# For continuous functions, preimage of open set is open.

Let $f$ be a continuous function from a metric space $X$ into $Y$. If $V\subset Y$ and $V$ is open, then show that $f^{-1}(V)$ is open.

The proofs I've seen of the fact that open sets have open preimages either use the fact that continuous functions map limit points to limit points, or they use a completely topological proof.

Is there a more basic metric feeling proof? Something that just uses the basic definition of open sets, and the basic definition of continuity? Or are these sequential/topological arguments the only arguments to make?

• Actually thats the definition of a continuous function $f$. Sep 6 '14 at 8:18
• @PenasRaul That is one definition of continuity, used in general topology. In metric spaces, the ε–δ style definition is (perhaps) more commonly used. It is, of course, equivalent, but that equivalence is what the question is about, no? Sep 6 '14 at 8:21
• Bottom of dpmms.cam.ac.uk/~wtg10/easyanalysis1.html feels metrical. Nov 11 '14 at 3:02

Let $$f$$ be a continuous function from a metric space $$X$$ into $$Y$$. If $$V\subset Y$$ and $$V$$ is open, then we shall prove that $$f^{-1}(V)$$ is open.
Suppose that $$p\in X$$ and $$f(p)\in V$$. Since $$V$$ is open, there exists $$\varepsilon>0$$ such that $$y=f(x) \in V$$ if $$d_{Y}(f(x),f(p))<\varepsilon$$, and since $$f$$ is continuous at $$p$$, there exists $$\delta>0$$ such that $$d_{Y}(f(x),f(p))<\varepsilon$$ if $$d_{X}(x,p)<\delta$$. Thus $$x\in f^{-1}{(V)}$$ as soon as $$d_X(x,p)<\delta$$.
• I find the logic in the proof not solid enough: because if e.g. $d_{X}(x,p)\geq\delta$ it is still possible that $d_{Y}(f(x),f(p))<\varepsilon$. Oct 1 '18 at 8:22
• That is not an issue. We only need that the membership hold true whenever $d_X(x,p)<\delta$. We do not care what happens otherwise. Sep 29 '19 at 15:08
• My question though is, what if the domain is not all of $X$? How can we guarantee that $\delta$ can be small enough such that the entire ball around $x$ is contained in the domain? Sep 29 '19 at 15:11
• @hicrypto123. Not sure if the domain "not all of $X$" matters here. The metric is defined on the domain. Sep 29 '19 at 15:25