Categorical Non-Properness in Wrapped Floer Theory: Sheel Ganatra
Categorical Non-Properness in Wrapped Floer Theory: Sheel Ganatra
Categorical Non-Properness in Wrapped Floer Theory: Sheel Ganatra
Sheel Ganatra
Abstract. In all known explicit computations on Weinstein manifolds, the self-wrapped Floer
homology of non-compact exact Lagrangian is always either infinite-dimensional or zero. We show
that a global variant of this observed phenomenon holds in broad generality: the wrapped Fukaya
category of any positive-dimensional Weinstein (or non-degenerate Liouville) manifold is always
either non-proper or zero, as is any quotient thereof. Moreover any non-compact connected exact
Lagrangian is always either a “(both left and right) non-proper object” or zero in such a wrapped
Fukaya category, as is any idempotent summand thereof. We also examine criteria under which the
argument persists or breaks if one drops exactness, which is consistent with known computations
of non-exact wrapped Fukaya categories which are smooth, proper, and non-vanishing (e.g., work
of Ritter-Smith).
1. Introduction
The wrapped Fukaya category W(X) is an important invariant of a Liouville (exact symplec-
tic convex at infinity) manifold X, whose objects are (exact, cylindrical at infinity) Lagrangian
submanifolds and whose cohomological morphisms are wrapped Floer homology groups. Unlike the
ordinary Lagrangian Floer cohomology between a pair of Lagrangians, wrapped Floer homology
allows further contributions to the chain complex beyond just (the finitely many transverse) inter-
section points of the Lagrangians, coming from Reeb chords between the Legendrian boundaries at
infinity. The presence of chords in these complexes means that wrapped Floer homology could be
infinite dimensional, but doesn’t force infinite-dimensionality.1 Indeed one of the motivations for
their introduction [AS] was precisely to mirror Ext groups of sheaves on singular or non-compact
varieties, which could be infinite dimensional. This paper is concerned with the question of to what
degree such wrapped Floer homology groups must be infinite dimensional (if they are non-zero).
In recent years, there has been an explosion of computations of wrapped Fukaya categories,
especially in the setting of Weinstein manifolds (those Liouville manifolds for which the Liouville
vector field Z, the symplectic dual of the primitive of the symplectic form is gradient like for a Morse
exhaustion function), where we have strong general control of the category. In all of the many
known computations to date on Weinstein manifolds (to the author’s knowledge), the following
phenomenon occurs: the self-wrapped Floer cohomology of a non-compact exact (cylindrical at
infinity) Lagrangian is either infinite dimensional or zero. An old folk question inquires whether this
could be always the case:
Question 1 (Folk). Let L be a non-compact connected exact Lagrangian with Legendrian ends
in a Liouville manifold X. Must the self-wrapped Floer homology HW ∗ (L, L) be either infinite
dimensional or zero?
Remark 2. Question 1 is very quickly false if one drops the connected hypotheses (take a
compact Lagrangian union a non-compact Lagrangian with zero wrapped Floer homology) When
1e.g., if there are only finitely many chords or the differential involves many cancellations
1
considering the more general question of when HW ∗ (K, L) could be finite non-zero, note that
HW ∗ (K, L) coincides with ordinary Floer cohomology HF ∗ (K, L) (which is finite and could be non-
zero) when one of K or L is compact, or more generally when K and L have Legendrian boundaries
on different components at infinity. However, the groups HW ∗ (K, L) do appear to ‘frequently’ be
infinite for non-compact K, L.
Question 1 is the open-string analogue of (and generalizes by setting L = ∆) an even older and
more well known folk question, about the nature of the often studied symplectic cohomology SH ∗ (X).
We recall that SH ∗ (X), by definition the (Hamiltonian) Floer cohomology of a Hamiltonian with
sufficiently large growth near infinity or the limit of Floer cohomologies of Hamiltonians which
increasingly grow near infinity, is generated (for a nice choice of Hamiltonian) by the co-chains of
X, a subcomplex, along with Reeb orbits on the boundary at infinity of X.
Question 3 (Folk). For a non-compact Liouville manifold X, must SH ∗ (X) be either infinite
dimensional or zero?
There are reasons to believe Questions 1 and 3 are either rather difficult, or maybe have negative
answers. For one, if the answer to Question 3 is yes in general, it would follow that Viterbo’s
“acceleration map” H ∗ (X) → SH ∗ (X) (induced by the inclusion of a subcomplex in the chain
model mentioned above) cannot be an isomorphism (the so-called “algebraic Weinstein conjecture”),
further implying Weinstein’s conjecture on the existence of a Reeb orbit on the boundary-at-infinity
of X for any non-degenerate contact form (which is currently unknown in this generality). Similarly,
an affirmative answer to Question 1 for any L would imply Arnold’s chord conjecture on the existence
of a Reeb self-chord on the Legendrian ∂∞ L. Even in the case of a cotangent bundle where one
can reduce to algebraic topology via the isomorphism SH ∗ (T ∗ Q) ∼ = Hn−∗ (LQ) 6= 0 [V] where LQ
denotes the free loop space2 Question 3, which then asks whether H∗ (LQ) must always be infinite-
dimensional for a closed Q, is to our knowledge unknown for all Q3. In all explicitly computed
examples on Weinstein manifolds (again to our understanding) the alternatives posed in Questions
1 and 3 hold. If one drops exactness however, the alternatives are known to fail, even if one retains
non-compactness of the target (see e.g., [Ri2, RS]).
Nevertheless, our main result is that a more global version of the empirically noticed “infinite or
zero” phenomenon in fact holds in broad generality for structural reasons. To state the main result,
recall from [Gan1] that a Liouville manifold is said to be non-degenerate if there exists a collection
of Lagrangians in the wrapped Fukaya category satisfying Abouzaid’s generation criterion [A]. We
will treat this hypothesis, which has strong implications [Gan1] (reviewed later), as a black box. A
series of works implies every Weinstein manifold is non-degenerate [CDGG, GPS2, Gan1, G].
Main Theorem. Let X be a Weinstein manifold (or more generally a non-degenerate Liouville
manifold in the sense of [Gan1]) of dimension greater than zero. Then
(1) The wrapped Fukaya category W(X) is either non-proper or it is zero. More generally,
(1’) Any quotient of W(X) is non-proper or zero, as is more generally any category admitting
a homological epimorphism from W(X).
(2) Any open exact Lagrangian submanifold L ∈ W(X) (i.e., all components of L are non-
compact) is either a (both right and left) non-proper object or it is zero. More generally,
(2’) Any idempotent summand (in the pre-triangulated split-closure perf (W(X))) of an L as
in (2) is either (both right and left) non-proper or it is zero.
Recall that an A∞ category C is proper over a field K if hom(X, Y ) ∈ perf (K) for all pairs of
objects X, Y ∈ C where, denoting by M od(K) the category of all chain complexes over K, perf (K) ⊂
M od(K) denotes the full subcategory of chain complexes with finite-dimensional total cohomology.
2implicitly using twisted coefficients on one or both sides in case Q is not Spin [K, Rmk. 1.3]
3though by direct computation it is known in many special cases, e.g., when Q is simply connected or when π Q
1
has infinitely many conjugacy classes; compare the discussion in [Ri1, §12.5]
2
An object L ∈ C is right (respectively left) proper if homC (X, L) (respectively homC (L, X)) lands
in perf (K) ⊂ M od(K) for every X ∈ C. Properness of a category, or of a (left or right) module
over a category such as the Yoneda modules homC (L, −) and homC (−, L), is a Morita invariant
notion, inherited by (split-)generating subcategories. Statement (1) of the Main Theorem therefore
implies that if W(X) 6= 0 there exist K, L ∈ W(X) with HW ∗ (K, L) infinite rank, including at least
one such pair in any split-generating collection of objects. The more general (1’) further precludes
phenomena like the existence of a non-trivial proper orthogonal summands of the wrapped Fukaya
category, which could a priori exist on a non-proper category. (2) correspondingly says that for any
open exact L which is not zero, there exist a K, K ′ with HW ∗ (K, L) and HW ∗ (L, K ′ ) of infinite
rank, including one such K and K ′ in any split-generating collection. (2’) further prohibits such
an L from having a non-trivial idempotent summand which is right or left proper (in contrast, any
compact exact Lagrangian Q is both non-zero and right and left proper). As explained in Corollary
25 and Remark 26, the Main Theorem also implies an alternative for the groups SH ∗ (X) and
HW ∗ (L, L) (under the stated hypotheses) which is slightly weaker than, but partly explains the
persistence in all known examples of, the alternatives raised by Questions 3 and 1.
Remark 4 (Comparing Statements (1) and (2)). To compare Statements (1) and (2) of the
Main Theorem in an instance, let’s suppose X has a (split-)generating collection {∆i }ki=1 , Then
Statement (1) implies that` if W(X) 6= 0 at least one of the groups HW ∗ (∆r , ∆s ) is infinite rank,
∗
( i ∆i , i ∆i ) := ⊕r,s HW ∗ (∆r , ∆s ) is infinite or zero, answering Question
`
and in particular HW `
1 affirmatively for L = i ∆i .
Let’s further assume that in the split-generating collection {∆i }ki=1 , all of the objects are open
submanifolds (e.g., take the cocores of some Weinstein presentation which generate by [CDGG,
GPS2]) and furthermore discard all zero objects from the collection (so W(X) = 0 iff k = 0).
Statement (2) then implies the stronger result that at least k of the groups {HW ∗ (∆r , ∆s )}kr,s=1
must be infinite rank, at least one for each r and each s.4
More generally, we see that Statement (1) actually follows from from Statement (2) provided
one knows that if W(X) 6= 0, at least one of the non-zero generators of X must be a non-compact
Lagrangian L, a fact which is true for Weinstein manifolds and more generally for all non-degenerate
Liouville manifolds by known geometric arguments (for disc confinement reasons, Abouzaid’s crite-
rion cannot be satisfied by compact Lagrangians alone, compare [Gan3]). One can similarly deduce
Statement (1’) from a strengthened form of (2) whose statement is omitted from the theorem for
simplicity (one shows the image of such an L under categorical quotients is non-proper or vanishing)
whose proof is also in this paper (see e.g., Lemma 19), along with similar checkable hypotheses
about non-compact generators. Alternatively, e.g., Statement (1) can be extracted from Statement
(2) applied to the diagonal ∆ ⊂ X − × X (making this perspective precise requires some technical
digressions which we opted to avoid here). Although (2) and variants therefore imply (1) and vari-
ants, we nevertheless found it cleanest to state (and give superficially independent arguments for)
(1) and (2), to distinguish “global” (whole category) statements from “local” (single object) ones.
Remark 5. For Weinstein manifolds, one geometric source of quotients (resp. more general
homological epimorphisms) comes from restricting to Weinstein subdomains [GPS2, Prop. 8.15]
(resp. Liouville subdomains which are independently Weinstein [Sy]); Statement (1’) for such quo-
tients/homological epimorphisms is simply a consequence of Statement (1) for the subdomain.
Remark 6. From a mirror symmetry lens Statement (1’) implies that if there is an HMS
equivalence W(X) = coh(Y ), then no component of Y can be simultaneously smooth and proper.
The proofs of the statements in the Main Theorem appeal to the structure and properties of some
TQFT operations in wrapped Floer theory, particularly the “degeneracy” of the geometric closed
respectively open string copairing for Liouville manifolds respectively their exact open Lagrangian
submanifolds [Ri1], and its interplay with the fact that (under the hypotheses imposed) W(X) is a
4There are examples where exactly k of the HW ∗ (∆ , ∆ ) are infinite, e.g., when HW ∗ (∆ , ∆ ) = 0 for i 6= j.
r s i j
3
(weak) smooth Calabi-Yau category [Gan1].5 Broadly the proof of (1), respectively (2) goes as fol-
lows. Suppose W(X) (resp. L ∈ W(X)) is proper (resp. left or right proper). Then, we observe that
(given X is non-degenerate) there must exist a pairing on SH ∗ (X) (respectively HW ∗ (L, L)) that
is compatible with the copairings in the sense of satisfying the snake relation in TQFT. This implies
the copairing is non-degenerate which (by simultaneous degeneracy) implies W(X) (respectively L)
is zero. The stronger (1’) and (2’) follow from translating the degeneracy of geometric copairings
into the degeneracy of certain algebraic copairings which exist on W(X) respectively L under the
stated hypotheses, a property (here called algebraic openness) that both implies the non-properness
or vanishing alternative and is inherited by quotients (respectively summands).
One can ask to what degree the Main Theorem holds in other studied non-compact settings6. If
one turns on a stop, partially wrapped Fukaya categories (or wrapped Fukaya categories of Liouville
sectors like T ∗ Rn ) can be proper non-zero (it may be interesting to articulate a relevant criterion
which sometimes fails and sometimes holds). Another question is to what extent the result holds
in non-exact (non-compact) cases when all of the structures are defined. Here we note that either
of the key properties used, non-degeneracy of W(X) and the degeneracy of the copairings, could
independently fail or hold in non-exact non-compact settings. For instance, as spelled out in Remark
9, work of Albers-Kang [AK] implies the failure of degeneracy of the copairing for negative line
bundles, making our arguments inapplicable in such settings. Computations by Ritter/Ritter-Smith
[Ri2, RS] show that the outcome of Main Theorem indeed does also fail in such settings, despite
a version of nondegeneracy holding. See Remarks 9, 11, and 27 for more exploration of non-exact
settings, including structural hypotheses under which the result would continue to hold.
The paper is organized as follows. In §2, we review the definition of the geometric closed and
open-string copairings and their degeneracy property, following [Ri1, Thm 13.3] with a slightly
different exposition. In §3, we turn towards the more abstract setting of weak smooth Calabi-
Yau categories and review the fact that (as a special case of the idea that these categories carry
suitable open-closed TFT operations) such categories carry algebraic “closed-string” and “open-
string” copairings (on Hochschild homology and the endomorphisms of any object respectively). We
also show that — when these copairings are degenerate in a suitable sense — the category/object
in question, and in fact any quotient/summand thereof, must be non-proper or zero, due to the
automatic existence of a dual compatible pairing when proper (as constructed in [Sh, BD], though
the object case requires some further arguments spelled out here). In §4 we tie the threads together
and recall that the wrapped Fukaya category indeed has a canonical (weak) smooth Calabi-Yau
structure [Gan1], whose associated algebraic closed and open-string copairings recover the geometric
closed and open-string copairings (and are therefore degenerate) by [Re] and [Gan1] respectively
(again with some further spelling out in the open-string case). The proof of the Main Theorem
appears in §5. An Appendix collects a few additional needed results about weak smooth Calabi-Yau
structures which we expect are well known (but for which we could not find a precise reference).
Conventions. For simplicity in our main argument below, we work over a fixed arbitrary field
K; however, the Main Theorem also holds over Z (where properness is defined in terms of taking
values in perfect dg Z-modules in perf (Z) ⊂ M od(Z), the subcategory of those chain complexes
which are quasi-isomorphic to a bounded complex of finitely-generated free abelian groups), as can
be most simply deduced from the case of fields by the universal coefficient theorem. We are agnostic
about grading structures, e.g., Floer-theoretic and categorical structures in sight could be Z/2 graded
or Z graded or have some other grading (degrees and degree shifts should be interpreted mod 2 in
the former case). While at one point in §2 we appeal to cohomological degrees, this degree need not
coincide with gradings in wrapped Floer theory. We indicate the K-linear dual of a vector space
V by V ∨ . All (dg or A∞ ) categories considered in this paper are (cohomologically graded and)
cohomologically unital, meaning there are cohomological identity morphisms.
5W(X) further has a (non-weak) smooth Calabi-Yau structure [Gan3], but we only use the weak structure here.
6in compact cases, Fukaya categories are proper by construction
4
Acknowledgments. I am grateful to Mohammed Abouzaid, Alexander Efimov, Yuan Gao,
Oleg Lazarev, Yankı Lekili, Daniel Pomerleano, Vivek Shende, Zachary Sylvan, Alex Takeda, Sara
Venkatesh, and Wai-Kit Yeung for helpful conversations and comments. I was partly supported by
NSF grant DMS–1907635.
Remark 9. It’s worth examining how the degeneracy Lemma (Theorem 7) could fail or alterna-
tively succeed on a non-exact X (assume one has a definition of SH ∗ (X) with its TQFT structures),
seeing as it is a key component in categorical “non-properness or vanishing” phenomena. If X is
compact the argument breaks because the difference between Hc∗ (X) and H ∗ (X) vanishes; indeed
the copairing is non-degenerate in such cases as one can construct a pairing. For more general
non-exact X the presence of J-holomorphic spheres means that the acceleration map from H ∗ (X)
to SH ∗ (X), a non-compact version of the PSS morphism [PSS], should be an algebra map for the
quantum product (rather than classical cup product) on H ∗ (X). Classes of positive cohomologi-
cal degree may no longer be nilpotent for the quantum product on H ∗ (X), which could break the
argument (even if X is non-compact).
On the other hand, in the absence of strictly positive Chern number holomorphic spheres, the
quantum product should only preserve or increase cohomological degree, meaning (provided each
component of X is non-compact) the image of Hc∗ (X) → H ∗ (X) should remain nilpotent and
Theorem 7 would seem to be applicable. For instance, this would seem to be the case on open
non-exact symplectic manifolds X for which c1 (X) = 0 and SH ∗ (X) is defined (e.g., on a toric
Calabi-Yau). We thank Abouzaid for discussions about this case.
Note that on some negative line bundles, which (are non-compact and) have positive Chern
number holomorphic spheres, Theorem 7 indeed is known to fail: Ritter showed SH ∗ (X) can be
non-zero [Ri2] while Albers-Kang [AK, §4] compute that nevertheless in such cases Rabinowitz
Floer homology vanishes, or equivalently cSH is non-degenerate (see Remark 8).
Turning to the open-string setting, we recall that there is correspondingly an open TQFT
structure on wrapped Floer cohomology (also developed in [Ri1, Thm. 6.13]). For any triple of exact
Lagrangians L0 , L1 , L2 ∈ W(X), counting maps from a disc with two inputs and one output induces a
composition map in wrapped Floer cohomology [µ2 ] : HW ∗ (L1 , L2 )⊗HW ∗ (L0 , L1 ) → HW ∗ (L0 , L2 )
a cohomological shadow of the chain-level A∞ structure (see e.g., [A]). Counting maps from an
(unstable) disc with one output induces cohomological identity morphisms [idL ] ∈ HW ∗ (L, L) for
each L which along with [µ2 ] (restricting to L0 = L1 = L2 = L) again equip HW ∗ (L, L) with the
structure of a unital K-algebra. There is also a geometric open-string copairing on the wrapped
Floer cohomology which cohomologically gives an element
(2.1) cK,L : K → HW ∗ (K, L) ⊗ HW ∗ (L, K)[n]
defined for any pair of exact Lagrangians K and L. We’ll equate cK,L with cK,L (1) and further
indicate cL,L by simply cL . The element (2.1) is constructed in exactly the same manner as the
closed-string copairing, by counting maps from an (unstable) disc with two outputs (or equivalently
from a stable disc with two outputs and one interior point with fixed cross ratio, with the interior
point unconstrained); this is a special case of the construction in e.g., [Ri1, Thm 6.13], or a variant
of the coproduct map in [A] with no inputs).
Theorem 10 (“Open-string degeneracy Lemma”, open-string analogue of Ritter [Ri1] Thm
13.3). Let L ∈ W(X) be an exact open Lagrangian (i.e., every component of L is non-compact).
Then the image of c∗L of any class in HW ∗ (L, L)∨ [−n] is nilpotent in HW ∗ (L, L). In particular, cL
is non-degenerate if and only if L = 0 (i.e., equivalently HW ∗ (L, L) = 0).
Proof. The argument is completely identical to the proof of Theorem 7. The salient points
are: the open-string copairing factors through constant paths, forcing the induced map c∗L to factor
6
as HW ∗ (L, L)∨ [−n] → H ∗ (L)∨ [−n] → H ∗ (L) → HW ∗ (L, L). Since every component of L is non-
compact, the image of the middle map consists of nilpotent elements with respect to the cup product.
Since L is exact the rightmost map is an algebra map (for the cup product and [µ2 ]).
Again this degeneracy property can be thought of as providing (unless the group in question
vanishes) a obstruction to the existence of a compatible pairing.
Remark 11. Continuing Remark 9, we examine how Theorem 10 could fail or succeed if the
hypotheses are relaxed (supposing ordinary and wrapped Floer homology are still well-defined). If
L is a compact (say exact) Lagrangian L, the map Hc∗ (L) → H ∗ (L) is an isomorphism, breaking the
above proof. Indeed, standard Floer theoretic methods allow one to construct a pairing HF ∗ (L, L)⊗
HF ∗ (L, L) → K[−n]; which can be used (along with the usual snake relation) to establish non-
degeneracy of the geometric open-string copairing even though one has HF ∗ (L, L) ∼ = H ∗ (L) 6= 0.
In the technical construction of such moduli spaces, one sees that there is extra freedom in the
choice of “Floer data” which allows for operations without outputs when one or both Lagrangian is
compact (without violating the maximum principle). One expects for the same reasons the failure
of Theorem 10 to persist for compact L in more general non-exact settings, provided such an L has
well-defined non-vanishing Floer theory.
Supposing L ⊂ X is non-compact and non-exact (whether X is exact or not), and all of the above
structures are defined. A similar argument would imply the map c∗L now factors through the map
from HF ∗ (L− , L) to HF ∗ (L+ , L) where HF ∗ (L− , L) is possibly a deformation (by counting discs)
of Hc∗ (L) and HF ∗ (L+ , L) similarly possibly a deformation of H ∗ (L). These groups (if defined)
may not coincide with ordinary (compactly supported or usual) cohomology, and even if there is a
coincidence of vector spaces, the algebra structure may still be different/deformed. Theorem 10 may
continue to hold if the groups are still Z-graded or there is more generally some “non-positivity”
control over Maslov indices of discs (paralleling Remark 9).
3.1. The algebraic copairings. Recall that for any dg or A∞ category C over K (with coho-
mological composition denoted [µ2 ]), one has a dg category [C, C] of (A∞ ) C−C bimodules (defined as
bilinear K-linear A∞ functors from Cop ×C to chain complexes over K); see [S2,Gan1,Sh,BD,S3] for
some references. We indicate morphisms (derived hom) by homC−C (−, −), the operation of (derived)
one-sided tensor product by − ⊗C − (this produces another bimodule) and the operation of (derived)
7
two-sided tensor product by − ⊗C−C − (this produces a chain-complex), and for our purposes isomor-
phisms respectively inverses mean homological isomorphisms/inverses. In the category [C, C], there
are some canonical objects, the diagonal bimodule C∆ := homC (−, −) (sometimes just indicated by
C) and the representable (or Yoneda) bimodules YK,L := homC (K, −) ⊗ homC (−, L). Evaluating at
a pair of objects evK,L : B 7→ B(K, L) induces a dg functor from bimodules to chain complexes,
and in particular there is a chain map (evK,L )∗ : homC−C (B, B′ ) → homMod(K) (B(K, L), B′ (K, L)),
sending a morphism f to (evK,L )∗ f = fK,L (which in the bar model can be thought of as projection
to a length zero quotient complex.
The Hochschild homology HH∗ (C) of C is the (cohomology of the) self-bimodule (two sided) tensor
product of the diagonal bimodule, whereas the Hochschild cohomology HH∗ (C) is the (cohomology)
of the endomorphisms of the diagonal bimodule (our grading conventions follow [A, Gan1]). More
generally, one can take the Hochschild homology (= bimodule tensor product with diagonal of)
and Hochschild cohomology (=endomorphisms from diagonal to -) of any bimodule B; the results
are denoted HH∗ (C, B) and HH∗ (C, B) and generalize the previous case by setting B = C∆ . There
∩
is a natural cap product action HH∗ (C, B) ⊗ HH∗ (C) → HH∗ (C, B), coming from the functoriality
of bimodule tensor product in the bimodules being considered. For any object X ∈ C, there are
natural maps H ∗ (B(X, X)) → HH∗ (C, B) and HH∗ (C, B) → H ∗ (B(X, X)) which on the level of
bar complexes can be modeled as inclusion of or projection to the length 0 terms; hence we’ll call
the maps inclusion respectively projection. In the case of diagonal coefficients, the map HH∗ (C) →
H ∗ (homC (X, X)) = H ∗ (C∆ (X, X)) is a unital algebra map, and the map map H ∗ (homC (X, X)) →
HH∗ (C) can be used to define the Chern character chX ∈ HH0 (C) of X to be the image of the
identity [idX ]. (Implicitly all A∞ categories we consider are cohomologically unital, i.e., they have
cohomological identity elements).
We say C is (homologically) smooth if its diagonal bimodule C∆ is a perfect bimodule, meaning
it is split-generated by representable (or Yoneda) bimodules. Denoting perf (C−C) the subcategory
of perfect bimodules, Morita invariance implies that the bilinear embedding Cop × C → perf (C−C)
induces an isomorphism HH∗ (Cop ) ⊗ HH∗ (C) ∼ = HH∗ (perf (C−C)); furthermore there is an isomor-
phism HH∗ (Cop ) ∼ = HH∗ (C). When C is smooth, the Chern character of the diagonal bimodule chC∆
therefore gives an element in HH∗ (perf (C−C)) ∼ = HH∗ (C) ⊗ HH∗ (C), or equivalently a map from K
to HH∗ (C) ⊗ HH∗ (C) which we call the algebraic closed-string copairing calg [Sh].
Remark 12. Note that while the algebraic closed-string copairing only requires smoothness to
define, we will require a (weak smooth) Calabi-Yau structure to formulate the relevant “algebraic
openness” degeneracy condition; see Definition 15.
Recall that for any bimodule P, there is a bimodule dual P! , whose value on a pair of ob-
jects (K, L) is P! (K, L) := homC−C (P, YK,L ). We call C!∆ the inverse dualizing bimodule (note
H ∗ (C!∆ (K, L)) = HH∗ (C, YK,L )), and note there is a canonical map ev : HH∗ (C) → H ∗ (homC−C (C! , C))
sending an element α to a morphism of bimodules which (on the level of maps of chain complexes)
corresponds to capping with α. ev is an equivalence for smooth C (analogous to the fact that the
canonical map for vector spaces V ⊗ V → hom(V ∗ , V ) is an equivalence when V is finite). A (weak)
smooth Calabi-Yau structure of dimension n on a smooth category C is an element σ ∈ HH−n (C)
such that ev(σ) is an isomorphism of bimodules C! [n] → C; on the cohomology level for a pair (K, L)
∼
=
the map ev(σ) is the operation − ∩ σ : HH∗ (C, YK,L ) → HH∗−n (C, YK,L ) ∼ = C∆ (K, L). It follows that
∼
=
the cap product map − ∩ σ : HH∗ (C, B) → HH∗−n (C, B) is an isomorphism for any bimodule, as it
can be realized by taking the bimodule tensor product of ev(σ) with B, along with the isomorphism
H ∗ (C!∆ ⊗C−C B) ∼= HH∗ (C, B) that exists when C is smooth. A weak smooth Calabi-Yau category
is a pair (C, σ) of a (dg/A∞ ) category equipped with a weak smooth Calabi-Yau structure. For a
∼
=
weak smooth Calabi-Yau category (C, σ), we will denote by CYC : HH∗−n (C) → HH∗ (C) the inverse
equivalence to capping with σ.
8
We now turn to defining the open-string copairing associated to a weak Calabi-Yau structure.
First we define a copairing cfK,L ∈ H ∗ (homC (K, L)) ⊗K H ∗ (homC (L, K))[n] associated to any closed
morphism of bimodules f ∈ homnC−C (C∆ , C! [n]), called the copairing shadow of f . The morphism
f induces, for every pair of objects (L, L) a chain map fL,L : homC (L, L) → C! [n](L, L); taking
the image of any cocycle idL cohomologically representing the identity (any ‘cohomological unit’)
gives a cocycle fL,L (idL ) ∈ C! (L, L) = homC−C (C, homC (−, L) ⊗K homC (L, −)); now such a bimodule
morphism induces, for any object K, a map of chain complexes (fL,L (idL ))K,K : homC (K, K) →
homC (K, L) ⊗K homC (L, K); taking the image of idK induces a well defined cohomology class
[(fL,L (idL ))K,K (idK )] ∈ H ∗ (homC (K, L)) ⊗K H ∗ (homC (L, K)) which we define to be the copair-
ing shadow cfK,L (this only depends on the cohomology class [f ]). Let’s observe first that by Morita
invariance the categories of bimodules associated to C and its pre-triangulated split-closure perf (C)
are naturally quasi-equivalent in a way identifying (up to quasi-isomorphism) the diagonal bimodules
and inverse dualizing bimodules. Hence, such an f induces a canonical copairing shadow cfK,L for
any K, L ∈ perf (C), coinciding with the previously defined copairing shadow for objects in C.
Finally, given a weak smooth Calabi-Yau structure σ on C so ev(σ) gives an isomorphism of
∼
bimodules C! [n] → C, we define the algebraic open-string copairing associated to (C, σ) to be the
ev(σ)−1
copairing shadow of the inverse to ev(σ): cσK,L := cK,L . As before we will note that these are
defined for any K, L ∈ perf (C) and we will abbreviate cσL := cσL,L .
3.2. Properness implies non-degeneracy. We will recall here that one has strong control
over the closed respectively open string algebraic copairings on a weak smooth Calabi-Yau category
when the given category respectively object is proper. The closed string case is directly in the
literature and doesn’t require the Calabi-Yau structure:
Proposition 13 (Shklyarov [Sh], where the result is attributed to Kontsevich-Soibelman).
When a smooth category C is further proper, then its algebraic closed-string copairing is perfect,
meaning c∗alg is an isomorphism.
Sketch. When C is proper, loc. cit. proves HH∗ (C) carries a canonical pairing (which can be
thought of as essentially the pushforward map induced by hom : Cop ⊗ C → perf (K) after using
Morita invariance to deduce HH∗ (perf (K)) = HH∗ (K) = K and appealing to the Künneth formula
for HH∗ and op-invariance HH∗ (Cop ) ∼ = HH∗ (C)). loc. cit. further proves that when C is smooth,
the closed-string algebraic copairing and pairing are related by the usual snake relation, implying
in particular that the map V → V ∗ → V induced first by the pairing and then the copairing is the
identity map. This implies c∗alg : V ∗ → V is an isomorphism as desired.
Turning to the case of open-string copairings, we shall now explain that similarly, (right or
left) properness of an object L implies non-degeneracy of its algebraic copairing. The main idea
is to appeal to a more general result of Brav-Dyckerhoff [BD, Thm 3.1] that smooth Calabi-Yau
structures (“left Calabi-Yau structures” in loc. cit.) induce proper Calabi-Yau structures (“right
Calabi-Yau structures” in loc. cit.) on any subcategory of (right respectively left) proper objects,
and in particular a perfect pairing on endormorphisms of any such object. We will recall this below
and further show this induced pairing is compatible (fits into a snake relation with) with the algebraic
open-string copairing defined above:
Proposition 14. Let C be a (dg/A∞ ) category equipped with a weak smooth Calabi-Yau struc-
ture σ ∈ HH−n (C), and let cσ be the σ-induced open-string copairing. If P ⊂ C denotes any collection
of right-proper objects, then for any K, L in P, cσK,L is perfect. The same conclusion holds if P ⊂ C
is any collection of left-proper objects.
Proof of Proposition 14. [BD, Thm 3.1] shows in particular that any collection of right
proper objects P (in this case P = {L}) inherits (from the weak smooth Calabi-Yau structure σ
on C) a weak proper Calabi-Yau structure (the main result cited concerns non-weak Calabi-Yau
9
structures, but we only need the weaker version, where the crux of the proof lies). Recall that a
weak proper Calabi-Yau structure on a category P is a map trP : HH∗ (P) → K[−n] such that under
the natural identification HH−n (P)∨ → homP−P (P∆ [n], P∨ ) (where P∨ := homP (−, −)∨ denotes
the (linear) dual of the diagonal bimodule), trP corresponds to a bimodule quasi-isomorphism.
Equivalently, the map ptr ∗ ∗
K,L : H (homP (K, L)) ⊗ H (homP (L, K)) → K[−n] induced by composing
P
[µ2 ] tr
H ∗ (homP (X, Y )) × H ∗ (homP (Y, X)) → H ∗ (hom(Y, Y )) → HH∗ (P) →P K[−n] perfect pairing; in
particular a weak proper Calabi-Yau structure has an associated perfect pairing ptr K,L . To define
P
the weak proper Calabi-Yau structure on a collection of right proper objects P in a weak smooth
Calabi-Yau category (C, σ), consider the Shklyarov pairing HH∗ (C) ⊗ HH∗ (P) → K associated to
the hom pairing Cop × P → perf (K) (this uses the fact that P consists of right proper objects),
then define trP := hσ, −i. As it induces a weak proper Calabi-Yau structure, σ therefore induces
hσ,−i
a pairing on objects of P, defined as pσK,L := pK,L . The argument in loc. cit. was written in the
setting of dg categories but the same argument with the same proof carries through in the setting of
A∞ categories; or we can simply reduce to the dg case by replacing each A∞ category with a quasi-
equivalent dg category (noting that the notions of weak smooth Calabi-Yau structure, associated
copairing, and proper object all transfer over).
It remains to check the snake relation between the algebraic open-string copairing cσK,L and
pK,L . Equivalently we will show (cσK,L )∗ : H ∗ (homP (L, K))∨ [−n] → H ∗ (homP (K, L)) and (pσK,L )∗ :
σ
H ∗ (homP (K, L)) → H ∗ (homP (L, K))∨ [−n] are inverse for any K, L ∈ P. To do so, we will first
recall that in the proof from loc. cit. the following commutative diagram appears:
∼
= ∼
=
Φ
homC−C (C! [n], C∆ ) / homP−P (P∆ [n], P∨ )
where the top horizontal arrow is the map α 7→ hα, −i using the Shklyarov-type pairing between
C and P defined above. The vertical arrows are natural identifications (the left arrow for instance
sends α to ev(α)) and the bottom horizontal arrow, which we will expand upon, sends bimodule
quasi-isomorphisms to bimodule quasi-isomorphisms (as it is induced by a composition of various
functors between bimodule categories). It follows that the image of ev(σ) induces a bimodule quasi-
∼
isomorphism p∗σ : P → P∨ [−n] (which implies by the commutativity that trP := hσ, −i is a weak
proper Calabi-Yau structure). In particular, by evaluating at the pair of objects K, L ∈ P, we obtain
a quasi-isomorphism of chain complexes (pσ )∗K,L : hom(K, L) → hom(L, K)∨ [−n], which is (passing
to the adjoint hom(K, L) ⊗ hom(L, K) → K[−n]) the desired pairing by definition.
If we now apply the same functorial map Φ to the (homologically) inverse quasi-isomorphism of
bimodules CY = ev(σ)−1 ∈ homC−C (C∆ , C! [n]), and then evaluate at the pair of objects K, L, we will
obtain a (homologically) inverse quasi-isomorphism of chain complexes ((pσ )∗K,L )−1 : hom(L, K)∨ [−n] →
hom(K, L). It suffices to show that this inverse ((pσ )∗K,L )−1 (which is induced by CY = ev(σ)−1 ) can
be computed as (ev(σ)−1 )K,K ([idK ])L,L ([idL ])∗ = (cσK,L )∗ = (− ⊗ id)(cσK,L ). This is an immediate
consequence of verifying more generally that the following diagram is cohomologically commutative:
Φ
(3.2) homC−C (C, C! [n]) / homP−P (P∨ , P[n])
where the left vertical arrow takes a map f : C → C! [n] and looks at its copairing shadow, the bottom
horizontal arrow takes c 7→ (c∗ : φ 7→ (φ ⊗ id) ◦ c), and the right vertical arrow simply considers
10
the underlying map of chain complexes associated to an element homP−P (P∨ , P[n]) for a pair (K, L)
both in P.
To verify this, let us describe the map Φ from loc. cit. in some more detail. First, one can
restrict a C−C bimodule along the inclusion i : P → C on the right; we’ll call the right restriction of
a bimodule B B|P following [BD]. Restriction is functorial, inducing a map
(3.3) resP : homC−C (C, C! [n]) → homC−P (C|P , C! [n]|P );
the image of CY = ev(σ)−1 is simply CY |P , and evidently for K, L in P, this restriction commutes
with the natural maps that take CY respectively CY |P to their copairing shadow for a pair of objects
in P.
We now apply the functor homC (−, C) (where homC is taken in the category of left modules),
a contravariant functor from C−P bimodules to P−C bimodules (note the right module structure
on the target C survives as does the right P module structure on the source which becomes a left
module structure by dualization):
(3.4) Φ1 : homC−P (C|P , C! [n]|P ) → homP−C (homC (C! |P [n], C), homC (C|P , C));
On the level of chain complexes, for an object K ∈ P
(3.5)
(Φ1 )K,K := homK (C(K, K), C! [n](K, K)) → homK (homK (C! (K, K), C(K, K)), homK (C|P (K, K), C(K, K)))
f 7→ (−) ◦ f
Next we observe that by properness of the bimodule C|P (as P consists of right-proper objects),
∼
=
there is a natural isomorphism of bimodules ev : C|P → (C|∨ ∨
P ) . Composing with ev, we see
ev⊗id◦−
that C! [n]|P = homC−C (C, C|P ⊗ C)[n] −→ homC−C (C, (C|∨ ∨
P ) ⊗ C)[n]. Using the canonical map
∨
α : M ⊗K N → homK (M, N ) which exists for any M , N (modules, chain complexes etc), which is an
isomorphism if M is proper (applicable here because C|P is proper), we can map this latter complex
∼
=
isomorphically to → homC−C (C, homK ((C∨ P ), C))[n], which by hom-tensor adjunction is equal to (in
fact coincides on the level of chain complexes built from the bar construction) homC (C ⊗C C∨
P , C)[n].
Then there is a natural “collapse” quasi-isomorphism of bimodules
∼
=
(3.6) C ⊗C C|∨ ∨
P → C|P ;
On the level of chain complexes, the homologically inverse chain map C∨ ∨
P (K, L) → (C ⊗C CP )(K, L)
∨ ∨
sends x 7→ idK ⊗ x ∈ C(K, K) ⊗K C|P (K, L) ⊂ (C ⊗C C|P )(K, L) (using the bar model). Composing
∼
=
with (3.6) induces a quasi-equivalence homC (C|∨ ∨
P , C) → homC (C ⊗C C|P , C); inverting this we all
! ∼ ∨
together get a quasi-isomorphism η : C [n]|P = homC (C|P , C); which on the level of chain complexes
(by above) admits the following description
ηK,K : homC−C (C, C|P (−, K) ⊗ C(K, −))[n] → homC (C|∨
P (−, K), C(−, K))
(3.7)
g 7→ (φ 7→ (φ ⊗ id) (g(idK )))
The quasi-isomorphism η induces
(3.8)
∼
=
Φ2 : homP−C (homC (C! |P [n], C), homC (C|P , C)) → homP−C (homC (homC (C|∨
P , C)[n], C), homC (C|P , C))
which on the level of chain complexes sends h 7→ (ψ 7→ h(ψ ◦ η)). Finally there is a canonical
morphism from C|∨ ∨
P to its double left-C-module dual homC (homC (C|P , C), C) which is an isomorphism
∨
along any object p ∈ P for which C|P (−, p) is a perfect C-module; this holds in our case for all p ∈ P
seeing as C|∨
P (−, p) is a proper C-module and C is smooth (over smooth categories, proper modules
are automatically perfect; compare [GPS1, Lemma A.8]):
∼
=
(3.9) C|∨ ∨
P → homC (homC (C|P , C), C).
On the chain level the above map sends φ 7→ (z 7→ z(φ)).
11
Precomposing with the inverse of (3.9) (shifted by −n) we obtain an equivalence
(3.10) Φ3 : homP−C (homC (homC (C|∨ ∨
P , C)[n], C), homC (C|P , C)) → homP−C (C|P [−n], homC (C|P , C))
3.3. A criterion for non-properness. Given a graded unital algebra A, we will say a copair-
ing c ∈ A ⊗ A[k] is entirely degenerate if the induced map c∗ : A∨ [−k] → A has nilpotent image. It
is easy to see that an entirely degenerate c can only be non-degenerate if 1 is nilpotent i.e., if A = 0.
On a weak smooth Calabi-Yau category C, in order to talk about the degeneracy properties of
its algebraic closed-string copairing, we transfer the copairing to Hochschild cohomology, a unital
algebra, via the Calabi-Yau isomorphism HH∗−n (C) ∼ = HH∗ (C) (or one can equivalently transfer the
ring structure to HH∗ (C)). The following terminology is motivated by the usage of “open manifold”
to describe manifolds all of whose components are non-compact:
Definition 15. A (weak) smooth Calabi-Yau category C := (C, σ) is algebraically open if its
algebraic closed-string copairing, thought of as living on HH ∗ (C) via the Calabi-Yau isomorphism,
is entirely degenerate.
Similarly, we say an object X in a (weak) smooth Calabi-Yau category C is algebraically open
if its algebraic open-string copairing on H ∗ (homC (X, X)) is entirely degenerate.
An immediate Corollary of the previous section is that algebraic openness provides a non-
properness or vanishing criterion:
Corollary 16. Let C := (C, σ) be an algebraically open weak smooth Calabi-Yau category.
Then C is either non-proper or zero. Similarly, if X ∈ C is an algebraically open object in a weak
smooth Calabi-Yau category, X is either (both left and right) non-proper or zero.
12
Proof. Suppose such a C is proper. Then Proposition 13 implies that its algebraic closed-
string copairing calg is non-degenerate, but it is also entirely degenerate, hence HH∗ (C) is zero,
which implies C is zero (as the cohomological endormophisms of any object are a unital module over
HH∗ (C)). Similarly, suppose X ∈ C is right proper. Proposition 14 implies the algebraic open string
copairing is non-degenerate, but by algebraic openness this can only happen if X = 0. The left
proper case is the same.
We conclude this section by showing that the categorical and object-wise algebraic openness
conditions are sufficiently strong so as to persist under quotients/homological epimorphisms and
idempotent summands.
Proposition 17. Let C be (weak) smooth Calabi-Yau, and suppose C is algebraically open.
Then any quotient/homological epimorphism D of C inherits from C a (weak) smooth Calabi-Yau
structure, with respect to which it is also algebraically open.
Proof. The algebraically open condition is equivalent to the map CYC ◦ c∗alg : HH∗ (C)∨ →
HH∗+n (C) having nilpotent image, so we will show this condition transfers along quotients/homological
epimorphisms. Let f : C → D denote the quotient map (or homological epimorphism). Lemmas
A.3 and A.4 imply that D is again smooth, that the algebraic closed-string copairing on D factors
through the one on C (i.e., (f∗ ⊗ f∗ )(cC D
alg ) = calg ), that D inherits a weak smooth Calabi-Yau struc-
ture from the one on C, and that there is a commutative diagram (using the inherited Calabi-Yau
structure to define CYD )
f∗
HH∗ (C) / HH∗ (D)
∼
= CYC ∼
= CYD
f♯
HH∗ (C) / HH∗ (D)
∗ ∨ C ∗
It follows that CYD ◦ (cD
alg ) therefore factors as f∗ followed by CYC ◦ (calg ) (whose image is
nilpotent by hypothesis) followed by the map f♯ (an algebra map by Lemma A.2); hence it has
nilpotent image.
Lemma 18. Let {cfL,K }K,L∈perf (C) be the collection of copairing shadows associated to any closed
morphism of bimodules f : C∆ → C! [n] above. If L is algebraically open with respect to cfL , and K is
any idempotent summand of L in perf (C), then K is algebraically open with respect to cfK .
In particular, with respect to the collection of algebraic open-string copairings {cσL,K }K,L∈perf (C)
induced by a weak smooth Calabi-Yau structure, the condition of algebraic openness is inherited by
idempotent summands.
Proof. By Morita invariance we replace C with perf (C) and think of f as a morphism of
bimodules over perf (C). Let [p] : L → K and [i] : K → L be the homological projections of L onto
K and inclusion of K into L, so [i] ◦ [p] : L → L is an idempotent, and [p] ◦ [i] = [idK ]. Composition
([µ2 ]) with [p] on the right ([µ2 ]([p], −)) and precomposition with [i] on the left ([µ2 ](−, [i])) gives
a map πL,K : H ∗ homC (L, L) → H ∗ homC (K, K) and hence a map πL,K ⊗ πL,K : H ∗ homC (L, L) ⊗
H ∗ homC (L, L) → H ∗ hom∗C (K, K) ⊗ H ∗ homC (K, K). The main claim that this map sends cL to
cK . Supposing it did, the proof can be completed as follows: since πL,K ⊗ πL,K sends cL to cK , it
∗
πL,K c∗ πL,K
follows that c∗K factors as H ∗ homC (K, K)∨ [−n] → H ∗ homC (L, L)∨ [−n] →
L
H ∗ homC (L, L) →
∗
H homC (K, K). Since the last map πL,K is an algebra map and hence sends nilpotent elements to
nilpotent elements, we are done.
Finally, the claim that the map πL,K sends cL to cK is an elementary consequence of the
map f being a closed morphism of A∞ bimodules. To spell this out, it is convenient to recast
the induced map on chain complexes fA,B (−)C,D (by hom-tensor adjunction) as a map fA,B,C,D :
H ∗ (hom(A, B)) ⊗ H ∗ (hom(C, D)) → H ∗ (hom(A, D)) ⊗ H ∗ (hom(C, B)) for every tuple of objects
13
A,B,C,D, compatible with the homology compositions on the right and left (since f is a closed mor-
phism). By definition, cL := fL,L,L,L([idL ] ⊗ [idL ]), and πL,K ⊗ πL,K = (µ2 (−, [i]) ◦ µ2 ([p], −))⊗2 ,
Hence, iteratively appealing to the compatibility with multiplication (using the abuse of notation µ2
for [µ2 ]), we eventually learn that πL,K ⊗πL,K ◦fL,L,L,L([idL ]⊗[idL ]) = fK,K,K,K (µ2 ([i], µ2 ([idL ], [p]))⊗
µ2 ([i], µ2 ([idL ], [p]))) = fK,K,K,K (µ2 ([i], [p]) ⊗ µ2 ([i], [p])) = fK,K,K,K ([idK ] ⊗ [idK ]) = cK .
Similarly, algebraic openness of an object is preserved by quotients/homological epiomorphisms:
Lemma 19. If (C, σ) is a weak smooth Calabi-Yau category and let f : C → D any quo-
tient/homological epimorphism. If X ∈ C is algebraically open, then f X ∈ D is algebraically open
(with respect to the weak smooth Calabi-Yau structure f∗ σ induced by the map from C by Lemma
A.4).
Proof. Lemma A.5 implies that f sends the open-string copairing on X ∈ (C, σ) to the one on
f X ∈ (D, f∗ σ), after which the proof follows the same lines as the start of the previous Lemma.
Lemma A.1. [Compare [E] Cor. 3.8 and Prop 3.4, [GPS3] Lemma 3.15] If f : C → D is a (dg
or A∞ ) quotient functor, then the canonical morphism c : f∗ (C∆ ) → D∆ is an isomorphism.
Recall that the quotient of an A∞ category C is another A∞ category D equipped with a functor
C → D which is in some sense initial among all functors sending a subcategory to zero; the definition
and various explicit models are discussed in [D, LO, LM] (the first reference in the dg case). We
say f is a homological epimorphism (compare [E]) if c : f∗ C∆ → D∆ is an isomorphism; Lemma A.1
ensures this is a weaker condition than asking f be a quotient.
Recall that Hochschild homology is covariantly functorial, meaning that functors f : C → D
induce pushforward maps f∗ : HH∗ (C) → HH∗ (D). While Hochschild cohomology is not generally
functorial, we do have covariant functoriality under (dg or A∞ ) quotient functors:
Proof Sketch. For the first statement, note f induces a functor f∗ : [C, C] → [D, D] and hence
a map from HH∗ (C) to (cohomological) endomorphisms of f∗ C∆ . In light of Lemma A.1, localization
implies there is a quasi-isomorphism c ◦ (−) ◦ c−1 : homD−D (f∗ C∆ , f∗ C∆ ) → homD−D (D∆ , D∆ ). The
second assertion follows immediately from the (cohomologically) commutative diagram, for any
[α] ∈ H ∗ (C∆ ⊗C−C C∆ ) = HH∗ (C):
C∆ ⊗C−C C∆ / f ∗ D∆ ⊗C−C C∆ / D∆ ⊗C−C (f∗ C∆ ) o D∆ ⊗D−D D∆
−1
c
where the first horizontal map is induced by the canonical map d : C∆ → f ∗ D∆ , the second
∼
horizontal map comes from the natural equality f ∗ D∆ ⊗C−C C∆ ← f ∗ (D∆ ⊗D D∆ ⊗D D∆ )⊗C−C C∆ =
D∆ ⊗D−D (D∆ ⊗C C∆ ⊗C D∆ ) = D∆ ⊗D−D f∗ C∆ , the first and second vertical maps are induced
applying α to the right C∆ factor, the third vertical map induced by applying f∗ (α), and the fourth
induced by the commutative diagram, i.e., by (cohomologically) applying c ◦ f∗ (α) ◦ c−1 = f♯ (α).
Lemma A.3. If f : C → D is a quotient functor (or more general homological epimorphism) with
C homologically smooth, then D is smooth too, and f∗ sends the algebraic closed-string copairing for
C to the algebraic closed-string copairing for D.
The following result recalls that localizations or more general homological epimorphisms out of
(weak) smooth Calabi-Yau categories inherit (weak) smooth Calabi-Yau structures:
f∗ Φf
ev / H ∗ (homD−D (D! , D∆ ))
HH∗ (D) ∆
(see e.g., [BD, Prop 4.3], where the results are stated for dg categories with minor notational
differences; the proof is the same in the A∞ case or one can simply perform a dg replacement) where
Φf can be computed as the composite
f∗
homC−C (C!∆ , C∆ ) → homD−D (f∗ (C∆ )! , f∗ (C∆ )) → homD−D (D!∆ , D∆ );
for the middle term space we are using [BD, Lemma 4.2] to equate f∗ (C!∆ ) with f∗ (C∆ )! and the
second arrow is induced by the canonical map c : f∗ (C∆ ) → D∆ , which is an isomorphism by
hypothesis. Since the first arrow (the map on morphism spaces induced by a functor) always sends
isomorphisms to isomorphisms, we conclude Φf sends isomorphisms to isomorphisms. In light of the
commutative diagram, we conclude that if σ ∈ HH−n (C) is a (weak) smooth Calabi-Yau structure,
then so is f∗ (σ) ∈ HH−n (D).
The second statement (commutative diagram) is an immediately corollary of the assertion from
Lemma A.2 that with respect to f♯ , f∗ is a map of HH∗ (C)-modules, i.e., for any σ ∈ HH∗ (C) and
α ∈ HH∗ (C), f∗ (α ∩ σ) = f♯ (α) ∩ f∗ (σ).
Lemma A.5. In the setting of Lemma A.4, applying f to morphism spaces sends the open-string
algebraic copairing cσK,L on (C, σC ) to the open-string algebraic copairing cff∗K,f
σC
L on (D, f∗ σ).
Proof. All such f considered are cohomologically unital by convention, i.e., f sends [idL ] to
[idf L ]. Hence, the Lemma is an immediate consequence of verifying the commutative diagram, for
any pair of objects (applying f∗ and using the isomorphism between f∗ C and D and between f∗ C!
and D! for the two leftmost vertical arrows):
(ev(σC ))−1
K,K
H ∗ (C∆ (K, K)) / H ∗ (C! (K, K)) / H ∗ (homC (K, L)) ⊗ H ∗ (homC (L, K))
[f ]⊗[f ]
(ev(f∗ σC ))−1
f K,f K
H ∗ (D∆ (f K, f K)) / H ∗ (D! (f K, f K)) / H ∗ (homD (f K, f L)) ⊗ H ∗ (homD (f L, f K)).
References
[A] M. Abouzaid, A geometric criterion for generating the Fukaya category, Publ. Math. Inst. Hautes Études
Sci. 112 (2010), 191–240.
[AK] P. Albers and J. Kang, Vanishing of Rabinowitz Floer homology on negative line bundles, Math. Z. 285
(2017), no. 1-2, 493–517.
18
[AS] M. Abouzaid and P. Seidel, An open string analogue of Viterbo functoriality, Geom. Topol. 14 (2010), no. 2,
627–718.
[B] R. Berger, Gerasimov’s theorem and N -Koszul algebras, J. Lond. Math. Soc. (2) 79 (2009), no. 3, 631–648.
[BD] C. Brav and T. Dyckerhoff, Relative Calabi-Yau structures, Compos. Math. 155 (2019), no. 2, 372–412.
[BGMS] R. Buchweitz, E. L. Green, D. Madsen, and Ø. Solberg, Finite Hochschild cohomology without finite global
dimension, Math. Res. Lett. 12 (2005), no. 5-6, 805–816.
[CDGG] B. Chantraine, G. Dimitroglou Rizell, P. Ghiggini, and R. Golovko, Geometric generation of the wrapped
Fukaya category of Weinstein manifolds and sectors, 2017. https://arxiv.org/pdf/1712.09126.
[CF] K. Cieliebak and U. Frauenfelder, A Floer homology for exact contact embeddings, Pacific J. Math. 239
(2009), no. 2, 251–316.
[CFO] K. Cieliebak, U. Frauenfelder, and A. Oancea, Rabinowitz Floer homology and symplectic homology, Ann.
Sci. Éc. Norm. Supér. (4) 43 (2010), no. 6, 957–1015.
[D] V. Drinfeld, DG quotients of DG categories, J. Algebra 272 (2004), no. 2, 643–691.
[E] A. Efimov, Homotopy finiteness of some DG categories from algebraic geometry, J. Eur. Math. Soc. (JEMS)
22 (2020), no. 9, 2879–2942.
[ELS] A. Elagin, V. Lunts, and O. Schnürer, Smoothness of derived categories of algebras, Mosc. Math. J. 20
(2020), no. 2, 277–276.
[Gan1] S. Ganatra, Symplectic cohomology and duality for the wrapped Fukaya category, 2012. Ph.D. Thesis, MIT.
http://arXiv.org/abs/1304.7312 .
[Gan2] , Automatically generating Fukaya categories and computing quantum cohomology, 2016.
http://arXiv.org/abs/1605.07702.
[Gan3] , Cyclic homology, S 1 -equivariant Floer cohomology, and Calabi-Yau structures, 2019.
https://arxiv.org/abs/1912.13510.
[GPS1] S. Ganatra, J. Pardon, and V. Shende, Microlocal Morse theory of wrapped Fukaya categories, 2019.
https://arxiv.org/abs/1809.08807.
[GPS2] , Sectorial descent for wrapped Fukaya categories, 2019. https://arxiv.org/abs/1809.03427.
[GPS3] , Covariantly functorial wrapped Floer theory on Liouville sectors, Publ. Math. Inst. Hautes Études
Sci. 131 (2020), 73–200. https://doi.org/10.1007/s10240-019-00112-x.
[G] Y. Gao, Functors of wrapped Fukaya categories from Lagrangian correspondences, 2017.
https://arxiv.org/pdf/1712.00225.
[K] T. Kragh, Parametrized ring-spectra and the nearby Lagrangian conjecture, Geom. Topol. 17 (2013), no. 2,
639–731. With an appendix by Mohammed Abouzaid.
[LM] V. Lyubashenko and O. Manzyuk, Quotients of unital A∞ -categories, Theory Appl. Categ. 20 (2008), No.
13, 405–496.
[LO] V. Lyubashenko and S. Ovsienko, A construction of quotient A∞ -categories, Homology, Homotopy Appl. 8
(2006), no. 2, 157–203.
[PSS] S. Piunikhin, D. Salamon, and M. Schwarz, Symplectic Floer-Donaldson theory and quantum cohomology,
Contact and symplectic geometry (Cambridge, 1994), 1996, pp. 171–200.
[Re] S. Rezchikov, Generalizations of Hodge-de-Rham degeneration for Fukaya categories. In preparation.
[Ri1] A. F. Ritter, Topological quantum field theory structure on symplectic cohomology, J. Topol. 6 (2013), no. 2,
391–489.
[Ri2] , Floer theory for negative line bundles via Gromov-Witten invariants, Adv. Math. 262 (2014), 1035–
1106.
[RS] A. F. Ritter and I. Smith, The monotone wrapped Fukaya category and the open-closed string map, Selecta
Math. (N.S.) 23 (2017), no. 1, 533–642.
[S1] P. Seidel, Fukaya categories and deformations, Proceedings of the International Congress of Mathematicians,
Vol. II (Beijing, 2002), 2002, pp. 351–360.
[S2] , A∞ -subalgebras and natural transformations, Homology, Homotopy Appl. 10 (2008), no. 2, 83–114.
[S3] N. Sheridan, Formulae in noncommutative Hodge theory, J. Homotopy Relat. Struct. 15 (2020), no. 1, 249–
299.
[Sh] D. Shklyarov, Hirzebruch-Riemann-Roch-type formula for DG algebras, Proc. Lond. Math. Soc. (3) 106
(2013), no. 1, 1–32.
[SS] A. A. Sharapov and E. D. Skvortsov, Hochschild cohomology of the Weyl algebra and Vasiliev’s equations,
Lett. Math. Phys. 107 (2017), no. 12, 2415–2432.
[Sy] Z. Sylvan, Orlov and Viterbo functors in partially wrapped Fukaya categories, 2019.
https://arxiv.org/abs/1908.02317.
[V] C. Viterbo, Functors and computations in Floer homology with applications. I, Geom. Funct. Anal. 9 (1999),
no. 5, 985–1033.
19