Abstract
Parallel block methods of BDF(backward differentiation formulae)type are studied for solving stiff differential equations. In order to treat errors of parallel and serial methods on a unified basis, the parallel block methods are reformulated in the form of the general linear method, which leads to a natural definition of global errors. The computational costs, errors and speed-up ratios of the proposed block methods are compared with those of the conventional BDF. It showld be noted that there exists an efficient method which reduces the computational cost of the conventional BDF of the same order by a factor of 5 using only two processors. The reason of the achievement is discussed. Experimental results which demonstrate the validity of the theoretical analysis are presented.
