On partial polynomial interpolation
Maria Chiara Brambilla, Giorgio Ottaviani

TL;DR
This paper generalizes the Alexander-Hirschowitz theorem to include arbitrary zero-dimensional schemes in general unions of double points, providing a comprehensive understanding of polynomial interpolation with derivatives.
Contribution
It extends the classical theorem to broader schemes and characterizes the dimension of polynomial spaces with derivative conditions, including detailed exceptions.
Findings
The affine space of polynomials has the expected dimension in most cases.
Five exceptional cases occur when the degree is not 2.
All exceptions for degree 2 are fully described.
Abstract
The Alexander-Hirschowitz theorem says that a general collection of double points in imposes independent conditions on homogeneous polynomials of degree with a well known list of exceptions. We generalize this theorem to arbitrary zero-dimensional schemes contained in a general union of double points. We work in the polynomial interpolation setting. In this framework our main result says that the affine space of polynomials of degree in variables, with assigned values of any number of general linear combinations of first partial derivatives, has the expected dimension if with only five exceptional cases. If the exceptional cases are fully described.
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsPolynomial and algebraic computation · Algebraic Geometry and Number Theory · Tensor decomposition and applications
On partial polynomial interpolation
Maria Chiara Brambilla and Giorgio Ottaviani
Abstract
The Alexander-Hirschowitz theorem says that a general collection of double points in imposes independent conditions on homogeneous polynomials of degree with a well known list of exceptions. We generalize this theorem to arbitrary zero-dimensional schemes contained in a general union of double points. We work in the polynomial interpolation setting. In this framework our main result says that the affine space of polynomials of degree in variables, with assigned values of any number of general linear combinations of first partial derivatives, has the expected dimension if with only five exceptional cases. If the exceptional cases are fully described.
AMS Subject Classification: 14C20; 15A72; 65D05; 32E30
Both authors are partially supported by Italian MUR and are members of GNSAGA-INDAM.
1 Introduction
Let be the vector space of polynomials of degree in variables over an infinite field . Note that . Let be general points and assume that over each of these points a general affine proper subspace of dimension is given. Assume that . Let be the graph of and be its tangent space at the point . Note that for any . Consider the conditions
[TABLE]
When , the assumption (1) means that the value of at is assigned. When , (1) means that the value of at and the values of all first partial derivatives of at are assigned. In the intermediate cases, (1) means that the value of at and the values of some linear combinations of first partial derivatives of at are assigned.
Consider now the affine space
[TABLE]
The polynomials in this space solve a partial polynomial interpolation problem. The conditions in (1) correspond to affine linear conditions on . Our main result describes the codimension of the above affine space. Since the description is different for and , we divide the result in two parts.
Theorem 1.1
Let and . For a general choice of points and subspaces , the affine space has codimension in equal to
[TABLE]
with the following list of exceptions
[TABLE]
In particular when there is a unique polynomial in , with the above exceptions a), b’), c), d). In the exceptional cases the space is empty.
The “general choice” assumption means that the points can be taken in a Zariski open set (i.e. outside the zero locus of a polynomial) and for each of these points the space can be taken again in a Zariski open set. On the real numbers this assumption means that the choices can be done outside a set of measure zero. Our result is not constructive but it ensures that in the case the linear system computing the interpolating polynomial with general data has a unique solution. Hence any algorithm solving linear systems can be successfully applied. Actually our proof shows that Theorem 1.1 holds on any infinite field, with the possible exception of finitely many values of (see the appendix). For finite fields the genericity assumption is meaningless.
The case in which for all was proved by Alexander and Hirschowitz in [AH1, AH2], see [BO] for a survey. The most notable exception is the case of seven points with seven tangent spaces for cubic polynomials in four variables, as in c). This example was known to classical algebraic geometers and it was rediscovered in the setting of numerical analysis in [Rei]. The case of curvilinear schemes was proved as a consequence of a more general result by [CG] on and by [CM] in general.
The case follows from elementary linear algebra. The case is easy and well known: in this case the statement of Theorem 1.1 is true with the only requirement that the points are distinct and the spaces are not vertical, that is their projections on satisfy .
Assume now . We set for . For any we denote
[TABLE]
Theorem 1.2
Let be an infinite field. For a general choice of points and subspaces , the affine space has codimension in equal to
[TABLE]
if and only if one of the following conditions takes place:
either for all ; 2. 2.
or .
In particular when there is a unique polynomial in if and only if, for any , we have
[TABLE]
The first nontrivial example which explains Theorem 1.2 is the following. Consider and . Then the affine space is given by quadratic polynomials with assigned tangent spaces , at two points , . This space is not empty if and only if the intersection space is not empty and its projection on contains the midpoint of , which is a codimension one condition. In order to prove this fact restrict to the line through and and use a well known property of the tangent lines to the parabola. In this case \delta_{n,n}(i)=\left\{\begin{array}[]{cl}0&i\neq 1\\ 1&i=1\end{array}\right. and the two conditions of Theorem 1.2 are not satisfied. In Section 3 we will explain these two conditions in graphical terms.
Let be the projection of on . For we consider the ideal
[TABLE]
Notice that we have and the ring corresponds to a zero-dimensional scheme of length , supported at and contained in the double point . When is not empty, its associated vector space (that is its translate containing the origin) consists of the hypersurfaces of degree through . Moreover, when this vector space has the expected dimension, it follows that has the expected dimension too.
The space can be embedded in the projective space . Since the choice of points is general, we can always avoid the “hyperplane at infinity”. In order to prove the above two theorems, we will reformulate them in the projective language of hypersurfaces of degree through zero-dimensional schemes. More precisely we refer to Theorem 3.2 for , Theorem 4.1 for and Theorem 5.6 for . This reformulation is convenient mostly to rely on the wide existing literature on the subject. In this setting Alexander and Hirschowitz proved that a general collection of double points imposes independent conditions on the hypersurfaces of degree (with the known exceptions) and our result generalizes to a general zero-dimensional scheme contained in a union of double points. It is possible to degenerate such a scheme to a union of double points only in few cases, in such cases of course our result is trivial from [AH1].
Our proof of Theorem 5.6, and hence of Theorem 1.1, is by induction on and . Since it is enough to find a particular zero-dimensional scheme which imposes independent condition on hypersurfaces of degree , we specialize some of the points on a hyperplane, following a technique which goes back to Terracini. We need a generalization of the Horace method, like in [AH1], that we develop in the proof of Theorem 5.6. The case of cubics, which is the starting point of the induction, is proved by generalizing the approach of [BO], where we restricted to a codimension three linear subspace. This case is the crucial step which allows to prove the Theorem 1.1. Section is devoted to this case, which requires a lot of effort and technical details, in the setting of discrete mathematics. Compared with the quick proof we gave in [BO], here we are forced to divide the proof in several cases and subcases. While the induction argument works quite smoothly for , it is painful to cover many of the initial cases. In the case we need the help of a computer, by a Montecarlo technique explained in the appendix.
A further remark is necessary. In [AH1, BO] the result about the independence of double points was shown to be equivalent, through Terracini lemma, to a statement about the dimension of higher secant varieties of the Veronese varieties, which in turn is related to the Waring problem for polynomials. Here the assumption that is algebraically closed of zero characteristic is necessary to translate safely the results, see also Theorem and Remark in [IK]. For example, on the real numbers, the closure in the euclidean topology of the locus of secants to the twisted cubic is a semi-algebraic set, corresponding to the cubic polynomials which have no three distinct real roots, which is defined by the condition that the discriminant is nonpositive. Indeed a real cubic polynomial can be expressed as the sum of two cubes of linear polynomials (Waring problem) if and only if it has two distinct complex conjugate roots or a root of multiplicity three.
2 Preliminaries
Let be a scheme contained in a collection of double points of . We say that the type of is if contains exactly subschemes of a double point of length , for . For example the type of double points is . The degree of is . A scheme of type corresponds to a collection of linear subspaces with with a marked point on each . Algebraic families of such schemes can be defined over any field with the Zariski topology.
We recall now some notation and results from [BO].
Given a zero-dimensional subscheme , the corresponding ideal sheaf and a linear system on , the Hilbert function is defined as follows:
[TABLE]
If , we say that is -independent, and in the case , we say -independent.
A zero-dimensional scheme is called curvilinear if it is contained in a non singular curve. Notice that a curvilinear scheme contained in a double point has length or .
Lemma 2.1** (Curvilinear Lemma [Ch, BO])**
Let be a zero-dimensional scheme of finite length contained in a union of double points of and a linear system on . Then is -independent if and only if every curvilinear subscheme of is -independent.
Let us denote and . The expected dimension of the vector space is
For any scheme and any hyperplane , the residual of with respect to is denoted by and it is defined by the ideal sheaf . We have, for any , the well known Castelnuovo sequence
[TABLE]
Remark 2.2
If are zero-dimensional schemes, then
- •
if is -independent, then so is ,
- •
if , then .
It follows that if any zero-dimensional scheme with is -independent, then any scheme contained in imposes independent conditions on hypersurfaces of degree in .**
Remark 2.3
Fix and . Assume that if a scheme with degree does not impose independent conditions on hypersurfaces of degree in , then it is of type for some given . It follows that any subscheme of is -independent. Indeed any proper subscheme of is also a subscheme of a scheme with degree and of type , for some and since is -independent, so is . Moreover any scheme containing impose independent conditions on hypersurfaces of degree if it contains a scheme with degree and of type for some . Indeed since imposes independent conditions on hypersurfaces of degree , also does. **
3 Quadratic polynomials
Assume that is a scheme of type . Let us fix an order on the irreducible components of (where ) such that
[TABLE]
and for any let us denote by the length of and by the point where is supported. Set for . For any let us denote
[TABLE]
Remark 3.1
Notice that if is positive for some , then the scheme does not impose independent conditions on quadrics. Indeed let be a basis of and be the symmetric matrix defining a quadric in passing through the scheme . We may assume that for all . Then, the condition implies that the elements in the last columns and rows of the matrix outlined below are all equal to [math]. Hence the scheme has degree , but imposes only conditions on quadrics. In fact the quadrics containing are exactly all the quadrics singular along the linear space spanned by . **
The following result describes the schemes which impose independent conditions on quadrics.
Theorem 3.2
A general zero-dimensional scheme contained in a union of double points of type imposes independent conditions on quadrics if and only if one of the following conditions takes place:
either for all ; 2. 2.
or .
Proof. First we prove that if does impose independent conditions on quadrics, then either condition 1 or 2 hold. Assume that both conditions are false and let us prove that has not the expected dimension . Indeed choose such that and and consider the family of quadratic cones with vertex containing the linear space spanned by . Of course we have
[TABLE]
Now we compute
[TABLE]
By assumption and
[TABLE]
Hence the dimension of is higher than the expected dimension.
Now we want to prove that if either condition 1 or condition 2 hold, then imposes independent conditions on quadrics. We work by induction on . If it is easy to check directly our claim.
Consider a scheme in which satisfies condition 1 and fix a hyperplane . We specialize all the components of on in such a way that the residual of each of the components is (if they are not zero) and the residual of the remaining components is zero. Then we get the Castelnuovo sequence
[TABLE]
where is the residual given by at most simple points and is the trace in . Hence we conclude by induction once we have proved that the trace satisfies condition 1 or 2.
In fact we prove now that if does not satisfy condition 1, then it satisfies 2. Assume that for in we have for all , while for in we have for some . Choose the index such that is maximum. This is possible only if there is some index such that . Since and we obtain
[TABLE]
from which it follows that
[TABLE]
Now in order to prove that satisfies 2 we need to show that
[TABLE]
Notice that
[TABLE]
hence if we prove the following inequality we are done:
[TABLE]
i.e.
[TABLE]
which reduces to
[TABLE]
By using inequality (3) it is enough to prove, for any , any and any , that
[TABLE]
and we prove this inequality by induction on .
First fix and choose . In this case (4) becomes
[TABLE]
which is true. Now if we assume that (4) is verified for , it is easy to check it for , thus completing the proof of (4).
It remains to prove that if satisfies condition 2, then the system of quadrics containing is empty.
If then we are in the previous case. We may assume that there exists such that . If the sequence is nondecreasing then in particular . This implies easily that the quadrics containing the first components are singular along the hyperplane , so the only existing quadric is the double hyperplane . By assumption , hence there is an extra condition and as we wanted.
We may assume that there exists such that and we pick the first such . In particular it follows
[TABLE]
As above, all the quadrics containing are singular along the linear space . Let . By definition .
Let be the projection from on a linear space . By (5) we have . Hence there is a bijective correspondence between and .
Note that
[TABLE]
hence if for we conclude again by the first case. If there exists such that , notice that in such a case we have
[TABLE]
hence
[TABLE]
We get that
[TABLE]
This means that satisfies the assumption 2 on and then by (complete) induction on we get that as we wanted.
A straightforward consequence of the previous theorem is the following corollary.
Corollary 3.3
A general zero-dimensional scheme contained in a union of double points with imposes independent conditions on quadrics if and only if for all .
Theorem 3.2 provides a classification of all the types of general subschemes of a collection of double points of which do not impose independent conditions on quadrics. For example in , the only case is given by two double points. In and in we have the following lists of subschemes which do not impose independent conditions on quadrics.
[TABLE]
Table 1: List of exceptions in
[TABLE]
Table 2: List of exceptions in
4 Cubic polynomials
In this section we generalize the approach of [BO, Section 3] to our setting and we prove the following result.
Theorem 4.1
A general zero-dimensional scheme contained in a union of double points imposes independent conditions on cubics with the only exception of and given by double points.
First we give the proof of the previous theorem in cases .
Lemma 4.2
Let be or . Then a general zero-dimensional scheme contained in a union of double points imposes independent conditions on cubics with the only exception of and given by double points.
Proof. By Remark 2.2 it is enough to prove the statement for with degree .
Let and a subscheme of a collection of double points with . Fix a line in and consider the Castelnuovo exact sequence
[TABLE]
It is easy to prove that it is always possible to specialize some components of on so that and that the residual does not contain two double points. The last condition ensures that for . Hence we conclude by Corollary 3.3.
In the case , the scheme has degree . Since there are no cubic surfaces with five singular points we can assume that contains at most three double points. Indeed if contains double points we can degenerate it to a collection of double points. We fix a plane in and we want to specialize some components of on so that and that the residual imposes independent conditions on quadrics. By looking at Table 1, since , it is enough to require that is not of the form or . It is easy to check that this is always possible: indeed specialize on the components of starting from the ones with higher length and keeping the residual as minimal as possible until the degree of the trace is or . If the degree of the trace is and there is in a component with length or we can obviously complete the specialization. The only special case is given by of type and in this case we specialize on the two double points and two components of length so that each of them has residual .
If the case of double points is exceptional. Assume that has degree and contains at most double points. We fix a hyperplane of and we want to specialize some components of on so that and that the residual imposes independent conditions on quadrics. By looking at Table 2, it is enough to require that does not contain two double points, does not contain one double point and two components of length and it is not of the form . It is possible to satisfy this conditions by specializing the components of in the following way: we specialize the components of on starting from the ones with higher length and keeping the residual as minimal as possible until the degree of the trace is maximal and does not exceed . Then we add some components allowing them to have residual in order to reach the degree . It is possible to check that this construction works, except for the case where we have to specialize on all the double points and of the components with length so that both have residual . It is easy also to check that following the construction above the residual has always the desired form, except for of the form , where the above rule gives a residual of type . In this case we make a specialization ad hoc: for example we can put on six components of length and the unique component of length in such a way that all them have residual and we obtain a residual of type which is admissible.
Now we have to check the schemes either contained in double points or containing double points. But this follows immediately by Remark 2.3.
We want to restrict a zero dimensional scheme of to a given subvariety . We could define the residual as a subscheme of the blow-up of along ([AH3]), but we prefer to consider just as an integer associated to and . More precisely given a subvariety , we denote . In particular we will use this notion in the following cases:
[TABLE]
where are three general subspaces of codimension three. We also recall that
[TABLE]
and
[TABLE]
[TABLE]
The proof of Theorem 4.1 relies on a preliminary description, which is inspired to the approach of [BO]. More precisely the proof is structured as follows:
in Proposition 4.3 below we generalize [BO, Proposition 5.2],
- -
in Proposition 4.7 and Proposition 4.8 we generalize [BO, Proposition 5.3],
- -
the analogue of [BO, Proposition 5.4] is contained in Lemma 4.9, Lemma 4.10, Lemma 4.11 and Proposition 4.12.
Proposition 4.3
Let and let be general subspaces of codimension . Let be a general scheme contained in a union of double points, where (resp. , ) is supported on (resp. , ), such that the triple is one of the following
- (i)
** 2. (ii)
** 3. (iii)
** 4. (iv)
** 5. (v)
**
then there are no cubic hypersurfaces in which contain and which contain .
Proof. For it is an explicit computation, which can be easily performed with the help of a computer (see the appendix).
For the statement follows by induction on . Indeed if it is easy to check that there are no quadrics containing . Then given a general hyperplane the Castelnuovo sequence induces the isomorphism
[TABLE]
hence specializing the support of on the hyperplane , since the space is empty, we get
[TABLE]
then our statement immediately follows by induction.
Remark 4.4
It seems likely that the previous proposition holds with much more general assumption. Anyway the general assumption is too weak, indeed the triple cannot be added to the list of the Proposition 4.3. Indeed there are two independent cubic hypersurfaces in , containing , , , two general double points on and seven general double points on , as it can be easily checked with the help of a computer (see the appendix). Quite surprisingly, the triple could be added to the list of the Proposition 4.3, and we think that this phenomenon has to be better understood. In Proposition 4.3 we have chosen exactly the assumptions that we will need in the following propositions, in order to minimize the number of the initial checks.
For the specialization technique we need the following two easy remarks.
Remark 4.5
Let , be two codimension three subspaces of , for . Let be a general scheme contained in a double point supported on such that , . Then there is a specialization of such that the support of is on , and .
Remark 4.6
Let be a codimension three subspaces of . Let be a scheme contained in a double point .
- i)
If then there is a specialization of which is supported at such that . 2. ii)
If then there are two possible specializations of which are supported at such that or . 3. iii)
If then there are three possible specializations of which are supported at such that , or . 4. iv)
If then there are four possible specializations of which are supported at such that , , or [math].
Proposition 4.7
Let and let be subspaces of codimension three. Let be a scheme contained in a union of double points such that (resp. ) is supported on (resp. ) and it is general among the schemes supported on (resp. ) and is general. Assume that the following further conditions hold:
[TABLE]
Then there are no cubic hypersurfaces in which contain and which contain .
Proof. For it is an explicit computation (see the appendix).
For , the statement follows by induction from to . Indeed given a third general codimension three subspace , we get the exact sequence
[TABLE]
where the dimensions of the three spaces in the sequence are respectively , and .
We will specialize now some components of on and some components of on . We denote by the union of the components of supported on and by the union of the components of supported on . Since we may assume also that . Analogously let and denote the corresponding subschemes of . Now we describe more explicitly the specialization.
From the assumption
[TABLE]
it follows that in particular has at least three irreducible components and so we may specialize all the components of on in such a way that .
Notice that the degree of the trace satisfies the same inductive hypothesis
[TABLE]
and we have
[TABLE]
If , by using that
[TABLE]
we get
[TABLE]
[TABLE]
then we can specialize and in such a way that and , indeed the conditions
[TABLE]
[TABLE]
are true for and guarantee that the inductive assumptions are true on the trace.
Now if , we have
[TABLE]
[TABLE]
and we can specialize in such a way that and . Indeed we have, for
[TABLE]
[TABLE]
In any of the previous cases, the residual satisfies the assumptions of Proposition 4.3, while the trace satisfies the inductive assumptions on . In conclusion by using the sequence
[TABLE]
we complete the proof.
The following proposition is analogous to the previous one, with a different assumption on . In this case we need an extra assumption on and , namely that in one of them there are enough irreducible components with residual different from . The reason for this choice is that it makes possible to find a suitable specialization with residual , or , by the Remark 4.5 (if all the components have residual , this should not be possible).
From now on we denote by (resp. ) for the union of the irreducible components of (resp. ) such that (resp. ).
Proposition 4.8
Let and let be subspaces of codimension three. Let be a scheme contained in a union of double points such that (resp. ) is supported on (resp. ) and it is general among the schemes supported on (resp. ) and is general. Assume that either the number of the irreducible components of , or that the number of the irreducible components of is at least . Assume that the following further conditions hold:
[TABLE]
Then there are no cubic hypersurfaces in which contain and which contain .
Proof. For it is an explicit computation (see the appendix), and the thesis is true even without the assumption on .
For the statement follows by induction from to , by using possibly also Proposition 4.7. As in the previous proof, given a third general codimension three subspace , we get the exact sequence
[TABLE]
We will specialize now some components of on and some components of on . We use the same notations as in the previous proof, and we describe more precisely the specialization in the following two cases.
Assume first that
[TABLE]
In particular has at least four irreducible components and we may specialize all the components of on in such a way that
[TABLE]
and so we have
[TABLE]
In particular it follows
[TABLE]
[TABLE]
We divide into two subcases.
In the first one we assume that the number of the irreducible components of is at least . In this case we can specialize and in such a way that and . Moreover there exists a specialization such that has at least components with residual or . Indeed in we keep at most one of these components, and if we are forced to keep three components of length one, it means that there are no components of length in , which implies our claim.
Notice that the conditions
[TABLE]
[TABLE]
are true for . They are also true for as soon as , so we need only to check the cases and , which implies .
In this case we specialize and in such a way that , and has at least components with residual or . The conditions
[TABLE]
[TABLE]
are true if or if and .
So the remaining cases to be considered are when , , and , that is when the triple
[TABLE]
is one of the following: , , , which have been checked with random choices (see the appendix) with a computer.
In the second subcase, we know that the number of the irreducible components of is at least . Then we can specialize and in such a way that and . As above it is easy to check that there exists a specialization such that has at least components with residual or .
Notice that the conditions
[TABLE]
[TABLE]
are true for as soon as one of the following conditions is satisfied
- (a)
, which implies .
- (b)
, , which implies
Assume then that (a) and (b) are not satisfied.
We have and we specialize and in such a way that and . The conditions
[TABLE]
[TABLE]
are true for or if and .
So the remaining cases to be considered are when , and (by case (b)) . The only remaining case are
[TABLE]
which we have checked with a computer. 2. 2.
Assume now that
[TABLE]
which implies
[TABLE]
In particular has at least five irreducible components and we may specialize all the components of on in such a way that .
In this case we have
[TABLE]
[TABLE]
and we can specialize and in such a way that and . Notice that the conditions
[TABLE]
[TABLE]
are true for and also for as soon as .
Assume now that and , which implies .
In this case we specialize and in such a way that and . The conditions
[TABLE]
[TABLE]
are true for and also for if .
The only remaining cases to be considered are then
, , and that is when the triple
[TABLE]
is one of the following: , , which we have checked with a computer.
In conclusion in any the previous cases we conclude by using the sequence
[TABLE]
since the trace satisfies the inductive assumptions on and the residual satisfies the hypotheses of Proposition 4.3.
Let be a scheme, contained in a union of double points, of degree with and be a subspace of codimension three. Assume that and that contains at most one component of degree . Let be the number of components of of degree for and let () be the degree of the component of of degree . Note that . Let us choose an order on the irreducible components of in such a way the length of any component is non increasing.
We consider one of the following two specializations where is supported on and is supported outside :
we choose as the union of the irreducible components of , starting from the ones with maximal length, in such a way that and it is minimal. By construction . Let be the number of components of of degree for . Then
[TABLE]
we choose as the union of the irreducible components of , starting from the ones with maximal length, in such a way that and it is minimal. By construction . Let be the number of components of of degree for . Then
[TABLE]
In both the specializations let us denote: and note that we have some freedom to specialize on , according to Remark 4.6. If we have a specialization with and another specialization with then for any value between and there is a suitable specialization such that attains that value. We will use often this technique by evaluating the maximum (resp. the minimum) possible value of under a specialization.
Lemma 4.9
If in the specialization we have
[TABLE]
then we have and there exists a specialization of type such that .
Proof. From the assumptions it follows that for any and with . Then consists of points of maximal length with at most one component of of length and at most one component of length . Hence consists only of double points, hence we get , hence we have , which is an integer, so that is an integer, so that .
It follows that , hence the maximum degree of is , the minimum degree is , and we can choose .
Lemma 4.10
If in the specialization we have
[TABLE]
then there exists a specialization of type such that either or .
Proof. Assume first . Since , from the assumption it follows
[TABLE]
Notice also that
[TABLE]
hence
[TABLE]
These two conditions imply that we have only the following possibilities:
[TABLE]
In all these cases, by performing the specialization of type , we have or . Moreover it is easy to check that if , if , and for any . In any case the difference between the maximum degree of the trace and the minimum degree satisfies
[TABLE]
We have and so
[TABLE]
In the first two cases, where and , we assume first , then the maximal degree of the trace is
[TABLE]
since , moreover and so we have that either , or work. It remains the case where we get that in we have three points of length , then either and , or and . By substituting in the hypothesis of our lemma the values we get and . In this case the maximal degree of the trace satisfies
[TABLE]
and, since , the choice works.
Now consider the case , where . Assume first , then the maximal degree of the trace is
[TABLE]
since , moreover so that either , or work. It remains the case , where we have either and , or and . By substituting in the hypothesis of our lemma the values , for , we get and . Then we have and so the maximal degree of the trace is
[TABLE]
and since the difference , the choice works.
In the case , if , then the maximal degree of the trace is
[TABLE]
and since it follows that either , or work. It remains the case where or . By substituting in the hypothesis of our lemma the values we get and . In this case the maximal degree of the trace is
[TABLE]
and since we can choose .
In the last case , if , then the maximal degree of the trace is
[TABLE]
and since it follows that either , or work. It remains the case where or . By substituting in the hypothesis of our lemma the values we get and , which is a contradiction. Then this case is impossible.
Now assume that . In this case we have also , hence it follows and for any . By assumption we have
[TABLE]
and, as in the first case, we also have
[TABLE]
These two inequalities imply that lies in the tetrahedron with vertices , , , . The only integer points in this tetrahedron are and .
In the case the maximal degree of the trace is
[TABLE]
and clearly the minimal degree is , thus one of the choices or works. In the case the maximal degree of the trace is
[TABLE]
and the minimal degree is obviously , so that one of the choices or works.
Lemma 4.11
If all the assumptions of Lemma 4.9 and Lemma 4.10 are not satisfied, then there exists satisfying , and every can be attained by a convenient specialization of type .
Proof. The maximal degree of the trace is
[TABLE]
.
Since the assumption of Lemma 4.10 are not satisfied, we have .
The minimal possible degree of the trace is
[TABLE]
[TABLE]
where we use the fact that . This is true because either or and we have
[TABLE]
Hence if we choose . Otherwise if we choose .
Both cases work because of the assumption
[TABLE]
We can now prove the last preliminary proposition. Recall that we denote by for the union of the irreducible components of such that .
Proposition 4.12
Let and let be a subspace of codimension three. Let be a scheme contained in a union of double points such that is supported on and is general among the schemes supported on and is general. Assume that , and that , for . We also assume that the number of the irreducible components of is . Then there are no cubic hypersurfaces in which contain and which contain .
Proof. For it is a direct computation (see the appendix).
For the statement follows by induction, and by the sequence
[TABLE]
where is a general codimension three subspace. We get
[TABLE]
First by Lemmas 4.9, 4.10, 4.11 we can specialize in such a way that (we will call in the following ), is supported on and , where , , and . Notice also that we have . It follows that
[TABLE]
Moreover let us specialize where is supported on and is supported on . We may also assume that the number of irreducible components of is . We may assume that
[TABLE]
indeed note that and there exist at least irreducible component in . Note that by using the minimal number of irreducible component in , at least components remain in , preserving our inductive assumption. It follows that
[TABLE]
moreover we have clearly
[TABLE]
and we may apply Proposition 4.7 and Proposition 4.8, since the scheme satisfies the corresponding assumptions. Then we conclude by induction, indeed the scheme satisfies our assumptions with respect to the spaces and . Precisely we have (by subtraction)
[TABLE]
and , where .
We are finally in position to give the proof of the main theorem.
Proof of Theorem 4.1. We fix a codimension three linear subspace and we prove the statement by induction by using the exact sequence
[TABLE]
We prove the claim by induction on from to . By Lemma 4.2 we know that the theorem holds for . Let be a general scheme contained in a collection of double points and with .
Since we can assume that contains at most one component of length . Fix a codimension three linear subspace and consider the exact sequence
[TABLE]
We want to specialize on some components of so that and apply Proposition 4.12.
We keep outside the irreducible components of starting from the ones with maximal length in such a way that and it is minimal. We get by construction that . Let be the number of components of of degree for and let be the degree of the component of of length . Then .
After the specialization, the minimum degree of the trace is
[TABLE]
if or
[TABLE]
if . On the other hand the maximum degree of the trace is
[TABLE]
We want to prove that belongs to the range between the minimum and the maximum of . This is implied by the inequalities
[TABLE]
and
[TABLE]
In order to prove the inequality (7), consider first the case . Then and we have
[TABLE]
[TABLE]
as we wanted. If we get
[TABLE]
[TABLE]
which is if , as we wanted.
In order to prove the inequality (8), notice that
[TABLE]
then if we conclude since , while if we conclude by the inequality , which is true if , in particular if .
Consider the last case and . If , so that is an integer, then contains at least irreducible components and this confirms the inequality. If , even double points and one component of length are not enough to cover all . Then contains at least irreducible components and again the inequality is confirmed.
Then a suitable specialization of exists such that . We denote again by for the union of irreducible components of such that .
In order to apply Proposition 4.12 we need only to show that the irreducible components of are at least . If this condition is not satisfied, we show now that it is possible to choose another suitable specialization such that again but the number of irreducible components of is . We assume that the number of irreducible components of is . Indeed we may perform the following operations, that leave the degree of the trace and of the residual both constant.
- •
Pull out a component from to and push down another component from to .
- •
Pull out a component from to and push down a component of .
- •
Pull out two components from to and push down a component from to .
After such operations have been performed, we get that is still a specialization of a subscheme of , allowing our semicontinuity argument.
If no of the above operations can be performed, then contains only components of length , contains only components of length contains only components of length and components of length .
Then we get
[TABLE]
hence
[TABLE]
On the other hand, we have also
[TABLE]
[TABLE]
[TABLE]
where the last inequality is true for . This contradiction concludes the proof.
5 Induction
In order to prove Theorem 1.1 we will work by induction on the dimension and the degree. In the following lemmas we describe case by case the initial and special instances, while in Theorem 5.6 below we present the general inductive procedure, which involves the differential Horace method.
Lemma 5.1
A general zero-dimensional scheme contained in a union of double points imposes independent conditions on for any , with the only exception of and given by the collection of double points.
Proof. Assume that is a general subscheme of a union of double points with . If is a collection of double points the statement follows from the Alexander-Hirschowitz theorem on (for an easy proof see for example [BO, Theorem 2.4]).
If is not a collection of double points, fix a hyperplane . Since contains at least a component of length or , it is clearly always possible to find a specialization of such that the trace has degree exactly . Then we conclude by induction from the Castelnuovo sequence
[TABLE]
Notice that any subscheme of double points and any scheme containing double points impose independent conditions on quartics, by Remark 2.3.
We give now an easy technical lemma that we need in the following.
Lemma 5.2
Assume that is a general zero-dimensional scheme contained in a union of double points of , which contains at least components of length less than or equal to . Then if it is possible to specialize some components of on a fixed hyperplane in such a way that .
Proof. By assumption there exist at least components with . Specialize on the hyperplane in such a way that the residual of each component is zero. Then specialize other components so that
[TABLE]
is minimal. If the claim is proved, so assume . Obviously we have , where is the minimal length of the components of which lie outside . Let be a component with length . Now we make the first components having residual with respect to and we specialize on with residual . Notice that this is possible since .
Lemma 5.3
Fix . A general zero-dimensional scheme contained in a union of double points imposes independent conditions on , with the following exceptions:
- •
* and either is the union of double points, or is the union of double points and a component of length ;*
- •
* and is the union of double points.*
Proof. If is a collection of double points, the statement holds by the Alexander-Hirschowitz theorem. Assume that is a scheme with degree which is not a union of double points. Let us denote by the number of double points in and by the number of the components with length less than or equal to .
If and , then and is the exceptional case of the statement. If and , then