THE CONVERGENCE OF THE GRP SCHEME

This paper deals with the convergence of the second-order GRP (Generalized Riemann Problem) numerical scheme to the entropy solution for scalar conservation laws with strictly convexfluxes. The approximate profiles at each time step are linear in each cell, with possible jump discontinuities (of functional values and slopes) across cell boundaries. The basic observation is that the discrete values produced by the scheme are exact averages of an approximate conservation law, which enables the use of properties of such solutions in the proof. In particular, the "total-variation" of the scheme can be controlled, using analytic properties. In practice, the GRP code allows "saw- teeth" profiles (i.e., the piecewise linear approximation is not monotone even if the sequences of averages is such). The "reconstruction" procedure considered here also allows the formation of "saw teeth" profiles, with an hypothesis of "Godunov Compatibility", which limits the slopes incases of non-monotone profiles. The scheme is proved to converge to a weak solution of the conservation law. In the case of a monotone initial profile it is shown (under a further hypothesis on the slopes) that the limit solution is indeed the entropy solution. The constructed solution satisfies the "finite propagation speed", so that no rarefaction shocks can appear in intervals such that the initial function is monotone in their domain of dependence. However, the characterization of the limit solution as the unique entropy solution for general initial data is still an open problem.