3
$\begingroup$

Let $X\subset\mathbb R$ be a subset and $f:X\to\mathbb R$ be a continuous function. Can we always find an open subset $U\subset\mathbb R$ and a continuous function $g:U\to\mathbb R$ such that $f|_{X\cap U} = g|_{X\cap U}$? Does restricting the carnality of $X$ to $\aleph_0:=|\mathbb N|$ or $\mathfrak c:=2^{\aleph_0}=|\mathbb R|$ changes the answer?

If $X$ is not dense in $\mathbb R$ or if $X$ contains an open subset of $\mathbb R$, then the answer is, of course, YES. This is not so obvious otherwise; for instance, we may take $X$ as the set of the irrational numbers.

Notice this is not the same as asking for a function that is continuous on $\mathbb R\setminus\mathbb Q$ and discontinuous on $\mathbb Q$, such as Thomae's function, since the restriction of this function to $\mathbb R\setminus\mathbb Q$ is the null function, which can be continuously extended to $\mathbb R$.

$\endgroup$
5
  • $\begingroup$ when $X$ is countable you can construct $g$ by linear interpolation. For an uncountable $X$ it becomes more tricky I think $\endgroup$
    – Levie B
    Commented Feb 10 at 1:00
  • $\begingroup$ @LevieB How would that interpolation work? Given any two $x,y\in X$ with $x<y$, there will typically be infinitely many $z,w\in X$ such that $x<z<w<y$. $\endgroup$ Commented Feb 10 at 1:15
  • $\begingroup$ I mean when $X =\{x_n, n \in\mathbb{N}\}$ define $g$ as the linear interpolation of $f$ to $Y$. Works only for $X$ countable, yes. $\endgroup$
    – Levie B
    Commented Feb 10 at 2:10
  • $\begingroup$ @LevieB that doesn't quite work, there is no reason for the $x_n$'s to be arrangeable in a consecutive sequence, so what would linear interpolation even mean? $\endgroup$
    – M W
    Commented Feb 10 at 6:47
  • $\begingroup$ Oh yeah if it is like $\mathbb{Q}$ then not, I was thinking of $\mathbb{N}$ $\endgroup$
    – Levie B
    Commented Feb 10 at 10:10

1 Answer 1

2
$\begingroup$

We can in fact never guarantee such a continuous extension on some $U\cap X$, unless $X$ has interior or is not dense, in which case we trivially can as you have observed.

Proof.

Suppose $X$ is dense and has no interior, so that $\mathbb R\backslash X$ is dense.

For ease of exposition, we will suppose that $X\subseteq\mathbb R\backslash \mathbb D$, where $\mathbb D=\left\{\frac{k}{2^n}\mid k\in\mathbb Z,n\in \mathbb N\right\}$ are the dyadics, and explain how to modify the construction to the general case at the end.

Let $I_n^k=\left(\frac{k}{2^n},\frac{k+1}{2^n}\right)$, and for each $n$ define $f_n\colon X\to \mathbb R$ so that

$$f_n(x)= \begin{cases} 3^{-n} & x\in I_n^k\cap X, k \text { odd}\\ 0 & x\in I_n^k\cap X, k \text { even}. \end{cases} $$ Let $f(x)=\sum_{n=1}^\infty f_n(x)$, and note that $f$ is continuous, since on each $I_n^k$, $f$ varies by at most $\sum_{m=n+1}^{\infty} 3^{-m} = \frac{1}{2}\left(3^{-n}\right)$, and each $x\in X$ is contained in the interior of some $I_n^k$ for each $n$.

On the other hand, $f$ cannot be uniformly continuous in any open subset of $X$, since every such set includes $(I_n^{2k}\cup I_n^{2k+1})\cap X$ for some $n$ and $k$, and then $$\left|f(x)-f(y)\right|\geq \left|3^{-n}-\frac{1}{2}\left(3^{-n}\right)\right| = \frac{1}{2}\left(3^{-n}\right)$$ for each $x\in I_n^{2k}\cap X$ and $y\in I_n^{2k+1}\cap X$, despite that such $x$ and $y$ may be arbitrarily close to each other, by density of $X$.

Since $f$ is not uniformly continuous on any open subset of $X$, $f$ cannot be extended to a continuous function on any open set in $\mathbb R$, since every continuous function on an open set in $\mathbb R$ is locally uniformly continuous.

Finally, note that we can mimic this construction for any dense subset $X$ with no interior, by simply choosing intervals $I_n^k$ to be nested in the same way as above, with diameters converging uniformly to $0$ with each generation, and with endpoints not in $X$.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .