Spatial And Frequency Decomposed Parallel Computing In Specific-domain Applications

Numerical computing plays an important role in modern scientific research. It helps researchers recognize and set up more and more complex process models. Parallel computing is used to solve time-consuming and scaling-constraint problems in scientific simulation. Parallel computing involves supporting hardware/system platform and numerical applications, where the multi-disciplinary applications have enormous influence in the scientific research and engineering areas in practice.To develop practical parallel computing numerical software, we need transform the original serial computing design into parallel computing model. Domain decomposition is the most popular parallelization method. In this approach, the simulation domain can be divided into spatial and frequency domain and others in real-world problems. It is the first step to design appropriate domain decomposition method before developing a parallel computing package. This article focuses on the practical application of these two parallel modes. Based on the spatial domain decomposition and frequency domain decomposition methods, an underground environment simulation problem and an electromagnetic field calculation problem using above two modes are developed in this paper.This paper proposed the design and implementation of parallel computing with the two practical problems. The research on the underground environment simulation was supported by China Geological Survey Program “National CO2 geological storage potential evaluation and demonstration project”. Based on US Lawrence Berkeley Laboratory TOUGHREACT simulation model, the author worked with the inter-disciplinary research group from environmental science and computer science to construct THC-MP, which is the code running the distributed memory computing platform. And THC-MP successfully supported the simulation of large scale 3D model made from Ordos basin reservoir.Research on the electromagnetic field research was supported by joint-doctoral program. The author collaborated with electromagnetic research group and realized the high performance HBFEM method. Aim to efficiently calculate the HBFEM using the harmonic decomposition in multi-frequency domain. This paper proposed the parallel algorithm and applied the method to a DC bias transformer sample. Frequency domain decomposed parallel computing provides an approach to effectively simulate the steady-state electromagnetic and industrial modeling problems.The main contributions of this paper are:1. Effective parallel simulation of the coupling multiphase flow, solute transport and chemical reaction in environmental science was done. Based on the spatial domain decomposition method, a completed implementation of parallel computing in the coupled process was done. The flow, transport and chemical reaction data was rearranged according to domain decomposition results. We designed coupled parallel computing framework based the uniform spatial decomposition. Hybrid computing design of matrix operation and grid processing at the core equations solving stage was proposed. Also the load balance of the parallel computing in coupled process was optimized from the runtime data.2. A coupling process T-H-C(Thermal-hydro-chemical) parallel computing framework was developed, and the author further implemented the stable THC-MP simulator. Combined with the ECO2 N module for the carbon dioxide sequestration simulation, a reservoir model from Songliao Basin was calculated to verify the reliability of the numerical parallel computing. The simulator was also applied to geological storage of carbon dioxide process simulation of Shiqianfeng reservoir formation at Ordos Basin on the distributed memory parallel computing platform. The simulation results show the performance of the large scale model simulation is well.3. Calculation of HBFEM problem with the frequency domain decomposition method. This paper designed a block-matrix based quasi-Jacobi parallel iteration method with harmonic decomposition for solving the equations arising from the HBFEM. And a high-frequency DC-bias transformer contains multiple harmonics problem was investigated. Parallel computing with the parallelism that equivalent to the number of harmonics was to verify the numerical accuracy. The experiment shows the parallel computing with decomposition in the frequency domain successfully reduced the HBFEM matrix size and simulation time.