3

Let $A = \{k2^{-n} : 0 \leq k \leq 2^n, n \geq 1\}$ and $f$ be a function defined on $A$ satisfying $$\sup_{n,k} c^n|f((k+1)2^{-n}) - f(k2^{-n})| < \infty$$ for some $c > 1.$ Then there exists unique a continuous function $g : [0,1] \rightarrow \mathbb{R}$ such that $g(t) = f(t) $ for all $t \in A$. Moreover, $g$ is also Holder continuous with exponent $\log_2c.$

I think that this might be a well-known theorem, but I cannot find its name or reference about it. I will be appreciate if anyone can suggest me how to prove it, or provide me the reference about its proof.

user117375
  • 1,335
  • See here for one reference: https://fabricebaudoin.wordpress.com/2012/03/26/lecture-6-the-kolmogorov-continuity-theorem/ – shalin Mar 26 '16 at 03:02

1 Answers1

3

Define $M$ to be the supremum of the quantity above.

For $n \in \Bbb N$, write $A_n:=\{ k2^{-n} : 0 \leq k \leq 2^n\}$, so $A = \bigcup_n A_n$.

From now on fix some $n \in \Bbb N$ and some $x,y \in A_n$ with $x<y$.

For $k \leq n$ define $x_k:= \min\{ r \in A_k: r\geq x\}$ and $y_k:= \max\{ r \in A_k: r \leq y\}$. Let $N:= \min\{ k: x_k < y_k\}$. Note that $|x_{k+1}-x_k| \leq 2^{-k}$, similarly $|y_{k+1}-y_k| \leq 2^{-k}$, and also $|x_N-y_N| =2^{-N}$. Also note that $x_n=x$ and $y_n=y$.

Note that $$|f(y)-f(x)| \leq \sum_{k=N}^{n-1} |f(y_{k+1})-f(y_k)|+|f(y_N)-f(x_N)| + \sum_{k=N}^{n-1}|f(x_{k+1})-f(x_k)|$$$$\leq \sum_{k=N}^{n-1} Mc^{-k} +Mc^{-N} + \sum_{k=N}^{n-1} Mc^{-k} \leq 3 M \sum_{k=N}^{\infty} c^{-k} = \frac{3M}{1-c^{-1}} \cdot c^{-N}$$

But by the definition of $N$, it follows that $|y-x| \geq |y_N-x_N| =2^{-N}$, and therefore $c^{-N} = 2^{-N \log_2 c} \leq |y-x|^{\log_2 c}$. Therefore, we showed that for all $x,y \in A$ $$|f(y)-f(x)| \leq \frac{3M}{1-c^{-1}} |y-x|^{\log_2 c}$$ so it follows that we can extend $f$ continuously to a Holder-$\log_2 c$ function. Uniqueness is an easy topology exercise.

shalin
  • 14,313
  • I suppose that this problem is related to Komogorov Continuity theorem (as you provided in the link). I try to connect what you provided with the theorem, but I think the theorem is for probability version. So I need to view this $f$ as a random function ? To be honest, I am confusing. The probability version also has expected value in the condition to use Komogorov's cont. thm, but above you just show the inequality of $f$ in absolute value ? (Does that absolute value means to norm of the $L^p$ space ?). In all, I try to understand, and still very confusing. Can you point out more ? – Both Htob Mar 26 '16 at 19:45
  • @BothHtob: The Kolmogorov Continuity Theorem doesn't follow immediately from the theorem I proved here. You need to first reduce the expected value inequality to something which looks like the condition in this theorem. In order to do that you use the Borel Cantelli Lemma. See that link above. – shalin Mar 30 '16 at 00:22