Analysis of (shifted) piecewise quadratic polynomial collocation for nonlocal diffusion model

TL;DR

This paper introduces a modified shifted-symmetric quadratic polynomial collocation method for nonlocal diffusion models, ensuring symmetric positive definite systems and satisfying the discrete maximum principle, with detailed convergence analysis and numerical verification.

Contribution

The paper develops a shifted-symmetric collocation scheme that guarantees symmetric positive definite systems and discrete maximum principle satisfaction for nonlocal diffusion models, with comprehensive convergence proofs.

Findings

Global error is rac{h^{ ext{min}\left\u00{2,1+eta ight"}}

Scheme is asymptotically compatible with error rac{h^{ ext{min}\left\u00{2,2eta ight"}}

Numerical experiments verify theoretical convergence rates

Abstract

The piecewise quadratic polynomial collocation is used to approximate the nonlocal model, which generally obtain the {\em nonsymmetric indefinite system} [Chen et al., IMA J. Numer. Anal., (2021)]. In this case, the discrete maximum principle is not satisfied, which might be trickier for the stability analysis of the high-order numerical schemes [D'Elia et al., Acta Numer., (2020); Leng et al., SIAM J. Numer. Anal., (2021)]. Here, we present the modified (shifted-symmetric) piecewise quadratic polynomial collocation for solving the linear nonlocal diffusion model, which has the {\em symmetric positive definite system} and satisfies the discrete maximum principle. Using Faulhaber's formula and Riemann zeta function, the perturbation error for symmetric positive definite system and nonsymmetric indefinite systems are given. Then the detailed proof of the convergence analysis for the nonlocal models with the general horizon parameter $\delta=\mathcal{O}\left(h^\beta\right)$, $\beta\geq0$ are provided. More concretely, the global error is $\mathcal{O}\left(h^{\min\left\{2,1+\beta\right\}}\right)$ if $\delta$ is not set as a grid point, but it shall recover $\mathcal{O}\left(h^{\max\left\{2,4-2\beta\right\}}\right)$ when $\delta$ is set as a grid point. We also prove that the shifted-symmetric scheme is asymptotically compatible, which has the global error $\mathcal{O}\left(h^{\min\left\{2,2\beta\right\}}\right)$ as $\delta,h\rightarrow 0$. The numerical experiments (including two-dimensional case) are performed to verify the convergence.