The word problem and Cayley graphs

Introduction
In this paper, we are going to explore the "word problem" and visualize it with
Cayley graphs. The word problem was rst proposed by Max dehn in 1912. It
states: Given a group G and a nite generating set S, is it possible decide which
words in

represent the identity? A variant of this problem is the

equality problem which asks if two words

, and

are same where

. One thing to note is that the word problem is concerned with

groups represented by nitely many generators.

Denitions
A nite generating set S: If G is a group and S is a subset of elements, then S
generates G if every element of G can be expressed as a product of elements
from S and inverses of elements of S. A group G is nitely
generated if it has a nite generating set.
Word:

Given a set S, a nite sequence of elements from S, possibly with

repetition, is called a word. The original set S is called alphabet.

Free monoid on a set S:

If S is a alphabet, then the collection of all words from

S including the empty word, is called free monoid on S, denoted by S*.

Cayley graph : Suppose that is agroup and is agenerating set. The Cayley
graph

is acolored directed graph constructed as follows:

Each element of

is assigned a vertex.

Each generator s

is assigned a color,

for every g

, the vertices corresponding to the elements

joined by direct edge of color

and

are

Eg:

Fig: Cayley graph of dihedral group Dih_4 on

where

= red and

blue.

Graph terms:
Distance: Let be a connected graph. The distance from a vertex v to a vertex w
is the minimum number of edges in a combinatorial path joining v to w. For
example, if v and w are distinct, but form Ends(e)
for some edge e, then the distance from v to w is 1. We denote this distance
function by d(v, w).
Sphere : For a xed vertex v V() and a xed n N, the sphere of radius
centered at is the set of vertices

S(v,n) = {w V() | d(v,w) = n}.

Ball : The ball of radius n centered at the vertex v is the subgraph formed as the
union of all paths in of length n that start at the vertex v. We denote this
subgraph by

. The vertices of

are the union of the vertices in S(v, i)

for i n.
Constructible: A Cayley graph is constructible if given n N one can
construct B(e,n) in a nite amount of time.

Fig: The ball of radius 2 in the Cayley graph of Z Z consists of all the
dark vertices and solid arrows.

Well known results and historical context

source

In 1910 Max Dehn publishedber die Topologie des dreidimensionalen Raumes*(On

the topology of 3-dimensional space) . In this paper he considered the problem of
when two knots are the same. Here he dealt with the fundamental group as a
key invariant and this group occurs naturally as given by a presentation. He
quickly realised that the knot theory problems were special cases of much more
general questions about nitely presented groups. He made the group theory
problems explicit in his 1911 paperber unendliche diskontinuierliche Gruppen. We
quote from the paper giving Dehn's formulation of the word problem for
groups:The general discontinuous group is given by n generators and m relations
between them, as dened by Dyck(Math. Ann.,20and22). The results of those
works, however, relate essentially to nite groups. The general theory of groups
dened in this way at present appears very undeveloped in the innite case.
Here there are above all three fundamental problems whose solution is very
difcult and which will not be possible without a penetrating study of the
subject.

**1. The Identity Problem [now called the Word Problem]: An element of the
group is given as a product of generators. One is required to give a method
whereby it may be decided in a nite number of steps whether this element is
the identity or not.
2. The Transformation Problem[now called the Conjugacy Problem]: Any two
elements S and T of the group are given. A method is sought for deciding the
question of whether S and T can be transformed into each other, i.e. whether
there is an element U of the group satisfying the relationS*=UTU-1.
3. The Isomorphism Problem: Given two groups, one is to decide whether they are
isomorphic or not *(and further, whether a given correspondence between the
generators of one group and elements of the other is an isomorphism or not).
These problems have very different degrees of difculty. ... One is already led to
them by necessity with work in topology. Each knotted space curve, in order to
be completely understood, demands the solution of the three above problems in
a special case.
Notice that there is a simple connection between the Conjugacy Problem and the
Word Problem. If one can solve the Conjugacy Problem in a particular group
then one can solve the Word Problem. For if a word w is conjugate to the
identity, then it is equal to the identify of the group. Dehn also proved in this
1911 paper that a nitely presented group can have a subgroup which is not
nitely presented. He also solved the isomorphism problem and the conjugacy
problem for nitely presented groups with the property that each generator
occurs at most twice in each of the dening relations.

In 1912 Dehn studied the word problem and the conjugacy problem for the
fundamental groups of orientable closed 2-dimensional manifolds. These are 1relator groups, that is groups with a presentation with a single dening relation.
He showed in Transformationen der Kurven auf zweiseitigen Flchen (1912) that in
special cases one could solve the word problem using a direct approach - called
Dehn's algorithm today. In this case one could construct a nite list of words in
the group generators, u1 , v1 , u2 , v2 , ... u**n , v**n having the following
properties. One has u**i = v**i as elements of the group, v**i is shorter
than u**i for each i, and one has the property that if w is any word in the
generators which represents the identity element, then at least one of theu**iis a
subword ofw. When a list with these properties exists one can easily solve the
word problem. For, given a wordwone looks to see if it contains a ui. If it does
not, thenwis not the identity. If it does contain au**ithen replaceu**ibyvit o
obtain a new word w'. Then continue the process, looking to see if w' contains
au**i. The process must terminate in a nite number of steps, since each time
au**iis replaced by av**ia shorter word is obtained.Dehn knew that the word
problem was difcult and posed an entirely new type of question in
mathematics.
He wrote:Solving the word problem for all groups may be as impossible as solving all mathematical
problems.
It was shown by Pyotr Novikov in 1955 that there exists a nitely presented
group G such that the word problem for G is undecidable.[4] It follows
immediately that the uniform word problem is also undecidable. A different
proof was obtained byWilliam Boonein 1958.[5]

Word problem and Cayley graph

(section 1.9)
We start by noticing that given any word {S S1}, there is an associated
edge path
in the Cayley graph

. The path starts at the vertex corresponding to the

identity and then traverses edges of

as dictated by . For example, consider

Z Z generated by x = (1, 0) and y = (0, 1). Then

the word =

describes the edge path illustrated in the diagram

below:

From this example, we can make the following observation.

Given a group G and a nite generating set S there is a one-toonecorrespondence between
1. nite edge paths in the Cayley graph
corresponding to the identity and
2. words in the free monoid {S S1}.

which begin at thevertex

From this observation, we can see that a word

corresponding path

represents an identity if the

is closed.

Going further, we will make the following claim and prove it.

Let G be a group with nite generating set S. Then the word problem is solvable if and
only if the Cayley graph

is constructible.

Proof:
() Let {SS1} and let n = ||. Construct the subgraphB(e, n) G,S ,
and trace out the path p inside this ball. The terminalvertex of this path is ve if
and only if () = e G. Thus one can solvethe word problem.
() The proof that B(e, n) is constructible is by induction on n. Thebase case is
immediate: B(e, 0) consists of a single vertex correspondingto the identity.
Assume that B(e, n) has been constructed, so that our goal is to con-struct B(e, n
+ 1). Every vertex in G,S is incident with exactly k edgescorresponding to the
generators S S1. If v B(e, n 1), then v isalready incident with k edges
labelled by S S1. If d(ve, v) = n, letLabelsn(v) be the subset of S S1
corresponding to edges that are attached to v in B(e, n). In forming B(e, n + 1) we
need to add an (ori-ented) edge to v, labelled s, for each s

S S1

Labelsn(v).The trick is to discover what the other vertex is that bounds this
edge.Notice that none of the missing edges can join vi to a vertex in B(e, n1).If
one did, then the edge in question would have been present in B(e, n),and so its
label would be in Labelsn(vi).
For convenience, list the elements in the sphere of radius n in somexed order:
S(e, n) = {v1, v2, . . . , vl}. To each vi associate a word iwhose corresponding
path in B(e,n) joins the vertex associated to e tovi. Starting with vertex v1, list the
generators in
S S1 \ Labelsn(v1) = {s1, s2, . . . , sm}.

The edge of G,S that is attached to v1 and is labelled si cannot joinv1 to a vertex
in B(e, n 1), so it must either join v1 to some othervj S(e,n), or join v1 to a
vertex in S(e,n + 1). This edge joins v1 tovj S(e,n) if and only if the word
1si(j)1 describes a circuit in G,S,that is, if and only if 1si(j)1 = e G.
Since we can solve theword problem, we can check this condition. If 1si(j)
1

= e Gfor each j, then the edge labelled si joins v1 to a new vertex, not

seenin B(e, n).

Let i > 1 and let sj be a missing label for the vertex vi, that is,sj S S1 \
Labelsn(vi). The edge labelled sj , which is attachedto vi in G,S, joins vi to one
of three types of vertices:
1. anothervk S(e,n);
2. a vertex in S(e, n+1) that is joined to vm S(e, n), where m < i;
or
3. a vertex in S (e, n + 1) that is not attached to any of the previous
vertices {v1, . . . , vi1}.

(The second case does not occur when i = 1.)

Case 1 occurs if andonly if is j(m)1 = e G, as before. Similarly, case 2
occurs ifand only if isj(msn)1 = e G for some sn S S1 , andthis can
be checked since we can solve the word problem. If none of theappropriate
words represent the identity, then we know we are in Case3, and a new vertex
needs to be added in forming B(e, n + 1). As anyvertex in B(e, n + 1) is joined to a
vertex in B(e, n), this process willterminate in a nite number of steps.

Open problem
Bibliography