15
$\begingroup$

Here I want to get the closed form solution of the following summation

$$ \sum_{n=1}^\infty \frac{\sin\sqrt{n^2+1}}{\sqrt{n^2+1}} \qquad(1) $$

Or the more general form ($x$ be an arbitrary real number, and $a\geq0$ is a constant):

$$ f_a(x) = \sum_{n=1}^\infty \frac{\sin\left(x\sqrt{n^2+a^2}\right)}{\sqrt{n^2+a^2}}\qquad(2) $$

I tried the numeircal simulations before I post the question. I truncated the first $1,000,000$ terms of equation (1) and it turned $0.781233190560320$.

Anyone can help me?

In fact, I made it the reduced case when $a=0$. It can be proved by Fourier series: $$ f_0(x)=\sum_{n=1}^\infty \frac{\sin nx}{n} = \left\{ \matrix{\dfrac{\pi-x}{2}, 0<x<2\pi\\0, x=0,2\pi} \right. $$ And the function is periodical: $$ f_0(x) = f_0(x+2\pi) $$

Edit: How about this one? $$ g_a(x) = \sum_{n=1}^\infty \frac{\cos\left(x\sqrt{n^2+a^2}\right)}{\sqrt{n^2+a^2}}\qquad(3) $$

We get the "diverging wave solution" in physics when we combine the equation (2) and (3): $$ h_a(x)=g_a(x)+\text{i}f_a(x)=\sum_{n=1}^\infty \frac{\exp\left(\text{i}x\sqrt{n^2+a^2}\right)}{\sqrt{n^2+a^2}}\qquad(4) $$

Edit #2: I tested the solution solved by Random Variable (see the most ranked answer and thousands thanks to it!) compared with the truncating results:

$$ \sum_{n=1}^N\frac{\sin \left(x\sqrt{n^2+a^2}\right)}{\sqrt{n^2+a^2}}, N = 1,000,000, a=1 $$

Here is the solution by @Random Variable: the solution of equation (2) as the following:

$$ \sum_{n={1}}^\infty \frac{\sin\left(x\sqrt{n^2+a^2}\right)}{\sqrt{n^2+a^2}} = \frac{\pi}{2} J_0(ax) -\frac{\sin(ax)}{2a}, a>0, 0<x<2\pi\qquad(2*) $$ where $J_0(ax)$ is the Bessel function of the first kind of order zero.

Here is the comparison: enter image description here

It can be found that the both agree well when $0 <x<2\pi$,but differ in other domain. So, how about the solution beyond $(0,2\pi)$?

Edit #3:

Inspired by Random Variable's answer, I found the solution of equation (3) as the following:

$$ \sum_{n={1}}^\infty \frac{\cos\left(x\sqrt{n^2+a^2}\right)}{\sqrt{n^2+a^2}} = -\frac{\pi}{2} Y_0(ax) -\frac{\cos(ax)}{2a}, a>0, 0<x<2\pi\qquad(3*) $$ where $Y_0(ax)$ is the Bessel function of the second kind of order zero.

Here is the comparison: enter image description here Note that equation (3) is divergent when $x=0$.


Possible relating QUESTIONS:

Does there exist a closed form for the non-integer shifted sinc-function series: $\frac{\sin(n+a)x}{(n+a)x}$?

$\endgroup$
  • $\begingroup$ It's equal to $$\sum_{n=1}^\infty \prod_{k=1}^\infty \left(1-\frac{n^2+1}{k^2\pi^2}\right)$$ but I don't immediately see how that helps. $\endgroup$ – orlp Nov 27 '17 at 14:28
  • $\begingroup$ Does $(3*)$ seem to hold numerically? I wasn't sure if the Abel-Plana formula would be applicable due to the fact that $\frac{\cos(x \sqrt{z^{2}+a^{2}})}{\sqrt{z^{2}+a^{2}}} $ has branch points on the imaginary axis. $\endgroup$ – Random Variable Nov 28 '17 at 5:09
  • $\begingroup$ I just reasoned out the solution by analogy according to the equation 10.9.9 HERE. The numerical results does hold on $(-2\pi,2\pi)$! $\endgroup$ – Jiaxin Zhong Nov 28 '17 at 5:17
  • $\begingroup$ There was a missing condition about the behavior of $f(z)$ as $\operatorname{Re}(z) \to +\infty$. I'm not quite sure how to justify that the Abel-Plana formula is applicable here. And the justification I used in my answer to that other question won't work here. $\endgroup$ – Random Variable Nov 28 '17 at 12:26
  • $\begingroup$ Do you have a link to the original question/answer? There's some context missing here $\endgroup$ – Dylan Nov 28 '17 at 12:50
17
+50
$\begingroup$

UPDATE:

To apply the Abel Plana formula, the behavior of $f(z)$ as $\operatorname{Re}(z) \to + \infty$ is also important. This was omitted from my answer.

A sufficient condition, now stated HERE, is $f(z) \sim O(e^{2\pi|\Im z|}/|z|^{1+\epsilon}) $ as $\operatorname{Re}(z) \to \infty$.

The function here does not satisfy that condition. But as achille hui explains, this condition is about ensuring that $\lim_{b\to\infty} f(b) = 0$ and $$\lim_{b\to\infty}\int_0^\infty \frac{f(b+it)-f(b-it)}{e^{2\pi t} - 1}dt = 0. $$

And achille hui informed me that that latter is indeed satisfied here.

I will ask achille hui to post a brief answer to explain this.


We can use the version of the Abel-Plana formula stated in achille hui's answer HERE.

Alternatively, we could also use the approach I used HERE to evaluate a series involving Bessel functions. Both approaches are related.

(EDIT: I used the DCT in that other answer, which won't work here.)

First notice that the singularities of $f(z) = \frac{\sin \left( x\sqrt{z^{2}+a^{2}}\right)}{\sqrt{z^{2}+a^{2}}}$ are removable.

Also, for $x>0$, $\left|\sin \left( x\sqrt{z^{2}+a^{2}}\right)\right| \sim \frac{e^{x\left|\operatorname{Im}(z)\right|}}{2} $as $\operatorname{Im}(z) \to \pm \infty$.

So if $0 < x < 2 \pi$, the conditions of the Abel-Plana formula are satisfied, and we get $$\begin{align} \sum_{{\color{red}{n=0}}}^{\infty} \frac{\sin \left( x\sqrt{n^{2}+a^{2}}\right)}{\sqrt{n^{2}+a^{2}}} &= \int_{0}^{\infty} \frac{\sin \left( x\sqrt{t^{2}+a^{2}}\right)}{\sqrt{t^{2}+a^{2}}} \, dt + \frac{1}{2} f(0) + i (0) \\ &= \int_{0}^{\infty} \frac{\sin \left( x\sqrt{t^{2}+a^{2}}\right)}{\sqrt{t^{2}+a^{2}}} \, dt + \frac{\sin (ax)}{2a}. \end{align}$$

But from this answer, we know that $$\int_{0}^{\infty} \frac{\sin \left( x\sqrt{t^{2}+a^{2}}\right)}{\sqrt{t^{2}+a^{2}}} \, dt = \frac{\pi}{2} J_{0}(ax), \quad (a>0, \ x>0), \tag{1}$$ where $J_{0}(x)$ is the Bessel function of the first kind of order zero.

(To see that $(1)$ is related to the Mehler–Sonine integral representation of the Bessel function of the first kind, you only need to make the initial substitution in that answer).

Therefore, $$\sum_{{\color{red}{n=0}}}^{\infty} \frac{\sin \left( x\sqrt{n^{2}+a^{2}}\right)}{\sqrt{n^{2}+a^{2}}} = \frac{\pi}{2} J_{0}(ax) + \frac{\sin (ax)}{2a}, \quad (a>0, \ 0<x < 2 \pi).$$

To recover the $a=0$ case, you'll have to pull out the $n=0$ term and take the limit on both sides of the equation as $a \to 0^{+}$.


It should be noted that the series converges by Dirichlet's test since $$\sin \left( x\sqrt{t^{2}+a^{2}}\right) \sim \sin(tx) + \mathcal{O} \left(\frac{1}{t} \right)$$ as $t \to \infty$, which can be shown by expanding $\sqrt{t^{2}+a^{2}} = t \sqrt{1+ \frac{a^{2}}{t^{2}}}$ at $t= \infty$ and using the trig identity for $\sin(\alpha +\beta)$.

$\endgroup$
  • 1
    $\begingroup$ awesome, i wouldn't have bet a penny on a closed form but here we go....(+10) $\endgroup$ – tired Nov 27 '17 at 22:56
  • $\begingroup$ it is interesting that $\sum...=\int...$ becomes exact for $x=k\pi/a$ for all $k\in \mathbb{Z}$. Any idea why this is the case? $\endgroup$ – tired Nov 27 '17 at 23:00
  • $\begingroup$ @tired I could have evaluated that Bessel function series using the Abel-Plana formula, but at the time I didn't know that you could weaken the conditions. I think the proof would be similar to what I did in that answer. The contour would just be a bit different. $\endgroup$ – Random Variable Nov 27 '17 at 23:21
  • $\begingroup$ Impressive ... But it will take me some time to understand your answer... You may consider the other case as seen in my last edit. $\endgroup$ – Jiaxin Zhong Nov 28 '17 at 2:01
  • $\begingroup$ What does the growth conditions mean? $\endgroup$ – Jiaxin Zhong Nov 28 '17 at 7:20
6
$\begingroup$

Per request, this is a supplement to Random Variable's answer.

In Frank W. J Olver's book: Asymptotics and Special Functions, the Abel-Plana formula on finite sum appears in essentially following form:

Let $S$ be the strip $a \le \Re z \le b$ where $a, b \in \mathbb{N}$. For any function $f(z)$

  1. continuous on $S$ and analytic on interior of $S$.
  2. $f(z) \sim o(e^{2\pi|\Im z|} )$ as $\Im z \to \pm \infty$, uniformly with respect to $\Re z$.

We have $$\begin{align}\sum_{n=a}^b f(n) = &\int_a^b f(x) dx + \frac12\left( f(a) + f(b)\right) \\& + i \int_0^\infty \frac{f(a+it) - f(a-it) - f(b+it) + f(b-it)}{e^{2\pi t}-1} dt\end{align}$$

For $f(z) = \frac{\sin(x\sqrt{z^2+a^2})}{\sqrt{z^2+a^2}}$ with $0 < x < 2\pi$, above conditions is satisfied for $a = 0$ and any $b \in \mathbb{Z}$. To obtain the version of AP formula for infinite sum used in Random Variable's answer, we just need:

$$\lim_{b\to \infty}f(b) = 0\quad\text{ and }\quad \lim_{b\to\infty}\int_0^\infty \frac{f(b+it) - f(b-it)}{e^{2\pi t}-1} dt = 0$$ The first condition is trivial. For the second condition, notice for any $n > 0$, $$\left|\sqrt{(n\pm it)^2+a^2}\right| = \left|\sqrt{(n \pm i(t+a))(n \pm i(t-a))}\right| \ge n$$ We find for large $b$ and $t$, $$\frac{\left|f(b\pm it)\right|}{e^{2\pi t}-1} \le \frac{\left|\sin\left(x(b\pm it) + O\left(\frac{a^2}{b}\right)\right)\right|}{b(e^{2\pi t}-1)} \sim \frac{1}{2b}e^{-(2\pi - x)t}\left( 1 + O\left(\frac{a^2}{b}\right)\right) $$ For large $b$ but small $t$, we have $$\frac{\left|f(b + it) - f(b - it)\right|}{e^{2\pi t}-1} \sim O\left(\frac{1}{b}\right)$$ instead (the pole at $t = 0$ from denominator is cancelled by the differences in numerator).

Combine these, we have following estimate of the integral appears in second condition:

$$\int_0^\infty \frac{f(b+it)-f(b-it)}{e^{2\pi t}-1} dt = O\left(\frac{1}{b(2\pi - x)}\right)$$ The second condition is satisfied and the use of AP formula in answering this question is justified.

$\endgroup$
  • $\begingroup$ Thanks for your supplement and concern! $\endgroup$ – Jiaxin Zhong Nov 29 '17 at 5:53
1
$\begingroup$

To go beyond the limitation $-2\pi<x<2\pi$ for the sine series, we can use the representation G&R (6.677.6) (or Ederlyi TI p.57 1.13.47) \begin{equation} \frac{\sin x\sqrt{n^2+a^2}}{\sqrt{n^2+a^2}}=\int_0^x J_0\left( n\sqrt{x^2-t^2} \right)\cos at\,dt\tag{1} \end{equation} valid for $x>0$. (For $x<0$, we will use the fact that the series is an odd function of $x$, as remarked by @RandomVariable). The summation can be computed using the Schlömilch series (G&R 8.521.1): \begin{equation} \sum_{n=1}^\infty J_0(nz)=-\frac{1}{2}+\frac{1}{z}+2\sum_{m=1}^p\frac{1}{\sqrt{z^2-4\pi^2m^2\pi^2}} \end{equation} for $2p\pi<z<2(p+1)\pi$, which defines $p=\lfloor \frac{z}{2\pi}\rfloor$. Choosing $z=\sqrt{x^2-t^2}$, \begin{align} S(x)&=\sum_{n=1}^\infty\frac{\sin x\sqrt{n^2+a^2}}{\sqrt{n^2+a^2}}\\ &=\int_0^x\sum_{n=1}^\infty J_0\left( n\sqrt{x^2-t^2} \right)\cos at\,dt\\ &=\int_0^x\left[-\frac{1}{2}+\frac{1}{\sqrt{x^2-t^2}}+2\sum_{m=1}^{\lfloor \frac{\sqrt{x^2-t^2}}{2\pi}\rfloor}\frac{1}{\sqrt{x^2-t^2-4\pi^2m^2\pi^2}}\right] \cos at\,dt\\ &=\int_0^x\left[-\frac{1}{2}+\frac{1}{\sqrt{x^2-t^2}}\right]\cos at\,dt+2\sum_{1\le m< \lfloor \frac{x}{2\pi}\rfloor}\int_0^{\sqrt{x^2-4\pi^2m^2}}\frac{\cos at\,dt}{\sqrt{x^2-4\pi^2m^2-t^2}} \end{align} (The summation does not exist if $\lfloor \frac{x}{2\pi}\rfloor=0$). With the classic integral representation \begin{equation} \frac{\pi}{2}J_0(s)=\int_0^1\frac{\cos sx}{\sqrt{1-x^2}}\,dx \end{equation} the result can be written as \begin{equation} \sum_{n=1}^\infty\frac{\sin x\sqrt{n^2+a^2}}{\sqrt{n^2+a^2}}=\frac{\pi}{2}J_0(ax)-\frac{\sin ax}{2a}+\pi\sum_{1\le m< \lfloor \frac{x}{2\pi}\rfloor}J_0\left( a\sqrt{x^2-4\pi^2m^2-t^2}\right) \end{equation} which seems to be numerically correct. Unfortunately, no corresponding form of (1) for the cosines seems to exist.

$\endgroup$
  • $\begingroup$ Great. This is more of a complete answer compared to others. $\endgroup$ – Fengshan Xiong Sep 29 at 5:52
0
$\begingroup$

COMMENT to users: Achille-Hui and Random-Variable

How about this ?:

$$\sum _{n=0}^{\infty } \frac{\sin \left(x \sqrt{n^2+a^2}\right)}{\sqrt{n^2+a^2}}=\frac{1}{2} \pi J_0(a x)+\frac{\sin (a x)}{2 a}$$

$$\sum _{n=0}^{\infty } \mathcal{L}_x\left[\frac{\sin \left(x \sqrt{n^2+a^2}\right)}{\sqrt{n^2+a^2}}\right](s)=\frac{1}{2} \pi J_0(a x)+\frac{\sin (a x)}{2 a}$$

$$\mathcal{L}_s^{-1}\left[\sum _{n=0}^{\infty } \frac{1}{a^2+n^2+s^2}\right](x)=\frac{1}{2} \pi J_0(a x)+\frac{\sin (a x)}{2 a}$$

$$\mathcal{L}_s^{-1}\left[\frac{1}{2 \left(a^2+s^2\right)}+\frac{\pi \sqrt{-a^2-s^2} \cot \left(\pi \sqrt{-a^2-s^2}\right)}{2 \left(a^2+s^2\right)}\right](x)=\frac{1}{2} \pi J_0(a x)+\frac{\sin (a x)}{2 a}$$

$$\mathcal{L}_s^{-1}\left[\frac{1}{2 \left(a^2+s^2\right)}\right](x)+\mathcal{L}_s^{-1}\left[-\frac{\pi \cot \left(\pi \sqrt{-a^2-s^2}\right)}{2 \sqrt{-a^2-s^2}}\right](x)=\frac{1}{2} \pi J_0(a x)+\frac{\sin (a x)}{2 a}$$

for $a>0$ and $s>0$

$$\frac{\sin (a x)}{2 a}+\mathcal{L}_s^{-1}\left[\frac{\pi \coth \left(\pi \sqrt{a^2+s^2}\right)}{2 \sqrt{a^2+s^2}}\right](x)=\frac{1}{2} \pi J_0(a x)+\frac{\sin (a x)}{2 a}$$ $$\mathcal{L}_s^{-1}\left[\frac{\pi \coth \left(\pi \sqrt{a^2+s^2}\right)}{2 \sqrt{a^2+s^2}}\right](x)=\frac{1}{2} \pi J_0(a x)$$

$$\mathcal{L}_x\left[\mathcal{L}_s^{-1}\left[\frac{\pi \coth \left(\pi \sqrt{a^2+s^2}\right)}{2 \sqrt{a^2+s^2}}\right](x)\right](s)=\mathcal{L}_x\left[\frac{1}{2} \pi J_0(a x)\right](s)$$

$$\frac{\pi \coth \left(\pi \sqrt{a^2+s^2}\right)}{2 \sqrt{a^2+s^2}}\neq \frac{\pi }{2 \sqrt{a^2+s^2}}$$

and then:

$$\sum _{n=0}^{\infty } \frac{\sin \left(x \sqrt{n^2+a^2}\right)}{\sqrt{n^2+a^2}}\neq \frac{1}{2} \pi J_0(a x)+\frac{\sin (a x)}{2 a}$$

EDITED:

Comparison a numeric InverseLaplaceTransform and Bessel function:

enter image description here

$\endgroup$
  • $\begingroup$ OP's simulation already tell us $f_a(x) \ne \frac12\pi J_0(ax) + \frac{\sin(ax)}{2a}$ outside the interval $(0,2\pi)$. If you want to show this also fails for $x \in (0,2\pi)$, what you have done is not enough, you need to compute the Laplace transform with functions clipped within $(0,2\pi)$. $\endgroup$ – achille hui Nov 29 '17 at 18:43
  • $\begingroup$ @achillehui.Yes you're right is true only in interval 0<x<2*Pi $\endgroup$ – Mariusz Iwaniuk Nov 29 '17 at 20:13
  • $\begingroup$ Thanks for your concerns! $\endgroup$ – Jiaxin Zhong Nov 30 '17 at 13:56

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.