An introduction to knot theory

Anthony Conway

Contributors: Keita Allen, Ilani Axelrod-Freed, Robert Koirala, Jacob Lerma,

Jeffrey Lu, Ayodeji Lindblad, Abe Montes, Mohan Richter-Addo, Kimi Sun, Xunjing Wei,
Alex Weiler, Wendy Wu, Liane Xu, William Zhao, Margaret Zheng, Daniel Zhu

Spring 2022

1
Contents

1 Knots, links and invariants 6

1.1 First steps in knot theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.1.1 Links and link diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.1.2 Linking and unlinking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.1.3 The Jones polynomial . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2 Manifolds and an introduction to concordance . . . . . . . . . . . . . . . . . . . . 18
1.2.1 Manifolds, surfaces and embeddings . . . . . . . . . . . . . . . . . . . . . 18
1.2.2 Seifert surfaces and Seifert’s algorithm . . . . . . . . . . . . . . . . . . . . 23
1.2.3 Knot concordance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.3 Classical invariants via Seifert matrices . . . . . . . . . . . . . . . . . . . . . . . . 34
1.3.1 Seifert matrices and classical invariants . . . . . . . . . . . . . . . . . . . 34
1.3.2 Properties of the Alexander polynomial and the signature . . . . . . . . . 39
1.3.3 Applications to knot concordance . . . . . . . . . . . . . . . . . . . . . . . 44
1.3.4 The sign refined Alexander polynomial . . . . . . . . . . . . . . . . . . . . 47
1.4 More polynomials, braids and skein relations . . . . . . . . . . . . . . . . . . . . 51
1.4.1 The HOMFLY polynomial . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
1.4.2 Braids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
1.4.3 A too brief introduction to quantum topology . . . . . . . . . . . . . . . . 61

2 The algebraic topology of the link exterior 65

2.1 The link exterior and the link group . . . . . . . . . . . . . . . . . . . . . . . . . 66
2.2 The Alexander Module . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
2.3 Satellite knots and the limitations of classical invariants . . . . . . . . . . . . . . 74
2.4 Fox calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
2.5 A too brief introduction to twisted knot invariants . . . . . . . . . . . . . . . . . 80

3 Knots and 4-manifolds 83

3.1 Exotica and knot concordance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.2 Topologically but not smoothly slice knots . . . . . . . . . . . . . . . . . . . . . . 86

A A user’s guide to homology theory 90

B A bit of algebra 97

C Covering space theory 99

D Bibliography 101

2
Context and warning

These notes are based on a class that was taught during the spring of 2022 at MIT (18.904:
Seminar in Topology). The 16 students of this class each gave four talks (as groups of two)
based on a skeletal version of these notes (which initially only contained a prompt for each
presentation as well as which theorems and definitions to cover) and TeXed what they planned
to present. These notes were later substantially revised leading to the current document.
Despite these revisions, these notes are far from being polished and no doubt still
contain numerous innacuracies. Some caution is therefore needed while reading
what follows. I hope to be able to improve these notes further next time I teach a knot theory
class.

3
Introduction

One exciting aspect of knot theory, beyond its geometric and aesthetic appeals, lies in its variety:
knot theory blends combinatorial methods with topological techniques while also intersecting
with a field known as quantum topology. Here are some more down to earth reasons for learning
knot theory.

• Informally, a knot is obtained by gluing both ends of a (very thin) piece of string together.
One goal of this class is to go from this intuition to a more formal definition of the form “a
knot is a locally flat embedding of S 1 in S 3 ". This is what we work towards in Sections 1.1
and 1.2. More generally, Chapter 1 introduces the fundamentals of knot theory as well as
some first knot invariants.

• Knot theory is a part of topology and fits more broadly in the theory of manifolds. Starting
in Section 1.3 and throughout Chapter 2, we describe this perspective in more detail and
study how tools from algebraic topology apply to knot theory.

• Insights into knot theory lead to a better understanding of 3 and 4-manifolds, and vice
versa. In Chapter 3, we learn about some of these interactions by outlining how a result
in knot concordance implies the existence of non-standard differentiable structures on R4 .

Knot theory is also studied for its own sake. Indeed despite applications ranging from 4-
manifold topology to biology, many knot theorists will admit to studying knots with no external
motivation. Keeping this in mind, here are some questions that will drive parts of this class.

• How difficult is it to unknot a knot?

• What kind of surface can a knot bound in 3-space? How about in 4-space?

As the class progresses, we will both make these questions more precise and develop methods
to approach them.
On a more practical level, what is assumed throughout these notes?

• A formal prerequisite is an introductory class in topology. In particular, it is expected that

the reader be familiar with the following notions: topological spaces, homeomorphisms,
homotopy equivalences and the fundamental group.

• It is helpful to be familiar with some group theory. Since group theory is not a formal
prerequisite, some of the contents of this class may vary depending on the participants’
background in algebra.

• As the classes progresses, we will make occasional uses of homology theory, covering spaces,
rings and modules. These notions are introduced in Appendices A, B and C.

4
 Summarising, the goal of this class is to get a sense of some techniques and questions in
knot theory, while omiting the proofs of some difficult results that underpin the theory.
Finally, here are the main knot theory textbooks that were used to prepare these lecture
notes: [Lic97, Liv93, BZ03, Kaw96, Cro04, Rol90, Rob15]. When references are mentioned at
the beginning of a section, it is understood that the reader will most likely be able to read
more about the topic in any of these textbooks. Other references that were used include [Tu11,
KT08, Ada04].

5
Chapter 1

Knots, links and invariants

This first chapter introduces knots and links as well as some of the first tools, known as in-
variants, to differentiate links. In more detail, Section 1.1 introduces knots and links as well
as a powerful invariant known as the Jones polynomial. Section 1.2 discusses the necessary
terminology from manifold topology that will then allow us in Section 1.3 to study further link
invariants such as the Alexander polynomial and the signature. Section 1.4 will then introduce
even more invariants such as the HOMFLY polynomial while also mentioning braids.

1.1 First steps in knot theory

Knot theory is a subject whose first steps require surprisingly much care to treat carefully. The
more the reader is familiar with manifold toplogy, the more natural all the underlying concept
might seem. However, since we are only assuming point set topology, there are some subtleties
we will temporarily avoid before coming back to them in Section 1.2. Additionally, we will skip
some early proofs, not because they are overly difficult but simply for the same reason one does
not spend a semester studying set theory and logic before one’s first analysis class.
Once the definitions of links, ambient isotopy and diagrams are treated (in Subsection 1.1.1),
we will discover our first note invariants, namely the linking number (in Subsection 1.1.2) and
the Jones polynomial (in Subsection 1.1.3).

1.1.1 Links and link diagrams

We introduce knots and links and what it means for them to be equivalent. One of the main
references is [Cro04, Chapter 1]. Many knot theory textbooks start by defining knots and
links either as smooth or polygonal (i.e. piecewise linear) objects. A word about these different
approaches will be mentioned in Subsection 1.2.1 but the bottom line is that while all approaches
lead to the same theory, they also each have their advantages and inconvenients. Indeed, if one
works in the piecewise-linear setting, polygonal knots are not hard to define, but setting up
piecewise-linear homeomorphisms require a bit more work; on the other hand, working in the
smooth setting is much more natural but only once one is familiar with some notions from
differential topology. Working in the topological (locally flat) setting, as we do here, bypasses
these concerns provided the reader is familiar with point set topology.

Definition 1.1. A link is a subspace L ⊂ S 3 that is homeomorphic to a disjoint union of circles.

A link with a single component is called a knot.

Figure 1.1 depicts several knots and links such as the unknot, the trefoil knot, the figure
eight knot, the hopf link and the n-component unlink. Before spending more time with these
examples, some comments on Definition 1.1 are in order.

6
 Figure 1.1: Some first examples of knots and links

Remark 1.2. We make a couple of remarks on Definition 1.1.

• Knot theorists work with links in S 3 instead of in R3 mostly for convenience: the fact
that S 3 is compact simplifies a number of arguments. From a broader perspective, knot
theory can be studied in any 3-manifold, but this is beyond the scope of this class.

• Definition 1.1 is in fact too general as it allows for knots such as the one depicted in
Figure 1.2. To eliminate this kind of behavior, we say that a link is locally flat if for ev-
ery p ∈ L there exists an open ball B ⊂ S 3 containing p such that (B, L) is homeomorphic
as a pair to (D3 , D1 ). Knots that are not locally flat, e.g. the one in Figure 1.2, are called
wild.

Figure 1.2: A knot that is not locally flat.

Convention 1.3. From now on, all links are assumed to be locally flat.

Next, we need to formalise what it means for two links to be “the same".

Definition 1.4. Two links L1 and L2 are called ambiently isotopic if there is a continuous
map F : S 3 × [0, 1] → S 3 such that ft := F (−, t) : S 3 → S 3 is a homeomorphism for every t ∈
[0, 1] and such that f0 = idS 3 and f1 (L1 ) = L2 .

Since ft (L1 ) is a link for each t ∈ [0, 1] with f0 (L1 ) = L1 and f1 (L1 ) = L2 , we see that
ambient isotopy formalises the notion of being able to “deform one link into another". Note
that it is very common to hear knot theorists saying that two links are isotopic or equivalent or
of the same type instead of ambiently isotopic.

Remark 1.5. A quick remark that was not mentioned during class. It is known that two links
are ambiently isotopic if and only if there is an orientation-preserving homeomorphism h : S 3 →
S 3 that takes L1 to L2 . We do not give the full details of this fact, but here is the idea: in one
direction, if one has an ambient isotopy ht , then one can verify that h := h1 : S 3 → S 3 is the
required orientation-preserving homeomorphism; the converse takes more work.1
1
For the curious reader: the converse relies on the fact that every orientation-preserving homeomorphism
S → S 3 is isotopic to the identity." One reference for this fact is [Fis60].
3

7
Remark 1.6. One goal of knot theory is to understand whether two links are ambiently isotopic.
To prove that two links are not ambiently isotopic, the idea is to use invariants: to every
link L, one associates a quantity I(L) (e.g. an integer, a group, a polynomial,...) such that
if L is ambiently isotopic to L0 , then I(L) = I(L0 ). Thus, to show that two links L and L0 are
not ambiently isotopic, a knot theorist will typically pick their favorite invariant I and show
that I(L) 6= I(L0 ). Some invariants distinguish many links but are hard to calculate (e.g. the
homeomorphism type of knot complement S 3 \ K), while others are easier to calculate but do
not distinguish many links (e.g. the number of components of a link). One of the goals of this
class is to learn link invariants that strike a balance between these extremes.

Links are subsets of a 3-dimensional space and yet we always draw them in 2-dimensionally.
The next couple of propositions and definitions formalise and justify this idea.

Proposition 1.7. Given a link L ⊂ R3 ⊂ S 3 , there exists a plane Π ⊂ R3 such that the
orthogonal projection pΠ : R3 → Π satisfies the following property: for every x ∈ pΠ (L), the
set p−1
Π (x) contains either one or two elements.

Proof. Proofs can be found in [Cro04, Theorem 3.2.1 on page 52] as well as in [Kaw96, Propo-
sition 1.1.1 on page 7] and [BZ03, Proposition 1.12].

Figure 1.3: Projections as those mentioned in Proposition 1.7 are often referred to as regular
projections. This figure illustrates a regular projection of the trefoil knot. Source:http://pi.
math.cornell.edu/~mec/2008-2009/HoHonLeung/page3_knots.htm

Projections such as the ones described in Proposition 1.7 are called regular; see Figure 1.3
for an illustration. Points x ∈ pπ (L) such that |p−1
π (x)| = 2 are called double points. Figure 1.3
also illustrates that each double point can be represented as an overcrossing or an undercross-
ing depending on the height of the segments with respect to the projection. Summarising,
Proposition 1.7 states that every link L admits a regular projection R3 → R2 .

Definition 1.8. A link diagram is the image of a link L by a regular projection together with
under/over information at each double point.

Exactly as in the 3-dimensional setting, a (planar) ambient isotopy between two closed curves
c1 , c2 ⊂ R2 refers to a continuous function H : R2 × [0, 1] → R2 such that each ht := H(−, t) is
a homeomorphism and such that h0 = id and h1 (c1 ) = c2 .
The next proposition explains why no information is lost by working with diagrams: deciding
whether two links are equivalent can be determined by considering link diagrams.

Proposition 1.9. Given a diagram D for a link L and a diagram D0 for a link L0 , the following
assertions are equivalent:

8
 • the links L and L0 are ambiently isotopic;

• the diagrams D and D0 are related by a finite number of ambient isotopies in R2 and by
the following local moves known as Reidemeister moves:

Proof. A proof can be found in [Cro04, Theorem 3.8.1 on page 69] as well as in [Kaw96, Ap-
pendix A] and [BZ03, Proposition 1.14].

1.1.2 Linking and unlinking

One of the goals of knot theory is to develop invariants that are able to efficiently distinguish
knots and links; recall Remark 1.6. In this section, we introduce our first invariant, known
as the linking number, and apply it both to the detection problem and to the study of the
unlinking number. References include [Lic97, page 7 and 11].

Definition 1.10. An oriented knot is a knot together with the choice of one of the two direc-
tions that we can travel along it. An oriented link is a link with an orientation of each of its
components.

Figure 1.4: Examples of oriented links.

Remark 1.11. We collect some remarks on oriented links.

• The choice of an orientation is depicted using an arrow, both for links and for their
diagrams, as illustrated in Figure 1.4.

• Two oriented links are deemed equivalent as oriented links if they are equivalent via an
ambient isotopy that preserves the orientation of each component of the link.2

• An oriented version of Reidemeister’s theorem holds for oriented links. The relevant
oriented Reidemeister moves are shown in Figure 1.5. The fact that these four moves
suffice is a result of Polyak [Pol10, Theorem 1.1].

2
Note that an oriented link need not be equivalent to its reverse, the oriented link obtained from L by reversing
the orientation of each component of L.

9
 Figure 1.5: The oriented Reidemeister moves.

We now make the final preparations needed to introduce our first link invariant. Given an
oriented link diagram, the sign of a crossing c, denoted sgn(c) ∈ {±1} is defined using the rule
illustrated in Figure 1.6 below.
Definition 1.12. Let D be a diagram for two oriented knots K1 ∪ K2 ⊂ S 3 (for example two
components of a link). The linking number of K1 and K2 is defined as
1X
`k(K1 , K2 ) = sgn(c),
2 c

where the sum ranges accross all the crossings in D between (the projections of) K1 and K2 .

Figure 1.6: The sign of a crossing.

While we observe that `k(K1 , K2 ) = `k(K2 , K1 ), note that it is à priori unclear whether
the linking number `k(K1 , K2 ) is independent of the diagram D chosen to represent K1 ∪ K2 .
The next result shows that this is indeed the case leading to our first invariant.
Proposition 1.13. The linking number is a Z-valued invariant of oriented links. More pre-
cisely, if L = K1 ∪ . . . ∪ Kn and L0 = K10 ∪ . . . ∪ Kn0 are oriented ordered links that are equivalent
as oriented ordered links, then `k(Ki , Kj ) = `k(Ki0 , Kj0 ).

Proof. To check that the linking number is an invariant, the reader can verify that the definition
is invariant under the three oriented Reidemeister moves. We prove that `k(Ki , Kj ) is a an
integer. Let D be an oriented diagram for L. Write the linking number as
1
`k(Ki , Kj ) = (ni + nj ),
2
where ni (resp. nj ) denotes the number of crossings in D where the over arc belongs to Ki
(resp. Kj ). Observe that ni and nj are integers as well as the three following facts :

10
 1. The quantity z(L) := ni − nj is unchanged both by crossing changes and Reidemeister
moves. Here a crossing change refers to the move described in Figure 1.7.

Figure 1.7: A crossing change.

2. For the n-component unlink Un , this invariant vanishes: z(Un ) = 0.

3. Any link diagram can be turned into a diagram for the unlink by performing crossing
changes.
Combining these three facts, we deduce that ni = nj and therefore the linking number, which
equals 21 (ni + nj ) = ni = nj is an integer, as claimed.

Figure 1.8: An example of the linking number of two disjoint knots.

Example 1.14. Figure 1.8 shows a sample calculation of the linking number on the positive
and negative Hopf links: the result is ±1. It therefore follows from Proposition 1.13 that the
Hopf link is not equivalent to the unlink. Note that the linking number cannot distinguish all
2-component links from one another: for example, the Whitehead link and the trivial link both
have vanishing linking numbers (regardless of orientations) but are not equivalent (we can’t
prove this yet); see Figure 1.9.

Figure 1.9: The Whitehead link.

Next, we apply the linking number to the question of unlinking. In order to formalise what
it means to “unlink" a link, recall that a crossing change on a diagram refers to the local move
performed in Figure 1.7.
Definition 1.15. The unlinking number u(L) of a link L is the minimal number of crossing
changes needed to turn L into the unlink. More precisely, u(L) = minD u(D) where D ranges
across all diagrams for L and u(D) denotes the minimal number of crossing changes in D needed
to turn D into a diagram for the unlink.

11
Remark 1.16. We make some remarks on the unlinking number.

1. The unlinking number is a link invariant. This follows because the minimum is taken over
all diagrams for the link.

2. The unlinking number is finite: u(L) < ∞ for every link L. Indeed, as we mentioned
in the proof of Proposition 1.13, any link diagram can be turned into a diagram for the
unlink by performing crossing changes.

3. Finding an upper bound on the unlinking number u(L) requires finding crossing changes
that turn a diagram for L into a diagram for the unlink; this can be done fairly explicitly.
Finding lower bounds on the unlinking number is more challenging and therefore such
calculations still remain challenging problems to this day.3

Example 1.17. The unlinking number of an n-component link L is zero if and only if L is
(ambiently isotopic to) the n-component unlink. Figure 1.10 shows that the unlinking number
of the Hopf link H and the unknotting number of the trefoil knot are at most 1. In fact u(H) = 1
because H is not the unlink thanks to our unlinking number calculation. In the first problem
set, we will show that the trefoil knot T is non-trivial and so we also deduce that u(T) = 1.

Figure 1.10: An illustration of how to calculate upper bounds on the unlinking number. As
explained in Example 1.17, these figures explain why the unlinking number of the Hopf link and
the unknotting number of the trefoil knot are at most 1.

Proposition 1.18. The sum of the absolute values of the linking number is a lower bound on
the unlinking number: if L = K1 ∪ . . . ∪ Kn is a link, then for any orientation of L, the following
inequality holds: X
|`k(Ki , Kj )| ≤ u(L).
i<j

Proof. Note that I(L) := i<j |`k(Ki , Kj )| is an invariant of oriented links that satisfies |I(J)−
P

I(J 0 )| = 1 for any two oriented links J and J 0 that differ by a single crossing change. Now let
L = L(0) , . . . , L(u) be an unlinking sequence, i.e. a sequence of links where L(i) and L(i+1) differ
by a unique crossing change and L(u) is the unlink. The proposition now follows from the fact
that I(L(u) ) = 0, the triangle inequality and the bounded behavior of I(L) ≥ 0 under crossing
changes:
u−1
X u−1
X
I(L) = |I(L)| = (I(L(i) ) − I(L(i+1) )) ≤ |I(L(i) ) − I(L(i+1) )| ≤ u.
i=0 i=0

3
For example the calculation of the unlinking numbers of prime links up to 9 crossings was only completed
in 2015 [NO15] and there are still links with 10 crossings whose unlinking number is unknown [Bul18]. The
reader interested in the literature on unknotting numbers can consult https://knotinfo.math.indiana.edu/
descriptions/unknotting_number.html.

12
 As an example of Proposition 1.18, the reader is invited to verify that u(L) = 2 for the link
illustrated in the figure below:

Figure 1.11: A link with unlinking number 2.

1.1.3 The Jones polynomial

This subsection takes a leap forwards both in historical terms and in our capacity to distinguish
links from one another: we introduce the Jones polynomial. Despite the ease with which we
are able to define it, it is worth keeping in mind that the Jones polynomial was only discover
in the eighties [Jon85, Jon87] (using von Neuman algebras) and reinterpreted in the form we
will learn about by Kauffman [Kau87]. The reader might therefore keep in mind that for a long
time the only polynomial available to knot theorists was the Alexander polynomial that we will
learn about later during this class. References for the upcoming material include [Lic97, pages
23-26] and [Liv93, pages 217-220] and [Rob15, Chapter 4].

Definition 1.19. The Kauffman bracket is the unique function from unoriented link diagrams
in R2 (considered up to homeomorphism of R2 ) to Z[A−1 , A] that satisfies the following three
properties for any diagram D:

where # is the diagram of the unknot with no crossings and D t # is the disjoint union of D
with #.

We describe how to apply Definition 1.19 in practice and illustrate this with an example.
When looking at an (unoriented) link diagram, we can always rotate it so that a given crossing
has the âĂIJover strand" crossing left to right. It can be smoothed in the two ways, by âĂIJType
1" smoothings and “Type 2" smoothings; this is illustrated in the figure below:

13
 We find the bracket by using Definition 1.19 recursively, smoothing each intersection by
Type 1 and Type 2 smoothings. Then, for each of those partially smoothed knots, we continue
by smoothing the other crossings, multiplying by the appropriate factor from rule 2 when a
circle appears. After the procedure is completed, we are left with a sum of polynomials from
our individual cases, which are summed. We illustrate this with an example.
Example 1.20. We calculate the Kauffman bracket for the figure eight knot by recursively
applying Definition 1.19. A first application of the definition gives

Applying the definition a second time to smooth the crossings displayed in the figure and
factoring the result gives

Applying Definition 1.19 recursively until we obtain unlinks and using that h##i = −A2 −
A−2 and h#i = 1, we obtain that for the given diagram D of the Figure eight knot

hDi = A8 − A4 + 1 − A−4 + A−8 .

Lemma 1.21. The Kauffman polynomial is invariant under the second and third Reidemeis-
ter moves and when a link diagram is changed by a Type I Reidemeister move, its Kauffman
polynomial is multiplied by either −A3 or −A−3 .

Proof. We have already encountered this computation in Example 1.20, but we spell this out
for concreteness. However, since both Reidemeister moves are treated similarly, we only give
details in one case. Namely, using twice Definition 1.19, we obtain

Recall from Definition 1.4 that two links are equivalent (ambiently isotopic) if and only if any
two of their diagrams are related to one another by a sequence of Reidemeister moves. Therefore
Lemma 1.21 shows that the Kauffmann bracket is not a link invariant: it is not invariant under

14
the first Reidemeister move. The idea is now to “fix" this by introducing another quantity whose
behavior under the first Reidemeister move will “compensate" that of the Kauffman bracket.

Definition 1.22. The writhe w(D) of a diagram D of an oriented link is the sum of the signs
of the crossings of D, where each crossing has sign +1 or −1 following the rule in Figure 1.6.

A right-hand rule can be used to remember this convention, where you take your index finger
to follow the upper arrow, and your middle finger to follow the lower arrow. Then, whichever
way your thumb ends up pointing decides the sign: +1 for upwards and -1 for downwards.

Lemma 1.23. The writhe is unchanged by all oriented Reidemeister 2 and 3 moves and changes
as illustrated under oriented Reidemeister 1 moves:

Proof. The verifications are made using Figure 1.5 and the reader is invited to refer to it during
this proof. We first verify invariance under the second and third Reidemeister moves, before
moving on to the first.
For the second Reidemeister move, the writhe of the left diagram in Figure 1.5 is (locally) −1
from the top crossing plus +1 from the bottom crossing, making a (local) total of 0. The writhe
of the right diagram is also (locally) 0, since there are no crossings. Similar argument show that
all various of the second oriented Reidemeister 2 move leave the writhe unchanged.
The verification for the third Reidemeister is analogous: in both diagrams there are three
crossings. Going from left to right, we have −1, +1, and +1 for the left diagram. For the right
diagram, we have +1, +1, and −1. Thus, both diagrams have a writhe of 1, so Reidemeister 3
moves leave the writhe unchanged.
We conclude by studying the effect of one of the two oriented Reidemeister moves on the
writhe. For RIa, the orientation given in Figure 1.5 is an over then under. We see that the left
diagram is +1, the right diagram is 0, and if we introduced a over then under loop, as in RIb,
then the writhe would be −11. Thus, Type 1 Reidemeister moves change the writhe by ±1.

Next, we calculate the writhe on a couple of diagrams.

Example 1.24. The left-hand side of Figure 1.12 is a diagram D of a trefoil and the right-
hand side, while also depicting a diagram D0 for the trefoil, contains an additional Reidemeister
1 move. We see that for D, all three crossings result in pointing into the page, resulting
in w(D) = −3. However, for D0 , the three crossings from before are still into the page, and
there is now an additional crossing that also points into the page on the bottom right. Thus,
we have w(D0 ) = −4.

Figure 1.12: Two practice oriented knot diagrams for calculation. As explained in Example
1.24, the writhe of the left diagram D is w(D) = −3, and the writhe of the right diagram D0 is
w(D0 ) = −4.

15
 Using the writhe and the Kauffman bracket, we can now define an invariant of an oriented
link known as the Jones polynomial.

Definition 1.25. Let L be an oriented link, and let D be a diagram of L. The Jones polynomial
of L is  
V (L) = (−A)−3w(D) hDi −1 .
A=t 4

It can be verified that the Jones polynomial V (L) belongs to Z[t−1/2 , t1/2 ], i.e. it is a Laurent
polynomial in t1/2 . In fact, for knots, the Jones polynomial even belongs to Z[t−1 , t].4 At first
glance, it may seem like the Jones polynomial depends on our choice of D, but the next result
shows that it indeed does not.

Theorem 1.26. The Jones polynomial is an invariant of oriented links.

Proof. It suffices to prove that the Jones polynomial is invariant under the three oriented Reider-
meister moves (short-handed R1a, R1b,R2, R3 below). First, recall from Lemmas 1.21 and 1.23
that the Kauffman bracket and the writhe are both invariant under R2 and R3. It follows that
V (L) is invariant under R2 and R3.
We now examine the effect of a R1 move on the Jones polynomial. By Lemma 1.23, if a
diagram is changed by a Type I Reidermeister move (R1), its Kauffman bracket polynomial
changes as follows, respectively for right-handed and left-handed R1 moves:

= −A3 ,
= −A−3 .
We now only show that the R1a move does not change V (L), as the argument for the R1b
move follows similarly. Let D and D0 be two oriented diagrams of L, where D0 results from
performing R1a to D, thus adding a crossing to D. Since R1a creates an additional positive
crossing, w(D0 ) = w(D) + 1, we deduce that
0
(−A)−3w(D ) hD0 i = (−A)−3(w(D)+1) (−A3 )hDi
= (−A)−3 (−A)−3w(D) (−A3 )hDi
= (−A)−3w(D) hDi.

Reasoning similiarly for R1b„ the quantity V (L) = (−A)−3w(D) is invariant under all oriented
Reidermeister moves and is therefore an oriented link invariant.

Example 1.27. Using Definition 1.25, we can calculate the Jones polynomial for a link L by
directly calculating its Kauffman Bracket. Here are the Jones polynomials of some familiar
knots (where, as a shorthand, we directly indicate in the notation which diagram we are using
for the calculation):

) = (−A)0 h1i = 1.


1. V (

) = (−A)−3(−3) (A7 − A3 − A−5 ) = −t−4 + t−3 + t−1 .

2. V ( 1
A=t− 4

) = (−A)−3(3) (A−7 − A−3 − A5 ) = −t4 + t3 + t.

3. V ( 1
A=t− 4

By examples 2 and 3 above, we have shown that the left-handed trefoil knot and right-handed
trefoil knot are not the equivalent.
4
Note that some authors choose to work with the variable q = t1/2 .

16
 We can also calculate the Jones polynomial through the next proposition, also known as the
Skein relation.

Theorem 1.28. The Jones polynomial has the following properties:

1. V(unknot) = 1,

2. If L+ , L− and L0 are three oriented links, which are isotopic outside of a neighborhood of
a point, where they are as in Figure 1.13, then
1 1
t−1 V (L+ ) − tV (L− ) = (t 2 − t− 2 )V (L0 ).

Figure 1.13: L+ , L− and L0 for the Skein relation, from [Lic97, page 28].

Proof. We showed in Example 1.27 that V (unknot) = 1, so it suffices to prove the relation
between V (L+ ), V (L− ) and V (L0 ). This follows from the definitions of the Kauffman bracket,
the Jones polynomial and the writhe. In particular, by the definition of the Kauffman bracket,
we have

h i = Ah i + A−1 h  i,


h i = A−1 h i + Ah  i.


Manipulating these equations gives us

Ah i − A−1 h i = (A2 − A−2 )h i.



Using the definition of the Jones polynomial (from Definition 1.25), we can express each of
these Kauffman brackets in terms of the Jones polynomial and of the writhe:

AV (L+ ) A−1 V (L− ) (A2 − A−2 )V (L0 )

− = .
(−A)−3w(L+ ) (−A)−3w(L− ) (−A)−3w(L0 )

Since L+ has one more positive crossing than L0 , w(L+ ) = w(L0 ) + 1. Similarly, L− has one
more negative crossing than L0 , so w(L− ) = w(L0 ) − 1. Thus, the equation above simplifies to

−A4 V (L+ ) + A−4 V (L− ) = (A2 − A−2 )V (L0 ).

1
Substituting in A = t− 4 results in
1 1
t−1 V (L+ ) − tV (L− ) = (t 2 − t− 2 )V (L0 ),

as desired.

In fact, the conditions in Theorem 1.28 fully characterize the Jones polynomial. We omit
the proof but an interested reader can see [Lic97, pages 28-29] for reference.

17
 Figure 1.14: An example for using the Skein relation to find V (L0 ).

Example 1.29. We illustrate Theorem 1.28 with an example. Let L+ , L− and L0 be defined
as in Figure 1.14. Both L+ and L− are related to the unknot by a Reidemeister move I, so
V (L+ ) = V (L− ) = 1. Then by the Skein relation from Theorem 1.28, we obtain
1 1
(t 2 − t− 2 )V (L0 ) = t−1 V (L+ ) − tV (L− ) = t−1 − t.

Rearranging gives us the Jones polynomial for L0 :

t−1 − t 1 1
V (L0 ) = 1 = −(t 2 + t− 2 ).
− 21
t −t
2

1.2 Manifolds and an introduction to concordance

In the introduction, we mentioned two questions that drive these notes: how difficult is it to
unlink a link and what kind of surfaces can a link bound. We introduced the first question in
Section 1.1.2 and the goal of this section is to introduce the latter. We will proceed as follows.
First, Subsection 1.2.1 aims to improve our understanding of links and ambient isotopy, by
introducing some terminology related to manifolds and embeddings. Then, Subsection 1.2.2
studies surfaces in 3-space, while Subsection 1.2.3 is concerned with concordance and surfaces
in 4-space.

1.2.1 Manifolds, surfaces and embeddings

Now that we have introduced links and their diagrams as well as some combinatorial invariants,
we take a step back and spend a little more time with the definition of a link. There are at
least three reasons to do this. Firstly, this allows to note some of the connections between
knot theory and the broader field of topology. Secondly, this second take will lead us to a
better understanding of why the definition of “ambient isotopy" is the correct equivalence to
study. Thirdly, taking the time to underscore the difference between locally flat and smooth
embeddings will be helpful when we study concordance and 4-manifolds.

Manifolds

In this section, we define manifolds, present the classification theorem of surfaces. References
include [Rob15, MA21, FNOP19].
Definition 1.30. A (topological) n-manifold is a second countable Hausdorff topological
space M such that for every x ∈ M , there exists open neighborhood U 3 x with a corre-
sponding homeomorphism Φ : U → Rn .

The reader should note that some references ask for V to be an open subset of Rn instead
of Rn itself. A rapid verification shows that the two resulting definitions of a manifold are
equivalent.
Terminology 1.31. We collect some terminology that is often used when studying manifolds.

• The pairs (U, Φ) from Definition 1.30 are called charts.

18
 • For later reference, we briefly review the definition of a smooth manifold. Charts are called
smoothly compatible if for all i, j ∈ J, the following composition is a diffeomorphism

Φj ◦ Φi−1 : Φi (Ui ∩ Uj ) → Φj (Ui ∩ Uj ).

A collection of smoothly compatible charts {(Ui , Φi )}i∈I such that the Ui form an open
cover of M is called an atlas. An atlas A is maximal if every chart that is smoothly
compatible with all the charts of A itself belongs to A. A smooth manifold is a topological
manifold together with a maximal atlas (also kwown as a smooth structure).

• In what follows, a surface refers to a 2-manifold.

Example 1.32. Next, we describe several examples of manifolds.

• Any open subset of Rn is a manifold with a single chart. In particular, Rn itself is a

manifold.

• The n-dimensional sphere S n , given in coordinates by

{(x1 , . . . , xn+1 ) ∈ Rn+1 | x21 + x22 + . . . x2n+1 = 1},

is a manifold. For example, around each point in the 2-sphere shown on the left hand side
of Figure 1.15, a small enough open disk around it looks like an open subset of Rn , thus
the sphere is a manifold.

• It can be shown that the product of two (smooth) manifolds is a (smooth) manifold.

Figure 1.15: On the left: a sphere, on the right: a torus.

• The torus can alternatively described as the product S 1 × S 1 or for any r < R as
( q 2 )
(x, y, z) ∈ R3 x2 + y 2 − R + z 2 = r2 .

The first description has the advantage of making it immediate that the torus is a manifold
(use the previous two examples) while the second description has the advantage of being
very concrete; see also the right hand side of Figure 1.15.

• The connected sum M1 #M2 of two n-manifolds M1 and M2 is defined as

((M1 \ D1n ) t (M2 \ D2n )) / ∼,

where the equivalence relation identifies x with ϕ(x) for every x ∈ S n−1 , for some (fixed)
homeomorphism ϕ : S n−1 → S n−1 . 5

• The connected sum of g tori is denoted Σg . It can be checked that the connected sum
of Σm and Σn is homeomorphic to Σm+n . Figure 1.16 shows Σ2 .
5
The fact that this definition does not depend on the choice of ϕ is a very hard theorem, far beyond the scope
of this class. The interested reader can nonetheless look at [FNOP19, Theorem 5.11] to get a sense of the proof
which relies on the so-called “Annulus Theorem". In the smooth setting, the independence is less difficult (but
still highly non-trivial).

19
 Figure 1.16: A connected sum of two tori, also known as a genus two surface

• The cylinder, illustrated in Figure 1.17a, can be described as S 1 × [0, 1] or as

n o
(x, y, z) ∈ R3 | x2 + y 2 = 1, 0 ≤ z ≤ 1 .

Using this second description, a point (x, y, z) on the cylinder is contained in an open
neighborhood homeomorphic to Rn whenever z ∈ / {0, 1}. However, the points z = 0
or z = 1 admit no such neighborhood and therefore, the cylinder is not a manifold.
However, we can extend the definition a bit to include this case.

In order to obtain a notion which allows very natural spaces such as the cylinder to be
manifolds, we slighty generalise Definition 1.30.

Definition 1.33. A (topological) n-manifold with boundary is a second countable Hausdorff

topological space M such that for every x ∈ M , there exists an open neighborhood U 3 x with a
corresponding homeomorphism Φ : U → V where V is either Rn or Hn = {(x1 , . . . , xn )|xn ≥ 0},
in which case Φ(x) is required to lie on ∂Hn = {(x1 , . . . , xn )|xn = 0}.

Remark 1.34. We present some remarks on the above definition.

• Observe that each point on a manifold with boundary can satisfy exactly one of the items
in Definition 1.33, not both. The boundary of M , denoted as ∂M , consists of all points
satisfying the second statement.

• The boundary of an n-dimensional manifold is an (n − 1)-dimensional manifold.

• Smooth manifolds with boundary can be defined similarly; we omit the details.

• Compact manifolds without boundary are often called closed. This unfortunate terminol-
ogy is not related to the definition of a closed set from general topology.

Example 1.35. The Möbius strip is the quotient of the unit square [0, 1] × [0, 1] by the equiva-
lence relation which identifies (x, 0) with (1−x, 1) for every x ∈ [0, 1]. The space, which is shown
in Figure 1.17b, is a manifold with boundary but, unlike the manifolds from Example 1.32, it
only has one side; i.e. it is non-orientable.

The Möbius strip from Example 1.35 might remind the reader of other familiar surfaces
which “only have one side" such as the Klein bottle and the real projective space. During this
class however, we will actively try to avoid such “non-orientable surfaces" whence the following
definition.

Definition 1.36. A surface is orientable if it does not contain a Möbius strip.

Given the number of compact oriented surfaces presented so far, it is natural to ask if we
have listed them all. The next theorem completly answers this question.

20
 (b) A Mobius Strip [Ben]

(a) A Cylinder

Figure 1.17: Two examples of manifolds with boundary

Theorem 1.37 (Classification of orientable surfaces). For a fixed integer b ≥ 0, every compact
orientable surface with b boundary components is homeomorphic to the connected sum of g tori
with b discs removed.

For a proof of Theorem 1.37, we refer to [Rob15, Section 6.8].

Remark 1.38. We now make some remarks on Theorem 1.37.

• If a surface is homeomorphic to the connected sum of g tori with b discs removed, we say
that it has genus g. This way, Theorem 1.37 says that two compact oriented surfaces are
homeomorphic if and only if they have the same genus and the same number of boundary
components.

• From now on, we use Σg,b to denote the surface of genus g with b boundary components.
For a closed surface, having b = 0, we write Σg .

Embeddings

In Section 1.1, we defined knots as locally flat subspaces of S 3 . The main goal of this sub-
section is to understand how to reformulate this definition in terms of (locally flat) embed-
dings. The second goal is to briefly mention that knot theory could have been set up using
smooth embeddings instead of merely locally flat embeddings. References for this section in-
clude [Mun00, Bro62, Cro04] as well as [Tu11, Hir12] in the smooth setting.

Definition 1.39. A continuous map : X → Y between topological spaces is an embedding if X

and f (X) are homeomorphic under f , where f (X) ⊂ Y has the subspace topology.

Example 1.40. We give a couple of examples of embeddings.

• For m ≤ n, the map Rm → Rn , (x1 , . . . , xn ) 7→ (x1 , . . . , xm , 0 . . . , 0) is an embedding.

• For S 1 = [0, 1]/0 ∼ 1, the map S 1 → R2 , t 7→ (cos(2πt), sin(2πt)) is an embedding.

• The right hand side of Figure 1.18 illustates a map D2 → R3 that is not an embedding.

Remark 1.41. Embeddings are continuous injections, but the converse does not hold. For
instance, if we think of S 1 as the subspace of R2 given by pairs (x, y) satisfying x2 +y 2 = 1, then
the continuous injective map g : [0, 1) → S 1 ⊂ R2 , t 7→ (cos 2πt, sin 2πt) is not an embedding
because [0, 1) is not homeomorphic to S 1 .

21
Figure 1.18: The left hand side illustrates an embedding of a 2-sphere in R3 whereas the right
hand side of this figure shows (the image of) a map D2 → S 3 that is not an embedding.

Another example of an embedding is given by the “wild knot" depicted in Figure 1.2. While
this is (the image of) an embedding S 1 → S 3 , we now impose similar restrictions on embeddings
as we did on knots, in order to avoid wild embeddings.
Definition 1.42. An embedding f : M → N is locally flat if for every point x ∈ M there is a
neighborhood U of f (x) ∈ N such that (U, U ∩ f (M )) is homeomorphic to (Rn , Rm ).
Example 1.43. Here are a couple of examples of locally flat embeddings.

• According to Definition 1.1 (and Convention 1.3) knots are images of locally flat embed-
dings S 1 → S 3 . The local flatness condition here coincides with the one from Remark 1.2.
In particular all the knots we have learnt about so far give examples of locally flat em-
beddings.
• A locally flat embedding of a surface in S 3 requires that at every point, the embedding
looks like R2 in R3 . The embedding S 2 → S 3 on the left hand side of S 3 of Figure 1.18 is
locally flat whereas the curious reader could search for the “Alexander’s horned sphere" for
an example of an embedding S 2 → S 3 that is not locally flat. As we already mentioned,
the wild knot from Figure 1.2 is an embedding S 1 → S 3 that is not locally flat.

Now that we see how knots fit into the broader setting of manifolds and embeddings, we
now describe the analogous framework for ambient isotopies.
Definition 1.44. We say that two embeddings f, g : M → N are

• isotopic if there is a continuous function H : M × [0, 1] → N , called an isotopy, such

that Ht := H(−, t) : M → N is an embedding for all t ∈ [0, 1], H0 = f , and H1 = g.
• ambiently isotopic if there is a continuous function H : N × [0, 1] → N , called an ambient
isotopy, such that Ht := H(−, t) : N → N is a homeomorphism for all t ∈ [0, 1], H0 = idN ,
and H1 ◦ f = g.
Remark 1.45. We make a few remarks about isotopies and ambient isotopies.

• An isotopy is a one-parameter family of embeddings.

• If f and g are ambiently isotopic embeddings, then they are isotopic.
• The converse is false: while all knots are isotopic to one another (as we can shrink the
knotting via an isotopy as illustrated in Figure 1.16), not all knots are ambiently isotopic
to one another (for instance, we saw how the Jones Polynomial shows that the unknot
and the trefoil are not ambient isotopic).

22
 Figure 1.19: Isotopy shrinking a knot.

We now move on to the second main goal of this section: briefly desribe how one can define
knots using a couple of notions from differential topology. Since differential topology is typically
not a prerequisite to learning knot theory, we start off by collecting a couple of terms from that
topic without going into details.
Definition 1.46. Let M and N be smooth manifolds.

• A continuous map F : M → N is smooth if at every point p ∈ M there are charts (U, φ)

about p ∈ M and (V, ψ) about F (p) ∈ N such that the composition ψ◦F ◦φ−1 : Rn → Rm
is smooth at φ(p).

• A smooth map F : M → N is a smooth embedding if it is an embedding and if the

differential dFp : Tp M → TF (p) N is injective at every point p ∈ M .

• Two smooth embeddings f, g : M → N are

– smoothly isotopic if there is a smooth map H : M × [0, 1] → N , called a smooth

isotopy, such that Ht := H(−, t) is a smooth embedding for all t ∈ [0, 1], H0 = f ,
and H1 = g.
– ambiently isotopic if there is a smooth function H : N × [0, 1] → N , called a smooth
ambient isotopy, such that Ht := H(−, t) : N → N for alll t ∈ [0, 1], H0 = idM ,
and H1 ◦ f = g.
Remark 1.47. We make a few remarks about smooth isotopies.

• Smooth embeddings are particular cases of locally flat embedding which in turn are par-
ticular cases of embeddings. Similarly, note that smooth embeddings can be shown to be
locally flat and that smooth ambient isotopy is a generalization of ambient isotopy.

• While we saw that ambient isotopy and isotopy are two distinct notions, a result known as
the isotopy extension theorem ensures that two smooth embeddings are ambient isotopic
if and only if they are isotopic.

A smooth knot refers to a smooth embedding S 1 → S 3 and one can consider smooth knots
up to smooth ambient isotopy (or equivalently up to isotopy as mentionned in Remark 1.47).
The next result ensures that the the resulting theory is identical to the one we introduced in
Section 1.1.
Proposition 1.48. There is a bijective correspondence between the set of locally flat knots up
to ambient isotopy and smooth knots considered up to smooth ambient isotopy.

1.2.2 Seifert surfaces and Seifert’s algorithm

We now start to study one of our driving questions: what kind of surfaces can a given link bound
in 3-space? To answer this question, we introduce Seifert surfaces, and use Seifert’s algorithm
to show how to construct a Seifert surface for any given link. We then define the 3-genus which
gives a more quantitative way of formulating our driving question. References for this section
include [Lic97, Chapter 2].

23
Definition 1.49. A Seifert surface for an oriented link L in S 3 is a connected, compact, oriented
surface embedded in S 3 that has L as its oriented boundary.

Figure 1.20: Example Seifert surfaces.

Figure 1.51 shows Seifert surfaces for the unknot and for the 2-component unlink. An
oriented link L is always the boundary of multiple different Seifert surfaces. For example, given
any Seifert surface, gluing a hollow tube to the surface will preserve its boundary but increase
its genus, showing that it is a different Seifert surface for the link! Even more concretly, the
unknot bounds both the disk and any punctured genus g > 0 surface.
Remark 1.50. The idea of checkerboard surfaces, which we now decsribe, gives a simple method
which sometimes works to obtain a Seifert surface for a link L. Take a diagram of L and color
it in a checkerboard fashion (spaces sharing a corner but not an edge a line). One of the colors
can then be viewed as your shaded in surface.

Figure 1.21: A Seifert Surface for the Hopf link.

This is successful in Figure 1.21 (a diagram of the Hopf link). Notably this surface consists
of a band with two twists, so it is not a Möbius strip and is in fact oriented. However sometimes
the surface produced is not oriented, as in the following example:

Figure 1.22: The checkerboard method does not produce a Seifert Surface for the figure 8 knot.

This raises the question, can we find a Seifert surface for every link? It turns out we can.

24
Theorem 1.51. Every oriented link bounds a Seifert Surface.

Proof. To prove the theorem, we describe Seifert’s algorithm, which gives a way to construct a
Seifert surface for an oriented link L by performing the following steps:

1. Fix and draw an oriented link diagram of L (so that the underlying 4-valent graph is
connected).
2. Pick a point on the oriented diagram and trace around it in the direction of the orientation.
At crossings, switch which strand of the diagram you follow, continuing in the direction of
the orientation on this new strand. Continue this process until you return to your initial
point.
3. Pick a new point on an untraced portion of diagram and repeat the previous step. Repeat
until all of the diagram is traced.
4. You now have a number of loops traced out, each emerging from a single starting point
you picked, called Seifert circles. Fill in each circle to get disks.
5. Raise the disks to different heights in R3 , leaving them parallel to the xy-plane.
6. Connect the disks with twisted bands at the places where crossings were removed. The
twist should be in the direction such that the project onto the xy-plane looks the same as
the crossing which was removed.

The disks together with the bands connecting them form a compact surface with boundary
L. Since we chose a “connected" diagram, the surface produced this way is connected. It can
also be checked that the surface is oriented and so this concludes the proof of the theorem.

Figure 1.23: Example of following Seifert’s algorithm to obtain a Seifert surface for the 3-unlink.

Figure 1.24: Surface constructed by Seifert algorithm for the trefoil.

25
 The next result describes the genus of a Seifert surface produced by Seifert’s algorithm.

Proposition 1.52. Given a connected link diagram with n crossings, s Seifert cycles, and m
link components, Seifert’s algorithm yields a surface homeomorphic to Σ n−s−m +1,m .
2

Before we prove this result, we will first review a few facts about the Euler characteristic.
For a graph G with V vertices and E edges, its Euler characteristic is χ(G) = V − E. It can
be checked that this definition is consistent with the topological definition given in Section A
in terms of homology. As a result, since the Euler characteristic is invariant under homotopy
equivalence, we may determine the Euler characteristic of a surface by finding a graph to which
it is homotopy equivalent, and then computing the Euler characteristic of said graph. The proof
of Proposition 1.52 will also use the fact that χ(Σg,b ) = 2 − 2g − b, which can be deduced by
combining Definition A.8 and Example A.5.

Proof of Proposition 1.52. Suppose Seifert’s algorithm produces a surface S. Since S is con-
nected, compact, and oriented, Theorem 1.37 ensures that it is homeomorphic to Σg,b for some
integers g, b . Furthermore, the boundary of S is the original link, with m components. Since
the boundary of Σg,b consists of b components, we conclude that b = m.
We now proceed to compute the Euler characteristic of S. Seifert’s algorithm begins by
drawing s disjoint disks in space. Then a band is added between two disks at every crossing.
As a result, by contracting each disk to a point and band to an edge, we conclude that S is
homotopy equivalent to a graph with s vertices and n edges. Therefore χ(S) = s − n.
However, we also know that χ(S) = χ(Σg,m ) = 2 − 2g − m. Therefore s − n = 2 − 2g − m,
implying that g = n−s−m2 + 1, as desired.

Having now introduced Seifert surfaces, we may now introduce the main quantity of this
section:

Definition 1.53. The 3-genus g3 (L) of an oriented link L is the minimum genus of any Seifert
surface with oriented boundary L.

The 3-genus is an invariant since we are taking the minimum over all Seifert surfaces for
the given link. The reader can also check that there are links (such as the torus link T (2, 4)
mentioned on the second problem set) on which different orientations lead to different genera.
Including the word “oriented" in Definition 1.53 is therefore crucial.

Remark 1.54. We list a couple of remarks on Definition 1.53.

• Upper bounds on the 3-genus can be obtained by using Proposition 1.52: if an m-

component link L has an oriented diagram with n crossings and s Seifert circles,
then g3 (L) ≤ n−s−m
2 + 1. Lower bounds on the 3-genus are harder to come by.

• A knot has 3-genus 0 if and only if it is the unknot. This follows from the fact that, by
definition, a knot is the unknot if it bounds an embedded disc in S 3 .

• The previous remark is false if the word “knot” is replaced by “link” and “unknot” replaced
by “unlink”. A counterexample is the Hopf link whose genus is zero, see Example 1.55
below.

Example 1.55. We calculate the 3-genus on several examples.

1. The 3-genus of any unknot or unlink is 0, since removing m open disks from a sphere leads
to a connected, compact, and oriented surface with a boundary composed of m unlinked
circles. As we mentioned in Remark 1.54, the converse is true for knots.

26
 2. The Hopf link L has g3 (L) = 0 as it bounds a annulus; recall Figure 1.51.

3. We argue that the trefoil has genus 1. The standard diagram of the trefoil knot has 2 Seifert
cycles and 3 crossings. Therefore, by Proposition 1.52 we find that applying Seifert’s
algorithm produces a surface homeomorphic to Σ1,1 , implying that the 3-genus of the
trefoil is at most 1. Moreover, since the trefoil is knotted it cannot have 3-genus 0, by
Remark 1.54.

4. We argue that the figure eight knot has genus 1. The standard diagram of the figure eight
knot has 3-Seifert circles and 4 crossings. Therefore we find that the 3-genus of the figure
eight knot is at most 12 (4 − 3 − 1) + 1 = 1. The claim again follows from the fact that the
figure eight knot is knotted.

In the final portion of this section we discuss one more property of the genus. To do so we
need to define connected sum of knots.

Definition 1.56. The connected sum K1 #K2 , of two oriented knots K1 and K2 is defined as
follows. For i = 1, 2 pick a pair (Di3 , Di1 ) ∼
= (Di1 × Di2 , Di1 ) of 3-balls around xi ∈ Di such that
Di ⊂ Ki is an unknotted arc in Ki and set
˚3 , K \D
(S 3 , K1 #K2 ) := (S 3 \D ˚1 ) t (S 3 \D
˚3 , K \D
˚1 )/ ∼
1 1 1 2 2 2

where x ∼ y if and only if ϕ(x) = y where ϕ : (∂D13 , ∂D11 ) → (∂D23 , ∂D21 ) is a fixed orientation
reversing homeomorphism,

Figure 1.25: An illustration of the connected sum of knots

One can argue that the connected sum of knots is independent of the choices of the
arcs D11 , D21 and of the homeomorphism ϕ. Informally, the connected sum is defined by cutting
each knot and connecting them in a way that produces an oriented knot.
Connected sum is related to the genus via the following theorem.

Theorem 1.57. The 3-genus is additive: for any oriented knots K1 , K2 one has

g3 (K1 #K2 ) = g3 (K1 ) + g3 (K2 ).

Proof. A proof of this theorem can be found in [Lic97, page 17].

1.2.3 Knot concordance

In the previous section, we started to investigate the surfaces a knot can bound in 3-space. Now
we ask the same question in dimension 4. In this section, all knots are assumed to be oriented.
References for this section include [Lic97, page 86] and [Liv93, pages 25-26, 186-189, 193-195].

27
 Recall that a knot is a locally flat embedded circle in the 3-sphere S 3 and that S 3 can be
thought of as the boundary of the 4-dimensional ball D4 . Additionally, recall that an embedded
2-dimensional disk D in D4 with boundary a knot K ⊂ S 3 is said to be locally flat if for any point
x in the interior of D, there exists an open neighborhood U of x in B 4 and a homeomorphism
∼
→ R4 such that h(U ∩ D) ∼
=
h: U − = R2 ⊂ R4 .

Definition 1.58. A knot K ⊂ S 3 is (topologically) slice if it bounds a locally flat embedded

2-dimensional disk D in B 4 . Such a disk D is called a slice disk for K.

Remark 1.59. We make a couple of remarks about sliceness.

• The same definition can be made with smoothly embedded disks instead of locally flat
discs in which case we say that a knot is smoothly slice. A smoothly slice knot is topolog-
ically slice but there are topologically slice knots that are not smoothly slice (this latter
statement requires the work of two Fields medallists, Freedman and Donalson).

• Every knot K bounds a 2-dimensional disk in D4 : taking the cone on K provides the
required disk (which is not locally flat at the cone point). Therefore, if we removed the
condition on the disk being locally flat from Definition 1.58, then then every knot would
be slice.

Example 1.60. Our next goal is to give several examples of knots that are slice.

• The simplest example of a slice knot is the unknot. Indeed, since the the unknot bounds
a locally flat disc in S 3 it also bounds a locally flat disc in D4 .

• We will prove in Example 1.62 below that the knot 946 illustrated in Figure 1.26 is slice.

Figure 1.26: The knot 946 .

• In Subsection 1.3.3 below, we will see that the trefoil knot the figure eight knot are not
slice.

Figure 1.27: A band move.

We now explain how so-called band moves give a diagramatic way to prove that knots are
slice. While an illustration of a band move can be found in Figure 1.27, the first thing to note
is that performing a band move on a knot K results in a two component link K1 ∪ K2 (and
vice versa provided the band move “connects" K1 and K2 ). The key observation is that if K

28
and K1 ∪ K2 are a related by a band move, then they form the boundary of a genus 0 surface
P ⊂ S 3 × [0, 1] with 6
∂P = K t (K1 ∪ K2 ) ⊂ S 3 × {0} t S 3 × {1}.
The situation is again illustrated in Figure 1.28 and for visually evocative reasons, the genus 0
surface P with three boundary components is frequently referred to as a pair of pants.

Figure 1.28: If a knot K and a 2-component link K1 ∪ K2 are related by a band move, then
they form the boundary an embedded pair of pants P ⊂ S 3 × [0, 1]. The upper left figure shows
a local picture of the pair of pants; the middle uper figure shows the situation when K is the
unknot and K1 ∪ K2 is a 2-component link, the upper right picture shows an abstract pair of
pants. The bottom picture shows slices of the pair of pants as time progresses, where by “time"
we mean the [0, 1]-factor of S 3 × [0, 1].

We summarise these observations in the form of a lemma.

Proposition 1.61. A knot is slice if it can be converted into a 2-component unlink using a
single band move.

Proof. If the outcome L = K1 ∪ K2 of the band move on K happens to bound two disjoint
locally flat disks D1 ∪ D2 in D4 (e.g. if L is the 2-component unlink), then a slice disc D ⊂ D4
for K is obtained as
(D4 , D) = (S 3 × [0, 1], P ) ∪(S 3 ×{1},L) (D4 , D1 t D2 ).
In other words, the slice disc D is obtained by filling in the bottom of the pair of pants with
the two discs; this is illustrated in Figure 1.29 below.

We apply this method to prove that 946 is slice, as promised in Example 1.60.
Example 1.62. To prove that the knot 946 is slice, perform the following band move and
observe that the outcome is the 2-component unlink which we know to be slice.

6
The careful reader will note that at this point we are ignoring orientations and might want to come back to
this discussion once familiar with the notion of the reverse of a knot.

29
Figure 1.29: If knot can be converted into a 2-component unlink using a single band move, then
the resulting pair of pants can be capped off using the two disjoint discs that the unlink bound.

To describe further examples of slice knots, we introduce some terminology. The reverse of
a knot K, denoted −K, is K with the reverse orientation. The mirror image of K, denoted K,
is the reflection of K about some plane in R3 . If D is a diagram for a knot K, then a diagram
for K is obtained by switching the over/under information at each crossing in D.
The next proposition leads to many more examples of slice knots.

Proposition 1.63. For every knot K, the knot K# − K is slice.

Proof. Let B 3 be a small open 3-ball that intersects K in a small open unknotted interval A.
Observe that D := (K \ A) × [0, 1] is a disc that sits inside the 4-ball (S 3 \ B 3 ) × [0, 1] and that
the boundary of D is precisely K# − K because

∂D = (∂(K \ A) × [0, 1]) ∪ ((K \ A) × ∂[0, 1]) = K# − K.

Thus K#−K is slice with slice disk D ⊂ D4 as also schematically illustrated in Figure 1.30.

Figure 1.30: A schematic illustrating why K# − K is slice.

Next, we describe how slice knots allow us to define an abelian group whose elements are
equivalence classes of knots with the connected sum inducing the group law and the class of
the unknot as the identity element.7 Informally, this group will consists of the sets of knots
modulo those that are slice. To make this more formal, we introduce an appropriate equivalence
relation.

Definition 1.64. Two knots K1 and K2 are concordant if there is a locally flat embedding
S 1 × [0, 1] ,→ S 3 × [0, 1] of an annulus whose image C has boundary

∂C = K1 t −K2 ⊂ S 3 × {0} t S 3 × {1}.

7
Observe that the set of isotopy classes of knots is not a group under connected sum: a non-trivial knot K1
cannot have an inverse K2 , otherwise by Theorem 1.57 we would have 0 < g3 (K1 ) + g3 (K2 ) = g3 (K1 #K2 ) = 0,
a contradiction.

30
 As we mentioned above, the idea is that concordance will define an equivalence relation on
the set of knots, and that quotienting out by this relation will having the effect of considering
“knots modulo slice knots". Before getting into the details however, we first give some examples
of concordances.
Example 1.65. Here are some examples of concordances.

• For every knot K, the product K × [0, 1] ⊂ S 3 × [0, 1] gives a concordance from K to itself.

• Ambiently isotopic knots are concordant: if K1 and K2 are ambiently isotopic via an
ambient isotopy ht : S 3 → S 3 , then the required annulus is the image of the embedding
K1 × [0, 1] → S 3 × [0, 1], (p, t) 7→ (ht (p), t). This is the embedding where the slice S 3 × {t}
contains ht (K1 ).

The next proposition relates concordance to sliceness and, in doing so, implicitly gives many
more examples of concordant knots.
Proposition 1.66. Given knots K1 , K2 , the following assertions are equivalent:

1. K1 is concordant to K2 ,

2. K1 # − K 2 is slice.

In particular K is slice if and only if it is concordant to the unknot.

Proof. Assume that K1 and K2 are concordant via a concordance C in S 3 ×I. Pick an interval A
in K1 such that A × I ⊂ S 3 × I meets K2 ⊂ S 3 × {1} in an interval and such that the 4-
ball B := νA × I intersects S 3 × {0}, S 3 × {1} in 3-balls. As one can see in Figure 1.31,
the pair (S 3 × I \ B, C \ A × I) is homeomorphic to a pair (D4 , D), where D is a slice disk
for K1 # − K 2 .
The converse is essentially the same proof as so we omit the details: if D is a slice disk
∼ D3 × I, then appropriately gluing back in a (standard) copy of (D4 , D2 )
for K1 # − K 2 in D4 =
recovers a concordance in S 3 × I between K1 and K2 . The final sentence of the proposition
follows by writing K as K = K# − U and applying the statement we just proved.

Since our goal is to study “the set of knots modulo concordance", we had better verify that
concordance is an equivalence relation. We also check it respects the connected sum operation.
Lemma 1.67. Concordance is an equivalence relation on the set of knots. Moreover, the
connected sum is well-defined on the set of concordance classes of knots.

Proof. The relation is reflexive since K × I provides a concordance from a knot K to itself.
Symmetry is a consequence of the fact that a concordance between K1 and K2 is (tautologically)
a concordance between K2 and K1 . Finally, if K1 is concordant to K2 and K2 is concordant
to K3 , then a concordance from K1 to K3 is obtained by stacking the concordances one on top
of the other, see Figure 1.32.

31
Figure 1.31: Going from a concordance in S 3 × I to a slice disk in D4 .

Figure 1.32: Concordance is an equivalence relation.

32
 To show the second statement, assume that K1 is concordant to K2 via a concordance CK
and that J1 is concordant to J2 via a concordance CJ . Our aim is to provide a concordance C
between K1 #J1 and K2 #J2 . Picking intervals AK ⊂ K1 and AJ ⊂ J1 as in Figure 1.33, the
desired concordance C is CK \ (AK × I) ∪ CJ \ (AJ × I) (informally, we are “gluing the cylinders
to one eacher in the horizontal direction"). Everything plays out well on the level of ambient
manifolds, since the pair (S 3 × I, C) is homeomorphic to the pair (S 3 × I \ (νAK × I) ∪ S 3 ×
I \ (νAJ × I), CK \ (AK × I) ∪ CJ \ (AJ × I)). This concludes the proof of the lemma.

Figure 1.33: If K1 is concordant to K2 and J1 is concordant to J2 , then K1 #J1 is concordant

to K2 #J2 .

As a corollary, we are finally able to define the knot concordance group C.

Corollary 1.68. The set C of concordance classes of knots forms an abelian group under the
connected sum. The identity element is the class of the unknot while the additive inverse of [K]
is [−K].

Proof. Using Lemma 1.67, we know that concordance is an equivalence relation over which
connected sum is well-defined. Since connected sum is already associative and commutative on
isotopy classes of knots, it is still associative and commutative on C. The fact that the unknot is
the identity element for the connected sum also descends immediately to C. Finally the inverse
of a class [K] is given by [−K]. Indeed, by the third point of Proposition 1.66, K# − K is slice
i.e. concordant to the trivial knot, by the second point of Proposition 1.66.

Figure 1.34: Using band moves provides a good method to find upper bounds on the 4-genus
as abstractly illustrated on the left hand side of this figure. The right hand side figure uses two
band moves to show that for the trefoil K = 31 one has g4 (K) ≤ 1.

33
 We conclude this section by briefly mentioning a generalisation of the notion of sliceness.

Remark 1.69. The (topological) 4-genus g4 (K) of a knot is the minimal genus of a locally
flat surface Σ ⊂ D4 with boundary. Naturally, the definition can again be made in the smooth
setting (in which case, we speak of the smooth 4-genus g4sm (K)) we observe that a knot K has
g4 (K) = 0 (resp. g4sm (K) = 0) if and only if it is slice (resp. smoothly slice). Note that

g4 (K) ≤ g4sm (K) ≤ g3 (K)

since smoothly embedded surfaces are locally flat and because pushing a Seifert surface for a
knot into the 4-ball leads to a smoothly embedded surface in D4 with boundary K. Apart
from using the 3-genus, upper bounds on the 4-genus can again be obtained by performing
band moves, as illustrated in Figure 1.34. So once again, finding upper bounds on this quantity
is doable, while finding lower bounds is more challenging. Finally, we note that contrarily to
the 3-genus (recall Theorem 1.57 the 4-genus is not additive since g4 (K# − K) = 0 for any
knot K (by Proposition 1.63). The reader might however try to show that the 4-genus remains
subbadditive, i.e. g4 (K1 #K2 ) ≤ g4 (K1 ) + g4 (K2 ).

1.3 Classical invariants via Seifert matrices

At this point, we have introduced the topological quantities we are curious about: the unlinking
number, the 3-genus and the 4-genus. Additionally, we have seen that upper bounds for each
of these integers can be found using diagrams, but so far we have no lower bounds. The aim
of this section is to introduce invariants that will allow us to obtain such lower bounds, namely
the Alexander polynomial and the signature. In Subsection 1.3.1, we introduce Seifert matrices
and define the signature and Alexander polynomial, while Subsection 1.3.2 lists some properties
and applications of these two invariants. The main reference is [Lic97, Chapters 6 and 8].

1.3.1 Seifert matrices and classical invariants

We introduce Seifert matrices and use them to give a definition of the Alexander polynomial
and signature. From now on, all links are assummed to be oriented.
The 3-genus of a link gives us some information about the complexity of the surface a link
can bound. For example, we saw in the examples 1.55 that the Hopf link has genus zero,
meaning that it can bound an annulus. However, if we look at how the Hopf link bounds the
annulus in Figure 1.35, there is a twist that simply cannot be done away with.

Figure 1.35: On the left, a Seifert surface for the Hopf Link with the orientation colored in. On
the right, the curve a and its pushoff a+ .

34
 A natural way to capture this twist is to take a vector normal to the surface in the positive
direction and track how many times it winds around as we move around the surface. In the
picture above, tracking the normal vector along the circle shown, shows that it twists once. The
curve created by the tip of the normal vector is called the pushoff of the curve labeled a above.
The pushoff of a curve is denoted with a superscript, a+ . Another way to picture the pushoff
is just to lift the curve off of the surface in the positive direction. 8
Expanding this idea to generators of the first homology group of a Seifert surfaces leads to
the notion of the Seifert form.
Definition 1.70. The Seifert form of a surface F ⊂ S 3 is the bilinear form

β : H1 (F ) × H1 (F ) → Z
([a], [b]) 7→ `k(a, b+ ).

A Seifert matrix for F is a matrix representing its Seifert form, i.e. Aij = β([fi ], [fj ]) where
(fi )ni=1 is a basis for H1 (F ). A Seifert matrix for a link L is a Seifert matrix for any of its Seifert
surfaces.
Remark 1.71. We collect a couple of remarks on Definition 1.70.

• The Seifert form is well defined on homology: if two simply closed curves x, a (resp. y, b)
represent the same homology classes in H1 (F ), then `k(a, b+ ) = `k(x, y + ).

• Neither Seifert surfaces nor Seifert matrices are link invariants. Indeed we already saw
that any given link admits multiple non-isotopic Seifert surfaces (e.g. adding a hollow
tube increases the genus) and the situation is even worse for Seifert matrices since there
is an additional choice of a basis for homology as well as the choice of oriented curves
representing those homology classes.

We work through two examples below.

 
0 −2
Example 1.72. We show that −1 0 is a Seifert matrix for the knot K = 946 illustrated on
the left hand side of Figure 1.37. To achieve this, we use the Seifert surface for K illustrated
on the right hand side of that same figure.

Figure 1.36: The left hand side shows an oriented link diagram for K = 946 while the right
hand side shows a Seifert surface F for K.

We then choose two loops a, b whose homology classes generate H1 (F ) = ∼ Z2 , give them an
orientation, determine their pushoffs a+ , b+ and calculate the linking numbers to obtain the
required result: β([a], [a]) = `k(a, a+ ) = 0 and similarly β([a], [b]) = −2, β([b], [a]) = −1 and
β([b], [b]) = 0; all of this is illustrated in Figure 1.37 below.
8
More formally, any embedding F ,→ S 3 of an oriented surface extends to an embedding F × [−1, 1] ,→ S 3
withs F × {0} = F and now the push-off a+ of a curve a is simply a × {1} ⊂ F × {1}.

35
Figure 1.37: Cycles a, b generating homology, and their pushoffs a+ , b+ . On the right, we have
lifted them off of the surface F in the positive direction, for the reader’s viewing pleasure.

Since a fixed link can admit many different Seifert matrices, it is natural to ask if any two
Seifert matrices for a given link are nonetheless related. The idea is that Seifert matrices are
related to each other in a way analogous to how Seifert surfaces of a link are related. The
following theorem gives a more precise answer; a proof can be found in [Lic97, Theorem 8.4].

Theorem 1.73. Any two Seifert matrices for a link are related by a finite sequence of the
following moves:

1. Elementary enlargement/reduction: Given a square integer matrix A, an elementary en-

largement of A is a matrix B of the form
   
A η 0 A 0 0
 t
B =  0 0 1 or η 0 0 (1.1)
  
0 0 0 0 1 0

for some column vector η. Conversely, if B is a matrix of one of the forms given above
for some column vector η, we say A is an elementary reduction of B.

2. Unimodular congruences: Two integer matrices A, B are said to be unimodularly congru-

ent if A = P t BP for some integer matrix P such that det(P ) = ±1.

One says that the two matrices are S-equivalent if they related in such a way.

Remark 1.74. Here are a couple of remarks related to Theorem 1.73.

1. Unimodular congruence captures the idea that the matrix is dependent on the chosen
basis: if two matrices A, B represent a same bilinear form β, then they are congruent.

2. Elementary enlargement captures the idea that if F is a Seifert surface for a link L, then
so any surface F 0 obtained from F by attaching a hollow tube (which might link other
parts of F ).

3. Theorem 1.73 shows that while Seifert matrices are not link invariants, the S-equivalence
class of a link is an invariant. The S-equivalence class of the link therefore refers to the
S-equivalence class of any of its Seifert matrices.

36
 Next, we extract more manageable link invariants from the S-equivalence class of a link,
namely the Alexander polynomial and the signature. We will define them, prove that they are
link invariants, and calculate them for the two trefoil knots and 946 as examples. The references
for this section are [Liv93, Lic97, Cro04].
 
Definition 1.75. The Alexander polynomial ∆L of an oriented link L is ∆L (t) = det tA − AT
where A is any Seifert matrix for L.

Proposition 1.76. The Alexander polynomial of an oriented link, considered up to multiplica-

tion by ±tk for some integer k, is independent of the choice of Seifert matrix and is thus a link
invariant.

Proof. We know from Theorem 1.73 that any two Seifert matrices for a link are S-equivalent,
so proving that the Alexander polynomial is the same up to a factor of ±tk for any S-equivalent
A, B will give the desired result. It suffices to show that conjugating by an integer matrix
with determinant ±1 and adding rows and columns in the described way will not impact the
polynomial. We first assume that B = M AM T for an integer matrix M with determinant ±1.
We have

det(tB − B T ) = det(tM AM T − M AT M T )
= det(M (tA − AT )M T )
= det(M ) det(tA − AT ) det(M T )
= det(tA − AT ),

so we get the desired result in this case. The case where B is obtained by elementary enlarg-
ment from A (or A is obtained from elementary reduction from B) also follows from a direct
calculation: Indeed, since
   
A ξ 0 tA − AT tξ 0
 T T
B = 0 0 1  =⇒ tB − B =  −ξ T 0 t .
  
0T 0 0 0T −1 0

the determinant of tB − B T is simply the determinant of tA − AT multiplied by t, as can be

seen by expanding the last two rows and columns.

Figure 1.38: Surface constructed by Seifert algorithm for the left trefoil.

Example 1.77. We will now calculate the Alexander polynomial of some links. For each knot
K, we write A for a Seifert matrix of K and use it to calculate the Alexander polynomial of K.

37
 • The left-handed trefoil L has Seifert surface shown in Figure 1.38 which gives rise to
! !
1 0 t−1 1
A= =⇒ tA − AT = .
−1 1 −t t − 1

Hence, the Alexander polynomial is ∆L (t) = t2 − t + 1. Similarly, the right-handed trefoil

R has ! !
−1 1 T −t + 1 t
A= =⇒ tA − A = .
0 −1 −1 −t + 1

• For the knot 946 , using the calculation from Example 1.72, we have
! !
0 −2 0 −2t + 1
A= =⇒ tA − AT = .
−1 0 −t + 2 0

Hence, the Alexander polynomial is ∆946 (t) = (t − 2)(2t − 1).

Next, we use Seifert matrices to introduce the signature of a link. The signature of a real
symmetric matrix A is the number of positive eigenvalues of A minus the number of negative
eigenvalues. This is well-defined because real symmetric matrices have real eigenvalues. For
later use, we recall Sylvester’s theorem, which states that if real symmetric matrices A and B
satisfy B = M AM T for some invertible matrix M , then sign(A) = sign(B).
 
Definition 1.78. The signature of an oriented link L is σ(L) = sign A + AT , where A is any
Seifert matrix for L.
Proposition 1.79. The signature of an oriented link is independent of the choice of a Seifert
matrix and is thus a link invariant.

Proof. It suffices to show the signature is preserved under the S-equivalence from Theorem 1.73.
Under a change of basis of H1 (F ), the Seifert matrices A and B are related by a unimodular
congruence B = M AM T for some invertible M . Then,

B T = M AT M T =⇒ B + B T = M (A + AT )M T .

Thus, the signatures of the matrices are the same by Sylvester’s theorem.
We now check that elementary enlargements and reductions preserve signature. In this case,
the Seifert matrices A and B are related by
   
A η 0 A + AT η 0
T
B = 0T 0 1  =⇒ B + B =  η T 0 1
   
0T 0 0 0T 1 0

for some unknown vector η. We can apply the same row and column elimination moves to
eliminate η and η T . This amounts to multiplying elementary matrices and their transposes to
both sides.
 
A + AT 0 0 !!
T  T T 0 1
B + B = E  0T 0 1  E = E (A + A ) ⊕ ET .

1 0
0T 1 0

Now, by Sylvester’s theorem and the additivity of the signature under direct sums, we see that
!! !!
T 0 1
T

T
 0 1
sign(B + B ) = sign (A + A ) ⊕ = sign A + A + sign = sign(A + AT )
1 0 1 0

since the eigenvalues of the 2-by-2 matrix is ±1, one of each sign.

38
Example 1.80. We calculate the signatures of the examples given in the previous section.

• Recall that the left-handed trefoil L has Seifert matrix

! !
1 0 2 −1
A= =⇒ A + AT = .
−1 1 −1 2

The eigenvalues are 1 and 3, so σ(L) = 2.

• The right-handed trefoil R has Seifert matrix

! !
−1 1 T −2 1
A= =⇒ A + A = .
0 −1 1 −2

The eigenvalues are −1 and −3, so σ(R) = −2. Hence, signature can tell that the two
trefoils are not equivalent, as we saw with the Jones polynomial, but not the Alexander
polynomial.

• The knot 946 has Seifert matrix

! !
0 −2 0 −3
A= =⇒ A + AT = .
−1 0 −3 0

The eigenvalues are 3 and −3, so σ(946 ) = 0.

1.3.2 Properties of the Alexander polynomial and the signature

In this section, we present several properties of the Alexander polynomial and signature. The
highlights are Theorem 1.84 and Theorem 1.88 where we show that the Alexander polynomial
and signature respectively give lower bounds on the 3-genus and unlinking number of a link.
We start by describing properties of the Alexander polynomial. Recall that the Alexander
polynomial of an oriented link L is defined as ∆L (t) = det(tA − AT ), where A is any Seifert
matrix for L, and that it is independent of the choice of Seifert matrix up to multiplication
by ±tk for k ∈ Z. For ease of reading, given Laurent polynomials p, q ∈ Z[t±1 ], we write
.
p=q

if p and q are equal up to multiplication by ±tk with k ∈ Z. Using this notation, we show
that the Alexander polynomial is symmetric and characterize its behavior on reversals, mirror
images and connected sums.

Proposition 1.81. Fix a link L and knots K1 , K2 .

.
1. The Alexander polynomial ∆L (t) of L is symmetric: ∆L (t) = ∆L (t−1 ).
.
2. The Alexander polynomial of L and its mirror image L agree: ∆L (t) = ∆L (t).
.
3. The Alexander polynomial of L and its reverse rL also agree: ∆L (t) = ∆rL (t).

4. The Alexander polynomial is multiplicative under connected sums: ∆K1 #K2 (t) = ∆K1 (t)∆K2 (t).

Proof. Let A be a n × n Seifert matrix of L. To prove the first statement, we use the definition
of the Alexander polynomial and factor out −t:
. .
∆L (t) = det(tA − AT ) = (−t)n det(t−1 A − AT ) = ∆L (t−1 ).

39
 To prove the second statement, we claim that the Seifert matrix A of L is −A. To see
this, recall that we can obtain a diagram for L from a diagram of L by changing overcrossings
to undercrossings and undercrossings to overcrossings. This changes the sign of lk(ai , a+ j ) for
each [ai ], [aj ] in the basis of H1 (F ), where F is a Seifert surface bounded by L. Since taking
the mirror image of L flips the sign of each entry in A,

A = −A.

Then by the definition of the Alexander polynomial and properties of the determinant,
. .
∆L (t) = det(tA − AT ) = (−1)n det(t(−A) − (−A)T ) = ∆L (t).

The proof of the third statement is similar. We omit the details but interested readers can see
[Lic97, page 58] for reference. For final statement, the reader can check that a Seifert matrix
for the connected sum K1 #K1 can be obtained as the direct sum of Seifert matrices for K1
and K2 ; the result then follows.

Example 1.82. In Example 1.77, we found that the Alexander polynomial for both the left and
right trefoil knot is t2 −t+1 by calculating a Seifert matrix for each and then using the definition
of the Alexander polynomial. Now we only need to calculate the Alexander polynomial of the
left trefoil: since the right trefoil is the mirror image of the left trefoil, it follows from Proposition
1.81 that the two knots must have the same Alexander polynomial.

Proposition 1.83. If an oriented link L bounds a Seifert surface which is not connected, then

∆L (t) ≡ 0.

Proof. Let Σ be a disconnected surface with boundary L. If Σ has m components, then we

can obtain a Seifert surface F (i.e. a connected surface) with boundary L by connecting the
components of Σ with m − 1 tubes.

Figure 1.39: Adding a tube to construct another Seifert surface F from a disconnected Seifert
surface Σ.

A Mayer-Vietoris argument that we omit shows that H1 (F ) = Zm−1 ⊕ H1 (Σ) where

the Zm−1 -summand is generated by the loops f1 , . . . , fm−1 that go around the m − 1 tubes;
see Figure 1.39 for an illustration. Using (fi )Ni=m to denote a basis for H1 (Σ), we observe
that for all i, we have lk(f1 , fi+ ) = 0 and lk(fi , f1+ ) = 0. Thus, the first row and column
of A have only zero entries. It follows that the first row of tA − AT is all zeros, and so
∆L (t) = det(tA − AT ) = 0.

40
 (a) A right-handed cinquefoil knot

(b) A Seifert surface for the cinquefoil

knot

Figure 1.40: A cinquefoil knot and its Seifert surface

We now describe a lower bound on the 3-genus of a link. To that effect, the breadth of a
Laurent polynomial p ∈ Z[t±1 ], denoted breadth(p), is the difference between its largest and
lowest degree. For example, breadth(t3 − t2 + 2t) = 3 − 1 = 2, since t3 − t2 + 2t has highest
degree term t3 of degree 3, and lowest degree term 2t of degree 1. Next, note that since the
Alexander is defined up to a unit in Z[t±1 ], the breadth of the Alexander polynomial is a link
invariant.
The next result gives our first lower bound on the 3-genus.
Theorem 1.84. For any m-component link link L, the following inequality holds,
1
(breadth(∆L ) − m + 1) ≤ 2g3 (L).
2
Proof. Let S be a Seifert surface for L of genus g, so that S is homeomorphic to Σg,m and in
particular, H1 (S) ∼
= Z2g+m−1 . Since a Seifert matrix A for this surface has size 2g + m − 1 and
∆L (t) = det(tA − AT ), we deduce that ∆L (t) is a polynomial of degree 2g + m − 1, so it also
has breadth at most 2g + m − 1, i.e. breadth(∆L ) ≤ 2g + m − 1. The theorem now follows by
rearranging the terms.

Example 1.85. The cinquefoil knot, also known as the knot 51 , is shown in Figure 1.40a. A
Seifert surface for it is shown in Figure 1.40b. Since this Seifert surface has genus two, its first
homology group has 4 generators.
To calculate its Seifert matrix, we draw some generators a1 , . . . a4 of the first homology
group of the Seifert surface, as shown in Figure 1.41a. This picture looks really complicated, so
we simplify it by pretending that the blue and pink faces on the top and bottom are pentagons,
with 5 green bands connecting them on the vertices, as shown in 1.41b. Then, after assigning
the Seifert surface to orient upwards, we can see that the pushoff of a1 , shown in grey in Figure
1.41b, starts outside of the cylinder on the plane above, then twists in front of the Seifert surface
to get inside the cylinder when it reaches the plane below. We see that a+ 1 can actually be pulled
out to just loop once around the green band on the left, as shown by the small grey loop on the
left. From this, we can calculate that,

`k(a1 , a+
1 ) = −1
`k(a2 , a+
1)=0

41
 Matrix Simplified.jpg

(a) Generators of the first homology group

of the Seifert surface
(b) Simplification of Seifert surface and
pushoffs

Figure 1.41: Calculation of the Seifert matrix for the cinquefoil knot

Similarly, the pushoff of a2 is shown in dark blue in Figure 1.41b, so it, in particular, has linking
number 1 with a1 .
Using similar reasoning for the other pushoffs, we can determine the Seifert matrix for the
right-handed cinquefoil knot is:
 
−1 0 0 0
 1 −1 0 0
A=
 
 0 1 −1 0 


0 0 1 −1

The Alexander polynomial is then:

∆L (t) = det(At − AT )
 
1 − t −1 0 0
 t 1 − t −1 0 
= det 
 
 0 t 1 − t −1 


0 0 t 1−t
= (1 − t)4 − 3(1 − t)2 (t)(−1) − t2 (−1)2
= t4 − t3 + t2 − t + 1.

In particular, its Alexander polynomial has breadth 4. Therefore, by Proposition 1.84, we know
that the cinquefoil knot must have 3-genus at least 2. Since the Seifert surface shown in Figure
1.40b has genus 2, we can conclude that the cinquefoil knot’s 3-genus is exactly 2.

We now move on to properties of the signature. Following a similar pattern as for the
Alexander polynomial, we first describe properties of the signature and then prove a more
interesting property, establishing a lower bound on the unlinking number.
Proposition 1.86. Fix a link L and knots K1 , K2 . The signature satisfies the following prop-
erties:

• the signature is unchanged under orientation reversal: σ(rL) = σ(L);

42
 • the signature changes sign under mirror images: σ(L) = −σ(L);
• the signature is additive under connected sums: σK1 #K2 = σK1 + σK2 .

Proof. Each of these properties follows from the definition of the signature: if A is a Seifert
matrix for L, then AT is a Seifert matrix for rL and −A is a Seifert matrix for L and a Seifert
matrix for the connected sum K1 #K1 can be obtained as the direct sum of Seifert matrices
for K1 and K2 .
Remark 1.87. A link called is amphichiral if it is equivalent to its mirror image. The second
item of Proposition 1.86 shows that the signature of an amphichiral link must vanish, so the
signature is an obstruction to amphichirality (whereas the Alexander polynomial is not). For
example, for the trefoil K, the fact that σ(K) 6= 0 implies that K is not amphichiral. In the
fourth problem set, we will prove that the figure eight knot is amphichiral.
The following theorem gives our first lower bound on the unknotting number.
Theorem 1.88. The absolute value of the signature is a lower bound on twice the unlinking
number: for every oriented link L, the following inequality holds:
|σ(L)| ≤ 2u(L).
Proof. We claim that if L0 is a link obtained from L by performing a unique crossing change, the
signature changes by at most 2. Without loss of generality, assume that (diagrams for) L = L+
and L0 = L− look locally as in the top of Figure 1.42, where L0 is an auxiliary link obtained by
smoothing the crossing. Seifert surfaces F+ , F0 , F− for L+ , L0 , L− can be chosen to locally look
like the bottom of Figure 1.42 around that crossing.

Figure 1.42: The links L+ , L0 and L− and local pictures for Seifert surface we chose for them.

It follows that H1 (F± ) = H1 (F0 ) ⊕ Z and if A+ , A0 , A− are Seifert matrices for these Seifert
surfaces with respect to some basis and if we abridge H• = A• + AT• , then H• = ( H∗0 ∗∗ )
so |σ(L± ) − σ(L0 )| ≤ 1. The claim now follows from the triangle inequality:
|σ(L) − σ(L0 )| ≤ |σ(L) − σ(L0 )| + |σ(L) − σ(L0 )| ≤ 2.
The end of the proof is now identitcal to the proof of Proposition 1.18. Let L = L(0) , . . . , L(u) be
an unlinking sequence, i.e. a sequence of links where L(i) and L(i+1) differ by a unique crossing
change and L(u) is the unlink. The theorem follows from the fact that σ(L(u) ) = 0, the triangle
inequality and the claim:
u−1
X u−1
X
|σ(L)| = (σ(L(i) ) − σ(L(i+1) )) ≤ |σ(L(i) ) − σ(L(i+1) )| ≤ 2u.
i=0 i=0

43
 The lower bound from Theorem 1.88 is in fact not the best one can obtain using signature-
like invariants. We will expand on this and on related points in Remark 1.90 below, but first
some examples.

Example 1.89. We use Theorem 1.88 to calculate the unknotting number of some knots.

• We show that the torus knot K := 51 has unknotting number 2. Two crossing changes
needed to turn K into the unknot are illustrated in Figure 1.43, while the reader can
verify that |σ(K)| = 4; the claim then follows from Theorem 1.88.

• We show that the connected sum Kn of n trefoils has unknotting number n. To see this,
note that for any knots K1 , K2 , we have u(K1 #K2 ) ≤ u(K1 ) + u(K2 ) (it is an open
question whether this is an equality) and so u(Kn ) ≤ n · u(K) = n. The equality now
follows by combinining Theorem 1.88 with the additivity of the signature under connected
sums from Proposition 1.86.

Figure 1.43: The unknotting number of the torus knot 51 = T (5, 2) is at most 2: this can be
seen by changing the two circled crossings.

Remark 1.90. We collect a couple of further remarks on the signature for the interested reader.

• The signature also provides a lower bound on the 3-genus (although it is usually weaker
than the one provided by the Alexander polynomial) namely we have |σ(L)| + 1 − m ≤
2g3 (L). This follows from the fact that |σ(L)| ≤ b1 (F ) for any Seifert surface F bound-
ing L; here b1 (F ) denotes the first Betti number of F .

• Finally, we briefly describe a generalisation of the signature. The Levine-Tristram signa-

ture of L at ω ∈ S 1 is the signature of H(ω) := (1 − ω)A + (1 − ω)AT .9 This quantity,
denoted σL (ω) is also called the ω-signature or the Tristram-Levine signature. Regardless
of the terminology, observe that σL (ω) and the nullity ηL (ω) := nullity(H(ω)) are link
invariants (one readily checks the invariance under S-equivalence) and additionally note
that σ(L) = σL (−1). The Levine-Tristram signature satisfies many of the properties listed
in Proposition 1.86 and Theorem 1.88 and in particular this latter result can be improved
to
|σL (ω)| + |ηL (ω) + m − 1| ≤ u(L)
for all ω ∈ S 1 . The proofs are (essentially) identical.

1.3.3 Applications to knot concordance

The goal of this section is to apply Seifert matrices, the Alexander polynomial and the signature
in order to obstruct knots from being slice. References include [Lic97, Liv93].
9
This makes sense because Hermitian matrices have real eigenvalues.

44
 The strongest obstruction to sliceness based on the Seifert matrix reads as follows.

Theorem 1.91. If a knot is slice, then each of its Seifert matrices is congruent to a matrix of
the form !
0 B
A= , (1.2)
C D
where 0, B, C, and D are integer-valued matrices whose size is half that of A.

In practice, if one is trying to obstruct a knot from being slice, it is often difficult to prove
the Seifert matrix of a knot is not congruent to one as in the statement of Theorem 1.91. The
next theorem states more manageable criteria.

Theorem 1.92. If K is a slice knot, then

.
1. the Alexander polynomial of K factors as a norm, i.e. ∆K (t) = f (t)f (t−1 ) for some f ∈
Z[t±1 ];

2. the signature of K vanishes: σ(K) = 0.

Before we discuss the proof of these theorems, we first give several examples.

Example 1.93.

• We start with a reality check on the knot 946 which we already know to be slice by 
0 −2
Example 1.62. Recall from Example 1.72 that a Seifert matrix for 946 is given by −1 0 .
The conclusion of Theorem 1.91 holds since the upper left 1 × 1 block of this matrix is
indeed all 0s. The conclusions of Theorem 1.92 also hold since the Alexander polynomial
.
of 946 is (t − 2)(2t − 1) = (t − 2)(t−1 − 2) and its signature of 946 vanishes.

• Next, we show that the (left-handed) trefoil knot K is not slice. Recall from Section 1.3.1
1 0 , which gives ∆ (t) = t2 − t + 1, and σ(K) = 2. Both


that a Seifert matrix of K is −1 1 K
criterions of Theorem 1.92 now imply that K is indeed not slice. To see that ∆K (t) is not
a norm, one can for example notice that ∆K (−1) = 3 is not a square.

• As another reality check, we verify that the conclusion of Theorems 1.91 and 1.92 hold
for every knot of the form K # −K̄, which we know to be slice by Proposition 1.63. First,
note that the conclusions of Theorem 1.92 hold for K# − K̄ thanks to the properties of
the Alexander polynomial and signature from Proposition ?? and Proposition 1.86:
. .
∆K#−K̄ (t) = ∆K (t)∆−K̄ (t) = ∆K (t)∆K (t) = ∆K (t)∆K (t−1 )
σK#−K̄ = σK + σ−K̄ = σK − σK = 0.

Next we check that the conclusion of Theorem 1.91 holds when K is the trefoil (but the
same proof works in general). To start, recall that for any two knots K1 , K2 with Seifert
matrices A1 and A2 , a Seifert matrix for K1 # K2 is the direct sum A1 ⊕ A2 . In addition,
if A is a Seifert matrix for K, then −AT is a Seifert matrix for −K̄. Thus a Seifert matrix
for K # −K̄ is10  
1 0 0 0
−1 1 0 0
A= .
 
 0 0 −1 1 
0 0 0 −1
10
AC: todo: think.

45
 It is not immediately obvious that this is congruent a matrix of the form in Theorem 1.91,
but this is in fact true. To see this, define u = (1, 0, 1, 0) and v = (0, 1, 0, 1); then we find
that
uT Au = uT Av = v T Au = v T Av = 0.
Since u and v can be extended to a basis of Z4 (say, by adding (1, 0, 0, 0) and (0, 1, 0, 0)),
we deduce that A is congruent to a matrix as in Theorem 1.91.
Remark 1.94. We note that the converses of the the above theorems are not true.

• The Alexander polynomial of the connected sum K#K of two identical trefoils has Alexan-
.
der polynomial (t2 − t + 1)2 = (t2 − t + 1)(t−2 − t−1 + 1) but K#K is not slice since the
absolute value of its signature equals 4.

• The figure eight knot K has signature 0 but is not slice since its Alexander polynomial
∆K (t) = t2 − 3t + 1 is a not a norm, e.g. because ∆K (−1) = 5 is not a square.

• There are knots that satisfy the condition of Theorem 1.91 (and therefore the conditions
of Theorem 1.92) but are nevertheless not slice; this is a difficult result that was first
proved by Casson and Gordon in the late seventies.

Let us now prove Theorems 1.91 and 1.92, starting with the latter.

Proof of Theorem 1.92. Suppose the knot K has a Seifert matrix A. We first prove that the
Alexander polynomial of K factors as f (t)f (t−1 ) with f (t) ∈ Z[t±1
]. Theorem 1.91 implies
the existence of an invertible matrix M such that M AM T = C0 D B . Using successively the

definition of the Alexander polynomial and the fact that det(M ) = ±1, we deduce that
!
. 0 tB − C T .
∆K (t) = det(tA−AT ) = det(tM AM T −M AT M T ) = = det(tB−C T )det(tC−B T )
tC − B τ tD − DT
.
In order to obtain a factorisation ∆K (t) = f (t)f (t−1 ) we set f (t) = det(tB − C T ).
We now prove that the signature of a slice knot vanishes. Recall that the signature of a knot
is the signature of the matrix A + AT for any Seifert matrix A. Using ∼ to denote congruence
and setting S := B + C T and R := D + DT , we see that
" #
T 0 S
(A + A ) ∼ τ
S R

Since A−AT is invertible [Lic97], so is A+AT and it follows that S must

 be invertible. Therefore,
0 Ig
row and column operations can give us a matrix of the form Ig R and then we can perform
 
0 Ig
more row and column operations to transform this matrix into Ig 0 . The reader can check
that this matrix has signature equal to zero.

To prove Theorem 1.91, we will state a short lemma whose proof can be found in [Lic97,
pages 87-89].
Lemma 1.95. If two knots K1 , K2 bound disjoint oriented embedded surfaces in D4 , then `k(K1 , K2 ) = 0.

We use this lemma to outline the proof of Theorem 1.91.

Proof. Let D ⊂ D4 be a slice disk for K and let F ⊂ S 3 be a genus g Seifert surface for K.
It is possible to show that there exists a compact connected oriented 3-manifold M ⊂ D4 such
that ∂M = F ∪∂ D, see [Lic97, Lemma 8.14]. Another argument beyond the scope of this class
shows that ker(H1 (∂M ; Q) → H1 (M ; Q)) is a half dimensional subspace of H1 (∂M ; Q) [Lic97,

46
Lemma 8.15]. It follows that there is a Z-basis [f1 ], . . . , [f2g ] for H1 (∂M ) such that [f1 ], . . . , [fg ]
are sent to zero in H1 (M, Q), see [Lic97, Corollary 8.16].
Since ∂M = F ∪∂ D and D is simply connected, the fi can be homotoped to lie in F .
We claim that [f1 ], . . . , [fg ] span the desired half-basis of the Seifert form. Since each [fi ]
maps to zero in H1 (M ; Q), there exists integers ni such that ni [fi ] vanishes in H1 (M ). It
follows that ni fi bounds a surface Ni in M . Moreover, if fi− is a push-off of fi , then (ni fi )−
bounds a surface Ni− which is disjoint from the Nj . Consequently, `k(fi− , fj ) = Ni− · Nj = 0
for i, j = 1, . . . , g, where Ni · Nj denotes the algebraic intersection of the surfaces in the 4-ball
(which vanishes by Lemma 1.95 since Ni− is disjoint from Nj ).

Remark 1.96. We conclude this subsection by briefly describing how Theorem 1.91 leads to
insights into the structure of the knot concordance group C. First, note that it is known that a
matrix A with integer coefficients is the Seifert matrix of a knot if and only if det(A−AT ) = ±1.
As a consequence, if one defines Calg as the group of such matrices modulo those as in (1.2) (i.e.
A1 and A2 are declared equivalent if A1 ⊕ −A2 is of that form) then Theorem 1.91 shows that
there is a surjective homomorphism
s : C  Calg .
In the sixties, Levine showed that Calg is isomorphic to Z∞ ∞ ∞
2 ⊕Z4 ⊕Z . The question of whether
s is injective remained open until Casson and Gordon (in the late seventies) developped new
invariants to prove that ker(s) 6= 0. While we now know that ker(s) is in fact very large, still
very little is known about the knot concordance concordance group C? Does it only contain
2-torsion or is there other torsion? Could it contain a Q-summand? Could it be that C splits
as Calg ⊕ ker(s)? All of these questions remain open.

1.3.4 The sign refined Alexander polynomial

Since the Alexander polynomial of an oriented link L is only well defined up to multiplication
by ±tk , we can always pick a representative f (t) of ∆L (t) that satisfies f (t) = ±f (t−1 ) but
the sign ambiguity a priori remains. In this section we show that a minor modification in
the definition of ∆L (t) makes it possible to fix this sign ambiguity. This seemingly minor
modification leads to several interesting consequences, most strikingly the fact that this sign-
refined version of the Alexander polynomial admits a skein relation, just as the Jones polynomial
did. References for this section include [Lic97, Chapter 8].
We start with the foreshadowed modification in the definition of the Alexander polynomial.

Definition 1.97. The Alexander-Conway polynomial of an oriented link L is defined as

∇L (t) = det(t1/2 A − t−1/2 A> )

where A is any Seifert matrix for L.

Normalising the Alexander polynomial as such leaves no ambiguity of sign or units as proved
in the next result.

Theorem 1.98. The Alexander-Conway polynomial is a well defined invariant on an oriented

link L: it does not depend on the Seifert matrix of choice.

Proof. We only need to check that ∇L (t) is invariant under S−equivalence. Recall that matri-
ces A and B are S−equivalent if they are related via a finite sequence of unimodular congru-
ences and elementary enlargements/reductions. It therefore suffices to prove that the expression

47
det(t1/2 A − t−1/2 A> ) is unchanged under each of these two moves. For unimodular congruence,
this follows from the fact that det(P ) = ±1, exactly as in the proof of Proposition 1.76:

det(t1/2 B − t−1/2 B > ) = (det P )2 det(t1/2 A − t−1/2 A> ) = det(t1/2 A − t−1/2 A> ).

We now move on to elementary enlargements and reductions. Since both versions of this move
(and its inverse) are extremely similar, we only make the verification in one case. Namely, we
suppose that  
A η 0
B =  0 0 1 ,
 
0 0 0
where η is a column vector so that

t1/2 A − t−1/2 A> t1/2 η

 
0
1/2 −1/2 >
t B−t B =  −t−1/2 η > 0 t1/2  .
 

0 −t−1/2 0

One can then check that det(t1/2 B − t−1/2 B > ) = det(t1/2 A − t−1/2 A> ) by cofactor expansion.
In conclusion, ∇L (t) is invariant under S− equivalence, as required.

Note that for a knot K, the Alexander-Conway polynomial belongs to Z[t±1 ] and, more
generally, for an m-component oriented link L, we have t(m−1)/2 ∇L (t) ∈ Z[t±1 ].

Example 1.99. We calculate the Alexander-Conway polynomial on a couple of examples.

 
−1 0
1. For the right-handed trefoil knot, using the Seifert matrix A = −1 −1 , we have
!
−1/2 > −t1/2 + t−1/2 t−1/2
∇L (t) = det(t 1/2
A−t A ) = det = t − 1 + t−1 .
−t1/2 −t1/2 + t−1/2

 
0 −2
2. For the knot 946 , using the Seifert matrix B = −1 0 , we have
!
0 −2t1/2 + t−1/2
∇L (t) = det = −2t + 5 − 2t−1 .
−t1/2 + 2t−1/2 0

Next, we collect some properties of the Alexander-Conway polynomial.

Proposition 1.100. For an oriented m-component link L, the sign-refined Alexander polyno-
mial, ∇L (t), satisfies the following properties:

1. (−1)m−1 -symmetry: ∇L (t) = (−1)m−1 ∇L (t−1 );

2. behavior under mirror images: ∇L (t) = (−1)m−1 ∇L (t);

3. invariance under reversal: ∇L (t) = ∇rL (t);

4. multiplicativity under connected sums: ∇K1 #K2 (t) = ∇K1 (t)∇K2 (t).

5. sufficient condition for vanishing: if L bounds a non-connected Seifert surface, then

∇L (t) ≡ 0.

48
Proof. We start with observations that we will use to prove the first two properties. First, if L
bounds a Seifert surface with genus g, then its first homology has 2g + m − 1 generators and
hence L admits a Seifert matrix A of size (2g + m − 1) × (2g + m − 1). For later use, note that
if we set n = 2g + m − 1 = n, then we have (−1)n = (−1)m−1 .
To prove (−1)m−1 -symmetry we factor a sign from the determinant and take the transpose,

∇L (t) = det(t1/2 A − t−1/2 A> ) = (−1)n det(t−1/2 A> − t1/2 A)) = (−1)m−1 ∇L (t−1 ).

Similarly, a Seifert matrix for the mirror L is −A, and thus we need only factor a sign from the
determinant:

∇L (t) = det(t1/2 A − t−1/2 A> ) = (−1)n det(−t1/2 A + t−1/2 A> ) = (−1)m−1 ∇L (t).

The proof for orientation reversal is similar to the one in Proposition 1.81: a Seifert matrix of
rL is AT , so we can take the transpose of the argument of the determinant to obtain

∇L (t) = det(t1/2 A − t−1/2 A> ) = det(t1/2 A> − t−1/2 A) = ∇rL (t).

We move on to the fourth property. If L bounds a non-connected Seifert surface, by Proposi-

tion 1.83, the Alexander polynomial of L is 0, therefore the sign-refined Alexander polynomial
vanishes as well.
Lastly, if A1 and A2 are Seifert matrices for knots K1 and K2 , respectively, then A = A1 ⊕A2
is a Seifert matrix for K1 #K2 . The determinant is multiplicative for direct sums, hence

∇K1 #K2 (t) = det(t1/2 A−t−1/2 A> ) = det(t1/2 A1 −t−1/2 A>

1 ) det(t
1/2
A2 −t−1/2 A>
2 ) = ∇K1 (t)∇K2 (t).

We have proved all five properties and so this concludes the proof of the proposition.

The most surprising advantage of fixing the sign of the Alexander polynomial is that the
result admits a skein relation, just as the Jones polynomial did. In particular, it is possible to
calculate the sign refined Alexander polynomial without computing a Seifert matrix.

Theorem 1.101. The Alexander-Conway polynomial satisfies the two following properties:

1. For the unknot U , ∇U (t) = 1.

2. If three oriented links L+ , L− and L0 are the same exept in a small neighborhood where
they are as in Figure 1.13, then the following skein relation holds:

∇L+ (t) − ∇L− (t) = (t−1/2 − t1/2 )∇L0 (t). (1.3)

Additionally, these two properties completely characterise the Alexander-Conway polynomial of

a link.

Proof. The proof of the first proposition follows from the definition. Choose a Seifert surface
F0 for L0 . Attaching twisted bands to F0 , we may obtain Seifert surfaces F+ and F− for L+
and L− as shown in Figure 1.44.
A Mayer-Vietoris argument which we omit shows that H1 (F± ) = H1 (F0 ) ⊕ Z[a0 ] where
the additional generator a0 is illustrated in Figure 1.44. We may therefore choose a basis
a1 , . . . , an ∈ H1 (F0 ) for the first homology group of F0 and extend it to a basis of H1 (F− ) and
H1 (F+ ), the first homology groups of F− and F+ , using the cycle a0 .

49
Figure 1.44: Seifert surfaces F0 , F+ , and F− for the links L0 , L+ , and L− related by adding
a band. In the left two cycles, one also sees a loop a0 that represents an element in the first
homology groups of F+ and F− .

It can then be seen that if A0 is the Seifert matrix of F0 with respect to the basis a1 , . . . , an
then the Seifert matrices A+ and A− for F+ and F− (with respect to the aforementioned bases)
are related as below, where η, ξ are integer column vectors and N is an integer:
! !
N ηt N + 1 ηt
A+ = A− =
ξ A0 ξ A0

To see this, notice that the twist in Figure 1.44 only affects the linking behavior of a0 with its
own push off and that in this case only a full twist has been added. Using these Seifert matrices
to compute the left hand side of (1.3) gives us

∇L+ (t) − ∇L− (t) = det(t1/2 A+ − t−1/2 At+ ) − det(t1/2 A− − t−1/2 At− )
!
t−1/2 − t1/2 0
= det 1/2 (1.4)
t ξ − t η t A0 − t−1/2 At0
1/2 1/2

= (t−1/2 − t1/2 )∇L0 (t),

This establishes the skein relation.

The fact that the skein relation, and value on the unknot, completely determine the poly-
nomial can be proved by induction on the number of crossings, but the details are omitted; we
refer the interested reader to [Lic97] for details.

Example 1.102. We apply the skein relation from Theorem 1.101 to the right trefoil shown in
Figure 1.45 in order to show that ∇L (t) = t − 1 + t−1 = (t−1/2 − t1/2 )2 + 1.

Figure 1.45: A link diagram L for the right trefoil.

Focusing on the upper right crossing, we can see that it has positive sign. So if we let L− ,
L0 be as in Figure 1.46, we obtain that ∇L (t) − ∇L− (t) = (t−1/2 − t1/2 )∇L0 (t). But we note
that L− is the unknot, so ∇L− (t) = 1. It remains to determine ∇L0 (t).

50
 Figure 1.46: Links L− , L+ obtained from crossing changes on our trefoil

Let L0 be a diagram for L0 . Taking a look at the lower crossing, in L0 we again note that it is
positively oriented. So if we let L0− , L00 to be as in Figure 1.47, we have that ∇L0 (t) − ∇L0− (t) =
(t−1/2 − t1/2 )∇L00 (t). But L00 is again the unknot, so ∇L00 (t) = 1. Furthermore ∇L0− (t) = 0,
because L0− is the two component unlink and bounds disconnected surface. We deduce that
∇L0 (t) = (t−1/2 − t1/2 )∇L0 (t) = (t−1/2 − t1/2 ) and making relevant substitutions, we obtain
∇L (t) = (t−1/2 − t1/2 )2 + 1 as claimed.

Figure 1.47: Links L0− , L00 obtained from crossing changes on L0 := L− .

By applying repeatedly Theorem 1.101 one can prove that ∇L (t) is a (non-Laurent) poly-
nomial in z := t−1/2 − t1/2 ( Example 1.102 provides a reality check). It is then sometimes con-
venient to define the Conway polynomial of an oriented link L by the formula CL (z) := ∇L (t).

Remark 1.103. We describe some of the properties of the Alexander-Conway polynomial ∇L (t)
in terms of the Conway polynomial CL (z) := ∇L (t). For example, if m is odd, the first item of
Proposition 1.100 shows that CL (−z) = CL (z) whereas the skein relation from Theorem 1.101
can be rewritten as
CL+ (z) − CL− (z) = zCL0 (z).
We also note that if K is a knot, then CK (0) = 1: crossing changes do not affect CK (0)
(indeed evaluating the skein relation at z = 0 we get that CL+ (0) = CL− (0)) and we noted in
Theorem 1.101 that ∇K ≡ 1 when K is the unknot.

1.4 More polynomials, braids and skein relations

Here are the main invariants we have learnt about so far: the linking number, the signature,
the Alexander polynomial and the Jones polynomial. Arguably, the latter two invariants are
the most efficient at telling knots apart and interestingly they both share a skein relation. Are
there are other polynomials that satisfy such relations? The goal of this section is two-fold.
Firstly, we will learn about a polynomial, known as the HOMFLY polynomial (or HOMFLY-
PT polynomial) that also satisfies a skein relation and generalises both the Alexander and Jones
polynomials; this will be the content of Subsection 1.4.1. Secondly, in Subsection 1.4.3, we will

51
sketch very briefly how the search for further polynomial invariants that satisfy skein relations
leads to the theory of quantum groups; this will require introducing braids which will be done
in Subsection 1.4.2. By the end of this section, the leading question will have shifted from "how
to find more knot invariants?" to "what is the topological meaning of the multitude of invariants
we now have at our disposal?" This shift in perspective has fundamentally shifted research in
knot theory.

1.4.1 The HOMFLY polynomial

In Subsections 1.1.3 and 1.3.4, we learnt that the Jones, and Alexander-Conway polynomials
can be characterized by their Skein relations and their value on the unknot. In this section,
we will introduce and study a two variable polynomial that generalises these invariants and
satisfies similar looking skein relation and normalisation properties. References for this section
include [Lic97, Liv93, Cro04].
±1
Definition 1.104. The HOMFLY polynomial, PL (l, m) ∈ Z[l±1,m ], of a link L is a two-
variable Laurent polynomial defined by the following three properties:

• invariance: PL (l, m) is invariant under ambient isotopy of L;

• normalization: PU (l, m) = 1 where U is the unknot;

• skein relation: For L0 , L± as in Figure 1.13, the following equation holds

l−1 PL+ (l, m) − lPL− (l, m) = mPL0 (l, m). (1.5)

When the context is clear, we will write PL instead PL (l, m). Regardless of notation, it is
à priori unclear that a Laurent polynomial satisfying the conditions of Definition 1.104 exists
(and is unique). This is a non-trivial fact whose proof we only sketch; a detailed argument can
be found in [Lic97, Theorem 15.2].

Theorem 1.105. There exists a unique function

P : {Oriented links in S 3 } → Z[l±1 , m±1 ]

that satisfies the three properties of Definition 1.104.

Proof. Informally, the proof of the existence of PL (l, m) proceeds by induction on the number of
crossings of link diagrams so that (1.104) holds and the function is invariant under Reidemeister
moves. To make this this more precise, let Dn be the set of all oriented links diagrams with at
most n crossings modulo orientation preserving homeomorphism of the plane. The induction
hypothesis is that for every n ≥ 0, there exists a unique function P : Dn → Z[l±1 , m±1 ] such
that

• the skein relation (1.104) holds.

• P (D) is unchanged under Reidemester moves on D that do no involve more than n

crossings.

• a technical condition involving so-called “ascending diagrams." Without going into details,
this condition (in particular) fixes the value of P on unlinks.

52
Example 1.106. We calculate the HOMFLY polynomial of a few relatively simple links and
build upon previous examples as we go. The desired link will often be denoted L+ and we will
use the skein relation in the form P (L+ ) = l(lP (L− ) + mP (L0 )). Also, in what follows we set
µ := m−1 (l−1 − l).

1. We claim that the n-component unlink has HOMFLY polynomial µn−1 . For n = 2,
applying Reidemeister I moves to the diagram of an unknot to obtain L+ and L− , Figure
1.48 shows that the two-component unlink has HOMFLY polynomial µ. The claim follows
by induction using a similar idea, as also illustrated in Figure 1.48.

Figure 1.48: The skein relation used to calculate the HOMFLY polynomial of the n-component
unlink.

2. Let L+ be the oriented Hopf link with positive linking number, so that applying the
skein relation to the crossing illustrated in Figure 1.49, L− is the unknot and L0 is the
two-component unlink. Applying the previous example, we obtain

P (Hopf link) = l(lP (two-component unlink) + mP (unknot))

= l(lµ + m)
= (l − l3 )m−1 + lm.

3. Let L+ be the (right) trefoil, so that applying the skein relation to the crossing illustrated
in Figure 1.49, L− is the unknot and L0 is the positive Hopf link. This time applying our
calculation of the HOMFLY polynomial for the positive Hopf link, we obtain

P (31 ) = l(lP (unknot) + mP (Hopf link))

= l(l + m((l − l3 )m−1 + lm))
= 2l2 − l4 + l2 m2 .

4. Let L+ be the two component link with oriented link diagram

Applying the skein relation to the crossing illustrated in Figure 1.49, L− is positive Hopf
link with and L0 is the right trefoil 31 . Applying our calculation for the trefoil, we get

P (L+ ) = l(lP (Hopf link) + mP (31 ))

= l(l((l − l3 )m−1 + lm) + m(2l2 − l4 + l2 m2 ))
= (l3 − l5 )m−1 + (3l3 − l5 )m + l3 m3 .

53
 Figure 1.49: A table of diagrams for L+ , L− , and L0 for the links in Example 1.106

Before moving on to properties of the HOMFLY polynomial, we describe in what sense it

generalises both the Alexander-Conway and Jones polynomial.

Proposition 1.107. For an oriented link L,

1. the Alexander-Conway polynomial can be recovered from the HOMFLY polynomial as

∇L (t) = PL (1, t−1/2 − t1/2 ).

2. the Jones polynomial can be recovered from the HOMFLY polynomial as

VL (t) = PL (t, t1/2 − t−1/2 ).

Proof. This follows from Theorems 1.105 and 1.101 (resp. Theorem 1.28) by noting that both
polynomials are equal to 1 one the unknot, comparing the skein relation of PL (1, t−1/2 − t1/2 )
(resp. PL (t, t1/2 − t−1/2 )) with the one for the Alexander-Conway polynomial (resp. the Jones
polynomial) and verifying that they agree. Indeed, we know that all three polynomials are
entirely characterised by their value on the unknot and their skein relation.

Example 1.108. We perform a reality check on the right trefoil K. We saw in Example 1.106
that PK (l, m) = 2l2 − l4 + l2 m2 . Setting l = 1 and m = t−1/2 − t1/2 , we recover the result of
Example 1.77, namely

PK (1, t−1/2 − t1/2 ) = 1 + (t−1/2 − t1/2 )2 = t − 1 + t−1 = ∇K (t)

while if we set l = t and m = t1/2 − t−1/2 , a similar verification shows that we recover the result
from Example 1.27 namely VK (t) = −t4 + t3 + t.

Next, we list some properties of the HOMFLY polynomial. Note that applying the evalua-
tions from Proposition 1.107 recovers previously mentioned properties of the Jones and Alexan-
der polynomial.

Proposition 1.109. The HOMFLY polynomial satisfies the following properties:

54
 1. behavior under mirror image: PL (m, l) = PL (l−1 , −m);11

2. invariance under orientation reversal: PrL = PL ;

3. multiplicativity under connected sums: PK1 #K2 = PK1 PK2 ;

l−1 −l
4. behavior under disjoint unions: PL1 tL2 = m PL1 PL2 .

Proof. For the first property, a diagram for the mirror image is obtained from a diagram of
the original link by changing all crossings, so applying successively the skein relation exchanges
the roles of l and l−1 and picks up a minus sign on the m variable. For the second property,
reversing orientations does not affect the skein relation so the outcome is identical. For the
third property, apply the skein relation recursively to K2 to obtain
n
X
PK2 (l, m) = fi (l, m)PUi (l, m),
i=1

where fi (l, m) is some Laurent polynomial, n > 0 is a positive integer and Ui denotes the i-
component unlink (with the convention that U0 is the empty link); the result now follows by
applying the skein relation to the K2 portion of K1 #K2 and observing that
n
X n
X
PK1 #K2 = fi PK1 tUi−1 = fi PK1 PUi = PK1 PK2 .
i=1 i=1

The fourth and final property follows from the third and from the fact that the skein relation
implies that (l−1 − l)PK1 #K2 = mPK1 tK2 (l, m); see Figure 1.50.

Figure 1.50: Applying the skein relation to relate K1 #K2 and K1 t K2 .

1.4.2 Braids
As we have already mentioned, our goal is now to briefly introduce quantum topology and
how it leads to generalisations of the Alexander, Jones and HOMFLY polynomials. Before
jumping into the topic however, we need to introduce braids and the braid group Bn . After
listing the definitions, we then describe a presentation of Bn by generators and relations, before
moving on to the relation between braids and links via the classical theorems of Alexander and
Markov. References include [Lic97, pages 183 and 187], [KT08, Sections 2.3 and 2.5], and [BB05,
Example 2.1].
Let D2 be the closed unit disc in R2 . Fix a set of n ≥ 1 punctures pi in the interior of D2 .
We will assume that the pi lie in (−1, 1) = Int(D2 ) ∩ R and p1 < . . . < pn .

Definition 1.110. An n-stranded braid consists of n embedded intervals in D2 × [0, 1] whose

boundary is ni=1 pi × {0} t ni=1 pi × {1} and where each strand intersects the disk D2 × {t}
F F

exactly once for every t ∈ [0, 1].

11
AC: Not sure about this. References list m instead of m, but that doesn’t make sense given the behavior or
the Alexander-Conway polynomial under the mirror image operation.

55
 As for knots and links, we represent braids by drawing diagram for them, as can be seen in
Figure 1.51.

Figure 1.51: A braid (left) and a diagram for it (right).

Next, we define what it means for two braids to be the same. Intuitively, this is the case
if we can stretch and slide the strands of one braids into those of the other while leaving the
endpoints fixed. The precise definition is as follows.

Definition 1.111. Two braids β1 and β2 are isotopic if there exists an ambient isotopy (ht : D2 ×
[0, 1] → D2 × [0, 1])t∈[0,1] with h0 = id, h1 (β1 ) = β2 and ht |∂(D2 ×[0,1]) = id for every t ∈ [0, 1].

Figure 1.52: An example of two isotopic braids.

Next, we describe how isotopy classes of braids give rise to a group. Firstly, the identity
braid is simply {p1 , . . . , pn } × [0, 1]. Secondly, the composition of two braids β1 and β2 consists
of gluing β1 on top of β2 and shrinking the result by a factor two, as illustrated in Figure 1.53.
Thirdly, if β is a braid, then we write β −1 for the braid obtained from β obtained from β by a
reflection with respect to the horizontal disc D2 × { 12 } as illustrated in Figure 1.54.

56
 Figure 1.53: Composing two braids A1 and A2 .

Definition 1.112. The braid group on n strands, denoted Bn , is the set of isotopy classes
of n-braids with braid composition as the group law.

B =

= = BB −1 = e

B −1 =

Figure 1.54: An xample of a braid B and its inverse.

It is not difficult to verify that the braid group is indeed a group. Instead of going through
the details, we describe a presentation of Bn by generators and relations.

Proposition 1.113. The braid group on n strands admits the following presentation:

Bn = hσ1 , . . . , σn−1 | σi σj = σj σi for |i − j| > 1 and σi σi+1 σi = σi+1 σi σi+1 for 1 ≤ i ≤ n − 2i.

Proof. Temporarily use G to denote the group defined by the above presentation. To prove
∼
=
the proposition, it suffices to define an isomorphism G − → Bn . Define a map φ : G → Bn by
mapping the generator σi to the (isotopy class of the) braid illustrated in Figure 1.55 where the
only crossing occurs when the ith strand crosses over the i + 1st ; all other strands are vertical:

φ(σi ) :

Figure 1.55: The braid φ(σi ).

57
 It can be verified that φ respects the relations of G and therefore defines a homomor-
phism G → Bn ; see Figure 1.56 below for an illustration. We omit the proof that φ is injective
but note that φ can be seen to be surjective by noting that any braid can be written as a compo-
sition of various φ(σi )±1 . Thus G is isomorphic to Bn , concluding the proof of the proposition.

Figure 1.56: An illustration of the fact that the map G → Bn satisfies the braid relations.

In what follows, we will slightly abuse notation by writing σi both for the generators of the
presentation from Proposition 1.113 and for the corresponding braid illustrated in Figure 1.55.

Remark 1.114. We briefly mention a second interpretation of the braid group in terms of
homeomorphisms of the punctured disk. Let Dn be D2 with n punctures, i.e. with n distinct
points in the interior of D2 removed. We choose an orientation for Dn . It now turns out that Bn
is isomorphic to group of isotopy classes of orientation-preserving homeomorphisms of Dn that
restrict to the identity on ∂Dn [Bir74].

Next we relate braids to links. In more detail, we say that the closure of a braid is the link
obtained by connecting each upper point of the braid to the lower point right beneath it, as
illustrated below:

Figure 1.57: The closure of a braid.

While the closure of any trivial braid is a trivial link the reader might also verify that the
closure of the braid σ13 ∈ B2 is the trefoil, while the closure of the braid σ1 σ2−1 σ1 σ2−1 is the
figure eight knot. In fact Alexander proved that every link arises as a braid closure.

Theorem 1.115. Any link can be represented as the closure of a braid.

Proof. Instead of presenting Alexander’s original proof from 1923, we outline an elegant proof
that was discovered independently in the eighties by Yamada and Vogel. We illustrate their
proof, which is essentially an algorithm that explictly converts a link diagram into a braid, with
a running example, namely the knot 52 . A diagram for 52 is illustrated on the left of Figure 1.58.

58
Figure 1.58: On the left: a diagram for the knot 52 ; in the center: the result of applying Seifert’s
algorithm to this diagram and connecting the Seifert circles with the appropriate labeled edges;
on the right: the result of performing an isotopy on the central figure.

The first step of the algorithm is to apply Seifert’s algorithm to a diagram for the link and
to connect the Seifert circles by labelled edges instead of the usual twisted band; these edges are
labelled either by a + or a − depending on the sign of the underlying crossing. At the end of
this first step, we have converted our diagram into a graph D where the edges that correspond
to the crossings are labelled; this is illustrated on the right hand side of Figure 1.58.

Figure 1.59: Successive reducing moves applied to a graph resulting from Seifert’s algorithm.
Once we obtain a diagram whose associated graph D satisfies h(D) = 0, it is possible to isotope
it (in S 2 = R2 ∪ {∞}) so that all the Seifert circles are concentric, as can be seen in the bottom
right of this diagram.

Before describe the second step, we first introduce some terminology. We call two Seifert
circles in the plane compatible if they are concentric and share the same orientation or are non-
concentric and have opposite orientations. We denote the number of Seifert circles in a graph D
as above by n(D), and the number of pairs of incompatible Seifert circles of D by h(D). We
refer to h(D) as the height of D. We now describe two key points in this proof.

• a diagram represents a closed braid if and only if the graph D obtained by applying
Seifert’s algorithm as in the first step satisfies h(D) = 0.12

• the height of a diagram decreases by one if we perform a reducing move such as the one
depicted on the on the upper left of Figure 1.59. A very careful description of this move
can be found found in [KT08], but we simply note that on the level of the underlying link
diagram, it is a Reidemeister two move.
12
AC: To do: add reference. Change “bend" to reducing move in figure.

59
 Note that reducing moves do not change the type of the underlying link. As a consequence,
using the second of the two assertions above, after a sequence of reducing moves, we obtain a
diagram for our link whose corresponding graph D satisfies h(D) = 0. The first assertion then
implies that this diagram is a braid closure.

Figure 1.60: This figure shows how a braid word can be read off from a graph D with h(D) = 0.
In this case, we obtain σ1−1 σ2 σ3−1 σ2 σ1 σ2 σ3 σ2 σ2 .

More concretely, once h(D) = 0, performing isotopies in S 2 = R2 ∪ {∞} makes it is possible

to arrange all the Seifert circles concentrically (while keeping track of the positions of the labeled
edges, as in Figure 1.59) and then to read off the braid word from this diagram; this is illustrated
in Figure 1.60.

Summarising, the proof of Theorem 1.115 leads to the following algorithm to obtain a braid
closure:

1. Apply Seifert’s algorithm to the diagram and add the appropriate labeled edeges between
the circles.

2. Reduce the height of the graph by successive reducing moves

3. Once the height is zero, arrange that the diagram lie in concentric circles and read off a
braid word.

We invite the reader to verify that applying the first step of the algorithm to the knots 31 , 41
and 51 immediately leads to graphs with h(D) = 0 making it fairly quick to read off a braid
word.

Figure 1.61: The first Markov move M1 on the left, and the second Markov move M2 on the
right. M2 is shown with σn+1 .

Every link can be described as the closure of a braid however this description is highly non-
unique. For example, given any braid β, it is not hard to convince oneself that β and βσn±1 share
the same closure (use the first Reidemeister move) as do β and γβγ −1 for any braid γ Both of

60
these moves are described in Figure 1.61 and are referred to as Markov moves. A striking result
due to Markov shows that two braids share the same closure if and only if they are related by
a sequence of these moves and their inverses.
Theorem 1.116. Two braids have isotopic closures if and only if they are related by a sequence
of the following Markov moves and their inverses, as pictured in Figure 1.61.

• The first Markov move, M1 : β 7→ γβγ −1 for β, γ ∈ Bn .

• The second Markov move, M2 : β 7→ ιn (β)σn±1 , where ιn : Bn → Bn+1 is the inclusion.

1.4.3 A too brief introduction to quantum topology

In this section, we use braids to define so-called quantum knot invariants. While these gen-
eralise the HOMFLY polynomial, they are in general difficult to calculate in general, despite
often satisfying a generalised skein relation. We emphasise that this subsection is only a small
glimpse into a very broad field of mathematics which include quantum topology and quantum
algebra. The main reference for this section is [Lic97, Chapter 16] while the interested reader
can consult [Tur94] (and the references within) for a detailed account of quantum invariants of
knots.
In the previous section, we saw that a way to define link invariants is to first define braid
invariants that remain unchanged under both Markov moves. With that aim in mind, we
consider the following question: how can can one define representations of the braid groups?
While one such representation is the Burau representation from the sixth problem set, another
approach is to search for algebraic operators whose behavior mimicks the braid relations.
Definition 1.117. Given a commutative ring Λ and a free Λ-module V , a Yang-Baxter operator
is an isomorphism R : V ⊗ V → V ⊗ V that satisfies the following Yang-Baxter relation:

(R ⊗ idV )(idV ⊗R)(R ⊗ idV ) = (idV ⊗R)(R ⊗ idV )(idV ⊗R).

As we hinted at above, any Yang-Baxter operator gives rise to a representation of the

braid group. Indeed, using Propsotion 1.113 and Definition 1.117, one verifies that the assign-
ment σi 7→ I ⊗(i−1) ⊗ R ⊗ I ⊗(n−i−1) satisfies the braid relations and therefore leads to a group
homomorphism
Φ : Bn → Aut(V ⊗n ).
We now start introducing the machinery that will allow us to go from a braid invariant
to a link invariant. For that, we recall some notation on tensor products, referring to the
appendix for more detail. Given a free Λ-module V with basis e1 , . . . , en and a Λ-linear map
R : V ⊗ V → V ⊗ V , we write
X
kl
R(ei ⊗ ej ) = Rij ek ⊗ el .
kl

We introduce a shorthand to avoid using indices overly frequently. Namely, given such
a operator R, one can associate an operator Tr2 (R) : V → V defined in coordinates by
Pn kj ⊗2 ) → End(V ) can also be described
Tr2 (R)ik = j=1 Rij . This partial trace Tr2 : End(V
in a more coordinate-free fashion, but we will avoid this digression: instead we introduce one
of the main definitions of this section.
Definition 1.118. Fix a commutative ring Λ. An enhanced Yang-Baxter operator is a quadru-
plet (R, µ, a, b) where R : V ⊗ V → V ⊗ V is a Yang-Baxter operator, µ : V → V is a
diagonalisable Λ-isomorphism, and a, b ∈ Λ are units, all of which are subject to the following
relations:

61
 1. R(µ ⊗ µ) = (µ ⊗ µ)R,
2. Tr2 R±1 (I ⊗ µ) = a±1 bI.

We delay giving examples and instead describe a construction whose input is an enhanced
Yang-Baxter operator (over a ring Λ) and a braid β and whose output T (β) ∈ Λ will later be
shown to only depend on the closure of β.
Construction 1.119. Let (R, µ, a, b) be an enhanced Yang-Baxter operator over a ring Λ. To
an n-braid β (whose writhe is denoted w(β)), we associate the quantity
T (β) = a−w(β) b−n Tr Φ(β)µ⊗n ∈ Λ.

Since we delayed giving examples, we cannot immediately make this definition concrete. On
the other hand, we can already notice that it simplifies for any braid representing the unknot:
for example, for the braid σ1 , a short calculation using Definition 1.118 shows that T (σ1 ) =
b−1 Tr(µ).
We now show that T is a link invariant, allowing us to write T (L) for any oriented link L.
Theorem 1.120. Let Λ be a commutative ring and let (R, µ, a, b) be an enhanced Yang-Baxter
operator on a free Λ-module V . For any braid β ∈ Bn , the quantity T (β) ∈ Λ depends only on
the closure of the braid β, making it a link invariant.

Proof. By Theorem 1.116, we need only prove that T (β) is invariant under the two Markov
moves. To check invariance under the first Markov move, we let β, γ ∈ Bn be two braids and
write β 0 = γβγ −1 . We must show that T (β) = T (β 0 ). Using the definition of T , we have
0
T (β 0 ) = a−w(β ) b−n Tr(Φ(γ)Φ(β)Φ(γ)−1 µ⊗n ).
Observe that w(β 0 ) = w(γ) + w(β) + w(γ −1 ) = w(β). Also, the Yang-Baxter operator R
commutes with µ⊗2 and Φ commutes with µ⊗n , so using the conjugacy invariance of the trace,
we conclude that T (β 0 ) = T (β) from the following calculation:
Tr(Φ(γ)Φ(β)Φ(γ)−1 µ⊗n ) = Tr(Φ(γ)Φ(β)µ⊗n Φ(γ)−1 ) = Tr(Φ(β)µ⊗n ).
It remains to check invariance of T under the second Markov move. We omit the detailed
argument since it involves a long but not particularly instructive calculation: given a braid β ∈
Bn , one must show that T (ι(β)σn±1 ) = T (β) where ι(β) ∈ Bn+1 denotes β with one additional
strand. Thus T is invariant under both Markov moves and therefore defines a link invariant,
concluding the proof.

In addition to being a link invariant, T also satisfies a skein relation of sorts. Note however
that unlike the skein relations we have seen so far, there is no guarantee that this relation will
be useful for computing T (L): it is possible that the relation will only express T (L) in terms
of T (L0 ) for links L0 that are more complicated than L. To state this “generalised skein relation",
we will be assuming that L−n , . . . , L−1 , L0 , L1 , . . . , Ln are links which are isotopic outside of a
neighbhorhood of a point where they are as in Figure 1.62.

Figure 1.62: The generalised skein relation satisfied by certain quantum invariants involves links
that locally differ by a certain number of twists.

62
 Observe that when n = 1, we recover the situation that we have been considering when
studying the skein relations satisfied by the Jones, Alexander-Conway and HOMFLY polyno-
mials. In those cases, we wrote L± instead of L±1 .
We omit the proof of the following lemma referring instead to [Lic97] for details.

Lemma 1.121. Let Λ be a commutative ring and let T be the quantum invariant associated to
an enhanced Yang-Baxter invariant (R, µ, a, b) as in Construction 1.119. Suppose n > 0 is a
positive integer and assume that k−n , k−n+1 , . . . , kn ∈ Λ satisfy the following relation:

k−n R−n + · · · + k−1 R−1 + k0 I + k1 R1 + · · · + kn Rn = 0.

If L−n , . . . , L−1 , L0 , L1 , . . . , Ln are links that are isotopic outside of a neighbhorhood of a point
where they are as in Figure 1.62, then the quantum invariant T satisfies
n
X
ki ai T (Li ) = 0.
i=−n

So far, we have not yet given a single example of a Yang-Baxter operator, let alone of an
enhanced Yang-Baxter operator. As a consequence, the rest of this section is devoted to showing
how, with the right choice of an enhanced Yang-Baxter operator, we obtain a polynomial T (L)
which, when normalized, recovers an evaluation of the HOMFLY polynomial.

Construction 1.122. For every m > 1, we define what will turn out to be a Yang-Baxter
operator over the free Z[q ±1 ]-module of rank m. Writing the canonical basis of V := Z[q ±1 ]m
as e1 , e2 , . . . , em , we consider the operator Rm : V ⊗ V → V ⊗ V defined on the basis {ei ⊗ ej }i,j
of V ⊗ V by the formulas

−q(ei ⊗ ej ) i = j,


Rm (ei ⊗ ej ) = e ⊗e +
j i (q −1 − q)(ei ⊗ ej ) i < j,

e ⊗ e

i > j.
j i

The proof of the following proposition follows from a direct computation involving the
definition of Rm and so we omit the details.

Proposition 1.123. The operator Rm from Construction 1.122 satisfies the following proper-
ties:

(a) Rm is a Yang-Baxter operator.

−1 = (q −1 − q)I, where I is the identity matrix.
(b) Rm − Rm

(c) Rm has an enhancement µm = diag(q −m+1 , q −m+3 , . . . , q m−3 , q m−1 ) with am = −q m

and bm = 1.

In particular, (Rm , µm , am , q) is an enhanced Yang-Baxter operator for every m ≥ 0 and so

there is an associated quantum link invariant Tm .

We note that the link invariant Tm from Proposition 1.123 does not vanish on the unknot U .
To see this, use the definition of Tm to note that

Tm (U ) = a−w(U ) b−n Trace(φ(U )µ⊗n ) = a0 b−n Trace(Iµ⊗n )

is non-zero because a and b are units, and µ⊗n has all positive multiples of q on its diagonal.
As a consequence, for any link L, we deduce that Tm (L)/Tm (U ) is a link invariant whose value
on the unknot is 1. The next theorem relates this invariant to the HOMFLY polynomial.

63
Theorem 1.124. For every link L, the link invariant Tm associated to the enhanced Yang-
Baxter operator (Rm , µm , am , bm ) from Construction 1.122 is related to the HOMFLY polyno-
mial as follows:
Tm (L)
= PL (q m , q −1 − q),
Tm (U )
where PL is the HOMFLY polynomial of L.

Proof. Since by definition TTm

m (L)
(U ) is equal to one for L = U , we need only verify that it satifies the
same skein relation as PL (q , q −1 − q). The required equality then follows from Theorem 1.105
m

which ensures that the skein relation and the normalisation condition entirely determine this
polynomial.
We know from the second item of Proposition 1.123 that Rm satisfies the polynomial equa-
tion Rm − Rm −1 + (q − q −1 )R0 = 0. Applying Lemma 1.121 and recalling that a = −q m , we
m
conclude that Tm satisfies the skein relation

−q m Tm (L+ ) + q −m Tm (L− ) = q −1 − qTm (L0 )

Dividing out by Tm (U ), we conclude that TTm

m (L) m −1 −
(U ) satisfies the same skein relation as PL (q , q
q) and, since both equal one on the unknot, the theorem follows.

64
Chapter 2

The algebraic topology of the link

exterior

At this point, we have defined several link invariants including the Alexander polynomial, the
signature, the Jones polynomial, the HOMFLY polynomial and we in fact hinted at the fact
that the theory of quantum groups leads to a myriad of more mysterious link invariants. The
Alexander polynomial and signature were defined using Seifert matrices, while quantum invari-
ants were obtained by a mix of diagrammatic and algebraic means. A question remains: why
does this work?
Answering this question for the Alexander polynomial is the main goal of this chapter.
Namely, we explain how the Alexander polynomial can obtained by using the algebraic topology
of the link exterior. While a similar complete understanding the signature is available, the
topological meaning of quantum invariants remains elusive to this day and is the focus of much
current research in knot theory.
To motivate the structure of this section, start by noting that the homeomorphism type of
the knot complement S 3 \ L is a link invariant. This seemingly inocuous observation implies
that every homeomorphism invariant of S 3 \ L is in fact a link invariant. With this in mind,
how would an algebraic topologist extract invariants from a link? Here the first observation
is that for every link n-component link L, a Mayer-Vietoris argument (or Alexander duality)
shows that 


 Z i = 0,

Zn

i = 1,
Hi (S 3 \ L) = n−1


 Z i = 2,

0

otherwise.
This implies that the homology of the link complement is a useless invariant. On the other
hand, as we will see in Section 2.1, the link group π1 (S 3 \ L) is a very strong invariant, but
it’s unfortunately difficult to apply in practice (distinguishing two groups up to isomorphism is
often no easier than distinguishing two links up to isotopy).
As the link group is an extremely good invariant but is hard to calculate, the next idea is,
informally, to “simplify" S 3 \ L by looking at its covers and then to take homology. As we will
see in Section 2.2, this idea is extremely fruiful and in particular leads to a nice topological
interpretation of the Alexander polynomial in terms of the algebraic topology of S 3 \ L.
The Alexander polynomial and other invariants obtained from the Seifert matrix are un-
fortunately not all powerful. Some of these limitations will be described in Section 2.3 when
we learn about satellite knots: it will become routine to produce distinct knots with identical
Seifert matrices (and therefore identical Alexander polynomials and signatures).
Apart from quantum invariants (including the Jones and HOMFLY polynomials), mod-

65
ern knot theory offers many alternatives to the Alexander polynomial and the signature.
Arguably, the most successful are Heegaard-Floer homology [OS04c, OS04a, OS03a, OS04b,
OS04b, OS03b] and Khovanov homology [Kho00] which were introduced at the turn of the cen-
tury. While studying these “knot homologies" is beyond the scope of this class, Section 2.5. will
instead briefly introduce another (relatively) recent type of knot invariant: twisted Alexander
polynomials.

2.1 The link exterior and the link group

In this section, given a link L, we study the group π1 (S 3 \ L) and describe an algorithmic way
to obtain a presentation for it. References include [Lic97, pages 110-111], [Liv93, pages 105-108]
and [Rob15, Section 8.4].

Definition 2.1. The group of a link L is the fundamental group of the complement of L in S 3 ,
denoted as π1 (S 3 \ L).

We briefly describe an equivalent description of the link group. By a tubular neighborhood

of a link L, we mean a open neighbhoorhood νL of L that is homeomorphism to L × int(D2 ).
With this notation, the exterior of a link L refers to XL := S 3 \ νL and we note that this
3-manifold S 3 − L deformation onto XL and therefore π1 (S 3 \ L) = π1 (XL ).

Theorem 2.2. The group of a link is a link invariant.

Proof. Let L and L0 be two ambiently isotopic links. The isotopy (ht )t∈[0,1] in particular gives a
homeomorphism h1 : S 3 → S 3 that takes L to L0 which therefore restricts to a homeomorphism
= S 3 \ L0 . Thus, we obtain π1 (S 3 − L) ∼
S3 \ L ∼ = π1 (S 3 − L0 ) as required.

As the next proposition shows, the knot group is a strong invariant: it can be used to
characterise the unknot.

Proposition 2.3. A knot K is the unknot if and only if π1 (S 3 − K) is infinite cyclic.

Proof. We first prove backward direction, which is the simpler of the two. Namely, we assume
that K is the unknot and prove that π1 (XK ) = Z. To see this, we note that

S3 ∼
= ∂D4 ∼
= ∂(D2 × D2 ) ∼
= (D2 × S 1 ) ∪ (S 1 × D2 )

so that S 3 is the union of 2 solid tori one either of which can be thought of a the exterior (resp.
a closed tubular neighoorhood) of K. Thus, if K is the unknot, then XK ∼ = S 1 × D2 and so
π1 (XK ) = Z, as required.
The converse is harder and relies on (the following consequence of) Dehn’s lemma (a difficult
result in 3-manifold topology): If K is not the unknot, then the inclusion ∂XK ,→ induces
an injective homomorphism π1 (∂XK ) → π1 (XK ). We do not prove this result but refer the
interested reader to [Lic97, pages 112-113]. Instead, we conclude the proof of the proposition.
Suppose for the sake of contradiction that K is not the unknot but has π1 (XK ) ∼ = Z: since
∂XK is a torus, Dehn’s lemma implies that there is an injection Z2 ∼ π
= 1 (∂X) ,→ π ∼
1 (XK ) = Z,
which is clearly false. Thus K must be the unknot.

We now present an algorithm to find the presentation of the group of a link.

Proposition 2.4. Let L be a link that has a diagram with m segments and n crossings. The
group of L admits a presentation with

• One generator for each strand; use gi to denote the generator associated to the i-th strand.

66
 • One relator for each crossing: for each crossing c, add the relator rc = gk gi gk−1 gj−1 if the
crossing is negative and rc = gk−1 gi gk gj−1 if the crossing is positive, where gi , gj and gk
are defined as in Figure 2.1.

This presentation is commonly known as Wirtinger’s presentation.

Figure 2.1: Notation for crossings for the Wirtinger presentation, from [Lic97, page 110].

Remark 2.5. Before giving a detailed proof of Proposition 2.4, we provide some intuition for
why it holds.

• Each gi corresponds to a loop from a basepoint around the ith segment in a positive
direction, i.e., the loop creates positive crossings with the ith segment. See Figure 2.2 for
an example.

Figure 2.2: Loop corresponding to g1 .

• For each negative crossing, from Figure 2.3, we see that gj = gk gi gk−1 . Similarly, for each
positive crossing, gj = gk−1 gi gk . Therefore, each rc is trivial. The challenge is to show
that these rc are indeed all of the relators.

Figure 2.3: Diagrams to show gj = gk gi gk−1 .

Example 2.6. Here we will look at Wirtinger presentation of three knots.

67
 1. We calculate a Wirtinger presentation for the group of the trefoil illustrated in Figure 2.4.
Since there are three arcs, we have three generators, say x, y, z. Using Proposition 2.4, we
see that the relations between these generators are: xz = zy and yx = xz and zy = yx.
Therefore, the Wirtinger presentation is

W = hx, y, z | xz = zy, yx = xz, zy = yxi.

x
z

Figure 2.4: A diagram for the trefoil on which the arcs are labelled.

2. We calculate a Wirtinger presentation for the group of the square knot SK illustrated in
Figure 2.5. Let u, v, w, x, y, z be the generators. Then using Proposition 2.4, we see that
the Wirtinger presentation is

WSK = hu, v, w, x, y, z | zu = uy, vz = zu, uy = yz, yx = wy, xw = vx, wy = xwi.

(2.1)

u w
y

z x
v

Figure 2.5: A diagram for the square knot on which the arcs are labelled.

3. We calculate a Wirtinger presentation for the group of the granny knot GK illustrated in
Figure 2.6. Let u, v, w, x, y, z be the generators. Then using Proposition 2.4, we see that
the Wirtinger presentation is

WGK = hu, v, w, x, y, z | zu = uy, vz = zu, uy = yz, yw = wx, xv = vw, wx = xvi.

(2.2)

u w
y

z x
v

Figure 2.6: A diagram for the granny knot GK on which the arcs are labelled.

Remark 2.7. Note that in both the presentation in (2.1) and the presentation in (2.2), we
have y = v. On the other hand, in (2.1), we can write u = yzy −1 and x = wyw−1 . Therefore,
we can reduce the presentation for WSK to

WSK = hw, y, z | zyz = yzy, ywy = wywi.

Similarly, noticing u = yzy −1 and w = xyx−1 in (2.2), we can reduce WGK to

WGK = hx, y, z | zyz = yzy, yxy = xyxi.

68
In particular, we deduce that WSK = ∼ WGK . However, note that despite having isomorphic
groups, these two knots are distinct, distinguished by the sginature: σ(SK) = 0 but σ(GK) =
±4. More generally„ the knot group fails to distinguish K1 #K2 from K1 # − K̄2 for any two
knots K1 and K2 ; see [Kaw96, Corollary 6.1.10]).

Proof of Proposition 2.4. idea of the proof is to construct the link exterior in steps and use van
Kampen to see how the fundamental group changes at each step. Thinking of S 3 as R3 ∪ {∞},
any loop in S 3 may be homotoped to avoid ∞, so the link group is the same as π1 (R3 \ νL).
To make the construction simpler, we will actually construct R3 \ νL, but this has the same
fundamental group as R3 \ νL, since L and L are related by a reflection in R3 . The construction
proceeds in the following steps. Let Xi denote the resulting space after step i is performed.

• Step 1: Let X1 be the closed half-space {(x1 , x2 , x3 ) ∈ R3 : x3 ≤ 0}. Draw the diagram
of L on the top {x3 = 0} of X1 (we are not adding anything to X1 ).

• Step 2: Add half-cylinder caps across each strand of L to obtain X2 . The caps will "encase"
νL.

• Step 3: Add "pavilions" to connect the half-cylinder caps at each crossing to obtain X3 .
This allows the top strand of L (the bottom strand of L) to pass over the existing half-
cylinder cap.

• Step 4: Add the upper half of R3 to obtain X4 = R3 \ νL.

69
 Now we see how the fundamental group changes at each step. The Xi are all path-connected,
so the choice of basepoint is not important.

• If the link diagram has n strands, the space X2 is homotopy equivalent to a wedge of
n circles, and hence π1 (X2 ) is free on generators [g1 ], . . . , [gn ]1 . Geometrically, given a
basepoint x ∈ X2 , the gi go directly from x to the ith strand, around the half-cylinder
cap, and back to x; we choose the gi so that the linking number of gi with the component
containing the ith strand of L is −1.

• The "pavilions" are homeomorphic to the 2-disc D2 and are attached to X2 along their
boundaries, i.e. they correspond to 2-cell attachments. By van Kampen’s theorem, if a
2-disc is attached along its boundary ϕ to a path-connected space X, the effect on the
fundamental group is adding the relation ϕ to π1 (X). Thus, as shown in Figure 2.7, at
positive crossings we add the relations gj = gk−1 gi gk . Similarly, at negative crossings, we
add the relations gj = gk gi gk−1 .

Figure 2.7: The relations in the Wirtinger presentation arising from a 2-cell attachment.

• Finally, adding the upper half of R3 to X3 to produce X4 = R3 \ νL does not change the
fundamental group.

Thus the link group is given by the Wirtinger presentation.

2.2 The Alexander Module

As we mentioned at the beginning of this chapter, deciding whether two groups are isomorphic
is a challenging problem and this is a serious drawback for the link group. The idea is then to
1
By abuse of notation, we might use gi and [gi ] interchangeably.

70
“simplify" the link exterior by considering some of its covers. More precisely, in this section, we
introduce the infinite cyclic cover and the Alexander module of a link. We then describe how
the Alexander polynomial can be extracted from the Alexander module. References for this
section include [Lic97, Chapters 6 and 7] and [BZ03].
Recall the classification of covering spaces from Theorem C.1 in the appendix: for most
spaces, specifying a subgroup of the fundamental group uniquely determines a covering space
of said space. The next definition applies this to the exterior XL of a link L.
Definition 2.8. The infinite cyclic cover of a link L = K1 ∪. . .∪Kn , is the covering space XL∞ →
XL , that corresponds to the kernel of the homomorphism φ : π1 (XL ) → Z, [γ] 7→ `k(K1 , γ) +
· · · + `k(Kn , γ).
Before giving a concrete example, we recall one additional fact from covering space theory.
For a normal subgroup, N / π1 (X), the deck transformation group of the corresponding cover,
denoted G(XN ) is isomorphic to π1 (X)/N . As a consequence, in the case where N = ker(φ) for
some surjective homomorphism φ : π1 (X) → G, the the first isomorphism theorem shows that
G(XN ) ∼
= π1 (X)/ ker(φ) ∼
= im(φ) = G.
In particular, the infinite cyclic cover of a link exterior has deck transformation group Z.
Example 2.9. We show that the infinite cyclic cover of the unknot K is homeomorphic to R ×
D2 . Since K is the unknot, recall that S 3 \νK is homeomorphic to the solid torus: XK =
S 1 × D2 . Since D2 is contractible, π1 (S 1 × D2 ) ∼
= π1 (S 1 ) ∼
= Z, generated by the curve µ
illustrated in Figure 2.8.

Figure 2.8: A tubular neighbhorhood of the unknot.

∞ is
As φ([µ]) = 1, we see that φ is an is isomorphism, so ker(φ) is trivial and therefore XK
∞ ∼ 2
the universal cover of XK , i.e. XK = R × D , as asserted.
We can think of XU∞ ∼= R × D2 as countably many [0, 1] × D2 glued together. This makes
∞
the action of Z on XK apparent: an element a ∈ Z translates the cylinder a units to the right.
Next, we explain why H1 (XL∞ ) supports the structure of a module over the Laurent poly-
nomial ring Z[t±1 ]; we refer to Appendix B for some recollections on modules. To describe this
module structure, we use the fact that Z acts on XL∞ by deck transformations and define the
action as follows.

• The deck transformation t acts on H1 (X∞ ) by t · [γ] = t∗ ([γ]) = [t(γ)].

• It follows that Z = hti = G(XL∞ ) acts on H1 (XL∞ ) by tk · [γ] = tk∗ ([γ]).
• The group ring Z[Z] = Z[t±1 ], formed by integral linear combinations of elements of hti ∼
=
Z, acts on H1 (X∞ ) by linearly extending the action of hti, i.e. for [γ] ∈ H1 (XL∞ ) and
nk ∈ Z,  
X X  
 nk tk  · [γ] = nk tk · [γ]
k∈Z k∈Z

71
 We now come to the main definition of this section.

Definition 2.10. The Alexander module of a link L is the Z[t±1 ]-module H1 (XL∞ ).

The Alexander module is in general difficult to calculate directly from its definition because
the homeomorphism type of the link exterior and its infinite cyclic cover can rarely be described
in a simple way. For the unknot however, everything can be understood explicitly.

Example 2.11. We show that the Alexander module of the unknot U is trivial. By Example 2.9,
the infinite cyclic cover of the unknot is XU∞ ∼
= R × D2 . Since this space is contractible, the
conclusion now follows readily: H1 (XU∞ ) = H1 (R × D2 ) = 0.

The question is now “how to extract invariants from the Alexander module?". This is
a purely algebraic question which we can study for more general rings and modules. The
next definition requires some familiarity with homomorphisms and quotients of modules; see
Appendix B for a brief refresher.

Definition 2.12. Let R be a commutative ring. A presentation matrix for a finitely gener-
ated R-module M is a matrix of a homomorphism ϕ : Rn → Rm such that M is isomorphic
to Rm / im(ϕ).

The name “presentation matrix" is not coincidental. One can think of the (canonical)
basis of Rm as being the generators of M , and the images of the (canonical) basis of Rn that
we quotient by are the relations imposed. So a presentation matrix for a module contains
information analogous to that of a presentation of a group, which may be a bit more familiar.

Example 2.13. We describe examples of matrices presenting some familiar modules.

1. Both (2) and ( 20 01 ) present Z2 , thought of as a module over Z.

∼ Z2 ⊕ Z3 so
2. The matrix (6) presents the Z-module Z6 , but using the isomorphism Z6 =
2 0
does ( 0 3 ) .

3. The Z[t±1 ]-module M = Z[t±1 ]/(t − 1) (which is isomorphic to Z with the trivial Z[t±1 ]-
action) admits (t − 1) as a presentation matrix.

As the preceding examples illustrate, the same module can have different presentation ma-
trices. It is of interest to see whether these different matrices are related. A complete answer is
provided by the following result.

Proposition 2.14. Any two presentation matrices A1 and A2 for a finitely generated R-
module M are related by a sequence of the following moves and their inverses:

1. Change of basis, i.e., A1 = P A2 Q−1 for invertible matrices P, Q.

!
A1 0
2. Addition of a trivial generator, i.e, A2 = .
0 1
 
3. Adding of a trivial relation, i.e., A2 = A1 0 .

Proof. We only describe the less challenging part of the proof: none of these moves changes the
isomorphism class of the module M . To achieve this, we think of presentation matrices using
generators and relations as described above this proposition.

1. Multiplying a matrix by an invertible matrix corresponds to changing rewriting the gen-

erators and relations, but this doesn’t affect the isomorphism type of the module.

72
 2. In the second move, we add one generator and one relation, but the relation we impose
is that the new generator is trivial, so again the isomorphism type of the module is
unchanged.

3. For the third, move adding a column of zeros does not change the image of the map, so
we have not changed the quotient.

Let us return to the task of extracting invariants from the Alexander module of an oriented
link, the task that we initially set out to carry out. The main theorem in this setting is the
following; a proof can be found in [Lic97, Theorem 6.5].
Theorem 2.15. If A is a Seifert matrix for an oriented link L, then tA − AT is a presentation
matrix for the Alexander module H1 (XL∞ ).

This theorem is great because it gives a concrete and manageable way to calculate a pre-
sentation matrix for the Alexander module of any link. We describe an example.
Example 2.16. We use Theorem 2.15 to show that the Alexander module of the trefoil is
isomorphic to Z[t±1 ]/(t2 − t + 1). We look at the right trefoil knot K, which we know has a
Seifert matrix !
−1 1
A= .
0 −1
 
We deduce that tA − AT = 1−t t ∞ ) by Theorem 2.15. Since ±tn are the
presents H1 (XK
−1 1−t
units in Z[t±1 ], we deduce that tA − AT is congruent to
! ! ! !
−t −1 1−t t 1 − t−1 −t−1 t2 − t + 1 0
=
0 1 −1 1 − t −t−1 −t−1 0 1

and use Proposition 2.14 to deduce that that (1 − t + t2 ) is also a presentation matrix for the
∞ ). We conclude that H (X ∞ ) ∼ Z[t±1 ]/(t2 − t + 1), as claimed.
module H1 (XK 1 K =

One may notice that the polynomial t2 − t + 1 is actually the Alexander polynomial of the
knot we were considering. This seems a bit too good to be a coincidence, and indeed the general
theory does show us that this is not entirely so.
The next definition could be made over much more general rings than R = Z[t±1 ] but we
restrict ourselves to that ring for the sake of simplicity and concreteness.
Definition 2.17. The first elementary ideal of a Z[t±1 ]-module M is the ideal ε1 (M ) ⊂ Z[t±1 ]
generated by all (n × n)-minors of any (m × n)-presentation matrix for M with m ≥ n, and the
order of M , denoted ∆M , is a generator for the smallest principal ideal containing ε1 (M ).
Remark 2.18. A couple of remarks on this definition are needed.

1. A priori, it is not clear that ε1 (M ) is independent of the choice of a presentation matrix,

Proposition 2.14 and some linear algebra show that this is indeed the case.

2. The order of M is only well defined up to multiplication by units of Z[t±1 ], i.e. up to

multiplication by ±tn : indeed a principal ideal does not admit a unique generator, but
any two generators are related by mulitplication by units.

3. The situation simplifies drastically if M admits a square presentation matrix A: in this

.
case ε1 (M ) is the principal ideal generated by det(A) and ∆M = det(A).

Finally, using this terminology, we are able to describe a more topological and intrinsic
definition of the Alexander polynomial.

73
Theorem 2.19. The Alexander polynomial of an oriented link L is the order of its Alexander
module.

Proof. Theorem 2.15 implies that for any Seifert matrix A of L, the matrix tA − At presents
the Z[t±1 ]-module H1 (XL∞ ). This matrix is square so Remark 2.18 and our definition of the
. .
Alexander polynomial now give the required result: ∆H1 (XL∞ ) = det(tA − At ) = ∆L (t).

This alternate definition provides us with a topological interpretation of the Alexander

polynomial as an invariant derived from a cyclic cover of the exterior of our link, and also hints
towards the fact that there are in fact many other polynomial invariants we can define, although
they may be a bit hard to access.

2.3 Satellite knots and the limitations of classical invariants

We define satellite knots and give several examples. We also state the behavior of the Alexander
polynomial under satellites. Finally, we will illustrate how this leads to swaths of distinct knots
with identical Alexander polynomial and signature. References include [Lic97, page 10 and 60-
62].
We start with some set-up and terminology. In what follows, the we consider the standardly
embedded solid torus T ⊂ R3 :
( 2  2 )
1
q
3 2
T := (x, y, z) ∈ R : x2 + y2 −1 +z ≤ .
2

The circle {(x, y, 0) ∈ R3 : x2 + y 2 = 32 } ⊂ T is called the standard longitude of T and is denoted

by L.

Figure 2.9: Solid torus T and circle L.

A meridional disc of an oriented knot K is a disc D ⊂ ν(K) that intersects K once. A

longitude of a knot K ⊂ S 3 , denoted as L(K), is an oriented circle in ∂(νK) that intersects any
meridional disc once and satisfies `k(K, L(K)) is equal to 0.

Figure 2.10: An embedding ι of the standardly embedded solitd torus T onto a tubular neighb-
hoorhood of the trefoil K that takes the standard longitude of T to the longitude of K.

74
 Now comes the main definition of this section, also illustrated in Figure 2.10.

Definition 2.20. Given a knot P ⊂ T and a knot K ⊂ S 3 , the satellite knot P (K) with
pattern P is constructed as follows: pick an embedding ι : T → S 3 that is a homeomorphism
onto ν(C) and that sends the standard longitude L of T to a longitude L(C) of C; the knot
P (K) is defined as ι(K).

Example 2.21. Here, we give some examples of satellite knots.

1. For (p, q)-coprime, the (p, q)-cable of a knot K is the satellite knot Kp,q obtained by
taking the torus knot Tp,q as a pattern and K as the companion. This is illustrated for
p = 2, q = 1 in Figure.

Figure 2.11: The (2, 1)-cable on the trefoil.

2. We let the reader verify that the connected sum is a particular case of the satellite oper-
ation (take P = K1 , wraping once around T so that P (K1 ) = K1 #K2 ).

3. The Whitehead double W h(K) of a knot K is obtained by applying the Whitehead pattern
on the left hand side of Figure 2.12

Figure 2.12: The Whitehead double of the trefoil and figure eight knots.

4. The knot on right hand side in Figure 2.13 is obtained by tying the trefoil (knot J1 := 31 )
into the left band of the knot R := 946 and tying the figure-8 knot (knot J2 := 41 ) into the
right band of R. Naturally, as illustrated in the left hand side of Figure 2.13 one can tie
arbitrary knots J1 , J2 in place of the trefoil and figure eight knot and, in particular, for the
unknot U , note that R = R(U, U ). Thus, the resulting knot R(J1 , J2 ) is both a satellite
with pattern R(J1 , U ) and companion J2 and a satellite knot with pattern R(U, J2 ) and
companion J1 . Both R(J1 , U ) and R(U, J2 ) are also satellite knots with pattern U .

We state the behavior of the Alexander polynomial under satellites. For that we say that the
winding number of a pattern P ⊂ T is the homology class [P ] ∈ H1 (T ) = Z. Less formally, it’s
the number of times, counted with sign, that P winds around the solid torus. As example, the
(2, 1)-pattern has winding number two, whereas the Whitehead pattern has winding number 0.

Proposition 2.22. Let P ⊂ T be a pattern of winding number n, and let K ⊂ S 3 be a knot.

Then
∆P (K) (t) = ∆P (t)∆K (tn ) σP (K) (ω) = σP (ω) + σK (ω n ).
In these formulas, when we write ∆P and σP (ω), we think of P ⊂ T ⊂ S 3 as a knot in S 3 .

75
Figure 2.13: Another example of the satellite process consists in tying in knots into the bands
of R := 946 .

Before moving on to examples, we record one remark about the formula for the signature.
Recall that the “classical" signature of a link is defined as σ(L) = sign(A + AT ) where A is any
Seifert matrix for L. so that σ(L) = σL (−1). Setting ω = −1 in the formula of Proposition 2.22,
we see that σ(P (K)) = σ(P ) if the winding number is even and σ(P (K)) = σ(P ) + σ(K) if the
winding number is odd.

Example 2.23. We give some examples for the formulas above.

1. The connected sum K1 #K2 of two knots is a satellite with pattern K1 , companion K2
and winding number 1 so that we recover the familiar formulas

∆K1 #K2 (t) = ∆K1 (t)∆K2 (t),

σ(K1 #K2 ) = σ(K1 ) + σ(K2 ).

2. The torus knot pattern has winding number p and Alexander polynomial ∆Tp,q =
(tpq −1)(t−1)
(tp −1)(tq −1) (as we will see in Example), so that the Alexander polynomial of a cable
is given by
(tpq − 1)(t − 1)
∆Kp,q (t) = ∆K (tp ).
(tp − 1)(tq − 1)

3. The Whitehead double W h(K) of any knot K has winding number zero and therefore,
since the Whitehead pattern is unknotted when viewed as a knot in S 3 , we obtain

∆W h(K) (t) = ∆K (t0 )∆U (t) = ∆K (1) = 1,

σ(W h(K)) = σ(U ) = 0.

In particular, neither the Alexander polynomial nor the signature can detect the difference
between any Whitehead double and the unknot.

4. Using Proposition 2.22, the family of knots R(J1 , J2 ) from Example 2.21 to have the same
Alexander polynomial and signature than the slice knot R = 946 (in fact with respect to
appropriate bases, they even admit the same Seifert matrix). Thus non of the obstructions
we have learnt up to now can obstruct R(J1 , J2 ) from being slice. One possible way to
prove this knot is not slice is to use the final theorem of the introduction of [COT03]
which states that R(J1 , J2 ) is not slice for any J1 and J2 whose signature functions both
have nonzero integrals over S 1 . For example, taking J1 = J2 = 31 , we have that R(J1 , J2 )
is algebraically slice but not slice.

76
Proof outline of Proposition 2.22. We claim (without proof; the interested reader can con-
sult [Lic97, Theorem 6.15]) that there is a Seifert matrix A for the pattern P and a Seifert
matrix C for the companion K such that the satellite knot P (K) has A ⊕ B as a Seifert matrix
where  
C C ... C
 t
C C . . . C 
B= .
 .. . . .
 . . . . 

Ct Ct . . . C
We describe how this claim implies the proposition. Using our definition of the Alexander
polynomial and some tedious linear algebra to calculate deg(tB − B T ) (an intermediate step
makes use of the fact that det(C − C T ) = ±1 if C is the Seifert matrix of a knot), we obtain
the required result:

∆P (K) (t) = det(tA − AT ) det(tB − B T ) = ∆P (t) det(tn C − C T ) = ∆P (t)∆K (tn ).

This concludes our proof outline.

2.4 Fox calculus

At this point, we have two equivalent definitions for the Alexander polynomial ∆K of a knot
and K understand some of its limitations. The goal of this section is to introduce a third
definition of ∆K that is both practical for calculations and will be useful when we briefly study
twisted knot invariants. To this effect, we first introduce Fox calculus, and then show how it
can be used it to calculate the Alexander polynomial of a link from a presentation of its group.
References for this section include [Lic97, pages 116-119] as well as [BZ03, Chapter 9].

Definition 2.24. Let Fn be the free group on generators g1 , . . . , gn . The Fox derivatives are
the maps ∂g∂ j : Fn → Z[Fn ] defined by

∂gi
• ∂gj = δij , where δij is the Kronecker delta,

∂(gi−1 )
• ∂gj = −δij gi−1 ,

∂(uv) ∂u ∂v
• ∂gj = ∂gj + u ∂g j
.

∂
The existence and uniqueness of the ∂gj is not obvious; a proof is given in [BZ03, Proposi-
tion 9.6].

Construction 2.25. Let L be a link with k components L1 , . . . , Lk , exterior XL , and G =

π1 (XL ) the group of L. Recall the homomorphism α : G → Z ∼ = hti, [γ] 7→ t`k(γ,L1 )+...+`k(γ,Lk ) .
Suppose G has presentation hg1 , . . . , gn |r1 , . . . , rn−1 i given by a Wirtinger presentation; the
Wirtinger presentation is defined so that there are the same number of generators and relations,
but it is known that any single relation may be obtained from the others and hence omitted.
Fox derivatives may be used to calculate the Alexander polynomial of L as follows:
 
∂ri
1. Write down the (n − 1) × n matrix A = ∂gj of Fox derivatives.

e : Z[G] → Z[t, t−1 ]. Apply α

2. Extend α : G → hti linearly to a map α e to each entry of A
and call the new matrix A.
e

3. Take any (n − 1) × (n − 1) minor of A,

e i.e. remove a column of Ae and take the determinant
±1
of the resulting matrix. Up to a unit of Z[t ], this is the Alexander polynomial ∆L (t).

77
Remark 2.26. If G is given by any presentation hg1 , . . . , gn |r1 , . . . , rn−1 i where there is one
more generator than relation, not necessarily a Wirtinger presentation, the same method can
be used with a slight adjustment. In step 3, one must take the ideal in Z[t, t−1 ] generated by all
(n − 1) × (n − 1) minors of A;e up to units, the Alexander polynomial is an element generating
this ideal. It just happens that if a Wirtinger presentation is used, all of the (n − 1) × (n − 1)
minors of Ae are the same up to units. See [Lic97, Pages 117-118] and [BZ03, Corollary 9.11].

Example 2.27. We calculate the Alexander polynomial of the figure eight knot 41 . As we did
in the seventh problem set, we have the following Wirtinger presentation of the knot group:
D E
w, x, y, z | x−1 yxz −1 , z −1 wzx−1 , wxw−1 y −1 , yzy −1 w−1 .

We reduce the number of generators and relations by substitution. Note z = x−1 yx and
w = zxz −1 = x−1 yxy −1 x. Since one relation is always redundant, we ignore wxw−1 y. Hence,
the only relation we have left is

r = wyz −1 y −1 = x−1 yxy −1 xyx−1 y −1 xy −1 .

We can now apply the algorithm to calculate the Alexander polynomial.

1. We take the Fox derivatives by breaking off the right one-by-one to get
∂r
= x−1 yxy −1 xyx−1 y −1 − x−1 yxy −1 xyx−1 + x−1 yxy −1 + x−1 y − x−1 ,
∂x
∂r
= −x−1 yxy −1 xyx−1 y −1 xy −1 − x−1 yxy −1 xyx−1 y −1 + x−1 yxy −1 x − x−1 yxy −1 + x−1 .
∂y

2. On abelianizing, both generators x and y of the Wirtinger presentation are sent to t from
the definition of the generators in the Wirtinger presentation, so
 
∂ri
= 1 − t + 1 + 1 − t−1 −1 − 1 + t − 1 + t−1 = −t−1 + 3 − t t−1 − 3 + t .



α
e ∂gj

3. Since we used a Wirtinger presentation, delete the first column to get

˙ t−1 − 3 + t.
∆41 (t) =

Example 2.28. For relatively prime positive integers p and q, the (p, q)-torus knot T (p, q) is
the image of the embedding f : S 1 → S 1 × S 1 ⊂ R3 , z 7→ (z m , z n ), where the torus S 1 × S 1 is
embedded in R3 in the standard way. As in [Hat02, Example 1.24], one can use Van Kampen’s
Theorem to deduce the following presentation for the knot group of T (p, q):s

hx, y | xp y −q i

We can now apply the algorithm to calculate the Alexander polynomial of T (p, q).

78
 1. We take the Fox derivatives by breaking off the right one-by-one to get
∂(xp y −q )
= xp−1 + xp−2 + · · · + 1,
∂x
∂(xp y −q )  
= xp −y −q − y −q+1 − · · · − y 2 − y −1 .
∂y

2. On abelianizing, it can be seen that x has linking number q with the knot and y has
linking number p, so x is sent to linking number tq and y is sent to tp , so
−tpq t−p (1−t−pq )
     
∂ri 1−tpq 1−tpq pq
α
e ∂gj = 1−tq 1−t−p
= 1−tq − 1−t
1−tp .

3. Hence, the Alexander polynomial is the generator of the principal ideal

1 − tpq 1 − tpq
 
, − .
1 − tq 1 − tp
Since p and q are relatively prime, the greatest common divisor of 1 − tp and 1 − tq is
1 − t, so the Alexander polynomial is
(1 − t)(1 − tpq )
∆T (p,q) (t) =
˙ .
(1 − tp )(1 − tq )

Now, let us see why the algorithm described above allows us to compute the Alexander
polynomial of a link. For below, L will be a link, XL will denote its exterior, G = π1 (XL ) will
denote the link group of XL , and p : XL∞ → XL will denote its infinite cyclic cover. The main
result is the following:
Proposition
  2.29. If hg1 , . . . , gn |r1 , . . . , rn−1 i is a Wirtinger presentation for G, the Jacobian
∂rj
matrix α̃ ∂gi gives us a presentation matrix for H1 (XL∞ ) after deleting any column. In
particular, the determinant of such a matrix yields ∆L (t).

Here, the map α̃ denotes the composite Zhg1 , . . . , gn i →

− ZG → − Z[t± ]. The first map is
obtained by extending the natural map hg1 , . . . , gn i → G and the second map is obtained by
extending the map π1 (X) → Z whose kernel defined the infinite cyclic cover.

Proof outline. The idea of the proof is to show that the Jacobian matrix presents a module
isomorphic to H1 (XL∞ ) ⊕ Z[t±1 ] and then to note that deleting a column from this matrix
corresponds to going from H1 (XL∞ ) ⊕ Z[t±1 ] to H1 (XL∞ ).
We give some more details. Use p : XL∞ → XL to denote the infinite cyclic cover. Fix a
point x0 ∈ XL and consider P := p−1 (x0 ) ⊂ XL∞ ; this consists of infinitely many points, one for
each element of Z. A short verification shows that the long exact sequence of the pair (XL∞ , P )
reduces to
0 → H1 (XL∞ ) → H1 (XL∞ , P ) → Z[t±1 ] → 0
Since Z[t±1 ] is free, the short exact sequence splits leading to an isomorphism

H1 (XL∞ , P ) = H1 (XL∞ ) ⊕ Z[t±1 ].

Thus as mentioned above, obtaining a presentation matrix for H1 (XL∞ , P ) will lead to a pre-
sentation matrix for H1 (XL∞ ) after removing a column.
We now describe why the Jacobian matrix gives a presentation matrix for H1 (XL∞ , P ). By
definition of homology, we have H1 (XL∞ ) = ker(∂1 )/ im(∂2 ) where the maps ∂1 and ∂2 come
from the relative chain complex
∂ ∂
C2 (XL∞ , P ) −→
2
C1 (XL∞ , P ) −→
1
C0 (XL∞ , P ).

79
Since XL and XL∞ are connected, one can deduce that C0 (XL∞ , P ) = 0. We deduce that
ker(∂1 ) = C1 (XL∞ , P ) and therefore H1 (XL∞ , P ) is presented by a matrix for ∂2 .
It remains to show that ∂2 is represented by the Jacobian matrix. To show this, it is more
convenient to work with the CW complex Y obtained from the 0-cell x0 by wedging of n one-
cells (one for each generator) by attaching 2-cells (one for each relator) along the generators.
By construction, we have π1 (XL∞ ) = π1 (Y ). Use Y ∞ → Y to denote the cover corresponding to
the subgroup of π1 (XL∞ ) = π1 (Y ) used to define XL∞ . One then deduce that

H1 (Y ∞ , P ) = H1 (Y ∞ , P ) and H1 (XL∞ ) = H1 (P ∞ ).

The advantage of the complex Y is that C1 (Y ∞ ) is now freely generator by lifts of the 1-cells
whereas C2 (Y ∞ ) is generated by lifts of the 2-cells. To understand ∂2 one looks at the boundary
of each 2-cell (this is given by the lifting to the cover the word describing the boundary of the
2-cell). A careful inspection shows that the Fox derivatives pop out.

Remark 2.30. Some care is required to show that deleting any column yields a presentation
of H1 (XL∞ ); this fact actually relies on our presentation being a Wirtinger presentation. It does
not work for an arbitrary presentation of the group!

2.5 A too brief introduction to twisted knot invariants

The upshot of this chapter is the following: the link group is a powerful invariant that is difficult
to use efficiently, while the Alexander polynomial is less difficult to apply but easier to fool.
The idea behind twisted Alexander polynomials is to define an invariant that is similar to the
Alexander polynomial but captures more information from the link group.
More concretely, given an r-component oriented link L, and a representation ρ : π1 (XL ) →
GLn (R) of the link group, we will mimick the construction of the Alexander polynomial in order
to obtain a (quotient of) polynomial(s) ∆ρL (t) which depends both on L and ρ.
Arguably the most enlightening definition ∆ρL (t) involves using twisted homology to gener-
alise the covering space approach from Subsection 2.2. Since this approach is somewhat abstract
however, we instead start by constructing ∆ρL (t) using Fox calculus exactly as in Section 2.4.
Only after learning about this first approach do we move on to the second version that involves
twisted homology. References for this section include [FV11] and [Wad94].
Fix an r-component oriented link L. As usual, π1 (XL ) denotes the group of L and we
consider once again the homomorphism α : π1 (XL ) → Z ∼
= hti, [γ] 7→ tlk(γ,K1 )+···+lk(γ,Kr ) which
we can extend linearly to obtain a ring homomorphism

e : Z[π1 (XL )] → Z[t±1 ].

α

Before introducing our first definition of the twisted Alexander polynomial, we recall how to
calculate the classical Alexander polynomomial from a Wirtinger presentation hx1 , x2 , . . . , xs |
r1 , r2 , · · · , rs−1 i of π1 (XL ) using Fox calculus. The three steps were as follows.
 
∂ri
• Consider the (s−1)×s matrix A = α̃( ∂x j
) from our presentation of π1 (XL ) by applying
Fox calculus.

• Consider the (s − 1) × (s − 1) minors of A.

• The Alexander polynomial ∆L (t) is the greatest common divisor of these minors.

We will emulate this procedure to define twisted Alexander polynomials. To start, fix a
unique factorization domain R (think of R = Z, R or C for concreteness) and a representation

80
ρ : π1 (XL ) → GLn (R). Combine α and ρ to obtain a group homomorphism ρ ⊗ α : π1 (XL ) →
GLn (R[±1 ]) by defining (ρ⊗α)(γ) := α(γ)ρ(γ) for any γ ∈ π1 (XL ). We then extend this linearly
to obtain a ring homomorphism

Φ : Z[π1 (XL )] → Mn (R[t±1 ]).

Consider the (s − 1) × s matrix M whose (i, j) component is the n × n matrix

!
∂ri
Φ ∈ Mn (R[t±1 ]).
∂xj

For 1 ≤ j ≤ s, denote by Mj the (s − 1) × (s − 1) matrix obtained from M by removing the

j-th column. Now regard Mj as a (s − 1)n × (s − 1)n matrix with coefficients in R[t, t−1 ].

Lemma 2.31. The following assertions hold:

1. det Φ(1 − xj ) 6= 0 for some j.

2. If det Φ(1 − xj ) and det Φ(1 − xk ) are non-zero Laurent polynomials, then

det Mj det Mk
=± .
det Φ(1 − xj ) det Φ(1 − xk )

In the untwisted setting, the Alexander polynomial was obtained by taking the greatest
common divisor of the size (n − 1)-minors. In this level of generality, a small additional step is
needed in order to get an invariant of (L, ρ) [Wad94].

Definition 2.32. Let L be an oriented link and let hx1 , x2 , . . . , xs | r1 , r2 , · · · , rs−1 i be a

Wirtinger presentation of π1 (XL ). Wada’s definition of the twisted Alexander polynomial of the
link L associated to a representation ρ : π1 (XL ) → GLn (R) is

det Mj
WL,ρ (t) := ,
det Φ(1 − xj )

where j is any index with det Φ(1 − xj ) 6= 0.

A priori, Wada’s definition of the twisted Alexander polynomial depends on the choice of
the presentation for π1 (XL ), but Wada shows that this is not the case [Wad94]. Indeed, Wada
generalises Definition 2.32 to include arbitrary presentations of π1 (XL ) and proves the following
result.

Theorem 2.33. Wada’s definition of the twisted Alexander polynomial WL,ρ (t) is independent
of the choice of the presentation P of π1 (XL ), up to a unit in R[t±1 ].

Two remarks about this construction: first WL,ρ is not necessarily a Laurent polynomial
(although it is so if L has r ≥ 2 components) and secondly, WL,ρ (t) is an invariant of (L, ρ), not
of L. We illustrate how Wada’s invariant generalizes the Alexander polynomial:

Example 2.34. For R = Z and ρ : π1 (XL ) → GL1 (R) the trivial representation, it immedi-
ately follows from the definitions that Wada’s invariant recovers a quotient of the Alexander
polynomial:
. ∆L (t)
WL,ρ (t) = .
1−t

81
 We now describe Wada’s invariant using a more intrinsic viewpoint, generalising our ap-
proach to the Alexander polynomial using covering spaces; recall Section 2.2. Deck transforma-
tions provide a left action by π1 (XL ) on the universal cover X
e L thus endowing C∗ (X
e L ) with the
structure of a left Z[π1 (XL )]-module structure. Given a representation ρ : π1 (XL ) → GLn (R),
we can also turn R[t±1 ]n into a right Z[π1 (XL )]-module by extending linearly the action
x · g = xρ(g)α(g) by matrices on row vectors. Therefore we can form the chain complex
R[t±1 ]n ⊗Z[π1 (X)] C∗ (X
e L ; Z) and taking homology, we get the twisted homology modules

Hi (XL ; R[t±1 ]n ) := Hi (R[t±1 ]n ⊗Z[π1 (XL )] C∗ (X

e L )).

For i = 1, one sometimes refers to H1 (XL ; R[t±1 ]n ) as the twisted Alexander module of L. Here,
note that these twisted homology are R[t±1 ]-modules. The second main definition of this section
is now the following.

Definition 2.35. Let L be an oriented link and let ρ : π1 (XL ) → GLn (R) be a representation.
The twisted Alexander polynomial ∆ρL is the order of the R[t±1 ]-module H1 (XL ; R[t±1 ]k ).

As in Example 2.34 we outline why Definition 2.35 generalises the classical Alexander poly-
nomial. For R = Z and ρ : π1 (XL ) → GL1 (R) the trivial representation, the twisted Alexander
polynomial reduces to the classical Alexander polynomial. The idea is the projection X eL → X ∞
L
can be shown to induce a chain isomorphism C∗ (X e L ; Z) ⊗ ±1 ∼ ∞
Z[π1 (XL )] Z[t ]) = C1 (XL ). It then
follows that H1 (XL ; Z[t±1 ]) = H1 (XL∞ ) and the claim follows by taking orders.
Since Definitions 2.32 and 2.35 both generalise the Alexander polynomial, it is natural to
expect them to be related. The next theorem, due to Kirk and Livingston relates Wada’s
invariant to the twisted Alexander polynomial [KL99].

Theorem 2.36. Let L be an oriented link and let ρ : π1 (XL ) → GLn (R) be a representation.
Wada invariant’s definition of the twisted Alexander polynomial, WL,ρ , is related the twisted
Alexander polynomial via the formula

. ∆ρL (t)
WL,ρ (t) = .
Ord(H0 (XL ; R[t±1 ]k ))

82
Chapter 3

Knots and 4-manifolds

The goal of this chapter is to give a hint of how knot theory (and more specifically knot
concordance) connects to 4-dimensional topology. Indeed, as we mentioned in the introduction,
one of the attractions of knot theory lies in its connections to the study of 3-manifolds and 4-
manifold. Since each of these connections could each be discussed during semester long courses,
we restrict ourselves to outlining how a subtle phenomenon in knot concordance leads to the
existence of 4-manifolds homeomorphic to R4 but not diffeomorphic to R4 .
This chapter is organised as follows. In Section 3.1, we briefly introduce handle decomposi-
tions and prove the existence of exotic R4 s modulo a difficult result of Quinn and the existence
of topologically slice knots that are not smoothly slice. In Section 3.2, we explain one method to
prove the existence of topologically slice knots that are not smoothly slice modulo two difficult
theorems: the slice-Bennequin inequality and the fact that Alexander polynomial one knots are
topologically slice.

3.1 Exotica and knot concordance

In this section, we learn about handlebody decompositions and in particular how knots can
be used to construct 4-manifolds. After illustrating these concepts with some examples, we
use them to outline the proof of why there are manifolds homeomorphic but not diffeomorphic
to R4 . The main reference for this section is [GS99].
This section makes frequent use of the notions of smooth manifolds and smooth embeddings;
see Section 1.2.1 for a refresher. In what follows, manifolds and embeddings are assumed to be
smooth unless otherwise specified.

Definition 3.1. Let 0 ≤ k ≤ n, an n-dimensional k-handle is Dk × Dn−k attached to the

boundary of an n-manifold, X, by an embedding ϕ : ∂Dk × Dn−k → ∂X.

The space we get by attaching a handle, h, with the attaching map ϕ to the space X is
denoted X ∪ϕ h. By smoothing the corners of the attachment [GS99, Page 99] we get a smooth
n-manifold. The attaching sphere of the handle is ∂Dk × {0} while the core of the handle
is Dk × {0}.

Example 3.2. We list some examples of handle attachments:

1. A 2-dimensional 1-handle D1 × D1 attached to D2 looks like a band attached to the

boundary of a disk, see Figure 3.1.

83
 Figure 3.1: A 1-handle attached to a 2-dimensional disk.

2. A 3-dimensional 1-handle D1 × D2 attached to D3 results in a solid torusa s shown in

Figure 3.2.

Figure 3.2: A 1-handle attached to a 3-dimensional disk.

In rest of the section, we will focus on the case where X is a 4-ball D4 and h is the 4-
dimensional 2-handle. Note that the attaching sphere is an embedding of ∂D2 × {0} ∼
= S 1 into
4 3
∂D = S and hence just a knot.
Remark 3.3. For the purpose of studying smooth structures on manifolds, we just need to
study handle attachments up to diffeomorphism of the resulting manifold. It is a fact that,
if ϕ and ϕ0 are two attaching maps for h that are isotopic, then X ∪ϕ h and X ∪ϕ0 h are
diffeomorphic [GS99, page 100]. Thus we only care about the embedding ϕ up to isotopy. It
turns out that an embedding S 1 × D2 ,→ ∂D4 is specified up to isotopy by two choices:

• The attaching sphere, a knot, ϕ0 : ∂D2 → ∂D4 ;

• An integer called the framing of the knot ϕ0 (∂D2 ). The curious reader should see [EHI19]
for a discussion of framings and [GS99, page 99-100] for an argument of why a knot and
a framing specify the attaching map up to isotopy. In some sense, the knot determines
the embedding of ∂D2 , but there is a degree of freedom in how ∂D2 × D2 embeds. The
framing removes this degree of freedom. For a knot, there are only integer-many framings,
and we can specify the same framing for any knot, the so-called zero-framing. We shall
focus on the need to specify a knot and gloss over the framing whenever possible.
Definition 3.4. For a knot K, the knot trace, X(K), is the 4-manifold obtained by attaching
a 0-framed 2-handle to D4 with attaching sphere K.

We leave it to the reader to prove that the knot trace of the unknot is S 2 × D2 . Instead we
now use knot traces to outline why there are manifolds homeomorphic but not diffeomorphic
to R4 . Such manifolds are known as exotic R4 .
Theorem 3.5. There is a smooth manifold homeomorphic to R4 but not diffeomorphic to R4 .

For n 6= 4, there is a unique smooth structure on Rn , so this theorem is quite astonishing.

To prove the theorem, we need the following lemma, known as the Trace Embedding Lemma:

84
 Figure 3.3: A schematic of X(K), drawn in three dimensions instead of four.

Lemma 3.6. Given a knot K, the following assertions are equivalent:

1. K is smoothly slice;

2. there is a smooth embedding X(K) ,→ R4 .

The same equivalence holds in the topological category.

Temporarily assuming this lemma, we show how the existence of topologically but not
smoothly slice knots (and a difficult result due to Quinn) gives rise to exotic R4 s.

Proof of Theorem 3.5. Pick a knot K that is topologically slice but not smoothly slice (the
existence of such knots will be proved in the next section). Since K is topologically slice, by
the trace embedding lemma (Lemma 3.6), we have a locally flat embedding f : X(K) ,→ R4 .
We then consider the non-compact topological manifold

Z := R4 \ f (X(K)).

Since X(K) is compact, Z is a non-compact topological manifold. Therefore a deep result of

Quinn [Qui82] ensures that there is a smooth structure on Z; i.e. Z is in fact a smooth manifold.
Now, consider the smooth manifold

X := Z ∪∂ X(K).

Note that X is homeomorphic to R4 , as we simply removed X(K) then put it back. We argue
that X is not diffeomorphic to R4 . If it were, we would get a smooth embedding X(K) ,→
X∼=diff R4 . However, we know from the trace embedding lemma that such an embedding does
not exist, as then K would be smoothly slice.
Summarising, X is a smooth manifold that is homeomorphic but not diffeomorphic to R4 .

We prove Lemma 3.6 using an adaptation of the the proof of [Pic20, Lemma 1.3].

85
Proof of Lemma 3.6. We prove that K being smoothly slice gives rise to a smooth embedding
X(K) ,→ R4 . Take a hyperplane H = ∼ R3 which divides R4 into upper and lower quadrants, R4 ,
+
and R− . Since K is smoothly slice, we can find a smoothly embedded disk D ⊂ R4+ which has
4

boundary ∂D = K ⊂ H; we can also assume that a tubular neighbhorhood ν(D) ∼ = D2 × D2 lies

in R4+ with ν(D) ∩ H = ν(K). Next, pick a smoothly embedded 4-ball B ∼ = R4− that contains
ν(K) in its boundary, as illustrated in Figure 3.4 This figure also illustrates the smooth 4-
manifold obtained the union
X = B ∪ νD ⊂ R4
We conclude by observing that X ∼ = X(K), since X consists of a 0-framed 2-handle νD ∼ =
D × D attached to the boundary of B ∼
2 2
= B 4 along ∂D2 × D2 ∼ = νK. Thus X is precisely
(diffeomorphic to) the knot trace of K.
The proof of the converse is of a similar level of difficulty, but we omit the details. This
concludes the proof of Lemma 3.6.

Figure 3.4: B ∪ D embedded in R4 .

3.2 Topologically but not smoothly slice knots

In this section, we prove the existence of topologically slice knots that are not smoothly slice
modulo two difficult theorems. The first is due to Freedman and asserts that Alexander poly-
nomial one knots are topologically slice [Fre84, FQ90], while the second is known as the slice-
bennequin inequality and gives a lower bound on the smooth 4-genus (recall Remark 1.69 for
the definition of the 4-genus). The main reference for this section is [Sto07]
We start by observing a relation between crossing changes and the 4-genus.

86
Proposition 3.7. If knots K and K 0 are related by a single crossing change, then

|g4sm (K 0 ) − g4sm (K)| ≤ 1.

Proof. Recall from Figure 1.28 that if a knot K and a link L are related by a band move, then
K and L bound a “pair of pants" in S 3 × [0, 1]. Furthermore, observe from Figure 3.5 that if
knots K and K 0 are related by a single crossing change, then K and K 0 are related by two band
moves.

Figure 3.5: If the knots K and K 0 are related by a single crossing change, they are related by
two band moves.

Therefore, we can construct a genus one surface Σ ⊂ S 3 × [0, 1] with boundary the disjoint
union ofK ⊂ S 3 × {0} and K 0 ⊂ S 3 × {1} by stacking two “pairs of pants" together, as in
Figure 3.6.

Figure 3.6: A genus one surface Σ ⊂ S 3 × [0, 1] bounded by K t K 0 .

Let SK ⊂ B 4 be a smoothly embedded orientable surface with boundary K and genus

It follows that Σ ∪ SK is a smoothly embedded orientable surface with boundary K 0
g4sm (K).
and genus g4sm (K) + 1 which sits inside of D4 = S 3 × [0, 1] ∪S 3 ×{0} D4 . We deduce that

g4sm (K 0 ) ≤ g4sm (K) + 1.

By a similar argument, g4sm (K) ≤ g4sm (K 0 ) + 1, so the statement of the proposition holds.

The next theorem introduces a lower bound on the smooth 4-genus, known as the slice-
Bennequin inequality [Rud93]. This lower bound is striking by its simplicity.

87
Theorem 3.8. Given an oriented knot K with oriend diagram D, we have

w(D) − s(D) + 1 ≤ 2g4sm (K),

where w(D) denotes the writhe of D, and s(D) is the number of Seifert circles.

The quantity b(D) = w(D)−s(D)+1

2 of a knot diagram D is sometimes called its Bennequin
number. Since a knot K is smoothly slice if and only if g4sm (K) = 0, we deduce that the
Bennequin number provides an obstruction to sliceness.
Next, we verify the slice-Bennequin inequality with an example.

Example 3.9. We recover the fact (that we first proved using the signature in Example 1.93)
that the right-handed trefoil is not slice. Indeed, looking at the diagram in Figure 3.7, we
see that b(T ) = w(T )−s(T
2
)+1
= 3−2+1
2 = 1, which means gsm4 (trefoil) ≥ 1. Therefore, the

right-handed trefoil knot is not slice.

Figure 3.7: The Slice-Bennequin inequality shows that the trefoil is not smoothly slice.

Using the slice-Bennequin inequality and our previous knowledge of the Alexander poly-
nomial, following [Sto07, page 172], we will now look at a more involved example and try to
understand why it is topologically slice but not smoothly slice.

Proposition 3.10. The knot K = 161335658 , pictured in Figure 3.8, is topologically slice but
not smoothly slice.

Figure 3.8: The 161335658 knot.

Proof. A tedious calculation that we omit shows that ∆K = 1. Work of Freedman then implies
that K is topologically slice [Fre84, FQ90]. It remains to prove that K is not smoothly slice. The
strategy is to find a knot K 0 with g4sm (K 0 ) ≥ 2 by changing one crossing in K: Proposition 3.8
then immediately implies that K is not smoothly slice.
To carry this out, we perform the crossing change on K that is depicted on the left hand side
of Figure 3.9. After some isotopies, the result of this crossing change is the knot K 0 = 121609
that illustrated on the right hand side of Figure 3.9.

88
Figure 3.9: A crossing change on the knot K along the pink crossing results in a knot K 0 which
we will show to have smooth 4-genus greater than 1.

We calculate the Bennequin number of the diagram D for the diagram of K 0 shown on the
right of Figure 3.9. Picking an arbitrary orientation, we calculate that w(D) = 8, s(D) = 5 as
depicted below:

The Bennequin number of D is therefore b(D) = (8 − 5 + 1)/2 = 2. Therefore the slice-

Bennequin inequality (recall Theorem 3.8) implies that g4sm (K 0 ) ≥ 2. Thus, as we already
mentioned above, Proposition 3.7 then implies that K is not slice.

89
Appendix A

A user’s guide to homology theory

Recollections on the fundamental group

In order to motivate homology, we recall some facts about the fundamental group in a slighty
unconventional way. Namely, instead of recalling the definition of π1 (X), we list some of its
properties and its values on some familiar spaces.

Remark A.1. First the intuition: the fundamental group of a path-connected space X is
denoted π1 (X) and its elements are equivalence classes of loops in X.1

Interestingly, one can calculate the fundamental group of several familair spaces once one
knows that π1 (S 1 ) = Z as well as remembers several other key properties. Since we will learn
about homology in this way (without spending too much time dwelling on definitions), we recall
some facts about the fundamental group in a similar way, to stress analogies.

Proposition A.2. For every path-connected space X there is a group π1 (X) and for ev-
ery continuous map f : X → Y , there is a group homomorphism f∗ : π1 (X) → π1 (Y ) such
that (idX )∗ = idπ1 (X) and (f ◦ g)∗ = f∗ ◦ g∗ that satisfy the following properties:

1. Value on some familiar spaces:

• if X = {∗}, then π1 (X) is trivial; we write π1 (X) = 1;

• if X = S 1 is a circle, then π1 (S 1 ) = Z.

2. Invariance under homotopy equivalences:

X ' Y ⇒ π1 (X) ∼
= π1 (Y ).

3. Calculation for unions (i.e. van Kampen’s theorem): the fundamental group of A ∪ B can
be calculated in terms of π1 (A), π1 (B), π1 (A ∩ B) and the inclusion induced maps

iA : π1 (A ∩ B) → π1 (A)
iB : π1 (A ∩ B) → π1 (B)

Namely, if we write N for the smallest normal subgroup of π1 (A) ∗ π1 (B) generated by
the iA (x)iB (x)−1 with x ∈ π1 (A ∩ B), then

π1 (A) ∗ π1 (B)
π1 (A ∪ B) = .
N
1
We will ignore basepoint considerations for simplicity.

90
Example A.3. Using these facts, we suggest the reader try to flesh out the following calcula-
tions.

1. We have π1 (Dn ) = 1 and π1 (Rn ) = 1: both spaces are homotopy equivalent to a point.

2. We have π1 (S n ) = 1 for n ≥ 2: this follows from π1 (Dn ) = 1 and van Kampen’s theorem.

3. The Moebius band and the cylinder have π1 = Z as they are homotopy equivalent to S 1 .

4. Van Kampen’s theorem gives π1 (S 1 ∨ . . . ∨ S 1 ) = Fn , the free group on n generators.

5. The genus g surface with b > 0 boundary components Σg,b ' ∨2g+b−1
i=1 S 1 has

π1 (Σg,b ) = F2g+b−1 .

6. Applying van Kampen to the genus g surface Σg = Σg,1 ∪∂ D2 with no boundary gives

π1 (Σg ) = ha1 , b1 , . . . , ag , bg | [a1 , b1 ] · · · [ab , bg ]i.

Crash course in homology

This section is an informal user’s guide to homology. We skip over the definition an explain
how to calculate homology for some familiar spaces.

Remark A.4. Again, we start with the intuition. Homology groups are defined for all topolog-
ical spaces, but in order to build intuition we will focus on manifolds. Given a manifold M and
a non-negative integer k ≥ 0, the k-th homology group of M is denoted Hk (M ), it is an abelian
group and its elements are equivalence classes of k-submanifolds of M . For the moment, we do
not dwell on said equivalence relation.
Let’s make this more concrete by considering low values of k:

• H0 (M ) is an abelian group whose elements are equivalence classes of disjoint unions of

points in M ; two points are equivalent in H0 (M ) if they are joined by a segment in M .
Addition in H0 (M ) is given by the disjoint union. For any space X, it is a fact that

H0 (X) = Z#{path-components of X} .

In particular, if X is a path-connected space, then H0 (X) = Z.

• H1 (M ) is an abelian group whose elements are equivalence classes of disjoint unions of

circles in M ; two circles are equivalent in H1 (M ) if they are the boundary of a surface
in M . Addition in H1 (M ) is given by the disjoint union. It is a fact that the the first
homology group is isomorphic to the abelianisation of the fundamental group of X:

H1 (X) = π1 (X)/[π1 (X), π1 (X)] =: π1 (X)ab .

In particular, if π1 (X) is abelian, then H1 (X) ∼

= π1 (X).

• H2 (M ) is an abelian group whose elements are equivalence classes of disjoint unions of

surfaces in M ; two surfaces are equivalent if they are the boundary of a 3-manifold in M .
Addition in H2 (M ) is given by the disjoint union.

• If M is an n-manifold, then Hi (M ) = 0 for i > n.

91
 • If M is a connected oriented n-manifold, then
(
Z if M is compact and has no boundary,
Hn (M ) =
0 if M is not compact or has a non-empty boundary.

• Homology groups are invariant under homotopy equivalences:

X ' Y ⇒ Hk (X) ∼
= Hk (Y ) for all k ≥ 0.

Example A.5. We describe the homology groups of several examples using these properties:

1. We calculate the homology groups of a point. A point X = {∗} is a connected 0-manifold,

so H0 (X) = Z and Hi (X) = 0 for i > 0. Using the invariance under homotopy equivalence,
the same assertions hold for Rn and Dn for every n ≥ 0.

2. We calculate the homology groups of a circle S 1 . The circle is connected so H0 (S 1 ) = Z

while H1 (S 1 ) = Z either because S 1 is a compact connected oriented manifold with no
boundary or because π1 (S 1 ) = Z abelianises to Z. Using the invariance under homotopy
equivalences, the same assertions hold for the Moebius band and the cylinder.

3. For the 2-sphere S 2 , we have H0 (S 2 ) = Z because S 2 is connected, H1 (S 2 ) = 0 because S 2

is simply-connected and H2 (S 2 ) = Z because S 2 is a connected compact oriented surface
with no boundary.

4. For the genus g surface Σg,b with b > 0 boundary components, we have H0 (Σg,b ) = Z
because Σg,b is connected, H1 (Σb,g ) = Z2g+b−1 because π1 (Σg ) = F2g+b−1 abelianises
to Z2g+b−1 and H2 (Σg,b ) = 0 because Σg is a connected compact oriented surface with
boundary.

5. For the genus g surface Σg (with no boundary), we have H0 (Σg ) = Z because Σg is con-
nected, H1 (Σg ) = Z2g because π1 (Σg ) = ha1 , b1 , . . . , ag , bg | [a1 , b1 ] · · · [ab , bg ]i abelianises
to Z2g ) and H2 (Σg ) = Z because Σg is a connected compact oriented surface with no
boundary.

We delay defining the homology groups some more and list more computational properties
in a way that mimicks Proposition A.2. Some of these properties were already mentioned in
Remark A.4, but we repeat them here to stress the analogy with the fundamental group.
Theorem A.6. For every space X and every k ≥ 0, there is an abelian group Hk (X) and for
every continuous map f : X → Y , there is a group homomorphism f∗ : Hn (X) → Hk (Y ) such
that (idX )∗ = idHn (X) and (f ◦ g)∗ = f∗ ◦ g∗ and that satisfy the following properties:

1. Value on points: If X = {∗}, then Hn (X) = 0 for n > 0 and H0 (X) = Z;

2. Invariance under homotopy equivalences:

X ' Y ⇒ Hk (X) ∼
= Hk (Y ) for all k ≥ 0.

3. Mayer-Vietoris sequence: If a space X decomposes as X = A ∪ B, then there is a homo-

morphism ∂ : Hn (X) → Hn−1 (A ∩ B) that fits an exact sequence2
 
(iA )∗
−(iB )∗ ( (jA )∗ (jB ))∗ ) ∂
. . . → Hn (A ∩ B) −−−−−−−→ Hn (A) ⊕ Hn (B) −−−−−−−−−→ Hn (X) −
→ Hn−1 (A ∩ B) . . . ,
2 fn+1 fn fn−1 fn−2
A sequence . . . → An → An−1 → An−2 → . . . of groups and group homomorphisms is called exact
if im(fi ) = ker(fi−1 ) for every i.

92
 where iA , iB , jA , jB are the following inclusion maps:

iA 9A jA
!
A∩B = X.
iB % jB
B

Example A.7. Using these facts, we suggest the reader try to flesh out the following calcula-
tions.

1. For the d-dimensional sphere S d (with d > 0), we claim that Hk (S d ) = Z for n = 0, d
and Hk (S d ) = 0 otherwise. Since S d is connected, we have H0 (S d ) = Z. View S d as
the union of two hemispheres (each hemisphere is homeomorphic to Dd and they are
glued along the equatorial S d−1 ). We now apply the third item of Theorem A.6 to the
decomposition S d = Dd ∪S d−1 Dd to get

. . . → Hk (S d−1 ) → Hk (Dd ) ⊕ Hk (Dd ) → Hk (S d ) → . . .

Since Dd is homotopy equivalent to a point, we have H0 (Dd ) = Z and Hi (Dd ) = 0

for i > 0. The result now follows by induction on d.

2. Assume that X is obtained as the wedge of two connected spaces A and B, meaning that
they intersect in a single point: A ∩ B = {∗}. We claim that Hk (X) = Hk (A) ⊕ Hk (B)
for n > 0. To see this, apply the third item of Theorem A.6 to obtain

. . . → Hk ({∗}) → Hk (A) ⊕ Hk (B) → Hk (X) → Hk−1 ({∗}) → . . .

The result now follows from the calculation of the homology of a point.

3. For the closed genus g surface, we use Mayer-Vietoris to give a second proof of the fact
that Hi (Σg ) = Z for i = 0, 2, H1 (Σg ) = Z2g and Hk (Σg ) = 0 otherwise; recall Exam-
ple A.5. The Mayer-Vietoris exact sequence applied to the decomposition Σg = Σg,1 ∪ D2
gives the long exact sequence

. . . → Hk (S 1 ) → Hk (Σg,1 ) ⊕ Hk (D2 ) → Hk (Σg ) → Hk−1 (S 1 ) → . . .

The case k = 0 follows by connectedness, so we focus on k = 1, 2. In that case, we get

j
0 → H2 (Σg ) → H1 (S 1 ) →
− H1 (Σg,1 ) → H1 (Σg ) → 0.

Since H1 (Σg,1 ) = Z2g by Example A.5 it remains to prove that j is the zero homomor-
phism. This can be seen geometrically by understanding how to express [∂Σg,1 ] as a linear
combination of generators of H1 (Σg,1 ).

Betti numbers and the Euler characteristic

We delay giving the definition of homology groups some more and instead mention a couple of
invariants that can be extracted from them. To this effect, we introduce the notion of the rank
of an abelian group. The most concrete way of doing this is to use the (non-trivial) fact that
every abelian group decomposes as G ∼ = F ⊕ T where F = Z ⊕ Z ⊕ . . . consists of r ∈ N ∪ {∞}
copies of Z and T is a finite abelian group. The rank of G, denoted rk(G), is then the number r
of copies of Z that appears in F . For example, we have rk(ZN ) = N and rk(Z ⊕ Z2 ) = 1

93
Definition A.8. The i-th betti number of a space X is defined as

bi (X) := rk(Hi (X))

and the Euler characteristic of a space X is defined as

X
χ(X) = (−1)i bi (X)
i≥0

whenever its Betti numbers are all finite and vanish beyond a certain index.

Both of these integers are invariant under homotopy equivalences and we suggest the reader
calculate the Euler characteristic of the spaces from Example A.5.
Example A.9. We verify that the Euler characteristic of a connected graph Γ with V edges
and E vertices is χ(Γ) = V − E. Deformation retracting along a maximal tree, Γ is homotopy
equivalent to a wedge of E −V +1 circles and therefore has Euler characteristic 1−(E −V +1) =
E−V.

Getting a sense of the definition

Homology groups can be defined for any space X using singular homology. The definition is
somewhat abstract and therefore to gain some intuition, it is much easier to work with simplicial
homology and ∆-complexes (though both definitions agree if X admits the structure of a ∆-
complex). In what follows, we closely follow [Hat02, pages 102-106].

Figure A.1: An illustration of a 1-simplex and a 2-simplex.

First we fix some terminology. An n-simplex [v0 , . . . , vn ] is the smallest convex set containing
n + 1 (ordered) vectors v0 , . . . , vn ∈ Rm such that all the v0 − vi are linearly independent.
The vi are called the vertices of the simplex and the i-th face of this simplex is the (n − 1)-
simplex [v0 , . . . , vbi , . . . , vn ] obtained by taking the smallest convex set containing all the vertices
but vi ; see Figure ?? for an illustration. For the sake of concreteness, note that any simplex is
homeomorphic to the standard n-simplex
n
X
∆n = {(t1 , . . . , tn ) ∈ Rn+1 | ti = 1 and ti ≥ 0}.
i=1

Next we introduce ∆-complexes. Informally ∆-complexes are spaces that are built inductively
by gluing simplices of higher and higher dimension. This means that we start with a set X 0 of
points (with the discrete topology) and for 1 ≤ k ≤ n, we set
G
X k = (X k−1 t ∆kα )/ ∼
α

94
where each k-simplex ∆kα is glued on using an equivalence relation x ∼ σα (x), where
the σα : ∆k → X k−1 are continous maps that satisfy a couple of conditions (that we will list
below); finally, one sets X = X n .
To make the definition more formal, we list the conditions on the gluing maps σα :

• The restriction σα |∆
˚ k is injective and each point of X is in the image of one such restriction.
˚
Here, we write ∆ := ∆k \ ∂∆k , where ∂∆k denotes the union of all the faces.
k

• The restriction of each σα to a face coincides with one of the σβ : ∆k−1 → X;

The topology on X is given by declaring A ⊂ X to be open if and only if σα−1 (A) is open in ∆n
for each α. We say that a space admits the structure of a ∆-complex if it is homeomorphic to
a ∆-complex.

Figure A.2: This figure illustrates ∆-complex structures for the circle and the torus.

Example A.10. We list a couple of examples of ∆-complexes.

1. The top of Figure A.2 describes a ∆-complex structure for the torus with one vertex and
one edge (i.e. one 1-simplex) where the endpoints of the edge are identified with the
unique vertex.

2. Figure A.2 describes a ∆-complex structure for the torus with one vertex, two 1-simplices
and two 2-simplices.

We now turn to homology. The i-th chain group Ck∆ (X) of a ∆-complex X is the free abelian
group generated by the (open) k-simplices in X; in other words, the elements of Ck∆ (X) are
formal linear combinations of the k-simplices of X. Regormulating, we set

˚ k ) not.
M M
Ck∆ (X) = Zσα (∆ = Zσα ,
α α

where the direct sum ranges over all k-simplices of X and the second equality is simply notation.
For example, as we will see in more detail in Example below, for the ∆-complex structure of
the torus described in Example A.10, we have C1 (X) = Za ⊕ Zb ⊕ Zc.
Next, for every k, we define a homomorphism

∂k : Ck∆ (X) → Ck−1

∆
(X).

˚ k ) ⊂ X, we only have to define ∂k

Since Ck (X) is freely generated by the open k-simplices σα (∆
on each such simplex and then extend linearly. On an k-simplex σα := σα (∆ ˚ k ), we define ∂k as
X
∂k (σα ) = (−1)i σα |[v0 ,...,bvi ,...,vk ] .
i

95
One checks that ∂ 2 = 0 so that (Ck∆ (X), ∂k )k≥0 forms what is called a chain complex.
Since (Ck∆ (X), ∂k )k≥0 is a chain complex, im(∂k+1 ) is a subgroup of ker(∂k ) for every k and
now the k-th homology group of X is defined as

ker(∂k )
Hk∆ (X) = .
im(∂k+1 )

Nearly all spaces we will encounter in this class admit the structure of a ∆-complex, but it is a
non-trivial fact that homology groups of a space X only depend on X and not on the choice of
the ∆-complex structure.

Example A.11. We calculate the homology groups of the circle and of the torus from this
perspective.

1. For the ∆-complex structure on the circle X described in Example A.10, we have C0 (X) =
Z, C1 = Ze and the boundary map is the zero map. It follows that Hi (X) = Z for i = 0, 1
and vanishes otherwise. This agrees with Example A.5.

2. For the ∆-complex structure on the torus X described in Example A.10, we have C0 (X) =
Zhvi while C1 (X) = Za ⊕ Zb ⊕ Zc and C2 (X) = ZU ⊕ ZL. We now have ∂1 ≡ 0
and ∂2 (U ) = a + b − c = ∂(L) so a calculation now recovers what learnt in Example A.5,
namely H0 (X) = Z = H2 (X) and H1 (X) = Z2 .

96
Appendix B

A bit of algebra

This chapter reviews a bit of algebra that we mostly used in Subsection 1.4.3 as well as 2.2
and 2.5. In Section B, we briefly review rings and modules, wheres in Section B, we review
tenso products. Since the definitions of rings and modules can be found in many textbooks
(and on wikipedia), our approach is slightly informal and emphasise ideas that we use in the
core of the notes instead of dwelling on axioms.

Rings and modules

We informally recall some notions concerning rings and modules. As we mentioned above, we
refer to algebra textbooks or the internet for detailed definitions or lists of axioms and instead
we simply recall some terminology “for the working knot theorist".
A ring R essentially consists of an abelian group (R, +) together with a multiplication ·R ×
R → R that is suitably compatible with the addition.

Remark B.1. While familiar rings include fields and the integers, an example that comes up
often in knot theory is the group ring Z[G] which consists of all finite Z-linear combinations
of elements of G. Observe that for G = Z, the ring Z[G] concides with the ring of Laurent
polynomials Z[t±1 ] but that in general the ring Z[G] is far less well behaved: it need not be an
integral domain (e.g. Z[Z2 ]) nor even commutative (e.g. Z[G] for non-abelian G).

Given a commutative ring R, an R-module M essentially consists of an abelian group (M, +)

together with an action · : R × M → M (“multiplication by a scalar") that satisfies the same
axioms as for vector spaces over fields.

Remark B.2. We collect some examples and remarks regarding modules.

• If R = F is a field, then an F-module is precisely an F-vector space, while for R = Z,

a Z-module is merely an abelian group.

• Modules over non-commutative rings can be defined similarly but it is then necessary to
keep track of whether we are talking about left multiplication or right multiplication by R.

• If a group G acts on a space X by homeomorphisms (on the left), then its homology groups
become (left) Z[G]-modules by taking the induced action on homology and extending Z-
linearly.

• A module can have both a left action by a ring R and a right action by a ring S in which
case we say M is an (R, S)-bimodule.

97
Tensor products
We briefly review tensor products of modules over a commutative ring R. We focus on giving
a couple of examples instead of listing more involved properties and alternate defiintions.

Definition B.3. The tensor product of R-modules A and B is the R-module

M
A ⊗R B := R ( ab ) / ∼
a∈A,b∈B

where the equivalence relation is defined so that (a1 + a2 , b) = (a1 , b) + (a2 , b), (a, b1 + b2 ) =
(a, b1 ) + (a, b2 ) and (ar, b) = (a, rb).

In Definition B.3, when we write a∈A,b∈B R ( ab ), we mean that the reader should take one
L

copy of R for each element of A × B and then take the direct sum of all these copies of R.

Remark B.4. We collect some remarks on Definition B.3.

• The definition of the tensor product is identical if R is not commutative provided A is a

left R-module and B is a right R-module.

• The equivalence class of (a, b) in A⊗R B is denoted a⊗b. This way, all elements of A⊗R B
are linear combinations of such elementary tensors.

• The module a∈A,b∈B R ( ab ) is sometimes referred to as the free module on A × B and is

L

denoted F (A × B).

Example B.5. Next we collect a couple of examples of tensor products:

1. Z2 ⊗Z2 Z2 = Z2 . Here F (Z2 ⊗Z2 Z2 ) = Z42 and then quotienting out gives the result.1

2. Cn ⊗ Cm = Cnm . The reasoning is similar and if (ei ) (resp. (fj )) is a basis for Cn (resp.
Cm ), then ei ⊗ fj is a basis for the tensor product Cn ⊗ Cm .

3. ZN ⊗ Q = 0. It is enough to show that all elementary tensors x ⊗ p/q vanish. For this,
note that x ⊗ p/q = N x ⊗ p/(qN ) = 0.

4. Here is an example that will be elaborated more on in Section 2.5. Given a (nice enough)
space X, each chain group of the chain complex C∗ (X) e of the universal cover of X in-
herits the structure of a left Z[π1 (X)] via deck transformations (more on this, in the
next chapter). Given a (R, Z[π1 (Mk )])-bimodule M we can then consider the chain com-
plex M ⊗Z[π1 (X) C∗ (X)
e of left R-modules whose homology is referred to as the twisted
homology of X with coefficients in M .

1
More generally, A ⊗R R ∼
= A for any R-module A.

98
Appendix C

Covering space theory

We collect a couple of facts about covering space theory. Namely, while we assume that the
reader is familiar with the first definitions and examples of covering spaces, Section C states the
correspondence between covering spaces of X and subgroups of π1 (X), while Section C collects
a couple of facts on the deck transformation group of a covering space.
Once again, our goal is to make these notions accessible to the blossoming knot theorist,
not to list all possible details and axioms. In practice, this means that we will mostly avoid
ignoring basepoints and that we fix once and for all a connected, path-connected semi-locally
simply-connected space X, which the hasty reader just read as “let X be a path-connected
manifold". We also fix a basetpoint x0 ∈ X.

The classification theorem

While we will not recall the definition of a covering space, it is helpful to keep in mind the
covering spaces of the circle S 1 given by z 7→ z n and the exponential map R → S 1 . In fact the
following classification theorem implies that those are the only covering spaces of S 1 .

Theorem C.1. Mapping a covering space p : (X, e xe0 ) → (X, x0 ) to the subgroup group p∗ (π1 (X,
e xe0 ))
of π1 (X, x0 ) gives rise to a bijection between the two following sets:

1. the set of basepoint preserving isomorphisms classes of path-connected covers of (X, x0 );

2. the set of subgroups of π1 (X, x0 ).

In this class, we will mostly be concerned with the surjectivity part of this statement:
specifying a subgroup of π1 (X, x0 ) determines a unique (up to isomorphism) covering space
of X. For example, the covering space corresponding to the subgroup Z ⊕ 0 ≤ Z2 = π1 (T 2 )
is the infinite cylinder S 1 × [0, 1]. In this class, given a link L = K1 ∪ · · · ∪ Kn with exterior
XL , we will be mostly concerned with the covering space of XL arising from the kernel of the
homomorphism π1 (XL ) → Z, γ 7→ `k(γ, K1 ) + . . . + `k(γ, Kn ); see Section 2.2.

The deck transformation group

In Sections 2.2 and 2.5, we use abundantly the fact that many coverings of a space X are acted
upon by quotients of π1 (X). To make sense of this, we briefly review the definition of the deck
transformation group of a covering space, before focusing on the case of normal covers, i.e. cover
Xe → X for which the subgroup p∗ (π1 (X))
e of π1 (X) is normal.

99
 e → X consists of all isomorphisms of X:
The deck transformation group of a cover p : X e

e = {ϕ : X
G(X) e →X
e | ϕ is a homeomorphism with p ◦ ϕ = p}.

Familiar examples include the fact that for the cover R → S 1 , t 7→ e2πit , the deck transformation
group is Z, while for the cover S 1 → S 1 , z 7→ z k , the deck transformation group is Z/kZ.
Knot theorists are mostly concerned with normal covers, and in this case, the deck trans-
formation can be explicitly described in terms of the fundamental group.
e → X is a normal cover, with group H := p∗ (π1 (X))
Proposition C.2. If X e then

e ∼
G(X) = π1 (X)/H.

Instead of describing the (very explicit) isomorphism in Proposition C.2, we mention a

quick reality check and focus on what we will use when studying knots and links. For the
reality check, observe that Proposition C.2 implies that the deck transformation group of the
cover R → S 1 , t 7→ e2πit is Z, which coincides with the fact that we mentioned before.

Remark C.3. We collect two consequences of Proposition C.2.

• If X
e → X is the universal cover of X, then G(X)
e = π1 (X).

• If ϕ : π1 (X)  G is a surjective homomorphism, then the covering space X

e corresponding
to ker(ϕ) has deck transformation group π1 (X)/ ker(ϕ) = im(ϕ) = G.

Acknowledgments
Thanks to Gaëtan Simian for pointing out several typos and inaccuracies in an ealier version of
these notes.

100
Appendix D

Bibliography

[Ada04] Colin C. Adams. The knot book. American Mathematical Society, Providence, RI,
2004. An elementary introduction to the mathematical theory of knots, Revised
reprint of the 1994 original.

[BB05] Joan S. Birman and Tara E. Brendle. Braids: a survey. In Handbook of knot theory,
pages 19–103. Elsevier B. V., Amsterdam, 2005.

[Ben] David Benbennick. Mobius_strip.

[Bir74] Joan S. Birman. Braids, links, and mapping class groups. Annals of Mathematics
Studies, No. 82. Princeton University Press, Princeton, N.J.; University of Tokyo
Press, Tokyo, 1974.

[Bro62] Morton Brown. Locally flat imbeddings of topological manifolds. Annals of Mathe-
matics, pages 331–341, 1962.

[Bul18] Lavinia Bulai. Unlinking numbers of links with crossing number 10. Involve,
11(2):335–353, 2018.

[BZ03] Gerhard Burde and Heiner Zieschang. Knots, volume 5 of De Gruyter Studies in
Mathematics. Walter de Gruyter & Co., Berlin, second edition, 2003.

[COT03] Tim D. Cochran, Kent E. Orr, and Peter Teichner. Knot concordance, Whitney
towers and L2 -signatures. Ann. of Math. (2), 157(2):433–519, 2003.

[Cro04] Peter R. Cromwell. Knots and links. Cambridge University Press, Cambridge, 2004.

[EHI19] Mohamed Elhamdadi, Mustafa Hajij, and Kyle Istvan. Framed knots. 2019.

[Fis60] Gordon M. Fisher. On the group of all homeomorphisms of a manifold. Trans.

Amer. Math. Soc., 97:193–212, 1960.

[FNOP19] Stefan Friedl, Matthias Nagel, Patrick Orson, and Mark Powell. A survey of the
foundations of four-manifold theory in the topological category. ArXiv 1910.07372,
2019.

[FQ90] Michael H. Freedman and Frank Quinn. Topology of 4-manifolds, volume 39 of

Princeton Mathematical Series. Princeton University Press, Princeton, NJ, 1990.

[Fre84] Michael H. Freedman. The disk theorem for four-dimensional manifolds. In Proceed-
ings of the International Congress of Mathematicians, Vol. 1, 2 (Warsaw, 1983),
pages 647–663, Warsaw, 1984. PWN.

101
 [FV11] Stefan Friedl and Stefano Vidussi. A survey of twisted Alexander polynomials. In
The mathematics of knots, volume 1 of Contrib. Math. Comput. Sci., pages 45–94.
Springer, Heidelberg, 2011.

[GS99] 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.

[Hat02] Allen Hatcher. Algebraic topology. Cambridge University Press, Cambridge, 2002.

[Hir12] Morris W Hirsch. Differential topology, volume 33. Springer Science & Business
Media, 2012.

[Jon85] Vaughan F. R. Jones. A polynomial invariant for knots via von Neumann algebras.
Bull. Amer. Math. Soc. (N.S.), 12(1):103–111, 1985.

[Jon87] V. F. R. Jones. Hecke algebra representations of braid groups and link polynomials.
Ann. of Math. (2), 126(2):335–388, 1987.

[Kau87] Louis H. Kauffman. State models and the Jones polynomial. Topology, 26(3):395–
407, 1987.

[Kaw96] Akio Kawauchi. A survey of knot theory. Birkhäuser Verlag, Basel, 1996. Translated
and revised from the 1990 Japanese original by the author.

[Kho00] Mikhail Khovanov. A categorification of the Jones polynomial. Duke Math. J.,
101(3):359–426, 2000.

[KL99] Paul Kirk and Charles Livingston. Twisted Alexander invariants, Reidemeister tor-
sion, and Casson-Gordon invariants. Topology, 38(3):635–661, 1999.

[KT08] Christian Kassel and Vladimir Turaev. Braid groups, volume 247 of Graduate Texts
in Mathematics. Springer, New York, 2008. With the graphical assistance of Olivier
Dodane.

[Lic97] W. B. Raymond Lickorish. An introduction to knot theory, volume 175 of Graduate

Texts in Mathematics. Springer-Verlag, New York, 1997.

[Liv93] Charles Livingston. Knot theory, volume 24 of Carus Mathematical Monographs.

Mathematical Association of America, Washington, DC, 1993.

[MA21] Powell Mark and Ray Arunima. Topological manifolds.

2021. https://maths.dur.ac.uk/users/mark.a.powell/
Topological-manifolds-lecture-notes-after-seminar.pdf.

[Mun00] James R. Munkres. Topology. Prentice Hall, Inc., Upper Saddle River, NJ, 2000.
Second edition of [ MR0464128].

[NO15] Matthias Nagel and Brendan Owens. Unlinking information from 4-manifolds. Bull.
Lond. Math. Soc., 47(6):964–979, 2015.

[OS03a] Peter Ozsváth and Zoltán Szabó. Absolutely graded Floer homologies and intersec-
tion forms for four-manifolds with boundary. Adv. Math., 173(2):179–261, 2003.

[OS03b] Peter Ozsváth and Zoltán Szabó. Knot Floer homology and the four-ball genus.
Geom. Topol., 7:615–639, 2003.

102
[OS04a] Peter Ozsváth and Zoltán Szabó. Holomorphic disks and knot invariants. Adv.
Math., 186(1):58–116, 2004.

[OS04b] Peter Ozsváth and Zoltán Szabó. Holomorphic disks and three-manifold invariants:
properties and applications. Ann. of Math. (2), 159(3):1159–1245, 2004.

[OS04c] Peter Ozsváth and Zoltán Szabó. Holomorphic disks and topological invariants for
closed three-manifolds. Ann. of Math. (2), 159(3):1027–1158, 2004.

[Pic20] Lisa Piccirillo. The Conway knot is not slice. Ann. of Math. (2), 191(2):581–591,
2020.

[Pol10] Michael Polyak. Minimal generating sets of Reidemeister moves. Quantum Topol.,
1(4):399–411, 2010.

[Qui82] Frank Quinn. Ends of maps. III. Dimensions 4 and 5. J. Differential Geometry,
17(3):503–521, 1982.

[Rob15] Justin Roberts. Knot notes. 2015. https://mathweb.ucsd.edu/~justin/

newknotes.pdf.

[Rol90] Dale Rolfsen. Knots and links, volume 7 of Mathematics Lecture Series. Publish or
Perish, Inc., Houston, TX, 1990. Corrected reprint of the 1976 original.

[Rud93] Lee Rudolph. Quasipositivity as an obstruction to sliceness. Bull. Amer. Math. Soc.
(N.S.), 29(1):51–59, 1993.

[Sto07] A. Stoimenow. Some examples related to knot sliceness. J. Pure Appl. Algebra,
210(1):161–175, 2007.

[Tu11] Loring W. Tu. An introduction to manifolds. Universitext. Springer, New York,

second edition, 2011.

[Tur94] V. G. Turaev. Quantum invariants of knots and 3-manifolds, volume 18 of De

Gruyter Studies in Mathematics. Walter de Gruyter & Co., Berlin, 1994.

[Wad94] Masaaki Wada. Twisted Alexander polynomial for finitely presentable groups. Topol-
ogy, 33(2):241–256, 1994.

103