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.

4 An Introduction to Integer <strong>and</strong> Large-Scale Linear <strong>Optimization</strong> 95<br />

B −1 is a bottleneck operation in the simplex method. (It is worth noting that modern<br />

simplex implementations do not compute B −1 directly, but instead use numerical<br />

linear algebra methods that improve numerical stability <strong>and</strong> reduce computational<br />

effort by exploiting the sparsity of the A matrix.) Therefore, if special structures<br />

exist in the constraint matrix, it may be possible to compute B −1 by specialized<br />

algorithms rather than by using general methods from linear algebra. Basis partitioning<br />

algorithms compute B −1 by decomposing B into specially structured blocks,<br />

<strong>and</strong> then employing a specialized algorithm to find corresponding blocks to its inverse.<br />

Only a small submatrix of B −1 needs to be computed under these schemes,<br />

which reduces both computational <strong>and</strong> memory requirements for solving problems<br />

by the simplex method. Foundational work in this area was performed by Schrage<br />

[25], <strong>and</strong> has been applied to network flow problems in several studies; see [18, 19]<br />

for multicommodity flow applications.<br />

4.3.4.2 Interior Point Methods<br />

As mentioned previously, interior-point methods were originally developed to solve<br />

nonlinear optimization problems. These algorithms had a dramatic implication in<br />

the field of linear programming in the early 1980s by yielding the first polynomialtime<br />

algorithm for linear programming problems. Interior point algorithms are more<br />

sophisticated than simplex-based algorithms. Due to worst-case time-complexity<br />

functions that are polynomial in the input size of the linear program, they often<br />

outperform simplex approaches for very-large-scale problems (roughly 10,000 variables<br />

or more), depending on problem-specific details such as the presence of special<br />

constraint structures. See, e.g., [12, 21, 31] for discussion on these methods on<br />

linear <strong>and</strong> nonlinear optimization problems. There is currently much development in<br />

computation <strong>and</strong> theory for interior-point algorithms, some of which is specifically<br />

focused on interior-point approaches for large-scale problems (e.g., [9]).<br />

4.3.4.3 Heuristics<br />

Exact optimization techniques such as the ones described above can become impractical<br />

because of the size <strong>and</strong>/or complexity of real-world instances. Rather than<br />

eschewing optimization as an option, one can turn to heuristic methods for solving<br />

problems. Heuristics are designed to quickly provide good-quality feasible solutions,<br />

but without a guarantee of optimality, or in some cases, without a guarantee<br />

of feasibility.<br />

Naturally, heuristic techniques are as diverse in complexity <strong>and</strong> effectiveness as<br />

are exact optimization techniques. Many heuristic algorithms are tailored specifically<br />

for the problem at h<strong>and</strong>, employing specific properties of the problem to obtain<br />

fast <strong>and</strong> effective heuristic techniques. There are too many such approaches to name,<br />

but one of the most famous of these is the Lin-Kernighan algorithm [16] for the Euclidean<br />

traveling salesman problem. A common trend over the last few decades in

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

Saved successfully!

Ooh no, something went wrong!