Metric Spaces and Continuity m303 - 4
Metric Spaces and Continuity m303 - 4
Metric Spaces and Continuity m303 - 4
Note to reader
Mathematical/statistical content at the Open University is usually provided to students in
printed books, with PDFs of the same online. This format ensures that mathematical notation
is presented accurately and clearly. The PDF of this extract thus shows the content exactly as
it would be seen by an Open University student. Please note that the PDF may contain
references to other parts of the module and/or to software or audio-visual components of the
module. Regrettably mathematical and statistical content in PDF files is unlikely to be
accessible using a screenreader, and some OpenLearn units may have PDF files that are not
searchable. You may need additional help to read these documents.
Contents
1 Introducing metric spaces 5
1.1 The definition of a metric space 5
1.2 Examples of metric spaces 7
1.3 Understanding the geometry of metric spaces 12
Index 27
1 Introducing metric spaces
5
Metric spaces and continuity 1
Remarks
A ‘point’ may be nothing like a 1. Mathematicians usually refer to the members of the set X as points,
dot in the plane: X could be a to emphasise the analogy with the points of a line, a plane or
set of functions and then a point three-dimensional space. Similarly, mathematicians usually refer to
in X would be a function.
d(a, b) as the distance between a and b. The theory of metric spaces
is the study of those properties of sets of points that depend only on
distance.
2. Condition (M1) says that distance is a non-negative quantity, and that
the only point of a metric space that is at zero distance from a given
point is that point itself.
3. Condition (M2) says that the distance from a point a to a point b is
precisely the same as the distance from b to a. (The metric is
symmetric. It doesn’t matter whether you are going from a to b or
b from b to a – the distance is the same.)
d(b, c)
4. Condition (M3) tells us that d(a, c) gives the ‘shortest’ distance
d(a, b) c
between a and c. For if we go directly from a to c, that gives a
distance of d(a, c). If we make a detour to b along the way, we must go
d(a, c)
a distance d(a, b) to get to b, and then an additional distance d(b, c) to
a
get from b to c (Figure 1.1). Condition (M3) tells us that the total
Figure 1.1 Shortest distance distance travelled must then be at least as great as the ‘direct’
between a and c distance d(a, c).
5. Since our general definition is modelled on the corresponding
properties of the Euclidean distance function d(n) , it follows that
(Rn , d(n) ) is a metric space, for each n ∈ N. It is known as Euclidean
n-space. Furthermore, in the context of metric spaces, the Euclidean
distance function d(n) is often referred to as the Euclidean metric
for Rn . These are our first examples of metric spaces.
If we look back at the proof of the Reverse Triangle Inequality for the
Euclidean metric on Rn (Proposition ??), we see that the proof makes use
only of properties (M1)–(M3) of the Euclidean metric. So it is no surprise
to learn that there is a Reverse Triangle Inequality for any metric space.
6
1 Introducing metric spaces
Remarks
1. When X = ∅, the empty set, the definition still makes sense but is not
very interesting.
2. The d0 -distance between any two distinct points of a set X is always
equal to 1. In particular, if we consider the case when X = R, the real
line, we find that d0 is a different metric from the usual Euclidean
distance. (To see this, observe that d0 (0, 2) = 1 whereas
d(1) (0, 2) = 2.) Thus the discrete metric gives us a second notion of
distance on R. In fact, if (X, d) is any metric space whose metric d is
not d0 , then (X, d0 ) is a second metric space with underlying set X.
The discrete metric is about the simplest possible notion of distance
there can be and acts to ‘isolate’ points of a space: they are all at
distance 1 from one another.
7
Metric spaces and continuity 1
Proposition 1.4
Let X be a set. Then (X, d0 ) is a metric space.
8
1 Introducing metric spaces
Exercise 1.1
Let X = {x, y, z} and define d : X × X → R by
d(x, x) = d(y, y) = d(z, z) = 0,
d(x, y) = d(y, x) = 1,
d(y, z) = d(z, y) = 2,
d(x, z) = d(z, x) = 4.
Determine whether d is a metric on X.
Let us verify that e1 does define a metric: we must check that it satisfies
conditions (M1)–(M3).
(M1) For each a = (a1 , a2 ), b = (b1 , b2 ) ∈ R2 , |b1 − a1 | and |b2 − a2 | are
non-negative, and hence so is their sum: thus e1 (a, b) ≥ 0.
For each a = (a1 , a2 ) ∈ R2 ,
e1 (a, a) = |a1 − a1 | + |a2 − a2 | = 0 + 0 = 0.
Conversely, suppose a = (a1 , a2 ), b = (b1 , b2 ) ∈ R2 are such that
e1 (a, b) = 0. Then
0 = |b1 − a1 | + |b2 − a2 |,
which implies that both |b1 − a1 | = 0 and |b2 − a2 | = 0. Thus a1 = b1
and a2 = b2 ; that is, a = b.
This proves that condition (M1) is satisfied.
(M2) Let a = (a1 , a2 ), b = (b1 , b2 ) ∈ R2 . Using that the modulus function
The modulus function is the has property (M2), we obtain
Euclidean distance function on
R, and so is a metric. e1 (a, b) = |b1 − a1 | + |b2 − a2 |
= |a1 − b1 | + |a2 − b2 | = e1 (b, a).
This proves that condition (M2) is satisfied by e1 .
(M3) Let a = (a1 , a2 ), b = (b1 , b2 ), c = (c1 , c2 ) ∈ R2 . Then, using that the
modulus function has property (M3),
e1 (a, c) = |c1 − a1 | + |c2 − a2 |
≤ (|c1 − b1 | + |b1 − a1 |) + (|c2 − b2 | + |b2 − a2 |)
= (|b1 − a1 | + |b2 − a2 |) + (|c1 − b1 | + |c2 − b2 |)
= e1 (a, b) + e1 (b, c).
This proves that condition (M3) is satisfied by e1 .
We conclude that e1 is a metric on the plane.
Exercise 1.3
Determine which of the following functions define a metric on R:
√ p
(a) d(a, b) = b2 − a2 ; (b) d(a, b) = |b|3 + |a|3 ;
(c) d(a, b) = 3 |b3 − a3 |.
p
10
1 Introducing metric spaces
Exercise 1.4
Show that d has property (M1).
Exercise 1.5
Show that d has property (M2).
11
Metric spaces and continuity 1
12
1 Introducing metric spaces
Bd [a, 0] = {x ∈ X : d(a, x) ≤ 0}
= {x ∈ X : d(a, x) = 0} = {a},
and
Sd (a, 0) = {x ∈ X : d(a, x) = 0} = {a}.
We now discover what open balls, closed balls and spheres look like for
some of the metric spaces we have met already.
Let us start by determining the open and closed balls for the discrete
metric, d0 .
Exercise 1.6
Let X be a non-empty set and a ∈ X. Determine Bd0 [a, r] for r ≥ 0.
Exercise 1.7
Let X be a non-empty set and a ∈ X. Determine Sd0 (a, r) for r ≥ 0.
Next, let us look at open balls defined using the taxicab metric.
13
Metric spaces and continuity 1
x2
(0, 1) Worked Exercise 1.9
Consider the metric space (R2 , e1 ) – that is, the plane with the
taxicab metric. Find the unit open ball Be1 (0, 1).
(x1 , 1 − x1 )
Solution
x1 + x2 = 1
The centre is 0 = (0, 0), and we want to find all points x = (x1 , x2 )
that satisfy
1
Be1 (0, 1)
Exercise 1.8
0 x1 Sketch the open ball Be1 ((2, 3), 2), briefly justifying your answer.
−1 1
−1 Let us finish this discussion by looking at the balls for our mixed metric in
the plane, d.
Figure 1.6 The open ball
Be1 (0, 1)
Worked Exercise 1.10
Let a = (a1 , a2 ) ∈ R2 . Determine Bd (a, r) for r > 0, where d denotes
the mixed metric in the plane given by
d(a, b) = |b1 − a1 | + d0 (a2 , b2 ) for a = (a1 , a2 ), b = (b1 , b2 ) ∈ R2 .
14
1 Introducing metric spaces
Solution
Let a = (a1 , a2 ) ∈ R2 and suppose that r > 0. Then, using the
definition of an open ball,
Bd (a, r) = {x = (x1 , x2 ) ∈ R2 : d(a, x) < r}
= {x = (x1 , x2 ) ∈ R2 : |x1 − a1 | + d0 (a2 , x2 ) < r}.
We now consider the cases x2 = a2 and x2 6= a2 separately. If x2 = a2 ,
then d0 (a2 , x2 ) = 0 and so
{x = (x1 , a2 ) ∈ R2 : |x1 − a1 | < r} = {x1 ∈ R : |x1 − a1 | < r} × {a2 }
= (a1 − r, a1 + r) × {a2 }.
If x2 6= a2 , then d0 (a2 , x2 ) = 1 and
{x = (x1 , x2 ) ∈ R2 : x2 6= a2 and |x1 − a1 | + d0 (a2 , x2 ) < r}
= {x = (x1 , x2 ) ∈ R2 : x2 6= a2 and |x1 − a1 | < r − 1}.
Let us unpack this set expression: x2 can be any value other than a2
but if r ≤ 1, then there are no possible x1 for which |x1 − a1 | < r − 1,
and so the expression is the empty set, ∅. Whereas if r > 1, then the
set of x1 ∈ R for which |x1 − a1 | < r − 1 is the interval
(a1 − (r − 1), a1 + (r − 1)). Hence for r > 1, the set expression is
(a1 − (r − 1), a1 + (r − 1)) × (R − {a2 }). Summarising,
{x = (x1 , x2 ) ∈ R2 : x2 6= a2 and |x1 − a1 | + d0 (a2 , x2 ) < r}
∅, if r ≤ 1,
=
(a1 − (r − 1), a1 + (r − 1)) × (R − {a2 }), if r > 1.
Hence, as shown in Figure 1.7,
(a1 − r, a1 + r) × {a2 }, if r ≤ 1,
Bd (a, r) = ((a1 − r, a1 + r) × {a2 })
∪ ((a1 − (r − 1), a1 + (r − 1)) × R), if r > 1.
2r − 2
a = (a1 , a2 ) a = (a1 , a2 )
(a1 − r, a2 ) (a1 + r, a2 ) (a1 − r, a2 ) (a1 + r, a2 )
r−1
Bd (a, r) Bd (a, r)
(a) (b)
15
Metric spaces and continuity 1
Exercise 1.9
Let a = (a1 , a2 ) ∈ R2 . Determine Bd [a, r] for r ≥ 0, where d denotes the
mixed metric in the plane given by
d(a, b) = |b1 − a1 | + d0 (a2 , b2 ) for a = (a1 , a2 ), b = (b1 , b2 ) ∈ R2 .
Note that this definition of a sequence does not require that we impose any
additional structure (such as a metric) on the set X.
16
2 Sequences in metric spaces
Remarks
1. Note that the limit a must be a point in X. For example, if X is the It is not hard to check that this
open interval (0, 1) and we give it the usual notion of distance defines a metric on (0, 1) – we
d(a, b) = |a − b|, then the sequence ( n1 ) is not d-convergent, since the return to the general problem of
defining metrics on subsets of
only possible choice of limit is 0 – which is not a point of X. metric spaces in the next
2. In order to show that a sequence (ak ) converges to a for the metric d, chapter.
we must show that d(ak , a) → 0 as k → ∞ – that is, for each ε > 0,
there is an N ∈ N for which d(ak , a) < ε for all k > N .
Exercise 2.1
Let (R2 , e1 ) be the plane with the taxicab metric, and let (an ) be the
1 1
sequence given by ak = (1 + 2k , 2 − 2k ). Show that (ak ) converges to (1, 2)
with respect to e1 .
Convergent sequences in (Rn , d(n) ) have unique limits – that is, a sequence
cannot simultaneously converge to two different limits. The next result
establishes this as a fact in any metric space.
17
Metric spaces and continuity 1
18
3 The definition of continuity in metric spaces
On the other hand, we know that the sequence ( k1 ) does converge to 0 with
respect to the Euclidean metric d(1) on R. The important conclusion to
draw from this example is:
convergence depends on how we measure distance.
In other words, different metrics on the same set can give rise to different
convergent sequences.
From the solution to Worked Exercise 2.5, we know that eventually
constant sequences (ak ) in a metric space (X, d0 ) are d0 -convergent. The
next exercise asks you to show that the eventually constant sequences are
the only convergent sequences for the d0 metric on a set X.
Exercise 2.2
Let X be a set and d0 the discrete metric for X. Suppose that (ak ) is a
sequence in X that is d0 -convergent. Show that (ak ) is an eventually
constant sequence.
19
Metric spaces and continuity 1
Remarks
1. When (X, d) = (Rn , d(n) ) and (Y, e) = (Rm , d(m) ), we recover the
definition of continuity given in Definition ?? of Subsection ??.
2. We have not yet defined what it means to be continuous on a subset A
See Theorem ?? of Chapter 15. of the metric space X; we return to this issue in the next chapter.
Our next worked exercise shows that this definition can make some
surprising functions continuous.
This is a rather artificial example and it tells us that every function from
In fact one can use a similar R to R is (d0 , d(1) )-continuous on R. However, it does illustrate that our
argument to show that any intuitive notion of what continuity means breaks down when looking at
f : X → Y is (d0 , e)-continuous metrics different from the Euclidean ones, and so highlights the
on X, no matter the choice of
metric e for Y .
importance of working from the definition.
20
3 The definition of continuity in metric spaces
At the moment our stock of metric spaces is quite small: Euclidean spaces,
the plane with the taxicab metric, the plane with a particular ‘mixed’
metric, and arbitrary sets with the discrete metric. In the next chapter we
will look at more examples of metric spaces and examine further the notion
of continuity. What we can do at this point, though, is prove a useful
result that applies to all continuous functions and which is an extension of
the Composition Rule for continuous functions between Euclidean spaces.
21
Metric spaces and continuity 1
This means that equation (1) holds, because g(yk ) = (g ◦ f )(ak ) and
g(y) = (g ◦ f )(a).
Thus g ◦ f is continuous at the arbitrary point a ∈ X, and so we conclude
that g ◦ f is continuous on X.
Exercise 3.1
Let (X, d) and (Y, e) be metric spaces, and let b ∈ Y be fixed. Suppose
that f : X → Y is (d, e)-continuous on X.
Use the Reverse Triangle Inequality to prove that g : X → R given by
g(x) = e(f (x), b) is (d, d(1) )-continuous on X.
22
Metric spaces and continuity 1
23
Solutions and comments on exercises
and so
d(a, c)3 ≤ (d(b, c) + d(a, b))3 = (d(a, b) + d(b, c))3,
as required.
Hence (M1)–(M3) hold and we conclude that d is a metric for R.
24
Metric spaces and continuity 1
(2, 3) The point (2, 3) is the centre of the ball, and its radius is 2. If we imagine
3 the origin of coordinates at (2, 3) and put the vertices of the diamond two
units from the centre along the axes, the required figure results (as shown).
25
Solutions and comments on exercises
26
Index
Index
Bd (a, r), the open ball of radius r with centre a, 11 Euclidean distance function, 4
Bd [a, r], the closed ball of radius r with centre a, 11 Euclidean n-space, 5
balls eventually constant, 17
closed, 11
open, 11 function
(d, e)-continuous, 19
centre, 11 (d, e)-discontinuous, 19
closed ball, 11
Composition Rule limit of a sequence, 16
for continuous functions between metric spaces, 20
continuity at a point metric, 4, 5
for metric spaces, 19 discrete, 6
continuity for metric spaces, 19 Euclidean, 5
continuous function mixed (on the plane), 10
between metric spaces, 19 taxicab, 8, 13
continuous functions between metric spaces metric space, 5
Composition Rule, 20 points, 5
convergent sequence Reverse Triangle Inequality, 6
in a metric space, 16 mixed metric in the plane, 10
d-closed ball, 11 open ball, 11
definition
d-convergent sequence, 16 point in a metric space, 5
d-divergent sequence, 16
closed ball, 11 radius, 11
continuous function between metric spaces, 19 Reverse Triangle Inequality
convergent sequence in a metric space, 16 for metrics, 6
discontinuous function between metric spaces, 19
eventually constant, 17 Sd (a, r), the sphere of radius r with centre a, 11
metric space, 5 sequence
open ball, 11 eventually constant, 17
sequence in a metric space, 15 sequence in a metric space, 15
sphere, 11 convergence, 16
discontinuous function divergence, 16
between metric spaces, 19 limit, 16
discrete metric, 6 sphere, 11
distance
in a metric space, 5 taxicab metric, 8
distance function, 4 open balls, 12
d-open ball, 11 Triangle Inequality
d-sphere, 11 condition for metrics, 5
divergence, 16 in Rn , 4
divergent sequence
in a metric space, 16 unit closed ball, 11
unit open ball, 11
Euclidean metric, 5 unit sphere, 11
27