6

I've been trying to calculate $$\int_0^\infty\frac{1}{x^a(1-x)}\,dx\quad\text{with }0<a<1.$$I haven't had much luck. I tried taking the branch cut with of the positive reals and estimating that way, but I wasn't sure how to handle the poles at $z=0$ and $z=1$ when dealing with branch cuts. I also looked at the Wikipedia page to try to get some insight, but I didn't see any similar examples with the contour closing in on poles that are on a branch cut.

Semiclassical
  • 15,842

2 Answers2

4

You use an almost-keyhole contour, except that you indent both paths above and below the real axis with a small semicircle to avoid the pole at $z=1$:

enter image description here

In doing this, you end up with not $4$, but $8$ contour segments. I will avoid writing them all out by noting that the integrals over the outer circular arc and inner circular arc at the origin vanish in the limits of their radii going to $\infty$ and $0$, respectively. We are left with

$$\oint_C dz \frac{z^{-a}}{1-z} = \int_{\epsilon}^{1-\epsilon} dx \frac{x^{-a}}{1-x} + i \epsilon \int_{\pi}^0 d\phi\, e^{i \phi} \frac{(1+\epsilon e^{i \phi})^{-a}}{-\epsilon e^{i \phi}} + \int_{1+\epsilon}^{\infty} dx \frac{x^{-a}}{1-x} \\+e^{-i 2 \pi a} \int_{\infty}^{1+\epsilon} dx \frac{x^{-a}}{1-x} +e^{-i 2 \pi a} i \epsilon \int_{2 \pi}^{\pi} d\phi\, e^{i \phi} \frac{(1+\epsilon e^{i \phi})^{-a}}{-\epsilon e^{i \phi}} +e^{-i 2 \pi a} \int_{1-\epsilon}^{\epsilon} dx \frac{x^{-a}}{1-x} $$

Combining like terms, we get

$$\oint_C dz \frac{z^{-a}}{1-z} = \left ( 1-e^{-i 2 \pi a}\right ) PV\int_{0}^{\infty} dx \frac{x^{-a}}{1-x} + \left ( 1+e^{-i 2 \pi a}\right ) i \pi = 0$$

because of Cauchy's Theorem. $PV$ denotes the Cauchy principal value. After a little algebra, the result is

$$PV\int_{0}^{\infty} dx \frac{x^{-a}}{1-x} = -i \pi \frac{1+e^{-i 2 \pi a}}{1-e^{-i 2 \pi a}}=-\pi \cot{\pi a}$$

EXAMPLE

Let's check the result for $a=1/2$. This would imply that

$$PV \int _{0}^{\infty} dx \frac{1}{\sqrt{x} (1-x)} = 0$$

Consider

$$\begin{align}\underbrace{\int_0^{1-\epsilon} dx \frac{1}{\sqrt{x} (1-x)}}_{x=1/u} &= \int_{1/(1-\epsilon)}^{\infty} \frac{du}{u^2} \frac{\sqrt{u}}{1-(1/u)} \\ &= -\int_{1+\epsilon}^{\infty} du \frac{1}{\sqrt{u} (1-u)}\end{align}$$

Thus

$$\int_0^{1-\epsilon} dx \frac{1}{\sqrt{x} (1-x)} + \int_{1+\epsilon}^{\infty} du \frac{1}{\sqrt{u} (1-u)} = 0$$

or

$$PV \int _{0}^{\infty} dx \frac{1}{\sqrt{x} (1-x)} = 0$$

as was to be demonstrated.

Ron Gordon
  • 138,521
  • Do we ever calculate the residue? Or is that more of a by-product for using the almost-keyhole contour, as when doing integration before learning about the residue theorem? – anon271828 Apr 26 '13 at 05:05
  • No residue here, because we excluded the pole at $z=1$ from the interior of the contour. I would say that we just get contributions from the segments we use to avoid integrating over a pole. It is these contributions that give rise to the value of the integral we seek. This happens more often than you think, and mastery of this technique will provide you with a powerful tool. BTW the integral we sought is the Cauchy principal value because it avoids the pole on the real axis. – Ron Gordon Apr 26 '13 at 05:11
  • Is there ever convergence of the Cauchy principal value and not of the ordinary integral (assuming they're different)? According to Mhenni's comments above, the integral should diverge, say, for $a=\frac{1}{2}$. According to this solution, it is $0$. – anon271828 Apr 26 '13 at 05:13
  • Mhenni speaks of the integration right through the pole, which obviously diverges. My result is the principal value, which converges to the value I posted. – Ron Gordon Apr 26 '13 at 05:14
  • I see; thank you very much for your help! As an update, Mhenni retracted his statement about divergence. I'll appreciate this answer more and more as I work through the details. – anon271828 Apr 26 '13 at 05:16
  • @anon271828: You're welcome. If you check in Mathematica, make sure you use the "PrincipalValue->True" option. – Ron Gordon Apr 26 '13 at 05:17
  • @RonGordon: As an aside, by saying principal value, this means you are restricting the argument to be between $0$ and $2\pi$? This isn't something that has been introduced in class, so I want to have a better understanding. Wikipedia was helpful, but perfect. – anon271828 Apr 26 '13 at 05:21
  • @RonGordon: Compare with my answer, I think you are missing a factor. – Mhenni Benghorbal Apr 26 '13 at 05:28
  • @anon271828: the point of the keyhole contour is to restrict the multivalued function in the integrand to have an argument between $0$ and $2 \pi$. That is a separate, but not unrelated, issue from the principal value thing. The PV is a mechanism we use to express that part of the integral with a pole stripped away. – Ron Gordon Apr 26 '13 at 05:28
  • @MhenniBenghorbal: I disagree and wonder how you miraculously got a convergent integral from a simple substitution like that. You need to be careful in the neighborhood of the singularity at $x=1$. BTW, as I mentioned above, my result checks with Mathematica. – Ron Gordon Apr 26 '13 at 05:30
  • @MhenniBenghorbal: thanks, BTW. – Ron Gordon Apr 26 '13 at 05:46
  • @RonGordon: I'll be checking this problem tomorrow. It is getting late now. – Mhenni Benghorbal Apr 26 '13 at 05:47
  • can you please tell me what parametrization you used for the bottom semicircle at the pole at z=1? – Madhav Asthana Mar 01 '24 at 07:03
1

$\newcommand{\bbx}[1]{\,\bbox[15px,border:1px groove navy]{\displaystyle{#1}}\,} \newcommand{\braces}[1]{\left\lbrace\,{#1}\,\right\rbrace} \newcommand{\bracks}[1]{\left\lbrack\,{#1}\,\right\rbrack} \newcommand{\dd}{\mathrm{d}} \newcommand{\ds}[1]{\displaystyle{#1}} \newcommand{\expo}[1]{\,\mathrm{e}^{#1}\,} \newcommand{\ic}{\mathrm{i}} \newcommand{\mc}[1]{\mathcal{#1}} \newcommand{\mrm}[1]{\mathrm{#1}} \newcommand{\on}[1]{\operatorname{#1}} \newcommand{\pars}[1]{\left(\,{#1}\,\right)} \newcommand{\partiald}[3][]{\frac{\partial^{#1} #2}{\partial #3^{#1}}} \newcommand{\root}[2][]{\,\sqrt[#1]{\,{#2}\,}\,} \newcommand{\totald}[3][]{\frac{\mathrm{d}^{#1} #2}{\mathrm{d} #3^{#1}}} \newcommand{\verts}[1]{\left\vert\,{#1}\,\right\vert}$ $\ds{\bbox[5px,#ffd]{\large An\ Alternative:}}$ \begin{align} &\bbox[5px,#ffd]{% \left.{\rm P.V.}\int_{0}^{\infty}{\dd x \over x^{a}\pars{1 - x}} \,\right\vert_{\,a\ \in\ \pars{0,1}}} = -\,\Re\int_{0}^{\infty}{x^{-a} \over x - 1 + \ic 0^{+}}\,\dd x \\[5mm] = &\ \Re\int_{\infty}^{0}{\ic^{-a}\,\,y^{-a} \over \ic y - 1} \,\ic\,\dd y = -\,\Im\bracks{\ic^{-a}\int_{0}^{\infty} {\,y^{\pars{\color{red}{1 - a}} - 1} \over 1 - \ic y}\,\dd y} \end{align} I'll evaluate the last integral with the Ramanujan's Master Theorem. Note that $\ds{{1 \over 1 - \ic y} = \sum_{k = 0}^{\infty}\pars{\ic y}^{k} = \sum_{k = 0}^{\infty}\color{red}{\Gamma\pars{1 + k}\expo{-k\pi\ic/2}}\,\,{\pars{-y}^{k} \over k!}}$.

Then, \begin{align} &\bbox[5px,#ffd]{% \left.{\rm P.V.}\int_{0}^{\infty}{\dd x \over x^{a}\pars{1 - x}} \,\right\vert_{\,a\ \in\ \pars{0,1}}} \\[5mm] = &\ -\,\Im\pars{\ic^{-a}\,\,\Gamma\pars{1 - a}\braces{\Gamma\pars{1 -\bracks{1 - a}}\expo{\pars{1 - a}\pi\ic/2}\,}} \\[5mm] = &\ -\,\Re\bracks{\expo{-\pi a\ic}\,\, \Gamma\pars{1 - a}\Gamma\pars{a}} = -\cos\pars{\pi a}\,{\pi \over \sin\pars{\pi a}} \\[5mm] = &\ \bbx{-\pi\cot\pars{\pi a}} \\ & \end{align}

Felix Marin
  • 89,464