Metric Spaces Lecture 3

Examples of metric spaces

(1) Let S = C
n
= { (x
1
, x
2
, . . . , x
n
) | x
i
C}, and let p > 1. For x = (x
1
, x
2
, . . . , x
n
)
and y = (y
1
, y
2
, . . . , y
n
) in C
n
dene
d
p
(x, y) =
_
n

k=1
|x
k
y
k
|
p
_
1/p
.
Then d
p
is a metric.
To prove this one must check the axioms. First, since |x
k
y
k
| = |y
k
x
k
| 0, it is
obvious that d
p
(x, y) = d
p
(y, x) 0 for all x and y. Furthermore, since

k
|x
k
y
k
|
p
= 0
if and only if all the terms |x
k
y
k
|
p
are zero, we see that d
p
(x, y) = 0 if and only if x = y.
To verify the remaining axiom we use Minkowskis Inequality.
Let x, y, z C
n
, and dene a
k
= y
k
x
k
and b
k
= x
k
z
k
, where x
k
, y
k
and z
k
are
the k-th coordinates of x, y and z respectively. Then a
k
+ b
k
= y
k
z
k
, and so
d
p
(y, z) =
_
n

k=1
|a
k
+ b
k
|
p
_
1/p

_
n

k=1
|a
k
|
p
_
1/p
+
_
n

k=1
|b
k
|
p
_
1/p
= d
p
(y, x) + d
p
(z, x)
as required.
Note that we could have used R
n
rather than C
n
, and everything would have worked
in the same way.
Two particular cases should be singled out. If p = 2 we get the usual Euclidean
metric,
d(x, y) =
_
|x
1
y
1
|
2
+|x
2
y
2
|
2
+ +|x
n
y
n
|
2
.
If p = 1 we get
d(x, y) = |x
1
y
1
| +|x
2
y
2
| + +|x
n
y
n
|,
which is the distance from x to y if you can only move parallel to the coordinate axes.
(2) Let S be any set. For x, y S, dene
d(x, y) =
_
0 if x = y
1 if x = y.
This is easily shown to be a metric; it is known as the standard discrete metric on S.
(3) Let d be the Euclidean metric on R
3
, and for x, y R
3
dene
d(x, y) =
_
d(x, y) if x = sy or y = sx for some s R
d(x, 0) + d(0, y) otherwise.
This says that d
1
(x, y) is the distance from x to y if you can only travel along rays through
the origin. Then d
1
is a metric.
1
(4) Let S = C
n
and d
p
as in the rst example above. We considered the case p = 1; so
it is natural to ask about the case q = 1, where p and q are related as in our discussion
of Holders Inequality. The relationship is 1/p + 1/q = 1, and if q = 1 this gives 1/p = 0,
which is nonsense of course. However, since 1/p 0 as p , perhaps it makes sense
to dene
d

(x, y) = lim
p
d
p
(x, y)
and perhaps d

will be a metric.
Let x = (x
1
, x
2
, . . . , x
n
), y = (y
1
, y
2
, . . . , y
n
) C
n
, and dene
s = max{ |x
1
y
1
|, |x
2
y
2
|, . . . , |x
n
y
n
| }.
Then for all p > 1,
s
p
|x
1
y
1
|
p
+|x
2
y
2
|
p
+ +|x
n
y
n
|
p
ns
p
,
and so taking p-th roots we obtain
s (|x
1
y
1
|
p
+|x
2
y
2
|
p
+ +|x
n
y
n
|
p
)
1/p
n
1/p
s.
That is, s d
p
(x, y) n
1/p
s, for all p > 1. But lim
p
n
1/p
= 1; so by the squeeze law
we deduce that lim
p
d
p
(x, y) = s.
Anyway, it is easy to check directly that d

dened by d

(x, y) = max
i
|x
i
y
i
| is
a metric on C
n
.
More background
Recall that a subset X of R is said to be bounded above if there exists a B R such
that x B for all x X. It is an axiom of the real number system that every subset X
of R which is bounded above has a least upper bound, or supremum, supX. Thus if X is
bounded above the supremum of X has the following two properties:
(i) for all x X, x supX;
(ii) if B R satised x B for all x X then supX B.
More examples
Let p 1, and dene
p
to be the set of all innite sequences (x
k
)

k=1
in R such that

k=1
|x
k
|
p
converges. If a = (a
k
)

k=1
and b = (b
k
)

k=1
are any elements of
p
, then for
each positive integer n Minkowskis Inequality gives
_
n

k=1
|a
k
b
k
|
p
_
1/p

_
n

k=1
|a
k
|
p
_
1/p
+
_
n

k=1
|b
k
|
p
_
1/p
A
1/p
+ B
1/p
,
where A =

k=1
|a
k
|
p
and B =

k=1
|b
k
|
p
. Thus

n
k=1
|a
k
b
k
|
p
(A
1/p
+ B
1/p
)
p
for
all n, and so the series

k=1
|a
k
b
k
|
p
is convergent. Dening
d(a, b) =
_

k=1
|a
k
b
k
|
p
_
1/p
;
2
makes
p
into a metric space. The only nontrivial thing to prove is the triangle inequality,
and this is quite easy also: if (x
k
), (y
k
), (z
k
)
p
then by Minkowskis Inequality
_
n

k=1
|y
k
z
k
|
p
_
1/p

_
n

k=1
|y
k
x
k
|
p
_
1/p
+
_
n

k=1
|x
k
z
k
|
p
_
1/p

k=1
|x
k
y
k
|
p
_
1/p
+
_

k=1
|x
k
z
k
|
p
_
1/p
= d(x, y) + d(x, z)
for all n, and taking the limit as n gives d(y, z) d(x, y) + d(x, z).
Let [, ] be any nite closed interval in R, and C[, ] the set of all continuous
real-valued functions on [, ]. Let p 1, and for f, g C[, ] dene
d(f, g) = d
p
(f, g) =
_
_

|f(x) g(x)|
p
dx
_
1/p
.
With this denition of distance, C[, ] becomes a metric space.
Again, the proof of the triangle inequality uses Minkowskis Inequality. By the de-
nition of the Riemann integral, if is any element of C[, ] then
_

(x) dx = lim
n
n

i=1
(x
i
)/n
where x
i
= + (i( )/n). Hence
lim
n
_
n

i=1
(x
i
)/n
_
1/p
=
_
_

(x) dx
_
1/p
.
Now let f, g, h be arbitrary elements of C[, ], and let n Z
+
. By Minkowskis identity
applied with a
i
= (g(x
i
) f(x
i
))/n
1/p
and b
i
= (f(x
i
) h(x
i
))/n
1/p
,
_
n

i=1
|g(x
i
) h(x
i
)|
p
/n
_
1/p

_
n

i=1
|g(x
i
) f(x
i
)|
p
/n
_
1/p
+
_
n

i=1
|f(x
i
) h(x
i
)|
p
/n
_
1/p
.
Taking limits as n , we deduce that
_
_

|g(x) h(x)|
p
dx
_
1/p
= lim
n
_
n

i=1
|g(x
i
) h(x
i
)|
p
/n
_
1/p
lim
n
_
n

i=1
|f(x
i
) g(x
i
)|
p
/n
_
1/p
+ lim
n
_
n

i=1
|f(x
i
) h(x
i
)|
p
/n
_
1/p
=
_
_

|f(x) g(x)|
p
dx

_
1/p
+
_
_

|f(x) h(x)|
p
dx

_
1/p
,
3
so that d(g, h) d(f, g) + d(f, h), as required.
The metric d in this example is called the L
p
metric on C[, ]. In both this example
and the previous one we can put p = , obtaining a metric on the set of bounded
innite sequences given by
d

(a, b) = sup
k
|a
k
b
k
|
and a metric on C[, ] given by
d

(f, g) = sup
x[,]
|f(x) g(x)|.
It is easy to check by direct verication of the axioms that these are metrics.
The following proof was not done in lectures, and is added here just for interests
sake. We show that lim
p
d
p
(f, g) = d

(f, g) for all f, g C[, ], provided that < .

Let f, g C[, ] and let
S = d

(f, g) = sup
x[,]
|f(x) g(x)|.
Let > 0 be arbitrary. There exists x
0
[, ] such that |f(x
0
) g(x
0
)| > S

2
, and
by continuity of |f(x) g(x)| it follows that |f(x) g(x)| > S

2
for all x close enough
to x
0
. That is, there exist x
1
, x
2
with x
1
< x
2
such that |f(x) g(x)| > S

2
for all x [x
1
, x
2
]. Dene (x) on [, ] by
(x) =
_
S

2
if x [x
1
, x
2
]
0 otherwise,
and observe that (x) |f(x) g(x)| S for all x [, ]. A typical example of this
situation is illustrated in the diagram below, in which the graphs of (x), |f(x) g(x)|
and the constant function S on the interval [, ] are shown.
x
1
x
2

x
S
S /2
y
For all p 1 we have (x)
p
|f(x) g(x)|
p
S
p
for all x [, ], and so
(x
2
x
1
)(S

2
)
p
=
_

(x)
p
dx

|f(x) g(x)|
p
dx

S
p
dx = ( )S
p
.
4
Taking p-th roots gives
(x
2
x
1
)
1/p
(S

2
)
_
_

|f(x) g(x)|
p
dx
_
1/p
( )
1/p
S,
and since the limit as p of the pth root of any positive number is 1, the left hand
side approaches S

2
and the right hand side approaches S as p . So there exists
P R such that (x
2
x
1
)
1/p
(S

2
) > S and ( )
1/p
S < S + for all p > P.
Hence for all p > P,
S <
_
_

|f(x) g(x)|
p
dx
_
1/p
< S + ,
and since was arbitrary, this shows that lim
p
_
_

|f(x) g(x)|
p
dx
_
1/p
= S, as claimed.
Let (X, d) be any metric space and let Y be an arbitrary subset of X. We can dene
a distance function d
Y
: Y Y R by
d
Y
(a, b) = d(a, b) for all a, b Y .
In other words, d
Y
is the restriction to Y Y of the distance function d: XX R. It is
clear that d
Y
satises the distance function axioms, since d does; thus (Y, d
Y
) is a metric
space. We say that (Y, d
Y
) is a subspace of (X, d), and d
Y
is called the metric induced on
Y by the metric d on X.
Topology
Let (X, d) be a metric space. Recall that if A X then a point a A is called an
interior point of A if there exists > 0 such that B(a, ) A. The set of all interior
points of A is Int(A), the interior of A. It is clear that Int(A) A for all subsets A of X.
Lemma. Let A be an arbitrary subset of the metric space X. Then Int(Int(A)) = Int(A).
Proof. The inclusion Int(Int(A)) Int(A) is immediate from the comment preceding
the statement of the lemma; so we just have to prove the reverse inclusion.
Let a Int(A) be arbitrary. Choose > 0 such that B(a, ) A. We shall show that
in fact B(a, ) Int(A). For, suppose that b B(a, ). Since B(a, ) is open, there exists
> 0 such that B(b, ) B(a, ). Thus B(b, ) A, which shows that b Int(A). As
this holds for all b B(a, ), we have shown that B(a, ) Int(A), as claimed. However,
this statement says that a is an interior point of Int(A), and since a was originally chosen
as an arbitrary point of Int(A), we have shown that all points of Int(A) are interior points,
as required.
5