This action might not be possible to undo. Are you sure you want to continue?
https://www.scribd.com/doc/60982226/ExamplesofTopologicalSpaces
09/23/2013
text
original
NEIL STRICKLAND
This is a list of examples of topological spaces. I am distributing it for a variety of reasons.
First and foremost, I want to persuade you that there are good reasons to study topology; it is a
powerful tool in almost every ﬁeld of mathematics. I want also to drive home the disparate nature
of the examples to which the theory applies. This means, on the one hand, that we achieve a
great economy of eﬀort, because we need only give one proof and it will apply in many contexts.
On the other hand, we need to be careful and rigorous, because our arguments are supposed to
be valid in situations far removed from our intuition.
Another reason for distributing these examples is to help you to understand the general theory.
I will always try to give examples of abstract theorems, and there will be questions about examples
on the problem sets (possibly drawn from this list) but it is always worthwhile to analyse further
cases on your own initiative.
I have tried to include examples from a wide range of ﬁelds of mathematics. This means that
there will probably be a number of examples which you do not have the neccessary background
to understand. Do not worry about this. Nothing in this list will be examinable unless I actually
lecture on it.
1. Euclidean Examples
The most basic example is the space R with the order topology. The open sets are the sets
U ⊂ R such that every point in U lies in an open interval wholly contained in U; in symbols
x ∈ U ⇒ ∃a, b ∈ R x ∈ (a, b) ⊆ U
This topology is also deﬁned by the metric
d(x, y) = x −y
A subset is compact if and only if it is bounded and closed. A subset A is connected if and
only if it is convex, i.e.
a < b < c and a, c ∈ A ⇒ b ∈ A
Almost as basic is the space R
n
with the product topology. There are many diﬀerent metrics
which induce this topology. For example, we can consider three norms on R
n
:
v
1
=
¸
k
v
k
 (1)
v
2
=
¸
k
v
k

2
(2)
v
∞
= max
k
v
k
 (3)
We then deﬁne d
1
(u, v) = u − v
1
and so on. This gives three diﬀerent metrics d
1
, d
2
and d
∞
.
However, they all deﬁne the same topology.
In fact, it is an interesting theorem that every norm whatsoever induces the product topology.
To explain a little: a function v of vectors v is a norm if:
v ≥ 0 (4)
v = 0 ⇔ v = 0 (5)
av = av (6)
u +v ≤ u +v (7)
1
2 NEIL STRICKLAND
Given a norm, we deﬁne a metric by d(u, v) = u − v. This metric induces a topology, and the
claim is that this is always the same as the product topology, no matter what norm we start with.
2. Examples from Functional Analysis
The examples in this section are all spaces of functions with various diﬀerent topologies. They
are important for analysing the convergence of Fourier series, the existence and uniqueness of
solutions to diﬀerential equations, the spectral theory of operators in quantum mechanics, and
many other things.
2.1. Continuous Functions.
C[0, 1] = {continuous functions f : [0, 1] −→R}
This is a normed space with the following norm:
f
∞
= sup{f(x)  0 ≤ x ≤ 1}
This is ﬁnite because a continuous real valued function on a compact space is bounded. From it
we derive a metric:
d(f, g) = f −g
A sequence of functions (f
n
)
∞
n=0
converges to a function f with respect to this metric if and only
if (in the usual language of real analysis) it converges uniformly. It follows (using the Weierstrass
Mtest) that C[0, 1] is complete as a metric space.
The polynomial functions from [0, 1] to R (such as f(x) = 5x
2
+ 6) form a subspace P[0, 1] of
C[0, 1]. It is a dense subspace, by the StoneWeierstrass theorem.
An interesting example of a continuous function from C[0, 1] to R is given by integration:
I : C[0, 1] −→R I(f) =
1
0
f(x)dx
Another is the evaluation function ˆ a for a ∈ [0, 1]:
ˆ a: C[0, 1] −→R ˆ a(f) = f(a)
This idea of regarding f(a) as a function of f rather than of a is certainly curious at ﬁrst sight,
but it turns out to be strikingly useful.
2.2. SquareIntegrable Functions.
L
2
(R/Z) = {f : R −→C such that f(x + 1) = f(x) and
1
0
f(x)
2
dx < ∞}
This space is the natural home of the theory of Fourier series. To make the deﬁnition of L
2
(R/Z)
precise, we need to mention that the integration sign means the Lebesgue integral, which is studied
in courses on measure theory. However, this is merely a technicality; the Lebesgue integral agrees
with any more elementary deﬁnition when the latter makes sense.
As in the case of C[0, 1], we deﬁne a norm and thence a metric:
f
2
=
1
0
f(x)
2
dx
d(f, g) = f −g
2
This is not quite a metric because it is possible to have d(f, g) = 0 even when f = g. For example,
we could have g = 0 and
f(x) =
1 if x ∈ Z
0 otherwise
This is again just an annoying technicality, which can be suppressed.
The basic examples of elements of L
2
(R/Z) are the functions
e
n
(x) = exp(2nπix)
EXAMPLES OF TOPOLOGICAL SPACES 3
and the basic example of a continuous function from L
2
(R/Z) to C is the Fouriercoeﬃcient
function
C
n
(f) =
1
0
f(x)e
n
(x)dx
The fundamental theorem about Fourier series is that for any f ∈ L
2
,
f =
¸
n∈Z
C
n
(f)e
n
where the sum converges with respect to the metric just described.
In fact, still more is true, as described in the next example.
2.3. SquareSummable Sequences.
L
2
(Z) = {series c = (c
n
)
∞
n=−∞
such that
¸
n
c
n

2
< ∞}
c
2
=
¸
n
c
n

2
d(b, c) = b −c
There are continuous maps
L
2
(Z)
F
−→ L
2
(R/Z)
C
−→ L
2
(Z)
deﬁned by
F(c) =
¸
n
c
n
e
n
C(f) = (C
n
(f))
∞
n=−∞
=
1
0
f(x)e
n
(x)dx
∞
n=−∞
These are in fact mutually inverse isometric isomorphisms:
FC(f) = f CF(c) = c
d(C(f), C(g)) = d(f, g) d(F(b), F(c)) = d(b, c)
This means that the two L
2
spaces can be identiﬁed in a very strong sense.
2.4. Smooth Functions. The function spaces described above are good for studying things like
integration, and diﬀerential equations can often be converted into integral equations by cunning
means; but to study diﬀerentiation directly we need a diﬀerent kind of space.
C
∞
(R) = { inﬁnitely diﬀerentiable functions f : R −→R}
Given a compact subset K ⊂ R, we let R
K
(f) denote the restriction of f to K:
R
K
: C
∞
(R) −→ C(K) R
K
(f) = f
K
We also write D for the function from C
∞
(R) to itself sending a function to its derivative:
D: C
∞
(R) −→ C
∞
(R) D(f) = f
We give C
∞
(R) the coarsest possible topology such that the maps R
K
(for all compact sets K)
and D are continuous. This topology is generated by a rather ugly metric, as follows:
P
n
(f) = min(1, sup{f(x) such that −n ≤ x ≤ n})
d(f) =
∞
¸
n=0
∞
¸
m=0
2
−m−n
P
n
(D
m
f)
d(f, g) = d(f −g)
Fortunately, one can usually use the characterisation of the topology in terms of R
K
and D and
ignore the metric.
Understanding this space and certain closely related spaces is the ﬁrst step towards the theory
of distributions, which is the proper home of the Dirac delta function and similar beasts.
4 NEIL STRICKLAND
3. Examples from Complex Analysis
3.1. The Riemann Sphere. The Riemann sphere C
∞
is the onepoint compactiﬁcation C∪{∞}.
The open subsets are the open sets in C together with the sets U ∪ {∞} such that U is an open
subset of C whose complement in C is compact.
If p and q are complex polynomials (not both zero) then the expression r(z) = p(z)/q(z) can
be made sense of as a continuous function from C
∞
to itself, although a certain amount of work
needs to be done to justify this. This is much the most natural context in which to think about
such functions.
3.2. Spaces of Analytic Functions. If U is a connected open subset of C, we let A(U) denote
the space of analytic functions on U. If K ⊂ U is compact and f ∈ A(U) then we write
f
K
= max{f(z) such that z ∈ K}
B(f, K, ) = {g ∈ A(U) such that f −g
K
< }
The sets B(f, K, ) form a basis for a topology on A(U), called the topology of locally uniform
convergence.
This topology has remarkably good properties, much stronger than the corresponding ones
for the space of merely continuous functions on U. Firstly, it follows from the Cauchy integral
formulae that the diﬀerentiation function is continuous:
D: A(U) −→ A(U) D(f) = f
If Γ is a simple closed contour whose interior is contained in U and F is the set {f ∈
A(U)  f has no zeros on Γ} then we can deﬁne a function
v
Γ
: F −→N
v
Γ
(f) = number of zeros of f inside Γ
Here zeros are counted by multiplicity in the usual way, so that f(z) = (z −1)
2
counts as having
two zeros at z = 1. This function turns out to be continuous, and thus (as N is discrete) constant
on the connected components of F.
Using this, we can prove another rather interesting theorem. Let G be the set of injective
analytic functions f : U −→C, so G ⊂ A(U). The theorem is that the closure is given by
G = G∪ { constant functions }
Still more interesting is the following theorem of Montel. Let us say that a set F ⊂ A(U) is
locally bounded if for every compact set K ⊂ U there is a constant M with f
K
≤ M for every
f ∈ F. Montel’s theorem states that F is compact if and only if it is locally bounded and closed.
The power of the above two results is revealed by the fact that Riemann mapping theorem is a
relatively simple consequence. This theorem states that any simply connected proper open subset
of C (no matter how wild its boundary) is conformally equivalent to the unit disc.
4. Examples from Differential Geometry and Algebraic Topology
The main interest of most of the following examples is their global topology, in other words,
what sort of holes they have in them and how the holes twist around each other and so on. This
course will lay important foundations for the study of such questions, and if we have time towards
the end we will address a few of the simpler ones. However, to understand such things fully we
would need the apparatus of algebraic topology; while this is particularly profound and beautiful,
it will have to wait until future courses. Nonetheless, we can at least take a quick look at some of
the phenomena which occur.
Disclaimer: my enthusiasm for this section has of course nothing whatever to do with the
subject of my research, honest.
EXAMPLES OF TOPOLOGICAL SPACES 5
4.1. Spheres.
S
n
= the nsphere = {x = (x
0
, . . . x
n
) ∈ R
n+1

¸
x
2
k
= 1}
While these spaces are rather simple, they are in an important sense the building blocks from
which most other spaces of interest are constructed. It turns out that to understand the process
of construction, one has to study the continuous maps from one sphere to another. An interesting
example is the Hopf map:
η: S
3
−→ S
2
To deﬁne it, we think of S
3
as a subset of R
4
= C
2
:
S
3
= {(z, w) ∈ C
2
such that z
2
+w
2
= 1}
On the other hand, we think of S
2
(which is an ordinary sphere, like the surface of a basketball)
as the Riemann sphere C ∪ ∞. The Hopf map is then just division:
η(z, w) = z/w ∈ C ∪ ∞
One interesting property is that the inverse image of any point in S
2
is a circle in S
3
. Any two
such circles are linked, like the links in a chain.
4.2. The Projective Plane. Our next example is the real projective plane:
RP
2
= { lines through the origin in R
3
}
Any such line crosses the unit sphere S
2
in two opposite points. Using this we can identify RP
2
with the space of opposite pairs of points, that is
RP
2
= S
2
/ ∼ x ∼ y iﬀ x = ±y
We give RP
2
the quotient topology coming from this identiﬁcation. This makes it a compact,
connected, Hausdorﬀ space.
Here is an example of a useful geometric construction involving this space. Suppose X is a nice
smooth surface in R
3
. For any point x ∈ X, there are two unit normal vectors to X at x, say n
and −n. It is not always possible to designate one of these as the positive normal in a way which
is consistent over the whole surface. Nonetheless, both unit normals deﬁne the same point in RP
2
,
so we get an unambiguous map
g : X −→RP
2
This is called the Gauss map.
4.3. Conﬁguration Spaces. Our next example is called the unordered conﬁguration space of k
points in C:
B
k
= { ﬁnite sets S ⊂ C with k elements }
We can describe a topology on this space in two diﬀerent ways (they turn out to be the same
topology). One way is to consider the ordered conﬁguration space
F
k
= {z = (z
1
, . . . z
k
) ∈ C
k
 z
i
= z
j
when i = j}
There is a surjective map from F
k
to B
k
which sends the ordered ktuple (z
1
, . . . z
k
) to the un
ordered set {z
1
, . . . z
k
}. Each point in B
k
comes from k! points in F
k
, corresponding to the diﬀerent
orders which could be imposed. For example, the six preimages of the point
{i, π, e} ∈ B
3
are the following six points in F
3
:
(i, π, e) (i, e, π) (π, i, e) (π, e, i) (e, i, π) (e, π, i)
We can thus topologise B
k
as a quotient space of F
k
. In fact, F
k
is a covering space of B
k
; I hope
to discuss covering spaces towards the end of the course.
Another approach to the topology on B
k
is as follows. Take k = 3 for simplicity. Given a set
S = {u, v, w} ⊂ C
6 NEIL STRICKLAND
consider the polynomial
p
S
(t) = (t −u)(t −v)(t −w) = t
3
+at
2
+bt +c
The numbers
a = −(u +v +w) b = uv +uw +vw c = −uvw
depend only the set S and not on the order in which I listed the elements. We thus get a well
deﬁned map
g : B
3
−→C
3
g(S) = (a, b, c)
This is injective, because S is precisely the set of roots of p
S
and so is determined uniquely by
(a, b, c). The image of g can be shown to be an open set in C
3
. We can use this to deﬁne a
topology on B
3
, in which the open sets are precisely the sets g
∗
(U) where U is open in C
3
. As
stated previously, this is the same as the quotient topology coming from F
3
.
The space F
3
is actually quite simple; you can check that the map
f(u, v, w) = (u, v −u, (w −u)/(v −u))
gives a homeomorphism
f : F
3
−→C ×C \ {0} ×C \ {0, 1}
However, the spaces F
k
for k > 3 are rather complicated, and the spaces B
k
are still worse. They
are rumoured to have an important relationship with the physics of string theory, which is one
good reason to study them.
You can show that B
k
and F
k
are connected, Hausdorﬀ, locally compact and metrisable topo
logical manifolds, but that they are not compact.
4.4. Loop Spaces. Next, we consider loop spaces on spheres:
Λ
n
S
m
= { continuous maps λ: S
n
−→ S
m
}
We give this the compactopen topology, which is deﬁned as follows. Given a compact set K ⊂ S
n
and an open set U ⊂ S
m
we write
W(K, U) = {λ ∈ Λ
n
S
m
 λ(K) ⊂ U}
These sets W(K, U) form a subbasis for the compactopen topology, so the open sets in this
topology are precisely the arbitary unions of ﬁnite intersections of sets of the form W(K, U).
The important point about the compact open topology is that the following “evaluation” map is
continuous:
ev: S
n
×Λ
n
S
m
−→ S
m
ev(x, λ) = λ(x)
Here is an interesting map σ: S
1
− → Λ
1
S
2
. Think of S
2
as the globe. A point in Λ
1
S
2
is a
function from S
1
to the globe, that is, a parameterised loop on the globe. The great circles which
start at the North pole, run down to a point x on the equator, then down to the South pole and
back up the other side, form a family of such loops. There is one such loop λ
x
for each point x
on the equator. On the other hand, we can also identify the equator with S
1
. We obtain a map
σ sending x ∈ S
1
= equator to the loop λ
x
. You can show that this is continuous.
A very hard, but very important, problem is to understand the connected components of Λ
n
S
m
when n ≥ m. A great deal of partial information is known, but the general case remains intractable.
The simplest case is that of Λ
1
S
1
, which is the space of continuous maps from the circle to itself.
It is simplest here to think of S
1
as the unit circle in the complex plane. Let C
0
(S
1
, R) be the
subspace of C(S
1
, R) consisting of continuous functions f : S
1
−→ R such that f(1) = 0. There is
a continuous function
γ : Z ×S
1
×C
0
(S
1
, R) −→ Λ
1
S
1
given by
γ(n, w, f)(z) = z
n
wexp(if(z))
which turns out to be a homeomorphism. As S
1
×C
0
(S
1
, R) is connected (why ?) this shows that
the set of components of Λ
1
S
1
bijects naturally with Z.
EXAMPLES OF TOPOLOGICAL SPACES 7
4.5. Matrix Groups. The last example in this section is the orthogonal group:
O
3
= {3 ×3 matrices such that A
T
= A
−1
}
(here A
T
denotes the transposed matrix).
This is topologised as a subspace of R
9
. You can show that it is compact. It is also a group
under matrix multiplication. You can show that the group operations are given by continuous
maps:
µ: O
3
×O
3
−→ O
3
µ(A, B) = AB
χ: O
3
−→ O
3
χ(A) = A
−1
Such a matrix has determinant ±1. As the determinant gives a continuous map
det: {3 ×3 matrices} −→R
we see that O
3
is disconnected. It falls into two parts:
SO
3
= O
+
3
= {A  det(A) = 1} O
−
3
= {A  det(A) = −1}
It can be shown that O
+
3
is the space of rotation matrices.
Suppose that L is a line through the origin in R
3
. Let P be the plane orthogonal to L. There
is a map ρ
L
from R
3
to itself, sending a vector v ∈ R
3
to its mirror image after reﬂection in P. If
the two unit vectors in L are n and −n, then you can check that the formula is
ρ
L
(v) = v −2(n.v)n = A
L
v
Here A
L
is the matrix whose (i, j) entry is δ
ij
−2n
i
n
j
, and δ
ij
is the Kronecker symbol:
δ
ij
=
1 if i = j
0 otherwise
Using this, we can see that there is a continuous map R: RP
2
−→ O
−
3
sending L to A
L
.
Here is another interesting map. A matrix A ∈ O
3
satisﬁes Av = v for every vector v ∈ R
3
,
so the action of A gives a continuous map α
A
: S
2
−→ S
2
. We thus get a map
α: O
3
−→ Λ
2
S
2
α(A) = α
A
You can check that this map is again continuous.
5. Fractal Examples
5.1. The Cantor Set. The simplest example of a fractal is the Cantor set. We deﬁne
U
l
=
3
l−1
¸
k=−3
l−1
((3k −
1
2
)3
−l
, (3k +
1
2
)3
−l
) ⊂ [−
1
2
, +
1
2
]
The Cantor set is then
X = [−
1
2
, +
1
2
] \
¸
l≥0
U
l
Another description is as follows: we start with the interval [−
1
2
, +
1
2
] and remove the middle third
(−
1
6
, +
1
6
) to leave two closed intervals [−
1
2
, −
1
6
] and [
1
6
,
1
2
]. We remove the middle thirds of each of
these to get four closed intervals of length 1/9, and so on. What we get in the limit is the Cantor
set again. At the n’th stage we have 2
n
closed intervals, each of length 3
−n
, so the total length is
(2/3)
n
. There is a well behaved concept of the “total length” of a subset of the real line (called
Lebesgue measure), which works even for curious sets like the Cantor set; from the above we can
see that the Lebesgue measure of X must be zero.
It is easy to see that X is compact and Hausdorﬀ. It is also totally disconnected: you can show
that the connected components of X are points. It is also perfect: every point a ∈ X lies in the
closure of X \ {a}. All these properties are quite typical of fractals.
8 NEIL STRICKLAND
The Cantor set is actually homeomorphic to an inﬁnite Cartesian product of copies of the two
point discrete space {−1, 1}. Indeed, you can show that the map
f : {−1, 1}
Z
+
−→ X f(a) =
¸
k>0
a
k
3
−k
is a homeomorphism.
Another typical fractal property is selfsimilarity: every neighbourhood of every point in X con
tains a homeomorphic copy of the whole set. To be more speciﬁc, consider a basic neighbourhood
(a − , a + ) of a point a ∈ X. For large l we have 3
−l
< . It follows from the homeomor
phism in the last paragraph that there is a unique integer n such that n is not divisible by 3 and
3
l
a −n ≤
1
2
. It then follows that the map
g : X −→ X ∩ (a −, a +)
g(b) = n + 3
−l
b
gives a homeomorphism between X and a small neighbourhood of a in X.
Such exact selfsimilarity is not actually very typical. In more complicated cases, there is
approximate selfsimilarity. To make this precise, we need to say what it means for two sets to
be almost the same, in other words, to impose a topology on a suitable collection of subsets of a
given space X. We shall consider several such topologies elsewhere in these notes.
5.2. Attractors for Newton’s Method. Consider what happens if we try to look for complex
roots of the equation z
3
−1 = 0 using Newton’s method. We start with some initial guess z
0
, and
recursively deﬁne
z
n+1
= z
n
−(z
3
n
−1)/3z
2
n
If z
n
= 0 we take z
m
= ∞ for all m > n. The hope is that this sequence will converge to some
number z which is a root of the equation. Of course, we know what the roots are — there are
three of them:
z = 1 z = ω = e
2πi/3
z = ω = e
−2πi/3
If our initial guess z
0
is close to ω (say) then the sequence (z
k
) will converge rapidly to ω. However,
if z
0
is intermediate between two of the roots, then the sequence need not converge at all. If it
does converge then the limit depends in a very sensitive and intricate way on the precise position
of z
0
.
Possibly the simplest example of a point for which the sequence does not converge is as follows:
ρ = 10
−1/6
θ = cos
−1
(−1/
5/32)/3
α = ρ exp(iθ)
If z
0
= α then the sequence is just
α, α, α, α, . . .
Before proceeding further, we change notation a little. We write
g(z) = z −(z
3
−1)/3z
2
=
2z
3
+ 1
3z
2
g
(3)
(z) = g(g(g(z))) g
(4)
(z) = g(g(g(g(z)))) etc.
These functions are called the iterates of g. In the old notation, we have
z
n
= g
(n)
(z
0
)
We can divide the complex plane into four parts, as follows:
F
1
= {w  g
(n)
(w) −→ 1 as n −→ ∞}
F
ω
= {w  g
(n)
(w) −→ ω as n −→ ∞}
F
ω
= {w  g
(n)
(w) −→ ω as n −→ ∞}
F = F
1
∪ F
ω
∪ F
ω
J = (C ∪ {∞}) \ F ⊂ C ∪ {∞}
EXAMPLES OF TOPOLOGICAL SPACES 9
It is not diﬃcult to write a program to plot these sets and colour them in four diﬀerent colours.
You will ﬁnd that they are extremely intricate.
The sets J and F are called the Julia set and the Fatou set respectively. The Julia set is
uncountable, closed, and perfect, and has empty interior. The Fatou set is open and has inﬁnitely
many connected components. The proofs of these facts involve subtle arguments in both complex
analysis and general topology.
If we write bU for the boundary of U, then we have the following very curious fact:
bF
1
= bF
ω
= bF
ω
= J
You should try to ﬁnd another example of three disjoint open sets with the same boundary, to see
quite how curious this fact is.
There is an interesting characterisation of F in terms of the topology of spaces of analytic
functions. For any open set V ⊂ C we write
G
V
= {g
(n)

V
such that n ∈ N}
We regard this as a subspace of the space of continuous maps from V to the Riemann sphere
C∪ {∞}, endowed with the compactopen topology. It turns out that F is the largest set V such
that G
V
has compact closure. This is the appropriate deﬁnition of the Fatou set for a more general
rational function g.
5.3. The Mandelbrot Set. Given a complex number c, deﬁne q
c
(z) = z
2
+c and
f
n
(c) = q
(n)
c
(0) = q
c
(q
c
(q
c
. . . (0) . . .))
where q
c
is applied n times. In other words:
f
0
(c) = 0
f
n+1
(c) = q
c
(f
n
(c)) = f
n
(c)
2
+c
The Mandelbrot set M is deﬁned as
M = {c ∈ C  f
n
(c) ≤ 2 for all n}
Many of you have probably seen pictures of this set  there are a number of computer programs
available to plot it, and indeed you can quite easily write such a program yourself. It has a very
intricate fractal boundary. Many small parts of the set contain approximate copies of the whole
set — in other words the Mandelbrot set is aproximately selfsimilar.
A number of interesting topological properties of M are known. Firstly, it is compact — this is
easy to see. Secondly, it is connected. This is at ﬁrst sight implausible — if you look at pictures
of M, you will see many small “islands” well separated from the main body of the set. However,
on closer inspection there appear to be thin tendrils linking the islands to the centre. The proof
that M is connected is quite formidable, involving very powerful methods from complex analysis.
However, it is also true that the complement of the Mandelbrot set is connected, and this is
comparatively straightforward to prove. This has a natural geometric interpretation. As M is
compact, the complement M
c
has a single unbounded component. If M had any holes in it (e.g.
if M were something like the closed annulus 1 ≤ z ≤ 2) then M
c
would also have a bounded
component, and so would be disconnected. Thus the fact that M
c
is connected just means that
M has no holes.
Another interesting fact is that M is the closure of its interior. It is an open question whether
M is locally connected — many other things would follow if it were.
6. Examples from Algebraic Geometry
Frequently, when one is faced a geometric problem involving sets of points in R
n
, the sets in
question are deﬁned by polynomial equations, and any relevant functions between them are also
given by polynomials. A situation like this can be analysed using the usual topology on R
n
— the
sets are then closed and the functions are continuous. However, polynomials are far more rigid
than arbitary continuous functions, so we could hope to replace the usual topology with a coarser
10 NEIL STRICKLAND
topology which would give more information. We will explore this idea in this section, but we will
use C rather than R for technical reasons.
6.1. The Zariski Topology on C
n
. We write C[z
1
, . . . z
n
] for the set of polynomial functions
from C
n
to C, for example the function
f(z
1
, z
2
, z
3
) = z
4
1
+iz
3
is an element of C[z
1
, z
2
, z
3
]. Given such a function f, we write
D(f) = {z  f(z) = 0}
It is easy to see that
D(f) ∩ D(g) = D(fg)
These sets D(f) form a basis for a new topology on C
n
, called the Zariski topology. Of course,
the sets D(f) are open in the usual topology, which implies that the Zariski topology is coarser
than the usual one.
The closed sets for the Zariski topology are all of the form
V (f
1
, . . . f
m
) = {z ∈ C
n
 f
1
(z) = . . . f
m
(z) = 0}
In principle, it seems that we ought to also allow sets like V (f
1
, f
2
, . . .) with inﬁnitely many f’s.
However, it is a consequence of the important Hilbert Basis Theorem that any such set can be
rewritten as V (g
1
, . . . g
m
) for some ﬁnite list of polynomials {g
1
, . . . g
m
}.
Let us write X
n
for the space C
n
equipped with the Zariski topology. It is not the same,
incidentally, as the product topology on C
n
= C×. . . C derived from the Zariski topology on each
factor.
The space X
n
has a number of properties which are strikingly diﬀerent from those of the spaces
considered previously. Firstly, it is not Hausdorﬀ. Indeed, any two nonempty open sets intersect
nontrivially, or in other words, every nonempty open set is dense. In fact, it is this example and
related ones which provide the main reason for bothering to study nonHausdorﬀ spaces.
However, every set consisting of a single point is closed, so that X
n
does satisfy the separation
axiom T
1
.
The next curious property of X
n
is that very many subspaces are compact. In particular, every
open subspace is compact. This is very unlike the situation with Hausdorﬀ spaces, in which every
compact set is closed.
6.2. Prime Spectra of Rings. Our next example requires some rather more sophisticated alge
bra. Consider a commutative ring A. We let X = spec(A) denote the set of prime ideals in A.
Given any ideal a ≤ A, we deﬁne
V (a) = {p ∈ spec(A)  a ≤ p}
These subsets of spec(A) satisfy
V (0) = spec(A) V (A) = ∅
V (Σ
i
a
i
) =
¸
i
V (a
i
)
V (a ∩ b) = V (ab) = V (a) ∪ V (b)
The sets D(a) = spec(A) \ V (a) are the open sets for a topology on X = spec(A), which we
again refer to as the Zariski topology. It is also good to consider the subset max(A), consisting of
the maximal ideals of A — we give this the subspace topology.
In particular, we can consider the case A = C[z
1
, . . . z
m
]. For any point z ∈ C
n
there is a
surjective evaluation homomorphism
ˆ z : A −→C ˆ z(f) = f(z)
The kernel of this is a maximal ideal:
m
z
= ker(ˆ z : A −→C) = {f ∈ A  f(z) = 0}
EXAMPLES OF TOPOLOGICAL SPACES 11
In fact, the map
C
n
− → max(A) z → m
z
is a homeomorphism if we give the left hand side the Zariski topology, as in the last section.
The space spec(A) is always compact. The closed points correspond to maximal ideals of A,
so the space max(A) is always T
1
. The larger space spec(A) is almost never T
1
, however. For
example, if A = Z then there is a point in spec(A) corresponding to the prime ideal {0}, and
the closure of this point is the whole space. On the other hand, the weaker axiom T
0
is always
satisﬁed. This and related examples are the main reason for studying spaces which are not T
1
.
7. Examples from Algebraic Number Theory
Number theory is to a large extent the study of Diophantine equations, that is, polynomial
equations whose solutions are required to be integers. The most famous example is of course the
Fermat equation
x
n
+y
n
= z
n
It was ﬁnally proved in June 1993 by Andrew Wiles that there are no nonzero integer solutions
to this when n > 2. This had been stated by Fermat over 300 years previously, but although he
claimed to have a proof, Fermat did not write it down — it is generally believed that he must
have been mistaken.
Anyway, one might ask how it is possible to attack such problems. One method is to work
modulo m for some convenient integer m. For example, consider the equation x
2
+ x + 1 = 0.
By considering the two cases in which x is even or odd, we see that the left hand side is always
odd and the right hand side is zero so there can be no integer solutions. We can rephrase this
argument: there are no solutions mod 2, and hence none integrally.
For a slightly diﬀerent example, consider the equation x
2
+ 2 = 0. This does have a solution
mod 2 (we can take x = 0) but it has no solution mod 4.
This all leads up to the idea that we should look for solutions modulo various large numbers
m (better still, numbers m with many factors) and view these as “approximate solutions” to the
original equation. It actually turns out to be technically convenient to focus on one prime p at a
time, and consider solutions modulo p
m
for large m. This is the core idea behind the constructions
of this section.
7.1. The padic Metric. Let p be a prime number. For any nonzero integer n, we can repeatedly
divide by p until this is no longer possible, and thus write n = p
v
m for uniquely determined integers
m and v ≥ 0. We then write
n
p
= p
−v
0
p
= 0
d
p
(l, n) = l −n
p
We ﬁnd that d
p
is a metric on Z, called the padic metric. It is very diﬀerent from the usual
metric. For example, as k −→ ∞ the numbers p
k
converge to zero padically but diverge to inﬁnity
in the usual sense.
Here is an analogy which may make this idea seem more natural. A formal power series is
a formal expression
¸
∞
k=0
a
k
x
k
with the a
k
being real numbers and x a symbol. There is no
requirement of convergence. The set of formal power series forms a ring in an obvious way  it is
called R[[x]]. It is usual to think of two power series f and g as being close to each other if they
agree to a high order, in other words, if f −g is divisible by a high power of x. This is analogous
to the padic topology, in which n and m are close if n −m is divisible by a high power of p.
The space Z with this metric has few good properties. Things improve greatly if we consider
the completion of Z with respect to d
p
, which is called the space of padic integers, denoted Z
p
.
This space is compact, Hausdorﬀ, and totally disconnected.
12 NEIL STRICKLAND
7.2. The padic Rationals. We can generalise the above slightly. If 0 = a ∈ Q then we can write
a = p
v
b/c where b and c are integers not divisible by p, and then write a
p
= p
−v
. This gives a
metric on Q. The completion is Q
p
, the ﬁeld of padic rationals. It is locally compact, Hausdorﬀ
and totally disconnected. All algebraic operations are continuous. There is a good notion of the
Fourier transform of a suitable function f : Q
p
−→ C, which is extremely important in modern
number theory.
7.3. The Ring of Ad`eles. It turns out to be convenient to regard ∞ as a prime and write a
∞
for the usual absolute value, so Q
∞
= R. Write C for the set of ﬁnite primes, and C
∞
= C ∪ {∞}.
We then have a topological ring:
¸
p∈C
∞
Q
p
= R ×
¸
C
Q
p
An element a = (a
p
) of this product is said to be an ad`ele if a
p
∈ Z
p
⊂ Q
p
for all but ﬁnitely many
ﬁnite primes p ∈ C. The set of ad`eles forms a ring A. You can check that it is locally compact,
although the inﬁnite product ring is not. If a ∈ Q then there is an ad`ele ˆ a with ˆ a
p
= a for all
p (why is this an ad`ele ?). This identiﬁes Q with a subspace of A. It is discrete in the subspace
topology. We can form the quotient additive group:
A/Q = A/ ∼ a ∼ b iﬀ a −b ∈ Q
it is an important fact that this is compact in the quotient topology.
On the other hand, we can consider the analogous construction involving only the ﬁnite primes:
A
= {(a
p
) ∈
¸
C
Q
p
 a
p
∈ Z
p
for almost all p}
It is another important fact that Q is dense in A
. This is closely related to the Chinese Remainder
Theorem.
1]. Continuous Functions. 2. v) = u − v . 1]. the Lebesgue integral agrees with any more elementary deﬁnition when the latter makes sense. and many other things. SquareIntegrable Functions. C[0. the existence and uniqueness of solutions to diﬀerential equations. 1]: ˆ a : C[0. The basic examples of elements of L2 (R/Z) are the functions en (x) = exp(2nπix) . the spectral theory of operators in quantum mechanics. 1] to R is given by integration: 1 I : C[0.2 NEIL STRICKLAND Given a norm. 1] → R − Another is the evaluation function a for a ∈ [0. we need to mention that the integration sign means the Lebesgue integral. It follows (using the Weierstrass M test) that C[0. Examples from Functional Analysis The examples in this section are all spaces of functions with various diﬀerent topologies. It is a dense subspace. To make the deﬁnition of L2 (R/Z) precise. 1] to R (such as f (x) = 5x2 + 6) form a subspace P [0. 1 L2 (R/Z) = {f : R − C such that f (x + 1) = f (x) and → 0 f (x)2 dx < ∞} This space is the natural home of the theory of Fourier series. no matter what norm we start with. This metric induces a topology. 1] = {continuous functions f : [0. 1] is complete as a metric space. 1] − R ˆ → I(f ) = 0 f (x)dx a(f ) = f (a) ˆ This idea of regarding f (a) as a function of f rather than of a is certainly curious at ﬁrst sight. As in the case of C[0. we deﬁne a norm and thence a metric: 1 f 2 = 0 f (x)2 dx 2 d(f. However. The polynomial functions from [0. An interesting example of a continuous function from C[0.1. 1] of C[0. which is studied in courses on measure theory. and the claim is that this is always the same as the product topology. by the StoneWeierstrass theorem. They are important for analysing the convergence of Fourier series. we deﬁne a metric by d(u. this is merely a technicality. 2. g) = f − g This is not quite a metric because it is possible to have d(f. 2. From it we derive a metric: d(f. which can be suppressed. g) = 0 even when f = g. we could have g = 0 and 1 if x ∈ Z f (x) = 0 otherwise This is again just an annoying technicality. 1] − R} → This is a normed space with the following norm: f ∞ = sup{f (x)  0 ≤ x ≤ 1} This is ﬁnite because a continuous real valued function on a compact space is bounded. but it turns out to be strikingly useful.2. For example. g) = f − g A sequence of functions (fn )∞ converges to a function f with respect to this metric if and only n=0 if (in the usual language of real analysis) it converges uniformly.
sup{f (x) ∞ ∞ such that − n ≤ x ≤ n}) 2−m−n Pn (Dm f ) d(f ) = n=0 m=0 d(f. This topology is generated by a rather ugly metric. . c) = b − c There are continuous maps L2 (Z) − L2 (R/Z) − L2 (Z) → → deﬁned by F (c) = n 1 ∞ F C cn en f (x)en (x)dx 0 n=−∞ C(f ) = (Cn (f ))∞ n=−∞ = These are in fact mutually inverse isometric isomorphisms: F C(f ) = f CF (c) = c d(C(f ). 2. as described in the next example. L2 (Z) = {series c = (cn )∞ n=−∞ such that n cn 2 < ∞} c 2 = n cn  2 d(b. which is the proper home of the Dirac delta function and similar beasts. g) = d(f − g) Fortunately. F (c)) = d(b. The function spaces described above are good for studying things like integration. In fact. one can usually use the characterisation of the topology in terms of RK and D and ignore the metric. f= n∈Z Cn (f )en where the sum converges with respect to the metric just described. c) This means that the two L2 spaces can be identiﬁed in a very strong sense. we let RK (f ) denote the restriction of f to K: RK : C ∞ (R) − C(K) → ∞ RK (f ) = f K D(f ) = f We also write D for the function from C (R) to itself sending a function to its derivative: D : C ∞ (R) − C ∞ (R) → ∞ We give C (R) the coarsest possible topology such that the maps RK (for all compact sets K) and D are continuous. as follows: Pn (f ) = min(1.3. 2. SquareSummable Sequences.EXAMPLES OF TOPOLOGICAL SPACES 3 and the basic example of a continuous function from L2 (R/Z) to C is the Fouriercoeﬃcient function 1 Cn (f ) = 0 f (x)en (x)dx The fundamental theorem about Fourier series is that for any f ∈ L2 . Understanding this space and certain closely related spaces is the ﬁrst step towards the theory of distributions. and diﬀerential equations can often be converted into integral equations by cunning means. still more is true.4. C ∞ (R) = { inﬁnitely diﬀerentiable functions f : R − R} → Given a compact subset K ⊂ R. g) d(F (b). but to study diﬀerentiation directly we need a diﬀerent kind of space. C(g)) = d(f. Smooth Functions.
Nonetheless. This topology has remarkably good properties. and thus (as N is discrete) constant on the connected components of F . The Riemann Sphere. If p and q are complex polynomials (not both zero) then the expression r(z) = p(z)/q(z) can be made sense of as a continuous function from C∞ to itself. so G ⊂ A(U ). so that f (z) = (z − 1)2 counts as having two zeros at z = 1. what sort of holes they have in them and how the holes twist around each other and so on. K.1. Using this. The theorem is that the closure is given by → G = G ∪ { constant functions } Still more interesting is the following theorem of Montel. although a certain amount of work needs to be done to justify this. K. Firstly. in other words. ) = {g ∈ A(U ) such that < } The sets B(f. Examples from Differential Geometry and Algebraic Topology The main interest of most of the following examples is their global topology. Let us say that a set F ⊂ A(U ) is locally bounded if for every compact set K ⊂ U there is a constant M with f K ≤ M for every f ∈ F . Montel’s theorem states that F is compact if and only if it is locally bounded and closed. ) form a basis for a topology on A(U ). Disclaimer: my enthusiasm for this section has of course nothing whatever to do with the subject of my research. 4. This is much the most natural context in which to think about such functions. If U is a connected open subset of C. This function turns out to be continuous. . we let A(U ) denote the space of analytic functions on U .2. Examples from Complex Analysis 3. while this is particularly profound and beautiful. Let G be the set of injective analytic functions f : U − C. and if we have time towards the end we will address a few of the simpler ones. If K ⊂ U is compact and f ∈ A(U ) then we write f K = max{f (z) such that z ∈ K} f −g K B(f. The power of the above two results is revealed by the fact that Riemann mapping theorem is a relatively simple consequence. called the topology of locally uniform convergence. However. 3. we can at least take a quick look at some of the phenomena which occur. much stronger than the corresponding ones for the space of merely continuous functions on U . to understand such things fully we would need the apparatus of algebraic topology. Spaces of Analytic Functions. honest. The open subsets are the open sets in C together with the sets U ∪ {∞} such that U is an open subset of C whose complement in C is compact.4 NEIL STRICKLAND 3. we can prove another rather interesting theorem. The Riemann sphere C∞ is the onepoint compactiﬁcation C∪{∞}. it will have to wait until future courses. This theorem states that any simply connected proper open subset of C (no matter how wild its boundary) is conformally equivalent to the unit disc. This course will lay important foundations for the study of such questions. it follows from the Cauchy integral formulae that the diﬀerentiation function is continuous: D : A(U ) − A(U ) → D(f ) = f If Γ is a simple closed contour whose interior is contained in U and F is the set {f ∈ A(U )  f has no zeros on Γ} then we can deﬁne a function vΓ : F − N → vΓ (f ) = number of zeros of f inside Γ Here zeros are counted by multiplicity in the usual way.
xn ) ∈ Rn+1  x2 = 1} k While these spaces are rather simple. e) (i. . Hausdorﬀ space. .1. one has to study the continuous maps from one sphere to another. Take k = 3 for simplicity. connected. . It is not always possible to designate one of these as the positive normal in a way which is consistent over the whole surface. I hope to discuss covering spaces towards the end of the course. like the surface of a basketball) as the Riemann sphere C ∪ ∞. say n and −n. S n = the nsphere = {x = (x0 . .EXAMPLES OF TOPOLOGICAL SPACES 5 4. π. Suppose X is a nice smooth surface in R3 . like the links in a chain. i) We can thus topologise Bk as a quotient space of Fk . . Given a set S = {u.2. Spheres. An interesting example is the Hopf map: η : S3 − S2 → To deﬁne it. In fact. i. v. Any two such circles are linked. . e) (π. i) (e. corresponding to the diﬀerent orders which could be imposed. Conﬁguration Spaces. so we get an unambiguous map g : X − RP 2 → This is called the Gauss map. Nonetheless. there are two unit normal vectors to X at x. Each point in Bk comes from k! points in Fk . zk ) to the unordered set {z1 . . Using this we can identify RP 2 with the space of opposite pairs of points. π. . 4. e. Our next example is called the unordered conﬁguration space of k points in C: Bk = { ﬁnite sets S ⊂ C with k elements } We can describe a topology on this space in two diﬀerent ways (they turn out to be the same topology). both unit normals deﬁne the same point in RP 2 . . 4. w} ⊂ C . .3. Our next example is the real projective plane: RP 2 = { lines through the origin in R3 } Any such line crosses the unit sphere S 2 in two opposite points. For example. we think of S 2 (which is an ordinary sphere. π. Fk is a covering space of Bk . i. One way is to consider the ordered conﬁguration space Fk = {z = (z1 . Here is an example of a useful geometric construction involving this space. The Projective Plane. For any point x ∈ X. that is RP 2 = S 2 / ∼ 2 x ∼ y iﬀ x = ±y We give RP the quotient topology coming from this identiﬁcation. . e} ∈ B3 are the following six points in F3 : (i. π) (e. they are in an important sense the building blocks from which most other spaces of interest are constructed. The Hopf map is then just division: η(z. zk }. we think of S 3 as a subset of R4 = C2 : S 3 = {(z. w) = z/w ∈ C ∪ ∞ One interesting property is that the inverse image of any point in S 2 is a circle in S 3 . w) ∈ C2 such that z2 + w2 = 1} On the other hand. . e. zk ) ∈ Ck  zi = zj when i = j} There is a surjective map from Fk to Bk which sends the ordered ktuple (z1 . Another approach to the topology on Bk is as follows. It turns out that to understand the process of construction. the six preimages of the point {i. This makes it a compact. π) (π.
6 NEIL STRICKLAND consider the polynomial pS (t) = (t − u)(t − v)(t − w) = t3 + at2 + bt + c The numbers a = −(u + v + w) b = uv + uw + vw c = −uvw depend only the set S and not on the order in which I listed the elements. We obtain a map σ sending x ∈ S 1 = equator to the loop λx . There is → a continuous function γ : Z × S 1 × C0 (S 1 . b. R) − Λ1 S 1 → given by γ(n. which is deﬁned as follows. . A very hard. 4. which is the space of continuous maps from the circle to itself. but very important. The simplest case is that of Λ1 S 1 . On the other hand. then down to the South pole and back up the other side. There is one such loop λx for each point x on the equator. this is the same as the quotient topology coming from F3 . form a family of such loops. R) be the subspace of C(S 1 . that is. It is simplest here to think of S 1 as the unit circle in the complex plane. a parameterised loop on the globe. Loop Spaces. w. The important point about the compact open topology is that the following “evaluation” map is continuous: ev : S n × Λn S m − S m → ev(x. You can show that this is continuous. w) = (u. The space F3 is actually quite simple. U ) form a subbasis for the compactopen topology. The great circles which start at the North pole. run down to a point x on the equator. R) is connected (why ?) this shows that the set of components of Λ1 S 1 bijects naturally with Z. we can also identify the equator with S 1 . U ) = {λ ∈ Λn S m  λ(K) ⊂ U } These sets W (K. b. but the general case remains intractable. We thus get a well deﬁned map g : B 3 − C3 → g(S) = (a. A great deal of partial information is known. f )(z) = z n w exp(if (z)) which turns out to be a homeomorphism. problem is to understand the connected components of Λn S m when n ≥ m. you can check that the map f (u. R) consisting of continuous functions f : S 1 − R such that f (1) = 0. which is one good reason to study them. Think of S 2 as the globe. 1} → However. v. v − u. and the spaces Bk are still worse. so the open sets in this topology are precisely the arbitary unions of ﬁnite intersections of sets of the form W (K. They are rumoured to have an important relationship with the physics of string theory. A point in Λ1 S 2 is a − function from S 1 to the globe. Next. You can show that Bk and Fk are connected. U ). As S 1 × C0 (S 1 . λ) = λ(x) Here is an interesting map σ : S 1 → Λ1 S 2 . locally compact and metrisable topological manifolds. c). Given a compact set K ⊂ S n and an open set U ⊂ S m we write W (K. As stated previously. we consider loop spaces on spheres: Λn S m = { continuous maps λ : S n − S m } → We give this the compactopen topology. c) This is injective. We can use this to deﬁne a topology on B3 . Hausdorﬀ.4. but that they are not compact. the spaces Fk for k > 3 are rather complicated. The image of g can be shown to be an open set in C3 . (w − u)/(v − u)) gives a homeomorphism f : F3 − C × C \ {0} × C \ {0. Let C0 (S 1 . in which the open sets are precisely the sets g ∗ (U ) where U is open in C3 . because S is precisely the set of roots of pS and so is determined uniquely by (a.
This is topologised as a subspace of R9 . Let P be the plane orthogonal to L. Suppose that L is a line through the origin in R3 . All these properties are quite typical of fractals. + 2 ] 2 2 The Cantor set is then X = [− 1 . A matrix A ∈ O3 satisﬁes Av = v for every vector v ∈ R3 . + 1 ) to leave two closed intervals [− 2 . You can show that the group operations are given by continuous maps: µ : O3 × O3 − O3 → µ(A. we can see that there is a continuous map R : RP 2 − O3 sending L to AL . If the two unit vectors in L are n and −n. It falls into two parts: + SO3 = O3 = {A  det(A) = 1} − O3 = {A  det(A) = −1} + It can be shown that O3 is the space of rotation matrices. then you can check that the formula is ρL (v) = v − 2(n. It is also perfect: every point a ∈ X lies in the closure of X \ {a}. Fractal Examples 5. so the action of A gives a continuous map αA : S 2 − S 2 . B) = AB χ : O3 − O3 → χ(A) = A−1 Such a matrix has determinant ±1. The last example in this section is the orthogonal group: O3 = {3 × 3 matrices such that AT = A−1 } (here AT denotes the transposed matrix).EXAMPLES OF TOPOLOGICAL SPACES 7 4. from the above we can see that the Lebesgue measure of X must be zero. The Cantor Set. . each of length 3−n . → − Here is another interesting map. There is a map ρL from R3 to itself. It is easy to see that X is compact and Hausdorﬀ. You can show that it is compact. (3k + 1 )3−l ) ⊂ [− 1 . + 1 ] \ 2 2 l≥0 Ul Another description is as follows: we start with the interval [− 1 . It is also a group under matrix multiplication. As the determinant gives a continuous map det : {3 × 3 matrices} − R → we see that O3 is disconnected. There is a well behaved concept of the “total length” of a subset of the real line (called Lebesgue measure). It is also totally disconnected: you can show that the connected components of X are points. We remove the middle thirds of each of 6 6 6 6 2 these to get four closed intervals of length 1/9. What we get in the limit is the Cantor set again. We thus get a map → α : O3 − Λ2 S 2 → You can check that this map is again continuous. − 1 ] and [ 1 . + 1 ] and remove the middle third 2 2 1 (− 1 . The simplest example of a fractal is the Cantor set. and so on.5. which works even for curious sets like the Cantor set. sending a vector v ∈ R3 to its mirror image after reﬂection in P . 1 ]. At the n’th stage we have 2n closed intervals. We deﬁne 3l−1 α(A) = αA Ul = k=−3l−1 1 1 ((3k − 2 )3−l . so the total length is (2/3)n . 5. Matrix Groups. j) entry is δij − 2ni nj . and δij is the Kronecker symbol: δij = 1 0 if i = j otherwise Using this.v)n = AL v Here AL is the matrix whose (i.1.
.8 NEIL STRICKLAND The Cantor set is actually homeomorphic to an inﬁnite Cartesian product of copies of the two point discrete space {−1. as follows: F1 = {w  g (n) (w) − 1 as n − ∞} → → Fω = {w  g (n) (w) − ω as n − ∞} → → Fω = {w  g (n) (w) − ω as n − ∞} → → F = F1 ∪ Fω ∪ Fω J = (C ∪ {∞}) \ F ⊂ C ∪ {∞} . 1}Z+ − X f (a) = k>0 ak 3−k is a homeomorphism. 5. We start with some initial guess z0 . we have g(z) = z − (z 3 − 1)/3z 2 = zn = g (n) (z0 ) We can divide the complex plane into four parts. we change notation a little.2. Another typical fractal property is selfsimilarity: every neighbourhood of every point in X contains a homeomorphic copy of the whole set. α. 1}. To make this precise. . α. In more complicated cases. If it does converge then the limit depends in a very sensitive and intricate way on the precise position of z0 . we need to say what it means for two sets to be almost the same. then the sequence need not converge at all. Before proceeding further. to impose a topology on a suitable collection of subsets of a given space X. you can show that the map → f : {−1. These functions are called the iterates of g. α. there is approximate selfsimilarity. We shall consider several such topologies elsewhere in these notes. consider a basic neighbourhood (a − . a + ) of a point a ∈ X. In the old notation. To be more speciﬁc. if z0 is intermediate between two of the roots. For large l we have 3−l < . Consider what happens if we try to look for complex roots of the equation z 3 − 1 = 0 using Newton’s method. Indeed. Such exact selfsimilarity is not actually very typical. We write 2z 3 + 1 3z 2 g (3) (z) = g(g(g(z))) g (4) (z) = g(g(g(g(z)))) etc. . Possibly the simplest example of a point for which the sequence does not converge is as follows: ρ = 10−1/6 θ = cos−1 (−1/ 5/32)/3 α = ρ exp(iθ) If z0 = α then the sequence is just α. However. Attractors for Newton’s Method. in other words. a + ) → g(b) = n + 3−l b gives a homeomorphism between X and a small neighbourhood of a in X. Of course. we know what the roots are — there are three of them: z=1 z = ω = e2πi/3 z = ω = e−2πi/3 If our initial guess z0 is close to ω (say) then the sequence (zk ) will converge rapidly to ω. It then follows that the map 2 g : X − X ∩ (a − . It follows from the homeomorphism in the last paragraph that there is a unique integer n such that n is not divisible by 3 and 3l a − n ≤ 1 . and recursively deﬁne 3 2 zn+1 = zn − (zn − 1)/3zn If zn = 0 we take zm = ∞ for all m > n. The hope is that this sequence will converge to some number z which is a root of the equation.
The Fatou set is open and has inﬁnitely many connected components. the sets in question are deﬁned by polynomial equations. This is the appropriate deﬁnition of the Fatou set for a more general rational function g. It turns out that F is the largest set V such that GV has compact closure. deﬁne qc (z) = z 2 + c and (n) fn (c) = qc (0) = qc (qc (qc . The proof that M is connected is quite formidable. The Mandelbrot Set. However. it is compact — this is easy to see.g. If we write bU for the boundary of U . For any open set V ⊂ C we write GV = {g (n) V such that n ∈ N} We regard this as a subspace of the space of continuous maps from V to the Riemann sphere C ∪ {∞}. If M had any holes in it (e. Given a complex number c. then we have the following very curious fact: bF1 = bFω = bFω = J You should try to ﬁnd another example of three disjoint open sets with the same boundary. A situation like this can be analysed using the usual topology on Rn — the sets are then closed and the functions are continuous. In other words: f0 (c) = 0 fn+1 (c) = qc (fn (c)) = fn (c)2 + c The Mandelbrot set M is deﬁned as M = {c ∈ C  fn (c) ≤ 2 for all n} Many of you have probably seen pictures of this set . (0) .)) where qc is applied n times. This is at ﬁrst sight implausible — if you look at pictures of M . Secondly.there are a number of computer programs available to plot it.3. endowed with the compactopen topology. the complement M c has a single unbounded component. you will see many small “islands” well separated from the main body of the set. 6. if M were something like the closed annulus 1 ≤ z ≤ 2) then M c would also have a bounded component. . so we could hope to replace the usual topology with a coarser . Examples from Algebraic Geometry Frequently. on closer inspection there appear to be thin tendrils linking the islands to the centre. . Thus the fact that M c is connected just means that M has no holes. It is an open question whether M is locally connected — many other things would follow if it were. This has a natural geometric interpretation. The proofs of these facts involve subtle arguments in both complex analysis and general topology. Many small parts of the set contain approximate copies of the whole set — in other words the Mandelbrot set is aproximately selfsimilar. involving very powerful methods from complex analysis. The sets J and F are called the Julia set and the Fatou set respectively. and so would be disconnected. and indeed you can quite easily write such a program yourself. There is an interesting characterisation of F in terms of the topology of spaces of analytic functions.EXAMPLES OF TOPOLOGICAL SPACES 9 It is not diﬃcult to write a program to plot these sets and colour them in four diﬀerent colours. it is connected. However. . As M is compact. polynomials are far more rigid than arbitary continuous functions. However. when one is faced a geometric problem involving sets of points in Rn . A number of interesting topological properties of M are known. . it is also true that the complement of the Mandelbrot set is connected. You will ﬁnd that they are extremely intricate. closed. and any relevant functions between them are also given by polynomials. Firstly. to see quite how curious this fact is. 5. Another interesting fact is that M is the closure of its interior. It has a very intricate fractal boundary. The Julia set is uncountable. and has empty interior. and perfect. and this is comparatively straightforward to prove.
z3 ]. Let us write Xn for the space Cn equipped with the Zariski topology. Given any ideal a ≤ A. gm ) for some ﬁnite list of polynomials {g1 . gm }. Indeed. For any point z ∈ Cn there is a surjective evaluation homomorphism z: A − C ˆ → The kernel of this is a maximal ideal: mz = ker(ˆ : A − C) = {f ∈ A  f (z) = 0} z → z (f ) = f (z) ˆ . It is also good to consider the subset max(A). . the sets D(f ) are open in the usual topology. we can consider the case A = C[z1 . . it is this example and related ones which provide the main reason for bothering to study nonHausdorﬀ spaces. . z2 . In particular. so that Xn does satisfy the separation axiom T1 . every set consisting of a single point is closed. we deﬁne V (a) = {p ∈ spec(A)  a ≤ p} These subsets of spec(A) satisfy V (0) = spec(A) V (Σi ai ) = i V (A) = ∅ V (ai ) V (a ∩ b) = V (ab) = V (a) ∪ V (b) The sets D(a) = spec(A) \ V (a) are the open sets for a topology on X = spec(A). . z2 . . . . zm ]. any two nonempty open sets intersect nontrivially. . Our next example requires some rather more sophisticated algebra. Of course. fm ) = {z ∈ Cn  f1 (z) = . The Zariski Topology on Cn .2. Prime Spectra of Rings. every nonempty open set is dense. The next curious property of Xn is that very many subspaces are compact. . In particular. but we will use C rather than R for technical reasons. The space Xn has a number of properties which are strikingly diﬀerent from those of the spaces considered previously. . . Given such a function f . . z3 ) = z1 + iz3 is an element of C[z1 . which implies that the Zariski topology is coarser than the usual one. . we write D(f ) = {z  f (z) = 0} It is easy to see that D(f ) ∩ D(g) = D(f g) These sets D(f ) form a basis for a new topology on Cn . C derived from the Zariski topology on each factor. We will explore this idea in this section. consisting of the maximal ideals of A — we give this the subspace topology. . However. Firstly. zn ] for the set of polynomial functions from Cn to C. We let X = spec(A) denote the set of prime ideals in A. . We write C[z1 . for example the function 4 f (z1 . . it is a consequence of the important Hilbert Basis Theorem that any such set can be rewritten as V (g1 .1. it is not Hausdorﬀ. . . as the product topology on Cn = C × . 6. . . it seems that we ought to also allow sets like V (f1 . called the Zariski topology. fm (z) = 0} In principle. or in other words. incidentally. It is not the same. Consider a commutative ring A. f2 .10 NEIL STRICKLAND topology which would give more information. .) with inﬁnitely many f ’s. However. which we again refer to as the Zariski topology. in which every compact set is closed. The closed sets for the Zariski topology are all of the form V (f1 . In fact. every open subspace is compact. . 6. This is very unlike the situation with Hausdorﬀ spaces.
On the other hand. 7. in which n and m are close if n − m is divisible by a high power of p. The closed points correspond to maximal ideals of A. . Things improve greatly if we consider the completion of Z with respect to dp . For example. so the space max(A) is always T1 . This had been stated by Fermat over 300 years previously. One method is to work modulo m for some convenient integer m. Hausdorﬀ. Let p be a prime number. in other words. We then write np = p−v 0p = 0 dp (l. Here is an analogy which may make this idea seem more natural. Anyway. however. Examples from Algebraic Number Theory Number theory is to a large extent the study of Diophantine equations. consider the equation x2 + x + 1 = 0. It is usual to think of two power series f and g as being close to each other if they agree to a high order. There is no requirement of convergence. The set of formal power series forms a ring in an obvious way . It is very diﬀerent from the usual metric. This is analogous to the padic topology. and totally disconnected. This and related examples are the main reason for studying spaces which are not T1 . This does have a solution mod 2 (we can take x = 0) but it has no solution mod 4. but although he claimed to have a proof.it is called R[[x]]. we can repeatedly divide by p until this is no longer possible. The space Z with this metric has few good properties. numbers m with many factors) and view these as “approximate solutions” to the original equation. n) = l − np We ﬁnd that dp is a metric on Z. called the padic metric.1. if A = Z then there is a point in spec(A) corresponding to the prime ideal {0}. For example. The space spec(A) is always compact. that is. and thus write n = pv m for uniquely determined integers m and v ≥ 0. This space is compact. consider the equation x2 + 2 = 0. The larger space spec(A) is almost never T1 . This all leads up to the idea that we should look for solutions modulo various large numbers m (better still. as in the last section. For a slightly diﬀerent example. A formal power series is ∞ k a formal expression k=0 ak x with the ak being real numbers and x a symbol. By considering the two cases in which x is even or odd. We can rephrase this argument: there are no solutions mod 2. Fermat did not write it down — it is generally believed that he must have been mistaken. and hence none integrally. For example. This is the core idea behind the constructions of this section. For any nonzero integer n. polynomial equations whose solutions are required to be integers. and the closure of this point is the whole space. if f − g is divisible by a high power of x. the map Cn → max(A) − z → mz is a homeomorphism if we give the left hand side the Zariski topology. and consider solutions modulo pm for large m. denoted Zp . The most famous example is of course the Fermat equation xn + y n = z n It was ﬁnally proved in June 1993 by Andrew Wiles that there are no nonzero integer solutions to this when n > 2. one might ask how it is possible to attack such problems. we see that the left hand side is always odd and the right hand side is zero so there can be no integer solutions. 7. which is called the space of padic integers. the weaker axiom T0 is always satisﬁed. as k − ∞ the numbers pk converge to zero padically but diverge to inﬁnity → in the usual sense.EXAMPLES OF TOPOLOGICAL SPACES 11 In fact. The padic Metric. It actually turns out to be technically convenient to focus on one prime p at a time.
so Q∞ = R. which is extremely important in modern → number theory. The completion is Qp . This gives a metric on Q. If 0 = a ∈ Q then we can write a = pv b/c where b and c are integers not divisible by p.12 NEIL STRICKLAND 7. we can consider the analogous construction involving only the ﬁnite primes: A = {(ap ) ∈ C Qp  ap ∈ Zp for almost all p} It is another important fact that Q is dense in A . Hausdorﬀ and totally disconnected. The set of ad`les forms a ring A. There is a good notion of the Fourier transform of a suitable function f : Qp − C. and then write ap = p−v . the ﬁeld of padic rationals. It turns out to be convenient to regard ∞ as a prime and write a∞ e for the usual absolute value. The padic Rationals. .3. We can generalise the above slightly.2. and C∞ = C ∪ {∞}. This is closely related to the Chinese Remainder Theorem. If a ∈ Q then there is an ad`le a with ap = a for all e ˆ ˆ p (why is this an ad`le ?). All algebraic operations are continuous. Write C for the set of ﬁnite primes. 7. This identiﬁes Q with a subspace of A. It is discrete in the subspace e topology. It is locally compact. The Ring of Ad`les. You can check that it is locally compact. e although the inﬁnite product ring is not. We can form the quotient additive group: A/Q = A/ ∼ a ∼ b iﬀ a − b ∈ Q it is an important fact that this is compact in the quotient topology. We then have a topological ring: Qp = R × Qp p∈C∞ C e An element a = (ap ) of this product is said to be an ad`le if ap ∈ Zp ⊂ Qp for all but ﬁnitely many ﬁnite primes p ∈ C. On the other hand.
This action might not be possible to undo. Are you sure you want to continue?