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

Create successful ePaper yourself

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

4 An Introduction to Integer <strong>and</strong> Large-Scale Linear <strong>Optimization</strong> 81<br />

also say that it is fathomed by integrality in case 3 since it is an integer solution.)<br />

Subproblem 6 yields another integer solution x 6 = (0,5) which gets fathomed both<br />

by bound <strong>and</strong> by integrality as well. Because there are no more active branches left,<br />

the algorithm terminates, <strong>and</strong> we conclude that the incumbent solution x ⋆ = (2,3)<br />

is an optimal solution for our problem.<br />

It is important to note that the number of subproblems generated within branch<strong>and</strong>-bound<br />

can grow exponentially in terms of the number of integer variables in the<br />

corresponding IP. Therefore, a practical concern is that the algorithm will not terminate<br />

within a reasonable amount of time. The execution time for branch-<strong>and</strong>-bound<br />

can be reduced by adding certain valid inequalities that cut off fractional solutions,<br />

which would otherwise need to be explored <strong>and</strong> eliminated by branch-<strong>and</strong>-bound.<br />

Algorithmic considerations may attempt to improve computational time by prescribing<br />

“node selection” rules (i.e., the order in which subproblems are branched on <strong>and</strong><br />

explored) <strong>and</strong> “variable selection” rules (i.e., the choice of variables to branch on<br />

when case 4 above is encountered). The field is far too large <strong>and</strong> complex to be discussed<br />

in any detail here, <strong>and</strong> so we refer to [20] for a comprehensive treatment of<br />

the subject.<br />

However, some problems appear to be too difficult to solve, either because of<br />

the problem complexity or size, regardless of the algorithmic tricks used to solve<br />

them. If the algorithm must be terminated early, then we can still attempt to assess<br />

the quality of the incumbent solution. Recall that, for a maximization problem, the<br />

incumbent solution’s objective function value is a lower bound (LB) on the optimal<br />

IP solution. The upper bound (UB) is the largest objective function value present<br />

at any active node of the branch-<strong>and</strong>-bound tree. We say that the optimality gap is<br />

the difference UB − LB, <strong>and</strong> the relative optimality gap is (UB − LB)/LB × 100%.<br />

When the relative optimality gap is sufficiently small, we may terminate the branch<strong>and</strong>-bound<br />

procedure with the knowledge that the incumbent solution is within a<br />

certain percentage of optimality. (The relative optimality gap is given with respect to<br />

the incumbent solution’s objective function value. Thus, for minimization problems,<br />

the incumbent solution’s objective function value yields an upper bound, <strong>and</strong> the<br />

relative optimality gap replaces LB with UB in the denominator.) In the example<br />

above, after subproblems 2 <strong>and</strong> 3 are solved, we have that LB = 5 <strong>and</strong> UB = 11/2.<br />

Thus, the relative optimality gap is (0.5/5)×100% = 10%.<br />

4.3 Large Scale <strong>Optimization</strong><br />

Many real-world optimization applications require very large mathematical programming<br />

models for which classical optimization techniques require too much<br />

time or memory to solve to optimality, due to a prohibitively large number of variables<br />

<strong>and</strong> constraints. In some formulations, though, the problem would be separable<br />

(solvable as a series of smaller optimization problems) if not for the presence of<br />

variables that are present in multiple sets of constraints, or constraints that jointly<br />

constrain several sets of variables. In this situation, we say that the problem con-

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

Saved successfully!

Ooh no, something went wrong!