# Topological invariants of knots: three routes to the Alexander Polynomial Edward Long

```Topological invariants of knots: three routes to the
Alexander Polynomial
Edward Long
Manchester University
MT4000 Double Project
Supervisor: Grant Walker
May 14, 2005
O time! thou must untangle this, not I;
It is too hard a knot for me to untie!
William Shakespeare
Twelfth Night, Act II, Scene 2
4
Contents
Introduction
7
1 Knots, links and their invariants
1.1 History and knot basics . . . . . . . . .
1.2 Knot diagrams . . . . . . . . . . . . . .
1.3 Reidemeister moves and knot invariants
1.4 Links . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
2 The
2.1
2.2
2.3
Alexander Polynomial: the combinatorial
Calculating the Alexander Polynomial . . . . .
The Alexander polynomial of 31 . . . . . . . . .
The Alexander polynomial of 52 . . . . . . . . .
3 The
3.1
3.2
3.3
invariance of the Alexander
The index of a region . . . . . .
Obtaining the square matrix . .
-equivalent matrices . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
route
19
. . . . . . . . . . . . 19
. . . . . . . . . . . . 21
. . . . . . . . . . . . 22
polynomial
25
. . . . . . . . . . . . . . . . . . . . . 25
. . . . . . . . . . . . . . . . . . . . . 26
. . . . . . . . . . . . . . . . . . . . . 28
4 Seifert surfaces: the geometric route
4.1 Seifert’s algorithm . . . . . . . . . .
4.2 Seifert surfaces for 31 and 52 . . . .
4.3 Seifert matrices . . . . . . . . . . . .
4.4 Simplifying the Seifert surface . . . .
4.5 Forming the Seifert matrix . . . . . .
4.6 The Seifert matrix of 31 . . . . . . .
4.7 The Seifert matrix of 52 . . . . . . .
5 The
5.1
5.2
5.3
5.4
.
.
.
.
9
9
11
13
14
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
fundamental group: the algebraic route
Abstract groups and group presentations . . . . . .
Application to knots . . . . . . . . . . . . . . . . .
Generators and relations of the fundamental group
Presentations of G(31 ) and G(52 ) . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
35
36
37
38
39
41
42
43
.
.
.
.
45
45
46
48
49
6 Labellings of diagrams
51
6.1 Labelling 31 and 52 with group elements . . . . . . . . . . . . . . . . 52
6.2 Invariant properties of labellings . . . . . . . . . . . . . . . . . . . . 54
5
6.3
6.4
6.5
7 The
7.1
7.2
7.3
7.4
7.5
Relation to presentation of the knot group . . . . . . . . . . . . . . . 56
The groups of 31 and 52 generated by labellings . . . . . . . . . . . . 57
Determining a labelling with fewer generators . . . . . . . . . . . . . 58
Fox algorithm
Fox derivatives . . . . . . . . . . . . . . . . . . . . . . . . .
Obtaining the Alexander polynomial . . . . . . . . . . . . .
The Alexander polynomial of 31 using Fox derivatives . . .
The Alexander polynomial of 52 using Fox derivatives . . .
Using Fox derivatives on the alternative group presentation
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
61
61
62
62
63
64
8 Knot theory redeemed
67
8.1 Applications in molecular biology . . . . . . . . . . . . . . . . . . . . 67
8.2 Applications in statistical mechanics . . . . . . . . . . . . . . . . . . 68
A table of knots
69
6
Introduction
This project was originally entirely based around JW Alexander’s 1927 Paper Topological invariants of knots and links, in which the author introduces the Alexander
polynomial. While doing background reading on the subject, however, I became
aware that calculation of the polynomial could be approached from three different
viewpoints: combinatorially, as in Alexander’s original formulation; geometrically,
via constructions called Seifert surfaces and algebraically, by considering the group
of the knot.
In considering these different viewpoints, I have increased the original scope of
the project in order to show—pun intended—how knot theory ties together different areas of mathematics.
Because of the increased breadth of this project, I do not prove all assertions
in detail, but attempts to sketch a proof are made where possible.
I also intend this project to be readable, in the most part, by someone with little
mathematical experience. Because of this, there is extra explanation of mathematical concepts such as groups and topological surfaces; informal descriptions are
used where possible and I have tried to include useful analogies along the way.
To show application of all the theories and to maintain a sense of continuity,
all of the examples in this document feature two knots: 31 and 52 . This is so
that the reader becomes familiar with the knots and so the different mathematical
viewpoints as mentioned above can be more easily compared.
Edward Long
7
8
Chapter 1
invariants
1.1
History and knot basics
Knots are objects that we are all familiar with in everyday life and it comes as a
surprise to some that there is a considerable amount of research devoted to their
study in a mathematical context. The origins of knot theory are linked to physics;
in the latter part of the 19th century a physical theory associated to Lord Kelvin
proposed that the universe was filled with a substance known as ether and it was
the way matter intertwined with this substance that brought about properties of
the chemical elements. It was therefore believed by some that the study of knots
would enlighten physicists as to the deepest mysteries of the universe. Because of
this, there was a drive to tabulate and enumerate as many knots as possible and
to be able to tell, especially in the case of more complex knots, whether two knots
were the same, or indeed whether something was knotted at all or could be unravelled to what is referred to by knot theorists as the unknot. The Scottish physicist
Peter Tait spent years compiling tables of knots in an attempt to produce what he
believed could be a table of chemical elements defined through this theory.
The order in which knots are tabulated is by crossing number, which is the number
of times the curve of the knot crosses itself when the knot is drawn in its simplest
form. Of course, finding the simplest form of the knot is a difficult task in itself
and many knots in Tait’s table were later found to have simpler diagrams or to
be repeats of other knots in the table. Tabulation of the knots also leads to the
traditional notation for a knot in the form Nm , where N is the crossing number
and the knot appears as the mth knot with that crossing number in the knot table.
The knot table lists only what are known as the prime knots. Knots which are
not prime are called composite knots and these are knots that can be decomposed
by cutting through two strands of the ‘string’ and retying the ends to give two
separate nontrivial knots. The trefoil or clover leaf (shown below) is the simplest
nontrivial prime knot and is the simplest to tie. It is the only prime knot with
crossing number 3 and is denoted 31 . Two trefoils can be combined to form a reef
9
knot, which is an example of a composite knot. There are two prime knots with
crossing number 5 and below is shown the knot 52 . These knots will feature in all
of the examples in the rest of this document. A table of all prime knots of crossing
number up to 7 is given at the back of this document.
Figure 1.1: The knots 31 and 52
Unfortunately for Tait, the ethereal theory was discredited in 1897 by experimental
evidence gathered at Case-Western Reserve University by Albert Abraham Michelson and Edward Morley. Added to this the advancements made in atomic theory
(for example Ernest Rutherford’s nuclear model of the atom and Joseph John
Thomson’s discovery of the electron), the physics community soon lost all interest
in knots and study that followed was by pure mathematicians and amateur puzzlesolvers interested in properties of the knots themselves.
So what is a mathematical knot? In the real world we think of a knot as a length
of string or rope wound around itself with the ends fastened so that it cannot
be unravelled again. In the mathematical sense, we prevent the knot from being
unravelled by ‘glueing’ the ends of the string together to form a loop. We also
think of the string as having no thickness (ie. a 1-dimensional object). What we
are left with is a one-dimensional curve embedded in three-dimensional space that
has no self intersections. If working from a geometrical point of view, the curve
can be thought of as made up of a number of straight sections, joined end to end
(a polygonal knot) but these can be made to be so small that we usually think of
what is called a smooth knot.
Definition 1.1 A knot K is a locally flat subset of points homeomorphic to a
circle.
In this definition, the condition of local flatness requires that at each point of the
curve, within some arbitrarily small spherical neighbourhood of the point, the arc of
the knot contained within the sphere is homeomorphic to a diameter of the sphere.
(Intuitively: if we look at the curve of the knot close enough then each section
looks flat). The reason for this constraint is to prevent the occurrence of entities
called wild knots. These are knots that have knotted features at arbitrarily small
scales in a similar way that detail can be found in fractal pictures however much
the picture is zoomed in on. Incidentally, polygonal knots can never be wild so an
alternative way around this problem is to only consider the class of polygonal knots.
10
Note that a knot is usually thought of as having an orientation. That is, we
travel around the curve of the knot in a particular direction.
So, when are two knots the same? The definition above gives describes a knot
as a set of points, but we want to think of two knots as equivalent even if they are
not equivalent sets. Formally, we describe two knots as being equivalent (or of the
same knot type) if they are ambient isotopic.
Definition 1.2 An isotopy is a continuous map h : X × [0, 1] → R3 where each
ht = h | X × {t} is one-to-one. By convention, h0 is the identity map.
Definition 1.3 Two knots K1 and K2 are ambient isotopic if there is an isotopy
h : R3 × [0, 1] → R3 such that h(K1 , 0) = h0 (K1 ) = K1 and h(K1 , 1) = h1 (K1 ) =
K2 .
These definitions allow us to deform our knot in the expected manner: the arcs
can be bent and moved through space without passing through one another, the
entire knot can be shrunk or grown and we are not permitted to pull the knot so
tight that it unknots itself by disappearing into a point.
1.2
Knot diagrams
We now have an adequate definition of a knot in 3-space, but in order to work with
knots more easily we want to be able to represent them in a diagram. Intuitively,
we do this by projecting the knot downwards (casting a ‘shadow’) onto the plane
and marking it in some way to show whether an arc is passing over or under
another arc where there is a crossing. We also require, to prevent confusion, that all
singularities are double points with the approaching arcs having distinct tangents
(see Figure 1.2). A diagram satisfying these conditions is called a regular projection
of the knot.
There are two common conventions for denoting which arc is the overpass and which
the underpass. The easiest to understand is to show the underpass as a broken line
and I shall use this convention in this document. In Alexander’s paper, he marks
the crossing point by placing two dots to the left hand side of the underpass as
you follow the given orientation of the knot (marked by an arrow in the diagram).
This convention is useful in later calculations and in those cases I will redundantly
use both marking styles for clarity.
Notice that the orientation of the knot naturally leads to an orientation of a crossing
point. Imagine that the knot is an electric train set and the trains move in the
direction of the knot’s orientation. The overpass is a bridge over another line. If we
sit on the overpass, facing in the direction of the train, then the trains underneath
will either pass from right to left or from left to right. In the first case we call the
crossing right handed and in the second case we call it left handed.
11
(a) Triple point singularity
(b) Regular projection
(c)
Double
point
with
indistinct
tangents
(d)
Regular
projection
Figure 1.2: Examples of illegal singularities
•
•
(a) Broken line
(b) Alexander’s notation
Figure 1.3: Styles of marking crossing points
(a) Left-handed crossing
(b) Right-handed crossing
Figure 1.4: Crossing points of a diagram
Although using diagrams makes it easier for us to visualise a knot, they introduce a
complication in that projecting the knot from different angles will result in different
diagrams for the knot. Of course we can allow the arcs of the diagrams to be
12
continuously deformed but we have to define new rules because of our constraints
on the types of singularity we allow.
1.3
Reidemeister moves and knot invariants
Luckily, there are only a small number of cases in which deforming the knot results
in illegal singularities. Specifically, there are three moves that can be made in the
neighbourhood of a crossing point which do not alter the knot type. These are
called Reidemeister moves after the German topologist and group theorist Kurt
Reidemeister. It can be proved that whenever two diagrams represent equivalent
knots, there exists a sequence of Reidemeister moves to transform one diagram into
the other.
←→
(a) Before move
(b) After move
Figure 1.5: Reidemeister I move
←→
(a) Before move
(b) After move
Figure 1.6: Reidemeister II move
Reidemeister moves provide a tool for directly showing whether two knots are
equivalent and can be easily applied in the case of simple knots with few crossings.
With larger knots, however, simply trying moves to see if one knot diagram can be
turned into another becomes very inefficient. What is required is an algorithmic
method that can be used to prove, in a finite number of steps, that two knots are
either of the same knot type or of different knot types. For this, we turn to knot
invariants.
13
←→
(a) Before move
(b) After move
Figure 1.7: Reidemeister III move
Definition 1.4 An invariant of an object, with respect to some transformation
of the object, is some quantity or characteristic that does not change under the
transformation.
In the case of knots, an invariant is something that is not changed under ambient
isotopy or, when dealing with diagrams, something that is not changed under any
of the Reidemeister moves. This means that any two equivalent knots will have
the same value for any particular invariant.
Since the number of Reidemeister moves is so small, it makes it relatively simple to prove whether something is an invariant or not: we need only to apply the
three moves in turn and see whether it remains unchanged. This provides the
motivation for most of the proofs in the rest of the document.
1.4
The title of Alexander’s paper on which this project is based mentions another
mathematical object: links. This generalises the idea of a knot to an entity with
more than one component. I will give these only a brief treatment.
Definition 1.5 A link is a finite disjoint union of knots: L = K1 ∪ . . . ∪ Kn .
That is, a link has a number of components, each of which is a knot. We call the
number of components of a link its multiplicity and so any knot is just a link of
multiplicity 1.
The name link suggests that the components of the link are embedded in such
a way that they cannot be pulled apart but, just as a knot need not necessarily
be knotted, a link does not have to be linked. A link that is just n copies of the
unknot sitting in 3-space is called the trivial link of multiplicity n.
link. Another interesting example is the Borromean rings. These are a 3-component
link with the property that if any one of the rings (unknots) is removed then the
14
remaining two rings become unlinked. Historically, this link appears as a heraldic
symbol to represent the notion that the strength of a group of people depends on
each of the individuals and the loss of any one would undermine the strength of
the whole.
Figure 1.9: The Borromean rings
As in the case of knots, link invariants are employed in studying whether two links
are equivalent. Again, two equivalent link diagrams can always be transformed
into one another using the Reidemeister moves and so these can be used to verify
our invariants.
components before moving on to the main focus of this document.
Definition 1.6 Given an oriented link diagram D, choose two components of the
link Di , Dj . For each crossing cr in which Di and Dj cross, set εr = +1 if the link
is right-handed and εr = −1 if the link is left-handed. Then, the linking number
lk(Di , Dj ) of the two components is the sum:
1X
εr
2 r
For example, our Hopf link has two right-handed crossings. So ε1 = ε2 = +1 and
lk(D1 , D2 ) = 21 (1 + 1) = 1. The Whitehead link has two right-handed crossings
and two left-handed crossings so their sum is zero. Hence lk(D1 , D2 ) = 0.
It is simple to demonstrate that the linking number is not changed by Reidemeister moves. First, a Reidemeister I move only creates a crossing point within
one component so this crossing is not counted in the linking number.
15
With a Reidemeister II move, assuming that the two arcs belong to two different link components, we either gain or lose one right-handed crossing and one
left-handed crossing. These cancel each other out leaving the sum unaltered.
R : ε1 = +1
←→
L : ε2 = −1
(b) After move
(a) Before move
Figure 1.10: Reidemeister II move (on oriented link diagram)
With a Reidemeister III move, the diagram before the move has three right-handed
crossings and also has three after the move. Hence, whichever components the arcs
of the diagram belong to, the sum of the values εr is unchanged. This is not quite
enough: the diagram only shows one possible orientation of the arcs. By changing
the orientation of the arcs in the diagram, the handedness of the crossing points
would be different but it is a simple matter to show that the sum of the εr ’s is the
same in each case and so the linking number is unchanged.
R
R
R
←→
R
R
R
(a) Before move
(b) After move
Figure 1.11: Reidemeister III move (on oriented link diagram)
Since the linking number of two link components is unchanged under each of these
and the Whitehead link are different, we know that we cannot apply a sequence
of Reidemeister moves to transform one diagram into the other and so the links
cannot be equivalent.
16
In the remainder of this document, we will restrict our attention to knots, referring
again to links only in the final chapter.
17
18
Chapter 2
The Alexander Polynomial: the
combinatorial route
James Waddell Alexander was an American Mathematician, born in New Jersey
in 1888. He studied Mathematics and Physics at Princeton University and was
awarded his PhD in 1915. During World War I, Alexander contributed his mathematical proficiency by working with the military at a weapons testing site. During
World War II he also worked at the US Air Forces Office of Scientific Research
and Development. Alexander held various professorships at Princeton and was one
of the first members of the Institute for Advanced Study. Being descended from
the president of the Equitable Life Assurance Company, however, he had become
a millionaire through inheritance and did not take salaries while in these positions.
In the 1950s, the political environment under Joseph R McCarthy coupled with
Alexander’s left-wing political views brought him under suspicion and he became
somewhat of a recluse, last appearing in public in 1954. He died in 1971.
Alexander’s main contribution to knot theory was a polynomial invariant that
can be calculated from the diagram of a knot. In overview: each crossing point
of the diagram yields an equation in variables ri . These equations can then be
represented in a matrix from which we can derive a polynomial by operating on
the matrix and taking the determinant. The resulting polynomial in powers of t
must then be normalised, and it is this normalised polynomial which is invariant
for equivalent knots.
This chapter gives an outline of the steps involved in calculating the polynomial
in the manner given in Alexander’s paper and demonstrates the calculation in the
cases of the knots 31 and 52 . All required results are proved in Chapter 3.
2.1
Calculating the Alexander Polynomial
We start our process with an oriented diagram D of a knot K. Let there be v
crossing points of the diagram: c1 , c2 , . . . , cv . Then, by Eulers theorem, it follows
that the arcs of the diagram divide the plane up into v + 2 regions (including the
19
region outside of the knot). We label the regions r0 , r1 , . . . , rv+1 .
We denote the underpasses of the diagram with the second convention mentioned
in Chapter 1: the two dots to the left hand side of the underpass. Now consider
an arbitrary crossing point, ci .
rj
rm
•
•
rk
rl
Figure 2.1: A dotted crossing point
Let the four regions surrounding it be rj , rk , rl and rm where we go around the
crossing point anticlockwise and where the dots lie in regions rj and rk . We can
now define the linear equation:
ci (r) = trj − trk + rl − rm = 0
by taking an alternating sum of the symbols representing the four regions in their
cyclic order and multiplying the dotted regions by t.
Defining such an equation for each of the crossings in the diagram yields a system of v equations in v + 2 variables, which we can then represent in a v × (v + 2)
matrix, M , where each entry is either ±t, ±1 or 0. In the matrix constructed
as just described, each row of the matrix corresponds to a crossing point of the
diagram and each column corresponds to a region. The next step in this process is
to choose two neighbouring regions rp , rq and delete their respective columns vp , vq
from the matrix. Any two neighbouring regions may be chosen and it is proved in
the next chapter that the regions chosen will not affect the resulting invariant.
Deleting columns vp , vq leaves us with a square v × v matrix, Mp,q . The matrix Mp,q is called the Alexander matrix of the knot K. Now let ∆p,q (t) be the
determinant of this square matrix, which will be a polynomial in powers of t with
integer coefficients.
Theorem 2.1 The polynomial ∆p,q (t) obtained as described above, computed from
any other equivalent knot diagram of K differs only by a factor of ±tk for some
integer k.
This theorem is proved in the following chapter.
The fact that the obtained polynomial may differ by a factor of ±tk when computed
20
from a different diagram of the knot suggests that we need some normal form so
that a unique polynomial can be associated to each knot. One possible form is
setting ∆K (t) = ±tn ∆p,q (t) so that the term of lowest degree in ∆K (t) is a positive
constant. This is the required normal form which gives us our knot invariant and
is called the Alexander polynomial.
2.2
The Alexander polynomial of 31
Consider the diagram of the trefoil. Examining crossing c1 we see that regions r2
and r0 are dotted and that the anticlockwise cyclic order is r0 , r3 , r4 , r1 .
r0
r0
r1
c1
r1
c2
•
r4
•
r2
r3
r3
r4
c3
Figure 2.2: Crossing c1 of the trefoil knot
This yields the equation:
c1 (r) = tr0 − tr3 + r4 − r1 = 0
Repeating the same process for crossing points c2 and c3 gives us the remaining
equations:
c2 (r) = tr0 − tr1 + r4 − r2 = 0
c3 (r) = tr0 − tr2 + r4 − r3 = 0
Altogether, we represent these equations in the matrix:


t −1 0 −t 1
M =  t −t −1 0 1 
t 0 −t −1 1
Two neigbouring regions are r3 and r4 so we delete the last two columns of the
matrix and take the determinant of the square matrix M3,4 :
t −1 0 −t −1
∆3,4 (t) = det(M3,4 ) = t −t −1 = t 0 −t
t 0 −t = t3 − t2 + t
t −1
+
t −t
= t(1 − t + t2 )
21
We then take out the factor of t to give the normalised polynomial:
∆K (t) = 1 − t + t2
This is the standard Alexander polynomial for the trefoil knot and so, by Theorem
2.1, calculating ∆K from any other diagram of the trefoil will give the same answer.
2.3
The Alexander polynomial of 52
The process for this knot follows the exact steps as for the trefoil but is made
more complicated by the larger number of crossing points, which lead to a bigger
Alexander matrix. Again, examine the crossing c1 in the diagram. The regions r1
and r2 are dotted and the cyclic order of the regions surrounding the crossing is
r1 , r 2 , r 3 , r 0 .
c1
r0
r1
r0
r2
r3
r1
r4
r5
•
r3
•
r2
r6
Figure 2.3: Crossing c1 of 52
Hence the equation derived is:
c1 (r) = tr1 − tr2 + r3 − r0 = 0
Applying the same process to the crossings c2 , . . . , c5 yields the matrix:



M =


−1
0
−t
−t
−1
t −t 1 0 0 0
t −1 1 −t 0 0
t 0 0 −1 1 0
0 0 1 −1 0 t
0 0 0 −t 1 t
22






In this diagram, we choose neighbouring regions r4 and r5 and delete their columns
to give the square matrix M4,5 . Then:
−1 t −t 1 0 −1 0 1 − t 0 0 0 t −1 1 0 0 t −1 1 0 0
0 0 ∆4,5 (t) = det(M4,5 ) = −t t 0 0 0 = −t t
−t 0 0 1 t −t 0
0
1 t −1 0 0 0 t −1 0
0
0 t = − t −1 1 0 t 0 0 0 +
(1
−
t)
0 0 1 t 0 0 0 t
  0 0 0
= − t 0 1 t
 0 0 t
t 0 0
+ 0 1 t
0 0 t

−t 0 0

+(1 − t) (−t) −t 1 t

−1 0 t
0
−t
−t
−1
t
t
0
0
1
0
1
0
0
0
t
t
t 0 0
+ 0 0 t
0 0 t



−t t 0
+ −t 0 t
−1 0 t



= −t2 + (1 − t)(−t2 + 2t3 )
= −2t3 + 3t3 − 2t4
We then normalise the polynomial by dividing by a factor of −t2 to give the polynomial:
∆K (t) = 2 − 3t2 + 2t2
You can see that the polynomial for 52 is different from the polynomial for 31 .
After the next chapter, we will be able to use this fact to prove that the two knots
are of different types.
23
24
Chapter 3
The invariance of the
Alexander polynomial
At the end of the previous chapter, we demonstrated that a polynomial can be
calculated from a diagram of a knot and that the polynomials calculated from the
knots 31 and 52 are different. In this chapter, we show that a polynomial calculated
in such a way is an invariant of a knot and hence knots with different Alexander
polynomials are necessarily of distinct knot types.
The proof follows that given in Alexander’s 1927 paper Topological invariants of
knots and links, and the argument centres on defining an equivalence between matrices and showing first that equivalent diagrams lead to equivalent matrices and
then that equivalent matrices have determinants which differ only by powers of
±tk . Hence, when normalised, the Alexander polynomial is invariant.
3.1
The index of a region
Alexander assigns an integer to each region of the knot diagram called the index
of the region. These integers are determined by assigning any integer p to a chosen
region and then determining the indices of the remaining regions by setting an
index to p + 1 if we cross into the region from right to left (with respect to the
orientation of the diagram) and to p − 1 if we cross from left to right.
Clearly, since all regions can be reached by crossing over the arcs of the diagram,
this process determines the indices of all the regions of the diagram. Also the process will always produce a consistent indexing.
Consider now the crossing points of the diagram. Clearly, at each point there will
be two regions with the same index, say p, one of index p + 1 and one of index p − 1.
At a left-handed crossing, the first dotted region is of index p and the second
is of index p + 1. At a right-handed crossing, the first dotted region is of index
p + 1 and the second is of index p (recall the cyclic order is anticlockwise).
25
p−1
p+1
p
Figure 3.1: Indexed regions
p+1
p
•
p−1
p
•
•
p
p−1
•
p+1
(a) Left-handed crossing
p
(b) Right-handed crossing
Figure 3.2: Indices around crossing points
Since each region has an index associated to it, when the equations of the diagram
are represented in a matrix each column of the matrix also has a corresponding
index.
3.2
Obtaining the square matrix
Recall that our process of finding the Alexander polynomial involved deleting two
columns from the matrix corresponding to adjacent regions of the diagram. By the
indexing process, any two adjacent regions will have indices differing by 1 and, in
fact, any two columns with indices differing by 1 may be deleted.
Proposition 3.1 If we reduce M to a square matrix Mp,q by deleting two of its
columns of index p and p + 1 then the determinants of the two matrices will differ
only by a factor of ±tk for any two such columns.
To prove this claim, let Rp denote the sum of all columns of index p. Then, since
each row of the matrix has one t, one −t, one 1 and one −1, we have:
X
Rp = 0
p
where 0 denotes the column of zeroes. For example, in the case of the trefoil:
26
0
1
2
1
1
We set the index of r0 to be zero and then apply the indexing rules to find that
r1 , r2 and r3 all have index 1 and r4 has index 2. Then:






−1
t+1
−t
R2 =  −1  , R1 =  t + 1  , R0 =  −t 
−1
t+1
−t
And so R2 + R1 + R0 = 0.
Now multiply each column of index p by a factor t−p . Since each row of the
matrix corresponds to a crossing point; and at each crossing point the indices of
the regions is determined, we have:
cL (r) = t.t−p rj − t.t−(p+1) rk + t−p rl − t−(p−1) rm
in the case of a left-handed crossing and:
cR (r) = t.t−(p+1) rj − t.t−p rk + t−(p−1) rl − t−p rm
in the case of a right-handed crossing. Clearly, in both cases the sum of the
coefficients is zero and so the sum of the columns of the matrix will again be the
zero vector. ie.
X
t−p Rp = 0
p
And so we can combine the above two sums to give:
X
(t−p − 1)Rp = 0
p
Note that since t0 = 1, the terms in R0 in the sum cancel each other out. Hence we
see, from the above sum, that if rj is a region of index p with corresponding column
vj then (t−p − 1)vj is expressible as a linear combination of the other columns with
nonzero index. Also, the coefficients of the columns in the linear combination are
of the form −(t−q − 1) for each column of index q.
Now consider the matrices M0,j and M0,k where the columns vj and vk have indices
p and q respectively. Because of the above result and by properties of determinants,
we see that:
(t−q − 1)∆0,j (t) = ±(t−p − 1)∆0,k (t)
27
Then, since the indices of the regions are determined only up to an additive constant
(we can set the initial p to be any number we choose), if vl and vm are two more
columns of M of index r and s respectively then we obtain the relations:
(tr−q − 1)∆l,j (t) = ±(tr−p − 1)∆l,k (t),
(tq−s − 1)∆k,l (t) = ±(tq−r − 1)∆k,m (t)
which we can combine to give:
∆l,j (t) = ±
(tq−r )(tr−p − 1)
∆k,m (t)
tq−s − 1
Finally, setting p = r + 1 and s = q + 1 we obtain:
∆l,j = ±tq−r ∆k,m
ie. Whenever we remove two columns from the matrix of consecutive index, the
determinant of the resulting matrix differs by ±tq−r , proving the proposition.
3.3
-equivalent matrices
Different diagrams of the same knot will give different matrices when we apply the
procedures outlined in Chapter 2. So we need a way of defining an equivalence
between matrices so that a knot always yields a matrix in the same equivalence
class.
Definition 3.2 Two matrices1 M1 and M2 are -equivalent if it is possible to
transform one into the other by a sequence of the following operations:
(α) Multiplying a row or column by -1
(β) Swapping two rows or columns
(γ) Adding one row or column to another
(δ) Either adding or removing a border where the corner element is 1 and all other
elements are 0, as shown below:




1 0 0 0
a b c

∼ 
 0 a b c 
 d e f  ←→
 0 d e f 
g h i
0 g h i
()Multiplying or dividing a column by t
By properties of determinants, it is simple to verify that the operations α– will
change the determinant of a matrix by at most a factor of ±tk . Hence any two
-equivalent matrices have determinants which differ by at most a factor of ±tk .
Recall that if two diagrams of knots are equivalent, then one diagram can be
transformed into the other via a sequence of Reidemeister moves. We use these to
show that equivalent knots have -equivalent Alexander matrices.
1
With entries which are polynomials in t with integer coefficients
28
Theorem 3.3 If two diagrams D1 , D2 represent knots of the same type then their
square matrices M1 , M2 are -equivalent.
We prove this by looking at the effect of the Reidemeister moves on the matrix of
the diagram.
(I) The diagram begins with regions r1 , r2 , . . . and the formation of a loop creates a new region r∗ and adds a new crossing point to the diagram.
r20
r2
r10
r1
•
r∗
•
(a) Before move
(b) After move
Figure 3.3: Reidemeister I move
Say the Alexander matrix of the knot before the transformation is M . The Reidemeister I move has the effect then of adding a new row and a new column to M.
This new matrix will have the form:





r∗ r10
r20
···
−t −1 t + 1 0 · · · 0
0
..
.
M
0





Since regions r10 and r20 are adjacent we may delete these from the matrix without
affecting the knot invariant. This leaves us with the matrix:
 r∗ · · ·
−t 0 · · · 0
 0

 ..
 .
M1,2
0





We can then divide the r∗ column by t (operation ) and multiply by −1 (operation
α). This leaves us with the matrix:
 r∗ · · ·
1 0···0
 0

 ..
 . M1,2
0
29





Finally, removing the border using operation δ leaves us with M1,2 : a viable square
matrix for the Alexander polynomial of the original knot. Hence ∆K is invariant
under Reidemeister I moves.
Note that the diagram I use for the Reidemeister moves could be oriented or dotted
differently, resulting in slightly different matrices. The methods outlined, however,
will be similar in all cases and a full treatment is omitted for the sake of space.
(II) In this case, we begin with a diagram with regions r1 , r2 , r3 , . . . and the transformation creates another two crossing points, a new region r∗ and splits the region
r2 into r20 and r200 .
r200
•
r1
r10
r3
•
r∗
•
r2
r30
•
r20
(b) After move
(a) Before move
Figure 3.4: Reidemeister II move
The matrix after the transformation will have the form:
0
0
00
···
 r ∗ r 1 r 2 r 2 r3
−t t 0 −1 1 0 · · · 0
 t −t 1 0 −1 0 · · · 0

 0
| |
|

 ..
 .
u v w
M1,2
0
| |
|







In this matrix, u is the column for r1 in the original matrix and the entries for r2
are divided between v and w (since the region has been divided in two). We shall
choose to delete the columns corresponding to regions r10 and r200 . This leaves us
with:


−t 0 1 0 · · · 0
 t 1 −1 0 · · · 0 


 0 |



 ..

 . v

M
1,2
0
|
30
We may divide the first column by
cancel the entries to get:

−1
 0

 0

 ..
 .
0
t and then add the first row to the second to
0 1 0···0
1 0 0···0
|
v
|
M1,2







Then add column r∗ to r30 , multiply column r∗ by −1 and remove the border to
get:


1 0 0···0

 |


 v
M1,2 
|
We may cancel all of the entries of v using multiple applications of operations α–
. This is done by multiplying row 1 of the matrix by the appropriate power of
t, adding row 1 to another row so that the entry in column 1 cancels and then
dividing row 1 by the same power of t so it keeps a 1 in the first entry. This will
then leave us with M1,2 bordered as described in operation δ and we can remove
the border to leave us with M1,2 again.
(III)In the case of a Reidemeister III move, the number of regions is unchanged
but the entries around the crossing points differ.
r1
r2
r4
r20
•
•
•
•
r3
•
r5
r10
•
•
•
r6
•
•
•
r50
r7
r30
r40
•
r70
r60
(b) After move
(a) Before move
Figure 3.5: Reidemeister III move
Before the transformation, the matrix obtained from the diagram will have the
form:
··· 
 r 1 r 2 r 3 r 4 r5 r 6 r 7
−1 t −t 1
0
0 0 0···0
 0 t −1 0 −t 1 0 0 · · · 0 




M =  0 0 t −1 0 −t 1 0 · · · 0 

 |
| 0
|
|
| |




.
 u v ..
w x
y z
X 
|
|
0
|
31
|
|
|
where X is the remaining portion of the matrix defined by the procedure in Chapter
2. After the transformation, the resulting matrix is:
0
 r1
0
 t


M 0 =  −1
 |


 u
|
r20
0
0
t
|
r30
−1
−t
1
0
.
v ..
| 0
r40 r50 r60
0
t −t
−1 0
0
0 −t 0
|
|
|
w
|
x
|
y
|
r70
···
1 0···0
1 0···0
0 0···0
|
z
|
X









Since this example involves such large matrices, we will use a convenient result to
simplify them.
Proposition 3.4 The matrix N obtained by changing the signs of all the negative
elements of M is -equivalent to the matrix M .
To see why this is true, recall the arrangement of the indices of the regions around
each crossing. As you go around the crossing, the indices will alternate between
odd and even. Hence, as entries in the matrix, the odd regions at a crossing will
either both be positive or both negative (and the corresponding even regions will
have the opposite parity). So if we multiply each odd column by −1 then each row
will have only positive entries or only negative entries. Finally we can multiply all
negative rows by −1 to give an entirely positive matrix. Note also that this process
is reversible so we can recover our original matrix M from the positive matrix N .
Applying this result to M gives us:
 r1
1
 0


N=  0
 |


 u0
|
r2
t
t
0
|
r3
t
1
t
0
..
.
r4
1
0
1
|
r5
0
t
0
|
r6
0
1
t
|
r7
···
0 0···0
0 0···0
1 0···0
|
v0
w0 x0 y 0 z 0
| 0 |
| | |
X0









Here w0 –z 0 are the columns w–z with the signs of all negative elements changed
(analogously for X 0 ) and we aim to find a sequence of operations α– that will
transform this matrix into:
0
 r1
0
 t


N0 =  1
 |


 u0
|
r20
0
0
t
|
r30
1
t
1
0
..
.
r40
0
1
0
|
r50
t
0
t
|
r60
t
0
0
|
r70
···
1 0···0
1 0···0
0 0···0
|
v0
w0 x0 y 0 z 0
| 0 |
| | |
32
X0









To save space, in the following calculation I will display only the first three rows
and first seven columns of N but we must take into account the nature of the rest of
the matrix. To avoid changing the entries of w0 –z 0 , only the column r3 is permitted
to be added to the other columns or multiplied by −1 or t. We begin by swapping
rows 1 and 3:




1 t t 1 0 0 0
0 0 t 1 0 t 1
 0 t 1 0 t 1 0  → 0 t 1 0 t 1 0 
0 0 t 1 0 t 1
1 t t 1 0 0 0


0 0
t 1 0
t 1
Times row 2 by −t →  0 −t2 −t 0 −t2 −t 0 
1
t
t 1 0
0 0


0 0
t 1 0
t 1
Add row 1 to row 2 →  0 −t2 0 1 −t2 0 1 
1
t
t 1 0 0 0


0 0 1 1 0
t 1
Divide column 3 by t →  0 −t2 0 1 −t2 0 1 
1
t
1 1 0 0 0


0 0 1 0 0
t 1
Subtract column 3 from column 4 →  0 −t2 0 1 −t2 0 1 
1
t
1 0 0 0 0


0 0
t 0 0
t 1
Times column 3 by t →  0 −t2 0 1 −t2 0 1 
1
t
t 0 0 0 0


0 0 t 0 0
t 1
Add t times row 3 to row 2 →  t 0 t2 1 −t2 0 1 
1 t t 0 0 0 0


0 0 t 0 t t 1
Add column 3 to column 5 →  t 0 t2 1 0 0 1 
1 t t 0 t 0 0


0 0 1 0 t t 1
Divide column 3 by t →  t 0 t 1 0 0 1 
1 t 1 0 t 0 0
This is our required matrix N 0 .
Then, by the result quoted above, we can recover the original matrix M 0 by multiplying the appropriate rows and columns by factors of −1. We can then choose
any two neighbouring regions and delete their columns from the matrices to find
the respective determinants. This shows that the determinant of the Alexander
matrix is not changed by a Reidemeister III move of the type illustrated in the
figure. Again, other orientations of the arcs are possible but the method of proof
33
would be the same as above in all cases2 .
In overview then, we have demonstrated that, given a particular diagram of a
knot, we will always derive the same polynomial ∆K (t), whichever columns we
choose to omit from the matrix. We have also shown that any other diagram from
the same knot will lead to the same polynomial since two diagrams of the same
knot can always be transformed into one another by a series of Reidemeister moves.
By the above proof, whenever we perform a Reidemeister move on a diagram the
resulting square matrix will be -equivalent to the original.
This proof validates the demonstration in the previous chapter that the knots
31 and 52 are topologically distinct. In fact, the Alexander polynomial is different
for all prime knots with eight or fewer crossings. If you allow knots with a larger
number of crossings, however, we begin to find repetitions of the same Alexander
polynomial and so it cannot be used to distinguish between knots with a higher
number of crossings.
It should also be noted that the Alexander polynomial does not distinguish handedness. The trefoil used in the above example is a right-handed trefoil, so called
because all of its crossing points are right handed. There exists a corresponding
left-handed trefoil (a mirror image of the right-handed trefoil) but the Alexander
polynomial for this knot can be shown to be the same as for the right handed case.
2
The proof above is my own. Alexander gives a shorter proof in his paper but it requires a
certain amount of work by the reader to verify. I wanted to give an explicit sequence of operations
α– in my demonstration which may take up a lot of space but is easier for the reader to check.
34
Chapter 4
Seifert surfaces: the geometric
route
So far we have constructed our knots explicitly, but knots also arise naturally in
other situations such as the closure of braids or random walks or, importantly for
this chapter, the boundaries of surfaces. Here, we refer to a surface in the mathematical sense as a two dimensional manifold. A surface may be orientable or
nonorientable and may be with or without boundary. Although a surface has no
thickness, it is intuitively helpful to think of orientable surfaces as those that have
two sides (which we could paint two different colours). A sphere is an example of
an oriented surface without boundary (we can paint the inside red and the outside
blue). A cylinder is another oriented example, but with a boundary consisting of
two circles. A Möbius band (a strip connected end-to-end with one half-twist) has
a boundary and is nonorientable; if we begin to paint in red then we are forced to
cover the entire strip in red. The boundary has one component which looks like a
circle with a twist.
(a) A cylinder
(b) A Möbius band
Figure 4.1: Examples of orientable and nonorientable surfaces
Knots occur as the boundaries of both orientable and nonorientable surfaces. It
is easy to construct a surface from the diagram of the knot by producing what
are known as checkerboard colourings. Begin by colouring a region adjacent to
the ‘outside’ of the diagram in red and then follow over the crossing points to the
opposite region at the crossing. Alternately colour the regions you reach in red or
35
leave them uncoloured. If the entire knot is coloured at the end of this process then
the surface is nonorientable. If there are uncoloured regions left then these can be
coloured in blue; in this case the surface is orientable. (NB. regions that can be
reached by crossing over from the outside of the knot should remain uncoloured
and are interpreted as empty space).
For example, in the case of the trefoil, regions r1 , r2 and r3 are all coloured red and
r4 is left uncoloured. The surface we form looks like a band with three half-twists
and, since only one colour is used, it is nonorientable. In the case of 52 , we start
by colouring r3 and r5 red but r1 and r6 are still uncoloured. We then colour these
blue. The regions r2 and r4 appear as empty space and are left uncoloured. Since
two colours are used, this surface is orientable. It looks like two discs connected by
two bands with one half-twist and one band with three half-twists.
(a) The trefoil
(b) The knot 52
Figure 4.2: Checkerboard colourings of knots
In 1930, Frankl and Pontrjagin proved that for all knots there exists a connected
orientable surface with the knot as its boundary. The German mathematician Herbert Seifert improved on this result in 1934 by giving a separate proof which also
included an algorithm for creating such a surface. An orientable surface with a
given knot as its boundary is now called a Seifert surface and it should be noted
that for one knot there is associated more than one Seifert surface, since the exact
nature of the surface depends on which diagram of the knot is used.
4.1
Seifert’s algorithm
The following algorithm applied to a knot K gives a Seifert surface F for the knot:
Algorithm 4.1
(1) Choose an oriented diagram D for the knot.
(2) Beginning somewhere on the curve of the diagram, trace the orientation until
a crossing point is reached.
(3) Switch to the other arc at the crossing point, still following the orientation.
(4) Repeat (3) until a closed loop is formed.
(5) Repeat (2)-(4) until all arcs of the diagram are traced, leaving a collection of
the circles in the plane.
36
(6) Fill the circles in to form discs and connect them by bands with half-twists that
correspond to the direction of the original crossing point in the knot diagram.
Remarks 4.2
After step (5) we are left with a collection of oriented circles. Some of these will
lie next to each other and some will be nested within each other. When creating
the surface, we imagine the nested circle as lying on a higher level than the outer
circle, with the bands connecting across the two levels. Note that if two circles
in the plane are connected by a band then their boundaries will have opposite
orientations (ie. one clockwise and one anticlockwise) but if a circle is connected
to a circle one level up then their boundaries will have the same orientation. It is
impossible for circles to be directly connected across more than one level.
(a) Circles in the plane
(b) Circles on two levels
Figure 4.3: Raising nested circles to a higher level
Theorem 4.3 Every knot is the boundary of an orientable surface.
We give an intuitive proof using the idea of an orientable surface having two sides
that we can paint red and blue. The algorithm produces a surface with the original
knot as its boundary. Then we can use the property noted in the remarks to colour
the discs with two colours. View the stack of planes from above: for all of the
discs in each level, if the orientation of the circle on its boundary is clockwise then
colour the ‘top’ of the disc red. If the orientation is anticlockwise then colour it
blue. Finally, colour the ‘bottoms’ of the discs with the opposite colour to the tops.
We see that we have consistently coloured the whole surface with two colours, since
if we start on the red side of a disc and move on the same level to another disc then
the half-twist takes us to the bottom of the next disc, which is also red. If we move
up a level, then the half-twist keeps us on the top of the next disc, which is red.
Hence the red sides of the discs form one whole side of the Seifert surface. Similarly
the blue sides form the other side and the surface is orientable as required.
4.2
Seifert surfaces for 31 and 52
If we follow Seifert’s algorithm for 31 then we form first a circle that goes around
the edge of the whole knot and then a second circle inside the first. Hence we fill
the circles in, raise the smaller disc to a higher level and join the two together
with three bands (one for each crossing point). Each disc has a clockwise oriented
37
boundary and each band has one right-handed half-twist.
(a) 31 after the Seifert algorithm
(b) As
colours
a
surface
with
two
Figure 4.4: A Seifert surface for the trefoil
In the case of 52 , tracing the orientation of the knot as described in the algorithm
results in four circles that lie next to each other in the plane. We fill these in
to form discs. The top two are then connected by two bands, each with a righthanded half-twist and the bottom two are part of a chain that link the top two
discs together. (We will later see that structures like this chain can be simplified).
Figure 4.5: A Seifert surface for 52
4.3
Seifert matrices
From the Seifert surface F formed using the algorithm in the previous section, it is
possible to construct a matrix from which we can derive the Alexander polynomial.
The construction of the matrix involves loops in the Seifert surface which form a
basis for a structure called the first homology group, H1 (F ), of the surface. The
theory involved is beyond the scope of this document but I will give a description
of the methods used to form the matrix, and demonstrate that using these methods
on the knots 31 and 52 result in the same Alexander polynomial as calculated in
Chapter 2.
38
4.4
Simplifying the Seifert surface
We are looking for n loops that lie in F that form a basis for H1 (F ). From the
information we get from these loops, we can form a n × n matrix, with each row
and column corresponding to one of the loops.
It can be shown that a Seifert surface made up of a number of discs and bands can
be transformed into a single disc with a number, say n, of bands that connect back
to itself (the final picture rather resembles a 2n-legged octopus with its legs glued
together in pairs). It can also be shown that if we fix some point in that disc and
define n loops by paths which go down the centre of the bands and come back to
the starting point, then those loops form a basis for H1 (F ) and hence are sufficient
for our purposes.
To produce this simplified diagram of the surface we first put all of the discs onto
the bottom level. In doing this we may have to allow some of the connecting bands
to cross each other and it is important to keep track of which passes over which.
To make manipulation of the diagrams easier I have devised my own system of
notation, which I will use in the examples. Each disc is represented by a circle and
a band between discs by a line. In the middle of the line, to show the direction
of the twist I place a box containing a +1 if the original crossing of the knot was
right handed and a −1 if it was left handed. I mark the disc in which we later plan
to fix our base point with an asterisk.
∗
+1
(b) In my notation
(a) Two discs with a right handed twist
Figure 4.6: Notation for Seifert surfaces
If we can find a chain of discs connected by bands that forms a loop from our
base disc and where no disc in the chain is connected to any other disc, then we
can clearly replace the chain by a band with multiple twists. In my notation: we
replace a chain of circles and lines by a line connecting the base circle back to itself
and the number in the box is the sum of all the numbers in the chain. Call this
number the degree of the line.
We also require a way to reduce the number of discs in the surface so that we are
left with only one.
39
+1
+1
∗
+3
+1
(a) A chain of discs
(b) The simplified diagram
Figure 4.7: Suppressing chains of discs
Consider what happens if we have two discs joined by a half twist. If we then
cut down the middle of one disc and along the twist and then treat the two halfdiscs as discs in their own right, we have the base disc joined to two other discs by
bands with a half-twist. But one band will pass over the other, corresponding to
the direction of the original twist. (Note also that the two new bands are twisted
the same way as the original).
Figure 4.8: Splitting one disc into two
This appears to create more discs but notice that if two discs are joined by three
bands, then cutting down the centre of the middle band forms two disc-chains that
we can simplify to twisted bands. Also, if two discs are joined by two bands but
each has a looped band attached to it, we can cut down the centre of the bands
that join the discs together and turn the second disc with its loop into two loops
attached to the first disc.
In my notation, we represent this operation by splitting a disc in two and replacing the straight line by two crossed lines with the same degree. Now, as we
move out from the base disc, if the order of the original line was −1 then the left
line passes over the right. If the order of the original was +1 then the right passes
over the left. Similarly, for larger orders, the lines cross again in the same way.
Remarks 4.4
This transformation does not change the topological type of the surface. Although
it is described as ‘cutting’ the disc, it can equally be viewed as a continuous deformation by pushing the boundary of the disc in and down the twisted band until it
40
+1
∗
+1
Figure 4.9: Splitting discs with my notation
is level with the boundary of the base disc.
I do not give an algorithm here for creating a simplified Seifert surface but the
methods outlined above are sufficient to create simplified surfaces for knots of low
crossing numbers without much difficulty. In particular, the previous examples: 31
and 52 .
4.5
Forming the Seifert matrix
We are given a simplified Seifert surface with a base disc (circle) and n bands
(lines) radiating from it which twist on themselves and around each other. It is
the way in which these bands twist that give us the required information to form
a matrix from which we can derive the Alexander polynomial in a different way.
Since we began with an orientable surface and the simplifying process has not altered the surface topologically, each band must have an even number of half-twists
in it (equivalently: a whole number of full twists). If a band had an odd number
of twists, we would be forced to paint the bottom of the base disc the same colour
as the top and the surface would be nonorientable.
Label the bands attached to the base disc with symbols a1 , . . . , an . We create
an n × n matrix V and label each row and column with a1 , . . . , an . The number
of full twists of each band determines the leading diagonal of the matrix. ie. If the
line representing band a1 has order +6 then the element v11 of V will be 3.
The number of times a band crosses another determines the other elements of
the matrix. ie. As we move out from the base disc, if band a1 crosses over band a2
from left to right m times then the element v12 of V will be m. If it crosses from
right to left m times then v12 will be −m.
If a band passes only underneath another band or if bands do not cross each other
at all then the corresponding elements of the matrix are zero.
Definition 4.5 A matrix with its entries filled in in the manner described above
is called a Seifert matrix for a given Seifert surface F .
Theorem 4.6 If V is a Seifert matrix for a Seifert surface of a knot K, then we
can obtain its Alexander polynomial by the formula:
∆K (t) = det(V − tV T )
41
(V T denotes the transpose of the matrix V )
The proof of the theorem is omitted.
4.6
The Seifert matrix of 31
The Seifert surface we earlier obtained from the diagram of the trefoil consisted of
two discs, one above the other, joined together by three bands, each with a righthanded twist. We level the surface to get two discs next to each other, joined again
by three bands with right-handed twists. We represent this by two circles, joined
by three lines of order +1 each.
+1
∗
+1
+1
Figure 4.10: The surface for the trefoil
Choose the left-hand circle as our base and split the right-hand circle along the
middle band. The middle line is then replaced by two crossed lines of order +1
with the right passing over the left. The two circles joined to the base circle can
then be suppressed and the orders of the lines added to leave two lines looping back
to the base circle of order +2 each. Call the top a1 and the bottom a2 .
+1
∗
+1
+1
+2
a1
+2
a2
∗
+1
Figure 4.11: Obtaining the simplified diagram
Then a1 passes over a2 from right to left and so the matrix element v12 is −1. As
each band has one full positive loop we have v11 = v22 = 1.
Hence the Seifert matrix is:
1 −1
V =
0 1
42
We then calculate the Alexander polynomial using the above theorem:
∆31 (t) = det(V − tV
T)
= det
1 −1
0 1
1 − t −1
= t
1−t
−t
1 0
−1 1
= (1 − t)2 + t
= 1 − t + t2
and we see that we obtain the same polynomial as calculated in Chapter 2.
4.7
The Seifert matrix of 52
The Seifert surface we obtain from 52 is made up of four discs, which can immediately be simplified to two discs joined by two bands with a right-handed half-twist
and one band with three right-handed half-twists. In my notation, we represent
this by two circles joined by three lines of order +1, +1 and +3.
+1
∗
+1
+3
Figure 4.12: The surface for 52
In the same way as in the case of the trefoil, we choose the left-hand circle as the
base and split the right-hand circle along the middle band. Again, we replace the
middle line with crossed lines of order +1 with the right crossing over the left.
Finally, we suppress the circles to give a line of order +2 and a line of order +4.
+1
∗
+1
+1
+2
a1
+4
a2
∗
+3
Figure 4.13: Obtaining the simplified diagram
43
Labelling the bands a1 and a2 as before we form the Seifert matrix:
1 −1
V =
0 2
And the Alexander polynomial can similiarly be found by:
1 −1
1 0
T
−t
∆52 (t) = det(V − tV ) = det
0 2
−1 2
1−t
−1
= t
2 − 2t
= 2(1 − t)2 + t
= 2 − 3t + 2t2
which also agrees with the original calculation.
44
Chapter 5
The fundamental group: the
algebraic route
In Alexander’s paper, he also discusses a second invariant of knots: the group of the
Knot. In this section, instead of assigning a polynomial to each knot, we assign an
algebraic structure. These structures form invariants in that whenever two knots
have the same type, there exists an isomorphism between their groups.
5.1
Abstract groups and group presentations
Informally, a group is a collection of elements with a single operation. The operation is typically addition or multiplication, but transformations of geometric
objects can also form a group such as the group of rotations and reflections of a
regular polygon. (Note that rotation and reflection are not two different group operations: rather, each is an element of the group and the operation is composition,
ie. performing one after the other). A group must also obey certain constraints on
its elements and its operation.
Formally:
Definition 5.1 A group G is a pair (S, ·) of a set together with an binary operation
on the elements of the set. G must be closed under this operation and also satisfy
the following:
(1) G is associative: for all elements a, b, c, we require that (a · b) · c = a · (b · c)
(2) There exists an identity element denoted by 1 such that for all a ∈ G, 1 · a =
a·1=a
(3) Each element a must have an inverse, a−1 , such that a · a−1 = a−1 · a = 1
For example, consider a regular heptagon lying in the plane with its corners numbered. We can rotate the heptagon by multiples of 2π
7 or flip it on any of the axes
passing through one of its corners and its centre. These transformations form a
group. The group identity is leaving the heptagon as it is (we can think of this as
a rotation by zero). Clockwise rotations have anticlockwise rotations as inverses
45
and all flips are self-inverse. That is, if you perform them twice then you return to
1
7
2
7
6
5
4
1
6
3
(a) The original
3
5
2
4
2
4
3
(b) After rotation
clockwise by 2π
7
5
1
6
7
(c) After
flipping
on the axis passing
through corner 2
Figure 5.1: Transformations of a heptagon
In the case of abstract groups, we do not require the elements of the group to represent ‘things’ in the real world (such as numbers, rotations, functions etc.) and
instead simply use formal symbols. The operation on the symbols is concatenation
to form words. For example, the combination of symbols a · b · a−1 · c forms the
word aba−1 c.
A presentation of an abstract group is denoted by:
ha1 , a2 , . . . , an i
which represents the group of all words formed by the symbols a1 , a2 , . . . , an and
their inverses. The symbols ai are called the generators of the group. We can also
impose a further structure on the group with a presentation:
ha1 , a2 , . . . , an | c1 (a), c2 (a), . . . , cm (a)i
where each ci is an identity of the form:
±λN
1 ±λ2
a±λ
i1 ai2 · · · aiN
The identities ci are referred to as the relations of the group and denote words
which are equal to the identity.
For example, the group presentation ha, b, c | a2 , b2 , c2 , abci represents a group with
three generators. All elements of the group are words in a, b, c and their inverses
and a2 = b2 = c2 = abc = 1.
5.2
Application to knots
In order to find a group presentation associated to a knot, we need to find a way of
encoding the geometric properties of a knot in an algebraic structure. We do this
46
x0
Figure 5.2: Paths in R3 − K
by considering loops in the space R3 − K: the complement of the knot.
Fix a base point x0 somewhere in R3 − K and consider the collection of all paths
that begin and end at that point. We can define a composition of two paths by
travelling down the first path and then down the second.
We consider two paths to be equivalent if one may be continuously deformed into
the other within R3 − K (ie. without passing through the knot). In the diagram,
p1 and p01 are equivalent, as are p2 and p02 . This means that any path that does not
pass around an arc of the knot can be shrunk down to the base point (a constant
path). In the diagram, loops p2 and p02 are both equivalent to the constant path.
Formally, the continuous function which maps one path onto an equivalent path is
called a homotopy of paths. Call the equivalence class of all loops equivalent to a
particular loop p a loop class. The composition of two loop classes is well-defined
and is taken to be the class of the composition of the loops.
p2
p02
p1
p01
Figure 5.3: Equivalent paths
If we consider the collection of all loop classes in our space then we see that there
arises a natural inverse for each class. If a composite loop is formed by travelling
down a loop in one direction and then returning to the base point down the same
path but in the other direction then we may continuously deform the composite
path to the base point. If the original loop was p, call the same loop with the
opposite orientation p̄. Then the composite loop pp̄ lies in the loop class of the
constant path at the base point.
Theorem 5.2 The collection of all loop classes in R3 − K forms a group with
composition of loop classes as the group operation.
47
We have seen that each class has an inverse. We take the class of the constant
path at the base point to be the group identity. To make these concepts precise it
is necessary to define explicit homotopies between paths in the space. Using these
homotopies we can also show that composition of loop classes is an associative operation, but I shall omit it here. A full treatment can be found in any introductory
text on algebraic topology.
5.3
Generators and relations of the fundamental group
We can read off a set of generators for the knot group from the diagram of the
knot. Indeed, each region corresponds to a generator of the group. By convention,
we place the base point for the loops in the outside region r0 and so this element
is the group identity. We define the loops corresponding to each region as starting
from some point in r0 , passing ‘above’ the knot diagram, through some region ri
and back ‘underneath’ the diagram (or the loops can be thought of as originating
from the eye of the reader, passing through the region of the knot and returning
from underneath the page). For simplicity, we represent the group element by the
same symbol as the label of the region.
But what if a loop passes through a number of regions? Can we verify that it
is equivalent to a word in the generators r0 , . . . , rv+1 ? This is easy to show. If a
loop r∗ passes through region r1 from top to bottom, then through region r2 from
bottom to top and finally through region r3 from top to bottom before returning
to the base point, we can imagine continuously deforming the loop so that it visits
the base point again in between each region (shown by the dotted lines in the diagram). Hence r∗ is in fact equivalent to the word r1 r2−1 r3 . Clearly, this argument
can be applied to any possible loop around the knot and the reader is invited to
try a number of examples to verify the fact.
r1
r2
r∗
r3
Figure 5.4: A compound loop through three regions
We also need to define a number of relations to give a full presentation of the knot
group. Firstly, we must denote the outside region as the identity of the group,
since any loop staying in that region can be shrunk to the base point. To show
this we place the relation r0 in the presentation of the knot group. Now consider
a crossing point of the diagram with surrounding regions rj , rk , rl , rm in the same
cyclic order as outlined in Chapter 2. For each such crossing point, we also obtain
48
the identity1 :
−1
rj rk−1 rl rm
=1
−1 to the group presentation.
and so we add the relation rj rk−1 rl rm
We see the reasoning for this relation by attempting to draw the loop it represents. The loop passes below the overpass at the crossing and above the underpass.
Hence the entire loop can be pulled free to lie outside the knot. So compound loops
of this type are all equal to the identity.
−1
rj rk−1 rl rm
rj
rm
rk
rl
Figure 5.5: A loop that can be pulled free from the knot
Denote these identities by ci (r), analogously to the equations at the crossing points
in Chapter 2 and we have a group presentation for a knot K with v crossing points
and v + 1 regions:
G(K) = hr0 , r1 , . . . , rv+1 | r0 , c1 (r), . . . , cv (r)i
5.4
Presentations of G(31 ) and G(52 )
Recall that the defining equations of 31 from Chapter 2 are:
c1 (r) = tr0 − tr3 + r4 − r1 = 0
c2 (r) = tr0 − tr1 + r4 − r2 = 0
c3 (r) = tr0 − tr2 + r4 − r3 = 0
From these, we can read off the appropriate relations:
c1 (r) = r0 r3−1 r4 r1−1
c2 (r) = r0 r1−1 r4 r2−1
c3 (r) = r0 r2−1 r4 r3−1
But since r0 is the identity, we can suppress all instances of it and its inverse.
Hence the presentation of the group of the trefoil is:
G(31 ) = hr0 , r1 , r2 , r3 , r4 | r0 , r3−1 r4 r1−1 , r1−1 r4 r2−1 , r2−1 r4 r3−1 i
1
Alexander uses additive notation in his paper but I shall use multiplicative notation to stress
the fact that the knot group is noncommutative
49
In the case of 52 we obtain the relations:
c1 (r) = r1 r2−1 r3 r0−1
c2 (r) = r1 r4−1 r3 r2−1
c3 (r) = r1 r0−1 r5 r4−1
c4 (r) = r6 r0−1 r3 r4−1
c5 (r) = r6 r4−1 r5 r0−1
which, when simplified, give the group presentation:
G(52 ) = hr0 , r1 , r2 , r3 , r4 , r5 , r6 | r0 , r1 r2−1 r3 , r1 r4−1 r3 r2−1 , r1 r5 r4−1 , r6 r3 r4−1 , r6 r4−1 r5 i
50
Chapter 6
Labellings of diagrams
An alternative algebraic treatment of knots is to view the diagram as a collection
of separate arcs, breaking where they pass underneath a crossing, and to label the
arcs of the diagram with elements from a group.
Definition 6.1 A labelling of an oriented knot diagram D with elements from a
group G consists of assigning an element of the group to each arc of the diagram
subject to the following conditions:
(1) Consistency: suppose the overpass at a crossing is labelled with element x, the
underpass before the crossing is labelled with element y and the underpass after the
crossing is labelled with element z.
y
z
z
x
y
x
(a) Left-handed crossing
(b) Right-handed crossing
Figure 6.1: Labelled crossing points
Then the group elements must satisfy:
xzx−1 = y at a right-handed crossing
and xyx−1 = z at a left-handed crossing.
(2) Generation: the collection of labels used in the diagram must generate the whole
group.
A simple example of a group with which we can try to label knot diagrams is
Sn : the group of permutations on n elements. For example, S3 is the group of all
51
permutations of three elements:
1 2 3
1 2 3
1 2 3
s1 =
s2 =
s3 =
1 2 3
2 1 3
3 1 2
s4 =
1 2 3
1 3 2
s5 =
1 2 3
2 3 1
s6 =
1 2 3
3 2 1
where the permutation sends the top number in the bracket to the number underneath it. Notice that s1 is the identity permutation.
Permutations are combined by applying one permutation after another. For example:

 1→2→1
2 → 1 → 3 , hence s2 · s3 = s4
s2 · s3 :

3→3→2
6.1
Labelling 31 and 52 with group elements
It is possible to label 31 with elements from the group S3 as shown. Each element
used in the labelling interchanges two of the elements in {1, 2, 3}. Permutations of
this form are called transpositions.
1 2 3
2 1 3
c2
c1
1 2 3
1 3 2
c3
1 2 3
3 2 1
Figure 6.2: A labelling of 31 with elements of S3
We verify the two necessary conditions for the labelling shown to be legitimate:
(1) Consistency: at the three crossing points we check the consistency condition.
c1 :
c2 :
c3 :
1 2 3
2 1 3
1 2 3
3 2 1
1 2 3
1 3 2
1 2 3
1 3 2
1 2 3
2 1 3
1 2 3
3 2 1
52
1 2 3
2 1 3
−1
1 2 3
3 2 1
−1
1 2 3
1 3 2
−1
1 2 3
3 2 1
1 2 3
1 3 2
1 2 3
2 1 3
=
=
=
(2) We must also show that these elements generate the group. Composing any
transposition with itself gives the identity so we only need to find the remaining
two group elements in terms of these generators. We find that:
1 2 3
1 2 3
1 2 3
=
,
3 1 2
3 2 1
2 1 3
and
1 2 3
2 3 1
=
1 2 3
2 1 3
1 2 3
3 2 1
Hence the labels in the diagram generate the whole group and it is a legitimate
labelling.
52 can be labelled with elements from the group of permutations on seven elements: S7 . For this example it will be convenient to introduce a different form
of notation for our permutations. To denote a transposition, we write the two
transposed numbers between parentheses. For example:
1 2 3 4 5 6 7
can be written more simply as (37)
1 2 7 4 5 6 3
The group elements we use to label the arcs of 52 are all products of transpositions.
We denote products of transpositions by writing a string of parentheses of the form
above. So an element which switches 1 with 6, 2 with 5 and 3 with 4 is denoted
(16)(25)(34). The labelling of 52 is shown below:
(16)(25)(34)
(27)(36)(45)
c1
(13)(47)(56)
c2
c3
c4
(12)(37)(46)
(15)(24)(67)
c5
Figure 6.3: A labelling for 52
It remains to verify that these elements satisfy the conditions of consistency and
generation. Recall that a transposition is self-inverse. If a product of transpositions
involves distinct elements then it is also self-inverse. This observation will help in
verifying consistency.
53
(1) Checking all crossing points (see diagram):
c1 : (27)(36)(45) · (16)(25)(34) · (27)(36)(45) = (13)(47)(56)
c2 : (13)(47)(56) · (27)(36)(45) · (13)(47)(56) = (15)(24)(67)
c3 : (16)(25)(34) · (13)(47)(56) · (16)(25)(34) = (12)(37)(46)
c4 : (15)(24)(67) · (12)(27)(46) · (15)(24)(67) = (27)(36)(45)
c5 : (12)(37)(46) · (15)(24)(67) · (12)(37)(46) = (16)(25)(34)
(2) The elements of S7 chosen in this labelling do not generate the entire group, in
fact they generate a subgroup of S7 that we looked at in the previous chapter: the
group of transformations of a regular heptagon. Each group element describes flipping the heptagon along a particular axis. For example, the element (13)(47)(56)
describes the flip along the axis that passes through corner number 2 of the heptagon, shown in the figure in Chapter 5. Similarly, the other group elements are
flips along axes which pass through corners 1,3,5 and 7.
The composition (12)(37)(46) · (27)(36)(45) gives us the permutation:
1 2 3 4 5 6 7
7 1 2 3 4 5 6
This is a clockwise rotation of the heptagon by 2π
7 . Composing this rotation with
itself gives us all the rotations of the group. The only remaining group elements
we need now are the flips in axes passing through corners 4 and 6. For corner 4,
we can do this by first rotating the heptagon by 6π
7 to put corner 1 in 4’s position,
then reflecting in the axis through corner 1 and finally using the inverse rotation
to return 4 to its starting place. The process is analogous for the flip at corner 6.
Hence this is also a legitimate knot labelling.
6.2
Invariant properties of labellings
These diagram labellings also form a knot invariant.
Theorem 6.2 If a labelling exists for a diagram D1 of a knot K, then a labelling
exists with elements of the same group for any other diagram D2 of K.
The proof of this involves performing Reidemeister moves on a labelled diagram
and seeing whether the labelling can be reconstructed with elements from the
same group after the move. Because the consistency condition varies depending on
whether a crossing is right or left handed, there are a fairly large number of cases to
consider. We will examine two of these as the treatment of the other cases is similar.
First, consider a Reidemeister II move in which both arcs of the diagram are oriented from the bottom of the diagram to the top. The left arc is labelled with
54
element x and the right arc is labelled with element y. After the move, a lefthanded crossing point and a right-handed crossing point are formed. Both crossing
points force the same labelling on the new arc by the consistency condition. We
label it with the element yxy −1 . Note that if the original labelling generated the
group then so does the new labelling since x and y still appear and no new generators are included.
x
x
y
yxy −1
y
x
(a) Before move
(b) After move
Figure 6.4: Reidemeister II move with arc labels
For another case, consider a Reidemeister III move with the arcs oriented as shown
in the diagram. Labelling the arcs x, y and z as shown forces the labelling of the
remaining arcs of the diagram. The arc leaving the top left of the diagram has
label x−1 y −1 zyx and the arc leaving the top right has label x−1 yx. Now, if we
label the same arcs with x, y and z after the Reidemeister move, we see that there
is a new arc in the middle of the diagram with label y −1 zy but the arcs leaving the
top left and top right of the diagram end up with the same labels as before. Again,
we use the same set of generators so this new labelling satisfies both consistency
and generation and the labels come from the same group.
x−1 y −1 zyx
x−1 y −1 zyx
x−1 yx
x−1 zx
x−1 yx
x
y −1 zy
x
y
z
y
(a) Before move
z
(b) After move
Figure 6.5: Reidemeister III move with arc labels
55
6.3
Relation to presentation of the knot group
We can use the conditions for a labelling of a knot diagram to give an alternative
presentation of the knot group from that given in Chapter 5. Again, treat the
group elements which label the diagram as formal symbols and assign one symbol
to each arc of the knot diagram. Then, treating these symbols as the generators
of the group, clearly the generation condition of the labelling holds. We enforce
the consistency condition by using the equations at the crossing points to form the
relations of the group presentation.
For example, if a relation at a crossing is:
xyx−1 = z
then we insert the relation xyx−1 z −1 in the presentation of the group.
In fact, it is simple to demonstrate that these two group presentations lead to
isomorphic groups.
Consider a right-handed crossing point. The regions surrounding the point are
rj , rk , rl and rm . The overpass is labelled with symbol x and on the underpass,
symbol y labels the back end and symbol z labels the front end.
Then, consider x representing a loop, as described in Chapter 5, which passes
around the arc it labels. Note that two such loops are possible: one in each direction. We will take x to be the loop that goes clockwise around the arc as you look
along the direction of orientation. (To help visualise: if the right hand grips the arc
with the thumb pointing in the direction of orientation then the loop x is oriented
in the same sense as the direction the fingers curl around). Similarly define loops
for y and z.
rj
x = rk rj−1
y
rm
x
rk
rl
Figure 6.6: A loop around an arc of the diagram
In our diagram, we see that x can be expressed in terms of rj and rk . ie. x = rk rj−1 .
But note that the relation corresponding to this crossing point gives us an alternative expression:
−1
−1
rj rk−1 rl rm
= 1 ⇔ rk rj−1 = rl rm
56
−1 . Similarly, y = r r −1 and z = r r −1 . With
Hence x can also be expressed as rl rm
m j
l k
these new definitions, we see that the consistency relation still holds since:
−1
xzx−1 = (rl rm
)(rm rj−1 )(rj rk−1 ) = rl rk−1 = y
At a left-handed crossing, the x arc goes in the other direction and so, to preserve
the orientation of the loops, we have that x = rj rk−1 = rm rl−1 .
The expressions for y and z are the same as in the right-handed case. In this
case, then, the consistency relation is:
xyx−1 = (rm rl−1 )(rl rk−1 )(rk rj−1 ) = rm rj−1 = z
as required.
So considering the knot group as generated by loops around the arcs instead of
loops through the regions provides a different generating set but with equivalent
relations to the original presentation. Note that the relation r0 does not appear in
this second presentation as the identity element is the symbolic identity, 1, which
is not an arc label. When using the regions of the diagram as generators for the
group, the group presentation will have v + 2 generators and v + 1 relations (one
for each crossing point and one to define the identity element). When we use arc
labellings, the group presentation will have v generators and v relations. So using
labellings leads to a simpler presentation of the knot group than using the original
definition. We will see that the presentation of the group using labellings can also
be further simplified.
Note that in the case of a group presentation generated by labellings, one of the
group relations can always be deduced from the others (or if there are only two
group relations then they are equivalent). Hence we can always omit one of the
relations without losing any information about the group. This means that a presentation from a labelling will really have v generators and v − 1 relations. So in
both cases there is one more generator than relation. We state this by saying the
group has defect 1, and this is in fact a property of all knot groups.
6.4
The groups of 31 and 52 generated by labellings
In the case of the trefoil, we label the three arcs of the diagram with symbols x, y
and z.
Then the three crossing points require that zxz −1 = y, xyx−1 = z and yzy −1 = x.
Hence the presentation of the group with respect to this particular labelling is:
G(31 ) = hx, y, z | zxz −1 y −1 , xyx−1 z −1 , yzy −1 x−1 i
In the case of 52 , the diagram is broken into five arcs which we label u, v, x, y, z.
57
x
c2
c1
c3
z
y
Figure 6.7: Labelling the arcs of 31
c1
u
v
x
c2
c3
c4
y
z
c5
Figure 6.8: Labelling the arcs of 52
At crossing c1 the consistency condition requires that vuv −1 = x. Hence we obtain
the relation vuv −1 x−1 in the group presentation. By considering the condition at
the remaining crossing points we see that the group presentation is:
G(52 ) = hu, v, x, y, z | vuv −1 x−1 , xvx−1 z −1 , uxu−1 y −1 , zyz −1 v −1 , yzy −1 u−1 i
6.5
Determining a labelling with fewer generators
It is not necessary to use a generator for each arc of the diagram. Indeed, if we
start with a smaller number of generators then we can begin at one point in the
diagram and use the consistency condition to determine the labels of the arcs each
time we come to a crossing point.
For example, in the case of the trefoil labelling two of the arcs with symbols x
and y forces the third arc to be labelled with xyx−1 .
We then simply require the consistency condition to be met at the remaining two
crossing points. ie.
c2 : (xyx−1 )(x)(xy −1 x−1 ) = y
c2 : (y)(xyx−1 )(y −1 ) = x
So our group representation is:
G(31 ) = hx, y | xyxy −1 x−1 y −1 , yxyx−1 y −1 x−1 i
58
x
c2
c1
xyx−1
c3
y
Figure 6.9: Determining a labelling of 31 with generators x,y
It has only two generators and two relations and so is a simpler presentation of the
group of 31 .
In the case of 52 , if we label the top two arcs of the diagram with symbols x
and y then the consistency condition at crossing points c1 , c2 and c3 determines
the labelling of all the remaining arcs in the diagram as shown.
c1
x
y
yxy −1
c2
c3
c4
xyxy −1 x−1
yxyx−1 y −1
c5
Figure 6.10: Determining a labelling of 52 with generators x,y
Once all arcs have been labelled, we obtain the relations for our group presentation
from crossing points c4 and c5 which require:
c4 : (yxyx−1 y −1 )(xyxy −1 x−1 )(yxy −1 x−1 y −1 ) = y
c5 : (xyxy −1 x−1 )(yxyx−1 y −1 )(xyx−1 y −1 x−1 ) = x
These lead to the group presentation:
xyx−1 y −1 xyxy −1 x−1 yxy −1 x−1 y −1 ,
G(52 ) = x, y yxy −1 x−1 yxyx−1 y −1 xyx−1 y −1 x−1
59
60
Chapter 7
The Fox algorithm
In this final chapter, we see how the presentation of a knot group can be related
back to the central theme of this document, namely the Alexander polynomial.
This is via a process developed by the American mathematician RH Fox in which
the Alexander polynomial can be recovered from a presentation of a knot group
using a form of calculus called Fox derivatives.
7.1
Fox derivatives
Fox derivatives are a means of defining a partial derivative for a monomial in
noncommuting symbols, such as the words which form the relations in our group
presentation. It should be noted that the derivative formed in this way cannot be
treated as a word in the same sense as before, but will be a formal sum of words
in the same symbols.
Definition 7.1 Suppose w1 and w2 are words in symbols x1 , . . . , xi , xj , . . . , xn and
their inverses. Then we find the Fox derivative of the words using the following
rules:
∂
(1) ∂x
(xi ) = 1
i
∂
(2) ∂xi (xj ) = 0 if i 6= j
∂
(3) ∂x
(1) = 0
i
−1
∂
(4) ∂xi (x−1
i ) = −xi
∂
∂
∂
(5) ∂x
(w1 · w2 ) = ∂x
(w1 ) + w1 · ∂x
(w2 )
i
i
i
The general method is to differentiate the left-hand symbol at each stage and use
rule (5) to obtain a sum of terms until the end of the word is reached.
For example, consider the word xyxy −1 x−1 y −1 . Then, applying rules (1) and (5)
we see that:
∂
∂
(xyxy −1 x−1 y −1 ) = 1 + x ·
(yxy −1 x−1 y −1 )
∂x
∂x
Applying rules (2) and (5):
∂
∂
(yxy −1 x−1 y −1 ) = 0 + y ·
(xy −1 x−1 y −1 )
∂x
∂x
61
Applying (1) and (5) again, twice:
∂
∂ −1 −1
(xy −1 x−1 y −1 ) = 1 + x · (0 + y −1 ·
(x y ))
∂x
∂x
Finally, rules (4) and (5) give:
∂ −1 −1
(x y ) = −x−1
∂x
Combining the above results we find the Fox derivative of the word:
∂
−1 −1 −1
∂x (xyxy x y )
= 1 + x(0 + y(1 + x(0 + y −1 (−x−1 ))))
= 1 + xy − xyxy −1 x−1
With a little practice, Fox derivatives can be calculated relatively easily and, in a
similar manner, we find that the derivative of the above word with respect to y is
x − xyxy −1 − xyxy −1 x−1 y −1 .
7.2
Obtaining the Alexander polynomial
Given a presentation of a knot group, we will have n words acting as the relations of
the group in n variables. Then, using Fox’s calculus we may calculate the derivative
of each word with respect to each variable in turn and represent their values in an
n × n matrix, called the Jacobian of the group presentation:
 ∂w1 ∂w1

1
· · · ∂w
∂x1
∂x2
∂xn
∂wi

..
.. 
..
=  ...
J=
.
.
. 
∂xj
∂wn
∂wn
∂wn
· · · ∂xn
∂x1
∂x2
We then delete any one row and any one column from the matrix and substitute t
for all the variables. For example, for the derivative calculated above we substitute
1 + xy − xyxy −1 x−1 with 1 + t2 − t3 . . . t−1 , which simplifies to 1 − t + t2 . (Recall
that this is the Alexander polynomial for the trefoil: this same calculation will be
used in the example later in the chapter).
The matrix we are left with is equivalent to the Alexander matrix of Chapter
2 and hence taking its determinant and normalising the resulting polynomial as
before gives us back the Alexander polynomial. I shall not prove this result.
7.3
The Alexander polynomial of 31 using Fox derivatives
We use the presentation of the group using arc labellings as discussed in Chapter
6. The trefoil has three arcs which we label x, y and z. Recall that the consistency
condition at the crossing points of the diagram gives us the relations:
w1 = zxz −1 y −1
62
w2 = xyx−1 z −1
w3 = yzy −1 x−1
∂w2
−1 and ∂w3 =
1
Then, using the Fox calculus we find that ∂w
∂x = z, ∂x = 1 − xyx
∂x
−yzy −1 x−1 By similar calculations for w2 and w3 we find the Jacobian matrix:


z
1 − xyx−1 −yzy −1 x−1
x
1 − yzy −1 
J(31 ) =  −zxz −1 y −1
−1
−1
−1
1 − zxz
−xyx z
y
Deleting the last row and column and substituting t for x, y and z we get:
t 1−t
M=
−1
t
And then clearly det(M ) = 1 − t + t2 = ∆31 (t) as required.
7.4
The Alexander polynomial of 52 using Fox derivatives
From Chapter 6, the relations in the group presentation of 52 gives us the words
w1 , . . . , w5 in variables u, v, w, x, y and z:
w1 = vuv −1 x−1
w2 = xvx−1 z −1
w3 = uxu−1 y −1
w4 = zyz −1 v −1 ,
w5 = yzy −1 u−1
Then the Fox derivatives of w1 are:
∂w1
= v,
∂u
∂w1
= 1 − vuv −1 ,
∂v
∂w1
= −vuv −1 x−1 ,
∂x
∂w1
∂w1
=
=0
∂y
∂z
A series of similar calculations for the words w2 , . . . , w5 gives the Jacobian matrix:


v
1 − vuv −1 −vuv −1 x−1
0
0

0
x
1 − xvx−1
0
−xvx−1 z −1 


−1
−1 y −1

1
−
uxu
0
u
−uxu
0
J(52 ) = 


−1
−1
−1


0
−zyz v
0
z
1 − zyz
−1
−1
−1
−yzy u
0
0
1 − yzy
y
We delete the final row and column and map all variables to t to get the matrix:


t
1 − t −1
0
 0
t
1−t 0 

N =
 1−t
0
t
−1 
0
−1
0
t
63
Finally:
t
1 − t −1
0 0
t
1 − t 0 det(N ) = 0
t
−1 1−t
0
−1
0
t 0
1
0
−t 0
t
1 − t 0 = 0
t
−1 1−t
t
1 − t −1
0 0
0
1 − t 0 t
1−t
t
−1 + t 1 − t
0
t
= − 1 − t
t
t
−1
0 1 − t −1
1 − t −1 1−t t
2
= (1 − t) − t t
0
t
−1
0
+ t(1 − t) 1 − t
t
1−t
= t(1 − t) − t2 (t − 1) − t2 + t(1 − t)3
= 2t − 3t2 + 2t3
Which normalises to 2 − 3t + 2t2 giving the required Alexander polynomial for 52 .
7.5
Using Fox derivatives on the alternative group presentation
Recall that in Chapter 6 we created an alternative group presentation for both 31
and 52 that had only two generators in each case. The advantage of using the Fox
calculus on these presentations is that the Jacobian matrix is only a 2 × 2 matrix
and so when one row and column has been struck out there is only one element left.
This basically means that the Fox derivative of either of the relations in the group
presentation, with respect to either generator will yield the Alexander polynomial.
In the presentation of the group of the trefoil, the two words in the list of relations
are:
w1 = xyxy −1 x−1 y −1
w2 = yxyx−1 y −1 x−1
Then, as calculated in our earlier example:
∂w1
= 1 + x(0 + y(1 + x(0 + y −1 (−x−1 )))) = 1 + xy − xyxy −1 x−1
∂x
Mapping all symbols to t we are left with 1 − t + t2 as required. Similarly:
∂w2
= 1 + y(0 + x(1 + y(0 + x−1 (−y −1 )))) = 1 + yx − yxyx−1 y −1
∂y
64
Which again maps to 1−t+t2 . In the same way, it can be shown that the remaining
two Fox derivatives also yield the Alexander polynomial.
For the group of 52 our two words in x and y are:
w1 = xyx−1 y −1 xyxy −1 x−1 yxy −1 x−1 y −1
w2 = yxy −1 x−1 yxyx−1 y −1 xyx−1 y −1 x−1
Then the derivative of w1 with respect to x is:
∂w1
∂x
= 1 + x(y(−x−1 + x−1 (y −1 (1 + x(y(1+
+x(y −1 (−x−1 + x−1 (y(1 + x(y −1 (−x−1 ))))))))))))
= 1 − xyx−1 + xyx−1 y −1 + xyx−1 y −1 xy − xyx−1 y −1 xyxy −1 x−1
+xyx−1 y −1 xyxy −1 x−1 y − xyx−1 y −1 xyxy −1 x−1 yxy −1 x−1
Mapping all symbols to t we get:
1 − t + 1 + t2 − t + t2 − t = 2 − 3t + 2t2
as required.
So in choosing which group presentation to use, we must consider the payoff between the more complicated algebra and the more complicated calculus. If we
begin with a larger number of generators then the relations are all very simple to
differentiate but we are left with a large matrix for which it takes a considerable
effort to find the determinant (if we are not using a computer!). If, on the other
hand, we begin with a minimal number of generators, we may not need to calculate determinants at all but the calculation of the Fox derivative is much more
cumbersome since the relations of the group are longer words.
65
66
Chapter 8
Knot theory redeemed
Alexander’s achievements in defining a polynomial invariant for knots paved the
way for the development of a number of other polynomial knot invariants including the HOMFLY polynomial and the Jones polynomial. These are calculated in
a different manner from the Alexander polynomial. The calculation of the Jones
polynomial, in particular, involves observing the effect on the knot of performing a
kind of local surgery by cutting the strands of the knot in a small neighbourhood
and reattaching them in a different manner. Altering the knot in this way is called
a skein operation.
The Jones polynomial was the first polynomial knot invariant to be discovered
since the Alexander polynomial and it took until 1984 for this development to take
place. The Jones polynomial is considered to be a more powerful knot invariant
than the Alexander polynomial as it distinguishes a larger proportion of knot types
(it distinguishes all prime knots with fewer than 9 crossings and also recognises the
handedness of knots) although it is still not a complete invariant and there even
exist pairs of knots that have the same Jones polynomial but different Alexander
polynomials.
Also, despite spending the most part of a century relegated to the attention of
only a select number of pure mathematicians and enthusiasts, knot theory has recently enjoyed a return to popularity within the wider scientific community. This
is because the mathematical tools developed for the study of knot types such as the
invariants we have discussed have been found to have uses in studying a number
of phenomena in physics, biology and chemistry. Given that the early interest in
the study of knots was amongst the physics community it is fitting that finally the
mathematics has finally found a practical use again in that field.
8.1
Applications in molecular biology
Every cell in the human (or indeed non-human) body contains all of the genetic
information needed to build the entire body. This information is coded on strands
of DNA, the famous double-helix structure discovered by James Watson, Francis
Crick and Rosalind Franklin in 1953 and for which Watson and Crick were awarded
67
the Nobel Prize for Medicine in 1962.
Topologically, DNA molecules take the form of long strands which can occur as
single strands or as coiled double helices. These either lie along an axis or have the
ends attached to form a ring.
In the process of DNA replication, the strands of the DNA are cut up by small
enzymes that occur inside the cell called topoisomerases. The action of these topoisomerases on strands of DNA is similar to the skein operations on knots described
above in calculating the Jones polynomial and the product molecules formed after
the action of the enzyme can have the form of various knots and links.
The enzymes may act on a molecule of DNA a number of times, forming a series of topologically different objects with each repeated recombination. Biologists
can then use knot theoretic techniques to analyse the topological types of a sample
of DNA molecules in a cell and then use this information to infer the specific change
to the structure of the DNA molecule each time the topoisomerase interacts with
it.
8.2
Applications in statistical mechanics
All matter is made up of a vast number of tiny particles which interact with each
other in a very complicated way. To understand the macroscopic properties of
matter by examining the mechanics of its constituent particles is almost impossible
and so what is required is a statistical approach: describing the probability that a
particle within a larger structure will have a certain property.
Statistical mechanics seeks to model the macroscopic properties of matter using
idealised statistical models and considering the respective probabilities that various states of the model will be realised.
The diagrams used to represent this idealised model of matter are in the form
of lattices of intersecting lines, which can be interpreted as arcs of a knot or link
and the points of intersection as crossing points.
Finally, the relations between the different states of the model have been found to
correspond again to skein operations so the mathematical properties of the Jones
polynomial can give insights into the properties of statistical mechanical models.
68
A table of prime knots of 7 or fewer crossings with their
Alexander polynomials
01
1
31
1 − t + t2
41
1 − 3t + t2
51
1 − t + t2 − t3 + t4
52
2 − 3t + 2t2
61
2 − 5t + 2t2
62
1 − 3t + 3t2 − 3t3 + t4
63
1 − 3t + 5t2 − 3t3 + t4
71
1 − t + t2 − t3 + t4 − t5 + t6
72
3 − 5t + 3t2
73
2 − 3t + 3t2 − 3t3 + 2t4
74
4 − 7t + 4t3
75
2 − 4t + 5t2 − 4t3 + 2t4
76
1 − 5t + 7t2 − 5t3 + t4
77
1 − 5t + 9t2 − 5t3 + t4
69
70
Bibliography
[1] JW Alexander, Topological invariants of knots and links, Transactions of
the American Mathematical Society, Volume 30, 1928, pp275–306
[2] C Livingston, Knot theory, Cambridge University Press, 1996
[3] P Cromwell, Knots and Links, Cambridge University Press, 2004
[4] RH Fox, A quick trip through knot theory, from Topology of 3-manifolds, MK
Fort Jr editor, Prentice-Hall, 1962
[5] K Murasugi, Knot Theory and Its Applications, translated by Bohdan
Kurpita, Birkhäuser Boston, 1996
[6] JJ O’Connor and EF Robertson, James Waddell Alexander, University
of St Andrews History of Mathematics Archive, http://www-groups.dcs.stand.ac.uk/∼history/Mathematicians/Alexander.html
[7] JJ O’Connor and EF Robertson, Karl Johannes Herbert Seifert, University of St Andrews History of Mathematics Archive,
http://www-groups.dcs.st-and.ac.uk/∼history/Mathematicians/Seifert.html
[8] JJ O’Connor and EF Robertson, Peter Guthrie Tait, University of St
Andrews History of Mathematics Archive,
http://www-history.mcs.st-and.ac.uk/history/Mathematicians/Tait.html
[9] Knot theory, Wikipedia, http://en.wikipedia.org/wiki/Knot theory
[10] Knot invariant, Wikipedia, http://en.wikipedia.org/wiki/Knot invariant
[12] Knot polynomial, Wikipedia, http://en.wikipedia.org/wiki/Knot polynomial
[13] Knots, Mathworld, http://mathworld.wolfram.com/Knot.html
[14] Alexander polynomial, Wikipedia,
http://mathworld.wolfram.com/AlexanderPolynomial.html
[15] The Knotplot Site,
http://www.cs.ubc.ca/nest/imager/contributions/scharein/KnotPlot.html
71
Notes
Text typeset in LATEX using TEXshop. The knot diagrams were drawn using the
Knotplot program, all hand-drawn diagrams were drawn using Stardraw. The
front cover graphics were produced using The GIMP (the knot graphic is from the
Knotplot site).
72
```