Dual Gauss--Legendre polynomials

Technical Summary: Dual Gauss–Legendre Polynomials

Problem Statement
The paper addresses the need for dual polynomial bases associated with the recently introduced Gauss–Legendre (GL) polynomials and curves. While GL polynomials and curves possess attractive properties for computer-aided geometric design (CAGD) and numerical analysis—such as excellent shape control and close adherence to control polygons—they do not lie within the convex hull of their control points. To fully leverage these structures in approximation theory, specifically for solving least-squares approximation problems and degree reduction, explicit representations of their dual bases are required. The authors aim to define and investigate these dual polynomials with respect to the Jacobi inner product, extending the utility of duality concepts previously established for Bernstein polynomials to the GL context.

Methodology
The authors employ the theory of dual bases within the space of univariate polynomials $\Pi_n$. The methodology relies on the following steps:

1. Definition of Primal Bases: The paper utilizes two families of polynomials: $G^n_i(t)$ (defined via integrals of Legendre polynomials) and $F^n_i(t)$ (defined as differences of $G^n_i$). These form bases for $\Pi_n$.
2. Duality Framework: The dual basis $D_n = \{d_0, \dots, d_n\}$ for a basis $B_n = \{b_0, \dots, b_n\}$ is defined relative to the Jacobi inner product $\langle f, g \rangle_{\alpha,\beta} = \int_{-1}^1 (1-x)^\alpha(1+x)^\beta f(x)g(x) dx$. The defining condition is $\langle b_i, d_j \rangle_{\alpha,\beta} = \delta_{ij}$.
3. Derivation via Orthogonal Expansion: Using Theorem 2.1, the authors express the Jacobi polynomials (the orthogonal basis) in terms of the GL bases ($G^n_i$ and $F^n_i$). By inverting this relationship, they derive explicit formulas for the dual polynomials.
4. Kernel Utilization: The Christoffel–Darboux kernel is utilized to derive closed-form expressions for the dual polynomials, particularly for the $G^n_j$ family, by differentiating the kernel with respect to specific nodes (zeros of Legendre polynomials).

Key Contributions and Results
The paper presents the following specific mathematical results:

• Explicit Representation of Dual $G^n_j$ Polynomials:

  • Theorem 3.3: Provides an expansion of the dual polynomials $dG^n_j(t; \alpha, \beta)$ in the Jacobi polynomial basis. The coefficients depend on the Jacobi weights, the zeros of the Legendre polynomial ($\tau_i$), and the values of Jacobi polynomials at these zeros.
  • Theorem 3.5: Derives a closed-form expression for $dG^n_j(t; \alpha, \beta)$ for $j \geq 1$ involving the Christoffel–Darboux kernel $K^{(\alpha,\beta)}_n$. This formula allows for evaluation in $O(n)$ time using the Clenshaw algorithm.
  • Special Case: The paper notes simplifications occur when $\alpha = \beta = 0$ (the Legendre case).

• Explicit Representation of Dual GL Polynomials ($F^n_j$):

  • Theorem 3.6: Provides the expansion of dual GL polynomials $dF^n_j(t; \alpha, \beta)$ in the Jacobi basis. The coefficients involve a summation of weighted Jacobi polynomial evaluations.
  • Theorem 3.7: Establishes a recurrence-like relationship between the dual GL polynomials and the dual $G^n$ polynomials: $dF^n_j = dF^n_{j-1} + dG^n_j$. This implies a computational complexity of $O(nj)$ for evaluating $dF^n_j$.

• Generalization to Lagrange Bases:

  • Theorem 4.2: The technique used to derive the dual GL polynomials is generalized to construct dual bases for arbitrary Lagrange polynomial bases with respect to any inner product. The result expresses the dual Lagrange polynomials in terms of the orthogonal polynomials associated with that inner product.

• Applications to Approximation:

  • Degree Reduction: The paper outlines how these dual bases solve the degree reduction problem for polynomials expressed in $G^n_i$ and $F^n_i$ bases. The optimal lower-degree approximation $p^*$ is constructed directly using the dual basis coefficients: $p^* = \sum \langle g, d_k \rangle b_k$.
  • GL Curve Reduction: The authors note that these results apply to the degree reduction of parametric GL curves, a common task in CAGD.

Significance and Claims
The paper claims that the derived dual polynomials provide a powerful tool for numerical analysis and CAGD, specifically for:

1. Solving Approximation Problems: They allow for the construction of optimal least-squares approximations without requiring the transformation to an orthogonal basis, which is often necessary when working in specific non-orthogonal bases like GL polynomials.
2. Computational Efficiency: The derived formulas (particularly Theorem 3.5) enable efficient evaluation of dual polynomials, potentially reducing computational complexity compared to naive methods.
3. Structural Insight: The work demonstrates that GL polynomials inherit structural properties similar to Bernstein polynomials, where dual bases have historically been effective for problems like degree reduction.

The authors remain modest regarding future implications, stating that while the results suggest a method for degree reduction of GL curves similar to that used for Bézier curves, deriving a fully efficient and numerically stable algorithm for this specific application requires further study of the new dual families' properties. They explicitly leave the development of such a detailed algorithm for future research.