# Finding $\int^{1}_{0}\frac{\ln^2(x)}{\sqrt{4-x^2}}dx$

Finding $$\int^{1}_{0}\frac{\ln^2(x)}{\sqrt{4-x^2}}dx$$

Try: Let $$I=\frac{1}{2}\int^{1}_{0}\frac{\ln^2(x)}{\sqrt{1-\frac{x^2}{4}}}=\frac{1}{2}\int^{1}_{0}\sum^{\infty}_{n=0}\binom{-1/2}{n}\bigg(-\frac{1}{4}\bigg)^nx^{2n}\ln^2(x)dx$$

$$I=\frac{1}{2}\sum^{\infty}_{n=0}\binom{-1/2}{n}\bigg(-\frac{1}{4}\bigg)^n\int^{1}_{0}x^{2n}\ln^2(x)dx$$

Using By parts , We have

$$I=\frac{1}{2}\sum^{\infty}_{n=0}\binom{-1/2}{n}\bigg(-\frac{1}{4}\bigg)^n\frac{2}{(2n+1)^3}$$

But answer given as $\displaystyle\frac{7\pi^3}{216}$

I am not understand How can i get it. could some help me , Thanks

• But $\ln ^2(x)$ blows up at $x=0$... – glowstonetrees May 13 '18 at 2:46
• If you let $x=2\sin u$, then the denominator is cleared. – Frank W. May 13 '18 at 2:54
• Answered here. – metamorphy May 16 '18 at 4:02
• Nice metamorphy. can you explain How $$\sum^{\infty}_{i=0}\frac{4^{-i}\Gamma(1/2+i)}{(1+2i)^3\sqrt{\pi}i!}$$ – DXT May 16 '18 at 6:37
• To be clear about what you are asking, you already have $I$ as a hypergeometric series. Are you asking how to evaluate it to $7(\pi/6)^3$? A computer algebra system can sum the series in closed form without giving any hint of how to do it "by hand". – Somos May 19 '18 at 4:39

A sketch (this approach is generalizable to arbitrary even powers of the logarithm):

We use the standardbranch of $\log$ throughout

$$\Re \log^2(1-e^{2 ix})=\log^2(2 \sin(x))-(x-\frac{\pi}{2})^2 \quad (\star)$$

Integrating the LHS over a large rectangle in the upper half plane with verticies $\{(0,0),(\frac{\pi}{6},0),(\frac{\pi}{6},i\infty),(0,i\infty)\}$ we obtain (the integral over the imaginary line cancels, since we are only interested in real components)

$$\Re\int_0^{\pi/6}\log^2(1-e^{2 ix})=-\Im\int_0^{\infty}\log^2(1-ae^{-2x})=-\Im(G(a))$$

where $a=e^{-i \pi/3}$. Substituting $e^{-2x}=q$ we get $$2 G(a)=\int_0^1 \frac{\log^2(1-a q)}{q}$$

using repeated integration by parts we get ($\text{Li}_n(z)$ denotes the Polylogarithm of order $n$)

$$2 G(a)=-2\text{Li}_3(1-a)+2\text{Li}_2(1-a)\log(1-a)+\log(1-a)^2\log(a)+2\text{Li}_3(1)$$

adding some Polylogarithm wizardy we find that (see Appendix)

$$\Im(G(a))=-\color{blue}{\frac{\pi^3}{324}}$$

using furthermore the trivial integral

$$\int_0^{\pi/6}dx(x-\frac{\pi}{2})^2=\color{green}{\frac{19 \pi^3}{648}}$$

we find indeed

$$\int_0^{\pi/6}dx\log^2(2 \sin(x))=\\ \color{green}{\frac{19 \pi^3}{648}}+\color{blue}{\frac{\pi^3}{324}}=\frac{7\pi^3}{216}$$

which is equivalent to the claim in question

Following OP we might rewrite the integral as an infinte sum, which gives us the hardly to believe corollary

$$\sum_{n\geq 0}\frac{1}{16^n(2n+1)^3}\binom{2n}{n}=\frac{7\pi^3}{216}$$

Appendix

The functional equations of the Dilogarithm immediately delievers

$$\Re\text{Li}_2(1-a)= \frac{1}{2}(\text{Li}_2(1-a)+\text{Li}_2(1-a^{-1})\\=-\frac{1}{4}\log^2(a)=\frac{\pi^2}{36}$$

The Trilogarithm part is a bit trickier,

$\Im\text{Li}_3(1-z)=\Im\int_0^{1-z}\text{Li}_2(x)/x=-\Im\text{Li}_3(z)$ together with the inversion formula $\text{Li}_3(-z)-\text{Li}_3(-1/z)=-\frac{1}{6}(\log^3(z)+\pi^2\log(z))$gives us that

$$\Im\text{Li}_3(1-a)=\frac{5\pi^3}{162}$$

since $\log(1-a)=\frac{i\pi}{3}$ we find

$$2\Im(G(a))=-2\frac{5\pi^3}{162}+2\frac{\pi^2}{36}\frac{\pi}{3}+\frac{\pi^2}9\frac{\pi}3$$

or $$\Im(G(a))=-\frac{\pi^3}{324}$$

• I love how you color coded your individual answer for each integral! – Frank W. May 21 '18 at 18:24
• Can you help me understand why$$\Re\int\limits_0^{\pi/6}dx\,\log^2(1-e^{2ix})=-\Im\int\limits_0^{\infty}dx\,\log(1+ae^{-2x})$$? I drew a rectangular contour with vertices $(0,0)$, $(\pi/6,0)$, $(\pi/6, R)$, and $(0,R)$. However, as $R\to\infty$, I only get$$\int\limits_0^{\pi/6}dx\,\log^2(1-e^{2ix})=i\int\limits_0^{\infty}dx\,\log^2(1-e^{-2x})-i\int\limits_0^{\infty}dx\,\log^2(1-e^{\pi i/3-2x})$$I don't see where the $+$ sign comes in and how you simplify the right-hand side. And how to you get $\log^2(1-aq)$, shouldn't it be $\log^2(1+aq)$ on the next line? I am so confused... – Frank W. May 29 '18 at 19:31
• the first integral on the rhs can be neglected since we are looking for real contributions and the integrand itself is clearly real. The $+$ you are talking about should be a $-$. Sry for that stupid mistake! – tired May 31 '18 at 19:50
• Okay, that make sense. But the $a$ still doesn’t match. For me, $a=e^{\pi i/3}$ while you let $a=e^{-\pi i/3}$... >:( – Frank W. May 31 '18 at 20:02
• @FrankW. i have to go to bed. i will come back to ypur question soon. Essentially it shouldn't matter much if we take $a$ or its conjugate – tired May 31 '18 at 22:41

We will integrate $\,f_p(x) := \ln^2(x)/\sqrt{p-x^2}\,$ using hypergeometric functions. Let $\,y:=x^2/p\,,\,$ and $\,h_0(x) := \!_1F_0(\frac12;;y) = 1/\sqrt{1-y},\,$ $\,h_1(x) := \!_2F_1(\frac12,\frac12;\frac32;y),\,$ $\,h_2(x) := \!_3F_2(\frac12,\frac12,\frac12;\frac32,\frac32;y),\,$ and $\,h_3(x) := \!_4F_3(\frac12,\frac12,\frac12,\frac12;\frac32,\frac32,\frac32;y).\,$ Note that $\,x h_n^{'}(x) = h_{n-1}(x)-h_n(x)\,$ if $\,n>0.\,$ Let $\,g_p(x) := \int f_p(x)\,dx = \sqrt{y}\,(2h_3(x) -2\ln(x)h_2(x) +\ln(x)^2h_1(x)).\,$ Now $\,g_p(0) = 0,\,$ and thus $\, I:=\int_0^1 f_p(x)\,dx = g_p(1) = 2h_3(1/p)/\sqrt{p}.\,$ In our case $\,p=4\,$ and $\,I = h_3(1/4) = 7\pi^3/216.\,$

• I would love to see how you found that hypergeometric sum "by hand". To me, it is highly non-trivial. – omegadot Feb 14 at 8:59

I arrive at another result, where is the error?

$$I = \int^{1}_{0}\frac{\ln^2(x)}{\sqrt{4-x^2}}dx$$

Proof :

\begin{align} \int_0^1 \frac{\ln^2 x}{\sqrt{4-x^2}}\,\mathrm dx &=\int_0^1 \frac{\ln^2 x}{2\sqrt{1-\left(\frac{x}{2}\right)^2}}\,\mathrm dx\\[7pt] &=\int_0^{1/2} \frac{\ln^2 (2t)}{2\sqrt{1-t^2}}2\,\mathrm dt=\int_0^{1/2} \frac{\left(\ln(2) + \log(t)\right)^2}{\sqrt{1-t^2}}\,\mathrm dt\tag{1}\\[7pt] &= \int_0^{1/2} \frac{\ln^2(t) + 2 \ln(2)\ln(t) + \ln^2(2)}{\sqrt{1-t^2}}\,\mathrm dt\\[7pt] &= \underbrace{\int_0^{1/2} \frac{\ln^2(t)}{\sqrt{1-t^2}}\,\mathrm dt}_{I_1} + \underbrace{\int_0^{1/2} \frac{2\ln(2)\ln(t)}{\sqrt{1-t^2}}\,\mathrm dt}_{I_2} + \underbrace{\int_0^{1/2} \frac{\ln^2(2)}{\sqrt{1-t^2}}\,\mathrm dt}_{I_3}\\[7pt] \\ I_3&= \ln^22\int_0^{1/2} \frac{1}{\sqrt{1-t^2}}\,\mathrm dt = \frac{\pi}{6}\ln^2(2) \\\\[7pt] I_2 &=2\ln(2)\int_0^{1/2} \frac{\ln(t)}{\sqrt{1-t^2}}\,\mathrm dt =2\ln(2)\int_0^{1} \frac{\ln(2x)}{2\sqrt{1-(2x)^2}}\,\mathrm dx \tag{1}\\ &=\ln(2)\int_0^{\pi/2} \frac{\ln(\sin\theta)}{\sqrt{1-\sin^2\theta}}\,\frac{1}{2}\cos\theta\, \mathrm d \theta \tag{2}\\[7pt] \quad&= \frac{\ln(2)}{2}\int_0^{\pi/2} \ln(\sin\theta)\,\mathrm d \theta =\frac{\ln(2)}{2}\left(-\frac{\pi}{2}\ln(2)\right) = -\frac{\pi}{4}\ln^2(2)\tag{3} \\\\[7pt] I_1 &= \int_0^{1/2} \frac{\ln^2(t)}{\sqrt{1-t^2}}\,\mathrm dt = \int_0^{1} \frac{\ln^2(2x)}{2\sqrt{1-(2x)^2}}\,\mathrm dx \tag{1}\\[7pt] &=\frac{1}{4}\int_0^{\pi/2}\frac{\ln^2(\sin\theta)}{\sqrt{1-\sin^2\theta}}\,\cos\theta\,\mathrm d\theta \tag{2}\\[7pt] \quad&= \frac{1}{4}\int_0^{\pi/2}\ln^2(\sin\theta)\,\mathrm d\theta = \frac{1}{4}\left(\frac{1}{24} \left(\pi^3 + 3\pi \ln^2(4)\right)\right)\tag{4}\\[7pt] \quad&= \frac{1}{4}\left(\frac{\pi^3}{24} + \frac{\pi}{8}\ln^2(4)\right) = \frac{1}{4}\left(\frac{\pi^3}{24} + \frac{\pi }{2}\ln ^2\left(2\right)\right)\\[7pt] &= \frac{\pi^3}{96} + \frac{\pi}{8}\ln^2\left(2\right) \\[7pt] \\ I&=\frac{\pi^3}{96} + \frac{\pi}{8}\ln^2\left(2\right) -\frac{\pi}{4}\ln^2(2) +\frac{\pi}{6}\ln^2(2)\\[7pt] \end{align}

$$I = \frac{\pi ^3+4\pi \ln ^2\left(2\right)}{96}$$

Explanation :

$(1)\;$ Use substitution $\;\displaystyle \frac{x}{2}=t \implies \mathrm d x = 2\mathrm d t$

$(2)\;$ Use substitution $\;\displaystyle x=\frac{1}{2}\sin\theta\quad\implies\quad dt=\frac{1}{2}\cos\theta\;\mathrm d\theta$

$(3)\;$ Use Euler log-sine integral $\;\displaystyle \int_0^{\pi/2}\ln\sin\theta\;\mathrm d\theta=-\frac{\pi}{2}\ln2$

$(4)\;$ Use $\int_0^{\pi/2}\ln^2(\sin\theta)\,\mathrm d\theta = \frac{1}{24} \left(\pi^3 + 3\pi \ln^2(4)\right)$

• When you substitute $t=\frac x2$, the limits change from zero to one to zero to $1/2$. Therefore, all the work after that substitution has been rendered moot. – Frank W. May 21 '18 at 3:29
• @FrankW. I have corrected the limits and not yet reached the result. what is wrong now :( – Darío A. Gutiérrez May 21 '18 at 11:16
• @FrankW. Double check wherever you changed variables. Your integration limits after change of variables are mostly wrong. – Szeto May 21 '18 at 14:38
• wrong limits....what a pity after all this nice work – tired May 21 '18 at 14:56
• @Szeto I didn’t write this answer. I think you meant Dario – Frank W. May 21 '18 at 15:48

This is as far as I've gotten with the integral. I will admit, I underestimated the difficulty of this integral. The four in the denominator proved to be a lot more of a nuisance than I thought. Nevertheless, this integral can be painstakingly computed by hand with the help of our best friend

$$\log\sin x=-\sum\limits_{n\geq1}\frac {\cos 2nx}n-\log 2\tag1$$

First, we make the substitution $x\mapsto 2\sin x$ to clear the four in the denominator\begin{align*}I & =\int\limits_0^{\pi/6}dx\,\log^2(2\sin x)\\ & =\int\limits_0^{\pi/6}dx\,\log^22+\log 4\int\limits_0^{\pi/6}dx\,\log\sin x+\int\limits_0^{\pi/6}dx\,\log^2\sin x\tag2\end{align*}Call the remaining integrals in (2) $I_1$, $I_2$, and $I_3$ respectively. The first integral $I_1$ is trivial

$$I_1\color{blue}{=\frac {\pi}6\log^22}\tag3$$

The second and third integrals $I_2$ and $I_3$ don't have to be fully computed to simplify the result. First, let's tackle $I_2$. Using the expansion for (1), we get

\begin{align*}I_2 & =-\log 4\int\limits_0^{\pi/6}dx\,\sum\limits_{n\geq1}\frac {\cos 2nx}n+\log 2\\ & \color{red}{=-\frac {\pi}3\log^22-\log 4\int\limits_0^{\pi/6}dx\,\sum\limits_{n\geq1}\frac {\cos 2nx}n}\tag4\end{align*}

Leave $I_2$ as is, because we can expand $I_3$ courtesy of the square and find out that a portion of $I_3$ cancels out with $I_1+I_2$. Doing the math gives

\begin{align*}I_3 & \color{brown}{=\sum\limits_{n\geq1}\sum\limits_{m\geq1}\int\limits_0^{\pi/6}dx\,\frac {\cos 2mx\cos 2nx}{mn}+\log 4\int\limits_0^{\pi/6}dx\,\sum\limits_{n\geq1}\frac {\cos 2nx}n+\frac {\pi}6\log^22}\tag5\end{align*}

Immediately, notice how the two infinite series cancel out, leaving us with a much more nicer sum to deal with. Adding (3), (4), and (5) together and canceling out all the terms leaves us with

$$I=\sum\limits_{n\geq1}\sum\limits_{m\geq1}\int\limits_0^{\pi/6}dx\,\frac {\cos 2mx\cos 2nx}{mn}$$

Now, what's left is to show that the nested sum actually equals $\frac {7\pi^3}{216}$.

• that is the difficult part :( – tired May 21 '18 at 11:49
• i think this route is rather non-trivial (i tried myself). i now changed to polylogarithm wizardy and all worked out pretty smoothly – tired May 21 '18 at 13:29
• @tired D:$\phantom{}$ – Frank W. May 21 '18 at 13:49
• @tired I’ve been doing a bit more practice with polylogs and trying to prove some of their identities and I gotta say, they’re pretty interesting! It’s really cool how they appear in infinite sums and definite integrals$$\sum\limits_{n\geq1}\frac {H_n x^n}{n^2}=\operatorname{Li}_3(x)-\operatorname{Li}_3(1-x)+\frac 12\log^2(1-x)\log x+\operatorname{Li}_2(x)\log(1-x)+\zeta(3)$$ – Frank W. May 27 '18 at 1:42
• Yeah the connection between Polylogs and the generating function of harmonic numbers is really interesting. Very handy for the calculation of many wired Euler-sums... – tired May 27 '18 at 10:41