3
$\begingroup$

Maybe it's a very simple question, but I have a problem with the following series $$\sum\limits_{n=1}^{\infty}\frac{1}{n^2+pn+q},$$ where $p, q \in \mathbb{R}$. I know about five ways how to calculate the series $$\sum\limits_{n=1}^{\infty}\frac{1}{n^2+r^2}, \,\, r>0.$$ The most "beautiful" way for me is a Poisson summation formula: $$\sum\limits_{n=-\infty}^{+\infty} f(n) = \sum\limits_{n=-\infty}^{+\infty}\int\limits_{-\infty}^{+\infty} f(x)e^{2\pi nix}dx$$ Hence, for $f(x)=(x^2+r^2)^{-1}$ $$\int\limits_{-\infty}^{+\infty} \frac{e^{2\pi nix}dx}{x^2+r^2} = \frac{\pi}{r}e^{-2\pi|n|r}$$ (this integral can be taken by residue). Therefore, $$\sum\limits_{n=-\infty}^{+\infty} f(n)= \sum\limits_{n=-\infty}^{+\infty}\frac{1}{n^2+r^2}= \frac{1}{r^2}+2\sum\limits_{n=1}^{+\infty}\frac{1}{n^2+r^2} = \sum\limits_{n=-\infty}^{+\infty}\frac{\pi}{r}e^{-2\pi|n|r}=\frac{\pi}{r}+\frac{2\pi}{r}\frac{e^{-2\pi r}}{1-e^{-2\pi r}}.$$ Finally, $$\sum\limits_{n=1}^{+\infty}\frac{1}{n^2+r^2}=\frac{1}{2r}(\pi+\frac{2\pi e^{-2\pi r}}{1-e^{-2\pi r}}-\frac{1}{r}).$$ Also I know how to calculate the series $$\sum\limits_{n=1}^{\infty}\frac{1}{(n+r)^2}, \,\, r>-1.$$ I prefer a Calabi's method. $$\frac{1}{(n+r)^2}=\int\limits_0^1\int\limits_0^1 (xy)^{n+r-1}dxdy.$$ Then $$\sum\limits_{n=1}^{\infty}\frac{1}{(n+r)^2}=\sum\limits_{n=1}^{\infty}\int\limits_0^1\int\limits_0^1 (xy)^{n+r-1}dxdy=\int\limits_0^1\int\limits_0^1(\sum\limits_{n=1}^{\infty} (xy)^{n+r-1})dxdy=\int\limits_0^1\int\limits_0^1 \frac{(xy)^r dxdy}{1-xy}.$$ After rotation of the coordinate system by $\pi/4$ and separation of the initial integral into two components, we can obtain a very complex recursive formula.

But what about the series of the form $$\sum\limits_{n=1}^{\infty}\frac{1}{n^2+pn+q},$$ are there any appropriate methods at least for some $p, q \in \mathbb{R}$?

$\endgroup$
4
  • 1
    $\begingroup$ According to sage closed form is $\frac{\psi\left(\frac{1}{2} \, p + \frac{1}{2} \, \sqrt{p^{2} - 4 \, q} + 1\right) - \psi\left(\frac{1}{2} \, p - \frac{1}{2} \, \sqrt{p^{2} - 4 \, q} + 1\right)}{\sqrt{p^{2} - 4 \, q}}$ $\endgroup$
    – joro
    Commented Jan 10, 2014 at 12:37
  • $\begingroup$ @joro Am I right that $\psi$-function in your (sage) notation is a derivative of the logarithm of the Gamma function? Unfortunately, for me it is not an option. Analysis of the properties of the Gamma function or Euler's totient function is more complex than initial problem. I need some numerical method (maybe with some level of accuracy), where I can quickly estimate this series. $\endgroup$
    – Mariarty
    Commented Jan 10, 2014 at 13:24
  • $\begingroup$ "The digamma function, psi(x), is the logarithmic derivative of the gamma function". Is is efficiently computable numerically. $\endgroup$
    – joro
    Commented Jan 10, 2014 at 13:27
  • $\begingroup$ In parametric form, I think, it isn't so efficient. I need a form like I have represented for special cases: through elementary functions in parametric form. I had hoped that this work with given level of accuracy has already been done previously. Anyway, thank you. $\endgroup$
    – Mariarty
    Commented Jan 10, 2014 at 14:59

1 Answer 1

4
$\begingroup$

This can be expressed in terms of elementary functions, if $p/2$ is an integer. Suppose, for example that it is a positive integer. Then your sum is $$S:=\sum_{m=p/2+1}^\infty\frac{1}{m^2+c}=\frac{1}{2}\sum_{|m|>p/2}\frac{1}{m^2+c},\quad c=q-p^2/4,$$ where summation in the last sum is over positive and negative integers. The sum in the right hand side differs by finitely many summands from the sum $$\sum_{-\infty}^\infty\frac{1}{m^2+c},$$ which you know how to compute. The simplest way to do this, by the way, is by the residue theorem: integrate $$\frac{\pi\cot\pi z}{z^2+c}$$ over appropriate expanding contours. The answer will be slightly different, depending on whether $\sqrt{-c}$ is an integer or not. Thus you obtain a closed form answer in elementary functions. The case when $p/2$ is a negative integer it treated similarly, by first dropping finitely many terms of your sum, and then adding them back.

If $p/2$ is not an integer, I afraid that you have to use Gamma function, as suggested in the comments, and there is no way around this.

$\endgroup$
1
  • $\begingroup$ It's exactly, what I have just obtained while asking a question, but I hoped that there is more general result. Thank you. $\endgroup$
    – Mariarty
    Commented Jan 10, 2014 at 20:04

You must log in to answer this question.

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