14

There is numerical evidence that $$I=\int_{\arccos(1/4)}^{\pi/2}\arccos\left(\cos x\left(2\sin^2x+\sqrt{1+4\sin^4x}\right)\right)\mathrm dx=\frac{\pi^2}{40}$$

How can this be proved?

I was trying to answer another question, and I got it down to this integral.

Wolfram does not evaluate the indefinite integral.

I tried techniques from a roughly similar integral. Letting $u=\tan \frac{x}{2}$, I got

$$I=\int_{\sqrt{3/5}}^1 \dfrac{2\arccos{\left(\frac{(1-u^2)\left(8u^2+\sqrt{u^8+4u^6+70u^4+4u^2+1}\right)}{(1+u^2)^3}\right)}}{1+u^2}\mathrm du$$

but I don't know what to do with this.

Zacky
  • 27,674
Dan
  • 22,158

1 Answers1

12

Some algebra to arrive at a manageable integral

Notice that the $\arccos $ argument is the "$+$" solution of a quadratic equation:

$$\cos x\left(2\sin^2x\pm\sqrt{1+4\sin^4x}\right)=\frac{2\sin x \sin(2x)\pm\sqrt{4\sin^2 x\sin^2(2x)+4\cos^2 x}}{2}$$

More exactly, with $b=-2\sin x\sin(2x)$ and $c=-\cos^2x$ in $y_{\pm}=\frac{-b\pm\sqrt{b^2-4c}}{2}$.

We also have $\arccos y_-\pm\arccos y_+=\arccos\left(y_-y_+\mp\sqrt{1-(y_-+y_+)^2 +2y_-y_+ +y_-^2y_+^2}\right)$, and since Vieta's relations allows us to utilize $y_-+y_+=2\sin x\sin(2x)$ and $\ y_-y_+=-\cos^2 x$, it makes sense to consider the following integrals:

$$I_{\pm}=\int_{\arccos(1/4)}^{\pi/2}\arccos\left(\cos x\left(2\sin^2x\pm\sqrt{1+4\sin^4x}\right)\right) dx$$

The original integral can then be rewritten as $I_+=\frac12\left((I_- + I_+)-(I_- - I_+)\right)$, where:

$$I_- \pm I_+=\int_{\arccos(1/4)}^{\pi/2}\arccos\left(-\cos^2 x\mp\sin^2 x\sqrt{1-16\cos^2 x}\right)dx$$

$$\overset{\sqrt{1-16\cos^2 x}\to x}=\int_0^1 \frac{x\arccos \left(-\frac{1-x^2}{16}\mp x\frac{15+x^2}{16}\right)}{\sqrt{1-x^2}{\sqrt{15+x^2}}}dx$$

Moreover, substituting $x\to -x$ in $I_- - I_+$, gives us:

$$I_+=\frac12\int_{-1}^1 \frac{x\arccos \left(-\frac{1-x^2}{16}- x\frac{15+x^2}{16}\right)}{\sqrt{1-x^2}{\sqrt{15+x^2}}}dx=\int_{-1}^1 \frac{x\operatorname{arctan}\sqrt{\frac{1-x^2}{15+x^2} \left(1+\frac{16}{(1-x)^2}\right)}}{\sqrt{1-x^2}{\sqrt{15+x^2}}}dx$$

In the last step it was utilized that $\arccos x =2\arctan \sqrt{\frac{1-x}{1+x}}$.


Evaluation of the integral

$$\int_{-1}^1 \frac{x\operatorname{arctan}\sqrt{\frac{1-x^2}{15+x^2} \left(1+\frac{16}{(1-x)^2}\right)}}{\sqrt{1-x^2}{\sqrt{15+x^2}}}dx=\int_{-1}^1 \int_0^{\sqrt{1+\frac{16}{(1-x)^2}}} \frac{x}{(15+x^2)+(1-x^2) y^2}dy dx$$

$$\overset{(*)}=16\int_{-1}^1 \int_0^{x} \frac{x}{(15+x^2)+(1-x^2) \left(1+\frac{16}{(1-y)^2}\right)}\frac{1}{(1-y)^2\sqrt{16+(1-y)^2}} dy dx$$

$$=16\int_{-1}^1 \frac{1}{(1-y)^2\sqrt{16+(1-y)^2}}\int_y^1 \frac{x}{(15+x^2)+\left(1+\frac{16}{(1-y)^2}\right)(1-x^2)} dx dy$$

$$=-\frac12\int_{-1}^1 \frac{\ln\left(\frac{1-y}{2}\right)}{\sqrt{16+(1-y)^2}}dy\overset{\frac{1-y}{2}\to y}=-\frac12 \int_0^1 \frac{\ln y}{\sqrt{4+y^2}}dy$$

$$\overset{y \to \frac{1-y^2}{y}}=\frac12 \int_{\large \frac{1}{\phi}}^1 \frac{\ln\left(\frac{y}{1-y^2}\right)}{y}dy =\frac14\operatorname{Li}_2(1) - \frac14\operatorname{Li}_2\left(\frac{1}{\phi^2}\right) - \frac{1}{4}\ln^2\phi \overset{(**)}=\, \boxed{\frac{\pi^2}{40}}$$


In $(**)$, the following dilogarithm values were employed:

$$\operatorname{Li}_2(1)=\frac{\pi^2}{6},\quad \operatorname{Li}_2\left(\frac{1}{\phi^2}\right) = \frac{\pi^2}{15}-\ln^2\phi,\quad \phi=\frac{1+\sqrt 5}{2}$$

Also, the $(*)$ step is not necessary, however I didn't know how to change the order of integration directly, so I had to make the substitution $y^2\to 1+\frac{16}{(1-y)^2}$ after noticing that the $x$-integral is odd which allows to write $\int_{-1}^1 \int_0^{f(x)}dydx = -\int_{-1}^1 \int_{f(x)}^\infty dydx$. Finally, $-\int_{-1}^1 \int_x^1 dydx$ was also rewritten as $\int_{-1}^1 \int_0^x dydx$ in order to change the order of integration easier.

Zacky
  • 27,674
  • 2
    Excellent, as always! Just a minor observation: Mathematica wouln't solve the integral $-\frac12 \int_0^1 \frac{\ln y}{\sqrt{4+y^2}}dy$ but the simple substitution $y\to 2z$ led to sucess : $-\frac12 \int_{0}^{\frac{1}{2}} \frac{\ln 2z}{\sqrt{1+z^2}}dz = \frac{\pi^2}{40}$ – Dr. Wolfgang Hintze Mar 14 '24 at 13:53
  • @Dr.WolfgangHintze thank you! It's quite weird that such a small substitution matters for Mathematica. For more clarity, the change of variable $y\to \frac{1-y^2}{y}$ is just an Euler substitution, equivalent to $\sqrt{1+\frac{y^2}{4}}-\frac{y}{2}\to y$. – Zacky Mar 14 '24 at 14:24
  • 4
    @ Zacky What's more, the antiderivative of the integral is found by Mathematica without problems. And the correcponding limits as well. But let me, after some futile attempts of my own, make my adorement of your solution more explicit. The initial integral looks as if someone started from a simple integral and made it artificially complicated. To show how to unwrap it with a long chain of substitutions is really a marvellous achievement of yours; and it is beautiful. – Dr. Wolfgang Hintze Mar 14 '24 at 15:51
  • Thanks again for the compliment and the kind words! – Zacky Mar 14 '24 at 17:19