9

I need to prove that $\delta_n(x)=\sin (nx)/\pi x$ is a delta function. That is, to prove that: $$\underset{n\rightarrow \infty}{\lim}\int^\infty_{-\infty}f(x)\frac{\sin( nx)}{\pi x}dx=f(0).$$ For that I made $y=x/n$ and took the limit under the integral since $f(x)$ is supposed to be analytic in $\mathbb{R}$. I have a fundamental limit for the $\sin$ leaving me with $$\int^\infty_{-\infty}f(0)\cdot1 \;dy.$$ But this is equal to $f(0)$ $\iff$ $f(x)$ have contribuition only at $x=0$, but here the only assumption is that $f(x)=0$ when $x\rightarrow \pm \infty$.

What is wrong here?

Gary
  • 31,845

4 Answers4

7

I thought that it might be instructive to present a way forward that uses a regularization of the Dirac Delta. To that end we proceed.



PRELIMINARIES:

Let $\displaystyle \delta_n(x)=\frac1{2\pi}\int_{-n}^ne^{ikx}\,dk$. Then, we can write

$$\delta_n(x)=\begin{cases}\frac{\sin(nx)}{\pi x}&,x\ne0\\\\\frac n\pi&,x=0\tag1\end{cases}$$

The function $\delta_n(x)$ has the following properties:

  1. For each $n$, $\delta_n(x)$ is an analytic function of $x$.
  2. $\lim_{n\to \infty} \delta_n(0)= \infty$
  3. $\left|\int_{-\infty}^x \delta_n(x')\,dx'\right|$ is uniformly bounded.
  4. $\lim_{n\to \infty}\int_{-\infty}^{x}\delta_n(x')\,dx'=u(x)$, where $u$ is the unit step funciton.
  5. For each $n>0$, $\int_{-\infty}^\infty \delta_n(x)\,dx=1$

While heuristically $\delta_n(x)$ "approximates" a Dirac Delta when $n$ is "large," the limit of $\delta_n(x)$ fails to exist. However, if we interpret this limit in the distributional sense, then $\lim_{n\to\infty}\delta_n(x)\sim\delta(x)$. We will now show that this is indeed that case.



ANALYSIS:

Let $\phi(x)\in S$ where $S$ is the Schwarz Space of functions.

We will now evaluate the limit

$$\begin{align} \lim_{n\to \infty}\int_{-\infty}^\infty \delta_n(x)\phi(x)\,dx=\lim_{n\to \infty}\int_{-\infty}^\infty \frac{\sin(nx)}{\pi x}\phi(x)\,dx\tag1 \end{align}$$

Integrating by parts the integral on the right-hand side of $(1)$ with $u=\phi(x)$ and $v=\int_{-\infty}^{nx}\frac{\sin(x')}{\pi x'}\,dx'$ reveals

$$\begin{align} \lim_{n\to \infty}\int_{-\infty}^\infty \delta_n(x)\phi(x)\,dx&=-\lim_{n\to \infty}\int_{-\infty}^\infty \phi'(x)\int_{-\infty}^{nx}\frac{\sin(x')}{\pi x'}\,dx'\,dx\tag2 \end{align}$$

Using Property 3 in the Preliminaries section, there exists a number $C$ such that $\left|\phi'(x)\int_{-\infty}^{nx}\frac{\sin(x')}{x'}\,dx'\right|\le C\,|\phi'(x)|$. Inasmuch as $C|\phi'(x)|$ is integrable, the Dominated Convergence Theorem guarantees that

$$\begin{align} \lim_{n\to \infty}\int_{-\infty}^\infty \delta_n(x)\phi(x)\,dx&=-\lim_{n\to \infty}\int_{-\infty}^\infty \phi'(x)\int_{-\infty}^{nx}\frac{\sin(x')}{\pi x'}\,dx'\,dk\tag3\\\\ &=-\int_{-\infty}^\infty \phi'(x)\lim_{n\to \infty}\left(\int_{-\infty}^{nx}\frac{\sin(x')}{\pi x'}\,dx'\right)\,dx\\\\ &=- \int_{-\infty}^\infty \phi'(x)\underbrace{u(x)}_{\text{Unit Step}}\,dx\\\\ &=-\int_0^\infty \phi'(x)\,dx\\\\ &=\phi(0) \end{align}$$

Therefore, in the sense of distributions as given by $(3)$, we assert that $\lim_{n\to\infty}\delta_n(x)\sim \delta(x)$ as was to be shown!

Mark Viola
  • 179,405
5

Since $\delta_n(x)=n\delta_1(nx)$,$$\begin{align}\int_{\Bbb R}\delta_n(x)f(x)dx&=\int_{\Bbb R}\delta_1(y)f(y/n)dy\\\implies\lim_{n\to\infty}\int_{\Bbb R}\delta_n(x)f(x)dx&=\lim_{n\to\infty}\int_{\Bbb R}\delta_1(y)f(y/n)dy\\&\stackrel{*}{=}\int_{\Bbb R}\delta_1(y)\lim_{n\to\infty}f(y/n)dy\\&=f(0)\underbrace{\int_{\Bbb R}\delta_1(y)dy}_{1},\end{align}$$provided $f$ has the right behaviour for $\stackrel{*}{=}$ to be legal. For example, while $f$ being a Schwartz function suffices, $f$ being analytic doesn't.

J.G.
  • 115,835
  • Hi JG. I hope you're doing well and staying safe and healthy. This isn't quire correct as is. However, integration by parts first permits use of the Dominated Convergence Theorem. I've posted a solution herein. ;-) – Mark Viola Feb 24 '21 at 21:14
  • @MarkViola Your answer uses IBP to show why Schwartz suffices. I agree my answer is incomplete due to stating this sufficiency without proof, but it's unclear to me where it's incorrect. (You might have been referring to a typo I've fixed, where a $\delta_1$ was unfortunately missing its $_{1}$.) – J.G. Feb 24 '21 at 22:02
2

Make your substitution, then break up your

$$\int_{-\infty}^\infty f\left(\frac{u}{n}\right) \frac{\sin u}{\pi u} d u$$

as $I_1 + I_2,$ where $I_1$ is the integral from $-\sqrt{n}$ to $\sqrt{n}$ and $I_2$ is the rest. The point is that $I_1$ converges to $$\int_{-\infty}^\infty f(0)\frac{\sin u}{\pi u} du = f(0),$$ and $I_2$ converges to $0.$

Angelo
  • 12,328
Igor Rivin
  • 25,994
  • 1
  • 19
  • 40
  • You need to justify the interchange of the limit and integral. That is THE challenge in this problem. I've posted a rigorous solution herein. ;-) – Mark Viola Feb 24 '21 at 21:15
  • @MarkViola you seem to be misreading what I wrote. – Igor Rivin Feb 24 '21 at 22:05
  • I did not read incorrectly. Your claim that $I_1$ converges to $f(0)$ need justification. Moreover, you need to restrict the class of functions to which $f$ belongs. That omission along with the lack pf proof renders this post rather superficial. – Mark Viola Feb 24 '21 at 23:19
2

There are indeed many ways to understand the conclusion, and they are all worthwhile. But some may be less reassuring or persuasive than others... of course depending on taste and prior experience.

For me, at this moment, ... let $\langle,\rangle$ be the pairing of Schwartz functions and tempered distributions. By an extended form of Plancherel, $$ \langle f,u\rangle \;=\; \langle \widehat{f},\widehat{u}\rangle $$ for Schwartz $f$ and tempered $u$. If we get things normalized properly, for a suitable version $u_n(x)={\sin nx\over x}$, $\widehat{u}_n$ is the characteristic function of $[-n,n]$. This computation is most easily done by invoking Fourier inversion after computing the Fourier transform of that characteristic function. These converge as tempered distributions to $1$, which is the Fourier transform of $\delta$...

paul garrett
  • 52,465