I have stumbled into a issue with gamma function, I will show the approach first: $$\Gamma(x)=\int_0^{\infty}t^{x-1}e^{-t}dt$$ subtituting $t=iu^2$ gives: $$\Gamma(x)=2\int_0^{\infty}(iu^2)^{x-1}e^{-iu^2}iudu\rightarrow\frac{\Gamma(x)}{2i^x}=\int_0^{\infty}u^{2x-1}e^{-iu^2}du$$ Doing the same thing using $t=-iu^2\,$results in$$\frac{\Gamma(x)}{-2i^x}=\int_0^{\infty}u^{2x-1}e^{iu^2}du$$ Now, summing those two and using that $i^x=e^{\frac{i\pi}{2}x} \,$gives $$\frac{\Gamma(x)}{2}(e^{\frac{i\pi}{2}x}+e^{\frac{-i\pi}{2}x})=\int_0^{\infty}u^{2x-1}(e^{iu^2}+e^{-iu^2})du$$ which is just $$\frac{\Gamma(x)}{2}\cos(\frac{\pi}{2}x)=\int_0^{\infty}u^{2x-1}\cos(u^2)du$$ plugging $x=-\frac{1}{2}$ we get that$$\int_0^{\infty}\frac{\cos(x^2)}{x^2}dx=-\sqrt{\frac{\pi}{2}}$$ Well, obviously this integral diverges... But if Instead of summing we subtract we get that $$\frac{\Gamma(x)}{2}\sin(\frac{\pi}{2}x)=\int_0^{\infty}u^{2x-1}\sin(u^2)du$$ simmilarly with $$x=-\frac12 \rightarrow \int_0^{\infty}\frac{\sin(x^2)}{x^2}dx=\sqrt{\frac{\pi}{2}}$$ So its not that completely garbage. Now my question is, what goes wrong when I use the first substitution? And how do I prove that I am allowed to use this substitution for the sine integral?

  • 1
    $\begingroup$ Your upper integration limit is incorrect after the change of variables. The correct intrgration limit are $u=0$ and $u=e^{-i\pi/4}\infty$. Attempt to deform the contour onto the real line fails. $\endgroup$ – Mark Viola Jul 24 '18 at 20:24

When you make a change of variables in a definite integral, the integration limits change as well.

The question now is why your formulas are correct. The original integral converges for $\operatorname{Re} x > 0.$ When additionally $\operatorname{Re} x < 1$, the integral of $u^{2x-1} \exp(-i u^2)$ over the arc of a large circle between $\arg u = -\pi/4$ and $\arg u = 0$ is negligible and we don't have to take any singularities into account, therefore the integral over $[0, e^{-i \pi/4} \infty)$ is the same as the integral over $[0, \infty)$.

Similar reasoning for the second integral $-$ this time starting from $[0, e^{i \pi/4} \infty)$ $-$ shows that both the sine and the cosine formulas are correct for $0 < \operatorname{Re} x < 1$.

The integral of $u^{2x-1} \sin(u^2)$ is an analytic function on $-1 < \operatorname{Re} x < 1$, and the reason why the sine formula is correct for $-1 < \operatorname{Re} x < 1$ is the uniqueness of analytic continuation.

The cosine formula will be correct for $x = -1/2$ if you choose the regularization of the divergent integral that coincides with the analytic continuation. For $-2 < \operatorname{Re} x < 0$, that regularization is the integral of $u^{2x-1} (\cos(u^2) - 1)$.

  • $\begingroup$ Thanks. Well I changed the limits but I took $e^{-i\pi/4}\infty$ as $\infty$. My contour integration knowledge is low as I studied only from wikipedia. Is there a proof for the uniqueness of analytic continuation? $\endgroup$ – Number Jul 30 '18 at 8:09
  • $\begingroup$ It's easier to see how the upper limit changes if you write the original integral as $\lim_{R \to \infty} \int_0^R$. There is a proof for the identity theorem here. $\endgroup$ – Maxim Jul 30 '18 at 13:10

$\small\underline{\text{For}\,\,Re\{s\}\gt0}$ : $$ \Gamma(s)=\int_{0}^{\infty}\frac{x^{s-1}}{e^x}\,dx $$ Substitute $\{\,x=+it^2\quad\text{&}\quad x=-it^2\,\}$, subtract, add, and simplify to get: $$ \begin{align} \frac{\Gamma(s)}{2}\,\sin\left(\frac{\pi}{2}s\right) &=\int_{0}^{\infty}\frac{\sin\left(x^2\right)}{x^{1-2s}}\,dx \\[2mm] \frac{\Gamma(s)}{2}\,\cos\left(\frac{\pi}{2}s\right) &=\int_{0}^{\infty}\frac{\cos\left(x^2\right)}{x^{1-2s}}\,dx \end{align} $$

$\small\underline{\text{For}\,\,-1\lt Re\{s\}\lt0}$ : $$ \Gamma(s)=\int_{0}^{\infty}x^{s-1}\left(\frac{1}{e^x}\color{red}{-1}\right)\,dx $$ Substitute $\{\,x=+it^2\quad\text{&}\quad x=-it^2\,\}$, subtract, add, and simplify to get: $$ \begin{align} \frac{\Gamma(s)}{2}\,\sin\left(\frac{\pi}{2}s\right) &=\int_{0}^{\infty}\frac{\sin\left(x^2\right)}{x^{1-2s}}\,dx \\[2mm] \frac{\Gamma(s)}{2}\,\cos\left(\frac{\pi}{2}s\right) &=\int_{0}^{\infty}\frac{\cos\left(x^2\right)\color{red}{-1}}{x^{1-2s}}\,dx \end{align} $$

$\{\,\color{red}{-1}\,\}$ cancelled each other in sine, and added to each other in cosine.

Nevertheless, do not forget to re-calculate the integration limits whenever you change the integration variable.

$$ \small\Gamma(s-N)=\int_0^\infty x^{s-1-N}\,\left[\,\frac1{e^x}-\sum_{n=0}^N (-1)^n\,\frac{x^n}{n!}\,\right]\,dx \quad\colon -1\lt Re\{s\}\lt0,\,\,N\in\{0,\,1,\,2,\,\dots\,\} $$

  • $\begingroup$ Thank you, but how did you see that its valid for $-1\lt Re\{s\}\lt0$? $\endgroup$ – Number Jul 30 '18 at 8:08
  • $\begingroup$ @HazemOrabi Note that your second and third formulas are not valid for all $s$ in the right half-plane, the integrals diverge for $\operatorname{Re} s \geq 1$. $\endgroup$ – Maxim Jul 30 '18 at 13:16

What you get when you substitute, say, $t=iu$ into an integral $$ \int_a^b f(t)\; dt$$ where $a, b \in \mathbb R$ is an integral over a path in the complex plane $$ i \int_C f(iu)\; du$$ where $C$ is a path consisting of points mapped into the interval $[a,b]$ by the mapping $u \mapsto iu$.

  • $\begingroup$ I didnt learn about mapping. So basically like an integral transform? I looked here en.wikipedia.org/wiki/Map_(mathematics)#Maps_as_functions $\endgroup$ – Number Jul 24 '18 at 20:12
  • $\begingroup$ Simply the straight line from $a/i = -ai$ to $b/i = -bi$, i.e. all $u$ such that $iu \in [a,b]$. $\endgroup$ – Robert Israel Jul 24 '18 at 23:35
  • $\begingroup$ Sorry, but I dont see how does this help me. $\endgroup$ – Number Jul 26 '18 at 18:09

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.