0
$\begingroup$
  1. I was told that the multivariate Dirac delta function $\delta(\mathbf x), \mathbf x \in \mathbb R^n$ with

\begin{align} 0 &= \delta(\mathbf x) \quad \forall \mathbf x \neq (0,0,...,0) \\ 1 &= \int_{-\infty}^\infty \cdots \int_{-\infty}^\infty \delta(\mathbf x) \; \text d x_1 \cdots \text d x_n \end{align}

is equivalent to the product of the it's marginals, so

\begin{align} \delta(\mathbf x) &= \delta(x_1,x_2,...,x_n) = \delta(x_1) \cdot \delta(x_2) \cdots \delta(x_n) = \prod_{i=1}^n \delta(x_i) \end{align}

Why is that the case?

  1. Also, I was told that a Dirac delta function that not only non-zero for the origin

\begin{align} 0 &= \delta(x_1, x_2, ..., x_m, ... , x_n) = \prod_{i=1}^m \delta(x_i) \quad \forall x_1, x_2, ... , x_m \neq 0 \\ 1 &= \int_{-\infty}^\infty \cdots \int_{-\infty}^\infty \prod_{i=1}^m \delta(x_i) \; \text d x_1 \cdots \text d x_n \end{align}

for $m<n$. Why does the second equation (still) hold (while still integrating over the entire n-dimensional set/space)?


EDIT: For illustrative purposes I imagine

For illustration I image the full case in 1 to be equivalent to

\begin{align} \delta(\mathbf x) = \begin{cases} \lim\limits_{a\rightarrow0} \quad \dfrac{1}{a^n} & \forall x_i \in [-\frac a2,\frac a2], 1\le i\le n \\[6pt] \quad 0 & \text{otherwise} \end{cases} \end{align}

Thais how it was introduced to me at university.

So, I was guessing I would need for the second case something like

\begin{align} \delta'(\mathbf x) = \begin{cases} \lim\limits_{a\rightarrow0} \quad \dfrac{a^{n-m}}{a^m} & \forall x_i \in [-\frac a2,\frac a2], x_j, \in [-\frac2a,\frac2a], 1\le i\le m < j\le n, \\[6pt] \quad 0 & \text{otherwise} \end{cases} \end{align}

so that the "volume" / integral over $\delta'$ would be the support multiplied by the "height" of the Dirac:

$$ \frac{a^m}{a^{n-m}} \cdot \frac{a^{n-m}}{a^m} = 1 $$

independent of $a$.

So, if we consider $\delta'$ to be measures for a measure space $(X, \Sigma, \delta')$, we take the $\delta'$-measure for an $S_1\subset X$ and then translate $S_1$ only in directions of $x_i, m<i\le n$ to $S_2$, then the $\delta'$-measure of $S_2$ should be the same as for $S_1$.

But I do not see how this relates to what I wrote in 2 above (before this edit). My main issue is, that $\delta(x_1) \cdot \delta(x_2) \cdots \delta(x_m)$ seems to "lack" the "normalization" so that the integral becomes 1: For example if we have $m=1, n=2$, then shouldn't the "height" of the Dirac delta function be 1? Because we have $a$ in one direction and $1/a$ in the other, so to get 1 for the integral, we need 1 as the height. But this does not seem to be incorporated into $\delta(x_1) \cdot \delta(x_2) \cdots \delta(x_m)$.

$\endgroup$
6
  • $\begingroup$ The Dirac $\delta$ isn't a function. $\endgroup$
    – Arthur
    Nov 28 '20 at 22:49
  • $\begingroup$ These are 'working rules' for the Dirac 'function', you would need the theory of distributions to make precise. $\endgroup$
    – copper.hat
    Nov 28 '20 at 23:34
  • $\begingroup$ @Arthur: I am not sure what you mean: en.wikipedia.org/wiki/Dirac_delta_function $\endgroup$
    – Make42
    Nov 29 '20 at 16:59
  • $\begingroup$ @Make42 It's called a function, we use function-like notation when we use it, but it isn't a function. It doesn't have a well-defined value at the origin. Even your Wikipedia link says "[T]here is no function that has these properties". $\endgroup$
    – Arthur
    Nov 29 '20 at 17:05
  • $\begingroup$ @Arthur: Ok, what is it then? $\endgroup$
    – Make42
    Nov 29 '20 at 17:44
0
$\begingroup$
  1. Observe that \begin{align} \delta(x_1) \cdots \delta(x_n)=0 \Leftrightarrow x_1,...,x_n\neq0. \end{align} Just like it should be for $\delta(x_1,...,x_n)$. Also: \begin{align} \int_{-\infty}^\infty \cdots \int_{-\infty}^\infty \delta(x_1) \cdots \delta(x_n) dx_1\cdots dx_n= 1\cdots 1=1=\int_{-\infty}^\infty \cdots \int_{-\infty}^\infty \delta(x_1,...,x_n) dx_1\cdots dx_n \end{align} (they are equivalent functions/distributions)

  2. As the function $\Pi_{i=0}^m \delta(x_i)$ is not defined in the last n-m Variabels it is also integrated over, the integration is an integration over a Null Set. I think this is what is bothering you. But note that the function $\delta(x)$ itself only has a value other than zero at x=0, which also is a Null Set in the domain of the function (e.g. reel Numbers). One must conclude that the value of $\delta(0)$ is something like infinity (just for intuition), or else the integral would be zero. This is how you can think about why integrating over the dirac delta function is always going to evaluate to 1, regardless if you are integrating over Null Sets.

$\endgroup$
1
  • $\begingroup$ Thank you for your insights. I do not think that the integration over the null set is the issue regarding 2 for me. It is more how, we get the integration to be 1. I added my understanding so far into the question. The issue comes from the lack of "normalization" if you will. $\endgroup$
    – Make42
    Nov 29 '20 at 19:10

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.