MATH5004M: Knot Theory

Bradley Ryan
University of Leeds

School of Mathematics

Submitted in accordance with the requirements for the degree of

Master of Mathematics

May 2020
Abstract
Knots are closed curves in R3 that do not self-intersect. It is not uncommon to work almost
exclusively with so-called knot diagrams, that is the projection of a knot onto the plane. This
paper will begin with the definition of a knot from a differential geometric foundation before
considering such projections, the purpose of which is to give a rigorous groundwork to build
the theory upon in a simpler way. This will be followed by in-depth discussion of a number
of knot invariants, namely tri-colourability, the knot determinant (with reference to general
p-colourability) and skein relations generating a number of polynomials. We will then consider
the so-called knot group, from a combinatorial group-theoretic perspective, and use the theory
of surfaces from topology to determine the characteristics of manifolds with a given knot as its
boundary. The following section will focus on braid theory, in which we will see that knots and
braids are closely related. The paper then concludes with the relevant set-up and proof of the
Fáry-Milnor Theorem.

Acknowledgements
My thanks goes to Martin Speight for supervising me during this project. It is only due to the
many meetings with him that I stood a chance at understanding this interesting topic. I am also
grateful to João Faria Martins for a useful conversation on some technical linear algebra.

i
Contents
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i

1 Elementary Topology 1
1.1 Parametrised Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Topology and Isotopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Knots and Links 9

2.1 Knot Diagrams and Reidemeister Moves . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Arithmetic of Knots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

3 Knot Invariants 15
3.1 Colourability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 Alexander Polynomial . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Bracket and Jones Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

4 The Knot Group 29

4.1 The Fundamental Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.2 Combinatorial Group Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.3 The Knot Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

5 Seifert Surfaces 44
5.1 Combinatorial Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.2 Seifert Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.3 Seifert Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

6 Braid Theory 63
6.1 The Braid Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
6.2 Alexander’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

7 The Fáry-Milnor Theorem 70

7.1 The Tangent Indicatrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
7.2 Crofton’s Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

8 Summary 76

References 77

Appendices 80

ii
A The Proof of Lemma 3.16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
B The Proof of Theorem 3.21 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
C Graphs and Combinatorial Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

iii
1 Elementary Topology
We begin with an introduction to a mathematical knot, via concepts seen in differential calculus.
This will pave the way towards a beautiful result by [Mil50] which will provide us with a sufficient
condition to test if a knot can be unknotted to a circle. In practice, this section will not be referred
to all that often and theory will instead be established for so-called knot projections as mentioned
in the abstract. However, it is important to understand the calculus and topological angle now
in preparation for Sections 4 and 7. We first follow the framework of [GP94].

1.1 Parametrised Curves

Definition 1.1 A parametrised curve in Rn is a smooth map γ : I → Rn , where I ⊆ R is some
interval. It is called regular if γ ′ (t) ̸= 0, ∀t ∈ I.

Example 1.2 Consider the curve γ : R → R2 , given by γ(t) = (t2 , t5 ). This is a parametrised
curve because the functions t2 and t5 are polynomial and hence smooth. However, it is clear
that γ ′ (t) = (2t, 5t4 ) = 0 if and only if t = 0, and so it is not a regularly parametrised curve.

Definition 1.3 Let γ : I → Rn be a parametrised curve. Then, the velocity of the curve is γ ′ , the
speed of the curve is ||γ ′ || and the acceleration of the curve is γ ′′ .

Example 1.4 Consider the curve γ : [0, 1] → R2 , given by γ(t) = (t3 − 3t, t3 + t). This is a
regularly parametrised curve. Indeed, the velocity is γ ′ (t) = (3t2 − 3, 3t2 + 1), which is zero if
and only if both co-ordinates are zero, but 3t2 − 3 = 0 only if t = ±1; in this case, the second
coordinate is non-zero. Note that velocity is also a parametrised curve and that the acceleration
is γ ′′ (t) = (6t, 6t), which is zero when t = 0.

It is clear by Example 1.4 that the condition of regularity is not necessarily preserved under
taking derivatives. Furthermore, it is clear that a curve is regular if and only if its velocity, and
thus speed, is everywhere non-vanishing.

Definition 1.5 Let γ : I → Rn be a parametrised curve. The arc length along [a, b] ⊆ I is
Z b
L(γ|[a,b] ) = ||γ ′ (t)|| dt.
a

Example 1.6 Consider the curve γ : [0, π] → R2 , given by γ(t) = (cos t, sin t). Its velocity is
γ ′ (t) = (− sin t, cos t), so it has speed ||γ ′ (t)|| = 1. Thus, the arc-length along the curve is
Z π
L(γ|[0,π] ) := L(γ) = 1 dt = π.
0

Note that parametrised curves live up to their name: they are parametrised by some variable. It is

1
now possible to make precise the idea of reparametrising a curve, which has some soon-to-be-seen
useful applications.

Definition 1.7 A reparametrisation of a parametrised curve γ : I → Rn is a parametrised curve

γ̃ : J → Rn such that γ̃ = γ ◦ φ, where φ : J → I is a smooth surjection with φ′ (s) > 0, ∀s ∈ J.

It is important to have both the surjection and positive derivative assumptions in Definition 1.7
so as to exclude the possibility of taking a regular curve and reparametrising it to one that is
non-regular. This result will be proven after we first study an example of a reparametrisation.

Example 1.8 Consider the curve γ : [0, 1] → R2 , given by γ(t) = (t2 , t). An example of a
reparametrisation is the curve γ̃ : [0, 2] → R2 , given by γ̃(s) = (s2 /4, s/2). Indeed, it is clear
that γ̃ = γ ◦ φ, where φ(s) = s/2 is the so-called parameter transformation. This does satisfy
the hypotheses of Definition 1.7 because φ is surjective and φ′ (s) = 1/2 > 0.

Lemma 1.9 Suppose γ : I → Rn is a regularly parametrised curve. Every reparametrisation of γ

is also regular.

Proof : Suppose γ̃ : J → Rn is such a reparametrisation. By definition, γ̃(s) = γ(φ(s)), where

φ is the parameter transformation. Consequently, γ̃ ′ (s) = γ ′ (φ(s))φ′ (s) ̸= 0 for every s ∈ J
because γ ′ (φ(s)) ̸= 0 by regularity and φ′ (s) > 0 by assumption.

Lemma 1.10 Suppose γ : I → Rn is a regularly parametrised curve. The arc length along γ is
invariant under reparametrisation.

Proof : Suppose γ̃ : J → Rn is such a reparametrisation with parameter transformation φ. Then,

for every s0 , s1 ∈ J, it is clear that
Z s1
L(γ̃|[s0 ,s1 ] ) = ||γ̃ ′ (s)|| ds
Zs0s1
= ||γ ′ (φ(s))||φ′ (s) ds, by the Chain Rule and using that φ′ (s) > 0,
s0
Z φ−1 (t1 )
= ||γ ′ (φ(s))||φ′ (s) ds
φ−1 (t0 )
Z t1
′
= ||γ (t)|| dt, by changing variables to t = φ(s),
t0

= L(γ|[t0 ,t1 ] ),

where we define t0 = φ(s0 ) and t1 = φ(s1 ).

2
Definition 1.11 Let γ : I → Rn be a regularly parametrised curve. The curvature vector is the
image of the function k : I → Rn defined as

1 
′′ γ ′ (t) · γ ′′ (t) ′ 
k(t) = γ (t) − γ (t) .
||γ ′ (t)||2 ||γ ′ (t)||2

By construction, the curvature vector is orthogonal to velocity. Indeed, this is clear since

γ ′ (t) · γ ′′ (t) γ ′ (t) · γ ′′ (t) ′

k(t) · γ ′ (t) = − ||γ (t)||2 = 0.
||γ ′ (t)||2 ||γ ′ (t)||4

Example 1.12 Let γ : [0, 2π] → R2 be given by γ(t) = (et cos t, et sin t); this is a parametrised
curve. Then, we can compute the following:

γ ′ (t) = (et cos t − et sin t, et cos t + et sin t).

γ ′′ (t) = (−2et sin t, 2et cos t).
γ ′ (t) · γ ′′ (t) = 2e2t .
||γ ′ (t)||2 = 2e2t .

Consequently, the curvature vector k(t) = − 2e1t (cos t + sin t, cos t − sin t).

Lemma 1.13 Suppose γ : I → Rn is a regularly parametrised curve. The curvature vector

attributed to γ is invariant under reparametrisation.

Proof : Suppose k̃ : I → Rn is the curvature vector of the reparametrisation γ̃ : J → Rn . Then,

1 
′′ γ̃ ′ (s) · γ̃ ′′ (s) ′ 
k̃(s) = γ̃ (s) − γ̃ (s)
||γ̃ ′ (s)||2 ||γ̃ ′ (s)||2
1 
′′ ′2 ′ ′′ γ ′ (φ)φ′ · [γ ′′ (φ)φ′2 + γ ′ (φ)φ′′ ] ′ ′

= γ (φ)φ + γ (φ)φ − γ (φ)φ
||γ ′ (φ)||2 φ′2 ||γ ′ (φ)||2 φ′2
1 
′′ ′2 ′ ′′ [γ ′ (φ) · γ ′′ (φ)]φ′4 + ||γ ′ (φ)||2 φ′2 φ′′ ] ′ 
= γ (φ)φ + γ (φ)φ − γ (φ)
||γ ′ (φ)||2 φ′2 ||γ ′ (φ)||2 φ′2
1 
′′ ′2 γ ′ (φ) · γ ′′ (φ)φ′2 ′ 
= γ (φ)φ − γ (φ)
||γ ′ (φ)||2 φ′2 ||γ ′ (φ)||2
1 
′′ γ ′ (t) · γ ′′ (t) ′ 
= γ (t) − γ (t)
||γ ′ (t)||2 ||γ ′ (t)||2
= k(t),

where t = φ(s) and we have used φ = φ(s) to avoid a notation calamity.

Definition 1.14 Let γ : [a, b] → Rn be a regularly parametrised curve, with k : [a, b] → Rn its

3
curvature vector. The total curvature of γ is defined as
Z b
µ(γ) = ||k(t)|| ||γ ′ (t)|| dt.
a

Example 1.15 Following on from Example 1.12, we can compute the total curvature:
Z 2π
1 √
µ(γ) = √ · 2et dt = 2π.
0 2et

Lemma 1.16 Let γ : [a, b] → Rn be a regularly parametrised curve. The total curvature µ(γ) is
invariant under reparametrisation.

Proof : Suppose γ̃ : [c, d] → Rn is such a reparametrisation, with parameter transformation φ

and curvature vector k̃ : [c, d] → Rn . Then,
Z d
µ(γ̃) = ||k̃(s)|| ||γ̃ ′ (s)|| ds
c
Z d
= ||k(φ(s))|| ||γ ′ (φ(s))||φ′ (s) ds, by Lemma 1.13 and using φ′ (s) > 0,
c
Z φ−1 (b)
= ||k(φ(s))|| ||γ ′ (φ(s))||φ′ (s) ds
φ−1 (a)
Zb
= ||k(t)|| ||γ ′ (t)|| dt, by changing variables to t = φ(s),
a
= µ(γ),

where we define a = φ(c) and b = φ(d).

Definition 1.17 Let γ : [a, b] → Rn be a parametrised curve. It is called closed if γ(a) = γ(b)
and γ (k) (a) = γ (k) (b) for every k ∈ Z+ . It is called simple if γ|[a,b) is injective.

Remark 1.18 Now, Definition 1.17 allows us to view a simple closed curve as a smooth injective
map of the form f : S 1 → Rn , where S 1 is the one-dimensional sphere (or unit circle). This is a
so-called embedding of the unit circle into real space, see Definition 1.35 for a generalisation of
this concept. A number of texts on knot theory use such a function as the definition of a knot,
where the co-domain is R3 .

1.2 Topology and Isotopy

Topology is a powerful tool for the study of knots, in particular for defining what a mathematical
knot is. We first recall some fundamental ideas from topology. Note that any text introducing

4
point-set topology worth its salt will include most of the definitions and results here (and
many more) but we select only the ones most beneficial to us, proving them independently
by definition-chasing. For the sake of completeness, see [May00].

Definition 1.19 A topology on a set X is a collection τ of subsets of X satisfying these:

(i) ∅, X ∈ τ .

(ii) If U, V ∈ τ , then U ∩ V ∈ τ .
S
(iii) If {Uλ }λ∈Λ is a family of subsets where Uλ ∈ τ , for all λ ∈ Λ, then Uλ ∈ τ .
λ∈Λ

Example 1.20 Consider {1, 2, 3, 4, 5}. One possible topology we can endow upon this set is
τ = {∅, {1, 2}, {3, 4, 5}, {1, 2, 3, 4, 5}}. It is quite trivial to verify that this is a topology, given
that the only non-trivial subsets within this collection partition our superset.

Definition 1.21 Let (X, τ ) be a topological space and U ⊆ X some subset. The subset is said to
be open if U ∈ τ and the subset is said to be closed if X \ U ∈ τ .

Example 1.22 Consider the closed interval [0, 1] ⊆ R. We better expect that this be closed in
the standard Euclidean topology (the topology generated by open intervals). Indeed, this would
be the case. However, [0, 1] ⊆ [0, 1] is both open and closed in any topology we endow upon the
interval (the whole space is always both open and closed by definition).

Note that openness and closure are not opposites of each other and not every subset can be
classified as either of them, that is a subset can be open, closed, both or neither.

Definition 1.23 Let X and Y be topological spaces. A function f : X → Y is continuous if for

all U ⊆ Y open, it follows that f −1 (U ) ⊆ X is open.

Example 1.24 Consider the map π : R2 → R, defined as π(x, y) = x, where the topologies are
induced by the standard Euclidean metric (meaning that the respective topologies consist of
unions of open balls). This so-called projection map is continuous; take U ⊆ R an arbitrary open
subset and note π −1 (U ) = {(x, y) : x ∈ U } = U × R, which itself is open because U ⊆ R is open
by assumption and R ⊆ R is open by Definition 1.19.

Proposition 1.25 Let X and Y be topological spaces. A function f : X → Y is continuous if and

only if for all C ⊆ Y closed, it follows that f −1 (C) ⊆ X is closed.

Proof : (⇒) Let C ⊆ Y be closed. So, this means that f −1 (Y \ C) = X \ f −1 (C) ⊆ X is open
by continuity, because Y \ C ⊆ Y is open. Thus, f −1 (C) ⊆ X is closed by definition.

(⇐) Let U ⊆ Y be open. Then, it must be that f −1 (Y \ U ) = X \ f −1 (U ) ⊆ X is closed by

assumption, since Y \ U ⊆ Y is closed. Hence, f −1 (U ) ⊆ X is open by definition.

5
Lemma 1.26 Let X, Y , Z be topological spaces and suppose f : X → Y and g : Y → Z are
continuous. Then, f ◦ g : X → Z is continuous.

Proof : Take U ⊆ Z open. Then, by the continuity of g, it follows that g −1 (U ) ⊆ Y is open but
then, by the continuity of f , it follows that f −1 (g −1 (U )) ⊆ X is open.

Definition 1.27 Let (X, τ ) be a topological space and A ⊆ X. The subspace topology is defined
as τA = {U ∩ A : U ∈ τ }. The topological space (A, τA ) is then called a subspace.

Example 1.28 We shall consider one example but we will prove it is indeed a subspace. Let’s
consider Z ⊆ R. The subspace topology induced by the standard Euclidean topology τ will be
the so-called discrete topology, that is where every subset is open (and therefore also closed by
Definition 1.21). Indeed, any interval of the form (n − 1/2, n + 1/2) ⊆ R is open, for every n ∈ Z.
As such, we have (n − 1/2, n + 1/2) ∩ Z = {n} ∈ τZ by Definition 1.27.

Lemma 1.29 Suppose X and Y are topological spaces, A ⊆ X is a subspace and f : X → Y is

continuous. Then, the restriction f |A : A → Y is continuous.

Proof : Consider the inclusion map ιA : A → X, given by ιA (x) = x. It is enough to show that
this is continuous. Indeed, take U ⊆ X open. Then, ι−1
A (U ) = U ∩ A, which is open in the
subspace by definition. Now, f |A = f ◦ ιA and Lemma 1.26 gives the result.

Lemma 1.30 Let X and Y be topological spaces and B ⊆ Y be a subspace. If f : X → Y is

continuous and is such that f (x) ∈ B, for all x ∈ X, then fˆ : X → B is continuous.

Proof : Take U ⊆ B open. By definition, U = V ∩ B, where V ⊆ Y is open. Consequently,

fˆ−1 (U ) = fˆ−1 (V ∩ B) = fˆ−1 (V ) ∩ fˆ−1 (B) = f −1 (V ), using the assumption that f (x) ∈ B,
equivalent to x ∈ f −1 (B). Thus, f −1 (V ) = fˆ−1 (U ) ⊆ X is open by the continuity of f .

Definition 1.31 Let (X, τ ) be a topological space and ∼ be an equivalence relation on X. Then,
the quotient topology on the set of equivalence classes is τq = {U ⊆ X/∼ : q −1 (U ) ∈ τ }, where
q : X → X/∼ given by q(x) = [x] is the quotient map. We call (X/∼, τq ) a quotient space.

A special type of continuous function is the analogue of an isomorphism between topological

spaces. This will now be defined, but first the intuition: two topological spaces are the ‘same’ if
one can be continuously deformed into the other without ‘cutting’ or ‘glueing’.

Definition 1.32 Let X and Y be topological spaces. These spaces are homeomorphic if there
exists a continuous bijection f : X → Y whose inverse is continuous, denoted X ∼
= Y . Such a
function is then called a homeomorphism.

6
Example 1.33 Consider the unit circle S 1 and the interval [0, 1]. These are not homeomorphic.
However, we can define an equivalence relation between elements x, y ∈ [0, 1] as follows:

x ∼ y if and only if x = y or x, y ∈ {0, 1}.

Thus, we see that [0,1]/∼ is a so-called quotient space consisting of the equivalence classes

[0] = [1] and [z] = {z}, for all z ∈ (0, 1).

Now, we can construct a homeomorphism between [0,1]/∼ and S 1 , namely

f : [0,1]/∼ → S 1 , given by f ([x]) = (cos(2πx), sin(2πx)).

It is known that cos and sin are continuous; therefore f is continuous. As for the inverse, it is

f −1 ((y1 , y2 )) = [tan−1 (y2 /y1 )/2π].

By the continuity of tan−1 , and noting that the quotient map is itself continuous and surjective,
it must be that f −1 is a continuous inverse; this shows that f is a homeomorphism.

Lemma 1.34 Let X, Y , Z be topological spaces and suppose f : X → Y and g : Y → Z are

homeomorphisms. Then, f ◦ g : X → Z is a homeomorphism.

Proof : By definition, f −1 and g −1 are continuous and Lemma 1.26 applies, meaning both f ◦ g
and g −1 ◦ f −1 are continuous. Finally, as the composition of bijections is a bijection, we have
that f ◦ g is a continuous bijective function with continuous inverse (f ◦ g)−1 .

Definition 1.35 Suppose X and Y are topological spaces. An embedding of X into Y is a

continuous injective map f : X → Y such that f : X → f (X) is a homeomorphism.

In fact, this is a slightly weaker notion than a homeomorphism.

Example 1.36 Consider the map f : R → R2 where f (x) = (x, x3 ). This is an embedding.
Indeed, this map is continuous as it is a polynomial function and so Lemma 1.30 applies by
considering f (X) ⊆ Y as a subspace, implying that the restriction is continuous. Now, assuming
f (x) = f (y), we have that (x, x3 ) = (y, y 3 ) so comparing the first element of the two-tuples
immediately gives that x = y, hence injectivity. The restriction is certainly surjective. Lastly,
the inverse is the projection map π : f (R) → R, given by π(x, x3 ) = x. This is continuous by
Example 1.24.

Definition 1.37 Let X and Y be topological spaces where f, g : X → Y are embeddings. An

isotopy from f to g is a continuous map F : X × [0, 1] → Y where F (x, 0) = f (x), F (x, 1) = g(x)

7
and F (x, t) is an embedding for each t ∈ [0, 1].

Example 1.38 Consider f, g : R → R2 given by f (x) = (x, 0) and g(x) = (x, x3 ), with standard
Euclidean topologies. The map F : R×[0, 1] → R2 where F (x, t) = (x, tx3 ) is an isotopy. Indeed,
Example 1.36 gives that g is an embedding and it is clear that f is also. Furthermore, we can
see immediately that F (x, 0) = f (x) and F (x, 1) = g(x). Finally then, for any fixed t ∈ [0, 1],
Ft (x) = F (x, t) is a two-tuple consisting of polynomial functions, meaning it is continuous; it
is injective since Ft (x) = Ft (y) is to say (x, tx3 ) = (y, ty 3 ), thus x = y; it is surjective as
Ft : R → Ft (R) is the map Ft (x) = (x, tx3 ); its inverse is the projection map π : Ft (R) → R,
given by π(x, tx3 ) = x, continuous by Example 1.24 and Lemma 1.29. Consequently, Ft is an
embedding and thus F is an isotopy.

A stronger notion is the following, which will allow us to give a useful definition of what it means
for a closed curve to be either knotted or unknotted.

Definition 1.39 Let X and Y be topological spaces where f, g : X → Y are homeomorphisms.

An ambient isotopy from f to g is a continuous map F : X × [0, 1] → Y such that F (x, 0) = x,
F (f (x), 1) = g(x) and F (x, t) is a homeomorphism for each t ∈ [0, 1].

Proposition 1.40 Ambient isotopy defines an equivalence relation.

Proof : Throughout the proof, suppose X and Y are topological spaces with f, g, h : X → Y
homeomorphisms. First, it is clear that f is ambient isotopic to itself via F (x, t) = f (x). So, we
have reflexivity. Next, suppose f is ambient isotopic to g via F (x, t). It then follows that g is
ambient isotopic to f via G(x, t) = F1−1 (F (x, 1 − t)). Indeed, from the definition, we have

G(x, 0) = F1−1 (F (x, 1)) = F1−1 (F1 (x)) = x

and
G(g(x), 1) = F1−1 (F (g(x), 0)) = F1−1 (g(x)) = f (x),

the final part of the definition of ambient isotopy a result of Lemma 1.34. Hence, we have proven
symmetry. Finally, again suppose f is ambient isotopic to g via F (x, t) and g is ambient isotopic
to h via G(x, t). Thus, f is ambient isotopic to h by H(x, t) = G(F (x, t), t), since

H(x, 0) = G(F (x, 0), 0) = G(x, 0) = x

and
H(f (x), 1) = G(F (f (x), 1), 1) = G(g(x), 1) = h(x),

again applying Lemma 1.34 to complete the verification, which thus gives transitivity.

8
2 Knots and Links
The theory developed in Section 1 suffices to now give a rigorous definition of a knot.

Definition 2.1 A knot is an equivalence class of closed curves, where the equivalence relation is
ambient isotopy.

Definition 2.2 We say that K is knotted if there does not exist an ambient isotopy on R3 such
that K is ambient isotopic to the unit circle S 1 . Otherwise, we say that K is unknotted; we then
identify any unknot with S 1 .

We require ambient isotopy in Definition 2.1 since isotopy is not a strong enough condition; any
embedding of a knot according to our definition is always isotopic to the unknot (this follows
from a procedure called a bachelor’s unknotting whereby ‘pulling’ on a curve makes the knot
shrink to a point and vanish; this is mentioned in [Cro04]).

Remark 2.3 The literature makes a distinction between so-called tame and wild knots, they are
knots that can either be represented as a finite polygonal chain (tame) or behave pathologically
(wild). In fact, Definition 2.2 precludes wild knots since we specify that a knot is the image of
smooth map (this is how we defined a curve) and wild knots fail to have well-defined derivatives,
as noted in [Liv93].

2.1 Knot Diagrams and Reidemeister Moves

We next restrict to looking at knots from a projective point of view. There is a lot to be said
about knot projections and isotopy; we will see that it is not at all trivial to determine if a given
simple regular closed curve is knotted or unknotted.

Definition 2.4 Let K1 , ..., Kn ⊆ R3 be knots. Then, a link is the disjoint union of knots, that is
L = K1 ⊔ ... ⊔ Kn . The knots forming said link are called the components of L.

In particular, any knot is a link with a single component. It is also important to visualise knots
and links, as will be possible by the next definition; this will be immensely useful throughout
the rest of the discussion.

Definition 2.5 Let L ⊆ R3 be a link. A link diagram is a projection of L onto R2 with crossings
consisting of one continuous curve (over-crossing) and one discontinuous curve (under-crossing).

Example 2.6 Consider the diagrams of the so-called unknot, trefoil knot and Hopf link:

9
 Figure 1: Respective diagrams of the unknot, trefoil knot and Hopf link.

Definition 2.7 Let L be a diagram of a link. An orientation of L is a choice of direction along

each component of L in which the diagram is traversed.

Example 2.8 There are two possible orientations of the trefoil knot, shown in Figure 2 below.

Figure 2: The two orientations of the trefoil knot.

Definition 2.9 Let L be an oriented diagram of a link. The writhe of L is the sum of the signs of
the crossings, denoted ω(L), subject to the following conventions of Figure 3.

+1 −1

Figure 3: The signs of a link diagram’s crossings.

Example 2.10 Consider the first of the orientations for the trefoil K as in Figure 2. One can
easily compute ω(K) = −3. Notice this is true irrespective of the orientation of K.

Lemma 2.11 The writhe of a knot diagram is independent of its orientation.

Proof : Let K be a diagram of a knot with some arbitrary orientation. By considering Figure 3,
it is clear that changing said orientation preserves the sign at a crossing.

10
Remark 2.12 Of course, Lemma 2.11 will fail if we instead refer to a diagram of a link with more
than one component. An easy example is the Hopf link as in Example 2.6; it can have writhe
either 2 or −2, depending on the orientation of each component.

Lemma 2.13 Let K be the diagram of a knot and K its mirror image. Then, ω(K) = −ω(K).

Proof : By taking mirror images, it is obvious that the types of crossing in Figure 3 will interchange
roles, meaning that the sum of the signs of the crossings will be of opposite sign to ω(K), that
is ω(K) = −ω(K) as needed.

Now, we wish to develop some theory as a means of determining if two knots are ambient isotopic
without the technical construction as in Definition 1.39. It will very quickly become clear that a
vast majority of this paper is dedicated to building machinery for this very purpose. However,
the technicality we face when studying a class of curves shan’t be understated. We use the
following definition which allows us to work only with knot diagrams; this is significantly nicer.

Definition 2.14 The Reidemeister moves are the following, applied to an arc of a link:

(R0) Planar isotopy of an arc.

(R1) Twisting part of an arc.

(R2) Moving an arc in front of or behind another arc.

(R3) Moving an arc in front of or behind a crossing.

R1 R2 R3
←→ ←→ ←→

Figure 4: The Reidemeister moves from a knot diagram perspective.

Definition 2.15 Two link diagrams L and L′ are called isotopic if they can be obtained from each
other by a finite sequence of Reidemeister moves. They are called regularly isotopic if the they
are isotopic via only the second and third Reidemeister moves.

Theorem 2.16 (Reidemeister’s Theorem) Two links L and L′ are ambient isotopic if and only if
their diagrams are related by a finite sequence of Reidemeister moves.

The proof of this theorem is beyond the scope of this paper and is omitted but can be found in
[Rei27]. Nevertheless, Reidemeister’s work allows us to focus primarily on the three operations
in Definition 2.14 and see how properties are affected by each of the Reidemeister moves.

11
Definition 2.17 A knot diagram K is called achiral if it is isotopic to its mirror image K and is
called chiral if it is not isotopic to its mirror image.

Example 2.18 One of the most basic achiral knots is the figure-eight knot, which we can show
directly by using the Reidemeister moves to get from the left-handed diagram to the right-handed
diagram, and appealing to Theorem 2.16, as is done so in Figure 5.

≃ ≃

Figure 5: The ambient isotopy between mirrors of the figure-eight knot.

2.2 Arithmetic of Knots

One may ask if it be possible to join two links together in some way; it is. We will refer back to
these notions later to show the interplay between properties of two links and this arithmetic.

Definition 2.19 Let L1 and L2 be two disjoint unoriented link diagrams. The connected sum of
these diagrams, denoted L1 #L2 , is a combination of these diagrams in the following way:

(i) Find some quadrilateral region of R2 such that two of its sides are parts of arcs of each
link diagram but otherwise it is disjoint from the diagrams.

(ii) Remove the parts of the arcs that coincide with the quadrilateral and add arcs on the other
two sides of the quadrilateral.

Example 2.20 Let K1 and K2 be trefoil knots of opposite chirality. So, Figure 6 shows K1 #K2 .

12
 #
−
→

Figure 6: The connected sum of chiral-opposite trefoil knots.

Remark 2.21 The connected sum is more generally a topological operation which applies to
manifolds. In the case where we deal with knots, the diagrams can be treated as one-manifolds.
However, every knot is homeomorphic to S 1 so this more general connected sum fails to recognise
the importance of the way our knots are embedded into R3 .

We can adapt Definition 2.19 to account for the links being oriented. This situation has one
additional condition; the quadrilateral chosen in the definition must be such that the orientation
of the two sides which are parts of arcs oppose each other. We can then orientate the new arcs,
the ones we add in (ii) of Definition 2.19. This is made visual by the next example.

Example 2.22 Let K1 and K2 be oriented trefoil knots of opposite chirality. Then, Figure 7
shows K1 #K2 .

#
−
→

Figure 7: The connected sum of oriented chiral-opposite trefoil knots.

Theorem 2.23 The set of knots with the operation connected sum forms a commutative monoid.

Sketch of Proof : We must show that the connected sum is a commutative, associative binary
operation and that there is an identity element. It is intuitively clear that K# = K by
isotopy of the diagrams. This gives us the existence of the identity, namely the unknot.

For commutativity, let K1 and K2 be disjoint knot diagrams. Simply appeal to the R2 and
R3 Reidemeister moves to ‘push’ K1 along K1 #K2 until it is to the right-hand side of K2 . By
Reidemeister’s Theorem, it follows that K1 #K2 = K2 #K1 .

13
For associativity, suppose we have (K1 #K2 )#K3 . Either K3 is connected straight onto K2 , in
which the result follows, or K3 is not connected directly to K2 , in which we can proceed as with
commutativity and ‘pushing’ K3 along (K1 #K2 )#K3 until it is adjoined to K2 . Either way, it
follows that (K1 #K2 )#K3 = K1 #(K2 #K3 ).

Finally, that the connected sum of any number of knots is also a knot is intuitively clear.

14
3 Knot Invariants
Now we begin the discussion on distinguishing between different knots and their diagrams by
developing properties of knots that remain invariant under the Reidemeister moves.

Definition 3.1 A property preserved under the Reidemeister moves is an isotopy invariant.

We can now evaluate such invariants in comparison with others; as will soon be evident, some
are much stronger than others. First we prove a small result which will be useful later.

Lemma 3.2 The writhe is a regular isotopy invariant.

Proof : The result is intuitive by considering Figure 4 prescribed with any orientations. Indeed,
R2 will remove two crossings of opposite sign, so said crossings are ignored when calculating the
writhe anyway, and R3 clearly doesn’t change the writhe.

3.1 Colourability
The first property considered is that of colourability. We will use results from linear algebra to
define tri-colourability and prove that it is a knot invariant. The discussion will then be extended
to p-colourability, for p > 2 any prime.

Definition 3.3 A knot diagram is tri-colourable if it can be coloured with three colours such that
the following are satisfied:

(i) At least two colours are used.

(ii) If two colours meet at a crossing, they meet a third colour also.

Example 3.4 Consider a diagram of the trefoil knot. It is tri-colourable, as shown in Figure 8.

Figure 8: An example of a tri-colouring of the trefoil knot.

Example 3.5 The unknot is not tri-colourable as there is one arc so only one colour is used.

Theorem 3.6 Tri-colourability is an isotopy invariant.

15
Proof : We will simply demonstrate that for a given tri-coloured diagram of a knot, K say, the
Reidemeister moves R1, R2, R3 preserve the tri-colourability conditions of the diagram. For
R1, note that an arc of K can only have one colour and inserting a twist introduces a crossing on
that arc, but there is only one colour at that crossing, so the second condition of Definition 3.3
is satisfied. If K is tri-colourable, then the first condition of Definition 3.3 holds automatically.
Hence, R1 preserves tri-colourability. For R2 and R3, there are two and four cases to consider,
respectively. This is done by simply applying the moves locally to the pre-tri-coloured knot K
and observing that tri-colourability still holds, as in Figure 9, which completes the proof.

R2 R2
←→ ←→

R3 R3
←→ ←→

R3 R3
←→ ←→

Figure 9: Tri-colourability being preserved under the Reidemeister moves.

Corollary 3.7 If a knot is tri-colourable, then it is not ambient isotopic to the unknot.

Note that Theorem 3.6 is somewhat useful for seeing if a knot is ambient isotopic to the unknot.
However, there are a few immediate shortcomings. Firstly, some knots are not tri-colourable but
are still not ambient isotopic to the unknot (the figure-eight knot for example). Secondly, two
knot diagrams being tri-colourable does not mean they are ambient isotopic (the trefoil and its
mirror image, for example). However, we can improve on this notion slightly. In order to extend
colourability of a knot diagram, as hinted at in [GP94], to a more general setting, define the set
of n colours used in a colouring of a knot diagram to be Zn . We re-state the second condition of
Definition 3.3. Indeed, let x, y, z ∈ Z3 where z is the colour of an over-crossing. Then, we have

16
 x + y − 2z ≡ 0 (mod 3).

The generalisation now follows, and underpins a very surprising numerical isotopy invariant.

Definition 3.8 Let p > 2 be prime. A knot diagram is p-colourable if it can be coloured with p
colours such that the following are satisfied:

(i) At least two colours are used.

(ii) For x, y, z ∈ Zp , at any crossing with over-crossing z and under-crossings x and y, it must
be that x + y − 2z ≡ 0 (mod p).

Remark 3.9 We can solve an exercise in [Bos19] to see that no knot can be 2-colourable. Indeed,
if p = 2 was allowed in Definition 3.8, it would mean that x + y − 2z ≡ 0 (mod 2), which is to
say x ≡ y (mod 2). But then, we have x ≡ z (mod 2). Here we have considered an arbitrary
crossing, so only one colour is used; the first part of Definition 3.8 fails.

We can appeal to linear algebra to better understand the concept of p-colourability and the
(somewhat limited) power it has. For that, we must define and prove a number of things.

Definition 3.10 Let K be a diagram of a knot with n crossings. A colouring matrix MK of K is

an n × n matrix with rows and columns representing the crossing and arcs of K, respectively,
where an entry of 2 denotes an over-crossing, an entry of −1 denotes an under-crossing and an
entry of 0 denotes exclusion from a crossing.

Remark 3.11 There is a particular case to discuss, that is when an arc crosses itself as in the
R1 move. Here, there is seemingly no well-defined notion of an over-crossing and under-crossing
since one arc acts as both. As such, we assign an entry of 1 for such an occurrence.

Example 3.12 Consider the the trefoil knot as in Figure 1, with arcs labelled x1 , x2 , x3 and
crossings labelled 1, 2, 3, anti-clockwise, with arc xn leading to crossing n. The colouring matrix
with respect to these conventions is
 
−1 −1 2
MK =  2 −1 −1.
 

−1 2 −1

Remark 3.13 The colouring matrix can be viewed as a matrix of coefficients for a system of
n equations, the so-called colouring equations. As is evident from Example 3.12, the rows and
columns of the colouring matrix will sum to zero, meaning the rows and columns are linearly
dependent as vectors in R3 ; linear algebra tells us that det(MK ) = 0 as a result. Sadly then,

17
this fails to be any form of invariant, as one may initially hope.

Definition 3.14 Let K be a diagram of a knot. The knot determinant det(K) of K is defined as
the absolute value of any minor of the matrix MK . The unknot has determinant 1 by convention.

Example 3.15 Continuing on from Example 3.12, we can consider the minor of MK formed by
removing the first row and second column, which gives us
!
−1 −1
det(K) = det = 3.
2 −1

In fact, it is rather easy to see that all minors of the colouring matrix MK coincide up to absolute
value. We must be sure that this is always true and so this is what we demonstrate next.

Lemma 3.16 The knot determinant is well-defined.

Proof : We actually prove a stronger result as in [Cam08], namely if some n × n matrix M has
the property that the entries along every row sum to zero, and the entries down every column
sum to zero, then the absolute value of every minor is equal. Suppose that U is an n × n matrix
consisting only of ones. We will compute det(M + U ) in terms of an arbitrary minor Mij of M .
To do this, we will appeal to classical row operations as in linear algebra. If mij represents the
entries of M , then
 
m11 + 1 · · · m1j + 1 · · · m1n + 1
 .. .. .. 

 . . . 

M + U =  mi1 + 1 · · · mij + 1 · · · min + 1 .
 
.. .. ..
 
. . .
 
 
mn1 + 1 · · · mnj + 1 · · · mnn + 1

We assume that the rows and the columns of M each sum to zero, that is
n
X n
X
mij = 0, for fixed i, and mij = 0, for fixed j,
j=1 i=1

respectively. We now proceed with elementary row and column operations on M + U which
transforms it into a matrix which is identical apart from the j th column being all-zero except
for an n on the ith row down. See Appendix A for the explicit computations. If we call
 
the resulting matrix M , we know det(M + U ) = n det M , the factor of n a result of the
f f
 
elementary operations we used. But now, det Mf = n(−1)i+j det(Mij ), so det(M + U ) =
n2 (−1)i+j det(Mij ). It must be that det(Mij ) is independent of the choice of i and j up to
sign.

18
We are getting closer to being able to prove that the knot determinant is an isotopy invariant,
which is surprising given that it is relatively simple to define and it is numerical. Before we do
so, we shall show that there is a connection between knot determinants and the colourability of
the corresponding knot.

Proposition 3.17 Let K be a diagram of a knot. K is p-colourable if and only if p | det(K).

Proof : Now, K is p-colourable if and only if the system of colouring equations has a solution,
which occurs if and only if there exists v ∈ Znp where v ̸= λ1, for any λ ∈ Z, such that MK v = 0.
By a classical result of linear algebra, it follows that rank(MK ) < n, since det(MK ) = 0. Every
solution of this system will lie in ker(MK ) yet we demand a non-zero solution, so null(MK ) > 0
which implies that rank(MK ) < n−1, by the Rank-Nullity Theorem. This is equivalent to saying
that the determinant of each (n − 1) × (n − 1) minor vanishes modulo p by a result of [MZ19].
This is precisely the condition det(K) ≡ 0 (mod p) and so p | det(K).

Example 3.18 Appealing to Proposition 3.17 in the context of Example 3.15, it is clear that the
trefoil is tri-colourable but not (5, 7, 11, ...)-colourable; it is only tri-colourable.

The consequence of Example 3.18 shouldn’t be so surprising; we can convince ourselves that it
is not possible to colour the trefoil for p ≥ 5 by looking at the usual diagram. However, this
triviality is not always the case, as we will now see.

Example 3.19 Consider the following labelled diagram of the so-called endless knot:

x3
x6 5 3 x1

1
2 7
x5

6 4
x2 x4
x7
Figure 10: Labelling for the colouring matrix of the endless knot.

19
This gives rise to the colouring matrix
 
−1 −1 0 0 2 0 0
0 −1 −1 00 0 2
 

 

 2 0 −1 −1 0 0 0
MK = 0 0 0 −1 −1 0 2 .
 
 

 0 0 2 0 −1 −1 0 

0 2 0 0 0 −1 −1
 

−1 0 0 2 0 0 −1

We need only calculate the determinant of a minor and take its absolute value. For the sake of
time, we will state the result: det(K) = 15. Accordingly, the endless knot is both 3-colourable
and 5-colourable, as a result of Proposition 3.17.

We next state and prove another linear algebra result. It will provide us with the means of
justifying why the knot determinant is invariant under the Reidemeister moves.

Lemma 3.20 Let M be an n × n matrix. Adjoining an all-zero row and an all-zero column except
for a one at their intersection, occurring on the leading diagonal, will not change det(M ).

Proof : Consider the determinant of M as an expansion by minors on the ith row, that is
n
X
det(M ) = (−1)i+j mij det(Mij ),
j=1

where mij is an entry of M and Mij is the minor formed by removing the ith row and j th column.
If M ′ is this extended matrix with the ith row being all zero except for one on the diagonal, then

n+1
X
′
(−1)i+j m′ij det Mij′ = det Mii′ ,




det M =
j=1

as the only position in which there is something non-zero is when j = i. Moreover, it follows
that det(Mii′ ) = det(M ) and this completes the proof.

Theorem 3.21 The knot determinant is an isotopy invariant.

Proof : It suffices to show that the Reidemeister moves preserve the knot determinant. Let K
be the knot on which such manipulations are performed. For R1 and R2, let MK be the knot’s
colouring matrix and KR1 and KR2 be the new diagrams formed by performing each of these
moves. We claim that via the use of elementary row and column operations, the new matrices

20
will take the following form:
 
  0 0
0  .. .. 
..   M . .
 K
MK .
  
MKR1 ∼ and MKR2 ∼ .
 
 0 0
0
   
 0 · · · 0 1 0 
0 ··· 0 1

0 ··· 0 0 1

The full detail is given in Appendix B. As a result of Lemma 3.20, the determinant and nullity
is unchanged; we know that the nullity of the original matrix is the dimension of the space of
solutions to MK v = 0 and this is clearly preserved by the above forms of each colouring matrix.
It follows from this that det(KR1 ) = det(K) = det(KR2 ). Lastly, it is clear that an R3 move on
a diagram K, resulting in the diagram KR3 , wouldn’t change the knot determinant since MKR3
is also an n × n matrix, obtainable from MK by interchanging rows and columns.

Example 3.22 We again consider our favourite trefoil and an equivalent diagram in Figure 11.
x1

x1 2 1

2 1
5
≃
x5
4
x2 3 x3 x2 x3
x4
3

Figure 11: Applying the second Reidemeister move to the trefoil.

We have computed MK in Example 3.15. It remains to compute MK ′ and in doing so we achieve

   
−1 −1 2 0 0 −1 −1 2 0 0
 2 −1 −1 0 0  2 −1 −1 0 0
   
  φ  
MK ′ =
0  7−
2 −1 −1 0  →−1 2 −1 0 0,

0 2 0 −1 −1 0 0 0 1 0
   

−1 2 0 0 −1 0 0 0 0 1

where φ represents the below sequence of row and column operations:

R3 7→ R3 + R5 , R3 7→ R3 − R4 , R4 7→ R4 − R5 ,

C1 7→ C1 + C4 − C5 , C2 7→ C2 + 2C5 , C4 7→ −C4 , C5 7→ −C5 .

21
This example demonstrates the property we claimed in the proof of Theorem 3.21.

We now make a small deviation. It was proven in [Deh87] that the trefoil is chiral. There are a
number of complicated arguments but we shall give a relatively modern result from which this
follows rather quickly. Note however that the proof of this result does include some higher-level
concepts but the base statement is rather accessible.

Proposition 3.23 Let p ≡ 3 (mod 4) be a prime number. If K is an achiral knot, then either

p ̸ | det(K) or p2 | det(K).

Proof : The proof relies on so-called linking forms and homology. It is presented in [FMP17].

Corollary 3.24 The trefoil is chiral.

Proof : Recall from Example 3.15 that for the trefoil K, we have det(K) = 3. As such, we use
Proposition 3.23 with p = 3 and see that p | det(K) and p2 ̸ | det(K). As such, the contrapositive
implies that K is not achiral, i.e. the trefoil is chiral.

3.2 Alexander Polynomial

We use the work of the previous subsection to develop the Alexander polynomial in an intuitive
manner. Alternatively, we could approach this from an algebraic angle in which we consider
rings, ideals and the Wirtinger presentation as in Section 4 to derive the Alexander polynomial
from a more general background. For simplicity, we shall state some interesting results but will
not prove them yet as they require alternate discussion.

Definition 3.25 Let K be an oriented diagram of a knot. The Alexander polynomial ∆K (t) of K
is the knot determinant of K whereby the colouring matrix uses the conventions of Figure 12.

t 1−t t 1−t

−1 −1

Figure 12: The colouring conventions for computing the Alexander polynomial.

Corollary 3.26 (of Theorem 3.21) The Alexander polynomial is an isotopy invariant.

Example 3.27 Consider the first oriented trefoil knot in Figure 2. We can easily compute the

22
colouring matrix with the conventions of Figure 12:
 
t −1
1−t
MK = 1 − t t −1 .
 

−1 1 − t t

Hence, we can now compute the Alexander polynomial, namely ∆K (t) = t2 − t + 1.

Lemma 3.28 Let K be an oriented diagram of a knot. Then, |∆K (−1)| = det(K).

Proof : Clearly, taking t = −1 in Figure 12 recovers Definition 3.10 from which the result follows.
Note that we include the absolute value in the statement of the result because of a non-trivial
multiplicative property of Alexander polynomials which is next discussed.

We now appeal to the work of [Bad16] to understand the properties of the Alexander polynomial.

Proposition 3.29 Let K be an oriented diagram of a knot and L an oriented diagram of a link.
Then, we have the properties listed below.

(i) The Alexander polynomial is symmetric modulo ±tn , that is ∆L (t) = ±tn ∆L (t−1 ).

(ii) The Alexander polynomial satisfies ∆K (1) = 1, possibly by using (i).

(iii) The Alexander polynomial satisfies ∆K (t) = ∆K (t).

Proof : (of (iii) only) Consider the mirror images of the labelling convention in Figure 12 and note
that the two are interchanged when reflecting across some line. This corresponds to elementary
row and column operations on MK which do not change the determinant.

As such, the Alexander polynomial cannot be used to detect whether or not a knot is achiral.

Remark 3.30 We mentioned in the proof of Lemma 3.28 that we require an absolute value; this
is because of part (i) of Proposition 3.29. Indeed, we computed the Alexander polynomial of the
trefoil in Example 3.27 and we see that ∆K (−1) = 3, agreeing with Example 3.15 even in sign.
However, suppose we had computed det(K) by considering a different minor, one that gives us
∆K (t) = −t2 + t − 1, for example. In this case, we see that ∆K (−1) = −3.

3.3 Bracket and Jones Polynomials

There are a number of ways to assign a polynomial to a knot or link, one of the simplest being that
of the bracket polynomial. Before we discuss the polynomial and its use, we will first construct
it from an operation which acts on the crossings of a knot. Note that such a construction is
possible for other polynomials as well, as will be seen.

23
Definition 3.31 Let L be a diagram of a link. A skein relation is an operation which establishes
correspondence between the given diagram of L and the diagram of another link L′ which differs
from L at only one crossing.

Example 3.32 Consider the following, where L is a link diagram and A, B, C are indeterminate
variables:

1. ⟨ ⟩ = 1.

2. ⟨ ⊔ L⟩ = C⟨L⟩.

3. ⟨ ⟩ = A⟨ ⟩ + B⟨ ⟩.

The first tells us that whenever we have the unknot under this skein relation, it evaluates to one
and the second tells us that whenever we have the disjoint union of the unknot with a link, we
can almost disregard the unknot, but with it comes a factor of the indeterminate C. The third
(a skein relation) provides a means of removing a crossing of the form written on the left from
a diagram of a link. What if we wanted to impose that this skein relation be invariant under
regular isotopy? Well, this is precisely to force that
D E D E
= and = .

Working with the second Reidemeister move, we can use the skein relation to see that

=A +B

= A2 + AB + AB + B2

= (A2 + ABC + B 2 ) + AB

= if and only if AB = 1 and A2 + ABC + B 2 = 0,

which has solution B = A−1 and C = −A2 − A−2 . It turns out that this solution coincides with
that when solving for the third Reidemeister move and thus we have enforced invariance under
regular isotopy.

Definition 3.33 The bracket polynomial of a link diagram L is a polynomial given as follows:

1. ⟨ ⟩ = 1.

2. ⟨ ⊔ L⟩ = (−A2 − A−2 )⟨L⟩.

3. ⟨ ⟩ = A⟨ ⟩ + A−1 ⟨ ⟩.

24
Lemma 3.34 The bracket polynomial is invariant under regular isotopy but not isotopy.

Proof : By construction, it is invariant under R2, so we need only show invariance under R3.
Indeed, we calculate directly from Definition 3.33 and see that

=A + A−1 and =A + A−1 ,

where we use the invariance under R2 in the first bracket on the right side of the above equations.
Clearly, these objects are equal and thus it is invariant under R3 and thus regular isotopy. Finally,
we expect that R1 fails. Indeed, this is clear from direct computation:

=A + A−1 = −A−3 ̸=

and thus the result is proven.

Note that for a link diagram with k crossings, there are 2k ways of removing the crossings under
the skein relation of the bracket polynomial, which results in the unlink on k-components. In
fact, we have this formula by applying the first and second rules of the bracket polynomial:

⟨⊔kn=1 ⟩ = ⟨⊔k−1
n=1 ⊔ ⟩ = (−A2 − A−2 )⟨⊔k−1
n=1 ⟩ = · · · = (−A2 − A−2 )k−1 .

Definition 3.35 Let L be an oriented diagram of a link. The normalised bracket polynomial is

fL (A) = (−A)−3ω(L) ⟨L⟩,

where ⟨L⟩ is the usual bracket polynomial, ω(L) is the writhe and A is an indeterminate variable.

Theorem 3.36 The normalised bracket polynomial fL (A) is an isotopy invariant.

Proof : The fact that fL (A) is invariant under R2 and R3 follows from Lemmata 3.2 and 3.34.
Suppose that L′ is an equivalent diagram of L, except we obtain L by applying R1 to L′ , that
is L′ has a twist on one of its arcs. Then, we can see that

ω(L′ ) = ω(L) ± 1 and ⟨L′ ⟩ = (−A)∓3 ⟨L⟩,

with sign depending on orientation. So, the definition of the normalised bracket polynomial gives

fL′ (A) = (−A)−3ω(L)±3 ⟨L′ ⟩

= (−A)−3ω(L) (−A)±3 (−A)∓3 ⟨L⟩

25
 = (−A)−3ω(L) ⟨L⟩
= fL (A).

and thus it is invariant under R1 and so the result is proven.

Example 3.37 Consider the diagram of the trefoil in Figure 2. It was Example 2.10 which showed
us that the writhe of the trefoil is ω(K) = −3. It remains to compute the corresponding bracket
polynomial. We shall demonstrate this now:

=A + A−1

= A2 + + A−1

= A3 + 2A + A−1 + + A−2

= A3 + 3A + 3A−1 + A−3

= A3 (−A2 − A−2 )2 + 3A(−A2 − A−2 ) + 3A−1 + A−3 (−A2 − A−2 )

= A7 − A3 − A−5 .

Finally then, it follows that fK (A) = (−A)9 (A7 − A3 − A−5 ) = −A16 + A12 + A4 .

Remark 3.38 If we proceed as in Example 3.37 with the diagram of the trefoil of opposite
chirality, that is its mirror image, but not opposite orientation, it is natural to ask the question
what would happen?. Indeed, if K is the mirror image of the diagram used in the above example,
we would see that
fK (A) = −A−16 + A−12 + A−4 .

We soon discuss the connection between the normalised bracket polynomial and mirror images.

Definition 3.39 Let L be an oriented diagram of a link. The Jones polynomial of L is

VL (t) = fL (t−1/4 ).

The advantage of this definition of the Jones polynomial is that it is relatively simple and we will
soon see some immediate consequences. However, another definition will come from the study of

26
braid theory, as in Section 6, which has its own set of advantages, including a nice generalisation.

L+ L− L0

Figure 13: Types of crossing in oriented link diagrams, as in Proposition 3.40.

Proposition 3.40 The Jones polynomial of the oriented diagram of link L satisfies these:

(i) V (t) = 1, where is an oriented unknot.

(ii) t−1 VL+ (t) − tVL− (t) = (t1/2 − t−1/2 )VL0 (t), where L+ , L− , L0 are as in Figure 13.

Proof : It is clear that

f (A) = (−A)0 ⟨ ⟩ = 1,

which means the first formula in the statement of the proposition is recovered. As for the second,
we will apply the definition of the bracket polynomial to the diagrams L+ and L− and substitute
in the relevant variable. To begin, we see that

fL+ (A) = −A−2 − A−4 and fL− (A) = −A4 − A2 .

Therefore, by making the substitution as in Definition 3.39, we see that

t−1 VL+ (t) − tVL− (t) = −t−1/2 − + + t1/2

= (t1/2 − t−1/2 )VL0 (t),

by definition. Thus, we have the second formula and the result is proven.

Example 3.41 Continuing on from Example 3.37, we see the Jones polynomial for the second
trefoil diagram K as in Figure 2 is VK (t) = −t4 + t3 + t.

Lemma 3.42 Let K be a diagram of an oriented knot and K

e its reverse. Then, V e (t) = VK (t).
K

Proof : It was Lemma 2.11 that established well-definedness for the writhe of an unoriented
knot, since it is unchanged by reversing orientation. Further, the bracket polynomial ignores

27
 e = ω(K) and ⟨K⟩
orientation, so it is clear that ω(K) e = ⟨K⟩, which means that f e (A) = fK (A)
K
and as a consequence, we have VKe (t) = VK (t).

We end the discussion of these two polynomials with an immediate consequence of some earlier
work. We will then briefly evaluate the invariants seen thus far.

Corollary 3.43 (of Theorem 3.36) The Jones polynomial is an isotopy invariant.

Of course, the bracket polynomial is not an isotopy invariant as was shown in Lemma 3.34 and
thus its use is somewhat limited to developing intuition for skein relations and providing the
basis of Definition 3.39. The benefit of distinguishing between Definitions 3.35 and 3.39 is that
the Jones polynomial is, by design, a Laurent polynomial with integer coefficients in t1/2 , that
is it lives in the ring Z[t1/2 , t−1/2 ]. This leads to generalisations that sadly we will not consider
in this paper.

We conclude with an interesting result from the theory we have developed, that being a (more
standard) sufficient test for chirality.

Lemma 3.44 Let K be a diagram of an oriented knot and K its mirror. Then, VK (t) = VK (t−1 ).

Proof : It was Lemma 2.13 that shows diagrams of opposing chirality have writhes of opposite
sign, that is ω(K) = −ω(K). However, since the bracket polynomial ignores orientation, ⟨K⟩ =
⟨K⟩, it follows that fK (A) = fK (A−1 ) and, as a consequence, VK (t) = VK (t−1 ).

Corollary 3.45 If K is achiral, then VK (t) = VK (t−1 ).

Example 3.46 It was Example 2.18 that established the figure-eight knot was achiral. We will
state but not prove that the Jones polynomial of this knot is VK (t) = t2 − t + 1 − t−1 + t−2 , by
way of [Jon14]. We see that VK (t) = VK (t−1 ), which demonstrates Corollary 3.45.

Example 3.47 We can use the Jones polynomial, specifically the contrapositive of Corollary 3.45
in conjunction with Example 3.41, to once again recover that the trefoil is chiral. Indeed, if K is
the trefoil knot, we know that VK (t) = −t4 + t3 + t and we know that VK (t) = −t−4 + t−3 + t−1
from Lemma 3.44. Therefore, VK (t) ̸= VK (t−1 ) implies that K is chiral.

This is better, in some senses, than the Alexander polynomial, since it can detect some chirality,
but sadly the previous result isn’t both necessary and sufficient; it may be that the mirror images
of a diagram share the same Jones polynomial yet they are not isotopic. More generally, we
may have two knots whose Jones polynomials coincide yet they are non-isotopic (the real killer
comes in [Kan86], where we see an example of two knots whose Jones polynomials agree but are
distinguished by the Alexander polynomial).

28
4 The Knot Group
Next, we concern ourselves with a more technical study of topological spaces. Indeed, we will
develop the idea of the so-called fundamental group as to convincingly discuss the titular topic.
We shall also consider an overview of free groups, generators and the Wirtinger presentation.

4.1 The Fundamental Group

The first path on the journey to defining the knot group is to understand a very important
topological tool known as the fundamental group. This is now introduced, with a few useful
results proven along the way.

Definition 4.1 Let X and Y be topological spaces and f, g : X → Y some maps. It is said that f
and g are homotopic if there exists a continuous map F : X × [0, 1] → Y such that the following
conditions hold:

(i) F (x, 0) = f (x).

(ii) F (x, 1) = g(x).

In this case, we write f ≃ g to denote the fact these maps are homotopic.

Lemma 4.2 Homotopy is an equivalence relation.

Proof : Throughout, suppose that X and Y are topological spaces and f, g, h : X → Y are maps.
It is clear that f ≃ f via F (x, t) = x. Next, assuming that f ≃ g via F , we also have g ≃ f via
G(x, t) = F (x, 1 − t). Finally, if f ≃ g and g ≃ h via F and G respectively, then f ≃ h via

F (x, 2t), t ∈ [0, 1/2]
H(x, t) = ,
G(x, 2t − 1), t ∈ [1/2, 1]

which completes the proof.

Definition 4.3 Let X and Y be topological spaces. They are called homotopically equivalent if,
for any f : X → Y , there exists g : Y → X such that g ◦ f ≃ idX and f ◦ g ≃ idY .

Example 4.4 We solve an exercise in [GM12]. Let R2 be endowed with the standard Euclidean
topology and consider S 1 and R2 \ {(0, 0)} as subspaces. Suppose that f : S 1 → R2 \ {(0, 0)}
is the inclusion map and that g : R2 \ {(0, 0)} → S 1 is given by g(x) = x/||x||. Then, these
subspaces are homotopically equivalent. Indeed, we verify explicit homotopies below:

ˆ g ◦ f ≃ idS 1 via F (x, t) = x. Indeed, F (x, 0) = x = idS 1 and F (x, 1) = x = (g ◦ f )(x); as a

matter of fact, this is exactly idS 1 since ||x|| = 1, ∀x ∈ S 1 .

29
 ˆ f ◦ g ≃ idR2 \{(0,0)} via G(x, t) = tx/||x|| + (1 − t)x. Indeed, G(x, 0) = x = idR2 \{(0,0)} and
G(x, 1) = x/||x|| = (f ◦ g)(x).

Note that Definition 4.1 is very similar to that of Definition 1.37, that is the definition of isotopy.
In fact, a homotopy can be thought of as a continuous deformation from one map to another
without the embedding condition that we have for an isotopy. This is detailed in the next
example.

Example 4.5 Consider the interval [−3, 3] ⊆ R as a subspace and maps f, g : [−3, 3] → R where
f (x) = x and g(x) = −x. It is clear that these maps are homotopic via F (x, t) = (1 − 2t)x but
they are not isotopic because there is no such homotopy that is injective for all t. Take the one
we stated here; at t = 1/2, the map is the zero-map F (x, t) = 0 which is certainly not injective.

Definition 4.6 Let X be a topological space. A path is a continuous map α : [0, 1] → X, where
[0, 1] has the subspace topology induced by the standard Euclidean topology on R. A path is
called a loop based at p if α(0) = p = α(1).

Definition 4.7 Let X be a topological spaces and α, β : [0, 1] → X some paths with endpoints x
and y. We call α and β path-homotopic if there is a continuous map F : [0, 1] × [0, 1] → X such
that the following conditions hold for all s ∈ [0, 1] and t ∈ [0, 1]:

(i) F (s, 0) = α(s). (iii) F (0, t) = x.

(ii) F (s, 1) = β(s). (iv) F (1, t) = y.

In this case, we write α ≃ β to denote the fact these paths are path-homotopic.

Example 4.8 Let α, β : [0, 1] → Rn be any paths with end points x and y. Then, they are
path-homotopic via the so-called straight-line homotopy, that is F (s, t) = (1 − t)α(s) + tβ(s).

Lemma 4.9 Path-homotopy is an equivalence relation.

Sketch of Proof : One can almost repeat the argument of the proof of Lemma 4.2 in the context
of paths, taking care to verify the end-point conditions.

Definition 4.10 Let X be a topological space. It is called path-connected if for every x, y ∈ X,

there exists a path α : [0, 1] → X from x to y.

Example 4.11 The space Rn with the standard Euclidean topology is path-connected. Indeed,
let x, y ∈ Rn . Then, α(s) = (1 − s)x + sy is a path from x to y. On the other hand, the
space {(0, 0)} ∪ {(x, sin(1/x)) : x ∈ (0, 1]} with topology induced from the Euclidean topology,
commonly known as the topologist’s sine curve, is famously not path-connected, since there is
no way to continuously link the sine curve part of the space to the origin (proven in [Connd]).

30
The goal of this section is to explain and use this so-called knot group, so it will be good form
to enter the mindset of a group-theorist at this point. The key to defining (the relevant objects
needed to discuss) the knot group is to understand operations on paths, more specifically loops
but these are just a special case by Definition 4.6.

Definition 4.12 Let X be a topological space along with α : [0, 1] → X a path from x to y and
β : [0, 1] → X be a path from y to z. Then, we define the following:

α(2s), s ∈ [0, 1/2]
(i) The join of α and β is the path α∗β : [0, 1] → X where (α∗β)(s) = .
β(2s − 1), s ∈ [1/2, 1]

(ii) The reverse of α is the path α : [0, 1] → X where α(s) = α(1 − s).

(iii) The constant path at x is the path ex : [0, 1] → X where ex (s) = x, ∀s ∈ [0, 1].

Proposition 4.13 The join and reverse operations are well-defined on path homotopy equivalence
classes. Also, we have associativity of the join and the existence of an identity and inverses.

Proof : The proof consists of definition-chasing and some clever but not-too-difficult choices of
path homotopies. We simply refer to [Rog18] for this result in its gory detail.

As promised, Proposition 4.13 seems to give a group-like structure on the set of path homotopy
equivalence classes. In fact, when we restrict to loops based at a specified point, we get just that.

Definition 4.14 Let X be a topological space and α : [0, 1] → X be some loop based at x ∈ X.
The fundamental group is π1 (X, x) = {[α] : α is a loop based at x} with the join operation ∗.
The literature also calls this the first homotopy group.

Remark 4.15 It is possible to also define the nth homotopy group of a topological space X based
at x ∈ X, which is πn (X, x) = {[f ] : f : S n → X is continuous with f ((1, 0, ..., 0)) = x}. We will
not dwell on this much because π1 has much nicer properties and is better understood for our
purposes. Notice that the fundamental group can be thought of in this way also; simply consider
the equivalence relation we did in Example 1.33 which means the domain of our loops can be
identified with S 1 .

We next state a result by [McM13] and insert its proof, which was omitted in the original text.

Theorem 4.16 Let X be a topological space and α : [0, 1] → X be a path from x to y. The map
φ : π1 (X, x) → π1 (X, y) defined as φ([γ]) = [α ∗ γ ∗ α] is a group isomorphism.

Proof : Let γ1 and γ2 be two loops based at x. Then, we see that

φ([γ1 ]) ∗ φ([γ2 ]) = [α ∗ γ1 ∗ α] ∗ [α ∗ γ2 ∗ α]

31
 = [α ∗ γ1 ∗ α ∗ α ∗ γ2 ∗ α]
= [α ∗ γ1 ∗ ex ∗ γ2 ∗ α]
= [α ∗ (γ1 ∗ γ2 ) ∗ α]
= φ([γ1 ] ∗ [γ2 ]),

which shows that the map is at least a group homomorphism. For bijectivity, it is enough
to state and show that a certain map is its inverse; φ−1 : π1 (X, y) → π1 (X, x) defined as
φ−1 ([γ]) = [α ∗ γ ∗ α] will do. It is easy to see that φ ◦ φ−1 = idπ1 (X,y) and φ−1 ◦ φ = idπ1 (X,x) .

Corollary 4.17 If X is a path-connected topological space, then π1 is independent of the base point
x. In this case, one may simply denote the fundamental group by π1 (X).

Definition 4.18 Let X be a path-connected topological space such that π1 (X) ∼

= {e}, that is it
has trivial fundamental group. Then, it is called simply-connected.

Example 4.19 By Example 4.8, it is clear that π1 (Rn ) ∼

= {e}. Hence, Rn is simply-connected.

Lemma 4.20 If U ⊆ Rn is convex, then π1 (U ) ∼

= {e}, that is U is simply-connected.

Proof : Let α : [0, 1] → U be some loop based at x ∈ U . Then, F (s, t) = (1 − t)α(s) + tx is a

path homotopy that maps into U by convexity. Hence, α ≃ ex and so π1 (U ) ∼ = {e}.

We will not write a detailed proof of the classification of π1 (S 1 ) since this would quickly become
a surgery of homotopy theory as opposed to that of knot theory. Instead, we provide some of
the groundwork which leads to the proof of this result.

Definition 4.21 Let X be a topological space. A covering space of X is a topological space C

such that there exists a continuous surjective map p : C → X where, for each x ∈ X, there
exists an open neighbourhood Ux ⊆ X of x such that p−1 (Ux ) = ⊔λ∈Λ Vλ and p|Vλ : Vλ → Ux
is a homeomorphism for every λ ∈ Λ, where (Vλ )λ∈Λ is a family of open subsets of C. The
terminology used is as follows:

(i) X is the base space.

(ii) p is the covering map.

(iii) p−1 (x) is the fiber over x.

Example 4.22 Let p : R → S 1 be given as p(x) = (cos(2πx), sin(2πx)). Visually, this gives a
helix with centre the real line. This map acts as the covering map from the covering space R to
the base space S 1 . It was Example 1.33 that this map is continuous. Moreover, it is clearly one
periodic whereby the restriction p|(y,y+1) : (y, y + 1) → S 1 \ {p(y)} is a homeomorphism.

32
Two important results are the so-called Path-Lifting and Homotopy-Lifting Lemmata. It is from
these that we get the next result; full detail can be found in [Tho04].

Theorem 4.23 The fundamental group π1 (S 1 ) ∼

= (Z, +).

It would be immensely useful to be able to determine the fundamental group of a space which
looks ‘similar’ to that of a space for which the fundamental group is already known. We first
claim that the fundamental groups of the spaces in Figure 14 coincide.

SDR
−−−→
x x

Figure 14: A contrived example of a strong deformation retract of S 1 .

This can and will be proven using the next definition and a useful result that comes of it.

Definition 4.24 Let X and Y be topological spaces with x ∈ X and y ∈ Y and h : X → Y be

continuous such that h(x) = y. Then, the induced homomorphism is h∗ : π1 (X, x) → π1 (Y, y)
which is given by h∗ ([α]) = [h ◦ α], where α : [0, 1] → X is a loop based at x.

Proposition 4.25 Let X, Y, Z be topological spaces with f : X → Y and g : Y → Z continuous.

Inducing homomorphisms is functorial, that is (g ◦ f )∗ = g∗ ◦ f∗ and (idX )∗ = idπ1 (X,x) , ∀x ∈ X.

Proof : Let α : [0, 1] → X be a loop based at x. By definition, we have that (g◦f )∗ ([α]) = [g◦f ◦α]
and that f∗ ([α]) = [f ◦ α], which implies g∗ (f∗ ([α])) = g∗ ([f ◦ α]) = [g ◦ f ◦ α]. Consequently, we
see that (g ◦ f )∗ = g∗ ◦ f∗ . Furthermore, (idX )∗ ([α]) = [idX ◦ α] = [α], by definition. As a result,
we see that (idX )∗ = idπ1 (X,x) .

It is now possible to prove a result which will mean that our knot group is an isotopy invariant
(see Theorem 4.46 for the full details, but note the heavy-lifting is done here).

Theorem 4.26 The fundamental group is invariant under homeomorphism.

Proof : Let X and Y be topological spaces and f : X → Y a homeomorphism such that f (x) = y,
for some x ∈ X and y ∈ Y . By definition, f∗ : π1 (X, x) → π1 (Y, y) is a group homomorphism.

33
We now claim that (f∗ )−1 = (f −1 )∗ . Indeed, both f and f −1 are continuous; we can appeal to the
functorial property in Proposition 4.25 to obtain f∗−1 ◦f∗ = (idπ1 (X,x) )∗ and f∗ ◦f∗−1 = (idπ1 (Y,y) )∗ .
Therefore, it follows that f∗ is bijective and thus a group isomorphism.

Definition 4.27 Let X be a topological space and A ⊆ X some subspace. It is said that A is a
strong deformation retract (SDR) of X if there exists a continuous map H : X × [0, 1] → X called
a strong deformation retraction such that the following conditions hold:

(i) H(x, 0) = x, ∀x ∈ A.

(ii) H(x, 1) ∈ A, ∀x ∈ A.

(iii) H(a, t) = a, ∀a ∈ A, ∀t ∈ [0, 1].

Example 4.28 An easy example is that the singleton {a} is a strong deformation retract of the
interval [−a, a] via H(x, t) = (1−t)x+ta. Another such example is the map G we saw in Example
4.4; it implies that S 1 is a strong deformation retract of the punctured plane R2 \ {(0, 0)}. This
map also implies the more general result that S n−1 is a strong deformation retract of Rn \ {0}.

Proposition 4.29 Let X be a topological space and A ⊆ X a strong deformation retract of X.

For all a ∈ A, the inclusion ιA : A → X induces the isomorphism (ιA )∗ : π1 (A, a) → π1 (X, a).

Proof : Suppose that H is the strong deformation retraction of A in X. For notation’s sake, we
define h1 : X → A to be h1 (x) = H(x, 1). By definition, this is a continuous map. Furthermore,
note that h1 ◦ ιA : A → A is the identity map on A, which means that (h1 ◦ ιA )∗ = (h1 )∗ ◦ (ιA )∗
is the identity map on π1 (A, a) by Proposition 4.25. Next, we consider ιA ◦ h1 : X → X. Let
α : [0, 1] → X be a loop based at a. Then, we see that [α] = [ιA ◦ h1 ◦ α] via the path homotopy
F (s, t) = H(α(s), t). Therefore, we also have that (ιA ◦ h1 )∗ = (ιA )∗ ◦ (h1 )∗ is the identity map
on π1 (X, a). Consequently, (ιA )−1
∗ = (h1 )∗ and so (ιA )∗ is a group isomorphism.

If we re-visit Figure 14, it is clear from Proposition 4.29 that the fundamental group of the space
based anywhere on the circle coincides with the fundamental group of the circle, that is (Z, +).

Example 4.30 Consider the so-called Klein bottle (one of the coolest shapes in Mathematics).
Although it is non-embeddable in three-dimensions, we consider its planar drawing in Figure 15.

34
 Figure 15: A planar drawing of the Klein bottle with gluing relations.

Remove one point from the Klein bottle, say the centre point without loss of generality. Applying
a strong deformation retract takes us to a bouquet of two circles, as is shown in Figure 16.

SDR SDR SDR

−−−→ −−−→ −−−→
t=0.5 t=0.9 t=1

x x x x

Figure 16: Performing a strong deformation retract on the punctured Klein bottle.

This lights the flame at the heart of the upcoming discussion in Section 5 on using such basic
sketches as devices to build intuition and understand strange topological spaces.

4.2 Combinatorial Group Theory

It is now time to lay the groundwork which will lead to the Wirtinger presentation of a group,
which is an immensely useful and relatively simple algorithmic tool for determining what we call
the knot group.

Definition 4.31 Let A = {g1 , ..., gn }, which we call the alphabet. A word is a finite concatenation
of the symbols gi and gi−1 , where the empty word is denoted e, that is the word with no symbols.

It is possible to define two operations on the set of words of an alphabet:

ˆ The product of two words is their concatenation.

ˆ The inverse of a word is given by reversing the order and interchanging gi and gi−1 .

Example 4.32 Consider the English alphabet A = {a, b, ..., z}. From this, we construct the word
hello = hel2 o, where we use the usual power notation to denote how many of that letter is
concatenated. Then, we see that (hello)−1 = o−1 l−2 e−1 h−1 . Alternatively, another word we
may construct is hel. Hence, the product (hello)−1 hel = o−1 l−1 = (lo)−1 . Overcoming the fact
this example is somewhat perverse, the usefulness of these operations cannot be overstated.

35
Lemma 4.33 Let A = {g1 , ..., gn } be an alphabet and W = {r1 , ..., rm } a finite set of words from
the alphabet A. This is an equivalence relation: r1 ∼ r2 if and only if r1 is obtainable from r2 by

inserting/removing gi g −1 and gi−1 gi or concatenating/removing r and r−1 .

Proof : It is clear that r1 ∼ r1 simply by concatenating the empty word e at the end, say. If
r1 ∼ r2 , then r2 ∼ r1 because we can concatenate r−1 onto where an r had been concatenated or
we can insert gi−1 gi after a place where gi gi−1 had been inserted. Finally, if r1 ∼ r2 and r2 ∼ r3 ,
then r1 ∼ r3 simply by performing the inserts/removals in one fell swoop.

Proposition 4.34 The quotient W/∼ = ⟨g1 , ..., gn ; r1 , ..., rm ⟩ is a group under concatenation.

Hence, we now say that the gi are generators and the rj are relations. It may not always be that
we have any relations. This is now explored and follows the framework of [MI02], culminating
in a neat application to the theory of the fundamental group.

Definition 4.35 Let F be a group and S ⊆ F some subset. We say that S is a free basis of F if,
for any group G and any map φ : S → G, there exists a unique group homomorphism φ
e:F →G
such that φ|
e S = φ. In this case, F is called a free group.

Example 4.36 The group (Z, +) is a free group with basis {1}. Indeed, if φ : {1} → G is a map
e : Z → G defined as φ(n)
where φ(1) = g, then φ e = g n is a group homomorphism with φ|
e {1} = φ.

A free group can now be understood as in Proposition 4.34, that is a free group F with basis S
has presentation ⟨S; ·⟩, that is where the set of relations is empty. But why can we make such an
identification? Because an application of the First Isomorphism Theorem to the unique group
homomorphism φ e ∼
e in Definition 4.35 gives F/ ker(φ) = G. Because of this combinatorial approach
to defining free groups, it is possible to state the celebrated Seifert-van Kampen Theorem in this
way, as is done in [Bignd].

Theorem 4.37 (Seifert-van Kampen Theorem) Let X be a topological space where U, V ⊆ X are
open, path-connected subsets with X = U ∪ V and U ∩ V ̸= ∅ also path-connected. Suppose that
ιU : U ∩ V → U and ιV : U ∩ V → V are inclusion maps and that for x ∈ U ∩ V , we have these
group presentations:

π1 (U, x) ∼
= ⟨a1 , ..., an ; r1 , ..., ri ⟩,
π1 (V, x) ∼
= ⟨b1 , ..., bm ; s1 , ..., sj ⟩,
π1 (U ∩ V, x) ∼
= ⟨c1 , ..., cl ; t1 , ..., tk ⟩.

Then, it is possible to write the presentation for the fundamental group of X as follows:

π1 (X, x) ∼
= ⟨a1 , ..., an , b1 , ..., bm ; r1 , ..., ri , s1 , ..., sj , (ιU )∗ (c1 )(ιV )∗ (c1 )−1 , ..., (ιU )∗ (cl )(ιV )∗ (cl )−1 ⟩.

36
Example 4.38 Let X = S 1 ⊔S 1 and identify one point on each circle via some equivalence relation
∼. Then, X/∼ is the so-called bouquet of two circles that intersect only at the identified point,
[p] ∈ X/∼, say. The goal is to apply the Seifert-van Kampen Theorem; this can be done with the
open sets shown in Figure 17.

[p] [p] [p]

U V U ∩V

Figure 17: One choice of open subsets of the bouquet of two circles.

We see that U and V are open and path-connected, with non-empty path-connected intersection.
Furthermore, S 1 is a deformation retract of both U and V , meaning π1 (U ) ∼
= π1 (S 1 ) ∼
= π1 (V ).
Example 4.36 implies π(S 1 ) ∼
= ⟨a; ·⟩, where a is some generator for loops based in S 1 . Therefore,

π1 (U ) ∼
= ⟨a; ·⟩ and π1 (V ) ∼
= ⟨b; ·⟩.

Also, U ∩ V will deformation retract to a point, which means π1 (U ∩ V ) ∼

= ⟨·; ·⟩. It then follows
from the Seifert-van Kampen Theorem that

π1 (X/∼) ∼
= ⟨a, b; ·⟩.

Remark 4.39 The work of Example 4.38 can be applied inductively, that is to discuss the so-called
bouquet of n-circles with an equivalence relation identifying one point on each copy. It turns out
that if X/∼ is our bouquet, then π1 (X/∼) ∼
= ⟨a1 , ..., an ; ·⟩, the free group of n generators.

Example 4.40 Let us make use of Example 4.30 in order to calculate π1 (K 2 ), where K 2 is a
Klein bottle. We can find two open subsets of K 2 , drawn as planar projections in Figure 18.
b

U V U ∩V

Figure 18: One choice of open subsets of the Klein bottle.

37
It was Example 4.30 that established the bouquet of two circles was a deformation retract of
U , so π1 (U ) ∼= ⟨a, b; ·⟩ by Example 4.38. Furthermore, it is clear that V is simply connected, so
π1 (V ) = ⟨·; ·⟩. Finally, U ∩V will deformation retract to S 1 , meaning that π1 (U ∩V ) ∼
∼ = ⟨c; ·⟩. If we
consider the inclusion maps as in the Seifert-van Kampen Theorem, it is clear that (ιV )∗ (c) = e
since V is simply connected and that (ιU )∗ (c) = aba−1 b. Consequently,

π1 (K 2 ) ∼
= ⟨a, b; aba−1 b⟩.

4.3 The Knot Group

The work on combinatorial group theory allows for us to comfortably introduce the so-called
Wirtinger presentation, which turns out to be a very useful device when considering the knot
group; this is equivalent to computing a fundamental group, a point discussed further towards
the end of this section. But first, let us finally live up to the name of Section 4 and discuss the
knot group.

Definition 4.41 Let K be a diagram of a knot. The knot group is gr(K).

This is far from a useful definition, but we will soon prove a useful form of the knot group, which
can then be interpreted geometrically and topologically.

Definition 4.42 Let K be an oriented diagram of a knot with n arcs labelled xi and m crossings.
The Wirtinger presentation is gr(K) ∼
= ⟨x1 , ..., xn ; r1 , ..., rm ⟩, with relations given in Figure 19.

xj xk xi xk

xi xj
L+ L−

xk xi x−1 −1
k xj = e xk x−1 −1
j xk xi = e

Figure 19: The Wirtinger relations as loops above positive and negative crossings.

Example 4.43 Yet again, we consider the first trefoil knot as in Figure 2, i.e. with clockwise
orientation. The Wirtinger relations are as follows:

x3 x−1 −1
2 x3 x1 = e ⇒ x1 = x−1
3 x2 x3

x1 x−1 −1
3 x1 x2 = e ⇒ x2 = x−1
1 x3 x1

x2 x−1 −1
1 x2 x3 = e ⇒ x3 = x−1
2 x1 x2

38
Thus, gr(K) ∼ = ⟨x1 , x2 , x3 ; x3 x−1 −1 −1 −1 −1 −1
2 x3 x1 , x1 x3 x1 x2 , x2 x1 x2 x3 ⟩. It is possible to eliminate one
of the relations by substituting in one of the above expressions. We will do so for x3 :

x1 = x−1 −1
2 x1 x2 x1 x2 and x2 = x−1 −1
1 x2 x1 x2 x1 .

Therefore,

gr(K) ∼
= ⟨x1 , x2 ; x−1 −1 −1 −1 −1 −1
1 x2 x1 x2 x1 x2 , x2 x1 x2 x1 x2 x1 ⟩

= ⟨x1 , x2 ; (x1 x2 x1 )−1 x2 x1 x2 , (x2 x1 x2 )−1 x1 x2 x1 ⟩

= ⟨x1 , x2 ; x1 x2 x1 = x2 x1 x2 ⟩
= ⟨a, b; a2 = b3 ⟩,

where we define a = x2 x1 x2 and b = x1 x2 .

Remark 4.44 It is possible to imagine the trefoil as embedded on the surface of a torus, where
the trefoil winds twice around the meridian (in the direction of the longitude) and thrice around
the longitude (in the direction of the meridian). This is visualised in Figure 20. Note that the
picture is a sort-of push-off of the knot, in that it is slightly lifted away from the torus so we can
better see how it is embedded. As such, the trefoil may be named the (2, 3)-torus knot. Note
the naming convention with regard to the indices that appear in the Wirtinger presentation in
Example 4.43. Our goal is to prove that this is no coincidence.

Figure 20: A birds-eye view of the trefoil embedded onto T 2 , with a slight push-off.

Proposition 4.45 Let K be a diagram of a knot. Then, gr(K) ∼

= π1 (R3 \ K).

Proof : Let p ∈ R3 \ K. Then, it is possible to consider loops based at p that enclose each of the
arcs in the crossing drawn in Figure 19 such that they have positive signs at each crossing in the
sense of Definition 2.9. Let the loop that encircles arc xi be denoted αi . Then, we deal with three
such loops, namely αi , αj , αk . We now show that the Wirtinger presentation coincides with the

39
path homotopy class involving these loops. Indeed then, Figure 21(a) demonstrates this in one
case. From this, we see that αk ∗ αi ∗ αk−1 ≃ αj , which agrees with the first Wirtinger relation.
We next consider Figure 21(b) from which we see that αk−1 ∗ αj ∗ αk ≃ αi , which agrees with the
second Wirtinger relation.

xk xk

αk−1

αi αj
xj xi xj xi
≃

αk
p p

(a) A positive crossing and the join of loops that encode the presentation.

xk xk

αk

αj αi
xi xj xi xj
≃

αk−1
p p

(b) A negative crossing and the join of loops that encode the presentation.

Figure 21: Wirtinger relations encoded by loops in the space R3 \ K.

From here, it is possible to appeal to the Seifert-van Kampen Theorem with strategically chosen
open and path-connected subsets on which the generators α1 , ..., αn have relations r1 , ..., rm as
in Definition 4.42, establishing our correspondence between the groups in the statement.

Theorem 4.46 The knot group is an isotopy invariant.

Proof : Recall that homeomorphic spaces have isomorphic fundamental groups, via Theorem 4.26
If K1 and K2 are isotopic knots, that is K1 ≃ K2 , then this implies R3 \ K1 ∼
= R3 \ K2 and so

40
π1 (R3 \ K1 ) ∼
= π1 (R3 \ K2 ). We then appeal to Proposition 4.45 to see that gr(K1 ) ∼
= gr(K2 ).

Example 4.47 We will determine the knot group of the unknot. We know that π1 (S 1 ) ∼
= (Z, +)
but we must determine π1 (R3 \ S 1 ). We proceed as in the proof of Proposition 4.45, by choosing
some base point p ∈ R3 \ S 1 of classes of loops. Of course, if a loop at p doesn’t link with S 1 ,
then the loop is contractible to the constant path ep . But of course, if a generates the loop α
that links with S 1 once, then another loop β that links with S 1 twice, say, will be generated
by a2 , as in Figure 22. Therefore, we see a familiarity with Example 4.36 and conclude that
π1 (R3 \ S 1 ) ∼
= ⟨a; ·⟩ ∼
= (Z, +).

α β

Figure 22: Representatives of two classes of loops that appear in π1 (R3 \ S 1 ).

Proposition 4.48 Let K be a diagram of a knot. Then, K is p-colourable if and only if there
exists a group homomorphism φ : gr(K) → D2p to the dihedral group of order 2p.

Proof : Recall that D2p = ⟨a, b; ap , b2 , abab⟩, where a generates rotations and b generates reflections,
by way of a result in [CSW16]. We flesh out the homomorphism briefly mentioned in this source.
Define φ(xi ) = aci b where ci is the colour of the arc xi . We must now verify that this is a group
homomorphism. Indeed, we apply this to the Wirtinger presentation of gr(K):

φ(xj ) = φ(xk xi x−1

k )
⇔ acj b = ack baci −ck
⇔ bacj b = back baci −ck
⇔ a−cj = aci −2ck .

But now, this is true if and only if ci + cj − 2ck ≡ 0 (mod p), which is precisely the p-colourability
condition (note that the Wirtinger presentation has arc xk as the over-crossing). Conversely, all
non-trivial homomorphisms between these groups arise in this way by tracing the above chain of
implications in reverse.

We took some time in Remark 4.44 to introduce the concept of a torus knot, where we also
alluded to a deeper interplay between these such knots and the Wirtinger presentation of their

41
knot groups. Proving this will allow us to conclude our work having an idea on how to use the
objects looked at throughout this section. First, we will define these special knots properly.

Definition 4.49 Let K be the diagram of a knot. It is called a (p, q)-torus knot if it is embeddable
on the surface of a torus such that it winds p times around the meridian and q times around the
longitude, where also gcd(p, q) = 1. In the case that gcd(p, q) ̸= 1, we have a (p, q)-torus link,
with number of components gcd(p, q). Such a knot will be denoted Tp,q .

Lemma 4.50 Let Tp,q be an oriented (p, q)-torus knot. Then, the following are true:

(i) T−p,−q ≃ Tep,q , i.e. it has reversed orientation.

(ii) Tp,−q ≃ T p,q , i.e. it is its mirror image.

(iii) Tp,q ≃ Tq,p .

Sketch of Proof : Now, (i) is intuitive and trivial. For (ii) we refer to [Kaw96]. As for (iii), we
follow the process as done in [Ada04] but generalising to arbitrary (p, q)-torus knots. Indeed,
suppose D ⊆ T 2 is a closed disk of a torus such that D ∩ Tp,q = ∅. Then, T 2 \ D deformation
retracts to a bouquet of two circles, but as T 2 is deformed in this way, Tp,q ends up being carried
by this deformation to two closed strips. It is possible to invert these (turn them inside out); the
roles of meridian and longitude thus swap roles during this process. Therefore, inserting back
the removed disk gives a torus on which the knot is a (q, p)-torus knot; a lovely illustration can
be found in the aforementioned source.

We are ready to reap the rewards of studying the knot group and Seifert-van Kampen Theorem.
As with some of the previous knot-theoretic concepts, we turn to [GP94], adapting their proof
slightly.

Theorem 4.51 The knot group π1 (R3 \ Tp,q ) ∼

= ⟨a, b; ap = bq ⟩.

Proof : Let T be the solid torus, which has boundary T 2 , and take some (p, q)-torus knot Tp,q ⊆ T 2
as in Definition 4.49. For some ε > 0, let Nε ⊆ R3 be the tubular neighbourhood of Tp,q .
Notice that the closure N ε deformation retracts to Tp,q , meaning π1 (R3 \ N ε ) ∼
= π1 (R3 \ Tp,q ) by
Proposition 4.29. We next define the following:

U = T \ Nε and V = (R3 \ T ) \ N ε

Geometrically, U is the solid part of our solid torus with a groove carved into it where which
the tubular neighbourhood would otherwise occupy; we can see that it deformation retracts to
the longitudinal circle at the centre of the solid part of the torus, that is π1 (U ) ∼
= ⟨a; ·⟩. Next,
we see that the fundamental group of V is generated by a loop through the hole of the torus,

42
from which it follows that π1 (V ) ∼
= ⟨b; ·⟩, with b being a generating loop passing through the hole
formed by removing the point x. Finally, we see that U ∩ V is an open annulus which therefore
has a circle as a deformation retract, so π1 (U ∩ V ) ∼
= ⟨c; ·⟩. The hypotheses of the Seifert-van
Kampen Theorem are satisfied:

π1 (R3 \ N ε ) ∼
= ⟨a, b; (ιU )∗ (c)(ιV )∗ (c)−1 ⟩ ∼
= ⟨a, b; ap b−q ⟩,

which occurs since the generator c corresponds to loops at the centre of the open annulus, winding
p times in the longitudinal direction and q times in the meridional direction.

Sadly, the knot group isn’t a complete isotopy invariant, in that there are instances of non-ambient
isotopic knots whose knot groups are isomorphic; we will show this by way of an example.

Example 4.52 The mirror image of our trefoil knot T2,3 is the knot T2,−3 by Lemma 4.50. As
such, we can appeal to Theorem 4.51 to see that

gr(T2,−3 ) ∼
= ⟨a, b; a2 = b−3 ⟩ ∼
= ⟨c, d; c2 = d3 ⟩ ∼
= gr(T2,3 ).

where the isomorphism in the above equation comes from mapping a 7→ c and b 7→ d−1 . But as
we have mentioned prior to this, we know that the trefoil is chiral, something which the knot
group fails to detect here.

Remark 4.53 The only knot groups with non-trivial centres are those of the torus knots, which
is a result discussed in [BZ66]. This is not a group theory paper and thus we stop here, but it is
interesting nevertheless.

43
5 Seifert Surfaces
We can use the classification of surfaces to discuss the interesting question what can we say
about a surface whose boundary is a given knot?. First, we will discuss the necessary theory as
to develop the titular concept in a rigorous manner.

Definition 5.1 A (topological) surface without boundary is a topological space Σ such that for
any x ∈ Σ, there exists an open neighbourhood Ux ⊆ Σ of x such that Ux ∼ = R2 . The boundary
∂Σ of a topological surface Σ is defined to be the set of points whose open neighbourhood is
homeomorphic to R2≥0 .

Example 5.2 It follows that R2 ⊆ R3 is a surface. Indeed, for any x ∈ R2 , a simply-connected

open neighbourhood Ux ⊆ R2 will certainly be homeomorphic to R2 , a consequence of the
well-known Riemann Mapping Theorem of complex analysis, where we make the identification
R2 ∼
= C, which gives a bijection between such a neighbourhood to the open unit disk.

We now deviate the discussion to the study of surfaces and will emerge with the required
knowledge to understand how knots relate to this area. This will be done in a more general
context by appealing to homological methods before restricting to the two-dimensional case that
we need. But first, we note one property of general topological spaces that we relate to surfaces.

Definition 5.3 Let X be a topological space and (Uλ )λ∈Λ a collection of open subsets.

ˆ The sets (Uλ )λ∈Λ form an open cover if λ∈Λ Uλ = X.

S

ˆ The sets (Uλ )λ∈Λ′ form a finite subcover if Λ′ ⊆ Λ has finite size and λ∈Λ′ Uλ = X.
S

The topological space X is called compact if every open cover admits a finite subcover.

Remark 5.4 When we talk about compact surfaces, we mean precisely that it is a compact
topological space with the property stated in Definition 5.1. In the case that a surface is both
compact and has no boundary, we call it closed and in the case that they have boundary, we can
think of them as being closed yet having a finite number of open disks removed. This idea will
be re-visited in Theorems 5.35 and 5.41; having boundary is formalised in Definition 5.39.

5.1 Combinatorial Topology

We now discuss the so-called triangulation of a surface; this delves into what is called homology.
This provides a means of defining orientability and will lay the groundwork for these Seifert
surfaces we wish to study.

Definition 5.5 Let p1 , ..., pn+1 ∈ Rn be affinely independent, that is p2 − p1 , ..., pn+1 − p1 are

44
linearly independent. The n-simplex generated by these points is the set

∆n = {λ1 p1 + ... + λn+1 pn+1 : λ1 + ... + λn+1 = 1 and λi ≥ 0}.

A neat way to think about these n-simplices are as complete graphs on n + 1 vertices with a
convex hull, that is the graph is the ‘frame’ of a polytope. This becomes evident by looking
at Example 5.6. Although we only require low-dimensional forms for the relation of surfaces to
knots, it is interesting to remain as general as possible.

Example 5.6 We can consider a number of simple examples which are sketched in Figure 23:

(i) The 0-simplex is a point (or vertex).

(ii) The 1-simplex is a line (or edge).

(iii) The 2-simplex is a triangle (or face).

(a) The 0-simplex. (b) The 1-simplex. (c) The 2-simplex.

Figure 23: The simplices up to two-dimensions viewed pictorially.

We aim to relate this discussion to surfaces; although Definition 5.5 is general for any dimension,
it is redundant for the purpose of our work to consider n > 3.

Definition 5.7 A k-face of an n-simplex is a subset of ∆n , where k ≤ n. It is proper when it is a

proper subset, that is not an n-face.

Example 5.8 Consider the 2-simplex. It is clear that it has three 0-faces (vertices) and three
1-faces (edges) and only one 2-face (the simplex itself).

It is possible to take a number of simplices and ‘glue’ them together on corresponding faces, by
which we mean identify faces via homeomorphisms.

Definition 5.9 An n-dimensional pseudo-simplicial complex (or n-complex) is a finite collection ∆

e
k
of simplices, where n = max{k : ∆ ⊆ ∆},
e that satisfy the following:

(i) For any simplex in ∆,

e all faces of that simplex are in ∆.
e

(ii) The intersection of any two simplices in ∆

e is either a common face or is empty.

45
Example 5.10 We form a 3-complex by adjoining a 2-simplex to a 3-simplex as in Figure 24:

Figure 24: One possible three-dimensional complex.

Definition 5.11 A triangulation of a surface Σ is a homeomorphism between the topological space

Σ and some simplicial complex.

Definition 5.12 Let Σ1 and Σ2 be surfaces. We say they are isomorphic if there is a bijection
between the vertices of the triangulation of Σ1 and the vertices of the triangulation of Σ2 such
that triangles map to triangles and the identification of sides is preserved.

Example 5.13 The sphere S 2 can be identified in the two slightly different ways shown below.

(a) The first gluing relation of S 2 . (b) The second gluing relation of S 2 .

Figure 25: Triangulations of S 2 .

It turns out these gluing relations are not isomorphic as noted in [Rob15], despite the obvious
fact that S 2 ∼
= S 2 . Therefore, something similar to but stronger than isomorphism is needed if
we are to classify surfaces in this combinatorial way. Another thing to note: although the images
above represent a surface via triangles, they are not triangulations. Look at the first one, say;
each 2-simplex shares three edges and three points; the picture is not that of a 2-complex since
their intersection is non-empty and is not a common face, so the second condition of Definition
5.9 fails. In fact, this is an example of a CW-complex, which has a less-restrictive definition than
that of our simplicial complexes.

46
Definition 5.14 Let Σ be a surface defined by some gluing relation. A subdivision of the gluing
relation is a decomposition into triangles.

Example 5.15 Consider the gluing relation in Figure 26 applied to a hexagon. By subdividing,
it is clear the surface is that of a torus T 2 .

∼
= ∼
= ∼
=

Figure 26: A subdivision which yields the usual projection of the torus T 2 .

We now introduce the way in which we identify two surfaces by only their gluing relations.

Definition 5.16 Two gluing relations are called equivalent if there exists a finite sequence of
isomorphisms and subdivisions relating them.

The next result is useful for subsequent proofs but is difficult to show rigorously; it is mentioned
and used in [Pet07].

Theorem 5.17 Two surfaces Σ1 and Σ2 are homeomorphic if and only if their gluing relations
are equivalent.

Definition 5.18 An orientation of a surface Σ is an assignment of a clockwise or anti-clockwise

circulation to each face of a triangulation such that circulations across any edge oppose one
another. In this case, Σ is called orientable.

Example 5.19 The sphere S 2 is orientable. Indeed, consider the following triangulation with the
assigned circulations (and note that the condition laid out in Definition 5.18 is satisfied).

Figure 27: An orientation of the sphere, given via triangulation.

47
Furthermore, we see that the Möbius strip M 2 is non-orientable. Indeed, take a triangulation of
M 2 and apply a circulation to one face; we immediately see a contradiction to Definition 5.18 if
we try to extend this across neighbouring faces – the orientation across the identified edges at
either end oppose each other.

Figure 28: An attempt at an orientation of the Möbius strip, given via triangulation.

Proposition 5.20 Orientability is a topological invariant.

Proof : It suffices by Theorem 5.17 to show that orientability is preserved under the equivalence
of gluing relations, in particular triangulations. Indeed, let Σ be an orientable surface have
some gluing relation. Any subdivision is also orientable; this is achieved by prescribing the same
circulation to any triangle of the face which it comes from. Of course, isomorphisms will preserve
circulations and thus we are done.

Definition 5.21 Let Σ be a surface. A subsurface of Σ is that which is obtained by deleting some
faces and their gluing instructions from the gluing relation of Σ.

Remark 5.22 It is useful to now assume a graph-theoretic approach in the style of [Rob15]. Given
the gluing relations, we can very naturally introduce two different graphs which are defined as
follows (see Appendix C for further details on this):

ˆ The identification graph G is one whose vertices and edges represent the vertices and edges
of the gluing relation, respectively.

ˆ The identification dual graph G∗ is one whose vertices and edges represent faces and adjacency
of faces given by the gluing relation, respectively.

Lemma 5.23 A surface Σ is non-orientable if and only if the Möbius strip is a subsurface of Σ.

Proof : (⇐) If M 2 is a subsurface, then Σ is non-orientable by Example 5.19.

(⇒) We proceed with a proof using graph-theoretic methods. Consider the identification dual
graph G∗ of Σ. Because this graph is connected, it is possible to choose some spanning tree T
of G∗ . Now, simply choose a circulation on one face and extend this along the edges of T , that

48
is to the other faces in the gluing relation. Let G be the identification graph of Σ and H the
subgraph of G where we removes the edges of G that intersect those of T . If the circulations
across any edge of H do not oppose each other, then Σ is non-orientable. If this occurs, then
we have a cycle in G∗ , which corresponds to a cycle of faces and thus the existence of M 2 as a
subsurface.

Example 5.24 Consider the so-called projective plane P2 as now sketched (we can think of it as
antipodal points of S 1 identified but we consider it also as a quotient of [0, 1] × [0, 1], which is
slightly more useful in this context).

∼
=

Figure 29: Two representations of the projective plane P2 .

Now, it is possible to subdivide the second image of Figure 29, which is shown in Figure 30. We
can then see that M 2 is a subsurface and so Lemma 5.23 applies, giving that P2 is non-orientable.

∼
= ∪

Figure 30: A pictorial justification that P2 = M 2 ∪ D, where D is a closed disk.

We are getting close to being able to classify topological surfaces. But first, there is an important
concept that must be discussed.

Definition 5.25 Let ∆

e be an n-complex. The Euler characteristic of ∆
e is defined as

n
X
χ(∆)
e = (−1)k ∆
e (k) ,
k=0

e (k) represents the set of k-faces of ∆.

where ∆ e

49
Example 5.26 Consider the 3-simplex ∆3 . Then, χ(∆3 ) = 4 − 6 + 4 = 2. This should be
familiar to those with a knowledge on the Platonic solids. However, Definition 5.25 is a deeper
generalisation which holds in arbitrary dimensions.

Example 5.27 Consider a truncated icosahedron X consisting of p pentagons and h hexagons.

Then, there are (5p + 6h)/3 vertices and (5p + 6h)/2 edges and p + h faces. As such, χ(X) = p/6.
But now, this is a convex shape, so it must be that χ(X) = 2, meaning that p = 12. We have
just proven that every football consists of precisely 12 pentagonal faces.

Lemma 5.28 Let X = A ∪ B, where A and B are n-complexes. Then,

χ(X) = χ(A) + χ(B) − χ(A ∩ B).

Proof : Using Definition 5.25, we formulate the right-hand-side of the equation in the statement:
n
X  
k (k) (k) (k)
χ(A) + χ(B) − χ(A ∩ B) = (−1) A + B − (A ∩ B) + (−1)n (A ∩ B)(n) ,
k=0

but A ∩ B is an (n − 1)-complex, meaning it has no n-faces, that is the final term above vanishes.
The Inclusion-Exclusion Principle gives |A ∪ B| = |A| + |B| − |A ∩ B|, which applies to the above:
n
X n
X
χ(A) + χ(B) − χ(A ∩ B) = (−1)k (A ∪ B)(k) = (−1)k X (k) = χ(X),
k=0 k=0

which is precisely the formula we require.

Proposition 5.29 The Euler characteristic is a topological invariant.

Proof : It again suffices by Theorem 5.17 to show that the Euler characteristic is preserved under
the equivalence of gluing relations; a subdivision adds an edge and adds a vertex but it also
splits another edge into two and adds a face. Thus, if Σ is a surface with Euler characteristic
χ(Σ) = V − E + F , then the Euler characteristic of the subdivision of this surface is

(V + 1) − (E + 2) + (F + 1) = V − E + F,

so χ(Σ) is preserved. Finally, isomorphisms trivially preserve χ(Σ) and thus we are done.

Example 5.30 Let X = S 2 ⊔ S 2 . Then, Lemma 5.28 gives that

χ(X) = χ(S 2 ) + χ(S 2 ) − χ(S 2 ∩ S 2 ) = 2 + 2 − 0 = 4,

using that the copies of S 2 are disjoint. Note that Example 5.26 gives the Euler characteristic of

50
S 2 , since the 3-simplex is the triangulation of the sphere and Proposition 5.29 means they have
equal Euler characteristic. Inductively, χ(S 2 ⊔ · · · ⊔ S 2 ) = 2n, where there are n copies of S 2 .

Another useful property of surfaces that can be discussed are that of through-holes. It seems
reasonable to assert that a torus cannot be homeomorphic to a sphere because a sphere has no
through-holes yet the torus has one. This concept can now be made rigorous.

Definition 5.31 Let Σ be a surface. The genus of the surface g(Σ) is the supremum of the number
of non-intersecting simple closed curves on Σ that preserve connectedness of the surface.

Example 5.32 Any closed curve on T 2 that is homotopically equivalent to a point will disconnect
the surface. However, one closed curve through the hole of the torus will preserve connectedness
(and a second such curve will give a disconnection). Therefore, g(T 2 ) = 1.

Recall in Section 2 we defined the connected sum of link diagrams; we noted in Remark 2.21
that this is a specific case of a more general topological operation, which is next discussed.

Definition 5.33 Let Σ1 and Σ2 be surfaces. The connected sum of these surfaces, denoted Σ1 #Σ2 ,
is found by identifying the boundary of a single face in each gluing relation.

Lemma 5.34 Let A and B be n-complexes. Then,

χ(A#B) = χ(A) + χ(B) − χ(S n ).

Proof : (of the n = 2 case) We shall prove this result for two surfaces Σ1 and Σ2 . Let X1 and
X2 be the surfaces’ respective triangulations, that is they are 2-complexes. By Definition 5.33,
the connected sum of these complexes comes about by identifying two 2-faces, one from each of
X1 and X2 . If we say that X is the 2-complex representing Σ1 #Σ2 , then we have the following:

ˆ The number of vertices VX = VX1 + VX2 − 3.

ˆ The number of edges EX = EX1 + EX2 − 3.

ˆ The number of faces FX = FX1 + FX2 − 2.

Therefore, we can easily see that χ(Σ1 #Σ2 ) = χ(Σ1 ) + χ(Σ2 ) − 2. The general case follows by
doing a similar analysis of the vertices, edges and faces of higher-dimensional complexes.

Theorem 5.35 (Classification of Surfaces without Boundary) Every closed connected surface Σ
can be represented in precisely one of the following ways:

(i) As a sphere, in which χ(Σ) = 2.

(ii) As the connected sum of n-tori, in which χ(Σ) = 2 − 2n.

(iii) As the connected sum of n-projective planes, in which χ(Σ) = 2 − n.

51
Remark 5.36 We omit the proof since it would detract from the intentions of this paper; we can
however see the full detail in [Zee66]. We shall note that the facts about the Euler characteristics
in the statement of Theorem 5.35 follow from inductively applying Lemma 5.34, using χ(T 2 ) = 0
and χ(P2 ) = 1.

Definition 5.37 If P2 is a subsurface of a surface Σ, then we say that it is a cross-cap of Σ.

From Theorem 5.35, we can state formulae for the genera of closed surfaces Σ without boundary:


1 − 1 χ(Σ), if Σ is orientable
2
g(Σ) = . (†)
2 − χ(Σ), if Σ is non-orientable

Example 5.38 We know that the torus T 2 is orientable by Lemma 5.23 and we easily compute
χ(T 2 ) = 1 − 2 + 1 = 0 and so g(T 2 ) = 1 (this agrees with our previous answer in Example 5.32).

Definition 5.39 Let D ⊆ R2 be a closed disk and Σ a surface. We say that D is a cuff of the
surface Σ#D. In other words, Σ#D is a surface with one boundary component.

Corollary 5.40 (of Lemma 5.34) Let Σ be a surface and D a cuff on the surface. Then,

χ(Σ#D) = χ(Σ) − 1.

Proof : It is clear that χ(D) = 1. Therefore, χ(Σ#D) = χ(Σ) + 1 − 2 = χ(Σ) − 1.

Theorem 5.41 (Classification of Surfaces with Boundary) Every connected surface Σ can be
represented in precisely one of the following ways:

(i) As a sphere with k-cuffs, in which χ(Σ) = 2 − k.

(ii) As the connected sum of n-tori with k-cuffs, in which χ(Σ) = 2 − 2n − k.

(iii) As the connected sum of n-projective planes with k-cuffs, in which χ(Σ) = 2 − n − k.

Remark 5.42 It is clear that a topological surface Σ is completely determined by the Euler
characteristic, its orientability and the number of boundary components. It is also clear that
formula (†) for the genus holds for surfaces with boundary, noting that n in the statement of
Theorem 5.41 is the (un-)orientable genus.

5.2 Seifert Surfaces

We can now relate the work done in Section 5 thus far to knot theory, in which we consider
special types of surfaces which also yield some useful knot invariants.

52
Definition 5.43 Let K be a knot and ΣK be a connected orientable surface with singular
boundary component ∂ΣK = K. Then, ΣK is called the Seifert surface of K.

Proposition 5.44 For any knot K, there exists a Seifert surface ΣK .

Sketch of Proof : We leave it to [Dan15] for the details; we shall discuss what is known as Seifert’s
Algorithm, that is the procedure which yields a Seifert surface and indeed this works with any
knot. The algorithm proceeds as follows:

(i) Choose an orientation of K.

(ii) Connect every arc entering a crossing to the adjacent arc leaving the crossing.

(iii) Connect these disks with half-twist bands at each crossing.

Choosing an orientation on one disk and propagating via the half-twist bands to the other disks
will yield a consistent orientation.

Example 5.45 We shall apply Seifert’s Algorithm to the usual trefoil diagram K and appeal to
the Classification of Surfaces with Boundary as to identify it with a more ‘standard’ surface.

Seifert’s Seifert’s
−−−−−−→ −−−−−−→
Algorithm Algorithm

Figure 31: Applying Seifert’s Algorithm to the trefoil.

Given what we have in Figure 31, we can see the Euler characteristic is χ(ΣK ) = 2χ(D) + 3χ(B),
where D is the disk and B is the half-twist band. But now, we know that χ(D) = 1 and we can
see that χ(B) = −1. Therefore, χ(ΣK ) = −1. Furthermore, we know that this surface has one
cuff by definition of Seifert surface. Because it is orientable, it can only be the connected sum
of n-tori with one cuff, by the Classification of Surfaces with Boundary. Substituting the known
values into the equation given in the statement of the theorem, we see that n = 1. Consequently,
the Seifert surface of the trefoil is a punctured torus.

Definition 5.46 Let K be a knot. The genus of K is defined to be the non-negative integer

g(K) = inf{g(ΣK ) : ΣK is a Seifert surface of K}.

53
The genus of a knot is trivially an isotopy invariant since it is a topological invariant. The main
problem comes from determining the genus of a knot; it is not at all trivial to do so. That being
said, one interesting point shall now be mentioned.

Lemma 5.47 Let K be a knot. Then, g(K) = 0 if and only if K is the unknot.

Proof : Well, g(K) = 0 if and only if K is the boundary of a disk.

Therefore, given some complicated diagram of a knot, the genus immediately tells us if that class
of closed curve is indeed knotted. We can refer to Lemma 5.47 to acquire the genus of what can
only be described as our favourite knot.

Example 5.48 By Example 5.45, we see that the genus of that particular Seifert surface of K is
g(ΣK ) = 1. By Lemma 5.47, g(K) ̸= 0 so taking the infimum must give that g(K) = 1.

We have used [Rob15] as a basis for some of this discussion. However, there are a number of
exercises left to the reader, which we will now solve in the next few results as to deepen our
understanding of the genus of a knot.

Proposition 5.49 Let K be a knot and ΣK the corresponding Seifert surface. If ΣK consists of
d disks and b half-twist bands, then χ(ΣK ) = d − b.

Proof : This is a generalisation of the Euler characteristic argument of Example 5.45. Indeed,
χ(ΣK ) = dχ(D) + bχ(B), again where D is a disk and B is a half-twist band. We know already
that χ(D) = 1 and χ(B) = −1.

Corollary 5.50 Let K be a knot with n crossings. Then, g(K) ≤ n/2.

Proof : Let ΣK be the corresponding Seifert surface. Because it has boundary, the number of
disks d ≥ 1, which is to say 0 ≥ 1 − d. Using the formula established for the genus, we know that
g(ΣK ) = 1 − 21 χ(ΣK ) which can be written as g(ΣK ) = 12 (1 − d + n). Hence, using the inequality
we have just established, it must be that g(ΣK ) ≤ 12 n. Minimising the crossing number over all
diagrams of K preserves this bound. As such, we have that g(K) ≤ 12 n.

The next result is rather powerful but a full proof is beyond the scope of our discussion. Despite
this, we fully detail half the argument and sketch the rest.

Theorem 5.51 Let K = K1 #K2 be a knot. Then, g(K) = g(K1 ) + g(K2 ).

Sketch of Proof : We first show the more simple inequality g(K) ≤ g(K1 ) + g(K2 ). Indeed, let
ΣK1 and ΣK2 be Seifert surfaces of the respective knots with minimal genera. We can apply
the more general connected sum to these surfaces, which gives us a surface ΣK1 #K2 whose

54
boundary is K1 #K2 . Indeed, we now proceed with a quick study on the genus of this new
surface. By Lemma 5.34, we have χ(ΣK1 #K2 ) = χ(ΣK1 ) + χ(ΣK2 ) − 1. Then, using the formula
relating the genus to the Euler characteristic as presented in Theorem 5.41, it is clear that
g(ΣK1 #K2 ) = g(ΣK1 ) + g(ΣK2 ). As such,

g(K1 #K2 ) ≤ g(ΣK1 #K2 ) = g(ΣK1 ) + g(ΣK2 ) = g(K1 ) + g(K2 ),

where we use the assumption that the Seifert surfaces for each knot are minimal.

We now sketch the proof to show the more difficult g(K) ≥ g(K1 )+g(K2 ), following the structure
of [Rin14]. Here, we start with ΣK1 #K2 being the Seifert surface of the connected sum of knots
which has minimal genus. This part of the proof hinges on the study of intersecting surfaces in
R3 . We construct a sphere S containing K1 but not K2 ; we should have S intersecting K1 #K2
in exactly two points p and q, say. From here, we consider S ∩ ΣK1 #K2 . It is certainly non-empty
because p, q ∈ S ∩ ΣK1 #K2 . Also, we can construct a path between p and q on this intersection.
The technical part now follows: we can suppose that this path is the only arc in the intersection
and that there are no loops in the intersection; this is achieved by repeatedly modifying S. If we
then ‘cut’ along this arc, we get the disjoint union of connected orientable surfaces with boundary
K1 and K2 respectively, i.e. Seifert surfaces for each knot. Call them ΣK1 and ΣK2 . Then,

g(K1 #K2 ) = g(ΣK1 #K2 ) = g(ΣK1 ) + g(ΣK2 ) ≥ g(K1 ) + g(K2 ),

which completes the proof.

Definition 5.52 Let K be a knot. It is called composite if there exist non-trivial knots K1 and
K2 such that K = K1 #K2 . Otherwise, it is called prime.

Proposition 5.53 Let K be a knot where g(K) = 1. Then, K is prime.

Proof : Suppose to the contrary that K = K1 #K2 for non-trivial knots K1 and K2 . We can use
Theorem 5.51 to see that g(K) = g(K1 ) + g(K2 ) = 1 which implies that either g(K1 ) = 0 or
g(K2 ) = 0, so one of them is trivial, a contradiction.

Corollary 5.54 Any composite knot K has a factorisation into prime knots.

Proof : By definition, there exist non-trivial knots K1 and K2 such that K = K1 #K2 . If each
of K1 and K2 are prime, we are done. Otherwise, proceed to decompose the composite knots in
the connected sum. Note that this process will terminate because Theorem 5.51 guarantees that
K cannot be the connected sum of greater than g(K) non-trivial knots.

55
Corollary 5.55 (of Theorem 2.23) The only invertible knot under connected sum is the unknot.

Proof : Suppose K1 and K2 are arbitrary knots. If K1 #K2 = , then Theorem 5.51 gives that

0 = g( ) = g(K1 ) + g(K2 ).

As such, it must be that g(K1 ) = 0 and g(K2 ) = 0; both K1 and K2 are trivial. This shows the
only way to obtain the unknot through connected sum is via the connected sum of the unknot
with itself. This is why knots only form a commutative monoid under connected sum.

There are other ways to argue for the next result, but we will use the theory of Seifert surfaces
(in particular the idea of the knot genus) to show we can create any number of distinct knots.

Theorem 5.56 There are infinitely many distinct knots.

Proof : Let K be any non-trivial knot and define a sequence (Kn )n∈Z+ of knots whereby Kn is
the connected sum of n-copies of K. Then g(Kn ) = ng(K) by Theorem 5.51. But now, the Kn
are pairwise distinct because g(Kn ) ̸= g(Km ), whenever n ̸= m by the fact that K is non-trivial
and Corollary 5.55 implies that the connected sum of non-trivial knots remains non-trivial.

We conclude this subsection with an example that not only shows how to compute the Seifert
surface of a link, but that a diagram can have more than just the one Seifert surface.

Example 5.57 Consider the link L as drawn in Figure 32.

Figure 32: An alternating diagram of some two-component link L.

Depending on the orientation we prescribe (of which there are four), we can generate different
Seifert surfaces. This is shown considering two possible orientations, one in which each component
is traversed clockwise and one in which they oppose each other. Call the Seifert surfaces ΣL,1
and ΣL,2 respective of the ordering of the figures below.

56
 Seifert’s
−−−−−−→
Algorithm

Figure 33: The Seifert surface ΣL,1 .

Using Proposition 5.49, we see that χ(ΣL,1 ) = 4 − 4 = 0 and, because L is has two components,
g(ΣL,1 ) = 0. We proceed similarly with the second orientation.

Seifert’s
−−−−−−→
Algorithm

Figure 34: The Seifert surface ΣL,2 .

Again by Proposition 5.49, we have χ(ΣL,2 ) = 2 − 4 = −2 and g(ΣL,2 ) = 1. Hence, ΣL,1 ≇ ΣL,2
since the genera do not agree, yet we have ∂ΣL,1 = L = ∂ΣL,2 by construction. Thus, this
example proves that a link diagram can have more than one corresponding Seifert surface.

Remark 5.58 Another approach considered was to attack this section from the point of view of
simplicial homology. Given some n-simplex ∆n as in Definition 5.5 on vertices {p1 , ..., pn+1 }, we
can consider the free Abelian groups Cn which are generated by the n-faces (more specifically,
they are linear combinations over Z of the n-faces); we call the elements of each of these groups
n-chains. This provides a means to define the boundary of a combinatorial surface without

57
referring to homeomorphisms as in Definition 5.1. Indeed, we say that the boundary maps are

n+1
X
∂n : Cn → Cn−1 , ∂n ([p1 , ..., pn+1 ]) = (−1)i [p1 , ..., pi−1 , pi+1 , ..., pn+1 ].
i=1

From here, we can define the k th homology group of some n-simplex as follows:

ker(∂k )
Hk (∆n ) = .
im(∂k+1 )

The intuition here is that these groups can detect any k-dimensional holes in a topological space.
Perhaps this is a little naı̈ve and it is better to think of the k th homology group as detecting
cycles that are not boundary components.

5.3 Seifert Matrices

We will now discuss yet another knot-related matrix. Given a knot K and Seifert surface ΣK ,
we saw that such a surface can be constructed using only disks and half-twist bands. In fact, we
can construct a Seifert surface of a knot using only one disk with (possibly more than half-twist)
bands attached to that disk. For this, we shall loosely follow [Mur07] but prove a number of
things independently.

Example 5.59 Consider Figure 31. We can slide the middle half-twist band along one of the
others, the right one say, so that it ends on the bottom disk which results in a one-twist band.
Then, we can shrink the top disk slightly to a strip which connects the ends of the left and right
half-twist bands. This results in another one-twist band. We see the end result in Figure 35.

Figure 35: A modified version of the Seifert surface of the trefoil.

Definition 5.60 Let L be an oriented link with at least two components, Ci and Cj say. Then,
the linking number of these components lk(Ci , Cj ) is defined as half the sum of the signs of the
crossings between these components.

Example 5.61 Consider our most basic link, the Hopf link as in Figure 1. Then, if we orientate

58
both clockwise, we see that lk(C1 , C2 ) = −1.

Definition 5.62 Let K be a knot and ΣK the modified one-disk Seifert surface as discussed above.
A Seifert matrix of the knot is SK = [lk(xi , x+
j )]i,j , where xi , xj are loops based on the disk which
circumnavigate the ith and j th bands respectively, where x+
j is the push-off of such a loop, i.e.
where we have lifted xj a distance ε away from the surface at each point in the direction of the
normal vector.

Example 5.63 Let us consider two loops on our modified Seifert surface from Example 5.59 (we
will be very well acquainted with this example by the end of this section); call them x1 and x2
and say they travel around the centre of their respective bands. With a bit of careful thought,
we can compute the Seifert matrix as follows:
!
1 1
SK = .
0 1

Remark 5.64 This remark is very much paired with Remark 5.58 in that there is a significant
homological angle that can be taken here; the loops we defined in Definition 5.62 are actually
generators of the first homology group of the Seifert surface, denoted H1 (ΣK ). Again, we shall
not discuss this too much (unfortunately) but we are clearly aware of this deeper approach. We
will however refer to the loops as first homology generators.

Given some knot K, the Seifert matrix SK isn’t remotely invariant; we can choose a different
set of first homology generators from which to get a Seifert matrix. In fact, Example 5.57
demonstrates that the Seifert surface may not always be unique (yes, this was done for a link but
everything we discuss here also extends to links with more than one component as is discussed
in [Cro04]). That aside, Seifert matrices do lead to a number of invariants.

Example 5.65 We again consider the usual trefoil K. We can use Example 5.63 to see that
! ! !
1 1 T 1 0 T 2 1
SK = and SK = which means SK + SK = .
0 1 1 1 1 2

T | = 3 = det(K). This is not a coincidence.

The eagle-eyed reader will notice that | det SK + SK
T |.


Proposition 5.66 Let K be a knot. Then, det(K) = | det SK + SK

Note that the vast majority of the literature uses Proposition 5.66 as a definition so we shall omit
a proof for now. Recall in Section 3 that we saw the Alexander polynomial as a generalisation
of the knot determinant. In a similar vein, we can define the Alexander polynomial in terms of
a Seifert matrix of a knot and it does coincide with that of Definition 3.25.
T .


Definition 5.67 Let K be a knot. Then, the Alexander polynomial is ∆K (t) = det SK − tSK

59
We can now finally prove part (i) of Proposition 3.29 using this alternate definition.

Proof : From our equivalent definition, we can quickly see that

T T



∆K (t) = det SK + tSK = det tSK − SK ,

since the transpose doesn’t affect the determinant. As such, we can remove a factor of t, giving
T − t−1 S
us ∆K (t) = t2g(ΣK ) det SK 2g(ΣK ) ∆ (t−1 ), where g(Σ ) is the genus.


K =t K K

We make one observation: the Seifert matrix SK constructed using ΣK will be of dimension
2g(ΣK ) × 2g(ΣK ); this is justified by Remark 5.58 in that the first homology group will detect
cycles on a surface that aren’t boundary components (also thought of as the number of holes
g(ΣK )) and we can use the formula relating the genus and Euler characteristic along with
Proposition 5.49 with d = 1 to conclude that the number of bands b = 2g(ΣK ).

Definition 5.68 Let M be a symmetric matrix. Then, the signature σ(M ) of M is defined as the
number of positive entries minus the number of negative entries on the leading diagonal of the
diagonalisation of M .

From linear algebra, we know that this is just the difference between the number of positive and
negative eigenvalues of a matrix. It will not be much of a surprise to see how we can relate this
to Seifert matrices and knot theory.

Example 5.69 We proceed with a quick bout of linear algebra; the characteristic polynomial of
T , where K is the usual trefoil, is λ2 − 4λ + 3 which has roots λ = 1 and λ = 3. Hence,
SK + SK
the diagonalisation of the matrix is !
1 0
.
0 3
T ) = 2, since there are no negative eigenvalues.
We can immediately read off that σ(SK + SK
T.
Definition 5.70 Let K be a knot. The signature σ(K) of K is the usual signature of SK + SK

Theorem 5.71 The signature of a knot is an isotopy invariant.

Sketch of Proof : This result is fully proven in our motivating reference [Mur07]. The idea is
that we use so-called S-equivalence of Seifert matrices which corresponds to performing surgery
on the surface ΣK , i.e. adding a handle under certain conditions. The punchline is that Seifert
matrices can be changed through congruence or enlargement; showing that the signature remains
unchanged when one of these occurs will give that it is an isotopy invariant.

We now proceed with proving some exercises left to us in [Liv93].

60
Lemma 5.72 Let K be a knot and K its mirror image. Then, σ(K) = −σ(K).

Proof : Let SK be a Seifert matrix of K. Then, taking mirror images interchanges positive and
negative crossings, meaning that the linking numbers change sign. As such, we have SK = −SK .
Diagonalising will preserve this change of sign, meaning we have DK = −DK for the diagonalised
matrices. Consequently, the number of positive and negative eigenvalues have swapped roles and
so the signature will change sign.

Corollary 5.73 If K is achiral, then σ(K) = 0.

Proof : If K is achiral, then K ≃ K and Theorem 5.71 implies that σ(K) = σ(K). Using this in
Lemma 5.72 means that we have σ(K) = −σ(K), so the only option is to have σ(K) = 0.

Proposition 5.74 Let K1 and K2 be knots. Then, we have that det(K1 #K2 ) = det(K1 ) det(K2 )
and that σ(K1 #K2 ) = σ(K1 ) + σ(K2 ).

Proof : Let ΣK1 and ΣK2 be respective Seifert surfaces on which Seifert matrices will be computed.
Note that the orientable surface formed by joining these via connected sum will give a Seifert
surface which we call ΣK1 #K2 . As such, the Seifert matrix of the connected sum of knots will
have the form !
SK1 0
SK1 #K2 = ,
0 SK2
since the first homology generators of ΣK1 will not interact with those of ΣK2 and vice versa.
As such, we can compute the symmetric matrix formed by adding the transpose, namely
!
T
SK1 + SK 0
T
SK1 #K2 + SK 1 #K2
= 1
T
.
0 SK2 + SK 2

Therefore, the non-trivial entries in the diagonalisation of the above matrix will be those from the
T and S
diagonalisations of SK1 +SK T
1 K2 +SK2 . As such, we can immediately read off the additivity
of the signature, i.e. σ(K
 1 #K2 ) = σ(K1 ) +
 σ(K2 ). As for multiplicity of the determinant, by
T
basic linear algebra, det SK1 #K2 + SK1 #K2 is merely the product of the entries on the leading
diagonal of its diagonalisation. As we already noted, this will consist only of the leading diagonal
entries of the other two matrices, which means we have det(K1 #K2 ) = det(K1 ) det(K2 ).

Corollary 5.75 Let K1 and K2 be knots. Then, ∆K1 #K2 (t) = ∆K1 (t)∆K2 (t).

We conclude this section by proving a somewhat surprising result, which will provide us with
another means to argue why no knot can be 2-colourable.

Theorem 5.76 Let K be a knot. Then, det(K) is odd and σ(K) is even.

61
 T = 1.


Proof : By Proposition 3.29, we know that ∆K (1) = 1, which is to say that det SK − SK
T ≡ S − S T (mod 2) and so taking determinants of both sides
As such, we can see that SK + SK K K
T
gives that det(K) ≡ 1 (mod 2). From this, it must follows that the diagonalisation of SK − SK
contains no zeros on the leading diagonal. Because it is a matrix of even dimension, if there
is an even (or odd) number of positive eigenvalues, there must be an even (or odd) number of
negative eigenvalues. As such, the signature is always even.

Remark 5.77 We gave a reason in Remark 3.9 as to why it is never going to be possible to colour
a knot with only two colours; we have made this notion rigorous. Indeed, Theorem 5.76 implies
that det(K) is never even, so 2 ̸ | det(K) but Proposition 3.17 implies therefore that K is never
2-colourable.

62
6 Braid Theory
The topic of knot theory, although relatively new, is still very deep. As such, there are a number
of interesting related areas that can be discussed. We will look at one of these closely related
disciplines called braid theory from geometric, algebraic and topological viewpoints, after which
we explain why these give equivalent notions of the same thing. We conclude with a constructive
proof of Alexander’s Theorem and explain how to use it in the context of some of our favourite
knots.

Definition 6.1 A braid on n-strings (or an n-braid) consists of n arcs x1 , ..., xn with integer
endpoints 1, ..., n. More specifically, each xi connects a point Pi+ = (i, 0, 1) in the so-called upper
−
plane to a point Pπ(i) = (i, 0, 0) in the so-called lower plane, where π ∈ Sn is a permutation. In
the case that π is trivial, we call it a pure braid on n-strings (or a pure n-braid).

To avoid ‘nasty’ braids, we stipulate that the strands of an n-braid must cross at a finite number
of places and must not intersect, with crossings occurring on different hyperplanes (i.e. different
z-values). As with links, two n-braids σ and σ ′ are equivalent if there exists an ambient isotopy
of R3 taking σ to σ ′ . We draw these as planar drawings, analogous to Definition 2.5.

Example 6.2 Consider the following 4-braid diagram, in which the permutation π = (1)(3)(24).

P1+ P2+ P3+ P4+

P1− P2− P3− P4−

Figure 36: An example of a 4-braid.

6.1 The Braid Group

Naturally, one may consider what would happen if we identify the upper and lower planes of
a braid diagram. This will become clear soon, but first we discuss properties of braids in an
algebraic context. We can use the structure of a group to describe interactions between n-braids.
This will be done as in the work of [Art47] before branching into a more topological approach.

Definition 6.3 The n-braid alphabet is the set A = {σk , σk−1 : 1 ≤ k ≤ n}, where the symbols
are given in Figure 37 below. An n-braid word is the concatenation of the symbols σk and σk−1 ,

63
where the trivial n-braid is denoted en , taken in order from the upper plane to the lower plane.

+ +
Pk+ Pk+1 Pk+ Pk+1

... ... and ... ...

− −
Pk− Pk+1 Pk− Pk+1

Figure 37: The geometric interpretations of σk and σk−1 respectively.

This looks similar to Definition 4.31 and it will come as little surprise that this will give us a
presentation of a group whose underlying set consists of n-braids.

Example 6.4 The word corresponding to the 4-braid σ in Example 6.2 is σ = σ3 σ2−1 σ3 .

Intuitively, the concatenation of these symbols corresponds to identifying the points at the ends
of each braid diagram, that is they are ‘stacked’ onto each other; this is shown in Figure 38.

Figure 38: The join of σ1 and σ2 .

We shall now develop the theory of equivalence of braids. We begin with three important moves,
the latter two of which seem to correspond with the Reidemeister moves R2 and R3 respectively.

Definition 6.5 The Artin moves are the following, applied to arcs of a braid:

(A1) We have σi σj = σj σi whenever |i − j| > 1.

(A2) We have σi σi−1 = e = σi−1 σi .

(A3) We have σi σi+1 σi = σi+1 σi σi+1 .

64
 ≃ and ≃

Figure 39: The geometric interpretations of A2 and A3.

Example 6.6 Consider the 4-braid given by σ = σ2−1 σ1 σ3−1 σ1−1 σ2 . Thus, the Artin moves give

σ ≃ σ2−1 σ1 σ1−1 σ3−1 σ2 ≃ σ2−1 σ3−1 σ2 .

Definition 6.7 Let σ be some n-braid. We can define an equivalence relation ∼ on the braid
diagram, given by Pi+ ∼ Pj− if and only if i = j, to get the link via braid closure Lσ .

Example 6.8 Any section of this paper would be incomplete without our favourite knot so we
will show that, by closing the following 2-braid, we achieve the trefoil.

which closes to ≃

Figure 40: The trefoil via braid closure.

Naturally, we may wonder can we close two different braids to the same link?. The answer is yes
and there are two relatively simple justifications. Indeed, let σ be an n-braid represented by the
word w. Then, Lσ can also be formed from the (n + 1)-braid σ ′ represented by wσn+1 , that is Lσ
and Lσ′ are ambient isotopic. Alternatively, Lσ can be formed from the n-braid σ ′′ represented
by σi wσi−1 , that is Lσ and Lσ′′ are ambient isotopic.

Definition 6.9 Let σ be an n-braid represented by word w. The Markov moves are the above
adaptations of an n-braid, called stabilisation and conjugation, respectively.

We will now state a result by Markov but we do not include the complicated proof; see [MJ36].

Theorem 6.10 (Markov’s Theorem) Two words represent the same link via braid closure if and
only if they can be obtained from each other via a finite sequence of Markov and Artin moves.

65
Example 6.11 Consider a braid and apply the stabilisation move. We clearly see that the links
via braid closures are indeed ambient isotopic (each is a Hopf link); we apply R1 to the coloured
sections in Figure 41 and stretch/shrink them slightly to pass between the diagrams.

Figure 41: Applying the stabilisation Markov move to the closure of a braid.

Definition 6.12 The braid group on n-strands with respect to the Artin presentation is

Bn = ⟨σ1 , ..., σn−1 ; σi σi+1 σi = σi+1 σi σi+1 ∀i, σi σj = σj σi ∀|i − j| > 1⟩.

We see the braid group on n-strands is a group on n − 1 generators with relations A1 and A3.

Proposition 6.13 Artin’s presentation of Bn is equivalent to the geometric interpretation of

defining an equivalence relation identifying the upper and lower planes of braid diagrams.

Proof : Because we stipulate that a braid diagram has crossings for different z-values, it is possible
to identify an n-braid with the joining of multiple diagrams of the form of Figure 37. This is
precisely a concatenation of members of the n-braid alphabet, by Definition 6.3. Conversely,
the identification of the upper and lower planes of multiple braid diagrams will yield an n-braid
word. The fact that different braid diagrams correspond to the same braid differ by the relations
in Definition 6.12 is clearly a result of Markov’s Theorem.

We now develop a more topological viewpoint of braid groups. However, we will also need to
refer to some deeper group-theoretic content in order to achieve this.

Definition 6.14 For a topological space X, the ordered configuration space of X on n points is
Conf n (X) = {(x1 , ..., xn ) ∈ X n : xi ̸= xj ∀i ̸= j}, with subspace topology induced by X n , which
in turn has the product topology induced by X.

Definition 6.15 Let G be a group and X a set. We have a left G-action on X if there exists a
map ρ : G × X → X given by ρg (x) satisfying the following, for all g, h ∈ G and x ∈ X:

ˆ (ρg ◦ ρh )(x) = ρgh (x).

66
 ˆ ρeG (x) = x.

Of course, we could define a right G-action similarly but generally, we just say G acts on X. We
can apply Definition 6.15 to a context involving the ordered configuration space as is now done.

Lemma 6.16 For a topological space X, we have an action of Sn on the set Conf n (X).

Proof : Define the map ρ : Sn × Conf n (X) → Conf n (X) by ρπ ((x1 , ..., xn )) = (xπ(1) , ..., xπ(n) ).
We need only verify the conditions of Definition 6.15. Indeed,

(ρπ1 ◦ ρπ2 )((x1 , ..., xn )) = (x(π1 ◦π2 )(1) , ..., x(π1 ◦π2 )(n) ) = ρπ1 ◦π2 ((x1 , ..., xn )),

noting that π1 ◦ π2 ∈ Sn by closure. As for the second condition, we see that

ρe ((x1 , ..., xn )) = (x1 , ..., xn ),

where e is the trivial permutation.

For some group G acting on a set X, we can define an equivalence relation on X whereby we say
two elements are equivalent precisely when the action maps one to the other. In particular, for
x, y ∈ X, we say x ∼ y if and only if there exists g ∈ G such that ρg (x) = y.

Definition 6.17 Let some group G act on a set X and x ∈ X. The G-orbit of x is the set
OrbG (x) = {ρg (x) : g ∈ G}, i.e. the equivalence class of x under the previous relation.

Example 6.18 Consider the group action in Lemma 6.16 and let (x1 , ..., xn ) ∈ Conf n (X). Then,
OrbSn ((x1 , ..., xn )) = {(xπ(1) , ..., xπ(n) ) : π ∈ Sn }, which is the set of n-tuples containing precisely
x1 , ..., xn in any order.

Definition 6.19 For a topological space X, the unordered configuration space of X on n points is
UConf n (X) = Conf n (X)/Sn , that is the set of Sn -orbits equipped with the quotient topology.

From [Coh13], we get the next definition, which finally ties the group theory we have done back
to the titular concept of Section 6. We then prove it is consistent with that which came before.

Definition 6.20 The braid group on n-strands with respect to the theory of fundamental groups
is Bn = π1 (UConf n (R2 )), where R2 is endowed with the standard Euclidean topology.

Proposition 6.21 The topological interpretation of Bn agrees with the geometric interpretation
of defining an equivalence relation identifying the upper and lower planes of braid diagrams.

Sketch of Proof : Consider a loop in the space UConf n (R2 ). We can see that such a loop will
trace n curves in space-time, that is a subset of R2 × [0, 1], which do not intersect. As such,

67
one recovers an n-braid. Conversely, an n-braid can be thought of in this way because a loop in
UConf n (R2 ) will act as a representative of a class of loops in π1 (UConf n (R2 )).

6.2 Alexander’s Theorem

We finish this section with an important result by Alexander, as is found in [BZ13].

Theorem 6.22 (Alexander’s Theorem) Every link is isotopic to a closed braid.

Sketch of Proof : Let L be the diagram of an oriented link and p ∈ R2 \ L some point through
which we draw the so-called braid-axis (not passing through a crossing of L). Choose a circulation
(orientation) of R2 . If all arc of L agree in orientation with that of R2 , we are done. Otherwise,
suppose that (part of) an arc xi is opposed to the orientation of R2 . Then, we can apply
Reidemeister moves to adjust xi in such a way that its orientation agrees with that of the plane.
Doing this for all such arcs and tracing the arcs that protrude from the braid-axis will yield a
braid, from which its closure is L.

Example 6.23 Consider the best knot to come out of Mathematics (we know which one we’re
talking about). We shall apply the constructive proof of Alexander’s Theorem to demonstrate
how to get a braid from which the trefoil is obtained via braid closure.

Alexander’s
p −−−−−−−→
Theorem

p
Figure 42: Applying Alexander’s Theorem to the trefoil.

With p chosen as in Figure 42 and anti-clockwise circulation, we see that all arcs of K agree with
this orientation of R2 . Thus, we can immediately trace the arcs from the braid-axis, sweeping
said axis around in the orientation given. Notice that our conclusion agrees with Example 6.8.

Example 6.24 Consider the usual diagram of the figure-eight knot as in Figure 5 and choose an
anti-clockwise circulation. Then, with the point p chosen as in Figure 43, we see that an arc is
in opposition to the orientation of R2 . As such, we can manoeuvre it via ambient isotopy so that
the arc is traced anti-clockwise around p.

68
 p p

Figure 43: Adjusting the figure-eight knot so that we can use Alexander’s Theorem.

Then, we need only trace the arcs from the braid-axis as we did in Example 6.23 and this gives
us what we see in Figure 44 below.

p
Alexander’s
−−−−−−−→
Theorem

Figure 44: Applying Alexander’s Theorem to the modified figure-eight knot.

69
7 The Fáry-Milnor Theorem
One of the most fascinating results in the study of knots comes from differential geometry. It
provides a relationship between the knottedness of a curve and its curvature, which is not at all
obvious. We have to develop some theory in the style of Section 1 if we have any hope of proving
the titular result but without further ado, we state the theorem of [Mil50].

Theorem 7.1 (Fáry-Milnor Theorem) Suppose γ : [0, 1] → R3 is a simple regular closed curve.
If the total curvature µ(γ) ≤ 4π, then γ is ambient isotopic to S 1 .

The intuition here is that any knot will wrap around itself at least twice, exhibiting a total
curvature bounded below by 4π, so it seems very plausible. The proof will require some additional
theory first, and we will even prove a slightly lesser result which is nevertheless interesting and
is perhaps the first step on the journey to the Fáry-Milnor Theorem.

7.1 The Tangent Indicatrix

We now introduce another curve-theoretic concept which allows for some lateral movement in
understanding the problem, the use of which will soon become clear.

Definition 7.2 Let γ : [0, 1] → Rn be a regular closed curve. The tangent indicatrix is the curve

γ ′ (t)
Γ : [0, 1] → S n−1 given by Γ(t) = .
||γ ′ (t)||

We shall begin with some intuition on the tangent indicatrix; it will become an immensely useful
tool for the proving of our main theorem.

Proposition 7.3 Let γ : [0, 1] → Rn be a regular closed curve. Then, the total curvature of γ
coincides with the length of the tangent indicatrix, that is µ(γ) = L(Γ).

Proof : First, Lemmata 1.10 and 1.16 imply L(Γ) and µ(γ) are invariant under reparametrisation.
As such, let γ be a unit speed curve without loss of generality, meaning that ||γ ′ (t)|| = 1. Under
this hypothesis, we see that the unit tangent vector is γ ′ (t) and the curvature vector is γ ′′ (t).
Consequently, we see that
Z 1
µ(γ) = ||k(t)|| ||γ ′ (t)|| dt
0
Z 1
= ||γ ′′ (t)|| dt
0
Z 1
= ||Γ′ (t)|| dt
0

70
 = L(Γ),

where we note that Γ′ (t) = γ ′′ (t) by the unit speed assumption.

We can now relate the tangent indicatrix to n-dimensional spheres, following the work of [Kok18]
but expanding on the ideas present in this source by inserting proofs where they are omitted.

Definition 7.4 Let γ : [0, 1] → S n−1 be a regular closed curve. Then γ is a great circle if it is of
the form γ(t) = cos(2πt)e1 + sin(2πt)e2 , where e1 , e2 ∈ S n−1 are orthonormal vectors.

Definition 7.5 Let γ be a great circle on S n−1 and p, q ∈ im(γ); these points disconnect γ into
two arcs, γ1 and γ2 say. The spherical distance between them is dS n−1 (p, q) = min{L(γ1 ), L(γ2 )}.

Lemma 7.6 Let γ : [0, 1] → S n−1 be some closed curve. If L(γ) < 2π, then there exists p ∈ S n−1
such that for every q ∈ im(γ), the spherical distance satisfies dS n−1 (p, q) ≤ L(γ)/4.

Proof : Let x, y ∈ im(γ) split the curve into arcs of equal length and p be the midpoint with
respect to the spherical distance. For every q ∈ im(γ) with dS n−1 (p, q) < π/2, it follows that

2dS n−1 (p, q) ≤ dS n−1 (x, q) + dS n−1 (y, q) ≤ L(γ1 ) + L(γ2 ) = L(γ)/2,

where γ1 and γ2 are arcs between x & q and y & q, respectively. Here, we use the fact that the
spherical distance is a lower bound on the length of a curve between two points on the sphere
(in fact, one can define the spherical distance between two points as the infimum over all curves
with those endpoints).

Remark 7.7 If we change the hypothesis of Lemma 7.6 to have L(Γ) = 2π and that Γ is not the
union of two great semi-circles, then the bound still holds. Furthermore, the spherical distance
is a known metric, meaning it satisfies the Triangle Inequality.

Now, suppose we have two points p, q ∈ S n−1 . If we consider the plane which contains both p
and q as well as the origin, then the angle θ between p and q will satisfy cos(θ) = p · q and the
infimum over all arcs between these points will have length θ. As such, dS n−1 (p, q) = cos−1 (p · q).
This is a simple case of the so-called Haversine Formula.

Lemma 7.8 Perpendicular vectors on S n−1 have spherical distance π/2.

Proof : Suppose that p · q = 0. Then, dS n−1 (p, q) = cos−1 (p · q) = cos−1 (0) = π/2.

Theorem 7.9 (Fenchel-Borsuk Theorem) Let γ : [0, 1] → Rn be a simple regular closed curve.
Then, the total curvature µ(γ) ≥ 2π with equality if and only if γ is plane convex.

71
Proof : Assume to the contrary that L(Γ) < 2π. Then, Lemma 7.6 applies, giving the existence
of p ∈ S n−1 such that for every t ∈ [0, 1], the spherical distance dS n−1 (p, Γ(t)) ≤ L(Γ)/4 < π/2.
Next, we define the function f (t) = p · γ(t). Because γ is a closed curve, we can assume it is
1-periodic, meaning that f (t) is also 1-periodic. By the Extreme Value Theorem, there exists
s ∈ [0, 1] such that f ′ (s) = 0. Therefore, differentiating at t = s gives us

π
p · γ ′ (s) = 0 ⇒ p · Γ(s) = 0 ⇒ dS n−1 (p, Γ(s)) = ,
2

the final implication a consequence of Lemma 7.8. This contradicts the strict inequality just
established. By Proposition 7.3, L(Γ) = µ(γ) ≥ 2π. We shall omit the proof of the equality
case since it requires use of other results – the Hopf Index Theorem and the Whitney-Graustein
Theorem – but we leave it to [Eva10] for the full details.

7.2 Crofton’s Formula

We give one final auxiliary result which leads directly to the proof of the Fáry-Milnor Theorem.
This will be done by following the work of [Che67] for both Crofton’s Formula and the proof,
which we expand upon through more detailed explanations than were provided in this source.

Definition 7.10 Let C : [0, 1] → S 2 be an oriented great circle. We call the unique point P ∈ S 2
a pole if it is positive-normal to the plane containing the great circle which has a norm induced
by the orientation. As such, we may identify the great circle with said unique pole by CP .

Proposition 7.11 (Crofton’s Formula) Let γ : [0, 1] → S 2 be a regularly parametrised curve and
C : [0, 1] → S 2 an oriented great circle. Then, the length of γ will satisfy
ZZ
nγ (C) dA = 4L(γ),

where nγ (C) is the number of intersection points of γ and C and the domain of the integral is
the set of poles of all oriented great circles.

Proof : Without loss of generality, assume γ has unit speed parametrisation and let γ(s) := e1 (s),
which we can complete to a positively-oriented orthonormal basis {e1 (s), e2 (s), e3 (s)} for our
curve γ(s) at each s ∈ [0, L(γ)]. We can use properties of orthogonality and bases to see that

ei · ei = 1 ⇒ e′i · ei = 0.

Furthermore, for every i ̸= j, we also see that

ei · ej = 0 ⇒ e′i · ej = −ei · e′j ,

72
that is we have the skew-symmetric property. Combining these properties gives the following:

e′1 = a2 e2 + a3 e3 ,
e′2 = −a2 e1 + a1 e3 ,
e′3 = −a3 e1 − a1 e2 ,

where the coefficients are real number depending on s, i.e. functions ai : [0, L(γ)] → R. At some
time, choose a point on the curve given by e1 . As such, the oriented great circles that contain
e1 have poles on the equator, which is the e2 e3 -plane, since this is the tangent plane. Therefore,
each pole will be of the form
Pφ,s = cos(φ)e2 + sin(φ)e3 , (∗)

where φ ∈ [0, 2π). As such, we have parametrised the domain of these such poles in terms of
the local coordinates (φ, s). Additionally, note that a22 + a23 = 1, which means, for some angle
θ ∈ [0, 2π) depending on s, we have the following forms for the coefficient functions:

a2 = cos(θ(s)) and a3 = sin(θ(s)).

Using partial differentiation on (∗) in conjunction with the above expressions, the area element
corresponding to these local coordinates is dAφ,s = ||∂φ P × ∂s P || dφds. Explicitly, we get

dAφ,s = |a2 cos(φ) + a3 sin(φ)| dφds,

= | cos(θ) cos(φ) + sin(θ) sin(φ)| dφds
= | cos(φ − θ)| dφds,

by a standard trigonometric identity. Let CP be the oriented great circle with pole Pφ,s . Then,
ZZ ZZ
nγ (CP ) dA = 1 dAφ,s
Z L(γ) Z 2π
= | cos(φ − θ)| dφds
0 0
Z L(γ) Z 2π
= | cos(x)| dxds,
0 0
Z L(γ) Z π/2
= 4 cos(x) dxds
0 0
Z L(γ)
= 4 ds
0

= 4L(γ),

73
where we made the substitution x = φ − θ which is such that dx = dφ.

Remark 7.12 The tangent indicatrix of a closed space curve satisfies the conditions of Crofton’s
Formula, which in conjunction with Proposition 7.3 gives us a means to tackle Theorem 7.1.

Proof of the Fáry-Milnor Theorem : Let γ have unit speed parametrisation and P be the pole of
great circle CP . We wish to analyse the value nΓ (CP ), where Γ is the tangent indicatrix of γ on
S 2 . Assume without loss of generality that the curvature of γ is non-vanishing; this is equivalent
to Γ being regular. We can now make two very useful observations:

ˆ nΓ (CP ) is positive on [0, 2π] × [0, L(γ)].

ˆ nΓ (CP ) is even. The intuition is that if Γ crosses over CP , then it must cross back because
it is closed. The situation where Γ is tangential to CP will occur at an isolated number of
points, meaning the measure of such oriented great circles as in Crofton’s Formula is 0.

Define the so-called height map hP : [0, L(γ)] → R given by hP (s) = γ(s) · P . We can clearly see
that h′P (s) = γ ′ (s) · P , which means any critical point of the height map occurs precisely when
P is orthogonal to some tangent plane of γ, i.e. when the tangent indicatrix intersects CP . As
such, nΓ (CP ) is the number of extrema of the height map. By Proposition 7.3, the hypothesis
of Theorem 7.1 is to say L(Γ) < 4π and so Crofton’s Formula implies that
ZZ
nΓ (CP ) dA < 16π.

Recall the surface area of S 2 is 4π, meaning there exists a pole P where nΓ (CP ) < 4π. If not,
ZZ ZZ
4 dA < 16π ⇒ dA < 4π,

contradicting the area of the sphere. It can only be that nΓ (CP ) = 2. This means hP has two
extrema. We can suppose that P = (0, 0, 1). In this case, since the height map is defined on
a compact set, the Extreme Value Theorem implies there is precisely one maximum and one
minimum point. These points divide our original curve γ into two arcs, along one of which the
vertical component increases and along the other of which it decreases. Hence, taking a plane Σ
orthogonal to P and considering its intersection whilst varying it between the extremal points
yield sets consisting of only two points. Connecting each point via a line segment generates a
surface bounded by γ which is homeomorphic to the closed disk. Therefore, Lemma 5.47 implies
that γ thought of as a knot has genus zero and must be the unknot.

Remark 7.13 We have shown a weaker version of the Fáry-Milnor Theorem, that is µ(γ) < 4π
implies that γ is the unknot. For the equality case, we refer to the original paper [Mil50] by Milnor

74
who proved the stronger result by estimating curves with closed polygons. In this estimation, he
defines the total curvature as the infimum of the polygonal curvature over all possible inscribed
polygons. He then showed that, for the bound we proved above, namely µ(γ) ≥ 4π for a knotted
curve, it is always possible to reduce the total curvature. The implication is that for non-trivial
isotopy classes of closed curves, the infimum 4π is never attained (if it was, it could be reduced,
contradicting the lower bound). These examples are found in [Coo19].

Example 7.14 We see from [Cla02] that any (p, q)-torus knot can be parametrised as follows:

γTp,q (t) = (cos(pt) cos(qt) + 2 cos(pt), sin(pt) cos(qt) + 2 sin(pt), − sin(qt)).

More specifically, let us consider for the final time the knot that has been our friend through the
thick and thin of this paper; I am obviously talking about the trefoil knot which we recall is a
(2, 3)-torus knot. As such, the trefoil has parametrised equation

γT2,3 (t) = (cos(2t) cos(3t) + 2 cos(2t), sin(2t) cos(3t) + 2 sin(2t), − sin(3t)).

We use software to compute the total curvature to be µ(γT2,3 ) ≈ 17.822 > 12.567 ≈ 4π so the
conditions of the Fáry-Milnor Theorem hold, as expected.

Going back to Corollary 3.7 for instance, we know that the trefoil is distinct from the unknot,
but the Fáry-Milnor Theorem does not guarantee this at all – we have only demonstrated the
contrapositive of Theorem 7.1 in Example 7.14.

Example 7.15 Consider the (1, 4)-torus knot, which is ambient isotopic fo S 1 . However, using

γT1,4 (t) = (cos(t) cos(4t) + 2 cos(t), sin(t) cos(4t) + 2 sin(t), − sin(4t)),

we can compute the total curvature to be µ(γT1,4 ) ≈ 24.274 > 4π, so the Fáry-Milnor Theorem
certainly doesn’t give us both a necessary and sufficient condition on the knottedness of a
parametrised curve in terms of its total curvature, since there exist unknots that have total
curvature larger than 4π.

75
8 Summary
The discussion on knot theory is concluded and we now summarise the findings established in
this paper and make comment on the areas we could develop further given the time.

We have seen a number of link invariants, ranging from numerical ones (number of components,
p-colourability, the knot determinant, the genus, the signature) to polynomial invariants (bracket
polynomial, normalised bracket and Jones polynomials) and even topological ones (knot group).
However, we haven’t discussed anything that can be considered a complete knot invariant; there
were always examples showing the limitations of the invariants we have. It would have been
nice to delve into some additional technicality, studying at least one complete invariant, but alas
we have developed sufficient theory to show that many of the knots and links we have used are
indeed distinct.

Arguably the most powerful tool to come of this paper is the fundamental group; we have seen
that it encodes information on the knot group, p-colourability and braid presentations. Its
outreach is surprising and it is relatively easy to gain intuition on this object.

One riveting aspect was the discussion on braid theory. Given more time, it would have been
preferable to flesh out Section 6 more. One possible route would be to use [Fas05] as a basis of
the extended discussion which would have seen us re-discover the Jones polynomial through the
so-called Temperley-Lieb Algebra. It would also have been preferable to elaborate on Remarks
5.58 and 5.64 given more time; homology is an interesting concept in topology and, as we saw on
a rudimentary level, it can even be used to discuss Alexander polynomials and Seifert matrices.
Nevertheless, we included the aforementioned remarks as a way to note that this alternative
route exists.

One of the most interesting results was the Fáry-Milnor Theorem. The fact this requires no real
intuition beyond a knot will be quite twisted is rather impressive; we could prove this theorem
(given the auxiliary results) pretty much at the end of Section 1. That being said, leaving the
statement and its resolution until the end hopefully gives any admirer of differential geometry an
opportunity to finish this paper with a sense of joy, having endured a lengthy discussion mostly
on knot diagrams.

76
References
[Ada04] Colin Adams. The Knot Book. American Mathematical Society, 2004.

[Art47] Emil Artin. Theory of Braids. Annals of Mathematics, 48(1):101–126, 1947.

[Bad16] Nate Bade. Properties of the Alexander Polynomial. Northeastern University, 2016.
URL: https://web.northeastern.edu/beasley/MATH7375/Lecture18.pdf.

[Bignd] Alessandro Bigazzi. The Seifert-van Kamepn Theorem. University of Warwick, n.d.

[Bos19] Anthony Bosman. Knot theory, 2019. URL: https://drive.google.com/file/d/

1fPwXmiYRhR3ZMa_7p8RNP0zBQRzoXFGz/view.

[BZ66] Gerhard Burde and Heiner Zieschang. Eine Kennzeichnung der Torusknoten.
Mathematische Annalene, 167:169–176, 1966. doi:10.1007/BF01362170.

[BZ13] Gerhard Burde and Heiner Zieschang. Knots, volume 5 of De Gruyter Studies in
Mathematics. Walter de Gruyter, 2013.

[Cam08] Peter Cameron. Matrices with Zero Row and Column Sum. Queen Mary University of
London, 2008. URL: http://www.maths.qmul.ac.uk/~pjc/odds/zero.pdf.

[Che67] Shiing Shen Chern. Curves and Surfaces in Euclidean Spaces. Studies in Global
Geometry and Analysis, 4(3):16–56, 1967.

[Cla02] David Clark. Transforming Trigonometric Knot Parameterizations into Rational Knot
Parameterizations. Mount Holyoke College, 2002. URL: https://www.mtholyoke.
edu/~adurfee/reu/02/dc.pdf.

[Coh13] Fred Cohen. Introduction to Configuration Spaces and their Applications. University
of Rochester, 2013. URL: https://www.mimuw.edu.pl/~sjack/prosem/Cohen_
Singapore.final.24.december.2008.pdf.

[Connd] Keith Conrad. Spaces that are Connected but not Path-Connected. Stanford University,
nd. URL: https://kconrad.math.uconn.edu/blurbs/topology/connnotpathconn.
pdf.

[Coo19] John Cook. Total Curvature of a Knot. 2019. URL: https://www.johndcook.com/

blog/2019/10/01/total-curvature-of-a-knot/.

[Cro04] Peter Cromwell. Knots and Links. Cambridge University Press, 2004.

[CSW16] Scott Carter, Daniel Silver, and Susan Williams. Three Dimensions Of Knot Coloring.
2016. URL: https://arxiv.org/pdf/1301.5378.pdf.

77
[Dan15] Supreedee Dangskul. A Construction of Seifert Surfaces by Differential Geometry.
University of Edinburgh, 2015.

[Deh87] Max Dehn. Papers on Group Theory and Topology. Springer-Verlag New York, 1987.

[Eva10] Charles M. Evans. Curves, Knots and Total Curvature. Wake Forest
University, 2010. URL: https://pdfs.semanticscholar.org/d035/
84eab92242313c316a34d19ca74d0f7bbd30.pdf.

[Fas05] Jordan Fassler. Braids, the Artin Group and the Jones Polynomial. University of
California Los Angeles, 2005. URL: https://www.math.ucla.edu/~radko/191.1.
05w/jordan.pdf.

[FMP17] Stefan Friedl, Allison N. Miller, and Mark Powell. Determinants of Amphichiral Knots.
University of Regensburg, 2017. URL: http://www.mathematik.uni-regensburg.
de/friedl/papers/determinant-of-amphichiral-knots-june-19-2017.pdf.

[GM12] Moritz Groth and Ieke Moerdijk. Homotopy Theory Exercise Sheet 1. Radboud
University, 2012. URL: https://www.math.ru.nl/~mgroth/teaching/htpy13/
Exercise01.pdf.

[GP94] Nick Gilbert and Timothy Porter. Knots and Surfaces. Oxford University Press, 1994.

[Jon14] Vaughan Jones. The Jones Polynomial for Dummies. University of California Berkley,
2014. URL: https://math.berkeley.edu/~vfr/jonesakl.pdf.

[Kan86] Taizo Kanenobu. Examples on Polynomial Invariants of Knots and Links.

Mathematische Annalen, 275(42):555–572, 1986. doi:10.1007/BF01459137.

[Kaw96] Akio Kawauchi. A Survey of Knot Theory. Birkhäuser Verlag, 1996.

[Kok18] Gerasim Kokarev. MATH5113M: Advanced Differential Geometry. University of Leeds,

2018.

[Liv93] Charles Livingston. Knot Theory, volume 24. Cambridge University Press, 1993.

[May00] J. Peter May. An Outline Summary of Basic Point Set Topology. University of Chicago,
2000. URL: http://www.math.uchicago.edu/~may/MISC/Topology.pdf.

[McM13] Curtis McMullen. Topology. Harvard University, 2013. URL: http://people.math.

harvard.edu/~ctm/papers/home/text/class/harvard/131/course/course.pdf.

[MI02] Charles Miller III. Combinatorial Group Theory. Heriot-Watt University, 2002. URL:
http://www.macs.hw.ac.uk/~lc45/Teaching/kggt/miller.pdf.

78
[Mil50] John Milnor. On the Total Curvature of Knots. Annals of Mathematics, 52(2):248–257,
1950. doi:10.2307/1969467.

[MJ36] Andrey Andreyevich Markov Jr. On the Free Equivalence of Closed Braids.
Matematicheskii Sbornik, 43(1):73–78, 1936.

[Mur07] Kunia Murasugi. Knot Theory and Its Applications. Springer Science and Business
Media, 2007.

[MZ19] Andreas Müller and Dimiter Zlatanov. Singular Configurations of Mechanisms and
Manipulators, volume 589 of CISM International Centre for Mechanical Sciences.
Springer, 2019.

[Pet07] Carlo Petronio. Combinatorial and Geometric Methods in Topology. 2007. URL:
https://arxiv.org/pdf/0706.4368.pdf.

[Rei27] Kurt Reidemeister. Elementare Begründung der Knotentheorie, volume 5. Springer

Berlin Heidelberg, 1927. doi:10.1007/BF02952507.

[Rin14] Andrea Rincon. Seifert Surfaces and Genus. 2014. URL: https://www.mathi.
uni-heidelberg.de/~lee/Andrea.pdf.

[Rob15] Justin Roberts. Knots Knotes. University of California San Diego, 2015. URL: http:
//math.ucsd.edu/~justin/Papers/knotes.pdf.

[Rog18] John Rognes. Lecture Notes on Topology for MATH3500/4500. University of

Oslo, 2018. URL: https://www.uio.no/studier/emner/matnat/math/MAT4500/
h18/dokumenter/topology.pdf.

[Tho04] Anne Thomas. Covering Spaces. University of Chicago, 2004. URL: http://math.
uchicago.edu/~womp/2004/athomas04.pdf.

[Wil96] Robin J. Wilson. Introduction to Graph Theory. Longman, 4th edition, 1996.

[Zee66] Erik Christopher Zeeman. An Introduction to Topology. University of Warwick, 1966.

79
Appendices

A The Proof of Lemma 3.16

Here, we will explicitly compute the row operations discussed in the proof of Lemma 3.16. Indeed,

 
m11 + 1 · · · m1j + 1 · · · m1n + 1
 .. .. .. 

 . . . 

M + U =  mi1 + 1 · · · mij + 1 · · · min + 1 
 
.. .. ..
 
. . .
 
 
mn1 + 1 · · · mnj + 1 · · · mnn + 1
 
m11 + 1 · · · n · · · m1n + 1
 .. .. .. 

 . . . 

1. 
7−
→ n · · · n2 · · · n 

.. .. ..
 
. . .
 
 
mn1 + 1 · · · n · · · mnn + 1
 
m11 + 1 · · · n · · · m1n + 1
 .. .. .. 

 . . . 

2. 
7−
→ 1 ··· n ··· 1


.. .. ..
 
. . .
 
 
mn1 + 1 · · · n · · · mnn + 1
 
m11 · · · 0 · · · m1n
 . .. .. 
 .. . . 
 
3. 
7−
→ 1 ··· n ··· 1 ,

 .. .. .. 
 
 . . . 
mn1 · · · 0 · · · mnn

where we have applied the following elementary matrix operations numbered as follows:

1. Ri 7→ Ri + Rk1 for every k1 ̸= i and Cj 7→ Cj + Ck2 for every k2 ̸= j.

2. Ri 7→ n1 Ri .

3. Rk 7→ Ri − Rk for every k ̸= i.

80
It is clear that the determinant of the resulting matrix can be written as an expansion by minors
down the j th column, which is all-zero bar one entry, as we wrote in the proof of Lemma 3.16.

B The Proof of Theorem 3.21

We give the linear algebra details as to how we arrived at the conclusions regarding the colouring
matrices of a knot which has had the Reidemeister moves applied to it. First, we see that the
original colouring matrix will be of the form

xn−1 xn
 
 
 
MK =  .
n−1

 −1 −1 

n −1

We apply R1 to the nth arc of a knot diagram K to form KR1 , assuming this arc passes between
the (n − 1)th and nth crossings. This gives us the following:

xn−1 xn xn+1
 
0
 ..

 .

0 
 
MKR1 =

 
n−1 
 −1 0 −1 

−1 0 
 
n 
n+1 0 ··· 0 0 −1 1

xn−1 xn xn+1
 
0
 ..

 .

0 
 
1.
7−
→

 
n−1 
 −1 −1 −1 

−1 0 
 
n 
n+1 0 ··· 0 0 0 1

81
 xn−1 xn xn+1
 
0
 .. 

 . 

0
 
2.
7−
→ ,
 

n−1 
 −1 −1 0 

−1 0
 
n  
n+1 0 ··· 0 0 0 1

where we have applied the following elementary matrix operations numbered as follows:

1. Cn 7→ Cn + Cn+1 .

2. Rn−1 7→ Rn−1 + Rn+1 .

We apply R2 to the (n − 1)th and nth arcs of a knot diagram K to form KR2 , where the nth arc
now passes underneath the (n − 1)th arc. This gives us the following:

xn−1 xn xn+1 xn+2

 
0 0
 .. ..

 . .

 

 0 0 

MKR2 = n−1

 −1 0 −1 0 
 
n 
 −1 0 0 
n+1
 0
 ··· 0 2 0 −1 −1 

n+2 0 ··· 0 2 −1 0 −1

xn−1 xn xn+1 xn+2

 
0 0
 .. ..

 . .

 
 0 0 
1.  
7−
→ n−1

 −1 −1 −1 0 
 
n 
 −1 0 0 

n+1
 0
 ··· 0 2 −1 −1 −1 

n+2 0 ··· 0 2 −1 0 −1

82
 xn−1 xn xn+1 xn+2
 
0 0
 .. .. 

 . . 

 
 0 0 
2.  
7−
→ n−1

 −1 −1 −1 0 

 
n 
 −1 0 0 

n+1
 0
 ··· 0 0 0 −1 0 

n+2 0 ··· 0 2 −1 0 −1

xn−1 xn xn+1 xn+2

 
0 0
 .. .. 

 . . 

 
 0 0 
3.  
7−
→ n−1

 −1 −1 0 0 ,

 
n 
 −1 0 0 

n+1
 0
 ··· 0 0 0 −1 0 

n+2 0 ··· 0 0 0 0 −1

where we have applied the following elementary matrix operations numbered as follows:

1. Cn 7→ Cn + Cn+1 .

2. Rn+1 7→ Rn+1 − Rn+2 .

3. Rn−1 7→ Rn−1 − Rn+1 and Cn−1 7→ Cn − 2Cn+2 and Cn 7→ Cn − Cn+2 .

C Graphs and Combinatorial Surfaces

We provide more information to gain better understanding of the graph-theoretic approach we
took in Remark 5.22 and Lemma 5.23. The graph theory we present is found in [Wil96]. We
conclude this section with an explicit construction of the identification graph and its dual in
relation to the Klein bottle.

Definition C.1 A graph is the pair G = (V, E), where V is a (finite) set of vertices and E is a set
of pairs of vertices we call edges.

Example C.2 Consider the graph G = (V, E) := ({1, 2, 3, 4}, {12, 13, 14, 34}). This means that
we have edges between vertices 1&2, 1&3, 1&4, 3&4. This can be drawn in Figure 45 below.

83
 1 2

4 3

Figure 45: A representation of a graph on four vertices.

This particular graph is connected since there exists a path (a sequence of edges) between any
two vertices. We can see that some vertices are contained at the ends of more than just one edge;
the number of edges where this occurs we call the degree of a vertex. However, we can clearly
see that the graph G′ = (V ∪ {5}, E) is disconnected because there are no edges involving vertex
5, let alone a path between that and any other vertex.

Definition C.3 Let G be a graph. A cycle is a closed path in G containing at least one edge, in
other words a non-trivial path which starts and ends at the same vertex.

Definition C.4 Let G be a graph. It is called a tree if it is connected and has no cycles.

Lemma C.5 A graph on n vertices G is a tree if and only if G has no cycles and n − 1 edges.

Proof : (⇒) If G is a tree, then there are no cycles by Definition C.4. Now, note that removing
any edge will disconnect the tree into two subtrees by the assumption that G contains no cycles.
Here, we proceed by induction on the number of edges, noting that the number of edges across
the two subtrees is one less than the number of vertices. Consequently, there are n − 1 edges to
begin with.

(⇐) Suppose to the contrary that G is not a tree. Then, it must contain (at least) one cycle.
Removing an edge from this cycle will not change connectedness but will give us a connected
graph on n vertices with n − 2 edges, which is clearly a contradiction.

There is a natural way to relate the Euler characteristic to graphs. Suppose G = (V, E) is a
graph where |V | = v and |E| = e. We can define the Euler characteristic of a graph as

χ(G) = v − e.

Remark C.6 A special type of graph is that of a planar graph, meaning it can be embedded in
R2 in such a way that edges do not intersect (they meet only at vertices). For such a graph,
we can define a face to be a region which is enclosed by a series of edges. From this, we

84
recover a more familiar form of the Euler characteristic, but if G is planar, we always have that
χ(G) = v − e + f = 2, were f is the number of faces, including the unbounded face surrounding
the graph. However, there is clearly a restriction here on what types of graph we can assign
faces and we have no reason to impose this. As such, we work in the more general setting laid
out above.

Proposition C.7 Let G be a connected graph. Then, χ(G) ≤ 1 with equality if and only if G is a
tree.

Proof : Without loss of generality, let G be a graph on v vertices and e edges that contains no
vertex of degree one. If this is the case, removing the vertex and corresponding edge will clearly
not affect the Euler characteristic, so we can always reduce our graphs ones of this form. Then,
either deg(x) ≥ 2 for every x ∈ V or there is only one vertex and no edges.

(i) In the first case, we appeal to the Handshaking Lemma [Wil96, page 12]; this result provides
P
us with the following relationship: x∈V deg(x) = 2e. From here, we can conclude that

X X
2v = 2 ≤ deg(x) = 2e,
x∈V x∈V

which rearranges to χ(G) = v − e ≤ 0.

(ii) In the second case, it is immediate that χ(G) = 1.

As such χ(G) ≤ 1. In the case that G is a tree, Lemma C.5 implies that it has v − 1 edges and so
χ(G) = v − (v − 1) = 1. Conversely, if χ(G) = 1, then removing the degree one vertices in turn
from G yields one vertex and no edges as the above working shows; the reverse of this process
is to add one edge and one vertex at each stage, which does not introduce any cycles, so our
original G must have been a tree.

Consider a triangulation of some surface Σ. Then, it is possible to draw the identification graph
G and the identification dual graph G∗ as in Figure 46.

85
 (a) The identification graph. (b) The identification dual graph.

Figure 46: The two graphs that we can relate to surfaces introduced in Remark 5.22.

In the proof of Lemma 5.23, we chose a spanning tree T of G∗ and a subgraph H of G formed
by omitting the edges of G that intersect those of T . We shall now demonstrate this procedure
with the Klein bottle.

Example C.8 We justify why the Klein bottle K 2 is non-orientable using this graph-theoretic
technique. Recall we have seen its planar drawing of K 2 in Figure 15, which we can split into
2-simplices, forming a triangulation. On this triangulation, we can construct the graphs G and
G∗ , as in Figure 47. If we prescribe a circulation on one of the faces and extend this along our
tree T , we will see Figure 28 appearing as a subsurface. In fact, a careful look at the failing of
our attempted orientation will convince us that K 2 = M 2 ∪ M 2 .

Figure 47: A triangulation of the Klein bottle, with shaded region a Möbius strip.

86