11
$\begingroup$

Let $J_{0}(z)$ be the Bessel function of the first kind of order zero, and assume that $\alpha$ and $\beta_{m}$ are positive real parameters.

$J_{0}(z)$ is an even function that is real-valued along the real axis.

And when $z$ approaches infinity at a constant phase angle, $J_{0}(z)$ has the asymptotic form $$J_{0}(z) \sim \sqrt{\frac{2}{\pi z}} \cos \left(z-\frac{\pi}{4} \right), \quad |\arg(z)| < \pi. $$

So by integrating the entire function $$e^{i \alpha z} \prod_{m=1}^{n} J_{0}(\beta_{m}z) , \quad \sum_{m=1}^{n} \beta_{m} < \alpha,$$ around a contour that consists of the real axis and the infinitely large semicircle above it, it would seem to follow that $$\int_{0}^{\infty} \cos(\alpha x) \prod_{m=1}^{n} J_{0}(\beta_{m} x) \, \mathrm dx =0 \, , \quad \sum_{m=1}^{n} \beta_{m} < \alpha. \tag{1} $$

(For the cases $n=1$ and $n=2$, you would need to appeal to Jordan's lemma.)


Is there way to prove $(1)$ that doesn't involve contour integration?


EDIT:

A similar approach also shows that $$\int_{0}^{\infty} \frac{\cos(\alpha x)}{1+x^{2}} \prod_{m=1}^{n} J_{0}(\beta_{m} x) \, \mathrm dx = \frac{\pi e^{-a} }{2} \prod_{m=1}^{n}I_{0}(\beta_{m}), \quad \sum_{m=1}^{n} \beta_{m} \le \alpha,$$ where $I_{0}(z)$ is the modified Bessel function of the first kind of order zero.

$\endgroup$
8
  • $\begingroup$ The integration extends only over the positive reals. How do you intend to deform the contour so that it enclosed the upper half plane? $\endgroup$ – Mark Viola Feb 12 '16 at 22:41
  • $\begingroup$ @Dr.MV $J_{0}(x)$ is an even function $\endgroup$ – Random Variable Feb 12 '16 at 22:43
  • $\begingroup$ So, are you starting with $\frac12 \text{Re}\left(\int_{-\infty}^\infty e^{i\alpha x}\prod_{m=1}^{n}J_0(\beta_m x)\,dx\right)$? $\endgroup$ – Mark Viola Feb 12 '16 at 22:54
  • $\begingroup$ Oh sorry ... just saw the aside ... ;-)) $\endgroup$ – Mark Viola Feb 12 '16 at 23:01
  • $\begingroup$ @Dr.MV Yes. If $\int_{-\infty}^{\infty} e^{i\alpha x} \prod_{m=1}^{n} J_{0}(\beta_{m} x) \, dx =0$ under the conditions on the parameters stated, then we also know that $\int_{0}^{\infty} \cos(\alpha x)\prod_{m=1}^{n} J_{0}(\beta_{m} x) \, dx =0$ under the same conditions since $J_{0}(x)$ is an even function that is real-valued along the real axis. $\endgroup$ – Random Variable Feb 12 '16 at 23:15
6
$\begingroup$

It's because the Fourier transform of $\mathrm{J}_0$ vanishes outside $[-1,1]$.

Let $I$ be the integral $$ \def\J{{\mathrm{J}_0}}\def\dd{{\,\mathrm{d}}}\def\ii{{\mathrm{i}}} \def\ee{{\mathrm{e}}} I(\alpha) = \int_0^\infty \cos\alpha x\prod_k \J(\beta_k x)\,\dd x. $$ I will use the integral representation $$ \J(x) = \int_0^\pi \cos(x\sin\theta) \frac{\dd\theta}{\pi} = \int_0^1 \cos(x u)\frac{2\dd u}{\pi\sqrt{1-u^2}} $$ together with the Fourier transform of the Heaviside sign function in the form $$ \int_0^\infty e^{\ii ax}\,\dd x = \text{P.V.}\frac{\ii}{a} + \pi\delta(a). $$

Expanding each Bessel function, we get $$ I(\alpha) = \int_0^\infty\dd x\int_0^1 \Big( \prod_k \frac{2\dd u_k}{\pi\sqrt{1-u_k^2}}\Big) \cos\alpha x \prod_k \cos(\beta_k x u_k). $$

Now expand each cosine as $\cos x = \frac12(\ee^{\ii x} + \ee^{-\ii x})$: $$ \cdots = \int_0^\infty \dd x\int_0^1 \Big( \prod_k \frac{2\dd u_k}{\pi\sqrt{1-u_k^2}} \Big) \sum_{s\in\{\pm1\}^{n+1}} 2^{-n-1} \exp\Big( \ii s_0\alpha x + \sum_k \ii s_k \beta_k u_k x \Big), $$ where the sum is taken over all $2^{n+1}$ choices of signs $s_0,\ldots,s_n = \pm1$ that come from expanding the cosines in exponentials.

The integral over $x$ now can be done directly, as above: $$ \cdots = \frac{1}{2\pi^{n-1}} \int_0^1 \Big( \prod_k \frac{\dd u_k}{\sqrt{1-u_k^2}} \Big) \sum_{s\in\{\pm1\}^{n+1}} \delta\Big(s_0\alpha + \sum_k s_k \beta_k u_k \Big). $$ (The imaginary part has to vanish so only the $\delta$ term remains.)

This makes it clear why the integral vanishes: the integral representation of the $n$ Bessel functions integrates over the $n$-cube $[0,1]^n$, but the $2^{n+1}$ hyperplanes $$ s_0\alpha + \sum_k s_k \beta_k u_k = 0 $$ do not intersect this cube at all when $$ \sum_k \beta_k < |\alpha|. $$

$\endgroup$
0

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.