23
$\begingroup$

When I showed to my brother how I proved \begin{equation} \int_{0}^{\!\Large \frac{\pi}{2}} \ln \left(x^{2} + \ln^2\cos x\right) \, \mathrm{d}x=\pi\ln\ln2 \end{equation} using the following theorem by Mr. Olivier Oloa \begin{equation}{\large\int_{0}^{\!\Large \frac{\pi}{2}}} \frac{\cos \left(\! s \arctan \left(-\frac{x}{\ln \cos x}\right)\right)}{(x^2+\ln^2\! \cos x)^{\Large\frac{s}{2}}}\, \mathrm{d}x = \frac{\pi}{2}\frac{1}{\ln^{\Large s}\!2}\qquad,\;\text{for }-1<s<1.\end{equation} He showed me the following interesting formula

\begin{equation} \int_{0}^{\!\Large \frac{\pi}{2}} x\csc^2(x)\arctan \left(\! \alpha \tan x\right)\, \mathrm{d}x =\frac{\pi}{2}\, \ln\left(\left[1 + \alpha\right]^{1 + \alpha} \over \alpha^\alpha\right)\,,\qquad \mbox{for}\ \alpha > 0\tag{✪}. \end{equation}

I tried several values of $\alpha$ to check its validity (since he always messes around with me) and the numerical results match the output of Mathematica $9$. The problem is how to prove this formula since he didn't tell me (as always). I tried Feynman's integration trick and I arrived to the following result: \begin{equation}\partial_\alpha\int_{0}^{\!\Large \frac{\pi}{2}} x\csc^2(x)\arctan \left(\! \alpha \tan x\right)\, \mathrm{d}x = \int_{0}^{\!\Large \frac{\pi}{2}} \frac{x\cot x}{\cos^2x+\alpha^2 \sin^2 x}\, \mathrm{d}x\end{equation} but I am having difficulty to crack the very last integral. Could anyone here please help me to prove the formula $(✪)$ preferably with elementary ways (high school methods)? Any help would be greatly appreciated. Thank you.

$\endgroup$
  • $\begingroup$ Just correct your work and finish the integral alone. It's easy. $\endgroup$ – user 1357113 Oct 16 '14 at 14:11
  • $\begingroup$ @Chris'ssis Thanks. I'll try $\endgroup$ – Anastasiya-Romanova 秀 Oct 16 '14 at 14:22
  • $\begingroup$ Mathematica can find a closed form for the indefinite integral corresponding to $(✪)$ in terms of elementary functions and dilogarithms. Then you can prove its correctness by a direct differentiation. $\endgroup$ – Vladimir Reshetnikov Oct 16 '14 at 16:41
  • $\begingroup$ @VladimirReshetnikov and could we have some nice dilogarithm special value identitity via this approach? $\endgroup$ – user153012 Oct 16 '14 at 18:36
  • 1
    $\begingroup$ @user153012 I'm afraid, no. Dilogarithms disappear when you take limits of the antiderivative at lower and upper bounds $0, \pi/2$. $\endgroup$ – Vladimir Reshetnikov Oct 16 '14 at 18:59
15
$\begingroup$

This approach is similar to Ron Gordon's answer. We introduce a new variable $\beta$ before differentiating.

By substitution $\tan x=u$, $$\int_{0}^{\!\Large \frac{\pi}{2}} x\csc^2(x)\arctan \left(\! \alpha \tan x\right)\, \mathrm{d}x=\int_{0}^{\infty}\frac{\tan^{-1}u\tan^{-1}\alpha u}{u^2}\,\mathrm{d}u$$

We will prove $$\int_0^{\infty}\frac{\tan^{-1}\alpha u\tan^{-1}\beta u}{u^2}\,\mathrm{d}u=\frac\pi2\log\left(\frac{(\alpha+\beta)^{\alpha+\beta}}{\alpha^\alpha\beta^\beta}\right)\tag{1}$$

Differentiation gives \begin{align}\partial_\alpha\partial_\beta\int_0^{\infty}\frac{\tan^{-1}\alpha u\tan^{-1}\beta u}{u^2}\,\mathrm{d}u&=\int_{0}^{\infty}\frac{\mathrm{d}u}{(1+\alpha^2 u^2)(1+\beta^2 u^2)}\\&=\frac1{\alpha^2-\beta^2}\int_{0}^{\infty}\frac{\alpha^2}{1+\alpha^2u^2}-\frac{\beta^2}{1+\beta^2u^2}\,\mathrm{d}u\\&=\frac1{\alpha^2-\beta^2}\int_{0}^{\infty}\frac{\alpha}{1+u^2}-\frac{\beta}{1+u^2}\,\mathrm{d}u\\&=\frac\pi{2(\alpha+\beta)}\end{align} for $\alpha\ne\beta$. The case $\alpha=\beta$ can be proved by letting $\alpha\to\beta$. Then $$\partial_\beta\int_0^{\infty}\frac{\tan^{-1}\alpha u\tan^{-1}\beta u}{u^2}\,\mathrm{d}u=\frac{\pi}{2}(\log(\alpha+\beta)-C_1(\beta))$$ If $\alpha\to0$ we have $0=\frac\pi2(\log\beta-C_1(\beta))$ so $C_1(\beta)=\log\beta$. Integrating by $\beta$,$$\int_0^{\infty}\frac{\tan^{-1}\alpha u\tan^{-1}\beta u}{u^2}\,\mathrm{d}u=\frac{\pi}{2}((\alpha+\beta)\log(\alpha+\beta)-\beta\log\beta-C_2(\alpha))$$ Let $\beta\to0$ and we find $C_2(\alpha)=\alpha\log\alpha$. Thus $(1)$ is proven and putting $\beta=1$ gives the conclusion $$\int_{0}^{\!\Large \frac{\pi}{2}} x\csc^2(x)\arctan \left(\! \alpha \tan x\right)\, \mathrm{d}x=\frac\pi2\log\left(\frac{(\alpha+1)^{\alpha+1}}{\alpha^\alpha}\right)$$

$\endgroup$
  • $\begingroup$ Nice answer, but in what way is my answer not as elementary as yours? $\endgroup$ – Ron Gordon Oct 17 '14 at 10:55
  • 1
    $\begingroup$ @RonGordon The last part where you decide $C$. I'm not aware of any elementary way of computing $\displaystyle\int_0^{\pi/2}\log\sin x dx$. I know three ways(fourier expansion of $\log\sin x$, differentiating the beta function, cauchy's integral formula) but none of them are 'high school methods'. Is there actually an elementary solution that I missed? $\endgroup$ – karvens Oct 17 '14 at 12:20
  • 1
    $\begingroup$ To me, 'high school methods' term means without using residue method. If you're looking the easy way to evaluate $C$, you may refer to: math.stackexchange.com/q/690644/133248. It's obviously a high school method in general meaning (ô‿ô) CC: Mr. @RonGordon $\endgroup$ – Anastasiya-Romanova 秀 Oct 17 '14 at 13:20
  • $\begingroup$ Thanks for the link! So it was easier than I thought. @RonGordon Sorry for the confusion I've caused... $\endgroup$ – karvens Oct 17 '14 at 13:26
  • $\begingroup$ @karvens: I disagree. See the addendum to my answer below. The simple technique I employ is hardly original and is used elsewhere in this site. (I see Anastasiya noted this as well.) $\endgroup$ – Ron Gordon Oct 17 '14 at 19:07
15
$\begingroup$

Note that $$ \begin{align} 1-(1-\alpha^2)\sin^2(x) &=1+(1-\alpha^2)\frac{e^{2ix}-2+e^{-2ix}}{4}\\ &=\frac{1-\alpha^2}4\left[e^{2ix}+2\frac{1+\alpha^2}{1-\alpha^2}+e^{-2ix}\right]\\ &=\frac{1-\alpha^2}4e^{-2ix}\left[e^{2ix}+\frac{1+\alpha}{1-\alpha}\right]\left[e^{2ix}+\frac{1-\alpha}{1+\alpha}\right]\\ &=\frac{(1+\alpha)^2}4\left[1+\frac{1-\alpha}{1+\alpha}e^{2ix}\right]\left[1+\frac{1-\alpha}{1+\alpha}e^{-2ix}\right]\tag{1} \end{align} $$ Taking $\frac12$ of the log of $(1)$ yields $$ \frac12\log\left(1-(1-\alpha^2)\sin^2(x)\right) =\log\left(\frac{\alpha+1}2\right)-\sum_{n=1}^\infty\left(\frac{\alpha-1}{\alpha+1}\right)^n\frac{\cos(2nx)}n\tag{2} $$ Subtracting $\log(\alpha)$ from $(2)$ and sending $\alpha\to\infty$ gives $$ \log(\sin(x))=-\log(2)-\sum_{n=1}^\infty\frac{\cos(2nx)}n\tag{3} $$ Now, we are ready to tackle the integral given. Taking the derivative, using partial fractions, integrating by parts, then applying $(2)$ and $(3)$, yields $$ \begin{align} &\frac{\mathrm{d}}{\mathrm{d}\alpha}\int_0^{\pi/2}x\csc^2(x)\arctan(\alpha\tan(x))\,\mathrm{d}x\\ &=\int_0^{\pi/2}\frac{x}{\sin(x)}\frac{1}{1-(1-\alpha^2)\sin^2(x)}\,\mathrm{d}\sin(x)\tag{4a}\\ &=\int_0^{\pi/2}\left[\frac1{\sin(x)}-\frac{\raise{5mu}{\frac12}\sqrt{1-\alpha^2}}{1+\sqrt{1-\alpha^2}\sin(x)}+\frac{\raise{5mu}{\frac12}\sqrt{1-\alpha^2}}{1-\sqrt{1-\alpha^2}\sin(x)}\right]x\,\mathrm{d}\sin(x)\tag{4b}\\ &=\int_0^{\pi/2}x\,\mathrm{d}\left[\log(\sin(x))-\tfrac12\log\left(1-(1-\alpha^2)\sin^2(x)\right)\right]\tag{4c}\\ &=-\frac\pi2\log(\alpha)-\int_0^{\pi/2}\left[\log(\sin(x))-\tfrac12\log\left(1-(1-\alpha^2)\sin^2(x)\right)\right]\mathrm{d}x\tag{4d}\\ &=-\frac\pi2\log(\alpha)+\frac\pi2\log(2)+\frac\pi2\log\left(\frac{\alpha+1}2\right)\tag{4e}\\ &=\frac\pi2\log\left(\frac{\alpha+1}\alpha\right)\tag{4} \end{align} $$ Explanation:
$\mathrm{(4a)}$: differentiate with respect to $\alpha$
$\mathrm{(4b)}$: partial fractions
$\mathrm{(4c)}$: prepare to integrate by parts
$\mathrm{(4d)}$: integrate by parts
$\mathrm{(4e)}$: apply $(2)$ and $(3)$ noting that $\int_0^{\pi/2}\frac{\cos(2nx)}{n}\,\mathrm{d}x=\left[\frac{\sin(2nx)}{2n^2}\right]_0^{\pi/2}=0$

Integrating $(4)$ gives $$ \int_0^{\pi/2}x\csc^2(x)\arctan(\alpha\tan(x))\,\mathrm{d}x =\frac\pi2\log\left(\frac{(\alpha+1)^{\alpha+1}}{\alpha^\alpha}\right)\tag{5} $$

$\endgroup$
  • $\begingroup$ Nice answer. Very clever expansion. Thank you for your answer, +1 of course ≥Ö‿Ö≤ $\endgroup$ – Anastasiya-Romanova 秀 Oct 17 '14 at 13:22
14
$\begingroup$

This integral may be done by observing the following: Let

$$I(\alpha) = \int_0^{\pi/2} dx \, x\, \csc^2{x} \, \arctan{\left ( \alpha \tan{x}\right)} $$

Then $I(\alpha)$ satisfies the differential equation

$$\begin{align}\frac{d}{d\alpha}\left ( \frac{I(\alpha)}{\alpha} \right ) &= -\frac1{\alpha^2} \int_0^{\pi/2} dx \frac{\arctan{\left (\alpha \tan{x} \right )}}{\tan{x}} \\ &= -\frac1{\alpha^2} \int_0^{\infty} du \frac{\arctan{\alpha u}}{u (1+u^2)} \\ &= -\frac1{\alpha^2} J(\alpha)\end{align}$$

One may prove this by integrating the original integral by parts on the one hand, and differentiating with respect to $\alpha$ on the other.

The integral on the RHS may be evaluated by, again, differentiating with respect to $\alpha$ as follows:

$$\begin{align} J'(\alpha) &= \int_0^{\infty} \frac{du}{(1+u^2) (1+\alpha^2 u^2)}\end{align}$$

This integral is easily done using partial fractions; the result is

$$J'(\alpha) = \frac{\pi}{2} \frac1{1+\alpha}$$

which implies that

$$J(\alpha) = \frac{\pi}{2} \log{(1+\alpha)}$$

because $J(0)=0$. We are now left to solve

$$\frac{d}{d\alpha}\left ( \frac{I(\alpha)}{\alpha} \right ) = -\frac{\pi}{2} \frac{\log{(1+\alpha)}}{\alpha^2}$$

This reduces to a simple integration by parts, the result of which is

$$I(\alpha) = \frac{\pi}{2} \left [(1+\alpha) \log{(1+\alpha)} - \alpha \log{\alpha} \right ] + C \alpha $$

Now we just need to show that $C=0$. This is equivalent to showing that

$$\int_0^{\pi/2} dx \frac{x^2}{\sin^2{x}} = \pi \log{2} $$

which is indeed true, as the integral may be shown, by integrating by parts, to be

$$-2 \int_0^{\pi/2} dx \log{\sin{x}} = -\int_0^{\pi} dx \log{\sin{x}}$$

which is indeed $\pi \log{2}$.

ADDENDUM

This last integral may in fact be done by elementary methods. For example, consider that

$$\begin{align}\int_0^{\pi} dx \log{\sin{x}} &= 2 \int_0^{\pi/2} dx \, \log{\sin{2 x}} \\ &= 2 \int_0^{\pi/2} dx \, \log{(2 \sin{x} \cos{x})} \\ &= \pi \log{2} + 2 \int_0^{\pi/2} dx \, \log{\sin{x}} + 2 \int_0^{\pi/2} dx \, \log{\cos{x}} \\&= \pi \log{2} + 4 \int_0^{\pi/2} dx \, \log{\sin{x}} \\&= \pi \log{2} + 2 \int_0^{\pi} dx \, \log{\sin{x}}\end{align}$$

A simple rearrangement proves the result.

$\endgroup$
  • $\begingroup$ @Anastasiya-Romanova herself is already smart so his brother obviously smarter than her. No doubt. +1 for your solution, excellent work as always! $\endgroup$ – Venus Oct 16 '14 at 18:13
  • $\begingroup$ Thanks for the answer Mr. Ron G, +1. I tried to make my own answer but right now I'm in school & the class is going to start. @Venus I take that as a compliment although ... Nevermind, forget it $\endgroup$ – Anastasiya-Romanova 秀 Oct 17 '14 at 0:23

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.