12

I want to compute:

$$\int^{1}_{0}\frac{1}{\sqrt[n]{1-x^n}}dx$$

for natural $n>1$ using Residue Calculus.

I am thinking of using some kind of a keyhole or bone contour that could go around the $n$th roots of unity (singularities in this case). The problem is I believe it is not clear how to define a suitable branch (or branches) of $\log$ in this region for it to work, also considering we only care about the segment from $0$ to $1$.

user91500
  • 5,606
  • 1
  • @Lucian Are you suggesting to try to do the contour integration to the integral seen as a Beta function? I see that by substituting $t=x^n$ you can get $Beta(1/n,1-1/n)$ which would even give the answer to the integral, but that would not use any contour integration method. – Luke Skywalker Jul 30 '14 at 04:36
  • The beta function has an analytic continuation, which is probably what Lucian is referring to – ClassicStyle Jul 30 '14 at 04:43
  • @TylerHG Ah, I see, thanks. So I take it the same kind of analytic continuation might work here? As I recall the continuation was not very transparent in that it involved the connection with the Gamma function... – Luke Skywalker Jul 30 '14 at 04:57
  • Just for your info, using Mathematica the solution is $\frac{\pi \csc \left(\frac{\pi }{n}\right)}{n}$. See Here – Mojtaba Golshani Jul 30 '14 at 04:59
  • @MojtabaGolshani Thanks, as I commented above, indeed the answer is $\frac{Beta(1/n,1−1/n)}{n}=\frac{\pi}{n \sin \left(\frac{\pi }{n}\right)}$; I am trying to get to it using residues with a suitable contour. – Luke Skywalker Jul 30 '14 at 05:05
  • Just for your info, the antiderivative is given by $$x , _2F_1\left(\frac{1}{n},\frac{1}{n};1+\frac{1}{n};x^n\right)$$ – Claude Leibovici Jul 30 '14 at 06:36
  • @ClaudeLeibovici Thanks, I was gonna ask what was $_2F_1$, but I see now it stands for Hypergeometric Function. – Luke Skywalker Jul 30 '14 at 06:57
  • @LukeSkywalker. And from there, the relation to the $Beta$ function. – Claude Leibovici Jul 30 '14 at 07:11

5 Answers5

13

Here is another contour integration using the contour that arises from not using any substitution.

Using the diagram and contour below, where $f(z)=\frac1{(1-z^n)^{1/n}}$

$\hspace{4cm}$enter image description here

it is easy to see that the integral along the red contour is $$ \begin{align} &\int_0^1\frac{\mathrm{d}t}{(1-t^n)^{1/n}}-e^{2\pi i/n}\int_0^1\frac{\mathrm{d}t}{(1-t^n)^{1/n}}\\ &=\left(1-e^{2\pi i/n}\right)\int_0^1\frac{\mathrm{d}t}{(1-t^n)^{1/n}}\tag{1} \end{align} $$ The integrand along each successive arm clockwise is $e^{2\pi i/n}$ times the integrand from the previous arm. $\mathrm{d}z$ along each successive arm is $e^{-2\pi i/n}$ times $\mathrm{d}z$ from the previous arm. Therefore, the integral along each successive arm is the same as the integral from the previous arm. Thus, the total of the integral along all the arms is $$ n\left(1-e^{2\pi i/n}\right)\int_0^1\frac{\mathrm{d}t}{(1-t^n)^{1/n}}\tag{2} $$ Now, if $z$ follows a circle of radius $R$ as $R\to\infty$, $f(z)\sim\frac{e^{\pi i/n}}z$ so the integral along a large clockwise circle is $$ -2\pi ie^{\pi i/n}\tag{3} $$ Since $(2)$ is equal to $(3)$, we have $$ \begin{align} \int_0^1\frac{\mathrm{d}t}{(1-t^n)^{1/n}} &=\frac{-2\pi ie^{\pi i/n}}{n\left(1-e^{2\pi i/n}\right)}\\ &=\frac\pi{n\sin(\pi/n)}\tag{4} \end{align} $$


The Branch Cut

Consider the function defined by $$ \begin{align} g(z) &=\pi i-\log(1-2^{-n})+\int_2^z\left(\vphantom{\sum_{k=0}^{n-1}}\right.\overbrace{\vphantom{\sum_{k=0}^{n-1}}\frac nw}^{\text{residue$=n$}}+\overbrace{\sum_{k=0}^{n-1}\frac1{e^{2\pi ik/n}-w}}^{\text{residue$=-n$}}\left.\vphantom{\sum_{k=0}^{n-1}}\right)\,\mathrm{d}w\\ &=\log\left(\frac{z^n}{1-z^n}\right)\tag{5} \end{align} $$ $g$ is well defined as long as the path of integration does not circle any of the poles of the integrand, or circles them all (since the sum of the residues is $0$); that is, the poles at $0$ and $\{e^{2\pi ik/n}:k\in\mathbb{Z}\}$. Circling none or all of these points is guaranteed by the branch cut in the diagram above. Therefore, with the branch cuts in the diagram above, $(5)$ allows us to define $$ \frac1ze^{g(z)/n}=\frac1{(1-z^n)^{1/n}}\tag{6} $$

robjohn
  • 345,667
  • Nicely done. +1 Working backwards, I knew that the integral along each successive had to be the same as the previous arm. But I was having trouble seeing that. – Random Variable Aug 02 '14 at 15:18
  • Consider $1-z^n$ near $z=1$. This looks like $1-((z-1)+1)^n\sim-n(z-1)+O(z-1)^2$. Thus, $\frac1{(1-z^n)^{1/n}}$ looks like $\frac{c}{(z-1)^{1/n}}$. Thus, going around a small clockwise circle centered at $1$ multiplies $f(z)$ by $e^{2\pi i/n}$. Therefore, going from the top to the bottom of $[0,1]$ multiplies $f(z)$ by $e^{2\pi i/n}$. – robjohn Aug 02 '14 at 16:53
  • Since $f(z)$ near $z=0$ is near $e^{2\pi ik/n}$ for some $k$, $f(z)$ is the same on the inside of both arms bordering a given sector. All we need to know about the behavior of $f(z)$ near the arms is determined by looking near $0$ or $e^{2\pi ik/n}$. – robjohn Aug 02 '14 at 16:56
  • This is exactly the kind of answer I was looking for when I asked the question, well done and thank you very much! – Luke Skywalker Aug 03 '14 at 02:02
  • 1
    How do you know that you could define a branch of $f$ on the region you specified? It's still not clear why a continuous branch must exist outside of those cuts you made. – Mark Aug 03 '14 at 02:28
  • One way to avoid the problem of defining $f$ is not mentioning it. I agree with Mark. Moreover, while this answers the expectations of the OP, it doesn't work for non integral $n$'s. – Omran Kouba Aug 03 '14 at 04:15
  • @Mark: I have added a justification for the branch cut used. – robjohn Aug 03 '14 at 06:05
  • @OmranKouba: the question does specify natural $n\gt1$. – robjohn Aug 03 '14 at 06:21
  • @robjohn, Exactly, that is what I meant, this is an unnecessary restriction, for which the method works. You also need to verify that your definition of $f$ coincides with what we expect for $t\in[0,1)$ and to say what happens if the path of integration encircles all the exceptional points. – Omran Kouba Aug 03 '14 at 16:15
  • 1
    @OmranKouba: Certainly, your answer and my first answer address general $n\gt1$. However, from Mark's comments, he is interested in a contour integration that uses the $n^\text{th}$ roots of unity. As for the definition of $f$, on any connected domain where $f(z)=(1-z^n)^{-1/n}$ is well defined, there are exactly $n$ possibilities for $f$; each differing by a factor of $e^{2\pi ik/n}$ from the others. All we need verify is the value at one point. That point for my definition is $z=2$. – robjohn Aug 04 '14 at 10:28
  • @robjohn, Connectedness is not sufficient, you need simple connectedness, and your domain is not simply connected. Apply your reasoning to the example of $z\mapsto\log z$ on $\mathbb{C}^*$. – Omran Kouba Aug 04 '14 at 13:45
  • @OmranKouba: I see that perhaps you are worried about the integral that leads to $g(z)$. I have improved the justification of the branch cut regarding circling the whole of the part about the origin. – robjohn Aug 04 '14 at 14:23
  • @OmranKouba $\widehat{\mathbb{C}} \setminus \bigcup\limits_{k=0}^{n-1} e^{2\pi ik/n}\cdot[0,1]$ is simply connected. $\infty$ is not a branch point since $\frac{1}{1-z^n}$ has a zero of order $n$ there. – Daniel Fischer Aug 04 '14 at 16:41
  • 1
    @DanielFischer, What I meant to say, is that all these comments and explanations, should have been part of the solution in order to make it complete. – Omran Kouba Aug 05 '14 at 05:28
  • @OmranKouba: I've added a section on the branch cut and improved it. Is there something else I should add to the answer? I don't really want it to be a tutorial on contour integration. – robjohn Aug 05 '14 at 06:45
  • @robjohn, being a professor, I always look for points that need extra explanations, or justification. Note that Daniel has a different justification for the definition of $g$, just saying that we need to be careful at some point is good enough. – Omran Kouba Aug 05 '14 at 06:54
  • @OmranKouba Thank you for raising the issue that the branch cut should be explained/justified carefully. I believe this has helped to complete the argument appropriately; thanks for those edits robjohn. Indeed, when posting the question, one of my main concerns was how to define the branch; I think it is more clear now. – Luke Skywalker Aug 05 '14 at 07:22
  • Someone unupvoted. Is there a problem? – robjohn Aug 06 '14 at 00:43
  • @DanielFischer: we could also consider $\frac{z^n}{1-z^n}$ which has no zeros outside of the contour (not even at $\infty$). We can take that to any power, not only $1/n$. – robjohn Aug 06 '14 at 00:47
10

Using the substitution $$ x^n=\frac{z^n}{1+z^n}\qquad\text{and}\qquad\mathrm{d}x=\frac{\mathrm{d}z}{(1+z^n)^{1+1/n}} $$ we get $$ \begin{align} \int_0^1\frac{\color{#00A000}{\mathrm{d}x}}{\color{#C00000}{(1-x^n)^{1/n}}} &=\int_0^\infty\color{#C00000}{(1+z^n)^{1/n}}\color{#00A000}{\frac{\mathrm{d}z}{(1+z^n)^{1+1/n}}}\\ &=\int_0^\infty\frac{\mathrm{d}z}{1+z^n}\\[6pt] &=\frac\pi{n\sin(\pi/n)} \end{align} $$ Using the result of the contour integration in this answer, which says $$ \int_0^\infty\frac{x^m}{1+x^n}\,\mathrm{d}x=\frac{\pi}{n}\csc\left(\pi\frac{m+1}{n}\right) $$

robjohn
  • 345,667
  • The motivation for using this substitution was solely to get things into a form to apply the cited result. – robjohn Aug 01 '14 at 17:42
  • Ah, this is another nice way! I actually had done before this integral you get and was wondering why the answer was the same csc! This explains it, thanks! – Luke Skywalker Aug 02 '14 at 05:31
9

In questions like this one, in order to avoid the problems of defining the right branch of the logarithm or the $n$th root, I suggest to, first start with a change of variables and to use Residue Theorem afterwards.

So, here how I do this. First the change of variables $x^n=\dfrac{e^t}{1+e^t}$ we get $$ I_n~{\buildrel {\rm def}\over =}~\int_0^1\frac{dx}{\root{n}\of{1-x^n}}=\frac{1}{n}\int_{-\infty}^\infty \frac{e^{t/n}}{1+e^t}dt $$ Next we integrate $F(z)=\dfrac{e^{z/n}}{n(1+e^z)}$ on the rectangle $\Gamma_R$ with vertexes $-R$, $R$,$R+2i\pi$ and $-R+2i\pi$. Letting $R$ tend to $+\infty$ we get $$ I_n-e^{2i\pi/n}I_n=2i\pi~\hbox{Res}(F(z),i\pi)=-2i\frac{\pi}{n} e^{i\pi/n}. $$ This yields $$I_n=\frac{\pi}{n\sin(\pi/n)}.$$ Note that we didn't use the fact that $n$ is a natural number. This is valid for any real $n>1$.

Omran Kouba
  • 28,772
  • I have not seen this change of variables before. Any intuition for why it works? – Mark Jul 30 '14 at 15:25
  • Yeah, great change of variables! It is basically transforming to another form of the Beta function as the comments were pointing out, and the one you chose admits this nice rectangular contour. – Luke Skywalker Jul 30 '14 at 16:03
  • I was kind of looking forward to an answer that would take on the definition of the branch and sum all the residues but this one works fine. Thanks for the suggestion in this kind of problems! – Luke Skywalker Jul 30 '14 at 16:17
5

$\newcommand{\angles}[1]{\left\langle\, #1 \,\right\rangle} \newcommand{\braces}[1]{\left\lbrace\, #1 \,\right\rbrace} \newcommand{\bracks}[1]{\left\lbrack\, #1 \,\right\rbrack} \newcommand{\ceil}[1]{\,\left\lceil\, #1 \,\right\rceil\,} \newcommand{\dd}{{\rm d}} \newcommand{\ds}[1]{\displaystyle{#1}} \newcommand{\expo}[1]{\,{\rm e}^{#1}\,} \newcommand{\fermi}{\,{\rm f}} \newcommand{\floor}[1]{\,\left\lfloor #1 \right\rfloor\,} \newcommand{\half}{{1 \over 2}} \newcommand{\ic}{{\rm i}} \newcommand{\iff}{\Longleftrightarrow} \newcommand{\imp}{\Longrightarrow} \newcommand{\pars}[1]{\left(\, #1 \,\right)} \newcommand{\partiald}[3][]{\frac{\partial^{#1} #2}{\partial #3^{#1}}} \newcommand{\pp}{{\cal P}} \newcommand{\root}[2][]{\,\sqrt[#1]{\vphantom{\large A}\,#2\,}\,} \newcommand{\sech}{\,{\rm sech}} \newcommand{\sgn}{\,{\rm sgn}} \newcommand{\totald}[3][]{\frac{{\rm d}^{#1} #2}{{\rm d} #3^{#1}}} \newcommand{\verts}[1]{\left\vert\, #1 \,\right\vert}$ \begin{align}& \overbrace{\color{#c00000}{\int_{0}^{1}{\dd x \over \root[n]{1 - x^{n}}}\dd x}} ^{\ds{x^{n}\ \mapsto\ x}}\ =\ \int_{0}^{1}\pars{1 - x}^{-1/n}\,{1 \over n}\,x^{1/n - 1}\,\dd x ={1 \over n}\color{#00f}{\int_{0}^{1}x^{1/n - 1}\pars{1 - x}^{-1/n}\,\dd x} \,\,\,\,\,\pars{1} \end{align}

Now, we study the integral $$ \oint_{\rm DB}{z^{1/n}\pars{1 - z}^{-1/n} \over z}\,\dd z $$ where $\ds{\rm DB}$ is a "dumbbell contour" ( see picture ). enter image description here

Also, $$ \begin{array}{rclrcccl} z^{1/n} & = & \verts{z}^{1/n}\expo{{\rm Arg}\pars{z}/n}\,, & -\pi & < & {\rm Arg}\pars{z} & < & \pi\,,\quad z \not= 0 \\[2mm] \pars{1 - z}^{-1/n} & = & \verts{1 - z}^{-1/n}\expo{-{\rm Arg}\pars{1 - z}/n}\,, & 0 & < & {\rm Arg}\pars{1 - z} & < & 2\pi\,,\quad z \not= 1 \end{array} $$ It's clear that the product $\ds{z^{1/n}\pars{1 - z}^{-1/n}}$ has a branch-cut in $\bracks{0,1}$.

The integral above the "upper line' is given by $$ \int_{1}^{0}x^{1/n - 1}\pars{1 - x}^{-1/n}\expo{-2\pi\ic/n}\,\dd x = -\expo{-2\pi\ic/n}\color{#00f}{\int_{0}^{1}x^{1/n}\pars{1 - x}^{-1/n}\,\dd x} $$ and in the "lower line" by $\ds{\color{#00f}{\int_{0}^{1}x^{1/n}\pars{1 - x}^{-1/n}\,\dd x}}$. Integration around the semicircles, with centers at $\ds{z = 0}$ and $\ds{z = 1}$, vanishes out when their radius $\ds{\to 0^{+}}$.

Then, \begin{align} &\pars{1 - \expo{-2\pi\ic/n}}\color{#00f}{% \int_{0}^{1}x^{1/n - 1}\pars{1 - x}^{-1/n}\,\dd x} =-2\pi\ic\, {\rm Res}_{\verts{z}\ \to\ \infty}\bracks{z^{1/n - 1}\pars{1 - z}^{-1/n}} \\[3mm]&=-2\pi\ic\,{\rm Res}_{z\ =\ 0}\bracks{% -\,{1 \over z^{2}}\,z^{1 - 1/n}\pars{1 - {1 \over z}}^{-1/n}} =2\pi\ic\,{\rm Res}_{z\ =\ 0}\bracks{\pars{z - 1}^{-1/n} \over z} \\[3mm]&=2\pi\ic\, {\rm Res}_{z\ =\ 0}\bracks{\verts{z - 1}^{\,-1/n}\expo{-\pi\ic/n} \over z} =2\pi\ic\expo{-\pi\ic/n} \end{align}

Then $$ \color{#00f}{\int_{0}^{1}x^{1/n - 1}\pars{1 - x}^{-1/n}\,\dd x} ={2\pi\ic\expo{-\pi\ic/n} \over 1 - \expo{-2\pi\ic/n}} =\pi\,{2\ic \over \expo{\pi\ic/n} - \expo{-\pi\ic/n}} ={\pi \over \sin\pars{\pi/n}} $$ which is replaced in expression $\pars{1}$:

$$ \color{#00f}{\large\int_{0}^{1}{\dd x \over \root[n]{1 - x^{n}}}\dd x ={\pi \over n\sin\pars{\pi/n}}} $$

Felix Marin
  • 89,464
2

From Omran Kouba's answer, we have \begin{eqnarray} I_n&=&\frac{1}{n}\int_{-\infty}^\infty \frac{e^{t/n}}{1+e^t}dt=\frac{1}{n}\int_{0}^\infty \frac{e^{t/n}}{1+e^t}dt+\frac{1}{n}\int_{-\infty}^0 \frac{e^{t/n}}{1+e^t}dt\\ &=&\frac{1}{n}\int_{0}^\infty \frac{e^{t/n-1}}{1+e^{-t}}dt+\frac{1}{n}\int_{-\infty}^0 \frac{e^{t/n}}{1+e^t}dt\\ &=&\frac{1}{n}\int_{0}^\infty \sum_{k=0}^\infty(-1)^ke^{t/n-t-kt}dt+\frac{1}{n}\int_{-\infty}^0 \sum_{k=0}^\infty(-1)^ke^{t/n+kt}dt\\ &=&\sum_{k=0}^\infty(-1)^k\frac{1}{nk+1}+\sum_{k=0}^\infty(-1)^k\frac{1}{nk+n-1}\\ &=&\sum_{k=-\infty}^\infty(-1)^k\frac{1}{nk+1}=\frac{\pi}{n\sin\frac{\pi}{n}}. \end{eqnarray}

xpaul
  • 44,000