23

...
...
...
...
...
...
...
...
...
...
...
Metric Spaces
...
...
...
C H A P T E R .

There are many ways to measure distance in the spaces in which

we live and work. For example, if you want the shortest distance
between two geographical places (the distance “as the crow flies”),
you follow the line segment joining them. But in real life this isn’t
always possible. If you are driving your car through a city or across
your campus, you need to go around solid objects and not through
them. So how do we calculate distance in those cases? Measuring
distance in a set X is a very small (but interesting) part of a branch
of mathematics known as “point set topology,” and we will look at it
in detail in this chapter. We will now often refer to the elements of
X as points.
So let’s go back to the first time you measured distance. It was
probably in R, on a number line, and you learned that the distance
between two points x and y was the absolute value of the difference
of the two numbers. If we write d(x, y)  |x − y|, then d is a function
and d : R × R → R. That’s straightforward enough, but now we want
to generalize our concept of distance. So let’s turn to the essential
properties of a distance function.
First, distance shouldn’t be negative, so d(x, y) ≥ 0 for two points
x and y, and if the distance satisfies d(x, y)  0, then you didn’t
move anywhere, so x  y. You also surely believe that distance from

283
284 23. Metric Spaces

x to y should be the distance from y to x. And finally, in Theorem 5.5

(Problem 5.9) you learned the triangle inequality, which said that “if
x and y are two real numbers, then |x + y| ≤ |x|+|y|.” In Problem 19.1,
you showed how to switch the triangle inequality into a statement
about distances. We recall the result of that problem here: For real
numbers x, y, and z

|x − y| ≤ |x − z | + |z − y|.
In English, this means that our path will be shorter if we go
directly from x to y as opposed to taking a detour through z, which
is as it should be. So we would want our general distance function
to satisfy something like this too; that is, in our new “d” notation we
want d(x, y) ≤ d(x, z) + d(z, y) for arbitrary points x, y, and z. So now
we will define something that acts like a distance on an arbitrary set
X and does all the important things that a distance should do.
Let X be a nonempty set. Then a metric on X is a function
d : X × X → R satisfying (i)–(iv) below.

(i) (Nonnegativity) For all x, y ∈ X, the function d satisfies

d(x, y) ≥ 0.
(ii) (Definiteness) For all x, y ∈ X, the function d satisfies d(x, y) 
0 if and only if x  y.
(iii) (Symmetry) For all x, y ∈ X, the function d satisfies d(x, y) 
d(y, x).
(iv) (Triangle inequality) For all x, y, z ∈ X, the function d satisfies

d(x, y) ≤ d(x, z) + d(z, y).

A metric is also called a distance function. A set X together with

the metric d is called a metric space and is denoted by (X, d), or just
X when it is clear which distance function we are using. Conditions
(i) and (ii) together are usually called positive definiteness.
When you learn the definition, don’t forget to say “Let X be a
nonempty set. Then a metric on X is a function d : X × X → R . . .”
These sentences tell us something about d, and cannot be omitted.
In the introduction, we showed that a metric can be defined on
R by du (x, y)  |x − y|. Though we outlined how to show that du
is a metric, you should write out the details to complete the proof.
This metric is often called the usual metric (hence the subscript u)
 23. Metric Spaces 285

or the Euclidean metric on R, and it is the one upon which your

intuition is almost certainly based. A set can have lots of metrics.
The next example is a metric on R that is not the same as the metric
given by the absolute value.

Example 23.1.
Define a metric dd : R × R → R by

0 if x  y
dd (x, y)  .
1 if x
 y

We will show that dd is a metric on R. This metric is called the

discrete metric, and it can really challenge your intuition.

Proof.
It is clear that dd is a function from R × R → R. Now let x and
y be points of R. We begin with nonnegativity: dd (x, y)  0 or
dd (x, y)  1, so clearly dd (x, y) ≥ 0. Thus, the nonnegativity con-
dition holds. Furthermore, since dd (x, y)  0 if and only if x  y,
the definiteness condition holds. For symmetry, note that if x
 y,
then y
 x and consequently dd (x, y)  1  dd (y, x). If x  y, then
dd (x, y)  0  dd (y, x), establishing symmetry. Finally, to establish
the triangle inequality, note that if z is a point of R, then we have two
cases to consider. In the first case, if x  y, then dd (x, y)  0 and the
nonnegativity condition implies that dd (x, y)  0 ≤ dd (x, z) + dd (z, y).
In the second case, x
 y, which implies that z
 x or z
 y (or
both). Therefore, either dd (x, z)  1 or dd (z, y)  1 (or both). Thus,
dd (x, y)  1 ≤ dd (x, z) + dd (z, y), completing the proof of the triangle
inequality.

The discrete metric can be defined on every space: the distance

between two distinct points is one, and the distance from a point to
itself is necessarily zero. The proof that this is a metric on a set X
is indistinguishable from the one above. Thus we have an example
of a metric on R2 . Example 23.2 and Exercise 23.3 provide us with
some other metrics on R2 .
286 23. Metric Spaces

Example 23.2.
On R2 define a metric by


du ((x1 , x2 ), (y1 , y2 ))  (x1 − y1 )2 + (x2 − y2 )2 .

Using Project 27.9, it can be shown that this is actually a metric on

R2 . For now, you may accept this fact. This metric is referred to as
the usual metric or the Euclidean metric on R2 . In fact, one may
also define the usual metric on Rn by


 n
du ((x1 , x2 , . . . , xn ), (y1 , y2 , . . . , yn ))   (xj − yj )2 .
j1

Exercise 23.3.
We now have two examples of metrics on R and two on R2 . Here are
two more metrics on R2 . Before you begin the exercise, familiarize
yourself with the metrics by computing various distances. For exam-
ple, try to find the distance from the point (1, 3) to the points (−3, 4)
using the various metrics below.
(a) Show that dtc ((x1 , x2 ), (y1 , y2 ))  |x1 − y1 | + |x2 − y2 | is a metric
on R2 . This metric, dtc , is called the taxicab metric. Why would
it be called that?
(b) Show that dm ((x1 , x2 ), (y1 , y2 ))  max{|x1 − y1 |, |x2 − y2 |} is also
a metric on R2 . The metric dm is sometimes called the max
metric.

The two examples introduced in Exercise 23.3 will appear again

in the near future.
A metric tells us when points are close. We studied the notion of
“closeness” in Chapter 19 when we studied convergent sequences.
You can picture convergence of a sequence to the number L in the
following way: a sequence converges to L if for every  > 0, the
sequence eventually lies in the open interval (L − , L + ); more
precisely, the definition of convergence said, “There exists a real
 23. Metric Spaces 287

number L such that for every  > 0, there exists a real number
N such that |xn − L| <  for all n ≥ N.” We return to the idea of
finding the limit of a sequence, but this time in a metric space. So
given a sequence (xn ) of points in a metric space (X, d), then (as
we did before) we say that (xn ) converges in X if there exists a
point x ∈ X such that for every  > 0, there exists a real number N
such that d(xn , x) <  for all n ≥ N. The value x is called the limit
of the sequence, we say that the sequence converges to x and, as
before, we write xn → x or limn→∞ xn  x. If the sequence does
not converge, we say that it diverges. If we consider X  R with
the usual metric, this is exactly the same definition that we had in
Chapter 19. Since we allow all sorts of choices for X now, we would
like to take this opportunity to point out that the point x must be in
the space X—not in some larger space that happens to contain X. If
it is clear that x belongs to X, we will often say that the sequence
converges, rather than “the sequence converges in X.” Also, note
that as the metric d changes, the distance between pairs of points
changes as well. Therefore, it is conceivable that some sequences
will converge in one metric, but not in another.

Exercise 23.4.
Complete the sentences.
(a) Let (xn ) be a sequence in a metric space X with metric d. Let
x ∈ X. Then (xn ) does not converge to x if . . . .
(b) Let (xn ) be a sequence in a metric space (X, d). Then (xn ) does
not converge if . . . .

We’ll break tradition and give you the answer to part (a) of the
above exercise here, because we need it: A sequence (xn ) does not
converge to x if there exists an  > 0 such that for every real number
N, there exists m ∈ N such that m ≥ N and d(xm , x) ≥ . While an
answer to (b) might read “a sequence (xn ) does not converge if for
every x ∈ X, the sequence does not converge to x,” this will probably
not be the most useful formulation of the answer. We leave the more
useful version to you.

2 8 8 23. Metric Spaces

Example 23.5.
We know that 1/n > 0 in R with the usual metric. Show that
(1/n, 1/n) (0, 0) in R2 with the usual metric.

Proof.
Let e > 0, and let N be a real number with N > \[2/e. If n is an
integer with n N, then
du((1/n, 1/n), (0, 0)) = (1/n 0)2 + (1/n 0)2

< 2/N (since n N)

(as N

See Figure 23.1 for a graphical illustration of this convergent

sequence. •

You may wonder where we came up with y/2/e. We did it by un-

derstanding the problem and devising a plan by working backwards.
So what you see here is what happened after we went to a separate
sheet of paper, and started with the inequality V2/n < e.

Example 23.6.
In Chapter 19, we showed that the sequence (1/n) converges to 0
in R with the usual metric. Does (1/n) converge to 0 in the discrete
metric?

R2

FIGURE 23.1 (1/n, 1/n) (0, 0)

 23. Metric Spaces 289

We claim that the sequence (in R with the discrete metric) does
not converge to 0. To see this, let   1/2. For every N ∈ R, there
exists an integer n ≥ N. Since 1/n
 0, we know that dd (1/n, 0)  1.
Hence for   1/2, and for every N ∈ R, there exists an integer
n ≥ N such that xn  1/n satisfies dd (xn , 0)  dd (1/n, 0)  1 ≥ 1/2.
Thus (1/n) does not converge to 0.

In the discrete metric, every point is “far” from every other point.
This makes it very hard to converge.

Exercise 23.7.
Consider R with the discrete metric. Describe the convergent
sequences in this metric space.

Sequences have many important properties, some of which we

discuss in the problems. The proofs are often quite similar to the
proofs we did in Chapter 19. At this point, we give one example of a
theorem with such a proof.

Theorem 23.8.
If a sequence (xn ) in a metric space (X, d) converges, then the limit is
unique.

The proof of this is the same as the proof of Theorem 19.7.

Solutions to Exercises
Solution to Exercise (23.3).
Parts (a) and (b) are very similar, so we will work part (a) only.
By definition dtc is a function from R × R to R. Now let (x1 , x2 )
and (y1 , y2 ) be elements of R2 . We will first show the nonneg-
ativity. Because |a| ≥ 0 for all real numbers a, we know that
dtc ((x1 , x2 ), (y1 , y2 ))  |x1 − y1 | + |x2 − y2 | ≥ 0, showing that
nonnegativity of dtc holds. For definiteness, we point out that
dtc ((x1 , x2 ), (y1 , y2 ))  0 if and only if |x1 − y1 | + |x2 − y2 |  0. This
last equality holds if and only if |x1 − y1 |  0 and |x2 − y2 |  0.
290 23. Metric Spaces

This, in turn, holds if and only if x1  y1 and x2  y2 ; in other

words, (x1 , x2 )  (y1 , y2 ). This string of equivalences establishes the
definiteness of dtc . Symmetry is shown as follows:

dtc ((x1 , x2 ), (y1 , y2 ))  |x1 − y1 | + |x2 − y2 |

 |y1 − x1 | + |y2 − x2 |
 dtc ((y1 , y2 ), (x1 , x2 )).

To prove that the triangle inequality holds for dtc , let (z1 , z2 ) ∈ R2 .
Then

dtc ((x1 , x2 ), (y1 , y2 ))  |x1 − y1 | + |x2 − y2 |

≤ |x1 − z1 | + |z1 − y1 | + |x2 − z2 | + |z2 − y2 |
(by the triangle inequality in R)
 (|x1 − z1 | + |x2 − z2 |) + (|z1 − y1 | + |z2 − y2 |)
 dtc ((x1 , x2 ), (z1 , z2 )) + dtc ((z1 , z2 ), (y1 , y2 )).

This shows that dtc is a metric on R2 . The taxicab metric between

two points measures the distance you have to travel from one point
to the next in a city built with rectangular blocks, assuming you stay
on the streets, do not take detours, and don’t have to worry about
one-way streets.

Solution to Exercise (23.4).

The solution to part (a) was given earlier following the exercise, so
here is the solution to part (b).
The sequence (xn ) does not converge in the metric space (X, d) if
for every x ∈ X there exists a real number  > 0 such that for every
real number N, there exists m such that m ≥ N and d(xm , x) ≥ .

Solution to Exercise (23.7).

We claim that a sequence (xn ) in (R, dd ) converges if and only if there
exist real numbers x and M such that xn  x for all n ≥ M. (Such a
sequence is called an eventually constant sequence.)
First assume that (xn ) is a sequence for which there exist real
numbers x and M satisfying xn  x for all n ≥ M. We will show that
xn → x. Let  > 0 and let N  M. Then for n ≥ N we know that
dd (xn , x)  dd (x, x)  0 < , which shows that (xn ) converges.
 23. Metric Spaces 291

For the converse, assume that (xn ) converges. Consider   1/2.

Then there exists N such that dd (xn , x) < 1/2 for n ≥ N. By the
definition of dd , the only way that this can happen is if xn  x for
n ≥ N. Taking M  N, we have shown that there exist x and M such
that xn  x for all n ≥ M, as desired.

Problems
Unless otherwise specified, assume that you are working in a general
metric space (X, d).

Problem 23.1.
(a) Suppose a student writes the following: A metric is a function
satisfying (i)–(iv) below.
(i) (Nonnegativity) d(x, y) ≥ 0,
(ii) (Definiteness) d(x, y)  0, if and only if x  y,
(iii) (Symmetry) d(x, y)  d(y, x), and
(iv) (Triangle inequality) if z is a point in X, then

d(x, y) ≤ d(x, z) + d(z, y).

Write this student a letter indicating what was omitted

from the definition, what must be inserted, and what else (if
anything) needs to be changed to make it a correct definition.
(b) Suppose the student had exactly the same definition as in the
text, except for the triangle inequality, where the student has “
d(x, y) ≤ d(x, z) + d(z, y) for some z ∈ X.”
Write a correct, careful, and complete response to this
student.

Problem 23.2.
(a) In R, find the distance of the number 1 to the number 3 in the
usual metric and in the discrete metric.
(b) In R2 , find the distance of the point (1, 3) to the point (2, 5) in
the usual metric, the taxicab metric, the max metric, and the
discrete metric.
292 23. Metric Spaces

Problem 23.3.
(a) Sketch the set {(x, y) ∈ R2 : du ((x, y), (0, 0)) < 1}, where du is
the usual metric.
(b) Sketch the set {(x, y) ∈ R2 : dtc ((x, y), (0, 0)) < 1}, where dtc is
the taxicab metric.
(c) Sketch the set {(x, y) ∈ R2 : dm ((x, y), (0, 0)) < 1}, where dm is
the max metric.
(d) Sketch the set {(x, y) ∈ R2 : dd ((x, y), (0, 0)) < 1}, where dd is
the discrete metric.
(e) Sketch the set {(x, y, z) ∈ R3 : du ((x, y, z), (0, 0, 0)) < 1}, where
du is the usual metric. (See Example 23.2 for the definition if
you need it.)

Problem 23.4.
(a) We defined the max metric on R2 . Define the max metric on
Rn and prove that it is a metric.
(b) We defined the taxicab metric on R2 . Define the taxicab
metric on Rn and prove that it is a metric.

Problem 23.5.
(a) Show that d : R × R → R defined by
d((x1 , x2 ), (y1 , y2 ))  |x1 − y1 |
is not a metric on R2 .
(b) Is d : R × R → R defined by
d((x1 , x2 ), (y1 , y2 ))  (x1 − y1 )2 + (x2 − y2 )2
a metric on R2 ?

Problem 23.6.
Let (X, d) be a metric space. Let α be a real number and define a
new function dα on X × X by dα (x, y)  αd(x, y). Is dα a metric on
X? If not, what assumptions must be placed on α to assure that dα is
a metric? Prove your answer.

Problem 23.7.
A set F in a metric space (X, d) is bounded if there exists a positive
number M such that d(x, y) ≤ M for all x, y ∈ F.
 23. Metric Spaces 293

(a) Consider the following “not a definition” of a bounded set.

“A set is bounded if for each x, y ∈ F there exists a positive
number M such that d(x, y) ≤ M.”
Give a complete, clear, concise explanation of the problems
with this definition.
(b) Give an example of a metric space and an infinite set that is
bounded in that metric. Prove that it is bounded.
(c) Complete the following definition: Let X be a metric space with
metric d and let F be a subset of X. Then F is not bounded if
....
(d) Give an example of a metric space and a set that is not bounded
in that metric. Prove that it is not bounded.

Problem 23.8.
Let X be a set with a metric d. Define a function db : X × X → R by

db (x, y)  min{d(x, y), 1}.

(a) Show that db is a metric on X. This metric is called the bounded

metric associated with d on X.
(b) (This part uses Problem 23.7.) Consider the metric space
(X, db ). Show that in this space, every subset of X is bounded.

Problem 23.9.
Show that in a metric space (X, d) the metric satisfies

|d(x, z) − d(y, z)| ≤ d(x, y),

for all x, y, z ∈ X.

Problem 23.10.
Let X be the space of polynomials with real coefficients. Define a
function d from X × X → R by d(p, q)  |p(0) − q(0)|. Is d a metric?
If so, prove it. If not, why not?

Problem 23.11.
The following exercise is only appropriate if you have had integration in
calculus.
2 9 4 23. Metric Spaces

Let X be the space of real-valued continuous functions defined

on the interval [0,1]. Define a function d : X × X R by

=fJo
for all f,g X.
(a) Show that d is a metric.
(b) Find the distance between ex and sin(πx/2).

Problem 23.12.
Choose a fixed point X0 in R2. If du denotes the usual (or Euclidean)
metric on R , then we define d : R × R2

I
2 2
R by
du(x, y) if x and y are on a straight

line through x0
du(x, X0) + du(x0, y) otherwise.
Figure 23.2 illustrates this function for three pairs of points in the
plane. Prove that d is a metric on R2.
This metric is sometimes called the "French railway system met-
ric." (See [39, p. 56].) Why? Think of x0 as Paris, and you'll note that
all trains pass through Paris, whether they need to or not.
Problem 23.13.
Prove each of the following.
(a) Consider R2 with the max metric. Prove that (1/n, 2/n) >
(0,0).
(b) Consider R with the usual metric. Prove that ( 1) n n/(3n+1) •
0.

x3

FIGURE 23.2 The metric: d(xi, yi) for i = 1, 2, 3

 23. Metric Spaces 295

(c) Consider R2 with the max metric. Does ((−1)n , 2/n)) converge
in this space?

Problem 23.14.
Consider Z with the usual metric.
(a) Show that a sequence that is eventually constant converges;
that is, if there exists an integer m such that xn  xk for all
n, k ≥ m, then the sequence converges.
(b) Can you give other examples of convergent sequences in
(Z, du )? Explain your answer.

Problem 23.15.
Let (X, d) be a metric space, and let (xn ) be a convergent sequence
in X.
(a) Prove that there exists x ∈ X and a natural number K such that
d(xn , x) ≤ K for all n ∈ N.
(You should know a similar problem.)
(b) Prove that the set {xn : n ∈ N} is bounded, as defined in Problem
23.7; that is, prove that there exists a positive number M such
that d(xn , xm ) ≤ M for all n, m ∈ N.

Problem 23.16.
In Problem 19.14, part (c), we defined the term Cauchy sequence
and proved some facts about such sequences. This problem asks you
to do the same in a general metric space.
(a) Define Cauchy sequence in a metric space (X, d).
(b) Prove that if (xn ) converges in (X, d), then (xn ) is Cauchy.

Problem 23.17.
(This problem uses Problem 23.16.) Let X  R \ Q with√the usual
metric du . Prove that the sequence (xn ), where xn  2/n, is a
Cauchy sequence in X, but (xn ) does not converge in X.