Recently, I ran across a product that seems interesting.

Does anyone know how to get to the closed form:


I tried using the identity $\cos(x)=\frac{\sin(2x)}{2\sin(x)}$ in order to make it "telescope" in some fashion, but to no avail. But, then again, I may very well have overlooked something.

This gives the correct solution if $n$ is odd, but of course evaluates to $0$ if $n$ is even.

So, I tried taking that into account, but must have approached it wrong.

How can this be shown? Thanks everyone.

  • 1
    $\begingroup$ It's finite, isn't it? $\endgroup$ – draks ... Jul 20 '12 at 12:20
  • $\begingroup$ You should have $sin(n\pi/2)$. $\endgroup$ – Mhenni Benghorbal Jul 20 '12 at 12:35
  • 1
    $\begingroup$ In the argument of the $\sin$, instead of $k$, it should be $n$. $\endgroup$ – Mhenni Benghorbal Jul 20 '12 at 12:38
  • $\begingroup$ Yes, I know. It is a careless typo. Sorry. $\endgroup$ – Cody Jul 20 '12 at 13:52
  • 2
    $\begingroup$ This is almost identical to the second part of this question $\endgroup$ – robjohn Jul 20 '12 at 18:03

The roots of the polynomial $X^{2n}-1$ are $\omega_j:=\exp\left(\mathrm i\frac{2j\pi}{2n}\right)$, $0\leq j\leq 2n-1$. We can write \begin{align} X^{2n}-1&=(X^2-1)\prod_{j=1}^{n-1}\left(X-\exp\left(\mathrm i\frac{2j\pi}{2n}\right)\right)\left(X-\exp\left(-\mathrm i\frac{2j\pi}{2n}\right)\right)\\ &=(X^2-1)\prod_{j=1}^{n-1}\left(X^2-2\cos\left(\frac{j\pi}n\right)X+1\right). \end{align} Evaluating this at $X=i$, we get $$(-1)^n-1=(-2)(-2\mathrm i)^{n-1}\prod_{j=1}^{n-1}\cos\left(\frac{j\pi}n\right),$$ hence \begin{align} \prod_{j=1}^n\cos\left(\frac{j\pi}n\right)&=-\prod_{j=1}^{n-1}\cos\left(\frac{j\pi}n\right)\\ &=\frac{(-1)^n-1}{2(-2\mathrm i)^{n-1}}\\ &=\frac{(-1)^n-1}2\cdot \frac{\mathrm i^{n-1}}{2^{n-1}}. \end{align} The RHS is $0$ if $n$ is even, and $-\dfrac{(-1)^m}{2^{2m}}=-\dfrac{\sin(n\pi/2)}{2^{n-1}}$ if $n$ is odd with $n=2m+1$.

  • $\begingroup$ Thanks Davide. Nice thoughts. $\endgroup$ – Cody Jul 20 '12 at 13:58
  • $\begingroup$ Also, thanks to joriki and Thomas. I appreciate everyones input. $\endgroup$ – Cody Jul 20 '12 at 14:06
  • 3
    $\begingroup$ Corrected some misprints, missing occurrences of $\mathrm i$, and wrong indexations, and added a one-sentence at the end. Hope the result suits you. $\endgroup$ – Did Jul 20 '12 at 14:24
  • $\begingroup$ I knew this all seemed too familiar. I think the second part of this answer is in the same vein as yours. $\endgroup$ – robjohn Jul 20 '12 at 18:06

The idea of using $\cos x=\frac{1}{2}\frac{\sin 2x}{\sin x}$ is a nice one, and works quickly. We should not use this when $x=\frac{n\pi}{n}$,because of the $\frac{0}{0}$ issue. Also, as you observe, the product is $0$ if $n$ is even, so we needn't bother. So let $n$ be odd.

Look at the product from $k=1$ to $k=n-1$. As $k$ ranges over these values, the numbers $2k$ range, modulo $n$, over all numbers from $1$ to $n-1$. So the $\cos(2k\pi/n)$ (apart from sign) range in some order over the $\cos(\pi j/n)$. Thus, apart from sign, there is cancellation and the product has absolute value $\frac{1}{2^{n-1}}$.

There is no sign issue. If $n\equiv 1\pmod {4}$, the product from $1$ to $n-1$ has an even number of negative terms. The term $\cos(\pi n/n)$ then gives us an extra $-1$, and the product is negative. For the same reason, if $n\equiv 3\pmod{4}$ then the product is positive. The $-\sin(n\pi/2)$ term captures these sign facts, and also produces the right answer of $0$ when $n$ is even.

  • $\begingroup$ Thanks Andre. I was on the right path afterall. I noticed the 0/0 issue you mention but failed to see the forest for the trees. Thanks. I managed to see the $2^{n-1}$ from this method, but did not see how the $\sin(\pi n/2)$ was revealed. $\endgroup$ – Cody Jul 20 '12 at 13:55
  • $\begingroup$ @Cody: Corrected a few typos. One would think such a short argument would be typo-free! As you can see, there is basically term by term cancellation if one looks at things right. $\endgroup$ – André Nicolas Jul 20 '12 at 14:01
  • $\begingroup$ @AndréNicolas: now that I read your answer, I see that mine is just a paraphrase into $\prod$s. (+1) $\endgroup$ – robjohn Jul 20 '12 at 16:55

If $n$ is even, then the term with $k=n/2$ makes the product on the left $0$ and $\sin\left(\frac{n}{2}\pi\right)=0$. So assume that $n$ is odd. $$ \begin{align} \prod_{k=1}^n\cos\left(\frac{k\pi}{n}\right) &=-\prod_{k=1}^{n-1}\cos\left(\frac{k\pi}{n}\right)\tag{1}\\ &=-\prod_{k=1}^{n-1}\frac{\sin\left(\frac{2k\pi}{n}\right)}{2\sin\left(\frac{k\pi}{n}\right)}\tag{2}\\ &=\frac{-1}{2^{n-1}}\frac{\prod\limits_{k=\frac{n+1}{2}}^{n-1}\sin\left(\frac{2k\pi}{n}\right)}{\prod\limits_{k=1}^{\frac{n-1}{2}}\sin\left(\frac{(2k-1)\pi}{n}\right)}\tag{3}\\ &=\frac{(-1)^{\frac{n+1}{2}}}{2^{n-1}}\frac{\prod\limits_{k=1}^{\frac{n-1}{2}}\sin\left(\frac{(2k-1)\pi}{n}\right)}{\prod\limits_{k=1}^{\frac{n-1}{2}}\sin\left(\frac{(2k-1)\pi}{n}\right)}\tag{4}\\ &=-\frac{\sin\left(n\frac\pi2\right)}{2^{n-1}}\tag{5} \end{align} $$ $(1)$: $\cos(\pi)=-1$

$(2)$: $\sin(2x)=2\sin(x)\cos(x)$

$(3)$: cancel $\sin\left(\frac{j\pi}{n}\right)$ in the numerator and denominator for even $j$ from $2$ to $n-1$

$(4)$: in the numerator, change variable $k\mapsto k+\frac{n-1}{2}$ and use $\sin(x+\pi)=-\sin(x)$

$(5)$: for odd $n$, $\sin\left(n\frac\pi2\right)=(-1)^{\frac{n-1}{2}}$

  • $\begingroup$ Thanks RobJohn. That is how I started it, but was too dumb to finish :(. Thanks a bunch. $\endgroup$ – Cody Jul 20 '12 at 17:19
  • $\begingroup$ @Cody: It is not easy to see all the cancellations. I had to picture where the points on the circle were and note where the terms from the numerator and denominator overlapped to get step $(3)$. Good question! (+1) $\endgroup$ – robjohn Jul 20 '12 at 17:55
  • $\begingroup$ Thanks RobJohn. I noticed in the link you provided this has been asked before. I always look before I post to see if the same question as been asked. But, I did not find this one. Often, it is difficult to locate something in particular. I thought this was a cool problem. Thanks for all the input and different methods. $\endgroup$ – Cody Jul 21 '12 at 10:31

The monic Chebyshev polynomial of the second kind,

$$\hat{U}_n(x)=\frac{\sin((n+1)\arccos\,x)}{2^n \sqrt{1-x^2}}$$

can be easily seen to have the roots $x_k=\cos\dfrac{\pi k}{n+1}$ for $k=1,\dots,n$. By Vieta, the constant term of $\hat{U}_n(x)$ is equal to

$$\hat{U}_n(0)=\prod_{k=1}^n \cos\dfrac{\pi k}{n+1}$$

and thus

$$\prod_{k=1}^n \cos\dfrac{\pi k}{n}=-\hat{U}_{n-1}(0)=-\frac{\sin\frac{n\pi}{2}}{2^{n-1}}$$

  • $\begingroup$ Thanks JM. I certainly would have not thought of this. I have heard of Chebyshev polynomials, but that's about it. I will have to look into them. $\endgroup$ – Cody Jul 20 '12 at 14:15
  • $\begingroup$ Big guns work fast :-) (+1) $\endgroup$ – robjohn Jul 20 '12 at 16:52
  • $\begingroup$ We both have seen this before. $\endgroup$ – robjohn Jul 20 '12 at 20:34
  • $\begingroup$ @rob, oh dear, precisely the same argument, except I hadn't mentioned Vieta in the other answer... :) $\endgroup$ – J. M. is a poor mathematician Jul 20 '12 at 22:03

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.