Abstract
The minimal aliasing local spectral (LS) method is a numerical technique that embodies features of both finite-difference (FD) and spectral transform (ST) methods. Anderson first described this method in the context of the one-dimensional advection-diffusion equation. In the current paper, we describe the extension of the LS method to multidimensions. First, we review the one-dimensional version of the LS method from a more rigorous view. In addition, we describe interpolation, differentiation, and dealiasing fitters for the LS method based on Lagrange polynomials. Without the dealiasing filters, this version of the LS method collapses to a standard high-order Taylor series FD scheme. When filter lengths span the integration domain and the dealiasing stage is retained, the LS method becomes an ST method, as described by Anderson. Issues concerning the implementation of the LS method in multidimensions are also discussed. These issues include the form of the high-resolution grid, the implementation of the interpolation stage, and the implementation of the dealiasing stage. Then, we test the LS method with a two-dimensional nonlinear density current problem using idealized boundary conditions. Comparisons are made with a high-resolution reference solution from a reference model, as well as with solutions from a high-order FD model. Results from simulations of the test problem demonstrate that the LS method is more accurate than high-order FD schemes at coarse grid resolutions, and as accurate at finer grid resolutions. Furthermore, the results show that solutions from LS models are more robust than solutions from FD models. After this, we show that dealiasing the nonlinear advection tendencies plays an important role in the success of the LS method, especially for simulations with sharp boundaries that are marginally resolved. For adequately resolved flows, dealiasing does not necessarily improve solutions for the short-term integrations that are presented. However, aliasing errors still must be controlled to prevent a catastrophic buildup of energy at the smallest resolvable wavelengths. Finally, the LS method is tested using open lateral boundary conditions. As the LS method is a higher-order scheme, special treatment of the vertical and lateral boundaries is required. One possibility is to use lower-order versions of the LS method as boundaries are approached, and outflow conditions at the lateral boundaries. This simple treatment results in solutions that compare very favorably to the reference solution of the test problem.
