An analysis of cubic approximation schemes for conic sections (Q1815864)

The author develops algorithms to approximate a piece of a conic section by piecewise cubic polynomials. His approach is based on the fact that a conic section is given as a rational quadratic Bézier curve determined by three planar control points \(p_0\), \(p_1\), \(p_2\) and a positive weight \(w\), and thus also possesses an implicit representation as the zero set of a bivariate function involving the weight \(w\) and the barycentric coordinates of a point with respect to the triangle formed by \(p_0\), \(p_1\), \(p_2\). As an approximation to the conic, cubic Bézier polynomials with control points \(p_0\), \((1-\lambda)p_0+\lambda p_1\), \((1-\lambda)p_2+\lambda p_1\), \(p_2\) are considered, where \(\lambda\) is a positive parameter. The different approximation schemes are then classified by different functions \(\lambda(w)\), describing how the parameter of the approximating cubic is determined by the weight of the conic. It is then possible to split the conic section into two pieces in such a way that two cubics which approximate the separate pieces are guaranteed to form a piecewise cubic approximant with \(G^2\) smoothness overall. This approximation procedure can be iterated and an error estimate for the Hausdorff distance between the conic section and the resulting piecewise cubic approximants is provided. It is shown that depending on the behaviour of the function \(\lambda(w)\) at \(w=1\), the approximation is of second, fourth, or sixth order in the recursion depth. As an example, apart from some known cubic Hermite interpolants, a specific choice of \(\lambda(w)\) is presented which produces a new sixth-order approximation which is even \(G^4\) overall.