9

It is widely know the summation theorem via residues:

Suppose $f(z)$ analytic on $\mathbb{C}$ except for a finite number of poles. And let $f(z)$ be such that along the path $C_{N}$, the square with vertices at $\left(N+\frac{1}{2}\right)(-1+i), \left(N+\frac{1}{2}\right)(-1-i), \left(N+\frac{1}{2}\right)(1-i), \left(N+\frac{1}{2}\right)(1+i)$.

$$|f(z)|\leq\frac{M}{|z|^k}$$

where $k>1$ and $M$ are constants independent of $N$ we have

$$\sum_{n=-\infty}^{\infty} f(n) = -\sum \left\{\textrm{ residues of } \pi\cot(\pi z)f(z) \textrm{ at the poles of } f(z) \right\}$$

We can extend this theorem for functions with infinitely many poles:

Consider the path $C_{N}$. Let $f(z)$ with infinitely many poles with the conditions:

$$|f(z) |\leq \frac{M}{|z|^{k}} \tag{*}$$ along the path $C_{N}$ where $k>1$ and $M$ are constants independent of $N$ and

$$\lim_{N \to \infty} \sum_{a\in A_{N}} \operatorname{Res}\left(\pi\cot(\pi z)f(z),a\right)<\infty$$

where

$$A_{N} = \left\{ \textrm{poles of } f(z) \textrm{ inside } C_{N} \right\}$$

then:

$$\sum_{n=-\infty}^{\infty} f(n) = \lim_{N \to \infty} \sum_{a\in A_{N}} \operatorname{Res}\left(\pi\cot(\pi z)f(z),a\right)$$

the set of $C_{N}$ is called "Big squares".

  1. Can we weaken the condition $(*)$? For example, $f(z)$ to be $\mathcal{O}(z^{-2})$ at infinity?

$$|f(z)| \leq \frac{M}{|z|^{2}} \quad z \to \infty$$ or there are even weaker conditions?

  1. What is a good reference for the summation theorem for functions with infinitely many poles?

Here is my try for question 1:

Given that $f(z)$ is $\mathcal{O}(z^{-2})$ :

$\exists \delta,M>0$ so that whenever $|z|>\delta$

$$|f(z)|\leq\frac{M}{|z|^2}$$

If we choose our first square $C_{N}$ for $N$ sufficiently large such that $|z|>\delta$ then

$$\oint_{C_{N}} \pi \cot(\pi z)f(z) dz = \sum_{n=-N}^{N} f(n) + \sum_{a\in A_{N}} \operatorname{Res}\left(\pi\cot(\pi z)f(z),a\right) $$

We just have to prove that

$$\oint_{C_{N}} \pi \cot(\pi z)f(z) dz \to 0 \quad N \to \infty$$

But it can be proven that $$|\cot(\pi z)|$$ is bounded in $C_{N}$:

$$|\cot(\pi z)|\leq A$$

Then

$$\left| \oint_{C_{N}} \pi \cot(\pi z) f(z) dz\right| \leq \frac{\pi A M}{N^2}(8N+4)$$

since the lenght of the path $C_{N}$ is $8N+4$.

Hence, the integral vanishes as $N \to \infty$.

Bertrand87
  • 2,171

1 Answers1

2

EDIT: I reorganized my answer.


The issue is that a meromorphic function with infinitely many poles can't be $O(z^{-k}), \ k >1, $ as $|z| \to \infty$ since if would require the poles of the function to all be located inside a circle of finite radius centered at the origin.

The poles would invariabily accumulate near some point, contradicting the fact that $f(z)$ only has isolated singularities.

It would be better to write $f(z)$ as the product $f(z) = g(z)h(z)$, where $g(z)$ is a meromoprhic function with infintiely many poles, and $h(z)$ is $O(z^{-k}), \ k>1$, as $|z| \to \infty$.

You would then need to argue that $g(z)$ is uniformly bounded on the contour like $\cot(\pi z)$.


If you stay away from the negative real axis, then the polygamma function $\psi^{(2)}(z)$ is a meromoprhic function with infinitely many poles that is $O(z^{-2})$ as $|z| \to \infty$.

However, you would need to examine how $\psi^{2}(z)$ behaves on the part of the contour that passes halfway between adjacent poles on the negative real axis as $N$ increases.


And when using the kernel $\pi \csc(\pi z)$ to evaluate an alternating series, you only need $h(z)$ to be $O(z^{-k}), \ k>0,$ as $|z| \to \infty$.

Basically it has to do with the fact that the magnitude of $\csc(\pi z)$ decays exponentially to zero as $\Im(z) \to \pm \infty$. See the answers to this question.

And depending on how $g(z)$ behaves as $\Re(z) \to \pm \infty$, $h(z) =1$ might suffice.

  • 1
    The issue is the existence of such a function. I changed my answer rather significantly. – Random Variable Nov 12 '21 at 02:38
  • Ok, I also rewrote my comment becuase I did no like it either. Suppose that the jordan curve $C_{N}$ (the square described in the question) contains the infinitely many poles of $f(z)$ when $N$ approachs infinity. I think this Jordan curve fit well for many functions with infinitely many poles: $\operatorname{coth}(\pi z), \operatorname{sech}(\pi z)$ and some others. Then a necessary condition for $$ \int_{C_{N}} \coth(\pi z) f(z) dz \to 0 ; \textrm{ as } N \to \infty$$ is that $f(z)$ is uniformly bounded in $C_{N}$ Can we conclude that? – Bertrand87 Nov 12 '21 at 03:08
  • 1
    @zinoviev Not quite. Going back to my previous answer, the reason $\lim_{N \to \infty} \int_{C_{N}} \frac{\cot(\pi z) \coth(\pi z)}{z^{3}} , \mathrm dz$ vanishes is because $|\cot(\pi z)||\coth(\pi z)| \le B$ on the contour, while $\frac{1}{|z|^{3}}$ is going to zero faster that $\frac{1}{|z|^{1+\epsilon}}$. The argument is similar if $\coth(\pi z)$ is replaced with $\operatorname{sinh}(\pi z)$. But if $\coth(\pi z)$ is replaced with $\operatorname{sech}(\pi z)$, then the contour would need to be adjusted since $\operatorname{sech}(\pi z)$ has poles at the imaginary half-integers. – Random Variable Nov 12 '21 at 04:13
  • Sorry, I meant $\operatorname{csch}(\pi z)$ not $\operatorname{sech}(\pi z)$ in my comment. So, if we suppose that $C_{N}$ encloses the poles of $f(z)$ the best condition continues being that $f$ is such that along the path $C_{N}$ $$|f(z)|\leq \frac{M}{|z|^k}$$ where $k>1$ and $M$ are constants independient of $N$ in order to get $$\oint_{C_{N}} \pi\coth(\pi z)f(z) dz \to 0$$ . Do you know standard literature for this theorem for functions with infinitely many poles? – Bertrand87 Nov 12 '21 at 06:07
  • 1
    Yes, in the case that $f(z)$ has infinitely many poles, the behavior of $f(z)$ on the contour as $N \to \infty$, and not just how $f(z)$ behaves when $|z|$ is large, would seem to be important. I asked a question here a long time ago that is sort of related. – Random Variable Nov 12 '21 at 07:31
  • @zinoviev I meant $\operatorname{csch}(\pi z)$ in my previous comment, not $\sinh(\pi z)$, which is definitely not uniformly bounded on the contour. – Random Variable Nov 12 '21 at 13:59
  • Take a look at Lema 1.1 (page 3) at this paper https://arxiv.org/pdf/1708.02236.pdf Look at the hypothesis of the Lemma (the author proposed an alternative conditon below the brief proof). Tha author used this Lemma to evaluate series involving functions with infinite poles in table 2.1 (pages 5 and 6 ). What is your take on it? – Bertrand87 Nov 12 '21 at 15:28