• $(\Omega,\mathcal A)$ be a measurable space
  • $\omega\in\Omega$
  • $\delta_\omega$ denote the Dirac measureat $\omega$ on $(\Omega,\mathcal A)$
  • $E$ be a $\mathbb R$-Banach space
  • $\mathcal M$ denote the set of strongly $\mathcal A$-measurable $f:\Omega\to E$

If $f\in\mathcal M$, $$\langle\delta_\omega,f\rangle :=\int f\:{\rm d}\delta_\omega=f(\omega)$$ is well-defined. In that sense, $\delta_\omega$ can be thought of as an element of the algebraic dual space $\mathcal M^\ast$ of $\mathcal M$.

Now, assume

  • $(\Omega,\mathcal A)=(\mathbb R,\mathcal B(\mathbb R))$
  • $E=\mathbb R$

In the context of distribution theory, we can find the so-called scaling property of the Dirac delta function $$\langle\delta(a\;\cdot\;),f\rangle=\frac1{\left|a\right|}\left\langle\delta,f\left(\frac{\;\cdot\;}a\right)\right\rangle=\frac1{\left|a\right|}f(0),\tag1$$ where $a\in\mathbb R\setminus\left\{0\right\}$ and $f\in C^\infty(\mathbb R,\mathbb C)$.

My problem is that I don't understand how $\delta(a\;\cdot\;)$ is defined. Moreover, I'm aware of the usual proof of $(1)$ relying on the substitution rule. However, since the Dirac delta function is not a function, I don't understand why that rule is applicable.

It seems like one is assuming that there is a function $\delta$ such that the measure $\delta_x$ is Radon-Nikodým differentiable with respect to the Lebesuge measure $\lambda$ on $(\mathbb R,\mathcal B(\mathbb R))$ with $$\delta(\;\cdot\;-x)=\frac{{\rm d}\delta_x}{{\rm d}\lambda}\tag2$$ for all $x\in\mathbb R$. With that assumption, it's easy to see that $$\int f(y)\delta(a(y-x))\:\lambda({\rm d}y)=\frac1{|a|}\int f\left(\frac{\;\cdot\;}a\right)\delta(\;\cdot\;-ax)\:{\rm d}\lambda=\frac1{|a|}f(x)\tag3$$ for all $x\in\mathbb R$.

However, since $\delta_x$ is singular with respect to $\lambda$ for all $x\in\mathbb R$, $(2)$ is not well-defined. So, how can we state and prove $(1)$ in a rigorous way?

Please note that I know almost nothing about distribution theory. So, it would be great if there would be a purely measure theoretic answer.


In fact (1) is the definition of $\delta(a\cdot)$. This is exactly how such things are "always" defined.

To save typing, let's say once and for all $\phi$ and $\psi$ are test functions on $\Bbb R$ and $u$ is a distribution. If $D$ is the derivative we note that $$\int(D\phi)\psi=-\int\phi D\psi,$$so we define $Du$ by $$(Du)(\psi)=-u(D\psi).$$If $\tau_a$ is the translation $\tau_a\phi(t)=\phi(t-a)$ then $$\int(\tau_a\phi)\psi=\int\phi\tau_{-a}\psi,$$hence the definition $$(\tau_au)(\psi)=u(\tau_{-a}\psi).$$Giving our dilation operators a name, say $a>0$ and define $\Delta_a \phi(t)=\phi(at)$. Then $$\int(\Delta_a\phi)\psi=\frac1{|a|}\int\phi\Delta_{1/a}\psi,$$hence the definition $$(\Delta_au)(\psi)=\frac1{|a|}u(\Delta_{1/a}\psi).$$

  • 1
    $\begingroup$ I think I've got it. In the concrete case of the question $\Delta_a\delta_x$ can even be extended to $\mathcal M$ (as the right-hand side of the definition is well-defined for any function of $\mathcal M$), right? (And note that you've missed to take the absolute value of $a$ in the denominator.) $\endgroup$ – 0xbadf00d Sep 30 '18 at 16:26

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.