A single Dirac delta has a distributional derivative $\delta'$ defined in the sense that $$ \int_{-\infty}^{\infty} \textrm{d}x \, \delta'(x) \, \phi(x) = -\int_{-\infty}^{\infty} \textrm{d}x \, \delta(x) \, \phi'(x) \, . $$ I understand this comes from the integral of the total derivative of $\delta(x) \, \phi(x)$ vanishing, $$ \int_{-\infty}^{\infty} \textrm{d}x \, \frac{\textrm{d}}{\textrm{d} x} \big( \delta(x) \, \phi(x) \big) = \delta(x) \, \phi(x) \, \big|_{-\infty}^{\infty} = 0 \, , $$ which means the product rule for differentation gives the above expression defining $\delta'$. If we instead consider the Dirac comb $$ \mathrm{III}(t) = \sum_{k=-\infty}^{\infty} \, \delta(x - kT) \, , $$ which is periodic with period $T$, is the same total derivative trick valid - $$ \int_{-\infty}^{\infty} \textrm{d}x \, \frac{\textrm{d}}{\textrm{d} x} \big( \textrm{III}(x) \, \phi(x) \big) = 0 \, ? $$ Allowing us to define the distributional derivative of the Dirac comb analogously to the Dirac delta $$ \int_{-\infty}^{\infty} \textrm{d}x \, \textrm{III}'(x) \, \phi(x) = -\int_{-\infty}^{\infty} \textrm{d}x \, \textrm{III}(x) \, \phi'(x) \, . $$


In fact, if stop reasoning in terms of integrals ans start reasoning in terms of distributions and distributional derivatives, then this result is quite easy to show. Let us note $\phi $ - am arbitrary test function (i.e. compactly supported and $C^\infty$)

We start with a delta-distribution $\delta_p$ given by $$\langle \delta_p,\phi\rangle = \phi(p)$$ and its derivative $\delta_p'$, for which the definition of distributional derivative (yes, it has common roots with integration by parts) gives $$\langle \delta_p' ,\phi\rangle=-\langle \delta_p,\phi'\rangle = -\phi'(p).$$

Now the Dirac comb $B = \sum_{k\in\Bbb Z} \delta_{kT}$ is a well-defined distribution, because $$\langle B,\phi\rangle = \sum_{k\in\Bbb Z} \phi(kT),$$ and the latter series converges because $\phi$ is compactly supported. I will omit the continuity part of the definition, it is an easy exercise.

Since $B$ is a distribution, it has a distributional derivative - it is one of the major resons to use distributions. Moreover, we have an explicit definition: $$\langle B',\phi\rangle=-\langle B,\phi'\rangle = -\sum_{k\in\Bbb Z} \phi'(kT),$$ and the latter series converges for the exact same reasons as we mentioned earlier.

Finally, by testing against all possible test functions $\phi$ you can conclude that $B'$ has the form $ \sum_{k\in\Bbb Z} \delta'_{kT}$.


Another approach would be to show that the distributions $B_n= \sum_{k\in\Bbb Z,\,|k|\le n} \delta_{kT}$ converge to $B$ in the distributional sense (easy to show by definition of this convergence). It implies - another great property of distributional derivatives - that $B'_n$ converges to $B'$, and the explicit formula for $B'_n$ is quite obvious.

| cite | improve this answer | |
  • 1
    $\begingroup$ Well done. And it is good to see one dispel the notion that the Dirac Delta is a distribution and not a function, and that the linear functional $\langle \phi,\delta\rangle$ is not an integral. (+1) $\endgroup$ – Mark Viola Apr 4 '17 at 23:08

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.