Recall that, back in section 8.3, I proved that f : R R is continuous i f 1 () for every open set . Its time to generalize.

. First we need a lemma, though. This says that there is a statement about relatively open sets which is analogous to the denition of a (plain vanilla) open set. This is part of Remark 8.27, section 8.3, which we didnt need until now. Lemma: U E is relatively open in E i a U there exists > 0 so that B (a) E U . Proof: "=" First suppose that U is relatively open in E . Then there exists an open set A so that U = E A. Since A is open and U is a subset of A, for any a U there is an > 0 so that B (a) A. But then B (a) E A E = U . "=" Conversely, suppose that U has the property that a U there exists > 0 so that B (a) E U . For each a U , nd the appropriate , and call it (a). Set A = aA B(a) (a). This is certainly an open set, since its the union of open sets. I claim that U = A E . Certainly if a U , then a B(a) (a), hence in A. Also, since U E , a E . Thus a A E , and U A E . For the other containment, take x A E . Since x A, x must be in B(a) (a) for some a A. Since x E as well, x B(a) (a) E . But by choice of , we must have x U . Theorem: f : E Rm is continuous on E i f 1 () is relatively open in E for every open set in Rm . Proof: First suppose that f is continuous on E , and that is open. If f 1 () is empty, its open, so assume that f 1 () is nonempty, and take a f 1 (). Then f (a) . Since is open, there is > 0 so that B (f (a)) . Since f is continuous at a, there is > 0 so that x a < , x E implies f (x) f (a) < . In other words, everything in B (a) E gets taken by f to within of f (a), hence into . This says that B (a) E f 1 (). This is true for arbitrary a f 1 (), so by the lemma, f 1 () is relatively open in E . Now suppose that inverse images of open sets are relatively open. Take an arbitrary a E and an > 0. We have that B (f (a)) is an open set, hence f 1 (B (f (a))) is relatively open in E . Since a f 1 (B (f (a))), by the lemma, there is a > 0 so that B (a) E f 1 (B (f (a))). Now untangle this statement. What its saying is that if you take f of any point of E thats within of a, you must end up in the ball around f (a). This is just the statement that x E , x a < implies f (x) f (x) < , i.e., that f is continuous at a, and hence on E since a was arbitrary.

c 2012, T. Vogel

73

Proposition: f 1 () is relatively open for each open set Rm i f 1 (C ) is relatively closed for each closed set C Rm . This gives a third way of characterizing continuity, but you dont see it nearly as much as the statement about relatively open sets. Proof: "=" Take a closed set C . Then C c is open, thus f 1 (C c ) is relatively open in E . But I claim that f 1 (C c ) = E f 1 (C ). Its just chasing the denition: a f 1 (C c ) i f (a) C c , i.e., f (a) / C, 1 1 1 which is the same as a / f (C ), i.e., a E f (C ). So, E f (C ) is relatively open in E , which weve seen is the same as f 1 (C ) is relatively closed in E . "=" Take an open set and look at complements, just as in the previous part. Note: Images of open sets under continuous maps need not be open. Heres an example. Example: f : R R given by f (x) = sin x. Then f ((0, 2 )) = [1, 1]. We can say something about images of compact and connected sets, however. Theorem: If H is compact, f :H Rm continuous, then f (H ) is compact. Proof: Suppose that {V : A} is an open cover of f (H ). We want to show that there is a nite subcover. I claim that the collection {f 1 (V ) : A} covers H . The reason: if x H , then f (x) f (H ), so f (x) V0 for some 0 , thus x f 1 (V0 ). Now, the collection {f 1 (V ) : A} is not an open cover of H , since all we know about each f 1 (V ) is that they are relatively open in H , but this will be enough. Since each f 1 (V ) is relatively open in H , there there is an open set U so that Ua H = f 1 (V ). The collection {U : A} is an open cover of H . Therefore, there is a nite subcover U1 , , Uk . Since these cover H , clearly their intersections with H cover H , so f 1 (V1 ), , f 1 (Vk ) cover H . But then V1 , , Vk cover f (H ): y f (H ) i x H so that f (x) = y (maybe more than 1 x), so there is Vj with x f 1 Vj , so y Vj . Note:

Continuous images of closed sets need not be closed: let f (x) = 1 . If C = [1, ) (closed, not compact), then f (C ) = (0, 1] is x not closed. Continuous images of bounded sets need not be bounded: take the same f as before, and look at f ((0, 1]). This is [1, ). c 2012, T. Vogel 74