19

I'm new in the area of the series involving Bessel function of the first kind. What are
the usual tools you would recommend me for computing such a series? Thanks.

$$\sum_{n=1}^{\infty} \frac{J_0(2n)}{n^2}$$

Olivier Oloa
  • 122,789
user 1591719
  • 44,987

2 Answers2

25

Here is a route.

Recall that

$$ J_0(2n)=\frac 1\pi \int_0^\pi \cos (2n \sin x)\:{\rm d}x \tag1 $$

and that $$\sum_{n=1}^\infty\frac{\cos nt}{n^2}=\frac{\pi^2}{6}-\frac{\pi t}{2}+\frac{t^2}{4},\quad 0\leq t\leq 2\pi. \tag2 $$ Then, due to normal convergence of the series on $[0,2\pi] $, we may write $$ \begin{align} \sum_{n=1}^\infty\frac{J_0(2n)}{n^2} & =\frac 1\pi \int_0^\pi \sum_{n=1}^\infty\frac{\cos (2n \sin x)}{n^2}{\rm d}x \\\\ & =\frac 1\pi \int_0^\pi\left(\frac{\pi^2}{6}-\pi \sin x+\sin^2 x\right){\rm d}x \\\\ & =\frac{\pi^2}{6}-\frac{3}{2}\\\\ & =0.144934066848226436... \end{align} $$ thus

$$ \sum_{n=1}^{\infty} \frac{J_0(2n)}{n^2} = \frac{\pi^2}{6}-\frac{3}{2}. $$

Using a similar technique, one may obtain the following result.

$$ \begin{align} \sum_{n=1}^\infty\frac{J_0(2 n \alpha )}{n^2} &=\frac{\pi^2}{6}-2\alpha+\frac{\alpha^2}{2},\qquad \alpha \in [0,\pi).\tag3 \end{align} $$

Olivier Oloa
  • 122,789
  • 5
    Good job there! (+1) – user 1591719 Sep 18 '14 at 22:11
  • hehe, beautiful results. Yesterday when I created it I didn't even know it's a known series. – user 1591719 Sep 19 '14 at 20:59
  • I checked $(4)$ and I got some different results with Mathematica. I'll recheck that again. – user 1591719 Sep 19 '14 at 21:15
  • You are right, Chris's sis and Juan Ospina, I gave a hasty answer concerning some general results, I've deleted it. Thank you very much! – Olivier Oloa Sep 20 '14 at 21:10
  • Chris's sis pointed me to this answer recently, since I found myself in need of what amounts to the $\alpha$-derivative of this series. Unfortunately, the result $(3)$ appears to be valid only on the interval $(0,\pi)$ (see this WA plot), and it's the behavior outside that interval which I'm crucially interested in. So I'd be quite happy if you would revisit this result. – Semiclassical Jun 12 '15 at 20:14
  • @Semiclassical Long time no see! Thank you for your remark! Edited. – Olivier Oloa Jun 13 '15 at 12:17
  • Thanks! I'm still curious about the $\alpha\notin(0,\pi)$ case, so I may put it up as its own question. – Semiclassical Jun 13 '15 at 15:47
  • Hey Olivier, do you think does also work if the index of the Bessel function is not equal to zero? i have some trouble to apply it... – tired Feb 18 '17 at 13:37
  • @tired I did not take a look at it, but I don't consider it as not feasible... – Olivier Oloa Feb 19 '17 at 21:59
6

We could seemingly use Olivier Oloa's approach to also evaluate $$\sum_{n=1}^{\infty} {\color{red}{(-1)^{n-1}}} \frac{J_{0}(2n)}{n^{2}} \tag{1}.$$

But I thought it might be worthwhile to mention that $(1)$ can also be evaluated using the residue theorem from complex analysis.

The first thing to recognize is that $J_{0}(z)$ is an entire function.

Now let $a$ be a positive parameter, and consider the contour integral $$\int_{|z|=N+\frac{1}{2}} \frac{\pi \csc(\pi z)J_{0}(2az)}{z^{2}} \, dz. $$

As $\text{Im}(z) \to + \infty$, the magnitude of $\csc(\pi z)$ decays like $2 e^{- \pi \, \text{Im}(z)}$.

And as $\text{Im}(z) \to - \infty$, the magnitude of $\csc(\pi z)$ decays like $2 e^{\pi \, \text{Im}(z)}$.

Combining this with the asymptotic behavior of $J_{0}(z)$ as $|z| \to \infty$, we can conclude that that as $N \to \infty$ discretely, the contour integral will vanish if $0 < a \le \frac{\pi}{2}$.

(And since $J_{0}(0)=1$, the contour integral will also vanish if $a=0$.)

Summing the residues, and using the fact that $\pi \csc(\pi z)$ has simple poles at the integers with residues that alternate between $1$ and $-1$, we get $$0=2 \pi i \left( -2 \sum_{n=1}^{\infty} (-1)^{n-1} \frac{J_{0}(2an)}{n^{2}} + \text{Res} \left[\frac{\pi \csc(\pi z)J_{0}(2az)}{z^{2}},0 \right] \right). $$

But $$ \begin{align}\frac{\pi \csc(\pi z)J_{0}(2az)}{z^{2}} &= \frac{1}{z^{2}} \left(\frac{1}{z} + \zeta(2) z + \mathcal{O}(z^{3})\right) \left(1-a^{2}z^{2} + \mathcal{O}(z^{4}) \right) \tag{2} \\ &=\frac{1}{z^{3}} + \frac{{\color{red}{\zeta(2) - a^{2}}}}{z} + \mathcal{O}(z) . \end{align}$$

Therefore, $$\text{Res} \left[\frac{\pi \csc(\pi z)J_{0}(2az)}{z^{2}},0 \right] = \zeta(2) - a^{2} = \frac{\pi^{2}}{6}-a^{2}, $$

and

$$\sum_{n=1}^{\infty} (-1)^{n-1} \frac{J_{0}(2an)}{n^{2}} = \frac{\pi^{2}}{12} - \frac{a^{2}}{2} \, , \quad 0 \le a \le \frac{\pi}{2}. $$

(Since $J_{0}(x)$ is an even function, $(3)$ actually holds for $ -\frac{\pi}{2} \le a \le \frac{\pi}{2}$.)


$(2)$ https://en.wikipedia.org/wiki/Trigonometric_functions#Series_definitions


EDIT:

More generally, if $\alpha_{k}$ is a real parameter, $$\sum_{n=1}^{\infty} (-1)^{n-1} \frac{\prod_{k=1}^{m} J_{0}(2 \alpha_{k}n) }{n^{2}} = \frac{\pi^{2}}{12} - \frac{\sum_{k=1}^{m}\alpha_{k}^{2}}{2} \, , \quad \sum_{k=1}^{m} \left| \alpha_{k} \right| \le \frac{\pi}{2}. $$

  • Nice answer (+1) Good to know the way by residue theorem. – user 1591719 Feb 19 '16 at 09:58
  • Good answer! (+1) do you see any chance to apply this approach to the non-alternating sums here: http://math.stackexchange.com/questions/2149457/closed-form-for-an-infinite-series-of-bessel-functions?noredirect=1&lq=1? The usual kernels like $\text{cotan}$ give diverging contriubtions from the contour... – tired Feb 19 '17 at 19:43
  • @tired I don't know. As you said, the contour integral won't vanish in the limit if we use the typical $\pi \cot (\pi z)$ kernel. – Random Variable Feb 19 '17 at 20:07
  • @tired I don't know if you saw that I posted an answer to that question, so I thought I'd ping you. Originally the answer was incomplete, but it's complete now. – Random Variable Mar 14 '17 at 06:25