scientific article; zbMATH DE number 1092704

Publication date 2 December 1997

Published in Advances in Numerical Mathematics (Search for Journal in Brave)

This book deals with the numerical solution of systems of hyperbolic conservation laws, mainly in two space dimensions. A large class of schemes consists of first approximating the solution at each time level using cell averages. From these averages a higher-order approximation in each cell is reconstructed using the cell averages also from neighbouring cells. Then the evolution in time by a small time step gives again cell averages on the new time level. A large portion of the book, five out of seven chapters, concentrate on the reconstruction phase of these algorithms. Introducing the reconstruction of smooth functions from cell averages reduces the improvement of the discretisation error to a problem in approximation theory. The problem is now decoupled from the finite volume (FV) method. The theory of optimal recovery is applied to FV methods. This has not been done before. If one uses polynomial reconstruction one obtains trivial reconstructions only. However, the author uses radial basis functions and is lead to completely new FV methods, which use nontrivial reconstructions in Hilbert spaces with reproducing kernels. In the following the contents is briefly reviewed. In the first chapter a brief review of the mathematical theory of hyperbolic conservation laws is given. In addition special systems which are rotational invariant and the system for inviscid compressible gas flow are presented. The next chapter deals with finite volume methods. For simplicity all triangulations are restricted to two space dimensions. Two types of meshes are considered, a primary one which consists of triangles and a secondary mesh which is sometimes called the dual mesh. The author refers to the control volumes as boxes. Then the overall scheme is described as mentioned above. If the time step is infinitesimal this leads to the semi-discretisation. The author concentrates mainly on a time step where the fluxes are integrated along the cell sides. This is done by Gauss quadrature, but for most cases a very low order quadrature is used. Instead of giving a whole zoo of flux functions the author concentrates on the Steger-Warming flux and the Osher-Solomon flux. In Theorem 2.2. the relation between the flux approximation error, the Gauss quadrature and the order in the space approximation is given. The chapter ends with a discussion of the numerical solution of the ordinary differential equation system originating from the semi-discretisation. The Runge-Kutta schemes of Shu and Osher are presented which respect the total variation diminishing property of a space discretisation. The rest of the book is devoted to the reconstruction and the preparation of the mathematical theory for this reconstruction. The longest Chapter 3 deals with polynomial reconstruction and reviews to a large extend existing techniques. For linear and quadratic polynomials structures of programs are given, as well as numerical examples. Applications to compressible flow problems are discussed and the author's DLR-tau method is described. In Chapter 4 the theory of optimal recovery as put forward by Golomb and Weinberger, Micchelli and Rivelin and Sard is presented and extended. The results are used to investigate optimal polynomial reconstruction. It is seen that this leads to trivial algorithms. Semi Hilbert spaces and abstract splines are introduced and it is shown that interpolating splines can be constructed. Chapter 5 deals with global radial functions. First Beppo-Levi spaces are introduced. Since one needs that function evaluations are continuous, one is restricted in 2 dimensions to Beppo-Levi spaces of order larger or equal to 2. A reconstruction by thin plate splines is done for FV methods. Numerical tests using the advection equation and the forward facing step problem with the Euler equations are performed. In Chapter 6 other spaces of radial functions which support splines are presented. In particular conditionally positive definite radial functions are used. One relies on the theory of W. R. Madych and S. A. Nelson. The chapter ends with Wu-Schaback optimality. Since the radial functions used are not local, one has to restrict these to a few control volumes. This leads to strong oscillations in the reconstruction. Hence in the final chapter the author uses local radial functions. The author always supports his theoretical findings with numerical examples. The author states as his main conclusion: `The presented algorithms which use radial function reconstructions will at the moment not be used in practical applications, since these are computationally much more involved than polynomial reconstruction'.

Recommendations

Cited in

(10)

This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4367606)