Polynomial approximation of rational Bézier curves with constraints (Q411526)

Given are weights \(\omega_i\in\mathbb{R}_+\) and coefficients \(r_i\) of a rational function of degree \(n\) with Bézier representation \[ R_n(t)=\sum_{i=0}^n \omega_i r_iB_i^n(t)/\sum_{i=0}^n \omega_i B_i^n(t),\quad t\in[0,1] \] (the \(B_i^n\) are Bézier basis functions). The problem is to compute the coefficients \(p_i\) of an approximating polynomial \(P_m(t)=\sum_{i=0}^m p_iB_i^m(t)\) of degree \(m\). The approximation is in a least squares sense and under the end-point constraints that \(R_n^{(i)}(0)=P_m^{(i)}(0)\), \(i=0,\ldots,k-1\) and \(R_n^{(j)}(1)=P_m^{(j)}(1)\), \(j=0,\ldots,l-1\) with \(k+l\leq m\). These constraints fix the coefficients \(p_0,\ldots,p_{k-1},p_{m-l+1},\ldots,p_m\) (explicit expressions are given and can be computed efficiently) while the remaining coefficients \(p_i\) are available to minimize the least squares error. The formulas are derived using a dual (i.e., a biorthogonal) set of Bézier polynomials which results in \(p_i\) being expressed as inner products. After working these out, only simple integrals of rational functions are needed which are evaluated with a Clenshaw-Curtis quadrature formula [cf. \textit{C. W. Clenshaw} and \textit{A. R. Curtis}, Numer. Math. 2, 197--205 (1960; Zbl 0093.14006)]. The method is illustrated by approximating several parametric curves in the plane.