which in conjunction with Equation (2) indicates that the i (x) are bounded between zero and
unity and satisfy the convex hull property. The non-negative condition leads to the variation
diminishing property and to positive-definite mass matrices [12]. Convex approximation
schemes are not prone to the Runge phenomena [21], which occurs when using higher-order
one-dimensional Lagrange interpolation on uniform grids. In addition, optimal conditioning
can be established for non-negative basis functions [22–24]. The adoption of NURBS-based
convex basis functions to ensure geometric exactness in finite element analysis has been
recently introduced by Hughes and co-workers [25]. MLS approximants [5], which are
widely used in meshfree Galerkin methods, are not convex approximants since MLS basis
functions can be negative.