[BCP] Incomprehensible values in Clp dual rays

Anders Dohn Hansen adh at imm.dtu.dk
Mon Jan 26 16:51:24 EST 2009


Hi everybody,

I am currently building a Branch-and-Price application in BCP 1.1.3 and
I am experiencing some problems when using CLP. In the instance that I
am currently testing my application on, I start out with a highly
infeasible master problem and I generate columns to restore feasibility
(using the restore_feasibility function). This works fine when using
Cplex, but when switching to CLP, I do not understand the dual rays that
I am given. 
In the instance I have a number of constraints with rhs: =1. Initially,
no columns contain any non-zero coefficients for these constraints.
Using Cplex, I get the following dual ray (actually my example contains
more constraints than this, but I have removed most here for clarity).
The first set of values is for the constraints that are known to be
infeasible initially. The second set is for constraints that will never
be infeasible.

dual_ray = [ -1 -1 -1 -1 -1], [ 0 0 0 0 0]

I add a column with a 1-coefficient in the first constraint and get the
following dual ray:

dual_ray = [ 0 -1 -1 -1 -1], [ 0 0 0 0 0]

The corresponding dual rays for CLP are:

dual_ray = [ 0 0 0 -1 0], [ 0 0 0 0 0]

dual_ray = [ 100000 100000 100000 0 100000], [ 0 0 0 0 0]

Here, I naturally generate a column with a 1-coefficient in constraint 4
as this is the only constraint which I am told is infeasible. The first
dual ray returned by both solvers are valid dual rays, but I do not
understand how CLP can return the second one.
I understand from a prior post
(http://list.coin-or.org/pipermail/coin-discuss/2004-July/001019.html)
that there can be some confusion with the sign of the dual ray, but here
it seems that CLP itself switches?
According to the (discontinued) AAP-example presolve should be turned
off to use Cplex. Should I do something similar for CLP, or am I doing
something else wrong?

Thank you very much.
Best regards,
Anders Dohn

---------------------------------------
Anders Dohn
Ph.D. Student
DTU Management

Technical University of Denmark
Department of Management Engineering
Richard Petersens Plads
2800 Kongens Lyngby, Denmark
Direct: +45 4525 3388
adh at imm.dtu.dk
www.imm.dtu.dk/~adh/
---------------------------------------




More information about the BCP mailing list