For the purpose of my research on persistent random walks I need to compute the inverse Laplace transform of $$ F(s)=\frac{\mathrm e^{-b\sqrt{s^2-1}}}{s^2-1}.$$ I looked up in tables of integral transforms such as Erdélyi et al., Gradsteyn & Ryzhik, Prudnikov et al. with no success. So I have looked for a proper contour for the Bromwich integral. I came up with this one

Double keyhole contour

$\color{green}A$ is the integral we want to compute $$f(t)=\frac1{2\pi\mathrm i}\int_{\gamma-\mathrm i\infty}^{\gamma+\mathrm i\infty}\mathrm e^{st}\frac{\mathrm e^{-b\sqrt{s^2-1}}}{s^2-1}\mathrm ds,$$ $\color{blue}B$ and $\color{blue}K$ vanish when the radius of the large circle goes to infinity. The contributions of $\color{brown}F$ and that of $\color{brown}D$ and $\color{brown}H$ are given by simple poles and yield together a contribution of $\sinh t$. The contribution of $\color{red}C$ and $\color{red}J$ cancel each other.

To compute the contributions of $\color{red}E$ and $\color{red}G$, I set $s=-\cos\theta+\mathrm i\varepsilon$ for $\color{red}E$ and $s=\cos\theta-\mathrm i \varepsilon$ for $\color{red}G$. I end up with the following integral $$\color{red}{E+G}\to\frac1{\pi\mathrm i}\;\text{vp.}\int_0^\pi \frac{\mathrm e^{\mathrm ib\sin\theta}}{\sin\theta}\sinh(t\cos\theta)\mathrm d\theta\tag{1}$$ that I didn't manage to reduce to a handier result.

Can someone tell me if I'm the right track and perharps help me to come up with a better expression of (1) ?

  • $\begingroup$ Maybe it helps to take the derrivative of (1) wrt. to $b$. Then you get rid of the sin in the denominator and get something that seems to go in the direction of Bessel functions. $\endgroup$
    – Urgje
    Sep 21 '15 at 11:08
  • $\begingroup$ @Urgje. Thank you for your comment. Unfortunately I already know the Laplace inverse of $s\mapsto\mathrm e^{-b\sqrt{s^2-1}}/\sqrt{s^2-1}$, it is $t\mapsto I_0(\sqrt{t^2-b^2})\Theta(t-b)$ that I can't integrate with respect to $b$ under this form. $\endgroup$
    – Tom-Tom
    Sep 21 '15 at 11:48

I think you played a little fast and loose with the contributions over $E$ and $G$. In considering the contour integral over $E$, I let $z=e^{i \pi} x$ and, over $G$, I let $z=e^{-i \pi} x$. The integrals I get, after taking appropriate limits, is

$$e^{i \pi} \int_1^{-1} dx \frac{e^{i b \sqrt{1-x^2}}}{-(1-x^2)} e^{-x t} + e^{-i \pi} \int_{-1}^1 dx \frac{e^{-i b \sqrt{1-x^2}}}{-(1-x^2)}e^{-x t}$$

Note that the $\pm i$ in the exponentials comes from the assignment of $\sqrt{s-1} = \sqrt{e^{\pm i \pi} (x+1)}$ from the respective branch. The result is

$$-i 2 \int_{-1}^1 dx \frac{\sin{b \sqrt{1-x^2}}}{1-x^2}e^{-x t}$$

This integral converges, although I have not been able to unwind it into a closed form. (I suspect it is some Bessel of complex argument.)

Anyway, I get that

$$\frac1{i 2 \pi} \int_{c-i \infty}^{c+ i \infty} ds \frac{e^{-b \sqrt{s^2-1}}}{s^2-1} e^{s t} = \sinh{t} + \frac1{\pi} \int_{-1}^1 dx \frac{\sin{b \sqrt{1-x^2}}}{1-x^2}e^{-x t}$$

  • $\begingroup$ Thanks a lot ! Indeed, I used a bad parametrization for $\color{red}E$, I should have used $s=\cos\theta+\mathrm i\varepsilon$ instead. Finding a closed form might be impossible, but the expression you have found will be definitely useful. $\endgroup$
    – Tom-Tom
    Sep 22 '15 at 7:50

I have a copy of G.E.Roberts and H.Kaufman "Table of Laplace Transforms", 1966.

On page 252, Item 3.2.55 is a more complex example but might apply if $v = 0$.

Inverse of $(s + (s^2 - a^2)^{1/2})^v \exp(-b\sqrt{s^2-a^2})/\sqrt{s^2 - a^2}$ is given as:

$0$ for $0 < t < b$

and $a^v ((t-b)/(t+b))^{v/2} I_v(a\sqrt{t^2-b^2})$ for $t > b$

With the normal restrictions that $b>0$, $|\operatorname{Re}(v)| < 1$, $\operatorname{Re} s > |\operatorname{Re} a|$.

$I_v$ is the modified Bessel function of order $v$.

I assume that the expressions would be valid for $v = 0$.

  • $\begingroup$ Thank you for your answer. I also found this inverse in the tables of Erdélyi et al (Bateman) or Prudnikov et al. But in this case, the denominator is $\sqrt{s^2-a^2}$ and not $s^2-a^2$ as in my original question (see my reply to Urgje below the question). $\endgroup$
    – Tom-Tom
    Nov 25 '15 at 8:18

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.