Authors: Laura Pertusi, Paolo Stellari
Categories: Article, 14F05, 14J28, 14J45, 18E30, Derived categories, Semiorthogonal decompositions, Torelli theorems
Source: Rendiconti Del Circolo Matematico Di Palermo
We survey some recent results concerning the so called Categorical Torelli problem. This is to say how one can reconstruct a smooth projective variety up to isomorphism, by using the homological properties of special admissible subcategories of the bounded derived category of coherent sheaves of such a variety. The focus is on Enriques surfaces, prime Fano threefolds and cubic fourfolds.
Keywords: Derived categories, Semiorthogonal decompositions, Torelli theorems
During the last decades, derived categories of coherent sheaves on smooth projective varieties have played a special role in algebraic geometry. In particular, their use in birational geometry and for the study of the geometry of moduli spaces has produced important and unexpected results.
One natural and related question is if a smooth projective variety can be reconstructed, up to isomorphism, from its derived category. Due to the seminal work by Bondal and Orlov [29] we know that this is indeed a theorem when the variety has canonical bundle which is either ample or anti-ample (meaning that its dual is ample). On the other hand, Mukai [114] showed that this is no longer the case when the canonical bundle is trivial.
Of course, one may start wondering how one can study the derived category of coherent sheaves and how one can extract geometric information from it. This is a fast growing research area where several fruitful ideas have come into the picture. Important results in this direction are due to the Russian school. The idea is to decompose the derived category in smaller pieces provided by nontrivial admissible subcategories which naturally generate the derived category and whose meaning is intrinsically connected to the geometry of the variety. This led to the notion of semiorthogonal decomposition which is certainly one of the main characters in this survey.
Semiorthogonal decompositions are not always available and when available they are not, in general, canonical. For example, again when the canonical bundle is trivial, the derived category is indecomposable. Nonetheless, when a semiorthogonal decomposition is given, then its components turn out to be extremely interesting. One special case, which is prominent in this paper, is when the derived category Db(X) of a smooth projective variety X contains a bunch of very simple objects, which are called exceptional, and a geometrically meaningful residual category, which we call Kuznetsov component.
In this paper we focus our attention on these components. Indeed, the problem we want to deal with can be now formulated in the following slightly vague
Categorical Torelli problem. Let X1 and X2 be smooth projective varieties over a field, in the same deformation class and with Kuznetsov components Ku(X1) and Ku(X2). Is it true that X1 and X2 are isomorphic if and only if their Kuznetsov components are equivalent?
As we mentioned above, semiorthogonal components are not in general canonical. Thus when such a problem has a positive answer, the corresponding Kuznetsov components have to emerge from very special geometric situations. This will be extensively explained in the examples of interest.
The Categorial Torelli theorems discussed in this paper are indeed the results that provide a positive answer to the Categorical Torelli problem above. As it turns out, we need to be more precise about the equivalence between the Kuznetsov component in the sense that, in some cases, it has to satisfy some additional property. We will discuss this along the paper and discover that some of these assumptions are probably removable once some related conjectures are proved. Others, unfortunately, cannot be avoided. Just to give a short summary, the Categorical Torelli theorems that we will review are the following geometric
One important feature of the above results is the variety of techniques that are used to prove them. Indeed, (CT1) is a consequence of a general statement (see Propsition 4.1) which allows us to extend a Fourier– Mukai equivalence between the Kuznetsov components of two Enriques surfaces to an equivalence of their bounded derived categories and then to apply what we call Derived Torelli theorem. This is somehow related to the possibility to enhance exact functors at the dg level. On the other hand, (CT2) and (CT3) use in an extensive way the notion of stability conditions which only recently were constructed on Kuznetsov components (see Section 6.1). The case of cubic fourfolds (CT4) can be handled either by using Hodge theoretic techniques as in [60] or, again, by using stability conditions [14]. Both approaches will be discussed in Section 7.2. The reader may actually view this survey as an occasion to review the most recent developments in so many different directions and to appreciate their power in combination with the theory of semiorthogonal decompositions.
In a negative direction, the Categorical Torelli theorem in the above formulation does not hold in the case of Fano threefolds of index 1. For instance, a consequence of the results in [90] provides the existence of non-isomorphic but birational Gushel– Mukai threefolds with equivalent Kuznetsov components. A refined version, which takes into account the preservation of special objects in the Kuznetsov components, has been recently proved in [67, 69] (see Section 6.5 for more details). It becomes then natural to ask whether the existence of an isomorphism between the given varieties with equivalent Kuznetsov components in (CT1)–(CT4) is just a special instance of the following more general Birational Categorical Torelli problem. Let X1 and X2 be smooth projective varieties over a field, in the same deformation class and with Kuznetsov components Ku(X1) and Ku(X2). Is it true that if Ku(X1) and Ku(X2) are equivalent, then X1 and X2 are birational?
This will be carefully discussed in the paper but it is worth mentioning that a converse to the Birational Categorical Torelli problem should not hold true (see Remark 7.9).
As a related line of investigation, we recommend the reader to consult [94], where the authors study the problem of recovering the birational class of a smooth projective variety from its bounded derived category. More precisely, they conjecture that two smooth projective varieties are birational if there exists a (strongly) filtered exact equivalence between their bounded derived categories (see [94, Section 10] for some motivic foundations of this conjecture and more generally [73] for reconstruction problems). Another very interesting research topic which is tightly related to the discussion in this paper but which will not be covered in this survey is the so called infinitesimal version of the Categorical Torelli theorems for Fano threefolds (see [67, 68]).
We conclude this presentation by pointing out that the paper is accompanied by a list of open problems in the form either of questions or of conjectures. Their relation to the existing results will be carefully explained, but we take the opportunity to stress that their role in this paper is as important as the one of the main results.
The survey is organized as follows. In Section 2 we recollect some basic definitions and examples of semiorthogonal decompositions, and of Fourier–Mukai functors.
In Section 3 we focus on some special examples of smooth projective varieties of low dimension having a semiorthogonal decomposition with a nontrivial component, known as the Kunzetsov component. In particular, we consider Enriques surfaces, prime Fano threefolds of index 1 and 2 and cubic fourfolds, and we recall the properties of their Kuznetsov component.
Section 4 is devoted to the proof of the Categorical Torelli theorem for Enriques surfaces following [97, 100]. The proof makes use of a general criterion, explained in Section 4.1, which allows us to extend Fourier–Mukai equivalences among admissible subcategories appearing in semiorthogonal decompositions under suitable assumptions. Then in Section 4.2 we characterize 3-spherical objects in the Kuznetsov component; this is used in the proof of the main theorem given in Section 4.3.
Section 5 provides a quick introduction to the notion of (weak) stability conditions, the definition of stability manifold together with the associated group actions, the construction via tilt-stability on surfaces and on some threefolds, and the conjectural approach through generalized Bogomolov inequalities.
Section 6 is devoted to the case of cubic threefolds. In Section 6.1 we explain the first method to construct stability conditions on the Kuznetsov component of a cubic threefold and how to apply this result to prove the Categorical Torelli theorem; the main reference is [20]. In Section 6.2 we review the more recent method, introduced in [14], to induce stability conditions on admissible subcategories which are left orthogonal components to an exceptional collection in a triangulated category with a Serre functor. Then we introduce the notion of Serre-invariant stability conditions. This is applied to cubic threefolds in Section 6.3 to construct stability conditions on the associated Kuznetsov component which are Serre-invariant. In Section 6.4 we explain some applications of this result on Serre-invariant stability conditions to the study of the geometry of moduli spaces and to give an alternative proof of the Categorical Torelli theorem; the main references are [8, 51, 129]. In Section 6.5 we recall the state of art about these questions on Serre-invariant stability conditions and Categorical Torelli theorem for the Kuznetsov component of prime Fano threefolds of index 2 and 1.
Finally, in Section 7 we analyze the higher dimensional case of cubic fourfolds. We recall the construction of stability conditions on the Kuznetsov component with the method of [14]. Then we explain the two known ways to prove the Categorical Torelli theorem from [60] and [14], and how to deduce the Classical Torelli theorem from it. We end by discussing the analogous questions in the case of Gushel–Mukai fourfolds.
All varieties and schemes appearing in this paper are defined over an algebraically closed field K. All categories are linear over such a field K as well. Depending on the results we are going to discuss, we will need to make further either that char(K) is different from 2 or sufficiently large or equal to 0. We tried to point this out along the paper in the most careful way. But it is certainly true that the nonexpert reader might get confused at first sight. Such a reader is then encouraged to read the paper assuming that K is algebraically closed of characteristic 0 and come back to the discussion about the possible more general statements only during a second read.
This section is devoted to a quick discussion about some basic facts concerning semiorthogonal decompositions (see Section 2.1) and Fourier– Mukai functors (see Section 2.3). We also recall some preliminary examples in Section 2.2.
Despite its relatively simple definition and even if it is one of the simplest examples of triangulated category, the bounded derived category Db(A) of an abelian category A has a very rich and often mysterious structure. This remains true when A is the category Coh(X) of coherent sheaves on a smooth projective variety X defined over a field K.
In the latter case, there are several approaches to the study of the structure of Db(X):=Db(Coh(X)):
The first strategy has a long history, initiated by the beautiful paper [31] (see also [132]). More recently, it was discovered that this is intimately related to the way Db(X) can be enhanced to higher categorical structures (see, for example, [28, 34, 39, 105]).
The second viewpoint has been widely adopted in the case of K3 surfaces, abelian varieties or, more generally, varieties with trivial canonical bundle (see, for example the seminal papers [116, 121]). The reason being that, when the canonical bundle is trivial, Db(X) is indeed indecomposable by [23] and thus the third strategy cannot be pursued. On the contrary, the autoequivalences group is usually very rich and intimately related to the topology of the stability manifold which we will discuss later.
In this paper, we are mainly interested in the third approach which can be made precise by introducing the following definition. Let T be a triangulated category which, for simplicity, we assume from now on to be linear over a field K. A semiorthogonal decomposition for T, denoted by
is a sequence of full triangulated subcategories D1,⋯,Dm of T such
We call the subcategories Di components of the decomposition.
It is a nice and relatively easy exercise to verify that (1) above yields that the factors πi(F) in (2) are uniquely determined and functorial, for all F∈T and for all i=1,…,m. Hence, in presence of a semiorthogonal decomposition, we get the i-th projection functor πi:T→Di.
Given a semiorthogonal decomposition for T as above, denote by αi:Di↪T the inclusion. We say that Di is admissible if αi has left adjoint αi∗ and right adjoint αi!. In presence of a semiorthogonal decomposition
then π1 and π2 coincide with the left adjoint α1∗ and the right adjoint α2!. Furthermore, if D is an admissible subcategory of T, we set
to be the left orthogonal and right orthogonal subcategories of D, respectively. These triangulated subcategories yield semiorthogonal decompositions
when D is admissible.
One special feature of the triangulated category T=Db(X), for X a smooth projective variety, is that it has Serre functor. Recall that a Serre functor of a triangulated category T is an exact autoequivalence ST:T→T inducing, for all A and B in T, an isomorphism
which is natural in both arguments. Such a functor is unique up to isomorphism of exact functors and thus we will refer to ST as ‘the’ Serre functor of T.
If X is a smooth projective variety, then the Serre functor SX:=SDb(X) takes the following explicit form
where ωX is the dualizing sheaf of X. If α:D↪Db(X) is an admissible subcategory, then
The latter is a general if T has Serre functor ST, and it contains an admissible triangulated subcategory D, then D has a Serre functor SD as well. The shape of SD is exactly the one above with SX replaced by ST (see [27]).
Recall that T is K-linear. An object E∈T is exceptional if Hom(E,E[p])=0, for all integers p≠0, and Hom(E,E)≅K. A set of objects {E1,…,Em} in T is an exceptional collection if Ei is an exceptional object, for all i, and Hom(Ei,Ej[p])=0, for all p and all i>j. An exceptional collection {E1,…,Em} is
Finally, assume that T is a proper K-linear triangulated category. This means that
for all F and G in T. Its numerical Grothendieck group N(T) is defined as the quotient
Here χ denotes the Euler form on K(T) defined by
and K(T) stands for the Grothendieck group of T, which is the free abelian group generated by isomorphism classes [F] of objects F∈D modulo the relation [F]=[E]+[G] for every exact triangle E→F→G. Note that K(-) and N(-) are additive with respect to semiorthogonal decompositions.
In this section we discuss the first examples of semiorthogonal decompositions of the derived categories of simple smooth projective varieties.
(Points and exceptional objects) It is an easy exercise to show that if E is an exceptional object in a triangulated category T, then the smallest full triangulated subcategory ⟨E⟩ of T containing E is equivalent to the bounded derived category of a point. On the other hand, if T is a proper triangulated category and E∈T is exceptional, then ⟨E⟩ is an admissible subcategory of T (see, for example, [111, Proposition 2.6]).
In higher dimension we have two important and classical examples.
(Projective spaces) In the case of the n-dimensional projective space Pn, a classical result of Beilinson [18] shows that the set of line bundles
forms a full exceptional collection and so it yields a semiorthogonal decomposition
It should be noted that the collection (2.3) is also full and strong.
(Quadrics) Assume now that Q is an n-dimensional smooth quadric in Pn+1 defined by an equation {q=0}. We assume char(K)≠2. According to [71], the category Db(Q) has a semiorthogonal decomposition by exceptional bundles whose explicit form depends on the parity of n. More precisely, if n=2m+1 is odd,
where S is the spinor bundle on Q defined as coker(ϕ|Q)(-1) and ϕ:OPn+1(-1)2m+1→OPn+12m+1 is such that ϕ∘(ϕ(-1))=q·Id:OPn+1(-2)2m+1→OPn+12m+1.
If n=2m is even, then we get
where S-:=coker(ϕ|Q)(-1), S+:=coker(ψ|Q)(-1), and ϕ,ψ:OPn+1(-1)2m→OPn+12m are such that ϕ∘(ψ(-1))=ψ∘(ϕ(-1))=q·Id. See [124] for more details on spinor bundles.
We now briefly recall how to define special classes of exact functors between admissible subcategories. The reader can have a look to [36] for a survey or to [55] for an extensive treatment. Note that all functors are derived.
Let X1 and X2 be smooth projective varieties over a field K with admissible embeddings
for i=1,2.
An exact functor F:D1→D2 is of Fourier– Mukai type (or a Fourier– Mukai functor) if there exists E∈Db(X1×X2) such that the composition α2∘F is isomorphic to the restriction
Here the exact functor ΦE is given by
where pi is the i-th natural projection.
By [79, Theorem 7.1], the projection functor onto an admissible subcategory D↪Db(X) is of Fourier– Mukai type. This motivates [85, Conjecture 3.7] which says that any exact equivalence F:D1→D2 between admissible subcategories Di↪Db(Xi), for Xi smooth projective over K, is of Fourier– Mukai type. Indeed, [79, Theorem 7.1] is a special case of this conjecture for F=id.
Here we propose the following
Is any exact fully faithful functor F:D1→D2 between admissible subcategories Di↪Db(Xi), for Xi smooth projective over K, of Fourier– Mukai type?
This is motivated by Orlov’s result that any fully faithful exact functor F:Db(X1)→Db(X2) is of Fourier– Mukai type when Xi is smooth projective over a field K (see [121]). It should be noted that in [35, 38] the assumptions on F were weakened. In particular, assuming full is enough. More recently, [120] extended Orlov’s result to the smooth and proper case. Note that Question 2.7 is related to [85, Conjecture 3.7].
Motivated by the recent work [39, 105], we can actually state a weaker version of Question 2.7:
Let F:D1→D2 be an exact equivalence between admissible subcategories Di↪Db(Xi), for Xi smooth projective over K. Is there a Fourier– Mukai equivalence D1≅D2?
The problem above is mainly motivated by the recent developments about uniqueness of enhancements [105, Corollary 9.12] (see also [39, Section 7.2] and the recent improvements in [34]). The idea is that, given an equivalence F:Perf(X1)→Perf(X2) between the categories of perfect complexes on quasi-compact and quasi-separated schemes, one can replace F with another equivalence which can be lifted to any dg model of Perf(Xi). Under additional suitable assumptions on Xi (e.g. if Xi is noetherian), the latter condition is equivalent to being of Fourier– Mukai type, due to [106, 143].
We expect Question 2.7 and Question 2.8 to have negative answers in general also due to the results in [34, Section 6.4] (see, in particular, Corollary 6.12 and Remark 6.13 there). We will discuss later when one can positively answer them in the geometric settings we are interested in.
In this section we would like to go beyond Example 2.4 and Example 2.5 and consider semiorthogonal decompositions for Db(X) for special but very interesting smooth projective varieties. In most of the examples discussed in this section, the derived category will contain a small set of exceptional objects of geometric origin and a nontrivial admissible subcategory right orthogonal as in (2.1) to the exceptional objects and which we will call Kuznetsov component.
It is well known that if C is a smooth complex projective curve, then Db(C) determines C up to
(Derived Torelli theorem for curves) Let C1 and C2 be smooth complex projective curves. Then Db(C1)≅Db(C2) if and only if C1≅C2.
As it is explained in the proof of [55, Corollary 5.46], the delicate case which requires using the cohomology of the curve and then the classical Torelli theorem is when the genus of the curves is 1. All the other cases follow from the following beautiful result by Bondal and Orlov [29].
(Derived Torelli theorem for (anti)Fano manifolds) Let X1 and X2 be smooth projective varieties with ample or antiample canonical bundle (i.e. either ωXi or ωXi∨ is an ample line bundle). Then Db(X1)≅Db(X2) if and only if X1≅X2.
Apart from the case of P1 which is covered by Example 2.4, in genus greater than 0 the triangulated category Db(C) does not have nontrivial semiorthogonal decompositions by [119].
Note that Theorem 3.1 remains true over any algebraically closed field K when the genus of the curves is not 1. This is because Theorem 3.2 holds in this more general setting. On the other hand, when the genus is 1, the assumption K=C cannot be removed as there are nonisomorphic smooth projective curves of genus 1 with equivalent derived categories (see [4, 139]).
Let K be an algebraically closed field of characteristic different from 2. An Enriques surface is a smooth projective surface X defined over K such that H1(X,OX)=0 and the dualizing line bundle ωX is nontrivial but 2-torsion. An Enriques surface X can be equivalently characterized as a quotient of a K3 surface by an involution acting without fixed points.
The derived category Db(X) of an Enriques surface determines the surface up to isomorphism in view of the following result which is a rewriting of [26, Proposition 6.1] and [54, Theorem 1.1]:
(Derived Torelli theorem for Enriques surfaces) Let X1 and X2 be smooth projective surfaces defined over an algebraically closed field K of characteristic different from 2. If X1 is an Enriques surface and there is an exact equivalence Db(X1)≅Db(X2), then X1≅X2.
Clearly the situation becomes much more involved when K has characteristic 2. In this case the definition has to be slightly an Enriques surface is a minimal smooth projective surface whose canonical bundle is numerically trivial and such that the second Betti number is 10. If char(K)≠2, this definition coincides with the one above. But when char(K)=2, then one gets three
In the first case the canonical bundle is nontrivial and 2-torsion while, in the latter two cases, the canonical bundle is trivial. Singular Enriques surfaces are again realized as quotients of K3 surfaces. The reader can have a look at [44] and [48] for an extensive treatment of Enriques surfaces and [47] for a shorter but informative one.
It is then natural to raise the following
Is Theorem 3.4 still true when char(K)=2 for some/all of the three families above?
It is clear that we should not expect an analogue of Theorem 3.4 to hold for all smooth projective surfaces. Indeed, already for abelian surfaces [114] and K3 surfaces [118, 140], this is known to be false.
For the rest of this section we stick to the case where K is algebraically close and char(K)≠2. We can further analyze Db(X) by means of the following result which is certainly well-known to experts.
([97, Proposition 3.5]) Let X be an Enriques surface over K as above. Then Db(X) contains an admissible subcategory L=⟨L1,⋯,Lc⟩, where L1,⋯,Lc are orthogonal admissible subcategories and
Here:
We can illustrate the geometry attached to the collection L in some interesting cases.
A generic Enriques surface does not contain (-2)-curves. Thus the collection in Propsition 3.7 gets much simplified. In particular, with the above notation, we have ni=1 for every i=1,⋯,c, and we get 10 completely orthogonal blocks Li=⟨Li⟩, where Li:=L1i is an exceptional line bundle, for all i=1,⋯,10.
If K=C and X does not contain (-2)-curves, then [146] gives a very geometric interpretation of these orthogonal line bundles. Indeed, any ample polarization on X of degree 10 yields 10 elliptic pencils each containing 2 double fibers. Denote them by Fi+ and Fi-. Then we can take Li:=OX(-Fi+). One can prove that, for all i, we have the relation Fi+=Fi-+KX, where KX is the canonical class. Using the Serre functor (see Example 2.2), one immediately sees that it is possible to change any Li to OX(-Fi-) and still get a completely orthogonal collection of 10 line bundles. In particular, Db(X) contains many distinct collections of exceptional objects, as we have at least 210=1024 possible choices of orthogonal exceptional collections of line bundles in Db(X) (see [97, Example 3.4] for a more detailed discussion).
Thus, if X is an Enriques surface and L is a collection of exceptional line bundles as in Propsition 3.7, then we get a semiorthogonal decomposition
The admissible subcategory Ku(X,L):=L⊥ is referred to as the Kuznetsov component of X. As the notation suggests, it is important to keep in mind that Ku(X,L) depends on L and not just on X.
The Kuznetsov component is, at the moment, quite a mysterious subcategory. On one hand, it is certainly easy to show that it is nonzero. Indeed, its numerical Grothendieck group can be easily
Furthermore, if
denotes the embedding with right adjoint κ!, then the object
is nontrivial in Ku(X,L), for L1i the line bundles in Propsition 3.7. On the other hand, three potentially interesting and related open problems are summarized by the
(i) Does the Serre functor SKu(X,L) have an explicit and computable description (other than the abstract one in Example 2.2)?
(ii) Is Ku(X,L) indecomposable2?
(iii) Does Question 2.7 or Question 2.8, with Di:=Ku(Xi,Li), have a positive answer for the Kuznetsov component Ku(X,L)?
In Section 4.2 we will comment on the action of SKu(X,L) on some special objects of Ku(X,L) and we will mention why a positive answer to Question 3.9 (ii) may be interesting to provide yet another counterexample to the Jordan– Hölder conjecture. A positive answer to Question 3.9 (iii) (in either of the two forms) would yield a simpler statement for the Categorical Torelli theorem discussed later.
As we observed in the introduction, not all surfaces admit nontrivial semiorthogonal decompositions. Already for surfaces of Kodaira dimension 0, K3 and abelian surfaces have indecomposable derived categories. Additional interesting results in dimension 2 are contained in [72, Section 5].
We turn now to the case of Fano threefolds, i.e. smooth projective threefolds X defined over an algebraically closed field K and such that ωX∨=OX(-KX) is ample. We stick to the examples where the rank ρX of the Picard group Pic(X) of X is 1, which are called prime Fano threefolds. Under these assumptions, we denote by HX the primitive ample generator of Pic(X).
The classification of prime Fano threefolds was achieved in [65, 117] in characteristic 0. These classification results have been extended to positive characteristic in [138]. These threefolds are classified by two numerical invariants. The first one is the index which is the positive integer iX such that
The second one is the degree which is the positive integer d:=HX3.
It turns out that iX∈{1,2,3,4}. The cases iX=3 and iX=4 correspond to X=Q and X=P3, respectively, where Q is a 3-dimensional quadric. In both cases, we know all about Db(X) in view of Example 2.4 and Example 2.5 (when char(K)≠2), respectively. Thus we can stick to iX=1,2.
Prime Fano threefolds with index 1 are organized in 10 deformation types. Moreover, the degree of these Fano threefolds dX=2gX-2 is even and the deformation type is characterized by the choice of 2≤gX≤12 but gX≠11, where gX is called the genus of X.
Now, for gX=2,3,4,5 we consider the semiorthogonal decomposition
On the other hand, if gX is even and greater than 4, then we have
where E2 is a rank-2 stable vector bundle on X whose existence is claimed in [115] (see [14, Section 6] for a careful proof of this fact which is valid for algebraically closed fields of characteristic either 0 or sufficiently large). For odd genus gX=7,9, we use again [115] which yields a rank-5 and rank-3 vector bundle E5 and E3 and semiorthogonal decompositions
when X has genus 7 and 9, respectively.
In all the above cases, the residual category Ku(X) is called Kuznetsov component. It is worth pointing out that Ku(X) can be better understood in some interesting cases. What is known is summarized in the following table, where the third column indicates the reference where the semiorthogonal decomposition in the second column is provided. As above, the base field has characteristic either zero or sufficiently large.
In the table, Xd denotes a prime Fano threefold of index 1 and degree d=2g-2, Cg denotes a smooth curve of genus g while Ei refer to vector bundles which are explicitly described in the references in the third column. It is worth to point out that in some cases the semiorthogonal decomposition of Db(X6) can be refined. More precisely, note that X6 is a complete intersection of a quadric hypersurface and a cubic hypersurface in P5. When the quadric cutting X6 is smooth, there is a semiorthogonal decomposition
where S is the restriction to X of a spinor bundle on the quadric.
Prime Fano threefolds with index 2 are usually referred to as del Pezzo threefolds. They all have a canonical semiorthogonal decomposition
where Ku(X) is, as usual, the Kuznetsov component. The degree dX is subject to the bounds 1≤dX≤5 and in some cases the Kuznetsov component can be further analyzed according to the following
In the table, C2 is a smooth curve of genus 2 and F3,F2 are vector bundles of rank 3 and 2, respectively, described explicitely in the reference in the third column. In the above list, when iX=2 and dX=3 we get the celebrated case of cubic threefolds (i.e. smooth degree 3 hypersurfaces in P4) which are going to be important examples that we will analyze in full detail. The Kuznetsov components of cubic threefolds yield examples of the following special class of triangulated categories.
The following result is going to be relevant later in the
Let X be a cubic threefold. Then the Kuznetsov component Ku(X) in (3.6) is a fractional Calabi– Yau of fractional dimension 53.
The admissible subcategory Ku(X) has Serre functor by Example 2.2. We can then simply apply [81, Corollary 4.4] (see also [81, Corollary 4.3] for a more general statement).
The following question is then natural.
Is Ku(X) indecomposable, when X is a cubic threefold?
A list of other prime Fano threefolds with Kuznetsov component which is a fractional Calabi– Yau category can be deduced from [84] (see Section 2.4 therein) and [91]. One interesting example is provided by quartic threefolds, i.e. smooth hypersurfaces of degree 4 in P4. According to the first table above (ρ=i=1 and g=3), if X is a quartic threefold, we have a semiorthogonal decomposition as in (3.3). In this case, Ku(X) is a fractional Calabi– Yau category of fractional dimension 104. We will comment on this case later.
After discussing the Categorical Torelli theorem for cubic threefolds, we will explain that Question 2.8 has a positive answer for Ku(X), when X is a cubic threefold (see Corollary 6.5).
A cubic fourfold is a smooth cubic hypersurface in P5. Here we assume that X is defined over an algebraically closed field K such that char(K)≠2.
The derived category of a cubic fourfold X has again a natural semiorthogonal decomposition
where H is the class of a hyperplane section of X. The admissible subcategory Ku(X) is the Kuznetsov component of X.
If X is a cubic fourfold, then the Kuznetsov component Ku(X) is an indecomposable 2-Calabi– Yau category for which Question 2.7 has a positive answer.
The fact that Ku(X) is a 2-Calabi– Yau category is again a consequence of [81, Corollary 4.3]. Moreover, it is an indecomposable admissible subcategory due to the well-known argument in [84, Section 2.6]. The fact that Question 2.7 has a positive answer is the content of [98] (based on [34, 37]).
It was observed by Kuznetsov in [80] that, in many cases, one can realize the Kuznetsov component of a cubic fourfold as the derived category of a K3 surface. For example, this happens for Pfaffian cubic fourfolds. Subsequent recent work carried out in [2] and [13] completely classified all cubic fourfolds whose Kuznetsov component is equivalent to the derived category of a K3 surface. This body of work is very much related to the following very influential conjecture (see [80, Conjecture 1.1]) which we state even though it is not directly related to the rest of this paper.
(Kuznetsov) A cubic fourfold X is rational if and only if Ku(X) is equivalent to the bounded derived category of a K3 surface.
None of the two implications is clear but the conjecture perfectly matches the classical Hodge theoretic still conjectural characterization of rational cubic fourfolds due to Harris and Hassett. Some important results in this direction are contained in [133, 134].
As it was first pointed out in [80], there are cubic fourfolds X such that Ku(X) is equivalent to the derived category Db(S,α) of twisted coherent sheaves, where S is a K3 surface and α is an element in the Brauer group Br(S):=H2(S,OS∗)tor of S (see [33, Chapter 1] for an extensive introduction to twisted coherent sheaves and their derived categories). The complete classification of all cubic fourfolds for which this is true was carried out in [13, 57]. In view of Conjecture 3.16 it is certainly interesting to understand which geometric property of X corresponds to having an equivalence Ku(X)≅Db(S,α).
Because of the many similarities with the derived categories of K3 surfaces—only few of which have been discussed here—the Kuznetsov component Ku(X) is an example of so called noncommutative K3 surfaces.
For actual K3 surfaces, the description of the autoequivalences group of their bounded derived categories is a challenging problem (see, for example, [9, 59, 116, 121]). One could of course try to describe the autoequivalences group of a noncommutative K3 surface as well. We will go back to this issue later in the paper. For later use, we content ourselves with the observation from [81] that, for a cubic fourfold X, the category Ku(X) has always an autoequivalence
where ι∗ is the left adjoint of the inclusion ι:Ku(X)↪Db(X) and H is a hyperplane class. Such an autoequivalence is called degree shift functor. It is not difficult to see that, by definition, OX is of Fourier– Mukai type.
Assume now that K=C. It was observed in [2] (see also [111, Section 3.4]) that the Kuznetsov component Ku(X) of a cubic fourfold X is equipped with an even unimodular lattice H~(Ku(X),Z) with a weight-2 Hodge structure induced by the Hodge decomposition of H4(X,C). Such a lattice is usually referred to as the Mukai lattice of X.
For the convenience of the reader let us spell out some details in the construction. First observe that the topological K-theory Ktop(X) of X comes equipped with the pairing
where p:X→pt. As a group H~(Ku(X),Z) is defined as the set of classes in Ktop(X) which are orthogonal, with respect to χ(-,-) to the classes of the three line bundles OX(iH), with i=0,1,2. This is nothing but the topological K-theory Ktop(Ku(X)) of the admissible subcategory Ku(X). Then the restriction of (-,-):=-χ(-,-) to Ktop(Ku(X)) defines a pairing on it, called the Mukai pairing.
It was proved in [2] that the lattice H~(Ku(X),Z) is deformation invariant. Moreover, when Ku(X) is equivalent to the derived category of a K3 surface, H~(Ku(X),Z) is Hodge isometric to the Mukai lattice of the K3 surface. In particular, as a lattice, H~(Ku(X),Z)≅U⊕4⊕E8(-1)⊕2, where U is the hyperbolic lattice and E8(-1) is the twist by -1 of the lattice corresponding to the root system E8.
The lattice H~(Ku(X),Z) comes with a weight-2 Hodge structure defined as follows. Consider the natural map
which is usually called Mukai vector. See [2, Section 2] for more details. Taking its complexification, we then set
Set HHodge(Ku(X),Z):=H1,1(Ku(X))∩H~(Ku(X),Z).
For later use, we are interested in describing special classes in H~Hodge(Ku(X),Z). To this extent, if we denote by ι∗:Db(X)→Ku(X) the projection functor as above and by ℓ any line in X, we can consider the classes
in the numerical Grothendieck group N(Ku(X)) of Ku(X).
The integral Hodge conjecture for cubic fourfolds was originally proved in [144, Theorem 18] and then reproved in [13, Corollary 29.8] (see also [125] for a general treatment). As for the Kuznetsov component, one can prove that H~Hodge(Ku(X),Z) is naturally isometric to the numerical Grothendieck group N(Ku(X)) of the Kuznetsov component and thus not only it contains interesting classes as observed above but it entirely consists of algebraic classes (see [13, Theorem 29.2]).
It is then clear that the classes λ1 and λ2 are in HHodge(Ku(X),Z)⊆H(Ku(X),Z) as well. We set
to be the primitive sublattice of H~Hodge(Ku(X),Z) generated by the classes λ1 and λ2. With the choice of these generators, A2 is the free Z-module Z⊕Z with intersection form given by the matrix
In this section we want to state and prove the Categorical Torelli theorem for Enriques surfaces as in [97, 100]. As we will see, we need to consider the semiorthogonal decompositions in Section 3.2. The key idea is to prove and use an extension result for Fourier– Mukai equivalences between admissible subcategories. In particular, no stability conditions are needed here.
In this section we illustrate a general criterion which was proved in [97]. It allows us to extend Fourier– Mukai equivalences between admissible subcategories under some assumptions on the nature of the semiorthogonal decompositions.
([97, Propositions 2.4 and 2.5]) Let α1:D1↪Db(X1) be an admissible embedding and let E∈⊥D1 be an exceptional object. Let ΦE:Db(X1)→Db(X2) be a Fourier– Mukai functor with the property that ΦE(⊥D1)≅0. Suppose further that
Then there exists a Fourier– Mukai functor ΦE~:Db(X1)→Db(X2) satisfying
In general, such a criterion is not easy to apply. We will try to clarify this with a brief vague discussion. Assumption (b) is hard to verify in concrete examples and, in some cases, it might happen that (b) is not satisfied by a given Fourier– Mukai equivalence ΦE which must then be composed with some additional autoequivalence of D2 (or D1). Heuristically, we should expect to be able to apply such a result either when we have a good grip on the autoequivalence groups of D1 and D2 or when we have a good understanding of the object α1α1!(E). We will show in the next section that for Enriques surfaces we are in the second scenario.
Of course, the criterion can be iterated when the orthogonal complement of Di consists of more than one exceptional object. As we will see, this makes computations more complicated.
We conclude this section by pointing out that the above criterion is very much related to the gluing theory for dg categories and dg functors developed in [88]. The reader can have a look at [97, Section 2.3] for an extensive discussion.
As we mentioned, the Categorical Torelli theorem for Enriques surfaces will be obtained as an application of Theorem 3.4 and Propsition 4.1. And, as we commented above, this is made possible by a complete understanding of the projection into the Kuznetsov component of the 10 exceptional line bundles in L.
Let us begin with a general discussion.
Let T be a triangulated category that is linear over a field K and with Serre functor ST.
Spherical objects were introduced and studied in [137]. They often appear in triangulated categories of Calabi– Yau type and, more specifically, in the derived category of smooth projective Calabi– Yau varieties. They naturally define special autoequivalences which are called spherical twists and which correspond, under Mirror Symmetry, to Dehn twists in the mirror Fukaya category.
The notions of spherical object and spherical twist have been widely extended and generalized. Actually, n-pseudoprojective objects are part of this more general picture. Indeed, the graded vector space of derived endomorphisms is, up to multiplying by 2 the degree, the same as the graded vector space of the total cohomology of an n-dimensional complex projective space. Hence, n-pseudoprojective objects are slight generalizations of the kind of objects studied in [61, 76].
Let us go back to the geometric setting and let us assume that X is an Enriques surface. As we explained in Section 3.2, we have a semiorthogonal decomposition
as in (3.1), where the 10 exceptional line bundles in L are as in Propsition 3.7. Without loss of generality, we can reorganize these exceptional objects to get a semiorthogonal decomposition
into blocks such that, if c≠10, then there is a positive integer 1≤d≤c such that Lj consists of more than one object if 1≤j≤d and of just one object if d<j≤c.
Consider now the corresponding objects Si∈Ku(X,L) defined in (3.2). They provide a complete classification of 3-spherical and 3-pseudoprojective objects in Ku(X,L) according to the following result.
([97], Proposition 4.10 & [100], Theorem 2.7) In the setting above, if F is an object in Ku(X,L), then
Furthermore, all these 3-spherical and 3-pseudoprojective objects are not isomorphic.
As we pointed out in [100, Remark 2.9], this result (see also its easy consequence [100, Corollary 2.8]) together with a positive answer to Question 3.9 (ii) would provide another counterexample to the Jordan– Hölder property for semiorthogonal decompositions. Roughly, such a property predicts that if X is a smooth projective variety then the semiorthogonal decompositions of Db(X) are essentially unique, up to a reordering of the components and up to equivalence. The fact that this property does not hold in general is not new due to the counterexamples in [7, 77].
We are now ready to state the main result of this section which is a combination of [97, Theorem A] and [100, Theorem A] (where we refer to the result below as the Refined Derived Torelli theorem).
(Categorical Torelli theorem for Enriques surfaces) Let X1 and X2 be Enriques surfaces over an algebraically closed field K of characteristic different from 2. If they possess semiorthogonal decompositions
where i=1,2 and Ni is defined as in Section 4.2 (with L replaced by Ni), and there exists an exact equivalence F:Ku(X1,N1)→∼Ku(X2,N2) of Fourier– Mukai type, then X1≅X2.
For simplicity, we illustrate the proof of Theorem 4.5 when N1 and N2 consist of 10 completely orthogonal line bundles. Hence,
for i=1,2. By Theorem 4.4 (1), the projections Si,j of Li,j into Ku(Xi,Ni) are, up to shift and isomorphism, the only 3-spherical objects in Ku(Xi,Ni). The more general case where Ni contains blocks with more then one object is dealt similarly by using Theorem 4.4 (2).
The idea is to extend the Fourier– Mukai equivalence F:Ku(X1,N1)→Ku(X2,N2) step by step by adding all the 10 exceptional objects in Ni. It is not difficult to see that, since these objects are completely orthogonal, it is enough to show how to add one of them. Thus, let us consider L1,1 and its projection S1,1. Since being 3-spherical is a property which is invariant under equivalence, the object F(S1,1) is 3-spherical as well. By Theorem 4.4 (1), up to shift, there is j∈{1,⋯,10} and an isomorphism
Again, by orthogonality, we can permute the exceptional objects L2,j’s and assume, without loss of generality, that j=1.
Now, a direct application of Propsition 4.1 implies that the Fourier– Mukai equivalence F extends to a Fourier– Mukai equivalence
The same argument can be applied again for the other exceptional objects in N1 and, in a finite number of steps, we get an equivalence Db(X1)≅Db(X2). Now we can just invoke Theorem 3.4 and deduce that X1≅X2. The careful reader might have noticed that, a priori, the argument gets more complicated when we add more exceptional one should classify 3-spherical objects in categories which are larger that the original Kuznetsov component. But this is in reality much simpler because the line bundles Li,j are completely orthogonal and the projection of Li,j onto
is the same as the projection onto Ku(Xi,Ni). Furthermore the situation gets even more involved when we deal with Enriques surfaces whose Kuznetsov component contains pseudoprojective objects. The detailed explanation about how these problems can be overcome are not suited for this survey and we refer to the original papers [97] and [100].
We conclude this section by observing that our proof of Theorem 4.5 is almost completely characteristic free. The only point where we use that K is not only algebraically closed but also such that char(K)≠2 is when we invoke Theorem 3.4. Thus, if we can answer Question 3.5 in the positive, then we can extend Theorem 4.5 to Enriques surfaces over fields of characteristic 2. Furthermore, if we provide a positive answer to Question 3.9 (iii), we can avoid assuming that the equivalence between the Kuznetsov components is of Fourier– Mukai type.
Stability conditions on triangulated categories have been defined for the first time by Bridgeland in [24], generalizing the notion of slope stability for sheaves on curves. Since then, the development of the theory has led to applications in classical algebraic geometry and in the study of moduli spaces of stable objects in admissible subcategories of the bounded derived category. In this section, we review the definition of (weak) stability conditions and the construction in the case of the bounded derived category of a smooth projective variety via tilt stability. Our main references are [11, 13, 24].
Let K be an algebraically closed field of arbitrary characteristic. Let X be a smooth projective variety over K and D be a full admissible subcategory of Db(X). A (weak) stability condition on D is essentially the data of the heart of a bounded t-structure and of a (weak) stability function detecting the semistable objects, satisfying certain compatibility conditions.
The heart of a bounded t-structure on D is a full subcategory A⊂D such that
Note that the heart of a bounded t-structure A is not a triangulated category. By [19] we have that A is an abelian category. We denote by K(A) the Grothendieck group of A. As an example, the abelian category Coh(X) of coherent sheaves on X is the heart of a bounded t-structure on D=Db(X).
Given E∈D, the objects Ai in Definition 5.1 are uniquely determined and functorial (also the integers ki are unique). They are called the cohomology objects of E in the heart A.
Let A be the heart of a bounded t-structure on D. A group homomorphism Z:K(A)→C is a weak stability function on A if for any 0≠E∈A we have ℑZ(E)≥0, and in the case that ℑZ(E)=0, we have ℜZ(E)≤0. A stability function on A is a weak stability function Z such that for any 0≠E∈A with ℑZ(E)=0, we have ℜZ(E)<0.
Given a (weak) stability function Z, the slope of E∈A is
and the phase of E is
We point out that if Z(E)=0, then μZ(E)=+∞ and ϕ(E)=1. If F=E[k] for E∈A, then ϕ(F)=ϕ(E)+k.
Let K(D) be the Grothendieck group of D. It is not difficult to see that K(D)=K(A). Fix a finite rank free abelian group Λ and a surjective morphism v:K(D)↠Λ.
A weak stability condition (with respect to v) on D is a pair σ=(A,Z), where A is the heart of a bounded t-structure on D and Z:Λ→C is a group morphism called central charge, satisfying the following
If Z∘v is a stability function, we say that σ is a Bridgeland stability condition on D (with respect to v).
The notion of weak stability condition is very useful for the construction of Bridgeland stability conditions, as we will explain in Section 5.3.
A Bridgeland stability condition is a stability condition in the sense of [24]. Note that it is not clear whether there exist moduli spaces parametrizing semistable objects with a fixed class in Λ, since they do not have a GIT description. Following the recent developments in [13] about the theory of families of stability conditions and [3] about the existence of good moduli spaces, we introduce the notion of stability condition with moduli spaces, which is a Bridgeland stability condition with “well-behaved” moduli functors.
Assume that the base field K is of characteristic 0. Given a Bridgeland stability condition σ on D with respect to v, fix v∈Λ and ϕ∈R such that Z(v)∈R>0eiπϕ. Consider the functor
from the category of schemes over K to the category of groupoids, which associates to T∈Ob(Sch) the groupoid Mσ(D,v)(T) of all perfect complexes E∈D(X×T) such that, for every point t∈T, the restriction Et of E to the fiber X×Spec(k(t)) belongs to D, is σ-semistable of phase ϕ and v(Et)=v.
A stability condition with moduli spaces on D (with respect to v) is a Bridgeland stability condition σ=(A,Z)
If σ is a stability condition with moduli spaces on D with respect to v, then by [13, Theorem 21.24(3)], which makes use of [3], for every v∈Λ it follows that Mσ(D,v) admits a good moduli space Mσ(D,v) which is a proper algebraic space over K.
A natural choice for the lattice Λ is the numerical Grothendieck group N(D) of D. In fact, the numerical Grothendieck group is a free abelian group of finite rank. This follows from the fact that D is an admissible subcategory of Db(X), thus we have the semiorthogonal decomposition Db(X)=⟨D,⊥D⟩. Since N(-) is additive, we have that N(D) is a subgroup of the numerical Grothendieck group of X, which is a free abelian group of finite rank [52, 19.3.2]. This motivates the following definition.
A numerical stability condition on D is a stability condition with respect to the numerical Grothendieck group N(D) of D.
(Slope stability) Let X be a smooth projective variety of dimension n with ample class H. Define
where ch0(E) and ch1(E) stand for the rank and the first Chern class of E, respectively. Set ΛH:=Im(v). Then the pair σH=(Coh(X),ZH), where
defines a weak stability condition on Db(X) with respect to ΛH. Indeed, if E is a sheaf on X, then Hnch0(E)≥0 and if it is 0 (i.e. E is a torsion sheaf), then Hn-1ch1(E)≥0. Moreover, by [24, Lemma 2.4] the HN property holds, and by [14, Remark 2.6] the trivial form Q=0 fulfills the support property.
If n=1, i.e. X is a curve, then σH is a numerical stability condition on Db(X), recovering the classical notion of slope stability.
Assume K is an algebraically closed field of arbitrary characteristic3. We denote by StabΛ(D) the set of stability conditions on D with respect to v. We consider on StabΛ(D) the coarsest topology such that the maps (A,Z)↦Z, (A,Z)↦ϕ+(E), (A,Z)↦ϕ-(E) are continuous for every 0≠E∈D. A celebrated result of Bridgeland states that StabΛ(D) has the structure of complex manifold.
(Bridgeland Deformation Theorem, [13, 24], Theorem 1.2) The continuous map Z:StabΛ(D)→Hom(Λ,C) defined by (A,Z)↦Z, is a local homeomorphism. In particular, the topological space StabΛ(D) is a complex manifold of dimension rk(Λ).
The support property implies that if we fix an element v∈Λ, then there is a locally-finite set of real codimension one submanifolds with boundary in StabΛ(D), called walls, where the set of semistable objects with class v changes. The connected components of the complement in StabΛ(D) of the union of walls for v are called chambers.
On StabΛ(D) we have the following group
(i) (Right action of GL2+(R)) Consider the connected group GL2+(R) of 2×2 real matrices with positive determinant. Note that GL2+(R) acts on the right by multiplication on Hom(Λ,C) via the identification C≅R2. In order to lift this action to the stability manifold, we consider the universal covering space GL2+(R) of GL2+(R), whose objects are pairs (M, g) with M∈GL2+(R), g:R→R an increasing function satisfying g(ϕ+1)=g(ϕ)+1, such that the induced actions of M and g on (R2{0})/R>0=S1 are the same. For σ=(A,Z)∈StabΛ(D) and (M,g)∈GL~2+(R), we define σ·(M,g) as the stability condition with heart P((g(0),g(1)]) and central charge Z′=M-1∘Z (see [24, Lemma 8.2]). Concretely, the stability conditions σ and σ·(M,g) have the same set of semistable objects, but with different phases.
(ii) (Left action of AutΛ(D)) Consider the group AutΛ(D) of pairs (Φ,ΦΛ), where Φ is an exact autoequivalence of D and ΦΛ is an endomorphism of Λ such that ΦΛ∘v=v∘Φ∗. Here Φ∗ is the automorphism of K(D) induced by Φ. For (Φ,ΦΛ)∈AutΛ(D) and σ∈StabΛ(D), we define the stability condition (Φ,ΦΛ)·σ=(Φ(A),Z∘ΦΛ-1). Note that if Λ=N(D), then the endomorphism ΦΛ is determined uniquely by Φ, hence AutΛ(D)=Aut(D) and one can talk about an action of Aut(D).
Assume that σ is a stability condition with moduli spaces. We observe that, by definition, moduli spaces with respect to σ·(M,g) and Φ·σ are isomorphic to moduli spaces with respect to σ.
(i) Let X be a smooth projective curve of genus ≥1 and set Λ:=N(Db(X)). Then
By [24, 109] the action of GL2+(R) is free and transitive, thus there is a unique orbit of numerical stability conditions with respect to the GL2+(R)-action, i.e.
where σH is the slope stability defined in Example 5.8.
(ii) (Action of the Serre functor) The Serre functor SD on D (see Example 2.2) defines an element in AutN(D)(D)=Aut(D), thus we can consider its action on numerical stability conditions. In Section 6.2 we will introduce the notion of Serre-invariant stability conditions, which are numerical stability conditions preserved by the Serre functor, up to the GL~2+(R)-action. In fact, this notion plays an important role in a proof of the Categorical Torelli theorem for cubic threefolds and in the study of the geometry of moduli spaces of stable objects in the Kuznetsov component, as we will see in Section 6.3.
The construction of stability conditions is a difficult task, even in the case of Db(X) for a smooth projective variety X of dimension n≥3 over an algebraically closed field K. A conjectural approach is via the notion of tilt stability, whose definition is summarized in this section and which works perfectly for X of dimension n=2 (see [1, 25]).
Let (X, H) be a polarized smooth projective variety. We have seen in Example 5.8 that slope stability σH=(Coh(X),ZH) defines a stability condition on Db(X) when X is a curve. However, in higher dimension this is no longer true, as ZH vanishes on torsion sheaves supported in codimension ≥2, and in higher dimensions slope stability is only a weak stability condition. Actually, the choice of Coh(X) as heart is not the correct one, since by [141, Lemma 2.7] it cannot be the heart of a numerical stability condition if dim(X)≥2.
To overcome this problem, one can consider a new heart by tilting Coh(X). More precisely, let σ=(A,Z) be a weak stability condition on a triangulated K-linear category. Fix s∈R and define the following subcategories of A:
As proved in [53], the category
is the heart of a bounded t-structure on T and is called the tilting of A with respect to σ at slope s. In this context ⟨-,-⟩ means the smallest full subcategory closed under extensions and containing the two additive subcategories Tσs and Fσs[1]. Note that if s>s′, then we have
Indeed, consider F∈A and σ-semistable with μZ(F)>s, which is an object in Aσs. Then μZ(F)>s′, so F∈Aσs′. Otherwise, consider F∈A and σ-semistable with μZ(F)≤s, so F[1]∈Aσs. If μZ(F)≤s′, then F[1]∈Aσs′, while if μZ(F)>s′, then F[1]∈Aσs′[1]. By the definition of Aσs, we deduce the desired property.
In the case of slope stability, we consider the heart
obtained by tilting Coh(X) with respect to σH at slope s. In analogy to the curve case, define
where ch2(E) denotes the degree-2 part of the Chern character of E, and set ΛH:=Im(v). For s,q∈R, define Zs,q:ΛH→C by
The slope associated to Zs,q is denoted by μs,q. We can now state the main result of this section.
([1, 11, 12, 24]) Let X be a smooth projective variety of dimension n≥2. There is a continuous family of weak stability conditions with respect to v, parametrized by
defined as
with a locally-finite wall and chamber structure.
If n=2, i.e. X is a surface, then {σs,q} is a continuous family of stability conditions on X.
The proof of Theorem 5.11 is slightly involved and we refer to [112, Sections 6.2, 6.3] for a detailed summary. We only mention that a key ingredient is the classical Bogomolov Inequality, which implies that every σH-semistable sheaf E satisfies
Moreover, one can choose ΔH as the quadratic form fulfilling the support property for σs,q. This also implies that σs,q satisfies well-behaved wall-crossing: for fixed v∈ΛH, the boundary of the locus in Δ where an object of class v is stable is defined by a locally-finite set of submanifolds of real codimension one.
The weak stability conditions σs,q defined in Theorem 5.11 are called tilt-stability conditions. A useful way to visualize tilt stability conditions has been introduced in [101, Section 1]. Note that given E∈Db(X) such that Zs,q(E)≠0, its truncated at degree 2 Chern character
defines the point
in a projective plane PR2. If ch0(E)≠0, we consider the affine coordinates
By (5.4) and [11, Theorem 3.5], we have that σs,q-semistable objects correspond to points below the parabola q=12s2 and points above parametrize tilt stability conditions. Furthermore, the phase of a σs,q-semistable object E∈Cohs(X) is the angle between the line connecting (s, q) and (s(E), q(E)) and the vertical half-ray from (s, q) to -∞ divided by π (see Figure 1).
Fig. 1 We can compare the σs,q-slope of σs,q-semistable objects E and F, using the E has larger slope than F if and only if the ray connecting E to (s, q) is after the ray connecting F with (s, q) moving from the dashed ray in counterclockwise direction (see [102, Lemma 2])
It is now natural to wonder whether the above procedure can be generalized to construct stability conditions when X has dimension >2. In this case, it is easy to see that Zs,q does not define a stability function, as it vanishes on objects supported in codimension ≥3.
Assume X is three-dimensional. In analogy to the surface case, in [12] the authors consider a new heart obtained by tilting Cohs(X) with respect to the tilt stability σs,q and a weak stability function on it, involving the third Chern character of the objects. The key observation, made in [11, 12], is that this new weak stability condition is a stability condition on Db(X) if and only if tilt semistable objects satisfy some quadratic inequality, called generalized Bogomolov inequality, involving the Chern characters till degree 3. Following this approach, stability conditions have been constructed on the bounded derived categories of Fano threefolds [12, 21, 95, 110, 135], abelian threefolds [11, 107, 108], some resolutions of finite quotients of abelian threefolds [11], and threefolds with nef tangent bundle [21, 74].
A further progress in this direction has been made in [96], where a stronger Bogomolov inequality, namely an inequality involving the Chern characters up to degree 2 of slope semistable torsion free sheaves (see [96, Theorem 1.1] for the precise statement), has been proved in the case of quintic threefolds, giving the first highly nontrivial example of stability conditions on a strict (i.e. simply-connected) Calabi–Yau threefold. The involved strategy makes use (among other things) of a restriction lemma, first appeared in [50], which allows one to reduce to show a Clifford type bound for the dimensions of global sections of stable vector bundles on curves defined as complete intersections of two quadrics and a quintic hypersurface in P4. The existence of stability conditions is then obtained by using this stronger Bogomolov inequality to prove the following statement (see [11, Conjecture 4.1]): if E is σs,q-semistable for a certain choice of (s,q)∈Δ, then E satisfies
where ch1s(E)=ch1(E)-sHch0(E), ch2s(E)=ch2(E)-sHch1(E)+s22H2ch0(E), ch3s(E)=ch3(E)-sHch2(E)+s22H2ch1(E)-s36H3ch0(E).
This method has been recently generalized to the case of some three-dimensional weighted hypersurfaces in weighted projective spaces [75], and to three-dimensional complete intersections of quartic and quadric hypersurfaces in P5 [103]. An interesting problem would be trying to adapt this argument to treat other examples of Calabi–Yau complete intersections in PN. For instance, the next case to address could be the
Let X be a complete intersection of a cubic and two quadric hypersurfaces in P6. Is it possible to show a stronger Bogomolov inequality for slope semistable torsion free sheaves on X, similarly to [96, Theorem 1.1], [103, Proposition 1.1], which implies (5.5)?
We end this section with a sort of negative twist. Indeed, while the possibility to prove the generalized Bogomolov inequality (5.5) is still plausible for threefolds with trivial canonical bundle, we know that (5.5) is false in general. The first counterexample was provided in [136]. A replacement for this too optimistic guess was recently provided in [10]. In order to state the revised conjecture, we proceed as follows. Let (X, H) be a polarized projective variety of dimension n. We denote by CHnum∙(X) the Chow ring of X modulo numerical equivalence. Let B∈NS(X)⊗R, where NS(X)=CHnum1(X), let Γ∈CHnum2(X)⊗R be such that Γ·Hn-2=0, and pick γ3,⋯γn to be arbitrary classes with γi∈CHnumi(X)⊗R. Let
and set chγ(-):=γ·ch(-). Using the class γ and the modified Chern character, one can repeat the discussion above, define γ-slope semistable objects and thus the notion of γ-tilt stability σs,qγ for (s,q)∈Δ (see [10, Section 4] for more details).
For an object E∈Db(X) we then set
Set viγ to be the i-th component of vHγ(E). We can then state the following which is [10, Conjecture 4.7]:
(Bayer– Macrì) Let (X, H) be a smooth complex projective polarized variety. There exists a class γ as in (5.6) and an upper semicontinuous function f:=fγ:R→R such that, for any σs,qγ-semistable object E,
for all (s,q)∈Δ, such that q>fγ(s).
We do not explain the origin of this conjecture here. We only remark that it is proved for the above mentioned cases (prime Fano threefolds, abelian threefolds, quintic threefolds, complete intersections of quadratic and quartic hypersurfaces in P5, the blow-up of P3 at a point, threefolds with nef tangent bundle) which justifies the way it is formulated. For an extensive and very interesting discussion, one can have a look at [10, Section 4]. The key point is that, if n=3, the conjecture implies the existence of stability conditions on Db(X). Finally, it is clear that the original conjecture in [11, 12] implies Conjecture 5.13 for some particular function fγ while the counterexamples mentioned above do not apply in this case.
As we mentioned in Section 3.3, if X is a cubic threefold, we have a semiorthogonal decomposition
where H is a hyperplane section. In this section we investigate the existence of stability conditions on Ku(X) and discuss several applications including, of course, the Categorical Torelli theorem for these hypersurfaces.
In this section we want to recall the approach in [20] to the construction of stability conditions on the Kuznetsov component of a cubic threefold. Even though a much more modern technology is, at the moment, available (see Section 6.2 and Section 6.3 below), this circle of ideas have interesting applications to geometric problems and it is an important illustration of what will be proven later about cubic fourfolds.
Assume that X is defined over a field K which is algebraically closed with char(K)≠2. The starting point is the following observation. Fix a line ℓ0 in a cubic threefold X↪P4 and consider a plane P2⊆P4 which is skew with respect to ℓ0. The rational map π0:X⤏P2 given by the projection from ℓ0 can be resolved as follows. Consider the blow-up P4 of P4 along ℓ0 and the strict transform X of X inside P4. They all sit in the commutative diagram where q is the P2-bundle induced by the projection from ℓ0 and its restriction π0 to X~ is the corresponding conic fibration. We denote by h the pull-back to X~ of a hyperplane section of P2, by H the pull-back to X~ of a hyperplane section of P4 and by e the exceptional divisor.
As a beautiful application of [83] (see also [6] for the case of arbitrary odd characteristic), the conic bundle π0:X→P2 yields a sheaf B0 on P2 of even part of Clifford algebras. The explicit description of B0 is not needed here but it is worth remembering that we have an isomorphism
of OP2-modules. Moreover, B0 is noncommutative and is an Azumaya algebra away from the discriminant of the conic bundle. One can further consider the abelian category Coh(P2,B0) of coherent B0-modules and the corresponding derived category Db(P2,B0).
It is clear from the discussion above that we can regard X~ both as a blow-up of X and as a conic fibration on P2. In the first case, [123] yields a semiorthogonal decomposition
where Ku(X) is embedded via the pullback along the blowup morphism X~→X. On the other hand, [83, Theorem 4.2] provides another semiorthogonal decomposition
(see [20, (2.7)]). These decompositions are obtained from the standard semiorthogonal decompositions by a sequence of mutations, using the equalities e=H-h and KX~=-H-h. We should note here that, in general, it is important to give an explicit description of the way Ku(X) and Db(P2,B0) are embedded in Db(X~) in (6.2) and (6.3) respectively. Here we can safely ignore this and just observe that a direct comparison of the semiorthogonal decompositions in (6.2) and (6.3) yields a new
where E:=OX~(h-H) is an additional exceptional object. Again, we are harmlessly ignoring the fact that Ku(X) is embedded in Db(P2,B0) in a nontrivial way.
Now, the semiorthogonal decomposition (6.4) and the observation that Db(P2,B0) behaves like the derived category of a (twisted) surface are the keys to apply the following general (and as such, at the moment, vague) *Dimension reduction * to study the geometric/homological properties of the Kuznetsov component Ku(X) of a Fano threefold/fourfold X, embed Ku(X) in the derived category of a smaller dimensional (twisted) variety.
This is particularly rewarding when we want to construct stability conditions on Ku(X). Indeed, as we explained in the previous sections, we have standard strategies to construct stability conditions on surfaces, the situation is less clear but, by now, manageable for threefolds and mostly obscure in higher dimension. We show now how this works for the Kuznetsov component of cubic threefolds. Then in Section 6.2 we will see that there is a more direct way to contruct stability conditions in the cubic threefold case, without relying on a dimension reduction trick. On the other hand, we will observe later that the dimension reduction trick is crucial while dealing with cubic fourfolds.
Let us now go back to X a cubic threefold. As we observed in the previous section, a standard way to construct (weak) stability conditions on surfaces is by using slope stability and tilt with respect to it.
In the case of the abelian category Coh(P2,B0), the fact that B0 is an OP2-algebra (see (6.1)), allows us to define a forgetful functor
and thus the functions
where N(P2,B0) stands for the numerical Grothendieck group of Db(P2,B0). For F∈Coh(P2,B0) with rk(F)≠0, we define the slope
As usual we can recast μh in terms of a central charge
Given F∈Coh(P2,B0), when we say that F is either torsion-free or torsion of dimension d, we always mean that Forg(F) has this property. It is then not difficult to show (see [20, Section 2.3]) that the pair σh=(Coh(P2,B0),Zh) is a weak stability condition on Db(P2,B0).
Note that if F∈Coh(P2,B0) has rank 0, then μh(F)=+∞. Given this, we can apply the tilting procedure described in the previous section and define the two full additive subcategories
Tilting with respect to the pair (F0,T0) yields the abelian category Coh-54(P2,B0), which is the heart of a bounded t-structure. Set
It turns out that A0 is the heart of a bounded t-structure on Ku(X) (see [20, Lemma 3.4]).
A useful computation shows that if ℓ⊆X is any line, then the ideal sheaf Iℓ is contained in A0 (see [20, Proposition 3.3]).
Consider now the function
on the numerical Grothendieck group N(P2,B0) of Db(P2,B0). The restriction Z0:=Z|N(Ku(X)) of Z to the numerical Grothendieck group of Ku(X) is a stability function. For this, see [20, Lemma 3.5]. This means that it satisfies the condition in Definition 5.3.
These preliminary observations allow us to conclude that we have stability conditions on Ku(X). Assume that the base field K has characteristic 0. Applying the recent results in [3, 13, Proposition 25.3] one can further show that σ0 is a stability condition with moduli spaces. For later use, we are also interested in studying special moduli spaces of stable objects. The precise result is the
([20], Theorems 3.1 and 4.1) In the assumptions above
The construction of (weak) stability conditions on Db(P2,B0) and on Ku(X), for X a cubic threefold, was realized in [92] in a similar fashion as above, and it has been used to answer some very geometric questions. In particular, in [92, Theorem B] the authors reprove and generalize the main result of [40] about the nonemptiness of moduli spaces of Ulrich bundles of arbitrary rank on cubic threefolds. Let us recall that an Ulrich bundle E is an aCM (arithmetically Cohen– Macaulay) bundle whose graded module ⨁m∈ZH0(X,E(m)) has 3rk(E) generators in degree 1. On the other hand, a vector bundle E on a cubic threefold X is aCM if dimHi(X,E(j))=0, for i=1,2 and all j∈Z. We will come back to related questions in Section 6.4 where the results mentioned above are stated in a precise form (see also [92, Remark 2.20] for comments on the used normalization in the definition of Ulrich bundle).
The result above is the key to prove the following which is indeed [20, Theorem 1.1].
(Categorical Torelli theorem for cubic threefolds) Let X1 and X2 be cubic threefolds defined over an algebraically closed field K with char(K)≠2,3. Then there exists an exact equivalence Ku(X1)≅Ku(X2) if and only if X1≅X2.
(Idea of proof) One implication is trivial. Indeed, if X1≅X2, then the isomorphism preserves the line bundles OXi and OXi(HXi). Thus we get an induced exact equivalence Ku(X1)≅Ku(X2).
Let us now start with an exact equivalence F:Ku(X1)→Ku(X2). Let ℓ⊆X1 be a line, seen as a point in F1(X1). By Theorem 6.2 (2), the object Iℓ is σ0-stable in Ku(X1). We would like to conclude that F(Iℓ) is again the ideal sheaf of a line in X2, but this is not true in general. Nevertheless, by the result of the discussion in [20, Section 5.1] based on a delicate argument in [20, Section 4], we can compose F with a suitable power of the Serre functor SKu(X2) and a shift, in order to get another exact equivalence F′:Ku(X1)→Ku(X2) such
Thus, Theorem 6.2 (2) implies that F′ induces a bijection between F1(X1) and F1(X2). By [20, Section 5.2] such a bijection can be turned into an isomorphism f:F1(X1)→F1(X2). It was observed in [42] that F1(Xi) is a surface of general type whose canonical bundle coincides with the ample class induced by the Plücker embedding of the Grassmannian of lines in P4. Thus, the isomorphism f must preserve the induced ample class given by the Plücker embedding. The trick in [41, Proposition 4] applies and thus f induces an isomorphism X1≅X2.
One interesting consequence is the following.
Question 2.8 has a positive answer for cubic threefolds.
Let X1 and X2 be cubic threefolds with an exact equivalence F:Ku(X1)→Ku(X2). By Theorem 6.4, there is an isomorphism f:X1→X2 and, as observed in the proof of the result above, f yields an exact equivalence between the Kuznetsov components. For the same reason as in Remark 4.6 (i), such an equivalence is of Fourier– Mukai type.
Let us conclude this section with the following
Is it possible to prove Theorem 6.4 by using the extension techniques that we discussed in Section 4.1?
In this section we aim at explaining the methods introduced in [14], which allow us to induce stability conditions on the Kuznetsov component of prime Fano threefolds defined in Section 3.3, so in particular of cubic threefolds, and, with a dimension reduction trick as in Section 6.1, of cubic fourfolds. Then we introduce the notion of Serre-invariant stability conditions, whose existence in the context of cubic threefolds will have interesting consequences on the study of moduli spaces.
Let T be a proper triangulated category which is linear over an algebraically closed field K and with Serre functor ST. Assume T has an exceptional collection E1,⋯,Em. Then we have a semiorthogonal decomposition of the form
where D:=⟨E1,⋯,Em⟩⊥. Note that rk(N(D))=rk(N(T))-m by additivility of the numerical Grothendieck group. This could vaguely suggest that D has “smaller dimension” than T, so it might be easier to construct Bridgeland stability conditions on D than on T.
In fact, the following key result provides a criterion which guarantees that a weak stability condition σ on T restricts to a Bridgeland stability condition on D.
([14], Proposition 5.1) Let σ=(A,Z) be a weak stability condition on T. Assume that the exceptional collection {E1,⋯,Em} satisfies the following
Then the pair σ′=(A′,Z′) is a Bridgeland stability condition on D.
Note that, if the weak stability condition σ in the statement is with respect to a lattice Λ, then σ′ is defined over the sublattice Λ′ of Λ determined by the image of K(A′).
We point out that the conditions (1),(2) are used to show that the restriction A′=A∩D is the heart of a bounded t-structure on D, since they imply that the cohomology factors with respect to A of an object in D belong to D as well. By (4) the restriction Z′ is a stability function on A′. For the proof of the Harder– Narasimhan property and the support property the authors make use of the notions of Harder– Narasimhan polygon and mass of an object in the heart. We suggest the interested reader to consult [14, Section 5] for more details.
We should not expect Propsition 6.7 to be applicable in all situations where Db(X) has semiorthogonal decomposition with a Kuznetsov component which is residual to finitely many exceptional objects. One crucial example is when X is an Enriques surface with a semiorthogonal decomposition as in (3.1). Indeed, more should be if that X is very general, then Ku(X,L) should not carry Bridgeland stability conditions at all. The heuristic reason is the following. The objects Si∈Ku(X,L) in (3.2) are, at the same time, 3-spherical and numerically trivial. The latter condition implies that the stability function of any stability condition on Ku(X,L) would map the class of Si to 0. On the other hand, motivated by the case of K3 surfaces, one would expect that, if a stability condition σ on Ku(X,L) exists, then the fact that Si is spherical implies that it is also σ-stable. These two facts together would immediately lead to a contradiction and thus to the fact that such a σ cannot exist. Of course, it would be very interesting to make the previous argument rigorous and show that Si is stable in any stability condition on the Kuznetsov component.
An interesting consequence of Propsition 6.7 is the existence of Bridgeland stability conditions on the whole category T, using results in [43] about gluings of t-structures.
([14], Proposition 5.13) With the assumptions of Propsition 6.7, the pair σ″=(A′′,Z′′) on T, where
is the extension-closure and
is a Bridgeland stability condition on T.
When T=Db(X), slope stability and tilt-stability, defined in Section 5.3, are weak stability conditions and we could ask whether they satisfy the conditions in Propsition 6.7. We will actually use the following tilting of σs,q defined in Theorem 5.11. For μ∈R, let u be the unit vector in the upper half plane with μ=-ℜ(u)ℑ(u). We denote by
the heart of a bounded t-structure obtained by tilting Cohs(X) with respect to σs,q at μs,q=μ. Set
If we were working with Bridgeland stability conditions, the above process would correspond to consider u as an element of GL1(C)⊂GL2+(R), lift it to u~∈GL2+(R) and act on σs,q by u. On the other hand, the GL~2+(R)-action is not a priori well-defined on weak stability conditions, because of the existence of objects with vanishing central charge. For instance, consider an extension E of the form
where F is σs,q-semistable in Cohs(X) with μs,q(F)≤μ and T∈Coh(X) supported in codimension >2. Then E∈Cohs,qμ(X), but E is not in Cohs(X) and thus cannot be σs,q-semistable. This suggests that there could be objects which are semistable with respect to σs,qμ, but not with respect to σs,q (see [131, Lemma 5.4] for a concrete example). This prevents us to have a well-defined GL~2+(R)-action on these weak stability conditions. Nevertheless, we have the following result.
([14], Proposition 2.15) Let X be a smooth projective variety. The pair σs,qμ=(Cohs,qμ(X),Zs,qμ) is a weak stability condition on Db(X) with respect to v defined in (5.3).
We end this paragraph by introducing a notion which will be very useful in the case of prime Fano threefolds.
A (Bridgeland) stability condition σ on a triangulated category T with Serre functor ST is Serre-invariant (or ST-invariant), if there exists g~∈GL~2+(R) such that
Note that the property of ST-invariance is stable under the GL~2+(R)-action.
One may wonder why we use the word invariant in the previous definition, which would be more suitable to identify a stability condition which is fixed by the Serre functor. However, recall that stability conditions in the same orbit with respect to the GL2+(R)-action have the same set of semistable objects. In particular, (if they exist!) the corresponding moduli spaces are isomorphic. From this viewpoint, we are interested in distinguishing or identifying stability conditions, up to the GL2+(R)-action.
If X is a curve of genus g(X), then the slope stability σH (see Example 5.8) is a Serre-invariant stability condition on Db(X). Actually, if g(X)≥1, then there is a unique GL~2+(R)-orbit of stability conditions on Db(X), so the condition in Definition 6.11 is trivially satisfied. Anyway, the fact that slope stability is Serre-invariant can be checked directly as follows. The Serre functor of Db(X) is
Tensoring by the line bundle ωX preserves the slope stability of a coherent sheaf and SX(Coh(X))=Coh(X)[1]. On the level of central charges, since (rk(ωX),deg(ωX))=(1,2g(X)-2), we have
Then the matrix M=-12g(X)-20-1 satisfies ZH∘(SX-1)∗=M-1∘ZH. Now note that
This implies that, if P is the slicing defined by σH, then there exists a lifting (g,M)∈GL~2+(R) of M, such that P((g(0),g(1)])⊂Coh(X)[1] (since Coh(X)=P((0,1]) and Meiπϕ∈R>0eiπg(ϕ)). Since they are both hearts of bounded t-structures, we have the equality P((g(0),g(1)])=Coh(X)[1].
One could first wonder whether an admissible subcategory D of Db(X) of a smooth projective variety X admits Serre-invariant stability conditions. This has been recently proved to be false in [91], using the notion of Serre dimension, in the case of the Kuznetsov component of almost all Fano complete intersections. Anyway, we could focus on prime Fano threefolds and consider the following less general
When does the Kuznetsov component of a prime Fano threefold with index 1 or 2 admit Serre-invariant stability conditions?
We will see in the next section that Question 6.15 has a positive answer for the Kuznetsov components of cubic threefolds and some other prime Fano threefolds. In these cases, we will also see that there is a unique GL~2+(R)-orbit of Serre-invariant stability conditions. Motivated by this, we could also ask the
Assume that D is an admissible subcategory of Db(X) which admits Serre-invariant stability conditions. If N(D) has rank 2, is there a unique orbit of Serre-invariant stability conditions with respect to the GL~2+(R)-action?
The question above is very much related to several expectations concerning the topology of the stability manifold of D. In particular, one my wonder whether such a manifold is connected. A positive answer to Question 6.16 may give a positive indication in the direction of this expectation as well.
Let us now apply the methods introduced in Section 6.2 to the Kuznetsov component of a cubic threefold X defined over an algebraically closed field K.
Consider first the slope stability σH=(Coh(X),ZH). Unfortunately, it does not satisfy the assumptions in Propsition 6.7 using the exceptional collection OX, OX(HX). Indeed, we see for instance that
Let us try with the weak stability conditions σs,q=(Cohs(X),Zs,q) on Db(X) for (s, q) in the set Δ=(s,q)∈R2:q>12s2 (see Section 5.3). Note that OX, OX(HX) are slope stable, as they are line bundles, with slope 0 and 1, respectively. Choosing s<0, we have OX, OX(HX)∈Cohs(X). On the other hand, we observe that if s<-1, then SX(OX(HX))=OX(-HX)[3]∈Cohs(X)[3], while if -1≤s<0, then OX(-HX)[3]∈Cohs(X)[2]. For later use, we point out that we have
Even if we cannot apply directly Propsition 6.7, this computation suggests that if we tilt another time, we could find a suitable heart. In fact, consider the weak stability condition σs,qμ defined in Propsition 6.10.
([14, Theorem 6.8]) Assume that (s,q)∈Δ and q<-12s. Then the pair
is a stability condition with moduli spaces on Ku(X) with respect to the lattice N(Ku(X)), for every μ∈R satisfying
Moreover, for μ, μ′ satisfying (6.6), the stability conditions σs,qμ|Ku(X), σs,qμ′|Ku(X) belong to the same orbit with respect to the GL~2+(R)-action.
By (6.5) we have that OX, OX(HX), OX(-2HX)[1], OX(-HX)[1] belong to Cohs(X) for -1≤s<0. Since they define points on the parabola q=12s2, by [11, Corollary 3.11(a)], they are σs,q-stable for every (s,q)∈Δ. Comparing the slopes with respect to σs,q via a picture as in Figure 2 (or by a direct computation), we deduce that
for (s, q) as in the statement. Thus if μ satisfies (6.6), then OX, OX(HX), OX(-2HX)[2], OX(-HX)[2] belong to Cohs,qμ(X). Note that an object E∈Cohs,qμ(X) with Zs,q(E)=0 is a torsion sheaf supported in codimension 3 by [14, Lemma 2.16], which is not in Ku(X) since Hom(OX,E)≠0. We can then apply Propsition 6.7 which proves that σs,qμ|Ku(X) is a Bridgeland stability condition on Ku(X) with respect to
Then σs,qμ|Ku(X) is a numerical stability condition on Ku(X) and [3, 13] (see, in particular, Proposition 25.3 in the latter paper) implies that it is a stability condition with moduli spaces. We leave to the reader the check of the second part of the statement, using [14, Lemma 4.3] and considering the rotation between two sides of the sharp angle on Figure 2 which is provided by the segment connecting the point (s, q) to OX and the segment which is the continuation of the one connecting OX(-HX) to (s, q).
Fig. 2 The boundary of the region of points (s, q) inducing stability conditions on Ku(X) is represented in bold. For (s, q) in this region the slope with respect to σs,q of OX is bigger than the slope of OX(-HX)[1]. Note that the sharp angle in grey between the interval connecting the point (s, q) to the point OX and continuation of the interval connecting OX(-H) to (s, q) beyond (s, q) is the region that corresponds to the values of μ for which σs,qμ|Ku(X) is a stability condition on Ku(X)
We denote the induced stability conditions on Ku(X) by
where
As we explain below, these stability conditions satisfy the nice property of being Serre-invariant, answering positively Question 6.15 for cubic threefolds.
From now on, we assume K=C as in the papers we are referring to, although the same proof should apply in more generality over an algebraically closed field of characteristic 0.
([129], Corollary 5.5) Let σ be a stability condition on Ku(X) in the same orbit of σ(s,q,μ) defined in (6.7), with respect to the GL~2+(R)-action. Then σ is Serre-invariant.
(Idea of proof) Since the Serre functor commutes with the GL~2+(R)-action, it is enough to show the statement for σ(-12,q,-12) with 18<q<14. By [81, Lemma 4.1] the Serre functor of Ku(X) satisfies the relation
where Φ(-)=(-)⊗OX(HX). Thus we can further reduce to prove the statement for the autoequivalence LOX∘Φ. Recall that LOX is the left mutation in the exceptional object OX. This means that if α denotes the embedding of the admissible subcategory generated by OX into Db(X), the functor LOX is defined by the canonical triangle of exact functors
where the first arrow is given by adjunction.
The second step is to show that the heart LOX(Φ(A(-12,q,-12))) is a tilt of A(-12,q,-12). This is the hardest part in the proof. One key ingredient is [102, Lemma 3], which allows us to control the slope of the semistable factors with respect to σs′,q′ of a σs,q-semistable object in Cohs(X), when deforming (s, q) to (s′,q′).
Finally, it is not difficult to show that there exists M∈GL2+(R) such that Z(s,q,-12)∘(LOX∘Φ)∗-1=M-1Z(s,q,-12). Then we can use a similar argument to the one in the proof of the second part of Theorem 6.17, to get the statement.
In this case we can also give a positive answer to Question 6.16, as a consequence of the following general criterion, which provides sufficient conditions in order for a fractional Calabi– Yau category to admit a unique GL~2+(R)-orbit of Serre-invariant stability conditions.
([51], Theorem 3.2) Let T be a proper K-linear triangulated category over a field K. Assume T satisfies the following
Let σ1 and σ2 be Serre-invariant numerical stability conditions on T. Then there exists g~∈GL2+(R) such that σ1=σ2·g.
Note that [51, Theorem 3.2] also deals with the case r=2 and k=4. In that situation, we need the additional condition that there are two objects Q1,Q2∈T satisfying (6.9) such that Q1 is not isomorphic to Q2 or Q2[1], Hom(Q2,Q1)≠0 and Hom(Q1,Q2[1])≠0.
The assumptions in the statement could seem unnatural at a first glance, but actually they are not. Indeed, in (1) we require T to be fractional Calabi– Yau of dimension <2, while in (2) we require N(T) to have rank 2 as a noncommutative curve (although the analogy with curves is not totally correct, as for instance in one case the numerical Grothendieck group is negative definite, while in the latter it is not, as pointed out by the referee). Condition (3) is also inspired by the case of the objects Q and ST(Q) for T play a similar role as the one of the skyscraper sheaves and line bundles on a curve in Macrì’s proof [109], e.g. they are stable with controlled phase.
We now see how to apply the above criterion to Ku(X) of a cubic threefold.
Question 6.16 has a positive answer for the Kuznetsov component Ku(X) of a cubic threefold X.
By Propsition 3.12 we know that Ku(X) is fractional Calabi– Yau of dimension 5/3<2. By [20, Proposition 2.7] and [82], the numerical Grothendieck group N(Ku(X)) is a rank-2 lattice and a basis is given by
where Iℓ is the ideal sheaf of a line ℓ in X, with respect to which the Euler form χ is represented by
An easy computation shows that χ(v,v)≤-1 for every 0≠v∈N(Ku(X)), thus ℓKu(X)=-1. Since dimExt1(Iℓ,Iℓ)=2, we see that
Thus Theorem 6.19 applies to Ku(X) and implies the statement.
In the next section, we will see how to apply these results on Serre-invariant stability conditions to study the properties of moduli spaces of stable objects in Ku(X) and to reprove the Categorical Torelli theorem.
Assume K=C. The first application of the results presented in the previous section concerns the relation between the stability conditions σ(s,q,μ), defined in (6.7), and the stability condition σ0, introduced previously in Theorem 6.2. Similarly to Theorem 6.18, we can show the following statement.
([51], Theorem 5.4) The stability condition σ0 on Ku(X) introduced in Theorem 6.2 is Serre-invariant.
The stability conditions σ0 and σ(s,q,μ) defined in (6.7) on Ku(X) are in the same GL~2+(R)-orbit.
It is a consequence of Theorem 6.18 and Theorem 6.22 and Corollary 6.21.
Consider the continuous family of Bridgeland stability conditions on Db(P2,B0), parameterized by
given by
By using Propsition 6.7, one can show [51, Proposition 5.3] that for any (b,w)∈U with -54≤b<-34, the pair
is a stability condition on Ku(X) which is in the GL~2+(R)-orbit of σ0 (see Figure 3).
Fig. 3 We represent the region U parametrizing the weak stability conditions σ¯b,w, which is the region w>12b2+1132. The strip with bold boundary corresponding to the region above the parabola in the interval -54≤b<-34 represents those inducing the stability conditions σ¯(b,w) on Ku(X)
An interesting consequence of these results is that all the known stability conditions on Ku(X) are Serre-invariant, hence belong to the same GL~2+(R)-orbit. A positive answer to the following (hard) question would complete the analogy of Ku(X) with the curve case in [109].
Let σ be a stability condition on Ku(X) of a cubic threefold X. Is σ in the same orbit of σ(s,q,μ)? Equivalently, is σ Serre-invariant?
The second application, which makes use of Corollary 6.23, concerns the study of Ulrich bundles on X, whose definition is recalled in Remark 6.3. Note that an Ulrich bundle E belongs to Ku(X). Indeed, by [92, Lemma 2.19] we have
where d is the rank of E. Note that Ulrich bundles are Gieseker semistable [40, Proposition 2.8]. The Gieseker stability of E implies that Hom(OX(kHX),E)=0 for k≥0. Since χ(OX(kHX),E)=0 for k=0,1, the definition of Ulrich bundles implies that Hom(OX(kHX),E[3])=0. We conclude that E∈Ku(X).
By using the embedding of Ku(X) in Db(P2,B0), the following existence result has been proved.
([92], Theorem B) The moduli space of stable Ulrich bundles of rank d≥2 on X is non-empty and smooth of dimension d2+1.
The property of Serre-invariance of σ0 and σ(s,q,μ) and the uniqueness result allows us to prove the irreducibility of the moduli space of Ulrich bundles.
([51], Theorem 6.1) The moduli space MdU of Ulrich bundles of rank d≥2 on X is irreducible.
(Idea of proof) The first step consists in proving that all Ulrich bundles E of rank d are σ(s,q,μ)-semistable. Since the condition of being Ulrich is open, we get an open embedding
where the latter is the moduli space of σ(s,q,μ)-semistable objects in Ku(X) with numerical class d[Iℓ] (see Section 5.1). Now using that σ(s,q,μ) is Serre-invariant and that σ0 is in the same GL~2+(R)-orbit of σ(s,q,μ), we have the isomorphisms between moduli spaces
where v:=[SKu(X)(Iℓ)] in N(Ku(X))⊂N(Db(P2,B0)).
The last step is to show that Mσ0(Ku(X),dv) is identified to the moduli space Md of Gieseker semistable sheaves in Coh(P2,B0) with class dv. A standard argument (see [92, Theorem 2.12], [70, Lemma 4.1]) whose key point is the vanishing of Ext2 for sheaves in this moduli space, allows us to show that Md is irreducible, and thus MdU is such (see [51, Proposition 6.4] for more details).
Let us now discuss some interesting properties of moduli spaces of stable objects in Ku(X) with respect to a Serre-invariant stability condition σ, e.g. for σ(s,q,μ) and σ0. We have already seen in Theorem 6.2 that the Fano surface of lines is identified with a moduli space of σ0-stable objects in Ku(X). We mention the following more general statement.
([51], Theorem 4.5, [129], Lemma 5.16) Let σ be a Serre-invariant stability condition on Ku(X).
The moduli spaces Mσ(Ku(X),[Iℓ]), Mσ(Ku(X),[SKu(X)(Iℓ)]), Mσ(Ku(X),[SKu(X)2(Iℓ)]) are isomorphic to the Fano surface of lines F1(X) in X.
Note that ±[Iℓ], ±[SKu(X)(Iℓ)], ±[Iℓ]-[SKu(X)(Iℓ)] are the only vectors of square -1 in N(Ku(X)). We also have the following result.
Let σ be a Serre-invariant stability condition on Ku(X). Then non-empty moduli spaces of σ-stable objects in Ku(X) are smooth.
Let E be a σ-stable object in Ku(X). Up to shifting, we can assume that E is in the heart of σ. Since σ is Serre-invariant, we have that SKu(X)(E) is σ-stable. By using the relation SKu(X)3=[5], it is not difficult to show (see [129, Lemma 5.9]) that the phases with respect to σ satisfy
As a consequence, by Serre duality we have
for i≥2, i<0. Indeed, the objects E[i], SKu(X)(E) are σ-stable and if i≥2, then ϕ(E[i])>ϕ(SKu(X)(E)) by (6.12). Thus there are no nontrivial morphisms from E[i] to SKu(X)(E). The vanishing for i<0 follows from E being in the heart. Since E is stable, we have that dimHom(E,E)=1, so dimExt1(E,E)=1-χ(E,E) is constant. Since Ext1(E,E) is identified with the tangent space to the moduli space at the point corresponding to E and the obstruction is given by Ext2(E,E) (see [93, 125, Lemma 4.8]) which is vanishing, we conclude that the moduli space Mσ(Ku(X),[E]), where [E]∈N(Ku(X)), is smooth of dimension 1-χ(E,E) at the point corresponding to E.
Last but not least, we focus on the Categorical Torelli theorem. Using the same strategy as in Theorem 6.4, it is possible to reprove this theorem by using any Serre-invariant stability condition and by applying the modular description of the Fano surface of lines by Theorem 6.28 (see [129, Theorem 5.17]). We end this section by explaining an alternative beautiful proof, given in [8], which is based on a description of the desingularization of the theta divisor of the intermediate Jacobian of X as a moduli space of stable sheaves on X, and then of semistable objects in Ku(X).
Recall that the intermediate Jacobian of X is the complex torus
It has the structure of a principally polarized abelian variety of dimension 5 and plays a key role in the seminal paper [42] for the proof of the (classical) Torelli Theorem and the nonrationality of X. If T denotes the closure in the Hilbert scheme of the subvariety parametrizing twisted cubic curves in X, then we have the Abel– Jacobi map α:T→J(X) defined by α(t)=t-HX2 (which is an element of J(X) through the cycle class map). Then by [16, Proposition 4.2] the image of α in J(X) is a theta divisor Θ⊂J(X) and its generic fiber is isomorphic to P2. In fact, the linear span of a twisted cubic C is ⟨C⟩≅P3, so C is contained in the cubic surface S=⟨C⟩∩X for C general. Then the generic fiber of α is the P2 of twisted cubic curves which are linearly equivalent to C on S.
Let MX(v′) be the moduli space of Gieseker stable sheaves on X with Chern character 3-H-12H2+16H3, hence having numerical class v′:=[Iℓ]+[SKu(X)(Iℓ)]. Note that the projection Kx in Ku(X) of the skyscraper sheaf of a point x∈X has numerical class v′ and it turns out to be σ-stable with respect to any Serre-invariant stability condition σ on Ku(X), which motivates the choice of this particular moduli space to recover the isomorphism class of X. Consider the Abel– Jacobi morphism β:MX(v′)→J(X) defined by β(E)=c2(E)-HX2, where c2(E) corresponds to the second Chern class c2(E) of E via the cycle map. The key ingredient is the following result.
([8], Theorem 7.1) The moduli space MX(v′) is smooth and irreducible of dimension 4. The Abel-Jacobi morphism β maps MX(v′) birationally onto the theta divisor Θ. More precisely, MX(v′) is the blow-up of Θ in its singular point. The exceptional divisor is isomorphic to the cubic threefold X, and parametrizes non-locally free sheaves in MX(v′).
The proof of the above theorem makes use of classical results on the properties of Θ together with techniques of wall-crossing to describe the objects in MX(v′). In particular, the embedding of X in MX(v′) is obtained by sending a point x∈X to the projection Kx in Ku(X) of the corresponding skyscraper sheaf. The objects Kx sweep a divisor in this moduli space. Also note that the Abel–Jacobi map α factors through β.
The second key ingredient is the identification of MX(v′) with a moduli space of semistable objects in Ku(X).
([8], Theorem 8.7) Let σ be a Serre-invariant stability condition on Ku(X). Then we have the isomorphism
(Idea of proof) Since σ is Serre-invariant, the Serre functor SKu(X) induces an isomorphism
where 2[Iℓ]-[SKu(X)(Iℓ]=SKu(X)-1(v′). The moduli space Mσ(Ku(X),2[Iℓ]-[SKu(X)(Iℓ])) has been fully described in [5, Theorem 1.2] (see also [8, Proposition 8.5]). The proof then follows by computing the image via SKu(X) of the objects.
As a consequence, the authors get a new proof of the Categorical Torelli Theorem 6.4 (see [8, Theorem 8.1]):
(New proof of Theorem 6.4) As usual, there is only one nontrivial implication to prove. Assume that there is an exact equivalence Φ:Ku(X1)→∼Ku(X2) (not necessarily of Fourier–Mukai type). Up to composing with a power of the Serre functor and shifting, we can assume Φ∗(v′)=v′, namely that Φ maps the numerical class of the projection Kx1 of the skyscraper sheaf of a point x1∈X1 to the numerical class of Kx2 for a point x2∈X2, using that the Serre functor acts transitively on all classes that have the same square as v′ with respect to χ (see [8, Lemma 8.3]). Since σ is Serre invariant and the Serre functor commutes with equivalences, we have that Φ·σ is Serre invariant. Theorem 6.31 applied two times gives the isomorphisms between moduli spaces
By [8, Lemma 7.5] we have that X1 and X2 are the unique rationally connected divisors in MX1(v′) and MX2(v′), respectively, which get contracted by any morphism to a complex abelian variety. Thus the above chain of isomorphisms implies X1≅X2.
After the success of the above techniques in the case of cubic threefolds, it is natural to wonder whether we can get similar results for other prime Fano threefolds of index 1 and 2 (see Section 3.3 for the classification and properties). The goal of this section is to summarize the state of art in these cases concerning the existence and uniqueness of Serre-invariant stability conditions and the Categorical Torelli theorem. We assume the base field is K=C for simplicity.
Let us start by considering the index 2 case. We denote by Yd a prime Fano threefold of index 2 and degree 1≤d≤55 and take its Kuznetsov component Ku(Yd):=⟨OYd,OYd(HYd)⟩⊥. The strategy of [14], reviewed in Section 6.3 for cubic threefolds, works more generally for Yd and allows one to construct stability conditions on Ku(Yd), which we denote by σ(s,q,μ). We have the following
The case d=1 remains mysterious. Recall that Y1 is a hypersurface of degree 6 in the weighted projective space P(1,1,1,2,3), known as the Veronese double cone. The Serre functor of Ku(Y1) satisfies SKu(Y1)3=[7] by [78, Corollary 4.2]. As a consequence, we cannot apply the criterion of Theorem 6.19 and the homological dimension of the heart of a Serre-invariant stability condition is 3. This makes this case quite different from the others. A remarkable difference is that this time the Fano surfaces of lines is an irreducible component of a moduli space of stable objects in Ku(Y1) and in [131] the authors classify all the objects in this moduli space. We can then formulate the following
Is there a unique GL~2+(R)-orbit of Serre-invariant stability conditions on Ku(Y1)?
Does the Categorical Torelli theorem holds for Y1?
We now focus on the index 1 case and denote by Xd such a prime Fano threefold of degree d=2g-2, where 2≤g≤12, g≠11. We first consider g≥6, so that the Kuznetsov component is defined by Ku(Xd):=⟨Er,OXd⟩⊥, where Er is an exceptional vector bundle of rank r on Xd (see Section 3.3). Again by [14], there are stability conditions σ(s,q,μ) on Ku(Xd) defined following the same procedure explained in Section 6.3 for cubic threefolds. In these cases we have the following
As noted in the introduction to [67], the Categorical Torelli theorem for cubic threefolds and the Kuznetsov’s correspondence, discussed in item (2) above, imply that the birational class of X14 is determined by its Kuznetsov component. Inspired by [90, Conjecture 1.7], we can formulate the following question, which is nothing but the Birational Categorical Torelli theorem for Gushel– Mukai
([67], Question 1.1(1)) Let X1 and X2 be prime Fano threefolds of index 1 and degree 10. Assume that there is an equivalence Φ:Ku(X1)→∼Ku(X2) (of Fourier– Mukai type). Is it true that X1 and X2 are birational?
As noted by the referee X12, X16, X18, X22 are rational (over an algebraically closed field), so the above question is void in these cases. Furthermore, this question has been positively answered in [67, Theorem 1.5] for general ordinary Gushel– Mukai threefolds.
Let us consider the remaining Fano threefolds Xd with d=2,4,6,8, where Ku(Xd):=⟨OXd⟩⊥. Even if the construction in [14] still applies and there are stability conditions σ(s,q,μ) on Ku(Xd), the situation is far less understood. Note that N(Ku(Xd)) has rank 3 in this case, so the stability manifold of Ku(Xd) has a bigger dimension than in the previous cases and Ku(Xd) does not look like a noncommutative curve. Up to now, we know the following
We can focus on the case of quartic threefolds and ask the following
If X is a quartic threefold, are the stability conditions σ(s,q,μ) Serre-invariant for some (s, q)?
Is there a version of the Categorical Torelli theorem for quartic threefolds? Does the birational Categorical Torelli theorem hold for quartic threefolds?
In these notes we have seen many methods to show the Categorical Torelli theorem for the Kuznetsov components of Enriques surfaces and cubic threefolds. The first tentative to approach Question 6.36 could be trying to adapt one of these techniques to the case of quartic threefolds. This is the line of investigation we are following in the work in progress [63].
Let us now move back to the case when X is a cubic fourfold defined over a field K which is algebraically closed with char(K)≠2. Let us recall from Section 3.4 that we have a semiorthogonal decomposition
where H is a hyperplane class. The aim of this section is to show that Ku(X) carries stability conditions which we will use to prove both a categorical and a classical Torelli theorem for these Fano fourfolds. We conclude this section with a brief discussion on Gushel– Mukai fourfolds (see Section 7.3) which, quite surprisingly, are at the same time very close in spirit to cubic fourfolds but very different for some key features.
In this section we want to prove that Ku(X) carries stability conditions. Ideally, we would like to apply the techniques discussed in Section 6.2 but what prevents us from a successful output is the fact that X has dimension 4 while the inducing strategy works perfectly fine for threefolds.
Thus the idea is to embed Ku(X) into the derived category of a new threefold whose construction is intimately related to the geometry of X as in Section 6.1.
More precisely, we follow [14, Section 7]. Now, as in the case of cubic threefolds, we pick a line ℓ⊆X which is not contained in a plane in X and we consider the projection πℓ:X⤏P3 onto a skew 3-dimensional projective plane. We further consider the blow-up X~:=Blℓ(X) of X in ℓ which makes the rational map πℓ into an actual morphism
whose fibers are conics.
On the categorical side, the conic fibration πℓ yields a sheaf B0 of even parts of Clifford algebras which is analogous to the one considered in Section 6.1 (in particular, it is noncommutative and generically Azumaya). As in Section 6.1 the fact that we can either view X as a conic fibration or as a blow-up, gives two fully faithful embeddings
A direct comparison between the two inclusions shows that the embedding of the Kuznetsov component can be made compatible by mutations with the one of Db(P3,B0). Hence, according to [14, Proposition 7.7], we get a semiorthogonal decomposition
where Ei is an exceptional B0-coherent sheaf. As in (6.4), we omit the explicit description of the embedding Ku(X)↪Db(P3,B0) which is indeed relevant for computations but useless for the purposes of this paper.
Despite the analogy between (6.4) for cubic threefolds and (7.1) for cubic fourfolds, the complexity of the twisted projective space (P3,B0) and of its derived category has secretly increased a lot. Indeed, while the numerical Grothendieck group of Db(P2,B0) has always rank 3 for all cubic threefolds (see [20, Proposition 2.12]), the rank of the numerical Grothendieck group of Db(P3,B0) varies when the cubic fourfold varies.
This is, indeed, not surprising because the Kuznetsov component behaves like a noncommutative curve (with a rather simple cohomology) for cubic threefolds while it behaves like a noncommutative K3 surface (hence with a rich cohomology) for cubic fourfolds.
In order to construct stability conditions on Ku(X) we are now in a good position to apply the dimension reduction trick described in Section 6.1: indeed, Ku(X) is now an admissible subcategory of the derived category of a (twisted) threefold with residue category generated by three exceptional objects.
To make this precise, we proceed as in the cubic threefold case and take the forgetful functor Forg:Db(P3,B0)→Db(P3). The twisted Chern character is then defined as
where L is the class of a line in P3. We denote by chB0,i the degree i component of chB0. Since this is a cohomology class of P3, it gets naturally identified with a rational number.
The mysterious numerical correction 1132 is needed to provide a Bogomolov inequality for the twisted derived category Db(P3,B0). The role of such an inequality is to give the correct quadratic inequality in the support property for the Kuznetsov component. This is similar to (5.4). For the purposes of this paper the numerical correction above can be ignored.
Next, we define
and consider the lattice ΛB0:=Im(v′). Finally, we denote by ΛB0,Ku(X) the image of the composition of the natural maps K(Ku(X))→K(Db(P3,B0))→ΛB0. The main result is then the following.
([14], Theorem 1.2, [13], Proposition 25.3) If X is a cubic fourfold, then there are stability conditions with moduli spaces on Ku(X) with respect to ΛB0,Ku(X).
It is worth pointing out that the result is actually more precise. Following Bridgeland’s notation, a full numerical stability condition on Ku(X) is a stability condition on Ku(X) with respect to the lattice HHodge(Ku(X),Z)≅N(Ku(X)) whose definition is recalled in Section 3.4 (see Remark 3.20). For later use, a Mukai vector is an element in the image of the restriction v:K(Ku(X))→HHodge(Ku(X),Z) of v′. We denote by Stabm(Ku(X)) the set of full numerical stability conditions with moduli spaces on Ku(X). Note that the twisted Chern character induces a natural map u:H~Hodge(Ku(X),Z)→ΛB0,Ku(X).
It is explained in [14, Section 9], that if σ=(A,Z) is a stability condition constructed in Theorem 7.3, then the pair σKu(X):=(A,Z∘u) is in Stabm(Ku(X)) which is then nonempty.
Theorem 7.3 and Propsition 6.9 yield the following
Let X be a cubic fourfold. Then Db(X) has a stability condition.
Furthermore, we consider the natural continuous map
defined in the following way. First, using the pairing on H~(Ku(X),Z)) (see Remark 3.19), we get a natural identification between Hom(HHodge(Ku(X),Z),C) and the vector space HHodge(Ku(X),C)=H~Hodge(Ku(X),Z)⊗ZC. Then η is nothing but the continuous map Z in Theorem 5.9 composed with such an identification.
Set now P⊆HHodge(Ku(X),C) to be the open subset consisting of vectors whose real and imaginary parts span positive-definite two-planes in HHodge(Ku(X),R). Then we set
where Δ:={δ∈H~Hodge(Ku(X),Z):(δ,δ)=-2}. Note that for σKu(X) as in Example 7.4 we have, by [14, Proposition 9.10],
Here A2 is the lattice in (3.10). Let P0+ be the connected component of P0 which contains η(σKu(X)), for σKu(X) as in Example 7.4. In addition, let Stab†(Ku(X)) be the connected component of Stabm(Ku(X)) which contains σKu(X).
The complete result is then the following.
([13], Theorem 29.1 and [14], Proposition 9.9) The preimage η-1(P0+) contains the connected component Stab†(Ku(X)). Moreover, the restriction η:Stab†(Ku(W))→P0+ is a covering map.
In this section we want to discuss the following result which is one of the main results of [60].
(Categorical Torelli theorem for cubic fourfolds) Let X1 and X2 be cubic fourfolds. Then X1≅X2 if and only if there is an equivalence Φ:Ku(X1)→∼Ku(X2) such that OX2∘Φ=Φ∘OX1.
Here OXi is the degree shift autoequivalence of Ku(Xi) described in Remark 3.18. Note that we do not require Φ being of Fourier–Mukai type. The assumption concerning the compatibility between the equivalence Φ and the degree shift functors OXi is crucial. Indeed, it was proved in [127, Theorem 1.1], that given any positive integer N one can find N non-isomorphic cubic fourfolds with equivalent Kuznetsov components. This is another striking similarity with the case of K3 surfaces (see, for example, [140]). It is then natural to raise the following question (see [111, Question 3.25] and the discussion therein):
(Huybrechts) Does a Birational Categorical Torelli theorem for cubic fourfolds hold? Namely, let X1 and X2 be cubic fourfolds. Is it true that the existence of an equivalence Ku(X1)→∼Ku(X2) implies that X1 and X2 are birational?
Note that Question 7.8 is compatible with Conjecture 3.16.
Note that the reverse implication of Question 7.8 is not true. More precisely, this amounts to asking whether two birational cubic fourfolds X1 and X2 have equivalent Kuznetsov components. However, as suggested by the referee, two general Pfaffian cubic fourfolds are birational (as both are rational), but their Kuznetsov components are not equivalent.
Of course, one could continue the analysis of the analogies with K3 surfaces. Assume that K=C. In this case, one knows that the derived categories of two K3 surfaces are equivalent if and only if there is an orientation preserving Hodge isometry of the Mukai lattices of the two surfaces (see [59, 121]). It is then natural to ask whether the same happens for the Kuznetsov components of cubic fourfolds (see [111, Question 3.24]).
We will skip this discussion and, in the rest of this section, we will deal with two proofs of Theorem 7.7 based on two different approaches.
Let us first explain the original approach in [60] which is close in spirit to the one in [49]. For this we have to stick to the case K=C.
Let us first introduce the main ingredients in the proof. If Y is a smooth hypersurface in Pn+1 described as the zero locus of a homogeneous polynomial F, then the Jacobian ring of Y is
If the degree d of Y is such that d≤n+1 (i.e. Y is a Fano manifold), then there is a semiorthogonal decomposition
where H is a hyperplane section. Let Ku(Y)(-(n+1-d)) be the admissible subcategory of Db(Y) obtained by tensoring Ku(Y) by OY(-(n+1-d)H).
Denote by Ku(Y)(-(n+1-d))⊠Ku(Y) the full subcategory of Db(Y×Y) which is generated by objects of the form E1⊠E2, where E1∈Ku(Y)(-(n+1-d)) and E2∈Ku(Y)6. Note that it is admissible by [79, Theorem 5.8]. Denote by
its fully faithful embedding and set P0:=jY∗OΔ, where jY∗ is the left adjoint of jY. Similarly, for a given n≥1, we set Pn to be the Fourier– Mukai kernel of the Fourier– Mukai functor obtained by composing OY with itself n times. It worth pointing out that OY is a Fourier– Mukai functor with Fourier– Mukai kernel given by the convolution of P0 and OΔ(1) [60, Remark 1.10]. By construction Pn∈Ku(Y)(-(n+1-d))⊠Ku(Y). Note that P0 and Pn are related by convolution, namely P1≅P0∘OΔ(1)∘P0 and Pn≅P1∘n by [60, Remark 1.11].
Now, assume that d>2, set N=(n+2)(d-2) and
This comes with the ring structure induced by the composition (see [60, Section 3.1]). Indeed, applying the convolution with Pi, we have a natural map RHom(P0,Pj)→RHom(Pi,Pi+j) and thus
requiring that the multiplication is trivial if i+j>N. Standard arguments show that this endows L(Y) with the structure of a graded ring. We denote by Lj(Y) the graded piece of degree j. We set HH∗(Ku(Y),OY) to be the graded subalgebra of L(Y) generated by L1(Y)7.
We are now ready to relate Jac(Y) and HH∗(Ku(Y),OY).
([60], Theorem 1.1) Let Y⊆Pn+1 be a smooth hypersurface of degree d≤n+22. Then there exists a natural surjection
of graded rings which is an isomorphisms if n+2 is divisible by d.
Now let Y1 and Y2 be smooth hypersurfaces in Pn+1 of degree 2≤d≤n+22 and assume that d divides n+2 so that πYi is an isomorphism, for i=1,2. If Φ:Ku(Y1)→∼Ku(Y2) is a Fourier– Mukai equivalence such that OY2∘Φ=Φ∘OY1, by [60, Proposition 3.9], Φ induces an isomorphism of graded algebras HH∗(Ku(Y1),OY1)≅HH∗(Ku(Y2),OY2). By Theorem 7.10, this lifts to an isomorphism of graded algebras Jac(Y1)≅Jac(Y2). By Yau– Mather theorem (see [49, Proposition 1.1], we get Y1≅Y2. Thus we
Let Y1 and Y2 be smooth hypersurfaces in Pn+1 of degree 2≤d≤n+22 and such that d divides n+2. Then Y1≅Y2 if and only if there is an equivalence of Fourier– Mukai type Φ:Ku(Y1)→∼Ku(Y2) such that OY2∘Φ=Φ∘OY1
Since cubic fourfolds satisfy the assumptions of Corollary 7.11, we immediately get Theorem 7.7 in view of Propsition 3.15.
We now outline the strategy of proof via stability conditions which is pursued in the appendix to [14] and, as explained later, allows one to get another proof of the classical Torelli theorem as well.
In this second proof, we can assume that K is an algebraically closed field with char(K)≠2. As above, there is only one implication in the statement that needs to be proved.
The strategy is very close in spirit to the one in Section 6.1 and Section 6.3 where we proved the Categorical Torelli theorem for cubic threefolds. Indeed, let X be a cubic fourfold and let F1(X) be the Fano varieties of lines in X. By [17], when K=C, the variety F1(X) is a 4-dimensional smooth and projective irreducible symplectic manifold (i.e. a simply-connected manifold such that H0(F1(X),ΩF1(X)2) is generated by an everywhere nondegenerate holomorphic 2-form). In general, as F1(X) is embedded in the Grassmannian of lines in P5, it carries a privileged ample polarization which is the restriction of the Plüker polarization. To shorten the notation we will refer to such a polarization on F1(X) as the Plücker polarization.
The key point is that one can interpret F1(X) as a moduli space of stable objects in the Kuznetsov component Ku(X). This approach was initiated in [113] and pursued in [14, 99]. Let X be a cubic fourfold and fix a stability condition σ∈Stab†(Ku(X)) such that η(σ)∈(A2⊗C)∩P⊆P0+. As we observed in (7.2) any stability condition constructed in the proof of Theorem 7.3 would work. By the general theory of moduli spaces of (semi)stable objects in the Kuznetsov component Ku(X) which we discussed in Section 5.1, we can take the moduli space Mσ(Ku(X),v) for every Mukai vector v. By the results in [13], any such moduli space Mσ(Ku(X),v) carries a natural ample polarization ℓσ.
The result is then the following (recall the class λ1 defined in (3.9) and that δ∈H~Hodge(Ku(X),Z) such that (δ,δ)=-2 is called (-2)-class):
([14], Theorem A.8) Let X be a cubic fourfold such that H~Hodge(Ku(X),Z) does not contain (-2)-classes. For any σ∈Stab†(Ku(X)) such that η(σ)∈(A2⊗C)∩P⊆P0+, the Fano variety of lines in X is isomorphic to the moduli space Mσ(Ku(X),λ1) of σ-stable objects with numerical class λ1. Moreover, the ample line bundle ℓσ on Mσ(Ku(X),λ1) is identified with a multiple of the Plücker polarization by this isomorphism.
In [99, Theorem 1.1] the previous result has been generalized by showing that the Fano variety of lines of every cubic fourfold is isomorphic to a moduli space of σ-stable objects for σ as in Section 7.1.
Now, let X1 and X2 be cubic fourfolds with an equivalence Φ:Ku(X1)→∼Ku(X2) commuting with the rotation functors. By Propsition 3.15, Φ is of Fourier– Mukai type. Then Φ induces a Hodge isometry
between the Mukai lattices. It is a simple exercise, using our assumption OX2∘Φ=Φ∘OX1 and [57, Proposition 3.12], to show that ΦH sends the A2-lattice of X1 to the corresponding one of X2.
It is clear that, for this argument, one can weaken the assumption OX2∘Φ=Φ∘OX1 in Theorem 7.7 to its cohomological version OX2H∘ΦH=ΦH∘OX1H. Here we have that OXiH:HHodge(Ku(Xi),Z)→∼HHodge(Ku(Xi),Z) denotes the Hodge isometry induced by OXi.
Recall that if X is a cubic fourfold, the middle cohomology H4(X,Z) has a natural lattice and Hodge structure (see [56] for an excellent introduction). If H is the class of a hyperplane section then the self-intersection H2 is an algebraic class in H4(X,Z). Then we denote by Hprim4(X,Z) the orthogonal to H2 in H4(X,Z). Clearly, such a sublattice inherits a Hodge structure from H4(X,Z). By [2] there is (up to Tate twist) a Hodge-isometry
In our setting, using the above identification and the fact that ΦH preserves the A2-lattices of X1 and X2, it follows that ΦH induces a Hodge isometry
Theorem 7.7 then follows from the following beautiful classical result. Its first proof was a masterpiece in Hodge theory due to Voisin [145]. Alternative and more recent proofs are due to Looijenga [104] and Charles [41].
(Classical Torelli theorem) Two smooth complex cubic fourfolds X1 and X2 are isomorphic if and only if there exists a Hodge isometry Hprim4(X1,Z)≅Hprim4(X2,Z).
Since we promised that the second proof of Theorem 7.7 would have been based on stability conditions, we are going to provide an alternative proof of Theorem 7.15 based on these techniques and following the Appendix of [14].
Of course, if X1≅X2, then there is a Hodge isometry Hprim4(X1,Z)≅Hprim4(X2,Z). For the other implication, we start with a Hodge isometry
The argument proceeds now by taking a local deformation of Xi. Indeed, as explained in [60], in view of the local Torelli theorem, φ extends to the bases of the universal deformation spaces Def(X1)≅Def(X2), which are considered as open subsets of the period domain. More precisely, one can find an identification Def(X1)≅Def(X2) such that parallel transport induces a Hodge-isometry
where Xi,d is the local deformation of Xi parametrized by d∈Def(Xi). Then a lattice theoretic argument (see [60, Proposition 4.2]) shows that for every d∈Def(X1)≅Def(X2) the Hodge-isometry φd lifts to an orientation preserving Hodge isometry
which commutes with the action of the degree shift functors on the Mukai lattices.
Consider the set Di of points of Def(Xi) corresponding to cubic fourfolds X such that Ku(X)≅Db(S,α), for S a K3 surface and α∈Br(S), and H~Hodge(Ku(X),Z) does not contain (-2)-classes. Since the condition of having Kuznetsov component equivalent to the bounded derived category of a twisted K3 surface is determined by the Mukai lattice [57, Theorem 1.4], [13, Proposition 33.1], we see that ϕd preserves this property. Analogously, ϕd preserved the property of not having (-2)-classes. We thus conclude that Def(X1)≅Def(X2) restricts to an isomorphism D1≅D2. Thus we can set D:=D1≅D2. As explained in the appendix to [14], the set D is dense. Moreover, for all d∈D, the Hodge isometry ϕd can be lifted to an equivalence
by the derived Torelli theorem for twisted K3 surfaces [62, Theorem 0.1]. By construction, the isometry ΦdH commutes with the action of the degree shifts functors in cohomology.
Now, for d∈D, pick σ1∈Stab†(Ku(X1,d)) such that η(σ1)∈(A2⊗C)∩P⊆P0+ and set σ2:=Φd(σ1). By [58, Theorem 1] the stability manifold Stab†(Ku(Xi,d)) has a unique connected component of maximal dimension. Since the action of Φd on the stability manifolds exchanges components of the same dimension, it follows that the image σ2 of σ1 belongs to Stab†(Ku(X2,d)). Moreover, by definition, η(σ2)∈(A2⊗C)∩P as well. Thus we can apply Theorem 7.12 twice and obtain a string of isomorphisms
The isomorphism in the middle is induced by Φd and, as explained in [14, Appendix A], it sends the polarization ℓσ1 to ℓσ2. Thus, by Theorem 7.12, the whole sequence of isomorphisms sends the Plücker polarization on F1(X1,d) to the Plücker polarization on F1(X2,d). The proof then continues as in the one of Theorem 6.4 and it consists in applying Chow’s trick (see [41, Proposition 4]) in order to conclude that X1,d≅X2,d for all d∈D. Since D is dense, separatedness of the moduli space of cubic fourfolds implies that X1 and X2 have to be isomorphic (since any open neighborhood of the point corresponding to X1 in Def(X1) intersects any open neighborhood of the point corresponding to X2 in Def(X2), precisely in the points which belong to D). This ends the proof.
We could wonder whether the techniques explained in the previous sections may be adapted to other classes of Fano fourfolds. This turns out to be true in the case of Gushel– Mukai fourfolds (more generally for Gushel– Mukai varieties of even dimension).
Recall that a general complex Gushel– Mukai (GM) fourfold X is a smooth four-dimensional intersection of the form
where Gr(2,5) is the Plücker embedded Grassmannian and Q is a quadric hypersurface in a hyperplane section of P9. GM fourfolds share many similarities with cubic fourfolds. For instance, from a geometric viewpoint, there are known examples of rational GM fourfolds, but it is still unknown whether the very general one is irrational or rational. From the point of view of derived categories, Kuznetsov and Perry proved in [89] that the bounded derived category of X has a semiorthogonal decomposition of the form
where UX is the restriction of the tautological bundle of rank 2 on the Grassmannian Gr(2,5) and Ku(X):=⟨OX,UX∨,OX(1),UX∨(1)⟩⊥. The residual component Ku(X) is the Kuznetsov component of X and is a noncommutative K3 surface.
Full numerical stability conditions on Ku(X) have been constructed in [126], using a dimension reduction trick. More precisely, the authors show that X is birational to a conic fibration over a quadric threefold Y and then provide an embedding of Ku(X) in the bounded derived category Db(Y,B0), where (Y,B0) is a twisted quadric threefold.
On the other hand, it is known that the Torelli Theorem does not hold for GM fourfolds. In fact, in this case the period map has four-dimensional fibers by [45]. Nevertheless, we can still wonder whether a (refined) Categorical Torelli theorem holds for GM fourfolds.
More precisely, note that GM fourfolds in the same fiber of the period map have equivalent Kuznetsov components by [90, Theorem 1.6]. On the other hand, by [32, Theorem 1.3] there are examples of GM fourfolds with equivalent Kuznetsov components, but defining different period points. These considerations suggest that we have to impose some additional conditions to an equivalence between the Kuznetsov components to recover the period point or the isomorphism class of a GM fourfold.
Recall that the degree shift functor of Ku(X) is defined by
where L⟨OX,UX∨⟩ is the left mutation in ⟨OX,UX∨⟩ (see, for example, [78, Section 2] for the precise definition). We can formulate the following
Let X1 and X2 be Gushel– Mukai fourfolds. Assume that there is an equivalence Φ:Ku(X1)→∼Ku(X2) which commutes with the degree shift functors of X1 and X2. Then under which assumptions on Φ we have that X1 and X2 are isomorphic?
To address this question, it may be helpful to use the stability conditions defined on Ku(X) and the associated moduli spaces having the structure of hyperkähler manifolds [126], as done for cubic fourfolds in the last part of Section 7.2.
Let X1 and X2 be Gushel– Mukai fourfolds. Assume that there is an equivalence Φ:Ku(X1)→∼Ku(X2) which commutes with the degree shift functors. One could ask the intermediate question whether Φ induces a Hodge isometry Hvan4(X1,Z)≅Hvan4(X2,Z) between the degree-four vanishing cohomologies of X1 and X2 (see [46, Section 3.3] for the definition). This is answered positively by [15, Proposition 1.12].
Note that if X1 and X2 are very general, then the above statement holds for every exact equivalence Φ:Ku(X1)→∼Ku(X2). This can be proved exactly as in [57, Corollary 3.6]. Indeed, by [98] every equivalence as above is of Fourier-Mukai type. Then by [57, Corollary 3.5], it induces a Hodge isometry ΦH between the Mukai lattices of Ku(X1) and Ku(X2). Note that the condition of being very general means that the algebraic part of the Mukai lattice of Ku(Xi) is generated by two classes λ1, λ2 (see [89] and [128, Lemma 2.4] for the precise definitions). As a consequence, the Hodge isometry ΦH restricts to a Hodge isometry Hvan4(X1,Z)≅Hvan4(X2,Z). Of course, we cannot say X1 and X2 are isomorphic (and in general they are not), since there is no Torelli theorem.
We finally recall the following conjecture about Birational Categorical Torelli theorem for Gushel– Mukai varieties. The same comment in Remark 7.9 applies to this setting.
([90], Conjecture 1.7) If X1 and X2 are Gushel– Mukai varieties of the same dimension such that there is an equivalence Ku(X1)→∼Ku(X2), then X1 and X2 are birational.
This project started when the second author was visiting the Max Planck Institute (Bonn) whose hospitality is gratefully acknowledged. It is our pleasure to thank Arend Bayer, Chunyi Li, Zhiyu Liu, Alex Perry, Emanuele Macrì for carefully reading a preliminary version of this paper and for many insightful comments. We are mostly grateful to the anonymous referee who carefully read this paper and helped us with several interesting comments which allowed us to improve the presentation.
Open access funding provided by Università degli Studi di Milano within the CRUI-CARE Agreement.
Laura Pertusi, Email: laura.pertusi@unimi.it, //www.mat.unimi.it/users/pertusi.
Paolo Stellari, Email: paolo.stellari@unimi.it, //sites.unimi.it/stellari.