8

Evaluate the following definite integral:

$$ \int_0^\pi \left( \frac{2 + 2\cos (x) - \cos((k-1)x) - 2\cos (kx) - \cos((k+1)x)}{1-\cos(2x)}\right) \mathrm{d}x, $$ where $k \in \mathbb{N}_{>0}$.

Dal
  • 8,214

3 Answers3

18

Let

$$I_k = \int_0^\pi \left( \frac{2 + 2\cos x - \cos(k-1)x - 2\cos kx - \cos(k+1)x}{1 - \cos2x}\right) \mathrm dx.$$

Then, we have:

  • $I_0 = 0$;

  • $I_1 = \pi$.

For any $ k \in \mathbb{N}^*$, we have: $$ \begin{align} I_{k+1} - 2I_k + I_{k-1} &= \int_0^\pi \left( \frac{\cos(-2+k)x - 2\cos kx + \cos (2 + k)x}{-1 + \cos 2x}\right) \mathrm dx \\ &= \int_0^\pi \left( \frac{2\cos kx(-1 + \cos^2x)}{-1 + \cos 2x}\right) \mathrm dx =0 \end{align}$$ Since $ I_{k+1} - I_k = I_k - I_{k-1} $, $(I_k)$ is an arithmetic progression. Hence $ I_k = k\pi $.

  • N.B.: I remember that this trick was used to solve a similar integral from Teodora-Liliana T. Rădulescu, Vicenţiu D. Rădulescu, Titu Andreescu, Problems in Real Analysis: Advanced Calculus on the Real Axis, Springer, New York, 2009. Later, I will check and write a precise reference. –  Oct 09 '14 at 16:00
  • That's clever. Thank you very much. – Dal Oct 09 '14 at 16:46
  • 3
    Nice use of recursion. – Ron Gordon Oct 10 '14 at 13:17
  • 1
    Bonus problem: for half-integral values of $k$ the answers are rational numbers. Can you prove this/give a closed form? – Igor Rivin Oct 10 '14 at 13:41
  • @IgorRivin by half-integral you mean of the form $n+\frac{1}{2}$? – Dal Oct 10 '14 at 21:28
  • @Dal yes, exactly. – Igor Rivin Oct 10 '14 at 23:46
  • @RonGordon Thank you very much. –  Oct 11 '14 at 08:37
  • @IgorRivin Thanks for your interesting question: I will surely look into the problem when I have some free time. –  Oct 11 '14 at 08:49
4

I think we can get the answer into a relatively elementary form. Observe that

$$\cos{(k-1) x} + 2 \cos{k x} + \cos{(k+1) x} = 2 \cos{k x} (1+ \cos{x})$$

Thus the integrand simplifies into

$$4 \int_0^{\pi} dx \cos^2{\frac{x}{2}} \frac{\sin^2{k \frac{x}{2}}}{\sin^2{x}} $$

which is equal to

$$\frac12 \int_0^{2 \pi} du \left (\frac{\sin{k u}}{\sin{u}} \right )^2 $$

This integral may be evaluated using the residue theorem by converting it to a complex integral:

$$-\frac{i}{2} \oint_{|z|=1} \frac{dz}{z^{2 k-1}} \left (\frac{z^{2 k}-1}{z^2-1} \right )^2$$

The only (non-removable) pole is at $z=0$, but this pole has multiplicity $2 k-1$. Thus, by the residue theorem, the integral is equal to

$$\frac{\pi}{(2 k-2)!} \left [\frac{d^{2 k-2}}{dz^{2 k-2}} \left (\frac{z^{2 k}-1}{z^2-1} \right )^2 \right ]_{z=0} = \frac{\pi}{(2 k-2)!} \left [\frac{d^{2 k-2}}{dz^{2 k-2}} \left (1+z^2+z^4+\cdots z^{2 k-2} \right )^2 \right ]_{z=0}$$

Now, this may be evaluated as follows. The derivative term is equal to

$$2 \frac{d^{2 k-3}}{dz^{2 k-3}} \left (1+z^2+z^4+\cdots z^{2 k-2} \right )^2 \left (2 z+4 z^3+\cdots+(2 k-2) z^{2 k-3} \right )$$

This is obviously a polynomial, and at $z=0$, the only surviving term will be the $z^{2 k-3}$ term. The coefficient of that term is

$$2+4+6+\cdots+(2 k-2) = k (k-1)$$

Therefore, the integral is

$$\frac{2 \pi}{(2 k-2)!} k (k-1) (2 k-3)! = \pi k$$

ADDENDUM

In response to @Igor's challenge, I can at least show that the value of the integral for half-integer values of $k$ is a rational number. (Note that the above analysis is only valid for integer $k$.) For these values of $k$, the above contour integration does not work because the extension of the integration interval to $[0,2 \pi]$ does not apply. Thus, the recurrence approach of @Nico is better here. If we define

$$J_m = I_{m+\frac12}$$

Then the recurrence relation we seek is

$$J_{m+1}-2 J_m + J_{m-1} = 4 \frac{(-1)^m}{2 m+1}$$

Thus, to show that $J_m \in \mathbb{Q}$, we need only show that $J_0 \in \mathbb{Q}$ and $J_1 \in \mathbb{Q}$. So we evaluate the integrals.

$$\begin{align}J_0 &= 2 \int_0^{\pi/2} du \frac{\sin^2{(u/2)}}{\sin^2{u}} \\ &= \frac12 \int_0^{\pi/2} du \, \sec^2{\frac{u}{2}} \\ &= \left [ \tan{\frac{u}{2}}\right ]_0^{\pi/2} = 1 \end{align}$$

$$\begin{align}J_1 &= 2 \int_0^{\pi/2} du \frac{\sin^2{(3 u/2)}}{\sin^2{u}} \\ &= \frac12 \int_0^{\pi/2} du \, \sec^2{\frac{u}{2}} (1+2 \cos{u})^2 \\ &= \left [ \tan{\frac{u}{2}} + 4 \sin{u}\right ]_0^{\pi/2} = 5 \end{align}$$

Thus, the $J_m \in \mathbb{Q}$ as asserted.

Ron Gordon
  • 138,521
  • It seems a little too complicated for me, but thank you very much for your help. – Dal Oct 09 '14 at 16:47
  • Sorry to hear that. If you get to complex analysis, you will see that your integral is a natural for the methods taught there. – Ron Gordon Oct 10 '14 at 13:10
  • I see. Thank you very much again. I appreciate your skills, your great work and kindness very much :) – Dal Oct 10 '14 at 13:14
  • By the way, you have been mentioned in an answer to this question of mine and I directly saw your skills. So I would like to ask you if you mind answering or commenting on that question and sharing some of your experiences. – Dal Oct 10 '14 at 13:19
  • @Dal: Oh my! I'll have a hard time getting my head through the door today. – Ron Gordon Oct 10 '14 at 13:29
  • Extraordinary work as always. –  Oct 11 '14 at 08:40
  • 2
    My guess is:

    $$ J_k= \frac{ -2 (-1)^k + (1 + 2k)(\pi + 2(-1)^k \Phi_k)}{2}, $$ where $\Phi_k = \Phi \Big(-1, 1, \frac{3+2k}{2}\Big)$ [see 1].

    –  Oct 11 '14 at 13:37
  • @Nico: that's pretty much what Mathematica produces. The LerchPhi is just a fancy way of saying the sum on my RHS. I doubt you're going to get any simpler than that. That's why I stopped at the proof of rationality. – Ron Gordon Oct 11 '14 at 13:39
  • You are perfectly right. –  Oct 11 '14 at 14:21
1

Mathematica says:

$$ -\frac{1}{2} i e^{-i \pi k} \left(2 \pi e^{i \pi k} k \cot (\pi k)+k \psi ^{(0)}\left(\frac{1}{2}-\frac{k}{2}\right)-k \psi ^{(0)}\left(-\frac{k}{2}\right)+e^{2 i \pi k} \left(k \psi ^{(0)}\left(\frac{k}{2}\right)-k \psi ^{(0)}\left(\frac{k+1}{2}\right)+1\right)+1\right) $$

Igor Rivin
  • 25,994
  • 1
  • 19
  • 40
  • I have not tried it with WolframALpha – Igor Rivin Oct 09 '14 at 16:57
  • 2
    Assuming[Element[k, Integers], Integrate[(2 + 2 Cos[x] - Cos[(k - 1) x] - 2 Cos[k x] - Cos[(k + 1) x])/(1 - Cos[2 x]), {x, 0, Pi}]] works for Mathematica – Igor Rivin Oct 09 '14 at 17:04
  • No offense @IgorRivin, but this is a clear example of why it may not be a good idea to just leave evaluation of tough integrals and series to Mathematica/Maple. – Ron Gordon Oct 10 '14 at 13:16
  • @RonGordon If you don't mean offense, your statement is completely gratuitous. – Igor Rivin Oct 10 '14 at 13:36
  • @IgorRivin: how so? You have posted output from Mathematica which suggests a seriously complicated result, when in fact the result is as simple as can be. Perhaps I missed your point. – Ron Gordon Oct 10 '14 at 13:48
  • 1
    @RonGordon I posted output from Mathematica, because that is the answer, for any $k$. Nico's solution is very nice, and immediately shows that this is a made-up question, with a made-up trick to do it. So, not very interesting. In fact, there is a lot more going on (see, for example, my comment on the accepted answer).

    On a different level, I saw the question, I had Mathematica open, I knew that I could potentially help the OP by using 30 seconds of my time. What exactly is the problem with that? Should I know a priori that this is a context problem?

    – Igor Rivin Oct 10 '14 at 13:56
  • @IgorRivin: perhaps it is a more general result. But you yourself never said that. In fact, you stated the opposite when explaining your input into Mathematica: your input states explicitly to assume that $k$ is an integer. Someone like Dal who is looking for guidance, if this were the only answer, may get discouraged or confused by such a complicated result when in fact the result is simple. That's all I am saying. – Ron Gordon Oct 10 '14 at 14:00
  • @RonGordon You are correct that the integer assumption is not necessary, and as for "confused or discouraged", I maintain that having a slick trick like the one in the accepted answer teaches the OP absolutely nothing. If s/he were to get the simple answer out of the mathematica formula (not so trivial), s/he might actually learn a lot. – Igor Rivin Oct 10 '14 at 14:11
  • Please, there is no need to argue. I just want to say that I appreciate all three answers because they give me different viewpoints and approaches and therefore all will (hopefully) help me get better. – Dal Oct 10 '14 at 14:14
  • @Dal: no need to worry. I respect Igor and his terrific contributions here. I was critical of this particular answer, and he disagrees. I stated my peace as has he. Typical sort of exchange here. – Ron Gordon Oct 10 '14 at 14:20
  • Indeed, the OP has a healthy approach - every different approach contributes something. – Igor Rivin Oct 10 '14 at 14:23