Is it possible to evaluate this integral in a closed form? $$I=\int_0^1\frac{\arctan^2x}{\sqrt{1-x^2}}\mathrm dx$$ It also can be represented as $$I=\int_0^{\pi/4}\frac{\phi^2}{\cos \phi\,\sqrt{\cos 2\phi}}\mathrm d\phi$$

up vote 35 down vote accepted

Okay, finally I was able to prove it.

Step 0. Observations. In view of the following identity

$$ \int_{0}^{\frac{\pi}{2}} \arctan (r \sin\theta) \, d\theta = 2 \chi_{2} \left( \frac{\sqrt{1+r^{2}} - 1}{r} \right), $$

Vladimir's result suggests that there may exists a general formula connecting

$$ I(r, s) = \int_{0}^{\frac{\pi}{2}} \arctan (r \sin\theta) \arctan (s \sin\theta) \, d\theta $$

and the Legendre chi function $\chi_{2}$. Indeed, inspired by Vladimir's result, I conjectured that

$$ I(r, s) = \pi \chi_{2} \left( \frac{\sqrt{1+r^{2}} - 1}{r} \cdot \frac{\sqrt{1+s^{2}} - 1}{s} \right). \tag{1} $$

I succeeded in proving this identity, so I post a solution here.

Step 1. Proof of the identity $\text{(1)}$. It is easy to check that the following identity holds:

$$ \arctan(ab) = \int_{1/b}^{\infty} \frac{a \, dx}{a^{2} + x^{2}}. $$

So it follows that

\begin{align*} I(r, s) &= \int_{1/r}^{\infty} \int_{1/s}^{\infty} \int_{0}^{\frac{\pi}{2}} \frac{\sin^{2}\theta}{(x^{2} + \sin^{2}\theta)(y^{2} + \sin^{2}\theta)} \, d\theta dy dx \\ &= \int_{1/r}^{\infty} \int_{1/s}^{\infty} \int_{0}^{\frac{\pi}{2}} \frac{1}{x^{2} - y^{2}} \left( \frac{x^{2}}{x^{2} + \sin^{2}\theta} - \frac{y^{2}}{y^{2} + \sin^{2}\theta} \right) \, d\theta dy dx \\ &= \frac{\pi}{2} \int_{1/r}^{\infty} \int_{1/s}^{\infty} \frac{1}{x^{2} - y^{2}} \left( \frac{x}{\sqrt{x^{2} + 1}} - \frac{y}{\sqrt{y^{2} + 1}} \right) \, dy dx. \end{align*}

For the convenience of notation, we put

$$ \alpha = \frac{\sqrt{r^{2} + 1} - 1}{r} \quad \text{and} \quad \beta = \frac{\sqrt{s^{2} + 1} - 1}{s}. $$

Then it is easy to check that $\mathrm{arsinh}(1/r) = - \log \alpha$ and likewise for $s$ and $\beta$. Thus with the substitution $x \mapsto \sinh x$ and $y \mapsto \sinh y$, we have

\begin{align*} I(r, s) &= \frac{\pi}{2} \int_{-\log\alpha}^{\infty} \int_{-\log\beta}^{\infty} \frac{\sinh x \cosh y - \sinh y \cosh x}{\sinh^{2}x - \sinh^{2}y} \, dy dx. \end{align*}

Applying the substitution $e^{-x} \mapsto x$ and $e^{-y} \mapsto y$, it follows that

\begin{align*} I(r, s) &= \pi \int_{0}^{\alpha} \int_{0}^{\beta} \frac{dydx}{1 - x^{2}y^{2}} \\ &= \pi \sum_{n=0}^{\infty} \left( \int_{0}^{\alpha} x^{2n} \, dx \right) \left( \int_{0}^{\beta} y^{2n} \, dx \right) = \pi \sum_{0}^{\infty} \frac{(\alpha \beta)^{2n+1}}{(2n+1)^{2}} \\ &= \pi \chi_{2}(\alpha \beta) \end{align*}

as desired, proving the identity $\text{(1)}$.


EDIT. I found a much simpler and intuitive proof of $\text{(1)}$. We first observe that $\text{(1)}$ is equivalent to the following identity

$$ \int_{0}^{\frac{\pi}{2}} \arctan\left( \frac{2r\sin\theta}{1-r^{2}} \right) \arctan\left( \frac{2s\sin\theta}{1-s^{2}} \right) \, d\theta = \pi \chi_{2}(rs). $$

Now we first observe that from the addition formula for the hyperbolic tangent, we obtain the following formula

$$ \operatorname{artanh}x - \operatorname{artanh} y = \operatorname{artanh} \left( \frac{x - y}{1 - xy} \right) $$

which holds for sufficiently small $x, y$. Thus

\begin{align*} \arctan\left( \frac{2r\sin\theta}{1-r^{2}} \right) &= \frac{1}{i} \operatorname{artanh}\left( \frac{2ir\sin\theta}{1-r^{2}} \right) = \frac{\operatorname{artanh}(re^{i\theta}) - \operatorname{artanh}(re^{-i\theta})}{i} \\ &= 2 \Im \operatorname{artanh}(re^{i\theta}) = 2 \sum_{n=0}^{\infty} \frac{\sin(2n+1)\theta}{2n+1} r^{2n+1}. \end{align*}

We readily check this holds for any $|r| < 1$. Therefore

\begin{align*} &\int_{0}^{\frac{\pi}{2}} \arctan\left( \frac{2r\sin\theta}{1-r^{2}} \right) \arctan\left( \frac{2s\sin\theta}{1-s^{2}} \right) \, d\theta \\ &\quad = 4 \sum_{m=0}^{\infty} \sum_{n=0}^{\infty} \frac{r^{2m+1}s^{2n+1}}{(2m+1)(2n+1)} \int_{0}^{\frac{\pi}{2}} \sin(2m+1)\theta \sin(2n+1)\theta \, d\theta\\ &\quad = 2 \sum_{m=0}^{\infty} \sum_{n=0}^{\infty} \frac{r^{2m+1}s^{2n+1}}{(2m+1)(2n+1)} \int_{0}^{\frac{\pi}{2}} \{ \cos(2m-2n)\theta - \cos(2m+2n+2)\theta \} \, d\theta\\ &\quad = \pi \sum_{m=0}^{\infty} \sum_{n=0}^{\infty} \frac{r^{2m+1}s^{2n+1}}{(2m+1)(2n+1)} \delta_{m,n} \\ &\quad = \pi \chi_{2}(rs). \end{align*}

  • 3
    Very nice result! – Vladimir Reshetnikov Nov 11 '13 at 22:05
  • How did you realize it would be incredibly convenient to define the parameters $\alpha$ and $\beta$ in such a manner? I would never have thought of that. – Random Variable Nov 20 '13 at 21:21
  • 2
    @RandomVariable As I revealed at the beginning of my answer, the choices of $\alpha$ and $\beta$ came out of the observation from a 1-variable version of this formula as an analogy. When I first derived this, however, I was not sure if this analogy does hold indeed. – Sangchul Lee Nov 20 '13 at 22:33
  • I was assuming that your evaluation of the 1-variable version was similar. A while ago I evaluated $I(1)$ by differentiating inside of $I(r)$. But I didn't actually evaluate $I(r)$ because at the time it didn't seem possible. mymathforum.com/… – Random Variable Nov 20 '13 at 23:25
  • 1
    +1. Great answer which represents a lot of work. Thanks. – Felix Marin Jun 30 '14 at 20:05

This is a beautiful integral with a beautiful result: $$\int_0^1\frac{\arctan^2x}{\sqrt{1-x^2}}dx=\frac\pi2\Big(\operatorname{Li}_2\left(3-\sqrt8\right)-\operatorname{Li}_2\left(\sqrt8-3\right)\Big).$$ One could expect it to have an elegant solution too, giving an insight into why the result is of this form.

Alas, I was not able to find such a solution. Mine was a long, semi-manual with a lot of assistance from Mathematica, lookups into Gradshteyn-Ryzhik Table of Integrals and huge intermediate expressions not fitting on a single screen. I would rather not post it here now. Instead, I hope somebody comes up with an elegant one.


Update: An alternative and perhaps simpler form of the result is: $$\frac{7\pi^3}{48}-\frac\pi8\,\ln^22-\frac\pi2\,\ln(1+\sqrt2)\cdot\ln2-\pi \operatorname{Li}_2\left(\tfrac1{\sqrt{2}}\right)$$

  • Note also that the integral is equivalent to the polylogarithmic ladder $$\int_0^1\frac{\arctan^2x}{\sqrt{1-x^2}}dx=\frac{\pi}{2}\left(2\operatorname{Li}_2(\beta^2)-\tfrac{1}{2}\operatorname{Li}_2(\beta^4)\right)$$ where $\beta=1-\sqrt{2}$. – Tito Piezas III Oct 16 '16 at 15:58

Your Answer

 

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.