, , 6 January 2005, revised 14 March 2005

Theorem 1.10.

H a m s

has the extension property.

Because

H a m s

is the kernel of a crossed homomorphism it follows from standard theory that one can use this homomorphism to define a class

O M

that measures the obstruction to lifting a map

φ : B \to B Symp

B H a m s

. Here

O M \in H^{2} (B Symp; H^{1} (M; P ω)) = H^{2} (B Symp; π_{1} (A)),

where we think of

A = H^{1} (M; R) / H^{1} (M; P ω)

as a quasitopological group: see Remark 2.6 . Thus the local coefficient system

π_{1} (A)

B Symp

has fibers isomorphic to the discrete group

H_{Q} : = H^{1} (M; P ω)

The first part of the next result restates Theorem 1.10 .

Theorem 1.11. (i) There is a class

O M \in H^{2} (B Symp; H_{Q})

such that the symplectic fibration

P \to B

classified by

φ : B \to B Symp

has a closed connection form iff

φ^{*} (O M) = 0

. In particular,

O M \in H^{2} (B Symp; H_{Q})

is zero iff every symplectic

M

-bundle has a closed connection form.

(ii) There is a crossed homomorphism

\tilde{F} : Symp \to H^{1} (M; R) / Γ

that extends Flux if and only if

O M

lies in the image of

H^{2} (B Symp; Γ)

H^{2} (B Symp; H_{Q})

Corollary 1.12. The following conditions are equivalent:

(i) Every symplectic

M

-bundle has a closed connection form.

(ii)

Γ = 0

and there is a crossed homomorphism

\tilde{F} : Symp \to H^{1} (M; R)

extending Flux.

For example, when

(M, ω)

is monotone, the first Chern class of the vertical tangent bundle of

P \to B

provides an extension of

[ω]

. Therefore the obstruction class

O M

must vanish. This is consistent with the corollary since the Kotschick–Morita homomorphism

F_{K M}

extends Flux. Observe further that, because

O M \in H^{2} (B Symp)

, Theorem 1.11 gives a second proof of Lemma 1.2 .

The next result clarifies the conditions under which Flux can be extended.

Proposition 1.13. The following conditions are equivalent.

(i) There is an extension

\tilde{F} : Symp (M, ω) \to H^{1} (M; R) / Γ

of Flux. (ii) For every closed Riemann surface

Σ

every representation of

π_{1} (Σ)

π_{0} (Symp)

lifts to a representation into the group

Symp / Ham

. (iii) For every product of commutators

[u_{1}, u_{2}] . . . [u_{2 p - 1}, u_{2 p}]

u_{i} \in Symp

, that lies in

{Symp}_{0}

, there are elements

g_{1}, . . ., g_{2 p} \in {Symp}_{0}

such that

[u_{1} g_{1}, u_{2} g_{2}] . . . [u_{2 p - 1} g_{2 p - 1}, u_{2 p} g_{2 p}] \in Ham .

(iv) For every symplectic

M

-bundle

P \to Σ

there is a bundle

Q \to S^{2}

such that the fiberwise connect sum

P # Q \to Σ # S^{2} = Σ

has a closed connection form.

Remark 1.14. (i) It is immediate that (ii) is equivalent to (iii). It is also easy to see that (i) implies (ii): if we write

ℋ : = ker \tilde{F}

, then the quotient group

ℋ / Ham

is isomorphic to

π_{0} (Symp)

because

ℋ \cap {Symp}_{0} = Ham

. Hence any representation in

π_{0} (Symp)

can be lifted to the subgroup

ℋ / Ham

Symp / Ham

. Our proof of the converse involves some topology since it uses the relation between

\tilde{F}

and the obstruction cocycle

O M

. On the other hand, if we restrict to the subgroup

{Symp}_{H}

Symp

that acts trivially on

H^{1} (M; R)

then (iii) is equivalent to saying that

\begin{matrix} [{Symp}_{H}, {Symp}_{H}] \cap {Symp}_{0} = Ham . \end{matrix}

(1.3)

and, as we show in Lemma 2.8 , there are immediate algebraic reasons why this is equivalent to the existence of an extension of

Flux

{Symp}_{H}

We show in Corollary 4.12 that equation ( 1.3 ) holds when

[ω]

vanishes on tori and lies in the subring of

H^{*} (M)

generated by

H^{1}

. However, it is so far unknown whether it always holds. If not, then

\tilde{F}

cannot always exist. On the other hand, there are indications that ( 1.3 ) might always hold. It seems that a large part of

π_{0} (Symp)

can be generated by Dehn twists about Lagrangian spheres: cf. Seidel [15,1.7] . In dimensions

> 2

these are well defined up to Hamiltonian isotopy and act trivially on

H^{1} (M)

, and so one might be able to take

ℋ \cap {Symp}_{H}

to be the group generated by Dehn twists. In any case, it does not seem that the methods used in this paper are sufficiently deep to resolve this question.

(ii) Recall that an

M

-bundle

P \to S^{2}

has a closed connection form iff it is constructed from a loop of symplectomorphisms with trivial flux. Therefore the equivalence of (i) and (iv) in Proposition 1.13 is a geometric restatement of Theorem 1.11 (ii).

³ This is the natural identity to use for a crossed homomorphism $G \to A$ when the group $G$ acts contravariantly on the coefficients $A$ . Note also that when $Symp$ acts nontrivially on $H^{1} (M; R)$ it is not possible to extend Flux to a group homomorphism: see Remark 2.7 .

⁴ The group cohomology of a discrete group $G^{δ}$ as as originally defined by Eilenberg–MacLane equals the singular cohomology of its classifying space $B G^{δ}$ . If $G$ is a topological group, then it has a (continuous) group cohomology defined using the (continuous) Eilenberg–MacLane complex. Since these are quite different from the singular cohomology of $B G$ , we will for the sake of clarity denote the group cohomology by $H_{E M}^{*}$ , adding a $c$ wherever appropriate to emphasize continuity.

1.3 Further results and remarks

After discussing stability, we describe a few cases where it is possible to extend the Flux homomorphism. We end by discussing the integral case, and the question of uniqueness.

Stability under perturbations of

ω

. It was shown in Lalonde–McDuff [7] that Hamiltonian bundles are stable under small perturbations of

ω

. One cannot expect general symplectic bundles to be stable under arbitrary small perturbations of

ω

since

π_{1} (B)

may act nontrivially on

H^{2} (M; R)

. Given a symplectic bundle

(M, ω) \to P \to B

let us denote by

V_{2} (P)

the subspace of

H_{2} (M; Q)

generated by the elements

g_{*} (C) - C

, where

C \in H_{2} (M; Z)

and

g

is any symplectomorphism of

M

that occurs as the holonomy of a symplectic connection on

P \to B

around some loop in

B

. (Since

g_{*} (C)

depends only on the smooth isotopy class of

g

, it does not matter which connection we use.) The subspace

H^{2} (M; R)^{i n v}

, consisting of classes

a \in H^{2} (M; R)

that are fixed by all such

g

, is the annihilator of

V_{2} (P)

. The most one can expect is that the existence of a symplectic structure on

P \to B

is stable under perturbations of

[ω]

in this subspace. For example, if

ω

is generic in the sense that it gives an injective map

H_{2} (M; Z) / T o r \to R

then

V_{2} (P)

is torsion and

H^{2} (M; R)^{i n v (P)} = H^{2} (M; R)

Proposition 1.15. Let

(M, ω) \to P \to B

be a symplectic

M

-bundle over a finite simplicial complex

B

. Then there is a neighborhood

N (ω)

ω

in the space of all closed

2

-forms on

M

that represent a class in

H^{2} (M; R)^{i n v (P)}

such that for all

ω^{'} \in N (ω)

(i)

P \to B

has the structure of an

ω^{'}

-symplectic bundle, and (ii) if there is a closed extension of

ω

, then the same is true for

ω^{'}

Part (i) of this proposition follows by the arguments in [7,Cor. 2.5] . Part (ii) was also proved in [7] in the case when

B

is classified by a map into

B {Symp}_{0} (M, ω)

. The proof of the general case is given at the end of § 3 . The next corollary is an immediate consequence of (ii).

Corollary 1.16. If

P \to B

has a

{Ham}^{s}

-structure then the image of the restriction map

H^{2} (P; R) \to H^{2} (M; R)

is the subspace

H^{2} (M; R)^{i n v (P)}

H^{2} (M; R)

that is invariant under the action of

π_{1} (B)

This result implies that the differential

d_{2}^{2, 0}

in the Leray–Serre cohomology spectral sequence for

P \to B

vanishes,⁵ and so is a partial generalization of the vanishing results in [7] .

Remark 1.17. Proposition 1.15 is proved using the Moser homotopy argument and so works only over compact pieces of

Symp

. This is enough to give stability for bundles over finite bases

B

but is not enough to allow one to make any statements about properties that involve the full group

Symp

. Hence even if

Flux

extends to

\tilde{F} : Symp \to H^{1} (M; R) / Γ

for the manifold

(M, ω)

, it is not clear that it also extends for sufficiently close forms

ω^{'}

whose cohomology class is invariant under

Symp (M, ω)

For one thing, however close

ω^{'}

is, there may be new components of

Symp (M, ω^{'})

containing elements that are far from those in

Symp (M, ω)

Manifolds with

Γ = 0

. One expects that for most manifolds

Γ = 0

. Whether

O M

then vanishes is still not clear. We now discuss some special cases in which

Γ = 0

and Flux extends to a crossed homomorphism defined either on the whole group

Symp

or on some large subgroup. Note that if Flux extends to

Symp

, then

O M = 0

by Theorem 1.11 (ii).

The first case is when

(M, ω)

is (strongly) monotone, i.e. the symplectic class

[ω]

is a multiple of the first Chern class. In this case

O M = 0

since one can always choose a closed connection form in the class of a multiple of the vertical first Chern class. Kotschick–Morita [5] observed that Flux always extends. We shall give an explicit formula for

\tilde{F}

in Theorem 4.8 .

Another somewhat tractable case is when

(M, ω)

is atoroidal, i.e.

\int_{T^{2}} ψ^{*} ω = 0

for all smooth maps

ψ : T^{2} \to M

. Note that

Γ = 0

for such manifolds, because for each loop

{f_{t}}

{Symp}_{0}

the value of the class

Flux ({f_{t}})

on the

1

-cycle

γ

is obtained by integrating

ω

over the torus

\cup_{t} f_{t} (γ)

. In the next proposition, we denote by

{Symp}_{π}

the subgroup of

Symp

consisting of elements that are isotopic to a symplectomorphism that fixes the basepoint

x_{0}

M

and induces the identity map

π_{1} (M, x_{0}) \to π_{1} (M, x_{0})

.⁶

Proposition 1.18. If

(M, ω)

is atoroidal then

Γ = 0

and

Flux

extends to a homomorphism

\tilde{F} : {Symp}_{π} \to H^{1} (M; R)

We shall see in § 4.2 that in the above situation

\tilde{F}

can be extended to a crossed homomorphism defined on the whole of

Symp

but at the cost of enlarging the target group.

Proposition 1.18 gives a partial answer to Kedra–Kotschick–Morita's question [3] of whether the usual flux homomorphism

{Symp}_{0} \to H^{1} (M; R)

extends to the full group

Symp

when

[ω]

is a bounded class. This condition means that

[ω]

may be represented by a singular cocycle that is uniformly bounded on the set of all singular

2

-simplices.⁷ If

[ω]

is bounded, then

(M, ω)

is atoroidal since an arbitrary multiple of a toric class

C

can be represented by the sum of just two singular

2

-simplices. Another interesting atoroidal case is that of symplectically hyperbolic manifolds. There are various possible definitions here.

We shall use Polterovich's definition from [14] in which

(M, ω)

is called symplectically hyperbolic if the pullback

\tilde{ω}

ω

to the universal cover

\tilde{M}

M

has bounded primitive, i.e.

\tilde{ω} = d β

for some

1

-form

β

that is bounded with respect to any metric on

\tilde{M}

that is pulled back from

M

. For example,

(M, ω)

might be a product of Riemann surfaces of genus

> 1

with a product symplectic form. Because in the covering

R^{2} \to T^{2}

the boundary of a square of side

N

encloses

N^{2}

fundamental domains, it is easy to check that any

2

-form on

T^{2}

whose pullback to

R^{2}

has bounded primitive must have zero integral over

T^{2}

. Hence we find:

Lemma 1.19. Proposition 1.18 applies both when

[ω]

is bounded and when

(M, ω)

is symplectically hyperbolic.

§ 4.2 contains a few other similar results that are valid in special cases, for example when

ω

vanishes on

π_{2} (M)

. We end the introduction with some general remarks.

Remark 1.20. (The integral case.) There is an analogous group

{Ham}^{s Z}

which is the kernel of a crossed homomorphism

{\hat{F}}_{s}^{Z}

with values in

H^{1} (M; R / P Z ω)

where

P Z ω

denotes the set of values of

[ω]

on the integral

2

-cycles

H_{2} (M; Z)

. In many respects the properties of this group are similar to those of

{Ham}^{s}

. However, there are some interesting differences. If

T o r

denotes the torsion subgroup of

H_{1} (M; Z)

, then the analog of the group

A

occurring in equation ( 1.2 ) is

A^{Z} : = H^{1} (M; R / P Z ω) \sim = Hom (T o r, R / P Z ω) \oplus H^{1} (M; R) / H^{1} (M; P Z ω) .

Hence Theorem 1.11 (i) does not immediately generalize; the proof of Lemma 3.4 shows that the obstruction to the existence of a

{Ham}^{s Z}

-structure is twofold, the first coming from the finite group

π_{0} (A^{Z})

(see Lemma 4.1 ) and the second an obstruction cocycle similar to

O M

coming from

π_{1} (A^{Z})

. Nevertheless, since every

H a m s

-bundle over a compact base

B

has a finite cover with a

{Ham}^{s Z}

-structure, the latter group has the modified extension property.

The group

{Ham}^{s Z}

is most interesting in the case when

P Z ω = Z

, i.e. when

[ω]

is a primitive integral class. In this situation one might expect

B {Ham}^{s Z}

to classify bundles

P \to M

that have a closed and integral connection form. We show in Proposition 3.3 that this is the case when

π_{0} (Symp)

acts trivially on

H_{1} (M; Z)

. Gal–Kȩdra [1] show that this remains true when

π_{0} (Symp)

acts trivially on the torsion classes in

H_{1} (M; Z)

. (This hypothesis is equivalent to saying that the action of

Symp

on the set of prequantum line bundles is trivial.) In general, however, we show in Example 4.2 that there may be no such classifying space. To get a well posed classification problem one must specify a particular integral lift of

[ω]

; equivalently, one must choose a prequantum line bundle.

This question is pursued further in Gal–Kȩdra [1] . We show here only that the choice of

s

determines a prequantum line bundle and that

{Ham}^{s Z}

is homotopy equivalent to the automorphism group of this bundle. (See Proposition 4.5 .)

Remark 1.21. (Issues of uniqueness) (i) Because we are interested in the algebraic and geometric properties of the symplectomorphism group we restricted ourselves above to the case when

ℋ

is a subgroup of

Symp

. However, from a homotopy theoretic point of view it would be more natural to look for a group

K

that classifies pairs consisting of a symplectic

M

-bundle

π : P \to B

together with an extension

\tilde{a} \in H^{2} (P; R)

of the fiberwise symplectic class

[ω]

. Here we should either normalize

\tilde{a}

by requiring

π_{!} ({\tilde{a}}^{n + 1}) = 0

(where

π_{!}

denotes integration over the fiber) or consider

\tilde{a}

to be well defined modulo elements in

π^{*} H^{2} (B)

. Then the homotopy class of

B K

would be well defined and there would be a forgetful map

ψ : B K \to B Symp

which is well defined up to homotopy (assuming that we are working in the category of spaces with the homotopy type of a CW complex). In general,

ψ

would not be a homotopy equivalence since the extension class

\tilde{a} \in H^{2} (P; R)

could vary by an element in

H^{1} (B; H^{1} (M; R))

. Further, in this scenario,

K

need not be a subgroup of

Symp

. (Cf. the discussion in Lalonde–McDuff [7] of the classification of Hamiltonian structures.) (ii) If we insist that

K

be a subgroup of

Symp

then there are several possible notions of equivalence, the most natural of which is perhaps given by conjugation by an element in

{Symp}_{0}

. With this definition equivalent groups would be isomorphic. We show in § 2 that the groups

{Ham}^{s}

are equivalent in this sense, though when

H_{1} (M; Z)

has torsion the integer versions

{Ham}^{s Z}

may not be. It is also not clear whether any two groups

ℋ_{1}, ℋ_{2}

that intersect each component of

Symp

and satisfy

ℋ_{1} \cap {Symp}_{0} = ℋ_{2} \cap {Symp}_{0} = Ham

must be isomorphic as abstract groups, although any such group must be isomorphic to an extension of

π_{0} (Symp)

Ham

. However, there is no immediate reason why they should be conjugate. For example, suppose that the group

π_{0} (Symp)

is isomorphic to

Z

, generated by the component

{Symp}_{α}

Symp

. Then because

Ham

is a normal subgroup of

Symp

the subgroup

ℋ_{g}

Symp

generated by

Ham

together with any element

g \in {Symp}_{α}

intersects

{Symp}_{0}

Ham

and therefore has the required properties. Any two such groups

ℋ_{g_{i}}, i = 1, 2,

are isomorphic, though they are conjugate only if there is

h \in Symp

such that

g_{1} h g_{2}^{- 1} h^{- 1} \in Ham

. On the other hand, because

g_{1}

and

g_{2}

can be joined by an isotopy, there is a smooth family of injective group homomorphisms

ι_{t} : ℋ_{g_{1}} \to Symp

t \in [1, 2]

, that starts with the inclusion and ends with an isomorphism onto

ℋ_{g_{2}}

. Thus the homotopy properties of the inclusions

ℋ_{g_{i}} \to Symp

are the same.

(iii) Instead of looking for subgroups of

Symp

with the extension property one could look for covering groups

ℋ \to Symp

with this property. Notice that if

Λ

is a discrete subgroup of an abelian topological group

A

and if the continuous crossed homomorphism

F : G \to A / Λ

extends the composite

F_{0} : G_{0} \to A \to A / Λ

, where

f : G_{0} \to A

is a homomorphism defined on the identity component of

G_{0}

, then the fiber product

\tilde{G} : = {(g, a) \in G \times A | F (g) = a + Λ}

G

and

A

over

A / Λ

is a covering group of

G

that contains a copy of

G_{0}

, namely the graph of

f

. Moreover, the obvious projection

\tilde{G} \to A

lifts

F_{0}

. This approach is particularly relevant in the integral case mentioned in Remark 1.20 above, as well as the cohomologically symplectic case, where the analog of the Hamiltonian group is already a covering group of

{Diff}_{0}

. For further discussion see § 4.3 and Gal–Kȩdra [1] .

⁵ Note that $d_{3}^{2, 0} = 0$ by Lemma 1.2 .

⁶ One can check that $g \in {Symp}_{π}$ iff for any path $γ$ in $M$ from $x_{0}$ to $g (x_{0})$ the induced maps $γ_{*}, g_{*} : π_{1} (M, x_{0}) \to π_{1} (M, g x_{0})$ differ by an inner automorphism. Thus, loosely speaking, ${Symp}_{π}$ consists of all symplectomorphisms that act trivially on $π_{1} (M)$ .

⁷ The (smooth) cocycle represented by integrating $ω$ can never be bounded because bounded cocycles vanish on cylinders as well as tori.

Contents

Acknowledgements. The author thanks Swiatoslav Gal, Jarek Kȩdra, Jack Milnor, Leonid Polterovich and Zhigang Han for useful discussions. She also thanks Gal and Kȩdra for showing her early drafts of their paper [1] and making various helpful comments about earlier versions of this paper. In particular, they helped streamline the proof of Lemma 3.4 .

2 Definition and Properties of ${\hat{F}}_{s}$

Define

P Z ω

(resp.

P ω : = P Q ω

) to be the set of values taken by

[ω]

on the elements of

H_{2} (M; Z)

(resp.

H_{2} (M; Q)

). To define

{\hat{F}}_{s}

we follow a suggestion of Polterovich (explained in Lalonde–McDuff [7] ). Define the homology group

S H_{1} (M, ω; Z)

to be the quotient of the space of integral

1

-cycles in

M

by the image under the boundary map

\partial

of the integral

2

-chains with zero symplectic area. Then there is a projection

π_{Z} : S H_{1} (M, ω; Z) \to H_{1} (M; Z)

and we set

S H_{1} (M, ω) : = S H_{1} (M, ω; Z) \otimes Q .

We shall consider

S H_{1} (M, ω)

and

P Q ω

Q

-vector spaces. Given a loop (or integral

1

-cycle)

ℓ

M

we denote its image in

H_{1} (M; Z)

H_{1} (M; Q)

[ℓ]

and its image in

S H_{1} (M, ω; Z)

S H_{1} (M, ω)

〈 ℓ 〉 .

We usually work over the rationals and shall omit the label

Q

unless there is a possibility of confusion.

Lemma 2.1. There are split exact sequences

\begin{matrix} 0 \to R / P Z ω \to S H_{1} (M, ω; Z) \overset{π_{Z}}{\to} H_{1} (M; Z) \to 0, \end{matrix}

(2.1)

and

\begin{matrix} 0 \to R / P Q ω \to S H_{1} (M, ω) \overset{π}{\to} H_{1} (M; Q) \to 0 . \end{matrix}

(2.2)

Proof. Choose a continuous family of integral $2$ -chains $f_{t} : D \to M$ for $t \in R$ with $\int_{D} f_{t}^{*} ω = t$ . If $γ_{t} : = f_{t} |_{\partial D}$ denotes the boundary of $f_{t}$ , then the elements $〈 γ_{t} 〉, t \in R,$ generate the kernel of the projection $π_{Z} : S H_{1} (M, ω; Z) \to H_{1} (M, Z)$ . Moreover they represent different classes in $S H_{1} (M, ω; Z)$ if and only if $t - t^{'} / \in P Z ω$ . Hence the sequence $0 \to R / P Z ω \to S H_{1} (M, ω; Z) \overset{π_{Z}}{\to} H_{1} (M; Z) \to 0,$ is exact. To see that it splits, we just need to check that each element $λ = [ℓ]$ of finite order $N$ in $H_{1} (M; Z)$ is the image of some element of order $N$ in $S H_{1} (M; Z)$ . But if $W$ is an integral $2$ -chain such that $\partial W = N ℓ$ and if $μ : = \int_{W} ω$ then $N (〈 ℓ 〉 - 〈 γ_{μ / N} 〉) = 0 and π (〈 ℓ 〉 - 〈 γ_{μ / N} 〉) = [ℓ] .$ In fact every element of order $N$ in the coset $π_{Z}^{- 1} ([ℓ])$ has the form $〈 ℓ 〉 - 〈 γ_{ν} 〉$ where $N ν \in μ + P Z ω$ . The proof for ( 2.2 ) is similar. □

We explain in § 4.1 a natural way to understand splittings of

π_{Z}

in the case when

P Z ω = Z

: cf. Definition 4.3 . Note also that in the previous lemma there is no need for

ω

to be nondegenerate; it suffices for it to be closed. However if it were an arbitrary closed form it would not have many isometries, and so the next lemmas would have little interest.

Lemma 2.2. The group

Symp (M, ω)

acts on

S H_{1} (M, ω; Z)

and

S H_{1} (M, ω)

The induced action of

{Symp}_{0}

on the set of splittings of

π

is transitive. When

H_{1} (M; Z)

has no torsion

Symp

also acts transitively on the splittings of

π_{Z}

Proof. Again, we shall work with the sequence over $Z$ . The group $Symp (M, ω)$ acts on these spaces because it preserves $ω$ . To prove the transitivity statement, note first that any splitting $s$ of $π_{Z}$ has the form $s λ_{i} = 〈 ℓ_{i} 〉$ where $ℓ_{1}, . . ., ℓ_{k}$ are loops (i.e. integral $1$ -cycles) in $M$ that project to the basis $λ_{1}, . . ., λ_{k}$ of $H_{1} (M; Z)$ . Suppose given two such splittings $s, s^{'}$ corresponding to different sets $L, L^{'}$ of representing $1$ -cycles for the $λ_{i}$ . Suppose also that $dim M > 2$ . Since Hamiltonian isotopies have zero flux, we may move the loops in $L$ and $L^{'}$ by such isotopies, without affecting their images in $S H_{1} (M, ω; Z)$ and so that no two intersect. Now choose $T_{1}, . . ., T_{k} \in R^{+}$ such that $〈 ℓ_{i}^{'} 〉 = 〈 ℓ_{i} 〉 + 〈 γ_{T_{i}} 〉, 1 \leq i \leq k,$ where the $γ_{t}$ are as in Lemma 2.1 . For each $i$ there is a symplectic isotopy $h_{i, t}$ such that for all $t \in [0, T_{i}]$ , $h_{i, t} |_{ℓ_{j}} = i d, j < i, h_{i, t} |_{h_{j, T_{j}} ℓ_{j}} = i d, j > i, \int_{W_{i}} ω = T_{i},$ where $W_{i} : = \cup_{0 \leq t \leq T_{i}} h_{i, t} (ℓ_{i})$ . (Take the $h_{i, t}$ to be generated by closed $1$ -forms $α_{i}$ that vanish near the appropriate loops and are such that $\int_{ℓ_{i}} α_{i} \neq 0$ . Here we are using the fact that $[ℓ_{i}]$ is not a torsion class.) Then $h : = h_{1, T_{1}} \circ \cdot \cdot \cdot \circ h_{k, T_{k}}$ takes $s$ to $s^{'}$ .
To extend this argument to the case $dim M = 2$ , it is convenient to describe the splitting by its effect on a standard basis $λ_{i}$ of $H_{1} (M; Z)$ . Thus we may assume that $ℓ_{i}$ and $ℓ_{j}$ are disjoint unless $(i, j) = (2 k - 1, 2 k)$ in which case they intersect in a single point. If $s_{0}$ is the splitting defined by these loops, it suffices to show that for any numbers $T_{i}$ there are representatives $ℓ_{i}^{'}$ for the $[ℓ_{i}]$ such that for each $i$ there is a cylinder of area $T_{i}$ with boundary $ℓ_{i}^{'} - ℓ_{i}$ . One achieves this by first isotoping the $ℓ_{i}$ for $i$ odd (fixing the other loops), and then adjusting the $ℓ_{i}$ for even $i$ . □

Choose a splitting

s

for

π_{Z}

. If

h \in Symp

and

λ \in H_{1} (M; Z)

, then the element

h_{*} (s λ) - s (h_{*} λ)

lies in the kernel of

π_{Z} : S H_{1} (M, ω) \to H_{1} (M; Z)

and one can define a map

{\hat{F}}_{s}^{Z} : Symp (M) \to A^{Z} : = Hom (H_{1} (M; Z), R / P Z ω)

by setting

\begin{matrix} {\hat{F}}_{s}^{Z} (h) (λ) : = h_{*} (s λ) - s (h_{*} λ) \in R / P Z ω, λ \in H_{1} (M; Z) . \end{matrix}

(2.3)

Explicitly, if we denote by

\bar{λ}

the image

s (λ)

λ \in H_{1} (M)

, then

\begin{matrix} {\hat{F}}_{s} (h) (λ) = a (h \bar{λ} - \bar{h λ}), \end{matrix}

(2.4)

where

a (〈 ℓ^{'} 〉 - 〈 ℓ 〉)

is the symplectic area of any cycle with boundary

ℓ^{'} - ℓ

. Similarly, for each splitting

s

of ( 2.2 ) we define

{\hat{F}}_{s} : Symp (M) \to A : = Hom (H_{1} (M), R / P Q ω) = H^{1} (M; R) / H^{1} (M; P Q ω)

{\hat{F}}_{s} (h) (λ) : = h_{*} (s λ) - s (h_{*} λ) \in R / P Q ω, λ \in H_{1} (M; Q) .

Proposition 2.3. (i)

{\hat{F}}_{s}^{Z}

is a crossed homomorphism that equals the composite

{Symp}_{0} \overset{Flux}{\to} H^{1} (M; R) / Γ \to A^{Z} : = Hom (H_{1} (M; Z), R / P Z ω)

{Symp}_{0}

. Moreover the class

[{\hat{F}}^{Z}] : = [{\hat{F}}_{s}^{Z}] \in H_{c E M}^{1} (Symp, A^{Z})

is independent of the choice of

s

(ii) The analogous statements hold for

{\hat{F}}_{s}

Proof.

{\hat{F}}_{s}^{Z}

is a crossed homomorphism because for all

g, h \in Symp

\begin{matrix} {\hat{F}}_{s}^{Z} (g h) (λ) & = & a (g h \bar{λ} - \bar{g h λ}) \end{matrix}

\begin{matrix} = & a (g h \bar{λ} - g \bar{h λ}) + a (g \bar{h λ} - \bar{g h λ}) \end{matrix}

\begin{matrix} = & a (h \bar{λ} - \bar{h λ}) + a (g \bar{h λ} - \bar{g h λ}) \end{matrix}

\begin{matrix} = & {\hat{F}}_{s}^{Z} (h) (λ) + {\hat{F}}_{s}^{Z} (g) (h λ) \end{matrix}

\begin{matrix} = & {\hat{F}}_{s}^{Z} (h) (λ) + h^{*} {\hat{F}}_{s}^{Z} (g) (λ) . \end{matrix}

The rest of the first statement in (i) is immediate from the definition.

To prove the second statement in (i) observe that two choices of splitting

s, s^{'}

differ by the element

α \in A^{Z} : = Hom (H_{1} (M; Z), R / P Z ω)

given by

α (λ) : = s^{'} (λ) - s (λ) \in R / P Z ω, λ \in H_{1} (M; Z) .

It follows easily that

\begin{matrix} {\hat{F}}_{s^{'}}^{Z} (h) - {\hat{F}}_{s}^{Z} (h) = α - h^{*} α, \end{matrix}

(2.5)

and so is a coboundary in the Eilenberg–MacLane complex.

The proof of (ii) is similar. □

Definition 2.4. Given a splitting

s : H_{1} (M; Q) \to S H_{1} (M, ω)

we define the enlarged Hamiltonian group

H a m s (M, ω)

to be the kernel of

{\hat{F}}_{s}

. Similarly, we define

{Ham}^{s Z} (M, ω)

to be the kernel of the integral crossed homomorphism

{\hat{F}}_{s}^{Z}

Lemma 2.5. Let

s, s^{'}

be two splittings and define

{Symp}_{H}

to be the subgroup of

Symp

that acts trivially on rational homology.

(i)

{Symp}_{H} \cap H a m s = {Symp}_{H} \cap {Ham}^{s^{'}}

(ii) The map

π_{0} ({Ham}^{s}) \to π_{0} (Symp)

is surjective.

(iii) The subgroups

H a m s

and

{Ham}^{s^{'}}

are conjugate in

Symp

by an element in

{Symp}_{0}

(iv) When topologized as a subspace of

Symp

, the path component of

H a m s

containing the identity element is

Ham (M, ω)

Proof. (i) is an immediate consequence of the identity ( 2.5 ). (iii) follows from the fact that ${Symp}_{0}$ acts transitively on the set of splittings and the description of $H a m s$ as the subgroup of $Symp$ whose action on $S H_{1} (M, ω)$ preserves the image of $s$ . To prove (ii), we must show that any element $h \in Symp$ is homotopic to an element in ${Ham}^{s}$ . This holds because the splittings $s^{'} = h_{*} (s)$ and $s$ are conjugate by an element in ${Symp}_{0}$ . To prove (iv) consider a continuous path $h_{t} \in Symp$ that starts at the identity and is such that ${\hat{F}}_{s} (h_{t}) (λ) = 0 \in R / P ω$ for all $t$ . By Proposition 2.3 , the path $t \mapsto {\hat{F}}_{s} (h_{t}) (λ) \in R / P ω$ has the continuous lift $t \mapsto Flux (h_{t}) (λ) \in R$ . Since $P ω$ is totally disconnected this lift must be identically zero; in other words the path $h_{t}$ is a Hamiltonian isotopy. □

Part (iv) of Lemma 2.5 holds for the group

{Ham}^{s Z}

, and (i) holds if one replaces

{Symp}_{H}

by the group that acts trivially on

H_{1} (M; Z)

. However, one must take care with the other two statements. For further details see § 4.1 .

Remark 2.6. (Topologies on

H a m s

and

R / P ω

.) The intersection

H a m s \cap {Symp}_{0}

is disconnected. In fact it is everywhere dense in

{Symp}_{0}

. Hence the subspace topology

τ_{s}

H a m s

is rather counterintuitive and it is better to give

H a m s

a finer topology in which its path components are closed. Therefore, although we give the group

Symp

the usual

C^{\infty}

-topology (which is the subspace topology it inherits from the diffeomorphism group), we give

H a m s

the topology

τ_{c}

that it inherits from the Hamiltonian topology on

Symp

. Then the identity map

(H a m s, τ_{c}) \to (H a m s, τ_{s})

is continuous and is a weak homotopy equivalence. Thus this change in topology does not affect the homotopy or (co)homology of the space.

Correspondingly we shall always think of

P ω

as a discrete group. Further we think of quotients such as

R / P ω

as quasitopological spaces, i.e. we specify which maps

f : X \to R / P ω

are continuous, where

X

is a finite simplicial complex. This gives enough structure so that we can talk of homotopy groups. In the present situation we say that

f

is continuous iff

X

has a subdivision

X^{'}

such that the restriction of

f

to each simplex in

X^{'}

has a continuous lift to

R

. Hence

π_{1} (R / P ω) \sim = P ω, π_{j} (R / P ω) = 0, j > 1 .

Thus

R / P ω

is (weakly homotopic to) the Eilenberg–MacLane space

K (P ω, 1)

. (Another way to deal with this technical problem — that also arises when one deals with spaces of germs — is to replace

R / P ω

by an appropriate semisimplicial complex. But then one has to replace all spaces and groups by their semisimplicial analogs.)

□

Remark 2.7. If

g \in Symp

and

h \in {Symp}_{0}

then it is easy to check that

Flux (g^{- 1} h g) = g^{*} (Flux h)

. Hence, if

{Symp}_{H}

denotes the subgroup of

Symp

acting trivially on

H^{1} (M; R)

then

[{Symp}_{0}, {Symp}_{H}] = Ham .

On the other hand

[{Symp}_{0}, Symp] = Ham

only if

Symp = {Symp}_{H}

. Hence when

Symp \neq {Symp}_{H}

the flux homomorphism does not extend to a homomorphism

Symp \to H^{1} (M) / Γ

There is another relevant subgroup, namely

{Symp}_{H^{Z}}

, consisting of elements that act trivially on

H_{1} (M; Z)

. Note that

[{Symp}_{H^{Z}}, {Symp}_{H^{Z}}] \cap {Symp}_{0}

lies in

{Ham}^{s Z}

because

{\hat{F}}_{s}^{Z}

restricts to a homomorphism on

{Symp}_{H^{Z}}

and so vanishes on the commutator subgroup

[{Symp}_{H^{Z}}, {Symp}_{H^{Z}}]

. But this is the best we can say; in particular, it is not clear whether

[{Symp}_{H}, {Symp}_{H}] \cap {Symp}_{0}

must always equal

Ham

Lemma 2.8. The following statements are equivalent.

(i)

[{Symp}_{H}, {Symp}_{H}] \cap {Symp}_{0} = Ham

; (ii) For every product of commutators

y : = [u_{1}, u_{2}] . . . [u_{2 p - 1}, u_{2 p}]

u_{i} \in {Symp}_{H}

, that lies in

{Symp}_{0}

, there are elements

g_{1}, . . ., g_{2 p} \in {Symp}_{0}

such that

f : = [u_{1} g_{1}, u_{2} g_{2}] . . . [u_{2 p - 1} g_{2 p - 1}, u_{2 p} g_{2 p}] \in Ham .

(iii) The flux homomorphism

Flux : {Symp}_{0} \to H^{1} (M; R) / Γ

extends to a continuous homomorphism

F : {Symp}_{H} \to H^{1} (M; R) / Γ

Proof. Clearly (iii) implies (i), which in turn implies (ii). To see that (ii) implies (i), note the identity

[u g, v h] = g^{u} h^{u v} (g^{- 1})^{u v} (h^{- 1})^{u v u^{- 1}} [u, v],

where

g^{a} : = a g a^{- 1}

. It follows that

f y^{- 1}

may be written as a product of terms of the form

g_{2 i - 1}^{'} g_{2 i}^{'} (g_{2 i - 1}^{''})^{- 1} (g_{2 i}^{''})^{- 1}

where

g_{j}^{'}

and

g_{j}^{''}

are conjugate to

g_{j}

by products of the

u_{i}

Since the

u_{i}

lie in

{Symp}_{H}

Flux (g_{j}^{'}) = Flux (g_{j}^{''}) = Flux g_{j}

. Hence

Flux y = Flux f = 0

, and

y \in Ham

It remains to show that (i) implies (iii). As in the proof of Proposition 1.8 given in §1, it suffices to find a section

σ : π_{0} ({Symp}_{H}) \to {Symp}_{H}, α \mapsto σ_{α} \in {Symp}_{α},

such that

\begin{matrix} σ_{α β} σ_{β}^{- 1} σ_{α}^{- 1} \in Ham, α, β \in π_{0} (Symp) . \end{matrix}

(2.6)

We first define

σ

on the commutator subgroup

[π_{0} ({Symp}_{H}), π_{0} ({Symp}_{H})]

. When

α

lies in this group then the component

{Symp}_{α}

contains elements that are products of commutators. We define

σ_{α}

to be such an element. Then

σ_{α}

is well defined modulo an element in

Ham

because

[{Symp}_{H}, {Symp}_{H}] \cap {Symp}_{0} = Ham

by assumption.

Hence ( 2.6 ) holds for these

α

. Now we extend by hand, defining a lift on the abelian group

π_{0} ({Symp}_{H}) / [π_{0} ({Symp}_{H}), π_{0} ({Symp}_{H})] .

This is easy to do on the free part, and on the torsion part one uses the divisibility of

H^{1} (M; R) / Γ

. Note that

F

is necessarily continuous since it is continuous on

{Symp}_{0}

. □

3 Bundles with structural group $H a m s$

This section contains the proofs of the main results about the group

{Ham}^{s}

and the obstruction class. In § 3.1 we give a simple proof that

{Symp}^{H t o p}

has the extension property (Proposition 1.1 ). Because

H a m s

is geometrically defined, a similar argument shows that

H a m s

has the extension property when restricted to bundles

P \to B

where

π_{1} (B)

acts trivially on

H_{1} (M; Q)

(Proposition 3.3 (i)). It also shows that bundles with structural group

{Ham}^{s Z}

have an integral closed connection form (Lemma 3.2 ), and that conversely, any bundle with a closed integral connection form and such that

π_{1} (B)

acts trivially on

H_{1} (M; Z)

has a

{Ham}^{s Z}

-structure (Proposition 3.3 (ii)).

We start § 3.2 by defining the obstruction cocycle

O M

and then use it to prove that

H a m s

has the extension property in all cases. For completeness, we shall give most of the details of the proof. (As pointed out by Tsemo [16] , this is a special case of a more general theory that can be nicely expressed in the language of gerbes.) We then prove Propositions 1.6 and 1.5 . In § 3.3 we discuss what it means for the obstruction class to vanish and prove Theorem 1.11 and Proposition 1.13 . The section ends with a proof of the stability result Proposition 1.15 .

3.1 Groups with the extension property

We begin by proving Proposition 1.1 which states that the group

{Symp}^{H t o p}

has the extension property.

Proof of Proposition 1.1 . Suppose first that a smooth

M

-bundle

P \to B

has a closed connection form

Ω .

Because the holonomy of the corresponding connection is Hamiltonian round all contractible loops, it defines a continuous map from the space of based loops in

B

to the group

{Symp}^{H t o p}

. This deloops to a lift

B \to B {Symp}^{H t o p}

of the classifying map for

P \to B

. Therefore the classifying map of any bundle with a closed connection form does lift to

B {Symp}^{H t o p}

Conversely, consider the universal

M

-bundle

M_{{Symp}^{H t o p}} \to B {Symp}^{H t o p} .

It suffices to show that the fiberwise symplectic class

a = [ω]

extends to a class

\tilde{a} \in H^{2} (M_{{Symp}^{H t o p}}; R)

. If not, there is a map of a finite CW complex

X \to B {Symp}^{H t o p}

such that the fiberwise symplectic class in the pullback bundle

M \to P_{X} \to X

does not extend to

P_{X}

. By embedding

X

in Euclidean space and replacing it by a small open neighborhood, we may assume that

X

is an smooth (open) manifold. Hence we may suppose that

M \to P_{X} \to X

is smooth. Since the structural group is

{Symp}^{H t o p}

this bundle has a symplectic connection with holonomy in

{Symp}^{H t o p}

. The holonomy round contractible loops lies in the identity component of

{Symp}^{H t o p}

and hence is Hamiltonian. Therefore the Guillemin–Lerman–Sternberg construction provides a closed connection form

τ

P_{X}

that defines this connection: see [10,Thm 6.21] . Since

[τ] \in H^{2} (P_{X})

extends

[ω]

, this contradicts our initial assumption.

□

Corollary 3.1. Let

ℋ

be any subgroup of

Symp

whose identity component is contained in

Ham

. Consider the universal

M

-bundle

M \to M_{ℋ} \to B ℋ .

Then the fiberwise symplectic class

a : = [ω]

extends to

\tilde{a} \in H^{2} (M_{ℋ}; R)

Proof. The hypothesis on $ℋ$ implies that the inclusion $ℋ \to Symp$ factors continuously through ${Symp}^{H t o p}$ . Therefore the class $\tilde{a} \in H^{2} (M_{{Symp}^{H t o p}}; R)$ constructed above pulls back to $H^{2} (M_{ℋ})$ . □

Lemma 3.2. If

P Z ω = Z

then the universal

M

-bundle over

B {Ham}^{s Z}

carries an integral extension of

[ω]

. In other words, we may choose

\tilde{a}

to lie in the image of

H^{2} (M_{ℋ}; Z)

H^{2} (M_{ℋ}; R)

Proof. The universal $M$ -bundle over $B {Ham}^{s Z}$ carries a connection with holonomy in ${Ham}^{s Z}$ . Since this has Hamiltonian holonomy round closed loops, the GLS construction shows that it is given by a closed connection form $Ω$ . We claim that there is $b \in H^{2} (B; R)$ such that $[Ω] + π^{*} (b)$ is integral.
To see this, it suffices to consider the pullback bundle $P \to Σ$ over any map $φ : Σ \to B {Ham}^{s Z}$ , where $Σ$ denotes a Riemann surface. There are three kinds of integral classes in $H_{2} (P; R)$ : cycles lying entirely in the fiber, cycles that project nontrivially to $H_{2} (Σ)$ and cycles $C (γ, δ)$ formed as follows. Suppose that $γ$ is a closed path in the base with holonomy $m_{γ} : M \to M$ that fixes the class $δ \in H_{1} (M; Z)$ . Choose a loop $ℓ_{δ}$ in $M$ such that $〈 ℓ_{δ} 〉 = s (δ)$ , and define $C (γ, δ)$ to be the union of the cylinder $C^{'}$ formed by the parallel translation of $ℓ_{δ}$ around $γ$ with a chain $C^{''}$ in $M$ with boundary $ℓ_{δ} - m_{γ} (ℓ_{δ})$ .
Since $Ω = 0$ on $C^{'}$ and $m_{γ} \in {Ham}^{s Z}$ , it follows from equation ( 2.4 ) that $\int_{C (γ, δ)} Ω = \int_{C^{''}} ω = - {\hat{F}}_{s}^{Z} (m_{γ}) \in P Z ω \subset Z .$ Thus $[Ω]$ takes integral values on cycles of the first and third kinds.
The class $[Ω] + π^{*} (b)$ is represented in the $E_{2}$ -term of the Leray–Serre spectral sequence for the cohomology of $P \to Σ$ by a sum whose $E_{2}^{02}$ entry is $[ω]$ and whose $E_{2}^{11}$ entry is the unique element of $H^{1} (Σ, {H^{1} (M; R)})$ that equals $[Ω]$ on the cycles $C (γ, δ)$ . Both these entries are integral. By suitable choice of $b \in H^{2} (B; R)$ we can also arrange that the $E_{2}^{20}$ entry is integral. Hence result. □

Proposition 3.3. (i) Let

P \to B

be a symplectic bundle over a finite simplicial complex

B

such that

π_{1} (B)

acts trivially on

H_{1} (M; R)

. Then

P

has a closed connection form iff the classifying map for

P \to B

lifts to

B H a m s

(ii) Let

P \to B

be a symplectic bundle over a finite simplicial complex

B

such that

π_{1} (B)

acts trivially on

H_{1} (M; Z)

. Then

P

has a closed and integral connection form iff the classifying map for

P \to B

lifts to

B {Ham}^{s Z}

Proof. Corollary 3.1 shows that every $H a m s$ -bundle has a closed connection form.
Conversely, suppose that $P \to B$ has a closed connection form. Then the restriction map $H^{2} (P; R) \to H^{2} (M; R)$ contains $[ω]$ in its image. Because $Q$ is a field, the restriction map $H^{2} (P; P Q ω) \to H^{2} (M; P Q ω)$ also contains $[ω]$ in its image. Choose a class $a \in H^{2} (P; P Q ω)$ that extends $[ω]$ . Thurston's construction (cf. [10,Thm 6.3] ) provides a closed extension $Ω$ in class $a$ . We claim that the holonomy of $Ω$ round loops $γ$ in the base $B$ lies in $H a m s$ . Granted this, one can use the local trivializations given by $Ω$ to reduce the structural group to $H a m s$ .
To prove the claim, observe that because $π_{1} (B)$ acts trivially on $H_{1} (M; R)$ one can use the connection defined by $Ω$ to construct for each loop $γ$ in $B$ a $2$ -cycle $C (γ, δ_{i})$ as in Lemma 3.2 , where $[δ_{i}]$ runs through a basis of $H_{1} (M; Q)$ . Then, the $Ω$ -holonomy $m_{Ω} (γ) : M \to M$ round the loop $γ$ in $B$ satisfies the identity: ${\hat{F}}_{s} (m_{Ω} (γ)) (δ_{i}) = - \int_{C (γ, δ_{i})} Ω \in P Q ω .$ Hence $m_{Ω} (γ) \in H a m s$ . This completes the proof of (i).
The proof of (ii) is similar and is left to the reader. □

We show in the next section that part (i) of this proposition extends to arbitrary bundles. However the integral case is more subtle. Example 4.2 shows that even if

[ω]

is integral there might be manifolds for which there is no group that classifies symplectic bundles with integral closed connection form.

3.2 The obstruction class

Denote by

A

the quasitopological abelian group

H^{1} (M; R) / H^{1} (M; P ω),

where

P ω : = P Q ω

. As explained in Remark 2.6 , this is a

K (π, 1)

with

π_{1}

isomorphic to the free (discrete) abelian group

H^{1} (M; P ω)

. To simplify notation we shall often write

H_{R} : = H^{1} (M; R), H_{Q} : = H^{1} (M; P ω),

so that

A : = H_{R} / H_{Q}

Lemma 3.4. There is an obstruction class

O M \in H^{2} (B Symp; H^{1} (M; P ω)),

such that the classifying map

φ : B \to B Symp

of a symplectic bundle lifts to

B H a m s

iff

φ^{*} (O M) = 0

. Moreover

O M = 0

if there is a continuous crossed homomorphism

{\tilde{F}}_{s} : Symp \to H^{1} (M; R)

that lifts

{\hat{F}}_{s}

Proof. Consider the fibration sequence $H a m s \to Symp \overset{{\hat{F}}_{s}}{\to} A,$ that identifies $A$ as the homogeneous space $Symp / H a m s$ . There is an associated homotopy fibration $A \to B H a m s \to B Symp .$ Because $A$ is a $K (π, 1)$ , there is a single obstruction to the existence of a section of this fibration, namely a class $O M \in H^{2} (B Symp; π_{1} (A))$ . Since $π_{1} (A) \sim = H^{1} (M; P ω))$ , this proves the first statement.
To prove the second, denote by $ℋ$ the kernel of ${\tilde{F}}_{s}$ . Then $ℋ$ is a subgroup of $H a m s$ and the composite map $ℋ \to H a m s \to Symp$ is a homotopy equivalence. Hence every map $φ : B \to Symp$ lifts to $B ℋ$ and hence to $B H a m s$ . □

Proposition 3.5. Suppose that the symplectic fibration

(M, ω) \to P \overset{π}{\to} B

that is classified by

φ : B \to B Symp

has a closed connection form

Ω

. Then the obstruction class

φ^{*} (O M)

vanishes.

Proof. Consider the change of coefficients exact sequence $. . . H^{1} (B; H_{R} / H_{Q}) \overset{δ}{\to} H^{2} (B; H_{Q}) \overset{ι}{\to} H^{2} (B; H_{R}) . . .$ (where $π_{1} (B)$ may act nontrivially on the coefficients.) The map $ι$ is injective because $H_{Q}$ is a $Q$ -vector space. (Here it is essential to work over $Q$ , not $Z$ .) We shall use the connection $τ_{Ω}$ given by $Ω$ to construct a $H a m s$ -structure over the restriction $P_{1} \to B_{1}$ of $P$ to the $1$ -skeleton of $B$ and a cocycle $α \in C^{1} (B; H_{R} / H_{Q})$ such that the vanishing of $δ ([α]) \in H^{2} (B; H_{Q})$ implies that the $H a m s$ -structure extends over $P_{2} \to B_{2}$ . But $δ ([α])$ always vanishes since $ι$ is injective. Hence the classifying map $φ : B_{2} \to B Symp$ lifts to $B H a m s$ . It follows that $φ^{*} (O M) = 0$ .
The construction of the cocycle $α$ is essentially tautological. We give $B$ the structure of a CW complex with one vertex $v$ and with $1$ -cells that are embedded loops. Then it suffices to define the values $α (γ) \in A : = H_{R} / H_{Q}$ for the $1$ -cells $γ$ in this structure, where we identify $M$ with the fiber at the vertex $v$ . The $1$ -cells are loops in $B$ and we set $α (γ) = {\hat{F}}_{s} (m_{Ω} (γ)),$ where $m_{Ω} (γ) \in Symp$ is the holonomy of the connection $τ_{Ω}$ round $γ$ . To see that $α$ is a cocycle we need to check that $α (\partial e) = 0$ where $\partial e$ is the boundary of a $2$ -cell. Because $Ω$ is closed, the $τ_{Ω}$ -holonomy round any contractible loop (such as $\partial e$ ) is Hamiltonian and so lies in the kernel of ${\hat{F}}_{s}$ . This immediately translates into the fact that $α (\partial e) = 0$ .
To be more precise, we may write the boundary of $e$ as a word in generators $γ_{i}$ corresponding to the $1$ -cells in $B$ : $\partial e = γ_{i_{p}}^{ɛ_{p}} * \dots * γ_{i_{1}}^{ɛ_{1}}$ where $ɛ_{j} = \pm 1$ and we think of going first around $γ_{i_{1}}$ in some direction and then around $γ_{i_{2}}$ and so on. Denote by $f_{j}$ the $Ω$ -holonomy round $γ_{i_{j}}^{ɛ_{j}}$ . Then the identity ${\hat{F}}_{s} (g f) = {\hat{F}}_{s} (f) + f^{*} ({\hat{F}}_{s} (g))$ implies that ${\hat{F}}_{s} (f_{p} \circ \dots \circ f_{1}) = {\hat{F}}_{s} (f_{1}) + (f_{1})^{*} ({\hat{F}}_{s} (f_{2})) + \dots + (f_{1})^{*} \dots (f_{p - 1})^{*} ({\hat{F}}_{s} (f_{p})) .$ Because the total holonomy round $\partial e$ is Hamiltonian, this must vanish. But this is precisely the cocycle condition in the twisted cohomology group $H^{1} (B; A)$ .
This defines the class $[α] \in H^{1} (B; A)$ . Putting a $H a m s$ -structure over the $1$ -skeleton $B_{1}$ is equivalent to putting a flat connection $ψ$ over each $1$ -cell $γ$ whose holonomy $m_{ψ} (γ)$ lies in $H a m s$ . For each $γ$ we choose ${\tilde{g}}_{γ} \in {\tilde{Symp}}_{0}$ so that $m_{Ω} (γ) g_{γ} \in H a m s$ . Then we define $ψ$ on $γ$ so that it differs from $τ_{Ω}$ by the path ${\tilde{g}}_{γ}$ . In particular,
$\begin{matrix} m_{ψ} (γ) : = m_{Ω} (γ) g_{γ} . \end{matrix}$ (3.1)
(This step uses the fact that $H a m s$ intersects every component of $Symp$ .) The next step is to describe the obstruction to extending this $H a m s$ -connection over a given $2$ -cell $e$ . Denote by $π : P_{1} \to B_{1}$ the pullback of $P$ over $B_{1}$ . This fibration is extended over $e$ by attaching the product $D^{2} \times M$ to $π^{- 1} (B_{1})$ along the boundary loop $λ \equiv \partial e$ in $B_{1}$ by a family $Φ$ of symplectomorphisms $φ_{t} : {t} \times M \to π^{- 1} (λ_{t}), t \in S^{1} .$ The connection $τ_{Ω}$ provides a map $ι_{Ω} : [0, 1] \times M \to π^{- 1} (λ), (t, x) \mapsto ι_{Ω, t} (x)$ such that $ι_{Ω, 1} = m_{Ω} (\partial e) \circ ι_{Ω, 0} .$ To say that $Ω$ extends over $π^{- 1} (e)$ is equivalent to saying that the path $t \mapsto (ι_{Ω, t})^{- 1} φ_{t}$ is isotopic with fixed endpoints to a Hamiltonian path from the identity to $m_{Ω} (\partial e)^{- 1}$ . Similarly, if $ι_{ψ}$ denotes the trivialization along $λ = \partial e$ provided by $ψ$ , the $H a m s$ structure extends over $π^{- 1} (e)$ iff $t \mapsto (ι_{ψ, t})^{- 1} φ_{t}$ is isotopic rel.
endpoints to a Hamiltonian path from the identity to $m_{ψ} (\partial e)^{- 1}$ . Since $Ω$ does extend by hypothesis, this happens iff the path $p_{e} : t \mapsto (ι_{Ω, t})^{- 1} ι_{ψ, t}$ is isotopic rel. endpoints to a Hamiltonian path from the identity to $m_{Ω} (\partial e)^{- 1} m_{ψ} (\partial e)$ .
As above, denote the $Ω$ -holonomy round $\partial e$ by $m_{Ω} (\partial e) = f_{p} \circ \dots \circ f_{1}$ . In obvious notation the $ψ$ -holonomy is $m_{ψ} (\partial e) : = f_{p} g_{p} \circ \dots \circ f_{1} g_{1}, g_{i} \in {Symp}_{0} .$ Since each $f_{i} g_{i} \in H a m s$ , $m_{ψ} (\partial e) \in H a m s$ . Now observe that $m (\partial e) : = {(m_{Ω} (\partial e))}^{- 1} m_{ψ} (\partial e) = {(f_{p} \circ \dots \circ f_{1})}^{- 1} m_{ψ} (\partial e)$ is a product of elements of the form $(f_{k - 1} . . . f_{1})^{- 1} g_{k} (f_{k - 1} . . . f_{1})$ . Lift $m (δ)$ to an element $\tilde{m} (\partial e) \in {\tilde{Symp}}_{0}$ by replacing the elements $g_{i}$ by ${\tilde{g}}_{i} \in {\tilde{Symp}}_{0}$ . It is not hard to check that the obstruction path $p_{e}$ mentioned in the previous paragraph represents the element $\tilde{m} (\partial e) \in {\tilde{Symp}}_{0}$ . Since $\tilde{m} (\partial e) \in {\tilde{Symp}}_{0}$ , $Flux (\tilde{m} (\partial e))$ is defined as an element in $H^{1} (M; R)$ . It lies in $H_{Q} : = H^{1} (M; P ω)$ since its endpoint lies in $H a m s$ .
Moreover, the remarks in the previous paragraph imply that $Flux (\tilde{m} (\partial e)) = 0$ iff the $H a m s$ -structure extends over $π^{- 1} (e)$ .
We now relate this information to the cocycle $α$ . To find a representative for $δ ([α]) \in H^{2} (B; H_{Q})$ , we first choose a lift $\tilde{α} \in C^{1} (B; H_{R})$ of the cocycle $α$ . Then we set $δ α (e) : = \tilde{α} (\partial e)$ . This takes values in $H_{Q}$ because it vanishes in the quotient $H_{R} / H_{Q}$ .
Because, for each $1$ -cell $γ$ $0 = {\hat{F}}_{s} (m_{ψ} (γ)) = {\hat{F}}_{s} (m_{Ω} (γ)) + {\hat{F}}_{s} (g_{γ}),$ we may choose $\tilde{α} (γ) : = - Flux ({\tilde{g}}_{γ}) \in H^{1} (M; R),$ where ${\tilde{g}}_{γ}$ is as above. It is then immediate that $\tilde{α} (\partial e) = - Flux (\tilde{m} (\partial e)) \in H_{Q} .$ Hence this value determines whether or not the chosen $H a m s$ -connection extends over $π^{- 1} (e)$ . Clearly, changing the lift $\tilde{α}$ corresponds to changing the choice of ${\tilde{g}}_{γ}$ satisfying ( 3.1 ). In particular, since $δ [α] = 0 \in H^{2} (B; H_{Q})$ we may choose $ψ$ over $B_{1}$ in such a way that it extends to a $H a m s$ -structure over $B_{2}$ . This completes the proof. □

Proof of Theorem 1.10 . Corollary 3.1 shows that every

H a m s

-bundle has a closed connection form. The converse follows from Lemma 3.4 and Proposition 3.5 .

□

Remark 3.6. In the above arguments we did not use the specific form of the crossed homomorphism

{\hat{F}}_{s}

but just that its restriction to

{Symp}_{0}

may be lifted to a (possibly discontinuous) homomorphism

\tilde{{Symp}_{0}} \to H^{1} (M; R)

whose kernel is the universal cover of

Ham

. In particular, the analog of Lemma 3.4 holds if we replace

{\hat{F}}_{s}

by any crossed homomorphism

\tilde{F} : Symp \to H^{1} (M; R) / Γ

that extends Flux. Thus if

ℋ : = ker \tilde{F}

, the obstruction to the existence of an

ℋ

-structure on the bundle

P \to B

is an element of

H^{2} (B; Γ)

. Therefore it vanishes if

Γ = 0

. This is to be expected since the inclusion

ℋ : = ker (\tilde{F}) \to Symp

is then a homotopy equivalence. As we shall see in the proof of Proposition 1.6 below, most of Proposition 3.5 also goes through. All of it goes through if we consider a crossed homomorphism into a divisible group, such as

H^{1} (M; R) / Γ \otimes Q

Proof of Proposition 1.6 . Let

ℋ

be a subgroup of

Symp

with identity component

Ham

and consider the image

Im (π_{0} (ℋ))

π_{0} (ℋ)

π_{0} (Symp)

. If

ℋ

has the modified extension property then we must show that the intersection with

Im (π_{0} (ℋ))

of every finitely generated subgroup

G

π_{0} (Symp)

has finite index in

G

. But otherwise there would be a map of a finite wedge

V

of circles into

B (π_{0} (Symp))

such that no finite cover

\tilde{V}

V

lifts into the image of

B ℋ

B (π_{0} (Symp))

. Since any bundle over a

1

-complex has a closed extension form, this contradicts our assumption on

ℋ

Conversely, assume that the cokernel of

Im (π_{0} (ℋ))

π_{0} (Symp)

has the stated finiteness properties and let

{Symp}_{ℋ}

be the subgroup of

Symp

consisting of elements isotopic to

ℋ

. If

P \to B

is classified by a map into

B ℋ

then it has a closed connection form by Corollary 3.1 . Therefore we just need to see that if

φ : B \to B Symp

classifies a bundle with a closed extension form its pullback over some finite cover

\tilde{B} \to B

lifts to

B ℋ

. Observe first that the composite map

π_{1} (B) \to π_{0} (Symp) / π_{0} ({Symp}_{ℋ})

has finite image by hypothesis. (Recall that we always assume

π_{1} (B)

is finitely generated.) Therefore we may replace

B

by a finite cover such that the pullback bundle

\tilde{P} \to \tilde{B}

is classified by a map into

B {Symp}_{ℋ}

. Note that this bundle still has a closed connection form. Set

Λ \subset H^{1} (M; R)

equal to the (discrete) group

Flux (ℋ \cap {Symp}_{0})

, and then define a crossed homomorphism

F : {Symp}_{ℋ} \to H^{1} (M; R) / Λ

as in the proof of Proposition 1.8 given in § 1.2 . By Remark 3.6 there is an obstruction class

O M Λ \in H^{2} (B Symp; Λ)

whose pullback

φ^{*} (O M Λ) \in H^{2} (B; Λ)

vanishes iff

φ

lifts to

B ℋ

. As in the proof of Proposition 3.5 , consider the change of coefficients exact sequence

\cdot \cdot \cdot \to H^{1} (B; H_{R} / Λ) \overset{δ}{\to} H^{2} (B; Λ) \overset{ι}{\to} H^{2} (B; H_{R}) \to . . .

As before, we may construct an element

α \in H^{1} (B; H_{R} / Λ)

such that

O M Λ = δ ([α])

But now

ι

need not be injective: its kernel consists of torsion elements that will vanish when pulled back over a suitable finite covering map. Therefore there is an

ℋ

-structure on the pullback of

\tilde{P} \to \tilde{B}

over yet another finite cover.

□

Proof of Proposition 1.5 . The first claim is that

Ham

has the modified restricted extension property. This is a corrected statement of the conclusions that one can draw from the proof of Theorem 1.1 in [7] . The claim also follows by arguing as in the proof of Proposition 1.6 using

Flux

instead of

{\hat{F}}_{s}

; the argument can be greatly simplified because the group

{Symp}_{0}

acts trivially on the coefficients. Here one should also note that if the cover

\tilde{B} \to B

is chosen so that

H_{1} (\tilde{B}; Z)

has no torsion, then the boundary map

δ : H^{1} (B; H_{R} / H_{Q}) \to H^{2} (B; H_{Q})

vanishes.

The second claim is that when

Γ \neq 0

the group

Ham

does not have the extension property. To see this choose a nonzero element

β \in H^{1} (M; R) \ Γ

such that

2 β \in Γ

and then choose

g \in {Symp}_{0}

with

Flux (g) = β

. Consider the bundle

P \to R P^{2}

that is formed from the mapping torus bundle

M_{g} : = M \times [0, 1] / (x, 1) \sim (g x, 0) ⟶ S^{1}

by attaching

M \times D^{2}

by the map

(x, e^{2 π i t}) \mapsto (g_{t} (x), 2 t)

where

g_{t}

is a path in

Ham

from the identity to

g_{1} : = g^{- 2} \in Ham

. The flat connection on

M_{g}

pulls back to a Hamiltonian connection on the boundary

M \times \partial D

and so extends over the rest of

P

to a connection with the property that its holonomy around contractible loops in the base is Hamiltonian. Hence the GLS construction provides a closed connection form

Ω

(Alternatively one can directly construct such a form: see the proof of Proposition 3.7 .) We claim that this bundle has no Hamiltonian structure. To see this consider the classifying map

φ : R P^{2} \to B {Symp}_{0}

. Just as in the discussion before Lemma 3.4 the homomorphism

Flux : {Symp}_{0} \to H_{R} / Γ

defines an obstruction class

O M Γ \in H^{2} (B {Symp}_{0}; Γ)

such that

φ^{*} (O M Γ)

vanishes iff the bundle

P \to R P^{2}

has a Hamiltonian structure. Since

B (H_{R} / Γ)

is a

K (Γ, 2)

, this class is the pullback to

B {Symp}_{0}

of the canonical generator of

H^{2} (K (Γ, 2); Γ) .

We claim that the composite map

R P^{2} \to B {Symp}_{0} \to B (H_{R} / Γ) = K (Γ, 2)

is not null homotopic. Since

R P^{2}

is the

2

-skeleton of

R P^{\infty} = K (Z / 2 Z; 1)

and

K (Γ, 2)

is homotopy equivalent to a product of copies of

B S^{1}

, this assertion is equivalent to saying that under the map

B (Z / 2 Z) \to B S^{1}

induced by the obvious inclusion

{\pm 1} \to S^{1}

the generator of

H^{2} (B S^{1}; Z / 2 Z)

pulls back to a nonzero element of

H^{2} (R P^{2}; Z / 2 Z)

This is well known. For a direct proof identify the

2

-skeleton of

B S^{1} = C P^{\infty}

with the quotient

S^{3} / S^{1}

and observe that the

Z / 2 Z

-equivariant map

S^{2} \to S^{3}, (r, s, t) \mapsto (r + i s, \frac{1}{\sqrt{2}} (t + i t)) \in S^{3} \subset C^{2},

descends to a map

R P^{2} \to S^{2}

of (mod 2) degree

1

□

3.3 Vanishing of the obstruction class

The next proposition explains what it means for the obstruction class

O M

to vanish.

Proposition 3.7. Suppose that

(M, ω)

is a symplectic manifold. Then the following conditions are equivalent.

(i) Every symplectic

M

-bundle over a closed Riemann surface

Σ

has a closed connection form.

(ii) Every symplectic

M

-bundle has a closed connection form.

(iii) The obstruction class

O M \in H^{2} (B Symp; H^{1} (M; P ω))

vanishes.

(iv) The flux subgroup

Γ

vanishes. Further, for every closed Riemann surface

Σ

every representation of

π_{1} (Σ)

π_{0} (Symp)

lifts to a representation into a subgroup of

Symp / Ham

Proof. If $O M \in H^{2} (B Symp; H_{Q})$ does not vanish, then it evaluates on some integral $2$ -cycle to be nonzero. Since a multiple of this $2$ -cycle can be represented by a Riemann surface, Proposition 3.5 implies that (i) will not hold. Therefore (i) implies (iii).
Conversely if (iii) holds then the classifying map of any a symplectic bundle $P \to Σ$ lifts to $B H a m s$ by Lemma 3.4 . Therefore (ii) holds by Corollary 3.1 . Since (ii) obviously implies (i) the first three conditions are equivalent.
Suppose now that (iv) holds and consider any symplectic bundle $P \to Σ_{g}$ . Then one can decompose the bundle $P$ into the union of a trivial bundle $M \times D \to D$ over a $2$ -disc with a symplectically flat bundle $P^{'} \to Σ_{g} \ D$ whose holonomy round the generators of $π_{1} (Σ_{g})$ is given by elements $u_{i} \in Symp$ such that⁸ $[u_{1}, u_{2}] \dots [u_{2 g - 1}, u_{2 g}] \in {Symp}_{0}$ . The $u_{i}$ determine a representation $ρ$ of $π_{1} (Σ) \to π_{0} (Symp)$ . By hypothesis, we may homotop the $u_{i}$ to elements $f_{i} \in Symp$ such that $f : = [f_{1}, f_{2}] \dots [f_{2 g - 1}, f_{2 g}] \in Ham$ . Correspondingly there is a new flat connection on $P^{'}$ with boundary holonomy $f$ . In other words we may identify the boundary $\partial P^{'}$ with $M \times [0, 1] / (x, 1) \equiv (f (x), 0)$ . Note also that the connection form ${\tilde{ω}}^{'}$ on $P^{'}$ determined by this second connection vanishes in the horizontal directions and is closed because the connection is flat.
The bundle $P$ is reconstructed from this data by gluing the boundary $M \times [0, 1] / (x, 1) \equiv (x, 0)$ of $M \times \partial D$ to $\partial P^{'} = M \times [0, 1] / \sim$ by a map $φ : M \times S^{1} \to \partial P^{'}, (x, t) \mapsto (g_{t} (x), t)$ where $g_{1} = f^{- 1}$ . Since $f \in Ham$ and the flux group vanishes, the path ${g_{t}}_{t \in [0, 1]}$ has zero flux and we may homotop it (fixing its endpoints) into $Ham$ . It is now easy to check that $φ^{*} ({\tilde{ω}}^{'}) - p r^{*} ω$ is exact on $M \times S^{1}$ (where $p r$ is the obvious projection onto $M$ ). Hence ${\tilde{ω}}^{'}$ extends to a closed connection form on $P \to Σ$ . Thus (iv) implies (i). To see that (i) implies (iv) note first that the flux subgroup $Γ$ vanishes iff every symplectic bundle over $S^{2}$ has a closed connection form. (This is well known and may be proved by explicit calculation as in the last step in the previous paragraph.) Now consider the second condition in (iv). A representation $ρ : π_{1} (Σ_{g}) \to π_{0} (Symp)$ is determined by a collection of elements $u_{1}, . . ., u_{2 g} \in Symp$ such that $[u_{1}, u_{2}] \dots [u_{2 g - 1}, u_{2 g}] \in {Symp}_{0}$ . Given such a collection, one can, as described above, build a symplectic fibration $P \to Σ_{g}$ with fiber $M$ and holonomy $[u_{i}] \in π_{0} (Symp)$ round the $i$ th element $γ_{i}$ in a standard basis for $π_{1} (Σ_{g})$ . If (i) holds this must have a closed connection form.
The holonomy of this connection round two homotopic loops differs by an element in $Ham$ and hence determines a lift of $ρ$ to $Symp / Ham$ . Thus (i) implies (iv). □

The second part of condition (iv) above is rather hard to understand. The next results spell out some of its implications.

Lemma 3.8. The following conditions are equivalent.

(i) For every closed Riemann surface

Σ

every representation of

π_{1} (Σ)

π_{0} (Symp)

lifts to a representation into

Symp / Ham

(ii) for every symplectic

M

-bundle

P \to Σ

there is a bundle

Q \to S^{2}

such that the fiberwise connect sum

P # Q \to Σ # S^{2} = Σ

has a closed connection form.

(iii) The class

O M

takes values in

Γ

, i.e. it lies in the image of

H^{2} (B Symp; Γ)

H^{2} (B Symp; H_{Q})

Proof. We argue as in the proof of Proposition 3.7 (iv). Consider a bundle $P \to Σ$ classified by $φ : Σ \to B Symp$ , decompose it as a union of $P^{'} \to Σ \ D$ with $M \times D \to D$ and construct a flat symplectic connection on $P^{'}$ as before. Denote its connection form by ${\tilde{ω}}^{'}$ and boundary holonomy by $f \in {Symp}_{0}$ . Choose a path $f_{t}$ from $f : = f_{1}$ to the identity. Then the obstruction $φ^{*} O M ([Σ]) \in H^{1} (M; P ω)$ to extending ${\tilde{ω}}^{'}$ over the rest of $P$ is the sum of $Flux {f_{t}} \in H^{1} (M; R)$ with an element of $Γ$ given by the flux of the attaching map of $M \times D$ . If (i) holds, then as explained earlier we may assume that $f \in Ham$ . Hence in this case $φ^{*} O M ([Σ]) \in Γ$ for all $2$ -cycles $φ : Σ \to B Symp$ , i.e. (iii) holds. Conversely, if (iii) holds, we may replace $P$ by its fiber sum with a bundle $P^{''} \to S^{2}$ so that the obstruction for $P # P^{''} \to Σ$ vanishes.
Then there is a closed connection form on $P # P^{''}$ . We saw earlier that this implies that (i) holds. The equivalence of (ii) and (iii) is obvious. □

Proof of Theorem 1.11 . Part (i) is an immediate consequence of Lemma 3.4 and Theorem 1.10 . If there is an extension

\tilde{F} : Symp \to H^{1} (M; R) / Γ

Flux

, then Remark 3.6 implies that

O M

takes values in

Γ

. Hence it remains to prove the converse, i.e. that

Flux

extends if

O M

takes values in

Γ

As in the proof of Proposition 1.8 given in §1, it suffices to find a section

σ : π_{0} (Symp) \to Symp, α \mapsto σ_{α} \in {Symp}_{α},

such that

\begin{matrix} (σ_{α β})^{- 1} σ_{α} σ_{β} \in Ham, α, β \in π_{0} (Symp) . \end{matrix}

(3.2)

To do this, consider the fibration sequence

A \to B H a m s \overset{π}{\to} B Symp

of Lemma 3.4 .

By assumption the obstruction to the existence of a section

s : B Symp \to B H a m s

is an element of

H^{2} (B Symp; Γ)

where

Γ

is identified with its image in

H_{Q} = π_{1} (A)

This means that for any compatible CW structures put on

B H a m s

and

B Symp

one can choose a map

s : (B Symp)_{1} \to (B H a m s)_{1}

(where

B_{1}

denotes the

1

-skeleton of

B

) so that

π \circ s \sim i d

and so that the corresponding obstruction cocycle takes values in

Γ

. Choose a CW structure on

B Symp

with one vertex, and one

1

-cell

I \times g_{α}

for each component

α \in π_{0} (Symp)

. (This is possible because

π_{0} (Symp)

is countable.) Then for each pair

α, β

π_{0} (Symp)

there is a

2

-cell

c_{α, β}

with boundary

(I \times g_{α β})^{- 1} (I \times g_{α}) (I \times g_{β})

(There are other

2

-cells in

B Symp

coming from the

1

-skeleton of a CW decomposition for

Symp

, but these are irrelevant for the current argument.) We define a CW structure on

B H a m s

in a similar way. Then the map

s

takes each

1

-cell

I \times g_{α}

B Symp

to a loop in

(B H a m s)_{1}

. This loop is given by a word

w_{α}

in the elements of

H a m s

that represents an element

h (w_{α})

H a m s \cap {Symp}_{α}

The obstruction to extending

s

over the

2

-cell

c_{α, β}

is the homotopy class in

B H a m s

of the loop corresponding to the word

w_{α β})^{- 1} w_{α} w_{β}

. This can be identified with the homotopy class

[h (w_{α β})^{- 1} h (w_{α}) h (w_{β})] \in π_{0} (H a m s \cap {Symp}_{0}) \sim = H^{1} (M; Q) / Γ

of the element

h (w_{α β})^{- 1} h (w_{α}) h (w_{β})

. To say the obstruction

O M (c_{α, β})

takes values in

Γ

means that this class lies in

π_{0} (H a m s) = Ham

. Hence it is always possible to extend

s

over these

2

-cells (though it may not extend over the other

2

-cells in

B Symp

). Further if we define the section

σ : π_{0} (Symp) \to Symp

σ (α) : = h (w_{α})

then the identity ( 3.2 ) holds. This completes the proof.

□

Proof of Proposition 1.13 . This follows by combining Lemma 3.8 with part (ii) of Theorem 1.11 .

□

3.4 Stability

We finally discuss the question of stability.

Proof of Proposition 1.15 .

Let

N (ω)

be a path connected neighborhood of

ω

in the space of forms annihilating

V_{2} (P)

such that

P \to B

has an

ω^{'}

-symplectic structure for all

ω^{'} \in N (ω)

. Our aim is to shrink

N (ω)

so that each such

ω^{'}

has a closed extension to

P

. We claim that for each map

ψ : Σ \to B

of a Riemann surface into

B

there is a homologous map

ψ^{'} : Σ^{'} \to B

such that the pullback bundle over

Σ^{'}

admits a closed extension of

ω^{'}

, provided that

ω^{'}

is sufficiently close to

ω

. Granted this, we may choose

N (ω)

so that this holds for a finite set of

ψ_{i}

that represent a set of generators for

H_{2} (M; R)

and all

ω^{'} \in N (ω)

It follows that the obstruction class

O_{ω^{'}}^{M}

must vanish when pulled back to

B

, i.e. that

(M, ω^{'}) \to P \to B

has a closed connection form when

ω^{'} \in N (ω)

To prove the claim, consider a map

ψ : Σ : = Σ_{p} \to B

. By Proposition 3.7 we may assume that the pullback bundle

ψ^{*} P \to Σ

has a flat

ω

-symplectic connection over

Σ \ D^{2}

whose holonomy

y

around the boundary of the disc

D^{2}

may be expressed as:

y : = [u_{1}, u_{2}] \dots [u_{2 p - 1}, u_{2 p}] \in Ham (M, ω), u_{i} \in Symp (M, ω) .

Since

Ham (M, ω)

is a perfect group, we may, by increasing the genus of

Σ

and choosing the flat connection on the extra handles to have Hamiltonian holonomy, assume that

y = i d .

By hypothesis on the deformation

ω^{'}

, we can choose:

∙

a path

ω_{t}

from

ω_{0} : = ω

ω_{1} : = ω^{'}

N ([ω])

, and

∙

C^{1}

-small paths⁹

g_{i t} \in {Diff}_{0} (M)

such that

u_{i} g_{i t} \in Symp (M, ω_{t})

for all

i

and

t \in [0, 1]

Since

y = i d

the smooth path

y_{t} = [u_{1} g_{1 t}, u_{2} g_{2 t}] \dots [u_{2 p - 1} g_{(2 p - 1) t}, u_{2 p} g_{(2 p) t}]

C^{1}

-small and lies in

{Symp}_{0} (M, ω_{t})

for all

t

. If we could arrange that

y_{t} \in Ham (M, ω_{t})

for each

t

then the connection could be extended to a Hamiltonian connection over the disc for all

t

and the proof would be complete.

We show below how to modify

ψ : Σ \to B

to a map

ψ^{'} : Σ^{'} \to B

so that

y_{t} \in {Ham}^{s} (M, ω_{t}) \cap {Symp}_{0} (M, ω_{t})

for all

t

. But this finishes the proof, because the

y_{t}

must then be in

Ham (M, ω_{t})

for all

t

by the following continuity argument. Observe that for each

[ℓ] \in H_{1} (M)

and

t \in [0, 1]

, the number

Φ (t) [ℓ] : = \int_{[0, t] \times S^{1}} φ_{t, ℓ}^{*} ω_{t} \in R, where φ_{t, ℓ} (r, s) : = y_{r} (ℓ (s)),

projects to

{\hat{F}}_{ω_{t}}^{s} (y_{t}) ([ℓ]) \in R / P_{ω_{t}}

. Since, by assumption,

y_{t} \in {Ham}^{s} (M, ω_{t})

, we find that

Φ (t) [ℓ] \in P_{ω_{t}}

for all

t

. But

Φ (t)

varies continuously with

t

and

Φ (0) = 0

. Hence the fact that

V_{2} (ω_{t}) \supseteq V_{2} (ω)

implies that

Φ (t) = 0

for all

t

. It remains to check that

Φ (t)

projects to

{Flux}_{ω_{t}} (y_{t}) \in H^{1} (M; R) / Γ_{ω_{t}}

for all

t \in [0, 1]

. But because the

y_{t}

are

C^{1}

-small, for each fixed

t \in [0, 1]

the path

{y_{r}}_{r \in [0, t]}

may be canonically homotoped to a path

{y_{r t}^{'}}_{r \in [0, t]}

Ham (M, ω_{t})

by a Moser process that fixes its endpoints.

{Flux}_{ω_{t}} (y_{t})

is given by integrating

ω_{t}

over the corresponding chain

φ_{t, ℓ}^{'} : [0, t] \times S^{1} \to M

. Since this is homotopic to

φ_{t, ℓ} : [0, t] \times S^{1} \to M

mod boundary, we find that

Φ (t) = {Flux}_{ω_{t}} (y_{t})

mod

Γ_{ω_{t}}

, as required.

To complete the proof we must construct an appropriate modification of

ψ : Σ \to B

Consider the subspace

V^{1}

H^{1} (M; R)

generated by the elements

u_{i}^{*} α - α

, where

i = 1, . . ., 2 p,

and

α

runs through the elements of

H^{1} (M; R)

. If the elements

a, b \in Symp (M, ω)

are each homotopic to some

u_{i}, i = 1, . . ., 2 p

then

\begin{matrix} {\hat{F}}^{s} ([a, b]) & = & {\hat{F}}^{s} (b^{- 1}) + (b^{- 1})^{*} {\hat{F}}^{s} (a^{- 1}) + (a^{- 1} b^{- 1})^{*} {\hat{F}}^{s} (b) + (b a^{- 1} b^{- 1})^{*} {\hat{F}}^{s} (a) \end{matrix}

\begin{matrix} = & - b^{*} {\hat{F}}^{s} (b) + (a^{- 1} b^{- 1})^{*} {\hat{F}}^{s} (b) - (b^{- 1})^{*} a^{*} {\hat{F}}^{s} (a) + (b a^{- 1} b^{- 1})^{*} {\hat{F}}^{s} (a), \end{matrix}

which is easily seen to lie in

V^{1} / (V^{1} \cap H^{1} (M; P ω))

. Hence

{\hat{F}}_{ω_{t}}^{s} (y_{t}) \in V^{1} / (V^{1} \cap H^{1} (M; P_{ω_{t}})), t \in [0, 1] .

By compactness we can therefore find a finite collection of smooth families

(v_{j t}, α_{j t}), j = 1, . . ., m

, such that

\sum_{j = 1}^{m} v_{j}^{*} α_{j t} - α_{j t} \in {\hat{F}}_{ω_{t}}^{s} (y_{t}) + H^{1} (M; P_{ω_{t}}), t \in [0, 1],

where each

v_{j t} \in Symp (M, ω_{t})

is a product of the elements

(u_{i} g_{i t})^{\pm 1}, i = 1, . . ., 2 p,

and

α_{j t}

is a path in

H^{1} (M; R)

with initial point

α_{j 0} = 0

For each pair

(v_{j t}, α_{j t})

choose a path

{\tilde{h}}_{j t}

{\tilde{Symp}}_{0} (M, ω_{t})

starting at

i d

such that

{Flux}_{ω_{t}} {\tilde{h}}_{j t} = α_{j t}

. Then

{Flux}_{ω_{t}} [v_{j t}^{- 1}, {\tilde{h}}_{j t}^{- 1}] = α_{j t} - v_{j t}^{*} α_{j t} = : β_{j t} .

Therefore there is a fibration

M \to Q_{j} \to T^{2}

that admits an

ω_{t}

-symplectic structure for each

t

and a flat connection over

T^{2} \ D^{2}

whose boundary holonomy has flux

β_{j t}

. As an

ω_{t}

-symplectic bundle,

Q_{j}

is pulled back from a bundle over

S^{1}

with holonomy

v_{j t}

Our choice of

v_{j t}

implies this bundle is a pullback of

P \to B

by some map

ψ_{j} : S^{1} \to B

that we can assume to be independent of

t

(since the holonomy

β_{j t}

depends only on the homotopy class of

v_{j t}

.) However the connection varies smoothly with

t

. Therefore we can change the flux

{\hat{F}}_{ω_{t}}^{s} (y_{t})

of the boundary

ω_{t}

-holonomy of the chosen flat connection on

f^{*} P \to (Σ \ D^{2})

{\hat{F}}_{ω_{t}}^{s} (y_{t}) + β_{j t}

by replacing

ψ : Σ \to B

by the homologous map

ψ # ψ_{j} : Σ # T^{2} \to B .

Repeating this process for

j = 1, . . ., m

allows us to perform the required modification.

□

⁹ These may be constructed using the Moser method.

4 Further considerations

We begin by collecting together various observations about the groups

{Ham}^{s Z}

in the case when

P Z ω = Z

. We then explain some situations in which

{\hat{F}}_{s}

lifts to a crossed homomorphism with values in

H^{1} (M; R) / Γ

. This is followed by a short discussion of

c

-Hamiltonian bundles and covering groups.

4.1 The integral case

We shall assume throughout this section that

P Z ω = Z

. Many (but not all) of our results have some analog in the general case.

We begin by considering the integral analog of Lemma 2.5 . If

H_{1} (M; Z)

is torsion free, then Lemma 2.2 applies and the whole of this lemma extends. But if this group has torsion then it is possible that (ii) does not hold.

Lemma 4.1. Suppose that

P Z ω = Z

and set

T o r : = T o r (H_{1} (M; Z))

. Then: (i)

{\hat{F}}_{s}^{Z}

induces a crossed homomorphism

C_{s} : π_{0} (Symp) \to Hom (T o r, R / Z)

, whose kernel equals the image of

π_{0} ({Ham}^{s Z})

π_{0} (Symp)

(ii) The image

[C_{s}]

C_{s}

H_{E M}^{1} (T o r; R / Z)

is independent of the choice of

s

. In particular, if

Symp

acts trivially on

T o r

then the kernel of

C_{s}

is independent of the choice of splitting

s

(iii) There is a splitting such that

C_{s} = 0

iff

[C_{s}] = 0

Proof. We saw in Lemma 2.5 that if $[ℓ]$ has order $N$ there are precisely $N$ distinct elements of order $N$ in the coset $π_{Z}^{- 1} ([ℓ])$ , namely $〈 ℓ 〉 - 〈 γ_{(i + μ) / N} 〉$ for $i = 0, . . ., N - 1$ , where $μ$ is the area of a chain $W$ that bounds $N ℓ$ . Since ${Symp}_{0}$ is a connected group it must act trivially on these elements. Therefore, for each $g \in Symp$ the restriction of ${\hat{F}}_{s}^{Z} (g)$ to the torsion elements in $H_{1} (M; Z)$ depends only on the image of $g$ in $π_{0} (Symp)$ .
This shows that $C_{s}$ exists. Its kernel obviously contains the image of ${Ham}^{s Z}$ . To complete the proof of (i) we must show that if ${\hat{F}}_{s}^{Z} (g)$ vanishes on $T o r$ then $g$ may be isotoped to an element in ${Ham}^{s Z}$ . But this holds by the proof of Lemma 2.2 . (Note that we may assume that $dim M > 2$ here since otherwise $T o r = 0$ .) Statement (ii) holds by the argument of Proposition 2.3 : given a splitting $s$ of $π_{Z}$ over $T o r$ , any other splitting $s^{'} : T o r \to π_{Z}^{- 1} (T o r)$ has the form $s + 〈 γ_{β ([ℓ])} 〉$ where $β \in Hom (T o r; R / Z)$ . (iii) is an immediate consequence of (ii). □

Example 4.2. (i) We again assume that

P Z ω = Z

. Suppose that for some

g \in Symp

there is a loop

ℓ

such that

[ℓ]

has order

N > 1

T o r

and there is a

2

-chain

W

with boundary

g (ℓ) - ℓ

and area

i / N

, where

0 < i < N

. Then

C_{s} (g) \neq 0

for all splittings

s

, and there is no splitting such that

g

is isotopic to an element in

{Ham}^{s Z}

. Note that the corresponding mapping torus bundle

M_{g} : = M \times [0, 1] / (x, 1) \sim (g x, 0) ⟶ S^{1}

has a closed connection form but not one that is integral.¹⁰ Equivalently,

g

does not fix any integral lift

τ \in H^{2} (M; Z)

[ω]

(ii) The following yet more intriguing situation cannot be ruled out in any obvious way. Suppose that

T o r = Z / 2 \oplus Z / 2

is generated by the elements

[ℓ]

and

[ℓ^{'}]

which are interchanged by two symplectomorphisms

h_{1}, h_{2}

. Suppose further that

(h_{1})_{*} 〈 ℓ 〉 \neq (h_{2})_{*} 〈 ℓ 〉, (h_{1}^{2})_{*} 〈 ℓ 〉 = (h_{2}^{2})_{*} 〈 ℓ 〉 = 〈 ℓ 〉 .

Then

g : = h_{1} h_{2}

fixes

[ℓ]

but acts nontrivially on

〈 ℓ 〉

and so has the properties assumed in (i) above. Now consider the splittings

s_{i}

defined by

s_{i} [ℓ] = 〈 ℓ 〉, s_{i} [ℓ^{'}] = 〈 h_{i} (ℓ) 〉, i = 1, 2 .

Then

h_{i} \in {Ham}^{s_{i} Z}

by construction. The corresponding mapping tori

M_{h_{i}}

have

{Ham}^{s_{i} Z}

-structures and so Lemma 3.2 implies that each supports a closed integral connection form. But their fiber connect sum

P \to V : = S^{1} \lor S^{1}

does not, since one of its pullbacks is the bundle

M_{g} \to S^{1}

considered in (i). Indeed, the torsion in

H_{1} (M; Z)

creates new terms in

H_{2} (P; Z)

on which any closed connection form is nonintegral. But if every

M

-bundle with closed integral connection form is pulled back from some universal bundle

P \to ℬ

with this property, then

P \to V

would also be such a pullback, and hence would also have a closed integral connection form. Thus, if

Symp (M, ω)

contains elements

h_{1}, h_{2}

as above there is no universal

M

-bundle

P \to ℬ

with closed integral connection form.

A similar argument applies whenever there are two splittings

s_{i}, i = 1, 2,

such that the images of

{Ham}^{s_{i} Z}

π_{0} (Symp)

are different. It follows from Proposition 4.5 that this happens iff there are integral lifts

τ_{i}, i = 1, 2

, of

[ω]

that are stabilized by different subgroups of

π_{0} (Symp)

. To get around this difficulty, one must reformulate the classification problem: see Gal–Kȩdra [1] .

We next explain a very natural way to think of a splitting

s^{Z}

π_{Z}

when

P Z ω = Z

Denote by

τ \in H^{2} (M; Z)

some integer lift of

[ω]

and by

ρ : L_{τ} \to M

the corresponding prequantum complex line bundle. Choose a connection

1

-form

α

L : = L_{τ}

with curvature

d α = - ρ^{*} (ω)

. Then

α

determines a splitting

s : = s_{α}

as follows. By Stokes' theorem the

α

-holonomy round a nullhomologous loop

ℓ

M

is multiplication by

\begin{matrix} exp (2 π i \int_{W} ω), where \partial W = ℓ . \end{matrix}

(4.1)

(To see this observe that any such

W

can be cut open until it is a disc and so can be lifted to a

2

-disc

\tilde{W}

L

whose boundary projects to the union of

ℓ

with some arcs that are covered twice, once in each direction.) Therefore the holonomy of

α

defines a homomorphism

m_{α}

from the group

Z_{1} (M)

of integral

1

-cycles in

M

R / Z

that factors through

S H_{1} (M, ω; Z)

. Now set

s_{α} [ℓ]

to be the unique element in

π_{Z}^{- 1} [ℓ]

in the kernel of

m_{α}

. In other words, we choose

s_{α} [ℓ]

so that the

α

-holonomy round this loop vanishes. Two connection forms

α, α^{'}

L_{τ}

differ by the pullback of a closed

1

-form

β

M

. Hence if

[ℓ]

is a torsion class the element

s_{α} [ℓ]

is independent of the choice of

α

. Therefore each integer lift

τ

[ω]

determines a family of splittings

s : = s_{α}

that give rise to the same element

C_{s} \in Hom (T o r, R / Z)

Definition 4.3. We say that a splitting

s : = s^{Z}

τ

-canonical if it has the form

s_{α}

for some connection form

α

on the prequantum bundle

ρ : L_{τ} \to M

Lemma 4.4. (i) Each splitting

s

τ

-canonical for a unique bundle

L_{τ} \to M

(ii) Two connections forms

α, α^{'}

L \to M

give rise to the same splitting iff

α - α^{'}

is the pullback of an exact

1

-form on

M

(iii) Any two

τ

-canonical splittings

s, s^{'}

π_{Z}

are homotopic. Moreover the corresponding groups

{Ham}^{s Z}

and

{Ham}^{s^{'} Z}

are conjugate by an element of

{Symp}_{0}

(iv) If

s

τ

-canonical, the image of

π_{0} ({Ham}^{s Z})

π_{0} (Symp)

is the stabilizer of

τ

π_{0} (Symp)

Proof. Statement (ii) is immediate from the construction. To prove (iii) note that two connection forms $α, α^{'}$ on $L_{τ}$ differ by the pullback of a closed $1$ -form $β$ on $M$ . Hence the corresponding splittings $s_{α}$ and $s_{α^{'}}$ can be joined by a path of splittings that are constant on the torsion loops. The proof of Lemma 2.2 shows that this path can be lifted to an isotopy in ${Symp}_{0}$ . Therefore ${Ham}^{s Z}$ and ${Ham}^{s^{'} Z}$ are conjugate as in Lemma 2.5 .
To prove (i) note that the set of integer lifts $τ$ of $[ω]$ is a coset of the torsion subgroup $T o r H^{2} (M; Z)$ , while the set of splittings of $π_{Z}$ is a coset of $Hom (H_{1} (M; Z); R / Z) = H^{1} (M; R / Z)$ . By (iii) we have set up a correspondence $τ \mapsto s_{α}$ between the set of integer lifts and the components of the space of splittings. Since these are finite sets with the same number of elements, we simply have to check that this correspondence is injective.
In other words, we need to see that the isomorphism class of $L$ is determined by the set of loops that are homologically torsion and have trivial $α$ -holonomy (where $α$ is any connection $1$ -form.) But this is an elementary fact about complex line bundles.
In fact, given bundles $L, L^{'}$ with connections $α, α^{'}$ that have the same curvature and have trivial holonomy round a set of loops generating $T o r (H_{1} (M; Z))$ , one can adjust $α^{'}$ so that the monodromies agree on a full set of generators for $H_{1} (M; Z)$ and then construct an isomorphism between the two bundles by parallel translation.
Finally note that by Lemma 4.1 the image of $π_{0} ({Ham}^{s Z})$ in $π_{0} (Symp)$ is the kernel of $C_{s}$ . The proof of (i) shows that if $s$ is $τ$ -canonical then $g \in ker C_{s}$ iff $g^{*} (τ) = τ$ . This proves (iv). □

We end this section by showing that when

s

τ

-canonical the group

{Ham}^{s Z}

has a natural geometric interpretation in terms of the bundle

L_{τ} \to M

. As in Gal–Kȩdra [1] , consider the group

G : = G_{τ}

of all

S^{1}

-equivariant automorphisms of the prequantum line bundle

L : = L_{τ}

that cover a symplectomorphism of

M

. Then there is a fibration sequence of groups and group homomorphisms

M a p (M, S^{1}) \to G_{τ} \to {Symp}_{τ},

where

{Symp}_{τ}

consists of all elements in

Symp

that fix the given lift

τ \in H_{2} (M; Z)

[ω]

. Thus

{Symp}_{τ}

is a union of components of

Symp

. (Note that every

g \in {Symp}_{τ}

does lift to an element in

G_{τ}

. Indeed, since

g^{*} (τ) = τ

there is a bundle isomorphism

ψ : L \to g^{*} (L)

. But this induces an isomorphism

ψ (x) : L_{x} \to (g^{*} L)_{x} = L_{g x}

for all

x

and so is a lift of

g

.) Consider the quotient group

{\bar{G}}_{τ} : = G_{τ} / N

where

N \subset M a p (M, S^{1})

is the contractible space consisting of all null homotopic maps

M \to S^{1}

. The above fibration descends to give a central extension

\begin{matrix} H^{1} (M; Z) \to {\bar{G}}_{τ} \to {Symp}_{τ} . \end{matrix}

(4.2)

We give a formula for its defining

2

-cocycle in Remark 4.16 below. The next result explains its relation to

{Ham}^{s Z}

Proposition 4.5. Suppose that

P Z ω = Z

. Given any

τ

-canonical splitting

s

, the inclusion

{Ham}^{s Z} \to {Symp}_{τ}

lifts to a continuous group homomorphism

ι : {Ham}^{s Z} \to {\bar{G}}_{τ}

that is a homotopy equivalence.

We shall give a very explicit proof, some of whose details will be useful in the proof of Theorem 4.8 . Gal–Kedra give a shorter proof by lifting

{\hat{F}}_{s}^{Z}

to a crossed homomorphism

{\bar{G}}_{τ} \to H^{1} (M; R)

: see [1,§4] . We begin by proving the following lemma.

Lemma 4.6. The boundary map

\partial : π_{1} ({Symp}_{τ}) \to π_{0} (H^{1} (M; Z)) = H^{1} (M; Z)

in the long exact sequence for ( 4.2 ) is given by the usual flux

Flux : π_{1} ({Symp}_{0}) \to Γ \subset H^{1} (M; Z) .

Proof. Choose a connection $α$ on $L$ . Observe that the space $P_{*} {Symp}_{0}$ of based paths $(h_{t})$ in ${Symp}_{0}$ acts on $L$ by taking $(x, θ)$ to its image under $α$ -parallel translation along the path $h_{t} (x)$ . If $(h_{t})$ and $(h_{t}^{'})$ are two paths with endpoint $h$ , then $(h_{t}) \cdot (x, θ) = λ_{f} ((h_{t}^{'}) \cdot (x, θ)),$ where $λ_{f} (h (x), θ) = (h (x), θ + f (x))$ and $f (x)$ is the holonomy of $α$ around the loop based at $h (x)$ that is formed by first going back along $h_{t}^{'} (x)$ and then forward along $h_{t} (x)$ . When $(h_{t})$ and $(h_{t}^{'})$ are homotopic, the function $f : M \to S^{1} = R / Z$ is obviously null homotopic. Therefore this gives a homomorphism
$\begin{matrix} \tilde{ȷ} : \tilde{Symp} \to {\bar{G}}_{τ} . \end{matrix}$ (4.3)
We claim that $f^{*} [d s] = Flux (φ) \in H^{1} (M; R),$ where $φ = {h_{t} \circ (h_{t}^{'})^{- 1}}$ . Therefore $f$ is null homotopic if and only if the loop $φ$ is Hamiltonian. Moreover the boundary map $\partial$ has the desired form.
To prove the claim note first that the function $f : M \to S^{1}$ is homotopic to the function $f_{φ}$ defined by $f_{φ} (x) = α -holonomy round φ_{t} (x), x \in M .$ But for any loop $ℓ$ in $M$
$\begin{matrix} \int_{ℓ} f_{φ}^{*} (d s) = Flux (φ) [ℓ] . \end{matrix}$ (4.4)
To see this, choose any lift $\tilde{f} : [0, 1] \to R$ of the function $S^{1} \to S^{1} : s \mapsto f_{φ} (ℓ (s)) .$ Then equation ( 4.1 ) implies that $\tilde{f} (1) - \tilde{f} (0) = \int_{[0, 1] \times S^{1}} Ψ^{*} ω = Flux (φ) [ℓ],$ where $Ψ (t, s) = φ_{t} ℓ (s)$ . Since $f^{*} [d s] = f_{φ}^{*} [d s]$ this proves the claim. □

Next, fix a base point

x_{0}

M

and denote by

{Ham}_{*}^{s Z}

the subgroup of

{Ham}^{s Z}

that fixes

x_{0}

. Similarly for

{Symp}_{*}

Lemma 4.7. If

s

τ

-canonical then the inclusion

{Ham}_{*}^{s Z} \to Symp

lifts to a group homomorphism

\tilde{ι} : {Ham}_{*}^{s Z} \to G_{τ} .

Proof. Denote by $\tilde{M} \to M$ the maximal abelian covering of $M$ (i.e. the covering associated to the homomorphism $π_{1} (M) ⟶ H_{1} (M; Z)$ ) and let $\tilde{L} \to \tilde{M}$ be the pullback of $L_{τ} \to M$ . Denote by $G_{\tilde{τ}}$ the group of $S^{1}$ -equivariant diffeomorphisms of $\tilde{τ}$ that cover symplectomorphisms of $M$ . Denote by $\tilde{α}$ the pullback to $\tilde{L}$ of a connection $α$ on $L$ such that $s : = s_{α}$ .
For each $g \in {Symp}_{*}$ define $\tilde{g} : \tilde{L} \to \tilde{L}$ as follows: the points of $\tilde{M}$ are pairs $(x, [γ_{x}])$ where $[γ_{x}]$ is an equivalence class of paths from $x_{0}$ to $x$ and we set $\tilde{g} (x, [γ_{x}], θ) : = (g x, [g γ_{x}], θ^{'})$ where $θ^{'} \in {\tilde{L}}_{(g x, [g γ_{x}])}$ is the image of $θ \in {\tilde{L}}_{(x, [γ_{x}])}$ under $\tilde{α}$ -parallel translation, first back along $γ_{x}$ to $x_{0}$ and then forwards along $g (γ_{x})$ . This is independent of the choice of representative $γ_{x}$ for $[γ_{x}]$ because $g \in {Symp}_{*}$ . (Use equation ( 4.1 ) and the fact that $g$ preserves $ω$ .) Further the map $g \mapsto \tilde{g}$ is a group homomorphism. (The bundle $\tilde{L} \to \tilde{M}$ is independent of the choice of $τ$ because $H_{1} (\tilde{M}) = 0$ .) The group of deck transformations of the covering $\tilde{L} \to L$ is $H_{1} (M; Z)$ acting by $κ \cdot (x, [γ_{x}], θ) = (x, [γ_{x} * ℓ_{κ}], θ),$ where $ℓ_{κ}$ is a based loop representing $κ \in H_{1} (M; Z)$ and $*$ denotes concatenation of paths. We claim that the map $\tilde{g} : \tilde{L} \to \tilde{L}$ commutes with this action iff $g \in {Ham}_{*}^{s Z}$ . To see this, we choose $ℓ_{κ}$ so that $〈 ℓ_{κ} 〉 = s (κ) .$ Since $s = s_{α}$ this is equivalent to saying that the $α$ -holonomy round $ℓ_{κ}$ is zero. Then the $α$ -holonomy round $g (ℓ_{κ})$ is zero for all $ℓ_{κ}$ iff $g \in {Ham}_{*}^{s Z}$ . But the definitions imply that $g (κ \cdot (x, [γ_{x}], θ)) = κ \cdot (g x, [g γ_{x} * g (ℓ_{κ})], θ^{'} + m)$ where $θ^{'}$ is the last coordinate in $g (x, [γ_{x}], θ))$ and $m$ is the holonomy round $g (ℓ_{κ})$ .
Thus when $g \in {Ham}_{*}^{s Z}$ the element $\tilde{g}$ descends to $G_{τ}$ . Thus we may define ${\tilde{ι}}_{*} (g)$ to be the image of $\tilde{g}$ in $G_{τ}$ . □

Proof of Proposition 4.5 . Composing the homomorphism

\tilde{ι}

constructed above with the projection

p r : G_{τ} \to {\bar{G}}_{τ}

gives a section

ι : {Ham}_{*}^{s Z} \to {\bar{G}}_{τ}

π : {\bar{G}}_{τ} \to {Symp}_{τ}

. To extend this to the full group, observe that every element in

{Ham}^{s Z}

can be written as a product

h g

where

h \in Ham

and

g \in {Ham}_{*}^{s Z}

. By Lemma 4.6 ,

{\bar{G}}_{0 τ} : = π^{- 1} ({Symp}_{0})

is the covering space of

{Symp}_{0}

determined by the homomorphism

Flux : π_{1} ({Symp}_{0}) \to H^{1} (M; Z)

. Hence there is a unique lift

ȷ : Ham \to {\bar{G}}_{τ}

of the inclusion

Ham \to {Symp}_{0}

such that

ȷ (i d) = i d

. Uniqueness implies that

ȷ

is equivariant under conjugation, i.e.

\begin{matrix} ȷ (g h g^{- 1}) = ι (g) ȷ (h) ι (g)^{- 1}, g \in {Ham}_{*}^{s Z}, h \in Ham . \end{matrix}

(4.5)

We show below that

ȷ = ι

{Ham}_{*} : = Ham \cap {Ham}_{*}^{s Z}

. Hence the map

ι (h g) : = ȷ (h) ι (g), h \in Ham, g \in {Ham}_{*}^{s Z},

is well defined. It is a homomorphism by ( 4.5 ).

It remains to check that

ȷ (h) = ι (h), h \in {Ham}_{*} .

(Since

{Ham}_{*}

may not be connected, this does not follow from the uniqueness properties of the lift

ȷ

.) To this end, fix a point

(x, [γ_{x}], θ) \in \tilde{L}

and a path

(h_{t})

Ham

with

h_{1} : = h \in {Ham}_{*}

. Consider the map

Y_{x} : [0, 1] \times [0, 1] \to M, Y (s, t) = h_{t} (γ_{x} (s)) .

The boundary

Y_{x} (\partial I^{2})

contains two opposite arcs

γ_{x}, h (γ_{x})

(at

t = 0, 1

) whose

α

-holonomy determines

\tilde{ι} (h) (x, [γ_{x}], θ)

, one extra loop

t \mapsto h_{t} (x_{0})

(at

s = 0

) and another arc

t \mapsto h_{t} (x)

(at

s = 1

) whose

α

-holonomy determines

\tilde{ȷ} (h) (x, θ)

. Here

\tilde{ȷ} : \tilde{Ham} \to {\bar{G}}_{τ}

is the homomorphism defined in equation ( 4.3 ). It lifts

ȷ

by the uniqueness of

ȷ

. Let us write

(x, [γ_{x}], θ)_{3} : = θ

for the third coordinate of a point in

\tilde{L}

. Then, by ( 4.1 )

f_{h} (x) : = {(\tilde{ι} (h) (x, [γ_{x}], θ))}_{3} - {(\tilde{ȷ} (h) (x, [γ_{x}], θ))}_{3} = \int_{J} Y_{x}^{*} (α) \in R / Z,

where

J = \partial I^{2} \ {s = 0} .

If we show that the homotopy class of

f_{h}

is independent of the chosen path

(h_{t})

and the choice of

γ_{x}

, it will follow that

\tilde{ι} (h)

and

\tilde{ȷ} (h)

have the same image in

{\bar{G}}_{τ}

, i.e. that

ι (h) = ȷ (h)

as desired. We first consider the effect of changing

γ_{x}

Clearly, changing

γ_{x}

by a homotopy makes no difference. Therefore we may suppose that the integral of

α

along the concatenated loop

ℓ : = (- γ_{x}) * γ_{x}^{'}

vanishes. But then

\int_{h (ℓ)} α

also vanishes: since

h

is isotopic to the identity this integral is the flux of

h

through

ℓ

. Therefore

f_{h} (x)

does not depend on the choice of

γ_{x}

. But if we change

(h_{t})

(h_{t}^{'})

then their difference is a loop

φ

, and

f_{h} (x) - f_{h^{'}} (x) = : f_{φ} (x)

is null homotopic by equation ( 4.4 ). Hence

ι = ȷ

{Ham}_{*}

as required. This completes the definition of the homomorphism

ι : {Ham}^{s Z} \to {\bar{G}}_{τ}

. To check that

ι

is a homotopy equivalence, note that there is an exact sequence

H^{1} (M; Z) \to {\bar{G}}_{τ} \to {Symp}_{τ}

, and hence a homotopy commutative diagram

\begin{matrix} {Ham}^{s Z} & ⟶ & {Symp}_{τ} & \overset{{\hat{F}}_{s}^{Z}}{⟶} & H^{1} (M; R) / H^{1} (M; Z) \end{matrix}

\begin{matrix} ι ↓ & i d ↓ & ≃ ↓ \end{matrix}

\begin{matrix} {\bar{G}}_{τ} & ⟶ & {Symp}_{τ} & ⟶ & B H^{1} (M; Z) \end{matrix}

in which the rows are fibration sequences. (Use Lemma 4.4 (iv) to see that

π_{0} ({Ham}^{s Z})

surjects onto

π_{0} ({Symp}_{τ})

.) Hence

ι

is a homotopy equivalence by the

5

-lemma.

□

4.2 Lifting ${\hat{F}}_{s}$

We now investigate situations in which the flux homomorphism extends to a continuous crossed homomorphism

\tilde{F}

with values in

H^{1} (M; R) / Γ

. One obvious situation when this happens is when

Γ = H^{1} (M; P Z ω)

and

H_{1} (M; Z)

has no torsion since then we can take

\tilde{F} = {\hat{F}}_{s}^{Z}

. For example, we can take

M

to be the product of the standard

2 n

-torus with a simply connected manifold. The other cases in which we know that

\tilde{F}

exists have

Γ = 0

, so that in fact

{\hat{F}}_{s}

lifts to

H^{1} (M; R)

As observed by Kotschick–Morita [5] , such a lift exists in the monotone case. In this case

ω

always has a closed extension — by a multiple of the first Chern class of the vertical tangent bundle — and so the flux group vanishes by Proposition 3.7 . We now explain their argument.

Kotschick and Morita consider the symplectomorphism group

{Symp}^{δ}

with the discrete topology and observe that the universal

M

-bundle

M_{{Symp}^{δ}} \to B {Symp}^{δ}

has a natural flat connection (or foliation) that is transverse to the fibers. Therefore the fiberwise symplectic form

ω

has a closed extension

\tilde{ω}

given by the associated connection form.

Suppose now that

(M, ω)

is monotone, i.e. that the symplectic class

[ω] \in H^{2} (M)

is a nonzero multiple

λ c_{1} (T M)

of the first Chern class of the tangent bundle

T M

Then

[ω]

has another extension

λ v

where

v : = c_{1}^{V e r t}

denotes the first Chern class of the vertical tangent bundle. By construction the class

[\tilde{ω}] - λ v \in H^{2} (M_{{Symp}^{δ}})

vanishes when restricted to the fiber. Therefore it projects to a class

F_{K M} \in E_{\infty}^{1, 1} = E_{2}^{1, 1} = H^{1} (B {Symp}^{δ}; H^{1} (M; R)) = H_{E M}^{1} ({Symp}^{δ}; H^{1} (M; R)),

where

E_{\infty}^{1, 1}

is the

(1, 1)

term in the Leray–Serre spectral sequence. Each class

[F] \in H^{1} (B {Symp}^{δ}; A)

may be represented by a crossed homomorphism

F : {Symp}^{δ} \to A

whose restriction to the subgroup

{Symp}_{H}^{δ}

that acts trivially on

A

is unique. Therefore it makes sense to talk about the restriction of

F_{K M}

to the identity component

{Symp}_{0}^{δ}

Kotschick and Morita show that this is just the usual flux homomorphism.

We now show that the Kotschick–Morita class

F_{K M}

also lifts the class

[{\hat{F}}_{s}]

. We shall use the notation of § 4.1 .

Theorem 4.8. Suppose that

(M, ω)

is monotone, i.e. that the symplectic class

[ω] \in H^{2} (M)

is a nonzero multiple

λ c_{1}

of the first Chern class

c_{1} : = c_{1} (T M)

. Choose a splitting

s

that is

τ

-compatible where

τ = [c_{1}] \in H^{2} (M; Z)

Then

{\hat{F}}_{s} : Symp (M, ω) \to A

lifts to a crossed homomorphism

{\tilde{F}}_{s} : Symp (M, ω) \to H^{1} (M; R) .

Moreover this lift represents

F_{K M}

H^{1} (B {Symp}^{δ}; H^{1} (M; R))

Proof. By rescaling

[ω]

we may suppose that

P Z ω = Z

. Then

c_{1} (T M) = N [ω]

where

| N |

is the minimal Chern number. Denote by

\tilde{M} \to M

the cover of

M

associated to the homomorphism

π_{1} (M) \to H_{1} (M; Z)

, let

L \to M

be the complex line bundle with first Chern class

c_{1} (T M)

and denote by

\tilde{L} \to \tilde{M}

its pullback to

\tilde{M}

. Fix a basepoint

x_{0}

M

and let

{\tilde{x}}_{0} = (x_{0}, [γ_{0}])

be the corresponding base point in

\tilde{M}

where

γ_{0}

denotes the constant path at

x_{0}

As in Lemma 4.7 , fix a connection

α

L

, and denote by

\tilde{α}

the connection on

\tilde{L}

that is pulled back from

α

L

. (Thus

d \tilde{α}

is the pullback of

- N ω

.) Then

{Symp}_{*}

acts on the connection forms on

\tilde{L}

by pullback via the homomorphism

\tilde{ι} : {Symp}_{*} \to G_{τ}, \tilde{ι} (g) = : \tilde{g} .

For

g \in {Symp}_{*}

let

f_{g} : \tilde{M} \to R

be the unique function such that

f_{g} (x_{0}) = 0

and

{\tilde{g}}^{*} (\tilde{α}) = \tilde{α} - d f_{g} .

Then set

{\tilde{F}}_{s} (g) ([ℓ]) : = \frac{1}{N} (f_{g} (\tilde{ℓ} (1)) - f_{g} (\tilde{ℓ} (0))) \in R, g \in {Symp}_{*}, [ℓ] \in H_{1} (M; Z) / T o r,

where

ℓ

is a based loop in

M

representing

[ℓ]

and the path

\tilde{ℓ}

is its lift to

\tilde{M}

. In other words,

{\tilde{F}}_{s} (g) ([ℓ]) : = \frac{1}{N} (m_{{\tilde{g}}^{*} \tilde{α}} (\tilde{ℓ}) - m_{\tilde{α}} (\tilde{ℓ})),

where

m_{\tilde{β}} (\tilde{ℓ})

denotes the holonomy of the connection

\tilde{β}

along the path

\tilde{ℓ}

. (This is only defined mod

Z

, but the difference between two connections gives an element in

R

.) It is easy to check that

{\tilde{F}}_{s} (g) ([ℓ])

does not depend on the choice of lift

\tilde{ℓ}

It is immediate from the definitions that

{\tilde{F}}_{s} : {Symp}_{*} \to R

is a crossed homomorphism.

We claim that its mod

Z

reduction coincides with

{\hat{F}}_{s}

. Suppose first that

g [ℓ] = [ℓ] \in H_{1} (M; Z)

. Then the lifted paths

\tilde{ℓ}

and

\tilde{g} (\tilde{ℓ})

have the same endpoint and form the boundary of a

2

-chain

\tilde{C}

\tilde{M}

that lifts a chain

C

M

with boundary

C = g (ℓ) - ℓ

Hence we can calculate the mod

Z

-reduction of

{\tilde{F}}_{s} (g) ([ℓ])

as follows:

\begin{matrix} {\tilde{F}}_{s} (\tilde{g}) ([ℓ]) & = & \frac{1}{N} (m_{{\tilde{g}}^{*} \tilde{α}} (\tilde{ℓ}) - m_{\tilde{α}} (\tilde{ℓ})) \end{matrix}

\begin{matrix} = & \frac{1}{N} (m_{\tilde{α}} (\tilde{g} (\tilde{ℓ})) - m_{\tilde{α}} (\tilde{ℓ})) \end{matrix}

\begin{matrix} = & \frac{1}{N} \int_{\partial \tilde{C}} - \tilde{α} = \int_{C} ω \end{matrix}

\begin{matrix} = & {\hat{F}}_{s} (g) [ℓ] \in R / Z . \end{matrix}

In general,

{\hat{F}}_{s} (g) [ℓ]

is defined to be the area of a chain with boundary

g (ℓ) - ℓ^{'}

, where

ℓ, ℓ^{'}

are chosen to have zero

α

-holonomy (i.e. so that they project to elements in

I m s

) and

ℓ^{'}

is homologous to

g (ℓ)

. But now

\tilde{g} (\tilde{ℓ})

and

{\tilde{ℓ}}^{'}

have the same endpoint and so

\begin{matrix} {\tilde{F}}_{s} (\tilde{g}) ([ℓ]) & = & \frac{1}{N} (m_{{\tilde{g}}^{*} \tilde{α}} (\tilde{ℓ}) - m_{\tilde{α}} ({\tilde{ℓ}}^{'})), \end{matrix}

\begin{matrix} = & \frac{1}{N} (m_{\tilde{α}} (\tilde{g} (\tilde{ℓ})) - m_{\tilde{α}} ({\tilde{ℓ}}^{'})), \end{matrix}

\begin{matrix} = & {\hat{F}}_{s} (g) [ℓ] \in R / Z, \end{matrix}

as before. Thus

{\tilde{F}}_{s}

lifts

{\hat{F}}_{s}

{Symp}_{*}

We claim that

{\tilde{F}}_{s}

vanishes on

{Ham}_{*}

.¹¹ Given

g \in Symp

we now set

{\tilde{F}}_{s} (g) : = {\tilde{F}}_{s} (h g),

where

h \in Ham

is any element such that

h g \in {Symp}_{*}

. This is independent of the choice of

h

because

{\tilde{F}}_{s} = 0

{Ham}_{*}

. Moreover

{\tilde{F}}_{s} : Symp \to H^{1} (M; R)

is a crossed homomorphism. It lifts

{\hat{F}}_{s}

, since this also vanishes on

Ham

. This completes the proof of the first statement in the proposition, modulo the claim.

To prove the claim, let

h \in {Ham}_{*}

and choose any path

h_{t}

from the identity to

h : = h_{1}

such that the

α

-holonomy round the loop

h_{t} (x_{0})

is trivial. Then

h_{t}

lifts to a path in

\tilde{M}

and thence to a path

{\hat{h}}_{t} : \tilde{L} \to \tilde{L}

given by taking the

\tilde{α}

-holonomy along

h_{t} (x)

. Let

s \mapsto ℓ (s)

be a based loop in

M

with lift

\tilde{ℓ}

. Consider the map

Y_{η} : I^{2} \to \tilde{M}, (s, t) \mapsto {\hat{h}}_{t} (\tilde{ℓ} (s)),

and trivialize the bundle

Y^{*} \tilde{L}

by parallel translation along the horizontal line

t = 0

and the verticals

s = c o n s t .

Then in this trivialization the map

{\hat{h}}_{1} : {\tilde{L}}_{Y ([0, 1] \times {0})} \to L_{Y ([0, 1] \times {1})}

has the form

(s, 0, θ) \to (s, 1, θ) \in Y^{*} (\tilde{L})

while the corresponding map defined by

\tilde{h} : = \tilde{ι} (h)

has the form

\tilde{h} : (s, 0, θ) \mapsto (s, 1, θ - f (s))

where

f (s)

is the area of the rectangle

Y ([0, s] \times [0, 1])

. (Here we have used the fact that the

α

-holonomy round the loop

h_{t} (x_{0})

is trivial so that our trivialization gives the obvious identification of the fiber

L_{x_{0}}

over

(0, 0)

with that (also

L_{x_{0}}

) over

(0, 1)

Further, because

s \mapsto Y (s, 0, θ)

\tilde{α}

-parallel, the path

s \mapsto \hat{h} (Y (s, 0, θ))

is also

\tilde{α}

-parallel by definition of

\hat{h}

. Therefore, in this trivialization the restriction of

{\tilde{h}}^{*} \tilde{α}

\tilde{ℓ}

\tilde{α} - d f (s)

Hence

{\tilde{F}}_{s} (h) [ℓ] = f (1)

. But because

h_{t}

is a Hamiltonian path with no flux through

ℓ

we must have

f (1) = 0

It remains to check that

{\tilde{F}}_{s}

represents the class

F_{K M}

H_{E M}^{1} ({Symp}^{δ}; R)

. This is easy to see on the subgroup

Ham

since both crossed homomorphisms vanish there.

Hence it suffices to check this statement for

{Symp}_{*}

. The difficulty in proving this is to find a suitable way to calculate the class

v : = c_{1}^{V e r t}

We first consider the subgroup

S_{* H} : = {Symp}_{* H}

{Symp}_{*}

that acts trivially on

H_{1} (M; Z)

. The

E_{1, 1}^{2}

-term of the integral homology spectral sequence for the pullback of

M_{{Symp}^{δ}} \to B {Symp}^{δ}

B S_{* H}^{δ}

is isomorphic to the product

H_{1} (B S_{* H}^{δ}) \otimes H_{1} (M; Z)

It is generated by cycles

C_{g, ℓ} : = Z_{g, ℓ} \cup W_{g, ℓ},

where

W_{g, ℓ}

is a

2

-chain in the fiber

M_{*}

over the base point with boundary

g (ℓ) - ℓ

, and

Z_{g, ℓ}

is a cylinder lying over the loop in the base corresponding to

g \in {Symp}^{δ}

with boundary

ℓ - g (ℓ)

. Since the class

\tilde{ω}

vanishes on

Z_{g, ℓ}

\int_{C_{g, ℓ}} \tilde{ω} = \int_{W_{g, ℓ}} ω .

(Note that the mod

Z

reduction of this class is

{\hat{F}}_{s} (g) [ℓ]

as one would hope.) Since

S_{* H}

does not act on

L

but does act on the pullback

\tilde{L}

, to understand the vertical Chern class

v = c_{1}^{V e r t}

we should think of

M

as the quotient of

\tilde{M}

by the group

G : = H_{1} (M; Z)

and consider the corresponding

G

-equivariant pullback line bundle

{\tilde{L}}_{S^{δ}} \to {\tilde{M}}_{S^{δ}} .

(Here we denote

S^{δ} : = S_{* H}^{δ}

. Then

C_{g, ℓ}

lifts to a cycle

{\tilde{C}}_{g, ℓ} : = {\tilde{Z}}_{g, ℓ} \cup {\tilde{W}}_{g, ℓ}

{\tilde{M}}_{S^{δ}}

. There is a trivialization of

{\tilde{L}}_{S^{δ}}

over

{\tilde{Z}}_{g, ℓ}

that restricts to

\tilde{α}

over

\tilde{ℓ}

and to

(\tilde{g})_{*} (\tilde{α})

over

\tilde{g} (\tilde{ℓ})

Extend this to any connection

\tilde{β}

\tilde{L}

over

{\tilde{W}}_{g, ℓ}

. Then

v (C_{g, ℓ})

is given by integrating the curvature of this connection over

{\tilde{C}}_{g, ℓ}

. Since this connection is flat over

{\tilde{Z}}_{g, ℓ}

, the relevant part of the integral is over

{\tilde{W}}_{g, ℓ}

. Then because

- N \tilde{ω}

pulls back to

d \tilde{α}

\begin{matrix} 〈 \tilde{ω} - \frac{1}{N} v, C_{g, ℓ} 〉 & = & \frac{1}{N} \int_{{\tilde{W}}_{g, ℓ}} (- d \tilde{α} + d \tilde{β}) \end{matrix}

\begin{matrix} = & \frac{1}{N} (\int_{\tilde{g} (\tilde{ℓ})} \tilde{β} - \tilde{α} - \int_{\tilde{ℓ}} \tilde{β} - \tilde{α}) \end{matrix}

\begin{matrix} = & \frac{1}{N} \int_{\tilde{g} (\tilde{ℓ})} \tilde{β} - \tilde{α} \end{matrix}

\begin{matrix} = & \frac{1}{N} \int_{\tilde{ℓ}} (\tilde{g})^{*} (\tilde{β} - \tilde{α}) \end{matrix}

\begin{matrix} = & \frac{1}{N} \int_{\tilde{ℓ}} \tilde{α} - (\tilde{g})^{*} \tilde{α} = {\tilde{F}}_{s} (g) [ℓ] . \end{matrix}

To extend this argument to the full group

S : = {Symp}_{*}

we consider the

2

-chains

C_{g, ℓ}^{'} = Z_{g, ℓ} \cup W_{g, ℓ}^{'},

where

ℓ

is now a loop with trivial

α

-holonomy,

Z_{g, ℓ}

is as before, and

W_{g, ℓ}^{'}

is any

2

-chain in

M_{*}

with boundary

g (ℓ) - ℓ^{'}

, where

ℓ^{'}

is homologous to

g (ℓ)

and also has zero

α

-holonomy. Thus

\partial C_{g, ℓ}^{'} = ℓ - ℓ^{'}

, where both loops have zero

α

-holonomy.

Any element in

E_{2}^{1, 1} = H^{1} (B S^{δ}; H^{1} (M; R))

is determined by its values on the integral

2

-cycles

Z

that are sums of chains of the form

C_{g, ℓ}^{'}

with chains in the fiber

M_{*}

whose boundary consists of sums of loops with trivial

α

-holonomy. Again we lift each such cycle to a cycle

\tilde{Z}

{\tilde{M}}_{S^{δ}}

, and evaluate

v

\tilde{Z}

by integrating the curvature of a suitable connection form for the pullback of

\tilde{L}

\tilde{Z}

. As before we suppose that this connection equals

\tilde{α}

on the “free” boundary arcs

\tilde{ℓ}, {\tilde{ℓ}}^{'}

. Since these have trivial

\tilde{α}

-holonomy by construction, the pieces of

\tilde{Z}

formed from chains in

M_{*}

do not contribute to

[ω] - \frac{1}{N} v

, while the contribution of

C_{g, ℓ}^{'}

〈 \tilde{ω} - \frac{1}{N} v, C_{g, ℓ}^{'} 〉 = {\tilde{F}}_{s} (g) [ℓ]

as before. This completes the proof. □

We end this section by discussing the situation when

ω

vanishes on tori and/or spheres. Recall from §1 that the group

{Symp}_{π}

consists of all symplectomorphisms that are isotopic to an element in

{Symp}_{*}

that acts trivially on

π_{1} (M, x_{0})

, where

x_{0}

is the base point in

M

. This is equivalent to saying that the elements in

{Symp}_{π} \cap {Symp}_{*}

act on

π_{1} (M, x_{0})

by inner automorphisms. We denote by

{Symp}_{* π}

the subgroup of

{Symp}_{π} \cap {Symp}_{*}

that acts trivially on

π_{1} (M, x_{0})

. The (disconnected) group

{Ham}_{*}

is a subgroup of

{Symp}_{* π}

because the evaluation map

π_{1} (Ham) \to π_{1} (M, x_{0})

is trivial: see [9] for example. Note also that although

Γ

may not vanish when

ω = 0

π_{2} (M)

, the Flux homomorphism vanishes on loops in

{Symp}_{* π} \cap {Symp}_{0}

since its value is then given by evaluating

[ω]

2

-spheres. Therefore, in this case Flux is a well defined as a homomorphism

{Symp}_{* π} \cap {Symp}_{0} \to H^{1} (M; R)

Lemma 4.9. If

ω = 0

on spheres then

Flux

extends to a homomorphism

F_{π} : {Symp}_{* π} \to H^{1} (M; R)

Proof. Denote by

S π_{1} (M)

the group formed by equivalence classes

〈 γ 〉

of based loops in

(M, x_{0})

where two loops

γ_{0}, γ_{1}

are equivalent iff they may be joined by a based homotopy

γ_{t}

of zero symplectic area, i.e.

\int_{C} ψ^{*} ω = 0, ψ : C = S^{1} \times [0, 1] \to M, (s, t) \mapsto γ_{t} (s) .

Because

ω

vanishes on spheres, the symplectic area of a based homotopy between two homotopic loops

γ_{0}, γ_{1}

is independent of the choice of homotopy. As in Lemma 2.1 it follows readily that there is an exact sequence of groups

\begin{matrix} 0 \to R \to S π_{1} (M) \overset{π}{\to} π_{1} (M) \to {1} . \end{matrix}

(4.6)

We now define a map

s : π_{1} (M) \to S π_{1} (M)

such that

π \circ s = i d

. Since

ω

vanishes on spheres it lifts to an exact form

\tilde{ω}

on the universal cover

\tilde{M}

M

. Choose a base point

{\tilde{x}}_{0}

\tilde{M}

that projects to

x_{0}

and choose a

1

-form

β

\tilde{M}

such that

d β = \tilde{ω}

Then define

s : = s_{β}

by setting

s_{β} ([γ]) = 〈 \bar{γ} 〉

where the based loop

\bar{γ}

represents

[γ]

and is such that

β

integrates to zero over the unique lift of

\bar{γ}

to a path in

\tilde{M}

starting at

{\tilde{x}}_{0}

. Note that

s : = s_{β}

need not be a group homomorphism because

β

is not invariant under deck transformations.

For each

g \in {Symp}_{* π}

define a function

F_{π} (g) : π_{1} (M) \to R

by setting

F_{π} (g) ([γ])

equal to the symplectic area of a based homotopy joining

\bar{g (γ)} = \bar{γ}

g \bar{γ}

. We claim that the map

F_{π} (g) : π_{1} (M) \to R

is a homomorphism for each

g \in {Symp}_{* π}

and so defines an element

F_{π} (g)

H^{1} (M; R)

. To see this, define

A (γ, δ)

to be the

ω

-area of any homotopy from

\bar{γ} * \bar{δ}

\bar{γ * δ}

, where

γ, δ \in π_{1} (M)

and

\bar{γ} * \bar{δ}

denotes the loop that first goes round

\bar{γ}

and then

\bar{δ}

. Now

F_{π} (g) ([γ * δ])

is the area of any homotopy from

\bar{g (γ * δ)}

g (\bar{γ * δ})

. Consider the three part homotopy that goes first from

\bar{g (γ * δ)}

\bar{g γ} * \bar{g δ}

, then to

g \bar{γ} * g \bar{δ}

and finally to

g (\bar{γ * δ})

. The first of these homotopies has area

- A (g γ, g δ) = - A (γ, δ)

since

g

acts trivially on

π_{1}

. Therefore this cancels out the area of the third homotopy. The middle homotopy can be chosen to be the juxtaposition of the homotopies used to define

F_{π} (g) [γ]

and

F_{π} (g) [δ]

and so has area equal to their sum. Thus

F_{π} (g) ([γ * δ]) = F_{π} (g) ([γ]) + F_{π} (g) ([δ])

One now shows that

F_{π} : {Symp}_{* π} \to H^{1} (M; R)

is a homomorphism as in Proposition 2.3 .

□

Corollary 4.10. If

ω

vanishes on tori then

F_{π}

extends to a homomorphism

{Symp}_{π} \to H^{1} (M; R)

Proof. Since each

g \in {Symp}_{π}

may be joined to a point

g h_{1} \in {Symp}_{* π}

by a Hamiltonian path

h_{t} \in Ham

, we extend

F_{π}

by setting

\begin{matrix} F_{π} (g) : = F_{π} (g h), g \in {Symp}_{π}, g h \in {Symp}_{* π}, h \in Ham . \end{matrix}

(4.7)

This is independent of the choice of

h_{t} \in Ham

because

ω = 0

on tori. □

The above proof was written using geometric language, to imitate the definition of

{\hat{F}}_{s}

. However, when

g \in {Symp}_{* π}

it is perhaps more illuminating to write

F_{π} (g) \in M a p (π_{1} (M), R)

in the form

\begin{matrix} F_{π} (g) (γ) = κ_{g} (\hat{γ}), γ \in π_{1} (M, x_{0}), g \in {Symp}_{* π} \end{matrix}

(4.8)

where

κ_{g} : \tilde{M} \to R

is the unique function such that

{\tilde{g}}^{*} β = β + d κ_{g}, κ_{g} ({\tilde{x}}_{0}) = 0 .

Here

\tilde{g}

denotes the lift of

g

\tilde{M}

that fixes the base point

{\tilde{x}}_{0}

, and

\hat{γ} \in π^{- 1} (x_{0}) \subset \tilde{M}

is the end point of the lift of any representative of

γ \in π_{1} (M)

. Hence

κ_{g h} = κ_{h} + {\tilde{h}}^{*} κ_{h}, F_{π} (g h) = F_{π} (h) + h^{*} F_{π} (g) .

The above identities hold for all

g \in {Symp}_{*}

. However, when

g \in {Symp}_{* π}

we saw above that the map

F (g)

is a group homomorphism

π_{1} (M) \to R

for each

g

. In general this is not true. To explain the algebra, denote by

ψ_{γ} : \tilde{M} \to \tilde{M}

the deck transformation corresponding to

γ \in π_{1} (M)

and define the function

f_{γ} : \tilde{M} \to R

(ψ_{γ})^{*} β = β + d f_{γ}, f_{γ} ({\tilde{x}}_{0}) = 0 .

Then

\begin{matrix} f_{γ δ} - f_{δ} - (ψ_{δ})^{*} f_{γ} = f_{γ} (\hat{δ}), \end{matrix}

(4.9)

where we think of the right hand side as a constant function on

M

. Therefore

\begin{matrix} A (γ, δ) = \int_{ψ_{γ} \bar{δ}} β = \int_{\bar{δ}} (ψ_{γ})^{*} (β) = \int_{\bar{δ}} d f_{γ} = f_{γ} (\hat{δ}) . \end{matrix}

(4.10)

Hence

A : π_{1} \times π_{1} \to R

satisfies the cocycle condition

A (δ, ɛ) - A (γ δ, ɛ) + A (γ, δ ɛ) - A (γ, δ) = 0 .

(When checking this, it is useful to think of

A (γ, δ) = (ψ_{ɛ})^{*} A (γ, δ)

as a constant function on

\tilde{M}

and to use ( 4.9 ).) Then the identity

\tilde{g} \circ ψ_{γ} = ψ_{g γ} \circ \tilde{g}

implies that

f_{g γ} \circ \tilde{g} - f_{γ} = κ_{g} \circ ψ_{γ} - κ_{g} - κ_{g} (\hat{γ}) .

It follows that

F_{π} (g) (γ δ) - F_{π} (g) (γ) - F_{π} (g) (δ) = f_{g γ} (\hat{g} δ) - f_{γ} (\hat{δ}) = g^{*} A (γ, δ) - A (γ, δ) .

Therefore, if we modify ( 4.8 ) by setting

F_{π} (g, γ) (δ) : = κ_{g} (\hat{γ}) + g^{*} A (γ, δ) - A (γ, δ),

we find that

F_{π} (g, 1) (γ δ) = F_{π} (g, 1) (γ) + F_{π} (g, γ) (δ) .

This discussion applies when

ω

vanishes on

π_{2} (M)

. If

ω

also vanishes on tori¹² then

Γ = 0

and

F_{π} (g) = Flux (g)

for

g \in {Symp}_{* 0}

, so that we can extend

F_{π}

to the whole group

Symp

by equation ( 4.7 ). Hence we have shown:

Proposition 4.11. Suppose that

ω

vanishes on tori. Then Flux extends to a crossed homomorphism

F_{π} : Symp \to M a p (π_{1} (M); R),

such that

F_{π} (g) \in Hom (π_{1} (M); R)

for

g \in {Symp}_{π}

Because of the rather complicated algebraic structure of the map

F_{π} (g)

when

g

acts nontrivially on

π_{1}

, the kernel of

F_{π}

will not in general intersect every component of

Symp

. Note also that if

ω

lifts to an exact form on the abelian cover of

M

given by the homomorphism

π_{1} (M) \to H_{1} (M; Z) / T o r

, one can play the same game there, defining

F_{H}

and

A

by equations ( 4.8 ) and ( 4.10 ). Thus:

Corollary 4.12. Suppose that

[ω] \in {(H^{1} (M; R))}^{2} \subset H^{2} (M; R)

and that

ω = 0

on tori. Then

Flux

extends to a crossed homomorphism

F_{H} : {Symp}_{H} \to H^{1} (M; R),

where

{Symp}_{H}

is the subgroup of

Symp

that acts trivially on

H_{1} (M; R)

¹² If $ω$ does not vanish on tori, we still have $F (g) = Flux (g)$ mod $Γ$ , provided that the evaluation map $π_{1} {Symp}_{0} \to π_{1} (M)$ is surjective. So in this case we also get an extension of $Flux$ .

4.3 $c$ -Hamiltonian bundles and covering groups

Suppose now that

(M, a)

is a

c

-symplectic manifold, that is, a closed oriented

2 n

-dimensional manifold equipped with a class

a \in H^{2} (M; R)

such that

a^{n} > 0

. Then the analogue of the symplectic group

Symp

is the group

{Diff}^{a}

of all diffeomorphisms whose action on

H^{2} (M; R)

preserves

a

. Thus its identity component is the full group

{Diff}_{0}

. Since this is a simple group, there is no subgroup corresponding to the Hamiltonian group. However, as noted in Kȩdra–McDuff [4] there is a covering group, which may be defined as follows.

Consider the

a

-Flux homomorphism

{Flux}^{a} : π_{1} ({Diff}_{0}) \to H^{1} (M; R),

whose value on a loop

λ

{Diff}_{0} (M)

is the cohomology class

{Flux}^{a} (λ)

, where, for a

1

-cycle

γ

M

{Flux}^{a} (λ) (γ)

is given by evaluating

a

on the

2

-cycle defined by the map

T^{2} \to M, (s, t) \mapsto λ (t) (γ (s))

. Set¹³

Γ_{a} : = I m ({Flux}_{a}) \subset H^{1} (M; P_{a}^{Z}) .

We define the

c

-Hamiltonian group

H D i f f_{0}^{a}

to be the corresponding covering space of

{Diff}_{0} : = {Diff}_{0} (M)

. Thus there is a group extension:

Γ_{a} \to H D i f f_{0}^{a} \to {Diff}_{0} .

A similar construction in the symplectic case gives a group extension

Γ \to ℋ_{0} \to {Symp}_{0},

such that the inclusion

Ham ↪ {Symp}_{0}

has a natural lift to a homotopy equivalence

ι : Ham \to ℋ_{0}

. To see this, note that if

h \in Ham

, the element in

ℋ_{0}

represented by the pair

(h, γ)

, where

γ

is any path in

ℋ

from the identity to

h

, is independent of the choice of

γ

because the difference between two such paths lies in the kernel of

{Flux}^{[ω]}

Here are some analogs for

{Diff}^{a}

of the questions considered earlier.

Question 4.13. Suppose that

M \to P \to B

is a bundle with structural group

{Diff}^{a}

. When does the class

a \in H^{2} (M)

extend to a class

\tilde{a} \in H^{2} (P) ?

Question 4.14. When is there a group extension

Γ_{a} \to H D i f f^{a} \to {Diff}^{a}

that restricts over

{Diff}_{0}

to the extension

Γ_{a} \to H D i f f_{0}^{a} \to {Diff}_{0} ?

Neither question has an obvious answer. It is also not clear what relation they have to each other. For example, consider the case when

Γ_{a} : = Im ({Flux}^{a}) = 0

. Then the cover is trivial and so always extends, but

a

may not always extend. We shall see below that Question 4.14 is, at least to some extent, analogous to Question 1.7 which asks when Flux extends. However, in the present situation there is so far no analog of Theorem 1.11 which shows the close relation between the existence of an extension of Flux and the obstruction cocycle

O M

The cocycle

ɛ_{0} \in H_{E M}^{2} ({Diff}_{0}^{a}; Γ_{a})

that determines the extension

Γ_{a} \to H D i f f_{0}^{a} \to {Diff}_{0}

is defined as follows. For each

g \in {Diff}_{0}

, choose a path

γ_{g}

from the identity element to

g

and then define

ɛ_{0} (g, h)

to be the value of Flux

^{a}

on the loop formed by going along

γ_{h}

h

then along

γ_{g} h

g h

and then back along

γ_{g h}

. In the symplectic case

ɛ_{0} (g, h)

can be defined as the sum:

ɛ_{0} (g, h) = Flux (γ_{h}) + Flux (γ_{g}) - Flux (γ_{g h}) .

The next result shows that

ɛ_{0}

extends to a cocycle

ɛ

Symp

when Flux extends.

Proposition 4.15. Suppose that

Flux

extends to a crossed homomorphism

\tilde{F} : Symp \to H_{R} / Γ

, and set

ℋ_{\tilde{F}} : = ker \tilde{F}

. For each

g \in Symp

, pick an element

x_{g} \in ℋ_{\tilde{F}}

that is isotopic to

g

and choose a path

γ_{g}

{Symp}_{0}

from the identity to

g x_{g}^{- 1}

. Then: (i) The formula

\begin{matrix} ɛ (g, h) = Flux (γ_{h}) + h^{*} Flux (γ_{g}) - Flux (γ_{g h}), \end{matrix}

(4.11)

defines an element

[ɛ] \in H_{E M}^{2} (Symp; Γ)

that is independent of choices.

(ii) The inclusion

ℋ_{\tilde{F}} \to Symp

lifts to a homotopy equivalence between

ℋ_{\tilde{F}}

and the covering group

\tilde{G}

Symp

defined by

ɛ

Proof. To check (i) first observe that

x_{g}

is determined modulo an element in

Ham

and so the element

Flux (γ_{h})

is independent of choices modulo an element

c_{h} \in Γ

that depends on the choice of path

γ_{h}

. The three paths

γ_{h}, γ_{g} h x_{h}^{- 1}, - γ_{g h}

no longer make up a triangle; to close them up into a loop one needs to add a path from

g x_{g}^{- 1} h x_{h}^{- 1}

g h x_{g h}^{- 1}

. But there are elements

k, k^{'} \in Ham

such that

g x_{g}^{- 1} h x_{h}^{- 1} = g h x_{g}^{- 1} k x_{h}^{- 1} = g h x_{g h}^{- 1} k^{'}

and so we can choose this path to lie in

Ham

. It follows that formula ( 4.11 ) does define an element

ɛ (g, h) \in Γ

. The cocycle condition

δ ɛ (g, h, k) = ɛ (h, k) - ɛ (g, h k) + ɛ (g h, k) - k^{*} ɛ (g, h)

follows by an easy calculation. Moreover different choices of the paths

γ_{h}

change

ɛ

by a coboundary. This proves (i).

To prove (ii) observe that the extension

\tilde{G}

defined by

ɛ

has elements

(g, a) \in Symp \times Γ

where

(g, a) (h, b) = (g h, h^{*} a + b + ɛ (g, h)) .

Now consider the map

\tilde{f} : \tilde{G} \to H_{R}

given by

\tilde{f} (g, a) = \tilde{Flux} (γ_{g}) - a \in H^{1} (M; R),

where

\tilde{Flux} (γ_{g})

is the flux along the path

γ_{g}

in the universal cover

\tilde{Symp}

. It follows immediately from the definitions that this is a crossed homomorphism with kernel

\tilde{ℋ}

, say. Thus there is a commutative diagram with exact rows and columns

\begin{matrix} Γ & \overset{i d}{\to} & Γ \end{matrix}

\begin{matrix} ↓ & ↓ \end{matrix}

\begin{matrix} \tilde{ℋ} & \to & \tilde{G} & \overset{\tilde{f}}{\to} & H_{R} \end{matrix}

\begin{matrix} ↓ & ↓ & ↓ \end{matrix}

\begin{matrix} ℋ_{\tilde{F}} & \to & Symp & \overset{\tilde{F}}{\to} & H_{R} / Γ . \end{matrix}

Because

H_{R}

is contractible the

5

-lemma implies that the map

\tilde{G} ≃ \tilde{ℋ} \to ℋ_{ℱ}

is a homotopy equivalence. □

Remark 4.16. (i) Suppose that

P Z ω = Z

, and repeat the above argument using the crossed homomorphism

{\hat{F}}_{s}^{Z}

, where

s

τ

-compatible. By Proposition 4.5 ,

{Ham}^{s Z}

maps onto

π_{0} ({Symp}_{τ})

. Hence

{Ham}^{s Z}

is homotopy equivalent to the covering group of

{Symp}_{τ}

defined by the cocycle

ɛ_{s} \in H_{c E M}^{2} ({Symp}_{τ})

of ( 4.11 ). It follows from Proposition 4.5 that this covering group may be identified with

{\bar{G}}_{τ}

(ii) If

F : G \to A / Λ

is a continuous crossed homomorphism then there is an associated (possibly discontinuous) extension cocycle

ɛ_{F} \in H_{E M}^{2} (G; Λ))

defined by

ɛ_{F} (g, h) = h^{*} f (g) + f (h) - f (g h)

, where

f : G \to A

is any lift of

F

. It is not hard to check that the corresponding extension of

G

coincides with the covering group

\tilde{G}

constructed from

F

in Remark 1.21 (iii).

Observe finally that we can relax Question 4.14 by asking for an extension of

{Diff}^{a}

by the group

H^{1} (M; P_{a}^{Z})

rather than by

Γ

. If

a

is a primitive integral class (i.e.

P_{a}^{Z} = Z

), there are natural candidates for such an extension just as in the symplectic case.

Moreover the existence of these groups should have some bearing on Question 4.13 .

Since these questions are very similar to those already discussed in connection with the group

{Ham}^{s Z}

, we shall not pursue them further here.

References

S. Gal and J. Kȩdra, Symplectic Configurations, in preparation.
M. Gotay, R. Lashof, J. Sniatycki, and A. Weinstein, Closed forms on symplectic fiber bundles. Commentarii Mathematicae Helveticae, 58 (1983), 617–21.
J. Kȩdra, D. Kotschick and S. Morita, in preparation.
J. Kȩdra and D. McDuff, Homotopy properties of Hamiltonian group actions, v.2 SG/0404539, Geometry and Topology (2004).
D. Kotschick and S. Morita, Signatures of foliated surface bundles and the symplectomorphism groups of surfaces, to appear in Topology.
D. Kotschick and S. Morita, Characteristic classes of foliated surface bundles with area-preserving holonomy, SG/0409214.
F. Lalonde and D. McDuff, Symplectic structures on fiber bundles, SG/0010275, Topology 42 (2003), 309–347.
D. McDuff, Symplectic diffeomorphisms and the flux homomorphism, Inventiones Mathematicae, 77, (1984), 353–66.
D. McDuff, A survey of topological properties of groups of symplectomorphisms, in Topology, Geometry and Quantum Field Theory, Proceedings of 2002 Symposium in honor of G.B. Segal, ed U.L. Tillmann, Cambridge Univ. Press (2004), 173–193.
D. McDuff and D. Salamon, Introduction to Symplectic Topology, 2nd edition (1998) OUP, Oxford, UK.
D. McDuff and D. Salamon, $J$ -holomorphic curves and Symplectic Topology, Colloquium Publications, Amer. Math Soc. (2004).
D. McDuff and D. Salamon, Erratum to Introduction to Symplectic Topology, preprint (2004).
K. Ono, Floer–Novikov cohomology and the flux conjecture, 2004.
L. Polterovich, Growth of maps, distortion in groups and symplectic geometry, Invent Math. 150 (2002), 655–686.
P. Seidel, Lectures on four-dimensional Dehn twists, SG/0309012
A. Tsemo, Gerbes, $2$ -gerbes and symplectic fibrations, in preparation.

Department of Mathematics, Stony Brook University, Stony Brook, NY 11794-3651, USA E-mail address : dusa@math.sunysb.edu URL: http://www.math.sunysb.edu/~dusa

¹³ By McDuff [8,Thm 6] $Γ_{a}$ need not be discrete.

6 January 2005, revised 14 March 2005

Dusa McDuff