45

How to show the following equality? $$\sum_{n=0}^\infty\frac{1}{a^2+n^2}=\frac{1+a\pi\coth a\pi}{2a^2}$$

Spenser
  • 19,469

5 Answers5

24

Related problems: (I), (II). This problem is a direct application of Fourier transform and Poisson summation formula. Recalling the definition of Fourier transform and the Poisson summation formula respectively

$$ F(w) = \frac{1}{\sqrt{2\pi}}\int_{-\infty}^{\infty} f(x) e^{-ixw} dx \,,$$

$$ \sum_{-\infty}^{\infty} f(n) = \sqrt{2\pi}\sum_{-\infty}^{\infty} F(2n\pi)\,, $$

where $F$ is the Fourier transform of $f$. Advancing with our problem, first, we compute the Fourier transform of $ f(x)=\frac{1}{x^2+a^2} $ which is equal to

$$ F(w) = \sqrt{\frac{\pi}{2}}\frac{1}{a}e^{-a|w|}\,.$$

Applying Poisson formula, we have

$$ \sum_{n=0}^{\infty}\frac{1}{n^2+a^2} = \frac{\pi}{a}\sum_{n=0}^{\infty}e^{-2an\pi} = \frac{\pi}{a} \sum_{n=0}^{\infty}r^{n}=\frac{\pi}{a}\frac{1}{1-r}\,,\quad r = e^{-2 \pi a} \,,$$

$$\Rightarrow \sum_{n=0}^{\infty}\frac{1}{n^2+a^2} = \frac{\pi}{a} \frac{1}{1-e^{-2a\pi}}=\frac{\pi}{a} \frac{e^{2a\pi}}{e^{2a\pi}-1} \,. $$

Now, I leave it to you to manipulate the above expression to reach the form

$$ \sum_{n=0}^\infty\frac{1}{a^2+n^2}=\frac{1+a\pi\coth a\pi}{2a^2} $$

You can use the identity

$$ \coth x = \frac{\cosh x}{\sinh x} = \frac {e^x + e^{-x}} {e^x - e^{-x}} = \frac{e^{2x} + 1} {e^{2x} - 1} \,. $$

15

It is well known that

$$\sum_{n=-\infty}^\infty f(n)= -\sum_{j=1}^k \operatorname*{Res}_{z=j}\pi \cot (\pi z)f(z) $$

Assume $a \neq 0$.

To find the residues of $g(z) := \pi \cot (\pi z)\frac{1}{a^2+n^2}$, we see

$$\frac{1}{a^2+n^2} = \frac{1}{(n+ia)(n-ia)}$$

so $g$ has poles at $z_1 = ia$ and $z_2 = -ia$. Their respective residues, $b_1$ and $b_2$ can be found:

$$b_1 = \operatorname*{Res}_{z=ia}\,g(z) = \lim_{z \to ia} \pi \cot (\pi z)\frac{(z-ia)}{(z+ia)(z-ia)} = \pi \cot (\pi i a)\frac{1}{2ia} = -\frac{\pi \coth (\pi a)}{2a}$$

$$b_2 = \operatorname*{Res}_{z=-ia}\,g(z) = \lim_{z \to -ia} \pi \cot (\pi z)\frac{(z+ia)}{(z+ia)(z-ia)} = -\pi \cot (-\pi i a)\frac{1}{2ia} = -\frac{\pi \coth (\pi a)}{2a}$$

And finally:

$$\sum_{k=-\infty}^\infty \frac{1}{a^2+k^2} = -(b_1+b_2)=\frac{\pi \coth (\pi a)}{a}$$

To change the starting number from $-\infty$ to $0$, we divide the series, as it is symmetrical (i.e. $g(n)=g(-n)$):

$$ \sum_{k=-\infty}^\infty \frac{1}{a^2+k^2}= \frac{\pi \coth (\pi a)}{a}=\\ \sum_{k=-\infty}^{-1} \frac{1}{a^2+k^2}+\frac{1}{a^2}+\sum_{k=1}^\infty \frac{1}{a^2+k^2}=\\ \frac{1}{a^2}+2\sum_{k=1}^\infty \frac{1}{a^2+k^2}=\\ \frac{1}{a^2}+2\left(\sum_{k=0}^\infty \frac{1}{a^2+k^2}-\frac{1}{a^2}\right)=\\ 2\sum_{k=0}^\infty \frac{1}{a^2+k^2}-\frac{1}{a^2} $$

Thus

$$\sum_{k=0}^\infty \frac{1}{a^2+k^2} = \frac{\pi \coth (\pi a)}{2a}+\frac{1}{2a^2} = \frac{\pi a\coth (\pi a)+1}{2a^2}$$

Argon
  • 25,303
  • 4
    Another interesting thing to do with this sum is to start it from $k=0$ and then let $a \to 0$ to get $$\zeta(2) = \frac{\pi^2}{6}$$ – Argon Oct 07 '12 at 13:27
  • +1, great answer, but why you didn't take the pole at z=0 when you calculate the residues ? also at wolfram they didnt take the pole at z=0 but why ? – mnsh Sep 14 '13 at 18:47
  • @hmedan.mnsh The residue at zero is $\frac{1}{a^2+0^2}$, i.e. it is a term of the sum. – Argon Sep 15 '13 at 12:49
  • A rigorous treatment should include the proof of the vanishing of the contour as it approaches infinity, which this proof lacks. – Hans Feb 07 '18 at 08:19
  • @Hans It's a pretty widely known result, I didn't think it was necessary to include. – Argon Feb 11 '18 at 02:34
  • The least you should do is to cite that "well know result" to justify the procedure. In fact, this is crucial, because without it all that residues amount to naught. – Hans Feb 11 '18 at 23:36
  • I believe that top comment should say "Another interesting thing to do with this sum is to start it from $k=1$... " – Mason Sep 16 '18 at 01:40
  • @Mason Just so, thanks. – Argon Sep 16 '18 at 02:17
  • @Argon hello after this long time , I don't understand your first comment how can it be a solution for the bessel problem where did you find that $\frac{\pi^2}{6}$ and how did you get rid of $a$ –  Dec 04 '20 at 08:47
  • 1
    @mugiwara Move the $k=0$ term to the RHS and let $a\to0$. The left side becomes the sum $1/1^2+1/2^2+\cdots$ and the right side becomes $\pi^2/6$. – Argon Dec 04 '20 at 13:07
11

Now, a real analytic proof. This one has no flaws (I hope).

Lemma 1. Integration by parts gives: $$\frac{1}{a}\int_{0}^{+\infty}\cos(n x)\,e^{-a x}\,dx = \frac{1}{a^2+n^2} = \int_{0}^{+\infty}\frac{\sin(n x)}{n}\,e^{-a x}\,dx.$$

Lemma 2. The series $$\sum_{n=1}^{+\infty}\frac{\sin(nx)}{n}$$ converges on $\mathbb{R}\setminus 2\pi\mathbb{Z}$ to the function: $$ f(x) = \pi\left(\frac{1}{2}-\left\{\frac{x}{2\pi}\right\}\right).$$

Lemma 3. The dominated convergence theorem hence gives: $$\sum_{n=1}^{+\infty}\frac{1}{a^2+n^2}=\pi\int_{0}^{+\infty}\left(\frac{1}{2}-\left\{\frac{x}{2\pi}\right\}\right)e^{-ax}\,dx,$$ and by splitting $[0,+\infty)$ as $[0,2\pi)\cup[2\pi,4\pi)\cup\ldots$ we have:

$$\sum_{n=1}^{+\infty}\frac{1}{a^2+n^2}=\frac{e^{2a\pi}}{e^{2a\pi}-1}\int_{0}^{2\pi}\frac{\pi-x}{2}e^{-ax}dx=\frac{\pi a \coth(\pi a)-1}{2a^2}.$$

Jack D'Aurizio
  • 353,855
8

This one is a proof I gave when I was attending my high school, before studying complex analysis. It is a bit flawed, but just a little.

Step 1. If $p(x)$ is a real polynomial satisfying $p(0)=1$ and its roots are simple and real, $$\sum_{\xi:p(\xi)=0}\frac{1}{\xi}=-\frac{p'(0)}{p(0)}$$ follows from Vieta's theorem.

Step 2. All the roots of $\frac{\sin x}{x}$ are simple and real. Moreover, $$\frac{\sin x}{x}=\prod_{n=1}^{+\infty}\left(1-\frac{x^2}{\pi^2 n^2}\right)$$ holds. It is the Weierstrass product for the sine function.

Step 3. $\{a^2+1,a^2+2^2,a^2+3^2,\ldots\}$ is the zero set of the function: $$f(x)=\frac{\sinh\left(\pi\sqrt{a^2-x}\right)}{\pi\sqrt{a^2-x}}.$$

Step 4. Since $$f(0)=\frac{\sinh(\pi a)}{\pi a},\qquad f'(0)=-\frac{\cosh(\pi a)}{2a^2}+\frac{\sinh(\pi a)}{2\pi a^4},$$ Step 1 gives:

$$\sum_{n=1}^{+\infty}\frac{1}{n^2+a^2}=\frac{\pi a \coth(\pi a)-1}{2a^2}.$$


Known issues: the determination of the square root function and the fact that we can treat $\frac{\sin x}{x}$ like an "infinite degree" polynomial with known roots. Beyond the naif approach, this shows that the Vieta's theorem for polynomials and the residue theorem for meromorphic functions are very closely related.

Jack D'Aurizio
  • 353,855
5

This is what I have from an essay I wrote. I don't know if there's a more elementary way (or if it's completely correct).

Consider $f(z) = \dfrac{\cot{\pi z}}{z^2 + k}$. This will have residues at $z = \pm i \sqrt{k}$, and at $z = n$ for $n \in \mathbb{Z}$. At $z = n$, we can compute the residues as \begin{align*} \textrm{Res}_{z=n} f(z) & = \lim_{z \rightarrow n} \dfrac{(z-n) \cot{\pi z}}{z^2 + k} = \lim_{z \rightarrow n} \dfrac{(z-n)}{(z^2 + k) \tan{\pi z}} \\ & = \lim_{z \rightarrow n} \dfrac{1}{\pi (z^2 + k) \sec^2{\pi z} + 2z \tan{\pi z}} \\ & = \dfrac{1}{\pi (n^2 + k)}. \end{align*} We can calculate the residues at $z = \pm i \sqrt{k}$: $\displaystyle \textrm{Res}_{z=i\sqrt{k}} f(z) = \lim_{z\rightarrow i\sqrt{k}}\dfrac{(z-i\sqrt{k})\cot{\pi z}}{z^2 + k}$.

This equals:

$\lim_{z \rightarrow i\sqrt{k}} \dfrac{\cot{\pi z}}{z + i\sqrt{k}} = \dfrac{\cot{\pi i\sqrt{k}}}{2i\sqrt{k}}.$

It can be shown that the residue at $z = -i \sqrt{k}$ is the same, because $\cot{\pi z}$ is an odd function. And so the residue contribution from the two poles at $z = \pm i \sqrt{k}$ is

$-\dfrac{\cot{\pi i \sqrt{k}}}{i\sqrt{k}} = -\dfrac{1}{2\sqrt{k}} \dfrac{e^{2\pi \sqrt{k}} + 1}{e^{2\pi \sqrt{k}} - 1}$.

Hence, we have

$\displaystyle \int_\gamma f(z) dz = 2\pi i \left(\sum_{n \in \mathbb{Z}} \dfrac{1}{\pi(n^2 +k)} -\dfrac{1}{2\sqrt{k}} \dfrac{e^{2\pi \sqrt{k}} + 1}{e^{2\pi \sqrt{k}} - 1}\right)$.

It is tempting for the left-hand side to go to zero, which we can arrange. Take the large square contour centered at the origin with sidelength $2R$. Observe that since

$\cot{z} = i\dfrac{e^{2iz} + 1}{e^{2iz}-1}$,

in the limit as $|z| \geq R \rightarrow \infty$, we will have $|\cot{z}| \rightarrow 1$ since the numerator and denominator of $\cot{z}$ grow equally fast. Moreover, we have that:

$|z^2 + k| \geq |z^2| \geq R^2$,

and so the maximum modulus of $f(z)$ on $\gamma$ is $1/R^2$. By the ML-inequality, we have that

$\left|\displaystyle \int_\gamma f(z) dz\right| \leq 8R \cdot \dfrac{1}{R^2}$.

So as $R \rightarrow \infty$, the integral goes to zero. And thus, \begin{align*} \sum_{n \in \mathbb{Z}} \dfrac{1}{\pi(n^2 +k)} -\dfrac{1}{2\sqrt{k}} \dfrac{e^{2\pi \sqrt{k}} + 1}{e^{2\pi \sqrt{k}} - 1} & = 0\\ \sum_{n \in \mathbb{Z}} \dfrac{1}{\pi(n^2 +k)} & = \dfrac{1}{2\sqrt{k}} \dfrac{e^{2\pi \sqrt{k}} + 1}{e^{2\pi \sqrt{k}} - 1} \\ \sum_{n=1}^\infty \dfrac{1}{(n^2 +k)} & = \dfrac{\pi}{2\sqrt{k}} \dfrac{e^{2\pi \sqrt{k}} + 1}{e^{2\pi \sqrt{k}} - 1} - \dfrac{1}{2k}. \end{align*}

Taking $k = a^2$, this formula becomes

$\dfrac{a \pi \coth{\pi a} -1}{2a^2}$.

Hmm.. not sure about -1 or +1.

notes
  • 1,843
  • Thanks a lot. Its $-1$ if you start at $n=1$ and $+1$ if you start at $n=0$. – Spenser Oct 06 '12 at 19:13
  • Ah, I didn't see that. Awesome then! – notes Oct 06 '12 at 21:11
  • 1
    Your contour has some problems. For example "we will have $|\cot z|\to 1$" is not true when $R$ is a generic real number as $\cot(\pi z)$ blows up at $z = n$ where $n$ is an integer. If you restrict $R$ to be on the form $R = N + \frac{1}{2}$ with $N$ integer then $\cot(\pi z)$ is bounded (by $2$) on this contour and everything should be fine. – Winther Sep 12 '16 at 11:49