5

The question asks to show that $$\mathcal{I}=\int_0^\infty \dfrac{x\tan(ax)}{x^2+b^2}\mathrm dx=\frac{\pi}{e^{2ab}+1}$$ for $a>0$, $b>0$.

I found this on the internet but searched using Approachzero but found no question. I have been trying this to evaluate using Real methods. Substituting $ax=u$, I got \begin{align}\mathcal{I}&=a^2\int_0^\infty \frac{u\tan u}{u^2+a^2b^2}\, \mathrm du\\&=\frac{a^2}{2}\int_0^\infty \tan u\, \mathrm d(\ln(a^2b^2+u^2))\end{align} Applying Integration by part here seems problematic since the first term seems to diverge. Probably this method cannot be applied. Then I searched Wolfram Alpha to find ''Computational time exceeded.'' With Wolfram, I found one result that $$\mathcal{M}\bigg(\frac{1}{x^2+b^2}\bigg)=\int_0^\infty \frac{x^{s-1}}{x^2+b^2}\mathrm dx=\frac{\pi}{2}b^{s-2}\csc\bigg(\frac{\pi s}{2}\bigg)$$ , where $\mathcal{M}$ denotes the Mellin Transform. I have just learned some basics of Mellin Transform so I don't know how this can help with this question. How can I solve this question or at least can you give some hints?

Edit: After comments by @Dr. Wolfgang Hintze, I searched the book ''Table of Integrals, Series and Products by I S Gradshteĭn'' book and found the same integral without proof.

  • 2
    The integral doesn't exist as written, there is a periodic $x^{-1}$ singularity. – Ninad Munshi Jul 01 '21 at 09:51
  • If the integral does not exist, Wolfram should be able to point out that but it just says ''Computational time exceeded''. That raises suspicion. –  Jul 01 '21 at 09:55
  • 1
    That's not a good argument that a software can't handle it. Proving this integral diverges is an exercise for first year calculus students. – Ninad Munshi Jul 01 '21 at 09:56
  • Yes, you seem to be correct. This integral diverges. Probably the question is wrong. –  Jul 01 '21 at 10:03
  • 1
    That's not necessarily the case, either. My original objection was the integral doesn't exist as written, but maybe its principal value is given by this formula. But principal value was not denoted on the formula. That or the problem intended something different. – Ninad Munshi Jul 01 '21 at 10:07
  • The problem comes from $\tan$ which does not exist on the whole $\left[0;+\infty\right[$ – Atmos Jul 01 '21 at 10:11
  • 3
    This integral is given in Gradshteyn-Ryshik 3.749.1 without comments. Also $\int_0^\infty \dfrac{x\cot(ax)}{x^2+b^2}\mathrm dx=\frac{\pi}{e^{2ab}-1}$ They are cited from "Gröbner W., Hofreiter N., Integraltafel, Bestimmte Integrale, Springer-Verlag, Wien und Innsbruck, 1958" – Dr. Wolfgang Hintze Jul 02 '21 at 12:30
  • @Dr.WolfgangHintze Thanks for the comment! I didn't know that. –  Jul 02 '21 at 12:52

1 Answers1

6

Due to the simple poles of $\tan(ax)$ and the slow decay of $\frac{x}{x^2+b^2}$, the integral should be interpreted in a very careful way. So here is one way to make sense of the integral:

Let $(R_k)_{k\geq1}$ be an increasing sequence in $(0, \infty)$ satisfying the following conditions:

  • $R_k \to +\infty$ as $k \to \infty$, and
  • there exists $\delta > 0$ such that $\lvert R_k - \frac{\pi}{2a}(2n-1) \rvert \geq \delta $ for any $k, n \geq 1$. That is, each $R_k$ is at least at a distance $\delta$ from any of the poles of $\tan(ax)$. (For example, $R_k = \frac{\pi k}{a}$ works.)

Then $$ \mathcal{I} := \lim_{k\to\infty} \operatorname{PV}\!\!\int_{0}^{R_k} \frac{x \tan(ax)}{x^2+b^2} \, \mathrm{d}x = \frac{\pi}{e^{2ab} + 1}. $$

For the interested user, here is the graph of $R \mapsto \operatorname{PV}\!\!\int_{0}^{R} \frac{x \tan(ax)}{x^2+b^2} \, \mathrm{d}x $ when $a = b = 1$:

Graph of the principal value


Sketch of the 1st Proof. We find that $2\mathcal{I}$ can be written as

\begin{align*} 2\mathcal{I} &= \lim_{k\to\infty} \operatorname{PV}\!\!\int_{-a R_k}^{a R_k} \frac{y \tan y}{y^2+(ab)^2} \, \mathrm{d}x \tag{$y = ax$}\\ &= \lim_{N\to\infty} \sum_{k=-N}^{N} \int_{-\pi/2}^{\pi/2} \frac{1}{2} \left( \frac{1}{y + k\pi + iab} + \frac{1}{y + k\pi - iab} \right) \tan y \, \mathrm{d}y \end{align*}

Then by utilizing the partial fraction decomposition of the cotangent,

$$ \cot z = \lim_{N\to\infty} \sum_{k=-N}^{N} \frac{1}{z - \pi k}, $$

we get

$$ 2\mathcal{I} = \int_{-\pi/2}^{\pi/2} \frac{\cot(y + iab) + \cot(y - iab)}{2} \tan y \, \mathrm{d}y. $$

By utilizing trigonometric identities, we find that the integrand simplifies as follows:

$$ \frac{\cot(y + iab) + \cot(y - iab)}{2} \tan y = \frac{\sin^2 y}{\cosh^2(ab) - \cos^2 y}. $$

Therefore

\begin{align*} 2\mathcal{I} &= \int_{-\pi/2}^{\pi/2} \frac{\sin^2 y}{\cosh^2(ab) - \cos^2 y} \, \mathrm{d}y \\ &= \int_{-\infty}^{\infty} \left( \frac{1}{t^2+1} - \frac{\sinh^2(ab)}{\cosh^2(ab) t^2 + \sinh^2(ab)} \right) \, \mathrm{d}y \tag{$t = \tan y$} \\ &= \pi - \pi \tanh(ab) = \frac{2\pi}{e^{2ab} + 1}. \end{align*}

Therefore the desired equality follows.


Sketch of the 2nd Proof. Let me only sketch the proof of this equality. We will instead study

$$ 2\mathcal{I} = \lim_{k\to\infty} \operatorname{PV}\!\!\int_{-R_k}^{R_k} \frac{x \tan(ax)}{x^2+b^2} \, \mathrm{d}x. $$

To this end, let $0 < \epsilon < b < T$ and $R > 0$. If $\Gamma$ is a counter-clockwise rectangular contour with the four corners $\pm R + i\epsilon$ and $ \pm R + iT$, then

$$ \int_{\Gamma} \frac{z \tan(az)}{z^2+b^2} \, \mathrm{d}z = 2\pi i \, \underset{z=ib}{\mathrm{Res}} \, \frac{z \tan(az)}{z^2+b^2} = -\pi \tanh(ab). $$

Moreover, as $R \to \infty$ along the sequence $(R_k)_{k \geq 1}$ and $\epsilon \to 0^+$,

$$ \int_{-R+i\epsilon}^{R+i\epsilon} \frac{z \tan(az)}{z^2+b^2} \, \mathrm{d}z \quad \longrightarrow \quad 2\mathcal{I} - \pi i \lim_{k\to\infty} \sum_{\substack{\xi : |\xi| < R; \\ \tan(a\xi) = 0 }} \underset{z=\xi}{\mathrm{Res}} \, \frac{z \tan(az)}{z^2+b^2} = 2\mathcal{I}, $$

since

$$ \sum_{\substack{\xi : |\xi| < R; \\ \tan(a\xi) = 0 }} \underset{z=\xi}{\mathrm{Res}} \, \frac{z \tan(az)}{z^2+b^2} = \sum_{\substack{\xi : |\xi| < R; \\ \tan(a\xi) = 0 }} \frac{a \xi \sec^2(a\xi)}{\xi^2+b^2} = 0 $$

by the symmetry. Finally, consider the integral

$$ \int_{\gamma} \frac{z \tan(az)}{z^2+b^2} \, \mathrm{d}z $$

along the polygonal path $\gamma$ from $R+i\epsilon$ to $R+iT$ to $-R+iT$ to $-R+i\epsilon$. To study the behavior of this integral, we make the following observations:

  • $ \tan(z) \to i $ uniformly as $\operatorname{Im}(z) \to +\infty$.

  • The argument change along $\gamma$, viewed from any given point, is $\pi + o(1)$ as $R, T \to \infty$ and $\epsilon \to 0^+$

So we expect that

$$ \int_{\gamma} \frac{z \tan(az)}{z^2+b^2} \, \mathrm{d}z \approx \frac{i}{2} \int_{\gamma} \left( \frac{1}{z+ib} + \frac{1}{z-ib} \right) \, \mathrm{d}z \approx \frac{i}{2} (i\pi + i\pi) = -\pi $$

and the error in the above approximate equalities will vanish along the limit. Therefore

$$ 2\mathcal{I} - \pi = \lim_{\substack{R_k\to\infty \\ T \to \infty \\ \epsilon \to 0^+}} \int_{\Gamma} \frac{z \tan(az)}{z^2+b^2} \, \mathrm{d}z = -\pi \tanh(ab), $$

and solving this for $\mathcal{I}$ yields the desired equality.

Sangchul Lee
  • 167,468
  • 1+ Excellent! There is an interesting paper on principal value integrals with application to some integrals in Gradsteyn-Ryzhik: https://arxiv.org/ftp/arxiv/papers/1503/1503.01175.pdf It also contains the intergral in question in the OP, which was already treated by G.H. Hardy back in 1902. – Dr. Wolfgang Hintze Jul 03 '21 at 07:00
  • Principal value problems can also be treated numerically in Mathematica: in our case we can write with a=b=1 and 100 summands: {Sum[NIntegrate[y Tan[y]/(y^2 + 1), {y, [Pi] n, [Pi] (n + 1)}, Method -> PrincipalValue, Exclusions -> Evaluate[Table[[Pi]/2 (2 k + 1), {k, 0, n}]]], {n, 0, 100}], [Pi]/(E^2 + 1) // N} resulting in {0.372303, 0.374487} – Dr. Wolfgang Hintze Jul 03 '21 at 08:02
  • Did you consider $\int_{0}^{\infty} \frac{z \sec(a z)}{b^2+z^2},dz$? – Dr. Wolfgang Hintze Jul 03 '21 at 11:33
  • @Dr.WolfgangHintze, Thank you for your comments! As for your last comment, do you mean $$\int_{0}^{\infty}\frac{x \color{red}{\cot(ax)}}{b^2+x^2},\mathrm{d}x \quad ?$$ – Sangchul Lee Jul 06 '21 at 07:00
  • @ Sangchul Lee You're welcome. No, see my comment to the OP, and I have just made a question out of my last comment here: https://math.stackexchange.com/questions/4191727/give-a-meaning-to-the-integral-int-0-infty-frac-secx1x2-dx – Dr. Wolfgang Hintze Jul 06 '21 at 13:17