero of that
>     function needs some
>     special work.  One approach is to expand in a Taylor series about
>     that zero...
>
>
> Yes, we could.  But we don't do that for sin, so why do we need that 
> for Bessel?
>
> Ray
>
>
Sin, cos, are handled by (very accurate) range reduction, so a series 
only has to be good for
something like 0<x<pi/8  or so.

exp and log can be dealt with by  'doubling' eg   exp(x) = (exp(x/2))^2  
recursively.

Bessel functions have nice identities, but not as nice as those.