It's a very straightforward problem. I have to compute the Fourier series of


I have done it using residues and I have got the following coefficients:

$$c_n = 2 ^{-|n|}$$

I think they are correct, but when I checked with Wolfram using the FourierSeries function, the result seems to be wrong. But if I compute the integral directly in Wolfram, the output then is ok. Could it be that the function FourierSeries and FourierCoefficient have a bug? Anyone has used them?

Thanks a lot, because I have consumed a lot of time trying to figure out my mistake...

For example, using this url.

And this one.

You can see a different result, and I think it should give the same output.

EDIT: The people from Technical Support in Wolfram have confirmed by mail there was a bug and are trying to fix it.


There does seem to be an error with WA's Fourier Series calculator for the function $\frac{3}{5-4\cos(x)}$. See this link.

We can proceed to develop the Fourier coefficients by noting that

$$\begin{align} \frac1{2\pi}\int_0^{2\pi}\frac{3}{5-4\cos(x)}e^{-inx}\,dx&=\frac1{2\pi i}\left(\oint_{|z|=1}\frac{z^{-n}}{z-1/2}\,dz-\oint_{|z|=1}\frac{z^{-n}}{z-2}\,dz \right)\tag1 \end{align}$$

For $n\le 0$, the first and second integrals on the right-hand side of $(1)$ have first order poles at $z=1/2$ and $z=2$, respectively. Hence, we have from the residue theorem

$$\frac1{2\pi}\int_0^{2\pi}\frac{3}{5-4\cos(x)}e^{-inx}\,dx= \left(\frac12\right)^n$$

For $n\ge 1$, we can make use of symmetry (evenness and $2\pi$-periodicity) and write

$$\begin{align} \frac1{2\pi}\int_0^{2\pi}\frac{3}{5-4\cos(x)}e^{-inx}\,dx&=\frac1{2\pi}\int_0^{2\pi}\frac{3}{5-4\cos(x)}e^{-i|n|x}\,dx\\\\ \end{align}$$

from which we immediately find that for $n\ge 1$

$$\frac1{2\pi}\int_0^{2\pi}\frac{3}{5-4\cos(x)}e^{-inx}\,dx= \left(\frac12\right)^{-n}$$

Therefore, we have

$$\begin{align} \frac{3}{5-4\cos(x)}&=\sum_{n=-\infty}^\infty \left(\frac12\right)^{|n|} e^{inx}\\\\ &=1 +2\sum_{n=1}^\infty \left(\frac12\right)^n\cos(nx) \end{align}$$

  • $\begingroup$ Thanks a lot! Yes, that's the way I derived the coefficients, so it seems WA is giving a wrong output. At first of course I thougth that the mistake was on my part, because the FourierSeries calculator its surely commonly used, and the function it's not a complicated one... I have written an email to them to report the apparent bug. $\endgroup$ – Jaime Arboleda Castilla May 10 '18 at 7:04
  • $\begingroup$ You're welcome. My pleasure. $\endgroup$ – Mark Viola May 10 '18 at 13:41
  • $\begingroup$ Ok, I voted up, but I did not know that I needed to tick also to accept the answer (I'm more or less new here). Thanks again! :) $\endgroup$ – Jaime Arboleda Castilla May 10 '18 at 20:34
  • $\begingroup$ No, there can only be one answer? I did not know. $\endgroup$ – Jaime Arboleda Castilla May 13 '18 at 18:47

A more direct computation without Fourier integrals uses partial fraction decomposition after inserting $\cos x = \frac12(e^{ix}+e^{-ix})$ and $e^{ix}=z$ and geometric series for the factors \begin{align} \frac{3}{5-2(z+z^{-1})}&=-\frac{(4z-2)-(z-2)}{(2z-1)(z-2)}=\frac{1}{2z-1}-\frac2{z-2}\\ &=\frac1{2z}\sum_{k=0}^\infty \left(\frac{1}{2z}\right)^k+\sum_{k=0}^\infty \left(\frac{z}{2}\right)^k\\ &=\frac12+\sum_{k=1}^\infty\frac{z^k+z^{-k}}{2^{k}}\\ &=\frac12+\sum_{k=1}^\infty\frac{\cos(kx)}{2^{k-1}}. \end{align}


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.