Parallel Processing for Matrix ComputationMakoto Natori, Takashi Nodera |
From inside the book
Results 1-3 of 3
Page 37
... nodes are numbered in such a way that the initial node of each edge is labeled with a number smaller than that of its terminal node . We recall that any no - fillin ( M ) ILU factorizations are algebraically equivalent if they are based ...
... nodes are numbered in such a way that the initial node of each edge is labeled with a number smaller than that of its terminal node . We recall that any no - fillin ( M ) ILU factorizations are algebraically equivalent if they are based ...
Page 39
... node pij . Using stencil expression , the remainder R ( = M - A ) for unmodified factorization is written as Cij fi - 1j / 8i - 1j ( R ) ij = bijeij - 1 / 8ij - 1 Subtracting these remainder terms from the unmodified diagonal ;; results ...
... node pij . Using stencil expression , the remainder R ( = M - A ) for unmodified factorization is written as Cij fi - 1j / 8i - 1j ( R ) ij = bijeij - 1 / 8ij - 1 Subtracting these remainder terms from the unmodified diagonal ;; results ...
Page 40
... nodes in PB ( m ) and PD ( m ) orderings . Subtracting these remainder terms from the diagonal § ;; for red - black ILU factorization results in red - black MILU factorization ( 3.3 ) Sij = dij - ( bij fij - 1 / dij − 1 + Cij¤i − 1 ...
... nodes in PB ( m ) and PD ( m ) orderings . Subtracting these remainder terms from the diagonal § ;; for red - black ILU factorization results in red - black MILU factorization ( 3.3 ) Sij = dij - ( bij fij - 1 / dij − 1 + Cij¤i − 1 ...