Generalized Characteristic Polynomials and Gaussian Cubature Rules

Yuan XuSIAM J. Matrix Anal. Appl 2015 · DOI

TL;DR

This paper introduces a novel construction of multivariate orthogonal polynomials derived from generalized characteristic polynomials of specific matrix families. These new polynomials possess a maximal number of real common zeros, which are crucial for constructing Gaussian cubature rules, a powerful technique for accurate numerical integration in higher dimensions.

What Problem Does This Paper Address?

Accurate numerical integration is a fundamental challenge in many scientific and engineering fields. Gaussian cubature rules offer a highly efficient way to approximate integrals, but their existence in multiple dimensions is notoriously rare. This rarity stems from a stringent requirement: the underlying orthogonal polynomials must have a maximal number of real, distinct common zeros. While some families of such polynomials are known, finding new ones is a significant research problem. This paper addresses this by exploring a connection between generalized characteristic polynomials of certain matrices and multivariate orthogonal polynomials, aiming to generate new families of these polynomials and, consequently, new Gaussian cubature rules.

Methods

The core of the paper lies in establishing a link between generalized characteristic polynomials and orthogonal polynomials. The authors begin by defining generalized characteristic polynomials for matrices, extending the concept from single-variable polynomials. They then investigate these polynomials for specific classes of matrices, particularly those that are "banded Toeplitz" and "centrohermitian."

A key technique is the use of three-term recurrence relations. For univariate orthogonal polynomials, satisfying such a relation (and certain conditions on its coefficients) guarantees orthogonality via Favard's theorem. The paper extends this idea to multivariate polynomials, where the recurrence relations involve matrix coefficients.

The authors then focus on a specific family of centrohermitian matrices, which are perturbations of a known matrix (related to Chebyshev polynomials). By analyzing the generalized characteristic polynomials of these perturbed matrices, they demonstrate that these polynomials satisfy the necessary three-term relations for orthogonality. They also investigate the nature of their common zeros, a critical step for Gaussian cubature.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Bridging Matrix Theory and Orthogonal Polynomials: The paper establishes a concrete link between the algebraic properties of generalized characteristic polynomials of specific matrices and the analytical properties of multivariate orthogonal polynomials. This connection is not merely theoretical; it provides a constructive method for generating new orthogonal polynomials.
  • Maximal Real Zeros for Cubature: The central insight is that by carefully choosing the matrix family (banded Toeplitz and centrohermitian), the resulting generalized characteristic polynomials can be made to have a maximal number of real, distinct common zeros. This is the "holy grail" for constructing Gaussian cubature rules.
  • A New Family of Orthogonal Polynomials: The paper introduces a new family of orthogonal polynomials, parameterized by complex numbers $a$ and $c$. This family is a perturbation of known Chebyshev polynomials, offering a generalization and expanding the toolkit for multivariate orthogonal polynomial research.
  • Guaranteed Gaussian Cubature: The existence of these new orthogonal polynomials with real common zeros directly implies the existence of Gaussian cubature rules for a corresponding family of integrals. This is a significant advancement, as such rules are rare.

Key Results

  • The paper shows that generalized characteristic polynomials of a specific family of centrohermitian matrices, parameterized by $a$ and $c$, are orthogonal polynomials.
  • Under certain conditions on $a$ and $c$ (specifically, $c(\bar{c} - a) \in \mathbb{R}$ and $|c| \ge 2|c-a|$), these polynomials possess a maximal number of real, simple common zeros.
  • This guarantees the existence of Gaussian cubature rules for integrals against a corresponding family of positive Borel measures.
  • The paper provides explicit formulas for the three-term recurrence relations satisfied by these new polynomials.
  • Numerical examples (Figure 2) illustrate the distribution of common zeros for different parameter values, showing how they remain real and within a specific region under the stated conditions, and how they deviate when these conditions are not met.

Limitations and Future Work

  • Explicit Measure Formula: While the paper guarantees the existence of a Gaussian cubature rule and its associated measure, it notes that the explicit formula for this measure is not known when $a \neq c$. This is a significant area for future research.
  • Generalization of Conditions: The conditions on $a$ and $c$ for the existence of real common zeros are specific. Exploring whether these conditions can be relaxed or generalized would be valuable.
  • Higher Dimensions: The paper focuses primarily on the case of two variables for clarity. Extending the results and constructions to higher dimensions would be a natural next step.
  • Conjectures: The paper mentions several conjectures regarding generalized characteristic polynomials and their properties. Further investigation into these conjectures, particularly those related to the orthogonality of polynomials from banded matrices, would be beneficial.

Summary updated

BibTeX
@article{xu2015generalized,
  author = {Yuan Xu},
  title = {{Generalized Characteristic Polynomials and Gaussian Cubature Rules}},
  journal = {SIAM J. Matrix Anal. Appl.},
  year = {2015},
  doi = {10.1137/140972810},
}

Loading related content...

Figure