I was playing around with some integrals and noticed that some integrals of the form: $$I(a,b,c)=\int_a^\infty \frac{\arctan(x+b)}{x^2+c}dx$$ Does have a closed form. I am trying to find for what constant $c$ will this work. In case you wonder why only $c$ is problematic, I will try to show by an example. $$I=I(1,3,16)=\int_1^\infty \frac{\arctan(x+3)}{x^2+16}dx$$ Let's start by letting $x-1=t\,$ thus: $$I=\int_0^\infty \frac{{\arctan(\color{blue}{t+4})}}{t^2+2t+17}dt$$ With $\displaystyle{t=\frac{17}{y}\rightarrow dt=-\frac{17}{y^2}dy}$ $$I=\int_0^\infty \frac{{\arctan\left(\color{red}{\frac{17}{y}+4}\right)}}{\left(\frac{17}{y}\right)^2 +\frac{34}{y}+17 }\frac{17}{y^2}dy\overset{y=t}=\int_0^\infty \frac{{\arctan\left(\color{red}{\frac{17}{t}+4}\right)}}{t^2+2t+17}dt$$ $$2I=\int_0^\infty \frac{{\arctan(\color{blue}{t+4})+{\arctan\left(\color{red}{\frac{17}{t}+4}\right)}}}{t^2+2t+17}dt$$ $${\arctan(\color{blue}{t+4})+{\arctan\left(\color{red}{\frac{17}{t}+4}\right)}}=\arctan\left(\frac{\color{blue}{t+4}+\color{red}{\frac{17}{t}+4}}{1-(\color{blue}{t+4})\left(\color{red}{\frac{17}{t}+4}\right)}\right)$$ $$=\arctan\left(\frac{x^2+8x+17}{x}\frac{x}{-4(x^2+8x+17}\right)=\pi-\arctan\left(\frac14\right)$$ Above follows since the original integral is positive so we take $\arctan(-x)$ as $\pi-\arctan x $ and therefore getting a negative answer will not be an issue. $$I=\frac12 \left(\pi -\arctan\left(\frac14\right)\right)\int_0^\infty \frac{1}{t^2+2t+17}dt$$ Well, now the inner integral is not hard to compute and the final answer happens to be: $$I=\frac12 \left(\pi -\arctan\left(\frac14\right)\right)\frac14\arctan\left(\frac{t+1}{4}\right)\bigg|_0^\infty =\frac{\pi^2}{16}-\frac{3\pi}{16}\arctan\left(\frac14\right)+\frac18\arctan^2\left(\frac14\right) $$ There are more examples that I found by checking and try such as: $$I(1,2,9)=\int_1^\infty \frac{\arctan(x+2)}{x^2+9}dx$$ $$I(2,1,6)=\int_2^\infty \frac{\arctan(x+1)}{x^2+6}dx$$ $$I(2,2,13)=\int_2^\infty \frac{\arctan(x+2)}{x^2+13}dx$$ And so on... All those can be solved by the same method: First substitute $x-a=t$, then let $t=\frac{\alpha}{y}$, where $\alpha$ is the "free of x" coefficient from the denominator.

The problem is that I tried more than $100$ combinations to get those integrals which is not that nice. How can we "smartly" find $c$ so that $I(a,b,c)$ is evaluable by symmetry? Or put in other word what should be $c$ if one wants to compute by symmetry $I(7,13,c)$?

  • 1
    $\begingroup$ By Feynman's trick $I(a,b,c)$ is related to the dilogarithm, which has a reflection formula. $\endgroup$ – Jack D'Aurizio Nov 18 '18 at 18:05
  • $\begingroup$ That's nice to hear, but the dilogarithm it's abit advanced for me. $\endgroup$ – Zacky Nov 18 '18 at 18:07
  • $\begingroup$ It doesn't change the fact that your integral is what it is. You are like asking for $\sum_{n\geq 1}\frac{(-1)^n}{n!^2}$ then stating "sorry, Bessel functions are a bit too advanced for me" :D $\endgroup$ – Jack D'Aurizio Nov 18 '18 at 18:09
  • 1
    $\begingroup$ Well, this is how I thought. First we have $$\arctan(\tan \pi) -\arctan\left(\frac14\right) =\arctan \left(\frac{\tan (\pi) - \frac14}{1+\tan(\pi) \frac14}\right)$$ Since $\tan(\pi)= 0 $ the above is just $\arctan\left(-\frac14\right)$, and I did this because the original integral is positive thus we don't want to arrive at a negative value, and since $\arctan $ is a multivalued function I choosed it to become positive. Sorry if I explain poorly. $\endgroup$ – Zacky Nov 18 '18 at 18:34
  • 1
    $\begingroup$ Ah, now it is clear to me. I would say it would be worth to be added but I guess it is enough that it is explained with the comment section. $\endgroup$ – mrtaurho Nov 18 '18 at 18:36

The simplest approaches are sometimes the right ones. Therefore choose any arbitrarily numbers $a,b$ and $c$ and apply your algorithm. So by firstly setting $t=x-a$ we get

$$\begin{align} I(a,b,c)=\int_a^{\infty}\frac{\arctan(x+b)}{x^2+c}dx=\int_0^{\infty}\frac{\arctan(t+a+b)}{t^2+2at+(c+a^2)}dt \end{align}$$

Now set $\alpha=c+a^2$ and then $\displaystyle t=\frac{c+a^2}{y}$ to further get

$$\begin{align} I(a,b,c)=\int_0^{\infty}\frac{\arctan(t+a+b)}{t^2+2at+(c+a^2)}dt&=\int_0^{\infty}\frac{\arctan\left(\frac{c+a^2}{y}+a+b\right)}{\left(\frac{c+a^2}{y}\right)^2+2a\left(\frac{c+a^2}{y}\right)+(c+a^2)}\frac{c+a^2}{y^2}dy\\ &\stackrel{y=t}{=}\int_0^{\infty}\frac{\arctan\left(\frac{c+a^2}{y}+a+b\right)}{t^2+2at+(c+a^2)}dt \end{align}$$

Adding the first and the second form up results in using the addition theorem of the inverse tangent function. This addition looks like the following

$$\small\begin{align} \arctan(\color{blue}{t+a+b})+\arctan\left(\color{red}{\frac{c+a^2}{y}+a+b}\right)&=\arctan\left(\frac{\color{blue}{t+a+b}+\color{red}{\frac{c+a^2}{y}+a+b}}{1-(\color{blue}{t+a+b})\left(\color{red}{\frac{c+a^2}{y}+a+b}\right)}\right)\\ &=\arctan\left(\frac{t^2+2(a+b)t+a^2+c}{-(a+b)\left(t^2+\frac{2a^2+2ab+b^2+c-1}{a+b}t+a^2+c\right)}\right) \end{align}$$

In order to make the polynomial $t^2+2(a+b)t+a^2+c$ vanish the following condition has to be fulfilled


From hereon we can deduce a relation between $a,b$ and $c$ that has to be satisfied. To be precise

$$\begin{align} \frac{2a^2+2ab+b^2+c-1}{a+b}=2(a+b)&\Leftrightarrow 2a^2+2ab+b^2+c-1=2(a+b)^2\\ &\Leftrightarrow 2a^2+2ab+b^2+c-1=2a^2+2b^2+4ab\\ &\Leftrightarrow c=b^2+2ab+1 \end{align}$$

And indeed your given case $I(\color{red}{1},\color{blue}{3},\color{green}{16})$ fulfills this relation as $\color{green}{16}=\color{blue}{3}^2+2\cdot\color{blue}{3}\cdot\color{red}{1}+1$. Another example would be $I(\color{red}{1},\color{blue}{1},\color{green}{4})$ which again turns out to work since $\color{green}{4}=\color{blue}{1}^2+2\cdot\color{blue}{1}\cdot\color{red}{1}+1$.

Therefore for your given case $I(7,13,c)$ you have to choose $c=352$ in order to let the integral be solvable via symmetry.

To finish the evaluation of $I(a,b,c)$, now under the restriction of $c=b^2+2ab+1$, we arrive at


following the same argumentation as you did. So for $I(a,b,c)$ as a whole we get

$$\begin{align} 2I(a,b,c)&=\left(\pi-\arctan\left(\frac1{a+b}\right)\right)\int_0^{\infty}\frac{dt}{t^2+2at+(c+a^2)}\\ &=\left(\pi-\arctan\left(\frac1{a+b}\right)\right)\int_0^{\infty}\frac{dt}{(t+a)^2+c}\\ &=\left(\pi-\arctan\left(\frac1{a+b}\right)\right)\left[\frac1{\sqrt{c}}\arctan\left(\frac{t+a}{\sqrt{c}}\right)\right]_0^{\infty}\\ \Leftrightarrow I(a,b,c)&=\frac1{2\sqrt{c}}\left(\pi-\arctan\left(\frac1{a+b}\right)\right)\left[\frac{\pi}2-\arctan\left(\frac a{\sqrt{c}}\right)\right] \end{align}$$

Where the final formula produces the right value for your example integral $I(1,3,16)$.

I do not claim that this deduced relation between $a,b$ and $c$ is the only one for which the integral can be evaluated via symmetry but in fact it is one possibilty.

  • 1
    $\begingroup$ This is perfect, thank you! $\endgroup$ – Zacky Nov 18 '18 at 20:03
  • $\begingroup$ No problem. I am glad that I could help :) $\endgroup$ – mrtaurho Nov 18 '18 at 20:05

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.