From: Nicolas N. <Nic...@iw...> - 2002-09-26 13:39:28
|
"Michael A. Koerber" <ma...@ll...> writes: > Nicolas, > > 1. (M/ A) eventually calls (GESV A I), where I is the identity > matrix, which intern calls the FORTRAN ?GESV which performs the LU > decomposition on the system of equations [A I] to form the inverse > of A. > > 2. Your statement concerning inverses is correct...and it is correctly > implimented in MATLISP. > > > The point is that computing the inverse is often not recommended > > because of numerical stability. Usually one uses the LU decomposition > > directly which is done by the getrf/getrs-pair. Of course, if you > > only need to apply the inverse once you could be satisfied with m/. > > The problem with using only GESV and M/ is that it computes an LU decomposition (which is the major O(N^3) part of the work). To use it again (e.g. in a time-stepping scheme) you need GETRS. Nicolas. |