The closed-open string map for S 1 -invariant Lagrangians
TTHE CLOSED-OPEN STRING MAP FOR S -INVARIANTLAGRANGIANS DMITRY TONKONOG
Abstract.
Given a monotone Lagrangian submanifold invariant under a loop ofHamiltonian diffeomorphisms, we compute a piece of the closed-open string mapinto the Hochschild cohomology of the Lagrangian which captures the homologyclass of the loop’s orbit.Our applications include split-generation and non-formality results for realLagrangians in projective spaces and other toric varieties; a particularly basicexample is that the equatorial circle on the 2-sphere carries a non-formal Fukaya A ∞ algebra in characteristic two. Introduction
Overview of main results.
Let X be a compact monotone symplectic man-ifold, L ⊂ X a monotone Lagrangian submanifold, and K be a field. We assumethat L satisfies the usual conditions making its Floer theory well-defined over K ,namely, L has Maslov index at least 2, and is oriented and spin if char K (cid:54) = 2. Inthis case, one can define a unital algebra over K , the Floer cohomology HF ∗ ( L, L ),which is invariant under Hamiltonian isotopies of L . A larger amount of informa-tion about L is captured by the Fukaya A ∞ algebra of L , and given this A ∞ algebra,one can build another associative unital algebra called the Hochschild cohomology HH ∗ ( L, L ). There is the so-called (full) closed-open string map CO ∗ : QH ∗ ( X ) → HH ∗ ( L, L ) , which is a map of unital algebras, where QH ∗ ( X ) is the (small) quantum cohomol-ogy of X . This map is of major importance in symplectic topology, particularlyin light of Abouzaid’s split-generation criterion [1, 46, 39, 2], one of whose ver-sions in the case char K = 2 says the following: if the closed-open map is injective ,then L split-generates the w -summand F uk ( X ) w of the Fukaya category, where w = w ( L ) ∈ K is the so-called obstruction number of L . (When char K (cid:54) = 2, thehypothesis can be weakened to say that CO ∗ is injective on a relevant eigensum-mand of QH ∗ ( X ); we will recall this later.)Split-generation of the Fukaya category F uk ( X ) w by a Lagrangian submani-fold L is an algebraic phenomenon which has important geometric implications.For example, in this case L must have non-empty intersection with any othermonotone Lagrangian submanifold L (cid:48) which is a non-trivial object in F uk ( X ) w ,namely such that HF ∗ ( L (cid:48) , L (cid:48) ) (cid:54) = 0 and w ( L (cid:48) ) = w . Another application, thoughnot discussed here, is that split-generation results are used in proofs of homologicalmirror symmetry. a r X i v : . [ m a t h . S G ] J un DMITRY TONKONOG
The present paper contributes with new calculations of the closed-open map,motivated by the split-generation criterion and the general lack of explicit calcula-tions known so far. (The closed-open map is defined by counting certain pseudo-holomorphic disks with boundary on L , which makes it extremely hard to computein general.)There is a simplification of the full closed-open map, called the “zeroth-order”closed-open map, which is a unital algebra map CO : QH ∗ ( X ) → HF ∗ ( L, L ) . It is the composition of CO ∗ with the canonical projection HH ∗ ( L, L ) → HF ∗ ( L, L ),and if CO is injective, so is CO ∗ (but not vice versa). Although CO generallycarries less information than CO ∗ , it is sometimes easier to compute. For exam-ple, we compute CO when L is the real locus of a complex toric Fano variety X ,see Theorem 1.13. This map turns out to be non-injective in many cases, e.g. for R P n +1 ⊂ C P n +1 over a characteristic 2 field. The aim of the present paper is tostudy the higher order terms of the full closed-open map CO ∗ , and to find exampleswhen CO ∗ is injective but CO is not.Specifically, let us consider the following setting: a loop γ of Hamiltonian sym-plectomorphisms preserves a Lagrangian L setwise. Let S ( γ ) ∈ QH ∗ ( X ) be theSeidel element of γ , then from Charette and Cornea [12] one can see that CO ( S ( γ )) = 1 L , the unit in HF ∗ ( L, L ). Our main result, Theorem 1.7, is a tool for distinguishing CO ∗ ( S ( γ )) from the Hochschild cohomology unit in HH ∗ ( L, L ); this way it cap-tures a non-trivial piece of the full closed-open map CO ∗ not seen by CO . Weapply Theorem 1.7 to show that CO ∗ is injective for some real Lagrangians in toricmanifolds, and also for monotone toric fibres which correspond to (non-Morse) A -type critical points of the Landau-Ginzburg superpotential.After this paper had appeared, Evans and Lekili [20] proved split-generation forall orientable real toric Lagrangians, and all monotone toric fibres in zero charac-teristic, by completely different methods. They make use of the fact that theseare homogeneous Lagrangians (i.e. they are orbits of Hamiltonian group actions),while we only use the fact these Lagrangians are invariant under certain Hamilton-ian loops.We will now mention our examples regarding real Lagrangians, and postpone alldiscussion of monotone toric fibres, along with an introductory part, to Section 4. Proposition 1.1.
Let K be a field of characteristic and R P n be the standard realLagrangian in C P n . Then CO ∗ : QH ∗ ( C P n ) → HH ∗ ( R P n , R P n ) is injective forall n . In contast, CO : QH ∗ ( C P n ) → HF ∗ ( R P n , R P n ) is injective if and only if n is even. Corollary 1.2.
Over a field of characteristic , R P n split-generates F uk ( C P n ) . As hinted above, this corollary leads to a result on non-displaceability of R P n from other monotone Lagrangians which are Floer-theoretically non-trivial. Thishas been known due to Biran and Cornea [9, Corollary 8.1.2], and Entov andPolterovich [19]. Very recently Konstantinov [30] showed that the Chiang La-grangian in C P admits a higher rank local system making it Floer-theoreticallynon-trivial over a characteristic two field; then he concludes via Corollary 1.2 that HE CLOSED-OPEN MAP FOR S -INVARIANT LAGRANGIANS 3 the Chiang Lagrangian is non-displaceable from R P . It is possible that for La-grangians with higher rank local systems, a generalisation of [9, 19] can be invokedinstead of Corollary 1.2, but we have not checked this.We can extract another interesting consequence about projective spaces fromour main computation of the closed-open map. Proposition 1.3.
The Fukaya A ∞ algebra of the Lagrangian R P n +1 ⊂ C P n +1 is not formal over a characteristic 2 field, for any n ≥ . Here, formality means an existence of a quasi-isomorphism with the associativealgebra HF ∗ ( R P n +1 , R P n +1 ) ∼ = K [ u ] / ( u n +2 − A ∞ algebrawith trivial higher-order structure maps. In particular, the Fukaya A ∞ algebraof the equator S ⊂ S is not formal in characteristic 2; we devote a separatediscussion to this fact in Section 3 where explicitly exhibit a non-trivial Masseyproduct which provides an alternative proof of the non-formality. Below is anotherexample of split-generation which we can prove using the same methods. Proposition 1.4.
Let K be a field of characteristic , X = Bl C P C P the blow-upof C P along a complex line which intersects R P in a circle, and let L ⊂ X be theblow-up of R P along that circle. Then CO ∗ : QH ∗ ( X ) → HH ∗ ( L, L ) is injectivealthough CO : QH ∗ ( X ) → HF ∗ ( L, L ) is not. Consequently, L split-generates F uk ( X ) . (The manifold Bl C P C P is the first instance among Bl C P k C P n for which L ismonotone of Maslov index at least 2, and such that CO is not injective — thelast requirement makes the use of our general results essential in this example.)In general, it is known that the real Lagrangian in a toric Fano variety is notdisplaceable from the monotone toric fibre: this was proved by Alston and Amorim[6]. Proposition 1.4 implies a much stronger non-displaceability result, like the onewhich has been known for R P n ⊂ C P n . Corollary 1.5.
Let K and L ⊂ X be as in Proposition 1.4, and L (cid:48) ⊂ X any othermonotone Lagrangian, perhaps equipped with a local system π ( L ) → K × , withminimal Maslov number at least 2 and such that HF ∗ ( L (cid:48) , L (cid:48) ) (cid:54) = 0 . If w ( L (cid:48) ) (cid:54) = 0 ,we also assume the technical Hypothesis 3.9, which is expected to hold following [23] . Then L ∩ L (cid:48) (cid:54) = ∅ . Here HF ∗ ( L (cid:48) , L (cid:48) ) denotes the Floer cohomology of L (cid:48) with respect to the localsystem ρ , so a better notation would be HF ∗ (( L (cid:48) , ρ ) , ( L (cid:48) , ρ )). For brevity, we de-cided to omit ρ from our notation of Floer and Hochschild cohomologies throughoutthe article, when it is clear that a Lagrangian is equipped with such a local system.The point of allowing local systems in Corollary 1.5 is to introduce more freedomin achieving the non-vanishing of HF ∗ ( L (cid:48) , L (cid:48) ).Note that Corollary 1.5 does not require that the obstruction number of L (cid:48) matches the one of L , namely zero. If w ( L (cid:48) ) (cid:54) = 0, we can pass to X × X noticingthat w ( L (cid:48) × L (cid:48) ) = 2 w ( L (cid:48) ) = 0 and similarly w ( L × L ) = 0, so we have well-definedFloer theory between the two product Lagrangians. This trick was observed byAbreu and Macarini [4] and has also been used in [6]. So it suffices to showthat L × L split-generates F uk ( X × X ) ; this follows from Proposition 1.4 by thegeneral expectation that the condition of the Abouzaid’s split-generation criterionis “preserved” under K¨unneth isomorphisms. As we explain later, this generalexpectation is contingent upon a certain commutative diagram which we formulate DMITRY TONKONOG as Hypothesis 3.9, and which is largely substantiated by Ganatra [23]; see also[3, 7].As in the case with R P n , we also prove a non-formality statement. Proposition 1.6.
The Fukaya A ∞ algebra of the Lagrangian Bl R P R P ⊂ BL C P C P from Proposition 1.4 is not formal over a characteristic 2 field. Although we cannot prove that CO ∗ is injective for the real locus of an arbitrarytoric Fano variety, we are able to do this in a slightly wider range of exampleswhich we postpone to Section 3. We will prove Proposition 1.1 and Corollary 1.2at the end of the introduction, and the remaining statements from above will beproved in Section 3. Now we state the main theorem; the new pieces of notationare explained straight after the statement. Theorem 1.7.
Let X be a compact monotone symplectic manifold, L ⊂ X amonotone Lagrangian submanifold of Maslov index at least , possibly equippedwith a local system ρ : H ( L ) → K × . If char K (cid:54) = 2 , assume L is oriented and spin.Let γ = { γ t } t ∈ S be a loop of Hamiltonian symplectomorphisms of X , and denoteby S ( γ ) ∈ QH ∗ ( X ) the corresponding Seidel element. Suppose the loop γ preserves L setwise, that is, γ t ( L ) = L . Denote by l ∈ H ( L ) the homology class of an orbit { γ t ( q ) } t ∈ S , q ∈ L . Finally, assume HF ∗ ( L, L ) (cid:54) = 0 .(a) Then CO ( S ( γ )) = ( − (cid:15) ( l ) · ρ ( l ) · L where L ∈ HF ∗ ( L, L ) is the unit.(b) Suppose there exists no a ∈ HF ∗ ( L, L ) such that ( ∗ ) µ ( a, Φ( y )) + µ (Φ( y ) , a ) = ( − (cid:15) ( l ) ρ ( l ) · (cid:104) y, l (cid:105) · L for each y ∈ H ( L ) . Then CO ∗ ( S ( γ )) ∈ HH ∗ ( L, L ) is linearly independent from the Hochschildcohomology unit.(c) More generally, suppose Q ∈ QH ∗ ( X ) and there exists no a ∈ HF ∗ ( L, L ) suchthat ( ∗∗ ) µ ( a, Φ( y )) + µ (Φ( y ) , a ) = ( − (cid:15) ( l ) ρ ( l ) · (cid:104) y, l (cid:105) · CO ( Q ) for each y ∈ H ( L ) . Then CO ∗ ( S ( γ ) ∗ Q ) and CO ∗ ( Q ) are linearly independent in the Hochschildcohomology HH ∗ ( L, L ) . Here µ is the product on HF ∗ ( L, L ), (cid:104)− , −(cid:105) is the pairing H ( L ) ⊗ H ( L ) → K ,and S ( γ ) ∗ Q is the quantum product of the two elements. Next,Φ : H ( L ) → HF ∗ ( L, L )is the PSS map of Albers [5], which is canonical and well-defined if HF ∗ ( L, L ) (cid:54) = 0.Its well-definedness in a setting closer to ours was studied by e.g. Biran and Cornea[10], and later we discuss it in more detail. Note that Φ is not necessarily injective,although in our applications, when HF ∗ ( L, L ) ∼ = H ∗ ( L ), it will be. Finally, in thetheorem we have allowed L to carry an arbitrary local system, which modifies theFukaya A ∞ structure of L by counting the same punctured holomorphic disks asin the case without a local system with coefficients which are the values of ρ on theboundary loops of such disks. The algebras HF ∗ ( L, L ), HH ∗ ( L, L ) get modifiedaccordingly, although their dependence on ρ is not reflected by our notation, asmentioned earlier. We allow non-trivial local systems in view of our application totoric fibres, and will only need the trivial local system ρ ≡ HE CLOSED-OPEN MAP FOR S -INVARIANT LAGRANGIANS 5 To complete the statement of Theorem 1.7, we need to explain the sign ( − (cid:15) ( l ) = ± T L over the loop γ t ( q ) ⊂ L : the one induced from a fixed basis of T q L bythe Hamiltonian loop γ , and the one determined by the spin structure on L . Weput (cid:15) ( l ) to be 0 if the two trivialisations agree, and 1 otherwise. Outline of proof.
It has been mentioned earlier that part (a) of Theorem 1.7is an easy consequence of the paper by Charette and Cornea [12]. The proof ofparts (b) and (c) also starts by using a result from that paper, and then the mainstep is an explicit computation of CO ( S ( γ )) | CF ( L,L ) : CF ( L, L ) → CF ( L, L )on cochain level, which turns out to be dual to taking the γ -orbit of a point up tothe factor ( − (cid:15) ( l ) ρ ( l ): this is Proposition 2.8. The final step is to check whetherthe computed nontrivial piece of the Hochschild cocycle CO ∗ ( S ( γ )) survives tocohomology; this is controlled by equations ( ∗ ), ( ∗∗ ). Remark 1.8.
In our examples we will never encounter a non-trivial sign ( − (cid:15) ( l ) :for real Lagrangians we shall be working over characteristic two, and for toric fibreswith the standard spin structure, this sign is easily seen to be +1. The exampleswhen the sign ( − (cid:15) ( l ) is negative have been found by J. Smith; they occur for P SU ( N − The split-generation criterion.
We will now briefly discuss the split-gener-ation criterion in more detail, particularly because we wish to pay attention to bothchar K = 2 and char K (cid:54) = 2 cases. We continue to denote by L ⊂ X a monotoneLagrangian submanifold with minimal Maslov number at least 2, which is orientedand spin if char K (cid:54) = 2. If char K = 2, we allow L to be non-orientable. Consider thequantum multiplication by the first Chern class as an endomorphism of quantumcohomology, − ∗ c ( X ) : QH ∗ ( X ) → QH ∗ ( X ). If K is algebraically closed, wehave an algebra decomposition QH ∗ ( X ) = ⊕ w QH ∗ ( X ) w where QH ∗ ( X ) w is thegeneralised w -eigenspace of − ∗ c ( X ), w ∈ K .Recall that w ( L ) ∈ K denotes the obstruction number of L , i.e. the count ofMaslov index 2 disks with boundary on L . By an observation of Auroux, Kontsevichand Seidel, CO (2 c ) = 2 w ( L ) · L , which in char K (cid:54) = 2 implies that CO ( c ) = w ( L ) · L , see e.g. [46]. Now suppose that char K = 2 and c ( X ) lies in the imageof H ( X, L ; K ) → H ( X ; K ), which is true if L is orientable (because the Maslovclass goes to twice the Chern class under H ( X, L ; Z ) → H ( X ; Z ), and the Maslovclass of an orientable manifold is integrally divisible by two). In this case, the sameargument shows again that CO ( c ) = w ( L ) · L . This way one deduces the followinglemma, which is well-known but usually stated only for char K (cid:54) = 2. Lemma 1.9.
For K of any characteristic, if L is orientable, then CO : QH ∗ ( X ) → HF ∗ ( L, L ) vanishes on all summands except maybe QH ∗ ( X ) w ( L ) . (cid:3) (If w ( L ) is not an eigenvalue of − ∗ c ( X ), then CO vanishes altogether, andit follows that HF ∗ ( L, L ) = 0. Recall that L is required to be monotone.) Thesame vanishing statement is expected to hold for the full map CO ∗ . Keeping thisvanishing in mind, we see that the “naive” version of the split-generation criterion DMITRY TONKONOG stated in the introduction, that CO ∗ : QH ∗ ( X ) → HH ∗ ( L, L ) is injective, canonly be useful when char K = 2 and L is non-orientable. In other cases it mustbe replaced by a more practical criterion which does not ignore the eigenvaluedecomposition; we will now state both versions of the criterion. Let F ( X ) w denotethe Fukaya category whose objects are monotone Lagrangians in X with minimalMaslov number at least 2, oriented and spin if char K (cid:54) = 2, and whose obstructionnumber equals w ∈ K . Theorem 1.10.
Let L , . . . , L n ⊂ X be Lagrangians which are objects of F uk ( X ) w ,and G ⊂ F uk ( X ) w be the full subcategory generated by L , . . . , L n . Then G split-generates F uk ( X ) w if either of the two following statements hold.(a) char K (cid:54) = 2 , and CO ∗ | QH ∗ ( X ) w : QH ∗ ( X ) w → HH ∗ ( G ) is injective.(b) K is arbitrary, and CO ∗ : QH ∗ ( X ) → HH ∗ ( G ) is injective. (cid:3) In the monotone case, this theorem is due to Ritter and Smith [39] and Sheridan[46]. It is more common to only state part (a), but it is easy to check the sameproof works for part (b) as well. (In part (a), we could also allow char K = 2, if L isorientable.) Theorem 1.10 is most easily applied when QH ∗ ( X ) w is 1-dimensional:because CO ∗ is unital, it automatically becomes injective. We are going to applythis theorem in more complicated cases. Before we proceed, let us mention oneeasy corollary of split-generation. We say that L , . . . , L n split-generate the Fukayacategory when G does. Lemma 1.11.
If Lagrangians L , . . . , L n ⊂ X split-generate F uk ( X ) w , and L ⊂ X is another Lagrangian which is an object of F uk ( X ) w with HF ∗ ( L, L ) (cid:54) = 0 , then L has non-empty intersection, and non-zero Floer cohomology, with one of theLagrangians L i . (cid:3) CO for real toric Lagrangians. In this subsection we state a theorem thatcomputes CO for real Lagrangians in toric manifolds. Using it, it is easy to identifythe cases when CO is injective (and the split-generation follows immediately), andthe cases when CO is not injective and therefore a further study of CO ∗ is requiredto establish the split-generation. Our subsequent goal is to apply the main result,Theorem 1.7, to some examples of the later type.Let X be a (smooth, compact) toric Fano variety with minimal Chern numberat least 2, i.e. (cid:104) c ( X ) , H ( X ; Z ) (cid:105) = N Z , N ≥
2. As a toric manifold X has acanonical anti-holomorphic involution τ : X → X . Its fixed locus is the so-calledreal Lagrangian L ⊂ X which is smooth [17, p. 419], monotone and whose minimalMaslov number equals the minimal Chern number of X [24]. When speaking ofsuch real Lagrangians, we will always be working over a field K of characteristic 2.In particular, there is the Frobenius map: F : QH ∗ ( X ) → QH ∗ ( X ) , F ( x ) = x . Because char K = 2, F is a map of unital algebras. We have reflected in our notationthat F multiplies the Z / N -grading by two. A classical theorem of Duistermaat[17] constructs, again in char K = 2, the isomorphisms H i ( L ) ∼ = H i ( X ). We canpackage these isomorphisms into a single isomorphism of unital algebras, D : H ∗ ( X ) ∼ = −→ H ∗ ( L ) . Let us now recall a recent theorem of Haug [24].
HE CLOSED-OPEN MAP FOR S -INVARIANT LAGRANGIANS 7 Theorem 1.12. If char K = 2 , then HF ∗ ( L, L ) ∼ = H ∗ ( L ) as vector spaces. Usingthe identification coming from a specific perfect Morse function from [24] , and alsoindentifying QH ∗ ( X ) ∼ = H ∗ ( X ) , the same map D : QH ∗ ( X ) ∼ = −→ HF ∗ ( L, L ) is again an isomorphism of unital algebras. (cid:3) It turns out that it is possible to completely compute CO for real toric La-grangians. This rather quickly follows by combining the works of Charette andCornea [12], Hyvrier [26], and McDuff and Tolman [33]; we explain this theoremin Section 3. Theorem 1.13.
The diagram below commutes. QH ∗ ( X ) F (cid:47) (cid:47) CO (cid:38) (cid:38) QH ∗ ( X ) D ∼ = (cid:15) (cid:15) HF ∗ ( L, L )In particular, CO is injective if and only if F is injective.1.4. Split-generation for the real projective space.
We conclude the intro-duction by proving Proposition 1.1 and Corollary 1.2. The crucial idea is thatwhen n is odd, the kernel of CO : QH ∗ ( C P n ) → HF ∗ ( R P n , R P n ) is the idealgenerated by the Seidel element of a non-trivial Hamiltonian loop preserving R P n ;this allows to apply Theorem 1.7 and get new information about CO ∗ . Recall that QH ∗ ( X ) ∼ = K [ x ] / ( x n +1 −
1) and w ( R P n ) = 0, because the minimal Maslov numberof R P n equals n + 1 (when n = 1, we still have w ( S ) = 0 for S ⊂ S ). Proof of Proposition 1.1. If n is even, the Frobenius map on QH ∗ ( C P n ) is injective,so by Theorem 1.13, CO : QH ∗ ( C P n ) → HF ∗ ( R P n , R P n ) is injective, and hence CO ∗ too.Now suppose n is odd and denote n = 2 p −
1. Given char K = 2, we have QH ∗ ( C P n ) ∼ = K [ x ] / ( x p + 1) , so ker F = ker CO is the ideal generated by x p + 1.Consider the Hamiltonian loop γ on C P n which in homogeneous co-ordinates ( z : . . . : z p ) is the rotation (cid:0) cos t sin t − sin t cos t (cid:1) , t ∈ [0 , π ], applied simultaneously to the pairs( z , z ) , . . . , ( z p − , z p ). Note that t runs to π , not 2 π . This loop is Hamiltonianisotopic to the loop( z : . . . : z p − ) (cid:55)→ ( e it z : z : . . . : e it z p − : z p ) , t ∈ [0 , π ] , so S ( γ ) = x p , see [33]. The loop γ obviously preserves the real Lagrangian R P n ⊂ C P n , and its orbit l is a generator of H ( R P n ) ∼ = K . Taking y ∈ H ( R P n ) tobe the generator, we get (cid:104) y, l (cid:105) = 1, and the right hand side of equation ( ∗∗ ) fromTheorem 1.7 equals CO ( Q ). On the other hand, the product on HF ∗ ( R P n , R P n )is commutative by Theorem 1.12, so the left hand side of ( ∗∗ ) necessarily vanishes.We conclude that the hypothesis of Theorem 1.7(c) is satisfied for any Q / ∈ ker CO .Let us prove that CO ∗ ( P ) (cid:54) = 0 for each nonzero P ∈ QH ∗ ( C P n ). If CO ( P ) (cid:54) = 0,we are done, so it suffices to suppose that CO ( P ) = 0. It means that P = ( x p + 1) ∗ Q = ( S ( γ ) + 1) ∗ Q DMITRY TONKONOG for some Q ∈ QH ∗ ( C P n ). Note that if Q ∈ ker CO = ker F then P ∈ (ker F ) = { } . So if P (cid:54) = 0, then CO ( Q ) (cid:54) = 0, and thus CO ∗ ( P ) (cid:54) = 0 by Theorem 1.7(c) andthe observation earlier in this proof. (cid:3) Remark 1.14.
When n is even, c ( C P n ) is invertible in QH ∗ ( C P n ), so the 0-eigenspace QH ∗ ( C P n ) is trivial; but L is non-orientable, so this does not contra-dict Lemma 1.9. On the other hand, when n is odd, L is orientable but c ( C P n )vanishes in char K = 2, so the whole QH ∗ ( C P n ) is its 0-eigenspace; this is alsoconsistent with Lemma 1.9. Proof of Corollary 1.2.
This follows from Proposition 1.4 and Theorem 1.10(b). (cid:3)
The same trick of finding a real Hamiltonian loop whose Seidel element generatesker CO works for some other toric manifolds which have “extra symmetry” inaddition to the toric action, like a Hamiltonian action of SU (2) dim C X/ whichwas essentially used above. As already mentioned, we will provide more explicitexamples in Section 3. Acknowledgements.
The author is most grateful to his supervisor Ivan Smithfor many useful comments as well as constant care and enthusiasm. Yankı Lekiliand Jack Smith have provided valuable feedback and pointed out two inaccuraciesin the previous versions of this paper. The paper has also benefitted from dis-cussions with Mohammed Abouzaid, Lino Amorim, Fran¸cois Charette, GeorgiosDimitroglou Rizell, Jonny Evans, Alexander Ritter, Paul Seidel, Nick Sheridanand Renato Vianna. The referee’s suggestions on improving the exposition havebeen very useful.The author was funded by the Cambridge Commonwealth, European and Inter-national Trust, and acknowledges travel funds from King’s College, Cambridge.2.
Proof of Theorem 1.7
Let X be a monotone symplectic manifold and w ∈ K . We recall that the objectsin the monotone Fukaya category F uk ( X ) w are monotone Lagrangian submanifolds L ⊂ X with minimal Maslov number at least 2, oriented and spin if char K (cid:54) =2, equipped with local systems ρ : π ( L ) → K × , whose count of Maslov 2 disks(weighted using ρ ) equals w . We will use the definition of the Fukaya category basedon achieving transversality by explicit Hamiltonian perturbations of the pseudo-holomorphic equation. This setup was developed by Seidel [41] for exact manifoldsand carries over to monotone ones, see [39, 46, 2]. There is a notion of boundingcochains from [22], generalising the notion of a local system, and all results areexpected carry over to them as well.2.1. A theorem of Charette and Cornea.
Suppose γ = { γ t } t ∈ S is a loopof Hamiltonian symplectomorphisms on X . As explained by Seidel in [41, Sec-tion (10c)], the loop γ gives rise to a natural transformation γ (cid:93) from the identityfunctor on F uk ( X ) w to itself. Any such natural transformation is a cocycle ofthe Hochschild cochain complex CC ∗ ( F uk ( X ) w ) [41, Section (1d)]. Denote thecorresponding Hochschild cohomology class by[ γ (cid:93) ] ∈ HH ∗ ( F uk ( X ) w ) . HE CLOSED-OPEN MAP FOR S -INVARIANT LAGRANGIANS 9 We denote, as earlier, the closed-open map by CO ∗ : QH ∗ ( X ) → HH ∗ ( F uk ( X ) w )and the Seidel element by S ( γ ) ∈ QH ∗ ( X ). The following theorem was proved byCharette and Cornea [12]. Theorem 2.1.
If we take for F uk ( X ) w the Fukaya category of Lagrangians withtrivial local systems only, then CO ∗ ( S ( γ )) = [ γ (cid:93) ] . (cid:3) Let us now restrict to a single Lagrangian L which is preserved by the Hamilton-ian loop γ , and denote by l ∈ H ( L ) the homology class of an orbit of γ on L . Let CC ∗ ( L, L ) denote the Hochschild cochain complex of the A ∞ algebra CF ∗ ( L, L ),and let HH ∗ ( L, L ) be its Hochschild cohomology. (The definition of Hochschildcohomology will be reminded later in this section.) We will now need to recall theproof of Theorem 2.1 for several reasons: first, we wish to see how Theorem 2.1gets modified in the presence of a local system on L ; second, we shall see the ap-pearance of a sign “hidden” in γ (cid:93) ; third and most importantly, we will recall thedefinition of the moDuli spaces computing γ (cid:93) in the process. Eventually, for lateruse we need a form of Theorem 2.1 expressed by formula (2.2) below, which takesthe local system and the sign into account.Pick some Floer datum { H s , J s } s ∈ [0 , and perturbation data defining an A ∞ structure on Floer’s complex CF ∗ ( L, L ) [41]. Recall that the maps CO k ( S ( γ )) : CF ∗ ( L, L ) ⊗ k → CF ∗ ( L, L )count 0-dimensional moduli space of disks satisfying a perturbed pseudo-holomorphicequation (with appropriately chosen perturbation data) with k + 1 boundary punc-tures ( k inputs and one output) and one interior marked point. These disks satisfythe Lagrangian boundary condition L , and their interior marked point is con-strained to a cycle dual to S ( γ ), see Figure 1(a) (in this figure, we abbreviate thedatum { H s , J s } simply to H ). A disk u is counted with coefficient ± ρ ( ∂u ) wherethe sign ± comes from the orientation on the moduli space and ρ ( ∂u ) ∈ K × is themonodromy of the local system. The collection of maps CO ∗ ( S ( γ )) := {CO k } k ≥ isa Hochschild cochain in CC ∗ ( L, L ), if all perturbation data are chosen consistentlywith gluing.The argument of Charette and Cornea starts by passing to a more convenientdefinition of the closed-open map in which CO k count holomorphic disks with k + 1boundary punctures and one interior puncture (instead of a marked point). Wecan view the neighbourhood of the interior puncture as a semi-infinite cylinder,then the pseudo-holomorphic equation restricts on this semi-infinite cylinder to aHamiltonian Floer equation with some Floer datum { F t , J t } t ∈ S . We input thePSS image of S ( γ ) to the interior puncture, see Figure 1(b), given as a linearcombination of some Hamiltonian orbits z (in the figure, we abbreviate the datum { F t , J t } simply to F t ).The PSS image of S ( γ ) counts configurations shown in the upper part of Fig-ure 1(b), consisting of disks with one output puncture (say, asymptotic to an orbit y ), and a cylinder counting continuation maps from ( γ − t ) ∗ y seen as an orbit ofFloer’s complex with datum pulled back by the loop γ − t [40, Lemmas 2.3 and 4.1],to another orbit z of the original Floer’s complex with datum { F t , J t } . Let usglue the z -orbits together, passing to Figure 1(c), and then substitute each lowerpunctured pseudo-holomorphic disk u in Figure 1(c) by ˜ u defined as follows:(2.1) ˜ u ( re πit ) = γ − t ◦ u ( re πit ) , Figure 1.
A computation of CO ∗ ( S ( γ )) by Charette and Cornea.assuming that the interior puncture is located at 0 ∈ C and the output punctureat 1 ∈ C . Let us look at the effect of this substitution.First, [ ∂u ] = [ ∂ ˜ u ] + l ∈ H ( L ) so the count of configurations in Figure 1(c)(before substitution) is equal to the count of configurations in Figure 1(d) (aftersubstitution) multiplied by ρ ( l ).Second, ˜ u satisfies the same boundary condition L because γ t L = L , but the per-turbation data defining the pseudo-holomorphic equation get pulled back accord-ingly. In particular, the Lagrangian Floer datum { H s , J s } s ∈ [0 , and the asymptoticchord at a strip-like end corresponding to the boundary puncture at t i ∈ S getpulled back by γ t i .Third, (2.1) gives an abstract bijection u (cid:55)→ ˜ u between the respective zero-dimensional moduli spaces, but we should discuss how this bijection behaves withrespect to the signs attached to u, ˜ u by the orientations on the moduli spaces.Assume for simplicity that the Hamiltonian perturbation is small enough, so thatwe can canonically deform ∂u and ∂ ˜ u to loops inside L . Take the trivialisation of T L | ∂u defined by the spin structure on L , and push it forward by γ to a trivialisationof T L | ∂ ˜ u . Obviously, (2.1) preserves the signs computed using these trivialisations;we remark that such trivialisation of T L | ∂ ˜ u is used in the general definition of γ (cid:93) . In our specific case, ˜ u is again a curve with boundary on L (rather than witha moving Lagrangian boundary condition), and we wish to consider a differentorientation scheme for ˜ u , namely the usual scheme for orienting moduli spaces ofcurves with boundary on L using the given spin structure. We shall be using thisorientation scheme from now on, and we observe that it uses the trivialisation of HE CLOSED-OPEN MAP FOR S -INVARIANT LAGRANGIANS 11 T L | ∂ ˜ u coming from the spin structure, which may be different from the pushforwardtrivialisation of T L | ∂ ˜ u mentioned before. We denote the sign difference betweenthe two orientations for ˜ u by ( − (cid:15) = ±
1; this sign equals +1 is and only if the twotrivialisations of
T L | ∂ ˜ u from above are homotopic. (Ultimately, we are going touse the fact that for our choice of orientation scheme, the moduli space of constantunconstrained disks is positively oriented; this may not be true for the γ push-forward orientation scheme.) Remark 2.2.
In general, the number (cid:15) does not necessarily equal (cid:15) ( l ) from theintroduction. This equality holds when ∂ ˜ u is contractible, as easily seen from thedefinitions.Fourth, near the interior puncture ˜ u satisfies the Hamiltonian Floer equationwith original datum { F t , J t } at the interior puncture, and is asymptotic orbit y . Sowe can glue the y -orbits, passing to Figure 1(e), and Figure 1(f) is another drawingof the same domain we got after gluing: namely, the disk with k + 1 boundarypunctures and one interior unconstrained marked point, fixed at 0 ∈ C . Let usexplain the presence of the marked point 0 ∈ C : it is carried over from a markedpoint on the upper disk in Figure 1(b), where the interior marked point serves tostabilise the domain; such a marked point is present in the definition of the Seidelelement.Summing up,(2.2) CO k ( S ( γ ))( x ⊗ . . . ⊗ x k ) = ( − (cid:15) · ρ ( l ) · (cid:88) (cid:93) M γ ( x , . . . , x k ; x ) · x where M γ ( x , . . . , x k ; x ) is the 0-dimensional moduli space of disks shown in Fig-ure 1(f) which satisfy the inhomogeneous pseudo-holomorphic equation defined bydomain- and modulus-dependent perturbation data in the sense of [41] such that: • the disks carry the unconstrained interior marked point fixed at t = 0, theoutput boundary puncture fixed at t = 1, and k free input boundary puncturesat t i ∈ S , i = 1 , . . . , k ; • on a strip-like end corresponding to a boundary puncture t i ∈ S , perturbationdata restrict to the Floer datum which is the γ t i -pullback of the original Floerdatum { H s , J s } s ∈ [0 , , and the asymptotic chord for this strip must be the γ t i -pullback of the asymptotic chord x i of the original Floer datum; • the data must be consistent with gluing strip-like ends at t i ∈ S to strip-likeends of punctured pseudo-holomorphic disks carrying the γ t i -pullbacks of theperturbation data defining the A ∞ structure on CF ∗ ( L, L ); back by γ t i . • we use the standard orientation scheme for curves with boundary on L to orientthe M γ s, and the sign ( − (cid:15) was explained above.The counts (cid:93) M γ are signed and weighted by ρ ; the third condition guaranteesthat CO ∗ ( S ( γ )) is a Hochschild cocycle. Formula (2.2) coincides with the formulafrom [41, Section (10c)] defining the natural transformation [ γ (cid:93) ] up to ( − (cid:15) ρ ( l ),and we have clarified this difference. Remark 2.3.
The fixed interior marked point at t = 0 and the fixed boundarymarked point at t = 1 make sure our disks have no automorphisms, so the values t i ∈ S of the other boundary punctures are uniquely defined.Before proceeding, note that we are already able to compute CO ( S ( γ )). Corollary 2.4. If { γ t } t ∈ S is a Hamiltonian loop such that γ t ( L ) = L , then on thechain level, CO ( S ( γ )) = ( − (cid:15) ( l ) · ρ ( l ) · L ∈ HF ( L, L ) . Here L is a chain-level representative of the cohomology unit, see (2.3) below.Proof. When k = 0, the moduli space in formula (2.2) is exactly the moduli spacedefining the cohomological unit in CF ∗ ( L, L ), see e.g. [46, Section 2.4]. The equality (cid:15) = (cid:15) ( l ) holds by Remark 2.2: for a small Hamiltonian, the curves computing theunit are close to being constant and therefore have contractible boundary. (cid:3) Proof of Theorem 1.7(a).
This is the homology-level version of Corollary 2.4. (cid:3)
The PSS maps in degree one.
Our goal will be to compute a “topologicalpiece” of CO ( S ( γ )). This subsection introduces some background required for thecomputation: in particular, we recall that there is a canonical map Φ : H ( L ) → HF ∗ ( L, L ) which was used in the statement of Theorem 1.7. This is the LagrangianPSS map of Albers [5], and the fact it is canonical was discussed, for instance, byBiran and Cornea [10, Proposition 4.5.1( ii )] in the context of Lagrangian quantumcohomology.First, recall that once the Floer datum is fixed, the complex CF ∗ ( L, L ) acquiresthe Morse Z -grading. This grading is not preserved by the Floer differential orthe A ∞ structure maps, but is still very useful. Assume that the Hamiltonianperturbation, as part of the Floer datum, is chosen to have a unique minimum x on L , which means that CF ( L, L ) is one-dimensional and generated by x . Wedenote by(2.3) 1 L ∈ CF ( L, L )the chain-level cohomological unit defined in [46, Section 2.4], which is proportionalto x . Now pick a metric and a Morse-Smale function f on L with a single minimum;together they define the Morse complex which we denote by C ∗ ( L ). Consider the“Maslov index 0” versions of the PSS maps, which are linear (but not chain) maps:(2.4) Ψ : CF ∗ ( L, L ) → C ∗ ( L ) , Φ : C ∗ ( L ) → CF ∗ ( L, L ) . These maps are defined as in the paper of Albers [5], with the difference thatΦ , Ψ count configurations with Maslov index 0 disks only. For example, the mapΨ counts configurations consisting of a Maslov index 0 pseudo-holomorphic diskwith boundary on L and one input boundary puncture, followed by a semi-infinitegradient trajectory of f which outputs an element of C ∗ ( L ). Similarly, Φ countsconfigurations in which a semi-infinite gradient trajectory is followed by a Maslovindex 0 disk with an output boundary puncture. The maps Ψ , Φ preserve Z -gradings on the two complexes.Let d : CF ∗ ( L, L ) → CF ∗ +1 ( L, L ) be the “Morse” part of the Floer differentialcounting the contribution of Maslov 0 disks, see Oh [35]. Denote by d Morse : C ∗ ( L ) → C ∗ +1 ( L ) the usual Morse differential. The lemma below is a version of [5, Theo-rem 4.11]. Lemma 2.5. Φ , Ψ are chain maps with respect to d and d Morse , and are cohomol-ogy inverses of each other. (cid:3)
Lemma 2.6.
Suppose HF ∗ ( L, L ) (cid:54) = 0 . If y ∈ C ( L ) is a Morse cocycle (resp. cobound-ary) then Φ( y ) is a Floer cocycle (resp. coboundary). HE CLOSED-OPEN MAP FOR S -INVARIANT LAGRANGIANS 13 Proof.
This follows from the fact that the image and the kernel of d and the fullFloer differential d coincide on CF ( L, L ) if HF ∗ ( L, L ) (cid:54) = 0, by Oh’s decompositionof the Floer differential [35]. (cid:3) Consequently, if HF ∗ ( L, L ) (cid:54) = 0, we get a mapΦ : H ( L ) → HF ∗ ( L, L ) . For Ψ, we have a weaker lemma using [35] (this lemma is not true for coboundariesinstead of cocycles).
Lemma 2.7. If y ∈ CF ∗ ( L, L ) is a Floer cocycle, then Ψ( y ) ∈ C ∗ ( L ) is a Morsecocycle. (cid:3) By Lemmas 2.6, 2.7, given HF ∗ ( L, L ) (cid:54) = 0, we have the following induced mapswhich we denote by the same symbols Ψ , Φ, abusing notation:(2.5) Ψ : CF ∗ ( L, L ) → H ∗ ( L ) , Φ : H ∗ ( L ) → HF ∗ ( L, L ) . In particular, Theorem 1.7 in Section 1 refers to this cohomological version of themap Φ. We remind that Ψ does not necessarily descend to a map from HF ∗ ( L, L ).2.3.
Computing the topological part of CO ( S ( γ )) . We continue to use theabove conventions and definitions, namely we use the Z -grading on CF ∗ ( L, L ),the maps Φ , Ψ, and the choice of a Hamiltonian perturbation on L with a uniqueminimum x . From now on, we assume HF ∗ ( L, L ) (cid:54) = 0. Recall that CO ∗ ( S ( γ )) isdetermined via formula (2.2) by the moduli spaces M γ ( x , . . . , x k ; x ). The con-nected components of M γ ( x , . . . , x k ; x ) corresponding to disks of Maslov index µ have dimension | x | + k + µ − k (cid:88) i =1 | x i | where | x i | are the Z -gradings of the x i ∈ CF ∗ ( L, L ). Consequently, CO k ( S ( γ )) : CF ∗ ( L, L ) ⊗ k → CF ∗ ( L, L )is a sum of maps of degrees − k − mN L , m ≥ , where N L is the minimal Maslov number of L . In particular, the restriction of CO ( S ( γ )) to CF ( L, L ) is of pure degree −
1, that is, its image lands in CF ( L, L ): CO ( S ( γ )) | CF ( L,L ) : CF ( L, L ) → CF ( L, L ) . Moreover, this map is determined by the moduli space consisting of Maslov index 0disks only, and can be computed in purely topological terms. This is the maintechnical computation which we now perform; recall that l ∈ H ( L ) is the homologyclass of an orbit of γ in L , and the sign ( − (cid:15) ( l ) was defined in Section 1. Proposition 2.8.
Suppose HF ∗ ( L, L ) (cid:54) = 0 . If x ∈ CF ( L, L ) is a Floer cocycle,then on the chain level, CO ( S ( γ ))( x ) = ( − (cid:15) ( l ) · ρ ( l ) · (cid:104) Ψ( x ) , l (cid:105) · L . Here Ψ( x ) ∈ H ( L ) is from (2.5), (cid:104)− , −(cid:105) denotes the pairing H ( L ) ⊗ H ( L ) → K ,we consider CO on the chain level, and L is a chain-level cohomology unit asin (2.3) . Proof.
All disks with boundary on L we consider in this proof are assumed to haveMaslov index 0. We identify the domains of all disks that appear in the proof withthe unit disk in C , and their boundaries are identified with the unit circle S ⊂ C . In the subsequent figures, punctured marked points will be drawn by circlesfilled white, and unpunctured marked points by circles filled black. According toformula (2.2), for a generator x ∈ CF ( L, L ) we have CO ( S ( γ ))( x ) = ( − (cid:15) · ρ ( l ) · (cid:93) M γ ( x ; x ) , where M γ ( x ; x ) consists of (perturbed pseudo-holomorphic) Maslov index 0 diskswhose domains are shown in Figure 2(a). Step 1. Perturbation data producing bubbles with unpunctured points.
Recall thatthe domains appearing in the moduli space M γ ( x ; x ) are disks with the interiormarked point 0 and boundary punctures 1 , t , where t ∈ S \ { } . For further use,we will choose perturbation data defining M γ ( x ; x ) whose bubbling behaviouras t → t →
1, with the gluing shown in Figure 2(a) → (b),where the bubble meets the principal disk along a puncture, meaning that near thispuncture it satisfies a Floer equation and shares an asymptotic Hamiltonian chordwith the corresponding puncture of the principal disk. On the other hand, we willuse perturbation data consistent with gluing shown in Figure 2(a) → (c), where thebubble is attached to the principal disk by an unpunctured marked point. Nearthe unpunctured marked point, the disks satisfy a holomorphic equation with noHamiltonian term. Figure 2.
Two types of gluings for M γ ( x ; x ), and a way to inter-polate between the glued perturbation data.Let us explain how to define both types of data more explicitly. The domainin Figure 2(a), with free parameter t close to 1, is bi-holomorphic to the domainshown in Figure 3 whose boundary marked points are fixed at 1 and some t ∈ S ,upon which a stretching procedure along the strip labelled (b) is performed. Thisstretching procedure changes the complex structure on the disk by identifying thestrip with [0 , × [0 , , × [0 , r ].The parameter r ∈ [1 , + ∞ ) is free and replaces the free parameter t , so that tending r → + ∞ replaces the collision of two marked points t → → (b), one requires the perturbed pseudo-holomorphic equa-tion to coincide, on the strip [0 , × [0 , r ], with the usual Floer equation definingthe Floer differential, which uses a Hamiltonian perturbation translation-invariant HE CLOSED-OPEN MAP FOR S -INVARIANT LAGRANGIANS 15 Figure 3.
Collision of two boundary marked points seen as stretch-ing the strip (b) with parameter r → + ∞ .in the direction of [0 , r ]. In order to get perturbation data producing the bub-bling pattern Figure 2(a) → (c), we simply put an unperturbed pseudo-holomorphicequation on the strip [0 , × [0 , r ], without using a Hamiltonian perturbation atall.Both ways of defining perturbation data are subject to appropriate gluing andcompactness theorems, which precisely say that as we tend r → ∞ , the solutionsbubble in one of the two corresponding ways shown in Figure 2. The standardchoice is used, for example, to prove that [ γ (cid:93) ] (obtained from the counts of various M γ ) is a Hochschild cocycle in the Fukaya A ∞ algebra of L defined using Seidel’ssetup with Hamiltonian perturbations. The other choice will be more convenientfor our computations. Note that the two different types of perturbation data givethe same count (cid:93) M γ ( x ; x ): this is proved by interpolating between them using thetwo-parametric space of perturbation data obtained from gluing together the disksin Figure 2(d) with different length parameters. Recall that all disks in M γ ( x ; x )have Maslov index 0 as the Morse index | x | = 1, therefore no unnecessary bubblingoccurs. (Since we do not want to compute the moduli spaces M γ other than the M γ ( x ; x ) for | x | = 1, we do not have to worry about extending our unusual typeof perturbation data to the other moduli spaces.)In addition, we will assume that the Hamiltonian perturbation vanishes over theprincipal disk in Figure 2(c), making this disk J -holomorphic and hence constant,because the disk has Maslov index 0. Such configurations can be made consistentwith gluing: for this, one just needs to make the Hamiltonian perturbation vanishover subdomain (a) in Figure 3, for all t close to 1. Note that regularity can beachieved by perturbing the pseudo-holomorphic equation over the subdomain tothe right of the strip (b) in Figure 3. Step 2. A one-dimensional cobordism from M γ ( x ; x ) . In what follows, we will use(and assume familiarity with) the theory of holomorphic pearly trees developed bySheridan in his Morse-Bott definition of the Fukaya category [46]; see also theearlier work of Cornea and Lalonde [16]. Sheridan performs the analysis based onextending Seidel’s setup of Fukaya categories from [41]. Although [46] considersexact Lagrangians instead of monotone ones, all the analysis works equally wellin the non-exact case if we only consider disks of Maslov index 0, because hereunpunctured disk bubbles cannot occur just like in the exact case. Techniques fordealing with holomorphic pearly trees (or “clusters”) with disks of arbitrary Maslovindex have appeared in [16, 11], but we will not actually need to appeal to them.We will now define a family of domains depending on two parameters s ∈ [0 , π ], l ∈ [ − , + ∞ ]. When s / ∈ { , π } , the domains are shown in Figure 4(a)–(e), wherewe denoted t = e is ; we discuss the case s ∈ { , π } later. When l = − Figure 4.
The domains for s ∈ (0 , π ), l ∈ [ − , ∞ ], where t = e is .is the disk from the definition of M γ ( x ; x ). When l ∈ ( − , l : the first point lies on the line segment[0 , t ], the second one lies on the line segment [0 , l from 0. When l = 0, the domain consists of the principal disk with markedpoints 0 , , t , and two bubble disks attached to the principal disk at points 1 and t . The first bubble disk has marked points 0 , − , − < l < ∞ , the domain contains the same three disks, now disjoint from each other,plus two line segments of length l connecting the bubble disks to the principal onealong the boundary marked points at which the disks used to be attached to eachother. (So far, the length is just a formal parameter associated with the domain,but soon it will become the length of the flowline corresponding to the segment.)When l = ∞ , we replace each line segment by two rays [0 , + ∞ ) (cid:116) ( −∞ , s = 0 or s = 2 π , the domains obtain extra bubbles as those discussed abovein the definition of M γ ( x ; x ), which correspond to the parameter t = e is ∈ S approaching 1 ∈ S from the two sides. These domains are shown in Figure 5:as l goes from − l ; this does not cause any difficulty with the definitionsbecause these marked points are only used to represent varying perturbation dataconsistent with the types of bubbling we prescribe in the figures. We will soonmention what these varying data are in terms of stretching certain strips insidea fixed disk. When l >
0, the length of the paths equals l . When l = ∞ , oneintroduces broken lines [0 , + ∞ ) (cid:116) ( −∞ ,
0] as above.Having specified the domains, we briefly explain how to equip the disks withsuitable perturbed pseudo-holomorphic equations, and line segments with suitablegradient equations to get a moduli space of solutions. When l = −
1, we choosethe equations defining M γ ( x ; x ) as discussed above in Step 1, in particular whichis consistent with bubbling at the unpunctured point as s → s → π . When − < l <
0, we choose the equations with the same properties as for M γ ( x ; x ), HE CLOSED-OPEN MAP FOR S -INVARIANT LAGRANGIANS 17 Figure 5.
The domains for s ∈ { , π } , l ∈ [ − , ∞ ].which are additionally constistent with bubbling at unpunctured points as l → l ≥
0, we choose the equation on the disk with an input puncture to bethe γ t -pullback of the one appearing in the definition of the PSS map Ψ, and theequation on the disk with an output puncture to be exactly the equation fromthe PSS map Φ. Finally we fix two generic Morse-Smale functions f, g : L → R .On the line segments and rays, the equation is the gradient equation for g or the γ t -pullback of the gradient equation for f , as shown in Figure 4.In general, we can arrange the equations on all disks to have a Hamiltonianperturbation. For the disks with punctures, we must do so anyway; however, for thedisks without punctures (with unpunctured marked points only), we can choose theequation to be J -holomorphic without a Hamiltonian term, provided that we checkthat our moduli space can be made regular with this restricted choice. In order tocarry out the computations below, we choose the zero Hamiltonian perturbationon all disks without boundary punctures, namely: • the central disk in Figures 4(c), (d), (e), • the left disks in Figures 5(a)–(d) and the central disks in Figures 5(c), (d).We will now specify one more property of the equations that we choose. When s = 0 and l <
0, we require the perturbation data on the twice-punctured disk inFigure 5(a) to be obtained by π -rotation from the perturbation data on the similardisk for s = 2 π (and the same parameter l ), if we identify the two punctures withpoints 1 , − s = 0 and l ≥
0, we make a similar symmetricchoice.Finally, we specify the asymptotic conditions at the punctures. If x is a generatorof CF ( L, L ), we specify that the input puncture in Figures 4 must be asymptoticto the γ t -pullback of x (as usual, if x is a linear combination of generators, we takethe disjoint union of the relevant moduli spaces). The output puncture must beasymptotic to the unique generator x ∈ CF ( L, L ). When l = ∞ , the first pairof rays in Figure 4(e) must be asymptotic to a point p such that γ − t ( p ) ∈ C ( L )(that is, p is an index 1 critical point of f ◦ γ t ) and the second pair of rays must beasymptotic to q ∈ C ( L ); we assume q is the unique minimum of g . The interiormarked points on the disks are unconstrained. Above, we have specified a 2-dimensional space of domains and the equationsover them. This gives us a moduli space of solutions (“pearly trees”) which is 1-dimensional, by our choice of indices. We remind the reader that a formal definitionof this moduli space falls into the setup of moduli spaces of pearly trajectories givenby Sheridan [45]. For our purposes, its description given above will suffice.The boundary of our 1-dimensional moduli space consists of: • solutions whose domains have parameter l = − l = ∞ , • solutions whose domains have parameter s = 0 or s = 2 π .We claim that solutions of the second type cancel pairwise. Indeed, recall thatthe disks without boundary punctures in Figures 5(a)–(d) are constant, and theperturbation data on the punctured disks for s = 0 , π are chosen in a way toprovide the same solutions, after a π -rotation on each disk. Let us describe moreexplicitly what happens when l <
0, as the case when l ≥ l < s >
0, where t = e is , by a disk with fixed boundarypunctures, stretched with length parameters − /l and 1 /s along the three stripsshown in Figure 6(a). The stretching procedure was described earlier, and ourchoice of perturbation data says that the stretched strips, and the sub-domain tothe left of the 1 /s -strip, carry an unperturbed pseudo-holomorphic equation. Sofor s = 0 we get the disks shown in Figure 6(b), with the unpunctured boundarymarked point attached to a constant disk, which means this boundary marked pointunconstrained. (As usual, the domain is considered up to complex automorphisms,so the unconstrained point does not prevent us from having rigid solutions.) Thisway, Figures 5(a) and 6(b) are drawings of the same configuration, for any l <
0. Ifwe rotate the disk in Figure 6(b) by π , we get precisely the disk with perturbationdata we would have got for s = 2 π , except that the boundary marked point (withthe attached constant disk) is on the different side of the boundary. Figure 6.
Left: the domains for l < s > s = 0when the principal disk is constant.We have shown that the solutions for s = 0 and s = 2 π are in a natural bijection;we claim that this bijection reverses the signs associated to those solutions asparts of the moduli space. This is ultimately related to the fact that the constantdisk is being glued to those solutions at the opposite boundary components. Onecan adopt a proof (which we will not provide in detail here) from the followingclassical example where an analogous sign issue has been treated. Suppose L , L S -INVARIANT LAGRANGIANS 19 are monotone Lagrangian submanifolds with obstruction numbers m ( L i ) ∈ Z , i =0 ,
1. Then the Floer differential d on CF ∗ ( L , L ) satisfies: d = m ( L ) − m ( L ).This relation arises from Maslov index 2 disks bubbling off the two sides of a 1-dimensional moduli space of Floer strips [34], see also [46, (2.3.9)] and [47, Figure 2].Here, indeed, the gluing of the side bubbles contributes with the opposite signs andresults in the term m ( L ) − m ( L ). The reference for the signs in this relationis [22], see specifically Remark 3.2.21(1), Formula (3.3.4) and Chapter 8 from thatbook. Although in our case we would be gluing a constant disk rather than Maslovindex 2 disk, and the non-constant curve in Figure 6(b) satisfies a different equationthan the standard Floer one (e.g. our equation is not R -invariant), the requiredorientation analysis is essentially the same.The outcome of the cancellation discussed above is that the count of configu-rations in Figure 4(a), i.e. (cid:93) M γ ( x ; x ), equals the count of configurations in Fig-ure 4(e), and it remains to compute the latter. Step 3. A Morse-theoretic computation.
Let us look at Figure 4(e). Recall that q ∈ L is the minimum of g , so the semi-infinite flowline of ∇ g flowing into q mustbe constant. Second, we have arranged the principal (central) disk to be constant,as well. So the configurations in Figure 4(e) reduce to those shown in Figure 7. Figure 7.
The domains when l = + ∞ and the principal disk to-gether with a flowline are constant. Here p is an index 1 criticalpoint of f ◦ γ t , and q is the minimum of f .The free parameter t = e is ∈ S \ { } is “unseen” by the domain after theprincipal disk became ghost (i.e. constant), but the equations still depend on it.First, consider the left disk and the left flowline in Figure 7, forgetting the rest ofthe configuration. Those disk and flowline satisfy the γ t -pullback of the equationdefining the PSS map Ψ, so for each t the linear combination of points p appearingas limits of such configurations equals γ t (Ψ( x )), where Ψ( x ) ∈ C ( L ) is the PSSimage which is a linear combination of index 1 critical points of f , so that γ t (Ψ( x ))is a combination of critical points of f ◦ γ t .Let us now add back the middle flowline, still forgetting the right flowline andthe right disk, and count the resulting configurations. The middle flowline is asemi-infinite flowline of ∇ ( f ◦ γ t ) ending at the point q ; note that q is not a criticalpoint of f . Suppose for the moment that we allow the right end of the middleflowline to be free (not constrained to q ) and denote the moduli space of suchconfigurations by P . Then we can consider the evaluation map at the right end ofthe flowline, ev : P → L . The image of ev, as a chain, is a linear combination ofunstable manifolds, with respect to the function f ◦ γ t , associated with the linearcombination of the critical points p which we have previously computed. Recallthat this linear combination of points p equals γ t (Ψ( x )). Consequently, if we denoteby C Ψ( x ) ⊂ L the disjoint union of (oriented, codimension 1) unstable manifolds ofthe Morse cochain Ψ( x ) ∈ C ( L ) with respect to f , then P = ( S \ { } ) × C Ψ( x ) , ev( t, z ) = γ t ( z ) . Those configurations which evaluate at q ∈ L are the intersection points C Ψ( x ) ∩ l ,where l = { γ t ( q ) } t ∈ S is the orbit of q . By perturbing γ t and f , the intersectionscan be easily made transverse, and we get: (cid:93) ( P × ev { q } ) = [ C Ψ( x ) ] · [ l ] = (cid:104) Ψ( x ) , l (cid:105) . Recall this is the count of the part of confugurations in Figure 7 which end up at q . Finally, the count of the rightmost flowlines (emerging from q ) plus the rightdisks in Figure 7 equals 1 L ∈ CF ( L, L ). Indeed, the g -unstable manifold of theminimum q is the whole manifold L (minus a codimension 2 subset), so the countis the same as the count of the rightmost disks only, and the latter by definitionproduces 1 L .Putting everything together and noting that (cid:15) = (cid:15) ( l ) by Remark 2.2 (since thetotal boundary in Figure 7 is contractible, for a small Hamiltonian perturbation),we get the statement of Proposition 2.8. One last thing is to argue that the modulispaces we have been using were regular.According to [46], the regularity of moduli spaces of pearly trajectories consistingof pseudo-holomorphic disks and flowlines is equivalent to the regularity of theseparate disks and flowlines not constrained to satisfy the incidence conditions,plus the transversality of the evaluation maps which account for the incidenceconditions.The non-constant disks in the proof carry the pseudo-holomorphic equation witha Hamiltonian perturbation which makes them regular. The constant disks areknown to be regular on their own; and it is easy to see that for generic f , g , theflowlines are transverse to the evaluation maps for all appearing configurations. (cid:3) Checking non-triviality in Hochschild cohomology.
In this subsectionwe prove Theorem 1.7(b), (c) (recall that part (a) was proved earlier, see Corol-lary 2.4). We have computed in Proposition 2.8 the map CO ( S ( γ )) | CF ( L,L ) , andit remains to see when the result survives to something non-trivial on the level ofHochschild cohomology, and thus distinguishes CO ∗ ( S ( γ )) ∈ HH ∗ ( L, L ) from theunit in HH ∗ ( L, L ).First, let us quickly recall the definition of Hochschild cohomology. Let A bean A ∞ algebra, and assume it is Z / K (cid:54) = 2. When char K = 2,we consider A as an ungraded algebra. The space of Hochschild cochains is, bydefinition CC ∗ ( A, A ) = (cid:89) k ≥ Hom ( A ⊗ k , A ) . If A is Z / CC ∗ ( A, A ) is Z / CC r ( A, A ) = (cid:81) k ≥ Hom ( A ⊗ k , A [ r − k ]). If h = { h k } k ≥ ∈ CC ∗ ( A ), h k : A ⊗ k → A , then the Hochschild differential of h is the sequence of maps( ∂h ) k ( a k , . . . , a ) = (cid:80) i + j ≤ k ( − ( r +1)( | a | + ... + | a i | + i ) · µ k +1 − i ( a k , . . . a i + j +1 , h j ( a i + j , . . . , a i +1 ) , a i , . . . , a )+ (cid:80) i + j ≤ k ( − r +1+ | a | + ... + | a i | + i · h k +1 − i ( a k , . . . a i + j +1 , µ j ( a i + j , . . . , a i +1 ) , a i , . . . , a ) . Here r is the Z / h . (When k = 0, the agreement is that Hom ( A ⊗ , A ) = A , so h is an element of A .) If char K = 2, we do not need the gradings as thesigns do not matter.Let us return to the A ∞ algebra CF ∗ ( L, L ). We continue to use the Morse Z -grading on the vector space CF ∗ ( L, L ) keeping in mind this grading is not respected
HE CLOSED-OPEN MAP FOR S -INVARIANT LAGRANGIANS 21 by the A ∞ structure. If L is oriented, the reduced Z / A ∞ structure so CF ∗ ( L, L ) is a Z / A ∞ algebra. If L is not oriented, wemust suppose char K = 2. Proof of Theorem 1.7(b).
We continue to work with CO ∗ on chain level. Becausethe homological closed-map is unital [46, Lemma 2.3], the Hochschild cohomologyunit is realised by the cochain 1 HH := CO ∗ (1) ∈ CC ∗ ( L, L ), where 1 is the unitin QH ∗ ( X ). The A ∞ category CF ∗ ( L, L ) need not be strictly unital, so the maps(1 HH ) k = CO k (1) need not vanish for k >
0. However, because the identity Hamil-tonian loop preserves L and has homologically trivial orbits on it, Proposition 2.8applies to 1 = S (Id) ∈ QH ∗ ( X ) and says that (1 HH ) ( x ) = 0 for any Floer cocycle x ∈ CF ( L, L ).Suppose CO ∗ ( S ( γ ))+ α · HH is the coboundary of an element h ∈ CC ∗ ( L, L ), forsome α ∈ K . By comparing ( ∂h ) with CO ( S ( γ )) + α · (1 HH ) , see Corollary 2.4,we get: µ ( h ) = ( − (cid:15) ( l ) ρ ( l ) · L + α · L . Here µ is the Floer differential and 1 L is a chain-level cohomology unit (2.3).The assumption HF ∗ ( L, L ) (cid:54) = 0 implies that the Floer cohomology unit 1 L cannotbe killed by the Floer differential. Therefore, we cannot solve the above equa-tion unless α = − ( − (cid:15) ( l ) ρ ( l ) and µ ( h ) = 0. Next, by comparing ( ∂h ) with CO ( S ( γ )) + α · (1 HH ) , see Proposition 2.8, for any Floer cocycle x ∈ CF ( L, L )we get( − | h | +1 µ ( x, h ) + ( − ( | h | +1)( | x | +1) µ ( h , x )+ µ ( h ( x )) + ( − | h | +1 h ( µ ( x )) = ( − (cid:15) ( l ) ρ ( l ) · (cid:104) Ψ( x ) , l (cid:105) · L . Here we are using the version of Ψ as in (2.5). Because x is a Floer cocycle,the last summand of the left hand side vanishes. If char K = 2, redenote a := h ∈ CF ∗ ( L, L ). If char K (cid:54) = 2, let a ∈ CF odd ( L, L ) be the odd degree part of h . By computing the signs in the above equality we get, for any Floer cocycle x ∈ CF ( L, L ): µ ( x, a ) + µ ( a, x ) + µ ( h ( x )) = ( − (cid:15) ( l ) ρ ( l ) · (cid:104) Ψ( x ) , l (cid:105) · L . Recall that µ ( h ) = 0 so µ ( a ) = 0 as well, and we get the following equality forFloer cohomology classes [ x ] , [ a ] ∈ HF ∗ ( L, L ) and Ψ( x ) ∈ H ( L ): µ ([ x ] , [ a ]) + µ ([ a ] , [ x ]) = ( − (cid:15) ( l ) ρ ( l ) · (cid:104) Ψ( x ) , l (cid:105) · L ∈ HF ∗ ( L, L ) . Now put x = Φ( y ), where y ∈ C ( L ) is a Morse cochain and Φ is the chain-levelmap from (2.4). The above equality means that for all [ y ] ∈ H ( L ), µ (Φ([ y ]) , [ a ]) + µ ([ a ] , Φ([ y ])) = ( − (cid:15) ( l ) ρ ( l ) · (cid:104) [ y ] , l (cid:105) · L ∈ HF ∗ ( L, L ) . This time, we have used the homology-level version of Φ from (2.5). The aboveequality is exactly prohibited by the hypothesis of Theorem 1.7(b), so Theo-rem 1.7(b) is proved. (cid:3)
Proof of Theorem 1.7(c).
Note that, on the homology level, CO ( S ( γ ) ∗ Q ) = CO ( S ( γ )) · CO ( Q ) = 1 L · CO ( Q ) = CO ( Q ) (here the dot denotes the µ product), so the only possible linear relation between CO ∗ ( S ( γ ) ∗ Q ) and CO ∗ ( Q ) is that CO ∗ (( S ( γ ) − ∗ Q ) = 0 , where 1 is the unit in QH ∗ ( X ). We have CO ∗ (( S ( γ ) − ∗ Q ) = CO ∗ ( S ( γ ) − (cid:63) CO ∗ ( Q ), where the symbol (cid:63) denotes the Yoneda product in Hochschild cohomol-ogy.Let us now return to working with CO ∗ on the chain level. Recall that if φ = { φ k } k ≥ , ψ = { ψ k } k ≥ ∈ CC ∗ ( L, L ) are Hochschild cochains, the k = 1 part oftheir Yoneda product by definition equals( φ (cid:63) ψ ) ( x ) = ± µ ( φ ( x ) , ψ ) ± µ ( φ , ψ ( x )) . There is an explicit formula for the signs which we do not need. Let us apply thisformula to CO ∗ ( S ( γ ) −
1) and CO ∗ ( Q ). We know that ( CO ∗ ( S ( γ ) − = 0 byCorollary 2.4, and ( CO ∗ ( S ( γ ) − ( x ) = CO ( S ( γ ))( x ) is given by Proposition 2.8for any Floer cocycle x ∈ CF ( L, L ). Consequently, we get: CO (( S ( γ ) − ∗ Q )( x ) = ( − (cid:15) ( l ) ρ ( l ) · (cid:104) Ψ( x ) , l (cid:105) · CO ( Q ) . From this point, the rest of the proof follows the one of Theorem 1.7(b). (cid:3) The closed-open map for real toric Lagrangians
In this section, after a short proof of Theorem 1.13, we look for further examplesof real toric Lagrangians where Theorem 1.7 can be effectively applied. We alsodiscover that Proposition 2.8, after additional work, allows to show that the Fukaya A ∞ algebra of some of the considered Lagrangians is not formal. In particular,we prove the results about real toric Lagrangians stated in Section 1 (except forProposition 1.1 and Corollary 1.2, which have been proved therein). We work witha coefficient field K of characteristic two throughout this section.3.1. A proof of Theorem 1.13.
Let X be a compact, smooth toric Fano variety,and D ⊂ X be a toric divisor corresponding to one of the facets of the polytopedefining X . There is a Hamiltonian circle action γ on X associated with D , whichcomes from the toric action by choosing a Hamiltonian which achieves maximumon D . A theorem of McDuff and Tolman [33] says the following. Theorem 3.1.
We have S ( γ ) = D ∗ , where D ∗ ∈ QH ∗ ( X ) is the Poincar´e dualof D . (cid:3) The loop γ never preserves the real Lagrangian L ⊂ X , but if we parametrise γ = { γ t } t ∈ [0 , then γ / ( L ) = L , see [24]. Consequently, α = { γ t ( L ) } t ∈ [0 , / isa loop of Lagrangian submanifolds, and moreover we have α = { γ t ( L ) } t ∈ [0 , inthe space of Lagrangian loops. There is an associated Lagrangian Seidel element S L ( α ) ∈ HF ∗ ( L, L ), which counts pseudo-holomorphic disks with rotating bound-ary condition α , and a single boundary puncture which evaluates to an element of HF ∗ ( L, L ). A theorem of Hyvrier [26, Theorem 1.13], based on the disk doublingtrick, computes S L ( α ). Theorem 3.2.
We have S L ( α ) = [ L ∩ D ] ∗ , where L ∩ D is the clean intersectionthat has codimension 1 in L , and [ L ∩ D ] ∗ ∈ H ( L ) ⊂ HF ∗ ( L, L ) is its dualclass. (cid:3) HE CLOSED-OPEN MAP FOR S -INVARIANT LAGRANGIANS 23 The inclusion H ( L ) ⊂ HF ∗ ( L, L ) is the PSS map Φ from Section 2, which isinjective because HF ∗ ( L, L ) ∼ = H ∗ ( L ) by Theorem 1.12. Proof of Theorem 1.13.
It suffices to prove that CO ( D ∗ ) = D ( F ( D ∗ )), where D ⊂ X is a toric divisor as above and D ∗ ∈ QH ∗ ( X ) is its dual class, because such D ∗ generate QH ∗ ( X ) as an algebra [33]. Let γ be the Hamiltonian loop correspondingto D as above, and α be the Lagrangian loop as above, such that α = { γ t ( L ) } t ∈ [0 , .It follows from Theorem 2.1 that CO ( S ( γ )) = S L ( α ) , and the latter can be rewritten as F ( S L ( α )), where F is the Frobenius map on HF ∗ ( L, L ). By Theorem 3.1, S ( γ ) = D ∗ , and by Theorem 3.2, S L ( α ) = [ L ∩ D ] ∗ .Finally, if we look at Haug’s construction [24] of the Duistermaat isomorphism D ,we will see that [ L ∩ D ] ∗ = D ( D ∗ ). Putting everything together, we get CO ( D ∗ ) = F ( D ( D ∗ )) . Because D is a ring map, it commutes with the Frobenius maps on HF ∗ ( L, L ) and QH ∗ ( X ), and the theorem follows. (cid:3) Split-generation for toric varieties with Picard rank 2.
It is knownthat the unique toric variety with Picard number 1 is the projective space. By atheorem of Kleinschmidt [29], see also [15], every n -dimensional toric Fano varietywhose Picard group has rank 2 (i.e. whose fan has n + 2 generators) is isomorphicto the projectivisation of a sum of line bundles over C P n − k :(3.1) X ( a , . . . , a k ) := P C P n − k ( O⊕O ( a ) ⊕ . . . ⊕O ( a k )) , a i ≥ , k (cid:88) i =1 a i ≤ n − k − . (The imposed conditions on the a i are equivalent to X being toric Fano). The n + 2vectors in Z n generating the fan of X ( a , . . . , a k ) are the columns of the followingmatrix:(3.2) I n × n − a ... ... − a k − ... ... − The minimal Chern number of X ( a , . . . , a k ) equals gcd( k + 1 , n − k + 1 − (cid:80) a i ),see [37]. Some of these varieties provide further examples where, using Theo-rems 1.13 and 1.7, we can prove the injectivity of CO ∗ and deduce split-generation. Theorem 3.3.
Let X := X ( a , . . . , a k ) be as above, L ⊂ X the real Lagrangian, K a field of characteristic 2. Suppose all a i are odd and gcd( k +1 , n − k +1 − (cid:80) a i ) ≥ .(a) If n − k + 1 is odd, then CO : QH ∗ ( X ) → HF ∗ ( L, L ) is injective.(b) If n − k + 1 is even, k is even and the numbers a i come in equal pairs, then CO ∗ : QH ∗ ( X ) → HH ∗ ( L, L ) is injective while CO is not.In both cases L split-generates F uk ( X ) .Proof. Let x, y ∈ H ( X ) be the generators corresponding to the last two columnsof the matrix (3.2). They generate QH ∗ ( X ) as an algebra and satisfy the followingrelations when char K = 2: x ( x + y ) k = 1 , y n − k +1 ( x + y ) − (cid:80) a i = 1 . (For brevity, we no longer use the symbol ∗ to denote the quantum product.) If n − k + 1 is odd, one can show that the Frobenius endomorphism F on QH ∗ ( X ) isan isomorphism, so CO is injective by Theorem 1.13. It follows that CO ∗ is alsoinjective, and split-generation follows from Theorem 1.10. Part (a) is proved.In the rest of the proof we work with the case (b), so let us redenote: n − k + 1 =2 r , k = 2 q , (cid:80) a i = 2 p . The rewritten relations in QH ∗ ( X ) are:(3.3) x ( x + y ) q = 1 , y r ( x + y ) − p = 1 . Lemma 3.4.
For the ring QH ∗ ( X ) as in (3.3), ker F is the ideal generated by y r ( x + y ) − p + 1 .Proof. Equations (3.3) are equivalent to x − p = y rq , y rq +2 r + y rp +2 p + 1 = 0 , where the second equation is rewritten from the second equation in (3.3) using thesubstitution x − p = y rq . This means if we denote R ( y ) = y rq + r + y rq + p + 1 , then R ( y ) = y r ( x + y ) − p + 1. Denote g = gcd(2 rq, p ) and let α, β ∈ Z be such that − rq · α + p · β = g. Consider the map φ : K [ u ] → QH ∗ ( X ) given by u (cid:55)→ x α y β ; this map is onto becausewe get(3.4) φ ( u p/g ) = y, φ ( u qr/g ) = x − using the given relations (note that the powers p/g , 2 qr/g are integral). Further,ker φ is obviously the ideal generated by V ( u ) where V ( u ) := R ( u p/g ), and weconclude that φ provides an isomorphism(3.5) φ : K [ u ] /V ( u ) ∼ = −→ QH ∗ ( X ) , V ( u ) = u pg (2 rq + r ) + u pg (2 rq + p ) + 1 . It is clear that V ( u ) generates the kernel of the Frobenius map on K [ u ] /V ( u ) .Because V ( u ) corresponds to y r ( x + y ) − p + 1 under φ , Lemma 3.4 follows. (cid:3) We continue the proof of Theorem 3.3(b). It turns out that, similarly to the caseof R P n ⊂ C P n studied in the introduction, the generator of ker F from Lemma 3.4equals S ( γ ) + 1 for a real Hamiltonian loop γ on X which preserves L setwiseand has homologically non-trivial orbits on it. To construct γ , we will need theadditional assumption that the a i come in equal pairs, so we assume the sequence( a i ) qi =1 is ( a , a , . . . , a q , a q ).Recall that X , being a toric manifold, is a quotient of C r +2 q +1 minus somelinear subspaces determined by the fan, by an action of ( C ∗ ) . Using the commonnotation, this action is given by z (cid:55)→ t v t v z , where z ∈ C r +2 q +1 and v , v are thevectors in Z r +2 q +1 given by the following two rows:(3.6) q entries r − entries entries a a . . . a q a q − . . . − − − − . . . − − . . . − z , . . . , z r +2 q +1 ) be the co-ordinates on C r +2 q +1 . The action of ( C ∗ ) on C r +2 q +1 commutes with the action of G = SU (2) q × SU (2 r ) , HE CLOSED-OPEN MAP FOR S -INVARIANT LAGRANGIANS 25 where the SU (2) factors act respectively on ( z , z ) , . . . , ( z q − , z q ), and SU (2 r )acts on ( z q +1 , . . . , z q +2 r − , z q +2 r +1 ), note we have omitted z q +2 r . (If we view X as a projective bundle over C P r − as in (3.1), the co-ordinates on which SU (2 r )acts are the homogeneous co-ordinates on the base.) Denote by G R = SO (2) q × SO (2 r ) the real form of G . Because all a i are odd, the action of ( − , +1) ∈ ( C ∗ ) coincides with the action of − I ∈ G . Consequently, the action of G descends toa Hamiltonian action of G/ ± I on X . Its real form G R / ± I preserves the realLagrangian L ⊂ X , and we let γ be the S -subgroup of G R / ± I defined as follows.This subgroup lifts to the path from I to − I in G R which is the image of therotation (cid:0) cos t sin t − sin t cos t (cid:1) ∈ SO (2), t ∈ [0 , π ], under the diagonal inclusions SO (2) ⊂ SO (2) q × SO (2) r ⊂ SO (2) q × SO (2 r ) = G R . Recall that we are assuming char K = 2. Lemma 3.5.
The homology class of γ -orbits on L is non-zero in H ( L ; K ) .Proof. Indeed, L is a real projective bundle over R P r − , and the orbits project tothe non-trivial cycle on the base, provided char K = 2. (cid:3) Lemma 3.6.
We have S ( γ ) + 1 = y r ( x + y ) − p + 1 (which is the generator of ker F from Lemma 3.4).Proof. Inside the complex group G/ ± I , the loop γ is homotopic to the loop γ (cid:48) liftingto the path from I to − I in G which is the image of the path (cid:16) e it e − it (cid:17) ∈ SU (2), t ∈ [0 , π ], under the diagonal inclusions SU (2) ⊂ SU (2) q × SU (2) r ⊂ SU (2) q × SU (2 r ) = G. By using the action of C ∗ ⊂ ( C ∗ ) corresponding to the first vector in (3.6), we seethat γ (cid:48) descends to the same Hamiltonian loop in X as the loop γ (cid:48)(cid:48) in G which actson C r +2 q +1 as follows:( z , . . . , z r +2 q +1 ) (cid:55)→ ( e it a z , e it a − z , . . . , e it aq +12 z q − , e it aq − z q ,z q +1 , e − it z q +2 . . . , e − it z r +2 q − , z r +2 q − , z r +2 q , e − it z r +2 q +1 ) , t ∈ [0 , π ] . Note that here t runs through [0 , π ], hence the -factors. Because all a i are odd, γ (cid:48)(cid:48) is now a closed loop in G , not only in G/ ± I . So by [33] its Seidel element S ( γ (cid:48)(cid:48) ) ∈ QH ∗ ( X ) can be computed as the quantum product of powers of thedivisors corresponding to the co-ordinates on C r +2 q +1 , where the powers are themultiplicities of rotations. Given char K = 2, and recalling that S ( γ (cid:48)(cid:48) ) = S ( γ (cid:48) ) = S ( γ ), we get: S ( γ ) = ( x + y ) a ( x + y ) a − . . . ( x + y ) aq +12 ( x + y ) aq − y − . . . y − = ( x + y ) p y − r . This element squares to 1 by (3.3) (in agreement with the fact γ has order 2 in π ( G/ ± I ) ∼ = Z / y r ( x + y ) − p , which proves Lemma 3.6. (cid:3) We conclude the proof of Theorem 3.3(b). By Lemmas 3.4 and 3.6, ker F isthe ideal generated by S ( γ ) + 1. Suppose P ∈ QH ∗ ( X ) such that CO ∗ ( P ) = 0 ∈ HH ∗ ( L, L ). Then CO ( P ) = 0, so P ∈ ker F by Theorem 1.13. Consequently P = ( S ( γ ) + 1) ∗ Q , and if P (cid:54) = 0 then Q / ∈ ker F (because otherwise we would get P ∈ (ker F ) = { } ). Apply Theorem 1.7(b) to the product ( S ( γ ) + 1) ∗ Q ; the lefthand side of ( ∗∗ ) vanishes because µ is commutative on HF ∗ ( L, L ) [24], and theright hand side is non-trivial for some y by Lemma 3.5 and because CO ( Q ) (cid:54) = 0. It follows that CO ∗ ( P ) (cid:54) = 0. We have shown that CO ∗ is injective, and split-generationfollows from Theorem 1.10. Note that w ( L ) = 0 holds for all real Lagrangians,as Maslov 2 disks come in pairs because of the action of the anti-holomorphicinvolution, see [24]. (cid:3) The following corollary in particular implies Proposition 1.4 from the introduc-tion.
Corollary 3.7.
Let X = Bl C P q − C P r +2 q − , and L ⊂ X be the real Lagrangian(diffeomorphic to Bl R P q − R P r +2 q − ). Assume gcd(2 q + 1 , r − q ) ≥ and thateither r or q are odd. Then CO ∗ : QH ∗ ( X ) → HF ∗ ( L, L ) is injective, although CO is not. Consequently, L split-generates F uk ( X ) .Proof of Proposition 1.4. Take X as in (3.1) with a = . . . = a k = 1, then X = Bl C P k − C P n , see e.g. [18, Proposition 11.14]. The additional hypotheses of thecurrent corollary make sure X satisfies all conditions of Theorem 3.3(b), whichtogether with the split-generation criterion (Theorem 1.10(b)) implies the corollary. (cid:3) In order to deduce non-displaceability results between the real Lagrangian L and other Lagrangians with arbitrary obstruction numbers, we need the followinglemma. Lemma 3.8.
Suppose char K = 2 , L is an object of F uk ( X ) w and CO ∗ : QH ∗ ( X ) → HH ∗ ( L, L ) is injective. Assuming Hypothesis 3.9 below, L × L split-generates F uk ( X × X ) . Note that by Lemma 1.9, the condition of Lemma 3.8 can only hold if QH ∗ ( X ) = QH ∗ ( X ) w or L is non-orientable. Proof.
First, observe that w ( L × L ) = 2 w ( L ) = 0. By [46], the injectivity of CO ∗ isequivalent to the fact that the open-closed map OC ∗ : HH ∗ ( L, L ) → QH ∗ ( X ) hitsthe unit 1 ∈ QH ∗ ( X ). Hypothesis 3.9.
There is a commutative diagram HH ∗ ( L, L ) ⊗ HH ∗ ( L, L ) (cid:47) (cid:47) OC ∗ ⊗OC ∗ (cid:15) (cid:15) HH split ∗ ( L × L, L × L ) OC ∗ prod (cid:15) (cid:15) HF ∗ ( X ) ⊗ HF ∗ ( X ) = (cid:47) (cid:47) HF ∗ ( X × X ) where OC ∗ prod is the open-closed map on the product, and the HF ∗ are HamiltonianFloer cohomologies, isomorphic to the quantum cohomologies of the correspondingspaces. Here HH split ∗ ( L × L, L × L ) indicates that the A ∞ structure on L × L is computedusing a split Hamiltonian perturbation and a product almost complex structure;such a choice can be made regular. We expect the hypothesis to hold followingGanatra [23, Remark 11.1] who stated it on chain level, in the setup of the wrappedFukaya category of an exact manifold. A slight complication is that the A ∞ algebraof L × L appearing in the top right corner of the diagram had to be equippedwith so-called one-sided homotopy units; their presense is denoted by a tilde in[23, Remark 11.1]. This does not affect the diagram on the homology level [23, HE CLOSED-OPEN MAP FOR S -INVARIANT LAGRANGIANS 27 Proposition 10.10], but we have not checked how this subtlety carries over to themonotone setup; therefore we leave Hypothesis 3.9 as a conjecture.Given Hypothesis 3.9, if OC ∗ hits the unit, then OC ∗ ⊗ OC ∗ and OC ∗ prod also do.The latter fact implies that CO ∗ is injective on the product, and split-generationfollows from Theorem 1.10(b). (cid:3) Corollary 1.5 from the introduction is a particular case of the following.
Corollary 3.10.
Let K be a field of characteristic 2 and L ⊂ X be as in Theo-rem 3.3(a) or (b), or as in Corollary 3.7. Suppose L (cid:48) ⊂ X another monotone La-grangian, perhaps equipped with a local system π ( L ) → K × , with minimal Maslovnumber at least 2 and such that HF ∗ ( L (cid:48) , L (cid:48) ) (cid:54) = 0 . If w ( L (cid:48) ) (cid:54) = 0 , assume Hypothe-sis 3.9. Then L ∩ L (cid:48) (cid:54) = ∅ .Proof. If w ( L (cid:48) ) = 0, this follows from the fact L split-generates F uk ( X ) andLemma 1.11. If w ( L (cid:48) ) (cid:54) = 0, we have that w ( L (cid:48) × L (cid:48) ) = 2 w ( L (cid:48) ) = 0, so L (cid:48) × L (cid:48) isan object of F uk ( X × X ) which is split-generated by L × L by Lemma 3.8. Then( L × L ) ∩ ( L (cid:48) × L (cid:48) ) (cid:54) = 0 by Lemma 1.11, and so L ∩ L (cid:48) (cid:54) = 0. (cid:3) An application to non-formality.
Recall that if
A → A (cid:48) is a quasi-iso-morphism of A ∞ categories, it induces an isomorphism HH ∗ ( A ) → HH ∗ ( A (cid:48) ), seee.g. Seidel [43, (1.14)]. We will need an explicit chain-level formula for this iso-morphism, which can be obtained by combining Seidel’s argument with Ganatra’sfunctoriality formulas [23, Section 2.9], and this requires a short account. We areassuming the reader is familiar with the basic language of A ∞ categories frome.g. [41, 46, 23], so that we can skip some basic definitions and present the otherones rather informally. For simplicity, we are working with char K = 2 so we won’thave to worry about signs, and restrict to A ∞ algebras rather than categories.Recall that if A is an A ∞ algebra, its Hochschild cohomology HH ∗ ( A ) can beseen as Hochschild cohomology HH ∗ ( A , A ) of A as an A−A bimodule. If F : A →A (cid:48) is a quasi-isomorphism between A ∞ algebras, it induces quasi-isomorphisms(3.7) CC ∗ ( A , A ) F ∗ −→ CC ∗ ( A , F ∗ A (cid:48) ) F ∗ ←−− CC ∗ ( A (cid:48) , A (cid:48) ) , which proves that HH ∗ ( A , A ) ∼ = HH ∗ ( A (cid:48) , A (cid:48) ). Chain-level formulas for the twointermediate quasi-isomorphisms, which we will now recall, were written downe.g. by Ganatra [23, Section 2.9] (in the context of Hochschild homology, but theseare easily adjusted to cohomology).If B , B (cid:48) are two A − A bimodules, a morphism G : B → B (cid:48) is a sequence of maps G k : A ⊗ i ⊗ B ⊗ A ⊗ j → B (cid:48) , i + j + 1 = k , satisfying a sequence of relations whichwe informally write down as (cid:88) (cid:63) G (cid:63) (Id ⊗ (cid:63) ⊗ µ (cid:63) A or B ⊗ Id ⊗ (cid:63) ) = (cid:88) (cid:63) µ (cid:63) B (cid:48) (Id ⊗ (cid:63) ⊗ G (cid:63) ⊗ Id ⊗ (cid:63) ) . Here (cid:63) are positive integers which are mutually independent but are such that thetotal number of inputs on both sides of the equation is the same; the sum is overall such possibilities; and the structure map on the left is µ (cid:63) A or µ (cid:63) B depending onwhether one of its arguments is in B . In its full form, the above relation should be written as follows: (cid:88) i + i + i + i = l G i + i − i +1 ( a , . . . , a i , µ i − i +1 B ( a i +1 , . . . , a i , b, a i +1 , . . . a i ) , a i +1 , . . . , a i )+ (cid:88) i + i + i + i = l G i + i − i +2 ( a , . . . , a i , µ i − i A ( a i +1 , . . . , a i ) , a i +1 , . . . , a i , b, a i +1 , . . . a i )+ (cid:88) i + i + i + i = l G i + i − i +2 ( a , . . . , a i , b, a i +1 , . . . , a i , µ i − i A ( a i +1 , . . . , a i ) , a i +1 , . . . a i )= (cid:88) i + i + i + i = l µ i + i − i +1 B (cid:48) ( a , . . . , a i , G i − i +1 ( a i +1 , . . . , a i , b, a i +1 , . . . , a i ) , a i +1 , . . . , a i ) , for a i ∈ A and b ∈ B . We will keep the informal style of notation, in which theinputs are omitted and the valencies are replaced by (cid:63) , further. The induced map G ∗ : CC ∗ ( A , B ) → CC ∗ ( A , B (cid:48) ) is defined by(3.8) ( G ∗ ( h )) (cid:63) = (cid:88) (cid:63) G (cid:63) (Id ⊗ (cid:63) ⊗ h (cid:63) ⊗ Id ⊗ (cid:63) )where h (cid:63) : A ⊗ (cid:63) → B and ( G ∗ ( h )) (cid:63) : A ⊗ (cid:63) → B (cid:48) . If G is a quasi-isomorphism, so is G ∗ .If A , A (cid:48) are two A ∞ algebras, a morphism F : A → A (cid:48) is a sequence of maps F (cid:63) : A ⊗ (cid:63) → A (cid:48) such that (cid:88) (cid:63) µ (cid:63) A (cid:48) ( F (cid:63) ⊗ . . . ⊗ F (cid:63) ) = (cid:88) (cid:63) F (cid:63) (Id ⊗ (cid:63) ⊗ µ (cid:63) A ⊗ Id ⊗ (cid:63) ) . Next, if B is an A (cid:48) −A (cid:48) bimodule, its two-sided pull-back F ∗ B is an A−A bimodulebased on the same vector space B , whose structure maps are [23, Section 2.8](3.9) µ (cid:63)F ∗ B = (cid:88) (cid:63) µ (cid:63) B ( F (cid:63) ⊗ . . . ⊗ F (cid:63) ⊗ Id B ⊗ F (cid:63) ⊗ . . . ⊗ F (cid:63) )There is also a morphism F ∗ : CC ∗ ( A (cid:48) , B ) → CC ∗ ( A , F ∗ B ) defined by(3.10) ( F ∗ ( h )) (cid:63) = (cid:88) (cid:63) h (cid:63) ( F (cid:63) ⊗ . . . ⊗ F (cid:63) )where h (cid:63) : ( A (cid:48) ) ⊗ (cid:63) → B and ( F ∗ ( h )) (cid:63) : A ⊗ (cid:63) → B . The total number of inputs herecan be zero, and F ∗ ( h ) = h . If F is a quasi-isomorphism, so is F ∗ .If, again, F : A → A (cid:48) is a morphism of A ∞ algebras, let F ∗ A (cid:48) be the A − A bimodule which is the pull-back of A (cid:48) seen as an A (cid:48) − A (cid:48) bimodule. Lemma 3.11.
The same sequence of maps F (cid:63) : A ⊗ (cid:63) → A (cid:48) provides a morphismof A − A bimodules
A → F ∗ A (cid:48) , also denoted by F .Proof. We must check that (cid:80) (cid:63) F (cid:63) (Id ⊗ (cid:63) ⊗ µ (cid:63) A ⊗ Id ⊗ (cid:63) ) = (cid:80) (cid:63) µ (cid:63)F ∗ A (Id ⊗ (cid:63) ⊗ F (cid:63) ⊗ Id ⊗ (cid:63) ).If we apply formula (3.9) to rewrite the right hand sum, the unique Id-factor in(3.9), which in our case is Id A (cid:48) , gets applied to the F (cid:63) -factor. So our right handsum equals (cid:80) (cid:63) µ (cid:63) A (cid:48) ( F (cid:63) ⊗ . . . ⊗ F (cid:63) ⊗ . . . ⊗ F (cid:63) ) which is exactly the condition that F is a morphism of A ∞ algebras A → A (cid:48) . (cid:3) This lemma explains the precise meaning of (3.7): if F : A → A (cid:48) is a morphism of A ∞ algebras, then the first map F ∗ from (3.7) is the push-forward of F consideredas a morphism of modules A → F ∗ A (cid:48) , given by formula (3.8). The second map in(3.7) is the pull-back as in (3.10).Next, recall that a general property of quasi-isomorphisms between A ∞ algebras(bimodues, etc.) is that they have quasi-inverses [41, Chapter 1]. If F : A → A (cid:48) is a quasi-isomorphism, then the
A − A bimodule morphism F from Lemma 3.11 HE CLOSED-OPEN MAP FOR S -INVARIANT LAGRANGIANS 29 is also a quasi-isomorphism, hence there is an A − A bimodule quasi-isomorphism G : F ∗ A (cid:48) → A which is a quasi-inverse of F , so we have quasi-isomorphisms:(3.11) CC ∗ ( A , A ) G ∗ ←−− CC ∗ ( A , F ∗ A (cid:48) ) F ∗ ←−− CC ∗ ( A (cid:48) , A (cid:48) ) . Their composition acts on Hochschild cochains by:( G ∗ F ∗ ( h )) (cid:63) = (cid:88) (cid:63) G (cid:63) (Id ⊗ (cid:63) ⊗ h (cid:63) ( F (cid:63) ⊗ . . . ⊗ F (cid:63) ) ⊗ Id ⊗ (cid:63) )where h (cid:63) : ( A (cid:48) ) ⊗ (cid:63) → A (cid:48) and ( G ∗ F ∗ ( h )) (cid:63) : A ⊗ (cid:63) → A . In particular, ( G ∗ F ∗ ( h )) = G ( h ), and if h = 0 ∈ A (cid:48) then(3.12) ( G ∗ F ∗ ( h )) ( u ) = G ( h ( F ( u ))) , u ∈ A (cid:48) . Note that G : A (cid:48) → A , F : A → A (cid:48) are chain maps with respect to µ A , µ A (cid:48) andare cohomology inverses of each other.Assume L ⊂ X is a Lagrangian preserved by a Hamiltonian loop γ which togethersatisfy the conditions of Theorem 1.7 (those conditions which are common to allparts of the theorem). Assume the A ∞ algebra CF ∗ ( L, L ) is formal, i.e. there isan A ∞ quasi-isomorphism F : HF ∗ ( L, L ) → CF ∗ ( L, L ). Denote(3.13) h := ( G ∗ F ∗ )( CO ∗ ( S ( γ )) − ∈ CC ∗ ( HF ∗ ( L, L ) , HF ∗ ( L, L )) , where G is a quasi-inverse of F , and G ∗ , F ∗ are as in (3.11). So h is a Hochschildcochain for the accociative algebra HF ∗ ( L, L ). Then by Corollary 2.4, h = 0,Proposition 2.8 and equation (3.12), h ( x ) = ρ ( l ) · (cid:104) Ψ( F ( x )) , l (cid:105) · G (1 L ) . (We have dropped the extra sign, working in characteristic two.) Let us additionallyassume that L is wide [10, Definition 1.2.1], i.e. there is a vector space isomorphismbetween H ∗ ( L ) and HF ∗ ( L, L ), and that L admits a perfect Morse function. Theseconditions enable us to identify CF ∗ ( L, L ) ∼ = HF ∗ ( L, L ) as vector spaces. Because G is cohomologically unital, G (1 L ) = 1 L ∈ HF ∗ ( L, L ), so(3.14) h ( x ) = ρ ( l ) · (cid:104) Ψ( F ( x )) , l (cid:105) · L ∈ HF ∗ ( L, L ) . Under our identifications, Ψ becomes an isomorphism between the vector spacesbelow, and F can be considered as algebra isomorphism from HF ∗ ( L, L ) to itself:(3.15) HF ∗ ( L, L ) F −−−−−−−→ algebra iso. HF ∗ ( L, L ) Ψ −−−−−−−→ v. space iso. H ∗ ( L ) . We now turn the discussion to Hochschild cohomology of monic algebras. Let f ( u ) ∈ K [ u ] be a polynomial and A := K [ u ] / ( f ) be the quotient algebra; it iscalled a monic algebra. This is an algebra in the ordinary associative sense; we canconsider it as an A ∞ algebra by equipping it with trivial higher structure maps.The Hochschild cohomology algebra HH ∗ ( A ) was computed by Holm [25]. Recallthat Hochschild cohomology of ungraded associative algebras is Z -graded (unlikeHochschild cohomology of non- Z -graded A ∞ algebras): cochains A ⊗ k → A are saidto have degree k , and the differential has degree 1. By [25, Proposition 2.2], HH k ( A ) = A, if k = 0 Ann A ( f (cid:48) ) if k > A/ ( f (cid:48) ) if k > . Although we already know the answer, let us compute HH ( A ) explicitly, as thiswill be helpful later. Lemma 3.12.
For an h : A → A which is a Hochschild cocycle in CC ( A ) , wemust necessarily have (3.16) h ( u m ) = amu m − for some fixed a ∈ A . Note that a = h ( u ) .Proof. Let us compute the Hochschild differential ∂h : A ⊗ A → A on the twoelements: u and u m − , for some m ∈ N . Because ∂h vanishes by assumption, weget: 0 = ( ∂h )( u, u m − ) = uh ( u m − ) + u m − h ( u ) + h ( u m ) . The desired formula follows by induction on m . (cid:3) So any cocycle h ∈ CC ( A ) is completely determined by a single element a = h ( u ) ∈ A , and h must meet an additional condition that h ( f ( u )) = h (0) = 0, whichis equivalent to a ∈ Ann A ( f (cid:48) ). As the differential CC ( A ) → CC ( A ) vanishes, weget an isomorphism HH ( A ) → Ann A ( f (cid:48) ), h (cid:55)→ h ( u ).We will further assume that char K = 2 and f (cid:48) = 0. The latter condition meansthat f is a sum of even powers of u . Denote by ψ : HH ( A ) → A the isomorphism φ ( h ) = h ( u ) from above. Note that if s ( u ) ∈ A is an arbitraryelement given by a polynomial with derivative s (cid:48) ( u ), then by (3.16) we get(3.17) ψ ( h ) = s (cid:48) ( u ) · h ( s ( u )) . For k >
1, we also have isomorphisms ψ : HH k ( A ) → A , all of which we denote bythe same letter by abusing notation; we will not need an explicit formula for theseisomorphisms when k > HH ∗ ( A ). In par-ticular, given h , h ∈ HH ( A ), their Yoneda product h (cid:63) h is determined by(3.18) ψ ( h (cid:63) h ) = ψ ( h ) · ψ ( h ) (cid:88) j odd f j u j − ∈ A, where f = (cid:80) j f j u j , f j ∈ K .The two strands of discussion can be combined in the following theorem. Theorem 3.13.
Let K be a field of characteristic 2, L ⊂ X a Lagrangian preservedby a Hamiltonian loop γ which together satisfy the conditions of Theorem 1.7 (thoseconditions which are common to all parts of the theorem). Assume there is an alge-bra isomorphism HF ∗ ( L, L ) ∼ = K [ u ] / ( f ) where f ( u ) = (cid:80) j ≥ f j u j is a polynomial,and also that L is wide and admits a perfect Morse function, so that we can identifythe vector spaces HF ∗ ( L, L ) ∼ = CF ∗ ( L, L ) , and Ψ : HF ∗ ( L, L ) → H ∗ ( L ) becomesan isomorphism of vector spaces. Further, assume: • f (cid:48) = 0 , and (cid:80) j odd f j u j − is invertible in K [ u ] / ( f ) ; • (cid:104) Ψ( r ( u )) , l (cid:105) = 1 for an element r ( u ) ∈ K [ u ] / ( f ) ∼ = HF ∗ ( L, L ) which generates HF ∗ ( L, L ) as an algebra; • S ( γ ) = 1 ∈ QH ∗ ( X ) .Then the Fukaya A ∞ algebra of L is not formal over K . HE CLOSED-OPEN MAP FOR S -INVARIANT LAGRANGIANS 31 Proof.
Supposing CF ∗ ( L, L ) is formal, let h be as in (3.13) and F be as in(3.15). Then there exists s ( u ) ∈ HF ∗ ( L, L ) (we view this element as a polyno-mial in K [ u ] / ( f )) such that F ( s ( u )) = r ( u ). Then by (3.14) h ( s ( u )) = ρ ( l ) · ∈ HF ∗ ( L, L ), so by (3.17), ψ ( h ) = ρ ( l ) · s (cid:48) ( u ) ∈ HF ∗ ( L, L ) . Further, note that h (cid:63) h = 0 because ( S ( γ ) + 1) = 0, so (3.18) yields ρ ( l ) · ( s (cid:48) ( u )) (cid:88) j odd f j u j − = 0 ∈ HF ∗ ( L, L ) . By hypothesis, this implies ( s (cid:48) ( u )) = 0, so s (cid:48) ( u ) ∈ ker F where F : K [ u ] / ( f ) → K [ u ] / ( f ) is the Frobenius endomorphism. In general, over char K = 2 it is alwaystrue that s (cid:48) ( u ) is a sum of even powers of u , so s (cid:48) ( u ) is a square of another poly-nomial: s (cid:48) ( u ) = ( t ( u )) . Then t ( u ) ∈ ker F , which implies t ( u ) ∈ ker F becauseker F , being an ideal in K [ u ] / ( f ), is necessarily prime. Consequently, s (cid:48) ( u ) = 0. So s ( u ) is a sum of even powers of u , hence the subalgebra generated by s ( u ) lies in thesubalgebra of K [ u ] / ( f ) generated by u , which is smaller than the whole K [ u ] / ( f ):for example, it does not contain the element u . (Recall that f is also a sum of evenpowers of u .) On the other hand, we know that F is an algebra isomorphism, F ( s ( u )) = r ( u ) and r ( u ) generates the whole HF ∗ ( L, L ) by hypothesis, so s ( u )should also generate HF ∗ ( L, L ), which is a contradiction. (cid:3)
Proof of Proposition 1.3.
Take the real loop γ preserving R P n +1 defined in theproof of Proposition 1.1 (see Section 1), and denote L = R P n +1 , X = C P n +1 .Recall that, if x ∈ H ( X ) is the generator, then QH ∗ ( X ) ∼ = K [ x ] / ( x n +2 + 1) and S ( γ ) = x n +1 , so S ( γ ) = 1. Also recall that l ∈ H ( L ) ∼ = K is non-zero. ByTheorem 1.12, HF ∗ ( L, L ) ∼ = K [ u ] / ( u n +2 + 1) where u ∈ CF ( L, L ) ∼ = K , and wehave (cid:104) Ψ( u ) , l (cid:105) = 1. Now apply Theorem 3.13 taking r ( u ) = u to conclude theproof. (cid:3) Proposition 3.14.
Let X = Bl C P q − C P r +2 q − , L ⊂ X be the real Lagrangian(diffeomorphic to Bl R P q − R P r +2 q − ). Assume that gcd(2 q + 1 , r − q ) ≥ andthat either r or q are odd. Then the A ∞ algebra of L is not formal over a charac-teristic 2 field.Proof. We recall that all real Lagrangians are wide by Theorem 1.12 and admita perfect Morse function by [24]. The fact that gcd(2 q + 1 , r − q ) ≥ k = 2 q = 2 p , a = . . . = a k = 1.We have already seen (3.5) that HF ∗ ( L, L ) ∼ = K [ u ] / ( f ) with f (cid:48) = 0, and it iseasy to check that (cid:80) j odd f j u j − is invertible provided that either r or q is odd(otherwise this element would vanish). Moreover, via (3.4) and Haug’s isomorphism(Theorem 1.12), u p/g corresponds to the generator of CF ( L, L ) ∼ = K such that (cid:104) Ψ( u ) , l (cid:105) = 1. Now apply Theorem 3.13 taking r ( u ) = u p/g . (cid:3) Non-formality of the equator on the sphere.
Proposition 1.3 says inparticular that the A ∞ algebra of an equatorial circle on S is not formal overchar K = 2. This is an especially simple case which can be verified by hand, and itis worth discussing it in more detail. Let L ⊂ S be a fixed equator, and L , L , . . . be a sequence of its small Hamiltonian perturbations; assume | L i ∩ L j | = 2 for each i, j . Then CF ( L i , L j ) ∼ = K is generated by an element which we denote by 1 (thisis the cohomological unit), and CF ( L i , L j ) ∼ = K is generated by an element which we denote by u (we use the same letter for all i, j ). Of the two intersection points L i ∩ L j , the point u is the one at which T u L j is obtained from T u L i by a smallpositive rotation with respect to the ω -induced orientation on S . Consider the A ∞ structure maps between the consequtive Lagrangians:(3.19) µ k : CF ∗ ( L k , L k +1 ) ⊗ . . . ⊗ CF ∗ ( L , L ) → CF ∗ ( L , L k +1 )given by counting immersed polygons as in [41]. These give a model for the A ∞ algebra of L , because all the L i differ small perturbations and we can canonicallyidentify the spaces CF ∗ ( L i , L i +1 ) with each other. The A ∞ maps will depend onthe particular arrangement of the L i , although up to quasi-isomorphism they givethe same A ∞ algebra. Remark 3.15.
The fact the A ∞ algebra of L defined using the count of polygons isquasi-isomorphic to the one defined using Hamiltonian perturbations seems not tohave been written down in detail but is widely accepted. An approach is sketchedin [42, Remark 7.2], and also performed in [45] in a slightly different setup.Let us compute some of the A ∞ structure maps using a specific choice of the L i .Fix a Hamiltonian H whose flow is the rotation of S ⊂ R around an axis which isnot orthogonal to the plane intersecting S along the equator L . Let L , L , . . . beobtained from L by applying that rotation by small but consequtively increasingangles, i.e. L i are time- t i push-offs of L under the flow of H , 0 = t < t < t . . . .The first four resulting circles L i are represented in Figure 8(a). The pairwiseintersections of the L i are contained in two opposite patches of the sphere; thosepatches are shown in the top and bottom of Figure 8(a) together with the L i onthem, which are depicted by straight lines. Both patches are drawn as if we lookat them from the same point “above” the sphere, so that the positive rotation(with respect to the orientation on S ) is counter-clockwise on the upper patchand clockwise on the lower patch. For this particular choice of perturbations, andfor each i < j , all degree-one points u ∈ CF ∗ ( L i , L j ) are located on the upperpatch, and all points 1 ∈ CF ∗ ( L i , L j ) are on the lower patch. Figure 8.
Two different configurations (a) and (b) consisting offour small Hamiltonian push-offs L , . . . , L (marked by numbers)of an equatorial circle on S . The image of the disk contributing to µ ( u, u, u ) = 1 is shaded.We claim that in this model we get: µ ( u, u, u ) = 1 , µ ( u, u,
1) = 0 , µ ( u, , u ) = 0 , µ (1 , u, u ) = 0 . HE CLOSED-OPEN MAP FOR S -INVARIANT LAGRANGIANS 33 For grading reasons, µ k ( u, . . . , u ) is a multiple of 1, and is determined by countingMaslov index 2 disks. There is a unique such disk; for k = 3 it is shown in grayshade in Figure 8(a) on the two patches; away from the patches this disc is just astrip between L and L . Also for grading reasons, the only other products whichcan possible be non-trivial are µ k ( u, . . . , u, , u . . . , u ) ∈ { , u } , where exactly oneinput is 1. It possible to check that these vanish for our configuration of the circles L i , at least when k = 3. Now note that µ (1 + u, u ) = 0 , µ (1 + u, u, u ) = 1 . The latter equality exhibits a non-trivial Massey product, seen as a well-definedelement of K [ u ] / (1 + u ) ∼ = K . An explanation where the Massey products generally belong to is found in [41,Remark 1.2] and explains the quotient by 1+ u above. The presence of a non-trivialMassey product is invariant under quasi-isomorphisms. To see this, recall that theanalogous fact for dg algebras is easy, and any A ∞ algebra is quasi-isomorphic toa dg algebra. Moreover, the Massey products for the A ∞ and dg models satisfy asimple relation [32, Theorem 3.1 and Corollary A.5], in particular, if triple Masseyproducts of an A ∞ algebra are non-trivial, they remain non-trivial for its dg-model.This gives us an alternative proof of the fact that the A ∞ algebra of the equatoron S is not formal.For any other arrangement of the L i , we will necessarily have µ (1 + u, u, u ) = 1 modulo 1 + u because of invariance of Massey products, meaning that µ (1 + u, u, u ) ∈ { , u } . For example, another possible configuration of L , . . . , L is shown in Figure 8(b);it is simply obtained from the earlier configuration by changing the ordering of the L i . In this new model, the maps µ k from (3.19) are now: µ ( u, u, u ) = 1 , µ ( u, u,
1) = 0 , µ ( u, , u ) = u, µ (1 , u, u ) = u. The unique disk contributing to µ ( u, u, u ) is shown in Figure 8(b) by gray shade.It is an immersed disk, and the domain over which it self-overlaps has darker shade.Note that in this model, the degree-one generators u ∈ CF ( L , L ), CF ( L , L ), CF ( L , L ) correspond to the intersection points on the upper patch, and thedegree-one generator u ∈ CF ( L , L ) corresponds to the intersection point on thelower patch. We see that we again get µ (1 + u, u, u ) = u .The existence of the Massey product above crucially required char K = 2, becauseotherwise we would not get µ (1 + u, u ) = 0, which is necessary to speak of thetriple Massey product of 1 + u with itself. If char K (cid:54) = 2, then HF ∗ ( L , L ) ∼ = K [ u ] / ( u − ∼ = K [ u ] / ( u − ⊕ K [ u ] / ( u + 1)is a direct sum of fields, whose Hochschild cohomology as an ordinary algebravanishes [28] in degree 2, in contrast to the case char K = 2. So any A ∞ algebra over K [ u ] / ( u − ⊕ K [ u ] / ( u + 1) is formal by [27] or [44, Section 3], in particular the A ∞ algebra of the equator on S is formal. For example, the product µ (1+ u, u, u )can be made to vanish after a formal diffeomorphism. Because of the non-trivialMassey product in characteristic 2, such a formal diffeomorphism, say over Q , willnecessarily involve division by 2, and cannot be realised by any geometric choiceof the push-offs L i . In comparison, the topological A ∞ algebra of the circle is formal over a fieldof any characteristic. Indeed, the topological A ∞ algebra is Z -graded, so if wemake this algebra to be based on the cohomology ring H ∗ ( S ) ∼ = K [ x ] /x where | x | = 1, the only possibly non-trivial products will be µ k ( x, . . . , x, , x, . . . , x ) forgrading reasons. On the other hand, every A ∞ algebra is quasi-isomorphic toa minimal, strictly unital one over a field of any characteristic [41, Lemma 2.1],[31, Theorem 3.1.1]. In a minimal strictly unital model, those products vanish bydefinition when k ≥ The closed-open map for monotone toric fibres
The mechanism of Theorem 1.7 for toric fibres.
Let X be an n -dimen-sional compact toric Fano variety, and T ⊂ X the unique monotone toric fibre.Evans and Lekili [20] proved (after this paper had appeared as preprint) that, ifchar K = 0, the Fukaya category F uk ( X ) w is split-generated by several copies of T ,equipped with the local systems corresponding to the critical points of the Landau-Ginzburg superpotential with critical value w ∈ K . We shall recall the formula forthe superpotential of a toric manifold in the next subsection; the common generalreferences are [14, 8, 21].Prior to [20], the split-generation by toric fibres had been proved only in thecase when the superpotential is Morse, see Ritter [38]. (For Ritter, proving split-generation requires considerable effort even in the Morse case, if W has severalcritical points with the same critical value. However, the difficulty is mainly relatedto the fact that he allows some non-compact toric varieties, where the injectivity of CO ∗ is no longer a criterion for split-generation and one must look at OC ∗ instead.If we work with compact manifolds, checking that CO ∗ is injective for an arbitraryMorse potential is easy: see Corollary 4.4). An example of a toric Fano varietywith non-Morse superpotential over C has been obtained by Ostrover and Tyomkin[36], and one can check that the superpotential in their case has an A singularity.To complete the literature overview, we should mention the work in progress byAbouzaid, Fukaya, Oh, Ohta and Ono [2] that will prove the split-generation resultfor toric manifolds that are not necessarily Fano.Because the toric fibre T is invariant under all the Hamiltonian loops comingfrom the torus action, it is an obvious example where Theorem 1.7 can be put to thetest. It turns out that it does allow to prove split-generation away from the Morsecase, though not too far from it: the superpotential is required to have at worst A singularities, and an extra condition char K (cid:54) = 2 , ∗ ) from Theorem 1.7depends on whether W is Morse or not. Equip T with a local system ρ whichcorresponds to a critical point of W , then ( T, ρ ) is wide and we can identify thevector spaces HF ∗ ( T, ρ ) ∼ = H ∗ ( T ) via the PSS map Φ. For convenience, let usrewrite equation ( ∗ ) from Theorem 1.7:( ∗ ) µ ( a, y ) + µ ( y, a ) = ρ ( l ) · (cid:104) y, l (cid:105) · T for each y ∈ H ( T ) . We are using the standard spin structure on T hence the sign ( − (cid:15) ( l ) is positive,see Remark 1.8. Recall that Theorem 1.7(b) can be applied if there exists no a ∈ HF ∗ ( T, ρ ) making ( ∗ ) hold. The Floer cohomology algebra of ( T, ρ ) is aClifford algebra determined by the Hessian of W at the point ρ , so the left-handside of ( ∗ ) is equal to Hess ρ W ( a, y ) · T , at least when a ∈ H ( T ); we are using HE CLOSED-OPEN MAP FOR S -INVARIANT LAGRANGIANS 35 informal notation for the moment. Therefore, finding an element a solving ( ∗ )reduces to finding an a such that(4.1) Hess ρ W ( a, − ) = const · (cid:104)− , l (cid:105) . The ability to find such an a depends on how degenerate Hess ρ W is. If ρ is a Morsepoint of W , such an a can always be found, so Theorem 1.7(b) does not apply.However, the Morse case can actually be covered by Theorem 1.7(a), as we explainbelow. On the other hand, when Hess ρ W has kernel, we will have some elements l ∈ H ( T ) for which equation (4.1) has no solution a . If we consider the S -actionwhose orbit is such an element l , Theorem 1.7(b) can be applied to the Seidelelement of this S -action to get some new information on CO ∗ which is not seenby CO . This information turns out to be sufficient only when the superpotentialhas A singularities, however, there is a possible way of improvement which wespeculate upon in the end of this section.4.2. The results.
Recall [13, 14, 21, 22] that the Landau-Ginzburg superpotentialof X is a Laurent polynomial W : ( K × ) n → K is given by W ( x , . . . , x n ) = (cid:88) e n (cid:88) j =1 x e j j where the first sum is over the outer normals e ∈ Z n to the facets of the polyhedrondefining X , and e j ∈ Z are their co-ordinates. (Sometimes, the superpotential iswritten down with a Novikov parameter, but we can ignore it because we will onlybe working with the monotone torus T .) We identify ( K × ) n with the space ofall local systems H ( T ; Z ) → K × . For ρ ∈ ( K × ) n , we write ( T, ρ ) for the torusequipped with this local system. Also, we will abbreviate HF ∗ ( T, ρ ) := HF ∗ (( T, ρ ) , ( T, ρ )) , and the same for Hochschild cohomology. It is known, see for example [36, Propo-sition 3.3], that(4.2) QH ∗ ( X ) ∼ = K [ x ± , . . . , x ± n ] /J ac ( W ) = O ( Z ) , where the Jacobian ideal J ac ( W ) is generated by ( ∂W/∂x , . . . , ∂W/∂x n ), and Z is the subscheme of Spec K [ x ± , . . . , x ± n ] defined by the ideal sheaf J ac ( W ). Then Z is a 0-dimensional scheme supported at the critical points of W , { ρ , . . . ρ q } = Crit W, ρ i ∈ ( K × ) n . The obstruction number of the torus is given by w ( T, ρ ) = W ( ρ ) ∈ K . Under the isomorphism (4.2), the quantum product is the usual product on O ( Z ),and the first Chern class of X is given by the function W itself. The generalisedeigenspace decomposition with respect to − ∗ c ( X ) is simply the decompositioninto the local rings at the points ρ i : K [ x ± , . . . , x ± n ] /J ac ( W ) ∼ = (cid:77) ρ i ∈ Crit W O ρ i ( Z ) , the eigenvalue of the ρ i -summand being the critical value W ( ρ i ). From Lemma 1.9,we see that HF ∗ ( T, ρ ) = 0 if ρ / ∈ Crit W . On the other hand, it is known that ( T, ρ i ) is wide for ρ i ∈ Crit W , i.e. HF ∗ ( T, ρ i ) is isomorphic as a vector space to H ∗ ( T ). Lemma 4.1.
Under the isomorphism (4.2), the map CO : QH ∗ ( X ) → HF ∗ ( T, ρ i ) is given by CO ( f ) = f ( ρ i ) · T . Here f ( x , . . . x n ) ∈ QH ∗ ( X ) , f ( ρ i ) ∈ K is the value of the function at ρ i ∈ Crit W ,and T ∈ HF ∗ ( T, ρ i ) is the unit.Proof. Because CO is a map of algebras, it suffices to prove the lemma when f = x k is a linear function, 1 ≤ k ≤ n . By [33], f = S ( γ ) for a Hamiltonian loop γ coming from the Hamiltonian torus action, such that the value of the local system ρ i on an orbit of γ equals the k th co-ordinate ρ ki , which is the same as the value f ( ρ i ). So CO ( f ) = f ( ρ i ) · T by Theorem 1.7(a). (cid:3) Corollary 4.2.
For ρ i (cid:54) = ρ j ∈ Crit W , the map CO ∗ | O ρi ( Z ) → HH ∗ ( T, ρ j ) van-ishes. Remark 4.3. If W ( ρ i ) (cid:54) = W ( ρ j ), Corollary 4.2 follows from Lemma 1.9. When W ( ρ i ) = W ( ρ j ), the statement is implicit in [38, Proof of Theorem 6.17] where itis shown that, dually, OC : HH ∗ ( T, ρ j ) → QH ∗ ( X )hits at most one summand of the form O ρ i , and we know by Lemma 4.1 that thissummand must actually be O ρ j . The proof in [38] is very different and relies onthe variation of the symplectic form. Proof.
Let f ∈ K [ x ± , . . . , x ± n ] be such that f ( ρ i ) (cid:54) = 0 and f ( ρ j ) = 0. Then, asan element of O ρ i ( Z ), f is invertible. If the corollary does not hold, CO ∗ ( f ) isalso invertible. On the other hand, CO ( f ) = 0 ∈ HF ∗ ( T, ρ j ) by Lemma 4.1. Themap HH ∗ ( T, ρ j ) → HF ∗ ( T, ρ j ), which takes a Hochschild cochain to its zeroth-order term, is a map of unital algebras, by the formula for the Yoneda productand because the Hochschild cohomology unit is represented by a cochain whosezeroth-order term is the Floer cohomology unit (this follows, for example, from theunitality of CO ∗ ). We have determined that f lies in the kernel of HH ∗ ( T, ρ j ) → HF ∗ ( T, ρ j ), but that contradicts the fact that f is invertible. This implies thecorollary. (cid:3) For w ∈ K , denote Crit w W = { ρ ∈ Crit W : W ( ρ ) = w } the set of all critical points of W with the same critical value w . We will sometimesdenote the restrictions of CO and CO ∗ to subalgebras of QH ∗ ( X ) by the samesymbol, when it is otherwise clear that we are considering a restriction. Corollary 4.4. If char K (cid:54) = 2 , the map CO : QH ∗ ( X ) w −→ (cid:77) ρ i ∈ Crit w W HF ∗ ( T, ρ i ) is injective if and only if all points of Crit w W are Morse.Proof. By Corollary 4.2, CO is injective if and only if its restrictions CO : O ρ i ( Z ) → HF ∗ ( T, ρ i ) are injective for each ρ i . The map O ρ i ( Z ) → K which takes f ∈ O ρ i ( Z )to its value f ( ρ i ) is injective if and only if O ρ i ( Z ) is a field, which is equivalent tothe fact that ρ i is a Morse point of W when char K (cid:54) = 2. Now apply Lemma 4.1. (cid:3) HE CLOSED-OPEN MAP FOR S -INVARIANT LAGRANGIANS 37 Proposition 4.5.
Suppose char K (cid:54) = 2 , and W has an A singularity at a point ρ , then CO ∗ : O ρ ( Z ) → HH ∗ ( T, ρ ) is injective.Proof. After an integral linear change of co-ordinates, we may assume that theHessian of W at ρ is the diagonal matrix: Hess ρ W = diag(1 , . . . , , O ρ ( Z ) is generated, as a vector space, by the two elements 1 and x n , wherethe linear function x n corresponds to the kernel of Hess ρ W . Indeed, after a furthernon-linear change of coordinates with the identity linear part, we can bring W tothe canonical form W (˜ x , . . . , ˜ x n ) = W ( ρ ) + n − (cid:88) i =1 (˜ x s − ρ i ) + (˜ x n − ρ n ) . Here ρ i ∈ K are the co-ordinates of ρ . Then J ac ( W ) = ((˜ x − ρ ) , . . . , (˜ x n − − ρ n − ) , (˜ x n − ρ n ) ), so O ρ ( Z ) is generated, as a vector space, by 1 and ˜ x n . Because x n , as a function of ˜ x , . . . , ˜ x n , equals ˜ x n plus terms of order at least 2, it is easyto see that the elements 1 , x n also generate the vector space O ρ ( Z ).Let us identify HF ∗ ( T, ρ ) with H ∗ ( T ) via the PSS map Φ. Recall that, ingeneral, HF ∗ ( T, ρ ) is the algebra generated by y , . . . , y n ∈ H ( T ) with relations y p y q + y q y p = ∂ x p x q W ( ρ ) . In particular, in our case we get y p y n + y n y p = 0 for any 1 ≤ p ≤ n , so y n ∈ HF ( T, ρ ) anti-commutes with any element of HF ∗ ( T, ρ ) of odd degree. Conse-quently, the left hand side of equation ( ∗ ) from Theorem 1.7 vanishes if we put y = y n , and allow a to be of arbitrary odd degree.Returning to our generator x n ∈ O ρ ( Z ), we have x n = S ( γ ) for a Hamiltonian S -action (coming from the toric action) such that the element y n ∈ HF ( T, ρ ) isdual to the orbit l ∈ H ( T ) of γ , so that (cid:104) y n , l (cid:105) = 1. So if we put y = y n , the righthand side of equation ( ∗ ) from Theorem 1.7 becomes ρ n · T (cid:54) = 0. Hence ( ∗ ) has nosolution, and Theorem 1.7(b) says that CO ∗ ( x n ) and 1 HH = CO ∗ (1) are linearlyindependent. (cid:3) Combining the above discussion with the split-generation criterion, we get thefollowing corollary.
Corollary 4.6.
Suppose char K (cid:54) = 2 , and each critical point ρ i ∈ Crit w W iseither Morse or an A singularity. Then the copies of the monotone toric fibrewith local systems { ( T, ρ i ) } ρ i ∈ Crit w W split-generate F uk ( X ) w . (cid:3) A way of extending Theorem 1.7.
It is in fact not surprising that The-orem 1.7 turned out to be efficient only for A singularities. The main result onwhich Theorem 1.7 is based upon is Proposition 2.8, which computes the linear part CO of the closed-open map, while the only non-Morse singularity whose localJacobian is generated as a vector space by constant and linear functions is the A singularity (for which the Jacobian is generated by 1 and x n as above). One couldextend the computation in Proposition 2.8 to all orders of CO ∗ when applied toproducts of 1-cochains on L ; we conjecture that the following holds. Conjecture 4.7.
The restriction CO k ( S ( γ )) | CF ( L,L ) ⊗ k : CF ( L, L ) ⊗ k → CF ( L, L ) equals (4.3) ( − (cid:15) ( l ) ρ ( l ) · ( l ∗ ) ⊗ k · L on symmetrised tensor products of Floer 1-cocycles. Here l ∗ : CF ( L, L ) → K isgiven by l ∗ ( x ) = (cid:104) Ψ( x ) , l (cid:105) , and l ∈ H ( L ) is the orbit of γ . Remark 4.8.
As in Proposition 2.8, part of the statement is that the image ofthis restriction necessarily lands in CF ( L, L ): this follows for degree reasons.Although the proof of the above formula should be analogous to Proposition 2.8,one new issue arises which we have not checked in detail. Consider the modulispaces M γ ( x , . . . , x k ; x ) from Section 2 and the pearly moduli spaces analogousto Figure 7 but with more inputs. The new issue is a different type of domaindegenerations coming from the collision of input points: e.g. several puncturedinputs for a curve in M γ ( x , . . . , x k ; x ) may collide and create a bubble. To prove(4.3), one would need to argue that these collisions cancel out when the input stringis symmetrised.Formula (4.3) is a chain level computation, and whether it survives to somethingnon-trivial in Hochschild cohomology will be governed by equations generalisingequation ( ∗ ) from Theorem 1.7; those equations will be determined by the A ∞ structure maps on L up to order k + 1. When L is the monotone toric fibre, the A ∞ structure maps have been related to higher-order partial derivatives of W byCho [13], and intuitively, the more degenerate the superpotential is, the more non-trivial information from (4.3) survives to Hochschild cohomology. Consequently,these observations are a possible starting point for proving split-generation resultsfor toric Fano varieties with other degenerate superpotentials. However, furtherdevelopment of this discussion seems both complicated and not particularly de-manded, given the general results of [2, 20]. References [1] M. Abouzaid. A geometric criterion for generating the Fukaya category.
Publ. Math. Inst.Hautes ´Etudes Sci. , 112(1):191–240, 2010.[2] M. Abouzaid, K. Fukaya, Y.-G. Oh, H. Ohta, and K. Ono. In preparation. 2016.[3] M. Abouzaid and I. Smith. Homological mirror symmetry for the four-torus.
Duke Math. J. ,152:373–440, 2010.[4] M. Abreu and L. Macarini. Remarks on Lagrangian intersections in toric manifolds.
Trans.Amer. Math. Soc. , 365:3851–3875, 2013.[5] P. Albers. A Lagrangian Piunikhin-Salamon-Schwarz morphism and two comparison homo-morphisms in Floer homology.
Int. Math. Res. Notices , 2008.[6] G. Alston and L. Amorim. Floer cohomology of torus fibers and real Lagrangians in Fanotoric manifolds.
Int. Math. Res. Notices , 12:2751–2793, 2012.[7] L. Amorim. The K¨unneth theorem for the Fukaya algebra of a product of Lagrangians. arXiv:1407.8436 , 2014.[8] D. Auroux. Mirror symmetry and T-duality in the complement of an anticanonical divisor.
J. G¨okova Geom. Topol. , 1:51–91, 2007.[9] P. Biran and O. Cornea. A Lagrangian Quantum Homology. In
New Perspectives and Chal-lenges in Symplectic Field Theory , volume 49 of
CRM Proc. Lecture Notes , pages 1–44. CRM-AMS, 2009.[10] P. Biran and O. Cornea. Rigidity and uniruling for Lagrangian submanifolds.
Geom. Topol. ,13:2881–2989, 2009.[11] F. Charest. Source Spaces and Perturbations for Cluster Complexes. arXiv:1212.2923 , 2012.[12] F. Charette and O. Cornea. Categorification of Seidel’s representation.
Israel J. Math ,211(1):67–104, 2016.
HE CLOSED-OPEN MAP FOR S -INVARIANT LAGRANGIANS 39 [13] C.-H. Cho. Products of Floer cohomology of torus fibers in toric Fano manifolds. Commun.Math. Phys. , 260:613–640, 2005.[14] C.-H. Cho and Y.-G. Oh. Floer cohomology and disc instantons of Lagrangian torus fibers inFano toric manifolds.
Asian J. Math. , 10(4):773–814, 2006.[15] S. Choi, M. Masuda, and D.Y. Suh. Quasitoric manifolds over a product of simplices.
OsakaJ. Math. , 47:109–129, 2010.[16] O. Cornea and F. Lalonde. Cluster homology: an overview of the construction and results.
Electron. Res. Announc. Amer. Math. Soc. , 12:1–12, 2006.[17] J. J. Duistermaat. Convexity and tightness for restrictions of Hamiltonian functions to fixedpoint sets of an antisymplectic involution.
Trans. Amer. Math. Soc. , 275(1):417–429, 1983.[18] D. Eisenbud and J. Harris. .Draft of a book.[19] M. Entov and L. Polterovich. Rigid subsets of symplectic manifolds.
Compositio Math. ,145:773–826, 2009.[20] J. D. Evans and Y. Lekili. Generating the Fukaya categories of Hamiltonian G-manifolds. arXiv:1507.05842 , 2015.[21] K. Fukaya, Y.-G. Oh, H. Ohta, and K. Ono. Lagrangian Floer theory on compact toricmanifolds: survey. arXiv:1011.4044 , 2010.[22] K. Fukaya, Y.-G. Oh, H. Ohta, and K. Ono.
Lagrangian Intersection Floer Theory: Anomalyand Obstruction , volume 46 of
Stud. Adv. Math.
American Mathematical Society, Interna-tional Press, 2010.[23] S. Ganatra. Symplectic cohomology and duality for the wrapped Fukaya category. arXiv:1304.7312 , 2013.[24] L. Haug. On the quantum homology of real Lagrangians in Fano toric manifolds.
Int. Math.Res. Notices , 14:3171–3220, 2013.[25] T. Holm. Hochschild cohomology rings of algebras k [ X ] / ( f ). Beitr¨age Algebra Geom. , 41:291–301, 2000.[26] C. Hyvrier. Lagrangian circle actions.
Algebr. Geom. Topol. , 16:1309–1342, 2016.[27] T. V. Kadeishvili. The structure of the A ∞ -algebra, and the Hochschild and Harrison coho-mologies. Trudy Tbiliss. Mat. Inst. Razmadze Akad. Nauk Gruzin. SSR , 91:19–27, 1988.[28] C. Kassel. A K¨unneth formula for the cyclic cohomology of Z / Math. Ann. ,275(4):683–699, 1986.[29] P. Kleinschmidt. A classification of toric varieties with few generators.
Aequationes Math. ,35(2-3):254–266, 1988.[30] M. Konstantinov. Higher rank local systems in Lagrangian Floer theory. arXiv:1701.03624 ,2017.[31] K. Lef`evre-Hasegawa.
Sur les A ∞ cat´egories . PhD thesis, Universit´e Paris 7, 2002.[32] D.-M. Lu, J. H. Palmieri, Q.-S. Wu, and J. J Zhang. A-infinity structure on Ext algebras. J.Pure Appl. Alg , 213(11):2017–2037, 2009.[33] D. McDuff and S. Tolman. Topological properties of Hamiltonian circle actions.
Int. Math.Res. Pap. , 2006.[34] Y.-G. Oh. Addendum to Floer cohomology of Lagrangian intersections and pseudoholomor-phic discs, I.
Comm. Pure Appl. Math. , 48(11):1299–1302, 1995.[35] Y.-G. Oh. Relative Floer and quantum cohomology and the symplectic topology of Lagrangiansubmanifolds. In
Proceedings for the 1994 Symplectic Topology program , Publ. of the NewtonInstitute. Cambridge University Press, 1996.[36] Y. Ostrover and I. Tyomkin. On the quantum homology algebra of toric Fano manifolds.
Sel.Math., New ser. , 15(1):121–149, 2009.[37] Z. Qin and Y. Ruan. Quantum cohomology of projective bundles over P n . Trans. Amer.Math. Soc. , 350(9):3615–3638, 1998.[38] A. Ritter. Circle-actions, quantum cohomology, and the Fukaya category of Fano toric vari-eties.
Geom. Topol. , 20:1941–2052, 2016.[39] A. Ritter and I. Smith. The monotone wrapped Fukaya category and the open-closed stringmap.
Selecta Math. , 23(1):533–642, 2017.[40] P. Seidel. π of symplectic automorphism groups and invertibles in quantum homology rings. Geom. Funct. Anal. , 7:1046–1095, 1997. [41] P. Seidel.
Fukaya Categories and Picard-Lefschetz Theory . European Mathematical Society,Zurich, 2008.[42] P. Seidel. Homological Mirror Symmetry for the genus two curve.
J. Algebraic Geom. , 20:727–769, 2011.[43] P. Seidel.
Abstract analogues of flux as symplectic invariants , volume 137 of
M´emoires de laSMF . 2014.[44] P. Seidel.
Homological Mirror Symmetry for the Quartic Surface , volume 236, number 1116of
Mem. Amer. Math. Soc.
J. DifferentialGeom. , 29(2):271–367, 2011.[46] N. Sheridan. On the Fukaya category of a Fano hypersurface in projective space.
Publ. Math.Inst. Hautes ´Etudes Sci. , 2016.[47] I. Smith. A symplectic prolegomenon.
Bull. Amer. Math. Soc. , 52:415–464, 2015.[48] J. Smith. Discrete and continuous symmetries in monotone Floer theory. arXiv:1703.05343 ,2017.
E-mail address : [email protected], [email protected]@cam.ac.uk, [email protected]