7
$\begingroup$

How to find an asymptotic expansion, for $\epsilon$ near $0$, of the following integral $$ I(\epsilon):=\int_{0}^{+\infty}\frac 1{\sinh^2 (\epsilon \sqrt{x^2+1}) } {\rm d}x. $$ As $\epsilon \rightarrow 0$, I have easily obtained $$ I(\epsilon) \sim \frac{\pi}{2\:\epsilon^2}. $$

Some clear steps leading to an extended expansion would be appreciated.

$\endgroup$
11
$\begingroup$

My calculation shows that

$$I(\epsilon) := \int_{0}^{\infty} \frac{dx}{\sinh^{2}(\epsilon\sqrt{x^{2}+1})} = \frac{\pi}{2\epsilon^{2}} - \frac{1}{\epsilon} + \pi \epsilon \sum_{n=1}^{\infty} \frac{1}{(\pi^{2}n^{2} + \epsilon^{2})^{3/2}} \tag{1}. $$

In particular, if we expand the infinite sum on the RHS, we get

$$ I(\epsilon) = \frac{\pi}{2\epsilon^{2}} - \frac{1}{\epsilon} + \sum_{n=0}^{\infty} \binom{-3/2}{n} \frac{\zeta(2n+3)}{\pi^{2n+2}} \epsilon^{2n+1}. $$

Indeed, from the following expansion

$$ \frac{1}{\sinh^{2}z} = \sum_{n=-\infty}^{\infty} \frac{1}{(z-i\pi n)^{2}} = \frac{1}{z^{2}} + 2 \sum_{n=1}^{\infty} \frac{z^{2} - \pi^{2}n^{2}}{(z^{2} + \pi^{2} n^{2})^{2}}, $$

it follows from term-wise integration that

\begin{align*} \int_{0}^{R} \frac{dx}{\sinh^{2}(\epsilon\sqrt{x^{2}+1})} = \frac{\arctan R}{\epsilon^{2}} + \sum_{n=1}^{\infty} &\Bigg( \frac{2\epsilon \arctan \left( R\epsilon \big/ \sqrt{\pi^{2}n^{2} + \epsilon^{2}} \right)}{(\pi^{2}n^{2} + \epsilon^{2})^{3/2}} \\ &\quad + \frac{2}{R}\frac{1}{\pi^{2}n^{2}+\epsilon^{2}} \\ &\quad - \frac{2(R^{2}+1)}{R(\pi^{2}n^{2} + (R^{2}+1)\epsilon^{2})} \Bigg). \tag{2} \end{align*}

Here, term-wise integration is possible from Fubini's theorem together with the following estimate:

$$ \int_{0}^{R} \left| \frac{\epsilon^{2}(x^{2}+1) - \pi^{2}n^{2}}{(\epsilon^{2}(x^{2}+1) + \pi^{2} n^{2})^{2}} \right| = \frac{\arctan \left( R\epsilon \big/ \sqrt{\pi^{2}n^{2} + \epsilon^{2}} \right)}{\epsilon\sqrt{\pi^{2}n^{2}+\epsilon^{2}}} \lesssim_{\epsilon, R} \frac{1}{n^{2}}. $$

(Notice that the arctan term also contributes to order $n^{-1}$. It means that this argument fails if we consider $R = \infty$. This is why we consider proper integral first.)

Finally, taking $R \to \infty$ to (2) yields (1). When doing this, the only non-trivial calculation is to check that

$$ \lim_{R\to\infty} \sum_{n=1}^{\infty} \frac{2(R^{2}+1)}{R(\pi^{2}n^{2} + (R^{2}+1)\epsilon^{2})} = \frac{1}{\epsilon}. $$

But this follows from the squeezing lemma combined with the following inequality

$$ C(1, R) \leq \sum_{n=1}^{\infty} \frac{2(R^{2}+1)}{R(\pi^{2}n^{2} + (R^{2}+1)\epsilon^{2})} \leq C(0, R), $$

where

\begin{align*} C(a, R) &:= \int_{a}^{\infty} \frac{2(R^{2}+1)}{R(\pi^{2}x^{2} + (R^{2}+1)\epsilon^{2})} \, dx \\ &= \frac{\sqrt{R^{2}+1}}{R\epsilon} \frac{\arctan\left( \epsilon\sqrt{R^{2}+1} \big/ a\pi \right)}{\pi/2}. \end{align*}

$\endgroup$
  • $\begingroup$ sos440: Perfect! +1. $\endgroup$ – Olivier Oloa Dec 29 '14 at 14:34
  • $\begingroup$ on the third equation there is a typo: the limits should be from $n=1$ to $n=\infty$ $\endgroup$ – glS Feb 26 '15 at 18:03
  • $\begingroup$ @glance, Thank you! I fixed it. $\endgroup$ – Sangchul Lee Feb 26 '15 at 22:12
1
$\begingroup$

What about considering more terms of the Weierstrass product of $\sinh$?

The first term of the asymptotics is easily recovered from: $$ I = \int_{0}^{\pi/2}\frac{d\theta}{\cos^2\theta\sinh^2\frac{\epsilon}{\cos\theta}}$$ by approximating $\sinh t$ with $t$. If we approximate $\sinh t$ with $t\left(1+\frac{t^2}{\pi^2}\right)$ or, even better, with $ t\, e^{t^2/6}$, we get: $$ I \approx \frac{1}{\varepsilon^2}\int_{0}^{\pi/2}\exp\left(-\frac{\epsilon^2}{3\cos^2\theta}\right)\,d\theta=\frac{\pi}{2\epsilon^2}\operatorname{Erfc}\left(\frac{\epsilon}{\sqrt{3}}\right)\approx\color{red}{\frac{\pi}{2\epsilon^2}\left(1-\frac{2\epsilon}{\sqrt{3\pi}}\right)}.$$

$\endgroup$
  • $\begingroup$ What is the meaning here of the symbol $\approx$? $\endgroup$ – Olivier Oloa Dec 29 '14 at 1:03
  • $\begingroup$ @OlivierOloa: approximated equivalence up to $O(1)$. $\endgroup$ – Jack D'Aurizio Dec 29 '14 at 1:06

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.