Two Condensed Macro-Elements with Full Approximation Power.pdfVIP

Two Condensed Macro-Elements with Full Approximation Power Peter Alfeld 1), Larry L. Schumaker 2) , and Tatyana Sorokina 3) Abstract. A variant of the classical C1 Powell-Sabin-12 macro-element is defined that has the same approximation power, but has fewer degrees of freedom and only requires vertex data. The same idea is then applied to a related C1 trivariate macro-element. §1. Introduction Suppose △ is a triangulation of a domain Ω ⊂ IR2 or a tetrahedral partition of 3 a domain in IR , and that △R is a refinement of △ obtained by applying a given splitting procedure to each triangle or tetrahedron in △. Suppose in addition that S (△R) is a polynomial spline space defined on △R such that every spline s ∈ S(△R ) is uniquely determined by the values of s and its gradients at some collection of points in Ω. Then (cf. [4]) S (△R) is called a macro-element space provided that for each triangle (tetrahedron) in △, s|T is uniquely determined by the data at points in T . There are many examples of both bivariate and trivariate macro-element spaces. For complete details, see [4]. Macro-element spaces are particularly im- portant in applications since they generally have stable local bases and have full approximation power. This means that they approximate sufficiently smooth func- tions to O(|△|d+1), where d is the degree of the splines, and |△| is the mesh size of △, i.e., the length of the longest edge in △. In applications, and in particular for scattered data fitting and the solution of boundary-value problems by the finite-element method, it is advantageous to work with macro-element spaces that produce a given order of approximation while involving the least numbe