How do I find $$\int_{-\infty}^{\infty}\frac{\arctan\sqrt{x^2+2}}{(x^2+1)\sqrt{x^2+2}}dx=\zeta(2)$$ without Feynman integration? I saw this video, which gives $$\int_{0}^{1}\frac{\arctan\sqrt{x^2+2}}{(x^2+1)\sqrt{x^2+2}}dx=\frac{5\pi^2}{96}$$ Via Feynman integration, but I would like to know another method.
-
See also https://math.stackexchange.com/questions/3562771/evaluate-int-0-infty-frac-tan-1-left-sqrta2x2-right-leftx – user170231 Dec 17 '23 at 19:32
-
I'm not sure how likely this is to help find another solution, but through a combination of substitutions (based on the identity $\left.\arctan\sqrt{\dfrac{1-x}{1+x}}=\dfrac12\arccos x\right)$ and integration by parts, I found that this integral is equivalent to$$\frac{\pi^2}4-\frac12\int_0^1\frac{\arccos x}{\sqrt x \sqrt{1+x} (1+2x)},dx$$and the remaining integral here also evaluates to $\dfrac{\pi^2}6$. – user170231 Dec 17 '23 at 20:14
2 Answers
Notice that $\frac{\arctan x}{x} = \int_{0}^{1} \frac{dy}{1+x^2y^2}$. So
\begin{align*} \int_{-\infty}^{\infty} \frac{\arctan\sqrt{x^2+2}}{(x^2+1)\sqrt{x^2+2}} \, dx &= \int_{-\infty}^{\infty} \int_{0}^{1} \frac{1}{(x^2+1)(1+(x^2+2)y^2)} \, dydx. \end{align*}
Notice that
$$ \frac{1}{(x^2+1)(1+(x^2+2)y^2)} = \frac{1}{(x^2+1)(y^2+1)} - \frac{y^2}{(y^2+1)(1+(x^2+2)y^2)}. $$
So interchanging the order of integration,
\begin{align*} \int_{-\infty}^{\infty} \frac{\arctan\sqrt{x^2+2}}{(x^2+1)\sqrt{x^2+2}} \, dx &= \int_{0}^{1} \left( \frac{\pi}{y^2+1} - \frac{\pi y}{(y^2+1)\sqrt{2y^2+1}} \right) \, dy \\ &= \pi \left[ \arctan(y) - \arctan\sqrt{2y^2+1} \right]_{0}^{1} \\ &= \frac{\pi^2}{6}. \end{align*}

- 167,468
-
1
-
The solution of Sangchul Lee is using Feynman's trick in disguise. It's the same that to consider: $F(a)=\int_{-\infty}^{+\infty} \frac{\arctan\left(a\sqrt{x^2+2}\right)}{(x^2+1)\sqrt{x^2+2}} , dx$ then, derive wrt a and so on. – FDP Oct 15 '18 at 07:53
-
@FDP, I agree that my technique is Feynman's trick in disguise. In some sense, Fubili's theorem and Feynman's trick are like integration and differentiation, so they are the same technique in different aspects. – Sangchul Lee Oct 15 '18 at 08:22
This integral can also be evaluated via complex residues.
First, we rewrite
$$\begin{align*} \mathcal I &= \int_{-\infty}^\infty \frac{\arctan \sqrt{x^2+2}}{\left(x^2+1\right) \sqrt{x^2+2}} \, dx \\ &= 2 \int_0^\infty \frac{\arctan \sqrt{x^2+2}}{\left(x^2+1\right) \sqrt{x^2+2}} \, dx \\ &= 2 \underbrace{\int_{\sqrt2}^\infty \frac{\arctan x}{\left(x^2-1\right) \sqrt{x^2-2}} \, dx}_{=:I} & \left[x\mapsto\sqrt{x^2-2}\right] \end{align*}$$
Next, introduce the complex function
$$\begin{align*} f(z) &= \frac{\arctan z}{\left(z^2-1\right) \sqrt{z^2-2}} \\ &= -\frac i2 \frac{\log\left\lvert\frac{i-z}{i+z}\right\rvert + i \arg \left(\frac{i-z}{i+z}\right)}{\left(z^2-1\right) \sqrt{\left\lvert z^2-2\right\rvert} \cdot e^{\tfrac i2 \left[\arg\left(z-\sqrt2\right)+ \arg\left(z+\sqrt2\right)\right]}} \end{align*}$$
where we use the principal branches for $\arctan z$ and $\sqrt{z+\sqrt2}$ such that $\arg\left(\dfrac{i-z}{i+z}\right)\in(-\pi,\pi)$ and $\arg\left(z+\sqrt2\right)\in(-\pi,\pi)$; for $\sqrt{z-\sqrt2}$ we take a branch cut along the positive real axis on $\left[\sqrt2,\infty\right)$ so that $\arg\left(z-\sqrt2\right)\in(0,2\pi)$. We integrate $f(z)$ along an indented circular contour in the counterclockwise direction:
The integrals along the circular portions will all vanish in their respective limits (proof omitted). The remaining contributions are outlined below.
- Around the cut $\left[\sqrt2,\infty\right)$:
$$\begin{align*} \int_A f(z) \, dz &= \int_{\sqrt2+\varepsilon}^R f(x+i\varepsilon) \, dx \\ &\to \int_{\sqrt2}^\infty \frac{\arctan x}{\left(x^2-1\right) \sqrt{x^2-2} \cdot e^{\tfrac i2\left[0+0\right]}} \, dx \\ \int_{A'} f(z) \, dz &= \int_R^{\sqrt2+\varepsilon} f(x-i\varepsilon) \, dx \\ &\to - \int_{\sqrt2}^\infty \frac{\arctan x}{\left(x^2-1\right) \sqrt{x^2-2} \cdot e^{\tfrac i2 \left[2\pi+0\right]}} \, dx \\[2ex] \implies \int_{A\cup A'} f(z) \, dz &\to 2I \end{align*}$$
- Around the cut $i[1,\infty)$:
$$\begin{align*} \int_B f(z) \, dz &= i \int_{1+\varepsilon}^R f(-\varepsilon+ix) \, dx \\ &\to -\frac 12 \int_1^\infty \frac{\log\left\lvert\frac{1-x}{1+x}\right\rvert - i \pi}{\left(x^2+1\right) \sqrt{x^2+2} \cdot e^{\tfrac i2\left[\tfrac\pi2+\tfrac\pi2\right]}} \, dx \\ \int_{B'} f(z) \, dz &= i \int_R^{1+\varepsilon} f(\varepsilon+ix) \, dx \\ &\to \frac 12 \int_1^\infty \frac{\log\left\lvert\frac{1-x}{1+x}\right\rvert + i \pi}{\left(x^2+1\right) \sqrt{x^2+2} \cdot e^{\tfrac i2\left[\tfrac\pi2+\tfrac\pi2\right]}} \, dx \\[2ex] \implies \int_{B\cup B'} f(z) \, dz &\to \pi \int_1^\infty \frac{dx}{\left(x^2+1\right) \sqrt{x^2+2}} \, dx = \frac{\pi^2}{12} \end{align*}$$
Around the cut $\left(-\infty,-\sqrt2\right]$, a similar analysis to the $A/A'$ pair indicates that $$\int_{C\cup C'} f(z) \, dz \to 2I;$$ the only difference here is that $\arg\left(z-\sqrt2\right)+\arg\left(z+\sqrt2\right)\to-\pi+\pi$ along $C$, and $\to\pi+\pi$ along $C'$.
Around the cut $i[-1,-\infty)$, in the same vein as $B/B'$, $$\int_{D\cup D'} f(z) \, dz \to \frac{\pi^2}{12},$$ with $\arg\left(z-\sqrt2\right)+\arg\left(z+\sqrt2\right)\to\dfrac{3\pi}2-\dfrac\pi2$ along both paths.
Now, $f(z)$ has two simple poles at $z=\pm1$, with residues
$$\underset{z=\pm1}{\operatorname{Res}} f(z) = -\frac{i\pi}8$$
so by the residue theorem,
$$i2\pi \left(-\frac{i\pi}8 - \frac{i\pi}8\right) = 2I + \frac{\pi^2}{12} + 2I + \frac{\pi^2}{12} \\ \implies I = \frac{\pi^2}{12} \implies \boxed{\mathcal I = \frac{\pi^2}6}$$

- 19,334