Professional Documents
Culture Documents
Quantum Steenrod Operations and Fukaya Categories: Zihong Chen
Quantum Steenrod Operations and Fukaya Categories: Zihong Chen
Zihong Chen
Abstract
arXiv:2405.05242v1 [math.SG] 8 May 2024
This paper is concerned with quantum cohomology and Fukaya categories of a closed monotone
symplectic manifold X, where we use coefficients in a field k of characteristic p > 0. The first main
result of this paper is that the quantum Steenrod operations QΣ admit an interpretation in terms of the
Fukaya category of X, via suitable versions of the open-closed maps. Using this, we show that QΣ, whose
definition is intrinsic to characteristic p, is compatible with certain structures inherited from the quantum
connection in characteristic 0. We then turn to applications of these results. The first application is an
arithmetic proof of the unramified exponential type conjecture for X that satisfies Abouzaid’s generation
criterion over Q, which uses a reduction mod p argument. Next, we demonstrate how the categorical
perspective provides new tools for computing QΣ beyond the reach of known technology. We also explore
potential connections of our work to arithmetic homological mirror symmetry.
Contents
1 Introduction 2
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Quantum Steenrod operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Relation to cyclic homology and the cyclic open-closed map . . . . . . . . . . . . . . . . . . . 5
1.4 Cyclic open-closed map and the t-connection . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 Change of base ring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1
7 A conjectural B-side formula 42
7.1 Example: AN singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Appendix B Grading 49
B.1 Z-grading on the monotone Fukaya category . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
B.2 From the q-connection to the t-connection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
1 Introduction
1.1 Motivation
This paper is concerned with quantum cohomology and Fukaya categories with coefficients in a field k of
characteristic p > 0. For simplicity, we work in the context of a closed monotone symplectic manifold (X, ω),
where the technical aspects of the relevant Floer-theoretic structures are straightforward. Why is there a
particular interest in working over positive characteristics?
1) As a most straightforward answer, even if one is only interested in characteristic 0 phenomena, reducing
coefficients mod p can be a useful tool in studying them, an idea that is ubiquitous and fruitful in
algebraic geometry. One goal of this paper is to demonstrate how this idea can be applied to symplectic
topology. In particular, we use the following conjecture of [KKP] and [GGI] as a motivating question.
Conjecture 1.1 (The unramified exponential type conjecture). The quantum connection (QH ∗ (X, C)((t)), ∇QH
d ),
dt
cf. (1.11), admits a finite direct sum decomposition
M d λ
∇QH
d = (C((t)), − ) ⊗ ∇reg
d , (1.1)
dt dt t2 dt
λ
2
In fact, motivated by mirror symmetry, one can apply a similar idea to show that the canonical
connection on the (formal) twisted de Rham cohomology associated to a pair (Y, W ), where Y is an
affine variety and W is a regular function with isolated critical points, has unramified exponential type,
cf. Theorem C.1. A reader more familiar with matrix factorizations than quantum Steenrod operations
may find Theorem C.1, whose proof is short and elementary, as motivation for the argument in section
6.
2) The effectiveness of characteristic p methods often comes from the fact that there are structures that
only exist in characteristic p, such as the Steenrod operations in classical topology, the Frobenius map
and Cartier operators in algebraic geometry, and the p-curvature in differential equations.
The same phenomenon appears in symplectic topology, and the characteristic p structures we are
concerned with are the quantum Steenrod operations QΣ : QH ∗ (X, k) ⊗ QHZ/p ∗
(X, k) → QHZ/p∗
(X, k);
cf. section 3.3 for their definition. In fact, quantum Steenrod operations are (or are expected to be)
related to all of the structures mentioned in the previous paragraph, and we will explore some of these
relations in this paper:
• By definition, QΣ’s are deformations of the classical Steenrod operations.
• QΣc1 is related to the p-curvature of the quantum connection, cf. Lemma 6.8 and Conjecture
6.10; this plays an important role in the proof of Theorem 1.2.
• We also expect QΣ to be ‘mirror’ to certain operations on differential forms in characteristic p on
the B-side, cf. section 7 for a discussion.
3) It is well known that Fukaya categories and homological mirror symmetry have interesting arithmetic
aspects [LPe], [LPo], [GHHPS]. Even simplistic symplectic manifolds such as the cylinder, when one
considers its Fukaya category over a general coefficient ring R, can have interesting homological mirrors
that encode the arithmetic properties of R, cf. [EL].
As a concrete example, take X to be the intersection of two quadrics in CP 5 . Over C, a celebrated result
of Ivan Smith [Smi, Theorem 1.1] states that a component of the Fukaya category of X is equivalent
to the Fukaya category of a genus 2 curve. In section 5, we show that there is a simple obstruction
for the prior mentioned equivalence to hold over any coefficient field not containing a square root of
−1, cf. Proposition 5.2. We do not know whether this is the √ only obstruction, i.e. whether Smith’s
equivalence holds over any coefficient field that contains a −1.√ However, what we can show is that
if Smith’s equivalence were true over some field k (containing a −1) of odd positive characteristic, it
imposes strong constraints on the structure of the quantum cohomology of X over k. In fact, having
such an equivalence will determine the quantum Steenrod operations of X over k, cf. Proposition 5.3.
The idea behind Proposition 5.3, and the main result of this paper, is that quantum Steenrod operations
admit an interpretation in terms of the Fukaya category. More precisely, for any A∞ -category, there
is an action of its Hochschild cohomology on its Z/p-equivariant Hochschild homology, which we call
the Z/p-equivariant cap product, cf. Definition 2.12. When the A∞ -category in question is the Fukaya
category of X, this action recovers QΣ via suitable versions of the open-closed maps. Using the
categorical interpretation,
• We show that certain structures induced by the quantum connection in characteristic 0 impose
constraints on the QΣ, which are operations intrinsic to characteristic p, cf. Corollary 1.10.
• We give new ways to compute QΣ using ideas from categorical enumerative invariants, cf. section
5.
Finally, we should mention that the three perspectives above are interconnected. For instance, the categorical
interpretation of QΣ is the key step in establishing nilpotence for the p-curvature of the quantum connection,
and which ultimately allows us to prove Theorem 1.2. With these motivations in mind, let us now state the
main results of the paper.
3
1.2 Quantum Steenrod operations
Quantum Steenrod operations are a collection of operations on equivariant quantum cohomology that come
from genus 0 Gromov-Witten theory with mod p coefficients. First introduced by Fukaya [Fuk], and system-
atically developed by Wilkins [Wil], they have seen various links and applications to Hamiltonian dynamics
[Shel], arithmetic mirror symmetry [Sei3] and representation theory [Lee]. They have the property of being
covariantly constant with respect to the quantum connection in characteristic p [SW], which gives effective
computation in low degrees. Nonetheless, our understanding of quantum Steenrod operations in the general
case is still limited:
• Relatively few computations of quantum Steenrod operations have been made beyond low degrees;
• The relationship between quantum Steenrod operations and Gromov-Witten theory in characteristic 0
remains mysterious.
This paper aims to advance our understanding of both questions above by systematically studying quantum
Steenrod operations on a closed monotone symplectic manifold from the perspective of its Fukaya category.
We now describe the setup and state the main results. Throughout this paper, (X, ω) will be a monotone
symplectic manifold of dimension 2n. Let A be some base ring. Let QH ∗ (X, A) denote the quantum
cohomology of X with coefficient in A; for λ ∈ A, let Fuk(X, A)λ denote the monotone Fukaya category with
disk potential λ ∈ A, see section 3 for a review. In this paper, the coefficient ring A will come in two flavors:
1) Characteristic 0: we consider R ⊂ K ⊂ √ Q the ring of integers of some number field K, up to inverting
finitely many elements, e.g. Z[ 12 ], Z[ 12 , 5, i]. Throughout the paper, we always assume that 2 is inverted
in R.
2) Positive Characteristics: we consider, for an odd prime p, fields k of characteristic p.
Let us start with the story in characteristic p. In section 2, we will introduce a p-fold Hochschild chain
complex p CC(A) for any A∞ -category A over k. This chain complex is quasi-isomorphic to the usual
Hochschild chain complex CC(A), but has a chain level Z/p-action, which can be thought of as induced
from the S 1 -action on CC(A) via the inclusion of the p-th roots of unity Z/p ⊂ S 1 . We denote its associated
negative Z/p-equivariant complex (or Z/p-homotopy fixed point) as CC Z/p (A). There is an action
Z/p
\
: HH ∗ (A) × HH∗
Z/p Z/p
(A) → HH∗ (A), (1.2)
which we call the Z/p-equivariant cap product. It has the following properties:
p(p−1) TZ/p
(ϕ ∪ φ, a) = (−1)|ϕ||φ| 2
TZ/p TZ/p
C1) It is a graded multiplicative action: (ϕ, (φ, a)), where ∪
denotes the cup product on Hochschild cohomology.
C2) It is additive in the second variable, and becomes additive in the first variable after multiplying by t
(the formal S 1 -equivariant variable of degree 2), the latter meaning t
TZ/p TZ/p
(ϕ + φ, −) = t (ϕ, −) +
TZ/p
t (φ, −).
C3) If A is cohomologically unital, in which case HH ∗ (A) is unital, then
TZ/p
is a unital action.
TZ/p
(aφ, −) = ap
T Z/p
C4) It is Frobenius p-linear: for a ∈ k, (φ, −).
In section 4, we define a Z/p-equivariant open-closed map (when context is clear, we omit λ from notation)
Z/p Z/p ∗+n
OCλ : HH∗ (Fuk(X, k)λ ) → QHZ/p (X, k), (1.3)
where the Z/p-action on QH ∗ is trivial. The right hand side of (1.3) can be more explicitly written as
where |t| = 2, |θ| = 1, θ2 = 0. (1.3) is a Z/p-analogue of the Ganatra’s cyclic open-closed map, and it will be
a crucial ingredient in proving the categorical formula for quantum Steenrod operations.
4
Following [SW], quantum Steenrod operations can be viewed as an action of QH ∗ (X, k) on QH ∗ (X, k)Z/p .
In particular, fixing a cohomology class b ∈ QH ∗ (X, k) there is a quantum Steenrod action associated to b
We review the definition and some of its properties in more detail in section 3.
Our first main result is that the Z/p-equivariant open-closed map intertwines the quantum Steenrod opera-
tions (1.5) with the categorical action (1.2).
Theorem 1.3. In the above setting, for all b ∈ QH ∗ (X, k) and λ ∈ k, the diagram
OC Z/p
QH ∗+n (X, k)[[t, θ]]
Z/p
HH∗ (Fuk(X, k)λ )
CO(b)
TZ/p
− QΣb (1.6)
OC Z/p
QH ∗+p|b|+n (X, k)[[t, θ]]
Z/p
HH∗+p|b| (Fuk(X, k)λ )
as vector spaces [Gan2]. CC∗nu (A) has an k[ϵ]/ϵ2 ≃ C∗ (S 1 )-action given by the non-unital Connes’ operator
1
B nu , cf. (2.28). The t-complex for negative cyclic homology CC S (A) is defined as CC∗nu (A)[[t]], with |t| = 2
1
and differential bnu + tB nu , whose cohomology we denote by HH∗S (A). [Gan2] defined the (negative) cyclic
open-closed map
1 1
OC S : HH∗S (Fuk(X, k)λ ) → QH ∗+n (X, k)[[t]]. (1.8)
The following theorem compares the cyclic and Z/p-equivariant open-closed maps, and is the key to applying
characteristic zero techniques in computing quantum Steenrod operations.
Theorem 1.4. ([Che, Theorem 1.6]) 1) For any cohomologically unital A∞ -category A, there exists a quasi-
1 1
isomorphism Φp : CC S (A) ⊕ CC S (A)θ ≃ CC Z/p (A), where θ is a formal variable of degree 1.
2) The following diagram is homotopy commutative:
OC Z/p
CC Z/p (Fuk(X, k)λ ) QH(X, k)[[t, θ]]
1 1
OC S ⊕OC S θ
Φp (1.9)
1 1
CC S (Fuk(X, k)λ ) ⊕ CC S (Fuk(X, k)λ )θ
The proof of Theorem 1.4 is rather technical, and was obtained in a prequel to this paper [Che].
Remark 1.5. From an abstract point of view, Theorem 1.4. 1) is an analogue of the classical Z/p-Gysin
sequence in topology, which relates the homology of the S 1 -homotopy fixed points of an S 1 -space with that
of the induced Z/p ⊂ S 1 -homotopy fixed points.
5
1.4 Cyclic open-closed map and the t-connection
In this subsection, we review the classical definition of t-connections in symplectic topology and properties
1
of the cyclic open-closed map over Q. Recall that the negative cyclic homology HH S (A) of an A∞ -category
A is equipped with the Getzler-Gauss-Manin t-connection given on the chain level1 by
P
GGM d Γ i{ k≥0 (2 − k)mk }
∇d/dt := + − , (1.10)
dt 2t 2t2
where Γ is the length operator on Hochschild chains and i is the contraction of a cyclic chain by a Hochschild
cochain, cf. [Hug, section 3.2], [OS, section 5]. We remark that in the literature, the S 1 -equivariant variable
is often named u (and (1.10) is named the ‘u-connection’), whereas we use t.
There is an analogous t-connection on S 1 -equivariant quantum cohomology, called the quantum t-connection
given by
d µ c1 ⋆
∇QH
d/dt := dt + t − t2 , (1.11)
This was originally conjectured by [GPS]. It was first proved by [Hug, Theorem 1.7] in a slightly simplified
technical setting. The proof that is closer to our setting is [PS, Theorem 6.3.5], which however considers the
so-called ‘q-connection’; we explain how Theorem 1.4 can be derived from the results in loc.cit. in Appendix
B.
We recall the following general fact about formal t-connections.
Lemma 1.7. [Hug, Lemma 2.13] Write {ej } for a generalized eigenbasis of c1 ⋆, and write ∇ = dt d
+ At20 + At1
in this basis. In particular, A0 is in Jordan normal form with diagonal entries the eigenvalues of c1 ⋆. Then,
there exists a basis {vj } of QH ∗ (X)[[t]] such that
1) vj |u=0 = ej .
d
P∞
2) Write ∇ = dt + t12 i=0 Ãi ti in the basis {vj }, then Ã0 = A0 and all Ãi ’s respect the generalized
eigen-decomposition of A0 .
3) The diagonal blocks of Ã1 and A1 agree.
Lemma 1.7 implies that there exists a decomposition of the quantum t-connection
M
QH ∗ (X, Q)[[t]] = QH ∗ (X, Q)[[t]]λ (1.13)
λ∈spec(c1 ⋆)
such that QH ∗ (X, Q)[[t]]λ |t=0 = QH ∗ (X, Q)λ . We call the decomposition (1.13) the elementary Hukuhara-
Levelt-Turittin (HLT) decomposition. On the categorical side, as λ ranges over the eigenvalues of c1 ⋆, we
1
consider the big direct sum λ∈spec(c1 ⋆) CC S (Fuk(X, Q)λ ), where each summand is equipped with its own
L
1
Getzler-Gauss-Manin t-connection. A consequence of Theorem 1.6 is that OC S must also intertwine the
two decompositions (cf. [Hug, Cor 6.5.]), i.e.
1 1
OC S (HH∗S (Fuk(X, Q)λ )) ⊂ QH ∗+n (X, Q)[[t]]λ . (1.14)
1 We remark that to be precise, in order for this formula to work for A = Fuk(X) (which in our convention is an uncurved
λ
A∞ -category, i.e. m0 = 0, cf. section 3.2), we need to pass to a quasi-equivalent A∞ -category Fuk(X)λ → Fuk(X)+ λ which
contains a strict unit e+ , and set m0 = λ · e+ (which makes Fuk(X)+λ a weakly curved A ∞ -category). This can be achieved, for
instance, by using Fukaya’s homotopy unit construction, cf. [Gan1, section 10]. An alternative approach is to stick to Fuk(X)λ ,
− λ2
but use the connection obtained by tensoring (1.10) with the one-dimensional connection E t , cf. (6.4).
6
We are interested in whether the decomposition in (1.13) holds over some finitely generated extension R ⊂ Q
of Z (which allows us to reduce mod p, for almost all primes p). We make the following simple observation.
Lemma 1.8. Suppose R is a ring such that QH ∗ (X, R) has a decomposition QH ∗ (X, R) = λ∈spec(c1 ⋆) QH ∗ (X, R)λ
L
into generalized eigenspaces of c1 ⋆, and that the difference between two distinct eigenvalues of c1 ⋆ is invert-
ible in R. Then this generalized eigen-decomposition extends to a unique decomposition of the t-connection
(1.13) over R.
Proof. By assumption, we can find a basis {ej } as in Lemma 1.7 for which c1 ⋆ has Jordan normal form. As
in the proof of Lemma 1.7 (cf. [Hug, Lemma 2.13]), the desired change of basis over R[[t]] is obtained by
inductively choosing matrices Tm such that 1) Tm has no diagonal blocks with respect to the Jordan normal
form of c1 ⋆ under {ej } and 2) the off-diagonal blocks of [A0 , Tm ] cancel with those of Am . Given these
two conditions, one can inductively solve for the Tm ’s uniquely, and in particular, the only denominators
appearing in the entries of Tm are the differences of two distinct eigenvalues of A0 (equivalently the eigenvalues
of c1 ⋆).
is an isomorphism.
We now say a few words about conditions A1)-A3). Condition A2) guarantees that the following (cf. section
2.9 [Sh1] for the result over C) holds over R: for λ, λ′ ∈ spec(c1 ⋆), the image of QH ∗ (X, R)λ′ under
CO : QH ∗ (X, R) → HH ∗ (Fuk(X, R)λ ) is 0 if λ ̸= λ′ . (1.16)
Assuming conditions A1) and A2), Lemma 1.8 implies that there is a unique elementary HLT decomposition
over R, i.e. a decomposition
M
QH ∗ (X, R)[[t]] = QH ∗ (X, R)[[t]]λ (1.17)
λ∈spec(c1 ⋆)
Assumption A3) is related to Abouzaid’s generation criterion, by work of [Gan1]. In particular, results in
loc.cit. implies that if the total open-closed map (1.15) hits the unit in QH ∗ (X, R), then it is an isomorphism.
Since R ⊂ Q, (1.14) continues to hold over R, i.e. we have
1 1
OC S (HH∗S (Fuk(X, R)λ )) ⊂ QH ∗+n (X, R)[[t]]λ . (1.18)
By a standard spectral sequence argument with respect to the t-filtration, A3) implies that the cyclic open-
closed map
1 1
OC S : HH∗S (Fuk(X, R)λ )) → QH ∗+n (X, R)[[t]]λ (1.19)
7
is an isomorphism for each λ ∈ spec(c1 ⋆).
Now suppose we have a homomorphism of rings π : R → k, where k is a field k of odd characteristic p. For
a fixed R, such homomorphism (usually not unique) exists for almost all p.
Consider the diagram of abelian groups induced by the change of base rings π:
1
1 OC S /k
HH∗S (Fuk(X, k)π(λ )) QH ∗+n (X, k)[[t]]
π∗ π∗ (1.20)
1
1 OC S /R
HH∗S (Fuk(X, R)λ ) QH ∗+n (X, R)[[t]]
The c1 ⋆-eigenvalues over k are the images under π of the c1 ⋆-eigenvalues over R and the generalized c1 ⋆-
eigenspaces satisfy (note by condition A2) π is injective on spec(c1 ⋆))
This always lies in QH ∗ (X, k)[[t, θ]]π(λ) by Corollary 1.9. This proves part 2). If λ ̸= λ′ , then COπ(λ′ ) (b) = 0,
which proves part 1).
8
Corollary 1.10 reveals a surprising aspect of quantum Steenrod operations: even though QΣ is defined
intrinsically in characteristic p, it must preserve certain structures inherited from characteristic 0. This will
be a key ingredient in our reduction mod p approach to the unramified exponential type conjecture in section
6.
Organization
The organization of this paper is as follows. Section 2 will be devoted to concepts in homological algebra. We
start by reviewing the definitions of A∞ -categories, bimodules, and their Hochschild (co)homology. Then,
we introduce the p-fold Hochschild complex and its Z/p-equivariant complex, and define the main algebraic
operation of this paper: the Z/p-equivariant cap product. In section 3, we review the definition of the
monotone Fukaya category and the construction of the quantum Steenrod operations. In section 4, we define
the Z/p-equivariant open-closed map and prove the main result Theorem 1.3. In section 5, we apply our
main results to computing the quantum Steenrod operations in the example of intersection of two quadrics in
CP 5 , conditional on Smith’s categorical equivalence over k. In section 6, we discuss a reduction mod p proof
of the unramified exponential type conjecture, under the assumption that X satisfies Abouzaid’s generation
criterion over Q. In section 7, we discuss a conjectural B-side formula for quantum Steenrod operations.
Acknowledgements
First and foremost, I would like to thank my advisor Paul Seidel for suggesting this line of inquiry and for
his patient and helpful guidance. I would also like to thank Denis Auroux, Sheel Ganatra, Jae Hee Lee, Ivan
Smith, and Nicholas Wilkins for helpful discussions at various points. This research was partially supported
by the Simons Foundation, through a Simons Investigator grant (256290).
9
of degree 1 − r − s, where we denote
Definition 2.3. Let A be an A∞ -category. The diagonal bimodule A∆ is the A − A bimodule defined by
A∆ (X, Y ) = homA (Y, X), and
r|1|s −1
r+1+s
µA∆ (xr , · · · , x1 , a, y1 , · · · , ys ) = (−1)✠−s +1 µA (xr , · · · , x1 , a, y1 , · · · , ys ). (2.6)
That is, we sum over all possible ways to insert φ into ϕ, and as before ✠s denotes the sum of degrees of
xs , · · · , x1 , using reduced degree unless it is a bimodule entry.
When one of xn , · · · , x1 is a C −D bimodule entry and φ is the bimodule structure map, we use ϕ◦(µC , φ, µD )
to denote the above expression plus summing over first applying µC or µD and then applying ϕ, with the
appropriate Koszul signs. For instance, the A∞ -category structural identity and the A∞ -bimodule structural
identity can be expressed as µA ◦ µA = 0 and µM ◦ (µC , µM , µD ) = 0, respectively.
Definition 2.4. A pre-morphism of C − D bimodules of degree k F : M → M′ is the data of:
of degree k − r − s.
The pre-morphisms between M, M′ form a chain complex homC−D (M, M′ ) where the differential is given
by
δ(F) = µM′ ◦ F − (−1)|F | F ◦ (µC , µM , µD ). (2.9)
A pre-morphism F is closed if δ(F) = 0 (also called a morphism of bimodules). The operation ◦ in (2.7)
defines a composition
M ⊗D N (2.11)
10
is a C − E bimodule whose underlying graded vector space is M ⊗ T D[1] ⊗ N , where T denotes the tensor
algebra (where we consider composable morphisms), and D[1] denotes using the reduced degree on homD ;
0|1|0
the differential µM⊗D N is given by
0|1|0
µM⊗D N (m, d1 , · · · , dk , n) =
−(t+1)
0|1|t s|1|0
X ✠
X
(−1) −(k+1) µM (m, d1 , · · · , dt ) ⊗ dt+1 ⊗ · · · ⊗ dk ⊗ n + m ⊗ d1 ⊗ · · · ⊗ dk−s ⊗ µN (dk−s+1 , · · · , dk , n)
X −(j+i+1)
✠
+ (−1) −(k+1) m ⊗ d1 ⊗ · · · ⊗ dk−s ⊗ µiD (dj+1 , · · · , dj+i ) ⊗ dj+i+1 ⊗ · · · ⊗ dk ⊗ n; (2.12)
for r > 0 or s > 0, structure maps
−(t+1)
r|1|0 r|1|t
X ✠
µM⊗D N (c1 , · · · , cr , m, d1 , · · · , dk , n) = (−1) −(k+1) µM (c1 , · · · , cr , m, d1 , · · · , dt ) ⊗ dt+1 ⊗ · · · dk ⊗ n,
(2.13)
0|1|s j|1|s
X
µM⊗D N (m, d1 , · · · , dk , n, e1 , · · · , es ) = m ⊗ d1 ⊗ · · · dk−j ⊗ µN (dk−j+1 , · · · dk , n, e1 , · · · , es ) (2.14)
r|1|s
and µM⊗D N = 0 if r > 0, s > 0.
where we use reduced degree for hom spaces of A and usual degree for M. The differential is given by
defined by
X
(−1)† µM (x1 , · · · , xk , a, y1 , · · · , yi , ϕ(yi+1 , · · · , yl−s ), yl−s+1 , · · · , yl ),
Ψ(ϕ)(x1 , · · · , xk , a, y1 , · · · , yl ) :=
(2.19)
Pl
where † = |ϕ| · ( j=l−s+1 ∥yj ∥). Moreover, Ψ is a quasi-isomorphism whenever A is cohomologically unital,
cf.[Gan1, Prop 2.5.].
There is a product structure on the Hochschild cochain complex CC ∗ (A, A) called the cup product. It is
given by
X
ϕ ∪ ψ(x1 , · · · , xk ) := (−1)♣ µk (x1 , · · · , xi , ϕ(xi+1 , · · · , xi+r ), · · · , ψ(xj+1 , · · · , xj+l ), · · · , xk ), (2.20)
Pk Pj Pk
where ♣ = |ϕ| · ( s=j+l+1 ∥xs ∥ + t=i+r+1 ∥xt ∥) + |ψ| · ( s=j+k+1 ∥xs ∥). The cup product defines an
algebra structure on HH ∗ (A), which is unital when A is cohomologically unital.
There is also an algebra structure on 2 HH ∗ (A) whose product is given by composition of bimodule mor-
phisms. On cohomology, Ψ of (2.18) is a map of algebras and is unital when A is cohomologically unital.
11
2.3 Hochschild homology and cyclic homology
Let A be an A∞ -category and M be an A-bimodule. The Hochschild chain complex of A with coefficients
in M is M
CC∗ (A, M) := M(Xk , X0 ) ⊗ A(X0 , · · · , Xk ), (2.21)
X0 ,X1 ,··· ,Xk
Pk
with grading given by deg(m ⊗ x1 ⊗ · · · ⊗ xk ) = |m| + i=1 ∥xi ∥ and differential given by
X i
b(m ⊗ x1 ⊗ · · · ⊗ xk ) = (−1)♯j µM (xk−j+1 , · · · , xk , m, x1 , · · · , xi ) ⊗ xi+1 ⊗ · · · ⊗ xk−j
X −(s+j+1)
+ (−1)✠−k m ⊗ x1 ⊗ · · · ⊗ xs ⊗ µjA (xs+1 , · · · , xs+j ) ⊗ xs+j+1 ⊗ · · · ⊗ xk , (2.22)
where the new symbol ♯ denotes
k k−j
−(i+1)
X X
♯ij =
∥xs ∥ · |m| + ∥xt ∥ + ✠−(k−j) . (2.23)
s=k−j+1 t=1
Remark 2.6. In words, the sign is given by the parity of: if a cyclic permutation is involved, we sum the
degrees of elements that get moved in front, and multiply with the sum of degrees of all other elements; if
furthermore a structure map (e.g. µA or µM ) is applied, we sum the degrees of elements to the right of µ,
and add to the previous sum. Reduced degree is used except for a bimodule entry.
The cohomology of this complex is called Hochschild homology, and the above chain model is also called the
cyclic bar complex. We denote CC∗ (A) := CC∗ (A, A∆ ).
Next, we recall from [Gan2] that the non-unital Hochschild complex is defined as
CC∗nu (A) := CC∗ (A) ⊕ CC∗ (A)[1]. (2.24)
To define the differential, one considers the following two operations.
X s
b′ (xd ⊗ · · · ⊗ x1 ) := (−1)✠1 xd ⊗ · · · ⊗ xs+j+1 ⊗ µ(xs+j , · · · , xs+1 ) ⊗ xs ⊗ · · · ⊗ x1 +
X d−j
(−1)✠1 µ(xd , · · · , xd−j+1 ) ⊗ xd−j ⊗ · · · ⊗ x1 , (2.25)
and d d d−1
d∧∨ (xd ⊗ · · · ⊗ x1 ) := (−1)✠2 +∥x1 ∥·✠2 +1 x1 ⊗ xd ⊗ · · · ⊗ x2 + (−1)✠1 xd ⊗ · · · ⊗ x1 . (2.26)
nu
Then, the differential b is defined as
nu b d∧∨
b := . (2.27)
0 b′
When A is cohomologically unital, the natural inclusion CC∗ (A) → CC∗nu (A) is a quasi-isomorphism, see
[Gan1, Prop. 2.2].
The crucial property of the non-unital Hochschild complex is that there is a chain level S 1 -action on CC∗nu (A),
for a not necessarily strictly unital A∞ -category A. The S 1 -action takes the form of the (non-unital) Connes
operator, given by
X i k k
B nu (xk ⊗ · · · ⊗ x1 , yl ⊗ · · · ⊗ y1 ) := (−1)✠1 ✠i+1 +∥xk ∥+✠1 +1 (0, xi ⊗ · · · ⊗ x1 ⊗ xk ⊗ · · · ⊗ xi+1 ). (2.28)
i
The Connes operator satisfies (B ) = 0 and bnu B nu + B nu bnu = 0, and therefore can be viewed as a unital
nu 2
12
2.4 The N -fold Hochschild complex
Let N be any positive integer. As seen in the previous subsection, the Hochschild homology of an A∞ -
category is equipped with an S 1 -action. It is natural to ask the following question: consider the inclusion
Z/N ⊂ S 1 as N -th roots of unities, what is the induced Z/N -action on Hochschild homology? To answer
this question at the chain level, we consider the following variant of the Hochschild chain complex.
Definition 2.7. The N -fold Hochschild chain complex N CC∗ (A) is defined as
where the tensor product is N -fold. For N = 2, this was considered by e.g. [Gan1, Definition 2.30], [LT,
section 3.7.2].
More explicitly, the underlying graded vector space of N CC∗ (A) is given by
where diag means we consider cyclically composable sequence of morphisms. The differential can be de-
scribed schematically by the following picture, where the boldface marked points correspond to distinguished
(bimodule) entries.
defined by
X PkN N Pp l Pkl l
(−1)( j=i+1 ∥xj ∥)(1+ l=1 (|x |+ j=1 ∥xj ∥)) µ(xN N 1 1 1 2 2 2 2 N N N
i+1 , · · · , xkN , x , x1 , · · · , xk1 , x , x1 , · · · , xj )⊗xj+1 ⊗· · ·⊗x ⊗x1 ⊗· · ·⊗xi ,
(2.33)
where the first term µ(· · · ) in (2.33), together with x3 , · · · , xN , becomes the new distinguished bimodule
entries. When A is homologically unital, ϵ0N,N −1 is a quasi-isomorphism for all N ≥ 2, cf. the proof of [Gan1,
Proposition 2.2]. Taking the composition of ϵ0k,k−1 for k = 2, 3, · · · , N , one obtains a quasi-isomorphism
13
There is a chain level Z/N -action on N CC∗ (A), where the generator τ ∈ Z/N acts by cyclically permuting
the N ‘blocks’:
−1
(−1)† xN ⊗ xN N 1 1 1
1 ⊗ · · · ⊗ xkN ⊗ x ⊗ x1 ⊗ · · · ⊗ xk1 ⊗ · · · ⊗ x
N −1
⊗ xN
1 ⊗ · · · ⊗ xkNN−1
−1
, (2.35)
where
kN p−1 N kj
−1 X
X X X
N
∥xN j
∥xji ∥
† = |x | + i ∥ · |x | + (2.36)
i=1 j=1 j=1 i=1
given by
(F, m ⊗ xn ⊗ · · · ⊗ x1 ) 7→ (−1)† F(xr , · · · , x1 , m, xn , · · · , xs ) ⊗ xs−1 ⊗ · · · ⊗ xr+1 , (2.38)
Pr Pn s−1 ′ ⊗N
where † = ( i=1 ∥xi ∥) · ( i=r+1 ∥xi ∥ + |m|) + |F|✠r+1 . When M = M = A∆ := A∆ ⊗A · · · ⊗A
A∆ (n times), precomposing (2.37) with the canonical chain map 2 CC ∗ (A)⊗N = EndA−A (A∆ )⊗N →
EndA−A (A⊗N
∆ ), we get a chain map
Y
N : 2 CC ∗ (A)⊗N ⊗ N CC∗ (A) → N CC∗ (A), (2.39)
which can be schematically represented as in figure 2. This action descends to cohomology, which we call
the N -fold 2-pointed cap product. Precompose with the ring map Ψ : HH ∗ (A) → 2 HH ∗ (A), we obtain an
action \
N : HH ∗ (A)⊗N ⊗ N HH∗ (A) → N HH∗ (A), (2.40)
which we call the N -fold cap product.
It is straightforward from the definition that the N -fold 2-pointed cap product (and hence the N -fold cap
product) gives a unital graded associative algebra action on the chain level, i.e.
Y Y Y
(ϕ1 ◦ ϕ′1 , · · · , ϕN ◦ ϕ′N ), x = (−1)† N ((ϕ′1 , · · · , ϕ′N ), x)
N (ϕ1 , · · · , ϕN ), N (2.41)
Y
N (id, id, · · · , id), x = x, (2.42)
14
PN
where † = i=1 |ϕ′i | · (|ϕi+1 | + · · · + |ϕN |) is the Koszul sign of reordering (ϕ1 , · · · , ϕN , ϕ′1 , · · · , ϕ′N ) into
(ϕ1 , ϕ′1 , · · · , ϕN , ϕ′N ). In addition, we make the following observation: let Z/N act on 2 CC ∗ (A, A)⊗N by
cyclically permuting the tensor product Q (with appropriate Koszul signs) and take the diagonal Z/N -action
on 2 CC ∗ (A)⊗N ⊗ N CC∗ (A), then N is Z/N -equivariant.
Let X be a cohomologically graded complex with a Z/p-actions. The associated negative Z/p-equivariant
complex, or Z/p homotopy fixed point, is defined as X Z/p := Rhomk[Z/p] (k, X). An explicit chain complex
computing X Z/p is given by X[[t, θ]], where |t| = 2, |θ| = 1, θ2 = 0 Let τ ∈ Z/p be the standard generator,
then the differential is given by
(
deq (x) = dx + (−1)|x| (τ x − x),
(2.43)
deq (xθ) = dx θ + (−1)|x| (x + τ x + · · · + τ N −1 x)t.
Suppose we have two Z/p-complexes X, Y , then X ⊗ Y is also a Z/p-complex equipped with the diago-
nal action. There is a canonical map
LXY : X Z/p ⊗ Y Z/p → (X ⊗ Y )Z/p (2.44)
given explicitly by (writing X Z/p ⊗ Y Z/p explicitly as X ⊗ Y [[t1 , t2 , θ1 , θ2 ]])
Let X = p CC(A) for some A∞ -category A. We shorthand p CC Z/p (A) as just CC Z/p (A), and denote its
homology by HH Z/p (A). Combining (2.44) with (2.39), we obtain a chain map
Y Z/p
: 2 CC ∗ (A)⊗p
Z/p Z/p
⊗ CC∗ (A) → CC∗ (A), (2.46)
eq
L(X⊗Y )Z
(X ⊗ Y )Z/p ⊗ Z Z/p (X ⊗ Y ⊗ Z)Z/p
is homotopy commutative,
15
• unital by (2.42) and that the unit in HZ/p
∗ ∗ ⊗p
2 HH (A) has a chain representative id ⊗ id ⊗ · · · ⊗ id.
Lemma 2.10. [SW, Lemma 2.5] Let X be a cochain complex. Taking a cocycle x ∈ X to its p-th power
x⊗p ∈ (X ⊗p )Z/p gives a well defined map
H ∗ (X) → HZ/p
∗
(X ⊗p ), (2.49)
which is Frobenius p-linear and additive in the first entry after multiplying by t.
QZ/p
Definition 2.11. is called the Z/p-equivariant 2-pointed cap product.
Precomposing with the unital algebra map Ψ : HH ∗ (A) → 2 HH ∗ (A), we obtain a unital multiplicative
Frobenius linear action
Z/N
\
: HH ∗ (A) ⊗ HH∗ (A) → HH∗ (A),
Z/p Z/p
(2.51)
which is Frobenius p-linear and additive in the first entry after multiplying by t.
TZ/p
Definition 2.12. is called the Z/p-equivariant cap product.
TZ/p
The above arguments show that satisfies the conditions C1)-C4) on page 2.
3.1 Monotonicity
Definition 3.1. A symplectic manifold (X, ω) is monotone if
ω = 2τ c1 ,
[ω] = λµ : H2 (X, L) → Z
for some nonzero constant λ (automatically λ = τ , see [Oh, Rmk 2.3.]), where µ is the Maslov class, and that
either H 1 (X) = 0 or the image of π1 (L) in π1 (X) is trivial. One important consequence of monotonicity is
that for holomorphic disks/spheres of a fixed index, then there is a uniform bound on their energy, cf. [Oh,
Proposition 2.7]. This allows us to appeal to Gromov compactness without using Novikov coefficients.
16
3.2 The monotone Fukaya category
In this subsection, we review the definition of the monotone Fukaya category associated with a closed
monotone symplectic manifold (X, ω), over an arbitrary ring R. More precisely, to each λ ∈ R, one can
associate a k-linear Z/2-graded A∞ -category Fuk(X, R)λ .
Let L be an oriented spin monotone Lagrangian submanifold L ⊂ X equipped with a R∗ -local system.
Recall that monotonicity means that µ(L) = [ω] when considered as classes in H 2 (X, L), where µ denotes
the Maslov class. Orientability implies that the minimal Maslov number is ≥ 2. By an abuse of notation,
we denote this datum simply by its underlying Lagrangian L. Let J denote the space of compatible almost
complex structures and H := C ∞ (X, R) the space of Hamiltonians. For each L, we fix JL ∈ J . For each pair
(L0 , L1 ), we fix Jt ∈ C ∞ ([0, 1], J ) and Ht ∈ C ∞ ([0, 1], H) such that Jt = JLt when t = 0, 1. If the R∗ local
systems on both Lagrangians are trivial, the morphism space CF ∗ (L0 , L1 ) is the free R-module generated
by time-1 Hamiltonian chords of Ht from L0 to L1 ; in general, it is the direct sum of hom spaces between
the fibers of the local systems at the startpoint and endpoint of the chord.
Fix Lagrangians L0 , L1 . By standard transversality arguments, for generic almost complex structure JL0 , JL1
and one parameter family (Ht , Jt ), t ∈ [0, 1] such that J0 = JL0 , J1 = JL1 :
R1) The moduli space M1 (L0 ) of Maslov index 2 J-holomorphic disks with one boundary marked point
and boundary on L0 is regular.
R2) The moduli space M1 (Jt ) of pairs (t, u), where t ∈ [0, 1] and t is a Chern number 1 Jt -holomorphic
sphere with one marked point, is regular.
R3) For any time 1 Hamiltonian chord γ : [0, 1] → X starting on L0 and ending on L1 , the map
More generally, ev∗ is weighted by the monodromy of the local system around the boundary of the disc, and
w(L) defines an element of k.
The Floer differential µ1 : CF ∗ (L0 , L1 ) → CF ∗ (L0 , L1 ) is defined as follows. Let x− , x+ ∈ CF ∗ (L0 , L1 ),
then the coefficient of x− in µ1 (x+ ) is the signed count (weighted by monodromy) of rigid elements of
M(x− , x+ )/R, where M(x− , x+ ) is the moduli space of u : R × [0, 1] → X such that
∂s u + Jt (∂t u − XHt ) = 0
u(s, 0) ∈ L0 , u(s, 1) ∈ L1 (3.4)
lims→±∞ u(s, ·) = x± .
By Gromov compactness and monotonicity, when M(x− , x+ ) is one dimensional (i.e. the Maslov index of t
is 1), the space M(x− , x+ )/R is compact. When M(x− , x+ ) is 2-dimensional, its Gromov compactification
consists of broken strips u1 , u2 , each with Maslov index 1, as well as a Maslov index 2 disk bubbling off
17
a Maslov index 0 (hence constant in s) strip. For generic Jt , sphere bubbling cannot occur by regularity
assumption (R3). Therefore, we have
where Hζ , Jζ are the chosen Hamiltonian and almost complex structure for the pair of Lagrangians
meeting at ζ. We also require J = JL when restricted to a boundary component labeled L. The pair
(K, J) is called a perturbation datum.
The higher A∞ -operations of Fuk(X, R)λ are governed by the Delign-Mumford moduli space of disks with
boundary marked points. Let Rd+1 be the moduli space of disks with one boundary output and d boundary
d+1
inputs. It admits a compactification to a manifold with corners R given by
d+1 a
R = RT , (3.7)
T
where T ranges over all planar stable d-leafed trees and RT := R|v| .
Q
v∈Ve(T )
We make a consistent choice of labeled Floer data for Rd+1 , d ≥ 2, meaning it is compatible with the product
′
of Floer data of lower dimensional Rd ’s near a boundary stratum, see [Sei1, section (9g),(9i)].
The higher operations µd , d ≥ 2 are then defined by counting isolated elements of the parametrized moduli
space M(y1 , · · · , yd ; y− ), which is the space of (r, u), r ∈ Rd+1 , u : Sr → X satisfying
(du − YK )0,1
J = 0, (3.8)
with appropriate Lagrangian boundary and asymptotic conditions, where YK is the one-form on S with value
the Hamiltonian vector field associated to K. For a generic choice of Floer data, this moduli space is regular
([Sei1, section (9k)]).
18
moduli space of maps u : C = CP 1 → X with three marked points z0 , z1 , z∞ ∈ C such that t lies in class
A, satisfy Floer’s equation (du − YK )0,1J = 0, where (K, J) denotes a perturbation datum, and z0 , z1 , z∞
are constrained on W u (x0 ), W u (x1 ) and W s (x∞ ), respectively. Equivalently, it is the moduli space of J-
holomorphic u : CP 1 → X together with gradient half-flowlines
satisfying
yk′ = ∇f (yk ), yk (0) = u(zk ), lim yk (s) = xk ,
s→−∞
′
y∞ = ∇f (y∞ ), y∞ (0) = u(z∞ ), lim y∞ (s) = x∞ . (3.10)
s→∞
For generic choice of (K, J), the moduli space MA (C, x0 , x1 , x∞ ) is regular of dimension 2c1 (A) + |x∞ | −
|x0 | − |x1 |. We define the coefficient of x∞ in the quantum cup product of x0 with x1 by counting isolated
elements of MA (C, x0 , x1 , x∞ ), over all A ∈ H2 (X, Z). For a fixed c1 (A), by monotonicity and Gromov
compactness, there are only finitely many homology class A admitting J-holomorphic curves. Thus there is
a well defined map ⋆ : CM ∗ (f ) ⊗ CM ∗ (f ) → CM ∗ (f ), which is easily seen to descend to cohomology, and
will be called the quantum cup product.
Quantum Steenrod operations. Fix a field k of odd characteristic. To define the quantum Steenrod
operations, following [SW, section 4a], we consider a moduli problem with fixed domain but parametrized
(equivariant) Floer data. The relevant Floer data will be parametrized by
For a prime p, there is a Z/p-action on S ∞ where the standard generator τ ∈ Z/p acts by
Let MA (∆i × C, x0 , x1 , · · · , xp , x∞ ) denote the moduli space of pairs (w, u), where w ∈ ∆i and u : C → X
in class A satisfying
(du − YKw )0,1
Jw = 0 (3.18)
and incidence conditions to W u (x0 ), · · · , W u (xp ), W s (x∞ ) as before. We have
MA (τ j (∆i ) × C, x0 , x1 , · · · , xp , x∞ ) ∼
= MA (∆i × C, xp−j+1 , · · · , xp , x1 , · · · , xp−j , x∞ ) (3.20)
19
given by (w, u) 7→ (τ −j (w), u ◦ σ −j ). This defines maps, for i ≥ 0,
Fix a Morse cocycle b ∈ CM ∗ (f ), and let CM (f )[[t, θ]] be the Z/p-equivariant complex of CM ∗ (f ) (where
Z/p acts trivially). One can combine the maps in (3.21) into a chain map
given by X
x 7→ (−1)|b||x| ΣA (∆2k , x, b, · · · , b) + (−1)|b|+|x| ΣA (∆2k+1 , x, b, · · · , b)θ tk . (3.23)
k
Up to homotopy, (3.23) only depends on the cohomology class [b] ∈ QH ∗ (X, k), cf. [SW, Lemma 4.4.].
Finally, summing over A (which is well defined by monotonicity) and extending (t, θ)-linearly, we obtain a
chain map
Σb : CM ∗ (f )[[t, θ]] → CM ∗ (f )[[t, θ]] (3.24)
of degree p|b|. We denote the cohomology level map of Σb as QΣb . Then, QΣ defines a Frobenius p-linear
action of QH ∗ (X, k) on QHZ/p
∗
(X, k) such that
p(p−1)
|b||b′ |
QΣb ◦ QΣb′ = (−1) 2 QΣb⋆b′ , (3.25)
cf. [SW, Proposition 4.8]. (3.25) is sometimes called the Quantum Cartan relation.
be the moduli space of disks with one interior output marked point yout and k1 + · · · + kp + p boundary input
marked points z 1 , z11 , · · · , zk11 , z 2 , z12 , · · · , zk22 , · · · , z p , z1p , · · · , zkpp in counterclockwise order such that up to
automorphisms of the disk, yout , z 1 , z 2 , · · · , z p lie at 0, ζ, ζ 2 , · · · , ζ p , where ζ = e2πi/p . A representative of an
element in R1k1 ,··· ,kp is called standard when yout , z 1 , · · · , z p satisfy the previous constraints. z 1 , · · · , z p are
called the distinguished inputs. We fix the orientation on (4.2) induced by −dz11 ∧· · ·∧dzk11 ∧· · ·∧dz1p ∧· · ·∧dzkpp
on a standard representative.
20
Figure 3: An element of R13,1,3
R1k1 ,··· ,kp has a Deligne-Mumford compactification given by the following. Let Tp denote equivalence classes
of rooted planar trees with k1 + · · · + kp + p − 1 for some k1 , · · · , kp ≥ 0 leaves, with one internal vertex
marked as main and p − 1 leaves marked as distinguished. Those p − 1 leaves, together with the root, are
called the p distinguished semi-infinite edges. Moreover, we require that no two distinguished semi-infinite
edges are adjacent to the same non-main vertex and each non-main vertex is adjacent to ≥ 3 edges. In
particular, one has |main| ≥ p. Then,
1 G
Rk1 ,··· ,kp = RT , (4.3)
T ∈Tp
where Y
RT := R1kT ,··· ,kT × R|v| . (4.4)
1 p
v∈T \{main}
The kiT ’s
are defined as follows: among the edges adjacent to the main vertex, there are p special ones that
are contained in the respective paths from the main vertex to the p distinguished semi-infinite edges. kiT is
defined as the number of edges in between the i-th and i + 1-th special edge in counter-clockwise order.
1
In particular, the codimension 1 boundary of Rk1 ,··· ,kp is covered by images of
′
Rki +1 × R1k1 ,··· ,ki −ki′ +1,··· ,kp , 1 ≤ i ≤ p, (4.5)
′ ′
Rki−1 +ki +2 × R1k1 ,··· ,ki−1 −ki−1
′ ,ki −ki′ ,··· ,kp , 1 ≤ i ≤ p. (4.6)
under the natural inclusions. In (4.5), the output of the first disk can be glued to any one of the inputs of
the main component labeled by z1i , · · · , zki i −k′ +1 , whereas in (4.6), the output of the first disk is glued to the
i
distinguished input z i , and the ki + 1-th input of the first disk will become the new distinguished i-th input
after gluing.
For each Lagrangian labeling of the universal families Sk11 ,··· ,kp → R1k1 ,··· ,kp , k1 , · · · , kp ≥ 0, we choose a
smoothly varying Floer data. Moreover, the Floer data are required to be consistent in the sense that over
each boundary strata of R1k1 ,··· ,kp that decomposes as a product, it agrees (up to conformal equivalence) with
the product of Floer data on the lower dimensional strata (it is understood that we have fixed consistent
labeled Floer data for the Rd ’s).
21
We briefly discuss the existence of such a consistent choice, even though this is a standard argument, cf.
[Sei1, Section (9g),(9i)]. For each Lagrangian labeling L, there is a fiber bundle
whose fiber over r ∈ (R1k1 ,··· ,kp )L := R1k1 ,··· ,kp is the space FSLr of labeled Floer datum on Sr . Note that FSLr
is contractible. A choice of Floer datum for (R1k1 ,··· ,kp )L is just a section of (4.7). The existence argument
is done inductively as follows.
Suppose we have chosen consistent labeled Floer data for R1k1 ,··· ,kp for all k1 + · · · + kp < N . Fix some
k1 + · · · + kp = N . Then for each codimension 1 boundary component of R1k1 ,··· ,kp (corresponding to a tree
type with one interior node), one defines a section of F L (R1k1 ,··· ,kp ) → (R1k1 ,··· ,kp )L in a collar neighborhood
of that boundary component by gluing at the node. Do this for all codimension 1 boundary components,
then by the inductive hypothesis, these local sections agree whenever their domain of definition overlaps
(which is a neighborhood of some codimension 2 corner). Therefore, we obtained a section of (4.7) in a collar
neighborhood of the entire codimension 1 boundary. Since the fibers of (4.7) are contractible, we can extend
this section to all of (R1k1 ,··· ,kp )L . By construction, these inductive choices of Floer data are consistent.
1
Fix a consistent choice of Floer data for Rk1 ,··· ,kp . Let yout ∈ CM ∗ (f ) be a Morse cochain and
be the moduli space of (r, u), where r ∈ R1k1 ,··· ,kp and u : Sr → X satisfies Floer’s equation (3.6) with
appropriate Lagrangian boundary conditions and asymptotic conditions specified by x, and that the interior
marked point is constrained at W u (yout ). For generically chosen Floer data, one can ensure that the above
moduli space is regular of dimension
When ind(u) = n − |yout | − k1 − · · · − kp , counting rigid elements in M(R1k1 ,··· ,kp , yout , x) gives rise to a map
22
4.2 The Z/pZ-equivariant open-closed map
There is Z/p-action on a
R1k1 ,··· ,kp (4.11)
L,k1 ,··· ,kp
such that for r ∈ R1k1 ,··· ,kp , the standard generator τ ∈ Z/p acts by rotating the standard representative of
r by e2πi/p , and the Lagrangian labels are rotated accordingly. It is clear that this action uniquely extends
` 1
to the compactification L,k1 ,··· ,kp Rk1 ,··· ,kp . Let σr : Sr → Sτ (r) be the rotation map, where Sr , Sτ (r) are
standard representatives of r, τ (r), respectively. Then, there is a Z/p-action on the set of consistent choices
1
of Floer data for Rk1 ,··· ,kp , denoted Fp OC , where the standard generator τ ∈ Z/p acts by
ϵτ (w),r = ϵw,τ (r) ◦ σr , Jτ (w),r = Jw,τ (r) ◦ σr , Kτ (w),r = σr∗ Kw,τ (r) , (4.13)
where w ∈ S ∞ .
Proposition 4.2. There exists an S ∞ -dependent consistent choice of Z/p-equivariant Floer data for p OC.
1
Proof. The argument is by induction over the spaces Rk1 ,··· ,kp and cells ∆i ⊂ S ∞ . We first fix a consistent
1
choice of Floer data for the spaces Rk1 ,··· ,kp constructed in section 4.1, and designate that to be the choice
1 1
of Floer data for Rk1 ,··· ,kp × ∆0 . By Z/p-equivariance, we define the Floer data for Rk1 ,··· ,kp × τ ∆0 to be
1
pulled back from Rk1 ,··· ,kp × ∆0 via the action of τ on Fp OC , cf. (4.13).
Suppose we have chosen Floer data that is consistent with boundary decomposition and Z/p-equivariant for
1
all Rk1′ ,··· ,kp′ × τ j ∆i′ with (k1′ , · · · , kp′ , i′ ) < (k1 , · · · , kp , i) and 0 ≤ j ≤ p − 1 if i′ is odd and 0 ≤ j ≤ 1 if
1
i′ is even. Now we inductively determine the Floer data for Rk1 ,··· ,kp × τ j ∆i where 0 ≤ j ≤ 1 if i is even
and 0 ≤ j ≤ p − 1 if i is odd. We discuss the case when i = 2k + 1 is odd, as the even case is completely
analogous.
1
The boundary of Rk1 ,··· ,kp × ∆2k+1 is covered by
ki′ +1 1
(R × Rk1 ,··· ,ki −ki′ +1,··· ,kp ) × ∆2k+1 , 1 ≤ i ≤ p, (4.14)
′
ki−1 +ki′ +2 1
(R × Rk1 ,··· ,ki−1 −ki−1
′ ,ki −ki′ ,··· ,kp ) × ∆2k+1 , 1 ≤ i ≤ p. (4.15)
and
1 1
Rk1 ,··· ,kp × ∆2k , Rk1 ,··· ,kp × τ ∆2k . (4.16)
By induction hypothesis, we have chosen Floer data for (4.14)-(4.16). Fix a small ϵ and apply gluing near the
codimension 1 boundary using the prescribed strip-like ends to (4.14) and (4.15), we obtain a choice Floer
1 1
data for Uϵ (∂Rk1 ,··· ,kp ) × ∆2k+1 , where Uϵ (∂Rk1 ,··· ,kp ) denotes an ϵ-collar neighborhood of the boundary.
We then extend the Floer data on
1 1 1
Uϵ (∂Rk1 ,··· ,kp ) × ∆2k+1 ∪ Rk1 ,··· ,kp × ∆2k ∪ Rk1 ,··· ,kp × τ ∆2k (4.17)
1
smoothly to the entire Rk1 ,··· ,kp × ∆2k+1 ; this can be done since the space of Floer data on Sr for each
1
r ∈ R1k1 ,··· ,kp is contractible. Then, we define the Floer data on Rk1 ,··· ,kp × τ j ∆2k+1 , 0 ≤ j ≤ p − 1 to be
1
pulled back from Rk1 ,··· ,kp ×∆2k+1 via the action of τ j on Fp OC . Note that this does not cause inconsistencies:
1 1
for i < j, the intersection of Rk1 ,··· ,kp × τ i ∆2k+1 with Rk1 ,··· ,kp × τ j ∆2k+1 is empty unless i = j − 1, in which
23
1 1
case it is Rk1 ,··· ,kp × τ j−1 ∆2k . Moreover, the restriction of the Floer data on Rk1 ,··· ,kp × τ l ∆2k+1 , l = j − 1, j
1
to Rk1 ,··· ,kp × τ j−1 ∆2k agree, and are both equivalent to the inductively chosen Floer data on this stratum.
This essentially uses the feature that the Z/p-action on S ∞ is free.
Let yout ∈ CM ∗ (f ) and
x = {x1 , x11 , · · · , x1k1 , · · · , xp , xp1 , · · · , xpkp } ∈ p CC∗ (F).
We define
M(∆i × R1k1 ,··· ,kp , yout , x) (4.18)
to be the moduli space of (w, r, u), where w ∈ ∆i , r ∈ R1k1 ,··· ,kp and u : Sr → X satisfies
By definition, p OC 0 = p OC.
The codimension one boundary strata of the Gromov compactification M(∆i × R1k1 ,··· ,kp , yout , x) consists
of contributions coming from domain breaking (4.5) and (4.6), the corresponding moduli space where the
parameter w is constrained to ∂∆i , semi-stable strip breaking, and Morse trajectory breaking. Note that
there is a canonical identification
∼
=
M(τ j (∆i ) × R1k1 ,··· ,kp , yout , x) −
→ M(∆i × R1kp−j+1 ,··· ,kp ,k1 ,··· ,kj , yout , τ j (x)) (4.23)
coming from the Z/p-equivariance of Floer data. Combining all of the above, and using formula (3.13) and
(3.14), we deduce that for i ≥ 0:
(
i−1
i i p OC ◦ (τ − 1), when i is odd
dCM ∗ (f ) ◦ p OC − p OC ◦ dp CC∗ = i−1
(4.24)
p OC ◦ (1 + τ + · + τ p−1 ), when i is even.
We define p OC −1 = 0 so that the above relation is satisfied for i = 0 (since p OC 0 = p OC is a chain map by
Proposition 4.1).
X p−2
X
2k ∥x∥ 2k+1
(x + τ (x) + · + τ j (x))t tk ,
xθ 7→ p OC (x)θ + (−1) p OC (4.27)
k j=0
and extended t-linearly.
Proposition 4.4. OC Z/p is a chain map.
Proof. This is an immediate consequence of (4.24).
24
4.3 The closed-open maps
We briefly review two versions of the closed-open string maps, cf. [Gan1, section 5.4-5.6],[RS, section 5.5].
Let
Rd1,1 (4.28)
be the moduli space of disks with
• d+1 boundary marked points z0− , z1 , · · · , zd labeled counterclockwise such that z0− is marked as output
and z1 , · · · , zd are marked as input,
• one interior marked point yin marked as input.
After choosing consistent regular Floer data for (4.28), counting rigid solutions to the Floer equation with
the usual conditions gives rise to a chain map
CO : CM ∗ (f ) → CC ∗ (F) (4.29)
(2 CO(b)⊗p ,−)
QZ/p
Σb
OC Z/p
CM ∗ (f )[[t, θ]]
Z/p
p CC∗ (F)
is homotopy commutative.
TZ/p
Proof of Theorem 1.3 given Theorem 4.6. By definition of we have
Z/p
\ Y
(CO(b), −) = (Ψ(CO(b))⊗p , −). (4.32)
eq
But Lemma 4.5 implies that Ψ(CO(b)) and 2 CO(b) are homologous in 2 CC ∗ (F), and Lemma 2.10 further
implies that Ψ(CO(b))⊗p and 2 CO(b)⊗p are homologous in (2 CC ∗ (F)⊗p )Z/p . This proves Theorem 1.3.
25
The rest of section 4.4 will be devoted to the proof of Theorem 4.6. We first introduce a new parameter
space of disks that will be used in constructing the homotopy in Theorem 4.6. Let
26
Figure 4: An element of Q1p,k1 ,··· ,kp degenerates as r → 0 or r → 1
We a priori fix consistent Floer data for the parameter spaces Rd+1 , R1,1
r,s as well as consistent Z/p-equivariant
(S ∞ -dependent) Floer data for R1k1 ,··· ,kp and Sp+1
1
.
A choice of S ∞ -dependent Floer data for the spaces Q1p,k1 ,··· ,kp , k1 , · · · , kp ≥ 0, denoted (ϵw,r , Kw,r , Jw,r ), w ∈
1
S ∞ , r ∈ Qp,k1 ,··· ,kp , is called consistent and Z/p-equivariant if:
• It is Z/p-equivariant, i.e.
ϵτ (w),r = ϵw,τ (r) ◦ σr , Jτ (w),r = Jw,τ (r) ◦ σr , Kτ (w),r = σr∗ Kw,τ (r) , (4.38)
1
where τ denotes the action of the standard generator of Z/p on Qp,k1 ,··· ,kp .
• When the underlying domain r approaches a boundary strata of type (4.35)-(4.37), the Floer data
(ϵw,r , Kw,r , Jw,r ), w ∈ S ∞ decomposes as a product of (ϵw,r′ , Kw,r′ , Jw,r′ ), w ∈ S ∞ , r′ ∈ Q (of lower
dimension) and the pre-chosen non-equivariant Floer data on the other components.
• When the underlying domain r approaches a boundary strata of type (4.34), we require that the Floer
data (ϵw,r , Kw,r , Jw,r ), w ∈ S ∞ decomposes as the product of (ϵw,r′ , Kw,r′ , Jw,r′ ), r′ ∈ R1p,k1 ,··· ,kp and
1
(ϵw,C , Kw,C , Jw,C ), C ∈ Sp+1 (note the S ∞ -parameter on both components agree with w).
The following proposition follows a similar inductive argument as in Proposition 4.2.
27
1
Proposition 4.7. An S ∞ -dependent, consistent and Z/p-equivariant Floer data on Qk1 ,··· ,kp , k1 , · · · , kp ≥ 0
exists.
satisfying
0,1 0,1
(du1 − YKσ1 (w),Sr )Jσ1 (w),Sr = 0, (du2 − YKσ2 (w),C )Jσ2 (w),C = 0,
appropriate Lagrangian boundary conditions for Sr ,
boundary marked points of Sr are asymptotic to x, (4.41)
the points z∞ , z1 , · · · , zp on C are constrained to W u (y∞ ), W s (y1 ), · · · , W s (yp ),
u1 (yout ) = u2 (z0 ).
be the corresponding operation defined by counting rigid solutions of (4.39). By linearity, we can define
(Σ♯p OC)σ for σ any k-coefficient chain in S ∞ × S ∞ . Similarly, we define
satisfying (all except the last condition are the same as (4.41))
(du1 − YKσ1 (w),Sr )0,1
Jσ1 (w),Sr = 0, (du2 − YKσ2 (w),C )J0,1
σ2 (w),C
= 0,
appropriate Lagrangian boundary conditions for Sr ,
boundary marked points of Sr are asymptotic to x, (4.45)
the points z∞ , z1 , · · · , zp on C are constrained to W u (y∞ ), W s (y1 ), · · · , W s (yp ),
u1 (yout ) is constrained to W u (y), u2 (z0 ) is constrained to W s (y).
28
be the moduli space of
w ∈ ∆i ⊂ S ∞ , r ∈ Q1p,k1 ,··· ,kp , u : Sr → X
satisfying
0,1
(du − YKw,Sr )Jw,Sr = 0,
appropriate Lagrangian boundary conditions for S ,
r
(4.48)
boundary marked points of Sr are asymptotic to x,
interior marked points yout , y1 , · · · , yp are constrained to W u (aout ), W s (a1 ), · · · , W s (ap ).
Fixing a generic, Z/p-equivariant and consistent choice of Floer data, (4.47) is regular of dimension
Let
Hi : CM ∗ (f )⊗p ⊗ p CC∗ (F) → p CC∗ (F) (4.50)
be the operation of degree n − i − 1 (mod 2) obtained by counting rigid elements of (4.47). If b ∈ CM ∗ (f )
is a Morse cocycle, we also write
p times
z }| {
Hbi i
:= H (b, · · · , b, −) : p CC∗ (F) → p CC∗ (F), (4.51)
M(∆i × Q1p,kp−j+1 ,··· ,kp ,k1 ,··· ,kj , yout , b, · · · , b, τ j (x)). (4.53)
As usual, we can put the Hbi together into a t-linear map Hb : CC∗ (F) → CM ∗ (f )[[t, θ]] by
Z/p Z/p
(
x 7→ k Hb2k (x) + (−1)∥x∥ Hb2k+1 (x)θ tk
P
(4.54)
xθ 7→ k Hb2k (x)θ + (−1)∥x∥ Hb2k+1 ((1 − τ )p−2 x)t tk .
P
Z/p
Note that since we are in characteristic p, 1 + τ + · · · + τ p−1 = (τ − 1)p−1 . (4.52) implies that Hb provides
a homotopy between
Z/p
Y
OC Z/p ◦ (2 CO(b)⊗p ⊗ −) (4.55)
29
Step 2. The second homotopy involves a standard trick of inserting a finite length Morse trajectory (and
letting its length go to infinity), cf. [PSS].
Z/p Z/p
The goal is to construct a homotopy between (Σ♯p OC)δ(∆),b and (Σ ⋆ p OC)δ(∆),b .
Fix Morse cocycle b, Morse cochain y∞ and p-fold Hochschild chain x of type (k1 , · · · , kp ) as before. Let
satisfying
γ̇ = −∇f,
(du1 − YKw,Sr )0,1 (du2 − YKw,C )0,1
Jw,Sr = 0, Jw,C = 0,
appropriate Lagrangian boundary conditions for S ,
r
(4.58)
boundary marked points of Sr are asymptotic to x,
the points z∞ , z1 , · · · , zp on C are constrained to W u (y∞ ), W s (b), · · · , W s (b),
u1 (yout ) = γ(0), u2 (z0 ) = γ(T ).
Then, as T → 0, we obtain the nodal moduli space defining (Σ♯p OC)δ(∆i ),b and as T → ∞, the finite
Morse trajectory breaks into two semi-infinite Morse trajectories meeting at some critical point, giving rise
30
to (Σ ⋆ p OC)δ(∆i ),b . Putting these together into an equivariant operation (cf. (4.54)), we obtain a homotopy
Z/p Z/p
between (Σ♯p OC)δ(∆),b and (Σ ⋆ p OC)δ(∆),b .
Z/p
Step 3. Finally we show that (Σ ⋆ p OC)δ(∆),b is homotopic to Σb ◦ OC Z/p , hence completing the proof
of Theorem 4.6. The basic idea is elementary topology: one would like to decompose the diagonal cell
δ(∆i ) ⊂ S ∞ × S ∞ , which parametrizes the Floer data for the operation (Σ ⋆ p OC)δ(∆),b , into product cells,
Z/p
For our purpose, a chain level enhancement of Lemma 4.8 is needed, for the reason that a general element
x ∈ p CC∗ (F) is not Z/p-invariant and thus a cohomology class in S ∞ /Z/p × S ∞ /Z/p does not give rise to
a well-defined operation. The chain level refinement of Lemma 4.8 is the following.
Lemma 4.9. Fix a ground field k of characteristic p. There exists a sequence of k-coefficient chains Ci of
dimension i in S ∞ × S ∞ , i ≥ 0, satisfying the recursive property that ∂Ci+1 =
X X
δ(∆i ) − ∆i1 × ∆i2 − ∆i1 × (τ − 1)∆i2 − (τ × τ − 1)Ci , if i is odd,
i1 +i2 =i i1 +i2 =i
X i1Xodd X
δ(∆i ) − ∆ i 1
× ∆ i 2
− τ k ∆i1 × τ j ∆i2 − (1 + τ × τ + · · · + (τ × τ )p−1 )Ci , if i is even,
i1 +i2 =i i1 +i2 =i 0≤k<j≤p−1
ik even i1 odd
(4.60)
where τ × τ denotes the diagonal action τ × τ (w1 , w2 ) := (τ (w1 ), τ (w2 )).
The Proof of Lemma 4.9 will be given in Appendix D.
Proof of Theorem 4.6 given Lemma 4.9. Let
be the operation obtained by counting rigid elements of the moduli space (cf. (4.39))
Then, by considering the boundary strata of the 1-dimensional component of (4.62), and using (4.60), we
obtain that
(Σ ⋆ p OC)Ci+1 ,b ◦ dp CC ∗ ± dCM ∗ (f ) ◦ (Σ ⋆ p OC)Ci+1 ,b = (Σ ⋆ p OC)δ(∆i ),b ±
X X
(Σ ⋆ p OC)∆i1 ×∆i2 ,b ± (Σ ⋆ p OC)∆i1 ×(τ −1)∆i2 ,b ± (Σ ⋆ p OC)(τ ×τ −1)Ci ,b , i odd
i1 +i2 =i
i1 +i2 =i
X i1Xodd X
(Σ ⋆ p OC) ∆ i1 ×∆ i2 ,b ± (Σ ⋆ p OC)τ k ∆i1 ×τ j ∆i2 ,b ± (Σ ⋆ p OC)(1+τ ×τ +···+(τ ×τ )p−1 )Ci ,b , i even.
i1 +i2 =i
i1 +i2 =i 0≤k<j≤p−1
ik even i1 odd
(4.63)
Since
(Σ ⋆ p OC)∆i1 ×∆i2 ,b = Σib2 ◦ p OC i1 (4.64)
and the Floer data are Z/p-equivariant, we can simplify (4.63) as
31
X
Σib2 ◦ p OC i1 ± (Σ ⋆ p OC)Ci ,b ◦ (τ − 1), i odd
i1 +i2 =i
X X
Σbi2 ◦ p OC i1 ± Σib2 ◦ p OC i1 ◦ (τ − 1)p−2 ± (Σ ⋆ p OC)Ci ,b ◦ (1 + τ + · · · + τ p−1 ), i even.
i1 +i2 =i
i1 +i2 =i
ik even i1 odd
(4.65)
Z/p Z/p ∗
We let H̃bi := (Σ ⋆ p OC)Ci+1 ,b (notice the shift in i) and define : H̃b p CC∗ (F) → CM (f )[[t, θ]] by
(
x 7→ k H̃b2k (x) + (−1)∥x∥ H̃b2k+1 (x)θ tk
P
(4.66)
xθ 7→ k H̃b2k (x)θ + (−1)∥x∥ H̃b2k+1 ((τ − 1)p−2 x)t tk .
P
By (4.65),
Z/p Z/p Z/p
H̃b ◦ dp CC Z/p ± dCM ∗ (f ) ◦ H̃b ± (Σ ⋆ p OC)δ(∆),b ± Σb ◦ p OC Z/p (4.67)
∗
with x 7→ (τ − 1)p−2 x. Since p > 2, Lemma 2.9 implies that (τ − 1)p−2 is nullhomotopic as an endomorphism
Z/p
of CC∗ (F). Therefore, (4.68) is nullhomotopic and the proof is complete.
Remark 4.10. When p = 2, the analogue of (4.60) is
X X
∂Ci+1 = δ(∆i ) + ∆i1 × ∆i2 + ∆i1 × (1 + τ )∆i2 + (1 + τ × τ )Ci .
i1 +i2 =i i1 +i2 =i
i1 odd
A celebrated result of [Smi] (cf. Theorem 1.1 loc.cit.) shows that there is a quasi-equivalence F : T wπ Fuk(X, C)0 ≃
T wπ Fuk(Σ2 , C), where T wπ denotes first taking twisted complexes and then idempotent completion. Since
applying T wπ does not change the Hochschild invariants, by abuse of notation, we also denote by F the
induced isomorphism on Hochschild (co)homology/cyclic homology.
32
In this example, we take R = Z[ 12 ]. Condition A1) and A2) follow from the computations in Lemma 5.4
below. Moreover, condition A3) follows from [Smi, Corollary 4.16]. We remark that even though loc.cit. uses
complex coefficients, the results therein regarding generation and the open-closed map being an isomorphism
(specifically Corollary 3.11, Lemma 4.15 and Corollary 4.16 in loc.cit.) hold more generally over R. An
interesting question is whether Smith’s equivalence holds over any field of characteristic not 2.
Conjecture 5.1. Over a field K of characteristic not 2, there is an A∞ -quasi-equivalence
F : T wπ Fuk(X)0 ≃ T wπ Fuk(Σ2 ). (5.2)
Unfortunately, Conjecture 5.1 cannot hold at this level of generality. In fact, we will show in section 5.1 that
Smith’s equivalence imposes the following simple arithmetic property on K.
Proposition 5.2. If Conjecture 5.1 holds, then K must contain a square root of −1.
Now we go to positive characteristics. Fix an odd prime p and a field k of characteristic p. Given
√ Proposition
5.2, it is natural to ask whether Smith’s quasi-equivalence holds over any k that contains a −1. While we
do not obtain such a result in this paper, we show in section 5.2 that Smith’s equivalence over k determines
the quantum Steenrod powers of all odd cohomology classes of X.
√
Proposition 5.3. Let k be a field of odd characteristic that contains a −1. Assuming Conjecture 5.1 holds
over k, then over a field k of odd characteristic p,
p−1 p − 1 p−1 1 2 81 4
QStX (η(γi )) = (−1) 2 ( !)t 2 (1 + t + t + · · · )η(γi ). (5.3)
2 256 262144
p−1 p−1
We remark that the t 2 -coefficient of QStX (η(γi )) in (5.3) is (−1) 2 ( p−1
2 !), which implies that QΞX =
p−1
(−1) 2 id when k = Fp , p = 1 mod 4, cf. [Sei3, Definition 1.8]. This recovers the computation of QΞX in
[Sei3, Example 9.8].
33
−1
Lemma 5.5. Suppose Conjecture 5.1 holds over K. We denote Ξ = COΣ 2
◦ F ◦ COX and Θ = OCΣ2 ◦
−1
F ◦ OCX , where OCX and COX are understood to be restricted to QH (X, K)0 and HH ∗ (Fuk(X, K)0 ),
∗
defined by ch(L) := [idL ]. Where [idL ] denotes the cohomological unit of L viewed as a length 0 Hochschild
homology class. Shklyarov showed that (cf. Proposition 4.4 loc.cit.)
where χ(L, L′ ) = i (−1)i dim HF i (L, L′ ) denotes the Euler pairing and ⟨−, −⟩Sh denotes the Shklyarov
P
pairing on Hochschild homology. We say that an object L has integral Chern character if OC(ch(L)) ∈
im(H 3 (X, Z) → H 3 (X, K)).
Lemma 5.6. 1) Any object L ∈ Fuk(X, K)0 has integral Chern character.
34
2) Assume we are in the situation of Lemma 5.5. Let {γ1 , γ2 , γ3 , γ4 } be a standard symplectic basis of
H1 (Σ2 , Z) with respect to the intersection pairing. Then with respect to the basis {η(γi )}4i=1 of H 3 (X, K)
and {PD(γi )}4i=1 of H 1 (Σ2 , K),
Θ|H 3 (X,K) : H 3 (X, K) → H 1 (Σ2 , K) (5.13)
has integral coefficients.
3) Assume we are in the situation of Lemma 5.5. There exists an autoequivalence G ∈ Auteq(Fuk(Σ2 , K))
such that if we replace F by G ◦ F , then
I 0
Θ= 2 (5.14)
0 −I2
with respect to the basis {η(γi )}4i=1 and {PD(γi )}4i=1 .
Proof. 1) Take an integral basis {γi }4i=1 of H1 (Σ2 , Z). These give rise to vanishing cycles Vγi ⊂ X, 1 ≤ i ≤ 4,
cf. [Smi, Section 4.2]. As Vγi bounds no Maslov index 2 disk, OC(ch(Vγi )) is the image of PD([Vγi ]) under
H 3 (X, Z) → H 3 (X, K). By (5.12), and the fact that OC intertwines the Shklyarov pairing with the Poincare
pairing up to a sign, we conclude that for any L ∈ Fuk(X, K)0 , the Poincare pairing of OC(ch(L)) with
PD([Vγi ]) is integral, for any 1 ≤ i ≤ 4. Since [Vγi ] form a basis for H3 (X, Z), OC(ch(L)) must be integral.
2) By [AS, Lemma 2.19], any object in Fuk(Σ2 , K) has integral Chern character. In particular, for 1 ≤ i ≤ 4,
OCΣ2 (HH∗ (F )(chVγi )) = OCΣ2 (chF (Vγi ) ) has integral coefficients when expanded with respect to the basis
{PD(γi )}4i=1 . Since OCX (ch(Vγi )) = PD([Vγi ]), 1 ≤ i ≤ 4 form an integral basis of H 3 (X), 2) follows.
n(n+1)
3) Since OC intertwines pairings up to (−1) 2 (where n is the complex dimension), we conclude that with
respect to the basis {η(γi )}4i=1 and {PD(γi )}4i=1 , the matrix form of Θ is in Sp−
4 (Z), the set of anti-symplectic
matrices. Therefore, to prove 3), it suffices to show that the natural map
≃
Auteq(Fuk(Σ2 , K)) → Aut(HH0 (Fuk(Σ2 , K)), ⟨−, −⟩Sh ) −−→ Aut(H 1 (Σ2 , K), ⟨−, −⟩Σ2 ) = Sp4 (K) (5.15)
OC
surjects onto the image of Sp4 (Z) → Sp4 (K). This can be seen by the following argument.
Let Γ(Σ2 ) denote the symplectic mapping class group of Σ2 . Then, there is a natural map
Γ(Σ2 ) → Aut(H 1 (Σ2 , K), ⟨−, −⟩Σ2 ) = Sp4 (K) (5.16)
whose image is Sp4 (Z); this is because Sp4 (Z) is generated by algebraic Dehn twists, which can be lifted to
symplectic Dehn twists along simple closed curves. Moreover, (5.16) factors as a homomorphism (depending
on a choice of balancing, cf. [AS, Section 2.6]) Γ(Σ2 ) → Auteq(Fuk(Σ2 , K)) composed with (5.15). Hence
the image of (5.15) must also contain Sp4 (Z).
At this point, we make a few observations. Firstly, since Σ2 has no genus 0 Gromov-Witten invariants, the
quantum Steenrod operations is classical:
p − 1 p−1 (1)
QΣΣ2 Σ2
y (c) = St (y) ∪ c = ( !)t 2 y ∪ c, (5.21)
2
35
for y ∈ H 1 (Σ2 , k). Here, (−)(1) denotes the relative Frobenius along H 1 (Σ2 , Fp ) → H 1 (Σ2 , k). Concretely,
expanding y in terms of a k-basis coming from H 1 (Σ2 , Fp ), then y (1) is obtained by raising each coefficient
to the p-th power. Secondly, by Theorem 1.4, we have
1 1
OCX ◦ F −1 ◦ (OCΣ2 )−1 = OCX
S
◦ F −1 ◦ (OCΣS2 )−1 (⊗k[[t]] k[[t, θ]]).
Z/p Z/p
(5.22)
−1 p − 1 p−1 X
QΣΣ cn Ξ(η(γi ))(1) tn .
Z/p Z/p
Ξ(η(γi )) (OCΣ2 ◦ F ◦ (OCX ) pr0 (1)) = ( !)t 2 (5.23)
2
2
n≥0
Finally, we observe that the t-connection restricted to the summand η(H1 (Σ2 ))[[t]] is trivial (as both c1 ⋆ and
S1 1
µ are trivial when applied to the image of η). As a consequence, OCX ◦ F −1 ◦ (OCΣS2 )−1 |H 1 (Σ2 )[[t]] = Θ−1 [[t]]
because a regular connection over Z has no nontrivial automorphism whose constant term is the identity.
By Lemma 5.6 2), Θ(η(γi )) ∈ im(H 1 (Σ2 , Fp ) → H 1 (Σ2 , k)). In particular, Θ(1) = Θ since the p-th power of
an element in Fp is itself. Therefore, QStX (η(γi )) is equal to
Therefore, to finish the computation, it suffices to determine cn . These coefficients can be extracted from the
so-called R-matrix, cf. Lemma 5.7; it turns out the R-matrix will also depend on the sign ϵ, which cancels
out the ϵ in (5.24).
Lemma 5.7. [Hug, Lemma B.1] Any isomorphism φ : QH ∗ (X, R) ∼ = R ⊕ QH ∗ (Σ2 , R) ⊕ R that intertwines
the operations c1 (X)⋆ and (8, c1 (Σ2 )⋆, −8) can be uniquely extended (meaning is the t = 0 term of a unique
k[[t]]-linear map) to an isomorphism
8 −8
φ̃ : QH ∗ (X, R)[[t]] ∼
= E t ⊕ QH ∗ (Σ2 , R)[[t]] ⊕ E t (5.25)
that intertwines the t-connections. If one chooses basis for both sides of (5.25), the matrix form of φ̃ is called
the R-matrix with constant term φ.
We remark that 1) [Hug, Lemma B.1] was originally proved over C, but the computations involved make
clear they work over any coefficient ring with 2 inverted and 2) it suffices to determine φ̃ on the even degree
part of QH ∗ (X, R)[[t]], this is because on the odd degree part η(H1 (Σ2 ))[[t]], the t-connection is trivial.
Consider the isomorphism
36
of R ⊕ QH even (Σ2 , R) ⊕ R, the inverse of (5.26) is given by the matrix
1 0 0 0
0 −4 0 0 √
1
−1ϵ. (5.29)
0 0 0
2
0 0 0 1
Since the cyclic open-closed map preserves connection, by the uniqueness property of R-matrix in Lemma
S1 1
5.7, one concludes that when restricted to the even degree part, OCX ◦ F −1 ◦ (OCΣS2 )−1 agrees with the
R-matrix whose constant term is (5.29). The next lemma follows from using the computation in [Hug,
Appendix B.1].
Lemma 5.8. With notations as above, the R matrix with constant term (5.29) is given by
7 1 1
15 7 9 1
1 0 0 0 − 64 2 32 0 − 8192 − 128 2048 1024
0 −4 √ 1 0 − 512 5 1 √ 27 7 27 √
0 0 −1ϵ − 161 16 −1ϵt + − 2048 128 0 2048 −1ϵt2
1 1 1 3 1
0 0
2 0 − 8 −2 0 8
32 0 − 512 32
1 1 7 1 7 9 15
0 0 0 1 0 2 − 32 64 1024 128 2048 − 8192
389 135 393 65
38421
3069 11907 3537
− 524288 − 16384 262144 65536 − 134217728 − 1048576 16777216 8388608
33 435 33 √ 7533 999 7533 √
0 − 1048576 8192 −1ϵt3 + − 4194304
0 4194304 −1ϵt4 ±· · ·
− 8192
− 9 3 9 9
262144
63 9
2048 128 0 2048
8192 0 − 524288 8192
65 135 393 389 3537 3069 11907 38421
− 65536 − 16384 − 262144 524288 8388608 1048576 16777216 − 134217728
(5.30)
Proof of Proposition 5.3. By (5.24), it suffices to compute the coefficients cn , which we obtain via the
following steps. First, express 1 in terms of our chosen basis (5.27) (viewed as a basis of QH even (X, R)[[t]]
over k[[t]])
1 −h6 + 4h4 − 3h2 + 4 1 h3 + 4h2
1= · +0·w+1·v+ · . (5.31)
8 4 8 16
Then, apply the R-matrix of Lemma 5.6 to the resulting vector ( 18 , 0, 1, 18 ). Next, apply the matrix
0 1 0 0
, (5.32)
0 0 1 0
in terms of the basis {−2H, 1} of QH even (Σ2 , R)[[t]], and n≥0 cn tn is simply the coefficient in front of 1.
P
Unfolding the computations we obtain Proposition 5.3.
Things to note about the coefficients in (5.3):
1) All the denominators are powers of 2, hence the expression in fact makes sense over Z[ 12 ].
2) For fixed p, all high enough coefficients have numerators divisible by p; in principle, one can compute
exactly how large that is. Therefore when reduced mod p, this gives a polynomial expression in t.
3) If we add back the Novikov variable q (which we have set to 1 throughout the paper) of degree 2, then
for degree reasons (since QStX (η(γi )) has total degree 3p) (5.3) involve no powers of q greater than p.
Geometrically, this says that p-fold cover curves do not contribute to QStX (η(γi )).
37
6 The unramified exponential type conjecture for quantum con-
nection
In this section, we apply the Fukaya categorical interpretation of quantum Steenrod operations to give
an arithmetic proof of the unramified exponential type conjecture for X satisfying Abouzaid’s generation
criterion.
First, we recall some basic definitions of formal connections and describe the setup of this theorem. Let K
be a field, and M be a finite dimensional K((t))-vector space equipped with a connection ∇ d : M → M .
dt
Definition 6.1. (M, ∇) has regular singularity at t = 0 if there exists an K((t))-basis of M under which
the connection matrix of ∇ d has at most a simple pole.
dt
Definition 6.2. (M, ∇) has unramified exponential type if, up to gauge transformation, there is a finite
direct sum decomposition M −λ
∇d = E t2 ⊗ ∇regλ (6.1)
dt
λ
λ
where each ∇reg
λ has regular singularity at t = 0 and E − t2 denotes the one-dimensional connection with
d
quadratic pole ((K(t)), dt − tλ2 ). Moreover, we say (M, ∇) has quasi-unipotent regularized monodromy if the
reg
monodromy of ∇λ around t = 0 is quasi-unipotent (i.e. eigenvalues are roots of unity).
Conjecture 6.3 (The unramified exponential type conjecture). For a closed monotone symplectic manifold
X, the quantum t-connection (QH ∗ (X, C)((t)), ∇QH
d ) of (1.11) has unramified exponential type and quasi-
dt
unipotent regularized monodromy at t = 0.
The main theorem of this section is the following.
Theorem 6.4. Let X be a closed monotone symplectic manifold. Suppose R ⊂ K is the ring of integers
of some number field (up to inverting finitely many elements) which satisfies the assumptions A1)-A3) of
section 1.5. Then, the quantum connection (QH ∗ (X, K)((t)), ∇QHd ) has unramified exponential type and
dt
quasi-unipotent regularized monodromy at t = 0.
In particular, after making the base-change − ⊗K((t)) C((t)), Theorem 6.4 implies Conjecture 6.3 for any
closed monotone symplectic manifold X such that there exists such an R ⊂ K satisfying A1)-A3). In fact,
we show that the existence of such an R is ensured if one assumes Abouzaid’s generation criterion (i.e. OC
hits the unit of QH ∗ (X)) over Q:
Proof of Theorem 1.2 given Theorem 6.4. First, we note that there always exists R ⊂ K ⊂ Q satisfying
condition A1) and A2). Namely, we take the number field K generated by the eigenvalues of c1 ⋆, and let
R be its ring of integers. Up to inverting finitely many elements in R, we can assume that c1 ⋆ admits a
Jordan decomposition over R in the sense of A1). We furthermore invert all the differences among distinct
eigenvalues of c1 ⋆, which guarantees A2).
Hence, it remains to prove that there exists a further finitely generated enlargement of R over which
Abouzaid’s generation criterion for X is satisfied, which by results of [Gan1] implies A3). By assump-
tion, OC/Q hits 1 ∈ QH ∗ (X, Q). We first show that there exists some number field K such that OC/K hits
1 ∈ QH ∗ (X, K). The idea is that, by definition, Abouzaid’s generation criterion is detectable at finitely many
objects. That is, for each λ ∈ spec(c1L ⋆), there exists a subcategory Fλ ⊂ Fuk(X, Q)λ with finitely many
objects, such that OC/Q restricted to λ∈spec(c1 ⋆) HH∗ (Fλ ) still hits 1. Write each object as (L, E) ∈ Fλ ,
∗
where L denotes the underlying Lagrangian (with orientation and spin structure) and E a Q local system
on L. Fixing a basis for H1 (L, Z), we may express the monodromy representation of E as a finite sequence of
∗
elements in Q . We take K to be the number field generated by these elements, ranging over all (L, E) ∈ Fλ
and λ ∈ spec(c1 ⋆). Then, each (L, E) can be viewed as an object of Fuk(X, K)λ , and it is clear that OC/K
also hit 1.
Finally, we show that OC/K hits 1 implies that there exists some R = OK [ α1 ], α ∈ K ∗ such that OC/R
hits 1. The argument is analogous to the previous paragraph. In this case, we let α′ be the product of all
38
the numerators and denominators appearing in the monodromy representation of E, as (L, E) ranges over
Fλ , λ ∈ spec(c1 ⋆). In particular, the (L, E)’s are well-defined objects of Fuk(X, R[ α1′ ])λ . Hence, OC/R[ α1′ ]
hits 1 after base-changing to K, and thus must hit 1 after inverting finitely many elements of R[ α1′ ] (because
the input of OC/R[ α1′ ], which is a Hochschild chain, only involves finitely many denominators).
Remark 6.5. 1) We remark that [PS]’s proof of the unramified exponential type conjecture assumes the
existence of a smooth anticanonical divisor whose complement is Weinstein, cf. Assumption 1.2.1 in
loc.cit. Furthermore, in the process of passing to the complement of the divisor, their proof also re-
quires certain results, cf. Conjecture 1.2.3, Conjecture 1.2.4 in loc.cit., regarding the Borman-Sheridan
deformation of (equivariant) symplectic cohomology. The proof in our paper does not require those
assumptions: in fact, we will solely work with the monotone Fukaya category/quantum cohomology of
X.
2) However, as a trade-off to 1), there is a different assumption required in our Theorem 1.2, namely
Abouzaid’s generation criterion for X over Q. The reason we need this assumption is that our proof
relies essentially on arguments at the level of the Fukaya category. More precisely, the key tool we
use in the proof of Theorem 6.4 is Corollary 1.10, whose proof is categorical and assumes that the
open-closed map is an isomorphism. However, we note that the statement of Corollary 1.10 is purely
closed-string, and it would be interesting to see if it has a closed-string proof.
We now reduce the proof of Theorem 6.4 to a computation of the p-curvature of the quantum connection,
using Katz’s local monodromy theorem, cf. Theorem A.1. Let R ⊂ K be as in the assumption of Theorem
6.4. Recall from Lemma 1.7, up to a gauge transformation (which only involves formal power series), we
have a decomposition of free R((t))-modules with connections
M
∇QHd = ∇QH,λ
d , (6.2)
dt dt
λ∈spec(c1 ⋆)
where
d λI + Nλ X
∇QH,λ
d = − 2
+ Am tm (6.3)
dt dt t
m≥−1
λ
where E − t2 = (R((t)), dt
d
− λ
t2 ) and
˜ QH,λ d Nλ X
∇ d = − 2 + Am t m , (6.5)
dt dt t
m≥−1
which we call the residual connection. By definition of unramified exponential type, Theorem 6.4 is an
immediate consequence of the following proposition.
Proposition 6.6. In the same setting as Theorem 6.4, for each λ ∈ spec(c1 ⋆), the residual connection
˜ QH,λ
(QH ∗ (X, R)((t))λ , ∇ d ) of (6.5) has regular singularity at t = 0 and quasi-unipotent monodromy after
dt
based-changed to K((t)).
By Theorem A.1, Proposition 6.6 would follow if for all p ∈ mSpec(R), the p-curvature of ∇ ˜ QH ⊗R((t)) κ(p)((t))
λ
is nilpotent, or equivalently (since the p-curvature of ∇QH,λ
t 2 d and ˜ QH,λ
∇ t 2 d differ by −λp
):
dt dt
Proposition 6.7. In the same setting as Theorem 6.4, for each λ ∈ spec(c1 ⋆) and p ∈ mSpec(R), the sum
of the p-curvature (where p is the characteristic of κ(p)) of ∇QH
λ
d
⊗R[[t]] κ(p)[[t]] along the vector field t2 dt
d X
Ftλ2 d := (∇QH,λ
t2 d
)p = (t2 − (λI + Nλ ) + Am tm+2 )p . (6.6)
dt dt dt
m≥−1
with λp is nilpotent.
39
The proof of Proposition 6.7 goes in two steps. Step 1) is a result that relates the p-curvature of ∇QH d to the
t2 dt
quantum Steenrod operation QΣc1 . This result was inspired by an observation of Jae Hee Lee in the case of
symplectic resolutions [Lee], and its proof in the closed monotone case (cf. Lemma 6.8) is communicated to
us by Paul Seidel. Step 2) uses Corollary 1.10 to show that QΣc1 − λp acts nilpotently on the λ-summand
(1.25).
Lemma 6.8 (Seidel). Let X be a closed monotone symplectic manifold. Then over any field k of odd
characteristic p,
QΣc1 + FtQH
2 d (6.7)
dt
is nilpotent, where
QH QH p
Ftq d := (∇
tq d
) − tp−1 ∇QH
tq d
(6.9)
dq dq dq
d
is the p-curvature of the quantum q-connection along the ‘vector field’ tq dq . The proof is an application of
Seidel-Wilkins’ covariant constancy for quantum Steenrod operations, [SW, Theorem 1.4].
QH
First we note that the classical (q = 0) terms of both QΣc1 ,q and Ftq d are the classical Steenrod operation
dq
(−tp−1 c1 + c∪p QH
1 )∪, and since both commute with ∇tq d (the former is a consequence of [SW, Theorem 1.4]
dq
and the latter is obvious as the p-curvature is just a combination of the connection), their difference consists
of terms involving q p or higher powers of q. Moreover, their non-equivariant (t = θ = 0) terms are both c⋆p1 ⋆.
So their difference consists of terms involving tq p -terms or higher. Since the total degree of both QΣc1 and
QH QH
Ftq d is 2p, QΣc1 − F
tq d
decreases the degree in H ∗ (X) by at least 2, so it is nilpotent.
dq dq
is nilpotent.
λ λ
P
To show that (6.12) is nilpotent, we write c1 = λ∈spec(c1 ⋆) c1 , where c1 is the projection of c1 onto
QH ∗ (X, κ(p))π(λ) . Similarly write e =
P
λ∈spec(c1 ⋆) eλ , where eλ is the projection of the unit e onto
QH ∗ (X, κ(p))π(λ) . First we note that cλ1 − λeλ is a nilpotent element of QH ∗ (X, κ(p))π(λ) ⊂ QH ∗ (X, κ(p)).
Indeed, by definition, there exists an integer k such that (cλ1 − λeλ )⋆k ⋆ = 0 on QH ∗ (X, κ(p))π(λ) , and thus
(cλ1 − λeλ )⋆k = (cλ1 − λeλ )⋆k ⋆ e = (cλ1 − λeλ )⋆k ⋆ eλ = 0. (6.13)
40
Since QΣ is a Frobenius-linear algebra action of QH ∗ (X, κ(p)) on QH ∗ (X, κ(p))[[t, θ]], we conclude that
By Corollary 1.10,
(6.14)
(QΣc1 − λp QΣe )k |QH ∗ (X,κ(p))[[t,θ]]π(λ) = (QΣcλ1 − λp QΣeλ )k |QH ∗ (X,κ(p)[[t,θ]]π(λ) = 0. (6.15)
Theorem 6.11. In the same setting of Theorem 6.4, if we further assume X satisfies Conjecture 6.10 and
˜ QH of (6.5) becomes regular
that c1 ⋆ is semi-simple, then for each λ ∈ spec(c1 ⋆), the residual connection ∇ λ
ν
after base-changing along t 7→ t for some positive integer ν.
Proof. Let p ∈ mSpec(R), and p the characteristic of the finite field κ(p). Since c1 ⋆ is semi-simple, we have
X
c1 = λeλ . (6.17)
λ∈spec(c1 ⋆)
˜ QH of (6.5) is zero. Since this is true for all p, the result follows from
or equivalently, the p-curvature of ∇ λ
Katz’s finite local monodromy theorem (cf. Theorem A.4).
Remark 6.12. We remark that Conjecture 6.10 is known to hold in the following special cases.
1) The proof of Lemma 6.8 shows that Conjecture 6.10 holds in cases where covariant constancy of QΣ
with respect to the quantum q-connection determines QΣ. A notable class of examples is any closed
monotone X such that H 2 (X) generates its quantum cohomology ring, cf. [SW, Proposition 6.4].
41
2) When c1 ⋆ has simple spectrum (i.e. all eigenvalues have multiplicity 1), so does FtQH
2 d , cf. [Lee,
dt
and 1
(Ω∗A [[t]], td − df ∧) ≃ CC∗S (MF(A, f )), (7.2)
where MF denotes the Z/2-graded dg category of matrix factorizations. See [CT, Theorem 1.8, Corollary
1.13] for flavors of these results when f has isolated singularities.
1 1 Z/p
Conjecture 7.1. Under the identification CC∗S ⊕ CC∗S θ ≃ CC∗ of Theorem 1.4 1), the HKR-type
quasi-isomorphisms (7.1) and (7.2) intertwine the following two Frobenius p-linear graded multiplicative
actions:
: HH ∗ (MF(A, f )) ⊗ HH∗ (MF(A, f )) → HH∗ (MF(A, f ))
TZ/p Z/p Z/p
1) The Z/p-equivariant cap product
2) An action (after applying ⊗k[[t]] k[[t, θ]])
∗
^
H ∗ (( T A, ιdf )) ⊗ H ∗ (Ω∗A [[t]], td − df ∧) → H ∗ (Ω∗A [[t]], td + df ∧) (7.3)
where
• ‘twisted functions’ f ∈ A/ιdf (T A) act as multiplication by f p ;
V2
• ‘twisted vector fields’ D ∈ ker(ιdf : T A → A)/ιdf ( T A) act as
[p] p−1
iD := (ιDp − Lp−1
D ιD )t
2 . (7.4)
42
concentrated in degree 0, where it is just the twisted functions k[[z]]/N z n−1 . As an algebra, this is generated
by the element z. On the other hand, H ∗ (Ω∗A [[t]], td−df ∧) is concentrated in degree 1 and is freely generated
as a k[[t]]-module by the elements dz, zdz, · · · , z N −2 dz. One can easily calculate the action of z on these
generators (recall that z acts as multiplication by z 3 ), and the result is that at the level of cohomology,
• when 0 ≤ k < N − 4,
[z 3 · z k dz] = [z k+3 dz]; (7.5)
(φ1 , −), φ11 ∈ HH ∗ (AzN ) (the definition of 3 R̃∂0 , · · · , 3 R̃∂N −2 and φ11 is given in (7.21) and (7.28)) is
TZ/3 1
given by the matrix
0 0 IN −4
t 0 0 0
, (7.11)
− 0
N 0 0
t
0 N
where I denotes the identity matrix.
43
In particular, under the identifications φ11 ↔ [z] and 3 R̃∂k ↔ [z N −2−k dz], 0 ≤ k ≤ N − 2, this agrees with
the computation in (7.5) and (7.6).
Hochschild homology. Let CC ∗ (AzN ) denote the normalized Hochschild chain complex of AzN , i.e. the
quotient of CC∗ (AzN ) by all chains of the form x0 | · · · |1| · · · |xn , where the unit 1 is not in the 0-th position.
Here, we have used the shorthand notation a0 |a1 | · · · |an for a0 ⊗ a1 ⊗ · · · ⊗ an . Since AzN is strictly unital,
the homology of CC ∗ (AzN ) computes HH∗ (AzN ). As a Z/2-graded vector space,
where k⟨· · · ⟩ denotes the free k-vector space generated by element(s) in the bracket. The Hochschild differ-
ential is given by
k k k−N +1
z }| { z }| { z }| {
b(1| ∂| · · · |∂) = 0 , b(∂| ∂| · · · |∂) = N · 1| ∂| · · · |∂ . (7.13)
From (7.13) it is straightforward to conclude that
Therefore,
∗ even odd
CC (AzN ) = CC (AzN ) ⊕ CC (AzN ) (7.17)
k k
Y z }| { Y z }| {
= Hom(k⟨∂| · · · |∂⟩, k⟨1⟩) ⊕ Hom(k⟨∂| · · · |∂⟩, k⟨∂⟩). (7.18)
k≥0 k≥0
k
z }| {
Let φk1 denote the even Hochschild cochain that sends ∂| · · · |∂ to 1 and all other generators to 0; similarly,
k
z }| {
let φk∂
denote the odd Hochschild cochain that sends ∂| · · · |∂ to ∂ and all other generators to 0. Then the
Hochschild cochain differential is given by
X X X
ak φk1 7→ 0 , ak φk∂ 7→ (N an+1−N )φn1 . (7.19)
k≥0 k≥0 n≥0
44
and moreover the Hochschild cocycles
φk1 , 0≤k ≤N −2 (7.21)
descend to a set of generators for HH even (AzN ). Finally, we note that under the cup (Yoneda) product,
(φ11 )∪k = φk1 .Thus, HH even (AzN ) is generated by φ11 as an algebra.
Z/p-equivariant Hochschild homology. We demonstrate the computation for p = 3; in this subsection,
k will be a field of characteristic 3. First, we find explicit generators for the 3-fold Hochschild homology of
AzN . Since AzN is unital, there exists an isomorphism
(
∼ 0 , if ∗ = even
3 HH∗ (Az N ) = HH∗ (Az N ) = (7.22)
k⊕N −1 , if ∗ = odd
induced by the quasi-isomorphism Φ03 of (2.34). Therefore, it suffices to find lifts of the chain level generators
(7.15) under Φ03 . For 0 ≤ k < N − 1, we define
k1 k2 k3
X z }| { z }| { z }| {
k
3 R∂ := 1| ∂| · · · |∂ |1| ∂| · · · |∂ |∂| ∂| · · · |∂ ∈ 3 CC ∗ (AzN ). (7.23)
k1 +k2 +k3 =k
In the expression (7.23), the three entries 1, 1, ∂ not under the overbrace are the three bimodule entries.
Since k < N − 1, if we apply the 3-fold Hochschild differential to 3 R∂k , the only terms involve applying an
m2 to an expression that contains a unit 1. Hence, it is easy to see that 3 R∂k , 0 ≤ k < N − 1 are cycles.
Moreover, if (k1 , k2 ) ̸= (0, 0), then
k1 k2 k3
z }| { z }| { z }| {
Φ03 (1| ∂| · · · |∂ |1| ∂| · · · |∂ |∂| ∂| · · · |∂) = 0. (7.24)
Thus,
k k
z }| { z }| {
Φ03 (3 R∂k ) = Φ03 (1|1|∂| ∂| · · · |∂) = ∂| ∂| · · · |∂, (7.25)
and the cycles {3 R∂k }0≤k<N −1 descend to a set of generators for 3 HH∗ (AzN ).
Z/3 Z/3
Now we proceed to find generators for HH∗ (AzN ). By Theorem 1.4 1), we know that abstractly HHeven (AzN ) =
k[[t, θ]]⊕N −1 θ and HHodd (AzN ) = k[[t, θ]]⊕N −1 , so it suffices to find lifts of the generators 3 R∂k , 0 ≤ k < N −1
Z/3
Z/3 t,θ=0
under HHodd (AzN ) −−−→ HHodd (AzN ).
Z/3
Let τ ∈ Z/3 be the standard generator, which acts on CC ∗ (AzN ) via (2.35). Then,
k1 k2 k3 k1 k2 k3
X z }| { z }| { z }| { z }| { z }| { z }| {
(τ − 1)3 R∂k = (∂| ∂| · · · |∂ |1| ∂| · · · |∂ |1| ∂| · · · |∂) − (1| ∂| · · · |∂ |1| ∂| · · · |∂ |∂| ∂| · · · |∂)
k1 +k2 +k3 =k
k1 k2 k3
1 X z }| { z }| { z }| {
3
=b ∂| ∂| · · · |∂ |1| ∂| · · · |∂ |∂| ∂| · · · |∂ (7.26)
N
k1 +k2 +k3 =k+N −1
k1 k2 k3
3
1 X z }| { z }| { z }| {
b − 2 ∂| ∂| · · · |∂ |∂| ∂| · · · |∂ |∂| ∂| · · · |∂ . (7.27)
N
k1 +k2 +k3 =k+2(N −1)
45
Moreover, in (7.27), the term inside the bracket of b3 (· · · ) is Z/3-invariant. In particular, for 0 ≤ k < N − 1
k1 k2 k3
k k 1 X z }| { z }| { z }| {
3 R̃∂ := 3 R∂ + ∂| ∂| · · · |∂ |1| ∂| · · · |∂ |∂| ∂| · · · |∂ θ
N
k1 +k2 +k3 =k+N −1
k1 k2 k3
1 X z }| { z }| { z }| { Z/3
+ 2 ∂| ∂| · · · |∂ |∂| ∂| · · · |∂ |∂| ∂| · · · |∂ t ∈ CC odd (AzN ) (7.28)
N
k1 +k2 +k3 =k+2(N −1)
Z/3
is a cycle in CC ∗ (AzN ) whose constant term is 3 R∂k . As a result, {3 R̃∂k }0≤k<N −1 descends to a set of
Z/3
generators for HH∗ (AzN ) as a free k[[t, θ]]-module.
The Z/p-equivariant cap product. Continuing the above discussion, we would like to determine the
TZ/3 TZ/3
action of Definition 2.12 for the A∞ -algebra AzN , 3 ∤ N . Since is a multiplicative and Frobenius-
linear action, and HH ∗ (AzN ) is generated by φ11 as an algebra, it suffices to compute
TZ/3 1
(φ1 , 3 R̃∂k ), for
0 ≤ k < N − 1.
Proof of Theorem 7.2. First of all, observe that by the A∞ -structure equations (7.7) and φ11 (∂) = 1, on the
chain level:
Z/3 k1 k2 k3 k1 −1 k2 −1 k3 −1
\ z }| { z }| { z }| { z }| { z }| { z }| {
1
(φ1 , x| ∂| · · · |∂ |y| ∂| · · · |∂ |z| ∂| · · · |∂) = x| ∂| · · · |∂ |y| ∂| · · · |∂ |z| ∂| · · · |∂, (7.29)
where we set the term to be 0 if ki − 1 < 0 for some i ∈ {1, 2, 3}. Now we discuss a few cases.
Case I: 3 ≤ k < N − 1. As an immediate consequence of (7.29) and the definition of 3 R̃∂k (7.28), we have
Z/3
\
(φ11 , 3 R̃∂k ) = 3 R̃∂k−3 , 3 ≤ k < N − 1. (7.30)
TZ/3
Case II: k = 0. By (7.28) and (7.29), we compute that (note (φ11 , 3 R∂0 ) = 0)
Z/3 k1 k2 k3
\ 1 X z }| { z }| { z }| {
(φ11 , 3 R̃∂0 ) = ∂| ∂| · · · |∂ |1| ∂| · · · |∂ |∂| ∂| · · · |∂ θ
N
k1 +k2 +k3 =N −4
k1 k2 k3
1 X z }| { z }| { z }| { Z/3
+ 2 ∂| ∂| · · · |∂ |∂| ∂| · · · |∂ |∂| ∂| · · · |∂ t ∈ CC ∗ (AzN ). (7.31)
N
k1 +k2 +k3 =2N −5
where
k1 k2 k3
X z }| { z }| { z }| {
Q1 = − k3 · 1| ∂| · · · |∂ |1| ∂| · · · |∂ |∂| ∂| · · · |∂ . (7.34)
k1 +k2 +k3 =N −3
46
Next, we compute that
k1 k2 k3 k1 k2 k3
X z }| { z }| { z }| { z }| { z }| { z }| {
2
(1 + τ + τ )Q1 t = − k3 · 1| ∂| · · · |∂ |1| ∂| · · · |∂ |∂| ∂| · · · |∂ +k1 · ∂| ∂| · · · |∂ |1| ∂| · · · |∂ |1| ∂| · · · |∂
k1 +k2 +k3 =N −3
k1 k2 k3
z }| { z }| { z }| {
+ k2 · 1| ∂| · · · |∂ |∂| ∂| · · · |∂ |1| ∂| · · · |∂ . (7.35)
where
k1 k2 k3
X z }| { z }| { z }| {
Q2 = − k1 · ∂| ∂| · · · |∂ |1| ∂| · · · |∂ |∂| ∂| · · · |∂ . (7.37)
k1 +k2 +k3 =2N −4
where
k1 k2 k3
X z }| { z }| { z }| {
Q3 = − k2 · 1| ∂| · · · |∂ |∂| ∂| · · · |∂ |∂| ∂| · · · |∂ . (7.39)
k1 +k2 +k3 =2N −4
where
k1 k2 k3
X z }| { z }| { z }| {
Q4 = k2 · ∂| ∂| · · · |∂ |∂| ∂| · · · |∂ |∂| ∂| · · · |∂, (7.42)
k1 +k2 +k3 =3N −5
and
k1 k2 k3
X z }| { z }| { z }| {
2
(1 + τ + τ )Q4 = ∂| ∂| · · · |∂ |∂| ∂| · · · |∂ |∂| ∂| · · · |∂ . (7.43)
k1 +k2 +k3 =3N −5
47
Combining all of the above, we obtain that on the chain level,
Z/3
\
(φ11 , 3 R̃∂0 ) + 3 R̃N −3 t = dCC Z/3 (Q1 θ + (Q2 + Q3 )t − Q4 tθ), (7.44)
48
Theorem A.3 (Fuchs, Turrittin, Lutz). Let (M, ∇) be a free R((t))-module with integrable connection that
contains a cyclic vector v ∈ M . Suppose (M, ∇) ⊗R((t)) K((t)) does not have a regular singularity at t = 0,
then for every multiple a of n! and after base changing along t 7→ ta , there exists an R((t))-basis f of M such
that
∇t d f = t−N (A + tB)f (A.2)
dt
∇t d m = (A + tB)m, (A.3)
dt
where A ∈ Mn (R), B ∈ Mn (R[[t]]). After adjoining to R the eigenvalues of A and inverting finitely many
elements, we may assume without loss of generality that A is in Jordan normal form A = D + N , where D
is diagonal, N is nilpotent and [D, N ] = 0. By assumption, for all p ∈ mSpec(R),
Comparing diagonal terms we deduce that D ∈ Fp ⊂ κ(p) for all p, and hence by Kronecker’s density
theorem, D ∈ R is a rational number. Comparing off-diagonal terms, we obtain that N p − N = 0. Since N
is nilpotent, 1 − N p−1 is invertible and thus in fact N = 0. Thus, (A.3) becomes
∇t d m = (D + tB)m, (A.6)
dt
where D is diagonal with rational eigenvalues. By applying gauge transformations of the form tm , we may
assume without loss of generality that the differences among the eigenvalues of D are non-integers. Then,
we can applying a further gauge transformation to simplify (A.6) into
d
∇t d = t + D. (A.7)
dt dt
Let ν be the least common multiple of the denominators of the eigenvalues of D. After pulling back along
t 7→ tν , (A.7) has a full set of solutions.
B Grading
B.1 Z-grading on the monotone Fukaya category
An R-linear pre-graded A∞ -category A is the data of a collection of objects obA; for each pair of objects
X0 , X1 , an R-module homA (X0 , X1 ); multilinear structure maps
satisfying the A∞ -relations (2.2). A Z-grading on a pre-graded A∞ -category A is a map, for each pair of
objects X0 , X1 ,
Gr : homA (X0 , X1 ) → homA (X0 , X1 ) (B.2)
49
such that
[Gr, µdA ] = (2 − d)µdA . (B.3)
Let X be a closed monotone symplectic manifold, and R a base coefficient ring. Recall from section 3.2
that for each λ ∈ R there is a Z/2-graded A∞ -category Fuk(X, R)λ over R. In this section, following [Hug,
Appendix A.1], we upgrade F uk(X, R)λ to a Z-graded A∞ -category Fuk(X, R[q, q −1 ])λ over R[q, q −1 ], where
q is a formal variable of degree 2.
The objects of Fuk(X, R[q, q −1 ])λ are oriented spin Lagrangian submanifolds L equipped with R∗ -local
systems, such that the Maslov index 2 disk potential (3.3) is λ. For simplicity, for the rest of this section,
we will ignore Hamiltonian perturbations when defining Lagrangian intersections and moduli spaces (and
assume they are already transverse), as they are standard and can be done in the same way as in section
3.2.
Let LX → X denote the oriented Lagrangian Grassmannian bundle over X. Any oriented Lagrangian
submanifold L ⊂ X gives rise to a section sL : L → LX|L defined by x 7→ Tx L. Let L0 , L1 be two objects of
Fuk(X, R[q, q −1 ])λ . For simplicity, assume they are equipped with trivial local systems. For each p ∈ L0 ∩L1 ,
let L(L0 , L1 , p) denote the homotopy classes of paths p̃ : [0, 1] → Lp X with p̃(0) = Tp L0 and p̃(1) = Tp L2 .
Define M
CF ∗ (L0 , L1 ) = R⟨L(L0 , L1 , 0)⟩. (B.4)
p∈L0 ∩L1
Given a generator p̃ ∈ L(L0 , L1 , p), we define its Z-grading to be the Maslov index of p̃ composed with the
canonical short path from Tp L1 to Tp L0 , cf. [Aur, Definition 1.8], which we denote by ind(p̃). In other words,
Gr(p̃) := ind(p̃)p̃. Note that the homotopy class of p̃ is uniquely determined by ind(p̃). The R[q, q −1 ]-module
structure on CF ∗ (L0 , L1 ) is defined by letting q send p̃ to the unique homotopy class of paths with Maslov
index ind(p̃) + 2.
We now define the A∞ -structure maps. Let L0 , · · · , Ld be objects of Fuk(X, R[q, q −1 ])λ with trivial local
systems, and let p̃i ∈ L(Li−1 , Li , pi ) for some pi ∈ Li−1 ∩ Li , i = 1, 2, · · · , d. For p0 ∈ Ld ∩ L1 and a
homotopy class α ∈ π2 (X, L0 ∪ · · · ∪ Ld ), let
M(p0 , · · · , pd ; α) (B.5)
denote the moduli space of pairs (r, u), where r ∈ Rd+1 , u : Sr → X a map of homotopy class α satisfying
du0,1
J = 0 and the following asymptotic and Lagrangian boundary conditions: let D2 \{z0 , · · · , zd } be a
representative of Sr , then we require that t sends zi to pi and the arc ∂i−1 D2 = (zi−1 , zi ) to Li−1 . Then,
the structure maps µdq of Fuk(X, R[q, q −1 ])λ is defined as
X
µdq (p̃1 , · · · , p̃d ) := #M(p0 , · · · , pd ; [u])p̃0 (p̃1 , · · · , p̃d ; [u]), (B.6)
p0 ∈Ld ∩L0
u|ind([u])=2−d
where p̃0 (p̃1 , · · · , p̃d ; [u]) ∈ L(Lk , L0 , p0 ) is given as follows. Fixing a trivialization of u∗ LX, we may view
the concatenation γ = (sL0 )|∂0 D2 ◦ p̃1 ◦ · · · ◦ p̃d ◦ (sLd )|∂d D2 as a path from Tp0 L0 to Tp0 Ld in Lp0 X.
We define p̃0 (p̃1 , · · · , p̃d ; [u]) to be the unique homotopy class of paths from Tp0 Ld to Tp0 L0 such that
p̃0 (p̃1 , · · · , p̃d ; [u]) ◦ γ has Maslov index 0. By definition of the Maslov index, it is immediate that (B.6)
defines an R[q, q −1 ]-multilinear operation of degree 2 − d (i.e. satisfies equation (B.3)). It is also clear from
definition that after restricting to q = 1 (which has the additional effect of collapsing the Z-grading to a
Z/2-grading), we recover the Z/2-graded A∞ -category Fuk(X, R)λ from section 3.2.
There is also a Z-graded R[q, q −1 ]-linear version of quantum cohomology (QH ∗ (X, R[q, q −1 ]), ⋆q ), cf. [SW,
section 3a] for a definition using the Morse chain model CM ∗ (f, R[q, q −1 ]), which is the model we use in
this paper. Recall that the underlying graded vector space of CM ∗ (f, R[q, q −1 ]) is freely generated by the
critical points of f over R[q, q −1 ]. There is a quantum q-connection on QH ∗ (X, R[q, q −1 ])[[t]] given by
d
∇QH
tq d
:= tq + c1 ⋆q . (B.7)
dq dq
50
We now describe a (chain level) Z-graded R[q, q −1 ]-linear open-closed map
that recovers the usual open-closed map when restricted to q = 1. Let L0 , · · · , Ld be objects of Fuk(X, R[q, q −1 ])λ
and p̃i ∈ L(Li−1 , Li , pi ) for some pi ∈ Li−1 ∩ Li . Let yout be a critical point of f . Let R1d+1 be the parameter
space of disks with d + 1 boundary marked points and 1 interior marked point, up to automorphism; let S
be the associated universal curve. For a homotopy class α ∈ π2 (X, L0 ∪ · · · ∪ Ld ), we define
to be the moduli space of pairs r ∈ R1d+1 , u : Sr → X of homotopy class α satisfying (du)0,1 J = 0 and the
following asymptotic, incidence and Lagrangian boundary conditions: let D2 \{z0 , · · · , zd }, y0 ∈ D2 be a
representative of Sr , then we require that u(zi ) = pi , u(∂i D2 ) ⊂ Li and that u(y0 ) ∈ W u (yout ). Define
X ind(p˜0 ,··· ,p̃d )
OCq (p̃0 ⊗ · · · ⊗ p̃d ) := #M(p0 , · · · , pd , yout ; [u])q 2 yout , (B.10)
yout ∈crit(f)
[u]|ind([u])=n−d−|yout |
where ind(p˜0 , · · · , p̃d ) is defined as the Maslov index of p̃0 ◦ (sL0 )|∂0 D2 ◦ p̃1 ◦ · · · ◦ p̃d ◦ (sLd )|∂d D2 , viewed
as a loop in Lp0 X after fixing a trivialization of u∗ LX. Using essentially the same argument as in [Gan1,
Proposition 5.1], taking into account the Maslov index of p̃i , one can show that (B.9) is a Z-graded R[q, q −1 ]-
linear chain map. One can furthermore enhance this into an S 1 -equivariant version, following [Gan2], and
obtain a Z-graded R[q, q −1 ]-linear chain map
1 1
OCqS : CC∗S (Fuk(X, R[q, q −1 ])λ → CM ∗+n (f, R[q, q −1 ])[[t]]. (B.11)
1
Theorem B.1. ([PS, Theorem 6.3.5]) On the level of cohomology, OCqS intertwines the Getzler-Gauss-
Manin q-connection with the quantum q-connection (B.7).
We refer the readers to [Hug, Definition 3.16] for a definition of the Getzler-Gauss-Manin q-connection on
1
HH∗S (Fuk(X, R[q, q −1 ])λ .
d µ c1 ⋆
∇QH
t2 d
= t2 + − 2 , (B.12)
dt dt t t
which has a second order pole at t = 0. In fact, one can obtain the quantum t-connection from the quantum
q-connection, and the process is crucially related to the issue of Z-grading. Define the total degree operator
Deg : QH ∗ (X, R[q, q −1 ])[[t]] → QH ∗ (X, R[q, q −1 ])[[t]] by
d d
Deg := 2(q + t + µ). (B.13)
dq dt
As its name suggests, the effect of Deg applied to an element β ∈ QH ∗ (X, R[q, q −1 ])[[t]] is to multiply β by
its total Z-grading (i.e. combining the grading from H ∗ (X), from q and from t), shifted by n = dimC X (cf.
the definition of µ in (1.10)). Then, by the formulae (B.7) and (B.12), the t-connection is obtained as
1
∇QH
t2 d
= t Deg − ∇QH d .
tq dq
(B.14)
dt 2
To be more precise, (1.10) is taken to be the restriction of (B.14) to q = 1, but we will abuse terminology
and call both the quantum t-connection.
51
Similarly, the Getzler-Gauss-Manin t-connection can be obtained from its q-version via
1
∇GGM
t2 d
:= tGr− − ∇GGMd
tq dq
. (B.15)
dt 2
In (B.15), the total grading operator Gr− is defined as
d
Gr− := LGr + Γ + 2t , (B.16)
dt
where LGr denotes the Lie action of Gr (B.2), viewed as a length 1 Hochschild cochain, on Hochschild chains,
and Γ(x0 ⊗ x1 ⊗ · · · ⊗ xk ) = −kx0 ⊗ x1 ⊗ · · · ⊗ xk is the length operator on Hochschild chains. The effect of
1
Gr− on an element α = x0 ⊗ x1 ⊗ · · · xk tl ∈ CC∗S (Fuk(X, R[q, q −1 ])λ ) is to multiply it by
which is the total Z-grading of α viewed as a cyclic chain (note that k is absorbed into the reduced grading
convention (2.21) on Hochschild chains). As a result of the above discussion, we obtain the following.
1
Corollary B.2. On the level of cohomology, OC S intertwines the quantum t-connection with the Getzler-
Gauss-Manin t-connection.
1
Proof. From the fact that OCqS of (B.11) is Z-graded, i.e. it intertwines the total grading operators Gr− with
Deg, Corollary B.2 immediately follows from Theorem B.1 and formulae (B.14), (B.15) for the t-connections
(after setting q = 1).
For the rest of this section, we work over k a field of odd characteristic p.
Lemma B.3. For any b ∈ H ∗ (X, k), QΣb commutes with ∇QH
t2 d
.
dt
Proof. By [SW, Theorem 1.4], QΣb commutes with the quantum q-connection. Thus, by formula (B.14), it
suffices to show that it commutes with Deg. However, QΣb is an operation of total degree p|b|. As a result,
FtQH QH
2 d = −Ftq d . (B.19)
dt dq
FtQH p QH
2 d = t Ft d
dt dt
= tp ((∇QH
td
)p − ∇QH
td
)
dt dt
1 1
= t (( Deg − ∇QH
p
q dq
p QH
d ) − ( Deg − ∇ d ))
q dq
2 2
1 1
= tp (( Deg)p − (∇QH p QH
d ) − ( Deg − ∇ d ))
q dq q dq
(since ∇QH
q d
has degree 0, it commutes with Deg)
2 2 dq
QH
= −Ftq d , (B.20)
dq
where in the last inequality, ( 12 Deg)p − 12 Deg = 0 because Deg is multiplication by an integer.
52
C An unramified exponential type result for matrix factorizations
Motivated by homological mirror symmetry [Sh1], we consider the following B-side situation. Let Y be an
algebraic variety over Q equipped with a function W : Y → A1 with isolated singularities. One can associate
to the pair (Y, W ) its triangulated category of singularities
Y
Db Sing(Y, W ) = Db Sing(W −1 (λ)). (C.1)
λ∈crit(W)
The decomposition (C.1) should be thought of as the B-side analogue of the fact that there is one monotone
Fukaya category Fuk(X)λ associated to each eigenvalue λ of c1 ⋆, cf. section 3.2. By classical results of Orlov
[Or], there is an exact equivalence of triangulated categories
We further restrict ourselves to the following simplified situation. Let Y = Spec Q[z1 , · · · , zn ] and W : Y →
A1 such that W (0, · · · , 0) = dW (0, · · · , 0) = 0; moreover, assume that (0, · · · , 0) is the only zero of dW .
[Shk2, Theorem 1.1] showed that the Getzler-Gauss-Manin t-connection on the periodic cyclic homology of
MF(Y, W ) is equivalent to the connection
d W Γ′
∇Wd := + 2 + (C.3)
dt dt t t
on H ∗ (Ω(Y )((t)), −dW + td), where Γ′ |Ωq (Y ) = − 2q . Theorem C.1 below is an analogue of Theorem 1.2 in
the context of matrix factorizations. This result is well known, see e.g. [Sab, Theorem 1.1], but we give a
different approach using reduction p methods.
Theorem C.1. The connection ∇Wd of (C.3) has regular singularity and quasi-unipotent monodromy at
dt
t = 0. Moreover, if W is contained in the Jacobian ideal ( ∂W ∂W W
∂z1 , · · · , ∂zn ), then ∇ dt
d is regular after passing
to a finite cover.
Proof. For the first statement, by Theorem A.1, it suffices to show that there exists an R ⊂ Q finitely
generated over Z, over which W is defined and such that when reduced mod each p ∈ mSpec(R), ∇Wd has
dt
nilpotent p-curvature (where p equals the characteristic of the residue field κ(p)).
By Nullstellensatz, our assumption on W implies that over Q there exists a positive integer N such that
∂W ∂W
WN ∈ ( ,··· , ). (C.4)
∂z1 ∂zn
In particular, there exists R ⊂ Q finitely generated over Z, over which W is defined and that (C.4) holds.
Fixing such an R (and assuming 21 ∈ R), the argument goes in two steps.
d
1) For p ∈ mSpec(R), we compute the p-curvature of ∇Wd over κ(p) along the vector field t2 dt . Because
dt
W and Γ′ commute,
d d
FtW
2 d = (t
2
+ W + tΓ′ )p = W p + (t2 + tΓ′ )p . (C.5)
dt dt dt
On the other hand,
d d d
(t2 + tΓ′ )p = tp ((t + Γ′ )p − (t + Γ′ ))
dt dt dt
d d
= t (t + (Γ ) − t − Γ′ )
p ′ p
dt dt
= 0, (C.6)
d p
where in the last equality, we used that (t dt d
) = t dt and (Γ′ )p = Γ′ (since Γ′ always takes half-integer
values) in characteristic p > 2.
53
2) By 1), it suffices to show that multiplication by W p defines a nilpotent operation on H ∗ (Ω(Y )((t)), −dW +
td) . The key observation is that there is a Frobenius p-linear algebra action of ‘twisted functions’
[f ] ∈ κ(p)[z1 , · · · , zn ]/ιdW on ‘twisted de Rham cohomology’ [α] ∈ Ω(κ(p)[z1 , · · · , zn ])((t)), −dW + td)
given by
Actp[f ] ([α]) := [f p · α]. (C.7)
We check this is indeed well-defined. d(f p ) = 0 implies that f p commutes with −dW + td. On the
other hand, if f = ιdW (D) = D(W ) for some vector field D, a straightforward but tedious computation
shows that −D(W )p α + tp−1 Dp (W )α + tp LDp α =
X (p − 1)!2 e(m1 )
[−dW + td, Qs Dm1 (W )e1 · · · Dms (W )es LlD ιD ]α, (C.8)
Ps Ps l! j=1 (mj !)ej (ej !)
j=1 ej = j=1 ej mj +l=p−1
ej >0, l≥0, 0≤m1 <m2 <···<ms
54
Combining (D.1) and (D.2), we have
X X
∂ δ(∆i ) − ∆i1 × ∆i2 − ∆i1 × (τ − 1)∆i2 − (τ × τ − 1)Ci
i1 +i2 =i i1 +i2 =i
i1 odd
X
(τ − 1)∆i1 −1 × ∆i2 + (−1)i1 ∆i1 × (1 + τ + · · · + τ p−1 )∆i2 −1
=−
i1 +i2 =i
i1 odd
X
(1 + τ + · · · + τ p−1 )∆i1 −1 × ∆i2 + (−1)i1 ∆i1 × (τ − 1)∆i2 −1
−
i1 +i2 =i
i1 even
X X X X
− (τ − 1)∆i1 −1 × (τ − 1)∆i2 + (τ × τ − 1) ∆i1 × ∆i2 + τ k ∆i1 × τ j ∆i2
i1 +i2 =i i1 +i2 =i−1 i1 +i2 =i−1 0≤k<j≤p−1
i1 odd ik even i1 odd
X X
=− (−1)i1 ∆i1 × (1 + τ + · · · + τ p−1 )∆i2 −1 − (1 + τ + · · · + τ p−1 )∆i1 −1 × ∆i2
i1 +i2 =i i1 +i2 =i
i1 odd i1 even
X X
+ (τ × τ − 1) τ k ∆i1 × τ j ∆i2
i1 +i2 =i−1 0≤k<j≤p−1
i1 odd
= 0. (D.3)
If i is even, we have
X X X
τ k ∆i1 × τ j ∆i2 − (1 + τ × τ + · · · + τ × τ p−1 )Ci
∂ δ(∆i ) − ∆i1 × ∆i2 −
i1 +i2 =i i1 +i2 =i 0≤k<j≤p−1
ik even i1 odd
55
Combining (D.5) and (D.6), we obtain
X X X
τ k ∆i1 × τ j ∆i2 − (1 + τ × τ + · · · + τ × τ p−1 )Ci
∂ δ(∆i ) − ∆i1 × ∆i2 −
i1 +i2 =i i1 +i2 =i 0≤k<j≤p−1
ik even i1 odd
X X
=(1 + τ × τ + · · · + τ × τ p−1 )
∆i1 × ∆i2 + ∆i1 × (τ − 1)∆i2
i1 +i2 =i−1 i1 +i2 =i−1
i1 odd
X
(1 + τ + · · · + τ p−1 )∆i1 −1 × ∆i2 + (−1)i1 ∆i1 × (1 + τ + · · · + τ p−1 )∆i2 −1
−
i1 +i2 =i
ik even
X X
τ k (τ − 1)∆i1 −1 × τ j ∆i2 + (−1)i1 τ k ∆i1 × τ j (τ − 1)∆i2 −1 .
− (D.7)
i1 +i2 =i 0≤k<j≤p−1
i1 odd
and
X
τ k ∆i1 × τ j (τ − 1)∆i2 = (1 + τ + · · · + τ p−1 )∆i1 × ∆i2 − (1 × τ )(1 + τ × τ + · · · + τ × τ p−1 )(∆i1 × ∆i2 ).
0≤k<j≤p−1
(D.9)
Plugging (D.8) and (D.9) into (D.7) we obtain
X X X
τ k ∆i1 × τ j ∆i2 − (1 + τ × τ + · · · + τ × τ p−1 )Ci = 0 (D.10)
∂ δ(∆i ) − ∆i1 × ∆i2 −
i1 +i2 =i i1 +i2 =i 0≤k<j≤p−1
ik even i1 odd
References
[AS] D. Auroux, I. Smith, Fukaya categories of surfaces, spherical objects, and mapping class groups. Forum
of Mathematics, Sigma , Volume 9 , 2021 , e26.
[AT] L. Amorim, J. Tu, Categorical primitive forms of Calabi–Yau A∞ -categories with semi-simple coho-
mology. Selecta Mathematica (2022).
[Aur] D. Auroux, A beginner’s introduction to Fukaya categories. arXiv: 1301.7056.
[BK] R. Bezrukavnikov, D. Kaledin, Fedosov quantization in positive characteristic. Journal of the A.M.S.,
Volume 21, No.2, 409-438 (2008).
[Che] Z. Chen, On operadic open-closed maps in characteristic p. arXiv: 2402.06183.
[CT] A. Caldararu, J. Tu, Curved A-infinity algebras and Landau-Ginzburg models. New York J. Math. 19
(2013), 305-342.
[Don] S.K. Donaldson, Gluing techniques in the cohomology of moduli spaces. In Topological methods in
modern mathematics, 1993.
56
[Dyc] T. Dyckerhoff, Compact generators in categories of matrix factorizations. Duke Math. J. 159, no. 2
(2011), 223-274.
[EL] J. Evans, Y. Lekili, Noncommutative crepant resolutions of cAn singularities via Fukaya categories.
arXiv: 2307.06592.
[Fuk] K. Fukaya, Morse homotopy and its quantization. Geometric topology (Athens, GA, 1993), pages
409–440. Amer. Math. Soc., 1997.
[Gan1] S. Ganatra, Symplectic cohomology and duality for the wrapped Fukaya category. Ph.D. thesis, arXiv:
1304.7312, 2012.
[Gan2] S. Ganatra, Cyclic homology, S 1 -equivariant Floer cohomology, and Calabi-Yau structures. Preprint,
arXiv:1912.13510, 2019.
[GGI] S. Galkin, Vasily Golyshev, H. Iritani. Gamma classes and quantum cohomology of Fano manifolds:
gamma conjectures. Duke Math. J., 165:2005–2077, 2016.
[GHHPS] S.Ganatra, A. Hanlon, J. Hicks, D. Pomerleano, N. Sheridan, Integrality of mirror maps and
arithmetic homological mirror symmetry for Greene–Plesser mirrors. arXiv: 2312.01949, 2023.
[GPS] S. Ganatra, T. Perutz, N.Sheridan, Mirror symmetry: from categories to curve counts. Preprint,
arXiv: 1510.03839, 2015.
[Hug] K.Hugtenberg, The cyclic open-closed map, u-connections and R-matrices. Sel. Math. New Ser. 30,
29 (2024).
[Ka] N. Katz, Nilpotent connections and the monodromy theorem: applications of a result of Turrittin.
Publications mathématiques de l’I.H.É.S., tome 39 (1970), p. 175-232.
[Ka1] N. Katz, A simple algorithm for cyclic vectors. Am. J. Math, Vol 109, No.1, Feb 1987, pp. 65-70.
[KKP] L. Katzarkov, M. Kontsevich, T. Pantev, Hodge theoretic aspects of mirror symmetry. In From Hodge
theory to integrability and TQFT tt*-geometry, volume 78 of Proc. Sympos. Pure Math., pages 87–174.
Amer. Math. Soc., 2008.
[Lee] J. Lee, quantum Steenrod operations of symplectic resolutions. arXiv:2312.02100
[LP] Y.Lekili, T. Preutz, Arithmetic mirror symmetry for the 2-torus. Select. Math., Volume 23, pages
1851-1907 (2017).
[LPe] Y. Lekili, T. Perutz, Arithmetic mirror symmetry for the 2-torus. arXiv: 1211.4632.
[LPo] Y. Lekili, A. Polishchuk, Arithmetic mirror symmetry for genus 1 curves with n marked points. Selecta
Mathematica (2017), 1851–1907.
[LT] R. Lipshitz, D. Treumann, Noncommutative Hodge-to-de Rham spectral sequence and the Heegaard
Floer homology of double covers. Journal of the E.M.S (2016), Vol.18, No.2, 281-325.
[NS] T. Nikolaus, P. Scholze, On topological cyclic homology. Preprint, arXiv:1707.01799, 2017.
[Oh] Y.G. Oh, Floer cohomology of Lagrangian intersections and pseudo-holomorphic disks I. Comm. Pure
Appl. Math. 46(7), 949–993 (1993).
[Or] D. Orlov, Triangulated categories of singularities and equivalences between Landau–Ginzburg models.
Mat. Sb., 197:12 (2006), 117–132.
[OS] H. Ohta, F. Sanda, Meromorphic connections in filtered A∞ categories. Pure Appl. Math. Q.16(2020),
no.3, 515–556.
[PS] D. Pomerleano, P.Seidel, The quantum connection, Fourier-Laplace transform, and families of A-
infinity-categories. arXiv:2308.13567.
[PSS] S. Piunikhin, D. Salamon, M. Schwarz, Symplectic Floer-Donaldson theory and quantum cohomology.
Proceedings of the Symposium on Symplectic Geometry (1996).
57
[PVV] A. Petrov, D. Vaintrob, V. Vologodsky, The Gauss–Manin connection on the periodic cyclic homology.
Sel. Math. New Ser. 24, 531–561 (2018).
[RS] A.F. Ritter, I. Smith. The monotone wrapped Fukaya category and the open-closed string map. Sel.
Math. New Ser. 23, 533–642 (2017).
[Sab] C. Sabbah, On a twisted de Rham complex II. arXiv: 1012.3818.
[Sh1] N. Sheridan, On the Fukaya category of a Fano hypersurface in projective space. Publications
mathématiques de lÍHES, 2016, 1–153.
[Sh2] N. Sheridan, Homological mirror symmetry for Calabi-Yau hypersurfaces in projective space. Invent.
Math. 199 (2015), no. 1, 1–186.
[Sei1] P. Seidel, Fukaya categories and Picard-Lefschetz theory. Zurich Lectures in Advanced Mathematics,
European Mathematical Society, 2008.
[Sei2] P. Seidel, Fukaya A∞ -structures associated to Lefschetz fibrations. I.. J. Symplectic Geom. 10 (2012),
325–388.
[Sei3] P.Seidel, Formal groups and quantum cohomology. Geom.Topol. 27 (2023) 2937-3060.
[Shel] E. Shelukin, Pseudorotations and Steenrod squares. J. Mod. Dyn. (2020), 289-304.
[Shk] D. Shklyarov, Hirzebruch–Riemann–Roch-type formula for DG algebras. Proceedings of the Lon. Math.
Soc, Volume 106, Issue 1, Jan 2013, 1-32.
[Shk2] D. Shklyarov, Non-commutative Hodge structures: towards matching categorical and geometric ex-
amples. Transactions of the Amer. Math. Soc, vol. 366, no. 6, 2014, pp. 2923–74.
[Smi] I. Smith, Floer cohomology and pencils of quadrics. Invent. Math. 189.1 (2012), pp. 149–250.
[SW] P. Seidel, N. Wilkins, Covariant constancy of quantum Steenrod operations. J. Fixed Point Theory
Appl. 24, 52 (2022).
[VWX] S. Venugopalan, C. Woodward, G. Xu, Fukaya categories of blowups. arXiv: 2006.12264.
[Wil] N. Wilkins, A construction of the quantum Steenrod squares and their algebraic relations. Geom. Topol.
24 (2020), 885-–970.
58