A new method for constructing Sylvester-type resultant matrices for multivariate elimination is proposed. Unlike sparse resultant constructions discussed recently in the literature or the Macaulay resultant construction, the proposed method does not explicitly use the support of a polynomial system in the construction. Instead, a multiplier set for each polynomial is obtained from the Dixon formulation using an arbitrary term for the construction. As shown in [KS96], the generalized Dixon resultant formulation implicitly exploits the sparse structure of the polynomial system. As a result, the proposed construction for Sylvester-type resultant matrices is sparse in the sense that the matrix size is determined by the support structure of the polynomial system, instead of the total degree of the polynomial system.The proposed construction is a generalization of a related construction proposed by the authors in which the monomial 1 is used [CK00b]. It is shown that any monomial (inside or outside the support of a polynomial in the polynomial system) can be used instead insofar as that monomial does not vanish on any of the common zeros of the polynomial system. For generic unmixed polynomial systems (in which every polynomial in the polynomial system has the same support, i.e., the same set of terms), it is shown that the choice of a monomial does not affect the matrix size insofar as it is in the support.The main advantage of the proposed construction is for mixed polynomial systems. Supports of a mixed polynomial system can be translated so as to have a maximal overlap, and a term is selected from the overlapped subset of translated supports. Determining an appropriate translation vector for each support and a term from overlapped support can be formulated as an optimization problem. It is shown that under certain conditions on the supports of polynomials in a mixed polynomial system, a monomial can be selected leading to a Dixon multiplier matrix of the smallest size, thus implying that the projection operator computed using the proposed construction is either the resultant or has an extraneous factor of minimal degree.The proposed construction is compared theoretically and empirically, on a number of examples, with other methods for generating Sylvester-type resultant matrices.

No abstract

No abstract

No abstract

Different matrix based resultant formulations use the support of the polynomials in a polynomial system in various ways for setting up resultant matrices for computing resultants. Every formulation suffers, however, from the fact that for most polynomial systems, the output is not a resultant, but rather a nontrivial multiple of the resultant, called a projection operator . It is shown that for the Dixon-based resultant methods, the degree of the projection operator of unmixed polynomial systems is determined by the support hull of the support of the polynomial system. This is similar to the property that the Newton polytope of a support determines the degree of the resultant for toric zeros.The support hull of a given support is similar to its convex hull (Newton polytope) except that instead of the Euclidean distance, the support hull is defined using relative quadrant (octant) position of points. The concept of a support hull interior point with respect to a support is defined. It is shown that for unmixed polynomial systems, generic inclusion of terms corresponding to support hull interior points does not change the size of the Dixon matrix (hence, the degree of the projection operator). The support hull of a support is the closure of the support with respect to support-interior points.The above results are shown to hold both for the generalized Dixon formulation as well as for Sylvester-type Dixon dialytic matrices constructed using the Dixon formulation.It is proved that for an unmixed polynomial system, the size of the Dixon matrix is less than or equal to the Minkowski sum of the alternating sums of the successive projections of the support of the polynomial system. This is a refinement of the result in Kapur and Saxena 1996 about the size of the Dixon matrix of a polynomial system, where it was shown that for the unmixed polynomial system, the size of the Dixon matrix is less than or equal to the Minkowski sum of the successive projection of the support.Many other combinatorial properties of the size of the Dixon matrix and the structure of the Dixon polynomial of a given polynomial system are related to the support hull of the polynomial system and their projections along different dimensions.This research is supported in part by NSF grant nos. CCR-0203051, CDA-9503064 and a grant from the Computer Science Research Institute at Sandia National Labs.