Examples of smooth components of moduli spaces of stable sheaves

Let M be a projective fine moduli space of stable sheaves on a smooth projective variety X with a universal family E. We prove that in four examples, E can be realized as a complete flat family of stable sheaves on M parametrized by X, which identifies X with a smooth connected component of some moduli space of stable sheaves on M.


Introduction
Background. The starting point of the article is a classical result on the moduli space of stable vector bundles on curves. Let C be a smooth complex projective curve of genus g 2. We denote the moduli space of stable vector bundles on C of rank n with a fixed determinant line bundle L d of degree d by M.
If n and d are coprime, then it is known by [MN68,Tju70] that M is a fine moduli space, namely, there exist a universal vector bundle E on C × M with the property that the fiber E| C×{m} over a closed point m = [E] ∈ M is isomorphic to the bundle E itself. But one can also take a closed point c ∈ C and consider the fiber which is a vector bundle on M. In [NR75] the authors proved that E c is a simple bundle for every closed point c ∈ C and that the infinitesimal deformation map T c C −→ Ext 1 M (E c , E c ) is bijective. In fact, for all closed points c ∈ C, the bundles E c are stable and pairwise non-isomorphic by [BBPN97,LN05].
Thus if we define M to be the moduli space of stable vector bundles on M, acquiring the same numerical classes as E c , then the classifying morphism identifies C with a smooth connected component of M, as explained in [LN05].
Other examples in a similar spirit appear in the pioneering work of Mukai [Muk81,Muk99] on abelian varieties and K3 surfaces. In the case of K3 surfaces, Mukai considered a general polarized K3 surface S of a certain degree, along with a 2dimensional fine moduli space M of stable vector bundles of rank at least 2 on S, admitting a universal family E on S × M. It turns out that M is also a K3 surface, and E can also be realized as a family of stable bundles on M parametrized by S.  Question 0.1. Let X be a smooth projective variety and M a projective fine moduli space of stable sheaves on X with universal family E on X × M. Then • Is E also a flat family of stable sheaves on M parametrized by X?
• If so, does the classifying map embed X as a smooth connected component of some moduli space of stable sheaves on M?
A positive answer to the above question, especially when X is of low dimension and M is of higher dimension, would be interesting from two perspectives. • X is a smooth projective variety of dimension d 2 and M = Hilb 2 (X) is the Hilbert scheme of 2 points on X; • X is K3 surface and M = Hilb n (X) is the Hilbert scheme of n points on X; • X is an abelian surface and M = Kum n (X) is the generalized Kummer variety of dimension 2n associated to X for any n 2; • X is a K3 surface of Picard rank 1 and M is some fine moduli space of stable torsion sheaves of pure dimension 1 on X.
Our proof in the first of the above cases will be completely elementary. In all other cases, the moduli space M is in fact an irreducible holomorphic symplectic manifold, and our proof will be divided into two steps: we first establish the flatness of E over X and the stability of the fibers E p over any closed point p ∈ X, then apply some very convenient results about P n -functors (see [Add16]) to conclude that X is in fact a component of some moduli space of stable sheaves on M.
It would be much more interesting to study Question 0.1 in more general settings, especially when X and M have trivial canonical classes and E is torsion free (or even locally free) of higher rank. However, it could be then much more difficult to prove the stability of E p for any closed point p ∈ X. Moreover, the corresponding results about P n -functors are not yet known to us (see [ This article consists of four sections, which are devoted to the four cases in Theorem 0.2 respectively. The notion of P n -functors will be briefly recalled in the beginning of §2, followed immediately by a list of P n -functors relevant to our discussion. All schemes are defined over the field of complex numbers C.

Hilbert squares of smooth projective varieties
Let X be a smooth projective variety of dimension d, and M = Hilb 2 (X). We denote by Z ⊆ X × M the universal closed subscheme and I Z the universal ideal sheaf on X × M. Then we have a commutative diagram where π is a flat morphism.
By [FGI + 05, Remark 7.2.2.], we have Z = Bl ∆ (X × X), the blow-up of X × X along the diagonal ∆. The projection τ can be interpreted as a composition of the blow-up b and the projection q 1 to the first factor. Moreover, the group Σ 2 = Z/2Z acts on Z by switching the two factors, with a fixed-locus given by the exceptional divisor. By [FGI + 05, Example 7.3.1(3)], π is the quotient of Z by Σ 2 .
For any closed point p ∈ X, we write Then we have the following results regarding the fibers of τ : Lemma 1.1. We have S p ∼ = F p ∼ = Bl p (X), and the morphism τ is flat.
Proof. The morphism π| Fp can be factored into a composition hence π induces an isomorphism from F p to its image S p . The canonical isomorphism F p ∼ = Bl p (X) is well known. Finally, since Z and X are both smooth and the fibers F p of τ are irreducible of dimension d for all closed points p ∈ X, we deduce from [Sch77, Lemma, p.675] that τ is flat.
By the description of F p as a blow-up in Lemma 1.1, we denote the exceptional divisor by E p α ֒−→ F p , then E p ∼ = P d−1 . This allows us to state the following result: Lemma 1.2. π −1 (S p ) has simple normal crossing singularities with two irreducible components Proof. This property can be verified analytically locally. Without loss of generality we assume that X = A n , and p = (0, · · · , 0) ∈ X. Then X × X = A n × A n with coordinates (x 1 , · · · , x n , y 1 , · · · , y n ). We perform an affine change of coordinates: for each 1 i n, we write s i = x i + y i and d i = x i − y i . Then the diagonal ∆ is given by ∆ = {(s 1 , · · · , s n , d 1 , · · · , d n ) | d 1 = · · · = d n = 0}.
Therefore we have g(F 1 p ) = (s 1 , · · · , s n , d 1 , u 2 , · · · , u n ) s 1 − d 1 = 0 s i = u i s 1 for 2 i n and the quotient Bl ∆ (X × X) 1 /Σ 2 is given by coordinates where e 1 = d 2 1 . We write the image of F 1 p under the quotient map by then it follows that S 1 p = (s 1 , · · · , s n , e 1 , u 2 , · · · , u n ) It is now clear that Therefore the intersection of the two components is transverse, and given by which gives precisely the exceptional divisor E p in the first affine chart, namely, The same argument also applies to all other affine charts of Bl ∆ (X × X), which finishes the proof.
In the following discussion, for any closed embedding U ֒→ V , we denote the corresponding ideal sheaf, conormal sheaf and normal sheaf by I U/V , C U/V and N U/V respectively. We consider now a smooth variety W and two smooth closed subvarieties Y and Z, which fit in the following commutative diagram of closed embeddings: where the intersection and the union are scheme theoretic. The following lemma will be required in our next result: Proof. We obtain by the second isomorphism theorem that By the third isomorphism theorem, this is equivalent to Since β is a closed embedding, this implies Therefore we obtain In our situation we pick W = Z, Y = g(F p ) and Z = F p in (3), then the morphism α becomes E p α ֒−→ F p . Lemma 1.3 immediately yields The following result is the key to the main theorem of this section: Proof. We divide the proof in two steps.
Step 1. We claim that N Sp/M fits into the exact sequence We consider the chain of closed embeddings By [Gro67, Proposition 16.2.7], we get the exact sequence of conormal sheaves Furthermore since S p ֒→ M is a regular embedding of codimension d, the sheaf C Sp/M is locally free of rank d. It follows by the flatness of π : Z → M that is also locally free of rank d. Therefore the first two terms in (5) are locally free sheaves of rank d and the third one is by Corollary 1.4 torsion with support E p . It follows that the first arrow in (5) is injective. By dualizing (5) we obtain Together with (6), (7) and Corollary 1.4 we obtain the claim (4).
Step 2. We claim that Dualizing this exact sequence shows Using E p ∼ = P d−1 we finally get: We conclude the proof by combining the long exact sequence in cohomology associated to (4) and the vanishing result (8).
The following lemma is the main source for finding components of moduli spaces. The proof follows literally from [BBPN97, Theorem 3.6].
Lemma 1.6. Let X be a smooth projective variety of dimension d and Y a projective scheme. Assume that a morphism f : X → Y is injective on closed points, and , it follows that Y is smooth of dimension d at each closed point y ∈ f (X) by [GW10, Theorem 6.28], hence f (X) must be a smooth irreducible component of Y , which is also a connected component of Y .
Finally, since f : X → f (X) is a morphism between smooth projective varieties and bijective on closed points, it is an isomorphism by Zariski's Main Theorem.
Combining the above results, we can now give our first main result: Theorem 1.7. Any smooth projective variety X of dimension d 2 is isomorphic to a smooth connected component of a moduli space of stable sheaves with trivial determinants on Hilb 2 (X), by viewing I Z as a family of coherent sheaves on Hilb 2 (X) parametrized by X.
Proof. By Lemma 1.1, Z is flat over X hence I Z can be viewed as a flat family of sheaves on Hilb 2 (X) parametrized by X. For each closed point p ∈ X, let (I Z ) p be the restriction of I Z on the fiber {p} × Hilb 2 (X). Then (I Z ) p is the ideal sheaf I Sp of the closed embedding of S p into Hilb 2 (X), hence is a stable sheaf of rank 1. Therefore we obtain an induced classifying morphism where M denotes the moduli space of stable sheaves on Hilb 2 (X) of the class of I Sp with trivial determinants. By [KPS18, Lemma B.5.6], M is isomorphic to the Hilbert scheme of subschemes of Hilb 2 (X) which have the same Hilbert polynomials as S p . It is easy to see that f is injective on closed points. On the other hand, for any closed point p ∈ X, we have Hence by Lemma 1.5, we have Therefore we conclude by Lemma 1.6 that the morphism (9) embeds X as a smooth connected component of M.

Hilbert schemes of points on K3 surfaces
What is particular interesting to us is the case of K3 surfaces. The technique of P nfunctors allows us to obtain similar results for their Hilbert schemes of 0-dimension subschemes of arbitrary length. We first recall the following notion of P n -functors and its implications.  We will focus on the case where A = D b (X) and B = D b (Y ) for two smooth projective varieties X and Y such that F = Φ F is an integral functor with kernel F ∈ D b (X × Y ). In fact, we are mostly interested in the case where F is actually a sheaf on X × Y and the autoequivalence H = [−2]. In this case condition (a) can be stated as RF ∼ = id ⊗H * (P n , C).
We will use the following simple consequence under this setting Proposition 2.2. [ADM16, §2.1] Assume X and Y are smooth projective varieties and F is a coherent sheaf on X × Y , flat over X, such that the integral functor F = Φ F with kernel F is a P n -functor with associated autoequivalence H = [−2]. Then for any closed points x, y ∈ X there is an isomorphism: where F x and F y are fibers of F over the closed points x and y respectively.
The following list of P n -functors will be of interest to us: We give a first application of P n -functors to our problem: let S be a K3 surface and M = Hilb n (S) for some positive integer n. Then M is a fine moduli space and the ideal sheaf I Z of the universal family Z is the universal sheaf on S × M. It is well-known that M is an irreducible holomorphic symplectic manifold. The flatness of I Z over S follows immediately from the following result: In particular, when s 0 = s 1 , it follows from (11) that which implies that (10) is injective on closed points; when s 0 = s 1 = s, it follows from (11) that Therefore we conclude by Lemma 1.6 that the morphism (10) embeds S as a smooth connected component of M, as desired.

Generalized Kummer varieties
In this section we apply the technique of P n -functors to study a component of the moduli space of stable sheaves on generalized Kummer varieties.
Let A be an abelian surface and Hilb n+1 (A) the Hilbert scheme parametrizing closed subschemes of A of length n + 1. Let the morphism Σ be the composition of the Hilbert-Chow morphism and the summation morphism with respect to the group law on A, namely Proof. It suffices to show that the morphism ϕ : Z → A is flat. First of all, we claim that the fiber ϕ −1 (a 0 ) is of dimension 2n − 2 for any closed point a 0 ∈ A.
On the one hand, since A is smooth, the closed point a 0 ∈ A is locally defined by two equations. Therefore locally near any point x ∈ ϕ −1 (a 0 ), the fiber ϕ −1 (a 0 ) is also defined by two equations, hence is of codimension at most 2 by Krull's height theorem; see [Mat80, §12.I, Theorem 18]. In other words, we have On the other hand, we have For any such ξ, we can write the associated 0-cycle [ξ] as where a 0 , a 1 , · · · , a k are pairwise distinct closed points, and n 1 · · · n k > 0 are the multiplicities. It is clear that we have k i=0 n i = n + 1 (13) which in particular implies k n, and which utilizes the group law on A. We call the partition of n n = (n 0 , n 1 , · · · , n k ) the type of ξ. Let ϕ −1 (a 0 , n) the set of all closed points ξ ∈ ϕ −1 (a 0 ) of type n, then we have a decomposition We then compute the dimension of ϕ −1 (a 0 , n) for each n.
When k 1, every ξ ∈ ϕ −1 (a 0 , n) corresponds to a configuration {a 1 , · · · , a k } of pairwise distinct points satisfying (14). We can choose the first (k − 1) points freely, then a k is uniquely determined up to n k -torsion. Hence there is a 2(k − 1)dimensional family of configurations {a 1 , · · · , a k }. For any fixed configuration, the possible scheme structures on ξ is classified by the product of punctual Hilbert schemes Hilb n 0 a 0 (A) × · · · × Hilb n k a k (A). By [Iar72, Corollary 1] and (13), we obtain Combining the two cases, we have by (15) that It then follows from (12) and (17) that all fibers ϕ −1 (a 0 ) are equidimensional of dimension 2n − 2.
Moreover, since ψ is a surjective flat morphism and Kum n (A) is smooth of dimension 2n, we know Z is Cohen-Macaulay of dimension 2n by [Eis95,Corollary 18.17]. Since A is smooth, we conclude that ϕ : Z → A is flat by [Sch77, Lemma, p.675], which implies that its ideal sheaf I Z is flat over A, as desired.
Remark 3.2. It is easy to see that the statement of Lemma 3.1 fails for n = 1, due to the failure of (16). In fact, in such a case, ϕ −1 (a 0 ) is either a smooth rational curve or a single point, depending on whether a 0 is a 2-torsion point of A.
The above result allows us to obtain a smooth component of the moduli space of stable sheaves on Kum n (A) as follows: where M denotes the moduli space of all stable sheaves on Kum n (A) of the class of (I Z ) a 0 . For any pair of closed points a 0 , a 1 ∈ A, we obtain by [Mea15,Theorem 4.1] and Proposition 2.2 that From here, a similar argument as in Theorem 2.4 shows that the morphism (18) embeds A as a smooth connected component of M.

Moduli spaces of pure sheaves on K3 surfaces
In this section we extend our discussion to the fine moduli spaces of stable sheaves of pure dimension 1 on a K3 surface of Picard number 1.
Let S be a K3 surface with Pic(S) = ZH where H is an ample line bundle of degree 2g − 2. Let P(V ) ∼ = P g be the complete linear system of H where V = H 0 (S, H). Since S has Picard number 1, every curve C in the linear system P(V ) is reduced and irreducible of genus g with planar singularities, hence its compactified Jacobian Jac d (C) is reduced and irreducible of dimension g by [AIK77, Theorem (9)]. We denote by C the universal curve of the linear system P(V ).
Let M be the moduli space of stable sheaves on S with Mukai vector v = (0, H, d + 1 − g).
We assume gcd(2g − 2, d + 1 − g) = 1, then M is a smooth fine moduli space of stable torsion sheaves of pure dimension 1, hence admits a universal family U. In fact, M is an irreducible holomorphic symplectic manifold. The corresponding support morphism sends a stable sheaf to its support curve.
Alternatively, M can also be interpreted as the relative compactified Jacobian Jac d (C/P(V )) of the family C → P(V ). Hence the support of the universal family U is given by It is more convenient to consider the universal family as a sheaf on T , so we define where ι : T ֒→ S × M is the closed embedding. Then we have U ∼ = ι * E by [GW10,Remark 7.35].
The relation among the various spaces and morphisms introduced above can be summarised in the following commutative diagram where both squares on the left are cartesian.
Moreover, for any closed point s ∈ S, we denote the fiber ψ −1 (s) by T s , with the corresponding closed embedding i s : T s ֒→ T . We also denote the pullback of E to the fiber T s by E s , and the pullback of U to the fiber {s} × M by U s .
The following properties will be used later: Lemma 4.1. Both T and T s for each closed point s ∈ S are Gorenstein.
Proof. We first note that C is smooth and irreducible of dimension g + 1, as it is a P g−1 -bundle over S. Moreover, since both M and P(V ) are smooth, and all closed fibers of η are compactified Jacobians, which are irreducible of dimension g, the morphism η is flat by [Sch77,Lemma,p.675]. It follows that ϕ is also flat, and every closed fiber of ϕ is reduced and irreducible. Therefore T is reduced and irreducible of dimension 2g + 1 by [ For any closed point s ∈ S, the restriction of ϕ to the fibers over s is given by We can apply a similar argument as above on ϕ s to show that T s is Gorenstein.
Now we turn to properties of the universal sheaf: Lemma 4.2. The sheaf E on T is flat over S, and the sheaf E s on T s is stable for each closed point s ∈ S.
Proof. We observe that the morphism C → P(V ) (the composition of the morphisms in the middle column of (20)) is projective, flat and Gorenstein of pure dimension 1. After the base change along η, the morphism π : T → M (the composition of the morphisms in the left column of (20)) is also projective, flat and Gorenstein of pure dimension 1. Furthermore E is flat over M, and for any point m ∈ M, the restriction of E to the fiber π −1 (m) is torsion free. It follows by [BK06, Corollary 2.2] that Ext i T (E, O T ) = 0 for every i > 0. Since T is irreducible and Gorenstein, this implies that E is a maximal Cohen-Macaulay sheaf on T .
We have seen that ϕ and τ are both flat morphisms, hence ψ is also a flat morphism. The closed embedding {s} ֒→ S is a morphism of finite Tor dimension. After a flat base change along ψ, we see that i s : T s ֒→ T is also of finite Tor dimension. Since T is irreducible and Gorenstein by Lemma 4.1, [Ari13, Lemma 2.3 (1)] implies Li * s E = i * s E for every closed point s ∈ S, where Li * s is the derived pullback functor. It follows by [Huy06,Lemma 3.31] that E is flat over S.
By Lemma 4.1 we also know T s is Gorenstein, hence is in particular Cohen-Macaulay. By [Ari13, Lemma 2.3 (2)], E s is also maximal Cohen-Macaulay, which by [HK71, Satz 6.1, a) ⇒ d)] implies that E s is reflexive, and hence in particular torsion free on T s . Therefore E s is stable since it is of rank 1.
The above result allows us to obtain again a smooth component of the moduli space of stable sheaves on M as follows: Proof. By Lemma 4.2, we know that the sheaf U = ι * E is also flat over S, and the fiber U s is a stable sheaf on M of pure dimension 1 for each closed point s ∈ S. Therefore U is a flat family of stable sheaves on M parametrized by S, with an induced classifying morphism given by where M is the moduli space of all stable sheaves on M of the class of U s . For any pair of closed points s 0 , s 1 ∈ S, we obtain by [ADM16, Theorem A] and Proposition 2.2 that Ext * M (U s 0 , U s 1 ) ∼ = Ext * S (O s 0 , O s 1 ) ⊗ H * (P g−1 , C).
From here, a similar argument as in Theorem 2.4 shows that the morphism (20) embeds S as a smooth component of M.