I am studying myself some facts about $\alpha$-Hölder-continuous functions but I don't get any further by proving the following:

$(1)$ $\forall\alpha\in ]0,1]$ is $C^{0,\alpha}$ dense in $C^0(D)$ concerning the uniform norm and $D\subset\mathbb R^n$.

$(2)$ $\forall\alpha\in ]0,1]$ and compact set $K\subset\mathbb R^n$ is $(C^{0,\alpha}(K),||\cdot||_{C^{0,\alpha}(K)})$ a complete space (with $||u||_{C^{0,\alpha}(K)}:=||u||_{\sup}+\sup\limits_{{x,y\in K\space\&\space x\ne y}}\frac{|u(x)-u(y)|}{|x-y|^\alpha}$ and $C^{0,\alpha}(K)$ the space of all $\alpha$-Hölder-continuous functions)

$(3)$ All bounded closed subsets of $(C^{0,\alpha}(K),||\cdot||_{C^{0,\alpha}(K)})$ are compact.

So how do you prove one $(1),(2),(3)$ ?


(1) Assume $D$ compact then by Weiertrass approximation you get that polynomials are dense in $C(D)$, since $D\subset B_R(0)$ for some $R>0$ and polynomials are $\alpha$-Hölder in $B_R(0)$ we get the result.

If $D$ is not compact the result is not true: Take $D=(0,1)$ and $f(x)=\sin(1/x)$. Take $0<r$ and $g\in C^{0,\alpha}$ such that $\| g-f\|_\infty <r$, then for every $n\in \mathbb{N}$ there exists $x_n,y_n\in (0,1)$ with $|x_n-y_n|<r$ and $f(x_n)=1$, $f(y_n)=-1$. Then $|g(x_n)-g(y_n)| \geq 2(1-r)>r^{\alpha}>|x_n-y_n|^{\alpha}$ if $r$ is small enough, contradicting $g\in C^{0,\alpha}$.

If $D=\mathbb{R}$ a similar argument works if we use something like $f(x)= \sin(2n\pi(x-n))$ if $x\in [n,n+1/n]$, $n\in \mathbb{N}$ and zero otherwise.

(2) We know that $C(K)$ is complete so we only need to check that the limit function of a Cauchy sequence in $C^{0,\alpha}$ is in this space and that the seminorms converge. For this notice $$ |f(x)-f(y)|\leq |f(x)-f_n(x)|+|f_n(x)-f_n(y)|+|f_n(y)-f(y)| $$ So, picking $n$ so that $\| f-f_n\|_{\infty} <\epsilon$ we get, and remembering that $\frac{|f_n(z)-f_n(w)|}{|z-w|^\alpha} \leq [f_n]_\alpha \leq C$, $$ \frac{|f(x)-f(y)|}{|x-y|^\alpha} \leq C+2\epsilon/|x-y|^\alpha $$ which gives $[f]_\alpha <C<\infty$.

To see that the seminorms converge just notice $$ \frac{|f(x)-f_n(x)-f(y)+f_n(y)|}{|x-y|^\alpha} \leq\limsup_m [f_m-f_n]_\alpha $$ which gives $$ \limsup_n [f-f_n]_\alpha \leq \limsup_n \limsup_m [f_m-f_n]_\alpha \to 0. $$ (3) I don't know about this statement as is, but Arzela-Ascoli gives that bounded closed sets in $C^{0,\alpha}(K)$ are precompact in $C(K)$

  • 1
    $\begingroup$ what's $[f_n]_\alpha$ in (2) ? $\endgroup$ – sheldoor Jun 16 '12 at 20:10
  • $\begingroup$ @sheldoor: In general $$[g]_\alpha:= \sup_{x,y} \frac{|g(x)-g(y)|}{|x-y|^\alpha}$$ $\endgroup$ – Jose27 Jun 17 '12 at 5:13
  • $\begingroup$ Nice, but one should add the subtle observation that in the inequality that gives the $[f]_\alpha <C$, epsilon is independent of $x, y$ so, one FIRST lets epsilon go to zero, and then takes sup over $x\neq y$. $\endgroup$ – Behnam Esmayli Mar 30 '18 at 15:26

(1) You should be more specific regarding $D$. Compact sets are fine. In general you want to make sure that the elements of your space of continuous functions are bounded and uniformly continuous; otherwise it's hard to approximate them uniformly... Here is one possible approach. Given a bounded uniformly continuous $f\colon D\to\mathbb R$, consider $f_n(x)=\inf_{y\in D}(f(y)+n|x-y|)$. Each $f_n$ is Lipschitz and $f_n\le f$ everywhere. Moreover, $f_n\to f$ uniformly, as you should be able to prove using the uniform continuity of $f$ (the idea is that only $y$ closed to $x$ affect the infimum).

Re (2): it's a good idea to begin by studying the proof that $C(K)$ is complete.

(3) is false: take $K=[0,1]$ and $f_k(x)=\sqrt{\max(x-1/k,0)}$. The sequence $f_k$ is bounded in $C^{0,1/2}$ but has no convergent subsequence. Indeed, the limit would have to be $f(x)=\sqrt{x}$, but $\|f-f_k\|_{C^{0,1/2}}=1$ for all $k$.


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.