High Performance Computing 1
MG 1D
•Let v(j) be the j-th eigenvector of T(i), and V the eigenvector matrix V=[v(1),v(2),...] whose j-th column is v(j). V is an orthogonal matrix, so any vector x can be written as a linear combination of the v(j).
•alpha(j) the j-th frequency component of x,
•On the upper half of the frequency domain, multigrid acting on P(m), damps the error in the solution by the solution operator S(i). On the next coarser grid, with half as many points, multigrid damps the upper half of the remaining frequency components in the error. On the next coarser grid, the upper half of the remaining frequency components are damped, and so on, until we solve the exact (1 unknown) problem P(1).
•