Orientations for DT invariants on quasi-projective Calabi–Yau 4-folds
Arkadij Bojko∗
Abstract For a Calabi–Yau 4-fold (X, ω), where X is quasi-projective and ω is a nowhere vanishing section of its canonical bundle KX , the (derived) moduli stack of compactly supported perfect complexes MX is −2-shifted symplectic and thus has an orientation ω bundle O → MX in the sense of Borisov–Joyce [11] necessary for defining Donaldson– Thomas type invariants of X. We extend first the orientability result of Cao–Gross– Joyce [15] to projective spin 4-folds. Then for any smooth projective compactification Y , such that D = Y \X is strictly normal crossing, we define orientation bundles on
the stack MY ×MD MY and express these as pullbacks of Z2-bundles in gauge theory, constructed using positive Dirac operators on the double of X. As a result, we relate the ω orientation bundle O → MX to a gauge-theoretic orientation on the classifying space of compactly supported K-theory. Using orientability of the latter, we obtain orientability of MX . We also prove orientability of moduli spaces of stable pairs and Hilbert schemes of proper subschemes. Finally, we consider the compatibility of orientations under direct sums.
1 Introduction
1.1 Background and results Donaldson-Thomas type theory for Calabi–Yau 4-folds has been developed by Borisov– Joyce [11], Cao–Leung [22] and Oh–Thomas [73] . The construction relies on having Serre duality Ext2(E•,E•) =∼ Ext2(E•,E•)∗ . for a perfect complex E•. This gives a real structure on Ext2(E•,E•), and one can take 2 • • its real subspace Ext (E ,E )R. Borisov–Joyce [11, Definition 3.25] define orientations as a square root of the natural isomorphism arXiv:2008.08441v3 [math.AG] 1 Jul 2021 2 ∼ det(LX ) = OX ∗Address: Mathematical Institute, Andrew Wiles Building, Radcliffe Observatory Quarter, Woodstock Road, Oxford, OX2 6GG, U.K., E-mail: [email protected]
1 for any −2-shifted symplectic derived scheme X. On moduli stacks of sheaves, this is equiv- 2 • • alent to the orientation of Ext (E ,E )R or in the language of [73] to a choice of a class of positive isotropic subspaces.
For the Borisov–Joyce class or Oh–Thomas class to be defined in singular homology, re- −1 spectively Chow homology with Z[2 ] coefficients, one needs to show that these choices can be made continuously. This was proved for compact Calabi–Yau 4-folds in Cao–Gross–Joyce [15]. However, many conjectural formulas have been written down in [17, 18, 19, 20, 21, 23, 25] and [26] for DT4 invariants of non-compact Calabi–Yau 4-folds, where additionally the moduli spaces can be non-compact and one has to use localization (see Oh–Thomas [73, §6] and Cao–Leung [22, §8]). The construction of virtual structure sheaves of Oh–Thomas [73] for non-compact Calabi–Yau 4-folds and its localization also depends on existence of global orientation. Moreover, in string theory the DT4-invariants over Hilbert (respectively Quot- ) schemes of points were studied by Nekrasov [69] and Nekrasov–Piazzalunga [70]. The dependence on signs (orientations) is explained in [70, §2.4.2]. The main goal of this work is to justify these results geometrically by proving ori- entability of the moduli stack MX of compactly supported perfect complexes for any quasi- projective Calabi–Yau 4-fold extending the result of Cao–Gross–Joyce [15]. We also prove orientability of moduli spaces of stable pairs (as in [24, 21]) and of Hilbert schemes of proper subschemes. Our result is also used by Oh–Thomas [73] to extend their theory to quasi- projective Calabi–Yau 4-folds and allows for the construction of global virtual structure sheaves even over quasi-projective moduli spaces when X is also quasi-projective. Of particular importance are general toric Calabi–Yau 4-folds and due to work of 4 Nekrasov and his collaborators especially C which was considered from a mathematical point of view in Cao–Kool [17] and Cao–Kool–Monavari [20]. Recently. Kool–Rennemo 4 [56] described explicitly the orientations on Quot-schemes of points on C by entirely dif- ferent means giving evidence for our more general result. A Calabi–Yau n-fold (X, Ω) here is a smooth quasi-projective variety X of dimension n over C, such that Ω is a nowhere vanishing (algebraic) section of its canonical bundle KX . An important feature of our results is that we are able to express our orientations on MX in terms of differential geometric ones. This allows us to consider compatibility of ω choices of trivializations of the orientation bundle O → MX under direct sums, which is related to constructing vertex algebras and Lie algebras on the homology of MX as done by Joyce in [49] and used for expressing wall-crossing formulae by Gross–Joyce–Tanaka in [43]. As another consequence we obtain that if Mα is a moduli scheme or stack of sheaves 0 with given compactly supported K-theory class α ∈ Kcs (X), then there is a natural choice of orientations on it up to a global sign coming from differential geometry independent of how many connected components Mα has after fixing a choice of compactification Y of X by a strictly normal crossing divisor in the sense of Definition 2.21.
2 1.2 Content of sections and main theorems In §2 we introduce briefly the language of derived algebraic geometry and recall the results of Pantev–To¨en–Vaqui´e–Vezzosi [74] and Brav–Dyckerhoff [13] about existence of shifted symplectic structures. For any smooth (quasi-)projective variety X, we introduce twisted virtual canonical bundles ΛL for any coherent sheaf L on the moduli stack of (compactly supported) perfect complexes MX . If (X, Θ) is a spin manifold as in Definition 2.10, then the virtual canonical bundle is defined by KMX = ΛΘ, where Θ is the corresponding S theta characteristic. In Definition 2.11, we define the orientation bundle O → MX over the moduli stack of perfect complexes on a projective spin 4-fold X and prove that it is trivializable:
Theorem 1.1 (see Theorem 2.14). Let (X, Θ) be smooth, projective and spin of complex dimension 4, and S O → MX (1.1) top the orientation bundle from Definition 2.11. Let ΓX :(MX ) → CX be as in Definition 2.13 and apply the topological realization functor (−)top from Blanc [8, Definition 3.1] to 1.1 S top top to obtain a Z2-bundle (O ) → (MX ) . There is a canonical isomorphism of Z2-bundles
/ S top ∼ ∗ D+ (O ) = ΓX (OC ) ,
D/ + where OC → CX is the Z2-bundle from Joyce–Tanaka–Upmeier [51, Definition 2.22] ap- plied to the positive Dirac operator D/ + : S+ → S− as in Cao–Gross–Joyce [15, Theorem S 1.11]. In particular, O → MX is trivializable. We describe examples where this can be used to construct orientations on non-compact Calabi–Yau 4-folds using Corollary 2.16. These include
4 a) C .
b) Tot(KV → V ) for every smooth projective variety V , where KV is the canonical line bundle and dimC(V ) = 3.
c) Tot(L1 ⊕ L2 → S) for any smooth projective surface S, where L1,L2 are line bundles ∼ such that L1 ⊗ L2 = KS. d) Tot(E → C) for any smooth projective curve C, where E is a rank 3 vector bundle, 3 such that Λ E = KC . This is a simple generalization of the work of Cao–Gross–Joyce [15], . Instead of trying to answer the question of existence of “spin compactifications” of Calabi–Yau 4-folds, we define in Definition 2.24 for any non-compact Calabi–Yau 4-fold X and its smooth projective compactification Y the moduli stack MY ×MD MY of perfect complexes identified at the divisor D = Y \X which we require to be strictly normal crossing ./ (see Definition 2.21). We define an orientation bundle O → MY ×MD MY depending on
3 extension data ./, which contains the information about a choice of order ord of the smooth SN irreducible divisors in D = i=1 Di. Under the inclusion into the first component
ζ : MX → MY ×MD MY
./ ω the bundle O pulls back to O → MX . This construction mimics the excision techniques from gauge theory as in Donaldson–Kronheimer [30], Donaldson [29] and Upmeier [95] and is more natural. The analog in topology would be considering vector bundles E,F → X+ identified on +, where (X+, +) is the one point compactification of X by the point +. 0 This is known to generate Kcs (X) which is the natural topological analog of compactly supported perfect complexes on X. We collect here the orientability results stated in §2 (see Theorem 2.26 and Theorem 2.28):
Theorem 1.2 (Theorem 2.26). Let (X, Ω) be a smooth Calabi–Yau 4-fold and Y its smooth projective compactification by a strictly normal crossing divisor D. Choose ord and the extension data ./ as in Definition 2.23, then the Z2-bundle
./ O → MY ×MD MY is trivializable. Let CD, CY be the topological spaces of maps from Definition 2.13. Let C DO → CY ×CD CY be the trivializable Z2-bundle from (4.13). If
top Γ:(MY,D) → CY ×CD CY is the natural map from Definition 2.25, then there exists a canonical isomorphism
./ ∗ C ∼ ./ top I :Γ (DO) = (O ) .
As a consequence of this we prove:
Theorem 1.3 (Theorem 2.28). Let (X, Ω) be a Calabi–Yau 4-fold, then the Z2-bundle
ω O → MX (1.2) from Definition 2.9 is trivializable. Moreover, fix a smooth projective compactification Y , such that D = Y \X is a strictly ω top top normal crossing divisor. Let (O ) → (MX ) be the Z2-bundle obtained by applying the topological realization functor (−)top to (1.2), then there is a canonical isomorphism
cs ∗ cs ∼ ω top IY : (ΓX ) (O ) = (O ) ,
cs cs cs + where O → CX is the trivializable Z2-bundle from Definition 2.27, CX = MapC0 (X , +), (BU× 0 cs top cs Z, 0) is the topological space classifying Kcs (X) and Γ :(MX ) → CX is the natural map from Definition 2.27.
4 We argue in Remark 2.29, that with some extra care, the isomorphism IY could be shown to be independent of Y , but leave proof of this for once applications arise. If X is quasi-projective and M is a moduli stack of pairs of the form OX → F , where F is compactly supported, we need a modification, because the complex is not compactly supported. Let E → M be the universal family, then for fixed compactification X ⊂ Y there is a natural isomorphism from Definition 2.30:
∼ ∗ det HomM(E, E) 0 = det HomM(E, E) 0 , where (−)0 denotes the trace-less part and we use the notation HomZ (E,F ) for two perfect ∨ complexes E,F on X × Z to denote π2 ∗(E ⊗ F ), where π2 : X × Z → Z is the projection. 0 To this, there is an associated Z2-bundle O → M. We can now state the result. Theorem 1.4 (Thm. 2.31). Let i : X → Y be a compactification with Y \X = D strictly ¯ ¯ normal crossing. Let η : M → MX ×MD MY be given by [E] →, [E, OY ], where E is the extension by a structure sheaf to the divisor, then there is a canonical isomorphism
η∗(O./) =∼ O0 .
In particular, O0 is trivializable.
Consequentially, if all stable pairs parameterized by M are of a fixed class OX + α, 0 J K where α ∈ Kcs(X), this determines unique orientations up to a global sign for a fixed compactification Y of X. In §3, we recall some background from Cao–Gross–Joyce [15] and their orientability result [15, Theorem 1.11]. We then develop some technical tools for transport along complex determinant line bundles of complex pseudo-differential operators generalizing the work of work of Upmeier [95], Donaldson [29], [30] and Atiyah–Singer [4]. We require such generalizations, because in allowing Y to be any compactification of X, we lose the ability to work with real structures and real line bundles, as there is no analog for the real Dirac operator D/ + on Y . We can still do excisions natural up to contractible choices of isotopies and use these to restrict everything back into X where the operator D/ + exists. In §4.1 we construct the doubled spin manifold Y˜ from X, and its moduli space (topo- ˜ ˜ logical stack) BY,˜ T˜ of connections on pairs of principal bundle P,Q → Y identified on T . This space has a product orientation as defined in Definition 4.2, which corresponds to the product of the orientation Z2-torsors at each pair of connections. The resulting orientation bundle is denoted DO(Y˜ ) and its pullback to pairs of vector bundles on Y identified on D is denoted by DO. ./ The bundles DO and O restricted to holomorphic vector bundles generated by global sections are related in §4.3 as strong H-principal Z2-bundles (see Definition 3.4) by con- structing isotopies (natural up to contractible choices) between the real structures used to obtain them. This tells us that O./ is trivializable and allows us to express compatibility under direct sums in §5. As the theory developed for identifying the Z2-bundles relies on constructing natural isotopies of algebraic and gauge theoretic isomorphisms of complex
5 determinant line bundles, and this part of it works in great generality in any dimension, the author hopes to use it to construct orientation data as in Joyce–Upmeier [52] for any non-compact Calabi–Yau 3-fold. Finally, in §5, we discuss the relations of orientations under direct sums in K-theory. For given extension data ./, we obtain the result in Proposition 5.3, which expresses choices ./ of trivialization of O → MY ×MD MY in terms of orientations on the K-theoretic space 0 CY ×CD CY with the property π0(CY ×CD CY ) = K (Y ∪D Y ). The signs comparing the orientations are more complicated, but restrict to the expected result under the inclusions cs cs ζ : MX → MY ×MD MY and κ : CX → CY ×CD CY : cs cs Theorem 1.5 (Theorem 5.4). Let Cα denote the connected component of CX corresponding 0 cs cs cs ω cs to α ∈ Kcs(X) = π0(CX ) and Oα = O |Cα . There is a canonical isomorphism φ : ω ω ∗ ω cs cs O O → µ (O ) such that for fixed choices of trivialization oα of Oα , we have ω cs ∗ cs cs ∗ cs ∼ cs ∗ cs φ I (ΓX ) oα I (ΓX ) oβ = α,βI (ΓX ) oα+β .
χ¯(α,α)¯χ(β,β)+¯χ(α,β) where the α,β ∈ {±1} satisfy β,α = (−1) α,β and α,βα+β,γ = β,γα,β+γ 0 for all α, β, γ ∈ Kcs(X). Using this, we formulate a version of [51, Theorem 2.27] for the compactly supported cs orientation group ΩX adapted to the compactly supported case. As a result, we obtain that for all three stacks MZ , MY ×MD MY and MX for (Z, Θ) projective spin 4-fold, (X, Ω) a non-compact Calabi–Yau 4-fold and Y a smooth compacti- fication of X, one can construct Joyce’s vertex algebras on their homology as in [49] using the signs from Proposition 5.3 and Remark 2.15. For MX the signs α,β from Theorem 1.5 are used in constructing the vertex algebra, which could be used to express wall-crossing formulas for (localized) DT4 invariants using the framework of Gross–Joyce–Tanaka [43].
Acknowledgments
The author wishes to express his gratitude to Dominic Joyce for his supervision and many helpful suggestions and ideas. We are indebted to Markus Upmeier for patient explanation of his work. We also thank Chris Brav, Yalong Cao, Simon Donaldson, Mathieu Florence, Michel van Garrel, Jacob Gross, Maxim Kontsevich, Martijn Kool, Edwin Kutas, Sven Meinhardt, Hector Papoulias, Yuuji Tanaka, Richard Thomas, Bertrand To¨enand Andrzej Weber. Finally, we would like to thank the referee for invaluable remarks. The author was supported by a Clarendon Fund Scholarship at the University of Oxford.
Notation
dim If E → X is a complex vector bundle, we denote det(E) = Λ C E its complex determinant ∗ dim line bundle. We write det (E) to denote its dual. If E is real we write detR(E) = Λ R E.
6 If L1,L2 → Y are two line bundles we will sometimes use the notation L1L2 = L1 ⊗ L2. ∗ If P : H0 → H1 is an operator between Hilbert spaces, we write P to denote its adjoint.
If D ⊂ X is a divisor, we write E(D) = E ⊗OX OD. If s : V → W is a map of vector bundles V,W → X and Q → X another vector bundle, then we will write s = s ⊗ idQ : V ⊗ Q → W ⊗ Q. an We often do not distinguish between a scheme X over C and its analytification X .
2 Orientation on non-compact Calabi–Yau 4-folds
In this section, we review the definition of orientation on the moduli stack of perfect com- plexes of a Calabi–Yau 4-fold which uses the existence of −2-shifted symplectic structure as defined by Pantev–To¨en–Vaqui´e–Vezzosi in [74]. We introduce some new notions of orien- tation in more general cases which we use together with compactifications of non-compact Calabi–Yau 4-folds to prove that the moduli stack of compactly supported perfect complexes is orientable.
2.1 Moduli stacks of perfect complexes and shifted symplectic structures Here we recall the construction of moduli stacks of perfect complexes that we will be work- ing with and the shifted symplectic structures on them.
Let X be a smooth algebraic variety. Its category Db Coh(X) of complexes of sheaves with coherent cohomologies does not have a moduli stack of its objects in the setting of standard algebraic stacks. Instead, one needs to rely on the methods provided by the theory of higher stacks and derived stacks. For a thorough discussion of these two terms, one can look at To¨enand Vezzosi [94, 91, 93] in the setting of model categories (see Hovey [48] and Hirschhorn [46] ) and Lurie [62] in the setting of ∞-categories (see Lurie [64] or [63]). Both higher stacks and derived stacks over C form the ∞-categories HStaC and DStaC. There exist an inclusion ∞-functor i : HStaC → DStaC and its adjoint truncation ∞-functor t0 : DStaC → HStaC, which relate the two categories. Note that the C-points of the stacks are left invariant under these functors. Let T be a dg-category (for background on dg-categories see Keller [54] and To¨en[90]). To¨en[90] introduces homotopy theory of dg-categories which is then used in To¨enand Vaqui´e [92] to define for T its associated moduli stack as a derived stack MT which classifies the pseudo-perfect dg-modules of T op. This defines a functor from the homotopy category op of dg-categories to the homotopy category of derived stacks M(−) : Ho(dg - Cat) → Ho(DSta). One can compose this with the truncation functor t0 : Ho(DSta) → Ho(HSta) mapping to the homotopy category of higher (infinity) stacks. We denote the composition by M(−) = t0 ◦ M(−) .
7 Definition 2.1. Let X be quasi-projective variety over C, then we use the notation MX ,
MX for MLpe(X), respectively MLpe(X), where Lpe(X) is the dg-category of perfect com- plexes on X. When X is smooth and projective, it is shown by To¨en–Vaqui´ein [92, Corollary 3.29] that MX is a locally geometric derived stack which is locally of finite type. A locally geometric derived stack corresponds to a union of open geometric sub-stacks, and thus it has a well defined cotangent complex LMX in the dg-category of perfect modules on MX denoted by Lpe(MX ).
For X smooth and projective, both MX and MX can be expressed as mapping stacks in the homotopy categories of the ∞-categories DStaC and HStaC. Let PerfC be the derived stack of perfect dg-modules/complexes over C as defined in [91, Definition 1.3.7.5] when applied to complexes of vector spaces over C as the chosen homotopy algebraic geometric context. Taking its truncation PerfC = t0(PerfC), we have
MX = Map(X, PerfC) MX = Map(X, PerfC) . (2.1) Definition 2.2. If X is just a quasi-projective variety that is not smooth or is not projective, then we will denote the mapping stacks using a superscript:
X X M = Map(X, PerfC) M = Map(X, PerfC) . (2.2) X Remark 2.3. One can show that if X is not smooth, then MX will not be equal to M in general, as the former can classify objects of QCoh(X) which are not perfect. When X is a quasi-projective smooth variety over C, then Lpe(X) is no longer a finite type dg-category and its moduli stack MX behaves differently. For any affine scheme Spec(A) ∈ AffC, the Spec(A)-points of MX are families of perfect complexes on X × Spec(A) with proper support in X.
If X is proper, we can use the description of MX as a mapping stack to construct a universal complex on X × MX : If
u : X × MX → PerfC (2.3) is the canonical morphisms for MX as a mapping stack, and U0 is the universal complex on PerfC used by Pantev–To¨en–Vaqui´e–Vezzosi in [74], then one defines the universal complex ∗ UX = u (U0) on X ×MX . When X is quasi-projective and not necessarily proper, we need a different construction of the universal complex. Definition 2.4. Let iX : X,→ Y be a smooth compactification of X and UY ∈ Lpe Y × MY the universal complex. Let
ξ = M ∗ : M → M (2.4) Y (iX ) X Y ∗ be the image of the pullback iX : Lpe(Y ) → Lpe(Y ), then it acts on Spec(A)-points by ∗ the right adjoint of iX × idSpec(A) as follows from its construction in To¨en–Vaqui´e[92,
8 §3.1] and therefore by the pushforward (iX × idSpec(A))∗ of families of compactly supported perfect complexes on X. We define UX → X × MX by
∗ UX = ξY (UY ) . It is independent of the choice of a compactification∗.
Q Q 0 When Z = i∈I Zi, we will use πI0 : Z → i∈I0 Zi for I ⊂ I to denote the projection to I0 components of the product. We use this also for general fiber products. Let Y now be any projective smooth four-fold and L a coherent sheaf on Y and UY ∈ Lpe(Y × MY ) its universal complex. We define
Ext = π (π∗ U ∨ ⊗ π∗ U ⊗ π∗L) , P = ∆∗ Ext . (2.5) L 2,3 ∗ 1,2 Y 1,3 Y 1 L MY L
As pushforward along π2,3 : Y × MY × MY → MY × MY maps (compactly supported) ! perfect complexes in Y to perfect complexes, it has a right adjoint π2,3 by Lurie’s adjoint functor theorem Gaitsgory–Rozenblyum [39, Thm. 2.5.4, §1.1.2], Lurie [64, Cor. 5.5.2.9]. ! ∗ Moreover, π2,3 = π2,3(−) ⊗ KY [4], which gives us the usual Serre duality in families ∼ ∗ ∨ Ext = σ (Ext ∨ )[−4] , (2.6) L (KY ⊗L ) where σ : MY × MY → MY × MY is the map interchanging the factors. Definition 2.5. Let Y be smooth and and L a coherent sheaf on Y , then as (2.5) are perfect, we construct the L-twisted virtual canonical bundle
ΣL = det(ExtL) , ΛL = det(PL) .
Moreover, ΣL, ΛL are Z2-graded with degree given by a map deg(ΛL): MX → Z2, such that
deg(ΣL)|Mα×Mβ ≡ χ(α, β · L) (mod 2) , deg(ΛL)|Mα ≡ χ(α, α · L) (mod 2) .
0 where α, β ∈ K (X) and Mα is the stack of complexes with class E = α. See Definition 3.12 and (5.5) for more details. J K
From the duality (2.6), we obtain the isomorphisms
∼ ∗ ∗ ∨ ∗ Σ = σ (Σ ∨ ) , θ : −→ ∨ [−4] , i :Λ −→ Λ ∨ . (2.7) L KX ⊗L L PL P(KX ⊗L ) L L (KX ⊗L )
It follows from [13, Proposition 3.3] that the tangent complex of MX can be expressed as T (MX ) = POX [1] (2.8) We use the term Calabi–Yau manifold in the following sense.
∗Simply choose a common compactification Y ← Y 00 → Y 0 and compare the resulting universal com- plexes.
9 Definition 2.6. A Calabi–Yau n-fold is a pair (X, Ω), where X is a smooth quasi-projective variety of dimension n over C and Ω is a global non-vanishing algebraic section of the canonical bundle KX of X. In this case, we use the notation KMX = ΛOX . Then applying the above to L = OX , we obtain the isomorphisms ω θ : T(MX ) → L(MX )[−2] , (2.9) iω : K → K∗ . (2.10) MX MX Brav and Dyckerhoff prove in [13, Proposition 5.3] and [13, Theorem 5.5 (1)] that the ω isomorphism θ comes from a −2-shifted symplectic form ω on MX .
This extra condition to have a −2-shifted symplectic stack (MX , ω) is necessary for constructing Borisov–Joyce fundamental classes as in [11] and Oh–Thomas classes from [73].
2.2 Orientation bundles on moduli stacks of perfect complexes In this subsection, we review the known results of orientation on compact Calabi–Yau 4- folds, and we define new orientation bundles which we will use in later sections to prove the orientability for the non-compact case.
There is a different but equivalent approach to constructing a Z2-bundle other than taking a real line bundle (or a complex line bundle with a real structure) and its associated orientation bundle. Definition 2.7. Let L → X be a complex line bundle over some variety, stack or a topo- logical space. Additionally, let us assume, that there is an isomorphism I : L → L∗. Then consider the adjoint isomorphism J which is the composition of
idL⊗I ∗ L ⊗ L −−−−→ L ⊗ L → C , where the second isomorphism is the canonical one. Then one can define the square root I Z2-bundle associated with I denoted by O . This bundle is given by the sheaf of its sections (which we denote the same way): I ∼ O (U) = {o : L|U −→ CU : o ⊗ o = J} . As this will be important for the proof of Theorem 2.26, we mention here how this construction is related to the one using real structures. Remark 2.8. Let I : L → L∗ be an isomorphism and h−, −i a metric on L. There is the induced isomorphism µ0 : L → L¯ such that h−, µ0(l)i = I(l) for all l ∈ L. As an anti- linear endomorphism, we take its second power (µ0)2 which is given by a multiplication by a 0 √µ strictly positive real function s on X. Then µ = s is a well defined real structure on L. Let Lµ denote the real line bundle of fixed points and or(Lµ) its associated orientation bundle, I then it is easy to see that or(Lµ) is canonically isomorphic to O . For a nice exposition of orientations on O(n, C) bundles see Oh-Thomas [73].
10 We recall the definition of orientations on −2-shifted derived stacks (see Borisov–Joyce [11, Definition 2.12]), which can now be applied also to non-compact Calabi–Yau 4-folds.
ω Definition 2.9. Let (S, ω) be a −2-shifted symplectic derived stack. Let θ : T(S) → L(S)[−2] be the isomorphism associated to the −2 shifted symplectic form. Taking the ω ∗ determinants and inverting, one obtains the isomorphism i : det(LS) → det(LS) . The ω ω orientation bundle O → S is the square root Z2 bundle associated to i .
Let (X, ω) be a Calabi–Yau 4-fold as in Definition 2.6, then we have KMX = det(LMX ). The isomorphisms θω and iω from (2.9) and (2.10) are associated to the −2-shifted sym- plectic derived stack from Definitions 2.1 and 2.6. We denote in this case the orientation ω bundle by O → MX .
From now now on we will restrict ourselves to working only in higher stacks HStaC. By this we mean that we take truncations S = t0(S) everywhere and restrict bundles and their isomorphism constructed on S to S by the canonical inclusion S,→ S. When X is a compact Calabi–Yau 4-fold, Cao–Gross–Joyce [15, Theorem 1.15] prove ω that O → MX is trivializable. One could generalize their result by replacing the require- ment of X being Calabi–Yau by a weaker one.
Definition 2.10. Let X be a smooth projective variety and KX its canonical divisor class. A divisor class Θ, such that 2Θ = KX is called a theta characteristic. We say that (X, Θ) for a given choice of a theta characteristic Θ is spin.
Using this, we construct orientation for the case when X is smooth projective and spin.
Definition 2.11. When X is spin and a choice of Θ is made, we will also use the notation
KMX = ΛΘ. Remark 2.12. A choice of Θ is equivalent to a choice of spin structure on Xan (see Atiyah [5, Proposition 3.2]).
Before we state the generalization of Cao–Gross–Joyce [15, Theorem 1.15] to projective spin varieties, we recall some terminology used to formulate it. Blanc [8] and Simpson [83], define the topological realization ∞-functor: (−)top : an HStaC → Top as the simplically enriched left Kan extension of the functor (−) : AffC → Top , which maps every finite type affine scheme over C to its analytification.
Z Definition 2.13. Let Z be a projective variety over C. Let M , be the mapping stack Z top from (2.2). Let uZ : Z × M → PerfC be the canonical map. Applying (−) and using Blanc [8, 4.2], we obtain (u )top : Zan ×(MZ )top → BU × , where BU = lim BU(n). § Z Z −→n→∞ This gives us Z top an ΓZ :(M ) → MapC0 (Z ,BU × Z) , by the universal property of MapC0 (−, −), where MapC0 (−, −) denotes the mapping space bifunctor in Top. For any topological space T we use the notation CT = MapC0 (T,BU ×Z) .
11 S Theorem 2.14. Let (X, Θ) be spin with the orientation bundle O → MX . Let ΓX : top top S top (MX ) → CX be as in Definition 2.13 and apply (−) to obtain a Z2-bundle (O ) → top (MX ) . There is a canonical isomorphism of Z2-bundles / S top ∼ ∗ D+ (O ) = ΓX (OC ) ,
D/ + where OC → CX is the Z2-bundle from Joyce–Tanaka–Upmeier [51, Definition 2.22] ap- plied to the positive Dirac operator D/ + : S+ → S− as in Cao–Gross–Joyce [15, Theorem S 1.11]. In particular, O → MX is trivializable by the aforementioned theorem. Proof. This is a simple generalization of the proof of [15, Theorem 1.15] relying on the fact that Theorem 3.10 requires Xan to be a spin manifold to trivialize the orientation bundle on BX . We mention it, as we hope that it will be useful in the future for generalization of DT4 invariants. We only discuss the corresponding real structure on the differential geometric side replacing [15, Definition 3.24]. We have the pairing ∧S :(A0,k ⊗ Θ) ⊗ (A0,4−k ⊗ Θ) → A4,4 , and the corresponding spin Hodge star ?S : A0,q ⊗ Θ → A0,n−q ⊗ Θ
S S 0,k ∞ (β ⊗ t) ∧ ?k (α ⊗ s) = hβ ⊗ s, α ⊗ tiΩ α, β ∈ A , s, t ∈ Γ (Ω) , 4,4 S 0,even where Ω ∈ A is the volume form. As a result, we have the real structures: #1 : A ⊗ 0,even S 0,odd 0,odd S q S Θ → A ⊗ Θ and #2 : A ⊗ Θ → A ⊗ Θ, where again #1 |A0,2q⊗Θ = (−1) ? S q+1 S S and #2 |A0,2q+1⊗Θ = (−1) ? . The Dolbeault operator commutes with these DΘ ◦ #1 = S #2 ◦DΘ and its real part is the positive Dirac operator D/ : S+ → S− by Friedrich [35, §3.4]. As twisting by connections only corresponds to tensoring symbols of operators by identity, this extends also to real structures on det(D∇End(E) ).
Remark 2.15. Note that one can also state the equivalent of Cao–Gross–Joyce [15, Theo- rem 1.15 (c)], expressing the comparison of orientations under direct sums on MX in terms of the comparison on CX . Suppose, that X is Calabi–Yau and that there exists Y smooth with an open embedding iY : X,→ Y , where Y is spin. Recall that we have the map ξY : MX → MY from Definition 2.4. We say that Y is a spin compactification of X. We now state the weaker result about orientability for non-compact Calabi–Yau 4-folds. Corollary 2.16. Let X be a Calabi–Yau 4-fold, and let Y be a spin compactification of ∼ X with a choice of Θ and an isomorphism φ : OX −→ Θ|X , then there exists an induced isomorphism of Z2 bundles on MX : ω ∼ ∗ Θ O = ξY (O ) .
In particular, MX is orientable.
Proof. Let E be a perfect complex on X with a proper support, then the Z2 torsors over a Spec(A)-point [E] of both of the above Z2-bundles are given by ∼ ω {oE : det Hom(E,E) −−→ C s.t. oE ⊗ oE = ad(i )|[E]}
12 where iω is the Serre duality, and we used the isomorphism E ⊗ Θ =∼ E induced by φ. Thus we have a natural identification of both Z2-bundles. The last statement follows from Theorem 2.14.
Remark 2.17. If Y is a spin compactification of X and Y \X = D is a divisor. Let N D = ∪i=1Di be its decomposition into irreducible components. If we can write the canonical PN divisor of Y as KY = i=1 aiDi, where ai ≡ 0(mod 2), then one can take
N X ai Θ = D 2 i i=1
1 as the square root. After choosing a meromorphic section Ω¯ 2 of Θ with poles and zeros on ∼ D, one obtains an isomorphism φ : OX −→ Θ|X . Then the condition of Corollary 2.16 is satisfied.
4 4 Example 2.18. The simplest example is C . While its natural compactification P is not 1 3 1 ×4 spin, one can choose to compactify it as P × P or (P ) which are both spin, both of which satisfy the property in Remark 2.17.
Example 2.19. Let S be a smooth projective variety 0 ≤ dimC(S) = k ≤ 4 and let E → S be a vector bundle, s.t. det(E) = KS. Then X = Tot(E → S) is Calabi–Yau. Taking its smooth compactification Y = P(E⊕OS) with the divisor at infinity D = P(E) ⊂ P(E⊕OS), one can show that KY = −(rk(E) + 1)D. If rk(E) ∈ 2Z + 1, we see that we can choose (rk(E)+1) Θ = OY (− 2 D) which satisfies the property of Remark 2.17. Then if rk(E) + k = 4, this is an example of Corollary 2.16, when rk(E) = 1, 3. If X = Tot(L1 ⊕ L2 → S) for a smooth projective surface S and its line bundles L1,L2, s.t. L1L2 = KS, then the spin compactification can be obtained as P(L1⊕OS)×SP(L2⊕OS). Example 2.20. Suppose we have a toric variety X (see Fulton [36], Cox [28]) given by a n n fan in the lattice Z ⊂ R . Suppose it is smooth and it contains the natural cone spanned n by (ei)i=1. Define the hyperplanes
n n X Hi = {(x1, . . . , xn) ∈ R : xj = i} . j=1
Then X is Calabi–Yau if and only if all the primitive vectors of rays of the fan lie in H1 and all the cones are spanned by a basis. A simple generalization of this well known statement shows that X is spin if and only if all the primitive vectors lie in H = S H . odd i∈2Z+1 i Starting from a toric Calabi–Yau X, one can compactify X to a projective smooth toric variety Y by adding divisors corresponding to primitive vectors. In general, we will not 2 have spin compactifications: Consider the fan in R with more than 3 primitive vectors in H1, then any compactification will be consecutive blow ups of a Hirzebruch surface at points, with at least one blow up.
13 A common way of constructing Calabi–Yau manifolds is by removing anti-canonical divisors from a Fano manifold. To further illustrate the scarceness of spin-compactifications in even dimension, we study the classification of toric projective Fano fourfolds by Batyrev [6]. Using the condition described above, we can show that there exist only 4 smooth toric 1 3 1 Fano fourfolds with a spin structure. These are PP3 (O ⊕ O(2)), P × P , P × PP2 (O ⊕ O(1)) 1 and 4P corresponding to the polytopes B2,B4,D12 and L8 respectively. Note that there are 123 smooth projective toric Fano fourfolds in total.
To avoid having to answer the question of existence of spin-compactifications, we develop a different general approach in the next section.
2.3 Orientations for a non-compact Calabi–Yau via algebraic excision principle Let (X, Ω) be a Calabi–Yau fourfold, then we fix a compactification Y with D = Y \X a strictly normal crossing divisor, where we are using the following convention.
Definition 2.21. A strictly normal crossing divisor is a normal crossing divisor, which is a union of smooth divisors with transversal intersections (any k-fold intersection is in particular smooth).
Let (X, Ω) be a Calabi–Yau fourfold, then we fix a compactification Y with Y \D a strictly normal crossing divisor. By Hironaka [45, Main Theorem 1], Bierstone–Milman [7] there exists such a compactification by embedding into a projective space and taking ∼ resolutions. Consider triples complexes (E, F, φ), where E,F ∈ Lpe(Y ) and φ : E|D −→ F |D . We will take the difference of the determinants det Hom(E,E) and det Hom((F,F ) and cancel the contributions which live purely on the divisor. One could think of this as an algebraic version of the excision principle defined for complex operators in §3.3. Let us now make the described method more rigorous. Let X, Y and D be as in the paragraph above, then we can write D as the union
N [ D = Di (2.11) i=1 where each Di is a smooth divisor. We require Ω to be algebraic, then there exists a unique meromorphic section Ω¯ of KY , s.t. Ω¯|X = Ω. The poles and zeroes of Ω¯ express KY uniquely PN in the following form KY = i=1 aiDi, where ai ∈ Z. We may write for the canonical line bundle: N N O O ⊗|ki| KY = O(kiDi) = O(sgn(ki)Di) . (2.12) i=1 i=1
Let ND be the free lattice spanned by the divisors Di which we from now on denote by the N elements ei ∈ ND. For a line bundle L = ⊗i=1O(aiDi) we write La, where a = (a1, . . . , aN ).
14 We will also use the notation Lk = KY . Then for a non-zero global section si of O(Di) one has the usual exact sequence
·si 0 La La+ei La+ei ⊗OY ODi 0 .
As all the operations used to define Exta = ExtLa and Pa = PLa in Definition 2.5 are derived, we obtain distinguished triangles
[1] Exta Exta+e Ext Exta[1] , i La+ei ⊗OY ODi (2.13) [1] a a+e a[1] . P P i PLa+ei ⊗OY ODi P By (2.2) both MY and MD can be expressed as mapping stacks Map Y, Perf and i C Map Di, PerfC , respectively. Let incDi : Di → Y be the inclusion, then we denote by ∗ ρi : MY → MDi the morphisms induced by the pullback (incDi ) : Lpe(Y ) → Lpe(Di), where we are using that both stacks are mapping stacks. For each divisor Di we set
La|Di = La,i and
∗ ∨ ∗ ∗ ∗ Exta,i = π2,3 ∗(π1,2 UDi ⊗ π1,3 UDi ⊗ π1La,i) , Pa,i = ∆ Exta,i , Lemma 2.22. We have the isomorphism
∼ ∗ Ext = (ρi × ρi) (Exta+e ,i) , La+ei ⊗OY ODi i ∼ ∗ = ρ ( a+e ,i) . PLa+ei ⊗OY ODi i P i where we use the same notation for the complexes P on MY and MDi .
Proof. For universal complexes UY , UDi on Y × MY , Di × MDi , we have UY |Di×MY = ∗ (idDi × ρi) UDi as follows from the commutative diagram
Y × MY PerfC
incDi ×idMY
idDi ×ρi Di × MY Di × MDi .
For the dual, we also have U ∨| = (id × ρ )∗U ∨ . Thus we have the following Y Di×MY Di i Di
15 equivalences
Ext La+ei ⊗OY ODi ∗ ∨ ∗ ∗ =π2,3 ∗ π1,2(U ) ⊗ π1,3(U) ⊗ π1(La+ei ⊗OY ODi ) ∼ ∨ ∗ =π2,3 ∗ (incDi × idMY ×MY )∗ U ⊗ U ⊗ π1La+ei |Di×MY ×MY ∼ ∗ ∗ ∨ ∗ ∗ =π2,3 ∗ (incDi × idMY ×MY )∗ ◦ (idDi × ρi × ρi) (π1,2(UDi ) ⊗ π1,3(UDi ) ⊗ π1La+ei,i) ∼ ∗ ∗ ∨ ∗ ∗ =π2,3 ∗ (idDi × ρi × ρi) (π1,2(UDi ) ⊗ π1,3(UDi )) ⊗ π1(La+ei,i) ∼ ∗ =(ρi × ρi) (Exta+ei,i) , the first isomorphism is the projection formula [38, Lem. 3.2.4] and the last step follows ∗ ∼ ∗ from the base change isomorphism π2,3 ∗ ◦ (idDi × ρi) = (ρi × ρi) ◦ π2,3 ∗ using Gaitsgory [38, Prop. 2.2.2]† and that the diagram
D × MX × MX D × MD × MD D
MX × MX MD × MD ∗ consists of Cartesian diagrams by the pasting law in ∞-categories. The second formula follows using (ρ × ρ ) ◦ ∆ = ∆ ◦ ρ : M → M × M . i i MY MDi i X Di Di After taking determinants of (2.13), we obtain the isomorphisms
∼ ∗ ∼ ∗ Σa+ei = Σa ⊗ ρi Σa+ei,i , Λa+ei = Λa ⊗ ρi Λa+ei,i) , (2.14)
where we omit writing L. We have the maps iDi : Di → Y , iD : D → Y inducing
D ρi :MY −→ MDi , ρD : MY −→ M , N N sp Y Y sp MD := MDi and ρ = ρi : MY −→ MD i=1 i=1
D Note that we have the obvious map M → MDi induced by the inclusion Di ,→ D. This gives sp sp : MY × D MY = MY,D −→ MY × sp MY = M . (2.15) M MD Y,D
Definition 2.23. For given X, Y as above let Ω¯ be a meromorphic section of KY restricting to Ω. Let ord denote the decomposition of D into irreducible components as in (2.11), which also specifies their order, such that there exist 0 ≤ N1 ≤ N2 ≤ N, such that ai = 0 for 0 < i ≤ N1, ai > 0 for N1 < i ≤ N2 and ai < 0 for N2 < i ≤ N, where ai are the coefficients
†These references are stated for derived stacks. So we should work with derived stacks until we construct the isomorphisms in Definition 2.24, which we can then restrict to its truncation.
16 from (2.12). For the construction, we may assume N1 = 0. We define extension data as the following ordered collection of sections ./= (si,k)i∈{1,...N2}, (tj,l)j∈{N2+1,...N} , si,k : OY → OY (Di) , tj,l : OY → OY (Dj) . 1≤k≤ai 1≤l≤−aj
Q Q −1 such that i∈{1,...N2} si,k j∈{N2+1,...,N}(tj,l) = Ω and si,k, tj,k are holomorphic with zeros 1≤k≤ai 1≤l≤−aj only on Di, resp. Dj.
This leads to a definition of a new Z2-bundle: sp Definition 2.24. On MY,D we have the line bundle
∗ ∗ ∗ LY,D = π1Λ0 ⊗ (π2Λ0) , (2.16)
π1 sp π2 where MY ←−MY,D −→MY are the natural projections. For a fixed choice ./, there is a natural isomorphism
sp ∼ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ϑ./ : LY,D = π1Λ0 ⊗ π1 ◦ ρ (ΛD) ⊗ π1 ◦ ρ (ΛD) ⊗ (π2Λ0) ∼ ∗ ∗ ∗ ∼ ∗ ∗ ∗ ∼ ∗ = π1Λk ⊗ π2(Λk) = π1(Λ0) ⊗ π2(Λ0) = LY,D , (2.17)
sp Here ΛD → MD are line bundles, and we used the commutativity of
sp π2 MY,D MY ρ π1 ρ sp MY MD
∗ in the first step. The bundles ΛD = ΛD,− ⊗ ΛD,+ appear as the result of using chosen si,k to construct isomorphism (2.14) for the first N2 divisors, then ΛD,− is obtained from using −1 tj,k and (2.14). Thus we will have the expressions:
ΛD,+ =Λ PN2−1 ... ⊗ Λ PN2−1 ⊗ ... k− i=1 aiei+(aN2 −1)eN2 ,N2 (k− i=1 aiei),N2
⊗ Λk−(a−1)e1,1 ⊗ ... ⊗ Λk,1
ΛD,− =Λ PN2 ⊗ ...... ⊗ Λ−eN ,N . k− i=1 aiei,N2+1
The second to last step uses (2.7). We define the Z2-bundles by using Definition 2.7:
sp ∗ ./ sp ϑ./ : LY,D →(LY,D) ,Osp → MY,D , ./ ∗ ./ O = sp Osp (2.18)
./ sp where Osp associated to ϑ./ .
17 ./ The important property of the Z2-bundle O is that it is going to allow us to use index theoretic excision on the side of gauge theory to prove its triviality. One should think of the triples [E, F, φ] which are the points in MY,D as similar objects to the relative pairs in [95, Definition 2.5] with identification given in some neighborhood of the divisor D. The ./ Z2-bundle O only cares about the behavior of the complexes in X. top Definition 2.25. Recall that from Definition 2.13 we have the maps ΓY :(MY ) → CY D top and ΓD :(M ) → CD, We define Γ as the composition
top top h top (MY,D) −→ (MY ) ×(MD)top (MY ) −→ C ×h C 'C × C = C . (2.19) Y CD Y Y CD Y Y,D The first map is induced by the homotopy commutative diagram obtained from applying (−)top to the Cartesian diagram
MY,D MY .
D MY M
The second map uses homotopy commutativity of
top D top top (MY ) (M ) (MY ) . ΓY ΓD ΓY
CY CD CY
an an an The final homotopy equivalence is the result of the map (incD) :(D) → Y being a cofibration for the standard model structure on Top. The map CY → CD is a fibration so the homotopy fiber-product is given by the strict fiber-product up to homotopy equivalences.
We now state the theorem which follows from Proposition 4.15 below and is the main tool in proving orientability of MX . Theorem 2.26. For X,Y and D fix ord and the extension data ./ as in Definition 2.24, then the Z2-bundle ./ O → MY,D (2.20) C is trivializable. Let DO → CY,D be the trivializable Z2-bundle from (4.1), then there exists a canonical isomorphism ./ ∗ C ∼ ./ top I :Γ (DO) = (O ) . (2.21) We now reinterpret this result to apply it to the orientation bundle of interest Oω → MX .
18 Definition 2.27. Let ζ : MX → MY,D induced by the map ξY × [0] : MX × SpecC → MY,D. We have the commutative diagram
top ζtop MX MY,D
Γcs Γ (2.22) cs ics CX CY,D , where cs cs κ : CX = CY ×CD {0} ,→ CY,D, + cs + and CY ×CD {0} = MapC0 (X , +), (BU × Z, 0) . The space C = MapC0 (X , +), (BU × X Z, 0) is the classifying space of compactly supported K-theory on X (see Spanier [85], cs 0 Ranicki–Roe [78, §2], May [67, Chapter 21]): π0(CX ) := Kcs(X) . We define cs cs ∗ C O := (κ ) (DO) . (2.23) The following is the first important consequence of Theorem 2.26 and leads to the construction of virtual fundamental classes using Borisov–Joyce [11] or Oh–Thomas [73] using the −2-shifted symplectic structures. It also gives preferred choices of orientations at fixed points up to a global sign when defining/computing invariants using localization as in [73, 22] for moduli spaces of compactly supported sheaves Mα with a fixed K-theory class 0 α ∈ Kcs and for a given compactification Y . ω Theorem 2.28. Let (X, Ω) be a Calabi–Yau fourfold then the Z2-bundle O → MX is trivializable. Moreover, for a fixed choice of embedding X → Y , with D = Y \X strictly normal crossing, there exists a canonical isomorphism cs ∗ cs ∼ ω top I : (ΓX ) (O ) = (O ) . Proof. We prove this in 3 steps: ∗ ./ ∼ ω 1. We have a natural isomorphism ζ (O ) = O : Consider a Spec(A)-point [E] ∈ MX , then at the corresponding point ([E˜], 0) ∈ MY,D, the isomorphism ϑ./ is given by ˜ ˜ ∼ ∗ ˜ ˜ ∼ ∗ ˜ ˜ det(Hom(E, E)) ⊗ C = det (Hom(E, E ⊗ KY )) = det (Hom(E, E)) , ∼ where we use that ΛL|[0] = C, the first isomorphism is Serre duality, and the second one si,k tj,l is the composition of isomorphisms induced by E −−→ E(Di) and E(−Dj) −−→ E. As Ω E is compactly supported in X, these isomorphisms compose into E −→ E ⊗ KY by the ω assumption on ./. Therefore ϑ./|([E˜],0) coincides with i |[E] and their associated Z2-bundles are identified. 2. By Lemma 4.8, we know Ocs, Oω are independent of the choice of ./. We define a family of I(./) for fixed ord:
(2.22) 1. cs ∗ cs ∼ top ∗ ∗ C ∼ top ∗ ./ top ∼ ω top I(./) : (ΓY ) (O ) = (ζ ) ◦ Γ (DO) = (ζ ) ◦ (O ) = (O ) .
19 ∗ Any two choices of si,k differ by C (and same holds for tj,l). Therefore the set of ./ ∗ PN |a |−1 corresponds to (C ) i i which is connected and I(./) does not depend on ./. 3. For simplicity, let us assume we only have two different divisors D1,D2. We then have the isomorphism obtained from applying (2.14) twice
∗ ∼ ∗ ∗ ∗ π1 Λe1+e2 =π1(Λ0)(ρ2 ◦ π1) Λe2,2(ρ1 ◦ π1) Λe1+e2,1 , (2.24) ∗ ∼ ∗ ∗ ∗ π1 Λe1+e2 =π1(Λ0)(ρ1 ◦ π1) Λe1,1(ρ2 ◦ π1) Λe1+e2,2 , (2.25) ∗ ∼ ∗ ∗ ∗ π2 Λe1+e2 =π2(Λ0)(ρ2 ◦ π2) Λe2,2(ρ1 ◦ π2) Λe1+e2,1 , (2.26) ∗ ∼ ∗ ∗ ∗ π2 Λe1+e2 =π2(Λ0)(ρ1 ◦ π2) Λe1,1(ρ2 ◦ π2) Λe1+e2,2 . (2.27)
To show that there is no difference between the chosen two orders, we use the commutative diagram, where all rows and columns fit into distinguished triangles:
∗ P Pe1 ρ1 Pe1,1
∗ (2.28) Pe2 Pe1+e2 ρ1 Pe1+e2,1
∗ ∗ ∗ ρ2 Pe2,2 ρ2(Pe1+e2,2) ρ1,2 Pe1+e2,1,2
We used in the bottom right corner term the restriction ρ1,2 : MY → MD1∩D2 and e +e ,1,2 = . Taking determinants of all four corner terms of the diagram P 1 2 PO(D1+D2)|D1∩D2 (see Knudsen–Mumford [55, Prop. 1]) and pulling back by π1, we get both (2.24) and (2.25) ∗ ∗ deg (ρD ◦π ) ΛO(D )| deg (ρ1◦π1) ΛO(D )| where the latter comes with (−1) 2 1 1 D1 2 D2 . This holds also for (2.26), (2.27). By commutativity of the diagram, we see that choosing the steps (2.24), (2.26) or (2.25), (2.27) we obtain the same as the signs cancel. As this permutes any two divisors, we obtain independence in the general case. From Lemma 4.8, Proposition 4.7 and C Proposition 4.15, DO and I are independent of the order. Note that this should be all con- sp sidered under pull-back by sp : MY,D → MY,D to have natural isomorphism independent of choices on the smooth intersection D1 ∩ D2: ∗ ∗ ∗ ∼ ∗ ∗ ∗ sp π ◦ ρ1,2(Pe1+e2,1,2) = sp π2 ◦ ρ1,2(Pe1+e2,1,2) .
Remark 2.29. If the need arises to show independence of compactification, one can relate two compactifications Y1 ← Y˜ → Y2 by a common one obtained as a blow up of the closure of X,→ Y1 × Y2. Then one could use [79, Thm. 1.2] comparing Hodge cohomologies for locally free sheaves under blow up in hopes of showing independence of the isomorphism I.
Let us discuss another straight-forward consequence of the framework used in Theo- rem 2.26. For (X, Ω) a quasi-projective Calabi–Yau fourfold, let M be a moduli scheme of
20 stable pairs OX → F where F is compactly supported (see [24, 21, 26, 50, 89]) or ideal sheaves of proper subvarieties. To make sense out of Serre duality, generalizing the ap- proach in Kool–Thomas [57, §3] and Maulik–Pandharipande–Thomas [65, §3.2], we choose a compactification Y as in Theorem 2.26.
Definition 2.30. Let E = (OX×M → F) → X × M be the universal perfect complex. Using the inclusion iX : X → Y we obtain the universal complex E¯ = OY ×M → (iX × idM )∗(F) → Y ×M. We have the following isomorphism, where (−)0 denotes the trace-less part: ω ∼ ¯ ¯ ∼ ∗ ¯ ¯ iM : det HomM (E, E)0 = det HomM (E, E)0 = det HomM (E, E ⊗ Lk)0 κ ∼ ∗ ¯ ¯ ∼ ∗ = det HomM (E, E)0 = det HomM (E, E)0 , where κ is constructed using the isomorphisms ¯ ¯ ∼ ¯ ¯ HomM (E, E ⊗ La)0 = HomM (E, E ⊗ La−ei )0 (2.29) ω in each step determined by ./ as in Definition 2.24. The orientation bundle OM → M is ω defined as the square root Z2-bundle associated with iM . Let M¯ be a moduli stack of stable pairs or ideal sheaves on Y of subvarieties with ¯ proper support in X with the projection πGm : M → M which is a [∗/Gm] torsor. We have an inclusion η : M¯ → MY,D given on Spec(A)-points by mapping [E] 7→ ([E,¯ OY ]), where E¯ = iX ∗(E). The following result leads to the construction of virtual fundamental classes when M is compact (assuming one believes the existence of shifted symplectic structures on pairs as in Preygel [76] or Bussi [14]) and preferred choices of orientations up to a global sign at fixed 0 points when using localization for a fixed K-theory class OX + α for α ∈ Kcs(X) and a choice of compactification Y . J K Theorem 2.31. Let (X, Ω) be a quasi-projective Calabi–Yau fourfold and let Y be its com- ω pactification as in Theorem 2.26. Let OM → M be the orientation bundle from Definition 2.30 for M a moduli scheme of stable pairs or ideal sheaves of proper subschemes of X. There is a canonical isomorphism of Z2-bundles π∗ (Oω ) ∼ η∗(O./) . Gm M = ω In particular, OM → M is trivializable. ¯ ∗ ¯ Proof. The universal perfect complex EM¯ on M is given by (idY × πGm ) (E). We have: ∼ ∗ ∼ ∗ γ : det HomM¯ (EM¯ , EM¯ )0 = det HomM¯ (EM¯ , EM¯ ) det HomM¯ (O, O) = η (LY,D) , such that the following diagram of isomorphism commutes:
∗ ∼ ∗ ∗ η (LY,D) η (LY,D) η∗(ϑ./) ∼ γ γ−∗ ∼