3

I want to prove that $$\lim_{x\to1^-}(1-x)\sum_{n=1}^{\infty}(-1)^{n-1}\frac{nx^n}{1-x^{2n}} = \frac14$$

So far I tried to manipulate the series for instance using $$\sum_{n=1}^{\infty}(-1)^{n-1}\frac{nx^n}{1-x^{2n}} = -\sum_{n=1}^{\infty}(-1)^{n}nx^n\sum_{m=0}^{\infty}\left(x^{2n}\right)^m$$ since $x < 1$. Interchanging the two sums (not sure if allowed) I obtained, assuming I did not make mistakes, the sum $$\sum_{m=0}^{\infty}\frac{x^{2m+1}}{(1+x^{2m+1})^2}$$ I am unable to continue from this point. Perhaps my work isn't actually useful at all. Can you help me?

Peanut
  • 1,664

2 Answers2

2

Your computation is very close to the answer. If we denote the sum by $S(x)$, then

$$ S(x) = (1 - x) \sum_{m=0}^{\infty} \frac{x^{2m+1}}{(1+x^{2m+1})^2} = \frac{1}{1+x} \sum_{m=0}^{\infty} \frac{x^{2m+1} - x^{2m+3}}{(1 + x^{2m+1})^2}. $$

Now the idea is that $S(x)$ can be regarded as a Riemann sum for $\frac{1}{(1+t)^2}$ over $[0, 1]$. To make use of this idea, note that $t \mapsto \frac{1}{(1+t)^2}$ is decreasing for $t \geq 0$, and so

$$ x^2 \int_{x^{2m+1}}^{x^{2m-1}} \frac{\mathrm{d}t}{(1+t)^2} \leq \frac{x^{2m+1} - x^{2m+3}}{(1 + x^{2m+1})^2} \leq \int_{x^{2m+3}}^{x^{2m+1}} \frac{\mathrm{d}t}{(1+t)^2}. $$

Summing this over $m = 0, 1, 2, \dots$, we obtain

$$ \frac{x^2}{1+x} \int_{0}^{1/x} \frac{\mathrm{d}t}{(1+t)^2} \leq S(x) \leq \frac{1}{1+x} \int_{0}^{x} \frac{\mathrm{d}t}{(1+t)^2}. $$

Therefore, letting $x \to 1^-$ yields

$$ \lim_{x \to 1^-} S(x) = \frac{1}{2} \int_{0}^{1} \frac{\mathrm{d}t}{(1+t)^2} = \frac{1}{4}. $$

Sangchul Lee
  • 167,468
  • Wonderful explanation! As always! – Peanut Nov 23 '20 at 16:31
  • Simplicity at its best (+1)! The sum in question belongs to the theory of theta as discussed in my answer. – Paramanand Singh Nov 24 '20 at 07:47
  • @ParamanandSingh, Thank you! I am aware that this kind of series opens up a door to theta functions, Lambert series, etc. But I am by no means an expert to those topics, so I always resort to some elementary means. – Sangchul Lee Nov 24 '20 at 07:50
1

The sum in question belongs more properly to the theory of theta functions and elliptic integrals and this is an approach which makes use of standard results from this theory.


Let's put $q=-x$ so that the expression under limit becomes $$-(1+q)\sum_{n=1}^{\infty} \frac{nq^n} {1-q^{2n}}$$ The sum above can be written as $$\sum_{n\geq 1}\left(\frac{nq^{n}}{1-q^{n}}-\frac{nq^{2n}}{1-q^{2n}}\right)$$ which in terms of Ramanujan function $$P(q)=1-24\sum_{n\geq 1}\frac{nq^n}{1-q^n} $$ becomes $$\frac{P(q^2)-P(q)}{24}$$ and it follows that the original expression under limit is $$(1-x)\cdot\frac{P(-x)-P(x^2)}{24}$$ Let's replace this variable $x$ again by $q$ (for convention) and the limit we seek is $$\lim_{q\to 1^-}(1-q)\cdot\frac{P(-q)-P(q^2)} {24}$$ Treating $q$ as the nome we use the following standard results \begin{align} q&=\exp\left(-\pi\frac{K'} {K} \right)\notag\\ P(-q) &=\left(\frac{2K}{\pi}\right)^2\left(\frac{6E}{K}+4k^2-5\right)\notag\\ P(q^2) &=\left(\frac{2K}{\pi}\right)^2\left(\frac{3E}{K}+k^2-2\right)\notag \end{align} (for proofs see this post) and the expression under limit equals $$(1-e^{-\pi K'/K}) \frac{1}{8}\left(\frac{2K}{\pi}\right)^2\left(\frac{E} {K} - k'^2\right)$$ where moduli $k, k'$ and elliptic integrals $K, K'$ correspond to nome $q$. As $q\to 1^-$ we have $k\to 1^-,k'\to 0^+$ and $K\to\infty, K'\to\pi/2,E\to 1$ and the desired limit equals the limit of $$\frac{\pi K'} {K} \cdot\frac{K} {2\pi^2}\cdot(E-k'^2K)$$ This works out to be $1/4$ if we can show that $k'^2K\to 0$. This is an easy consequence of the asymptotic $$K=\log(4/k')+o(1)$$ as $k\to 1^-$ (for proof see this post).

  • Can you please suggest some reference books for this topic? – PNDas Nov 24 '20 at 07:57
  • 1
    @PNDas: see the last part of one of my answers which lists many references. It also has a reference to my blog where you can search for theta, elliptic, Ramanujan, Jacobi, Gauss, AGM. – Paramanand Singh Nov 24 '20 at 08:21
  • @PNDas: btw this is a difficult topic and requires a hell lot of patience to understand coherently. My blog was a result of my efforts to understand this topic. – Paramanand Singh Nov 24 '20 at 08:27
  • I was thinking about the same. I want to ask , is this covered in standard university course ? (I mean in MSc or PhD course in India). Did you self study ?Where did you learn this? Anyway thank you for your comment. I am your fan from last three years. I have seen your answers, and admired your wisdom. You are one of my inspiration. Also yours and mine first name is almost same . I have decided that as soon as I complete my BSc course I'll try to study advanced integrations topics , special functions. – PNDas Nov 24 '20 at 08:38
  • 1
    @PNDas: thanks for your kind words. I am an engineering graduate (btech in computer science) and work in IT. So I don't have dedicated math education (like MSc, PhD). But I have great interest in mathematics and have learnt some stuff from various online sources (this website being my favorite). – Paramanand Singh Nov 24 '20 at 09:34
  • 1
    @PNDas: wish you best of luck for your higher studies! – Paramanand Singh Nov 24 '20 at 09:35
  • I will read and understand this answer when I'll have some mathematical knowledge. I really appreciate your help though! And thanks for the references! +1 – Peanut Nov 24 '20 at 13:39
  • @Peanut: I myself prefer Sangchul Lee's answer. The advantage of my approach is more when you want to evaluate the sum for specific values of $x$. If $x=\pm e^{-\pi\sqrt{r} }$, $r$ being a positive rational number then the sum can be explicitly evaluated. – Paramanand Singh Nov 24 '20 at 13:42