An Algebraic Approach to Symmetry

with Applications to Knot Theory

by

David Edward Joyce

A dissertation in mathematics

Presented to the Graduate Faculty of the University of Pennsylvania in partial

fulfillment of the requirements for the degree of Doctor of Philosophy

1979

[signed]
Peter J. Freyd
Supervisor of Dissertation

[signed] Jerry L. Kazdan

Graduate Group Chairman

1
 Preface

This edition of my dissertation differs little from the original 1979 version. This edition
is typeset in LATEX, whereas the original was typed on a typewriter and the figures were
hand drawn. The page numbers and figure numbers are changed, the table of contents
is expanded to include sections, a list of figures is included, and the index appears at the
end instead of the front. Ive corrected a few typos (and probably added others), and I
added figure 4.5 that was missing from the original.

David Joyce
June, 2009

i
 Acknowledgments

I thank my friends for their tolerance, nay, for their encouragement, of the inves-
tigation of the mathematics herein presented. I owe much to Julian Cole for many
encouraging discussions and for some suggestions concerning nomenclature. I owe much
more to my adviser, Peter Freyd, for his continued guidance of my mathematical matu-
rity and for many discussions in many fields, mathematical and otherwise. I give special
thanks to Janet Burns for the fine job of typing this dissertation.

Abstract

The usual algebraic construction used to study the symmetries of an object is the
group of automorphisms of that object. In many geometric settings, however, one may
interpret the symmetries in a more intimate manner by an algebraic structure on the
object itself. Define a quandle to be a set equipped with two binary operations, (x, y) 7
x . y and (x, y) 7 x .-1 y, which satisfies the axioms
Q1. x . x = x.
Q2. (x . y) .-1 y = x = (x .-1 y) . y.
Q3. (x . y) . z = (x . z) .(y . z).
Call the map S(y) sending x to x . y the symmetry at y.
To each point y of a symmetric space there is a symmetry S(y) of the space. By
defining x . y = x .-1 y to be the image of x under S(y), the symmetric space becomes a
quandle. Call a quandle satisfying x . y = x .-1 y an involutory quandle. Loos [1] has
defined a symmetric space as a manifold with an involutory quandle structure such that
each point y is an isolated fixed point of S(y).
The underlying set of a group G along with the operations of conjugation, x . y =
y 1 xy and x .-1 y = yxy 1 form a quandle Conj G. Moreover, the theory of conjugation
may be regarded as the theory of quandles in the sense that any equation in . and .-1
holding in Conj G for all groups G also holds in any quandle. If the center of G is trivial,
then Conj G determines G.
Let G be a group and n 2. The n-core of G is the set

{(x1 , x2 , . . . , xn ) Gn | x1 x2 . . . xn = 1}

along with the operation

(x1 , x2 , . . . , xn ) .(y1 , y2 , . . . , yn ) = (yn1 xn y1 , y11 x1 y2 , . . . , yn1

1
xn1 yn ).

The n-core is an n-quandle, that is, each symmetry has order dividing n. The group G
is simple if and only if its n-core is a simple quandle.
Let G be a noncyclic simple group and Q a nontrivial conjugacy class in H viewed
as a subquandle of Conj G. Then Q is a simple quandle.

ii
 Let Q be a quandle. The transvection group of Q, Trans Q, is the automorphism
group of Q generated by automorphisms of the form S(x)S(y)1 for x, y in Q. Suppose
Q is a simple p-quandle where p is prime. Then either Trans Q is a simple group, or else
Q is the p-core of a simple group G and Trans Q = Gp .
Consider the category of pairs of topological spaces (X, K), K X, where a map
f : (X, K) (Y, L) is a continuous map f : X Y such that f 1 (L) = K. Let
(D, O) be the closed unit disk paired with the origin O. Call a map from (D, O) to
(X, K) a noose in X about K. The homotopy classes of nooses in X about K form
the fundamental quandle Q(X, K). The inclusion of the unit circle to the boundary of
D gives a natural transfromation from Q(X, K) to the fundamental group 1 (X K).
A statement analogous to the Seifert-Van Kampen theorem for the fundamental group
holds for the fundamental quandle.
Let K be an oriented knot in the 3-sphere X. Define the knot quandle Q(K) to be
the subquandle of Q(X, K) consisting of nooses linking once with K. Then Q(K) is a
classifying invariant of tame knots, that is, if Q(K) = Q(K 0 ), then K is equivalent to
K 0 . The knot group and the Alexander invariant can be computed from Q(K).

[1] Loos, O., Symmetric Spaces, Benjamin, New York, 1969.

iii
Contents

1 Definitions and Examples 1

1.1 Quandles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Involutory quandles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Representations and the general algebraic theory of quandles 6

2.1 The algebraic theory of conjugation . . . . . . . . . . . . . . . . . . . . . 6
2.2 Automorphism groups of quandles . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Representation of quandles as conjugacy classes . . . . . . . . . . . . . . 7
2.4 Representation of quandles as cosets . . . . . . . . . . . . . . . . . . . . 8
2.5 Algebraic connectivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.6 The transvection group . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.7 n-Cores . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.8 Examples of simple quandles . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.9 Classification of simple p-quandles . . . . . . . . . . . . . . . . . . . . . . 14
2.10 Augmented quandles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.11 Quotients of augmented quandles described by normal subgroups of the
augmentation group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3 Involutory quandles 23
3.1 Involutory quandles and geodesics . . . . . . . . . . . . . . . . . . . . . . 23
3.2 Involutory quandles generated by two points . . . . . . . . . . . . . . . . 25
3.3 Group cores . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.4 Distributive quandles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.5 Involutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.6 Moufang loop cores . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.7 Distributive 2-quandles with midpoints . . . . . . . . . . . . . . . . . . . 34

4 Algebraic topology and knots 36

4.1 The fundamental quandle of a pair of spaces. . . . . . . . . . . . . . . . . 36
4.2 The fundamental quandle of a disk . . . . . . . . . . . . . . . . . . . . . 41
4.3 The Seifert-Van Kampen theorem . . . . . . . . . . . . . . . . . . . . . . 42
4.4 Applications of the Seifert-Van Kampen theorem . . . . . . . . . . . . . 44
4.5 Knot quandles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

iv
 4.6 A presentation of the knot quandle . . . . . . . . . . . . . . . . . . . . . 47
4.7 The invariance of the knot quandle . . . . . . . . . . . . . . . . . . . . . 48
4.8 A presentation of the knot quandle (continued) . . . . . . . . . . . . . . 50
4.9 A representation of the knot quandle . . . . . . . . . . . . . . . . . . . . 52
4.10 The Alexander invariant of a knot . . . . . . . . . . . . . . . . . . . . . . 53
4.11 The cyclic invariants of a knot . . . . . . . . . . . . . . . . . . . . . . . . 56
4.12 The involutory knot quandle . . . . . . . . . . . . . . . . . . . . . . . . . 56

Bibliography 59

Index 62

v
List of Figures

1.1 The trefoil knot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.1 A singular 2-quandle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

3.1 QG3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2 Geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3 Singular quandles Cs(4) and CS(8) . . . . . . . . . . . . . . . . . . . . . 25
3.4 Distributivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.5 Midpoints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

4.1 The noose N . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

4.2 A noose homotopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.3 q and q 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.4 Seifert-Van Kampen noose homotopy . . . . . . . . . . . . . . . . . . . . 43
4.5 Noose boundary loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.6 The figure-8 knot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.7 A knot crossing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.8 Basic knot deformations . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.9 Invariance under 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.10 Invariance under 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.11 Invariance under 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.12 The loop i . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.13 The trefoil knot 31 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.14 The figure-8 knot 41 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.15 The knot 10124 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.16 Q2 (10124 ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.17 Two links with homeomorphic complements but different quandles . . . . 59

vi
Chapter 1

Definitions and Examples

1.1 Quandles
Let Q be a set equipped with a binary operation, and denote this operation by (x, y) 7
x . y. We use the nonsymmetric symbol . here since the two variables will play different
roles in the following discussion. Also, it will distinguish this binary operation from
others that Q may have, in particular, addition and multiplication.
For z in Q, let S(z) be the function on Q whose value at x is x . z. It will be more
convenient for us to use the notation xS(z) = x . z rather than S(z)(x) = x . z. For S(z)
to be a homomorphism, we require
(x . y)S(z) = xS(z) . yS(z),
that is,
(1) (x . y) . z = (x . z) .(y . z).
When (1) holds for all x, y, z in Q, S is a function from Q to End Q, the set of endomor-
phisms of Q. If S(z) is also a bijection for all z, then S maps Q to Aut Q, the group of
automorphisms of Q. Any group, in particular Aut Q, has the operation of conjugation,
f . g = g 1 f g, which satisfies (1). Then S : Q Aut Q is itself a homomorphism.
That is S(y . z) = S(z)1 S(y)S(z), equivalently, S(z)S(y . z) = S(y)S(z), which is a
restatement of (1). The requirement that S(z) be a bijection for all z is equivalent to
the existence of another binary operation

(x, y) 7 x .-1 y

that satisfies
(2) x . y = z x = z .-1 y.
An equational identity equivalent to (2) is
(3) (x . y) .-1 y = x = (x .-1 y) . y.
From (1) and (2) we may derive the identities

1
 (x . y) .-1 z = (x .-1 z) .(y .-1 z),
(x .-1 y) . z = (x . z) .-1 (y . z),
(x .-1 y) .-1 z = (x .-1 z) .-1 (y .-1 z).

In all the applications that follow, S(z) will not only be an automorphism, but one which
fixes z.
Definition. A quandle is a set Q equipped with two binary operations (x, y) 7 x . y and
(x, y) 7 x .-1 y which satisfies three axioms
Q1. x . x = x.
Q2. (x . y) .-1 y = x = (x .-1 y) . y.
Q3. (x . y) . z = (x . z) .(y . z).
The map S(z) is called the symmetry at z, and x . z may be read as x through z. The
axioms taken together say that the symmetry at any point of Q is an automorphism of
Q fixing that point. The order of a quandle is the cardinality of its underlying set. The
elements of a quandle will be frequently referred to as points.
Example 1. A group G is a quandle, denoted Conj G, with conjugation as the operation.
x . y = y 1 xy, x .-1 y = yxy 1 . Any conjugacy class of G is a subquandle of Conj G as is
any subset closed under conjugation. When G is Abelian, the operation becomes simply
the first projection operation, x . y = x.
Definition. A quandle Q is said to be Abelian if it satisfies
QAb. (w . x) .(y . z) = (w . y) .(x . z).
It follows from the definition that an Abelian quandle also satisfies the identities

(w .-1 x) .(y .-1 z) = (w . y) .-1 (x . z)

(w .-1 x) .-1 (y .-1 z) = (w .-1 y) .-1 (x .-1 z)

Example 2. Let T be a nonsingular linear transformation on a vector space V . Then V

becomes a quandle with the operations x . y = T (x y) + y and x .-1 y = T 1 (x y) + y.
Moreover, V is an Abelian quandle.
It should be noted that quandles are seldom associative. In fact, the identity
(x . y) . z = x .(y . z) is equivalent to the identity x . y = x. One associativity equa-
tion which does hold for any quandle is (x . y) . x = x .(y . x). To reduce the number of
parentheses we use the notation x . y . z for (x . y) . z.
By far the most interesting axiom for quandles is the distributivity axiom Q3. The first
study of self-distributivity is that of Burstin and Mayer [5]. They define distributive
groups, or in modern terminology, distributive quasigroups. A quasigroup is a set G
equipped with a binary operation (x, y) 7 xy such that for all a, b in G there exist

2
unique solutions x, y to the equations xa = b and ay = b. A quasigroup is distributive if
it satisfies the two identities

(xy)z = (xz)(yz), and

x(yz) = (xy)(xz).

It follows that a distributive quasigroup is idempotent, xx = x. Hence, quandles are a

generalization of distributive quasigroups. Burstin and Mayer define an Abelian dis-
tributive group to be one satisfying

(wx)(yz) = (wy)(xz).

This axiom goes by the names entropy, mediality, surcommutativity, and sym-
metry.

1.2 Involutory quandles

An important class of quandles are those in which the symmetries S(z) are all involutions,
S(z)2 is the identity. In this case x . z = x .-1 z, which allows us to dispense with the
second quandle operation. An equivalent condition is the identity
QInv. x . y . y = x.
Definition. A quandle satisfying QInv is called an involutory quandle or 2-quandle.
Alternatively, an involutory quandle may be defined as a set equipped with a binary
operation (x, y) 7 x . y which satisfies Q1, Qinv, and Q3. Analogously an n-quandle is
a quandle such that for all x, y, xS(y)n = x. In any quandle let x .n y denote xS(y)n .
Example 1. Let G be a group. The set of involutions in G, Inv G = {x G | x2 = 1},
forms an involutory quandle with conjugation as the operation.
Example 2. Any group G has an involutory quandle structure given by x . y = yx1 y. The
underlying set of G along with this operation is called the core of G and is denoted Core G.
Note that Core G, Conj G, and Inv G are all distinct unless G consists of involutions only.
Example 3. Let M be a Riemannian symmetric space, that is, a connected Riemannian
manifold M in which each point y is an isolated fixed point of an involutive isometry S(y).
In a neighborhood of y, S(y) is given in terms of the exponential map exp : Ty M
(Ty = tangent space at y) as

xS(y) = exp( exp1 (x)).

Since M is connected, this local involutive isometry is uniquely extendable to M . M is

an involutive quandle with the operation x . y = xS(y). Indeed, Q1 holds since y is fixed
by S(y), and QInv holds since S(y) is an involution. To show Q3 it suffices to show

3
S(y . z) = S(z)S(y)S(z). But S(z)S(y)S(z) is an involutive isometry having y . z as an
isolated fixed point, and S(y . z) is described by this property.
A more descriptive construction of x . y is the following. If x = y, let x . y = x.
Otherwise, pass a geodesic through x and y, and let d be the length along the geodesic
from x to y. Let x . y be the point on the geodesic extended through y by the same
length d.
Symmetric spaces give examples of 2-quandles which are not cores of groups. Three
basic two-dimensional symmetric spaces are the sphere, Euclidean plane, and hyperbolic
plane. The Euclidean plane is the core of the Abelian group R2 , but neither the sphere
nor the hyperbolic plane are cores of toplogical groups.
Example 4. The involutory quandle of a knot. Consider a regular projection of a knot
K, such as the trefoil knot in figure 1.1, and label the arcs a, b, c, . . ., where by arc
is meant a segment from one underpass, over whatever overpasses there may be, to the
next underpass. At each underpass, read a relation on the arcs, as a under b gives c
a . b = c. Let Q(K) be the quandle generated by the arcs with relations given by the
underpasses. For instance,

Q(trefoil) = (a, b, c : a . b = c, b . c = a, c . a = b)

which is isomorphic to Core(Z/3Z). The order of Q(K) need not be equal to the number
of arcs in the projection; it need not even be finite. A different regular projection of
K will give the same Q(K) up to isomorphism. Moreover, if K and K 0 are equivalent
knots, then Q(K) is isomorphic to Q(K). Proofs and precise definitions will be supplied
in Chapter 4.

Figure 1.1: The trefoil knot

c.a = b
a.b = c

c
b.c = a
b

A similar construction gives the (non-involutory) quandle of a knot. An orientation

of the knot is used to determine the relations. As expected, the knot quandle holds more
information about the knot than the involutory knot quandle.

4
 M. Takasaki [16] defined involutory quandles under the name kei. Takasakis mo-
tivation derives form the net (web) theory of Thomsen [17,18]. This theory is described
in the book of Blaschke and Bol [2]. A similar geometric basis underlies Moufangs [9]
study of loops. Bruck [3] defined the core of a Moufang loop as the underlying set of the
loop along with the operation (x, y) 7 yx1 y (example 2 above). See Chapter 3 more
more on loop cores.
Loos discovered the intrinsic algebraic structure of symmetric spaces as explained in
Loos [7,9]. Not only are Riemannian symmetric spaces determined by their algebraic
structure, but so are affine symmetric spaces. This allows Loos to define a symmetric
space as a differentiable involutory quandle in which every point is an isolated fixed point
of the symmetry through it.

5
Chapter 2

Representations and the general

algebraic theory of quandles

There are various ways that groups may be used to represent quandles. First of all,
Conj G, for G a group, is a quandle. Many quandles may be represented as subquandles of
Conj G for appropriate G. Free quandles, for example, may be so represented. Secondly,
homogeneous quandles may be represented as cosets H\G for H a subgroup of G where
an automorphism of G fixing H is needed to describe the quandle operations on H\G.
Non-homogeneous quandles are representable as a union H1 \G H2 \G where
several automorphisms are used to describe the quandle operations. Finally, a quandle
may be given by a set Q along with an action of a group G and a function  : Q G
that describes the symmetries of the points of Q. Such a construction will be called an
augmented quandle. We will be able to study some varieties of quandles by means of
augmented quandles.

2.1 The algebraic theory of conjugation

In this section we show that the theory of quandles may be regarded as the theory of
conjugation. Consider the two binary operations of conjugation, (x, y) 7 y 1 xy = x . y
and (x, y) 7 yxy 1 = x .-1 y, on a group. We ask whether there are any equations
involving only these two operations which hold uniformly for all groups other than those
which hold in all quandles. To this end we show that free quandles may be faithfully
represented as unions of conjugacy classes in free groups.
Proposition. Let A be a set and F be the free group on A. Then the free quandle on A
appears as the subquandle Q of Conj F consisting of the conjugates of the generators of
F.
Proof. We use the notation of quandles in Conj F . Each element of Q is named as

a .e1 b1 .e2 .en bn

6
where a, b1 , . . . , bn A and e1 , . . . , en {1, 1}. That is to say, the conjugates of a are
of the form
bnen be e1 en
1 ab1 bn .
1

The equivalence on names is generated by two cases.

1. If a = b1 , then a .e1 b1 .e2 .en bn names the same element as a .e2 b2 .e3 .en bn .

2. If bi = bi+1 and ei + ei+1 = 0, then a .e1 b1 .e2 .en bn names the same element as
a .e1 .ei1 bi1 .ei+2 bi+2 . .en bn .
Now let each a in A be assigned to a point f (a) in a quandle P . If f extends to Q, then
we must have

f (a .e1 b1 .e2 .en bn ) = f (a) .e1 f (b1 ) .e2 .en f (bn ).

We must only show that this extension is well-defined. But this follows directly from the
fact that in P the analogues of 1) and 2) hold for f (a) .e1 f (b1 ) .e2 .en f (bn ). 
Corollary. any equation holding in Conj G for all groups G holds in all quandles.
Proof. Let E be an equation holding in Conj G for all groups G. In particular E holds
in Conj F for free groups F , hence, E holds in free quandles. Whence, E holds in all
quandles. 

2.2 Automorphism groups of quandles

Let Q be a quandle. We define three automorphism groups for A. First, there is the
group consisting of all automorphisms, the full automorphism group of Q, Aut Q. Second,
there is the subgroup of Aut Q, generated by all the symmetries of Q, called the inner
automorphism group of Q, Inn Q. Third, there is the subgroup of Inn Q generated by
automorphisms of the form S(x)S(y)1 for x, y Q, called the transvection group of Q,
Trans Q. Inn Q is a normal subgroup of Aut Q, and Trans Q is normal in both Inn Q and
Aut Q. The quotient group Inn Q/ Trans Q is a cyclic group. The elements of Trans Q
are the automorphisms of the form S(x1 )e1 S(xn )en such that e1 + + en = 0.
To illustrate these groups let Q be R2 with x . y = 2y z consered as a quandle in the
category of topological spaces. Then Aut Q consists of the continuous automorphisms
of R2 , that is, the affine transformations. Inn Q includes symmetries at points and
translations. Trans Q includes only translations.

2.3 Representation of quandles as conjugacy classes

Two points x and y of a quandle Q are said to be behaviorally equivalent if z . x = z . y for
all z in Q. An equivalent condition is that S : Q Inn Q identifies x and y. Behavioral
equivalence is a congruence relation, b , on the quandle, and Q/ b is isomorphic to the

7
image S(Q) as a subquandle of Conj Inn Q. The elements of Q are behaviorally distinct
if and only if S is an injection, in which case Q is isomorphic to a union of conjugacy
classes in Inn Q.
Even if the points of a quandle are not all behaviorally distinct, the quandle may be
isomorphic to a union of conjugacy classes of some
Q group. For instance, any quandle
satisfying x . y = x may be embedded in Conj( I Z2 ) for sufficiently large I. There is
a universal group in which to represent a quandle as a subset closed under conjugation.
As noted in example 1 of Chapter 1, every group G may be considered to be a quandle
Conj G, with conjugation as the quandle operation. Adjointly, every quandle Q gives rise
to a group Adconj Q, generated by the elements of Q modulo the relations of conjugation.
Precisely, Adconj Q has the presentation

(x, for x Q : x . y = y 1 x y, for x, y Q}.

The map : Q Conj Adconj Q sending x to x is a quandle homomorphism whose

image is a union of conjugacy classes of Adconj Q. The map has the following universal
property: for any quandle homomorphism h : Q Conj G, G a group, there exists a
unique group homomorphism H : Adconj Q G such that h = H . Thus, if any
h : Q Conj G is monic, then is monic.
But need not be injective in general. Consider the 2-quandle of three elements given
1
in the table in figure 2.1. Since b . a = b, a commutes with b. But a . b = c, so b ab = c.
Therefore, a = c, and is not injective.

Figure 2.1: A singular 2-quandle

. a b c
a a c a
b b b b
c c a c

Later, when we consider the quandle associated to a knot, the non-injectivity of

will be important. For example, the quandles associated to the square and granny knots
are distinct, but the Adconj groups of these quandles (which are the knot groups) are
isomorphic, and for each, is not injective.

2.4 Representation of quandles as cosets

Let s be an automorphism on a group G. We may define a quandle operation on G
by x . y = s(xy 1 )y. Verification is straightforward. Let H be a subgroup of G whose
elements are fixed by s. Then H\G inherits this quandle structure

Hx . Hy = Hs(xy 1 )y.

8
Denote this quandle as (H\G; s). G acts on the right on (H\G; s) by (Hx, y) 7 Hxy, and
the action is by quandle automorphisms. Since G acts transitively on H\G, it follows that
(H\G; s) is a homogeneous quandle, that is, there is a quandle automorphism sending
any point to any other point of the quandle.
We are mainly interested in the case when s is an inner automorphism of G, s(x) =
z xz for some fixed element z of G. Then x . y = z 1 xy 1 zy. When H contains z, the
1

operation of (H\G; z) = (H\g; s) is

Hx . Hy = Hxy 1 zy.

Proposition. Every homogeneous quandle is representable as (H\G; z).

Proof. Let Q be a homogeneous quandle and G = Aut Q. Fix p Q. Let z = S(p),
symmetry at p, and let s be conjugation by z in G. Then e : (G; s) Q, evaluation
at p, defined by sending the element x to its value at p, is a quandle homomorphism.
Indeed, e(x . y) = e(z 1 xy 1 zy) = p xy 1 zy = (p xy 1 . p)y = p xy 1 y . py = e(x) . e(y).
Since Q is homogeneous, e is surjective. Let H be the stability subgroup of p, H = {x
G | px = p}. Then e factors through (H\g; z) since p = pH. Moreover, (H\G; z) Q is
injective, for if pHx = pHy, then pxy 1 = p, xy 1 H, and so Hx = Hy. Thus, Q is
isomorphic to (H\G; z). 
Some adjustments are needed to represent non-homogeneous quandles. Given a group
G, elements z1 , z2 , . . . of G, and subgroups H1 , H2 , . . . of G such that for each index i, Hi
is contained in the centralizer of zi , we form a quandle (H1 , H2 , . . . \G; z1 , z1 , . . .) as the
disjoint union of H1 \G, H2 \G, . . . with the quandle operation

Hi x . Hj y = Hi xy 1 zj y.


Proposition. Every quandle is representable as (H1 , H2 , . . . \G; z1 , z1 , . . .).
Proof. Let Q be a quandle and G = Aut Q. Let Q1 , Q2 , . . . be the orbits of the action of G
on Q. For each index i choose pi Qi , let zi = S(pI ), and let Hi be the stability subgroup
of pi . Then for each i, Hi is contained in the centralizer of zi , and so we have a quandle
P = (H1 , H2 , . . . \G; z1 , z1 , . . .) as described above. Define e : P Q by Hi x 7 pi x. As
in the proof of the previous propsition, e may be shown to be an isomorphism. 
In the case of involutory quandles, the automorphism s of G must be an involution
on G while the elements z, z1 , z2 , . . . of g must be involutions in G.

2.5 Algebraic connectivity

We say that a quandle Q is algebraically connected (or just connected when there will be
no confusion with topological connectivity) if the inner automorphism group Inn Q acts

9
transitively on Q. In other words, Q is connected if and only if for each pair a, b in Q
there are a1 , a2 , . . . , an in Q and e1 , e2 , . . . , en {1, 1} such that
a .e1 a1 .e2 .en an = b.
Let Q be a quandle and q a point of Q. The q-fibre of a map g : Q Q00 is the
subquandle Q0 = {p Q | g(p) = g(q)} of Q. Suppose that Q00 is a quotient of Q, that
is, Q00 is given by a congruence on Q. In general the q-fibre does not determine Q00 ; just
consider quandles whose operation is the first projection.
Proposition. Let Q be an algebraically connected quandle and q a point of Q. Then
every quotient of Q is determined by its q-fibre. Consequently, every congruence on Q is
determined by any one of its congruence classes.
Proof. Let Q00 be a quotent of Q with q-fibre Q0 . Let a, b Q. By the connectivity of
Q there is an inner automorphism x such that ax = q. Since homomorphisms respect
inner automorphisms, it follows that g(a) = g(b) if and only if g(ax) = g(bx). Hence,
g(a) = g(b) if and only if x Inn Q such that ax = q and bx Q0 . Thus, Q0 determines
Q00 . 

2.6 The transvection group

As defined above, the transvection group Trans Q of a quandle Q is the subgroup of Inn Q
generated by automorphisms of the form S(x)S(y)1 . Trans Q is a normal subgroup of
Inn Q with cyclic quotient. Alternatively, we may define a transvection on Q as an
automorphism of Q of the form S(x1 )e1 S(xn )en with xi Q, ei Z, i = 1, . . . , n,
such that e1 + + en = 0. Then Trans Q is the group of transvections on Q.
Some the the properties of Q are reflected in Trans Q.
Proposition. A quandle is Abelian if and only if its transvection group is Abelian.
Proof. Let Q be a quandle with transvection group T . By definition, Q is Abelian if and
only if
(w . x) .(y . z) = (w . y) .(x . z).
Equivalently,
S(x)S(z)1 S(y) = S(y)S(z)1 S(x).
On the other hand, T is Abelian if and only if
S(x)S(z)1 S(y)S(t)1 = S(y)S(t)1 S(x)S(z)1 .
By setting t = z we find that if T is Abelian then Q is Abelian. From Q Abelian follows
S(x)S(z)1 S(y)S(t)1 = S(y)S(z)1 S(x)S(t)1
= S(y)S(t)1 S(x)S(z)1 .
which implies that T is Abelian. 

10
2.7 n-Cores
The core of a group has the property that all its symmetries are involutions. In this
section we define an n-core of a group wherein the n-th power of each symmetry is the
identity. This agrees with the usual core in the case n = 2.
Let G be a group and n a positive integer. The wreath product G o Zn consists of
n + 1-tuples (x0 , . . . , xn1 , k) with xi G, k Zn . The index i is to take values in Zn .
Multiplication in G o Zn is given by

(x0 , . . . , xn1 , k) (y0 , . . . , yn1 , l) = (x0 yk , x1 yk+1 , . . . , xn1 yk1 , k + l).

Let Q be the conjugacy class of (1, . . . , 1, 1) in G o Zn . Then

Q = {x0 , . . . , xn1 , 1) | x0 xn1 = 1}

and the quandle operation on Q is given by

1
(x0 , . . . , xn1 , 1) .(y0 , . . . , yn1 , 1) = (yn1 xn1 y0 , y)1 x0 y1 , . . . , 1).

This quandle is called the n-core of G. The 2-core of G is isomorphic to the core of G.

2.8 Examples of simple quandles

A quandle is said to be simple if its only quotients are itself and the one-point quandle.
We will show in this section that n-cores of noncyclic simple groups are simple. In fact,
a noncyclic group is simple if and only if its n-core is simple. Also, nontrivial conjugacy
classes of simple groups are simple. We proceed with some lemmas.
Lemma 1. Let H be a group with commutator H 0 . An element x in H lies in H 0 if and
only if there exist x1 , x2 , . . . , xk in H such that x = x1 x2 . . . xk and xk . . . x2 x1 = 1. 
Lemma 2. Let H be a perfect group, H = H 0 . Then the n-core Q of H as a subset of
G = H o Zn generates G.
Proof. Let b = (1, 1) = (1, . . . , 1, 1) Q. Let x H. By lemma 1, x =
x1 . . . xk , xk . . . x1 = 1. Then

(x1 , x1 1 1
1 , 1, . . . , 1)b (x2 , x2 1, . . . , 1)b
1
. . . (xk , x1
k , 1, . . . , 1)b
1

1
= (x1 x2 xk , x1 1
1 x2 xk , 1, . . . , 1, 0)
= (x, 1, . . . , 1, 0).

Since b and each (xi , x1

i , 1, . . . , 1) lie in Q, so (x, 1, . . . , 1, 0) is a member of the subgroup
generated by Q. The rest follows easily. 
Lemma 3. If the center of a nontrivial group H is trivial, then the center of H o Zn is
trivial. 

11
Lemma 4. Let G be a group with trivial center, and let Q be a conjugacy class that
generates G. Then G = Inn Q, and G0 = Trans Q.
Proof. For each x G let S(x) be conjugation by x, and regard S(x) as an automorphism
of Q. Then S is a group homomorphism S : G Aut Q. Note that S(x) = 1 if and
only if for all q in Q, x1 qx = q. Since Q generates G, S(x) = 1 if and only if x Z(G).
Therefore, S is injective. The image of S is Inn Q. Hence, S is an isomorphism G = Inn Q.
0
We show next that S(G ) = Trans Q. Let p, q Q. Then [S(p), S(q)] =
S(p)1 S(p . q) Trans Q. Trans Q is a normal subgroup of Inn Q, so S(G0 ) = (Inn Q)0
Trans Q. Since Q is a conjugacy class in G, there is an x in G such that x1 px = q.
Therefore, S(p)1 S(q) = [x, S(q)] (Inn Q)0 . Thus, Trans Q S(G0 ). 
Lemma 5. Under the hypotheses of lemma 4 the following statements are equivalent.
(1) Q is a simple quandle.
(2) G0 is the smallest nontrivial normal subgroup of G.
(3) G0 is a minimal nontrivial normal subgroup of G.
Proof. (1) = (2). Let N be a normal subgroup of Inn Q. Define an equivalence relation
on Q by
p q n N such that pn = q.
We show that is an congruence. Assume p q, pn = q. For r in Q we have
q . r = pn . r = (p . r)m where m = S(r)1 nS(r) N . Hence, q . r p . r. Also
r . q = r . pn = (r . p)m1 n where m1 n N . Hence, r . q = r . p. Therefore, is a
congruence. By the simplicity of Q we have only two cases.
Case 1. is equality. Let n N . For all q in Q, qn = q, so n1 S(q)n = S(q). From
the hypotheses of the lemma it follows that n = 1. Thus, N = 1.
Case 2. relates all points of Q. For p, q in Q there is an n in N such that pn = q.
Then n1 S(p)n = S(q). Therefore, S(p)S(q)1 N . Hence, Trans Q N .
Now (2) follows from the conclusions of lemma 4.
(2) = (3). Clear.
(3) = (1). Assume (3). Let be a congruence on Q. Conjugation by elements of
Q respects , that is, p q implies p . r q . r. Since Q generates G, conjugation by
elements of G respects . Let

N = {x G0 | ps(x) p for all p Q}.

Then N is a normal subgroup of G contained in G0 . By (3), either N = 1 or N = G0 .

Assume is not equality. Then q, r Q such that q r but q 6= r. It follows that
1 6= qr1 N . Hence, N = G0 . Now let q, r be arbitrary in Q. x G0 such that
x1 qx = r. Therefore, q qS(x) = p. Thus, if is not equality, then relates any two
elements. 
Lemma 6. Let H be a noncyclic simple group and G = H o Zn , n 2. Then K =
{(x0 , . . . , xn1 , 0) G} is a minimal nontrivial normal subgroup of G.

12
Proof. Let (x0 , . . . , xn1 , 0) be a nontrivial element of K. We will show the smallest
normal subgroup N containing this element is K. For some i, xi 6= 1, say i = 0. There
is an element w of G such that [x0 , w] = z 6= 1. Then [(x0 , . . . , xn1 , 0), (w, 1, . . . , 1)] =
(z, 1, . . . , 1, 0) lies in N . As z 6= 1 and H is simple, it follows that (y, 1, . . . , 1, 0) N for
all y in H. Now

(y, 1, . . . , 1, 0) .(1, . . . , 1, k) = (1, . . . , y, . . . , 1, 0)

also lies in N for all y in H and k in Zn . Hence N = K. 

Theorem 1. Let H be a noncyclic group and Q be the n-core of H (n 2). Then Q is
simple if and only if H is simple, in which case Inn Q
= G = H o Zn and

Trans Q
= G0 = {(x0 , . . . , xn , 0) G}
= H n.

Proof. Assume H is simple. According to lemma 2, Q generates G, and by lemma 3

z(G) = 1. Since the hypotheses of lemma 4 hold, we have G = Inn Q, and G0 = Trans Q.
Lemma 6 says K = {(x0 , . . . , xn1 , Q) T } is a minimal nontrivial normal subgroup of
G. Hence, G0 = K. Finally, we conclude from lemma 5 that Q is simple.
Any normal subgroup N of H gives a quandle congruence on Q defined by

(x, 1) (y, 1) xi yi1 N for i = 0, . . . n 1.

Moreover, if N 6= 1, then is not equality. Hence, the simplicity of Q assures that of

H. 
The n-core of a noncyclic simple group retains, therefore, more information about the
group than just its simplicity. It can, in fact, be reconstructed from its n-core.
Remark. The 4-core of the cyclic simple group Z2 is not a simple quandle.
Corollary. The core (2-core) of a group is simple if and only if the group is simple.
Proof. The only groups not covered by the theorem are cyclic groups for which the
statement is easily verified. 
There are two other ways that simple quandles derive from simple groups besides
n-cores. We will show that any nontrivial conjugacy class in a simple group is a simple
quandle. The following lemma is a direct consequence of lemmas 4 and 5.
Lemma 7. Under the hypotheses of lemma 4, if G0 is a simple group, then Q is a simple
quandle. 
Theorem 2. Let H be a noncyclic simple group and Q a nontrivial conjugacy class in H.
Then Q is a simple quandle. Also, Inn Q = Trans Q
= H.
Proof. Q generates H, and z(H) = 1. So by lemma 4, Inn Q = Trans Q = H. By lemma
7, Q is a simple quandle. 

13
Theorem 3. Let H be a noncyclic simple group, p a prime integer, and s an outer
automorphism of G of order p. Let G be the semidirect product H n Zp , (x, y) (y, l) =
(xsk (y), k + l). Then Q is a simple quandle, Inn Q
= G, and Trans Q
= H.
Proof. First we show that Q generates G. Let (Q) be the subgroup generated by Q.
(Q) is normal in G. (1, 1) (Q), so (1, k) (Q) for all k in Zp . Also, (1, 1) .(y, 0) =
(y 1 s1 (y), 1) (Q). Since s 6= 1, y H such that 1 6= (y 1 s1 (y), 1) (Q). Also
(y 1 s1 (y), 0) (Q). Hence (Q) H 6= 1. Therefore, (Q) H = H. It follows that
(Q) = G.
Next we show Z(G) = 1. Suppose (a, k) Z(G). Then zsk (y) = yz for all y in H.
Thus, sk = S(z 1 ). If p divides k, then 1 = sk = S(z 1 ), which gives (z, k) = (1, 0).
Otherwise, (p, k) = 1. Then for some m, km 1 mod p, so s = skm = S(z m ). in
contradiction to the hypothesis that s is not an inner automorphism. This, Z(G) is
trivial.
We have shown that Q and G satisfy the hypotheses of lemma 4. Hence, G = Inn Q,
0
and G = Trans Q.
Clearly, G0 = H n 0 = H, so by lemma 7, Q is simple. 

2.9 Classification of simple p-quandles

In this section we examine the problem of classifying simple quandles. In the case of p-
quandles, p a prime integer, we solve the problem in terms of simple groups. Throughout
this section let Q be a simple quandle and G = Inn Q.
Lemma 1. Either S : Q G is injective or the order of Q is less than three.
Proof. The behavioral equivalence on Q is either equality or else relates any two elements
of Q. In the former case S is injective. In the latter case Q satisfies the identity x . y = x.
But the only simple quandles satisfying x . y = x have fewer than three elements. 
Assume for the rest of this section that the order of Q is greater than two. Since the
set of connected components of Q is a quotient of Q, it follows that Q is algebraically
connected. Also, S(Q) is a conjugacy class in G since it is closed under conjugation and
generates G.
Lemma 2. The center of G is trivial.
Proof.
z Z(G) q Q, zS(q) = S(q)z
q Q, S(qz) = S(q)
q Q, qz = q.
But the only automorphism fixing all the points of Q is 1. Therefore, Z(G) = 1. 
By lemma 4 of section 2.9 we have Trans Q = G0 . Hence, G/G0 is a cyclic group.
Moreover, if Q is an n-quandle, then the order of G/G0 divides n. Since S(Q)
= Q is

14
a simple quandle, by lemma 5 of section 2.9 it follows that G0 is the smallest nontrivial
normal subgroup of G.
At this point we must break the classification into cases. If G0 = G, then G is a
simple group, and Q is isomorphic to the nontrivial conjugacy class S(Q) in the simple
group G. For the rest of this section we assume G0 6= G. We will also assume that Q is
a p-quandle, p a prime integer. Then G/G0 = Zp .
Fix q0 in Q. Let x0 = S(q0 ) G, and let s be conjugation by x0 as an automorphism
of G0 . Then x0 6= 1, xpo = 1, s 6= 1, sp = 1. Let K be the semidirect product G0 n Zp
where
(x, k)(y, l) = (xsk (y), k + l).
There is an isomorphism f : K G, f (x, k) = xxk0 . In particular, f (1, 1) = x0 , and
f 1 (S(Q)) is the conjugacy class in K of (1, 1).
Suppose G0 is a simple group. From the fact that Z(G0 ) = 1 it follows that s is not
an inner automorphism of G0 . Thus, Q is isomorphic to the conjugacy class of (1, y)
in k = G0 n Zp where G0 is a simple group and G0 n Zp is constructed from an outer
automorphism of G0 of order p. This is the situation encountered in theorem 3 of section
2.8.
We have yet to consider the case where G0 is not simple.
Lemma 3. Let H be a group with a smallest nontrivial normal subgroup T such that
[H : T ] = p is prime. Assume T is not simple. Then T is isomorphic to N p for some
simple group N .
Proof. Let N be a nontrivial proper normal subgroup of T . Fix x in H T . Let s
be conjugation by x as an automorphism of T . Then sp (N ) = N since xp T . More
generally, sp+i (N ) = si (N ) for any integer i. Since p is prime and N is not normal in N ,
we have p distinct conjugates of N , namely,

N, s(N ), . . . , sp1 (N ).

Claim. For k = 0, 1, . . . , p 2, there exist nontrivial proper normal subgroups Nk of

T such that 0 6= |i j| k implies si (Nk ) sj (Nk ) = 1.
Define N0 = N . Inductively define Nk+1 (k + 1 p 1) as follows. The group

Nk sk+1 (Nk ) s(k+1)(p1) (Nk )

is normal in G since k+1 is relatively prime to p, and, being strictly contained

in T , is, therefore, trivial. Let l be least such that

Nk sk+1 (Nk ) s(k+1)l (Nk ) = 1.

Then
Nk+1 = Nk sk+1 (Nk ) s(k+1)(l1) (Nk )
satisfies the requirements of the claim.

15
We may assume N = Np1 . That is, the p conjugates of N ,

N, s(N ), . . . , sp1 (N ),

have pairwise trivial intersection and generate T . Hence

T = N s(N ) sp1 (N )
= N p.

Also, N is simple. Indeed, if M is a proper normal subgroup of N , then M s(M )

sp1 (M ) is normal in H and strictly contained in T and, therefore, is trivial as is
M. 
Theorem 1. Let Q be a simple p-quandle, p a prime, and let G = Inn Q. As noted above,
G0 = Trans Q. Assume G0 is not a simple group. Then G is a wreath product of a simple
group N with Zp , G0 = N p , and Q is isomorphic to the p-core of N .
Proof. By lemma 3, G0 = N p where N is a simple group. As noted above G is isomorphic
to G n Zp . Therefore, G = G0 n Zp = N o Zp . Also, Q is isomorphic to the conjugacy
0

class of (1, 1) in G0 n Zp and, therefore, to the p-core of N . 

Scholium. simple p-quandles of order greater than two arise in three ways:

1. a nontrivial conjugacy class in a simple group,

2. the conjugacy class of (1, 1) in H n Zp where H is a simple group and H n Zp is

constructed from an outer automorphism of H,

3. the p-core of a simple group.

The three cases are distinguished by the structure of the inner automorphism group of
the quandle.

2.10 Augmented quandles

Let G be a group acting on a quandle Q by quandle automorphisms. That is, for x, y G
and p, q Q we have

q(xy) = (qx)y, and

(p . q)x = px . py.

Assume that G contains representatives of the symmetries of Q, that is, there is a function
 : Q G satisfying
p(q) = p . q.
In particular, we have
AQ1. p(p) = p, for p Q.

16
Assume further that  satisfies the coherency condition
AQ2. (px) = x1 (p)x, for p Q, x G.
Then we have a group action on Q, Q G Q, and a function  : Q G which satisfy
AQ1 and AQ2.
Conversely, given a group action of G on a set Q, QG Q, and a function  : Q G
satisfying AQ1 and AQ2, we can define quandle operations on Q as x . y = x(y) and
x .-1 y = x(y)1 so that the action of G on Q is by quandle automorphisms.
Definition. An augmented quandle (Q, G) consists of a set Q and a group G equipped
with a right action on the set Q and a function  : Q G called the augmentation map
which satisfy AQ1 and AQ2.
With the operations mentioned above Q is a quandle, and the augmentation map is
a quandle homomorphism  : Q Conj G.
A morphism of augmented quandles from (Q, G) to P, H) consists of a group homo-
morphism g : G H and a function f : Q P such that the diagram

Q G Q G

gy

f g y

fy


P H P H
commutes. It follows that f is a quandle homomorphism.
Examples. Fix a quandle Q. Two examples of augmented quandles with underlying
quandle Q are (Q, Aut Q) and (Q, Inn Q). the augmentation in each case is the function
that has been denoted S. The action is the natural one. In the category of augmentations
of Q, (Q, Aut Q) is the terminator. That is, for each augmentation (Q, G), there is a
unique homomorphism f : G Aut Q such that

QG Q G


1f y

fy
1y

S
Q Aut Q Q Aut Q
commutes. The map f is readily defined from Q G Q.
Another example of an augmentation of Q is (Q, Adconj Q), (see section 2.3). The
function representing symmetries of Q is : Q Adconj Q, while the group action is
defined by
z(y e11 y enn ) = z .e1 y1 .e2 .en yn ,
where z Q and y e11 y enn is an arbitrary element of Adconj Q, yi Q, ei {1, 1} for
i = 1, . . . n. To show that this is a well-defined group action, it suffices to note that
z(x . y) = x .(x . y)
= z .1 y . x . y
= zy 1 x y

17
The axiom AQ1 clearly holds. Since (Q) generates Adconj Q, AQ2 reduces to the fact
that : Q Conj Adconj Q is a quandle homomorphism as noted in section 2.3.
In the category of augmentations of Q, (Q, Adconj Q) is the coterminator. That is,
for each (Q, G) there is a unique group homomorphism f : Adconj Q G such that

Q Adconj Q Q Adconj G


1f y

fy
1y


QG Q G
commutes. According to the right square, f must be the map H : Adconj Q G
described in section 2.3. To show the commutativity of the left square it suffices to show
zy = zf (y) for y, z in Q, since such y generate Adconj Q. But zy = z . y = z(y) = zf (y).
We consider now constructions in the category AQ of augmented quandles. Products,
equalizers, and limits in general are of the usual sort. For instance, the product of (Q, G)
and (P, H) has as its augmentation group GH and has as its underlying quandle QP .
However, it will take more work to describe colimits.
Let U be the forgetful functor from AQ to the category of groups, U (Q, G) = G. U
has both a left adjoint T and a right adjoint V . That U has a left adjoint T is automatic
and uninteresting. T (G) = (, G). On the other hand, the existence of a right adjoint V
is unexpected. Let G be a group. Then V (G) = (Conj G, G) where G acts on Conj G by
conjugation and the function  : Conj G G is the identity. We show that (Conj G, G)
satisfies the appropriate universal property. Let (G, H) be an augmented quandle and
f : H G a group homomorphism. We must show there exists a unique function
g : Q Conj G such that

QH Q H

gy

gf y

fy

Conj G G Conj G G
commutes. Since Conj G G is the identity, the function g must be f . The commu-
tativity of the left square states
() g(qx) = g(q) f (x)
for q Q, x H. Here, g(q) f (x) denotes conjugation of g(q) by f (x), so equals
f (x)1 g(q)f (x) where the multiplication occurs in G. Then () is equivalent to

f (qx) = f (x)1 (f )(q)f (x)

= f (x1 (q)x),

and this follows from axiom AQ2 for (Q, H).

The existence of a right adjoint for U simplifies the construction of colimits in AQ. If
(Q, G) is the colimit, lim(Qi , Gi ), then G is the colimit, lim Gi , in the category of groups.

18
Unfortunately, the forgetful functor from AQ to the category of quandles has no right
adjoint. We need another construction of augmented quandles to describe their colimits.
Let (Q, G) be an augmented quandle and f : G H a group homomorphism. Then
Q H is a right H-set with action (q, x)y = (q, xy) for q in Q and x, y in H. Define an
H-set congruence on Q H by (q, y) (p, z) if and only if yz 1 = f (x) and p = qx for
some x G. Let q y denote the congruence class of (q, y), and let Q G H, or more
simply Q H, denote the set of congruence classes. We have

(q y)z = q yz, and

q f (x) = qx 1.

Define  : Q H H by (q x) = x1 (f )(q)x.  is well-defined since axiom AQ2

holds for (Q, G). Then (Q H, H) is an augmented quandle, as can be directly verified.
We also have a function i : Q Q H given by q 7 q 1, which along with f gives a
map (i, f ) : (Q, G) (Q H, H) of augmented quandles.
Proposition. Let (Q, G) be an augmented quandle and f : G H a group homomor-
phism. Then (i, f ) : (Q, G) (Q H, H) satisfies the following universal property. For
each map of the form (g, h f ) : (Q, G) (P, K), there exists a unique map of the form
(k, h) : (Q H, H) (P, K) such that (g, h f ) = (k, h) (i, f ). Otherwise said, in the
category of augmented quandles
(, G) (Q, G)


(1,f )y
(i,f )
y
(, H) (Q H, H)
is a pushout diagram.
Proof. Let (g, hf ) : (Q, G) (P, K) be given. Denote the required function QH P
by k. There are four requirements on k. In order that k be well-defined we need
1) k(q f (x)y) = k(qx y), for q Q, x G, y H.
In order that (k, h) be a map in AQ we need
2) ( k)(q y) = (h )( q y), for q Q, y H,
and
3) k(q yz) = k(q y)h(z), for q Q, y, z H.
And so that (q, h f ) = (k, h) (i, f ) we need
4) g(q) = k(q 1), for q Q.
Together, 3) and 4) show that k must be defined as k(q y) = g(q)h(y), giving the
uniqueness of k. With this definition of k, 1) states

g(q)h(f (x)y) = g(qx)h(y).

19
This reduces to
g(q)(h f )(x) = g(qx),
which holds since (g, h f ) is a map in AQ.
Finally, 2) states that
(g(q)h(y)) = (h )(q y).
But

(g(q)h(y) = h(y)1 ( g)(q)h(y)

= h(y)1 (h f )(q)h(y)
= h(y 1 (f )(q)y)
= (h )(q y).


We will denote the function k in the proposition by g in spite of the confusion it may
cause. In this notation (g, h f ) = (g, h) (i, f ). In the case that H = K and h is the
identity function, 1 : H H, we have (g, f ) = (g, 1) (i, f ). Note also that when H = G
and f is the identity, 1 : G G, then the augmented quandle (Q G G, G) is the original
augmented quandle (Q, G). Hence, Q G G = Q.
We now consider an arbitrary colimit (Q, G) = lim(Qj , Gj ) in the category AQ.

As noted above G is the colimit, lim Gj , in the category of groups. By the preceding

proposition, for each j, (Qj , Gj ) (Q, G) factors uniquely through (Qj , Gj ) (Pj , G)
where Pj denotes Qj Gj G. Consequently, (Q, G) = lim (P , G). This reduces the
j
construction of colimits to the case where a single group G acts on all the sets Pj , and
all maps (Pj , G) (Pk , G) are of the form (f, 1).
In this case let P = lim Pj in the category of sets. Then P has a unique right

G-action determined by the G-actions on the Pj , so we might just as well have taken
this colimit in the category of G-sets. There is also a function  : P G determined
by the functions  : Pj G. It may be directly verified that with this action and 
that (P, G) is an augmented quandle. By the definition of P there is a unique function
(f, 1) : (P, G) (Q, G) determined by the maps (Pj , G) (Q, G). Also the function
(f, 1) satisfies the commutativity conditions to be a map in AQ since all the maps
(Pj , G) (Q, G) satisfy these conditions. Furthermore, all the maps (Pj , G) (P, G)
lie in AQ, so (P, G) = lim(Pj , G).

We summarize these results.
Theorem. A colimit, lim(Qj , Gj ) in AQ is isomorphic to lim(Qj Gj G, G) where G =

lim Gj in the category of groups. It is also isomorphic to (lim Qj Gj G, G) with lim Qj Gj

G taken in the category of sets. 

20
2.11 Quotients of augmented quandles described by
normal subgroups of the augmentation group
Let (Q, G) be an augmented quandle and N be a normal subgroup of G. Let G denote
the quotient group G/N with elements denoted by x for x in G. Let Q and Q/N denote
the quandle Q G G. The elements of Q are equivalence classes q of elements of Q where
q = {qn Q | n N }. The action Q G Q is given by q x = qx and the augmentation
 : Q Q is given by (q) = (q).
Let (Q, G) be an augmented quandle. In order that the quandle Q be Abelian we
need
(p . q) .(r . s) = (p . r) .(q . s).
Equivalently, (q)(r(s)) = (r)((p)(s)). That is, every element in G of the form
() (q)(s)1 (r)(q)1 (s)(r)1
equal 1. Let N be the normal subgroup of G generated by such elements. Then the
quotient (Q, G) of (Q, G) is assured to be Abelian. It is evident that (Q, G) has the
universal property that each map (Q, G) (P, H) factors uniquely through (Q, G)
(Q, G) whenever p is an Abelian quandle.
Proposition 1. Let (Q, G) be an augmented quandle such that (Q) generates G. Let N
and (Q, G) be defined as above. Then Q is the Abelianization of the quandle Q.
Proof. Let P be an Abelian quandle and f : Q P be a quandle homomorphism. We
must show that f : Q P given by f (q) = f (q) is well defined, that is, f (qn) = f (q)
for n N . If n is of the form (), then f (qn) = f (q) since P is Abelian. Since G is
generated by (Q) we may assume n is of the form (p)1 n0 (p) where f (q 0 n0 ) = f (q 0 ) for
all q 0 in Q. Then

f (qn) = f ((q .-1 p)n0 . p)

= f ((q .-1 p)n0 ) . f (p)
= f (q .-1 p) . f (p)
= f (q).

Thus, f is well-defined on Q. 
Corollary 1. Let A be a set and G the group generated by A modulo relations ab1 c =
cb1 a for conjugates a, b, c of the generators of G. Then the free Abelian quandle on A
consists of the conjugates of the generators of G. 
What has been done here for Abelian quandles can be done for many other varieties
of quandles. The method works for any variety defined by equations of the form

p .e1 1 .e2 .em m = p .f1 1 .f2 .fn n

21
where the i and j are expressions not involving p. For example, the identity for
n-quandles, p .n q = p, is of this form.
Proposition 2. Let (Q, G) be an augmented quandle such that (Q) generates G, and N
be a positive integer. Let Nn be the normal subgroup of G generated by (q)n for q in
Q. Then Q/Nn is the largest quotient of Q which is an n-quandle 
Corollary 2. Let A be a set and G = (a, a A : an = 1, a A). The free n-quandle on
A consists of the conjugates of the generators of G.
Corollary 3. The free involutory quandle on two points is isomorphic to Core Z with
generators 0 and 1.
Proof. Let A = {a, b} and G = (a, b : a2 = b2 = 1). Let Q be the quandle of conjugates
of a and b in G. Let x = ab. Then G = (a, x : a2 = 1, axa = x1 ). Each element
of G is uniquely represented as ae xk with k Z and e {0, 1}. The conjugates of a
and b are those elements of the form axk , k Z. Q = {axk | k Z}. Verification that
axn . axm = ax2nm yields an isomorphism of quandles f : Q Core Z, f (axn ) = n.
Also, f (a) = 0, f (b) = 1. 
Proposition 3. The free Abelian involutory quandle on n + 1 generators appears as

A = {(k1 , . . . , kn ) Zn | at most one ki is odd}

as a subquandle of Core Zn with generators e0 = (0, . . . , 0), e1 = (1, 0, . . . , 0), e2 =

(0, 1, 0, . . . , 0), . . . , en = (0, . . . , 0, 1).
Proof. Let G be the group presented as

(ao , ..., an : a2i = 1, ai aj ak = ak aj ai , all i, j, k),

and let Q include the conjugates of the generators of G. Then Q is the free Abelian
involutory quandle on a0 , . . . , an . As A is an Abelian quandle, there is a unique map
h : Q A such that h(ai ) = ei , i = 0, . . . , n. We will show h is an isomorphism.
Let tj = ao aj , j = 0, . . . , n. Then tj tk = tk tj . The conjugates of ai are of the form
ai . aj1 . . ajr = ai . aj1 ajr , and r may be taken to be even since ai . ai = ai .
Then ai . aj1 ajr = ai . t1 1
j1 tj2 tjr1 tjr . Thus, Q consists of elements of the form
ai . tk11 tknn with ki Z, i = 1, . . . , n. Now h(ai . tk11 tknn = ei + 2k1 (e1 e0 ) + 2k2 (e2
e0 ) + + 2kn (en e0 ) = ei + (2k1 , . . . , 2kn ). Clearly, h is surjective and injective. 
Alternatively, we may describe the free Abelian involutory quandle on n+1 generators
as
B = {(k0 , . . . , kn ) Zn+1 | exactly one ki is odd}
as a subquandle of Core Zn+1 .

22
Chapter 3

Involutory quandles

3.1 Involutory quandles and geodesics

The fact that symmetric spaces are involutory quandles and that their structure is de-
termined by distance along geodesics suggests that involutory quandles in general be
determined by some kind of geodesic. Consider, for example, the integral line quandle,
L = Core Z. Interpret L as the integral points on a line. Then for m, n L, m . n is
found by moving along the line from m through n the same distance beyond n as m is
beyond n.
The suggestion may be formalized as follows. Define an involutory quandle with
geodesics as a set Q of points with a collection of functions, called geodesics, g : L Q,
where L is the integral line quandle Core Z, satisfying three axioms

QG1. Every pair of points lies in the image of some geodesic.

QG2. Whenever a pair of points x, y lie in the image of two geodesics, f (m) =
x, f (n) = y, g(m0 ) = x, g(n0 ) = y, it is the case that f (m . n) = g(m0 . n0 ).
We denote this point f (m . n) as x . y.
QG3. A geodesic reflected through a point is a geodesic; precisely, if x is
a point and f a geodesic, then there exists a geodesic g such that for all
m, n L, there exist p, q L such that f (m) . x = g(p), f (n) . x = g(q), and
f (m . n) . x = g(p . q). See figure 3.

It is easily seen that an involutory quandle with geodesics is an involutory quandle.

The operation . is as defined in QG2.
Proposition. Every involutory quandle is representable as an involutory quandle with
geodesics.
Proof. Recall corollary 3, section 2.11, which states that L is the free involutory quandle
on two points. Let Q be the given quandle. For each pair of points x, y in Q there is a
unique quandle map f : L Q such that f (0) = x and f (1) = y. Take all such maps

23
 Figure 3.1: QG3

f (m) f (n) f (m . n)
q q q f
@  

@  
@  
q
 x
@
 @
  @
  @
q q @q

 g
g(p . q) g(q) g(p)

as geodesics. Clearly, QG1 holds. For points x, y, if f is a geodesic such that f (m) = x
and f (n) = y, then f (m . n) = x . y, hence, QG2 holds. Finally, given a geodesic f
and a point x, the geodesic g required for QG3 is that such that g(0) = f (0) . x and
g(1) = f (1) . x.
Example. Figure 3.2 displays a 2-quandle by means of geodesics. Note that some pairs of
points of the quandle lie on distinct geodesics. This particular example is algebraically
connected but does not have behaviorally distinct elements.

Figure 3.2: Geodesics

r r

r r

r r

r r

r r

24
3.2 Involutory quandles generated by two points
At this point it is appropriate to classify the involutory quandles generated by two points.
They will all be quotients of the free involutory quandle on two points, L = Core Z.

Figure 3.3: Singular quandles Cs(4) and CS(8)

1q

2 6q q
q 04
5
q7 3q
02
3q q q1

Cs(4) Cs(8)

Proposition. Any involutory quandle generated by two points is isomorphic to one of the
following

i). L = Core Z.
ii). C(n) = Core Zn , the (nonsingular) cyclic quandle of order n.
iii). Cs(4n), the quotient of C(4n) given by the congruence 2k 2k + 2n for
all k Zn , the singular cyclic quandle of order 3n.

Remark. Figure 3.3 illustrates Cs(4) and Cs(8).

Proof. It is straightforwad to check that the list induces only quotient quandles of L.
Assume now that Q is the proper quotient of L, Q = L/ . Let d be the least difference
between any two distinct equivalent points of L. d = |m n| 6= 0, m n. By using
a translation by m on L, which is an isomorphism of L, we may assume m = 0.
d = |n| = 6 0. Now n = n . 0 0 . 0 = 0, so 0 d. For all k, k = k . 0
k . d = k + 2d. Therefore, Q is a quotient of C(2d) = Core Z2d . Note that for all k,
2k = 0 . k d . k = 2k d. Similarly,
() 2k d 2k 2k + d.
We consider two cases depending on the parity of d.
Case 1. d is odd. We show p p + d for all p. If p is even, p = 2k, then () implies
p p + d. If p is odd, p = 2k d, then () implies p 2k p + d. Therefore, Q is a
quotient of C(d) = Core Zd . By the minimality of d, Q = C(d).

25
 Case 2. d is even. Let d = 2c. We see from () that Q is a quotient of Cs(4c). Assume
that Q is a proper quotient of Cs(4c). Let p be the least nonnegative integer congruent
to an element in Q from which it is distinct in Cs(4c). Reflection through p 1 shows
that p 2 has the same property unless p = 0 or p = 1. However, p cannot be 0, as the
elements equivalent to 0 in Cs(4c) are already the minimal distance d apart. Thus, p = 1.
Then there is some q, 1 < q 2n such that 1 q. We have 1 2d+1 in Cs(4c), so by the
minimality of d, q = d+1, and 1 d+1. For all k, 2k1 = 1 . k (d+1) . k = 2k1d.
Coupled with () we now have Q = C(d) = Core Zd . 
It may be asked why other axioms were not included in the definition of quandle in
order to eliminate the singular cyclic quandles as examples of quandles. There are two
responses to this question. One is that the axioms could not remain equational without
adding more operations. The other is these singular examples occur as the involutory
quandles associated to certain links (as defined in chapter 4).

3.3 Group cores

In this section we will examine some more properties of the core of a group. We have
already demonstrated (in section 2.8) the equisimplicity of a group and its core. In fact, if
the core is simple, then the core determines the group. Bruck in [3] has shown, however,
that different groups may have isomorphic cores. In particular, a nilpotent group of class
two all of whose elements have odd finite order h has a core isomorphic to that of an
Abelian group. Nonetheless, we have the following proposition.
Proposition 1. If the cores of two finitely generated Abelian groups are isomorphic, then
the groups themselves are isomorphic.
Proof. Let f : Core G Core H be an isomorphism between the cores of the finitely
generated Abelian groups G and H. By composing f with the translation by f (0) in
H (translation, y 7 y f (0), is a quandle isomorphism of Core H), we may assume that
f (0) = 0. Then f (x) = f (x) and f (2x) = 2f (x). Moreover,
() f (x + 2y) = f (x) + 2f (y).
The bijection f restricts to an isomorphism from 2G = {2x | x G} onto 2H. Using the
structure theorem for finitely generated Abelian groups, we conclude that

G
= Zr (Z2m1 Z2mk ) Odd G

where 1 m1 mk , and Odd G is the subgroup of G of elements of odd order.

Similarly,
H = Zs (Z2n1 Z2nl ) Odd H.
Now, 2G = 2Zr (2Z2m1 2Z2mk ) Odd G, and we have a similar isomorphism for
2H. Since 2G = 2H, we have Odd G = Odd H, r = s, and beginning at the first mi > 1
and the first nj > 1, the sequence m1 , m2 , . . . , mk is the same as n1 , n2 , . . . , nl . We only

26
have to show there are the same number of ones occurring in the sequence m1 , . . . , mk
as in the sequence n1 , . . . , nl . Using () we see f induces a bijection form G/2G onto
H/2H. But G/2G = Zr2 Zk2 and H/2H = Zs2 Zl2 . Since r = s and G/2G has the
same cardinality as H/2H, we have k = l. Hence G = H. 
The next proposition interprets the Abelianness of a group core. Distributivity will
be considered in section 3.4.
Proposition 2. A group G is nilpotent of class at most 2, that is, its commutator G0 is
contained in its center Z, if and only if its core is Abelian.
Proof. First note that for group cores, x . y = z if and only if xw . yw = zw. Core G
is Abelian when the identity (w . x) .(y . z) = (w . y) .(x . z). Multiplying this equation
by z 1 on the right yields

(wz 1 . xz 1 ) .(yz 1 . 1) = (wz 1 . yz 1 ) .(xz 1 . 1).

Thus, the group core is Abelian if and only if it satisfies

(w . x) . y 1 = (w . y) . x1 .

that is, y 1 w1 wx1 y 1 = x1 y 1 wy 1 x1 , which may be rewritten as

() w[x, y] = [x1 , y 1 ]w.
Assume Core G is Abelian. then for w = 1, [x, y] = [x1 , y 1 ], and so, generally

w[x, y] = [x, y]w.

Hence, G0 Z.
Now assume G0 Z. In order to show (), it suffices to show [x, y] = [x1 , y 1 ]. But

[x1 , y 1 ] = xyx1 y 1 = yx[x, y]x1 y 1 = [x, y]yxx1 y 1 = [x, y].


Proposition 2 generalizes Soublins result [15] page 101, which, in the nomenclature
of quandles, states that for any group G of exponent 3, Core G is Abelian if and only if
G is nilpotent of class at most 2.

3.4 Distributive quandles

A property of quandles which is weaker than Abelianness is distributivity, satisfaction
of
QDist. x .(y . z) = (x . y) .(x . z).
For each x in a distributive quandle, the map

P (x) : y 7 x . y

27
is a quandle homomorphism, called the projection from x. Projections need not be either
injective or surjective.
Lemma 1. For an involutory quandle, distributivity is equivalent to satisfaction of either
(1) or (2).
(1) x . z . y = x . y . z . x.
(2) x . y . x . z = x . z . x . y.

Figure 3.4: Distributivity

q x.y
q x.z.x


 HH
J
q
H
 J HH
y q " Jq Hq y

x .q y "


HHH
PP y "
  "
J
  J PP q"
J
HH
q x.z J"" PP q
q x Hq
" Jq
x
 J

q  
  

J 
q   z q
"
x   " zJ

J 
@   x.z J
q

Jq 
@   Jq


 z
q x.y.z
y.z @
@ 


q




@  x.y.x
@q

QDist (1) (2)

Proof. (1) = (2).

(x . y) . x . z = (x . y) . z . x .(x . y) by (1)
= (x . y . z) . x . y . x . y
= (x . z . y . x) . x . y . x . y by (1)
= x . z . x . y.

(2) = QDist.

(x . y) .(x . z) = x . y . x .(z . x) . x
= x .(z . x) . x . y . z by (2)
= x .(y . z).

28
QDist = (1).
x . z . y = x .(y . z) . z
= (x . y) .(x . z) . z by QDist
= x . y . z . x.

Proposition 1. The core of a group is distributive if and only if every element of the group
commutes with each of its conjugates.
Proof. Simplify the distributivity condition QDist by multiplying on the right by z 1 .
xz 1 .(yz 1 . 1) = (xz 1 . yz 1 ) .(xz 1 . 1).
This yields the identity involving two variables
u .(v . 1) = (u . v) .(u . 1),
which reduces to the identity
u(v 1 uv) = (v 1 uv)u.

Groups in which conjugate elements commute have been studied by Burnside and
others. If such a group is generated by two elements, then its commutator is contained
in its center, and so its core is Abelian. This suggests that a distributive 2-quandle
generated by three points is Abelian.
Proposition 2. The free distributive 2-quandle on three points appears as
Q = {(m, n) Z Z | at most one of m and n is odd}
as a subquandle of Core(Z Z) with (0, 0), (1, 0), and (0, 1) as generators.
Proof. Let D be a distributive 2-quandle and f (0, 0), f (1, 0), f (0, 1) three points of D.
We extend f to all of Q. Since Core Z is the free 2-quandle on two points, we can extend
f uniquely to Z0 and 0Z. Inductively define f (2m, n) as f (2m1, n) . f (2m1, 0)
for positive integers m, and f (2m, n) as f (2m + 2, n) . f (2m + 1, 0) for negative integers
m. For n = 0 this agrees with the previous definition of f (2m, 0). We show
(1) f (2m, n) . f (p, 0) = f (2p 2m, n)
by induction on d = |2m p|. (1) holds for d = 0. It suffices to prove (1) for 2m > p
by symmetry at f (p, 0). (1) holds for d = 1 by definition of f (2p 2m, n). Assume (1)
holds for d 1 and d 2. Then
f (2m, n) . f (p, 0) = f (2m 2, n) . f (2m 1, 0) . f (p, 0)
= (f (2m 2, n) . f (p, 0)) .(f (2m 1, 0) . f (p, 0))
= f (2p 2m + 2, n) . f (2p 2m + 1, 0)
= f (2p 2m, n).

29
Analogously, we may define f 0 (m, 2n) where f 0 (0, q) = f (0, q) and f 0 (p, 0) = f (p, 0) so
that
(10 ) f 0 (m, 2n) . f 0 (0, q) = f 0 (m, 2q 2n).
From distributivity, we have the identity of lemma 1,

x . z . x . y = x . y . x . z.

Taking x = f (0, 0), z = f (0, n), and y = f (m, 0), we find that f 0 (2m, 2n) = f (2m, 2n).
Thus, we may eliminate the primes. We have defined f on all of Q. It remains to show
that f is a homomorphism. We will show
(2) f (2m, n) . f (0, q) = f (2m, 2q n).
Now, (2) holds for q = 0, and by reflection through (0, 0), it suffices to show (2) for q > 0.
Assume for a moment that (2) holds for q = 1. Then by induction on q > 1, we have

f (2m, n) . f (0, q) = f (2m, n) .(f (0, q 2) . f (0, q 1))

= f (2m, n) . f (0, q 1) . f (0, q 2) . f (0, q 1)
= f (2m, 2q n).

Thus, it suffices to show (2) for q = 1:

(3) f (2m, n) . f (0, 1) = f (2m, 2 n).
By a similar induction it suffices to show (3) holds for n = 0 and n = 1. The n = 0 case
is a special case of (10 ), and the n = 1 case follows from projecting f (m, 0), f (0, 0), and
f (m, 0) from f (0, 1). Thus, (2) holds. Similarly, we have
(20 ) f (m, 2n) . f (p, 0) = f (2p m, 2n)).
Finally, we show for (x, y) in Q that
(4) f (x, y) . f (2p, q) = f (4p x, 2q y)).

f (x, y) . f (2p, q) = f (x, y) .(f (0, q) . f (p, 0))

= f (x, y) . f (p, 0) . f (0, q) . f (p, 0)
= f (2p x, y) . f (0, q) . f (p, 0)
= f (x 2p, y 2q) . f (p, 0)
= f (4p x, 2q y).

Along with (40 ) we have shown that f is a homomorphism. 

As a corollary, we have that any distributive 2-quandle generated by three points is
Abelian.
Soublin [15] constructs a nonAbelian distributive quandle M81 of order 81. It is the
smallest nonAbelian distributive quandle satisfying x . y = y . x.

30
3.5 Involutions
A natural occurrence of involutory quandles is that of the set of involutions in a group
G. More generally, for n a positive integer

Qn (G) = {x G | xn = 1}

is an n-quandle with conjugation as the quandle operation. Qn is a functor: (groups)

(n-quandles). Adjoint to Qn is the functor AdQn : (n-quandles) (groups). For an
n-quandle Q, AdQn (Q) is a group presented as

AdQn (Q) = (p, for p Q : pn = 1, p . q = q 1 p q, for p, q Q).

The degree to which AdQn relates n quandles to groups may be seen in part by the
following proposition.
Proposition. For an n-quandle Q, the order of the group AdQn (Q) is no greater than n
raised to the order of Q. |AdQn (Q)| n|Q| .
Lemma. Let the elements of an n-quandle Q be well ordered. Than any element of
AdQn (Q) may be written as a finite product of the generators in nondecreasing order.
Proof. The proposition follows directly from the lemma. We prove the lemma by a double
induction.
Let z = x1 xm be an element of G = AdQn (Q), each xi in Q, not necessarily
distinct. By induction on m, the length of the product, we may assume that a product
of length less than m may be written with the xi s in nondecreasing order. So we may
assume x1  x2   xm1 . By transfinite induction on xm , we may assume that
products of length m whose first m 1 terms are in order and whose n-th term is less
than xm may be written in nondecreasing order. We show now that z may be rewritten
in order without increasing its length. If xm1  xm , then Z is already in order.
Otherwise, xm1  xm . Let y = xm .-1 xm1 . Then xm1 xm = y xm1 , and so
z = x1 xn2 y xn1 . By the first induction we may write x1 xn2 y in nondecreasing
order as y 1 y n1 , so z = y 1 y n1 xn1 . By the second induction, using the fact that
xn1 xn , we may write z in nondecreasing order. 
The bound of 2|Q| is achieved for finite 2-quandles satisfying x . y = x. Since AdQn (Q)
maps onto Inn Q, we have as a corollary that | Inn Q| n|Q| .

3.6 Moufang loop cores

The functor Core : (groups) (2-quandles) may be extended from groups to Moufang
loops. Recall that a loop is a set G equipped with a binary operation (usually written
multiplicatively) with an identity element 1, x1 = 1x = x, such that for all a, b in G
there are unique solutions to the equations xa = b and ay = b. Thus, a loop is a

31
quasigroup with an identity element. A loop has the inverse property when it has an
operation x 7 x1 satisfying (xy 1 )y = x and y(y 1 x) = x. Such loops also satisfy
x1 x = xx1 = 1, (x1 )1 = x, (xy)1 = y 1 x1 , and x1 (xy) = y = (yx)x1 .
A loop is a Moufang loop if it satisfies
(1) (xy)(zx) = (x(yz))x.
Moufang loops have the inverse property, and they satisfy the identities
(2) ((xy)z)y = x(y(zy)), and
(3) x(y(xz)) = ((xy)x)z.
Although a Moufang loop need not be a group, for it need not be associative, any subloop
generated by two elements is a group. Also, if x, y, z are three elements of a Moufang
loop which associate, that is, (xy)z = x(yz), then the subloop which they generate is a
group.
For a discussion of Moufang loops and proofs of the above statements see chapter vii
of Brucks book [3]. A basic problem of Moufang loops (and of loops and quasigroups
in general) is to determine when two loops are isotopic. An isotopy of two quasigroups
G, H consists of three bijections f, g, h : G H such that for all x, y in G
f (x)g(y) = h(xy).
Bruck defined the core of a Moufang loop as the underlying set of the loop along with
the binary operation (x, y) 7 yx1 y (which we denote x . y) in order to have a property
of Moufang loops invariant under isotopy. If two Moufang loops are isotopic, then their
cores are isomorphic. Work on cores of loops more inclusive than Moufang loops may be
found in Robinson [12] and Burn [4].
Proposition 1. The core of a Moufang loop is a 2-quandle.
Proof. Axioms Q1 and QInv hold since they only involve two variables, and they hold
in the case of a group. To show axiom Q3 first observe that xz . yz = (x . y)z. Indeed,
xz . yz = (yz)(xz)1 (yz)
= (yz)(z 1 x1 )(yz)
= ((yz)z 1 )(x1 (yz)) by (1)
= y(x1 (yz))
= (yx1 y)z by (3)
= (x . y)z.
Similarly, zx . zy = z(x . y). Also (x . y)1 = x1 . y 1 . Hence,
(x . z) .(y . z) = (zx1 z) .(zy 1 z)
= z(x1 . y 1 )z
= z(x . y)1 z
= (x . y) . z

32
 
A Moufang loop is commutative if xy = yx. We will write commutative Moufang
loops additively. We have from (2) the identity
((x + y) + z) + y = x + (y + (z + y)),
which is equivalent to
(4) (x + y) + z = ((2z + z) + x) y.
Proposition 2. The core of a commutative Moufang loop is distributive.
Proof. x . y = 2y x. 2(x . y) = 2x . 2y.
(x . y) .(w . z) = (2y x) .(2z x)
= (2y . 2z) x
= 2(y . z) x
= x .(y . z)

Proposition 3. The core of a Moufang loop is distributive if and only if every element
commutes with each of its conjugates.
Proof. The proof is identical to that of proposition 1 in section 3.4. 
Proposition 4. The core of a commutative Moufang loop G is Abelian if and only if G is
a group.
Proof. The proof that the core of an Abelian group is an Abelian quandle is direct.
Let G be a commutative Moufang loop with an Abelian core. Abelianness gives
(4z 2y) (2x w) = (4z 2x) (2y w).
Setting z = 0 and negating x and y, we find
(w + 2x) + 2y = (w + 2y + 2x).
Since the subloop generated by w, 2x and 2y is associative, w, 2x, and 2y associate in
any order. Note
(w + x) + 2y = ((2x + 2y) + w) x by (4)
= ((2y + w) + 2x) x
= (2y + w) + x.
Hence, w, x, and 2y associate. Finally,
(w + y) + x = ((2y + x) + w) y by (4)
= ((x + w) + 2y) y
= (x + w) + y.
Thus, G is associative. 

33
3.7 Distributive 2-quandles with midpoints
Definition. Let Q be a 2-quandle. A midpoint between two points x and y of Q is a
point m such that x . m = y (and so y . m = x). Q is said to have midpoints if there is
a midpoint between any two of its points.
If Q is a finite 2-quandle with midpoints, then midpoints are unique. Midpoints need
not be unique in the infinite case. Consider, for example, Q = Core R/Z. Between 0 and
1
2
lie the midpoints 14 and 43 .
The assumption that a commutative Moufang loop is 2-divisible, that is, for each
element x there exists an element y such that 2y = x, implies that its core has midpoints.
Proposition. Let Q be a distributive 2-quandle with midpoints and 0 be a fixed element
of Q. Then Q has the structure of a 2-divisible commutative Moufang loop, L(Q), by
taking x + y = 0 . m where m is any midpoint between x and y.
Lemma. Let Q be a distributive 2-quandle with midpoints. Let x, y Q. Then any two
midpoints between x and y are behaviorally equivalent.
Proof of lemma. Let m be a midpoint between x and y. Let z be a point of Q. We show
that z . m depends only on z, x, and y, not on m. Take n to be a midpoint between z
and y. Using a variant of the identity (1) in lemma 1, section 3.4, we have
z.m = z.n.m.z.n
= y.m.z.n
= x . z . n.
The last expression does not depend on m. 
Proof of proposition. Addition is well-defined by the lemma. Clearly, addition is commu-
tative, and 0 is an additive identity. To show that L(Q) is a loop, we must show that for
all x, y Q, there is a unique z in Q such that x + z = y. Let m be a midpoint between 0
and y, and set z = x . m. Then x + z = y. Conversely, if x + z = y, and m is a midpoint
between x and z, then m is a midpoint between 0 and y, and by the lemma, z = x . m
Next we show the Moufang identity for commutative loops:
(x + y) + (x + z) = x + (x + (y + z)).
Let n be a midpint between y and z, m a midpoint between x and n, and p a midpoint
between x and z. As shown in figure 3.5, by projecting p, m, and p . m from x, we
conclude that p . m is a midpoint between x and y. Projecting p, m, and p . m from 0,
we find 0 . m is a midpoint between x + y and x + z. Hence, 0 .(0 . m) = (x + y) + (x + z).
Now take q to be a midpoint between 0 . m and n. Then x . q = n . m . q, which by
lemma 1, section 3.4, equals n . q . m . n. So
x.q = n.q.m.n
= 0.m.m.n
= 0 . n = y . z.

34
 Figure 3.5: Midpoints
y
r
n
r
zr
A
y+z r
A x+ y 0.m
x +rz
x + (y + z)
A r r


r
A  ,
,
,


,



A  ,


A  
,

n r
P
,  ,
A  ,
 ,

p . m Ar m r, r
PPr,




, 
p ,PPPr


0.m
A

 A,,

,, q
 ,A

,

 , A

,

r
, A
r
,

0 A
0

A
r

r
A
x
x

Therefore, q is a midpoint between x and y + z. As indicated in figure 3.5, we find that

0 . m is a midpoint between x and x + (y + z) by reflecting 0, n, and y + z through q.
Hence, 0 . m also equals x + (x + (y + z)).
Thus, L(Q) is a Moufang loop. Finally, L(Q) is 2-divisible, since if m is a midpoint
between 0 and x, then x = 2m. 
The functors Core and L are inverse to each other, so we have an isomorphism of
categories; the category of 2-divisible commutative Moufang loops is isomorphic to the
category of pointed distributive 2-quandles with midpoints. By proposition 4 of sec-
tion 3.6, this isomorphism restricts to an isomorphism from the category of 2-disvisible
Abelian groups to the category of pointed Abelian 2-quandles with midpoints.

35
Chapter 4

Algebraic topology and knots

4.1 The fundamental quandle of a pair of spaces.

Consider the category P 0 of pairs of topological spaces (X, K), K a subspace of X, where
a map f : (X, K) (Y, L) in P 0 is given by a continuous map f : X Y such that
f 1 (L) = K. Two maps f, g : (X, K) (Y, L) in P 0 are said to be homotopic, written
f g, if there is a map h : (X I, K I) (Y, L) in P 0 , where I is the unit interval,
such that h(x, 0) = f (x) and h(x, 1) = g(x) for all x in X. This concept of homotopy
in P 0 yields a category [P 0 ] of pairs of topological spaces where the maps are homotopy
classes of maps in P 0 .
For our purposes we must consider pointed spaces. Let P be the category whose
objects are pairs of spaces (X, K) along with a distinguished point in X K, called
the basepoint, and whose maps between pairs preserve the basepoint. If (X, K) is such
a pointed pair, let X I be the quotient (X I)/({} I) of X I. Two maps
f, g : (X, K) (Y, L) in P are homotopic, written f g, if there is a map h : (X
I, K I) (Y, L) in P such that h(x, 0) = f (x) and h(x, 1) = g(x) for all x in X. Let
[P] be the resulting homotopy category for P.
One object of [P] is the circle S = (S 1 , ) where S 1 is the unit circle in the complex
plane with the basepoint at 1. S is a cogroup in [P], that is, S has a group structure
in [P op ]. Since we will be dealing with this cogroup in some detail, let us describe this
structure explicitly. We need a comultiplication : S S S, a coinversion : S S,
and a coidentity S (1, ). There is only one map S (1, ) so it is the coidentity.
Define by (z) = z 1 . Represent S S as S {1, 2} with the points (1, 1) and (1, 2)
identified. Then define by
 2it
it (e , 1) for 0 t
(e ) =
(e2it , 2) for t 2.

With these definitions S becomes a cogroup. The homotopy classes of maps from S to
any object (X, K) form a group (X, K) which is just the fundamental group of X K,
1 (X K).

36
 A map : S (X, K) is called a loop in X K. If and are two loops, we let
1 = and = ( ). Then 1 1 and () ().
Unfortunately, (X, K) gives only partial information about the way that K is situ-
ated in X, no more than the circumstantial information concerning X K. For instance,
when K is a knot in 3-space X, (X, K) is the knot group, and although the knot group
distinguishes many knots, it fails to distinguish the square knot from the granny knot.
We will replace the pair S = (S 1 , ) by a pair where the subspace forms an integral part
of the whole. In doing so, we will not have a cogroup, but only a coquandle.
Let N be the object (X, 0) in P where X is the subspace of the complex numbers
consisting of the union of the closed unit disk {z C | |z| 1} and the rope {z
C | z real and 1 z 5}, where 0 denotes {0 C}, and the basepoint of X is 5.

Figure 4.1: The noose N

'$N
q q
0
5=
&%

We will show that N is a coquandle, but not directly. Instead, we will show that S and
N together form a co-augmented-quandle. This entails the construction of two maps in
P, a : N N S and d : S N so that in [P]op the two axioms AQ1 and AQ2,
as stated in section 2.10, are satisfied. Once this is done, S and N will represent a
contravariant functor from [P] to the category of augmented quandles which will extend
the fundamental group functor.
Let d : S N wrap the circle around the disk of N by way of the rope of N .

5 8t/ for 0 t /2,
d(eit ) = e2i(t/2) for /2 t 3/2,
8t/ 11 for 3/2 t 2.

The map d may be interpreted as the boundary of N . Let the map a : N N S place
the disk of N onto the disk of N S, then stretch the rope of N along the rope of N S
to the basepoint and around the circle of N S.

zN if |z| 1,
a(z) = 2z 1 N if 1 z 3,
i(z3)
e S if 3 z 5.
In order to show that a gives a group action, we must show that the diagram
a
N N S


ay

ya1
1
N S N S S
37
commutes up to homotopy. Both maps f = a(a 1) and g = a(1 ) place the disk of
N onto the disk of N S S then stretch the rope of N along the rope of N S S and
around each circle of N S S. Restricted to the disk of N , f equals g. They only differ
with regard to the rate that they stretch the rope. A homotopy h : N I N S S
between f and g is


f (z) if |z| 1,
f ((z 1)(1 t/2) + 1) if 1 z 3,

h(z, t) =

f (z t) if 3 z 4,
f (5 (5 z)(1 + t)) if 4 z 5.

In order to show AQ1 holds we must show that the diagram

a 1d
N N S N N
PP
P

PP
P1 PP
q
P
N

commutes up to homotopy. Represent N N as N {1, 2} with basepoints identified.

Then the map : N N N is given by (z, n) = z. Let f = a(1 d). Then
f : N N is formulated as


z if |z| 1,
2(z 1) + 1 if 1 z 3,


f (z) = 5 8(z 3) if 3 z 3.5
exp(2i(z 3.5)) if 3.5 z 4.5




8(z 4.5) + 1 if 4.5 z 5.

Verbally described, f places the disk of N onto itself, then stretches the rope along and
back itself, around the disk, and back to the basepoint. In the category P a homotopy is
not allowed to pass any point of X A through A; in particular, the rope of N may not
pass through the origin. The required homotopy, f 1, may be made by rotating the
disk counterclockwise one revolution while contracting the string to its initial position.
Figure 4.2 illustrates such a homotopy.
In order to show the axiom AQ2 holds, we must verify that the diagram
1
S S S S 
S S
0
d 10
0d

N a N S

commutes up to homotopy. Both of the compositions describe a loop in N S starting

at the basepoint, clockwise around the circle of N S, around the boundary of N in
N S, and counterclockwise around the circle of N S back to the basepoint. The two
loops only differ by their rate, and hence are homotopic.

38
Figure 4.2: f = a(1 d) : N N is homotopic to 1
'$

t=0 q-
(f ) &%
'$

t = 1/4 q-
&%
'$

t = 1/2 q-
&%
'$

t = 2/3 AKq
&%
'$

t = 5/6 q


&%
'$

t=1 q-
(1) &%

39
 Thus, S and N together represent a contravariant function AQ from [P] to the cat-
egory of augmented quandles. Let (X, K) be an object of P. A noose about K is a
map : N (X, K) in in P. If is a loop in X K and is a noose about K, let
= a( ) and () = d. We call () the boundary loop of . If is another loop
in X K, we have
() (), and
() 1 ().
If and are nooses about K, let . = (). Then the homotopy classes of nooses
in X about K form a quandle, Q(X, K). We name Q(X, K) the fundamental quandle
of (X, K), and we name AQ(X, K) the fundamental augmented quandle of (X, K).
AQ(X, K) is the fundamental quandle augmented by the fundamental group of X K.

Figure 4.3: q and q 1

N 1 q N 1 q

N 2 q N 2 q
q q 1

As N is a coquandle, a presentation of its structure is in order. The structure is given

by two maps, q, q 0 : N N N , q used to represent . and q 0 to represent .-1 . The map
q is to be homotopic to a(1 d). Let N N be N {1, 2} with basepoints identified.
Then such a map is


(z, 1) if |z| 1,
(4(z 1) + 1, 1) if 1 z 2,


q(z) = (5 4(z 2), 2) if 2 z 3,
(exp(2i(z 3)), 2) if 3 z 4,




(4(z 4) + 1, 2) if 4 z 5.

The map q places the disk of N onto the first disk of N N and stretches the rope of
N along the first rope of N N and around the boundary of the second N . See figure
4.3. The map q 0 is defined similarly except that (exp(2i(z e)), 2) is used in the case
3 z 4.
In forthcoming proofs we will have occasion to compose nooses with paths as well as
loops. If is a noose in X about K with basepoint , and is a path in X K from
to 0 , then let denote the composition. is a noose in X about K with basepoint

40
0 . Let Disk denote the noose with basepoint (1), (Disk : N (X, K) is constantly
(1) on the rope of N ). Let Rope denote the path from (1) to along the image of .
Then (Disk )(Rope ).

4.2 The fundamental quandle of a disk

Let (D, 0) be the object in P where D is the closed unit disk in the complex plane and
0 is the center of the disk.
Proposition. Any element of Q(D, 0) is uniquely representable as a noose f : N (D, 0)
 in
re if |z| 1, z = rei ,
f (z) =
1 if 1 z 5,

for a unique integer n. n is the winding number of the boundary of f around 0.

Proof. In order to simplify computations, we may take N to be (D, 0) as they are
homotopically equivalent in the category P. To further simplify the description of the
homotopies, we will take D to be S 1 [0, ) union a point at infinity (to correspond to
the center of D).
Let f : D D be continuous such that f 1 () = . First we construct a homotopy
so that we may assume f (S 1 0) S 1 0. Since f (S 1 0) is compact and misses ,
there is an x0 such that f (S 1 0) S 1 [0, xo ]. We use a homotopy to squeeze S 1 [0, x0 ]
to S 1 0. Using the notation f (, x) = (f 1 (, x), f 2 (, x)), define h : D I D by
 1
(f (, x), f 2 (, x) tx0 ) if x tx0
h(, x, t) =
(f 1 (, x), 0) if x tx0 .

Then h(, x, 0) = f (, x) while h(, 0, 1) S 1 0. We may therefore assume that

f (S 1 0) S 1 0.
Now f restricted to S 1 0 gives a loop in S 1 0, so is homotopic to the map (, 0)
(n, 0), where n is the winding number. Then there is a map h : S 1 0 I S 1 0 such
that h(, 0, 0) = f (, 0) and h(, 0, 1) = (n, 0). We construct a homotopy H : DI D
extending h. Let 
f (, x t) if x t,
H(, x, t) =
h(, 0, t x) if x t.
Then H(, x, 0) = f (, x) while H(, 0, 1) = (n, 0). Thus, we may assume f (, 0) =
(n, 0).
Finally, we show f is homotopic to the map (, x) 7 (n, x). We will define a
homotopy H : D I D which spreads the influence of the restriction of f to S 1 0
down the entire cylinder S 1 [0, ). Let

H(, x, t) = (f 1 (, (1 t)x), f 2 (1 t)x + tx).

41
Then H(, x, 0) = (f 1 (, x), f 2 (, x)) = f (, x), and H(, x, 1) = (f 1 (, 0), f 2 (, 0) + x) =
(n, x). We should check that setting H(, t) to leaves H continuous. Given N > 0,
we must show there exists an M > 0 such that for all x > M , H 2 (, x, t) > N . Let M1
be such that if x > M1 , then f 2 (, x) > N . Let M = max{2M1 , 2N }. Suppose x > M .
Then x/2 > M1 and x/2 > N .
Case 1. t 21 . H 2 (, x, t) = f 2 (, (1 t)x) + tx tx x/2 > N .
Case 2. t 12 . H 2 (, x, t) = f 2 (, (1 t)x) + tx f 2 (, (1 t)x) which is greater
than N since (1 t)x x/2 > M1 .
Thus, H gives a homotopy of f to the map (, x) 7 (n, x). 
The quandle operation on Q(D, 0) is first projection, that is x . y = x.

4.3 The Seifert-Van Kampen theorem

Recall the Seifert-Van Kampen theorem for the fundamental group. Let X be an arc-
wise connected topological space with a basepoint, and let {Ui } be a covering of X by
arcwise connected open sets closed under pairwise intersections such that each open set
Ui contains the basepoint. Then
1 (X) = lim 1 (Ui ).

We will prove an analogous theorem for the fundamental quandle of a pair of spaces.
Theorem. Let (X, K) be an object in P. Let {Ui } be a covering of X closed under
pairwise intersection. Assume for each index i that Ui is a neighborhood of Ui K and
that Ui k is arcwise connected and contains the basepoint of X. Then
AQ(X, K) = lim AQ(Ui , Ui K).

Proof. Let AQ(X, K) = (Q, G). Then by the Seifert-Van Kampen theorem, G = 1 (X
K) = lim 1 (Ui K). By the theorem of section 2.10, the lim AQ(Ui , Ui K) is then

of the form (L, G). By the universal property of (L, G), there is a unique : L Q
determined by the maps AQ(Ui , Ui K) AQ(X, K) = (Q, G). We will show is an
isomorphism.
Surjectivity of . By the theorem of section 2.10, it suffices to show that every noose
about K is homotopic to some where is a noose in some (Ui , Ui K), and is a
loop in X K. Let be an arbitrary noose about K. (0) lies in some Ui . Since 1 (Ui )
is a neighborhood of 0 in N , there is some r > 0 such that |z| r implies (z) Ui .
Using the homotopy

(zr) if |z| 1
H(z) =
((z 1)(5 4)/r + r) if 1 z 5,
we may adjust so that we may assume (z) Ui for |z| 1. Choose a path in
Ui K from (1) to . Let be the noose (Disk ) in (Ui , Ui K) and be the loop

42
 1 (Rope ) in X K. Then = (Disk ) 1 (Rope ) (Disk )(Rope ) as
required.
Injectivity of . It suffices to show that if 0 0 where is a noose in (Ui , Ui K),
and 0 are loops in X K, and 0 is a noose in (Ui0 , Ui0 K), then as elements of L,
= 0 0 . Let H effect the homotopy 0 0 ; H : N I X, H(z, 0) = ()(z),
H(z, 1) = (0 0 )(z). The inverse images of the open sets Ui under H(0, t) cover the unit
interval I. Hence, we may divide I into subintervals 0 = t0 < t1 < < tn = 1 so that
for each j = 1, . . . , n there is an index i(j) for which H(0, t) Ui(j) when ti1 t ti .
We may take i(1) = i and i(n) = i0 . There is an r > 0 so that for |z| r and tj1 t tj
we have H(z, t) Ui(j) . One r suffices for all j. By appropriate adjustments of , 0 ,
and H we may assume r = 1.
Figure 4.4 illustrates the remainder of the proof. For j = 0, 1, . . . , n define the nooses
j = H(z, tj ) in (X, K). Also choose paths j from j (1) to in Ui(j1) Ui(j) K for
j = 1, 2, . . . , n 1, and set 0 = Rope and n = Rope 0 .

Figure 4.4: Seifert-Van Kampen noose homotopy

0 1 2 n
5

=0 0 =n

0 1 2 n
=0 1 2 0 =n
1

q q q q

I
0 = t0 t1 t2 tn = 1

For j = 0, . . . , n define j to be the noose (Disk j )j in (Ui(j1) Ui(j) K, Ui(j1)

Ui(j) ) and j to be the loop j1 (Rope j ) in X K. Then = 0 , 0 = n . As nooses

43
in (Ui , Ui K) and (Ui0 , Ui0 K), 0 0 and 0 0 n n , respectively. In order to
show equals 0 0 in L, we will show for j = 1, . . . , n that j1 j1 equals j j .
1
Fix j between 1 and n. We must show j1 = j j j1 . Let j be the path
from j1 (1) to j (1) in Ui(j1) K given by j (t) = H(1, tj1 + t(tj tj1 )). The
portion of the homotopy H on [1, 5] [tj1 , tj ] yields a homotopy in X K of j to
(Rope j1 )(Rope j )1 . Thus
j1 j1 = j1
1
(Rope j1 )(Rope j )1 j
1
= j1 j j
1
as elements of 1 (X K). The noose j j1 j j lies entirely in Ui(j1) as does j1 .
Moreover, the restriction of H to {|z| 1} [tj1 , tj ] has an image in Ui(j1) and yields
1 1
a homotopy of j1 to j j1 j j in (Ui(j1) K, Ui(j1) ). Thus, j1 = j j1 j j =
1
j j1 j as elements of L. 
Corollary. Let (X, K) be an object in P. Let U and V be an open covering of X such
that U K, V K, and U V K are arcwise connected and contain the basepoint of
X. Then AQ(X, K) is the pushout
AQ(U V, U V K) AQ(V, V K)


y y
AQ(U, U K) AQ(X, K)


4.4 Applications of the Seifert-Van Kampen theo-

rem
The proposition in section 4.2 implies that if K is a point in the plane X, then
AQ(X, K) = (Conj F, F ) where F is the free group on one element. Consider a gen-
eralization where X is a 2-manifold.
Proposition 1. Let K be a point in a 2-manifold X, and let G = 1 (X K). Let U be
the union of a small disk around K and a path to the basepoint, and let be a loop
in U winding once around K. Let F be the free group generated by one element and
f : F G send the generator of F to the homotopy class of . Then
AQ(K, X) = (Conj F F G, G).
Proof. Let V = X K. Then by the Seifert-Van Kampen theorem,
AQ(U V, ) = (, F ) AQ(V, ) = (, G)


y y
AQ(U, K) = (Conj F, F ) AQ(X, K)

44
is a pushout square. Hence, by the theorem of section 2.10, we have AQ(K, X) =
(Conj F F G, G). 
Another generalization of the same proposition is where X remains the plane and K
is a discrete subset of X. For each point k in K let (k) be a small loop winding once
around k. Then G = 1 (X K) is free on {(k) | k K}. For k in K let Fk be the free
group on (k) and Fk G the inclusion homomorphism.
Proposition 2. X, K, G, Fk as above. Then
[
AQ(X, K) = ( Conj Fk Fk G, G).
kK

Proof. For k in K let Uk = (X K) {k}. Then AQ(Uk , {k}) = (Conj Fa Fa G, G) by

proposition 1. If k, l K, k 6= l, then AQ(Uk Ul , ) = (, G). Thus, AQ(X, K) is the
limit of the diagram

*
 (Conj Fk Fk G, G)


 :


(, G)

   ..
XXX .
HH XXX
HH XXz
H
HH
j
H (Conj Fl Fl G, G)

S
which is ( kK Conj Fk Fk G, G). 
In most of the applications that follow K will be a submanifold of X of codimension
two. When this is the case we will try to restrict our attention to those nooses in X
winding once about K. For instance if X is an oriented 2-manifold and K is a discrete
subset of X, winding once is well-defined.
Let X be the plane and K a discrete subset of X. Orient X and let Q consist
of the nooses which wind exactly once (in the positive sense) around a point of K,
and let AQ(X, SK) be Q augmented by G = 1 (X K). Then by proposition 2,
AQ(X, K) = ( kK {(k)} Fk G, G). Then Q is just the free quandle on |K| elements,
and is isomorphic to the quandle consisting of the conjugates of the generators of G.

4.5 Knot quandles

Recall that a knot is a subspace K of the 3-sphere X = S 3 which is homeomorphic to
a circle. A link is a subspace K of X homeomorphic to a disjoint union of circles. Two
knots or links K and K 0 are equivalent if there is a homeomorphism h of X such that
h(K) = K 0 , that is, in the category P, (X, K) is homeomorphic to (X, K 0 ). We will

45
deal with oriented spaces; we assume X and K are endowed with orientations. (If K is
a link, then we assume each component circle is oriented.) An oriented equivalence of K
and K 0 is an orientation-preserving homeomorphism h of X such that h(K) = K 0 and h
preserves the orientation of each component of K. An equivalence class of oriented knots
or links (under oriented equivalence) is called an oriented knot type or oriented link type,
respectively. We assume all knots, links, and equivalences are oriented and henceforth
omit the adjective oriented.
The fundamental group of X K, 1 (X K), is called the knot group (or link group).
This definition assumes either that X K has a designated basepoint or else that the
knot group is only defined up to noncanonical isomorphism; we assume a basepoint.
Recall that a tame knot is a knot equivalent to a closed polygonal curve in X. Some
of the results below are restricted to tame knots and tame links.
Associated to a knot (X, K) we have the fundamental quandle Q(X, K). An element
of Q(X, K) is represented by a noose about K. The boundary loop () may or may
not link with K as shown in figure 4.5.

Figure 4.5: Noose boundary loops

q
K
q

In order to decide when a loop in X K links once with K, it suffices to choose a

generator of H1 (X K) = Z. Then loops homologous to that generator have linking
number 1 with K. Since we assume K and X have orientations, such a generator may
be naturally chosen (say, by the right-hand rule).
Let f be the compostion

Q(X, K) 1 (X K) H1 (X K),

and let Q = Q(X, K) = f 1 (generator). Then Q consists of the nooses linking once
with K. Q is an invariant of the knot type of K; if (X, K)
= (X, K 0 ), then Q(X, K)
=
0
Q(X, K ). The boundaries of the nooses in Q are called meridians of K. 1 (X k) acts
on Q(X, K) as well as Q(X, K). Call Q the knot quandle of the knot (X, K).

46
4.6 A presentation of the knot quandle

Figure 4.6: The figure-8 knot

-
a3
x1 x2
n
@
@
x@
3@ a2
a1

@
R
@
@
R
@

? 6
a4
6
 x4

Let (X, K) be a tame knot, such as the figure-8 knot shown in figure 4.6. Recall
Wirtingers presentation for the knot group.

Figure 4.7: A knot crossing

x x3
 1 @
a1 R
@
  a3
B x3 a2
BBN  x2





Project the knot onto a suitably chosen plane so that the image contains no triple
points and only finitely many, n, double points. Such a projection is called a regular
projection. . The n underpoints (one for each double point) divide the knot into n
arcs, an arc going from one underpass, over whatever overpasses there may be, to another
underpass. Label the arcs a1 , a2 , . . . , an , placing the labels each to the right of the knot
(using the orientation of (X, K)). For each arc ai pass a loop xi in X K under the arc

47
ai from right to left. These loops generate the knot group. Each underpass yields one
relation among the loops. For instance, the circled intersection of the knot in figure 4.6,
blown up in figure 4.7, yields the relation x1
3 x1 x3 = x2 . Together these n generators and
n relations give a presentation of the knot. For the figure-8 knot we have the presentation
G = (x1 , x2 , x3 , x4 : x1 1 1 1
3 x1 x3 = x2 , x4 x2 x4 = x3 , x1 x3 x1 = x4 , x2 x4 x2 = x1 ).

Since each relation states that one generator is a conjugate of another, we may give a
presentation of a quandle just by using quandle notation. For the figure-8 knot we then
have
Q = (x1 , x2 , x3 , x4 : x1 . x3 = x2 , x2 .-1 x4 = x3 , x3 . x1 = x4 , x4 .-1 x2 = x1 ).
By construction, Adconj Q = G. We may arrive at the same presentation more sim-
ply. Take a regular projection of the knot. Label the arcs putting the labels always
on the one side of the knot. For each intersection derive a relation on the arcs, as

b b
a c gives a . b = c, while a c gives a .-1 b = c.

The n relations on the n generators give a presentation of the quandle.

In section 4.7 we will give a direct combinatorial demonstration that this quandle is
an invariant of the knot (that is, it does not depend on the choice of regular projection).
In 4.8 we show that this quandle is isomorphic to the knot quandle defined in 4.5. In 4.9
we represent the knot quandle in terms of the knot group and show that it is a complete
knot invariant. That is, if Q(K) = Q(K 0 ), then K is equivalent to K 0 .

4.7 The invariance of the knot quandle

In this section we directly demonstrate the invariance of the quandle of a tame knot given
by generators and relations as described in the previous section. There are three basic
deformations of regular projections of knots which do not change the knot type. The
deformation 1 removes or adds a kink, 2 slides one arc under another, and 3 slides
an arc under an intersection.
These three deformations account for the equivalences among tame knots in the fol-
lowing sense. If two tame knots are equivalent, then for any regular projections of the
two knots, there is a sequence of basic deformations transforming one projection into
the other. A detailed proof of this fact is proved by Alexander and Briggs [1]. See also
Reidemeister [11]. In order to show the invariance of the knot quandle it suffices to show
its invariance under these basic deformations.
For the deformation 1 , we have two cases depending on which side of the arc is
labelled. Figure 4.9 indicates that the knot quandle is invariant under 1 since in a
quandle, the identities x . x = x and x .-1 x = x are satisfied.

48
 Figure 4.8: Basic knot deformations

1 - -

2 -

3 -

Figure 4.9: Invariance under 1

a a a .-1 a
- -

a .-1 a a a

a a a.a
- -

a.a a a

Figure 4.10: Invariance under 2

a a
a b b a b b
- -
a b b . a .-1 a a b a b .-1 a .a
a

49
Deformation 2 requires that x . y .-1 y = x and x .-1 y . y = x as shown in figure 4.10.

Figure 4.11: Invariance under 3

b.-1c b.-1c
c a.-1c c
(a.-1c) .(b.-1c)
a -
a a.b.-1c
b b a.b

b.c b.c
c a.c c
(a.c) .(b.c)
a -
a a.b . c
b a.b
b

The deformation 3 gives four requirements depending again on the labeling. Figure
4.11 illustrates two of these requirements.
The requirements for the invariance of the knot quandle under the basic deformations
are all satisfied. Thus, the quandle is an invariant of the knot type.

4.8 A presentation of the knot quandle (continued)

Proposition. Let K be a tame knot and Q be the quandle presented in terms of a regular
projection as described in section 4.6. Then Q is isomorphic to the knot quandle Q
defined in section 4.5.
Proof. Assume that the projection p of the knot to the plane is projection from the
basepoint . Label the arcs in order a1 , a2 , . . . , an . For i = 1, . . . , n, let bi be a path down
from to the center of the arc ai . Let i be the loop which travels from down bi , along
ai and ai+1 , then up bi+1 to as illustrated in figure 4.12.
Let Ui be a small toroidal neighborhood of i . Then Ui1 Ui is a neighborhood of bi .
Let V be X K. Then X = V U1 Un . In order to construct Q, we need to
know only the quandles of (Ui , Ui K), (Ui Ui+1 , Ui Ui+1 K), (V, ), (Ui V, ), and
(Ui Ui+1 V, ).
Let i be a noose in Ui1 Ui linking once about ai . Then 1 (Ui1 Ui K) = (xi )
is the free group on one element xi = (i ). Thus

AQ(Ui1 Ui , Ui1 Ui K) = (Conj(xi ), (xi )).

50
 Figure 4.12: The loop i

bi  r Ibi+1


- -
ai ai+1

Restricting to the nooses linking once with K, we have

AQ(Ui1 Ui , Ui1 Ui K) = ({xi }, (xi )).

Let G = 1 (X K) and Gi = 1 (Ui K). Then

AQ(Ui , Ui K) = (Conj(xi ) (xi ) Gi , Gi ), and

AQ(Ui , Ui K) = ({xi } (xi ) Gi , Gi ).

In order to find AQ(X, K), we may first tensor the various AQ with G = 1 (X K).
According to the theorem of section 2.10 and the Seifert-Van Kampen theorem, upon
taking the colimit of the various AQ G, we will have AQ(X, K).
Both AQ(Ui1 Ui , Ui1 Ui K) and AQ(Ui , Ui K) become ({xi } (xi ) G, G) when
tensored with G. Thus, Q is generated by the x1 , . . . , xn modulo the relations induced
by tensoring with G. These relations are determined by the action of the generators of
G on Q and the relations among the generators of G. It is exactly these relations which
were used in the definition of Q in section 4.6. Thus, Q = Q. 
We have already noted and used the fact that Adconj Q is the knot group G. As Q is
isomorphic to the knot quandle Q, we have Adconj Q
= G. In particular, (Q) generates
G.
Corollary. Let K be a tame knot, G its knot group, and Q its knot quandle. Then
Q G Q is a transitive group action.
Proof. In order to show G acts transitively on Q it suffices to show that for generators
a, b of Q there is an x in G such that ax = b. But by passing under sufficiently many
arcs of the regular projection, a becomes b, as suggested in the drawing

51
 a1
a2 ak b
a

Hence, b = a(a1 )1 (ak )1 . 

The theorems in sectons 4.5 through 4.8 hold not only for knots, but also for links
when careful attention is paid to the orientation of the components of the links. An
exception is the preceding corollary which holds only for knots. The quandle of a link is
algebraically connected if and only if the link is a knot.

4.9 A representation of the knot quandle

Recall that a meridian about a knot K is a loop in the complement of K that links once
with K and bounds a disk intersecting K at one point. Equivalently, a meridian is a
boundary of a noose in the knot quandle. Let U be a regular neighborhood of K in X,
that is, U is the image of S 1 (disk) embedded in X with K = image(S 1 {0}). The
boundary U of U is a torus. Connect U to the basepoint by a path in X U .
Then the inclusion U X K induces a homomorphism from 1 (U ) = ZZ
to G = 1 (X K). This is a monomorphism unless K is a trivial knot. The image of
this map is called a peripheral subgroup of the knot group G. Each peripheral subgroup
P contains exactly one meridian. Another distinguished element of P is the longitude l.
l is a generator of the subgroup of P consisting of loops which are not linked with K.
Proposition. Let K be a tame knot with group G and quandle Q. Let Q and
G = {x G | x = }. Then G is a peripheral subgroup of G.
Proof. Let U be a regular neighborhood of K containing the disk of the noose . Connect
U to by Rope . Then the loops in U Rope form a peripheral subgroup P of G. We
show P = G . Without loss of generality we may assume lies on U and Rope = {}.
Let U = f (S 1 disk), U = f (S 1 S 1 ), = f (1, 1). We may also assume () = f (S 1 1).
P is generated by the meridian () and the longitude = f (1 S 1 ). is homotopic
to (slide the disk of around the solid torus U by one revolution). Hence, P G .
Let be a loop in X K such that . The homotopy H of to may be chosen
so that the disk portion of the homotopy lies inside U .

52


q
6

-
q q

Let be the loop from (1) to (1) given by (t) = H(1, t). Then (Rope )(Rope )1
which lies in P . Thus, G = P . 
Corollary 1. Let K be a knot with knot group G and knot quandle Q. Let P be a
peripheral subgroup of G containing the meridian m. Then (P \G; m), as described in
section 2.4, is isomorphic to the knot quandle.
Proof. Follows from the corollary of secton 4.8. 
Thus, the knot quandle contains the same information as the triple (G, P, m) consist-
ing of the knot group G, a peripheral subgroup P , and a meridian m in P .
Neuwirth [10] remarks that if two tame knot groups are isomorphic by a map which
sends a meridian to a meridian and the group system (the conjugate peripheral sub-
groups) of one onto the group system of the other, then the (unoriented) knots are
equivalent. Conway and Gordon [6] use a slightly stronger principle to construct a group
that classifies oriented knots. If two tame knot groups are isomorphic by a map which
sends a meridian and corresponding longitude of one onto those of the other, then the
oriented knots are equivalent. A proof of this principle may be found in Waldhausen [19].

Corollary 2. If the knot quandles of two tame knots are isomorphic, then the (unoriented)
knots are equivalent. 
Other algebraic characterizations of knots have been described by Simon [14] and
Whitten [20]. The constructions given by Conway and Gordon, Simon, and Whitten are
not functorial, unlike the knot quandle.

4.10 The Alexander invariant of a knot

Let K be a knot in X = S 3 and Y = X K. Let Y be the infinite cyclic cover of Y .
Then 1 (Y ) = G0 = [G, G] where G = 1 (Y ) is the knot group. Also G/G0 = Z and
H1 (Y )
= G0 /G00 . If x in G is such that the map G G/G0 = Z sends x to 1, then

conjugation by x gives an automorphism t of H1 (Y ), and t is independent of x. Thus,
H1 (Y ) is not only an Abelian group, but also a module over the ring = Z[t, t1 ]. This
-module is called the Alexander invariant A of the knot K. We will show in this section
that A carries the same information as the Abelian knot quandle.

53
 The usual presentation of the Alexander invariant is by means of a matrix. Take a
Seifert surface for the knot K and let n be the genus of the surface. Let a1 , . . . , an be
generating cycles for the homology of the surface. and let vij be the linking number of
ai with aj . The matrix (vij ) is called a Seifert matrix of the knot. P = (vij tvji ) is an
Alexander matrix of the knot. The entries of P lie in . Then the Alexander invariant
is the cokernel
P
n n A 0.
The determinant of P is called the Alexander polynomial (t) of the knot. (t) is defined
only up to a unit of . One important property of (t) is that |(1)| = 1. A more direct
definition of (t) is that it is a generator of the annihilator ideal of A.
We first show that A may be constructed from the Abelian knot quandle Ab Q. As
noted in section 4.8, Adconj Q = G. Hence Adconj Ab Q = G/N where N is the normal
subgroup of G generated by elements of the form

ab1 ca1 bc1 , with a, b, c (Q).

Lemma 1. N = G00 .
Proof. Part 1. N G00 . Let a, b, c (Q). Since Q is algebraically connected, there exist
x, y in G such that b = x1 ax and c = y 1 ay. Then

ab1 ca1 bc1 = (ax1 a1 x)(y 1 aya1 )(x1 axy 1 a1 y).

Both ax1 a1 x and y 1 aya1 lie in G0 , so, modulo G00 ,

ab1 ca1 bc1 (y 1 aya1 )(ax1 a1 x)(x1 axy 1 a1 y) = 1.

Hence, ab1 ca1 bc1 G00 . Thus, N G00 .

Part 2. G00 N . Let a, b G0 . a is of the form ae11 aenn with ai (Q) and
P
ei = 0.
1 1 -1
Note that if x, y (Q), then xy = yz where z = x . y (Q). Hence, a may be
written as
a = a1 a1 1
2 a3 a4 an1 an
1

with each ai in (Q). Similarly,

b = b1 b1 1 1
2 b3 b4 bm1 bm

with each bi in (Q). Therefore, modulo N ,

ab = a1 a1 1 1 1 1 1
2 a3 a4 an1 an b1 b2 b3 b4 bm1 bm
b1 b1 1 1 1 1 1
2 b3 b4 bm1 bm a1 a2 a3 a4 an1 an = ba.

Hence, [a, b] N . Thus, G00 = N . 

00
Therefore, Adconj Ab Q = G/G is constructable from AQ, and, whence, its commutator
G0 /G00 is also. The symmetry at a point a0 in Ab Q is an automorphism of Ab Q which
induces conjugation by (a0 ) on G0 /G00 , the required -structure on H1 (Y )
= G0 /G00 .

54
 Next, we show that the Alexander invariant A of a knot determines Ab Q.
Theorem. Let A be given the quandle structure x . y = t(xy)+y, x .-1 y = t1 (xy)+y.
Then with this structure A is isomorphic to Ab Q.
Proof. Choose a0 Ab Q. Let P be the peripheral subgroup for a0 . By corollary 1 in
section 4.9, : P \G
= Q, (P x) = a0 x, is a quandle isomorphism where . on P \G
is given by P x . P y = P xy 1 (a0 )y. Now Ab Q is Q/N where N is defined above and
shown in the lemma to be G00 . Hence

(P/P G00 )\(G/G00 ) = (P/P N )\(G/N )

= Ab Q.

Now, the map G Q sending x to a0 x is still surjective when restricted to G0 , so

(P G0 /P G00 )\(G0 /G00 )

= Ab Q.

Let l be a longitude in P . Then P G0 = (l).

Lemma 2. l G00 .
Proof of lemma 2. Let m = (ao ). As l P , m1 lm = l. In the notation of A, l A and
tl = l. Hence, (t 1)l = 0. A is presented as
p
n n A 0.

As an element of n , (t 1)l = P a, some a n . Evaluate at t = 1 to get 0 = P (1) a(1).

Now P (1) is invertible as a matrix in Mn (Z) since det P (1) = (1) = 1. Hence,
0 = a(1), that is, (t 1) divides a, and a = (t 1)b. So (t 1) = P (t 1)b. Therefore,
l = P b image P . So l = 0 in A, l G00 . 
Proof of theorem continued. So P G00 = (l) = P G0 . Therefore, G0 /G00 = Ab Q.
Examine the quandle structure on G0 /G00 . In P \G, . is

P x . P y = P xy 1 my = P mxy 1 my = P (x . m)(y . m)1 y.

So in G0 /G00 , . is
x . y = (x . m)(y . m)1 y
which in the notation of A becomes

x . y = tx ty + y = tx + (1 t)y.


The Alexander invariant is not enough to distinguish all knots from the trivial knot.
For instance, the Alexander invariant of any doubled knot is trivial.

55
4.11 The cyclic invariants of a knot
In this section let K be a knot in X = S 3 and Y the complement of the knot. Let n be
a positive integer. Let Yn be the n-fold cyclic cover of Y and n be the n-fold branched
cyclic cover. Then H1 (Yn )
= H1 (n ) Z. H1 (n ) is a finite Abelian group. The n-th
torsion numbers of K are the subscripts in the canonical decomposition

H1 (n )
= Zk1 Zkr , ki | ki+1 , i = 1, . . . , r 1.

The order of H1 (2 ) is called the determinant of K, det K. Let P (t) = (vij tvji ) be
an Alexander matrix for K. Then P (1) = (vij + vji ) presents H1 (2 ) as a module over
Z. The determinant of P (1) gives det K; det K = | det P (1)|. Hence, the Alexander
polynomial (t) evaluated at 1 gives det K; det K = |(1)|.
H1 (n ) has an automorphism induced from conjugation by an x in G = 1 (Y ) for
which G Z sends x to 1. That is, H1 (n ) is a module over n = Z[t]/(tn 1). Then
H1 (n ) has a quandle structure given by the formula x . y = tx + (1 t)y. With this
structure H1 (n ) is an n-quandle, in fact, it is the largest quotient of the Alexander
invariant which is an n quandle. For n = 2, H1 (n ) is the involutory Abelian knot
quandle.

4.12 The involutory knot quandle

Let K be a knot with quandle Q. The involutory knot quandle Q2 results from imposing
the identity (x . y) . y = x on Q.
Example 1. The trefoil knot 31 .

Figure 4.13: The trefoil knot 31

a
'$
aq
qb
c q
b &%
c

Q = (a, b, c : a . b = c, b . c = a, c . a = b). Q2 has the same presentation as Q as long as

the relations (x . y) . y = x are understood. Viewed with geodesics (see section 3.1), Q2
has three points shown to the right.
Example 2. The figure-8 knot 41 .
Q = (a, b, c, d : a . c = b, b .-1 d = c, c . a = d, d .-1 b = a).
Q2 = (a, b, c, d : a . c = b, b . d = c, c . a = d, d . b = a).
Very little work shows |Q2 | = 5.

56
 Figure 4.14: The figure-8 knot 41

a b q
'$
bq qd
d q qa
&%
c

For these two examples Q2 is Abelian as well as involutory. The next example has a
nonAbelian Q2 .
Example 3. The knot 10124 .

Figure 4.15: The knot 10124

a b f
c j
g

h e

The Alexander polynomial of 10124 is (t) = t4 t3 + t1 1 + t t3 + t4 , so the

determinant of 10124 is 1. Thus, Ab Q2 is trivial. A few computations will show that Q2
may be faithfully represented as the 30 edges of a dodecahedron projected onto a sphere.
Figure 4.16 displays these points in stereographic projection in the plane.
Proposition. The link quandle is not an invariant of the complement of the link.
Proof. We examine the involutory link quandles of the links K1 and K2 displayed
in figure 4.17. As described on page 49 in [13], the complements of K1 and K2 are

57
 Figure 4.16: Q2 (10124 )

r rb
h
f r
r r r
r
r r j
r r r

r r rc r

dr ra r
r r e
r
r r r
i g
r

r r

58
 Figure 4.17: Two links with homeomorphic complements but different quandles

a a

e b g b
e
d c c
K1 K2

d f

homeomorphic.

Q2 (K1 ) = (a, b, c, d, e : a . c = b, b . d = a, c . a = d, d . c = e, e . a = c)
= Core Z4 .
Q2 (K2 ) = (a, . . . , g : a . c = b, b . e = a, c . f = d, d . a = e, e . c = f, f . d = g, g . a = c).

Q2 (K2 ) has order 8. It may be represented with geodesics as

cr

rf
r
e
ar rb

dr
r

r
g

Since the involutory quandles of K1 and K2 are distinct, so are their quandles. 

59
Bibliography

[1] Alexander, J. W., and G. B. Briggs, On types of knotted curves. Ann. of Math. 28
(1926), 562566.

[2] Blaschke, W., and G. Bol, Geometrie der Gewebe. Springer, Berlin, 1938.

[3] Bruck, R., H., A Survey of Binary Systems. Springer, Berlin, 1958.

[4] Burn, R. P., Finite Bol loops. Math. Proc. Cambridge Phil. Soc. 84 (1978), 377385.

[5] Burstin, C., and W. Mayer, Distributive Gruppen, J. Reine Angew. Math. 160
(1929), 111130.

[6] Conway, J. H., and C. McA. Gordon, A group to classify knots, Bull. London Math.
Soc. 7 (1975), 8488.

[7] Loos, O., Spiegelungsra

ume und homogene symmetrische Raume, Math. Z. 99
(1967), 141170.

[8] Loos, O., Symmetric Spaces. Benjamin, New York, 1969.

[9] Moufang, R., Zur Struktur von Alternativekorpern, Math. Ann. 110 (1935), 416430.

[10] Neuworth, L. P., Knot Groups. Princeton Univ. Press, 1965.

[11] Reidemeister, K., Knotentheorie. Chelsea, New York, 1949 (reprint).

[12] Robinson, D. A., Bol loops, Trans. Amer. Math. Soc. 123 (1966), 341354.

[13] Rolfsen, D., Knots and Links. Publish or Perish, Berkeley, 1976.

[14] Simon, J., An algebraic classification of knots in S 3 , Ann. of Math. 93 (1973), 113.

[15] Soublin, J.-P., Etude algebrique de la notion de moyenne, J. Math. Pures Appl. (9)
50 (1971), 53264.

[16] Takasaki, M, Abstractions of symmetric functions, Tohoku Math. J. 49 (1943), 143

207, [Japanese]. Math. Reviews 9, p. 8.

60
[17] Thomsen, G., Un teorema topologico sulle schiere de curve e una caratterizzazione
geometrica delle superficie isotermo-asintotiche, Boll. Union Mat. Ital. Bologna 6
(1927), 8085.

[18] Thomsen, G., Grundlagen der Elentargeometrie in gruppen algebraischer Behand-

lung, Hamburger Math. Einzelschr., no. 15, Teubner, Leipzig, 1933.

[19] Waldhausen, F., On irreducible 3-manifolds which are sufficiently large, Ann. of
Math. 87 (1968), 5688.

[20] Whitten, W., Characterization of knots and links, Bull. Amer. Math. Soc. 80 (1974),
12651270.

61
Index

10124 knot, 55 Disk, 39

n-core, ii, 11 distributive quasigroup, 3
n-quandle, ii, 3 distributivity, 27
p-quandle, 14
equivalent knots or links, 44
2-quandle, 3
fibre, 10
Abelian quandle, 2 figure-8 knot, 45, 54
Alexander and Briggs, 47 free quandle, 6
Alexander invariant, 51 fundamental augmented quandle, 39
Alexander matrix, 52 fundamental quandle, iii, 39
Alexander polynomial, 52
algebraically connected quandle, 9 geodesic, 23
associate, 31
homogeneous quandle, 9
associative, not, 2
homotopic maps, 36
augmentation map, 17
augmented quandle, 17 inner automorphism group, 7
fundamental, 39 inner automorphism of a quandle, 7
morphism of, 17 integral line quandle, 23
automorphism group, 7 inverse property, 31
involutory quandle, ii, 3, 23
basepoint, 36
with geodesics, 23
behaviorally equivalent, 7
isotopy of quasigroups, 31
boundary, 39
Bruck, 5, 26, 31 kei, 5
Burn, 31 knot, 44
Burstin and Mayer, 2 10124 , 55
commutative, 32 determinant, 54
conjugacy classes, 7 figure-8, 45, 54
conjugation, ii, 2 trefoil, 4, 54
Conway and Gordon, 51 knot group, 44
core, ii, 3, 26, 31 knot quandle, iii, 45
cosets, 8 link, 44
cyclic quandle, 24 link group, 44
determinant of a knot, 54 longitude, 50

62
loop, 31, 37 symmetry, ii, 2
Loos, ii, 5
Takasaki, 5
meridian, 45, 50 tame knot, 44
midpoint, 33 torsion numbers, 54
Moufang, 5 transvection, 10
Moufang loop, 31 transvection group, iii, 7, 10
trefoil knot, 4, 54
Neuwirth, 51
noose, iii, 39 Waldhausen, 51
Whitten, 51
order of a quandle, 2
Wirtinger, 45
pairs of topological spaces, 36
peripheral subgroup, 50
point, 2
pointed spaces, 36
projection, 27
quandle, ii, 2
Abelian, 2
algebraically connected, 9
augmented, 17
cyclic, 24
distributive, 27
free, 6
fundamental, 39
homogeneous, 9
involutory, 3, 23
knot, iii, 45
simple, 11
quandle, involutory, ii
quasigroup, 2, 31
distributive, 3
regular neighborhood, 50
regular projection, 45
Reidemeister, 47
Robinson, 31
Seifert matrix, 52
Simon, 51
simple quandle, 11
Soublin, 27, 30
symmetric space, 3

63