[Coin-lpsolver] Using ClpCholeskyUfl with choldmod

[Kish Shen]

Hi,

I have been trying out CLP's barrier solver with both UFL_BARRIER and CLP_USE_CHOLDMOD defined, compiled with ClpCholeskyUfl.cpp,  as I was interested in seeing how this compares with using UFL_BARRIER on its own (I assume this means comparing using CHOLMOD versus AMD on its own for doing the Cholesky factorization.

Unfortunately, I seem to get some problems with using CHOLMOD -- I get multiple copies of this warning message from CHOLDMOD:

CHOLMOD warning: not positive definite

when I tried the solver on some test problems, and I don't seem to get the correct solution at the end...
Does anyone have any idea what the problem might be?

I am able to solve these problems when CLP is compiled with just UFL_BARRIER defined with ClpCholeskyUfl, which just use Ufl AMD for the factorization.

I am using the trunk branch of Coin-Cbc (with CLP as the linear solver) that I downloaded 9 Nov, and the various UFL modules used by CHOLMOD within the last two days. I am doing the tests on a (x86 based) Linux machine, using gcc and g77 as the C and Fortran compilers respectively.

Thanks in advance for any information!

-- Kish Shen