1
On Horn-Kapranov uniformisation of the discriminantal loci
Susumu TANABÉ
November 27, 2006
Abstract. In this note we give a rational uniformisation equation of the discriminant loci associated to a non-degenerate affine complete intersection variety. To show this formula we establish a relation of the fibre-integral with the hypergeometric function of Horn and that of Gel'fand-Kapranov-Zelevinski.
0 Introduction
In this note we give a concrete rational uniformisation equation for the discriminantal loci of non-degenerate affine complete intersection depending on deformation parameters.
First of all, let us fix the situation. For the complex varieties
and
we consider the mapping,
such that
Let
be polynomials that define a non-degenerate complete intersection (CI) in the sense of Danilov-Khovanski [3] with the following specific form:
|
(0.2)
|
where
Let
be the dimension of the variety
Then it is known that the discriminantal loci of
coincides with that of
That is to say, the study of the discriminantal loci of a CI can be reduced to that of an hypersurface associated with the original CI in a special manner. This fact has been discovered by Arthur Cayley [5] and thus the method to reduce the geometric study of a CI to that of an hypersurface is named ”Cayley trick” in general, even in contexts apart from the study of discriminantal loci (e.g. the description of the mixed Hodge structure of the former by means of the latter given by T.Terasoma, A.Mavlyutov [9] and others). Here we return to the initial spirit of Cayley who treated the question of the discriminantal loci.
The main idea is based on that of the paper [6] which states that the singular loci of the linear differential operators annihilating the fibre integrals of
coincide with the discriminantal loci of
In the modern terminology of the A-hypergeometric functions (HGF), it is equivalent to say that A-discriminantal loci are singular loci for generalized A-HGF. This fact has been proven in [7] and we give a more precise description of the discriminantal loci by means of combinatorial data of the polynomial mapping
and the toric geometry of
(see Theorem 2.6).
AMS Subject Classification: 14M10, 55R80, 32S40, 33C65.
Key words and phrases: complete intersection, discriminantal set, Horn hypergeometric function. Let us review the contents of the note in short. In §1 we remind us of some basic facts on the Cayley trick and Néron-Severi torus. In §2, we calculate the Mellin transform of the fibre integral in an explicit manner.In making use of a representation of the Mellin transform we show that fibre integral satisfies the Horn type system of differential equations (Theorem 2.4). From this expression of the Horn type system, we get the discriminantal loci as the boundary of a convergence domain of solutions to the system. In §3, we show that the fibre integral calculated in §2 is nothing but the quotient of the Gel'fandKapranov-Zelevinski generalized hypergeometric function (HGF) by the torus action. In §4 we give two computational examples: discriminantal loci for the
type singularity and the simplest non-quasihomogeneous complete intersection.
Finally we remark that this note is an abridged version of some parts from [13 ]
where one can find more details.
1 Cayley trick and Néron-Severi torus
Throughout this section we keep the notation of §0. Further we introduce the following notations.
Let
be the complex algebraic torus of dimension
We denote by
the monomial
with multi-index
and by
the
volume form
We shall also use the notations
and
and their analogies for each variable. In this section we consider an extension of the mapping
to that defined from
to
We follow the construction by [2] and [9] . Let us define
as the dimension of a minimal ambient space so that we can quasihomogenize simultaneously the polynomials
by multiplying certain terms by new variables:
Let us denote by
the new polynomials obtained in such a way. These polynomials are quasi-homogeneous with respect to certain weight system i.e. there exists a set of positive integers
such that their G.C.D. equals 1 and the following relation holds:
where
is some positive integer and
|
(1.1)
|
an Euler vector field.
Example We modify the polynomial
with
in adding a new variable
so that the new polynomial
becomes quasihomogeneous with respect to the weight system
.
In general there are of course many choices of terms that we modify to realize the quasihomogeneiety.
From now on we will use the notation
and that of the polynomial
If we introduce the Euler vector field,
we have the following relation:
From now on we denote
Let
be an integer lattice of rang
and
be its dual,
We denote by
(resp.
) the natural extension of
(resp.
) to its real space. Let us take
a set of generators of one dimensional cones such that
We can define a simplicial fan
in
as a set of simplicial cones spanned by the above
. Our construction of the Euler vector field
correspond to the superstructure
with a basis of generators
such that
Here we have
for the projection
While the dimension of the vector space
must be minimal i.e.
We introduce a polynomial,
|
(1.2)
|
in adding new variables
Let
be the elements of the set
We define a simplicial rational fan
in
as a set of simplicial cones generated by
We consider the injective homomorphism
for
defined by
The cokernel of this mapping is a free abelian group,
for which the following group can be defined
|
(1.3)
|
As a matter of fact this group
is isomorphic to an algebraic torus
One can define the toric variety
associated to the affine space,
To this end we proceed following way after the method initiated by M.Audin. Let
be a monomial defining a coordinate plane and the ideal
Let
be the variety defined by the ideal
We construct the toric variety
as the quotient of
by the group action
:
with
Definition 1
This group
is called the Néron-Severi torus associated to the fan
We introduce the following polynomial (named phase function below),
|
(1.4)
|
that will play essential rôle in our further studies. In §3, we treat the following affine variety defined for
:
|
(1.5)
|
Further on we shall prepare several lemmata on combinatorics which are useful for the derivation of the discriminant loci equation. We denote by
the number of monomials in
that take part in the phase function
for
. That is to say
Here we introduce new variables
that satisfy the following relations,
|
(1.6)
|
Each
represents the
th monomial present in
(see (2.4) below). We introduce the following matrix
whose column is a vertex of the Newton polyhedron
|
(1.7)
|
Further we assume that
We always assume the inequality
for
. In this situation we can define a non-negative integer
as the minimal number of variables
|
(1.8)
|
to make the number of variables present in the expression
equal to
. That is to say
For example, the relation
may be modified into the following form:
|
(1.6′)
|
In other words, proper addition of new variables
to
make the polynomial
quasihomogeneous. In this way we have
Further we shall consider a simple parametrisation of the variety
|
(1.10)
|
Namely we denote,
|
(1.11)
|
|
(1.12)
|
|
(1.13)
|
Then we have, for example, a linear equation equivalent to
that can be written down as follows,
|
(1.14)
|
Let us write down the relation between
and
by means of a matrix
|
(1.15)
|
Below the columns
(resp.
) of the matrix
(resp.
) shall always be ordered in accordance with (1.11), (1.12), (1.13) if not especially mentioned on the exception.
For the polynomial mapping (0.2), the choice of monomials to be modified by supplementary variables is a bit delicate. Namely, we have to observe the following rules to avoid the degeneracy of the matrix
of the relation
Lemma 1.1
For (0.2) and
, we get a non-degenerate matrix
if we observe the following rules:
a. For the fixed index
, it it necessary to choose at least one of monomials
that remains without modification.
b. For the fixed index
it it necessary to choose at least one of monomials
that remains without modification.
We recall here the notion of non-degenerate hypersurface,
Definition 2
The hypersurface defined by a polynomial
is said to be non-degenerate if and only if for any
the following inclusion takes place,
where
We call the CI
for (0.2) non-degenerate if the hypersurface
is non-degenerate.
The following is an easy consequence of the above Definition.
Proposition 1.2
If the matrix
is non-degenerate, the hypersurface
and the CI
are non-degenerate in the sense of the Definition 2 .
2 Horn's hypergeometric functions
From this section, we change the name of variables
into
We use both of the notations
Let us consider the Leray's coboundary (see [14] ) to define the fibre integral,
such that
. Further on central object of our study is the following fibre integral,
|
(2.1)
|
and its Mellin transform,
|
(2.2)
|
for certain cycle
homologous to
which avoids the singular loci of
(cf. [11] ). After Definition 1 above, we understand that
is a variable on the Néron-Severi torus. Thus the fibre integral
is a ramified function on the torus
It is useful to understand the calculus of the Mellin transform in connection with the notion of the generalized HGF in the sense of Mellin-Barnes-Pincherle [1] , [10] . After this formulation, the classical HGF of Gauss can be expressed by means of the integral,
Next we modify the Mellin transform
with
Here we made use of notations
and
the
volume form on
In the above transformation we used a classical interpretation of Dirac's delta function as a residue:
We introduce the notation
One shall not confuse it with the thimble of Lefschetz, because
is rather a tube without thimble. We will rewrite the last expression,
where
|
(2.3)
|
in which each term
stands for a monomial in variables
of the phase function
We transform the above integral into the following form,
|
(2.4)
|
Here
means the image of the chain in
into that in
induced by the transformation
We define
The second equality of
follows from Proposition 2.1, 3) below that can be proven in a way independent of the argument to derive
We will denote the set of columns and rows of the matrix
by
Here we remember the relation
The following notion helps us to formulate the result in a compact manner.
Definition 3
A meromorphic function
is called
periodic for
if
for some rational function
For the simplicial CI
(i.e. we can construct
for which the matrix
is non-degenerate), we have the following statement.
Proposition 2.1
1)For any cycle
the Mellin transform
can be represented as a product of
function factors up to a
periodic function factor
,
with
|
(2.5)
|
Here the following matrix
has integer elements,
|
(2.6)
|
with
for all
In this way
is uniquely determined. The coefficients of (2.5) satisfy the following properties for each index
:
Either
i.e.
Or
2) For each fixed index
the following equalities take place:
|
(2.7)
|
3) The following relation holds among the linear functions
:
Proof 1) First of all we recall the definition of the
function,
for the unique non-trivial cycle
that turns around
with the asymptotes
We consider a transformation of the integral
induced by the change of cycle
defined by the relation,
By the aid of this action the chain
turns out to be homologous to a chain,
with
This fact explains the appearance of the factor
apart from the factors of type
In the sequel we analyze the
function factors that arise from the integral (2.4). To this end, we represent the matrix
(resp.
) as a set of
columns properly ordered:
|
(2.8)
|
The interior product of vectors
and
defines the linear function in question:
|
(2.9)
|
The vector columns of
are divided into 3 groups:
1 the columns with all formally non-zero elements.
2 with unique non-zero element
that produces
and
in (2.9).
3 with the non-zero elements that produce a function linear in
after (2.5). In the further argument, only the first two groups of columns are important.
The column that corresponds to
of
consists in the unique non-zero element
at the position
Meanwhile the column of
that corresponds to the variable
consists also of an unique non-zero element
outside the positions
Let us denote this correspondence by
that yields in
Here the mappings
are injections that send the number of columns corresponding to the variables
to the total set of indices
. We divide the columns of
into
groups
each of which corresponds to
For this group, one can claim following assertions.
The column
with
non-zero elements
For the vectors
of the case
above,
|
(2.10)
|
and there exists another vector of the same group
that satisfies:
|
(2.11)
|
where
is the Kronecker delta symbol. The vector
corresponds to the group 2.
Thus the columns of the group
(resp.
) give rise to the linear functions of the group
(resp.
).
2) The 1-st
th vector rows of the matrix
are orthogonal to the vectors
above. This means the relations (2.7).
3) The statement can be deduced from 2). Q.E.D. In view of the Proposition 2.1 , we introduce the subsets of indices
as follows.
Definition 4
The subset
(resp.
) consists of the indices
such that the coefficient
of
(2.5) is positive (resp. negative, zero). Analogously we define the subset
(resp.
) that consists in such indices
that the coefficient
of
is positive (resp. negative, zero).
To assure the convergence of the Mellin inverse transform of
from (2.1) in a properly chosen angular sector in the variables
, we shall verify that the Mellin transform
admits the following estimation modulo multiplication by a
periodic function
for certain
Here we remember an elementary lemma for the integral:
|
(2.12)
|
Lemma 2.2
If one chooses one of the following functions
(resp.
) in terms of
then the integrand of (2.12) is exponentially decaying as
tends to
within the sector
(resp.
)
avec
Proof It is enough to recall
while
Here we used the formula of Binet:
if
. The factor
for
gives the exponentially decreasing contribution in each cases. Q.E.D. Let us introduce a simplified notation,
Lemma 2.3
The sufficient conditions so that
|
(2.13)
|
defines a polynomially increasing function with
a properly chosen
periodic function (including the infinity
) are the following.
i) For every
ii) The real number
is non negative.
To see the exponential decay property of the integrand, one shall make reference to Nörlund's trick [10] . Further we apply the Stirling's formula on the asymptotic behaviour of the
function (Whittaker-Watson, Chapter XII, Example 44).
If we apply this lemma to our integral, we see that there exists a cycle
such that
|
(2.14)
|
with a function
rational with respect to
. Here we remember the relation
Thus we get the theorem on the Horn type system.
Theorem 2.4
The integral
satisfies the hypergeometric system of Horn type as follows:
|
(2.15−1)
|
with
|
(2.15 − 2)
|
|
(2.15 − 3)
|
where
are the sets of indices defined in Definition 4 .
|
(2.15−4)
|
|
(2.15 − 5)
|
|
(2.15 − 6)
|
where
are the sets of indices defined in the Definition 4 . The degree of two operators
are equal. Namely,
|
(2.16)
|
Analogously,
The proof is mainly based on the Proposition 2.1 . To deduce
from the Mellin transform
we use the following well known recurrence relation:
if
a positive integer.
if
a negative integer.
The evident compatibility (i.e. integrability) of the above system
in the sense of Ore-Sato ([12] ) can be formulated like the following cocycle condition. To state the proposition we introduce the notation
Proposition 2.5
The rational expression
|
(2.17)
|
defined for the operators
satisfies the following relation:
|
(2.18)
|
Similarly for
|
(2.19)
|
satisfies the following relation:
|
(2.20)
|
We introduce here the main object of our study: the discriminantal loci of the CI defined by the polynomials
|
(2.21)
|
As it is easy to see [5] ,
coincides with the discriminantal loci of
Let us define the
th roots of rational functions associated with the linear functions
as follows.
|
(2.22)
|
|
(2.23)
|
|
(2.24)
|
By virtue of the property
the rational function
(resp.
) is of weight zero with respect to the variables
and thus it is possible to consider the mapping
defined on
instead of
Let
be a polynomial that defines the discriminantal loci
without multiplicity.
Theorem 2.6
The image of
is identified with the discriminantal loci
if we choose a proper
th branch in the equations
Proof From the system of equations
we see that
is contained in the set:
|
(2.25)
|
here we use the notation
The existence of
in
is equivalent to the existence of
Thus the set
admits a representation,
While after Theorem 2.1, a) and Remark 2.4 of [7] , this set
coincides with
if
As for the case
it is natural to consider the
covering
of the mapping
while the branch of the image of
shall be specified in a proper way. To do that we remark that
where the difference
consists of the divisors that arise from the
branching effect
In considering
we shall discard the superfluous
branching effect
Q.E.D. The mapping
is nothing but the inverse mapping of the logarithmic Gauss map;
This is a direct consequence of the cocycle property
of the operators
and
see [7] , Theorem 2.1, b).
3
Hypergeometric function of Gel'fand-Kapranov-Zelevinski
Let us consider the set of polynomials with deformation parameter coefficients
associated to the polynomial system
|
(3.1)
|
For the sake of simplicity we will further make use of the notation
We consider the Leray coboundary
of a cycle
of the CI
Then we can define the
hypergeometric function
introduced by Gel'fand-Zelevinski-Kapranov [4] associated to the polynomials,
Namely it is defined as a kind of multiple residue along
|
(3.2)
|
We impose here the non-degeneracy condition of the Definition 2 for the complete intersection
after the procedure described in §1.
In the sequel we consider a lattice
of
vectors defined by the system of following linear equations:
Here we denoted by
a
basis of
For the subset
such that the columns
of the matrix
(1.7) span
over
and
we define the set of indices (a generalisation of the Frobenius' method) after [4] ,
which satisfy the following system of equations,
Let
be a triangulation of the Newton polyhedron
for
of (1.4) after the definition [4] , 1.2. Here we impose that
for
Let
be two different simplices of the triangulation
We suppose that
for
with
We introduce the condition of
non-resonance on
|
(3.3)
|
for any pair
An adaptation of theorem 3 [4] to our situation can find the following formulation.
Theorem 3.1
1) The
HGF
satisfies the following system of equations.
|
(3.4)
|
2) The dimension of solutions of the system above at a generic point
is equal to
if the
non-resonant condition
is satisfied.
In the sequel we shuffle the variables
in accordance with the order of their appearance and we define anew the indexed parameters
Let us introduce notations analogous to
|
(3.5)
|
We consider the equation
where the matrix
is constructed as follows. The columns
with vectors
defined like the column of the matrix
in (1.15). For the columns of number
to
The columns
the matrix
is obtained after implementation of the matrix
into the transposed matrix
between the
th and the
th column up to necessary permutations necessary after the implementation.
Proposition 3.2
There exists a cycle
such that the following equality holds for the integral defined in
|
(3.6)
|
here
The exponents
are determined by the following relation,
|
(3.7)
|
The transition of the cycle
to
is controled by the transformations,
Proof It is enough to remark the following property,
Q.E.D. One can thus conclude (at leat locally on the chart
for
)
HGF of GZK
is expressed by means of a fibre integral annihilated by the Horn system
One can find a similar statement in [7] where Kapranov restricts himself to a power series expansion of the solution to
.
Corollary 3.3
The dimension of the solution space of the system (3.3) at the generic point is equal to
if the
non-resonance condition
is satisfied.
Proof We shall consider the convex hull of vectors that correspond to the vertices of the Newton polyhedron of the polynomial
That is to say
They are located on the hyperplane
Thus it is possible to measure
dimensional volume
that is equal to
The Euler characteristic admits the following expression
after Khovanski [8] . Q.E.D. We define the
discriminantal loci
in
like following,
|
(3.8)
|
As it is seen from
the uniformisation equations
give rise to an uniformisation of
discriminantal loci
without
branching effect.
Corollary 3.4
We have the following relations among
located on the discriminantal loci
|
(3.9 − 1)
|
|
(3.9 − 2)
|
This allows us to express
by means of the deformation parameters
and
4 Examples
4.1 Deformation of
Let us consider the versal deformation of
singularity of the following form,
|
(4.1)
|
By means of the resultant calculus on computer, we get a defining equation of the discriminantal loci as follows,
|
(4.2)
|
This is a polynomial with quasihomogeneous weight
if we assign to the variables
the weights
Here we remark that
branch of the discriminantal locus
corresponds to the deformation of
singularity.
On the other hand, our Theorem 2.6 states that the uniformisation equation of the discriminantal loci for the deformation (i.e. torus action quotient of the deformation parameter space
on the chart
),
has the following form,
|
(4.3)
|
If we eliminate the variables
from the expressions
we get an equation
We recall here that our method requires that the expression
contains so much terms as the variables in it. The reason why the value
has been chosen is of purely technical character. In substituting the special value
for
in
we get,
4.2 Deformation of a non-quasihomogeneous complete intersection.
Let us consider the following pair of polynomials that define a non-degenerate complete intersection
in
,
|
(4.4)
|
The discriminant of this CI in
can be calculated as follows,
|
(4.5)
|
Evidently the fibres corresponding to the parameter values on the divisor
are contained in
Thus the discriminant of CI
is given by the third factor of
After Theorem 2.6, we can find an uniformisation equation of the discriminantal loci
,
|
(4.6)
|
If we eliminate the variables
from the expressions
we get an equation of
where
is a polynomial whose Newton polyhedron is contained in a four sided rectilinear figure with vertices
This factor contains the image of
outside of
References
-
P.Appell and J.Kampe de Feriet, Fonctions hypergéométriques et hypersphériques, Paris, 1926.
-
V.V.Batyrev and D.Cox, On the Hodge structure of projective hypersurfaces in toric varieties , Duke Math.J. 75,no.2, (1994), 293-338.
-
V.Danilov, A.G.Khovanski, Newton polyhedra and an algorithm for computing Hodge-Deligne numbers , Math.USSR Izv. 29 (1987),No.2, pp.279-298.
-
I.M.Gel'fand, M.M.Kapranov, A.V.Zelevinsky, Hypergeometric functions and toric varieties , Functional analysis and its appl. 23, no.1, (1989), pp. 12-26.
-
I.M.Gel'fand, M.M.Kapranov, A.V.Zelevinsky, Discriminants of polynomials in several variables and triangulations of Newton polytopes , Leningrad Math. J. 2, no.3, (1989), pp. 449-505.
-
J.Horn, Über die Convergenz der hypergeometrischen Reihen zweier und dreier Veränderlichen, Math. Ann.34,(1889),pp.544-600.
-
M.Kapranov, A characterisation of
discriminantal hypersurfaces in terms of the logarithmic Gauss map , Math. Ann.290,(1991),pp.277-285.
-
A.G.Khovanski, Newton polyhedra and the genus of complete intersections, Functional Anal.Appl. 12 (1978), pp. 5161.
-
A. Mavlyutov, Cohomology of complete intersections in toric varieties, Pacific J.Math. 191, (1999), No.1, 133-144.
-
I.Norlund, Hypergeometric functions, Acta Math. 94, (1955/56), pp.289-349.
-
M.Passare, T.Sadykov, A.G.Tsikh, Nonconfluent hypergeometric functions in several variables and their singularities, preprint Max Planck Institut für Mathematik, (2000).
-
T.Sadykov, On the Horn system of partial differential equations and series of hypergeometric type, Math.Scand. 91A(2002), No.1, pp.127-149.
-
S.Tanabe, Transformée de Mellin des intégralesfibres associées à l'intersection complète non-dégénérée, preprint math.AG/0405399
-
V.A.Vassiliev, Ramified integrals, singularities and Lacunas, Kluwer Academic Publishers, Dordrecht, 1995.
Indepent University of Moscow Bol'shoj Vlasievskij pereulok 11, Moscow, 121002, Russia E-mails: tanabe@mccme.ru, tanabesusumu@hotmail.com