Re: LU factorization in Mathematica
- To: mathgroup at smc.vnet.net
- Subject: [mg20660] Re: [mg20637] LU factorization in Mathematica
- From: Daniel Lichtblau <danl at wolfram.com>
- Date: Sun, 7 Nov 1999 02:10:00 -0500
- References: <199911040713.CAA02740@smc.vnet.net.>
- Sender: owner-wri-mathgroup at wolfram.com
Peltio wrote: > > I apologise for this question but I did not find what iI was looking for > on the Big Book. > Any hint will be appreciated. > > LUDecomposition returns a matrix that is a combination of the L and U > matrices. But what kind of combination? > How can I recover the two matrices L and U? > What is the meaning of the pivoting vector (Does P[[k]]=j means that row > k has been swapped with row j ?) > > And what are the meaning of the elements returned by LUfactor (in the > GaussianElimination package ), which returns a different pivot vector > (and matrix as well)? > > Thanks in advance, > Peltio > > peltio AT iname DOT com > warning: Antispam return address <ON> I quote from the Further Examples for LUDecomposition in the Help Browser, under the subheading "Recovering the lower and upper triangular matrices from the LUDecomposition" "LUDecomposition satisfies the following relation: Given a matrix M, determine a unit lower triangular matrix L, an upper triangular matrix U and a permutation vector P such that Part[M, P] == L. U . Part[M, P] rearranges the rows of M." There is code provided in that section to do the extractions you want. As for the permutation vector, P[[k]]==j means that the kth pivot row was originally row j. But it may have moved prior to being swapped into row k, that is, it need not have been the jth row at the time of the pivot selection. LUFactor is an older package function that predates the appearance of LUDecomposition in Mathematica. The pivot vector indeed records actual swaps at the time the pivot was chosen, so it does not represent a permutation. LUFactor is retained primarily for pedagogical reasons because it is tied to related arithmetic packages and hence can effectively demonstrate e.g. LU factorization of a matrix of real intervals. Daniel Lichtblau Wolfram Research