1

The question: The function $f(x) = x^\alpha $ belongs to the Lipschitz class of order $\alpha$ gives an example of $\alpha$-lipschitz function for $\alpha \in ]0,1[$. This is $x^{\alpha}$.

I guess this does not hold when $\alpha > 1$. What is an easy argument to show this? Also, what are examples of $\alpha$-Lipstchitz functions when $\alpha > 1$?

Recall $\alpha$-Lipschitz means there is a $K$ such that $\|f(x)-f(y) \| \le K \| x-y \|^{\alpha}$.

Bernard
  • 179,256
user1868607
  • 6,243

1 Answers1

2

The correct name is $\alpha$-Hölder, not Lipschitz. Lipschitz refers to the case $\alpha=1$. Anyway, for $\alpha>1$, you can show that $\alpha$-Hölder functions defined on "good sets" (see below) are constant.


Assume that $E$ is a convex set. Obviously, $\mathbb{R}$, $\mathbb{R}^n$ or open balls in these spaces all are convex sets.

$\forall y \in E$, our assumption implies $$\forall x\in E: \,\,\,\|f(x)-f(y) \| \le K \| x-y \|^{\alpha}$$

From the properties of real numbers we know that when $\alpha>1$,

$$\forall r\in\mathbb{R}:0<r<1 \implies r^{\alpha}<r$$

Now given $\epsilon>0$, take $\delta=\frac{1}{2}\min\{1,\frac{2\epsilon}{K}\}$. This ensures that $\delta < 1$ holds. Therefore,

$$\| x-y \|< \delta \implies \| x-y \|^{\alpha}<\delta^{\alpha}<\delta<\frac{\epsilon}{K}$$

$$\|f(x)-f(y) \| \le K \| x-y \|^{\alpha} < K \frac{\epsilon}K = \epsilon$$

Since the above inequality is true for all $\epsilon >0$, we get $\|f(x)-f(y) \|=0$ when $\|x-y\|< \delta$.

Hence $f(x)=f(y)$ for all $x\in E \cap B_{\delta}(y)$ for all $y\in E$, i.e. $f$ is locally constant.

If $d(x,y) \geq \delta$, since $d(x,y)$ is a finite number and $E$ is convex, you can choose $n$ points $\{z_0=x,z_1,\cdots,z_n=y\} \in E$ to connect $x$ to $y$ such that $$d(z_i,z_{i+1})<\delta \text{ for } i=0,1,\cdots,n-1$$

Then $\forall i\in\{1,\cdots,n-1\}: \,f(x)=f(z_i)=f(y)$, which means that $f(x)=f(y)$ for all $x,y \in E$. Hence, $f$ is constant on $E$.


Generalization to connected sets

Using the theorem in topology that locally constant functions on connected spaces are constant, we can assume $E$ to be connected only. Our argument will be simplified then. We won't need intermediate points anymore. Instead, we argue that we have shown $f$ to be locally constant at each point $y \in E$. Since $E$ is connected, it must be constant on all of $E$.


An example of a bad set

To see an example of a bad set $E$, take $E = (-\infty,-1] \cup [+1,+\infty) \subset \mathbb{R}$ and let $f: E \to \{\pm 1\}$ be $$f(x)=\begin{cases} +1 & \text{if}\, x\geq +1 \\ -1 & \text{if}\, x\leq -1\end{cases}$$

Then the assumptions of being $\alpha$-Hölder are satisfied for $K=1$, but $f$ is not constant obviously. So, more generally, one can say that if $E$ has several connected components, then $f$ is constant on each connected component.

stressed out
  • 8,330
  • You really need to say what $E$ is to make this valid. – Umberto P. Mar 03 '19 at 19:24
  • @UmbertoP. You're exactly right. I was editing my answer before I read your comment. So, I'm glad that my suspicion wasn't wrong. I think $E$ has to be convex, but maybe a weaker assumption that $E$ is star-shaped works too. – stressed out Mar 03 '19 at 19:33
  • @UmbertoP. I just realized that we can use the theorem that says locally constant functions on connected sets are constant instead. So, if $E$ is connected, then the proof still works. Am I correct? – stressed out Mar 03 '19 at 19:39
  • i really appreciate your detailed answer. the name i was giving comes from the lectures notes from a fractal geometry course i folllow. it is precisely stated in the notes although not explicit in the accompanying book (by the same author). Anyway I would appreciate if there are any references you can recommend on alpha-holder functions. i feel this kind of functions will come into play in a later course on differential equations i have to take – user1868607 Mar 04 '19 at 14:18
  • 1
    @Javier You're welcome. I don't know a resource specifically on $\alpha$-Hölder functions. I don't know if there's an independent theory about them or not. I personally find Wikipedia entries on them very comprehensive but you have to prove the claims on your own. One interesting case is $\alpha=1$ which gives Lipschitz functions that are absolutely continuous. Another interesting case that I remember from a course on stochastic differential equations is for $\alpha=1/2$. It turns out that Brownian motion approximately satisfies $|W(t)-W(s)| \approx |t-s|^{1/2}$ when $t$ and $s$ are very close – stressed out Mar 04 '19 at 15:56
  • @Javier Brownian motion, which is called the Wiener process in mathematics, is the basis of stochastic calculus. It turns out that unlike the cases when $\alpha\geq 1$ which make the function (almost everywhere) differentiable, Brownian motion is almost everywhere continuous while being nowhere differentiable. So, when $0<\alpha<1$ things seem to be more interesting. – stressed out Mar 04 '19 at 16:00