Asymptotically compatible discretization of multidimensional nonlocal diffusion models and approximation of nonlocal Green's functions. (5th April 2018)

• Record Type: Journal Article
• Authors: Du, Qiang; Tao, Yunzhe; Tian, Xiaochuan; Yang, Jiang
• Is Part Of: IMA journal of numerical analysis. Volume 39:Number 2(2019)
• Page Start: 607
• Page End: 625
• Publication Date: 2018-04-05
• Subjects: nonlocal models -- nonlocal diffusion -- peridynamics -- nonlocal gradient -- asymptotic compatibility -- quadrature collocation approximations -- nonlocal Green's function; Numerical analysis -- Periodicals; 519.405
• DOI: 10.1093/imanum/dry011
• Languages: English
• ISSNs: 0272-4979

Abstract

Abstract: Nonlocal diffusion equations and their numerical approximations have attracted much attention in the literature as nonlocal modeling becomes popular in various applications. This paper continues the study of robust discretization schemes for the numerical solution of nonlocal models. In particular, we present quadrature-based finite difference approximations of some linear nonlocal diffusion equations in multidimensions. These approximations are able to preserve various nice properties of the nonlocal continuum models such as the maximum principle and they are shown to be asymptotically compatible in the sense that as the nonlocality vanishes, the numerical solutions can give consistent local limits. The approximation errors are proved to be of optimal order in both nonlocal and asymptotically local settings. The numerical schemes involve a unique design of quadrature weights that reflect the multidimensional nature and require technical estimates on nonconventional divided differences for their numerical analysis. We also study numerical approximations of nonlocal Green's functions associated with nonlocal models. Unlike their local counterparts, nonlocal Green's functions might become singular measures that are not well defined pointwise. We demonstrate how to combine a splitting technique with the asymptotically compatible schemes to provide effective numerical approximations of these singular measures.