Originally posted by: g.liuzzi

Hi everybody,
I am interested in solving MIQP problems using CPLEX with callbacks.
I would like to use a solveCallback to pass CPLEX, at every node,
an initial solution computed using my routine.
The CPLEX barrier optimizer seems not to use the given solution but
it reoptimizes the node problem from scratch.
To pass the initial solution I am using CPXcopystart().
Any help or suggestions ?

Thanks.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

The barrier algorithm does not support warm starts. Hence, every starting solution that you provide will be ignored.

Note that you can also solve QPs with the simplex algorithm. This is able to exploit warm start information. But if you only have a primal/dual vector, the QP simplex first needs to go through a crossover step to generate a basic solution of the underlying KKT system.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: vpiccialli

I have a similar question: If for some reason we know in advance that the solution I want to pass to cplex is optimal at the current node, is it possible to tell cplex not to waste time in certifying optimality?This is to avoid additional computational time.

Similarly, is it possible to use a different relaxation solver in the solve callback?

Thank you for any help.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: RWunderling

CPLEX depends on the status code in the node LP as you can query with CPXgetstat() to continue to work.
This status can only be changed by calling an optimization function. This implies that any solution
provided e.g. in terms of a starting vector will be verified. Note, however, that this optimization is
in general no wasted time, since CPLEX uses more information than just the x vector to conduct the MIP
optimization (e.g. for reduced cost fixing).

• Roland

#CPLEXOptimizers
#DecisionOptimization