Subdivision surfaces with isogeometric analysis adapted refinement weights

TL;DR

This paper introduces optimized subdivision weights for isogeometric analysis on subdivision surfaces, significantly reducing discretization errors around extraordinary vertices and improving convergence constants.

Contribution

It develops a method to optimize subdivision weights to minimize finite element discretization errors, enhancing surface quality and analysis accuracy.

Findings

Reduced discretization errors by over 50% in energy and L2 norms.

Improved convergence constants without changing convergence rates.

Optimized weights adapt to local shape features around extraordinary vertices.

Abstract

Subdivision surfaces provide an elegant isogeometric analysis framework for geometric design and analysis of partial differential equations defined on surfaces. They are already a standard in high-end computer animation and graphics and are becoming available in a number of geometric modelling systems for engineering design. The subdivision refinement rules are usually adapted from knot insertion rules for splines. The quadrilateral Catmull-Clark scheme considered in this work is equivalent to cubic B-splines away from extraordinary, or irregular, vertices with other than four adjacent elements. Around extraordinary vertices the surface consists of a nested sequence of smooth spline patches which join $C^1$ continuously at the point itself. As known from geometric design literature, the subdivision weights can be optimised so that the surface quality is improved by minimising short-wavelength surface oscillations around extraordinary vertices. We use the related techniques to determine weights that minimise finite element discretisation errors as measured in the thin-shell energy norm. The optimisation problem is formulated over a characteristic domain and the errors in approximating cup- and saddle-like quadratic shapes obtained from eigenanalysis of the subdivision matrix are minimised. In finite element analysis the optimised subdivision weights for either cup- or saddle-like shapes are chosen depending on the shape of the solution field around an extraordinary vertex. As our computations confirm, the optimised subdivision weights yield a reduction of $50\%$ and more in discretisation errors in the energy and $L_2$ norms. Although, as to be expected, the convergence rates are the same as for the classical Catmull-Clark weights, the convergence constants are improved.