Is there a closed form for: $$f(x)=\sum_{k=1}^\infty (-1)^k \ln \left( \tanh \frac{\pi k x}{2} \right)=2\sum_{n=0}^\infty \frac{1}{2n+1}\frac{1}{e^{\pi (2n+1) x}+1}$$

This sum originated from a recent question, where we have:

$$f(1)= -\frac{1}{\pi}\int_0^1 \ln \left( \ln \frac{1}{x} \right) \frac{dx}{1+x^2}=\ln \frac{\Gamma (3/4)}{\pi^{1/4}}$$

If we differentiate w.r.t. $x$, we obtain:

$$f'(x)=\sum_{k=1}^\infty (-1)^k \frac{\pi k}{\sinh \pi k x}$$

There is again a closed form for $x=1$ (obtained numerically):


So, is there a closed form or at least an integral definition for arbitrary $x>0$?

The series converges absolutely (numerically at least):

$$\sum_{k=1}^\infty \ln \left( \tanh \frac{\pi k x}{2} \right)< \infty$$

Thus, this series can also be expressed as a logarithm of an infinite product:

$$f(x)=\ln \prod_{k=1}^\infty \tanh (\pi k x) - \ln \prod_{k=1}^\infty \tanh \left( \pi (k-1/2) x \right)$$

$$e^{f(x)}= \prod_{k=1}^\infty \frac{\tanh (\pi k x)}{\tanh \left( \pi (k-1/2) x \right)}$$

This by the way leads to:

$$\prod_{k=1}^\infty \frac{\tanh (\pi k)}{\tanh \left( \pi (k-1/2) \right)}=\frac{\pi^{1/4}}{\Gamma(3/4)}$$

I feel like there is a way to use the infinite product form for $\sinh$ and $\cosh$:

$$\sinh (\pi x)=\pi x \prod_{n=1}^\infty \left(1+\frac{x^2}{n^2} \right)$$

$$\cosh (\pi x)=\prod_{n=1}^\infty \left(1+\frac{x^2}{(n-1/2)^2} \right)$$

  • 1
    $\begingroup$ In my answer to the linked question I have shown that the series on right equals $$\frac{1}{2}\log\vartheta_{3}(q)$$ where $q = e^{-\pi x}$. I don't think there is any closed form different from theta functions and their cousins elliptic integrals. $\endgroup$ – Paramanand Singh Feb 26 '17 at 16:14
  • 2
    $\begingroup$ One can get a closed form if one considers the function $$g(x) = \sum_{n = 1}^{\infty}\log\tanh \frac{n\pi x}{2} = \log\vartheta_{4}(q)$$ (see math.stackexchange.com/a/1793756/72031) and then $$g(x) - f(x) = \frac{1}{4}\log\frac{\vartheta_{4}^{4}(q)}{\vartheta_{3}^{4}(q)} = \frac{1}{4}\log(1 - k^{2})$$ so that if $x = \sqrt{r}, r \in\mathbb{Q}^{+}$ then $k$ is algebraic and we have a closed form for $g(x) - f(x)$. $\endgroup$ – Paramanand Singh Feb 26 '17 at 16:42
  • $\begingroup$ @ParamanandSingh, oh, I didn't read your answer carefully enough. I haven't noticed that you give the general closed form $\endgroup$ – Yuriy S Feb 26 '17 at 16:45
  • 2
    $\begingroup$ The general form is proved there, but it is evaluated in terms of Gamma values only for specific value of $q = e^{-\pi}$ (which is specific to that integral like your $f(1)$ instead of general $f(x)$). $\endgroup$ – Paramanand Singh Feb 26 '17 at 16:51

Let’s use $~\displaystyle\prod\limits_{k=1}^\infty (1+z^k)(1-z^{2k-1}) =1~$ . $\enspace$ (It's explained in a note below.)

For $~z:=q^2~$ and $~q:=e^{-\pi x}~$ with $~x>0~$ we get

$\displaystyle e^{f(x)} = \prod\limits_{k=1}^\infty\frac{\tanh(k\pi x)}{\tanh((k-\frac{1}{2})\pi x)} = \prod\limits_{k=1}^\infty\frac{ \frac{q^{-k}-q^k}{q^{-k}+q^k} }{ \frac{q^{\frac{1}{2}-k}-q^{k-\frac{1}{2} }}{q^{\frac{1}{2}-k}+q^{k-\frac{1}{2}}} } = \prod\limits_{k=1}^\infty\frac{(1-q^{2k})(1+q^{2k-1})}{(1+q^{2k})(1-q^{2k-1})} =$

$\displaystyle = \prod\limits_{k=1}^\infty (1-q^{2k})(1+q^{2k-1})^2 = \sum\limits_{k=-\infty}^{+\infty} q^{k^2} = \vartheta(0;ix)$

The “closed form” for $\,f\,$ is:

$$f(x) = \ln\vartheta(0;ix)$$

Please see e.g. Theta function .


$\displaystyle\prod\limits_{k=1}^\infty (1+z^k)(1-z^{2k-1}) =1$

$\Leftrightarrow\hspace{2cm}$ (logarithm)

$\displaystyle \sum\limits_{k=1}^\infty \sum\limits_{v=1}^\infty \frac{(-1)^{v-1}z^{kv}}{v} = \sum\limits_{k=1}^\infty \ln(1+z^k) = -\sum\limits_{k=1}^\infty \ln(1-z^{2k-1}) = \sum\limits_{k=1}^\infty \sum\limits_{v=1}^\infty \frac{z^{(2k-1)v}}{v}$

$\Leftrightarrow\hspace{2cm}$ (exchanging the sum symbols which is valid for $~|z|<1~$

$\hspace{2.7cm}$ and using $~\displaystyle\frac{x}{1-x}=\sum\limits_{k=1}^\infty x^k~$)

$\displaystyle\sum\limits_{v=1}^\infty \frac{(-1)^{v-1}}{v}\frac{z^v}{1-z^v} = \sum\limits_{v=1}^\infty \frac{1}{v}\frac{z^v}{1-z^v} - 2\sum\limits_{v=1}^\infty \frac{1}{2v}\frac{z^{2v}}{1-z^{2v}} = \sum\limits_{v=1}^\infty \frac{1}{v}\frac{z^v}{1-z^{2v}}$

  • 1
    $\begingroup$ Where does the first identity come from? "Let's use..." $\endgroup$ – Diger Mar 22 '19 at 11:27
  • 2
    $\begingroup$ @Diger : I've added a note . ;) $\endgroup$ – user90369 Mar 22 '19 at 12:54
  • 1
    $\begingroup$ Very nice answer, thank you! I haven't checked numerically yet, but I trust you $\endgroup$ – Yuriy S Mar 25 '19 at 8:03
  • 1
    $\begingroup$ @YuriyS : You are welcome! And thank you for paying attention to my answer although my answer is 2 years too late. ;) Explicit values can be seen e.g. at the web page for Theta function (the link above). $\endgroup$ – user90369 Mar 25 '19 at 9:08

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.