Is this proof formal enough? I plan on being a theoretical physicist one day, so I want to get into the good habit of being mathematically strict.

My proof:

$u=x$; $du=dx$

$v = \delta (x)$; $dv = -\delta (x)$

$$\int x \frac{d}{dx}(\delta (x))dx = x\delta (x) - \int \delta (x)dx = \int -\delta (x) dx$$

$$x\delta (x) = 0$$

We now integrate both sides in order to properly use the Kronecker delta function.

$$\int x \delta (x) dx = \int 0dx$$

It is known that the $\int_{-\infty}^{+\infty} f(x)\delta (x)=f(0)$. Thus,


  • $\begingroup$ Mathematically, the Dirac delta is a distribution (or what physicists call a generalized function), which is meaningful when you pair this with test functions: if $\varphi$ is any smooth compactly supported function, then as Zachary Selk computed, we get $$ \langle x \delta'(x), \varphi(x) \rangle = \int x \delta'(x) \varphi(x) \, \mathrm{d}x = -\varphi(0). $$ And we know exactly what distribution gives this result: $-\delta(x)$! $\endgroup$ Sep 3 '16 at 23:53
  • 2
    $\begingroup$ more explicitly, by definition of the distributional derivative $\langle T',\varphi \rangle = -\langle T,\varphi' \rangle$ and the multiplication of a distribution $T$ by a function $f \in C^\infty$ : $ \ \ \langle f T,\varphi \rangle = \langle T,f\varphi \rangle$ we have $\langle x \delta', \varphi \rangle =\langle \delta', x \varphi \rangle =-\langle \delta, (x\varphi)' \rangle = -\langle \delta, \varphi+x \varphi' \rangle =-\langle \delta, \varphi \rangle$ @SangchulLee $\endgroup$
    – reuns
    Sep 4 '16 at 0:17
  • $\begingroup$ @user1952009, That's right. In order to be precise, we should work on paring itself and using the definition of derivative of distribution. Thank you for pointing out this. $\endgroup$ Sep 4 '16 at 0:32
  • 1
    $\begingroup$ @SangchulLee $Generalized\ Functions$ were introduced, in 1935, by the russian Sergei L. Sobolev who WAS a mathematician. $\endgroup$ Sep 4 '16 at 4:08
  • $\begingroup$ @FelixMarin Thank you for correcting that! It's interesting to see that the seemingly intuitive term generalized functions is less popular in math literature than more ambiguous term distribution, which has at least two more independent meanings. $\endgroup$ Sep 4 '16 at 5:28

Let $f$ be any smooth compactly supported function. Then:

$$\int x\delta'(x)f(x)\ dx=\int\delta'(x) xf(x)\ dx=-\int\delta(x)(xf'(x)+f(x))\ dx=-0f'(0)-f(0)=-f(0)$$

  • 1
    $\begingroup$ @whatwhatwhat Integration by parts gives that. The intermediate term vanishes because $f$ is compactly supported. $\endgroup$ Sep 3 '16 at 23:52
  • 2
    $\begingroup$ @whatwhatwhat What's your definition of $\delta$? $\endgroup$
    – user223391
    Sep 3 '16 at 23:59
  • 1
    $\begingroup$ @whatwhatwhat, the Kronecker delta is meaningfully only under summation: $ \sum_x f(x) \delta(x) = f(0) $. Since integrals are intuitively sums of infinitesimals, say $\sum_i f(x_i) \Delta x_i$, the corresponding version of $\delta$ must be adjusted so that $ \sum_i f(x_i)\delta(x_i) \Delta x_i = f(0)$. In other words, the Dirac delta must compensate the infinitesimal factor $\Delta x$. The issue is that, there is no legitimate function that do this job. So we have to consider a wider class of objects that generalizes functions, and distributions are exactly designed for this purpose. $\endgroup$ Sep 4 '16 at 0:09
  • 1
    $\begingroup$ nice discussion .. Dirac, Kronecker .. resembles old times philosophycal discussions: would be the case to fix the postulates first ? $\endgroup$
    – G Cab
    Sep 4 '16 at 1:09
  • 2
    $\begingroup$ @whatwhatwhat, Although one conventionally use function notation for the Dirac delta, it is not a function and thus its value at a point is meaningless. So, in general, information on distribution cannot be extracted pointwise by feeding a point as argument. Instead, we can use test functions to extract information. This also means that essentially the only way we manipulate and identify a distribution is to feed them any possible test functions (by the way which we call 'pairing') and see what happens. Zachary Selk's answer demonstrates how this can be done. $\endgroup$ Sep 4 '16 at 2:13

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.