Estimating Discretization Error with Preset Orders of Accuracy and Fractional Refinement Ratios

TL;DR

This paper introduces POEM, a grid refinement method that uses preset accuracy orders and fractional ratios to reliably estimate discretization errors with lower computational costs and controlled uncertainty.

Contribution

The paper proposes POEM, a novel grid refinement approach that guarantees optimal accuracy orders and incorporates fractional ratios to improve error estimation reliability and efficiency.

Findings

POEM guarantees optimal accuracy orders through iterative refinement.

Using fractional ratios with MIDAS reduces computational cost and error uncertainty.

POEM effectively estimates discretization errors in advection and diffusion problems.

Abstract

Verification of solutions is crucial for establishing the reliability of simulations. A central challenge is to find an accurate and reliable estimate of the discretization error. Current approaches to this estimation rely on the observed order of accuracy; however, studies have shown that it may alter irregularly or become undefined. Therefore, we propose a grid refinement method which adopts constant orders given by the user, called the Preset Orders Expansion Method (POEM). The user is guaranteed to obtain the optimal set of orders through iterations and hence an accurate estimate of the discretization error. This method evaluates the reliability of the estimation by assessing the convergence of the expansion terms, which is fundamental for all grid refinement methods. We demonstrate these capabilities using advection and diffusion problems along different refinement paths. POEM requires a lower computational cost when the refinement ratio is higher. However, the estimated error suffers from higher uncertainty due to the reduced number of shared grid points. We circumvent this by using fractional refinement ratios and the Method of Interpolating Differences between Approximate Solutions (MIDAS). As a result, we can obtain a global estimate of the discretization error of lower uncertainty at a reduced computational cost.