12

I came across the following non-trivial improper integral while I was elaborating on a fluid mechanical problem involving porous media: $$ f(r,z,a) = \int_0^\infty \frac{u^2}{\sqrt{u^2+a^2}} J_1(ru) e^{-z\sqrt{u^2+a^2}} \mathrm{d}u , $$ wherein $r$, $z$, and $a$ are positive real numbers. Physically, $a$ is a measure of the resistance of the porous medium against flow (impermeability coefficient.) Using a different solution route of the original problem (which does not involve integral transforms), I was able to find that $f(r,z,a)$ is actually given by $$ f(r,z,a) = \frac{r(1+aR)}{R^3} e^{-aR} , $$ with $R = \sqrt{r^2+z^2}$. I can check that this expression is correct by numerical substitution but I have no clue how this could be rigorously proven.

What I tried is to use Poisson's representation of the Bessel function but this does not lead to the desired result. It would be highly appreciated if someone here could be of help to clarify how this could be the case. Thank you!

In particular, it can readily be checked that in the limit case $a=0$ both expressions lead to the same value, namely, $$ f(r,z,0) = \frac{r}{\left( r^2+z^2\right)^{3/2}} . $$

Zacky
  • 30,116
Eulerian
  • 454
  • "Using a different solution route of the problem, I was able to find a closed-form" contradicts with "but I have no clue how this could be rigorously proven"? – MathFail May 31 '23 at 09:42
  • @MathFail The different route of the original problem does not involve integrals. The solution can be obtained easily. I want to build on the integral approach to be able to solve another problem which cannot be solved using this different route... I hope i am clear now! – Eulerian May 31 '23 at 09:46
  • 1
    More generally, $$\int_{0}^{\infty} J_{\mu}(ru) , \frac{K_{\nu} \left(z\sqrt{u^{2}+a^{2}} \right)}{(u^{2}+a^{2})^{\nu/2}} , u^{\mu +1} , \mathrm du = \frac{r^{\mu}}{z^{\nu}} \left(\frac{\sqrt{z^{2}+r^{2}}}{a} \right)^{\nu-\mu-1}K_{\nu-\mu-1} \left(a\sqrt{z^{2}+r^{2}} \right).$$

    See page 416 in A Treatise of the Theory of Bessel Functions. For your integral, let $\mu=1$ and $\nu = 1/2$.

    – Random Variable May 31 '23 at 17:20
  • The integral $\int\limits_0^\infty \frac{u^2}{\sqrt{u^2+a^2}}, J_1(ru), e^{-z\sqrt{u^2+a^2}},du$ is an order one Hankel transform $\int\limits_0^{\infty} u, f(u), J_1(r u),du$ where $f(u)=\frac{u}{\sqrt{u^2+a^2}}, e^{-z\sqrt{u^2+a^2}}$ which I believe can be simplified slightly to an order zero Hankel transform via the relationship $\int\limits_0^{\infty} u, g'(u), J_1(r u),du=-r \int\limits_0^{\infty} u, g(u), J_0(r u) , du$ where $g'(u)=f(u)$ and $g(u)=-\frac{1}{z}e^{-z \sqrt{u^2+a^2}}$. However I'm not sure how to evaluate either the original or simplified Hankel transform. – Steven Clark Jun 01 '23 at 00:19
  • Since the Hankel transform is its own inverse, another approach to the problem is to prove the order one Hankel transform $\int_0^\infty r, f(r,z,a), J_1(r u),dr=\frac{u}{\sqrt{u^2+a^2}}, e^{-z\sqrt{u^2+a^2}}$ which again I believe can be slightly simplified as the order zero Hankel transform $-u \int\limits_0^\infty r, (-\frac{e^{-a \sqrt{r^2+z^2}}}{\sqrt{r^2+z^2}}), J_0(r u),dr$. – Steven Clark Jun 01 '23 at 03:58
  • @RandomVariable citing a reference is sufficient for my purpose. Thanks for helping with this! – Eulerian Jun 01 '23 at 07:17
  • @StevenClark Thanks for the hint. I would just cite Watson's monograph for my purpose yet a full solution would of course be helpful! – Eulerian Jun 01 '23 at 07:18
  • 1
    You said you solve it by a different way, so can you share your approach? @Staufenberg – MathFail Jun 01 '23 at 07:23
  • @MathFail Using a fundamentally different approach which does not involve any integral. I believe that this is outside the scope of the question and require background and preliminaries from fluid physics. – Eulerian Jun 01 '23 at 12:05
  • hmmm, can you post some references? where do you find this proof? @Staufenberg – MathFail Jun 01 '23 at 12:09

2 Answers2

6

$$\int_0^\infty x\operatorname{J}_0(rx)\frac{e^{-z\sqrt{x^2+a^2}}}{\sqrt{x^2+a^2}}dx=\frac{e^{-a\sqrt{r^2+z^2}}}{\sqrt{r^2+z^2}}$$ $$\overset{\large \frac{d}{dr}}\Rightarrow \int_0^\infty x^2\operatorname{J}_1(rx)\frac{e^{-z\sqrt{x^2+a^2}}}{\sqrt{x^2+a^2}}dx=\frac{r\left(1+a\sqrt{r^2+z^2}\right)}{(r^2+z^2)^{3/2}}e^{-a\sqrt{r^2+z^2}}$$


In order to obtain the first result we can start by using the following identity: $$\int_0^\infty \frac{\cos t}{t^2+a^2}dt=\frac{\pi}{2}\frac{e^{-a}}{a}$$ $$\Rightarrow \mathcal J=\int_0^\infty x\operatorname{J}_0(rx)\color{blue}{\frac{e^{-z\sqrt{x^2+a^2}}}{\sqrt{x^2+a^2}}}dx$$ $$=\color{blue}{\frac{2z}{\pi}}\int_0^\infty x\operatorname{J}_0(rx) \color{blue}{\int_0^\infty \frac{\cos t}{t^2+z^2(x^2+a^2)}dt}dx$$ $$=\frac{2}{z\pi} \int_{0}^\infty \cos t\int_0^\infty \frac{\color{red}{x}\color{green}{\operatorname{J}_0(rx)}}{\color{red}{t^2/z^2+a^2+x^2}}dxdt$$


Now, for the inner integral we will employ the following property of the Laplace transform: $$\int_0^\infty f(x)g(x) dx = \int_0^\infty \mathcal L(f(x))\mathcal L^{-1}(g(x))dx$$ $$\Rightarrow \mathcal J=\frac{2}{z\pi} \int_{0}^\infty \cos t\int_0^\infty\frac{\color{red}{\cos\left(x\sqrt{t^2/z^2+a^2}\right)}}{\color{green}{\sqrt{r^2+x^2}}}dxdt$$ $$\overset{x\sqrt{t^2/z^2+a^2}\to x}=\frac{2}{z\pi} \int_{0}^\infty \cos t\int_0^\infty \frac{\cos x}{\color{blue}{\sqrt{r^2(t^2/z^2+a^2)+x^2}}}dxdt$$


Next, we will get rid of the square root from the denominator by using: $$\int_0^\infty e^{-ay^2}dy = \frac{\sqrt \pi}{2\sqrt a}$$

$$\Rightarrow \mathcal J=\frac{\color{blue}{2}\cdot 2}{z\pi\color{blue}{\sqrt \pi}} \int_{0}^\infty \cos t\int_0^\infty \cos x \color{blue}{\int_0^\infty \exp\left({-\left(r^2t^2/z^2+r^2a^2+x^2\right)y^2}\right)dy}dxdt$$

$$=\frac{4}{z\pi\sqrt \pi} \int_{0}^\infty \cos t\int_0^\infty \exp\left({-\left(r^2t^2/z^2+r^2a^2\right)y^2}\right)\color{red}{\int_0^\infty \cos x e^{-x^2y^2}dx}dydt$$


Finally, to get rid of the cosine terms, we will use the following identity (twice in a row): $$\int_0^\infty \cos x e^{-y^2 x^2}=\frac{\sqrt \pi}{2y}e^{-1/(4y^2)}$$ $$\Rightarrow \mathcal J=\frac{2}{z\pi}\int_{0}^\infty \cos t\int_0^\infty \exp\left({-\left(r^2t^2/z^2+r^2a^2\right)y^2}\color{red}{-\frac{1}{4y^2}}\right)\color{red}{\frac{1}{y}}dydt$$ $$=\frac{2}{z\pi}\int_{0}^\infty\exp\left({-r^2a^2y^2}-\frac{1}{4y^2}\right)\frac{1}{y}\color{blue}{\int_0^\infty \cos t e^{-r^2y^2t^2/z^2}dt}dy$$ $$=\frac{1}{\sqrt \pi}\int_{0}^\infty\exp\left({-r^2a^2y^2}-\frac{1}{4y^2}\color{blue}{-\frac{z^2}{4r^2y^2}}\right)\frac{1}{y}\color{blue}{\frac{1}{ry}}dy$$ $$\overset{\large y\to\frac{1}{y}}=\frac{1}{r\sqrt \pi}\int_0^\infty \exp\left(-\frac{y^2}{4}\left(1+\frac{z^2}{r^2}\right)-\frac{r^2a^2}{y^2}\right) dy=\frac{e^{-a\sqrt{r^2+z^2}}}{\sqrt{r^2+z^2}}$$ Where the last result was obtained using Glasser's Master Theorem.

Zacky
  • 30,116
  • Do you know, if there is a closed form also for the even powers? Like you calculated the moment for $x^1$ from which you can derive all the other odd powers, not the even though. – Diger Jun 30 '23 at 08:39
  • @Diger I'm afraid I don't understand exactly what you mean. Are you asking about the even powers and in the same time to have $J_0(rx)$? Or to have something like $x^{2n} J_{2n}(rx)$? – Zacky Jun 30 '23 at 10:17
  • Hey, I opened a new question: https://math.stackexchange.com/questions/4728113/bessel-type-integral – Diger Jun 30 '23 at 10:19
4

Differentiating the integral formula $$\int_{0}^{\infty} \exp \left(-\alpha x^{2}- \frac{\beta}{4x^{2}}\right) \, \mathrm dx = \frac{\sqrt{\pi}}{2\sqrt{\alpha}} \, e^{-\sqrt{\alpha \beta}} \, , \quad (\alpha, \beta > 0)$$ with respect to the parameter $\beta$ twice, we get $$ \int_{0}^{\infty} \frac{1}{x^{4}} \, \exp \left(-\alpha x^{2}- \frac{\beta}{4x^{2}} \right) \, \mathrm dx = 2 \sqrt{\pi} \, \frac{1+\sqrt{\alpha \beta}}{ \beta^{3/2}} \, e^{-\sqrt{\alpha \beta}}.$$

(Differentiation under the integral sign is justified since for any positive $\delta < \beta$, $ \frac{1}{x^{2}}\exp \left(-\alpha x^{2}- \frac{\beta}{4x^{2}}\right) $ is dominated by the integrable function $\frac{1}{x^{2}}\exp \left(-\alpha x^{2}- \frac{\delta}{4x^{2}}\right) $, and $\frac{1}{x^{4}} \, \exp \left(-\alpha x^{2}- \frac{\beta}{4x^{2}}\right) $ is dominated by the integrable function $\frac{1}{x^{4}} \, \exp \left(-\alpha x^{2}- \frac{\delta}{4x^{2}}\right) $.)

We also have $$\begin{align} \int_{0}^{\infty} u^{2} J_{1}(ru) e^{-t^{2}u^{2}} \, \mathrm du &= \int_{0}^{\infty} u^{2} e^{-t^{2}u^{2}} \sum_{m=0}^{\infty} \frac{(-1)^{m}}{m!(m+1)!} \left(\frac{ru}{2} \right)^{2m+1} \, \mathrm du \\ &= \sum_{m=0}^{\infty} \frac{(-1)^{m}}{m!(m+1)!} \left(\frac{r}{2} \right)^{2m+1} \int_{0}^{\infty} u^{2m+3} e^{-t^{2}u^{2}} \, \mathrm du \\ &= \sum_{m=0}^{\infty} \frac{(-1)^{m}}{m!(m+1)!} \left(\frac{r}{2} \right)^{2m+1} \frac{(m+1)!}{2t^{2m+4}} \\ &= \frac{r}{4t^{4}}\sum_{m=0}^{\infty} \frac{(-1)^{m}}{m!} \left(\frac{r^{2}}{4t^{2}} \right)^{m} \\ &= \frac{r}{4t^{4}} \, \exp \left(-\frac{r^{2}}{4t^{2}} \right). \end{align}$$

Therefore,

$ \begin{align} \int_{0}^{\infty} \, \frac{u^2}{\sqrt{u^2+a^2}} \, J_1(ru) \, e^{-z \sqrt{u^{2}+a^{2}}} \, \mathrm du &= \frac{2}{\sqrt{\pi}} \int_{0}^{\infty} u^{2} J_{1}(ru) \int_{0}^{\infty}\exp \left(-(u^{2}+a^{2})t^{2}-\frac{z^{2}}{4t^{2}} \right) \, \mathrm dt \, \mathrm du \\ &= \frac{2}{\sqrt{\pi}} \int_{0}^{\infty} \exp \left(-a^{2}t^{2}- \frac{z^{2}}{4t^{2}} \right) \int_{0}^{\infty} u^{2} J_{1}(ru) e^{-t^{2}u^{2}} \, \mathrm du \, \mathrm dt \\ &= \frac{r}{2\sqrt{\pi}} \int_{0}^{\infty} \frac{1}{t^{4}} \, \exp \left(-a^{2}t^{2} - \frac{r^{2}+z^{2}}{4t^{2}} \right) \, \mathrm dt \\ &= \frac{r}{2\sqrt{\pi}} 2 \sqrt{\pi} \, \frac{1+a\sqrt{r^{2}+z^{2}}}{(r^{2}+z^{2})^{3/2}} e^{-a\sqrt{r^{2}+z^{2}}} \\ &= r \, \frac{1+a \sqrt{r^{2}+z^{2}}}{(r^{2}+z^{2})^{3/2}} \, e^{-a \sqrt{r^{2}+z^{2}}} . \end{align}$

Eulerian
  • 454