3
$\begingroup$

How do you show that: $$ \int_{0}^{\infty} \frac{\sin{x^{q}}}{x^{q}} dx = \frac{\Gamma{\frac{1}{q}}}{q-1}\cos{\frac{\pi}{2q}} \mbox{, q > 1} $$ Without using Gamma function?

$\endgroup$
  • 5
    $\begingroup$ I'm puzzled by how you're going to end up with a result with a Gamma function in it... without using the Gamma function. $\endgroup$ – Chappers May 11 '15 at 17:16
  • 2
    $\begingroup$ @Chappers I suppose get something like: $\left( \frac{1}{q}- 1 \right)! = \Gamma{\frac{1}{q}}$, but that is only the final part, I guess $\endgroup$ – antoniov.joel May 11 '15 at 17:20
  • 1
    $\begingroup$ Well, if $q$ isn't an integer, you don't have much choice. $\endgroup$ – Chappers May 11 '15 at 17:21
  • 2
    $\begingroup$ @Chappers That's because I wanna know if there exists a way to get it, if no there one solution, that's ok, it's just curious $\endgroup$ – antoniov.joel May 11 '15 at 17:27
  • $\begingroup$ As far as I know, there are two ways of doing this integral: contour integration, and interchanging an order of integration. I'm not sure which one you know of. $\endgroup$ – Chappers May 11 '15 at 17:29
4
$\begingroup$

Yet another approach: applying Ramanujan's master theorem which states that the Mellin transform of a function that admits the expansion of the form $$f(x)=\sum_{n=0}^\infty \frac{(-1)^n x^n}{n!}\phi(n) $$

is given by$$\int_0^\infty x^{s-1}f(x)dx = \Gamma(s)\phi(-s) .$$ Since $$\sin(x)= \sum_{n=0}^\infty \frac{(-1)^n x^{2n+1}}{(2n+1)!}$$ we have $$\frac{\sin(\sqrt{x})}{\sqrt{x}}=\sum_{n=0}^\infty \frac{(-1)^nx^n}{n!}\frac{\Gamma(n+1)}{\Gamma(2n+2)}$$ and therefore, for $0<s<1$: $$\int_0^\infty x^{s-1}\frac{\sin(\sqrt{x})}{\sqrt{x}}dx=\frac{\Gamma(s)\Gamma(1-s)}{\Gamma(2-2s)}=\frac{\pi}{\sin(\pi s)\Gamma(2-2s)}$$ Substituting $t=\sqrt{x}$, $2s=\frac{1}{q}$ and multiplying by $\frac{1}{q}$ we get $$I=\frac{1}{q}\int_0^\infty x^{\frac{1}{q}-2}\sin(x)dx=\frac{\pi}{2q\sin(\frac{\pi}{2q})\Gamma(2-\frac{1}{q})}$$ Which you can match to your proposed form by multiplying by $$\frac{\Gamma(\frac{1}{q}-1)}{\Gamma(\frac{1}{q}-1)}\frac{(\frac{1}{q}-1)}{(\frac{1}{q}-1)}$$

$\endgroup$
3
$\begingroup$

As said by Chappers in the comments, to prove a result involving a $\Gamma$ function without even mentioning it is quite pointless. Anyway assuming $q>1$, $$I=\int_{0}^{+\infty}\frac{\sin(x^q)}{x^q}\,dx = \frac{1}{q}\int_{0}^{+\infty}\frac{\sin z}{z^{2-\frac{1}{q}}}\,dz \tag{1}$$ and by exploiting the following Laplace (direct and inverse) transforms: $$\mathcal{L}^{-1}\left(\frac{1}{z^{2-\frac{1}{q}}}\right)=\Gamma\left(2-\frac{1}{q}\right)^{-1} s^{\frac{q-1}{q}},\qquad\mathcal{L}(\sin z)=\frac{1}{1+s^2}\tag{2}$$ we get: $$ I = \frac{1}{q\cdot\Gamma\left(2-\frac{1}{q}\right)}\int_{0}^{+\infty}\frac{s^{1-\frac{1}{q}}}{s^2+1}\,ds = \frac{1}{\Gamma\left(2-\frac{1}{q}\right)}\int_{0}^{+\infty}\frac{t^{2q-2}}{t^{2q}+1}\,dt\tag{3} $$ and the last integral can be easily computed through the residue theorem or the substitution $\frac{1}{t^{2q}+1}=u$, leading to a value of the Beta function that can be written as a cosine in virtue of the $\Gamma$ duplication formula.

$\endgroup$
  • $\begingroup$ (+1), but I suspect this is just using $\frac{1}{A^n} = \int_0^{\infty} \alpha^{n-1} e^{-A\alpha} \, d\alpha$ in disguise... $\endgroup$ – Chappers May 11 '15 at 18:02
  • $\begingroup$ @Chappers: it is, indeed! $\endgroup$ – Jack D'Aurizio May 11 '15 at 18:12
2
$\begingroup$

First, set $y=x^q$, to clear out the nasty argument of the sine. Then the integral is $$ \frac{1}{q}\int_0^{\infty} y^{1/q-2}\sin{y} \, dy $$ Write the sine in terms of complex exponentials as $\frac{e^{iy}-e^{-iy}}{2i}$ in the usual way. Now, I shall run through the $e^{iy}$ one: the other can be done in the same way in the lower half-plane. Set $p=1/q-1$ to save writing.

We consider the integral $$ \int_{\gamma} z^{p-1} e^{iz} \, dz, $$ where $\gamma$ is the contour made out of the real axis between $\varepsilon $ and $R$, the quarter-circle $z=Re^{i\theta}$, $0<\theta<\pi/2$, the imaginary axis from $iR$ to $i\varepsilon$, and finally the small quarter-circle $\varepsilon e^{i\pi/2-i\theta}$, $0<\theta<\pi/2$. It is easy to show that for $0<p<1$, the small semicircle integral tends to zero as $\varepsilon \to 0$. The large semicircle integral can be shown to converge to zero as $R \to \infty$ in the same way as Jordan's lemma is proven. Since the integrand has no singularities inside the contour, it follows that $$ \int_0^{\infty} y^{p-1} e^{iy} \, dy = - \left(- \int_{\infty}^{0} (e^{i\pi/2}y)^{p-1} e^{-y} \, (-i) dy \right) = i e^{i (p-1) \pi/2} \Gamma(p) \\ = i e^{i \pi / (2q)} \Gamma(1/q-1) = i e^{i \pi / (2q)} \frac{q}{q-1}\Gamma(1/q). $$ Nearly there: the sensible thing to do at this point is to notice that you can take the imaginary part of the left-hand side to get the original integral, and hence $$ \frac{1}{q} \int_0^{\infty} y^{1/q-2} \sin{y} \, dy = \frac{1}{q-1}\cos{\left(\frac{\pi}{2q}\right)} \Gamma(1/q) $$

$\endgroup$

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.