Sign up ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Compute the following limit: $$\lim_{n \to \infty}\int_{0}^{2\pi}\cos\left(x\right)\cos\left(2x\right)\ldots \cos\left(nx\right)\,{\rm d}x$$

Today I was working on a W. L. Putnam competition's problem containing this integral and wondered how I may compute its value when $n$ goes to $\infty$. So far I've found no answer.
Could you give me some suggestions about the way I should go?

share|cite|improve this question
It seems to me that the limit is zero. There are infinitely many choices of $n$ that produce exactly zero. Anyway, I believe that using the formulae in you can inductively compute a primitive. This primitive is a sum of sines (which produce zeroes after integration) and a term which is either a sine or a polynomial. But the coefficient of this polynomial becomes smaller and smaller as $n \to +\infty$. – Siminore Jul 13 '12 at 14:31
@Siminore: ma i non lo so parlare italiano. Grazie! – Chris's sis the artist Jul 13 '12 at 14:34
@Siminore, i also think that the limit is $0$ but i don't know how to prove it. Yeah, it could be a way. – Chris's sis the artist Jul 13 '12 at 14:36
This is in English. :) – J. M. Jul 13 '12 at 14:37
Sorry, it seems that the name of Werner is not very popular in the american literature. Anyway, mathematics formulae are written in a universal language ;-) – Siminore Jul 13 '12 at 14:43

6 Answers 6

up vote 18 down vote accepted

Let $u = e^{ix}$. Then $\cos x \cos 2x ... \cos nx = \frac{1}{2^n}\prod_{k=1}^n (u^k + u^{-k})$.

Write $$\prod_{k=1}^n (z^k + z^{-k}) = \sum_{j=-N}^N a_j z^j$$

where $N=\frac{n(n+1)}2$.

Then the integral above is just $\frac{2\pi}{2^n}a_0$.

Combinatorially, we can see that $a_0$ is the number of ways of picking a subset of $\{1,2,...,n\}$ which adds up to $\frac{n(n+1)}{4}$.

Now, the set of subsets of $\{1,...,n\}$ that add up to a particular value is an antichain, so we have a bound on the size of $a_0$, namely, $$0\leq a_0 \leq \binom {n}{\lfloor n/2\rfloor}$$.

And we know that $$\lim_{n\to\infty}\frac{1}{2^n}\binom {n}{\lfloor n/2\rfloor} =0$$

So we know your limit is zero.

(The maximum size of an antichain in the Boolean poset is Sperner's theorem.)

Note that when $n\equiv 1,2\pmod 4$, $\frac{n(n+1)}4$ is not an integer, so in those cases, $a_0=0$. Anybody have an "interesting" lower bound on $a_0$ for the other cases?

[In comments below, Robert Israel shows that when $n\equiv 0,3\pmod 4$, $a_0\geq 2^{\lfloor (n+1)/4\rfloor}$. I suspect that there is a polynomial, $q$, such that $a_0\geq \frac{2^n}{q(n)}$.]

This proof generalizes. Given any sequence of positive integers, $\{b_1,...,b_n\}$, we have the same result:

$$\int_{0}^{2\pi} \cos b_1 x \cos b_2 x ... \cos b_n x dx= \frac{2\pi}{2^n}A$$

where $A$ is the count of an antichain, so $0\leq A\leq \binom {n}{\lfloor n/2\rfloor}$

If $b_i=1$ for all $i$, then we get exact equality when $n$ is even, which is to say that:

$$\int_{0}^{2\pi} \cos^{2m} x dx = \frac{2\pi}{2^{2m}} \binom {2m}{m}$$

share|cite|improve this answer
Ah, I wasn't thinking in terms of posets! – anon Jul 13 '12 at 15:04
I was thinking exactly along these lines except I didn't think that much about the $a_0$ term. My logic was that whatever it would be, it would still be a polynomial in $n$, and $2^{n}$ would always grow faster than a polynomial in $n$ which implies that the limit would zero. Would this have sufficed on the exam? I don't think that they expect the people taking the exam to know these facts about chains and what-not. – Ahsan Jul 13 '12 at 18:28
But I agree, Sperner's theorem was definitely not something I learned until grad school (even though it is a relatively elementary thorem.) – Thomas Andrews Jul 13 '12 at 18:39
Note that for $n \equiv 0 \mod 4$, $a_0 \ge 2^{n/4}$ since in each four-tuple $(4j+1,4j+2,4j+3,4j+4)$ you can choose either $\{4j+1,4j+4\}$ or $\{4j+2,4j+3\}$. So there is certainly no polynomial upper bound. – Robert Israel Jul 13 '12 at 20:03
And similarly for $n \equiv 3 \mod 4$, $a_0 \ge 2^{(n+1)/4}$ since you can take $\{1,2\}$ or $\{3\}$ and then in each $\{4j,\ldots,4j+3\}$ either $\{4j+1,4j+2\}$ or $\{4j,4j+3\}$. – Robert Israel Jul 13 '12 at 20:08

We have

$$\begin{array}{c l}\int_0^{2\pi}\cos x\cdots\cos nx dx= & \int_0^{2\pi}\prod_{k=1}^n\frac{e^{ikx}+e^{-ikx}}{2}dx \\ & =\frac{1}{2^n}\int_0^{2\pi}e^{-n(n+1)/2\,x}\prod_{k=1}^n(1+e^{i2kx})dx \\ & = \frac{1}{2^n}\sum_{K\subseteq [n]}\int_0^{2\pi}\exp\left(-\frac{n(n+1)}{2}i+2i\sum_{k\in K}k\right)dx \\ & =\frac{2\pi|L_n|}{2^n}. \end{array}$$

where $L_n$ is the set of subsets $K$ of $[n]=\{1,\cdots,n\}$ such that $$2\sum_{k\in K}k=\frac{n(n+1)}{2}.$$

(Note this is zero when $n\ne 0,-1$ mod $4$.) Perhaps one can show $|L_n|=\Omega(2^n)$ with a combinatorial argument? Since it's $0$ infinitely often it would also suffice to show the limit exists.

share|cite|improve this answer
Are $L_n$ Lucas' numbers? – Pedro Tamaroff Jul 13 '12 at 15:05
@PeterTamaroff I just felt like using the letter $L$. – anon Jul 13 '12 at 15:06
Ah, OK. Because $F_n<\phi^n$ so you might have gotten somewhere. – Pedro Tamaroff Jul 13 '12 at 15:07

Another analysis proof:

It suffices to show that $$\lim_{n \rightarrow \infty} \int_0^{2\pi}|\cos x \cos 2x\cdots \cos nx|\,dx = 0$$ Define $f_n(x) = |\cos x \cos 2x\cdots \cos nx|$. Then since each $|\cos kx | \leq 1$, $f_n(x) \geq f_{n+1}(x)$ for all $n$ and $x$. Since $|f_n(x)| \leq 1$ for all $n$ and $x$, the Lebesgue dominated convergence theorem says $$\lim_{n \rightarrow \infty} \int_0^{2\pi} |\cos x \cos 2x\cdots \cos nx|\,dx = \int_0^{2\pi} \lim_{n \rightarrow \infty}|\cos x \cos 2x\cdots \cos nx|\,dx$$ So it suffices to show that the limit on the inside is zero for almost every $x$. But whenever $x$ is not a rational multiple of $\pi$, $\{nx: n = 1,2,..\}$ is dense in $[0,2\pi]$. Which means given such $x$, there is a sequence $n_1, n_2,...$ such that $$\lim_{k \rightarrow \infty} n_k x = {\pi \over 2}$$ So $$\lim_{k \rightarrow \infty} \cos(n_k x) = \cos({\pi \over 2}) = 0$$ Since $\lim_{n \rightarrow \infty}|\cos x \cos 2x\cdots \cos nx|$ is at most any given $|\cos (n_k x)|$, we conclude that $\lim_{n \rightarrow \infty}|\cos x \cos 2x\cdots \cos nx| = 0$ for all $x$ not a rational multiple of $\pi$. Since the multiples of $\pi$ are a set of measure zero, we are done.

share|cite|improve this answer
I like this one! – user117932 Jan 27 '14 at 0:55
In my opinion, this method is the best. – recursive recursion Apr 6 '14 at 23:21

You can prove instead that

$$\lim_{n\to\infty} \int_{0}^{2\pi} \left| \cos x \cos 2x\cdots \cos nx \space{dx} \right| = 0 \,.$$

Let $\epsilon >0$. Then for all $n$ we have

$$\int_{0}^{2\pi} \left| \cos x \cos 2x\cdots \cos nx \right| \space{dx} = \int_{0}^{\frac{\epsilon}5} \left| \cos x \cos 2x\cdots \cos nx \right| \space{dx} $$ $$+ \int_{\frac{\epsilon}5}^{\pi-\frac{\epsilon}5} \left| \cos x \cos 2x\cdots \cos nx \right| \space{dx} + \int_{\pi-\frac{\epsilon}5}^{\pi+\frac{\epsilon}5} \left| \cos x \cos 2x\cdots \cos nx \right| \space{dx}$$ $$ + \int_{\pi+\frac{\epsilon}5}^{2\pi-\frac{\epsilon}5} \left| \cos x \cos 2x\cdots \cos nx \right| \space{dx}+\int_{2 \pi-\frac{\epsilon}3}^{2\pi} \left| \cos x \cos 2x\cdots \cos nx \right| \space{dx}$$ $$\leq \frac{4\epsilon}{5}+\int_{\frac{\epsilon}5}^{\pi-\frac{\epsilon}5} \left| \cos x \cos 2x\cdots \cos nx \right| \space{dx}+ \int_{\pi+\frac{\epsilon}5}^{2\pi-\frac{\epsilon}5} \left| \cos x \cos 2x\cdots \cos nx \right| \space{dx}$$

I split it this way to avoid the points $x$ where $\cos(kx) \in \{ \pm 1 \}$.

Now, we claim that for all $x \in [\frac{\epsilon}{5}, \pi-\frac{\epsilon}{5}] \cup [\pi+\frac{\epsilon}{5}, 2\pi-\frac{\epsilon}{5}]$ we have

$$\lim_n \left| \cos x \cos 2x\cdots \right| =0$$

This is easy to prove, by splitting it in two cases: x rational and irrational with respect to $\pi$. In the first case, it follows from the Dirichclet Principle that there exists a $k$ so that for all $n$ one of $|\cos nx| , |\cos (n+1)x|, ..., |\cos (n+k)x|$ is less than $\frac{1}{2}$, while the second case is easy to threat by writing $x= \frac{k}m \pi$ with $gcd(m,k)=1$ and $m\neq 1$.

To complete the proof, all these functions are uniformly bounded by $1$ and they converge pointwise to $0$, thus by Lebesgue dominant convergence theorem we have

$$\int_{\frac{\epsilon}5}^{\pi-\frac{\epsilon}5} \left| \cos x \cos 2x\cdots \cos nx \right| \space{dx}+ \int_{\pi+\frac{\epsilon}5}^{2\pi-\frac{\epsilon}5} \left| \cos x \cos 2x\cdots \cos nx \right| \space{dx} \to 0$$

P.S. This proof is probably an overkill...

share|cite|improve this answer
What did you mean by Dirichlet Principle ? I did not find the reference about this Principle. – user119228 Feb 21 '14 at 12:24

Let $X_j$ be independent coin flips ($0$ or $1$ with equal probabilities) and $S_n = \sum_{j=1}^n j X_j$. From Thomas Andrews's answer, we see that $\int_0^{2\pi} \cos(x) \cos(2x) \ldots \cos(nx)\ dx = 2 \pi P(S_n = n(n+1)/4)$. We have $E[S_n] = n(n+1)/4$ and $\text{Var}(S_n) = \dfrac{1}{4} \sum_{j=1}^n j^2 = \dfrac{n(n+1)(2n+1)}{24}$. Moreover, Lindeberg's version of the Central Limit Theorem (which does not require the random variables to be identically distributed) applies. We conclude that for any $\alpha > 0$, $P(|S_n - n(n+1)/4| < \alpha \sqrt{n(n+1)(2n+1)/24}) \to \Phi(\alpha) - \Phi(-\alpha)$ as $n \to \infty$, where $\Phi$ is the CDF of the standard normal distribution. Since $\Phi(\alpha) - \Phi(-\alpha) \to 0$ as $\alpha \to 0$, we conclude that $P(S_n = n(n+1)/4) \to 0$.

share|cite|improve this answer

Choose $0<\epsilon<\pi$. Set $f_n(x)=|\cos(x)\cos(2x)\cdots\cos(nx)|$. Clearly the sequence of (continuous) functions $(f_n)$ converges pointwise to $0$ on $I_{\epsilon}=[\epsilon,2\pi-\epsilon]$, and $0\leq f_{n+1}\leq f_n\leq 1$. It follows from Dini's lemma that $f_n$ converges uniformly to $0$ on $I_{\epsilon}$, so there exists an integer $N_{\epsilon}$ such that for all $n\geq N_{\epsilon}$, $0\leq f_n\leq\epsilon$. Thus, for all $n\geq N_{\epsilon}$, $$\begin{align*}\left|\int_0^{2\pi}f_n \right|&\leq \left|\int_0^{\epsilon}1 \right|+\left|\int_{\epsilon}^{2\pi-\epsilon}f_n \right|+\left|\int_{2\pi-\epsilon}^{2\pi}1 \right| \\ &\leq (2+2\pi)\epsilon\end{align*}$$ Hence $$\int_0^{2\pi}\cos(x)\cos(2x)\cdots\cos(nx)\;dx\xrightarrow{n\to\infty}0\;.$$

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.