You are on page 1of 5

METRIC SPACES Math 441, Summer 2009 We begin this class by a motivational introduction to metric spaces.

The concepts of metric and metric space are generalizations of the idea of distance in Euclidean space. Well give some examples and define continuity on metric spaces, then show how continuity can be stated without reference to metrics. This material will motivate the definition of topology in Chapter 2 of your textbook.

1. Continuous functions between Euclidean spaces.


Lets recall what it means for a function to be continuous: Definition 1: We say that is continuous at a point iff lim = (), i.e.: > 0, > 0 , | | < => | () () | < .

In words: for every real number , no matter how small, but positive, there exist a positive value (possibly depending on ), such that whenever is within distance of a, its image () is within distance of . (In rather coarse intuitive terms, continuity says that if two points x, a are close in the domain, then their images (), () are close in the range. This intuitive formulation is much too vague to be acceptable in a proof, but sometimes it can help guide your intuition.) We say that f is continuous on if it is continuous at every point of . Note that the definition of continuity of a function depends on having a notion of distance between two points, in both the domain and the range. Here the distance between two points , is defined in the obvious way, by (, ) = | |. More generally, let m, n arbitrary positive integer numbers, and consider functions (where is the set of all n-tuples (1 , . . . , ) of real numbers, also known as n-dimensional Euclidean space). In this case the usual distance between two points = (1 , . . . , ) and = (1 , . . . , ) is given by:

, : =
=1

This formula is, of course, motivated by the most familiar 2 and 3-dimensional cases ( = 2, 3), but it works just as well in general. Once we have this distance function, we can define continuity exactly as we did in the 1-dimensional case: Definition 2: We say that : is continuous at a point a iff lim = () . i.e. > 0, > 0 , , < => , < .

2. Definition of a metric space.


This leads to the more abstract notion of a metric space. Definition: A metric space (X, d) is a set X together with a distance function (or metric) d on the set X, i.e. a function satisfying the following four conditions: (1) (2) (3) (4) (, ) (, ) (, ) (, ) 0 for all , ; = 0 if and only if = ; = (, ); (, ) + (, ) for all , , . (this property is called the triangle inequality)

Proposition/Example: ( , d) is a metric space, where , : = Proof: included in exercise 1 below. Exercise 1: do Problem 9, p. 128 of the text.

=1

Checking conditions (1) (2) and (3) is straightforward; the triangle inequality is a little tricky to prove.

We call this metric the usual metric, the standard metric, or the Euclidean metric on . Unless otherwise specified, well always assume the Euclidean metric on . We can generalize many concepts from real analysis to metric spaces. For instance, the concept of a convergent sequence can be extended to metric spaces in a straightforward way. Definition: If (X, d) is a metric space, and 1 is a sequence of elements of X, we say that converges to (shorthand notation: , or = ) iff for every > 0, there exists a positive integer such that for all integers , , < .

3. Continuous functions between metric spaces.


Definition 3: If (, ) and (, ) are metric spaces, we say that a function : is continuous if for all points , lim = (), i.e.: > 0, > 0 , , < => , < . It is also possible to express continuity in terms of convergent sequences. Def.: Given a map : , where X, Y are metric spaces as above, we call f sequentially continuous iff whenever in X, we also have ( ) () in Y. Theorem: Let : be a map between metric spaces. Then f is continuous f is sequentially continuous. (The proof is similar to the proof for : that you probably saw in Math 327) This result is useful because sometimes its easier to work with sequences. Later we will see that this theorem does not hold for arbitrary topological spaces.

4. Subspaces of a metric space.


Notice that if (X, d) is a metric space, and , then A itself is a metric space with the metric inherited from X. We call (A, d) a subspace of (X, d). For example, 3 {0} inherits the usual metric from 3 , and the function : 3 0 given by 1 , , = 2 + 2 + 2 is an example of a continuous, real-valued function on 3 {0}.

5. Different metrics on the same set.


It often happens that we have two or more different metrics on the same set X. Example I: Consider the unit sphere 2 in 3 . Two different metrics come to mind: On the one hand, there is the Euclidean metric d inherited from 3 . In this metric, the distance between the north and south poles is 2. On the other hand, we might prefer to measure distance strictly along the surface of the sphere; in other words, we can define another metric , where , = the length of the shortest great-circle arc on the sphere, connecting x to y. In this metric the distance between the poles is . (It is not obvious that is really a metric; take this on faith for now may see this in Math 442/443.)

Suppose now that we have a function f from 2 to some other metric space Y. For simplicity, lets think of the case Y = , with the usual metric d. Then that the continuity of f doesnt depend on which of the two metrics we use. Using sequential continuity (see the previous section), it is not hard to see why: Suppose is a sequence of points on the sphere, converging to a point x in the d-metric. Then it is intuitively clear that this sequence will also converge to x in the -metric. Conversely, if in the -metric, then in the d-metric. So the definition of convergent sequence doesnt depend on which of the two metrics we use. Using sequential continuity (see 3.), it follows that the continuity of f also does not depend on which of the two metrics we use. Example II: For a simpler example of this phenomenon, consider the following two metrics on 2 : (i) the usual metric d; and (ii) the metric given by ((x1 , y1 ), (x2 , y2 )) = | x1 x2 | + | y1 y2 | (Taxicab metric) Exercise 2: Prove that satisfies the four conditions for a metric. Exercise 3: Let (, ) be any metric space, and let 2 be a function. Prove that f is continuous with respect to the d metric if and only if f is continuous with respect to the -metric. (optional: prove the analogous statement for functions 2 ) At this point you might ask: Is it always true that the property is continuous is independent of the metrics we use on X and Y? The answer is a resounding NO. To find counterexamples to this assertion, we need to expand our list of examples of metric spaces.

6. More examples of metric spaces.


6a. The discrete metric. I pause to make a general comment on understanding new concepts. Whenever a new concept is introduced, you should take a moment to think about extreme cases of it, even if these extreme cases seem trivial or artificial. In the case of metric spaces, there is an extreme case called the discrete metric. Namely, take any set X whatsoever, and define a metric 0 on X by the rule 0, = 0 , = 1, Note that this is a metric. The proof is completely trivial (check!). Exercise 4: Let (, 0 ) be a metric space with the discrete metric. a) Prove that a sequence of points in X converges if and only if it is eventually constant. (start by rigorously defining what eventually constant means.) b) Prove that if (Y, d) is any metric space, and is any function, then f is continuous! Later in the course well also see that if X is a continuous function, then f is constant. Now we can give the first of our counterexamples. Let 0 denote the discrete metric on , and let d denote the usual metric. Then any function whatsoever , 0 (, ) is continuous, by Exercise 4b. But we know lots of examples of functions that are not continuous with respect to the usual metric on both source and target. 0, < 0 For example, the step function = is not continuous if we use the usual metric in the domain, 1, 0 but it is continuous if we use the discrete metric. 6b. Two interesting metrics on sets of functions. Perhaps the preceding example seems a little artificial. Here is a much more substantial counterexample, an example that occurs in nature, if you will.

Let : = : 0,1 | continuos be the set of all continuous real-valued functions on [0,1]. (Remember that unless otherwise specified, the default metric is the Euclidean one) Then X itself has many interesting metrics on it, depending on when you want to consider two functions to be close to one another. Perhaps the first idea that springs to mind is to say that two functions , 0, 1 are close if they are uniformly close. This leads to the uniform metric du, defined by (, ) = max | () () |
[0,1]

where the maximum is taken over the interval. (It is a nontrivial task to prove that such a maximum exists, and the existence depends heavily on the fact that the functions in question are continuous.) (to show that is a metric, do exercise 5 below.) On the other hand, we might wish to consider two functions to be close if they are close in average value. This suggests the averaging metric defined by
1

, =

Recall that continuous functions are Riemann-integrable, so this definition makes sense. The proof that is a metric is more interesting than it was for . For example, suppose we want to prove that , = 0 implies f = g. Then you need to use the fact that if a continuous function taking values in the non-negative real numbers has a zero integral, then it is identically zero. This should be familiar from Math 328. Exercise 5: Show that the uniform metric and the averaging metric really are metrics; i.e. they satisfy the four conditions for a metric stated above. Now lets consider two examples of real-valued functions on our space X. I will give these examples in the form of exercises. Keep in mind that X is a set whose elements are functions. Exercise 6: Define a function by =
1 0

. Show that I is continuous with respect to either

of the two metrics; in other words, both , and , are continuous. Exercise 7: Define a function : by simply evaluating the input at zero: () = (0). a) Show that E is continuous as a function , . b) Show that E is not continuous as a function , . (This part is fun!)
(Notice the nice intuition associated with this problem. Intuitively, the continuity of E with respect to the uniform metric just says that if two functions are close to each other everywhere on the interval, then in particular they are close at 0. The non-continuity of E with respect to the averaging metric reflects the fact that if two functions are close in average value, it does not follow that they are close at any particular point.)

Exercise 6 shows that the metrics and are fundamentally different. After some reflection we see that this is not so surprising. After all, what does it mean for a sequence of functions to converge in the uniform metric? It means precisely that the sequence converges uniformly on the interval [0, 1]. As you learned in Math 327, uniform convergence implies that
1 0

1 0

, but the converse is false. The meaning of

convergent sequence is quite different for these two metrics, and therefore we would expect continuity to behave differently as well. The preceding examples show that while our definition for continuity (so far) involves a metric d on X, if we have two different metrics on the same set X, sometimes they yield the same notion of continuous function, and sometimes they dont. Whats going on?

7. An alternative formulation of continuity.


In order to answer this question, we first reformulate the definition of continuity yet again. Definition: If (X, d) is a metric space, and a is a point of X, for any > 0 we define the (open) r-ball around a by : = , < } Definition: We then say that a subset is open in X iff , > 0 such that . Exercise 8: Show that for any metric space (X, d), the following subsets are always open in X: a) the empty set ; b) X itself; c) (), for any > 0 and any .
(When you prove (c), prove it directly from the axioms for a metric. As a general rule, pictures can guide your intuition, and can and should be used to illustrate your proofs, but be careful, because not all of the intuition you acquired in the Euclidean world carries over to general metric spaces. For example, when X has the discrete metric, () is either just the singleton {x} itself, or the entire space X!)

We can immediately translate our definition of continuity in terms of r-balls, as follows: Definition 4: A function between metric spaces is continuous at a point if and only if > 0, > 0 , ( ). Or, to state the inclusion in terms of pre-images, 1 ( ). But we want to formulate continuity in a way that pushes the metrics even further into the background. Here is our final formulation: Theorem (Definition 5): Let be a function between two metric spaces. Then f is continuous if and only if for every open set U in Y, its pre-image () is open in X. Proof: => First suppose that f is continuous, and let U be an open subset of Y. We must show that 1 () is open in X; that is, that for every 1 () we can find > 0 such that () 1 (). Since U is open in Y, and , there exists > 0 such that . Then 1 ( ) 1 (). Since f is continuous, we can then find > 0 such that 1 ( ). Then: 1 ( ) 1 () as required. <= Conversely, suppose that for every open set U in Y , the pre-image 1 () is open in X. Let , and let > 0. To show that f is continuous, we must find > 0 such that 1 . But is an open set by Exercise 7c, and hence 1 hypothesis. Hence the required exists, by the definition of an open set. Thought Experiment: We have seen several examples of different metrics on the same set. In which cases do the two metrics define the same open sets? In the case of the uniform and averaging metrics, the open sets are definitely different; can you say anything about how they compare? We will return to these questions later. Conclusion. the continuity of a function f between metrics spaces X, Y depends only on the open sets of X and Y In other words, if two metrics on X have the same open sets, then the continuity of f will not depend on which of the two metrics we use. This leads us to the concept of a topological space, and we will now proceed through our text, starting in Ch. 2.

is open by

You might also like