7
$\begingroup$

Does the inverse Laplace transform, defined by the integral, \begin{equation} F(t) = \mathscr L_s^{-1}\left[\sqrt s\right](t) = \int_{c - i\infty}^{c + i\infty} \sqrt s ~e^{-st} ds \end{equation} exist?

Here are a few things I've considered in trying to answer this question myself:

  1. $\mathscr{L}[t^p] = \frac{\Gamma (p+1)}{s^{p+1}}$, provided that $p>-1$. The latter condition precludes the case of interest, which would correspond to $p=-3/2$.

  2. The square root function can be written as $ \sqrt s = e^{\frac{1}{2}\log s}$. So $\sqrt s$ is not holomorphic at the origin.

  3. This topic has been discussed previously here without clear resolution.

  4. Mathematica throws the transform back at me without a solution.

$\endgroup$

5 Answers 5

6
$\begingroup$

I have encountered the same problem! But let me emphasize, I am a geophysicist, not a mathematician!

My observations:

  1. I do not find these (seemingly simple) transforms in most tables of Laplace transforms (not in Abramowitz and Stegun, for example). So, this is, to me, a "red flag".

  2. I can do the complex-s-plane contour integral to obtain inverse LT of $\sqrt{s}$. It seems simple, there is a branch cut but no pole. The result is $\frac{-1}{2\sqrt{\pi}}t^{-3/2}$

  3. Another way to evaluate the inverse LT of $\sqrt{s}$ is to use the convolution theorem and multiply by the LT of a Heaviside step function (1/s). The product is 1/$\sqrt{s}$ and the inverse LT of this IS found in standard tables, 1/$\sqrt{\pi t}$. If this is, then, differentiated with respect to t, we obtain $\frac{-1}{2\sqrt{\pi}}t^{-3/2}$

  4. However, the forward Laplace transform of $t^{-3/2}$ does not converge! The problem, as might be suspected, is at t=0. This can be seen by calculating the indefinite integral and noting that the equation "blows up" at t = 0.

  5. So this leaves us with what seems to me to be a paradox. The LT does not exist, but its inverse does! To me, this is also a red flag, but I don't know what it means.

  6. Subsequent comment: My thinking, now, is that the forward LT of $t^{-3/2}$ diverges because the LT is not exactly designed for simple transformation of quantities that are singular at t=0. One must properly accommodate the "pre-zero 0-" boundary condition with the "post-zero 0+" function. This might be addressed with "generalized functions" that can be discontinuous (But I don't, for the moment, know the details). Furthermore, I note that Abramowitz and Stegun do not layout mathematics that is explicitly compatible with this, see, for example section 29.1.1, where the lower limit of the LT is not very specifically identified as coming from the positive or negative direction. This might explain why they don't list the transforms of relevance to this discussion. Once the boundary condition at t=0 is handled, then the term that diverges in the LT of $t^{-3/2}$ might cancel, the remaining term is $\sqrt{s}$ -- thus there is the possibility of both forward LT compatibility with inverse LT. ;-)

$\endgroup$
4
$\begingroup$

As you understand, in the cases you are interested in, the direct and inverse Laplace transforms do not exist in a straightforward way. However both make sense within theory of distributions. See

A. H. Zemanian, Generalized Integral Transformations, Interscience, New York, 1968;

Yu. A. Brychkov and A. P. Prudnikov, Integral transforms of generalized functions. Gordon and Breach, New York, 1989.

$\endgroup$
2
  • $\begingroup$ Thanks for your comment. I don't really understand, though. Why do the Laplace transforms not exist in the typical sense? $\endgroup$ Jul 30, 2013 at 16:36
  • 1
    $\begingroup$ If $p\le -1$, then the integral defining the Laplace transform of $t^p$ diverges near the origin. $\endgroup$ Jul 30, 2013 at 19:28
4
$\begingroup$

According to Wolfram Alpha, it is $\frac{1}{\Gamma(-1/2)x^{3/2}}$. This, of course, involves regularization, so to get the full answer, we have to subtract the divergent part so that its Laplace transform to converge.

So, the answer is $ \frac{\sqrt{i}}2\delta ^{(1/2)}(x)+\text{f.p.}\frac{1}{\Gamma(-1/2)x^{3/2}}$, following Urs Graf's "Introduction to Hyperfunctions and Their Integral Transforms", page 36:

enter image description here

More explanation, using the theory of divergent integrals.

The Fourier transform of $x^n$ is $2\pi i^{n}\delta ^{(n)}(s )$. This means that at $s=0$ we have

$\int_{-\infty}^\infty x^ndx=2\pi i^n\delta ^{(n)}(0 )$.

If the integral is taken from zero, we have half of that value:

$\int_{0}^\infty x^ndx=\pi i^n\delta ^{(n)}(0 )$.

But there is an equality of divergent integrals

$\int_0^\infty \frac1{x^n} dx=\frac1{(n-1)!}\int_0^\infty x^{n-2} dx$

it is based on the fact that the following Laplace-based transform preserves the area under the integral, defining the equivalence class of divergent integrals:

$\int_0^\infty f(x)dx=\int_0^\infty\mathcal{L}_t[t f(t)]\left(x\right)dx=\int_0^\infty\frac1x\mathcal{L}^{-1}_t[ f(t)]\left(x\right)dx$

So, we have:

$\int_0^\infty \frac1{x^{n+1}} dx=\frac1{n!}\int_0^\infty x^{n-1} dx=\frac\pi{n!}i^{n-1}\delta ^{(n-1)}(0)$

For $n=3/2$, and adding coefficient $\frac{3/2}{\Gamma(-1/2)}$ to the both sides we get

$\frac{3/2}{\Gamma(-1/2)}\int_0^\infty \frac1{x^{5/2}} dx=\frac1{\Gamma(-1/2)\Gamma(1/2)}\int_0^\infty \sqrt{x} dx=\frac{\sqrt{i}\delta ^{(1/2)}(0)}{\Gamma(-1/2)\Gamma(1/2)}=-\frac{\sqrt{i}}2\delta ^{(1/2)}(0)$

So to normalize our equation we should take this term with an opposite sign.

$\endgroup$
4
  • $\begingroup$ but the Laplace transform of $1/x^{3/2}$ diverges at $x=0$ (the $p>-1$ condition mentioned by the OP). $\endgroup$ Oct 1, 2021 at 11:28
  • $\begingroup$ @CarloBeenakker I have added the term that makes the integral finite. Thank you for pointing this out. $\endgroup$
    – Anixx
    Oct 1, 2021 at 12:24
  • $\begingroup$ @CarloBeenakker actually, the integral diverges if you integrate the Hadamard finite part. If you integrate the generalized function, including $0^+$, the integral converges. $\endgroup$
    – Anixx
    Oct 1, 2021 at 16:03
  • $\begingroup$ @CarloBeenakker I have added an alternative derivation. $\endgroup$
    – Anixx
    Oct 1, 2021 at 17:52
3
$\begingroup$

Let us work first formally. You want to "calculate'' $ f(s)=\int_0^{+\infty}t^{1/2} e^{-st} dt. $ It is indeed possible to give a meaning to $ F(\tau)=\int_0^{+\infty}t^{1/2} e^{-i\tau t} dt $ as the Fourier transform of the homogeneous distribution $t_+^{1/2}$. $F$ is an homogeneous distribution of degree $-\frac 32$ and in fact $$ F(\tau)=C(\tau-i0)^{-3/2}. $$ As a result, $ f(s)=F(s/i)=C(-is-i0)^{-3/2}=C_1\frac{d}{ds}(s^{-1/2}). $ It was obvious from the beginning since for $s> 0$ $$ f(s)=\int_0^{+\infty}x e^{-sx^2} 2xdx=-2\frac{d}{ds}\int_0^{+\infty} e^{-sx^2} dx =-\sqrt{π}\frac{d}{ds}(s^{-1/2}). $$

$\endgroup$
1
  • 2
    $\begingroup$ It seems that this question and its cousins will recur forever. :) $\endgroup$ Jul 30, 2013 at 21:49
-1
$\begingroup$

The inverse transform of $\sqrt{s}$ doesn't exist, because the integral representing it,

$$ \int\limits_{\gamma -i\infty }^{\gamma +i\infty }\sqrt{s}e^{st}\mathrm{d}s, $$

diverges as $s \rightarrow \gamma \pm i\infty$. People typically don't notice this, and just go ahead with "evaluating" this integral (by closing the contour, etc.) -- and end up with a result which doesn't make sense (as the resulting function doesn't have a forward transform).

$\endgroup$
3
  • 1
    $\begingroup$ Even if the inverse Laplace transform does not exist as a function, it can exist as a distribution. The answer depends on which function spaces you are interested in. $\endgroup$ Nov 3, 2014 at 11:23
  • $\begingroup$ "Even if the inverse Laplace transform does not exist as a function, it can exist as a distribution. The answer depends on which function spaces you are interested in." This is generally true, of course, but I don't understand the relevance to the question at hand. $\endgroup$ Nov 3, 2014 at 11:32
  • $\begingroup$ I've got something like $\frac{\sqrt{i}}2\delta ^{(1/2)}(x)+\text{f.p.}\frac{1}{\Gamma(-1/2)x^{3/2}}$ I can be mistaken in the coefficient before $\endgroup$
    – Anixx
    Oct 1, 2021 at 17:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.