integrate ( exp (x^3), x, 0, 1) still wrong



On Sat, May 12, 2012 at 7:32 AM, Raymond Toy <toy.raymond at gmail.com> wrote:

> >
> > --------------------------------------------------
> > case exp(x^5) over (1,2)
> > domain:complex greatly improves answer
> > but large incorrect imaginary part
> > and real part only accurate to two sig. figures.
>
> Yes, that's a problem.  gamma_incomplete(1/5,-32) is not very accurate.
>  This can probably solved, but I'm not sure.  Using bigfloats with 32
> digits gives a much better answer.
>
>
I've updated the evaluation for gamma_incomplete(, and
gamma_incomplete1/5,-32) should give  about 15 digits now.  The bigfloat
version should give about fpprec precision too.  (Previously, for large
negative x, it was getting less and less accurate.)

Ray