4

Can anyone rigorously prove this? $$\int dx \, \delta(x-\alpha)\delta^{\prime} (x-\beta) = \delta^{\prime} (\alpha-\beta).$$

1 Answers1

3

In THIS ANSWER, I provided a primer on the Dirac Delta and the Unit Doublet distributions.

Here, we have for any test function $f$, the distribution $\int_{-\infty}^\infty\delta(x-\alpha)\delta'(x-\beta)\,dx$ satisfies the following:

$$\begin{align} \int_{-\infty}^\infty f(\alpha)\left(\int_{-\infty}^\infty\delta(x-\alpha)\delta'(x-\beta)\,dx\right)\,d\alpha&=\int_{-\infty}^\infty \delta'(x-\beta)\left(\int_{-\infty}^\infty f(\alpha)\,\delta(x-\alpha)\,d\alpha\right)\,dx\\\\ &=\int_{-\infty}^\infty \delta'(x-\beta)\,f(x)\,dx\\\\ &=-f'(\beta)\\\\ &=\int_{-\infty}^\infty \delta'(\alpha-\beta)\,f(\alpha)\,d\alpha \end{align}$$

We also have

$$\begin{align} \int_{-\infty}^\infty f(\beta)\left(\int_{-\infty}^\infty\delta(x-\alpha)\delta'(x-\beta)\,dx\right)\,d\beta&=\int_{-\infty}^\infty \delta(x-\alpha)\left(\int_{-\infty}^\infty f(\beta)\,\delta'(x-\beta)\,d\beta\right)\,dx\\\\ &=\int_{-\infty}^\infty \delta(x-\alpha)\,f'(x)\,dx\\\\ &=f'(\alpha)\\\\ &=\int_{-\infty}^{\infty}\delta'(\alpha -\beta)f(\beta)\,d\beta \end{align}$$

Therefore, in terms of Generalized Functions, we have the equivalence

$$\delta'(x-\beta)=\int_{-\infty}^\infty\delta(x-\alpha)\delta'(x-\beta)\,dx$$

Using a more compact notation, we have

$$\begin{align} \langle f,\langle \delta'_{\beta},\delta_{\alpha}\rangle \rangle_{\alpha}&=\langle \delta'_{\beta},\langle f,\delta_{\alpha}\rangle_{\alpha} \rangle\\\\ &=\langle \delta'_{\beta},f\rangle\\\\ &=-f'(\beta) \end{align}$$

A similar notation is applicable where we take the distribution over $\beta$.

Mark Viola
  • 179,405
  • 1
    This doesn't compute. What if you evaluate in a different order, i.e. $\langle \delta_{\alpha},\langle f,\delta'_{\beta}\rangle \rangle$? – A.S. Dec 03 '15 at 05:05
  • 1
    The two results are not equal unless $\alpha=\beta$, so I don't see how your definition of $\langle \delta'{\beta},\delta{\alpha}\rangle$ works. In some sense $\langle \delta'{\beta},\delta{\alpha}\rangle=0$ for $\alpha\ne \beta$. – A.S. Dec 03 '15 at 05:19
  • @A.S. If we wish to interchange the order, we have implicitly changed the parameter from $\alpha$ to $\beta$ over which we are computing the distribution. – Mark Viola Dec 03 '15 at 05:22
  • Then your test functions needs to be $f(\alpha,\beta)$ since we have s distribution in 2 parameters. – A.S. Dec 03 '15 at 05:24
  • 1
    Why not consider both as a fixed number then? That's a much more natural assumption than assuming one is variable and the other is a parameter - especially since $\alpha$ and $\beta$ are not symmetric. Either both are parameters or both are variables. – A.S. Dec 03 '15 at 05:28
  • @A.S. I added another development where the distribution is over $\beta$. This will hopefully clarify. Thank you. - Mark – Mark Viola Dec 03 '15 at 05:48
  • 1
    Mark, it clarifies the mechanics of what follows but not the original choice to treat one of $\alpha,\beta$ as a fixed parameter and the other as a variable of a distribution. The OP treats both as a parameter (as I think was intended), so the real question is if we can define $\int_R\delta(x-\alpha)\delta'(x-\alpha)$ - or more generally $\delta\delta'$ - which is a product of two distribution with non-disjoint singular supports. – A.S. Dec 03 '15 at 06:02
  • @A.S. Yes, I agree now. Apology for not heeding your advice earlier. ;-)) And +1 for your insightful comments. - Mark – Mark Viola Dec 03 '15 at 06:03
  • Good. One way to make sense of the formula (as Ian mentioned) is to view it as a convolution: $\delta_a*\delta'\beta=\delta{\beta-\alpha}'$ since otherwise (as a direct product) it's not defined for $\alpha=\beta$. – A.S. Dec 03 '15 at 06:45
  • Thanks for the reminder about how to formally deal with such things (I have long forgotten my applied math courses). One question though: if I approximate delta with a limit of a series of functions (e.g. Gaussian or others), am I always guaranteed the same answer regardless of what series I use? – Victor V Albert Dec 03 '15 at 16:03
  • You're welcome. My pleasure. And here is another answer I posted that discusses regularization of the Dirac Delta. Please let me know if this helps. It contains embedded links to yet other answers I've posted herein on the Dirac Delta. – Mark Viola Dec 03 '15 at 16:19
  • I've scanned your answers. Since you use many different regularizations, it seems like I can pick my favorite and work with it without checking the others then... – Victor V Albert Dec 03 '15 at 16:36
  • 1
    @VictorVAlbert Sure. Choose the one that best facilitates your specific problem. And remember, I provided only a small sample of regularizations - just the tip of the iceberg. - Mark – Mark Viola Dec 03 '15 at 17:08