Does anybody have any thoughts about how to integrate

$$I=\int_0^{\frac{\pi}{2}} x (\log\tan x)^{2n+1}\;dx$$

for integral $n$ where $n\ge 1$

In the case $n=0$

$$\int_0^{\frac{\pi}{2}} x \log\tan x \;dx=\lambda(3)=\frac{7}{8}\zeta(3)$$

I have managed to integrate the function when the exponent is even, that is $(\log\tan x)^{2n}$, using the substitution $y=\left(\frac{\pi}{2}-x \right)$ over the two intervals $[0,\frac{\pi}{4}]$ and $[\frac{\pi}{4},\frac{\pi}{2}]$, but the same trick does not apply in regard to the odd powers.

Basically via integration by parts I am left with the repeated integral

$$\int_0^{\frac{\pi}{2}} \int_0^x (\log\tan u)^{2n+1}\;du\;dx$$

As far as I know $(\log\tan u)^{2n+1}$ does not have a definite integral I can use, so I am stuck. I've tried a few substitutions and those have not helped. Any ideas how I might proceed?

Some Added Background Notes

  1. To obtain a function more suitable for numerical integration use the substitution $u=\log \tan x$ to give

$$\int_0^{\frac{\pi}{2}} x (\log\tan x)^{n}\;dx=\int_{-\infty}^{+\infty} \arctan(e^u)\frac{u^n}{e^u+e^{-u}} \;du$$

This shows that the integral $I$ is closely related to the standard integral for the $\beta(n)$ function. The analogous integral $\int_0^{\infty} x (\log\tanh x)^{n}\;dx$ via a similar change of variables is seen to be related to the standard integral for the $\lambda(n)$ function.


By setting $x=\arctan u$ we are left with

$$ \mathcal{I}(n) = \int_{0}^{+\infty}\frac{\arctan u}{1+u^2}\left(\log u\right)^{2n+1}\,du =\left.\frac{d^{2n+1}}{d\alpha^{2n+1}}\int_{0}^{+\infty}\frac{u^\alpha \arctan u}{1+u^2}\,du\right|_{\alpha=0}$$ but the integral in the RHS is related to the Beta function. By un-doing the previous substitution, $$ \int_{0}^{+\infty}\frac{u^\alpha \arctan u}{1+u^2}\,du = \int_{0}^{\pi/2}x\left(\sin x\right)^{\alpha}\left(\cos x\right)^{-\alpha}\,dx $$ where we may write $x$ as $$ \arcsin(\sin x)= \sum_{n\geq 0}\frac{(\sin x)^{2n+1}}{(2n+1)4^n}\binom{2n}{n}$$ leading to: $$ \int_{0}^{+\infty}\frac{u^{\alpha}\arctan u}{1+u^2}\,du=\sum_{n\geq 0}\frac{\binom{2n}{n}}{(2n+1)4^n}\cdot\frac{\Gamma\left(\frac{1}{2}-\frac{a}{2}\right)\, \Gamma\left(1+\frac{a}{2}+n\right)}{2\,\Gamma\left(\frac{3}{2}+n\right)}. $$ Now "it is enough" to differentiate both sides with respect to $\alpha$ the correct number of times and perform an evaluation at $\alpha=0$ in order to convert the original integral in a "twisted hypergeometric series", whose terms depend both on hypergeometric terms and generalized harmonic numbers (arising from the differentiation of the Beta function).

  • $\begingroup$ Amazing solution! $\endgroup$ – openspace Dec 14 '17 at 22:21
  • $\begingroup$ I am not really that experienced with this type of solution. I am still struggling on the RHS of the last formula line with the differentiations and the summation. Even for the $n=1$ case it is very hard. However the messy sum I have created so far numerically agrees with the integral, so I am happy that you have put me on the right track. I've struggled with this difficult integral on and off for over a year now and I would have never have found this method myself. Thank you. $\endgroup$ – James Arathoon Dec 15 '17 at 0:43

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.