Documente Academic
Documente Profesional
Documente Cultură
MOTIVATING EXAMPLES
IN
Benjamin Fine
1
2
INTRODUCTION
where
R = p1 ...ps e1 ...et [a1 , b1 ]...[ag , bg ].
The sequence (g; m1 , .., ms ; t) is called the signature of G. The real
number
s
X
(2.2) µ(G) = 2π(2g − 2 + t + (1 − 1/mi ))
i=1
SURFACE GROUPS
What I”d like to do today is go back from this and look at what motivates
most of discrete group theory - surface groups and examine surface groups
as prime motivating examples in combinatorial group theory.
Just as finite group theory dealt with the finite groups necessary to study
equations over fields and infinite continuous group theory dealt with those
groups necessary to study analysis combinatorial group theory grew out of
the need to study the infinite discrete groups necessary to understand the
combinatorial objects in low dimensional topology - specifically originally
surface groups .
TABLE OF CONTENTS
1. INTRODUCTION
5. TAME AUTOMORPHISMS
Historically the beginning point for questions of this type was the Frei-
heitssatz or Independence Theorem of Magnus. If G is an orientable surface
group of genus g then its rank is 2g. From Theorem 2.1 if H is a subgroup
of rank < 2g it cannot be of finite index. It follows that H must be a free
group. If G has the standard presentation (1.1) this implies that any proper
subset of the generators generates a free group. According to Magnus [CM]
this fact was known to Dehn who presented to Magnus the problem of
proving the general result for one-relator groups. This is the basis for the
Freiheitssatz.
Theorem 2.2. (Freiheitssatz) Let G = < x1 , . . . , xn ; R = 1 > where R
is a cyclically reduced word which involves all the generators. Then the
subgroup generated by x1 , . . . , xn−1 is free on these generators.
In proving the Freiheitssatz Magnus developed a general method, now
called the Magnus Method, to handle one-relator groups. This involved
using group amalgams coupled with induction on the length of the relator.
These techniques have become standard in combinatorial group theory.
8
From this point of view, for any group amalgam - free product with
amalgamation or HNN groups - an amalgam factor is a FHS factor. Thus
any factor in a free product with amalgamation and the base in an HNN
group embed as a FHS factors in the resulting groups. This becomes the
basic idea in Magnus’ method. The method is to embed the group into
an amalgam in such a way that the proposed FHS factor embeds into an
amalgam factor which in turn contains the proposed FHS factor as a FHS
factor. The result can then be obtained by applying the FHS for amalgams.
A set of very strong results due to Lev Shneerson that are very inacces-
sible and not well known in the west indicates that the Freiheitssatz is a
very general resuslt in the theory of free algebras of many different types.
Baumslag, Fine and Roseneberger [BFR] used this idea to give a short proof
of the original Freiheitssatz completely different from the standard group
tehoretic method.
9
The most extensive work on extending the Freiheitssatz has been to one-
relator products. Let {Ai }, i in some index set I, be a family of groups.
Then a one-relator product is the quotient, G = A/N (R), of the free
product A = ∗i Ai by the normal closure N (R) of a single non-trivial word R
in the free product. We assume that R is cyclically reduced and of syllable
length at least two. The groups Ai are called the factors while R is the
relator. In analogy with the one-relator group case we say R involves Ai
if R has a non-trivial syllable from Ai . If R = S m with S a non-trivial
cyclically reduced word in the free product and m ≥ 2, then R is a proper
power. We then also call S a relator.
As with one-relator groups, the starting off point for a study of one-
relator products is to determine a Freiheitssatz. The example above shows
that there is no such result in general and therefore some restrictions must
be imposed. There are two approaches. The first is to impose conditions
on the factors while the second is to impose conditions on the relator.
Clearly such a group is the free product of the free groups on a1 , ..., ap and
ap+1 , ..., an respectively amalgamated over the cyclic subgroups generated
by U and V .
Theorem 2.6. The isomorphism problem for any cyclically pinched one-
relator group is solvable; given a cyclically pinched one-relator group G
there is an algorithm to decide in finitely many steps whether an arbitrary
one-relator group is isomorphic or not to G.
More specifically let G be a non-free cyclically pinched one-relator group
such that at most one of U and V is a power of a primitive element in F1
respectively F2 . Suppose x1 , ..., xp+q is a generating system for G. Then
one of the following two cases occurs:
(1) There is a Nielsen transformation from {x1 , ..., xp+q } to a system
{a1 , ..., ap , y1 , .., yq } with y1 , ..., yq ∈ F2 and F2 =< V, y1 , .., yq >.
(2) There is a Nielsen transformation from {x1 , ..., xp+q } to a system
{y1 , ..., yp , b1 , .., bq } with y1 , ..., yp ∈ F1 and F1 =< U, y1 , .., yp >.
For x1 , ..., xp+q there is a presentation of G with one-relator. Further G
has only finitely many Nielsen equivalence classes of minimal generating
systems.
Groups of this type arise in many different contexts and share many of
the general properties of the cyclically pinched case. However many of the
proofs become tremendously more complicated in the conjugacy pinched
case than the cyclically pinched case. Further in most cases additional
conditions on the associated elements U and V are necessary. To illustrate
this we state a result ([FRR], see [FR]) which gives a partial solution to the
isomorphism problem for conjugacy pinched one-relator groups.
Theorem 2.7. Let G =< a1 , ..., an , t; tU t−1 = V > be a conjugacy pinched
one-relator group and suppose that neither U nor V is a proper power in
the free group on a1 , ..., an . Suppose further that there is no Nielsen trans-
formation from {a1 , ..., an } to a system {b1 , ..., bn } with U ∈ {b1 , ..., bn−1 }
and that there is no Nielsen transformation from {a1 , ..., an } to a system
{c1 , ..., cn } with V ∈ {c1 , ..., cn−1 }. Then:
(1) G has rank n+1 and for any minimal generating system for G there
is a one-relator presentation.
(2) The isomorphism problem for G is solvable, that is it can be decided
algorithmically in finitely many steps whether an arbitrary given
one-relator group is isomorphic to G.
(3) G is Hopfian
Using Nielsen and extended Nielsen reduction in free products with amal-
gamation G.Rosenberger [R 7] was then able to give a complete classifica-
tion of the subgroups of rank ≤ 4 of such cyclically pinched one-relator
groups.
We note that the 3-free part of the above theorem was reproven in a
different manner by G.Baumslag and P. Shalen [BS].
Then G is n-free.
and Reiwer [PR] and is used in the analysis of conjugacy pinched one-relator
groups. Important for applications of Peczynski and Reiwer’s results is the
case where the associated subgroups are malnormal in the base. Recall that
H ⊂ G is malnormal if xHx−1 ∩ H = {1} if x ∈ / H. For a cyclic subgroup
< U > of a free group F this requires that U is not a proper power in
F . Using this, Fine,Roehl and Rosenberger proved the following two-free
result.
Theorem 3.5. [FRR1] Let G =< a1 , ..., an , t; tU t−1 = V > be a conjugacy
pinched one-relator group. Suppose that neither U nor V are proper powers
in the free group on a1 , ..., an . If < x, y > is a two-generator subgroup of G
then one of the following holds:
(1) < x, y > is free of rank two
(2) < x, y > is abelian
(3) < x, y > has a presentation < a, b; aba−1 = b−1 >.
residually free groups of low rank [FGMRS]. Recall that a group G is resid-
ually free if for each non-trivial g ∈ G there is a free group Fg and an
epimorphism hg : G → Fg such that hg (g) 6= 1 and is fully residually free
provided to every finite set S ⊂ G \ {1} of non-trivial elements of G there
is a free group FS and an epimorphism hS : G → FS such that hS (g) 6= 1
for all g ∈ S.
Theorem 3.8. [FGMRS] Let G be a 2-free fully residually free group.
Then G is 3-free.
18
Tarski Problem. All free groups have the same elementary or first-order
theory.
Theorem 4.1. All free groups have the same elementary or first-order
theory.
Based on this we define
Definition 4.1. An elementary free group is a group which has the
same elementary theory as the class of free groups.
A question posed prior to the solution of the Tarksi was; given the cor-
rectness of the Tarksi (at the time) Conjecture
Theorem 4.2. The class of elementary free groups coincides with the class
of regular NTQ-groups.
A regular NTQ-group is the coordinate group of a regular NTQ-system
of equations over a free group.
19
Prominent among the NTQ-groups and hence among the elementary free
groups are the surface groups
5. TAME AUTOMORPHISMS
(2) This was improved upon and extended by Zieschang and Rosenberger
and others to more general one-relator groups and Fuchsian groups.
and
The following new result due to [FKMR] generalize the result for surface
groups.
Fully residually free groups have played a tremendous role in both the
study of algebraic geometry over groups and in the solution of the Tarksi
problem. The following remarkable theorem by Remeslennikov and inde-
pendently Gaglione-Spellman ties together fully residually free groups and
elementary free groups. First we must recall a few things.
It is easy to show that all free groups have the same universal theory,
that is they all satisfy the same universal first order sentences.
23
Theorem 6.2. Suppose that G is a residually free group. Then the follow-
ing are equivalent
(1) G is fully residually free
(2) G is commutative transitive
(3) G is universally free
In particular being elementary free clearly implies being universally free
and hence theorems about fully residually free groups apply to elementary
free groups.
Our results are based on extensive work on the structure of fully resid-
ually free groups also called limit groups by Kharlampovich, Myas-
nikov,Remeslennikov, Bumagin, Rips, Sela, Fine, Gaglione,Rosenberger,
Spellman, Serbin and others. In particular we rely heavily on a paper
of Bumagin, Kharlampovich and Myasnikov in which they prove that the
isomorphism problem is solvable for finitely generated fully residually free
groups. In this paper they give a characterization of automorphisms of fully
residually free groups.
(1) The existence of canonical cyclic JSJ decompositions for fully resid-
ually free groups
The same type of analysis using cyclic JSJ decompositions for fully resid-
ually free groups leads to some partial progress on what we call the surface
group conjecture:
It is known that
Theorem 7.1. All orientable surface groups of genus g ≥ 2 are fully resid-
ually free.
Then if n is odd for there exists normal subgroups of finite index which do
not have one-relator presentations. In particular if
then every subgroup o finite index is again a one-relator group if and only
if n is even and hence a surface group.
26
This essentially says that is a fully residually free group has Property IF
then each subgroup of finite index is a one-relator group.
Theorem 7.3. Let G be a finitely generated fully residually free group with
property IF. Then either G is hyperbolic or G is free abelian of rank 2.
In the preceding results we assumed that G was fully residually free and
used the JSJ decomposition. Hwoever property IF will imply the finite
27
index property if we assume not the fully residually free property but that
we start with a graph of groups decomposition. In particular which get
the following which gives further evidence towards the full surface group
conjecture;
This theorem uses the subgroup theorem for free products with amalga-
mation and for HNN groups in the form described by Karrass and Solitar.
Theorem 7.5. Suppose that R and S are elements of a free group which
have the same normal closure. Then R is conjuagte to S ±1 .
Recently Jim Howie ( anD independently ( Bogopolski) has proved the
same result for one-relator quotients of surface groups. Bogopolski calls
this the Magnus Property.
where the factors A and B are either one-relator groups or free groups and
H is free,
29
Dehn in 1912 [De 2] provided a solution to the word problem for a finitely
generated orientable surface group. In doing so he ultimately paved the
way for both the development of small cancellation theory and the theory
of hyperbolic groups.
The general idea of a Dehn algorithm is clearly that there is ”not much
cancellation possible in multiplying relators”. Although Dehn’s approach
was geometric the idea can be phrased purely algebraically. This is the
31
HYPERBOLIC GROUPS
In hyperbolic geometry there is a universal constant A such that triangles
are A-thin. By this we mean that if XY Z is any geodesic triangle then
any point on one side is at a distance less than A from some point on one
of the other two sides. Now suppose G is a finitely generated group with
fixed finite symmetric generating set X. We say that a generating system
X for a group is symmetric if whenever a ∈ X we also have a−1 ∈ X. Let
Γ be the Cayley graph of G relative to this symmetric generating set X
equipped with the word metric. A geodesic in the Cayley graph is a path
between two points with minimal length relative to the word metric. A
geodesic triangle is a triangle with geodesic sides. A geodesic triangle in
Γ is δ-thin if any point on one side is at a distance less than δ from some
point on one of the other two sides. Γ is δ-hyperbolic if every geodesic
triangle is δ-thin. Finally G is word-hyperbolic or just hyperbolic if G
is δ-hyperbolic with respect to some symmetric generating set X and some
fixed δ ≥ 0. Gromov further showed that being hyperbolic is independent
of the generating set although the δ may change, that is if G is hyperbolic
with respect to one finite symmetric generating set it is hyperbolic with
respect to all finite symmetric generating sets.
From the existence of the Dehn algorithm we get the following immediate
results again generalizing the results for surface groups.
Gromov [Gr] further has proved that hyperbolic groups have solvable
conjugacy problem while Sela [Sel 1] has shown that the isomorphism prob-
lem is solvable for the class of torsion-free hyperbolic groups.
34
Note the analogy in the above statement with Stallings theorem on free-
by-finite groups.
Nielsen [N] gave the first non-trivial example of a test word by show-
ing that in the free group on x, y the commutator [x, y] satisfies this
property. Other examples of test words have been given by Zieschang
[Z4,Z5],Rosenberger[R1,R2,R3] Kalia and Rosenberger[KR], Hill and Pride
[HP] and Durnev[D]. THese are related to the surface group relator word.
Gupta and Shpilrain[GuS] have studied the question as to whether the
commutator [x, y] is a test element in various quotients of the free group
on x, y.
RETRACTION THEOREM
Test elements in finitely generated free groups were completely char-
acterized by the retraction theoremdue to Ted Turner. Recall that a
subgroup H of a group G is a retract if there exists a homomorphism
f : G → H which is the identity on H. Clearly in a free group F any free
factor is a retract. However there do exist retracts in free groups which are
not free factors. T.Turner [T] characterized test words as those elements of
a free group which do not lie in any proper retract.
Then if p ≥ 2 is prime the element ap1 bp1 ...apg bpg is a test element.
Turner later proved [Tu 2] that the retraction theorem also holds in
certain torsion-free hyperbolic groups. Whether it holds in all is still an
open question.
39
n = A the abelian variety. Stallings [St] and Dold[Do] have given sufficient
conditions for an element of a free group to be n -generic. Using this it can
be shown that xn1 xn2 ...xnm is n -generic in the free group on x1 , ..., xm for all
n ≥ 2 and if m is even [x1 , x2 ], ...[xm−1 , xm ] is n -generic in the free group
on x1 , ..., xm for n = 0 and for all n ≥ 2. These facts are also consequences
of a result of Rosenberger [R2,R3].
Further, in general it is not true that being an APE implies being a test
word. Again let F = F (x, y) and let w = x2 yx−1 y −1 . Brunner,Burns and
Oates-Williams show that w is an APE. However Turner shows that w is
not a test word. Since generic elements are test words in a Hopfian group
this example shows further that APE does not imply generic in general.
This is really to be expected since test words are strongly non-primitive.
However the following result shows that many APE’s are indeed generic
and therefore test words.
Suppose that