One-Variable Equations In Torsion-Free Hyperbolic Groups

Alexei Kvaschuk

A dissertation submitted to the Graduate Faculty in Mathematics in partial fulllment of the requirements for the degree of Doctor of Philosophy, The City University of New York

2003

This manuscript has been read and accepted for the Graduate Faculty in Mathematics in satisfaction of the dissertation requirement for the degree of Doctor of Philosophy.

Date

Chair of Examining Committee

Date

Executive Ocer

Supervisory Committee

THE CITY UNIVERSITY OF NEW YORK

Abstract

One-Variable Equations In Torsion-Free Hyperbolic Groups

Alexei Kvaschuk Alexei Miasnikov

ii

Contents
1 Hyperbolic groups 1.1 Alternative denitions of hyperbolic groups (spaces) and examples of hyperbolic groups (spaces) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 1 1

Properties of hyperbolic group . . . . . . . . . . . . . . . . . . . . . . . . . 15 20

2 Free Constructions 2.1 2.2

Free products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 HNN extensions and Free Products with Amalgamation . . . . . . . . . . . 22 27 45 54

3 Equations in Groups 4 Elements of Algebraic Geometry Over Groups 5 Big Powers and Free Constructions 5.1 5.2 5.3 5.4 5.5 5.6

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 Elementary properties of BP-groups . . . . . . . . . . . . . . . . . . . . . . 58 Free products of BP-groups . . . . . . . . . . . . . . . . . . . . . . . . . . 64 Big Powers Condition Modulo Subgroups . . . . . . . . . . . . . . . . . . . 74 HNN Extensions and Big Powers Condition . . . . . . . . . . . . . . . . . 80 Free Products With Amalgamation and Big Powers . . . . . . . . . . . . . 88

iii

6 One-variable equations in torsion-free hyperbolic groups 6.1 6.2 6.3

90

Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 Main technical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

Description of solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 130

7 Resolution method 7.1 Application of resolutions for nding solutions of one-variable equations in

torsion-free hyperbolic groups . . . . . . . . . . . . . . . . . . . . . . . . . 136 8 Irreducible algebraic sets and their coordinate groups 8.1 8.2 8.3 8.4 8.5 140

Ultraproducts and coordinate groups of irreducible algebraic sets. . . . . . 140 Length functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142 Main technical results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144 Coordinate groups of irreducible algebraic sets . . . . . . . . . . . . . . . . 149 Description of irreducible algebraic sets . . . . . . . . . . . . . . . . . . . . 157

iv

List of Figures
1 2 3 4 5 6 7 8 Slim triangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Thin triangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Slim Thin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Slim Thin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 4 6 7

Canonical representatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 Lemma 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 Lemma 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 Theorem 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

Hyperbolic groups

The class of hyperbolic groups was rst introduced and studied by M. Gromov in his paper Hyperbolic Groups published in the volume Essays in Group Theory [G]. Since then hyperbolic groups have been a subject for intensive investigations. This class of groups is dened in geometric terms, making reference to the Cayley graph of a nitely generated group. The aim of the theory was to generalise results obtained for the fundamental groups of closed compact hyperbolic manifolds to some larger class, where techniques similar to those used in studying Kleinian groups may be useful. The principal reference in this area is the original text of Gromov [G], however several other authors have worked on producing more accessible versions [ABC], [Ger], [GH].

1.1

Alternative denitions of hyperbolic groups (spaces) and examples of hyperbolic groups (spaces)

We start with the notion of hyperbolicity for geodesic metric spaces. A metric space (X,d) is called geodesic if for all pairs of points x,y in X there is an isometric imbedding f : [0, d(x, y )] X taking the end points of the interval to x and y. The image of f is called a geodesic segment connecting these point, or more simply, a geodesic, and is denoted by [x,y].

The example of a geodesic metric space we are interested in is a Cayley graph of a nitely generated group G. Suppose that X is a nite set of generators for G, in the sense that every element of G can be written as a nite product of elements of X and their inverses. The Cayley graph X (G) has a vertex for each element of G, and an oriented edge labelled x from g to gx for each element g G and each x X . The group G acts on X (G) by multiplications on the left. A metric is dened by assigning unit length to each edge, and dening the distance between two points to be the minimum length of paths joining them. With this metric the graph X (G) becomes a geodesic metric space and left translations by elements of G become isometries. Now we give a collection of equivalent denitions of hyperbolic geodesic metric spaces. Denition 1 Inner (Gromov) product [G] . Let (X, d) be a metric space. Given a base point w X , we dene an inner product on X by

1 (x y )w = (d(x, w) + d(w, y ) d(x, y )). 2 If there is a constant 0 such that

x, y, z X, (x y )w min{(x z )w , (z y )w } 2

we say that the inner product is hyperbolic. It is not hard to prove [G] that this denition is independent of the base point, more precisely, if the inner product is -hyperbolic with respect to one base point, then it is 2 -hyperbolic with respect to any other base point. Denition 2 Slim triangles (Rips) [ABC] . Let (X, d) be a geodesic metric space. Given any three points x, y, z in X , we say that a triangle xyz of geodesics joining these points is -slim if for any point w on [x, y ] we have that min{d(w, [x, z ]), d(w, [y, z ])} . We say that triangles are slim in X if there is a constant such that all geodesic triangles in X are -slim. The geometric meaning of this denition is clear: a triangle is -slim if each side is contained in the union of -neighbourhoods of two other sides. Denition 3 Thin triangles [ABC] . Let (X, d) be a geodesic metric space. Given a triangle = xyz in X , let = x y z be a Euclidean comparison triangle with sides of the same lengths. Let f : be a natural identication map and t : T is a unique isometry of the onto a tripod T . Consider f = t f : T . 3

Figure 1: Slim triangle

y'
Cz' Cx'

x'
f

Cy'

z'
t

f(y) x z f

f(x)

f (z)

Figure 2: Thin triangle

We say that xyz is - thin if the bres of f have diameter at most in X . ( i.e. f (p) = f (q ) = dX (p, q ) ). We say that triangles are thin if there is a constant such that all geodesic triangles in X are -thin. Remark It is worth to note that in some literature slim triangles are called thin and vice versa. It turns out, and not hard to prove using elementary methods [ABC], that all three denitions above are equivalent. Theorem 1 [ABC] The following are equivalent for a geodesic metric space X . (i) Triangles are slim. (ii) Triangles are thin. (iii) The inner product on X is hyperbolic with any choice of base point. To demonstrate a technique of working with hyperbolic metric spaces we give a proof of the equivalence of the rts two denitions. Proof. It is clear that thin implies slim. Suppose now that all geodesic triangles in X are -slim. Let xyz be a geodesic triangle, and let cx , cy , cz be the internal points. 5

Consider the point cz on [x, y ]. There is a point t on [x, z ] [y, z ] such that d(cz , t) . Without loss of generality, suppose that t lies on [x, z ]. Then d(x, t) + d(cz , x) = d(cy , x) and d(x, t) d(x, cz ) + and so d(t, cy ) = d(x, cz ) d(x, t) , and d(cz , cy ) 2 . A similar argument shows that cx is at distance not more than 2 from one of cz and cy . It follows that diam{cx , cy , cz } 4 . Let u be a point on [x, cy ] and v a point on [x, cz ] such that d(u, x) = d(v, x). As geodesic triangles are -slim, d(u, [x, cz ] [cy , cz ]) . If there is a point t [x, cz ] such that d(u, t) , then d(t, v ) = d(u, x) d(x, t) d(u, t) , so that d(u, v ) 2 . Thus if d(u, v ) > 2 , it follows that there are points tu ,
z

Cy
t

Cx

Cz

Figure 3: Slim Thin 6

Cy

tu tv

Cz

Figure 4: Slim Thin tv [cy , cz ] such that d(u, tu ) and d(v, tv ) , and d(u, v ) d(u, tu ) + d(v, tv ) + d(tu , tv ) 6. Therefore triangles are thin. Examples. 1. It can be shown that in the hyperbolic upper half plane H triangles are ln(1 + slim. 2. Every simplicial tree (i.e a 1-dimensional contractible simplicial complex with the path metric, where each edge has length 1) is 0-hyperbolic. 3. Any bounded geodesic metric space is D-hyperbolic. 4. The Euclidean plane E is not -hyperbolic for any choice of . 2)-

5. R-trees (see [Ch] for the denition of R-trees) are 0-hyperbolic metric spaces. Denition 4 Word hyperbolic group. We say that a group G is word hyperbolic if it has a nite set of generators X such that the corresponding Cayley graph X (G) is a geodesic metric space with a -hyperbolic inner product, for some 0. To show that hyperbolicity is a group theoretic property we must show that the Cayley graph X (G) for any other nite set of generators X of G is -hyperbolic for some 0. This fact can be deduced from the properties of quasi-geodesics. Denition 5 Quasi-isometry. A (, ) quasi-isometric mapping f : (X, d) (X , d ) between two metric spaces (X, d), (X , d ) is a function f (in general discontinuous) satisfying the inequalities 1 d (f (x), f (y )) d(x, y ) d (f (x), f (y )) + for all x, y X , and some constants 1 and 0. A (, ) quasi-geodesic mapping of an interval [a, b] R into (X, d) is called a (, ) quasi-geodesic. Two metric spaces (X, d) and (X , d ) are quasi-isometric if there are numbers 1, 0, and C 0, and a (, ) quasi-geodesic mappings f : (X, d) (X , d ) and 8

f : (X , d ) (X, d) so that both compositions f f and f f are within C of the appropriate identity map. Thus d(f (f (x)), x) C for all x X and similarly for the other composition. Examples 1. (Z, d) and (R, d) are quasi-isometric, where d is the usual metric : d(x, y ) = |x y |. The natural embedding Z R is an isometry, so a (1, 0)- quasi-isometry. It is not surjective, but each point of R is at most
1 2

away from Z. We can dene a (1, 1 )2

quasi-isometry f : R Z by f (x) = x rounded to the nearest integer. 2. Every bounded metric space is quasi-isometric to a point. 3. If S and T are nite generating sets for a group G, then (G, dS ) and (G, dT ) are quasi-isometric. Indeed, let be the maximum length of any element of S expressed as a word in T or vice versa. Then the identity map G G is a (, 0)-quasiisometry from (G, dS ) to (G, dT ) and vice versa. Hence, when we are discussing quasi-isometry in the context of nitely generated groups, we can omit mention of the particular generating set, and make statements like G is quasi-isometric to H without ambiguity. We say that two subsets A and B of the metric space (X, d) are at nite Hausdor distance if there is a constant H > 0 such that B is contained in the H -neighborhood of 9

the set A and A is contained in the H -neighborhood of B . The basic result concerning quasi-geodesics in hyperbolic metric spaces is the following. Theorem 2 [ABC], [G] Let X be a -hyperbolic geodesic metric space. Then there is a function H (, ) > 0 so that for any two points x, y X , any (, ) quasi-geodesic f : [a, b] X with end points mapped to x, y , and any geodesic segment with the same end points x, y one has the image of f and the image of are at nite Hausdor distance at most H (, ) > 0 from each other. The next theorem states that a quasi-isometry preserves hyperbolicity. Theorem 3 [G] Let (X, d) and (X , d ) be quasi-isometric geodesic metric spaces. If (X, d) is hyperbolic, then there exists 0 so that (X , d ) is -hyperbolic. Using the theorem above, and the fact that any two Cayley graphs for the same nitely generated group are quasi-isometric geodesic metric spaces, we prove that hyperbolicity is indeed a group theoretic property. Another dierence of geometrical nature between hyperbolic and Euclidean geometries is the ratio of area to circumference of a polygon. It is well known that in the Euclidean

10

plane the area is a quadratics function of the circumference, whereas in the hyperbolic plane it is a linear function. Once we formulate an appropriate concept of area in a group this gives us another characterisation of a hyperbolic group. Denition 6 Linear isoperimetric inequality. Let < X | R > be a nite presentation of the group G. If w is a reduced word in F (X ) of length l(w) such that w =G 1 then there are words pi F (X ), relators ri R and i = 1 such that
i 1 w = N in F (X ). i=1 pi ri pi

If there is a constant K such that for all such words minimal number N is strictly less than K l(w), we say that G satises a linear isoperimetric inequality. It is known [G], [ABC] that the class of nitely presented groups which satisfy a linear isoperimetric inequality and the class of hyperbolic groups coincide. Denition 7 Dehns algorithm. A Dehn presentation for the group G is a nite presentation < X | R > such that any non-trivial word in F (X ) which represents the identity element of G contains more than half of some word in R. A group is said to have a Dehns algorithm if it has a Dehn presentation. 11

In other words, if a group has a Dehn presentation, then any word which represents the trivial element may be shortend by use of one relator (from a nite list of relations), which solves a word problem for this group. Indeed, given a word one reduces it until the word can not be shortened. If the obtained word is empty the original word represents the identity, otherwise the word represents a nontrivial element. A group is said to have a Dehns algorithm if it has a Dehn presentation. It is clear that a group with a Dehns algorithm satises a linear isoperimetric inequality (with multiplicative constant 1). It was established, for example, by Lysenok [Lys] that a group is hyperbolic if and only if it has a Dehns algorithm. Summarizing all of the above we have the following theorem. Theorem 4 Let G be a f.g. group, then the following are equivalent 1) G is word hyperbolic. 2) Triangles are slim in X (G). 3) Triangles are thin in X (G). 4) G satises a linear isoperimetric inequality. 5) G has a Dehn presentation. 12

Recall that an action of a group G on the topological space X is a mapping G X X , denoted by (g, x) gx, so that X is a G-set (this means that 1x = x, and g (gx) = (g g )x for all g, g G, x X ) and so that for each g G the map x gx is a homeomorphism of X onto itself. The action is called properly discontinuous if for each compact subset K of X the collection of groups elements {g G | gK K = } is nite. The action is called cocompact if the orbit space G \ X is compact. The metric space (X, d) is called proper if all balls of nite radius have compact closures. The following result of Troyanov has many important corollaries. Theorem 5 [T] Suppose that the group G acts properly discontinously and cocompactly by isometries on the proper geodesic metric space (X, d). Then G is nitely generated, and G with any word metric and (X, d) are quasi-isometric metric spaces. Therefore since quasi-isometry preserves hyperbolicity G is hyperbolic if and only if (X, d) is hyperbolic. The class of hyperbolic groups is fairly extensive, which is shown in the following examples. Examples. 1. It was proposed by Gromov [G] and proved by Olshanskii [Olsh], that in some 13

natural probabilistic sense almost every nitely generated group is hyperbolic. 2. Given a nite presentation P =< X | R >, let R denote the cyclic closure of R, i.e. the set of cyclic conjugates of elements of R and their inverses. A piece is a non-trivial word v F (X ) such that there are two dierent relators r1 , r2 R such that r1 = vr 1 and r2 = vr 2 . We say that P satises the C (p) condition if no element of R is a product of fewer
1 than p pieces. We say that P satises the C ( p ) if for each piece v occuring in the 1 relator r, p l(v ) < l(r). Thus if the C ( p ) condition holds, then so does the C (p + 1)

condition. ) (C(7)) small cancellation It is well known that a group which satises C ( 1 6 conditions satises a linear isoperimetric inequality and thus hyperbolic. It follows that fundamental groups of closed orientable surface of genus at least 2 or closed nonorientable surfaces of genus at least 4 are hyperbolic. Notice that the presentation P =< x, y, z | x2 y 2 z 2 > for the fundamental group of the nonorientable surface of genus 3 does not satisfy condition C ( 1 ). However using arguments of a 6 dierent kind it can be proved that this group is hyperbolic. 3. Finitely generated free group is 0-hyperbolic. 4. An arbitrary nite group is hyperbolic. 14

5. If G is the fundamental group of a closed Riemannian manifold all of whose sectional curvatures are strictly negative, then G is hyperbolic. The next two examples are consequences of Troyanovs theorem 5. 6. If H is a subgroup of nite index of the f.g. group G, then H is hyperbolic if and only if G is hyperbolic. 7. Hyperbolic n-space Hn is dened as {(x1 , . . . , xn ) Rn |xn > 0} with the Riemannian metric

dsn Hn =

2 dsn x2 1 + + xn En = , x2 x2 n n

where dsn En is the Euclidean metric. A (closed) hyperbolic n-manifold can be dened as the quotient of Hn by a cocompact properly discontinuous subgroup of isometries acting freely on Hn . Therefore by Troyanovs theorem if G is a subgroup of the group of isometries of Hn which acts properly discontinuously and cocompactly on Hn , then G is hyperbolic.

1.2

Properties of hyperbolic group

The following theorem shows that hyperbolic groups are nitely presented.

15

Theorem 6 [ABC] Let G be a hyperbolic, in the sense that geodesic triangles in the Cayley graph X (G) are -thin. Let

R = {w F (X ) | l(w) 8 and (w) = 1}, where : F (X ) F (X )/N = G is a natural epimorphism. Then < X | R > is a Dehn presentation for G. A simple consequence of the existence of a Dehn presentation is the following theorem which we state with a proof. Theorem 7 [GH] In a hyperbolic group, there are only nitely many conjugacy classes of elements of nite order. Proof. Let X be a nite set of generators for the group G and : F (X ) G is a natural surjection, where F (X ) is the free group on X . Consider a Dehn presentation of G. Let g be an element of nite order, and let [g ] be the set of all conjugates of g in G. Let w be a word in F (X ), chosen to be shortest over all v F (X ) such that (v ) [g ]. Let n be the order of (w). As (wn ) = 1, there must be some subword of wn which is more than half of a word r R. This implies that l(w) < l(r), otherwise w, or some 16

cyclic conjugate of w, can be shortened. (In fact w is contained in a cyclic conjugate of a relator.) Thus the number of conjugacy classes of elements of nite order is less than the number of elements of length at most max l(r).
r R

The following important technical result was proved by Gromov. Theorem 8 [G] Let g be an element of innite order in a hyperbolic group G. Let be a path from the vertex corresponding to the identity element to the vertex corresponding to g . Then the bi-innite path (..., g 1 , , g, ...) is a quasigeodesic. The next result shows that in a hyperbolic group a centralizer of an element of innite order is cyclic-by-nite. Theorem 9 [GH] Let G be a hyperbolic group and let g be an element of G of innite order. Let C (g ) denote the centralizer of g . Then C (g )/ < g > is nite. It easily follows from the theorem above that no hyperbolic group contains Z Z as a subgroup. The next two theorems describe some algorithmic properties of word-hyperbolic groups. Theorem 10 [DA] If a group has a Dehn presentation, then there is an algorithm for solving the word problem which is linear in terms of the length of the input word. 17

Theorem 11 [G] If elements g and h are conjugated in a -hyperbolic group G, then aga1 = h for some a G such that |a| < 2(|g | + |h| + 8 ). So there is an algorithm for solving conjugacy problem in a hyperbolic group. The next property of torsion-free hyperbolic groups is a very eective technical tool in solving equations over hyperbolic groups, and discrimination and completion of hyperbolic groups. Denition 8 A group G satises the big powers condition (BP) if for any nite sequence of elements u = (u1 , ..., uk ) such that [ui , ui+1 ] = 1 for every i = 1, ..., k 1 there exists number n(u) s.t. for any integers 1 , ..., k n(u) u1 1 . . . uk k = 1. Theorem 12 [Ol1] Every torsion-free hyperbolic group is a BP -group Denition 9 If X is a geodesic metric space then a subset A is -quasiconvex if for all geodesics [a, b] with endpoints a, b A, [a, b] N (A). A subgroup of a nitely generated group is said to be quasiconvex if the vertices in the subgroup form a quasiconvex subset of the Cayley graph. Lemma 1 [CDP] 18

A quasiconvex subgroup H of a hyperbolic group G is nitely generated, and in fact is also hyperbolic. A theorem of Rips [R] states that given any nitely presented group G there is a short exact sequence of groups

1 N E G 1 where E has a nite presentation satisfying the small cancellation condition C ( 1 ) and 6 where N is a normal subgroup which is nitely generated as a group. In Ripss construction the group N is never nitely presented except in trivial cases, so this gives examples of nitely generated subgroups of hyperbolic groups which are not nitely presented and therefore not hyperbolic. N. Brady [Brady] has given an example of a nitely presented subgroup H of a hyperbolic group G where H is not hyperbolic. The following well-known statement is due to Gromov. Theorem 13 [G] Let G be a torsion-free word-hyperbolic group. Then G contains only nitely many conjugacy classes of non-cyclic freely indecomposable two-generated subgroups. It is not hard to see, using Ripss construction, that the similar result fails for threegenerated subgroups. However, recently G. Arzhantseva [Arzh], and independently I. 19

Kapovich and R. Weidmann, proved the following general result. Theorem 14 For any 0 and an integer L > 0 there exists a number M = M (, L) > 0 with the following property. Let G be a -hyperbolic group and H be a subgroup of G generated by h1 , . . . , hL . Then at least one of the following assertions is true. (i) H is free on h1 , . . . , hL and quasiconvex in G; (ii) The tuple h1 , . . . hL is Nielsen equivalent to an L-tuple h 1 , . . . , h L with h 1 conjugate to an element of length at most M in G. The constant M = M (, L) can be calculated explicitly. Beyond these results there is almost nothing known about the subgroup structure of general hyperbolic groups.

Free Constructions

In this section we give the denitions and properties of the products of groups which are basic to combinatorial group theory. A motivation for studying these constructions came from topology as these constructions appear in the specail case of Seifert-van Kampen Theorem. We refer to two classical books on combinatorial group theory ([MKS], [LS]) for the detailed presentation of this subject. 20

2.1

Free products

Denition 10 Let A and B be groups with presentations A =< a1 , . . . | r1 , . . . > and B =< b1 , . . . | s1 , . . . > respectively. The free product, A B , of the groups A and B is the group A B =< a1 , . . . , b1 , . . . | r1 , . . . , s1 , . . . > . The groups A and B are called the factors of A B . Remark Free product is independent of the choice of the presentation chosen or A and B which are isomorphic to A and and B . Also, A B is generated by subgroups A B = 1. B respectively, and such that A Now we turn to the basic theorem about free products. Denition 11 A reduced sequence (or normal form) is a sequence g1 , . . . , gn , n 0, of elements of A B such that each gi = 1, each gi is in one of the factors, A or B , and successive gi , gi+1 are not in the same factor. ( We allow n = 0 for the empty sequnce). Theorem 15 [LS] (The Normal Form Theorem for Free Products) Consider the free product A B . Then the follwoing two equivalent statements hold. (I) If w = g1 . . . gn , n > 0, where g1 , . . . , gn is a rediced sequence, then w = 1 in A B . (II) Eache element w of A B can be uniquely expressed as a product w = g1 . . . gn where g1 , . . . , gn is a reduced sequence. 21

The Normal Form Theorem allows us to dene a length for elements of free products. If an element w of G = A B has normal form g1 . . . gn then the length of w, written |w|, is dened to be n.

2.2

HNN extensions and Free Products with Amalgamation

HNN extensions were introduced by G. Higman, B.H. Neumann and H.Neumann in 1949. We now turn to the denition. Denition 12 Let G be a group, and let A and B be subgroups of G with : A B an isomorphism. The HNN esxtension of G relative to A, B and is the group

G =< G, t | t1 at = (a), a A > . the group G is called the base of G , t is called stable letter, and A and B are called the associated subgroups. Now if w is an element of G , then we can write this element in a reduced form w = g0 t1 g1 . . . tn gn (n 0), where in the sequence g0 , t1 , g1 , . . . , tn , gn

22

there is no consecutive subsequence t, gi , t1 with gi A or t1 , gj , t with gj B . Now we state the so called Brittons lemma. Lemma 2 [LS] If the sequence g0 , t1 , g1 , . . . , tn , gn is reduced and n 1, then g0 t1 g1 . . . tn gn = 1 in G . We assign a length to each element z of G as follows. Let w be any reduced word of G which represents z . If w = g0 t1 g1 . . . tn gn , the length of z , written |z |, is the number n of occurences of t1 in w. It can be shown that |z | is well dened.

We now turn to free products with amalgamation (introduced by Schreier in 1926). Denition 13 Let G =< x1 , . . . | r1 , . . . > and H =< y1 , . . . | s1 , . . . > be groups. Let A G and B G be subgroups, such that there exists an isomorphism : A B . Then the free product of G and H , amalgamating the subgroups A and B by the isimorphism is the group

< x1 , . . . , y1 , . . . | r1 , . . . , s1 , . . . , a = (a), a A > . The basic idea of the free product with amalgamation is that the subgroup A is identied wit hits isomorphic image (A) H . The free product with amalgamation 23

depends on G, H , A, B and the isimorphism . The groups G and H are called factors of the free product with amalgamation, while A and B are called the amalgamated subgroups. We will simply write free products with amalgamation as P =< G H | a = (a), a A > . We can view P as the quotient of the free product G H by the normal subgroup generated by {a(a)1 | a A}. Denition 14 A sequnce c1 , . . . , cn , n 0, of elements of G H will be called reduced if (1) Each ci is in one of the factors G or H . (2) Successive ci , ci+1 come from dieren factors. (3) If n > 1, no ci is in A or B . (4) If n = 1, c1 = 1. It is clear that every element of P is equal to the product of the elements in a reduced sequence. On the other hand we have. Theorem 16 [LS] (Normal Form Theorem for Free Products with Amalgamation) If c1 , . . . , cn is a reduced sequence, n 1, then the product c1 . . . cn = 1 in P . In particular, G and H are embedded in P by the maps g g and h h. 24

We mentioned in the previous section the notion of a BP-group. The class of BPgroups is quite broad. In particular, any torsion-free hyperbolic group, or any subgroup of it, is a BP-group. The next results, proved in section 5, show that many groups, beyond the torsion-free hyperbolic ones, are BP-groups. Theorem (Kvaschuk, Miasnikov) A free product G H is a BP-group if and only if G and H are BP-groups. Now we introduce a relative form of the big powers condition, the BP-condition modulo a subgroup. Recall also that a subgroup A of a group G is called isolated if for every g G if g k A for some non-zero number k then g A. Let G be a group, A be a subgroup of G and u = (u1 , ..., uk ) be a generic sequence of elements of G. We say that: 1) u is A-generic if there exists i such that ui A; 2) u is A-independent if there exists an integer n such that for any integers 1 , . . . , k n
k u1 1 ...u k A.

Denition 15 Let G be a group. A subgroup A of G is called strongly isolated if every A-generic sequence of elements of G is A-independent. Observe, that a strongly isolated subgroup is isolated. 25

Theorem (Kvaschuk, Miasnikov) Let G be a torsion-free hyperbolic group and A be a quasi-convex isolated subgroup of G. Then A is strongly isolated. To state the similar result results about free product with amalgamation and HNNextensions we need to introduce some denitions. Recall rst that a subgroup M of a group G is malnormal if for any g G \ M g 1 M g M = 1. Following [KM1] we say that two subgroups A, B of a group G are conjugately separated if for any g G g 1 Ag B = 1. Theorem (Kvaschuk, Miasnikov) Let G and H are BP -groups and A, B are isomorphic subgroups of G and H . If A and B are strongly isolated and, at least, one of them is malnormal in G or H , then the free product with amalgamation GA=B H is a BP-group. Theorem (Kvaschuk, Miasnikov) Let G be a BP-group and A, B be conjugately separated isomorphic subgroups of G. If A and B are strongly isolated in G and either A or B is malnormal in G then the HNN-extension G, t | tAt1 = B is a BP-group.

26

Equations in Groups

Let G be an arbitrary group and F (X ) be a free group freely generated by X = {x1 , . . . , xn }. Then we can view G[X ] = G[x1 , . . . , xn ] = G F (X ) as a non-commutative analogue of a polynomial algebra over a unitary commutative ring in nitely many variables. We think of subsets S G F (X ) as systems of equations with coecients in G and say that a tuple (g1 , . . . , gn ) Gn is a solution of a system S if it satises each equation from S . We refer to the papers by Roger C. Lyndon [Lyn] and Grigorchuk and Kurchanov [GK2] for a survey of general results about equations in groups. The two natural questions about systems of equations in groups are: 1. Is it algorithmically decidable whether a given system has at least one solution ? (Diophantine problem) 2. How to describe the solution set ? Both these questions have been answered in armative in the case of free groups. In 1982 Makanin [Mak] proved the crucial result about the algorithmic decidability of Diophantine problem over free groups. He proved that if a given equation over a free 27

group F has a solution in F , then this equation has a solution of bounded length ( and this bound can be eectively computed from the equation itself). In his proof Makanin introduced, what is called now, Makanins process. This method was further developed by I. Rips to prove classication theorem of free actions of nitely generated groups on R-trees. It is worth to note that Makanins proof is extremely complicated and gives a very ineffective decision algorithm. Indeed, Ko scielski and Pacholsky [KP] noticed that it is not even primitively recursive. G. Sabbagh and independently L. Comerford asked the following question: does an equation in F has a solution if and only if it has a solution in any nite quotient of F ? A positive answer to this question would have yielded an alternative solution of Diophantine problem. T. Coulbois and A. Kh elif [CK] showed that the question of Sabbagh-Comerford has a negative answer. They considered the folowing equation [x2 a, y 1 z 2 by ] = t3 were a, b are distinct members of a set of free generators of F , and gave a short and elegant proof of the fact that this equation has a solution in any nite quotient of F , but has no solution in F . A. Razborov [Raz] described solution sets of arbitrary systems of equations over free 28

groups in terms of, so called, fundamental sequences. More precisely, let G be a free non-Abelian group, and S G[X ] be a system of equations over G, and Y = VG (S ) be its solution set. Then we dene the radical ideal of system S (or of algebraic set Y ) R(S ) = Rad(S ) = {f G[X ] | f (p) = 1p VG (S )}. The factor group GR(S ) = GRad(S ) = G[X ]/Rad(S ) is termed the coordinate group of the system S (of algebraic set Y ). Denition 16 A fundamental sequence of length k for a system of equations S is a triple (M, Hom, Aut), where M consists of k systems of equations S1 = 1, . . . , Sk = 1, S = S1 and Sk is an empty system. Hom is a collection of k 1 homomorphisms 1 , . . . , k1 where i : FR(Si ) FR(Si+1 ) , and i is a retract on F . Aut is a collection of k nitely generated automorphism groups P1 , . . . , Pk of the groups FR(S1 ) , . . . , FR(Sk ) respectively. A fundamental sequence S = (M, Hom, Aut) is eectively given if the systems in M, homomorphisms from Hom, and automorphisms from Aut are eectively given. To eectively dene a homomorphism from FR(S ) FR(T ) means to dene the images of the generators of the group FR(S ) . If is some fundamental sequence of length k for the system S = 1, : FR(Sk ) F a homomorphism of free groups, and 1 , 2 , . . . k are automorphisms from P1 , P2 . . . , Pk 29

respectively, then the composition

1 1 2 2 k FR(S ) FR(S ) FR(S2 ) FR(S2 ) . . . FR(Sk ) FR(Sk ) F (1)

of the system S . equals X for some solution X

We say that describes a solu-

of the system S if X tion X can be represented in the form (1) for some choice of 1 , . . . , k , 1 , 2 , . . . k . Razborov [Raz] proved that one can eectively construct a nite family of fundamental sequences such that any solution of the equation can be obtained from one of these sequences. Now we turn to a more precise description of solution sets which is based on the notion of quadratic equations. Denition 17 Let G be an arbitrary group and S (x1 , . . . , xn ) G[x1 , . . . , xn ]. We say that equation S (x1 , . . . , xn ) = 1 is quadratic (strongly quadratic) if every variable appears in S (x1 , . . . , xn ) not more than twice (exactly twice). The study of quadratic equations in non-free groups was initiated by P. Schupp [Schupp] . Let F be a free non-Abelian group. Let X1 , . . . , Xm be disjoint tuples of variables. A system U (X1 , . . . , Xm ) = 1 (with coecients from F ) of the following type S1 (X1 , X2 , . . . , Xm ) = 1 30

S2 (X2 , . . . , Xm ) = 1 ... Sm (Xm ) = 1 is said to be triangular quasi-quadratic if for every i the equation Si (Xi , . . . , Xm ) = 1 is quadratic in the variables from Xi . Denote by Gi the coordinate group of the subsystem Si = 1, . . . , Sm = 1 of the system U = 1: Gi = F [Xi , . . . Xm ]/R(Si (Xi , . . . , Xm ), . . . , Sm (Xm )) (i = 1, . . . , m + 1), in particular, Gm+1 = F and G1 = FR(U ) . The system U = 1 is said to be non-degenerate if for each i the equation Si (Xi , . . . , Xm ) = 1 has a solution in Gi+1 (with elements from Xi considered as variables and elements from Xi+1 , . . . , Xm as coecients from Gi+1 ). Observe, that if the system U = 1 is non-degenerate then the coordinate group Gi+1 is embeddable into Gi (i = 1, . . . , m) ([KM2]), i.e., we have a chain of groups F = Gm+1 Gm . . . G1 = FR(U ) . To solve the system U = 1 over F one needs to solve the last quadratic equation Sm (Xm ) = 1 over Gm+1 = F , then the previous one (which is again quadratic!) Sm1 (Xm1 , Xm ) = 1 31

over the coordinate group Gm , and continue the process going up along the triangular system until the rst equation S1 (X1 , . . . , Xm ) = 1 has been solved in the group G2 . F (in this case to specialize means to take G2 into F , that xes elements from F , and apply it to the obtained set of solutions The crucial result of O. Kharlampovich and A. Miasnikov [KM2], [KM3] describes the solution set in a free group F of an arbitrary system S (X ) = 1 with coecients from F : for any such S (X ) = 1 one can eectively nd a nite family of non-degenerate triangular quasi-quadratic systems U1 (Y1 ) = 1,. . . , Un (Yn ) = 1 (here Yi s are disjoint tuples of variables of, possibly, dierent length) and word mappings p1 (Y1 ), . . . , pn (Yn ) such that VF (S ) = p1 (VF (U1 )) . . . pn (VF (Un )), where VF (S ) is a solutions set of a system S in F . This description shows that solution sets dened by quadratic equations are building blocks for construction of arbitrary solution sets over F . We remind that a standard quadratic equation over a group G is an equation of the one of the following forms:

[xi , yi ] = 1, n > 0;
i=1

32

[xi , yi ]
i=1 i=1

zi 1 ci zi d1 = 1, n, m 0, m + n 1;

xi 2 = 1, n > 0;
i=1

xi 2
i=1 i=1

zi 1 ci zi d1 = 1, n, m 0, m + n 1;

where d, ci are non-trivial elements from G. A description of solutions of standard quadratics equations in free groups was obtained independently by Comerford and Edmonds [CE] and R. Grigorchuk and P. Kurchanov [GK]. We will give the precise formulation of this description later. This nished o the quadratic case, because it follows from the work of A. Hoar, A. Karras and D. Solitar [HKS1], [HKS2] that every strictly quadratic equation is automorphically equivalent to a standard one. More precisely: let S (X, Y ) = 1 be a strictly quadratic equation in variables X over F . Then there is an automorphism of the free group F F (X Y ) such that xes all the letters from Y and all the elements from F and such that (S ) = 1 is a standard quadratic equation over F . R.Lyndon [Lyn1] investigated one-variable systems of equations over free groups and proved that the set of solutions of a single equation can be dened by a nite system of

33

parametric words. Further progress in this direction was made by Lorents [Lor] and Appel [Ap] who described the exact form of the parametric words. Lorents showed that the set of solutions is given by a nite set of parametric words of the form ab cd in two parameters and . Appel showed that the set of solutions for a single equation in one unknown is given by a nite set of words of the form ab c in a single parameter , and Lorents established the same for nite systems of equations in one unknown. Unfortunately Appels published proof has a gap and Lorents announced his results without proof. I. Chiswell and V. Remeslennikov [CR] gave a full argument, and suggested a new approach for describing the set of all solutions of a one-variable system in a free group. Instead of working directly with solutions, they work only with the coordinate groups of irreducible algebraic sets. We will discuss this approach in details in the next section. It worth to mention that none of the results above gives an eective description of the solution set of a one-varible equation in a free group. Therefore the following problem arises : Problem. How to nd these one-parametric words eectively ? It was proved by Appel [Ap1] that an equation in two (or more) unknowns need not have its solutions given by any nite set of parametric words. Specically, over F (a, b)

34

free with basis {a, b}, the equation [x, y ] = [a, b] admits among its solutions all instances of the following parametric words:

(a, b), (b1 a, b), (b1 a, (b1 a)1 b), (((b1 a)1 b)2 b1 a, (b1 a)1 b), . . . . It is not dicult to see that not all instances of these are instances of any nite set of parametric words. However Ozhigov [Ozh] was able to handle equations in two variables. Denition 18 A group G is said to be equationally Noetherian if for any system of equations S G[x1 , . . . , xn ] there is a nite subset S0 of S which is equivalent to S (i.e. S and S0 have the same solution sets). The paper [BMRO] contains a detailed discussion about equationally Noetherian groups as well as examples of various kinds. It was proved by Guba [GV] that free groups are equationally Noetherian. Now we will discuss the technical trick which shows how to reduce an arbitrary system of equations over a free group to a single equation. It was conjectured by Vaught that x2 y 2 z 2 = 1 35

implies xy = yx, or, equivalently, that x, y , z all lie in a cyclic subgroup of F . The rst proof of Vaughts conjecture was given by R. C. Lyndon [Lyn2]. Let F be a non-Abelian free group and a, b be arbitrary non-commuting elements in F . Then it is not hard to prove, using Lyndons result, that the equation x2 ax2 a1 = (ybyb1 )2 has only a trivial solution in F , i.e. x = 1 and y = 1. Now a system which consists of two equations S1 (x1 , . . . , xn ) = 1 and S2 (x1 , . . . , xn ) = 1 is equivalent to the equation (S1 (x1 , . . . , xn )2 a)2 a2 = ((S2 (x1 , . . . , xn )b)2 b2 )2 , and a general case follows by induction. In the class of hyperbolic groups the situation is following. E. Rips and Z. Sela [RS] were able to develop a theory of, so termed, canonical representatives, which allows one to reduce a system of equations in a torsion-free hyperbolic group to a nite set of systems in a free group. Let G =< G | R > be a nite presentation of a torsion-free hyperbolic group, F (G ) be a the free group generated by the elements of G and : F (G ) G be a canonical epimorphism. We know that geodesic triangles in the Cayley graph of G are thin, however the combinatioral distinction between the three geodesic segments dening the 36

triangle is not completely clear. Canonical representative are a family of quasigeodesics constructed to have several stability properties which are usefull whenever one needs to control combinatorial ( or symbolic) properties of paths in the Cayley graph. Let S G[X ] where X = {x1 , . . . , xn } be a nite system of equations in G: sj (x1 , . . . , xn , g1 , . . . , gk ) = 1 1jq

Since G is torsion-free, and in particular has no 2-torsion, it is always possible to write S using triangular equations i.e. equations containig exactly 3 terms. Namely, if l = n + k and we let z1 = x1 , . . . , zn = xn and zn+1 = a1 , . . . , zl = ak we can write each of the equations in the triangular form as:

sj (z1 , . . . , zl ) = zi(j,1) zi(j,2) zi(j,3) = 1

1 j q.

()

From now on we assume that equations from S are triangular and S has a form (). Due to a great number of technical details we omit a precise denition of canonical representatives and give a description of canonical representatives of an arbitrary solution of system S . Let w = (w1 , . . . , wl ), wi G be an arbitrary solution of the system S (clearly wn+1 = a1 , . . . , wl = ak ). Proposition (Corollary 4.4. [RS])

37

Let w be an arbitrary solution of S . There exist a constant (S ),

j words ya , cj a F (G ); 1 j q ; 1 a 3,

and a map m0 : G F (G ); m0 {1, . . . , (S ) } (such that m0 = idG ) for which: (i)

j j j c 1 y2 m0 (wi(j,1) ) = y1 1

j j j m0 (wi(j,2) ) = y2 c 2 y3

j j j m0 (wi(j,3) ) = y3 c 3 y1

(ii) length(cj a ) C (S, G) where C (G, S ) is some constant which depends only on S and G (iii)
j j cj 1 c2 c3 ncl (R)

The words m0 (wi(j,k) ) are called canonical representatives of elements wi(j,k) with respect to the section m0 . We see that canonical representatives are chosen so that they agree outside some globally bounded neighborhood of the inscribed triangle (Fig. 5). 38

y2
m
W1 W2

c1
W3

c2 c3 y
3

y1

Figure 5: Canonical representatives

j Now let ya denote the unknowns for the systems Tr (1 j q ; 1 a 3) over

the free group F (G ). Although, canonical representatives are assigned to elements in the hyperbolic group G and not to the unknowns z1 , . . . , zl , we denote by m0 (z1 ), . . . , m0 (zl ) the canonical representatives assigned to a particulr solution of the original system S by proposition above (the particular index m0 depends on the solution we pick). For each j we get the following identities between the canonical representatives of the original
j unknowns z1 , . . . , zl in F (G ) and our new unknowns ya :

j j j m0 (zi(j,1) ) = y1 c 1 y2

j j j m0 (zi(j,2) ) = y2 c 2 y3

39

j j j m0 (zi(j,3) ) = y3 c 3 y1

j j j where cj 1 c2 c3 ncl (R) and length(ca ) is globally bounded. Since there is an eective

procedure to solve the word problem in G, one is able to produce all possible triples
j j for cj 1 , c2 , c3 . The equations belonging to each of our new systems Tr come from the

identities m0 (zi(j,a) ) = m0 (zi(j ,a ) ) if i(j, a) = i(j , a ). In addition, we have to substitute the canonical representatives for the coecients a1 , . . . , ak with respect to the criterion m0 , so that we will be able to add the equations:

m0 (zlk+1 ) = m0 (a1 ), . . . , m0 (zl ) = m0 (ak ). It is shown in [RS] that given m0 the canonical representatives for the coecients at (1 t k ) can be found eectively. Finally, each system Tr ; 1 r (S ) will have the equations indicated above, where for each r one picks a dierent choice for the coecients cj a (1 j q ; 1 a 3) and for the canonical representatives m0 (at ); (1 t k ). The way each system Tr is constructed, guarantess that every solution of such system projects into a solution of S under the natural homomorphism F (G ) G. Proposition above proves the converse, namely shows that for each solution w = (w1 , . . . , wl ) of the original sustem S there exists a solution of one of the Tr , such that m0 (zq ) projects on 40

wq (1 q l). In particular, the original system S is consistent if and only if one of the systems Tr over free group F (G ) is consistent. Since the latter is algorithmically decidable using Makanins algorithm there is an algorithm recognizing the solvability of any system of equations in a torsion-free hyperbolic group. Observe, that if S (X ) = 1 is a one-variable equation in a torsion-free hyperbolic group, then Rips-Sela reduction provides an equation in free group which is not, in general, onevariable. Recently Z. Sela (unpublished) proved that torsion-free hyperbolic groups are equationally Noetherian. In [Lys] an algorithm was given which determines the solvability of a quadratic equation in a hyperbolic group. And later R. Grigorchuk and I. Lysenok [GL] described the set of solutions of a quadratic equation in a hyperbolic group. In order to formulate this result we shall interpret quadratic equations and their solutions in the following way. Let C and X be two disjoint innite sets whose elements are called constants and variables, respectively. Elements of C X are called letters and elements of the group alphabet C 1 X 1 are called signed letters. For a word W , L(W ) denotes the set of all letters occuring in W . A quadratic word is a cyclically reduced word S in the alphabet C 1 X 1 such that every variable which occurs in S occurs exactly twice and every 41

constant which occurs in S occurs exactly once. Given a quadratic word S , we dene an associated group CS by CS =< L(S ) | S = 1 > . A quadratic equation in a group H , written as (S = 1, ), is a pair consisting of a quadratic word S and a coecient map : L(S ) C H . A solution of a quadratic equation (S = 1, ) in H is any homomorphism : CS H such that c = c for all c L(S ) C . Given a quadratic word S , let A(S ) be the group of all automorphisms Aut CS such that c = c for all c L(S ) C . Clearly if is a solution of a quadratic equation (S = 1, ) in a group H then for every A(S ), is also a solution of (S = 1, ). Thus the set of all solutions of quadratic equation (S = 1, ) in a group is divided into orbits under the action of A(S ). We introduce a class of parametric solutions of quadratic equations in a group H . Given a subset B H , a nite set T of parameters, and a map b : T B, we dene and HNN-extension HT, b of H by

HT, b = H, T | tb(t) = b(t)t (t T ) . Let (S = 1, ) be a quadratic equation in H . We call a homomorphism : CS HT, b a parametric solution of (S = 1, ) if c = c for all c L(S )C . A parametric evaluation 42

map is any homomorphism w : Ht, b H which is identity on H . If is a parametric solution of (S = 1, ) and w is a parametric evaluation map then w is obviously a solution of (S = 1, ). We call w a specialization of . Now we formulate the main result of [GL]. We call a set {i | i I } of parametric solutions of a quadratic equation (S = 1, ) in H fundamental if every solution of (S = 1, ) equals to i w for some A(S ), i I and parameters evaluation map w. Theorem 17 [GL]. Given a hyperbolic group H and a standard quadratic equation (S = 1, ) in H , there exists and can be constructed eectively a nite fundamental set of parametric solutions of the equation. If the group H and the quadratic word S are xed then the running time of the algorithm is bounded from above by a polynomial function on the total length of words representing the elements c , c L(S ) C . We nish this section with one of the main results of this paper. As we mentioned above, classical results about one-variable equations in free groups do not give an eective description of solution sets. The next two theorems, which we prove in section 6, solve this problem for the class of torsion-free hyperbolic groups.

43

Theorem (Kvaschuk, Miasnikov) Let G be a torsion-free hyperbolic group and S (x) = x1 u1 x2 u2 . . . xn un = 1 be an arbitrary one-variable equation in G. Then there is a constant C (which can be found eectively) and there are nitely many elements f and g (which can be found eectively) such that an arbitrary solution x of S (x) = 1 has one of the following forms: 1) |x| < C ; 2) x = f g for some g G such that |g | < C ; 3) x = (f g )g where |g | < C ; 4) x = (f g )f g for some g G such that |g | < C ; where and are integers. Notice that contrary to a free group case, two-parametric solutions appear in the case of an arbitrary tosion-free hyperbolic group G. In the section 6 we give particular examples of torsion-free hyperbolic groups which have two-parametric sets of the form f g or f gf as solutions of some one-variable equations. Notice also that the theorem above gives a necessary condition for x to be a solution of S (x) = 1. Then one needs to check that a given element x is indeed a solution of S (x) = 1, which is algorithmically solvable. To nd quickly parametric solutions of S (x) = 1 we 44

develop a, so termed, resolution method (see section 7). This method is based on the BP-property of torsion-free hyperbolic groups. Therefore, there is an algorithm which nds all solution of one-variable equations in torsion-free hyperbolic groups.

Elements of Algebraic Geometry Over Groups

In this section we present some basic notions of algebraic geometry over groups which were introduced and studied in the paper by G. Baumslag, A. Miasnikov and V. Remeslennikov [BMR1]. Denition 19 Let G be a xed group. A group H is called a G-group if G is a xed subgroup of H . The class of G-groups forms a category in the obvious way, in particular a group homomorphism : H1 H2 between two G-groups is termed a G-homomorphism if (g ) = g for all g G. Denition 20 Let X be a set and G a group. The group G F (X ), where F (X ) is the free group on X , is called the free G-group on X and is denoted by G[X ].

With the obvious embedding of G, G[X ] is a free object, in the usual sense, in the category of G-groups. A G-group H is called a free G-group if it is isomorphic as a G45

group to G[X ] for some set X , and X is then called a set of free G-generators for H . If X has a single element x, we denote G[X ] = G x by G[x]. Denition 21 The set Gn = {(g1 , . . . , gn ) | gi G} is called ane n-space over G; when n = 1, G1 = G. Suppose X = {x1 , . . . , xn } is a nite set. Using the normal form theorem for free products and for F (X ), an element f G[X ] may be viewed as a word in the elements of G and X 1 , which we view as a monomial in the variables X 1 with coecients in G.
1 1 Given p = (g1 , . . . , gn ) Gn , we can substitute gi for x in f to obtain an element i

f (p) G. (More formally, there is a homomorphism : G[X ] G which is the identity mapping on G and sends xi to gi , and f (p) means (f )). If f (p) = 1, we think of p as a solution of the equation f = 1, and sometimes call a subset of G[X ] a system of equations over G. The denitions which follow are made by analogy with the situation for ane algebraic sets over a eld. Denition 22 Let S G[X ], where X = {x1 , . . . , xn }. Then the set VG (S ) = {p Gn | f (p) = 1 for all f S } is termed the algebraic set over G dened by S .

Typical Examples. 46

(1) Every singleton {g } is an algebraic set: VG (xg 1 ) = {g } (here n = 1 and X = {x}). (2) For any subset M G, the centraliser CG (M ) is algebraic, being equal to VG ([x, m] | m M ). (3) More generally, for any g G, the coset CG (M )g = VG ([xg 1 , m] | m M ) is algebraic. Denition 23 Let S G[X ] be a system of equations over G, and put Y = VG (S ). Then we dene Rad(S ) = Rad(Y ) = {f G[X ] | f (p) = 1 p Y }. Call Rad(S ) the radical ideal dened by the system S (or by Y ). We can view Rad(S ) as a set of all consequences of a system S . Clearly Rad(S ) is always a normal subgroup of G[X ]. Denition 24 The factor group GY = GS = G[X ]/Rad(S ) is termed the coordinate group of the algebraic set Y .

Examples. Let G = F be a non-abelian free group and X = {x}. (1) If Y = VF (xg 1 ) = {g }, then GY = F [x]/ncl(xg 1 ) = F. 47

(2) If Y = CG (g ) = VF ([x, g ]), then GY = F [x]/ncl([x, g ]) = F, t | [g, t] = 1 . The proof is done by using the Normal Form Theorem for HNN extensions. Such an HNN extension is called a free extension of centralisers of rank 1 . Denition 25 Let G be a group and let Y Gn , Z Gp be algebraic sets. Then a map : Y Z is termed a morphism of the algebraic set Y to the algebraic set Z if there exist f1 , . . . , fp G[x1 , . . . xn ] such that for any point a1 , . . . an Y , (a1 , . . . an ) = (f1 (a1 , . . . an ), . . . fp (a1 , . . . an )) Z. Denition 26 Two algebraic sets Y , Z are said to be isomorphic if there exist morphisms : Y Z , : Z Y such that = 1Y and = 1Z . Let ASG be the category of all algebraic sets over G, with morphisms as dened above (so the use of the term isomorphic is consistent with that in category theory). Denote by AGG the category of all coordinate groups of the algebraic sets from ASG , where morphisms are G-homomorphisms. The following is Theorem 4 in [BMR1]. Theorem 18 The categories ASG and AGG are equivalent. From the theorem follows a very important fact: to classify algebraic sets up to isomorphism, it is enough to classify coordinate groups of the corresponding algebraic sets. 48

Typical Examples. (1) Let V be an algebraic set in Gn . Then for any two points p, q Gn the set pV q is also an algebraic set which is isomorphic to V . Let p = (p1 , . . . , pn ), q = (q1 , . . . , qn ), and let S (x1 , . . . , xn ) be the system of equations that denes V. Then S (y1 , . . . , yn ),
1 1 where yi = p denes pV q and the map : V pV q , (a1 , . . . , an ) = i xi q i

(p1 a1 q1 , . . . , pn an qn ) denes the isomorphism between V and pV q . In particular, (1a) Any two points p1 , p2 of Gn are isomorphic to each other (as singleton algebraic sets). (1b) If CG (M ) is the centralizer of a subset M and f , g G then CG (M ) is isomorphic to f CG (M )g . (2) If V G1 is an algebraic set and V is isomorphic to G1 , then V = G1 . For by Theorem 18, an isomorphism from V to G1 induces a G-isomorphism : G[x] G[y ] of two free G-groups of rank 1, where G[x] is the coordinate group of G and G[y ] is the coordinate group of V . Then (x) = py e q , where e {1, 1}, p, q G. Further, denes an isomorphism between G and V in the following way (see the proof of Theorem 18 given in [BMR1] ): (g ) = pgq . Therefore V = G. We recall a denition of an equationally Noetherian group.

49

Denition 27 A group G is said to be equationally Noetherian if, for every n > 0 and any subset S of G[x1 , . . . xn ], there exists a nite subset S0 of S such that VG (S ) = VG (S0 ). If G is a torsion-free hyperbolic group, it is possible to introduce on Gn the so-called Zariski topology, by taking the set of all algebraic sets as the set of closed subsets. One can therefore dene the notion of irreducible algebraic set. (A non-empty subset Y of a topological space X is called irreducible if it cannot be expressed as a union Y = Y1 Y2 of proper subsets, each of which is relatively closed in Y ). Theorem 19 [BMR1] Every algebraic set V over an equationally Noetherian group H is a nite union of irreducible algebraic sets, which are uniquely determined by V , up to order. (They are called the irreducible components of V ). Theorem 20 [BMR1] Let G be an equationally Noetherian group and let V = V1 . . . Vk be the decomposition of an algebraic set V into its irreducible components. Then the coordinate group of V is canonically embedded in the direct product of the coordinate groups of the Vi , GV GV1 . . . GVk . In the previous section we mentioned a result of Chiswell and Remeslennikov about description of solutions sets of one-variable equations over free groups. Instead of working

50

directly with solutions, they work only with the coordinate groups of irreducible algebraic sets. The following fundamental theorem on which this approach is based was proved in [BMR1]. Theorem 21 [BMR1] Let G be an equationally Noetherian group. Then for any countable innite set I and any non-principal ultralter D over I , the ultrapower GI /D has the following properties: 1. GI /D is a G group, where G is embedded in GI /D via the diagonal mapping; 2. Every coordinate group of an irreducible algebraic set over G is embeddable in GI /D; 3. Every nitely generated G-subgroup of GI /D is isomorphic to the coordinate group of some algebraic set Y over G. Let F be a free group and S any system of equations with one variable over F (i.e. S F x ), such that VF (S ) = . Theorem 22 [CR] Any coordinate group FY of an irreducible algebraic set Y F 1 satises one of the following. a) FY = F; 51

b) FY = F, t | [u, t] = 1 , where u is a root element of F ; c) FY =F x . Theorem 23 [CR] Let F be a non-abelian free group and let V = V1 . . . Vk be the decomposition of the algebraic set V from F 1 into its irreducible components. If V = F 1 , then every Vi has one of the following forms: a) Vi is a point; b) there exist elements f , g , h F such that Vi = f CF (g )h. The next two theorems, whose proofs are given in section 8, give a description of irreducible algebraic sets and their coordinate groups over torsion-free hyperbolic groups. Let G be a torsion-free hyperbolic group and S G x any system of equations with one variable over G such that VG (S ) = . Theorem (Kvaschuk, Miasnikov) Any coordinate group GY of an irreducible algebraic set Y G1 satises one of the following. a) GY = G; b) GY = G, t | [g, t] = 1 , where g is a root element of G; 52

c) GY =G x ; d) GY = G g = g f , g , where f , g = f, g for some root elements f, g G; e) GY = H, s | s1 f s = f where H = G g = g , where f , g = f, g for some root elements f, g G; Let M =< f, g | r(f, g ), r R > be a subgroup of G generated by root elements f, g G. We introduce the following systems
= {r(x1 f x, g ) = 1; r R}. SM = {r(f, x) = 1; r R} and SM

Theorem (Kvaschuk, Miasnikov) Let G be a torsion-free hyperbolic group and V = V1 Vk be the decomposition of an algebraic set V from G1 into irreducible components. If V = G1 , then every Vi , up to isomorphism, has one of the following forms: a) Vi is a point; b) Vi = f C (g ), for some elements f, g, G; c) Vi = VG (SM ) = x G | |x| < C1 , or x = (f x1 )1 , or x = f x1 f , or x =

f x1 f for some |x1 | < C1 , for some M =< f, g | r(f, g ), r R > and C1 = C1 (, f, g ) 53

d) Vi = VG (S M ) = x G | |x| < C2 , or x = f x1 or x = x1 g , or x = f x1 g for

1 1 some |x1 | < C2 such that r (x 1 f x1 , g ) = 1 r (x f x, g ) = 1 S M , for some

M =< f, g | r(f, g ), r R > and C2 = C2 (, f, g ) Corollary Let G be a torsion-free 2-free hyperbolic group, then every irreducible algebraic set V = G over G is isomorphic to a point or centralizer of some element. In particular if G is a non-abelian f.g. free group, then we get Chiswell-Remeslennikov theorem.

5
5.1

Big Powers and Free Constructions

Introduction

In this section we study a class of groups satisfying the so-called big powers condition. We show that HNN-extensions and free products with amalgamation preserve the big powers condition under some natural requirements on associated (amalgamated) subgroups. Let G be a group and u = (u1 , ..., uk ) be a sequence of non-trivial elements of G. We say that: 1) u is in a general position if neighbors in u do not commute: [ui , ui+1 ] = 1 f or every i = 1, ...k 1;

54

2) u is independent if there exists an integer n such that for any integers 1 , . . . , k n

k u1 1 ...u k = 1.

Sometimes we refer to a sequence in a general position as to a generic sequence, or a commutation-free sequence. The positive integer n from 2) (if it exists) is called a separation boundary or a no-return boundary for u. Denition 28 A group G satises the big powers condition (BP) if any sequence of elements of G in a general position is independent. Such groups are called BP-groups. It is clear that (BP) is a local property, i.e., a group G satises (BP) if and only if every nitely generated subgroup of G does. Since a single non-trivial element forms a generic sequence it implies that every BP-group is torsion-free. One can modify slightly the denition of a BP-group (requiring that all elements in a generic sequence should have innite order) to accommodate torsion, but we elected to consider here only torsionfree groups, leaving the general case for the future. This allows us to demonstrate basic techniques avoiding some painful technicalities. The class of BP-groups is quite broad. For example, any torsion-free hyperbolic group, or any subgroup of it, is a BP-group. To describe some other examples of BP-groups recall that a group G separates (discriminates) a group H if for any non-trivial element h H (any nite set of nontrivial elements h1 , . . . , hk H ) there exists a homomorphism 55

 : H G such that h = 1 (h i = 1 for i = 1, . . . , k ). It turns out that any group discriminated by a BP-group is a BP-group (Proposition 2). Every torsion-free abelian group satises (BP) since one-element sequences are the only generic ones in abelian groups. Moreover, according to Theorem 24 from Section 5.2 a BP-group satisfying a non-trivial identity has to be abelian (this puts also some restrictions on subgroups of BP-groups). The big powers condition has been used in various forms in dierent areas of group theory. Lyndon and Appel used it (though implicitly) for solving one-variable equations in free groups ([L1], [Ap]). Makanin applied (BP) in his seminal work [Mak] on decidability of equations in free groups in the form of Bulitkos lemma. G.Baumslag used some form of the (BP) condition in free groups to prove residual freeness of surface groups [BG]. In [Ol1] Olshanskii introduced the (BP) condition explicitly in the (equivalent) form of a separation condition (S) (see Section 5.2) and proved that (S) holds in torsion-free hyperbolic groups. Myasnikov and Remeslennikov showed that the Lyndons Z [t]-completion F Z [t] of a free group F is a BP-group, and used this to construct free Lyndon length functions on F Z [t] [MR3]. It turned out that a similar technique can be applied in studying Z [t]-completions of arbitrary torsion-free hyperbolic groups [BMR3]. The method of big powers became extremely useful in algebraic geometry over groups,

56

in particular, in characterizations of the coordinate groups of irreducible algebraic varieties over torsion-free hyperbolic groups [BMR1]. In [KM2] and [KM3] Kharlampovich and Myasnikov described some canonical forms of coordinate groups of irreducible varieties over free non-abelian groups, the BP-method was one of the key technical ingredients of their proof. The main goal of this paper is to show that many groups, beyond the torsion-free hyperbolic ones, are BP-groups. In Section 5.2 we discuss equivalent denitions and basic properties of BP-groups. The main result of Section 5.3 is Theorem 25 which states that a free product of two BP-groups is a BP-group. Here, following [MR3] we introduce an important technical notion - the separation condition (CS), which allows one to control cancellation in BPgroups. In Section 5.4 we introduce a relative form of the big powers condition, the (BP)condition modulo a subgroup. Subgroup of this type are called strongly isolated subgroups. It is not hard to see that centralizers in BP-groups are strongly isolated (Lemma 5). Theorem 26 claims that every isolated quasi-convex subgroup of a torsion-free hyperbolic group is strongly isolated. In Sections 5.6 and 5.5 we prove the following two theorems, which are the main results of the paper. To explain it recall that a subgroup M of a group G is malnormal if for any

57

g G \ M g 1 M g M = 1. Following [KM1] we say that two subgroups A, B of a group G are conjugately separated if for any g G g 1 Ag B = 1. Theorem 28 Let G and H are BP -groups and A, B are isomorphic subgroups of G and H . If A and B are strongly isolated and, at least, one of them is malnormal in G or H , then the free product with amalgamation GA=B H is a BP-group. Theorem 27 Let G be a BP-group and A, B be conjugately separated isomorphic subgroups of G. If A and B are strongly isolated in G and either A or B is malnormal in G then the HNN-extension G, t | tAt1 = B is a BP-group.

5.2

Elementary properties of BP-groups

The following condition (S) was introduced in [Ol1]. Denition 29 A group G satises the separation condition (S) if for any positive integer k and any tuples u = (u1 , . . . , uk ) and g = (g1 , . . . , gk+1 ) of elements from G such that [ui i+1 , ui+1 ] = 1 f or i = 1, . . . , k 1, there exists an integer n(u, g ) such that
k 1 g1 u 1 g2 . . . gk uk gk+1 = 1 g

(1)

58

for any integers 1 , . . . , k n(u, g ). It turns out that the conditions (BP) and (S) are equivalent. To prove this we need the following lemma. Lemma 3 Let G be a BP-group, u = (g1 , ...gk ) Gk be a commutation-free tuple, and h G. Then there exists a constant n(u/h) such that g1 1 . . . g k k = h for every 1 , . . . , k > n(u/h). Proof. Suppose that for any integer n there exist integers n1 , . . . , nk > n such that g1 n1 . . . gk nk = h. Observe that the tuple w = (g1 , . . . , gk , gk1 1 , . . . , g1 1 ) is commutation-free. Let n(w) be a boundary of separation for u. By our assumption there are 1 , . . . , k > n(w) such that g1 1 . . . gk k = h. Similarly, there are some 1 , . . . , k > n(w) + k such that g1 1 . . . gk k = h. 59

So g1 1 . . . gk k k (gk1 1 )
k 1

. . . (g1 1 )

=1

and all the exponents above are greater than n(w) - contradiction. Proposition 1 A group G satises the big powers condition (BP) if and only if it satises the separation condition (S). Proof. Obviously, (S) implies (BP) (put g = (1, . . . , 1) in (S)). Assume now that (BP) holds in G. Consider tuples u = (u1 , . . . , uk ) and g =

(g1 , . . . , gk+1 ) of elements from G and suppose that for any integer n(u, g ) there are integers 1 , . . . , k n(u, g ) such that

k 1 g1 u 1 g2 . . . gk uk gk+1 = 1. g

We need to show that [ui i+1 , ui+1 ] = 1 for some i (1 i k 1). Observe, that
k 1 1 1 2 1 = g1 u 1 g2 . . . gk uk gk+1 = (u1 ) (u2 g 1
1 g 1 g1 2

. . . (ukk

1 g 1 ...g1 k

) g1 . . . gk+1

(2)

Hence (u11 )1 (u22

g 1
1 g 1 g1 2

. . . (ukk

1 g 1 ...g1 k

= (g1 . . . gk+1 )1 .

It follows from Lemma 3 that the tuple (u11 , (u22

g 1
1 g 1 g1 2

) , . . . , (ukk 60

1 g 1 ...g1 k

) )

has commutation, i.e., [ui i

1 g 1 ...g1 i+1 , ui+1 1 g 1 ...g1

]=1

for some i (1 i k 1). Conjugating by g1 . . . gi we obtain

i+1 ] = [ui i+1 , ui+1 ], 1 = [ui , ui+1

g 1

as desired. The following proposition provides a host of examples of BP-groups. Proposition 2 1) A subgroup of a BP-group is a BP-group;

2) A group discriminated by a BP-group is a BP-group; 3) Every torsion-free hyperbolic group is a BP-group. Proof. 1) is obvious. To prove 2) suppose H is discriminated by a BP-group G. If u = (h1 , . . . , hk ) H k is a commutation-free sequence in H , then there exists a homomorphism : H G such that v = (u) is commutation-free in G. Therefore there exists a constant n(v ) such that
1 k (h . . . (h =1 1) k)

for any i n(v ). Hence

k 1 h 1 . . . hk = 1

61

for every such i . This shows that H is a BP-group. 3) has been proven in [Ol1]. Proposition 3 1) Direct product of two non-abelian groups is not a BP-group;

2) Direct product of a BP-group and a torsion-free abelian group is a BP-group. Proof. Let G and H be non-abelian groups. Suppose a, b G and x, y H are two pairs of non-commuting elements. Put u1 = (1, y ), u2 = (a, x), u3 = (b, x1 ), u4 = (1, y 1 ), u5 = (b1 , x), u6 = (a1 , x1 ). Then u = (u1 , . . . , u6 ) is commutation-free. But for an arbitrary large n we have
n un 1 . . . u6 = 1.

Hence G H is not a BP-group. 2) is straightforward. Remark 1 The rst statement above shows, in particular, that a group separated by a BP-group may not be a BP-group. And the second one implies that there are non-abelian BP-groups with non-trivial center. Lemma 4 Let G be a BP-group. Then the following holds:

62

1) For any elements x, y G and any non-zero integers n, m if [xn , y m ] = 1 then [x, y ] = 1; 2) Extraction of roots is unique (wherever dened) in BP-groups. Proof. Let [xn , y m ] = 1 for some non-zero integers n, m. If [x, y ] = 1 then the tuple (x, y, x1 , y 1 ) is commutation-free. But for any k > 0 we have xnk y mk xnk y mk = 1 - contradiction. This proves 1). 2) follows easily from 1). Theorem 24 Every BP-group satisfying a non-trivial identity is abelian. Proof. Suppose that a BP-group G satises a non-trivial identity W (x1 , ...xn ) = xi1 k1 . . . xil kl = 1,
1 where xij {x1 , . . . , xn }, xij = x ij +1 . If there exist n pair-wise non-commuting elements

a1 , ..., an G then the tuple (ai1 k1 , ..., ail kl ) is commutation-free, but for any k > 0 we have W (a1 k , ..., an k ) = ai1 k1 k ...ail kl k = 1 - contradicting (BP). Thus, it suces to show that if G is non-abelian then there are n pair-wise non-commuting elements in G. Suppose x, y G such that [x, y ] = 1. We 63

claim that the elements xy, xy 2 , ..., xy n are pair-wise non-commuting. Indeed if s = t and [xy s , xy t ] = 1 then 1 = [xy s , xy t ] = [xy t y st , xy t ] = [xy t , xy t ]
y s t

[y st , xy t ] =

= [y st , xy t ] = [y st , y t ][y st , x] = [y st , x] and by Lemma 4 [x, y ] = 1 - contradiction. This shows that G is abelian, as desired.

5.3

Free products of BP-groups

In this section we show that a free product of two groups is a BP-group if and only if every factor is a BP-group. Let G and H be groups. By G H we denote a free product of G and H . Every element w G H can be uniquely expressed in a reduced normal form w = u1 ... un , where ui = 1, ui is in one of the factors G or H, and ui , ui+1 are not in the same factor. The number n is called the length of w and it is denoted by |w|. If u, v G H such that the reduced forms of u1 and v begin with elements from dierent factors then we sometimes write u v instead of uv. Clearly uv = u v |uv | = |u| + |v |.

64

Put 1 c(u, v ) = (|u| + |v | |uv |). 2 Observe that c(u, v ) is either a non-negative integer n or a rational number of the form
1 . The number c(u, v ) measures the length of cancellation in the product uv . Now, for n+ 2

a non-negative rational number we denote by u v the product uv provided c(u, v ) , i.e., uv = u v c(u, v ) . In particular, if = 0 then u v = uv. Also, if u = u1 f1 , v = f2 v1 , where f1 , f2 from the same factor (G or H ) and f1 f2 = 1, then uv = u 1 v .
2

Lemma 5 The following equivalence holds for every positive integer n u n v u = u1 u2 , v = v1 v2 , |u2 |, |v1 | n, uv = u1 (u2 v1 ) v2 . Proof. By a straightforward verication. Denition 30 Let u, v G H . Set c (u, v ) = limsupk c(uk , v k ) If c (u, v ) is nite then we say that (u, v ) satises separation condition (SC) and denote by k = k (u, v ) a positive integer (in this case it exists) such that c (u, v ) = c(uk , v k ) 65

and for every m, n k c(un , v m ) c(uk , v k ). An element w = u1 . . . un G H is called cyclically reduced (c.r.) if un and u1 are in dierent factors or n 1. We call w weakly cyclically reduced (w.c.r.) if either
1 un = u 1 , or n 1. Every element w G H can be uniquely presented in the form 1 w = wc w wc ,

where w is w.c.r. This element w is called the w.c.r. part of w. Lemma 6 If u, v G H and either |u | 1 or |v | 1 then c (u, v ) < . Proof. If, say, |u | 1 then for any m
1 1 m uc | = |u|. uc | |u |um | = |u c u c u

Therefore for every m c(um , v n ) |um | |u|, hence c (u, v ) < |u|. Lemma 7 Let u G H and |u | 2 then: 1) |uk+1 | |uk | + |u | 1; 66

2) |u k | > k |u | k + 1. Proof is obvious. Lemmas 5 and 7 imply the following result. Lemma 8 Let u, v G H and |u |, |v | 2. If c (u, v ) < , then for every e > max{c (u, v ), k (u, v )} and every m, n > e the following holds un v m = une |uc |+ 1 (ue v e )|vc |+ 1 v me .
2 2

Proof. Let e > max{c (u, v ), k (u, v )} and let m, n > e. Then by denition c(un , v m ) e. We have
1 1 un = u ne 1 u e uc , v m = v c v me 1 v e vc . c u
2 2

Since |u |, |v | 2 it follows from Lemma 7 that |u e uc | > e c(un , v m ), |v e vc | > e c(un , v m ). By Lemma 5
1 1 e ne 1 ( un v m = u v ) 1 v u e uc v c me vc c u
2 2

The equality above can written in the form

1 e 1 e 1 1 v vc ) |vc |+ 1 (vc ne uc ) |uc |+ 1 (u uc v c v me vc ) = un v m = (u c u c u
2 2

67

= une |uc |+ 1 (ue v e ) |vc |+ 1 v me ,

2 2

as required. Lemma 9 Let u, v, w G H . Then c (u, v ) < c (w1 uw, w1 vw) < .
1 1 Proof. Let u = u uc , v = v c v vc . In view of Lemma 6 we can assume that c u

|u |, |v | 2. 1) We show rst that

1 c (u, v ) < c ( u, uc vu c ) < .

There exists l such that

1 l |vc v | > |uc |, |v l vc | > |uc |.

Then for any n 1

l 1 n l 1 uc v n+2l u c = uc v |vc |+ 1 v |vc |+ 1 v uc .
2 2

This implies that

1 1 1 c( um+l , uc v n+l u um+l uc , v n+l u um+l , uc v n+l u c ) c( c ) c( c ) + |uc |

(3)

Now, for every m, n k (u, v ) + 1

1 n+2l 1 1 c( um , (uc vu ) = c( um , uc v n+2l u um uc , v n+2l u c ) c ) c( c ) =

68

1 m1 n+l c(uc u |uc |+ 1 um1 , v n+l |vc |+ 1 v l u , v ) c (u, v ). c ) = c(u

2 2

It follows that
1 c ( u, uc vu c ) c (u, v ).

Reversing the argument above and using (3) one can show that
1 c ( u, uc vu c ) + |uc | c (u, v ).

This proves 1). 2) In view of 1) we can assume now that u = u . Similarly to the argument in 1), one can easily show that if u = p 1 q then
2

c (u, v ) < c (qp, p1 vp) < . 3) Observe, that the w.c.r. part of w1 uw is a cyclic permutation of u = u , i.e., w1 uw = qp for some p, q such that u = p 1 q. It follows from 1) and 2) now that
2

c (u, v ) < c (w1 uw, w1 vw) < . This proves the lemma. Lemma 10 Let u, v be cyclically reduced elements of G H such that |u|, |v | 2. If for some m, n > 1 c(um , v n ) |u| + |v | 69

then u and v are both powers of one and the same element from G H . In particular, if both u and v are not proper powers then u = v . Proof. If |u| = |v | 2 then obviously u = v . Let now |v | > |u| and let |v | = s|u| + k, for some integers 1 s, 0 k < |u|. Let u = u1 u2 where |u1 | = k . Then v = us u1 and hence u = u2 u1 . So [u2 , u1 ] = 1. Since |u| 2 we deduce that u2 , u1 both are powers of some w G H . Therefore u and v are also powers of w. It follows that u = v = w in the case when u and v are not proper powers. Proposition 4 Let G, H be arbitrary groups, and u, v be elements of G H such that [u, v ] = 1. Then c (u, v ) < . Proof. Suppose (u, v ) is a counterexample to the statement above with a minimal possible total length |u| + |v |. Observe that by Lemma 6 |u |, |v | 2. By Lemma 9 every pair conjugate to (u, v ) is also a counterexample. Assume that at least one of u, v , say u, is
1 1 not a w.c.r. element, i.e., u = u uc and uc = 1. Then the pair ( u, uc vu c u c ) is a

counterexample of the length

1 1 |u | + |uc vu | + |uc | + |v | + |u c | |u c | = |u| + |v |. 1 1 The equality above holds if and only if uc vu c = uc v uc . Therefore uc v = uc v and

hence for every m, n > 0 un v m = un v m , i.e., c (u, v ) = 0, - contradiction. This shows 70

that both u and v are w.c.r. Assume now that at least one of u, v , say u, is not cyclically reduced. In this event, u = a1 u a2 , where a1 , a2 are non-trivial elements from the
1 same factor, say G, and a1 = a 2 . Notice, that v starts with a non-trivial element a3

from the same factor G (otherwise there is no cancellation between u and v , as well as
1 between un and v m ), so v = a3 v . If a2 a3 = 1 then un v m = un 1 v m , and c (u, v ) 2 ,
2

1 1 - contradiction. Hence a3 = a 2 . In this case the element a2 ua2 = a2 a1 u has smaller 1 1 length then u, but the element a2 va = va has length at most equal to the length 2 2

of v - contradiction. It follows that both elements u and v are cyclically reduced. Since c (u, v ) = we have c(um , v n ) |u| + |v | for some big enough m, n. This implies that (u1 )m and v n has common initial segment of length |u| + |v |. By Lemma 10 [u, v ] = 1 contradiction. This shows that (u, v ) satises (SC). Theorem 25 A free product G H is a BP-group if and only if G and H are BP-groups. Proof. Let G and H be BP-groups. To prove that G H is a BP-group we have to show that an arbitrary generic sequence of elements from G H is independent. It suces to consider only sequences of elements which are not proper powers. Let u = (u1 , . . . , uk ) be such a generic sequence. We divide our proof into three cases.
1 1) Let ui = wi gi wi , and |gi | = 1 (i.e., gi G H ), i = 1, . . . , k .

For elements g, h of a torsion-free group with g = 1 by n(g/h) we denote a minimal positive number such that g n = h for any integer n n(g/h) (such number n(g/h) always 71

exists). Now, for every i we dene numbers mi , ni as follows. Suppose gi G. If the

1 reduced form of wi1 wi ends on an element from H then mi = 1; if this form ends on 1 some g G then mi = n(gi /g ); if wi1 wi = 1 and gi1 G then mi = n(gi /gi1 ); if 1 1 wi1 wi = 1 and gi1 H then mi = 1. Similarly, if the reduced form of wi wi +1 starts

with an element from H then ni = 1; if the form starts with an element g G then
1 ni = n(gi /g ); if wi wi +1 = 1 then ni = n(gi /gi+1 ). In the case when gi H we dene

numbers mi and ni in the same manner, just interchanging G and H . For uniformity put m1 = nk = 1. It is easy to see that n(u) = max{m1 , n1 , . . . , mk , nk } is a separation boundary for the tuple u, so u is independent. 2) Let |u i | 2 for all i = 1, . . . , k. By Proposition 4 for all pairs (ui , ui+1 ) one has c (ui , ui+1 ) < . Denote by ei any number satisfying the inequality from Lemma 8 for the pair ui , ui+1 . Put t = max{e1 , . . . , ek1 }. Then for arbitrary integers ni 2t + 2 the following holds
nk n1 t n2 2t k 1 t t 1 1 ut un 2 ut 1 . . . uk = u1 1 u2 2 u2 2 u3 . . . u k 1 n 2t nk t t k1 ut , k1 uk k uk

nk 1 1 where i < 2 |ui |. This shows that un 1 . . . uk = 1 and the sequence u is independent.

72

3) Suppose now that u = (u1 , . . . , uk ) is an arbitrary generic sequence. We can divide u into segments u = v1 . . . vm such that each segment vj is either of the type 1) or the type 2), and successive segments vi , vi+1 are of dierent types. As we showed above, for every vi there exists a separation boundary ki . We use these numbers ki to obtain a separation boundary for u. Suppose that vi is of the type 1) and vi+1 is of the type 2). We start with dening some auxiliary numbers qi , ri+1 . Let w1 gw be the last element of the sequence vi and f be the rst element of the sequence vi+1 . Denote by ri+1 a positive integer such that |f ri+1 | |w| + 1. If the reduced form of wf ri+1 starts with an element a from the same factor as g then put qi = n(g/a), otherwise qi = 1. Similarly, if vi is of the type 2) and vi+1 is of the type 1) then we dene numbers si , pi+1 as follows. Let now f be the last element of the tuple vi and w1 gw be the rst element of the tuple vi+1 . Put si be a positive integer such that |f si | |w| + 1. If the reduced form of f si w ends with an element a from the same factor as g then put pi+1 = n(g/a), otherwise pi+1 = 1. This way for every index i we dened some of the numbers pi , qi , ri , si . For uniformity uniform we assume all other yet undened numbers to be equal to 1. Put n(u) = max{ki + pi + qi + ri + si | i = 1, . . . , m}. It is not hard to see that n(u) is a separation boundary for u. This shows that G H is a BP-group, as desired. The converse statement is obvious, since G and H are subgroups of G H . This proves 73

the theorem.

5.4

Big Powers Condition Modulo Subgroups

Recall that a subgroup A of a group G is called isolated if for every g G if g k A for some non-zero number k then g A. Let G be a group, A be a subgroup of G and u = (u1 , ..., uk ) be a generic sequence of elements of G. We say that: 1) u is A-generic if there exists i such that ui A; 2) u is A-independent if there exists an integer n such that for any integers 1 , . . . , k n
k u1 1 ...u k A.

Denition 31 Let G be a group. A subgroup A of G is called strongly isolated if every A-generic sequence of elements of G is A-independent. Sometimes, if A is a strongly isolated subgroup of G we say that G satises (BP) modulo A. Observe, that a strongly isolated subgroup is isolated.

74

Lemma 11 Let A be a subgroup of a group G. If u = (u1 , . . . , uk ) is an A-independent sequence of elements from G then for any element h G there exists a number n = nA,h (u) such that for any integers i n u1 1 . . . uk k h A. Proof is similar to the one of Lemma 3. Proposition 5 Let G be a BP-group. Then any centralizer of G is strongly isolated. Proof. Let A = CG (X ) be a centralizer of a subset X G and let u = (u1 , . . . , uk ) be an A-generic sequence from G. Assume u is not A-independent and k is minimal with this property. Then u1 A. Indeed, if u1 A then the sequence (u2 , . . . , uk ) is still A-generic and it is not A-independent, since for any integers i
k k 2 1 [u 1 . . . uk , X ] = 1 [u2 . . . uk , X ] = 1.

This shows that there exists y X such that [u1 , y ] = 1. Now if [uk , y ] = 1 then for any integers i
k 1 k 1 1 [u 1 . . . uk , y ] = 1 [u1 . . . uk1 , y ] = 1.

Taking (u1 , . . . , uk1 ) instead of u, if necessary, we may assume that [uk , y ] = 1. Now, the sequence
1 1 v = (u k , . . . u1 , y, u1 , . . . , uk , y )

75

is generic, but it is not independent, since for every integer n there are integers 1 , . . . , k n such that
k 1 [u 1 . . . uk , X ] = 1

in particular for every integer m

k m 1 [u 1 . . . uk , y ] = 1.

This contradiction shows that A is strongly isolated. Proposition 6 A free factor of a BP-group is strongly isolated. Proof. Let G H be a BP-group. We need to show that G is strongly isolated. Let u = (u1 , . . . , uk ) be a generic sequence from G H . In the proof of Theorem 25 we showed that the syllable length of the element
k 1 u = u 1 . . . uk

goes to when all i go to , unless each element ui has the cyclically reduced length equal to 1, i.e., |u i | = 1. This shows that if u is G-generic and it is not G-independent then |u i | = 1 for i = 1, . . . , k. Moreover, it can be easily seen from the proof of the case 1) of Theorem 25 that in this case all u i have to be from the same free factor, in fact from G, otherwise |u | > 1 for big enough i . Further analysis shows that ui G for every i = . . . , k - contradiction with A-genericity of u. This proves the result. 76

Now let G be a torsion-free hyperbolic group with a nite generating set S . For elements g, h G denote by d(g, h) = dS (g, h) the length of a geodesic path from g to h in the Cayley graph (G, S ) of G with respect to S . Denote l(g ) = d(1, g ). Given x, y, w G we dene an inner product (x y )w by 1 (x y )w = (d(x, w) + d(y, w) d(x, y )). 2 In particular, if w = 1 then we denote (x y )1 by c(x, y ). We need the following known results. Lemma 12 [Ol1] Let G be a torsion-free hyperbolic group and g, h G such that [g, h] = 1. Then there exists a constant C = C (g, h) such that for any m, n c(g n , hm ) C . Lemma 13 [ABC] Let g be an element of innite order in a hyperbolic group G. Then there exists an integer N (G) such that l(g N m ) m for any positive integer m. For x, y G by [x, y ] we denote a geodesic segment from x to g in (G, S ). Let [x1 , ..., xn ] be a broken line which connects points x1 , . . . , xn in (G, S ) by the geodesic segments [x1 , x2 ], . . . , [xn1 , xn ]. In particular, [x1 , ..., xn , x1 ] is a geodesic n-gon in (G, S ). Lemma 14 [Ol2] Let G be a torsion-free hyperbolic group with a hyperbolicity constant . If c 14 , c1 12(c + ), and a geodesic n-gon [x1 , . . . , xn , x1 ] with n 3 satises the 77

following conditions: |xi1 xi | > c1 for i = 2, . . . , n and (xi2 xi )xi1 c for i = 3, . . . , n. Then the broken line p = [x1 , . . . , xn ] is contained in the closed 2c - neighborhood of the side [xn , x1 ] and the side [xn , x1 ] is contained in the closed 14 -neighborhood of p. Denition 32 A subset A (G, S ) is quasi-convex if there exists a real number > 0 such that every geodesic [a, b] with endpoints a, b A is contained in -neighborhood of A. Theorem 26 Let G be a torsion-free hyperbolic group and A be a quasi-convex isolated subgroup of G. Then A is strongly isolated. Proof. Assume that a quasi-convex subgroup A is not strongly isolated in G. Let u = (g1 , . . . , gk ) be an A-generic sequence from G which is not A-independent of minimal length k . It follows that g1 A. By our assumption for any positive integer n there exist integers n1 , . . . , nk > n such that g1 n1 . . . gk nk A. We consider now a k + 1-gon [x1 , . . . , xk+1 ], where x1 = 1 and for i = 2, .., k + 1 xi = g1 n1 g2 n2 . . . gi1 ni1 . Notice that (xi2 , xi )xi1 = c(gi1 ni1 , gi ni ), so by Lemma 12 there exists a constant Ci = Ci (gi1 , gi ) such that for any ni1 , ni we have (xi2 , xi )xi1 Ci . 78

Let be a constant of hyperbolicity of G. Put C = max{Ci | i = 1, . . . , k }, C1 = 12(C + ), C2 = max

r {|gi |}.

0ik, 0r<N

Lets take n = (C1 + C2 )N where N is the constant which appears in Lemma 13. Then for every ni n we have ni = kN + r, where k C1 + C2 and 0 r < N , so by Lemma 13

ni kN +r r d(xi1 , xi ) = |gi | = |gi | C1 + C2 |gi | C1 = 12(C + ), i = 2, . . . , k + 1,

This implies that there exists a number n such that for any m n the (k + 1)-gon
n1 [x1 , . . . , xk+1 ] satises the conditions of Lemma 14. Hence the element g1 is contained

in the closed 2C -neighborhood of the geodesic segment [1, g1 n1 . . . gk nk ], and therefore in + 2C -neighborhood of A (here is a quasi-convexity constant for A). It means that there exists u G such that g1 n1 u A and l(u) + 2C . Since there are only a nite number of elements u with length less than or equal to + 2C and there are innitely many such n1 (at least one for every n) we see that there are n = m such that g1 n u A and g1 m u A. It follows that g1 nm A, but because A is isolated we deduce that g1 A - contradiction. This shows that A is strongly isolated.

79

5.5

HNN Extensions and Big Powers Condition

In this section we show that under some conditions on associated subgroups HN N extension of a BP group is again BP . Consider HN N extension G of a group G relative to its associated subgroups A and B and isomorphism : A B . So G =< G, t|tat1 = (a), a A > . Now if w is an element of G , then we can write this element in a reduced form w = g0 t1 g1 . . . tn gn (n 0), where in the sequence g0 , t1 , g1 , . . . , tn , gn there is no consecutive subsequence t, gi , t1 with gi A or t1 , gj , t with gj B . Now we state the so called Brittons lemma. Lemma 15 [LS] If the sequence g0 , t1 , g1 , . . . , tn , gn is reduced and n 1, then g0 t1 g1 . . . tn gn = 1 in G . Reduced form of the element of G is not a normal form, which means that the products of the elements in two distinct reduced sequences may be equal in G . Let choose a set

80

of representatives of the right cosets of A in G, and a set of representatives of the right cosets of B in G. We will assume that 1G is the representative of both A and B . Denition 33 A normal form is a sequence g0 , t1 , g1 , . . . , tn , gn (n 0) where (i) g0 is an arbitrary element of G, (ii) if i = 1 then gi is a representative of a coset of B in G, (iii) if i = 1, then gi is a representative of a coset of A in G, and (iv) there is no consecutive subsequence t , 1, t . It is known that every element w of G has a unique representation as w = g0 t1 g1 . . . tn gn where g0 , t1 , g1 , . . . , tn , gn is a normal form. If now w = g0 t1 g1 . . . tn gn is a reduced form of element w, then we can assign to element w the length |w| = n which is a number of occurrences of t+1 in w. Then |w| is a well dened length function on G . Clearly this length function has a kernel which is the group G. 81

An element w = g0 t1 g1 . . . tn gn is cyclically reduced if all cyclic permutations of the sequence g0 , t1 , g1 , . . . , tn , gn are reduced. Clearly, every element of G is conjugate to a cyclically reduced element. As always we write w = w1 w w1 1 , where w is the cyclically reduced part of w. Like in a free product case we can dene the notion of separation condition for two elements of G and prove the following lemma. Lemma 16 Let u, v, w G . If (u, v ) satises (SC) then (w1 uw, w1 vw) satises (SC). Lemma 17 Let u, v G such that |u |, |v | 1 and [u, v ] = 1. Then these elements satisfy separation condition. Proof. As in the proof of similar lemma for free product we can reduce a general case to the cyclically reduced one. So let u and v be cyclically reduced elements where |u|, |v | 1. Lets assume that pair (u, v ) does not satisfy (SC) condition and cancellations go to innity. Suppose that element u has the following form u = u1 v k where the number |k | is maximal (case v = ul v1 is similar). Now consider the product u2 v k+2 = (u1 v k u1 )v v , 82

but u1 v = u1 |v| v so there may be cancellations if |u1 | < |v |. In this case cancellations will keep going if u1 = av1 1 where a G and v = v1 v2 . So we have u = av1 1 v k and v = v1 v2 . Now u2 v k+2 = (av1 1 v k )a(v2 v1 v2 ) = (av1 1 v2 1 . . . v2 1 v1 1 )a(v2 v1 v2 ). Case 1. |v1 | |v2 |. Then for further cancellations v1 1 has to have the following form v1 1 = v 1 v2 1 a1 . So now u2 v k+2 = (av1 1 v2 1 ... v2 1 v 1 )(v1 v2 ). But |v2 1 v 1 | = |v1 | and if there are further cancellations then we have v2 1 v 1 v1 = a where a G, so u2 v k+2 = (av1 1 v2 1 ... v1 1 )a (v2 ) = (av1 1 v2 1 ... v 1 v2 1 a1 )a (v2 ) , and therefore a = a. And we have (v2 1 v 1 )v1 = a so v1 = v 1 1 v2 a. So we have the following equation av2 v 1
1

= v1

v2 a = v1 .

Now well show that u = av1 1 v k and v = v1 v2 commute. It is enough to show that [av1 1 , v ] = 1. We have av1 1 v = (av 1 v2 1 a1 )(av2 v 1 83
1

v2 ) = av2

on the other hand v av1 1 = (av2 v 1

1

v2 )av 1 v2 1 a1 = av2 av2 v 1 v 1 v2 1 a1 = av2

, so [av1 1 , v ] = 1. Therefore [u, v ] = 1. Contradiction. Case 2. |v2 | > |v1 |. In this case we have v2 = a1 v1 v 2 . And applying the same considerations as above we have v2 = v 2 v1 a1 . So we have the following equation a1 v1 v 2 = v 2 v1 a1 . Again we can show that [av1 1 , v ] = 1. Indeed av1 1 v = av2 = v1 v 2 , on the other hand vav1 1 = v1 v2 av1 1 = v1 v 2 v1 a1 av1 1 = v1 v 2 . So again we have [u, v ] = 1. So what we get is the fact that starting from some big number k in the product uk v k all cancellations stop, which gives us the separation condition for u and v . If neither u = u1 v k nor v = ul v1 for some numbers k and l, then u|v|1 v n and un |u|1 v for any n and m. Obviously in this case (u, v ) satises (SC ). Denition 34 Let G be a torsion-free group and A, B are subgroups of G. We say that A and B are conjugately separated, if for any element g G gAg 1 B = {1G }. Theorem 27 Let G be a torsion-free BP-group, A, B are conjugately separated isomorphic subgroups of G with : A B an isomorphism, and either A or B is malnormal 84

in G. Suppose also that G has a BP condition modulo A and modulo B . Then G, t | tat1 = (a), a A is a BP-group. Proof. Suppose that B is malnormal in G. To prove that G is a BP-group we have to show that an arbitrary commutation-free tuple of elements of G satises the BP condition. Let u = (u1 , . . . , um ) be a commutation-free tuple of elements from G . We will show how to nd a separation boundary for u. First of all we divide our tuple u into subsequences of dierent types which we are going to describe. 1) Consider the subsequence u = (ui1 , ..., uij ) of maximal length where uik = wj gik wj 1 , k {1, ..., j }. Note that we can assume that wj ends on t , where = +1. Suppose that wj ends on t, then every element gik doesnt belong to A. In this case, using the fact that G is BP modulo A, we associate with the tuple u a number nA (gu ), where gu = (gi1 , ..., gij ). Clearly nA (gu ) is a separation boundary for the tuple u which we denote by n(u ). Similarly, if wj ends on t1 then n(u ) = nB (gu ).

85

2) Consider the subsequence u = (uj1 , ..., ujl ) of u of maximal length such that for every i {1, ..., l} u ji = u ji+1 , where uji = u ji gji u ji 1 . Now for every i we dene numbers mi , ni as follows. Consider u i gi u i 1 u i+1 gi+1 u i+1 1 .
| 1. Suppose that u ends on t, so g A. If u 1 u Case a. |u i i i i i+1 starts from g0 t,

then mi = 1. If u i 1 u i+1 starts from g0 t1 , then mi = nA,g0 (gi ). Suppose now that u i+1 ends on t. If u i 1 u i+1 ends on t1 g0 then ni = 1. If u i 1 u i+1 ends on tg0 , then ni = nA,g0 (gi+1 ). All other cases can be done in the same manner. Case b. u i = 1. If i = j1 , then we have gi u i+1 gi+1 u i+1 1 . Note that in this case u i+1 = u i = 1. In this case we take mi = ni = 1. Let now i > j1 . Consider u i1 gi1 u i1 1 gi u i+1 gi+1 u i+1 1 . Suppose that ui1 written in the normal form ends on t1 g and ui+1 starts from g 1 t. In this case g B and if gi B , then using malnormality of B in G we take mi = ni = 1. If gi B and ggi g 1 B , then since B is isolated we take mi = ni = 1. Suppose that ui1 ends on tg and ui+1 starts from g 1 t1 and gi B . Here we can use the fact that A and B are conjugately separated and take mi = ni = 1. Now suppose that ui1 ends on tg and ui+1 starts from g 1 t1 and gi = a A. So we have ui1 gi ui+1 = ...t gn1 tg a g 1 t1 g 1 t ....

86

Since A is not malnormal in G it might happen that gag 1 = a A so we have ta t1 = b B . Lets consider the case when = 1 and = 1. Clearly there exists l such that for any k l gn1 bk g 1 B . Indeed otherwise gn1 bs g 1 B for some s > 0 and gn1 bt g 1 B for some t > 0, so gn1 bs gn1 1 B , but B is malnormal in G. So we take mi = l, ni = 1. All other cases can be done using the same considerations. It is easy to see that n(u ) = max{m1 , n1 , ..., ml1 , nl1 } is a separation boundary for u . 3) Consider the subsequence u = (us1 , ..., usl ) of u of maximal length such that |u si | 1 for every i = 1, ..., l. Then like in free product case using Lemma 17 we can nd a separation boundary n(u ) for u . Now we can divide u into segments u = v1 ... vm such that each segment vj satises either 1) or 2) or 3) and successive segments vi , vi+1 are of dierent types. As we have shown above for every tuple vi there exists a separation boundary ki . Now for each pair (vi , vi+1 ) we dene numbers li and ri to construct a separation boundary for u. If vi 1) and vi+1 2) (or vi 2) and vi+1 1)) then considering the last element of vi and the rst element of vi+1 we dene numbers li and ri exactly in the same manner as we dened mi and ni in 2). Let now vi 3) and vi+1 2). Let uj be the last element of subsequence vi and w g w1 be the rst element of vi+1 , where |w| 1.

87

Case 1. Let w = uj k w . Then we can take s > k + 2. Consider the product uj s (w g w1 ) = uj sk x w g w1 where x < |u|. Now if w doesnt cancel out then we take li = s, ri = 1. Suppose that w cancels out then we can consider this product in the following form usk2 (u2 w )g w1 . If w1 starts from t (note that in this case g B ) and u2 w ends on t1 g0 then we take li = s, ri = nB,g0 (g ). All other cases can be done using the arguments above. Case 2. Suppose that we have uj = u w1 . So u2 (w g w1 ) = u u g w1 . And we can take li = 2 and nd ri using the arguments above. We take n(u) = max{r1 , l1 , n(v1 ), ..., rm1 , lm1 , n(vm1 ), n(vm )}. It is not hard to see that n(u) is a separation boundary for u.

5.6

Free Products With Amalgamation and Big Powers

Following the strategy outlined in [GKM] we obtain similar results for amalgamated products as corollaries of Theorem 27. Lemma 18 [GKM] Let A and B be subgroups of groups G and H respectively, : A B an isomorphism. The groups A and B can be considered as isomorphic subgroups of the

88

free product G H . Let us denote by E (G, H, ) = G H, t | t1 at = (a) the HNN-extension, associated with , of the group G H . The amalgamated product G H is embeddable in E (G, H, ). Proof. The subgroup Gt , H generated in E (G, H, ) by the t-conjugate of G and H is isomorphic to G H . It can be easily veried using the normal forms of elements in E (G, H, ). Theorem 28 Let G and H be BP -groups, A and B are strongly isolated subgroups respectively in G and H , and either A is malnormal in G or B is malnormal in H . Then the free product with amalgamation GA=B H is a BP-group. Proof. It suces to show that the HNN-extension E (G, H, ) satises all the conditions of Theorem 27. Indeed, in this case E (G, H, ) is a BP-group and by Lemma 18 GA=B H is a subgroup of E (G, H, ), hence it is also a BP-group. First, we claim that either A or B is malnormal in the free product GH . Malnormality is transitive: if X is a malnormal subgroup of Y and Y is a malnormal subgroup of Z , then X is a malnormal subgroup of Z . So we need only to point out that the factors of a free product are malnormal in the product [LS]. Then we need the fact that A and B 89

are conjugate separated, i.e. A B x = 1 for all x G H . It is enough to prove that G and H are mutually conjugate separated in their free product, and this is easily veried using normal forms. Next, we need the fact that the class of BP-groups is closed under free products (Theorem 25). To complete the proof, we note that A and B are strongly isolated in G H . Indeed, it is easy to see that strong isolation is transitive, and the result follows from Proposition 6, which says that free factors of BP-groups are strongly isolated.

One-variable equations in torsion-free hyperbolic groups

6.1

Preliminaries

In this subsection we list some properties of torsion-free hyperbolic groups which we will use in the sequel, some of these properties were already mentioned in the section 1. Let G be a -hyperbolic group, such that an inner product is -hyperbolic. We x the following notations l(g ) = |g | = d(1, g ), 1 c(g, h) = (g h)1 = (l(g ) + l(h) l(g 1 h)) 2 90

and g h c(g 1 , h) . Lemma 19 [G] For any nontrivial element g of a torsion-free hyperbolic group G there exists g that is not a true power in G and a natural number m such that g = g m ; g and m are uniquely determined. In addition, C (g ) =< g >. We call g G a root element if it is not a proper power in G. Lemma 20 Any torsion-free hyperbolic group satyses CSA condition, so commutativity relation is an equivalence relation on G {1G }. Lemma 21 [Ol1] Let G be a torsion-free hyperbolic group and g, h G such that [g, h] = 1. Then g n C (g,h) hm , g s C (g) g t and hs C (h) ht for some constants C (g, h), C (g ) and C (h) that doesnt depend on m,n,s,t provided st 0. Lemma 22 [G] If g is an element of innite order in a hyperbolic group G and be a geodesic path from the vertex corresponding to the identity element to the vertex corresponding to g . Then there is a constant H (g ) which is a Hausdor distance between [1, g n ] and bi-innite path (..., g 1 , , g, ...), for any integer n. 91

Lemma 23 [Lys] [G] If in -hyperbolic group G elements g and h are conjugated, then there exists an element a such that g = a1 ha and |a| < 2(|g | + |h| + 8 ). 1 and 2

6.2

Main technical results

Let G be a torsion-free hyperbolic group. To simplify computations we assume that = max {slim , thin , inner product }. Therefore, for example, we will have no qualms about saying that triangles are -thin while at the same time inner product is -hyperbolic. The next simple lemma follows from the fact that inner product is -hyperbolic. Lemma 24 Let (a, b, c) be a triple of natural numbers such that a min(b, c) , b min(a, c) , c min(a, b) . Then |x y | and z x, y , for some x, y, z {a, b, c} . Lemma 25 Let v1 , ..., vn G, n 2, |vi | > 2(c + ) and vi c vi+1 . Then (v1 . . . vn1 )c+ vn . In particular, |v1 . . . vn | > 2c + 4 and v1 . . . vn = 1. Proof. Proof is done by induction on n. Case n = 2 is trivial.

92

Let (v1 vn2 )c+ vn1 . So |v1 vn2 | + |vn1 | |v1 vn1 | 2(c + ). Consider the following triple c((v1 vn1 )1 , vn ), c((v1 vn1 )1 , vn1 1 ), c(vn1 1 , vn ) = (, , ). From the hyperbolicity of group G this triple satises conditions of the lemma above. We have 1 1 = (|v1 vn1 | + |vn1 | |v1 vn2 |) (2|vn1 | 2c 2 ) > c + 2 2 and c. Now the case when | | is impossible, therefore by the lemma above | | and hence + c + . Hence |v1 vn1 | + |vn | |v1 vn | 2(c + ), and |v1 vn | |v1 vn1 | + |vn | 2(c + ) > 2c + 4 + |vn | 2(c + ) 2c + 4 as desired. The next lemma shows that in the case of three factors we just need a middle element to be long. Lemma 26 (i) For any u, v, w G such that uc1 v c2 w if |v | > c1 + c2 + , then (uv )c2 + w and uc1 + (vw). (ii) For any u, v, w G if (uc1 v )c2 w, then uc1 v c1 +c2 w and uc1 +c2 (v c1 +c2 w). 93

Proof. (i) Clearly c(v 1 , (uv )1 ) > c2 + . Suppose now that c((uv )1 , w) > c2 + , then since the inner product is -hyperbolic c(v 1 , w) > c2 contrary to the conditions. Therefore c((uv )1 , w) c2 + . Similarly we show that c(u1 , vw) c1 + . (ii) By assumptions |u| + |v | |uv | 2c1 and |uv | + |w| |uvw| 2c2 . We have 2c(v 1 , w) = |v | + |w| |vw| |u| + |v | + |w| |uvw| |u| + |v | |uv | + 2c2 2(c1 + c2 ). Similarly we prove the second statement. The next important lemma says that if there is a big cancellation in the product tut1 then we can pick out a big power of u in element t, i.e. t = t c u for some constant c. Lemma 27 Let t, u G such that c(t1 , ut1 ) = 1 (13|u| + 68 ) + , then t = t c u 2 where C (u) =< u >, c < 4|u| + 32 and |u | > . Therefore c(t 1 , ut 1 )
31 |u| 2

+ 98 .

Proof. Suppose that d([a, b], [o, c]) < 5|u| + 34 + , then there exist y [a, b] and x [o, c] such that d(x, y ) < 5|u| +34 + . Clearly |[o, x]| |t||[x, a]| |t| d(x, y ) |u|. Similarly |[x, c]| |t| d(x, y ) |u|, so |[o, c]| 2(|t| d(x, y ) |u|). We have 2c(t1 , ut1 ) = |t| + |ut1 | |[o, c]| |ut1 | + |t| 2|t| + 2(d(x, y ) + |u|) = |ut1 | |t| + 2(d(x, y ) + |u|) < 94

. .b t v. t t . w. v ' t . . o. x z
a

{
1

-1
1

t -1

.c

Figure 6: Lemma 4 < 3|u| + 10|u| + 68 + 2 = 13|u| + 68 + 2. Contradiction. Therefore d([a, b], [o, c]) 5|u| + 34 + > 2 and by Lemma ?? [o, c] O2 ([o, a] [b, c]), so we have z [o, c] such that d(z, w) 2 and d(z, v ) 2 for some points w [o, a] and v [b, c]. Let |w a| |b v | then we take v [b, c] such that |w a| = |b v |. Then |v v | = |w a| |v b| |u| + 4 and |w v | 4 + |u| + 4 = 8 + |u|. Therefore there exist elements t1 and t2 such that t = t2 t1 and |t1 ut1 1 | |u| + 8 . And clearly 5|u| + 34 + d(z, [a, b]) |t1 | + 2 + |u| and hence |t1 | 4|u| + 32 + . By Lemma 23 there exist s G such that |s| < 2(|u| + |u| + 8 + 8 ) = 4|u| + 32 and t1 ut1 1 = sus1 . So s1 t1 = u and we have t = t2 ss1 t1 = t c u where c s < 4|u| + 32 . Note also that |u | = |s1 t1 | |t1 | |s| > .

95

Now, since |t | + |u | |t u | < 2c we have | + |uu t | |t ut | > 2c(t1 , ut1 ) = |t| + |ut1 | |t ut | = |t c u | 2c |t ut | = > |t | + |u | 2c + |ut | 2|u| + |u = 2c(t , ut ) + 2|u | 2|u| 4c > 2c(t , ut ) + 2 2|u| 4c. Therefore 1 1 1 c(t , ut ) < (13|u| + 68 ) + |u| + 2c < 2 31 1 < (13|u| + 68 ) + |u| + 8|u| + 64 = |u| + 98. 2 2 Lemma 28 Let t, u G and c(t1 , u ) = 3 + H (u) + |u| + , where > 0. Then t = t1 c1 u1 where c1 2 + H (u) + |u|, 1 > 0 and |u1 | . Therefore tu = t1 c u1 , where c 5 + 2H (u) + 2|u| + C (u). Proof. d(a, [o, b]) (o, b)a = c(t1 , u ). In particular d(a, [o, b]) > . Therefore there are t2 [o, a] and x [a, b] such that d(t2 , [o, c]) and d(x, [o, c]) . So |t2 x| 2 and by the Lemma 22 x = u1 v where v H (u) + |u|. We have t2 = gv 1 u1 where g = t2 x, and hence t = t1 (gv 1 u1 ) = t 1 c1 u1 , where c1 |g | + |v | 2 + H (u) + |u|. Now |u1 | |t2 | |gv 1 | |t2 | |g | |v | |t2 | (2 + H (u) + |u|) 2 + H (u) + |u| + (2 + H (u) + |u|) = . 96
1 1 1 1 1 1 1 1 1

t
t1 o

t2

x v

b c

97

Figure 7: Lemma 5

Now by Lemma 21 c(u , u ) C (u) for 0. We have 2c(t1 , u ) = |t1 c1 u1 | + |u | |t1 u1 | |t1 | + |u1 | 2c1 + |u1 u1 | |t1 u1 | |t1 | + |u1 | 2c1 + |u1 | + |u1 | 2C (u) |t1 u1 | = = 2c(t1 1 , u1 ) + 2|u1 | 2c1 2C (u) 2c(t1 1 , u1 ) + 2 2c1 2C (u). Therefore c(t1 1 , u1 ) 3 + H (u) + |u| + c1 + C (u) 5 + 2H (u) + 2|u| + C (u). Denition 35 A word W is termed reduced provided ||W || = |W |. A word W is called cyclically reduced if for any word V the conjugacy of V and W in G implies that ||W || ||V ||. Any subword of a power of some word B 1 is termed a B-periodic word. Lemma 29 [Ol1] There is a constant 0 = 0 (G) such that if W is a B-periodic word, where B is cyclically reduced word, then |W | > |W1 | + |W2 | 0 for every factorization W W1 W2 . 98

When element W is long enough the value of constant 0 is easy to nd. Lemma 30 Let W be a cyclically reduced word, such that |W | 2( +1) then c(W 1 , W ) < + 1. Proof. Since |W | 2( +1) there is a decomposition W W1 W2 W3 such that |W1 | = |W2 | = + 1. Therefore if c(W 1 , W ) + 1 then |W3 W1 | . Hence |W3 W W3 1 | = |W3 W1 W2 | |W2 | + < |W |. Contradiction. Let c( ) = 1 . 2 0 Notice that if u is a cyclically reduced element and H ( u) is a constant which appears in lemma 6, then it follows from theorem 2 (see section 1) that H ( u) = 1 (G) does not depend on element u . Lemma 31 Let t, u G where u is a cyclically reduced element and c(t1 , u ) = 3 + 1 + , where > 0 and 1 (G) = H ( u). Then t = t 1 +2 u1 1 u 1 and tu = t 5+21 +c() u1 1 u 1 , where |u1 1 u 1 | and u u 1 u2 . Proof. We may assume that > 0. As in the proof of Lemma 28 we consider geodesic triangle [o, a, b]. We have d(a, [o, b]) (o, b)a = c(t1 , u ) > . Therefore there 99

are t2 [o, a] and x [a, b] such that d(t2 , [o, c]) and d(x, [o, c]) . So t2 x = g and |t2 x| = |g | 2 . Path x = u 1 u1 v where |v | 1 and u u1 u2 . So t2 = gv 1 u1 1 u 1 and t = t1 (gv 1 u1 1 u 1 ) = t 1 +2 u1 1 u 1 . We have |u1 1 u 1 | |t2 | |gv 1 | c(t1 , u ) 2 1 = . Note that u = u 1 u 1 u 1 1 u 1 and |u | > |u 1 u1 | + |u1 1 u 1 | 0 . Now 2c(t1 , u ) = |t1 gv 1 1 +2 (u1 1 u 1 )| + |u | |t1 gv 1 u1 1 u 1 | | |t1 gv 1 u1 1 u 1 | 21 4 > |t1 gv 1 | + |u1 1 u 1 | + |u > |t1 gv 1 | + |u1 1 u 1 | + |u1 1 u 1 | + |u 1 u1 | 0 |t1 gv 1 u1 1 u 1 | 21 4 = = 2c (t1 gv 1 ) , u1 1 u 1 + 2|u1 1 u 1 | 0 21 4 2c (t1 gv 1 ) , u1 1 u 1 + 2 0 21 4. Therefore we have c((t1 gv 1 )1 , u1 1 u 1 ) < 3 + 1 + + c( ) + 1 + 2 = 5 + c( ) + 21 . Lets make the following notations: 1 c1 (u) = (13|u| + 68 ), c 1 (u) = 4|u| + 32, 2 c 1 (u) = 31 |u| + 98 , c2 (u) = 3 + H (u) + |u|, 2
1 1

c 2 (u) = 2 + H (u) + |u|, c 2 (u) = 5 + 2H (u) + 2|u| + C (u). 100

Lemma 32 Let t, u, v G such that [u, v ] = 1 and c(t, ut) > c1 (u) + c 1 (u) + C (u, v ) + . Then c(t, vt) < c1 (v ) + c 1 (v ) + C (u, v ) + .
Proof. By Lemma 27 t = u c 1 (u) + C (u, v ) + . If c(t, vt) c1 (v ) 1 (u) t , where |u | > c then we are done. Suppose now that c(t, vt) = c1 (v ) + , where > 0, then t = v c 1 (v ) t

and |v | > . By Lemma 26 (i) v t = v ( u t ) = v C (u,v)+ t. On the other hand 2c(v , t) = |v | + |t| |t | |v | |t | + |v | + |t | 2 c1 (v ) = = 2|v | 2 c1 (v ) > 2 2 c1 (v ). Therefore < c 1 (v ) + C (u, v ) + as desired. Lemma 33 Let c(t, ut) c, then c(u , t) c2 (u) + c + 2 c2 (u) + |u| for any number , where C (u) =< u >. Proof. If c(u , t) c2 (u), then we are done. Let c(u , t) = c2 (u) + b, b > 0. Then by
1 Lemma 28 we have t = u1 c b. So t1 ut = t1 1 ut1 . We have 2 (u) t1 and u

1 c(t, ut) = (|t| + |ut| |t1 ut|) = 2 101

1 1 1 c = (|u1 c 2 (u) t1 | + |u(u 2 (u) t1 )| |t1 ut1 |) 2 1 1 (|u1 | + |t1 | 2 c2 (u) + |u1 c 2 (u) t1 | |u| |t1 ut1 |) 2 1 (2|u1 | + 2|t1 | 4|c 2 (u)| |u| |t1 1 ut1 |) 2 1 (2|u1 | + 2|t1 | 4|c 2 (u)| |u| 2|t1 | |u|). 2 So we get c |u1 | 2 c2 (u) |u| and b c + 2 c2 (u) + |u|. Therefore c(u , t) c2 (u) + c + 2 c2 (u) + |u| as claimed. Let c3 (u) = c2 ( u) + c1 (u) + 2 c2 ( u) + |u|. Lemma 34 (i) Let c(t1 , ut1 ) c, then c(t1 , uk t1 ) < c1 (uk ) + c + |u| + 2 c1 (uk ), where C (u) =< u >. (ii) Let c(t1 , ut1 ) c1 (u), then c(t1 , u k t1 ) c4 (u), where c4 (u) = c1 2c3 (u) + 2 + c1 (u) + 2 c1 2c3 (u) + 2 + |u|. Notice that c4 (u) does not depend on k . Proof. (i) Suppose that c(t1 , uk t1 ) = c1 (uk ) + (otherwise lemma is obvious),
where > 0. Then by Lemma 27 t = t c 1 (uk ) u , where |u | > . We have now

c |t| + |ut1 | |t ut | 2(|t | + |u | 2 c1 (uk )) |u| |t ut | 2|u | 4 c1 (uk ) 2|u| > 2 (4 c1 (uk ) + 2|u|). 102

Therefore < c + 2 c1 (uk ) + |u| which proves (i). (ii) Since c(t1 , ut1 ) c1 (u) by the previous lemma c(t 1 , u ) c3 (u ) for any . Suppose rst that |uk | > 2c3 (u ) + . Then by Lemma 26 t c3 (u )+ (uk t 1 ). which clearly proves (ii). Let |uk | 2c3 (u ) + . By the part (i) of this Lemma we get c(t1 , uk t1 ) c4 (u). Let d = 5 + 0 + 21 , d = 2d + , d 1 = c3 (d ), c1 (2d ) + d , e = c 1 (d ), e1 = c4 (d ). d1 = c1 (2d ) + c1 (d ) + d + 2 Lemma 35 Let t G, then
1 1 d t (i) t = t1 d t 1 = t1 e (tt1 ), where t is some cyclically reduced element from G.

)d1 (t t1 ) and if c(t1 , t) = d1 + where > 0, then |t1 | . (ii) t t = (t1 t

1 1 n t n (iii) tn = t1 e1 (t 1 ) = t1 d 1 t d 1 t1 for any n.

Also 1 1 1 t |t 1 | > |t| d and |t1 t| > |t| d . 2 2 , Proof. First, note that element t is conjugated to some cyclically reduced element t t1 1 . so t = t1 t | > 2d + . Case (i) |t 103

 ) = 3 + 1 + where > 0. By Lemma Suppose that for some > 0 we have c(t1 1 , t 31 we have
1 1 1 1 1 1 +1 ) = t 1 +2 (t t1 = t 1 +2 t 1 t2 c( ) t1 t 1 t 1 1 1 t 1 t 2 . Therefore by Lemma 26 (ii) t 2+c()+1 t where t . 1 t2 or equivalently t2 t1 2 +c( )+1 t 1 1 = t 5+c()+21 t 2 t 1 . We have now Also by Lemma 31 we have t1 t , so t 5+0 +21 t 1 t 1 1 2 t 1 t 1 = t 5+0 +21 t 2 t 1 2+c()+1 t 1 t1 1 = t t 1 t 1 = t t t1 t t t1 t 1 t

2 t 1 is also cyclically reduced. Similarly we consider the case when c(t1 1 , t ) = and note that t 3 + 1 + where > 0 and < 0. ) 3 +1 for any . Then t1 3+c()+1 t and t1 3+c()+1 t 1 . Suppose now, that c(t1 1 , t So we have
1 3+c()+1 t t1 1 = t1 3+c()+1 t t1 t 1 .

d t1 1 and Summarizing all of the above we see that there is t1 G such that t = t1 d t )d+ t1 1 . t1 1 ) = (t1 t by Lemma 26 (i) t = t1 d+ (t And c() t d t1 1 = t1 d t c()+ (t t1 1 ) = (t1 t )c()+2 (t t1 1 ). t t = t1 d t Also if c(t1 , t) = d1 + then
1 2 | + |t t 2c(t1 , t) = 2|t| |t2 | 2|t1 | + |t1 t 1 | |t | < 2|t1 | + 2d1

104

and |t1 | > . n d t1 1 for any n and since |t | > 2d + > 2 + 2 by Lemma 30 t +1 t and Now t1 d t n | > 2d + . Therefore tn = t1 d+ (t n t1 1 ). by Lemma 25 |t | 2d + . Case (ii) |t t1 1 = t1 c (d ) (t t1 1 ) = (t1 t )c (d ) t1 1 = t1 d t d t1 1 for some Using Lemma 27 t = t1 t 1 1 t1 G. )d1 (t t1 1 ). Therefore And by Lemma 34 (i) t t = (t1 t | + |t t1 | |t2 | 2|t1 | + 2d1 2c(t1 , t) = 2|t| |t2 | 2|t1 | + |t1 t and |t1 | .
1 n t1 1 ) = t1 d 1 t n d 1 t Now by Lemma 33 and Lemma 34 (ii) tn = t1 e1 (t 1 for any n, where

e1 = c4 (d ) and d 1 = c3 (d ). For both cases above t1 1 | |t t1 1 | + 2d |t | + |t t1 1 | < 2|t t1 1 | + 2d d t1 1 | |t1 | + |t |t| = |t1 d t t1 1 | > 1 |t| d . Similarly |t1 t | > 1 |t| d . and hence |t 2 2

Corollary 1 (i) Let t be any non-trivial element from G then |t2 | > |t| 4d 2d1 . 105

1 In particular c(t1 , t) < 2 |t| + 2d + d1 .

(ii) |tn | > |t| 2 e1 + d 1 + c( ) for any number n. Proof. (i) From the previous lemma it follows that | 4d |t2 | |t1 t | + |t t1 | |t2 | 2d1 . 2|t1 | + 2|t Hence | 4d 2d1 |t| + |t | 4d 2d1 > |t| 4d 2d1 . |t2 | 2|t1 | + 2|t Therefore | + 4d + 2d1 = |t| |t | + 4d + 2d1 < |t| + 4d + 2d1 2c(t1 , t) = 2|t| |t2 | 2|u| |t| |t as desired. (ii)
1 n t n n |tn | = |t1 t 1 | |t1 t | + |t1 | 2e1 |t1 | + |t | + |t1 | 2(e1 + d 1 ) >

| + |t1 | 2(e1 + d 1 + c( )) |t| 2(e1 + d 1 + c( )). > |t1 | + |t Lemma 36 Let t, u G and c(t1 , ut) = d1 + |u| + where > 0. Then d t1 1 ) = u1 (t1 e (t t1 1 )) t = u1 (t1 d t and )d1 (t t1 1 )) where|t1 | . t u t = u1 ((t1 t 106

Proof. Since c(t1 , ut) = d1 + |u| + we have c((ut)1 , ut) d1 + |u| + |u| = d1 + . So d t1 1 where |t1 | . by the Lemma 35 ut = t1 d t And we get t1 1 )) d t1 1 ) = u1 (t1 e (t t = u1 (t1 d t )d1 (t t1 1 )). and t u t = u1 (ut ut) = u1 ((t1 t Recall [MR3], that a length function l : G Z is regular if for any elements u, v G we have u = u1 w, v = w1 v2 , u v = u1 v2 . Here w = u v w = uv & l(w) = l(u) + l(v ) Clearly, the natural length function on a non-abelian free group is regular. The theorem below is one of the main technical tools we use in this paper. It shows that a word length function d(1, g ) = |g | on a hyperbolic group G is almost regular. Theorem 29 Let u, v G and suppose that |u| c(u1 , v ) > c1 + 2 and |v | c(u1 , v ) > c2 + 2. Then u = u1 u2 and v = v1 v2 where c(u1 , v ) |u2 | = |v1 | c(u1 , v ) + 2, |u1 | > c1 , |v2 | > c2 . 107

u
u1 o

u2

v 1 v1

b c

Figure 8: Theorem 1 And uv = u1 3 (u2 v1 v2 ), where |u2 v1 | 4 . Proof. If c(u1 , v ) then u v and we take u2 = v1 = 1. Let now c(u1 , v ) > . Then |ac| c(u1 , v ) > . Then there exist c [o, b], u2 [o, a] and v 1 [a, b] such that d(c, u2 ) and d(c, v 1 ) . Let |u2 | > |v 1 |, then v1 = v 1 v 1 where |v1 | = |u2 |. Therefore |v 1 | = |u2 | |v 1 | 2 . So |u2 v1 | 4 and |u2 | = |v1 | |ac| c(u1 , v ) . Therefore we have uv = u1 (u2 v1 v2 ), where u = u1 u2 and v = v1 v2 . Notice that 2c(u1 , v ) = |u| + |v | |uv | 2c(u1 , v ) 2 + |u1 | + |v2 | |uv |, so |u1 | + |v2 | |uv | 2 and |u1 | + |u2 v1 v2 | |uv | 6 . Therefore c(u1 1 , u2 v1 v2 ) 3 . Suppose that |u2 | = |v1 | > c(u1 , v ) + 2 . Then 2c(u1 , v ) = |u| + |v | |uv | > |u1 | + |v2 | + 2c(u1 , v ) + 4 |u1 | |v2 | |u2 v1 | 108

 2c(u1 , v ) + 4 4, and we get a contradiction. We have now c(u1 , v ) |u2 | = |v1 | c(u1 , v ) + 2 . By assumptions |u| c(u1 , v ) > c1 + 2 , so |u1 | > c(u1 , v ) |u2 | + c1 + 2 c(u1 , v ) c(u1 , v ) 2 + c1 + 2 = c1 therefore |u1 | > c1 . Similarly |v2 | > c2 . Theorem 30 Let v1 , . . . , vn G such that 1 (|vi1 vi vi+1 | |vi1 | |vi+1 | + |vi |) > y f or every i {2, . . . , n 1}. 2 Then v1 . . .vn = v 1 3 . . . 3 v n2 3 v n1,1 3 (svn,2 ), where |v 2 |, . . . , |v n2 |, |v n1,1 | > y 8 and |s| 4 , vn = vn,1 vn,2 , vn1 = vn1,1 vn1,2 , s vn1,2 = v n1,1 v n1,2 , |s | 4 , |v n1,2 | = |vn,1 |, v n1,2 vn,1 = s. Proof. We use induction on n. Vn = v1 . . .vn = Vn1 vn = v 1 3 . . . 3 v n3 3 v n2,1 3 (s1 vn1,2 ) vn , where |v 2 |, . . . , |v n3 |, |v n2,1 | > y 8 , and |s1 | 4, vn1 = vn1,1 vn1,2 , vn2 = vn2,1 vn2,2 , s 1 vn2,2 = v n2,1 v n2,2 , |s 1 | 4, |v n2,2 | = |vn1,1 |, v n2,2 vn1,1 = s1 . 109

We have 2y < |vn2 vn1 vn | |vn2 | |vn | + |vn1 | |v n2,1 s1 vn1,2 vn | + |vn2,1 |+ +4 |vn2,1 | |s 1 (v n2,1 v n2,2 )| |vn | + |vn1,1 | + |vn1,2 | |v n2,1 s1 vn1,2 vn | + 4 |v n2,1 | |v n2,2 | + 4 |vn | + |vn1,1 |+ +|s1 vn1,2 | + 4 = |v n2,1 s1 vn1,2 vn | + 12 |v n2,1 | |vn | + |s1 vn1,2 |. So we get |v n2,1 s1 vn1,2 vn | |v n2,1 | |vn | + |s1 vn1,2 | > 2y 12 . Now, by the previous theorem (s1 vn1,2 )vn = v n1,1 3 (svn,2 ) where s1 vn1,2 = v n1,1 v n1,2 , and vn = vn,1 vn,2 , |s| 4 , |v n1,2 | = |vn,1 |. Therefore v n2,1 3 (s1 vn1,2 ) vn = v n2,1 3 v n1,1 3 (svn,2 ). Then 2y 12 < |v n2,1 s1 vn1,2 vn | |v n2,1 | |vn | + |s1 vn1,2 | |v n2,1 | + |v n1,1 | + |svn,2 | |v n2,1 | |vn,1 | |vn,2 | + |v n1,1 | + |v n1,2 | 2|v n1,1 | + 4, and hence |v n1,1 | > y 8 . Finally Vn = v 1 3 . . . 3 v n3 3 v n2,1 3 v n1,1 3 (svn,2 ) = 110
1

= v 1 3 . . . 3 v n2 3 v n1,1 3 (svn,2 ).

6.3

Description of solutions

Lets consider an arbitrary one-variable equation in G.

x1 u1 x2 u2 . . . xn un = 1 where u1 , . . . , un G and if ui = 1 for some index i then i = i+1 .

(4)

The following technical result says that a given one-variable equation S (x) = 1 in a torsion-free hyperbolic group G has a subword of length three in which the middle occurrence of x cancels out almost completely. Theorem 31 Let x1 u1 x2 u2 . . . xn un = 1, then there exists an index i such that 1 i1 (|x ui1 xi ui xi+1 | |xi1 ui1 | |ui xi+1 | + |xi |) 24 + 4l , 2 where l = max {|ui |}.
i=1,...,n

Proof. Let vi = xi ui . Notice that |vj | 2l |vi | |vj | + 2l for any i, j {1, . . . , n}. Now x1 u1 x2 u2 . . . xn un = v1 . . .vn .
1 Suppose that 2 (|vi1 vi vi+1 | |vi1 | |vi+1 | + |vi |) > 24 + 2l for every i {2, . . . , n 1}.

Then by theorem 30 v1 . . .vn = v 1 3 . . . 3 v n2 3 v n1,1 3 (svn,2 ), 111

where |v 2 |, . . . , |v n2 |, |v n1,1 | > 24 + 2l 8 = 16 + 2l and |s| 4 , vn = vn,1 vn,2 , vn1 = vn1,1 vn1,2 , s vn1,2 = v n1,1 v n1,2 , |s | 4 , |v n1,2 | = |vn,1 |, v n1,2 vn,1 = s. Since |vn,1 | + |vn,2 | = |vn | |vn1 | 2l |vn1,2 | 2l |s vn1,2 | 4 2l |v n1,1 | + |v n1,2 | 4 2l we have |svn,2 | |vn,2 | 4 |v n1,1 | 8 2l > 8. Similarly from the proof of the previous theorem it follows that |v 1 | |v 2 | 2l 4 and therefore |v 1 | 16 + 2l 2l 4 = 12 . Therefore the length of each element in the product v 1 3 . . . 3 v n2 3 v n1,1 3 (svn,2 ) is greater than 8 , and hence by Lemma 25 this product is not trivial. Contradiction. So there exists index i such that 1 (|vi1 vi vi+1 | |vi1 | |vi+1 | + |vi | 24 + 2l. 2 We have 48 +4l |xi1 ui1 xi ui xi+1 ui+1 ||xi1 ui1 ||xi+1 ui+1 |+|xi ui | |xi1 ui1 xi ui xi+1 | |ui+1 | |xi1 ui1 | |ui xi+1 | |ui | |ui+1 | + |xi | |ui |. 112

So |xi1 ui1 xi ui xi+1 | |xi1 ui1 | |ui xi+1 | + |xi | 48 + 8l. Dividing both sides by 2 we prove the theorem.

Now we turn to a description of the solutions of Equation 4. We need the following technical Lemma 37 Let u, v, w G such that 1 (|uvw| |u| |w| + |v |) y. 2 (i) Suppose that uvw = (u c1 v )c2 w where v = (v1 d1 v )d2 v2 , u = u u1 , w = w1 w and |u1 | = |v1 |, |v2 | = |w1 |. Then |v | y + c1 + c2 + d1 + d2 . (ii) Suppose that uvw = u c1 v c2 w where v = (v1 d1 v )d2 v2 , u = u u1 , w = w1 w and |u1 | = |v1 |, |v2 | = |w1 |. Then |v | y + c1 + c2 + + d1 + d2 . Proof. (i) We have 2y |uvw| |u| |w| + |v | |u v | + |w | 2c2 |u| |w| + |v | |u | + |v | 2c1 2c2 + |w | |u| |w| + |v | |u | + |v | + |w | 2(c1 + c2 ) |u | |u1 | |w1 | |w | + |v1 | + |v | + |v2 | 2(d1 + d2 ) = = 2|v | 2(c1 + c2 + d1 + d2 ). 113

So |v | y + c1 + c2 + d1 + d2 as desired. (ii) Suppose that |v | > y + c1 + c2 + 2 + d1 + d2 > c1 + c2 + 2. Then using Lemma 26 (i) we have uvw = (u c1 v )c2 + w and by the rst part |v | y + c1 + c2 + + d1 + d2 . Contradiction. Lets x the following notations 1 31 c1 = c1 (l) = (13l + 68 ), c 1 = c 1 (l) = 4l + 32, c 1 = c 1 (l) = l + 98, 2 2 where l = max {|ui |}. Also
i=1,...,n 1 c2 = max{ max c2 (ui ), max c2 (u i1 ui ui1 )}, i=1,...,n i=2,...,n 1 c 2 = max{ max c 2 (ui ), max c 2 (u i1 ui ui1 )}, i=1,...,n i=2,...,n 1 c 2 = max{ max c 2 (ui ), max c 2 (u i1 ui ui1 )}, i=1,...,n i=2,...,n

and c3 = c2 + c 1 + 2 c2 + l . Let also C (g ) for any g G be a constant which appears in Lemma 21 then we make the following notation
1 C = max C (u i1 ui ui1 ). i=2,...,n

Now we turn to a description of a solution x by considering all dierent forms of the triple xi1 ui1 xi ui xi+1 . 114

Lemma 38 Let 1 (|xui1 x1 ui x| |xui1 | |ui x| + |x|) 24 + 4l, 2 for some index i then x has one of the following forms: 1) |x| < C1 ;
2) x = u 1 x where |x | < C1 ; i c 3) x = x c 1 ui1 where |x | < C1 ; 4) x = (u 1 x )c 1 ui1 where |x | < C1 ; i c

for some constant C1 (, l). Proof. Lets consider the following product xui1 x1 ui x. Let c(x1 , ui1 x1 ) c1 and c(x, ui x) c1 . So we have xc1 (ui1 x1 )c1 +l ui x. Applying Lemma 37 (ii) we have |ui1 x1 | 24 + 4l + c1 + c1 + l + + l and therefore |x| 25 + 7l + 2c1 . 115

Suppose now that c(x1 , ui1 x1 ) c1 and c(x, ui x) = c1 + where > 0. By Lemma
27 x = u 1 x . So i c xc1 (ui1 x1 )ui x = (xui1 )c1 + c 1 (ui x ). c1 +2l x 1

By Lemma 37 (ii) we have |x | 24 + 4l + c1 + c 1 + 2l + c 1 + + c 1 .

Suppose that c(x1 , ui1 x1 ) = c1 + and > 0 then by Lemma 27 x = x c 1 ui1 1 where u = x c1 (ui1 x 1 ). i1 and xui1 x

We have now
xui1 x1 ui x = x c1 (ui1 x )ui (x c 1 ui1 ). 1

Let c(x , ui x ) c1 , then

xui1 x1 ui x = (x ui1 )c1 +l x c1 (ui x )c 1 +l ui1 . 1

Suppose that |x | > c1 + c 1 + l + + l then by Lemma 26 (i) xui1 x1 ui x = (x ui1 )c 1 +l x c1 + (ui x u i1 ). And by Lemma 37 1 . |x | + 4l + c 1 + l + c1 + + + c
Hence x = x c 1 . 1 ui1 where |x | 26 + 5l + c1 + c1 + c 1

116

 Let c(x , ui x ) = c1 + 1 and 1 > 0 then by Lemma 27 x = u 1 x where |ui | 1 i c

and x 1 ui x = x 1 c1 (ui x ). Now x has the following form

x = (u 1 x )c 1 ui1 = ui c 1 x 2 c1 ui1 . i c

and
c1 (ui x )2 xui1 x1 ui x = (x ui1 )c1 + c1 +l ui1 . c1 +2l x 1

Suppose that c1 + l + + l. |x | > c1 + 2 Then

xui1 x1 ui x = (x ui1 )c1 + c1 + (ui x u c1 +2l x i1 ). 1

And hence by Lemma 37 c1 . |x | 26 + 6l + 2c1 + 3 Lemma 39 Let 1 (|xui1 xui x1 | |xui1 | |ui x1 | + |x|) 24 + 4l, 2 for some index i then x has one of the following forms: 1) |x| < C2 ; 117

 2) x = x c 1 ui , where |x | < C2 ; 1 3) x = (u 2 x )c 1 ui , where |x | < C2 and ( ) 0; i1 ui ui1 c 1 4) x = (u 1 ui where |x | < C2 ; i1 ui ui1 C + c2 x )c

for some constant C2 (, l). Proof. Consider the following product xui1 xui x1 . Let c(x1 , ui x1 ) c1 . We have then
1 )d1 ( xx1 )) ui x1 = xui1 xui x1 = xui1 xc1 ui x1 = u i1 ((x1 x 1 = u )d1 +l ( xx1 )c1 +e+l ui x1 . i1 (x1 x 1 1

By Lemma 37 |x x 1 | 24 + 4l + d1 + l + c1 + e + l + + e + l = A, and |x| < 2|x x 1 | + 2(d + l) 2A + 2(d + l).

Suppose that c(x1 , ui x1 ) = c1 + 1 where 1 > 0. We have then x = x c 1 ui and 1 1

therefore
xui1 xui x1 = xui1 (x c1 ui x ) = x c . 1 ui ui1 x ui x 1 1

118

1 Let c(u i1 ui ui1 , x ) c2 . Then xui1 xui x1 = x c u x . 1 +l ui ui1 c2 +l x c 1 i If |u 1 + l + c2 + l + + l = B then applying the same arguments as in the previous i | c 1

case we get |x | 2(25 + 7l + d1 + 3B + c1 + 2e + c 1 ) + 2(d + B + l).

Let |u i | > B then by Lemma 26 (i) . xui1 xui x1 = (x u i ui1 )c2 +l+ x c 1 ui x 1

Therefore by Lemma 37 1 . |x | 24 + 4l + c2 + l + + c 1 + + c
1 Suppose that c(u i1 ui ui1 , x ) = c2 + 2 where 2 > 0, then by Lemma 28

1 x = u 2 x where > 0. i1 ui ui1 c

Notice that since x c1 ui x 1 by Lemma 33 x c3 u i for any . We have now

1 x = x c 1 ui = (ui1 ui ui1 c 2 x )c 1 ui ,

and
ui1 c2 +l x c u x . xui1 xui x1 = x c3 +l u 2 +c i 1 i 1

119

Suppose that
|u ui1 | > c3 + l + c2 + l + = C0 . i

Then by Lemma 26 (i)

ui1 )c2 +l+ x c u x . xui1 xui x1 = (x u 2 +c i 1 i 1

And applying Lemma 37 we have 2 + c 1 = C. 2 + c1 + + c |x | 24 + 4l + c2 + l + + c We have now

1 x = (u 2 x )c 1 ui , i1 ui ui1 c

where |x | C . Suppose that ( ) < 0. We have

1 1 x = (u ui1 )c 2 x c 1 ui = i1 ui ui1 C ui1 ui 1 = (u ui1 x ) c 1 ui . i1 ui C +l+ c2 (ui

Then x c1 ui x Therefore
xui1 xui x1 = x c3 +l+C u . ui1 x c +C +2l+ i c 2 ui x
1

1 1 = u ui1 x ) c1 ui x i1 ui ui1 C +l+ c2 (ui1 ui

120

Therefore by Lemma 37 (ii) we have

|u ui1 x | i

24 + 4l + c3 + l + C + c1 + C + 2l + c 2 + + C + l + c 2 + c 1 = D. Hence element x has the following form

1 x = (u 1 ui i1 ui ui1 C + c2 x )c

where |x | D + l.
ui1 | C0 . We have Lets consider the case when |u i 1 ). xui1 xui x1 = (u ui1 (x c 2 x ) ui 2 +c 1 ui x i1 ui ui1 c Let c(x 1 , u ui1 x ) d1 + C0 . Then i 1 . xui1 xui x1 = u ui1 x c 2 x d1 +C0 ui 2 +c 1 +C0 ui x i1 ui ui1 c 1 1

Suppose that |x | > c 2 + d1 + C0 + . Then

1 . xui1 xui x1 = (u ui1 x c 2 +c 1 +C0 ui x i1 ui ui1 x )d1 +C0 + ui 1

And
1 x = (u ui1 x )c 2 +C0 ui 1 ui i1 ui c

using Lemma 37 we have

ui1 x | 26 + 4l + d1 + 3C0 + 2 c2 + c 1 + c1 = E |u i

121

and element x has the following form

1 x = (u 2 +C0 +l x )c 1 ui i1 ui ui1 c

where |x | E .
Suppose that c(x 1 , u ui1 x ) = d1 + C0 + 3 where 3 > 0. Then by Lemma 36 i

1 x u ui1 x = u ((x 1 x )d1 ( x x 1 )), i i1 ui ui1 )1 (x 1 e ( x x 1 )). x = (u ui1 )1 (x 1 d x d x 1 ) = (u i i 1 1

Therefore we have
1 u ui1 )1 (x 1 x ) 2 +e+C0 (ui i1 ui ui1 c 1 1 1 and since x c we have x x . Suppose that 2 +c 1 ui x c 2 + c 1 ui x 1 e+C0 +

2 + e + C0 + d1 + C0 + + C0 |x x 1 | > c then applying Lemma 26 we get xui1 xui x1 =

1 = u ui1 )1 (x 1 x )d1 +C0 ( x x 1 )e+C0 + c 2 + c 1 ui x 2 +e+C0 (ui i1 ui ui1 c 1 = (u x 1 x )d1 +C0 + ( x x 1 )e+C0 + . c 2 + c 1 ui x i1 ui 1 1 1 1

And x has the following form

1 x = (u 2 x )c 1 ui = i1 ui ui1 c

122

 1 x 1 = u ui1 )1 x 1 C0 +e x 2 (ui i1 ui ui1 c

c 1 ui = 1

1 = (u x x 1 ) c ui1 )1 x 1 )c 1 ui 2 +e+C0 (ui 2 +C0 +e ( i1 ui ui1 c

where the last equality follows from Lemma 26 (ii). Now applying Lemma 37 we have 1 = F0 |x x 1 | 26 + 4l + d1 + 3C0 + 2e + 2 c2 + c1 + c and |x | < 2|x x 1 | + 2(d + C0 ) 2F0 + 2(d + C0 ) = F. This shows that element x has the following form
1 1 1 ui1 x )c x = (u 2 +C0 +l x )c 1 ui 1 ui = (ui1 ui ui1 c i1 ui ui1 ui1 ui 1 1

where |x | |x | + C0 + l F + C0 + l. Lemma 40 Let 1 (|xui1 xui x| |xui1 | |ui x| + |x|) 24 + 4l, 2 for some index i then x has one of the following forms: 1) |x| < C3 ; 123

 1 ) , where |x | < C3 ; 2) x = u c1 (x uj i1 uj e+

1 1 1 3) x = u ui1 ) , where |x | < C3 ; c1 (x ui1 uj i1 ui1 uj ui1 e+3

1 1 1 4) x = u c1 (x uj ui1 i1 uj ui1 e1 +2

) , where |x | < C3 ;

1 (5) x = u c1 (x ui1 uj i1 ui1 uj e1 +2

) , where |x | < C3 ;

for some constant C3 (, l) and index j . Proof. Consider the product xui1 xui x, and suppose that ui1 = ui . Let c(x1 , ui1 x) > d1 + l. Then
1 1 x = u d x1 1 ) = u )e x1 1 ), i1 (x1 d x i1 ((x1 x 1 1 and xui1 x = u )d1 ( xx1 1 ). Let c(x1 , ui u i1 (x1 x i1 x1 ) > 2c1 > c1 (2l ). Then x1 = 1 1 1 1 1 ui u c1 x 1 and x1 ui ui1 x1 = x 1 2c 1 (ui ui1 x 1 ). Put i1 2

A = 27 + 6l + d1 + 2d + 6 c1 + 2e and suppose that c1 + 4l + e > d + 2 c1 + 2c 1 + . c1 ) > 2c 1 + 2 + 2 |x 1 | > A + 2(d + 2 Applying Lemma 26

1 x 1 = x d +2 c1 +2l x c1 x 1 2c 1 (ui ui1 x 1 )e+2 1

124

1 = ( xx 1 )2c 1 + (ui u x 1 = c1 +2l x i1 x 1 )e+2 1 = ( xx 1 )2c 1 +2 (ui u x 1 ). i1 x 1 x 1

We have
1 1 xui1 xui x = u d1 +l+d +2 xx 1 )2c 1 +2 (ui u x 1 ). c1 ( i1 x1 x i1 x 1 x 1 1

And
1 1 x = (u x 1 c1 +d x i1 ui ui1 x 1 )e+l+2 1 Now by Lemma 37 (ii) |x x 1 | A and therefore 1 1 e+2 c1 ui ui1

|x 1 | |x x 1 | + 2(d + 2 c1 ) A + 2(d + 2 c1 ). The obtained contradiction shows that |x 1 | A + 2(d + 2 c1 )

1 1 Let x 1 ui ui1 x 1 = v , then

|v | 2(A + 2(d + 2 c1 )) + 2l = B. Let c( x1 , v x ) > d1 + B then

1 1 )e+B x )d1 +B (x x x = v 1 ( x1 x v x = v 1 ( x1 x 1 and x 1 ). 1 1 , where C = d1 + d + l + e + B . We Now (u )d1 +d +l x , therefore (u )C v 1 x 1 x i1 x1 x i1 x1 x

have
1 1 1 d1 +B x x )C v 1 x 1 x xui1 xui x = (u 1 d +e+B x1 . i1 x1 x

125

Suppose that | > 26 + 4l + 3C + d1 + B = D, |v 1 x 1 x

1 x then |x 1 | > d1 + B + d + e + B + . Under these assumptions applying Lemma 26 (i)

we have
1 1 1 d1 +B + (x x )C v 1 x 1 x xui1 xui x = (u 1 x1 ). i1 x1 x

Its not hard to show that

1 1 ))C ( x = (u 1 x x1 x1 1 ). i1 x1 C (v x

| D. Contradiction. Hence Therefore by Lemma 37 |v 1 x 1 x | + |x | + |v 1 x | + 2d + B 2D + 2d + B |x | |v 1 x 1 x 1 | |v 1 x 1 x 1 x and element x has the following form
1 1 1 x = u c1 +e (x ui ui1 i1 ui ui1 2

where |x | 2D + 2d + B + B 2l = E . Lets now consider the case when ui1 = ui . We have xui1 xui1 x = (ui1 x1 x )d1 +l+d x d1 +d x x 1 1 .
1 And x = (u )e+l x1 1 . Then by Lemma 37 i1 x1 d +l x

|x | 25 + 7l + 2d1 + 3d + e = F. 126

1 Lets substitute x = u x1 1 ) into the original equation 4 i1 (x1 x

x1 u1 x2 u2 . . . xn un = 1 and consider it as an equation on x1 . Then there is an index j such that in one of the following products middle element x1 1 concells out almost completely : (1) (x1 x )e x1 1 uj x1 or x1 1 uj x1 e ( x1 x1 1 ), where uj = 1;
1 1 1 xx1 1 ), where uj = 1; (2) (x1 x 1 )e x1 1 ui1 uj u i1 x1 or x1 ui1 uj ui1 x1 e ( 1 1 1 xx1 1 ), where ui1 = uj ; (3) (x1 x )e x1 1 uj u i1 x1 or x1 uj ui1 x1 e (

x1 x1 1 ) where ui1 = uj 1 ; (4) (x1 x 1 )e x1 1 ui1 uj x1 or x1 1 ui1 uj x1 e (

1 k )e1 x1 1 ui1 uj x1 , where ui1 = uj and ui1 = uj 1 (5) (x1 x k )e1 x1 1 uj u i1 x1 or (x1 x

respectively ; or
1 n x x1 u1 x2 u2 . . . xn un = x1 x 1 ui1 un = 1 1 1 n x ( x1 u1 x2 u2 . . . xn un = u 1 un = 1 ), i1 x1 x

in which case by Corollary 1 (ii) of Lemma 35 |x| < 2 e1 + d 1 + c( ) + l + l.

127

For cases (1)-(5) let L = n F . Then applying the same arguments as in Lemma 38 we get:
1 ) , where |x | H ; (1) |x| G or x = u c1 (x uj i1 uj e+ 1 1 1 ui1 ) , where |x | H ; (2) |x| G or x = u c1 (x ui1 uj i1 ui1 uj ui1 e+3 1 1 1 (3) |x| G or x = u c1 (x uj ui1 i1 uj ui1 e+2 1 (4) |x| G or x = u c1 (x ui1 uj i1 ui1 uj e+2

) , where |x | H ; ) , where |x | H ; ) or

1 1 1 (5) |x| G1 or x = u c1 (x uj ui1 i1 uj ui1 e1 +2

1 x = u c1 (x ui1 uj i1 ui1 uj e1 +2

) , where |x | H1 ;

where G = l + F + 2(25 + 4L + e + 3c1 ), H = F + 2(25 + 4L + e + 6c1 + 3c 1 ), and G1 = l + F + 2(25 + 4L + e1 + 3c1 ), H1 = F + 2(25 + 4L + e1 + 6c1 + 3c 1 ). The three lemmas above give the following description of solution sets of one-variable equations in torsion-free hyperbolic groups. Theorem 32 Let G be a torsion-free hyperbolic group and x is a solution of an arbitrary one-variable equation x1 u1 x2 u2 . . . xn un = 1

128

where u1 , . . . , un G and if ui = 1 for some index i then i = i+1 , and l = max {|ui |}.

i=1,...,n

Then there are constants C = C (, l) and D = D(, l) (which can be found eectively) such that one of the following holds: 1) |x| < C ;
1 2) x = (u i D x ) , where |x | < C ; 3) x = (u i x )D ui1 1

, where |x | < C ;
1

1 4) x = (u i1 ui ui1 x )D ui

, where |x | < C and ( ) 0; , where |x | < C ;

1

5) x = 6) x =

1 u ui x D uj i uj 1 u i 1 uj u i

D (x

1 uj u i

or
1

1 ui uj D (x ui uj ) x = u i

, where |x | < C ;

We can write the above theorem in the following compact form. Theorem 33 Let G be a torsion-free hyperbolic group and S (x) = x1 u1 x2 u2 . . . xn un = 1

129

be an arbitrary one-variable equation in G. Then there is a constant C (which can be found eectively) and there are nitely many elements f and g (which can be found eectively) such that an arbitrary solution x of S (x) = 1 has one of the following forms: 1) |x| < C ; 2) x = f g for some g G such that |g | < C ; 3) x = (f g )g where |g | < C ; 4) x = (f g )f g for some g G such that |g | < C ; where and are integers.

Resolution method

In this section we develop a method for nding parametric solutions of one-variable equations in torsion-free hyperbolic groups. In fact, the only two properties of hyperbolic groups we use here are BP (big powers) property and CSA (commutative-transitive) property. Let G be a CSA non-abelian BP-group (notice that in this case G is torsion-free). We need the following technical lemma.

130

Lemma 41 Let G be a commutative-transitive BP -group. Then for any tuple s Gk if the equality
n n sn 1 s2 . . . sk = 1

holds for innitely positive integers n then there exists a segment si,j = (si , . . . , sj ) of length at least 2 (j i 1), such that () si . . . sj = 1, [sp , sq ] = 1 f or any p, q {i, . . . , j }. Proof. We can assume that all si -s are non-trivial. Since G satises a BP condition s has a commutation, say [sm , sm+1 ] = 1. Now
n n n sn 1 s2 . . . (sm sm+1 ) . . . sk = 1.

If sm sm+1 = 1 then we are done. Otherwise, by induction, there exists a segment t = (sl1 , . . . , sl2 ) of the sequence (s1 , . . . , sm sm+1 , . . . sk ) for which the condition () holds. If sm sm+1 is not in t then we are done. Otherwise, from transitivity of commutation [sp , sm sm+1 ] = 1 & [sm sm+1 , sm ] = 1 [sp , sm ] = [sp , sm+1 ] = 1 for any p {l1 , . . . , l2 }. This proves the lemma. Now we turn to a description of what we call a BP-resolution. Let us describe precisely an elementary step (we call it a BP-reduction) of this method. Consider an arbitrary onevariable equation in G

131

f (x) = x1 u1 x2 u2 . . . xn un = 1

(5)

where u1 , . . . , un G and if ui = 1 for some index i then i = i+1 . An equation f (x) = 1 is called homogeneous if f (1) = 1, i.e., 1 VG (f ). Observe, that the equation 5 is homogeneous if and only if u1 . . . un = 1. Suppose that C = {f k } VG (f (x)) for innitely many integers k and some 1 = f G. Rewrite f (x) = 1 in the form (x1 )(x2 u1 )(x3 u2 Since (f 1 )k (f 2 u1 )k (f 3 u2
1 1 1 u1 1 1 1 u1

) . . . (xn un1 un2 ...u1 )u1 u2 . . . un = 1.

)k . . . (f n un1 un2 ...u1 )k u1 u2 . . . un = 1

holds for innitely many powers of f it follows from lemma ?? that u1 u2 . . . un = 1, hence f (x) = 1 is homogeneous. Consider now the equation s1 . . . sn = 1, sl = xl ul1 ul2 ...u1 (u0 = 1).
1 1 1

By lemma 41 there exists a segment sij = (si , . . . , sj ) with j > i (we call it a cancellation segment ) such that si . . . sj = 1, [sp , sq ] = 1 p, q {i, . . . , j }. 132 (6)

Rewriting equalities 6 in terms of x and ui -s we see that [xp up1 ...u1 , xp+1 up
1 1 1 1 ...u 1

]=1

for each p = i, . . . j 1. After conjugation we have [xp , xp+1 up ] = 1 which implies, due to CSA property, the following relations [x, ui ] = 1, . . . , [x, uj 1 ] = 1, [up , uq ] = 1, p, q {i, . . . , j 1}. Observe, that at least one of the coecients ui , . . . , uj 1 is non-trivial, otherwise the equality si . . . sj = 1 would imply that the equation f (x) = 1 is not reduced. Therefore the system 6 is equivalent (over C = {f k } !) to the condition i + . . . + j = 0 and the following system of equations [x, ui ] = 1, . . . , [x, uj 1 ] = 1. (7)
1

The equations in 7 are called resolvents of the ij -reduction. Now we can replace the equation s1 . . . sn = 1 by a shorter equation (obtained by deleting sij from the initial one) s1 . . . si1 sj +1 . . . sn = 1. (8)

We call this equation the residual equation and denote it by fij = 1. These arguments prove the following lemma, which is one of the main technical tools of what follows in the sequel. 133

Lemma 42 (Resolution Lemma) Let G be a CSA BP-group and f (x) = 1 be a onevariable equation in G. If the solution set of f (x) = 1 contains a set C which consists of innitely many powers of some non-trivial element f , then there exists an elementary ij -reduction of f (x) = 1 such that f (x) = 1 is equivalent over C to the system consisting of fij = 1 and all ij -resolvents of f (x) = 1. One can continue the resolution process and apply a BP-reduction to the equation fij = 1, provided it is non-trivial (observe, that fij = 1 also contains C in its solution set). In nitely many steps (which depend on C ) we split f (x) = 1 into a system of resolvents RC = 1 which is equivalent to f (x) = 1 over C . Any such particular way of splitting f (x) = 1 is called a resolution of f (x) = 1. Notice, that the system RC = 1 is completely dened by the choices of pairs of indexes i, j on the elementary steps. So there are only nitely many possibilities to make a resolution of f (x) = 1, and one can eectively produce them without any knowledge of the set C. We write down the output of the discussion above in following proposition. Proposition 7 Let G be a non-abelian CSA BP-group. Then for any one-variable equation f (x) = 1 over G one can eectively nd a nite set R(f (x)) = {R1 = 1, . . . , Rn = 1} of systems of BP-resolvents such that if a non-trivial set C which consisits of innitely many powers of some element f belongs to the solution set of f (x) = 1 then RC R(f (x)). 134

Notice that if a non-trivial element f satises some resolution from R(f (x)), then C (f ) VG (f (x)), where C (f ) is a centralizer of element f . Corollary. Let G be a non-abelian CSA BP-group, and f (x) = 1 be a one-variable equation in G such that C VG (f (x)) where C consists of innitely many powers of some non-trivial element f . Then C (f ) VG (f (x)). Now we discuss in more details the resolution process and the systems RC . Observe rst, that a cancellation segment has to satisfy the following conditions: the sum of x-exponents along the segment is equal to 0 and the coecients ui , . . . , uj 1 pairwise commute (the latter holds since G is CSA). An ij -reduction applied to f (x) = 1 gives one resolvent of the type: [x, up ] = 1. Deleting a cancellation segment during an ij -reduction may produce a new cancellation segment in fij which was not a segment in the initial equation. For example, after an ij -reduction the pair (si1 , sj +1 ) may form a cancellation segment for fij . These new types of cancellation segments result in the equations of the type [x, ui1 . . . uj ] = 1, [ui1 . . . uj , ul1 . . . um ] = 1, for suitable indexes i, j, l, m. It follows from the discussion above that the system RC consists of commutator equations

135

of the type [x, up . . . uq ] = 1, say RC = {[x, up up+1 . . . uq ] = 1 | (p, q ) I N 2 } Since the group G is CSA this system is equivalent over G to the union of a single arbitrary equation from RC and the following one
= {[up up+1 . . . uq , ul ul+1 . . . um ] = 1 | (p, q ), (l, m) I }. RC

The single equation can be always choosen of the type [x, up ] = 1 for some coecient up of f (x) = 1. Indeed, equations of this type always occur on the rst step of a resolution.

7.1

Application of resolutions for nding solutions of one-variable equations in torsion-free hyperbolic groups

Now we describe an algorithm for nding all solutions of one-variable equation 5. Note that by theorem 32 if x is a solution of 5 then it takes one of four possible forms: 1.) x has a bounded length |x| < C ; 2.) x = f x , for some f and |x | < C ; 3.) x = (f x f x )1 , for some f, x , x G, where |x |, |x | < C. 4.) x = (f x g x )1 , for some f, g, x , x G, and |x |, |x | < C ; 136

for some constant C which depends on equation 5, and elements f and g which are described precisely. Now, depending on the form of element x, we give an algorithm which checks, whether x is a solution of 5. Case 1.) Let |x| < C for some constant C which appears in theorem 32. Since there are only nitely many elements of length less than C , in nitely many steps we will nd all such solutions. Case 2.) Let x = f x , for some f = 1 and |x | < C . Consider the following equation f (y ) = (yx )1 u1 (yx )2 u2 . . . (yx )n un = 1. Then y is a solution of f (y ) = 1 if and only if x = x y is a solution of an original equation f (x) = 1. Let R(f (y )) be a nite set of BP-resolvents of f (y ) = 1 then by proposition 7 f is a solution of f (y ) = 1 for every Z (or we just say one-parametric solution ) if and only if f satises at least one resolvent Rj R(f (y )). Therefore, since there are only nitely many elements of length less than C , we can eectively nd all one-parametric solutions of f (x) = 1 of the form x = f x . Suppose now that x = f x is not a solution of f (x) = 1 for every Z. Nevertheless, there may be solutions of the form x = f k x for some k Z. Since G is a torsion-free hyperbolic group, there exists a constant c(f ), which can be found eectively ([Ol1]), such

137

that if f (f l ) = 1 for any l > c(f ) then f VG (f (y )). So we just need to check all elements of the form x = f k x for k c(f ). Case 3.) Let x = f x f x , for some f, x , x G, where |x |, |x | < C . Consider the following equation f (y ) = (yx y 1 x )1 u1 (yx y 1 x )2 u2 . . . (yx y 1 x )n un = 1 then, clearly, f k x f k x VG (f (x)) f k VG (f (y )), for any k Z. And we apply resolution method to f (y ) = 1 to check whether f x f x is a one-parametric solution of f (x) = 1. Again, there might be solutions of the form f k x f k x for some k Z. As in the previous case, there exist a constant c(f ), which can be found eectively, such that if f (f l ) = 1 for some l > c(f ) then f VG (f (y )). And we just need to check all elements of the form x = f k x f k x for k c(f ). Case 4.) Let x = f x g x , where |x |, |x | < C . Let
fy (x) = (xx yx )1 u1 (xx yx )2 u2 . . . (xx yx )n un = 1.

Considering y as a parameter we can eectively construct a set of all BP-resolvents of

fy (x) = 1: f f (x)) = {R1 }, R(fy , ..., Rm

138

f where Ri = {[x, u li (y )] = 1 Ri (y ) = 1}, where u j (y ) are coecients of equation fy (x) = 1.

Let f x g x is a two-parametric solution of f (x) = 1 then for every k equation f gk (x) =

f 1 has f in its solutionset. Therefore for some index l(k ) there exist Rl (k) R(fy (x)) such f k k that Rl (k) = {[f, u s(k) (g )] = 1 Rl(k) (g ) = 1}. Therefore there exists a BP-resolvent f f ) for innitely many integers k . Hence by corollary (x, y ) such that (f, g k ) VG (Rj Rj f f such ). And vice versa, if there is a BP-resolvent Rj of proposition 7 (f, g ) VG (Rj f that (f, g ) VG (Rj ), then x = f x g x is a two-parametric solution of f (x) = 1. And

we use resolution method again to nd such BP-resolvent. Now suppose that f x g x is not a two-parametric solution of f (x) = 1, then there may exist integers k and l such that f k x g l x is a solution of f (x) = 1. Since G is a torsion-free hyperbolic group and satises a BP-condition, there exists a constant c(f, g ), which can be found eectively, such that if f q x g r x VG (f (x)) for some q, r > c(f, g ), then f x g x VG (f (x)) is a two-parametric solution of f (x) = 1. Therefore it is enough to check all elements of the form x = f k x g l x for k, l c(f, g ).

139

Irreducible algebraic sets and their coordinate groups

In this section, following paper [CR], we use ultrapowers for studying coordinate groups of irreducible algebraic sets.

8.1

Ultraproducts and coordinate groups of irreducible algebraic sets.

Let I be a set and let P (I ) be the Boolean algebra of all subsets of I . An ultralter over I is a subset D of P (I ) such that (i) A D and A B I implies B D (ii) A, B D implies A B D (iii) for all A P (I ), exactly one of A, I \ A belongs to D. Denition 36 Let {Xi | i I } be a family of sets indexed by I , and let D be an ultralter over I . The ultraproduct
iI

Xi /D is dened to be the quotient set

iI

Xi / , where

(xi )iI (yi )iI if and only if {i I | xi = yi } D. This is easily checked to be an equivalence relation. We shall denote the equivalence class of an element (xi )iI
iI

Xi by xi

iI ,

and usually abbreviate this to xi . If

all Xi = X , a single set, then the ultraproduct X I /D is called an ultrapower of X . Note 140

that X embeds via the diagonal mapping in X I /D, by mapping x to xi , where xi = x for all i I . Denoting the ultrapower X I /D by X , given a function f : X n Y of n variables, : (X )n Y by dening f (x(1) , . . . , x(n) ) = we obtain an extension to a function f f (xi , . . . , xi ) , where x1 = x1 i , etc. Thus if X is a group, then we can dene a multiplication in X by xi yi = xi yi , and this makes X into a group. If X is abelian, so is X I /D, and if X is an ordered abelian group, then X I /D can be made into an ordered abelian group by dening xi yi if and only if {i I | xi yi } D. This can be proved directly, or using Lo ss Theorem (see [??]). Now the diagonal embedding of X into X I /D is clearly structure-preserving (if X is a group, it is a group homomorphism). An ultralter D over I is called principal if some nite subset of I belongs to D. This is equivalent to saying there is an element j I such that j A for all A D. In this case, the canonical embedding X X I /D is surjective and so identies X and X I /D. On the other hand, if I is countable and D is non-principal, the canonical embedding is not surjective. In the introduction we mentioned a result of Chiswell and Remeslennikov about description of solutions sets of one-variable equations over free groups. Instead of working directly with solutions, they work only with the coordinate groups of irreducible algebraic
(1) (n)

141

sets. The following fundamental theorem on which this approach is based was proved in [BMR1]. Theorem 34 [BMR1] Let G be an equationally Noetherian group. Then for any countable innite set I and any non-principal ultralter D over I , the ultrapower GI /D has the following properties: 1. GI /D is a G group, where G is embedded in GI /D via the diagonal mapping; 2. Every coordinate group of an irreducible algebraic set over G is embeddable in GI /D; 3. Every nitely generated G-subgroup of GI /D is isomorphic to the coordinate group of some algebraic set Y over G. Theorem 34 says that for any countable innite set I and any non-principal ultralter D over I , the ultrapower GI /D, in a sense, almost universal for the category of all coordinate groups of irreducible algebraic sets (a subcategory of AGG ).

8.2

Length functions.

Denition 37 Let G be a group and let be an ordered abelian group. A mapping L : G is called a Lyndon length function if 142

(1) L(1) = 0 (2) for all g G, L(g ) = L(g 1 ) (3) for all g , h, k G, c(g, h) min{c(h, k ), c(k, g )}, where c(g, h) is dened to be
1 (L(g ) 2

+ L(h) L(g 1 h)).

(Thus c(g, h) is an element of the ordered abelian group 1 , which in turn is a subgroup 2 of the ordered abelian group Q Z ). Axiom (3) means that at least two of c(g, h), c(h, k ), c(k, g ) are equal, and not greater than the third, for all g , h, k G. Such a triple c(g, h), c(h, k ), c(k, g ) is called admissible. It is easy to verify that these axioms imply: (4) for all g , h G, 0 c(g, h) L(g ). In particular, since c(g 1 , h) 0, we obtain the triangle inequality, (5) for all g , h G, L(gh) L(g ) + L(h). Example. Take a free group F with basis X , and dene L : F Z by L(g ) = d(1, g ) the length of the reduced word on X 1 representing g . Then L is a Lyndon length function, and c(g, h) is the length of the largest common initial segment of the reduced words for g and h (this is the reason for the factor 143
1 2

in the denition of c).

Denition 38 A Lyndon length function L : G is called free if for all g G, g = 1, L(g 2 ) > L(g ). Denition 39 Let L be a Lyndon length function on a group G. A sequence (g1 , . . . , gn ), of elements of G is reduced relative to L if L(g ) = L(g1 ) + . . . + L(gn ). We write g = g1 . . . gn to mean g = g1 . . . gn and (g1 , . . . , gn ) is reduced.

8.3

Main technical results.

Let G be a torsion-free hyperbolic group, then for an arbitrary countable set I and a non-principal ultralter D we can consider ultrapower G of G. Let l (g ) = d(1, g ). There is an obvious coordinate-wise extension of function l : G Z to ultrapower G, therefore we obtain a function l : G Z. Notice that l is not a length function on G, although due to hyperbolicity of G, l enjoys the following property c (g, h) min{c (g, f ), c (h, f )} where c (g, h) =
1 2

g, h, f G

(),

l (g ) + l (h) l (g 1 h) .

However we can introduce a length function on G by factoring out constant . Namely, it is easy to see that Z is a convex subgroup of an ordered group Z, therefore we can dene an ordered quotient group Z/Z. This denes a function l : G Z/Z, where 144

l(g ) = l (g ) + Z. For this function inequality () becomes c(g, h) min{c(g, f ), c(h, f )} where c(g, h) =
1 2

g, h, f G

(),

l(g ) + l(h) l(g 1 h) .

Thus l satises all Lyndon length function axioms. Clearly Ker l = {g G | l(g ) = 0} = G, therefore l is not free (in the sense of denition 38), however by Corollary 1 (see section 6) l(g 2 ) > l(g ) for any g G \ G. Also if t G \ G and u G then l(t) = l(tu) = l(ut) = 0. We introduce the following terminology. Elements t G \ G are termed big elements, and n Z \ Z are termed superinteger numbers. Element f G where Z/Z is called a big power of element f . The following lemma follows from lemma 25. Lemma 43 Let v1 , . . . , vn are big elements, and vi vi+1 for any i {1, . . . , n}. Then the sequence (v1 , . . . , vn ) is reduced. In particular v1 . . . vn is a big element. Lemma 44 Let f, g G and , Z. Then 1.) f g 2.) f f providing 0 3.) f u f for any u C (f ), u G 145

4.) g u f providing C (g ) = uC (f )u1 , u G. Proof. Parts 1.) and 2.) are direct consequences of lemma 21. To prove 3.) and 4.) its enough to notice that [f, uf u1 ] = 1 for any u C (f ). Denition 40 Let s, t G, then we say that pair (s, t) has big cancellation if 2c(s1 , t) = 0.
1 has a big cancellation. Then t = u1 (t 1 )

Denition 41 Let t, s G. Then t s if there are elements g, h G such that t = gsh. We denote equivalence class of element t G by [t]. Obviously if elements s and t are equivalent then < G, t >=< G, s >. The next theorem is a consequence of theorem 32. Theorem 35 Let t G and t1 u1 t2 u2 . . . tn un = 1 where u1 , . . . , un G and if ui = 1 for some index i then i = i+1 . Then one of the following holds: 1.) t G; 146

2.) [t] contains an element f - a big power of element f G; 3.) [t] contains an element f g for some f, g G and [f, g ] = 1; 4.) [t] contains an element f g f for some f, g G such that [g, f ] = 1; To prove this theorem we need the following simple lemma. Lemma 45 Let I be a set, D be an ultralter over I and D D. Suppose that

D = D1 . . . Dn , then Di D for some i {1, . . . , n}. Proof. It is sucient to give a proof for n = 2 and the general case follows by induction. Therefore, let D = D1 D2 where D D. Suppose that neither D1 nor D2 belongs to D, then both D1 and D2 are from D and D = D1 D2 D. Contradiction. Proof. (Theorem 35) Consider an equation

t1 u1 t2 u2 . . . tn un = 1. Let
2 n 1 D = {i I | t i u1 ti u2 . . . ti un = 1}

147

where t = ti , then D D. By theorem 32 for each i D there are constants C = C (, l) and D = D (, l) such that ti has one of the following forms: 1) l (ti ) < C ; 2) ti = f g for some g G where l (g ) < C ; 3) ti = (f D g )g for some f, g G where l (g ), l (f ), l (g ) < C and [f, g ] = 1; 4) ti = (f g )D f g for some f, g, g G where l (g ), l (f ), l (g ) < C ; Indeed, theorem 32 says that element ti has one of six possible forms. For example, it may have the following form

1 1 ti = u uj u i i

1 ) . D (x uj u i

We can rewrite element ti as follows

1 1 ti = u i uj ui ui

1 1 1 1 g u i x ui ui uj ui ui ui = (f g )D f

1 1 1 1 where f = u i uj ui ui , g = ui x ui and g = ui .

Therefore we have the following decomposition of the set D

D = D1 D2 D3 D4 , 148

where Dj = i I | ti has a form j {1, . . . 4} from the list above . By the lemma above there exists j such that Dj D which proves the theorem.

8.4

Coordinate groups of irreducible algebraic sets

Let G be a torsion-free hyperbolic group and S G x any system of equations with one variable over G such that VG (S ) = . Theorem 36 Any coordinate group GY of an irreducible algebraic set Y G1 satises one of the following. a) GY = G; b) GY = G, t | [g, t] = 1 , where g is a root element of G; c) GY =G x ; d) GY = f, g for some root elements f, g G; = G g = g f , g , where f , g e) GY = H, s | s1 f s = f where H = G g = g , where f , g = f, g for some root elements f, g G; Corollary Let G be a torsion-free 2-free hyperbolic group, then every irreducible algebraic set V = G over G is isomorphic to a point or centralizer of some element. In particular if G is a non-abelian f.g. free group, then we get Chiswell-Remeslennikov theorem. 149

Proof. (theorem 36) By theorem 34 coordinate groups of irreducible algebraic sets Y G1 are isomorphic to one generated G-subgroups of G and vice versa. Therefore it is sucent to classify up to isomorphism all groups of the form G, t where t G. When t G then G, t = G. Let now t be a big element. Then by theorem 35 G, t = G t unless [t] contains an element of one of the following forms: 1.) t G; 2.) t = f where f G and Z \ Z; 3.) t = f g for some f, g G such that [f, g ] = 1 and , Z \ Z; 4.) t = f g f for some f, g G such that [f, g ] = 1 and Z \ Z; Lemma 46 Let t G \ G and t f where Z \ Z. Then G, t = G, s | [s, f ] = 1 . Proof. Clearly G, t = G, f . Let H = G, s | [s, f ] = 1 then there is an obvious epimorphism : H G, f which sends s to f . Since H is an HNNextension of a group G, any non-trivial element h H can be presented in a reduced form, so that h = u1 s1 u2 s2 u3 . . . sn un , and if ui C (f ) then i = i+1 . To prove that is an isomorphism we need to show that if h is a reduced element then (h) =

150

u1 f 1 u2 f 2 u3 . . .n un = 1. Notice that by lemma 44 any pair (f i , ui+1 f i+1 ) is reduced. Therefore by lemma 43 (h) = u1 f 1 u2 f 2 . . . un1 f n un which proves that element (h) is big and hence is nontrivial. Lemma 47 (i) Let t f g where , Z \ Z, f, g G and g is not conjugated with f 1 . Then G, t = H, s | s1 f s = f where H = G g = g f , g , where f , g = f, g . (ii) Let t f uf where , Z \ Z such that + Z \ Z and f G. Then G, t = H, s | s1 f s = f where H = G uf u1
= g

f , g , where f , g = f, uf u1 .

Proof. (i) Notice rst that since g is not conjugated with f 1 a pair (f , ug ) is reduced for any element u G. Let H1 = G g = g f , g , where f , g = f, g and H = H1 , s | s1 f s = f - HNN-extension of group H1 . It is not hard to see what are dening relations of H . These are relations of G plus relations of the following type:

s1 f k1 sg l1 s1 f k2 s . . . s1 f kn sg ln = 1 where f k1 g l1 f k2 . . . f kn g ln = 1 is a denig relation of f, g . Clearly G, t = f g has relations of the same kind with s replaced by t. Therefore there is an obvious epimorphism : H G, t which sends s to t. To show that is an isomorphism we need to prove 151

that h = 1 implies (h) = 1. Since H is an HNN-extension of H1 any nontrivial element h can be presented in a reduced form h = g1 s1 g2 s2 g2 . . . gn1 sn gn where gi H1 for every i, and there are no subwords of the form sf k s1 or s1 f l s. Since H1 is a free product with amalgamation each gi can be written in the normal form gi = g li s1 r1 . . . smi rmi where sj is a representative of a coset of g in G and rj is a representative of a coset of g in f , g , in addition s1 or rmi may be trivial. Since (s) = t we have (f ) = g f g and therefore for any rj = 1 (rj ) = (f 1 g l1 f 2 g l2 . . . f p g lp ) = g f k1 g l1 f k2 g l2 . . . f kp g lp g = g hjg where hj g . Hence for any gi H1 (gi ) = g li s1 g h1 g s2 . . . smi g hmi g where each hj g for j {1, . . . , mi1 } and sj g for j {2, . . . , mi }. Lets show by induction on k that
k k k

(h) = h f g when k = 1,

(h) = h f when k = 1 for some h G. In particular element (h) is big and hence nontrivial. 152

Case 1.) Let k1 = 1, k = 1. a.) s1 = 1, hmk1 = 1. In this case

(h) = (h f g ) (gk1 ) f g = = (h f g )g lk1 s1 g h1 g s2 . . . smk1 g hmk1 g f g = = h f g g lk1 s1 g . . . smk1 g hmk1 g f g = h f g b.) s1 = 1, hmk1 = 1. In this case

(h) = (h f g )g lk1 g h1 g s2 . . . smk1 g hmk1 g f g = = h f g lk1 h1 g . . . smk1 g hmk1 g f g = h f g c.) s1 = 1, hmk1 = 1. In this case

(h) = (h f g )g lk1 s1 g h1 g s2 . . . g hmk2 g smk1 f g = = h f g g lk1 s1 g . . . g hmk2 g smk1 f g = h f g d.) s1 = 1, hmk1 = 1. This case is similar to the cases b.) and c.). Case 2.) Let k1 = 1, k = 1. Then (h) = (h f )(gk1 ) f g . Suppose that (gk1 ) G then since h is a reduced element (gk1 ) C (f ) and hence (h) = h f (gk1 )f g = h f g . 153

Let (gk1 ) G. Then (h) = (h f ) g lk1 s1 g h1 g s2 . . . smk1 g hmk1 g f g = = h f g lk1 s1 g . . . smk1 g hmk1 g f g = h f g Similarly we consider the cases k1 = 1, k = 1 and k1 = 1, k = 1. (ii) Let now g = uf u1 where u C (f ) and + Z \ Z. Lets show by induction on k that (h) = h (f uf u1 ), or (h) = h (f + uf u1 ) where h f for any Z when k = 1,

(h) = h (f u f ), or (h) = h (f u f (+ ) ) for some u C (f ) when k = 1. Case 1.) Let k1 = 1, k = 1. Then

(h) = (h f uf u1 ) g lk1 s1 g h1 g s2 . . . sk1 g hmk1 g (f uf u1 ). a.) Suppose that s1 = 1, hmk1 = 1. Similarly to the corresponding case in (i) we have (h) = h f uf u1 g lk1 s1 g h1 g . . . g hmk1 g f uf u1 = 154

= h (f uf u1 ), where h f for any Z (this is because h ends with g ). b.) Suppose that s1 = 1, hmk1 = 1. In this case (h) = (h f uf u1 ) uf lk1 u1 h1 uf u1 s2 . . . g hmk1 g (f uf u1 ) = = (h f )uf lk1 u1 h1 u f u1 s2 . . . sk1 g hmk1 g (f uf u1 ). When uf lk1 u1 h1 u C (f ) we have (h) = h f uf lk1 u1 h1 u f u1 s2 . . . sk1 g hmk1 g f uf u1 = = h (f uf u1 ) as desired. Suppose that uf lk1 u1 h1 u = f l k1 . In this case (h) = h f + u1 s2 . . . sk1 g hmk1 g f uf u1 = h (f uf u1 ). c.) Suppose that s1 = 1, hmk1 = 1. Then (h) = (h f uf u1 ) g lk1 s1 g h1 g . . . g hmk2 uf u1 sk1 f uf u1 . If u1 sk1 C (f ) then f u1 sk1 f and (h) = h (f uf u1 ). Let u1 sk1 C (f ) then (h) = h f + uf u1 . 155

All other cases are done by the straightforward verication using the arguments above. In any case element (h) is big and hence nontrivial. Notice that if t f gf where + Z then = + k , k Z and therefore t f gf . Lemma 48 Let t G\G and t f gf where Z\Z. Then G, t = G f where f , g = f, g . Proof. Consider a group H = G f
= f = f

f , g ,

f , g , where f , g = f, g . Then there is

an obvious epimorphism : H G, t = f gf which sends g to t. Any nontrivial element h H can be written in a normal form h = f k s1 t1 . . . sk tk where each si is a representative of a coset of f in G and each ti is a representative of a coset of f in f , g . Then (ti ) = f hi f , where hi C (f ). Therefore for any h = 1 we have (h) = f k s1 f h1 f . . . sk f hk f = 1, which proves that is an isomorphism. Summarizing the results of three lemmas above we get a proof of theorem 36.

156

8.5

Description of irreducible algebraic sets

Lets now describe the irreducible algebraic sets of G1 . First we nd systems of equations solutions sets of which are irreduible algebraic sets. Denition 42 Let G be a group, S G[X ] and VG (S ) = . Then we say that system S satises the Nullstellensatz over G if Rad(VG (S )) = nclG[X ] (S ). Theorem 37 [BMR1] Let G be a group, S G[X ] and VG (S ) = . Then system S satises the Nullstellensatz over G if and only if G[X ]/nclG[X ] (S ) is G-separated by G. Let M =< f, g | r(f, g ), r R > be a non-free subgroup of G generated by some root elements f, g G. Consider the following system SM = {r(f, x) = 1; r R}. Notice that GY = G g = g f , g , where f , g = f, g is isomorphic to G[X ]/ncl(SM ). But GY being a coordinate group of some irreducible algebraic set is G - discriminated by G. Therefore by the theorem above SM satises the Nullstellensatz and hence GY is the coordinate group of an algebraic set VG (SM ). Similarly, let M =< f, g | r(f, g ), r R > be a non-free subgroup of G generated by some root elements f, g G. Consider the system S M = {r(x1 f x, g ) = 1; r R}. Then GY = H, s | s1 f s = f where H = G g = g , where f , g = f, g is a coordinate group of VG (S (M )).

157

Now we turn to a description of solution sets of systems SM and S M . Since torsionfree hyperbolic groups are equationally Neotherian we can assume that SM and S M are nite systems. An arbitrary equation from SM has the following form xl1 f k1 xl2 f k2 . . . xln f kn = 1, where g l1 f k1 g l2 f k2 . . . g ln f kn = 1. Since the number of equations is nite, by theorem 32 there is a constant C1 = C1 (, f, g ) such that VG (SM ) = x G | |x| < C1 , or x = (f x1 )1 , or x = f x1 f , or x = f x1 f for some |x1 | < C1 .

An arbitrary equation from S M has the following form r (x1 f x, g ) = g l1 x1 f k1 xg l2 x1 f k2 . . . xg ln x1 f kn x = 1, where g l1 f k1 g l2 f k2 . . . g ln f kn = 1. By theorem 31 there are i and j such that in one of the following products xg li x1 f ki x, or x1 f kj xg lj+1 x1 the middle occurence of x is cancelled out almost completely. To describe a form of element x we use lemma 38 from section 6. 158

Lemma 38 Let 1 (|xux1 vx| |xu| |vx| + |x|) C (, l), 2 then x has one of the following forms: 1) |x| < D; 2) x = v x1 where |x1 | < D; 3) x = x1 u where |x1 | < D; 4) x = ( v x1 ) u where |x1 | < D; for some constant D = D(, l). Since the number of equations is nite, by lemma 38 there is a constant C2 = C2 (, f, g ) such that |x| < C2 , or x = f x1 , or x = x1 g , or x = f x1 g for some |x1 | < C2 .
1 It is easy to see that r (x 1 f x1 , g ) = 1. Indeed, for example, when x = f x1 g we have

1 kn 1 k2 1 k1 f f x1 g . f . . . g ln g x f f x 1 g g l2 g x 1 = r (x1 f x, g ) = g l1 g x 1 f 1 f 1 f

Therefore
1 1 kn l2 1 k 2 1 k1 . . . x 1 g ln x 1 = g l1 x 1 f x1 = r (x1 f x1 , g ). 1 f x1 g x1 f

159

Finally, VG (S M ) = x G | |x| < C2 , or x = f x1 , or x = x1 g , or x = f x1 g

1 1 for some |x1 | < C2 such that r (x 1 f x1 , g ) = 1 r (x f x, g ) = 1 S M .

Summarizing all of the above we get the following result. Let M =< f, g | r(f, g ), r R > be a subgroup of G generated by root elements f, g G. We introduce the following systems
= {r(x1 f x, g ) = 1; r R}. SM = {r(f, x) = 1; r R} and SM

Theorem 38 Let G be a torsion-free hyperbolic group and V = V1 Vk be the decomposition of an algebraic set V from G1 into irreducible components. If V = G1 , then every Vi , up to isomorphism, has one of the following forms: a) Vi is a point; b) Vi = f C (g ), for some elements f, g, G; c) Vi = VG (SM ) = x G | |x| < C1 , or x = (f x1 )1 , or x = f x1 f , or x =

f x1 f for some |x1 | < C1 , for some M =< f, g | r(f, g ), r R > and C1 = C1 (, f, g )

160

d) Vi = VG (S M ) = x G | |x| < C2 , or x = f x1 or x = x1 g , or x = f x1 g for

1 1 some |x1 | < C2 such that r (x 1 f x1 , g ) = 1 r (x f x, g ) = 1 S M , for some

M =< f, g | r(f, g ), r R > and C2 = C2 (, f, g )

References
[ABC] J.M. Alonso, T. Brady, D. Cooper, V. Ferlini, M. Lustig, M. Mihalik, M. Shapiro, H. Short. Notes on word hyperbolic groups. In Group Theory from a geometric viewpoint. World Scientic, Singapore. Proceedings of the ICTP conference in summer 1990. [Ap] K.I. Appel. One-variable equations in free groups. Proc. Amer. Math. Soc., 19 (1968), 912-918. [Ap1] K. I. Appel. On two-variable equations in free groups, Proc. Amer. Math. Soc. 21 (1969) 179-181. [Arzh] G. Arzhantseva, A dichotomy for nitely generated subgroups of word hyperbolic groups. Preprint University of Geneva, 2002. [BG] G. Baumslag. On generalized free products. Math. Zeitschr., 1962, v.7, 8, pp. 423 438.

161

[BMR1] G. Baumslag, A. Myasnikov, V. Remeslennikov. Algebraic geometry over groups I: Algebraic sets and ideal theory. Journal of Algebra, 1999, v.219, pp.16-79. [BMR2] G. Baumslag, A. Myasnikov, V. Remeslennikov. Discriminating and codiscriminating groups. J. of Group Theory, v. 3, no. 4 (2000), p. 467479 . [BMR3] G. Baumslag, A. Myasnikov, V. Remeslennikov, Discriminating completions of hyperbolic groups, To appear in Bull. of Austr. Math. Soc. [BMRO] G. Baumslag, A. Myasnikov, and V. Romankov, Two theorems about equationally Noetherian groups, J. Algebra 194 (1997), 654-664. [Brady] N. Brady, Branched coverings of cubical complexes and subgroups of hyperbolic groups, Jour. London Math. Soc., Vol. 60 (1999), No. 2, 461480. [Ch] Ian Chiswell, Introduction to - trees. Singapore, World Scientic 2001. [CDP] M. Coornaert, T. Delzant and A. Papadopolous, Notes sur les groupes hyperboliques de Gromov, Lecture Notes in Mathematics, # 1441, Springer-Verlag, 1990. [CE] L. P. Comerford and C. C. Edmunds, Solutions of equations in free groups. In Group Theory, Proceedings of the Singapore Group Theory Conference held at the National University of Singapore, 1987, 347-56. Walter de Gruyter, 1989.

162

[CK] T. Coulbois, A. Kh elif, Equations in free groups are not nitely approximable Proc. Amer. Math. Soc. 127 (1999), 963-965. [CR] I. M. Chiswell, V. N. Remeslennikov, Equations in free groups with one variable., J. of Group Theory, V. 3, no. 4 (2000), p. 445466. [DA] B. Domanski and M. Anshel, The complexity of Dehns algorithm for word problems in groups. Journal of Algorithms, 1985, v. 6, no. 4,p. 543-549. [Ger] S. M. Gersten Ban lectures on hyperbolic and automatic groups. Appeared in the Proceedings of the CRM Summer School in Combinatorial Group Theory, Ban 1996, editor O. Kharlampovich, published by the AMS in the CRM Proceedings and Lecture Notes Series, vol. 17 (1999) [GKM] D. Gildenhuys, O. Kharlampovich, and A. Myasnikov. CSA groups and separated free constructions. Bull. Austr. Math. Soc., 1995, v. 52, 1, p.63-84. [G] M. Gromov. Hyperbolic groups, Essays in Group Theory, ed. S. M. Gersten, M.S.R.I. Pub. 8, Springer,1987, pp. 75-263. [GH] E. Ghys and P. de la Harpe, Espaces Metriques Hyperboliques sur les Groupes Hyperboliques dapres Michael Gromov, Birkhauser, 1991.

163

[GK] R. I. Grigorchuk and P. F. Kurchanov, On quadratic equations in free groups. Contemporary Mathematics, 131 (1992) 159-171, (Part 1). [GK2] R. I. Grigorchuk and P. F. Kurchanov, Some questions of group theory connected with geometry. In Encyclopedia of mathematics, 1989. [GL] R. I. Grigorchuk and I. Lysenok, A description of solutions of quadratic equations in hyperbolic groups, Int. J. of Algebra and Computation, Vol. 2, No. 3 (1992), p. 237-274. [GV] V. Guba, Equivalence of innite systems of equations in free groups and semigroups to nite systems, Mat. Zametki 40, No. 3 (1986), 321-324. [HKS1] A. Hoare, A. Karras and D. Solitar Subgroups of nite index of Fuchsian groups, Math. Z. 120 (1971), p. 289-298. MR 44:2837 [HKS2] A. Hoare, A. Karras and D. Solitar Subgroups of innite index of Fuchsian groups, Math. Z. 125 (1972), p. 59-69. MR 45:2029 [KM1] O. Kharlampovich and A. Myasnikov. Hyperbolic groups and free constructions. Transactions of Math., 350(1998), 2, pp.571-613.

164

[KM2] O. Kharlampovich, A. Myasnikov. Irreducible ane varieties over a free group. I: Irreducibility of quadratic equations and Nullstellensatz. J. of Algebra, 1998, v. 200, n. 2, p.472-516. [KM3] O. Kharlampovich, A. Myasnikov. Irreducible ane varieties over a free group. II: Systems in triangular quasi-quadratic form and description of residually free groups. J. of Algebra, 1998,v. 200, n. 2, p. 517-570. [KP] A. Ko scielski and L. Pacholski, Complexity of unication in free groups and free semi-groups. In Proceedings of the 31-st IEEE FOCS, 824-829, 1990. [KvM1] A. Kvaschuk, A. Miasnikov Big powers and free constructions, to appear in Int. J. Algebra and Comput. [KvM2] A. Kvaschuk, A. Miasnikov One-variable equations in torsion-free hyperbolic groups:I. Descriptions of solution sets, preprint. [L1] R.C. Lyndon. Groups with parametric exponents. Trans. Amer. Math. Soc., 1960, v.96, pp.518533. [Lor] A. A. Lorenc The solution of systems of equations in one unkonown in free groups. Dokl. Akad. Nauk SSSR, 4:262-266, 1963.

165

[LS] R.C. Lyndon, and P.E. Schupp. Combinatorial group theory. Ergebnisse der Mathematik und ihrer Grenzgebiete 89, Springer-Verlag, Berlin, Heidelberg, New York (1977). [Lyn] R. C. Lyndon Equations in groups. Bol. Soc. Bras. Mat., 11:79-102. 1980. [Lyn1] R. C. Lyndon Equations in free groups. Trans. Amer. Math. Soc., 96:445-457, 1960. [Lyn2] R. C. Lyndon The equation a2 b2 = c2 in free groups. Michigan Math. J., V. 6, p. 155-164, 1959. [Lys] I.G. Lysionok. On some algorithmic properties of hyperbolic groups, Izvestia Acad. Nauk. SSSR, ser. mat. 53 (4) (1989), 814-832. [Mak] G.S. Makanin. Equations in a free group . Izv. Akad. Nauk SSSR, Ser. Mat., 46 (1982), 11991273 (English transl. in Math. USSR Izv., V. 21, 1983; MR 84m:20040). [MKS] W. Magnus, A. Karras, D. Solitar Combinatorial group theory: Presentations of groups in terms of generators and relations Interscience Publishers, A division of John Wiley & Sons, Inc. New York, London, Sydney (1966).

166

[MR1] Myasnikov A.G., Remeslennikov V.N., Exponential groups II: extensions of centralizers and tensor completion of CSA-groups, Interntional Journal of Algebra and Computation, 6(6), pp. 687711, 1996. [MR2] A.Myasnikov, V.Remeslennikov. Algebraic geometry 2: logical foundations. Journal of Algebra 234 (2000), p. 225-276. [MR3] A.G.Myasnikov, V.N.Remeslennikov. Length functions on free exponential groups. Proc. N 26. IITPM SO RAN, Omsk, 1996, 1-34. [Olsh] A. Yu. Olshanskii. Almost every group is hyperbolic, Int. J. Algebra and Comput. 2 (1) (1992) 1-17 [Ol1] A. Yu. Olshanskii. On residualing homomorphisms and G-subgroups of hyperbolic groups, Int. J. Algebra and Comput. 3 (4) (1993) 365-409. [Ol2] A. Yu. Olshanskii. Periodic quotients of hyperbolic groups, Mat. Zbornik 182 (4) (1991), 543-567 (in Russian), English translation in Math. USSR Sbornik 72 (1992), N2. [Ozh] Yu. I. Ozhigov, Equations with two unknowns in a free group, Soviet Math. Dokl., 27 (1983).

167

[R] E. Rips, Subgroups of small cancellation groups, Bull. London Math. Soc. 14 (1982), 45-47. [Raz] A. A. Razborov, On systems of equations in a free group, Math. USSR Izvestya, 25(1):115-162, 1985. [RS] E. Rips, Z. Sela, Canonical representatives and equations in hyperbolic groups, Invent. Math. 120 (1995), 489-512. [Schupp] P. E. Schupp, Quadratic equations in groups, cancellation diagrams on compact surfaces, and automorphisms of surface groups, Word Problems II, ed. S. I. Adian, W. W. Boone, G. Higman, North-Holland, 1980, p. 347-371. [T] M. Troyanov. Espaces a courbure n egative et groupes hyperboliques, in Sur les groupes hyperboliques dapres Mikhael Gromov, E. Ghys and P. de la Harpe, eds., vol. 83 of Progresses in Mathematics, Birkh auser, 1988, ch.3.

168