Show that: $$\int_0^1\frac{\arcsin^3 x}{x^2}\text{d}x=6\pi G-\frac{\pi^3}{8}-\frac{21}{2}\zeta(3)$$ I evaluated this by some Fourier series. Is there any other method? Start with substitution of $$u=\arcsin x$$ Then we have to integrate $$\int_0^{\frac{\pi}{2}}\frac{u^3\cos u}{\sin^2 u}\text{d}u=-\int_0^{\frac{\pi}{2}}u^3\csc u\text{d}u$$ Since $$\int\csc u\text{d}u=\ln (\csc u-\cot u)=\ln \left(\frac{1-\cos x}{\sin x}\right)=\ln 2+2\ln \left(\sin \frac{x}{2}\right)-\ln \sin x$$ Thus $$\int_0^{\frac{\pi}{2}}u^2\csc u\text{d}u=\int_0^{\frac{\pi}{2}}u^2\text{d}\left(2\ln \frac{\sin u}{2}-\ln \sin u\right)$$ $$=-\frac{\pi^2}{4}\ln 2-2\int_0^{\frac{\pi}{2}}u\left(2\ln \sin \frac{u}{2}-\ln \sin u\right)$$ $$=-\frac{\pi^2}{4}\ln 2-4\int_0^{\frac{\pi}{2}}u\ln \sin \frac{u}{2}\text{d}u+2\int_0^{\frac{\pi}{2}}u\ln \sin u\text{d}u$$ $$=-\frac{\pi^2}{4}\ln 2+4\int_0^{\frac{\pi}{2}}u\left[\ln 2+\sum_{n=1}^{\infty}\frac{\cos nu}{n}\right]\text{d}u-\int_0^{\frac{\pi}{2}}u^2\cot u\text{d}u$$ $$=\frac{\pi^2}{4}\ln 2+4\sum_{n=1}^{\infty}\frac{1}{n}\int_0^{\frac{\pi}{2}}u\cos nu\text{d}u-\frac{\pi^2}{4}\ln 2+\frac78\zeta(3)$$ $$=\frac78\zeta(3)+2\sum_{n=1}^{\infty}\frac{-2+2\cos \frac{n\pi}{2}+n\pi \sin \frac{n\pi}{2}}{n^3}$$ $$=\frac78\zeta(3)-4\zeta(3)-\frac38\zeta(3)+2\pi G=2\pi G-\frac72\zeta(3)$$ Combine these gives $$\int_0^1\frac{\arcsin ^3x}{x^2}\text{d}x=6\pi G-\frac{\pi^3}{8}-\frac{21}{2}\zeta(3)$$

  • 1
    $\begingroup$ The Taylor series $$ \arcsin^3x=6\sum_{m=0}^\infty\frac{(2m+1)!!^2}{(2m+3)!}\sum_{k=0}^m\frac1{(2k+1)^2}x^{2m+3} $$ (see OEIS sequence A001824) might help. $\endgroup$ – joriki Feb 13 '13 at 9:54
  • $\begingroup$ Is this a homework or something just of interest? $\endgroup$ – Christopher A. Wong Feb 13 '13 at 10:19
  • $\begingroup$ @ChristopherA.Wong : Just for interest! I am reading "irresistible integrals". It has a similar one but simpler $\endgroup$ – Ryan Feb 13 '13 at 10:26
  • 1
    $\begingroup$ A related problem. $\endgroup$ – Mhenni Benghorbal Feb 13 '13 at 14:48
  • $\begingroup$ @Ryan: Can you post your solution? $\endgroup$ – Mhenni Benghorbal Feb 13 '13 at 15:00

Integrating by parts twice, we get $$ \int_0^{\pi/2}x^2\,e^{ikx}\,\mathrm{d}x =i^{k-1}\frac{\pi^2}{4k}+i^k\frac\pi{k^2}+\frac2{k^3}\left(i^{k+1}-i\right) $$ Therefore, using $\sin(x)=\dfrac{e^{ix}-e^{-ix}}{2i}$, $$ \begin{align} &\int_0^1\frac{\arcsin^3(x)}{x^2}\mathrm{d}x\\ &=\int_0^{\pi/2}\frac{x^3}{\sin^2(x)}\mathrm{d}\sin(x)\\ &=-x^3\csc(x)\Big]_0^{\pi/2}+3\int_0^{\pi/2}\csc(x)\,x^2\,\mathrm{d}x\\ &=-\frac{\pi^3}{8}+3\int_0^{\pi/2}x^2\,\frac{2ie^{-ix}\,\mathrm{d}x}{1-e^{-2ix}}\tag{$\lozenge$}\\ &=-\frac{\pi^3}{8}+6i\sum_{k=0}^\infty\int_0^{\pi/2}x^2\,e^{-(2k+1)ix}\,\mathrm{d}x\\ &=-\frac{\pi^3}{8}+6i\sum_{k=0}^\infty\left((-1)^k\frac{\pi^2}{8k+4}-i(-1)^k\frac\pi{(2k+1)^2}-\left((-1)^k-i\right)\frac2{(2k+1)^3}\right)\\ &=-\frac{\pi^3}{8}+6\sum_{k=0}^\infty\left(\frac{(-1)^k\pi}{(2k+1)^2}-\frac2{(2k+1)^3}\right)\tag{$\ast$}\\ &=-\frac{\pi^3}{8}-\frac{21}{2}\zeta(3)+6\pi\sum_{k=0}^\infty\frac{(-1)^k}{(2k+1)^2}\\ &=-\frac{\pi^3}{8}-\frac{21}{2}\zeta(3)+6\pi G \end{align} $$ where $G$ is Catalan's Constant. In $(\ast)$, we drop the imaginary part (which should be $0$).

There is a question about the convergence in $(\lozenge)$. To handle this, we can consider $$ x^2\frac{2ie^{-ix}}{1-e^{-2ix}}=\lim_{r\to1^-}x^2\frac{2ire^{-ix}}{1-r^2e^{-2ix}} $$ and the convergence in the sums is uniform as $r\to1^-$.

| cite | improve this answer | |
  • $\begingroup$ Nice solution! Can you have a look on my other question about $\int_0^{\frac{\pi}{2}}x^3\csc x\text{d}x$. That was the last part when I evaluating $\int_0^1\frac{\arcsin ^4x}{x^4}\text{d}x$ $\endgroup$ – Ryan Feb 14 '13 at 1:54
  • $\begingroup$ @Ryan: I posted an answer. $\endgroup$ – robjohn Feb 14 '13 at 5:51
  • $\begingroup$ Could you explain why we left imaginary parts apart? $\endgroup$ – bianco Apr 22 at 20:45
  • $\begingroup$ $\int_0^1\frac{\arcsin^3(x)}{x^2}\,\mathrm{d}x$ is real, so the complex part is $0$. We could check that by computing the imaginary part of the right hand side, and if zero, it would give us the identity $$\pi^2\sum_{k=0}^\infty\frac{(-1)^k}{8k+4 }=2\sum_{k=0}^\infty\frac{(-1)^k}{(2k+1)^3}$$ and indeed both are $\frac{\pi^3}{16}$. But since the imaginary part of the left side is $0$, we can just drop the imaginary part of the right side. $\endgroup$ – robjohn Apr 23 at 3:02

Here's a suggestion that might be fruitful. It's not a full solution but it seems too messy to put into the comments. If somebody else completes the solution, feel free to add it to this.

You can use the substitution $\sin{t} = x$ to convert the integral to

$$ \int_0^{\pi/2} t^3 \csc{t} \cot{t} \, dt$$ and by applying integration by parts yields $$ \left[ - t^3 \csc{t} \right]_0^{\pi/2} + \int_0^{\pi/2} 3t^2 \csc{t} \, dt = - \frac{\pi^2}{8} + \int_0^{\pi/2} 3t^2 \csc{t} \, dt$$ This produces one of the three special terms that you want. From here, I am not sure. We could calculate the power series of $\csc{t}$, from which the remaining integral can be readily expressed as an infinite series. Mathematica tells me that $$ \int_0^{\pi/2} \log| \csc{t} + \cot{t} | \, dt = 2G,$$ and that integral can fall out by integration by parts once more, but that seems a little too difficult to form a power series from.


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.