Low-Dimensional Topology and Knot Theory

András Juhász
 Contents

Introduction 4
Chapter 1. Higher dimensional manifolds 6
1.1. Topological manifolds 6
1.2. Fibre bundles 10
1.3. Smooth manifolds 12
1.4. Embeddings, immersions, and submersions 17
1.5. Morse theory 20
1.6. Handle decompositions and surgery 23
1.7. Cobordisms 26
1.8. The Whitney trick 28
1.9. The h-cobordism theorem 30

Chapter 2. 3-manifolds 33
2.1. The Schönflies theorem 33
2.2. Heegaard decompositions and diagrams 34
2.3. Incompressible surfaces and the loop theorem 37
2.4. Haken manifolds 41
2.5. Normal surfaces and prime decomposition 42
Chapter 3. Knots and links 48
3.1. Knots and links 48
3.2. Reidemeister moves 49
3.3. Seifert surfaces 50
3.4. The Seifert form 53
3.5. Important classes of knots 56
3.6. The knot group 60
3.7. Fibred knots 62
3.8. The Jones polynomial 64
3.9. Constructing 3-manifolds using links 66

Chapter 4. 4-manifolds 69
4.1. Kirby calculus 69
4.2. The intersection form and the classification of 4-manifolds 72
Bibliography 77

3
 Introduction

This course gives an introduction to low-dimensional topology. This is the study

of manifolds of dimensions two, three, and four, and includes knot theory. Roughly
speaking, an n-dimensional manifold is a topological space that locally looks like
Rn . They arise as configurations spaces of mechanical systems or solution sets of
equations. They also appear in theoretical physics, for example as spacetimes in
different versions of string theory.
To understand the distinction between low (≤ 4) and high (> 4) dimensions,
we first introduce the general theory of topological and smooth manifolds, and give
an overview of high dimensions. A particularly fruitful tool is to look at the critical
points of smooth functions on a smooth manifold, which is called Morse theory.
This allows one to decompose manifolds into handles, which can be thought of as
thickened cells of a cell decomposition. The Whitney trick, which is a method to
eliminate pairs of intersection points of complementary-dimensional submanifolds
and only works in high dimensions, allows us to prove the celebrated h-cobordism
theorem. As a corollary, we obtain the generalised Poincaré conjecture, which
states that, if a smooth n-manifold for n ≥ 5 is simply-connected and has the same
homology as the n-sphere, then it is homeomorphic to it.
We then cover some results from the classical theory of 3-manifolds. This in-
cludes the Schönflies theorem (every smoothly embedded 2-sphere in R3 bounds
a disk), Dehn’s lemma and the loop theorem, unique prime decomposition of
3-manifolds via normal surface theory, and the classification of lens spaces (3-
manifolds that can be obtained by gluing together two solid tori).
3- and 4-manifolds can be described using links in the 3-sphere whose com-
ponents are labelled by integers. Hence, knot theory forms a fundamental part of
low-dimensional topology. We introduce some classical knot invariants, such as the
Alexander polynomial and the signature. These can be derived from surfaces that
the knot bounds. We introduce several important classes of knots, then study the
fundamental group of the knot complement. We give a brief overview of a more
modern knot invariant called the Jones polynomial, whose topological meaning is
still mysterious. Moving towards dimension four, we study surfaces that a knot
bounds in the 4-ball. Dehn surgery is a tool for removing a neighbourhood of a
knot and regluing it in order to obtain a new 3-manifold. One can obtain any
3-manifold using a sequence of such surgeries on knots in the 3-sphere. Another
way to obtain new 3-manifolds from a link in the 3-sphere is via branched covers
(meromorphic maps are analogues in dimension 2).
We conclude with looking at 4-manifolds. Kirby calculus gives us a diagram-
matic way to define and manipulate 4-manifolds using links. We then overview

4
 INTRODUCTION 5

the classification of simply-connected topological 4-manifolds in terms of their in-

tersection forms on second homology due to Freedman, and the obstructions to
topological 4-manifolds admitting smooth structures due to Donaldson.
It will become apparent by the end of this course that low-dimensional topol-
ogy uses tools from many areas of mathematics, including geometry, analysis, and
algebra.
 CHAPTER 1

Higher dimensional manifolds

Manifolds form a particularly nice class of topological spaces that appear in

many areas of mathematics. We first give a brief introduction to topological and
smooth manifolds, but will mostly focus on smooth manifolds thereafter. We do
assume some basic familiarity with 1-manifolds and surfaces. We then give an
overview of Morse theory, which allows one to study smooth manifolds using the
critical points of smooth functions they admit. This leads us to handle decomposi-
tions and the h-cobordism theorem, a key result that underlies the classification of
higher dimensional manifolds using surgery theory.

1.1. Topological manifolds

For a non-negative integer n, an n-dimensional manifold (or n-manifold, in
short) is a topological space that locally looks like Euclidean space Rn . We now
give a rigorous definition.
Definition 1.1.1. An n-dimensional (topological) manifold M is a topological
space such that
(1) each point of M has a neighbourhood homeomorphic to Rn ,
(2) M is second countable (i.e., it has a countable basis of open sets), and
(3) M is Hausdorff (i.e., different points have disjoint neighbourhoods).
Exercise 1.1.2. Give examples of topological spaces that satisfy exactly two
of conditions (1)–(3).
Topological manifolds form a subcategory of the category of topological spaces,
where the morphisms are continuous maps. As for topological spaces, isomorphisms
are the homeomorphisms.
We now look at some examples. Clearly, Rn is an n-manifold. So is the n-sphere
S n = {x ∈ Rn+1 : |x| = 1}.
The real projective space
RPn = (Rn+1 \ {0})/R∗ = S n /{±1}
is the set of lines in Rn+1 . The complex projective space
CPn = (Cn+1 \ {0})/C∗ = S 2n+1 /S 1
is the set of complex lines in Cn+1 , and has (real) dimension 2n. All the familiar
matrix groups GL(n), O(n), SO(n) etc. are manifolds (in fact, they are Lie groups).
Manifolds with boundary will also play an important role. We write
Rn+ := {(x1 , . . . , xn ) ∈ Rn : xn ≥ 0}.
6
 1.1. TOPOLOGICAL MANIFOLDS 7

Definition 1.1.3. An n-dimensional manifold with boundary is a topological

space M such that
(1) each point of M has a neighbourhood homeomorphic to Rn or Rn+ ,
(2) M is second countable, and
(3) M is Hausdorff.
The interior Int(M ) of M consists of those points of M that have neighbourhoods
homeomorphic to Rn , and its boundary is ∂M := M \ Int(M ).
Note that, if a point of M has a neighbourhood homeomorphic to Rn , then it
also has a neighbourhood homeomorphic to Rn+ , so we cannot define ∂M as the set
of points of M having a neighbourhood homeomorphic to Rn+ . The subspace ∂M
is a manifold without boundary.
For example, the n-disk
Dn = {x ∈ Rn : |x| ≤ 1}
is a manifold with boundary S n−1 . We will write B n for the open ball Int(Dn ).
Note that every manifold is a manifold with (empty) boundary. We say that a
manifold is closed if it is compact and has no boundary.
Definition 1.1.4. A subset N of an m-manifold M is called a submanifold
if it is a manifold with the subspace topology. We say that N is a locally flat
submanifold of M if each p ∈ N has a neighbourhood Up ⊂ M such that the pair
(Up , Up ∩ N ) is homeomorphic to (Rm , Rn ) for some n.
Example 1.1.5. Let K be a non-trivial knot in S 3 ; i.e., a connected, locally flat
1-dimensional submanifold of S 3 such that the pair (S 3 , K) is not homeomorphic
to (S 3 , S 1 ). Then the cone on K from the centre 0 of D4 is a submanifold of D4 ,
but it is not locally flat at 0.
Definition 1.1.6. A closed, connected n-manifold is called orientable if
Hn (M ) ∼
= Z.
If M is orientable, an orientation of M is a choice of generator of Hn (M ). This
generator is called a fundamental class.
Note that the above isomorphism is not canonical, and hence there is no pre-
ferred orientation. The above homological definition of orientation captures the
intuition of having a coherent system of local orientations in the following sense:
Let α ∈ Hn (M ) ∼ = Z be a generator. Then, for every point p ∈ M , the im-
age of α under the map Hn (M ) → Hn (M, M \ {p}) induced by the embedding
(M, ∅) → (M, M \ {p}) is a generator of Hn (M, M \ {p}) ∼ = Z.
More generally, if M is a compact, connected manifold with boundary, then M
is called orientable when Hn (M, ∂M ) ∼
= Z, and an orientation is a generator of this
group. A fundamental example of a non-orientable manifold is the Möbius band.
A 2-manifold is orientable if and only if it does not contain a Möbius band.
Exercise 1.1.7. Construct a cell decomposition of RPn , and use this to com-
pute the homology groups H∗ (RPn ; Z), H∗ (RPn ; Z2 ), and the cohomology ring
H ∗ (RPn ; Z2 ). For what n is RPn orientable?
One of the main problems of manifold topology is the classification of manifolds.
 1.1. TOPOLOGICAL MANIFOLDS 8

Exercise 1.1.8. Show that a 0-manifold is a discrete, countable topological

space.
We shall focus on connected manifolds, as every other manifold is a countable
disjoint union of such. We state the classification of 1-manifolds. For a proof, see
for example Fuks–Rokhlin [13] or Milnor [44].
Theorem 1.1.9. Every connected 1-manifold with boundary is homeomorphic
to one of S 1 , R, [0, ∞), or I := [0, 1].
If M and N are manifolds with boundary, we can form their product M × N ,
which has dimension dim(M )+dim(N ). For example, the n-torus T n is the product
of n copies of S 1 . We call S 1 × D2 a solid torus.
Exercise 1.1.10. Show that ∂(M × N ) = (∂M × N ) ∪ (M × ∂N ).
Remark 1.1.11. Clearly, (∂M × N ) ∩ (M × ∂N ) = ∂M × ∂N . In particular,
by applying the formula of Exercise 1.1.10 to S 3 = ∂D4 ≈ ∂(D2 × D2 ), we see that
S 3 = (S 1 × D2 ) ∪ (D2 × S 1 ).
In other words, we can obtain the 3-sphere by gluing two solid tori along their
boundaries. The gluing map interchanges the meridian {1} × S 1 and the longitude
S 1 × {1} of T 2 = ∂(S 1 × D2 ).
Another important operation on manifolds is the connected sum. Suppose that
M and N are manifolds of the same dimension n. Remove interiors of closed balls
from M and N , and glue them along their sphere boundaries. The resulting mani-
fold M #N only depends on which components of M and N we removed the balls
from, up to homeomorphism. If n > 2, then π1 (S n−1 ) ∼ = 1, and hence π1 (M #N )
is the free product of π1 (M ) and π1 (N ) by the Seifert–van Kampen theorem.
A manifold of dimension two is also called a surface. For a positive integer g, let
Σg be the connected sum of g copies of the 2-torus T 2 , and we write Σ0 = S 2 . The
surface Σg is orientable; i.e., two-sided. Here g is called the genus of the orientable
surface Σg . We can obtain the non-orientable surface Ng by taking the connected
sum of g ≥ 1 copies of the real projective plane RP2 . We now state the classification
of compact surfaces without proof.
Theorem 1.1.12. Every closed, connected surface is homeomorphic to either
Σg for some g ≥ 0, or Ng for some g ≥ 1. Compact surfaces with boundary are
obtained from these by removing finitely many open balls.
The proof for triangulated surfaces can be found in most introductory textbooks
(see for example Munkres [52]) and was covered in the Part A Topology option.
The fact that every topological surface can be triangulated is a non-trivial result
of Radó [57]; see also Hatcher [22]. This is not true in higher dimensions, as there
are topological manifolds in every dimension at least 5 that cannot be triangulated
by the work of Manolescu [39]. Surprisingly, the proof of this result relies on a
gauge-theoretic invariant of 3-manifolds called monopole Floer homology.
3-manifolds are much more complicated. While a lot is known about them,
we do not have a complete classification as in dimension 2 and below. The most
important 3-manifold invariant is the fundamental group. Perelman proved the
famous Poincaré Conjecture only 100 years after it was first formulated. This states
that the only closed simply-connected 3-manifold is S 3 . In fact, he proved the much
 1.1. TOPOLOGICAL MANIFOLDS 9

stronger Geometrisation Conjecture of Bill Thurston: Every closed 3-manifold can

be cut along embedded 2-spheres and 2-tori such that each of the resulting pieces
carries one of eight special geometric structures. These include spherical, Euclidean,
and hyperbolic. The most difficult to understand are the hyperbolic pieces, which
are determined by their fundamental groups by Mostow rigidity.
In dimensions 4 and higher, there is no hope for obtaining a complete classifi-
cation. This is due to the following two results:

Theorem 1.1.13. For every finitely presented group G and integer n ≥ 4, there
exists a closed n-manifold M with π1 (M ) ∼
= G.

Proof. Let h x1 , . . . , xk | r1 , . . . , rl i be a presentation of G. Consider the n-

manifold X obtained by taking the connected sum of k copies of S 1 × S n−1 . Then
π1 (X) is the free product of k copies of

π1 (S 1 × S n−1 ) ∼
= π1 (S 1 ) × π1 (S n−1 ) ∼
= Z;

i.e., π1 (X) ∼
= h x1 , . . . , xk i.
We are now going to change X to introduce the relations r1 , . . . , rl using an
operation called surgery that we will discuss in more detail later. Let γi be a curve
in X freely homotopic to the relation ri ∈ π1 (X) for i ∈ {1, . . . , l}. As n ≥ 3, we
can assume that the curves γ1 , . . . , γl are embedded, pairwise disjoint, and have
neighbourhoods N (γi ) homeomorphic to S 1 × Dn−1 by representing each generator
xi using curves of the form S 1 × {p} for some p ∈ S n−1 .
An application of the Seifert–van Kampen theorem shows that

π1 (X \ N (γ1 ∪ · · · ∪ γl )) ∼
= π1 (X).

Indeed, π1 (N (γi )) ∼
= Z is generated by [γi ], which is homotopic to the generator of

π1 (∂N (γi )) ∼
= π1 (S 1 × S n−2 ) ∼
= Z,

where we have used that π1 (S n−2 ) ∼ = 1 as n ≥ 4. Hence, gluing N (γi ) to X \N (γ1 ∪

· · · ∪ γl ) does not change the fundamental group, as it introduces a new generator
yi and a new relation yi = wi for an element wi ∈ π1 (X \ N (γ1 ∪ · · · ∪ γl )).
Finally, we glue a copy of D2 × S n−2 to ∂N (γi ) ≈ S 1 × S n−2 such that ∂D2 is
identified with S 1 × {p} for each i ∈ {1, . . . , l} for some p ∈ S n−2 . Again, by the
Seifert–van Kampen theorem, this kills the homotopy class of γi . 

Remark 1.1.14. In contrast, note that the fundamental group of every closed
3-manifold has a presentation with the same number of generators as relations, and
is hence not arbitrary. This holds since every closed 3-manifold admits a Heegaard
decomposition; see Chapter 2.2.

Theorem 1.1.15. There is no algorithm to decide whether two finitely presented

groups are isomorphic.

Hence, to classify manifolds of dimension at least 4, one has to put restrictions

on the fundamental group. A natural choice is the class of simply-connected man-
ifolds. Simply-connected topological 4-manifolds were classified by Freedman; see
Chapter 4.2.
 1.2. FIBRE BUNDLES 10

1.2. Fibre bundles

Here, we review the necessary background on fibre bundles without proofs. For
further details, see the books of Steenrod [66] and Milnor–Stasheff [46]. Intuitively,
a fibre bundle is a space E that is locally a product, but might be twisted globally.
Definition 1.2.1. A fibre bundle consists of a surjective, continuous map
π : E → B, where E is called the total space and B the base space, and a topological
space F called the fibre. We assume that B is path-connected, and that each point of
B has a neighbourhood U such that there is a homeomorphism h : π −1 (U ) → U ×F
that makes the following diagram commutative:

π −1 (U )
h / U × F,

π
pU
 y
U
where pU is projection onto U .
It immediately follows from the definition that Fb := π −1 ({b}) is homeomorphic
to the fibre F for every b ∈ B. However, this homeomorphism is not canonical.
Depending on the context, we will denote a fibre bundle using only the total space
E, by π : E → B, or by F → E → B, where F → E refers to a homeomorphism
between F and π −1 ({b}) for some b ∈ B.
Example 1.2.2. (1) A covering space is a fibre bundle where π is a local
homeomorphism, and consequently F is discrete.
(2) If µ is the Möbius band, then the projection π : µ → S 1 onto its core circle
is a fibre bundle with fibre [−1, 1].
(3) Let ϕ : F → F be a homeomorphism. Then
Mϕ := I × F/(1,x)∼(0,ϕ(x)) ,
called the mapping torus of ϕ, is a fibre bundle over S 1 with fibre F , where
π([(t, x)]) := t for t ∈ I and x ∈ F . When F = [−1, 1] and ϕ(x) = −x, we
recover the previous example.
Definition 1.2.3. Let π : E → B and π 0 : E 0 → B 0 be fibre bundles. A bundle
morphism consists of continuous maps ε : E → E 0 and β : B → B 0 such that the
following diagram commutes:
ε /
E E0
π π0
 
B
β
/ B0.
Bundles and morphisms between them form a category, and hence we can talk
about bundle isomorphisms. If the base space B is contractible, then every bundle
over it is trivial ; i.e., equivalent to a product.
Definition 1.2.4. We say that s : B → E is a section of the bundle π : E → B
if π ◦ s = IdB .
Not every bundle admits a section. For example, consider the nontrivial double
cover of S 1 .
 1.2. FIBRE BUNDLES 11

Definition 1.2.5. Let π : E → B be a fibre bundle with fibre F . Given a

continuous map φ : B 0 → B, we can form the pullback bundle π 0 : φ∗ E → B 0 by
setting
φ∗ E := {(b0 , e) ∈ B 0 × E : φ(b0 ) = π(e)}
and π 0 (b0 , e) := b0 . Then this is also a fibre bundle with fibre F . Furthermore, if
we set ε(b0 , e) := e, then (ε, φ) is a bundle morphism; i.e., the following diagram is
commutative:
φ∗ E
ε /E

π0 π
 
B0
φ
/ B.
If s is a section of π : E → B, then we can define its pullback as
φ∗ s(b0 ) := (b0 , s(φ(b0 ))).
Recall that a topological group is a group G that is endowed with a topology
such that the product G × G → G and the inverse G → G are both continuous.
π
Definition 1.2.6. Let F → E → B be a fibre bundle, and let G be a topolog-
ical group admitting a left homeomorphism action on F . A G-atlas on the bundle
is a set of local trivialisations
{(Ui , hi : π −1 (Ui ) → Ui × F ) : i ∈ I},
where I is an index set and {Ui : i ∈ I} is an open cover of B, such that the
transition maps
hj ◦ h−1
i : (Ui ∩ Uj ) × F → (Ui ∩ Uj ) × F
−1
are of the form hj ◦ hi (b, x) = (b, tij (b)x) for continuous transition functions
tij : Ui ∩ Uj → G.
Two G-atlases are equivalent if their union is also a G-atlas. A G-structure on
a fibre bundle is an equivalence class of G-atlases.
The transition functions satisfy the following conditions:
(1) tii ≡ 1,
(2) tij = t−1
ji , and
(3) tij tjk = tik .
Hence, the transition functions form a Čech 1-cocycle.
Closely related to the notion of G-bundles are principal bundles.
Definition 1.2.7. Let G be topological group. Then a principal G-bundle
is a fibre bundle π : P → B such that P admits a continuous right action of G
that preserves the fibres, and acts by homeomorphisms on each fibre freely and
transitively.
In particular, the fibre of a principal G-bundle is homeomorphic to G, though
not canonically. The homeomorphism becomes canonical once we fix the identity
element of the fibre. A principal bundle is trivial if and only if it admits a section.
Suppose that π : P → B is a principal G-bundle, and that G acts on a topo-
logical space F on the left by homeomorphisms. Then we can form the associated
bundle
P ×G F := P × F/(p,f )∼(pg,g−1 f ) ,
which is a fibre bundle with a G-structure.
 1.3. SMOOTH MANIFOLDS 12

Conversely, given a fibre bundle π : E → B with a G-structure, we can associate

to it a principal G-bundle over B by using the transition functions tij to glue Ui ×G
and Uj × G.
Definition 1.2.8. Let F be either R or C. A rank n vector bundle over F is a
π
fibre bundle Fn → E → B such that
(1) π −1 (b) has the structure of an F-vector space for every b ∈ B;
(2) every point of B has a neighbourhood U and a local trivialisation
h : π −1 (U ) → U × Fn
such that
h|π−1 ({b}) : π −1 ({b}) → {b} × Fn
is a linear isomorphism for every b ∈ U .
Equivalently, a real vector bundle is a fibre bundle with fibre Rn and a GL(n, R)-
structure. One can construct the associated principal GL(n, R)-bundle by consid-
ering the bundle of n-frames in the fibres of the vector bundle. Similarly, a complex
vector bundle is associated to the principal GL(n, C)-bundle of complex n-frames
in its fibres.
A fundamental operation on vector bundles is the direct sum or Whitney sum
operation. Given vector bundles ξ and η over the same base space B, we can define
their direct sum ξ ⊕ η as a vector bundle over B whose fibre over b ∈ B is ξb ⊕ ηb .
Its precise definition is the following:
Definition 1.2.9. Let ξ and η be vector bundles over B. Then
ξ ⊕ η := ∆∗ (ξ × η),
where ∆ : B → B × B is the diagonal map, and ξ × η is the product bundle over
B × B.
Given a real rank n vector bundle ξ, its structure group is GL(n, R). Fixing a
Riemannian metric on ξ; i.e., a positive definite, symmetric bilinear form on each
fibre of ξ that varies continuously, reduces the structure group to O(n). Indeed,
the associated principal O(n)-bundle consists of orthonormal n-frames in the fibres
of ξ. Every vector bundle admits a Riemannian metric, and so we can reduce its
structure group to O(n). Note that O(n) is a deformation retract of GL(n, R) by
the Gram–Schmidt process.
Fixing a consistent orientation on the fibres of ξ reduces the structure group
to GL+ (n, R). If we are also given a Riemannian metric and an orientation, the
structure group is SO(n). Recall that SO(2) ≈ S 1 and π1 (SO(n)) ∼ = Z2 for n > 2.
Definition 1.2.10. For n ≥ 2, the spin group Spin(n) is the connected double
cover of SO(n). Let ξ be an oriented rank n vector bundle over B, and write
P (ξ) for the associated principal SO(n)-bundle. A spin structure on ξ consists of
a principal Spin(n)-bundle P̃ → B, together with a bundle morphism P̃ → P (ξ)
that is fibrewise the nontrivial double cover Spin(n) → SO(n).

1.3. Smooth manifolds

We say that a function from Rn to Rm is smooth or C ∞ if it is infinitely
differentiable; i.e., its coordinate functions have continuous partial derivatives of
arbitrarily high order. Smooth manifolds are topological manifolds with some extra
 1.3. SMOOTH MANIFOLDS 13

structure that allows one to define smooth maps between them. We can think
of a smooth n-manifold as being glued together from open subsets of Rn such
that the gluing maps are all smooth with a smooth inverse (i.e., diffeomorphisms).
Alternatively, we can think of it as a topological manifold with a set of local charts
that form the pages of an atlas, and when converting coordinates from one chart to
the other, we apply a smooth map. For example, a usual atlas of the surface of the
Earth, which is topologically S 2 , consists of pages each showing a planar region.
Definition 1.3.1. Let M be a topological n-manifold. A chart on M is a
homeomorphism φ from an open subset U of M to an open subset S of Rn . A smooth
atlas of M is a set A of charts {(Ui , φi ) : i ∈ I} such that M = i∈I Ui and the
transition function
φj ◦ φi−1 : φi (Ui ∩ Uj ) → φj (Ui ∩ Uj )
is smooth for every i, j ∈ I. We say that the smooth atlases A and A0 are equivalent
if A ∪ A0 is also a smooth atlas. A smooth structure on the topological manifold M
is an equivalence class of smooth atlases.
Remark 1.3.2. In the above definition, we could replace the word “smooth”
with “C k -differentiable” to define C k -differentiable manifolds. However, there is
a bijection between C k -differentiable and smooth structures on any topological
manifold for k ≥ 1, so we will only study smooth manifolds. We will use the terms
“local coordinates”and “chart” interchangeably.
If we require the transition functions to be analytic, we obtain the class of
real analytic manifolds. If we are given charts to Ck with biholomorphic transition
functions, we obtain complex manifolds. A 1-dimensional complex manifold is called
a Riemann surface.
If the transition maps are piecewise linear, we obtain the notion of PL man-
ifolds. A triangulation of a topological manifold is a homeomorphism with the
topological realisation of a simplicial complex. Recall that the link of a vertex v of
a triangulation is the boundary of the union of closed simplices incident to v. A PL
structure corresponds to a triangulation where the link of every vertex is a sphere.
Every smooth manifold can be triangulated; see Munkres [51]. However, not
every triangulable manifold admits a smooth structure, even if we assume that the
boundary of the link of each vertex is a sphere.
Definition 1.3.3. Let M and N be smooth manifolds with smooth atlases
{(Ui , φi ) : i ∈ I} and {(Vj , ψj ) : j ∈ J }. We say that the map f : M → N is C r
for r ∈ N ∪ {∞} if the map
ψj ◦ f ◦ φ−1
i : φi (Ui ∩ f
−1
(Vj )) → ψj (f (Ui ) ∩ Vj )
is r-times continuously differentiable for every i ∈ I and j ∈ J . (This does not
depend on the choice of atlases.) We denote the set of C r maps from M to N by
C r (M, N ). We will often call C ∞ maps smooth.
We can endow C r (M, N ) with several topologies, but we will not give the
precise definition. Intuitively, a neighbourhood of a function f consists of functions
whose partial derivatives up to order r are close to those of f in local coordinate
systems.
Smooth manifolds with smooth maps between them form a category. An equiv-
alence in this category is called a diffeomorphism:
 1.3. SMOOTH MANIFOLDS 14

Definition 1.3.4. The map φ : M → N is a diffeomorphism if it is a homeo-

morphism such that both φ and φ−1 are smooth.
Two diffeomorphisms φ0 , φ1 : M → N are pseudo-isotopic if there is a diffeo-
morphism Φ : M ×I → N ×I such that Φ(x, 0) = (φ0 (x), 0) and Φ(x, 1) = (φ1 (x), 1)
for every x ∈ M . If Φ is also level preserving; i.e., Φ(x, t) ∈ N ×{t} for every x ∈ M ,
then φ0 and φ1 are called isotopic.
Given an isotopy Φ, we will use the notation Φt (x) := Φ(x, t) for t ∈ I. Then
we can think of Φ as a smooth 1-parameter family of diffeomorphisms Φt connecting
φ0 and φ1 . In fact, if we endow the space of diffeomorphisms Diff(M, N ) from M to
N with the C ∞ topology, then two diffeomorphisms are isotopic if and only if they
lie in the same path component of Diff(M, N ). The proof of this involves deforming
a continuous path of diffeomorphisms to a smooth one, which we omit.
We now define tangent vectors and tangent and cotangent spaces.
Definition 1.3.5. Let M be a smooth n-manifold and p ∈ M a point. Given
smooth curves γ1 , γ2 : (−ε, ε) → M such that γ1 (0) = γ2 (0) = p, we say that γ1
and γ2 are equivalent if their velocity vectors (φ ◦ γ1 )0 (0) and (φ ◦ γ2 )0 (0) are the
same in any (and hence all) chart (U, φ) with p ∈ U . A tangent vector of M at p is
an equivalence class of such curves.
We denote by Tp M the set of all tangent vectors of M at p, and call it the
tangent space of M at p. This is a vector space, where the operations are the
natural ones given in a coordinate chart; e.g., [γ1 ] + [γ2 ] = [γ3 ] if
(φ ◦ γ1 )0 (0) + (φ ◦ γ2 )0 (0) = (φ ◦ γ3 )0 (0)
S
for some chart (U, φ) about p. Finally, we write T M = p∈M Tp M for the tangent
bundle of M . This is a vector bundle with base space M . A vector field on M is a
section of the tangent bundle T M .
The cotangent bundle T ∗ M of M is obtained by taking the union of the dual
spaces Tp∗ M = (Tp M )∗ for p ∈ M .
To define the topology on T M , we choose an atlas A = {(Ui , φi ) : i ∈ I} on M ,
and glue together the product bundles Ui × Rn for i ∈ I, as follows: For i, j ∈ I,
a point x ∈ Ui ∩ Uj , and a vector v ∈ Rn , we identify (x, v) ∈ Ui × Rn with
x, D(φj ◦ φ−1 n


i )(φi (x))(v) ∈ Uj × R ,

where D(φj ◦ φ−1 −1

i )(φi (x)) is the Jacobian of φj ◦ φi at φi (x).
A subset of a smooth n-manifold is called a submanifold if it is locally modelled
on (Rn , Rk ) for some k ≤ n:
Definition 1.3.6. Let M be a smooth n-manifold and k ≤ n. We say that
N ⊂ M is a smooth k-dimensional submanifold if there is a chart (U, φ) about each
p ∈ N such that φ(N ∩ U ) = φ(U ) ∩ Rk . The charts (N ∩ U, φ|N ∩U ) define an atlas
of N , which becomes a smooth k-manifold.
If N is a submanifold of M , then T N is a subbundle of the restriction T M |N .
The normal bundle of N is defined as the quotient
νN ⊂M := (T M |N )/T N.
Definition 1.3.7. Let S be a smooth submanifold of M . We write E for the
total space of the normal bundle νS⊂M and 0S : S → E for the zero-section. Then
 1.3. SMOOTH MANIFOLDS 15

we say that the open subset N (S) ⊂ M is a tubular or regular neighbourhood of S

if there is a diffeomorphism ϕ : E → N (S) such that ϕ ◦ 0S = IdS .
Proposition 1.3.8. Every smooth submanifold has a tubular neighbourhood.
We omit the proof.
Definition 1.3.9. Let N and N 0 be smooth submanifolds of the manifold
M . We say that N and N 0 are transverse (or intersect transversely) if at each
intersection point p ∈ N ∩ N 0 , we have Tp N + Tp N 0 = Tp M .
If N and N 0 are transverse, then N ∩ N 0 is a smooth manifold. Furthermore,
dim(N ∩ N 0 ) = dim(N ) + dim(N 0 ) − dim(M ),
which follows from the observation that Tp (N ∩ N 0 ) = Tp N ∩ Tp N 0 for p ∈ N ∩ N 0 ,
and the dimension formula for the intersection of two linear subspaces of a vector
space.
Definition 1.3.10. If N and N 0 are smooth submanifolds of M , then they are
said to be ambient isotopic (or simply isotopic) if there is an isotopy Φ : M × I →
M × I such that Φ0 = IdM and Φ1 (N ) = N 0 .
Definition 1.3.11. Let f : M → N be a smooth map. The differential of f at
p ∈ M is a linear map dfp : Tp M → Tf (p) N , defined as follows: Let γ be a curve
representing a vector v ∈ Tp M . Then dfp (v) is the tangent vector of N at f (p)
represented by f ◦ γ. The maps dfp for p ∈ M assemble to a morphism of vector
bundles df : T M → T N .
Let C ∞ (M ) denote the vector space of smooth functions f : M → R. Given
a tangent vector v ∈ Tp M represented by a curve γ : (−ε, ε) → M and a smooth
function f ∈ C ∞ (M ), we can define the directional derivative vf as (f ◦ γ)0 (0).
This is independent of the choice of representative γ. The differential df of the
function f is a section of the cotangent bundle T ∗ M . By definition, df (v) := vf .
A smooth n-manifold is orientable if and only if it admits an atlas where all the
transition maps are orientation preserving, in the sense that they have differentials
in GL+ (n).
Let us write Cp∞ (M ) for the vector space of germs of smooth functions on M
at p. This consists of equivalence classes of smooth functions on M defined in a
neighbourhood of p such that f ∼ g if there exists an open set U in M containing
p such that f |U = g|U . A derivation D at p ∈ M is a linear functional on Cp∞ (M )
that satisfies the Leibniz rule; i.e.,
D(f g) = D(f )g + f D(g).
Then there is a bijection between Tp M and the space of derivations at p. Hence,
alternatively, we could have defined tangent vectors as derivations. Vector fields
correspond to linear transformations on C ∞ (M ) that satisfy the Leibniz rule. For
example, given a coordinate chart φ : U → Rn on M , we obtain the coordinate
vector fields ∂/∂xi on U for i ∈ {1, . . . , n}. For a function f ∈ C ∞ (U, R), we let
(∂/∂xi )(f ) := ∂(f ◦ φ−1 )/∂xi .
In Section 1.2, we defined the notion of Riemannian metric on a vector bundle.
A Riemannian metric on a smooth manifold M is simply a Riemannian metric on
its tangent bundle T M :
 1.3. SMOOTH MANIFOLDS 16

Definition 1.3.12. A Riemannian metric g on a smooth manifold M is a

positive definite, symmetric bilinear form gp on Tp M for each p ∈ M that varies
smoothly in the sense that, for every coordinate chart φ : U → Rn and i, j ∈
{1, . . . , n}, the functions gp (∂/∂xi , ∂/∂xj ) are smooth in p ∈ U .
Differential topology is the study of the category of smooth manifolds and
smooth maps between them. One of the main questions is the classification of
smooth structures on topological manifolds.
Theorem 1.3.13. Every topological manifold of dimension at most 3 has a
unique smooth structure, up to diffeomorphism.
Proof. Moise [48, 47] showed that every 2- and 3-manifold admits a unique
PL structure; see also Hamilton [20]. For the uniqueness of smoothing of the PL
structure, see the books of Hirsch and Mazur [25] and Thurston [69]. 

However, things suddenly change in dimension 4. There are topological man-

ifolds that admit no smooth structure, and there are some that admit infinitely
many. This will be the subject of Chapter 4.
Surprisingly, things get somewhat easier in dimensions 5 and higher, at least
once one restricts the fundamental group, and considers only say simply-connected
manifolds. This is because of the h-cobordism theorem that we will discuss in
Section 1.9. In the heart of the proof of the h-cobordism theorem lies the Whit-
ney trick, which is a geometric operation aimed at cancelling pairs of intersection
points between two submanifolds of complementary dimensions, and hence realiz-
ing algebraic intersection numbers as geometric intersection numbers. In particular,
Kervaire and Milnor [32] determined the set of smooth structures on all spheres of
dimension at least 5. There is a unique smooth structure on S 5 and S 6 , but there
are 28 on S 7 !
Given a smooth manifold M , a non-diffeomorphic smooth structure on M is
called exotic. It is not known whether there is an exotic smooth structure on S 4 .
This is called the smooth 4-dimensional Poincaré conjecture.
We have already encountered the connected sum operation for topological man-
ifolds. The corresponding operation in the smooth category is more subtle:
Definition 1.3.14. We define the connected sum of two smooth, oriented,
connected n-manifolds M1 and M2 as follows: Choose embeddings ei : Dn ,→ Mi
for i ∈ {1, 2} such that e1 is orientation-preserving and e2 is orientation-reversing.
We obtain M1 #M2 from the disjoint union
(M1 \ {e1 (0)}) t (M2 \ {e2 (0)})
by identifying e1 (tv) with e2 ((1 − t)v) for each unit vector v ∈ S n−1 and t ∈ (0, 1).
We orient M1 #M2 compatibly with M1 and M2 .
This is independent of the choice of embeddings e1 and e2 up to diffeomorphism,
by the work of Palais [54] and Cerf [6], who showed that, for any two embeddings
e, e0 : Dn → M , there is a diffeomorphism f : M → M such that e0 = f ◦ e.
Furthermore, the connected sum operation is commutative and associative up to
diffeomorphism, and S n is an identity element.
Boundary connected sum is a closely related operation for manifolds with
boundary:
 1.4. EMBEDDINGS, IMMERSIONS, AND SUBMERSIONS 17

Definition 1.3.15. Let W1 and W2 be oriented (n + 1)-manifolds with con-

nected boundary. Let H n+1 := Dn+1 ∩ Rn+1 + be a half-disk. Choose embed-
dings ei : (H n+1 , Dn ) ,→ (Wi , ∂Wi ) for i ∈ {1, 2} such that e2 ◦ e−1
1 is orientation-
reversing. We obtain the boundary connected sum W1 \W2 from
(W1 \ {e1 (0)}) t (W2 \ {e2 (0)})
by identifying e1 (tv) with e2 ((1 − t)v) for every v ∈ S n ∩ Rn+1
+ and t ∈ (0, 1).
Note that
∂(W1 \W2 ) = ∂W1 #∂W2 .

1.4. Embeddings, immersions, and submersions

Manifolds were first considered as subsets of some Euclidean space Ra , and
defined as common zero sets of a collection of smooth functions. In this section, we
will show that every abstract smooth manifold defined using atlases indeed embeds
into some Ra .
Definition 1.4.1. A smooth map f : M → N is an immersion if its differential
dfp : Tp M → Tf (p) N is injective for every p ∈ M .
We say that f is an embedding if it is an injective immersion that is a homeo-
morphism onto its image.
It is customary to use the notation f : M # N for immersions. An immersion
f is locally an embedding, and its image is locally a submanifold of N . However, it
might have self-intersections; i.e., distinct points p, q ∈ M such that f (p) = f (q).
For example, consider a map from S 1 to R2 parametrising a figure eight curve in
the plane.
Not every injective immersion is an embedding: The map f : R → R2 /Z2
given by f (t) = tv for v = (1, s) and s ∈ R \ Q has dense image in the torus
T 2 = R2 /Z2 . So f is not a homeomorphism onto its image when that is endowed
with the subspace topology. However, every injective immersion of a compact
manifold is an embedding. If f is an embedding, then f (M ) is a submanifold of N .
Definition 1.4.2. We say that the embeddings f0 , f1 : M → N are isotopic if
there is a smooth map F : M × I → N such that Ft := F (−, t) is an embedding for
every t ∈ I, and Fi = fi for i ∈ {0, 1}. We call F an isotopy.
We can think of an isotopy as a path in the subspace Emb(M, N ) of embed-
dings of M into N , considered as a subspace of C ∞ (M, N ). In Definition 1.3.10, we
introduced the notion of ambient isotopy for two submanifolds. There is an anal-
ogous notion for embeddings. According to the isotopy extension theorem, every
isotopy can be extended to an ambient isotopy:
Theorem 1.4.3 (Isotopy extension theorem). Let M be a compact submaifold of
N . Suppose that F : M ×I → N is an isotopy, and let us write Ft := F (−, t). Then
there is a smooth map G : N × I → N such that Gt := G(−, t) is a diffeomorphism
and
Gt ◦ F0 = Ft
for every t ∈ I.
 1.4. EMBEDDINGS, IMMERSIONS, AND SUBMERSIONS 18

Proof. We obtain G by integrating a vector field v on N × I. This will be of

the form vN + ∂/∂t, where vN is tangent to the N -direction. We first define vN
along the submanifold
L := {(ft (x), t) : (x, t) ∈ M × I}
of N × I. For (x, t) ∈ M × I, we let vN (ft (x), t) be the velocity vector of the curve
γx (t) := Ft (x) at t. Let N (L) be a tubular neighbourhood of L in N × I. We then
extend vN to N (L) and multiply it with a fibrewise bump function on N (L) that
is 1 along L and 0 on a neighbourhood of (N × I) \ N (L). We finally let vN = 0
on (N × I) \ N (L).
By construction, v is tangent to the curves t 7→ (γx (t), t) for (x, t) ∈ M × I.
Given (x, t) ∈ N × I, we obtain (G(x, t), t) by following the flow of v starting from
(x, 0) until we reach N × {t}. 

As the space of vector fields v tangent to F ({x} × I) for every x ∈ M and of

the form vN + ∂/∂t in the proof of Theorem 1.4.3 is convex and hence contractible,
the diffeomorphism (G(x, t), t) of N × I is unique up to isotopy.
Proposition 1.4.4. For every smooth n-manifold M , there is an embedding
f : M ,→ Ra
for some a ∈ N.
Proof. We only give the proof when M is compact. Then there are charts
ϕi : Ui → Rn for i ∈ {1, . . . , k} such that each Ui is diffeomorphic to the ball B n ,
and there are concentric balls Ui00 ( Ui0 ( Ui with M = ∪ki=1 Ui00 . There are smooth
functions µi : M → I and λi : M → I such that
• µi |Ui0 ≡ 1 and µi |M \Ui ≡ 0, and
• λi |Ui00 ≡ 1 and λi |M \Ui0 ≡ 0.
Then the maps
ψi := µi ϕi : M → Rn
are smooth for i ∈ {1, . . . , k}. Furthermore, if we set ψ = (ψ1 , . . . , ψk ), then
ψ : M → Rnk is an immersion. Indeed, for p ∈ M , there is an i ∈ {1, . . . , k} such
that p ∈ Ui00 , and (dψi )p has rank n.
To lift ψ to an embedding, let Λ := (λ1 , . . . , λk ), and set
f := (ψ, Λ) : M → Rnk × Rk .
Suppose that f (x) = f (y) for x 6= y ∈ M . Then x ∈ Ui00 for some i ∈ {1, . . . , k}.
As ψi (x) = ψi (y), and since ψi = µi ϕi = ϕi is injective in Ui0 , we have y 6∈ Ui0 . But
then λi (y) = 0 and λi (x) = 1, which is a contradiction. 

Definition 1.4.5. Let f : M → N be a smooth map. We say that p ∈ M is a

regular point of f if rk(dfp ) = dim(N ), and is a critical point otherwise. The point
q ∈ N is a regular value of f if every point of f −1 (q) is regular, and is a critical
value otherwise. The map f is a submersion if every p ∈ M is regular.
If q is a regular value, then f −1 (q) is a submanifold of M by the implicit
function theorem.
 1.4. EMBEDDINGS, IMMERSIONS, AND SUBMERSIONS 19

Exercise 1.4.6 (Ehresmann’s fibration lemma). Let f : M → N be a submer-

sion. Show that if f is proper; i.e., f −1 (K) is compact for every K ⊆ N compact,
then M is a fibre bundle over N with fibre f −1 ({q}) for q ∈ N . Give a counterex-
ample when f is not proper.
A subset of a smooth manifold is said to be measure zero if its intersection with
any coordinate chart has measure zero. This is well-defined since diffeomorphisms
map measure zero sets to measure zero sets, and change of coordinate maps are
diffeomorphisms. The following result of Sard plays an important role in differential
topology.
Theorem 1.4.7 (Sard’s theorem). Let f : M → N be a smooth map. Then the
set of critical values of f forms a measure zero subset of N .
For a proof, see the book of Milnor [44]. As a special case, when dim(M ) <
dim(N ), we obtain that Im(f ) has measure zero in N .
It is an important question in differential topology what the smallest a is such
that we can embed or immerse a given smooth n-manifold into Ra .
Theorem 1.4.8. Every smooth n-manifold M can be embedded into R2n+1 and
immersed into R2n .
Proof. Again, we only prove the case when M is compact. By Proposi-
tion 1.4.4, there is an embedding f : M ,→ Ra for some a ∈ N. For v ∈ S a−1 , let
pv be orthogonal projection onto v ⊥ . Our goal is to find a direction v ∈ S a−1 such
that pv ◦ f is also an embedding or immersion.
For pv ◦ f to be an immersion, d(pv ◦ f ) has to have rank n everywhere. To
ensure this, v should not be tangent to Im(f ). This means v 6= df (w)/|df (w)| for w
a non-zero tangent vector of M . The unit tangent bundle ST M := (T M \ 0M )/R∗
of M is (2n − 1)-dimensional, as it has fibre S n−1 , where 0M is the zero section.
Hence, by applying Sard’s theorem to df /|df | : ST M → S a−1 , if a > 2n, for v
outside a measure zero subset of S a−1 , the map pv ◦ f is an immersion. Repeatedly
projecting in this way, we obtain an immersion of M into R2n .
To make sure pv ◦ f is also injective, consider the map ϕ : (M × M ) \ ∆ → S a−1
given by
f (x) − f (y)
ϕ(x, y) =
|f (x) − f (y)|
for x 6= y ∈ M , where ∆ = {(x, x) : x ∈ M } is the diagonal. By Sard’s theorem,
the image of ϕ has measure zero in S a−1 whenever a − 1 > dim(M × M ) = 2n. If
v ∈ S a−1 \ Im(ϕ), then pv ◦ f is injective. Furthermore, if v also avoids the image of
ST M , the map pv ◦ f is also an immersion, and is hence an embedding. Repeating
this process, we obtain an embedding of M into R2n+1 . 

The above proof can be modified to show that, in fact, embeddings of an n-

manifold into R2n+1 are dense in the space of all smooth maps, which we endow
with the weak C ∞ topology. Similarly, the set of immersions into R2n is also dense.
In other words, any smooth map of an n-manifold into R2n+1 can be perturbed into
an embedding, and every smooth map into R2n can be perturbed into an immersion.
Using the Whitney trick (Section 1.8), Theorem 1.4.8 can be improved, and
every n-manifold embeds into R2n and, when n > 1, immerses into R2n−1 . However,
the subsets of embeddings and immersions are no longer dense in these dimensions.
 1.5. MORSE THEORY 20

Given a manifold, it is a hard problem to determine the smallest a such that it

embeds or immerses into Ra . The immersion conjecture states that if α(n) is the
number of ones in the binary expansion of n, then every n-manifold immerses into
R2n−α(n) . This has been shown by Cohen [8].

1.5. Morse theory

We can learn a lot about smooth manifolds by studying critical points of certain
nice generic functions on them, which are called Morse. We write C ∞ (M ) :=
C ∞ (M, R), and let f ∈ C ∞ (M ). By Definition 1.4.5, p ∈ M is a critical point of f
if dfp = 0; i.e., if vf = 0 for every v ∈ Tp M . Furthermore, c ∈ R is a critical value
of f if there is a critical point p of f with f (p) = c, and is a regular value otherwise.
Definition 1.5.1. Let f ∈ C ∞ (M ) be a smooth function on the n-manifold
M . We say that the critical point p is non-degenerate if, in a local coordinate
system (x1 , . . . , xn ) about p, the Hessian
 2 
∂ f
(0)
∂xi ∂xj i,j=1,...,n

is non-degenerate. The function f is called Morse if all of its critical points are
non-degenerate.
We will denote the set of critical points of a smooth function f by Crit(f ). The
non-degeneracy condition is equivalent to requiring that the section df : M → T ∗ M
is transverse to the 0-section at p. There exists a local coordinate system about
each non-degenerate critical point in which the function has a particularly nice
form. More specifically, if (x1 , . . . , xn ) are local coordinates, then 0 is a critical
∂f
point of f if and only if ∂xi
(0) = 0 for i ∈ {1, . . . , n}. Nondegeneracy of the critical
point amounts to nondegeneracy of the PnHessian, which is a real symmetric bilinear
form, and hence is diagonalisable as i=1 ai x2i for ai ∈ {1, −1} by Sylvester’s law
of inertia. The Morse lemma states that there is a local coordinate system about a
nondegenerate critical point in which the Hessian is diagonal with entries ±1, and
all higher order terms of the Taylor series vanish.
Lemma 1.5.2 (Morse Lemma). Let p be a non-degenerate critical point of f ∈
C ∞ (M ). Then there are local coordinates (x1 , . . . , xn ) about p such that
f (x1 , . . . , xn ) = f (0) − x21 − · · · − x2i + x2i+1 + · · · + x2n
for some i ∈ {0, . . . , n}.
We call i the index of the critical point p and denote the set of index i critical
points of f by Criti (f ).
Proof (Non-examinable). By choosing a chart about p, we can assume that
f is defined in a convex neighbourhood U of 0 ∈ Rn . Furthermore, by replacing f
with f − f (0), we can assume that f (0) = 0. Then
Z 1 Z 1Xn n Z 1
df ∂f X ∂f
f (x) = (tx)dt = (tx)xi dt = xi (tx)dt,
0 dt 0 i=1 ∂xi i=1 0 ∂xi

where we used the fundamental theorem of calculus and f (0) = 0 in the first step.
R 1 ∂f
Repeating the same computation with gi (x) := 0 ∂x i
(tx)dt in place of f (x), and
 1.5. MORSE THEORY 21

∂f
noting that gi (0) = ∂xi (0) = 0, we obtain functions Hi,j such that
n
X
(1.5.1) f (x) = xi xj Hi,j (x).
i,j=1

We can further assume that Hi,j = Hj,i by replacing Hi,j with 12 (Hi,j +Hj,i ). Then
2
Hi,j (0) = 21 ∂x∂i ∂x
f
j
(0).
We now adapt the proof of Sylvester’s law of inertia to find the desired coor-
dinate system. Suppose that there is a coordinate system (u1 , . . . , un ) about 0 in
which
X n
X
2
f (u) = ±ui + ui uj Hi,j (u),
i<r i,j=r
and Hi,j = Hj,i . This is true for r = 1 by equation (1.5.1), and the main result
follows from this by induction on r, as follows.
By non-degeneracy of the Hessian of f , there are i, j ≥ r such that Hi,j (0) 6= 0.
If i 6= j, then let yi = ui + uj , yj = ui − uj , and yk = uk for k 6∈ {i, j}, giving
1 2
(y − yj2 )Hi,j .
2ui uj Hi,j =
2 i
Hence, by possibly preindexing the coordinates, we can assume that Hr,r (0) 6= 0.
Write g(u) := |Hr,r (u)|. This is a smooth, positive function in some neigh-
bourhood U 0 of 0 in U . We now set
!
X
vr (u) := g(u) ur + ui Hi,r (u)/Hr,r (u) ,
i>r

∂v
and vi := ui for i 6= r. Then the Jacobian ∂u (0) 6= 0, so u 7→ v is a change
00 0
of coordinates in a neighbourhood U ⊂ U of 0 by the inverse function theorem.
Furthermore, X X
0
f (v) = ±vi2 + vi vj Hi,j (v)
i≤r i,j>r
for v ∈ U 00 . 
Corollary 1.5.3. A non-degenerate critical point is always isolated.
A subset of a topological space is called residual if it is an intersection of
countably many open dense subsets. We say that a property of elements of a
topological space is generic if it holds for a residual subset.
Theorem 1.5.4. Morse functions form a residual subset of C ∞ (M ).
For a proof, see the book of Milnor [43]. Since very residual set is dense,
every smooth manifold admits a Morse function, and every smooth function can be
perturbed to a Morse function.
Now suppose that f is a Morse function such that different critical points have
distinct values. This is a generic condition. If c is a regular value of f , then f −1 (c) is
a smooth submanifold of M by the implicit function theorem. This is the boundary
of M c := f −1 ((−∞, c]). As c increases, both f −1 (c) and M c change smoothly as
long as c does not pass a critical value. The key observation of Morse theory is that
if c passes a point of f (Criti (f )), the manifold M c changes by attaching a thickened
 1.5. MORSE THEORY 22

xi+1 , . . . , xn f −1 (c + ε)
f −1 (c − ε)

x1 , . . . , x i

Figure 1. This figure shows the level sets f −1 (c − ε), f (c), and
f −1 (c + ε) of the function f in Lemma 1.5.2, where c = f (0).
The shaded region is M c−ε union the i-handle Di × Dn−i , which
becomes diffeomorphic to M c+ε after smoothing the corners.

i-cell, called an n-dimensional i-handle. This is Di × Dn−i , which we attach along

an embedding
h : ∂Di × Dn−i ,→ ∂M c = f −1 (c)
and smooth the resulting corners. This follows from a simple local analysis of the
normal form of f in Lemma 1.5.2. Indeed, if c = f (0) and ε > 0 is small, then –
up to diffeomorphism – we obtain M c+ε from M c−ε by gluing the i-handle
N ({ x ∈ Rn : xi+1 = · · · = xn = 0 }) \ M c−ε
to M c−ε ; see Figure 1. We will study handle decompositions of manifolds in more
detail in Section 1.6.
In particular, if we are only interested in the homotopy type of M c , then it
changes by attaching an i-cell. Successively attaching such cells, we obtain a cell
complex homotopy equivalent to M . Indeed, by cellular approximation, we can
homotope the attaching map of the i-cell into the (i − 1)-skeleton of the complex
corresponding to M c−ε . Hence, using cellular homology, we obtain lower bounds
on the Betti numbers of M : If ci is the number of critical points of f of index i,
then
bi (M ) ≤ ci
for i ∈ {0, . . . , n}. These are called the weak Morse inequalities. Furthermore,
n
X
χ(M ) = (−1)n ci .
i=0
Let v be the gradient of f with respect to the metric g, which is defined by
the equation g(v, w) = wf for all vector fields w on M . If c, c0 ∈ R satisfy c < c0
and f (Crit(f )) ∩ [c, c0 ] = ∅, then the flow of v/vf gives a diffeomorphism between
f −1 ([c, c0 ]) and f −1 (c) × I.
If p is an index i critical point of f , choose local coordinates (x1 , . . . , xn ) about
p as in Lemma 1.5.2. Assume that g is the usual Euclidean metric in these coordi-
nates; i.e., g(∂/∂xi , ∂/∂xj ) = δi,j . Then, in this chart,
v = 2(−x1 , . . . , −xi , xi+1 , . . . , xn ).
We see that the stable manifold Ds (v, p) of v at p – the set of points of M on flow
lines of v converging to p – is an i-disk. Similarly, the unstable manifold Du (v, p)
 1.6. HANDLE DECOMPOSITIONS AND SURGERY 23

of v at p is a disk of dimension n − i. If c = f (p), the manifold M c+ε is obtained

from M c−ε by attaching a neighbourhood of the i-disk Ds (v, p) \ M c−ε .
Remark 1.5.5. It is often more convenient to work with gradient-like vector
fields instead of Riemannian metrics. Given a Morse function f on a smooth n-
manifold M , a gradient-like vector field for f is a vector field v such that vf (p) > 0
whenever p ∈ M \ Crit(f ); furthermore, there are local coordinates (x1 , . . . , xn )
about each critical point p in which
f (x1 , . . . , xn ) = f (p) − x21 − · · · − x2i + x2i+1 + · · · + x2n ,
and the vector field is the Euclidean gradient:
 
∂ ∂ ∂ ∂
v = 2 −x1 − · · · − xi + xi+1 + · · · + xn .
∂x1 ∂xi ∂xi+1 ∂xn
The advantage of working with gradient-like vector fields is that they are easier to
manipulate and deform than Riemannian metrics.
We conclude this section with the following application of Morse theory, due
to Reeb:
Theorem 1.5.6. Let M be a smooth, closed n-manifold that admits a Morse
function f with only two critical points. Then M is homeomorphic to S n .
Proof. Since f has only two critical points, one has to be the global minimum,
which has index 0, and the other the global maximum, which has index n. Hence
f can be obtained by gluing an n-handle hn = Dn to a 0-handle h0 = Dn along a
diffeomorphism φ : ∂hn → ∂h0 . Consider the hemispheres S+ n
:= S n ∩ {xn+1 ≥ 0}
n n n
and S− := S ∩ {xn+1 ≤ 0} and the points p± := S ∩ {xn+1 = ±1}. We define
the homeomorphism H : M → S n as follows: For x ∈ h0 , let
 p 
H(x) = x, − 1 − |x|2 .
We extend this to hn radially via
 p 
n
H(x) := |x|φ(x/|x|), 1 − |x|2 ∈ S+ . 

Remark 1.5.7. First, note that the above homeomorphism H is not smooth at
the centre of the n-handle. The gluing map of the two handles is determined by the
flow of a gradient vector field v on M between the minimum and the maximum.
The smooth structure on an n-manifold obtained by gluing two n-disks depends
on the isotopy class of the gluing map. This gluing map is called the clutching
function. Kervaire and Milnor showed that there are 28 different smooth structures
on S 7 , all arising from this construction of gluing two copies of D7 .
For a detailed account of Morse theory, see the excellent book of Milnor [43].
Morse homology is developed in the book of Schwarz [63]. Above, we just gave an
outline of the construction, and have omitted several deep analytical proofs.

1.6. Handle decompositions and surgery

When studying homotopy types of topological spaces, cell complexes play a
fundamental role. These are built from cells of varying dimensions. In order to
study n-manifolds, we instead study handle decompositions, which are obtained by
thickening cells so that they all become n-dimensional.
 1.6. HANDLE DECOMPOSITIONS AND SURGERY 24

Definition 1.6.1. An n-dimensional k-handle hk is Dk × Dn−k , which we can

think of as a thickened k-cell. We call k the index of the handle. The core of hk is
Dk × {0}, the co-core is {0} × Dn−k , the attaching sphere is A(hk ) := S k−1 × {0},
and the belt sphere is B(hk ) := {0} × S n−k−1 .
Given an n-manifold M with boundary and an embedding
ϕ : S k−1 × Dn−k ,→ ∂M,
called the attaching map, we can attach the k-handle Dk × Dn−k to M along ϕ as
follows: We take the disjoint union M t (Dk × Dn−k ), identify x ∈ S k−1 × Dn−k
with ϕ(x) ∈ ∂M , and round the corners. The result is unique up to diffeomorphism,
and we denote it by M (ϕ).
To specify ϕ, we need to define an embedding of the attaching sphere S k−1 ×{0},
together with a normal framing; i.e., an identification of a neighbourhood of the
image with S k−1 × Dn−k . So the normal bundle of the attaching sphere has to
be trivial, and we also need to specify a trivialization of the normal bundle (the
framing).
When attaching a handle, ∂M changes by removing the image of ϕ, and gluing
in Dk × S n−k−1 using ϕ|S k−1 ×S n−k−1 . This leads to the notion of surgery:
Definition 1.6.2. Let X be a smooth n-manifold, and S ⊂ X an embedded
k-sphere with trivial normal bundle and normal framing ν : S k × Dn−k → N (S).
Then the result of surgery on X along the framed sphere (S, ν) is
X(S, ν) := (X \ N (S)) ∪ν|Sk ×Sn−k−1 (Dk+1 × S n−k−1 ).

More generally, one could glue using any automorphism of S k × S n−k−1 , which
will lead to the notion of Dehn surgery in dimension 3. Sometimes we will only
write X(ν) instead of X(S, ν).
If Φ ∈ Diff(X) is an automorphism of X, then it induces a diffeomorphism
Φ(ν) : X(ν) → X(Φ ◦ ν).
We saw in Section 1.5 that, when we pass a critical point of index i of a Morse
function, the sub-level set changes by attaching an i-handle. The attaching map is
given by the negative gradient flow from the critical point. Consequently, the level
set changes by a surgery.
Recall that a cell complex is obtained by taking a collection of 0-cells (a discrete
topological space), then attaching 1-cells, followed by 2-cells, etc. The analogous
construction for handles is called a handlebody:
Definition 1.6.3. An n-dimensional handlebody is obtained by taking finitely
many n-dimensional 0-handles, and recursively attaching 1-handles, followed by
2-handles, etc.
A handle decomposition of the smooth n-manifold M consists of a handlebody
H, together with a diffeomorphism φ : H → M .
Analogously, we can define relative handle decompositions, built on an n-
manifold with boundary. Furthermore, the resulting handlebody can be a manifold
with boundary.
Lemma 1.6.4. If we attach a k-handle hk followed by an l-handle hl such that
k ≥ l, then we can isotope the attaching map of the l handle to be disjoint from hk .
 1.6. HANDLE DECOMPOSITIONS AND SURGERY 25

Proof. Note that dim(A(hl )) = l − 1, dim(B(hk )) = n − k − 1, and

(l − 1) + (n − k − 1) ≤ n − 2.
So we can perturb the attaching map of hl such that its image is disjoint from
B(hk ). Then we can push off hl from hk by isotoping its attaching map radially
along the rays of the core of hk . 
Proposition 1.6.5. Every closed smooth manifold has a handle decomposition.
Proof. Let M be a closed smooth n-manifold. By Theorem 1.5.4, there is a
Morse function f : M → R, and we can arrange that f |Crit(f ) is injective. Consider
M c := f −1 ((−∞, c]) for c ∈ R. When c < min(f ), we have M c = ∅. We saw in
Section 1.5 that, if c ∈ f (Criti (f )), then M c+ε is obtained from M c−ε by attaching
an n-dimensional i-handle. It follows that M can be constructed by recursively
attaching handles. To arrange that the handles are attached such that their indices
are nondecreasing, we apply Lemma 1.6.4. 
Any two handle decompositions of a smooth manifold are related by a set of
elementary moves. However, we have to pass through handle decompositions where
the handles are not necessarily attached with indices in increasing order during the
intermediate steps.
The first move is an isotopy of the attaching map of one of the handles. We
modify the attaching maps of subsequently attached handles suitably as well. A
special case is a handle slide. In this case, one isotopes the attaching map of an
i-handle hi1 over the belt sphere of another i-handle hi2 such that, along the isotopy,
the image of A(hi1 ) intersects B(hi2 ) at a single point, and the manifold traced by
A(hi1 ) is transverse to B(hi2 ) at the intersection point. Using the language of Morse
theory, this corresponds to a flow line between two index i critical points. This
does not happen for a generic gradient-like vector field, but occurs in 1-parameter
families.
The second move is a handle creation or cancellation. Here, we add or remove
an i-handle hi and an (i + 1)-handle hi+1 , where hi and hi+1 are attached consecu-
tively, and |A(hi+1 ) ∩ B(hi )| = 1. This is possible since hi ∪ hi+1 is diffeomorphic to
a disk Dn , after smoothing corners. This is intuitively clear, but requires a careful
analysis that we omit.
Theorem 1.6.6. Any two handle decompositions of a closed smooth manifold
can be connected by isotopies of the attaching maps of the handles, and handle
creations and cancellations.
Sketch of proof. To prove that the above two moves are sufficient to con-
nect any two handle decompositions of the same manifold, we again use Morse
theory. We first construct Morse functions f and f 0 and gradient-like vector fields
v and v 0 , respectively, such that (f, v) and (f 0 , v 0 ) induce the two handle decompo-
sitions. We obtain these by gluing model functions and vector fields on each handle.
We then choose a generic 1-parameter family of smooth functions ft for t ∈ I, such
that f0 = f and f1 = f 0 . The only singularity appearing in such a family is of the
form
(1.6.1) ft (x1 , . . . , xn ) = c − x21 − · · · − x2i + x2i+1 + · · · + x2n−1 + x3n ± txn .
This is called a birth-death singularity, and is the suspension of the family x3 ± tx
for t ∈ I. The sign determines whether a pair of critical points are born or die.
 1.7. COBORDISMS 26

The critical points have indices i and i + 1, and there is a single gradient flow line
connecting them. So, if we choose vt to be the Euclidean gradient of ft in this local
coordinate system, the attaching sphere of the higher index handle will intersect
the belt sphere of the lower index one in a single point.
As t increases, the attaching spheres of the handles corresponding to the critical
points of ft change by isotopies, except when ft has a birth-death singularity, as in
equation 1.6.1, which corresponds to a handle creation or cancellation. 

1.7. Cobordisms
Cobordism provides a coarser notion of equivalence between manifolds than
homeomorphism or diffeomorphism. The idea stems from an attempt of Poincaré
to define homology, was formalised by Pontryagin, and further studied by Thom.
Definition 1.7.1. Let M and N be closed n-manifolds. A cobordism from M
to N is a compact (n + 1)-manifold with boundary W such that
∂W = M t N,
where t denotes “disjoint union.” If M and N are oriented, we say that W is an
oriented cobordism if W is also oriented and
∂W = −M t N.
Two manifolds are called cobordant if there is a cobordism between them.
Remark 1.7.2. If we did not require W to be compact, every manifold M
would be cobordant to ∅ via M × [0, ∞).
We can define cobordisms in both the topological and the smooth category.
Given cobordisms W from M0 to M1 and W 0 from M1 to M2 , we can define their
composition W 0 ◦ W by gluing them along M1 . In the smooth category, we need to
smooth the corners along M1 ; however, the result is unique up to diffeomorphism
fixing M0 and M2 . Hence cobordism is a transitive relation.
If W is an oriented cobordism from M to N , then
∂W = −M t N = −(−N ) t −M,
so we can also view W as a cobordism from −N to −M . We denote this by W ,
and call it the reverse of W . Note that this is not the same as −W , which is a
cobordism from N to M . Clearly, W = I × M provides a cobordism from M to
M . We have obtained the following:
Proposition 1.7.3. Cobordism is an equivalence relation.
In fact, cobordism classes of manifolds with the operation of disjoint union form
an abelian group, called the cobordism group of n-manifolds, where the identity el-
ement is the class of the empty n-manifold. (This class consists of the n-manfolds
that are boundaries of compact (n + 1)-manifolds.) Furthermore, in the oriented
cobordism group, the inverse of [M ] is [−M ], since we can view I × M as a cobor-
dism from M t −M to ∅. The unoriented cobordism groups are denoted by Nn ,
and the oriented ones by ΩSO n . These groups have been determined using the pi-
oneering work of René Thom [68]. For the purposes of low-dimensional topology,
 1.7. COBORDISMS 27

the following groups will be important:




 Z if n = 0,
0 if n = 1,



ΩSO ∼ 0 if n = 2,
n =

0 if

 n = 3,


Z if n = 4.
Exercise 1.7.4. Prove that ΩSO ∼
0 = Z.

More generally, cartesian product of manifolds endows ΩSO SO

L
∗ := i≥0 Ωi with
SO
a ring structure. Thom proved that Ω∗ ⊗ Q is a polynomial ring generated by the
cobordism classes of the complex projective spaces CP2i for i > 0. More generally,
two oriented manifolds are oriented cobordant if and only if they have the same
Stiefel–Whitney and Pontryagin numbers (see the book of Milnor and Stasheff [46]
for an overview of characteristic classes).
Two 4-manifolds are cobordant if and only if they have the same signature,
which is a numerical invariant that we now define in full generality. Let M be
a closed oriented 4k-manifold, with fundamental class [M ] ∈ H4k (M ). The cup
product defines a symmetric bilinear form
QM : H 2k (M ) ⊗ H 2k (M ) → Z
by the formula QM (x ⊗ y) = h x ∪ y, [M ] i, which is unimodular by Poincaré duality.
Then σ(M ) ∈ Z, the signature of M , is the signature of QM ; i.e., the dimension of
a maximal positive definite subspace minus the dimension of a maximal negative
definite subspace. René Thom showed that the signature is a cobordism invariant.
We say that two cobordisms from M0 to M1 are equivalent if they are diffeomor-
phic relative to M0 t M1 . The cobordism category Cobn has objects closed (smooth
and/or orientable) n-manifolds and morphisms equivalence classes of cobordisms
between them. The identity cobordism of M is the class of I × M . A topological
quantum field theory, or TQFT in short, is a certain functor from Cobn to the cat-
egory of vector spaces and linear maps that take disjoint unions to tensor products.
To be completely rigorous, we should define cobordisms as follows:
Definition 1.7.5. A cobordism from M0 to M1 as a 5-tuple (W, N0 , N1 , φ0 , φ1 ),
where ∂W = N0 t N1 (or −N0 t N1 if we are working with oriented manifolds), and
φi : Ni → Mi are diffeomorphisms or homeomorphisms (depending on the category)
for i ∈ {0, 1}.
Why this is necessary becomes clear when trying to construct the identity
cobordism from M to M . Using the less rigorous definition, this would be a manifold
W with ∂W = M t M , which is impossible unless M = ∅.
Definition 1.7.6. The identity cobordism from M to M is given by the tuple
(I × M, {0} × M, {1} × M, e0 , e1 ), where ei (i, x) = x for x ∈ M and i ∈ {0, 1}.
More generally, Definition 1.7.5 makes it very easy to associate a cylindrical
cobordism to a diffeomorphism:
Definition 1.7.7. The cylindrical cobordism Wψ associated to a diffeomor-
phism ψ : M → N is the tuple (I ×M, {0}×M, {1}×M, φ0 , φ1 ), where φ0 (0, x) = x
and φ1 (1, x) = ψ(x) for x ∈ M .
 1.8. THE WHITNEY TRICK 28

Definition 1.7.8. The cobordisms (W, N0 , N1 , φ0 , φ1 ) and (W 0 , N00 , N10 , φ00 , φ01 )
from M0 to M1 are equivalent if there is a diffeomorphism (or homeomorphism)
Φ : W → W 0 such that Φ|Ni = (φ0i )−1 ◦ φi for i ∈ {0, 1}.
Morphisms in the cobordism category are equivalence classes of cobordisms,
since the composition of two cobordisms is only well-defined up to equivalence due
to the smoothing involved. Throughout this work, we usually use the less rigorous
definition of cobordism, as it is usually straightforward to make the arguments
precise.
Exercise 1.7.9. Let ψ, ψ 0 : M → N be diffeomorphism of n-manifolds. Show
that the cobordisms Wψ and Wψ0 are equivalent if and only if ψ and ψ 0 are pseudo-
isotopic.
If W is a cobordism from M0 to M1 , then we say that f : W → [a, b] is a
Morse function if it has only non-degenerate critical points, and f −1 (a) = M0 and
f −1 (b) = M1 . Using such a Morse function, we can obtain a relative handle decom-
position of W by successively attaching handles to I × M0 . Furthermore, we can
arrange that the handles are attached with nondecreasing indices by Lemma 1.6.4.
Definition 1.7.10. If S is an embedded k-sphere with normal framing ν in
the n-manifold M , we define the trace of the surgery on M along (S, ν) to be the
cobordism W (S, ν) from M to M (S, ν) obtained by attaching an (n+1)-dimensional
(k + 1)-handle to I × M along {1} × S using the framing 1 × ν.
Traces of surgeries admit Morse functions with a single critical point. They are
known as elementary cobordisms. Since every cobordism admits a Morse function,
every cobordism is a product of elementary cobordisms.

1.8. The Whitney trick

The Whitney trick plays a fundamental role in manifold topology. Its fail-
ure in lower dimensions is the main reason why the classification of manifolds in
dimensions 3 and 4 is more difficult and has a different flavour than in higher di-
mensions. While this course focuses on low-dimensional topology, it is important
to understand the reason behind this distinction. The Whitney trick is the key
component of the proof of the h-cobordism theorem, which is the subject of the
following section. However, it was originally used by Whitney [74] to prove that
every n-manifold embeds into R2n .
Let A and B be smooth submanifolds of the n-manifold M , and write a =
dim(A) and b = dim(B). If A and B are transverse and a + b = n, then the
intersection is a discrete set of points. When A, B, and M are oriented, each point
p ∈ A ∩ B has a positive or negative sign, which we denote by sgn(p) ∈ {±1}. We
have sgn(p) = +1 if and only if an oriented basis of Tp A followed by an oriented
basis of Tp B gives an oriented basis of Tp M = Tp A ⊕ Tp B. Note that this also
depends on the order of the submanifolds A and B. In B ∩ A, the intersection signs
are (−1)dim(A) dim(B) times those in A ∩ B, which is the sign of the permutation
swapping the bases of Tp A and Tp B.
If A and B are closed and oriented, they represent homology classes [A], [B] ∈
H∗ (M ). Their algebraic intersection is
X
#(A ∩ B) := sgn(p) ∈ Z.
p∈A∩B
 1.8. THE WHITNEY TRICK 29

Φ1 (ZA )

p ZB
sB
p0 q0
sA
q ZA
B
A

Figure 2. An illustration of the proof of the Whitney trick. The

left shows the trivialisation of the normal bundle of the Whitney
disk. The right shows the standard model in the disk.

If α, β ∈ H ∗ (M ) are the Poincaré duals of [A] and [B], respectively, then this agrees
with hα ∪ β, [M ]i. More generally, α ∪ β is dual to A ∩ B, even if we do not assume
that dim(A) + dim(B) = dim(M ). In particular, #(A ∩ B) only depends on the
homology classes that A and B represent.
Proposition 1.8.1 (Whitney trick). Let A and B be oriented, smooth submani-
folds of the oriented n-manifold M that intersect transversely, and write a = dim(A)
and b = dim(B). Suppose the following hold:
(1) a + b = n,
(2) n ≥ 5, a ≥ 1, and b ≥ 3,
(3) p, q ∈ A ∩ B have opposite signs,
(4) when a is 1 or 2, then the map π1 (M \ B) → π1 (M ) is injective,
(5) there are embedded paths sA and sB from p to q in A and B, respectively,
with interiors disjoint from A ∩ B, such that sA s̄B is contractible in M .
Then there is an isotopy ht : M → M for t ∈ I such that h0 = IdM and
h1 (A) ∩ B = A ∩ B \ {p, q}.
In other words, we can eliminate intersection points of opposite signs using an
ambient isotopy, under the assumptions above.
Proof (Non-examinable). Since the curve C := sA s̄B is 0-homotopic in M ,
there is a map h : D2 → M such that h|∂D2 = C. As n ≥ 5, we can perturb the
map h to be an embedding by Theorem 1.4.8 and the remark following it, and to
be transverse to A and B along Int(D2 ). When a, b ≥ 3, the latter implies that
h(Int(D2 ))∩(A∪B) = ∅, since both A and B have codimension at least 3, while D2
is 2-dimensional. When a is 1 or 2, condition (4) implies that C is also contractible
in M \ B, and hence we can choose h such that h(Int(D2 )) ∩ B = ∅. As the
codimension of A is still b ≥ 3, we have h(Int(D2 )) ∩ A = ∅ by transversality. We
can also assume that Tx h(D2 )∩Tx A = Tx sA for every x ∈ sA and Tx h(D2 )∩Tx B =
Tx sB for every x ∈ sB .
Let v1 , . . . , va−1 be linearly independent normal vector fields along sA tangent
to A; see the left-hand side of Figure 2. Then vi (p) and vi (q) are normal to both
h(D2 ) and B for i ∈ {1, . . . , a − 1}. We extend these vector fields to sB such
that they are normal to both h(D2 ) and B. This is possible since the normal to
 1.9. THE H-COBORDISM THEOREM 30

T h(D2 ) + T B over sB is a trivial rank a − 1 vector bundle. Furthermore, the

frames v1 (p), . . . , va−1 (p) and v1 (q), . . . , va−1 (q) lie in the same path component of
GL(a − 1) as sgn(p) = −sgn(q).
Since D2 is contractible, the normal bundle of h(D2 ) in M is a trivial (n − 2)-
bundle εn−2 . The vector fields v1 , . . . , va−1 give a map ϕ from the curve C to
the Stiefel manifold Va−1 (Rn−2 ), which is (b − 2)-connected. (The Stiefel manifold
Vk (Rl ) consists of k-frames in Rl , and πi (Vk (Rl )) = 0 for i < l − k.) Hence, the
frame v1 , . . . , va−1 extends to the normal bundle of h(D2 ) in M , giving a splitting
εn−2 ∼ = εa−1 ⊕ εb−1 (the complement of εa−1 is trivial since every bundle is trivial
over D2 ).
Now we consider the standard model D2 × Ra−1 × Rb−1 , together with properly
embedded arcs ZA , ZB ⊂ D2 that intersect transversely in a pair of points p0 and
q 0 . See the right-hand side of Figure 2. Let D0 be the closure of the component
of D2 \ (ZA ∪ ZB ) disjoint from ∂D2 . The above splitting of the normal bundle of
h(D2 ) allows us to define a diffeomorphism
ψ : D2 × Ra−1 × Rb−1 → N (h(D2 )),
such that
• ψ(D0 × {0} × {0}) = h(D2 ),
• ψ(p0 , 0, 0) = p and ψ(q 0 , 0, 0) = q,
• sA ⊂ ψ(ZA × {0} × {0}) and sB ⊂ ψ(ZB × {0} × {0}),
• ψ(ZA × Ra−1 × {0}) ⊂ A and ψ(ZB × {0} × Rb−1 ) ⊂ B.
Let Φt for t ∈ I be an isotopy of D2 such that Φ0 = IdD2 , Φ1 (ZA ) ∩ ZB = ∅,
and Φt |∂D2 = Id∂D2 for every t ∈ I (first define Φt on ZA , then extend it to D2
using Theorem 1.4.3). Furthermore, let λ : R → I be a smooth function such that
λ(0) = 1 and λ(t) = 0 whenever |t| ≥ 1. Then we define the desired isotopy ht of
M on ψ(D2 × {x} × {y}) to be ψ ◦ Φtλ(|(x,y)|) ◦ ψ −1 for x ∈ Ra−1 and y ∈ Rb−1 ,
and the identity outside N (h(D2 )). 

1.9. The h-cobordism theorem

The h-cobordism theorem is the key technical tool that allows one to reduce
the classification of simply-connected smooth and topological manifolds in dimen-
sion at least 5 to algebraic topology, proven by Smale [65]. It was extended to
simply-connected topological manifolds in dimension 4 by Freedman [11]. How-
ever, Donaldson [10] showed it fails in the smooth category in dimension 4. In
dimension 3, it is still open whether it holds, and is equivalent to the smooth
4-dimensional Poincaré conjecture. In dimension 2, it follows from the classical
3-dimensional Poincaré conjecture, proven by Perelman.
Definition 1.9.1. Let W be a cobordism from M0 to M1 . Then we say that W
is an h-cobordism (where “h” stands for “homotopy”) if the embeddings ei : Mi ,→
W are homotopy equivalences for i ∈ {0, 1}.
Remark 1.9.2. By the relative Hurewicz theorem, when W , M0 , and M1 are
simply-connected, the following are equivalent:
(1) the embedding e0 is a homotopy equivalence,
(2) H∗ (W, M0 ) = 0,
(3) H ∗ (W, M1 ) = 0,
(4) H∗ (W, M1 ) = 0,
 1.9. THE H-COBORDISM THEOREM 31

(5) e1 is a homotopy equivalence.

Theorem 1.9.3 (h-cobordism theorem). If n ≥ 6 and W is a simply-connected
h-cobordism between the (n − 1)-manifolds M0 and M1 , then W is diffeomorphic to
I × M0 . In particular, M0 and M1 are diffeomorphic.
Using the terminology of Definition 1.7.5, the conclusion of the h-cobordism
theorem is that W is equivalent to Wψ for some equivalence ψ : M0 → M1 .
The simple-connectivity assumption is necessary: Milnor showed that the 7-
manifolds L(7, 1) × S 4 and L(7, 2) × S 4 with fundamental group Z7 are h-cobordant
but not diffeomorphic. Here L(7, 1) and L(7, 2) are 3-dimensional lens spaces, which
we will study in detail in Section 2.2.
Donaldson gave an example of a 5-dimensional smooth h-cobordism W from
M0 to M1 that is not diffeomorphic to I × M0 . On the other hand, Freedman
proved the Whitney trick in the topologically locally flat category in dimension 4,
which implies the h-cobordism theorem for topological 5-dimensional h-cobordisms
as well, and gave his classification of simply-connected topological 4-manifolds that
we will review in Section 4.2.
As an application of the h-cobordism theorem, we first give a characterisation
of the smooth n-disc for n ≥ 6.
Proposition 1.9.4. Let W be a compact simply-connected smooth n-manifold
for n ≥ 6 with simply-connected boundary. Then the following are equivalent:
(1) W n is diffeomorphic to Dn .
(2) W n is homeomorphic to Dn .
(3) W n is contractible.
(4) W n has the same integral homology as a point.
Proof. As the four statements are respectively weaker, it suffices to prove
that (4) implies (1). Let D0 ⊆ Int(W ) be a smooth n-disk. By excision,
H∗ (W \ Int(D0 ), ∂D0 ) ∼
= H∗ (W, D0 ) = 0.
Since π1 (W \ Int(D0 )) = 1 and π1 (∂W ) = 1, the manifold W \ Int(D0 ) is a simply-
connected h-cobordism from ∂D0 ≈ S n−1 to ∂W by Remark 1.9.2. Hence, by the
h-cobordism theorem, W \ Int(D0 ) is equivalent to the product I × ∂D0 , and so W
is diffeomorphic to Dn . 
We now state and prove the Generalised Poincaré Conjecture in dimensions at
least six.
Theorem 1.9.5 (Generalised Poincaré Conjecture). Let M be a closed, simply-
connected, smooth n-manifold that has the same homology as S n , and suppose that
n ≥ 6. Then M is homeomorphic to S n .
Proof. Choose a handle decomposition of M . Analogously to Step 1 of the
proof of the h-cobordism theorem, we can arrange that there is a single 0-handle
h0 . Then
W := M \ Int(h0 )
is simply-connected and has the same homology as a point by excision and the
long exact sequence of the pair (M, W ). Hence W is diffeomorphic to Dn by
Proposition 1.9.4. Reattaching h0 gives a twisted sphere, which is homeomorphic
to S n by Theorem 1.5.6, as it admits a Morse function with only two critical
points. 
 1.9. THE H-COBORDISM THEOREM 32

Note that the Generalised Poincaré Conjecture holds in every dimension in the
topological category, but it is beyond the scope of this book to prove it in full
generality. The 1- and 2-dimensional cases simply follow from the classification
of manifolds in these dimensions. The 3-dimensional case is the classical Poincaré
conjecture, proven by Perelman using the Ricci flow. The 4-dimensional case is due
to Freedman. In dimension 5, the result follows from work of Kervaire and Milnor,
which in fact implies the result in the smooth category in dimensions 5 and 6.
However, as mentioned earlier, there are 28 non-diffeomorphic smooth structures
on S 7 by the work of Milnor. It is still unknown if there is an exotic smooth
structure on S 4 .
 CHAPTER 2

3-manifolds

Every topological 3-manifold admits a unique smooth structure up to diffeo-

morphism by the work of Moise [48], hence the categories of topological and smooth
manifolds are equivalent in dimension three.
In these notes, we only discuss classical results from 3-manifold topology, pre-
ceding Thurston’s revolutionary work. Most of the proofs use cut-and-paste tech-
niques. Some of these results will then be applied in the following section on knots
and links.

2.1. The Schönflies theorem

The Jordan–Schönflies theorem states that every Jordan curve in R2 bounds
a disk. One might wonder whether this generalises to higher dimensions: Does
an embedding of an n-sphere into Rn+1 bound a disk? The answer depends on
the category we are working in. The Alexander horned sphere is an example of a
continuous embedding of D3 into R3 whose complement has non-finitely generated
fundamental group; see Rolfsen [61] for the construction. Its boundary is hence a
continuously embedded S 2 in S 3 that does not bound a disk on one side.
However, if we restrict to the smooth or PL category, then the generalised
Schönflies theorem is known to hold for n 6= 3: A smoothly or PL embedded S n
bounds a Dn+1 in Rn+1 . In the smooth category, for n > 4 this follows from the
h-cobordism theorem; see Proposition 1.9.4. The case n = 3 is still open. The case
n = 2 is due to Alexander; we prove it in the smooth category:
Theorem 2.1.1 (Schönflies theorem). Any smoothly embedded 2-sphere S in
R3 bounds a 3-disk.
Proof. We follow the exposition of Casson [5]. We use the following observa-
tion: If M = M1 ∪ M2 , where M1 is a 3-manifold, M2 ≈ D3 , and
M1 ∩ M2 ≈ ∂M1 ∩ ∂M2 ≈ D2 ,
then M ≈ M1 .
We can perturb S such that the height function x3 |S is Morse for x3 : R3 → R.
Let M be the closure of the bounded component of R3 \ S. The proof proceeds by
induction on the number n of index 1 critical points of x3 |S .
If n = 0, then S has one local minimum and one local maximum. Let tmin :=
min(x3 |S ) and tmax := max(x3 |S ). Then, for every t ∈ (tmin , tmax ), the level set
(x3 |S )−1 (t) is a circle in R2 × {t} that bounds a 2-disk in R2 × {t} by the smooth
Jordan curve theorem, and hence M ≈ D3 .
If n = 1, then there are either two local minima and one local maximum, or
one local minimum and two local maxima. Without loss of generality, assume it is
the former. Consider the level sets of x3 |S . As we pass the two local minima, two
33
 2.2. HEEGAARD DECOMPOSITIONS AND DIAGRAMS 34

Figure 3. Two embeddings of S 2 in R3 with one saddle point and

two local minima.

circles appear. If these have disjoint interiors, then M is a regular neighbourhood of

a ∩-shaped curve; see the left of Figure 3. If one circle is contained in the interior of
the other, M is a “tilted bowl;” see the right of Figure 3. In both cases, M bounds
a disk.
Now suppose that n ≥ 2. Let t be a regular value of x3 |S such that there is
at least one saddle point of x3 |S on each side of H := R2 × {t}. Then S ∩ H is a
compact 1-manifold. Let C be an innermost component of H ∩ S. This bounds a
disk D ⊂ H such that D ∩ S = ∂D = C. Then C separates S into disks D1 and
D2 , so S1 = D ∪ D1 and S2 = D ∪ D2 are embedded 2-spheres in R3 .
If both x3 |S1 and x3 |S2 have saddles, then they bound disks M1 and M2 , re-
spectively, by the inductive hypothesis. There are three cases: If M1 ∩ M2 = D,
then M = M1 ∪ M2 ≈ D3 by the observation. If M1 ⊂ M2 , then we can apply the
observation to M2 = M ∪ M1 as M ∩ M1 = D1 is a disk. The case M2 ⊂ M1 is
analogous.
If S1 has no saddle, it bounds a disk M1 by the n = 0 case. As before, using
the observation, S bounds a disk if and only if S2 bounds a disk. We push S2 a bit
to eliminate C from S2 ∩ H, reducing the number of its components. We repeat the
above procedure with a new innermost circle, which has to terminate since there is
a saddle on each side of H. It follows that S also bounds a disk. 

Exercise 2.1.2. Using the same method as in the proof of the Schönflies the-
orem (and assuming the Schönflies theorem), show that if T is a smooth torus in
S 3 , then one of the components of S 3 \ T has closure S 1 × D2 . Give an example of
a torus T in R3 such that the closure of the bounded component of R3 \ T is not a
solid torus.

2.2. Heegaard decompositions and diagrams

A genus g handlebody is a 3-ball with g oriented 1-handles attached. Alterna-
tively, it can be described as a regular neighbourhood of a wedge of g unknotted
circles in R3 . Every closed, oriented, and connected 3-manifold can be obtained by
gluing together two handlebodies of the same genus along their boundaries.
Definition 2.2.1. Let M be a closed, connected, and oriented 3-manifold. A
Heegaard decomposition of M consists of a closed, oriented surface Σ ⊂ M that
separates M into the union of two handlebodies Hα and Hβ , where Σ is oriented
as the boundary of Hα . We call Σ a Heegaard surface. The Heegaard genus of M
is the minimal genus of a Heegaard surface Σ.
 2.2. HEEGAARD DECOMPOSITIONS AND DIAGRAMS 35

Proposition 2.2.2. Every closed, connected, and oriented 3-manifold M ad-

mits a Heegaard decomposition.
Proof. Choose a handle decomposition of M . Since M is connected, we can
assume that it has a single 0-handle and a single 3-handle. Indeed, if there is more
than one 0-handle, we can cancel one of them against a 1-handle until we are left
with just one. The union of the 0-handle and the 1-handles is one of the handlebod-
ies, and the union of the 3-handle and the 2-handles is another handlebody. The
two handlebodies have the same genus since they share the same boundary. 

Remark 2.2.3. An alternative proof often found in the literature uses the fact
that every 3-manifold admits a triangulation. Then a Heegaard decomposition can
be obtained by taking a regular neighbourhood of the 1-skeleton and its comple-
ment. The complement is a handlebody because it is a regular neighbourhood of
the 1-skeleton of the dual cell decomposition, whose 0-cells are the centres of the
tetrahedra of the original triangulation, and two 0-cells are connected by a 1-cell if
and only if the corresponding tetrahedra share a face. It is a rather technical result
that every smooth manifold admits a triangulation; see Munkres [51].
Given two genus g handlebodies, the diffeomorphism class of the 3-manifold ob-
tained by gluing them together only depends on the isotopy class of the gluing map.
The group of isotopy classes of orientation-preserving automorphisms of a genus g
surface Σg is called the mapping class group, and is denoted by MCG(Σg ). Due
to Proposition 2.2.2, and since every handlebody admits an orientation-reversing
symmetry, one can study 3-manifolds via the mapping class group.
Exercise 2.2.4. Show that if M is a 3-manifold, then π1 (M ) has a presenta-
tion with the same number of generators and relations. Show that Z4 is not the
fundamental group of a closed 3-manifold.
Since every orientation-reversing automorphism of S 2 is isotopic to a reflection,
the only 3-manifold of Heegaard genus zero is S 3 . Heegaard genus one manifolds
form an important class:
Definition 2.2.5. A lens space is a closed, oriented 3-manifold of Heegaard
genus one that is not S 1 × S 2 .
In other words, a lens space is a manifold that can be obtained by gluing two
solid tori along their boundaries, with the exception of S 3 and S 1 × S 2 . The gluing
map is described by an automorphism of T 2 . Since π0 (Diff(T 2 )) ∼ = GL(2, Z), an
orientation-reversing automorphism of T 2 is isotopic to a transformation given by
a matrix  
q r
p s
with determinant −1, so p and q are relatively prime. If m is the meridian and l is
the longitude, then m is mapped to qm + pl. This already determines the resulting
3-manifold L(p, q), since every automorphism of T 2 that preserves m extends to
S 1 × D2 .
Alternatively, we can also describe L(p, q) as the quotient of S 3 ⊂ C2 by the
action of the cyclic group Cp given by
(z, w) 7→ (ζz, ζ q w)
 2.2. HEEGAARD DECOMPOSITIONS AND DIAGRAMS 36

for a p-th root of unity ζ ∈ Cp . This is a free action, hence the universal cover of
L(p, q) is S 3 , and π1 (L(p, q)) ∼
= Cp . We orient L(p, q) such that this covering map
is orientation-preserving. Alternatively, one can compute the fundamental group
by applying the Seifert–van Kampen theorem to the first description of lens spaces.
Exercise 2.2.6. Show that the two descriptions of a lens space are equiva-
lent. (Hint: Construct a cell decomposition of L(p, q) using a Cp -equivariant cell
decomposition of S 3 .)
Theorem 2.2.7. Let (p, q) and (p0 , q 0 ) be relatively prime pairs of integers.
Then L(p, q) and L(p0 , q 0 ) are homotopy equivalent if and only if p = p0 and qq 0 ≡
±n2 mod p for some n ∈ Z. They are homeomorphic if and only if p = p0 and
q 0 ≡ ±q ±1 mod p.
This was shown by Reidemeister [59] using Reidemeister torsion. It follows from
Theorem 2.2.7 that the lens spaces L(7, 1) and L(7, 2) are homotopy equivalent but
not homeomorphic.
A Heegaard decomposition is not unique. For example, S 3 has a genus g
Heegaard decomposition for every g: just consider the standard genus g surface
in S 3 . By a result of Waldhausen [71], every genus g Heegaard splitting of S 3 is
isotopic to this.
Given 3-manifolds M and M 0 , together with Heegaard surfaces Σ and Σ0 , re-
spectively, we can take their connected sum: Choose 3-balls B ⊂ M and B 0 ⊂ M 0
such that both B ∩ Σ and B 0 ∩ Σ0 are 2-disks. Then, if we take the connected sum
of M and M 0 along B and B 0 , then Σ#Σ0 is a Heegaard surface of M #M 0 .
Definition 2.2.8. Let Σ be a Heegaard surface in M . Then a stabilisation of
Σ is obtained by taking the connected sum with (S 3 , T 2 ).
The following theorem is due to Reidemeister and Singer:
Theorem 2.2.9. Let Σ0 and Σ1 be Heegaard surfaces in the closed, oriented
3-manifold M . Then there is a Heegaard surface Σ that is isotopic to a stabilisation
of both.
If one would also like to encode how the two handlebodies are glued together,
one can record the belt circles of the 1-handles of the two handlebodies:
Definition 2.2.10. An abstract Heegaard diagram is a triple (Σ, α, β), where
Σ is an oriented, genus g surface, and α = (α1 , . . . , αg ) and β = (β1 , . . . , βg ) are two
g-tuples of pairwise disjoint simple closed curves in Σ that are linearly independent
in H1 (Σ).
An embedded Heegaard diagram of the closed, connected, and oriented 3-manifold
M is an abstract Heegaard diagram (Σ, α, β) such that Σ ⊂ M is a Heegaard sur-
face, each αi bounds a disk in Hα , and each βj bounds a disk in Hβ .
For example, (T 2 , µ, λ) is a genus 1 embedded Heegaard diagram of S 3 , where
µ is a meridian and λ is a longitude of T 2 ; see Remark 1.1.11. A stabilisation of a
Heegaard diagram is obtained by taking the connected sum with this.
An abstract Heegaard diagram H = (Σ, α, β) defines a 3-manifold as follows:
Take Σ × I, and attach 3-dimensional 2-handles along α × {0} and β × {1} for
every α ∈ α and β ∈ β. Since the α-curves are linearly independent in H1 (Σ) and
there is g of them, the lower boundary component, which is obtained by surgering
 2.3. INCOMPRESSIBLE SURFACES AND THE LOOP THEOREM 37

Σ along α, is a 2-sphere. Similarly, the upper boundary component is also S 2 .

After attaching two 3-balls to these two spheres, we obtain a closed, connected, and
oriented 3-manifold, which we call the 3-manifold defined by H. If H is an embedded
Heegaard diagram of M , then the 3-manifold defined by H is diffeomorphic to M
relative to Σ.
If H = (Σ, α, β) is a Heegaard diagram of M , then we say that α0 = α \ {αi } ∪
{αi } is obtained from α by sliding αi over αj if αi , αi0 , and αj is the boundary
0

of a pair-of-pants (i.e., S 2 \ P for |P | = 3) in Σ disjoint from the other curves in

α. (This corresponds to sliding the 3-dimensional 2-handle corresponding to αi
over the handle corresponding to αj in the 3-manifold defined by the diagram.)
We can define a handle slide among β analogously. Then we have the following
strengthening of the Reidemeister–Singer theorem:
Theorem 2.2.11. Suppose that H and H0 are embedded Heegaard diagrams of
the closed, connected, oriented 3-manifold M . Then they become isotopic after a
sequence of handle slides, stabilisations, and destabilisations.

2.3. Incompressible surfaces and the loop theorem

Definition 2.3.1. Let S be a surface in a 3-manifold M . Then we say that
S is compressible either if there is a disk D ⊂ M such that D ∩ S = ∂D and ∂D
does not bound a disk in S, or if S has an S 2 component that is the boundary of
a 3-ball in M , or if S has a boundary-parallel D2 component. We say that S is
incompressible otherwise. We call such a D a compressing disk.
Given a compressing disk D for a surface S in the 3-manifold M , we can
compress it along D as follows: Let N (D) ≈ D ×[−1, 1] be a regular neighbourhood
of D such that N (D) ∩ S ≈ ∂D × [−1, 1]. Then the compressed surface is
S(D) := S \ (∂D × (−1, 1)) ∪ (D × {−1, 1}).
One can think of this as a surgery operation for embedded surfaces.
Definition 2.3.2. A surface S in a 3-manifold M is called 2-sided if its normal
bundle is trivial. We say that S is π1 -injective if the map π1 (S) → π1 (M ) induced
by the inclusion is injective.
For example, if both M and S are orientable, then S is 2-sided. The following
fundamental result was first stated by Dehn in 1910, but gaps in his proof were
found by Kneser. The first complete proof was given by Papakyriakopoulos [55] in
1957.
Theorem 2.3.3 (Dehn’s lemma). Let M be a 3-manifold with boundary, and
f : D2 → M a continuous map that is an embedding near ∂D2 and f (∂D2 ) ⊂
∂M . Then there exists an embedding g : D2 → M such that f and g agree in a
neighbourhood of ∂D2 .
The following generalisation of Dehn’s lemma is known as the loop theorem,
also due to Papakyriakopoulos:
Theorem 2.3.4 (Loop theorem). Let M be a 3-manifold with boundary (not
necessarily compact). If ∂M is not π1 -injective, then it is compressible.
 2.3. INCOMPRESSIBLE SURFACES AND THE LOOP THEOREM 38

Di

τi

Mi

pi

Di−1

Fi−1

Vi−1

Figure 4. The double cover pi : Mi → Vi−1 in the proof of the

loop theorem. Here, Vi−1 is a regular neighbourhood of the disk
Di−1 and τi is the covering involution.

Proof (Non-examinable). We follow the exposition of Hatcher [21]. Let

f : (D2 , ∂D2 ) → (M, ∂M ) be a continuous map such that [f |∂D2 ] 6= 1 ∈ π1 (∂M ).
Choose a triangulation of M . By the relative version of the simplicial approxima-
tion theorem, there is a triangulation of D2 such that f is homotopic to a simplicial
map f0 : (D2 , ∂D2 ) → (M, ∂M ). Let D0 := f0 (D2 ), and V0 a regular neighbour-
hood of D0 obtained by taking the union of the simpices of the second barycentric
subdivision of M incident to D0 .
Let p1 : M1 → V0 be a connected double cover of V0 , if one exists. See Figure 4.
As π1 (D2 ) = 1, we can lift f0 to a map f1 : D2 → M1 . We write D1 := f1 (D2 ) and
V1 for a regular neighbourhood of D1 , as above. We repeat this procedure to obtain
a sequence of maps fi : D2 → Mi whose image is Di with regular neighbourhood
Vi , where pi : Mi → Vi−1 is a connected double cover. We can triangulate Mi by
lifting the triangulation of Vi−1 , and Di and Vi are subcomplexes.
We claim that this process terminates in finitely many steps, and write n for
the largest index for which Vn admits no connected double cover. Indeed, let
Ei := p−1i (Di−1 ), and note that the double cover Ei → Di−1 is connected as Mi is
connected and deformation retracts onto Ei by lifting the retraction of Vi−1 onto
Di−1 . If τi is the covering automorphism of Ei → Di−1 , then Ei = Di ∪ τi (Di ). As
Ei is connected, Di ∩τi (Di ) 6= ∅. So there is a simplex σ of Di such that τi (σ) ⊂ Di .
As τi has no fixed points, σ 6= τi (σ). This means that Di−1 , which is a quotient of
Di , has fewer simplices than Di . The number of simplices in each Di is bounded
from above by the number of simplices in D2 , which hence gives an upper bound
on the number of steps n.
We now show that each component of ∂Vn is a 2-sphere. It suffices to prove
that H1 (∂Vn ; Z2 ) = 0. As Vn has no connected double cover, π1 (Vn ) has no index
two subgroup, and hence H 1 (Vn ; Z2 ) ∼= Hom(π1 (Vn ), Z2 ) = 0. So, by the universal
 2.3. INCOMPRESSIBLE SURFACES AND THE LOOP THEOREM 39
0
D

Figure 5. Removing double point curves of an immersed disk.

Left: the preimages of the double point curve are nested circles.
Middle: the preimages of the double point curve are non-nested
circles. Right: the preimage of the double point of curve is con-
nected.

coefficient theorem, H1 (Vn ; Z2 ) = 0, and by Poincaré duality, H2 (Vn , ∂Vn ; Z2 ) = 0.

Combined with the exact sequence
H2 (Vn , ∂Vn ; Z2 ) → H1 (∂Vn ; Z2 ) → H1 (Vn ; Z2 )
of the pair (Vn , ∂Vn ), we see that H1 (∂Vn ; Z2 ) = 0, as claimed.
Let ∂0 Vi be the component of ∂Vi containing fi (∂D2 ), and let
Fi := (p1 ◦ · · · ◦ pi )−1 (∂M ) ∩ ∂0 Vi .
We denote the kernel of the homomorphism (p1 ◦ · · · ◦ pi )∗ : π1 (Fi ) → π1 (∂M ) by
Ni . As [f |∂D2 ] 6= 0 ∈ π1 (∂M ), we have [fi |∂D2 ] 6∈ Ni .
Since each component of ∂Vn is a sphere, the surface Fn is planar. Hence
π1 (Fn ) is normally generated by the components of ∂Fn . As Nn 6= π1 (Fn ), there is
a component C of ∂Fn that represents an element of π1 (Fn ) \ Nn . Each component
of ∂0 Vn \ Fn is a disk. We can push the interior of the disk bounding C into Vn to
obtain a smooth embedding gn : D2 ,→ Vn such that [gn |∂D2 ] 6∈ Nn .
Starting from gn , we recursively construct a sequence of embeddings gi : D2 ,→
Vi with [gi |∂D2 ] 6∈ Ni . Then g0 is the desired embedding of D2 into M whose
boundary is homotopically non-trivial.
Suppose we have already constructed gi . To obtain gi−1 , we consider the im-
mersion pi ◦gi . After a small perturbation of gi , this is an immersion with transverse
double point curves and arcs, since pi is a 2-fold cover. We use cut-and-paste tech-
niques to get rid of these, resulting in the embedding gi−1 .
Let C be a double point curve of pi ◦ gi (D2 ). Let N (C) be a regular neigh-
bourhood of C in pi ◦ gi (D2 ). Then N (C) is an X-bundle over S 1 . This can be
obtained from X × [0, 1] by idenfitying X × {0} and X × {1} via an automorphism
ϕ of the figure X. This automorphism cannot be a 90◦ rotation or a reflection in
one of the two lines in X as otherwise D2 would contain a Möbius band. So ϕ is
either the identity of the figure X, or a reflection in a vertical or horizontal line.
First, suppose that ϕ is the identity. Then the preimage of C in D2 consists of
two circles that bound disks D and D0 . If these disks are nested, say D ⊂ D0 , then
we replace pi ◦ gi |D0 with pi ◦ gi |D and smooth the corners; see the left of Figure 5.
If D ∩ D0 = ∅, then we swap pi ◦ gi |D and pi ◦ gi |D0 and again smooth the corners;
see the middle of Figure 5. If ϕ is a reflection in the horizontal axis, we smooth
each X-fibre as on the right of Figure 5, which replaces the annulus mapping to
N (C) with another annulus. The case of reflection in a vertical axis is analogous.
In each case, we remove C from the double point set, without introducing any new
double point curves or changing pi ◦ gi |∂D2 .
Now suppose that A is a double point arc of pi ◦ gi (D2 ). Its preimage in D2
consists of a pair of arcs a, a0 ⊂ D2 with boundary on ∂D2 ; see the left of Figure 6.
We label the components of ∂D2 \ (a ∪ a0 ) by α, β, γ, and δ counterclockwise such
 2.3. INCOMPRESSIBLE SURFACES AND THE LOOP THEOREM 40

δ
gi0

α a Q a0 γ A
pi ◦ gi

β gi00

Figure 6. Left: The preimage of a double point arc A. Right:

The two resolutions gi0 and gi00 .

that a and α and a0 and γ form bigons B and B 0 (and hence a, β, a0 , δ are sides of
a quadrilateral Q).
There are two ways of smoothing the double point arc A; see the right of
Figure 6. In the first case, we glue together the maps pi ◦ gi |B and pi ◦ gi |B 0 , smooth
the corners, and remove the quadrilateral Q, to obtain the map gi0 . The other
smoothing of A gives rise to a map we denote gi00 . This is obtained by removing
pi ◦ gi |Q and vertically reversing it, gluing the side a of B to the side a0 of Q, and
the side a0 of B 0 to the side a of Q.
We claim that either [gi0 |∂D2 ] 6∈ Ni−1 or [gi00 |∂D2 ] 6∈ Ni−1 . Choose an orientation
of A; this induces orientations on a and a0 . If exactly one of them is oriented
coherently with ∂Q, then
αβγδ = (αγ)δ −1 (αβ −1 γδ −1 )−1 (αγ)δ
= (gi0 |∂D2 )δ −1 (gi00 |∂D2 )−1 (gi0 |∂D2 )δ.
Otherwise, we have
αβγδ = (αγ −1 )(γδ)−1 (αγ −1 )−1 (αδγβ)(γδ)
= (gi0 |∂D2 )(γδ)−1 (gi0 |∂D2 )−1 (gi00 |∂D2 )(γδ).
As αβγδ 6∈ Ni−1 , at least one of [gi0 |∂D2 ] and [gi00 |∂D2 ] is not in Ni−1 , as claimed,
and we continue with this map until pi ◦ gi becomes an embedding. 

It is clear that every π1 -injective surface without S 2 components that bound

3-balls and without boundary-parallel D2 components is incompressible. The con-
verse also holds for 2-sided surfaces:
Theorem 2.3.5. If S is a 2-sided, incompressible surface in a 3-manifold M ,
then S is π1 -injective.
Proof. By contradiction, suppose that there is a homotopically non-trivial
curve γ : S 1 → S that is null-homotopic in M . Then there is a smooth map
u : D2 → M transverse to S such that u|S 1 = γ. The preimage u−1 (S) is a smooth
1-manifold C in D2 . Let C0 be an innermost component of C. Let D0 ⊂ D2 be
the disk bounded by C0 . If u|C0 is null-homotopic in S, then we can replace u on
D0 with the null-homotopy and push it off S such that we obtain a map v with
v −1 (S) = C \ C0 . Repeating this procedure, we can assume that u|C0 is not null-
homotopic. Note that u(D0 ) ∩ S = u(C0 ). Hence, we can assume the curve γ on S
is null-homotopic in the complement of S.
 2.4. HAKEN MANIFOLDS 41

Since S is 2-sided, N (S) ≈ S × [−1, 1]. The curve γ × {1} is null-homotopic in

M 0 := M \(S ×(−1, 1)). Hence, by Theorem 2.3.4 applied to M 0 , the surface S ×{1}
is compressible in M 0 , so S is compressible in M , which is a contradiction. 
Exercise 2.3.6. Let K denote the Klein bottle, and let K ×I
e be an orientable
I-bundle over K. Then ∂K ×I e is a 2-torus T . Show that we can attach S 1 × D2
to K ×I
e along T such that in the resulting 3-manifold M , the Klein bottle K is
incompressible but not π1 -injective.

2.4. Haken manifolds

In this section, we review some results on a class of particularly nice 3-manifolds
called Haken manifolds that includes knot exteriors. They were used by Haken in
his unknot detection algorithm.
Definition 2.4.1. A 3-manifold M is irreducible if every embedded 2-sphere
in M bounds a 3-ball. We say that M is boundary-irreducible if each component of
∂M is incompressible in M .
An oriented 3-manifold M is Haken or sufficiently large if it is irreducible, and
contains a properly embedded, orientable, incompressible surface.
We will need a result that is known as the “half lives, half dies lemma.”
Lemma 2.4.2. Let M be a compact, oriented 3-manifold. Then
rk (H1 (∂M ))
rk (ker(H1 (∂M ) → H1 (M ))) = .
2
Proof. From the homological and cohomological long exact sequences of the
pair (M, ∂M ) with real coefficients, we obtain the commutative diagram

H2 (M, ∂M ; R)
∂∗
/ H1 (∂M ; R) i∗
/ H1 (M ; R)

≈ ≈ ≈
 ∗
 
H 1 (M ; R)
i / H 1 (∂M ; R) δ / H 2 (M, ∂M ; R),

where the vertical arrows are Poincaré duality isomorphisms. Hence

dim coker(i∗ ) = dim coker(∂∗ ) = dim H1 (∂M ; R) − dim Im(∂∗ )
(2.4.1)
= dim H1 (∂M ; R) − dim ker(i∗ ).
Since we are working over the filed R, we have i∗ = Hom(i∗ , R), and hence
dim ker(i∗ ) = dim coker(i∗ ).
Together with equation (2.4.1), we obtain that 2 dim ker(i∗ ) = dim H1 (∂M ; R), as
claimed. 
Proposition 2.4.3. Let M be a compact, oriented, irreducible 3-manifold with
boundary with b1 (M ) > 0. Then M is Haken.
Proof. Since b1 (M ) > 0, we have
H 1 (M ) = Hom(H1 (M ), Z) 6= 0.
If a ∈ H 1 (M ) \ {0}, then there is a map f : M → S 1 such that f ∗ (1) = a, where 1 is
the generator of H 1 (S 1 ) dual to the fundamental class of S 1 , using the identification
between H 1 (S 1 ) and the set of homotopy classes [M, S 1 ]. Let x be a regular value
 2.5. NORMAL SURFACES AND PRIME DECOMPOSITION 42

of f . Then S := f −1 (x) is Poincaré dual to a. It is orientable since we can pull back

∂/∂θ ∈ T1 S 1 to give a trivialisation of the normal bundle of S, and M is orientable.
After compressing S, we obtain a properly embedded, orientable, incompressible
surface S 0 homologous to S. Since M is irreducible, every S 2 component of S 0
bounds a ball. So we can remove these and all boundary-parallel D2 components
to obtain a surface S 00 without changing the homology class. As [S 00 ] 6= 0, we have
S 00 6= ∅. So S 00 is incompressible, and hence M is Haken. 
A Haken manifold M admits a Haken hierarchy, where we successively cut it
along incompressible surfaces until we obtain 3-balls. This makes them amenable to
inductive proofs. Let S be a properly embedded, orientable, incompressible surface
in M . Then we cut M along S; i.e., consider M1 := M \ N (S). If a component M10
of M1 has a boundary component that is not a sphere, it has b1 (M10 ) > 0 by the half
lives, half dies lemma (Lemma 2.4.2), and is hence also Haken by Proposition 2.4.3.
So we can cut M1 along an incompressible surface. We can continue this procedure
until all boundary components are spheres. As M was irreducible, each such sphere
bounds a ball in M , and so all the resulting components are balls. The fact that
this procedure terminates in finitely many steps relies on Theorem 2.5.8 below; see
Hempel [24, Theorem 13.3].
According to the following result of Waldhausen [72], Haken manifolds are
determined by their fundamental groups:
Theorem 2.4.4. Let M and N be compact, orientable 3-manifolds that are
irreducible and boundary-irreducible. Suppose M is Haken. Let ψ : π1 (N ) → π1 (M )
be an isomorphism that “respects the peripheral structure;” i.e., for each component
F of ∂N there exists a component G of ∂M such that ψ(i∗ (π1 (F ))) ⊂ A and A is
conjugate in π1 (M ) to i∗ (π1 (G)), where the i∗ denote inclusion homomorphisms.
Then there exists a homeomorphism f : N → M which induces ψ.
The following deep result is due to Agol [1], building on work of Kahn and
Markovic [29][30], which was known as the virtually Haken conjecture:
Theorem 2.4.5. Every compact, orientable, irreducible 3-manifold with infinite
fundamental group is virtually Haken; i.e., has a finite cover that is Haken.

2.5. Normal surfaces and prime decomposition

This section is based on unpublished notes of Casson [5].
Definition 2.5.1. We say that the closed 3-manifold M is prime if, whenever
M = A#B, we have A ≈ M and B ≈ S 3 , or A ≈ S 3 and B ≈ M .
Every irreducible 3-manifold is clearly prime. Conversely, we have the following:
Exercise 2.5.2. Show that if a closed 3-manifold is prime, then it is either
irreducible, or S 1 × S 2 , or the non-orientable S 2 -bundle over S 1 .
Normal surfaces were introduced by Kneser [35] in order to prove that every
closed 3-manifold can be written as a connected sum of prime 3-manifolds, and
further developed by Haken [18] into what is now known as normal surface theory.
It is a fundamental tool for many 3-manifold algorithms, including the following
result due to Haken [19], Waldhausen [72], Jaco–Shalen [26], Johannson [27], and
Hemion [23]:
 2.5. NORMAL SURFACES AND PRIME DECOMPOSITION 43

Figure 7. A normal surface intersects each tetrahedron in trian-

gles (left) or quads (right).

Theorem 2.5.3. There is an algorithm to decide whether two Haken 3-manifolds

are homeomorphic.
Proof. A knot exterior is a Haken 3-manifold, so we can use Theorem 2.5.3 to
decide whether it is homeomorphic to S 1 × D2 , the complement of the unknot. 
While most of this book focuses on differential topological methods, normal
surface theory is specific to the PL category of triangulated 3-manifolds. In dimen-
sion 3, every topological 3-manifold has a unique smooth structure and a unique
PL structure, so the two approaches are in some sense equivalent.
Let M be a 3-manifold with triangulation T . We write T i for the i-skeleton
of T . Let F be a closed surface in M ; not necessarily connected. We can make F
transverse to T i for every i.
Definition 2.5.4. We say that the surface F in the 3-manifold M with tri-
angulation T is normal if, for every 3-simplex τ of T , each component of F ∩ τ is
either
(1) a triangle with vertices on three edges meeting at a vertex of τ , one vertex
on each edge, as on the left of Figure 7, or
(2) a quadrilateral (quad ) with vertices on four edges that are adjacent to an
edge of τ , one on each edge, as on the right of Figure 7.
There are four types of triangles, corresponding to the four vertices of τ , and
three types of quads, corresponding to disjoint pairs of edges of τ . For each τ , there
might be several components of F ∩ τ of each triangle type, but at most one type
of quad as otherwise F would have double points.
Definition 2.5.5. Let S ⊂ M be a union of 2-spheres. We say that these
spheres are independent if no component of M \ S is homeomorphic to S 3 \ P for
P ⊂ S 3 finite.
Lemma 2.5.6. Let S be a subsurface of the 3-manifold M with triangulation T .
(1) If S is incompressible and M is irreducible, then S is isotopic to a normal
surface.
(2) If S is a collection of k independent 2-spheres, then M also contains an
independent set of k 2-spheres that is normal with respect to T .
Proof. We first prove statement (1). We perturb S such that it becomes
transvere to all simplices of T . We then define the weight of S to be w(S) := |S∩T 1 |.
Suppose that S minimises w(S) in its isotopy class. Then we claim that, for each
3-simplex ∆ of T , every component C of S ∩ ∂∆ is a simple closed curve that either
(i) lies in a 2-simplex of T ,
 2.5. NORMAL SURFACES AND PRIME DECOMPOSITION 44

b a0 b
a0

Figure 8. A component C of S ∩ ∂∆ that is not of the required

form admits one of these configurations.

(ii) intersects three edges of T exactly once that meet at a vertex of ∆ (see the
left of Figure 7), or
(iii) intersects four edges of T exactly once that are adjacent to an edge of ∆ (see
the right of Figure 7).
If C is not of this form, then there is a sub-arc b of an edge of ∆ and a sub-arc a0 of
C such that |Int(a0 ) ∩ T 1 | ∈ {0, 3}; see Figure 8. The curve a0 ∪ b bounds a bigon D0
in ∂∆ that is either disjoint from the 1-skeleton, or contains a single vertex of ∆.
In either case, let a be an arc in S ∩ ∆ such that a ∩ ∂∆ = ∂a0 , and such
that a and a0 are parallel. Then there is a disk D in ∆ parallel to D0 such that
∂D ∩S = a and ∂D ∩∂∆ = b. (Note that the interior of D might intersect S.) Then
a Whitney move (see Section 1.8) supported in N (D) eliminates the intersection
points ∂b from S ∩ T 1 , contradicting the minimality of w(S).
Let m(S) be the sum of |S ∩ δ| over 2-simplices δ of T , and isotope S such
that m(S) is minimal. Suppose that C is an innermost component of S ∩ ∂∆ for a
3-simplex ∆ that lies in a 2-simplex of ∆; i.e., C is of type (i). Then C bounds a
disk D in the complement of S. Since S is incompressible, this means that C also
bounds a disk D0 in S. Then D ∪ D0 is a 2-sphere that bounds a 3-ball B in M ,
since M is irreducible. Hence, we can eliminate C from the intersection of S with
2-simplices of T by isotoping S across B, decreasing m(S). Hence we can assume
that no component of S ∩ ∂∆ is of the form (i).
Now suppose that there is a 3-simplex ∆ of T such that a component F of
S ∩ ∆ is not a disk. Let C be a component of ∂F ; this bounds a disk D0 in ∂∆. We
choose F , C, and D0 such that D0 does not contain in its interior the boundary of a
non-disk component of S ∩ ∆. Then there is a disk D ⊂ ∆ with D ∩ S = ∂D = C.
Since S is incompressible, we can again eliminate C by an isotopy of S, decreasing
m(S). Hence S is a normal suface when m(S) is minimal. This concludes the proof
of statement (1).
The proof of statement (2) proceeds similarly, with the exception of the last
two paragraphs. Suppose that C is an innermost component of S ∩ ∂∆ of type (i).
Let the components of S be S1 , . . . , Sk , and suppose that C lies in S1 . If we
compress S1 along D, we obtain the 2-spheres S10 and S100 . Then at least one of
S 0 = S10 ∪ S2 ∪ · · · ∪ Sk and S 00 = S100 ∪ S2 ∪ · · · ∪ Sk is independent. Indeed, suppose
there were components B 0 and B 00 of M \ S 0 and M \ S 00 , respectively, that were
punctured spheres. Since S1 , . . . , Sk are independent, we must have S10 ⊂ ∂B 0 and
S100 ⊂ ∂B 00 . If S1 ⊂ B 0 or S1 ⊂ B 00 , then S would not be independent by the
Schönflies theorem (Theorem 2.1.1). Otherwise, B 0 ∩ B 00 = D, so B 0 ∪ B 00 would
be a punctured sphere component of M \ S, which is again impossible. Note that
m(S 0 ) < m(S) and m(S 00 ) < m(S). Hence, when m(S) is minimal, there is no
 2.5. NORMAL SURFACES AND PRIME DECOMPOSITION 45

component C of S ∩ ∂∆ that lies in a 2-simplex. Similarly, we can arrange that

S ∩ ∆ has only disk components for every 3-simplex ∆ of T . 
Definition 2.5.7. Let S and S 0 be disjoint embedded surfaces in the 3-manifold
M . Then we say that S and S 0 are parallel if M \ (S ∪ S 0 ) has a component C with
∂C = S ∪ S 0 and C ≈ S × I (in particular, S ≈ S 0 ).
Theorem 2.5.8. For any compact, irreducible 3-manifold M , there exists a
constant h(M ) ∈ N, such that, for any incompressible surface S in M with |S| ≥
h(M ), there are two components of S that are parallel.
Proof. Let T be a triangulation of M with t tetrahedra. Then we let
h(M ) := 6t + 2b2 (M ),
where b2 (M ) = dim H2 (M ; Z2 ). By statement (1) of Lemma 2.5.6, we can isotope
S such that it becomes normal with respect to T .
If ∆ is a 3-simplex of T , a component of ∂∆ \ S is called good if it is an
annulus, and is bad otherwise. At most six components of ∂∆ \ S are bad. (The
worst-case scenario is when we have all four types of triangles and one type of
quad in ∆, in which case the four components containing the vertices and two
components between triangles and quads are bad.) A component X of M \ S is
good if every component of X ∩ ∂∆ is good for every 3-simplex ∆ of T , and is bad
otherwise. At most 6t components of M \ S are bad. If |S| ≥ 6t + 2b2 (M ), then
|M \ S| ≥ 6t + 1 + b2 (M ), so there are at least 1 + b2 (M ) good components.
A good component X is obtained by gluing regions homeomorphic to trian-
gle times I and square times I along their edges times I. So X is an I-bundle
over a surface. Every non-trivial I-bundle contributes one Z2 direct summand to
H2 (M ; Z2 ), so there is at least one trivial I-bundle X. The boundary components
of X are parallel components of S. 
This finiteness result allows one to enumerate all normal surfaces by writing
down so called matching equations. The following result and corollary are due to
Kneser [35]:
Theorem 2.5.9. Let M be a triangulated 3-manifold with t 3-simplices. If M
contains an independent set of k 2-spheres, then k ≤ 6t + 2b2 (M ).
Proof. The proof is analogous to the proof of Theorem 2.5.8, except now we
invoke statement (2) of Lemma 2.5.6. 
Corollary 2.5.10. Every closed 3-manifold can be expressed as a connected
sum of finitely many prime 3-manifolds.
Uniqueness was shown 30 years later by Milnor [42].
Theorem 2.5.11. Let M be a closed oriented 3-manifold. If M ≈ M1 # . . . #Mk
and M ≈ N1 # . . . #Nl with Mi , Nj prime and not S 3 , then k = l, and after rein-
dexing, Mi and Ni are orientation-preserving homeomorphic.
Proof (Non-examinable). First, suppose that every two-sphere in M is sep-
arating. Then every Mi and Nj is irreducible. Let S be the union of the connected
sum spheres in M1 # . . . #Mk . If Σ is a two-sphere separating N1 and N2 # . . . #Nl ,
then we can isotope it such that it is transverse to S. Furthermore, we choose S
 2.5. NORMAL SURFACES AND PRIME DECOMPOSITION 46

Sj Sk−1
Sj0
Σ
C
Mi∗ D P N1∗
C Mk∗
D0

M1∗ Mr∗
F0 ∗
Mr+1 Mk∗
00
F

S Fr S
F1 S

F0
D A D0

Figure 9. Illustrations for the proof of uniqueness of prime de-

compositions of three-manifolds.

to minimise |S ∩ Σ|. We write M1∗ , . . . , Mk∗ for the closures of the components of
M \ S, where Mi∗ is a punctured Mi . We define N1∗ , . . . , Nl∗ analogously.
If S ∩ Σ 6= ∅, then let C be a component of S ∩ Σ innermost in Σ, bounding a
disk D ⊂ Σ with D ∩S = C = ∂D; see the top of Figure 9. Suppose D ⊂ Mi∗ . Since
Mi is irreducible, one of the components of Mi∗ \ D has closure a punctured ball
P . Suppose that C lies in a component Sj of ∂Mi∗ , and let D0 = Sj \ P . Replace
Sj by D ∪ D0 , pushed slightly to eliminate the intersection component C. This
contradicts the choice of S to minimise the number of components of S ∩ Σ.
So S ∩ Σ = ∅. Suppose that S ∩ N1∗ 6= ∅. Since N1 is irreducible, some
component of S is parallel to Σ and Mi∗ ⊂ N1∗ with N1∗ \ Mi∗ ≈ S 2 × I for some
i ∈ {1, . . . , k}. If S ∩ N1∗ = ∅, then N1∗ ⊂ Mi∗ for some i. Since Mi is irreducible,
Mi∗ \ N1∗ is a punctured ball. In either case, Mi ≈ N1 and
M1 # . . . #Mi−1 #Mi+1 # . . . #Mk ≈ N2 # . . . #Nl .
 2.5. NORMAL SURFACES AND PRIME DECOMPOSITION 47

Hence, the result follows by induction.

Now suppose that F is a maximal collection of two-spheres in M such that
M \ F is connected, and let r = |F |. Furthermore, let S be as above. If F ∩ S 6= ∅,
then we choose a component C of F ∩ S innermost in S, and compress F along
the disk that C bounds in S. This gives rise to collections of spheres F 0 and F 00 ,
one of which – say F 0 – does not separate; see the middle of Figure 9. Then we
can isotope F 0 to eliminate C. Repeating this procedure, we obtain a sequence
of nonseparating collections of spheres F = F 0 , F 1 , . . . , F n such that F n ∩ S = ∅.
If Fjn ⊂ Mi∗ for a component Fjn of F n , then Mi ≈ S 1 × S 2 by Exercise 2.5.2,
since Mi is prime but not irreducible as it contains the nonseparating sphere Fjn .
Hence, after reindexing, M1 ≈ · · · ≈ Mr ≈ S 1 × S 2 . Furthermore, M \ N (F n ) is a
2r-punctured Mr+1 # . . . #Mk . The same applies to the factorisation N1 # . . . #Nl .
We claim that M \ N (F ) ≈ M \ N (F n ). It suffices to show that, if D is a
disk in M with D ∩ F = ∂D such that ∂D bounds a disk D0 ⊂ F and F 0 :=
(F \ D0 ) ∪ D does not separate M , then M \ N (F ) ≈ M \ N (F 0 ); see the bottom
of Figure 9. Indeed, since F was maximal, D × I separates M \ N (F ) into two
components, whose closures we denote by A and B. Then M \ N (F ) ≈ A ∪D B
and M \ N (F 0 ) ≈ A ∪D0 B. These are homeomorphic since both are obtained from
A t B by gluing disks in their boundaries.
We conclude that a 2r-punctured Mr+1 # . . . #Mk is homeomorphic to a 2r-
punctured Nr+1 # . . . #Nl , as they are both homeomorphic to M \ N (F ). Hence,
Mr+1 # . . . #Mk ≈ Nr+1 # . . . #Nl ,
and it does not contain a separating two-sphere, so the result now follows from the
first part of the proof. 
 CHAPTER 3

Knots and links

Knots and links play a fundamental role in low-dimensional topology. Their

exteriors provide an important class of 3-manifolds. One can obtain all 3-manifolds
using surgery on links in S 3 . Furthermore, all 4-manifolds can be obtained from
traces of surgeries on links in connected sums of copies of S 1 × S 2 (after adding
3-handles and a 4-handle in a unique way); see Section 4.1 on Kirby calculus for
more information. Invariants of 3- and 4-manifolds are often easier to understand
for links, and some of them are in fact defined as extensions of link invariants.

3.1. Knots and links

Definition 3.1.1. A k-component link in a 3-manifold Y is a smooth embed-
ding
[k
L: S 1 × {i} ,→ Y.
i=1
The links L0 and L1 are equivalent if they are ambient isotopic; i.e., there is an
isotopy Φ : Y × I → Y such that Φ0 = IdY and Φ1 ◦ L0 = L1 . A knot is a
1-component link.
One could also define piecewise linear (PL) knots and links, whose study is
amenable to combinatorial techniques. However, we will take the differential topo-
logical point of view. The two are equivalent for links.
The image Im(L) is a smooth 1-dimensional submanifold of M that is oriented
such that the map L is orientation-preserving. Given an orientation of Im(L), the
parametrisation L is uniquely determined up to isotopy as Diff+ (S 1 ) is connected.
Hence, we will often think of a link as an oriented smooth 1-dimensional submanifold
of M . We will denote by L the link L with its orientation reversed, called the reverse
of L.
Since π0 (Diff+ (S 3 )) = 0 by Cerf [7], any orientation-preserving automorphism
of S is isotopic to the identity. So two links in S 3 are ambient isotopic if and only
3

if they are orientation preserving diffeomorphic. We denote by −L the mirror of

the link L, obtained by reflecting L in a plane.
We can represent a link in R3 by looking at its projection onto R2 . If this
projection is an immersion with only transverse double point singularities, and at
each double point we record which strand is higher with respect to the coordinate z,
we obtain a link diagram. This allows one to reconstruct the link up to equivalence.
Definition 3.1.2. A link diagram is an immersed, oriented 1-manifold in R2
or S 2 that has only transverse double point singularities, and at each double point,
a designation of one of the two strands as the over-strand.
Every link has a diagram:
48
 3.2. REIDEMEISTER MOVES 49

R1 R2 R3

Figure 10. The three Reidemeister moves.

Proposition 3.1.3. Let L be a link in R3 . Then there is an isotopy {ψt : t ∈

I } of R3 such that ψ0 = IdR3 and orthogonal projection of ψ1 (L) onto R2 is an
immersion with only transverse double point singularities.
The proof is similar to the proof of Theorem 1.4.8, which we hence omit.

3.2. Reidemeister moves

Given two link diagrams, they represent equivalent links if and only if they
can be connected by a sequence of Reidemeister moves and planar isotopies. We
describe these next. For an illustration, see Figure 10.
We say that two diagrams are related by a planar isotopy if there is an isotopy
of R2 taking one diagram to the other.
Let D be a diagram of the link L. We can perform the first Reidemeister
move R1 if there is a component C of R2 \ D that contains a single crossing in its
boundary. The resulting diagram is obtained by smoothing D \ ∂C, and has one
fewer crossings than D. We will also refer to the reverse of this move as an R1
move.
Now suppose that R2 \ D contains a component C that has two crossings along
its boundary, and such that C ≈ D2 . We further assume that, as we traverse one of
the two strands along ∂R, both crossings are either overcrossings, or they are both
undercrossings. Then the second Reidemeister move R2 applied to the component
C removes a neighbourhood of C from D, and reconnects the resulting two pairs
of endpoints in N (C) without intersection points, as in the middle of Figure 10.
Again, we also call the reverse of this move an R2 move.
Finally, suppose that R2 \ D has a component C with C ≈ D2 whose boundary
contains three crossings, and one of the three strands along ∂C – call it s – has two
undercrossings; see the right-hand side of Figure 10. Let c be the crossing along
∂C opposite s, and let C 0 be the component of R2 \ D that meets C at c. Then we
can remove a small extension of s from D, and reconnect its endpoints across C 0 .
This is called the third Reidemeister move R3.
The following result is due to Reidemeister [58]:
Theorem 3.2.1. Let D0 and D1 be link diagrams. Then they represent equiva-
lent links if and only if they can be connected by a sequence of planar isotopies and
Reidemeister moves R1–R3.
Sketch of proof. We denote by π : R3 → R2 the projection. Let Li be a
link with diagram Di for i ∈ {0, 1}. Suppose there is an isotopy φt for t ∈ I of R3
such that φ0 = IdS 3 and φ1 (L0 ) = L1 . We write Lt := φt (L). If the isotopy φt is
generic, then π(Lt ) is an immersion with only transverse double points for every
t ∈ I outside a finite set S ⊂ (0, 1). Furthermore, for s ∈ S, the projection π(Ls )
has exactly one of the following singularities:
 3.3. SEIFERT SURFACES 50

(1) a cusp, corresponding to a point p ∈ Lt where Tp Lt is parallel to ∂/∂z,

(2) a self-tangency modelled on the graphs of x2 and −x2 ,
(3) a triple point where any two branches of the curve are transverse.
We write Dt for the diagram of Lt for t ∈ I \S. If t and t0 lie in the same component
of I \S, then the diagrams Dt and Dt0 are related by a planar isotopy. Furthermore,
for s ∈ S and ε sufficiently small, Ds−ε and Ds+ε are related by a Reidemeister
move R1 in case (1), move R2 in case (2), and move R3 in case (3).
Conversely, every planar isotopy and Reidemeister move can be lifted to an
isotopy of the knot. Furthermore, if two knots have the same diagram, they are
isotopic via linearly interpolating between the z-coordinate functions. 

A corollary of the above theorem is that any quantity that is defined for link
diagrams and is invariant under Reidemeister moves gives rise to a link invariant.
The Jones polynomial, which we will encounter later, is a particularly interesting
example of this construction.
Definition 3.2.2. Let D be a link diagram. We say that a crossing of D is
positive if, as we traverse the over-strand following the orientation, the under-strand
crosses from right to left, and is negative otherwise.
The writhe w(D) of the diagram D is the number of positive crossings minus
the number of negative crossings.
The winding number of the knot diagram D is the number of rotations the unit
tangent vector of D makes as we traverse the knot.
Writhe and winding number are not knot invariants, but are only changed by
move R1. Trace [70] showed that two diagrams of the same knot are related only
using moves R2 and R3 if and only if they have the same writhe and winding
number.
The writhe is closely related to the notion of linking. By Alexander duality,
H1 (S 3 \K) ∼
= Z for any knot K. Given disjoint knots K and K 0 , their linking number
lk(K, K ) is the integer corresponding to [K 0 ] ∈ H1 (S 3 \ K). Given a diagram for
0

the link K ∪ K 0 , this can be computed by taking the number of positive crossings
minus the number of negative crossings between K and K 0 , divided by two.
Definition 3.2.3. The crossing number c(D) of a diagram D of the link L is
the number of its crossings. We define the crossing number c(L) of the link L to
be the minimum of c(D) over all diagrams of L.
Knots and links are usually tabulated according to their crossing number. For
example, the Rolfsen table of knots [61] tabulates all knots up to ten crossings. The
unknot is represented by the standard circle S 1 in S 3 , and is the only knot with
crossing number zero. There is no knot of crossing number two. Up to mirroring,
there is just one knot of crossing number three: the trefoil (see Figure 12), and one
crossing number four knot: the figure eight. Even though the crossing number is
easy to define, it is unknown whether it is additive under connected sums.

3.3. Seifert surfaces

Surfaces bounding knots in the 3-sphere play an important role in defining
various knot invariants.
 3.3. SEIFERT SURFACES 51

Definition 3.3.1. Let L be a link in the 3-manifold Y . A Seifert surface for

L is a compact, oriented surface S embedded in Y without closed components such
that ∂S = L.
Proposition 3.3.2. Every link in S 3 admits a Seifert surface.
Proof. Let D be a diagram of L. We resolve each crossing according to
the orientation of D, obtaining an embedded, oriented 1-manifold C in R2 whose
components are called Seifert circles. We choose a collection S0 of disjoint disks in
R3 with boundary C. For each crossing in D, we add a half-twisted band to S0 ,
obtaining the desired Seifert surface S. 
We say that the diagram D is simple if it has no nested Seifert circles. The
above proof only works for links in S 3 . More generally, we have the following result:
Proposition 3.3.3. Let L be a closed 0-homologous oriented (n − 2)-manifold
in the n-manifold Y , and a ∈ Hn−1 (Y, L) a homology class such that ∂a = [L] ∈
Hn−2 (L), where ∂ : Hn−1 (Y, L) → Hn−2 (L) is the boundary map in the long exact
sequence of the pair (Y, L). Then there is a compact, connected, oriented (n − 1)-
manifold embedded in Y with boundary L representing a.
Proof. Let E := Y \ N (L) be the exterior of L. Since L is 0-homologous in
Y , there is a trivialisation ∂E ≈ L × S 1 such that L × {1} is 0-homologous in E.
Furthermore, there is a cohomology class α ∈ H 1 (E) dual to a ∈ Hn−1 (E, ∂E) ∼ =
Hn−1 (Y, L) such that α(µ) = 1 for any meridian µ of L. Since the Eilenberg–
MacLane space K(Z, 1) ∼ = S 1 , the map i : [E, S 1 ] → H 1 (E) given by i(f ) := f ∗ (1)
for [f ] ∈ [E, S ] and 1 ∈ H 1 (S 1 ) is an isomorphism. Let fα : E → S 1 be a map
1

such that i(fα ) = α and fα |∂E : L × S 1 → S 1 is projection onto the second factor.
We can perturb fα such that it is smooth, and such that 1 ∈ S 1 is a regular value.
After taking the connected sum of the components of f −1 ({1}), it extends in N (L)
to a compact, connected, oriented manifold with boundary L representing a. 
Seifert surfaces are not unique. For example, if S is a Seifert surface of a link
L in the 3-manifold Y , then one can always choose a 3-disk D ⊂ Y such that
D ∩ S ≈ D2 t D2 , and replace D ∩ S with an unknotted annulus. This operation
is called a stabilisation, and its reverse a destabilisation.
Proposition 3.3.4. Let L be a 0-homologous link in the 3-manifold Y and a ∈
H2 (Y, L) a homology class. Then any two Seifert surfaces of L representing the class
a can be connected by a sequence of isotopies, stabilisations, and destabilisations.
Proof. Let S0 and S1 be Seifert surfaces of L representing the class a, and
write E for the link exterior. Choose maps f0 , f1 : E → S 1 such that 1 is a common
regular value of both, f0−1 ({1}) = S0 and f1−1 ({1}) = S1 , and which are projections
onto the S 1 -factor along ∂N (L) ≈ L × S 1 .
Let {ft : t ∈ I} be a generic 1-parameter family of smooth functions connecting
f0 and f1 relative to ∂E. Then ft is a circle-valued Morse function, except for
finitely many values of t, when ft has a single birth-death critical point p, and
generically f (p) 6= 1; see equation (1.6.1).
Hence, the preimage ft−1 ({1}) only changes when there is a nondegenerate
critical point p of ft with ft (p) = 1. If 1 is a regular value of ft , we write St for
the connected sum of the components of ft−1 ({1}). When the value of an index 0
or 3 critical point passes through 1, the preimage ft−1 ({1}) changes by the birth or
 3.3. SEIFERT SURFACES 52

death of a small 2-sphere that bounds a 3-ball, hence St−ε and St+ε are isotopic.
When the value of an index 1 or 2 critical point passes through 1, the level set
ft−1 ({1}) changes by adding or removing a tube. Hence, the surfaces St−ε and
St+ε are related by an isotopy if the tube connects distinct components, and a
stabilisation or destabilisation otherwise. 
Remark 3.3.5. While a knot in S 3 has no unique Seifert surface, the normal
framing given by a Seifert surface is well-defined, and is called the Seifert framing.
This follows by applying the half lives half dies lemma to the knot exterior. The
blackboard framing of a knot diagram is given by the normal in the plane of the
diagram. The difference between the Seifert framing and the blackboard framing
is the writhe of the diagram.
Definition 3.3.6. Let K be a knot in S 3 . Then its Seifert genus g(K) is the
minimum of g(S), where S is a Seifert surface of K and g(S) is the genus of S.
The Seifert genus detects the unknot, in the following sense:
Lemma 3.3.7. Let K be a knot in S 3 . Then g(K) = 0 if and only if K is the
unknot U .
Proof. Clearly, g(U ) = 0. Conversely, suppose that g(K) = 0. Then there is
an embedding e : D2 ,→ S 3 such that e|S 1 = K. Then et := e|(1−t)·S 1 for t ∈ [0, 1−ε]
provides an isotopy of K to the curve e1−ε . If ε is small enough, then we can linearly
isotope e1−ε to the tangent plane e∗ (T0 D2 ) and obtain the unknot. 
The following celebrated result is due to Haken [18]:
Theorem 3.3.8. There is an algorithm to decide whether a knot in S 3 is trivial.
Proof. The only knot K in S 3 with exterior S 1 × D2 is the unknot, since
{1} × ∂D2 has to be a longitude of K for homological reasons. So g(K) = 0 and
K = U by Lemma 3.3.7. A knot exterior is a Haken three-manifold, so we can use
Theorem 2.5.3 to decide whether it is homeomorphic to S 1 × D2 . 
Given knots K1 and K2 , we can form their connected sum K1 #K2 , as follows:
Choose diagrams D1 and D2 , respectively, such that D1 ⊂ R2+ and D2 ⊂ R2− .
Let r : I × I ,→ R2 be an embedding such that r(I × I) ∩ D1 = r({0} × I) and
r(I × I) ∩ D2 = r({1} × I), and the orientation along ∂r(I × I) is coherent with the
orientations of D1 and D2 . Then K1 #K2 has diagram the symmetric difference
(D1 ∪ D2 )4(∂r(I × I)).
Proposition 3.3.9. Let K1 and K2 be knots in S 3 . Then
g(K1 #K2 ) = g(K1 ) + g(K2 ).
Proof. By considering the boundary connected sum of Seifert surfaces of K1
and K2 , we obtain that g(K1 #K2 ) ≤ g(K1 ) + g(K2 ).
We now show that g(K1 #K2 ) ≥ g(K1 ) + g(K2 ). Let S be a Seifert surface
for K1 #K2 . Make S transverse to the connected sum sphere C. Then C ∩ S is
a 1-manifold with a single arc component, and a collection of circle components.
Choose an innermost circle component γ of C ∩ S, which bounds a disk D ⊂ C
such that D ∩ S = γ. If we compress S along D, we obtain a surface S 0 such that
χ(S 0 ) = χ(S) + 2. If S 0 has an S 2 component, we remove it and obtain a surface
with Euler characteristic χ(S).
 3.4. THE SEIFERT FORM 53

Repeating this process, we obtain a surface S 00 with boundary K such that

S ∩ C is a single arc and χ(S 00 ) ≥ χ(S). Furthermore, S 00 has no S 2 component.
00

Hence, removing the closed components of S 00 , we obtain a Seifert surface S 000 of

K such that χ(S 000 ) ≥ χ(S), and so g(S 000 ) ≤ g(S). As S 000 ∩ C consists of a single
component, S 000 is a boundary connected sum of Seifert surfaces of K1 and K2 .
Hence g(S) ≥ g(S 000 ) ≥ g(K1 ) + g(K2 ), as claimed. 

Definition 3.3.10. We say that a knot K 6= U is prime if K = K1 #K2 implies

that K1 = U or K2 = U .
Corollary 3.3.11. Every nontrivial knot can be written as a connected sum
of prime knots.
Proof. We prove the claim by induction on g(K). If K is not prime, then
it can be written as K = K1 #K2 , where K1 6= U and K2 6= U . Furthermore,
g(K1 ) < g(K) and g(K2 ) < g(K) by Proposition 3.3.9 and Lemma 3.3.7. We
can now write K1 and K2 as a connected sum of prime knots by the inductive
hypothesis. 

Schubert [62] proved the uniqueness of prime decompositions of knots:

Theorem 3.3.12. Any two prime decompositions of a knot are related by a
permutation of the summands.

3.4. The Seifert form

We now introduce the Seifert form, which is the source of a number of powerful
knot invariants.
Definition 3.4.1. Let S be a Seifert surface of a knot K in S 3 . We define the
Seifert form
h , iS : H1 (S) × H1 (S) → Z
as follows. For a, b ∈ H1 (S), choose 1-cycles α, β ⊂ S representing them. We write
β + for the 1-cycle in S 3 \ S obtained by pushing β off S along the positive normal
of S. Then
ha, biS := lk(α, β + ).
If a1 , . . . , a2g is a basis of the free Z-module H1 (S), then the corresponding Seifert
matrix V has (i, j)-th entry hai , aj iS for i, j ∈ {1, . . . , 2g}.
The Seifert form is bilinear, but not necessarily symmetric. Given a Seifert
matrix V of K, we call V + V T the corresponding symmetrised Seifert matrix.
Definition 3.4.2. The Alexander polynomial of a knot K in S 3 is defined as
∆K (t) := det(V − tV T ),
where V is a Seifert matrix for K.
Proposition 3.4.3. The Alexander polynomial is well-defined up to multipli-
cation by tk for k ∈ Z.
Proof. We first show that det(V − tV T ) is independent of the choice of basis
a1 , . . . , a2g . Let b1 , . . . , b2g be another basis of H1 (S). Then the Seifert matrix V1
 3.4. THE SEIFERT FORM 54

obtained from b1 , . . . , b2g is of the form M T V M for an invertible integral matrix

M . Then det(M ) = ±1 as it is a unit in Z. Hence
det(V1 − tV1T ) = det(M T V M − tM T V T M ) =
det(M T ) det(V − tV T ) det(M ) =
det(V − tV T ).
By Proposition 3.3.4, it suffices to prove that det(V − tV T ) is unchanged by a
stabilisation of S. Suppose that S 0 is a stabilisation of S. If V is the Seifert matrix
of S, then    
V ∗ 0 V ∗ 0
V 0 =  ∗ ∗ 1  or  ∗ ∗ 0 
0 0 0 0 1 0
0
is a Seifert matrix of S , where we extend a basis of H1 (S) by adding a longitude
and a meridian of the stabilisation tube. After a change of basis, this becomes
   
V ∗ 0 V 0 0
 0 0 1  or  ∗ 0 0  .
0 0 0 0 1 0
Hence, we have det(V 0 − t(V 0 )T ) = t det(V − tV T ), and the result follows. 
.
The Alexander polynomial satisfies the symmetry relation ∆K (t) = ∆K (t−1 ),
. k
where “=” denotes equality up to multiplication by t for some k ∈ Z. The Con-
way normalization of the Alexander polynomial is the unique Laurent polynomial
satisfying ∆K (t) = ∆K (t−1 ). It is immediate from the definition that
deg ∆K (t) ≤ g(K).
Since V − V T agrees with the matrix of the intersection form of S in the basis
a1 , . . . , a2g , which is unimodular, we have
(3.4.1) ∆K (1) = det(V − V T ) = ±1.
We now introduce the signature of a knot, which, as we shall see, gives lower
bounds on the genera of orientable surfaces the knot bounds in the 4-ball.
Definition 3.4.4. The signature σ(K) of the knot K is defined as the signature
of the symmetrised Seifert matrix V + V T (i.e., the number of positive eigenvalues
minus the number of negative eigenvalues).
Proposition 3.4.5. The signature σ(K) of a knot K is well-defined.
Proof. For a given basis of H1 (S), the symmetrised Seifert matrix is the
matrix of the symmetric bilinear form
ha, bi := ha, biS + hb, aiS .
The signature is then the dimension of a maximal positive definite subspace of
H1 (S) minus the dimension of a maximal negative definite subspace of H1 (S).
This shows that the signature is independent of the choice of basis.
Invariance under stabilisation of the Seifert surface follows from the observation
that V +V T , up to change of basis, changes by taking the direct sum with the matrix
 
0 1
,
1 0
 3.4. THE SEIFERT FORM 55

Figure 11. The twist knot 61 is the first non-trivial slice knot.

which has signature 0. 

It immediately follows from the definition that the signature is additive under
connected sum. We have the following analogues of Seifert surfaces and the Seifert
genus in dimension four:
Definition 3.4.6. Let K be a knot in S 3 . A slice surface for K is a connected,
compact, oriented surface F embedded in D4 such that ∂F = K. The 4-ball genus
g4 (K) of K is the minimal genus of a slice surface for K.
Clearly, g4 (K) ≤ g(K). A knot is called slice if g4 (K) = 0. There are nontrivial
slice knots. The smallest one is the twist knot 61 in the Rolfsen table; see Figure 3.4.
Exercise 3.4.7. Show that the knot 61 is slice by attaching a single band,
resulting in a 2-component unlink.
Lemma 3.4.8. Let K and K 0 be disjoint knots in S 3 with Seifert surfaces S
and S 0 and slice surfaces F and F 0 , respectively. Then
lk(K, K 0 ) = S · K 0 = S 0 · K = F · F 0 ,
where · denotes the algebraic intersection number.
Proof. Since S is Poincaré–Lefschetz dual to the meridian of K in the exterior
of K, we have lk(K, K 0 ) = S · K 0 . Similarly, lk(K, K 0 ) = S 0 · K.
View D4 as S+ 4
, and push S 0 into S−4
. Then F ∪ −S and F 0 ∪ −S 0 are closed
4
oriented surfaces in S that are null-homologous, hence
0 = (F ∪ −S) · (F 0 ∪ −S 0 ) = F · F 0 + S · S 0 .
As S ∩ S 0 = S ∩ K 0 , and since the intersection signs are opposite, the result follows
from the previous paragraph. 
Proposition 3.4.9. Let K be a knot in S 3 . Then
|σ(K)|
≤ g4 (K).
2
Proof. Let g = g4 (K), and let F be a genus g slice surface for K. Choose an
arbitrary Seifert surface S for K. Then S ∪ F is a closed, oriented, null-homologous
surface in D4 , and hence bounds a Seifert 3-manifold M according to Proposi-
tion 3.3.3. Let
U = ker(H1 (∂M ) → H1 (M )).
By Lemma 2.4.2,
rk(H1 (∂M ))
rk(U ) = = g(S) + g(F ).
2
 3.5. IMPORTANT CLASSES OF KNOTS 56

Note that H1 (∂M ) ∼

= H1 (S) ⊕ H1 (F ), and write US := U ∩ H1 (S). Then
(3.4.2) rk(US ) ≥ rk(U ) − rk(H1 (F )) = g(S) − g(F ).
We claim that ha, biS = 0 for a, b ∈ US . Indeed, a and b are 0-homologous in
M , and let A and B be 2-chains in M with ∂A = a and ∂B = b. We write B +
for the 2-chain in D4 obtained by pushing B off M in the direction of the positive
normal of M . Then ∂B + = b+ and A ∩ B + = ∅, hence
ha, biS = lk(a, b+ ) = A · B + = 0,
where the second equality follows from Lemma 3.4.8.
On the other hand, if V is a Seifert matrix associated to S, then
det(V + V T ) ≡ det(V − V T ) ≡ 1 mod 2
T
by equation (3.4.1), so V + V is nondegenerate. Hence, if P ≤ H1 (S) and N ≤
H1 (S) are maximal positive and negative definite subspaces of the Seifert form,
respectively, then P ⊕ N = H1 (S). So, if we write p = rk(P ) and n = rk(N ), then
p + n = 2g(S).
As P ∩ US = {0} and N ∩ US = {0}, we have
max(p, n) ≤ 2g(S) − rk(US ) ≤ g(S) + g(F ),
where the last inequality follows from equation (3.4.2). Hence
σ(K) = p − n = (p + n) − 2n ≥ 2g(S) − 2(g(S) + g(F )) = −2g(F ).
Similarly, −σ(K) ≥ −2g(F ), and the result follows. 
The signature of a knot can also be computed from an unoriented spanning
surface for the knot due to the work of Gordon and Litherland [16].
Definition 3.4.10. Let S be a compact, connected, embedded surface in S 3 .
The Gordon–Litherland pairing
h , iS : H1 (S) × H1 (S) → Z
is defined as follows: Let a, b ∈ H1 (S), represented by oriented multi-curves α,
β ⊂ S. Consider the unit normal bundle pS : U N (S) → S of S in S 3 . Then
hα, βiS := lk(α, p−1
S (β)).

The Gordon–Litherland pairing is symmetric and bilinear. It agrees with the

symmetrised Seifert form when S is orientable. Let {b1 , . . . , bn } be a basis of H1 (S).
Then the Goeritz matrix GS is an n × n symmetric matrix with (i, j)-th entry
hbi , bj iS for i, j ∈ {1, . . . , n}. Furthermore, the normal Euler number e(S) of S is
defined to be −lk(K, K 0 ), where K 0 is the framing of K given by S. Gordon and
Litherland proved the following:
Theorem 3.4.11. Let S be an unoriented surface bounding the knot K in S 3 .
Then
e(S)
σ(K) = σ(GS ) + ,
2
where σ(GS ) is the signature of the Goeritz matrix.

3.5. Important classes of knots

In this section, we present several important classes of knots.
 3.5. IMPORTANT CLASSES OF KNOTS 57

3.5.1. Alternating knots.

Definition 3.5.1. The knot K is alternating if it admits a diagram D such that,
as we travel along D, we alternatingly encounter undercrossings and overcrossings.
Alternating knots have a number of nice properties. A large proportion of small
crossing knots are alternating, but their proportion tends to zero as the crossing
number goes to infinity.
Definition 3.5.2. We say that a crossing c of a knot diagram D is reducible
or nugatory if there is a circle C ⊂ R2 such that C ∩ D = {c}. The diagram D is
called reduced if it does not have a reducible crossing.
We can always remove such a crossing by applying ±π-rotation about a line in
R2 through the crossing to the part of D in the circle C. Consequently, if a diagram
realises the crossing number of a knot, it has to be reduced. A related operation
on knot diagrams is called a flype:
Definition 3.5.3. Let D be a knot diagram, and suppose that C ⊂ R2 is
a circle that intersects D in a crossing c and two additional points that are not
crossings. A flype on D is the operation of rotating the portion of D inside C by
angle ±π about a line in R2 passing through c such that we remove the crossing c,
but create a new crossing between the other two strands intersecting C.
The following results were conjectured by Tait. The first two statements were
shown by Kauffman [31], Murasugi [53], and Thistlethwaite [67] using the Jones
polynomial, and the third by Menasco and Thistlethwaite [40].
Theorem 3.5.4. Let D and D0 be reduced alternating diagrams of a knot K.
Then the following hold:
(1) c(D) = c(D0 ),
(2) w(D) = w(D0 ), and,
(3) if K is prime, we can obtain D0 from D using flypes.
In fact, any reduced alternating diagram of a knot has minimal crossing number.
Conversely, if K is a prime alternating knot, then any minimal crossing number
diagram of K is alternating. (This does not hold for nonprime knots; see for
example the square knot 31 # − 31 , which admits a nonalternating six-crossing
diagram.) The reduced alternating diagram of a prime alternating knot is unique
up to flypes according to statement (3), known as the Tait flyping conjecture. Since
a flype preserves both the crossing number and the writhe, statement (3) implies
statements (1) and (2).
Theorem 3.5.5. If we apply Seifert’s algorithm to a reduced alternating dia-
gram of a knot, we obtain a minimal genus Seifert surface.
3.5.2. Torus knots.
Definition 3.5.6. Let p and q be coprime integers. The (p, q)-torus knot Tp,q is
given by the curve on the standard torus T 2 in R3 that winds around the longitude
p times and the meridian q times.
In other words, Tp,q is the projection of the line h(p, q)i ⊂ R2 under the covering
map R2 → R2 /Z2 ∼= T 2 that sends (1, 0) to the longitude and (0, 1) to the meridian.
When p and q are not coprime and n is their highest common factor, we obtain
 3.5. IMPORTANT CLASSES OF KNOTS 58

Figure 12. The left shows the right-handed trefoil T2,3 , and the
right its mirror, the left-handed trefoil T2,−3 .

an n-component link, where each component is T(p/n,q/n) . For example, T2,2 is the
Hopf link.
When p = 1 or q = 1, the torus knot Tp,q is the unknot. The first nontrivial
torus knot is T2,3 , which is the trefoil knot, denoted 31 in the Rolfsen table; see
Figure 12.
Torus knots are typically far from being alternating. They are particularly
important as they arise from singularities of complex plane curves:
Exercise 3.5.7. The complex polynomial f (w, z) = wp +z q has a critical point
at the origin of C2 . Let the plane curve Vf ⊂ C2 be the zero-set of f . Show that,
for ε > 0 sufficiently small, the link of the singular point of Vf at the origin, defined
as Vf ∩ Sε3 , is the torus knot Tp,q .
The torus knots Tp,q and Tq,p are equivalent. Indeed, there is an orientation-
preserving automorphism of S 3 that swaps the solid tori on the two sides of T 2 ,
and interchanges the longitude and meridian. The torus knots Tp,q and Tp,−q are
mirror images.
Proposition 3.5.8. Let p, q > 0. Then
(p − 1)(q − 1)
c(Tp,q ) = min{(p − 1)q, (q − 1)p} and g(Tp,q ) = g4 (Tp,q ) = .
2
Exercise 3.5.9. Compute the signature of the right-handed trefoil using Seifert
matrices, and also using the formula of Gordon and Litherland from the checker-
board surfaces.
3.5.3. Satellite knots.
Definition 3.5.10. Let K 0 be a knot in the solid torus S 1 × D2 that does not
lie in a 3-ball, and is not isotopic to S 1 × {0}. Furthermore, let K be a knot in S 3 ,
called the companion, and n ∈ Z a framing coefficient. Choose a diffeomorphism
f : S 1 × D2 → N (K) representing the framing n. Then the n-twisted satellite of K
with pattern K 0 is f (K 0 ), which we denote by K(K 0 , n). We say that J is a satellite
knot if J = K(K 0 , n) for K 6= U .
Example 3.5.11. If there is a point p ∈ S 1 such that |({p} × D2 ) ∩ K 0 | = 1,
then K(K 0 , n) = K#K 0 for any n, where we view K 0 as a knot in S 3 by identifying
S 1 × D2 with the standard solid torus in S 3 .
Example 3.5.12. If K 0 = Tp,q , then K(K 0 , 0) is called the (p, q)-cable of K.
The (n, 0)-cable of K is also known as the n-cable of K (note that Tn,0 is an
n-component unlink).
 3.5. IMPORTANT CLASSES OF KNOTS 59

Figure 13. The Whitehead knot in the solid torus.

Example 3.5.13. If W is the Whitehead knot in S 1 × D2 shown in Figure 13,

then K(W, n) is called the n-twisted Whitehead double of K.
The algebraic winding number of an oriented knot in S 1 × D2 is its signed
intersection number with {1} × D2 (or, equivalently, its homology class in H1 (S 1 ×
D2 ) ∼
= Z), while its geometric winding number is the minimal number of intersection
points with {1} × D2 counted without signs. The algebraic winding number of W
is zero, while its geometric winding number is two.
Exercise 3.5.14. If K is a Whitehead double, it has a genus one Seifert surface.
Use this to show that ∆K (t) = 1.
Satellite knots are characterised by the property that their exterior E contains
a torus T not parallel to ∂E that is incompressible; i.e., there is no homotopically
nontrivial simple closed curve on ∂E that bounds an embedded disk in E.
3.5.4. Hyperbolic links.
Definition 3.5.15. A link L in S 3 is hyperbolic if S 3 \ L admits a complete
hyperbolic metric; i.e., a Riemannian metric of constant sectional curvature −1.
The following result is a special case of Thurston’s hyperbolisation theorem:
Theorem 3.5.16. Every knot in the 3-sphere is either hyperbolic, a torus knot,
or a satellite, and these classes are mutually exclusive.
In other words, we can build all knots from hyperbolic knots and torus knots
using satellite operations. By the Mostow–Prasad rigidity theorem [50][56], which
we now state, the hyperbolic structure is unique, and so any geometric quantity one
can assign to the hyperbolic structure on the knot complement is a knot invariant:
Theorem 3.5.17. Let M and N be complete finite volume hyperbolic manifolds
of dimensions at least 3. Then any isomorphism π1 (M ) → π1 (N ) is induced by a
unique isometry from M to N .
The most fundamental hyperbolic knot invariant is the volume. Other invari-
ants describe the shape of the cusp, such as the cusp volume, the lengths of the
longitude and meridian, and the longitudinal and meridional translations.
The hyperbolic structure can often be found using the computer software
SnapPy [9], which can also compute many of the associated invariants.
Every complete, connected hyperbolic n-manifold has universal cover Hn , hence
can be written as Hn /Γ, where Γ is a torsion-free discrete group of isometries of
Hn . When n = 3, this isometry group is PSL(2, C). Hence hyperbolic 3-manifolds
can also be studied using group theory.
 3.6. THE KNOT GROUP 60

3.6. The knot group

Given a knot K in S 3 , its complement is S 3 \ K, and its exterior is S 3 \
N (K). The complement and the exterior uniquely determine each other. Knot
complements are considered more often when studying hyperbolic structures, while
the exterior has the advantage of being compact.
By a deep result of Gordon and Luecke [17], knots in S 3 are determined by
their exteriors up to equivalence:
Theorem 3.6.1. The isotopy class of a knot in S 3 is determined by the orien-
tation-preserving homeomorphism type of its exterior.
While the longitude of the knot is homologically determined by the half lives,
half dies lemma (Lemma 2.4.2), there are infinitely many possible choices for the
meridian. What Gordon and Luecke actually showed is that there is exactly one
gluing of a solid torus to the knot exterior that gives S 3 . The fact that the knots are
in S 3 is important, since there are different knots in lens spaces with homeomorphic
complements; see Bleiler–Hodgson–Weeks [3].
The fundamental group of the knot complement is a powerful knot invariant.
A presentation, called the Wirtinger presentation, can be computed as follows:
Theorem 3.6.2. Let D be a diagram of the knot K in S 3 . Label the components
of D by a1 , . . . , ac as we follow the orientation of K, where c is the number of
crossings of D, and we call crossing i the endpoint of the arc ai . If aj is the
over-strand at crossing i, let
(
ai aj−1 a−1
i+1 aj if crossing i is positive,
wi = −1 −1
ai aj ai+1 aj otherwise.
Then
h a1 , . . . , ac | w1 , . . . , wc i
is a presentation of π1 (S \ K). In fact, S 3 \ K is homotopy equivalent to a cell
3

complex with a single 0-cell, 1-cells a1 , . . . , ac , 2-cells attached along w1 , . . . , wc ,

and a single 3-cell.
Proof. It clearly suffices to construct the claimed cell decomposition. Let D3
be the upper hemisphere of S 3 , which we view as a zero-handle, with D lying on
its boundary.
Let N (ai ) ≈ I × [−1, 1] be a regular neighbourhood of ai in S 2 , where ai ≈
I × {0} and ai × {−1} lies to the right of ai . We attach a two-dimensional one-
handle to D3 along I × {−1, 1}, which we view as a tunnel over ai ; see Figure 14.
We orient its core from I × {−1} to I × {1}. We repeat this for each arc ai , after
which we have a space homotopy equivalent to a wedge of c circles.
Over crossing i, we then attach a two-cell of the form [−1, 1] × [−1, 1]. Its
boundary goes over ai in the positive direction, passes over the tunnel over aj in
the positive or negative direction depending on the sign of crossing i, goes over ai+1
in the negative direction, and finally goes over aj in the opposite direction. Hence,
the attaching map of the two-cell is precisely wi .
Finally, we attach the three-dimensional three-handle corresponding to the
lower hemisphere of S 3 , and obtain a CW decomposition of S 3 \ K of the re-
quired form. We remark that we can thicken the one-cells and two-cells to obtain
a handle decomposition of the knot exterior. 
 3.6. THE KNOT GROUP 61

ai

aj

wi
ai+1

Figure 14. Illustration of the proof of the Wirtinger presentation.

The zero-handle is the upper half-space, and the tunnels are one-
handles attached below the plane. The two-cell over crossing i is
attached from underneath along the curve wi shown in grey.

Exercise 3.6.3. Show using the Wirtinger presentation that H1 (S 3 \ K) ∼

= Z.
Use Alexander duality to compute H1 (S 3 \ L) for a link L in S 3 .
The knot group is not a complete knot invariant. For example, the square knot
31 # − 31 and the granny knot 31 #31 have isomorphic groups, where 31 is the right-
handed and −31 is the left-handed trefoil. However, the knot group together with
the subgroup π1 (∂M ), called a peripheral system, is a complete knot invariant:
Theorem 3.6.4. Let K and K 0 be knots in S 3 with exteriors M and M 0 , respec-
tively. If there is an isomorphism π1 (M ) → π1 (M 0 ) that induces an isomorphism
π1 (∂M ) → π1 (∂M 0 ), then K and K 0 are equivalent. If K and K 0 are prime, then
they are equivalent if and only if they have isomorphic groups.
Proof. As M and M 0 are 3-manifolds with boundary other than D3 , they are
Haken, and we can apply Waldhausen’s theorem (Theorem 2.4.4) to see that M
and M 0 are homeomorphic. To conclude that K and K 0 are equivalent knots, we
now invoke the theorem of Gordon and Luecke (Theorem 3.6.1). 

We can compute the Alexander polynomial of a knot from a presentation of its

fundamental group via Fox calculus: Given a free group F generated by g1 , . . . , gn ,
the Fox derivatives ∂/∂gi : Z[F ] → Z[F ] is a group homomorphism characterised
by the following axioms:
(1) ∂gi /∂gj = δij ,
(2) ∂e/∂gi = 0, and
(3) ∂(uv)/∂gi = ∂u/∂gi + u(∂v/∂gi ) for any u, v ∈ F .
Exercise 3.6.5. Deduce from the axioms that
∂u−1 /∂gi = −u−1 (∂u/∂gi )
for any u ∈ F .
 3.7. FIBRED KNOTS 62

Now suppose that h g1 , . . . , gn | r1 , . . . , rm i is a presentation of G := π1 (S 3 \ K).

Let F be the free group on g1 , . . . , gn , and
H := G/[G, G] = H1 (S 3 \ K) ∼
= Z.
Consider the homomorphism φ : Z[F ] → Z[H], and write t for the generator of H.
We form the matrix J whose (i, j)-th entry is φ(∂ri /∂gj ) for i ∈ {1, . . . , m} and
j ∈ {1, . . . , n}. Then the greatest common divisor of all (n−1)×(n−1) minors of J
agrees with the Alexander polynomial of K, up to multiplication by ±tk for k ∈ Z.
(Recall that the Alexander polynomial is only well-defined up to multiplication by
tk for k ∈ Z.)
Exercise 3.6.6. Using Fox calculus, compute the Alexander polynomial of the
trefoil knot.

3.7. Fibred knots

Definition 3.7.1. A knot K in a 3-manifold Y is said to be fibred if its com-
plement Y \ K is a fibre bundle over S 1 such that the closure of each fibre is a
Seifert surface of K.
Example 3.7.2. The simplest example of a fibred knot is the unknot U , since
S 3 \ U ≈ S 1 × B 2 . The first non-trivial example is given by the trefoil knot. The
figure eight knot is also fibred; we will shortly introduce a method that will allow
us to easily prove this.
Example 3.7.3. The torus knot Tp,q is fibred for any pair of relatively prime
integers p and q. Indeed, if we use the description of Tp,q as the link of the singularity
wp + z q at the origin in C2 given in Exercise 3.5.7, then
wp + z q
: S3 \ K → S1
|wp + z q |
is a fibration. Indeed, the restriction of this map to the knot exterior is a proper
submersion, which is hence a fibre bundle by Ehresmann’s fibration lemma (Exer-
cise 1.4.6).
More generally, suppose that f : C2 → C is a polynomial such that f (0) = 0,
df ∂f 2
∂z (0) = 0, ∂w (0) = 0, and 0 ∈ C is an isolated common zero of these functions.
−1
Then V := f (0) is an algebraic variety with an isolated singularity at 0. The
link of the singularity is the knot K := V ∩ Sε3 for some ε > 0 small. Then K is a
fibred knot with fibration g/|g| : S 3 \ K → S 1 , where g := f |S 3 \K . This is called
the Milnor fibration of the singularity. For more detail, see the book of Milnor [41].
The following result of Stallings [4] gives a complete characterisation of fibred
knots:
Theorem 3.7.4. Let K be a knot in S 3 . Then K is fibred if and only if the
commutator subgroup of π1 (S 3 \ K) is finitely generated.
Proposition 3.7.5. Let K be a fibred knot in S 3 . Then the Alexander polyno-
mial ∆K (t) is monic.
Proof. Let S be a fibre surface, and fix a basis B = {b1 , . . . , b2g } of H1 (S).
Then we can write the exterior of K as a mapping torus Sϕ , where ϕ is called the
 3.7. FIBRED KNOTS 63

monodromy of the fibration. Let V be the Seifert matrix and M the matrix of
ϕ∗ : H1 (S) → H1 (S) with respect to B. Note that
lk(bi , b+ + +
j ) = lk(bi , ϕ∗ bj ) = lk(ϕ∗ bj , bi ),

by translating bi and b+ +
j along the mapping cylinder (S × I)/(x,1)∼(ϕ(x),0) until bj
reaches S ×{1}. This implies that V = M T V T and V is invertible, so M = V −1 V T .
Hence, the Alexander polynomial
∆K (t) = det(V − tV T ) = det(V ) det(I − tV −1 V T ) = det(I − tM ).
As ϕ is an automorphism of S, the matrix M is invertible over Z, so the leading
coefficient of ∆K (t) is det(M ) = ±1. 
If K is fibred, the surface obtained by taking the closure of a fibre of the
fibration S 3 \ K → S 1 is a minimal genus Seifert surface of K, and is unique up to
isotopy relative to K.
Gabai introduced sutured manifolds to give a simple method to check whether
a knot is fibred.
Definition 3.7.6. A sutured manifold is a pair (M, γ), where M is a compact,
oriented 3-manifold with boundary, and γ is a collection of thickened, oriented
simple closed curves in ∂M that divide ∂M into two subsurfaces R+ (γ) and R− (γ)
that meet along γ. Furthermore, R+ (γ) is oriented as ∂M , while R− (γ) is oriented
as −∂M , and γ is oriented as the boundary of both R+ (γ) and R− (γ).
Given a knot K in an oriented 3-manifold Y and a Seifert surface S for K,
the sutured manifold complementary to S is given by the pair (M, γ), where M =
Y \ (S × [−1, 1]) where S × [−1, 1] is a product neighbourhood of S in Y such
that S × {1} lies on the positive side of S. Furthermore, γ = ∂S × [−1, 1] and
R± (γ) = S × {±1}.
We say that (M, γ) is a product sutured manifold if M = R × [−1, 1], γ =
∂R × [−1, 1], and R± (γ) = R × {±1} for some compact, oriented surface R.
Proposition 3.7.7. The knot K in S 3 is fibred with fibre S if and only if the
sutured manifold complementary to S is a product.
Proof. This is a straightforward corollary of the definitions. 
Given a sutured manifold (M, γ), a product disk in (M, γ) is a properly em-
bedded disk D ⊂ M such that |D ∩ γ| = 2. We can decompose (M, γ) along
D by taking M 0 := M \ N (D), and γ 0 is obtained by reconnecting the ends of
γ \ N (D) along D × {−1} and D × {1}, where we identify N (D) with D × [−1, 1].
D
Then (M, γ) (M 0 , γ 0 ) is called a product decomposition. The following result is
straightforward:
Proposition 3.7.8. The connected sutured manifold (M, γ) is a product if and
only if it admits a sequence of product decompositions terminating at a 3-ball with
a single suture on it.
This gives us a practical method for showing that a knot K in S 3 is fibred:
Choose a Seifert surface S for K, usually using Seifert’s algorithm. Then con-
sider the product sutured manifold (S × [−1, 1], ∂S × [−1, 1]) by thickening S, and
considering K as the suture. We then look for product disks in the complement
of (S × [−1, 1], ∂S × [−1, 1]). Decomposing the complement along such a disk D
 3.8. THE JONES POLYNOMIAL 64

S + − S × [−1, 1]

+
+ − + − ≈
−

Figure 15. A proof that the trefoil knot is fibred using sutured manifolds.

amounts to adding D × [−1, 1] to (S × [−1, 1], ∂S × [−1, 1]), and reconnecting the
sutures along the boundary. If, at the end of this process, we end up with D3 with
a single suture, then K is fibred, and S is a minimal genus Seifert surface that is a
fibre.
Example 3.7.9. Figure 15 illustrates the above procedure in the case of the
right-handed trefoil knot. Consider the Seifert surface S shown in the upper left.
Its positive side is on the left. Thickening S results in the genus two handlebody
H := S ×[−1, 1] on the right. The original knot is our suture γ on ∂H. This divides
∂H into R+ (γ) and R− (γ). By construction, this is a product sutured manifold.
We now focus on the complementary sutured manifol (M, γ), where M :=
S 3 \ Int(H). We have shaded two product disks in (M, γ) on the bottom left. If we
decompose (M, γ) along these, then we obtain the sutured manifold in the middle
of the bottom row, which is diffeomorphic to D3 with a single suture.
Exercise 3.7.10. Use product decompositions to show that the figure eight
knot is fibred.

3.8. The Jones polynomial

Vaughan Jones [28] introduced a novel invariant of links in 1984 using von
Neumann algebras, which was the first new knot polynomial after the Alexander
polynomial. It admits a much simpler definition due to Kauffman [31]. Unlike the
Alexander polynomial, its relationship with the geometry of the link is much less
clear. It is conjectured to detect the unknot. It was the main tool for proving Tait’s
conjectures (Theorem 3.5.4). It admits a categorification called Khovanov homology
– a homology theory whose graded Euler characteristic is the Jones polynomial –
that has recently been shown to detect the unknot by Kronheimer and Mrowka [36].
Let L be a link in S 3 . The Jones polynomial VL (t) is a Laurent polynomial in
1/2 −1/2
Z[t , t ] characterised by VU (t) = 1 and the oriented skein relation
(t1/2 − t−1/2 )VL0 (t) = t−1 VL+ (t) − tVL− (t),
where L+ , L− , and L0 are link diagrams that agree outside a single crossing, L− is
obtained from L+ by changing a positive crossing to a negative one, and L0 is the
 3.8. THE JONES POLYNOMIAL 65

L+ L− L0

L∞ L0 L1

Figure 16. The top row shows the three links featuring in the
oriented skein relation, and the bottom row the links in the unori-
ented skein relation.

oriented resolution of the crossing; see the top row of Figure 16. When the number
of component of L is odd, then VL (t) ∈ Z[t, t−1 ].
It is not clear from this description that there exists such a polynomial, and
that it is unique. We now give the definition of the Jones polynomial using the
Kauffman bracket, which satisfies a different type of skein relation.
Definition 3.8.1. Let L be an unoriented link diagram in S 2 . Then its Kauff-
man bracket hLi ∈ Z[A, A−1 ] is characterised by the following relations:
(1) h i = 1,
(2) hL t i = −(A2 + A−2 )hLi, and
(3) if the diagrams L∞ , L0 , and L1 are as in the bottom row of Figure 16,
then hL∞ i = A−1 hL0 i + AhL1 i (unoriented skein relation).
If a diagram L has n crossings, there are two ways of resolving each, giving
rise to 2n unlink diagrams. The Kauffman bracket hLi is then recursively obtained
as a combination of the bracket polynomials of these unlinks. A c-component
unlink has bracket (−A2 − A−2 )c−1 . To see that this is independent of the order
of resolutions, one checks that the result is unchanged if we swap the order of two
adjacent crossings. The Kauffman bracket is not quite a link invariant:
Lemma 3.8.2. Let L be a link diagram in S 2 .
(1) Suppose that L0 is obtained from L via an R1 move. Then either the
0-resolution or the 1-resolution of L0 at the new crossing splits off an
unknot component. In the former case, hL0 i = −A−3 hLi, and, in the
latter, hL0 i = −A3 hLi.
(2) If L0 is obtained from L by an R2 or R3 move, then hLi = hL0 i.
Proof. Consider part (1), and assume that the 1-resolution splits off an un-
knot. If we set L∞ := L0 and apply the unoriented skein relation to it, then L0 = L
and L1 = L t , so we obtain that
hL∞ i = A−1 hLi + AhL t i = A−1 hLi − A(A2 + A−2 )hLi = −A3 hLi,
as claimed. The cased when the 0-resolution splits off an unknot is analogous.
Invariance under Reidemeister move R2 is obtained by applying the skein re-
lation to one of the two new crossings, followed by part (1).
 3.9. CONSTRUCTING 3-MANIFOLDS USING LINKS 66

Finally, R3-invariance can be seen by applying the skein relation to the topmost
crossing before and after the R3 move, and applying R2-invariance twice. We leave
the details as an exercise. 

It is immediate that the following is well-defined:

Definition 3.8.3. Let L be a link in S 3 and choose a diagram in S 2 . Then the

Jones polynomial VL (t) is obtained from (−A)−3w(L) hLi by substituting t1/2 = A−2 .

Unlike the Alexander polynomial, the Jones polynomial is hard to compute, and
current algorithms are exponential time in the crossing number n of the diagram
(corresponding to the 2n resolutions).
It is a simple consequence of the above characterisation of the Jones polynomial
that V−K (t) = VK (t−1 ). Hence, if VK (t) 6= VK (t−1 ), then the knot is chiral ; i.e.,
K 6= −K. For example, V31 (t) = t + t3 − t4 is not a symmetric polynomial, hence
the left- and right-handed trefoil are inequivalent knots. As for the Alexander
polynomial, the Jones polynomial is also alternating for alternating links, as shown
by Thistlethwaite.
It is clear from the definition that the span of the Jones polynomial gives
a lower bound on the crossing number of the link diagram (and hence on the
crossing number of the link). For alternating, reducible link diagram, Lickorish and
Millett [37] showed that the span is equal to the crossing number of the diagram.
This implies the first Tait conjecture (Theorem 3.5.4).

3.9. Constructing 3-manifolds using links

In this section, we introduce two methods for constructing 3-manifolds from
links in the 3-sphere: Dehn surgery and cyclic branched covers.
We have already encountered surgery in an arbitrary n-manifold M along
a framed, embedded k-sphere S. This described the outgoing boundary of the
handle attachment to I × M along {1} × S. Concretely, one removes a regu-
lar neighbourhood N (S) ≈ S k × Dn−k of S, and glues in Dk+1 × S n−k−1 along
∂N (S) = S k × S n−k−1 using the framing.
When we have a framed knot K in a 3-manifold Y , we are in the special
situation that ∂N (K) = S 1 × S 1 has a large automorphism group, namely SL(2, Z),
and we are not restricted to gluing via a framing. This leads to the notion of Dehn
surgery. As in the case of lens spaces, any automorphism of S 1 × S 1 that preserves
S 1 × {x} for some x ∈ S 1 extends to D2 × S 1 . Hence, the result of gluing D2 × S 1
to Y \ N (K) is determined, up to diffeomorphism, by the homology class of the
image of S 1 × {x} in ∂N (K). Given a slope α ∈ H1 (∂N (K)), we write Yα (K) for
the result of Dehn surgery along K with slope α.
If Y is a homology 3-sphere, then K is null-homologous in Y , and we have
a well-defined longitude l ∈ H1 (∂N (K)), which is the class null-homologous in
Y \ N (K). This exists by the half lives half dies lemma (Lemma 2.4.2). If we write
m ∈ H1 (∂N (K)) for the class of the meridian of K, the result of the surgery is
determined by an extended rational number p/q ∈ Q ∪ {∞}, where p and q are
coprime, and the image of S 1 × {x} is pm + ql ∈ H1 (∂N (K)). Gluing along a
framing of K corresponds to q = 1; i.e., to integer surgery. When p/q 6∈ Z, the
surgery does not correspond to a handle attachment.
 3.9. CONSTRUCTING 3-MANIFOLDS USING LINKS 67

More generally, one can consider surgery along a framed link whose components
are labelled by extended rational numbers. When the link is in a homology 3-
sphere, there is a canonical framing. We have the following result of Lickorish [38]
and Wallace [73]:
Theorem 3.9.1. Every closed, connected, and oriented 3-manifold Y can be
obtained by Dehn surgery along a link in S 3 . Furthermore, the surgery coefficients
can be chosen to be integers.
Proof. We have seen that the oriented cobordism group ΩSO 3 of 3-manifolds
vanishes. Hence Y bounds an oriented 4-manifold X. Choose a handle decompo-
sition of X. By cancelling 0- and 1-handles, we can assume that X has a single
0-handle. Similarly, we can assume that X has no 4-handles. Given a 1-handle,
we can cap off its core in the 0-handle to obtain an embedded S 1 . If we perform
surgery along it, we replace S 1 ×D3 with D2 ×S 2 , which corresponds to attaching a
2-handle. (In the language of Kirby calculus, this corresponds to replacing a dotted
circle representing a 1-handle with a 0-framed 2-handle.) If we turn the handle de-
composition upside down, we can similarly replace 3-handles with 2-handles. Hence,
X admits a handle decomposition with a single 0-handle and some 2-handles. This
means that M can be obtained by performing integer surgery along a link in S 3 . 
Given a 3-manifold M with a torus boundary component T , we can glue S 1 ×D2
to M along T . This operation is called Dehn filling. So one can think of Dehn
surgery as removing a neighbourhood of a knot, and Dehn filling the resulting
manifold with torus boundary.
Branched covers along links provide another useful operation for constructing
and studying 3-manifolds.
Definition 3.9.2. Let L be a k-component link in S 3 . Then the n-fold cyclic
branched cover Σn (S 3 , L) of S 3 along L is obtained by taking the n-fold cover of
S 3 \ N (L) corresponding to the kernel of the homomorphism
∼ Zk → Zn ,
π1 (S 3 \ N (L)) → H1 (S 3 \ N (L)) =
where the first map is abelianisation, and the second map is
(a1 , . . . , ak ) 7→ a1 + · · · + ak mod n.
We then fill the boundary with k solid tori, mapping their meridians to the lifts of
n times the meridians of each link component.
By a classical result of Alexander, every 3-manifold can be obtained as a
branched cover of the 3-sphere.
Definition 3.9.3. Let p, q, r ≥ 2 be integers. Then we define the Brieskorn
manifold
M (p, q, r) := { (z1 , z2 , z3 ) ∈ C3 : z1p + z2q + z3r = 0 } ∩ S 5 .
Exercise 3.9.4. Show that M (p, q, r) is a manifold of dimension 3.
The Brieskorn manifold M (p, q, r) is a homology sphere; i.e.,
H∗ (M (p, q, r)) ∼
= H∗ (S 3 ),
if and only if p, q, r are pairwise relatively prime. The manifold M (2, 3, 5) is
known as the Poincaré homology sphere, sometimes also denoted by ΣP . Poincaré
 3.9. CONSTRUCTING 3-MANIFOLDS USING LINKS 68

originally conjectured that every homology sphere is homeomorphic to S 3 , but

found this counterexample and reformulated his conjecture to homotopy spheres.
It can also be described by gluing the opposite faces of a dodecahedron with a twist.
The manifolds M (p, q, r) can also be described as branched coverings, due to
work of Milnor [45]:
Theorem 3.9.5. The Brieksorn manifold M (p, q, r) is homeomorphic to the
r-fold cyclic branched cover of S 3 , branched along the torus link Tp,q .
 CHAPTER 4

4-manifolds

4-manifold theory has a completely different flavour than 3-manifold topology.

This is the first dimension where the topological and the smooth classification
differ. Furthermore, since the Whitney trick, and hence the h-cobordism theorem
and surgery theory fail in dimension 4, there are numerous exotic phenomena not
present in higher dimensions.
By Theorem 1.1.13, every finitely presented group appears as the fundamental
group of a closed, smooth 4-manifold. Furthermore, there is no algorithm to decide
whether a finitely presented group is trivial. Hence, unless one restricts the fun-
damental group, there is no hope of obtaining a classification. Much of 4-manifold
topology focuses on simply-connected manifolds.
By the work of Freedman, the classification of simply-connected, topological
4-manifolds is governed by algebraic topology; namely, the intersection form on the
second cohomology group.
The classification of smooth 4-manifolds relies on invariants that can distin-
guish homeomorphic but non-diffeomorphic manifolds. One of the main tools for
constructing and manipulating 4-manifolds is Kirby calculus, which allows us to vi-
sualise a smooth 4-manifold as a framed link in the 3-sphere. Furthermore, it gives
a set of moves such that two 4-manifolds are diffeomorphic if and only if they are
related by a sequence of such moves. There are further constructions originating
from algebraic and symplectic geometry, such as blow-ups and fibre sums.
The two main invariants are the Donaldson and the Seiberg–Witten invariants.
These come from gauge theory. Roughly, one chooses a Riemannian metric on the
manifold, writes down some non-linear PDEs, and counts the solutions. Then one
shows this count is independent of the chosen metric. Heegaard Floer theory is a
more combinatorial version of Seiberg–Witten theory.

4.1. Kirby calculus

Kirby calculus gives us a way to represent and manipulate 4-manifolds using
framed links in the 3-sphere. If the 4-manifold has boundary, it can also be used
to study the boundary 3-manifold. More generally, one can even use it to study
4-dimensional cobordisms.
Let X be a smooth, connected 4-manifold (not necessarily closed). Then it
admits a handle decompositon with a single 0-handle. The 0-handle can be iden-
tified with D4 , which has boundary S 3 . Each 1-handle is attached along a framed
0-sphere, which one can represent by a pair of 2-spheres, with their boundaries
identified by an orientation-reversing diffeomorphism – usually a reflection.
Alternatively, one can visualise a 1-handle h1 as an unknotted circle with a
dot on it: If we attach a 2-handle h2 cancelling h1 , we can identify the resulting
manifold with D4 . The belt circle of h2 is an unknotted circle in S 3 , which we
69
 4.1. KIRBY CALCULUS 70

+1 −1 0 0

2
CP2 CP S2 × S2
2
Figure 17. Kirby diagrams for CP2 , CP , and S 2 × S 2 .

mark with a dot. If D is a 2-disk with boundary the dotted circle, we can push it
into D4 to obtain the co-core of h2 , and removing a neighbourhood of it amounts
to removing h2 , leaving us with h1 .
After attaching all the 1-handles, we obtain the boundary connected sum of a
number of copies of S 1 × D3 , whose boundary is #n S 1 × S 2 . The 2-handles are
attached along a framed link L in #n S 1 × S 2 . If we denote the 1-handles with pairs
of 2-spheres, a component of L is a collection of arcs in S 3 with endpoints on the
2-spheres, which are paired up using the reflections. In dotted circle notation, we
can visualise L as a usual link in S 3 . The framing can be encoded by a parallel
copy F of L. When there are no 1-handles, we can encode the framing by an
integer on each component of L, which gives the linking between the corresponding
components of L and F . It is unknown whether every closed, simply-connected
4-manifold admits a handle decomposition without 1-handles.
After the 2-handles are attached, we have a 4-manifold X2 with boundary a
3-manifold Y . Then the 1-handles with the framed link can be viewed as a diagram
of X2 , or as a diagram of the boundary 3-manifold Y . If X is closed, we further
have to attach some 3-handles and a single 4-handle. For this to be possible, Y has
to be a connected sum of copies of S 1 ×S 2 . If this is the case, by the following result
of Laudenbach and Poénaru, this uniquely determines X, up to diffeomorphism:

Proposition 4.1.1. Every automorphism of #k S 1 × S 2 extends to \k S 1 × D3 .

In other words, there is an essentially unique way to attach the 3- and 4-handles,
which we hence do not need to encode in the diagram.

Example 4.1.2. We now give examples of Kirby diagrams of some simply-

connected 4-manifolds; see Figure 17. One can obtain diagrams for connected sums
of these by taking their disjoint union.
The complex projective plane CP2 with the complex orientation is represented
by the unknot with framing +1. Indeed, if we remove a ball – thought of as a
4-handle – from CP2 , we are left with the total space of the tautological line bundle
over CP1 . (A point of CP1 is a complex 1-dimensional subspac L of C2 . The fibre
of the tautological line bundle over L is L itself.) A generic section intersects the
0-section algebraically once as two complex lines in CP2 intersect once positively.
Hence, the total space can be obtained by gluing two copies of D2 × D2 along
S 1 × D2 with framing +1. We identify one copy of D2 × D2 with the 0-handle B 4 .
Then the other is a 2-handle attached to B 4 along the unkot S 1 × {0} with framing
2
+1. We denote by CP the complex projective plane with its orientation reversed.
It can be represented by an unknot with framing −1.
 4.1. KIRBY CALCULUS 71

ri rj ri + rj ± 2lk(Li , Lj ) rj
Li Lj

Figure 18. Handle sliding the link component Li with framing ri

over the component Lj with framing rj .

Next, we describe a Kirby diagram for S 2 × S 2 . We write S 2 as the union of

2 2
the hemispheres S+ and S− . Then
S 2 × S 2 = (S−
2 2
× S− 2
) ∪ (S+ 2
× S− 2
) ∪ (S− 2
× S+ 2
) ∪ (S+ 2
× S+ ).
2 2 2 2 2 2 2 2
We view S− ×S− as a 0-handle, S+ ×S− and S− ×S+ as 2-handles, and S+ ×S+ as a
2 2
4-handle. The 2-handles are attached along {0} × ∂S− and ∂S− × {0}, respectively,
which form a Hopf link. Both components of the Hopf link have framing 0, since
the self-intersection of S 2 × {p} and {p} × S 2 is 0 for p ∈ S 2 .
By the Lickorish–Wallace theorem (Theorem 3.9.1), every closed, oriented 3-
manifold Y admits a Kirby diagram without 1-handles. Let L be the framed link
of this Kirby diagram, and we denote by ML the corresponding 4-manifold with
∂ML = Y . The boundary is unchanged under the following two Kirby moves:
(1) Blow-up: We add or remove an unknot with framing ±1 that is unlinked
from L. If L0 is obtained by adding a +1-framed unknot to L, then
2
ML0 = ML #CP2 . Similarly, ML0 = ML #CP if L0 is obtained from L by
adding a (−1)-framed unknot.
(2) Handle slide: We do a handle slide among the 2-handles. If Li and
Lj are the two components corresponding to the handles involved, with
framings Fi and Fj , respectively, then we choose a band connecting Li
and Fj (i.e., an embedding b : I × I ,→ ML such that b({0} × I) ⊂ Li and
b({1}×I) ⊂ Fj , while the rest of b(I ×I) is disjoint from L and F ), and we
replace Li with L0i := Li #b Fj . For k 6= i, we set L0k := Lk and Fk0 := Fk .
The framing of L0i is ri + rj ± 2lk(Li , Lj ), depending on whether the band
is orientation-preserving, where rk = lk(Lk , Fk ). For an illustration, see
Figure 18.
While blow-up changes the underlying 4-manifold, handle slides do not.
There are two S 2 -bundles over S 2 . Indeed, since D2 is contractible, the bundle
is trivial over the Northern and Southern hemispheres, and the bundle is defined by
the gluing map along the equator. This is given by an elment of π1 (SO(3)) ∼ = Z2 .
2 2 2 ∼ 2
We denote the non-trivial S -bundle over S by S × S .
Exercise 4.1.3. Using Kirby calculus, show that
2
(S 2 × S 2 )#CP2 ≈ CP2 #CP2 #CP .
∼ 2
Furthermore, S 2 × S 2 ≈ CP2 #CP .
The real power of Kirby calculus comes form the following result of Kirby [33]:
Theorem 4.1.4. Two Kirby diagrams without 1-handles represent diffeomor-
phic 3-manifolds if and only if they are related by a finite sequence of Kirby moves (1)
and (2).
 4.2. THE INTERSECTION FORM AND THE CLASSIFICATION OF 4-MANIFOLDS 72

There is also a Kirby calculus for 4-manifolds: Let X be a smooth 4-manifold,

and f0 , f1 two ordered Morse functions on it with a single index 0 critical point.
Then we can connect them with a generic 1-parameter family of smooth functions
ft for t ∈ I such that every ft is ordered for generic t, and has a unique index 0
critical point. If we consider the Kirby diagrams of X corresponding to ft , they
change using the following moves:
(1) An isotopy of the 1- and 2-handles.
(2) A 1-handle slides over a 1-handle, or a 2-handle slides over a 1- or 2-handle.
(3) An index 1-2 or 2-3 birth or death.
In dotted circle notation, an index 1-2 birth results in an unknotted 2-handle,
which is linked by a dotted circle. In general, one can cancel a 1- and a 2-handle
whenever the 2-handle passes through the 1-handle exactly once.
When we are considering Kirby diagrams of closed 4-manifolds; i.e., the bound-
ary of the 2-handlebody is a connected sum of copies of S 1 × S 2 that we fill with
3-handles and a 4-handle, an index 2-3 birth corresponds to the appearance of an
unknotted, 0-framed 2-handle.
For an in-depth study of 4-manifold topology from the point of view of Kirby
calculus, see the books of Gompf and Stipsicz [15] and Akbulut [2].

4.2. The intersection form and the classification of 4-manifolds

Let X be a closed, connected, oriented, and simply-connected 4-manifold. Then
H1 (X) = H3 (X) = 0. Hence H2 (X) and H 2 (X) are free abelian by the universal
coefficient theorem. We denote the fundamental class by [X] ∈ H4 (X). Then the
intersection form QX : H 2 (X) × H 2 (X) → Z of X is given by
QX (a, b) := ha ∪ b, [X]i
2
for a, b ∈ H (X). This is a non-degenerate, symmetric bilinear form. It is unimod-
ular ; i.e., induces an isomorphism
H 2 (X) → Hom(H 2 (X), Z) ∼
= H2 (X)
by Poincaré duality. Hence, its matrix in any basis is invertible over Z; i.e., has
determinant ±1.
Proposition 4.2.1. Let X be a 4-manifold, and s ∈ H 2 (X) a cohomology
class. Then there exists an oriented surface S properly embedded in X such that
[S] ∈ H2 (X) is Poincaré dual to s.
Proof. Since the Eilenberg–MacLane space K(Z, 2) is CP∞ , we have
H 2 (X) ∼
= [X, CP∞ ].
Then s corresponds to a continuous map fs : X → CP∞ such that f ∗ x = s, where
x is the generator of H 2 (CP∞ ) ∼
= Z[x], which is represented by the hypersur-
face CP∞−1 . We can homotope f such that it becomes smooth and transverse to
CP∞−1 . Then we let S := fs−1 (CP∞−1 ). 
If A and B are surfaces Poincaré dual to a, b ∈ H 2 (X), then QX (a, b) is the
algebraic intersection number of A and B. This justifies the term “intersection
form.”
We can easily read off the intersection form of a 4-manifold from its Kirby
diagram. Suppose that the 4-manifold X is given by a Kirby diagram without
 4.2. THE INTERSECTION FORM AND THE CLASSIFICATION OF 4-MANIFOLDS 73

1-handles and 3-handles; i.e., a framed link L = L1 ∪ · · · ∪ Ln in S 3 with framing

coefficient ri ∈ Z for i ∈ {1, . . . , n}. Then we obtain a basis b1 , . . . , bn of H2 (X) by
capping off a Seifert surface of Li with the core of the 2-handle attached along Li .
Furthermore, QX (bi , bj ) = lk(Li , Lj ) if i 6= j and QX (bi , bi ) = ri by Lemma 3.4.8.
We can also consider the intersection form on H 2 (X; R). Nondegenerate, sym-
metric bilinear forms over real vector spaces are completely determined by their
−
rank and signature. We write b+ 2 (X) and b2 (X) for the dimensions of maximal
positive and negative definite subspaces of H 2 (X; R). The signature of QX is
−
σ(X) = b+ 2 (X) − b2 (X). Signature is a cobordism invariant, and it vanishes if and
only if X is oriented 0-cobordant; see Section 1.7.
We say that QX is positive definite if b+ 2 (X) = b2 (X), negative definite if
−
b2 (X) = b2 (X), and is indefinite otherwise.
Definition 4.2.2. The intersection form QX is even if Q(x, x) is even for every
x ∈ H 2 (X), and is odd otherwise. We call this the type of X.
Definition 4.2.3. The Lie group Spin(n) is the connected double cover of
SO(n) for n ≥ 2. The oriented n-manifold M is spin if the structure group of T M
can be lifted from SO(n) to Spin(n). A spin structure on M is an equivalence class
of such lifts.
Proposition 4.2.4. The intersection form QX is even if and only if X is spin.
Proof. This is because the Spin structure of D4 extends to a 2-handle if and
only if the framing of the attaching 2-sphere is even, and the self-intersection of the
corresponding element of H2 (X) is precisely the framing. 
Indefinite intersection forms have a simple classification:
Theorem 4.2.5. Two indefinite, unimodular, symmetric bilinear forms over Z
are equivalent if and only if they have the same rank, signature, and type.
Not every value of the triple rank, signature, and type can be realised:
Proposition 4.2.6. The signature of an even form is divisible by 8.
However, this is the only restriction for indefinite forms. An important example
of a definite even form of signature 8 is the E8 form, given by the matrix
 
2 1 0 0 0 0 0 0
1 2 1 0 0 0 0 0
 
0 1 2 1 0 0 0 0
 
0 0 1 2 1 0 0 0
(4.2.1) E8 = 0 0 0 1 2 1 0 1 .

 
0 0 0 0 1 2 1 0
 
0 0 0 0 0 1 2 0
0 0 0 0 1 0 0 2
The 4-manifold PE8 with boundary defined by the Kirby diagram in Figure 19 has
intersection form E8 . Its boundary is the Poincaré homology sphere ΣP , which
is the Brieskorn manifold M (2, 3, 5); see Definition 3.9.3. By the work of Freed-
man [11], there is a topological 4-manifold ∆ with boundary ΣP that has the same
homotopy groups as D4 . The E8 manifold ME8 = PE8 ∪ ∆ is a closed topologyical
4-manifold with intersection form E8 .
 4.2. THE INTERSECTION FORM AND THE CLASSIFICATION OF 4-MANIFOLDS 74

+2

+2

+2 +2 +2

+2 +2

+2

Figure 19. A Kirby diagram for a 4-manifold whose intersection

form is the E8 lattice.

The hyperbolic form  

0 1
H=
1 0
has rank 2 and signature 0. (Note that QS 2 ×S 2 = H.) Hence, every symmetric,
unimodular, indefinite, even form is isomorphic to kE8 ⊕ lH for some k ∈ Z and
l ∈ Z+ . On the other hand, every indefinite odd form is isomorphic to k(1) ⊕ l(−1)
for k, l ∈ Z+ .
Not every symmetric, unimodular bilinear form can be the intersection form of
a smooth 4-manifold, due to the following result of Rokhlin [60]:
Theorem 4.2.7. Let X be a closed, smooth, spin 4-manifold (i.e., QX is even).
Then σ(X) is divisible by 16.
If M is a closed topological n-manifold, then the Kriby–Siebenmann class
κ(M ) ∈ H 4 (M ; Z2 ) vanishes if M admits a PL structure. When dim(M ) ≥ 5,
the converse also holds. For further details, see the book of Kirby and Sieben-
mann [34]. When X is a spin 4-manifold,
κ(X) ≡ σ(X)/8 mod 2.
A 4-manifold has a PL structure if and only if it has a smooth structure, and so a
smooth 4-manifold has κ(X) = 0.
Example 4.2.8. An example of a smooth spin 4-manifold with signature −16
is the K3 surface. It can be constructed as the solution to
x4 + y 4 + z 4 + w 4 = 0
in CP3 . Its intersection form is −2E8 ⊕ 3H.
From the point of view of complex geometry and algebraic geometry, there are
many different K3 surfaces, but they are all diffeomorphic by the work of Kodaira.
It follows from Rokhlin’s theorem (Theorem 4.2.7) that a closed topological 4-
manifold with intersection form E8 (e.g., ME8 ) does not admit a smooth structure.
By a celebrated result of Freedman [11][12], the pair (QX , κ(X)) is a complete
invariant of topological, simply-connected 4-manifolds X:
 4.2. THE INTERSECTION FORM AND THE CLASSIFICATION OF 4-MANIFOLDS 75

Theorem 4.2.9. Two closed, simply-connected topological 4-manifolds are home-

omorphic if and only if they have isomorphic intersection forms and the same
Kirby–Siebenmann class.
Conversely, for any symmetric, unimodular bilinear form Q over Z, there is a
unique topological 4-manifold X with QX = Q if Q is even, and two if Q is odd,
distinguished by κ(X).

Note that, when QX is odd and κ(X) = 1, then X does not admit a smooth
structure. However, κ(X) = 0 does not guarantee that X is smoothable.
In Theorem 4.2.5, we saw that the indefinite forms have a simple classification.
On the other hand, there is a wild world of definite forms. The following surprising
result of Donaldson [10] implies that most of these forms do not arise as intersection
forms of smooth 4-manifolds:

Theorem 4.2.10. Let X be a closed, smooth, simply-connected 4-manifold such

that QX is positive (or negative) definite. Then QX is diagonalisable; i.e., isomor-
phic to n(1) (or n(−1)).

Combined with Freedman’s theorem, we obtain that every smooth, definite

2
4-manifold is homeomorphic to either #n CP2 or #n CP .
By Theorem 4.2.5, indefinite forms are determined by their rank, signature, and
type. In particular, the indefinite odd forms are k(1) ⊕ l(−1), so every 4-manifold
2
with such an intersection form is homeomorphic to kCP2 #lCP for k, l > 0.
What remains open is the geography problem for smooth, indefinite, even (i.e.,
spin) 4-manifolds. Namely, which of these intersection forms are represented by
smooth 4-manifolds. Up to homeomorphism, they are all determined by their rank
and signature. And we have seen that they have intersection forms 2kE8 ⊕ lH
for k ∈ Z and l > 0. We can suppose that k ≥ 0 by possibly reversing the
orientation of the manifold. If l ≥ 3k, then 2kE8 ⊕ lH is the intersection form of
kK3#(l − 3k)(S 2 × S 2 ). The condition l ≥ 3k is equivalent to b2 (X) ≥ 11 8 σ(X).
11
The 8 -conjecture states that this condition is also necessary for 2kE8 ⊕ lH to be
the intersection form of a smooth, simply-connected 4-manifold. The best known
result, due to Furuta [14], states that b2 (X) ≥ 10
8 σ(X) for every smooth, indefinite,
even 4-manifold.
Another difficult question is the classification of smooth structures up to dif-
feomorphism on a given topological 4-manifold. We have seen that in higher di-
mensions, there are manifolds that admit more than one smooth structure, but the
number is always finite. In dimensions below 4, every manifold admits a unique
smooth structure. In contrast, Taubes has shown that R4 admits continuum many
non-diffeomorphic smooth structures. Note that Rn has a unique smooth structure
for n 6= 4.
If X is a closed topological 4-manifold, then it can be represented by a fi-
nite Kirby diagram, and hence admits at most countably infinite pairwise non-
diffeomorphic smooth structures. The K3 surface, for example, admits infinitely
2
many smooth structures, and so does CP2 #kCP for k ≥ 3. There is no known
example of a smooth 4-manifold that admits only finitely many smooth structures.
One of the most difficult open problems in topology is the smooth 4-dimensional
Poincaré conjecture, which asks whether S 4 admits a unique smooth structure. To
distinguish smooth structures, the main tool is the Seiberg–Witten invariant. For
 4.2. THE INTERSECTION FORM AND THE CLASSIFICATION OF 4-MANIFOLDS 76

an introduction, see the book of Morgan [49]. However, this is only defined when
b+
2 (M ) > 0.
For a thorough and entertaining overview of 4-manifold topology, see the book
of Scorpan [64].
 Bibliography

[1] Ian Agol. The virtual Haken conjecture. Doc. Math., 18:1045–1087, 2013. With an appendix
by Agol, Daniel Groves, and Jason Manning.
[2] Selman Akbulut. 4-manifolds, volume 25 of Oxford Graduate Texts in Mathematics. Oxford
University Press, Oxford, 2016.
[3] Steven A. Bleiler, Craig D. Hodgson, and Jeffrey R. Weeks. Cosmetic surgery on knots. In
Proceedings of the Kirbyfest (Berkeley, CA, 1998), volume 2 of Geom. Topol. Monogr., pages
23–34. Geom. Topol. Publ., Coventry, 1999.
[4] Gerhard Burde, Heiner Zieschang, and Michael Heusener. Knots, volume 5 of De Gruyter
Studies in Mathematics. De Gruyter, Berlin, extended edition, 2014.
[5] Andrew Casson. Three-dimensional topology. unpublished lecture notes.
[6] Jean Cerf. Topologie de certains espaces de plongements. Bull. Soc. Math. France, 89:227–
380, 1961.
[7] Jean Cerf. Sur les difféomorphismes de la sphère de dimension trois (Γ4 = 0). Lecture Notes
in Mathematics, No. 53. Springer-Verlag, Berlin-New York, 1968.
[8] Ralph L. Cohen. The immersion conjecture for differentiable manifolds. Ann. of Math. (2),
122(2):237–328, 1985.
[9] Marc Culler, Nathan M. Dunfield, Matthias Goerner, and Jeffrey R. Weeks. SnapPy, a
computer program for studying the geometry and topology of 3-manifolds. Available at
http://snappy.computop.org (23/05/2021).
[10] S. K. Donaldson. An application of gauge theory to four-dimensional topology. J. Differential
Geom., 18(2):279–315, 1983.
[11] Michael H. Freedman. The topology of four-dimensional manifolds. J. Differential Geometry,
17(3):357–453, 1982.
[12] Michael H. Freedman and Frank Quinn. Topology of 4-manifolds, volume 39 of Princeton
Mathematical Series. Princeton University Press, Princeton, NJ, 1990.
[13] D. B. Fuks and V. A. Rokhlin. Beginner’s course in topology. Universitext. Springer-Verlag,
Berlin, 1984. Geometric chapters, Translated from the Russian by A. Iacob, Springer Series
in Soviet Mathematics.
[14] M. Furuta. Monopole equation and the 11 8
-conjecture. Math. Res. Lett., 8(3):279–291, 2001.
[15] Robert E. Gompf and András I. Stipsicz. 4-manifolds and Kirby calculus, volume 20 of
Graduate Studies in Mathematics. American Mathematical Society, Providence, RI, 1999.
[16] C. McA. Gordon and R. A. Litherland. On the signature of a link. Invent. Math., 47(1):53–69,
1978.
[17] Cameron McA. Gordon and John Luecke. Knots are determined by their complements. J.
Amer. Math. Soc., 2(2):371–415, 1989.
[18] Wolfgang Haken. Theorie der Normalflächen. Acta Math., 105:245–375, 1961.
[19] Wolfgang Haken. Über das Homöomorphieproblem der 3-Mannigfaltigkeiten. I. Math. Z.,
80:89–120, 1962.
[20] A. J. S. Hamilton. The triangulation of 3-manifolds. Quart. J. Math. Oxford Ser. (2),
27(105):63–70, 1976.
[21] Allen Hatcher. Notes on basic 3-manifold topology. https://pi.math.cornell.edu/
~hatcher/3M/3Mfds.pdf, 2007.
[22] Allen E. Hatcher. The Kirby torus trick for surfaces. arXiv:1312.3518, 2022.
[23] Geoffrey Hemion. On the classification of homeomorphisms of 2-manifolds and the classifica-
tion of 3-manifolds. Acta Math., 142(1-2):123–155, 1979.
[24] John Hempel. 3-Manifolds. Princeton University Press, Princeton, N. J.; University of Tokyo
Press, Tokyo, 1976. Ann. of Math. Studies, No. 86.

77
 BIBLIOGRAPHY 78

[25] Morris W. Hirsch and Barry Mazur. Smoothings of piecewise linear manifolds. Princeton
University Press, Princeton, N. J.; University of Tokyo Press, Tokyo, 1974. Annals of Math-
ematics Studies, No. 80.
[26] William H. Jaco and Peter B. Shalen. Seifert fibered spaces in 3-manifolds. Mem. Amer.
Math. Soc., 21(220):viii+192, 1979.
[27] Klaus Johannson. Homotopy equivalences of 3-manifolds with boundaries, volume 761 of
Lecture Notes in Mathematics. Springer, Berlin, 1979.
[28] Vaughan F. R. Jones. A polynomial invariant for knots via von Neumann algebras. Bull.
Amer. Math. Soc. (N.S.), 12(1):103–111, 1985.
[29] Jeremy Kahn and Vladimir Marković. Counting essential surfaces in a closed hyperbolic
three-manifold. Geom. Topol., 16(1):601–624, 2012.
[30] Jeremy Kahn and Vladimir Markovic. Immersing almost geodesic surfaces in a closed hyper-
bolic three manifold. Ann. of Math. (2), 175(3):1127–1190, 2012.
[31] Louis H. Kauffman. State models and the Jones polynomial. Topology, 26(3):395–407, 1987.
[32] Michel A. Kervaire and John W. Milnor. Groups of homotopy spheres. I. Ann. of Math. (2),
77:504–537, 1963.
[33] Robion Kirby. A calculus for framed links in S 3 . Invent. Math., 45(1):35–56, 1978.
[34] Robion C. Kirby and Laurence C. Siebenmann. Foundational essays on topological mani-
folds, smoothings, and triangulations. Princeton University Press, Princeton, N.J.; Univer-
sity of Tokyo Press, Tokyo, 1977. With notes by John Milnor and Michael Atiyah, Annals of
Mathematics Studies, No. 88.
[35] Hellmuth Kneser. Geschlossen Flächen in dreidimensionalen Mannigfaltigkeiten. Jahres-
bericht der Deutschen Mathematiker Vereinigung, 38:248–260, 1929.
[36] P. B. Kronheimer and T. S. Mrowka. Khovanov homology is an unknot-detector. Publ. Math.
Inst. Hautes Études Sci., 113:97–208, 2011.
[37] W. B. R. Lickorish and K. C. Millett. The new polynomial invariants of knots and links.
Math. Mag., 61(1):3–23, 1988.
[38] William B. R. Lickorish. A representation of orientable combinatorial 3-manifolds. Ann. of
Math. (2), 76:531–540, 1962.
[39] Ciprian Manolescu. Lectures on the triangulation conjecture. In Proceedings of the Gökova
Geometry-Topology Conference 2015, pages 1–38. Gökova Geometry/Topology Conference
(GGT), Gökova, 2016.
[40] William W. Menasco and Morwen B. Thistlethwaite. The Tait flyping conjecture. Bull. Amer.
Math. Soc. (N.S.), 25(2):403–412, 1991.
[41] John Milnor. Singular points of complex hypersurfaces. Annals of Mathematics Studies, No.
61. Princeton University Press, Princeton, N.J.; University of Tokyo Press, Tokyo, 1968.
[42] John W. Milnor. A unique decomposition theorem for 3-manifolds. Amer. J. Math., 84:1–7,
1962.
[43] John W. Milnor. Morse theory. Based on lecture notes by M. Spivak and R. Wells. Annals of
Mathematics Studies, No. 51. Princeton University Press, Princeton, N.J., 1963.
[44] John W. Milnor. Topology from the differentiable viewpoint. The University Press of Virginia,
Charlottesville, Va., 1965. Based on notes by David W. Weaver.
[45] John W. Milnor. On the 3-dimensional Brieskorn manifolds M (p, q, r). In Knots, groups, and
3-manifolds (Papers dedicated to the memory of R. H. Fox), volume 84 of Ann. of Math.
Studies, pages 175–225. Princeton University Press, 1975.
[46] John W. Milnor and James D. Stasheff. Characteristic classes. Princeton University Press,
Princeton, N. J.; University of Tokyo Press, Tokyo, 1974. Annals of Mathematics Studies,
No. 76.
[47] Edwin E. Moise. Affine structures in 3-manifolds. V. The triangulation theorem and
Hauptvermutung. Ann. of Math. (2), 56:96–114, 1952.
[48] Edwin E. Moise. Geometric topology in dimensions 2 and 3. Springer-Verlag, New York-
Heidelberg, 1977. Graduate Texts in Mathematics, Vol. 47.
[49] John W. Morgan. The Seiberg-Witten equations and applications to the topology of smooth
four-manifolds, volume 44 of Mathematical Notes. Princeton University Press, Princeton, NJ,
1996.
[50] George D. Mostow. Strong rigidity of locally symmetric spaces. Princeton University Press,
Princeton, N.J.; University of Tokyo Press, Tokyo, 1973. Annals of Mathematics Studies, No.
78.
 BIBLIOGRAPHY 79

[51] James R. Munkres. Elementary differential topology, volume 1961 of Lectures given at Mas-
sachusetts Institute of Technology, Fall. Princeton University Press, Princeton, N.J., 1966.
[52] James R. Munkres. Topology. Prentice Hall, Inc., Upper Saddle River, NJ, 2000. Second
edition of [ MR0464128].
[53] Kunio Murasugi. Jones polynomials and classical conjectures in knot theory. Topology,
26(2):187–194, 1987.
[54] Richard S. Palais. Local triviality of the restriction map for embeddings. Comment. Math.
Helv., 34:305–312, 1960.
[55] C. D. Papakyriakopoulos. On Dehn’s lemma and the asphericity of knots. Ann. of Math. (2),
66:1–26, 1957.
[56] Gopal Prasad. Strong rigidity of Q-rank 1 lattices. Invent. Math., 21:255–286, 1973.
[57] Tibor Radó. Über den Begriff der Riemannschen Fläche. Acta Sci. Math. Szeged., 2:101–121,
1925.
[58] Kurt Reidemeister. Elementare Begründung der Knotentheorie. Abh. Math. Sem. Univ. Ham-
burg, 5(1):24–32, 1927.
[59] Kurt Reidemeister. Homotopieringe und Linsenräume. Abh. Math. Sem. Univ. Hamburg,
11(1):102–109, 1935.
[60] Vladimir A. Rohlin. New results in the theory of four-dimensional manifolds. Doklady Akad.
Nauk SSSR (N.S.), 84:221–224, 1952.
[61] Dale Rolfsen. Knots and links, volume 7 of Mathematics Lecture Series. Publish or Perish,
Inc., Houston, TX, 1990. Corrected reprint of the 1976 original.
[62] Horst Schubert. Knoten und Vollringe. Acta Math., 90:131–286, 1953.
[63] Matthias Schwarz. Morse homology, volume 111 of Progress in Mathematics. Birkhäuser
Verlag, Basel, 1993.
[64] Alexandru Scorpan. The wild world of 4-manifolds. American Mathematical Society, Provi-
dence, RI, 2005.
[65] Stephen Smale. Generalized Poincaré’s conjecture in dimensions greater than four. Ann. of
Math. (2), 74:391–406, 1961.
[66] Norman Steenrod. The topology of fibre bundles. Princeton Landmarks in Mathematics.
Princeton University Press, Princeton, NJ, 1999. Reprint of the 1957 edition, Princeton Pa-
perbacks.
[67] Morwen B. Thistlethwaite. A spanning tree expansion of the Jones polynomial. Topology,
26(3):297–309, 1987.
[68] René Thom. Quelques propriétés globales des variétés différentiables. Comment. Math. Helv.,
28:17–86, 1954.
[69] William P. Thurston. Three-dimensional geometry and topology. Vol. 1, volume 35 of Prince-
ton Mathematical Series. Princeton University Press, Princeton, NJ, 1997. Edited by Silvio
Levy.
[70] Bruce Trace. On the Reidemeister moves of a classical knot. Proc. Amer. Math. Soc.,
89(4):722–724, 1983.
[71] Friedhelm Waldhausen. Heegaard-Zerlegungen der 3-Sphäre. Topology, 7:195–203, 1968.
[72] Friedhelm Waldhausen. On irreducible 3-manifolds which are sufficiently large. Ann. of Math.
(2), 87:56–88, 1968.
[73] Andrew H. Wallace. Modifications and cobounding manifolds. Canad. J. Math., 12:503–528,
1960.
[74] H. Whitney. The self-intersections of a smooth n-manifold in 2n-space. Ann. of Math., 45:220–
246, 1944.