25
$\begingroup$

A disk of area $2n$ is divided into $n$ regions by drawing $n$ evenly spaced points on the perimeter and then drawing line segments joining one fixed point with all the other points. An example is shown with $n=8$.

Circle

What is the limit of the product of the areas of the regions, as $n$ approaches $\infty$ ?

Using basic trigonometry, I've got:

$$\lim_{n\to\infty}\prod_{k=1}^{n}\left(2-\frac{n}{\pi}\sin\left(\frac{2k\pi}{n}\right)+\frac{n}{\pi}\sin\left(\frac{2(k-1)\pi}{n}\right)\right)$$

I do not know how to evaluate this limit. Wolfram does not evaluate the limit, but tells me that when $n=10000$ the product is approximately $8.3$.

UPDATE1:

I am fairly confident that the limit is $4\cosh^2\left({\frac{\pi}{2\sqrt{3}}}\right)=8.29674...$

Here's why. I was trying to answer a similar question: A ball is divided into $n$ concentric shells of equal thickness. Can the average volume of the shells be fixed so that the product of the volumes converges to a positive number as $n\to\infty$? The answer turns out to be yes. If we fix the average volume of the shells to be $\frac{e^2}{3}$ then the product of the volumes converges to $2\cosh\left({\frac{\pi}{2\sqrt{3}}}\right)=2.8804...$ I noticed that this number seemed to be (to many decimal places) the square root of the answer to the question here. I think there must be a connection.

UPDATE2:

Another similar question is this. A disk is divided into $n$ regions by equally spaced parallel lines, with the perimeter of the disk being tangent to two of the lines. Can the average area of the regions be fixed so that the product of the areas converges to a positive number as $n\to\infty$? The answer seems to be yes. If we fix the average area of the regions to be $\frac{{\pi}e}{8}$ then the product of the areas seems to converge to $2\cos\left(\frac{\pi}{2\sqrt{3}}\right)$ (notice this is cos, not cosh). Again, I think there must be a connection.

(As for the source of the original question, I thought of the question by myself. It was inspired by an IB (high school) May 2021 exam question, which goes like this. A unit circle has $n$ evenly distributed points. Line segments are drawn joining one point with all the other points (like a seashell). The exam question led students to find the product of the lengths of the line segments, using complex numbers. The answer turns out to be $n$. Then I wondered, what is the product of the areas of the enclosed regions? Obviously the product approaches $0$ as $n\to\infty$ (since all the areas approach $0$). But could the average area be fixed so that the product of the areas converges to a positive number as $n\to\infty$? Definitely beyond the course syllabus. After experimenting on desmos, I found that the answer seems to be yes: if we fix the average area to be $2$, then the product of the areas seems to converge to a positive number. The question in this post is, what does it converge to?)

$\endgroup$
10
  • 1
    $\begingroup$ Have you evaluated the partial sum for various values of $n$? Just knowing that it's $\approx 8.3$ when $n=10000$ doesn't suggest it converges unless it's very stable at that and other magnitudes. $\endgroup$ Commented Dec 19, 2021 at 12:08
  • $\begingroup$ @JackNeubecker Good point. When $n=10, 100, 1000, 10000$, the products are approximately $7.08, 8.16, 8.28, 8.30,$ respectively. $\endgroup$
    – Dan
    Commented Dec 19, 2021 at 12:15
  • 1
    $\begingroup$ Please provide the source of this problem, which may give some hint as to the level of sophistication of the intended solution. (Add all context to the body of the question. Comments are easily overlooked and may in fact be hidden.) $\endgroup$
    – Blue
    Commented Dec 25, 2021 at 17:14
  • 1
    $\begingroup$ @Blue OK, done. Thanks for the suggestion. $\endgroup$
    – Dan
    Commented Dec 26, 2021 at 2:00
  • 1
    $\begingroup$ @runway44 See the part of my post labelled "EDIT1". I found the answer to the question about the ball using wolfram; wolfram directly gave the expression $2\cosh\left({\frac{\pi}{2\sqrt{3}}}\right)$. Then I noticed that this number seemed to be (to many decimal places) the square root of the numerical estimates of the answer to my original question. So I guessed that the answer to my original question is $4\cosh^2\left({\frac{\pi}{2\sqrt{3}}}\right)$. $\endgroup$
    – Dan
    Commented Feb 22, 2022 at 3:38

2 Answers 2

20
$\begingroup$

The limit is, as suspected, $$4\cosh^2\left(\frac{\pi}{2\sqrt 3}\right).$$ First, define $\epsilon_n=1-\frac n\pi\sin(\pi/n)$, so that $$\epsilon_n=\frac{\pi^2}{6n^2}-O(n^{-3})$$ and \begin{align*} a_n &:=\sum_{k=1}^n\log\left(2-\frac n\pi\sin\left(\frac{2k\pi}n\right)+\frac n\pi\sin\left(\frac{2(k-1)\pi}n\right)\right)\\ &=\sum_{k=1}^n\log\left(2-\frac n\pi\left[\sin\left(\frac{2k\pi}n\right)-\sin\left(\frac{2(k-1)\pi}n\right)\right]\right)\\ &=\sum_{k=1}^n\log\left(2-2\frac n\pi\sin\left(\frac\pi n\right)\cos\left(\frac{(2k-1)\pi}n\right)\right)\\ &=\sum_{k=1}^n\log\left(2-2\cos\left(\frac{(2k-1)\pi}n\right)+2\cos\left(\frac{(2k-1)\pi}n\right)\epsilon_n\right). \end{align*} Define $$b_n:=\sum_{k=1}^n\log\left(2-2\cos\left(\frac{(2k-1)\pi}n\right)\right),$$ so that $$c_n:=a_n-b_n=\sum_{k=1}^n\log\left(1+\frac{\epsilon_n\cos\left(\frac{(2k-1)\pi}n\right)}{1-\cos\left(\frac{(2k-1)\pi}n\right)}\right).$$ The summand here is symmetric about $k=\frac{n+1}2$, and is $\log(1-\epsilon_n/2)=o(1)$ at $k=\frac{n+1}2$, so $$c_n=o(1)+2\sum_{k=1}^{\lfloor n/2\rfloor}\log\left(1+\frac{\epsilon_n\cos\left(\frac{(2k-1)\pi}n\right)}{1-\cos\left(\frac{(2k-1)\pi}n\right)}\right).$$ Write $$c_{n,K}=\sum_{k=1}^K\log\left(1+\frac{\epsilon_n\cos\left(\frac{(2k-1)\pi}n\right)}{1-\cos\left(\frac{(2k-1)\pi}n\right)}\right).$$ We have, for large $n$ (being liberal with constant factors), that \begin{align*} |c_n-2c_{n,K}| &\leq 2\sum_{k=K+1}^{\lfloor n/2\rfloor}\left|\log\left(1+\frac{\epsilon_n\cos\left(\frac{(2k-1)\pi}n\right)}{1-\cos\left(\frac{(2k-1)\pi}n\right)}\right)\right|\\ &\leq 4\sum_{k=K+1}^{\lfloor n/2\rfloor}\frac{\epsilon_n}{1-\cos\left(\frac{(2k-1)\pi}n\right)}\\ &=2\epsilon_n\sum_{k=K+1}^{\lfloor n/2\rfloor}\frac1{\sin^2\left(\frac{(2k-1)\pi}{2n}\right)}\\ &\leq 4\epsilon_n\sum_{k=K+1}^{\lfloor n/2\rfloor}\frac1{\left(\frac{(2k-1)\pi}{2n}\right)^2}\\ &=\frac{16\epsilon_nn^2}{\pi^2}\sum_{k=K+1}^{\lfloor n/2\rfloor}\frac1{(2k-1)^2}\\ &\leq\frac 83\sum_{k=K+1}^\infty \frac1{(2k-1)^2}\leq \frac2{3K}. \end{align*} On the other hand, for $n$ large and $k<n^{1/4}$, \begin{align*} \log\left(1+\frac{\epsilon_n\cos\left(\frac{(2k-1)\pi}n\right)}{1-\cos\left(\frac{(2k-1)\pi}n\right)}\right) &=\log\left(1+\frac{\left(\frac{\pi^2}{6n^2}+O(n^{-3})\right)\left(1-O(n^{-1})\right)}{2\sin^2\left(\frac{(2k-1)\pi}{2n}\right)}\right)\\ &=\log\left(1+\frac{\left(\frac{\pi^2}{6n^2}+O(n^{-3})\right)\left(1-O(n^{-3/2})\right)}{\frac{(2k-1)^2\pi^2}{2n^2}+O(n^{-3})}\right)\\ &=\log\left(1+\frac1{3(2k-1)^2}+O(n^{-1})\right)\\ &=\log\left(1+\frac1{3(2k-1)^2}\right)+O\left(\frac 1n\right). \end{align*} This means that, setting $K=n^{1/4}$, $$c_n=2c_{n,K}+o(1)=o(1)+2\sum_{k=1}^K\log\left(1+\frac1{3(2k-1)^2}\right),$$ and thus $$\lim_{n\to\infty} c_n=2\sum_{k=1}^\infty \log\left(1+\frac1{3(2k-1)^2}\right).$$ Now, \begin{align*} e^{b_n} &=\prod_{k=1}^n\left(2-2\cos\left(\frac{(2k-1)\pi}n\right)\right)\\ &=\prod_{k=1}^n\left|1-e^{\frac{(2k-1)\pi i}n}\right|^2\\ &=\prod_{z^n=-1}\left|1-z\right|^2\\ &=(1-(-1))^2=4. \end{align*} This shows that $$\lim_{n\to\infty} a_n=2\ln 2+\lim_{n\to\infty}c_n=2\ln 2+2\sum_{k=1}^\infty \log\left(1+\frac1{3(2k-1)^2}\right).$$ So, we need only evaluate this infinite series, which we call $S$. We first expand the $\log$ to get $$S=\sum_{k=1}^\infty\sum_{m=1}^\infty \frac{(-1)^{m-1}}{m3^m(2k-1)^{2m}}=\sum_{m=1}^\infty \frac{(-1)^{m-1}}{m3^m}\sum_{k=1}^\infty \frac1{(2k-1)^{2m}},$$ where the interchange of summation is justified since the sum is absolutely convergent. Using that $$\sum_{k=1}^\infty \frac1{(2k-1)^{2m}}=\left(1-2^{-2m}\right)\zeta(2m),$$ we get $$S=\sum_{m=1}^\infty \frac{(-1)^{m-1}}m\left(3^{-m}-12^{-m}\right)\zeta(2m).$$ Write $$f(x)=\sum_{m=1}^\infty\frac{(-1)^{m-1}x^m\zeta(2m)}m.$$ We compute $$f'(x)=\sum_{m=1}^\infty(-x)^{m-1}\zeta(2m).$$ An identity here gives that $$\sum_{m=1}^\infty t^{2m}\zeta(2m)=\frac{1-\pi t\cot(\pi t)}{2},$$ and so $$f'(x)=-\frac{1-\pi i\sqrt{x}\cot(\pi i\sqrt x)}{2x}=\frac{\pi\sqrt x\coth(\pi\sqrt x)-1}{2x}.$$ This gives \begin{align*} S &=f\left(\frac13\right)-f\left(\frac1{12}\right)\\ &=\int_{1/12}^{1/3}\frac{\pi\sqrt x\coth(\pi\sqrt x)-1}{2x}dx\\ &=\int_{\frac1{2\sqrt3}}^{\frac1{\sqrt3}}\frac{\pi y\coth(\pi y)-1}ydy\\ &=\int_{\frac1{2\sqrt3}}^{\frac1{\sqrt3}}\pi\coth(\pi y)dy-\log 2\\ &=\log(\sinh(\pi y))\bigg|_{\frac1{2\sqrt3}}^{\frac1{\sqrt3}}-\log 2\\ &=\log\left(\frac{\sinh\left(\frac{\pi}{\sqrt 3}\right)}{2\sinh\left(\frac{\pi}{2\sqrt 3}\right)}\right)\\ &=\log\cosh\frac{\pi}{2\sqrt 3}. \end{align*} So, the limit is $$e^{2\ln 2+2S}=4\cosh^2\left(\frac{\pi}{2\sqrt 3}\right),$$ as desired.

$\endgroup$
8
  • 6
    $\begingroup$ Of course, why didn't I think of that ; ) $\endgroup$
    – Dan
    Commented Dec 26, 2021 at 7:17
  • $\begingroup$ In your first and last equations, I think you have an extra "2". $\endgroup$
    – Dan
    Commented Dec 26, 2021 at 7:17
  • 2
    $\begingroup$ @FShrike The goal is to be able to upper bound the sum of $\log$ terms, and the $x$ terms are easier to sum than anything similar. The precise claim is that, if $0<x<1/2$, then $|\log(1+x\cos y)|\leq 2x$. If $\cos y>0$, then you can more simply say $$|\log(1+x\cos y)|=\log(1+x\cos y)\leq x\cos y\leq x,$$ but the bound for when $x\cos y<0$ goes the other way. The factor of $2$ is just a matter of convenience, and can be replaced by $1+\delta$ for any $\delta>0$, provided that $x$ is small enough. $\endgroup$ Commented Dec 27, 2021 at 1:29
  • 2
    $\begingroup$ @Dan It was very useful to have an explicit answer to shoot for when checking my work / finding mistakes (when there are a lot of calculations, as there are here, some of them are bound to come out wrong the first time around). If I remember correctly, the expected form was also helpful in evaluating the final sum: once I saw $\coth$ popping up, I knew I was on the right track. $\endgroup$ Commented Jan 1, 2022 at 11:18
  • 3
    $\begingroup$ As for any applications of technology: Wikipedia was the source of the zeta-series identity (I'd remembered it vaguely from before, and probably could've rederived it if necessary, but I'm pretty sure that's where I saw it in the first place), and WolframAlpha and Desmos were very useful for verifying routine calculations and checking bounds on things. I also used Python for some of the computation-checking where I compared intermediate steps with the suspected answer. $\endgroup$ Commented Jan 1, 2022 at 11:20
4
$\begingroup$

I'll answer my own question.

Using $\sin{A}-\sin{B}=2\left(\sin{\frac{A-B}{2}}\right)\left(\cos{\frac{A+B}{2}}\right)$,

$$L=\lim_{n\to\infty}\prod_{k=1}^{n}\left(2-\frac{n}{\pi}\sin\frac{2k\pi}{n}+\frac{n}{\pi}\sin\frac{2(k-1)\pi}{n}\right)$$

$$=\lim_{n\to\infty}\prod_{k=1}^{n}\left(2-\left(\frac{2n}{\pi}\sin{\frac{\pi}{n}}\right)\cos{\frac{(2k-1)\pi}{n}}\right)$$

Using the Maclaurin series for sine,

$$L=\lim_{n\to\infty}\prod_{k=1}^{n}\left(2-\left(2-\left(\frac{\frac{\pi}{\sqrt3}}{n}\right)^2+O\left(\frac{1}{n^4}\right)\right)\cos{\frac{(2k-1)\pi}{n}}\right)$$

Since we are taking the limit, I think we can ignore the $O\left(\dfrac{1}{n^4}\right)$, but I'm not sure how to to justify this; feel free to comment or edit.

Then this answer to one of my other questions shows that
$$\lim_{n\to\infty}\prod_{k=1}^{n}\left(2-\left(2-\left(\frac{x}{n}\right)^2\right)\cos{\frac{(2k-1)\pi}{n}}\right)=4\cosh^2{\left(\frac{x}{2}\right)}$$

$$\therefore L=4\cosh^2{\left(\frac{\pi}{2\sqrt{3}}\right)}$$

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .