Study On Parallel Algorithm Of The Method Of Characteristics For Two-dimensional Neutron Problems

Two-dimensional neutron transport equation has been used in many disciplines,so how to solute for it efficiently and accurately has become the focus of the research.Compared with many other solutions for two-dimensional neutron transport equation,MOC can solve any geometric problems,but it also has its own problems,such as large amount of calculation and memory.There are many efficient acceleration algorithms can reduce the amount of calculation by accelerating convergence.With the enlargement of problem scale,the problems beyond the memory of a single processor cannot be solved.So a single processor has been unable to meet the research of MOC,and the research of efficient parallel algorithm of MOC is very important.Most of the parallel research to MOC cannot solve the problem of memory very well.This paper is to solve the problem of memory of MOC and further reduce the computation time based on the accelerating algorithm,so the parallel method we use is geometric domain decomposition.The parallel algorithm proposed in this paper is parallel pipelined MOC algorithm,the algorithm can not only make the result parallel before fully consistent with the after,but also largely reduce the communication after geometric division.The algorithm has been used in KYLIN-2 program using MPI,by setting parameter MsgMax can have better speedup,and we test three different scale problems all have good speedup and maximum speedup can be 6 with 64 processors.The accelerated method method of the method of characteristics(MOC)is the generalized coarse-mesh finite method(GCMFD).The parallel algorithm for accelerated algorithm,in the final is to solve the sparse matrix equations in parallel,also is the GMRES method in parallel.GMRES method include a series of process,such as inner operation,matrix-vector multiplication and matrix of data operation.This paper used two-dimensional grid block partition method.This method is applied to the KYLIN-2program,and test on three different scale model problems and maximum speedup can be13 with 64 processors.The two parallel algorithms are all applied to the KYLIN-2 program,and has very good speedup.We also test on three different scale model problems and the maximum speedup can be 14 with 64 processors.