Problems with improper integrals



On Mon, Mar 10, 2008 at 7:24 AM, Alexios Tamparopoulos
<alexis.maxima at gmail.com> wrote:
>  f2(x):=x/(1-cos(x)+x^2);
>  integrate(f2(x),x,0,2*%pi);
>  not result was presented (should be 8*%pi/3 if calculation I did by hand
>  was right).
>  Seems like Maxima can't handle this integral.

I'm afraid your hand calculation is incorrect -- this integral is
divergent.  You can see this as follows:

     f2: x/(1-cos(x)+x^2)$
     f2t: taylor(f2,x,0,0) => 2/(3*x)
         since integrate(1/x,x,0,a) is divergent, this is divergent

Now let's use Maxima to study the behavior of this integral when the
lower limit is near 0.  Let's separate the pole part and the rest:

    pole_part: integrate(f2t,x,a,2*%pi) => 2*(log(2*%pi)-log(a))/3

    nonpole_part: quad_qag( f2 - ratsimp(f2t), x, 0, 2*%pi, 1)[1] => - 0.094188
            (The ratsimp prevents Maxima from truncating f2 by contagion.)

So the whole thing looks like expand(float(pole_part+nonpole_part)) =>
1.131063 - 0.666667 * log(a) or 1.131063 - 2/3*log(a)

I have no idea if there is a closed form.

            -s