摘要

In recent years, Mao and his co-workers developed a new type of difference schemes for evolution partial differential equations. The core of the new schemes is to simulate, in addition to the original unknowns of the equations, some quantities that are nonlinear functions of the unknowns; therefore, they maintain additional nonlinear discrete structures of the equations. The schemes show a super-convergence property, and their numerical solutions are far better than that of traditional difference schemes at both accuracy and long-time behavior. In this paper, to understand the super-convergence properties of the schemes, we carry out a truncation error investigation on the scheme maintaining two conservation laws for the linear advection equation. This scheme is the simplest one of this type. Our investigation reveals that the numerical errors of the scheme produced in different time steps are accumulated in a nonlinear fashion, in which they cancel each other. As to our knowledge, such an error self-canceling feature has not been seen in other numerical methods, and it is this feature that brings the super-convergence property of the scheme.