I am trying to calculate integrals of the form:

$$ I(t, n) = \int_{-\infty}^{\infty} \Big(\frac{1}{1-jq}\big)^{n} e^{-jqt} dq $$

where $j = \sqrt{-1}$. In the case when $n=1$, I have:

$$ I(t, 1) = \int_{-\infty}^{\infty} \frac{1}{1-jq} e^{-jqt} dq $$

Now, my thought was to view this as a function of $t$, and use the Feynmann trick. Ie:

$$ \frac{d}{dt} I(t, 1) = \int_{-\infty}^{\infty} \frac{\partial}{\partial t} \big(\frac{1}{1-jq} e^{-jqt} \big) dq $$ $$ \frac{d}{dt} I(t, 1) = \int_{-\infty}^{\infty} \frac{-jq}{1-jq} e^{-jqt} dq $$

This looks promising, but I can't get it to go anywhere. An even more promising avenue seems to be to express the exponential as a power series around zero. This gives:

$$ I(t, 1) = \int_{-\infty}^{\infty} \frac{1}{1-jq} \sum_{k=0}^{\infty} \frac{(-jqx)^{k}}{k!} dq $$

Since the complex exponential is entire, we can interchange the sum and integral, giving:

$$ I(t, 1) = \sum_{k=0}^{\infty} \frac{x^{k}}{k!} \int_{-\infty}^{\infty} \frac{(-jq)^{k}}{1-jq} dq $$

However, I get stuck here too because I can't find an antiderivative for the integrand.

Any ideas? Am I missing some fundamental theoretical concept or trick or technique that makes all of this difficulty disappear?

Unfortunately, I am really not that good at integration, but I want to get better!


For simplicity we rename $j\to i, q \to -z $ to rewrite the integral as $$\begin{align} \int_{-\infty}^\infty\frac{e^{itz}dz}{(1+iz)^n}&=\int_\Gamma\frac{e^{itz}dz}{(1+iz)^n}=2\pi i\;\underset{z=i}{\text {Res}}\,\frac{e^{itz}}{(1+iz)^n}\\ &=\frac {2\pi}{ i^{n-1}}\left.\frac1 {(n-1)!}\frac {d^{n-1}}{dz^{n-1}} e^{itz}\right|_{z=i}=2\pi\frac{ t^{n-1}e^{-t}}{(n-1)!},\end{align} $$ where $\Gamma $ is the usual counterclockwise-oriented contour consisting of the real axis and a large semicircle in the upper complex half-plane. $t$ is assumed to be positive real number.

  • $\begingroup$ This seems to solve it. Thank you very much. I need to revisit my contour integration! So what happens if t is a nonpositive real number? Also, if q = -z, then shouldnt dq = -dz ? But it doesnt matter because of the limits of integration, right? Sorry, for the basic questions, but the only way to get better is to ask when I'm unsure. $\endgroup$ – The Dude Feb 21 at 22:30
  • $\begingroup$ If $t $ is negative the coutour should lie in the lower half-plane, the pole is outside of the contour and the integral is 0. The negation of the integration variable and interchange of the limits always compensate each other. $\endgroup$ – user Feb 21 at 22:45
  • $\begingroup$ Thanks for your reply. I will have to go back to my notes and go over this. This was very helpful. $\endgroup$ – The Dude Feb 21 at 22:58
  • $\begingroup$ Okay, I spent a few days reviewing this and now have a question -- where does the $\frac{1}{i^{n-1}}$ come from? Shouldn't that be in the numerator when you take the derivative N-1 times? Why do you divide it out beforehand? $\endgroup$ – The Dude Feb 26 at 16:07
  • $\begingroup$ @TheDude $\frac1{(1+iz)^n}=\frac1{i^n (z-i)^n}$. $\endgroup$ – user Feb 26 at 16:29

If we define the Fourier transform of a function $f$ and its inverse as $$\mathcal F(f)(\omega)=\int_{\mathbb R}f(x)e^{-j\omega x}dx \,\,\,\text{ and }\,\,\, f(x)=\frac 1 {2\pi}\int_{\mathbb R} \mathcal F(f)(\omega)e^{jx\omega}d\omega$$ then you're looking for the Fourier transform of $f_n(x)=f(x)^n$ where $f(x)=\frac{1}{1-jx}$.

Because the Fourier transform maps products to convolutions (times $\frac 1 {2\pi}$, given the Fourier definition we adopted), you're looking for the $n$-th self-convolution of $\mathcal F(f)$. Now, with $H$ denoting the Heaviside step function, we have $$f(x)=\frac{1}{1-jx}=\int_{\mathbb R}e^{-\omega}H(\omega)e^{jx\omega}d\omega$$ As a consequence, the Fourier transform of $f$ is $\omega\rightarrow 2\pi e^{-\omega}H(\omega)$.

This gives us $$I(t, 1)=2\pi e^{-t}H(t)$$ and you can verify that the $n$-th self-convolution is given by $$I(t,n)=2\pi\frac{t^{n-1}}{(n-1)!}e^{-t}H(t)$$

  • $\begingroup$ Where did you find this fourier transform pair? After looking this up, I only found the first result. I'm not sure still how the (n-1)! comes out of this ,,Fourier Matching'' as I like to call it. $\endgroup$ – The Dude Feb 22 at 14:34
  • $\begingroup$ It's a known Fourier transform. Then computing the convolution is not too difficult. You can prove it by induction after trying the first few values of $n$. $\endgroup$ – Stefan Lafon Feb 22 at 14:36
  • $\begingroup$ Well I guess I am just a n00b. Okay, fair enough. I gotta practice more. $\endgroup$ – The Dude Feb 22 at 14:38
  • $\begingroup$ Though the factorial is still confusing... $\endgroup$ – The Dude Feb 22 at 14:38
  • $\begingroup$ It comes from integrating powers of $\omega$. $\endgroup$ – Stefan Lafon Feb 22 at 14:41

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.