Open Access

ARTICLE

Adaptive Support Domain Implementation on the Moving Least Squares Approximation for Mfree Methods Applied on Elliptic and Parabolic PDE Problems Using Strong-Form Description

G. C. Bourantas¹, E. D. Skouras^2,3,4, G. C. Nikiforidis¹

1 Department of Medical Physics, School of Medicine, University of Patras, GR 26500, Rion, Greece.
2 Department of Chemical Engineering, University of Patras, GR 26500, Rion, Greece.
3 Institute of Chemical Engineering and High Temperature Chemical Processes - Foundation for Research and Technology, P.O. Box 1414, GR-26504, Patras, Greece.
4 Correspondence to Eugene.Skouras@iceht.forth.gr

Computer Modeling in Engineering & Sciences 2009, 43(1), 1-26. https://doi.org/10.3970/cmes.2009.043.001

Abstract

The extent of application of meshfree methods based on point collocation (PC) techniques with adaptive support domain for strong form Partial Differential Equations (PDE) is investigated. The basis functions are constructed using the Moving Least Square (MLS) approximation. The weak-form description of PDEs is used in most MLS methods to circumvent problems related to the increased level of resolution necessary near natural (Neumann) boundary conditions (BCs), dislocations, or regions of steep gradients. Alternatively, one can adopt Radial Basis Function (RBF) approximation on the strong-form of PDEs using meshless PC methods, due to the delta function behavior (exact solution on nodes). The present approach is one of the few successful attempts of using MLS approximation [Atluri, Liu, and Han (2006), Han, Liu, Rajendran and Atluri (2006), Atluri and Liu (2006)] instead of RBF approximation for the meshless PC method using strong-form description. To increase the accuracy of the MLS interpolation method and its robustness in problems with natural BCs, a suitable support domain should be chosen in order to ensure an optimized area of coverage for interpolation. To this end, the basis functions are constructed using two different approaches, pertinent to the dimension of the support domain. On one hand, a compact form for the support domain is retained by keeping its radius constant. On the other hand, one can control the number of neighboring nodes as the support domain of each point. The results show that some inaccuracies are present near the boundaries using the first approach, due to the limited number of nodes belonging to the support domain, which results in failed matrix inversion. Instead, the second approach offers capability for fully matrix inversion under many (if not all) circumstances, resulting in basis functions of increased accuracy and robustness. This PC method, applied along with an intelligent adaptive refinement, is demonstrated for elliptic and for parabolic PDEs, related to many flow and mass transfer problems.

---

Keywords

---