In this work we propose an efficient parallelization of multiple-precision Taylor series method with variable stepsize and fixed order. For given level of accuracy the optimal variable stepsize determines higher order of the method than in the case of optimal fixed stepsize. Although the used order of the method is greater then that in the case of fixed stepsize, and hence the computational work per step is greater, the reduced number of steps gives less overall work. Also the greater order of the method is beneficial in the sense that it increases the parallel efficiency. As a model problem we use the paradigmatic Lorenz system. With 256 CPU cores in Nestum cluster, Sofia, Bulgaria, we succeed to obtain a correct reference solution in the rather long time interval - [0,11000]. To get this solution we performed two large computations: one computation with 4566 decimal digits of precision and 5240-th order method, and second computation for verification - with 4778 decimal digits of precision and 5490-th order method.
翻译:在这项工作中,我们建议以可变步骤和固定顺序对多精度泰勒系列方法进行高效的平行。根据给定的精确度,最佳可变步骤决定了方法的更近顺序,而不是最佳固定步骤。虽然这种方法的用序比用定的阶梯确定得更高。在固定的阶梯化的情况下,方法的用序比用法的用序要高,因此每步的计算工作要大,因此,每步的减序使整体工作较少。此外,方法的更大顺序有利于提高平行效率。作为一个示范问题,我们使用了范式Lorenz系统。在保加利亚索菲亚的内斯图姆集群中,256个CPU核心在相当长的时间间隔里成功地获得了正确的参考解决方案 — [0,1000]。为了获得这一解决方案,我们进行了两次大型计算:一次计算,4,5,566位小数的精确数和5,240个顺序方法,以及第二次计算核查,4,778个十位精确数和5490个顺序方法。