No label defined (Q4454375)
From MaRDI portal
!
This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:
scientific article; zbMATH DE number 2052738
| Language | Label | Description | Also known as |
|---|---|---|---|
| default for all languages | No label defined |
||
| English | No label defined |
scientific article; zbMATH DE number 2052738 |
Statements
8 March 2004
0 references
semidiscretizations
0 references
skew-adjoint finite difference methods
0 references
nonuniform grids
0 references
differentiation matrices
0 references
stability
0 references
error bounds
0 references
numerical examples
0 references
wave equations
0 references
reaction-diffusion equations
0 references
0.7657780051231384
0 references
0.7519041895866394
0 references
0.7416219711303711
0 references
0.7399667501449585
0 references
Differentiation matrices are studied that are skew-adjoint with respect to a diagonal discrete inner product. It is shown that these can be used to construct conservative and hence nonlinearly stable semidiscretizations of certain partial differential equations. Families of such finite differences of any order are constructed on arbitrary periodic, one-dimensional grids and are studied for their performance with respect to the roughness of the grid for linear and nonlinear wave equations. The existence of finite difference methods which are skew-adjoint with respect to diagonal bilinear forms is established. To get this property and have order greater than 1 (the limit for the standard mapping method) on rough grids, one has to pay a price: the coefficients of the finite differences are globally coupled, and the bandwidth is increased. It is noted that the bandwidth 5 methods appear to improve to order 4 on \(C^1\) grids, which is the order of a standard bandwidth 5 method. As the grid becomes rougher or less differentiable the methods remain stable and their accuracy degrades gracefully. However, for very rough grids, e.g. if \(\frac{{h_i }} {{h_{i + 1} }}\) is too small, even these methods eventually lose stability, in the sense that none of the possible diagonal \(S\) matrices are positive. The periodic boundary conditions are mostly considered. Other boundary conditions remain an important problem for the future. They change the counting in the order conditions but not the overall approach. If the bandwidth of \(A\) is large, then one loses unknowns in \(A\) near an (e.g.) Dirichlet boundary, and the local order of approximation has to be reduced appropriately at such grid points. Another possibility is to consider differences formed from narrow stencils, e.g. using \(u_{i + 1} - u_i \) to approximate \(hu'\left( {x_{i + {1 / 2}} } \right)\). As in the uniform grid case, these are likely to be more accurate that the wide stencils. Relaxing the requirement that \(S\) be diagonal would give much more scope, indeed this is common in the summation-by-parts literature, but, as shown in Section 2, it does rule out the direct application to nonlinear equations, if one wants to preserve, e.g., energy. It remains to be seen if the present approach is practical for higher-dimensional equations such as fluid and reaction-diffusion equations on rough grids.
0 references