0
$\begingroup$

I am asked to compute this integral in a sheet that treats the Residue's theorem. I've found the n singularities, calculated the residues, but for example for n=3 and $\alpha = 1$ I get that the sum of the residues is equal to 0, while Wolfram alpha tells me that the integral is different from zero. I am taking into account the branch cut. Any help would be appreciated. I found that the singularities are located at $z_k = e^{i/n (\pi + 2\pi k)}$ and Residue(s)=$\frac{1}{n}e^{(i/n)(\alpha-n+1)(\pi + 2\pi k). To compute the integral I would choose a contour that avoids the branch cut, going from R to a distance epsilon from the 0 just above the branch cut, then do a circle around zer of radius epsilon, then go from there to R under the brunch cut, and then do a circle of Radius R to close everything. I get that $\int_0^{+\infty}\frac{x^{\alpha}}{1+x^n}dx = \frac{e^{(i/n)(\alpha-n+1)(\pi+2\pi k}}{n(1-e^{2\pi i\alpha)}}$, where in the numerator we obviously have to sum over k. But like I said this isn't correct because for $\alpha$ natural this explodes, and this doesn't hold for a particular case of $n,\alpha$

$\endgroup$
  • $\begingroup$ Why don't you show your work? $\endgroup$ – Botond May 27 '18 at 20:45
  • $\begingroup$ Just did it, thanks for the suggestion $\endgroup$ – Paco Lopez May 27 '18 at 20:54
  • $\begingroup$ I don't quite understand your path, but I think it will be interesting for you: math.stackexchange.com/q/110457/281471 $\endgroup$ – Botond May 27 '18 at 21:09
  • $\begingroup$ Are you forced to use the residue theorem? Or can you use real analysis? It’s much easier by substituting $x=\tan^{2/n}\theta$ $\endgroup$ – Frank W May 27 '18 at 21:19
  • $\begingroup$ Yes, I'm forced to us the residue's theorem. $\endgroup$ – Paco Lopez May 28 '18 at 5:44
0
$\begingroup$

Important Note:

I have come to realize that the sum of residues is actually NOT $0$, but have yet to figure out how to compute it. But as I have already written most of this answer, I shall post it anyway.


Since we are going to be working with complex numbers, we will need to define the function $x^\alpha$.

Here, I take the branch cut to be along the positive real axis, so that we can make the definition

$$x^\alpha=|x|^\alpha e^{i \alpha \arg (x)} \qquad \arg (x) \in (0,2\pi)$$

Next, consider the keyhole contour as shown below

enter image description here

where the radius of $\rho$ is $R$, the radius of $\gamma_\epsilon$ is $\epsilon$, and $\gamma_+$ and $\gamma_-$ are respectively just above and just below the positive real axis.

For each $R$ and $\epsilon$, lets call this contour $\Gamma _{R,\epsilon}$.

Let $f(x)=\dfrac{x^\alpha}{1+x^n}$

We see that the singularities of $f$ are at $$x=\exp \bigg[\frac{i\pi}{n}(1+2k) \bigg] \qquad k=0,\dots ,n-1$$

They are all lined up on the unit circle, so as long as $\epsilon<1$ and $R>1$, all of them will be inside our contour.

NOTE: The sum of residues is NOT $0$. I am still trying to figure out how to find the sum of the residues, but for now, lets suppose that the sum of residues is $\Omega$.

Then, by Cauchy's Residue Theorem,

\begin{align} & \int_{\Gamma_{R,\epsilon}}f(x)dx=2\pi i \Omega \qquad \forall \; 0<\epsilon<1 \; \forall \; R>1 \\ \implies & \lim_{R \rightarrow \infty} \lim_{\epsilon \rightarrow 0} \int_{\Gamma_{R,\epsilon}}f(x)dx=2\pi i \Omega \tag{1} \end{align}

Now lets have a look at each individual component of $\Gamma_{R,\epsilon}$

$f$ has no singularities inside $\gamma_\epsilon$. However, it does go from one end of the branch cut to the other, so we cannot just claim that its integral is $0$. Instead, we can use the fact that $\epsilon \rightarrow 0$:

\begin{align} \bigg| \int_{\gamma_\epsilon}\frac{x^\alpha}{1+x^n}dx \bigg| & ≤ \int_{t=0}^{t=2\pi}\bigg| \frac{\epsilon^\alpha e^{it\alpha}}{1+\epsilon^n e^{itn}}i\epsilon e^{it} \bigg| dt \\ & = \epsilon^{\alpha+1} \int_{t=0}^{t=2\pi}\bigg| \frac{1}{1+\epsilon^n e^{itn}}\bigg| dt \\ & ≤ \epsilon^{\alpha+1} \int_{t=0}^{t=2\pi} \frac{1}{1-\epsilon^n}dt \\ & = \frac{2\pi \epsilon^{\alpha+1}}{1-\epsilon^n} \rightarrow 0 \quad \text{as} \; \epsilon \rightarrow 0 \end{align}

Hence

$$\lim_{\epsilon \rightarrow 0} \int_{\gamma_\epsilon} f(x)dx=0 \tag{2}$$

Of course, we can use a similar argument for the big arc $\rho _R$

$$\bigg| \int_{\rho_R}\frac{x^\alpha}{1+x^n}dx \bigg| ≤ \int_{t=0}^{t=2\pi}\bigg| \frac{R^\alpha e^{it\alpha}}{1+R^n e^{itn}}iR e^{it} \bigg| dt ≤ \int_{t=0}^{t=2\pi}\frac{R^{\alpha +1}}{1-R^n}dt=\frac{2\pi R^{\alpha +1}}{1-R^n} \rightarrow 0 \; \text{as} \; R\rightarrow \infty$$

where we have used the fact that $\alpha +1<n$. It follows that

$$\lim_{R \rightarrow \infty} \int_{\rho_R} f(x)dx=0 \tag{3}$$

Next, we observe that the integral along $\gamma_+$ is the one we want. As for the integral along $\gamma_-$, we see that it is parameterised in the opposite direction as $\gamma_+$.

Moreover, look back at our definition of $x^\alpha$. If we were to take a point $x_+$ just above the real axis, and its "corresponding point" $x_-$ just below the real axis, we see that

$$x_-^\alpha = |x_-|^\alpha e^{i\alpha (2\pi)}=e^{i\alpha (2\pi)}\bigg(|x_+|^\alpha e^{i\alpha (0)}\bigg)=e^{i\alpha (2\pi)}x_+^\alpha$$

Hence, we have the relationship

$$\int_{\gamma_-}f(x)dx=-e^{i\alpha (2\pi)}\int_{\gamma_+}f(x)dx \tag{4}$$

FINALLY, if we plug $(2)$, $(3)$ and $(4)$ back into $(1)$, we get

\begin{align} & \lim_{R \rightarrow \infty} \lim_{\epsilon \rightarrow 0}\int_{\Gamma_{R,\epsilon}}f(x)dx=2\pi i \Omega \\ \implies & \lim_{R \rightarrow \infty} \lim_{\epsilon \rightarrow 0} \bigg(\int_{\gamma_\epsilon \, +\, \rho_R \,+\, \gamma_+ \,+\, \gamma_-} f(x)dx \bigg) =2\pi i \Omega \\ \implies & (0)+(0)+\lim_{R \rightarrow \infty} \lim_{\epsilon \rightarrow 0} \bigg(\int_{\gamma_+} f(x)dx + \int_{\gamma_+} f(x)dx \bigg)=2\pi i \Omega \\ \implies & \lim_{R \rightarrow \infty} \lim_{\epsilon \rightarrow 0} \big(1-e^{2\pi i \alpha} \big) \int_{\gamma_+}f(x)dx=2\pi i \Omega \\ \implies & \int_0^\infty \frac{x^\alpha}{1+x^n}dx=\frac{2\pi i \Omega}{1-e^{2\pi i \alpha}} \end{align}

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Thanks glowstonetrees, that's exactly what I did and the result that I got, but as you can see from your result for $\alpha$ natural the fraction explodes, and for n=3 $\alpha =1$ I get that the sum of the residues is 0. $\endgroup$ – Paco Lopez May 28 '18 at 5:43
0
$\begingroup$

Too long for a comment and moreover using real analysis (I faced this problem years ago).

As given by Wolfram Alpha $$\int\frac{x^a}{1+x^n}\,dx=\frac{x^{a+1} }{a+1}\, _2F_1\left(1,\frac{a+1}{n};\frac{n+a+1}{n};-x^n\right)$$ Using the given bounds $$\int_0^\infty\frac{x^a}{1+x^n}\,dx=\frac{\pi }{n}\csc \left(\frac{\pi (a+1)}{n}\right)\qquad \text{if}\qquad \Re(a-n)<-1\land \Re(a)>-1$$ Amazing is also $$\int_1^\infty\frac{x^a}{1+x^n}\,dx=\frac{H_{-\frac{a+1}{2 n}}-H_{-\frac{n+a+1}{2 n}}}{2 n}\qquad \text{if}\qquad \Re(a-n)<-1\land \Re(a)>-1$$

As Frank W. commented, the problem is much simpler using $x=\tan^{2/n}(t)$ which reduces the problem to $$\frac2n\int_0^{\frac \pi 2} \tan ^{\frac{2 (a+1)}{n}-1}(t)\,dt$$

| cite | improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.