On infinitely generated homology of Torelli groups
Abstract.
Let be the Torelli group of an oriented closed surface of genus , that is, the kernel of the action of the mapping class group on the first integral homology group of . We prove that the th integral homology group of contains a free Abelian subgroup of infinite rank, provided that and . Earlier the same property was known only for (Bestvina, Bux, Margalit, 2007) and in the special case (Johnson, Millson, 1992). Moreover, we show that the group contains a free Abelian subgroup of infinite rank generated by Abelian cycles and we construct explicitly an infinite system of Abelian cycles generating such subgroup. As a consequence, we obtain that an Eilenberg–MacLane CWcomplex of type cannot have a finite skeleton. The proofs are based on the study of the Cartan–Leray spectral sequence for the action of on the complex of cycles constructed by Bestvina, Bux, and Margalit.
Key words and phrases:
Torelli group, homology of groups, complex of cycles, Abelian cycles, Cartan–Leray spectral sequence2010 Mathematics Subject Classification:
20F34 (Primary); 57M07, 20J05 (Secondary)1. Introduction
Let be an oriented closed surface of genus . Recall that the mapping class group of is the group
where is the group of orientation preserving homeomorphisms of on itself. The action of the mapping class group on the first integral homology group of preserves the intersection form and yields the surjective homomorphism
The kernel of this homomorphism is called the Torelli group of and is denoted by .
It is a classical result that , and so is trivial, cf. [FaMa12, Theorem 2.5]. Mess [Mes92] proved that is an infinitely generated free group. (The fact that is not finitely generated was earlier proved by McCullough and Miller [MCM86].) Johnson [Joh83] showed that for , the group is finitely generated, and described explicitly a finite set of generators. Nevertheless, the structure of the Torelli groups , where , is still rather poorly understood.
An interesting and important problem is to study the homology of the Torelli groups , . For the sake of simplicity, we denote the homology groups with integral coefficients simply by . The group , i. e., the Abelianization of was computed by Johnson [Joh85b]. He showed that
and described explicitly the structure of an module on this group. For , the problem of explicit computation of seems to be extremely hard. So the reasonable questions are:

Which groups are trivial and which are not?

Which groups are finitely generated and which are not?
One of the first results towards the latter question was obtained by Akita [Aki01] who showed that for , the total rational homology group
is an infinitelydimensional vector space. For , it is known that either of the groups and contains a free Abelian subgroup of infinite rank, see [Mes92] and [Hai02], respectively.
In 2007 Bestvina, Bux, and Margalit [BBM07] developed a new method for studying the homology of the Torelli groups based on the Cartan–Leray spectral sequence for the action of on a special contractible cell complex called the complex of cycles. They proved that the cohomological dimension of is equal to and the topdimensional homology group is not finitely generated. (Recall that the cohomological dimension of a group is the largest integer for which there exists a module such that .) Moreover, the proof in [BBM07] actually yields that contains a free Abelian subgroup of infinite rank, see Remark 2.5 below. Until now, for no intermediate dimension , it has been known whether the group is finitely generated or not, except for the above mentioned case due to Mess. The main result of the present paper is as follows.
Theorem 1.1.
Suppose that and . Then the group is not finitely generated. Moreover, contains a free Abelian subgroup of infinite rank.
Recall the definition of an Abelian cycle. Let be pairwise commuting elements of a group . Consider the homomorphism that sends the generator of the th factor to for every . We denote by the image of the standard generator of the group under the homomorphism . Homology classes are called Abelian cycles.
Vautaw [Vau02] proved that the Torelli group does not contain a free Abelian subgroup of rank greater than . Since is torsionfree, this result implies that the group contains no nontrivial Abelian cycles, provided that . For , we can refine Theorem 1.1 in the following way.
Theorem 1.2.
Suppose that . Then the group contains a free Abelian subgroup of infinite rank whose generators are Abelian cycles.
Remark 1.3.
For , Theorem 1.1 exactly recovers the already mentioned result by Johnson and Millson (unpublished, see [Mes92]) that contains a free Abelian subgroup of infinite rank. Nevertheless, Theorem 1.2 seems to be a new result even for genus . For each pair such that and any , the assertion of Theorem 1.1 is a new result.
Remark 1.4.
Our proof of Theorem 1.2 is constructive, which means that we shall construct explicitly an infinite set of linearly independent Abelian cycles in . (Recall that elements of an Abelian group are linearly independent if they form a basis for a free Abelian subgroup of this group.) Notice also that this construction does not use the result of Bestvina, Bux, and Margalit [BBM07] claiming that the group is infinitely generated. On the contrary, our construction of an infinite set of linearly independent homology classes in , where , is in a sense a mixture of our construction of an infinite set of linearly independent Abelian cycles in and the Bestvina–Bux–Margalit description of an infinite set of linearly independent homology classes in . It is a debatable question whether the latter description can be regarded as an explicit construction. Our proof of Theorem 1.1 is constructive modulo this description.
The proofs of Theorems 1.1 and 1.2 are based on a more delicate study of the Cartan–Leray spectral sequence for the action of on the contractible complex of cycles constructed by Bestvina, Bux, and Margalit [BBM07]. Recall that this spectral sequence converges to the homology of , see Section LABEL:subsection_CL for details.
Remark 1.5.
Bestvina, Bux, and Margalit asked explicitly whether the groups are infinitely generated whenever (cf. Question 8.1 in [BBM07]), and suggested a possible approach towards obtaining the affirmative answer to this question. Though our main result (Theorem 1.1) is exactly the answer to this question, our proof does not follow the approach suggested in [BBM07]. Namely, Bestvina, Bux, and Margalit showed that the term of the Cartan–Leray spectral sequence for the action of on is infinitely generated, provided that , and concluded that if one could prove that this group remains infinitely generated after taking consecutive quotients by the images of the differentials , then he would obtain that is infinitely generated, since injects into . This plan seems to be very hard to realize because it requires computation of higher differentials of the spectral sequence. Instead, we show that every group , where , contains a free Abelian subgroup of infinite rank. This also implies that contains a free Abelian subgroup of infinite rank for , since the groups , , are consecutive quotients for certain filtration in . The crucial point in our approach is that we avoid explicit computation of higher differentials by mapping the Cartan–Leray spectral sequence to certain auxiliary spectral sequences in which the corresponding differentials are trivial by dimension reasons.
Kirbi’s list of problems in lowdimensional topology [Kir97] contains the following Problem 2.9(B) attributed to Mess: Given , what is the largest for which admits a classifying space with finite skeleton. (This is also Problem 5.11 in [Far06].) The best previously known estimate was obtained by Bestvina, Bux, and Margalit [BBM07]. The following is a direct consequence of Theorem 1.1.
Corollary 1.6.
If admits a classifying space with finite skeleton, then .
This paper is organized as follows. In Section 2 we describe explicitly an infinite linearly independent system of Abelian cycles in (Theorem 2.1) and an infinite linearly independent system of homology classes in for every (Theorem LABEL:theorem_explicit). The rest part of the paper contains the proof that these systems of homology classes are indeed linearly independent. The main tool is the Cartan–Leray spectral sequence for the action of on the complex of cycles . Section LABEL:section_background contains necessary information on the complex of cycles and on the Cartan–Leray spectral sequence. In the last three sections we study the Cartan–Leray spectral sequence for the action of on . In Section LABEL:section_aux_spectral we construct auxiliary spectral sequences and and morphisms of spectral sequences
This construction is needed to avoid explicit computation of higher differentials of the spectral sequence . In Sections LABEL:section_Abelian_proof and LABEL:section_main_proof we prove Theorems 2.1 and LABEL:theorem_explicit, respectively.
The author is grateful to S. I. Adian and A. L. Talambutsa for fruitful discussions.
2. Systems of linearly independent homology classes
A multicurve on the surface is a union of a finite number of pairwise disjoint simple closed curves such that no curve is homotopic to a point and no two curves and are homotopic to each other. A multicurve is said to be oriented if all components of are endowed with orientations. (Notice that, by definition, an oriented multicurve is not allowed to contain a pair of components that are homotopic to each other with the homotopy either preserving or reversing the orientation.)
Most constructions in the present paper concerning curves and multicurves are determined up to isotopy. For the sake of simplicity, we shall later not distinguish between notation for a curve (or a multicurve) and its isotopy class. For instance, the subgroup of consisting of all mapping classes that stabilize the isotopy class of a multicurve will be denoted by and will be called the stabilizer of the multicurve .
We denote by the homology class of an oriented simple closed curve in . We denote by the algebraic intersection index of homology classes . We denote by the subgroup of generated by homology classes . We denote by the (left) Dehn twist about a simple closed curve .
We choose a primitive element and fix it throughout the whole paper.
2.1. System of linearly independent Abelian cycles
Consider a splitting
(2.1) 
where every has rank and is orthogonal to with respect to the intersection form unless . Then there is a unique decomposition
(2.2) 
Let be the set of all splittings of form (2.1) such that all summands in the corresponding decomposition (2.2) are nonzero. Splittings obtained from each other by permutations of summands are regarded to be different, i. e., a splitting is an ordered tuple . Obviously, the set is infinite.
If a splitting belongs to , then every summand in decomposition (2.2) can be uniquely written as , where is primitive and . We denote by the set of all tuples such that and for every . (Notice that we do not choose elements and in and .)
It is easy to see that for each pair such that and , there exists a component multicurve , where
that satisfies the following conditions (see Fig. 1):

are separating simple closed curves,

consists of connected components such that

and are oncepunctured tori adjacent to and to , respectively,

for , is a twicepunctured torus adjacent to and ,


the image of the homomorphism induced by the inclusion is ,

and are oriented simple closed curve contained in ,

,

the connected component of that is adjacent to lies on the righthand side from .
It is easy to check that all multicurves satisfying these conditions for the fixed pair lie in the same orbit. Therefore the homology class
is well defined and independent of the choice of .
For and , we denote by and the tuples obtained from and by reversing the orders of ’s and ’s, respectively, i. e., , . It is easy to see that
Theorem 2.1.
Let be a subset containing exactly one splitting in every pair . Choose any representatives , where . Then the Abelian cycles , where runs over , are linearly independent in .
2.2. On the group
Consider an oriented compact surface of genus with one boundary component. Let be the mapping class group of , where is the group of all homeomorphisms that are identical on the boundary. Let be the corresponding Torelli group, that is, the kernel of the natural surjective homomorphism .
We shall need the following proposition, which is a reformulation of a result by Bestvina, Bux, and Margalit [BBM07].
Proposition 2.2.
Suppose that . Then and the top homology group contains a free Abelian subgroup of infinite rank.
Proof.
Bestvina, Bux, and Margalit considered the Torelli group for the oriented surface of genus with one puncture rather than one boundary component. (However, they denoted this group by rather than by .) Their result is precisely as follows: and the top homology group is infinitely generated, see Theorem C and Lemma 8.8 in [BBM07]. However, in fact, their proof yields a stronger result, namely, that contains a free Abelian subgroup of infinite rank. Indeed, their proof is by induction on and is based on the following two facts:
Fact 2.3.
for .
Fact 2.4.
injects into for .
The former fact is contained in the proof of Lemma 8.8 in [BBM07]. The latter fact can be extracted from [BBM07] in the following way. First, by Lemma 8.10 in [BBM07], the group , where is a nonseparating simple closed curve on , injects into . Second, in the proof of Theorem C in [BBM07], a torsion free Abelian group is constructed such that the group injects into . Finally, by the construction, is a subgroup of the first homology of a free group, hence, is free Abelian. Therefore injects into .
By Fact 2.3, . Since is an infinitely generated free group (cf. [Mes92]), we obtain that is a free Abelian group of infinite rank. Using Facts 2.3 and 2.4, we obtain by induction that contains a free Abelian subgroup of infinite rank.
Now, consider the short exact sequence
(2.3) 
Here the subgroup is generated by the Dehn twist about a simple curve isotopic to the boundary of , hence, is central in . Since , the Hochschild–Serre spectral sequence for central extention (2.3) yields the natural isomorphism , which completes the proof of the proposition. ∎
Remark 2.5.
The above proof also yields that the group , where , contains a free Abelian subgroup of infinite rank.
In the sequel, we shall need the following proposition, which is equivalent to Fact 2.4.
Proposition 2.6.
Let be a nonseparating simple closed curve on , where . Then any embedding induces an injection .
Proof.
We have the commutative diagram with exact rows:
(2.4) 
Here the first row is exact sequence (2.3), the second row is the restriction of the Birman exact sequence obtained in [BBM07, Lemma 8.3], and is the commutator subgroup of , where is a oncepunctured surface of genus obtained from by forgetting one of the two punctures. Obviously, is a free group. The group in the first row is generated by the Dehn twist , where is a simple curve isotopic to the boundary of . The element represents a nontrivial homology class in , see [BBM07, Lemma 8.6]. Since is torsionfree, we obtain that the homomorphism is an injection. This injection induces the homomorphism
(2.5) 
where is endowed with the structure of an module corresponding to the second row of (2.4). (Since lies in the centre of , the action of on the coefficient group is trivial.) It is proved in [BBM07, proof of Theorem C] that (2.5) is an injection. (Notice that the submodule that is constructed in [BBM07, Lemma 8.7] and used in the proof of Theorem C in [BBM07] is exactly .)
Finally, consider the Hochschild–Serre spectral sequences for the rows of (2.4). Since and , these spectral sequences yield the commutative diagram