科研成果详情

In this report, we present and analyze the results of applying parallel algorithms to a two dimensional gas dynamics code using front tracking. We also discuss the ideas to generalize the algorithm to three dimensions. The main purpose of this paper is to demonstrate that parallel computations can be applied to complex algorithms. We take front tracking as an example. A number of computations for two dimensional fluid flow have been successful to simulate chaotic mixing, shock interactions and oil reservoir simulation. In most of these cases, further progress depends upon three dimensional computations. Performing systematic studies of chaotic mixing in three dimensions will require the extensive computer resources which parallel computations can offer. In order to develop a reliable parallel programming paradigm for three dimensional studies, we first parallelize the relatively simpler serial two dimensional gas dynamics code on two representative distributed-memory MIMD parallel supercomputers, the iPSC/860 and the NCUBE/2. The main algorithmic issues for front tracking in three dimensions are: (1)the construction of surface grids, (2) the construction of volume grids which are adapted to (i.e. which respect, or do not overlap with) specified surfaces. (3) the efficient computation of interface topology, (4) the resolutin of self intersections in a tangled interface, and (5) parallel computation. Most of these issues are important for numerous methods of computation. We present methods for addressing these issues which are appropriate to the front tracking context.