14.07.2013 Views

X - UWSpace - University of Waterloo

X - UWSpace - University of Waterloo

X - UWSpace - University of Waterloo

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

optimal value <strong>of</strong> P. This is explained further below.<br />

Since it is very diffxcult and long to state the general N part decomposition method in<br />

algebraic terms in detail. we will focus on a Cpart case as shown in Figure 4.2. The extension<br />

to any number <strong>of</strong> parts Ncan be straightfoward from the demonstration <strong>of</strong> this Cpart case.<br />

PUU gives an upper bound to the original problem because Pu in the fint level gives an<br />

upper bound to the original problem. and the upper bound subproblem <strong>of</strong> Pu, which is Puv.<br />

provides an upper bound to Pu. With the same reasoning, Pu. provides a lower bound to the<br />

original problem. The upper bounds provided by Pvv are nonincreasing. as the iterations<br />

proceed. and the lower bounds from PLL are nondecreasing. The algorithm proceeds through<br />

iterations <strong>of</strong> parailel solution <strong>of</strong> PLL and PLU, by exchanges <strong>of</strong> primd and dual proposals.<br />

converging towards the optimal solution <strong>of</strong> PL. Simultaneously, PLI and Puu are solved<br />

iteratively. in parallel, converging towards the optimal solution <strong>of</strong> Pu.<br />

Figure 4.2 lpart decomposition principle and information flow.<br />

('Tol" is the predetermined srnall tolemce for judging convergence)

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!