Advances in Applied Mathematics 28, 17–39 (2002) doi:10.1006/aama.2002.0765, available online at http://www.idealibrary.com on
Generalized Bernstein Polynomials and Symmetric Functions
Robert P. Boyer
Department of Mathematics and Computer Science, Drexel University, Philadelphia, Pennsylvania 19104 E-mail: [email protected]
and
Linda C. Thiel
Society for Industrial and Applied Mathematics, University City Science Center, Philadelphia, Pennsylvania 19104 E-mail: [email protected]
Received February 20, 2001; accepted February 26, 2001
We begin by classifying all solutions of two natural recurrences that Bernstein polynomials satisfy. The first scheme gives a natural characterization of Stancu poly- nomials. In Section 2, we identify the Bernstein polynomials as coefficients in the generating function for the elementary symmetric functions, which gives a new proof of total positivity for Bernstein polynomials, by identifying the required determi- nants as Schur functions. In the final section, we introduce a new class of approxi- mation polynomials based on the symplectic Schur functions. These polynomials are shown to agree with the polynomials introduced by Vaughan Jones in his work on subfactors and knots. We show that they have the same fundamental properties as the usual Bernstein polynomials: variation diminishing (whose proof uses symplec- tic characters), uniform convergence, and conditions for monotone convergence. 2002 Elsevier Science (USA) Key Words: Bernstein polynomials; total positivity; variation diminishing; Schur functions; symplectic group; group characters; Young tableau; unitary group; Jones polynomials.
INTRODUCTION
In this paper, we study Bernstein polynomials and other similar polyno- mial schemes that retain the shape of the approximating function and that
17 0196-8858/02 $35.00 2002 Elsevier Science (USA) All rights reserved. 18 boyer and thiel are computed with recurrence relations. These important properties make such approximations desirable for applications in computer-aided geomet- ric design. We find all polynomial solutions for the Bernstein polynomial recurrence which are exhausted by the polynomials found by Stancu [17] and further investigated by many others including Goldman [8, 9, 11] who used urn models found in probability theory to develop their properties. We can visualize this recurrence with the Pascal triangle. We show that the only polynomial approximations that satisfy this Pascal recurrence and give a convergent approximation are exactly the Bernstein approximations. In particular, the Stancu approximations never converge. We give a new proof of the classical result that the Bernstein approximations are shape preserving by describing them as coefficients for the elementary symmetric functions and using the theory of the symmetric Schur functions. We study another one-dimensional approximation scheme, which comes from another family of symmetric functions: the symplectic Schur polynomi- als. We show that the corresponding symplectic Bernstein polynomials are also variation diminishing by showing that the polynomials form a totally positive basis. The proof that the symplectic Bernstein approximations con- verge uses both the ergodic theorem and the asymptotic character formula, which are described in the Appendix. This example is related to the algebra that gives the knot polynomial invariant found by Jones. This recurrence can also be visualized by a variant of the Pascal triangle. Some of our motivations for this study are the connections of poly- nomial approximations with special classes of symmetric functions, which are attached to the classical groups. Several other classical approximation operators such as the Baskakov and Sz´asz–Mirakyan approximations are associated with representations of the unitary group.
1. PASCAL TRIANGLE RECURRENCE AND GENERALIZED BERNSTEIN POLYNOMIALS Recall that the Bernstein polynomials B n t = n tk 1 − t n−k, for k k 0 ≤ k ≤ n, give polynomial approximations that converge to a continuous function f on the unit interval 0 1. We express these approximations as a linear operator: n k n n k B f t = f B n t = f tk 1 − t n−k n n k k n k=0 k=0
Bernstein polynomials obey a useful recurrence relation that is important in n+1 n n computer-aided design applications: Bk t = 1 − t Bk t +tBk−1 t . generalized bernstein polynomials 19
Stancu [17] introduced a generalized of Bernstein polynomials by n t t +α ··· t +k−1α 1−t 1−t +α ··· 1−t +n−k−1α S n t = k k 1+α 1+2α ··· 1+n−1α