27
$\begingroup$

Does the following equality hold?

$$\large \int_0^1 \frac{\tan^{-1}{\left(\frac{88\sqrt{21}}{215+36x^2}\right)}}{\sqrt{1-x^2}} \, \text{d}x = \frac{\pi^2}{6}$$

The supposed equality holds to 61 decimal places in Mathematica, which fails to numerically evaluate it after anything greater than 71 digits of working precision. I am unsure of it's correctness, and I struggle to prove it's correctness.

The only progress I have in solving this is the following identity, which holds for all real $x$:

$$\tan^{-1}{\left( \frac{11+6x}{4\sqrt{21}} \right )} + \tan^{-1}{\left( \frac{11-6x}{4\sqrt{21}} \right )} \equiv \tan^{-1}{\left(\frac{88\sqrt{21}}{215+36x^2}\right)}$$

I also tried the Euler Substitution $t^2 = \frac{1-x}{1+x}$ but it looks horrible.

Addition: Is there some kind of general form to this integral?

Side thoughts: Perhaps this is transformable into the Generalised Ahmed's Integral, or something similar.

$\endgroup$
  • 5
    $\begingroup$ Here is a link:math.stackexchange.com/questions/397742/… $\endgroup$ – gymbvghjkgkjkhgfkl Feb 24 '17 at 11:05
  • 2
    $\begingroup$ I'm still searching for an alternative solution too. Using the change of variable of Jack Lam it's equal to $\displaystyle 2\int_0^1 \dfrac{\arctan\left(\dfrac{88\sqrt{21}(1+x^2)^2}{251x^4+358x^2+251}\right)}{1+x^2}dx$ $\endgroup$ – FDP Feb 27 '17 at 10:28
  • $\begingroup$ @FDP Is it possible to apply Double Integral Exchange (DIE) or Differntiation Under The IntEgral Sign (DUTIES) in this case? $\endgroup$ – Jack Tiger Lam Feb 27 '17 at 10:58
  • $\begingroup$ The double integral you get using $\displaystyle \arctan x=\int_0^1 \dfrac{x}{1+x^2t^2}dt$ looks horrible. $\endgroup$ – FDP Feb 27 '17 at 12:08
  • $\begingroup$ Maybe some magic change of variable makes equal this integral to $\displaystyle \int_0^{\tfrac{\pi}{4}} \dfrac{32x}{6} dx$ $\endgroup$ – FDP Feb 27 '17 at 12:19
19
$\begingroup$

As pointed out in one of the comments, user @Start wearing purple demonstrated a very general approach for solving this kind of integral, see this. As an alternative approach, let me give a different argument that appeals to a specific property satisfied by OP's integral.

Step 1. (Reduction and the main claim) We begin by substituting $x = \cos(\theta/2)$. Then the integral equals

$$ \frac{1}{2} \int_{0}^{\pi} \arctan\left(\frac{88\sqrt{21}}{233+18\cos\theta}\right) \, d\theta = \frac{\pi}{4} - \frac{1}{2} \int_{0}^{\pi} \arctan\left(\frac{233+18\cos\theta}{88\sqrt{21}}\right) \, d\theta. $$

So it suffices to prove that

$$ \int_{0}^{\pi} \arctan\left(\frac{233+18\cos\theta}{88\sqrt{21}}\right) \, d\theta \stackrel{?}{=} \frac{\pi^2}{6}. \tag{1} $$

To evaluate this integral, let me give the punchline.

Claim. Let $0 < a <1$ and $b > 0$ satisfy $4a^2 - b^2 = \frac{4}{3}$. Then $$ \int_{0}^{\pi} \arctan(a + b\cos\theta) \, d\theta = \frac{\pi^2}{6}. $$

Notice that $(a, b) = \left( \frac{233}{88\sqrt{21}}, \frac{18}{88\sqrt{21}} \right)$ satisfies the relation in the assertion of Claim. So we focus on proving this claim.

Step 2. (Definition and properties of $I$) Now define $I(a, b)$ by

$$ I(a, b) = \int_{0}^{\pi} \arctan(a + b\cos\theta) \, d\theta. $$

From the substitution $\theta \mapsto \pi - \theta$, it is clear that $I(a,-b) = I(a, b)$. Then for $0 < a < 1$ and $0 < \theta < \pi$, we have

\begin{align*} &\arctan(a + b\cos\theta) + \arctan(a - b\cos\theta) \\ &\hspace{1em}= \arctan\left( \frac{2a}{1-(a^2-b^2\cos^2\theta)} \right) \\ &\hspace{2em}= \arctan\left( \frac{4a}{2-2a^2+b^2+b^2\cos(2\theta)} \right) \\ &\hspace{3em}= \frac{\pi}{2} - \arctan\left( \frac{2-2a^2+b^2}{4a} + \frac{b^2}{4a}\cos(2\theta) \right). \end{align*} Plugging this back and exploiting the symmetry of cosine, we have

$$ I(a, b) = \frac{\pi^2}{4} - \frac{1}{2}I\left( \frac{2-2a^2+b^2}{4a}, \frac{b^2}{4a} \right). \tag{2} $$

Step 3. Now here comes the central observation. Let $(a, b)$ satisfy $0 < a < 1$ and $b > 0$, and define the sequence $(a_n, b_n)$ recursively by

$$ (a_0, b_0) = (a, b), \qquad (a_{n+1}, b_{n+1}) = \left( \frac{2-2a_n^2+b_n^2}{4a_n}, \frac{b_n^2}{4a_n} \right). $$

Observation. Assume that $4a^2 - b^2 = \frac{4}{3}$. Then for all $n \geq 0$ we have $$ \frac{1}{\sqrt{3}} \leq a_{n+1} \leq a_n, \qquad 4a_n^2 - b_n^2 = \frac{4}{3}. $$

The proof is a tedious algebra, so we skip this. Now by this observation, we have $|a_n| < 1$ for all $n$. Then a recursive application of $\text{(2)}$ gives

$$ I(a, b) = \frac{\pi^2}{4}\sum_{k=0}^{n-1} \left(-\frac{1}{2}\right)^k + \left(-\frac{1}{2}\right)^n I(a_n, b_n). $$

Since $|I(a_n, b_n)| \leq \frac{\pi^2}{2}$ for all $n$, taking limist as $n\to\infty$ proves the claim.


Remark. (1) The condition $4a^2 - b^2 = \frac{4}{3}$ is crucial for our proof. For arbitrary starting point $(a, b)$, the sequence $(a_n, b_n)$ is dynamically unstable and hence the formula $\text{(2)}$ is not applicable.

(2) The claim is true for any $a > 0$ in view of the principle of analytic continuation.

(3) Again, @Start wearing purple's computation gives a more general result with a relatively economic computation: for all $a, b \in \Bbb{R}$,

$$ \int_{0}^{\pi} \arctan(a + b\cos\theta) \, d\theta = \pi \arg \left(1 + ia + \sqrt{b^2 + (1+ia)^2}\right). \tag{3} $$

This follows from the formula

$$ \int_{0}^{\pi} \log(1 + s \cos\theta) \, d\theta = \pi \log\left( \frac{1 + \sqrt{1-s^2}}{2} \right) $$

which is valid for any complex $s$ with $|s| < 1$. Our relation $4a^2 - b^2 = \frac{4}{3}$ ensures that the RHS of $\text{(3)}$ is always $\frac{\pi^2}{6}$, since $1 + ia + \sqrt{b^2 + (1+ia)^2} = (1+\sqrt{3}a)\left( 1 + \frac{i}{\sqrt{3}} \right)$.

| cite | improve this answer | |
$\endgroup$
  • 1
    $\begingroup$ Amazing! How did you come up with the remarkable recurrence relation?! $\endgroup$ – Jack Tiger Lam Apr 26 '17 at 3:26
  • 4
    $\begingroup$ @JackLam, Obviously my beginning point is the formula $\text{(2)}$. The value of $I$ is exactly $\pi^2/6$ when this formula is repeatedly applicable. That fact that $I$ is not constant for general $(a, b)$ means that this is only possibly true along the level curve $C : I(a, b) = \pi^2/6$. Then $(a_n, b_n)$ will be a sequence of points on $C$. I tried to figure out this curve by taming the dynamic instability of $(a_n)$, and luckily I succeeded. $\endgroup$ – Sangchul Lee Apr 26 '17 at 3:45

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.