X - UWSpace - University of Waterloo
X - UWSpace - University of Waterloo
X - UWSpace - University of Waterloo
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
END<br />
- othenvise. calculate the optimal (or feasible) primai and dual solutions for part t as<br />
k-1 k 1 k-1 k-1<br />
(X:-'A,:-~,Y~-'A~~-') and II*- 1; Stop-<br />
The heuristic algorithm soives the subproblems simultaneously and searches for the<br />
optimum by broadcasting the primal and dual solutions for k >l. Since the feasibility <strong>of</strong> the<br />
original problem and subpmblems are ensured by the step 1. each subproblem dways has a<br />
feasible solution for P, (which wiU be discwed in the next subsection). The algorithm terminates<br />
with an optimal solution <strong>of</strong> the original problem when the difference between the best upper<br />
bound ,' and the best lower bound :: <strong>of</strong> the whole problern gets less than the predetermined<br />
convergence tolerance ES. However. in the cases <strong>of</strong> the sarne repeated objective values three<br />
times in a row in dl upper bound and al1 Iower bound subproblems respectively. the heuristic<br />
aigorithm terminates with a feasible but not optimal solution <strong>of</strong> the original problem.<br />
4S3 Properties <strong>of</strong> the Algorithm for the Second Method<br />
Several properties <strong>of</strong> the parailel decomposition algorithm are discussed in this sechon.<br />
The arguments are sirniiar to those in section 4.4.<br />
Theorem 4.7 verifies chat the algorithm des out the possibility <strong>of</strong> unboundedness <strong>of</strong> the<br />
problem P and <strong>of</strong> any <strong>of</strong> the primal subproblems SP:.<br />
Theorem 4.7 Problem P and al1 subproblem SP: are bounded.