2

Question

In a proof about continuous $\mathbb C$-Automorphisms, it was used that two continuous functions $\mathbb C \rightarrow \mathbb C$ agreeing on a dense subset must be equal. I was curious to figure out whether that holds for topological spaces in general. I could prove the following

Lemma: Let $B$ be Hausdorff. If two continuous functions $f,g: A\rightarrow B$ agre on a dense subset $D\subset A$, they agree on the whole space $A$.

Proof. Assume there was a point $a\in A$ where they differed. By the Hausdorff axiom, we can find two disjoint open sets $F\ni f(a),G\ni g(a)$. Since the intersection of their preimages must be open and $D$ is dense, there must be a point $x \in f^{-1}(F)\cap g^{-1}(G)$ on which $f$ and $g$ agree, so $f(x)\in F$ and $f(x)=g(x)\in G$, contradicting the assumed disjointness.

Since the Hausdorff axiom is quite a strong restriction, I tried to look for counterexamples in general topological spaces and found this:

Example Consider the topological space $\mathcal T:=\left([0,2],\,\{\emptyset,[0,1),[0,2]\}\right)$ and $$f:=id\vert_{\mathcal T},\quad g:x\mapsto\begin{cases}x,& x\in [0,1)\\ 2,&\mathrm{else.}\end{cases}$$ We note that both functions are continuous, agree on the dense subset $\{1,2\}$, but are not equal.

Since this on the other hand is a very weak example ($\mathcal T$ is not even $T_0$), I was wondering:

What is the minimum separation axiom that must be fulfilled by $B$ in order for this to hold?

Edit: Answer

These are the conclusions I reached by clarifying parts of @user87690's answer. Upon writing this, I noted that this very question has been asked here before, effectively rendering this a duplicate.

Recall the following

Definition. A point $x$ is called isolated if $\{x\}$ is open.

We immediately observe the following

Lemma. Let $B:=(X, \mathcal T)$ be a topological space, $x\in X$. $U\setminus\{x\}$ is dense in $B$ if and only if $x$ is isolated.

Theorem. Let $B:=(X, \mathcal T)$ be a non-Hausdorff space. There is a space $A$ and continuous functions $f, g$ agreeing on a dense subset of $A$ that are not equal.

Proof. Let $x, y\in X$ be non-separable points. One of them is non-isolated, WLOG $x$. We will define (as set-functions) $$f := id\vert_X, \quad g: p\mapsto \cases{ y,\quad p=x\\ p,\quad else, }$$ in other words, $p$ differs from the identity by mapping $x$ to $y$.

Note that although the initial topology would make $f,g$ continuous, it is uninteresting in the sense that $U\setminus\{x\}$, the set on which both agree, is not dense (because $x$ is isolated). Instead we will use the coarsest topology making $f$ an $g$ continuous, which is the topology generated by $\mathcal T \cup g^{-1} [\mathcal T]$.

As we will prove in the next lemmata, $x$ is not isolated in this topology, and so $X\setminus \{x\}$ is dense . $\tag*{$\blacksquare$}$

Lemma. Let $x\in X$. If two $\cap$-stable generators $\mathcal G, \mathcal G^\prime \subset \mathcal P (X)$ don't contain $\{x\}$, $x$ will not be isolated in $\langle \mathcal G\cup \mathcal G^\prime \rangle$ if and only if for every choice $A\in \mathcal G, B\in \mathcal G^\prime$, $A\cap B \neq \{x\}$.

Proof. By $U_c$ we mean the hull operator taking countable unions, and by $I_f$ we mean the hull operator taking finite intersections. If a set $\mathcal G$ is $\cap$-stable, by induction over set cardinality we have $I_f(\mathcal G) = \mathcal G$. The coarsest topology containing a generating set $\mathcal G$ can be constructed explicitly by applying both hull operators and adding $\{X, \emptyset\}$.

We now note that $U \in I_f(\mathcal G \cup \mathcal G^\prime)$ can be represented as $$U = \bigcap_{i=1\\G_i\in \mathcal G}^n G_i \cap \bigcap_{i=1\\G_i\in \mathcal G^\prime \setminus \mathcal G} G_i,$$ and due to $I_f$-Stability of both sets, we can choose \begin{align} A&:=\bigcap_{i=1\\G_i\in \mathcal G}^n G_i \in \mathcal G \\ B&:=\bigcap_{i=1\\G_i\in \mathcal G^\prime \setminus \mathcal G}^n G_i \in \mathcal G^\prime, \end{align} So $U = A \cap B$ where $A, B$ are either empty or in $\mathcal G, \mathcal G^\prime$, respectively. This leaves us with $$I_f(\mathcal G \cup \mathcal G^\prime) = \mathcal G \cup \mathcal G^\prime \cup \{A\cap B\,\vert A\in \mathcal G, B\in\mathcal G^\prime \}.$$ We can now observe \begin{align} \{x\} &\in \langle \mathcal G\cup \mathcal G^\prime \rangle = \{X, \emptyset\} \cup U_c(I_f(\mathcal G \cup \mathcal G^\prime))\\ \Leftrightarrow \{x\} &\in U_c(I_f(\mathcal G \cup \mathcal G^\prime))\\ \Leftrightarrow \{x\} &\in I_f(\mathcal G \cup \mathcal G^\prime)\\ &= \mathcal G \cup \mathcal G^\prime \cup \{A\cap B\,\vert A\in \mathcal G, B\in\mathcal G^\prime \}\\ \Leftrightarrow \exists A\in \mathcal G, B\in\mathcal G^\prime:\ \{x\}&= A\cap B\\ \end{align} $\tag*{$\blacksquare$}$

We observe that \begin{align} g^{-1}[\mathcal T] &= \{U\setminus\{x\} \,\vert x\in U\not\ni y\}\\ &\cup \{U\cup\{x\} \,\vert x\not\in U\ni y\}\\ &\cup \{U \,\vert x\not\in U\not\ni y\}\\ &\cup \{U \,\vert x\in U\ni y\}\\ \end{align} Where the last two sets are already contained in $\mathcal T$. Therefore, our topology can be written as $\langle \mathcal T \cup \mathcal S \rangle$ where $$\mathcal S := \mathcal S_1 \cup \mathcal S_2 := \{U\setminus\{x\} \,\vert x\in U\not\ni y\} \cup \{U\cup\{x\} \,\vert x\not\in U\in y\}$$

We can now observe (proof omitted, do a case distinction as done with $U$ above) that

  1. $\mathcal S$ is $\cap$-stable
  2. no intersection of $A\in \mathcal G$ and $B\in \mathcal G^\prime$ yields $\{x\}$

Both of which can be used to deduce that $x$ cannot be isolated in our topology, completing the above proof of our theorem. $\tag*{$\blacksquare$}$

  • 1
    A little more generally if $B$ is $T_2$ and $f:A\to B$ and $g:A\to B$ are continuous then ${x\in A: f(x)=g(x)}$ is closed in $A,$ and I think it is easier to prove this directly (rather than by contradiction) by using the $T_2$ property of $B$ to show that ${x\in A: f(x)\ne g(x)}$ is open in $A.$..... It is a good Q. – DanielWainfleet Apr 04 '18 at 16:36

1 Answers1

2

Let us call a topological space $B$ good if for every topological space $A$ and two continuous maps $f, g\colon A \to B$ that agree on a dense set are equal, and let us call $B$ bad if it is not good.

You observed that every $T_2$ space is good and gave an example of a non-$T_0$ space that is bad. In fact, every non-$T_0$ space is bad, i.e. every good space is $T_0$. If $x ≠ y$ are to indistinguishable points, then consider $f$ and identity and $g$ being the modification of $f$ that maps $x$ to $y$. There are continuous maps and they agree on $B \setminus \{x\}$. Since $x$ is not isolated, we are done.

It seems to be a good idea to consider these one-point modifications of the identity. If $B$ is non-$T_2$, there are two points, $x ≠ y$ that cannot be separated by disjoint neighborhoods. One of them is non-isolated, wlog $x$. Let $A$ be $B$, $f$ the identity, and $g$ the same one-point modification of $f$. Now $g$ does not have to be continuous, but we may endow $A$ with the initial topology to fix this. If $x$ remains non-isolated in this new topology, then we have that $B$ is bad.

Let $U ⊆ B$ be open. If $x ∈ U ∌ y$, then $g^{-1}[U] = U \setminus \{x\}$. If $x ∉ U ∋ y$, then $g^{-1}[U] = U ∪ \{x\}$. Otherwise, $g^{-1}[U] = U$. Therefore, if $\{x\}$ becomes isolated in the finer topology, this is witnessed by $\{x\} = (U ∪ \{x\}) ∩ V$ where $x ∉ U ∋ y$ and $U, V$ are open in the original topology. We have $U ∩ V = ∅$, $x ∈ V$, and $y ∈ U$. This is a contradiction. Therefore, every non-$T_2$ space is bad, and so exactly $T_2$ spaces are good.

user87690
  • 9,133
  • Wait, I was a bit hasty with accepting this answer. If we view $B:=(X, \mathcal T)$ and $A:= (X, \mathcal P(X))$, i.e. the initial topology, then every point in $A$ is isolated. What's the exact topology you equip $X$ with to construct $A$? AFAICS we would want a topology such that some subset of $U\setminus {x}$ is dense but $f,g$ are continuous. – Lukas Juhrich Apr 08 '18 at 21:40
  • I think I got the essence, see my edit. Thanks a lot, the equivalence is a really neat fact. – Lukas Juhrich Apr 08 '18 at 22:41
  • I made a mistake, for id to be continuous, $A$ must be finer. As of now, I'm trying to prove that the Topology generated by ${\mathcal T, g^{-1} [\mathcal T]}$ still cannot contain $x$, which does not strike me as simple. – Lukas Juhrich Apr 09 '18 at 12:53
  • 1
    Look at the last paragraph. The subbasic new open sets are of form $U ∪ {x}$ and $U \setminus {x}$ for open set $U$. The point $x$ is isolated in the new topology iff ${x}$ can be written as a finite intersection of generating open sets, and these are the old open sets and the sets above. If we group some sets together, we obtain ${x} = (U ∪ {x}) ∩ V$. – user87690 Apr 09 '18 at 14:25
  • Thanks for your hint! I realized I lacked quite some understanding of the topology generated by some sets. I fixed my proof by proving the lemma you mentioned. I think it should be correct now. It's still more effort than I previously anticipated to write down a formal proof even when the Idea has been pretty clear for a while :-) – Lukas Juhrich Apr 09 '18 at 19:10
  • 1
    Several small comments: • You write that the initial topology would be uninteresting and you use $\mathcal{T} ∪ g^{-1}[\mathcal{T}]$, but that is the initial topology generated by $f$, $g$. • Topology is stable under all unions, not only countable ones, so you should replace the union operator. • You don't need to add $∅$ and $X$ explicitly – $X$ is the intersection of the empty family and $∅$ is the union of the empty family. This convention also simplifies the proof a bit. Otherwise, your proof is correct. – user87690 Apr 10 '18 at 08:03