16.11.2012 Views

Wireless Network Design: Optimization Models and Solution ...

Wireless Network Design: Optimization Models and Solution ...

Wireless Network Design: Optimization Models and Solution ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

78 J. Cole Smith <strong>and</strong> Sibel B. Sonuc<br />

containing the condition that xi ≤ ⌊ f ⌋, <strong>and</strong> the other with xi ≥ ⌈ f ⌉. The subproblems<br />

are solved recursively, <strong>and</strong> the best solution obtained from the two subproblems is<br />

optimal.<br />

Note that the recursive application of branch-<strong>and</strong>-bound implies that the branching<br />

operations may need to be applied to subproblems at multiple levels of the recursive<br />

algorithm. The branches created throughout this algorithm yield a branch<strong>and</strong>-bound<br />

tree.<br />

Before presenting the remainder of the branch-<strong>and</strong>-bound algorithm, we illustrate<br />

the first stages of the branching process on formulation (4.10). First, observe that the<br />

optimal solution to the LP relaxation of (4.10) is x0 = (12/7,27/7). Of course, x0 is<br />

infeasible to (4.10), because x 0 1 <strong>and</strong> x0 2<br />

are both fractional. Therefore, the next step<br />

in our algorithm is to branch <strong>and</strong> either eliminate the portion of the feasible region<br />

in which 1 < x1 < 2, or the portion in which 3 < x2 < 4. We (arbitrarily) choose to<br />

branch on x2. As a result, we create two subproblems:<br />

Subproblem 1: max x1 + x2<br />

(4.11a)<br />

s.t. constraints (4.9b) − (4.9e), (4.10c) (4.11b)<br />

x2 ≤ 3 (4.11c)<br />

Subproblem 2: max x1 + x2<br />

(4.12a)<br />

s.t. constraints (4.9b) − (4.9e), (4.10c) (4.12b)<br />

x2 ≥ 4. (4.12c)<br />

Figure 4.6 illustrates these two new subproblems, which we call R1 <strong>and</strong> R2, respectively<br />

associated with subproblems 1 <strong>and</strong> 2.<br />

Now, both regions R1 <strong>and</strong> R2 are active, meaning that they must be searched by<br />

recursively solving both of the associated subproblem LPs. This process is depicted<br />

in the branch-<strong>and</strong>-bound tree shown in Figure 4.7. We must solve subproblems over<br />

Fig. 4.6 Feasible regions for<br />

branch-<strong>and</strong>-bound algorithm<br />

for IP (4.10)<br />

x2<br />

R2<br />

R1<br />

x2 ≥ 4<br />

x2 ≤ 3<br />

x1

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

Saved successfully!

Ooh no, something went wrong!