27

How to prove the following equality ?: \begin{equation} \tag{*}\sum_{k=1}^\infty \frac{(4 k-1) 2^{8 k}}{k^5 (2 k-1) \binom{2 k}{k}^4} = 372 \zeta (5)-28 \pi ^2 \zeta (3)\label{*} \end{equation}


Some background:

  • (\ref{*}) is used to obtain a unique series for $\zeta(5)$ ( last equation here ), which is in turn used to calculate $\zeta(5)$ to extremely high precision (current record is 200 billion digits).
  • (\ref{*}) is a very-well-poised $2$-balanced $_9\!F_8$, so one might attempt Bailey's transformation on it. However, this approach has not yet been fruitful.
  • A proof of (\ref{*}) can be found in Example $\tt XV$ of this article, it is however quite contrived and complicated. I wonder if there exists more elegant proofs.

Any comment or suggestion is welcomed.

Zacky
  • 30,116
pisco
  • 19,748
  • 1
    $$=\int_{0}^{1}\sum_{k=0}^{\infty}\frac{32(x+1)x^{2k}}{\left(z^{k}^{-3/2}\right)^4}dx$$ I'm not sure that I should go this way. – Quý Nhân Dec 20 '24 at 15:04
  • I'm looking at Example XV, but not seeing that exact formula. Are you saying that () can easily be derived from the formulas in Ex. XV? If so, you already know a proof of the equality (), so what precisely are you asking for in this MSE question? Thanks in advance for the clarifications – D.R. Dec 24 '24 at 06:16
  • 1
    @D.R. Thanks for the question. This formula is the displayed equation just before Remark 5.3 of the article. As I think the proof there is complicated, I am asking for a more elegant proof on MSE. I know some users here are experts in producing clever/elegant solution to otherwise tricky integral/series. – pisco Dec 24 '24 at 10:08
  • 1
    Possible reference - https://math.stackexchange.com/a/2130679/1157207 – Amrut Ayan Dec 25 '24 at 05:56
  • 1
    As a suggestion, one might attempt to connect the series with the following integrals: $$\color{blue}{\int_0^1 (2x-1) K^4(x)dx}-\color{red}{\zeta(2)\int_0^1 (2x-1) \frac{K^4\left(\frac{2 \sqrt x}{1+\sqrt x}\right)}{(1+\sqrt x)^2}dx}=\frac{\color{blue}{372\zeta(5)}-\color{red}{28 \pi^2\zeta(3)}}{32}$$ Now after some integral manipulations we can also rewrite the original series as: $$S=16\int_0^1\int_0^1 \left(K^2(\tfrac{1+\sqrt{1-x^2}}{2})-K^2(\tfrac{1-\sqrt{1-x^2}}{2})\right)\frac{\arcsin(xt)}{\sqrt{1-x^2t^2}}dxdt$$ Note that I utilized the Wolfram variant of $K(k)$ not with modulus $k$. – Zacky Dec 26 '24 at 16:56
  • 1
    @Zacky Quite nice integral representation, you're very encouraged to post as a partial solution below. – pisco Dec 26 '24 at 18:23
  • Do you have a proof of $\sum_{n=1}^{\infty} \frac{\left ( 1 \right )_n^4 }{ \left ( \frac32 \right )_n^4 } \frac{4n+1}{n^4} =8(10-\pi^2)?$ – Setness Ramesory Jan 17 '25 at 15:26
  • @SetnessRamesory Probably can be proved in the same way as I how proved this series in the linked paper. I don't have time to do the computation though, – pisco Jan 17 '25 at 17:14
  • this should be easy to to and no elliptic integrals needed. Split the sum into two parts. The hypergeometric function of unit augment of 1/2 and 3/2 for the bases is easy to do as it's only a quadratic . All you do is find the seed integral that yields the binomial , inverse sine or arctan, and use integration by parts to 4 times on the $ln (1+x)/(1+bx+cx^2)$ and then you evaluate the resulting polylogarithims. The unit argument means it should be some linear combination of zeta 2,3,4 and 5 – CarP24 Feb 10 '25 at 01:06
  • @CarP24 Can you enlighten all of us by providing the details as an answer? – pisco Feb 10 '25 at 08:56
  • nevermind.my method only works after evaluating the long hypergeometric series of a general augment, and this takes more work. i still think it is doable without elliptic integrals and am working on it .maybe generating function of a 4th order equation which is doable and equate coefficients in closed form – CarP24 Feb 12 '25 at 22:19

2 Answers2

6

This is only meant to give some ideas on how to tackle the series via integrals. I hope that someone can further build up on this and finish the solution at some point. To get a first impression, I'd recommend to peek at $(1)$, $(2)$ and $(3)$ directly.

We will start by converting the series into a quadruple integral by utilizing the following two identities:

$$\frac{\arcsin \left( x\right)}{\sqrt{1- x^2}} =\frac12 \sum_{n=1}^\infty \frac{4^n x^{2n-1}}{n\binom{2n}{n}};\quad \int_0^\frac{\pi}{2} \sin^{2n-1}xdx= \frac12\frac{4^n}{n\binom{2n}{n}}$$

$$\Rightarrow \frac{1}{16}\sum_{n=1}^\infty \left(\frac{4^n}{n\binom{2n}{n}}\right)^4 t^n=\int_0^\frac{\pi}{2}\int_0^\frac{\pi}{2}\int_0^\frac{\pi}{2}\frac{\arcsin(t\sin x\sin y\sin z)}{\sqrt{1-t^2\sin^2 x\sin^2 y \sin^2 z }}dxdydz$$

$$ \Rightarrow \mathcal S=\sum_{n=1}^\infty \frac{(4n-1) 256^n}{(2n-1)n^5 \binom{2n}{n}}=\small{32\int_0^1\int_0^\frac{\pi}{2}\int_0^\frac{\pi}{2}\int_0^\frac{\pi}{2} \frac{(1+t)\arcsin(t\sin x\sin y\sin z)}{t\sqrt{1-t^2\sin^2 x\sin^2 y \sin^2 z }}dxdydzdt}$$

Next, we will substitute $\sin x\to x$, $\sin y\to y$ and $\sin z\to z$ followed by $xyz\to x$ to arrive at:

$$\mathcal S=32\int_0^1 \int_0^1 \int_0^1 \int_0^{yz} \frac{(1+t)\arcsin(tx)}{t\sqrt{1-t^2 x^2}} \frac{dxdydzdt}{\sqrt{y^2 z^2-x^2}\sqrt{1-y^2}\sqrt{1-z^2}}$$

$$=32\int_0^1 \frac{(1+t)\arcsin(tx)}{t\sqrt{1-t^2 x^2}} \underbrace{\int_{x}^1 \int_{x/z}^1 \frac{dydzdxdt}{\sqrt{y^2z^2-x^2}\sqrt{1-y^2}\sqrt{1-z^2}}}_{\mathcal J(x)}$$ $$\small =\boxed{16\int_0^1 \int_0^1 \left(K^2\left(\sqrt{\frac{1+\sqrt{1-x^2}}{2}}\right)-K^2\left(\sqrt{\frac{1-\sqrt{1-x^2}}{2}}\right)\right) \frac{(1+t)\arcsin(tx)}{t\sqrt{1-t^2 x^2}} dxdt} \tag 1$$

Where $K(k)$ is the Complete Elliptic Integral of the First Kind with elliptic modulus $k$. Note that you need to put $K(k)=K(k^2)$ if you want to verify numerically in the Wolfram language.


Note that the $\mathcal J(x)$ integral from above was calculated as follows:

$$\mathcal J(x)=\int_{x}^1 \frac{1}{z\sqrt{1-z^2}}\int_{x/z}^1 \frac{dydz}{\sqrt{y^2-x^2/z^2}\sqrt{1-y^2}}$$ $$\overset{y\to \sqrt{1-(1-x^2/z^2)y^2}}=\int_x^1 \frac{1}{z\sqrt{1-z^2}}\int_0^1 \frac{dydz}{\sqrt{1-y^2}\sqrt{1-(1-x^2/z^2)y^2}}$$

$$=\int_x^1 \frac{K\left(\sqrt{1-\frac{x^2}{z^2}}\right)}{z\sqrt{1-z^2}}dz\overset{\sqrt{1-\frac{x^2}{z^2}}\to z}=\int_0^\sqrt{1-x^2} \frac{zK(z)}{\sqrt{1-z^2}\sqrt{1-x^2-z^2}}dz$$

$$=\Re\int_0^1 \frac{zK(z)}{\sqrt{1-z^2}\sqrt{1-x^2-z^2}}dz \overset{\small \bigstar}= \Re \frac{1}{1+\sqrt{1-x^2}} K^2 \small{\left(\sqrt{\frac{2}{1+\sqrt{1-x^2}}}\right)}$$

$$\small \overset{\small \bigstar \bigstar}=\frac12\left(K^2\left(\sqrt{\frac{1+\sqrt{1-x^2}}{2}}\right)-K^2\left(\sqrt{\frac{1-\sqrt{1-x^2}}{2}}\right)\right)$$

The $\small \bigstar$ result was derived by Yajun Zhou in this article, see page 19 under eq. $(47)$. Whereas the $\small \bigstar \bigstar$ identity follows by squaring $K(1/k)=k(K(k)-iK'(k))$ and taking the real part.


Now, the most intuitively way to move forward I think would be to compute the integral with respect to $t$, which can be done with the substitution $tx\to t$ followed by $t\to \frac{2t}{1+t^2}$, to obtain:

$$\int_0^1 \frac{(1+t)\arcsin(xt)}{t\sqrt{1-x^2t^2}}dt=2\operatorname{Ti}_2\left(\frac{x}{1+\sqrt{1-x^2}}\right)+\frac{\arcsin^2 x}{2x}$$

Where $\operatorname{Ti}_2(x)$ is the inverse tangent integral. This gives yet another integral representation:

$$\small \mathcal S=32\int_0^1 \left(K^2\left(\sqrt{\frac{1+\sqrt{1-x^2}}{2}}\right)-K^2\left(\sqrt{\frac{1-\sqrt{1-x^2}}{2}}\right)\right)\left(\operatorname{Ti}_2\left(\frac{x}{1+\sqrt{1-x^2}}\right)+\frac{\arcsin^2 x}{4x}\right)dx$$

$$\overset{x\to \sin x}=\boxed{32\int_0^\frac{\pi}{2} \left(K^2\left(\cos \frac{x}{2}\right)-K^2\left(\sin \frac{x}{2}\right)\right)\left(\operatorname{Ti}_2\left(\tan \frac{x}{2}\right)+\frac{x^2}{4\sin x}\right)\cos x \, dx} \tag2$$

I was quite optimistic about the above integral, especially since it allows to utilize various series such as $\operatorname{Ti}_2(\tan x)-x\ln(\tan x)=\sum\limits_{n=0}^\infty \frac{\sin((4n+2)x)}{(2n+1)^2}$, but felt short in the end.


An even more promising idea is based on another article due to Yajun Zhou, which proves that:

$$\zeta(5)=-\frac{\pi^4}{372}\int_{-1}^1 x\left(P_{-1/2}(x)\right)^4 dx; \quad \zeta(3) = -\frac{\pi^4}{168}\int_{-1}^1 x\left(P_{-1/4}(x)\right)^4 dx$$

Where $P_n(x)$ is the Legendre function, defined via the Mehler-Dirichlet formula. Now, by rewriting $P_{-1/2}(\cos x)=\frac{2}{\pi} K\left(\sin \frac{x}{2}\right)$ and $P_{-1/4}(\cos x)= \frac{2}{\pi \sqrt{1+\sin \frac{x}{2}}}K\left(\sqrt{\frac{2\sin \frac{x}{2}}{1+\sin \frac{x}{2}}}\right)$ followed by substituting $\frac{1-x}{2}\to x$ we arrive at the following result, which surprisingly is identical with the result of our original series:

$$\boxed{32\int_0^1 (2x-1) K^4(\sqrt x)dx-32\zeta(2)\int_0^1 \frac{2x-1}{{(1+\sqrt x)^2}} \small{K^4\left(\sqrt{\frac{2 \sqrt x}{1+\sqrt x}}\right)}dx=372\zeta(5)-28 \pi^2\zeta(3)}$$

As such, it is reasonable to think that we can actually go ahead and connect our previous double integral seen in $(1)$ with the above result. Namely, to arrive at squared elliptic integrals in $(1)$ we integrated twice after essentially substituting $xy \to x$ and $xz\to x$, similarly we can expect that if we further substitute $xt \to x$ we will obtain third powers, and so on for fourth powers.

Anyway, continuing from $(1)$ we have:

$$\small \mathcal S =16\int_0^1 \int_0^1 \left(K^2\left(\sqrt{\frac{1+\sqrt{1-x^2}}{2}}\right)-K^2\left(\sqrt{\frac{1-\sqrt{1-x^2}}{2}}\right)\right) \frac{(1+t)\arcsin(tx)}{t\sqrt{1-t^2 x^2}} dxdt$$

$$\small \overset{xt \to x}=16\int_0^1 \frac{\arcsin x}{\sqrt{1-x^2}}\int_x^1 \left(\frac{1}{t}+\frac{1}{t^2}\right) \left(K^2\left(\sqrt{\frac{1+\sqrt{1-x^2/t^2}}{2}}\right)-K^2\left(\sqrt{\frac{1-\sqrt{1-x^2/t^2}}{2}}\right)\right) dtdx$$

$$\small \overset{\sqrt{1-x^2/t^2}\to t}=16\int_0^1 \frac{\arcsin x}{\sqrt{1-x^2}} \int_0^\sqrt{1-x^2} \left(\frac{t}{1-t^2}+\frac{t}{x\sqrt{1-t^2}}\right)\left(K^2\left(\sqrt{\frac{1+t}{2}}\right)-K^2\left(\sqrt{\frac{1-t}{2}}\right)\right) dtdx$$

$$\overset{\large {x\to \sin x \atop \large t\to\cos t}}=\boxed{16\int_0^\frac{\pi}{2} x \int_x^\frac{\pi}{2} \left(\frac{\cos t}{\sin t}+\frac{\cos t}{\sin \color{red}{x}}\right)\left(K^2\left(\cos \frac{t}{2}\right)-K^2\left(\sin \frac{t}{2}\right)\right) dtdx} \tag 3$$

Unfortunately, here I hit another dead end as I'm not sure how to further calculate the above $t$ integral, which I suspect that can be expressed as a sum of cubic elliptic integrals.

Zacky
  • 30,116
  • 1
    Maybe it's more direct to use $$\int_{0}^{1} x^n\left [ K\left ( \sqrt{\frac{1+\sqrt{1-x} }{2} } \right )^2
    • K\left ( \sqrt{\frac{1-\sqrt{1-x} }{2} } \right )^2 \right ]

    \text{d} x=\frac{4\cdot(1)_n^3}{\left ( 2n+1 \right )^3\left ( \frac12 \right )_n^3}.$$

    – Setness Ramesory Dec 29 '24 at 07:48
  • @SetnessRamesory interesting result, I was not aware of it. – Zacky Dec 29 '24 at 12:20
0

It seems correct.

See the Mathematica code.

Sum[(4 k - 1)*2^(8 k)/(k^5*(2 k - 1)*Binomial[2 k, k]^4), {k, 1, 
  Infinity}]

$$ 16 \left(\, _6F_5\left(1,1,1,1,1,1;\frac{3}{2},\frac{3}{2},\frac{3}{2},\frac{3}{2},2;1\right)+2 \, _6F_5\left(\frac{1}{2},1,1,1,1,1;\frac{3}{2},\frac{3}{2},\frac{3}{2},\frac{3}{2},\frac{3}{2};1\right)\right) $$

372 Zeta[5] - 28 Pi^2*Zeta[3] - % // N