Mathematical Sciences Research Institute Publications 1 FREED/UHLENBECK: Instantons and Four-Manifolds 2 CHERN (editor): Seminar on Nonlinear Partial Differential Equations 3 LEPOWSKY/MANDELSTAM/SINGER (editors): Vertex Operators in Mathematics and Physics 4 Kac (editor): Infinite Dimensional Groups with Applications 5 BLACKADAR: K-Theory for Operator Algebras 6 Moore (editor): Group Representations, Ergodic Theory, Operator Algebras, and Mathematical Physics 7 CHoRIN/MAJDA (editors): Wave Motion: Theory, Modelling, and Computation 8 GERSTEN (editor): Essays in Group Theory Forthcoming CHERN/GRIFFITHS: Exterior Differential SystemsS.M. Gersten Editor Essays in Group Theory With 46 Illustrations Springer-Verlag New York Berlin Heidelberg London Paris ‘TokyoS.M. Gersten Department of Mathematics University of Utah Salt Lake City, Utah 84112 USA AMS Subject Classification: 20XX Individual Articles: 20F0S/20F32, 20F06/20F06, 20F10/ 20F32, 05C25, 57M15/20E06, 20F10 Library of Congress Cataloging in Publication Data Essays in group theory. (Mathematical Sciences Research Institute publications; 8) Bibliography: p. 1. Groups, Theory of. I. Gersten. S. M. Il. Series. QAITLE79 1987 512'.22 87-24336 © 1987 by Springer-Verlag New York Inc. All rights reserved. This work may not be translated or copied in Whele-OF in part without the written permission of the publisher (Springer-Verlag, 175 Fifth Avenue, New York, NY 10010, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of in- formation storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden ‘The use of general descriptive names, trade names, trademarks, etc. in this publication, even if the former are not especially identified, is not to be taken as a sign that such names, as understood by the Trade Marks and Merchandise Marks Act, may accordingly be used freely by anyone. ‘Text prepared in camera-ready form by the MSRI. Printed and bound by R.R. Donnelley & Sons, Harrisonburg, Virginia. Printed in the United States of America. 987654321 IN 0-387-96618-8 Springer-Verlag New York Berlin Heidelberg 140-96618-8 Springer-Verlug Berlin Heidelberg New YorkPreface In late June 1985 G. Baumslag, M. Gromov, P. Shalen, J. Stallings and I were invited to MSRI, Berkeley. I organized a seminar and each of us spoke on a topic of current interest in group theory. Because of the high quality of the talks it seemed appropriate to retain a permanent record of the seminar. With the encouragement of Calvin Moore, who proposed the title, "Essays in Group Theory", I undertook the task of editing this volume. Roughly, Gromov's and Shalen's papers deal with hyperbolic groups and actions on R-trees. Stalling's paper and my own deal with applications of diagrams to group theory. The joint article of Baumsiag and Shalen studies presentations by the classical method of representations. I want to thank the referees, who provided me with invaluable assistance. [ should also like to thank Faye Yeager for the fine job of typing the articles and the staff of MSRI for their assistance. S. M. GerstenESSAYS IN GROUP THEORY Table of Contents Page Affine Algebraic Sets and Some Infinite 1 Finitely Presented Groups Gilbert Baumslag and Peter B. Shalen Reducible Diagrams and Equations 15 Over Groups S.M. Gersten Hyperbolic Groups xi) M. Gromov Dendrology of Groups: An Introduction 265 Peter B. Shalen Adian Groups and Pregroups 321 John R. StallingsAFFINE ALGEBRAIC SETS AND SOME INFINITE FINITELY PRESENTED GROUPS By Gilbert Baumslag and Peter B. Shalen 1 Introduction The obiect of this note is to prove the following two theorems. Theorem A: Let p be a prime and let the group G be given by the finite presentation a) If e>Oand 0 < m < (n-1)(p*1), then G is infinite; indeed G contains an element of infinite order. Theorem B: Let p be an odd prime and let G be given by the finite presentation (ay G = n2/4, where n is the minimal number of generators of G. Notice that if we choose e large enough, then in Theorem A we have m > n2/4. Nevertheless G is still infinite. Of course if G were a finite p-group, then the Golod-Shafarevich inequality would hold. In order to illustrate the scope of Theorem A, consider the groups G = and H = , which are infinite whenever 1/2 + 1/m + 1/n ¢ 1. So the fact that G is infinite is not new. But the fact that H is infinite does not seem to follow from known results, even though it is obtained from the infinite triangle group G44 4 by adding one more relation. The inequality in Theorem B is similar to, but a little more restrictive than the one in Theorem A. As a reward all of the groups involved, which are infinite by Theorem A, contain free subgroups of rank two. It is worth noting that G35 does not satisfy the Inequality demanded by Theorem B, which is as well because it is solvable. This serves to show that the inequality in Theorem B is necessarily strict--the same can be said of Theorem A since Gp 9 > is finite. 3. Some dimension estimates and the proof of Theorem A Let G be a finitely generated group, d a positive integer and K an algebraically ciosed field of characteristic zero. Lubotsky and Magid [7] have, among others (see also Culler and Shalen [3]), shown how to parametrise the equivalence classes of semi-simple representations of G in SL4(K) by the points of an affine algebraic set X,(G) (= X,(G.K)). Our objective here is to obtain an analogous parametrisation of the semi-simple representations of the groups G dofined by (1) and (2) (see Theorems A and B), in the case where K isan algebraically closed field of characteristic p. Our approach, which leans very heavily on work of Procesi [8], is ad hoc, depending on the presentations of the groups involved. This allows us to find a lower bound on the dimension dim X4(G) of this affine algebraic set. Suppose, for the moment, that K is an algebraically closed field and that d>1 and n> 41 are fixed integers. Furthermore, let ¢ denote the free K-algebra freely generated by x. and let M denote the ring of all d Xd matrices with entries in K. The set Rg(®) (= Rg(®K)) of representations of @ in M can be put in a one-to-one correspondence with the points of U = M®, an effine space of dimension nd@ over K. Indeed, if p € R,(#) then we associate to p the point (o(x;),...,0(x,)) € U. Conversely, given u = (my,....m,) € U, the corresponding representation , € Ry(¢) is defined by py(x;) = mj. The group GL,g(K) acts on U by conjugation: ‘¥e(my,...m,) = (¥my ¥m,¥7}), It follows that two points of U are in the same orbit under this action if and only if they correspond to equivalent representations of © in M. Procesi [8] has shown how to construct a “quotient” of U by this action of GLy(K). We review the basic facts here. For each element % € @ and (cf. Lubotsky and Magid [7] and also Culler and Shalen [3], for the characteristic zero case) each integer i, 0 < i < d-1, we define a function £5: U — K as follows: fpla) = the coefficient of the degree-i term of the characteristic polynomial of »,,(¥). It is not hard to see that fy is given by a polynomial in the coordinates of the affine space U and that each fy is constant on the GL4(K)-orbits of U. The polynomial functions on U (to K) form a K-algebra K(U] (see e.g. Hartshorne [6], p. 16). In fact K[U] is a polynomialalgebra in nd? variables over K. We view the functions fy (wy € @ 1 i and so there exist irreducible representations of F inS14(K). Let ug € Ug correspond to such a representation. Then v7 (plug) is the orbit of ug under the action of GLy(K). It is clear that p~l{pfug)) is also the orbit of ug under the action of SL,(K) on Uy. Hence dim p“'(plug)) < dim SL4(K) = a2-1. On the other hand dim p~'p(ug)) = dim Uy - dim Vp = n(d2-1)-dim Vo (nve Shafarevich [10]. Theorem 7 (1), p. 60). Thus a dim Vp > (n-1Nd?~1). (See Lubotsky and Magid [7] for the characteristic zero case.) Now suppose that G is a group defined by (1) (see the statement of Theorem A) and suppose that 0 < m < (n-1)(p°+1). In particular it follows that n > 1. We now choose d = p® and also choose K. already assumed algebraically closed, to be of characteristic p. Then a matrix in SLg(K) is of order dividing d if and only if its characteristic polynomial is (x-1)1 = x4 - 1, So a representation of F corresponding to a point u € Up defines a representation of G if and only if the coordinates of u satisfy m(d-1) additional polynomial equations. The polynomials involved are all contained in C. In other words if RgG) is the algebraic subset of Ug defined by these wquations, the points of R,(G) are in a one-to-one correspondence with Uhe representations of G. As already noted the extra polynomials tofining Ry(G) are all contained in C. So, precisely as in the discussion above of the definition of Ug, we find that Ky(G) = pK AG), where X4(G) is an algebraic subset of Vg defined by m(d-1) polynomials. {l follows that the points of X,(G) are in a one-io-one correspondence with the equivalence classes of semi-simplerepresentations of G. Now X,(G) is non-empty, since the trivial representation determines a point of X4(G). Moreover (see Hartshorne [6], p. 6) for each component Y of X4(G) we have dim Y > dim Vg - m(d-1) > (n-1Mp?°-1) ~ m(p®-1. Therefore (4) dim Xq(G) > (p®-1)((n-1p°+1)-m). This is the dimension estimate that we have been looking for. The proof of Theorem A follows easily from (4). Thus suppose that G is defined by (1) and that m < (n-1)(p°+1). Then, by (4), dim X4(G) > 0. So G has infinitely many inequivalent representations and therefore is infinite (see e.g. Blackburn and Huppert [2]). Since a finitely generated periodic group of matrices over a commutative ring is finite (see e.g. Wehrfritz [12]), G contains an element of infinite order. This completes the proof of Theorem A. 4. The proof of Theorem B The proof of Theorem B is similar to that of Theorem A, but a little more elaborate. Suppose that, as in the statement of Theorem B, (2) swe = ay WE Fe If G is any group then we denote by s4{G) the minimal number of generators of G. Notice that if G has a presentation of the kind given by (1) (or (2), or course), then u{G) =n. In particular, every presentation of this kind contains exactly n generators. Thus the only variable in such a presentation for G is m. Put@(G) = n(p-1)-m, where m is here chosen minimal. It follows from the Roidemeister-Schreier method that if H is any subgroup of finite index in G containing all of the elements of order p in G, then H also has a presentation of the same kind as that given by (2). So a(H) is also dofined. Indeed we find, again invoking Reidemeister-Schreier, that (5) a(H) > [G;H]- (G6). Now suppose that p is odd, that d = p - 1 and that K is again an algebraically closed field of characteristic p. In addition suppose that 0 < m < (n-1)p - m. Then a(G) > 0 and hence (6) a(H) > 0. ‘An analogous argument to that which yielded (4) gives us ”) dim X,_(G) > (o-2M(n-1)p ~ m) = (p-2)+ a(6). ‘Therefore (8) dim X,_ (6) > 0. So G is infinite, as before. We need somewhat more. The proof of Theorem B now proceeds by way of contradiction. ‘Thus suppose that G does not contain a free subgroup of rank two. ‘Then the image of every representation of G in SL,_1(K) is virtually solvable, by a theorem of Tits [11], i.e. it contains a subgroup of finite index which is solvable. In order to continue with the proof we need the following lomma, which can be demonstrated by a slight variation of the proof of the Lie-Kolchin-Mal'cev theorem as given, for example, in Robinson[9]. We give a proof here for completeness. Lemma: Let G be a finitely generated subgroup of the general linear group GL4(K) of degree d over a field of characteristic p. If G is virtually solvable and infinite and if d1. If G acts reducibly on K4 then either there ie a proper subspace V of K4 such that an infinite image of G acts faithfully on V or else there is a proper quotient space W of K4 such that an infinite image of G acte faithfully on W. In both cases the desired conclusion follows inductively. Consider now the homomorphism 6 from G into the multiplicative group of K defined by taking determinants: B: g — det(g). Then &(G) is a finitely generated subgroup of the multiplicative group of K. So either 8(G) is finite of order prime to p or else has an infinite cyclic factor group. This reduces the proof of the lemma to the case where G is a subgroup of SL,(K). So we are left with the case where d> 1, G acts irreducibly on K% and is a subgroup of SL4(K). Let A be any abelian normal subgroup of G. By Clifford's theorem, A acts completely reducibly on Ka, Since we may assume without loss of generality that K is algebraically closed, we can decompose K4 into a direct sum of subspaces Wy)...Wq such that A acts scalar-wise on each W;. Moreover we may also assume that the one-dimensional A-invariant subspaces of a given W; are all isomorphic as A-modules and that these one-dimensional A-invariant subspaces of the different W, are not isomorphic (see e.g. the proof of theLie-Kolchin-Mal'cev theorem given by Robinson [9]). The number of these W; is no more than d (< p). Moreover G acts as a group of permutations of these W;. Hence there is a subgroup G, of G of finite index dividing (p-1)! which leaves every one of the W, invariant. If there is more than one W; the desired conclusion again follows by Induction. Suppose then that there is one Wj, i.e. K4, This means that every abelian normal subgroup A of G is actually scalar. Since AC SL4(K), the entries on the main diagonal of each element of A are d-th roots of unity. Thus A is finite, ie. every abelian normal subgroup of G is finite. Now G contains a normal subgroup of finite index which is solvable. Among all solvable normal subgroups of finite index in G, choose one, say H, whose derived length is as small as possible; thus no finite-index normal subgroup of G has derived length less than £. Since G is infinite, we have H #1, ie. £>0. The £-th term H(2) of the derived series of H is an abelian normal subgroup of G. Honce H(£) is finite. Now G is residually finite by a theorem of Mal'cev (see [12]). So we can find a normal subgroup N of G of finite index such that NA Hg) = 1. Put K=NAH. Then K is a solvable normal subgroup of finite index in G of derived length at most #- 1. This contradicts the choice of H and the minimality of 2 and so completes the proof of the Lemma. We continue now with the proof of Theorem B. By the Lemma there is a subgroup H of finite index prime to p in G such that H,,, is infinite. By assumption G does not contain a free subgroup of rank Iwo. Hence, by a theorem of Bieri and Strebel [1], H is a descending HNN extension H = 0. Put Hy, = enit®,B). Then again H, contains all the elements of order p in H; furthermore CH:H,] =k and wy) < w+) for every k. But by (5) (9) a(H,) > k+ a(H) for every k. Hence a(H,) can be made as large as we please by making k sufficiently large. But as already indicated, Hy, can be presented on 4 + 1 generators by a presentation of the kind given by (2). Therefore it follows immediately from the definition of a(H,) that aH) < (u+1)(p-1) for every k. This contradicts (9) and thereby completes the proof of Theorem B. Part of this work was carried out in June 1985 at MSRI. We thank MSRI and the NSF for their support.References Bieri, R. and Strebel, R., Almost finitely presented solvable groups. Math. Helv. 53 (1978), 258-278. Blackburn, N. and Huppert, B., Finite groups II. Grundlehren der mathematischen Wissenschaften 242. Springer-Verlag, Berlin, Heidelberg, New York (1982). Culler, M. and Shalen, P.B., Varieties of group representations and splittings of 3-manifolds. Annals of Math. 117 (1983), 109-146. Golod, E.S. and Shafarevich, I.R., On the class field tower. Izzv. Akad. Nawuk SSSR 28 (1964), 261-272. Haboush, W.J., Reductive groups are geometrically reductive. Ann. Math. 102 (1975), 67-83. Hartshorne, R., Algebraic geometry. Graduate Texts in Mathematics 52. Springer-Verlag, New York (1977). Lubotsky, A. and Magid, A., Varieties of representations of finitely generated groups. To appear Memoirs American Math. Soc. 1986. Procesi, C., Finite dimensional representations of algebras. Israel J. Math. 19 (1974), 169-182. Robinson, D.J.S., Finiteness conditions and generalised solvable groups. Springer-Verlag, Berlin, Heidelberg, New York (1972). Shafarevich, I.R., Basic algebraic geometry. Grundlehren der mathematischen Wissenschaften in Einzeldarstellungen 213. Springer-Verlag, New York, Heidelberg, Berlin (1974).Tits, J., Free subgroups in linear groups. J. Algebra 20 (1972), 250-270. Wehrfritz, B.A.F., Infinite linear groups. An account of the group-theoretic properties of infinite groups of matrices. Ergebnisse der Math. Bd. 76. Springer-Verlag, Berlin, Heidelberg, New York (1971).REDUCIBLE DIAGRAMS AND EQUATIONS OVER GROUPS By S. M. Gersten University of Utah Salt Lake City, UT 84112 U.S.A. ABSTRACT Diagrammatic reducibility is related to the solution of equations over groups. Sufficient conditions for the reducibility of all spherical diagrams are given, unifying and generalizing work of Adian, Remmers, lyndon, and Sieradski. Hyperbolic 2-complexes are defined and the word problem is solved for their fundamental groups. Partially supported by NSF Grant DMS-8400882 and DMS-8120790.80. §1. 82. 83. 84, $5. $6. Append CONTENTS Introduction Quotients Corners Diagrammatic Reducibility Reducibility Criteria Extension Theorem and Staggered 2-Complexes Subdivision Theorems Hyperbolic 2-Complexes80. Introduction An early version of this paper was written in December 1984 to provide evidence for a conjecture of Stallings [22] about surfaces in handlebodies of dimension 3 which would have implied the Kervaire-Laudenbach conjecture [11], that a non-singular system of equations over a group G has a solution in an overgroup of G. Since [ was unable to prove Stallings’ conjecture in general, I decided to write up my partial results and move on to other things. While visiting MSRI at Berkeley in February 1985 I discovered a counterexample to Stallings’ conjecture [7], namely eleven elements & 1< is 11, in the free group F(a,4) = F of rank 2 such that ie mel bot 1¢€ im gy, where H = F(a,8,t)/<> This ‘counterexample is tilt” somewhat of a mystery, since the Kervaire conjecture is true for the equation at@bt~! = 1. Namely, for any group G and a,b € G, the natural map G—> G*/< is injective [13], where t generates an infinite cycle. My partial results now assumed much greater interest than previously. At the very least, they show where noz to look for counterexamples to the Kervaire-Laudenbach conjecture. My idea was to reformulate Stallings’ conjecture B [22] as a geometric statement about spherical diagrams in a 2~complex and apply woometric techniques to it. The equivalent version [6] is the statement that for a 2-complex X with H){X)=0 and for any spherical diagram f: C — X (that is, combinatorial map of a cell complex C for the 2-sphere to X), one has 1 € ved cor% yf (the "reciprocity law") where 2%,(f) is the vertex label of f at v € 0, which tells how f maps the link of v in C to the link of f{v) in X. In this language, the Kervaire-Laudenbach conjecture states that under the same hypotheses, any one vertex label is in the normal closure of the remaining vertex labels. Since the former conjecture is fuise and the Jatter is unknown in general, although partial results abound [5], [8], [11], [14], I asked for which X with H,(X) = 0 the reciprocity law is valid.A result (Theorem 3.3) I establish is that if X is diagrammatically reducible (Definition 3.1) then the reciprocity law is valid for X. Roughly speaking diagrammatic reducibility means any spherical diagram in X contains a pair of adjacent faces with an edge in common so that the faces are mapped mirror-wise across this edge. This notion was introduced by Sieradski [20] who gave a "coloring test" for it. I generalize Sieradski's result in Theorem 4.7, the "weight test", which includes some earlier results of Adian's and Lyndon's as special cases and which is strongly suggestive of spaces of non-positive curvature. Thus the weight test implies diagrammatic reducibility which in turn implies the reciprocity law. I prove in %5 that staggered 2-complexes satisfy the reciprocity law. This includes all cyclically reduced 1-relator presentations. An unpublished result of S.J. Pride's, Corollary 5.7, drops out on solving a single equation, cyclically reduced in its variable letters, in an overgroup. In $6 I examine combinatorial invariance of the reciprocity law, diagrammatic reducibility, and the weight test. The results are that all are invariant under subdivision, provided, for the reciprocity law, one restricts oneself to complexes X with H2(X) = 0. In an appendix we formulate a definition of hyperbolic 2-complexes and solve the word problem for the corresponding presentation of the fundamental group. T am grateful to MSRI, Berkeley, for inviting me to spend two weeks in June 1985 there. That was the occasion for the group theory seminar where these results were reported on and which was the genesis of this volume of essays on group theory.NOTATION If G is a group and g € G, then ¢° = Cxgx"!/x € G2, the conjugacy class of g. If S is a subset of G, then denotes the subgroup generated by S and <>q denotes the normal closure of S in G (we write <> for <>q if G is understood). If © is a group presentation then K(®) denotes the 2-complex canonically associated to @.$1. Quotients 1.1. Definition A cellular map f: X —+ Y of CW complexes is said to be combinatorial if the restriction of f to each open cell of X is a homeomorphism onto its image. We shall call a 2-complex X combinatorial if for each 2-cell a of X the attaching map fq: 8! — x") is combinatorial for a suitable subdivision of the 1-sphere st, Subcomplexes and covering spaces of combinatorial 2-complexes are themselves combinatorial, and we have a category of combinatorial 2-complexes where morphisms are combinatorial maps. If K is a subcomplex of the combinatorial 2-complex L, then the quotient complex L/K obtained by identifying K to a vertex is not necessarily combinatorial. However, with one small restriction it almost is. 1.2 Lemma Suppose I is a proper closed interval in the boundary OD of the 2-disc D. Then the quotient space D/I is homeomorphic to D. 1.3 Proposition Let K be a subcomplex of the combinatorial 2-complex L. Assume that for each 2 cell a of L-K the attaching map fg s1— Lb does not factor through the inclusion Kc {0), Then there is a combinatorial 2-complex X and a cellular map n: L/K — X which is universal for cellular maps L/K —+ Y where Y is combinatorial. Proof. We define x) = aK). tf £4: 8, — L!) is the attaching map of the 2 cell a of L+K, let f,: 8, —+ X" be the composition sh AS.) (imp!) = xf, Let qi 8, —+ 8) be the identification map which collapses the connected components of the singular set of f,, to distinct points. The map dq exists since f,(Sj) GK. Let ¢,: $4, — X'!) be thefactorization in the diagram s oe ! % zl 4 Ss We take the maps g, to be the attaching maps of 2-cells D, and construct X as the push out in the diagram H. US >. UD t x -----5X when the left vertical arrow maps $1 via gq. By repeated application of Lemma 1.2, we see that the map gi 84 — 84 extends to a map pg: Dg — Dg which is 1-1 off the singular set of qq and induces a homeomorphism of the decomposition space of Pa with Dy. Using the maps Pq we define the map 2: L//K —+ X. The universal property of the map 7 follows from properties of push outs. 1.4 Convention We shall denote the complex X constructed in 1.3 by L/K. The map 7: L/K —» L//K is a homotopy equivalence, so for homotopytheoretic purposes L/K and L//K are the same. However we shall make essential use of the combinatorial structure of L//K in the sequel.$2. Corners Let X be a combinatorial 2-complex and let f4: Dg —> X be the characteristic map of a 2-cell a of X. The subdivision of Si = ODq making £184 a combinatorial map selects a finite set of points vy,vp,...v, in 84. We define the oriented corner y, of X at v, to the oriented interval «oS; in, 8, ¢ = €x € Dy I d(v,x) = €2 for suffiently small € > 0. aq where “i % ain, / of Q Two corners are equivalent if they can be isotoped to each other in Dg orientations. We are only interested in equivalence classes of corners, holding the end points in 4 - Cvy,...v,2 and preserving the so we shall frequently identify equivalent corners. Bach corner ¥; determines its opposite v;' which is the same class of intervals with the opposite orientation. A corner of X is a corner of Dy for some 2-cell a of X. 2.1 Lemma A combinatorial map f: X —* Y of combinatorial 2-complexes induces a map of corners of X to corners of Y. This follows from the definitions and from invariance of domain. 2.2 Definition Let X be a combinatorial 2-complex. We let Fy be the free ¢roup with free basis consisting of a choice of one of each pair ofequivalence classes of oppositely oriented corners of X. If f: XK — Y is a combinatorial map, with X and Y combinatorial, then f induces a homomorphism fx: Fy —* Fy by Lemma 2.1. Thus X m— Fy is a functor from combinatorial 2-complexes to free groups. 2.3 Remark If X is a combinatorial 2-complex, let E be the link complex of the zero-skeleton of X. Thus = is the disjoint union of Link(v) for v € x 0), where X" is the second barycentric subdivision of X. Let £ be obtained from the graph £ by identifying all vertices of £ to a single point. Then there is a canonical identification Fy = ,(E) obtained by observing that oriented edges of = correspond 1-1 to oriented corners of X. The graph E is known to group theorists by various names, including "star graph" and “coinitial graph.” Suppose now that M is a combinatorial cellular structure on a closed oriented surface. If v is 0-cell of M, then the orientation on M determines an orientation in a neighborhood of v and this in turn selecte an oriented corner in each 2-cell of M incident with v. If Yy:Tor-0%_ are the oriented corners encountered in order in a circuit around v once in the positive direction from some starting point, we set £, = [v1 %9..¥,], the conjugacy class in Fy of the circuit VV 2. y: Observe that %, is insensitive to cyclic permutations of ¥4%p..%q 80 is well defined. 2.4 inition We set 2£(M) = my where the product is indexed by the O-cells of M. As a product of conjugacy classes £(M) is a subset of Fy and is independent of the order of factors in the product. If f:M-— X is a combinatorial map, then we set £y(C) = falZ,) and £(f) = faX(M). The former is contained in the conjugacy class of fe(vy)fe(¥2)..fely,) if 2%, = [vy 79..-%Q]. and the latter is a subset of Fy. In a commutative diagramno . Mea of combinatorial maps, we have 2th) = L(ef) = sex) as subsets of Fy. 2.5 Definition If f: C —+ X is a combinatorial map, where C is some cell structure on the 2-sphere with its standard orientation, we say that f satisfies the reciprocity taw if 1€ z(f) & Fy. We say that X satisfies the reciprocity law if f does for all such maps (and all cell structures on $4). In this case we write RL(X) to denote that X satisfies the reciprocity law. 2.6 Remark In [22] Stallings proposed "Conjecture B” which was shown in [6] to be equivalent to the assertion that if Hp(X) = 0, then the 2-complex X satisfies the reciprocity law. In [7] we exhibited a counterexample to RL(X) where X is the dunce hat, the 2-complex with one cell in each dimension and attaching map ttt of the unique 2 cell, where t is the 1-cell. The interest in "Conjecture B" is that it would have implied the Kervaire~Laudenbach conjecture, which in this setting states that if HA(X) = 0 and if f: C —+ X is a combinatorialmap for C a cell structure on S* with vertex labels # (0 (0) = oe Fy Oronty (0, where Cl! = €v9,V4...¥,2, then x6 is in the n normal closure of 2%, (f),...%, (f) in Fy. For if 1€ ay 2, (, then #105 mt, th ¢ is the natural map and if 84,8,. r H € F, h that 1€ OD og)", 2, € Fy are sue! Hos (e;) Waa, then 1€ Hef X, and is 2.8.3 "Property G": Let n: Fy —* G be an arbitrary homomorphism of groups and let #: G— H = G*T/> be the natural map. If 882.48, € G are such that 1 € Hote » then 1 € a ws In particular, observe the following consequence, ‘or r= 1,2 in 2.8.3. 2.9 Corollary If RL(X) holds then the map G— G*T/> is injective on conjugacy classes. In connection with the Kervaire-Laudenbach problem, let us make the following observation, which seems to have escaped notice previously. 2.10 Proposition Let X be a combinatorial 2 complex with H)(X)=0. Let n: Fy —* G be a homomorphism to a finite group G. Then the canonical map #: G — H = G*T/> is injective on conjugacy classes. Proof, Let xy €G be such that o(x)¥ = wy and let G —& Un) be a unitary representation of G. Consider the commutative diagram bolowu(n) 2 at where H' is the push out of p and 9. Thus H = Un)*T/>. The hypothesis H,(X) = 0 is equivalent to the assertion that the system of equations he , e € x"), is independent, so it follows from the result of Gerstenhaber and Rothaus [8] that there is a homomorphism o: H' —* U(n) such that 08 = Ty) It follows that o(x)U) = py)%) and hence that x pl) bats ps where Xp is the character of ». Since p is ee arbitrary, it follows that x°'= y®, since the characters of @ finite group separate conjugacy classes. 2.11 Remark The homomorphism ¢#: G—+ H of 2.10 is also injective on conjugacy classes if G is conjugacy-separable, where a group G is called conjugacy separable if distinct conjugacy classes can be separated in finite quotients of G. It is known that polycyclic-by-finite groups are conjugacy-separable [4], but Gl,(2) is not conjugacy-separable for n 2 3 [18]. We shall need another aan of the condition RL(X) in the sequel, which we discuss now. 2.12 Definition Let f: C+ X be a combinatorial map of combinatorial 2-complexes. An f-labeling of the oriented edges E(C) of C is a function y: E(C) + Fy such that (8) = Weyl, where & is the edge oppositely oriented to e, and where for each 2-cell a of C, the product of the ¥{e) and the fs(v) in order in one circuit around the boundary of a is 1, where e ranges over edges and Y ranges over corners. Forming this product from another starting point gives a conjugate element of Fy, so the condition is independent of starting point2.13 Theorem Let f:C — X be a combinatorial map where C is a cell structure $2. There is an f-labeling y of B(C) iff 1 € #(f) © Fy. : Proof. Let U be an open regular neighborhood of X) in X and let X' = X - U. Form the pull back C’ below: ct oc oo f peel Then C’ is a 2-sphere with holes whose boundary components are mapped into £, the link complex of X (see 2.3) which occurs as the boundary of U in X. The vertex labels #,(f) just describe how the boundary component of C' corresponding to a vertex v € cl is mapped into £, the graph © with all vertices identified. By [7] lomma 3.1 the composite map ac’ ELLOS, » ——--. & oxtends to a map fC’ — £ iff 1€ x(f) Fy = 4(2). Restricting £; to the 1-cells of C’ ~ QC’ and taking homotopy classes in w4(Z) = Py gives the desired f-labeling of E(C).2.14 Remark If f: C —+ X is a combinatorial map of a cellular subdivision C of 8% to the combinatorial 2-complex X and if f satisfies the reciprocity law, then so does ef, for any combinatorial map ¢: X — Y of combinatorial 2-complexes. This is evident from functoriality, since ga(2(f)) = £lef), so if 1 € #(£), then 1 € #( ef). In particular if RL(X) holds for a covering space X of X, then RL(X) also holds. The converse is false however. We shall see later that the reciprocity law holds for the real projective plane, but it does not hold for the identity map of the 2-sphere. 2.15 Theorem Let K be a connected subcomplex of the connected 2-complex L. If L//K = X is defined and satisfies the reciprocity law, then x4(K) injects into x4(L). If in addition L is aspherical, then K is also aspherical. Proof. We may assume that K has a single vertex by collapsing out a maximal tree of K if necessary. This does not change X. Thus all 1-cells of L - K are attached trivially. We define a homomorphism n: Fy — ,(K) as follows. Let a be a 2 cell of L -K and let fq: 84 — L') be the attaching map of a. If I is a maximal oriented subinterval of si mapped into K by f,, then I corresponds to a unique oriented corner y of X. We send y into the homotopy class in ~4(K) determined by f,1I, thus defining n: Fy —> 74(K). The map *4(K) + 7(L) induced by inclusion K&L is precisely the homomorphism 14K) > ey(KT/<> of 2.8.3. This is a consequence of the van Kampen theorem. If X_ satisfies the reciprocity law, then 2.8.3 for r= 1 implies that the map x4(K) — 74(L) is injective, as desired. If in addition L is aspherical, an elementary covering space argument shows that K is also aspherical. This completes the proof of 2.15.$3. Diagrammatic Reducibility 3.1 Definition By a diagram we shall mean a combinatorial map f: C — X, where C is a cell structure on S* (other authors call these spherical diagrams, but we shall never have need of other kinds of diagrams). We say that two distinct faces (= closed 2 cells) F and F' of C are opposite with respect to f if F and F’ have an edge e in common, and if there is an orientation reversing homeomorphism g: F' — F fixing F AF’ pointwise with f Sipe = flip. The diagram f: C — X is reducible if there is an opposite pair of faces and reduced otherwise. If the diagram f: C —+ X is reducible with faces F, F’, edge e, nnd homeomorphism g as above, then there is a well known method of obtaining diagrams from f with fewer faces. One cuts out F V F' and identifies points on the boundary via g. If F A F’ has c connected components (*), the result is at most c two-spheres joined together at certain vertices so that the total number of faces is reduced by two. We call the combinatorial 2-complex X diagrammatically reducible and write "DR(X) is satisfied” if every spherical diagram In X is reducible. This notion was first considered by Sieradski [20]. Unfortunately there are a number of different notions called diagrammatic asphericity in the literature [2] including the original notion of Lyndon's [16]. For example, Lyndon allows the real projective plane ([16], Proposition 11.1, page 161) to be diagrammatically aspherical, although it is not aspherical =np{RP?) = 2. The difference can be illustrated as follows. Take as the model for RP? = K(@), where @ is the presentation . Label the corners as follows in the unique 2 cell of K(@): (")The number c must be counted in the 2-disc mapped to F by the characteristic map of this 2-cell, and not in C itself.K(®): t Consider the following spherical diagram 2 in K(®), where the second face is at ~. 2 & = By Lyndon's definition, the two 2-cells of 2 have an edge in common and reading around the boundaries, beginning with that edge we get the same word t?, Hence Lyndon allows us to remove the pair of 2-cells by a cut and paste operation. However this removal cannot be done by a homotopy. In fact, the diagram 2 represents the generator for rp(RP), since it lifts to the identity map of $2. Following a suggestion of Sieradski's student, William Bogley, we have chosen to call his notion “diagrammatic reducibility". In the next section we shall generalize Sieradski's “coloring test" [20] to produce a useful sufficient condition for diagrammatic reducibility. 3.2 Remark Let us observe that if X is diagrammatically reducible, then no(X) = 0. For a transversality argument shows that p(X) isgenerated as a x,(X) module by classes of spherical diagrams. Given a spherical diagram, a pair of adjacent faces can be removed by a homotopy. An inductive argument then shows that any spherical diagram represents the zero element in 79(X). The converse however is false. An aspherical 2~complex need not be diagrammatically reducible. The simplest example is the dunce hat X = K(®) where @ = . A reduced diagram in X is shown below: 3.3 ‘theorem Let X be a combinatorial 2-complex. If every diagram f; C —» X is reducible, then X satisfies the reprocity law. Proof. The argument is by induction on the number of faces of C, with the induction starting when C has two faces, F and F’. F’ We define the labeling function ¥ on the edges of F so that the product of y of edges and fz of corners in order is 1 in Fy. This is possible since there is an edge e common to F and F’, so e is counted once in the product. We use the homeomorphism g: F' — F in thedefinition of reducibility to define ~ on the edges of F' by the rule w + g(e') = y(e'). This is possible since g fixes all edges in F f\ F'. Then y so defined is an f-labeling of E(C). We assume now that every diagram in X with fewer than n faces satisfies the reciprocity law and f: C —> X is a diagram with n faces, n > 2. Since f is reducible, we have a pair of opposite faces F and F’, an edge e in FF’, and an orientation reversing homeomorphism g fixing e of F' onto F satisfying f + gly: = fl py. We perform the cuts removing F V F'. The diagram below illustrates the general situation. The homeomorphism g here is reflection about the y axis. The shaded regions are cells which will be glued along parts of their boundaries to give spherical diagrams. These latter spherical diagrams possess f-labelings by the inductive hypothesis. When cuts are made on these latter diagrams, the result is a g - equivariant f-labeling y on the cut diagram above. Since e is a free edge in OF which is so far unlabeled, we may extend y over E(F). Using the homeomorphism g, we may extend ¥ over E(F'). This latter extension is possible since g fixes F M F' pointwise. The resulting extension ¢ is an f-labeling of the edges of C. From Theorem 2.13 it follows that f: C —» X satisfies the reciprocity law, and the induction is complete.$4. Reducibility Criteria 4.1 Definition An unoriented corner, or korner, of a combinatorial 2-complex X is an equivalence class of corners under the relation y ~ ¥. If D is a face of X we define the degree d(D) of D to be the degree of a boundary cycle of D. Hence d(D) is the number of korners incident with the interior D of D. If vy is a korner, we write y < D to indicate that v is incident with D. If C is a combinatorial cell structure on S* and x € cO, we define d(x) to be the valence of x. That is, d(x) is the number of korners of 2 cells of C at x. It can also be defined as the number of oriented edges of the 1-skeleton of C whose initial vertex is x. We d= write y < x to indicate the korner ¥ is a korner at x. 42 Definition A weight on a combinatorial 2-complex X is a real valued function on korners of X. 4.3 Example If X is a tesselation of a part of a Riemannian 2-manifold M by geodesic polygons, we may take w(y) to be the angle determined by the korner v, 0 < w(v) < 2x.4.4 Proposition Let C be a combinatorial cell structure on $2, Then no weight w on C can satisfy simultaneously 44.1 = wr) > 2, and x¥< xect® 44.2 = wy) < d(D) - 2. x p for all x € C and d(D) > q for all faces D of C. 4.6 Corollary (Lyndon [16])There are no (p,q) maps on S? if (p,q) = (6,3), (4,4) or (3,6). Proof. Let w be the constant function at constant value 1/3, 1/2, or 2/3 in each of the three cases respectively, and apply Proposition 4.4. 4.7 Theorem (Weight Test) , Suppose a combinatorial 2-complex X has a weight w satisfying 4.7.1 for each non-trivial reduced circuit z in the link complex E£ of X we have £ w(y) 2 2; here we write y , then KI) isdiagrammatically reducible. 4.14 Remark As in the case of Sieradski's coloring test, subcomplexes and coverings of complexes satisfying the hypotheses of the weight test also satisfy the weight test. In addition connected sums of 2-complexes which pass the weight test also pass the weight test (for the definitions of connected sum, see Sieradski [20], page 105). The coloring test is not invariant under subdivision, although it is so under barycentric subdivision. We shall see in &6 that the weight test is invariant under arbitrary subdivision, hence is a combinatorial invariant. We'd like to point out a variant on Proposition 4.12 which is of interest since one of the hypotheses is satisfied in applications to knots and to ribbon disc complements [10]. 4.15 Proposition Let @ be the Adian presentation ® = X has both a source vertex and a sink vertex. Then Remmers’ argument with T) or T,, applies. Denoting the group of the presentation ®@ by G, we see that the association a 1 1, a € A, extends to a homomorphism G — Z and hence determines an infinite cyclic cover X—2s xX. Since C is simply connected, f: C — X lifts a diagram f: C —> ¥. We give X the pull-back orientation on X!!) via p. Now the vertices P,, n € Z, of X are indexed by Z and for each a € A there is a unique edge a, of X connecting Py to Poel with p(a,) =a. Since C is compact, f(C) is compact, so let m=mintn € z21P, €Imf> and let M=maxCn € Z1P, € Im f}. It follows that P,, is a source vertex of f(C) and Py, is a sink. Since m < M and since P,, and Pyy are in f(C), it follows that C has both a source vertex and a sink vertex.4.16 Example In Sieradski's paper [20] he considers a spine for the square knot complement K(®) where @ = . He wasn’t able to decide where K(®) was diagrammatically reducible, although another 2-complex, equivalent to K(@) by 2- and 3- moves, was. We rewrite ® as an Adian presentation @ = . The left graph of @ is a tree, although the right graph has a cycle. It follows from 4.15 that K(®) is diagrammatically reducible. 4.17 Remark Diagrammatic reducibility is not an invariant of 2- and 3- moves. This is already clear from the dunce hat example 3.2. A more convincing example with cyclically reduced relators is the well known presentation of the trivial group © = which is not diagrammatically reducible (in fact, it can be shown that the reciprocity law fails for K(®)), although K(®) is equivalent by 2- and 3~ moves to K(@') where ®' = (that is, @ is "Andrews-Curtis trivial"). 4.18 Remark Corollary 4.6, due to Lyndon, can be used to prove the diagrammatic reducibility criteria arising in small cancellation theory [16]. The result is the following. Suppose @ = is a presentation where each relator R; is a cyclically reduced word in the alphabet A V A. Suppose no R; is a proper power and no R; is conjugate (in the free group) to Rj or R;| G# i. Suppose in addition that the symmetrized presentation obtained from @ by adding cyclic conjugates of the R, and Rj! satisfies the small cancellation condition C(p) and T(q) where (p,q) = (6,3), (4,4), or (3,6). Then K(®) is diagrammatically reducible.4.19 Example Let X= K(®) where @ = , For n = 2 or 3, it is diagrams can be constructed. For n 2 4, K(#4,) is known to be aspherical and %, presents an infinite group with no nontrivial finite quotients. We shall see that K(%,) is diagrammatically reducible for n2 4. The link complex = has been sketched below with weights indicated on its edges. In addition we've shown one 2-cell of K(#,) with the weights in the korners. (*)G. Higman, A finitely generated infinite simple group, J. London Math. Soc. 26 (1951), 61-64.Observe that 4.7.2 is satisfied, since for each 2-cell D, =_w(ly) = 3 = d(D) - 2. Every circuit of = with at least four Gdges has weight > 4(1/2) = 2. If a circuit of Z has exactly three edges, it must contain a segment connecting x; to X, which is given weight 1 (this is where n 2 4 is used). From this it follows that 4.7.1 is satisfied. Hence DR(K(¥,)) is satisfied for n > 4.$5. Extension Theorem and Staggered 2-Complexes 5.1 Theorem Suppose X is a sub-complex of the combinatorial 2-complex Y, where X and Y are both connected. Suppose Y//X is defined and RL{X) and RL(Y//X) are both satisfied. Then RL(Y) is satisfied. Proof. By identifying all vertices (Y)) to a single point, we may assume X and Y have a unique vertex. This does not change the link complexes of X or Y nor does it change the collection of spherical diagrams in X or Y. Let G = Fy and choose an orientation on the edges of Y. Let T be the free group on the set of oriented edges of X and let U be the free group on the set of oriented edges of Y - X. Let G— H be the natural homomorphism with H = G*T/>, The corners involved in w,, e € x2), are those of X. Next let H— K be the natural homomorphism with K = HPU/<>, Here n: Fy,;y —* H is defined as follows. Let e be a 2-cell of Y-X with attaching map fy: $)— ¥", Let I be a maximal oriented subinterval of S} mapped into X by f,. Observe that I corresponds to a unique corner y of Y//X. We associate to y the product in order of corners of L and oriented edges of X which occur in I in the boundary cycle of e.The accompanying diagram may clarify the definition of 7. The interval I is mapped to t,*f)+€, in the picture whereas x and y are in YD _ x) 79 the corner ¥ of Y//X we associate the coset of the word arty bef)-c-f-d in H. Observe that the composite homomorphism G —» H — K is canonically identified with the natural map G > GT*U/>. Suppose now that 84,8,...8,€ G are such that 1€ 0 eK. Since wx Banta the reciprocity law, we deduce from 2.8.3 that 1é a ei On the other hand X satisfies the reciprocity law, so 2.8.3 implies that 1 € a e)°. Proposition 2.8 then tells us that RL(Y) is satisfied. This completes the proof. 5.2 We review now some fundamental work of Howie's [9]. Howie defines a combinatorial 2-complex X to be staggered if there are linear orderings in the {unoriented) 1- and 2-cells of X, denoted <, such that (St1) For each 2 cell a, the attaching map Wa of ais a cyclically reduced word in the free semigroup W of oriented 1-cells of X, and (St2) If a and & are 2 cells with a < 4, then max a < max & and min a < min 4 where max a (resp. min a) denotes the largest (resp. smallest) 1-cell occurring in wa. Howie show that any subcomplex or Z-cover of a staggered 2-complex is staggered. In addition, if K is a finite staggered 2-complex with HK) = 0 in which no 2-cell is attached by a proper power (so Wq # x" for any x € W, n 2 2) then K is collapsible, [9] Corollary 3.1. Howie's argument actually shows more, which is worthstating. 5.3 Lemma (Howie) Let K be a finite staggered 2-complex with H(K) = 0. Let a n be a 2-cell of K and write wy = x4 with x, € W and ng maximal (so x, is not a proper power). Then (max alt! occurs precisely once in xq, where max a is the largest 1-cell occurring in Wa In other words, max a occurs ng times in wq with all occurrences of the same sign. In addition, if a, is the largest 2-cell of K, then K' = K - Ca, max d,? is a subcomplex of K satisfying H1(K') = 0. This gives an inductive method for constructing K. 5.4 emma Let @ = with n#0. Then K(®) satisfies the reciprocity law. Proof. This result is Proposition 5 of [6]. It was proved using the explicit nature of F. Levin's solution of the equation aytast...a,t = 1 over the group G (ay,a9,....a, € G). 5.5 Theorem Let X be a staggered 2-complex. Then RL(X) is satisfied. Proof. Let f: C —+ X be a spherical diagram and let f,: C — K be a maximal tower lift [11] of f where a tower is a composite of Z-covers and inclusions of finite subcomplexes. It suffices to establish the reciprocity law for K, by functionality (2.14). However K is a finite connected staggered complex with HK) = 0, by maximality. Let aj , for some n; #0. Hence K;,4//K; satisfies the reciprocity law by 5.4. By Theorem 5.1 and induction we deduce that K satisfies the reciprocity law and hence also X does. This completes the proof. 5.6 Corollary Let © = where the relator R is a cyclically reduced non-trivial word in the alphabet AV A. Then K(®) satisfies the reciprocity law. Proof. Clearly K(®) is staggered, where the ordering of 1-cells is arbitrary subject only to the restriction that the largest 1-cell occurs in R. 5.7 Corollary (S.J. Pride, Unpublished) Let G be a group and let WE GR (where tytg,..t, freely generate a free group) be such that w remains cyclically reduced (and non-trivial) when all “coefficients” of w in G are set equal to 1. Then the equation w= 1 is solvable in an overgroup of G. Proof. Let W be the cyclically reduced word in ty,tz....,t, obtained by setting the coefficients of w equal to 1. If @ = Sty tgs ety 1 Ww, then K(@) satisfies the reciprocity law by Corollary 5.7. The result follows from 2.8.3 in the special case r = 1. 5.8 Example Here is an application of Theorem 5.1 to a complex which is not staggered. Consider the presentation @ of an F-group ({16], page 126): O = 2 for all i, and where either n = 2g and F - eee : oe Eygiy¥ai] or n =e and a = yyy¥Q..¥e. With theexception of free products of cyclic groups every Fuchsian group has a presentation of type ®. Then Y = K(@) satisfies the reciprocity law if n > 0. To see this let X be the subcomplex of Y associated to the presentation = @"_= , X = K(@’). Clearly X is staggered. Also Y//X exists aa is the 2-complex associated to one of the presentations rs < i > OP = HV gVag |W [yg p¥2i) g 2 . or @" = Vor Vg 1M y;>, where g>0 since n> 0, In either is case 5.6 applies, so Y//X satisfies the reciprocity law. Hence RL(Y) is satisfied by Theorem 5.1.$6. Subdivision Theorems 6.1 Let X be a combinatorial 2-complex and let X' be a subdivision of X, so every cell of X' is contained in a unique cell of X. We wish to investigate the effect of the subdivision on the weight text, diagrammatic reducibility and on the reciprocity law. We point out at the outset that Howie has shown that DR(X) <> DR(X’) (oral communication). The argument makes use of “diamond moves” [3], page 161, to eliminate ramification at vertices introduced by subdivision. Let f: C —+ X be a spherical diagram in X. Then f determines by subdivision a diagram f': C' — X'. 6.2 Theorem If f' satisfies the reciprocity law (2.4), then so does f. If f satisfies the reciprocity law and if H)(X) = 0, then f' also satisfies the reciprocity law. A vertex of C’ or X' which is not vertex of C or X will be called a new vertex. Vertices of C or X will be called old vertices. 6.3 Lemma These are homomorphisms ¢: Fy —* Fy, p: Fy: —* Fy such that (a) <(£,(f) = £,(f) for each old vertex v of C and conjugacy class £,(f). (b) plat, tf") = 1 for each new vertex v' of c’. (c) pel y(f) = £,(f) for each old vertex v of C. Proof. Let y be a corner (oriented) of X. Under subdivision the corner ¥ is replaced by, say, n corners ee in order. These may be defined by choosing a suitable arc representing y (see$2) which intersects each new edge at the corner ¥ once transversely. We define cv) = ,72..7,. Observe that e(v7! = (¢x7}, so £ extends to a homomorphism ¢: Fy — Fy. It is a straightforward matter to check now that c(,(f)) = £,(€') from the definitions, Choose a free basis ¥4,¥p,...,.¥, of oriented corners for Fy. If u(y) = where ¥5, are defined as above, we set ply j,) =; and ply y,) = 1 for k > 1. If 8° is a comer of X’ incident on a new vertex, we set (5')= 1. Thus p extends to a homomorphism : Fy. — Fy. It is now a_ straightforward calculation to check conclusions (b) and (c) of Lemma 6.3. We can now prove the first assertion of Theorem 6.2. If 1€ 1 ef) = 0 ef) Mm kf"), v' vertex aula vold Y vt new of ct we may apply 2 to get 1€ Of plti(f') + a Py (E") = a? fy (£)) viold vt v ol = r # (f. Consequently, if f' satisfies the Feciprocity law, v vertex of C then so does f. 6.4 Suppose now that a is an oriented 2-cell of X whereHX) = 0. Let x be a point of a and let U be an open disc neighborhood of x entirely contained in a. We define a map a: X— 82 by collapsing X-U to a point and selecting a homeomorphism of X/(X-U) with s?. If we compose f: C —+ X with a: X— 8%, we get a map nf: S%—+ 8? which has atx) as a regular value. Since H,(X) = 0, this map nf has degree 0. Consequently if f(x) = Cyy¥q00¥_2 where y; is in the 2-cell a; of C (so a; # aj for i # i, since f is combinatorial), then as many of the cells €a;3 are mapped onto a by f with positive orientation as with negative orientation. These remarks from the geometrical basis of the next result. 6.5 Lemma We have 7. #,-(f') 3 1, where v" runs over new vertices of C' which are interior to 2-cells of C. Proof. If x" is a new vertex of X' interior to the 2 cell a of X, then x" has a Euclidean neighborhood in X'. Letting £,.. denote the conjugacy class of the product of corners of x'' in order around x" in one circuit, we have for each vertex y € fx"), £(f) = £514, with as many + signs occuring as - signs, as y varies. The lemma follows immediately. 6.6 Suppose that y is a new vertex of C’ which lies on edge (1 cell) @ of C. Then & is incident to two 2-cells a, and a of C (it can happen that a, = a). If we consider the product of those corners of C’ in a; incident with y, we see that #y(E) is the conjugacy class of an element of Fy: of the form vy * ¥p, where ¥; is the product of corners of X' in f(a;) incident with the vertex fly) = x. If we fix x and let y vary in f(x), the argument of 6.4 shows that as many factors y, occur with positive exponent as occur with negative exponent. Thus we shall have 1 € ney {{'), where y runs over all new vertices of C’ which lie on an edge of C, provided we can establish the following result about free groups. 6.7 LemmaLet Py,Pp,..P, be elements of a free group F where 8 Py = Yai mi: ei a 8; = 41. Here VyVgrY, are pairwise non-conjugate elements of F. Assume that for each j, the total number of occurrences of ‘Yj among the p; with exponent +1 equals the total number of occurrences of Yj; with exponent -1. Then a 1é @ OF. isl Proof. Assume that vj; occurs in p, with exponent + 1 and in pg with exponent -1,_ ik # &. Thus py ~ vZ'y; and Py ~ ante 80 pyPy 3 YZ'y2'. We may now replace the collection p4,P2....P, by the collection py,...Bj By Po (where "fi," means delete p, from the collection), where p, = vs'v2!. This reduces the number of elements by 1 and preserves the hypothesis of 6.7. The proof can now be completed by downward induction on n. 6.8 We are able now to complete the proof of Theorem 6.2. Assume that HA{X) = 0 and that f satisfies the reciprocity law. Write tr ef) = MW RY OLE) oO LE), Meverees et eateia va igen ty even of ct where v' runs over new vertices of C' which lie on edges of C and v" runs over new vertices of C' which lie in 2-cells of C. By Lemma 6.7, 1€ 1 #,(f). By Lemma 6.5, 1€ 0 fyulf). By ' vee v Lemma 6.3 (a), 0 fF) = ¢( z,(f). But we are v old V v vertex % of C assuming 1€ OO &,(f), so it follows that v old 1é€ 0 £,(f'). This completes the proof of the theorem. v vertex of c! 6.9 Theorem Let X' be a subdivision of the combinatorial 2-complex X satisfying H9(X) = 0. Then RL(X) <> RIX’).Proof. The implication RL(X') > RL(X) follows from Theorem 6.2. The converse is proved by analyzing an arbitrary subdivision as a composite of subdivisions of three types: (1) adding a new vertex to an edge, (2) adding a new vertex v in a 2 cell a and a new edge connecting v to a vertex of X, and (3) adding a new edge in a 2 cell a connecting two old vertices (possibly the same). The subdivision of type (3) only subdivides the link complex, so adds no new cycles. In addition it suffices to prove the reciprocity law for reduced diagrams in X' as one checks by a simple computation. From these two observations it's clear that RL(X) > RL{X') if X' is obtained from X by a type (3) subdivision since the reduced diagrams in X correspond 1-1 to those in X' here. Suppose X' is obtained from X by a type (2) subdivision as indicated in the figures below in which the 2 cell a is subdivided: Let f: C —> X' be a spherical diagram. The difficulty here is that f may ramify over v, in the sense that if y € f1(v), it's possible for the link of y in C to be mapped by a proper covering to the link of v in X’ (both links are circles). If f is unramified over v, then f is obtained from a diagram in X by subdivision, and Theorem 6.2 applies. Our strategy is to reduce to the unramified case by a “superdiamond move" and keep track of the reciprocity law in the process. Suppose then that y € f-1(v) and that the ramification index of f at v is n> 1. We've drawn a neighborhood of y in C below for n = 3. The edges of C over eat y are labeled Cpl grsOy: The diagram above contains also corner labels. Here the corner ¥ of a at v' is subdivided into v', y" in X'. The symbols L; in the diagram refer to the product of corner labels in Fy: in order, so for example Lyv'y" is the vertex label of f at the initial point of the edge &, 1<¢is DR(X’). One assumes f: C — X' is a reduced diagram. One does a huge superdiamond move on f to obtain a new diagram f': C' —+ X' which is unramified over all subdivision vertices of X', hence f' arises from a diagram in X. The point is to check that f reduced implies that f’ is also reduced. But this contradicts the assumption that all diagrams in X are reducible. 6.11 We assume now that the combinatorial 2-complex X has a weight w satisfying 4.7.1 and 4.7.2. An arbitrary subdivision of X is obtained by iteration of three operations, (1) adding a vertex to an edge, (2) adding a vertex v to a face D and an edge in D connecting v to a vertex of D, and (3) adding an edge in D connecting two vertices (possibly the same) of D. 6.12 Examine the effect of adding a new vertex v to an edge e of X to get X’. If a is a face with e* in w, (the attaching map of a), then a new korner y is created in a at v. We set w'(y) = 1, and w' = w on kornors of X. Since any non-trivial cyclein the link of v in X' must meet two such new corners at v, we see that 4.7.1 is satisfied for w'. On the other hand the subdivision has added one edge to the boundary cycle of the 2-cell a, and one new corner with weight 1. It follows that 4.7.2 is satisfied for w’. 6.13 Suppose a is a face of X and a new vertex v is introduced in the interior of a and a new edge e in a connecting v to the vertex v' of X The effect is to subdivide the korner y of a at v' into two new korners, say ¥' and ¥", and to create a new korner 5 at v. We set w'(8) = 2 and w'(e) = w(e) for all korners € # ¥ of X. The degree of the boundary cycle of a has been increased by 2 in the subdivision. In addition the effect on the link complex of X has been to add a vertex of valence 2. In particular any circuit without backtracking in the link complex of X' which contains ¥' must contain y" an equal number of times. Let w'(v') and w'(v") be chosen arbitrarily so that w'(v') + w'(v") = w(y). It follows from the remarks just made and from the choice of w'(5) = 2 that 4.7.1 and 4.7.2 will be satisfied for w'. 6.14 Suppose now a new edge e is added to the interior of a face a connecting distinct vertices v and v,Eo ~S In this case a is subdivided into 2 new faces a’ and a”. The Vv 1 he korners y and Y, of a . v, and Vv respectively are epee each into korners v',¥" and Wy respectively where y' and ‘; are in a’ and y” and vl are in a". We set w'(e) = wie) for korners € # ¥, € + v' of X. We must simultaneously satisfy the inequalities 6.14.1 wily’) + wv") = wly). 6.14.2 wiry) + wry) = wry) 6.14.3 wir’) + wir,) +a < dia’) - 2 6.14.4 wir") + wry) +b < dla") - 2, where az = wie) and ob = we). € korner of € korner of X in at X in a" Observe that d(a’) + d(a") = d(a) + 2, so addition of 6.14.3 and 6.14.4 gives { wie) < d(a) - 2, where 6.14.1 and 6.14.2 have been used. We claim the four inequalities 6.14.1-6.14.4 are consistent. To see this, choose w'(v') = 0, w'(v") = wy). Then 6.14.2-6.14.4 become respectively 6.14.2, 6.14.5 wir, )+a d(a') - 2, and6.14.6 wiv) + wiry) +b < d(a") - 2. Define wr,) = dla’) = a = 2, so 6.14.2 forces wiry) = wiry) +a + 2-- d(a’), We have only to check that 6.14.6 is satisfied, i.e. that 6.14.7 wy) + wl) + a+ 2 - dia’) +b ¢ d(a") - 2 But 6.14.7 is equivalent to the inequality = wle) < d(a) - 2 e DR(Y). If in addition H)(X) = 0, then RL(X) = RL{Y). Proof. The second barycentric subdivisions X" and Y" of X and Y respectively are simplicial complexes. By the Hauptvermutung for 2-complexes [17], X" and Y” have isomorphic simplicial subdivisions. The result follows from 6.9 and 6.10. 6.18 Proposition Let X be a subcomplex of the combinatorial 2-complex Y where X and Y are connected and Y is aspherical. If Y//X is homeomorphic to a closed 2-manifold M where M is not S* then X is aspherical. Proof. The 2-manifold M is homeomorphic to K(®) where @ is a certain well known 1-relator presentation. If M is not IP, then K(@) is diagrammatically reducible, for example by the weight test. It follows from 6.17 that Y//X is diagrammatically reducible and hence satisfies the reciprocity law by Theorem 3.3. If M is IRP2, then K(@) satisfies the reciprocity law by 5.4 and has Hy = 0. It follows from 6.17 that Y//X satisfies the reciprocity law. In either case RL(Y//X) is satisfied, so X is aspherical by Theorem 2.15. 6.19 Remark It is possible to give a direct cut-and-paste argument for 6.18. It is not possible to resist the temptation to observe that if 6.18 could be proved when Y/X is $2, then the special case of Whitehead’s conjecture (that subcomplexes of aspherical 2-complexes are aspherical [23]) for finite aspherical 2-complexes would follow as a corollary.6.20 Speculandum I'd like to close with an open question related to the Whitehead problem and to a class of examples that are not well understood, the ribbon disc complement presentations. I claim no originality here, since Howie first introduced me to these examples, yet I have my own peculiar point of view. In [10] Howie associates to each ribbon disc complement a label oriented tree T and a group presentation (I). A label oriented tree is a finite graph T in the sense of Serre which is a tree, with an orientation O on the edge set E(r) and a labeling function X: O — Vir), where V(I) is the vertex set. The presentation @(T) associated to T is the Adian presentation OT) = . Here i,7: E(t) —> V(r) are the initial and terminal vertex functions. It is easy to see that adjoining any vertex to the relations of @(Ir) gives a balanced presentation of the trivial group, hence a finite contractible 2-complex. Thus Whitehead's conjecture predicts that K(@()) is aspherical if T is a label-oriented tree. In general there are examples where K(@(F)) is not diagrammatically reducible. However all the examples I know arise from graph theoretic pathologies. It may happen that for some edge e € O either A(e) = ife) or d(e) = Te). In this case we collapse the edge e, identifying <¢(e) with 7(e), to get a presentation equivalent to @(f) by Andrews-Curtis moves. Or it may happen that there are two oriented edges e # e’ with i(e) = i(e’) (respectively 7(e) = 7(e')) and d(e) = Ae’). In these cases we may fold the edges e and e', identifying r(e) with 7(e') and again get a smaller label oriented tree whose presentation is Andrews-Curtis equivalent to or) I call TF reduced if neither of these two possibilities occurs (ie., no edge collapses nor folds are possible as just described). I should like to ask then whether a K(@(I)) is diagrammatically reducible for a reduced label-oriented tree f. The evidence is meagre. The study of these presentations has beenplagued by errors in the literature. Howie's own methods [10] break down distressingly early, for trees of diameter greater than three. It's still conceivable that this problem can be treated by diagrammatic techniques and will shed some light on Whitehead's conjecture.APPENDIX HYPERBOLIC 2-COMPLEXES Al. In this appendix we assume that X is a finite combinatorial 2-complex with a weight w (4.2) satisfying 4.7.1, and, in place of 4.7.2, the stronger inequality. All = wir) < d(a) - 2 for each face A of X, and in addition 0 < w(y) < 1 for all korners ¥ of X (in the angle interpretation, achieved by replacing w by xw, the last condition means all angles are less than a straight angle; A1.1 corresponds to negative curvature). Let f: D—> X be a reduced van Kampen diagram in X. Thus D is a connected and simply connected planar region and f is a combinatorial map. “Reduced” has the same meaning as in 3.1. We assume that D is not a single point. Then D together with the complementary cell ¢,, at infinity, e = s? - D, give a combinatorial structure on S%, We define d(D) to be d(e,,), the degree of the attaching map of e,,. Our goal is to prove Theorem A. There is a positive constant A depending only on X and w such that for any reduced van Kampen diagram f: D — X one has # < adi. This result yields a solution to the word problem for 74(X,x) by the following well known argument. If a circuit ¢ in X" at the base point x represents the identity in 1,(X,x), we represent c as a boundary circuit of a reduced van Kampen diagram f: D —+ X. Since #0) < Ad(D) = Alle), and since there are only finitely many such diagrams with a bounded number of vertices, we may construct a Turing machine to enumerate these diagrams and check whether c is a boundary circuit of one of them. We remark that a result similar to Theorem A was announcedby Gromov at the group theory seminar at MSRI in June 1985, in the ~ context of his hyperbolic groups. We find it convenient to introduce some terminology. A2. Definition. A hyperbolic structure on the combinatorial 2-complex X is a pair (X,w) where w is a weight satisfying 4.7.1, Al1.1, and 0 < w(y) < 1 for all korners y of X. There is a notion of morphism g: (X,w) —+ (X1,wy) of hyperbolic structures where g is a combinatorial map and e*(w,) = w. A finite covering map p: X — X induces a morphism (X,p*(w)) 2+ (X,w). Observe also that if f: D— X is a reduced van Kampen diagram, then (D,f*(w)) is a , hyperbolic structure and f: (D,f*(w)) —> (X,w) is a morphism. A3. Lemma. If (X,w) is a hyperbolic structure, then there is a weight w’ on X', the first derived subdivision of X, so that (X',w') is a ‘ hyperbolic structure. Proof. If y is a korner at thé O-cell x of X, then y is subdivided into two korners ¥,; and Yo in X’. We set w'(v;) = 1/2 wiv), i= 1,2. Observe 0 < wily) < 1/2. If & is a korner of X' at the barycenter of an edge of X, set w'(4) = 1/2. Next let v" be the barycenter of a face D of X. Let K(D) = d(D) - 2 - ve) > 0, the "defect" of D, and define €(D) = min (1-800, KO >0. Ifa is a korner of X' at v", then a determines a unique triangle A of X' containing a korner 4 at the barycenter of an edge of X and a korner +; of X' at the vertex of X.We set w(a)= 1/2 - w'(y;) - €(D) > 0. Observe that w'(a) + ws) + wily) <1 = (A) - 2, proving Al1 for w’. Also z= wig) == (1/2 - wily) - €(D)) = a(D) - = wr) a 1, X sufficiently close to 1, (X,Aw) is also a hyperbolic structure. Choose one such X > 1. Then it follows from 4.7.1 there is an € > 0 such that Z dAwly) 2 2+ €, where c is a non-trivial reduced circuit in £. We call the structure (X,Aw) adequate if this latter condition is satisfied, and the number € > 0 is called the adequacy of (X,dw). Summarizing, we have shown A5. Lemma. If (X,w) is a hyperbolic structure, then there is a number X > 1 such that (X,Aw) is an adequate hyperbolic structure. A6. Theorem ("Isoperimetric inequality"): Let (U,W) be an adequate hyperbolic structure on the finite simplical complex U which is a compact 2-manifold. Let ¢ > 0 be the adequacy of U. Then @ v< (425]v., - & where V = #0), v, = #(0U), and where x = x(U) is the Buler characteristic of U. Proof: Let S= x wv) = = zw) (2+ Wing, where Ving is the number of interior vertices of U. ii int Hence AGL. (2 + Wig < F Now V-E+F= x and 3F = 2E - E,, = 2E - V,,, where E (resp. E,,) denotes the number of edges of U (resp. OU) and where V (resp. V..) denotes the number of vertices of U (resp. OU). Eliminating E gives AG.2. F = -2x + 2V - V,. Combining A6.1 with A6.2 yields the desired result, v«< (48)v,, - 2% A.7. We can now prove Theorem A. If (X,w) is a hyperbolic structure and f:D— X is a reduced van Kampen diagram, + barycentrically subdivide twice to obtain a simplical map f": D” — X". = By A3 and A4, X" has an adequate hyperbolic structure (X",w,) and let € > 0 be its adequacy. If D were a disc we'd deduce that ree oo where V" is #(D") and where V., = d(D"). From this it follows that 9D) < (4E£}a0D) if D is a disc. The case of general van Kampen diagrams is easily reduced to the case of discs, with constant A = 4(4£5) in the statement of Theorem A. 8. Corollary: If (X,w) is a hyperbolic 2~complex, then the corresponding presentation of y([Link]) (for xq € Xl) has solvable word and conjugacy problems. For the word problem one applies Theorem A. For the conjugacy problem one considers reduced maps of an annulus into Xand one applies the isoperimetric inequality A6 (with x = 0). A9. Example. The standard examples of 2-complexes possessing hyperbolic structures are the usual presentations of closed surfaces which admit Riemannian metrics of constant negative curvature--that is, all except S, RP, T?, and the Klein bottle. It is an amusing exercise to associate to these presentations the weight functions which give hyperbolic structures in our sense (A2). This example goes back to Dehn(*). A9. Remark. It would be of interest if #(D )) could be bounded by a recursive function of d(D) for a reduced van Kampen diagram f: D —»* X, where X is a combinatorial 2-complex possessing a weight w satisfying 4.7.1 and 4.7.2 and in addition 0 < w(v) < 1 for all korners y of X. Then, for example, one could apply Dehn's algorithm to Adian presentations. In general a linear bound is not possible. Al0. Remark. It would be of interest if one could find a finite aspherical 2-complex whose fundamental group had an unsolvable word problem. The proof of the Higman imbedding theorem shows that a group G which has an aspherical recursive presentation imbeds in a finitely presented group H for which the Eilenberg MacLane space K(H,1) has a finite dimensional model. But, a priori, the dimension is greater than 2. : All. Example. Let n 2 3 and let aj,ap,...,a, be integers no one of which is zero. Then K(®) has a hyperbolic structure where a, a a 1 (42 n. Ixy) Xy% x, >. @ = is such that each R; is cyclically reduced and no conjugate of R; equals R, for j# i. It is allowed, however, that R be a proper power. Thus X = K() is not aspherical in general. A piece p is a reduced word which occurs as an initial segment of cyclic conjugates of R, and R,i if i = j we demand that these cyclic conjugates not be identical. We say that @ satisfies the small cancellation condition C'(1/6) if whenever the piece p occurs as an initial segment of some cyclic conjugate ¢ of Ry, then £(p) < ¢ 2(c). One can define similarly the condition C'(d) for any d > 0. Suppose that f: M —+ K(@) is a reduced diagram, where © is a C'(1/6) presentation and M is a compact surface. We shall describe a new cell structure M on the underlying manifold M by removing all internal vertices of M of valence 2. One must check that M is a cell complex; that is, each new edge begins and ends at a vertex (the open 2-cells are unchanged). One then checks the following assertions: (1) M and M have the same boundary vertices, (2) each internal vertex of M has valence > 3, and (3) each face D of M has at least 7 sides counted correctly (i.e., dy,(D) 2 7). This last assertion follows since each cyclic conjugate of a relator R, is a product of no fewer than 7 pieces. Next one assigns the constant weight w = 2/3 to each corner of M. One checks (4) (Mw) is a hyperbolic structure.The isoperimetric inequality hence applies to (M,w). Since ®M is unchanged in passing from M to M, this implies a solution of the word problem and conjugacy problem for *,(X,x). Postscript: Lyndon wrote me that he had made use of the idea of assigning angle measures to corners in his article "On the combinatorial Riemann-Hurwitz formula,” Symposia Mathematica XVII (1976), 435-439. He uses it to calculate the area u(G) of a fundamental domain for the Fuchsian group G and to show that {G:H] = s(H)/u(G) where H is a subgroup of finite index in G. Addendum Since this article was written nearly two years ago, it's worth attempting to bring up to date (August 1986). First, the condition that Hp(X)=0 can be dropped from Theorems 6.2 and 6.9. The reason is a 2-complex X satisfying the reciprocity law is automatically Cockcroft (recall that X is called Cockcroft if the Hurewicz map x>(X) —+ H>(X) is zero.) It's the fact that X is Cockcroft that makes the argument in 6.4 including the "Thom map” work. Second, we've given a homotopy theoretic interpretation for diagrammatic reducibility (S.M. Gersten, “Branched coverings of 2-complexes and diagrammatic reducibility,” preprint): DR(X) is satisfied iff every finite branched covering of X is Cockcroft. Third, Steve Brick (Thesis, University of California at Berkeley, 1986) has carried over many of the results of the paper, including the subdivision theorem, to develop a parallel theory of “Kervaire complexes." Brick calls a 2-complex X "Kervaire" if all systems of equations over all coefficient groups G, whose words in the variable letters are modeled on the attaching maps of X, are solvable in an overgroup of G. This notion promises to be useful in discussing equations over groups by virtue of Brick's results. Finally, Chuck Miller informed that he and Don Collins have constructed a sequence of groups