9
$\begingroup$

When reading about the unit step or Heaviside function and it's derivative, the ramp function I encountered the following characterisations of the Heaviside function $\mathscr{H}$:

  1. $x \cdot \mathscr{H}(x) = \int_{- \infty}^{x} \mathscr{H}(t) dt$
  2. $(\mathscr{H} \ast \mathscr{H})(x) = \int_{- \infty}^{x} \mathscr{H}(t) dt$, where $\ast$ is convolution.

Are there any other functions $\Phi: \mathbb{R} \to \mathbb{R}$ with $\Phi \not \equiv 0$ satisfying either of those conditions on $\mathbb{R}$? I wrote out the definition of convolution but didn't come any further since surely, $\Phi$ could to be defined piece-wisely.

Some ideas: Plugging $x = 0$ into (1) and (2) yields $$ \int_{-\infty}^{0} \mathscr{H}(t) dt = \int_{\mathbb{R}} \mathscr{H}(t) \mathscr{H}(-t) dt = 0, $$ meaning that the function and its reflection upon the $y$-axis "cancel out".


Update 1: $f$ is not differentiable in $x = 0$.

Case 1: $f$ is differentiable and fulfils one above requirement Now we can differentiate $$ x \cdot f(x) = \int_{-\infty}^{x} f(t) dt $$ to obtain $$ x \cdot f'(x) + f(x) = f(x) \implies x \cdot f'(x) = 0 $$ Since $x \not\equiv 0$ we know that $f'(x)$ is constant, implying that it is constant lets say $f(x) \equiv c \in \mathbb{R}$. (Since $f$ is differentiable it is continuous and therefore can't be piecewise constant with discontinuous jumps like a step function) But this is a contraction since $\int_{-\infty}^{x} c dx = \infty$ for all $x,c \in \mathbb{R}$.

Case 2: $f$ has an antiderivative $F$ defined on $\mathbb{R}$ Then, by the FTOC we can write $$ x \cdot f(x) = F(x) - F(-\infty) \implies x \cdot f(x) + F(-\infty) = F(x). $$ Since $f$ fulfils the above condition we deduce $F(-\infty) < \infty$. We know that $F$ is differentiable with $F' = f$. Since $F(-\infty)$ is a constant irrelevant to the differentiability of the LHS we conclude that $f(x)$ is differentiable on $\mathbb{R} \setminus \{0\}$. Because of case 1, $f$ can't be differentiable in zero.

Is this correct and if yes, how can we continue from here?


I also noticed that $x \cdot \Phi(x) = (\Phi \ast \Phi)(x)$ will also hold for the "reverse Heaviside function" $\Phi := \mathbb{1}_{\{x<0\}}$.

$\endgroup$
  • 2
    $\begingroup$ You may find the following interesting: $$\int_{-\infty}^x f(t)\,dt=\int_{-\infty}^\infty f(t)\mathscr{H}(x-t)\,dt=(f\ast \mathscr{H})(x).$$ Thus the first antiderivative of $f(x)$ can itself be expressed as a convolution against the Heaviside step function. (This in particular makes it obvious why $\int_{-\infty}^x \mathscr{H}(t)\,dt=(\mathscr{H}\ast \mathscr{H})(x).$) $\endgroup$ – Semiclassical Aug 20 at 21:34
3
+50
$\begingroup$

Suppose $\Phi(x)$ is a distribution, then the only solutions, $f(x)$, to equation 1 are multiples of the Heaviside function. This is because, as you already stated, $x\cdot f'(x)=0$, which has only $\delta (x)$, or multiples, as solutions for$f'(x)$. Note that $x\cdot \delta'(x)=\delta(x)$ not $0$ and similarly for higher derivatives. Of course $\int_{-\infty}^{x} \delta(t)dt=\mathscr{H}(x)$.

It is straightforward to show that $f(x)=e^{icx}\mathscr{H}(x)$ solves equation 2 for any $c$. Invoking the idea of $f(x)$ as a distribution on the space of smooth test functions which fall off rapidly at large $x$, we can Fourier transform equation 2. I note that distributions do not in general have convolutions with themselves, but assume that the convolution is ok. Then equation 2 becomes $$-i\tilde{f}'(\omega)=2\pi \tilde{f}^2(\omega),$$ which has solutions $$\tilde{f}(\omega) = \frac{i}{2\pi(\omega+c)},$$ where $\tilde{f}$ denotes the Fourier transform of $f$. Fourier transforming back, we get $f(x)=e^{icx}\mathscr{H}(x)$.

$\endgroup$

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.