Parallel computation for neutron diffusion equation

The parallel computation for neutron diffusion equation is discussed, especially the three dimensional problem for which rapid calculation is demanded. The computing time for the three dimensional diffusion code is mostly consumed by the calculation of neutron diffusion equation which is approximated by seven points difference equations, refferred to as inner iteration. Hence, the diffusion code is indeed able to make effective use of parallel capability, by vectorizing the inner most loop, if possible. Mainly investigated in this report are as follows: (1) Six general methods of solution of the seven points difference equations, SOR, SLOR, Modified SLOR, ADI, Cyclic reduction algorithm and Direct method are numerically studied from the viewpoint of their adaptability for parallel computations. (2) Neutron diffusion code ADC is analyzed to vectorize. It is seen that the modified SLOR which is newly developed is best to calculating the seven points difference equations, in the sense that the method well prevents the increase of iterations. It is also seen that the computing time for the ADC code can be reduced to thirty percent by vectorizing the inner iteration with the method of SOR, if it goes well. (author)