This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Mapping class groups of circle bundles over a surface

Lei Chen  and  Bena Tshishiku Lei Chen
Department of Mathematics
University of Maryland
4176 Campus Drive
College Park, MD 20742, USA
[email protected]
Bena Tshishiku
Department of Mathematics
Brown University
151 Thayer St.
Providence, RI, 02912, USA
bena_\_[email protected].
Abstract.

In this paper, we study the algebraic structure of mapping class group Mod(X)\operatorname{Mod}(X) of 3-manifolds XX that fiber as a circle bundle over a surface S1XSgS^{1}\to X\to S_{g}. There is an exact sequence 1H1(Sg)Mod(X)Mod(Sg)11\to H^{1}(S_{g})\to\operatorname{Mod}(X)\to\operatorname{Mod}(S_{g})\to 1. We relate this to the Birman exact sequence and determine when this sequence splits.

1. Introduction

For g1g\geq 1, let SgS_{g} denote the closed oriented surface of genus gg, and for kk\in\mathbb{Z}, let XgkX_{g}^{k} denote the closed 3-manifold that fibers

S1XgkSgS^{1}\to X_{g}^{k}\to S_{g}

as an oriented circle-bundle with Euler number kk. Assuming (g,k)(1,0)(g,k)\neq(1,0), the mapping class group Mod(Xgk):=π0(Homeo+(Xgk))\operatorname{Mod}(X_{g}^{k}):=\pi_{0}\big{(}\operatorname{Homeo}^{+}(X_{g}^{k})\big{)} fits into a short exact sequence

(1) 1H1(Sg;)Mod(Xgk)Mod(Sg)1.1\rightarrow H^{1}(S_{g};\mathbb{Z})\rightarrow\operatorname{Mod}(X_{g}^{k})\rightarrow\operatorname{Mod}(S_{g})\rightarrow 1.

This paper is motivated by the following question.

Question 1.1.

For which values of g,kg,k is the extension in (1) split?

Interestingly, the extension does split for k=22gk=2-2g, in which case XgkX_{g}^{k} is unit tangent bundle USgUS_{g}. In fact, there is a natural action of Mod(Sg)\operatorname{Mod}(S_{g}) on USgUS_{g} by homeomorphisms, which gives a splitting of (1) upon taking isotopy classes. For g2g\geq 2, this action comes from the action of the punctured mapping class group Mod(Sg,1)\operatorname{Mod}(S_{g,1}) on triples of points on the boundary of hyperbolic space 2\mathbb{H}^{2}. This construction dates back to the work of Nielsen. See [FM12, §5.5.4, §8.2.6] and [Sou10, §1].

In general, Question 1.1 reduces to a question about group cohomology. The extension (1) splits if and only if its Euler class eukH2(Mod(Sg);H1(Sg;))eu_{k}\in H^{2}\big{(}\operatorname{Mod}(S_{g});H^{1}(S_{g};\mathbb{Z})\big{)} vanishes [Bro82, §IV.3]. Here the coefficients are twisted via the natural action of Mod(Sg)\operatorname{Mod}(S_{g}) on H1(Sg;)H^{1}(S_{g};\mathbb{Z}). However, a computation of H2(Mod(Sg);H1(Sg;))H^{2}\big{(}\operatorname{Mod}(S_{g});H^{1}(S_{g};\mathbb{Z})\big{)} does not appear to be in the literature.

The extension (1) is related to the Birman exact sequence

1π1(Sg)Mod(Sg,1)Mod(Sg)1.1\to\pi_{1}(S_{g})\to\operatorname{Mod}(S_{g,1})\to\operatorname{Mod}(S_{g})\to 1.

By taking quotients by the commutator subgroup π[π1(Sg),π1(Sg)]\pi^{\prime}\equiv[\pi_{1}(S_{g}),\pi_{1}(S_{g})], we obtain the following extension

(2) 1H1(Sg)Mod(Sg,1)/πMod(Sg)1.1\to H_{1}(S_{g})\to\operatorname{Mod}(S_{g,1})/\pi^{\prime}\to\operatorname{Mod}(S_{g})\to 1.

Our main result relates the sequences (1) and (2).

Theorem A.

Fix g1g\geq 1 and kk\in\mathbb{Z}. Assume (g,k)(1,0)(g,k)\neq(1,0). There is a map between the short exact sequences (1) and (2)

1\textstyle{1}H1(Sg)\textstyle{H_{1}(S_{g})}Mod(Sg,1)/π\textstyle{\operatorname{Mod}(S_{g,1})/\pi^{\prime}}Mod(Sg)\textstyle{\operatorname{Mod}(S_{g})}1\textstyle{1}1\textstyle{1}H1(Sg;)\textstyle{H^{1}(S_{g};\mathbb{Z})}Mod(Xgk)\textstyle{\operatorname{Mod}(X_{g}^{k})}Mod(Sg)\textstyle{\operatorname{Mod}(S_{g})}1\textstyle{1}kδ\textstyle{k\delta}

The homomorphism kδk\delta is the Poincaré duality isomorphism δ\delta composed with multiplication by kk. In particular, when k=1k=1, the exact sequences (1) and (2) are isomorphic.

Theorem A implies the Euler classes of the extensions (1) satisfy euk=keu1eu_{k}=k\>eu_{1} for fixed gg. Next we determine the subgroup generated by eu1eu_{1} in H2(Mod(Sg);H1(Sg;))H^{2}\big{(}\operatorname{Mod}(S_{g});H^{1}(S_{g};\mathbb{Z})\big{)}.

Theorem 1.2.

Fix g1g\geq 1, and let eu1eu_{1} be the Euler class of the extension (1). Then eu1eu_{1} has order 2g22g-2 in H2(Mod(Sg);H1(Sg;))H^{2}\big{(}\operatorname{Mod}(S_{g});H_{1}(S_{g};\mathbb{Z})\big{)}. Furthermore, if g8g\geq 8, then eu1eu_{1} generates this group, i.e.

H2(Mod(Sg);H1(Sg;))/(2g2).H^{2}\big{(}\operatorname{Mod}(S_{g});H^{1}(S_{g};\mathbb{Z})\big{)}\cong\mathbb{Z}/(2g-2)\mathbb{Z}.

Combining Theorem A and Theorem 1.2 we obtain the following answer to Question 1.1.

Corollary 1.3.

For g2g\geq 2 and kk\in\mathbb{Z}, the extension (1) splits if and only if kk is divisible 2g22g-2. For g=1g=1 the extension splits for each kk.

When a splitting exists, the different possible splittings (up to the action of H1(Sg;)H^{1}(S_{g};\mathbb{Z}) on Mod(Xgk)\operatorname{Mod}(X_{g}^{k}) by conjugation) are parameterized by elements of H1(Mod(Sg);H1(Sg;))H^{1}\big{(}\operatorname{Mod}(S_{g});H^{1}(S_{g};\mathbb{Z})\big{)} [Bro82, Ch. IV, Prop. 2.3]. This group vanishes for g1g\geq 1 [Mor85, Prop. 4.1], so the splitting, when it exists, is unique.


Connection to Nielsen realization. Instead of Question 1.1, one can ask whether there is a splitting of the composite surjection

Homeo(Xgk)Mod(Xgk)Mod(Sg).\operatorname{Homeo}(X_{g}^{k})\to\operatorname{Mod}(X_{g}^{k})\to\operatorname{Mod}(S_{g}).

This is an instance of a Nielsen realization problem. Of course, if Mod(Xgk)Mod(Sg)\operatorname{Mod}(X_{g}^{k})\to\operatorname{Mod}(S_{g}) does not split, then neither does Homeo(Xgk)Mod(Sg)\operatorname{Homeo}(X_{g}^{k})\to\operatorname{Mod}(S_{g}), and Corollary 1.3 gives examples of this. Since Mod(Sg)\operatorname{Mod}(S_{g}) has a natural action on USgUS_{g}, the surjection Homeo(Xgk)Mod(Sg)\operatorname{Homeo}(X_{g}^{k})\to\operatorname{Mod}(S_{g}) does split for k=±(2g2)k=\pm(2g-2). This is somewhat surprising since mapping class groups are rarely realized as groups of surface homeomorphisms [Mar07, Che19, CS22]. We wonder whether this splitting is unique, or if a splitting exists for other values kk divisible by 2g22g-2 (for example, k=0k=0). We plan to study this in a future paper.


Previous work and proof techniques. Waldhausen [Wal68, §7] proved that the group π0(Homeo(Xgk))\pi_{0}\big{(}\operatorname{Homeo}(X_{g}^{k})\big{)} is isomorphic to the outer automorphism group Out(π1(Xgk))\operatorname{Out}\big{(}\pi_{1}(X_{g}^{k})\big{)}. From this, the short exact sequence (1) can be derived from work of Conner–Raymond [CR77] and the Dehn–Nielsen–Baer theorem; alternatively, see McCullough [McC91, §3]. The Dehn–Nielsen–Baer theorem also plays a central role in Theorem A, since it allows us to translate back and forth between topology and group theory. There is a mix of both in the proof of Theorem A in §3.

To prove Theorem A, we consider a version of Question 1.1 where XgkX_{g}^{k} and SgS_{g} are punctured. For the punctured manifolds, similar to (1), there is a short exact sequence

1H1(Sg;)Mod(Xg,1k)Mod(Sg,1)1,1\to H^{1}(S_{g};\mathbb{Z})\to\operatorname{Mod}(X_{g,1}^{k})\to\operatorname{Mod}(S_{g,1})\to 1,

and we construct a splitting

σ:Mod(Sg,1)Mod(Xg,1k).\sigma:\operatorname{Mod}(S_{g,1})\to\operatorname{Mod}(X_{g,1}^{k}).

See Corollary 3.1. A key part of our proof of Theorem A is to determine the image of the point-pushing subgroup π1(Sg)<Mod(Sg,1)\pi_{1}(S_{g})<\operatorname{Mod}(S_{g,1}) under σ\sigma. For this we relate three natural surface group representations π1(Sg)Mod(Xg,1k)\pi_{1}(S_{g})\to\operatorname{Mod}(X_{g,1}^{k}) that appear in the following diagram, where the diagonal map is point pushing on XgkX_{g}^{k} (not a commutative diagram).

π1(Sg)\textstyle{\pi_{1}(S_{g})}Mod(Sg,1)\textstyle{\operatorname{Mod}(S_{g,1})}H1(Sg;)\textstyle{H^{1}(S_{g};\mathbb{Z})}Mod(Xg,1k)\textstyle{\operatorname{Mod}(X_{g,1}^{k})}point-pushing on SgS_{g}σ\textstyle{\sigma}transvections

See Proposition 3.4 for a precise statement.

In order to deduce Corollary 1.3, we use a spectral sequence argument to prove that eu1eu_{1} generates a subgroup of H2(Mod(Sg);H1(Sg;))H^{2}\big{(}\operatorname{Mod}(S_{g});H^{1}(S_{g};\mathbb{Z})\big{)} isomorphic to /(2g2)\mathbb{Z}/(2g-2)\mathbb{Z}. A different spectral sequence computation proves that eu1eu_{1} generates H2(Mod(Sg);H1(Sg;))H^{2}\big{(}\operatorname{Mod}(S_{g});H^{1}(S_{g};\mathbb{Z})\big{)} when gg is large. These computations use several known computations, including work of Morita [Mor85].


Section outline. In §2 we collect the results we need about the manifolds XgkX_{g}^{k} and their mapping class groups, including Waldhausen’s work. Theorem A is proved in §3; this section is the core of the paper. In §4, we do two spectral sequence computations to prove Theorem 1.2.


Acknowledgement. Thanks to B. Farb for sharing the reference [McC91] and to D. Margalit for comments on a draft. The authors are supported by NSF grants DMS-2203178, DMS-2104346 and DMS-2005409.

2. Circle bundles over surfaces

Here we review some results about circle bundles over surfaces that we will need in future sections.

2.1. Classification

By an oriented circle bundle we mean a fiber bundle

S1EBS^{1}\to E\to B

with structure group Homeo+(S1)\operatorname{Homeo}^{+}(S^{1}), the group of orientation preserving homeomorphisms of S1S^{1}. The inclusion of the rotation group SO(2)\operatorname{SO}(2) in Homeo+(S1)\operatorname{Homeo}^{+}(S^{1}) is a homotopy equivalence, so circle bundles are in bijection with rank-2 real vector bundles. The classifying space BSO(2)B\operatorname{SO}(2) is homotopy equivalent to P\mathbb{C}P^{\infty}, which is an Eilenberg–Maclane space K(,2)K(\mathbb{Z},2). Thus each circle bundle is uniquely determined up to isomorphism by its Euler class eu(E)H2(B;)eu(E)\in H^{2}(B;\mathbb{Z}), which is the primary obstruction to a section of the bundle.

When B=SgB=S_{g} is a closed, oriented surface, H2(Sg;)H^{2}(S_{g};\mathbb{Z})\cong\mathbb{Z}, we can speak of the Euler number. We use XgkX_{g}^{k} to denote the total space of the circle bundle

S1XgkSgS^{1}\to X_{g}^{k}\to S_{g}

with Euler number kk. For example, for the unit tangent bundle eu(USg)=22geu(US_{g})=2-2g (the Euler characteristic), so USgXg22gUS_{g}\cong X_{g}^{2-2g}. We also note that XgkX_{g}^{k} and XgkX_{g}^{-k} are homeomorphic 3-manifolds, since the sign of the Euler number of a circle bundle over SgS_{g} depends on the choice of orientation on SgS_{g}.

2.2. Fundamental group 𝝅𝟏(𝑿𝒈𝒌)\pi_{1}(X_{g}^{k}) and its automorphisms

From the long exact sequence of a fibration, we have an exact sequence

1π1(Xgk)π1(Sg)1.1\to\mathbb{Z}\to\pi_{1}(X_{g}^{k})\to\pi_{1}(S_{g})\to 1.

The group π1(Xgk)\pi_{1}(X_{g}^{k}) has a presentation

(3) π1(Xgk)=A1,B1,,Ag,Bg,zz central, [A1,B1][Ag,Bg]=zk.\pi_{1}(X_{g}^{k})=\big{\langle}A_{1},B_{1},\ldots,A_{g},B_{g},z\mid z\text{ central, }[A_{1},B_{1}]\cdots[A_{g},B_{g}]=z^{k}\big{\rangle}.

Using this, one finds z\langle z\rangle\cong\mathbb{Z} is the center of π1(Xgk)\pi_{1}(X_{g}^{k}) as long as (g,k)(1,0)(g,k)\neq(1,0). When g2g\geq 2 follows from the fact that the group π1(Sg)\pi_{1}(S_{g}) has trivial center; the case g=1g=1 can be treated directly.

Given this computation of the center, any automorphism of π1(Xgk)\pi_{1}(X_{g}^{k}) induces an automorphism of z\langle z\rangle\cong\mathbb{Z} and descends to an automorphism of π1(Sg)\pi_{1}(S_{g}). The latter gives a homomorphism

Aut(π1(Xgk))Aut(π1(Sg))\operatorname{Aut}\big{(}\pi_{1}(X_{g}^{k})\big{)}\to\operatorname{Aut}\big{(}\pi_{1}(S_{g})\big{)}

that restricts to an isomorphism between the inner automorphism groups

(4) Inn(π1(Xgk))π1(Sg)Inn(π1(Sg))\operatorname{Inn}\big{(}\pi_{1}(X_{g}^{k})\big{)}\cong\pi_{1}(S_{g})\cong\operatorname{Inn}\big{(}\pi_{1}(S_{g})\big{)}

and hence descends to a homomorphism

(5) Out(π1(Xgk))Out(π1(Sg)).\operatorname{Out}\big{(}\pi_{1}(X_{g}^{k})\big{)}\to\operatorname{Out}\big{(}\pi_{1}(S_{g})\big{)}.

Orientations. It will be convenient to define

𝒜𝒰𝒯(π1(Sg))<Aut(π1(Sg))\operatorname{\mathcal{AUT}}\big{(}\pi_{1}(S_{g})\big{)}<\operatorname{Aut}\big{(}\pi_{1}(S_{g})\big{)}

as the subgroup that acts trivially on H2(π1(Sg);)H_{2}(\pi_{1}(S_{g});\mathbb{Z})\cong\mathbb{Z} (the “orientation-preserving” subgroup). We define

𝒜𝒰𝒯(π1(Xgk))<Aut(π1(Xgk))\operatorname{\mathcal{AUT}}\big{(}\pi_{1}(X_{g}^{k})\big{)}<\operatorname{Aut}\big{(}\pi_{1}(X_{g}^{k})\big{)}

as the group of automorphisms that project into 𝒜𝒰𝒯(π1(Sg))\operatorname{\mathcal{AUT}}\big{(}\pi_{1}(S_{g})\big{)} and that act trivially on the center z\langle z\rangle\cong\mathbb{Z}. In particular, 𝒜𝒰𝒯(π1(Xgk))\operatorname{\mathcal{AUT}}\big{(}\pi_{1}(X_{g}^{k})\big{)} has index 4 in Aut(π1(Xgk))\operatorname{Aut}\big{(}\pi_{1}(X_{g}^{k})\big{)}.

These orientation-preserving subgroups contain the (respective) inner automorphism groups, and we denote the quotients 𝒪𝒰𝒯(π1(Xgk))\operatorname{\mathcal{OUT}}\big{(}\pi_{1}(X_{g}^{k})\big{)} and 𝒪𝒰𝒯(π1(Sg))\operatorname{\mathcal{OUT}}\big{(}\pi_{1}(S_{g})\big{)}.

2.3. Mapping class group 𝐌𝐨𝐝(𝑿𝒈𝒌)\operatorname{Mod}(X_{g}^{k})

Fix g1g\geq 1 and kk\in\mathbb{Z}, and assume (g,k)(1,0)(g,k)\neq(1,0). Let Homeo+(Xgk)\operatorname{Homeo}^{+}(X_{g}^{k}) denote the group of homeomorphisms whose image in Out(π1(Xgk))\operatorname{Out}\big{(}\pi_{1}(X_{g}^{k})\big{)} is contained in 𝒪𝒰𝒯(π1(Sg))\operatorname{\mathcal{OUT}}\big{(}\pi_{1}(S_{g})\big{)}. Define

Mod(Xgk):=π0(Homeo+(Xgk)).\operatorname{Mod}(X_{g}^{k}):=\pi_{0}\big{(}\operatorname{Homeo}^{+}(X_{g}^{k})\big{)}.

Waldhausen [Wal68, Cor. 7.5] proved that the natural homomorphism

π0(Homeo(Xgk))Out(π1(Xgk))\pi_{0}\big{(}\operatorname{Homeo}(X_{g}^{k})\big{)}\to\operatorname{Out}\big{(}\pi_{1}(X_{g}^{k})\big{)}

is an isomorphism. Then, by the definitions, this homomorphism restricts to an isomorphism Mod(Xgk)𝒪𝒰𝒯(π1(Xgk))\operatorname{Mod}(X_{g}^{k})\cong\operatorname{\mathcal{OUT}}\big{(}\pi_{1}(X_{g}^{k})\big{)}. Waldhausen also proved that π0Homeo(Xgk)\pi_{0}\operatorname{Homeo}(X_{g}^{k}) is isomorphic to the group of fiber-preserving homeomorphisms modulo homeomorphisms that are isotopic to the identity through fiber-preserving isotopies; see [Wal68, Rmk. following Cor. 7.5]. Consequently, there is a homomorphism

(6) Mod(Xgk)Mod(Sg).\operatorname{Mod}(X_{g}^{k})\to\operatorname{Mod}(S_{g}).

Altogether, we have the following commutative diagram relating the maps (5) and (6).

(7) Mod(Xgk)\textstyle{\operatorname{Mod}(X_{g}^{k})}Mod(Sg)\textstyle{\operatorname{Mod}(S_{g})}𝒪𝒰𝒯(π1(Xgk))\textstyle{\operatorname{\mathcal{OUT}}\big{(}\pi_{1}(X_{g}^{k})\big{)}}𝒪𝒰𝒯(π1(Sg))\textstyle{\operatorname{\mathcal{OUT}}\big{(}\pi_{1}(S_{g})\big{)}}\textstyle{\cong}\textstyle{\cong}

The right vertical map is an isomorphism by the Dehn–Nielsen–Baer theorem [FM12, Thm. 8.1]. Furthermore, by Conner–Raymond [CR77, Thm. 8] that there is a short exact sequence

(8) 1Hom(π1(Sg),)𝒪𝒰𝒯(π1(Xgk))𝒪𝒰𝒯(π1(Sg))1.1\to\operatorname{Hom}(\pi_{1}(S_{g}),\mathbb{Z})\to\operatorname{\mathcal{OUT}}\big{(}\pi_{1}(X_{g}^{k})\big{)}\to\operatorname{\mathcal{OUT}}\big{(}\pi_{1}(S_{g})\big{)}\to 1.

This establishes the short exact sequence (1) in the introduction. We will give a concrete derivation of this exact sequence in Corollary 3.1 below.

3. Relating Mod(Xgk)\operatorname{Mod}(X_{g}^{k}) to the Birman exact sequence

In this section, we prove Theorem A. To construct the map of short exact sequences in Theorem A, our main task is to first define a homomorphism Mod(Sg,1)Mod(Xgk)\operatorname{Mod}(S_{g,1})\to\operatorname{Mod}(X_{g}^{k}) and then to compute that its kernel is the commutator subgroup of π1(Sg)<Mod(Sg,1)\pi_{1}(S_{g})<\operatorname{Mod}(S_{g,1}) (the point-pushing subgroup). We do this is §3.1 and §3.2.

3.1. A homomorphism 𝚿:𝐌𝐨𝐝(𝑺𝒈,𝟏)𝐌𝐨𝐝(𝑿𝒈𝒌)\Psi:\operatorname{Mod}(S_{g,1})\to\operatorname{Mod}(X_{g}^{k})

Fix a basepoint Sg*\in S_{g}, and set Sg,1=Sg{}S_{g,1}=S_{g}\setminus\{*\}. By the Dehn–Nielsen–Baer theorem, Mod(Sg,1)\operatorname{Mod}(S_{g,1}) is isomorphic to Out(F2g)\operatorname{Out}^{*}(F_{2g}), where F2gF_{2g} is the free group of rank 2g2g and Out(F2g)<Out(F2g)\operatorname{Out}^{*}(F_{2g})<\operatorname{Out}(F_{2g}) is the subgroup that preserves the conjugacy class corresponding to the free homotopy class of the curve around the puncture in Sg,1S_{g,1}. We construct Ψ\Psi as a composition

(9) Ψ:Mod(Sg,1)Out(F2g)𝜎𝒜𝒰𝒯(π1(Xgk))𝒪𝒰𝒯(π1(Xgk))Mod(Xgk).\Psi:\operatorname{Mod}(S_{g,1})\cong\operatorname{Out}^{*}(F_{2g})\xrightarrow{\sigma}\operatorname{\mathcal{AUT}}\big{(}\pi_{1}(X_{g}^{k})\big{)}\to\operatorname{\mathcal{OUT}}\big{(}\pi_{1}(X_{g}^{k})\big{)}\cong\operatorname{Mod}(X_{g}^{k}).

To define σ\sigma, fix a generating set α1,β1,,αg,βg\alpha_{1},\beta_{1},\ldots,\alpha_{g},\beta_{g} for F2gF_{2g} such that c=i=1g[αi,βi]c=\prod_{i=1}^{g}[\alpha_{i},\beta_{i}] represents the conjugacy class of the curve around the puncture. Let

(10) ι:F2gπ1(Xgk)\iota:F_{2g}\to\pi_{1}(X_{g}^{k})

be the homomorphism defined by αiAi\alpha_{i}\mapsto A_{i} and βiBi\beta_{i}\mapsto B_{i}. Given fOut(F2g)f\in\operatorname{Out}^{*}(F_{2g}), fix an automorphism f~:F2gF2g\tilde{f}:F_{2g}\to F_{2g} that represents ff, and assume that f~(c)=c\tilde{f}(c)=c (this can always be achieved by composing any lift with an inner automorphism of F2gF_{2g}). Next we define σ(f)\sigma(f) on generators of π1(Xgk)\pi_{1}(X_{g}^{k}) by

(11) σ(f)(Ai)=ιf~(αi),σ(f)(Bi)=ιf~(βi),σ(f)(z)=z.\sigma(f)(A_{i})=\iota\tilde{f}(\alpha_{i}),\>\>\>\sigma(f)(B_{i})=\iota\tilde{f}(\beta_{i}),\>\>\>\sigma(f)(z)=z.

To show that σ(f)\sigma(f) extends to a homomorphism of π1(Xgk)\pi_{1}(X_{g}^{k}), we check that the relation [A1,B1][Ag,Bg]=zk[A_{1},B_{1}]\cdots[A_{g},B_{g}]=z^{k} is preserved under σ(f)\sigma(f):

i[σ(f)(Ai),σ(f)(Bi)]=i[ιf~(αi),ιf~(βi)]=ι(c)=zk=σ(f)(zk).\prod_{i}[\sigma(f)(A_{i}),\sigma(f)(B_{i})]=\prod_{i}[\iota\tilde{f}(\alpha_{i}),\iota\tilde{f}(\beta_{i})]=\iota(c)=z^{k}=\sigma(f)(z^{k}).

The second equality uses the the fact that f~(c)=c\tilde{f}(c)=c. The map σ(f)\sigma(f) is independent of the choice of f~\tilde{f} because different choices of f~\tilde{f} differ by conjugation by powers of cc (because the centralizer of cc in F2gF_{2g} is the cyclic subgroup c\langle c\rangle)111Note that the centralizer is isomorphic to \mathbb{Z} and contains c\langle c\rangle. It is only bigger if c=uic=u^{i} for some uF2gu\in F_{2g} and i2i\geq 2. By contradiction, if c=uic=u^{i} for i2i\geq 2, then uu is cyclically reduced because cc is. This implies that uu is a subword of c=[αi,βi]c=\prod[\alpha_{i},\beta_{i}], which is absurd. and ι(c)=zk\iota(c)=z^{k} is central in π1(Xgk)\pi_{1}(X_{g}^{k}). The homomorphism σ(f):π1(Xgk)π1(Xgk)\sigma(f):\pi_{1}(X_{g}^{k})\to\pi_{1}(X_{g}^{k}) is an automorphism and belongs to 𝒜𝒰𝒯(π1(Xgk))\operatorname{\mathcal{AUT}}\big{(}\pi_{1}(X_{g}^{k})\big{)} by definition. Furthermore, fσ(f)f\mapsto\sigma(f) is a homomorphism, which is easy to check using the observation that if w=ιww=\iota w^{\prime}, then σ(f)(w)=ιf~(w)\sigma(f)(w)=\iota\tilde{f}(w^{\prime}).

Composing σ\sigma with 𝒜𝒰𝒯𝒪𝒰𝒯\operatorname{\mathcal{AUT}}\to\operatorname{\mathcal{OUT}} gives the desired homomorphism Ψ\Psi. As a corollary of this construction, we have proved the following.

Corollary 3.1.

Fix g1g\geq 1 and kk\in\mathbb{Z}, and assume (g,k)(1,0)(g,k)\neq(1,0). The natural map Φ:𝒜𝒰𝒯(π1(Xgk))𝒜𝒰𝒯(π1(Sg))\Phi:\operatorname{\mathcal{AUT}}\big{(}\pi_{1}(X_{g}^{k})\big{)}\to\operatorname{\mathcal{AUT}}\big{(}\pi_{1}(S_{g})\big{)} (see §2.2) fits into an exact sequence

(12) 1Hom(π1(Sg),)𝒜𝒰𝒯(π1(Xgk))Φ𝒜𝒰𝒯(π1(Sg))1,1\to\operatorname{Hom}(\pi_{1}(S_{g}),\mathbb{Z})\to\operatorname{\mathcal{AUT}}\big{(}\pi_{1}(X_{g}^{k})\big{)}\xrightarrow{\Phi}\operatorname{\mathcal{AUT}}\big{(}\pi_{1}(S_{g})\big{)}\to 1,

and this exact sequence splits.

Proof.

First we compute the kernel of Φ\Phi. Using the presentation for π1(Xgk)\pi_{1}(X_{g}^{k}) in (3), if fker(Φ)f\in\ker(\Phi), then

f(Ai)=Aizmi and f(Bi)=Biznif(A_{i})=A_{i}z^{m_{i}}\>\>\>\text{ and }\>\>\>f(B_{i})=B_{i}z^{n_{i}}

for some m1,n1,,mg,ngm_{1},n_{1},\ldots,m_{g},n_{g}\in\mathbb{Z}. The map aimia_{i}\mapsto m_{i}, binib_{i}\mapsto n_{i} extends to a homomorphism τ(f):π1(Sg)\tau(f):\pi_{1}(S_{g})\to\mathbb{Z}. It is elementary to check that the map ker(Φ)Hom(π1(Sg),)\ker(\Phi)\to\operatorname{Hom}(\pi_{1}(S_{g}),\mathbb{Z}) defined by fτ(f)f\mapsto\tau(f) is an isomorphism.

The homomorphism σ\sigma defined above shows that Φ\Phi is a split surjection. Note that the Mod(Sg,)Mod(Sg{})\operatorname{Mod}(S_{g},*)\cong\operatorname{Mod}(S_{g}\setminus\{*\}) (basepoint vs. puncture), so by Dehn–Nielsen–Baer there is an isomorphism 𝒜𝒰𝒯(π1(Sg))Out(F2g)\operatorname{\mathcal{AUT}}\big{(}\pi_{1}(S_{g})\big{)}\cong\operatorname{Out}^{*}(F_{2g}), and we use this isomorphism to view σ\sigma as a splitting of Φ\Phi. ∎

Remark 3.2.

We call elements of ker(Φ)H1(Sg;)\ker(\Phi)\cong H^{1}(S_{g};\mathbb{Z}) transvections.

Remark 3.3.

The homomorphism Ψ\Psi can be constructed on the level of topology as follows. Fix a regular neighborhood DD of the puncture on Sg,1S_{g,1} (so DD is a once-punctured disk). Given a mapping class fMod(Sg,1)f\in\operatorname{Mod}(S_{g,1}), choose a representing homeomorphism 𝔣\mathfrak{f}. Without loss of generality, we can assume that 𝔣\mathfrak{f} is the identity on DD. The bundle XgkSgX_{g}^{k}\to S_{g} can be trivialized over SDS\setminus D (because the classifying space BSO(2)B\operatorname{SO}(2) is simply connected). Fixing a trivialization (SD)×S1(S\setminus D)\times S^{1} over SDS\setminus D, we lift 𝔣\mathfrak{f} to the product homeomorphism 𝔣×idS1\mathfrak{f}\times\text{id}_{S^{1}}. This homeomorphisms is the identity on the boundary (SD)×S1\partial(S\setminus D)\times S^{1}, so we can extend by the identity to obtain a homeomorphism 𝔣~\tilde{\mathfrak{f}} of XgkX_{g}^{k}. The map sending fMod(Sg,1)f\in\operatorname{Mod}(S_{g,1}) to the isotopy class [𝔣~]Mod(Xgk)\left[\tilde{\mathfrak{f}}\right]\in\operatorname{Mod}(X_{g}^{k}) is the topological version of the homomorphism Ψ\Psi. Note that the isotopy class [𝔣][\mathfrak{f}] is only well-defined up to Dehn twists about D\partial D which is a loop around the puncture. This is analogous to the ambiguity encountered in the definition of σ\sigma, which ultimately does not affect the definition of Ψ\Psi.

Corollary 3.1 and equation (4) combine to give the short exact sequence of outer automorphism groups (8).


Warning. The splitting of the short exact sequence (12) does not give a splitting of the short exact sequence (8). Indeed we will show the latter sequence does not always split (Corollary 1.3). The subtlety comes from the fact that the inner automorphism group Inn(π1(Xgk))π1(Sg)\operatorname{Inn}\big{(}\pi_{1}(X_{g}^{k})\big{)}\cong\pi_{1}(S_{g}) does not coincide with the image of π1(Sg)Inn(π1(Sg))<Aut(π1(Sg))\pi_{1}(S_{g})\cong\operatorname{Inn}\big{(}\pi_{1}(S_{g})\big{)}<\operatorname{Aut}\big{(}\pi_{1}(S_{g})\big{)} under the section σ\sigma. Proposition 3.4 below describes the precise relationship.

3.2. Kernel of 𝚿:𝐌𝐨𝐝(𝑺𝒈,𝟏)𝐌𝐨𝐝(𝑿𝒈𝒌)\Psi:\operatorname{Mod}(S_{g,1})\to\operatorname{Mod}(X_{g}^{k})

Observe that the kernel of Ψ\Psi is contained in the point-pushing subgroup π1(Sg)<Mod(Sg,1)\pi_{1}(S_{g})<\operatorname{Mod}(S_{g,1}). This is because Ψ\Psi composed with the natural map Mod(Xgk)Mod(Sg)\operatorname{Mod}(X_{g}^{k})\to\operatorname{Mod}(S_{g}) is the natural map Mod(Sg,1)Mod(Sg)\operatorname{Mod}(S_{g,1})\to\operatorname{Mod}(S_{g}), whose kernel is the point-pushing subgroup. Thus we want to understand the image of the point-pushing subgroup under the section σ\sigma used to define Ψ\Psi. What we find is a simple relationship between three surface group representations:

π1(Sg)\textstyle{\pi_{1}(S_{g})}π1(Sg)\textstyle{\pi_{1}(S_{g})}π1(Sg)\textstyle{\pi_{1}(S_{g})}𝒜𝒰𝒯(π1(Xgk))\textstyle{\operatorname{\mathcal{AUT}}\big{(}\pi_{1}(X_{g}^{k})\big{)}}inner auts of π1(Sg)\pi_{1}(S_{g}), liftedinner auts of π1(Xgk)\pi_{1}(X_{g}^{k})σ\textstyle{\sigma}transvections

The main results are Proposition 3.4 and Corollary 3.5 below. In order to state Proposition 3.4, we need the following notation. Let

δ:H1(Sg;)H1(Sg;)\delta:H_{1}(S_{g};\mathbb{Z})\to H^{1}(S_{g};\mathbb{Z})

be the Poincaré duality map, given explicitly by γ,γ\gamma\mapsto\langle-,\gamma\rangle, where

,:H1(Sg;)×H1(Sg;)\langle-,-\rangle:H_{1}(S_{g};\mathbb{Z})\times H_{1}(S_{g};\mathbb{Z})\to\mathbb{Z}

is the algebraic intersection form. We use δ^\hat{\delta} denote the composition

δ^:H1(Sg;)𝛿H1(Sg;)𝒜𝒰𝒯(π1(Xgk)).\hat{\delta}:H_{1}(S_{g};\mathbb{Z})\xrightarrow{\delta}H^{1}(S_{g};\mathbb{Z})\hookrightarrow\operatorname{\mathcal{AUT}}\big{(}\pi_{1}(X_{g}^{k})\big{)}.

This map is given explicitly by δ^(γ)(w)=wz[w¯],γ\hat{\delta}(\gamma)(w)=w\cdot z^{\langle[\bar{w}],\gamma\rangle}, where w¯\bar{w} is the image of ww under π1(Xgk)π1(Sg)\pi_{1}(X_{g}^{k})\to\pi_{1}(S_{g}) and [w¯]H1(Sg;)[\bar{w}]\in H_{1}(S_{g};\mathbb{Z}) is the corresponding homology class.

Fix a basepoint Sg,1\star\in S_{g,1}. Recall that we have fixed a standard generating set {αi,βi}\{\alpha_{i},\beta_{i}\} of π1(Sg,1,)F2g\pi_{1}(S_{g,1},\star)\cong F_{2g} so that c:=i[αi,βi]c:=\prod_{i}[\alpha_{i},\beta_{i}] is a loop around the puncture * of Sg,1=Sg{}S_{g,1}=S_{g}\setminus\{*\}. Define

(13) Π:π1(Sg,1,)π1(Sg,)\Pi:\pi_{1}(S_{g,1},\star)\to\pi_{1}(S_{g},*)

by γϵ.γ.ϵ¯\gamma\mapsto\epsilon.\gamma.\bar{\epsilon}, where ϵ\epsilon is a fixed arc from * to \star.

Proposition 3.4.

Fix tπ1(Sg,)t\in\pi_{1}(S_{g},*), and let Push(t)Mod(Sg,1)Out(F2g)\operatorname{Push}(t)\in\operatorname{Mod}(S_{g,1})\cong\operatorname{Out}^{*}(F_{2g}) be the point-pushing mapping class. If t~π1(Sg,1,)\tilde{t}\in\pi_{1}(S_{g,1},\star) is any lift of tt (i.e. Π(t~)=t\Pi(\tilde{t})=t), then

(14) σ(Push(t))=Cιt~δ^([kt]),\sigma\big{(}\operatorname{Push}(t)\big{)}=C_{\iota\tilde{t}}\circ\hat{\delta}([kt]),

Here CxC_{x} denotes conjugation by xx, and the maps ι:F2gπ1(Xgk)\iota:F_{2g}\to\pi_{1}(X_{g}^{k}) and σ:Out(F2g)𝒜𝒰𝒯(π1(Xgk))\sigma:\operatorname{Out}^{*}(F_{2g})\to\operatorname{\mathcal{AUT}}\big{(}\pi_{1}(X_{g}^{k})\big{)} are defined in (10) and (11).

As a sanity check, observe that Cιt~C_{\iota\tilde{t}} does not depend on the choice of lift t~\tilde{t} because any two lifts differ by an element of the normal closure of cc in π1(Sg,1,)=F2g\pi_{1}(S_{g,1},\star)=F_{2g}, and conjugation by any such element is trivial on π1(Xgk)\pi_{1}(X_{g}^{k}).

Proof of Proposition 3.4.

It suffices to prove the lemma for tπ1(Sg,)t\in\pi_{1}(S_{g},*) that are represented by a non-separating simple closed curve. To see this, first note that π1(Sg,)\pi_{1}(S_{g},*) is generated by these curves. Furthermore, the groups Inn(π1(Xgk))\operatorname{Inn}\big{(}\pi_{1}(X_{g}^{k})\big{)} and H1(Sg;)H^{1}(S_{g};\mathbb{Z}) commute in Aut(π1(Xgk))\operatorname{Aut}\big{(}\pi_{1}(X_{g}^{k})\big{)}, so

[Cιt~1δ^([t1])][Cιt~2δ^([t2])]=Cι(t~1t~2)δ^([t1t2]).\big{[}C_{\iota\tilde{t}_{1}}\circ\hat{\delta}([t_{1}])\big{]}\circ\big{[}C_{\iota\tilde{t}_{2}}\circ\hat{\delta}([t_{2}])\big{]}=C_{\iota(\tilde{t}_{1}*\tilde{t}_{2})}\circ\hat{\delta}([t_{1}*t_{2}]).

Assume now that tπ1(Sg,)t\in\pi_{1}(S_{g},*) is represented by a non-separating simple closed curve. After an isotopy, we can assume that tt contains ϵ\epsilon as a sub-arc. Choose t~\tilde{t} as pictured in Figure 1.

\labellist
\pinlabel

* at 230 430 \pinlabel\star at 290 430 \pinlabeltt at 343 435 \pinlabelt~\tilde{t} at 343 459 \pinlabelϵ\epsilon at 260 425 \endlabellistRefer to caption

Figure 1. A small regular neighborhood of a loop representing tπ1(Sg,)t\in\pi_{1}(S_{g},*) and a lift t~π1(Sg,1,)\tilde{t}\in\pi_{1}(S_{g,1},\star).

We want to show that

σ(Push(t))(w)=[Cιt~δ^([t])](w)\sigma\big{(}\operatorname{Push}(t)\big{)}(w)=\big{[}C_{\iota\tilde{t}}\circ\hat{\delta}([t])\big{]}(w)

for each wπ1(Xgk)w\in\pi_{1}(X_{g}^{k}). Since this is obviously true for w=zw=z, it suffices to show this equality for w=ι(s)w=\iota(s) for sπ1(Sg,1,)s\in\pi_{1}(S_{g,1},\star); furthermore, it suffices to show the equality on any generating set of π1(Sg,1,)\pi_{1}(S_{g,1},\star). We use the (infinite) generating set consisting of curves of one of the forms pictured in Figure 2 (the intersection of these curves with the annulus around tt has one component).

\labellist
\pinlabel

cc at 235 237 \pinlabels1s_{1} at 105 337 \pinlabels2s_{2} at 263 337 \pinlabels3s_{3} at 495 300 \endlabellistRefer to caption

Figure 2. The group π1(Sg,1,)\pi_{1}(S_{g,1},\star) is generated by t~\tilde{t} and loops of the form pictured above.

Note that Push(t)\operatorname{Push}(t) fixes the basepoint \star, so we can compute the action of Push(t)\operatorname{Push}(t) on sπ1(Sg,1,)s\in\pi_{1}(S_{g,1},\star). We compute the action of Push(t)\operatorname{Push}(t) on the elements in Figure 2 as follows. See Figure 3 for an illustration.

s1(t~)1s1t~c1 and s2(t~)1s2t~ and s3c(t~)1s3t~c1 and cc.s_{1}\mapsto(\tilde{t})^{-1}s_{1}\tilde{t}c^{-1}\>\>\>\text{ and }\>\>\>s_{2}\mapsto(\tilde{t})^{-1}s_{2}\tilde{t}\>\>\>\text{ and }\>\>\>s_{3}\mapsto c(\tilde{t})^{-1}s_{3}\tilde{t}c^{-1}\>\>\>\text{ and }\>\>\>c\mapsto c.
\labellist
\pinlabel

Push(t)(s1)\operatorname{Push}(t)(s_{1}) at 100 100 \pinlabelPush(t)(s2)\operatorname{Push}(t)(s_{2}) at 263 100 \pinlabelPush(t)(s3)\operatorname{Push}(t)(s_{3}) at 465 60 \endlabellistRefer to caption

Figure 3. Action of point-pushing about tt on the loops in Figure 2. The curve cc is fixed up to isotopy (up to isotopy Push(t)\operatorname{Push}(t) is the identity on a neighborhood of tt that contains cc).

This proves that, for example, that

σ(Push(t))(ιs1)=(ιt~)1(ιs1)(ιt~)zk=[Cιt~δ^([kt])](ιs1).\sigma\big{(}\operatorname{Push}(t)\big{)}(\iota s_{1})=(\iota\tilde{t})^{-1}(\iota s_{1})(\iota\tilde{t})z^{-k}=\big{[}C_{\iota\tilde{t}}\circ\hat{\delta}([kt])\big{]}(\iota s_{1}).

We conclude similarly for the generators s2,s3s_{2},s_{3}. This proves the desired formula for σ(Push(t))\sigma\big{(}\operatorname{Push}(t)\big{)}. ∎

The following corollary is an immediate consequence of Proposition 3.4.

Corollary 3.5.

Consider the composition

(15) Ψ:𝒜𝒰𝒯(π1(Sg))𝜎𝒜𝒰𝒯(π1(Xgk))𝒪𝒰𝒯(π1(Xgk)).\Psi:\operatorname{\mathcal{AUT}}\big{(}\pi_{1}(S_{g})\big{)}\xrightarrow{\sigma}\operatorname{\mathcal{AUT}}\big{(}\pi_{1}(X_{g}^{k})\big{)}\to\operatorname{\mathcal{OUT}}\big{(}\pi_{1}(X_{g}^{k})\big{)}.

The restriction of Ψ\Psi to π1(Sg)Inn(π1(Sg))\pi_{1}(S_{g})\cong\operatorname{Inn}\big{(}\pi_{1}(S_{g})\big{)} factors as follows.

π1(Sg)\textstyle{\pi_{1}(S_{g})}𝒜𝒰𝒯(π1(Sg))\textstyle{\operatorname{\mathcal{AUT}}\big{(}\pi_{1}(S_{g})\big{)}}H1(Sg;)\textstyle{H^{1}(S_{g};\mathbb{Z})}𝒪𝒰𝒯(π1(Xgk))\textstyle{\operatorname{\mathcal{OUT}}\big{(}\pi_{1}(X_{g}^{k})\big{)}}conjugationΨ\textstyle{\Psi}kδab\textstyle{k\delta\circ\operatorname{ab}}transvections

Here ab\operatorname{ab} denotes the abelianization map π1(Sg,)H1(Sg;)\pi_{1}(S_{g},*)\to H_{1}(S_{g};\mathbb{Z}).

3.3. Proof of Theorem A

Using the isomorphisms between mapping class groups and automorphism groups, the desired diagram is equivalent to the following one.

1\textstyle{1}π1(Sg)ab\textstyle{\pi_{1}(S_{g})^{\operatorname{ab}}}𝒜𝒰𝒯(π1(Sg))/π\textstyle{\operatorname{\mathcal{AUT}}\big{(}\pi_{1}(S_{g})\big{)}/\pi^{\prime}}𝒪𝒰𝒯(π1(Sg))\textstyle{\operatorname{\mathcal{OUT}}\big{(}\pi_{1}(S_{g})\big{)}}1\textstyle{1}1\textstyle{1}Hom(π1(Sg),)\textstyle{\operatorname{Hom}(\pi_{1}(S_{g}),\mathbb{Z})}𝒪𝒰𝒯(π1(Xgk))\textstyle{\operatorname{\mathcal{OUT}}\big{(}\pi_{1}(X_{g}^{k})\big{)}}𝒪𝒰𝒯(π1(Sg))\textstyle{\operatorname{\mathcal{OUT}}\big{(}\pi_{1}(S_{g}\big{)})}1\textstyle{1}kδ\textstyle{k\delta}

The map Ψ\Psi in (15) descends to the middle vertical map and restricts to the left vertical map by Corollary 3.5. The fact that σ\sigma is a section (Corollary 3.1) implies that the middle vertical map descends to the identity map on 𝒪𝒰𝒯(π1(Sg))\operatorname{\mathcal{OUT}}\big{(}\pi_{1}(S_{g})\big{)}. When k=1k=1, the middle vertical map is an isomorphism by the five lemma. This concludes the proof of Theorem A.

4. Spectral sequence computation

In this section we prove Theorem 1.2. This is achieved by two different computations using the Lyndon–Hochschild–Serre (LHS) spectral sequence. Recall that this spectral sequence takes input a short exact sequence of groups 1NGQ11\to N\to G\to Q\to 1 and a GG-module AA, has E2E_{2} page

E2p,q=Hp(Q;Hq(N;A)),E_{2}^{p,q}=H^{p}\big{(}Q;H^{q}(N;A)\big{)},

and converges to Hp+q(G;A)H^{p+q}(G;A). For both computations we use the Birman exact sequence, but with different choices of the module AA.

Notational note. To simplify the notation, we use the convention that cohomology groups have \mathbb{Z} coefficients unless otherwise specified.

4.1. Euler class computation

Our goal in this section is to prove Proposition 4.1 below, which implies Corollary 1.3.

Proposition 4.1.

Fix g1g\geq 1. Let eukeu_{k} be the Euler class of the extension (1). Then euk=keu1eu_{k}=k\>eu_{1}, and eu1eu_{1} has order 2g22g-2 in H2(Mod(Sg);H1(Sg))H^{2}\big{(}\operatorname{Mod}(S_{g});H^{1}(S_{g})\big{)}.

Proof.

The relation euk=keu1eu_{k}=k\>eu_{1} already follows from Theorem A. Indeed, choosing a set-theoretic section for the sequence in the top row of the diagram in Theorem A gives a cocycle representative for eukeu_{k} that is kk times the cocycle representative for e1e_{1}.

Now we prove that eu1eu_{1} generates a cyclic subgroup isomorphic to /(2g2)\mathbb{Z}/(2g-2)\mathbb{Z} in H2(Mod(Sg);H1(Sg))H^{2}\big{(}\operatorname{Mod}(S_{g});H^{1}(S_{g})\big{)}. Our method is to apply the LHS spectral sequence to the Birman exact sequence with the module A=H1(Sg)A=H^{1}(S_{g}). Here

E2p,qHp(Mod(Sg);Hq(Sg;A)).E_{2}^{p,q}\cong H^{p}\big{(}\operatorname{Mod}(S_{g});H^{q}(S_{g};A)\big{)}.

A portion of the associated 5-term exact sequence is as follows.

0H1(Mod(Sg);H1(Sg))H1(Mod(Sg,1);H1(Sg))𝐴Hom(H1(Sg),H1(Sg))Mod(Sg)d20,1H2(Mod(Sg);H1(Sg))\begin{array}[]{rll}0\to H^{1}\big{(}\operatorname{Mod}(S_{g});H^{1}(S_{g})\big{)}\to H^{1}\big{(}\operatorname{Mod}(S_{g,1});H^{1}(S_{g})\big{)}&\xrightarrow{A}\operatorname{Hom}\big{(}H_{1}(S_{g}),H^{1}(S_{g})\big{)}^{\operatorname{Mod}(S_{g})}\\[5.69054pt] &\xrightarrow{d_{2}^{0,1}}H^{2}\big{(}\operatorname{Mod}(S_{g});H^{1}(S_{g})\big{)}\end{array}

This sequence has been studied by Morita. Morita [Mor85, Prop. 4.1] computes that the first term vanishes, so the map AA is injective. The group Hom(H1(Sg),H1(Sg))Mod(Sg)\operatorname{Hom}\big{(}H_{1}(S_{g}),H^{1}(S_{g})\big{)}^{\operatorname{Mod}(S_{g})} is isomorphic to \mathbb{Z} and generated the Poincaré duality isomorphism δ\delta. Morita [Mor85, proof of Prop. 6.4] shows that the image of AA is (2g2)(2g-2)\mathbb{Z}. Consequently, the differential d20,1d_{2}^{0,1} descends to an injection /(2g2)H2(Mod(Sg);H1(Sg))\mathbb{Z}/(2g-2)\mathbb{Z}\to H^{2}\big{(}\operatorname{Mod}(S_{g});H^{1}(S_{g})\big{)}.

It remains to show that d20,1d_{2}^{0,1} sends a generator to eu1eu_{1}. The differential d20,1d_{2}^{0,1} is the transgression; see e.g. [NSW08, Prop. 1.6.6, Thm. 2.4.3]. By standard knowledge of the transgression applied to our situation, we find that d20,1d_{2}^{0,1} sends a generator to δ(eu)\delta_{*}(eu), where eueu is the Euler class of the extension (2), and

δ:H2(Mod(Sg);H1(Sg))H2(Mod(Sg);H1(Sg))\delta_{*}:H^{2}\big{(}\operatorname{Mod}(S_{g});H_{1}(S_{g})\big{)}\to H^{2}\big{(}\operatorname{Mod}(S_{g});H^{1}(S_{g})\big{)}

is the isomorphism induced by the Poincaré duality isomorphism δ\delta. (For this property of the transgression, see [NSW08, §I.6, Exercise 1-2]. While that reference is mainly concerned with finite or profinite groups, the analysis of the transgression contained given there applies more generally.) Finally, we observe that δ(eu)=eu1\delta_{*}(eu)=eu_{1} by Theorem A. ∎

4.2. Computation of H2(Mod(Sg);H1(Sg))H^{2}\big{(}\operatorname{Mod}(S_{g});H^{1}(S_{g})\big{)}

Running the LHS spectral sequence with the trivial module A=A=\mathbb{Z}, we prove that if g8g\geq 8, then

(16) H2(Mod(Sg);H1(Sg))/(2g2).H^{2}\big{(}\operatorname{Mod}(S_{g});H^{1}(S_{g})\big{)}\cong\mathbb{Z}/(2g-2)\mathbb{Z}.

Combining this with Proposition 4.1 proves Theorem 1.2. The relevant portion of the spectral sequence appears below.

2{2}H0(Mod(Sg);H2(Sg)){H^{0}(\operatorname{Mod}(S_{g});H^{2}(S_{g}))}1{1}0{0}0{0}H2(Mod(Sg);H1(Sg)){H^{2}(\operatorname{Mod}(S_{g});H^{1}(S_{g}))}0{0}{\mathbb{Z}}0{0}H2(Mod(Sg)){H^{2}(\operatorname{Mod}(S_{g}))}H3(Mod(Sg)){H^{3}(\operatorname{Mod}(S_{g}))}H4(Mod(Sg)){H^{4}(\operatorname{Mod}(S_{g}))}0{0}1{1}2{2}3{3}4{4}d20,2d_{2}^{0,2}d22,1d_{2}^{2,1}

The computations in the first column are easy. In the second column, Morita [Mor85, Prop. 4.1] computed H1(Mod(Sg);H1(Sg))=0H^{1}\big{(}\operatorname{Mod}(S_{g});H^{1}(S_{g})\big{)}=0 for g1g\geq 1. The other computation H1(Mod(Sg))=0H^{1}\big{(}\operatorname{Mod}(S_{g})\big{)}=0 holds for g1g\geq 1 because the abelianization of Mod(Sg)\operatorname{Mod}(S_{g}) is finite [FM12, §5.1.2-3].

According to [BT01, Cor. 1.2],

H(Mod(Sg,1))H(Mod(Sg))[x]H_{*}\big{(}\operatorname{Mod}(S_{g,1})\big{)}\cong H_{*}\big{(}\operatorname{Mod}(S_{g})\big{)}\otimes\mathbb{Z}[x]

in degrees g2g\geq 2*. Here xx has degree 2. Applying this and using the universal coefficients theorem, we conclude that

Hi(Mod(Sg))Hi(Mod(Sg,1))H^{i}\big{(}\operatorname{Mod}(S_{g})\big{)}\to H^{i}\big{(}\operatorname{Mod}(S_{g,1})\big{)}

is an isomorphism if i=3i=3 and g6g\geq 6, and it is injective if i=4i=4 and if g8g\geq 8.

Since the map H4(Mod(Sg))H4(Mod(Sg,1))H^{4}\big{(}\operatorname{Mod}(S_{g})\big{)}\to H^{4}\big{(}\operatorname{Mod}(S_{g,1})\big{)} is injective, the differential d22,1d_{2}^{2,1} is zero. Since the map H3(Mod(Sg))H3(Mod(Sg,1))H^{3}\big{(}\operatorname{Mod}(S_{g})\big{)}\to H^{3}\big{(}\operatorname{Mod}(S_{g,1})\big{)} is an isomorphism, the differential d20,2d_{2}^{0,2} is surjective.

Thus, the filtration of H2(Mod(Sg,1))H^{2}\big{(}\operatorname{Mod}(S_{g,1})\big{)} coming from the EE_{\infty} page gives an exact sequence

0H2(Mod(Sg))H2(Mod(Sg,1))𝐹H0(Mod(Sg);H2(Sg))d20,2H2(Mod(Sg);H1(Sg))0.\begin{array}[]{rcl}0\to H^{2}\big{(}\operatorname{Mod}(S_{g})\big{)}\to H^{2}\big{(}\operatorname{Mod}(S_{g,1})\big{)}&\xrightarrow{F}&H^{0}\big{(}\operatorname{Mod}(S_{g});H^{2}(S_{g})\big{)}\cong\mathbb{Z}\\ &\xrightarrow{d_{2}^{0,2}}&H^{2}\big{(}\operatorname{Mod}(S_{g});H^{1}(S_{g})\big{)}\to 0.\end{array}

For g4g\geq 4,

H2(Mod(Sg))[e1] and H2(Mod(Sg,1))[e,e1]H^{2}\big{(}\operatorname{Mod}(S_{g})\big{)}\cong\mathbb{Z}[e_{1}]\>\>\>\text{ and }\>\>\>H^{2}\big{(}\operatorname{Mod}(S_{g,1})\big{)}\cong\mathbb{Z}[e,e_{1}]

and the map [e1][e,e1]\mathbb{Z}[e_{1}]\to\mathbb{Z}[e,e_{1}] is the obvious one e1e1e_{1}\mapsto e_{1}. We claim that F(e)=22gF(e)=2-2g. From this we deduce the desired isomorphism (16). The claim follows from the fact that the extension that defines ee, when restricted to the point-pushing subgroup π1(Sg)<Mod(Sg,1)\pi_{1}(S_{g})<\operatorname{Mod}(S_{g,1}), gives the extension

1π1(USg)π1(Sg)11\to\mathbb{Z}\to\pi_{1}(US_{g})\to\pi_{1}(S_{g})\to 1

where USgUS_{g} is the unit tangent bundle. See [FM12, §5.5.5]. This extension has Euler class 22g2-2g, so the claim follows.

References

  • [Bro82] K. S. Brown. Cohomology of groups, volume 87 of Graduate Texts in Mathematics. Springer-Verlag, New York-Berlin, 1982.
  • [BT01] C.-F. Bödigheimer and U. Tillmann. Stripping and splitting decorated mapping class groups. In Cohomological methods in homotopy theory (Bellaterra, 1998), volume 196 of Progr. Math., pages 47–57. Birkhäuser, Basel, 2001.
  • [Che19] L. Chen. On the nonrealizability of braid groups by homeomorphisms. Geom. Topol., 23(7):3735–3749, 2019.
  • [CR77] P. E. Conner and F. Raymond. Deforming homotopy equivalences to homeomorphisms in aspherical manifolds. Bull. Amer. Math. Soc., 83(1):36–85, 1977.
  • [CS22] L. Chen and N. Salter. Global fixed points of mapping class group actions and a theorem of Markovic. J. Topol., 15(3):1311–1324, 2022.
  • [FM12] B. Farb and D. Margalit. A primer on mapping class groups, volume 49 of Princeton Mathematical Series. Princeton University Press, Princeton, NJ, 2012.
  • [Mar07] V. Markovic. Realization of the mapping class group by homeomorphisms. Invent. Math., 168(3):523–566, 2007.
  • [McC91] D. McCullough. Virtually geometrically finite mapping class groups of 33-manifolds. J. Differential Geom., 33(1):1–65, 1991.
  • [Mor85] S. Morita. Family of Jacobian manifolds and characteristic classes of surface bundles. II. Proc. Japan Acad. Ser. A Math. Sci., 61(4):112–115, 1985.
  • [NSW08] J. Neukirch, A. Schmidt, and K. Wingberg. Cohomology of number fields, volume 323 of Grundlehren der mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]. Springer-Verlag, Berlin, second edition, 2008.
  • [Sou10] J. Souto. A remark on the action of the mapping class group on the unit tangent bundle. Ann. Fac. Sci. Toulouse Math. (6), 19(3-4):589–601, 2010.
  • [Wal68] F. Waldhausen. On irreducible 33-manifolds which are sufficiently large. Ann. of Math. (2), 87:56–88, 1968.