Comparison of moving least squares and RBF+poly for interpolation and derivative approximation Articles
Overview
published in
- JOURNAL OF SCIENTIFIC COMPUTING Journal
publication date
- October 2019
start page
- 486
end page
- 512
volume
- 81
Digital Object Identifier (DOI)
full text
International Standard Serial Number (ISSN)
- 0885-7474
Electronic International Standard Serial Number (EISSN)
- 1573-7691
abstract
- The combination of polyharmonic splines (PHS) with high degree polynomials (PHS+poly) has recently opened new opportunities for radial basis function generated finite difference approximations. The PHS+poly formulation, which relies on a polynomial least squares fitting to enforce the local polynomial reproduction property, resembles somehow the so-called moving least squares (MLS) method. Although these two meshfree approaches are increasingly used nowadays, no direct comparison has been done yet. The present study aims to fill this gap, focusing on scattered data interpolation and derivative approximation. We first review the MLS approach and show that under some mild assumptions PHS+poly can be formulated analogously. Based on heuristic perspectives and numerical demonstrations, we then compare their performances in 1-D and 2-D. One key result is that, as previously found for PHS+poly, MLS can also overcome the edge oscillations (Runge's phenomenon) by simply increasing the stencil size for a fixed polynomial degree. This is, however, controlled by a weighted least squares fitting which fails for high polynomial degrees. Overall, PHS+poly is found to perform superior in terms of accuracy and robustness
Classification
subjects
- Mathematics
keywords
- moving least squares; radial basis functions; rbf-fd; polyharmonic; splines; polynomial augmentation; local polynomial reproduction; interpolation; derivative approximation; runge’s phenomenon; meshless