A Taylor-series approach to numerical accuracy and a third-order scheme for strong convective flows (Q1094990)

The use of the Taylor-series theorem to study the accuracy of numerical discretization schemes is reexamined. A generalized equation is presented for a given derivative, to any required order of accuracy. The relationship between the Taylor-series truncation errors of each term in the parent differential equation and the solution error at each node is established through the concept of an inevitable error. Based on Taylor- series error analysis, a stable, third-order accurate numerical method for the convection term is presented which involves the same number of computational nodes as the second-order QUICK scheme. A two-dimensional test problem shows the new scheme generally to produce smaller solution and inevitable errors than other schemes.