30
$\begingroup$

Can we possibly compute the following integral in terms of known constants?

$$\int_0^1 \int_0^1 \frac{dx\,dy}{1-xy(1-x)(1-y)}$$ Some progress was already done here http://integralsandseries.prophpbb.com/topic279.html but still we have a hypergeometric function. What's your thoughts on it?

UPDATE: The question was also posted on mathoverflow here https://mathoverflow.net/questions/206253/calculate-in-closed-form-int-01-int-01-fracdx-dy1-xy1-x1-y#

$\endgroup$
  • $\begingroup$ Plouffe's Inverse Symbolic Calculator doesn't seem to find any evidence that there's a better representation for the answer than as a hypergeometric value. Though that in itself is nothing conclusive. $\endgroup$ – Igor Khavkine May 11 '15 at 16:17
32
+500
$\begingroup$

Let $x=\sin^2(t)$ and $y=\sin^2(w)$, we obtain \begin{align} I & = \int_0^{\pi/2}\int_0^{\pi/2} \dfrac{4\sin(t)\cos(t)\sin(w)\cos(w)dtdw}{1-\sin^2(t)\cos^2(t)\sin^2(w)\cos^2(w)}\\ & = 4\sum_{k=0}^{\infty}\int_0^{\pi/2}\int_0^{\pi/2} \sin^{(2k+1)}(t) \cos^{(2k+1)}(t)\sin^{(2k+1)}(w) \cos^{(2k+1)}(w)dtdw\\ & = 4 \sum_{k=0}^{\infty} \left(\int_0^{\pi/2}\sin^{(2k+1)}(t) \cos^{(2k+1)}(t)dt\right)^2\\ & = 4 \sum_{k=0}^{\infty} \dfrac1{4^{2k+1}}\left(\int_0^{\pi/2}\sin^{(2k+1)}(2t)dt\right)^2 = \sum_{k=0}^{\infty} \dfrac1{4^{2k}}\left(\int_0^{\pi}\sin^{(2k+1)}(t)\dfrac{dt}2\right)^2\\ & = \sum_{k=0}^{\infty} \dfrac1{4^{2k}}\left(\int_0^{\pi/2}\sin^{(2k+1)}(t)dt\right)^2 \end{align} Recall that $$\int_0^{\pi/2}\sin^{(2k+1)}(t)dt=\dfrac{4^k}{2k+1}\dfrac1{\dbinom{2k}k}$$ This gives us $$I = \sum_{k=0}^{\infty} \dfrac1{(2k+1)^2} \dfrac1{\dbinom{2k}k^2} = _3F_2\left(1,1,1;\frac32,\frac32;\frac1{16}\right)$$

$\endgroup$
  • $\begingroup$ Wolfram Alpha evaluates the hypergeometric in the last line to: 1.0289425662942443315755244192977092068490... , which cannot be identified by the Inverse Symbolic Calculator even in "advanced" mode. So a more "closed" solution than the given seems to be unlikely. $\endgroup$ – Johannes Trost May 11 '15 at 13:06
  • 2
    $\begingroup$ @JohannesTrost the fact that the inverse symbolic calculator in advanced mode doesn't find anything means almost nothing (I tell you that from my experience). $\endgroup$ – user 1591719 May 11 '15 at 13:56
17
$\begingroup$

An incomplete answer.

Tackling the inner integral with elementary methods, \begin{align} \int^1_0\frac{{\rm d}x}{1-ax+ax^2} &=\int^1_0\frac{{\rm d}x}{a\left(x-\frac{1}{2}\right)^2+1-\frac{a}{4}}\\&=\int^\frac{1}{2}_{-\frac{1}{2}}\frac{{\rm d}x}{ax^2+1-\frac{a}{4}}\\ &=\frac{1}{\sqrt{a\left(1-\frac{a}{4}\right)}}\int^\frac{\sqrt{a}}{2\sqrt{(1-\frac{a}{4})}}_{-\frac{\sqrt{a}}{2\sqrt{(1-\frac{a}{4})}}}\frac{{\rm d}x}{1+x^2}\\&=\frac{4}{\sqrt{a(4-a)}}\arctan\left(\sqrt{\frac{a}{4-a}}\right) \end{align} Thus \begin{align} \iint\limits_{[0,1]^2}\frac{{\rm d}A}{1-xy(1-x)(1-y)} &=4\int^1_0\arctan\left(\sqrt{\frac{x(1-x)}{x^2-x+4}}\right)\frac{{\rm d}x}{\sqrt{x(1-x)(x^2-x+4)}}\\ &=16\int^1_0\arctan\left(\sqrt{\frac{1-x^2}{15+x^2}}\right)\frac{{\rm d}x}{\sqrt{(1-x^2)(15+x^2)}}\\ &=\frac{16}{\sqrt{15}}\int^1_0\operatorname{F}\left(\arcsin{x}\right)\frac{x\ {\rm d}x}{\sqrt{(1-x^2)(15+x^2)}}\\ &=\frac{8}{15}\int^{\operatorname{K}}_{-\operatorname{K}}u\operatorname{sn}(u)\ {\rm d}u \end{align} where $\operatorname{sn}(u)$ is a Jacobi elliptic function with modulus $k=15^{-1/2}i$.

Let $R$ be the parallelogram with vertices $-\operatorname{K}-2i\operatorname{K}'$, $\operatorname{K}-2i\operatorname{K}'$, $\operatorname{K}+2i\operatorname{K}'$, $-\operatorname{K}+2i\operatorname{K}'$. Parameterising along the contour, \begin{align} \int_{\partial R}z^{\color{red}{2}}\operatorname{sn}(z)\ {\rm d}z &=-8i\operatorname{K}\left(\frac{4}{\sqrt{15}}\right)\int^{\operatorname{K}}_{-\operatorname{K}}u\operatorname{sn}(u)\ {\rm d}u+2i\int^{2\operatorname{K}'}_{-2\operatorname{K}'}(\operatorname{K}^2-t^2)\operatorname{cd}(it)\ {\rm d}t\\ &=2\pi i\sum_{\pm}\operatorname*{Res}_{z=\pm i\operatorname{K}'}z^2\operatorname{sn}(z)=-4\pi\sqrt{15}\operatorname{K}\left(\frac{4}{\sqrt{15}}\right)^2 \end{align} $\operatorname{cd}(t)$ has an anti-derivative, namely $$2i\operatorname{K}^2\int^{2\operatorname{K}'}_{-2\operatorname{K}'}\operatorname{cd}(it)\ {\rm d}t=-4i\sqrt{15}\operatorname{K}^2\left.\ln\left(\operatorname{nd}(t)+\frac{i}{\sqrt{15}}\operatorname{sd}(t)\right)\right|^{2i\operatorname{K}'}_0$$ Otherwise, this seems to be as far as I can go for now.

$\endgroup$
  • 1
    $\begingroup$ (+1) Well done. Now we just need to find the bravery to compute $I'(\alpha)$ :) $\endgroup$ – Jack D'Aurizio May 4 '15 at 12:00
  • 3
    $\begingroup$ (+1) A very good starting point! This is the way to go (at least for the first part of the way). However it might be an illusion that the last integral is easier than the initial one. :-) $\endgroup$ – user 1591719 May 4 '15 at 12:31
  • 1
    $\begingroup$ Thank you very much for your work to my question. $\endgroup$ – user 1591719 May 13 '15 at 19:29
14
$\begingroup$

From hypergeometric function to elliptic integrals

The hypergeometric function $_3F_2$ from your link (or user17762's answer) may be rewritten (using following integral representation and $_2F_1\left(1,1;\frac32; t\right)=\frac d{dt}\left[\left(\arcsin\sqrt{t}\,\right)^2\right]$) as : \begin{align} \tag{1}I&=\;_3F_2\left(1,1,1;\frac32,\frac32;\frac1{16}\right)=\frac 12\int_0^1\frac{_2F_1\left(1,1;\frac32;\frac x{16}\right)}{\sqrt{1-x}}dx\\ &=2\int_0^1\frac{\arcsin\frac{\sqrt{x}}4}{\sqrt{x\,(1-x)\left(1-\frac x{16}\right)}}dx\\ \tag{2}&=16\int_0^{1/4}\frac{\arcsin u}{\sqrt{(1-16 u^2)\left(1-u^2\right)}}\,du\\ &\tag{3}=16\int_0^{\arcsin(1/4)} \frac {y}{\sqrt{1-4^2\sin(y)^2)}}\;dy\\ &=\left.16\;y\;F\left(y, 4\right)\right|_0^{\arcsin(1/4)}-16\int_0^{\arcsin(1/4)}\;F\left(y, 4\right)\;dy\\ \tag{4}&=4\;\arcsin(1/4)\;K\left(\frac 1{4}\right)-16\int_0^{\arcsin(1/4)}\;F\left(y, 4\right)\;dy\\ \end{align}

With $\;\displaystyle F(\phi,k):=\int_0^{\phi} \frac {dy}{\sqrt{1-k^2\sin(y)^2)}},\;K(k):=F\left(\frac{\pi}2,k\right)\;$ the incomplete and complete elliptic integral of the first kind $F$ and $K$ (one should take care of replacing $F(\phi,k)$ by EllipticF[$\phi,k^2$] and $K(k)$ by EllipticK[$k^2$] while using Alpha/Mathematica).

'$I$' may be expressed with elliptic integrals but this doesn't seem to help for a closed form... Observing that $\;\displaystyle K\left(\frac 1{4}\right)=\frac{\pi}2\sum_{k=0}^\infty\frac{\binom{2k}{k}^2}{2^{8\,k}}\;$ let's start it all again using the

Squared central binomial series

We want $\;\displaystyle I=\sum_{k=0}^\infty \frac 1{(2k+1)^2\binom{2k}{k}^2}\;$ but Nected Batir proved in 2004 following more general formula (for $n\in\mathbb{N},\; n\ge3$) :

\begin{align} \tag{5}I_n(x)&:=\sum_{k=0}^\infty \frac{x^{2k+1}}{(2k+1)^n\,\binom{2k}{k}^2}\\ &=\frac{4\,(-1)^{n-3}}{(n-3)!}\int_0^{\pi/2}\int_0^{\arcsin(x\cos(y)/4)}\frac t{\sin(t)}\;\log^{n-3}\left(\frac{4\sin(t)}{x\cos(y)}\right)\;dt\,dy\\ \end{align} For $\,n=3\,$ this is simply $\;\displaystyle I_3(x)=\int_0^{\pi/2}\int_0^{\arcsin(x\cos(y)/4)}\frac t{\sin(t)}\;dt\,dy\;\;$ and after derivation \begin{align} I_2(x)&=x\,I_3(x)'=4\,x\int_0^{\pi/2} \frac {\arcsin(x\cos(y)/4)}{x\cos(y)/4} \frac {\partial}{\partial x}\left[\arcsin(x\cos(y)/4)\right]\;dy\\ \tag{6}I_2(x)&=4\int_0^{\pi/2} \frac {\arcsin(x\cos(y)/4)}{\sqrt{1-(x\cos(y)/4)^2}} \;dy\\ \end{align} Unfortunately this is merely $(2)$ in the case $x=1$ (the integral for $x=4$ is divergent and $I_3(4)=8\pi G-14\zeta(3)\;$ but this won't help here...) so nothing complete here either...

The journey may be of interest anyway...

$\endgroup$
  • 1
    $\begingroup$ Thank you for giving insights on the problem (+1). Very nice the formula by Nected Batir I wasn't aware of. $\endgroup$ – user 1591719 May 4 '15 at 23:27
  • 1
    $\begingroup$ Thanks @Chris'ssis and interesting question btw! Is this question linked to another of your quests? Excellent continuation, $\endgroup$ – Raymond Manzoni May 5 '15 at 15:41
  • 1
    $\begingroup$ @ Raymond Manzoni Thank you! :-) Indeed, I'm deeply involved in personal research that covers such integrals, series and limits. $\endgroup$ – user 1591719 May 5 '15 at 16:40
12
$\begingroup$

Our integral equals: $$\begin{eqnarray*} I=\sum_{n\geq 0}\iint_{[0,1]^2}x^n y^n(1-x)^n (1-y)^n\,dx\,dy &=& \sum_{n\geq 0}\left(\int_{0}^{1}x^n(1-x)^n\,dx\right)^2\\&=&\sum_{n\geq 0}\left(\frac{\Gamma(n+1)^2}{\Gamma(2n+2)}\right)^2\\&=&\sum_{n\geq 0}\frac{1}{(2n+1)^2\binom{2n}{n}^2}\tag{1}\end{eqnarray*}$$ and: $$ \sum_{n\geq 0}\frac{x^n}{(2n+1)\binom{2n}{n}}=\frac{4\arcsin\frac{\sqrt{x}}{2}}{\sqrt{x(4-x)}}\tag{2}$$ hence $I$ is the squared $L^2$-norm of a quite complicated function: $$ I = \frac{16}{\pi}\int_{0}^{\pi}\frac{\arcsin\frac{e^{it/2}}{2}\cdot\arcsin\frac{e^{-it/2}}{2}}{\sqrt{(4-e^{it})(4-e^{-it})}}\,dt.\tag{3}$$

$\endgroup$
  • $\begingroup$ @Chris'ssis Good to see you back on the main site. I remember you were taking a break from the site. $\endgroup$ – Adhvaitha May 2 '15 at 13:26
  • $\begingroup$ @user17762 hehe, indeed :-). I actually never take a break from doing math. What was your username in the past? I might recollect who you are. :-) $\endgroup$ – user 1591719 May 2 '15 at 13:29
  • $\begingroup$ @Chris'ssis marvis in my earlier avatar. $\endgroup$ – Adhvaitha May 2 '15 at 14:09
  • $\begingroup$ @user17762 long time I haven't seen you around. Nice to meet you again! :-) $\endgroup$ – user 1591719 May 2 '15 at 14:41
  • 1
    $\begingroup$ I've checked your (1) by hand and found it to be correct. Meanwhile M.N.C.E. has changed his answer. $\endgroup$ – Han de Bruijn May 10 '15 at 19:30

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.