3

I am stuck with the following integral. Does it converge?

$$ \int_{0}^{\infty}\left(J_1(x)^2+J_1(x)J_1(x)^{''}\right)\text{d}x $$

According to tables I find that the first term is divergent, so I assume it is overall divergent but it could very well be that the second term tames it.

Edit: It seems that only for the second term to be $a J_1(x)J_1(x)^{''}$ with $a=1$ the integral can be shown to converge, in other cases the divergent part $J_1(x)^2$, for example when $a=1/2$, cannot be substracted by the proposed identities, right?

Hamurabi
  • 629

3 Answers3

5

Sometimes, simpler is better ...

So, let's just start with the identity that Chappers provided, namely

$$\int_{0}^{\infty} (J_1(x)^2+J_1(x)J_1(x)'')\,dx= \frac{1}{4}\int_0^{\infty} J_1(x) (J_1(x)+J_3(x)) \, dx $$

Then, let's substitute the recurrence relation

$$J_2(x)=\frac{x}{4} (J_1(x)+J_3(x))$$

to find that

$$\int_{0}^{\infty} (J_1(x)^2+J_1(x)J_1(x)'') \, dx = \int_0^{\infty} \frac{J_1(x)J_2(x)}{x} \, dx $$

Finally, let's exploit the orthogonality relationship

$$\int_0^{\infty} \frac{J_{\alpha}(x)J_{\beta}(x)}{x} \, dx =\frac{2}{\pi}\frac{\sin (\frac{\pi}{2}(\alpha-\beta))}{\alpha^2-\beta^2}$$

with $\alpha =2$ and $\beta =1$ to obtain

$$\int_0^{\infty} \frac{J_{2}(x)J_{1}(x)}{x} \, dx =\frac{2}{\pi}\frac{\sin (\frac{\pi}{2}(2-1))}{2^2-1^2}=\frac{2}{3\pi}$$

which recovers the expected result!

Chappers
  • 67,606
Mark Viola
  • 179,405
4

Since $J_1$ is a solution of the Bessel differential equation: $$ x^2 f'' + x f' + x^2 f = f \tag{1}$$ by exploiting integration by parts we have that: $$ \int_{0}^{+\infty}\left(J_1(x)^2+J_1(x)\,J_1''(x)\right)\,dx = \frac{1}{2}\int_{0}^{+\infty}\left(\frac{J_1(x)}{x}\right)^2\,dx \tag{2} $$ so we just need to recall that the Fourier transform of $\frac{J_1(x)}{x}$ is given by: $$ \mathcal{F}\left(\frac{J_1(x)}{x}\right)(t) = \sqrt{\frac{2}{\pi}}\sqrt{1-t^2}\cdot\mathbb{1}_{(-1,1)}(t)\tag{3}$$ to be able to state: $$ \int_{0}^{+\infty}\left(J_1(x)^2+J_1(x)\,J_1''(x)\right)\,dx = \frac{1}{\pi}\int_{0}^{1}\left(1-t^2\right)\,dt = \frac{2}{3\pi}\tag{4}$$ as a consequence of Parseval's theorem.

Jack D'Aurizio
  • 353,855
  • 1
    As a colorful note, this is my second problem of the day for which the answer is given by $\frac{2}{3\pi}$, the first one being http://math.stackexchange.com/questions/1243160/expected-area-of-triangle-formed-by-three-random-points-inside-unit-circle/1243173#1243173 – Jack D'Aurizio Apr 20 '15 at 14:53
  • 2
    Funny, who would have thought that "The expected area of a triangle formed by three randomly chosen points inside the unit circle" would equal $\int_0^{+\infty} (J_1(x)^2+J_1(x)J_1''(x)),dx$? :) – mickep Apr 20 '15 at 15:10
  • 1
    +1 Back! Your solutions are most often full of creativity and breadth. – Mark Viola Apr 20 '15 at 15:17
  • @ Jack D'Aurizio: Do you have a comment on the edit for me? That would be cool. Thanks! – Hamurabi Apr 22 '15 at 12:02
  • 1
    @Hamurabi: yes, you are right, the last case the integral is converging only when $\alpha=1$. That follows from the fact that $J_1$ decays like $\frac{1}{\sqrt{x}}$ (it is straighforward to prove by exploiting the Fourier transform), hence $a=1$ is needed in order to grant integrability. – Jack D'Aurizio Apr 22 '15 at 16:15
3

Mathematica says it converges to $2/(3\pi)$. You can probably get this from this identity: $$ \int_{0}^{\infty}\frac{J_{\mu}(at)J_{\nu}(at)}{t^{\lambda}}dt= \frac{(\frac{1}{2}a)^{\lambda-1} \Gamma\left(\frac{1}{2}\mu+\frac{1}{2}\nu-\frac{1}{2}\lambda+\frac{1}{2}\right) \Gamma(\lambda)}{ 2\Gamma\left(\frac{1}{2}\lambda+\frac{1}{2}\nu- \frac{1}{2}\mu+\frac{1}{2}\right) \Gamma\left(\frac{1}{2} \lambda+\frac{1}{2}\mu-\frac{1}{2}\nu+\frac{1}{2}\right) \Gamma\left(\frac{1}{2}\lambda+\frac{1}{2}\mu+\frac{1}{2}\nu+\frac{1}{2}\right)} $$ with careful limits as $\lambda \to 0$, and the expression $$ J_1''(x) = \frac{1}{4} (J_3(x)-3 J_1(x)) $$ reduces to the integral to one of the form $$ \frac{1}{4}\int_0^{\infty} J_1(x) (J_1(x)+J_3(x)) \, dx, $$ which you can then do with the linked identity.

Chappers
  • 67,606
  • I posted a question asking for a proof of the above identity weeks ago, but found no satisfactory answer. – Lucian Apr 20 '15 at 14:11
  • @Lucian: maybe it is possible to use the argument in my answer to give an answer to that question, too. It looks reasonable that by switching to Fourier transforms we get a product of values for some Beta functions. – Jack D'Aurizio Apr 20 '15 at 14:55
  • @Lucian I'm surprised that Watson's book didn't contain a satisfactory answer! – Chappers Apr 20 '15 at 15:38
  • Thanks! And a "+1" Back! I was in the midst of solving this and had arrived at the intermediate identity when you posted. After pursuing a bit more I arrived at quite a simple way forward. – Mark Viola Apr 20 '15 at 15:48
  • @Chappers: That book is well above my pay-grade. :-$)$ I was hoping for an easier approach, one that even the likes of me could understand. :-$)$ – Lucian Apr 20 '15 at 16:34