4

I'm trying to prove the Riemann integral $$\int_0 ^ \infty \frac{\sin(x)} x \ dx = \frac{\pi}{2}$$ using tools that I, as a statistician, am likely not to forget. My general approach is

  1. Show that the Laplace transform $\psi(t) := \int_0 ^ \infty \frac{e^{-xt} \sin(x)} x \ dx$ is $-\arctan(t) + \frac \pi 2$ for $t \in (0, \infty)$ by differentiating under the integral (using dominated convergence stuff since the Lebesgue integral exists here) and solving the resulting differential equation.
  2. Argue that $\psi(t)$ is continuous at $0$.

I've hit a snag at step 2. So, under what conditions do we have that the Laplace transform of a function is continuous at $t = 0$? Any trivial garuntees that it should be for $\frac {\sin x} x$ in particular? I feel like this would be easier if the function was Lebesgue integrable since when I'm trying to bound $|\psi(0) - \psi(t)|$ I am having a hard time controlling the tail of the integral of $\psi(t)$ for $t$ uniformly in $t$ near $0$.

I'm also finding this somewhat interesting since it looks like $\frac{\sin x} x$ is sitting on a boundary of sorts: we are Lebesgue integrable up to a point, then Riemann integrable, and past that not integrable($t > 0, = 0, < 0$) So a bunch of Lebesgue integrable functions are converging to a function which has an improper Riemann integral but no Lebesgue integral. And yet interchanging limits produces the correct answer anyways when calculating the integral.

EDIT: We already have an answer below that suggests how to do this for $\frac {\sin x} x$ but I'd like to get something on the more general question of under what conditions we have right continuity of the Laplace transform at $0$. Is it always continuous provided that the function has a finite improper Riemann integral?

Mittens
  • 39,145
guy
  • 4,352
  • This is a very old question, I have jut posted a solution that I think answers the question in your edit if you are still interested. – Mittens May 24 '23 at 02:34

4 Answers4

3

You can get right continuity at $s=0$ along the real axis, at least: We have $$I(s,n):=\int_{n\pi}^\infty \frac{e^{-st}\sin t}{t}\,dt=\sum_{k=n}^\infty\int_{k\pi}^{(k+1)\pi} \frac{e^{-st}\sin t}{t}\,dt$$ and use the alternating nature of the series to show that the entire sum lies between $0$ and its first term. Thus $I(s,n)\to0$ uniformly in $s\ge0$ as $n\to\infty$. On the other hand, you have unifom convergence of the integrand for $t\in[0,n\pi]$, so you don't even need to invoke Lebesgue theory to get the needed convergence.

1

I would like to show my way to show that $ \int_0 ^ \infty \frac{\sin(x)} x \ dx=\frac{\pi}{2}$

$\psi(t) := \int_0 ^ \infty \frac{e^{-xt} \sin(x)} x \ dx$

$\psi '(t) := -\int_0 ^ \infty e^{-xt} \sin(x) \ dx$

From Laplace tables we can find sine's Laplace transform=$\frac{1}{t^2+1}$ . if you don't want to use Laplace table, replace $\sin(x)=\frac{e^{ix}-{e^{-ix}}}{2i}$ and find the same result easily from $ \int_0 ^ \infty e^{-ax} \ dx=\frac{1}{a}$.

$\psi '(t) := -\frac{1}{t^2+1} $

$\int \psi '(t)dt=-\int \frac{1}{t^2+1} dt$

$\psi(t) := -\arctan(t)+c $

if $\psi(t) := \int_0 ^ \infty \frac{e^{-xt} \sin(x)} x \ dx$
then $\psi(\infty)=\int_0 ^ \infty \frac{e^{-\infty t} \sin(x)} x \ dx =0$

if $\arctan(\infty)=\frac{\pi}{2}$
then

$c=\frac{\pi}{2}$

$\psi(0) := \int_0^\infty \frac {\sin(x)}{x} dx=-\arctan(0)+c=\frac{\pi}{2}$

Mittens
  • 39,145
Mathlover
  • 10,058
1

You can use an alternate result which is quite nice.

$${\mathcal L}\left\{ {\frac{{\sin t}}{t}} \right\}(s) = \int\limits_0^\infty {{e^{ - st}}\frac{{\sin t}}{t}dt} $$

But

$$\frac{{{e^{ - st}}}}{t} = \int\limits_s^\infty {{e^{ - tm}}dm} $$

So

$$\eqalign{ & {\mathcal L}\left\{ {\frac{{\sin t}}{t}} \right\}(s) = \int\limits_0^\infty {\left( {\int\limits_s^\infty {{e^{ - tm}}} dm} \right)\sin tdt} \cr & {\mathcal L}\left\{ {\frac{{\sin t}}{t}} \right\}(s) = \int\limits_s^\infty {\left( {\int\limits_0^\infty {{e^{ - tm}}} \sin tdt} \right)dm} \cr & {\mathcal L}\left\{ {\frac{{\sin t}}{t}} \right\}(s) = \int\limits_s^\infty {\frac{{dm}}{{1 + {m^2}}}} \cr} $$

This means that

$$\int\limits_0^\infty {{e^{ - st}}\frac{{\sin t}}{t}dt} = \int\limits_s^\infty {\frac{{dm}}{{1 + {m^2}}}} $$

Now let $s \to 0$.

Pedro
  • 122,002
  • 1
    The issue is that the change of order of integration needs some justification (Fubini-tonelli's does not apply here) – Mittens May 24 '23 at 18:18
1

Here is a general result that answers (2) in the OP:

Theorem: Suppose $f\in L^{loc}_1([0,\infty)$ (i.e., $f\mathbb{1}_{[a,b]}\in L_1([0,\infty)$ for all $0\leq a\leq b<\infty$), and suppose that $I=\int^\infty_0f$ converges (in the sense that $\lim_{A\rightarrow\infty}\int^A_0f$ exists). Then

  1. For any $s>0$, $I(s)=\int^\infty_0e^{-st}f(t)\,dt$ converges (that is $I(s):=\lim_{A\rightarrow\infty}\int^A_0e^{-s t}f(t)\,dt$ exists as a real number).
  2. $\lim_{s\rightarrow0}I(s)=I$.

Comments:

  1. Very often, $f$ is Riemann integrable on every closed interval $[a,b]\subset[0,\infty)$ and $I=\int^\infty_0f$ is defined as a Riemann improper integrable. The assumption in the Theorem above generalizes this situation.
  2. In the context of the OP, $f(t)=\frac{\sin t}{t}\mathbb{1}_{\{t\neq0\}}+\mathbb{1}_{\{t=0\}}(t)$ satisfies the condition of the Proposition since for any $n\in\mathbb{N}$, $$\int^{\pi (n+1)}_0\frac{\sin t}{t}\,dt=\sum^n_{k=0}(-1)^k\int^{(k+1)\pi}_{k\pi}\frac{|\sin t|}{t}\,dt$$ defines an alternating series with main term $a_k=\int^{(k+1)\pi}_{k\pi}\frac{|\sin t|}{t}\,dt=\int^{\pi}_0\frac{\sin t}{k\pi+t}\,dt\searrow0$ as $k\rightarrow\infty$. Notice that $f_s(t):= e^{-s t}f(t)$ is in fact Lebesgue integrable for any $s>0$, which is a much stronger condition than needed.

Proof of Theorem: Fix $s>0$. Since $f$ is (Lebesgue) integrable in any bounded interval in $[0,\infty)$ then so is $t\mapsto e^{-st}f(t)$. By the second mean value theorem of Lebesgue integrals (see here for example), for any $0<A<B$, there exists $x_0=x_0(A,B)\in[A,B]$ such that $$\int^B_A e^{-s t} f(t)\,dt= e^{-As}\int^{x_0}_A f(t)\,dt$$ Since $I$ is convergent, for any $\varepsilon>0$, there is $M>0$ such that \begin{align} \Big|\int^p_q f(t)\,dt\Big|<\varepsilon/2\quad\text{whenever}\quad M\leq p\leq q\tag{1}\label{one} \end{align} As $0<e^{-As}\leq1$, we have that for $B\geq A\geq M$, \begin{align} \Big|\int^B_A e^{-sA}f(t)\,dt\Big|=e^{-sA}\Big|\int^{x_0}_Af(t)\,dt\Big|<\frac{\varepsilon}{2}\tag{2}\label{two} \end{align} This implies that $J(m):=\int^m_0e^{-st}f(t)\,dt$ is Cauchy and so, $\lim_{m\rightarrow\infty}J(m)$ exists (as a real number). This proves (1). Observe that \eqref{two} is uniform in $s>0$.

To prove (2) let $M$ large enough so that \eqref{one} holds. Then, for $0\leq M\leq A\leq B$ \begin{align} \Big|\int^B_0 f(t)\,dt-\int^B_0 e^{-st}f(t)\,dt\Big|&\leq \Big|\int^A_0 f(t)(1-e^{-st})\,dt\Big|+\Big|\int^B_Af(t)\,dt\Big|+\Big|\int^B_Ae^{-st}f(t)\,dt\Big|\\ &<\int^A_0 |f(t)|(1-e^{-st})\,dt+\varepsilon \end{align} Letting $B\rightarrow\infty$ yields $$|I-I(s)|\leq\int^A_0 |f(t)|(1-e^{-st})\,dt+\varepsilon$$ As $|f(t)|(1-e^{-st})\mathbb{1}_{[0,A]}(t)\leq |f(t)|\mathbb{1}_{[0,A]}(t)\in L_1([0,\infty)$ and $\lim_{s\rightarrow0} f(t)(1-e^{-st})\xrightarrow{s\rightarrow0}f(t)$ pointwise, $$\limsup_{s\rightarrow0}|I-I(s)|\leq\varepsilon$$ by dominated convergence. Part (2) of the theorem follows immediately.

Mittens
  • 39,145
  • Nice. I will study this in more detail later – FShrike May 24 '23 at 18:24
  • 1
    @FShrike: I hope you find this result interesting. It provides justification to many applications of the Laplace method (or the Feynman technique). The answer by Mathlover, though entirely correct, misses to point out the convergence of the integral of interest. At issue is the fact that $\mathcal{L}(f)(s)$ may converge to some limit as $s\rightarrow0$ but the function $f$ may not be integrable (in the improper sense) For example $\int^\infty_0e^{-sa}\cos t,dt=\frac{s}{s^2+1}\xrightarrow{s\rightarrow0}0$ but $\int^\infty_0\cos t,dt$ does not converge. – Mittens May 24 '23 at 18:57