I'm interested in computing the sum for $y>0$: $$ \sum_{n=1}^{\infty}\left[ \tan^{-1}\left( \frac{y}{n} \right) - \frac{y}{n} \right] $$

In formula (42.1.6) of Hansen's A Table of Series and Products, there is a formula: $$ \sum_{n=-\infty\\(n\neq0)}^{\infty}\left[ \tan^{-1}\left( \frac{y}{n+x} \right) - \frac{y}{n} \right] \ = \ \tan^{-1}\big( \tanh(\pi y) \cot(\pi x) \big) - \tan^{-1}\left( \frac{y}{x} \right) $$

In taking $x \to 0^{+}$, the above formula becomes the following: $$ \sum_{n=-\infty\\(n\neq0)}^{\infty}\left[ \tan^{-1}\left( \frac{y}{n} \right) - \frac{y}{n} \right] \ = \ 0 $$

Which simplifies to $\sum_{n=1}^{\infty}\left[ -\tan^{-1}\left( \frac{y}{n} \right) + \frac{y}{n} \right] + \sum_{n=1}^{\infty}\left[ \tan^{-1}\left( \frac{y}{n} \right) - \frac{y}{n} \right] = 0$, which is obvious.

Is there a way to sum this series?


$\newcommand{\bbx}[1]{\,\bbox[15px,border:1px groove navy]{\displaystyle{#1}}\,} \newcommand{\braces}[1]{\left\lbrace\,{#1}\,\right\rbrace} \newcommand{\bracks}[1]{\left\lbrack\,{#1}\,\right\rbrack} \newcommand{\dd}{\mathrm{d}} \newcommand{\ds}[1]{\displaystyle{#1}} \newcommand{\expo}[1]{\,\mathrm{e}^{#1}\,} \newcommand{\ic}{\mathrm{i}} \newcommand{\mc}[1]{\mathcal{#1}} \newcommand{\mrm}[1]{\mathrm{#1}} \newcommand{\pars}[1]{\left(\,{#1}\,\right)} \newcommand{\partiald}[3][]{\frac{\partial^{#1} #2}{\partial #3^{#1}}} \newcommand{\root}[2][]{\,\sqrt[#1]{\,{#2}\,}\,} \newcommand{\totald}[3][]{\frac{\mathrm{d}^{#1} #2}{\mathrm{d} #3^{#1}}} \newcommand{\verts}[1]{\left\vert\,{#1}\,\right\vert}$ \begin{align} &\bbox[10px,#ffd]{\ds{\sum_{n = 1}^{\infty}\bracks{\arctan\pars{y \over n} - {y \over n}}}} = -\int_{0}^{y}\sum_{n = 1}^{\infty}{x^{2} \over n\pars{n^{2} + x^{2}}}\,\dd x \\[5mm] = &\ -\,\Im\int_{0}^{y}\sum_{n = 1}^{\infty}{x \over n\pars{n - \ic x}}\,\dd x = -\,\Im\int_{0}^{y}\sum_{n = 0}^{\infty} {x \over \pars{n + 1}\pars{n + 1 - \ic x}}\,\dd x \\[5mm] = &\ -\,\Im\int_{0}^{y} x\,{\Psi\pars{1} - \Psi\pars{1 - \ic x} \over \ic x}\,\dd x\qquad \pars{~\Psi:\ Digamma\ Function~} \\[5mm] = &\ \Re\int_{0}^{y} \bracks{-\gamma - \Psi\pars{1 - \ic x}}\,\dd x\qquad \pars{~\gamma:\ Euler\!-\!Mascheroni\ Constant~} \\[5mm] = &\ -\gamma\, y - \Re\bracks{\ic\,\ln\pars{\Gamma\pars{1 - \ic y}}} = \bbx{-\gamma\, y + \Im\ln\pars{\Gamma\pars{1 - \ic y}}} \end{align}

  • $\begingroup$ Hi Felix my friend. This is very similar to how I did this (and deleted after I saw your post). But I don't believe that the closed form we obtained is more useful than the original series for numerical computation. $\endgroup$ – Mark Viola Apr 3 '18 at 2:38
  • $\begingroup$ @MarkViola Hi, Mark my friend. I saw your answer and it's right. That's true: The involved $\Gamma$ will require a numerical evaluation as the original post. $\endgroup$ – Felix Marin Apr 3 '18 at 2:43

By the inverse Laplace transform, assuming $y\in(0,1)$,

$$ \sum_{n\geq 1}\left(\arctan\frac{y}{n}-\frac{y}{n}\right)=\int_{0}^{+\infty}\left(-y+\frac{\sin(sy)}{s}\right)\frac{ds}{e^s-1}$$ and the RHS can be represented as $$\sum_{n\geq 1}\frac{(-1)^n y^{2n+1} \zeta(2n+1)}{(2n+1)}\tag{"odd" series}$$ which, on the other hand, does not simplify nicely like the similar $$ \sum_{n\geq 1}\frac{(-1)^n y^{2n}\zeta(2n)}{2n} = \log\sqrt{\frac{\pi y}{\sinh(\pi y)}}\tag{"even" series}.$$ Indeed the "even" series is related to $\text{Re} \log\Gamma(1+iy)$ which simplifies thanks to the reflection formula, while your series is given by the "odd" series, related to $\text{Im}\log\Gamma(1+iy)$. For values of $y$ close to zero a numerical approximation by series acceleration is pretty simple; for very large values of $y$ Stirling's approximation is recommended.

  • $\begingroup$ I am quite interested how the last equation is obtained. Could you please provide some related references? $\endgroup$ – Szeto Apr 3 '18 at 2:27
  • $\begingroup$ @Szeto: the $\log\Gamma$ function is involved and $\Gamma(1-is)\Gamma(is)$ simplifies by the reflection formula. $\endgroup$ – Jack D'Aurizio Apr 3 '18 at 2:32
  • $\begingroup$ When $y = 1$, the result is $\approx -0.2756$ but your result yields $\approx -0.6509$. $\endgroup$ – Felix Marin Apr 3 '18 at 2:34
  • $\begingroup$ How did you use the ILT to arrive at the first equation? $\endgroup$ – Mark Viola Apr 3 '18 at 2:40
  • $\begingroup$ @FelixMarin: the series involving $\zeta(2n+1)$ returns the correct outcome, the second series is shown just for providing a similar example. $\endgroup$ – Jack D'Aurizio Apr 3 '18 at 2:41

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.