A parallel algorithm for partially separable non-convex global minimization: Linear constraints (Q2276880)

The global minimization of large-scale partially separable non-convex problems over a bounded polyhedral set using a parallel branch and bound approach is considered. The objective function is a sum of a separable concave part, an unseparated convex part and a linear part. These large- scale problems are characterized by having the number of linear variables much greater than the number of nonlinear variables. A convex underestimating function to the objective function is constructed and minimized over the feasible domain to get both upper and lower bounds on the global minimum function value. At each iteration of the algorithm the feasible domain is divided into subregions and convex underestimating problems over each subregion are solved in parallel. Branch and bound techniques are used. Computational results are also presented.