0

Prerequisites for the answer:

I am trying to understand which conditions makes the derivative of non-everywhere-differentiable continuous time-limited functions to be bounded $\sup_t |f'(t)|<\infty$, so the derivative $f'(t)$ could be discontinuous, but cannot diverge to infinity in these discontinuities (because the requested restriction will be instantly violated), so $\mathbf{\sup_t |f'(t)|<\infty}$ is a requirement. Also, if $f(t)$ is differentiable, the derivative will be bounded so $f(t)$ has to have at least one point where is not differentiable (as example, having one finite-size-gap jump-discontinuity).

Also, the function $f(t)$ must have a Fourier Transform, so the function $f(t)$ must be Lebesgue integrable, or equivalently, fulfill that $\int_{t_0}^{t_F} |f(t)|dt < \infty$, and also, the amount of discontinuities must be finite, so nowhere-differentiable functions are not allowed.

With saying that $f(t)$ is "Time-Limited" I am meaning that the function $f(t)$ has a starting point at a time $t_0$ and and ending point at a time $t_F$ with $t_F>t_0$, so, $f(t) = 0,\,t<t_0$ and $f(t) = 0,\,t>t_F$, and can be any function that fulfill the other requirements within the boundaries of its domain $\partial t =\{t_0,\,t_F\}$. Since at the discontinuity that could rise at the edges of the domain could made some problems with the Fourier Transform, if you wish you can assume that $f(t_0)=f(t_F)=0$, but this issue can be easily avoided using the transform $\mathring{\mathbb{F}}\{\cdot\}$ is introduced here, so I will continue assuming that this problem is nonexistent (and using $\mathbb{F}\{\cdot\}$ or $\mathring{\mathbb{F}}\{\cdot\}$ indistinguishable as I need it).

Please note that time-limited functions are compact-supported, so since I am interesting in continuous functions $f(t)$, they will be also bounded functions $\sup_t |f(t)| < \infty$, and since are also Lebesgue integrable, both conditions together imply also that $f(t)$ is of finite energy $\int_{t_0}^{t_F} |f(t)|^2 dt < \infty$.

The problem

Since the traditional upper bound of the derivative $||f'||_\infty \leq ||wF(w)||_1$ will diverge if $f'(t)$ have discontinuities $\Rightarrow ||wF(w)||_1 \to \infty$ (here $F(w)$ is the Fourier transform of $f(t)$), I am looking for alternative upper bounds that could be useful for time-limited continuous functions $f(t)$ with a finite number of discontinuities in $f'(t)$.

On the Wikipedia website of Norms appears some inequalities that fulfills some of the vector norms:

  1. $||x||_2 < ||x||_1 < \sqrt{n}\,||x||_2 $
  2. $||x||_\infty < ||x||_2 < \sqrt{n}\,||x||_\infty $
  3. $||x||_\infty < ||x||_1 < n\,||x||_\infty $

But through some questions I have already done here in SE (1, 2, 3, 4, 5), I know now that these inequalities are not general for functions, which some inequalities could be even inverted for the same function depending on the integration domain.

But experimenting with the following functions $f(t)=1-|t|,\,|t|\leq 1$ and $f(t)=e^{-|t|},\,|t|\leq 1$, that have bounded $||f'(t)|| < \infty$ but makes $||wF(w)|| \to \infty$ since they have bounded-size jump-discontinuities en $f'(t)$ (examples I am taking from the answer to 2), I believe that maybe some of the inequalities could apply for time-limited continuous functions.

The question:

Does any of these inequalities holds in general for time-limited continuous functions $\mathbf{f(t)}$ with bounded derivative $\mathbf{\sup_t |f'(t)|<\infty}$?

  1. $\sup_t |f'(t)| \leq \sqrt{(t_F - t_0)\,\int\limits_{t_0}^{t_F}|f'(t)|^2\,dt}$? (this is the main one I am interested in $||f'||_\infty \leq \sqrt{t_F-t_0}\,||f'||_2$)
  2. $\sup_t |f'(t)| \leq \sqrt{\frac{\pi}{2}\,\int\limits_{t_0}^{t_F}|f'(t)|^2\,dt}$?
  3. $\sup_t |f'(t)| \leq \int\limits_{t_0}^{t_F}|f'(t)|\,dt$?
  4. $\sup_t |f'(t)| \leq \int\limits_{t_0}^{t_F}|f(t)|\,dt$?

I am asking explicitly if they holds for general domain $\mathbf{-\infty<t_0<t_F<+\infty}$.

Joako
  • 1,380

1 Answers1

1

No, none of these bounds hold.

First remark, the derivative is a local object, so the fact to be time limited will not help to bound a derivative. The derivative at a point do not change if you change your function at other points.

Then, defining $g = f'$, you can see that what you are asking has (almost) nothing to do with derivatives and just with bounding the $L^\infty$ norm by the $L^2$ norm. This is in general false: there are a lot of unbounded functions that are square integrable, such as $g(t) = |t|^{-1/4}$ with $t\in[-1,1]$ and $g(t)=0$ on the complement.

The only thing that is special about $g$ being the derivative of a compactly supported function is the fact that you have to ensure that $∫_{t_0}^{t_f} g = 0$, and so you can replace the above example by $g(t) = |t|^{-5/4}t$ (for $t_0 = -1$ and $t_f=1$, but you can translate and dilate the coordinates to get any $t_0$ and $t_f$). In terms of $f$, it leads to $$ f(t) = \frac{4}{3} \,(t^{3/4} - 1) \ \text{ if } t\in[-1,1] $$ and $f(t)=0$ if $|t|>1$ as a counterexample to all your inequalities.

LL 3.14
  • 12,457
  • I see that is false since there are unbounded functions for which the integrals are finite, but in the case of bounded functions (as I asked in the question prerequisites that $\sup_t |f'(t)|<\infty$), All these inequalities still going to be false?... for me is really counter-intuitive/unexpected that there are no other known upper bound for the derivative than $||wF(w)||_1$, which fails to work if $f'(t)$ has discontinuities but bounded. – Joako Dec 08 '21 at 22:28
  • 1
    The derivative $g$ is unbounded, but the function $f$ is bounded ... usually in analysis, the derivative is a good thing to consider so one prefers to keep the norm $|f'|_{L^\infty}$ (which is called the Lipschitz semi-norm, sometimes written $\dot W^{1,\infty}$ or $\dot C^{0,1}$) instead of trying to bound it using the Fourier transform. – LL 3.14 Dec 08 '21 at 22:45
  • I understand that part, maybe I am explain myself wrong so far: if the derivative $g$ is continuous the function will be differentiable (because is compact-supported), and the bound $||wF(w)||_1$ will be useful; if the derivative $g$ is unbounded, well nothing will be useful since already has point with infinite value; but in between are functions with discontinuous but bounded derivative $g$, for which the upper bound $||wF(w)||_1$ is useless since it diverges, so for these functions of discontinuous but bounded derivative I am trying to figure out if there are an upper bound. – Joako Dec 08 '21 at 22:52
  • 1
    But an upper bound in terms of what? I could tell you that you can find upper bounds in terms of a lot a things like $|f'|{L^\infty} ≤ |f|{B^1_{1,\infty}}$ or $|f'|{L^\infty} ≤ |(1+|\ln t|)f'|{L^{\infty}}$ but it will not help you to have a more complicated expression. Perhaps this is what is not clear in your questions. For example, why is $|w, F(w)|{L^1}$ simpler than $|f'|{L^\infty} $ for your applications? – LL 3.14 Dec 08 '21 at 23:06
  • That is a good question because is part of what I am looking to understand... I am trying to understand what will make a function to have a bounded/unbounded derivative, where the doubt arise because of this: When seeing videos of relativity is always said that "massive objects can´t achieve the speed of light because it will require infinite energy", so speed is bounded, but through these questions I have done I realize that having a bounded derivative has nothing to do with the signal energy, as viewing them as functions. (...) – Joako Dec 08 '21 at 23:15
  • Even worst, so far, it looks axiomatic, so or I allow a model to have infinite speed or not, there aren´t any bounds from the "mathematics" of function itself that will tell me if is forbidden or not to possible achieve and infinite speed (not from its energy, neither its norms, neither from their Fourier transform)... so or I allow to have models like $f(t) = \sqrt{1-t^2},,|t|<1$ with infinite speed (violating any possible universe with finite causality speed), or I arbitrarily forbid them, but I didn't found (yet) any upper bound that create a restriction to this. – Joako Dec 08 '21 at 23:20
  • PS: Could you give a reference of what the upper bound $||f||{B{1,\infty}^1}$ mean? Is applicable to time-limited signals? – Joako Dec 08 '21 at 23:22
  • The example you gives in the answer dumps all the listed inequalities, but its derivative is unbounded... Do you know any other counter-example but with bounded derivative $sup_t |f'(t)|<\infty$? (with all the other restrictions of $f(t)$ being time-limited and continuous, and absolute integrable with finite energy and definite Fourier Transform, so with finite amount of discontinuities in $f'(t)$, but with at least one discontinuity in $f'(t)$). – Joako Dec 09 '21 at 23:23