19

I tried to solve it the Feynman way and defined:

$$I(a):=\int_0^{\infty} {\sin(\tan(a \cdot x)) \over x} \ dx$$

And look what happens when one substitutes $u=ax$ $(a>0)$:

$$I(a)=\int_0^{\infty} {\sin(\tan(u)) \over u} \ du = I(1)$$

Which implies that $I(a)=const$ for $a>0$. More generally $I(a)=c \cdot sign(a)$. I wonder whether this can help.

I recalled that in order to solve $\int_0^{\infty} {\sin(x) \over x} \ dx$ using the Feynman technique one had to define $I(a):=\int_0^{\infty} {\sin(x) \over x} e^{-a \cdot x}\ dx$ and differentiate it. Consequently I suspect we should define $I(a):=\int_0^{\infty} {\sin(\tan x) \over x} e^{-a \cdot x}\ dx$, but differentiation yields:

$$I'(a)=-\int_0^{\infty} {\sin(\tan x)} e^{-a \cdot x}\ dx$$

Which is another difficult integral.

Any help?

(please try to avoid gamma functions)

Chris
  • 1,075
  • Have you tried graphing that function? It has extremely pathological behavior (Not saying it can't be integrated). Also you should know that Feynman integration is usually very hard to do. You need to have a great deal of intuition to do it in any kind of effective way... – Zach466920 May 07 '15 at 16:12
  • i would be really surprise dif we could find a closed form solution – tired May 07 '15 at 16:17
  • There is a closed form, at least for the definite integral. I have a list of very hard integrals a friend of mine gave me a couple of years ago, and this one's in it. Unfortunately, I'm not in contact with that person anymore so I don't know what's the solution. I guess it should be solved by the Feynman technique since most of the integrals on that list are solved that way. – Chris May 07 '15 at 16:21
  • 1
    would you mind to show us the solutions, maybe helpful to find a way – tired May 07 '15 at 16:21
  • I don't have the solution, that's why I am asking you for help. – Chris May 07 '15 at 16:23
  • You don't have a value for the integral or you don't have a solution? I'm guessing neither. However if you do have a value for the integral, that would be helpful to know. – Simon S May 07 '15 at 16:48
  • I have nothing. – Chris May 07 '15 at 16:53
  • 1
    Shouldn't it be $I'(1) $, not $I(1)$, which is of course zero? Also it seems like this integral tends to $1$. – Cameron Williams May 07 '15 at 21:27
  • You're right, I guess I'm too tired. How do you see it tends to 1? – Chris May 07 '15 at 21:28
  • 2
    Observation: We have an even integrand, so we can change limits of integration to $-\infty, \infty$. Therefore we could possibly use contour integration – tired May 07 '15 at 21:35
  • 1
    I did some numerical tests and it seems to converge to $1$. – Cameron Williams May 08 '15 at 01:55
  • 3
    Hint: $\sum_{n=-\infty}^\infty \frac{1}{x-n\pi} = \frac{1}{\tan x}$ and the integral is $\frac{\pi}{2}(1-\frac{1}{e}) \approx 0.992932651899435760276275$ – achille hui May 08 '15 at 10:20

6 Answers6

26

Notice $\tan x$ is a periodic function with period $\pi$ and recall following expansion:

$$\frac{1}{\tan x} = \lim_{N\to\infty}\sum_{n=-N}^N \frac{1}{x + n\pi}$$

The integral we seek $$\mathcal{I} \stackrel{def}{=} \int_0^\infty \frac{\sin\tan x}{x} dx = \frac12 \int_{-\infty}^\infty \frac{\sin\tan x}{x} dx = \frac12 \left(\sum_{n=-\infty}^\infty \int_{(n-\frac12)\pi}^{(n+\frac12)\pi}\right)\frac{\sin\tan x}{x}dx $$ can be rewritten as $$ \mathcal{I} = \frac12 \int_{-\frac12\pi}^{\frac12\pi}\sin\tan x\left(\sum_{n=-\infty}^\infty\frac{1}{x+n\pi}\right) dx = \frac12\int_{-\frac12\pi}^{\frac12\pi}\frac{\sin\tan x}{\tan x} dx $$ Change variable to $t = \tan x$, we get

$$\mathcal{I} = \frac12\int_{-\infty}^{\infty} \frac{\sin t}{t(1+t^2)} dt = \frac12\Im\left[\int_{-\infty}^{\infty}\frac{e^{it}-1}{t(1+t^2)} dt\right]$$

We can evaluate the integral on RHS as a contour integral. By completing the contour in upper half-plane and using the fact the integrand has only two poles at $t = \pm i$, we get:

$$\begin{align} \mathcal{I} &= \frac12\Im\left[ 2\pi i \, \mathop{\text{Res}}_{z = i}\left(\frac{e^{it}-1}{t(1+t^2)}\right)\right] = \pi\Re\left[ \frac{e^{i(i)} - 1}{i(i+i)}\right] = \frac{\pi}{2}\left(1 - \frac1e \right)\\ &\approx 0.9929326518994357602762750999834... \end{align} $$

Update

If one don't want to use contour integral, we can replace the last step by a Feymann trick. Consider the function

$$J(a) = \int_0^\infty \frac{\sin(at)}{t(1+t^2)}dt $$

It is easy to see $\mathcal{I} = J(1)$ and $J(a)$ satisfies following ODE for $a > 0$.

$$\left( -\frac{d^2}{da^2} + 1 \right)J(a) = \int_0^\infty \frac{\sin(at)}{t} dt = \int_0^\infty \frac{\sin t}{t} dt = \frac{\pi}{2}$$

This implies $\displaystyle\;J(a) = \frac{\pi}{2} + A e^a + B e^{-a}\;$ for suitably chosen constants $A, B$. Notice $$\begin{align} J(+\infty) &= \lim_{a\to+\infty} \int_0^\infty \frac{\sin t}{t\left(1 + \left(\frac{t}{a}\right)^2\right)} dt = \int_0^\infty \frac{\sin t}{t} dt = \frac{\pi}{2}\\ J'(0^{+}) &= \lim_{a\to 0^{+}} \int_0^\infty \frac{\cos(at)}{1+t^2} dt = \int_0^\infty \frac{dt}{1+t^2} = \frac{\pi}{2} \end{align}$$

This fixes $\;A = 0$, $\displaystyle\;B = -\frac{\pi}{2}\;$ and hence

$$J(a) = \frac{\pi}{2}\left(1 - e^{-a}\right) \quad\implies\quad \mathcal{I} = J(1) = \frac{\pi}{2}\left( 1 - \frac{1}{e}\right)$$

achille hui
  • 122,701
  • Simply brilliant ! – Lucian May 08 '15 at 12:26
  • (+999) Great work – tired May 08 '15 at 12:44
  • Thanks for this great proof :) – Chris May 08 '15 at 13:50
  • After a cursory glance at your answer, I thought you were doing something different. Now that I've finished my answer, I see that the first part of your answer pretty much outlines what I did. I hope you don't mind me filling in some details. If you do, I will delete my answer. – robjohn May 08 '15 at 18:10
  • @robjohn I don't mind. the second part of you answer shows another way to perform the contour integral which others may find useful. – achille hui May 08 '15 at 18:31
11

Real Manipulations $$ \begin{align} \int_0^\infty\frac{\sin(\tan(x))}x\,\mathrm{d}x &=\frac12\int_{-\infty}^\infty\frac{\sin(\tan(\pi x))}x\,\mathrm{d}x\tag{1}\\ &=\frac12\sum_{k\in\mathbb{Z}}\int_0^1\frac{\sin(\tan(\pi x))}{x+k}\,\mathrm{d}x\tag{2}\\ &=\frac\pi2\int_0^1\sin(\tan(\pi x))\cot(\pi x)\,\mathrm{d}x\tag{3}\\ &=\frac\pi2\int_{-1/2}^{1/2}\sin(\tan(\pi x))\cot(\pi x)\,\mathrm{d}x\tag{4}\\ &=\frac12\int_{-\infty}^\infty\frac{\sin(u)}{u(1+u^2)}\,\mathrm{d}u\tag{5}\\ \end{align} $$ Explanation:
$(1)$: $\frac{\sin(\tan(x))}x$ is even, double domain and divide by $2$; substitute $x\mapsto\pi x$
$(2)$: break the domain into unit intervals
$(3)$: $\sum\limits_{k\in\mathbb{Z}}\frac1{x+k}=\pi\cot(\pi x)$ (see this answer)
$(4)$: $\sin(\tan(\pi x))\cot(\pi x)$ has period $\pi$
$(5)$: substitute $u=\tan(\pi x)$

Contour Integration

We will use the counter-clockwise contour $$ \gamma^+=[-R-i/2,R-i/2]\cup Re^{i[0,\pi]}-i/2 $$ and the clockwise contour $$ \gamma^-=[-R-i/2,R-i/2]\cup Re^{-i[0,\pi]}-i/2 $$ Then $$ \begin{align} \frac12\int_{-\infty}^\infty\frac{\sin(u)}{u(1+u^2)}\,\mathrm{d}u &=\frac12\int_{-\infty-\frac i2}^{\infty-\frac i2}\frac{\sin(u)}{u(1+u^2)}\,\mathrm{d}u\tag{6}\\ &=\frac1{4i}\int_{\gamma^+}\frac{e^{iz}}{z(1+z^2)}\,\mathrm{d}z -\frac1{4i}\int_{\gamma^-}\frac{e^{-iz}}{z(1+z^2)}\,\mathrm{d}u\tag{7}\\ &=\frac1{4i}\int_{\gamma^+}e^{iz}\left(\frac1z-\frac{1/2}{z-i}\color{#A0A0A0}{-\frac{1/2}{z+i}}\right)\,\mathrm{d}u\\ &-\frac1{4i}\int_{\gamma^-}e^{-iz}\left(\color{#A0A0A0}{\frac1z-\frac{1/2}{z-i}}-\frac{1/2}{z+i}\right)\,\mathrm{d}u\tag{8}\\ &=\frac{2\pi i}{4i}\left(1-\frac1{2e}\right)-\frac{2\pi i}{4i}\left(\frac1{2e}\right)\tag{9}\\ &=\frac\pi2\left(1-\frac1e\right)\tag{10} \end{align} $$ Explanation:
$(6)$: no singularities in $\small[-R,R]\cup[R,R-i/2]\cup[R-i/2,-R-i/2]\cup[-R-i/2,-R]$
$\hphantom{(6)\text{:}}$ integrand vanishes on vertical segments
$(7)$: $\sin(z)=\frac{e^{iz}-e^{-iz}}{2i}$, integrals vanish on semi-circular arcs
$(8)$: $\frac1{z(z^2+1)}=\frac1z-\frac{1/2}{z-i}-\frac{1/2}{z+i}$
$(9)$: singularities at $0$ and $i$ are in $\gamma^+$, singularity at $-i$ is in $\gamma^-$
$(10)$: simplify

robjohn
  • 345,667
6

Lobachevsky Integral Formula: If $f(x)$ meet $f(x+π)=-f(x)$ and $f(−x)=f(x)$.

Then $$\int_0^\infty f(x)\cdot\frac{\sin x}{x}dx=\int_0^\frac{\pi}{2}f(x)\cdot \cos xdx$$

so $$\int_0^\infty\frac{\sin(\tan x)}{x}dx=\int_0^\infty\frac{\sin(\tan x)}{\sin x}\cdot\frac{\sin x}{x}dx$$ where $$f(x)=\frac{\sin(\tan x)}{\sin x}$$

so $$\int_0^\infty\frac{\sin(\tan x)}{x}dx=\int_0^\frac{\pi}{2}\frac{\sin(\tan x)}{\sin x}\cdot \cos xdx=\int_0^\frac{\pi}{2}\frac{\sin(\tan x)}{\tan x}dx$$ $$=\int_0^\infty\frac{\sin \theta}{\theta(1+\theta^2)}d\theta =\int_0^\infty\frac{\sin\theta}{\theta}d\theta-\int_0^\infty\frac{\theta\sin\theta}{1+\theta^2}d\theta=\frac{\pi}{2}-\frac{\pi}{e\pi}=\frac{\pi}{2}\Big(1-\frac{1}{e}\Big)$$

3

Assume that $N$ is a large positive integer.

The complex function $$f(z) = \frac{e^{i \tan z}}{z} $$ has a simple pole at the origin and essential singularities at the half-integer multiples of $\pi$.

Let's integrate $f(z)$ around a rectangular contour that has vertices at $z= \pm \pi N$, $z=\pm \pi N+i\sqrt{N}$ with small clockwise-oriented semicircles of radius $\epsilon$ about the origin and about the points $z= \pi \left(k+\frac{1}{2} \right)$, $k=-N, -N+1, \ldots -1, 0, 1, \ldots N-2, N-1$.

The contour is similar to the contour used in this answer.

In the upper half-plane, $$|e^{i \tan z}| = |e^{i \tan (x+iy)}| =\exp \left(-\frac{\sinh 2y}{\cos 2x + \cosh 2y} \right) \le 1.$$

Since the height of the contour is $\sqrt{N}$ and the magnitude of $e^{i \tan z}$ is bounded in the upper half-plane, the estimation lemma tells us that integral vanishes on the left and right the sides of the rectangle vanish as $N \to \infty$.

And since the magnitude of $e^{i \tan z}$ is bounded in the upper half-plane, the integral vanishes on the small semicircles about the half-integer multiples of $\pi$ as $\epsilon \to 0$.

So we have

$$ \text{PV} \int_{-\infty}^{\infty} \frac{e^{i \tan x}}{x} \, dx - i \pi \,\text{Res}[f(z),0] -\lim_{N \to \infty} \int_{-N}^{N} \frac{e^{i \tan (t+i\sqrt{N})}}{t+i\sqrt{N}} \, dt =0. $$

But since the magnitude of $e^{i \tan z}- \frac{1}{e}$ tends to zero exponentially fast as $\Im(z) \to +\infty$, we can replace $e^{i \tan(t+ i \sqrt{N})}$ with $\frac{1}{e}$. (Specifically, it's going to zero like $\frac{2}{e} e^{-2 \, \Im(z)}$.)

Therefore,

$$\text{PV} \int_{-\infty}^{\infty} \frac{e^{i \tan x}}{x} \, dx - i \pi - \frac{1}{e} \lim_{N \to \infty} \int_{-N}^{N} \frac{1}{t+i\sqrt{N}} \, dt =0. $$

But if we integrate $g(z) = \frac{1}{z}$ around a similar contour (minus the semicircles about the half-integer multiples of $\pi$), we get

$$ \underbrace{\text{PV} \int_{-\infty}^{\infty} \frac{dx}{x}}_{0} - i \pi \underbrace{\text{Res}[g(z),0]}_{1} - \lim_{N \to \infty} \int_{-N}^{N} \frac{1}{t+i\sqrt{N}} \, dt = 0.$$

Therefore, $$ \lim_{N \to \infty} \int_{-N}^{N} \frac{1}{t+i\sqrt{N}} \, dt = - i \pi, $$ and

$$ \text{PV} \int_{-\infty}^{\infty} \frac{e^{i \tan x}}{x} \, dx - i \pi - \frac{1}{e} (-i \pi) =0.$$

Equating the imaginary parts on both sides of the equation, we have $$\int_{-\infty}^{\infty} \frac{\sin (\tan x)}{x} \, \mathrm dx = \pi \left(1- \frac{1}{e} \right).$$


UPDATE:

Integrating the function $\frac{z}{z^{2}+a^{2}} \, e^{i \tan z} $ around almost the same contour shows that $$\int_{-\infty}^{\infty} \frac{x}{x^{2}+a^{2}} \, \sin(\tan x) \, \mathrm dx = \pi \left(e^{- \tanh a}- \frac{1}{e} \right). $$

2

Ok i will try a contour method approach wihout a reference to the Mittag Leffler expansion of $\tan(x)$.

To begin, write $\sin(\tan[x])=\Im(e^{i\tan[x]})$.

$$ I=\Im\underbrace{\int_{\epsilon}^{\infty}\frac{e^{i\tan[x]}}{x}}_{I_1}+\Im\underbrace{\int_{-\infty}^{-\epsilon}\frac{e^{i\tan[x]}}{x}}_{I_2} $$

where the limit of $\epsilon\rightarrow 0$ is implicit.

Now we come to our crucial step :

We should keep in mind some possible contributions from complex infinity, because Jordan's Lemma is not sufficent here [the oscillations are not regular enough]

We can rewrite $I_1 $ using Cauchy's theorem ( $\tan[\pm ix]=\pm i \tanh[x]$)

$$I_1=\int_{QC^1_0}\frac{e^{-i\tan[z]}}{z}+\int_{QC^1_{\infty}}\frac{e^{-i\tan[z]}}{z}+\int_{\epsilon}^{\infty-\epsilon}\frac{e^{-\tanh[y]}}{iy}$$

Similiar $I_2$

$$I_2=\int_{QC^2_0}\frac{e^{-i\tan[z]}}{z}+\int_{QC^2_{\infty}}\frac{e^{-i\tan[z]}}{z}-\int_{\epsilon}^{\infty-\epsilon}\frac{e^{-\tanh[y]}}{iy}$$

Here $QC^{1,2}_{0,\infty}$ denotes a quarter circle around $0(\infty)$ in the first(second) quadrant and the remaining to straightline integrals are along the imaginary axis.

Adding back $I_1$ and $I_2$ we see that the two integrals along the positve/negative imaginary cancel out so we are left with $$ I_1+I_2=\int_{QC^1_0}\frac{e^{-i\tan[z]}}{z}+\int_{QC^1_{\infty}}\frac{e^{-i\tan[z]}}{z}+\int_{QC^2_0}\frac{e^{-i\tan[z]}}{z}+\int_{QC^2_{\infty}}\frac{e^{-i\tan[z]}}{z} $$

we can achive further simplifiction by exploting the fact that semi circles around the same point

a) enclose singularities in opposite direction

b) have twisted integration regions.

This results in two minus signs which multiply to an overall plus. Therfore:

$$ I_1+I_2=2\int_{QC^1_0}\frac{e^{-i\tan[z]}}{z}+\int_{QC^1_{\infty}}\frac{e^{-i\tan[z]}}{z} $$

Because this are first order poles the integral is finite in a principal value sense. It's now easy to show that the integral is given by

$$ I=2\Im \left[\frac{\pi i}{4}\left(\text{res}(0)+\text{res}(\infty)\right)\right] $$

the residues can be calculated as $\text{res}[0]=1 $ and $\text{res}[\infty]=\frac{-1}{e} $ and therefor

$$ I=\frac{\pi }{2}\left(1-\frac{1}{e}\right) $$

As expected from the other solutions

tired
  • 12,325
0

I give you two kind of representation; both gave me a good result.

  1. $$\frac{sin(tan(ax))}{x} = \frac{(tan(a x))}{x} \int_0^1 cos(t~tan(a x)) dt $$

  2. $$ \frac{sin(tan(ax))}{x} = \frac{-i}{2x \sqrt\pi} \int_{(-i \infty+\gamma)}^{(i \infty+\gamma)} \frac{(2^{(-1+2 s)} \Gamma(s) \tan^{(1-2 s)}(a x))}{(\Gamma(\frac{3}{2}-s))} ds$$ $$For (0<\gamma<1~and~\tan(a x)>0)$$

You can use the second easily if you are familiar with de $\Gamma$-function.

Good luck!

  • Okay, after intensive thinking I still don't know how to use number 1. Can you give more details? – Chris May 07 '15 at 20:36
  • I found $\frac{\pi}{2}(1-\frac{1}{e})$; It seems to be the answer for the different answers provided from my last visit. –  May 08 '15 at 16:27