The Research And Implementation Of Parallel Algorithms Of Molecular Dynamics For Solidification Of Liquid Metals

Metal solidification process is an extremely complicated process, which includesmacro process and micro process. Molecular dynamics method can simulate the liquidmetals solidification at the microscopic level. In the simulation of moleculardynamics, the simulation results are closer to the experiment results when thenumbers of atom involved in simulated system increases. And thus these moreaccurated simulation results can help people to observe the microscopic properties ofthe metals solidification processes and predict the trend of development, along withproviding producing practice with guidance. However, the serial program for MDcalculations in our study of liquid metals solidification processes simulation is at lowspeeds, and only a small number of atoms are involved in the simulation, leading tothat the simulated results are significantly different from the actual scenarios. Theadvent of the parallel computation has provided large-scale MD simulationcalculations with condition and platform, accelerating the MD simulation withaccurate results and promoting the development of the parallel molecular dynamics.In this paper, the theory of molecular dynamics and parallel computing aresimply reviewed. With these basic theory and method of parallel computing, theparallel program with PVM parallel architecture previously adopted is modified intothe parallel program adopting MPI+OpenMP model to simulate the solidificationprocesses. The main work of this paper contains the following three points:First of all, considering that the previous PVM parallel program can onlysimulate the system involving50000-100000atoms, along with low calculationefficiency and speed. We proposed the MPI+OpenMP parallel computing architecturefor large scale simulation with excellent computing performance of the Tianhesupercomputer. MPI is used to implement the parallel computing and communicationfor CPU heterogeneous computing platforms while OpenMP is used to implement theparallel implementation for multi-core platforms in MD simulations. Secondly, we realized the parallel algorithm with FORTRAN language. Thealgorithm is analyzed and compared with the previous programs in the aspects of theperformance of programs, showing better speedup, parallel efficiency, and scalability.The number of atoms in the simulating system incereases from50,000to5,000,000,enlarging up to100times.Finally, many microstructure analysis methods in numerical simulation analysis are adopted to verify the validity of the simulation system involving5,000,000-Alatoms. From these results, it is clear that the simulation results are in good agreementwith the experimental results. The microscopic properties of the metals solidificationprocesses can be observed. These properties can not be obtained by the realexperiment and small-scale simulation. And then the correctness of the algorithm canbe also proved by the results of our implementation.