Abstract:
A numerical study is carried out of the branch and cut method
adapted for solving the clique partitioning problem (CPP).
The problem is to find a family of pairwise disjoint cliques
with minimum total weight
in a complete edge-weighted graph.
The two particular cases of the CPP
are considered: The first is known as
the aggregating binary relations problem (ABRP),
and the second is
the graph approximation problem (GAP).
For the previously known class of facet inequalities of
the polytope of the problem,
the cutting-plane algorithm is developed.
This algorithm includes the two new basic elements:
finding a solution with given guaranteed accuracy
and a local search procedure
to solve the problem of inequality identification.
The proposed cutting-plane algorithm is used
to construct lower bounds in the branch and cut method.
Some special heuristics are used
to search upper bounds for the exact solution.
We perform a numerical experiment on randomly generated graphs.
Our method makes it possible
to find an optimal solution for the previously studied cases of the ABRP
and for new problems of large dimension.
The GAP turns out to be a more complicated case of the CPP
in the computational aspect.
Moreover, some simple and difficult classes of the GAPs are identified for our algorithm.
Tab. 5, illustr. 1, bibliogr. 32.
Keywords:branch and cut, facet inequality, local search.