Abstract
We discuss the space VPS(Q, H) of ideals with Hilbert function that are apolar to a full rank quadric Q. We prove that its components of saturated ideals are closely related to the locus of Gorenstein algebras and to the Slip component in border apolarity. We also point out an important error in [Citation20] and provide the necessary corrections.
1 Introduction
For a fixed nondegenerate quadric q in n variables, a polar simplex is a tuple of points, where are such that . The locus of all polar simplices is a principal homogeneous space for the group . Informally speaking, the variety of polar simplices is a compactification of this locus. As we discuss below, there is actually more than one possible compactification and choosing the correct one is subtle. The variety is important for two main reasons. First, it is a “simplest” example of a variety of sums of powers [Citation19], which are useful as examples of special projective varieties and for applications in tensors [Citation3–5, Citation10, Citation14, Citation21]. Second, as explained below, it serves as a mean to investigate the Gorenstein locus of the Hilbert scheme of points: it has smaller dimension, which is important for computations such as [Citation16] and at the same time contains each abstract Gorenstein subscheme of degree n.
Let be a polynomial ring which we view as a homogeneous coordinate ring of the . Fix a number d and a nonzero , then . A finite subscheme is apolar to F if lies in the span of the d-uple reembedding of Γ in . Apolarity may also be formulated algebraically, see Section 2, by requiring that , the ideal of forms in S that annihilates f by differentiation. This condition may be formulated for ideals in general; an ideal is apolar to F if .
Consider the special case d = 2 and being a full rank quadric. Let denote the closure of the locus of degree n zero-dimensional schemes satisfying . The result [Citation20, Corollary 2.2] and [Citation23, Proposition 6.5] both state that every is apolar to Q. It is not so, as the following example shows.
Example 1.1.
Let n = 4, let , where yi is the dual to xi, and consider the ideal
This ideal is saturated and has Hilbert function for every , hence for a finite, degree 4 subscheme ; in particular . Consider now a -action on corresponding to the grading by . The quadric Q is a semi-invariant for this action, so for every , the subscheme corresponds to a point in with ideal given by
The limit of those ideals at , taken degree by degree, is the ideal whose saturation is . Let , then is a limit of for , so while is not contained in , so is not apolar to Q.
The example implies in particular that several of the main results in [Citation20] are wrong as stated. The purpose of this article is to both discuss what can salvaged and how and, which is of independent interest, show unexpected connections between and the moduli of Gorenstein algebras.
Weronika Buczyńska and Jarosław Buczyński introduced the multigraded Hilbert scheme into the subject of apolarity and observed that it is better than the usual Hilbert scheme also when one considers VPS [Citation3, Section 7.6].
The multigraded Hilbert scheme , defined in [Citation15], parametrizes homogeneous ideals with a given Hilbert function H. For a fixed form f, the condition for an ideal I is a closed condition in the multigraded Hilbert scheme. Example 1.1 above shows that the apolarity condition is not closed in the usual Hilbert scheme. Hence, it is more natural to work in the multigraded Hilbert scheme.
We consider the Hilbert function and, for the full rank quadric Q, define
This is a closed subscheme of . Let be the locus of saturated ideals. This locus is open by [Citation17, Theorem 2.6]. Let be its closure, so is a union of irreducible components of . Let with its reduced scheme structure. In analogy with the locus of smoothable schemes in the Hilbert scheme, we say that is the locus of ideals that are Satura BLe, this is consistent with the notation of [Citation17] thanks to Corollary 3.9. The intersection is the locus of unsaturated limit ideals: apolar unsaturated ideals that are limits of saturated apolar ideals.
Associating to each ideal the space I2 of quadrics in the ideal defines a forgetful map
into the Grassmannian of -dimensional subspaces in . We denote
In this paper we will be concerned with these loci. The map πG restricted to is an isomorphism (see Proposition 4.5), while the restriction to the other loci, in general, is not.
The scheme is different from in general, and similarly for and . For , the schemes and coincide and are isomorphic to , see Proposition 3.2. As soon as , both the complement of in , i.e. the intersection , and the complement of in are nonempty. For n = 4 the ideals of the intersection have an unexpected beautiful connection to the geometry of the inverse, quadric . The latter defines the collineation inverse to the symmetric collineation defined by .
Example 1.2.
For n = 4, the loci are summarized in . In this case, is given by ideals , where is finite of length 4 and L is a line in . The intersection is 5-dimensional and consists of ideals where and L is a line in the quadric inverse to Q, see Section 3.
For , the locus is singular (because there are obstructed Gorenstein schemes of degree n, see Theorem 1.4). For , the two loci coincide, are isomorphic to , and are smooth by Proposition 3.2 and [Citation18, Proposition 10]. We prove:
Theorem 1.3.
(Corollaries 4.3 and 4.10) For n = 4, 5 both and are smooth.
There is a map from the multigraded Hilbert scheme to the Hilbert scheme. For , it sends the scheme to the variety of sums of powers . For , we do not know whether is smooth; as noted above, the Hilbert scheme compactification seems to be not as interesting as the other two: Being apolar is not a closed property in the Hilbert scheme so has no functorial interpretation, see Remark 4.4 for a bit more discussion.
1.1 Connections to the Gorenstein locus
A saturated apolar ideal of a finite scheme is locally Gorenstein [Citation2, Proof of Proposition 2.2]. Let be the locus of which satisfy . The locus of saturated ideals is closely connected to the locus of locally Gorenstein Γ in .
Theorem 1.4.
(Proposition 2.5) The natural map is smooth and its image is . In particular, the singularity types encountered on and on coincide.
The bridge given by the theorem allows us to apply results from both sides. First, we improve and make sharp the question about when is reducible, raised in [Citation20, Theorem 1.3].
Corollary 1.5.
(Theorem 2.7) The locus is irreducible exactly when .
Second, we use the result [Citation20, Corollary 5.16] to obtain properties of the Gorenstein locus itself.
Corollary 1.6.
The Gorenstein locus of every is reduced for every .
Proof.
Reducedness of a point depends only on the underlying Z, so is reduced if and only if is reduced. By Theorem 1.4 this holds if and only if is reduced. The reducedness of the latter is proven in [Citation20, Corollaries 5.12 and 5.16]. □
Theorem 1.4 extends also to the whole . The locus of saturated ideals in is open and its closure is called . This closure is a union of components of , for it is a single component, called . This locus has recently gained much attention thanks to applications in border apolarity [Citation3, Citation17]. The theorem extends as follows.
Theorem 1.7.
(Proposition 3.8, Corollaries 3.9 and 4.3) The natural map is smooth. For , is smooth and the component is smooth along the image of this map.
The component is not smooth in all points, not even for n = 4: the saturated ideal of a second-order neighborhood of a point of yields a singular point of .
1.2 Salvaging [Citation20]
In this section we report on the state of the results in the paper [Citation20].
Our notation differs from that of [Citation20], we follow rather the notation of [Citation19]. The paper [Citation20] uses for the Variety of Polar Simplices as a subscheme of the Hilbert scheme; what we call the Variety of Sums of Powers. The paper [Citation20] uses for the image in the Hilbert scheme of the scheme .
The crucial error is found in the proof of [Citation20, Corollary 2.2] that wrongly claims that while we only have an open immersion
We therefore discuss the statements that depend on this error.
1.2.1 Introduction
The second part of [Citation20, Theorem 1.1] asserts that for , the variety is singular, rational and -dimensional. This is true (after possibly reducing) with the same proof. The first part of Theorem 1.1 claims that for , the variety is additionally smooth of Picard rank 1 and is Fano of index 2. For n = 2, 3, we have and πG is an isomorphism onto the image . In particular both are isomorphic to and the argument of [Citation20] is correct. For n = 4, 5, we could ask the same question for and and their images under πG. In both cases, by Theorem 1.3, is smooth and admits a birational morphism onto the smooth , that contracts the boundary divisor (cf. Remark 3.11), hence the Picard rank of is at least two. For we do not know whether it is smooth, but if it were, see Remark 4.4, its Picard rank would also be at least two. When replacing by the Grassmannian subscheme , however, we salvage also the first part of [Citation20, Theorem 1.1].
Salvaged Theorem 1.1. (Corollary 4.10) For , the Grassmannian subscheme is a smooth rational -dimensional Fano variety of index 2 and Picard number 1.
The theorem [Citation20, Theorem 1.2] concerns , the Grassmannian subscheme. We were able to obtain the result of this theorem for n = 4, with a correct degree, using a more nuanced machinery of excess intersections. The case n = 5 remains open. We refer to [Citation20] for the notation regarding the Gauss map; we will not use it in the present article.
Salvaged Theorem 1.2. (Proposition 4.15) The variety contains the image of the Gauss map. When n = 4 the restriction of the Plücker line bundle generates the Picard group of and the degree is 362.
The theorem [Citation20, Theorem 1.3] concerns the linear span of , and is wrong. The image of unsaturated limit ideals does not lie in the span of (loc.cit.). Whether is a linear section of the Grassmannian therefore remains an open problem. It is true for n = 3, and we give a computational proof for n = 4.
Finally, [Citation20, Proposition 1.4] remains correct with the same proof.
1.2.2 Sections 2–5
The results of these sections are local and the proofs are not effected by mistakes concerning the compactifications up to restricting to the locus of linearly normal schemes. That is, up to replacing and by and its smoothable component, the statements are correct according to our knowledge, except the final part of Corollary 2.2.
1.2.3 Section 6
The theorem [Citation20, Theorem 6.3] concerns the degree of the Grassmannian subscheme . The presence of unsaturated limit ideals in means that the claim of the theorem is wrong, instead the degree formula is a contribution in a computation of the degree of using excess intersection. We show this in the case n = 4, see Proposition 4.15 and Remark 4.16.
1.3 Legend
In Section 2 we show that the locus of saturated apolar ideals is irreducible if and only if the quadric Q has rank at most 13. In the following Section 3, we discuss the unsaturated ideals in the boundary and give a precise characterization when n = 4, 5. In Section 4 we give our results on global properties both of the multigraded Hilbert scheme compactification and of the Grassmannian compactification of . The Appendices A and B contain computer code [Citation22] in Macaulay2 [Citation11], used in our computational arguments and a result on for n = 4, for which we only give a computational proof.
1.4 Notation
We let k be a field of characteristic zero. Our computations, see Appendices A and B, are performed over . Let be polynomial rings. We view and as dual bases of dual spaces S1 and T1. Differentiation defines bilinear pairings which induces isomorphisms for any . In this sense we say that T is dual to S. We let S be the coordinate ring of , and T be the coordinate ring of the dual space . Thus we may set the projective space of 1-dimensional subspaces in T1.
For a homogeneous polynomial let denote its apolar ideal. For a subscheme , let denote its homogeneous ideal; note that this ideal is saturated. We say that an ideal is apolar to if . We say that a zero-dimensional is apolar to F if its homogeneous ideal is apolar, that is, if . We usually denote quadratic forms in T by q; it defines a quadric and a collineation: . When q is nondegenerate, then denote by the quadratic form that defines the inverse collineation, and by the corresponding quadric.
We also identify with the associated map . If is a full rank quadric, then is generated by degree two elements and , so an ideal I is apolar to Q if and only if and . A subscheme Γ as above is apolar to Q if and only if its homogeneous ideal is apolar.
2 Good points of VPS
The following result essentially appears in [Citation20], but due to its importance we provide a full proof.
Proposition 2.1.
[Citation20, Lemma 2.6] Let have length n. The following are equivalent
is (locally) Gorenstein and linearly normal,
there exists a full rank quadric Q such that Γ is apolar to Q.
Proof.
Let . Suppose (1) holds, so the dualizing A-module is isomorphic to A. Let be its generator, then the pairing given by is symmetric and perfect. Since is linearly normal, the restriction of linear forms is an isomorphism and we obtain a symmetric perfect pairing . Let be the corresponding quadric. It has full rank. Moreover, it arises from the following commutative diagram so , hence Γ is apolar to Q.
Assume (2) holds. Consider the diagram as above, where is the functional induced by . Since q has full rank, the natural map has to be injective, so an isomorphism by comparing dimensions. This shows that from the diagram is a perfect pairing, hence ωA is isomorphic to A and, hence Γ is locally Gorenstein. □
Recall that is the multigraded Hilbert scheme parameterizing ideals with Hilbert function . Let denote the locus of ideals which are saturated and apolar to Q. Recall that we have a natural map . Let denote the open locus of Γ with ; equivalently those are the subschemes which span , or, in yet other words, the subschemes with .
Proposition 2.2.
The locus is locally closed in . Under the map it maps isomorphically onto the locus of apolar subschemes in .
Proof.
Let inside denote the subset of saturated ideals, this is an open subset [Citation17, Theorem 2.6] and so comes with a natural scheme structure. The condition is closed in and so is closed in . By [Citation17, Theorem 3.9] the subset maps isomorphically to under the natural map . Restricting this isomorphism to , we get the final claim. □
We now describe the topology of .
Proposition 2.3.
The space is connected.
Proof.
This follows from the argument in [Citation20, Section 6] which is correct if one restricts from to . Essentially the same argument is given independently in [Citation13, Proposition 4.1]: any k-point Γ of degenerates to a fixed one and this degeneration is possible inside because it preserves orientation of Γ, as defined in [Citation13, 2.1]. □
Before we give more refined information about the topology, we need a technical idea of orthogonalization, which links and .
2.1 Orthogonalization
The following key technical theorem shows that any infinitesimal deformation of an apolar ideal can be “orthogonalized” to a deformation in . In essence, it says the following: suppose that is an apolar ideal and I is its deformation (in the multigraded sense) over a base with a nilpotent ε so . It may happen that I does not lie in , for example when I comes from a infinitesimal change of coordinates that moves q. The theorem says that this is essentially the only possibility: there exists a coordinate change so that is contained in the ideal . It follows from the fact that the -action on full rank quadrics is transitive in the infinitesimal neighborhood of q.
Let us introduce the setup. For a k-algebra A let . For a surjection of k-algebras , we get a surjection which in down-to-earth terms reduces the coefficients of the polynomials modulo . Let T be a polynomial ring dual to S and let be a full rank quadric, as in the setup at the beginning of Section 2. Let be its apolar ideal. We denote by also its image in .
Theorem 2.4
(orthogonalization). Let be a surjection of finite local k-algebras with kernel J satisfying . Let be graded vector subspaces. Assume that
,
is a free A-module,
the image of I2 in is .
Then there exists an invertible matrix such that and .
Proof.
The duality of S and T induces a duality of free A-modules and and, as a consequence, the free -modules and . The space perpendicular to is . Since is a free -module of rank one, the surjections dualize to a composition which sends 1 to q.
The A-modules are free, hence
Since , the right-hand side contains the natural surjection . Let us lift it to a surjection . The surjections
dualize to (1) (1)
Let denote the image of . Since the sequences for I reduce to those for modulo J, the quadric satisfies . Let us identify quadrics with symmetric matrices. For an element , the action of g on (1) yields (2) (2) which maps to the quadric . We claim that there exists a g such that and . We view the vector space T2 as an algebraic variety. Since q has full rank, the orbit map is smooth. The quadric can be interpreted as a map . Let be the constant map to the identity matrix. Since , we have a commutative diagram
From the infinitesimal lifting criterion [Citation24, Tag 02H6], there is a lift
Taking yields the desired element such that . Dualizing (2) we get that . □
For a moment, we change focus from to . By Proposition 2.2 the reader can choose one of two perspectives: either think about or about the saturated locus of .
Proposition 2.5.
The action map is smooth and its image is the Gorenstein locus in .
Proof.
We view as a locally closed subset of , using Proposition 2.2. Let be the restriction of the universal family. The points of are linearly normal by Proposition 2.1. By Cohomology and Base Change, the -module is locally free and its fiber over a point is . The claim now follows from Theorem 2.4 and the infinitesimal lifting criterion [Citation24, Tag 02H6]. □
Remark 2.6.
As explained in the introduction, the result of [Citation20, Corollary 5.16] yields interesting news about the point , the so-called G-fat point, in .
Theorem 2.7.
The scheme is irreducible, when .
Proof.
First we show that is irreducible. For this, note that the deformation theory of a zero-dimensional subscheme is independent of its projective embedding, see for example [Citation1, p.4]. The Gorenstein locus is connected, see [Citation13, Proposition 4.1] or [Citation7], and the linear normality condition on the Gorenstein locus in is open in the Gorenstein locus . So when the latter is irreducible, so is the former. But the latter is irreducible when by [Citation6, Theorem A].
Finally we prove that is irreducible if and only if is irreducible. We observe that both spaces are connected (see Proposition 2.3) and related by smooth maps see Proposition 2.5, hence one of them contains a point which lies on two irreducible components if and only if the other does. □
3 Unsaturated limit ideals
Having discussed , in the following we analyze the boundary . We will see that the boundary is nonempty exactly when and give a necessary condition for a point of to lie in . We will classify the points of the boundary for n = 4 and for n = 5. Two key inputs for the classifications are the results of [Citation17] and the orthogonalization, see Proposition 3.8. An important geometric consequence of these, which is new, is the connection with the inverse quadric , see Corollary 3.5.
We begin with a general observation.
Proposition 3.1.
Each component of the boundary is a divisor in .
Proof.
Consider the natural projective map . By Proposition 2.2, the locus maps isomorphically onto its image. Let be the closure of the image, so that . Let be the locus of Γ which are not linearly normal. Since , the locus is a divisor. In fact, on the universal scheme , the map between vector bundles of rank n drops rank on the divisor . By Proposition 2.1, the intersection lies in the complement of in X. Thus the boundary is a union of divisors of X, so also is a union of divisors. □
Now we pass to the more detailed description of the boundary.
Proposition 3.2.
Let and let be an unsaturated limit ideal. Then contains a length 4 subscheme contained in a line. In particular, for there are no unsaturated limit ideals. Moreover, under the same assumptions, the Hilbert function of is or, for n = 5, .
Proof.
Consider the Hilbert function H of . Since I is unsaturated (in degree 2), , so . From Macaulay’s Growth Theorem, see [Citation6, Section 2E and Lemma 2.9], it follows that
when , we get for all , hence a contradiction,
when n = 4, we get for all , so defines a line L with ,
when n = 5, we get or for all , so is a line or a subscheme with Hilbert polynomial m + 2; in the latter case it is either a line L with embedded point or a line L with a disjoint point. In either case has length at least . □
The following is our main tool to discern from .
Proposition 3.3.
[Citation17, Example 4.2] Let I be an unsaturated homogeneous ideal such that the Hilbert function of S/I is while the Hilbert function of is . Assume furthermore that is a line L and some, possibly embedded, points (this is automatic when , see Proposition 3.2).
If I is a limit of saturated ideals, then , where IL is the ideal of L, is contained in I.
For a linear space , we denote by its perpendicular (this notion has nothing to do with q!). Recall also that the linear maps and are nondegenerate and inverses of each other.
Lemma 3.4.
Let be linear subspaces with . Then the following are equivalent
,
,
,
,
,
.
Proof.
The equivalences follows immediately from the definitions when noticing that . □
Corollary 3.5
(a geometric condition for being in the ). In the notation of Proposition 3.3 assume that . Let be the line in and let be the linear span of . Then .
Proof.
We have . The span N is equal to and L is equal to . By the assumption of Proposition 3.3, we have , so the claim follows from Lemma 3.4. □
Corollary 3.6.
Let be an unsaturated limit ideal. Then contains a subscheme of length 4 in a line L, and the line L is contained in the inverse quadric In particular, when n = 4, .
Proof.
Let , where, of course, . Then the linear span of contains L. But then , and hence, by Corollary 3.5, , which means . When n = 4, has Hilbert function while has Hilbert function , so . □
A finite apolar scheme of minimal length is Gorenstein. This property does not necessarily hold for unsaturated limit ideals.
Example 3.7
(Non-Gorenstein unsaturated limit ideal for n = 5). Let . Consider the family over given by the ideal
It is contained in and the fiber over is saturated. The fiber over t = 0 is non-saturated and abstractly isomorphic to , which in particular is not Gorenstein.
Recall that is the union of irreducible components of and it is defined as the closure of the locus of saturated . The following theorem allows us to relate smoothness a point and the same point . This is useful both theoretically and in computations, see cases n = 4, 5 below.
Proposition 3.8.
For every -stable locally closed subscheme of , the map is smooth and has -dimensional fibers.
Proof.
Smoothness of the map follows by Theorem 2.4 and infinitesimal lifting criterion [Citation24, Tag 02H6]. For the fiber, by -equivariance it is enough to compute the dimension of the fiber over any point . The fiber consists of pairs such that . The containment is equivalent to . Now, we have , so there is an n-dimensional space of quadrics such that . The fiber is isomorphic to the set of which map q into this space. The action of on quadrics is transitive, so the dimension of the fiber is as claimed. □
Corollary 3.9
(being a limit of saturated is independent of being in ). We have as schemes. For , we have as schemes.
Proof.
By definition, we have as schemes. To prove the other inclusion, pick a point and let be the complete local ring of . Since A is complete, by applying Theorem 2.4 to finite order truncations of the map , we obtain a pair of maps and with . Pick any generic point . Since is the closure of saturated locus, the ideal is saturated. We have , so also the ideal is saturated. This ideal is apolar to Q, thus lies in . This holds for every generic point η, so ψ factors through the closure of , that is, through . This proves the other containment from the statement. For , all Gorenstein algebras are smoothable, hence the locus is contained scheme-theoretically in and so its closure is contained in . This yields (still for ) a chain of inclusions of schemes which then have to be equalities. □
Corollary 3.10.
Let Γ be a scheme of length 4 contained in a line in and let , then I is an unsaturated limit ideal in .
Proof.
I is an unsaturated limit ideal by [Citation17, Proposition 4.2] and it is apolar to Q, hence lies in by Corollary 3.9. □
3.1 Unsaturated limit ideals in the case n = 5
With reference to [Citation17, Sections 4.4.2 and 4.4.3], we will describe when n = 5.
For , by Proposition 3.2, the length five scheme contains a length four subscheme contained in a line. It follows that is a union of two families , described as follows.
A general point of is an ideal I such that , where p is a point and has length four and is contained in a line. For general such and p, the Hilbert function of is , so ; the ideal I is determined by as in the case of n = 4. The family is irreducible and has dimension 6 + 4 + 4.
For the family , a general point is an ideal I such that is a length five scheme on a line. The ideal I is obtained from as the intersection where c is a cubic form such that . We do not know whether the family is irreducible or what is its dimension.
Let us now consider the saturable elements of both families. It follows from Corollaries 3.5 and 3.6 and from [Citation17, Section 4.4.2] that consists exactly of ideals I such that the line is contained in and p is contained in the plane polar to L, here p is such that counting multiplicities. The set of unsaturated limit ideals in has dimension 3 + 4 + 2. In the case , again by Corollary 3.6 the line must be contained in . Moreover by [Citation17, Section 4.4.3], the cube c must be of the form , where is a linear form on the line and μ is a linear form on the polar plane. The set of unsaturated limit ideals in has dimension 3 + 5 + 1.
Remark 3.11.
When n = 4, 5 and I is an unsaturated limit ideal, then the map is a forgetful map with positive dimensional fibers. The boundary is a divisor by Proposition 3.1, and has dimension 5 and 9 when n = 4 and n = 5 respectively. The ideal I2 vanishes on the line that contains a subscheme of length four (or five) of In cases n = 4 and n = 5 with , the ideal I2 depend only on a line, respectively a line and a point, so the family of ideals I2 in the image of πG has dimension one and five, respectively. In case n = 5 and , the image , depend on the line and a special cubic form, so the family of ideals I2 in the image of πG has dimension at most 6.
4 The schemes and for
As above, we let . It follows from Proposition 3.2 that for , the schemes and are both equal to and hence smooth. We show they are both smooth also for n = 4, 5. For , already is singular.
The idea of the proof is to prove that is smooth in the unsaturated limit ideals of , then use Proposition 3.8 and Corollary 3.9 for to deduce the same for itself and finally prove the smoothness for their projections in the Grassmannian. We begin by showing that is smooth at the unsaturated limit ideals of .
Proposition 4.1.
Let n = 4, 5 and consider a homogeneous ideal such that . Then is a smooth point on .
Proof.
We will make an upper bound on the dimension of the tangent space to at . Unfortunately, this space has no functorial interpretation, hence we begin with the tangent space . The short exact sequence yields
We have , while is one-dimensional, so and hence
The short exact yields
Using Proposition 3.2 and [Citation17, Example 4.1], we obtain that
Since , by [Citation17, Corollary 3.16] we obtain that the point is smooth. The saturated locus of this Hilbert scheme maps isomorphically to its image in , see [Citation17, Proposition 3.9], hence has dimension where comes from a choice of n – 4 points, a choice of a line and 4 from a choice of four points on it. In total we obtain a bound (3) (3)
We now pass to . The inclusion corresponds to the tangent map for the embedding of , where is the natural map.
By Proposition 3.3 we see that the intersection consists of I such that where L is the line in . This is true even scheme-theoretically and shows that the intersection above is the Grassmannian , which has dimension . Arguing as above, the estimate (3) yields
The variety is birational to the smoothable component of , hence has dimension . This proves that is smooth for n = 4, 5. □
Proposition 4.2.
Let n = 5 and consider a homogeneous ideal such that . Then is a smooth point on .
Proof.
The strategy of the proof is the same as in Proposition 4.1. As in that proposition, we argue that the tangent space at is 11-dimensional because the locus of saturated ideals with this Hilbert function is parameterized by 6-dimensional choice of line and 5-dimensional choice of five points. It follows that is an extension of the at most 12-dimensional space by . The latter space is the tangent space to the fiber of . This fiber is smooth and rational of dimension 14, its parameterization is given in [Citation17, Section 4.4.3]. Also by [Citation17, Section 4.4.3] the intersection is smooth and rational of dimension 8. It follows that inside the 26-dimensional space the intersection of with is 8-dimensional. The latter space is 14-dimensional, hence the dimension of is at most , which coincides with the dimension of , so equality holds and is a smooth point on . □
Corollary 4.3.
For n = 4, 5 the variety is smooth.
Proof.
Being smooth is preserved by smooth maps. It follows that the good part is smooth by Proposition 2.5. The variety is smooth at unsaturated limit ideals by Proposition 3.8 with Corollary 3.9 and Propositions 4.1–4.2. □
Some natural simplified versions of Proposition 4.1 are false: for example can be singular at the points as in the statement and can be singular away from the locus in the Proposition; both pathologies occur for n = 4. It would be interesting to know whether Proposition 4.1 holds for higher n.
Remark 4.4.
We do not know whether is smooth even for n = 4, 5. The map restricts to a map which is an isomorphism on . For n = 4, this map is also bijective on points, because for an unsaturated limit ideal I we have . This shows that for n = 4, the variety is the normalization of .
Now we prove that also the Grassmannian compactification is smooth for n = 4, 5. One cannot hope to have Proposition 3.8 for and because the latter lacks structure. But there is an analogue of the Hilbert scheme in this setup. We define the syzygetic locus as the closed subscheme given by the determinantal equations which on a point boil down to , where is the ideal generated by . The tangent space to at a point where is given by
The name compactification is justified by the following.
Proposition 4.5.
The map , given by is an open immersion. In particular, the map is an isomorphism onto .
Proof.
A point of corresponds to a saturated ideal I such that the quotient by I has Hilbert function H. It follows that the regularity of the quotient is one and the regularity of I is two. By [Citation8, Proposition 3.1] the complete local rings of and are isomorphic. It follows from [Citation24, Tag 039M] that πG is étale at . From the regularity it also follows that I is generated by I2, so πG is injective on points. It follows from [Citation24, Tag 025G] that this map is an open immersion. The result on follows by intersecting with the locus apolar to Q. □
Proposition 4.6.
Let n = 4, let be an unsaturated limit ideal and let . Then is a smooth point of the image of in .
Proof.
Every unsaturated limit ideal I defines a scheme of length four on a line on a quadric by Proposition 3.3 and Corollary 3.6. The ideal I itself is determined by Γ as . Using a torus action not changing the quadric, we may degenerate further so that Γ is supported only at a single point. There is then, up to coordinate change, only one unsaturated limit ideal. If , then we may take it as .
Using the package VersalDeformations [Citation16], see Appendix A, we check that near the point the scheme is reduced and it has two irreducible components passing through ; the dimensions are 12 and 10, respectively. Moreover, both these components are smooth at . The image of under is an integral 12-dimensional variety, hence it has to coincide with the larger component: . In particular, is a smooth point on it. □
Proposition 4.7.
Let n = 5 and let be an unsaturated limit ideal. Then its image is a smooth point of the image of in .
Proof.
The argument is analogous to Proposition 4.6, although reducing to finitely many possible limits requires much more work (we need finitely many of them to compute the tangent space dimensions). By Corollary 3.6 and Proposition 3.2, an unsaturated limit ideal I corresponds to a scheme of length five with a subscheme of length four on a line and a possibly embedded fifth point. Up to coordinate change, we may assume that and the line L is . We have . Consider two cases for :
the scheme Γ is contained in L,
the scheme Γ is not contained in L.
We begin with the case (2). By Corollary 3.5 the fifth point lies on the plane . Consider a coordinate change
It preserves and L and maps a point to . Using such an action, we may assume that the fifth point is not . Consider the torus action by
When , the fifth point converges to a point of L. The set of singular points of the image is closed, so we may reduce to the case of the fifth point on a line; either the point is embedded or we reduce to case (1). We consider the case (1) below, so assume that the point is embedded.
For every , the coordinate change maps , L and Π to themselves. Using such an action for general λ, we may assume that Γ is not supported at . Consider the torus action by
Again it preserves , L and Π. The limit with of any point
Hence, we reduce to the case of Γ supported at having an embedded point, so that and Γ spans Π. Here I is determined by . We check its smoothness in Section A. (This is the case from Example 3.7.)
The case (1) is similar, but slightly more algebraic, as in this case the coordinate changes need to take into account the structure of I rather than . By [Citation17, Proposition 4.4(3)], the space is spanned by and , where . Making the coordinate change as above, we may assume that is not supported at and that is not proportional to y2. Then a torus degeneration as above reduces us to the case and Γ supported at , thus I is determined as . Again, we check its smoothness in Section A. □
Remark 4.8.
Actually, according to the calculation in Section A, near the variety is equal to for n = 4, 5.
Proposition 4.9.
Let be the natural map. The map is smooth, so in particular is smooth for n = 4, 5.
Proof.
We have and the map πG sends I to I2, so the smoothness of the map follows by Theorem 2.4. For , the locus is smooth so is smooth by Proposition 4.5. Moreover, is smooth at the unsaturated limit ideals by Propositions 4.6–4.7, hence is smooth by the smoothness of the map. □
When n = 1 and n = 2 the variety is , respectively a 3-fold general linear section of , cf. [Citation18]. In particular, in both cases, the Grassmannian subscheme is a Fano variety of index 2. The same holds for n = 4, 5.
Corollary 4.10.
is a Fano variety of index 2, when n = 4, 5.
Proof.
We follow the argument of [Citation20, Theorem 6.1]. Consider the set where is a hyperplane. In [Citation20, Lemma 4.9] the set Hh is defined for schemes , but the definition extends naturally. Similarly, the proof of the lemma extends to show that Hh is the restriction of a Plücker divisor on . Notice that the divisors Hh contains the image of all unsaturated limit ideals when n = 4, 5, since in these cases contains a line that necessarily intersects the hyperplane h. By [Citation20, Proposition 5.11], the complement of Hh is isomorphic to an affine space, while Hh is very ample by Proposition 4.9. The image of the set of all unsaturated limit ideals is not a divisor, so the argument of the proof of [Citation20, Theorem 6.1] applies: Since the complement of Hh is isomorphic to an affine space, the Picard group of is as soon as the restriction Hh of the special Plücker divisor is irreducible. So assume .
Any apolar scheme Γ lies in the complement of some h, so the subset BL is the base locus of the special divisors Hh. But BL is not a divisor, so H1 and H2 must both move with base locus contained in BL.
Now for every line , so only one of the two components can have positive intersection with l. In particular, one of the components, say H2 has intersection and contains every line l that it intersects. By, [Citation20, Lemma 6.2], any two smooth apolar schemes are connected by a sequence of lines, so H2 would contain all of . Therefore Hh is irreducible.
The computation of the Fano-index in [Citation20, Theorem 6.1] when n = 4, 5, applies similarly to , see also Remark 4.14. □
4.1 The scheme for n = 4
In the case n = 4, we give a more precise description of the Grassmannian compactification, starting with the image of the family of unsaturated limit ideals in by the map . By Corollaries 3.6 and 3.10 each unsaturated limit ideal is contained in the ideal of a line in , and the degree two part of such an ideal depends only on the line: it is the space of those quadrics in the ideal of the line that lie in . Therefore is the disjoint union of two rational curves C1 and C2, one for each pencil of lines in .
Lemma 4.11.
Let n = 4. The two curves C1 and C2 are rational normal curves of degree 6 in .
Proof.
For each line , we let IL be the intersection of its ideal with the space of quadrics in the apolar ideal . Thus IL is 6-dimensional in , and its orthogonal is a 3-dimensional subspace . We proceed by arguing in . Projectively the subspace is the span of the conic under the veronese embedding . The rational -scroll X formed by these spans as L moves in one pencil of lines in is a rational normal scroll: For each triple of lines in one pencil in , the corresponding conics on X is not contained in any hyperplane, so the degree of X is at least 6, the minimal degree of a 3-fold in . On the other hand is linearly normal, hence also X, so X has degree 6. The Grassmannian embedding of a rational normal scroll of planes X in the Grassmannian of planes is a rational normal curve and has the same degree as X. The corresponding embedding in of the pencil of spaces dual to is projectively equivalent and therefore also has degree 6. □
Lemma 4.12.
For each rational normal curve in , the tangent bundle of restricted to Ci is and the normal bundle is .
Proof.
Fix coordinates so that and C is the rational normal curve containing the point . Explicitly, we have (4) (4)
Consider the -action on where a matrix acts by
The quadric Q is invariant under this action and so acts on and on C. Actually, the action on C is transitive; the point p is sent to ; thus the induced map is the usual one. The action above induces -linearizations on tangent bundle and normal bundle from the statement. The action of the standard torus of decomposes the degree two part of (4) into weight spaces of weights 0 and –2, respectively. The tangent space is equal to which is 9-dimensional. Let , it is a weight space of weight two. Any linear map ρ2 from the weight zero space in to W has weight two. To see that ρ2 lifts to a tangent vector ρ it is enough to check that it induces a linear map ρ3 in degree three that is zero on the linear syzygies of , since the linear syzygies generate all syzygies of . Since ρ2 has weight two, it induces, in degree three, a map ρ3 of the weight one space to a weight three space of cubics.
But any linear syzygy of has weight ±1 so extending ρ2 to the zero map on the weight space , means that ρ3 is zero on the syzygies. We conclude that ρ2 lifts to a tangent vector ρ.
The space of maps ρ2 is 9-dimensional, hence equal to , the tangent space to at p. It follows that the weights of the action of the standard torus of on are all equal to 2. The fiber of the tangent bundle from the statement is a subrepresentation of for this torus, so it also has all weights equal to 2. The fiber of the normal bundle is its quotient representation, so the same holds here. By the classification of vector bundles on , we find that the tangent bundle is , while the normal bundle is . □
Remark 4.13.
The degree and normal bundle of the curves C1 and C2 is also found in the Macaulay2 computation [Citation22].
Remark 4.14.
The normal bundle of the two curves may be used to compute the index of the former variety as a Fano variety. The Picard group is generated by H, so the canonical divisor is for some integer a. Since the normal bundle in of each of the two rational curves Ci in BL has degree 10 while the degree of each curve is 6, the adjunction formula implies , so a = 2 and has Fano index 2.
Proposition 4.15.
Let n = 4. The subscheme inside is an arithmetically Cohen-Macaulay variety and a smooth linear section of degree 362. The image of the set of unsaturated limit ideals form a union of two disjoint rational normal curves each of degree 6.
A general codimension 6 linear space containing intersects the locus in 310 points.
Proof.
It remains to prove the linear section and Cohen-Macaulay properties and the degree computations. This involves the Macaulay2 computation [Citation22], see Theorem B.1 in Appendix B. It proves that the linear section , where is the linear span of , is a 6-dimensional arithmetically Cohen-Macaulay variety of degree 362.
Now, we argue that the degree of X coincides with the degree of , so that is the only component of maximal dimension in X. But, since X is arithmetically Cohen-Macaulay, every component is maximal, so X and coincide if their degrees do.
The degree argument uses excess intersection. Let L be a general linear space of codimension 6 that contain the two curves . The excess intersection of the two curves in the intersection is computed by the formula [Citation9, Prop 9.1.1.(2)]: where NL is the normal bundle of L restricted to Ci and H is the class of a hyperplane. So the excess intersection along has degree . By the below remark, the linear space L intersects in 310 points. So the intersection of with a general linear space of codimension 6 is , which coincides with the computation of the degree of X. □
Remark 4.16.
In [Citation20] special Plücker hyperplanes are considered in a computation of . They already appeared in the proof of Corollary 4.10. For a hyperplane , the set is a Plücker hyperplane section of . Since contains a line for every unsaturated limit ideal I, every divisor Hh contains the image of all unsaturated limit ideals. According to [Citation20, Theorem 6.3], when n = 4, the intersection of six general hyperplanes Hh contains 310 ideals of polar simplices, i.e. points on . That theorem claims that this is also the degree of , based on the false assumption that this variety is closed. The locus which is contained in every hyperplane Hh. The intersection of six general hyperplane sections Hh therefore contains in addition to the 310 points on . So to compute the degree of using the formula in [Citation20, Theorem 6.3], becomes the excess intersection problem of finding the contribution of to the degree of the intersection.
Acknowledgments
We thank Tomasz Mańdziuk and Emanuele Ventura for pointing out an issue in an earlier version of the paper, and anonymous referees for helpful comments on presentation.
Additional information
Funding
References
- Artin, M. (1976). Deformations of Singularities. Notes by C.S. Seshadri and Allen Tannenbaum. Bombay, India: Tata Institute of Fundamental Research.
- Buczyńska, W., Buczyński, J. (2014). Secant varieties to high degree veronese reembeddings, catalecticant matrices and smoothable gorenstein schemes. Duke Math. J. 23(1): 63–69.
- Buczyńska, W., Buczyński, J. (2021). Apolarity, border rank, and multigraded Hilbert scheme. Duke Math. J. 170(16): 3659–3702. 10.1215/00127094-2021-0048
- Bolognesi, M., Massarenti, A. (2018). Varieties of sums of powers and moduli spaces of (1, 7)-polarized abelian surfaces. J. Geom. Phys. 125: 23–32. 10.1016/j.geomphys.2017.12.004
- Carlini, E., Catalisano, M. V., Oneto, A. (2017). Waring loci and the Strassen conjecture. Adv. Math. 314: 630–662. 10.1016/j.aim.2017.05.008
- Casnati, G., Jelisiejew, J., Notari, R. (2015). Irreducibility of the Gorenstein loci of Hilbert schemes via ray families. Algebra Number Theory 9(7): 1525–1570. 10.2140/ant.2015.9.1525
- Casnati, G., Notari, R. (2009). On the Gorenstein locus of some punctual Hilbert schemes. J. Pure Appl. Algebra 213(11): 2055–2074. 10.1016/j.jpaa.2009.03.002
- Erman, D. (2012). Murphy’s law for Hilbert function strata in the Hilbert scheme of points. Math. Res. Lett. 19(6): 1277–1281. 10.4310/MRL.2012.v19.n6.a8
- Fulton, W. (1998). Intersection Theory, volume 2 of Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics [Results in Mathematics and Related Areas. 3rd Series. A Series of Modern Surveys in Mathematics], 2nd ed. Berlin: Springer-Verlag.
- Gallet, M., Ranestad, K., Villamizar, N. (2018). Varieties of apolar subschemes of toric surfaces. Ark. Mat. 56(1): 73–99. 10.4310/ARKIV.2018.v56.n1.a6
- Grayson, D., Stillman, M. Macaulay2, a software system for research in algebraic geometry, ver. 1.21. Available at: https://macaulay2.com.
- Hauser, H. (1983). An algorithm of construction of the semiuniversal deformation of an isolated singularity. In: Singularities, Part 1 (Arcata, Calif., 1981), volume 40 of Proceedings of Symposia in Pure Mathematics. Providence, RI: American Mathematical Society, pp. 567–573.
- Hoyois, M., Jelisiejew, J., Nardin, D., Yakerson, M. (2022). Hermitian k-theory via oriented gorenstein algebras. J. Reine Angew. Math. 793: 105–142. 10.1515/crelle-2022-0063
- Huang, H., Michałek, M., Ventura, E. (2020). Vanishing Hessian, wild forms and their border VSP. Math. Ann. 378(3–4): 1505–1532. 10.1007/s00208-020-02080-8
- Haiman, M., Sturmfels, B. (2004). Multigraded Hilbert schemes. J. Algebraic Geom. 13(4): 725–769. 10.1090/S1056-3911-04-00373-X
- Ilten, N. O. (2012). Versal deformations and local Hilbert schemes. J. Softw. Algebra Geom. 4: 12–16. 10.2140/jsag.2012.4.12
- Jelisiejew, J., Mańdziuk, T. (2022). Limits of saturated ideals. arXiv:2210.13579.
- Mukai, S. (1992). Fano 3-folds. London Math. Soc., Lect. Notes Ser. 179: 255–263.
- Ranestad, K., Schreyer, F.-O. (2000). Varieties of sums of powers. J. Reine Angew. Math. 525: 147–181. 10.1515/crll.2000.064
- Ranestad, K., Schreyer, F.-O. (2013). The variety of polar simplices. Doc. Math. 18: 469–505. 10.4171/dm/406
- Ranestad, K., Voisin, C. (2017). Variety of power sums and divisors in the moduli space of cubic fourfolds. Doc. Math. 22: 455–504. 10.4171/dm/571
- Schreyer, F.-O. (2023). vsp4, a Macaulay2 package for the computation of the variety of polar simplices of a quadric in P 3. Available at: https://www.math.uni-sb.de/ag/schreyer/index.php/computeralgebra.
- Skjelnes, R. M., Ståhl, G. S. (2022). Explicit projective embeddings of standard opens of the Hilbert scheme of points. J. Algebra 590: 254–276. 10.1016/j.jalgebra.2021.10.007
- The Stacks project authors. (2023). The stacks project. Available at: https://stacks.math.columbia.edu.
Appendix A:
Computer code
In this section we exhibit the computer code used to prove smoothness of specific points in for n = 4, 5. The computations are performed over . See package [Citation16] for details.
loadPackage("VersalDeformations", Reload= >true);
computeComponents = I -> (
– check that Syz and Hilb agree
assert(hilbertFunction(3, I) == dim ring I);
J = ideal select(flatten entries mingens I, x->sum degree x< =2);
– the ideal I_2 + S_{\geq 4}
Jtr = ideal mingens(J + (ideal gens ring J)^4);
tg = normalMatrix(0, Jtr);
reducedtg = CT^1(0, Jtr);
ob = CT^2(0, Jtr);
(F,R,G,C) = versalDeformation(mingens Jtr,reducedtg,ob);
IG = ideal mingens ideal sum G;
pd = primaryDecomposition IG;
assert(#pd == 2);
– actually, those are linear spaces
assert(dim ideal singularLocus pd_0 == -1);
assert(dim ideal singularLocus pd_1 == -1);
use ring I; – reset the default ring
return (rank source tg - codim pd_0,
rank source tg - codim pd_1);
);
The code for n = 4 is as follows.
QQ[x_0. x_3];
Q = x_0*x_2 + x_1*x_3;
Qperp = inverseSystem(Q);
I = intersect(Qperp, ideal(x_1^4, x_2, x_3)); –case n = 4
computeComponents(I) == (12,10) – true
The code for n = 5 is very similar.
QQ[z_0. z_4];
Qz = z_0*z_2 + z_1*z_3 + z_4^2;
Qzperp = inverseSystem(Qz);
I = ideal(z_3*z_4, z_2*z_4, z_0*z_4, z_3^2, z_2*z_3, z_1*z_3,
z_0*z_3-z_4^2, z_2^2, z_1*z_2 - z_4^2, z_0*z_2, z_0^4);
computeComponents(I) == (20,20) – true
I = intersect(ideal(z_0, z_3, z_4, z_1^5),
inverseSystem(z_4*z_2^2), inverseSystem(Qz));
computeComponents(I) == (20,20) – true
The code for n = 4 is as follows.
QQ[x_1. x_4];
Q = x_1*x_3 + x_2*x_4;
Qperp = inverseSystem(Q);
I = intersect(Qperp, ideal(x_2^4, x_3, x_4)); –case n = 4
computeComponents(I) == (12,10) – true
The code for n = 5 is very similar.
QQ[x_1. x_5];
Qz = x_1*x_3 + x_2*x_4 + x_5^2;
Qzperp = inverseSystem(Qz);
I = ideal(x_4*x_5, x_2*x_5, x_1*x_5, x_4^2, x_1*x_4,x_3*x_4,
2*x_2*x_4-x_5^2, x_1^2,x_1*x_2, 2*x_1*x_3 - x_5^2, x_2^4);
computeComponents(I) == (20,20) – true
I = intersect(ideal(x_1, x_4, x_5, x_2^5),
inverseSystem(x_5*x_3^2), inverseSystem(Qz));
computeComponents(I) == (20,20) – true
Appendix B:
Remarks on the package vsp4.m2
In our package we prove computationally the following theorem. The computations are performed over .
Theorem B.1.
Consider with . The Grassmannian model is smooth. Then is the intersection of with the linear span of inside . It is an arithmetically Cohen-Macaulay variety of degree 362 with h-vector .
The proof is computational. The first step consists in computing the unfolding [Citation12] of the ideal and the computation of the flatness relations via a Gröbner basis computation. As it turns out the base space of the family consists of two components of dimension 8 and 6, respectively. The 8-dimensional family consists of lines in together with 4 points. The general element of the 6-dimensional family consists of 4 distinct points not on a line, they vary as sets of points apolar to q. The intersection of these families consists of 4 points on a line on the inverse quadric . So the intersection has two components corresponding to the two family of lines on . They are of dimension 5.
In the next step we compute the image of these families in the Grassmannian . It is computational accessible to compute the linear equations in . To see that image of the 8-dimensional family coincides with the 3-uple embedding of is straightforward.
Analyzing the image of the 6-dimensional family is computationally difficult. The linear span is a and the 1050 Plücker quadrics restrict to 380 independent quadrics on this . Moreover, the quadrics define a saturate ideal J of dimension 7 and degree 362. So J corresponds to a 6-dimensional scheme X of degree 362. The 6-th difference function of the Hilbert function of X takes values .
This lead us to conjecture that X is an arithmetically Gorenstein subscheme of . Note that . Thus if X is arithmetically Gorenstein, its homogeneous ideal in is generated by these 380 quadrics. However we could not check directly that the homogeneous ideal of the image of the 6-dimensional family is generated by quadrics.
Instead we establish that X is arithmetically Cohen-Macaulay by testing that for a carefully chosen sparse sequence of linear forms the ideals are still saturated for . Since (which we check computationally) the ideal defines the homogeneous ideal of a zero dimensional subscheme of a of degree 362. Thus there exists a final linear form such that are a regular sequence for the homogeneous coordinate ring of X.
This proves that X is arithmetically Cohen-Macaulay and therefore unmixed.
In Proposition 4.15, we apply excess intersection theory to show that the two varieties X and have the same degree, hence must coincide.
By the Macaulay2 computation, the image of the intersection of the two families consists of two rational normal curves C1, C2 of degree 6, as above. Furthermore it checks that X is smooth along the normal curves and that their normal bundles are isomorphic to , in accordance with Lemmas 4.11 and 4.12. Finally, by inspecting the 6 × 9 matrix which defines the map of the 6-dimensional family into the Grassmannian we see that is smooth outside .