This preview shows pages 1–8. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: Cholesky Factorization Parallel Dense Cholesky Parallel Sparse Cholesky Parallel Numerical Algorithms Chapter 7 Cholesky Factorization Prof. Michael T. Heath Department of Computer Science University of Illinois at UrbanaChampaign CSE 512 / CS 554 Michael T. Heath Parallel Numerical Algorithms 1 / 52 Cholesky Factorization Parallel Dense Cholesky Parallel Sparse Cholesky Outline 1 Cholesky Factorization 2 Parallel Dense Cholesky 3 Parallel Sparse Cholesky Michael T. Heath Parallel Numerical Algorithms 2 / 52 Cholesky Factorization Parallel Dense Cholesky Parallel Sparse Cholesky Cholesky Factorization Computing Cholesky Cholesky Algorithm Cholesky Factorization Symmetric positive definite matrix A has Cholesky factorization A = LL T where L is lower triangular matrix with positive diagonal entries Linear system Ax = b can then be solved by forwardsubstitution in lower triangular system Ly = b , followed by backsubstitution in upper triangular system L T x = y Michael T. Heath Parallel Numerical Algorithms 3 / 52 Cholesky Factorization Parallel Dense Cholesky Parallel Sparse Cholesky Cholesky Factorization Computing Cholesky Cholesky Algorithm Computing Cholesky Factorization Algorithm for computing Cholesky factorization can be derived by equating corresponding entries of A and LL T and generating them in correct order For example, in 2 2 case a 11 a 21 a 21 a 22 = 11 21 22 11 21 22 so we have 11 = a 11 , 21 = a 21 / 11 , 22 = q a 22 2 21 Michael T. Heath Parallel Numerical Algorithms 4 / 52 Cholesky Factorization Parallel Dense Cholesky Parallel Sparse Cholesky Cholesky Factorization Computing Cholesky Cholesky Algorithm Cholesky Factorization Algorithm for k = 1 to n a kk = a kk for i = k + 1 to n a ik = a ik /a kk end for j = k + 1 to n for i = j to n a ij = a ij a ik a jk end end end Michael T. Heath Parallel Numerical Algorithms 5 / 52 Cholesky Factorization Parallel Dense Cholesky Parallel Sparse Cholesky Cholesky Factorization Computing Cholesky Cholesky Algorithm Cholesky Factorization Algorithm All n square roots are of positive numbers, so algorithm well defined Only lower triangle of A is accessed, so strict upper triangular portion need not be stored Factor L is computed in place, overwriting lower triangle of A Pivoting is not required for numerical stability About n 3 / 6 multiplications and similar number of additions are required (about half as many as for LU) Michael T. Heath Parallel Numerical Algorithms 6 / 52 Cholesky Factorization Parallel Dense Cholesky Parallel Sparse Cholesky Parallel Algorithm Loop Orderings ColumnOriented Algorithms Parallel Algorithm Partition For i,j = 1 ,...,n , finegrain task ( i,j ) stores a ij and computes and stores ij , if i j ji , if i < j yielding 2D array of n 2 finegrain tasks Zero entries in upper triangle of L need not be computed or stored, so for convenience in using 2D mesh network, ij can be redundantly computed as both task...
View Full
Document
 Summer '09
 STAFF
 Algorithms

Click to edit the document details