Topological complexity of symplectic manifolds

We prove that the topological complexity of every symplectically atoroidal manifold is equal to twice its dimension. This is the analogue for topological complexity of a result of Rudyak and Oprea, who showed that the Lusternik--Schnirelmann category of a symplectically aspherical manifold equals its dimension. Symplectically hyperbolic manifolds are symplectically atoroidal, as are symplectically aspherical manifolds whose fundamental group does not contain free abelian subgroups of rank two. Thus we obtain many new calculations of topological complexity, including iterated surface bundles and symplectically aspherical manifolds with hyperbolic fundamental groups. Our result also applies in the greater generality of cohomologically symplectic manifolds.


Introduction
Topological complexity is a numerical homotopy invariant, originally defined by M. Farber in [Far03] and motivated by the motion planning problem from robotics. Let X be a topological space, and let P X = C 0 ([0, 1], X) be the space of paths in X endowed with the compact-open topology. Consider the free path fibration π : P X → X × X given by γ → (γ(0), γ(1)). The topological complexity of X, denoted TC(X), is the minimal number k (or infinity) for which there exists an open cover {U 0 , U 1 , . . . , U k } of X × X with the property that for each j ∈ {0, 1, . . . , k} there exists a continuous map s j : U j → P X with π • s j = incl j : U j ֒→ X × X. (Note that we use the reduced topological complexity, which is one less than Farber's original definition.) Due to its potential applicability and intrinsic interest, this invariant has attracted a lot of attention from homotopy theorists in recent years. Overviews of its main properties and applications are given in [Far06] and [Far08,Chapter 4].
The invariant TC(X) is similar in spirit and properties to the classical Lusternik-Schnirelmann category cat(X)-the minimal number k for which there exists an open cover {U 0 , U 1 , . . . , U k } of X such that each inclusion U j ֒→ X is null-homotopic. (Here again we are using the reduced version.) Both invariants possess upper bounds in terms of dimension and connectivity. Namely, if X is an (r − 1)-connected CW complex, then cat(X) ≤ dim X/r and TC(X) ≤ 2 dim X/r.
Proofs of these bounds may be found in [CLOT03, Theorem 1.50] for cat(X), and in [Far04,Theorem 5.2] for TC(X). Both invariants also admit elementary lower bounds in terms of cup-length in cohomology. Taking coefficients for cohomology in an arbitrary commutative ring, recall that the cup-length of X, denoted cl(X), is the minimal k such that any product of k +1 elements of H * (X) vanishes. The zero-divisors cup-length of X, denoted zcl(X), is the minimal k such that any product of k + 1 elements in the kernel of the homomorphism ∆ * : H * (X × X) → H * (X) induced by the diagonal vanishes. Then we have cat(X) ≥ cl(X) and TC(X) ≥ zcl(X).
There are analogous statements with local coefficients. These statements, which may be found in [CLOT03, Proposition 1.5] and [Far03,Theorem 7] respectively, are both specializations of [Sch66,Theorem 4]. Indeed, both cat(X) and TC(X) can be seen as special cases of the notion of Schwarz genus (or sectional category) of fibrations. The goal of this article is to give new computations of the topological complexity of symplectic manifolds. Our results will apply more generally to the class of cohomologically symplectic, or c-symplectic manifolds [LO95]. Our convention is that a c-symplectic manifold is a pair (M, ω) consisting of a closed manifold M of even dimension 2n, together with a closed 2-form ω ∈ Ω 2 (M ) such that the n-th power [ω] n ∈ H 2n (M ; R) of the cohomology class represented by ω is nonzero. (Choosing a representative of the c-symplectic class gives a slight refinement of the definition used in [LO95], which is unnecessary but simplifies some statements.) In particular, a closed symplectic manifold (M, ω) is c-symplectic. If M is a simply connected c-symplectic manifold, then the bounds given above (together with a simple cohomology calculation in the case of TC, carried out in [FTY03]) imply that cat(M ) = n and TC(M ) = 2n.
Our interest therefore lies in the non-simply connected case. Here, one can often use the notion of category weight of cohomology classes, introduced by E. Fadell and S. Husseini [FH92] and later refined by Y. Rudyak [Rud99] and J. Strom [Str97], in order to improve on the cup-length lower bound for cat(M ). Recall that a (c-)symplectic manifold (M, ω) is called (c-)symplectically aspherical if for all smooth maps f : S 2 → M . Using category weight, Rudyak and J. Oprea derived the following result, a key tool in Rudyak's proof of the Arnold Conjecture for symplectically aspherical manifolds. Hence for c-symplectically aspherical manifolds, the dimensional upper bound for cat(M ) is attained. The analogous statement for topological complexity (that TC(M ) = 2 dim M when M is c-symplectically aspherical) does not hold. For example, the torus T 2 = S 1 × S 1 equipped with its standard volume form is aspherical, therefore symplectically aspherical. However, as shown in [Far03], we have TC(T 2 ) = 2 < 4 = 2 dim T 2 .
Our main result identifies an extra condition which ensures that the topological complexity of a c-symplectically aspherical manifold achieves its dimensional upper bound, thereby giving an analogue of Theorem 1.1 for topological complexity. We say that a (c-)symplectic manifold for all smooth maps f : T 2 → M . Note that since there is a degree-one map T 2 → S 2 , every c-symplectically atoroidal manifold is c-symplectically aspherical. The torus itself is symplectically aspherical but not symplectically atoroidal, as are many symplectic nilmanifolds or solvmanifolds. J. Kȩdra has shown that every symplectically hyperbolic manifold is symplectically atoroidal [Kȩd09], while A. Borat has shown that every symplectically aspherical manifold whose fundamental group does not contain subgroups isomorphic to Z ⊕ Z is symplectically atoroidal [Bor16]. Thus our result applies to many examples, including iterated bundles of higher genus surfaces, symplectic manifolds of negative sectional curvature, and symplectically aspherical manifolds with hyperbolic fundamental group. We mention also that M. Brunnbauer and D. Kotschick [BK08] have given an example of a closed 4-manifold which is symplectically atoroidal without being symplectically hyperbolic.
Symplectic atoroidality may be viewed as imposing a condition of asymmetry on a symplectic manifold, in the following sense (we are indebted to John Oprea for the statement and its proof). Proposition 1.3. A symplectically atoroidal manifold (M, ω) does not admit any non-trivial symplectic S 1 -actions.
Thus our results tend to support the idea, explored in [Gra12], that symmetries of manifolds are responsible for lowering their topological complexity. Note however that D. Cohen and L. Vandembroucq have recently shown [CV17] that the Klein bottle K has TC(K) = 2 dim(K), despite admitting an effective S 1 -action.
The proof of Theorem 1.2 employs the notion of TC-weight of cohomology classes. This analogue of category weight was introduced and studied by M. Farber and the first author in [FG07] and [FG08], in order to improve on the zero-divisors cup-length lower bound for topological complexity. A class u ∈ H * (X × X) has wgt(u) ≥ 1 if and only if it is a zero-divisor, and has wgt(u) ≥ 2 if and only if it is in the kernel of the homomorphism induced by the fibrewise join p 2 : P 2 X → X × X of the free path fibration with itself. We show that for a closed atoroidal form ω ∈ Ω 2 (M ) the associated zero-divisor [ω] = 1 × [ω] − [ω] × 1 ∈ H 2 (M × M ; R) has wgt([ω]) = 2, by analyzing the Mayer-Vietoris sequence associated to the fibrewise join P 2 M .
Our methods use de Rham theory on infinite-dimensional manifolds of smooth paths and loops, for which we adopt the formalism of Kriegl and Michor [KM97]. Special care needs to be taken in passing between de Rham and singular cohomology, because the space P 2 M is not a manifold.
The paper is organized as follows. In Section 2, we recall the definition of the TC-weight of cohomology classes and its most important properties before relating it to fiberwise joins and their Mayer-Vietoris sequence. Section 3 looks at this Mayer-Vietoris sequence from a different angle in the case of manifolds, where we derive an explicit description of the connecting homomorphism in terms of differential forms. This description is applied to c-symplectic manifolds in Section 4 and yields a proof of Theorem 1.2. We apply this theorem in Section 5 to derive several new computations of the topological complexity of c-symplectic manifolds.

The TC-weight of a cohomology class and fiberwise joins
Throughout this section we fix a topological space X. We let P X = C 0 ([0, 1], X) and LX = C 0 (S 1 , X) denote the continuous path and loop spaces, resp. For a local coefficient system A on X × X, we will call a cohomology class where ∆ : X → X × X denotes the diagonal map.
In the following, we will briefly recall the notion of TC-weight from [FG07] and [FG08]. It is an analogue of the strict category weight defined by Y. Rudyak in [Rud99].
Definition 2.1. Let A be a local coefficient system on X × X and k ∈ N 0 . The TC-weight of a class u ∈ H * (X × X; A), denoted wgt(u), is defined to be the maximal k such that The following properties of TC-weight, proven by M. Farber and the first author in [FG07], illustrate its importance for obtaining estimates from below of topological complexity.
Theorem 2.2. Let A and B be local coefficient systems on X × X, and let u ∈ H * (X × X; A) and v ∈ H * (X × X; B) be cohomology classes.
A class in H * (X × X; A) has positive TC-weight if and only if it is a zero-divisor, see [FG08,p. 3361]. In the remainder of this section, we want to determine a criterion to decide if a given cohomology class has TC-weight at least 2. A key role is played by the fiberwise join of the fibration π : P X → X × X with itself, which we will denote by p 2 : P 2 X → X × X, see e.g. [FSGKV06, Section 2] for a detailed construction. More precisely, a result of Schwarz from [Sch66] applied to the fibration π shows the following: We want to take a closer look at the cohomology of P 2 X. We define continuous maps r 1 , r 2 : LX → P X by One checks without difficulties that (r 1 , r 2 ) : LX → P X × P X maps LX homeomorphically onto the pullback of P X π → X × X π ← P X, such that the following is a pullback diagram: The total space of the fibration P 2 X is a homotopy pushout of the diagram P X r1 ← LX r2 → P X, and we let i 1 , i 2 : P X → P 2 X and p 2 : P 2 X → X × X be the induced maps making the following diagram commutative up to homotopy: In fact, by taking P 2 X to be the double mapping cylinder of the maps r 1 and r 2 , and taking p 2 to be the whisker map induced by the constant homotopy across the diagram (2), we can arrange that the triangles in the above diagram strictly commute. We want to consider the Mayer-Vietoris cohomology sequence associated with this homotopy pushout, see [Str11,Chapter 21], which is of the form for any abelian group A.
Proposition 2.4. Let A be an abelian group and k ∈ N.
, where δ denotes the connecting homomorphism of the above Mayer-Vietoris sequence.
Proof. We consider the map e : P X → X, γ → γ(0), given by evaluating a path at its initial point. For each ν ∈ {1, 2}, the diagram commutes up to homotopy. Since the vertical map e is well-known to be a homotopy equivalence, it follows that 2} and the exactness of the Mayer-Vietoris sequence shows the claim.
We next combine the previous propositions to obtain a criterion for a cohomology class to have TC-weight two.
Corollary 2.5. Let A be an abelian group and k ∈ N with k ≥ 2. If the connecting homomorphism from the above Mayer-Vietoris sequence δ : H k−1 (LX; A) → H k (P 2 X; A) vanishes, then every nontrivial zero-divisor u ∈ H k (X × X; A) satisfies wgt(u) ≥ 2.
Proof. This is an immediate consequence of Propositions 2.3 and 2.4.

The connecting homomorphism for manifolds
We want to make use of Proposition 2.4 to investigate zero-divisors that are induced by degreetwo cohomology classes of a manifold. More precisely, given a manifold M and a closed 2-form ω ∈ Ω 2 (M ) we want to determine a 1-cocycle on LM whose cohomology class maps to the class To work in a de-Rham-theoretic setting, we briefly discuss differential forms and de Rham cohomology on smooth path and loop spaces of manifolds. Let M be a finite-dimensional smooth manifold. We consider the spaces  Proof. The inclusion of smooth paths into continuous paths gives a commuting diagram For k ∈ N we let Ω k (M ) denote the space of smooth real-valued k-forms on M . For each ω ∈ Ω k (M ) we letω ∈ Ω k (M × M ) denote the form given bȳ ω = 1 × ω − ω × 1 := pr * 2 ω − pr * 1 ω. It is easy to see that if ω is closed, thenω will be closed as well. Moreover, since pr 1 •∆ = pr 2 •∆, the class [ω] ∈ H k (M × M ; R) will be a zero-divisor for every closed ω ∈ Ω k (M ).
In analogy with the finite-dimensional case, we define the space of smooth k-forms on PM for each k ∈ N by Ω k (PM ) = Γ(L k a (T PM )), where L k a (T PM ) denotes the bundle of alternating k-linear forms on T PM . We define Ω k (ΛM ) similarly.
It is a consequence of [KM97,Theorem 34.7] and [KM97,Proposition 42.3] that the de Rham cohomology groups of PM and ΛM are well-defined, and that the de Rham Theorem holds for these manifolds. Indeed, for any smoothly paracompact smooth manifold M (finite or infinite dimensional) the composition of cochain maps Proof. By definition, the connecting homomorphism is obtained using the snake lemma in the first two rows of the following diagram: where C i (−; R) denotes singular cochains with real coefficients and ∂ denotes the respective singular codifferential. The (strict) commutativity of the bottom triangle follows from the remark made immediately after diagram (3).
To find a cocycle a ∈ C 1 (ΛM ; R) whose cohomology class is mapped to the class of p * 2 c by the connecting homomorphism for a given cocycle c ∈ C 2 (M × M ; R), it thus suffices to find such an a for which there exist b 1 , b 2 ∈ C 1 (PM ; R) such that a = r * 1 b 1 − r * 2 b 2 and ∂b 1 = ∂b 2 = π * c. Passing to differential forms, we have seen in Propositions 3.3 and 3.4 that dβ ω = π * ω and α ω = r * 1 β ω − r * 2 β ω . The result now follows from the naturality of the de Rham cochain equivalences Ψ M×M , Ψ PM and Ψ ΛM .

Applications
In this section we give some general settings in which our main result Theorem 1.2 applies.
Note that this result improves by one the bound given in [FM17,Theorem 8]. It applies in particular to the case of hyperbolic fundamental groups. Proof. By the Theorems of Cartan-Hadamard and Preissman, M is aspherical and π 1 (M ) does not contain any subgroups isomorphic to Z ⊕ Z. Thus Theorem 5.1 applies.
We next discuss a related geometric condition which implies symplectic atoroidality.
Definition 5.4. Let (M, ω) be a symplectic manifold and let p :M → M be its universal cover. The pair (M, ω) is called symplectically hyperbolic if there exists θ ∈ Ω 1 (M ), such that p * ω = dθ and such that sup where · q denotes the norm on T * qM that is induced by the lift of a chosen Riemannian metric on M .
This notion was introduced by L. Polterovich in [Pol02], generalizing the concept of Kähler hyperbolicity introduced by M. Gromov in [Gro91]. It has further been discussed by J. Kȩdra in [Kȩd09] and by G. Paternain under the name of weakly exact forms with bounded primitive in [Pat06]. The simplest examples of symplectically hyperbolic manifolds are oriented surfaces of genus at least two together with their volume forms. The connection to our results is given by the following observation. Hence, Theorem 1.2 applies to all closed symplectically hyperbolic manifolds.
In particular, if E is the total space of an iterated sequence of oriented surface bundles, where the fiber of each iteration step and the base space of the first fibration are oriented surfaces of genus at least two, then TC(E) = 2 dim E.
Finally, we give examples of c-symplectic but not symplectic manifolds to which our Theorem applies.
Proposition 5.7. Let (M, ω) be a c-symplectically atoroidal manifold (with its induced orientation), and let g : N → M be map of nonzero degree. Then the pair (N, g * ω) is c-symplectically atoroidal.
Proof. Since g induces an isomorphism on top cohomology, it follows that the pair (N, g * ω) is a c-symplectic manifold. If f : T 2 → N is a smooth map, then since ω is atoroidal. Hence g * ω is atoroidal.
Example 5.8. Let (M, ω) be a c-symplectically atoroidal manifold, and let g :M → M be a branched covering of nonzero degree. Then (M , g * ω) is c-symplectically atoroidal, and so TC(M ) = 2 dimM . Note that g * ω fails to be symplectic over the branch locus. We do not know ifM necessarily carries a symplectic structure.
Example 5.9. Let (M, ω) be a closed symplectically atoroidal 4-manifold, and let X be any almost complex 4-manifold (with the induced orientation). As shown in [Aud91, Proposition 1.3.1], the connected sum M #X does not possess any almost complex structure, and hence cannot possess any symplectic structure. However, there is a degree-one map g : M #X → M collapsing X to a point, which by Proposition 5.7 implies that M #X is c-symplectically atoroidal. Hence TC(M #X) = 8.