3

Suppose that $n$ is a positive integer, and let $k$ be a positive integer less than $n$.

How can one show that $\displaystyle\int_0^1 \frac{x^{k-1} - x^{n-k-1}}{1 - x^n} \; dx= \frac{\pi}{n} \cot{\frac{\pi k}{n}}$ without using the digamma (psi) function or complex analysis? I am pretty sure that I have seen such an approach (using basic calculus) at one point, but I can not quite recall how the argument worked. Thanks!

BDS
  • 98

1 Answers1

2

Begin by Taylor expanding the denominator:

$$\int_0^1 dx \; \frac{x^{k-1} - x^{n-k-1}}{1 - x^n} = \int_0^1 dx \; (x^{k-1} - x^{n-k-1}) \sum_{m=0}^{\infty} x^{m n} $$

Reverse the order of the sum and the integral:

$$ \begin{align} &= \sum_{m=0}^{\infty} \int_0^1 dx \; (x^{m n + k-1} - x^{(m+1) n-k-1}) \\ &= \sum_{m=0}^{\infty} \left ( \frac{1}{m n + k} - \frac{1}{(m+1) n - k} \right ) \\ &= \frac{1}{n} \sum_{m=0}^{\infty} \left ( \frac{1}{m + k/n} - \frac{1}{m+1 - k/n} \right ) \\&= \frac{1}{n} \sum_{m=-\infty}^{\infty} \frac{1}{m + k/n} \\ \end{align}$$

Write $k/n = z/\pi$ and rewrite the sum below:

$$ \begin{align} &= \frac{\pi}{n} \left ( \frac{1}{z} + 2 z \sum_{m=1}^{\infty} \frac{1}{z^2-\pi^2 m^2} \right ) \\ \end{align} $$

The expression in parenthesis is a well-known representation of $\cot(z)$ . This result was derived by Euler, and may be derived by considering Euler's product representation of $\sin{\pi z}/(\pi z)$ and differentiating $\log{\sin{\pi z}}$.

Using the definition of $z$, we find that

$$\int_0^1 dx \; \frac{x^{k-1} - x^{n-k-1}}{1 - x^n} = \frac{\pi}{n} \cot{\frac{\pi k}{n}} $$

EDIT

Simple proof that

$$\cot{z} = \frac{1}{z} + 2 z \sum_{m=1}^{\infty} \frac{1}{z^2-\pi^2 m^2} $$

Consider the relation

$$\cot{y} = \frac{1}{2} (\cot{\frac{y}{2}} - \tan{\frac{y}{2}}) $$

We may rewrite this as

$$\pi x \cot{\pi x} = \frac{\pi x}{2} \left (\cot{\frac{\pi x}{2}} + \cot{\frac{\pi (x \pm 1)}{2}} \right ) $$

Note that this forms a sort of recurrence. That is, we can apply the above relation to each term inside the parenthesis. We use the minus sign for the first term, and the plus sign for the second term. The result of a first pass is

$$\pi x \cot{\pi x} = \frac{\pi x}{2^2} \left (\cot{\frac{\pi x}{2^2}} + \cot{\frac{\pi (x - 1)}{2^2}} + \cot{\frac{\pi (x + 1)}{2^2}} + \cot{\frac{\pi (x + 2)}{2^2}} \right ) $$

We can repeat this as much as we like; after $r$ rounds of this, we get

$$\pi x \cot{\pi x} = \frac{\pi x}{2^r} \left (\cot{\frac{\pi x}{2^r}} + \sum_{m=1}^{2^{r-1}-1} \left [\cot{\frac{\pi (x - m)}{2^r}} + \cot{\frac{\pi (x + m)}{2^r}} \right ] + \cot{\frac{\pi (x + 2^{r-1})}{2^r}} \right ) $$

Now take the limit as $r \rightarrow \infty$. Use the facts that

$$\lim_{r \rightarrow \infty} \frac{\pi x}{2^r} \cot{\frac{\pi x}{2^r}} = 1 $$

$$\lim_{r \rightarrow \infty} \frac{1}{2^r} \cot{\frac{a}{2^r}} = \frac{1}{a} $$

$$\lim_{r \rightarrow \infty} \frac{1}{2^r} \cot{\frac{\pi(x+2^{r-1})}{2^r}} = 0$$

$\forall \, a \ne 0$. and we get the expansion

$$\pi x \cot{\pi x} = 1 + x \sum_{m=1}^{\infty} \left ( \frac{1}{x+m} + \frac{1}{x-m} \right ) $$

from which the cited result follows.

Ron Gordon
  • 138,521
  • It seems complex analysis creeps in when one invokes some well-known representation of cot. – Did Jan 25 '13 at 07:23
  • @Did: No, not necessarily. A very simple, non-complex derivation of this result is given in Knopp, Theory and Application of Infinite Series, Dover, Sec. 24, p.205. Knopp credits the derivation to Euler. – Ron Gordon Jan 25 '13 at 07:35
  • Then you might want to include this representation (or at least some hints about it) in your answer. – Did Jan 25 '13 at 07:45
  • Nice solution! I have seen a variation of this derivation (though the rewriting of the doubly infinite sum is clever) elsewhere. Now, is there a solution which does not use any series result? – BDS Jan 25 '13 at 07:48
  • @BDS: Thanks. Not anything that comes immediately to mind. You have something against series? :) – Ron Gordon Jan 25 '13 at 07:51
  • Absolutely not! I love series very much. I am simply trying to find an 'elementary' as possible proof of the quoted integral fact. I had such a solution bookmarked a while ago before my original laptop crashed! – BDS Jan 25 '13 at 08:00
  • So, you are appealing to Euler's product representation of the function sine. I know two approaches to this result, one uses the residue theorem and the other Liouville’s theorem. Hence, unless one provides a "real analysis" proof of Euler's product, this approach is in fact deeply entrenched in complex analysis. – Did Jan 25 '13 at 08:05
  • @Did: boy you drive a hard bargain. OK, I will reproduce said proof, sans reference to anything complex. – Ron Gordon Jan 25 '13 at 08:11
  • @Did: there you go, all real. – Ron Gordon Jan 25 '13 at 08:56
  • Wonderful. Much better answer now. (+1) – Did Jan 25 '13 at 12:22
  • @Did: Thanks for pushing for it. – Ron Gordon Jan 25 '13 at 13:39