3

I want to show that:

$$\lim_{\epsilon \to 0} \space \text{Im}\frac{1}{x+i \epsilon}=-\pi\delta(x)$$

This is my attempt:

  1. I assumed that $\text{Im}$ stands for the imaginary part. Therefore, $$\frac{1}{x+i \epsilon} \cdot \frac{x-i \epsilon}{x-i\epsilon}=\frac{x-i\epsilon}{x^2+\epsilon^2}=\frac{x}{x^2+\epsilon^2}-i \left (\frac{ \epsilon}{x^2+\epsilon^2} \right ) \implies\Im(z)=-\frac{1}{\frac{x^2}{\epsilon}+\epsilon}$$
  2. Taking the limit: $$\lim_{\epsilon \to 0} -\frac{1}{\frac{x^2}{\epsilon}+\epsilon}=0$$

But this would mean that $$\implies0=-\pi \delta(x) $$

However, this equality can't be true because:

$$\delta(x)=\begin{cases}\infty, \space \space\space x=0 \\ 0, \space \space \space x \not= 0 \end{cases}$$

which would mean that at $x=0$ the equality $0=-\pi \delta(x)$ is not true. What am doing wrong?

bluemoon
  • 969
  • 3
    What should you actually have at $0$? – Matias Heikkilä May 23 '16 at 19:50
  • 1
    You took the limit to infinity, rather than the limit to zero. – Neil May 23 '16 at 19:52
  • @Neil Whoops. Major mistake. Let me fix that. – bluemoon May 23 '16 at 19:53
  • @MatiasHeikkilä I am not sure what you mean. Is your question: What is the value of $$\frac{1}{\frac{x^2}{ \epsilon} +\epsilon}$$ at $\epsilon=0$? – bluemoon May 23 '16 at 19:57
  • 1
    Well, you show that the limit is $0$ if $x\neq 0$, this is consistent (I'm omitting to comment on the typo where you write $\varepsilon\to \infty$ instead of $\varepsilon\to 0$). But your whole argument, from 2. onwards, implicitly assumes $x\neq 0$: treat the case $x=0$ separately, now. – Clement C. May 23 '16 at 19:57
  • 1
    Consider evaluating the limit separately for the cases $x = 0$ and $x \not= 0$ – Neil May 23 '16 at 19:59
  • @ClementC. Ohh I see! At $x=0$ $$ \implies \lim_{\epsilon \to 0} -\frac{1}{\epsilon}= -\infty$$

    Thank you! :)

    – bluemoon May 23 '16 at 20:00
  • 3
    You need to look at the function's properties inside an integral. Simply saying the delta function is 0 at $x \neq 0$ and $\infty, x=0$ isn't going to cut it. – Paul May 23 '16 at 20:01
  • @Paul So you are saying my argument doesn't make sense? Are you saying I need to use properties like $\int_{-\infty}^{\infty} \delta(x)dx=1$ etc.? – bluemoon May 23 '16 at 20:04
  • 2
    Indeed the pointwise limit is quite irrelevant (and I do not know why some users send you on this road), rather you are supposed to show that, for every suitable test function $f$, $$\lim_{\epsilon\to0}\int_\mathbb Rf(x)\frac{-\epsilon}{x^2+\epsilon^2}dx=-\pi f(0).$$ Any idea to show this? – Did May 23 '16 at 20:04
  • 1
    (To react to the above: I (incorrectly, it seems) assumed $\delta(x)$ was used as a concise notation to express what the OP wrote, not in the distribution sense -- and that $\pi$ was the writer of the exercise being facetious. My bad.) – Clement C. May 23 '16 at 20:06
  • @Did Maybe this is a really stupid question but how does my initial statement: $$\lim_{\epsilon \to 0} \space \text{Im}\frac{1}{x+i \epsilon}=-\pi\delta(x)$$

    translate to : $$\lim_{\epsilon \to 0} \int_R f(x) \frac{-\epsilon}{x^2+\epsilon^2}dx=-\pi f(0)$$

    I just don't see the connection. In other words, how would I know that I have to show the satement you mentioned above?

    – bluemoon May 23 '16 at 20:10
  • 1
    Because this is the only rigorous definition of the Dirac distribution $\delta$. What were you trying to prove before I explained that? – Did May 23 '16 at 20:13
  • @Did Thank you for your help. I will try to read and understand the answer and links provided by Dr. MV. – bluemoon May 23 '16 at 20:16

1 Answers1

4

In THIS ANSWER and THIS ONE, I provide primers on the Dirac Delta.

For any smooth test function $\phi$, we have for any $\nu >0$, there exists a $\delta>0$ such that $|\phi(x)-\phi(0)|<\nu$ whenever $|x|<\delta$. Then, we can write

$$\begin{align} \lim_{\epsilon \to 0}\int_{-\infty}^\infty \phi(x)\text{Im}\left(\frac{1}{x+i\epsilon}\right)\,dx&=-\lim_{\epsilon \to 0}\int_{-\infty}^\infty \phi(x)\left(\frac{\epsilon}{x^2+\epsilon^2}\right)\,dx\\\\ &=-\lim_{\epsilon \to 0}\int_{|x|\ge \delta} \phi(x)\left(\frac{\epsilon}{x^2+\epsilon^2}\right)\,dx-\lim_{\epsilon \to 0}\int_{-\delta}^\delta \phi(x)\left(\frac{\epsilon}{x^2+\epsilon^2}\right)\,dx\\\\ &-\lim_{\epsilon \to 0}\int_{-\delta}^\delta \phi(x)\left(\frac{\epsilon}{x^2+\epsilon^2}\right)\,dx\\\\ &=-\lim_{\epsilon \to 0}\left(\int_{-\delta}^\delta (\phi(x)-\phi(0))\left(\frac{\epsilon}{x^2+\epsilon^2}\right)\,dx-\phi(0)\int_{-\delta}^\delta \left(\frac{\epsilon}{x^2+\epsilon^2}\right)\,dx\right)\\\\ &=-\pi \phi(0)-\lim_{\epsilon \to 0}\int_{-\delta}^\delta (\phi(x)-\phi(0))\left(\frac{\epsilon}{x^2+\epsilon^2}\right)\,dx\\\\ \end{align}$$

Note that we have for any $\nu>0$

$$\left|\int_{-\delta}^\delta (\phi(x)-\phi(0))\left(\frac{\epsilon}{x^2+\epsilon^2}\right)\,dx\right|\le 2\nu \arctan(\delta/\epsilon)\le \pi \nu $$

Therefore, for any smooth tests function $\phi$, we find

$$\lim_{\epsilon \to 0}\int_{-\infty}^\infty \phi(x)\text{Im}\left(\frac{1}{x+i\epsilon}\right)\,dx=-\pi \phi(0)$$

This is equivalent to the statement that

$$\lim_{\epsilon \to 0}\text{Im}\left(\frac{1}{x+i\epsilon}\right)\sim -\pi \delta(x)$$

in the sense of a regularization of the Dirac Delta.

Mark Viola
  • 179,405
  • Maybe overkill, but what about using the DCT to show that $$\lim_{\varepsilon\to 0} \int_{\mathbb{R}} \phi(x)\frac{\varepsilon}{x^2+\varepsilon^2}dx = \pi \phi(0)$$? Starting with $$\int_{\mathbb{R}} \phi(x)\frac{\varepsilon}{x^2+\varepsilon^2}dx=\int_{\mathbb{R}} \phi(\varepsilon u)\frac{du}{u^2+1}$$ for $\varepsilon > 0$ (with "$u=\varepsilon x$" and then applying the Dominated Convergence Theorem. (It looks more direct, even though it invokes a less elementary theorem.) – Clement C. May 23 '16 at 20:14
  • @ClementC. No overkill at all, on the contrary this might be the most natural way of thinking about the whole shebang if you have some notions of probability... And it works beautifully. – Did May 23 '16 at 20:16
  • Yes, that works after the substitution interestingly. One needs an indicator function as part of the integrand in that case. – Mark Viola May 23 '16 at 20:19
  • Really great answer. It will take me some time to work through it and through the links you have provided. Thank you very much. – bluemoon May 23 '16 at 20:19
  • @Did Thanks. Glad I did not stay astray up to the very end... – Clement C. May 23 '16 at 20:20
  • @bluemoon You're welcome. My pleasure. -Mark – Mark Viola May 23 '16 at 20:23
  • @Dr.MV Just a quick question. Can you recommend any "easy" and "accessible" textbooks that treat this in detail? I am quite new to this stuff and a bit shaky when it comes to the notation. – bluemoon May 23 '16 at 20:26
  • Here is one reference to Generalized Functions. I embedded links in the answers that I referenced in this post. – Mark Viola May 23 '16 at 20:28
  • @Dr.MV Thanks! Have a nice day! :) – bluemoon May 23 '16 at 20:36