It is clear that if $f : X \to Y$ is continuous then the graph of $f$ over every path connected subspace is path connected.

Is the converse true? That is, if the graph of $f : X \to Y$ over every path connected subspace is path connected, does it follow that $f$ is continuous?

The topologist's sine curve is a well known counterexample if we say "connected" instead of "path connected." Is there a standard counterexample in the path connected case? Is there a counterexample that works when $X = Y = \mathbb{R}$?

Part of why I ask is that continuity is often informally defined in introductory calculus classes as a function whose graph you could "walk along" as though it were a path. This, however, is quite different from the formal definition and, I suspect (but have not been able to prove), is not even true.

  • $\begingroup$ What are you assuming about $X$? I guess there will be some easy counterexamples where $X$ is a non-discrete space with no nontrivial path-connected subspaces, e.g., the space of irrational numbers. $\endgroup$ – bof Dec 19 '20 at 5:53
  • $\begingroup$ You’re going to want to make some assumptions about $X$ and $Y$; at the very least locally path-connected, maybe even manifolds. $\endgroup$ – Qiaochu Yuan Dec 19 '20 at 6:40
  • $\begingroup$ @bof It should have been obvious to me that we needed more assumptions on $X$ and $Y$. I'm really most interested in the case where $X = Y \mathbb{R}$. $\endgroup$ – Charles Hudgins Dec 19 '20 at 14:40

For arbitrary spaces it certainly isn't true. For instance, if $X$ is totally path-disconnected (e.g., $X=\mathbb{Q}$), then your condition is trivial, but not every map out of $X$ is continuous unless $X$ is discrete.

To characterize the spaces for which it is true, consider the following properties of a topological space $X$:

  • $X$ is well-graphed if it has your property for arbitrary $Y$: that is, a map $X\to Y$ is continuous iff its graph over every path-connected subspace of $X$ is path-connected.
  • $X$ is path-generated if for any $Y$, a map $X\to Y$ is continuous iff its composition with every continous path $[0,1]\to X$ is continuous. (For more about this property see this answer of mine; in particular, a path-generated space must be locally path-connected and sequential.)
  • $X$ is arc-generated if for any $Y$, a map $X\to Y$ is continuous iff its composition with every embedding $[0,1]\to X$ is continuous.

We then have the following theorem:

Theorem: Every arc-generated space is well-graphed, and the converse holds for Hausdorff spaces. Every well-graphed space is path-generated.

Proof: Suppose $X$ is well-graphed and $f:X\to Y$ has the property that the composition of $f$ with every path in $X$ is continuous. Then the graph of $f$ over every path-connected subspace is path-connected, since every path in $X$ lifts to a path in the graph. Thus $f$ is continuous, so $X$ is path-generated. Moreover, if $X$ is Hausdorff, then it would suffice to assume the composition of $f$ with every embedding $[0,1]\to X$ is continuous, since in a path-connected Hausdorff space any two points are connected by a path which is an embedding, so $X$ would be arc-generated.

Conversely, suppose $X$ is arc-generated. Let $f:X\to Y$ be a map whose graph over every path-connected subset is path-connected. To show $f$ is continuous, it suffices to show $fg$ is continuous for every embedding $g:[0,1]\to X$. Since $g$ is an embedding, the graph of $fg:[0,1]\to Y$ is homeomorphic to the graph of $f$ restricted to the image of $g$. By hypothesis, the latter graph is path-connected, so the graph of $fg$ is path-connected. Now take a path $h$ from $(0,f(g(0)))$ to $(1,f(g(1)))$ in this graph. The first coordinate of $h$ is a continuous surjection $p:[0,1]\to[0,1]$, which is automatically a quotient map since the domain is compact and the codomain is Hausdorff. The second coordinate of $h$ is then the map $fgp:[0,1]\to Y$ and is continuous since $h$ is. But since $p$ is a quotient map, this implies $fg$ is continuous, as desired. $\blacksquare$

In particular, $\mathbb{R}$ is arc-generated and thus well-graphed, so your statement is true for the motivating example of maps $\mathbb{R}\to\mathbb{R}$.

As a counterexample to show that a Hausdorff path-generated space need not be well-graphed (or equivalently, arc-generated), let $X$ be the 1-point compactification of the set $$A=[0,\infty)\times\{0\}\cup\mathbb{N}\times[0,1]\subset\mathbb{R}^2.$$ Then $X$ is path-generated (indeed, it is a quotient of $[0,1]$, by a path that goes from $(0,0)$ off to $\infty$ taking a detour on each $\{n\}\to[0,1]$ along the way). However, I claim $X$ is not well-graphed. Consider the map $f:X\to[0,1]$ that is the first projection on $A$ and maps $\infty$ to $0$. Then $f$ is not continuous at $\infty$, since the points $(n,1)\in A$ approach $\infty$ as $n$ gets large but $f(n,1)=1$ and $f(\infty)=0$. However, the graph of $f$ over any path-connected set is path-connected, since any path-connected subset that contains both $\infty$ and a point $(n,t)$ for $t>0$ must also contain the path from $(n,t)$ down to $(n,0)$ and then the horizontal path from $(n,0)$ out to $\infty$, and $f$ is continuous on both of these paths.

I don't know whether there is a (necessarily non-Hausdorff) well-graphed space that is not arc-generated.

  • $\begingroup$ My question could have been phrased better, but this answers everything. Thank you. I'm a bit surprised that turning "connected" into "path-connected" makes the statement true for $\mathbb{R}$. $\endgroup$ – Charles Hudgins Dec 19 '20 at 14:45

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.