Sei sulla pagina 1di 11

f

1
(x
0
), f
2
(x
0
), f
3
(x
0
), . . ., and to dene f(x
0
) as the limit of these values. But we dont
even know that this sequence converges for any value of x
0
! However, we can use the
fact that f
1
, f
2
, . . . is a Cauchy sequence with respect to the metric d

to see that
f
1
(x
0
), f
2
(x
0
), . . . is a Cauchy sequence of real numbers:
|f
m
(x
0
) f
n
(x
0
)| 6 sup{|f
m
(t) f
n
(t)| : t [0, 1]} = d

(f
n
, f
m
).
We know that f
1
, f
2
, . . . is Cauchy, so that for > 0 we can nd N such that d

(f
n
, f
m
) <
for n, m > N. It then follows that |f
m
(x
0
) f
n
(x
0
)| < for m, n > N, and therefore
(f
n
(x
0
)) is a Cauchy sequence in R. But R is complete, so (f
n
(x
0
)) converges to a limit
that we call f(x
0
).
At this point, we have dened f(x
0
) for all x
0
[0, 1], and so we have dened the
function f on all of [0, 1]. But we still need to prove that f is continuous, and that f
n
f
in (C[0, 1], d

).
First, we claim that, given > 0, we can nd N such that |f
n
(x) f(x)| < for all
n > N and x [0, 1]. To prove this, choose N large enough that d

(f
n
, f
m
) <

2
for all
n, m > N. Then for n > N and x [0, 1] we have
|f
n
(x) f(x)| 6 |f
n
(x) f
m
(x)| + |f
m
(x) f(x)|
<

2
+ |f
m
(x) f(x)|
and |f
m
(x) f(x)| 0 as m , so that by a familiar rule for limits and inequalities
we have |f
n
(x) f(x)| 6

2
< as claimed.
Now we can prove that f is continuous. Let x [0, 1] and let > 0. By the claim
above, we can choose n large enough that |f
n
(x) f(x)| <

3
for all x, and since f
n
is
continuous at x we can nd > 0 so that |f
n
(x) f
n
(y)| <

3
whenever |x y| < . Then
for |x y| < we have
|f(x) f(y)| 6 |f(x) f
n
(x)| + |f
n
(x) f
n
(y)| + |f
n
(y) f(y)| <

3
+

3
+

3
= .
Finally we must show that d

(f
n
, f) 0 as n . But now that we know that
f is continuous, this is exactly what the earlier claim says.
As you see, all our main examples of metric spaces, R, R
n
, C[0, 1], are all complete
(as long as we use the right metric!). This will be very useful indeed, when we study the
main application of general theory to convergence of iterations, in section 7.
7 The Contraction Mapping Principle.
Remember the original aim of our course. We want to be able to solve equations like
f(x) = x, i.e. to nd xed points of f, by picking an initial value x
0
and iterating
x
n+1
= f(x
n
)
to obtain a sequence (x
n
). If this sequence has a limit x, then x is indeed a xed point of
f, because
f(x) = f( lim
n
x
n
) = lim
n
f(x
n
) = lim
n
x
n+1
= x.
MAS331 38 Autumn 2011-12
The question is whether we can nd conditions to tell us when this idea will actually
work! In this chapter well prove a powerful theorem that tells us when this happens in
general, and well use it to prove lots of useful and rather general results. The idea is
illustrated by the following example, which you saw in MAS207:
Lemma 7.1. Given a continuous function f : [a, b] [a, b], there is a point c [a, b] such
that f(c) = c.
Proof. This is just a corollary of the Intermediate Value Theorem. Consider the function
g(x) = f(x) x. Since a 6 f(a) (since f takes values in [a, b]), g(a) > 0. Similarly, b >
f(b), so g(b) 6 0. And as f(x) and x are continuous, so is g(x). The Intermediate Value
Theorem tells us that there is some point c [a, b] such that g(c) = 0, i.e., f(c) = c.
Here, we have a map from a set to itself, and the lemma tells us that we can nd a
xed point. Our aim in this chapter is to prove that many maps on more general metric
spaces have xed points, and to show that we can nd these xed points using iterative
procedures.
7.1 Contractions.
What we want is a condition on a function f which guarantees that the iterative method
will lead to a solution of x = f(x).
If x is a solution of x = f(x), and x
1
is any guess, then wed like x
2
= f(x
1
) to be
closer to the solution than x
1
was. That is,
|f(x
1
) x| < |x
1
x|,
or equivalently,
|f(x
1
) f(x)| < |x
1
x
0
|.
Since we dont actually know in advance what x is, we will have to insist on something
like
|f(x) f(y)| < |x y|
for all x, y distinct (i.e. for all x 6= y).
Example 7.2. The cosine function satises this property. If x and y are distinct, then:
| cos x cos y| =

2 sin
_
xy
2
_
sin
_
x+y
2
_

6 2

sin
_
xy
2
_

< 2
|xy|
2
= |x y|.
Indeed, if we iterate x
n+1
= cos(x
n
), starting with x
0
= 1, then we nd that from x
20
onwards, all the x
n
s have value 0.739 to three decimal places. It seems that in this case
the iteration does converge to a solution of x = cos x.
Example 7.3. There are functions f(x) which do satisfy |f(x) f(y)| < |x y| for all
distinct x, y, but which do not have a xed point. On the example sheet, you will show
that if f : [1, ) [1, ) is dened by f(x) = x+
1
x
, then |f(x)f(y)| < |xy| whenever
x 6= y, but f(x) clearly has no xed point.
MAS331 39 Autumn 2011-12
In the last example, f(x) had no xed points even though it satised |f(x)f(y)| <
|xy| for x 6= y. Perhaps the problem in this case is that |f(x) f(y)| can get arbitrarily
close to |x y| when x and y are very large. Lets modify the example to x this, and
then see what happens:
Example 7.4. Fix a number k < 1 and dene f : [1, ) [1, ) by f(x) = k(x +
1
x
).
Then |f(x)f(y)| 6 k|xy| for all x, y. Since k < 1, this is actually a stronger condition
than just saying |f(x) f(y)| < |x y| for x 6= y.
Lets set k =
3
4
and iterate x
n+1
= f(x
n
), beginning with x
0
= 1. Then to three
decimal places the rst terms in the iteration are as follows.
x
1
= 1.500
x
2
= 1.625
x
3
= 1.680
x
4
= 1.707
x
5
= 1.719
x
6
= 1.726
x
7
= 1.729
x
8
= 1.730
x
9
= 1.731
Again, it seems here that the iteration is converging to a value around 1.731. Indeed,
we can solve x =
3
4
(x +
1
x
) algebaically by rearranging the equation to get a quadratic;
the only solution in [1, ) is x =

3, which is 1.73205080 . . ., which agrees to 2 decimal


places with the value we found after only 9 iterations. So in this case the method works!
Weve seen that the condition
|f(x) f(y)| < |x y|
for x 6= y is not enough to guarantee convergence of the iteration. Perhaps the stronger
condition that there should exist a k < 1 such that
|f(x) f(y)| 6 k|x y|
for x 6= y is enough, as in the last example.
Denition 7.5. Let f : X X be a function on the metric space (X, d). Then f is a
contraction if there exists a constant 0 6 k < 1 such that
d(f(x), f(y)) 6 kd(x, y).
Example 7.6. Here are some examples that we already know are or arent con-
tractions:
1. cos: [

2
,

2
] [

2
,

2
] is a contraction.
2. For 0 6 k < 1, the function f : [1, ) [1, ), f(x) = k(x +
1
x
) is a contraction.
3. The function f : [1, ) [1, ), f(x) = (x +
1
x
), is not a contraction.
It looks like any sequence dened by iterating a contraction is at least going to
satisfy the condition that the points get closer together. But weve already seen that
this might not be enough to guarantee convergence if our metric space is not complete,
then it will have holes, and even though terms of the sequence get closer and closer, the
sequence need not converge.
MAS331 40 Autumn 2011-12
7.2 The Contraction Mapping Principle.
By now, you should have the feeling that to guarantee convergence of an iterative proce-
dure, we should be working with contractions on a complete metric space. The following
theorem, the main theorem of the course, justies this:
Theorem 7.7 (Contraction Mapping Principle). Let f : X X be a contraction of the
complete metric space (X, d). Then f has a unique xed point.
Proof. Choose x
0
X, and dene (x
n
) inductively by iterating
x
n+1
= f(x
n
).
We will prove that (x
n
) is a Cauchy sequence as X is complete, the sequence will
converge.
First well bound d(x
n
, x
n+1
). Note that each d(x
i
, x
i+1
) = d(f(x
i1
), f(x
i
)) 6
kd(x
i1
, x
i
), so that
d(x
n
, x
n+1
) 6 kd(x
n
, x
n+1
) 6 k
2
d(x
n1
, x
n
) 6 6 k
n
d(x
1
, x
0
).
Now well suppose that m > n and nd a bound for d(x
n
, x
m
):
d(x
n
, x
m
) 6 d(x
n
, x
n+1
) + d(x
n+1
, x
n+2
) + + d(x
m1
, x
m
)
6 [k
n
+ k
n+1
+ + k
m1
]d(x
0
, x
1
)
6 [k
n
+ k
n+1
+ ]d(x
0
, x
1
)
=
k
n
1 k
d(x
0
, x
1
).
Since k < 1, it follows that
k
n
1k
0 as n . It follows that (x
n
) is a Cauchy sequence,
and since X is complete, it follows that x
n
x for some limit x X.
Lets now see that x is a xed point. Note that, since the sequence x
0
, x
1
, . . .
converges to x, then the sequence x
1
, x
2
, . . . also converges to x. Therefore:
f(x) = f( lim
n
x
n
) = lim
n
f(x
n
) = lim
n
x
n+1
= x,
so that f(x) = x as required.
Now well show that the xed point is unique. Suppose that both x and x
0
are xed
points. Then
d(x, x
0
) = d(f(x), f(x
0
)) 6 kd(x, x
0
).
As k < 1, the only way this can happen is if d(x, x
0
) = 0, and so x = x
0
.
We can look carefully at the proof of the theorem to explicitly nd a sequence that
converges to the xed point, and more importantly, to tell us how quickly that sequence
converges.
Proposition 7.8. Let f : X X be a contraction of the complete metric space (X, d), so
that d(f(x), f(y)) 6 kd(x, y) for some 0 6 k < 1, and let x
0
be any point of X. Then the
sequence (x
n
) dened by x
n+1
= f(x
n
) converges to the unique xed point x. Furthermore,
for any n we have
d(x
n
, x) 6
k
n
1 k
d(x
0
, f(x
0
)),
Thus we get a bound for the distance of x
n
from the limit x in terms of x
0
.
MAS331 41 Autumn 2011-12
Proof. The rst part is taken directly from the proof it is how we dened the xed
point x in the rst place. For the second part, we take the line d(x
m
, x
n
) 6
k
n
1k
d(x
0
, x
1
)
and let m to get
d(x
n
, x) 6
k
n
1 k
d(x
0
, x
1
).
Example 7.9. Question:
Dene a map
f : R
3
R
3
(x, y, z) 7

sin y
4
,
sin z
3
+ 1,
sin x
5
+ 2

Show that there exists a unique xed point, and nd an approximation to it,
explaining why your answer is within 0.01 of the xed point (measuring with
the Euclidean distance).
Well prove the rst part by using the Contraction Mapping Principle. First, R
3
is
complete. Second, we must check that the map f is a contraction. We need to nd a
k < 1 and then show that, for any two points (x, y, z) and (x
0
, y
0
, z
0
) in our space, we have
d(f(x, y, z), f(x
0
, y
0
, z
0
)) 6 kd((x, y, z), (x
0
, y
0
, z
0
)). Well use the fact that | sin x sin y| 6
|x y| for all x, y R. Now
d(f(x, y, z), f(x
0
, y
0
, z
0
)) = d

sin y
4
,
sin z
3
+ 1,
sin x
5
+ 2

sin y
0
4
,
sin z
0
3
+ 1,
sin x
0
5
+ 2

sin y
4

sin y
0
4

2
+

sin z
3

sin z
0
3

2
+

sin x
5

sin x
0
5

2
6

sin y sin y
0
3

2
+

sin z sin z
0
3

2
+

sin x sin x
0
3

2
6
1
3
_
(y y
0
)
2
+ (z z
0
)
2
+ (x x
0
)
2
=
1
3
d((x, y, z), (x
0
, y
0
, z
0
)).
It follows that f is a contraction on R
3
with contraction factor k =
1
3
.
So now we know that all of the hypotheses of the Contraction Mapping Principle
are satised. Thus the conclusion holds: f does have a unique xed point.
The proof of the Contraction Mapping Principle tells us that we can nd this
xed point (or at least, an approximation to it) by choosing any starting point, e.g.,
(x
0
, y
0
, z
0
) = (0, 0, 0) and iterating f. Lets do a few steps of this:
(x
1
, y
1
, z
1
) = (0.0000, 1.0000, 2.0000)
(x
2
, y
2
, z
2
) = (0.2104, 1.3031, 2.0000)
(x
3
, y
3
, z
3
) = (0.2411, 1.3031, 2.0418)
(x
4
, y
4
, z
4
) = (0.2411, 1.2970, 2.0478)
(x
5
, y
5
, z
5
) = (0.2407, 1.2961, 2.0478)
(x
6
, y
6
, z
6
) = (0.2406, 1.2961, 2.0477)
(x
7
, y
7
, z
7
) = (0.2406, 1.2961, 2.0477)
MAS331 42 Autumn 2011-12
and so on. It looks like we should be within 0.01 of the answer. We can use Proposition 7.8
to prove this, because it tells us that
d(x
n
, x) 6
k
n
1 k
d(x
0
, x
1
).
We can work out
d(x
0
, x
1
) = d((x
0
, y
0
, z
0
), (x
1
, y
1
, z
1
)) =
_
(0 0)
2
+ (1 0)
2
+ (2 0)
2
=

5 = 2.236 . . . .
Since k =
1
3
, to get
k
n
1k
d((x
0
, y
0
, z
0
), (x
1
, y
1
, z
1
)) to be less than 0.01, we need
(
1
3
)
n
1
1
3
.

5 < 0.01.
This is solved by
1
3
n
<
2
3

0.01

5
,
or, rearranging and evaluating,
3
n
> 335.41 . . . ,
which means that n > 6 should be enough. And indeed, the point (x
6
, y
6
, z
6
) is actually
correct to even greater accuracy (its within 0.001 of the xed point).
7.3 Applications.
Its rather amazing that such a simple theorem should have so many applications. Well
just give a few. Heres a simple application to linear algebra:
Theorem 7.10. Let M be a real n n matrix, all of whose entries are less than
1
n
in
modulus. Then I M is invertible.
Proof. Consider the metric space R
n
with the non-standard metric
d

(x, y) = max
16i6n
(|x
i
y
i
|).
The matrix M can be regarded as a map R
n
R
n
. Well explain that this map is a
contraction. Indeed, take x, y R
n
, and let x
0
= Mx, y
0
= My. Then
_
_
_
_
_
x
0
1
x
0
2
.
.
.
x
0
n
_
_
_
_
_
=
_
_
_
_
_
m
11
m
12
m
1n
m
21
m
22
m
2n
.
.
.
.
.
.
.
.
.
.
.
.
m
n1
m
n2
m
nn
_
_
_
_
_
_
_
_
_
_
x
1
x
2
.
.
.
x
n
_
_
_
_
_
and similarly for y
0
and y. Now it is easy to see that
d

(x
0
, y
0
) 6 n max
16i,j6n
|m
ij
| d

(x, y).
Since |m
ij
| <
1
n
for each entry m
ij
of M, it follows that max
16i,j6n
|m
ij
| <
1
n
, and so M
is a contraction. The Contraction Mapping Principle says that there is a unique point
MAS331 43 Autumn 2011-12
x with Mx = x, and clearly this must be x = (0, 0, . . . , 0). So Mx = x if and only if
x = (0, 0, . . . , 0). Thus
x ker(I M) (I M)x = 0
x = Mx
x = 0,
so the kernel of multiplication by (I M) is trivial and therefore I M is invertible.
Now lets return to consider real functions. Well give an easy criterion for a dier-
entiable function to be a contraction.
Theorem 7.11 (Dierential Criterion). Let f : [a, b] [a, b] be dierentiable. Then f is
a contraction of [a, b] if and only if there exists k < 1 with |f
0
(x)| 6 k for all x (a, b).
The result also holds for intervals [a, ), (, b] and for R as a whole.
Proof. Suppose f is a contraction. Then x x (a, b), and let x + x [a, b]. We have
|f(x + x) f(x)| 6 k|(x + x) x| = k|x|,
and so |
f(x+x)f(x)
x
| 6 k. If we let x 0, then this inequality becomes |f
0
(x)| 6 k as
required.
Conversely, suppose that |f
0
(x)| 6 k for all x [a, b], and let x, y [a, b]. By the
Mean Value Theorem, there exists c between x and y such that
f(x) f(y)
x y
= f
0
(c);
by the bound |f
0
(c)| 6 k, we conclude that

f(x) f(y)
x y

6 k,
and we deduce that f is a contraction.
Example 7.12. As an example of how to use this kind of result, lets think about the
function f : [1, 1] [1, 1] dened by f(x) =
1
6
(x
3
+ x
2
+ 1). Then
|f
0
(x)| =
3x
2
+ 2x
6
6
5
6
for all x (1, 1). It follows that f is a contraction of [1, 1]. We conclude that there
is a unique value of x satisfying x = f(x), or, rearranging, x
3
+ x
2
6x + 1 = 0, in the
interval [1, 1].
Example 7.13. Lets check that the function f : [1, ) [1, ) dened by f(x) =
1
2
(x +
2
x
) is a contraction. This is a little ddly using the original denition but since
f
0
(x) =
1
2

1
x
2
, we can see that |f
0
(x)| 6
1
2
when x (1, ). Theorem 7.11 now tells us
that f is a contraction.
The next result will be crucial for our main application to the existence of solutions
to dierential equations.
MAS331 44 Autumn 2011-12
Theorem 7.14. Let X be a complete metric space, and let f : X X have the property
that, for some N, the iterate f
N
is a contraction of X. Then f has a unique xed point,
and the usual iterative procedure converges to this xed point for any choice of starting
point.
Proof. Since f
N
is a contraction, it has a unique xed point x, i.e., f
N
(x) = x. Apply f
to this relation, and we see that f
N+1
(x) = f(x), i.e., f
N
(f(x)) = f(x), so f(x) is also a
xed point of f
N
. Since f
N
is a contraction, it only has one xed point, so that x = f(x).
Given a starting point x
0
, and the sequence x
n
= f
n
(x
0
), the sequence
x
0
, x
1
, x
2
, . . . , x
N
, x
N+1
, x
N+2
, . . . , x
2N
, x
2N+1
, x
2N+2
, . . . ,
is made up of subsequences x
0
, x
N
, x
2N
, . . . and x
1
, x
N+1
, x
2N+1
, . . ., etcetera. But these
subsequences are x
0
, f
N
(x
0
), f
2N
(x
0
), . . . and x
1
, f
N
(x
1
), f
2N
(x
1
), . . ., etcetera, all of which
converge to the unique xed point of f (i.e., of f
N
).
7.4 Dierential equations.
We have already seen applications of the Contraction Mapping Principle to real equations,
simultaneous equations and linear algebra. One of the examples of iteration in the rst
chapter was to an integral equation, and perhaps surprisingly, we can use the Contraction
Mapping Principle to prove that many dierential equations have a solution.
Before we prove a general result, well motivate it with an example.
Example 7.15. There is a solution of the dierential equation
df
dx
= (f(x) + x)x (0 6 x 6 1), f(0) = 0.
We rst observe that f is a solution of this dierential equation, together with its
initial condition f(0) = 0, if and only if it is a solution of the integral equation
f(x) =
x
_
0
(f(u) + u)u du.
Indeed, we can dierentiate this integral equation to get the original dierential equation,
and we can substitute in x = 0 to get f(0) = 0.
Now we dene F : C[0, 1] C[0, 1] by
F(f)(x) =
x
_
0
(f(u) + u)u du.
Note that the integral equation is now just F(f) = f. This means that the solutions of
the original dierential equation are the xed points of F.
To apply the Contraction Mapping Principle, we need to be working in a complete
MAS331 45 Autumn 2011-12
space, so C[0, 1] must be given the supremum metric. Lets prove that F is a contraction.
d

(F(f
1
), F(f
2
)) = sup
t[0,1]
|F(f
1
)(t) F(f
2
)(t)|
= sup
t[0,1]

t
_
0
(f
1
(u) + u)u du
t
_
0
(f
2
(u) + u)u du

= sup
t[0,1]

t
_
0
(f
1
(u) f
2
(u))u du

6 sup
t[0,1]
t
_
0
|f
1
(u) f
2
(u)|u du
6
1
_
0
|f
1
(u) f
2
(u)|u du
6 sup
t[0,1]
|f
1
(t) f
2
(t)|
1
_
0
u du
= sup
t[0,1]
|f
1
(t) f
2
(t)|/2
= d

(f
1
, f
2
)/2.
So the dierential equation has a unique solution. To nd it, we can start with any initial
guess, f
0
(x) = 0 for all x [0, 1], say, and then we can iterate the function F to get the
limit. Thus f
1
= F(f
0
) is
f
1
(x) = F(f
0
)(x) =
x
_
0
(f
0
(u) + u)u du =
x
_
0
u
2
du =
x
3
3
.
Similarly, set f
2
= F(f
1
):
f
2
(x) = F(f
1
) =
x
_
0
(f
1
(u) + u)u du =
x
_
0

u
4
3
+ u
2

du =
x
5
15
+
x
3
3
.
We can continue in this way, and nd the series solution
f(x) =
x
3
3
+
x
5
3.5
+
x
7
3.5.7
+
x
9
3.5.7.9
+ ,
which we can then check directly.
In fact, this method of proof works fairly generally, although we may need to use
some iterate of F in general.
Theorem 7.16. Consider the dierential equation
df
dx
= (f(x), x), a 6 x 6 b, f(a) = ,
MAS331 46 Autumn 2011-12
where (y, x): R [a, b] R is a continuous function of y and x. Suppose that there
exists a real number L such that
|(y
1
, x) (y
2
, x)| 6 L|y
1
y
2
|
for all y
1
, y
2
R, and x [a, b]. Then the dierential equation has a unique solution.
Proof. As in the example, we dene F : C[a, b] C[a, b] by
(F(f))(x) = +
x
_
a
(f(u), u) du.
If f is a solution of f = F(f), then f is a solution of the original dierential equation,
and the initial condition is visibly satised.
For two functions f
1
, f
2
C[a, b], and x [a, b], then
|(F(f
1
))(x) (F(f
2
))(x)| =

_
_
+
x
_
a
(f
1
(u), u) du
_
_

_
_
+
x
_
a
(f
2
(u), u) du
_
_

x
_
a
[(f
1
(u), u) (f
2
(u), u)] du

6
x
_
a
|(f
1
(u), u) (f
2
(u), u)| du
6
x
_
a
L|f
1
(u) f
2
(u)| du
6
x
_
a
L d

(f
1
, f
2
) du
= L d

(f
1
, f
2
) (x a).
Similarly, with F(f
1
), F(f
2
) replacing f
1
, f
2
:

F
2
(f
1
)(x) F
2
(f
2
)(x)

6
x
_
a
L|F(f
1
)(u) F(f
2
)(u)| du,
and so

F
2
(f
1
)(x) F
2
(f
2
)(x)

6
x
_
a
L|F(f
1
)(u) F(f
2
)(u)| du
6
x
_
a
L
2
d

(f
1
, f
2
)(u a) du
= L
2
d

(f
1
, f
2
)
(x a)
2
2
.
MAS331 47 Autumn 2011-12
Continuing in this way, we get the general result that
|F
N
(f
1
)(x) F
N
(f
2
)(x)| 6 L
N
d

(f
1
, f
2
)
(x a)
N
N!
6 L
N
d

(f
1
, f
2
)
(b a)
N
N!
since x [a, b]. So
d

(F
N
(f
1
), F
N
(f
2
)) 6
L
N
(b a)
N
N!
d

(f
1
, f
2
).
As N , the coecients
L
N
(ba)
N
N!
0. Eventually, then, this coecient is less than 1,
and for this value of N, we conclude that F
N
is a contraction, and the theorem follows.
It turns out that the condition on is almost automatic: one needs little more than
that is (partially) dierentiable with respect to y. The theorem can be extended to
simultaneous dierential equations and to higher-order dierential equations. For reasons
of time, we will omit these applications.
8 Compactness.
This chapter is not directly relevant to our aim of studying conditions for convergence
of iterative processes. Nevertheless it will, if anything, be more relevant to later courses
than some of the other denitions introduced in the course.
Think back to our proof of Theorem 6.10, that R is complete: a Cauchy sequence is
bounded, so it has a convergent subsequence, and so the whole sequence converges. The
key step used the Bolzano-Weierstrass Theorem 6.9, which stated that any sequence in
the interval [a, b] has a convergent subsequence with limit in the set. Generalisations of
this turn out to be so useful that we will isolate this property.
8.1 Compactness Using Subsequences.
Denition 8.1. Let A X be a subset of a metric space. We say that A is compact if
every sequence in A has a subsequence that converges to a point of A.
Example 8.2. All closed intervals of the form [a, b] are compact, by the Bolzano-Weierstrass
Theorem 6.9. However, the real numbers R, or any interval which is not bounded such as
[a, ) or (, b], are not compact. For example, in R there is the sequence 0, 1, 2, 3, 4, . . .
which has no convergent subsequence.
Heres a way to nd many compact sets:
Lemma 8.3. Let A X be a closed subset of a compact space X. Then A is compact.
Proof. Take a sequence in A. We may regard it as a sequence in X and, as X is compact,
there is a subsequence converging to a limit a X. But a X is the limit of a sequence
in the closed subset A, so a A, and so A is compact.
Now we are going to prove several properties of compact sets, in order to get a
feel for which sorts of set are compact. First, weve already noted that complete sets
are closed, so that the property of being complete is stronger than the property of being
closed. The property of being compact is stronger still.
MAS331 48 Autumn 2011-12

Potrebbero piacerti anche