I came across a definition of the δ-function as a generalized function in Mathematics Handbook for Science and Engineering which made me curious.

They first define a test function $\varphi: \mathbb R \to \mathbb C$ satisfying

1) $\varphi \in \mathcal C^\infty(\mathbb R)$

2) $\displaystyle\lim_{|t|\to\infty} t^p \frac{\mathrm d^q \varphi(t)}{\mathrm dt^q} = 0$ for all $p,q\geq 0$

and denote the class of all test functions $S$.

They then define a sequence $\displaystyle\{\varphi_k\}_{k=0}^\infty$ with $\varphi_k \in S$ as zero sequence in $S$ if and only if

$\displaystyle\lim_{n\to\infty} \max_{t \in \mathbb R} \left| \frac{\mathrm d^q \varphi(t)}{\mathrm dt^q}\right| = 0$ for all $p,q\geq 0.$

They then denote the value of a functional $f : S \to \mathbb C$ as $(f|\varphi)$ and define a distribution as a continuous linear functional on $S$, i.e. for all $\varphi, \psi \in S$ and $\alpha, \beta \in \mathbb C$ we have

$(f|\alpha \varphi + \beta \psi) = \alpha(f|\varphi) + \beta(f|\psi)$


$\displaystyle\lim_{k\to\infty} (f|\varphi_k) \equiv 0$ for all zero sequences $\displaystyle\{\varphi_k\}_{k=0}^\infty$ in $S$

They then let $g : \mathbb R \to \mathbb R$ be a piece-wise continuous function satisfying the integral equation

$\displaystyle\int_{\mathbb R} \left( 1+ t^2\right)^{-m} \left|g(t)\right| \; \mathrm dt < \infty$ for some $m \in \mathbb Z.$


$(f|\varphi) = \displaystyle\int_{\mathbb R} g(t) \varphi(t) \; \mathrm dt$

is said to define a regular distribution. Non-regular distributions are called singular.

Finally, they define the Dirac $\delta$-function as a singular distribution by

$(\delta|\varphi) = \varphi(0)$

Now, surely you can claim that this is a sound definition. Could one prove the existence of a such functional?

  • $\begingroup$ What do you mean by "eventual uniqueness"? Uniqueness is completely obvious, since we have defined exactly what $\delta$ is. All that needs to be checked is that this definition is actually continuous and linear. $\endgroup$ – Eric Wofsey Dec 16 '17 at 23:12
  • 1
    $\begingroup$ The distributions defined here are often called tempered distributions and are those for which the Fourier transform can be defined. A bigger class of distributions is defined as continuous linear functionals on $C_c^\infty$, the space of infinately differentiable functions with compact support. $\endgroup$ – md2perpe Dec 16 '17 at 23:14
  • $\begingroup$ @EricWofsey. Probably Markus has confused with the Swedish word 'eventuell' meaning 'possible'. Thus it should actually read "possible uniqueness". $\endgroup$ – md2perpe Dec 16 '17 at 23:16
  • $\begingroup$ @EricWofsey Yeah, obviously the uniqueness is trivial. What's not trivial is that it's a valid distribution. $\endgroup$ – Markus Klyver Dec 16 '17 at 23:23

If there was another distribution $\eta$ with $(\eta|\varphi) = \varphi(0)$ for all test functions $\varphi$, then $(\eta-\delta|\varphi) = (\eta|\varphi) - (\delta|\varphi) = \varphi(0) - \varphi(0) = 0$

Hence $\eta-\delta = 0\iff \eta = \delta$ which shows uniqueness.

Existence is simply a consequence of the definition. $\delta:S\to\mathbb C, \varphi \to \varphi(0)$ *is* a continuous linear functional on S. Hence $\delta$ exists as an element of the dual space $S^*$.

Remark: Obviously every element of $S^*$ is unique. (Either $a=b$ or $a\neq b$). My guess is that you rather confused this with the observation that for regular distributions, there might be different functions $f,g$ inducing the same distribution. If $\int \varphi(x) f(x) dx = \int \varphi(x) g(x) dx$ for all test functions $\varphi$, we can only conclude that $f(x)-g(x)$ is orthogonal to all test functions. Depending on the space of test functions such $f,g$ might exist or not. This however does not apply to the dirac delta. The reason we still write $\int \varphi(x)\delta(x) dx$ instead of say $\delta(\varphi)$ is because of Riesz's theorem.

The theorem states that if $H$ is a Hilbert space, then any coninuous linear functional $\psi$ on $H$ takes the form $\psi(\varphi) = (f_\psi,\varphi)$ for some $f_\psi\in H$. Note that $S$ is not a Hilbert-space! Since $\delta$ is provably non-regular, you cannot find $f_\delta\in S$ such that $\delta(\varphi) = (f_\delta,\varphi)$ for all $\varphi$. But it motivates why we carry on with the notation $\int \varphi(x)\delta(x) dx :=\delta(\varphi)$ as if the dirac delta existed as a function.

  • $\begingroup$ Thank you for you answer, but how does $(\eta - \delta|\varphi) = 0 \implies \eta = \delta$? $\endgroup$ – Markus Klyver Dec 16 '17 at 22:38
  • $\begingroup$ I also find it strange you can just "declare" the existence of $\delta$ without actually showing a such functional exists. How do we know a functional $\delta$ satisfying $\begin{cases}(\delta|\alpha \varphi + \beta \psi) = \alpha(\delta|\varphi) + \beta(\delta|\psi) \\ \displaystyle\lim_{k\to\infty} (\delta|\varphi_k) \equiv 0 \\ (\delta|\varphi) = \varphi(0) \end{cases}$ really exists? $\endgroup$ – Markus Klyver Dec 16 '17 at 22:43
  • $\begingroup$ @K.Power The function $x \mapsto x^2$ exists because we can show that a such function really satisfies the definition of a function (which may depend on which foundation you might use). So, to re-state my question: how do we know $\delta|\varphi) = \varphi(0)$ is actually valid and sound? That is, how do we know $(\delta|\varphi) = \varphi(0)$ really defines a singular distribution? $\endgroup$ – Markus Klyver Dec 16 '17 at 22:56
  • 1
    $\begingroup$ Tjenare, @MarkusKlyver! Two functions are equal iff their values are equal for all arguments. In the case of linear functionals this means that the functionals $u$ and $v$ are equal iff $(u|\varphi) = (v|\varphi)$ for all $\phi$ in the domain. $\endgroup$ – md2perpe Dec 16 '17 at 23:08
  • 1
    $\begingroup$ @MarkusKlyver Check out my remark. I am curious about your idea what 'existence' is supposed to mean. By definition, the dirac delta is the function that maps an element $\varphi$ of the function space $S$ onto its value at the origin $\varphi(0)$. This exists first and foremost simply as a set-theoretic map. There is no doubt about its existence. Now its easy to show that this map indeed defines a continuous linear functional on $S$. And quite a bit harder to prove it is non-regular. $\endgroup$ – Hyperplane Dec 16 '17 at 23:19

I want to give a second remark about the existence of $\delta$ as a function since this seems to cause you trouble.

Consider a map $f:A\to B$. Then $f$ can be identified with the set $\{(a,f(a)) \mid a\in A\} \subseteq A\times B$.

Hence questioning the existence of $\delta:S\to\mathbb C, \varphi \to \varphi(0)$ is the same as questioning whether or not the set

$$\{(\varphi,\varphi(0))\mid\varphi\in S\} \subset S\times \mathbb C$$



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.