6
$\begingroup$

Let $(X, d)$ be a complete and separable metric space. I am interested in the case where bounded subsets of $X$ are not necessarily compact. Let $f: X \to \mathbb R$ be bounded and continuous.

Is there a sequence $(f_n)$, where each $f_n: X \to \mathbb R$ is bounded and uniformly continuous, such that $f_n \to f$ uniformly on every compact subset of $X$?

Thank you for your elaboration.

$\endgroup$
3
  • 1
    $\begingroup$ At least for $X = \mathbb R$, convolution with standard mollifiers yields a sequence with the desired property. See e.g. Theorem 11 of these notes. $\endgroup$ Commented Jul 4 at 14:40
  • 3
    $\begingroup$ @JackEdwardTisdell I think that, when $X=\mathbb R$, you don't need mollifiers. You can take $f_n$ to coincide with $f$ on $[-n,n]$, to equal $f(n)$ on $[n,\infty)$, and to equal $f(-n)$ on $(-\infty,-n]$. $\endgroup$ Commented Jul 4 at 14:47
  • $\begingroup$ I am interested in the case where bounded subsets of $X$ are not necessarily compact. $\endgroup$ Commented Jul 4 at 14:48

2 Answers 2

12
$\begingroup$

$\newcommand\de\delta\newcommand\ep\varepsilon$This is true for any metric space $(X,d)$. In particular, the completeness and separability conditions are not needed. The $f_n$'s (which are actually Lipschitz continuous) are given by a simple explicit construction. The proof is completely elementary.

Indeed, for natural $n$ and $x\in X$, let $$f_n(x):=\inf_{y\in X}(f(y)+nd(y,x)).$$

We have $m\le f\le M$ for some real $m,M$. So, for all $n$ and all $x\in X$, $$m\le\inf_{y\in X}f(y)\le f_n(x)\le f(x)+nd(x,x)=f(x)\le M,$$ so that $m\le f_n\le M$.

For all $x,x_1,y$ in $X$, $$f_n(x_1)\le f(y)+nd(y,x_1)\le f(y)+nd(y,x)+nd(x,x_1),$$ so that $$f_n(x_1)\le \inf_{y\in Y}(f(y)+nd(y,x)+nd(x,x_1))=f_n(x)+nd(x,x_1),$$ so that $f_n$ is $n$-Lipschitz and hence uniformly continuous.

Take any compact $K\subseteq X$ and any real $\ep>0$. Since $f$ is continuous, for each $z\in X$ there is some real $\de_z>0$ such that $|f(w)-f(z)|<\ep$ for all $w\in V_z:=B_z(2\de_z)$, where $B_z(r)$ is the open ball in $X$ of radius $r$ centered at $z$. Since $K$ is compact, there is some finite subset $F$ of $K$ such that $\bigcup_{z\in F}U_z\supseteq K$, where $U_z:=B_z(\de_z)$. Let $\de:=\min_{z\in F}\de_z$.

Now take any $x\in K$ and any $y\in X$. Then $x\in U_{z_x}\subseteq V_{z_x}$ for some $z_x\in F$ and hence $|f(x)-f(z_x)|<\ep$.

If $d(y,x)<\de$, then $d(y,z_x)\le d(y,x)+d(x,z_x)<\de+\de_{z_x}\le2\de_{z_x}$, so that $y\in V_z$ and hence $|f(y)-f(z_x)|<\ep$, so that $$f(y)\ge f(x)-|f(x)-f(z_x)|-|f(y)-f(z_x)|>f(x)-2\ep.$$

So, if $d(y,x)<\de$, then $$f(y)+nd(y,x)\ge f(y)>f(x)-2\ep.$$ If now $d(y,x)\ge\de$, then $$f(y)+nd(y,x)\ge m+nd(y,x)\ge m+n\de\ge M\ge f(x)$$ if $n\ge N:=(M-m)/\de$, so that for such $n$ and all $x\in K$ we have $$f_n(x)=\inf_{y\in X}(f(y)+nd(y,x))>f(x)-2\ep.$$ On the other hand, $f_n\le f$. So, $f_n\to f$ uniformly on $K$. $\quad\Box$

$\endgroup$
3
  • $\begingroup$ Why are you restricting your attention only to $y\in K$? The definition of $f_n$ takes the infimum over all $y\in X$ so you can't exactly use uniform continuity. I think I see how to fix this by considering small balls and taking the finite subcover or something like that, but it will be a bit longer than what you wrote. $\endgroup$ Commented Jul 4 at 15:38
  • $\begingroup$ @AlekseiKulikov : Thank you for your comment. This is now fixed. $\endgroup$ Commented Jul 4 at 17:14
  • $\begingroup$ The downvoter: Would you please explain your reason(s)? $\endgroup$ Commented Jul 6 at 1:23
5
$\begingroup$

Yes, this follows from the Stone-Weierstrass theorem. This theorem holds on any topological space $X$ with respect to the compact-open topology on $C^0(X)$. When the space $X$ is a metric space, convergence with respect to this topology is uniform convergence on compact subsets. Second countability of $X$ implies that $C^0(X)$ is metrizable hence sequences can be used to describe closures. For the subalgebra separating points, you can take the bounded Lipschitz functions.

Dugundji, Topology is a standard reference for these results.

$\endgroup$

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.