Dear Mr. Brian Gough, I'm sorry for this late reply, but I spent some time to write a example program and make sure everything is correct. I found the problem with gsl_linalg_cholesky_decomp(A) and gsl_linalg_cholesky_svx(A,X) is that if you use a matrix A with upper triangle elements being zeroes, gsl_linagl_cholesky_decomp(A) will do calculation as if it were symmetric. Of course, cholesky decomposition can only apply for positive definite and symmetric matrix. However, when you udr a non-symmetric matrix as input, it should at least give some error message. I attach my test program to this email. I wish it may be of a little help. Thanks, Chuhu
On 9/18/06 2:55 PM, "Brian Gough" <[EMAIL PROTECTED]> wrote: > Chuhu Yang wrote: >> Dear Mir or Madam, >> Recently, I was using gsl. I found one problem with the function >> gsl_linalg_cholesky_svx(A,x). >> I want to obtain a solution to a linear function AX=b. I used >> gsl_linalg_cholesky_decomp(A) first. I obtained a correct Cholesky factor L. >> A after calling gsl_linalg_cholesky_decomp is a combination of L and L'. >> Then I used gsl_linalg_cholesky_svx(A,X) with A from the first step. I >> couldn't obtain correct solution for X. I guess there is a bug in >> gsl_linalg_cholesky_svx(A,X). >> However, I can get a correct solution by using >> gsl_linalg_HH_svx(A,X) directly. >> I hope this might help to make the function gsl_linalg_cholesky_svx better. >> Thanks, > > Hello, > Thank you for your email -- please could you send a small example > program that we can use to reproduce the problem.
test.cc
Description: Binary data
_______________________________________________ Bug-gsl mailing list [email protected] http://lists.gnu.org/mailman/listinfo/bug-gsl
