6
$\begingroup$

I recently saw this question ... https://math.stackexchange.com/questions/2393668/double-factorial-sum-k-0-infty-left-frac-2k-1-2k-right#2393668 ... & I am unable to show it.

The result $$1-\left( \frac{1}{2} \right)^{3}+\left( \frac{1\cdot 3}{2\cdot 4} \right)^{3}-\left( \frac{1\cdot 3\cdot 5}{2\cdot 4\cdot 6} \right)^{3}+\left( \frac{1\cdot 3\cdot 5\cdot 7}{2\cdot 4\cdot 6\cdot 8} \right)^{3}-...=\frac{\Gamma ^{2}\left( \frac{9}{8} \right)}{\Gamma ^{2}\left( \frac{7}{8} \right)\cdot \Gamma ^{2}\left( \frac{10}{8} \right)}$$

Now I know several results relating to those ratios of double factorials \begin{eqnarray*} \int_0^{\frac{\pi}{2}} \sin^{2n} \theta d \theta= \frac{ \pi}{2} \frac{(2n-1)!!}{(2n)!!} \tag{1} \\ \sum_{n=0}^{\infty} \frac{(2n-1)!!}{(2n)!!} y^n = \frac{1}{\sqrt{1-y}} \tag{2} \end{eqnarray*} From these two results it is reasonably easy to derive the series for the elliptic integral of the first kind \begin{eqnarray*} K(k)=\int_0^{\frac{\pi}{2}} \frac{d \theta}{\sqrt{1-k^2 \sin^{2}(\theta)}} =\frac{\pi}{2} \left(1+ \left(\frac{1}{2}\right)^2+ \left(\frac{1.3}{2.4}\right)^2+ \cdots \right) \end{eqnarray*} & it is well known that this can be evaluated for special values of $k$ in terms of Gamma functions whose arguements are rational values. See this question Can $\Gamma(1/5)$ be written in this form? and the reference cited there. So my first thought is to use $(1)$ three times, sum the geometric plum & we have \begin{eqnarray*} \int_{0}^{\frac{\pi}{2}} \int_{0}^{\frac{\pi}{2}} \int_{0}^{\frac{\pi}{2}} \frac{d \alpha d \beta d \gamma}{1+ \sin^2 \alpha \sin^2 \beta \sin^2 \gamma} \end{eqnarray*} This triple integral looks difficult so ...

Second thoughts ... use $(1)$ twice & then use $(2)$ to get the double integral \begin{eqnarray*} \int_{0}^{\frac{\pi}{2}} \int_{0}^{\frac{\pi}{2}} \frac{d \alpha d \beta }{\sqrt{1+ \sin^2 \alpha \sin^2 \beta }} \end{eqnarray*} now substitute $ \sqrt{s} =\sin \alpha$ and $ \sqrt{t} =\sin \beta$ (might have lost a factor of $4$) \begin{eqnarray*} \int_{0}^{1} \int_{0}^{1} \frac{d s d t }{\sqrt{s(1-s)t(1-t)(1+st) }} \end{eqnarray*} & I am not sure what to do with this.

In both of these attempts I feel I have taken a wrong turn.

Can someone either give me a Big hint or a reference to the original derivation of this result or a reasonably complete solution ?

Bonus question ... why was it stated with $\frac{10}{8}$ instead of $\frac{5}{4}$ ?

The result $$1-\left( \frac{1}{2} \right)^{3}+\left( \frac{1\cdot 3}{2\cdot 4} \right)^{3}-\left( \frac{1\cdot 3\cdot 5}{2\cdot 4\cdot 6} \right)^{3}+\left( \frac{1\cdot 3\cdot 5\cdot 7}{2\cdot 4\cdot 6\cdot 8} \right)^{3}-...=\frac{\Gamma ^{2}\left( \frac{9}{8} \right)}{\Gamma ^{2}\left( \frac{7}{8} \right)\cdot \Gamma ^{2}\left( \color{red}{\frac{5}{4}} \right)}$$

Hint from Jack D'Auirzio : look at equation (6) here http://mathworld.wolfram.com/CompleteEllipticIntegraloftheFirstKind.html and use $2kk'=i$ where $k'$ is the complementary modulus.

$\endgroup$
  • 1
    $\begingroup$ Extended hint: once $2kk'=i$ is solved, it is enough to convert the LHS of $(6)$ into a value of Euler's Beta function. And if someone wonders about how to prove $(6)$, that can be shown through different techniques: differential equations, Landen's transforms for hypergeometric functions, relations between elliptic integrals of the first kind and the AGM. $\endgroup$ – Jack D'Aurizio Aug 14 '17 at 23:20
  • 1
    $\begingroup$ They probably used $10/8$ to keep the pattern of $n/8$ going in those Gamma functions. $\endgroup$ – Simply Beautiful Art Aug 14 '17 at 23:38
  • $\begingroup$ Related: math.stackexchange.com/a/93878/44121 $\endgroup$ – Jack D'Aurizio Aug 15 '17 at 0:41
  • $\begingroup$ +1 for adding all that context to a poorly framed question. $\endgroup$ – Paramanand Singh Aug 15 '17 at 2:06
  • 1
    $\begingroup$ $$\sum_{k=0}^{\infty} \frac{(2n-1)!!}{(2n)!!} y^n = \frac{1}{\sqrt{1-y}}$$, Shouldn't it be $n = 0$ as the index ? $\endgroup$ – A---B Sep 23 '17 at 12:23
5
$\begingroup$

The sum evaluates to $(1-k^{2})(2K(k)/\pi)^{2}$ for $k=\sqrt{2}-1$. The value of $K$ is obtained from here. Using the value of $K$ we can get the desired closed form sum of the given series.


Using hypergeometric transformations we can prove the formula $$\left(\frac{2K(k)}{\pi}\right)^{2}=\sum_{n=0}^{\infty}\left(\frac{1\cdot 3\cdot 5\cdots (2n-1)}{2\cdot 4\cdot 6\cdots (2n)}\right)^{3}(2kk')^{2n}\tag{1}$$ (see complete proof here). Note that the above equation can be thought of as an identity involving the nome $q=e^{-\pi K'/K} $ also and then we can switch from $q$ to $-q$. In terms of Jacobi's theta functions we have $2K/\pi=\vartheta_{3}^{2}(q)$ and replacing $q$ by $-q$ leads us to $$\vartheta_{3}^{2}(-q)=\vartheta_{4}^{2}(q)=\frac{\vartheta_{4}^{2}(q)}{\vartheta_{3}^{2}(q)}\cdot\vartheta_{3}^{2}(q)=k'\cdot\frac{2K}{\pi}$$ Similarly we have $$(2kk')^{2}=4k^{2}k'^{2}=4\cdot\frac{\vartheta_{2}^{4}(q)\vartheta_{4}^{4}(q)}{\vartheta_{3}^{8}(q)}=4\cdot\frac{16q\psi^{4}(q^{2})\vartheta_{4}^{4}(q)}{\vartheta_{3}^{8}(q)}$$ where $\psi$ is one of Ramanujan's theta functions. Now changing $q$ into $-q$ interchanges $\vartheta_{3},\vartheta_{4}$ and hence the above expression becomes $$-4\cdot\frac{16q\psi^{4}(q^{2})\vartheta_{3}^{4}(q)}{\vartheta_{4}^{8}(q)} =-4\cdot\frac{\vartheta_{2}^{4}(q)/\vartheta_{3}^{4}(q)} {\vartheta_{4}^{8}(q)/\vartheta_{3}^{8}(q)}=-\frac{4k^{2}}{k'^{4}} $$ So we have the formula $$k'^{2}\left(\frac{2K}{\pi}\right)^{2}=\sum_{n=0}^{\infty}(-1)^{n}\left(\frac{1\cdot 3\cdots (2n-1)}{2\cdot 4\cdots(2n)}\right)^{3}(2k/k'^{2})^{2n}\tag{2}$$ And putting $k=\sqrt{2}-1$ we get the desired sum.

As indicated in a comment from Jack d'Aurizio the sum in question can also be expressed in terms of arithmetic-geometric-mean as $$\frac{2}{\operatorname {AGM} (\sqrt{2},\sqrt{1+\sqrt{2}})^{2}}\tag{3}$$ Writing in this manner helps to evaluate the sum very accurately with a few iterations of the AGM.

It is also interesting to know that Ramanujan used the series $(1)$ to obtain $$\frac{4}{\pi}=\sum_{n=0}^{\infty}\left(\frac{1\cdot 3\cdots (2n-1)}{2\cdot 4\cdots (2n)}\right)^{3}\cdot\frac{6n+1}{4^{n}}\tag{4}$$ and $$\frac{2}{\pi}=\sum_{n=0}^{\infty}(-1)^{n}(4n+1)\left(\frac{1\cdot 3\cdots (2n-1)}{2\cdot 4\cdots (2n)}\right)^{3}\tag{5}$$ was obtained from series $(2)$. You can get all the details here.


The value of $K(\sqrt{2}-1)$ has been obtained via non-obvious hypergeometric transformations and it would be nice to have an approach based on direct evaluation of integral involved.

$\endgroup$
  • $\begingroup$ (+1) Beautiful answer, I was waiting for Clausen's formula to appear. It might be an interesting addendum to mention that the origin series equals $$ \frac{2}{\text{AGM}(\sqrt{2},\sqrt{1+\sqrt{2}})^2}.$$ $\endgroup$ – Jack D'Aurizio Aug 15 '17 at 2:12
  • 1
    $\begingroup$ @JackD'Aurizio: I will add your comment in my answer (because at times people might feel a bit lazy to read comments). And thanks for your kind words. $\endgroup$ – Paramanand Singh Aug 15 '17 at 3:08
  • 1
    $\begingroup$ \begin{eqnarray*} \Gamma \left( \frac{1}{8} \right) =\pi ^{\frac{1}{8}} 2^{\frac{17}{8}} K \left( \frac{1}{\sqrt{2}} \right)^{\frac{1}{4}} K \left( \sqrt{2}-1 \right)^{\frac{1}{2}} \end{eqnarray*} I will read the two references in your post. I think I can piece it together from here. Thank you very much indeed for your efforts. $\endgroup$ – Donald Splutterwit Aug 15 '17 at 20:57
  • 1
    $\begingroup$ @JackD'Aurizio Thank you very much for helpful hints. $\endgroup$ – Donald Splutterwit Aug 15 '17 at 20:58

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.