Solving large systems of linear ordinary differential equations on a vector computer (Q1116662)

This communication is concerned with the numerical solution of large systems of ODEs on a computer consisting of two vector processors. Given the system: \(v'+Av=b\), the constant matrix A is expressed as the matrix sum \(A=D+L_ 1+L_ 2,\) where D is a diagonal matrix and \(L_ 1\) and \(L_ 2\) are respectively lower and upper triangular matrices. Then the method advances from \(v_ 0=v(t_ 0)\) to a new approximation \(v_ 1\) computing \(w_ i\), \(i=1,2\), from the independent triangular systems \(H_ iw_ i=K_ iv_ 0+b\) and then \(v_ 1\) is the arithmetic mean \(v_ 1=(w_ 1+w_ 2)/2.\) Here \(H_ i\) and \(K_ i\) are easily computable matrices in terms of \(L_ i\) and D. It is shown that the method is consistent and when A is a symmetric positive definite matrix, is stable for all \(h>0\). The paper ends with some numerical experiments which arise in solving an initial-boundary value problem for a diffusion equation on a rectangular domain by the method of lines. The problem is solved with a CRAY X-MP computer for different initial and boundary conditions and the speed using two processors is almost two times faster than the corresponding to one processor.