7
$\begingroup$

I recently encountered this integral $$\int_0^\infty\frac{\cos(a x+ 2b \arctan x)}{x^2+1}dx$$ which suspiciously close to 0 for nonzero integer values of $b$, as indicated by numerical calculations. When $b$ is not an integer it is not 0. Based on my experiments I conjecture that $$\int_0^\infty\frac{\cos(a x+ 2b \arctan x)}{x^2+1}dx=0,~~~~b\in \mathbb{Z}\setminus \{0\}.$$

Question. Is the above conjecture true?

$\endgroup$
11
  • $\begingroup$ So did you apply $cos(A+B)$? $\endgroup$
    – Mason
    May 22, 2018 at 14:14
  • $\begingroup$ @Mason and then what? There will be a sum of 2 integrals that doesn't seem to be easier to handle. $\endgroup$
    – Tyrell
    May 22, 2018 at 14:15
  • $\begingroup$ Also I am not sure that contour-integration is an appropriate tag. $\endgroup$
    – Mason
    May 22, 2018 at 14:16
  • 2
    $\begingroup$ If we take $u=\arctan(x)$ Then we have $ \int_0^{\pi/2} cos(a \tan(u)+bu) du$ where $a \tan$ means $a \times \tan()$ $\endgroup$
    – Mason
    May 22, 2018 at 14:40
  • 1
    $\begingroup$ It must be $2bu$, but else the transformation is correct. I have tried different values for $a$ and $b$ via WolframAlpha, and it always (and quickly!) yields $0$ if $b$ is an integer, so I suppose there must indeed be some transformation to simplify the integral. $\endgroup$
    – Thern
    May 22, 2018 at 15:04

3 Answers 3

8
$\begingroup$

This can, as the author suspected, be approached by contour integration.

Let $a>0$ and $n\in\mathbb{Z}$. As Mason pointed out, the integral in question is equal to $$\int_0^{\pi/2}\cos(a\tan(x)+2nx)dx$$ Using that the integrand is even we may extend the bounds to $(-\pi/2,\pi/2)$ then perform the substitution $t=2x$ to find this is $$\frac{1}{4}\int_{-\pi}^\pi \cos(a\tan(t/2)+nt)dt$$ Notice that this is is equal to $$\frac{1}{4}\int_{-\pi}^\pi \exp(ia\tan(t/2)+int)dt$$ since the odd part of $e^{ix}=\cos(x)+i\sin(x)$ cancels. Now by expanding the tangent in terms of exponentials and performing some algebra we find $$\tan(t/2)=\frac{1}{i}\frac{e^{it}-1}{e^{it}+1}$$ Thus we may rewrite our integral as a contour integral with $z=e^{i t}$, $dt=\frac{1}{iz}dz$. Since our bounds are $t\in(-\pi,\pi)$, the contour is the unit circle. Thus our integral is $$\frac{1}{4i}\oint \exp(a\frac{z-1}{z+1})z^{n-1}dz$$ (Notice $e^{int}=z^n$.) By the Residue Theorem, this contour integral is equal to $$\frac{\pi}{2}\text{ Res}_{z=0} \exp(a\frac{z-1}{z+1})z^{n-1}$$ This residue is the coefficient of $z^{-n}$ in $\exp(a\frac{z-1}{z+1})$. Since this function is analytic inside the unit disk, this coefficient is $0$ for $n>0$ (with $n\in\mathbb{Z}$). The value of the integral for negative integral $n$ is given by the $|n|$th Taylor coefficient of $\frac{\pi}{2}\exp(a\frac{z-1}{z+1})$.

$\endgroup$
1
  • $\begingroup$ Well done. Note that if $a$ and $n$ are negative, we can return to positive values of both because $\cos(-x)=cos(x)$. $\endgroup$
    – Thern
    May 23, 2018 at 9:37
3
$\begingroup$

Maybe not a full answer, but part of the way:

Let $a=0$. Then we have

$$\int_0^\infty\frac{\cos(0+ 2b \arctan x)}{x^2+1}dx=\left[\frac{\sin\left(2b \arctan(x)\right)}{2b}\right]_0^\infty\\ =\frac{\sin\left(2b\cdot\frac{\pi}{2}\right)}{2b}-\frac{\sin(2b\cdot0)}{2b}\\ = \frac{\sin(b\pi)}{2b}$$

This is obviously zero if and only if b is an integer (except zero).

Currently I am not fully aware why this behavior remains when $a\neq0$, maybe I find a solution for that later on.

$\endgroup$
1
  • 2
    $\begingroup$ I'd like to point out that: only when the sign of $a$ and $b$ are the same can the conjecture be true. For example this is not zero but this is zero. $\endgroup$ May 22, 2018 at 18:24
1
$\begingroup$

Assume that $a > 0$ and that $n$ is a positive integer.

Then we have

$$ \begin{align}\int_{0}^{\infty} \frac{\cos \left(ax + 2 n \arctan x \right)}{1+x^{2}} \, \mathrm dx &= \frac{1}{2} \, \Re \int_{-\infty}^{\infty} \frac{\exp \left(iax + i2n \arctan x \right)}{1+x^{2}} \, \mathrm dx \\ &= \frac{1}{2} \, \Re \int_{-\infty}^{\infty} e^{iax} \, \frac{ \exp \left(n \log(1+ix) - n\log(1-ix) \right)}{1+x^{2}} \, \mathrm dx \\ &= \frac{1}{2} \, \Re \int_{-\infty}^{\infty} e^{iax} \, \frac{(1+ix)^{n}}{(1-ix)^{n}(1+x^{2})} \, \mathrm dx \\ &= \frac{1}{2} \, \Re \int_{-\infty}^{\infty} e^{iax} \, \frac{(1+ix)^{n-1}}{(1-ix)^{n+1}} \, \mathrm dx. \end{align}$$

But the integrand is analytic in the upper half of the complex plane, and by Jordan's lemma (or the estimation lemma) the integral $$\int_{R\exp(i[0, \pi])} e^{iaz} \, \frac{(1+iz)^{n-1}}{(1-iz)^{n+1}} \, \mathrm dz$$ vanishes as $R \to \infty$.

Therefore, $$\int_{0}^{\infty} \frac{\cos \left(ax + 2 n \arctan x \right)}{1+x^{2}} \, \mathrm dx =0. $$

$\endgroup$

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.