This question is about inverse Laplace transform $\mathscr{L}^{-1}:s\rightarrow t$. Although I was not able to find appropriate contour to invert $K_0 \left(r s\right)$, I somehow know that $$\mathscr{L}^{-1}\{K_0 \left(r s\right)\}=\frac{\theta(t-r)}{\sqrt{t^2-r^2}}.$$

How do one show that statement rigorously? The main part of my question is: Is it possible, similarly, to express a similar inverse fourier transform $$\mathscr{L}^{-1}\left\{K_0 \left(r \sqrt{s^2-1}\right)\right\}$$

in terms of elementary functions? Thank you for suggestions. ($K_0$ is the modified Bessel function of second kind, $\theta$ is just Heaviside theta, $r>0$ is a positive real parameter).

Important note: The function $K_0$ in the second laplace transform is ill-defined for $s\in (0,1)$ and is ment to represent only its real part, equivalently, using common identities for Bessel functions,

$$K_0 \left(r \sqrt{s^2-1}\right) = -\frac{\pi}{2}Y_0\left(r\sqrt{1 - s^2}\right)$$

which extends the domain of the original function to $s\in (0,1)$.

  • $\begingroup$ I am not getting a closed form, but something like $$\frac{\pi}{2} \int_{-1}^1 dx \, J_0{\left ( r \sqrt{1-x^2} \right )} \, e^{-t x} $$ $\endgroup$
    – Ron Gordon
    Jul 25 '18 at 19:15
  • $\begingroup$ This looks very nice. However, it cannot be true since if you do the Laplace transform on it and compare with the original function, the graphs looks differently for $s>1$ - and even for $s<1$ they do not coincide. $\endgroup$
    – Machinato
    Jul 25 '18 at 19:39

The result given by Mariusz can be verified as follows. The substitution $t = r \cosh \tau$ gives $$F(s) = \int_0^\infty \frac {\cosh \sqrt {t^2 - r^2}} {\sqrt {t^2 - r^2}} \theta(t - r) e^{-s t} dt = \int_0^\infty e^{-r s \cosh \tau} \cosh(r \sinh \tau) d\tau, \\ \operatorname{Re} s > 1.$$ Converting $\cosh(r \sinh \tau)$ to $e^{\pm r \sinh \tau}$ and writing $a \sinh \tau + b \cosh \tau$ as $A \cosh(\tau + \tau_0)$ gives $$F(s) = \frac 1 2 \int_0^\infty e^{A \cosh(\tau - \tau_0)} d\tau + \frac 1 2 \int_0^\infty e^{A \cosh(\tau + \tau_0)}, \\ A = -r \sqrt {s^2 - 1}, \; \tau_0 = \operatorname{arcsinh} \frac 1 {\sqrt {s^2 - 1}}.$$ Since $\cosh$ is even, $$F(s) = \frac 1 2 \int_{-\tau_0}^\infty e^{A \cosh \tau} d\tau + \frac 1 2 \int_{\tau_0}^\infty e^{A \cosh \tau} d\tau = \int_0^\infty e^{A \cosh \tau} d\tau = \\ K_0(-A).$$


With CAS help:

$$\mathcal{L}_s^{-1}\left[K_0\left(r \sqrt{s^2-1}\right)\right](t)=\frac{\theta (t-r) \cosh \left(\sqrt{t^2-r^2}\right)}{\sqrt{t^2-r^2}}$$

for $r>0$

Maple code:

`assuming`([inttrans:-invlaplace(BesselK(0, r*sqrt(s^2-1)), s, t)], [r > 0])

  • $\begingroup$ The square root inside $\cosh$ is missing in the MathJax formula. $\endgroup$
    – Maxim
    Jul 26 '18 at 2:34
  • $\begingroup$ @Maxim. Thanks,it was a typo. $\endgroup$ Jul 26 '18 at 7:24
  • $\begingroup$ This is miraculous! Thank you, this is what I was looking for. $\endgroup$
    – Machinato
    Jul 26 '18 at 8:01

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.