Alicia Dickenstein
Abstract. We highlight the role of primary decomposition of binomial ideals in a commutative polynomial ring, in the description of the holonomicity, the holonomic rank, and the shape of solutions of multivariate hypergeometric differential systems of partial differential equations.
Partially supported by UBACYT X064, CONICET PIP 5617 and ANPCyT PICT 20569, Argentina
En honor a Mischa Cotlar, con afecto y admiración
There have been two main directions in the study of classical hypergeometric functions. The first of these is to study the properties of a particular series, going back to Euler [Eu1748] and Gauss [Gau1812]. The second one is to find a differential equation satisfied by the hypergeometric function, and to study all the solutions of that equation, going back to Kummer [Kum1836] and Riemann [Rie1857].
Recall that a (convergent) univariate series
In the generalization to hypergeometric functions in several variables, these two points of view suggest different approaches. A natural definition of hypergeometric power series in several variables was proposed by Horn [Hor1889, Hor31]. A series
in variables with complex coefficients is hypergeometric in the sense of Horn if there exist rational functions in variables such that

where denote the standard basis vectors of . As before, denote by the operator , . For all monomials and all polynomials it holds that , which implies again the following relation between the behaviour of the ratios of coefficients and the existence of hypergeometric differentials operators annihilating . If we write the rational functions as
where and are relatively prime polynomials and divides , the series satisfies the following Horn hypergeometric system of differential equations:

Gelfand, Graev, Kapranov, and Zelevinsky [GGZ87, GKZ89, GKZ90] developed a highly interesting point of view, by "dressing" the hypergeometric functions and operators with "homogeneities". This allows to understand the properties of classical hypergeometric systems via tools in algebraic geometry and combinatorics. They associated to a configuration of lattice points spanning (which we encode as the columns of a integer matrix, also called ) and a vector , a holonomic left ideal in the Weyl algebra in variables as follows.
]]> Definition 1.1. The hypergeometric system (or GKZhypergeometric system) with exponent is the left ideal in the Weyl algebra generated by the toric operators , for all such that , and the Euler operators for .We refer the reader to [SST00] for an account of hypergeometric systems with emphasis on computations.
A local holomorphic function is hypergeometric of degree if it is annihilated by . hypergeometric systems are homogeneous (and complete) versions of classical hypergeometric systems in variables [GGR92, DMS05], in the following sense. Let be a matrix whose colums span the integer kernel of , and also call the Gale dual lattice configuration given by the rows of . Consider the surjective open map
Let , be simply connected open sets and denote by the coordinates in . Given a holomorphic function , call , where . Then, , for . Moreover, for any , we have that if and only if , where and denote the following differential operators in and variables respectively:


where Note that system (1.2) corresponds to the choices (). On the side, this amounts to considering only codimension many binomial operators (corresponding to the columns of , that is to a basis of , and not the whole system of toric binomials in (1.1)).
Although the isomorphism sending to is only at the level of local holomorphic solutions and not at the level of modules, it preserves many of the pertinent features, including the dimensions of the spaces of local holomorphic solutions and the structure of their series expansions. This point of view allowed us in [DMS05] to study the holonomicity, the holonomic rank and the persistence of Puiseux polynomial solutions to Horn systems for via the translation into a multihomogeneous setting and the study of primary components of binomial ideals in a commutative polynomial ring. This work was widely generalized in the recent article [DMM06], where we propose the definition of a binomial module (see Definition 3.1 below) which comprises the notion of hypergeometric system. In this note we illustrate our perspective and results via examples and extended comments.
2. Examples: Holonomic rank and binomials
]]> Before giving more precise statements and definitions, we focus on classical examples under a modern "binomial" view.2.1. Horn and Appel hypergeometric systems in two variables.
We consider two classical bivariate Horn systems of hypergeometric partial differential equations as in (1.2):
 The system associated to Horn hypergeometric series is:
y_{1}(2θ_{y1}  θ_{y2} + a′)(2θ_{y1}  θ_{y2} + a′ + 1)  (θ_{y1} + 2θ_{y2} + a)θ_{y1}f = 0 , y_{2}(θ_{y1} + 2θ_{y2} + a)(θ_{y1} + 2θ_{y2} + a + 1)  (2θ_{y1}  θ_{y2} + a′)θ_{y}f = 0 , where are generic parameters.
 The system associated to Appell series is:
y_{1}(θ_{y1} + θ_{y2} + a)(θ_{y1} + b)  θ_{y1}(θ_{y1} + θ_{y2} + c  1)f = 0 , y_{2}(θ_{y1} + θ_{y2} + a)(θ_{y2} + b′)  θ_{y2}(θ_{y1} + θ_{y2} + c  1)f = 0, where and are generic parameters.
Recall that the holonomic rank of a system of linear differential operators with polynomial (or holomorphic coefficients) is the vector space dimension of the space of local holomorphic solutions around a point which is not in the singular locus of the associated module. Both systems are defined by two linear operators in two variables, so a first guess for the holonomic rank is in both cases. This is the right answer for Horn system, but Appell system has holonomic rank equal to . Moreover, Erdélyi noted in [Erd50] that, in a neighborhood of a given point, three linearly independent solutions of Horn system can be obtained through contour integral methods. He also found a fourth linearly independent solution: the Puiseux monomial . He remarked that the existence of this elementary solution was puzzling, and offered no explanation for its occurrence. Via the translation to a homogeneous system (which is part of an hypergeometric system, as in the Introduction), we gave in [DMS05] an explanation for these facts.
Look at the binomials
]]> in the commutative polynomial ring . The exponents are read from the (integer) coefficients of the linear forms in occuring in the system. Its zero set in has two irreducible components. One is the toric variety which is the closure of the solutions in the torus of the binomials , which has degree and corresponds to the fully supported solutions described by Erdélyi (we refer the reader to [Stu96] for background on toric varieties). The other component lies "at infinity", that is, in the union of the coordinate hyperplanes. Its multiplicity equals the intersection multiplicity at the origin of the system of binomials in variablesThe same result explains the holonomic degree for Appell system. The translation to the homogeneous setting gives the following two binomials:
For particular choices of parameters, Appell Horn system is not holonomic. Consider the matrix
and a vector of parameters . Denote by the partial derivative with respect to the variable and let , for . We can choose the following matrix , with Gale dual :
Set . The corresponding Horn system (1.2) looks in its "binomial incarnation" as follows:
]]>where , , , . If , then any (local holomorphic) function annihilated by the operators is a solution, for instance all monomials with satisfying . So, the space of such functions is infinitedimensional; in fact, it has uncountable dimension. Again, this phenomenon is explained in general in [DMS05].
2.2. Mellin hypergeometric system for algebraic functions.
Given coprime integers set

and The local roots of the generic sparse polynomial
viewed as functions of the coefficients, are algebraic solutions to the associated hypergeometric system [Ma37, Bir27, Stu00, CDD99, PT04]. In particular, any solution to this hypergeometric system has a double homogeneity property and can therefore be considered as a function of variables. This implies that we can arbitrarily prescribe the values of any two nonzero coefficients in without losing any essential information on the general solution to this equation. If we divide by and then set we obtain an algebraic equation of the form

A classical result of Mellin from 1921 (see [Mel21]) states that if is a local root of (2.1), then it satisfies the following system of partial differential equations:
]]> 
where and .
In [DS07] we studied the solutions of the Mellin system via homogenization and translation to the binomial setting. Mellin not only observed in [Mel21] that the roots of the algebraic equation (2.1) satisfy the Mellin system (2.2), but he also made the following remark. The solution around the origin which satisfies , is given by

Here and the empty product is defined to be All other solutions around the origin have the form

where we denote . It happens that the holonomic rank of Mellin system (2.2) equals . In the effort of getting equations such that in each of them a partial derivative of with respect to each of the variables is expressed in terms of other derivatives, the Mellin system has not only the roots of the algebraic equation (2.1) as solutions, but also the roots of the associated equations (2.4).
When we add homogeneities, we are led to an isomorphism between the solutions of the Mellin system and the solutions of a Horn system in variables associated to the following matrix :
]]> (2.5) 
Let and consider the Horn system , given by the operators defined in (1.4). The colums of generate a non saturated lattice of order and the binomials read in its columns
define a complete intersection variety with components not at infinity, given by torus translates of the toric variety associated to the matrix (which correspond to the associate algebraic equations (2.4). The degree of (and hence of all the components) is , so by [DMM06, Theorem 6.10], the holonomic rank of the Horn system equals , which explains the holonomic rank of the Mellin system. Moreover, we describe in [DS07, Theorem 4.3] the dimension of the space of algebraic solutions and the ocurrence of explicit non algebraic logarithmic solutions. This behaviour is related to the interactions among the different primary components of . The general pattern is given in [DMM06, Theorem 6.8].
3. Main definitions and results
3.1. What is a binomial module?.
We give now the main general definition of a binomial module, which contains the previous hypergeometric systems as special cases. The building blocks in the world of binomial modules are the hypergeometric systems (1.1) for different .
We consider an integer matrix such that the cone generated by the columns of contains no lines, all of the are nonzero, and . The matrix induces a grading of the polynomial ring by setting . An ideal of is graded if it is generated by elements that are homogeneous for the grading.
A binomial ideal is an ideal generated by binomials , where are column vectors and . A binomial ideal is graded precisely when it is generated by binomials each of which satisfies either or .
The Weyl algebra of linear partial differential operators is also graded by additionally setting . For each , the th Euler operator is defined as
]]>Given a vector , we write for the sequence Note that these operators are homogeneous of degree . A (local) holomorphic function is annihilated by precisely when is homogeneous of homogeneity , that is
where , with all close to .
Definition 3.1. Given an graded binomial ideal , we denote by the left ideal in the Weyl algebra . The binomial module associated to is the quotient .
Thus, a binomial Dmodule is a quotient by a left ideal generated by an graded binomial ideal with constant coefficients plus Euler operators of order associated to the row span of , which prescribe homogeneity infinitesimally. As we pointed out in the classical case, the binomial differential operators annihilate a (multivariate Puiseux) series if and only if the coefficients of the series satisfy (special) linear recurrences.
When equals the toric ideal , the associated binomial module is just the hypergeometric system (1.1). On the other side, when is generated by binomials , , with linearly independent over (in this case is called a lattice basis ideal), the corresponding binomial module is a Horn system (in binomial version).
3.2. Main questions and answers about binomial modules.
We summarize the main questions concerning binomial modules:
 For which parameters does the space of local holomorphic solutions around a nonsingular point of a binomial Horn system have finite dimension as a complex vector space? ]]> What is a combinatorial formula for the minimum holonomic rank, over all possible choices of parameters?
 Which parameters are generic, in the sense that the minimum dimension is attained?
 How do (the supports of) series solutions centered at the origin of a binomial Horn system look, combinatorially?
 When is a holonomic module?
 When is a regular holonomic module?
We now summarize the main answers to these questions. We refer to [DMM06] for precise definitions, statements and proofs. We explicitly describe and classify all primary components of (in particular, all monomials that are present), their multiplicities, their behaviour with respect to the grading (which splits the components into toral (when they don't have any further homogeneities, as in the component at infinity of the Horn system, and Andean (when they admit a new homogeneity as in the component at infinity in the Appell system), and their holonomic rank. Moreover, we explicitly define three subspace arrangements associated to the Andean components (Andean arrangement), to the pairwise intersections of two components (primary arrangement) and to the rankjumping parameters where the holonomic rank increases (jump arrangement), as the Zariski closure of parameters for which the corresponding piece in certain local cohomology modules is non zero, which account for non generic behaviour of the complex parameter .
Concretely,
 The dimension is finite exactly for not in the Andean arrangement.
 Given , denote by the submatrix of which consists only of the columns of for and denote by a sublattice of . We refer to the Introduction of [DMM06] for complete definitions and explanations, in particular for the definition of the multiplicities and the notion of toral associated sublattice. The generic (minimum) holonomic rank is , the sum being over all toral associated sublattices with , where is the volume of the convex hull of and the origin, normalized so a lattice simplex in has volume . ]]> The minimum rank is attained precisely when lies outside of an explicit affine subspace arrangement determined by certain local cohomology modules, containing the Andean arrangement.
 When the Horn system is regular holonomic and is general, there are linearly independent solutions supported on (translates of) the bounded classes (for the definition, see [DMM06, Subsection 1.6]), with hypergeometric recursions determining the coefficients.
 Only many Gamma series solutions have full support, where we denote the index of in its saturation .
 Holonomicity is equivalent to finite dimension of the (local holomorphic) solutions spaces.
 Holonomicity is equivalent to regular holonomicity when is standard graded—i.e., the rowspan of contains the vector . Conversely, if there exists a parameter for which is regular holonomic, then is graded.
We briefly discuss the main tools in the proofs of the previous results. As we said in the introduction, these results are generalizations of the beautiful theory of hypergeometric systems developed by Gelfand, Kapranov, and Zelevinsky in [GKZ89, GKZ90], see also [Ado94, SST00] for further developments of the basic theory. They defined the systems in terms of binomials and Euler operators, proved that they are holonomic, computed the holonomic rank for generic parameters, described the singular locus (in terms of sparse discriminants), and constructed bases of solutions for generic parameters in terms of Gamma series.
We also use the results in [DMM08], which give a precise description of the combinatorial and commutative algebra of primary components of binomial ideals in semigroup rings, based on the basic work of Eisenbud and Sturmfels on binomial ideals in [ES96]. In particular we have characterized the monomials in each primary component.
In characteristic zero, the primary decomposition of arbitrary binomial ideals is controled by the geometry and combinatorics of lattice point congruences, which also governs the module theoretic properties of binomial modules. In order to make this translation, we need to extend the results on EulerKoszul homology from [MMW05], which functorially translate the commutative algebra of graded primary decomposition into the module setting. The EulerKozsul homology functor is used to pull apart the primary components of binomial ideals, thereby isolating the contribution of each to the solutions of the corresponding binomial module. This allows us to integrate results from [GKZ89, GKZ90, Ado94, Hot91, SchW08, DMS05]. In particular, we prove that for parameters outside the (finite) primary subspace arrangement, the binomial module decomposes as a direct sum over the toral primary components of .
]]> 4. Further examples: binomials and the shape of the solutionsThe basic blocks in all these descriptions, besides the prime binomial ideals (i.e. the toric ideals), are the zero dimensional binomial ideals. We illustrate in this section how the information is assembled to give bases of series solutions for Horn binomial modules. We refer the reader to [DMM06, Section t] for general definitions and results.
4.1. Finding polynomial solutions of square binomial ideals.
Let be a square matrix such that each column has at least one negative and one positive entry. Such an defines an infinite graph with vertices in the points and edges for all pairs such that is a column of . Consider the system of constant coefficient binomial differential operators , where the operator is associated to the th column of as follows:
The number of bounded connected components of the graph corresponds precisely to the irreducible supports of polynomial solutions to . This number equals the dimension of the local quotient by at the origin. Moreover, let be the set of all the monomials occuring in a polynomial solution of . Then, equals precisely all monomials in the complement of a monomial ideal , which is the biggest monomial ideal (in the commutative ring generated by the partial derivatives) which lies inside .
Consider for instance the following matrix
The system is defined by the operators

The coefficients of these polynomial solutions are prescribed (up to constant) by the recurrence imposed by the binomial operators of the system. For instance, the bounded connected component gives rise to the quatrinomial solution of ;

The set has monomials, precisely those monomials which do not belong to the monomial ideal , which is the biggest monomial ideal inside . Note that in this case, the matrix is invertible, so its "Gale dual" matrix would be empty.
4.2. Description of solutions to complete intersection Horn binomial Dmodules.
Consider the matrices:
We can associate a Horn binomial module to the columns of and a parameter vector :
]]>where denote the binomials:
Note that , so there is only one associated primary component coming from this decomposition, with associated prime . Solutions to the binomial module associated to can be constructed from the polynomial solutions to and solutions to the hypergeometric system associated to the matrix , which, up to a rescaling in the homogeneities, equals the hypergeometric system associated to the matrix , whose columns span .
For instance, consider the quatrinomial in (4.1). Let be any hypergeometric function of degree . Then, the following function is a solution of :
Note that all these solutions depend polynomially on .
[Ado94] Alan Adolphson, Hypergeometric functions and rings generated by monomials, Duke Math. J. 73 (1994), no. 2, 269290. [ Links ]
[App1880] Paul Appell, Sur les séries hypergéometriques de deux variables et sur des équations différentielles linéaires aux dérivées partielles, Comptes Rendus 90 (1880), 296298. [ Links ]
[Bir27] R. Birkeland, Über die Auflösung algebraischer Gleichungen durch hypergeometrische Funktionen. Mathematische Zeitschrift 26 (1927), 565578. [ Links ]
[Bjö93] JanErik Björk, Analytic modules and applications, Mathematics and its Applications, vol. 247, Kluwer Academic Publishers Group, Dordrecht, 1993. [ Links ]
[Bjö79] Jan Erik Björk, Rings of differential operators, NorthHolland Mathematical Library, vol. 21, NorthHolland Publishing Co., Amsterdam, 1979. [ Links ]
[CDD99] E. Cattani, C. D'Andrea, and A. Dickenstein, The hypergeometric system associated with a monomial curve, Duke Math. J. 99 (1999), 179207. [ Links ]
[CD07] Eduardo Cattani and Alicia Dickenstein, Counting solutions to binomial complete intersections, J. of Complexity 23, Issue 1 (2007), 82107 [ Links ]
[DMM06] Alicia Dickenstein, Laura Matusevich, and Ezra Miller, Binomial modules, Preprint. Available at arXiv:math/0610353. [ Links ]
[DMM08] Alicia Dickenstein, Laura Matusevich, and Ezra Miller. Combinatorics of binomial primary decomposition, Preprint. Available at: arXiv:0803.3846. [ Links ]
[DMS05] Alicia Dickenstein, Laura Felicia Matusevich, and Timur Sadykov, Bivariate hypergeometric modules, Adv. Math. 196 (2005), no. 1, 78123. [ Links ]
[DSt02] Alicia Dickenstein and Bernd Sturmfels, Elimination in codimension two, J. Symbolic Comput. 34(2) (2002), 119135. [ Links ]
[DS07] Alicia Dickenstein and Timur Sadykov, Bases in the solution space of the Mellin system, Mat. Sb. 198:9 (2007), 5980. [ Links ]
[ES96] David Eisenbud and Bernd Sturmfels, Binomial ideals, Duke Math. J. 84 (1996), no. 1, 145. [ Links ]
[Erd50] Arthur Erdélyi, Hypergeometric functions of two variables, Acta Math. 83 (1950), 131164. [ Links ]
[Eu1748] Leonhard Euler, Introductio in Analysis Infinitorum, volume 1. Laussane, 1748. [ Links ]
[Gau1812] Carl Friedrich Gauss, Disquisitiones generales circa seriem infinitam. Thesis, Göttingen, 1812. In Ges. Werke, Göttingen, 1866. [ Links ]
[GGR92] I. M. Gel'fand, M. I. Graev, and V. S. Retakh, General hypergeometric systems of equations and series of hypergeometric type, Uspekhi Mat. Nauk 47 (1992), no. 4(286), 382, 235. [ Links ]
[GGZ87] I. M. Gel'fand, M. I. Graev, and A. V. Zelevinskiy, Holonomic systems of equations and series of hypergeometric type, Dokl. Akad. Nauk SSSR 295 (1987), no. 1, 1419. [ Links ]
[GKZ89] I. M. Gel'fand, A. V. Zelevinskiy, and M. M. Kapranov, Hypergeometric functions and toral manifolds. Functional Analysis and its Applications 23 (1989), 94106. [ Links ]
[GKZ90] I. M. Gel'fand, M. Kapranov, and A. Zelevinski, Generalized Euler integrals and hypergeometric functions. Advances in Mathematics 84 (1990), 255271. [ Links ]
[Hor1889] J. Horn, Über die konvergenz der hypergeometrischen Reihen zweier und dreier Veränderlichen, Math. Ann. 34 (1889), 544600. [ Links ]
[Hor31] J. Horn, Hypergeometrische Funktionen zweier Veränderlichen, Math. Ann. 105 (1931), no. 1, 381407. [ Links ]
[Hot91] Ryoshi Hotta, Equivariant modules, math.RT/980502. In P. Torasso, editor, Proceedings of ICPAM Spring School in Wuhan, Travaux en Cours. Paris, 1991. [ Links ]
[Kum1836] Ernst Eduard Kummer, Über die hypergeometrische Reihe , J. für Math. 15 (1836). [ Links ]
[MMW05] Laura Felicia Matusevich, Ezra Miller, and Uli Walther, Homological methods for hypergeometric families, J. Amer. Math. Soc. 18 (2005), no. 4, 919941. [ Links ]
[Ma37] K. Mayr, Über die Auflösung algebraischer Gleichungssysteme durch hypergeometrische Funktionen, Monatshefte für Mathematik und Physik 45 (1937), 280313. [ Links ]
[Mel21] Hjalmar Mellin, Résolution de l'équation algébrique générale à l'aide de la fonction , C.R. Acad. Sc. 172 (1921), 658661. [ Links ]
[MS05] Ezra Miller and Bernd Sturmfels, Combinatorial commutative algebra, Graduate Texts in Mathematics, vol. 227, SpringerVerlag, New York, 2005. [ Links ]
[PT04] M. Passare and A. Tsikh, Algebraic equations and hypergeometric series, in: "The Legacy of N.H. Abel", SpringerVerlag (2004), 563582. [ Links ]
[Rie1857] Georg Friedrich Bernhard Riemann, Beiträge zur Theorie der durch Gauss'sche Reihe darstellbaren Functionen. K. Gess. Wiss. Göttingen, 7, 124 (1857) [ Links ]
[Sad02] Timur Sadykov, On the Horn system of partial differential equations and series of hypergeometric type, Math. Scand. 91 (2002), no. 1, 127149. [ Links ]
[SchW08] Mathias Schulze and Uli Walther, Irregularity of hypergeometric systems via slopes along coordinate subspaces, Duke Math. J. 142,3 (2008), 465509 [ Links ]
[SST00] Mutsumi Saito, Bernd Sturmfels, and Nobuki Takayama, Gröbner Deformations of Hypergeometric Differential Equations, SpringerVerlag, Berlin, 2000. [ Links ]
[Stu96] Bernd Sturmfels, Gröbner bases and convex polytopes, American Mathematical Society, Providence, RI, 1996. [ Links ]
[Stu00] Bernd Sturmfels, Solving algebraic equations in terms of Ahypergeometric series, in: Formal power series and algebraic combinatorics (Minneapolis, MN, 1996), Discrete Math. 210 (2000), no. 13, 171181. [ Links ]
Alicia Dickenstein
Departamento de Matemática
FCEN, Universidad de Buenos Aires
C1428EGA Buenos Aires, Argentina.
alidick@dm.uba.ar
Aceptado: 26 de noviembre de 2008 ]]>