I want to evaluate the following integral $$\int_0^\infty x^{t-1}e^{-\beta x}dx$$ where $\beta$ is a complex number.

Now, if $\beta$ was real, we could just set $y = \beta x$ and we will reduce to the Gamma function. Since $\beta$ is complex, though, when I set $y = \beta x$, I am integrating over the line with $\arg \beta$ on the complex plane, so I can't reduce directly to the Gamma function, can I?

I have found after some calculations that $$\int_0^\infty x^{t-1}e^{-\beta x}dx = \Gamma(t)\beta^{-t}$$ which is exactly what one would find if it didn't bother with the previous observation.

So my questions are: Is my observation on the complex line correct? and 2) What is the best way to prove the result?

My work

Write $\beta = a + ib$. Consider the integral as a function of $t,a,b$ to get $$I(t,a,b) = \int_0^\infty x^{t-1}e^{-a x}e^{-ibx}dx$$.

Notice that $$\frac{\partial I}{\partial a}(t,a,b) = -I(t+1,a,b)$$ and $$\frac{\partial I}{\partial b}(t,a,b) = -iI(t+1,a,b)$$

Now since $\displaystyle I(t+1) = \frac t{a+ib}I(t)$, the previous two equations become $$\frac{\partial I}{\partial a} = -\frac t{a+ib}I$$ and $$\frac{\partial I}{\partial b} = -\frac{it}{a+ib}I$$

which put together yield $I(t,a,b) = C(t) (a+ib)^{-t}$. Also, since $\displaystyle I(t,1,0) = C(t) = \int_0^\infty x^{t-1}e^{-x}dx= \Gamma(t)$, we get

$$I(t,a,b) = \Gamma(t) (a+ib)^{-t} = \Gamma(t)\beta^{-t}$$

which seems like too much work!

  • 1
    $\begingroup$ You can also first prove that the integral is analytic e.g. using these methods and then the result follows from analytic continuation being unique. $\endgroup$ – Count Iblis Aug 16 '15 at 20:21

You're correct in the result. To verify this and to provide an instructive way forward, we present here a direct approach the relies on Cauchy's Integral Theorem.

To that end, we let $I$ be the integral in the complex plane given by

$$I=\oint_C z^{t-1}e^{-\beta z}\,dz$$

where $\text{Re}(\beta)>0$, $t>0$, and $C=C_1+C_2+C_3$ is the contour comprised of the $4$ components

$(i)$ $C_1$ is the real-line segment from $(\epsilon,0)$ to $(R,0)$;

$(ii)$ $C_2$ is the circular arc centered at the origin with radius $R$ from $(R,0)$ to $(R\cos \theta_0,R\sin \theta_0)$ where $\beta e^{i\theta_0}=1 \implies e^{i\theta_0}=\frac{1}{\beta}$.

$(iii)$ $C_3$ is the line segment from $(R\cos \theta_0,R\sin \theta_0)$ to $(\epsilon \cos \theta,\epsilon \sin \theta)$;

$(iv)$ $C_4$ is the circular arc centered at the origin with radius $\epsilon$ from $(\epsilon \cos \theta,\epsilon \sin \theta)$ to $(\epsilon,0)$.

Since $z^{x-1}e^{-\beta z}$ is analytic in $C$, then from Cauchy's Integral Theorem, we have $I=0$. Moreover, the contributions from $C_2$ and $C_4$ can easily be shown to vanish as $R\to \infty$ and $\epsilon \to 0$, respectively.

Therefore, we have

$$\begin{align} I&=\oint_C z^{t-1}e^{-\beta z}\,dz\\\\ &=\int_0^{\infty}(x)^{t-1}e^{-\beta x}\,dx+\int_\infty^0 (e^{i\theta_0}x)^{t-1}e^{-\beta e^{i\theta_0}x}e^{i\theta_0}dx=0\\\\ \implies \int_0^{\infty}(x)^{t-1}e^{-\beta x}\,dx&=\frac{1}{\beta^{t}}\int_0^\infty x^{t-1}e^{-x}\,dx\\\\ &=\frac{1}{\beta^{t}}\Gamma(t) \end{align}$$

which was to be shown!

  • $\begingroup$ veri nice indeed. Thank you! :-) $\endgroup$ – Ant Aug 18 '15 at 9:18
  • $\begingroup$ @ant Thank you! Much appreciative. And you're welcome. My pleasure. $\endgroup$ – Mark Viola Aug 18 '15 at 14:21

$$ f(\beta) = \int_0^\infty x^{t-1}e^{-\beta x}dx$$ is clearly an analytic function in $\beta$ if $\text{Re}(\beta) > 0$ (differentiate under the integral sign). $$ g(\beta) = \Gamma(t) \beta^{-t} $$ is also analytic in $\beta$. Since $f(\beta) = g(\beta)$ for all $\beta \in (0,\infty)$, it follows by the principle of isolated zeros that they must be equal on the intersections of their domains of definition.

  • $\begingroup$ Analytic continuation is one of the coolest things ever. I still get excited whenever I see it in practice. $\endgroup$ – Cameron Williams Aug 17 '15 at 1:20
  • 1
    $\begingroup$ I was under the belief that the OP wanted to verify that $f=g$. $\endgroup$ – Mark Viola Aug 17 '15 at 1:30
  • $\begingroup$ @Dr.MV It is trivial by change of variables if $\beta$ is real, as the OP pointed out. I used analytic continuation to show it is true for all other $\beta$. $\endgroup$ – Stephen Montgomery-Smith Aug 17 '15 at 1:49
  • $\begingroup$ Your answer states that $f=g$ for all $\beta$, but how does one reach that conclusion immediately. Perhaps one can say $f=g$ for real $\beta>0$ and then use AC to conclude that $f=g$ for all $\beta$ with positive real part. $\endgroup$ – Mark Viola Aug 17 '15 at 1:55
  • $\begingroup$ @Dr.MV As it happens, the integral is undefined for the real part of $\beta$ less than or equal to 0. Nevertheless, it if were defined outside of that range, one could still use analytic continuation. $\endgroup$ – Stephen Montgomery-Smith Aug 17 '15 at 2:01

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.