Saturday, July 18, 2015

Complex Analysis and the Basel Problem

Arguably the oldest and most famous infinite series in mathematics is the surprising and celebrated Basel series or Basel problem $$S = \sum_{n=1}^\infty\frac 1{n^2}.$$ For hundreds of years the closed form $$S=\frac{\pi^2}6$$ has been known, and has been further used to prove scores of other results involving integrals, special functions, and the like. However, the series is quite hard to evaluate - Euler's original proof involving Taylor expansions of trigonometric functions is ambiguous and in modern times, without rigorization, hardly more than a heuristic. In this post we will prove the identity above by means of complex analysis, using the residue theorem. The methods that we will use prove useful in the evaluation of all sorts of infinite series. Source: the ideas from this derivation are based upon those in this document.

Let's begin. For convenience, write $f(z)=z^{-2}$ so that $S=\sum_{n=1}^\infty f(n)\equiv \sum f(n)$. Define $$g(z)=f(z)\cot(\pi z)=\frac{\cos(\pi z)}{z^2\sin(\pi z)}.$$ We would like to study, and further understand, the pole structure of $g(z)$ by enumerating all poles and calculating the residues at these poles. This is not a particularly hard task as we just need to find all values that make the denominator vanish: i.e. we need to solve $$z^2\sin(\pi z)=0.$$ Obviously $\sin(\pi z)$ vanishes at all real integer values of $z$, in particular at $z=\pm 1,\pm 2,\dots$ and $z=0$. The only possible way for $z^2$ to vanish is $z=0$, and this is already accounted for by the sine. Thus we have a whole bunch of simple poles, one at each nonzero integer, and an order-three pole at the origin.

Figure 1: A plot of the denominator $z^2\sin(\pi z)$
on $\mathbb R$ showing the integral zeroes
(which correspond to poles in $g$).

Next, we should compute the residues at these poles (throughout $k$ denotes an integer). Since there's only an order one divergence corresponding to the poles $k\neq 0$, the poles are simple and the residue is then $$r_k = \operatorname{Res}_{z=k}\frac{\cos(\pi z)}{z^2\sin(\pi z)} = \lim_{z\to k}\frac{(z-k)\cos(\pi z)}{z^2\sin(\pi z)}.$$ To evaluate this take out the convergent part $\frac{\cos(\pi z)}{z^2}$: $$r_k=\frac{\cos(\pi k)}{k^2}\lim_{z\to k}\frac{z-k}{\sin(\pi z)}.$$ In the rightmost limit we can apply L'Hopital's rule to get $$r_k = \frac{\cos(\pi k)}{k^2}\lim_{z\to k}\frac 1{\pi\cos(\pi z)}=\frac 1{\pi k^2}$$ and we have completed our evaluation of the first-order residues.

What about the order-three residue $r_0$ at $z=0$? Here, we have some options - either do some ugly series expansions to find a certain coefficient, or do some ugly derivatives. We'll do the latter (exericse: show that the other way is equivalent!) Using the standard formula for a higher-order pole: $$r_0 = \frac 1{2!}\lim_{z\to 0}\frac{d^2}{dz^2}\left(\frac{(z-0)^3\cos(\pi z)}{z^2\sin(\pi z)}\right) = \frac 1 2\lim_{z\to 0}\frac{d^2}{dz^2}(z\cot(\pi z)),$$ we compute a juicy second derivative (using WolframAlpha if you're sane) to get $$r_0=\pi\lim_{z\to 0}\big[\csc^2(\pi z)(\pi z\cot(\pi z)-1)\big]=\pi\lim_{z\to 0}\big[\csc^2(z)(z\cot z-1)\big].$$ This is an easy limit using common truncated series expansions of $\csc^2(x)\approx\frac 1 {x^2}$ and $\cot(x)\approx\frac 1 x-\frac x 3$: $$r_0=\pi\lim_{z\to 0}\frac 1{x^2}\left(\frac x x-\frac{x^2}3-1\right)=\pi\lim_{z\to 0}-\frac 1 3=-\frac{\pi}3.$$ With these simple computational lemmas in hand, we can move on to the interesting part of the derivation.

The key idea is to integrate the function $g$ (that we've just computed the poles and residues of) around squares of ever-increasing side length around the origin indexed by some number $n$. From the residue theorem (and our work above) we will get a term proportional a partial sum of $S$. We'll then send $N\to\infty$ and show that this gives the result. The square contours $\Gamma_n$ will be positively oriented in the usual way, and we will set the vertices to be at $(1+i)(n+1/2),\;(1-i)(n+1/2),\;(-1+i)(n+1/2),\;(-1-i)(n+1/2).$ We multiply by $n+1/2$ rather than $n$ in order to pick up $n$ positive poles rather than $n-1$ and keep the calculation clean.

Figure 2: The concentric squares $\Gamma_i$
encircling the origin. Note that $\Gamma_n$
picks up $2n+1$ poles as is clear from the picture.

Using our calculations from above, applying the residue theorem to the integral on $\Gamma_n$ of $g$ gives $$\oint_{\Gamma_n}\frac{\cos(\pi z)}{z^2\sin(\pi z)}dz=2\pi i\left(-\frac\pi 3+2\sum_{i=1}^n\frac 1{\pi k^2}\right)$$ with the factor of $2$ in front of the sum compensating for the over-counting due to negative residues. If we could show that the integral vanished as $n\to\infty$, then we would have $$0=-\frac{2\pi^2i}3+4iS$$ by simple algebra which immediately tells us that $$S = \frac{2\pi^2 i}{3\cdot 4i}=\frac{\pi^2}6.$$ Thus it remains to prove:

Lemma 1. $$\oint_{\Gamma_n}\frac{\cos(\pi z)}{z^2\sin(\pi z)}dz\to 0\text{   as   }n\to\infty.$$

Proof. The remainder of this derivation is a rather boring and technical exercise in complex bounding using the length inequality $$\int_\Gamma f(z)dx\leq\operatorname{length}\Gamma\cdot\max_{z\in\Gamma}f(z).$$ Obviously $\operatorname{length}\Gamma_n = 4(2n+1) = 8n+4$, so it remains to bound the function $g$ on $\Gamma_n$ in such a way that that it assures that the maximum decreases strictly faster than $O(n)$.

Looking at Figure 2, the closest point on $\Gamma_n$ to the origin is any of the four points on the real/imaginary axes, which each have modulus $n+\frac 1 2$. Therefore on $\Gamma_n$ we have $$\vert z\vert\geq n+\frac 1 2\implies\left\vert\frac 1{z^2}\right\vert\leq\frac 1{(n+1/2)^2}\leq\frac 1{n^2},$$ so that $$\left\vert\frac{\cos(\pi z)}{z^2\sin(\pi z)}\right\vert\leq\frac 1{n^2}\vert\cot(\pi z)\vert$$ on $\Gamma_n$. This simplifies our problem drastically: we now need to show that the cotangent never blows up too fast. In particular we'll bound it from above by a constant $C$ which doesn't depend on $n$. This is quite easy to do using identities for the modulus of the cosine of some complex number. If you get stuck, see the link at the beginning of the post and scroll down. $\square$

Thus the proof is completed, and indeed $S=\frac{\pi^2}6.$

No comments:

Post a Comment