[Coin-lpsolver] CLP and QUADOBJ

[Matthew Guthaus]

Hi all,

I've been successfully using CLP to solve QP problems. However, I had  
a QP problem that seemed to be taking a while so I decided to  
writeMPS and run it in CPLEX for comparison. CPLEX complained when  
the MPS file specifies the full Q matrix with:

CPLEX Error  5011: Duplicate entry in 'objective' for pair 'C0000012'  
and 'C0000014'.

It is assuming that the matrix is symmetric and I should only specify  
one entry (C12,C14) or (C14,C12). My question is, is this part of the  
MPS specification or is this an assumption made in CPLEX? Is CLP  
making the wrong assumption? Is it because CLP handles arbitrary Q  
matrices and CPLEX can only handle semi-definite?

The other issue (which is actually a CPLEX problem, but maybe someone  
will know!) is that CPLEX then claims my matrix is not semi-definite  
when it surely is...

Q = (x2-x3)^2 + (x2-x4)^2 + (x2-x5)^2 + (x3-x4)^2 + (x3-x5)^2 + (x4- 
x5)^2

Q = [ 0 0 0 0 0 0; 0 0 0 0 0 0 ; 0 0 3 -1 -1 -1 ; 0 0 -1 3 -1 -1 ; 0  
0 -1 -1 3 -1 ; 0 0 -1 -1 -1 3 ]
octave:8> eig(Q)'
ans =

   0.00000  0.00000  0.00000  4.00000  4.00000  4.00000



Thanks,

Matt